**Preface to "Cyanotoxins in Bloom: Ever-Increasing Occurrence and Global Distribution of Freshwater Cyanotoxins from Planktic and Benthic Cyanobacteria"**

At present, cyanobacteria and their toxins (also known as cyanotoxins) constitute a major threat for freshwater resources worldwide. Cyanotoxin occurrence in water bodies around the globe is constantly increasing, whereas emerging, less studied or completely new variants and congeners of various chemical classes of cyanotoxins, as well as their degradation/transformation products are often detected. In addition to planctic cyanobacteria, benthic cyanobacteria, in many cases, appear to be important toxin producers, although far less studied and more difficult to manage and control. This Special Issue highlights novel research results on the structural diversity of cyanotoxins from planktic and benthic cyanobacteria, as well as on their expanding global geographical spread in freshwaters.

> **Triantafyllos Kaloudis, Anastasia Hiskia, and Theodoros Triantis** *Editors*

### *Editorial* **Cyanotoxins in Bloom: Ever-Increasing Occurrence and Global Distribution of Freshwater Cyanotoxins from Planktic and Benthic Cyanobacteria**

**Triantafyllos Kaloudis 1,2,\*, Anastasia Hiskia <sup>2</sup> and Theodoros M. Triantis <sup>2</sup>**


Toxic cyanobacteria in freshwater bodies constitute a major threat to public health and aquatic ecosystems [1]. Cyanobacterial blooms are increasing in frequency, magnitude and duration globally, while eutrophication, rising CO2 and climate change promote their global expansion [2,3]. Toxic cyanobacteria metabolites, known as cyanotoxins, comprise a wide range of compounds, including cyclic peptides (microcystins, nodularins) and alkaloids (cylindrospermopsins, anatoxins, saxitoxins) that can be hepatotoxic, cytotoxic, genotoxic or neurotoxic. In response to the risks associated to known cyanotoxins, the World Health Organization (WHO) has published guidelines for their monitoring and management, including provisional guideline values for exposure via drinking water and recreational activities [4]. Nevertheless, the high metabolic potential of cyanobacteria yields a plethora of secondary metabolites that are largely understudied. A recently developed database of cyano-metabolites reported in the literature (CyanoMetDB) contains more than 2000 molecules, including more than 300 microcystin congeners [5]. Still, research so far on the occurrence and impacts of cyano-metabolites has mostly focused on a small number of cyanotoxins, particularly on a few microcystin congeners.

This Special Issue aims to present novel research results on the presence and structural diversity of cyanotoxins and cyano-metabolites in freshwater bodies worldwide. We welcomed research and review papers that showcase the expanding global geographical spread of cyanotoxins, including reports from less-studied areas and on understudied cyanotoxins and cyano-metabolites. We particularly encouraged advances and novelties in the areas of cyanotoxin analysis and monitoring, structural elucidation of new cyanometabolites, biotic and abiotic factors linked to cyanotoxin production and the role of benthic cyanobacteria as cyanotoxin producers.

A number of published papers reported the presence of toxic cyanobacteria and cyanotoxins using diverse monitoring techniques, in freshwater bodies encompassing Central and South European, Mediterranean, Southeast Asian and North American regions. Van Hassel et al. [6] reported results from monitoring of cyanobacterial blooms in lakes of Wallonia, Flanders and Brussels, Belgium, using LC-MS/MS, PCR and sequencing techniques, to assess the risks associated to recreational waters. More than 20% of samples exceeded the WHO guideline value for microcystins, while the *mcyE* gene was detected in 76% of samples. Fournier et al. [7] investigated the deep-water, red-pigmented biomass occurrences in Lake Constance, which is the third largest lake in Central-Western Europe that borders Germany, Austria and Switzerland. Using 16S rRNA gene-amplicon sequencing and LC-MS/MS they showed that these blooms were contributed by microcystin-producing *Planktothrix* spp. A one-year monitoring study of Slovenian waterbodies using qPCR (*mcyE*, *cyrJ*, *sxtA* genes) and LC-MS/MS (microcystins, cylindrospermopsin, saxitoxin) was

**Citation:** Kaloudis, T.; Hiskia, A.; Triantis, T.M. Cyanotoxins in Bloom: Ever-Increasing Occurrence and Global Distribution of Freshwater Cyanotoxins from Planktic and Benthic Cyanobacteria. *Toxins* **2022**, *14*, 264. https://doi.org/10.3390/ toxins14040264

Received: 26 March 2022 Accepted: 1 April 2022 Published: 8 April 2022

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2022 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

conducted by Zupancic et al. [8]. Potentially toxic *Microcystis* and *Planktothrix* cells were detected by qPCR and microscopic analysis and a positive correlation between the numbers of *mcyE* gene copies and microcystin concentrations was observed. Furthermore, potential cylindrospermopsin and saxitoxin producers were detected by qPCR, showing the potential of molecular techniques to complement chemical and microscopic analysis in freshwater monitoring programs. Zervou et al. [9] reported the results of a 3-year monitoring of Lake Vegoritis in Northwestern Greece, which is used for irrigation, fishing and recreational activities. LC-MS/MS analysis showed the co-occurrence of cyanotoxins (seven microcystin congeners and cylindrspermopsin, at low levels, <1 μg/L) with other cyanobacterial peptides (anabaenopeptins, microginins). An investigation for the presence of cyanotoxins in Lake Karaoun, the largest artificial lake in Lebanon that serves multiple purposes, was conducted by Hammoud et al. [10], using complementary analytical techniques (LC-MS/MS, qPCR, ELISA and in vitro bioassays). A total of 11 microcystin congeners were detected in concentrations up to 211 and 199 μg/L for MC-LR and MC-YR, respectively. In addition, typical volatile and odorous cyanobacteria compounds were detected by GC-MS. Using a polyphasic approach, Ballot et al. [11] characterized cyanobacterial strains isolated from Meiktila Lake, a shallow reservoir close to Meiktila city in central Myanmar. The strains were classified morphologically and phylogenetically as *R. raciborskii*, and *Microcystis* spp. Cylindrospermopsins were detected by ELISA and LC–MS in 3 of the 5 *Raphidiopsis* strains, while *Microcystis* strains produced a wide range of microcystins, including 22 previously unreported congeners. Zastepa et al. [12] characterized nearshore deep chlorophyl layers from two embayments of Lake Huron, Canada. These layers were shown to be dominated by *Planktothrix* cf. *isothrix.* Microcystins, anabaenopeptins and cyanopeptolins were detected through the water column, along with the corresponding genes. The results also indicated that intersecting gradients of light and nutrient-enriched hypoxic hypolimnia are key factors in supporting deep chlorophyl layers in these embayments.

A serious incident involving dog deaths in Mandichosee, a mesotrophic reservoir of the River Lech, Germany, was investigated by Bauer et al. [13]. Anatoxin-a and dihydroanatoxina (dhATX) from benthic *Tychonema* sp. were detected by LC-MS/MS in the stomachs of two dogs in concentrations up to 1207 μg/L, while up to 68,000 μg/L anatoxins were present in lake samples containing large amounts of mat material. The findings of this study are extremely important as they underscore the role of less-studied benthic cyanobacteria in the production of potent toxins, such as the neurotoxic anatoxins.

Several studies reported results on new or less common cyanobacteria metabolites and cyanotoxin producers. Kust et al. [14] applied a molecular networking and dereplication approach in high-resolution mass spectrometry data using the open global natural product social networking (GNPS) web platform to putatively identify a wide range of cyanopeptides from eutrophic fishponds in the Czech Republic. Forty peptides belonging to the groups of anabaenopeptins, microcystins, cyanopeptolins, microginins, cyanobactins, radiosumins, planktocyclins and epidolastatins were identified. Zervou et al. [15] reported the occurrence and structural variety of anabaenopeptins in cyanobacterial blooms and cultured strains from Greek freshwaters using LC-MS/MS. Thirteen structures of anabaenopeptins were annotated based on interpretation of fragmentation spectra, including three structures not reported before. Cordeiro et al. [16] screened 157 strains from the Azorean Bank of Algae and Cyanobacteria (BACA) for cyanotoxin production (microcystins, saxitoxins and cylindrospermopsins) using qPCR, LC-MS/MS and 16S rRNA phylogenetic analysis. Cyanotoxin-producing genes were amplified in 13 strains, and 4 were confirmed as toxin producers by LC-MS/MS. Two nostocalean strains, possibly belonging to a new genus, were identified as new cylindrospermopsin producers, as they were positive for *cyrB* and *cyrC* genes and the presence of cylindrospermopsin was further confirmed by LC-MS/MS.

Two papers reported effects of nutrient and climate factors on the proliferation of cyanobacteria and the production of cyanotoxins. Barnard et al. [17] investigated the role of phosphorus and nitrogen limitation on microcystin and anatoxin production from *Microcystis* spp. and *Planktothrix* spp. in Western Lake Erie. The results showed the

importance of reducing both nitrogen and phosphorus to limit cyanotoxin and cyanobacterial biomass production. Le Moal et al. [18] analyzed 13 years of eutrophication and climatic data of Lac au Duc, one of the largest shallow water bodies in Brittany, Western France, which is used as recreational and drinking water reservoir. Analysis showed interannual variability of cyanobacterial composition, with dominant species shifting from *Planktothrix agardhii* towards *Microcystis* sp. and then *Dolichospermum* sp. due to climatic pressures and nitrogen limitation.

Paleolimnological studies based on analysis of sediment cores for cyanobacteria and cyanotoxins, can contribute historic data on the prevalence of toxic cyanobacterial blooms. Weisbrod et al. [19] explored the spatial variability and historical cyanobacterial composition in sediment cores from Lake Rotorua in the South Island of New Zealand, focusing on the abundance of *Microcystis*, *mcyE* gene copy numbers and microcystins. The results showed that toxin producing *Microcystis* blooms are a relatively recent phenomenon in Lake Rotorua, initiated after the 1950s. In addition, results indicated that a single sediment core sampling used by most paleolimnological studies in small to medium-sized lakes can capture dominant microbial communities.

The Special Issue includes three review papers that present emerging areas of toxic cyanobacteria and cyanotoxins research. Metcalf and Codd [20] reviewed and discussed cases were cyanobacteria and cyanotoxins co-occurred with additional hazards such as algal toxins, microbial pathogens, metals, pesticides and microplastics. The authors discussed challenges in assessment of toxicity in such cases and identified further research needs in this field. Sundaravadivelu et al. [21] reviewed the current methodologies for the analysis of freshwater cyanotoxins and prymnesins with emphasis in samples other than water. The authors discussed their limitations, especially with respect to accurate quantitation and structural confirmation of various cyanotoxins, where mass spectrometric techniques are advantageous as they can potentially be applied for detection and unambiguous identification of multiple toxins. Lastly, Monteiro et al. [22] reviewed the existing knowledge on the less-studied, structurally diverse cyclic hexapeptides anabaenopeptins that are increasingly detected in freshwaters in elevated concentrations and possibly play important roles in aquatic ecosystems.

**Funding:** This research received no external funding.

**Acknowledgments:** The editors are grateful to all the authors who submitted and contributed their work to this Special Issue. Special thanks to the expert peer reviewers for the rigorous evaluations of all submitted manuscripts.

**Conflicts of Interest:** The authors declare no conflict of interest.

### **References**


### *Article* **A Summer of Cyanobacterial Blooms in Belgian Waterbodies: Microcystin Quantification and Molecular Characterizations**

**Wannes Hugo R. Van Hassel 1,2,\*, Mirjana Andjelkovic 3, Benoit Durieu 2, Viviana Almanza Marroquin 4, Julien Masquelier 1, Bart Huybrechts 1,† and Annick Wilmotte <sup>2</sup>**


**Abstract:** In the context of increasing occurrences of toxic cyanobacterial blooms worldwide, their monitoring in Belgium is currently performed by regional environmental agencies (in two of three regions) using different protocols and is restricted to some selected recreational ponds and lakes. Therefore, a global assessment based on the comparison of existing datasets is not possible. For this study, 79 water samples from a monitoring of five lakes in Wallonia and occasional blooms in Flanders and Brussels, including a canal, were analyzed. A Liquid Chromatography with tandem mass spectrometry (LC-MS/MS) method allowed to detect and quantify eight microcystin congeners. The *mcyE* gene was detected using PCR, while dominant cyanobacterial species were identified using 16S RNA amplification and direct sequencing. The cyanobacterial diversity for two water samples was characterized with amplicon sequencing. Microcystins were detected above limit of quantification (LOQ) in 68 water samples, and the World Health Organization (WHO) recommended guideline value for microcystins in recreational water (24 μg L−1) was surpassed in 18 samples. The microcystin concentrations ranged from 0.11 μg L−<sup>1</sup> to 2798.81 μg L−<sup>1</sup> total microcystin. For 45 samples, the dominance of the genera *Microcystis* sp., *Dolichospermum* sp., *Aphanizomenon* sp., *Cyanobium/Synechococcus* sp., *Planktothrix* sp., *Romeria* sp., *Cyanodictyon* sp., and *Phormidium* sp. was shown. Moreover, the *mcyE* gene was detected in 75.71% of all the water samples.

**Keywords:** planktonic cyanobacteria; microcystin; blooms; monitoring; analysis; mass spectrometry; Liquid Chromatography with tandem mass spectrometry (LC-MS/MS)

**Key Contribution:** First assessment and comparison of microcystin content; the contribution of microcystin congeners; toxin-producing potential; and dominant genera with standardized methods in water samples from monitoring of five lakes and occasional sampling in the three Belgian regions.

### **1. Introduction**

A Belgian global picture of the diversity of cyanotoxins and the taxa producing them is currently lacking. This impairs a better understanding of their importance, their yearly variations and the need to prevent and mitigate blooms. Drinking water in Belgium is mostly provided by aquifers and barely depends on reservoirs [1,2]. Moreover, the water supply varies among the regions (Brussels regions, Flanders, and Wallonia). In Flanders, besides aquifers, the Meuse river and the Albert canal, only eight reservoirs supply water. If this is not sufficient, water is imported from Wallonia and neighboring

**Citation:** Van Hassel, W.H.R.; Andjelkovic, M.; Durieu, B.; Marroquin, V.A.; Masquelier, J.; Huybrechts, B.; Wilmotte, A. A Summer of Cyanobacterial Blooms in Belgian Waterbodies: Microcystin Quantification and Molecular Characterizations. *Toxins* **2022**, *14*, 61. https://doi.org/10.3390/toxins 14010061

Received: 19 November 2021 Accepted: 12 January 2022 Published: 16 January 2022

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2022 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

countries [2]. The drinking water in Brussels originates mainly (>90%) from Wallonia [3]. Water exploitation in Wallonia is dependent on ground water for 80% and is supplemented with water exploitation from the Meuse, some old mining sites and six dams [1,4,5]. Further, recreational waters are a sensitive issue, as there is an increasing societal demand for such areas in summer. External sources of eutrophication, such as untreated sewage discharge or agriculture run off, in these waterbodies may promote cyanobacterial bloom formation [6,7]. However, little information about eutrophication sources is available for fresh waterbodies in Belgium.

Ingestion of cyanotoxin contaminated water has been shown to be detrimental to human and animal health [8–14]. Yet in Belgium, no causative link has so far been found between toxic blooms and associated symptoms in humans and animals, such as gastroenteritis, vomiting, liver damage or convulsions [15,16]. However, suspicious bird deaths have been reported, which coincided with a toxic *Microcystis* bloom [17].

Between 1994 and 2008, a few studies identified the morphological and toxin diversity in toxic cyanobacterial blooms in Belgian lakes and ponds [17–24]. The concentrations of total microcystin (MC) measured by Ultra High Performance Liquid Chromatography (UHPLC) reached 18 to 2651 μg g−<sup>1</sup> dry weight (DW), and the bloom samples contained up to six variants [20,22]. *Microcystis* was found to be the most dominant genus, followed by *Planktothrix*. These results prompted a large scale study (BelSPO project B-BLOOMS2) from 2007 to 2010 [23,24]. During this study, 89% and 83% of the 162 samples tested showed the presence of *mcyE* and *mcyA* genes, respectively. *Microcystis*, *Anabaena* (now taxonomically identified as *Dolichospermum*), *Aphanizomenon*, *Planktothrix, and Woronichinia* were the primary bloom-forming cyanobacterial taxa. Furthermore, a quantitative toxin analysis of the samples showed that the total congeners concentration varied from 0.120 μg L−<sup>1</sup> to 37500 μg L−<sup>1</sup> total microcystin, analyzed with high performance liquid chromatography photodiode array detection (HPLC-DAD) and ELISA, in parallel [23–25]. Based largely on the B-BLOOMS2 report, the public authorities started to take action by informing the citizens and including cyanobacterial blooms in monitoring studies. Presently, Flemish and Walloon environmental agencies perform limited monitoring of recreational ponds (where bathing or other activities involving water contact are allowed) but using different protocols. This approach precludes the possibility of obtaining a global overview of blooms in Belgium. Furthermore, the data is limited to a small number of waterbodies, as bathing in Belgian surface waters is very restricted. Our study will extend the data from the B-BLOOMS2 study by providing new toxin, molecular and cyanobacterial occurrence data for water samples after a 10-year hiatus, using a uniform protocol. Moreover, the MC quantification revealed the existence of new microcystin congeners.

MCs belong to the most common cyanotoxins group found worldwide and are produced by multiple taxa (e.g., *Microcystis aeruginosa*, *Planktothrix* sp., *Anabaena*/*Dolichospermum*, *Oscillatoria* and *Nostoc*) [26,27]. The MCs covalently bind the protein phosphatases 1 and 2A (PP1 and PP2A) in eukaryotes, inhibiting their functions and eventually causing cell death [28–30]. Upon ingestion by mammals, the congeners are primarily transported in the liver cells through specific organic anion transporting polypeptides (OATPS) [31–34], which results in a hepatotoxic effect, causing nausea, intestinal problems and liver damage [29,30,35–41]. These toxins can also effect other organs such as the lungs and kidneys [34,39,42]. Human exposure to the MCs through multiple routes have been described (e.g., drinking water, recreational exposure, cyanobacteria-based food supplements, contaminated crops, ... ) [14,16,39,41,43–49]. The most prevalent congener found in Europe is MC-LR, though it is rarely detected in isolation [21,27,50–55]. Therefore, MC-LR is also commonly used in toxicological assays [36,37,40,56,57]. However, other congeners have single or multiple modifications in their structure, such as different amino acids in positions two and/or four, or methylations at different positions [58–60]. The structurally different congeners interact differently with the OATPS, PP1 and PP2A, resulting in different toxicities [31,58,61–64]. The half maximal inhibitory concentration (IC50) for PP2a and lethal dose for half of the test population (LD50) in vivo for MC-RR are shown to be

lower than MC-LR [61,62,64]. More efficient uptake for MC-LF and MC-WR than MC-LR is suggested to correspond to higher in vivo toxicity, while the PP-inhibiting capabilities are comparable [31,34,61,62]. Although differences in toxicity between the congeners are known, no uniform toxicity equivalency factors are available to adjust for the variation in their activity, as is the case for marine toxins [65,66]. An accurate risk assessment, when several congeners are present, is, therefore, difficult. Thus, the World Health Organization (WHO), Environmental Protection Agency (EPA) and other regulatory agencies use the sum of the concentrations for all MCs, described as MC-LR equivalent (MC-LR Equiv). The WHO published, in 1994, an initial tolerable daily intake (TDI) guideline of 0.04 μg kgbodyweight−<sup>1</sup> day<sup>−</sup>1, which translated to a concentration of 20 μg L−<sup>1</sup> in surface waters used for recreational activities [67]. In 2020, the WHO updated its provisional guideline value to 24 μg L−<sup>1</sup> [39]. The changed value results from a difference in calculations. The original value (20 μg L−1) was based on the proposed TDI, the body weight of an adult and the involuntary ingestion of 100 mL of water during swimming activities [68]. The new guideline value is calculated from the no observed adverse effect level (NOAEL) (40 μg kg−<sup>1</sup> bodyweight) [57], a ten times reduced uncertainty factor compared to the proposed TDI due to the short-term nature of recreational exposure, a volume of 250 mL of involuntarily ingested water, and taking into account the bodyweight of a child instead of an adult. The WHO guideline values are calculated to provide an adequate margin of safety [39]. The US E.P.A. also provided a new guideline for recreational waters in 2019. A reference dose (RfD) of 0.05 μg kg bodyweight−<sup>1</sup> day<sup>−</sup>1, the mean body weight of children between 6 and 10 years and an incidental ingestion factor were used to calculate 8 μg L−<sup>1</sup> as the recommended value [69].

Since the B-BLOOMS2 study was finalized a decade ago, no studies with standardized protocols have been performed to monitor cyanobacterial diversity and toxins in water samples from all over Belgium. For the first time in Belgium, we utilized Ultra High Performance Liquid Chromatography with tandem mass spectrometry (UHPLC-MS/MS) to identify and quantify the most frequent microcystin congeners in the water samples. By also detecting the genetic potential for synthesizing MCs and the dominant species in the samples, we tried to obtain more insights concerning the bloom characteristics in Belgium. Our cooperation with the three regional environmental agencies in Belgium achieved a sampling on a wide spatial scale of 79 water samples, covering 23 aquatic ecosystems. The species and toxin profiling by a standard set of analyses reveal the importance of monitoring in space and time. Furthermore, by targeting various waterbodies, we aim to identify the locations where monitoring would be needed because there is a health risk. Additionally, the data could help to design more effective prevention and mitigation measures.

### **2. Results**

### *2.1. Toxin Quantification*

In 86.08% (68/79) of the water samples, at least one of the quantified MCs (MC-RR, MC-LA, MC-LF, MC-LR, MC-LY, MC-LW, MC-YR, MC-WR) could be found above the limit of quantification (LOQ = 12.5 μg kg−<sup>1</sup> before correction for the total sample weight and volume). Moreover, 22.78% (18/79) of the samples contained toxin concentrations higher than 24 μg L−<sup>1</sup> total microcystin [39]. A complete overview of the results can be found in Table S1 in the Supplementary Materials. The concentration range was between 0.11 μg L−<sup>1</sup> and 2798.81 μg L−<sup>1</sup> total microcystin.

Furthermore, 82.61% (19/23) of the waterbodies contained, at least once during the summer, a quantifiable concentration of toxin (>LOQ), whereas concentrations higher than 24 μg L−<sup>1</sup> total microcystin were detected in 34.78% (8/23) of the waterbodies.

Interestingly, the canal samples (BV1, BV2, BV3) from Brussels contained congener concentrations higher than LOQ. One of the samples even reached 1831.32 μg L−<sup>1</sup> total microcystin. These results are the first reports on blooms in Belgium waterways.

Validation results for the UHPLC-MS/MS method used for analysis of the water samples can be found in Tables S5 and S6 in the Supplementary Materials.

### *2.2. Toxin Congener Diversity*

The detection frequency of MC-RR was the highest (84.81%), followed by the detection frequencies of MC-LR (81.01%) and MC-YR (50.63%).

MC-LR, MC-RR and MC-YR also contribute the most to the total MCs concentration in individual samples, compared to the other congeners. When comparing the proportions of the highest contributors in Belgium, the proportion of MC-RR is significantly higher than MC-LR, based on the Wilcoxon signed-ranked test (α < 0.05). Proportionally, MC-LR is the second-highest contributor, followed by MC-YR. The Wilcoxon test shows a significant difference (α < 0.05) between the proportional contributions of MC-YR compared to the other congeners (Figure 1a). Separate statistical analysis of samples containing a total MC concentration above or below the WHO guideline value for recreational water showed that the proportions of MC-YR are also significantly lower in relation to the proportions of MC-LR and MC-RR both above or below the 24 μg L−<sup>1</sup> total microcystin (Figure 1b,c).

**Figure 1.** (**a**) The distribution of the proportion of microcystin congeners (MCs) calculated at an individual sample level for all Belgian samples. (**b**) Samples with concentrations higher than 24 μg L−<sup>1</sup> total microcystin. (**c**) Samples with concentrations lower than 24 μg L−<sup>1</sup> total microcystin. (**d**) Proportions of MC-RR are compared in samples below and above the World Health Organization (WHO) guideline value. (**e**) Proportions of MC-LR are compared in samples below and above the WHO guideline value. (**f**) Proportions of MC-YR are compared in samples below and above the WHO guideline value. \* Proportion of MC is significantly different from MC-LR at α < 0.05 using the Wilcoxon test. \* Proportion of MC is significantly different from MC-YR at α < 0.05 using the Wilcoxon test. \* Proportion of MC is significantly different from the proportion of MC at concentration range > 24 μg L−<sup>1</sup> total microcystin with α < 0.05 using the Wilcoxon test.

When comparing samples with a total MC concentration above or below the guideline value, several observations can be made. There was a significant difference in MC-YR (Figure 1f) but no significant difference in the proportions of MC-LR and MC-RR (Figure 1d,e). Comparing the two concentration ranges for the proportions of MC-LA, MC-LY, MC-LF and MC-LW, a significant difference was shown using the Wilcoxon signedrank test (α < 0.05) (data not shown). A higher diversity of congeners contributed to the total MCs concentration when the concentration was above the WHO guideline value for recreational water (24 μg L−<sup>1</sup> total microcystin) (Figure 1b,c).

Additionally, the water samples were screened for six other MCs (MC-HtyR, dm-MC-LR, D-asp- MC-LR, dm-MC-RR, D-asp-Dhb-MC-RR and MC-HilR), which are also commonly detected in other studies [20,70,71]. These toxins were not included in the initially designed validation process. However, due to their prevalence and possible toxicity, they were screened. The congeners are identified based on molecular mass, production ions and elution time with the UHPLC-MS/MS method. However, dm-MC-LR and D-asp-MC-LR as well as dm-MC-RR and D-asp-Dhb-MC-RR could not be separated based on these parameters and are reported together (Table 1). We further establish limits of detection for the congeners, shown in Table 1. Overall, dm-MC-RR/D-asp-Dhb-MC-RR were the most abundant congeners in the water samples, followed by dm-MC-LR/D-asp- MC-LR, MC-HilR and MC-HytR, sequentially. An overview of their detection frequency can also be seen in Table 1, as well as their frequency related to the total quantified microcystin concentration in the samples. A complete overview of the results per sample can be found in Table S7 in the Supplementary Materials.

**Table 1.** Overview of precursor ion, product ions and limit of detection for not validated microcystin congeners. Additionally, the table also includes the detection frequency of the congeners in the analyzed samples at different total microcystin concentrations.


### *2.3. Molecular Analysis of Water Samples*

PCR amplification of the 16S rRNA fragment was attempted for 76 water samples. The fragment was successfully amplified for 45 samples. Some water samples were amplified twice (e.g., BL1.29, BV1.34, B04.29, I04.32), as can be seen in Table S3. Direct Sanger sequencing of the 16S rRNA fragment from the water samples resulted in 49 sequences of sufficient quality that could be analyzed with BLAST. They were of cyanobacterial origin, except for one plastid sequence in sample E04.32. The majority (42/49) of the analyzed 16S rRNA fragments showed 97% or higher similarities to sequences found in Genbank, as shown in Table S3. However, not all samples had a single dominant species, and 31 failed sequencings corresponded to mixtures of sequences that could not be analyzed. In four cases, the PCR with different primer pairs gave different dominant genera, and both are shown in Table S3. For all the samples analyzed with direct Sanger sequencing, *Microcystis* was the most dominant genus (12/76 or 15.79%), closely followed by *Dolichospermum* (11/76 or 14.47%). The third most abundant genus was *Aphanizomenon* (8/76 or 10.53%). Furthermore, the *Synechococcus* and *Planktothrix* genera were dominant in five and three samples, respectively. The *Cyanobium* genus was observed twice, while the *Cyanodictyon, Romeria* and, *Phormidium* genera were found once.

The amplicon sequencings by Illumina indicated a dominance of sequences from five OTUs belonging to the *Dolichospermum* genus for sample BL5.29 (71% of the reads), followed by *Microcystis* (20.5% of the reads) and a minor fraction of *Aphanizomenon* and *Cyanobium*/*Synechococcus*. This corresponds to the dominance of *Dolichospermum* inferred from the direct sanger sequencing. In contrast, sample VL1.36 was completely dominated

by sequences of *Microcystis* (99.1% of the reads) followed by 0.6% of the reads affiliated to *Dolichospermum*. However, the direct Sanger sequencing did not give any readable sequence to compare (Table S4).

The dominance of *Microcystis* in VL1.36 coincides with a high diversity (seven microcystin congeners) and a high concentration of total microcystin (128.93 μg L<sup>−</sup>1). In contrast, only two congeners and a lower total microcystin concentration (1.22 μg L<sup>−</sup>1) were found in the *Dolichospermum* dominated bloom BL5.29 (Table S1).

The *mcyE* gene amplification was tested in 70 water samples. The *mcyE* gene was detected together with MCs in 71.43% of the samples. Moreover, 4.29% of the samples contained the *mcyE* gene though the presence of MCs could not be detected by UHPLC-MS/MS method. However, the presence of the *mcyE* gene only implies the potential to produce MCs. Microcystin will not be produced when *mcyE* or other genes of the *mcy* gene cluster are lacking, are silenced or contain mutations.

### *2.4. Cyanobacteria Dominance at Different MCs Concentrations*

In the water samples, there was a dominance of the genera *Microcystis*, *Dolichospermum*, *Cyanobium/Synechococcus*, *Aphanizomenon* and *Planktothrix*. Sequences affiliated with *Cyanodictyon*, *Romeria*, and *Phormidium* were each observed in one of the samples. In samples with quantified MCs below the WHO guideline value, *Aphanizomenon*, *Dolichospermum*, *Microcystis*, *Synechococcus*/*Cyanobium*, *Planktothrix* and a plastid were identified as dominant (Figure 2a). For water samples that contained MCs concentrations above the WHO guideline value for recreational waters, primarily *Microcystis* or *Dolichospermum* could be identified as dominant species, based on the direct Sanger sequencing (Figure 2b). However, in both cases, it was not always possible to determine the dominant taxon, as the direct sequencing was not successful. When MCs were present in samples where *Dolichospermum* was dominant, concentrations ranged between 0.67 to 2420.91 μg L−<sup>1</sup> total microcystin, while for samples with *Microcystis* as dominant species, concentrations ranged from 1.07 to 2798.81 μg L−<sup>1</sup> total microcystin. Samples, dominated by *Aphanizomemon*, contained concentrations between 0.11 and 4.35 μg L−<sup>1</sup> total microcystin. In two water samples from lake H02, where *Planktotrix* was dominant, total microcystin concentrations were 4.05 and 2.80 μg L<sup>−</sup>1. In contrast, a concentration of 250.35 μg L−<sup>1</sup> total microcystin was quantified in a water sample from lake I04 containing *Planktothrix*. However, the origin of the MCs can be debated as a week earlier, *Microcystis* was abundant in lake I04, and the bloom could have been in decline, as further shown in Figure 3a. Our results cannot support definitive conclusions that would link high toxin concentrations to a specific cyanobacterial taxon. However, higher total microcystin concentrations are observed when *Microcystis* or *Dolichospermum* are the dominant taxa.

divided based on the WHO guideline value for recreational ponds (24 μg L−<sup>1</sup> total microcystin (MC)). The "n.d." abbreviation refers to not exploitable 16S rRNA sequences. (**a**) Species distribution for samples containing total MCs concentration below the WHO guideline value. (**b**) Species distribution for samples containing total MCs concentration above the WHO guideline value.

**Figure 3.** (**a**) Evolution of total microcystin concentrations in Lake I04 (lac de Bambois, Fosses-La-Ville) during the summer of 2019. Dominant genera detected in the samples are also indicated. (**b**) Evolution of total microcystin concentrations in Lake E04 (Grand large, Mons) during the summer of 2019. Dominant genera detected in the samples are also indicated.

### *2.5. Monitoring of Walloon Recreational Lakes*

The weekly sampling of the Walloon lakes provided an opportunity to look at the evolution of the toxin concentration and dominant species with time. The samples from lake I01 (Falemprise) showed a total microcystin concentration higher than 1 μg L−<sup>1</sup> (I01.31) only once, and when the direct Sanger sequencing was possible, the dominant cyanobacteria belonged to the unicellular *Synechococcus*/*Cyanobium* (Table S1). This lake was designated as a reference recreational lake during the B-BLOOMS2 study and has been regularly monitored since then. In lake B04 (Renipont plage), the concentration of MCs only rose slightly above 1 μg L−<sup>1</sup> in three instances when the potentially toxic cyanobacteria genera, *Aphanizomenon* and *Planktothrix* were found (Table S1). MC concentrations in the samples from H02 (Saint Léger sport complex) never reached the WHO guideline for recreational use but were slightly increasing over the summer and peaked at 4.35 μg L−<sup>1</sup> at the end of August, coinciding with the presence of the potentially toxic *Aphanizomemon* genus. However, the values decreased in the following weeks (Table S1). The two lakes where the WHO guideline value was exceeded were E04 (Grand Large, Mons) and I04 (Lac de Bambois). There was an increase in MCs over the summer, ending with a decrease in September in these samples. However, in the latter lake, the MC values were much higher, reaching 250.35 μg L<sup>−</sup>1, and the decrease was more gradual. For both lakes, *Microcystis* sp. was prevalent in the samples just before the MC increases (Figure 3a,b). However, in lake

I04, a higher diversity of potentially toxic genera was detected by direct Sanger sequencing, *Aphanizomenon*, *Dolichospermum* and *Planktothrix* (Figure 3a).

### **3. Discussion**

For the first time since the B-BLOOOMS2 study, microcystin congeners in Belgian surface waters have been reliably quantified by a standardized, state of the art analytical method in all three Belgian regions. Moreover, unmonitored waterbodies in the Brussels region were included in the study and provided additional proof of toxic blooms.

True monitoring data was achieved in Wallonia due to the weekly sampling of five recreational lakes during the summer of 2019. Three recreational waters (H02, I01, B04) showed MC values lower than the WHO guideline. Falemprise (I01) and Bambois (I04) had already served as a reference recreational lake with regular monitoring and a sporadically analyzed lake, respectively, during the B-BLOOMS2 study. Falemprise had shown quite a variable planktonic diversity, with a regular presence of potentially toxic taxa, the *mcy* genes and total microcystin concentrations ranging from 0.12 to 6.11 μg L<sup>−</sup>1. In Bambois, *Anabaena* (now *Dolichospermum*) formed blooms with MCs concentrations lower than 2.6 μg L−<sup>1</sup> [23]. In the present study, both lakes contained MCs, showing a persistent problem. Lake Bambois presented the highest values for the five Walloon lakes and exceeded the WHO guideline value for recreational use between the end of July to the middle of September (weeks 31 to 37). These results seem to indicate an increase in toxin concentration during the bloom events. *Aphanizomenon* and *Dolichospermum* were observed before the start of the MCs peaks.

Globally, 34.78% of the studied waterbodies contained total microcystin concentrations higher than the suggested guideline for recreational waters. The highest values were observed in waterbodies in Flanders. These samples also contained the highest fresh weights of filtered biomass (20.4 10−<sup>6</sup> g L−<sup>1</sup> average) compared to Brussels (4.6 10−<sup>6</sup> g L−<sup>1</sup> average) and Wallonia (2.3 10−<sup>6</sup> g L−<sup>1</sup> average). However, high concentrations of total microcystin could not always be linked to a high fresh weight of the biomass, as shown in Table S1. The total microcystin range (0.11 μg L−<sup>1</sup> to 2798.81 μg L<sup>−</sup>1) found in 68 of the 78 analyzed water samples is similar to the one found a decade ago during the B-BLOOMS2 study (0.120 μg L−<sup>1</sup> to 37,500 μg L−<sup>1</sup> total microcystin) by HPLC or ELISA [23]. The specific screening can explain the difference in certain MCs, resulting in underestimating the actual concentration, as the screening method only includes a selection of congeners and might miss certain ones. The ELISA assays, performed during the B-BLOOMS2 study, have the advantage of detecting all the possible hydrophobic adda groups specific for MCs (Figure S1) in the samples. In contrast, our triple quadrupole MS/MS method is very specific and will detect only the targeted toxins selected a priori in the detection method. This type of method is not well suited for a full scan analysis of samples. Worldwide, total microcystin concentrations can reach a maximum of 42.7 mg L<sup>−</sup>1, although concentrations are usually lower [51–54,70,72–80]. For instance, during a sampling campaign in the United Kingdom, only 18% of the samples contained a concentration above 20 μg L−<sup>1</sup> total microcystin [51].

In terms of congener diversity, MC-RR (84.81%), MC-LR (81.01%), and MC-YR (50.63%) were most prevalent in our study. Earlier, MC-LR was found to be the dominant congener (in 64% of analyzed water samples) in Wallonia and Luxemburg by Willame et al. [20]. MC-YR was the second most common congener, while MC-RR was not found [20]. Other reports confirm the presence of these most abundant congeners (MC-LR, MC-RR and MC-YR) in western Europe [26,51,53,54,70,71,81–83].

Moreover, five MCs (MC-LA, MC-LF, MC-WR, MC-LW and MC-LY) were also quantified in lower concentrations, while six other MCs (MC-HytR, dm-MC-LR, D-asp- MC-LR, dm-MC-RR, D-asp-Dhb-MC-RR and MC-HilR) were only screened and not quantified during our study. For the five quantified MCs, only a minor contribution to the total microcystins concentration at different concentration levels was shown. The MCs that were not quantified were primarily present in samples with total microcystin concentrations

above the WHO guideline limit (24 μg L−1). The lower contribution or presence of these eleven congeners at higher total microcystin concentrations are consistent in the samples from independent waterbodies and suggest that the biosynthesis dynamics of these toxins is different compared to the more abundant congeners (MC-LR, MC-RR, MC-YR). However, shifts in environmental factors, cyanobacterial strain diversity, growth phase and toxin production dynamics could influence the overall toxin diversity and quota in the blooms [84].

Overall, the minor contribution or presence of these eleven MCs (quantified or not) to the total MC concentration might be of minor importance for the degree of health risk compared to the more abundant MCs (MC-LR, MC-RR and MC-YR). In most cases, the sum of the MC-LR, MC-RR and MC-YR concentrations already exceeded the new WHO guidelines for microcystin, without the contributions of the other congeners (Table S1) [39]. However, different geographical origins, species and environmental factors might influence which congeners are more abundant. Analysis of the most common congeners during monitoring is therefore still advisable [20,53–55,71,76,77,85,86].

Incorporation of the toxic demethylated congeners in the quantification methodology would be similarly advisable in the future [16].

Besides toxins, *mcyE* was detected in almost 76% of the water samples. Comparing these results is difficult due to differences in the monitoring frequency and amplification method. For instance, the B-BLOOMS2 study found *mcyE* in 89% of their samples, but the number of samples was larger due to more regular monitoring in surface waters prone to bloom events and the addition of occasional bloom samples [23]. Similarly, Moreira et al. (2020) also found variations in *mcyE* presence in various lakes in Portugal. During their sampling from April to September, the *mcyE* detection rate varied from 33.3% to 83.3% for different lakes, but they used a different amplification method than our study [80]. A good correlation between toxin content and the *mcyE* compared to other genes has been shown in the multiple studies reviewed in Pacheco et al. [87]. The authors also pointed out that correlations between *mcyE* copy number and toxin concentration are still controversial. Nonetheless, this data supports the need for our approach of assessing the potential of toxin production by the detection of the *mcyE*.

With the direct Sanger sequencing approach, only the most abundant species would be detected if there was a strong dominance in the sample. Indeed, if the cyanobacteria populations were too heterogeneous, the direct sequencing would fail because the resulting chromatograms would not be interpretable. Therefore, only 45 out of 79 samples produced usable data due to species heterogeneity. Nevertheless, direct Sanger sequencing showed that *Microcystis*, *Dolichospermum*/*Anabaena*, *Synechococcus*/*Cyanobium*, *Planktothrix* and *Aphanizomenon* were generally observed in the samples, as earlier described in the B-BLOOMS2 study. However, *Woronichinia*, also present during B-BLOOMS2, was not detected. Our sampling season, which focused on the warmer months of the year, might have missed the occurrence of *Woronichinia*, which is more prevalent at temperatures below 15 ◦C [88]. Yet, they might also be present in the more heterogeneous samples but could not be detected with our current methods. Furthermore, the *mcyE* gene was found in samples containing *Cyanobium*, which raises the question of the possible toxicity of this genus, as discussed during B-BLOOMS [23]. Little information is available about the toxicity of picocyanobacteria in eutrophic waters. *Cyanobium* have previously been identified in freshwater ecosystems [89,90], although their toxicity has been put in question. Multiple studies showed low amounts of MCs being produced [15,90–92], while the sequenced genomes of *Cyanobium* strains seem to lack a complete NRPS/PKS gene clusters [93].

Amplicon sequencing by a high-throughput sequencing technique is an alternative to direct Sanger sequencing that allows the identification of species in complex populations, allowing the study of heterogeneous samples. However, amplicon sequencing protocol is more expensive and time-consuming. Thus, it was tested for in two water samples, of which one (BL5.29) was used for both sequencing types. This comparison shows that the

dominant taxon was quasi-identical with both methods. Indeed, the Sanger sequence was 99.8% similar to the most similar hit for the dominant OTU72 (*A. ellipsoides* Ana HB).

The samples from waterbodies were either collected through one of the regional monitoring programs or following a sporadic bloom notice by the regional environmental agencies in an unmonitored waterbody. The samples from Wallonia and some of the samples from Flanders were obtained from already established monitoring programs for recreational waterbodies, whereas the samples from ponds in Brussels were obtained specifically for the study when a bloom notice was received. In the cases with monitoring programs, the situation is followed by the public authorities and the recreational waterbodies in both regions get regularly closed if (toxic) blooms are observed. In the not monitored ponds in the Brussels region, we found that 66.67% and 16.67% of the samples were above LOQ and 24 μg L−<sup>1</sup> total microcystin, respectively [67,68]. By comparison, the total MC concentration rose above LOQ and 24 μg L−<sup>1</sup> total microcystin in 86.08% and 22.78% of the Belgian waterbodies, respectively, at one point in time. Even though the numbers seem to be lower in Brussels compared to the rest of the country, this is artificial. In Brussels, only a small number of waterbodies was sampled. Furthermore, these results were obtained outside the standardized protocol and from only three or fewer time points per site. Therefore, we could have missed bloom peaks during the summer. As we reported, the canal in Brussels also contained a total of 1831.32 μg L−<sup>1</sup> total microcystin at one sampling spot. Currently, there is a public dynamic in the region to make more waterbodies available for recreational use during the summer. Moreover, the unauthorized use of waterbodies as bathing areas was also common practice during hot summers. Without proper monitoring, this could create a public health risk for humans and domestic animals [9]. In addition, other usages of this water need to be considered. For instance, pond water could be used to irrigate urban vegetable gardens, and water from canals could be used in agriculture. Several research groups have already shown that plants could accumulate MCs when irrigated with contaminated water [45,47,94–96]. The connection between water usage and toxin accumulation in plants needs to be further investigated in the future.

Based on our results, it appears that monitoring of any potential bathing sites where unauthorized bathing occurs could be recommended. In the context of Belgium, this would be in the Brussels region. Monitoring could reveal links between public health issues and any potential hazard related to cyanobacterial blooms. The other two Belgian regions monitor their official bathing sites, using different sampling protocols and analysis techniques. Harmonizing the monitoring methods could provide insight into the species and toxin diversity in the blooms in Belgium. Moreover, it might help to uncover environmental drivers that promote the blooms. Techniques used during the monitoring could vary depending on the expertise and resources that are available. Cell counting and species identification are relatively low tech monitoring tools to determine the intensity of the blooms and quantify the potential toxin-producing species, but they are time-consuming and need taxonomic expertise. ELISA and *mcyE* amplification are fast, relatively cheap tests appropriate for screening MCs or MC-production potential, respectively, when a bloom is observed. However, these four techniques need to be supplemented with UHPLC-MS/MS to accurately determine MCs concentrations during and after the bloom to ensure public safety. When toxin equivalency factors become available for the different congeners, UHPLC-MS/MS approaches will also be crucial to accurately determine the risk.

More detailed information about the bloom incidences in a country can also benefit possible mitigation strategies in the future. Preventive strategies could be designed to reduce the influx of nutrients where this is feasible. This strategy requires information about the sources of eutrophication, which is still lacking for most fresh waterbodies in Belgium. During this study; the regional environmental agency listed agriculture, the discharge of purified sewage water, water influx by a canal and the feeding of fish during recreational fishing as potential sources of eutrophication in the Walloon lakes. For smaller ponds in Brussels, an overpopulation of waterfowl can cause nutrient loading due to an abundance of excrement. Another mitigation strategy is hydrogen peroxide treatments. This treatment can eliminate the bloom and MCs but needs to be optimized based on bloom density and the quantity of toxins. For toxin quantification and assistance to prevention, analytical methods, such as the one presented in this paper, would be suitable [97–105]. Approaches such as flock lock or flock sink techniques could be a viable solution to prevent bloom incidence by capturing phosphorous on the bottom of larger waterbodies with a greater depth. However, the sediments of shallow recreational lakes might be too frequently disturbed for this approach to be effective [106–109]. External phosphorous loading in the waterbodies after treatment (e.g., sewage disposal, floods, ... ) will also undermine the effectiveness of these treatments. To ensure public safety, monitoring of waterbodies will always be necessary and toxin quantification should be included as a part of the monitoring techniques.

### **4. Conclusions**

During this study, we validated and used a UHPLC-MS/MS method for the first time to analyze Belgian water samples from the three different regions with an identical method. The microcystin concentrations found clearly illustrate a persistent problem of toxic blooms throughout Belgium with a potential health impact. The three most abundant MCs (MC-RR, MC-LR, MC-YR) contributed the most to the total microcystin concentrations. Our fast and efficient method can be applied to monitoring programs in Belgium and other parts of the world. PCR amplification of the *mcyE* gene linked its occurrence to the toxin presence for 71.43% of the water samples. Moreover, the dominant blooming taxa were also determined in a number of samples. Interestingly, this study also characterized a cyanobacterial bloom in a Belgian canal for the first time. The abundance of water samples that contained MCs shows the need to enlarge the sampling of waterbodies where there could be a risk of human exposure and include them in existing or new monitoring programs.

### **5. Materials and Methods**

UPLC/MS grade solvents (Biosolve B.V., Valkenswaard, The Netherlands) were used for extraction or basis for the mobile phase. The MCs standards were ordered as a solid powder from Enzo Life Sciences (Antwerp, Belgium)®, except for D-asp-Dhb-MC-RR from Cyano Biotech GmbH (Berlin, Germany) and Dm-MC-RR from Novakits (Nantes, France). They were initially dissolved in 100% methanol and used to prepare mixed stock solutions in 50% methanol with 1% acetic acid. The dissolved cyanotoxin standards were kept at −20 ◦C. Whatman GF/C grade filters were obtained from Sigma Aldrich (Overijse, Belgium).

### *5.1. Sampling*

The sampling was performed from July until mid-September in 2019 at 23 different locations in the three Belgian regions: Wallonia (5 locations), Flanders (7 locations) and Brussels (11 locations). The sampling frequency was dependent on the region, type of waterbodies and access to the lakes (directly or via the environmental agency). Recreational waterbodies are defined as ponds and lakes where bathing is permitted. The water samples were either collected every week or only after a bloom was observed. Each sample was annotated by combining a three digit annotation of the sample site followed by a number giving the week of the year (e.g., XYZ.12). Names for the waterbodies can be found in Table S2 in Supplementary files.

In Wallonia, water samples were collected weekly in 5 recreational lakes (I01, I04, E04, B04 and H02), independently of the presence of a bloom following a standardized protocol. The environmental agency Institut Scientifique de Service Public (ISSEP) sampled the surface water with a 500 mL glass bottle at a fixed point. Samples were stored at 4 ◦C and later transported to Sciensano within three days of the collection for further processing.

In Flanders, water samples were taken from 3 recreational waterbodies (AN1, AN2 and AN3) by the environmental agency Vlaamse Milieu Maatschappij (VMM) only when a bloom was present. The samples were stored in plastic containers at 4 ◦C before further processing and analysis by Sciensano. Four ponds (GH1, VL1, VL2 and VL3) that were not used for recreation were sampled by Sciensano. GH1 is a sedimentation pond for wastewater, while VL1, VL2 and VL3 are shallow ponds in parks where fishing is allowed. The surface water was sampled with 500 mL sterilized glass bottles. They were processed the same day. In this case, public media had indicated the presence of the blooms. These waterbodies were only sampled a second time by Sciensano if a bloom was still present.

In Brussels, we performed samplings in ponds where bathing is not allowed and thus are not considered as recreational waterbodies. Each waterbody was initially sampled after a bloom notification by the regional environmental agency Environment.Brussels or Port.Brussels. The latter manages the port estate in the Brussels capital region. In total, 8 ponds (BL1-8) and 3 locations in the Brussels canal (BV1-3) were sampled. Each spot was sampled at least a second time independent of bloom presence, except for two spots in the canal. Samples were taken from the surface water with 500 mL sterilized glass bottles. They were processed the same day. An overview of all the sampling sites is shown in Figure 4. An overview of the sample sites with waterbody type can be found Supplementary Material (Table S2).

**Figure 4.** Map of Belgium showing the sample sites. The first three letters of the sample names are used as abbreviation. In Flanders, 7 sites were sampled (AN1-3, VL1-3 and GH1). In Wallonia, 5 recreational lakes were sampled (I01, I04, E04, B04 and H02). For clarity, the Brussels region is enlarged. Here 8 ponds were sampled (BL1-8), as well as the Brussels canal at 3 different sites (BV1-3). Place names for the waterbodies and their type can be found in Table S2 in Supplementary files.

In general, 150 mL of the sample was filtered on a GF/C Whatman® filter under vacuum to collect the biomass. Lower volumes were filtered due to clogging when dealing with high bloom density. The sample filters were weighed before and after filtration to determine the weight of the wet biomass. The sample filters were stored at −20 ◦C before analysis. Filtration was performed in duplicate. One filter was used for the quantification of the MCs, while the other was used for the molecular work. The filtrates were collected and stored at −20 ◦C to determine extracellular toxin concentration.

### *5.2. Quantitative Analysis of Microcystin Congeners*

### 5.2.1. Intracellular and Extracellular Microcystin Extraction

Only the most common toxins in Europe (MCs) were selected for our quantification method. Earlier studies in Belgium suggested that these toxins are the most prevalent public health threat [20,23]. To properly validate the method, only commercially available MCs were selected.

The method used for analysis was validated in house. Results of the validation can be found in Tables S5 and S6 in the Supplementary Materials. The filters, containing biomass, used for toxin extraction underwent a freeze-thaw step and liquid extraction. When the filters were initially stored at −20 ◦C, they only need to be defrosted. The filters were cut in half and weighted. For the liquid extraction, 4.5 mL 80% methanol was added together with the filter in 50 mL plastic tubes. Solvent and biomass contact was increased by regularly mixing during 1 h. The samples were centrifuged for 10 min at 3900 rpm.

The extract was filtered through a Phenomenex 0.2 μm RC syringe filter (Utrecht, The Netherlands) to remove debris. Samples were stored in a 15 mL plastic tube at −20 ◦C. Samples with high concentrations of the MCs were diluted after the initial analysis to fit within the range of the calibration curve. The calibration curve was made in a blank matrix.

The sample filtrates (extracellular fraction) were also purified using a Phenomenex 0.2 μm RC syringe filter and analyzed separately through direct injection of 10 μL the Xevo TQ-S, similar to Turner et al., 2018 [51].

### 5.2.2. Detection and Quantification of Cyanotoxins

The detection and quantification parameters were identical for intra- and extracellular toxins analysis. A Waters Acquity UPLC H-class (Eten-Leur, The Netherlands) connected to a Waters XEVO TQ-S was used for the detection of the cyanotoxins. A 1.7 μm, 2.1 mm × 100 mm Waters Acquity BEH C18 column fitted with a Waters Acquity BEH C18 1.7 μM VANGUARD PRE-Col separated the toxins under the influence of a gradient elution program. The fraction of acetonitrile (B) in the eluent changed as followed: 0 min, 2% B; 1.00 min, 40% B; 7.00 min, 55% B; 7.20 min, 98% B; 8.00 min, 98% B; 9.00 min; 2% B; 12 min, 2% B. Both Organic and water phases were supplemented with 0.025% formic acid. The flow rate was 0.5 mL min−1. The column was heated to 60 ◦C, and 10 μL of sample was injected.

Multiple reaction monitoring (MRM) was then used to detect the toxins by selectively quantifying compounds within complex samples. The triple quadrupole MS initially targeted the ions corresponding to the toxins of interest, referred to as the "precursor ion". Two product ions from the collision induced fragmentation were selected. One was used for quantification of the cyanotoxin, the other as a qualifier. The MS parameters were set according to the literature data and optimized to the instrument setting (Table 2).



After quantification, the concentration for each cyanotoxin was recalculated to μg L<sup>−</sup>1, corrected with the mass of the original mass of the filter and the filtered volume of the sample. The concentration of each MC in the filtrate (extracellular fraction) was added to

the final concentration of the MC extracted from the biomass. Thereafter, the sum of all the congeners was calculated to provide a μg L−<sup>1</sup> total microcystin value.

Congener proportions to the total MCs concentration were calculated in each sample. The differences in proportions for the separate congener were then compared for Belgium using the Wilcoxon test at α < 0.05. Additionally, the same statistical test was used to compare the difference in congener proportions for samples containing MCs concentration higher and lower than 24 μg L−<sup>1</sup> total microcystin, separately. The samples without MCs were excluded. Proportions of MC-LR, MC-RR and MC-YR were also compared between the two concentration ranges.

### *5.3. Molecular Analysis of the 16S rRNA and the mcyE Gene*

### 5.3.1. DNA Extraction

First, 0.8 mL lysis buffer (40 mM EDTA 5, 50nM Tris-HCl, 0.75 M sucrose) was added to each sample filter (containing biomass), and a bead-beating step (at 30 m s−<sup>1</sup> for 30 s) was performed. Then, a lysozyme (Sigma-Aldrich, St. Louis, MI, USA) (20 mg mL−1) digestion for 30 min at 37 ◦C was followed by a treatment with 22.22 mg mL−<sup>1</sup> proteinase K (Macherey-Nagel, Düren, Germany), supplemented with 80 μL SDS (100%), for 2 h at 55 ◦C. The lysate was transferred to a new Eppendorf tube. Subsequently, the filters were rinsed with 1 mL lysis buffer during a 10 min incubation at 55 ◦C. The second lysate was stored in another Eppendorf tube.

A Phenol-chloroform-isoamyl alcohol solution (25:24:1, pH 8) (VWR, Leuven, Belgium) was added in an equal volume to the extract volume (V:V) to both lysates. Next, the samples were centrifuged at 14,000× *g* for 15 min. The upper phase of each tube was transferred to a new Eppendorf tube, and chloroform-isoamyl alcohol (24:1, pH 8) was added V:V. The tubes were centrifuged again at 14,000× *g* for 15 min to collect the upper phase. For each sample, the two lysates were combined

Finally, the DNA was precipitated with 0.1 V:V of sodium acetate (3 M, pH 5.2) and 0.6 V:V of cold isopropanol. After centrifugation, the DNA was rinsed once with 300 μL icecold ethanol (Merck, Branchburg, NJ, USA) (100%) and once with 300 μL ice-cold ethanol (70%). The supernatant was removed, and the pellet was air dried. Finally, the DNA pellet was dissolved in 100 μL TE buffer (10mM Tris-HCl and 1mM EDTA, pH 8) and stored at −20 ◦C.

### 5.3.2. Gene Amplification of Partial rRNA and mcyE Gene Sequences

For the rRNA gene sequences, two protocols were tested. A long rRNA fragment was amplified with the cyano-specific primers 359F/23S30R [110] using the SuperTaq Plus© enzyme (HT Biotechnology, Cambridge, UK), buffer and dNTPs obtained from SpharoQ (NL). The amplification program was 95 ◦C—5 min, 35 times; 95 ◦C—30 s, 57 ◦C—45 s, 68 ◦C—1 min; followed by 69 ◦C—5 min, 16 ◦C—infinite. As a shorter PCR product could give a higher amplification efficiency, the primer pair 359F/781R [111] was tested later. However, the SuperTaq Plus© enzyme was no longer commercialized and was replaced by the Q5 High Fidelity polymerase (New England Biolabs, Ipswich, MA, USA) for the majority of the PCR reactions. The amplification program was: 98 ◦C—5 min, 35 times; 98 ◦C—30 s, 65 ◦C—45 s, 72 ◦C—1 min; 72 ◦C—5 min, 16 ◦C—infinite. The *mcyE* gene involved in the production of MCs was amplified with the primer pair mcyEF2/mcyER4 [112] using the SuperTaq Plus© enzyme. The amplification program was 94 ◦C—3 min, 30 times; 94 ◦C—30 s, 57 ◦C—45 s, 68 ◦C—1 min, followed by 68 ◦C—10 min and 16 ◦C infinite, as described in the final report of B-BLOOMS2 [23]. Amplifications were performed in a Thermal cycler T100 (Bio-Rad, Hercules, CA, USA). The presence of PCR products of the right size was visualized by electrophoresis on a 1.5% agarose gel during a 95 min run at 90 V.

### 5.3.3. Sanger Sequencing and Sequence Analysis

After PCR, the 16S rRNA amplicons were sent for Sanger sequencing with primers 359F, 781R or 23S30R at Giga Genomics (ULiege) [110,111]. Some sequences were of bad quality, which prohibited further analysis. These sequences probably resulted from a mixture of organisms without clear dominance by one taxon. The forward and reverse sequences were not obtained in all cases for each PCR product, and therefore, the individual sequences of a single strand were used for further analysis, admitting that some sequencing errors might be present but that the quality would be sufficient to determine the dominant genus. In three cases, the sequences obtained on different PCR products (short or longer ones) for the same sample were affiliated to different genera. The sequences used during the further analysis can be found in the supplementary files (Table S8).

The NCBI nucleotide BLAST (basic local alignment search tool) was used to identify the most closely related strain sequences for the 16S rRNA sequences, using individual sequences obtained by the different primers tested and the identification was based on this data, as shown in Table S3.

### 5.3.4. Amplicon Sequencing with the Illumina Technique

For samples BL5.29 and VL1.36, partial 16S rRNA gene sequences were obtained by PCR using the primer set CYA359F and CYA781Ra/CYA781Rb, which amplifies the V3-V4 region of the cyanobacterial 16S rRNA gene [111]. Primers were modified to include a 10-bp sample-specific barcode tag at the 5 end to allow samples to be multiplexed for sequencing. PCR reactions were performed in triplicates in order to minimize the influence of amplification biases. These were pooled to equivalent concentrations and purified using the NucleoSpin® Gel and PCR Clean-up kit (Macherey-Nagel, Düren, Germany). Purified samples were sent to Genewiz (South Plainfield, NJ, USA), where sequencing adapters were ligated to the amplicons and sequencing was performed using Illumina MiSeq (Illumina, San Diego, CA, USA) using 2 × 300 bp paired-end libraries. The bioinformatic analysis is adapted from a validated method by Pessi et al. and consists of processing raw reads to remove chimeric sequences, followed by the clustering into an operational taxonomic unit (OTU) [113]. Briefly, paired-end reads were merged, filtered and only reads containing both barcodes in the 3 and 5 ends were kept. Two and zero mismatches were allowed to the primer and barcode sequences, respectively, and reads with a maximum expected error of more than 0,5 and a length of less than 370 bp were removed. Singletons were removed, and remaining quality-filtered sequences were denoised to remove chimaeras and sequencing errors using unoise3 [114]. The denoised operational taxonomic units (ZOTUs) obtained were then clustered into OTUs at a 99% similarity threshold [115]. The representative sequence of an OTU is the most abundant unique sequence of each OTU cluster. Taxonomic classification was performed by extracting from Genbank the most closely related sequences of each OTU using BLAST.

