**Genomic Islands Confer Heavy Metal Resistance in** *Mucilaginibacter kameinonensis* **and** *Mucilaginibacter rubeus* **Isolated from a Gold/Copper Mine**


Received: 22 October 2018; Accepted: 19 November 2018; Published: 23 November 2018

**Abstract:** Heavy metals (HMs) are compounds that can be hazardous and impair growth of living organisms. Bacteria have evolved the capability not only to cope with heavy metals but also to detoxify polluted environments. Three heavy metal-resistant strains of *Mucilaginibacer rubeus* and one of *Mucilaginibacter kameinonensis* were isolated from the gold/copper Zijin mining site, Longyan, Fujian, China. These strains were shown to exhibit high resistance to heavy metals with minimal inhibitory concentration reaching up to 3.5 mM Cu(II), 21 mM Zn(II), 1.2 mM Cd(II), and 10.0 mM As(III). Genomes of the four strains were sequenced by Illumina. Sequence analyses revealed the presence of a high abundance of heavy metal resistance (HMR) determinants. One of the strain, *M. rubeus* P2, carried genes encoding 6 putative PIB-1-ATPase, 5 putative PIB-3-ATPase, 4 putative Zn(II)/Cd(II) PIB-4 type ATPase, and 16 putative resistance-nodulation-division (RND)-type metal transporter systems. Moreover, the four genomes contained a high abundance of genes coding for putative metal binding chaperones. Analysis of the close vicinity of these HMR determinants uncovered the presence of clusters of genes potentially associated with mobile genetic elements. These loci included genes coding for tyrosine recombinases (integrases) and subunits of mating pore (type 4 secretion system), respectively allowing integration/excision and conjugative transfer of numerous genomic islands. Further in silico analyses revealed that their genetic organization and gene products resemble the *Bacteroides* integrative and conjugative element CTnDOT. These results highlight the pivotal role of genomic islands in the acquisition and dissemination of adaptive traits, allowing for rapid adaption of bacteria and colonization of hostile environments.

**Keywords:** *Mucilaginibacer rubeus*; *Mucilaginibacter kameinonensis*; genomic island; evolution; heavy metal resistance; draft genome sequence; CTnDOT

### **1. Introduction**

Heavy metals (HMs) have a dualistic impact on living organisms. On the one hand, metal ions are essential for numerous biological processes mandatory for cellular activity, including homeostasis, enzyme activity, and protein functionality [1]. On the other hand, when present in excess in the environment, HM can have toxic effect hindering diverse cellular processes and thus cellular life.

Heavy metal pollution has been part of Earth's history as it can originate from natural processes such as volcanic eruption. Recent (over)industrialization and exploitation of Earth resources worldwide has accelerated HM release into the environment and led to high levels of water, air, and soil pollution. Especially, mine exploitation for metal extraction is one of the most important sources of heavy metal pollution [2]. This comes not only from excavating deep-buried HMs to be exposed to the surface, but also from extraction protocols that often rely on the use of other contaminants, including HMs [2].

Beyond its effects on people, HM toxicity was shown to have profound impacts on microbial communities, including fungi and bacteria [2]. Heavy metals were shown to have critical consequences on bacterial viability due to their pleiotropic effect on cellular processes. Excess of HM can disrupt the cell membrane, damage nucleic acids and proteins, impair enzymatic activities, and inhibit key processes such as transcription [1]. The presence of HM pollution exerts a high selective pressure on microbial communities, reducing their diversity, biomass, and activity, thus strongly impacting the biological activity of polluted environments [3].

In order to cope with the presence of elevated concentration of HMs, a myriad of bacterial genetic programs has been selected encoding functions that allow efflux and/or sequestration of HMs, and modification to inactivate or reduce reactivity of certain metal ions. The main mechanism to resist toxicity of HMs is efflux [1]. Important classes of HM transporters include PIB-type ATPases and cation diffusion facilitators (CDF). Both types of transporters translocate HM ions from the cytoplasm across the cytoplasmic membrane into the periplasm [4]. In the context described here with microbes having to handle very high external concentrations of HMs, P-type ATPases are much more relevant since they are much more powerful using ATP to pump HMs against their concentration gradient out of the cytoplasm [5]. In addition, HMs are translocated from the periplasm across the outer membrane into the extracellular space by resistance-nodulation-division (RND)-type transport systems. These multicomponent transporters of the RND type contain 3 RND transport proteins, 6 membrane fusion proteins (MFPs), and 3 outer membrane factor (OMF) proteins. The fascinating transport mechanism of the RND-type transport complex has been described in detail [4]. PIB-type ATPases and RND-type transport systems were described as being the most important systems to confer a high HM resistance (HMR).

Bacteria also show an astonishing capability to spread HMR genes within bacterial communities via horizontal gene transfer. Dissemination of genetic material conferring HMR is frequently associated with conjugative plasmids, genomic islands, and transposons [6]. Conjugative plasmids are extrachromosomal replicative entities able to transfer from a donor cell toward a recipient cell by conjugation [7]. Conjugative plasmids have been recognized as major contributors for the spread of adaptive traits such as antibiotic resistance, new metabolic capacities, and HMR [8]. Conjugative plasmid-borne HMR is associated with occurrence of large clusters of HMR genes that can span over several kb [9–15]. Portions of genomic DNA called genomic islands (GIs) were also shown to play a pivotal role into the horizontal dissemination of genetic material [16]. Although the mechanisms underlying the mobility of some GIs remain obscure, current knowledge describes different strategies that ultimately rely on conjugative transfer [17,18]. GI-associated HMR was described in *Enterobacteriaceae* and *Shewanellaceae* [19], *Listeria monocytogenes* [20], and *Acinetobacter baumannii* [21]. Also, HMR was shown to be conferred by an IncC-dependent mobilizable genomic island SGI1 variant called SGI1-K in *Salmonella enterica* [22–24]. Transposons are genetic entities able to move intra-molecularly (on the same replicon) or inter-molecularly (between different replicons) [25]. Most transposons can hitchhike by integrating into a conjugative plasmid or a GI for intercellular mobility. Transposons conferring HMR were described to be in association with other mobile genetic elements [13,22].

In this study, we describe the isolation and characterization of four heavy metal-resistant *Mucilaginibacter* strains isolated from a gold/copper mine in China. Genomes of these strains were sequenced and further in silico analysis revealed a high number of heavy metal resistance determinants. Moreover, at least part of these HMR gene clusters were shown to be potentially mobile as they are in the close vicinity of the core region of putative integrative and conjugative elements (ICEs).

### **2. Materials and Methods**

### *2.1. Bacterial Isolation*

Strains *Mucilaginibacter rubeus* P1, P2, and P3 were isolated from samples collected at 5–10 cm below the surface of a soil located near a waste water treatment dam of a copper-gold mine, and *Mucilaginibacter kameinonensis* P4, was isolated from a hillside with little human activity within the gold and copper mine (Zijin mining) in Longyan city of Fujian province, China (Table 1). After serial dilutions with 0.85% NaCl, the soil sample was spread on R2A (DSM medium 830) agar plates containing 2 mM CuSO4·5H2O. After incubation at 28 ◦C for 1 week, the strains were isolated and later stored at −80 ◦C in 20% glycerol (*w*/*v*).

**Table 1.** Characteristics of the heavy metal (HM)-contaminated soil from where the strains were isolated.


### *2.2. Taxonomic Analysis*

Strains were incubated at 28 ◦C for 24 h on R2A agar plates. As described in Brosius et al. [26], the universal primer pair 27F/1492R was used to amplify 16S sequences and the amplified PCR product was subsequently sequenced [26]. PCR products were sequenced by Biosune Company (Fuzhou, China) using the Sanger method. Based on the EzTaxon database (http://eztaxon-e.ezbiocloud.net) [27], pairwise sequence similarity and phylogenetic neighbors of the sequences of each individual strain (1382–1432 bp) were obtained through BLAST searches. In total, 19 *Mucilaginibacter* strains with publicly available 16S ribosomal RNA (rRNA) gene sequences were selected, with *Pedobacter africanus* DSM 12126T (AJ438171) as an out-group, to do the alignment via Mega 7.0 software [28]. A Neighbor-joining (NJ) tree was generated and the Kimura's two-parameter model was used to calculate evolutionary distances [29], and bootstrap analysis with 1000 replications was conducted to obtain confidence levels of the branches [30].

### *2.3. Determination of the Minimal Inhibitory Concentration*

To determine the level of resistance to various metals of all strains, *M. rubeus* P1, P2, and P3 and *M. kameinonensis* P4 were grown on Cu, As, Cd, and Zn agar plates containing different Cu(II), Zn(II), As(III), and Cd(II) concentrations to determine the minimal inhibitory concentration (MIC). The different R2A plates contained 0–10.0 mM of copper or arsenic, with 0.5 mM increments, 0–30.0 mM with 1.5 mM increments in case of zinc, and 0–2.0 mM cadmium with the increments being 0.2 mM. 1M CuSO4·5H2O, ZnCl2, NaAsO2, and CdCl2·5H2O stock solutions were prepared and stored after filtration through a 0.22 μm filter.

### *2.4. Cell Morphology and Flagella Observation*

Overnight cultures of strains *M. rubeus* P1, P2, and P3 and *M. kameinonensis* P4 were inoculated into 50 mL of R2A medium at 28 ◦C with 180 rpm shaking. After 24 h of growth with shaking, cells were centrifuged (1000× *g*, 10 min, 4 ◦C) and observed under scanning electron microscopy (SEM). Cells were harvested and washed three times with cold (4 ◦C) phosphate buffered saline (0.2 M PBS, pH 7.2). Fixation was performed with 2.5% glutaraldehyde (24 h, 4 ◦C). Fixed cells were dehydrated through a series of alcohol dehydration steps (30%, 50%, 70%, 85%, 95%, and 100%) and finally freeze dried and sputter coated. The samples were then viewed using a scanning electron microscope JSM-6390 SEM (JEOL, Tokyo, Japan).

### *2.5. Growth Conditions Optimization*

To optimize NaCl concentration and pH of the medium for growth of the *Mucilaginibacter* strains, 50 μL precultures were added to 5 mL R2A liquid medium supplemented with 0–3% NaCl at pH 7, or to R2A without any NaCl and with pH set to the range between pH 2–11. Cultures were incubated at 28 ◦C for 7 days, after which culture turbidities optical density (OD) at 600nm were evaluated. Anaerobic growth was tested by incubating R2A plates in an anaerobic chamber at 28 ◦C for 1 week. Optimal growth temperature was tested in the incubator on R2A agar plates at temperatures between 4 to 40 ◦C for 1 week.

### *2.6. Genomic DNA Extraction*

Genomic DNA (gDNA) was extracted by using a TIANamp Bacteria DNA Kit (Tiangen Biotech, Beijing, China) from cultures grown on R2A. The quantity and purity of gDNA were assessed using an UV spectrophotometry (Nanodrop ND-1000, J & H Technology Co., Ltd. Wilmington, USA). Genomic DNA with OD260/280 value higher than 1.80 was selected and examined on agarose gel electrophoresis (0.8%). Samples containing more than 25 μg of intact gDNA (fragment size > 20 kb) were sent out for whole-genome sequencing.

### *2.7. Whole-Genome Sequencing*

Whole-genome shotgun sequencing was preformed using an Illumina HiSeq X Ten System provided by Vazyme Biotech Co., Ltd. (Nanjing, China). The DNA library was constructed using the Illumina V3 VAHTS Universal DNA Library Prep Kit according to the VAHTS Universal DNA sample preparation protocol (Illumina, Santiago, USA). The insert size was 300 bp for all strains, and 16,980,768, 18,531,104, 18,306,636, and 20,005,292 read-pairs and 2.86, 3.12, 3.09, and 3.37 Gb of raw data were obtained for strains *M. kameinonensis* P4 and *M. rubeus* P1, P2, and P3, respectively.

### *2.8. De novo Genome Assembly and Annotation*

Illumina reads were quality-filtered, trimmed, and de novo assembled with default settings using CLC Genomic Workbench 11.0 (QIAGEN, Hilden, Germany). The draft genome sequences were annotated by NCBI PGAP, and are accessible under GenBank numbers QEYR0000000, QFKW0000000, QFKV0000000, and QFKU0000000 for *M. kameinonensis* P4 and *M. rubeus* P1, P2, and P3, respectively. *M. kameinonensis* P4 generated 78 contigs with an n50 value of 350.607 bp. *M rubeus* P1 generated 158 contigs with an n50 value of 139.339 bp. *M rubeus* P2 generated 118 contigs with an n50 value of 132.524 bp. *M rubeus* P3 generated 107 contigs with an n50 value of 148.541 bp.

### *2.9. TraG Proteins Phylogenetic Analyses*

Molecular phylogenetic analysis of TraG proteins was performed using MEGA6 [28]. The 807- to 850-amino acid sequences of TraG proteins were recovered from genome sequences of *Mucilaginibacter* isolated in this study. The corresponding sequence in CTnDOT (TraGDOT accession number: AAG17832.1) was added to the dataset as an outgroup. Analyses were computed using an amino acid

alignment generated by MUSCLE [31]. The evolutionary history was inferred by using the Maximum Likelihood method based on the Jone, Taylor and Thornton (JTT) matrix-based model [32]. Initial tree(s) for the heuristic search were obtained by applying the NJ method to a matrix of pairwise distances estimated using a JTT model. A discrete Gamma distribution was used to model evolutionary rate differences among sites (five categories (+G, parameter = 2.9848)). The analysis involved 18 amino acid sequences. All positions with less than 95% site coverage were eliminated, providing a total of 716 positions in the final dataset.

### **3. Results and Discussion**

### *3.1. Isolation of Four Heavy Metal-Resistant Mucilaginibacter*

We intended to isolate heavy metal resistant strains from the ZiJin copper-gold mine to gain insights into how bacterial strains adapt to high concentrations of HMs. We recovered four strains that were morphologically similar with a high tolerance to a number of HMs.

Based on phylogenetic analysis (NJ) of the 16S rRNA gene three strains (P1, P2, and P3) were closely related to *M. rubeus* EF23<sup>T</sup> (98.34–99.93%) and *M. gossypiicola* Gh-67<sup>T</sup> (98.12–99.01 %). The fourth strain (P4) grouped closely with *M. kameinonensis* SCK<sup>T</sup> (98.8 %) (Figure 1). All strains belonged to the *Sphingobacteriaceae* family in the class *Sphingobacteriia*.

**Figure 1.** Neighbour-joining phylogenetic tree constructed based on the 16S ribosomal RNA (rRNA) gene sequences from the draft genome sequence showing the phylogenetic relationships between strains *Mucilaginibacter rubeus* P1, P2 and P3 and *Mucilaginibacter kameinonensis* P4 and other species in the genus *Mucilaginibacter*. Values indicate percentages of identical branching in 1000 bootstrappings. The sequence of *Pedobacter koreensis* WPCB189T was used as an out-group. Bar, 0.01 substitutions per nucleotide position.

### *3.2. Phenotypic Characterization of Mucilaginibacter Strains Uncovered Multiple Heavy Metal Resistances*

The HM concentration of the soil is extremely high, even the lowest concentration of total Zn, As, Cd, and Cu was found to be 49.27, 1.43, 1.19, and 18.37 mg·kg−1, respectively. The MICs of the four strains reached up to 3.5 mM Cu(II), 21 mM Zn(II), 1.2 mM Cd(II), and 10.0 mM As(III). Strain *M. kameinonensis* P4 displayed higher Cd resistance compared to strains *M. rubeus* P1, P2, and P3 (Table 2). Related, not heavy metal resistant *Mucilaginibacter pedocola* sp. TBZ30T, cultured under similar conditions displayed MICs of 0.4 mM Cu(II), 3 mM Zn(II), 0.2 mM Cd(II), and 0.2 mM As(III) [33]. Such high resistance to multiple HMs as reported here has therefore not been observed before in the genus *Mucilaginibacter* [33].


**Table 2.** Minimal inhibitory concentration (MIC) of strains to Zn(II), As(III), Cd(II), and Cu(II) and respective concentrations of the HM in the soil where the strains were isolated from.

Note. ND means not found.

Strains *M. rubeus* P1, P2, and P3 and *M. kameinonensis* P4 formed a light orange or pink, moist, circular, and convex colony with smooth margins on R2A agar plates. All strains were Gram-negative and aerobic. Growth of strains was observed at 4–30 ◦C (optimum, 28 ◦C). Optimal growth occurred in absence of further NaCl, but the strains could still grow in R2A with up to 1.5% NaCl added. These characteristics are consistent with description of the genus *Mucilaginibacter* [34–36]. Medium pH for optimal growth (~pH 5.0) and pH tolerance (pH 5.0–9.0) varied slightly between the four strains (Table 3).

**Table 3.** General features of strains *M. rubeus* P1, P2, and P3 and *M. kameinonensis* P4.


*3.3. Mucilaginobacter Strains Exhibit an Arsenal of Genetic Determinants to Deal with High Concentrations of Heavy Metals*

To gain insight in the genetic basis of how the four strains were able to deal with these high HM concentrations, we determined draft genome sequences. Draft genomes were automatically annotated through RAST (Rapid Annotation using Subsystem Technology) database (http://rast.nmpdr.org/). Based on inferred protein homologies, between 6 and 16 putative P1B type-ATPase [37] were encoded in the four genomes (Table 4). All strains further encoded a variety of putative RND type metal transporter systems of the CzcCBAD type. Three strains further encoded putative CusCBA Cu(I) translocating RND-type transport systems, except *M. kameinonensis* P4 (Table 3). Multiple genes for putative multicopper oxidases were found on the different genomes, which may constitute the basis for the observed copper resistance (Table 3). Genes for putative multicopper oxidases were only taken into account if they were located adjacent to genes encoding P1B type Cu(I) translocating P-type ATPase. Finally, between 2 and 4 putative *ars* operons (*arsNCR, acr3, arsMCR*) were among the *Mucilaginibacter*

genomes (Table 3). The higher number of *ars* operons in strain P1 and P2 genomes correlated to their high MICs on As(III) (10.0 and 9.0 mM, respectively).

The number of HMR determinants in *Mucilaginibacter* genomes was unusually high, even in comparison to the well-known HM-resistant strain *Cupriavidus metallidurans* CH34 [5,38,39] (Table 4), suggesting a strong selection for HM resistance in their natural living environment. The HM resistance determinants are often clustered together, and often located adjacent to *tra* genes. They could be identified on many different contigs.

**Table 4.** Heavy metal related genes in the analyzed *Mucilaginibacter* strains in comparison to *Cupriavidus metallidurans* CH34.


RND: Resistance-nodulation-division.

### *3.4. Heavy Metal Resistance Is Associated with CTnDOT-Related Genomics Islands*

Tolerance to HMs is frequently acquired by horizontal transmission among and between bacterial populations. Given the important size of HMR clusters identified in *Mucilaginibacter* genomes (up to 150 kb), we wondered whether some might be encompassed by GIs (Figure 2). We examined the close vicinity of the HMR clusters for the hallmark of conjugative systems, also known as Type 4 secretion systems (T4SSs) [40]. T4SSs have been classified based on their VirB4 protein, a ubiquitous constituent of conjugative systems [41]. A total of 17 genes encoding VirB4 proteins (*traG*) exhibiting sizes between 807 to 850 amino acids were identified on the 4 genomes, most of them in the close proximity of HMR clusters: 6 in *M. rubeus* P1 (contigs 1, 4, 24, 29, 42, 55), 6 in *M. rubeus* P2 (contigs 24, 26, 29, 32, 34, 42), 2 in *M. rubeus* P3 (contigs 5 and 11), and 3 in *M. kameinonensis* P4 (contigs 6, 9, and 35).

TraG proteins of the putative conjugative GIs identified in *Mucilaginibacter* genomes were compared to the 838-amino acid TraG of CTnDOT (TraGDOT) and showed 32 to 53% of identity over 94 to 98% of their amino acid sequence. The evolutionary history of TraG proteins was inferred using TraGDOT as an outgroup (Figure 3). Two strongly supported clades were delineated, suggesting that they belong to two distinct lineages (Figure 3, green and red boxes). As expected, each one of the TraG proteins of strain *M. rubeus* P1 grouped with one TraG protein from strain *M. rubeus* P2, confirming that these identical strains contain the same 6 elements. More interestingly, TraGP1-1, TraGP2-26, TraGP3-5, and TraGP4-9 grouped together, and their gene sequences were identical.

Closer analysis revealed the presence of genes coding for other T4SS subunits adjacent to each one of the *traG* genes. This grouping of *tra* genes may be regarded as a conjugation module, i.e., genes and sequences implicated in the same biological process [42]. In particular the regions including *traGP1-1, traGP2-26, traGP3-5*, and *traGP4-9* were 100% identical across a circa 150-kb region, including an about 75-kb cluster coding for multiple HMR. The organization of these putative conjugation modules resembles the one encoded in CTnDOT, a protypical ICE of *Bacteroides* [41,43]. As in CTnDOT, putative conjugative modules encoded by GIs of *Mucilaginibacter* thus belong to the mating pair formation (MPF) category B (MPFB) [41]. Also, most of the putative conjugative modules identified in *Mucilaginibacter* strains are in the proximity of genes encoding a putative tyrosine recombinase related to IntDOT, the integrase of CTnDOT [44]. Moreover, the *Mucilaginibacter* GIs carried a gene predicted to encode a RteC-like protein reminiscent of the CTnDOT regulation system [45,46]. *Mucilaginibacter*

*Genes* **2018**, *9*, 573

GIs are thus likely to be ICEs, whose maintenance relies on integration into the chromosome and dissemination depends on its excision from the chromosome as a circular element that would transfer by conjugation [17,47,48]. The presence of at least one identical contiguous region over 150 kb (represented by the *traGP1-1* gene) in the four different *Mucilaginibacter* recovered strains suggest active mobility and recent transfer of this GI.

proteins and unknown functions are highlighted

 in gray.

**Figure 3.** Molecular phylogenetic analysis of TraG proteins of putative conjugative genomic islands (GIs) of *Mucilaginibacter*. The evolutionary history was inferred by using the Maximum Likelihood method based on the JTT matrix-based model [32]. The percentage of trees in which the associated taxa clustered together is shown next to the branches. The tree is drawn to scale, with branch lengths measured in the number of substitutions per site. Evolutionary analyses were conducted in MEGA6 [28]. Initial alignment of sequences was performed using Muscle for the presented tree. An identical tree with minor changes in bootstrap values was obtained using ClustalW for alignment. The VirB4 subunit of MPFB T4SS is named TraG [41]. For convenience and consistence in TraG protein identification, nomenclature is as follows: TraGPX-Y, where X is the strain number and Y the contig carrying the gene coding for TraG. TraG CTnDOT (TraGDOT) accession number: AAG17832.1.

Since the draft genomes were not completely curated to a single contiguous scaffold, we could not confidently delimit the boundaries of the putative conjugative GIs. As a matter of fact, a single GI might be spread over multiple contigs, or could be a defective element lacking flanking or internal parts of the original GI. Also, IntDOT was reported to not require strict homology between the recombining sites in contrast to the majority of tyrosine recombinases [46,49]. The integration/excision is, in that case, site-selective rather than site-specific, strongly impairing the precise identification of the right and left attachment sites (*attR* and *attL*, respectively).

### **4. Concluding Remark**

This work allowed the isolation and characterization of four heavy metal-resistant *Mucilaginibacter* strains recovered from polluted soil of gold mines. Sequencing of genomic content allowed inspection of HMR loci into the chromosome of these strains and their close association with loci coding for conjugation of CTnDOT-related GIs. Further genome closure and experimental investigation should allow testing the functionality of such putative ICEs found in the *Mucilaginibacter* strains. Notably, phenotypes such as their ability to excise from the chromosome and their capability to transfer toward a new host by conjugation will be monitored. In particular the 150-kb (at least) conserved putative ICE present in the four strains is an interesting candidate, likely to be functional given its complete identity among all four strains. Protein BLAST using TraG of this conserved element and search

*Mucilaginibacter* genomes did not give other perfect hits, suggesting that presence of this GI is restricted to sampling locations of this study (Table 1). Increasing availability of fully sequenced genomes should allow further data meaning in order to evaluate the abundance of specific GIs in *Mucilaginibacter* and predict their functionality.

The presence of putative CTnDOT-related ICEs into genomes of *Mucilaginibacter* strains does not seem uncommon. Protein BLAST analysis using TraGDOT as query and searching *Mucilaginibacter* genomes gave multiple hits with identity ranging from 70% over 99% of aa sequence, down to 30% over 79% of aa sequence (considering 96 hits with more than 75% of coverage). This observation highlights the presence of multiple putative CTnDOT-related GIs populating *Mucilaginibacter* genomes, most likely playing a key role in their genome evolution and adaptation. One can speculate that such GIs may be involved in conferring specific capabilities such as the ability to degrade pectin, xylan, and laminarin of *M. paludis* and *M. gracilis* [50] plant growth promotion capacity conferred by *M. gossypii* and *M. gossypiicola* [51], or yet to be discovered adaptive functions that may be conferred by such GIs. Further *in silico* analyses may reveal interesting features of such putative ICEs considering accessory functions they could confer, or functionality of their recombination, conjugation, or regulatory systems.

This exploratory work on HMR GIs of *Mucilaginibacter* together with other research done on *Mucilaginibacter* species so far constitutes a solid ground for future experimental research aiming at developing molecular tools. Such tools would greatly facilitate further investigation of *M. rubeus* and *M. kameinonensis* biology and could likely be extended to other *Mucilaginibacter* species.

**Author Contributions:** Y.P.L. isolated strains from mine and generated the phylogenetic tree of 16SrDNA, annotated the genome, mapped the gene cluster, and partially drafted the manuscript. N.C. contributed to the general concept of the study, performed gene sequence prediction, molecular phylogenetic analysis of TraG proteins, and partially drafted the manuscript. N.Y. contributed to the MICs determination and partially drafted the manuscript. B.L. performed growth conditions optimization and genomic DNA extraction. X.X. assembled the genomes. Q.S. and H.A.Al-W. contributed to data analysis. R.F. contributed to the general concept of the study and manuscript verification. J.R.v.d.M. contributed to the general concept of the study and manuscript verification. C.R. contributed to the general concept of the study, design of experiments and manuscript preparation.

**Funding:** This research was funded by National Natural Science Foundation of China (NSFC) grant number 31770123, State administration of Foreign Expert Project numbers GDT20173600005 and 110000214620170006, and Swiss National Science Foundation (SNSF) grant number 31003A\_175638, Twasol Research Excellence Program (TRE), King Saud University, Saudi Arabia for support.

**Acknowledgments:** We like to thank engineers Huaiguo Huang, Hongwen Li, and Xianzheng Chen of Zijin Mining Group Co., Ltd. for the technical support in the sampling.

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

### **References**


© 2018 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 (http://creativecommons.org/licenses/by/4.0/).

### *Review* **Heavy Metal Resistance Determinants of the Foodborne Pathogen** *Listeria monocytogenes*

**Cameron Parsons 1,\*,†, Sangmi Lee 2,† and Sophia Kathariou <sup>1</sup>**


Received: 16 November 2016; Accepted: 18 December 2018; Published: 24 December 2018

**Abstract:** *Listeria monocytogenes* is ubiquitous in the environment and causes the disease listeriosis. Metal homeostasis is one of the key processes utilized by *L. monocytogenes* in its role as either a saprophyte or pathogen. In the environment, as well as within an animal host, *L. monocytogenes* needs to both acquire essential metals and mitigate toxic levels of metals. While the mechanisms associated with acquisition and detoxification of essential metals such as copper, iron, and zinc have been extensively studied and recently reviewed, a review of the mechanisms associated with non-essential heavy metals such as arsenic and cadmium is lacking. Resistance to both cadmium and arsenic is frequently encountered in *L. monocytogenes*, including isolates from human listeriosis. In addition, a growing body of work indicates the association of these determinants with other cellular functions such as virulence, suggesting the importance of further study in this area.

**Keywords:** *Listeria monocytogenes*; heavy metal resistance; mobile genetic element; cadmium; arsenic

### **1. Introduction**

*Listeria monocytogenes* is a Gram-positive facultative intracellular pathogen and the causative agent of the disease listeriosis. In healthy individuals, listeriosis can manifest as gastroenteritis; however, in at-risk individuals such as the elderly, pregnant women, or immunocompromised patients, listeriosis can result in severe symptoms, including septicemia, meningitis, stillbirths and even death [1–3]. Listeriosis is responsible for approximately 1455 hospitalizations and 255 deaths in the United States annually [4]. *L. monocytogenes* is found ubiquitously in the environment, is capable of growing in the cold, and can persistently colonize food production facilities [2,5]. This, along with the severe outcomes and life-threatening potential of listeriosis, makes *L. monocytogenes* a major cause for food safety and public health concern.