**Supplementary Materials:** The following are available online at https://www.mdpi.com/article/10 .3390/toxins14010061/s1, Figure S1: Microcystin core structure with annotated adda group, Table S1: Overview of experimental data for the water samples, concentrations of MC congeners and total microcystin (μg L−1), presence of genes coding for 16S rRNA and *mcyE*. molecular identification, coverage and identity % of the most similar hit by BLAST and primer used for Sanger sequencing. The first three characters of the samples annotation indicate the sample location, while the last two numbers indicate the week of the year when the sample was collected. Samples are grouped by origin, Flanders. Brussels and Wallonia. N.A. indicates samples for which no PCR was performed. "-" represent samples were no visible bands were obtained after gel electrophoresis of the PCR products or sequencing results were off too pore a quality to provide a reliable result. "?" represent inconclusive PCR results where the band was visible but was too faint or not at the correct height. The asterisk \* indicates that the sequence was obtained on the basis of a PCR reaction produced with the primer pair 359F-781R, whereas the others sequences were obtained with 359F-32S30R. "i " denotes the samples that were analyzed using Illumina technology, Table S2: Overview of sampling sites in the different regions, water sample annotation, waterbody type and specific monitoring for

cyanobacteria, Table S3: Detailed information for taxonomic identification based on BLAST analysis. \* Samples for which PCR with different primers gave different dominant taxa, Table S4: Number of reads and BLAST analysis of the OTUs obtained with the Illumina amplicon sequencing, Table S5: Validation results for UHPLC-MS/MS method quantification method of 8 MCs and Nodularin in filtered cyanobacterial biomass. The different validation concentration levels Limit of detection (LOD) was set at lowest tested concertation where signal to noise ratio was higher than 3. Limit of Quantification (LOQ) was selected as lowest concentration for which the method was validated. Signal to noise ratio for LOQ should be above 20. Additionally, values for recovery, repeatability, reproducibility, measurement uncertainty (MU) and R2 of the linear curve are presented, Table S6: Results of the ion ratio for the validation of the UHPLC-MS/MS method quantification method of 8 MCs and Nodularin in filtered cyanobacterial biomass. Acceptation criteria are based on European Decision 2002/EC/657, Table S7: Detection results for six additional microcystin congeners with a limit of detection (LOD) at 0.1 μg L−1. Not detected is abbreviated by "n.d.". Detected MCs are annotated as > LOD, Table S8: Overview of the single sequences amplified by the Sanger method used for taxonomic identification based on BLAST analysis.

**Author Contributions:** Conceptualization, W.H.R.V.H., A.W., M.A. and B.H.; methodology, W.H.R.V.H., A.W. and B.H.; validation, W.H.R.V.H. and B.H.; formal analysis, W.H.R.V.H. and B.D.; investigation, W.H.R.V.H., A.W., B.D. and V.A.M.; resources, A.W., M.A., V.A.M. and J.M.; data curation, W.H.R.V.H.; writing—original draft preparation, W.H.R.V.H., A.W., B.D., B.H. and M.A.; writing—review and editing W.H.R.V.H., A.W., B.D., M.A., V.A.M. and J.M.; visualization, W.H.R.V.H.; supervision, A.W., M.A. and J.M.; project administration, W.H.R.V.H., A.W., M.A. and J.M.; funding acquisition, A.W. and M.A. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by the Federal Agency for the Safety of the Food Chain (FAVV-AFSCA-FASFC) in the framework of the work of WVH towards his PhD. FAVV-AFSCA-FASFC cannot be held liable for the use of the data nor the conclusions that could be drawn from their treatment. Annick Wilmotte is Senior Research Associate of the Belgian Funds for Scientific Research (FRS-FNRS).

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Conflicts of Interest:** The authors declare no conflict of interest.

### **References**


### *Article* **Anabaenopeptins from Cyanobacteria in Freshwater Bodies of Greece**

### **Sevasti-Kiriaki Zervou 1,\*, Triantafyllos Kaloudis 1, Spyros Gkelis 2, Anastasia Hiskia <sup>1</sup> and Hanna Mazur-Marzec <sup>3</sup>**


**Abstract:** Cyanobacteria are photosynthetic microorganisms that are able to produce a large number of secondary metabolites. In freshwaters, under favorable conditions, they can rapidly multiply, forming blooms, and can release their toxic/bioactive metabolites in water. Among them, anabaenopeptins (APs) are a less studied class of cyclic bioactive cyanopeptides. The occurrence and structural variety of APs in cyanobacterial blooms and cultured strains from Greek freshwaters were investigated. Cyanobacterial extracts were analyzed with LC–qTRAP MS/MS using information-dependent acquisition in enhanced ion product mode in order to obtain the fragmentation mass spectra of APs. Thirteen APs were detected, and their possible structures were annotated based on the elucidation of fragmentation spectra, including three novel ones. APs were present in the majority of bloom samples (91%) collected from nine Greek lakes during different time periods. A large variety of APs was observed, with up to eight congeners co-occurring in the same sample. AP F (87%), Oscillamide Y (87%) and AP B (65%) were the most frequently detected congeners. Thirty cyanobacterial strain cultures were also analyzed. APs were only detected in one strain (*Microcystis ichtyoblabe*). The results contribute to a better understanding of APs produced by freshwater cyanobacteria and expand the range of structurally characterized APs.

**Keywords:** anabaenopeptins; LC–qTRAP MS/MS; fragmentation spectra; structure elucidation; cyanopeptides; cyanobacterial metabolites; Greek freshwaters; cyanobacteria

**Key Contribution:** The first study of anabaenopeptins' occurrence and their structural variety in cyanobacterial blooms and isolated strains from freshwaters of Greece utilizing LC–qTRAP MS/MS. Possible structures for three novel anabaenopeptins are proposed, expanding the knowledge on this understudied class of cyanobacterial metabolites.

### **1. Introduction**

Anabaenopeptins (APs) are cyanobacterial metabolites with a cyclic peptide structure [1]. The presence of APs has been reported in freshwater [2–6] and marine cyanobacterial blooms [7–9], as well as in terrestrial environments, including the leaves of plants in a coastal forest [10] and the terrestrial mat in a bamboo forest [11]. APs are produced by freshwater, marine and terrestrial cyanobacteria [12] mainly belonging to the genera *Planktothrix* [13,14], *Anabaena* [15–18], *Microcystis* [19–23] and *Nostoc* [24,25]. They especially belong to the species *Planktothrix* (*Oscillatoria*) *agardhii* [3,26–32], *Planktothrix rubescens* [30,31,33,34], *Anabaena* (*Dolichospermum*) *flos-aquae* [35], *Anabaena lemmermannii* [36], *Microcystis aeruginosa* [3,37–39], *Microcystis flos-aquae* [36], *Microcystis ichthyoblabe*, *Microcystis wesenbergii* [19] and *Nostoc punctiforme*[40], as well as *Aphanizomenon flos-aquae*[41], *Nodularia spumigena* [40,42–45], *Woronichinia naegeliana* [46,47] and *Woronichinia compacta* [48]. Additionally, the cyanobacteria

**Citation:** Zervou, S.-K.; Kaloudis, T.; Gkelis, S.; Hiskia, A.; Mazur-Marzec, H. Anabaenopeptins from Cyanobacteria in Freshwater Bodies of Greece. *Toxins* **2021**, *14*, 4. https:// doi.org/10.3390/toxins14010004

Received: 12 October 2021 Accepted: 15 December 2021 Published: 21 December 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

*Lyngbya* sp. [49], *Lyngbya confervoides* [50], *Schizothrix* sp. [51], *Tychonema* sp. [52] and *Brasilonema* sp. [11] are known APs producers. APs are also produced by symbiotic cyanobacteria that have been isolated from the marine sponges *Theonella* sp. [53–55], *Theonella swinhoei* [56], *Psammocinia aff. bulbosa* [57], and from the lithistid family *Theonellidae* [58].

APs are hexapeptides with the general structure X1-CO-[Lys-X3-X4-MeX5-X6], characterized by the presence of the amino acid lysine (Lys), which contributes to ring formation by an *N*-peptide bond with the carboxy group of amino acid X6 and a C-peptide bond with the amino group of amino acid X3. A side chain of one amino acid unit is attached to a five-peptide ring via an ureido bond between the a-N of Lys and the a-N of the side chain amino acid [1,35] (Figure 1). Apart from Lys, all other amino acids are variable, while the amino acid in position X5 is usually *N*-methylated and the amino acid in position 4 is usually in homo form (Table S1).

The first characterized APs (i.e., Anabaenopeptin A and Anabaenopeptin B) were isolated in 1995 by Harada et al. from the freshwater cyanobacterial strain *Anabaena flos-aquae* NRC 525-17 [35], from which APs were named. Structural variants of APs are also known by other names such as Nodulapeptins [42], Oscillamides [27,30], Konbamide [54], Keramamides [53,55], Ferintoic acids [59], Mozamides [58], Schizopeptin [51], Brunsvicamides [52], Psymbamide A [57], Pompanopeptin B [50], Paltolides [56], Lyngbyaureidamides [49] and Nostamides [24,40]. These names were mainly based on the producing taxon or on the geographic location of discovery, complemented with suffixes describing the variety. As a consequence, nomenclature of this class is not fully systematic. Newly identified APs may also contain their molecular mass as part of their name for consistency; however, this approach may be problematic because several variants can have the same molecular mass due to the large number of possible combinations of the variable amino acid residues in the structure (Table S1).

The biosynthesis of APs is performed via nonribosomal peptide synthesis (NRPS) pathways (*aptABCD*) encoded in the genomes of a variety of cyanobacteria [40,60–62] and assembled by an NRPS enzyme complex, which has a modular structure [61]. Synthesis is performed stepwise by modules that each contain specific functional domains for the elongation of the peptide sequence through adenylation and thiolation of the activated amino acid residues, ring formation by the epimerization of Lys and *N*-methylation of amino acid X5 [61]. The gene clusters of APs can encode either a single starter module, as in *Nostoc* and *Nodularia*, or two alternative loading modules, as in *Anabaena* sp. 90, allowing the simultaneous synthesis of multiple AP variants [40]. Possibly, due to the relaxed substrate specificity of NRPSs, numerous structural variants of cyanobacterial peptides may be generated [61]. Up to now, more than 150 AP congeners have been reported in the literature (Table S1).

Bioactivity studies have shown that APs can inhibit the enzymes responsible for the regulation of several physiological and metabolic processes [63]. Specifically, AP congeners can inhibit carboxypeptidase A [2,9,29,32,41,64], serine proteases (chymotrypsin [27,65], trypsin [51,65] and elastase [65–67]), serine/threonine protein phosphatases (PP1 and PP2A [9,30,68]) and *Mycobacterium tuberculosis* enzyme MptpB [52]. Of great pharmacological interest is their high activity against the thrombin activatable fibrinolysis inhibitor (TAFIa) (carboxypeptidase), resulting in the stimulation of fibrin clot degradation, which may help to prevent thrombosis [69]. On the other hand, APs can cause toxic effects on microorganisms such as the nematode *Caenorhabditis elegans* [70], the amoeba *Acanthamoeba castellanii* [36] and the planktic crustacean *Daphnia pulex* [71]. Furthermore, APs could possibly control cyanobacterial population density as the presence of APs (i.e., Anabaenopeptin B and Anabaenopeptin F) has been correlated with the triggering of cell lysis that ends up in the collapse of cyanobacterial blooms [72].

**Figure 1.** General structure of anabaenopeptins (APs) and an overview of their variable amino acids. Ala = alanine, AcSer = acetyl-serine, Arg = arginine, Asn = asparagine, Bh-Trp = 2-bromo-5 -hydroxy-tryptophan, Br-Trp = bromo-tryptophan, Cl-Hty = chloro-homotyrosine, EtHph = ethyl-homophenylalanine, Gly = glycine, Har = homoarginine, Hph = homophenylalanine, Hph/MePhe = homophenylalanine/methyl-phenylalanine (isobaric compounds), Hty = homotyrosine, Leu/Ile = leucine/isoleucine (isobaric compounds), Lys = lysine, Met = methionine, MetO = methionine sulfoxide, Met(O)2 = methionine sulfone (S-dioxide), Me-5'-BrTrp = mehtyl-5'-bromotryptophan, Me-5'-hydroxyTrp = mehtyl-5'-hydroxy-tryptophan, MeAhpha = N-methyl -2-amino-6- (hydroxyl phenyl) hexanoic acid, MeAla = methyl-alanine, MeAsn = methyl-asparagine, MeCht = 6-chloro-5-hydroxy-N-methyl-tryptophan, MeCTrp = 6-chloro-N-methyl-tryptophan, Me-formyl kyn = methyl-formyl kynurenine, MeGly = methyl-glycine, MeHph = methyl-homophenylalanine, MeHty = methyl-homotyrosine, MeLeu/MeIle = methyl-leucine/methyl-isoleucine (isobaric compounds), MeTrp = methyl-tryptophan, OMeArg = arginine methyl ester, OMeGlu = glutamic acid methyl ester, OHTrp = hydroxyl-tryptophan, Phe = phenylalanine, PNV = 5-phenylnorvaline, PNL = 6-phenylnorleucine, Ser = serine, Trp = tryptophan, Tyr = tyrosine, Val = valine.

The occurrence of APs in cyanobacterial blooms and cultured strains from freshwater bodies has been reported more frequently during recent years in several countries worldwide, including Japan [2,26–28], Germany [3,13,19,20], Finland [16,18], Norway [73,74], Poland [47,75,76], Slovenia [33], Czech Republic [22,77], Austria [34], Hungary [23], Switzerland [78], Spain [4,6,79], Portugal [37,38], Italy [80–84], France [85,86], United Kingdom [85,87], Turkey [44], Israel [21,39,88], Brazil [89], Canada [59,86,90], USA [64,91–94], New Zealand [95], and India [96]. Recent studies indicate that APs could be more abundant in freshwaters than other toxic cyanobacterial metabolites such as the

known cyanotoxins microcystins [6,87,90]. APs have also been detected in wild-caught fishes (fish muscles) from Pike River, Canada [97].

Structural characterization of cyanobacterial metabolites, including APs, is an emerging issue due to the great diversity of molecules, their bioactivities and possible effects on ecosystems and on human health. Nuclear magnetic resonance (NMR), after the isolation of the compound, usually from a cyanobacterial strain culture, has been used for the structural elucidation of APs e.g., [2,26,27,35]. Mass spectrometric (MS) techniques such as Matrix-Assisted Laser Desorption/Ionization Time-of-Flight (MALDI-TOF) [3,22,95] or MS coupled with liquid chromatography such as liquid chromatography–hybrid triple quadrupole/linear ion trap mass spectrometry (LC–qTRAP) [8,23,25,40,45,48] or liquid chromatography–hybrid triple quadrupole/Time-of-Flight (LC–qTOF) [24,44] are nowadays widely used as they can be applied directly to extracts of field samples or strain cultures. A significant indicator of APs fragmentation spectrum is the characteristic fragment ion of lysine (Lys) at *m*/*z* 84 [3]. The typical fragmentation pattern of APs includes the loss of the amino acid and the CO of the side chain, resulting in the peptide ring ion [3,10,44].

Information regarding the presence of APs in Greek freshwater bodies is limited. Only three monitoring studies have been conducted so far, targeting no more than three AP congeners [5,98,99]. In the present study, an untargeted analysis approach utilizing a LC–qTRAP method was applied for the investigation of APs presence in cyanobacteria from Greece. The main aims were (i) to report, for the first time, the structural diversity of APs in cyanobacterial bloom samples collected from lakes of Greece, (ii) to assess the ability of Greek freshwater cyanobacterial strains to produce APs and (iii) to identify the possible new structures of APs, contributing to a better understanding of the existing variety of these hexapeptide cyanobacterial metabolites.

### **2. Results and Discussion**

### *2.1. Structural Elucidation of Anabaenopeptins*

Thirteen APs were detected in the samples of cyanobacteria from Greek freshwater bodies. The elucidation of proposed AP structures was based on their precursor ions from full scan (MS1) (Table S1) and fragmentation (MS2) spectra, enabling annotation of the compounds [100]. Among them, the possible structures of three AP congeners were proposed for the first time in the frame of the present study. The amino acid sequences of the detected APs with their precursor ions [M + H]+ and the retention time (tR) are provided in Table 1. The proposed structures, extracted ion chromatograms (EIC), full scan spectra (MS1) and fragmentation mass spectra (MS2), of the three newly annotated APs are shown in Figures 2–4, while the elucidation of their spectra are provided in the relevant captions.

**Table 1.** List of anabaenopeptins (APs) detected in cyanobacterial blooms and cyanobacterial strains from Greek lakes.


**Figure 2.** Extracted ion chromatogram (EIC) at *m*/*z* 838, full scan spectrum (MS1) at 9.25 min, fragmentation mass spectrum (MS2) and proposed structure of the new **AP 837** with [M + H]+ at *m/z* 838. (*m/z* 820 = [M + H <sup>−</sup> H2O]+, *m/z* 793 = [M + H <sup>−</sup> OCH2CH3] +, *m/z* 663 = [M + H <sup>−</sup> OEtGlu]+, *m/z* 661 = [M + H <sup>−</sup> Hty]+, *m/z* 637 = [Lys-Val-Hty-MeAla-Phe + H]+, *m/z* 635 = [M + H <sup>−</sup> OEtGlu <sup>−</sup> CO]+, *m/z* 562 = [M + H <sup>−</sup> Val-Hty]+, *m/z* 460 = [MeAla-Phe-Lys-Val + H]+, *m/z* 405 = [MeAla-Phe-Lys-CO-NH2 + H]+, *m/z* 362 = [Val-Hty-MeAla + H]+, *m/z* 320 = [Phe-Lys-CO-NH2 + H]+ and/or [Lys-Val-Hty + H]+, *m/z* 150 = Hty immonium ions, *m/z* 120 = Phe immonium ion, *m/z* 107 = Hty related ion, *m/z* 102 = Glu immonium ion, *m/z* 84 = Lys immonium ion).

**Figure 3.** Extracted ion chromatogram (EIC) at *m*/*z* 852, full scan spectrum (MS1) at 9.43 min, fragmentation mass spectrum (MS2) and proposed structure of the new **AP 851** with [M + H]+ at *m/z* 852. (*m/z* 834 = [M + H <sup>−</sup> H2O]+, *m/z* 807 = [M + H <sup>−</sup> OCH2CH3] +, *m/z* 677 = [M + H <sup>−</sup> OEtGlu]+, *m/z* 675 = [M + H <sup>−</sup> Hty]+, *m/z* 651 = [Lys-Leu/Ile-Hty-MeAla-Phe + H]+, *m/z* 649 = [M + H <sup>−</sup> OetGlu <sup>−</sup> CO]+, *m/z* 564 = [Phe-Lys-Leu/Ile-Hty + H]+, *m/z* 405 = [MeAla-Phe-Lys-CO-NH2 + H]+, *m/z* 376 = [Leu/Ile-Hty-MeAla + H]+, *m/z* 320 = [Phe-Lys-CO-NH2 + H]+, *m/z* 263 = [MeAla-Hty + H]+, *m/z* 150 = Hty immonium ions, *m/z* 120 = Phe immonium ion, *m/z* 107 = Hty related ion, *m/z* 102 = Glu immonium ion, *m/z* 86 = Leu/Ile immonium ion, *m/z* 84 = Lys immonium ion).

**Figure 4.** Extracted ion chromatogram (EIC) at *m*/*z* 895, full scan spectrum (MS1) at 8.92 min, fragmentation mass spectrum (MS2) and proposed structure of the new **AP 894** with [M + H]+ at *m/z* 895. (*m/z* 749 = [M + H <sup>−</sup> Lys]+, *m/z* 723 = [M + H <sup>−</sup> Lys <sup>−</sup> CO]+, *m/z* 705 = [Lys-Leu/Ile-Hty-MeHty-Lue/Ile <sup>−</sup> H2O + H]<sup>+</sup> or [M + H <sup>−</sup> Lys <sup>−</sup> CO <sup>−</sup> H2O]+, *m/z* 636 = [CO-Lys-Leu/Ile-Hty-MeHty]+, *m/z* 608 = [Lys-Leu/Ile-Hty-MeHty]+, *m/z* 530 = [Leu/Ile-Lys-Leu/Ile-Hty]+, *m/z* 482 = [Leu/Ile-Hty-MeHty + H]+, *m/z* 369 = [Hty-MeHty + H]+, *m/z* 305 = [MeHty-Leu/Ile + H]+, *m/z* 240 = [Lys-Leu/Ile + H]+, *m/z* 164 = MeHty immonium ion, *m/z* 107 = Hty related ion, *m/z* 84 = Lys immonium ion).

The detection of APs was based on the diagnostic fragment ion of lysine (Lys) *m*/*z* 84 [3], which was present in the fragmentation spectra of all APs (Figures 2–4 and S2–S11). Structural elucidation of APs was based on fragmentation patterns described in previous studies [3,9,10,23,25,44,48] and on immonium ions of the common amino acids. In Table 1, both leucine (Leu) and isoleucine (Ile) are provided in the proposed AP sequences as these amino acids are isobaric compounds with the same chemical formula (C6H13NO2) and they could not be distinguished.

Generally, one of the intense ions that is always present in the fragmentation spectrum of APs is the ion formed by the loss of the side chain amino acid, i.e., [M + H − X1] +. Fragment ions [M + H − X3] <sup>+</sup> and [M + H − X4] <sup>+</sup> are also commonly found in the APs

spectra. Furthermore, among the most intense fragment ions of APs is the five-peptide ring ion generated after the loss of the side chain, i.e., [Lys-X3-X4-MeX5-X6 + H]+.

The characteristic ions of annotated APs during this study were *m*/*z* 635 = [Lys-Leu/Ile-Hph-MeAla-Phe + H]+ for AP 842 and AP 870; *m*/*z* 637 = [Lys-Val-Hty-MeAla-Phe + H]+ for AP B, AP 837, AP A and AP 872; and *m*/*z* 651 = [Lys-Leu/Ile-Hty-MeAla-Phe + H]+ that was present in the fragmentation spectra of AP F, AP 851, Osc Y and AP 886. Other common fragment ions of elucidated AP structures were *m*/*z* 320 = [Phe-Lys-CO-NH2 + H]+ and *m*/*z* 405 = [MeAla-Phe-Lys-CO-NH2 + H]+ (Figure 2, Figure 3, Figures S4, S5 and S7–S10) that suggested the presence of Phe amino acid in position X6 and MeAla amino acid in position X5.

The fragment ions *<sup>m</sup>*/*<sup>z</sup>* 550 = [MeAla-Phe-Lys-CO-Tyr − H2O]+ and *<sup>m</sup>*/*<sup>z</sup>* 568 = [MeAla-Phe-Lys-CO-Tyr]+ were characteristic of APs with Tyr as the side chain, i.e., AP 842, AP A and Osc Y (Figures S4, S5 and S7), while the fragment ions *m*/*z* 578 = [MeAla-Phe-Lys-CO-MeHty − H2O]<sup>+</sup> and *<sup>m</sup>*/*<sup>z</sup>* 596 = [MeAla-Phe-Lys-CO-MeHty]<sup>+</sup> indicated the presence of MeHty as the side chain, i.e., AP 870, AP 872 and AP 886 (Figures S8–S10).

Congeners of APs with arginine (Arg) as the side chain amino acid (i.e., AP B, AP F, AP 820, AP KB906) had *m*/*z* 201 = [Arg + CO + H]+ as the most intense fragment ion. The other intense fragment ions of the spectra were the [M + H-Arg-CO]<sup>+</sup> and [M + H-Arg-CO-H2O]+. The characteristic Lys immonium ion, *m*/*z* 84, was present in the fragmentation spectra with low intensity (Figures S2, S3, S6 and S11).

Immonium ions of amino acids were also significant indicators of the peptide sequences. The presence of Phe was indicated by an intense peak at *m*/*z* 120 and Hph by *m*/*z* 134. Low intensity peaks at *m*/*z* 107 and *m*/*z* 150 suggested the presence of Hty and *m*/*z* 164 of MeHty, while *m*/*z* 136 was attributed to Tyr, *m*/*z* 58 to MeAla and *m*/*z* 86 corresponded to Leu/Ile.

### *2.2. Anabaenopeptins in Cyanobacterial Blooms from Greek Lakes*

Samples were collected from nine different lakes of Greece during cyanobacterial bloom events, which were mainly dominated by *Microcystis* and *Dolichospermum* species, and were analyzed for the presence of APs. The detected AP congeners and the dominant cyanobacterial species of each sample are presented in Figure S1, and details are provided in Table 2. In total, thirteen different AP congeners were detected, and their amino acid sequences are shown in Table 1.

The presence of APs was confirmed in the majority of the examined samples (91%). In addition, a large within-sample structural diversity of APs was observed as at least six AP congeners were detected in each of the 11 samples (48% of total samples). Two samples contained only one AP congener. The largest diversity of APs was observed in three samples collected from lakes Kastoria (5 October 1995), Kerkini (3 August 1999) and Zazari (5 August 1999); eight APs were detected in each of them. These samples were dominated by *Microcystis* species (Table 2). A large diversity of APs was also observed in samples collected from lakes Pamvotida, Mikri Prespa, and Vistonida.

APs were not detected in two samples collected from lakes Marathonas and Karla, although cyanobacterial species that possibly produce APs were present in both lakes (i.e., *Microcystis flos-aquae* at Lake Marathonas and *Planktothrix* cf. *agardhii* at Lake Karla).

The most frequently detected APs in Greek freshwater samples were AP F (87% of samples) and Osc Y (87%), followed by AP B (65%) and AP 886 (57%). AP A and AP 872 were also common congeners among the samples. AP 820 and AP KB906 were detected in one sample from Lake Kastoria and Lake Zazari, respectively.

AP 894, whose structure is proposed for the first time in the present study, was detected in two samples collected from lakes Kerkini and Zazari. The newly proposed APs, 837 and 851, were detected in one sample collected from Lake Mikri Prespa (4 November 2014).


**Table 2.** Dominant cyanobacterial species and anabaenopeptins detected in samples of cyanobacterial blooms from Greek lakes.

\* Samples were collected from two different sampling points for Lake Amrakia (19 August 1999), Lake Kastoria (5 October 1995) and Lake Pamvotida (18 August 2000). **BOLD:** New AP structures proposed in the frame of the present study.

In two previous monitoring studies targeting AP A and AP B by HPLC–PDA, in which cyanobacterial bloom samples were collected from up to 36 freshwater bodies of Greece, the presence of APs in lakes Zazari (AP A), Kastoria (AP A and AP B) and Pamvotis (AP A and AP B) was reported [5,98]. In the current study, both AP A and AP B were detected by mass spectrometry in lakes Kastoria, Pamvotis, Zazari, Kerkini, Mikri Prespa and Vistonida, along with several other APs congeners.

According to a three-year monitoring study of the Greek Lake Vegoritis targeting 25 cyanobacterial toxins and peptides, AP B and AP F were found to be the most frequently detected cyanobacterial metabolites; they were present in almost all the samples, followed by Osc Y [99]. These results are in agreement with the current study as AP F, Osc Y and AP B were the most commonly occurring AP congeners in the freshwaters of Greece.

The occurrence of cyanobacterial metabolites, including APs in freshwater blooms, has been investigated in a number of past studies. Analysis by MALDI-TOF MS showed the presence of AP B and AP F in samples collected from lakes in Italy [80,81,102], Germany [3], Spain [79] and Brazil [89]. In samples collected from a waterbody of Poland and analyzed by LC–qTRAP MS/MS, the most abundant AP congener was AP B, followed by AP A, AP

F, AP G, Osc Y, AP D and AP 915 [75]. The presence of AP A, AP B, AP F and Osc Y was also confirmed by LC–HRMS in samples collected from the freshwaters of Spain [6] and the Czech Republic [77], while AP B, AP A and Osc Y were identified in samples from the United Kingdom [87]. Based on the results of this study and of previous reports, it appears that AP B and AP F followed by AP A and Osc Y are the most frequently reported APs not only in Greece but also in the European continent.

AP F, Osc Y, AP B and AP A are protease inhibitors that possess activity against carboxypeptidase A and protein phosphatase 1 (PP1) [9,30,64]. AP B and AP F are also highly selective TAFIa inhibitors [69] and elastase inhibitors, with no activity towards chymotrypsin and trypsin [66], while Osc Y have presented inhibitory activity against chymotrypsin [27]. Additionally, AP A, AP B and AP F have had toxicity effects in the nematode *Caenorhabditis elegans* [70]. Even though APs toxicity effects on animal models and microorganisms have been reported, there remains a lack of data regarding their toxicity and impact on human health [12].

APs are the 3rd class of cyanopeptides with the highest structural diversity after microcystins and cyanopeptolins [103]. In the present investigation, thirteen structures of APs from the cyanobacteria of Greek freshwaters were detected, and they had a rather low diversity of variable amino acids (Figure 5). In particular, all the moieties that composed the ring structures were represented by only two different amino acids per site. Even though the diversity was limited, it is interesting that the two amino acids that were determined in each position are among the most commonly found in known AP congeners (Figure 1).

**Figure 5.** Diversity and frequency of variable amino acids in the structures of anabaenopeptins detected in Greek freshwaters.

Specifically, the currently known 42 APs from freshwater environments mainly consist of Val (45%) and Ile (29%) in position X3, Hty (64%) and Hph (29%) in position X4, MeAla (50%) and MeHty (38%) in position X5 and Phe (45%) and Ile (24%) in position X6 [12]. The 13 APs identified in Greek freshwaters consist of Val (38%) and Ile (62%) in position X3, Hty (69%) and Hph (31%) in position X4, MeAla (85%) and MeHty (15%) in position X5 and Phe (85%) and Ile (15%) in position X6 (Figure 5). A comparison of findings strongly supports that the variable amino acids of AP rings determined during this study are consistent with the most common ones of the known APs from freshwaters.

A higher diversity of amino acids was observed in the side chain. Arg (31%) was the most frequent, followed by Tyr (23%), MeHty (23%), OEtGlu (15%) and Lys (8%). Arg and Tyr are present in the side chains of commonly found AP congeners worldwide (i.e., AP B and AP F have Arg; AP A and Osc Y have Tyr). Contrarily, the presence of MeHty as a side

chain has been reported for only seven AP congeners that were detected in cyanobacteria from Lake Balaton, Hungary [23]. The proposed side chain of the three novel APs consists of infrequent amino acids (i.e., Lys and OEtGlu). Lys (AP 894) has been determined in the side chain of six known congeners (Figure 1, Table S1), while OEtGlu (AP 837 and AP 851) is proposed for the first time. A previous study reported the presence of OMeGlu occupying the side chain amino acid position in the AP MM823 [65]. In fact, AP MM823 and the newly proposed AP 837 also share the same five-peptide ring structure. Although methylated amino acids are frequently occurring in AP structures, ethylated ones have also been reported [10,25], indicating the metabolomic potential of cyanobacteria.

### *2.3. Anabaenopeptins in Cyanobacterial Strains Isolated from Greek Freshwaters*

Thirty cyanobacterial strains from the TAU-MAC culture collection [104], isolated from the freshwaters of Greece, were analyzed in order to evaluate their ability to produce APs (Table S2), i.e., fourteen strains of *Microcystis*, five of *Nostoc*, three of *Jaaginema*, two of *Synechococcus*, and one from the species of the genera *Anabaena*, *Calothrix*, *Chlorogloeopsis*, *Desmonostoc*, *Limnothrix* and *Nodosilinea*. APs were only detected in one strain extract out of the thirty examined. In particular, AP A and Osc Y were identified in the extract of *Microcystis ichtyoblabe* TAU-MAC 0510.

Although AP F and AP B along with Osc Y were the most frequently detected APs in cyanobacterial bloom samples in this study, they were not detected in any of the examined cyanobacterial strains. The diversity of APs in the isolated strains was limited compared to that of bloom extracts. This finding is in agreement with the results of previous studies as it was reported that *Microcystis* strains have a less diverse peptide pattern compared to that of the entire population of a bloom sample from a German lake [19], and that the *Planktothrix agardhii* samples from a Polish freshwater reservoir contained up to seven APs while the two strains isolated from the reservoir contained only one AP [75]. This was rather expected because the diversity of APs in field bloom samples reflects the high diversity of the chemotypes present in water bodies, therefore it cannot be compared with the diversity of the compounds in isolated strains [19,75]. The results of previous chemodiversity studies of freshwater cyanobacterial strains also indicate the limited presence of AP congeners in the samples. Welker et al. reported the presence of APs in only 9% of 850 examined *Microcystis* colonies with five AP structural variants in total [22] while, in another study, 165 *Microcystis* colonies were examined and only up to four APs were detected in 21% of analyzed samples [20]. Martins et al. have also reported a limited presence of APs in *Microcystis aeruginosa* strains where one to three APs were detected in the 30% of examined strains [38]. Furthermore, in an investigation of 18 *Planktothrix* clonal strains, APs were present in 11 of them, with one, two and three APs present in seven, three and one strain, respectively [13]. The limited presence of APs in cyanobacterial strains may also be correlated with the evidence that cyanobacterial strains could lose the ability to produce cyanopeptides under laboratory conditions [105].

In a previous chemo-diversity study including 24 *Microcystis* strains isolated from the same freshwater blooms or from different populations in various geographical areas (i.e., Netherlands, Scotland, France, Senegal, Burkina Faso), it was found that AP A, AP B, AP F and Osc Y were the most commonly detected AP congeners and were mainly produced by *Microcystis aeruginosa* strains, while all the examined *Microcystis wesenbergii*/*M. viridis* strains did not produce APs. A comparison of the specific chemical footprints of the examined strains showed that the metabolite content was influenced globally by microcystin production rather than sampling locality origins [106]. In another study, it was concluded that AP B and AP E/F were among the principal cyanopeptides detected in 165 *Microcystis* sp. colonies isolated from German lakes and that APs were mostly produced by *Microcystis ichthyoblabe* colonies than by *Microcystis aeruginosa* [20]. According to Fastner et al., AP B, AP F and Osc.Y were the most prominent APs in *Microcystis ichthyoblabe* colonies isolated from a German lake, followed by AP I and AP A, while APs were rarely detected in the *Microcystis aeruginosa* colonies and not detected at all in *Microcystis wesenbergii* colonies [19]. A common conclusion of the above

studies was that *Microcystis aeruginosa* colonies predominately produced microcystins; this was in contrast to *Microcystis ichthyoblabe* colonies that mainly produced APs rather than microcystins [19,20]. This is in agreement with the results of the present study where one strain belonging to cyanobacterial species *Microcystis ichthyoblabe* was found to be positive to APs while strains belonging to *Microcystis aeruginosa* and *Microcystis viridis* were negative to APs (Table S2).

In general, AP A, AP B, AP F and Osc Y are the most commonly detected APs both in *Microcystis* and *Planktothrix* strains isolated from several water bodies of European countries, such as Austria [34], the Czech Republic [22], Finland [14], Germany [13,19,20], Norway [74], Portugal [37] and Switzerland [31]. The current study constitutes the first investigation into APs presence in several cyanobacterial strains isolated from Greek freshwaters.

### **3. Conclusions**

The structural diversity of APs from bloom samples and cultured cyanobacteria strains of Greek freshwaters was investigated for the first time, utilizing LC–qTPAR MS/MS in IDA and EIP modes in order to structurally elucidate APs from their fragmentation spectra. Overall, thirteen APs were annotated, with three of these being reported for the first time (AP 837, AP 851 and AP 894). A variety of APs were found to occur in 21 out of 23 samples from cyanobacterial blooms from seven out of nine lakes that were mainly dominated by *Microcystis* and *Dolichospermum* species. The most frequently occurring APs in bloom samples were AP F and Osc Y, followed by AP B, AP 886 and AP A. On the other hand, in thirty samples of cultured cyanobacterial strains isolated from the freshwater bodies of Greece, APs (AP A and Osc Y) were only found in *Microcystis ichtyoblabe* TAU-MAC 0510. The results of this study are in general agreement with previous studies on the occurrence of APs in European freshwater bodies and contribute to the expansion of the range of known AP congeners by introducing three new AP structures and their mass fragmentation spectra. Considering that APs are a class of cyanobacterial bioactive metabolites that naturally occur in water bodies in high frequency and possibly in significant amounts, the results of this study highlight the need for further assessment of their environmental effects and impacts.

### **4. Materials and Methods**

### *4.1. Cyanobacterial Bloom Samples*

Samples were collected from nine Greek lakes (Amvrakia, Kastoria, Pamvotida, Kerkini, Zazari, Mikri Prespa, Vistonida, Karla, Marathonas) during episodes of cyanobacterial bloom (Table 2). General characteristics and location of the freshwater bodies are provided in the details of previous studies [98,107,108]. Water samples (100–1500 mL) were collected in airtight polyethylene bottles from the surface layer (0–35 cm) at the margins of the lakes where accumulation of cyanobacteria had been observed from May to October in 1995, 1999, 2000, 2010, 2014 and 2015, as previously described [5,98]. Samples were filtered through Whatman GF/C filters (Millipore, Cork, Ireland), lyophilized and stored at −20 ◦C until analysis. The cyanobacterial biomass of the samples ranged from 10–1000 mg/L. Dominant cyanobacterial species were characterized by microscopic analysis, as previously reported [5,98,109].

### *4.2. Source and Culture Conditions of Cyanobacterial Strains*

Thirty cyanobacterial strains isolated from Greek freshwaters from 1999 to 2010 [109] were identified and provided by Thessaloniki Aristotle University Microalgae and Cyanobacteria (TAU-MAC) Culture Collection [104]. Strains were planktic or benthic; details of their origin and isolation are provided in [109]. Cyanobacterial strains belonging to *Chroococcales, Synechococcales* and *Nostocales* based on polyphasic taxonomy were classified into 10 genera (*Anabaena*, *Microcystis*, *Nostoc*, *Synechococcus*, *Limnothrix*, *Calothrix*, *Nodosilinea*, *Desmonostoc*, *Chlorogloeopsis* and *Jaaginema*) and 16 taxa, as listed in Table S2 [110]. Cyanobacterial strain cultures were grown in BG11 medium with or without nitrogen (BG110 for the nitrogen-fixing strains, see Table S2), shaken manually once per day and maintained at

25 ◦C (*Microcystis* strains) or 20 ± 1 ◦C (strains of the rest genera) under cool white light fluorescent lamps (Sylvania Standard F36W/154-T8, SLI, Sylvania, Erlangen, Germany) with a light intensity of 20–25 μmol m−<sup>2</sup> s−<sup>1</sup> in a 16/8 h light/dark cycle. At the end of the exponential phase of culture growth (between days 20 and 35, depending on the strain, see Table S2), the whole liquid culture (250 mL) was centrifuged and the cyanobacterial cells collected, lyophilized and stored at −20 ◦C until analysis. Chlorophyll-*a* was extracted from 5 mL of wet biomass with 95% (*v/v*) acetone solution and spectrophotometrically quantified, as outlined in APHA (2005) [111]. The chlorophyll-*a* concentration of the strains at the time of the collection (as an estimate of their biomass) ranged from 6.21–6.77 mg/L.

### *4.3. Sample Preparation and LC–MS/MS Analysis*

Analysis of two different sample types, i.e., cyanobacterial blooms and cyanobacterial strain cultures, was performed. The same amount of each sample type was extracted and analyzed. Lyophilized biomass (~10 mg) of each sample was extracted with 1.5 mL of 75% methanol:25% water assisted by vortexing and sonication in an ice bath for 15 min. After centrifugation (10,000 rpm, 15 min), the supernatants were collected and further centrifuged (10,000 rpm, 5 min) prior to LC–MS/MS analysis.

Untargeted analysis was carried out with an Agilent 1200, liquid chromatography apparatus (Agilent Technologies, Waldboronn, Germany) coupled with a hybrid triple quadrupole/linear ion trap mass spectrometer (QTRAP5500, Applied Biosystems, Sciex; Concorde, ON, Canada) according to Mazur-Marzec et al., 2013 [44]. Chromatographic separation was achieved with a reversed phase column (Zorbax Eclipse XDB-C18 4.6 × 150 mm, 5 μm Agilent Technologies, Santa Clara, CA, USA) applying gradient elution. Mobile phases consisted of (A) acetonitrile and (B) 5% acetonitrile in MilliQ water, both containing 0.1% formic acid; flow rate was 0.6 mL min−<sup>1</sup> and injection volume was 5 μL. Ionization was performed with electrospray (ESI) source in positive mode. For MS detection, information-dependent acquisition (IDA) mode and enhanced ion product (EIP) mode were applied. In IDA mode, a full scan from 500 to 1200 Da was acquired for detection of the compounds. EIP mode was triggered when the signal of an ion was above a threshold of 500,000 cps; the ions were fragmented in the collision cell (Q2) and fragmentation spectra were recorded from 50 to 1000 Da with a scan speed of 2000 Da s−<sup>1</sup> and collision energy (CE) of 60 V with collision energy spread (CES) of 20 V. Analyst QS® 1.5.1 software was used for data acquisition and processing. Obtained fragmentation spectra were examined in order to elucidate the structures of occurring APs.

**Supplementary Materials:** The following are available online at https://www.mdpi.com/article/ 10.3390/toxins14010004/s1, Table S1: List of anabaenopeptins reported in the literature and their amino acid sequence, Table S2: List of the cyanobacterial strains from Greek freshwaters, examined for their ability to produce APs, Figure S1: Anabaenopeptins' presence in cyanobacterial blooms of Greek freshwaters and the dominant species, Figure S2: Fragmentation mass spectrum of Anabaenopeptin 820 with precursor ion at *m*/*z* 821 [M + H]+, Figure S3: Fragmentation mass spectrum of Anabaenopeptin B with precursor ion at *m*/*z* 837 [M + H]+, Figure S4: Fragmentation mass spectrum of Anabaenopeptin 842 with precursor ion at *m*/*z* 842 [M + H]+, Figure S5: Fragmentation mass spectrum of Anabaenopeptin A with precursor ion at *m*/*z* 844 [M + H]+, Figure S6: Fragmentation mass spectrum of Anabaenopeptin F with precursor ion at *m*/*z* 851 [M + H]+, Figure S7: Fragmentation mass spectrum of Oscillamide Y with precursor ion at *m*/*z* 858 [M + H]+, Figure S8: Fragmentation mass spectrum of Anabaenopeptin 870 with precursor ion at *m*/*z* 870 [M + H]+, Figure S9: Fragmentation mass spectrum of Anabaenopeptin 872 with precursor ion at *m*/*z* 872 [M + H]+, Figure S10: Fragmentation mass spectrum of Anabaenopeptin 886 with precursor ion at *m*/*z* 886 [M + H]+, Figure S11: Fragmentation mass spectrum of Anabaenopeptin KB 906 with precursor ion at *m*/*z* 907 [M + H]+.

**Author Contributions:** Conceptualization, S.-K.Z., T.K., S.G., A.H. and H.M.-M.; methodology, S.-K.Z., A.H. and H.M.-M.; software, S.-K.Z. and H.M.-M.; formal analysis, S.-K.Z.; investigation, S.-K.Z., T.K., S.G., A.H. and H.M.-M.; resources, T.K., S.G., A.H. and H.M.-M.; data curation, S.-K.Z. and H.M.-M.; writing—original draft preparation, S.-K.Z.; writing—review and editing, T.K., S.G., A.H. and H.M.-M.; visualization, S.-K.Z.; supervision, A.H. and H.M.-M.; funding acquisition, T.K., A.H. and H.M.-M. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was co-financed by Greece and the European Union (European Social Fund— ESF) through the Operational Programme «Human Resources Development, Education and Lifelong Learning» in the context of the project "Reinforcement of Postdoctoral Researchers—2nd Cycle" (MIS-5033021), implemented by the State Scholarships Foundation (IKΥ).

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** Data are contained within the article or Supplementary Material.

**Acknowledgments:** The authors acknowledge COST Action ES 1105 "CYANOCOST—Cyanobacterial blooms and toxins in water resources: Occurrence impacts and management" www.cyanocost.net for adding value to this study through networking and knowledge sharing with European experts in the field.

**Conflicts of Interest:** The authors declare no conflict of interest.

### **References**


### *Communication* **Investigation of the Occurrence of Cyanotoxins in Lake Karaoun (Lebanon) by Mass Spectrometry, Bioassays and Molecular Methods**

**Noura Alice Hammoud 1,2,3, Sevasti-Kiriaki Zervou 2, Triantafyllos Kaloudis 2,4, Christophoros Christophoridis 2, Aikaterina Paraskevopoulou 2,5, Theodoros M. Triantis 2, Kamal Slim 1, Joanna Szpunar 3, Ali Fadel 1, Ryszard Lobinski 3,6 and Anastasia Hiskia 2,\***


**Abstract:** Lake Karaoun is the largest artificial lake in Lebanon and serves multiple purposes. Recently, intensive cyanobacterial blooms have been reported in the lake, raising safety and aesthetic concerns related to the presence of cyanotoxins and cyanobacterial taste and odor (T&O) compounds, respectively. Here, we communicate for the first time results from a recent investigation by LC-MS/MS covering multiple cyanotoxins (microcystins (MCs), anatoxin-a, cylindrospermopsin, nodularin) in water and fish collected between 2019 and 2020. Eleven MCs were identified reaching concentrations of 211 and 199 μg/L for MC-LR and MC-YR, respectively. Cylindrospermopsin, anatoxin-a and nodularin were not detected. The determination of the total MCs was also carried out by ELISA and Protein Phosphatase Inhibition Assay yielding comparable results. Molecular detection of cyanobacteria (16S rRNA) and biosynthetic genes of toxins were carried out by qPCR. Untargeted screening analysis by GC-MS showed the presence of T&O compounds, such as β-cyclocitral, β-ionone, nonanal and dimethylsulfides that contribute to unpleasant odors in water. The determination of volatile organic compounds (VOCs) showed the presence of anthropogenic pollutants, mostly dichloromethane and toluene. The findings are important to develop future monitoring schemes in order to assess the risks from cyanobacterial blooms with regard to the lake's ecosystem and its uses.

**Keywords:** cyanotoxins (CTs); microcystins (MCs); volatile organic compounds (VOCs); taste and odor (T&O) compounds; SPE-LC-MS/MS; HS-SPME-GC/MS

**Key Contribution:** First report confirming the occurrence of MC congeners and cyanobacterial/algal T&O compounds in lake Karaoun; Lebanon.

### **1. Introduction**

Cyanobacteria are photosynthetic microorganisms commonly found in surface waters. They can produce a large variety of secondary metabolites, including toxic compounds,

**Citation:** Hammoud, N.A.; Zervou, S.-K.; Kaloudis, T.; Christophoridis, C.; Paraskevopoulou, A.; Triantis, T.M.; Slim, K.; Szpunar, J.; Fadel, A.; Lobinski, R.; et al. Investigation of the Occurrence of Cyanotoxins in Lake Karaoun (Lebanon) by Mass Spectrometry, Bioassays and Molecular Methods. *Toxins* **2021**, *13*, 716. https://doi.org/10.3390/ toxins13100716

Received: 23 July 2021 Accepted: 5 October 2021 Published: 10 October 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

known as cyanotoxins (CTs). CTs have various chemical structures and modes of toxicity (Figure S1). Microcystins (MCs) [1] and Nodularin (NOD) [2] are cyclic peptides, both containing the unique L-amino acid Adda ((2S,3S,8S,9S)-3-amino-9-methoxy-2,6,8-trimethyl-10-phenyldeca-4,6-dienoicacid), which is responsible for their hepatotoxic activity [3,4]. Cylindrospermopsin (CYN) is an alkaloid cyanotoxin with cytotoxic, dermatotoxic, hepatotoxic and possibly carcinogenic potency [5]. The alkaloid Anatoxin-a (ATX), also known as "Very Fast Death Factor", is a bicyclic secondary amine (2-acetyl-9-azabicyclo(1,4,2)non-2-ene) with acute neurotoxicity [6]. Among CTs, MCs is the most frequently reported class [7,8].

Under favorable environmental conditions of temperature, presence of nutrients and light, cyanobacteria can proliferate excessively and form blooms [9]. The presence of CTs in toxic cyanobacterial blooms poses a significant risk to aquatic ecosystems and drinking water sources, while it has been associated with lethal poisonings of wild animals, livestock and humans [10]. The World Health Organization (WHO) established guideline values for MCs, CYN and ATX for drinking-water (lifetime, short-term or acute) and recreational exposure and published guides to hazard identification and management of risks posed by cyanobacteria and their toxins at each step of the water-use system [11].

Lake Karaoun (Qaraoun) is the largest water body in Lebanon and has multiple uses. The reservoir was created in 1959 and is intended for irrigation, hydropower, and recreational activities, as well as an anticipated drinking water supply to the capital, Beirut [12]. However, nutrient enrichment due to the excessive use of fertilizers in the lake's watershed, untreated sewage and industrial waste water runoffs, are possibly contributing to the degradation of the reservoir's water quality [13]. In 2009, the cyanobacteria *Aphanizomenon ovalisporum* and *Microcystis aeruginosa* were first reported and were seen to dominate the lake's phytoplankton. Their proliferation was attributed to eutrophication and climate change resulting in the extension of dry periods and the rise in water temperature, as well as the pollution from multiple-industrial and agricultural sources [14]. In recent years, the blooms have become more intense, forming thick scums and emitting unpleasant odors in the area.

There is limited information on toxic cyanobacterial blooms in Lake Karaoum [14–16], with a previous study reporting the presence of CYN produced by *Aphanizomenon ovalisporum,* however, not confirmed by mass spectrometry [17]. There are no published data regarding the presence of MCs in the reservoir, especially during incidents of cyanobacterial blooms. Due to the multiple uses of the reservoir, including the supply of drinking water, there is an urgent need to definitively assess and confirm the presence of CTs in Lake Karaoun, in order to enable the design of future monitoring programs and to develop related management strategies.

In addition to CTs, cyanobacteria produce a plethora of volatile secondary metabolites, with several of them having a strong taste and odor (T&O). This can make reservoirs unacceptable as drinking water sources due to consumer complaints and can negatively impact recreational activities and tourism [18,19]. Indeed, concerning Lake Karaoun, there has been circumstantial reporting of bad smells, but the presence of T&O compounds has never been investigated and it is not clear if the origin of T&O is from cyanobacteria or from industrial pollution. T&O compounds can be present at very low concentrations and although they are generally not toxic, they are responsible for the unpleasant smell of drinking water. Their removal may require re-planning and additional investments in water treatment [20].

The primary aim of this communication is to report for the first time conclusive results of the presence of CTs (MCs, CYN, ATX) in Lake Karaoun, using mass spectrometric analytical techniques, in addition to bioassays and molecular methods. A further aim is to investigate the presence of T&O compounds, in order to understand whether they originate from cyanobacterial metabolism or from anthropogenic pollution. To meet these objectives, a set of complementary methods were used to produce reliable results for a variety of cyanotoxins and T&O compounds. Since Middle-Eastern lakes (with the exception of Lake Kinneret [21]) are under-studied in regard to toxic cyanobacteria, the results presented

give important information on the presence of CTs and T&O compounds in the area, enabling future monitoring programs and management strategies for Lake Karaoun.

### **2. Results**

### *2.1. Water Quality and Diversity of Cyanobacteria in Lake Karaoun*

During the period of field campaigns (2019–2020), the diversity of cyanobacteria in the reservoir was limited (Figure 1). *Aphanizomenon ovalisporum* was the dominant species (>95%) from October 2019 to April 2020, while *Microcystis aeruginosa* was the most abundant on July and October 2020. In August 2019, both *Microcystis aeruginosa* and *Aphanizomenon ovalisporum* were equally present with abundances of 50% and 49%, respectively. In addition, in July of the same year, *Microcystis ichtyoblable* and *Woronichia naegeliana* were identified in very low abundances, of 1% each. A microscope image of the dominant cyanobacterial species, i.e., *Microcystis aeruginosa* and *Aphanizomenon ovalisporum* is shown in Figure 2.

**Figure 1.** Abundance (%) of cyanobacterial species in Lake Karaoun per sampling month (August 2019–October 2020).

**Figure 2.** Dominant cyanobacteria species in Lake Karaoun: *Microcystis aeruginosa* and *Aphanizomenon ovalisporum.*

Physico-chemical parameters of the lake's water were also monitored during the sampling period by the National Litani River Authority, which is in charge of the river management (Table S5). In most of the sampling dates, concentrations of nitrate and phosphate exhibited elevated values up to 16.72, 0.28 and 0.16 expressed as mg/L of NO3-N, NO2-N and PO4-P, respectively. This is in agreement with the assessment of the ecological status of the Karaoun reservoir by Fadel et al., classifying the lake as hypereutrophic, with low phytoplankton biodiversity and regular blooms of toxic cyanobacteria [15,16]. Degradation of the reservoir's water quality is considered to be mainly due to the significant

loads of untreated sewage water and the discarding of agricultural and livestock remnants into the river stream.

### *2.2. Occurrence of Cyanotoxins (CTs) in Lake Karaoun*

Occurrence of CTs in samples from Lake Karaoun was assessed by liquid chromatography tandem mass spectrometry (LC-MS/MS), complemented by ELISA, Protein Phosphatase Inhibition Assay (PPIA) and molecular detection of cyanobacteria and toxin genes by qPCR.

### 2.2.1. LC-MS/MS Analysis of Water Samples

Water samples were analyzed by LC-MS/MS for detection and quantitative determination of extracellular and intracellular CTs, i.e., ATX, CYN, NOD and 12 MC variants (Table 1). Cyanotoxins were found in 50% of samples (8 out of 16), extending over 2019 and 2020. CYN, ATX and NOD were not detected in any of the samples. On the other hand, 11 out of 12 MC variants i.e., dmMC-RR, MC-RR, MC-YR, dmMC-LR, MC-LR, MC-HilR, MC-WR, MC-LA, MC-LY, MC-LW and MC-LF were detected in at least one sample, while MC-HtyR was not detected in any samples. The samples with the largest diversity of MCs were sample 2 (S2-October 19, 10 variants), sample 16 (S3-October 20, 9 variants) and sample 4 (S1-December 19, 5 variants). MC-RR was the most frequently detected toxin (8 out of 16 samples). The sample with the highest concentrations of MCs was sample 16 (S3-October 20), where total MC-LR (sum of extracellular and intracellular fractions) reached 211 μg/L and MC-YR 199 μg/L. In this sample, the dissolved (extracellular) amounts of MC-LR and MC-YR were roughly 100-fold of the intracellular amounts, in contrast to samples 2 (S2-October 19) and 4 (S1-December 19), where MC-LR, MC-RR and MC-YR are found mostly as intracellular. The high proportion of extracellular MCs in sample 16 can possibly be attributed to a decaying phase of the bloom, with disruption of cyanobacteria cells and release of MCs into water. An indicative MRM chromatogram (sample 2, S2-02/10/2019—intracellular fraction) is presented in Figure S2.


Concentrations ( μg/L) of extracellular/intracellular cyanotoxins (CYN, ATX, NOD, 12 MC variants) in water samples collected from Lake Karaoun

> **Table 1.**

 during

1 ND: Not detected; 2 NA: Not analyzed; 3 <LOQ: Values higher than limit of detection (LOD) and lower than limit of

quantitation

 (LOQ); LODs and LOQs are given in Table S1.

### 2.2.2. ELISA and PPIA Analysis of Water Samples

Since LC-MS/MS analysis targeted only a limited number of specific MC variants (12), assessment of the presence of MCs was complemented with ELISA and Protein Phosphatase Inhibition Assay (PPIA), in order to estimate total MCs concentration based on structural (ELISA) and functional (PPIA) similarities. Samples were analyzed for extracellular (dissolved) and intracellular MCs, similarly to LC-MS/MS. Results of ELISA and PPIA, together with the sum of MCs determined by LC-MS/MS, for comparison, are presented in Table 2. In general, samples in which either MCs were not detected by LC-MS/MS or their concentrations were lower than the LODs of ELISA/PPIA (0.10/0.25 μg/L), were negative by ELISA and PPIA with the exception of samples 8 (S5–17/2/2020), 11 (S1-15/04/2020), 12 (S2-22/04/2020) and 13 (S4-22/04/2020). Especially, in the latter sample, MCs were detected only by ELISA/PPIA as extracellular, in comparable concentrations (2.30/2.32 μg/L eq. MC-LR). MCs were found mostly as extracellular in sample 16 (S3-9/10/2020) by ELISA and PPIA, in agreement to LC-MS/MS, although in this sample ELISA and PPIA gave about 55% and 32% lower concentrations than LC-MS/MS.

**Table 2.** Total Extracellular/total intracellular MCs concentrations in water samples from Lake Karaoun by LC-MS/MS (sum of 12 extracellular MCs/sum of 12 intracellular MCs, μg/L), ELISA (μg/L MC-LR equivalents) and PPIA (μg/L MC-LR equivalents).


ND = Not detected, NA = Not Available. LODs: LC-MS/MS 0.001 μg/L; ELISA 0.10 μg/L; PPIA 0.25 μg/L. \* analysis with validated methods [22,23], extracellular toxins RSD < 16%, intracellular toxins RSD < 26%; \*\* duplicate analysis, RSD < 25%.

2.2.3. Molecular Detection of Cyanobacteria and Cyanotoxin Genes with qPCR

Molecular detection of cyanobacteria (16S rRNA) and biosynthetic genes of MCs, NOD, CYN, and Saxitoxins (STX) was carried out by qPCR. All samples were found positive for the presence of cyanobacteria. Samples 2 (S2—02/10/2019), 4 (S1—22/12/2019) and 16 (S3—09/10/2020) were found positive for MC producing genes (*mcyE*), with sample 4 being the most abundant in both cyanobacteria and MC genes among all samples. Genes associated with production of CYN (*cyrA*) and STX (*stxA*) were not detected in any of the samples. Results (gene copies per ml of sample) are presented in Table 3.

**Table 3.** Detection of cyanobacteria genes (16S rRNA) and biosynthetic genes of MCs and NOD (*mcyE),* CYN (*cyrA*) and STX (*stxA*) in water samples from Lake Karaoun (gene copies/mL).


NA: not available, ND: not detected.

### 2.2.4. LC-MS/MS Analysis of MCs in Fish Samples

Fish (*Cyprinus caprio*) flesh and liver samples were also analyzed for 12 MC variants by LC-MS/MS, using an in-house method. MCs were not detected in the analyzed samples. An indicative LC-MS/MS MRM chromatogram of a fish liver sample along with the chromatogram of a multi-toxin standard (Figure S3), shows the absence of MCs.

### *2.3. Taste and Odor (T&O) and Volatile Organic Compounds (VOCs)*

Untargeted screening of volatile compounds by Headspace Solid Phase Microextraction coupled to Gas Chromatography-Mass Spectrometry (HS-SPME-GC-MS) led to the detection and identification of 20 compounds belonging to the following chemical groups: terpenes/terpenoids, hydrocarbons, aldehydes/ketones, phenols, phthalates, alkyl sulfides and indoles (Table 4). The chemical structures of the identified volatile compounds are presented in Figure 3. Two typical cyanobacterial terpenoids T&O compounds, β-cyclocitral (tobacco/woody odor) and β-ionone (floral odor), were detected in samples 2, 4 and 16 and 4, 12 and 16, respectively (Table S2). Nonanal, a known cyanobacterial/algal aldehyde with a fishy odor [24], was detected in 9 samples. Dimethyl disulfide, dimethyl trisulfide and dimethyl tetrasulfide, which are alkyl sulfides associated with algae/cyanobacteria were detected in sample 4, which had a strong swampy/septic odor. Samples 2 and 4 also contained 3-methylindole (skatole), a fecal compound with a characteristic odor. A number of hydrocarbons were also detected, with 2,2,4,6,6-pentamethyl heptane being the most common, as it was found in all samples. Industrial pollutants 2,4-di-tert-butylphenol, diisobutyl phthalate and p-cresol were detected in samples 10, 4 and 2, respectively. Results of untargeted HS-SPME-GC/MS screening per sample are given in Table S2.