*L. monocytogenes* is well-adapted to survive both in the environment as well as within the body of humans and other animals [6,7]. One of the key adaptations for these dual survival modalities is metal homeostasis. Certain metals such as copper, zinc, and iron are required for essential cellular functions but become toxic at higher concentrations. In contrast, metals such as arsenic and cadmium appear to serve no cellular function and are considered toxic at any concentration [8]. In the environment, metals are typically found at low levels, but their concentrations can increase due to various anthropogenic interventions, including industrial pollution or agricultural practices [9,10]. In an animal host, metal concentrations are dependent on various factors, such as diet and tissue type [11–13]. The immune system can utilize metals in response to pathogens, either by restricting metal availability or by accumulating metals, to exert toxic effects on pathogens in the course of an infection [14–16]. For these reasons, the ability to import or export metals as needed is essential for *L. monocytogenes* to survive in its diverse environmental niches. Here we update the information currently available for functions related to the essential metals copper, iron, and zinc in *L. monocytogenes*, while providing the first comprehensive review of the widely-distributed resistances to toxic heavy metals, specifically cadmium and arsenic.

### **2. Essential Yet Potentially Toxic Metals**

Metals such as copper, iron, and zinc are cofactors for essential enzymes, and insufficient amounts of these metals can result in cellular death [8]. However, at excessive concentrations these metals become toxic to the cells, disrupting membrane potential, interfering with enzyme function, and creating reactive oxygen species [8,17,18]. *L. monocytogenes* has several determinants to acquire these metals at highly regulated levels, and to expel, sequester, or convert and detoxify these metals when they are in excess [8,18–20]. Both conditions can occur in the animal host. Substantial work has been done to elucidate these processes for essential metals in *L. monocytogenes*, culminating in several reviews [8,18,19]. In relation to iron, recent findings have clarified the role of FrvA, which is implicated in haem toxicity and pathogenicity and is a high-affinity Fe(II)-exporting P-type ATPase with specificity for elemental iron [21,22]. Additionally, a recent study by Yousuf, Ahire, and Dicks elucidated the likely mechanism underlying copper toxicity in *L. monocytogenes* in which copper disrupts the cell membrane through lipid peroxidation and protein oxidation, as shown in other organisms as well [23–25]. Additionally of note from this study was the finding that *L. monocytogenes* had the most pronounced resistance to copper of all the Gram-positive organisms tested (*L. monocytogenes*, *Streptococcus* spp., *Enterococcus* spp. and *Bacillus cereus*), which was considered by the authors to be worthy of further investigation [25]. Recent work identified a dual role for the penicillin-binding protein encoded by *pbp4* (*lmo 2229* homolog), both in tolerance of *L. monocytogenes* to β-lactam antibiotics and in copper homeostasis [26].

### **3. Cadmium and Arsenic: Non-Essential Toxic Metals**

In contrast to multiple reviews of *L. monocytogenes* determinants mediating homeostasis for essential metals, no comprehensive reviews are available on this pathogen's resistance to non-essential toxic metals, such as arsenic and cadmium, even though resistance to these agents has been one of the earliest-documented phenotypes of *L. monocytogenes* [27,28]. Such resistance was encountered frequently enough to be utilized as a subtyping tool before the advent of higher-resolution techniques such as ribotyping, pulsed-field gel electrophoresis and multilocus sequencing, and was often associated with epidemic-associated clones [29,30]. Determinants mediating resistance to these heavy metals are widely distributed within *L. monocytogenes*, both on the chromosome and on plasmids [31–36]; all such determinants described here are summarized in Table 1.


**Table 1.** Heavy metal resistance-associated determinants in *Listeria monocytogenes*.

<sup>1</sup> Cadmium resistance determinants can exhibit sequence divergence sufficient to be considered as different alleles of *cadA* and *cadC*. As discussed in this review, these have been designated with numbers, e.g., *cadA1 cadC1*, *cadA2 cadC2*, etc., based on the order in which they were identified or characterized.

### **4. Arsenic Resistance**

Arsenic resistance has been primarily associated with serotype 4b, which is over-represented among clinical isolates in comparison to those from foods and food processing environments [29,39,40]. Further studies on the distribution of arsenic-resistant isolates within serotype 4b revealed that

arsenic resistance was most frequently encountered among clones associated with outbreaks [30,36,39]. In'particular, clonal complex (CC) 2 (formerly epidemic clone (EC) Ia) displayed the highest prevalence of arsenic-resistant isolates, and CC1 (previously ECI) was the second highest in the percentage of arsenic-resistant isolates, with resistance also encountered in several other clones, including the hypervirulent serotype 4b clone CC4 and isolates of CC315 and CC9 [30,36,39]. Interestingly, however, no CC6 (former ECII) isolates were found to be resistant to arsenic [30,36,39].

Albeit infrequent, arsenic resistance can also be found in serotypes 1/2a, 1/2b and 1/2c [39]. In serotype 1/2a, for instance, approximately 2% of the isolates tested were resistant to arsenic [39]. Interestingly, non-pathogenic *Listeria* species seem to largely lack arsenic resistance, with one of the exceptions being the reference strain *Listeria innocua* CLIP 11262, which harbors arsenic resistance genes on its plasmid pLI100 [31]. These findings suggest that arsenic resistance is primarily encountered in *L. monocytogenes*, especially in serotype 4b.

Analysis of strains that were persistently isolated from a rabbit meat processing facility in Italy revealed that approximately 90% of the isolates of clone CC14 (serotype 1/2a) exhibited arsenic resistance [41]. The extent to which arsenic resistance may contribute to persistence of this or other *L. monocytogenes* clones in food-processing facilities remains to be elucidated.

### **5. Arsenic Resistance Determinants**

Typically, arsenic resistance cassettes are comprised of three (*arsRBC*) to five (*arsRDABC*) genes that are transcribed into a single polycistronic mRNA [37,42,43]. The genes *arsA* and *arsB* encode an ATPase and a membrane transporter, respectively, which form an ATP-dependent anion pump that exports arsenite from the cells [37,44]. The *arsA* gene product can function independently as a passive transporter of arsenite [37,42]. The *arsC* gene encodes a reductase that performs the conversion of arsenate to arsenite, which is then extruded by ArsA or the ArsA/ArsB complex [37,42,45,46]. Thus, deletion of *arsC* impairs resistance to arsenate but does not influence resistance to arsenite [45]. The *arsA*, *arsB*, and *arsC* genes are regulated by two regulatory proteins encoded by *arsR* and *arsD* [37,42,45,47]. The *arsR* gene product is a repressor that binds to the operator of the *ars* cassette in the absence of the inducer (arsenate and arsenite) but dissociates from the operator upon interaction with the inducer [37,45]. In other words, ArsR determines the basal expression level of the arsenic resistance cassette [37,45]. Meanwhile, the *arsD* gene product is not affected by the inducer and controls the maximal level of the *ars* operon, preventing the deleterious effects of *arsB* overexpression, such as hypersensitivity to arsenite [37,42,47].

Whole genome sequencing of *L. monocytogenes* has revealed three operons putatively associated with arsenic detoxification in *Listeria* spp. [31,34,35]. The first putative arsenic resistance operon (*arsR1D2R2A2B1B2*) was reported on plasmid pLI100, which is harbored in *L. innocua* CLIP 11262 [31,34]. As indicated above, no plasmid-borne arsenic resistance determinants have been reported in *L. monocytogenes*, which is consistent with earlier findings that arsenic resistance is chromosomally mediated in this species [29]. The other two putative arsenic resistance determinants are both located on the chromosome and were identified only in *L. monocytogenes*, each with a tendency to contribute to arsenic resistance in different serotypes. The first consists of the arsenic resistance cassette identified in pLI100 (*arsR1D2R2A2B1B2*) and two additional upstream genes, *arsD1* and *arsA1* [48]. The second consists of the arsenic resistance cassette *arsCBADR* harbored on a Tn*554*-like element [35].

The *arsR1D2R2A2B1B2* cassette, together with the upstream genes *arsD1* and *arsA1*, were initially identified on a 35-kb chromosomal island, termed *Listeria* genomic island 2 (LGI2), harbored by the CC2 strain Scott A [48]. Downstream of the arsenic resistance cassette, LGI2 also harbored the novel cadmium resistance determinant *cadA4* [35,36,48,49]. Therefore, all tested serotype 4b arsenic-resistant isolates were also resistant to cadmium [30,36].

Further studies using the LGI2 genes as genetic markers and whole genome sequence analysis showed that LGI2 was present in all tested arsenic-resistant isolates of serotype 4b, including isolates belonging to clones CC1 and CC2, and the hypervirulent clone CC4 [36,39,50]. Interestingly, the entire island was markedly diversified in a majority of arsenic-resistant CC1 strains; this diversified derivative was termed LGI2-1 and was inserted at the same chromosomal locus in all CC1 isolates that harbored it [36,39,50,51]. These findings suggest that arsenic resistance of serotype 4b can be attributable to arsenic resistance genes harbored on LGI2. Furthermore, regardless of the diversification, all serotype 4b arsenic-resistant isolates harboring LGI2 displayed tolerance to higher concentrations of arsenic (arsenite minimum inhibitory concentration (MIC) of 1.250 to 2.500 μg/mL) compared with susceptible strains (arsenite MIC of 250 to 500 μg/mL) [36,39]. However, direct experimental evidence is still warranted to explicitly demonstrate the involvement of arsenic resistance genes on LGI2 in arsenic detoxification.

LGI2 genes were rarely encountered among arsenic-resistant isolates that belong to serotypes other than 4b [39]. Even if strains were positive for LGI2 genes, both PCR typing and whole genome sequence analysis suggested sequence divergence from either LGI2 or LGI2-1, except in the case of a serotype 1/2a belonging to CC14, which harbored LGI2 that was highly conserved with that in the serotype 4b strain Scott A [39,41]. As mentioned earlier, CC14 isolates harboring this highly-conserved LGI2 were also found to be persistent in a rabbit meat processing plant [40].

While the genetic content of LGI2 is highly conserved in *L. monocytogenes*, genome analysis of arsenic-resistant isolates harboring LGI2 revealed this island to be inserted in at least eight different locations, primarily within open reading frames [39]. The GC content of LGI2 is lower than average (34% versus the *L. monocytogenes* average of 38%), and LGI2 also harbors a putative phage integrase gene. Such findings suggest that LGI2 was acquired via horizontal gene transfer from other bacterial genomes [36,39]. However, likely donors for LGI2 or LGI2-1 in *L. monocytogenes* remain unidentified.

The chromosomally-encoded arsenic resistance cassette (*arsCBADR*) that is associated with a Tn*554*-like transposon was first identified via whole genome sequencing of the serotype 1/2c strain SLCC 2372 [35]. When *arsA* associated with this Tn*554*-like element was used as a genetic marker, it was exclusively found among arsenic-resistant isolates of serotypes other than 4b, and approximately 90% of these isolates were negative for any LGI2-associated arsenic resistance genes, while positive for *arsA* harbored on the Tn*554*-like transposon [39]. These observations suggest that, in contrast to LGI2 which is predominantly found among serotype 4b isolates, the arsenic resistance cassette harbored on the Tn*554*-like transposon is responsible for arsenic resistance of strains of *L. monocytogenes* of other serotypes. Further experimental and *in-silico* evidence can deepen our understanding of the evolution and function of arsenic resistance associated with the Tn*554*-like element.

### **6. Cadmium Resistance**

Several studies examined prevalence of cadmium resistance in strains isolated from food and food processing facilities in different regions [52–55]. Prevalence ranged from 50 to 66%, suggesting that cadmium resistance is globally widespread and highly prevalent in food-associated isolates. One study also noted that isolates repeatedly isolated from milk and dairy foods in Northern Ireland were more likely to be cadmium-resistant than those that were only sporadically recovered [56], suggesting that food or food processing facilities may provide unique pressures that select for cadmium resistance in *L. monocytogenes*.

In contrast to arsenic resistance often associated with serotype 4b, cadmium resistance was frequently encountered among isolates of serotype 1/2a, which are over-represented among food isolates compared with those of clinical origin [29,40]. In congruence with this association, cadmium resistance was generally much more prevalent than arsenic resistance among *L. monocytogenes* from foods and food processing environments [29,52–55]. Even in serotype 4b, the prevalence of cadmium-resistant isolates surpassed that of arsenic-resistant isolates; however, approximately 50% of serotype 4b cadmium-resistant isolates were also resistant to arsenic due to LGI2, as discussed above. While genes putatively associated with both cadmium and arsenic detoxification are co-localized on

LGI2, multiple other cadmium resistance determinants have been identified in *L. monocytogenes* and are harbored both chromosomally and on plasmids, as will be discussed in the following section.

### **7. Plasmid-Associated Cadmium Resistance Determinants**

In a survey of *L. monocytogenes* plasmids from strains of diverse origins (food, environmental, and clinical), an estimated 28% of the isolates were found to harbor plasmids, and most (95%) of these plasmid-harboring strains were found to be cadmium-resistant [28]. This was consistent with a later survey of plasmids in the genus *Listeria* that spanned multiple species, serogroups, and origins which found that, besides the origin of replication, the most common plasmid-borne elements were cadmium resistance cassettes encountered in all plasmids that were analyzed [34]. The cadmium resistance cassettes on the plasmids encoded a cadmium efflux P-type ATPase (*cadA*) and its putative repressor *cadC*. Multiple *cadA* determinants have been identified in *L. monocytogenes* and have been serially numbered, e.g., *cadA1*, *cadA2*, *cadA3*, etc., in the order in which they were discovered. Many *cadA*-harboring plasmids also harbored putative copper-resistance determinants [34]. Intriguingly, plasmid-encoded *cadA* in conjunction with a cassette of genes for arsenic detoxification has only been encountered once, in the aforementioned pLI100 of *L. innocua* CLIP 11262 [31,35]. The evolutionary and ecological mechanisms mediating the scarcity of pLI100-like plasmids with genes for both cadmium and arsenic resistance remain unclear.

The first *cadA* determinant to be identified in *L. monocytogenes* was *cadA1*, which was genetically similar to the *cadA* characterized in *Staphylococcus aureus* [38]. While the *cadA1* in *S. aureus* conferred resistance to both cadmium and zinc, the plasmid-harbored *cadA1* in *L. monocytogenes* was specific to cadmium [38]. *CadA1* was harbored on the mobile genetic element Tn*5422* [57]. Interestingly, Tn*5422* was never detected chromosomally but appeared to integrate extensively into plasmids, leading Lebrun et al. to speculate that this element was responsible for much of the size variation of plasmids in *L. monocytogenes* [38]. Tn*5422*-associated *cadA1* has been subsequently identified on numerous other plasmids of *L. monocytogenes* (Figure 1) [57,58].

**Figure 1.** Resistance gene distribution across mobile genetic elements for *cadA1*, *cadA2*, and *cadA4*. *Cad* family members, benzalkonium chloride resistance determinants, toxic triphenylmethane dye resistance determinants, and putative arsenic detoxification determinants are in red, black, purple, and green, respectively.