**Table 4.** Compounds detected by untargeted HS-SPME-GC/MS screening, identification level of analysis, chemical group and retention time (tR) of detected compounds, number of samples in which they have been identified and their % peak areas.

A: Retention Index (RI) ± 20, spectral matching ≥80%; B: RI ± 20, spectral matching ≥90%; C: matching with a reference standard; D: spectral matching ≥80%.

**Figure 3.** Chemical structures of the volatile compounds identified.

Quantitative determination of 36 VOCs (given in 5.8.2) was carried out by HS-SPME-GC/MS. The targeted compounds are typical industrial pollutants of anthropogenic origin. Concentrations of VOCs which were detected in at least one sample are presented in Table 5. All samples were found to contain dichloromethane (25.1–34.0 μg/L). Toluene was present in 12 samples at concentrations up to 0.81 μg/L. Trichloromethane (chloroform) was detected at low concentrations in samples 11 and 16 (0.40 and 0.50 μg/L, respectively). Traces (<0.25 μg/L) of trichlorobenzenes were detected in sample 6, and traces of 1,2 dichloropropane, benzene, xylenes and styrene in sample 16.

**Table 5.** Concentration (μg/L) of VOCs detected in at least one sample by targeted HS-SPME-GC/MS analysis


ND: Not Detected. Sample 12 was not available for analysis.

### **3. Discussion**

A complementary set of analytical methods was used to investigate and confirm the occurrence of cyanotoxins in Lake Karaoun for the first time. Unambiguous determination of 12 MC congeners, NOD, ATX and CYN was carried out by LC-MS/MS, complemented by ELISA and PPIA for MCs and qPCR for MC, CYN and STX producing genes. Results confirm the presence of several congeners (variants) of microcystins in the lake at elevated concentrations during October 2019 and 2020. These early findings give important information on the occurrence, variants and levels of occurring CTs, and they can be used for the establishment of reliable future monitoring programs to support management of cyanobacteria, cyanotoxins and T&O compounds in the context of the multiple uses of the lake.

LC-MS/MS analysis confirmed the presence of dmMC-RR, MC-RR, MC-YR, dmMC-LR, MC-LR, MC-HilR, MC-WR, MC-LA, MC-LY, MC-LW, and MC-LF, with concentrations at 146.8, 25.8 and 429.3 μg/L in October 2019, December 2019 and October 2020, respectively, expressed as the sum of extra and intracellular fractions. MC-LR was the most abundant MC congener followed by MC-YR and MC-RR. CYN, ATX and NOD were not detected in any of the samples.

Results from the analysis of water samples using ELISA and PPIA for total MCs concentration were in general agreement with those obtained from LC-MS/MS (Table 2). In samples 11, 12 and 13, the higher values for extracellular fraction obtained by ELISA and

PPIA may indicate the presence of MC congeners other than those that were targeted by the LC-MS/MS method used in the study. This limitation, caused by the unavailability of commercial MC reference standards is well known in the literature, and it underscores the need for complementary quantitative analysis for total MCs (e.g., ELISA, PPIA), or for the development of non-targeted methods based on high-resolution mass spectrometry and related mass-spectral databases [25]. However, discrepancies between ELISA/PPIA and LC-MS/MS could also be attributed to the well-documented weaknesses of ELISA/PPIA. In particular, the shortcomings of ELISA are its high susceptibility to matrix effects, the variable cross-reactivity of different MC congeners and the response to degradation and/or transformation products of MCs that can result in overestimations [26]. In addition, PPIA based on the assessment of enzyme activity indicating the overall toxicity, does not have the same sensitivity to the different MCs and may also interfere with other unknown compounds present in the sample, thus resulting in under- or over-estimation of the concentration of toxins [27]. Furthermore, ELISA/PPIA lack the specificity of LC-MS/MS, as they respond to structurally/functionally similar molecules in total, and not to specific MC congeners. ELISA/PPIA are useful, quick, easy approaches for the screening of surface waters for MCs, especially where advanced laboratory infrastructure (e.g., LC-MS/MS) is not easily available. However, they must be considered as quantitative screening techniques for the detection of cyanotoxins [28,29].

Molecular analysis (qPCR) confirmed the presence of cyanobacteria (16s rRNA gene) in all the water samples collected. The gene *mcyE* which is associated with the production of MCs was found in 3 samples, i.e., samples 2 (S2, October 2019), 4 (S1, December 2019 and 16 (S3, October 2020), where the presence of MCs was also confirmed by LC-MS/MS, ELISA and PPIA. The gene responsible for the production of CYN (*cyrA*) was not detected in any of the samples (Table 3). This is in agreement with LC-MS/MS analysis where CYN was also not detected (Table 1).

Although significant concentrations of target MCs were detected in water samples, MCs were not detected in fish samples (flesh or liver). Previous studies also referred to cases where very low concentrations or total absence of MCs were reported in fish tissues collected during toxic cyanobacterial blooms [30,31]. It was shown that concentrations of MCs in fish tissues is strongly affected by local bloom conditions, fish feeding habits and ecosystem characteristics [32]. The occurrence of blooms and the presence of MCs in water are temporally and spatially variable, leading to different exposure of fish to MCs [33,34]. Inter-species differences in fish are also important factors for the intake and accumulation of MCs in their tissues. For example, zooplanktivorous fish were more likely to accumulate MCs than fish of other feeding guilds e.g., omnivorous, such as the studied *Cyprinous Carpio* [35]. The mechanisms of MC excretion and the timescales for MC elimination can also differ considerably among different fish species [36].

In the course of this study, *Microcystis aeruginosa* and *Aphanizomenon ovalisporum* were almost unique cyanobacterial species that were observed in autumn/spring and summer, respectively. These results are in agreement with a previous study reporting *Microcystis aeruginosa* and *Aphanizomenon ovalisporum* to be the most frequently encountered bloom-forming species in Lake Karaoun, either separately or together [14]. Furthermore, low phytoplankton biodiversity in Lake Karaoun was reported by Fadel and Slim who found that, since 2009, these two species had constituted >95% of the total phytoplankton biomass [16]. It was also reported that in Lake Karaoun, *Aphanizomenon ovalisporum* blooms formed both in spring and autumn, while *Microcystis aeruginosa* blooms formed at higher water temperatures observed during the summer months [15]. Although, *Aphanizomenon ovalisporum* is a well-known CYN producer, CYN was not detected during this study. On the contrary, the presence of CYN was previously reported during May to November 2012 and March to May 2013. The determination was performed with ELISA showing concentrations ranging from 0.5 to 1.7 μg/L in 2012, and from undetected to 1.7 μg/L in 2013 [17]. However, to date, the detection of CYN has never been confirmed by LC-MS/MS.

Monitoring studies of lakes and reservoirs in the Middle East are rare, with the exception of the closest natural lake, Lake Kinneret (Israel). Although Lake Kinneret is less eutrophic than the Karaoun reservoir, blooms of *Microcystis sp.* have been reported since the late 1960s [37], while *Aphanizomenon ovalisporum* and *Cylindrospermopsis raciborskii* were first detected in September 1994 and summer 1998, respectively [38,39]. The most frequently found cyanotoxins in Lake Kinneret have been CYN and MCs [21]. Despite the fact that CYN was not detected in this study, it is proposed that CYN should be included, together with MCs, in future monitoring programs to better evaluate the associated risks, since *Aphanizomenon ovalisporum* blooms continue to occur in lake Karaoun. NOD is mostly associated with brackish water cyanobacteria with a limited number of studies to report its occurrence in freshwater bodies [4,40]. Therefore, its presence is generally not expected in Lake Karaoun, while it can be determined along with MCs due to its structural similarity with them [41] in future monitoring programs. The presence of ATX should further be investigated, in cases where known cyanobacteria producers are present [4].

Screening for volatile compounds showed the presence of T&O cyanobacterial/algal terpenoids β-cyclocitral (in 3 samples) and β-ionone (in 3 samples). In particular, the presence of β-cyclocitral was reported to be associated with the occurrence of strains of *Microcystis* [42,43] which has a tobacco-woody odor with a rather high odor threshold concentration of 5 μg/L [44]. It was shown that β-cyclocitral is rapidly produced upon *Microcystis* cell rupture via a carotene oxygenase reaction and it subsequently affects grazer behavior, acting as a repellent and signal of low-quality food to grazers [43]. β-Ionone, commonly occurring in algae and higher plants, is also produced through the carotenoid cleavage of dioxygenases [45]. It has a characteristic flowery-woody odor, showing strong odor thresholds, at a level of 0.007 μg/L [46]. The release of β-ionone in lake water was positively correlated with microcystis biomass [47]. The common cyanobacteria/actimomycetes T&O compounds, geosmin and 2-methylisoborneol were not detected in any of the samples.

Nonanal, a known cyanobacterial/algal aldehyde with fishy odor [24] and an odor threshold of 1 μg/L in water [48] was detected in 9 samples. Aldehydes are mostly derivatives of polyunsaturated fatty acids and are common causes of odor in surface waters [18]. Dimethyl disulfide, dimethyl trisulfide and dimethyl tetrasulfide were detected in sample 4 that had a strong swampy/septic odor. Such undesirable odors have been a major concern for drinking water in several countries. In particular, dialkyl sulfides present strong odors described as swampy/septic, rotten, rancid and stinky, with very low odor threshold concentration levels of ng/L or less [49–51]. These and other organosulfur compounds can be produced both under oxic and anoxic conditions by a diversity of biota, biochemical pathways, enzymes and precursors [52]. 3-methylindole (skatole) was detected in samples 2 and 4. In particular, sample 4 had a strong swampy/septic odor due to the presence of dimethyl sulfides and skatole. The latter is a fecal compound with a characteristic odor that was previously reported to occur in algal cultures and field samples [18].

Several species of cyanobacteria can produce cyanotoxins and T&O compounds. For example, some strains of *Microcystis* produce microcystins together with β-cyclocitral and alkyl sulfides [42]. However, cyanobacterial T&O do not inevitably indicate the occurrence of cyanotoxins [53]. Nevertheless, since T&O can be sensed by the human nose at very low concentrations, they can serve as an early warning indicator for further investigations into the presence of toxic cyanobacteria [54]. The cyanobacterial/algal T&O detected in this study can serve as an initial list of compounds to be screened in future T&O episodes in Beirut's drinking water supplies.

The hydrocarbons detected, such as 2,2,4,6,6-pentamethyl heptane (all samples) and straight-chain hydrocarbons hexadecane (2 samples) and heptadecane (6 samples), could generally be of anthropogenic or biogenic origin. The hydrocarbon 2,2,4,6,6-Pentamethyl heptane has many industrial uses in anti-freeze products, coatings, fillers, lubricants, greases, etc. [55]. On the other hand, cyanobacteria and algae have long been known producers of alkanes, and their potential for biofuel production has been an area of increased research interest [56]. A study of volatile compounds associated with cyanobacteria and algae in freshwater by Juttner et al., [42] reported that biogenic hydrocarbons were represented exclusively by straight-chain components. Other compounds detected, such as 2,4-di-*tert*-butylphenol (antioxidant), diisobutyl phthalate (plasticizer) and p-cresol are common industrial pollutants, with uses in fuels, plastics, and production of chemicals. These were identified in samples 10, 4 and 2, respectively.

Quantitative determination of VOCs showed the presence of dichloromethane in all samples, at concentrations up to 34 μg/L. Dichloromethane is a common industrial solvent used in many chemical processes. Toluene was detected in the majority of samples at concentrations of up to 0.81 μg/L, while in sample 16 it co-occurred with traces of benzene and o,m,p-xylenes. These compounds are found in fuels and petroleum products and are also common industrial solvents. The "BTEX" volatiles (benzene, toluene, ethylbenzene, xylenes) are frequently used as an index to assess the impact of pollution caused by spills or leaks from fuel storage tanks into water bodies. Trichloromethane, detected in two samples at concentrations of up to 0.5 μg/L could indicate contamination of the lake with wastewater, since trihalomethanes, are commonly present in chlorine-treated water [57].

The above findings imply that, besides cyanobacteria and their metabolites, cyanotoxins and T&O compounds, anthropogenic pollution can also be a concern for Lake Karaoun, regarding its use as a drinking water reservoir, supporting the emerging need for studies and impact assessment of the co-occurrence of toxic cyanobacteria with other anthropogenic pollutants [58].

### **4. Conclusions**

The results of this study demonstrate for the first time the presence of multiple MC congeners in Lake Karaoun, i.e., dmMC-RR, MC-RR, MC-YR, dmMC-LR, MC-LR, MC-HilR, MC-WR, MC-LA, MC-LY, MC-LW, and MC-LF, with total MCs reaching up to 429 μg/L (sample 16). Additionally, T&O compounds such as β-cyclocitral, β-Ionone, nonanal and dimethylsulfides were identified, while industrial pollutants of anthropogenic origin including dichloromethane and toluene were determined up to 34 (sample 9) and 0.81 (sample 16) μg/L, respectively. Complementary methods were applied aiming at the reliable determination of cyanotoxins and T&O compounds in the lake. Since blooms of *Microcystis aeruginosa* and *Aphanizomenon ovalisporum* continue to occur in Lake Karaoun, monitoring of cyanobacterial blooms is necessary in the future, for the assessment of risks related to the intended uses of the lake. With regard to the cyanotoxin analysis, the results show that MCs should be a monitored target, especially when blooms of *Microcystis aeruginosa* occur. Quantitative screening by ELISA or PPIA could be used for the monitoring of MCs if LC-MS/MS facilities and expertise are not available. It is, however, recommended to confirm the findings by LC-MS/MS which has the advantage of being compound-specific. Despite the fact that CYN was not detected in this study, CYN should be included in the ongoing monitoring schemes, especially in the presence of *Aphanizomenon ovalisporum* in the lake. Whenever incidents of unpleasant odors occur in the lake or in water supplies, further analysis of T&O and VOCs should be carried out to identify the source of T&O.

### **5. Materials and Methods**

### *5.1. Study Site and Sampling*

The Middle-Eastern Lake Karaoun (Qaraoun) is the largest water body in Lebanon, located in the western part of the valley Bekaa (33.34◦ N, 35.41◦ E) (Figure 4). Lake Karaoun is an artificial reservoir, created in 1959 by construction of a dam on the Litani River. It has a surface of 12 km<sup>2</sup> at a full capacity of 220 million m3, a maximum depth of 60 m and a mean depth of 19 m [16]. The lake is considered to be located in a vulnerable arid to semi-arid zone, with winters being moderately cold (13 ◦C average temperature), the wet season extending from November to April, and summers being hot and dry, lasting from July to October [15,59].

During field campaigns that covered the wet and dry periods, 16 water samples along with a total of 9 specimens of *Cyprinus carpio* fish (weighing 215 g on average) were collected mainly by the national Litani River Authority. All water samples were collected in polyethylene bottles from 5 sampling points of Lake Karaoun (S1, S2, S3, S4 and S5). S1 was located close to the river input at the west bank, S2 was east of the dam, S3 was close to the river input at the east bank, S4 was in the middle of the lake and S5 was at the dam (Figure 4). Water samples were transported in coolers (at a temperature of 4 ◦C) for laboratory analysis in Beirut, Lebanon and Athens, Greece. The fish were kept on ice and were sacrificed within 24 h. Fish liver and muscle sub-samples were labeled and frozen separately at −20 ◦C. The frozen fish tissue samples were then lyophilized with a Labconco freeze-dryer for 48 h at −84 ◦C and 0.2 mbar and powdered using a pestle and a mortar.

**Figure 4.** Map and sampling points of Lake Karaoun, Lebanon.

### *5.2. Chemicals and Reagents*

Cyanotoxin standards of MC-RR, MC-LR, MC-YR, MC-LA and NOD were purchased from Sigma-Aldrich (Steinheim, Germany), [D-Asp3]MC-LR, [D-Asp3]MC-RR, MC-WR, MC-HtyR, MC-HilR, MC-LY, MC-LW and MC-LF from ENZO Life Science (Lausen, Switzerland), CYN from Abraxis (Warminster, PA, USA), and ATX fumarate from TOCRIS Bioscience (Bristol, UK). All toxin standards had purity >95%. A VOC 57 standard mix (44926-U) was purchased from Supelco (Darmstadt, Germany). β-Cyclocitral (C10H16O) (≥95.0%), β-ionone (C13H20O) (purity ≥ 97.0%), pentadecane (C15H32) (purity ≥ 99.8%), 2,4-di-*tert*-butylphenol (C14H22O) (99.0%), hexadecane (C16H34) (purity ≥99.8%), heptadecane (C17H36) (purity ≥ 99.5%), dimethyl disulfide (CH3SSCH3) (≥99.0%), dimethyl trisulfide (purity ≥ 98.5%) and 3-methylindole (C9H9N) (98.0%), were supplied by Sigma Aldrich (Steinheim, Germany). Acetonitrile (ACN) and methanol (MeOH) of HPLC grade (≥99.9%) as well as dichloromethane (DCM) and hexane (HXN) of analytical grade (99.9%) were supplied by Fisher Chemical (Loughborough, UK). High purity formic acid (HCOOH) (98–100%) and acetic acid (CH3COOH) (>99.8%) were obtained from Riedel-de Haën (Seelze, Germany). Sodium chloride (NaCl) of analytical purity (99.5%) and fuming (37%) hydrochloric acid (HCl) were purchased from Merck (Darmstadt, Germany). Ethylenediaminetetraacetic acid (EDTA) of analytical grade was supplied from Serva Electrophoresis. Sodium hydroxide (NaOH) 2M solution used for adjustment of pH was prepared from NaOH pellets (purity 98%) purchased from Sigma-Aldrich (Steinheim, Germany). Potassium carbonate (K2CO3) (>99.5%) was supplied from Carlo Erba Reagents. High purity water (18.2 MΩ cm at 25 oC) was produced in-house using a TEMAK TSDW10 water purification system (TEMAK, Athens, Greece).

### *5.3. Microscopic Examination*

Microscopic examination of samples was carried out within 24 h on 20 mL water samples collected during field campaigns with a phytoplankton net and kept at low temperature (4 ◦C). Examination was carried out in the LAEC laboratory in Beirut, Lebanon using a phase contrast microscope (Olympus, Munster, Germany), under a ×40 objective and ×100 immersion. Identification of cyanobacteria was based on taxonomic keys as cells structures and dimensions, mucilage features and the form of colonies. Estimation of cyanobacterial abundances was performed as described elsewhere [17].

### *5.4. LC-MS/MS Analysis of Cyanotoxins*

### 5.4.1. Sample Preparation of Water Samples

Preparation of water samples for LC-MS/MS analysis of cyanotoxins was based on the method of Zervou et al. [22]. For the determination of extracellular toxins, 150 mL of sample was filtered using a glass fiber filter (Millipore, Ireland). After adjusting the pH of the filtrate to 11, solid phase extraction (SPE) was performed using a dual cartridge assembly with a Supel-Select HLB (bed wt. 200 mg, volume 6 mL, Supelco) and a Supelclean ENVI-Carb (bed wt. 250 mg, volume 3 mL, Supelco) on a 12-port SPE vacuum manifold (Supelco) connected with a vacuum pump. Conditioning of cartridges was carried out with 6 mL DCM, followed by 6 mL of MeOH and 6 mL of pure water (pH 11). Sample passed through the two tandem cartridges at a 0.5 mL/min flow rate. After sample passing, cartridges were dried for 15 min (air under vacuum) and the sequence of cartridges in the assembly was reversed. Elution was done with (60:40) MeOH/DCM having 0.1% HCOOH. Eluents were evaporated to dryness under a gentle nitrogen stream, reconstituted with 150 μL of 5% (*v/v*) MeOH, and transferred into vials for LC-MS/MS analysis. For the determination of intracellular cyanotoxins, after sample filtration the filters were extracted, according to Chistophoridis et al. [23], with 9 mL 75% (*v/v*) MeOH. A 3-mL aliquot of the filtered supernatant was evaporated to dryness and the residue was re-dissolved with 500 μL of 5% (*v/v*) MeOH for LC-MS/MS analysis.

### 5.4.2. Sample Preparation of Fish Samples

Subsamples of 0.2 g of lyophilized powdered flesh or 0.25 g liver were extracted with 5 mL 80% MeOH containing 0.5% HCOOH by stirring for 15 min, followed by ultrasonication for 30 min (Bandelin Sonorex Super RK106). The mixture was then transferred to a Falcon tube and centrifuged for 15 min at 4500 rpm (DuPont RMC-14 Refrigerated Micro-Centrifuge, Sorvall Instruments, Newtown, CT, USA). The supernatant was extracted three times with 1 mL of hexane. The hexane phase was discarded, and the bottom layer was transferred to a flask and diluted with 100 mL of water containing 0.3% formic acid. The extract was cleaned-up by SPE with a Supel-Select HLB cartridge (bed wt. 200 mg, volume 6 mL, Supelco) conditioned with 6 mL MeOH followed by 6 mL acidified water (0.3% HCOOH). After extraction, the cartridge was washed with 6 mL of water, dried for 15 min under vacuum and eluted with 6 mL MeOH. The eluents were dried in a water bath at 40 ◦C under a gentle nitrogen stream. Reconstitution was carried out with 200 μL of 5% MeOH, followed by vortexing and sonication for 3 min, prior to LC-MS/MS analysis.

### 5.4.3. Determination by LC-MS/MS

A Finnigan Surveyor LC system, equipped with an AS autosampler (Thermo, Waltham, MA, USA) coupled with a TSQ Quantum Discovery Max triple-stage quadrupole mass spectrometer (Thermo, Waltham, MA, USA) with electrospray ionization (ESI) source, was used for LC-MS/MS analysis. Data was acquired and processed by Xcalibur software. Targeted analysis of CYN, ATX, NOD and 12 MCs (dmMC-RR, MC-RR, MC-YR, MC-HtyR, dmMC-LR, MC-LR, MC-HilR, MC-WR, MC-LA, MC-LY, MC-LW, MC-LF) was performed as described in a previous study [22]. Detection of CTs was carried out in multiple reaction monitoring (MRM) mode, using the three most intense and characteristic precursor/product ion transitions for each toxin. Confirmation of identity was based on criteria for retention time (tR), characteristic precursor/product ion transitions and two calculated ratios of precursor/product ion transitions. LC-MS/MS detection parameters of targeted cyanotoxins are given in Table S3. An example of MC-LR identification in a sample (S2, 2 October 2019) from Lake Karaoun is shown in Figure S4.

### *5.5. ELISA for Microcystins*

ELISA was carried out with the Microcystins-ADDA ELISA kit (Eurofins—Abraxis, Warminster, PA, USA) in 96 well microplates, using an Infinite M200 reader (Tecan, Männedorf, Switzerland). The kit was used according to the manufacturer's instructions and concentrations of toxins were calculated using calibration curves based on the absorbance at 450 nm. For the determination of extracellular MCs, water samples were filtered through 47 mm glass fiber filters (Millipore) and analyzed without any further treatment. Dilutions with ELISA sample diluent were carried out when samples exceeded 5 μg/L MC-LR equivalents. For the analysis of intracellular MCs, after sample filtration (see 5.4.1) the filter was extracted with 9 mL of 75% MeOH. Then, 3 mL of the extract was evaporated to dryness and the residue was re-dissolved in water to avoid false results due to the high percentage of MeOH [41]. Finally, samples were analyzed in duplicate and mean values were reported when RSD < 25%.

### *5.6. Protein Phosphatase Inhibition Assay (PPIA) for Microcystins*

PPIA was carried out using the Microcystins/Nodularins PP2A kit (Eurofins—Abraxis) in 96 well microplate using an Infinite M200 reader (Tecan, Männedorf, Switzerland) for measurements at 405 nm, according to manufacturer's instructions. Dilutions with ultrapure water were carried out when samples exceeded 2.5 μg/L MC-LR equivalents. Sample preparation prior to PPIA was the same as in ELISA. Samples were analyzed in duplicate and mean values were reported when RSD < 25%.

### *5.7. qPCR Assay for Total Cyanobacteria and Cyanotoxin Genes*

The Phytoxigene™ CyanoDTec (Diagnostic Technology, Belrose, Australia) multiplex quantitative real-time PCR assay was applied to determine the gene copies of the 16s rRNA gene (total cyanobacteria) and the *mcyE*, *cyrA*, *sxtA* genes (microcystins, cylindrospermopsin, saxitoxins, respectively). The kit was used according to the manufacturer's instructions and PCR was carried out in a Smartcycler II system (Cepheid). In brief, a volume of water sample (1 to 15 mL, depending on visual cell density) was filtered through a Nucleopore 25 mm, 0.8 μm filter (Whatman, Little Chalfont, UK) using a syringe and filter holder. DNA extraction of filters was carried out using BioGx bead lysis tubes (Diagnostic Technology) in a BeadBug bead beater (Benchmark Scientific). Extracts were centrifuged (MicroCL 21, Thermo Fisher Scientific, Waltham, MA, USA) and proceeded to qPCR. Quantitation of gene copies was based on calibrations with the Phytoxigene™ CyanoNAS standards (100–1,000,000 copies per reaction).

### *5.8. GC-MS Analysis of Volatile Compounds*

Analysis of volatile compounds was carried out in order to (a) screen samples to detect and identify typical cyanobacterial volatile and T&O compounds (untargeted method) and (b) to quantitatively determine a range of anthropogenic VOC pollutants (targeted method).

### 5.8.1. Untargeted Screening of Cyanobacterial Volatiles and T&O Compounds

Headspace Solid Phase Microextraction coupled to Gas Chromatography-Mass Spectrometry (HS-SPME-GC-MS) was used to screen water samples for volatile compounds with focus on cyanobacterial T&O. A 456 GC coupled to TQ mass spectrometer and equipped with an SPME autosampler (Bruker Daltonics, Bremen, Germany) was used. A (2–10 mL) aliquot of the sample was transferred into a 20 mL SPME glass vial containing 3 g NaCl, then adjusted to 10mL with ultrapure water and tightly sealed. HS-SPME was carried out automatically under the following conditions: 2 cm Divinylbenzene/Carboxen/Polydimethylsiloxane SPME fiber (Supelco, Bellefonte, PA, USA), equilibration 10 min at 60 oC, headspace extraction: 10 min at 60 oC, agitation 300 rpm and desorption time 2 min. GC analysis was carried out using a column RXI®- 5 Sil MS, 30 m, 0.25 mm ID, 0.25 μm df (Restek, Bellefonte, PA, USA). GC conditions were injector temp. 250 ◦C, splitless, constant flow 1 mL/min (He), column program: (a) 50 ◦C (1 min) to 250 ◦C at a rate of 15 ◦C/min, 250 ◦C (5 min) and (b) 35 ◦C (5 min) to 250 ◦C at a rate of 8 ◦C/min, 250 ◦C (5 min). MS conditions were: EI source (70 eV), scan 30–300 m/z, positive polarity. Fluorobenzene standard solution in methanol (Sigma – Aldrich, St. Louis, MO, USA) spiked at a concentration of 5 μg/L in the samples was used as a surrogate for evaluation of SPME efficiency.

Mass spectrometry data were processed using MSWS software (Bruker). Mass spectral deconvolution and identification of compounds was carried out with AMDIS (NIST, Gaithersburg, MD, USA) using the NIST MS library (2015) and retention index calibration with C7-C30 saturated alkanes reference standard (Sigma). Combined matching scores (retention index-RI and mass spectral matching) were used as criteria for identification (RI ± 20 of the reference RI and a spectral matching ≥80%). Identification was considered definite only if a reference standard of the suspect compound was available in the lab and retention times of the suspect and reference compounds matched within ±0.01 min, in addition to mass spectral matching.

### 5.8.2. Quantitative Determination of VOCs

Quantitative determination of a number of VOCs was carried out by HS-SPME-GC-MS according to EN ISO 17943:2016 (HS-SPME-GC-MS), using 5 ml samples [60]. The compounds determined were chloroform, bromoform, dibromochloromethane, bromodichloromethane, benzene, 1,2-dichloroethane, tetrachoroethene and trichloroethene, 2-ethoxy-2-methyl propane (ETBE), bromochloromethane, dibromomethane, toluene, 1,2 dibromoethane, chlorobenzene, styrene, bromobenzene, n-butyl benzene, hexachlorobutadiene, naphthalene, 1,1-dichloroethane, 1,1-dichloroethene, 1,1,2-trichloroethane, 1,1,2,2 tetrachloroethane, 1,2-dibromo-3-chloropropane, 1,2-dichloropropane, 1,2,3-trichlorobenzene, 1,2,4-trichlorobenzene, 1,3-dichloropropane, 2-methoxy-2-methyl-butane (TAME), ethylbenzene, n-propylbenzene, o-xylene, p+m-xylenes, sec-butylbenzene and tert-butylbenzene. A VOC 57 standard mix (Supelco) was used for calibration and fluorobenzene, toluene-d8 and benzene-d6 (Sigma) as internal standards. HS-SPME conditions were: 2 cm Divinylbenzene/Carboxen/Polydimethylsiloxane SPME fiber (Supelco), equilibration for 10 min at 40 ◦C, headspace extraction for 10 min at 40 ◦C, agitation 300 rpm and desorption time 2 min. GC conditions were: column RXI®- 624 Sil MS, 60 m, 0.32 mm ID, 1.8 μm df (Restek), injector temp. 250 ◦C, splitless, constant flow 1ml/min (He), column program: 35 ◦C (5 min) to 250 ◦C at a rate of 8 ◦C/min and at 250 ◦C (10min). MS conditions were: EI source (70 eV), Selected Ion Monitoring (3 ions per compound), positive polarity. LOD of each compound was 0.2 μg/L. Confirmation of determinations was carried out with the ISO criteria for retention time and ion ratios [60].

### *5.9. Method Validation and Quality Control Procedures*

The methods applied in this study were validated either previously or in the frame of the study and method performance parameters such as specificity, linearity, precision, accuracy, and limits of detection have been assessed. Furthermore, quality control procedures

including measurements of blank/negative and control/positive samples were followed in each batch of analyzed samples.

Validation data of the method applied for the determination of CTs in water by LC-MS/MS are reported in previous studies with RSD values of detected CTs being <16% and <26% for extra- and intra-cellular fractions, respectively [22,23]. This method is also accredited by ISO 17025 in the NCSR Demokritos laboratory [61].

The LC-MS/MS method for determination of CTs in fish was developed, optimized and validated in the frame of this study. For optimization, various combinations of extraction solvents, extraction/mixing times, clean-up steps and SPE cartridges were tested (Oasis HLB and HLB followed by Sep-Pak Vac, silica) as presented in Figures S5 and S6. Optimization experiments were performed with samples spiked with MC-LR and MC-RR at concentration levels of 100 and 400 ng/g dw for flesh and liver, respectively. Selection of the optimized conditions was based on maximization of % recoveries. As shown in Figure S7, method 2 provided the highest mean recoveries for the two spiked MCs in flesh (78.9% for MC-RR and 79.1% for MC-LR) and liver (77.1% for MC-RR and 75.4% for MC-LR). Subsequently, the optimized method was validated in-house. Recovery and precision were evaluated by analyzing toxins-free lyophilized flesh and liver fish samples spiked with a mixture of 12 MCs, at concentration levels of 100 and 400 ng/g dw in 3 replicates. Samples were extracted and analyzed as outlined in Section 5.4.2. Table S4 provides performance characteristics of this method. Briefly, mean recoveries of [D-Asp3]MC-RR, MC-RR, MC-YR, [D-Asp3]MC-LR, MC-LR, MC-HilR, MC-LA and MC-LY ranged from 68.5% to 81.6% for flesh, while liver recoveries ranged from 61.5% to 72.2%, with intra-day precision in the range of 6.8–16.5% for flesh and 5.5–15.2% for liver. LODs for fish flesh and liver samples ranged from 1.0 to 7.0 ng/g dry weight and from 0.8 to 5.6 ng/g dry weight, respectively.

In-house validation and method performance data of ELISA and PPIA for MCs in water were reported previously [41] and methods have been proven suitable for quantitative screening of MCs. Negative and positive control samples were included in each analysis batch.

The qPCR assay for total cyanobacteria and cyanotoxin genes included evaluation of negative and positive control samples in each batch of samples. The Phytoxigene™ CyanoNAS standards used for calibration and accurate quantitation were commissioned and developed by the National Measurement Institute (NMI), of the Australian Department of Industry which participates in the International Bureau of Weights and Measures (BIPM) Consultative Committee for Amount of Substance (CCQM).

The untargeted HS-SPME-GC-MS screening method for detection and identification of cyanobacterial volatile compounds has been tested with a wide range of compounds that are available in the lab of EYDAP SA and has been shown to be capable of detection/identification at concentrations generally <1μg/L. The method has also been tested successfully in interlaboratory trials for unknown odorous compounds in water. The targeted HS-SPME-GC-MS method for VOCs in water has been fully validated in-house, is accredited by ISO 17025 and has successfully been evaluated in interlaboratory tests.

**Supplementary Materials:** The following are available online at https://www.mdpi.com/article/ 10.3390/toxins13100716/s1, Figure S1. chemical structures of studied cyanotoxins, Figure S2. LC-MS/MS MRM chromatogram of sample S2, 02/10/2019 (intracellular fraction) from Lake Karaoun, Figure S3. MRM chromatogram of (a) a standard solution of 12 MCs at a concentration level corresponding to 50 ng/g dw and (b) liver sample from *Cyprinous Carpio* fish collected in September 2019, Figure S4. Example of identification of MC-LR in sample S2, 2 October 2019 (intracellular fraction) from Lake Karaoun, Figure S5. Experimental procedures tested in order to optimize the extraction of MCs in fish flesh, Figure S6. Experimental procedures tested in order to optimize the extraction of MCs in fish liver, Figure S7. Selection of method for (a) fish flesh and (b) fish liver: obtained recoveries of spiked MCs using different treatment processes; Table S1. Method LODs and LOQs for each cyanotoxin analysed in this study using LC-MS/MS, Table S2. Results of untargeted HS-SPME-GC/MS screening per sample, Table S3. LC-MS/MS detection parameters of targeted cyanotoxins, Table S4. Performance characteristics of the method for the analysis of target MCs in fish flesh and liver. Table S5. Physicochemical parameters of Lake Karaoun.

**Author Contributions:** Conceptualization, R.L., A.H., K.S.; Data curation, N.A.H., S.-K.Z., T.K., C.C., A.P., A.F.; Formal analysis, N.A.H., S.-K.Z., T.K., C.C., A.P., A.F.; Funding acquisition, R.L.; Investigation, R.L., J.S., T.K., T.M.T., K.S., A.H.; Methodology, T.K., S.-K.Z., A.H.; Project administration, K.S., A.H., R.L.; Resources, R.L., J.S., T.K., A.H.; Software, S.-K.Z., T.K., A.P.; Supervision, A.H., R.L.; Validation, S.-K.Z., T.K. and T.M.T.; Visualization, S.-K.Z. and T.M.T.; Writing—original draft, N.A.H., S.-K.Z., T.K. and A.H.; Writing—review & editing, S.-K.Z., T.K., T.M.T., A.F., K.S., J.S., R.L. and A.H. All authors have read and agreed to the published version of the manuscript.

**Funding:** Not applicable.

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** Data is contained within the article or Supplementary Material.

**Acknowledgments:** N.A.H. thanks the National Council for Scientific Research of Lebanon (CNRS-L) and E2S UPPA (France) for granting a doctoral fellowship. The authors acknowledge COST Actions ES 1105 "CYANOCOST–Cyanobacterial blooms and toxins in water resources: Occurrence impacts and management" www.cyanocost.net and CA 18225 "WaterTOP–Taste and Odor in early diagnosis of source and drinking Water Problems" https://watertopnet.eu for adding value to this study through networking and knowledge sharing with European experts in the field. S.-K.Z. acknowledges the Action titled "National Network on Climate Change and its Impacts—Climpact", which is implemented under the sub-project 3 of the project "Infrastructure of national research networks in the fields of Precision Medicine, Quantum Technology and Climate Change", funded by the Public Investment Program of Greece, General Secretary of Research and Technology/Ministry of Development and Investments.

**Conflicts of Interest:** The authors declare no conflict of interest.

### **References**


### *Article* **Is Toxin-Producing** *Planktothrix* **sp. an Emerging Species in Lake Constance?**

**Corentin Fournier 1,†, Eva Riehle 2,†, Daniel R. Dietrich 2,\* and David Schleheck 1,3,\***


**Abstract:** Recurring blooms of filamentous, red-pigmented and toxin-producing cyanobacteria *Planktothrix rubescens* have been reported in numerous deep and stratified prealpine lakes, with the exception of Lake Constance. In a 2019 and 2020 Lake Constance field campaign, we collected samples from a distinct red-pigmented biomass maximum below the chlorophyll-a maximum, which was determined using fluorescence probe measurements at depths between 18 and 20 m. Here, we report the characterization of these deep water red pigment maxima (DRM) as cyanobacterial blooms. Using 16S rRNA gene-amplicon sequencing, we found evidence that the blooms were, indeed, contributed by *Planktothrix* spp., although phycoerythrin-rich *Synechococcus* taxa constituted most of the biomass (>96% relative read abundance) of the cyanobacterial DRM community. Through UPLC–MS/MS, we also detected toxic microcystins (MCs) in the DRM in the individual sampling days at concentrations of ≤1.5 ng/L. Subsequently, we reevaluated the fluorescence probe measurements collected over the past decade and found that, in the summer, DRM have been present in Lake Constance, at least since 2009. Our study highlights the need for a continuous monitoring program also targeting the cyanobacterial DRM in Lake Constance, and for future studies on the competition of the different cyanobacterial taxa. Future studies will address the potential community composition changes in response to the climate change driven physiochemical and biological parameters of the lake.

**Keywords:** *Planktothrix*; *Synechococcus*; microcystins; temperate lakes

**Key Contribution:** Deep water, red-pigmented cyanobacterial blooms in Lake Constance were characterized by phylogenetic community sequencing during the summers of 2019 and 2020, demonstrating the low abundance of *Planktothrix* spp., and the predominance of picoplanktic phycoerythrin-rich *Synechococcus* spp., as well as very low concentrations of cyanobacterial toxins on individual sampling days. In response to climate change, changes in the physiochemical and biological parameters of the lake may, in future, support the establishment of toxic *Planktothrix* spp. blooms and/or the mass development of potentially toxic picocyanobacteria.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

### **1. Introduction**

The formation of cyanobacterial blooms involves the complex interplay of regional and biological variables, and blooms have been reported worldwide with an increasing frequency [1,2]. Although it has been shown that climate change and eutrophication are, in many cases, major contributors to bloom formation, the mechanisms through which nutrients and temperature interact to amplify blooms varies extensively between cyanobacterial groups. Moreover, the optimal growth temperature and nutrient availability of cyanobacteria is species specific, making inferences from studies with other species more complex [2,3]. Indeed, different cyanobacterial species predominate depending on the N/P ratio [3]. The

**Citation:** Fournier, C.; Riehle, E.; Dietrich, D.R.; Schleheck, D. Is Toxin-Producing *Planktothrix* sp. an Emerging Species in Lake Constance? *Toxins* **2021**, *13*, 666. https://doi.org/ 10.3390/toxins13090666

Received: 22 July 2021 Accepted: 15 September 2021 Published: 17 September 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

filamentous cyanobacterial genus, *Planktothrix*, occurs preferentially in prealpine and alpine lakes in temperate regions, and has been responsible for many blooms in the past (for example, in Lake Zurich [4,5], Lake Bourget [6], Lake Garda [7] and Lake Mondsee [8]), where its mass occurrence can evolve to become a major influence on the food web [9,10]. The success of *Planktothrix* spp. is attributed to its adaptability, including its ability to regulate buoyancy, and the use of phycobilins, in addition to chlorophyll-a [11,12], as chromophores. By virtue of the phycobilins, phycocyanin and phycoerythrin, *Planktothrix* spp. can absorb light in large parts of the electromagnetic spectrum; specifically, blue and green light, thus conferring its red appearance. Consequently, the free-floating *Planktothrix* spp. usually develop blooms at depths of 9–15 m [11,13]. Bloom depths within a stratified lake can be influenced by the internal waves, and can impact the growth of *Planktothrix* spp. by changing the light availability [14]. Similarly to many other cyanobacterial species, mass occurrences of *Planktothrix* spp. are supported by warmer water temperatures, mediating a more stable lake stratification in the summer [9,15]. Strikingly, *Planktothrix* spp. can, additionally, form prominent winter blooms and can thrive even beneath an ice cover, allowing for full year dominance [16,17]. Although microcystin production in winter blooms, on and under ice covers, has been reported, toxins appear to be less abundant in colder environments than in warmer environments [16,18,19].

The genus *Planktothrix* is currently distinguished in nine species, including red and green phenotypes, with the most prominent representatives being *P. rubescens* and *P. agardhii*, respectively, which mostly occur in the freshwater ecosystems of temperate regions [20]. Similarly to many other cyanobacteria, *Planktothrix* spp. are capable of producing microcystins (MCs), which are known to be toxic to humans, as well as other mammalian and non-mammalian species [15–18]. Hence, cyanobacterial blooms are often associated with the detection of increased intra- and extra-cellular toxin(s) [21]. Despite a plethora of efforts, neither the trigger for toxin production, nor the factors resulting in the development of toxic cyanobacterial blooms, have been elucidated. While toxins are suggested to be part of a defense mechanism against zooplankton or parasites [22–24], and toxin producing strains seem to have an advantage over non-toxic strains [25,26], the molecular 'switch' that turns toxin production on has not yet been discovered. Irrespective of the latter, the potential adverse impact of toxic cyanobacterial blooms on human health, society, economy and ecology highlights the importance of an improved understanding of cyanobacterial bloom formation, with or without concomitant toxin production [27,28].

Microcystins (MCs) share one common monocyclic structure with a molecular weight of approximately 1 kDa, which is composed of seven amino acids, (three *D*-amino acids, one *N*-methyldehydroalanine, two *L*-amino acids at the hypervariable positions two and four, and the unique amino acid ADDA (Figure S1) [29]). Variations in amino acid composition, and modifications such as methylations, create extreme structural diversity—at least 279 different congeners have been reported [30]. MC synthesis is encoded by 9–10 genes that constitute one *mcy* gene cluster [31]. The toxicity of MCs is induced by covalent binding, the inhibition of ser/thr protein phosphatases and the concomitant hyperphosphorylation of cellular proteins [32–34], whereby the toxicodynamically relevant biological availability of MCs highly depends on the route of exposure and the respective MC congener [35,36]. MC concentrations of <35 μg/L were reported in *Planktothrix* spp. blooms [37,38]. When considering the current World Health Organization (WHO) guideline value of 1 μg MC/L in drinking water [39], which translates to approximately 10 μg/L of raw water, largely depending on the type of water treatment used [40,41], it becomes obvious that toxinproducing *Planktothrix* spp. blooms must be taken as a serious threat to freshwater systems serving as drinking water resources.

Unlike other prealpine lakes, Lake Constance, a drinking water reservoir for more than four million inhabitants, has not yet seen prominent and recurring, well-documented blooms of *Planktothrix* spp. Although *P. rubescens* has the capability to dominate entire lake ecosystems, even at low nutrient concentrations (e.g., Lake Bourget [6]), analyses of Lake Constance samples revealed only low abundances of *Planktothrix* spp. to date [9,42]. Indeed, the first prominent appearance of *Planktothrix rubescens* was documented in 2016, when *P. rubescens* filaments were observed in various German, Austrian, and Swiss sampling sites at Lake Constance [42,43].

No molecular phylogenetic studies have been conducted, to date, to evaluate the composition of the red-pigmented cyanobacterial community in Lake Constance, in water depths that are typical for the blooms of *Planktothrix* spp. (i.e., below the chlorophyll-a maximum). In a field campaign in the Überlingen embayment of Lake Constance (47.7571◦ N 9.1273◦ E), we observed reddish colored plankton filters in the samples taken at 18 m water depth, in the summer of 2019. Subsequently, water samples were taken at the depths of maximal red pigment (phycoerythrin) concentration, as determined by a fluorescence probe, on each of the fortnightly sampling days, between June and October 2019, and between July and September 2020. The composition of the cyanobacterial community at this deep water red pigment maximum (DRM) was assessed through Illumina PCR-amplicon sequencing, using cyanobacteria-specific primers, while UPLC–MS/MS analyses were applied to determine the toxin concentrations. Potentially toxin-producing species were identified and quantified via quantitative PCR, using *Planktothrix*- and microcystin-biosynthesis-specific (*mcy*) primers, and Sanger sequencing.

### **2. Results**

### *2.1. Blooms of Red-Pigmented Phytoplankton at Water Depths below the Chlorophyll-a Maximum in Lake Constance*

Multiwavelength fluorometer profiles were taken along the water column, using a Moldaenke FluoroProbe, and were evaluated directly on the ship to determine the depths of the maxima of green pigment (chlorophyll-a maximum; predominantly diatoms and green algae) and red pigment concentrations (deep water red pigment maximum, DRM). An example of a depth profile for all of the recorded fluorescence channels, as well as of the interpretation of the abundance of different algae classes (as calculated by the Moldaenke FluoroProbe) is depicted in Figure 1. An increased abundance of redpigmented cyanobacteria, such as phycoerythrin-rich *Planktothrix* spp. or *Synechococcus* spp., was suggested by the elevated fluorescence intensity at 570 nm (and 525 nm) excitation wavelength (Figure 1). According to the distinction of algae classes used by the Moldaenke FluoroProbe, these phycoerythrin-rich algae were attributed to represent 'cryptophyta'.

**Figure 1.** Representative depth profile recorded with a Moldaenke FluoroProbe multichannel fluorimeter, indicating the high abundance of red-pigmented biomass at a water depth below the chlorophyll-a maximum in Lake Constance on 1 July 2019. (**A**) Absolute fluorescence intensities recorded at the different excitation wavelengths. (**B**) Abundance of the different algae classes as attributed by FluoroProbe. Red-pigmented cyanobacteria are attributed to 'cryptophyta'. In this example, the chlorophyll-a maximum was determined at 8–10 m water depth (Figure 1A, 470 nm), and a second maximum, indicating a red-pigmented biomass, at approximately 18–20 m water depth (Figure 1A, 570 nm; 1B, cryptophyta).

### *2.2. Phycoerythrin-Rich Synechococcus Phylotypes Dominated the DRM Cyanobacterial Community*

In order to characterize the cyanobacterial community composition that is presumably present at the DRM, the plankton biomass at the DRM was collected on Whatman GF6 glass fiber filters. The filters appeared reddish, compared to the yellow-green filters obtained from the chlorophyll-a maximum sample (see Supplementary Material, Figure S2). The total DNA was extracted from the filters, and a fragment of the 16S rRNA gene was amplified using the cyanobacteria-specific primers, CYA359F and CYA784R [44]. Illumina sequencing, with 300 bp paired-end reads, was employed. These primers amplified cyanobacteria phylotypes, which allowed for the collection of phylogenetic information at a finer resolution, and also of the low abundant cyanobacteria at the DRM. Taxonomic affiliation was carried out using two different reference databases: SILVA\_138 and Greengenes. The cyanobacteria taxonomy was consistent between both databases, with the exception of the *Synechococcus* genus in Greengenes (replaced by *Cyanobium*\_PCC-6307 in SILVA; *Cyanobium*\_PCC-6307 is a heterotypic synonym of *Synechococcus* sp PCC-6307).

Subsequently to bioinformatic processing and the removal (filtering) of the extremely low abundant phylotypes (i.e., phylotypes represented by less than three reads in at least 20% of all samples; see Material and Methods), 35 amplicon sequence variants (ASVs) were detected in 2019, and 37 were detected in 2020 (Figure 2). Each ASV was affiliated at the level of either genus or order, depending on the last common taxonomic rank between the SILVA and Greengenes databases (note that species rank could not be affiliated by the amplicon sequencing technique that we used).

For both the 2019 and 2020 sampling campaigns, the genus *Synechococcus* clearly dominated the cyanobacterial DRM community (as examined using amplicon sequencing), occupying 96% (SILVA) or 98% (Greengenes) of the total relative read abundance, and representing 65.7% (SILVA) and 67.6% (Greengenes) of the detected ASVs in the community (see Supplementary Material, Table S2). For 2019, five *Synechococcus* ASVs represented 78% of the total relative abundance, with ASV13 being the most abundant with 21% total relative abundance (Figure 2A). For 2020, only three of the ASVs affiliated to *Synechococcus* contributed to 77% of the total relative abundance, with ASV4 contributing to almost half (45%) throughout the year (Figure 2B).

Although *Synechococcus* dominated the cyanobacterial community at the DRM in 2019 and 2020, each of the two ASVs that are affiliated to *Planktothrix* could be detected in both years. *Planktothrix* (Oscillatoriophycideae in Figure 2A,B), with 0.4% (2019) and 0.9% (2020) of the total relative abundance (Figure 2A,B), represented only low abundant taxa, together with Nostocophycideae. In addition, two *Microcystis* spp. ASVs were detected in 2019, at 0.06% of the total relative read abundance. Only one *Microcystis* spp. ASV was detected in 2020, with a very low total relative abundance of 0.008%. Although the respective relative abundances of *Planktothrix* and *Microcystis* species are low, the strong bioinformatic filtration (for details, see Section 5.6) confirms the biological significance of these amplicon sequencing results. The ASVs of the most abundant *Synechococcus* spp., as well as the *Planktothrix* spp. and *Microcystis* spp. ASVs, were used for further analyses.

**Figure 2.** Krona plot of the DRM cyanobacterial community composition, as determined by 16S rRNA gene-amplicon sequencing across the sampling periods in 2019 (**A**) and 2020 (**B**). The community analysis was carried out by filtration of DRM water samples, total DNA extraction of the filters and PCR amplicon sequencing of the cyanobacteria-specific 16S rRNA gene region V3–V4 (380 bp length). For this Krona plot, and as an overview, the results shown are based on all samples combined per year. Taxonomic affiliation was carried out using the Greengenes reference database, and all taxonomic ranks are represented in the plot. Amplicon sequence variants (ASVs), as outputs of the Dada2 software package (see Section 5), distinguish sequence variations by a single nucleotide, giving ASVs a higher resolution than the operational taxonomic units (OTUs) typically used. Therefore, ASVs were used as the deepest taxonomic rank in our study. Total relative abundance was calculated by dividing the number of reads affiliated to an ASV in a sample by the total number of reads in the sample.

### *2.3. Synechococcus Rubescens and Cyanobium Gracile Clusters in 2019 and 2020*

We examined the phylogenetic relationship of the main *Synechococcus* ASVs that were detected in 2019 and 2020, with the reference sequences of (i) all the cultivated phycoerythrin-rich *Synechococcus* spp. of Lake Constance, as established by Ernst and colleagues in 2003 [45], and (ii) *Synechococcus rubescens* and phycoerythrin-rich *Cyanobium gracile* reference sequences, as established by NCBI (Figure 3). The relationship was established using the appropriate sequence fragments, representing the PCR amplicon of 380 bp, with the IQ-TREE program [46], based on a phylogenetic inference using the maximum likelihood, coupled with ModelFinder to determine the best-fitting nucleotide substitution model [47]. Although the target sequence was shorter than the full 16S rRNA gene sequences established by Ernst et al., 2003, the phylogenetic relationship between the reference sequences remained the same, thereby confirming our analyses. The ASVs from this study were always grouped in pairs, with one ASV from 2019 (Figure 3, green font) and another from 2020 (Figure 3, blue font), as a reflection of the reoccurring *Synechococcus* phylotypes across the two years, further confirming our analysis.

**Figure 3.** Illustration of the phylogenetic relationship of the *Synechococcus* spp. taxa observed in 2019 and 2020. The phylogeny is based on the cyanobacteria-specific V3–V4 (380 bp) 16S rRNA gene region. Colors correspond to the origin of the sequences, with green referring to ASVs observed in 2019 and blue referring to ASVs observed in 2020 (sampling year also in brackets). Sequences from *Synechococcus spp.* taxa, previously isolated from Lake Constance [45], and sequences of *Synechococcus rubescens* strain SAG3.81 and *Cyanobium gracile* PCC 6307 from NCBI were used as references (indicated in black). IQ-TREE was used for phylogenetic inference using maximum likelihood. The model of nucleotide substitutions used was TPM2+F + I [46], determined as the best-fit model by ModelFinder [47], based on the Bayesian information criterion scores and weights. Numbers at the internal nodes represent the percentage support of this specific node in 1000 bootstrap testing. The tree was rooted using the sequences of the cyanobacterium *Anacystis nidulans* PCC 6301 as an outgroup (not shown). The reference sequences are identified by their accession numbers in brackets.

Overall, the ASVs were grouped into two main clusters, either more closely related to the *S. rubescens* or to the *C. gracile* reference sequence (Figure 3). The top three most abundant ASVs for 2019 (ASV13, ASV14 and ASV15; see Figure 2 and below) and the top two most abundant for 2020 (ASV4 and ASV7; Figure 2), were more closely related to the *Synechococcus rubescens* NCBI reference sequence. For this group, two reference sequences of Lake Constance phycoerythrin-rich *Synechococcus* isolates [45] were available (Figure 3; BO8807 and BO9404), while for the *C. gracile* group, the reference sequences of seven Lake Constance isolates were available (Figure 3).

The megablast results were analogous to the phylogenetic tree shown, with the same ASVs in 2019 and 2020 being affiliated more closely to either *Synechococcus rubescens* or *Cyanobium gracile* (Supplementary Material, Table S3a,b), with the exception of ASV18 in 2019 and ASV10 in 2020. These ASVs showed identical percentage identities and E-values in the megablast for both *S. rubescens* and *C. gracile* (Table S3a,b) and were, therefore, grouped in between both clusters (Figure 3).

### *2.4. Dynamics of Synechococcus ASVs in 2019 and 2020*

We examined the change in the relative abundance of the *Synechococcus* ASVs detected, over time. Figure 4 illustrates the dynamics of the *Synechococcus* ASVs as a heatmap, using relative abundance data after the log10 transformation (Figure 4A,C) and the data distribution (Figure 4B,D) as the mean and standard deviation of the relative abundances in a combined bar and jitter plot. Each dot represents the relative abundance of the taxa at a specific date (relative abundance values (%) per ASV across the sampling dates are shown in Supplementary Material, Table S4a,b).

Overall, the observed changes in the phylogenetic structure across the sampling dates suggested a high degree of successional change within the *Synechococcus* spp. community at the DRM. Some ASVs varied from being almost undetectable in the beginning to a relative abundance maximum later in the year, or showed the opposite trend (being abundant at the beginning or in the middle of the sampling campaigns), while other ASVs showed a comparatively stable (and low) relative abundance across the sampling campaigns. Indeed, while 10 ASVs were registered in 2019, only nine ASVs were registered in 2020 (Figure 4A,C). Thus, different ASVs appeared to dominate, and, therefore, provide a successional change, in the two field campaigns. For example, ASV13, as the most prominent ASV in 2019 (Figure 4B), reached its maximum in July (up to 36%; Figure 4A) and decreased thereafter in its relative abundance (to approximately 10%) at the end of September, while the two second most abundant ASVs in 2019 (ASV 14 and 15; Figure 4B) reached their maxima in mid-August and later in the year (see Figure 4A and Table S4a). Similarly, in 2020, ASV7 represented almost 50% of the total relative abundance in early July, but decreased to approximately 6% by mid-August, while the most abundant taxon in the 2020 sampling campaign (ASV4; Figure 4C,D) reached its peak at the end of August (68%) and decreased to approximately 23% by the end of the campaign (Figure 4C). Furthermore, the third most abundant ASV21 in 2020 represented only approximately 1% of the total relative abundance at the beginning of the sampling campaign, but almost 20% at its end (see Figure 4C, Table S4b).

The statistical relevance of the differences in *Synechococcus* ASV relative abundance was confirmed by a Kruskal–Wallis analysis of variance, with a *p*-value far below the threshold of 0.05 (*p*-value of 3.95 × <sup>10</sup>−<sup>10</sup> and 1.27 × <sup>10</sup>−<sup>5</sup> for 2019 and 2020, respectively). In an attempt to group the taxa according to their relative abundance differences, a post hoc Conover–Iman test was performed using a Benjamini Yekutieli *p*-value adjustment method for False Discovery Rate control (Table S5a,b). For 2019, the results indicated two groups. First, a group comprising ASVs that comparatively stably dominated the cyanobacterial community at the DRM throughout July and October (together 63% to 87% of the total relative abundance); these were ASV13, 14, 15, 28 and 29. A second group statistically differed from the first group (i.e., ASV17, 18, 22, 30 and 33), for which the abundance differences appeared to be larger and/or occurred within shorter time intervals: For example, ASV22 reached its high relative abundance of 15% on only two sampling dates (in September and October, see Figure 4A and Figure S4a), as also discussed above. For 2020, the Conover–Iman test did not significantly separate the ASVs into two groups as for 2019 (Table S5b), although visually (Figure 4D), ASV4 dominated the cyanobacterial community with an average relative abundance of 45% throughout 2020.

**Figure 4.** Dynamics of relative abundance changes for the *Synechococcus* ASVs observed during the 2019 and 2020 sampling campaigns. Shown are heatmaps across all sampling dates (**A**,**C**) and the ASVs' average relative abundances (**B**,**D**) for each year (bars) with each individual data point indicated (grey dots). For the y-axes of heatmaps (**A**,**C**), and average relative abundances (**B**,**D**), the different ASVs were ordered from the most abundant (top) to the least abundant (bottom). The heatmap data was log10 transformed from the read count matrix, and, to avoid introducing infinity for zero read counts, we added one artificial read to every cell prior to log10 transformation (log10[x + 1]). Color schemes vary between dark blue for low log10 values to bright orange for high log10 values; a higher log10 value means a higher relative abundance. Please note that the x-axes for B and D do not have the same scale. The average relative abundance across all samples of each year was calculated as it was for the Krona plot (see Figure 2). Standard deviation was also calculated and represented for each ASV; if the graphical representation of the standard deviation was below 0, the minimum error bar value was set to 0.

### *2.5. Dynamics of Planktothrix and Microcystis ASVs, the Abundance of Microcystin Biosynthesis Genes and the Concentration of Microcystins in Samples Taken during 2019 and 2020*

In both the 2019 and the 2020 amplicon sequencing datasets, we detected two ASVs affiliated to the *Planktothrix* genus. The relative abundance of *Planktothrix* ASVs in 2019 increased from July to late September, with a peak on 31 July, where 0.75% of the total cyanobacterial community were *Planktothrix* ASVs (Figure 5A). Likewise, in 2020, *Planktothrix* ASVs were abundant from the end of July to the end of September, with a maximum

of approximately 2% relative abundance on 21 July (Figure 5B). While one of the *Planktothrix* ASVs was affiliated to *Planktothrix rubescens* (a toxin-producing *Planktothrix* species), the amplicon sequencing of the cyanobacteria-specific 16S rRNA-gene fragments of the other ASVs did not allow for taxonomical distinction at the species level (i.e., between the mostly non-toxin-producing *P. agardhii* and the toxin-producing *P. rubescens*). For example, a megablast alignment of the *Planktothrix* ASVs suggested *Planktothrix agardhii* and *Planktothrix rubescens* were the top hits, with identical query coverage and E-values, and percentage identities varying between 99.70% and 100% for *P. agardhii*, and 99.38% to 99.74% for *P. rubescens*.

Beyond the *Planktothrix* ASVs, two ASVs that are affiliated to the *Microcystis* genus were detected in 2019; however, only one *Microcystis* ASV was detected in 2020. As both genera, *Planktothrix* and *Microcystis*, are renowned for their toxin producing species, more in depth analyses were carried out regarding the toxin producing potential of the species found in the Überlingen embayment.