A second putative *cadA* (*cadA2*) was first identified on the large plasmid pLI100 of *L*. *innocua* CLIP 11262, followed by its discovery on the approximately 80 kb plasmid pLM80 of *L. monocytogenes* H7858, a strain implicated in a large, multistate outbreak in the U.S. in 1998–1999, which involved contaminated hotdogs [31,33] (Figure 1). The latter plasmid was later experimentally confirmed to confer not only cadmium resistance but also resistance to the quaternary ammonium compound (QAC) benzalkonium chloride, via the *bcrABC* efflux cassette that mediated enhanced tolerance to benzalkonium chloride and other QACs [32,59]. In addition, pLM80 conferred resistance to toxic triphenylmethane dyes such as crystal violet and malachite green via *tmr*, a determinant that detoxifies these dyes and appears to have been acquired from Gram-negative bacteria [32,60] (Figure 1).

Plasmids harboring *cadA1* and *cadA2* have been observed in both *L. monocytogenes* and other, non-pathogenic *Listeria* spp. [34,61]. Several studies have suggested that in *L. monocytogenes*, *cadA1* is more common than *cadA2* [53,54,62]. Data also suggest that *cadA1* may have been predominant earlier in *L. monocytogenes* [28,38,57], since *cadA2* was not identified until the characterization of strains implicated in the 1998–1999 hot dog outbreak [32,33]. It is thought-provoking that pLM80-like plasmids were not identified previously in *L. monocytogenes* and that *cadA2* determinants were not detected among the cadmium-resistant plasmid-harboring isolates in earlier studies [28,57]. It is tempting to speculate that the reasons are related to co-selection of *cadA2*-harboring plasmids for resistance to QACs, which were not recognized as disinfectants until 1934 [63] and which have been routinely and extensively employed for sanitation of food processing facilities only in the past few decades. A survey of *L. monocytogenes* isolates of primarily serotype 1/2a or 1/2b from turkey processing plants in the U.S. revealed that cadmium-resistant isolates that also exhibited enhanced QAC tolerance were more likely to harbor *cadA2,* either alone or together with *cadA1*, than isolates that were cadmium resistant but without enhanced tolerance to QACs [62].

Strains harboring both *cadA1* and *cadA2* were encountered in some surveys [54,62], while other studies reported little or no co-occurrence of *cadA1* and *cadA2* in the same strains [36,53]. Such findings suggest that certain environments may be more conducive to the co-occurrence of these two *cadA* determinants. Analysis of *Listeria* plasmid sequences failed to reveal plasmids that harbor both *cadA1* and *cadA2* [34], suggesting that strains positive for both *cadA1* and *cadA2* harbored these determinants on different plasmids.

Overall, *cadA1* and *cadA2* were far more prevalent in serotype 1/2a and 1/2b strains from food and food processing plants in comparison to serotype 4b [53,62]. This could potentially be explained by the overall greater prevalence of plasmids in serogroup 1 than in serogroup 4 [34]. Strains that harbored only *cadA1* were significantly more likely to be serotype 1/2a, while those harboring only *cadA2* were significantly more likely to be serotype 1/2b [62]. An unexpected finding has been the lack of detection of *cadA1* among serotype 4b clinical isolates belonging to the major clones CC2 and CC6, while *cadA2* was not detected in another leading clone, CC1, suggesting proclivity of different clonal groups for specific cadmium resistance determinants [36,39]. The underlying reasons for differential prevalence of different *cadA* determinants in various serotypes and clones of *L. monocytogenes* are worthy of further investigation and may reflect differences in their ecology, including the microbial community that may include donors for the plasmids, and the accompanying selective pressures.

### **8. Chromosomal Cadmium Resistance Determinants**

The proliferation of whole genome sequencing (WGS) data resulted in the discovery of several cadmium resistance determinants harbored chromosomally in *L. monocytogenes*. The first such determinant was *cadA3*, harbored on a mobile integrated conjugative element in strain EGDe [31]. While direct experimental evidence for its role in cadmium resistance is still lacking, the presence of this gene has been associated with tolerance to cadmium of >140 μg/mL [36]. Thus far, this determinant has been encountered infrequently, having been identified only in EGDe and a few additional strains [35,36,53]. A survey of 136 serotype 4b isolates from human sporadic listeriosis in the U.S. revealed 45 that were cadmium resistant, of which only one harbored *cadA3* [36].

Another chromosomal *cadA* family member (*cadA4*) was identified within the previouslydiscussed LGI2 in the chromosome of strain Scott A [35,36,48] (Figure 1). While *cadA1–3* are all associated with tolerance to cadmium of >140 μg/mL and can be detected through routine screening of the isolates on 70 μg/mL [29,52], *cadA4* only confers resistance to approximately 50 μg/mL of cadmium and thus can be detected by growth at 35 but not at 70 μg/mL [29,36,49,52,53]. Thus, resistance mediated by *cadA4* would have been undetected if isolates were screened at a level of 70 μg/mL as employed in studies prior to 2013 [28,29,52,53]. For this reason, the prevalence of cadmium resistance in *L. monocytogenes* may have been underestimated in earlier studies. Mechanisms underlying differences in resistance level between *cadA4* and *cadA1–A3* remain to be identified, but

may reflect the divergent nature of the deduced *cadA4* product. Pairwise comparisons revealed that the *cadA4* amino acid sequence had approximately 36% identity to those encoded by *cadA1–A3*, while the latter exhibited approximately 70% identity to each other [36,49].

Due to its more recent characterization, *cadA4* has not been surveyed as extensively as *cadA1* and *cadA2*, and its distribution was mostly investigated in serotype 4b strains often associated with the arsenic resistance island LGI2 [30,36,39]. In a survey of 136 serotype 4b isolates from sporadic cases of human listeriosis in the United States discussed above, *cadA4* accounted for approximately 10% of the isolates and 29% of the cadmium-resistant isolates [36]. Interestingly, *cadA4* was always located on LGI2 downstream of arsenic resistance genes [39]. Even though *cadA4* has been primarily reported in serotype 4b, it has also been detected in a few strains of other serotypes (1/2a, 1/2b and 1/2c), including the persistent clone CC14 (serotype 1/2a), where *cadA4* was located on LGI2 downstream to arsenic resistance genes, as in serotype 4b [39,41]. To date, we lack reports of *cadA4*-harboring isolates of *Listeria* spp. other than *L. monocytogenes*.

As discussed above, a diversified LGI2 derivative (LGI2-1) was observed in a subset of arsenic-resistant serotype 4b CC1 isolates, and another derivative was also identified in a serotype 1/2c strain of CC9 [39]. The *cadA4* homolog in these divergent islands exhibited ~90% amino acid identity with the *cadA4* of Scott A (Figure 2). Together, LGI2-associated *cadA4* and its divergent homolog in LGI2-1 accounted for half of serotype 4b cadmium-resistant isolates in a previous study [30]. Further studies need to be conducted to investigate the prevalence and functional characteristics of this or additional *cadA4* homologs.

**Figure 2.** Amino acid alignment between *cadA4* in strain Scott A and its divergent counterpart (here labeled as *cadA5*) encountered in LGI2-1. Amino acid alignment was generated using CLC Genomics Workbench 11.0 [64].

### **9. Impacts of Heavy Metal Resistance Determinants on other Adaptations, Including Virulence**

There is a growing body of evidence that suggests links between heavy metal resistance and the ability of *L. monocytogenes* to cause disease. For instance, an investigation of the prevalence of arsenic and cadmium resistance among serotype 4b isolates from human listeriosis patients in the United States found a high prevalence of resistance and a strong association with clones repeatedly implicated in outbreaks [30]. More direct evidence supporting the involvement of cadmium resistance with pathogenicity has been provided in several studies. The first experimental evidence for the possible involvement of cadmium resistance determinants in virulence was obtained from *in-vivo* transcriptional analysis of *L. monocytogenes* from the livers of mice infected with strain EGDe, which harbors *cadA3* [65]. The putative repressor encoded by *cadC3* was markedly upregulated in the liver of the infected mice, and deletion of this gene resulted in a decrease in virulence when the bacteria were administered intravenously [65]. Similar findings were also reported by Pombinho et al. in 2017, who again identified *cadC3* as being essential for virulence of *L. monocytogenes* [66]. They found that, in addition to repressing *cadA3*, *cadC3* also repressed *ispB,* which is involved in initiating an immune response, thus helping *L. monocytogenes* avoid detection by the host immune system [66]. Interestingly, a transposon insertion mutant of *cadA4* [49] showed increased virulence in the *Galleria mellonella* model, suggesting an inverse relationship between *cadA4* and virulence [49]. This finding is consistent with the previous study, which demonstrated that the putative repressor *cadC3* was required for full virulence [65]. Taken together, the results of these studies suggest an association between cadmium resistance determinants and the ability of *L. monocytogenes* to cause disease. It is important to note, however, that these studies have focused on *cadC3* and *cadA4*. We currently lack information on the potential virulence or pathogenicity roles of the predominant cadmium resistance determinants *cadA1* or *cadA2*, and their cognate *cadC* repressors (*cadC1* and *cadC2*, respectively).

The cadmium and arsenic resistance genes of *L. monocytogenes* discussed here are accompanied by transcriptional regulators of the ArsR family of metal-associated transcriptional regulators [67]. In other bacterial systems, these regulators have been found to regulate expression of single genes in some circumstances, while mediating a global transcriptional response in others [67]. Most often, they regulate expression of genes directly involved in metal detoxification, but they can also impact expression of genes with a variety of other functions including oxidative stress tolerance, acid adaptation, respiration and ribosome biogenesis [67].

Members of the ArsR family were involved in the regulation of virulence-associated genes in various species. For instance, the PhoPR two-component regulatory system which is responsible for the regulation of virulence and persistence genes in *Mycobacterium* spp. was shown to be under the control of an ArsR transcriptional factor [68]. In *L. monocytogenes*, the aforementioned cadmium resistance regulator *cadC* belongs to the ArsR family and, as previously stated, *cadC3* was involved in virulence via its impacts on *ispB* [49,65,67].

In the case of the arsenic resistance genomic island LGI2, it is tempting to speculate that this element may have roles in virulence and pathogenicity, based on the fact that, as discussed above, LGI2 has been detected exclusively in *L. monocytogenes* and primarily in serotype 4b, which makes significant contributions to human listeriosis [56]. However, direct experimental evidence is needed to assess LGI2's roles in virulence, e.g., by comparing virulence of isogenic strains with and without specific LGI2-associated genes. This was pursued with *cadA4* [49], but similar investigations with the arsenic resistance genes on LGI2 are lacking.

The potential contributions of heavy metal resistance to environmental persistence of *L. monocytogenes* remains to be elucidated. As discussed earlier, one study found that isolates repeatedly isolated from contaminated foods were more likely to be cadmium-resistant than those that were only sporadically encountered [56], and 74% or more of the serotype 1/2a and 1/2b isolates from turkey processing facilities were cadmium resistant, harboring *cadA1* and/or *cadA2* [52,62]. It is thus tempting to speculate that cadmium resistance via these determinants may enhance the capacity of the isolates

to persist in the contaminated food or food processing environments, but the underlying mechanisms remain unknown.

In addition to their overt impacts on enhanced tolerance to heavy metals, heavy metal resistance genomic islands may indirectly influence environmental fitness or pathogenicity by promoting the horizontal transfer of accessory genes. In bacteria, metal resistance genes have been found to co-localize with antibiotic and other resistance genes on mobile genetic elements, such as plasmids, genomic islands, and transposons [69–72] (Figure 1). As discussed above, pLM80 and related plasmids harbor not only cadmium resistance genes but also genes mediating enhanced tolerance to QACs and toxic dyes [32,34,60]. Sequence data suggest that these elements have been introduced to *L. monocytogenes* from other species [48,60], and there is direct experimental evidence for the transfer of such elements between *L. welshimeri* or *L. innocua* and *L. monocytogenes* [61]. This creates the possibility that metal contamination and/or metal resistance genes could facilitate the acquisition and transfer of other resistance genes to *L. monocytogenes*, or from *L. monocytogenes* and other *Listeria* spp. to other bacterial agents of public health concern. It has also been shown that extremely low levels of metals can induce transcription of metal resistance genes and exert sufficient selective pressure to result in the retention of these elements [49,57,73]. These data suggest that in minute amounts such as might be encountered in the environment or in an animal host, heavy metals can potentially exert selective pressure, which in turn could direct the acquisition or transfer of mobile genetic elements that can impact the environmental or *in-vivo* fitness of *L. monocytogenes*.

### **10. Conclusions**

Metals play a key role in the survival of *L. monocytogenes* both in the environment and in animal hosts [8]. Essential metals must be acquired, and toxic effects of excess metals must be mitigated. While the cellular functions of *L. monocytogenes* associated with essential metals have been extensively studied and reviewed, those involved with exclusively toxic metals such as cadmium and arsenic are poorly understood. The significance of these determinants is shown by their wide distribution within *L. monocytogenes,* as well as their association with food, food processing plants, clinical strains and clonal groups involved in outbreaks [36,53–55,62]. Evidence from other microorganisms suggests the involvement of metal resistance genes in a variety of functions beyond just metal detoxification [24,74]. Several studies discussed here would also suggest alternate and additional functions for these genes in *L. monocytogenes* [49,65]. Given their prevalence, potential involvement in selection and population dynamics, as well as their growing implication in important alternative cellular functions such as virulence, heavy metal resistance genes are an ideal candidate for further study.

**Author Contributions:** C.P. authored the initial draft of the manuscript and worked on later edits and revisions, S.L. contributed significant portions of original material to the manuscript and worked on later edits and revisions, S.K. contributed to initial concept and worked on later edits and revisions.

**Funding:** This research was funded by USDA NIFA grant 2012-67017-30218.

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

### **References**


© 2018 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 (http://creativecommons.org/licenses/by/4.0/).

### *Article*

### **Distribution of the** *pco* **Gene Cluster and Associated Genetic Determinants among Swine** *Escherichia coli* **from a Controlled Feeding Trial**

**Gabhan Chalmers 1, Kelly M. Rozas 2, Raghavendra G. Amachawadi 3, Harvey Morgan Scott 2, Keri N. Norman 4, Tiruvoor G. Nagaraja 5, Mike D. Tokach <sup>6</sup> and Patrick Boerlin 1,\***


Received: 24 September 2018; Accepted: 15 October 2018; Published: 18 October 2018

**Abstract:** Copper is used as an alternative to antibiotics for growth promotion and disease prevention. However, bacteria developed tolerance mechanisms for elevated copper concentrations, including those encoded by the *pco* operon in Gram-negative bacteria. Using cohorts of weaned piglets, this study showed that the supplementation of feed with copper concentrations as used in the field did not result in a significant short-term increase in the proportion of *pco*-positive fecal *Escherichia coli*. The *pco* and *sil* (silver resistance) operons were found concurrently in all screened isolates, and whole-genome sequencing showed that they were distributed among a diversity of unrelated *E. coli* strains. The presence of *pco*/*sil* in *E. coli* was not associated with elevated copper minimal inhibitory concentrations (MICs) under a variety of conditions. As found in previous studies, the *pco*/*sil* operons were part of a Tn*7*-like structure found both on the chromosome or on plasmids in the *E. coli* strains investigated. Transfer of a *pco*/*sil* IncHI2 plasmid from *E. coli* to *Salmonella enterica* resulted in elevated copper MICs in the latter. *Escherichia coli* may represent a reservoir of *pco*/*sil* genes transferable to other organisms such as *S. enterica*, for which it may represent an advantage in the presence of copper. This, in turn, has the potential for co-selection of resistance to antibiotics.

**Keywords:** copper; resistance; swine; *Escherichia coli*

### **1. Introduction**

As restrictions on the use of antimicrobial agents for the purpose of growth promotion and disease prevention in farm animals are increasing, alternatives to these agents are becoming more popular. Feed supplementation with copper is one of the most frequently used, particularly in the swine industry [1]. The copper concentrations used in swine feed for growth promotion are relatively high and usually in the range of 100 to 250 ppm [2].

Bacteria developed mechanisms to cope with high concentrations of copper. In Gram-positive bacteria, the most well-known mechanism is the *tcrB* gene [3,4], which provides a selective advantage to intestinal enterococci in swine and cattle [5–7]. It also seems to be involved in the co-selection of bacteria resistant to antimicrobial agents of importance for both veterinary and human medicine [6,8]. Several tolerance and homeostasis mechanisms were described in Gram-negative bacteria and in *Enterobacteriaceae* in particular (for a review, see, for instance, References [9–11]). Although most are chromosomally encoded and present in the majority of bacteria from the species in which they reside, one of them initially found in *Escherichia coli* was shown to be plasmid-borne and not present in every isolate of the species [12]. The *pco* gene cluster associated with this system was later characterized in more detail [13] and shown to consist of seven genes (*pcoA*, *B*, *C*, *D*, *R*, *S*, and *E* [14–16]). This cluster was found in a variety of *Enterobacteriaceae* species and, depending on bacterial species and strain, the associated copper tolerance phenotype was variable, both in terms of copper minimal inhibitory concentration and inducibility [13]. Since then, several studies showed that *pco* genes are not always plasmid-borne but can also regularly be found on the chromosome of *Enterobacteriaceae* species, including *Salmonella enterica* and *E. coli* [17–19]. This spread and mobility may be related to the location of the *pco* genes on a Tn*7*-like transposon [17]. This Tn*7*-like element frequently carries both the *pco* gene cluster and the *sil* gene cluster [17] associated with silver tolerance [20]. Investigations on silver and copper tolerance in *S. enterica* isolates from Portugal showed a clear association between the presence of *sil* genes and copper tolerance, while the presence of *pco* genes did not seem to show any evident correlation with this phenotype [18,21]. Similarly, recent experimental studies on the effect of feed supplementation with copper on fecal *E. coli* and on the fecal metagenome of swine did not demonstrate any clear or systematic selective effect for *pco* genes [2,22]. These results suggest that either the concentrations of copper used in feed (125 ppm) may have been too low to have such an effect, or the presence/absence of the *pco* genes did not affect the tolerance of *E. coli* and other bacteria to elevated copper concentrations under the conditions found in the gut of the animals. However, a negative association was observed between copper supplementation and resistance to antimicrobials, as well as resistance to extended-spectrum cephalosporins in particular [2]. Also, an association between *pco* genes and the *tet*(B) tetracycline resistance gene was detected in *E. coli*, while these two genes were negatively associated with the *bla*CMY and *tet*(A) genes encoding for extended-spectrum cephalosporins and tetracycline resistance, respectively [2].

Based on these observations, the objectives of this study were (a) to replicate the previous experiments of Agga and collaborators [2] and reassess the associations between the *pco* genes and *tet*(A), *tet*(B), *bla*CMY, and *bla*CTX-M among *E. coli* from groups of swine subjected to diverse combinations of copper and tetracycline feed supplementation; (b) to use whole-genome sequencing to assess the genetic diversity and clonal relationships of *E. coli* isolates recovered from these experiments and carrying diverse combinations of these genes; (c) to compare the copper susceptibility and genome sequences of selected isolates with plasmid-borne and chromosomally encoded *pco* genes; and (d) to transfer *E. coli* plasmids carrying the *pco* and *sil* gene clusters into *S. enterica* by conjugation, and assess the associated copper susceptibility. These objectives related to the use of copper in feed and its effect on copper tolerance in *E. coli* were part of a broader study on alternatives to antibiotics [23]. The latter also included the use of zinc and oregano oil, but is not discussed here.

### **2. Materials and Methods**

### *2.1. Experiment Design*

The Kansas State University Institutional Animal Care and Use Committee approved the protocol for this experiment (AUP # 3135). The study was conducted at the university's Segregated Early Weaning Facility in Manhattan, KS. Each pen (1.22 × 1.22 m) had metal tri-bar flooring, one four-hole self-feeder, and a cup waterer to provide ad libitum access to feed and water. This experiment was also described in a publication by Feldpausch and collaborators [23].

A total of 350 piglets (21 days old) were assigned to one of 70 pens (five piglets per pen), which were then randomly assigned to each of the 10 in-feed treatments arranged in a 2 × 2 × 2 (+2) factorial design. In detail, the ten dietary treatments were (1) a basal swine diet fully meeting National Research Council (NRC) nutritional guidelines, including 16.5 ppm of supplemental copper and 165 ppm of supplemental zinc (control group); (2) a basal diet supplemented with 125 ppm of copper provided by copper sulfate; (3) a basal diet supplemented with zinc at 3000 ppm of zinc provided by zinc oxide; (4) a basal diet supplemented with oregano premix containing 5% oregano oil (Regano 500; Ralco-mix Products, Marshall, MN, USA); (5) a basal diet with both 125 ppm of copper and zinc at 3000 ppm; (6) a basal diet with both 125 ppm of copper and oregano premix; (7) a basal diet with both zinc at 3000 ppm and oregano premix; (8) a basal diet containing copper, zinc, and oregano premix; (9) a basal diet containing a preventive level of chlortetracycline (CTC) (22 mg/kg body weight (BW); High CTC); and (10) a basal diet containing a subtherapeutic level of CTC (4 mg/kg BW; Low CTC). These latter treatment groups (9 and 10) did not interact with other main treatment factors (Zn, Cu, and oregano oil) in the study design, so as to assess the impact of antimicrobial alternatives versus both true negative controls and the "existing standard controls" represented by antimicrobial use groups. The basal diet consisted of corn, soybean meal, vitamins, amino acids, and trace mineral supplements per NRC requirements.

The study lasted 49 days with an initial seven days of acclimation, and 28 days of feeding trial, followed by 14 days of washout phase. Three fresh fecal samples were collected from random pigs in each pen by gentle rectal massage at days 0 and 28 of the feeding trial. Fecal samples were transported to the laboratory for further processing. The fecal samples were thoroughly mixed with 50% glycerol (1:1) and stored at −80 ◦C. Laboratory personnel were blinded to the treatment groups.

### *2.2. Selection of Isolates and Detection of* pco

A total of 420 samples, 210 from day 0 (for pre-treatment effect) and 210 from day 28 (maximum treatment effect), were subjected to bacteriological culture and quantified for *E. coli* using standard isolation techniques and spiral plating. Briefly, one gram of 50:50 glycerol and feces were diluted in 9 mL of phosphate-buffered saline (PBS). A 50-μL aliquot of the fecal suspension was spiral-plated onto each of MacConkey agar, MacConkey agar supplemented with 16 mg/L tetracycline, and MacConkey agar supplemented with 4 mg/L ceftriaxone using an Eddy Jet 2 spiral plater (Neu-tec Group Inc., Farmingdale, NY, USA). Crude quantification values were determined by the Flash & Go Automatic Colony Counter (Neu-tec Group Inc.). A single, randomly selected colony was used from a plain MacConkey plate and confirmed as *E. coli* by lactose fermentation and an indole test; the species identity was also later confirmed with Illumina-based DNA sequencing. Isolates were preserved at −80 ◦C in protectant CryoBeads™ for further characterization.

Antimicrobial susceptibility testing was conducted by broth microdilution using the Sensititre™ system (TREK, Thermo Scientific Microbiology, Oakwood Village, OH, USA) and Sensititre™ NARMS Gram-negative plates (CMV3AGNF) on 403 *E. coli* isolates. *Escherichia coli* ATCC 25922, *Escherichia coli* ATCC 35218, *Pseudomonas aeruginosa* ATCC 27853, *Staphylococcus aureus* ATCC 29213, and *Enterococcus faecalis* ATCC 29212 were used as quality control strains. Plates were incubated at 37 ◦C for 18 h and read on a Sensititre OptiRead™ (TREK). The results were interpreted according to Clinical and Laboratory Standards Institute (CLSI) guidelines [24]. Intermediate isolates were interpreted as susceptible for binary statistical analyses.

Detection of *pco*, tetracycline, and extended-spectrum cephalosporin resistance genes was performed by PCR with the primers described in Table 1. Thermocycling conditions were the same as those defined in the respective references. Amplicons were visualized by horizontal gel electrophoresis and ultraviolet (UV) imaging.


**Table 1.** PCR targets and primers used for detection of *pco*, *sil*, and antimicrobial resistance genes.

### *2.3. Copper Susceptibility*

Susceptibility to copper was analyzed by broth microdilution for four randomly selected *pco*/*sil*-positive *E. coli*, and two *pco*/*sil*-negative isolates. Isolates were from day 0 (KSC9, 27, 64, and 207) and day 28 (KSC857 and 1031), from animals within the copper treatment group (KSC27, 857, and 1031) and from those without (KSC9, 64, and 207). A stock solution of 400 mM copper(II) sulfate (Sigma-Aldrich, St. Louis, MO, USA) was prepared in double-distilled water (ddH2O), and filter-sterilized. A non-serial dilution range (0, 4, 8, 16, 20, 24, 36, 48, 64, and 100 mM) was prepared in Mueller–Hinton II broth, cation-adjusted (Becton Dickinson, Franklin Lakes, NJ, USA), and each dilution was adjusted to pH 7.2 using 5 M NaOH [11]. Bacterial suspensions of a 0.5 McFarland standard were diluted 1/100, and 50 μL of this suspension was inoculated in a 96-well plate with 50 μL of the copper dilutions, resulting in halving of the initial copper concentrations. Microplates were incubated at 37 ◦C for 16 h, under both aerobic and anaerobic conditions. Minimum inhibitory concentration (MIC) was defined as the first concentration without visible growth. Minimum bactericidal concentrations (MBCs) were determined by removing 10 μL from wells that showed no visible growth, and plating them on Mueller–Hinton II agar plates for incubation at 37 ◦C for 16 h. The ATCC 25922 *E. coli* strain (*pco*/*sil*-negative) was used as a negative control for susceptibility testing.

Minimum inhibitory concentrations were also determined using agar plate dilutions of copper, as described by Mourão and collaborators [21]. Briefly, copper dilutions of 0, 0.5, 1, 2, 4, 8, 12, 16, 20, 24, 28, 32, and 36 mM were prepared in Mueller–Hinton II agar, and the pH was adjusted as above. One microliter of an approximate 107 colony forming units (CFU)/mL culture was pipetted onto the surface of each plate. Growth at 37 ◦C in both aerobic and anaerobic conditions was assessed after 16 h.

### *2.4. Expression of* pco *by Complementary DNA Synthesis and Real-Time PCR*

Three *E. coli* isolates were selected randomly (two *pco*/*sil*-positive and one negative control, none of which were isolated from copper-treated animals) for determining the expression of *pco* under aerobic conditions, with and without induction with low concentrations of copper, performed as previously described [15,27]. Briefly, isolates were plated overnight at 37 ◦C on Luria–Bertani (LB) agar (Becton Dickinson) plates. A single loop of bacteria was inoculated into 1 mL of LB broth, and vortexed; 200 μL of this suspension was inoculated into 20 mL of LB broth supplemented with 0, 1 mM, and 5 mM copper(II) sulfate and incubated at 37 ◦C for approximately 2.5 h, until optical densities of 0.5 were reached at 600 nm. Broth microdilution MICs were performed again as above under aerobic conditions to observe any effect of this induction on copper tolerance. In parallel, 10 mL of broth was centrifuged, and the resulting pellet was resuspended in 1 mL of RNAlater (QIAGEN Inc., Valencia, CA, USA). Total RNA was extracted using an RNAeasy Mini kit (QIAGEN), according to the manufacturer's instructions. An additional DNase step was performed to ensure all traces of DNA were removed, and was verified by a *pco* PCR using 1 μL as a template. RNA was quantified

using a BioAnalyzer 2100 instrument (Agilent Technologies, Santa Clara, CA, USA), and 100 ng of each RNA preparation was used for complementary DNA (cDNA) synthesis using an Applied Biosystems High-Capacity cDNA Reverse Transcription kit (Thermo Fisher Scientific, Carlsbad, CA, USA).

PCR was performed to amplify gene fragments to be used for cloning into a plasmid vector, for use as a standard curve for real-time PCR. Amplicons of the *pcoA* and *pcoD* genes were produced using primers forward *pcoA* (pcoA\_F), CGGGTATGCAAAGTCATCCT; reverse *pcoA* (pcoA\_R), TTGATCAGCGTGATCCTGAG; and pcoD\_F, AAGCGGTGTCAGACATGAAA; pcoD\_R, GATGGGTCAGATCGCTCAGT, respectively. As controls, two housekeeping gene amplicons for *hcaT* (HcaT major facilitator superfamily transporter) and *rrsA* (16S ribosomal RNA) were amplified using primers hcaT\_F, CTGATGCTGGTGATGATTGG; hcaT\_R, CAATGCAGAATTTGCACCAC; and rrsA\_F, CGGACGGGTGAGTAATGTCT; rrsA\_R, GTTAGCCGGTGCTTCTTCTG, respectively. Each amplicon was cloned into a pCR 2.1-TOPO plasmid vector, using an Invitrogen TOPO-TA cloning kit (Thermo Fisher Scientific). Inserted sequences were confirmed by DNA sequencing, and plasmid DNA was prepared using a Plasmid Midi Kit (QIAGEN). Plasmid DNA was quantified using Quant-IT Picogreen dsDNA reagent (Thermo Fisher Scientific) and read using a DTX 880 Multimode detector (Beckman Coulter, Brea, CA, USA). Gene copy numbers were then predicted by the DNA concentration divided by the molecular weight of the plasmid.