To confirm the detection of the *Planktothrix* genus in Lake Constance at very low levels, particularly when compared to *Synechococcus*, quantitative PCR (qPCR) was performed to estimate the abundance of toxin-producing *Planktothrix* genotypes. Specifically, we used *Planktothrix*-specific primer pairs for the 16S rDNA gene and the *mcyBA1* gene, as described by Ostermaier and Kurmayer, 2009 (Table 1, [48]). *Planktothrix mcyBA1* encodes for the first adenylation domain in non-ribosomal peptide synthase (NRPS) gene clusters, and is present only in species that are capable of toxin production. Briefly, we created a standard curve using a dilution series of *Planktothrix* DNA (0.00001–100% *Planktothrix* DNA diluted in *Microcystis* DNA) and then calculated the relative abundance of *Planktothrix* DNA in our samples using a linear regression. The calculated relative abundance of *Planktothrix* ranged between 0.1 and 0.6% in 2019, and 0.1 and 0.01% in 2020 (Figure S3). Although the relative abundance calculated by qPCR differed from the relative abundances that were found with amplicon sequencing (Figure S3 and Figure 5), the trend aligned well between both methods. Statistical analyses of the differences between the 16S-rRNA gene and *mcyBA1* amplifications (ANOVA, see Ostermaier and Kurmayer, 2009) showed no significant difference in abundance with respect to toxin-producing or non-toxin-producing *Planktothrix* genotypes, suggesting that there was only one genotype of *Planktothrix* present in 2019 and 2020. To confirm the presence of toxin-producing cyanobacterial species in Lake Constance, we used universal *mcyE*-specific PCR primers (HEPF/R, Table 1, [49]). Being a member of the MC production gene cluster, *mcyE* is partly responsible for the synthesis of the ADDA chain in microcystins, as well as the incorporation and synthesis of *D*-Glu [50]. For the 2019 sampling campaign, the PCR yielded amplicons for every sample tested, suggesting the presence of potential microcystin producers throughout the year. The subsequent Sanger sequencing of the PCR products, and analysis of the consensus sequences with megablast, attributed toxin-producing capabilities to *Planktothrix* species (Figure 5A), except for the sample taken on 24 September 2019, where the *mcyE* consensus sequence had the highest alignment scores with *Microcystis*, thus matching the date with the highest relative abundance of the *Microcystis*-affiliated ASV. For 2020, *mcyE* amplicons were observed less consistently than in 2019, although, as seen in 2019, Sanger sequencing of these amplicons attributed the toxin-producing capabilities to *Planktothrix* spp. (Figure 5B).

Collectively, we provide evidence that toxin-producing *Planktothrix* spp. (and/or *Microcystis* spp.) are present in the Überlingen embayment of Lake Constance. However, our data did not allow us to conclude whether or not toxin production took place during the sampling campaigns, as many species can carry the gene cluster without actively producing the toxins [51]. Consequently, we analyzed the biomass samples that were collected independently on filters from the DRM for microcystins, using UPLC–MS/MS (Figure 5A,B). Microcystin concentrations peaked at the end of September 2019, where, in total, approximately 1.5 ng/L of intracellular microcystins were found. The microcystin variants present were MC-LR (leucine and arginine in hypervariable region) and MC-YR (leucine and tyrosine in hypervariable region), with MC-LR being almost solely responsible

for the peak in toxin concentration in September 2019 (Figure 5A). Strikingly, for the samples collected during 2020, no toxins were detected under the conditions we used.

**Figure 5.** Relative abundance of the *Planktothrix* and *Microcystis* ASVs observed in 2019 and 2020, and of microcystin concentrations for samples taken in 2019. The bar plots represent the total relative abundance observed for the two *Planktothrix* ASVs (red bars) detected in 2019 (**A**) and 2020 (**B**), for the two *Microcystis* ASVs (blue bars) detected in 2019, and the single *Microcystis* ASV detected in 2020. The *x*-axes represent the different sampling dates and the left *y*-axis the relative abundance (%). The error bars in B represent the standard deviation of the relative abundance, as calculated from biological triplicates (n = 3); no error bars are represented for 2019, as only one sample was collected per sampling date. The right *y*-axis represents the toxin concentration (ng/L) determined for independently collected DRM filters, for which the microcystin variant concentrations are indicated by dashed lines in black (MC-LR) and grey (MC-YR) (see main text). The text on top of each bar indicates the results of the PCR amplification of the microcystin synthesis gene *mcyE* (+, detected; −, not detected) and of the phylogenetic affiliation of the *mcyE* consensus sequence to either *Planktothrix* or *Microcystis*, as established by Sanger sequencing of the PCR amplicons (see main text).

**Table 1.** Primers used in this study.



**Table 1.** *Cont.*

<sup>1</sup> TaqMan probes contain 5 FAM (6-carboxyfluorescein, fluorescent reporter dye) and 3 TAMRA (6-carboxy-tetramethylrhodamine, fluorescent quencher dye).

### *2.6. Retrospective Evaluation of Depth Profiles for the Lake Überlingen Routine Sampling Site*

During our sampling campaigns in 2019 and 2020 (and the ongoing 2021 campaign), prominent DRM were observed from June onwards, particularly after long and stable good weather periods. This is best illustrated on 1 July 2019, when we observed a first prominent DRM at 18.4 m depth at the routine sampling site 'Wallhausen', in the Überlingen embayment of Lake Constance. This DRM appeared after the weather presented a stable window of approximately two weeks with predominant sunshine and no precipitation, low wind and elevated temperatures, as depicted in Figures S5, S7 and S8.

Three-dimensional (3D) plots of the FluoroProbe depth profiles for 'cryptophyta' content, as proxy, in the water column at 0–40 m depth (Figure 1) across the sampling campaigns in 2019 and 2020 are depicted in Figure 6A. Furthermore, we retrospectively evaluated the 'cryptophyta' depth profiles that were collected in the previous years from the routine sampling site, and transformed these into 3D plots using MATLAB (for years 2009–2018, see Supplementary Material, Figure S6). During the last twelve years, DRM have occurred multiple times (Figure S6, plots from 2011 and 2015–2020). Specifically, in 2016, we observed a prominent DRM in the summer, with maximum 'cryptophyta' concentrations of 6 μg/L on 6 September and 20 September (Figure 6B). Corresponding to the DRM, the high abundance of *Planktothrix rubescens* has been reported in various sampling sites at Lake Constance in 2016 [42,43].

Overall, prominent DRM were observed from July to October, with a maximum 'cryptophyta' content of approximately 2–4 μg/L (as estimated based on the FluoroProbe calibration), and at water depths ranging between 10 and 20 m (Figure 6A and Figure S6). The observed variation of the DRM depths likely follows the lake's internal waves, as previously found in Lake Ammer in Germany and Lake Bourget in France [14,52]. Although we could speculate that these peaks represent high abundances of *P. rubescens* in the Überlingen embayment, the absence of any appropriate samples allowing for DNA or toxin analyses available from that time preclude any corroboration.

**Figure 6.** Depth profiles recorded with the FluoroProbe across the sampling campaigns in 2019 and 2020 (**A**), in comparison to the year 2016, in which *P. rubescens* blooms were reported in Lake Constance (**B**). Depicted are the FluoroProbe profiles for 'cryptophyta' abundance (cf. Figure 1) (expressed in μg chlorophyll-a per liter), recorded as proxy of red pigment abundance, in the water column from 0–40 m depth at the routine sampling site 'Wallhausen', in the Überlingen embayment of Lake Constance. Prominent DRM are highlighted with dates and water depths. Coordinates of the study site: 47.7571◦N 9.1273◦E; for an illustration, see Figure S4. In 2016, blooms of *P. rubescens* were reported for Lake Constance in September–October at various sampling sites (i.e., of the German, Austrian and Swiss sections of Lake Constance [42,43]). For 2016, FluoroProbe data is available for the Überlingen embayment of Lake Constance (**B**).

### **3. Discussion**

The characterization of the recurring blooms of red-pigmented cyanobacteria in the Überlingen embayment of Lake Constance at water depths of 15–20 m by amplicon sequencing demonstrated the presence of *Synechococcus*, *Planktothrix* and *Microcystis*. Briefly, the relative abundance data suggested that *Synechococcus* taxa predominated the community (96–98%) at these water depths, while *Planktothrix* and *Microcystis* taxa were detectable only in very low abundances. For example, in 2020, up to 45% of the total relative abundance was represented by a single *Synechococcus* ASV (ASV4; Figures 2B and 4C,D). Moreover, the observed changes in the relative abundance of *Synechococcus* ASVs across the sampling dates suggest a high degree of successional change within the *Synechococcus* spp. community at the DRM (Figure 4).

Phycoerythrin-rich *Synechococcus* species from Lake Constance have been investigated previously [45,53,54], and, similarly to *Planktothrix* spp., they can express a large variety of phycobilins, thereby exploiting diverse light conditions [45,55]. Interestingly, the highest abundant *Synechococcus* taxa found in our studies in 2019 and 2020 are closely related to either *S. rubescens* or *C. gracile*, forming two main clusters (Figure 3, Table S3a,b). For the 2019 and 2020 sequences, a discrimination between *S. rubescens* and *C. gracile* is difficult (Figure 3), as the 380-bp PCR amplicon used in this study is cyanobacteria-specific, and allows us to affiliate taxa with high confidence only up to the genus rank. However, each *Synechococcus* taxon detected in 2019 is paired with a taxon detected in 2020, and their sequence alignment showed a 100% identity with no gap, suggesting recurring phylotypes/ecotypes, at least across the two years. The close association of our sequences with those from the early 2000s [45] suggests a stable deep water cyanobacterial community in Lake Constance. The latter interpretation is supported by the earlier finding that different lineages of *Synechococcus* spp. can adapt to, and thrive in, specific ecological niches [56].

The predominance of *Synechococcus* species over other cyanobacterial genera was reported for the experimental cocultures of *Synechococcus* and *Microcystis* strains, as well as varying phosphate and nitrogen concentrations [57,58]. Indeed, at low nutrient concentrations, which could also mirror the currently oligotrophic conditions of Lake Constance [59], *Synechococcus* outcompeted *Microcystis* in growth rate and final biomass [57,58]. The predominance of *Synechococcus* may be explained by its seemingly higher affinity for orthophosphates, more efficient nutrient uptake (due to a larger surface-to-volume ratio), as well as by the potential for competitive inhibition via the quorum sensing/quenching molecules between the two strains [57,58,60,61]. The latter laboratory findings were also corroborated in natural habitats, at least in regards to their trend [62].

Despite the seemingly stable predominance of *Synechococcus*in the deep water cyanobacterial community observed in Lake Constance over the last two decades [45], the de novo occurrence of *Planktothrix* spp. in 2016, and the reconfirmation of this earlier finding with our samples in 2019 and 2020, could suggest that *Planktothrix* spp. is in the process of establishing a stable presence in the Überlingen embayment of Lake Constance. The latter is of critical importance as mass occurrences of toxin-producing cyanobacteria at water depths of >20 m could become a threat to the water intake for the Sipplingen water treatment plant (https://www.bodensee-wasserversorgung.de, accessed on 7 May 2021 [63]) which serves >4 million people with drinking water. Indeed, we detected a low abundance of potentially MC-producing *Planktothrix* spp. using amplicon sequencing, as well as through *Planktothrix*-specific qPCR for 2019 and 2020 (Figure 5 and Figure S3). Furthermore, the presence of the microcystin biosynthesis gene cluster, *mcy*, was detected through the PCR amplification of *mcyE* (Figure 5). The subsequent Sanger sequencing of these amplicons identified *Planktothrix* spp. to be the main contributor to this gene sequence in our samples, with the exception of 24 September 2019, when the amplified *mcyE* consensus sequence identified the highest alignment scores in *Microcystis* spp. In *Planktothrix* spp., the *mcy* gene cluster can be inactivated by various mutations, including insertions or deletions, and, thus, non-toxic strains can develop [64]. Non-toxic strains are less successful in competition than their toxic relatives, and, thus, *Planktothrix* blooms are usually dominated by toxic strains [48]. Corresponding to the presence of *mcyE* in 2019, low amounts of the microcystins MC-LR and MC-YR were detected using UPLC–MS/MS. In 2020, although *mcyE* amplicons were detectable in some of our samples, concentrations of microcystins were below the detection levels of the UPLC–MS/MS that was used (LOD of UPLC–MS/MS method: 0.5 ng/mL [65]; therefore, the resulting LOD in our water samples: 62 pg/L lake water). Despite the low abundance of *Planktothrix* spp. and the low concentrations of detected toxins, the question does arise concerning whether the latter is a sign of a fundamental change in water body dynamics in the Überlingen embayment of Lake Constance (e.g., resulting from global warming). Indeed, the greater and prolonged stratification of water bodies, in conjunction with lowered nutrient levels, would promote a deep water euphotic ecosystem encompassing low-light specialized species, such as the picocyanobacterial *Synechoccocus* spp. and *Planktothrix* spp. [9,15], as was reported for other prealpine lakes (e.g., Lake Zurich [4], Lake Mondsee [8] and Lake Bourget [6]).

As *Planktothrix* spp. are stimulated by increased temperatures [15], the continuous shift toward higher overall temperatures could favor the perseverant establishment of toxinproducing *Planktothrix* spp., to the disadvantage of today's *Synechococcus*-dominated DRM ecosystems. Considering that *Synechococcus* spp. are currently markedly outcompeting other cyanobacterial species occurring at these water depths, this may suggest that the allelopathic compounds from *Synechococcus* spp. can have an adverse effect on co-occurring

species. Indeed, such effects have been observed for freshwater *Synechococcus* spp., which were able to impact the growth of other freshwater cyanobacteria or green algae [66,67]. Adverse effects caused by allelopathic compounds from marine *Synechococcus* spp. were also observed on various marine invertebrates [68], as well as other bacterial species [60]. In conclusion, the described effects indicate the widespread production of allelopathic compounds by *Synechococcus* species that can even influence other bacteria, plants and invertebrates.

In most cases, the co-occurrence of species occupying the same, or similar, ecological niche leads to the dominance of one species, largely depending on the individual species' competitive advantages (e.g., nitrogen fixation, uptake of inorganic phosphorus, regulation of buoyancy, allelopathic compounds, etc.), as also shown by Weisbrod et al. [69]. This suggests that, if *Synechococcus* outcompetes *Planktothrix* in its ecological niche, allelopathic compounds (such as toxins), amongst other factors, may be used to compete against the respective other species. Indeed, allelopathic activity is one of the major competitive strategies of freshwater *Synechococcus* against coexisting phytoplankton species [66], further supporting the potential role of picocyanobacterial exudates in competition with other cyanobacteria, such as *Planktothrix*. Counterintuitively, some freshwater *Synechococcus* spp. possess a positive allelopathic activity towards *Microcystis* spp., and no effect on *Phormidium* spp., whereby *Phormidium* spp. (like *Planktothrix* spp.) is part of *Oscillatoriaceae* [66], suggesting that *Synechococcus* spp. could even promote the growth of toxin-producing species. The complex interplay of species in competition in Lake Constance emphasizes the need for further studies regarding the co-occurrence and dominance of *Synechococcus* spp., relative to *Planktothrix* and *Microcystis* spp. in this lake.

The geographical setting of the Überlingen embayment of Lake Constance, with its minor wind influence (Figures S7 and S8), is considered an additional factor that could favor the continuous development of deep water *Planktothrix* spp. populations. Indeed, light winds and convective mixing are highly important in the seasonal cycling of *P. rubescens* communities within a strongly stratified medium-sized lake [70]. In consequence, this means that toxin-producing *Planktothrix* spp. could possibly establish themselves as a dominant cyanobacteria species at deeper water levels, and may become a relevant concern for the quality of the Überlingen embayment of Lake Constance as a drinking water resource in the not too distant future.

### **4. Conclusions**

Our study characterized the deep water red pigment maxima (DRM) in the Überlingen embayment of Lake Constance in 2019 and 2020 as being dominated by phycoerythrin-rich picocyanobacterial; namely, *Synechococcus rubescens* and *Cyanobium gracile.* Unlike other prealpine lakes, the DRM in Lake Constance is not dominated by the phycoerythrin-rich, filamentous and often toxin-producing *Planktothrix rubescens*. Indeed, the alignment of our results with the sequences from the Ernst et al. study (2003) demonstrates high sequence similarity, and suggests that the same species have been dominant at 15–20 m depth in Lake Constance for the past 20 years. However, we confirmed the reports from 2016 that *Planktothrix* spp. does occur in Lake Constance in the years 2019 and 2020, albeit at very low relative abundances. Nevertheless, microcystin concentrations of up to 1.5 ng/L were detected through UPLC–MS/MS in 2019, which appeared to be produced by *Planktothrix* and/or *Microcystis* spp. Hence, at present, Lake Constance seems to have a rather stable deep water cyanobacterial ecosystem, predominated by *Synechococcus* spp., although the geographical setting, as well as the continued climate warming, could favor the development and steady predominance of toxin-producing *Planktothrix* spp. This highlights the importance of a future monitoring program for Lake Constance, with emphasis on sequencing-based cyanobacterial community studies and microcystin monitoring, as well as the importance of competition studies regarding the different cyanobacterial taxa in relation to the physiochemical and biological parameters of the lake, particularly in respect to the ongoing climate change. Monitoring programs and hypothesis-driven competition

studies may provide the required database to predict future deep water mass occurrences of toxin-producing cyanobacteria and, thus, help to secure Lake Constance as the drinking water resource for millions of people in the future.

### **5. Materials and Methods**

### *5.1. Sample Collection*

Samples were taken every two weeks from 1 July to 8 October in 2019, and 7 July to 15 September in 2020. The initial sampling in 2019 included only one replicate/day, while samples were collected in biological triplicates (n = 3) during 2020. A bbe Moldaenke FluoroProbe (FP) (SN: 01709; recalibrated at bbe in 2009, 2012, 2016 and 2019) was used to determine the deep red maximum (DRM). Samples were collected at that depth every other week in Upper Lake Constance (47.7571◦ N 9.1273◦ E), using a messenger-released Free Flow Water Sampler 5 L (HYDRO-BIOS, Altenholz, Germany). An overview of the samples taken during this study is illustrated in Table S1. Water was filtered through a 180 μm nylon net filter to exclude zooplankton and larger particles. Two liters of water were then filtered on one Whatman GF6 glass fiber filter while applying 2 bars pressure for the collection of plankton biomass. Filters were stored at −20 ◦C until analysis. This sampling method was applied to DNA extractions in 2019, and toxin extractions in 2019 and 2020. In 2020, DNA samples were collected independently (n = 3) on 0.2 μm polycarbonate filters.

### *5.2. DNA Extraction and PCR*

DNA from the 2019 samples was extracted using the ZYMO Research Fecal/Soil/Microbe Microprep kit, following the manufacturer's instructions. DNA of 2020 samples was extracted using a phenol/chloroform/isoamylalcohol protocol, adapted from Rusch et al., [71] and the JGI protocol [72]. Standard PCR was performed with Taq polymerase, using 2X Taq MasterMix (NEB) and 30 cycles. The microcystin synthesis gene, *mcyE*, was amplified with the HEPF/R primer set [49], and DNA from cultured *Microcystis aeruginosa* strain 78 was used as a positive control for the presence of *mcyE*. Planktothrix-specific primer pairs PcPI+/− (PC-IGS) and peamso+/− (*mcyA*) [8] were used for additional PCR reactions, with *Microcystis aeruginosa* strain 78 genomic DNA as a negative control, and *Planktothrix rubescens* strain 101 genomic DNA as a positive control.

### *5.3. RT-PCR/TNA*

Quantitative Taq Nuclease Assays (TNA or TaqMan PCR) were performed with primers specific to *Planktothrix* spp. [48]. We quantified both *Planktothrix*-specific 16S rDNA and *mcyBA1*, which encodes the first adenylation domain of *mcyB*, and is indicative of all genotypes containing the *mcy* gene cluster. Both probes contained 5 FAM as a fluorescent marker and 3 TAMRA as a quencher dye. Each reaction contained 50 ng template DNA, 200 nM of primers and probes, and KAPA probe fast MasterMix. Amplification and quantification were carried out in triplicates in a Bio-Rad CFX96 cycler with the following protocol: 10 min at 95 ◦C to activate the hot start polymerase, followed by 50 cycles of 15 s at 95 ◦C, 60 s at 51.5 ◦C for 16S, 60 s at 57 ◦C for *mcyBA1*, 60 s at 68 ◦C and a final elongation for 5 min at 68 ◦C. The standard curve contained *Planktothrix* DNA mixed with *Microcystis* DNA in different ratios (0.00001–100% *Planktothrix* DNA diluted in *Microcystis* DNA). Data analysis using linear regression from the standard curve was performed using BioRad CFX manager, Microsoft Excel and GraphPad Prism 5.

### *5.4. Sanger Sequencing*

Sanger sequencing was performed with the amplicons produced by the HEPF/R primer pair to identify the main microcystin producer in the samples. PCR amplicons were purified using a QIAquick PCR cleanup kit and sent to Eurofins Genomics for analysis. The identity of the obtained sequences was determined using Nucleotide BLAST megablast [73].

### *5.5. Amplification and Illumina Sequencing*

Amplification of the V3–V5 hypervariable regions, and the cyanobacterial-specific V3–V4 hypervariable regions of the 16S rRNA gene, was performed with 0.02 U/μL of Phusion High Fidelity DNA polymerase, 1X Phusion HF Buffer, 200 μM of dNTPs (New England Biolabs, USA) and 0.5 μM of each primer. Primers targeting the V3–V5 hypervariable regions were 357F and 926R [74,75]. Cyanobacteria-specific primers were CYA359F and CYA784R [44]. Each PCR comprised an initial denaturation step of 3 min at 98 ◦C, followed by 30 cycles of denaturation for 45 s at 98 ◦C, annealing for 20 s at 62.4 ◦C (V3–V5) or 60 ◦C (cyanobacteria-specific) and extension for 8 s at 72 ◦C, and a final extension step for 5 min at 72 ◦C. Extracted DNA was added at a final concentration of 0.12 ng/μL. No purification step was performed; the PCR products were directly sent to Eurofins Genomics for sequencing using Illumina MiSeq 2 × 300 bp with the Microbiome Profiling Indexing only package. The data presented in this study are accessible on NCBI under the bioproject number PRJNA727470.

### *5.6. Bioinformatics Pipeline*

The analysis was carried out using the already merged dataset provided by Eurofins Genomics, Konstanz, Germany. The expected fragment sizes were 569 bp for the V3–V5 amplicon and 425 bp for the V3–V4 cyanobacterial-specific amplicon. Reads were first trimmed using Trimmomatic [76], removing all reads with a Phred quality below 3 for the start and the end of the reads, below an average quality of 10 on a window of 3 base within the reads, and below a size of 500 bp for the V3–V5 amplicons and 380 bp for the V3–V4 amplicons. FastQC was used for quality control of the reads before and after trimming [77]. The following steps were performed using QIIME2 2019.10 [78]. Filtration of chimeras using the consensus method, denoising and dereplication of the quality reads were performed using the denoise and dereplicate single-end sequences (Dada2, denoisesingle) and a reads learn of 2,000,000 reads for the training error model [79]. Quality trimming had already been performed using Trimmomatic, so no trimming step was performed here. Taxonomic affiliation was achieved using the classify-consensus-vsearch program and the databases SILVA\_138 and Greengenes, with a percentage identity of 80%, 90%, 97% and 100%. Taxonomic results were merged and the highest percentage identity taxonomic affiliation was retained. Reference databases were previously trained using the feature-classifier extract-reads script with the sequences of the primers used for the amplification of the hypervariable regions. After training, the databases only contained the part of interest of the 16S rRNA gene for taxonomic assignation, V3–V5 hypervariable regions for the general 16S rRNA gene amplification and V3–V4 hypervariable regions for the cyanobacteria-specific primers. Taxonomy was mostly consistent between the SILVA\_138 and Greengenes databases.

### *5.7. Phylogenetic Analysis*

A megablast (highly similar sequence) was performed on the sequences affiliated to the genera *Planktothrix*, *Microcystis* and the most abundant *Synechococcus* using the 16S rRNA sequences reference database on NCBI. The top ten hit sequences and the description table were extracted. The *Synechococcus* sequences from Lake Constance analyzed and sequenced by Anneliese Ernst [45] were downloaded from the NCBI database. Two other 16S rRNA gene sequences, belonging to *Synechococcus rubescens* and *Cyanobium gracile*, were also collected from NCBI and used as reference for the phylogenetic analysis, as previously carried out by Anneliese Ernst. The collected dataset was then compared to the most abundant *Synechococcus* sequences in 2019–2020. Sequences were merged with our sequences into a fasta file and aligned using SeaView [80]. IQ-TREE 2.1.2 [46] was used to calculate the phylogenetic tree by maximum likelihood using 1000 bootstrap parameters. The model of nucleotide substitution used was TPM2+I+F [46], determined as the best fit model by ModelFinder [47], based on the Bayesian information criterion scores and weights. The tree was visualized on R using the package ggtree [81].

### *5.8. Statistical and Network Analysis*

Statistical analysis was performed with R software [82], using the package Phyloseq [83], vegan [84] and visualized with the package ggplot2 [85]. Features represented by less than 3 reads in below 20% of the samples were discarded. Chloroplast-affiliated features were removed from both datasets, bacteria-affiliated taxa were also removed from the cyanobacterial-specific dataset only. No rarefaction has been applied to the dataset [86]. After filtration, the lowest number of reads in a sample was 19,175 for the cyanobacteriaspecific dataset and 24,329 for the general 16S rDNA dataset in 2019. The 2020 dataset consisted of only the cyanobacteria-specific amplicon and the lowest number of reads was 24,341. This number of reads instilled confidence in catching all the richness present in our dataset, as the rarefaction curves showed that we were in the stationary phase of the curve. The read count matrix was transformed in relative abundance by dividing the reads affiliated to one taxa by the total number of reads in the sample (x/sum(x)), and a logarithmic transformation was also applied. To avoid the presence of 0 in the matrix, one artificial read was added to each cells of the data (Log10(x + 1)). Alpha diversity was analyzed using the Observed richness, Pielou's evenness index, Shannon diversity index and the Inverted Simpson diversity index. Community composition was observed using barplot, jitter plot and krona plot, using the relative abundance data and heatmap with the Log10 transformed data. Statistical analyses were performed using a non-parametrical statistical test, as sequencing data did not fulfill the parametric test requirements (e.g., data normality and homoscedasticity, and independence of observation) using a *p*-value and False discovery rate threshold of 0.05. Kruskall–Wallis one-way analysis of variance was performed on the relative abundance sub-data of the main Synechococcus taxa to test if at least one taxa dominated the community. The hypothesis that the relative abundance distribution of the tested taxa are equal was the null hypothesis (H0) and the alternative hypothesis (H1) was that the relative abundance distribution of the tested taxa are not equal. If the Kruskall–Wallis test rejected H0, a post hoc test using the Conover–Iman squared rank test, with a Benjamini Yekutieli *p*-value adjustment procedure, was performed to determine which taxon or group of taxa dominate the community. The Benjamini Yekutieli procedure was chosen for *p* value adjustment because of the dependency between the taxa's relative abundance observation. Barplots of the taxa of interest (*Planktothrix spp.*, *Microcystis spp.*, *Synechococcales*) and correlation observation of the main node from the network analysis were produced.

### *5.9. Toxin Extraction and Analysis*

Toxins were extracted from filters using a methanolic extraction method and were, subsequently, analyzed via UPLC–MS/MS [24,65]. Briefly, 3 mL 50% (*v/v*) aqueous methanol was added to each filter and soaked for 30 min at room temperature. After vigorous mixing (vortex) for 10 min, samples were sonicated in an ultrasonic bath for 15 min, before centrifugation at 4000 × *g* for 10 min. The supernatant was collected in a separate tube and the above steps, excluding the initial 30 min soaking time, were repeated twice with the remaining pellet. The pooled supernatant was dried overnight using a speed vacuum system (Univapo 100H) and resuspended in 250 μL 50% aqueous methanol. Toxin samples were stored in glass vials at −20 ◦C until analysis.

Concentrations of different MCs were measured via UPLC–MS/MS, with an internal standard containing deuterated MC-LR and MC-LF (D5-MC-LF and D7-MC-LR) [24,65,87]. Three different MCs, MC-RR, MC-YR and MC-LR were used as external standards for analysis, with final concentrations of 2, 10 and 100 ng/mL. An Acquity H-class liquid chromatograph and a Waters XEVO TQ-S mass spectrometer were used. For UPLC, an Acquity BEH C18 1.7 μm column (2.1 × 50 mm) with a corresponding guard column, each kept at 40 ◦C, was used. Solvents A and B were composed of 10% and 90% acetonitrile, respectively, 100 mM formic acid and 6 mM NH3. Initial conditions were 25% B, held for 30 s, then 45% B within 30 s, 60% B within 180 s and 99% B within 12 s, which was held again for 30 s. The flow rate was 0.4 mL/min. Prior to application of the next sample, the

column was re-equilibrated to 25% B over 78 s and held for 60 s. Injection volume was 5 μL. As described in Altaner et al., 2019, simultaneous analysis of MC congeners was carried out using five analysis windows, maximizing the scan time for each congener [65].

### *5.10. Evaluation of bbe Moldaenke FluoroProbe Data*

For data acquisition, the Moldaenke FluoroProbe (FP) was slowly lowered (approximately 0.2 m/s) from the water surface to 100 m depth using a winch, then quickly pulled up again, while collecting data both ways. FP data was preprocessed and later plotted in 3D plots using MATLAB R2020a [88]. The datasets were chopped above 1 m depth and below 100 m depth and subsequently sorted. Data were interpolated to 0.1 m steps, and 570 nm fluorescence measurements and cryptophyta content (μg/L) data were extracted. Plotting was carried out using the MATLAB standard functions *surf* (3D) and *plot* (2D), and peaks were analyzed using *max*.

**Supplementary Materials:** The following are available online at https://www.mdpi.com/article/10 .3390/toxins13090666/s1, Figure S1: Common microcystin structure; Figure S2: Illustration of red colored filters; Figure S3: Relative abundance as quantified with qPCR; Figure S4: Localization of study site; Figure S5: Depiction of stable weather window prior to 1 July 2019; Figure S6: Depiction of depth profiles for 2009–2020; Figures S7 and S8: Mean wind speed and directions around the study site; Table S1: Sampling dates with corresponding sampling depths; Table S2: Assignment of ASVs with SILVA and Greengenes databases; Table S3: NCBI megablast of *C. gracile* and *S. rubescens* in 2019 and 2020; Table S4: Relative abundance of *Synechococcus* ASVs per date in 2019 and 2020; Table S5: Conover–Iman test of *Synechococcus* ASVs in 2019 and 2020.

**Author Contributions:** Conceptualization, E.R., C.F., D.R.D. and D.S.; methodology, E.R. and C.F.; formal analysis, E.R. and C.F; data curation, E.R. and C.F.; writing—original draft preparation, E.R. and C.F; writing—review and editing, D.R.D. and D.S.; project administration, D.R.D. and D.S.; funding acquisition, D.R.D. and D.S. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by DFG Research Training Group R3—Responses to biotic and abiotic Changes, Resilience and Reversibility of Lake Ecosystems (GRK 2272) and MTU Friedrichshafen GmbH (MTU Umweltstiftung).

**Data Availability Statement:** The Illumina sequence data presented in this study is accessible at NCBI under the bioproject number PRJNA727470.

**Acknowledgments:** We would like to highlight the excellent support received from Julia Schmidt, Microbial Ecology and Limnic Microbiology, with regard to DNA extraction from samples she helped to organize from the Überlinger embayment. The assistance with UPLC-MS/MS by Sarah Krassnig and Aswin Mangerich and the help in the laboratory by Leon Walther is thankfully acknowledged. We would also like to thank Beatrix Rosenberg for access to long-term data and Alfred Sulger, Angelika Seifried, Pia Mahler and Josef Halder at the Limnological Institute of University of Konstanz for managing the ship cruises, and Anneliese Ernst for helpful discussions. Further, we would like to thank the MTU Umweltstiftung and all members of the RTG-R3, especially Tina Romer and Frank Peeters, for their support.

**Conflicts of Interest:** The authors declare no conflict of interest.

### **References**


### *Article* **Toxins and Other Bioactive Metabolites in Deep Chlorophyll Layers Containing the Cyanobacteria** *Planktothrix* **cf.** *isothrix* **in Two Georgian Bay Embayments, Lake Huron**

**Arthur Zastepa 1,\*, Todd R. Miller 2, L. Cynthia Watson 1, Hedy Kling <sup>3</sup> and Susan B. Watson <sup>4</sup>**


**Abstract:** The understanding of deep chlorophyll layers (DCLs) in the Great Lakes—largely reported as a mix of picoplankton and mixotrophic nanoflagellates—is predominantly based on studies of deep (>30 m), offshore locations. Here, we document and characterize nearshore DCLs from two meso-oligotrophic embayments, Twelve Mile Bay (TMB) and South Bay (SB), along eastern Georgian Bay, Lake Huron (Ontario, Canada) in 2014, 2015, and 2018. Both embayments showed the annual formation of DCLs, present as dense, thin, metalimnetic plates dominated by the large, potentially toxic, and bloom-forming cyanobacteria *Planktothrix* cf. *isothrix*. The contribution of *P.* cf. *isothrix* to the deep-living total biomass (TB) increased as thermal stratification progressed over the ice-free season, reaching 40% in TMB (0.6 mg/L at 9.5 m) and 65% in South Bay (3.5 mg/L at 7.5 m) in 2015. The euphotic zone in each embayment extended down past the mixed layer, into the nutrient-enriched hypoxic hypolimnia, consistent with other studies of similar systems with DCLs. The co-occurrence of the metal-oxidizing bacteria *Leptothrix* spp. and bactivorous flagellates within the metalimnetic DCLs suggests that the microbial loop plays an important role in recycling nutrients within these layers, particularly phosphate (PO4) and iron (Fe). Samples taken through the water column in both embayments showed measurable concentrations of the cyanobacterial toxins microcystins (max. 0.4 μg/L) and the other bioactive metabolites anabaenopeptins (max. ~7 μg/L) and cyanopeptolins (max. 1 ng/L), along with the corresponding genes (max. in 2018). These oligopeptides are known to act as metabolic inhibitors (e.g., in chemical defence against grazers, parasites) and allow a competitive advantage. In TMB, the 2018 peaks in these oligopeptides and genes coincided with the *P.* cf. *isothrix* DCLs, suggesting this species as the main source. Our data indicate that intersecting physicochemical gradients of light and nutrient-enriched hypoxic hypolimnia are key factors in supporting DCLs in TMB and SB. Microbial activity and allelopathy may also influence DCL community structure and function, and require further investigation, particularly related to the dominance of potentially toxigenic species such as *P.* cf. *isothrix*.

**Keywords:** deep-chlorophyll layers (DCLs); cyanobacterial toxins; *Planktothrix*; allelopathy; bioactive metabolites; hypoxia; Georgian Bay

**Key Contribution:** Deep chlorophyll layers of *Planktothrix* cf. *isothrix* in two meso-oligotrophic embayments in Georgian Bay showed measurable concentrations of the cyanobacterial toxins microcystins (max. 0.4 μg/L) and the other bioactive metabolites anabaenopeptins (max. ~7 μg/L) and cyanopeptolins (max. 1 ng/L)—along with the corresponding genes—that can act as metabolic inhibitors (e.g., in chemical defence against grazers, parasites) and allow a competitive advantage. Supporting physicochemical data provide strongs evidence that opposing and intersecting physicochemical gradients of light and nutrients from a hypoxic hypolimnion are critical in supporting

**Citation:** Zastepa, A.; Miller, T.R.; Watson, L.C.; Kling, H.; Watson, S.B. Toxins and Other Bioactive Metabolites in Deep Chlorophyll Layers Containing the Cyanobacteria *Planktothrix* cf. *isothrix* in Two Georgian Bay Embayments, Lake Huron. *Toxins* **2021**, *13*, 445. https:// doi.org/10.3390/toxins13070445

Received: 26 March 2021 Accepted: 23 June 2021 Published: 27 June 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

the DCLs in these embayments. Microbial processes and allelopathy may also influence the DCL community structure and function and the dominance of large, toxigenic cyanobacteria such as *P.* cf. *isothrix*.

### **1. Introduction**

Deep chlorophyll layers (DCLs) are ecologically important but often overlooked phenomena in many deep lentic ecosystems. DCLs can account for a significant fraction of pelagic primary productivity, play a key role in biogeochemical cycling of nutrients, and influence the vertical movement of micrograzers [1–7]. DCLs are likely far more prevalent than reported, as most sampling efforts concentrate on surface water layers. Investigations of large and deep Canadian waters such as the Great Lakes, where offshore DCLs have been observed as deep as ca. 50 m, have documented assemblages dominated by diatoms, picoplankton, and cryptophytes, along with the presence of other flagellated autotrophs/mixotrophs (e.g., chrysophytes, dinoflagellates) [8–11]. Previous studies have identified the importance of light (euphotic depth, Zeu) and thermal stratification in predicting the depth and thickness of deep-living phytoplankton communities, but less is known about the influence of chemical (e.g., nutrient) gradients and even less about the chemical ecology of this deep-living biota [4,7,12–14]. Several mechanisms can produce chemical gradients conducive to DCL formation and progression, including upwelling, inter-flow, groundwater, and diffusion from surficial sediments, i.e., internal loading (e.g., [15]). Internal loading often occurs in productive, stratified systems within a hypoxic hypolimnion and can provide a supply of bioavailable dissolved nutrients that are directly accessible to deep-living phytoplankton particularly in shallow waterbodies where Zmix < Zeu [16]. Studies of DCLs in smaller and shallower (<30 m) lakes in Canada report flagellated autotrophs/mixotrophs, e.g., chrysophytes as being the most prominent [1,2,6,17]. However, a more recent study documents DCLs dominated by large cyanobacteria in two nearshore embayments of eastern Georgian Bay (Lake Huron, Ontario, Canada) (e.g., [16]).

Earlier work along the eastern coast of Georgian Bay reported DCLs in a number of sheltered embayments of intermediate size and depth (<30 m), many of which regularly undergo thermal stratification and establish pronounced chemical gradients with depth, [16]. Building on this research, we carried out two nearshore surveys (2014 and 2015) of 15 embayments along this coastline to further investigate the presence of DCLs, and selected two of these waterbodies for more focused work. In 2015, detailed measurements were done in these two embayments to characterize the seasonality and vertical structure of each documented DCL and its phytoplankton composition as well as the coincident physicochemical and nutrient gradients relevant to phytoplankton physiology. We also measured a suite of bioactive metabolites that can play a role in cyanobacterial dominance (the cyanobacterial toxins microcystins, nodularin, anatoxins, saxitoxins, and cylindrospermopsins, and several other metabolic inhibitors known to act at the food web level, anabaenopeptins, cyanopeptolins, and microginins) [13]. Samples from the DCL sites were also obtained in 2018 for additional investigations of these bioactive metabolites, including genetic screening. While the 'toxins' have been widely measured, studies of other metabolic inhibitors have been largely limited to Europe with a few recent reports from North America [13,18,19]. To our knowledge, however, none of these other bioactive metabolites have been characterized in DCLs.

### **2. Results and Discussion**

DCLs were documented in 2 of 15 embayments, Twelve Mile Bay (TMB) and South Bay (SB), during a spatial survey conducted in 2014 (Figure 1, Table 1). In 2015, the full survey was repeated in addition to a more detailed seasonal sampling of TMB and SB (Figure 2) where profiles and discrete depth samples were collected to characterize the vertical structure of phytoplankton composition (Table S1 in Supplementary Materials), the coincident physicochemical and nutrient gradients (Figure 3, Table 2), and a suite of bioactive metabolites (Table 3). In 2018, additional samples were collected from the two DCL sites to confirm previous results on bioactive metabolites (Table 3) and construct vertical water column profiles of corresponding genes (Figure 4).

**Figure 1.** Locations of study sites in Georgian Bay, Lake Huron (Ontario, Canada) denoted by red circles.

**Table 1.** Depth of mixed (Zmix) and euphotic zones (Zeu) during mixed and stratified conditions and presence (Y: Yes/N: No) of deep-chlorophyll layer (DCL) in different embayments of Georgian Bay in the spring (June) and summer (August) of 2015. During mixed conditions in spring (June), Zmix was equivalent to the site depth (Zsite).



<sup>a</sup> Not detected during this study but reported in 2012 and 2014 by [16].

*2.1. Spatiotemporal Changes in Phytoplankton Composition in Near-Surface and Deep Chlorophyll Layers in Twelve Mile Bay*

Overall, TMB phytoplankton showed seasonality and a near-surface biomass, which was composed of a diverse community of flagellates, diatoms and picoplankton, typical of many meso-oligotrophic zones of the Great Lakes (e.g., [20]). While some of these near-surface taxa were also present in the deeper strata, DCLs were often composed of a distinct community, and dominated by trichomes of the cyanobacteria *P.* cf. *isothrix,* which was never observed in near-surface samples. In the summer of 2014, TMB was stratified with a distinct DCL at 10 m depth at the metalimnion, exceeding a total biomass (TB) of 1600 μg/L, almost double that in the epilimnion (Table S1). The DCL was dominated by the buoyancy-regulating cyanobacteria *Planktothrix* cf. *isothrix* (>1300 μg/L, 80% TB), with pennate diatoms (7% TB) and cryptophytes (4% TB) also abundant (Table S1). This deep-living community differed markedly from the epilimnetic assemblage, which was dominated by large flagellates—dinoflagellates (13% TB), cryptophytes (24% TB) and colonial chrysophytes (10% TB).

By autumn 2014, lake turnover brought on isothermal conditions and a more uniform vertical distribution of TB (~1000 μg/L) and composition. Both near-surface (1 m) and deep (10 m) samples had a diverse community dominated by large pennate diatoms (*Fragilaria crotonensis*, up to ~40% TB) and chrysophytes (up to ~40% TB; *Dinobryon sertularia*, *Synura* sp., ochromonads). The deep-living population of *P.* cf. *isothrix* seen during the summer was reduced significantly (<50 μg/L, 4% TB) and, again, this species was not observed nearsurface. We also noted a small population of the nuisance flagellate *Gonyostomum semens* (<40 μg/L, 4% TB) near-surface. While commonplace in low numbers in softwater lakes, this raphidophyte can produce periodic blooms in boreal lakes (e.g., [21]) and outbreaks have been linked to high Fe availability [22].

During the winter ice cover of 2015, low near-surface biomass (<200 μg/L) was dominated by chrysoflagellates (55% TB), cryptophytes (24% TB) and dinoflagellates (3% TB), with low numbers of cyanobacteria also present (7% TB; *Microcystis* sp., *Planktothrix* sp.) (Figure 2, Table S1). Near-bottom (10 m) samples showed a sparse biomass (<40 μg/L), composed of mixotrophic chrysophytes (16% TB) and the large heterotrophic dinoflagellate *Gymnodinium helveticum* (14% TB), along with filaments of *Planktothrix* cf. *isothrix* (10% TB) and the diatom *Aulacoseira subarctica* (7% TB).

In spring 2015, there was a significant increase in biomass, particularly near-surface (>3300 μg/L TB) (Figure 2, Table S1). Near-surface (1 m) and deep (10 and 11.5 m) samples were all dominated by the large diatom *Fragilaria crotonensis* (up to 80% TB); large chrysoflagellates (up to 10% TB) were also important while cryptophytes (10% TB) were only present closest to the bottom (11.5 m). Low numbers of the colonial cyanobacteria *Aphanothece minutissima* and *Microcystis* sp. were present in the two deeper samples, but *Planktothrix* was not detected.

**Figure 2.** Seasonal (W, winter; Sp, spring; Su, summer; Au, autumn) changes in total phytoplankton biomass (line series) and extracted chlorophyll-*a* concentration (bars) in 2015. Near-surface (1 m) water column measurements are in panel (**A**) and at the depth of the recurring metalimnetic peaks, i.e., deep chlorophyll layers (DCLs) (~9 m in Twelve Mile Bay (TMB) and ~7 m in South Bay (SB), see Table 1 and Figure 3) are in panel (**B**). In the absence of DCLs during winter (W) and spring (Sp), i.e., during isothermal conditions, the sample was taken at the depth corresponding to summer DCL peak for comparison, i.e., ~9 m in TMB and ~7 m in SB. See Materials and Methods for sampling dates.

Thermal profiles in summer of 2015 revealed a stratified water column, with a mixed/epilimnetic layer (down to ~7 m) overlaying a steep thermocline to the bottom and no discernible hypolimnion (Figure 3). Epilimnetic biomass (>1700 μg/L) was again dominated by diatoms (65% TB; notably *F. crotonensis*), with large thecate dinoflagellates (13% TB) also being important (Table S1). Cyanobacteria constituted only a small fraction (14%) of this biomass, largely as colonial picocyanobacteria (*Aphanocapsa* sp., *Aphanothece minutissima*) and a few trichomes of the diazotrophic cyanobacteria *Dolichospermum* cf. *fuscum*. At depth, the phytoplankton community segregated into two distinct DCLs (~1 m apart), each showing strong peaks (>20 μg/L) in chlorophyll-*a* fluorescence (Figure 3). The more prominent DCL was concentrated within a ~1 m thick layer at the meta-hypolimnetic boundary (~9 m deep). FP mapping estimated that this DCL extended horizontally over an elliptical area of almost 0.2 km<sup>2</sup> until the end of the first (~9 m) contour (Figure 3, Figure A1). This biomass peak (>1400 μg/L) was dominated by cryptophytes (50%TB; *C. reflexa*), diatoms (13% TB; possibly settling from the epilimnion) and *P.* cf. *isothrix* (8% TB) (Figure 2, Table S1). While water samples from the smaller DCL at 10 m could not be obtained, FP fluorescence indicated an assemblage dominated by "brown/golden" phytoplankton (e.g., dinoflagellates, diatoms, chrysophytes) (Figure 3).

**Figure 3.** Water column profiles in Twelve Mile Bay (panel **A**) and South Bay (panel **B**) for: phosphate (PO4), iron (Fe), manganese, ammonium (NH4) (•), nitrite/nitrate (NO2/3) (-), temperature, chlorophyll-*a*, dissolved oxygen, pH, and redox in summer (August) of 2015. See Materials and Methods section for measurement and analytical protocols used.

In the autumn of 2015, the water column showed a deepened mixed/epilimnetic layer and a single metalimnetic DCL at ~9 m. Near-surface biomass had decreased to <700 μg/L and was dominated by large mixotrophic and non-mixotrophic chrysophytes (38% TB) and thecate dinoflagellates (14% TB) (Figure 2, Table S1). Small populations of colonial picocyanobacteria were also present but *Planktothrix* was not observed. The DCL contained double the near-surface biomass (>1400 μg/L) and was dominated by *Dinobryon* sp. (70% TB); *Peridinium* sp. (9% TB) and *P.* cf. *isothrix* (8% TB) were also important. A sample taken just above the sediments at 10 m showed a significantly lower total biomass (<700 μg/L) but an increased proportion of *P.* cf. *isothrix* (40% TB) (Figure 2, Table S1).

Although *P.* cf. *isothrix* was absent near-surface in TMB throughout all sampling, it was present in samples from both the sediment-water interface and surficial sediments throughout the seasons. Sediments—particularly those collected in winter—yielded (on qualitative inspection) viable trichomes of *P.* cf. *isothrix* under laboratory conditions simulating the summer metalimnion (Z8 media, <15 ◦C and <50 μmol m−<sup>2</sup> s<sup>−</sup>1). This suggests overwintering filaments could act as a reservoir of viable cells for the metalimnetic populations of this species. Recruitment of viable, vegetative cyanobacteria cells overwintering in sediments has been observed elsewhere with bloom-forming cyanobacteria, notably *Oscillatoria* (syn. *Planktothrix*), *Dolichospermum* (syn. *Anabaena*) and *Microcystis* [23–26]; however, in these cases, the benthic cyanobacteria were implicated in supporting surface blooms rather than DCLs. In TMB, our observations suggest that the sediments serve as a reservoir of viable *P.* cf. *isothrix* cells which seed the DCLs, and to our knowledge remain at depth and do not appear in any detectable numbers at or near the surface. While our evidence is supportive, a more direct measurement of this recruitment process is needed to confirm our hypothesis that sediment seeding may play an important role in the origins, seasonal dynamics, and fate of cyanobacteria-dominated DCLs in TMB.

### *2.2. Spatiotemporal Changes in Phytoplankton Composition Near-Surface and within Deep-Chlorophyll Layers in South Bay*

In summer of 2014, SB was stratified with a distinct metalimnetic DCL at 6.5 m, reaching >60 μg/L chlorophyll-*a*—six times that measured near-surface (1 m)—of which ~87% was derived from cyanobacteria (based on FP fluorescence). Although a detailed taxonomic identification and enumeration of the summer 2014 DCL was not possible, results from subsequent years (reported below) suggest *P.* cf. *isothrix* as a prominent and recurring member of these deep phytoplankton communities in SB. Near-surface (1 m) summer biomass in 2014 was ~1000 μg/L, and dominated by chrysophytes (35% TB), dinoflagellates (10% TB), and picocyanobacteria (7% TB) (Table S1). Large cyanobacteria were present but in very low biomass (~3% TB; notably *Dolichospermum*).

By autumn 2014, lake turnover brought on isothermal conditions and biomass had concentrated near-surface, increasing over three times to ~3000 μg/L; dominated by large diazotrophic cyanobacteria (*Aphanizomenon flos aquae* complex, 45% TB; *Dolichospermum planctonicum*, 4% TB), with *Euglena* sp. (9% TB), *Aulacoseira ambigua* (9% TB), and *Plagioselmis nanoplanktica* (8% TB) also being abundant (Table S1).

During the winter ice cover of 2015, biomass near-surface (>400 μg/L at 1 m) was dominated by large colonial, scaled chrysoflagellates (*Synura* sp., 77% TB), with small populations (<5% each) of dinoflagellates and picocyanobacteria (Table S1). In contrast, a much lower biomass at ~8 m depth (<200 μg/L) was composed of picocyanobacteria (*Synechococcus spp*., 45% TB), chrysophytes (37% TB), and dinoflagellates (9% TB).

By spring of 2015, under isothermal conditions, near-surface (1 m) biomass exceeded 1000 μg/L; dominated by a variety of chrysoflagellates (37% TB), cryptophytes (17% TB; mostly *C. reflexa*) and the large, horned dinoflagellate *Ceratium furcoides* (15% TB) (Figure 2, Table S1). At 5.5 m depth, a comparable biomass (>1000 μg/L) had an assemblage of scaled chrysophytes (36% TB; e.g., *Chrysosphaerella longispina*) and large dinoflagellates (*Ceratium* sp., 21% TB) with a few diazotrophic cyanobacteria (*Dolichospermum lemmermanni*, *Aphanizomenon flos-aquae*) also present. Near-bottom (10.5 m) samples had a less biomass

(<400 μg/L), composed mostly of *Cryptomonas* spp. (32% TB), *Chlamydomonas* sp. (14% TB), and *P.* cf. *isothrix* (9% TB).

Profiles in summer of 2015 revealed a stratified water column with near-surface (1 m) biomass (~800 μg/L) dominated by filamentous diatoms (34% TB; *Aulacoseira*), thecate dinoflagellates (15% TB)*,* and the diazotroph *D. planktonicum* (12% TB) (Figure 2, Table S1). A distinct and sizable DCL (>3500 μg/L) was concentrated within a ~1 m thick layer at the meta-hypolimnetic boundary (~7 m). FP mapping estimated that the DCL extended horizontally over an elliptical area of almost 0.03 km2, extending over an area bounded by the ~10 m depth contour (Figure 3, Figure A1). This deep biomass peak was dominated by *P.* cf. *isothrix* (65%); other cyanobacteria were also present (*Aph. flos-aquae* complex, 6% TB; *Microcystis novacekii*, 3% TB) along with dinoflagellates (7% TB) and diatoms (4% TB).

By autumn 2015, lake turnover brought on isothermal conditions and a more uniform vertical distribution of phytoplankton, i.e., no DCL was observed. A similar sized surface biomass (~600 μg/L) showed a shift in community composition from the summer assemblage towards an increased proportion of diazotrophic cyanobacteria (*D. planktonicum*, 32% TB; *Aph. flos-aquae*, 6% TB) and a decrease in the filamentous diatom *Aulacoseira* (16% TB) (Figure 2, Table S1). In the near-bottom sample (~9 m), a comparable biomass (~500 μg/L) showed similar composition (*Aulacoseira*, 17% TB; *D. planktonicum*, 15% TB). As seen in summer, *P.* cf. *isothrix* was again present at depth but at comparably lower proportion (9%), but other cyanobacteria (*Pseudanabaena* sp. 8% TB; *Aph. flos*-*aquae* complex, 7% TB) were also observed.

We note that while numerous other studies have reported DCLs that have been mixed into the epilimnion by physical processes such as turnover, storms, or artificial aeration/mixing and subsequently manifested as ephemeral surface blooms [4,18,27–32], there have been no reports of surface blooms in TMB nor SB near the locations of the DCLs, despite the high density of cottagers in the area. This suggests that the thick metalimnetic plates of *Planktothrix* remain essentially segregated from the surface community and are seeded and sustained by the physicochemical conditions in the bottom layers and sediments.

### *2.3. Light and Pigmentation*

Light is a major factor which often controls phytoplankton vertical structure and the formation of deep-living communities (e.g., [7]), and many taxa modify or supplement their effective light regimes via vertical migration, mixotrophy, and photoadaptation. The abundance of large flagellated mixotrophs in the deep layers in TMB and SB (Table S1) suggests that motility and bactivory may provide access to alternative supplies of energy and nutrients (e.g., [17]); however, these photophagotrophs rarely predominated the DCLs in these embayments. PAR measures showed that the DCLs were positioned within the euphotic zone (Table 1), and we therefore conclude that in situ light levels were sufficient to support the development of *P.* cf. *isothrix*, i.e., light availability was not a primary factor limiting these DCL communities (e.g., [10,33]). Although current literature suggests phycoerythrin-rich *P. rubescens* is more often the predominant *Planktothrix* species in deep-living maxima, the predominant species *P.* cf. *isothrix* remained green-pigmented, suggesting minimal cellular phycoerythrin (Figure 2, Figure A2). We are aware of only one other published report of a DCL dominated by green-pigmented *Planktothrix* [3] although broader geographical surveys are needed [9,14,34–38].

### *2.4. Nutrients and Physicochemical Conditions*

In SB, DCLs dominated by *P.* cf. *isothrix* were located at the metalimnion where sufficient PAR intersected with elevated levels of major nutrients and trace metals, particularly PO4, dissolved inorganic nitrogen (DIN) and dissolved Fe (Table 1, Table 2, Figure 3). Vertical gradients of these chemical constituents increased with depth through the hypolimnion, indicative of internal loading from surficial sediments [16,39–41]. PO4, Fe, NH4, Mn and Co concentrations were up to three orders of magnitude higher near the

sediments as conditions became hypoxic (dissolved oxygen <2 mg/L). Compared to TMB, bottom concentrations were generally higher and gradients were more pronounced in SB—which also exhibited a lower redox potential in the hypolimnion, particularly near sediments (−120 mV vs. −65 mV in TMB). Although similar gradients with depth through the hypolimnion were also evident in TMB for most measures taken, they were not as pronounced as in SB and this, along with markedly lower PO4, is consistent with the significant difference in DCL biomass between the two embayments (Figures 2 and 3). Interestingly, the sheathed bacteria *Leptothrix* were observed in the DCL of TMB but not SB (Figure A2). These chemoorganotrophs exploit DOC-rich environments in the presence of Fe2+ and Mn2+, which they precipitate as ferrous hydroxide and manganese oxide at the interface between oxic/anoxic zones, often co-precipitating PO4 and making it inaccessible to phytoplankton (e.g., [42,43]). Their PO4-precipitating activity in TMB may, to some extent, account for the minimal change in PO4 with depth despite conditions conducive to internal loading but further study to quantify the significance of this mechanism is needed. Differences in sediment chemistry (e.g., sulphate/sulphide or aluminum) could also account for the contrasting PO4 profiles but this has not yet been characterized nor quantified in the two embayments [39–41].

There is a considerable body of evidence to indicate that the development of opposing and intersecting vertical gradients of light and nutrients is strongly correlated with DCL formation in stratified, meso-oligotrophic systems [4,5,7,15,34,44]. Our survey showed that stratified Georgian Bay embayments lacking these opposing and intersecting vertical gradients also lacked discernible DCLs. Some, such as Sturgeon Bay, had a hypoxic hypolimnion and associated gradient in nutrients that reached the metalimnion but did not have sufficient light penetration to support a DCL, i.e., Zeu << Zmix (Table 1). Others, such as Cognashene Lake (actually an embayment), had sufficient light at the metalimnion (Table 1) and elevated nutrients in the hypoxic hypolimnion (e.g., >1400 μg/L Fe, >600 μg/L Mn, >200 μg/L NH4) but it was unclear if these opposing gradients intersected as we lacked sufficient resolution with depth. We speculate that the lake's greater depth and thicker hypolimnion rendered the hypolimnetic nutrients inaccessible to phytoplankton at sufficient PAR (Table 1); however, with more stable and prolonged periods of stratification, and associated hypoxia, as expected with climate change, the nutrient gradient would presumably strengthen.


**Table 2.** Nutrient concentrations (μg/L) and depth ratios in the hypolimnion (Hypo) and epilimnion (Epi) in each embayment (DON = dissolved organic nitrogen). Data are based on the summer 2015 sampling expedition during stratification.

Consistent with our observations, others have observed metalimnetic DCLs most commonly occuring at depths of less than 15 m, particularly those dominated by bloomforming cyanobacteria [4,5,7,31,34,38,44–46]. Although significant DCLs dominated by potentially toxic cyanobacteria such as *Planktothrix* have been reported at depths below 15 m (e.g., [47]), this appears relatively rare [7,46]. DCLs existing deeper than 15 m (close to 30 m depth) are generally dominated by picoplankton, flagellated autotrophs/mixotrophs (e.g., large colonial chrysophytes, dinoflagellates), and diatoms [9–11,15]. The occurrence of

these deeper DCLs has been documented in much larger (>19,000 km2) and deeper systems (>60 m) including offshore Lake Ontario [10,11], Lake Michigan, Lake Superior, and Lake Huron [8,9] and in marine systems (e.g., [15]). The significant depth of these systems and corresponding thickness of the hypolimnion suggests upwelling of bottom waters [15] and recycling from biomass [1,6,48] as more likely sources of nutrients contributing to the DCLs—this, in contrast to the direct access metalimnetic DCLs in the relatively shallower TMB and SB have to the nutrient gradient generated from internal loading at the sedimentwater interface. Although opposing and intersecting physicochemical gradients in nutrients and light are considered key drivers of DCL biomass formation in general; their effects on phytoplankton community composition and activity need further investigation.

### *2.5. Toxins Produced by Cyanobacteria*

Over the course of sampling (2014, 2015, and 2018), near-surface summer samples from 5 of 15 Georgian Bay embayments showed both detectable levels of microcystins and the genetic potential to produce these hepatotoxins (positive for *mcy* gene). The highest concentration observed in any sample (0.4 μg L−<sup>1</sup> in 2018) was well below the Canadian guidelines for safe drinking and recreational water (<1.5 μg/L; <20 μg/L respectively; Health Canada 2017) (Table 3, Figure 4). MC-LA was the most frequently observed microcystin variant and also present at higher concentrations than MC-LR and MC-RR. This is consistent with other reports suggesting MC-LA is more prevalent in systems with lower trophic status (e.g., [13,49]). In both TMB and SB, both the microcystin gene (*mcy*) and the microcystin toxins were present at the metalimnetic DCLs and in the hypolimnion. Based on our 2018 measurements, the *Planktothrix*-dominated DCL in TMB coincided with peaks in microcystin concentration and corresponding gene copies (*mcy*) (Figure 4, Table S1) suggesting these cyanobacteria as the primary source. Other species of *Planktothrix* are known for the production of a wide range of bioactive metabolites including microcystins [18]; however, production by *P.* cf. *isothrix* has yet to be directly confirmed.

**Figure 4.** Vertical water column gradients of bioactive metabolites (microcystins and anabaenopeptins) produced by cyanobacteria and corresponding genes (*mcy* and *apnDTe*, respectively) in Twelve Mile Bay in summer (August) of 2018. Note that higher concentrations of these bioactive metabolites (and their genes) co-occur with biomass peaks, notably of *Planktothrix isothrix*, at depth (Table S1, 2018 data).

The presence of the *mcy* gene and trace amounts of microcystin in the hypolimnion of Tadenac Bay is also noteworthy (Table 3) as this embayment is relatively unimpacted (TP < 10 μg/L, TDP < 5 μg/L) with minimal shoreline development. The presence and low-level expression of the *mcy* gene in Tadenac Bay suggests an innate potential for toxin production despite the oligotrophic nature of this system (Table 3). None of the other toxins for which the samples were screened were detected, i.e., anatoxins, saxitoxins, cylindrospermopsins, nodularins (Table 3). In North Bay, a single sample showed the presence of low copy numbers of the saxitoxin gene (*sxtA*), but not the toxin.

**Table 3.** Major bioactive metabolites (right column) <sup>1</sup> and corresponding genes (left column) <sup>2</sup> (detect "+"; non-detect "-"; not measured "NM"; or maximum concentration (μg/L), any depth) in water samples from embayments along eastern Georgian Bay during 2014, 2015, and 2018. MC: Microcystins, NOD: Nodularin, CYN: Cylindrospermopsins, STX: Saxitoxins, ATX: Anatoxins, APT: Anabaenopeptins, CPT: Cyanopeptolins, MG: Microginins. NOD, CYN, and ATX were not detected in any of the samples.


<sup>1</sup> ELISA and PPIA results verified by LC-MS/MS. <sup>2</sup> PCR primers listed in Table A2 (adapted from [18,50]. <sup>3</sup> Peaks in Twelve Mile Bay were measured in 2018 and the highest concentrations of these bioactive metabolites (and their genes) co-occurred with increasing biomass, notably of *Planktothrix* cf. *isothrix*, at the depth of the deep chlorophyll layer (Figure 4, Table S1, 2018 data).

### *2.6. Other Bioactive Metabolites Produced by Cyanobacteria*

When associated with cyanobacteria, the term 'toxins' is generally applied to the small fraction of bioactive metabolites that affect humans and other large vertebrates; however, many bioactive metabolites produced by cyanobacteria have no known effect on human, pet, or livestock health, but are active towards more proximal elements of the food web including competing phytoplankton, bacteria, pathogens (e.g., viruses, chytrids), and grazers (e.g., [13]). These bioactive metabolites, which include a variety of peptides, have been reported previously from some European and US lakes and are known to inhibit metabolic processes (e.g., protease inhibitors) and function in chemical defence [13,19]. Our study represents a first-time analysis of samples from Canadian waters for both cyanobacterial toxins and these other bioactive metabolites. As seen with microcystins, several anabaenopeptins and one cyanopeptolin (1007) as well as their corresponding genes (*apn, oci*) were detected in both TMB and SB (Table 3); however, clear peaks coincident with the *Planktothrix*-dominated DCL were only evident in TMB (Figure 4). Anabaenopeptins (A, B, and F combined) peaked in 2018 at 6.6 μg L−<sup>1</sup> in the DCL of the meso-oligotrophic TMB, which was comparable to concentrations found in eutrophic surface waters in the United States and Europe (e.g., [51–53]). Anabaenopeptins are thought to play a role in chemical defence via their ability to inhibit proteases used by parasites of cyanobacteria, such as chytrids, to digest their hosts [54]. This anti-parasitic activity could afford cyanobacteria in TMB an additional competitive advantage to establish dominance within the DCL.

Our results indicate that these (and potentially other) bioactive metabolites—typically reported from eutrophic systems (e.g., [51])—are also produced in meso-oligotrophic waters. Furthermore, while most studies have concentrated efforts on surface waters where visible blooms of cyanobacteria are most evident, our data demonstrate the potential importance of these semiochemicals in the often-overlooked communities in DCLs.

### **3. Conclusions**

We observed the annual formation of persistent DCLs in TMB and SB during the stratification period, with a predominance of the large cyanobacteria *Planktothrix* cf *isothrix*. These deep-living layers were formed near the hypoxic hypolimnion at the intersection of opposing vertical gradients of PAR and nutrients, notably Fe, Mn and PO4. Similar DCLs were not observed in the other (13) embayments surveyed along the same coastline, where analogous gradients of PAR and water chemistry were not detected. We also observed the Fe- and PO4-precipitating bacteria *Leptothrix* within the DCL of TMB but not SB, suggesting that microbial processes may contribute to differences in their metalimnetic chemical gradients and DCL composition and biomass. In both embayments, the *Planktothrix*-dominated DCLs coincided with measurable levels of cyanobacterial toxins and other bioactive metabolites (most notably microcystins and anabaenopeptins), along with the associated genes. This is the first report of these compounds in DCLs from mesooligotrophic Canadian lakes, and it merits more focused work to understand if and how these compounds function in the establishment and maintenance of these deep-living communities. The presence of significant numbers of live *Planktothrix* in surficial sediments and overlying water suggest these act as important seed populations to the DCLs, meriting further investigation. Overall, the prevalence of these significant DCLs dominated by large, potentially harmful cyanobacteria across different waterbodies is unknown, and their formation in apparently oligotrophic systems requires more research to understand the driving factors and track their change with changing environmental conditions and anthropogenic development in respective regions.

### **4. Materials and Methods**

### *4.1. Study Sites*

Fifteen embayments along the eastern shore of Georgian Bay, Lake Huron (Ontario, Canada) were surveyed, with a focus on sheltered embayments of intermediate size and depth (<30 m), particularly those that are reported to regularly undergo stratification and establish pronounced chemical gradients facilitated by hypolimnetic hypoxia (Figure 1) [16,55]. All 15 embayments were sampled during summer (July/August) and autumn (September/October) stratification of 2014 and 2015 (Figures 1 and A1, Table A1). Twelve Mile Bay (TMB) and South Bay (SB) were selected, based on the presence of DCLs, for additional and more detailed sampling and analysis in 2015 in winter (February/March), spring (June), summer (August), and autumn (September/October). Based on the earlier results, TMB and SB were sampled again during stratification in the summer of 2018 (July/August) and analyzed for a broader suite of bioactive metabolites beyond the cyanobacterial toxins commonly measured (Table 3).

### *4.2. Physicochemical Profiles of Water Column*

At each sampling, photosynthetically active radiation (PAR; 400–700 nm) was measured with a LI-193 spherical underwater quantum sensor (LI-COR Biosciences, NE, USA) at 0.5 m depth intervals. Measurements were corrected for variance in incident irradiance using a LI-190R quantum sensor (LI-COR Biosciences, NE, USA), and euphotic zone depth calculated (i.e., <1% sub-surface irradiance). A YSI multi-parameter sonde (Xylem Inc., New York, NY, USA) was used to obtain depth profiles of temperature, dissolved oxygen, pH, specific conductivity, redox, and turbidity at each site, while a FluoroProbe (FP) (bbe Moldaenke GmbH, Schwentinental, Germany) was used to measure depth profiles of in situ, fluorescence-based, phytoplankton pigment class-specific chlorophyll ('green algae', 'cyanobacteria', 'brown' algae' (diatoms, chrysophytes and dinoflagellates), and 'cryptophytes').

### *4.3. Water Sampling*

At each site and date, whole water samples were collected from near-surface (1 m) and bottom (1 m from sediments) using a 10 L Niskin water sampler, subsampled into acid washed polyethylene bottles and kept in the dark at 4 ◦C until processed (within 24 h). During stratification, samples were also collected from the middle of the thermocline (metalimnion), the epilimnion (1 m), and hypolimnion (1 m from sediments). Samples were also taken at additional depths where DCLs were detected in the FP profiles.

### *4.4. Sediment Sampling*

Surficial sediments were sampled using a modified gravity corer (Uwitech, Austria). Overlying water was siphoned and surficial 1.0 cm of the core was extruded on-site, placed in pre-labeled Whirl-Pak® bags, and immediately transported to the laboratory on ice in a dark cooler for microscopic imaging before being frozen in the dark at −20 ◦C.

### *4.5. Water Quality Analysis*

Dissolved and particulate components were analyzed using filtrate from a cellulose acetate filter (0.45 μm pore size, 47 mm diameter) and material collected on a Whatman GF/C filter (1.2 μm nominal pore size, 47 mm diameter), respectively. These included major nutrients (dissolved inorganic and organic carbon (DIC, DOC), particulate organic carbon (POC), nitrate/nitrite (NO2/3), ammonium (NH4), total dissolved Kjeldahl nitrogen (TKN), particulate organic nitrogen (PON), phosphorus (P), dissolved phosphorus (DP), phosphate (PO4)), dissolved silica, and extracted chlorophyll-*a,* all of which were analyzed at the National Laboratory for Environmental Testing (NLET, Burlington, Ontario) using standard methods [56]. Dissolved organic nitrogen (DON) was calculated by subtracting NH4 1+ from TKN.

### *4.6. Extraction and Analysis of Cyanobacterial Bioactive Metabolites*

Whole water samples were concentrated onto Whatman GF/C filters (1.2 μm nominal pore size, 47 mm diameter) and stored at −80 ◦C in the dark until extraction. Cyanobacterial toxins and other bioactive metabolites were extracted in 10 mL of analytical grade aqueous methanol (1:1 *v*/*v*) amended with analytical grade formic acid (0.1%) using probe sonication (three 30 sec pulses) (Fisher Scientific Co. Qsonica Sonicator Q500, Ontario, Canada). Samples were then centrifuged at 3000 rpm for 15 min to pellet debris. A PTFE syringe filter (1.0 μm pore size, 30 mm diameter) attached to a 10 mL gas-tight glass syringe was used to filter the resulting supernatant into a glass vial, which was then evaporated to dryness using nitrogen gas-flow and heat (30 ◦C). The resulting residue was reconstituted with 1 mL of analytical grade aqueous methanol (1:1 *v*/*v*) and vortexed. A final filtration was done using a PTFE syringe filter (0.45 μm pore size, 15 mm diameter) attached to a gas-tight glass syringe into a 1.5 mL HPLC amber glass vial and stored at −80 ◦C in the dark until analysis.

Nineteen cyanobacterial peptides and five alkaloids were analyzed by high performance liquid chromatography tandem mass spectrometry (HPLC-MS/MS), as described previously [51]. Certified reference standards of microcystin-LR, -[Dha7]LR and nodularin (NOD-R) were purchased from the National Research Council (NRC) of Canada Biotoxins program (Nova Scotia, Canada). Microcystin-RR, -LA, -LF, -YR, -WR, -LY, -LW, -HtyR, and -HilR were purchased from Enzo Life Sciences (New York, NY, USA). Anabaenopeptin-A, -B, and -F; cyanopeptolin-1007, -1021, and -1040; as well as microginin-690 were purchased from MARBIONC (NC, USA). Anatoxin-a fumarate was purchased from Tocris Bioscience (MN, USA) as a racemic mixture. Homo-anatoxin-a was purchased from Abraxis. Cylindrospermopsin was purchased from Enzo Life Sciences. Saxitoxin and neosaxtoxin were

purchased from the NRC. Each analyte exceeded 95% purity as per the manufacturer's certification.

HPLC-MS/MS analysis was performed on a Sciex 4000 QTRAP (AB Sciex, USA) tandem mass spectrometer equipped with a Shimadzu Prominence HPLC. Peptides were chromatographically separated in 20 μL injections of extracts on a Luna C8 column (Phenomenex, CA, USA) using gradient elution where the mobile phase consisted of A (0.1% formic acid and 5 mM ammonium acetate in HPLC grade water) and B (0.1% formic acid and 5 mM ammonium acetate in 95% acetonitrile). The gradient began at 30% B for 3 min, increasing over a linear gradient to 95% B at 9 min, and held at 95% B until 15 min at which point B was returned to the starting condition for 5 min. The mass spectrometer was operated in positive mode using a scheduled multiple reaction monitoring method. Alkaloids were separated by hydrophilic interaction liquid chromatography (HILIC) (SeQuant®, 5 μm, 150 × 2.1 mm I.D., EMD Millipore Corporation, MA, USA) with mobile phases of (A) HPLC water with 60 mM formic acid and (B) 100% acetonitrile. Isocratic elution (60% B) was used for the HILIC method. The concentration of target analytes was determined based on a linear regression model of peak area relative to the known concentration of an 8-point calibration curve prepared in 1:1 methanol:water. Regression coefficients of R > 0.98 were accepted at an accuracy of >90% at each calibration level.

### *4.7. Taxonomic Identification and Enumeration*

Unfiltered sample aliquots of 100 mL were preserved in Lugol's iodine solution (2% *v*/*v*) for later taxonomic identification, abundance, and biomass using the Utermöhl technique [57].

### *4.8. DNA Extraction and PCR Amplification*

Water samples were filtered onto 0.22 μm pore size, 47 mm polycarbonate filters and stored at −80 ◦C for DNA extraction. Samples were extracted using DNeasy PowerWater kit (Qiagen, Canada) following the manufacturer's protocol.

The availability of a commercial multiplex, quantitative PCR (qPCR) kit for genes of cylindrospermopsin (*cyrA*), microcystin/nodularin (*mcyE*/*ndA*), and saxitoxin (*sxtA*) facilitated the simultaneous measurement of respective gene copy numbers. The qPCR assays were performed using the PhytoxigeneTM kit (Diagnostic Technology, Australia) according to the manufacturer's protocol. Briefly, the lyophilized master mix was first spun down and then reconstituted in PCR-grade water. Each qPCR reaction consisted of 20 μL of the reconstituted master mix and 5 μL of template DNA (10–50 ng) and was carried out in a Bio-Rad CFX96 cycler (Bio-Rad, USA). The cycling conditions consisted of an initial denaturation at 95 ◦C for 2 min, followed by 40 cycles of denaturation at 95 ◦C for 15 sec and annealing-extension at 60 ◦C for 30 s. Gene copies per sample were calculated using a standard curve (target gene copy number vs. Ct) determined for each target gene. Standard curves for all target genes were constructed using standards purchased from the same manufacturer (correlation coefficient and efficiency: *mcyE*/*ndA*: R<sup>2</sup> = 0.999; E = 100.7%; *cyrA*: R<sup>2</sup> = 0.999; E = 100.7%; *sxtA*: R2 = 1.000; E = 102.1%).

All samples were screened by conventional endpoint PCR for the presence of anatoxina synthesis gene and several synthesis genes of oligopeptides know to be bioactive metabolites: aeruginoside, anabaenopeptin, cyanopeptolin, microcystin, microginin, and prenylagaramide. PCR was performed in a total volume of 25 μL composed of 5 μL of GoTaq 10X buffer (Promega), 0.5 μL of dNTPs (10 mM each), 0.5 μL each forward and reverse primers (10 μM), 0.125 μL GoTaq Polymerase (5 units μL−1), 2 μL of template DNA (10–50 ng), and 16 μL of nuclease-free water (see Table A2 for primer sequences). The thermal cycling condition consisted of an initial denaturation step at 95◦ C for 3 min followed by 35 cycles of denaturation at 95◦ C for 1 min, annealing for 30 sec (variable annealing temperature; see Table A2), and extension at 72◦ for 30 s. PCR products were evaluated on a 2% agarose gel.

Anabaenopeptin gene copies were quantified using digital PCR (dPCR). Each dPCR mixture consisted of 7.5 μL of QuantStudio 3D digital PCR master mix v.2 (Thermo Fisher Scientific, Canada), 1.4 μL of each 900 nM forward and reverse primer, 0.75 μL of 250 nM probe, 2 μL of DNA extract, and 195 μL nuclease-free water. The reaction mixtures were loaded into a QuantStudio 3D digital PCR 20K chip v2 (Thermo Fisher Scientific, Canada) using a QuantStudio 3D chip loader (Thermo Fisher Scientific, Canada). The reactions were carried out in a ProFlex thermocycler (Thermo Fisher Scientific, Canada) using the following cycling conditions: 96 ◦C for 10 min, 40 cycles of 60 ◦C for 2 min and 98 ◦C for 30 s, and a final step of 60 ◦C for 2 min. The chips were read on a QuantStudio 3D digital PCR instrument and the results analyzed using the QuantStudio AnalysisSuite software (V 3.2, 2019) (Thermo Fisher Scientific, Canada).

**Supplementary Materials:** The following are available at https://www.mdpi.com/article/10.3390/ toxins13070445/s1, Table S1: Taxonomic identification and enumeration of phytoplankton in Twelve Mile Bay and South Bay embayments of Georgian Bay, Lake Huron (Ontario, Canada).

**Author Contributions:** Conceptualization, A.Z. and S.B.W.; methodology, T.R.M., L.C.W. and H.K.; validation, A.Z.; formal analysis, A.Z. and S.B.W.; investigation, A.Z. and S.B.W.; resources, A.Z. and S.B.W.; data curation, A.Z., S.B.W., T.R.M., L.C.W. and H.K.; writing—original draft preparation, A.Z.; writing—review and editing, A.Z., S.B.W., T.R.M., L.C.W. and H.K; visualization, A.Z.; supervision, A.Z. and S.B.W.; project administration, A.Z. and S.B.W.; funding acquisition, A.Z. and S.B.W. All authors have read and agreed to the published version of the manuscript.

**Funding:** This study was supported by Environment and Climate Change Canada's Lake Simcoe/Southeastern Georgian Bay Clean-Up Fund.

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** Data is contained within the article or supplementary material.

**Acknowledgments:** We would like to extend our sincerest gratitude to the hardworking students and staff, Kaitlin Burek, Shelby Grassick, Flora (Zi Wan) Dong, Jay Guo, Joseph Gzik, Anqi Liang, and the skilled technical team of the Research Support Section, Adam Morden, Daniel Abbey, and Corey Treen, within the Watershed Hydrology and Ecology Research Division at Environment and Climate Change Canada for their assistance in the field and laboratory. Thank you also to Patricia Chow-Fraser and Stuart Campbell at McMaster University as well as Aisha Chiandet at the Severn Sound Environmental Association for their support in the field as well as for their knowledge, guidance, and navigation in planning and execution of our science in this complex watershed. A thank you also to Bob Rowsell, David Depew and Jacqui Milne for their support, encouragement, and engagement in the early stages of the work.

**Conflicts of Interest:** The authors declare no conflict of interest.

### **Appendix A**

**Figure A1.** Location within (**A**) Twelve Mile Bay (TMB), scale 52.16 m and (**B**) South Bay (SB), scale 0.105 km, in Georgian Bay, Lake Huron, Ontario, Canada where each of the deep-chlorophyll layers were sampled, denoted by a filled, red circle. Three-dimensional mapping by Fluoroprobe© revealed that each DCL extended horizontally to the end of the first contour demarcated on the map (~9 m in TMB and ~10 m in SB, approximate extent indicated by red lines) and therefore each of these contours were used to estimate the elliptical area over which each of the DCLs existed. Maps reproduced from Navionics Web API v2, www.navionics.com accessed on January 2021.

**Table A1.** Coordinates and depths of sites under study in Georgian Bay, Lake Huron (Ontario, Canada). All 15 embayments were sampled during summer (July/August) and autumn (September/October) stratification of 2014 and 2015. Twelve Mile Bay and South Bay were selected (based on presence of DCLs) for additional and more detailed sampling and analysis in 2015 in winter (February/March), spring (June), summer (August), and autumn (September/October). TMB and SB were again sampled during stratification in the summer of 2018 (July/August) and analyzed for a broader suite of bioactive metabolites beyond the cyanobacterial toxins commonly measured (Table 3).


**Figure A2.** Microscopic imaging of water sample at the DCL (~9 m) in Twelve Mile Bay showing *Planktothrix* cf. *isothrix* (black arrow) and *Leptothrix* with Fe oxide precipitates (five red arrows). Images are not to scale and are only for representative purpose.


**Table A2.** Primers used to amplify regions of genes coding for cyanobacterial toxins and other bioactive metabolites (oligopeptides).

<sup>1</sup> Primers for cylindrospermopsin, nodularin, and saxitoxin are not listed here since these toxins were only measured using the PhytoxigeneTM kit (Diagnostic Technology, Australia) and the primers are included in the kit's master mix. <sup>2</sup> [18]. <sup>3</sup> [50].

### **References**


### *Article* **Cyanobacterial Toxins and Peptides in Lake Vegoritis, Greece**

**Sevasti-Kiriaki Zervou 1, Kimon Moschandreou 2, Aikaterina Paraskevopoulou 1, Christophoros Christophoridis 1, Elpida Grigoriadou 3, Triantafyllos Kaloudis 1, Theodoros M. Triantis 1, Vasiliki Tsiaoussi <sup>2</sup> and Anastasia Hiskia 1,\***

	- <sup>2</sup> The Goulandris Natural History Museum—Greek Biotope/Wetland Centre, 14th km Thessaloniki-Mihaniona, Thermi P.O. Box 60394, 57001 Thessaloniki, Greece; kmosch@ekby.gr (K.M.); vasso@ekby.gr (V.T.)
	- <sup>3</sup> Water Resources Management Agency of West Macedonia, 50100 Kozani, Decentralized Administration of Epirus—Western Macedonia, Greece; grig.elpida@gmail.com

**Abstract:** Cyanotoxins (CTs) produced by cyanobacteria in surface freshwater are a major threat for public health and aquatic ecosystems. Cyanobacteria can also produce a wide variety of other understudied bioactive metabolites such as oligopeptides microginins (MGs), aeruginosins (AERs), aeruginosamides (AEGs) and anabaenopeptins (APs). This study reports on the co-occurrence of CTs and cyanopeptides (CPs) in Lake Vegoritis, Greece and presents their variant-specific profiles obtained during 3-years of monitoring (2018–2020). Fifteen CTs (cylindrospermopsin (CYN), anatoxin (ATX), nodularin (NOD), and 12 microcystins (MCs)) and ten CPs (3 APs, 4 MGs, 2 AERs and aeruginosamide (AEG A)) were targeted using an extended and validated LC-MS/MS protocol for the simultaneous determination of multi-class CTs and CPs. Results showed the presence of MCs (MC-LR, MC-RR, MC-YR, dmMC-LR, dmMC-RR, MC-HtyR, and MC-HilR) and CYN at concentrations of <1 μg/L, with MC-LR (79%) and CYN (71%) being the most frequently occurring. Anabaenopeptins B (AP B) and F (AP F) were detected in almost all samples and microginin T1 (MG T1) was the most abundant CP, reaching 47.0 μg/L. This is the first report of the co-occurrence of CTs and CPs in Lake Vegoritis, which is used for irrigation, fishing and recreational activities. The findings support the need for further investigations of the occurrence of CTs and the less studied cyanobacterial metabolites in lakes, to promote risk assessment with relevance to human exposure.

**Keywords:** cyanotoxins; microcystins; cylindrospermopsin; cyanopeptides; anabaenopeptins; microginins; aeruginosins; aeruginosamide; SPE; LC-MS/MS; Lake Vegoritis

**Key Contribution:** First report of the co-occurrence of multi-class cyanotoxins and cyanopeptides in Lake Vegoritis. Simultaneous determination of cyanotoxins and cyanopeptides in intra- and extracellular fractions with an extended and validated LC-MS/MS protocol. Three-year monitoring study that revealed the co-occurrence of microcystins and cylindrospermopsin with anabaenopeptins, microginins, aeruginosins, and aeruginosamide in Lake Vegoritis, Greece

### **1. Introduction**

Cyanobacteria are common photosynthetic microorganisms found in lakes and surface water reservoirs, which can, under favorable conditions, grow massively to form blooms [1]. Several cyanobacteria species produce potent toxic compounds as secondary metabolites, called cyanotoxins (CTs) [2,3]. Several incidents of wild and domestic animal poisoning as well as human health effects due to toxic cyanobacterial blooms have been reported [4–7].

**Citation:** Zervou, S.-K.; Moschandreou, K.; Paraskevopoulou, A.; Christophoridis, C.; Grigoriadou, E.; Kaloudis, T.; Triantis, T.M.; Tsiaoussi, V.; Hiskia, A. Cyanobacterial Toxins and Peptides in Lake Vegoritis, Greece. *Toxins* **2021**, *13*, 394. https://doi.org/10.3390/ toxins13060394

Received: 28 April 2021 Accepted: 27 May 2021 Published: 1 June 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

Cyanotoxins comprise a large number of compounds presenting a variety of chemical structures (Figure S1). Microcystins (MCs) [8] and nodularins (NODs) [9] are cyclic peptides typically characterized by the presence of the unique amino acid Adda ((2S,3S,8S,9S)-3 amino-9-methoxy-2,6,8-trimethyl-10-phenyl deca-4,6-dienoic acid) in their structure, which is associated with their hepatotoxicity [2,10,11]. The alkaloid cylindrospermopsin (CYN) is cytotoxic, dermatotoxic, hepatotoxic, and possibly carcinogenic [12,13]. Anatoxin-a (ATX) is a bicyclic secondary amine (2-acetyl-9-azabicyclo (1,2,4) non-2-ene) with acute neurotoxicity [14].

In addition to CTs, cyanobacteria can also produce a wide variety of other metabolites, including compounds of peptide structure such as microginins (MGs), aeruginosins (AERs), aeruginosamides (AEGs), and anabaenopeptins (APs) [15] (Figure S2). MGs are a group of linear oligopeptides characterized by the presence of a decanoic acid derivative, 3-amino-2-hydroxy-decanoic acid (Ahda) at the *N*-terminus [16]. Although not fully investigated, MGs were shown to present strong protease inhibition with MG variants displaying ecotoxicological effects [17]. AERs are linear peptides that include both a derivative of hydroxyl-phenyl lactic acid (Hpla) at the *N*-terminus, the amino acid 2-carboxy-6 hydroxyoctahydroindole (Choi) and an arginine derivative at the C-terminus [18]. Studies on their bioactivity revealed that they inhibit serine proteases trypsin and thrombin, while AER 828A was found to be toxic to *Thamnocephalus platyurus* [19]. The linear peptides AEGs, characterized by the presence of prenyl and thiazole groups, are an understudied group of cyanobacterial metabolites for which limited knowledge is available with regards to their occurrence in cyanobacterial blooms and their bioactivity [20,21]. APs are cyclic peptides with the general structure of X1-CO-[Lys-X3-X4-MeX5-X6]. Lysine (Lys) is present in all variants while X1, X3, X4, X5 and X6 are variable amino acids. A side chain of one amino acid is attached to the ring through an ureido bond with Lys [15]. Recently, it has been reported that APs can be very abundant in nature [22,23]. To date, little is known about the potential health effects of APs on animals and humans [17]. Anabaenopeptin F (AP F) is considered a protease inhibitor and it was shown to inhibit protein phosphatases similarly to MCs [24]. Additionally, Anabaenopeptin B (AP B) and AP F were shown to induce lysis of the cyanobacteria *Microcystis aeruginosa* that can drastically influence cyanobacterial community dynamics and trigger the release of toxins into surface waters [25].

Toxic cyanobacteria blooms occur worldwide [26], with climate predictions suggesting their increase in the future in terms of frequency and severity [27–29]. Therefore, there is an urgent need to monitor toxic cyanobacteria and their toxins, especially in water bodies intended to be used as drinking water supplies or for recreational activities, particularly by children [30]. At the same time, there is need to better assess the occurrence, bioactivity, and effects of other cyanobacterial peptides (CPs) in order to improve risk assessment and the development of management strategies for cyanobacterial blooms.

Lake Vegoritis is a large natural lake covering an area of 60 km<sup>2</sup> in the region of Western Macedonia, in north-western Greece. The banks of the lake are an ideal refuge for many wild birds and it has remarkable fish fauna, which includes a large variety of species. The lake's sensitive ecosystem belongs to the European Network of Protected Areas (NATURA 2000), due to its important habitats and rich biodiversity. A part of its littoral zone was also designated as a bathing area, according to Directive 2006/7/EC. The ecological and historical background of toxic cyanobacterial blooms of Lake Vegoritis, as well as the recreational activities that it offers, create a growing concern about the possible effect of CTs and CPs to its ecosystem and human health. Besides its ecological importance, the lake is used for irrigation, fishing, and recreational activities.

Recently, a multi-lake survey covering 14 lakes in Greece was conducted with the aim to assess the presence of a wide range of CTs from different classes including MCs, NODs, CYN and ATX, using liquid chromatography coupled to tandem mass spectrometry (LC-MS/MS) [31]. In the frame of that study, it was found that water from Lake Vegoritis contained MCs and traces of CYN. Although the occurrence of CTs in Lake Vegoritis was confirmed and documented, it was based only on individual samples and the study was

not designed to provide information concerning the spatial and temporal variation of CTs. Furthermore, there is complete lack of information regarding the co-occurrence of other cyanobacterial peptides such as MGs, AERs, AEGs, and APs.

In response to the above study, a 3-year monitoring program of Lake Vegoritis was initiated in 2018, with the aim to characterize the cyanobacterial species present and to assess the occurrence of various classes of CTs and CPs in the lake. Fifteen CTs (i.e., CYN, ATX, NOD and dmMC-RR, MC-RR, MC-YR, MC-HtyR, dmMC-LR, MC-LR, MC-HilR, MC-WR, MC-LA, MC-LY, MC-LW, and MC-LF) and ten CPs (i.e., MG FR1, MG FR3 MG T1, MG T2, AER 602/K139, AER 298A, AEG A, AP B, AP F and oscillamide (OSC Y)) were targeted. Selection of the targeted cyanobacterial metabolites was based on their frequency of detection in other Greek lakes [31,32]. To implement this monitoring program, a new method for simultaneous determination of various classes of CPs in addition to CTs was developed and validated, extending a previously validated LC-MS/MS analytical protocol [31,33] to include MGs, AERs, AEGs and APs. Using this new protocol, the detection of several classes of CTs and CPs would be possible in a single analytical run.

Results obtained by this 3-year study enable risk assessment and management of toxic cyanobacterial blooms by the lake's authorities. The study's findings also facilitate the reliable and effective communication of the risks to the general public and stakeholders, with regards to the uses of the lake, such as irrigation, fishing, and recreational activities.

### **2. Results and Discussion**

### *2.1. Physico-Chemical Parameters of Lake Vegoritis*

Lake Vegoritis is under pressure from point source and diffuse pollution. As reported by monitoring results from 2012–2015, the lake was in moderate ecological and good chemical status [34]. The concentrations of most of the physicochemical quality parameters, with the exception of nitrates and sulfates, did not fluctuate considerably during the study period (Table S1). The F−, NO2 −, Br− and PO4 <sup>3</sup><sup>−</sup> ions were measured below the method's limit of quantification (LOQ) during the whole study period. The Cl− ranged between 32– 40 mg/L, with a single high measurement (58 mg/L) in July 2020. The NO3 − ions ranged from below LOQ to 1.0 mg/L, with the highest values detected in the winter of 2018 and 2020 and in August 2020. The greatest variability was noticed for SO4 <sup>2</sup><sup>−</sup> levels that ranged between 84 and 200 mg/L, with no seasonal or other temporal pattern observed. The cation (Na+, K+, Mg2+, Ca2+) concentrations measured showed relative stability throughout the study period and no temporal distribution pattern.

In 2018, the highest levels of total phosphorus (TP) were measured. Concentrations ranged from 22 to 60 μg/L and displayed the highest values in February and April (60 and 50 μg/L, respectively). In June 2018, TP decreased to 43 μg/L, while in July 2018 the mean value of the TP concentration was 38 μg/L. Concentrations declined in the following months (<29 μg/L). The following years, 2019 and 2020, TP concentrations were measured at slightly lower levels, 28–50 μg/L and 16–38 μg/L, respectively.

The transparency of the water was measured using the Secchi disc and in June 2018 there was a significant reduction to 0.3 m, from 6.5 m and 6.0 m measured in February 2018 and April 2018, respectively. Such a low value was measured for the first time in the lake throughout the operation of the National Monitoring Water Network (2012–2018). Since then, the transparency of the water in the lake, including bathing area sampling points, presented a noteworthy increase (2.4 m on 4 July 2018 and 3.0 m on 12 July 2017 and 17 July 2018). The Secchi disc was visible to the bottom of the bathing area stations until the autumn. The following years 2019–2020, transparency of the water at the NMWN (National Monitoring Water Network) sampling point ranged from 1.6 m to 4.8 m.

The total suspended solids (TSS) exhibited a high value of 8.46 mg/L in July 2018. During the summer, their concentrations in the lake decreased both at the NMWN sampling point and the bathing area (Site 1). In the next two years, TSS did not exceed 2.54 mg/L, except for two cases in June 2019 (3.75 mg/L) and in September 2020 (5.67 mg/L).

### *2.2. Chlorophyll α*

The concentration of chlorophyll α in Lake Vegoritis in June 2018 was high (15.9 μg/L), with a decreasing trend during the following summer months at both the NMWN sampling point and the bathing area (Site 1) (3.0–8.0 μg/L). This reduction was in line with the improvement in water transparency values observed during the same period. Similar values of chlorophyll α concentration (4.2–7.7 μg/L) were measured during the following years, 2019 and 2020. In two samples collected on 10 September 2019 and 22 September 2020 much higher values (19.4 and 31.8 μg/L, respectively) were measured, but no discoloration of water was observed (Table S1).

### *2.3. Phytoplankton*

During June 2018, the total phytoplankton biomass was estimated at 2.3 mg/L. The most abundant taxa were the green alga *Sphaerocystis schroeteri* (39,889,429 cells/L), cyanobacteria of the genus *Dolichospermum* (14,864,850 cells/L) and the species *Aphanizomenon flos-aquae* (3,383,107 cells/L). Chlorophyta and cyanobacteria comprised 88% of the biomass. In the subsequent samplings, no bloom, mat or scum (as defined in Directive 2006/7/EC) [35] were visually observed. Furthermore, phytoplankton biomass in samples from NMWN gradually decreased, mainly due to the decrease in the biomass of chlorophytes and dinophytes. The *Dolichospermum* biomass also declined sharply (from 0.75 mg/L in June 2018 to 0.02 mg/L in July 2018 and 0.04 mg/L in August 2018).

At the same period, summer 2018, in the samples from the bathing area the cyanobacteria *Microcystis* spp., *Aphanocapsa* spp., *Dolichospermum* spp., and *Aphanizomenon flos-aquae* were dominant. However, their biomass did not exceed 0.7 mg/L (*Aphanizomenon flos-aquae* in July 2018). The biomass values of the dominant cyanobacteria were measured at higher levels in July (1.6 mg/L), followed by sharp decline in the next months. Regarding the chlorophyte *Sphaerocystis schroeteri*, its biomass was measured at lower levels compared to the measurement of June 2018 at the NMWN sampling point and showed a gradual further decrease during August and September.

During the warm period of 2019 the cyanobacteria biomass (four samples from NMWN sampling point) displayed the opposite trend. Biomass gradually increased during summer until the maximum value of 5.4 mg/L (September 2019), where *Aphanizomenon* spp. biomass was estimated at 2.1 mg/L, *Lemmermanniella* spp. at 2.1 mg/L, and *Raphidiopsis raciborskii* at 0.5 mg/L. In all the other three samples of 2019, *Aphanizomenon* spp. and *Dolichospermum* spp. were the main representatives, though with lower biomass values (up to 0.82 mg/L).

No specific trend was observed during the warm period of 2020. In June 2020 the biomass of cyanobacteria was minimal (0.1 mg/L). Over the next three months it increased up to 1.8 mg/L, as measured in August 2020. In July a *Dolichospermum* species dominated (1.1 mg/L), but low biomass values were measured for *Aphanocapsa* cf. *holsatica* and *Cyanodictyon* species (up to 0.2 mg/L). In August and September, species of the genus *Aphanizomenon* prevailed (1.6 and 0.9 mg/L, respectively). Low biomass values of *Microcystis* species (0.1 and 0.3 mg/L) were also measured. Total phytoplankton and cyanobacteria biomass concentrations measured during the study period are given in Table S1.

### *2.4. Occurrence of Cyanotoxins (CTs) in Lake Vegoritis*

A range of CTs (extracellular and intracellular fractions), including CYN, ATX, dmMC-RR, MC-RR, NOD, MC-YR, MC-HtyR, dmMC-LR, MC-LR, MC-HilR, MC-WR, MC-LA, MC-LY, MC-LW, and MC-LF were determined by LC-MS/MS in samples taken during the study (2018–2020). Results are presented in Tables S2 and S3.

In 2018, 14 samples were analyzed from July to November, seven from each one of the sampling sites. None of these samples were found to contain detectable amounts of CYN, ATX and NOD. The analysis of filtered water showed the presence of extracellular MCs, i.e., MC-LR, MC-RR and MC-YR, at concentrations up to 0.029, 0.023 and 0.014 μg/L, respectively. Intracellular (cell-bound) MCs were also detected, including MC-RR, MC-LR, MC-YR, and dmMC-RR at concentrations of up to 0.074, 0.055, 0.026 and 0.003 μg/L, respectively.

During 2019, a total of 24 samples from the two sampling sites were analyzed from April to October. ATX and NOD were not detected in any of the samples. Contrary to findings of 2018 monitoring, CYN was detected mainly in the intracellular (cell-bound) fraction, at concentrations ranging from 0.032 to 0.685 μg/L, and at a lower level in the extracellular (dissolved) fraction. MCs were also detected, with MC-LR and dmMC-LR up to 0.233 and 0.080 μg/L (extracellular fraction) and MC-RR, dmMC-LR, MC-LR, and MC-HilR up to 0.029, 0.055, 0.241, and 0.027 μg/L, respectively (intracellular fraction).

In 2020, 20 samples were analyzed from May to October. Co-occurrence of CYN and MCs was again observed with CYN present during this period at concentrations reaching 0.128 and 0.075 μg/L in extra- and intracellular fractions, respectively. MC-RR, MC-HtyR, dmMC-LR, and MC-HilR (extracellular fraction) were also present at concentrations of up to 0.315 μg/L (22 June 2020, Site 2), while dmMC-RR, MC-RR, MC-YR, MC-HtyR, dmMC-LR, MC-LR, and MC-HilR (intracellular fraction) were up to 0.674 μg/L (7 September 2020, Site 1).

The intracellular and extracellular fractions of MCs and CYN, as well as the total concentrations (sum of intracellular and extracellular) per sampling date and site, are presented in Figures 1 and 2. The occurrence of individual CTs in Lake Vegoritis are shown in Figure 3.

**Figure 1.** Intracellular and extracellular fractions of microcystins (MCs) and cylindrospermopsin (CYN) per sampling date; MCs at (**a**) Site 1 and (**b**) Site 2, and CYN at (**c**) Site 1 and (**d**) Site 2.

**Figure 2.** Total concentration (sum of intracellular and extracellular) of cyanotoxins (CTs) and cyanopeptides (CPs) detected per sampling date at (**a**) Site 1 and (**b**) Site 2.

**Figure 3.** Occurrence of CTs in Lake Vegoritis at (**a**) Site 1 and (**b**) Site 2.

Results for both sampling sites presented a similar seasonal trend in CT concentrations, with an increase during the summer followed by a second outbreak in the autumn. As expected, intracellular MCs tended to appear first, followed by extracellular MCs, due to the release of cell-bound MCs into water after the lysis of cyanobacterial cells, with the exception of June 2020 (Figure 1a,b). The maximum of total MCs, 1.02 μg/L, consisted of dmMC-RR, MC-RR, MC-YR, and MC-LR, and was measured on 7 September 2020 at Site 1, with MC-RR being the most abundant MC variant. The percentage of samples in which each CT was detected is shown in Table 1. MC-LR was detected in 79% of samples through the entire monitoring program, while MC-RR was detected in 50% of samples and was largely absent during 2019.

CYN was also detected in 71% of samples of Lake Vegoritis, while detections occurred only during 2019 and 2020, and not in samples taken in 2018. CYN was found in both extraand intracellular fractions. The maximum concentration of total CYN was 0.727 μg/L at Site 1 on 1 September 2019 (Figure 1c).


**Table 1.** Percentage of samples where cyanotoxins (CTs) and cyanopeptides (CPs) were detected during the monitoring period (2018–2020).

> The presence of MCs could be related to the dominant cyanobacteria *Dolichospermum* spp. (*Anabaena* spp.), *Aphanizomenon* spp., and *Microcystis* spp. that were identified in Lake Vegorits during the study period [36]. The presence of CYN may also be attributed to the cyanobacteria *Aphanizomenon* spp. and *Dolichospermum* spp. as well as to *Raphidiopsis raciborskii* [12,37]. These results concur with a previous study where Lake Vegoritis was found to be mainly dominated by *Dolichospermum* spp., *Aphanizomenon* spp., and *Microcystis* spp. [31]. In the same study, CYN, MC-RR and MC-LR were identified in one sample of biomass from Lake Vegoritis (September 2008) at concentrations of <LOQ, 0.118 and 0.049 μg/L, respectively. In a water sample (July 2014), extracellular dmMC-RR, MC-RR, MC-YR, dmMC-LR, MC-LR, and MC-LY (MC-RR: 104 μg/L and MC-LR 96.3 μg/L) were also detected [31]. Although the findings were based on only two samples, the toxin profile (MC-RR, MC-LR) is in agreement with the present study. However, the concentrations reported in the previous study were far higher, possibly because the sampling was targeted rather than systematic, aiming at localized bloom formations.

> While there are no recreational beach monitoring programs for toxins in Greece, this is the first study to investigate the presence, the concentration, and the diversity of CTs in a popular lake beach of North Greece devoted to recreational activities. In all cases, the measured concentration of CTs did not exceed the provisional guideline values proposed for recreational water by the World Health Organization (WHO) that have been recently updated and were set at 24 and 6 μg/L for MC-LR and CYN, respectively [38,39].

### *2.5. Detection, Identification and Occurrence of Cyanobacterial Peptides (CPs) in Lake Vegoritis*

Although the occurrence of MCs in fresh water bodies is well documented due to the development of analytical protocols [1,31,33,40], data bases [41,42], and a number of commercially available standards, less is known regarding the presence of other CPs. Recent studies showed the presence of CPs in water bodies [17,43], but analysis was mainly done in cyanobacterial biomass, not in water samples. Analytical protocols have not been developed and validated for water samples (extracellular–intracellular fractions), to include cleanup and pre-concentration steps [22]. In this study, we present method performance and validation results for targeted LC-MS/MS analysis of water samples for CPs, based on an analytical workflow previously used for analysis of CTs [33]. The targeted CPs were MG FR1, MG FR3 MG T1, MG T2, AER 602/K139, AER 298A, AEG A, AP B, AP F and OSC Y. The validated method was then used to analyze water samples from Lake Vegoritis.

### 2.5.1. Chromatographic Separation and MS/MS Identification of CPs

A sample of cyanobacterial mass from a bloom in Lake Kastoria, Greece (September 2014) containing all target CPs was used as a reference sample. Efficient chromatographic separation of target CPs was achieved with a reversed-phase C18 column (Atlantis T3, Waters), previously applied for MCs, NODs, CYN and ATX [33].

Identification of CPs was performed using tandem mass spectrometry in a multiple reaction monitoring (MRM) mode (Table 2). The total ion chromatogram (TIC) and MRM chromatograms of the selected quantifier transitions obtained from cyanobacterial mass extract from Lake Vegoritis (7 September 2020, Site1) are presented in Figure 4. MGs presented a characteristic fragment ion, at *m/z* 128.2, attributed to a part of Ahda and, at *m/z* 162.1, to a part of chlorinated Ahda [44,45]. The most intense common ions of MGs, which share the amino acid sequence proline (Pro)-tyrosine (Tyr)-tyrosine (Tyr) at the Cterminus in their structure (i.e., MG FR3, MG T1 and MG T2), were at *m/z* 233.0 [Pro-Tyr-CO + H]<sup>+</sup> and *m/z* 442.2 [Pro-Tyr-Tyr + H]<sup>+</sup> [45,46]. AERs were characterized by *m/z* 140.0 and *m/z* 122.0, which are the Choi immonium ion and dehydrated Choi immonium ion, respectively [47]. Additionally, *m/z* 221.2 was attributed to Leucine (Leu)-Choi fragment (or Isoleucine (Ile)-Choi fragment), while *m/z* 311.0 was indicative of the presence of the arginine derivative—argininol in the structure of AERs [45,48]. APs were characterized by *m/z* 84, which corresponds to the lysine (Lys) immonium ion [49]. The fragment ion *m/z* 201.0 is characteristic of APs that contain arginine (Arg) as a side chain [49]. Fragment ions from the loss of the side chain amino acid with the CO linkage (i.e., *m/z* 637.3 for AP B and *m/z* 651.4 for AP F) or amino acid from the ring (i.e., *m/z* 681.4 for OSC Y [50]) were also considered. AEGs were characterized by *m/z* 112.0 and was annotated in a previous study as TzlCO in case of AEG A, since it is a common ion in the fragmentation spectra of some other AEGs [20]. The structure of fragment ions *m/z* 86.0 and *m/z* 154.2 was proposed in the frame of this study as [PreNH3] <sup>+</sup> and [(Pre)2NH2] +, respectively (Figure 5). The fragmentation pathways of AEG A, involving the *m/z* 154.2 and 86.0 fragment ions based on in silico fragmentation (Mass Frontier 8.0, Thermo Scientific), are presented in Figure S3.



**<sup>Q</sup>** quantifier ion.

**Figure 4.** Example of TIC and MRM chromatograms of quantifier transitions for the intracellular fraction of CPs (sample taken on 7 September 2020, Site 1).

**Figure 5.** Structure of proposed fragment ions of AEG A.

### 2.5.2. Method Performance and Validation Results

The ability of the method for accurate quantification was assessed for AP B, for which an analytical reference standard was available. The response (quantification ion peak area) over the range 5–100 <sup>μ</sup>g L−<sup>1</sup> was linear (r2 ≥ 0.999). Precision, expressed as relative standard deviation (%RSD), was 8.6% under repeatability (*n* = 3) conditions and 15.9% under reproducibility (different days, *n* = 15) conditions. The limit of detection (LOD) of AP B was 0.001 μg/L and the LOQ was 0.003 μg/L. The LOD was estimated from measurements *(n* = 8) of standard solution (5 μg/L) using the formula: LOD = t (*n*−1, 0.95) × SD, where t (*n*−1, 0.95) was the t-test value for *n*—1 degrees of freedom at 95% confidence level, (1.895 for *n* = 8) and SD was the standard deviation of measurements. Limit of quantification (LOQ) was estimated as 3 × LOD.

Quantification of APs was carried out using the class equivalent approach with concentrations expressed as AP B equivalents, while for the rest CPs concentrations were expressed as MC-LR equivalents.

Recoveries of CPs (extracellular and intracellular fractions) were evaluated by analyzing spiked samples using CP-free water and cyanobacterial biomass as matrices and the reference sample from the Lake Kastoria bloom for spiking. Results are presented in Table 3. Recovery experiments were carried out in triplicate and mean recoveries in the extracellular fraction ranged from 77.0–129.2% for all target CPs, except for AEG A which was poorly recovered (17.1%) and AER 602/K139 that showed a recovery of 163.5%. Mean recoveries in the intracellular fraction were in the range of 73.4–98.3%, except for AEG A which had a low recovery (7.5%) (Table 3). In all recovery estimations, %RSD was <28.4%.

The validated analytical protocol can be used for detection and identification of target CTs and CPs of various chemical classes using a single analytical method. It could further serve as a basic template for analysis of cyanobacterial metabolites, expanding to more CTs and CPs in the future as they become commercially available as standards or included in mass spectral databases.


**Table 3.** Recoveries of target CPs from water (extracellular and intracellular).

2.5.3. Occurrence of CPs in Lake Vegoritis

Concentrations of extracellular and intracellular target CPs (MG FR1, MG FR3, MG T1, MG T2, AER 602/K139, AER 298A, AEG A, AP B, AP F, and OSC Y) during the monitoring period are presented in Tables S4 and S5, respectively.

In 2018, MG FR1, MG FR3, MG T1, MG T2, AER 602/K139, and OSC Y were found only in intracellular fraction, with MG T1 to be the most abundant one, reaching 1.16 μg/L. AEG A, AP B, and AP F were found mostly in the intracellular fraction. AER 298A was not detected in any sample. During 2019, none of the samples analyzed were found to contain detectable amounts of MG FR1, MG FR3, MG T1, MG T2, and AEG A. AER 602/K139 and AER 298A were found only in the intracellular fraction. Target APs were all present in both extracellular and intracellular fractions, with the intracellular being at higher concentrations than the extracellular. In 2020, MG FR1, MG FR3, MG T1, MG T2, AER 298A, and AEG A were present only in intracellular form with MG T1 found to be the most abundant, reaching 47.0 μg/L. AER 602/K139, AP B, AP F, and OSC Y were found in both extracellular and intracellular form. AP F was the most abundant, up to 1.382 μg/L in the intracellular fraction, while AER 602/K139 was the most abundant (0.154 μg/L) in the extracellular fraction. The results present a similar trend in the CPs profile in both sampling sites (Figure 6).

**Figure 6.** Occurrence of CPs detected in Lake Vegoritis at (**a**) Site 1 and (**b**) Site 2.

These findings consist of the first report of the occurrence of a variety of CPs, in addition to MCs and CYN, in Lake Vegoritis (Figure S4), showing that all 10 target CPs were detected, with MGs and Aps found to be the most abundant classes of CPs compared to other classes (Figure 7). The highest concentration of total MGs (sum of intra- and extra-cellular), 60.0 μg/L, was measured on 7 September 2020 (Site 1), and consisted of MG FR1, MG FR3, MG T1, and MG T2, where MG T1 was the most abundant. The maximum of total APs (AP B, AP F, and OSC Y), 2.83 μg/L (Figure 2), was measured on 27 May 2019 (Site 2), with AP F presenting the highest concentration (Figure 6).

**Figure 7.** Intracellular and extracellular fractions of APs, MGs, AERs and AEG A per sampling date; APs at (**a**) Site 1 and (**b**) Site 2; MGs at (**c**) Site 1 and (**d**) Site 2; AERs at (**e**) Site 1 and (**f**) Site 2; and AEG A at (**g**) Site 1 and (**h**) Site 2.

To date, there is only one report in the literature related to the presence of MGs in cyanobacterial bloom samples collected from Greek lakes other than Vegoritis, providing only qualitative data with no information on the presence of MGs in the water phase [32]. According to that study, the most frequently detected MGs were MG FR1 (70% of samples) followed by MG T1 (52%). In another study on the occurrence of APs in the freshwater bodies of Greece, the presence of AP B and AP A was reported, however, analysis was carried out using HPLC-UV without confirmation by mass spectrometry [51].

There is still lack of information concerning the occurrence of AERs and AEG A in Greek lakes. Furthermore, there is a gap in the knowledge related to the co-existence of CPs with CTs which may cause adverse health effects to humans and animals. In the frame of the present study, it was found that in water samples of Lake Vegoritis, AP B and AP F, which were detected in almost all samples (100% and 98%, respectively), co-existed with the frequently found MC-LR, CYN, and MC-RR (79%, 71%, and 50%, respectively), (Table 1). OSC Y was also present in 68% of the samples. The frequency of occurrence of the rest of the CPs ranged from 9% to 45%, while the detected CTs ranged from 5% to 24%, respectively.

Cyanobacteria possess a great metabolic potential and are able to co-produce several peptides from different classes [15]. MGs and AEG A exhibited a similar trend in Lake Vegoritis and their production could be attributed to *Microcystis* spp. [21,47,52], since both CP classes were detected in 2018 and 2020, but not in 2019, similarly to *Microcystis* spp. Low concentrations of AERs were determined in all sampling periods and their presence in Lake Vegoritis might be related to *Aphanizomenon* spp. and *Dolichospermum* spp. [53,54] as well as *Microcystis* spp. [18]. *Dolichospermum* spp. (*Anabaena* spp.), the dominant cyanobacterial species during all sampling periods, are possibly related to the high frequency of APs detection, especially AP B, which was present in all samples from Lake Vegoritis [55].

Our results are in agreement with other studies investigating the presence of CPs in inland water bodies. MCs, Aps, and AERs were the main cyanobacterial metabolites identified in biomass during a bloom episode in a dam for drinking water on Lake Occhito, near the town of Foggia in Southern Italy [56]. Similarly, MCs and APs were identified in the biomass from Siemianówka Dam Reservoir (northeast Poland) during a study from 2009 to 2012 [57]. In a more recent study of six eutrophic lakes in USA by Beversdorf et al., APs were detected in all lakes together with MGs at concentrations of the same order of magnitude found in Lake Vegoritis [22]. Similarly, both MCs and CPs (APs, MGs, and cyanopeptolins) were detected in surface and raw drinking waters from the eutrophic Lake Winnebago, Wisconsin [58]. In another study on the diversity and spatial distribution of MCs, NODs, Aps, and MGs in Green Bay, Lake Michigan, an important recreational resource, the presence of MCs (mainly MC-RR and -LR) and CPs (mainly APs and MGs) was reported with the mean of total MCs and APs as 1.28 and 0.20 μg/L, respectively [23].

The occurrence of cyanobacterial metabolites was also reported for lake water samples (mostly recreational) from Canada, in the frame of a collaborative citizen-science project. CTs were present in 75% of the samples, from ng/L up to μg/L, and AP A and AP B were in 38% of the samples at concentrations up to 10 μg/L [59]. High levels of APs (μg mg/L) were also detected in water samples from the Sau-Susqueda-El Pasteral reservoir system in Spain in the autumn of 2015, although MCs were <0.3 μg/L [60].

In most of the published studies, the intracellular (cell-bound) or total concentration (sum of extracellular and intracellular fractions) of CPs was reported. There is a general lack of knowledge on whether CPs are released from cyanobacterial cells into water or remain as cell-bound in the intracellular fraction [58]. In the frame of the current study, as intra- and extracellular fractions were quantified separately, it seems that the CPs studied are mostly intracellular, with the extracellular fraction concentrations being generally lower by an order of magnitude (Tables S4, S5 and Figure 7).

From Figure 2, it was also observed that the maximum concentrations of CPs occurred at the same time period with CTs, following the same trend. Taking into consideration the co-occurrence of CTs with CPs of different classes with unknown behavior, bioactivity and environmental levels in lake water, more research and monitoring programs are urgently needed for assessing possible threats to humans and the environment. This study was the first of this kind in Greece and in Lake Vegoritis, where both recreational and fishing activities take place.

### **3. Conclusions**

This study reports, for the first time, the co-occurrence of CTs and CPs in Lake Vegoritis, situated in the north of Greece. Furthermore, the study describes the variantspecific changes of CT and CP profiles over a 3-year monitoring period as well as the basic water quality parameters of the lake and phytoplankton-cyanobacteria composition.

In order to realize the study, a previously validated LC-MS/MS analytical protocol for the simultaneous determination of multi-class cyanotoxins in water was further extended and validated to be applied for the detection of multiple CPs, i.e., MGs (MG FR1, MG FR3 MG T1, and MG T2), AERs (AER 602/K139 and AER 298A), AEG A and APs (AP B, AP F, and OSC Y). Samples of Lake Vegoritis were analyzed for CTs and CPs in both extracellular and intracellular fractions.

Using the above protocol, CTs were detected in two sampling sites of Lake Vegoritis, during 2018–2020, consisting of MC-LR, MC-RR, MC-YR, dmMC-LR, dmMC-RR, MC-HtyR, MC-HilR and CYN, with MC-LR (79%) being the most frequently detected, followed by CYN (71%). The concentrations of MCs and CYN were generally low (<1 μg/L) and they did not exceed the guideline values proposed for recreational water by WHO (24 and 6 μg/L for MC-LR and CYN, respectively). CPs belonging to the classes of APs, AERs, MGs, and AEG A were also detected, with AP B and AP F present in almost all water samples. The co-occurrence of two potent cyanotoxin classes (MCs and CYN) with multiple CPs in the lake's water supports the need for future studies on the interactions between multiple cyanobacteria metabolites with regards to possible effects on human health.

### **4. Materials and Methods**

### *4.1. Study Area Description and Sample Collection*

Lake Vegoritis, one of the largest lakes in Greece, is located in Western Macedonia region in North-Western Greece and occupies the lowest area of the Ptolemaida basin (Figure 8a). It is considered as one of the most important water resources of Western Macedonia for its multiple uses and benefits for humans [34]. The investigative monitoring of the condition of Lake Vegoritis was designed and implemented from 2018 to 2020. In compliance with the WFD provisions [35], integrated samples (from the euphotic zone, 2.5× Secchi Depth) were taken from the pelagic zone of the lake (40◦44 40.70 'N, 21◦47 3.90 'E, National Monitoring Water Network sampling point, NMWN point), with a Nansen water sampler (Hydro-Bios, Germany). Physicochemical features were measured seasonally and each month during the growing season (May to October). For phytoplankton and chlorophyll α analysis, 2–4 samples were obtained during the growing season. Grab samples (1.5 L) for toxins analysis were collected in disposable plastic bottles from the surface layer of the lake (0–30 cm) at least monthly, from July to November, April to October, and May to October during 2018, 2019 and 2020, respectively. Sampling took place at two points: Site 1, 40◦43 29.5 'N, 21◦45 13.6 'E from the designated bathing area of the lake, and Site 2, 40◦43 12.1 'N, 21◦45 07.2 'E from the pier (Figure 8b). The samples were transported to the laboratory within 24 h of collection, in dark containers and at low temperature (≈4 ◦C).

**Figure 8.** The (**a**) map and (**b**) sampling points of Lake Vegoritis.

### *4.2. Chemicals and Instrumentation*

[D-Asp3]MC-LR, [D-Asp3]MC-RR, MC-WR, MC-HtyR, MC-HilR, MC-LY, MC-LW, MC-LF, and AP B standards were supplied by ENZO Life Science. MC-RR, MC-LR, MC-YR, MC-LA, and NOD standards were supplied by Sigma-Aldrich. CYN was purchased from Abraxis and (±) Anatoxin-a fumarate from TOCRIS Bioscience. All substances had a purity of >95%. Organic solvents i.e., acetonitrile (ACN) and methanol (MeOH) of HPLC grade (99.9%) and dichloromethane (DCM) of analytical reagent grade (99.9%), were supplied by Fisher Chemical. Formic acid (FA) was obtained from Riedel de Haen. Ultra-pure water (18.2 MΩ cm at 25 ◦C) was produced in the lab using a Temak TSDW10 water purification system (TEMAK, Athens, Greece).

Lake water samples were filtered through glass fiber filters with a 47 mm diameter and 0.7 μm pore size (Millipore, Cork, Ireland) and a glass vacuum filtration device. Solid-phase extraction (SPE) was carried out using a 12-port SPE vacuum manifold with large volume samplers connected through PTFE tubes (Supelco, Bellefonte, PA, USA) and a diaphragm vacuum pump (KNF Laboport, Freiburg, Germany). SPE cartridges used for cleanup and pre-concentration purposes were Supel-Select HLB (bed wt. 200 mg, volume 6 mL, Supelco, St. Louis, MO, USA) and Supelclean ENVI-Carb (bed wt. 250 mg, volume 3 mL, Supelco, St. Louis, MO, USA).