Real-time PCR was used to quantify the expression of each *pcoA*, *pcoD*, *hcaT*, and *rrsA* gene using primers internal to the fragments described above. In triplicate, 1 μL of cDNA or plasmid standards were added to 19 μL of LightCycler 480 SYBR Green I Master (Roche Diagnostics, Indianapolis, IN, USA) containing 250 nM of each primer. Primers used for the quantification of *pco* expression were RT\_pcoAF, TGGTTGATATGCAGGCGATG; RT\_pcoAR, TCCGCGTACGTGAGAACCTT; and RT\_pcoDF, GTCAGGCTCTGTGCCCTGTT; RT\_pcoDR, CCCACTCATCGTCATCAGCA. Housekeeping gene primers used for *hcaT* and *rrsA* were those described by Zhou and collaborators [28].

### *2.5. Next-Generation Sequencing*

A subset of 82 isolates was selected to represent suspected extended-spectrum β-lactamase (ESBL)-producing isolates and isolates with elevated ciprofloxacin MICs (based on Sensititre phenotypes and ciprofloxacin MICs of ≥0.05 mg/L; *n* = 26), isolates carrying the *bla*CMY gene (*n* = 26), and a representative sample of isolates with resistance phenotypes determined by Sensititre (*n* = 30). In addition, all 34 *E. coli* carrying the *pcoD* gene were also included. Genomic DNA was prepared for MiSeq sequencing (Illumina, San Diego, CA, USA) for all of these 116 isolates using a QIAamp DNA extraction kit (QIAGEN), and libraries were prepared using a Nextera XT kit (Illumina). Achtman sequence types were determined with the SRST2 plugin for BaseSpace Labs (Illumina) using the MiSeq paired-end reads, where sufficient read quality was obtained. These reads were also used for core-genome multilocus sequence typing (cgMLST) (EnteroBase typing scheme) using the wgMLST application for BioNumerics v7.6 (Sint-Martens-Latem, Belgium). Single-nucleotide polymorphism (SNP) analysis of each *pco* gene cluster (*pcoEABCDRSE*) was also performed using the wgSNP analysis tool from BioNumerics.

DNA was also prepared for four of the isolates tested for susceptibility to copper and harboring *pco* (KSC9, KSC64, KSC207, and KSC1031) using a MasterPure DNA Purification Kit (Epicentre, Madison, WI, USA) for PacBio RS II sequencing (Pacific Biosciences, Menlo Park, CA, USA). Sequencing and assembly of these four isolate genomes were performed at the McGill University and Génome Québec Innovation Centre, Montreal, QC, Canada. PacBio sequencing assembly was completed on chromosome and plasmid assemblies of sheared large inserts (~20 kbp) using the de novo genome assembly pipeline Hierarchical Genome Assembly Process (HGAP); alignments were further polished using the Quiver consensus algorithm. Genomes and/or plasmids (pMRGN207 and pMRGN1031) containing the *pco* gene cluster were uploaded to GenBank under BioProject

PRJNA355857. Annotations were performed using the NCBI Prokaryotic Genome Annotation Pipeline (PGAP) version 4.0.

### *2.6. Conjugation of a* pco*/*sil *Plasmid into Salmonella*

Based on the known plasmid-borne location in two of our sequenced *E. coli* isolates, we attempted to transfer these plasmids into *Salmonella* recipients to observe any change to copper susceptibility. Six *Salmonella* isolates were selected at random from a large collection of isolates maintained by the Canadian Integrated Program for Antimicrobial Resistance Surveillance (CIPARS) between 2008 and 2011. Isolates were screened for *pco* and *sil* by PCR, and three *pco*/*sil*-positive and three *pco*/*sil*-negative were used for copper susceptibility testing. *Escherichia coli* isolates KSC207 and KSC1031 were used as plasmid donor strains, with *pco*/*sil* plasmid incompatibility types HI2 and FII, respectively (as determined by DNA sequence analysis), while *pco*/*sil*-negative *Salmonella* isolates SA8197 (serovar Kentucky) and SA82540 (serovar Infantis) were used as recipients. Briefly, 50 μL of donor and 100 μL of recipient overnight broth cultures were mixed, plated on LB agar, and incubated overnight to facilitate conjugation. Growth was then resuspended in LB broth, and plated on Brilliant Green agar to inhibit the growth of donor *E. coli*, as both plasmids carried a tetracycline resistance determinant—*tet*(A) and *tet*(B), respectively. Additionally, 12 μg/mL tetracycline was included in the selective plates. Up to five pink-colored colonies were then sub-cultured for purity, and indole testing and PCR for *pco* and *sil* were used to confirm transfer. These transconjugant *Salmonella* were then used for copper susceptibility testing as before using anaerobic agar dilution, and performed in triplicate.

### *2.7. Statistics*

Descriptive and inferential statistical methods were performed using Stata version 15 (StataCorp, College Station, TX, USA). Categorical data (i.e., resistance phenotypes and presence of genes) were tabulated and cross-tabulated to explore bivariable associations between treatments and the following outcomes: (1) presence of phenotypic resistance to (a) tetracyclines, (b) cefoxitin, for preliminary classification of ESBL (susceptible) versus AmpC (resistant) producing β-lactamases, (c) third-generation cephalosporins, and (d) multidrug resistance (MDR) count (integer count), out of 14 antimicrobials tested on a broth microdilution panel; and (2) presence of resistance genes for (a) tetracyclines (*tet*(A) and *tet*(B)), (b) extended-spectrum cephalosporins (*bla*CMY-2 and *bla*CTX-M), and (c) metal resistance genes (*pco* and *sil*). Likelihood ratio χ<sup>2</sup> or Fisher's exact tests (when zero-cells were abundant) were used in bivariable analyses; statistical significance was determined at *p* < 0.05. Differences in growth (log10 CFU) of coliforms on plain versus antimicrobial (tetracycline or ceftriaxone) impregnated plates were examined by unpaired *t*-tests. Multivariable mixed logistic (binary outcomes) and linear models (count and log10 CFU outcomes) using a four-way factorial design (plus two additional indicator variables for low- and high-dose CTC) were built and assessed for each of the binary response (logistic) and quantitative (log10 CFU differences) endpoints. Full factorial models were subjected to reduction, firstly removing non-significant interaction terms and then main effects. Of note, in-feed copper (low versus high), sampling day (0 versus 28), and their interaction were always retained marginal means estimated with *p*-values representing the post hoc multiple comparisons adjusted using Bonferroni's correction.

### **3. Results**

### *3.1. Resistance Determinants*

Not all samples (*n* = 420) yielded lactose- and indole-positive isolates. A total of 403 *E. coli* isolates were included in this analysis. Thirty-four of the 403 isolates (8.4%) were positive for *pco* by PCR. Twelve isolates carried a *bla*CTX-M gene (3.0%), and 60 carried *bla*CMY (14.9%, all variant CMY-2). All but two isolates (99.5%) were positive for *tet*(A) or *tet*(B) (121 (30.0%) and 267 (66.3%), respectively); 13 isolates (3.2%) carried both of these tetracycline resistance genes.

All *bla*CTX-M sequences encoded the CTX-M-27 variant, and all *bla*CTX-M-positive isolates were *pco*-negative, but *tet*(B)-positive; 11 of these 12 isolates were ST744 and had an ampicillin/ceftriaxone/ciprofloxacin/nalidixic acid/tetracycline resistance phenotype.

### *3.2. Associations of Copper Treatment Groups with* pco *Prevalence*

Multi-level model-adjusted estimates of the occurrence of the *pco* gene were initially unstable in the presence of the full factorial specification. A reduced model containing copper (forced into model), day (and its interaction), and low- versus high-dose CTC yielded a model significant at *p* < 0.03. Copper did not select for the *pco* gene (*p* = 0.249); that is, copper supplemented at NRC requirements yielded 0.07 (95% confidence intervals (CIs): 0.04–0.11) of isolates with the gene, versus 0.11 (95% CIs: 0.04–0.17) in the group supplemented with copper beyond nutrient needs. Likewise, copper did not appear to select for any of the additional microbiological endpoints (*p* > 0.05; data not shown). Low-dose CTC did select (*p* = 0.001) for increased *pco* with 0.28 (95% CIs: 0.11–0.46) of the isolates in the low-dose group harboring the gene versus 0.06 (95% CIs: 0.04–0.09) in pigs not receiving any CTC. The difference between the high-dose CTC and each of the other two levels was not significant (0.12; 95% CIs: 0.00–0.24). Of note, though unexplained by the field trial study design, the *pco* gene was significantly (*p* = 0.012) associated with a lower MDR count; however, most of this difference was due to a lack of the highest MDR counts (maximum with *pco* present = 7 versus 11 in *pco*-negative isolates). No associations were significant (*p* > 0.05) among the genes tested, with the notable exception of *pco*/*sil* for which there was complete agreement (34/34; Fisher's exact test, *p* < 0.0001).

### *3.3. Susceptibility to Copper and Expression of* pco

Susceptibility to copper by broth microdilution ranged from 12 to 18 mM in all seven *E. coli* isolates tested (Table 2). The presence of the *pco* gene cluster did not appear to have an effect on susceptibility to copper under both aerobic and anaerobic conditions, nor did the use of broth or solid media. All MBC values were identical to their respective MIC results. The ATCC 25922 (*pco*/*sil*-negative) isolate had an MIC and MBC of 12 mM. For the *Salmonella* isolates, no systematic differences were observed when using broth dilution under anaerobic conditions (MICs of 12 to 18 mM), but the MICs of the *pco*/*sil*-positive and -negative isolates differed when tested on agar, with values of 24 mM and 4 mM, respectively (Table 2).


**Table 2.** Isolates used in this study for copper susceptibility testing. All reported minimum inhibitory concentration (MIC) values were determined under anaerobic conditions. MBC—minimum bactericidal concentration.

<sup>1</sup> concentration in mM; MICs are averages of three complete biological replicates. <sup>2</sup> ND (not done).

Induction with 1 mM and 5 mM under aerobic conditions appeared to have no effect on the MIC of each isolate tested (data not shown). RNA extraction and cDNA analysis also showed no significant change in *pcoA* or *pcoD* transcription, observed by real-time PCR. Using *hcaT* as the reference gene (expression of *rrsA* expression was considerably higher than all other genes, and was not used in the analysis), expressions of *pcoA* and *pcoD* were measured as the "target" in all samples using relative quantification. The average adjusted crossing point (CP) across all three induction concentrations was 31.9 (±1.7) for *pcoA*, and 32.1 (±1.9) for *pcoD*.

### *3.4. Next-Generation Sequencing and cgMLST*

All 116 *E. coli* isolates selected for MiSeq sequencing were analyzed using the BioNumerics software (Figure 1). All 34 *pco*-positive isolates recovered during this study also carried the *sil* gene cluster (determined by reference mapping of MiSeq reads), and none of the 82 *pco*-negative isolates carried any *sil* gene. Core-genome MLST analysis of the MiSeq data showed a random distribution of *pco*/*sil* among sequence types and no association with a specific clonal lineage (Figure 1). Further *pco* SNP analysis showed the gene cluster to be highly conserved within a sequence type (ST), but had some variation between most STs (Figure 2).

**Figure 1.** Minimum spanning tree of 116 *Escherichia coli* isolates, using core-genome multilocus sequence typing (MLST) analysis comprising 2513 genes (BioNumerics *E. coli*/*Shigella* EnteroBase scheme). A tree with the highest resampling support is shown, using 1000-resampling bootstrapping. (**A**) Isolates carrying the *pco* gene cluster are highlighted in orange; (**B**) isolates carrying the resistance genes *bla*CTX-M, *bla*CMY, *tet*(A), and *tet*(B) are indicated. Letters in 1A indicated single-nucleotide polymorphism (SNP) types found in Figure 2. Circles containing multiple sections indicate multiple isolates within a core-genome sequence type. Isolates used for minimum inhibitory concentration (MIC) testing are also highlighted and labeled in 1A.

Pacific Biosciences long-read sequencing and assembly showed the *pco* gene cluster to be both plasmid and chromosomally encoded (Figure 3). The gene cluster was accompanied by the *sil* gene cluster in all four *pco*-positive isolates, and was flanked by a Tn*7*-like transposable element in three of them. Using this *pco*–*sil*–*tns* sequence as a template, short-read Illumina sequences were successfully mapped onto all but one *pco*-positive isolate (KSC1031) for the complete structure, demonstrating the highly conserved nature of this transposable element across multiple STs and plasmid types found in this study.

**Figure 2.** Phylogenetic analysis of the *pcoEABCDRSE* gene cluster (5487 bp) of all *pco*-positive isolates in this study, using a categorical (differences) similarity coefficient and unweighted pair group with arithmetic mean (UPGMA) cluster analysis. Treatment groups, including Cu (copper), Zn (zinc), oreg (oregano oil), and high/low-dose chlortetracycline (CTC) are shown. Letters indicating identical SNP groups are also shown in Figure 1A. Isolates used for MIC testing are underlined. ST: sequence type.

Both of the two chromosomal Tn*7*-like elements were found in approximately the same position, 300 kbp downstream of the preferential *glmS* insertion point for Tn*7* [29], much farther than previously reported [30,31]. Neither plasmid harboring the *pco*/*sil* gene cluster contained any known virulence factors (using Virulence Finder v1.5 [32]). In one instance, the *pcoA* gene was interrupted by a transposase, while the *sil* gene cluster was always intact (Figure 3).

**Figure 3.** Genetic context of the *pco*–*sil*–*tns* area of four isolates from this study sequenced by PacBio. KSC1031 (plasmid pMRGN1031; GenBank accession number CP019561), KSC207 (plasmid pMRGN207; CP019559), KSC64 (CP018840), and KSC9 (CP018323) were compared to the pR478 (BXX664015) and pAPEC-O1-R (DQ517526) previously published sequences. Colors indicate the *pco* operon (orange), the *sil* operon (light blue), transposases (red), Tn*7* genes (green), and others (dark blue).

### *3.5. Conjugation of a* pco*/*sil *Plasmid into Salmonella*

The *pco*/*sil* plasmid (IncHI2) from *E. coli* isolate KSC207 was successfully transferred to both *Salmonella* recipient isolates, as confirmed by PCR. Using agar dilution under anaerobic conditions, a clear difference was observed between the transconjugants carrying the *sil*/*pco* gene clusters (MIC = 24), and the recipient isolates (MIC = 4; Table 2). The IncFII plasmid from *E. coli* isolate KSC1031 could not be successfully transferred to either *Salmonella* recipient.

### **4. Discussion**

The plasmid-borne nature of the *pco* gene cluster [12,13,19] and the recent demonstration of its linkage with important antimicrobial resistance genes [11,17] warrant further investigations on the potential medical and public health implications of copper use in animal feed. A previous set of experiments under controlled conditions failed to demonstrate any significant selection of *pco*-positive *E. coli* isolates [2] or increase in *pco* copy numbers [22] in feces from pigs fed copper after weaning. However, *pco*-positive isolates were more frequently associated with the tetracycline resistance gene *tet*(B) than with its *tet*(A) counterpart [2]. This suggested some possible gene linkage between *pco* genes and *tet*(B) on mobile elements or clonal expansion of strains carrying both genes. The replication of these experiments described here showed the same lack of selection of *pco*-positive *E. coli* with copper concentrations in feed (125 mM) similar to those used in the field (100–250 mM; [2]). Although this does not exclude some selection in the long term or the selection of other genes, no significant effect could be detected during a single feeding period. Since none of the 82 *pco*-negative isolates sequenced carried the *sil* operon (31 of which were from samples of animals receiving copper supplementation), it is also unlikely that the copper treatment would have selected for this latter operon alone. A *pco*–*tet*(B) positive association, as well as a negative association between *pco* and *bla*CMY, was observed previously by Agga et al*.* [2]; however, while the direction of association was similar here, the associations were not

significant (*p* = 0.344 and 0.087, respectively). We did see much higher levels of *pco* among isolates from pigs subjected to low doses of CTC (a dosage regimen for growth promotion purposes not permitted in the US since 1 January 2017), which may suggest that indirect selection of *pco* and *pco*-positive strains could occur when using tetracyclines in swine. In the present study, the prevalence of *tet*(A) was highest in the group receiving high-dose CTC (45.0% versus 22.5%) whereas the prevalence of *tet*(B) was highest in the group receiving low-dose CTC (77.5% versus 62.5%). This finding may help explain the relationship of tetracycline uses with *pco*, though such hypotheses are largely based on the previous findings of Agga et al*.* [2].

Analysis of genomic similarities between *pco*-positive isolates through cgMLST demonstrates that the *pco* genes are distributed across a variety of clonal lineages and do not cluster in only a few clear discrete groups of closely related isolates. The most frequent *pco* single-nucleotide polymorphism (SNP group C in Figures 1A and 2) is also present in several STs and unrelated clonal lineages. Both observations illustrate the active horizontal transfer of the *pco* gene cluster in *E. coli* populations. However, the associations between most of the other *pco* SNP groups and STs (Figure 2) or clonal lineages (Figure 1) suggest that both a combination of short-term or local clonal spread and broader long-term horizontal gene transfer (HGT) play a role in the distribution of this gene cluster in *E. coli* from the swine population examined. Similar to the *pco*–*sil* clusters, the *tet*(A), *tet*(B), and *bla*CMY genes which have been present in Enterobacteriaceae from farm animals in North America for several decades also appeared to be distributed randomly and did not cluster clearly together with *pco* genes in a discrete number of clonal lineages (Figure 1). Overall, these observations suggest that the positive and negative statistical associations observed between *pco* and *tet*(B) or *tet*(A) and *bla*CMY, respectively, do not rely on the expansion and contraction of a very limited number of major clonal lineages. This differs from the *bla*CTX-M-27 gene which was found mainly (11/12) in closely related isolates. CTX-M β-lactamases were reported in food animals much later in North America than in other continents [33] and may have emerged in swine in the US only recently. It may, therefore, be only in the early stages of its spread through HGT in bacteria from swine and still limited to a small number of clonal lineages.

The *pco* genes were located together with the *sil* cluster on a Tn*7*-like transposon structure [17,19,34] in all but one isolate in this study (KSC1031). The high transposition frequency of Tn*7* and related elements [29] may be an important reason for the distribution of the *pco*–*sil* cluster in a wide diversity of strains illustrated in the present study. Tn*7* transposons developed refined strategies to insert preferentially on mobile plasmids [29]. It may, therefore, not appear entirely surprising that the *pco* plasmid we were able to transfer by conjugation (pMRGN207) carried the full *pco*/*sil*/Tn*7*-like element, while pMRGN1031 missing the Tn*7* part of the element was not transferable. Coincidentally, the plasmid we were able to transfer was an IncHI2 plasmid, an incompatibility group already shown by others to carry *pco* genes in different geographic locations and bacterial species [17,18,35,36]. Tn*7* transposons also developed refined strategies to insert preferentially into the same selectively neutral *attTn7* chromosomal site located in proximity of the *glmS* gene [29]. However, the locations of the two chromosomal Tn*7*-like elements associated with the *pco*–*sil* cluster in the closed genome sequences generated with PacBio long reads show that this mobile element does not always insert in the same *attTn7* site or in the proximity of *glmS*. This may warrant further investigations on the transposition mechanisms of this Tn*7*-like transposable element. Together, these findings further stress the likely important role of IncHI2 plasmids and Tn*7*-like elements in the spread of the *pco*–*sil* gene clusters.

The overall structure of the region encompassing the *pco*–*sil* clusters was highly conserved and identical to pR478 [36] in two of the four isolates we investigated in detail (one plasmid-borne and the other chromosomal). This conserved region also included the *tns* gene cluster of Tn*7* and the intervening region between the *tns* and *sil* genes. This structure was described by others on several plasmids [17,19]. Insertions were present in the *pco*–*sil* clusters for the two other isolates. In one of them (chromosomal), three insertions were present in this region, but all were within open reading frames encoding putative proteins of unknown function, and were not affecting the *pco* or the *sil* gene clusters. However, in pMRGN1031, an insertion was disrupting the *pcoA* gene. This latter insertion would be expected to inactivate the copper resistance if a phenotype were detectable [14].

In addition to the loss of the *tns* gene cluster and parts of the genes upstream of the *sil* cluster already mentioned above, SNP analysis also showed that the *pco* genes in pMRGN1031 are clearly divergent from the majority of those from the other isolates of this study. This strongly supports the hypothesis that the *pco*–*sil* gene clusters on this plasmid have a longer or different evolutionary history than those found on other plasmids, and that parts of it may possibly be decaying.

The surprising initial lack of difference in susceptibility to copper between *pco*/*sil*-positive and *pco*/*sil*-negative isolates that we obtained in broth under aerobic growth conditions triggered further investigations under a variety of other conditions. Previous publications showed that the copper resistance phenotype of *pco*-positive isolates is inducible and can be triggered by preliminary incubation in subinhibitory concentrations of copper [13,36]. Subjecting our isolates to subinhibitory concentrations of copper similar to those described in these studies did not result in any change in copper MIC, and our isolates did not show any significant change in RNA transcription of the *pcoA* and *pcoD* genes after induction. Copper susceptibility of *E. coli* and *S. enterica* was tested by others with a variety of methods, including broth [17] and agar dilutions [21,37], as well as under aerobic [17,21] and anaerobic conditions [21,37]. No differences in copper MICs were observed by these authors between *pco*/*sil*-positive and -negative isolates under aerobic conditions, neither for *E. coli*, nor for *S. enterica*. However, differences were consistently observed for *S. enterica* when agar dilutions were used under anaerobic conditions [18,21]. Therefore, we also tested our *E. coli* and a few *S. enterica* isolates by agar dilution under anaerobic conditions. As expected, an evident dichotomization of MICs was visible under these conditions for *S. enterica*, but this was not the case for *E. coli*. These data are in agreement with results from others showing that copper resistance associated with the *pco* gene cluster is host-dependent [13]. The increase in MIC observed in *S. enterica* after transfer of a *pco*/*sil* plasmid from *E. coli* clearly confirmed this hypothesis.

Overall, the results from this study strongly suggest that the *pco*/*sil* gene clusters may have only a minor effect on copper MICs in typical wild-type intestinal *E. coli*, and may not represent a major selective advantage in this bacterial species in the gut of swine fed high concentrations of copper. Some of our findings are based on a relatively limited number of isolates, and confirmation on larger numbers of isolates is needed. *Escherichia coli* may represent a reservoir of mobile copper resistance determinants of potential importance for *S. enterica*. As illustrated here with pMRGN207 and by other researchers [17,18], transferable *pco*/*sil* plasmids concomitantly carry antimicrobial resistance determinants. These antimicrobial resistance determinants may help maintain these mobile plasmids, and indirectly, the *pco*–*sil* cluster in *E. coli* populations. Antimicrobial resistance may, in turn, be maintained and selected in *S. enterica* harboring these plasmids by the supplementation of feed with copper. Further animal experiments are needed to clarify the latter points. The role of IncHI2 plasmids in this context and the exact mechanisms and dynamics of transposition of Tn*7*-like transposons associated with the *pco*–*sil* gene clusters certainly also warrant further investigations, as do the respective roles and contribution of the *pco* versus *sil* genes in the observed copper resistance in *S. enterica*.

**Author Contributions:** H.M.S. and P.B. conceived and designed the experiments; H.M.S., R.G.A., T.G.N. and M.D.T. designed and performed the animal study; K.M.R., K.N.N., R.G.A. and G.C. performed the experiments; K.N.N., H.M.S., P.B. and G.C. analyzed the data; G.C., H.M.S. and P.B. wrote the original draft of the manuscript. All authors read and approved the final manuscript.

**Funding:** This work was supported by the USDA National Institute of Food and Agriculture, AFRI Food Safety Challenge Grant project #2013-68003-21257.

**Acknowledgments:** The contents of this manuscript are solely the responsibility of the authors and do not necessarily represent the official views of the USDA or NIFA.

**Conflicts of Interest:** The authors declare no conflicts 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**


© 2018 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 (http://creativecommons.org/licenses/by/4.0/).

### *Review* **Harnessing Rhizobia to Improve Heavy-Metal Phytoremediation by Legumes**

### **Camilla Fagorzi 1, Alice Checcucci 1,\*, George C. diCenzo 1, Klaudia Debiec-Andrzejewska 2, Lukasz Dziewit 3, Francesco Pini <sup>4</sup> and Alessio Mengoni 1,\***


Received: 28 September 2018; Accepted: 6 November 2018; Published: 8 November 2018

**Abstract:** Rhizobia are bacteria that can form symbiotic associations with plants of the Fabaceae family, during which they reduce atmospheric di-nitrogen to ammonia. The symbiosis between rhizobia and leguminous plants is a fundamental contributor to nitrogen cycling in natural and agricultural ecosystems. Rhizobial microsymbionts are a major reason why legumes can colonize marginal lands and nitrogen-deficient soils. Several leguminous species have been found in metal-contaminated areas, and they often harbor metal-tolerant rhizobia. In recent years, there have been numerous efforts and discoveries related to the genetic determinants of metal resistance by rhizobia, and on the effectiveness of such rhizobia to increase the metal tolerance of host plants. Here, we review the main findings on the metal resistance of rhizobia: the physiological role, evolution, and genetic determinants, and the potential to use native and genetically-manipulated rhizobia as inoculants for legumes in phytoremediation practices.

**Keywords:** soil bioremediation; heavy-metals; serpentine soils; serpentine vegetation; genome manipulation; *cis*-hybrid strains

### **1. Introduction**

Plants are colonized by an extraordinarily high number of (micro)organisms, which may reach numbers much larger than those of plant cells [1]. This is particularly evident in the rhizosphere, the thin layer of soil surrounding and influenced by plant roots, where a staggering diversity of microorganisms is present. The collective communities of plant-associated microorganisms are referred to as the plant microbiota, and include the microbial communities of the rhizosphere, as well as those of the external and internal (the endosphere) plant tissues (for examples see [1–4]). The rhizobiome refers specifically to the microbial community of the rhizosphere, and microbes from this community have been deeply studied for their beneficial effects on plant growth and health. These mainly include mycorrhizal fungi (AMF) and plant-growth promoting rhizobacteria (PGPR), with the latter including the nitrogen fixing legume endosymbiotic bacteria known as rhizobia [5]. Rhizobia are a paraphyletic group of nitrogen fixing bacteria belonging to the Alpha- and Betaproteobacteria classes. Rhizobia can penetrate plant tissues and establish an intracellular population within specialized tissue (known as a nodule) on the root (or stem in a few cases) of leguminous plants. Once inside the

cells, the rhizobia differentiate into forms known as bacteroids, which are able to perform nitrogen fixation (the formation of ammonia from di-nitrogen gas) [6]. This process, termed "symbiotic nitrogen fixation" (SNF), provides the plant with nitrogen to sustain its growth in nitrogen-deficient soils, and has been suggested as one of the factors contributing to the evolutionary success of the Fabaceae plant family [6]. Plant growth and crop yield in agricultural systems emerge as the net results of the interactions between the specific plant cultivar and its associated microbiome [7].

Heavy metals are naturally present in soils; however, their increase over certain thresholds has become a worldwide issue [8]. The major cause of heavy-metal contamination in soil is anthropogenic activities (i.e., atmospheric pollution, industrial and urban waste, mining, and some agricultural practices), while natural contamination is mainly due to weathering of metal-enriched rocks [9]. Plant-associated microbiomes play important roles in phytoremediation, allowing plants to thrive on contaminated soils, alleviating the stress associated with toxic levels of heavy-metals and metalloids (such as As), and increasing phytoextraction and phytostabilization [10–14]. Phytoextraction refers to the plants' ability to import soil contaminants through their roots, and to accumulate these compounds in the aboveground tissues [15]. Phytostabilization involves the immobilization of pollutants in the soil as a result of either their absorption and accumulation in the roots, their adsorption on the root surface, or their transformation within the rhizosphere into sparingly-soluble compounds [16]. In plants such as legumes, which are generally non-hyperaccumulating species, phytostabilization is likely the more relevant process when considering the remediation of contaminated soils [15–17]. Plant-associated bacteria may promote the chemical transformation, the chelation, or precipitation and sorption of heavy-metals [18] (Figure 1). For instance, some endophytic bacteria may reduce heavy-metal toxicity [19,20]. Improved growth and increased chlorophyll content were detected in several crop plants inoculated with siderophore-producing bacteria [19]. Additionally, enhanced plant biomass production and remediation has been observed in several hyperaccumulating plants following inoculation with rhizosphere or endophytic bacteria with plant growth promoting (PGP) capabilities [21], such as 1-aminocyclopropane-1-carboxylate (ACC) deaminase production (for detailed reviews, please see [11,12]).