The analysis of target analytes was carried out using a TSQ Quantum Discovery Max triple-stage quadrupole mass spectrometer (Thermo Electron Corporation, San Jose, PA, USA), with an electrospray ionization (ESI) source coupled to a Finnigan Surveyor LC system, equipped with a Surveyor AS autosampler (Thermo Electron Corporation, San Jose, PA, USA). Xcalibur software 2.0 was used to control the MS parameters for data acquisition and data analysis. The chromatographic column used was an Atlantis T3 (2.1 mm × 100 mm, 3 μm, Waters, Wexford, Ireland).

### *4.3. Physico-Chemical Parameters*

Ion chromatography (IC) was applied for separation, analysis, and quantification of both anions (F−, Cl−, NO2 −, Br−, NO3 −, PO4 <sup>3</sup>−, and SO4 <sup>2</sup>−) and cations (Na+, K+, Mg2+, and Ca2+) in water samples. Subsamples were filtered through 0.45 μm glass fiber filters within 48 h of sampling and kept at 2–6 ◦C for up to 4 days. For cation analysis, samples were acidified to a pH of 3 ± 0.5 with nitric acid. Dissolved anions and cations were determined using a Dionex Aquion IC System (Thermo Fisher Scientific, Waltham, MA, USA) according to EN ISO 10,304 and 14,911 guidelines, respectively [61,62].

Determination of total phosphorus (TP) was carried out in unfiltered water subsamples (about 250 mL, stored at −18 ◦C for up to six months), using the ascorbic acid method following persulfate digestion [63]. Orthophosphates were determined based on spectrophotometry (Hitachi U-5100 UV/VIS, Hitachi, Tokyo, Japan).

For determination of total suspended solids (TSS), a measured volume of a water subsample (about 500 mL) was filtered through a pre-weighed glass microfiber filter (Whatman Grade 934-AH® RTU circles). The filter was heated at 103–105 ◦C for an hour and then weighed. The TSS were calculated as the mass increase divided by the water volume filtered [63]. Analysis was conducted within seven days from sampling and the subsamples were kept at 4 ◦C in the dark.

### *4.4. Chlorophyll α Analysis*

Subsamples of 1 L were filtered through Whatman GF/F glass fiber filters within 48 h of sampling. Chlorophyll α was measured using 90% acetone and application of the trichromatic equation [63,64]. Absorbance was measured using a Cary 60 UV-Vis spectrophotometer (Agilent Technologies, Santa Clara, CA, USA).

### *4.5. Microscopic Analysis*

For phytoplankton analysis, subsamples (about 500 mL) were transferred to plastic bottles and preserved with acid Lugol's solution [65]. Additional samples were obtained by vertical plankton net hauls (20 μm mesh, Hydro-Bios, Altenholz, Germany) through the euphotic zone, and preserved with formaldehyde. Phytoplankton identification and enumeration were based on the Utermöhl settling technique [66] as described in EN 15,204 guidelines [65]. The analysis was carried out using sedimentation chambers (Hydro-Bios) and an inverted microscope Zeiss AxioObserver.A1 equipped with an AxioCam (Carl Zeiss, Oberkochen, Germany). The phytoplankton biomass was estimated following EN 16,695 guidelines [67].

### *4.6. Analysis of CTs and CPs*

### 4.6.1. Sample Preparation

Analysis of CTs (CYN, ATX, NOD, dmMC-RR, MC-RR, MC-YR, MC-HtyR, dmMC-LR, MC-LR, MC-HilR, MC-WR, MC-LA, MC-LY, MC-LW, and MC-LF) and CPs (MG FR1, MG FR3, MG T1, MG T2, AER 602/K139, AER 298A, AEG A, AP B, AP F, and OSC Y) in water samples was carried out by liquid chromatography coupled with tandem mass spectrometry (LC-MS/MS). For the determination of intra- and extracellular CTs and CPs, water samples were first filtered through GF/F filters and then the filters and filtered water were analyzed. Intracellular CTs and CPs were extracted from the filters' biomass by an extraction mixture containing 75% MeOH:25% H2O. After evaporation of the extract and reconstitution with MeOH: H2O (5:95 *v/v*), the final solution was injected into the LC-MS/MS for analysis [31]. Filtered water samples were pre-treated using the dual cartridge (HLB and Envi-Carb) SPE process [33]. Briefly, water samples, after adjustment to pH 11, were passed through a dual cartridge assembly of HLB and ENVI-Carb. Recovery of extracellular CTs and CPs was achieved by reversing the cartridges and eluting with a mixture of 10 mL DCM:MeOH (40:60, *v/v*), containing 0.5% FA. The extract was dried and the residue was re-dissolved with 400 μL MeOH: H2O (5:95, *v/v*) prior to LC-MS/MS analysis.

### 4.6.2. LC-MS/MS Analysis

A gradient elution program was applied for chromatographic separation with solvents (A) ACN and (B) water, both containing 0.5% FA. The gradient started at 5% A (held for 3 min), which increased to 20% A in 1 min (held for 2 min), further to 35% A in 1 min (held for 7 min), 70% A in 14 min, and finally 90% in 1 min (held for 3 min). An equilibration time of 10 min was kept after each sample run. The flow rate was set at 0.2 mL/min with 20 μL injection volume and the column temperature was set at 30 ◦C.

Electrospray ionization (ESI) in positive mode was used and the three most intense and characteristic precursor–product ion transitions were selected for detection and identification of each compound in MRM mode. LC-MS/MS detection parameters for the target CTs was set according to Zervou et al. [33]. Parameters for the CPs detection are given in Table 2. The selection of MRM transitions was based on fragmentation spectra from previous studies [46,49,50,57,68] or in the frame of this study. In all cases, single protonated [M+H]+ ions were set as the precursor ions. The most intense product ion was chosen to be the quantifier ion, while quantification was performed using external standards at concentrations of 5, 20, and 100 μg/L. For APs, since only AP B was commercially available as a standard, quantification was carried out using the class equivalent approach with AP F and OSC Y concentrations expressed as AP B equivalents. All other CPs for which no class equivalent standard was available, were expressed as MC-LR equivalents [69].

### 4.6.3. Validation of Methods for the Determination of CPs

The analytical work flow used for the target CTs [31] was also validated for its performance for CPs. Recovery studies were carried out by spiking samples with an extract of cyanobacterial mass (Lake Kastoria, September 2014) containing all target CPs. The TIC and MRM chromatograms of the selected quantification ions obtained from cyanobacterial mass extract from Lake Kastoria (September 2014) are shown in Figure S5.

To obtain the extract of cyanobacterial mass used for spiking, 10 mg of lyophilized biomass (Lake Kastoria, September 2014) was extracted two times with 1.5 mL of 75% MeOH: 25% H2O and a third time with 1.5 mL *n*-butanol. Each time the mixture was vortexed, sonicated for 15 min in a sonication bath (Bandelin Sonorex Super RK106) and then centrifuged at 4000 rpm for 10 min at room temperature (DuPont RMC-14 Refrigerated Micro-Centrifuge, Sorvall Instruments, Newtown, CT, USA) and the supernatant was separated from the pellet. All supernatants were pooled together. One milliliter of the extract was evaporated to dryness, the residue was re-dissolved in 400 μL of MeOH: H2O (5:95 *v/v*) and analyzed by LC-MS/MS. The rest of the extract was used for spiking CPfree biomass (retained on filter after passing 200 mL of lake water) and filtered water samples (400 mL) to carry out recovery experiments. Additionally, recovery experiments were carried out by spiking filtered CP-free water samples (400 mL) with AP B at the concentration level of 100 ng/L.

**Supplementary Materials:** The following are available online at https://www.mdpi.com/article/ 10.3390/toxins13060394/s1, Figure S1: structures of several classes of cyanotoxins. For the classes of *Microcystins* and *Nodularins*, Rx stands for variable L-amino acids, Figure S2: structures of several classes of cyanopeptides. For the classes of *Anabaenopeptins* and *Microginins*, Ax stands for variable L-amino acids, Figure S3: fragmentation pathways of AEG A involving the *m/z* 154.2 and 86.0 fragment ions. Fragmentation pathways predicted by Mass Frontier 8.0 software, Figure S4: heat maps showing all CP and CT concentrations detected in Lake Vegoritis throughout the two sampling points (Site 1 and Site 2), Figure S5: TIC and MRM chromatograms of quantifier ions for the cyanopeptides obtained from the biomass extract of Lake Kastoria (sampling September 2014), and Table S1: physico-chemical parameters, total phytoplankton biomass and cyanobacteria biomass concentrations measured during the study period. Table S2: concentrations of extracellular CTs from Lake Vegoritis (μg/L), Table S3: concentrations of intracellular CTs from Lake Vegoritis (μg/L), Table S4: concentrations of extracellular CPs from Lake Vegoritis (μg/L), and Table S5: concentrations of intracellular CPs from Lake Vegoritis (μg/L).

**Author Contributions:** Conceptualization, E.G. and A.H.; Data curation, S.-K.Z., K.M. and V.T.; Formal analysis, S.-K.Z., K.M., A.P., C.C. and V.T.; Funding acquisition, A.H.; Investigation, K.M., E.G., V.T. and A.H.; Methodology, S.-K.Z., K.M., E.G., V.T. and A.H.; Project administration, Theodoros M. Triantis and A.H.; Resources, K.M., E.G., V.T. and A.H.; Software, S.-K.Z.; Supervision, A.H.; Validation, S.-K.Z., T.K. and T.M.T.; Visualization, S.-K.Z. and T.M.T.; Writing—original draft, K.M., V.T. and A.H.; Writing—review & editing, S.-K.Z., T.K. and T.M.T. All authors have read and agreed to the published version of the manuscript.

**Funding:** The cyanotoxins data presented in this research were financed by the Decentralized Administration of Epirus—Western Macedonia through the program "Determination of cyanotoxins in water samples of Lake Vegoritida" (Duration: 4/2019-4/2021). The phytoplankton, chlorophyll α, and physicochemical data used in this research come from Act MIS 5001204 financed by the European Union Cohesion Fund (Partnership Agreement 2014–2020).

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** Data is contained within the article or Supplementary Material.

**Acknowledgments:** The authors acknowledge Vasileios Mixelakis, Coordinator of Decentralized Administration of Epirus-West Macedonia, for his administrative support and encouragement. S.-K.Z. acknowledges the Action titled "National Network on Climate Change and its Impacts—Climpact", which is implemented under the sub-project 3 of the project "Infrastructure of national research networks in the fields of Precision Medicine, Quantum Technology and Climate Change", funded by the Public Investment Program of Greece, General Secretary of Research and Technology/Ministry of Development and Investments.

**Conflicts of Interest:** The authors declare no conflict of interest.

### **References**


### *Article* **Is the Cyanobacterial Bloom Composition Shifting Due to Climate Forcing or Nutrient Changes? Example of a Shallow Eutrophic Reservoir**

**Morgane Le Moal 1, Alexandrine Pannard 1, Luc Brient 1, Benjamin Richard 2, Marion Chorin 1, Emilien Mineaud 1,3 and Claudia Wiegand 1,\***


**Abstract:** Cyanobacterial blooms in eutrophic freshwater is a global threat to the functioning of ecosystems, human health and the economy. Parties responsible for the ecosystems and human health increasingly demand reliable predictions of cyanobacterial development to support necessary decisions. Long-term data series help with identifying environmental drivers of cyanobacterial developments in the context of climatic and anthropogenic pressure. Here, we analyzed 13 years of eutrophication and climatic data of a shallow temperate reservoir showing a high interannual variability of cyanobacterial development and composition, which is a less occurring and/or less described phenomenon compared to recurrant monospecific blooms. While between 2007–2012 *Planktothrix agardhii* dominated the cyanobacterial community, it shifted towards *Microcystis* sp. and then *Dolichospermum* sp. afterwards (2013–2019). The shift to *Microcystis* sp. dominance was mainly influenced by generally calmer and warmer conditions. The later shift to *Dolichospermum* sp. was driven by droughts influencing, amongst others, the N-load, as P remained unchanged over the time period. Both, climatic pressure and N-limitation contributed to the high variability of cyanobacterial blooms and may lead to a new equilibrium. The further reduction of P-load in parallel to the decreasing N-load is important to suppress cyanobacterial blooms and ameliorate ecosystem health.

**Keywords:** cyanobacteria; eutrophication; long term monitoring; water quality

**Key Contribution:** Long term (13 years) data were used to explain high variability in cyanobacterial bloom development and composition in a lowland reservoir; N-reduction and climatic forces (draughts increasing residence time) were identified as the main drivers. The phytoplankton community may be shifting from being dominated by *Planktothrix* sp. in the past to *Microcystis* sp. and *Dolichospermum* sp. in future, with the potential of increased production of cyanobacterial toxins.

### **1. Introduction**

Recurrent and persistent mass developments (blooms) of cyanobacteria are one of the main outcomes of eutrophication of freshwater ecosystems, that is the natural increase in organic matter production and accumulation in an aquatic ecosystem, accelerated during the last decades by human activities [1–4]. Natural control of cyanobacterial blooms, e.g., by zooplankton, is limited by their low palatability owed to long filaments or colony formation and their low nutritional value, besides their toxicity [5]. Cyanobacterial dominance

**Citation:** Le Moal, M.; Pannard, A.; Brient, L.; Richard, B.; Chorin, M.; Mineaud, E.; Wiegand, C. Is the Cyanobacterial Bloom Composition Shifting Due to Climate Forcing or Nutrient Changes? Example of a Shallow Eutrophic Reservoir. *Toxins* **2021**, *13*, 351. https://doi.org/ 10.3390/toxins13050351

Received: 30 March 2021 Accepted: 11 May 2021 Published: 13 May 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

thus causes a misbalance and malfunctioning of the aquatic ecosystem by supressing the biodiversity of zooplankton, phytoplankton (via competition) and submerged macropytes (by shading), thereby disturbing trophic chains and fluxes [5–7]. The decay of phytoplankton including cyanobacteria consumes oxygen, affecting organisms in the water column and ultimately at the sediment. Bottom oxygen depletion limits phosphorus (P) fixation and enhances its release, thus accelerates eutrophication [8].

As various cyanobacteria produce bioactive or toxic metabolites, their mass development threatens organisms in the water body or depending on it [9–11]. To ensure human safety, many countries use thresholds of cyanobacterial cell densities combined with toxin concentration to restrict access for recreational activities, and the WHO has established guidelines of 1 μg L−<sup>1</sup> of total microcystins in drinking water [12,13]. Managers and actors responsible for ecosystem and human health increasingly demand reliable predictions of seasonal cyanobacterial development as those regulations may cause financial consequences, including losses in income via recreational activities or increasing costs for drinking water purification. The different cyanobacteria genera comprise distinctive capabilities to form blooms or to produce specific toxins [6], which underlines the necessity to predict bloom formation, composition, duration and heterogeneity in a given water body.

Despite the urge to return or shift towards the dominance of eukaryotic phytoplankton, this remains difficult, as several physiological mechanisms enable cyanobacteria to outperform eukaryotic phytoplankton. Cyanobacteria grow better in low-light conditions, which occur either during mixing events or due to high phytoplankton biomass, moreover some cyanobacterial species possess gas-vesicles enabling them to adjust their light requirements via buoyancy in the water column [14]. They often benefit from faster uptake kinetics for CO2, a higher temperature optimum and lower N concentrations [14]. Additionally, several species are able to directly use dinitrogen as an N-source [8]. All of these features lead to a higher growth rate during summer, by which cyanobacteria can outcompete eukaryotic phytoplankton, leading to recurrent bloom situations.

Eutrophication is the main driver influencing cyanobacteria blooms, but temperature intensifies bloom frequency, duration and intensity [15]. Shallow lowland lakes suffer more from cyanobacteria because here nutrient loadings meet higher temperatures, and P is easily resuspended from the sediment due to their often polymictic character [6,16]. This applies even more to shallow reservoirs, which continuously receive nutrients (P, N) from the inflowing rivers, charging their sediments as they act as nutrient trap. If not reduced, eutrophication in combination with climate change will thus in future increase cyanobacterial blooms [6,17].

Due to the interaction of the driving factors and the differences from one water body to another, it remains difficult to foresee the development and density of a phytoplankton bloom, and particularly its species composition. Long-term data series help identify environmental drivers of cyanobacterial blooms' occurrence and composition in the context of climatic and anthropogenic pressure and improve predictions for managers. Long term studies previously illustrated the trajectories of eutrophication. This knowledge was applied to reduce point source P pollution that led to a gradual decrease of phytoplanktonic biomass and/or changes in its composition towards eukaryotic species, as demonstrated for eight lakes or reservoirs of many examples by Fastner et al. [18]. The intensification of agricultural practices, however, increased again N and P flux towards adjacent waters, causing numerous lakes and reservoirs to experience or re-experience cyanobacterial blooms. Nevertheless, where effort has been made to reduce agricultural P and N emissions to a lake, it was followed by the amelioration of its water quality [19,20]. Long term (37 years) surveys showed changes in the cyanobacterial community composition towards diazotrophic species as consequence of N decrease while P remained constant, [21], whereas, in another lake, reduction in both N and P decreased also the proportion of N2 fixers in the phytoplankton community [20].

In accordance with their habitat characteristics in terms of depth, nutrient load, temperature, etc, the same species will dominate recurrent blooms in most lakes. *Microcystis sp.* continuously dominates for example Lake Taihu, China [22,23] or the Aguieira reservoir in Portugal [24], while *Planktothrix agardhii* perennially dominates shallow eutrophic water bodies in lowland areas of the Netherlands and Northern Germany [25,26]. Compared to large monospecific blooms, interannual variation amongst cyanobacterial dominance needs better understanding as it can be a good indicator for changes in climatic and anthropogenic pressure. Interannual variations may indicate a shift towards a new state of the composition of the bloom.

The investigated shallow reservoir has received its eutrophication with the incoming rivers during the past decades. It shows a high variability of cyanobacterial development between years, for which the driving factors are not yet identified. Aims were therefore 1) to characterize the seasonal dynamic of the nutrient pattern as well as the phytoplankton and zooplankton succession over a one-year period, and 2) to identify the driving factors of cyanobacterial development over 13 years of monitoring, testing the hypothesis that nutrients, phyto- and zooplankton in the reservoir differ along the transect from entrance to outlet (H1), and that even in this short time series, climatic forcing changes the dynamics of cyanobacterial blooms (H2).

### **2. Results**

### *2.1. Seasonal Cycle*

### 2.1.1. Nutrients' Concentration

Nutrients' concentrations were very similar at the entrance, in the middle and lower basin of the reservoir during the 2018–2019 seasonal cycle (Figure 1). No significant difference was observed between the three stations in terms of nutrient concentrations but nitrogen and phosphorus concentrations revealed different patterns of seasonal variability. Total dissolved nitrogen and nitrate–nitrite were highly correlated (Spearman correlations, rS > 0.92, *p* < 0.001) and reached maximum concentrations in winter. Particulate N was low, but still corresponded to the particulate P concentrations during that period. Using the theoretical composition of the phytoplankton with the Redfield ratio and the ratios given by Reynolds (2006), we find that 150 μg of P corresponds to 1.05 mg N/L (mass ratio of 47:7:1 for C:N:P) and are therefore consistent with particulate N concentrations of 1.25 mg N/L. During periods of high concentrations in N, dissolved nitrogen reached almost 100% of the total N from January to May, as particulate N remained below LOQ for several sampling dates. During that period, nitrate and nitrite accounted for 65% of the total dissolved nitrogen.

Particulate phosphorus, total dissolved phosphorus and phosphate maximum concentrations were recorded during summer, and both particulate P and TDP were correlated (Spearman correlations, rS = 0.65, *p* < 0.001). Phosphorus concentrations of all fractions were very low from January to May, but reached almost 180 μg L−<sup>1</sup> for particulate P and 125 μg L−<sup>1</sup> for TDP at the end of summer, in September 2018. From June to December, total phosphorus was mainly composed of particulate phosphorus for two thirds and TDP for one third. The dissolved fraction accounted on average for 49 ± 24%, 40 ± 10% and 40 ± 13% at the entrance, middle and in the lower basin of the reservoir, respectively. Among this dissolved fraction, more than half was composed of phosphate (mean = 53 ± 10%, 59 ± 23% and 53 ± 12% at the entrance, middle and in the lower basin of the reservoir, respectively). N/P Redfield ratios were low at the entrance of the reservoir (<10 with mean = 6 ± 4), while there were closer to 16 in the middle (>10 with mean = 12.6 ± 1.6) and the lower basin (>10 with mean = 14.9 ± 3.5; Table 1).

**Figure 1.** Nutrients concentration at the entrance, middle and in the lower basin of the reservoir during the 2018–2019 seasonal cycle.

**Table 1.** N/P Redfield ratios of the particulate matter at the entrance, middle and in the lower basin of the reservoir, respectively. N/P Redfield ratios were calculated only for the July–November period, as during the rest of the year, there was no particulate nitrogen and/or phosphorus concentrations were below limit of quantification.


### 2.1.2. Phytoplankton

The relative proportions of the different groups of phytoplankton and the total abundance of cyanobacteria were very similar at the entrance, middle and in the lower basin, showing no significant difference (Figure 2A). The relative contribution of phytoplankton is calculated from cell number, biovolume being unfortunately unknown, as counting was performed at the genus level. Cyanobacteria dominated planktonic community from June to December 2018 and from June to at least August 2019. There were no cyanobacteria in winter and spring, and their concentrations reached 1,500,000 cells mL−<sup>1</sup> in late summer (September 2018). A second bloom of cyanobacteria was observed during autumn 2018 in the middle and the lower basin, reaching 600,000 and 1,200,000 cells mL−1, respectively, in November. Cyanobacterial density was one order of magnitude lower during summer 2019. Phytoplankton eukaryotic community ranged from 200 cells mL−<sup>1</sup> in winter to 72,000 cells mL−<sup>1</sup> in early summer (July 2018). Chlorophytes were the most abundant

in this community almost all yearlong, except in spring (March–April) during which diatoms or in february, where euglenophytes were dominating. In January, there was a co-domnance of chlorophytes with Chrysophytes at the entrance but with diatoms in the middle and lower basin.

**Figure 2.** Phytoplankton and zooplankton dynamics over 2018–2019 seasonal cycle at the entrance, middle and in the lower basin of the reservoir. No statistical differences were observed between abundances of the three stations, except for *Polyarthra* (rotifer), copepods and *Daphnia*, for which the entrance present a lower abundance than the lower basin (*p* < 0.05).

### 2.1.3. Zooplankton

Zooplanktonic community abundance and composition was different between the entrance of the reservoir and middle–lower basin, apart from a common pattern of higher density during summers than winter (Figure 2B). Abundances of the rotifer *Polyathra* were significantly higher in the lower basin compared with the entrance (*p* < 0.05). The Cladocera *Daphnia* and copepods had also higher abundances in both the middle and lower basins compared with the entrance (*p* < 0.05). At the seasonal scale, the rotifer species *Keratella* strongly dominated the micro-zooplankton community during both summers at the entrance of the reservoir, reaching 6000 and 12,000 individuals L−<sup>1</sup> in July 2018 and August 2019, respectively. In the middle and in the lower basin, *Pompholyx* and *Polyarthra* species accompanied *Keratella*, reaching a total maximum concentration of 3000 and 6000 individuals L−<sup>1</sup> in July 2018 and August 2019, respectively. Concerning meso-zooplankton, the Cladoceran *Bosmina* genus dominated the community in the entire

reservoir, together with copepods in a smaller proportion (Figure 2C). The total concentration of meso-zooplankton reached 1200 and 1500 individuals L-1 in the middle and the lower basin during summer 2018, while they were 4 to 5 times lower at the entrance of the reservoir (300 individuals L<sup>−</sup>1). During summer 2019, *Daphnia* developed in the middle and lower basin, dominating at 51% the community in July 2019 in the lower basin. Concentration of Nauplii was quite homogenous in the whole reservoir, ranging between few individuals in winter to 800 individuals L−<sup>1</sup> in summer.

### *2.2. Inter-Annual Variability of Summer Periods*

### 2.2.1. Abiotic Parameters

With the exception of 2010, summers from 2013-2019 were globally warmer, sunnier and dryer compared to 2007–2012 (Table 2). GAMs were applied successfully to temperature, nitrate concentrations, residence time and flow showing their significant changes over the tested time period (Figure 3, Figure S1 and Table 3). Even during this relatively short time period with respect to long term data, a slight increase of temperature was observed (Figure 3A, Figure 4A) During summers 2010 and 2013–2019, either warm temperature (≥20 days above 20 ◦C) or high light intensity (≥10 days above 2800 J m<sup>−</sup>2) or dry months (less than 1.5 million m3 water entering into the reservoir) were recorded (Table 2). Before 2013 (except 2009), a strong and frequent wind was measured in the summers, with at least 19 days of wind with an average daily speed greater than4ms−<sup>1</sup> (Figure 4E, Table 2). After 2013, only 2019 was classified as a windy summer. GAM could not be adjusted on wind due to the high daily variability and explained less than 4% of the deviance (not shown).

The biggest differences between the years of the study concerned the amount of water entering the reservoir during the hydrological year, ranging from 19 to 155 million m<sup>3</sup> per year (Table 2). 2012, 2017 and 2019 were the driest hydrological years, having received less than 40 million m3. This increased the residence time and lowered the level, and water outflow stopped from July onwards (Figure 3B,C, Figure 4C,D). On the contrary, 2007, 2008 and 2014 presented high flow (Figure 3B) all year round (Table 2).

**Table 2.** Hydrological and meteorological characterisation of summer periods between 2007 and 2019. "Annual river inflow" was measured between November of the previous year and October, while "Summer river inflow" was measured between June and September. Dates of downstream flow stop (out of the lake) are indicated and depends on the water level (overflow). The date of the autumn increase in river inflow is also indicated in the last column.


**Figure 3.** (**left**) Time series of the abiotic factors (blue points), showing the adjusted GAM (Table 3) in red. (**right**) Smooth fit with confidence bands performed with mgcv's gam function, showing the effect of year and month (or time for Nitrates) on the abiotic factors.

**Table 3.** Summary of the GAM results performed on the abiotic parameters, with their smoothing functions. Plots are shown in Figure S1. Adj., adjusted.


**Figure 4.** Time series of (**A**) the air temperature, (**B**) the light intensity, (**C**) the residence time and the river inflow, (**D**) the lake water level, (**E**) the mean daily wind speed, (**F**) the nitrates and (**G**) the total phosphorus concentration over 13 years (2007–2019). The light grey areas mark the summer periods. The different dotted lines show tendencies over the 13 years period. The black line in plot D indicates the hight above sea level of the outflow.

### 2.2.2. Nutrients

A strong seasonal pattern was observed for nitrates (Figures 3E and 4F), with maximum concentrations in winter (11.5 mg N-NO3 <sup>−</sup> L−1) and minimum ones in summer. Owing to the low data points (n = 283), GAM was not able to discouple the seasonal pattern from the interannual one, and we smooth only one independent variable, the time (Figure 3D). Nitrate concentrations slightly decreased over the studied period, and since 2009 concentrations were at the LOD at the end of summer periods, except in 2014 (Figures 3D and 4F). Contrastingly, no seasonal pattern nor long-term tendency was observed for total phosphorus for the

2007–2019 period (Figure 4G). Total phosphorus concentration remained high with on average 100 ± 67 μg P-PO4 <sup>3</sup><sup>−</sup> L−<sup>1</sup> over the period studied.

### 2.2.3. Cyanobacterial Blooms

The cyanobacterial community varied strongly interannually, even including years without blooms, 2010, 2014 and 2015 (Figure 5). *Planktothrix agardhii* was the most abundant species, dominating 6 of the 13 monitored years and reaching a biomass of at least 35–55 mm3 L−<sup>1</sup> in 2007, 2011, 2012 and 2018. *Microcystis* (mainly *M. aeruginosa*) dominated in low densities in 2013, 2015 and early 2019, with associated detection of microcystin in low concentration in 2013 and 2015 (Figure 5, Table S1). *Microcystis* and microcystin (-LR, -RR, -YR, -LA, but not the demethylated congeners) were also detected in low density in 2018, a *Planktothrix* dominated year. *Dolichospermum* (mainly *spiroides*) dominated in huge densities in 2017 (75 mm3 L<sup>−</sup>1), and was present to a lesser extent in 2012, 2018 and 2019. Saxitoxin was detected in low concentration in 2017 and 2018. Anatoxin-a and cylindrospermopsin remained below LOD. *Aphanizomenon* dominated the cyanobacterial community in 2009. *Aphanothece* and *Aphanocapsa* (picocyanobacteria) dominated the cyanobacterial community in 2010 despite their very small biovolume per cell. All these species were present during summer 2014, but none dominated or bloomed.

**Figure 5.** Cyanobacterial genus (in biovolumes, areas) and toxins' concentrations (points) over 2007–2019 summer periods in the lower basin of the reservoir.

### 2.2.4. Coupling Blooming Species with Environmental Parameters

To link the cyanobacteria species composition depending on time and environmental parameters, a canonical correspondence analysis has been performed on 195 sampling dates performed during summer. The final CCA included four environmental variables and time, which explain 10.2% of the total variability in the species composition (Figure 6). The CCA discriminated the 12 cyanobacteria species of which 6 formed a bloom. The two first axes of the analysis presented here represent 7.5% of the total variability in species blooming (total inertia) and 73.9% of the constrained inertia. The first axis correlated with time (r = 0.92), hence when samples are grouped by sampling dates, old dates appear on the left while recent ones are on the right (Figure 6a). The first axis correlated moreover to air temperature (r = 0.35), residence time (r = 0.3) and light (r = 0.25) (all towards the positive side).

**Figure 6.** Distance biplot of the Canonical Correspondence Analysis (CCA) linking cyanobacterial biovolumes to environmental parameters measured during the six previous days. Data from summer periods between 2006 and 2017 were used. (**a**) Significant change of cyanobacteria composition with time was tested by permutation test (function envfit). (**b**) Histogramm of the permutation test for the ordination model. (**c**) Biomass of *Dolichospermum* depending on nitrates concentration, showing a threshold effect. ResTim: water residence time, AirTe: mean air temperature, Light: mean daily global radiation, Flow: mean river inflow, and time: sampling dates as open circles. The less significant environmental variables (rainfall, wind in intensity and variability) were eliminated from the analysis based on Akaike Information Criterion (AIC). Cyanobacteria genus by alphabetic order are: Aphaniz: *Aphanizomenon*; Aphca: *Aphanocpasa, Aphth: Ahanothece,* Chroo: *Chroococcus,* Coelom: *Coelomoron,* Dolicho: *Dolichospermum*; Gomo: *Gomphosphaeria*, Lemne: *Lemmermaniella*, Limno: *Limnothrix*, Microc: *Microcystis*; Oscill: *Oscillatoria*, Plank: *Planktothrix agardhii*, Pseud: *Pseudanabaena*, Woro: *Woronichinia.*

> The first axis is also explained by *Planktothrix* (23.7%) and *Gomphosphaeria* (20.3%) on the negative side and *Dolichospermum* (14.2%), *Aphanothece* (13.9%), *Lemmermaniella* (8.3%) and *Aphanocapsa* (5.6%) on the positive side.

> The second axis negatively correlates with air temperature (r = −0.73), residence time (r = −0.50) and light (r = −0.29), and positively with flow (r = 0.31) (Figure 6). Sampling dates with low flows thus coincide with high temperature, residence time and light. Air temperature and residence time thus contribute to both axes. The second axis is explained by *Microcystis* (28.2%) and *Aphanizomenon* (24.3%) on the negative side (bottom part − low flow), and by *Planktothrix* (18.2%), *Lemmermaniella* (11.7%), *Coelomoron* (7.0%) and *Limnothrix* (3.5%) on the positive side (upper part–high flow). All blooming species thus contributed at least to one of the two axes, while *Planktothrix* contributes negatively to both of them, in opposition with air temperature and residence time.

> It should be noticed that nutrients were not included in the CCA analyses because they were measured less frequently and not necessarily at the same times than the phytoplankton community, moreover P concentrations did not change during the period of the data

series. *Dolichospermum* sp. however seems to benefit below concentrations of 2 mg L−<sup>1</sup> of nitrogen (Figure 6c): blooms of *Dolichospermum* were indeed only observed at very low nitrate concentrations.

### **3. Discussion**

A low spatial variability but a high interannual variability in cyanobacteria biomass and composition have been revealed by this study. The seasonal cycle characterized this shallow reservoir as relatively homogeneous, with a similar evolution of nutrient concentrations and phytoplankton abundances, but different patterns for zooplankton along the increasing distance and depth from the entrance to the lower basin. These findings reject our first hypothesis assuming differeces concerning most parameters, but it could be accepted for the zooplankton dynamics after further investigations.

Maximum concentrations of phosphorus and nitrogen were both high, but with strong opposite seasonal patterns: the highest phosphorus concentrations occurred in summer when nitrogen was the lowest. Phosphorus concentrations of the lake were 3 times higher in the reservoir in summer 2018 compared to stations measured in headwaters of the incoming river Yvel [27]. As both nutrients usually enter during the wet winter months [28], their opposing seasonal pattern could indicate either enhanced uptake or denitrification for N, whereas the P concentrations exceeding those of the incoming water could have been released from the sediments during summer, as known from other lakes [20,29,30].

In summer, the Redfield N/P ratios in the total fraction were always below 20, confirming a deficiency of bioavailable N in the water column during this period [31]. This result is in line with an analyse of 369 German lakes concluding that N limitation seems to predominate during summer in shallow polymictic lakes [29].

Phytoplankton and zooplankton succession and densities were similar to many eutrophic water bodies. Total abundances of meso-zooplankton appeared high, but remained in agreement with other shallow eutrophic lakes [32–34]. Small zooplanktonic taxa seemed to dominate at the detriment of larger cladoceans, e.g., *Daphnia* sp., as known from eutrophic temperate lakes of both Europe [34] and the USA [35]. As sampling size was low in this study, more research focusing on zooplankton dynamics in that reservoir would be beneficial.

An interesting element in the seasonal cycle was the interannual variability between the two summers: while the bloom of cyanobacteria reached an exceptional high value of 1.5 million cells mL−<sup>1</sup> in 2018, exceeding the limit allowing bathing by a factor of 15 in France and other countries [36], it was one order of magnitude inferior in 2019. At the same time, in the lower basin the zooplanktonic community switched from a dominance of *Polyarthra* and *Bosmina* during 2018 to a dominance of *Keratella, Pompholyx* and *Daphnia* in 2019, which could have been provoked by decreasing cyanobacteria in the phytoplankton community. Some zooplankton taxa such as the raptorial rotifer *Polyarthra*, or to a lesser extent the cladocerans *Bosmina*, are highly selective feeders avoiding cyanobacteria compared to filter feeders like *Daphnia* that thrive better in the absence of cyanobacteria [37–40]. A zooplankton community can also be top-down controlled by zooplanktivorous fish preferring large zooplankton such as *Daphnia* [41]. Despite the Lac-au-Duc reservoir being a frequented fishing site, the lack of published data on the fish compartment does not allow us to discuss their potential role in structuring the planktonic community.

The contrast of cyanobacteria abundance between the summer of 2018 and 2019 is in the range of the interannual variability of blooms' intensity and species composition in the reservoir Lac-au-Duc observed during 13 years. Within the 2007–2019 period, densities of 40 to 80 mm3 L−<sup>1</sup> were reached during five summers, while in the other eight summers it ranged at maximum from 1 to 20 mm3 L<sup>−</sup>1.

Towards the end of the time series (2018–2019), a dominance or co-dominance of *Dolichospermum* replaced the dominance of *Planktothrix agardhii* until 2013 together with a co-occurance of *Microcystis* (mainly *aeruginosa*) in low densities in 2013, 2015 and early 2019. This shift in 2013 of cyanobacterial dominance seems to correlate to dryer and

warmer summers connected to an increase of water residence time, a slight decrease of wind and a decrease of N sources, which in total confirms our second hypothesis (climate forces as main drivers). Before 2013, with the exception of 2009, summers were indeed characterized by more windy days. Blooms of *Planktothrix agardhii* dominated during that first period, as this species tolerates low average insolation in turbid waters of polymictic lakes [26,41]. *P. agardhii* also grows best at temperatures between 10–20 ◦C [42], and may become disadvantaged by warmer temperatures. Additionally, at the regional scale in Brittany, *P. agardhii* was the dominant taxa of the freshwater cyanobacterial community between 2004 and 2011 [43].

Surprisingly, none of the measured microcystin congeners was detected during *Planktothrix agardhii* blooms despite *Planktothrix* sp. being able to form toxic blooms in temperate freshwater ecosystems [44,45]. Variation of toxin production can be explained by (i) presence/absence of *mcy* genes necessary for their synthesis and (ii) individual variation within *mcy* genotypes with inactivation or regulation at the level of genes expression [46,47]. Natural *Planktothrix agardhii* blooming populations can be composed of microcystin producing (*mcy*) and non-producing (non-*mcy*) genotypes, and their proportion can vary considerably [47–49]. Moreover, it was demonstrated on non-mcy strains of *P. agardhii* isolated from nine European freshwater bodies to have lost more than 90% of their *mcy* genes during evolutionary processes [50]. Based on the analyse of 138 *Planktothrix* strains from three continents, the variable spatial distribution of *mcy* and non-*mcy* genotypes was suggested to depend on ecophysiological adaptation [51]. In laboratory experiments, non-mcy strains of *P. agardhii* seem to have better fitness than *mcy* strains under non-limiting conditions [52]. Similar results were obtained for *Microcystis*: non-*mcy* strains dominated under optimal growth conditions [53]. The authors of these studies then hypothesized that when cyanobacteria grow under favourable environmental conditions, the cost of producing microcystins becomes too high compared to the advantages it can bring [52]. Based on these results, it is tempting to hypothesize that non-limiting growth conditions concerning P and N may have favoured the selection of non-*mcy Planktothrix agardhii* strains in the Lac-au-Duc reservoir. It would be interesting to verify the presence of *mcy* genes, especially since microcystin detections were demonstrated to negatively correlate to *Planktothrix* biovolume in several Brittany lakes [43], suggesting that non-*mcy P. agardhii* strains are prevalent at the regional scale. From the composition of the cyanobacterial bloom, also other toxins, such as anatoxin-a and cylindrospermopsin could have been expected [44,54], they were however never detected during the monitoring between 2007–2019.

Since summer 2013, *Microcystis* and *Dolichospermum* became dominant or codominant, related to generally calmer conditions. Both genera are known to benefit greatly from water column stability as they can regulate their buoyancy according to their requirements in the illuminated area of stable lakes [55–57]. *Microcystis* occurrence was moreover strongly related to light intensity and temperature, which enhances the stabilization of the water column and favoured the development and persistence of *Microcystis* blooms in other lakes as well [16,58]. Temperature also benefits directly cyanobacterial development through their growth rate [59], and has been the most important factor driving the development of *Microcystis* sp. as analysed in more than 1000 lakes [60]. At the regional scale, light intensity was identified as the main climatic driver for *Microcystis* sp. [43]. Despite dominant in 2013 and 2015, *Microcystis* abundances and associated toxin production remained low in Lac-au-Duc compared to other lakes (e.g., [16,24], suggesting that the favourable conditions for its development were not fully met. The optimal growth rate for *Microcystis* is well above 25 ◦C [59], a temperature rarely reached in Lac-au-Duc or in the surrounding region [43]. We can also hypothesize that *Microcystis* was limited by nitrogen in 2018 and 2019, when the species dominated in early summer but disappeared thereafter. Although *Microcystis* is able to use diverse forms of nitrogen, N availability appears to be an essential element controlling the development of this cyanobacteria and its toxin production [61,62].

This context of N limitation in Lac au Duc could also explain the emerging occurrence of the diazotrophic *Dolichospermum* in recent years, as underlined by the threshold above which its occurance decreases. Nitrate concentrations progressively declined during the last decade in the river entering the reservoir, as in other rivers at the regional scale [63]. Thus, N-fixation ability provides an ecological advantage for this cyanobacterium [41].

Indeed, the first occurrence of *Dolichospermum* in 2012 and its largest bloom in 2017 corresponded to the driest hydrological years of the time series, that is the years with the lowest nitrogen recharge. Drought also seems to have an impact on the hydrological functioning of the reservoir: from 2015 onwards, due to the succession of rather dry hydrological years, the water level remained low, when outflow stopped from beginning of July until autumn. Water movements were doubtless strongly reduced and this phenomenon was probably exacerbated during dry summers such as in 2017 and 2018, when the inflow of water ceased already in early summer. Air temperature, residence time and light contributed to both axis of the CCA, therefore jointly with time on the first axis, indicating that both parameters tend to increase over time. The occurrence of *Dolichospermum* was seems partly related to a long water residence time producing a favourable environment for these buoyant cyanobacteria. These results are in accordance with those of Hayes et al. (2015) who found evidence in 42 lakes from agricultural watersheds that droughts strongly influence the system towards N limitation and induce the development of diazotrophic cyanobacteria.

We hypothesize that the lake is in the progress to shift to a new equilibrium, dominated by N2 fixing cyanobacteria. This is potentially connected to toxin production, but evidence for a causal link between reduced N loading and diazotrophic cyanobacteria such as *Dolichospermum* abundance or biovolume is mixed [57]. In addition, evidence is increasing that N2 fixation cannot always compensate significantly for the N deficiency, underlying the need to continue reducing emission of both nitrogen and phosphorus in the catchment [20,64–67]. Thus, a further monitoring of the cyanobacterial community is recommended.

This study provides another proof for the necessity to apply a catchment wide approach to limit P and N entrance into lakes and reservoirs. While reducing N was successful, it remains difficult to reduce non-point source P, despite efforts undertaken by farmers in changing agricultural practices. Moreover, depending on the water-bodies' bathymetry and the P stocked in the sediment, the time to reach and restore less eutrophe conditions needs to be taken into consideration. Many environmental parameters can be intercorrelated (nutrient availability, residence time, surface water temperature, stratification, etc.), making it impossible to separate the individual effect of these factors on a community. The complementary approach to such environmental surveys is to use very long term data, a multiple sites approach or mesocosms experiments. The intensity (frequency, stations and parameters) of long term sampling campaigns may present shortcomings, as in our case due to the restauration attempts with applications of CuSO4 amongst others, thus we were forced to eliminate several of the years, which may have weakened the data set.

### **4. Conclusions**

To conclude, over only 13 years, two major shifts in the cyanobacterial community have been recorded: a first shift in 2013 from a mixing tolerant population of *Planktothrix* to a buoyant species, *Microcystis*, preferring more stable water conditions, influenced by generally calmer and warmer conditions. A second shift from 2017 onwards was driven simultaneously by droughts (representing the biggest change) and reduced N loadings favouring the diazothrophic *Dolichospermum*. Our study also points out that if P reduction is not successful, dominance of *Microcystis* sp. and *Dolichospermum* sp. potentially increase. It is tempting to predict that these buoyant cyanobacteria will replace the *Planktothrix agardhii* population permanently in the context of climate forcing favouring warmer and dryer conditions at the global scale if N and in particular P cannot be reduced below their requirements [45,68]. This could imply many changes in terms of potential toxin production and increased persistance of cyanobacterial populations from year to year

with the growing stock of cells in sediment as both *Microcystis* and *Dolichospermum* have dormant cells, while *Dolichospermum* possesses in addition akinetes.

Changes can also concern curative actions as buoyant cyanobacteria are sensitive to artificial mixing for example, in contrast to *Planktothrix* [69]. Thus, it will be necessary to follow the evolution of future years to confirm or deny the persistent installation of *Microcystis* or *Dolichospermum* in relation to climatic variation.

### **5. Materials and Methods**

### *5.1. Study Site*

The study site, called "Lac au Duc", is one of the largest shallow water bodies (250 ha, 2.6-m depth in mean) in Brittany (western France, Figure 7). This 3-million m<sup>3</sup> recreational and drinking water reservoir drains an agricultural catchment (37,000 ha) with the Yvel river as the main tributary. The lake is used for bathing, fishing and nautical activities.

**Figure 7.** Localization of the Lac-au-Duc in France and stations monitored during the 2018–2019 seasonal cycle (round) for the 13 summer period analyses (triangle). Maps extracted from Geoportail.

Two types of monitoring data were used in this study: (i) data acquired monthly over a 2018–2019 seasonal cycle at three points in the lake and (ii) data acquired almost weekly over 13 summers, from 2007 to 2019 in a bathing zone. Additional environmental data were collected from governmental and meteorological data bases.

### *5.2. Seasonal Cycle 2018–2019*

### 5.2.1. Sampling Sites and Sampling

To examine if nutrients, phyto- and zooplankton were evenly distributed in the reservoir we realized a seasonal cycle. Samples were collected from July 2018 to August 2019 at the entrance, in the middle and in the lower basin of the reservoir (Figure 7) from a boat. Duplicates of five liters of sub-surface water were collected witha1m vertical tube sampler at each point for nutrient, phytoplankton and zooplankton analyses. Duplicate samples for dissolved nutrient analyses were filtered on board on sterile Minisart CA 0.45 μm. Nutrient samples were kept at 4 ◦C until return to the lab, where they were stocked at −20 ◦C until analysis. Temperature and oxygen were measured along depth profiles with an Idronaut Ocean Seven 316 Plus CTD.

### 5.2.2. Nutrients

Particulate phosphorus, total dissolved phosphorus (TDP), particulate nitrogen and total dissolved nitrogen (TDN) were measured by colorimetry after digestion with persulfate according to [70], with a limit of detection of 6 μgPL−<sup>1</sup> and 50 μgNL−1. Orthophosphate (PO4 <sup>3</sup>−) was analysed by the ammonium molybdate method [71] with a limit of detection of 3 μgPL<sup>−</sup>1. After nitrate (NO3 −) reduction into nitrite (NO2 −) with vanadium chloride, NO2 − (originally present and reduced nitrates) was measured by colorimetry using sulphanilamide and N-1-naphthylethylenediamine dihydrochloride [72], with a limit of detection of 50 μgNL−1. Colorimetric measurements were realised using a Gallery Photometric Analyser Gallery Plus (Thermo Fisher, Saint Herblain, France).

### 5.2.3. Phytoplankton

For the seasonal cycle, microalgae and cyanobacteria composition was determined directly in the fresh sample when possible or were preserved in Lugols solution and stored at 4 ◦C. Identification were realized to the genus level according to Komárek. Phytoplankton cell counts were carried out with a Nageotte cell according to [73]. For large colonies, the number of cells was estimated using photos. Microscopic photos were used to assess the number of cells in colonies of picocyanobacterial, using Pegasus software. At least, 400 individuals were counted per sample. For both, a light microscope (Olympus BX 50, Rungis, France) was used. Prior to cell counts and identification, if necessary, microalgae and cyanobacteria were concentrated on a 1 μm Poretics polycarbonate membrane filter, thanks to a filtration pump but with low vacuum pressure. Transfer and resuspension of cells in 1 ml of water was done while the membrane was still wet. Intact colonial chlorophytes and cyanobacteria were checked on the microscope to ensure that damage due to the concentration of cells was low.

### 5.2.4. Zooplankton

Zooplankton was concentrated from two liters of water by filtering through a 50-μm net, then narcotized with soda water and preserved in 70% ethanol at 4 ◦C until counting. For counting, 1 mL of the concentrated sample was distributed on a Sedgewick-Rafter counting chamber. For each sample, a minimum of 400 rotifers were counted and identified at the genus or species level, according to USEPA protocol [74] with a Zeiss microscope, based on the identification manuals for rotifers [75] and cladocerans [76]. In low-abundance samples, the totality of the 1 mL was observed. In addition, crustaceans and cladocerans were counted.

### *5.3. Long Term Series 2007 to 2019*

### 5.3.1. Phytoplankton and Toxins

Each summer since 2004 the regional public health authorities (French Agence Régionale de Santé, ARS) of Brittany monitor water from a bathing zone located in the lower basin of the reservoir (Figure 7). Depending on the bloom density, samples were collected weekly from June to September, except in 2009 (20/7 to 31/8) and 2010 (24/6 to 30/8). Monitoring stopped in mid-September in 2015–2019, while it continued until end of October in 2011 and 2012. Cyanobacteria abundance and species composition as well as toxin concentrations were analysed by local laboratories. Toxins were analysed with LC-MS/MS [77] after extraction with methanol. Microcystins analysed comprised the congeners microcystin—LR, YR, RR, LF, LW, LY, LA, desmethyl-LR and desmethyl RR with a limit of quantification (LOQ) of 0.3 μg L<sup>−</sup>1. MC-LR-eqivalents were calculated according to Wolf and Frank [78] and Fastner and Humpage [79] with MC-LR (1 per definition), MC-LA, MC-YR, MC-YM (1.0) and MC-RR (0.1). Saxitoxin was determined with an LOQ of 0.6 μg L<sup>−</sup>1, anatoxin LOQ < 0.3 μg L−<sup>1</sup> and cylindrospermopsin LOQ < 0.6 μg L<sup>−</sup>1. All cyanobacterial cell densities data were converted to biovolume with the use of relevant geometric formulas and data literature [80,81], to account for differences in contribution from larger species and smaller species.

### 5.3.2. Abiotic Parameters

Daily meteorological data (air temperature, global radiation intensity and wind speed) were collected from Météo-France database from 2007 to 2019. Measurements were realised at the Météo-France station of Ploërmel, located 0.5 km from the reservoir (Figure 7). The

light energy sensor broke in July 2017, from this date measurements from the Rennes-St-Jacques Météo-France station, located 60 km from the reservoir, were used. The number of days with an air temperature above 20 ◦C, or with solar radiations greater than 2800 J cm−<sup>2</sup> were calculated. The water flow as well as nutrient concentrations were measured in the Yvel river, 3 km upstream the entrance of the reservoir (Figure 7). The water flow data were provided for the 2007–2019 period from the national database Hydro France [82] and corresponded to the daily mean values calculated from continuous stage records. From these data, we calculated the residence time of the water in the reservoir, knowing that its volume is approximately 3 million m3. To take into account the time needed between climate variation and phytoplankton response, hydrological and meteorological data were averaged over 6 days preceding the sampling date of cyanobacteria. Wind variation completed the average data. Windy summers were identified by the number of days of wind above a threshold of4ms−<sup>1</sup> during a 24 h average following [82]. Monthly data for nitrate and total phosphorus concentrations were provided from the database of the Yvel-Yvet watershed manager (the Syndicat Mixte du Grand Bassin de l'Oust, SMGBO) from 2007 to 2019. By combining these nutrient data with flow data, we calculated the quantities of NO3 − and Ptot entering the reservoir over the summer period or the complete hydrological year (November to November). The city of Ploërmel also provided reservoir water level data from the drinking water purification station.

### 5.3.3. Statistical Analyses

Despite being available, we discarded data from 2002 to 2005 as some copper sulphate treatments had been applied in the whole reservoir up to summer 2005. Treatments with calcium carbonate and hydrogen peroxide also took place, but locally in the bathing area, between 2013 and 2015 and in July 2018, respectively. During these years, except 2018, no significant difference was observed between the bathing area and the rest of the lake. We then chose to keep the data from the long-term monitoring bathing area for 2013–2015, while we used data from outside of the enclosed area only in 2018. Monitoring of nutrient concentrations started in 2007, thus data from 2006 were not used in analyses, leaving an extra year after the copper sulphate treatments.

All statistical analyses were performed in R Studio version 1.2.5019 [83]. In order to link cyanobacterial blooming genera with environmental parameters and characterize the sampling dates, a canonical correspondence analysis (CCA) was performed on Hellinger transformed genera biovolume (response matrix) and centred-reduced environmental parameters (explanatory variables). To reduce the number of explanatory variables, the ordistep function of the vegan package has been used to find the most parsimonious model based on Akaike Information Criterion (AIC). The less significant explanatory variables were eliminated from the CCA, in order to identify only the significant ones. The significance of the final CCA has been tested through a permutation test with the function envfit. A reference distribution under H0 from the data themselves is generated by permutations of rows and columns and by calculating the new percentage of constrained variance (sum of all canonical eigenvalues). The collinearity between explanatory parameters has also been checked.

We used generalized additive models (GAM) for time series of abiotic parameters, to analyze their temporal dynamics according to the season and the year. GAM takes into account the nonlinear response between the dependent variable (the abiotic parameter) and the explanatory variables (time, month of the year and year), using smooth functions called splines. Finally, the final shape of the relationship is determined by the data themselves. The independant variables in the initial model were the month of the year and the year, except for nitrates, for which it was only time (date) due to little data (Table 3). The best model was selected based on the gain in deviance explained (DE) relative to the initial model, while minimizing the Akaike's information criterion (AIC). Models were adjusted using the R package 'mgcv' and the function gam() with the selection of the Restricted

Maximum Likelihood (REML) method. The family was Gaussian and the link function was identity. A Log 10 transformation was done on the abiotic parameter when needed.

**Supplementary Materials:** The following are available online at https://www.mdpi.com/article/10 .3390/toxins13050351/s1, Figure S1. Plots from the function gam.check () for each abiotic parameters: At the top left, the Q—Q plot compares the model residuals to a normal distribution., Table S1. Microcystin types meas-ured over the 2007–2019 summer periods.

**Author Contributions:** Conceptualization, M.L.M., A.P. and C.W.; methodology, M.L.M., A.P. and C.W.; formal analysis, M.L.M. and A.P.; investigation, M.L.M., M.C., L.B. and E.M.; resources, B.R.; data curation, A.P.; writing—original draft preparation, M.L.M.; writing—review and editing, A.P., C.W., L.B. and B.R.; visualization, M.L.M. and A.P.; supervision, C.W. and L.B.; project administration, C.W.; funding acquisition, C.W. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by the Interreg CPES program 2017–2020 (https://www.cpesinterreg.eu; last accessed date 12 May 2021).

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** Long term data series were obtained from the following sources: Cyanobacteria: ARS Bretagne, France; Temperature, radiation and wind speed: Meteo France, Ploermel, respectively, Rennes, St Jaques, nutrients: Syndicat Mixte du Grand Bassin de l'Oust, SMGBO, Bretagne, France.

**Acknowledgments:** We thank the French regional public health agency of Brittany for providing the data on cyanobacteria, and Meteo France and DREAL Bretagne-Hydro France for meteorological and hydrological data, respectively. We also thank the Syndicat Mixte du Grand Bassin de l'Oust for providing data on nutrients, as well as the city of Ploërmel for the data on the water level in the reservoir. The nautical base of Ploërmel is thanked for its help collecting samples. We thank also André-Jean Francez for his kind help in zooplankton identification, as well as Enora Briand for her useful comments on the discussion on microcystin production. Nutrient analysis were conducted at the Plateforme EcoChimie (EcoChim, UMS OSUR 3343).

**Conflicts of Interest:** The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.

### **References**


### *Article* **Cyanotoxin Screening in BACA Culture Collection: Identification of New Cylindrospermopsin Producing Cyanobacteria**

**Rita Cordeiro 1,2,\*, Joana Azevedo 3, Rúben Luz 1,2, Vitor Vasconcelos 3,4, Vítor Gonçalves 1,2 and Amélia Fonseca 1,2**

	- <sup>3</sup> Interdisciplinary Centre of Marine and Environmental Research—CIIMAR/CIMAR, University of Porto, Terminal de Cruzeiros do Porto de Leixões, Av. General Norton de Matos s/n, 4450-208 Matosinhos, Portugal; joana.azevedo@ciimar.up.pt (J.A.); vmvascon@fc.up.pt (V.V.)
	- <sup>4</sup> Department of Biology, Faculty of Sciences, University of Porto, 4069-007 Porto, Portugal

**Abstract:** Microcystins (MCs), Saxitoxins (STXs), and Cylindrospermopsins (CYNs) are some of the more well-known cyanotoxins. Taking into consideration the impacts of cyanotoxins, many studies have focused on the identification of unknown cyanotoxin(s)-producing strains. This study aimed to screen strains from the Azorean Bank of Algae and Cyanobacteria (BACA) for MCs, STX, and CYN production. A total of 157 strains were searched for *mcy*, *sxt,* and *cyr* producing genes by PCR, toxin identification by ESI-LC-MS/MS, and cyanotoxin-producing strains morphological identification and confirmation by 16S rRNA phylogenetic analysis. Cyanotoxin-producing genes were amplified in 13 strains and four were confirmed as toxin producers by ESI-LC-MS/MS. As expected *Aphanizomenon gracile* BACA0041 was confirmed as an STX producer, with amplification of genes *sxt*A, *sxt*G, *sxt*H, and *sxt*I, and *Microcystis aeruginosa* BACA0148 as an MC-LR producer, with amplification of genes *mcy*C, *mcy*D, *mcy*E, and *mcy*G. Two nostocalean strains, BACA0025 and BACA0031, were positive for both *cyr*B and *cyr*C genes and ESI-LC-MS/MS confirmed CYN production. Although these strains morphologically resemble *Sphaerospermopsis*, the 16S rRNA phylogenetic analysis reveals that they probably belong to a new genus.

**Keywords:** microcystin; saxitoxin; cylindrospermopsin; ESI-LC-MS/MS; 16S rRNA phylogeny; Azores

**Key Contribution:** 16s rRNA phylogenetic analysis revealed evidence for new toxic cyanobacteriataxa, in two identified CYN producing nostocalean strains BACA0025 and BACA0031.

### **1. Introduction**

Due to their impacts on ecosystem degradation, public health risk, and associated economical losses, cyanotoxins are one of the most studied primary and secondary metabolites from cyanobacteria [1–4]. Microcystins (MCs), Saxitoxins (STXs), and Cylindrospermopsins (CYNs) are some of the more well-known cyanotoxins, for which a high number of studies were done covering different aspects such as their taxonomic distribution, genetic organization, biosynthesis pathways, and mechanisms of action [5–10].

Microcystins are hepatotoxins produced by non-ribosomal pathways, and the most common and more well-studied cyanotoxins, with over 240 variants (e.g., MC-LR, MC-LY, MC-RR). Mostly produced by *Microcystis aeruginosa* [11], MCs are, however, currently known to be produced by over 40 species [12]. The gene cluster responsible for the biosynthesis of MCs (*mcy*) contains 10 encoding genes organized in two operons (*mcy*A-C

**Citation:** Cordeiro, R.; Azevedo, J.; Luz, R.; Vasconcelos, V.; Gonçalves, V.; Fonseca, A. Cyanotoxin Screening in BACA Culture Collection: Identification of New Cylindrospermopsin Producing Cyanobacteria. *Toxins* **2021**, *13*, 258. https://doi.org/10.3390/toxins 13040258

Received: 8 February 2021 Accepted: 1 April 2021 Published: 3 April 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

and *mcy*D-J) in which each gene has a specific function for the toxin synthesis, regulation, and release [13,14]. This hydrophilic peptide has an Adda group in its structure, responsible for its toxicity [15,16]. With over 240 analogs, other groups as ADMAdda have been identified and characterized as responsible for MCs more toxic variants [17].

The alkaloid STX is one of the most potent natural neurotoxin [18] identified in both cyanobacteria and dinoflagellates [19]. In cyanobacteria, STXs have been identified in several species as *Aphanizomenon gracile*, *Cylindrospermum stagnale* [20], *Dolichospermum circinale* [21], and *Raphidiopsis raciborskii* [22]. The gene cluster *sxt* encodes 26 genes responsible for the biosynthesis of STXs [19] and its gene organization varies between species, which may indicate gene loss or transfer between species [6,23,24]. This scenario has been observed in the STX gene cluster where the *sxt*A gene, recognized as the encoding gene for the polyketide synthase-like structure involved in the first step of STXs production [6], was found in both STXs-producing and non-producing *A*. *gracile* strains, but all non-producing strains lost at least one of the other genes of the cluster (e.g. *sxt*G, *sxt*H, *sxt*I, or *sxt*X) [25].

Cylindrospermopsins are cytotoxic alkaloids initially identified in *Raphidiopsis raciborskii* [26] but also reported to be produced by other species as *Aphanizomenon gracile* [27] or *Chrysosporum ovalisporum* [28]. As MCs, CYNs biosynthesis is done by a non-ribosomal pathway, where the gene cluster *cyr* (15 encoding genes) has been identified as responsible for CYNs synthesis, regulation, and exportation in *R. raciborskii* [29].

Cyanobacteria have high morphological and ecological variability [30], and due to this, its taxonomical classification is relatively intricate and is in constant change [31]. New cyanobacteria families, genera, and species, as Oculatellaceae and Trichocoleaceae [32], *Aliinostoc* [33], *Halotia* [34], and *Lusitaniella* [35], and *Hyella patelloides* [35] and *Compactonostoc shennongjiaensis* [36] have been reported in recent years. These taxonomic redefinitions are challenging for the identification and distribution of toxins in many of the new genera [37], thus increasing the lack of knowledge on toxic strains. Although cyanobacteria are widely distributed through many environments [7,38], most cyanobacteria and cyanotoxins studies are focused on planktonic species from inland freshwater systems [7,20], with less information in extreme environments (e.g., hot springs, caves, hypersaline lakes, polar deserts), especially regarding their toxicity [7,39].

Taking into consideration cyanotoxins impacts, either in environmental or public health, many scientific studies have focused on the identification of unknown cyanotoxin(s) producing species, usually by a combination of methods as immunoassays (e.g., ELISA enzyme-linked immunosorbent assay), by targeting biosynthesis encoding genes (PCR polymerase chain reaction) and/or by analytical methods, as liquid chromatography [40,41]. The liquid chromatography-mass spectrometry (LC-MS) is a highly sensitive and selective method for cyanotoxin detection and identification [40,41]. It can be used for cyanotoxins confirmation even in very low concentrations, as well as analog identification, even without available standards [40].

The main aim of this study was to screen for cyanotoxin-producing cyanobacteria in strains isolated from several environments (freshwater, terrestrial and thermal) from the Azores islands using a polyphasic approach, based on the detection of MCs, STX, and CYN encoding genes, the identification of their production by LC-MS, and the phylogenetic distribution of toxic strains through 16S rRNA molecular identification.

### **2. Results**

### *2.1. Detection of Cyanotoxin(s)-Producing Cyanobacteria*

From the 157 screened cyanobacteria strains, 13 were identified as cyanotoxin(s) potential producers by PCR amplification of the STX, CYN, and/or MC encoding genes, and four were confirmed as toxin producers by ESI-LC-MS/MS (Figures S1–S6; Table 1; Table S2).


**Table 1.** Detection of genes involved in microcystin, cylindrospermopsin, and saxitoxin production, and ESI-LC-MS/MS analysis on BACA strains. Only strains with positive detection of either the genes or the toxins are shown.

"-": absence of biosynthesis-encoding genes amplification or absence of toxin in the ESI-LC-MS/MS analysis; C: *mcy*C, D: *mcy*D, E: *mcy*E, G: *mcy*G, A: *sxt*A, G: *sxt*G, H: *sxt*H, I: *sxt*I, B: *cyr*B, CC: *cyr*C, amplification of respective biosynthesis-encoding genes.

> STX producing genes were detected in six strains, however, the only strain with amplification of all the searched STX-encoding genes, *Aphanizomenon gracile* BACA0041, was the only one confirmed as saxitoxin producer by ESI-LC-MS/MS (Table 1).

> CYN producing genes were detected in five strains, although only BACA0025 and BACA0031 had amplification of both searched *cyr*B and *cyr*C genes. These two strains were also the only ones with CYN detection by ESI-LC-MS/MS. The remaining three strains with positive results for the detection of CYN encoding genes only amplified the *cyr*B gene (Table 1).

> MC-encoding genes were only detected in *Microcystis aeruginosa* BACA0148 (*mcy*C, *mcy*D, *mcy*E, and *mcy*G) and *Nostoc* sp. BACA0091 (*mcy*E). However, in the ESI-LC-MS/MS analysis, MC-LR ions were identified only in *M. aeruginosa* BACA0148.

### *2.2. Phylogenetic Characterization*

Strains BACA0025 and BACA0031 had similarities with sequences deposited in the GenBank NCBI between 96 and 97%, which shows the distinctiveness of these strains, while strains BACA0041 and BACA0148 had similarities greater than 98% with *Aphanizomenon gracile* and *Microcystis aeruginosa* strains, respectively (Table S3). In the 16S rRNA phylogenetic tree strains BACA0041 and BACA0148 are positioned in clades with strains close to their initial identification, while strains BACA0025 and BACA0031 had high phylogenetic distances from the closest morphological identified strains.

The strain BACA0041, morphologically identified as *Aphanizomenon gracile* (Figure 1F,G), was positioned near other *A. gracile* and *Aphanizomenon flos-aquae* strains, and closer to know STX producing strains as *A. gracile* PMC638.10 or *A. flos-aquae* NIES 81 (Figure 2).

Strains BACA0025 (Figure 1A–C) and BACA0031 (Figure 1E), morphologically identified as *Sphaerospermopsis* sp. are positioned in cluster II between *Nostoc* spp., *Fortiea* spp., *Desikacharya* spp., *Trichormus* spp., *Minunostoc cylindricum*, and *Desmonostoc* spp. strains, however, with significant phylogenetic distance from all these strains (Figure 2).

The strain BACA0148, morphologically identified as *Microcystis aeruginosa* (Figure 1D), was positioned near other *M. aeruginosa* strains, and closer to know MC producing strains as *M. aeruginosa* VN481 (AB666064) or *M*. *bengalensis* VN486 (AB666082) (Figure 3).

**Figure 1.** Cyanotoxin-producing strains included in the phylogenetic analysis: BACA0025 (**A–C**), *Microcystis aeruginosa* BACA0148 (**D**), BACA0031 (**E**), and *Aphanizomenon gracile* BACA0041 (**F**,**G**). Scale bars –10 μm.

**Figure 2.** Maximum likelihood tree cluster I of partial 16S rRNA gene sequences (1070 bp). Bootstrap (greater than 50%) and probabilities values (greater than 0.9) are presented in front of the pertinent nodes. **CYN+**: Cylindrospermopsin producing strains; STX+: Saxitoxin producing strains. \* Collapsed sequences (NCBI accession numbers): AJ630425, FN691925, FN691926, KC297496, FN691923, FN691907, FM242088, FN691908, AB551474, FN691909, KR154316, KR154314, AB551455, KR154298, FN691922, KR154304, AJ630413, AF092504, HQ407325, LR590627, JQ237770, AY763117, AJ582102, LR590629, LR746263, MG921181, MG921182, HQ730086, KT583658, AY701557, FM161349, FM161350, LC474825, LC474824, KT290381, AJ293110, AJ630428, KT290378, KM019920, KT290326, GU434226, AJ630446, MT294032, MT294033, AY196087, KM245026, MN381942, MN381943, DQ234830, AJ630457, AJ630456, KC346266, KC346265, KC346267, JN385287, AF160256, JF768744, EU076457, FJ234895, EF529489, EF529482, EF529488, EU076457, FJ234884, FJ234885, FJ234890, FJ234897, AB608023, JQ237772, LN997860, AJ133177, AJ781131, AJ781149, AJ781145, KM199731, FM177481, AY038033, AM773306, FR774773, MK503791, KY403996, MK503792, MK503793, MH497064, KJ737427, MN243143, AF334695, KY863521, AP018280, KT336439.2, KT336441, MG970541, MG970549, MG970536, MG970538, JN695681, JN695682, FR822753, AF334690, AF334692, JQ083655, KF017617,

KF934181, JX088105, HG970655, AF334701, AF334703, KY508609, KY508610, HG970652, MN626663, MN626664, KM199732, KY508607, KY508612, KY508611, KY508608, HG970653, KF417425, JN385289, JN385292, HQ847564, KM268886, KM268884, KY098849, KM268889, KF934180, KF052614, 125687, KF052599, 114701, GQ287650, KF052617, KF052610, KF052603, KF052605, KU161661, AM711524, HG004586, KT166436, MH497066, MH291266, MF642332, KX787933, KJ566945, KJ566947, KF052607, 125685, KF052616, HE797731, KB235930, MN400069, MN400070, MH598843, MT044191, MT044192, KX638483, KX638487, KX638489, KU161674, KU161676, KU161650, AY577534, MF002129, KM019950, AY785313, HQ847558, HQ847561.

**Figure 3.** Maximum likelihood tree of the partial 16S rRNA gene sequences (1324 bp). Bootstrap (greater than 50%) and probabilities values (greater than 0.9) are presented in front of the pertinent nodes. MC+: Microcystin-producing strains. \* Collapsed sequences (NCBI accession numbers): *Myxosarcina* sp. (AJ344562; AJ344561), *Dermocarpella incrassata* (AJ344559), *Synechocystis pevalekii* (KM350249), *Gloeocapsopsis crepidinum* (KF498710), *Foliisarcina bertiogensis* (KT731151), *Hyella patelloides* (HQ832901), *Chroococcidiopsis* sp. (AJ344557), *Xenococcus* sp. (AF132783), *Stanieria cyanosphaera* (112109), *Chroococcopsis gigantea* (KM019987).

### **3. Discussion**

The ability to produce cyanotoxins depends on the simultaneous existence of several genes involved in their biosynthesis pathways [42]. As seen in previous studies, the *sxt* cluster has suffered several modifications and its gene composition and organization varies between taxa [6,43–45]. Although the amplification of the *sxt*A gene and the absence of toxin production in *Aphanizomenon* strains was seen in other studies [10,25,46,47], in the studied strains only strains with *sxt*A amplification were able to produce STX (Table 1). Cirés et al. [47] states that the *sxt*A gene does not allow the distinction between STXproducing and non-producing *Anabaena*/*Aphanizomenon* strains, nonetheless the presence of *sxt*A in non-producing strains could have been due to gene loss/inactivation within the *sxt* cluster [46]. The presence of *sxt*G or *sxt*H amplifications in several non-STX producing genera (Table 1; Table S2; Figure S3) in this study are interesting new results that require further investigation to fully understand *sxt* gene distribution among cyanobacteria taxa. Nonetheless, to our knowledge, the presence of *sxt*G, and/or *sxt*H genes in *Kamptonema*, *Leptodesmis*, *Anathece minutissima*, or *Leptolyngbya* strains, have not been reported before.

*Aphanizomenon gracile* BACA0041 was confirmed as an STX producer with amplification of the *sxt*A, *sxt*G, *sxt*H, and *sxt*I genes. The ESI-LC-MS/MS spectra of *A*. *gracile* BACA0041 matched the fragmented pattern of STX standard spectra (Figure S4), with identification of the precursor ion 300 m/z and product ions 282 m/z, 265 m/z, 241 m/z, 240.25 m/z, 204 m/z, and 186 m/z [10]. Phylogenetic analysis confirms *Aphanizomenon gracile* BACA0041 identification's, this strain is positioned in a well-supported clade of several STX-producing *A. gracile* strains (Figure 2), as *A. gracile* UAM531 [47] and *A. gracile* PMC 638.10 [10]. In our study, we report another freshwater STX-producing *A. gracile* strain, the first STX producer identified in the Azores islands.

Strains BACA0025 and BACA0031 are quite similar, these were both initially identified as *Sphaerospermopsis* sp. (isolated from similar freshwater lakes from the same island; Table S1), with similar BlastN results (Table S3). Strains BACA0025 and BACA0031 were positioned in cluster II in the phylogenetic tree (Figure 2) close to Nostocaceae genera, however with significant phylogenetic distance to conclude that these strains might belong to a new genus. Both were confirmed as CYN producers with detection of both *cyr*B and *cyr*C genes (Table 1). The ESI-LC-MS/MS spectra of BACA0025 and BACA0031 both matched the fragmented pattern of CYN standard spectra (Figure S5), with identification of the precursor ion 416 m/z and product ions 336 m/z, 318 m/z, 274 m/z, and 194 m/z [48], confirming these two strains as CYN producers.

The *cyr*B gene was also amplified in *Nostoc* sp. BACA0109, and in two thermal *Leptolyngbya* sp. strains BACA0142 and BACA0146, however without CYN identification in the ESI-LC-MS/MS. Amplification of *cyr* genes without CYN production confirmation has been reported previously, as is the case of non-CYN producing *Chrysosporum bergii* and *Chrysosporum ovalisporum* strains with amplification of *cyr*A, *cyr*B, and *cyr*C genes [49]. *Nostoc* and *Leptolyngbya* strains are known to produce cyanotoxins, however, MCs [12,17,50] and not CYN, and as far as we know, the presence of the *cyr*B and *cyr*C genes has not been previously reported in *Nostoc* or *Leptolyngbya* strains.

Microcystins are the most common and more well-studied cyanotoxins, being MC-LR one of the most prevalent and toxic congeners [51]. *Microcystis aeruginosa* was the first cyanobacteria species identified as MCs producer and is the most studied species regarding MCs [11]. Our results show that the Azorean strain *M*. *aeruginosa* BACA0148 is also an MC-LR producer, with the detection of *mcy*C, *mcy*D, *mcy*E, and *mcy*G genes. Characteristic MC-LR fragmentation pattern was observed in *M*. *aeruginosa* BACA0148 (Figure S6), with identification of precursor ion 995 m/z and product ions 977 m/z, 866 m/z, 599 m/z, and 553 m/z [52]. The *mcy*E gene amplification in *Nostoc* sp. BACA0091, despite the absence of MC-LR ions in the ESI-LC-MS/MS, or the absence of the other searched *mcy* genes (*mcy*C, *mcy*D, and *mcy*G), can be explained due to gene(s) recombination or loss [6,14,53]. As stated by Dittmann et al. [6], the *mcy* gene cluster has high repetitive sequences, enabling recombination events that ultimately cause changes in the final product, which can be confirmed by the growing reported number of MCs congeners [54,55].

All strains identified as toxin producers (BACA0025, BACA0031, *A. gracile* BACA0041, and *M*. *aeruginosa* BACA0148) were isolated from lakes in Pico and São Miguel islands (Table S1, S2). The presence of toxic strains in these lakes represents environmental and public health hazards. Contrarily, none of the strains isolated from thermal and terrestrial habitats were identified as cyanotoxin producers, although in some of them, cyanotoxinencoding genes were detected, as in the thermal *Leptolyngbya* strains BACA0112, BACA0123, BACA0142, BACA0144 and BACA0146, and *Coleospermum* sp. BACA0119.

### **4. Conclusions**

Within the BACA collection, we identified and reported another MC-LR-producing *M*. *aeruginosa* strain (BACA0148) and another STX-producing *A*. *gracile* strain (BACA0041). Phylogenetic analysis revealed evidence for new cyanobacteria taxa BACA0025 and BACA0031, confirmed as CYN producers. Further studies are necessary to confirm and describe these new taxa, with morphological characterization and 16S rRNA and ITS analysis.

The identification of new cyanotoxin-producing strains, and unreported toxins, in the Azores, confirms the risk of toxicity and threat to environmental and public health; thus, an appropriate monitoring program should be implemented/updated to search MCs, STX, and CYN. Future efforts should also be made to avoid cyanobacteria blooms and consequently cyanotoxins released in high concentrations.

### **5. Materials and Methods**

### *5.1. BACA Strains and Growth Conditions*

A total of 157 strains, isolated from various environments (Table S1), were retrieved from the Azorean Bank of Algae and Cyanobacteria (BACA) created in the framework of the REBECA project (MAC/1.1a/060). For genetic analysis, 50 mL cultures were prepared without agitation, whereas for toxin extraction, the cultures were scaled up to 1 L, with filtered aeration. All strains were grown in liquid BG-11 media (with or without combined nitrogen) [56], in a climate-controlled room with a 14:10 h light: dark (170 μmol photons m−<sup>2</sup> s<sup>−</sup>1) photoperiod at 25 ◦C [56,57]. Cyanobacterial cells were harvested by centrifugation (4000× *g* for 15 min), after 3–5 weeks, and lyophilized. The lyophilized cyanobacteria biomass were stored at −20 ◦C.

### *5.2. Cyanotoxins Analysis*

### 5.2.1. Toxin extraction

The lyophilized cyanobacteria, 157 samples in total, were weighed (80–100 mg) to a glass vial and extracted with a 5% methanolic solution (2–10 mL). Solutions were then submitted to ultrasounds for 1–5 min, 60 Hz, in an ice bath, and transferred to falcons to be centrifuged (5000× *g*, 5 min, 4◦C). Pellet was then submitted to a second extraction and left in the dark at 4 ◦C overnight. Supernatants were pooled together and lyophilized (extraction solvent completely freeze-dried).

Residues were finally dissolved in 200–500 μL 50% methanol LC-MS grade acidified with 0.1% Formic Acid and filtered with a nylon membrane 0.2 μm before analysis (or centrifuged at 10,000× *g* for 5 min). Samples were protected from light through all the processes and stored at −80 ◦C until analysis.

### 5.2.2. ESI-LC-MS/MS analysis

Saxitoxin (CRM-00-STX, Lot 16-001, 99% purity), microcystin-LR (CRM-00-MC-LR, Lot 19-001, 96% purity), and cylindrospermopsin (CRM-03-CYN, Lot 16-001, 99% purity) standards were all supplied by Cifga (Lugo, Spain). Although MC-RR, MC-YR, and MC-LA standards were not used, their mass was searched in the spectra.

All the standards were injected individually and then as a standard mixture with a concentration interval from 10 ppb to 300 ppb. ESI-LC-MS/MS analysis was performed to confirm the presence or absence of these three cyanotoxins in the selected cyanobacteria strains.

Samples were injected in a Liquid Chromatograph Thermo Finnigan Surveyor HPLC System (Thermo Scientific, Waltham, MA, USA), coupled with a Mass Spectrometry LCQ Fleet™ Ion Trap Mass Spectrometer (Thermo Scientific, Waltham, MA, USA), with a column TSKgel® Amide-80 Phase carbamoyl (250 mm× 2 mm i.d., 5 <sup>μ</sup>m) (TOSOH Bioscience-Lot082B, Tokyo, Japan).

The eluents used were methanol (A) and water (B) both acidified with formic acid at 0.1% (v/v). The gradient program started at 10% B (held for 10 min), increasing to 50% B in 5 min, turning back to initial conditions in 5 min, equilibrating more 10 min with 10% B. The injection volume was 10 μL with a flow of 0.2 mL min−<sup>1</sup> and column kept at 30 ◦C.

Mass spectrometry analysis acquisition parameters were as follows: ESI source, positive ionization using collision-induced dissociation (CID). Table S4 resumes analysis parameters for each searched toxin.

### *5.3. DNA Extraction, PCR Amplification, and Sequencing*

Total genomic DNA was extracted with the PureLinkTM Genomic DNA Mini Kit (Invitrogen, Carlsbad, CA, USA), as previously described by Cordeiro et al. [58]. DNA samples were stored at −20 ◦C.

Genes *mcy*C, *mcy*D, *mcy*E, and *mcy*G were targeted for MCs production potential, *sxt*A, *sxt*G, *sxt*I, and *sxt*H for STX, and *cyr*B and *cyr*C for CYN, using specific primer pairs available in the literature (Table 2). For the 16S rRNA gene amplification primers 27F [59], CYA359F [60], and 1494R [59] were used, whereas for sequencing it was also used primers CYA781F [60] and CYA781R (Table 2).

**Table 2.** Primers used to amplify and/ or sequence cyanotoxins biosynthesis genes and 16S rRNA.


PCRs were carried out in a ProFlex™ 3 × 32-well PCR System (Thermo Fischer, Waltham, MA, USA), according to the literature [9,10,25,58,61,62]. The PCR products were visualized by electrophoresis on 1.5% agarose gels stained with SYBR™ SAFE (0.2 g mL<sup>−</sup>1) and visualized using the transilluminator Molecular Imager® Gel Doc™ XR+ (BioRad, Hercules, CA, USA).

16S rRNA amplification products were purified using the EXTRACTME® DNA cleanup kit (Blirt, Gda ´nsk, Poland), following the manufacturer's protocol. Sequencing was done by Macrogen Ltd. (Madrid, Spain). Nucleotide sequences were deposited in the NCBI Genbank under the accession numbers MT176703, MW776414, MT176711, and MT176750.

### *5.4. Phylogenetic Analysis*

Partial 16S sequences were amplified for the four strains with toxin identification by ESI-LC-MS/MS. All the sequences obtained in this study were compared with sequences deposited in the GenBank NCBI by BlastN tool.

The databases were constructed with the 16S rRNA sequences from this study and identified strains with MCs, STX, and CYN production retrieved from the literature [10,17,25,43,63–67] and GenBank. A final database of 267 OTUs (operational taxonomic units) were aligned for Nostocales and 50 OTUs for Chroococcales, using MAFFT v7.475 [68]. The sequence data matrixes with a final length of 1070 bp (Nostocales) and 1324 bp (Chroococcales) were used to infer phylogenetic distances.

The 16S rRNA gene phylogenetic relations were calculated using maximum likelihood (ML) and Bayesian inference (BI). jModelTest 2.1.10 [69] was used to select the best-fit nucleotide model for our database, on which the general time-reversible evolutionary model of substitution with gamma-distributed evolutionary rates and with an estimated proportion of invariable sites (GTR+G+I) was selected. ML was calculated using the IQ-Tree online version v1.6.12 [70] with 1000 ultrafast bootstrap and BI was calculated using MrBayes v3.2.7a [71], applying two separate runs with four chains each and 50,000,000 Markov chain Monte Carlo generations (sampling every 100 generations with a 0.25 burn-in). The tree was drawn with FigTree 1.4.4 (http://tree.bio.ed.ac.uk/software/figtree, accessed on 8 February 2021) and Inkscape 1.0.1 (https://inkscape.org/pt/, accessed on 8 February 2021). Only the ML tree is presented, with bootstrap percentages (ML) and BI probabilities for branch support, since ML and BI methods resulted in similar trees. Only probabilities above 0.9 and bootstrap percentages above 50 are shown at the branch nodes of the phylogenetic distance trees. *Gloeobacter violaceus* PCC 8105 (AF132791) was used as the out-group.

**Supplementary Materials:** The following are available online at https://www.mdpi.com/article/10 .3390/toxins13040258/s1, Figure S1: Electrophoresis gel photos of *mcy*C (674 bp), *mcy*D (647 bp), *mcy*E (755 bp), and *mcy*G (425 bp) biosynthesis genes amplifications, Figure S2: Electrophoresis gel photos of *cyr*B (650 bp) and *cyr*C (597 bp) biosynthesis genes amplifications, Figure S3: Electrophoresis gel photos of *sxt*A (602 bp), *sxt*G (893 bp), *sxt*H (812 bp), and *sxt*I (910 bp) biosynthesis genes amplifications, Figure S4: Total ion chromatograms and spectra of a STX standard solution (**A**) and sample *Aphanizomenon gracile* BACA0041 (**B**), Figure S5: Total ion chromatograms and spectra of a CYN standard solution (**A**), sample BACA0025 (**B**) and sample BACA0031 (**C**), Figure S6. Total ion chromatograms and spectra of an MC-LR standard solution (**A**) and sample *Microcystis aeruginosa* BACA0148 (**B**), Table S1: Strains information, Table S2: PCR amplifications of MC (*mcy*C, *mcy*D, *mcy*E, *mcy*G), STX (*sxt*A, *sxt*G, *sxt*H, *sxt*I) and CYN (*cyr*B, *cyr*C) biosynthesis-encoding genes and ESI-LC-MS/MS toxicity confirmation, Table S3: Sequence identity (%) of 16S rRNA gene fragment between BACA strains and other cyanobacterial sequences available in GenBank (NCBI), Table S4: ESI-LC-MS/MS analysis parameters for the identification of CYN, MC-LR, and STX.

**Author Contributions:** Conceptualization, R.C., V.V., V.G., and A.F.; methodology, R.C., J.A., and R.L.; formal analysis, R.C.; investigation, R.C.; resources, V.V., V.G., and A.F.; writing—original draft preparation, R.C.; writing—review and editing, R.C., J.A., R.L., V.V., V.G., and A.F.; visu-alization, R.C.; supervision, V.V., V.G., and A.F.; project administration, V.G.; funding acquisition, V.V., V.G., and A.F. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by FEDER funds through the Interreg-MAC 2014-2020 program under the projects REBECA—Red de Excelencia en Biotecnología Azul (algas) de la Región Macaronesia (MAC/1.1a/060) and REBECA-CCT—Red de Excelencia en Biotecnología Azul de la Región Macaronésica. Consolidación, Certificación y Transferencia (MAC2/1.1b/269). Rita Cordeiro was supported by a Ph.D. grant (M3.1.a/F/017/2015) from the Fundo Regional da Ciência e Tecnologia (FRCT). CIIMAR acknowledges the project H2020 RISE project EMERTOX—Emergent Marine Toxins in the North Atlantic and the Mediterranean: New Approaches to Assess their Occurrence and Future Scenarios in the Framework of Global Environmental Changes (grant agreement no. 778069), and FCT Projects UIDB/04423/2020 and UIDP/04423/2020. This work was also funded by FEDER funds through the Operational Programme for Competitiveness Factors - COMPETE and by National Funds through FCT - Foundation for Science and Technology under the UID/BIA/50027/2020 and POCI-01-0145-FEDER-006821. The APC was funded by REBECA-CCT (MAC2/1.1b/269).

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** Data is contained within the article or Supplementary Material. The data presented in this study are available in https://www.mdpi.com/article/10.3390/toxins13040258/s1.

**Acknowledgments:** We would like to thank Carmo Barreto from the University of the Azores, for the use of Biochemistry Lab facilities, without it, this work would not be possible.

**Conflicts of Interest:** The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.

### **References**


### *Article* **Potentially Toxic Planktic and Benthic Cyanobacteria in Slovenian Freshwater Bodies: Detection by Quantitative PCR**

**Maša Zupanˇciˇc 1,2,\*, Polona Kogovšek 3, Tadeja Šter 4, Špela Remec Rekar 4, Leonardo Cerasino 5, Špela Baebler 3, Aleksandra Krivograd Klemenˇciˇc <sup>4</sup> and Tina Eleršek <sup>1</sup>**


**Abstract:** Due to increased frequency of cyanobacterial blooms and emerging evidence of cyanotoxicity in biofilm, reliable methods for early cyanotoxin threat detection are of major importance for protection of human, animal and environmental health. To complement the current methods of risk assessment, this study aimed to evaluate selected qPCR assays for detection of potentially toxic cyanobacteria in environmental samples. In the course of one year, 25 plankton and 23 biofilm samples were collected from 15 water bodies in Slovenia. Three different analyses were performed and compared to each other; qPCR targeting *mcyE*, *cyrJ* and *sxtA* genes involved in cyanotoxin production, LC-MS/MS quantifying microcystin, cylindrospermopsin and saxitoxin concentration, and microscopic analyses identifying potentially toxic cyanobacterial taxa. qPCR analyses detected potentially toxic *Microcystis* in 10 lake plankton samples, and potentially toxic *Planktothrix* cells in 12 lake plankton and one lake biofilm sample. A positive correlation was observed between numbers of *mcyE* gene copies and microcystin concentrations. Potential cylindrospermopsin- and saxitoxin-producers were detected in three and seven lake biofilm samples, respectively. The study demonstrated a potential for cyanotoxin production that was left undetected by traditional methods in both plankton and biofilm samples. Thus, the qPCR method could be useful in regular monitoring of water bodies to improve risk assessment and enable timely measures.

**Keywords:** cyanotoxin detection; harmful cyanobacterial blooms; next-generation biomonitoring; real-time PCR; qPCR; LC-MS/MS; microcystin; cylindrospermopsin; saxitoxin

**Key Contribution:** Currently used biomonitoring methods are not sufficient for detection of cyano-toxic potential. In addition to planktic cyanobacteria, benthic species in biofilm can be a potential source of cyanotoxins and therefore both groups should be included in biomonitoring for risk assessment.

### **1. Introduction**

Cyanobacterial blooms and a subsequent release of cyanotoxins into the environment are becoming more frequent due to eutrophication, global warming and other anthropogenic pressures. They can have negative effects on all ecosystem services as well as human and animal health and can cause economical damage by affecting tourism, recreation, industry, agriculture and drinking water supply. On the European Union level, there is no legislation prescribing regular monitoring of cyanotoxin concentration in surface waters. The most frequently used guideline is the one for drinking water from the World Health Organisation, setting the upper limit of 1 μg/L of microcystin-LR equivalents [1].

**Citation:** Zupanˇciˇc, M.; Kogovšek, P.; Šter, T.; Remec Rekar, Š.; Cerasino, L.; Baebler, Š.; Krivograd Klemenˇciˇc, A.; Eleršek, T. Potentially Toxic Planktic and Benthic Cyanobacteria in Slovenian Freshwater Bodies: Detection by Quantitative PCR. *Toxins* **2021**, *13*, 133. https://doi.org/ 10.3390/toxins13020133

Received: 18 January 2021 Accepted: 9 February 2021 Published: 11 February 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

Early detection of cyanotoxin threat could help water resources managers take timely and appropriate measures. Current approaches for identification and quantification of cyanobacterial cells and cyanotoxins—microscopic count and analytical methods, such as high-performance liquid chromatography (HPLC), liquid chromatography–mass spectrometry (LC-MS), enzyme-linked immunosorbent assay (ELISA) or protein phosphatase 1A (PP1A) analyses—each have their advantages, but they can often be time-consuming, costly or technically demanding. Emerging molecular methods, such as quantitative PCR (qPCR), could enable fast, highly sensitive and cost-effective detection of potentially toxic cyanobacteria [2]. This approach is based on the extraction of community DNA from environmental samples and can therefore provide a full picture of the cyanobacterial diversity. However, for its use in regular monitoring programs, these methods have to be thoroughly tested and optimised in order to achieve comparability with current methods and thus applicability in monitoring schemes.

Analytical methods for detection of cyanotoxins can determine their concentration only at a certain point in time, whereas cyanotoxin content can vary significantly throughout the day (depending on hydrological conditions and the presence of bacterial decomposers). Besides, due to a high number of variants of cyanotoxins (e.g., over 248 variants of microcystins [3]) and a lack of standards for each of these variants, not all of them can be measured. These methods are also too expensive to monitor the concentration daily. On the other hand, detection and quantification of genes involved in cyanotoxin synthesis could enable cost-effective monitoring of the potential for cyanotoxin production on a daily basis at various locations. Moreover, all of the cyanotoxins with known genetic basis can be analysed. This can give us comprehensive information on the toxigenic potential of cyanobacterial communities in the environment.

The qPCR method has been applied in various studies, which have been summed up by Pacheco et al. [2]. Majority of the studies target genes involved in microcystin (MC) synthesis (e.g., [4], followed by cylindrospermopsin (CYN) (e.g., [5] and saxitoxin (SXT) synthesis (e.g., [6]). There have been attempts to optimise the method for its use directly in the field ([7] or to target various genes at once in multiplex reactions (e.g., [8]). However, there is still no consensus on the applicability of qPCR in regular monitoring since gene copy numbers reveal only the potential for cyanotoxin production, which is not always in correlation with actual cyanotoxin concentrations in the environment. Precisely this contrast between the methods indicates the advantage of qPCR over analytical methods focused on cyanotoxin measurement, as the former one could predict also future risk rather than assessing only the current situation.

The majority of the cyanobacterial qPCR studies are directed at microcystins, while detection of cylindrospermopsins and especially saxitoxins is still relatively rare [2]. Moreover, most of the studies focus on plankton samples, while cyanobacteria in biofilm are still underrepresented in molecular studies despite increasing evidence of their toxicity with potential acute effects on animals [9–13]. Furthermore, few of these studies include comparison of qPCR method and microscopy [2]. As microscopy is the preferred method of biomonitoring in many countries, it is important to investigate its effectiveness in detecting potential risk of cyanotoxin production. Additionally, all such studies are geographically limited with little or no focus on the central European region [2]. Taking into account the high genetic variability in naturally occurring cyanobacterial strains throughout the world (e.g., [14]), the assays should be tested in different regions and different water bodies to assure their wide applicability.

Therefore, the aim of this study was to expand the evaluation of qPCR assays for detection of cyanotoxin threat to understudied benthic cyanobacteria in biofilm samples, with the emphasis on comparison of results with microscopy as well as LC-MS/MS. We focused on water bodies in different regions of Slovenia (central Europe) and on three groups of cyanotoxins: microcystins, cylindrospermopsins and saxitoxins. We employed five previously published qPCR assays for detection of the microcystin- [15–17], cylindrospermopsin- [18] and saxitoxin-producing cyanobacteria [6]. Although the cyanotoxin potential is not necessarily linked to cyanotoxin concentration, we evaluated the correlation between the number of gene copies, microscopically determined cell number of potentially toxic species and cyanotoxin concentration. This is the first study in Slovenia aiming to detect cyanobacterial toxic potential with qPCR, and one of the few studies employing this method in environmental biofilm samples. Only thorough understanding of strengths and weaknesses of the qPCR method can enable its implementation into existing environmental monitoring strategies.

### **2. Results**

### *2.1. Evaluation of the qPCR Assays*

For our study, we have chosen previously published assays mcyE-Ana, mcyE-Mic, mcyE-Pla, cyrJ and sxtA, targeting microcystin-producers from genera *Dolichospermum* (ex *Anabaena*), *Microcystis* and *Planktothrix*, cylindrospermopsin-producers and saxitoxinproducers, respectively ([6,15–18]; Table 1). First, we evaluated the selected qPCR assays in terms of their specificity, sensitivity and robustness.



2.1.1. Selection and Specificity of the qPCR Assays

Based on a literature review (Supplementary file S1), we selected nine published assays for specificity evaluation. In addition to the assays shown in Table 1, two assays targeting all microcystin-producers (McyE-F2b/R4 [15,19] and DQmcy [20]), and assay anaC-gen targeting anatoxin-producers [21,22] were evaluated. Assay anaC-gen was excluded due to too high specificity indicated in the original paper [21] and too long amplicon, originally designed for end-point PCR application. The other eight assays were evaluated in vitro on test environmental samples. Assays mcyE-F2b/R4 and DQmcy were excluded based on suboptimal performance with Slovenian environmental samples (dissociation curves indicating non-target amplification (multiple peaks) and certain results inconsistent with

microscopic observations, which could be due to regional differences in cyanobacterial genotypes; data not shown)).

The remaining six assays were further characterised in silico and in vitro. Specificity evaluation done in the original papers demonstrated appropriate specificity of all assays, while BLAST analysis in this study revealed that four assays (namely mcyE-Ana, mcyE-Pla, cyrJ and sxtA) are specific for desired target organisms and genes. Assays mcyE-Mic and 16S-cyano showed non-target alignment; the former with *Pseudanabaena* sp. CCM-UFV065, and the latter with plant chloroplasts as well as some heterotrophic bacteria, such as *Chryseobacterium* or *Actinobacterium*, which can be found in freshwater habitats and could thus be amplified in environmental samples. In original studies, specificity of assays mcyE-Mic [16] and 16S-cyano [6] was in vitro tested with selected non-target genera, namely *Planktothrix*, *Dolichospermum* and *Nostoc*, and with several target cyanobacterial cultures, respectively. None of the assays showed any cross-reactivity, however, unspecific reaction predicted with in silico analysis was not evaluated.

In cyanobacterial cultures, specific amplification occurred only in strains producing target cyanotoxins and not in other strains (Table 2). In assay 16S-cyano, strong amplification was observed in selected plant samples, confirming cross-reactivity with plant 16S rRNA genes (Supplementary file S2). This could lead to false positive results and overestimation of cyanobacterial abundance in environmental samples; thus, this assay was used only for evaluation of DNA extraction and control of inhibition in qPCR reactions.


**Table 2.** Specificity of selected qPCR assays, with positive results shaded in grey. Average quantification cycle (Cq) values between three technical replicates of DNA in 10−<sup>2</sup> dilution and reference melting temperatures (Tm, in ◦C), calculated as an average of all DNA dilutions within quantification range are shown. Raw data is available in Supplementary file S4.

In some of the environmental samples, melting temperatures of the amplified products (Tm) obtained via dissociation curve analysis indicated the presence of non-target amplicons. This was mostly observed with assays mcyE-Ana and mcyE-Mic, which showed up to 12.3 ◦C and up to 10.0 ◦C higher Tm then the reference Tm (Table 2, Supplementary file S4), respectively. Additionally, primer dimers were detected in some samples with assays mcyE-Ana, mcyE-Mic and mcyE-Pla (Tm < 70 ◦C). Therefore, results for all cyanotoxinspecific assays were considered positive if their Tm values were within the expected range (±0.5 ◦C). For the assay 16S-cyano, however, a wider range of obtained Tm values (80.7–82.7 ◦C; Supplementary files S2–S5) were considered positive, as they were consistent between technical replicates and the Cq values were below 30. Reference Tm values from pure cultures corresponded closely to Tm of synthetic DNA fragments for all assays (± ≤0.5 ◦C, data not shown). Gel electrophoresis of the samples with multiple Tm peaks (18 samples for mcyE-Ana, 3 samples for mcyE-Pla and 2 samples for cyrJ) produced multiple bands of different lengths, in contrast to environmental samples with one distinctive Tm peak that produced a single band (data not shown), confirming non-specific amplification of various DNA fragments in the former group. Therefore, such samples were considered negative.

2.1.2. Sensitivity of the qPCR Assays.

Dilution series of the cyanobacterial culture DNA was prepared to evaluate the sensitivity of the assays (Table 3; calibration curves in Supplementary file S6). All six qPCR assays showed high sensitivity, ranging from 10 to 30 cells/mL for all assays, except sxtA, which showed the highest sensitivity, detecting less than 1 cell/mL (Table 3). Amplification efficiency determined from the dilution series of all assays was between 63% and 98%.

**Table 3.** Limit of detection (LOD), limit of quantification (LOQ), amplification efficiency and correlation coefficient for selected qPCR assays based on calibration curves of reference cyanobacterial strains DNA. LOQ and LOD are expressed in cells/μL DNA and in cells/mL sample, which depends on the individual sample volume. Individual calibration curves are shown in Supplementary file S6.


### 2.1.3. Robustness of the qPCR Assays

Performance of the assays was evaluated also on typical environmental samples with known cyanobacterial taxa composition (as determined by microscopy). Up to 14 samples of plankton community DNA (2 from lakes, 6 from urban ponds, 1 from urban stream and 5 cyanobacterial bloom samples) were tested with specific assays. The presence of the target organisms was mostly confirmed in samples where it was expected (Supplementary file S7), thus proving the suitability of the method for target gene detection in environmental samples sampled in our region.

For intra-assay variability, the absolute difference between Cq values from three technical replicates in positive samples was determined. Relatively high intra-assay variability was observed in samples with target gene concentration close to LOD of the assays, which is probably due to a stochastic effect, while lower variability was observed in the rest of the samples (Supplementary file S5). For the assay mcyE-Ana, we could not assess intra-assay variability as we did not get any positive qPCR results from the environmental samples. When using cyanobacterial monocultures as a template, the variability was in general lower than with environmental samples (Supplementary files S3 and S5). This is expected, as the presence of inhibitory compounds in environmental samples can interfere with qPCR amplification [23], which can result in higher intra-assay variability. Inter-assay variability, evaluated for assays mcyE-Pla, cyrJ and sxtA in two separate runs, differed between assays but showed reasonably good reproducibility, taking into account degradation of DNA due to freeze/thaw cycle and different inhibitory substances in environmental samples (data not shown).

Additionally, possible inhibition of qPCR reactions was checked by testing two subsequent dilutions of 10 randomly selected samples. Only one sample showed a potential for inhibition of a qPCR reaction (BL1.5, Supplementary file S8), therefore analysis of this sample was repeated with more diluted DNA. Robustness of the assays was demonstrated by testing DNA extracted from different matrices (cyanobacterial monocultures, synthetic DNA fragments, frozen or lyophilised bloom samples, environmental samples of plankton or biofilm), of variable purity and of variable DNA concentration (data not shown), where adequate performance was observed in all cases.

### *2.2. Presence of Target Genes in Environmental Samples*

For the analysis of the environmental samples, negative and positive controls were included in every qPCR run, where the former showed no amplification and the latter showed specific amplification in all cases. Additionally, negative controls were included in every DNA extraction and in every sampling, and their NanoDrop measurements showed no presence of DNA. This confirmed appropriate assay performance and absence of contamination during field sampling, DNA extraction and preparation of qPCR reaction mixtures. Successful DNA extraction and qPCR amplification were additionally confirmed by positive results of the assay 16S-cyano amplifying 16S rRNA genes in all samples with a Cq range between 14 and 24 (Supplementary file S5).

Detailed results of qPCR, LC-MS/MS and microscopic analyses are depicted in Supplementary file S8, raw data is available in Supplementary file S5. Potentially toxic *Microcystis* cells (assay mcyE-Mic) were detected in 10 lake plankton samples (all samples from Lake Vogrscek, Slivnica and Pernica). Potentially toxic *Planktothrix* cells (assay mcyE-Pla) were detected in 12 lake plankton samples (all samples from Lake Bled, in low amounts also in Lake Bohinj) and in low amounts in one biofilm sample (Lake Bled). Potentially toxic *Dolichospermum* species (assay mcyE-Ana) were not detected in any plankton nor biofilm sample. Potential cylindrospermopsin producers (assay cyrJ) were detected in three lake biofilm samples (Lake Bled, Sava River), while potential saxitoxin producers (assay sxtA) were detected in seven lake biofilm samples (Lake Bled, Koseze Pond) (Table 4, Supplementary file S8). In some samples, only 1/3 or 2/3 technical replicates were positive; all of these samples were close to LOD of the assay. This means that the target genes were present in low quantities and thus not amplified in every subsample (stochastic effect).

**Table 4.** Abundance of target gene copies in environmental samples. Plankton [gc/mL]: \* 1–102, \*\* 102–104, \*\*\* 104–106; biofilm [gc/g dry weight]: \* 103–105, \*\* 105–107. Due to low amplification efficiency, the gc values are not reliable and should be used as qualitative observation. Sample description and quantification data is available in Supplementary file S8. Sample BL1.4 has only been analysed for cyanotoxin content and not by qPCR and is thus not represented in this table. gc—gene copies, empty—below LOD.



### *2.3. Temporal Variability of Microcystin Abundance*

Cylindrospermopsins or saxitoxins were not detected with LC-MS/MS in any of the samples. Microcystins were detected in 16 out of 23 plankton samples (out of which one was uncertain as it was too close to the background noise) and 4 out of 22 biofilm samples (out of which 3 were uncertain) with 5 different variants observed (MC-RR, MC-RRdm, MC-HtyRdm, MC-LRdm, MC-LR). The highest concentrations of microcystins were measured in plankton samples from Lake Bled during winter months (up to 660 ng/L in February, BL1.2, Supplementary file S8), which complies with the highest cell concentration of the microcystin-producing species *Planktothrix rubescens* (See Section 2.5. Correlation between Parametres). Moreover, all three microcystin variants found in these samples are typically produced by *Planktothrix rubescens*. Figure 1 shows the microcystin diversity in plankton samples of Lake Bled, where a temporal trend can be observed. In all other samples, microcystins were either not detected or their concentrations were low; up to 6.9 ng/L in plankton samples (SL2, Supplementary file S8) and up to 11.9 ng/g dry weight in biofilm samples (BL2.6, Supplementary file S8) and thus their diversity is not represented in the Figure 1.

### *2.4. Microscopic Analyses*

With microscopic analyses, we found 17 potentially toxic taxa in plankton samples (Aphanizomenon sp., Aphanizomenon flos-aquae, Aphanizomenon issatschenkoi, Cylindrospermopsis raciborskii, Dolichospermum crassum, Dolichospermum flos-aquae, Dolichospermum lemmermannii, Dolichospermum planctonicum, Microcystis aeruginosa, Microcystis flos-aquae, Phormidium sp., Phormidium amoenum, Planktothrix agardhii, Planktothrix rubescens, Pseudanabaena sp., Pseudanabaena catenata, Pseudanabaena limnetica) and 4 in biofilm samples (Oscillatoria sp., Phormidium sp., Phormidium autumnale, Pseudanabaena catenata) (Figure 2, Supplementary file S9). The highest diversity of potentially toxic planktic cyanobacteria was found in Lake Pernica (13 taxa), while in Lake Bohinj none was detected. In the majority of the lakes, there was a higher diversity of potentially toxic taxa observed in summer months (June, July, August) than in the rest of the year. Out

of the 23 biofilm samples, in the majority of them there was one potentially toxic taxon detected under microscope, while six of them contained none and two of them contained two potentially toxic taxa (Koseze Pond, Lake Bled). The most common potentially toxic genus was Phormidium, which was detected in 17 samples.

55GP +W\5GP /5GP

**Figure 1.** Microcystin temporal variability and diversity in plankton samples from Lake Bled. The three congeners (RRdm, HtyRdm and LRdm) are all demethylated variants, typical of *Planktothrix rubescens*. The temporal changes of proportions can be explained by the succeeding of different chemotypes of the same species. Samples Feb–Dec correspond to samples BL1.2–BL1.12 (Supplementary file S8).

### *2.5. Correlation between Parametres*

To further evaluate qPCR as a method to detect cyanotoxin production potential, correlations (presented as Pearson correlation coefficient) between qPCR, LC-MS/MS and microscopy results were determined. For plankton samples (N = 25), the numbers of *mcyE* gene copies (sum of values produced by assays mcyE-Ana, mcyE-Mic and mcyE-Pla) were positively correlated with microcystin concentrations measured by LC-MS/MS (r = 0.8375, Figure 3A), while there was no correlation with cell number or biovolume of all potential microcystin-producing taxa. There was also no correlation found between *Microcystis*or *Planktothrix*-specific *mcyE* gene copies and cell numbers or biovolumes of *Microcystis* or *Planktothrix* cells, respectively. However, when results from Lake Bled were analysed separately, there was a positive correlation between *Planktothrix*-specific *mcyE* gene copies and both cell numbers and biovolumes of *Planktothrix* cells (r = 0.8831 in both cases, the former one is presented on Figure 3B). More detailed graphical representation of results from Lake Bled produced with different methods (Figure 4) shows similar temporal trend observed with qPCR, microscopy and LC-MS/MS. Elevated abundances of *Planktothrix* cells and microcystin concentrations in winter months correspond to a scarlet-coloured blooms of *Planktothrix rubescens* (Figure 2), which were observed on Lake Bled in February 2019 and January 2020.

**Figure 2.** Potentially toxic cyanobacterial taxa found in environmental samples. (**A**)—Aphanizomenon flos-aquae, (**B**)—Aphanizomenon issatschenkoi, (**C**)—Dolichospermum lemmermanii, (**D**)—C.r. Cylindrospermpsis raciborskii, D.c. Dolichospermum crassum, P.a. Planktothrix agardhii, (**E**)—D.f. Dolichorpermum flos-aquae, D.p. Dolichospermum planctonicum, M.a. Microcystis aeruginosa, (**F**)—M.a. Microcystis aeruginosa, M.f. Microcystis flos-aqaue, P.a. Planktothrix agardhii, (**G**)—Phormidium amoenum, (**H**)—Planktothrix rubescens, (**I**)—P.a. Planktothrix agardhii, P.c. Pseudoanabaena catenata. The photos were taken under light microscope with 160× (**A**), 400× (**C**,**H**), 640× (**B**,**D**,**E**,**F**) or 1600× magnification (**G**,**I**).

**Figure 3.** Scatterplots showing the correlations between different parameters for plankton samples. Solid lines represent linear regression curves, dotted lines represent 95% confidence band, and Pearson correlation coefficient (r) is given in the bottom right-hand corners. (**A**) correlation between *mcyE* copies concentration (the sum of concentrations obtained by assays mcyE-Ana, mcyE-Mic and mcyE-Pla) and MC concentration. (**B**) correlation between *Planktothrix*-specific *mcyE* gene copies concentration and cell concentration of all *Planktothrix* species (only samples from Lake Bled, N = 11).

**Figure 4.** *Planktothrix* cell abundance in Lake Bled in 2019 determined by qPCR (assay mcyE-Pla) and microscopic analyses. Due to low amplification efficiency and possible variability of gene copy numbers per cell, the calculated cell abundance values are not reliable and should be used as qualitative observation. Total MC concentration measured by LC-MS/MS is included for comparison. In microscopy, 10% variation is expected, which is the average standard deviation between technical replicates with Bürker Türk counting chamber in our laboratory.

For plankton samples, correlations could not be determined for assays mcyE-Ana, cyrJ and sxtA as we did not detect target genes in any of the plankton samples. For biofilm samples (N = 23), correlations between target gene copies and relative species abundance were determined and there was no correlation found for any of the assays. Species abundance was evaluated semi-quantitatively (values 1–5), which might have impacted the results.

### **3. Discussion**

The study aimed to evaluate the suitability of a qPCR method for early detection of potentially toxic cyanobacteria in surface water bodies in the central European region. Systematic search for publications describing molecular assays (Supplementary file S1) revealed several qPCR assays that were applied for detection of cyanobacteria-specific target genes. We performed a selection of the amplicons, where we took into consideration their specificity, sensitivity and suitability for qPCR reaction conditions. Furthermore, we tested the performance of selected assays (Table 1) in vitro. Even though all assays showed good performance in pure cultures (Table 2) and synthetic DNA fragments, some of them showed unspecific amplification in environmental samples. This is a consequence of heterogenous samples originating from water bodies and presents a high risk in application of SYBR Green chemistry detection in environmental samples, especially when the genome of the target organism is unknown. Nevertheless, we were able to filter the true positive samples from the unspecific samples with reference Tm values. Based on these results, all five selected cyanotoxin-specific assays are suitable for detection of cyanotoxin potential in water bodies. However, as we did not detect cylindrospermopsins or saxitoxins in any of the environmental samples, further research is needed to confirm the suitability of the assays for potential producers of these cyanotoxins. On the other hand, assay 16S-cyano

was shown to be inappropriate for detection of cyanobacteria, since it also amplifies plant chloroplast DNA (Supplementary file S2). Thus, we used it as a DNA extraction and qPCR inhibition control.

The qPCR results revealed some new information that was unknown up to date and could not be obtained by microscopy alone. The most novel finding is the potential for cylindrospermopsin- and saxitoxin-production in biofilm in certain water bodies (Table 4), where it has never been reported before. In Slovenia, cylindrospermopsins have been detected once in low amounts in a planktic sample (data not published), while saxitoxins have never been detected. Our study indicates that despite these cyanotoxins not being detected in the moment of sampling (Supplementary file S8), a potential for their production exists, which might be important information for the future monitoring schemes and research studies.

Another important finding is the discovery of cyanotoxic potential in biofilm (Table 4). Even though the first observation of potentially toxic cyanobacteria in biofilm samples was reported in 1997 [24] and cyanotoxins from benthic cyanobacteria are believed to have caused animal death on a few occasions [9–13,24], such studies are still scarce. These findings suggest anatoxin-a and microcytins as the prevalent cyanotoxins in cyanobacterial mats. However, our results indicate such microbial mats might also possess cylindrospermopsin- and saxitoxin-producing potential (found in over a third of the biofilm samples, Table 4), which could not be detected by either microscopy or LC-MS/MS. Although there have been some prior publications about cylindrospermopsin- [25–27] and saxitoxin-production [28–30] by benthic cyanobacteria, the issue is still poorly investigated. This information, together with some novel findings regarding microcystins in Slovenia (presence of potentially toxic *Microcystis* in Lake Slivnica and Vogrscek throughout the whole year, which has not been reported by regular monitoring before) might be a valuable guideline for future water management.

In addition to applying the qPCR method to environmental samples, our aim was also to compare its performance to traditionally used methods. While many studies have compared qPCR results with cyanotoxins measurements (e.g., [5,6,31]), comparisons with microscopic counts are harder to find, so one of our goals was to evaluate qPCR in comparison with microscopy-based biomonitoring methods. There was no correlation found between gene copy numbers and cell numbers or biovolumes for microcystin-producing cyanobacterial species, neither when analysed separately by genus nor as a whole group. Regarding *Microcystis* genus, there were four samples where *Microcystis*-specific *mcyE* genes were detected, while *Microcystis* cells were not observed under microscope (Supplementary file S8). This could mean that the numbers of *Microcystis* cells in these lakes were below LOD of microscopy, but could be detected by qPCR, which is expected due to much higher sensitivity of qPCR. Alternatively, discrepancy between results could be caused by crosscontamination of field equipment while transferring it between lakes or by low specificity of the assay mcyE-Mic (detecting *mcyE* genes in other genera or other non-target products). The possibility of non-target detection was confirmed also by BLAST analysis, revealing *Pseudanabaena* sp. as one of the assay's potential targets. However, the majority of the results cannot be explained by this, as *Pseudanabaena* was microscopically observed only in two of these samples. This could be further investigated by DNA sequencing of obtained qPCR products. On the other hand, there were also two samples where *Microcystis* cells were microscopically identified, while *Microcystis*-specific *mcyE* genes were not detected (Supplementary file S8). This might be due to the fact that toxic and non-toxic *Microcystis* cells cannot be distinguished morphologically [32].

For the *Planktothrix* species, elevated cell and microcystin concentrations in Lake Bled in the beginning of the year (Figure 4) correspond to a moderate *Planktothrix rubescens* bloom in 2019. Elevated concentrations at the end of the year contributed to a massive bloom formation that occurred in the end of January 2020 (field observations), which was influenced also by other nutritional factors taking place that month, so it cannot be explained only by our measurement in 2019. Despite the lack of correlation between microscopic and qPCR-based abundance of *Planktothrix* cells in the whole dataset, there was a positive correlation when the analysis was performed only with samples from Lake Bled (Figure 3). Those 11 samples represent a majority of qPCR-positive samples for this assay (mcyE-Pla; Table 4), which indicates that the lack of correlation in other samples is primarily caused by samples with negative qPCR and positive microscopy results. In most of such samples, the dominant *Planktothrix* species was *P. agardhii* (Lake Pernica, Supplementary file S9), which might suggest that the assay does not amplify target genes in the whole genus equally. Alternatively, the discrepancy might be caused by the presence of non-toxic *P. agardhii* strains and the inability of microscopic analyses to differentiate between them, which has been shown in previous studies [33]. Regarding *Dolichospermum* genus, the complete lack of genus-specific *mcyE* genes in all analysed samples (assay mcyE-Ana) despite some microscopic observations (Lake Pernica, Lake Bled; Supplementary file S9) might indicate that the assay does not amplify target genes in all *Dolichospermum* strains, or that the present taxa were in fact not possessing *mcyE* genes.

Furthermore, qPCR was also compared to LC-MS/MS results. A positive correlation was found between *mcyE* gene copy numbers (sum of all analysed genera) and microcystin concentrations (Figure 3), which corroborate numerous prior studies (e.g., [16,31,34]). However, there were also some discrepancies between qPCR and LC-MS/MS results. In some samples, target genes were detected (mostly below LOQ), while cyanotoxins were not (Supplementary file S8). Similar inconsistencies have been observed in other studies as well (e.g., [6,35].) Authors' potential explanations include low concentration of cyanotoxins (below LOD of analytical method), degradation of cyanotoxins in the samples, lack of gene expression or mutations leading to non-toxicity. It has to be noted that these results are not always expected to match, as analytical methods (such as LC-MS/MS) measure the actual cyanotoxin concentration in a particular moment of sampling, while qPCR detects only the potential for cyanotoxin production. It has been indicated that despite the presence of *mcy* genes, their expression can vary in time significantly [36]. The toxin production depends on physical parameters (e.g., temperature), growth phase [37] and nutrient content [38]. Therefore, quantification of gene copies alone cannot always predict toxin concentration.

Regarding the methodology itself, our study confirmed that qPCR has significantly higher sensitivity (LOD = 1.5–205.9 cells/mL, Table 3) than microscopic cell count (Bürker Türk counting chamber, LOD = 10.000 cells/mL) of Slovenian samples, which are not preconcentrated. Detection of less than 1 cell/mL (assay sxtA) could be explained by multiple gene copies of the target gene per cell [6] and possible free DNA in the sample. Specificity of all assays was tested and proved appropriate in the original publications. On top of that, our experiments showed that the assays are highly specific in cyanobacterial monocultures (Table 2), while there was some non-specific amplification observed in environmental samples—especially with assays mcyE-Ana and mcyE-Mic. Some of these amplicons are probably primer dimers (Tm < 70 ◦C), which were observed also in the original study [16]. In order to eliminate false positives, the authors measured fluorescence at a temperature higher than Tm of primer dimers (77 ◦C). On the other hand, we also observed non-target amplicons with higher Tm (mostly > 80 ◦C), which indicates amplification of non-target regions. These samples did not show distinctive Tm peaks, but rather multiple smaller peaks, and we confirmed the presence of various non-specific amplicons also by gel electrophoresis. Cq values of such samples were therefore a product of amplification of various templates and could not be considered positive. These false positive signals were excluded from further analyses. The results suggest that the SYBR Green chemistry might not be the most suitable for environmental samples. Specificity and quantification could be improved by using TaqMan chemistry (Roche Molecular Systems Inc., USA) with fluorescent probes or by complementing qPCR results with sequencing of the products.

In some of the assays, amplification efficiency was relatively low (Table 3). While in the original studies amplification efficiency exceeded 90% [6,16,18] for all evaluated assays, in our study that was the case only for assay sxtA. This difference might be caused by sequence variability of uncharacterised cyanobacterial cultures, which is even more significant between different geographical regions of sampling. Possible mismatches in the target regions due to high sequence variability between cyanobacterial strains can lead to low amplification efficiency. Moreover, the effect of the sample impurities and secondary structure of genomic DNA has to be considered as well. Therefore, the LOD and LOQ values, as well as calculated gene copy numbers and cell numbers, might not be fully reliable and should be taken only as a qualitative observation. The reliability of the assay mcyE-Pla might be additionally decreased by a narrow linear dynamic range (Supplementary file S6), which should be addressed in future studies. Moreover, the efficiency of DNA extraction from environmental samples should be evaluated for a more accurate quantification of cells and comparability of results.

For implementation in existing monitoring programs, it is important to quantify cells of the target organisms, not merely gene copies. This might be uncertain in the phylum of cyanobacteria due to unknown number of target gene copies per cell. Genetic cluster *mcy* is thought to appear in only one copy per genome [39,40], while gene *sxtA* appears on average in 3.58 copies per cell in strain *A. circinalis* AWQC131C [6]. Besides, cyanobacteria can contain up to 10 or even more copies of genome per cell [41]; ploidy level differs between species and strains, while it depends also on the growth phase and environmental parameters [42,43]. Therefore, it has to be noted that the calculated cell numbers represent an average for all the genotypes containing target genes from environmental samples, estimated based on assumption that they contain the same number of gene copies as the reference strains. This issue could be avoided if the risk assessment guidelines were adapted for operation with number of gene copies instead of cells.

To enable a thorough cyanotoxin risk assessment in regular monitoring, an assay for detection of anatoxin-a production potential should be designed and optimised. In this study, anatoxin-a was excluded, because literature search did not reveal any appropriate qPCR assays for detection of all anatoxin-producing cyanobacteria. What is more, it would be beneficial to optimise a single assay for detection of *mcyE* genes in all potential microcystin-producers instead of genus-specific assays in order to simplify the test and lower the costs.

### **4. Conclusions**

This is the first study in Slovenia aiming to detect cyanotoxic potential with qPCR, as well as one of the few studies employing this method in environmental biofilm samples. We conclude that the method is appropriate for detection of potentially toxic cyanobacteria in water bodies for the purpose of rapid screening and early warning, which could improve risk assessment and protection of human and ecosystem health. Its advantages are early risk detection, short time of analysis and cost effectiveness, while the main downside of the tested assays is suboptimal specificity in environmental samples as a result of SYBR Green chemistry used.

In particular, we aimed to expand the evaluation of the qPCR method also to understudied benthic cyanobacteria in biofilm samples, with the emphasis on comparison with microscopy and LC-MS/MS. The study demonstrated that in both plankton and biofilm samples there might be a potential for cyanotoxin production which is left undetected by traditional methods. This might be especially important in urban water bodies with regular human and animal visitors. In such water bodies, qPCR could provide additional information if implemented in biomonitoring programs, ensuring appropriate precautions to avoid negative effects of acute and chronic exposure to cyanotoxins.

Furthermore, the study indicated that microscopy as the preferred and often the only method of regular biomonitoring is not sufficient for detecting cyanotoxic potential. Similarly, LC-MS/MS did not detect cyanotoxins in all the samples with observed potential for their production. Implementation of the qPCR method with monitoring strategies could serve for assessing potential toxicity of cyanobacterial blooms or microbial mats and forming recommendations for visitors, as well as in evaluating the efficiency of implemented measures for removal or destruction of cyanobacterial cells or cyanotoxins.

### **5. Methods and Materials**

### *5.1. Cyanobacterial Cultures and Synthetic DNA Fragments*

For specificity testing and cell quantification, reference cyanobacterial cultures were used: microcystin-producing *Anabaena* sp. UHCC 0315 (University of Helsinki, Finland) [16], *Microcystis aeruginosa* PCC 7806 (Pasteur Institute, France) [16] and *Planktothrix agardhii* NIVA-CYA 126 (Norwegian Institute for Water Research, Norway) [17], cylindrospermopsin-producing *Aphanizomenon ovalisporum* ILC-164 (Israel Oceanographic and Limnological Research, Israel) [44] and saxitoxin-producing *Aphanizomenon gracile* NIVA-CYA 851 (Norwegian Institute for Water Research, Norway) [45]. The cultures were grown in standard medium BG11 (Gibco, ThermoFisher Scientific, Waltham, MA, USA) under natural light conditions at room temperature. For DNA extraction (see Section 5.3. DNA Extraction and Quality Control), the cultures were filtered through Sterivex columns (Milipore Sterivex-GP Pressure Filter Unit, Merck KGaA, Germany) with 0.22 μm pore size; the volume filtered was between 5 and 19 mL. We estimated cell concentration using Bürker Türk counting chamber and light microscopy.

For positive controls of qPCR reactions and gene copy quantification, synthetic DNA fragments (gBlocks, Integrated DNA Technologies, Coralville, IA, USA) specific to the selected target regions were used. Their nucleotide sequences were determined based on reference sequences from NCBI GenBank [46]; *Anabaena* sp. 90 (AJ536156.1) for mcyE-Ana, *Microcystis aeruginosa* PCC 7806 (AF183408.1) for mcyE-Mic, *Planktothrix agardhii* 213 (EU151891.1) for mcyE-Pla, *Aphanizomenon* sp. 10E6 (GQ385961.1) for cyrJ, *Anabaena circinalis* AWQC131C (DQ787201.1) for sxtA and *Anabaena circinalis* AWQC131C (AF247589.1) for 16S-cyano (Supplementary file S10).

### *5.2. Environmental Sampling*

In total, 25 plankton and 23 biofilm samples were collected from 7 lakes or reservoirs and 8 rivers or streams in Slovenia in 2019 (Supplementary file S8); one of them (plankton, BL1.4) was included only in cyanotoxin analysis. Plankton samples were collected in lakes with integrating water sampler (Hydro-Bios, IWS III, Germany) 11 times in the pilot area, Lake Bled, and 3–4 times elsewhere. Biofilm samples were collected once in rivers and selected lakes, by brushing biofilm off stones or—if stones were not available—macrophytes, wooden substrate or bricks. Sampling was performed according to national guidelines for monitoring of ecological state of water bodies intercalibrated in the frame of Water Framework Directive implementation. All field equipment used for community DNA samples was treated beforehand with 10% H2O2 solution and rinsed with distilled water. The sampling procedure was controlled in the field by using blank controls with Milli-Q water (Merck KGaA).

Each sample was collected in three aliquotes for community DNA extraction, microscopic cell count and cyanotoxin analysis. For DNA extraction from plankton samples, 60–1000 mL of water was filtered through Sterivex columns (Milipore Sterivex-GP Pressure Filter Unit, Merck KGaA) with pore size 0.22 μm. The columns were stored on –20 ◦C for up to 161 days. For biofilm, 10 mL of biofilm-Mili-Q water mixture was mixed with 40 mL of absolute ethanol and stored on 4 ◦C for up to 147 days. For cyanotoxin analysis, plankton (200–2000 mL) or biofilm (5 or 10 mL) samples were filtered through GF/C filters (Whatman, GE Healthcare, Chicago, IL, USA) with pore size 1.2 μm and dry weight was determined (for biofilm). Filters were stored on –20 ◦C for up to 14 months before analysed with LC-MS/MS. Cyanobacterial cell count was performed under light microscope with counting chambers (e.g., Hydro-Bios, Germany). For plankton samples, cell numbers were counted, and volumes were calculated for each species (i.e., biovolumes). For filamentous cyanobacterial species, where cell numbers cannot be directly counted, first their biovolumes were calculated based on the measurements of filament length and width, and then this was converted to cell numbers based on species-specific cell biovolumes known from literature or from our previous measurements. For biofilm samples, species abundance

was evaluated semi-quantitatively by assigning each species a value of 1–5 based on their abundance (0—not detected, 1—very rare, 2—rare, 3present, 4—frequent, 5—dominant).

### *5.3. DNA Extraction and Quality Control*

DNA extraction was performed using commercially available kits following manufacturers' instructions; for cyanobacterial cultures and plankton samples DNeasy PowerWater Sterivex kit (Qiagen, Germany) and for biofilm samples NucleoSpin Soil kits (Macherey-Nagel, Germany) were used. DNA was stored at –20 ◦C for further analyses. In each extraction, a blank control using sterile water (B. Braun, Germany) was included. DNA concentration and purity of the samples and blank controls were evaluated using spectrophotometer NanoDrop (Thermo Scientific, ThermoFisher Scientific) with 1.5–2 μL of DNA sample and elution buffers from DNA extraction kits as a background.

### *5.4. qPCR Setup, Assay Validation and Quantification*

After a literature review of previously designed qPCR assays for the detection of potentially toxic cyanobacteria, nine published assays were selected and evaluated based on their performance described in the published papers and in silico and in vitro characterisation in this study. In addition to the specificity evaluation done in the original papers (in silico and in vitro for all primers), in silico specificity check was performed with NCBI Primer-BLAST [47] using nr/nt database. Furthermore, specificity was tested in vitro on five different cyanobacterial cultures (see below) for all assays, and for assay 16S-cyano additionally with selected heterotrophic bacterial strains that could be present in freshwater habitats (*Salmonella enterica*, *Escherichia coli*, *Pseudomonas fluorescens*, *Brevundimonas* sp., *Arcobacter butzleri*) and with selected plant samples to check for amplification of plant chloroplasts (*Solanum lycopersicum*, *Vitis vinifera*, *Alnus glutinosa*, *Clematis* sp.). The final set of assays included five qPCR assays to target cyanotoxin-producing cyanobacteria and an additional assay to target 16S rRNA gene, which served as a DNA extraction and qPCR inhibition control (Table 1). Amplification was performed on qPCR cycler Applied Biosystems 7900HT (ThermoFisher Scientific). First, different primer concentrations (0.3 μM and 0.9 μM) and reaction volumes (10 μL and 20 μL) were tested and optimised for each assay. Final reaction volume was 10 μL, consisting of 5 μL of SYBR Green PCR Master Mix (Applied Biosystems, ThermoFisher Scientific), 0.9 μM (assays mcyE-Ana, mcyE-Pla, cyrJ, sxtA and 16S-cyano) or 0.3 μM (assay mcyE-Mic) of each primer and 2 μL of DNA template in 10−<sup>1</sup> dilution. Potential for qPCR inhibition was evaluated on up to five selected samples for assays mcyE-Pla, cyrJ and sxtA, by analysing two subsequent dilutions (10−<sup>1</sup> and 10−2) for each sample. Reactions were performed in 384-well clear PCR plates (Thermo Scientific, ThermoFisher Scientific), covered by MicroAmp™ optical adhesive film (Applied Biosystems, ThermoFisher Scientific). Temperature profile was as follows: 2 min on 50 ◦C, 10 min on 95 ◦C, followed by 45 cycles of 15 s on 95 ◦C and 1 min on 60 ◦C. Dissociation stage with initial denaturation for 15 s on 95 ◦C, followed by 15 s on 60 ◦C and a gradual increase up to 95 ◦C, was added at the end. Every reaction was performed in three technical replicates. Positive (specific synthetic DNA fragments) and negative controls (nuclease-free water, Sigma-Aldrich, St. Louis, MS, USA) were included in every experiment. qPCR amplification conditions were the same for synthetic DNA and DNA isolated from cultures or environmental samples.

qPCR results were analysed with software SDS (version 2.4.1, Applied Biosystems, ThermoFisher Scientific). Threshold values were set manually for each assay within linear part of exponential curve, allowing for the comparison of Cq values between runs. For each sample, the amplification curve (giving Cq value) and dissociation curve of the amplified product (giving Tm value) was checked. To exclude false positives, results were considered positive if the following two criteria were met: Cq was within detection range and there was a distinctive peak with appropriate Tm. Cut-off values for Cq (values at the highest dilution within detection range) and reference values for Tm (average Tm values of all DNA dilutions within quantification range) were determined based on results

from cyanobacterial cultures. Tm values served to control the specificity of amplification, therefore only values in the expected range were considered positive.

Samples that resulted in multiple peaks with one of them showing appropriate Tm were further analysed with agarose gel electrophoresis. qPCR products were run on 2% agarose gel with 1× Tris Acetate-EDTA (TAE) buffer stained with ethidium bromide at 100 V for 90 min. Samples contained 4 μL of each product and 1 μL of 6× Mass Ruler DNA Loading Dye (Thermo Scientific, ThermoFisher Scientific), and GeneRuler 100 bp DNA Ladder (Thermo Scientific, ThermoFisher Scientific) was used to determine the product length. Visualisation of amplicons was performed under UV light using UVP ChemStudio PLUS Imaging System (Analytik Jena, Germany).

Assay sensitivity (LOD, LOQ, linear dynamic range) of the assays was evaluated from the cyanobacterial culture calibration curves. LOD was defined as a number of cells where at least 2/3 technical replicates produced a positive result. LOQ was defined as number of cells in sample that gave Cq value at the lower end of the linear curve and the CV of technical replicates did not exceed 2%. Amplification efficiency (e) was calculated by the equation e = 10−1/S—1, where S represents the slope of the linear dynamic range of the calibration curve. Potential for qPCR inhibition was evaluated by comparing differences between average Cq values of two subsequent dilutions for the same sample.

For results with 3/3 positive technical replicates, average Cq value (which was the basis for quantification of cells and gene copies) and absolute difference between Cq values of technical replicates (which served for assessing intra-assay variability) were calculated. Quantification was performed using calibration curves approach. The calibration curves were generated from eight subsequent dilutions of cyanobacterial cultures DNA for quantification of target cells, or of synthetic DNA fragments for quantification of target gene copies. DNA concentration of stock solution of DNA fragments was 10 ng/μL (according to the manufacturer), from which gene copy numbers were calculated based on the following equation:

$$\text{geene copy number} = \frac{\text{DNA amount} [\text{ng}] \times 6.022 \times 10^{23}}{\text{DNA fragment length} [\text{bp}] \times 650 \times 10^9}$$

Results with calculated values below LOQ were given a value of LOQ/2. Results with 1/3 or 2/3 positive technical replicates were given a value of LOQ/10. These values were used for all following analyses. To ensure comparability of results despite variable volumes of plankton samples filtered prior to DNA extraction and variable densities of biofilm samples, gene copies per microliter of DNA were converted into values per millilitre of water (plankton) or gram of dry weight (biofilm).

Intra-assay variability was tested by using three technical replicates within each experiment. Repeatability (inter-assay variability) of qPCR reactions was tested by repeating the experiment two times with the same template DNA and under the same conditions on up to five selected samples for assays mcyE-Pla, cyrJ and sxtA, and comparing average Cq values of the two technical repetitions. Robustness of the assays was evaluated by testing them on diverse set of samples; cyanobacterial monocultures, synthetic DNA fragments, frozen or lyophilised cyanobacterial bloom samples and environmental samples (plankton, biofilm).

### *5.5. Cyanotoxin Analysis with LC-MS/MS*

Intracellular cyanotoxins were extracted from filters applying the protocol described by Cerasino and Salmaso [48] and quantified with LC-MS/MS. The extraction was carried out by using a mixture of acetonitrile in water (60/40 v/v), containing 0.1% formic acid. Extracted toxins were injected into a LC-MS/MS system, composed of a Waters Acquity UPLC system (Waters, Milford, MA, USA) coupled to a SCIEX 4000 QTRAP mass spectrometer (AB Sciex Pte. Ltd., Singapore). The mass detector was operated in scheduled MRM (Multiple Reaction Monitor) mode, using positive electrospray ionisation (ESI+). Quantification of microcystins was performed following the protocol from Cerasino and Salmaso [48], which has the capability of determining the 11 most common microcystin variants, namely

RR, [D-Asp3]-RR (RRdm), [D-Asp3]-HtyrR (HtyRdm), YR, LR, [D-Asp3]-LR (LRdm), WR, LA, LY, LW, LF. Analysis of cylindrospermopsins and saxitoxins was performed following the protocol from Ballot et al. [49], targeting CYN, STX, dcSTX, NeoSTX, GTX1, GTX4, GTX5, C1 and C2.

### *5.6. Data Analysis*

Calibration curves and their Pearson correlation coefficients were prepared in Microsoft Excel (2007). Correlations and linear regression curves between gene copy numbers and microcystin concentrations, cell numbers and biovolumes were determined with Pearson correlation coefficient, using Prism 6 (GraphPad Inc., San Diego, CA, USA) with 95% confidence interval. For biofilm samples, relative abundance was calculated by summing up the values of individual target species and normalising the summed value to 100%. In order to include negative values (below LOD) in correlation analysis, they were replaced with a minimum detected value for each assay divided by 100 (for microscopy and LC-MS/MS results) or by 10 (for qPCR results).

**Supplementary Materials:** The following are available online at https://www.mdpi.com/2072-665 1/13/2/133/s1.

**Author Contributions:** Š.R.R., T.Š., A.K.K. and M.Z. carried out the sampling campaign. Š.R.R., T.Š., A.K.K. and T.E. performed microscopic analyses. L.C. performed LC-MS/MS measurements. P.K., M.Z. and T.E. planned the qPCR experiments with the help of Š.B. M.Z. performed molecular laboratory work and associated data analyses. T.E. performed statistical analyses. P.K. and Š.B. contributed to interpretation of qPCR results. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by ARRS research programs P1-0245 and P4-0165, project Interreg Eco-AlpsWater (569) and MORS contract no. 848-23 and annex no. 1/2019.

**Data Availability Statement:** Majority of the data produced in this study is available in supplementary files. Additional data is available upon request to the corresponding author.

**Acknowledgments:** We are grateful to Rainer Kurmayer (UIBK) for contributing comments to the manuscript, Assaf Sukenik (IOLR) for providing us with cyanobacterial culture *Aphanizomenon ovalisporum* ILC-164, Zala Kogej (NIB) for technical help with laboratory work, Janja Zabukovnik for providing us with test environmental samples, Nataša Mehle (NIB) for providing us with plant samples, and Valentina Turk (NIB) and Olga Zorman Rojs (UL, Veterinary faculty) for providing us with heterotrophic bacteria samples.

**Conflicts of Interest:** The authors declare no conflict of interest.

### **References**


### *Article* **Roles of Nutrient Limitation on Western Lake Erie CyanoHAB Toxin Production**

**Malcolm A. Barnard 1,\*, Justin D. Chaffin 2, Haley E. Plaas 1, Gregory L. Boyer 3, Bofan Wei 3, Steven W. Wilhelm 4, Karen L. Rossignol 1, Jeremy S. Braddy 1, George S. Bullerjahn 5, Thomas B. Bridgeman 6, Timothy W. Davis 5, Jin Wei 7, Minsheng Bu <sup>7</sup> and Hans W. Paerl 1,\***


**Abstract:** Cyanobacterial harmful algal bloom (CyanoHAB) proliferation is a global problem impacting ecosystem and human health. Western Lake Erie (WLE) typically endures two highly toxic CyanoHABs during summer: a *Microcystis* spp. bloom in Maumee Bay that extends throughout the western basin, and a *Planktothrix* spp. bloom in Sandusky Bay. Recently, the USA and Canada agreed to a 40% phosphorus (P) load reduction to lessen the severity of the WLE blooms. To investigate phosphorus and nitrogen (N) limitation of biomass and toxin production in WLE CyanoHABs, we conducted in situ nutrient addition and 40% dilution microcosm bioassays in June and August 2019. During the June Sandusky Bay bloom, biomass production as well as hepatotoxic microcystin and neurotoxic anatoxin production were N and P co-limited with microcystin production becoming nutrient deplete under 40% dilution. During August, the Maumee Bay bloom produced microcystin under nutrient repletion with slight induced P limitation under 40% dilution, and the Sandusky Bay bloom produced anatoxin under N limitation in both dilution treatments. The results demonstrate the importance of nutrient limitation effects on microcystin and anatoxin production. To properly combat cyanotoxin and cyanobacterial biomass production in WLE, both N and P reduction efforts should be implemented in its watershed.

**Keywords:** cyanotoxins; Maumee Bay; Sandusky Bay; *Microcystis*; *Planktothrix*; microcystin; anatoxin

**Key Contribution:** Nutrient limitation of cyanobacterial harmful algal blooms (CyanoHABs) was investigated with respect to the production of the cyanotoxins microcystin and anatoxin in Maumee Bay and Sandusky Bay in Western Lake Erie. This is one of the first studies investigating nutrient limitation effects on anatoxin production in Lake Erie and one of the first studies to evaluate the effects of nutrient reduction on Western Lake Erie CyanoHABs using nutrient dilution assays. To reduce CyanoHABs and their toxicity, both N and P reductions are needed in the Western Lake Erie watershed.

**Citation:** Barnard, M.A.; Chaffin, J.D.; Plaas, H.E.; Boyer, G.L.; Wei, B.; Wilhelm, S.W.; Rossignol, K.L.; Braddy, J.S.; Bullerjahn, G.S.; Bridgeman, T.B.; et al. Roles of Nutrient Limitation on Western Lake Erie CyanoHAB Toxin Production. *Toxins* **2021**, *13*, 47. https://doi.org/ 10.3390/toxins13010047

Received: 10 December 2020 Accepted: 6 January 2021 Published: 9 January 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

### **1. Introduction**

Freshwater ecosystems are critical for sustaining life and supporting civilizations throughout history [1]. As the global human population grows, increased urbanization, agricultural and industrial productions, combined with insufficient wastewater treatment practices, have led to a widespread increase in nutrient pollution of these ecosystems, threatening clean and safe water supplies [2]. Excessive inputs of nitrogen (N) and phosphorus (P) have accelerated eutrophication, the process of increasing organic enrichment, which is largely attributable to increased microalgal and aquatic macrophyte growth [3]. The major detrimental impacts of eutrophication include harmful algal bloom (HAB) formation, decreased water transparency (increased turbidity), O2 depletion, and reduced biodiversity [3,4]. HAB formation has been a major water quality issue in the U.S. since the 1960s, as noted in a 1965 White House Report indicating HABs as a major source of environmental degradation [5]. Furthermore, nutrient-driven eutrophication of lakes and rivers is one the most significant causes of water quality decline globally [3,6–8]. In particular, there are growing concerns about the proliferation and diversification of N- and P-based fertilizers, as they are potent stimulants of aquatic primary production along the freshwater to marine continuum [9,10]. Additionally, climate change (e.g., warming and changing precipitation patterns) is increasing the likelihood of more expansive blooms, exposing human and animal populations (e.g., pets, wildlife, cattle, fish, birds) to waterborne and aerosolized toxins [7,11–14]. Despite CyanoHAB toxicity being a major human and ecosystem health hazard, the causes and controls of underlying toxicity mechanisms remain poorly understood [15].

Blooms of cyanobacteria in Lake Erie, largely dominated by filamentous heterocystous (N2-fixing) forms (*Anabaena/Dolichospermum*, *Aphanizomenon)*, were common in the late 1950s through to the 1970s. These blooms dissipated following the signing in of the *Great Lakes Water Quality Agreement* of 1972, which was updated in 2012. However, the blooms returned as non-N2 fixing *Microcystis* blooms in the early 2000s, which have continued and perhaps worsened [16,17], leading to major environmental degradation and increased human health risks [7]. In August 2014, a toxic *Microcystis* spp. bloom in Western Lake Erie (WLE) created a water crisis, forcing public water supplies to be shut down for over 400,000 people in Toledo, OH, USA [7,18]. Nutrient runoff from agricultural nonpoint sources has been a major factor promoting CyanoHABs in WLE [7]. Primary production in Maumee Bay of Lake Erie (largely dominated by *Microcystis* spp. in the summer) shifts from P-limitation to N-limitation with spatial nutrient limitation heterogeneity with Nand P-limitation occurring several km apart [19–21]. Prior studies revealed that during the summer months, N was often depleted in embayments such as Sandusky and Maumee Bay [22–26], where summertime molar N:P ratios for Sandusky Bay remained below the canonical Redfield ratio (16:1) [26–28]. This suggests the presence of strong N sinks, mediated by denitrification and/or active N cycling and N uptake by high amounts of algal biomass [28–30]. The primary fertilizers used in the agriculturally dominated drainage basin of Lake Erie are inorganic fertilizers (ammonium nitrate, urea, and phosphate) and manure, which has low N:P ratios (~5:1), is about 20% [31–34]. There is an urgent need to determine the linkage between different bioreactive forms of N and P and the promotion of toxic CyanoHABs, to establish the necessary reduction in these nutrient forms to ensure the security of surface potable water. Nutrient reduction will likely need to be even greater as climate change increases the N and P reduction thresholds required for CyanoHAB mitigation [35,36]. The *in situ* bioassay-based study reported here is among the first to use an experimental approach to investigate the response of a natural CyanoHAB community dominated by either *Microcystis* (Maumee Bay) or *Planktothrix* (Sandusky Bay) to actual reductions in N, P or both, under natural conditions in Lake Erie. Satellite and field images of the 2019 WLE blooms can be seen in Figure 1.

**Figure 1.** Images of the 2019 WLE CyanoHABs. (**A**) Satellite imagery from the NASA Terra satellite of the WLE CyanoHAB on 19 August 2019 as provided by NOAA MODIS [37]; (**B**) Maumee Bay *Microcystis*-dominated cyanobacterial harmful algal bloom (CyanoHAB) on 4 August 2019 during sampling for the August 2019 bioassays. Photo by H. Plaas; (**C**) Sandusky Bay *Planktothrix*-dominated bloom on 4 August 2019 during sampling for the August 2019 bioassays. Photo by H. Plaas.

A recent review suggested that management efforts to reduce P pollution without controlling N inputs have caused nutrient imbalances in eutrophic systems, which may favor toxic CyanoHABs incapable of fixing atmospheric N2 gas, i.e., requiring combined N sources [24]. Prior to P load reductions in the 1970s, CyanoHABs in Lake Erie were mostly the N2-fixers *Aphanizomenon* and *Dolichospermum*, formerly *Anabaena* [38]; now, CyanoHABs are primarily non-N2-fixing *Microcystis* and *Planktothrix* [16]. In WLE, molecular analysis of the *Microcystis* community indicates a shift from toxic to non-toxic strains that correlates with NO3 availability [39], although there appears to be a temporal disconnect as a multiyear analysis found no correlation between the proportion of microcystin-producing genotypes of *Microcystis* and the concentration of microcystin [40]. Recent work has strengthened links between N availability, dominant strain shifts, and toxicity by showing seasonal trends in these patterns [24]. The inability of these cyanobacteria to fix atmospheric N2, and their strong affinity for reduced N forms (e.g., NH4 and urea), suggests that N delivered through agricultural runoff and internal N recycling mechanisms play critical roles in modulating total phytoplankton biomass, CyanoHAB community composition, and toxicity [39,41].

The prominent cyanotoxins, microcystin and anatoxin, have molecular structures containing N, suggesting that their syntheses may be linked to N availability; hence, there is a need to investigate the potential roles N fertilizers (i.e., NH4, NO3, and urea) play in bloom dynamics and toxin production in Lake Erie [23,36,42]. A recent study showed that there are N concentration reduction thresholds at which bloom microcystin levels will decrease, leading to further evidence that N limitation may play a role in controlling cyanotoxin production in the WLE blooms [41]. Due to the shift to non-N2-fixing CyanoHABs, a major unknown concerning this shift in nutrient limitation is how specific microcystin and anatoxin production potentials are linked to nutrient input reductions.

The US Environmental Protection Agency (EPA) and Environment and Climate Change Canada have recommended a 40% reduction in springtime P loading into WLE to help control the blooms [43–46]. The 40% P load reduction was the result of a multiple modeling exercise included in the Great Lakes Water Quality Agreement between the US and Canada [47]. As both N and P have been shown to influence the WLE CyanoHABs, it is crucial to investigate the effects of both 40% reductions in both N and P in addition to the investigations of the effects of N and P addition. Here, we addressed the following questions: (1) how do nutrients influence WLE microcystin and anatoxin production? (2) Do the same nutrients limit toxin production and CyanoHAB biomass? (3) Will the 40% P reduction as recommended by the US EPA be effective in reducing CyanoHAB microcystin and anatoxin and biomass production in WLE? (4) Is P reduction alone enough to decrease WLE CyanoHAB biomass and microcystin and anatoxin production, or is a combined N and P reduction strategy needed? Given the relatively high content of N in the cyanotoxins microcystin and anatoxin, we predicted that cyanotoxin production is N-limited and that excessive N inputs promote toxicity of these non-N2-fixing CyanoHABs.

### **2. Results**

### *2.1. June 2019 Experiement*

June 2019 bioassay experiments were characterized by a late spring diatom bloom shortly before the onset of a summer *Microcystis* bloom in Maumee Bay and the very early Planktothrix bloom in Sandusky Bay. In both Maumee and Sandusky Bays, there were high N concentrations—over 200 μmol L−<sup>1</sup> nitrate plus nitrite in Maumee Bay and over 100 μmol L−<sup>1</sup> nitrate plus nitrite in Sandusky Bay and relatively low P concentrations of 1–2 μmol L−<sup>1</sup> dissolved reactive phosphorus (DRP) (Table 1).


**Table 1.** Initial nutrient concentrations in the June 2019 bioassay water collected from control Cubitainers. All data are *n* = 3.

In the June Maumee Bay experiment, growth rates significantly differed (*p* < 0.001) among nutrient treatments, but there was no difference between the undiluted and diluted treatments (*p* = 0.76). The +P and +P&N treatments resulted in a higher growth rate than the control and +N treatments, indicating P-limited growth, in both the undiluted and 40% dilution treatments, likely due to the high concentrations of N in the bay (Figure 2; Table 1, Tables S1 and S2). Cyanotoxins were not detected in the June Maumee Bay experiment.

**Figure 2.** Growth rates of phytoplankton, as determined by chlorophyll *a* accumulation during the course of incubation in the June 2019 bioassays. (**A**) Undiluted Maumee Bay water (also see Table S1); (**B**) undiluted Sandusky Bay water (also see Table S1); (**C**) 40% dilution Maumee Bay water (also see Table S1); (**D**) 40% dilution Sandusky Bay water (also see Table S1); (**E**) Maumee Bay growth rates under the various nutrient addition treatments at the two locations of T3 compared to T0 (also see Table S2). Error bars are standard error; (**F**) Maumee Bay growth rates under the various nutrient addition treatments at the two locations of T3 compared to T0 (also see Table S2). Error bars are standard error. Significances between treatments for (**E**,**F**) are from two-factor ANOVAs.

> In the June Sandusky Bay experiment, nutrient enrichment did not impact growth rates (*p* = 0.68), but growth rates were lower in the 40% diluted treatments (*p* = 0.013); Figure 2; Tables S1 and S2), which indicates nutrient-replete conditions. The initial undi

luted total microcystin concentration was 0.136 μg/L and total anatoxin concentration was 0.053 μg/L (Tables S3 and S5). Microcystin concentrations increased throughout the experiment in the no dilution treatments but not the 40% dilution (Figure 3). The Sandusky Bay microcystin production rate was slightly yet not significantly affected by nutrient enrichment (*p* = 0.067), becoming significant in the biomass-normalized analyses (*p* < 0.001). However, the production rate was lower in the 40% diluted treatments (Figure 3 and Tables S3—S6). The June 2019 Sandusky Bay anatoxin production rate was not affected by dilution treatment with a slight nutrient effect in the biomass-normalized analyses (*p* < 0.01) (Figure 4).

**Figure 3.** Production rates of microcystin during the June 2019 bioassays. Only Sandusky Bay produced microcystin in June. (**A**) Undiluted Sandusky Bay microcystin concentrations (also see Table S3); (**B**) undiluted Sandusky Bay biomassnormalized microcystin concentrations (also see Table S4); (**C**) 40% dilution Sandusky Bay microcystin concentrations (also see Table S3); (**D**) 40% dilution Sandusky Bay biomass-normalized microcystin concentrations (also see Table S4); (**E**) Maumee Bay microcystin production rates under the various nutrient addition treatments at the two locations of T3 compared to T0 (also see Table S5); (**F**) Maumee Bay biomass-normalized microcystin production rates under the various nutrient addition treatments at the two locations of T3 compared to T0 (also see Table S6). Error bars are standard error. Significance for (**E**,**F**) is from *n*-factor ANOVA analysis due to unbalanced data sets.

**Figure 4.** Chlorophyll *a*-based production rates of anatoxin during the June 2019 bioassays. Only Sandusky Bay produced anatoxin in June. (**A**) Undiluted Sandusky Bay anatoxin concentrations (also see Table S7); (**B**) undiluted Sandusky Bay biomass-normalized anatoxin concentrations (also see Table S8); (**C**) 40% dilution Sandusky Bay anatoxin concentrations (also see Table S7); (**D**) 40% dilution Sandusky Bay biomass-normalized anatoxin concentrations (also see Table S8); (**E**) Maumee Bay anatoxin production rates under the various nutrient addition treatments at the two locations of T3 compared to T0 (also see Table S9); (**F**) Maumee Bay biomass-normalized anatoxin production rates under the various nutrient addition treatments at the two locations of T3 compared to T0 (also see Table S10). Error bars are standard error. Significance for (**E**,**F**) is from *n*-factor ANOVA analysis due to unbalanced data sets.

### *2.2. August 2019 Experiment*

August experiments were characterized by dense blooms of *Microcystis* in Maumee Bay and *Planktothrix* in Sandusky Bay. Maumee Bay had high N concentrations—over 100 μmol L−<sup>1</sup> nitrate plus nitrite—but Sandusky Bay had low N concentrations with 6.5 μmol L−<sup>1</sup> nitrate plus nitrite (Table 2). Both Maumee and Sandusky Bay had low P concentrations of 0.03 to 0.20 μmol L−<sup>1</sup> DRP (Table 2).

**Table 2.** Initial concentrations of nutrients in the August 2019 bioassay water taken from T0 control Cubitainers. All data are *n* = 3.


Chlorophyll *a* concentrations decreased throughout incubation of the very dense bloom in the August Maumee Bay experiment, coinciding with negative growth rates (Figure 5). The diluted treatments had reduced algal mortality compared to the undiluted treatments likely due to lower initial starting biomass (*p* < 0.001). The growth rate was significantly affected by nutrients (*p* < 0.001) and the interaction between nutrients and dilution (*p* = 0.012), but there was no discernable pattern, leading to a lack of ecological significance. The initial undiluted total microcystin concentration in the August Maumee Bay experiment was 18.06 μg/L. Microcystin concentration and production rates (Figure 6) followed a similar pattern to chlorophyll with a non-significant nutrient effect (*p* = 0.14) and significant dilution effect without biomass-normalization (*p* < 0.001) and a non-significant effect in biomass-normalized analysis (*p* = 0.5452). Anatoxin was not detected in the Maumee Bay August experiment.

**Figure 5.** Growth rates of phytoplankton in the August 2019 bioassays. (**A**) Undiluted Maumee Bay Chlorophyll *a* (also see Table S1); (**B**) undiluted Sandusky Bay Chlorophyll *a* (also see Table S1); (**C**) undiluted Maumee Bay Chlorophyll *a* (also see Table S1); (**D**) 40% dilution Sandusky Bay Chlorophyll *a* (also see Table S1); (**E**) Maumee Bay growth rates under the various nutrient addition treatments at the two locations at T3 compared to T0 (also see Table S2). Error bars are standard error. Significance for (**E**) is from two-factor ANOVA analysis; (**F**) Maumee Bay growth rates under the various nutrient addition treatments at the two locations at T3 compared to T0 (Table S2). Error bars are standard error. Significance for (**F**) is from *n*-factor ANOVA analysis due to unbalanced data sets.

**Figure 6.** Production rates of microcystin during the August 2019 bioassays. Only Maumee Bay produced microcystin in all samples. (**A**) Undiluted Maumee Bay microcystin concentrations (also see Table S3); (**B**) undiluted Maumee Bay biomassnormalized microcystin concentrations (also see Table S4); (**C**) 40% dilution Maumee Bay microcystin concentrations (also see Table S3); (**D**) 40% dilution Maumee Bay biomass-normalized microcystin concentrations (also see Table S4); (**E**) Maumee Bay microcystin production rates under the various nutrient addition treatments at the two locations of T3 compared to T0 (also see Table S5); (**F**) Maumee Bay biomass-normalized microcystin production rates under the various nutrient addition treatments at the two locations of T3 compared to T0 (also see Table S6). Error bars are standard error. Significance for (**E**,**F**) is from 2-factor ANOVA analysis.

In the August Sandusky Bay experiment, chlorophyll concentration increased throughout the incubation in the three N-only treatments and the + N&P treatment, while it declined in the control and P-only treatment in both the diluted and non-diluted treatments (Figure 5), which indicates N was the primary limiting nutrient. The various forms of N did not exert a discernable difference on growth rates. The highest growth rates were measured in the +N&P treatments, which indicates a secondary P limitation. The dilution effect was also significant (*p* = 0.004). The initial undiluted anatoxin concentration was 0.596 μg/L (Figure 7). Anatoxin production was primarily N-limited both with and without biomass normalization (*p* < 0.001), like growth rates, but P was not secondarily limiting. Unlike chlorophyll, which decreased throughout incubation in the control and P-only treatment, anatoxin concentrations in the control and P-only treatment remained constant throughout the incubation due to production rates of anatoxin increasing throughout the incubation.

**Figure 7.** Production rates of anatoxin during the August 2019 bioassays. Only Sandusky Bay produced anatoxin in August. (**A**) Undiluted Sandusky Bay anatoxin concentrations (also see Table S7); (**B**) undiluted Sandusky Bay biomass-normalized anatoxin concentrations (also see Table S8); (**C**) 40% dilution Sandusky Bay anatoxin concentrations (also see Table S7); (**D**) 40% dilution Sandusky Bay biomass-normalized anatoxin concentrations (also see Table S8); (**E**) Maumee Bay anatoxin production rates under the various nutrient addition treatments at the two locations of T3 compared to T0 (also see Table S9); (**F**) Maumee Bay biomass-normalized anatoxin production rates under the various nutrient addition treatments at the two locations of T3 compared to T0 (also see Table S10). Error bars are standard error. Significance for (**E**,**F**) is from 2-factor ANOVA analysis.

### **3. Discussion**

Given that CyanoHABs and their associated cyanotoxins have led to adverse human and ecosystem health outcomes in WLE [18], it is important to clarify the major driver(s) of CyanoHAB toxicity. This study investigated nutrient limitation on biomass production and cyanotoxin production, focusing on microcystin and anatoxin. We found that high concentrations of both major nutrients, P and N, drove CyanoHAB growth and microcystin and anatoxin production in WLE. We also found times when the 40% reduction in nutrients could slow microcystin production during nutrient replete conditions (Figure 3E,F).

We found that the June 2019 late spring diatom bloom in Maumee Bay was P-limited, which was induced in both the undiluted and 40% dilution samples due to high ambient N concentrations (>100 μmol/L), while the June 2019 Sandusky Bay *Planktothrix* bloom was not affected by nutrient addition, but growth was slowed following a 40% reduction in nutrients. This is possibly explained by the rapid growth associated with the early bloom, with the 40% reduction in nutrients dropping below the threshold needed to support this bloom [48]. During the bloom maxima in August 2019, the Maumee Bay *Microcystis* bloom was nutrient replete under both undiluted and 40% dilution treatments, with less of a decline in the biomass due to the 40% lower starting biomass following dilution. Additionally, ammonium concentration was higher in the initial 40% dilution than the undiluted sample in both the June and August 2019 Maumee Bay, likely due to an initial die off in the subsample, leading to increased regenerated N as ammonium. These results are likely due to bottle effects attributable to the very high biomass; restricted exchange of gases and nutrients [49–51]. The August Sandusky Bay *Planktothrix* bloom was N-limited

in both the 40% reduction and the undiluted samples. All nutrient concentrations in the August 2019 Sandusky Bay 40% dilution were higher than concentrations in the undiluted treatment, likely due to the rapid growth of the *Planktothrix* bloom using up more nutrients in the undiluted control group prior to sample filtration, when compared to the reduced biomass in the 40% dilution. Differences between the effects of the different N species were not significant at either location during either experimental period, which has been seen previously in strongly N-limited blooms in WLE [52], but differs from past findings in WLE during periods of weaker N-limitation [22,53–55]. This could be due to the high ambient concentrations of NO3 paired with low NH4 (Tables 1 and 2). Our findings of N limitation contradict the previous assumption that P availability exclusively controls CyanoHABs [56–59]. Instead, these findings support the paradigm shift to also consider N input reductions to mitigate CyanoHABs [19,29,60,61].

During the early Sandusky Bay *Planktothrix* bloom (June 2019), microcystin production shifted from between N and P co-limitation in the undiluted samples to nutrient deplete conditions in the 40% dilution samples. This is likely due to the bloom's use of nutrient resources early on to support biomass production rather than produce secondary metabolites, e.g., cyanotoxins, possibly due to the genetic inability of the June populations to produce the microcystin as seen in prior years [62,63]. Alternatively, the cells could have lysed due to viral or other processes and the dissolved microcystin was not captured on the 0.7 μm porosity GF/F filters or degraded [64,65]. At its peak in August 2019, the *Microcystis* bloom in Maumee Bay was the only bloom that produced microcystin. This production of microcystin occurred under nutrient replete conditions, with less of a decline in microcystin concentrations with slight P limitation in the diluted samples and no apparent nutrient limitation in the undiluted samples. Neither experiments showed significant effects of the various forms of N.

Even though cyanobacteria require N to produce N-rich microcystin, P is also required for cellular growth to allow for higher microcystin concentrations. As the ratio of microcystin to chlorophyll *a* in both June and August was nearly linear (Figures 3 and 6), we conclude that the primary bloomers—*Planktothrix* in Sandusky Bay and *Microcystis* in Maumee Bay—were the primary producers of microcystin. The P requirement for microcystin production has been observed in prior studies in Lake Erie, and in several German lakes [66]. This deviates from previous studies that clearly demonstrated links between N availability and higher N:P and bloom toxicity in microcystin-producing blooms [7,67–69]. This could be due to microcystin being an "N bargain" with a C:N ratio of 4.9:1 compared to the average of 3.6:1 in a survey of 2000 proteins [69]. However, P-limitation of microcystin production has been shown to occur in chemostat experiments [70] and in a transcriptome experiment on Lake Erie blooms [40]. The microcystin congener pattern observed in these experiments followed what was expected for North American lakes, including Lake Erie, with microcystin LR, YR, RR being the dominant congeners [18].

We observed anatoxin production in the Sandusky Bay *Planktothrix* bloom during both early and peak blooms. This is the first study showing anatoxin production in Lake Erie, although it has been shown that anatoxin production can occur during *Planktothrix* blooms accompanied by other cyanobacteria, including *Cuspidothrix issatschenkoi*, which has previously been identified in Sandusky Bay [23,71–75]. This was likely the case, as the biomass normalized anatoxin production mirrors the anatoxin production in the nonnormalized analysis (Figures 4 and 7), meaning that secondary cyanobacterial species may be driving the anatoxin production in Sandusky Bay. During the early *Planktothrix* bloom in June 2019, there was no apparent nutrient limitation in the undiluted treatments. However, there was co-limitation by both N and P in the diluted treatments. During the peak bloom in August 2019, anatoxin production was N-limited in both the undiluted and 40% diluted samples. While no differences were found between forms of N added in the June bioassay, during the peak bloom in August, NO3 additions led to higher concentrations of anatoxin compared to NH4 and urea additions. Additionally, N limitation of anatoxin production has been shown previously [76]. As observed in this experiment, higher overall

N concentrations lead to higher anatoxin concentrations, with NO3 enrichment leading to the largest increase in anatoxin production, which parallels results from other limnetic anatoxin-producing CyanoHABs [77–81]. Anatoxin production in Sandusky Bay and other *Planktothrix*-dominated bodies of water needs further examination, given the neurotoxicity and potential developmental toxicity of anatoxin [82,83] as well as its multiple deleterious environmental effects [84,85].

Nutrient concentrations were very high during both the early and peak 2019 bloom in Maumee Bay with 223.67 ± 25.43 <sup>μ</sup>g L−<sup>1</sup> NO3 and 2.224 ± 1.008 <sup>μ</sup>g L−<sup>1</sup> DRP in June and 127.12 ± 10.82 <sup>μ</sup>g L−<sup>1</sup> combined NO3 and NO2 in August and 0.203 ± 0.199 <sup>μ</sup>g L−<sup>1</sup> DRP. Similar to Maumee Bay, Sandusky Bay exhibited high nutrient concentrations in June with 101.45 ± 5.95 <sup>μ</sup>g L−<sup>1</sup> NO3 and 0.203 ± 0.138 <sup>μ</sup>g L−<sup>1</sup> DRP in June, but had lower nutrient concentrations in August with 127.12 ± 10.82 <sup>μ</sup>g L−<sup>1</sup> NO3 in August and 0.032 ± 0.012 μg L−<sup>1</sup> DRP. This is likely due to larger nutrient loads from the Maumee River than from the Sandusky River, as seen previously in 2007 [86]. The high nutrient loads were exacerbated by elevated precipitation associated with a very wet winter in 2019 [87], which will likely continue to be an issue as high precipitation events are predicted to continue in the future [88–90]. Denitrification and assimilation draw down nitrate to concentrations below the threshold of detection (<0.5 μmol/L) throughout summer and fall in western Lake Erie and Sandusky Bay [28,91], which is a pattern that occurs independent of tributary nutrient loads [19]. Our Maumee Bay experiments occurred before nitrate depletion, and therefore, we would expect to have observed N-limited growth and microcystin production following the N depletion [25]. However, it remains to be seen how a 40% dilution in nutrients (N and P) would affect N-limited *Microcystis* in late summer. Therefore, nutrient input reductions need to target both N and P rather than just P as recommended by the US EPA and Environment and Climate Change Canada [43–45,92]. While P reduction is actively pursued [93], N management strategies are required as well [35,94,95].

### **4. Conclusions**

Our results suggest that nutrient dynamics play a crucial role in the WLE CyanoHABs for both biomass production as well as microcystin and anatoxin production in the eutrophic Sandusky and Maumee Bays. During the peak bloom periods when microcystin and anatoxin concentrations are highest, microcystin production was nutrient deplete and anatoxin production was N-limited. Maumee Bay biomass shifted from P-limited immediately prior to the *Microcystis* bloom to nutrient deplete during peak bloom, while the Sandusky Bay *Planktothrix* bloom shifted from nutrient deplete to N-limited from early bloom to peak bloom. A 40% reduction in N and P led to a slight reduction in biomass and microcystin and anatoxin production. However, further studies are needed to investigate the long-term nutrient reduction thresholds needed to control CyanoHABs. With N and P enrichment stimulating the WLE CyanoHABs, there is a need to constrain external loads of both N and P, and impose stricter nutrient-limited conditions in order to help mitigate the CyanoHAB problem in WLE [52,96–98]. Our study took place only in eutrophic bays and we showed that a 40% reduction might not be enough in Maumee and Sandusky Bay because growth and toxin production could still be nutrient-saturated. Future studies are needed to determine if a 40% reduction is adequate for the open waters of WLE. Furthermore, an adaptive management approach is needed to determine if the 40% reduction goal needs to be adjusted with changes in land use practices and climate change [99]. Additionally, future studies should focus on drawing direct functional links between nutrient enrichment and cyanotoxin production, e.g., Krausfeldt et al. [36]. Lastly, anatoxin should be more closely monitored in WLE, as it is a potent neurotoxin with human health-associated implications [100].

### **5. Materials and Methods**

### *5.1. Bioassay Methods*

We performed experimental manipulations of natural Maumee Bay (Oregon, OH, USA) and Sandusky Bay (Sandusky, OH, USA) phytoplankton communities that were collected from nearshore docks (Figure 8; Table S11). Water was pumped from 1 m below the surface into pre-cleaned (flushed with lake water) 20 L carboys using a non-destructive diaphragm pump and was transported to The Ohio State University Stone Laboratory on South Bass Island (Put-in-Bay, OH, USA) (Figure 8).

This experiment deployed in situ bioassays, using 4 L pre-cleaned polyethylene Cubitainers to which natural lake water was added from Maumee and Sandusky Bays using the methodology described in Paerl et al. [101] and Xu et al. [102]. Microcosm treatments were individually amended with either 100 μM N of NO3 (as KNO3), 100 μM N of NH4 (as NH4Cl), 6 μM PO4 (as KH2PO4), 100 μM N and 6 μMP added as a combined addition of 50 μM NO3, 50 μM NH4, and 6 μM PO4, and, in August 2019, urea (50 μM urea to achieve 100 μM N), yielding similar total dissolved nutrient concentrations (for each treatment) and falling within a range matching riverine dissolved inorganic nutrient discharge into Lake Erie nearshore waters. To avoid silica or dissolved inorganic carbon limitation in Cubitainers during the incubation period, we added 50 μM Si as Na2SiO3 and 10 mg L−<sup>1</sup> (83.25 μM) DIC as NaHCO3 based on previous Si and DIC values from Hanson et al. [103] and Rockwell et al. [104]. We used a major ion solution (MIS) specific to WLE to provide 40% dilutions to mimic the EPA-recommended reductions in P inputs to WLE as well as a parallel 40% reduction in N, as both N and P have been shown to influence WLE CyanoHAB bloom dynamics [22,39,43]. The 40% dilution control investigated a 40% reduction in both N and P. Incubations were run for 72 h at a lake site near the Stone Laboratory at ambient lake water temperatures and light conditions [23,101,102]. Based on previous work on eutrophic Lake Taihu, China [95], a 72 h maximum incubation period was chosen to minimize "bottle effects", while having ample time to examine phytoplankton growth, microcystin, and anatoxin production responses.

To perform nutrient dilutions, we developed a major ion solution (MIS) for WLE, which provided a N- and P-free dilution media to minimize hypertonic and hypotonic effects on the organisms in the samples by balancing major dissolved ions in the system (Table 3). As an example, artificial seawater is the MIS for the open ocean. For WLE, we based the ambient ion concentrations on a past study by Chapra et al. [105]. As there is substantial natural variability due to rainfall and evaporative effects and the ions in the MIS are in micromolar concentrations and pulse events change the ions in WLE, these deviations are considered reasonable. The compounds used in the MIS are found in Table S12.


**Table 3.** Concentrations of major ions in the ambient Lake Erie water and the major ion solution (MIS) used for the dilutions in the bioassays.

<sup>1</sup> Constituents of MIS can be found in Table S12; <sup>2</sup> lower concentration compared to ambient concentration; <sup>3</sup> higher concentration compared to average ambient concentrations.

**Figure 8.** Map of the sampling sites and the location of the incubation. Maumee Bay water was collected off a bulkhead dock near the University of Toledo Lake Erie Center in Oregon, OH, USA. Sandusky Bay sampling took place at a dock outside the Paper District Marina in Sandusky, OH, USA. Incubation took place at The Ohio State Stone Laboratory on South Bass Island (Put-In-Bay, OH, USA). GPS coordinates for the sampling and incubation sites can be found in Table S11. This figure was created with www.simplemappr.net [106].

### *5.2. Phytoplankton Biomass Determination*

Chlorophyll *a*, as an indicator of phytoplankton biomass, was measured on subsampled samples by filtering 50 mL of sample water onto Whatman glass fiber filters (GF/F). Filters were frozen at −20 ◦C and subsequently extracted using a tissue grinder in 90% acetone [107,108]. Chlorophyll *a* in extracts was measured using the non-acidification method of Welschmeyer [109] on a Turner Designs Trilogy fluorometer calibrated with pure Chlorophyll *a* standards (Turner Designs, Sunnyvale, CA, USA).

### *5.3. Nutrient Concentration Determination*

Nutrient samples were collected in 50 mL Falcon tubes by collecting the GF/F filtered water from the chlorophyll *a* sample collection and frozen at −20 ◦C. A continuous segmented flow auto-analyzer (QuAAtro SEAL Analytical Inc., Mequon, WI, USA) was used to quantify nitrate, nitrite, ammonium, dissolved reactive P, and silicate using standard U.S. EPA methods [110]. Urea concentration (as urea-N) was determined spectrophotometrically [52,111,112].

### *5.4. Anatoxin and Microcystin Determinations*

Cyanotoxins were measured on subsampled samples by filtering 50 mL of the sample water onto Whatman GF/F. Filters were frozen at −20 ◦C until extraction with ultrasonic sonication in 5 mL of 50% methanol and 1% acetic acid. Samples were centrifuged at 14,000× *g* for 10 min at 4 ◦C. The supernatants were filtered through 0.45 μm pore-size nylon syringe filters (Corning, CLS431225) and stored at −20 ◦C until analysis. Microcystin was quantified via coupled liquid chromatography/mass spectrometry using methods modified from Boyer [113] and Peng et al. [114]. Reverse-phase liquid chromatography using a Waters 2695 solvent delivery system (Waters, Milford, MA, USA) coupled to a Waters ZQ4000 mass spectrometer (Waters, Milford, MA, USA) (m/z 500–1250 amu) and a 2996 photodiode array detector (Waters, Milford, MA, USA) (210 to 400 nm wavelength) was used to screen for molecular ions of 22 common microcystin congeners (RR, dRR, mRR, H4YR, hYR, YR, LR, mLR, zLR. dLR, meLR, AR, FR, WR, LA, dLA, mLA, LL, LY, LW, LF, WR). Separation conditions used an ACE 5 C18, 150 × 3.0 mm column and a 30–70% aqueous acetonitrile gradient containing 0.1% formic acid at a flow rate of 0.3 mL min. Individual congener concentrations were quantified using the peak area of the extracted ion relative to standards of microcystin-LR (Enzo Life Sciences, Ann Arbor, MI, USA). This allows quantification of congeners where standards are not available. Detection of congeners was validated by co-occurring presence of the diagnostic UV signature from the ADDA group. Full methodological details and the standard operating protocols are available from Protocols.io [115].

Anatoxin-a, dihydro-anatoxin-a and homoanatoxin-a were determined by LC-MS/MS using one quantification ion and two confirmation ions for each compound. Separation was achieved with an ACE 5 4.6 × 150 mm column (MacMod Analytical, Chadds Ford, PA, USA) assembly with solvent flow of 0.5 mL/min from a Waters Alliance 2695 solvent system (Waters, Milford, MA, USA). The solvent system was: A, 0.1% formic acid in water; B, 0.1% formic acid in acetonitrile. The separation gradient was: 0 to 20% B from 0 to 10 min, 20% to 80% B from 10 to 20 min, and 80% to 100% B from 20 to 23 min, followed by equilibration back to 0% B for 7 min. Toxins were identified using a Waters Acquity TQD mass spectrometer (Waters, Milford, MA, USA) operated in positive mode with capillary voltage 3.5 kV, desolvation and cone gasses at 30 and 800 Lh−1, respectively, desolvation and source temperatures of 400 and 150 ◦C, respectively. Retention times and fragmentation patterns were determined using anatoxin-a (BioMOL International, Farmingdale, NY, USA), homoanatoxin-a isolated from natural sources and α and β dihydroanatoxin synthesized by catalytic hydrogenation/reduction of anatoxin-a [116]. Calibration was performed with anatoxin-a; dihydro-anatoxin-a and homoanatoxin-a concentrations were estimated using the anatoxin-a standard curve. A phenylalanine standard was run with each set to confirm the baseline resolution between anatoxin-a and phenylalanine. Multiple reaction monitoring quantitation transitions were: anatoxin-a (166.09 > 131.00, collision energy (CE) 15 eV), dihydro-anatoxin-a (168.20 > 43.10, CE 23 eV), homoanatoxin-a (180.10 > 163.10, CE 15 eV). Confirmation transitions were: anatoxin-a (166.09 > 148.90, CE 15 eV; 166.09 > 90.90, CE 17 eV), dihydro-anatoxin-a (168.20 > 55.90, CE 22 eV; 168.20 > 67.00, CE 26 eV), homoanatoxin-a (180.10 > 145.10, CE 15 eV; 180.10 > 105.00, CE 17 eV).

### *5.5. Data Transformation and Analysis*

To remove biomass effects on toxin to better measure nutrient effects on microcystin and anatoxin production, microcystin and anatoxin concentrations are normalized to biomass as proxied by chlorophyll *a*. Microcystin:chl *a* and anatoxin:chl *a* ratios are calculated using Equation (1):

$$\text{toxin}: \text{chl } a \text{ ratio } \left(\text{\(\mu g\)}\\\text{crosystem} \text{o} \text{ oznato}\\\text{in } \text{\(\mu g\)}\\\text{dol} \text{o} \text{syl} \text{h} \text{o}^{-1}\right) = \frac{[\text{toxin}]}{[\text{bi} \text{o} \text{mass}]} \qquad \text{(1)}$$

where [toxin] is the concentration of either microcystin or anatoxin (in μg L−1) and [biomass] is the concentration of chlorophyll *a* (in μg L<sup>−</sup>1).

For comparison between dilution treatments, we calculated production rates from the chlorophyll *a* and biomass-normalized microcystin and biomass-normalized anatoxin concentrations. Production rate (d−1) is a method to ln normalize the changes in concentrations, where a production of 0.693 d−<sup>1</sup> is a doubling of the concentration per day, a production of 0.0 d−<sup>1</sup> indicates no change, and a production of −0.693 d−<sup>1</sup> represents a halving of the concentration. Production is calculated using Equation (2):

$$Production\left(d^{-1}\right) = \ln\left(\frac{\mu\_{T3}}{\mu\_{T0}}\right) \* \frac{1}{t} \tag{2}$$

where *μT*<sup>0</sup> is the average value of the measurement for the initial time point (T0), *μT*<sup>3</sup> is the average value of the measurement for the time point of 3 days (T3), and t is the time difference between the samplings (in days), which in this case is *t* = 3 days. To calculate the standard deviation for the production, propagated standard deviation is used, as calculated by Equation (3):

$$\text{Propagonal Standard Deviation} = \sqrt{\left(\frac{\sigma\_{T0}}{\mu\_{T0}}\right)^2 + \left(\frac{\sigma\_{T3}}{\mu\_{T3}}\right)^2} \tag{3}$$

where *μT*<sup>0</sup> is the average value of the measurement for T0, *ςT*<sup>0</sup> is the standard deviation for the measurement at T0, *μT*<sup>3</sup> is the average value of the measurement for T3, and *ςT*<sup>3</sup> is the standard deviation for the measurement at T3. For error bars, standard error is used, which is calculated using Equation (4):

$$Standard\ Error = \frac{\sigma}{\sqrt{n}}\tag{4}$$

where *ς* is the standard deviation for Figure 2a–d, Figure 3a–d, Figure 4a–d, Figure 5a–d, Figure 6a–d, and Figure 7a–d, *ς* is the propagated standard deviation for Figure 2e–f, Figure 3e–f, Figure 4e–f, Figure 5e–f, Figure 6e–f, and Figure 7e–f, and *n* is the number of data points. The standard errors are available in the WLE\_Barnard\_et\_al\_Toxins GitHub repository [117].

### *5.6. Statistical Analysis*

To evaluate the source of the variation between the treatments, ANOVA analyses were performed. For this experiment, two-factor ANOVA analyses were run on balanced data sets (all data *n* = 3), and n-factor ANOVA analyses were run on unbalanced data sets (one or more treatments were characterized as *n* = 1 or *n* = 2) using MATLAB ver. R2018b [118]. Both the two-factor and n-factor ANOVA analyses calculate degrees of freedom (d.f.) as the number of treatments (*n*) minus one (d.f. = *n*−1). The homogeneity of variances was tested for with Levene's Absolute test using MATLAB ver. R2018b [118]. All data and corresponding n-values are in Tables S1, S3, S4, S7, and S8.

**Supplementary Materials:** The following are available online at https://www.mdpi.com/2072-6 651/13/1/47/s1, Table S1: Chlorophyll *a* data, Table S2: Chlorophyll *a* production rates, Table S3: Microcystin data, Table S4: Biomass-normalized microcystin data, Table S5: Microcystin production rates, Table S6: Biomass-normalized microcystin production rates, Table S7: Anatoxin data, Table S8: Biomass-normalized anatoxin data, Table S9: Anatoxin production rates, Table S10: Biomassnormalized anatoxin production rates, Table S11: GPS coordinates of the Western Lake Erie sampling sites, Table S12: Compounds comprising the major ion solution. The following are available online at www.doi.org/10.5281/zenodo.4281127, Code used to produce Figures 2–7, importable data file formatted for the code, Key to the importable data file.

**Author Contributions:** Conceptualization, M.A.B. and H.W.P.; data curation, M.A.B.; formal analysis, M.A.B., G.L.B. and B.W.; funding acquisition, M.A.B., J.D.C., G.L.B., S.W.W., G.S.B., T.B.B. and T.W.D.; investigation, M.A.B., J.D.C., H.E.P., G.L.B., B.W., S.W.W., K.L.R., J.S.B., G.S.B., T.B.B., T.W.D., J.W., M.B. and H.W.P.; methodology, M.A.B., J.D.C., H.E.P., G.L.B., B.W., K.L.R., J.S.B., G.S.B., T.B.B., T.W.D., J.W., M.B. and H.W.P.; project administration, G.S.B.; supervision, J.D.C., G.L.B., S.W.W., G.S.B., T.B.B., T.W.D. and H.W.P.; writing-original draft, M.A.B.; writing-review and editing, J.D.C., H.E.P., G.L.B., B.W., S.W.W., K.L.R., J.S.B., G.S.B., T.B.B., T.W.D., J.W., M.B. and H.W.P.; all authors have read and agreed to the published version of the manuscript. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by the United States National Science Foundation (OCE 0812913, OCE 0825466, OCE 1840715, CBET 0826819, IOS 1451528 and DEB 1831096), the United States National Institutes of Health (NIEHS P01ES028939), a Grant-in-Aid of Research from Sigma Xi, The Scientific Research Society (G201903158412545) [MAB], a Kenan Graduate Student Award from the University of North Carolina at Chapel Hill Department of Marine Sciences [MAB], and the NOAA/North Carolina Sea Grant Program R/MER-43, R/MER-47 [MAB, HEP, KLR, JSB, HWP].

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** The data formatted for analysis and executable MATLAB code used to produce Figures 2–7 can be found on GitHub at www.doi.org/10.5281/zenodo.4281127 [117]. The data presented in this study are also available in table form in the accompanying supplementary material: https://www.mdpi.com/2072-6651/13/1/47/s1.

**Acknowledgments:** We thank R. Sloup, N. Hall, and B. Abare of the UNC Institute of Marine Sciences, as well as laboratory technicians and students from The Ohio State University and University of Toledo Lake Erie Center at the for their help with experimental work.

**Conflicts of Interest:** The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.

### **References**


### *Article*