The association between leguminous plants and symbiotic rhizobia has stirred the attention of researchers involved in the restoration of heavy-metal-contaminated sites [22]. The possibility to cultivate legumes on marginal and nutrient-poor soils thanks to the intimate association with PGPR, particularly with nitrogen-fixing rhizobia, has been seen as an opportunity to increase phytoremediation efficiencies while simultaneously reducing its costs [23]. Heavy-metals play central roles in symbiotic nitrogen fixation (see [24] for a review of on the role of metals in the symbiosis). Notably, the nitrogenase enzyme is dependent on a cofactor containing molybdenum and iron (FeMo-co), vanadium and iron (VFe-co), or two iron molecules (FeFe-co). There is also evidence for the role of nickel in the symbiosis. For instance, plants inoculated with a deletion mutant of the rhizobium *Sinorhizobium meliloti* lacking the *nreB*-encoded Ni2+ efflux system displayed increased growth under controlled conditions [25]. Additionally, a treatment with low doses of Ni2+ as the amendment was shown to stimulate nitrogen fixation and plant growth in soybean, and to increase hydrogenase activity in *Rhizobium leguminosarum* bv. *viciae* [26,27]. However, an excess of heavy-metals negatively impacts the symbiosis, reducing the number of symbiotic nodules, the rate of nodulation, and the rate of nitrogen fixation [28,29]. Consequently, in order to promote legume-based phytoremediation through the improvement of the host plant-symbiont partnership, there is a need to discover metal-resistant rhizobia and/or to manipulate existing rhizobial inoculants to increase their level of metal resistance.

In this review, we summarize the main findings on metal resistance in rhizobia: the physiological role, evolution, and genetic determinants of metal resistance, and the perspective to use native and genetically-manipulated rhizobia as inoculants for legumes in phytoremediation practices.

**Figure 1.** The multiple roles of bacteria in helping plants cope with heavy metals. Plant-associated bacteria may have various roles in both phytostabilization and plant growth. They may influence metal solubility by directly producing molecules for metal chelation (e.g., siderophores), or by influencing plant root growth, resulting in increased production of root exudates. Moreover, both rhizospheric and endophytic bacteria can positively affect plant growth by producing phytohormone molecules (e.g., auxins), alleviating plant stress (e.g., plant ethylene production), or through nitrogen fixation. This latter activity is especially relevant when leguminous plants and their rhizobial microsymbionts are considered. PGP: Plant growth promotion.

### **2. Legumes in Heavy-Metal Contaminated Areas**

The family Leguminosae (Fabaceae) is one of the most diverse among land plants and includes over 700 genera and 20,000 species [30]. Legumes have been proposed as relevant species for phytoremediation, largely due to their ability to colonize marginal lands and nutrient-poor soils [28,31]. In particular, legumes are relevant for phytostabilization, as only a few species have been found to be metal hyperaccumulators (e.g., some species of the genus *Astragalus* isolated in the Western United States are selenium hyperaccumulators) for phytoextraction [23,28,32]. Normally, the symbiosis with rhizobia is inhibited by high levels of heavy-metals in the soil, and genetic engineering techniques have been suggested to improve symbiotic nitrogen fixation under such harsh environmental conditions [33]. However, although such biotechnological proposals are interesting in terms of molecular dissection of the system and theoretical application, currently, there are a number of limitations to the use of genetically-modified microorganisms, including their free release in nature. Analyses on legumes from heavy-metal-contaminated soils have led to the discovery of naturally-resistant rhizobia, which could be used as inoculants in these extreme environments. However, a deeper investigation of leguminous plants growing in metal-enriched sites is required to improve legume-based phytoremediation.

### *2.1. The Serpentine Vegetation: A Source of Legumes Evolved on Heavy-Metal Rich Soils*

Serpentine rocks are an array of ultramafic rock types composed of a hydrous magnesium iron phyllosilicate mineral that originates from metamorphic alterations of peridotite and pyroxene with water. The soils derived from these rocks are characterized by: (i) high levels of nickel, cobalt, and chromium, (ii) low levels of N, P, K, and Ca, and (iii) a high Mg/Ca ratio [34]. This chemical composition strongly limits the growth of most plant species [35], as well as many microorganisms [3]. The presence of serpentine outcrops is scattered across the planet. Along a geological timescale, serpentine outcrops have prompted the evolution of peculiar plant adaptation mechanisms (such as metal hyperaccumulation [36]), which then gave rise to plant differentiation and speciation in a classical

"ecological islands" model [37,38]. Serpentine vegetation in temperate ecosystems includes several leguminous species from various genera, including *Lotus*, *Lupinus*, *Trifolium*, *Vicia*, *Melilotus*, *Medicago*, *Lathyrus*, *Ononis*, *Dorychnium*, *Chamaecytisus*, *Astragalus*, *Anthyllis*, *Cytisus*, and *Acmispon* [39,40]. Serpentine endemic legumes have also been reported, such as *Errazurizia benthamii* [41] in North America, and *Serianthes calycina* [42] in New Caledonia. The microbiomes associated with serpentine plants contain a fraction of microorganisms that appear to have specifically evolved functions to cope with toxic levels of metals present in the soil and in the plant itself [3]. Moreover, some of these microorganisms have been shown to be effective in promoting host plant growth in serpentine soil and, for metal hyperaccumulating plants, to increase metal translocation to the aerial part [43]. Consequently, rhizobia from serpentine endemic legumes (such as Ni-resistant bradyrhizobia from *S. calycina* [42]) may already be adapted to optimizing the fitness of their host in serpentine environments through a long-term natural selection process [44]. Serpentine endemic legumes may therefore represent an ideal source of rhizobia that are naturally highly-competent symbiotic partners in heavy-metal contaminated soils.

### *2.2. The Search for Heavy-Metal Tolerant Rhizobia and Their Use as Inoculants*

Legumes growing in contaminated areas such as mine deposits and serpentine soils have been a source of symbiotic rhizobial strains displaying resistance to heavy-metals, including Zn, Pb, and Cu [45–48]. Table 1 summarizes the main studies on the (positive) effects of rhizobial inoculation on the heavy-metal tolerance of host plants.

*Anthyllis vulneraria* is one of the most relevant legumes for isolating rhizobia that promote metal-tolerance by the host plant. *A. vulneraria* is a perennial herb from boreo-temperate climate areas in Europe, and it can be found colonizing rocky outcrops and establishing populations on heavy-metal (mainly Zn)-contaminated sites. *Anthyllis* is characterized by determinate nodules, where the meristematic activity disappears shortly after nodule formation, resulting in nodules of spherical shape. *Anthyllis* nodules contain a multilayer cortex: a glycoproteic parenchyma for diffusion, an endodermis, and the outer cortex, which mainly serves as a barrier against pathogens [49]. Nodule bacterial population of leguminous plants grown in Morocco metal-polluted soil displayed a great biodiversity, suggesting that, in these conditions, metal resistant non-rhizobia may efficiently colonize the nodules as endophytes [50]. This highlights the importance of heavy-metal resistance in rhizobia for the establishment of an effective symbiotic interaction in contaminated soils. *A. vulneraria* has been found to be associated with rhizobial symbionts from the genera *Mesorhizobium*, *Rhizobium*, and *Aminobacter*. These include novel rhizobial species, such as *Mesorhizobium metallidurans*, *Rhizobium metallidurans*, and *Aminobacter anthyllidis* [45,47–51]. Interestingly, these novel rhizobial species have so far been identified only in Pb-contaminated environments and not in unpolluted soils [47]. The symbiosis between *A. vulneraria* and its possibly exclusive metal-resistant bacterial species may provide the basis for the establishment of phytoremediation practices. This could involve the use of *A. vulneraria* metal-resistant germplasms, together with its specific natural rhizobial symbionts. Alternatively, the heavy-metal-resistant rhizobia isolated from *A. vulneraria* could be modified, either through laboratory-based experimental evolution studies [52] or direct genetic manipulation, to be capable of entering into an effective symbiosis with other host legumes.

Legumes of the genus *Medicago* have also been deeply investigated for their application in phytoremediation (see Table 1 and references therein). This is mainly because species from this genus are important forage crops for which cultivation techniques and genetics are well established, providing important advantages for future cost-effective applications [53]. Genetically-modified [54,55] and natural [56,57] inocula of *Sinorhizobium* (syn. *Ensifer*) *meliloti* and *Sinorhizobium medicae* [54] have been examined for their abilities to improve plant growth and metal accumulation in the presence of toxic levels of heavy metals such as Cu, Cd, and Zn. However, genetic manipulation is not absolutely required, as interesting results have also been obtained using indigenous *S. meliloti* and *S. medicae* strains directly isolated from contaminated soils [56,57]. For example, inoculation of *Medicago sativa*

plants, grown under field conditions, with wild *S. meliloti* and *S. medicae* strains resulted in active nodulation and the promotion of metal bioaccumulation within the root nodules [56,57]. These results suggest that the exploitation of natural rhizobia could be a valuable tool for promoting land restoration and phytostabilization by legumes.

Legume-based phytoremediation may also be improved through inoculation with a consortium of metal-resistant rhizobia and other PGP bacteria. In metal polluted soil, inoculation of *Lupinus luteus* with *Bradyrhizobium* sp. 750 in consortium with *Pseudomonas* sp. Az13 and *Ochrobactrum cytisi* Azn6.2 increased plant biomass by greater than 100% with respect to uninoculated plants [10]. In contrast, inoculation with only *Bradyrhizobium* sp. 750 increased plant biomass by only 30%. Similarly, co-inoculation of *M. lupina* with *S. meliloti* CCNWSX0020 and *Pseudomonas putida* UW4 resulted in larger plants and greater total Cu accumulation than inoculation with just *S. meliloti* CCNWSX0020 [55]. Inoculation of *Vicia faba*, *Lens culinaris*, and *Sulla coronaria* with consortia of rhizobia and non-rhizobia was also effective at improving plant growth and pod yield when grown in metal-contaminated soil [58]. Moreover, the inoculated *S. coronaria* accumulated significantly more cadmium than uninoculated plants [58]. These results highlight the potential for root-associated microbial communities to influence the success of phytoremediation by rhizobium-inoculated legumes.

It may be concluded that there is great biotechnological potential in increasing the phytoremediation capabilities of legumes by their associated rhizobia. This may be mediated through at least two mechanisms: (i) reducing the toxic effects of the metals, and (ii) promoting the growth of the plant through PGP activities.



### **3. Genetics and Genomics of Heavy-Metal Resistance in Symbiotic Rhizobia**

A deep understanding of the genetics and molecular mechanisms of metal resistance remains one of the main goals in environmental biotechnology, with the final aim of promoting the bioremediation (including phytoremediation) of contaminated soils. Table 2 reports the main studies evaluating the genetic determinants of heavy metal resistance in rhizobia. Such studies have most commonly identified the presence of efflux systems that increase metal tolerance by reducing the intracellular concentrations of the metal(s). However, studies employing genome-scale methods, such as transcriptome analyses and transposon mutagenesis, have demonstrated that the cellular response to metal stress involves an intricate genetic network.

Mechanisms mediating resistance to Co and Ni have been identified in many metal resistant rhizobia through the identification of orthologs of metal resistance genes characterized in *Cupriavidus metallidurans* CH34 [63,64]. A gene encoding a DmeF ortholog has been identified in *R. leguminosarum* bv. *viciae* strain UPM791 [65]. DmeF proteins belong to the cation diffusion facilitator (CDF) protein family, which form metal/proton antiport systems to translocate heavy metals across the bacterial membrane [66]. Mutation of the *dmeRF* operon in *R. leguminosarum* resulted in increased sensitivity to Co and Ni, but not to Zn or Cu [65]. The mutant also appeared to be somewhat less effective in symbiosis with pea plants, but not lentil plants, when grown with high concentrations of Co or Ni [65]. Further experiments demonstrated that *dmeR* encodes a Ni- and Co-responsive transcriptional regulator that represses expression of the efflux system in the absence of these metals [65]. Despite being considered a metal-sensitive strain, the *S. meliloti* strain 1021 encodes various metal homeostasis mechanisms, including the DmeRF system, several P-ATPases that are highly common in bacteria, and an ortholog of the *C. metallidurans* NreB protein [25,65,67]. Mutation of *nreB*, encoding a Ni2+ efflux protein, resulted in increased sensitivity to Ni, Cu, and low pH, but increased tolerance to urea osmotic stress [25]. The P1B-5-ATPase of *S. meliloti*, termed Nia (nickel iron ATPase), is positively induced by the presence of Ni2+ and Fe2+, and its expression is higher within nodules relative to free-living cells, which may prevent toxic levels of iron accumulation in the symbiosomes. The wild type protein and recombinants with a deletion of the C-terminal Hr domain have been used to understand the metal specificity of the P1B-5-ATPase family [67].

Genome-wide analyses have been used to investigate the genetics of the resistance mechanisms in *S. meliloti* strain CCNWSX0020, which is resistant to high levels of various heavy-metals (Cu, Zn, Cd and Pb). Gene mutation and transcriptome analyses have suggested the involvement of dozens of genes in the metal-resistance phenotypes of CCNWSX0020, including housekeeping genes [68–70]. Of particular note are the following three operons: the multicopper oxidase (MCO), CopG, and YadYZ operons. The MCO operon is highly expressed following exposure to Cu, and it encodes an outer membrane protein (Omp), the multicopper oxidase CueO, a blue copper azurin-like protein, and a copper chaperone involved in Cu homeostasis [70]. It was proposed that the CueO protein (showing 40% similarity with the CueO protein of *E. coli*) catalyzes Cu(I) oxidation in the periplasmic space, followed by the export of the excessive Cu(II) across the outer membrane [70,71]. The CopG operon consists of four genes: CopG, a CusA-like protein, a FixH-like protein, and a hypothetical protein. Mutation of any of the latter three genes resulted in elevated sensitivity to Zn, Pb, Cd, and Cu, although the mechanism of resistance of this operon remains unknown [70]. The CusA-like protein appears to be a highly-truncated ortholog of the CusA protein of the CusCBA Cu(I) efflux system of *E. coli* [72,73], and may act as a metal binding protein [70]. The FixH-like protein displays similarity to the FixH protein of the FixHGI membrane-bound system, a likely cation transporter that has been shown to be essential for symbiotic nitrogen fixation [74,75]. A FixH-like homolog is also encoded by the pSinB plasmid of *Ensifer* sp. M14 (formerly *Sinorhizobium* sp. M14), where it was also experimentally shown to be involved in metal resistance [76]. Deletion of the *yedYZ* operon resulted in increased sensitivity to Zn, Pb, Cd, and Cu [70]. This was the first report suggesting that YedYZ may be involved in heavy-metal tolerance. In *E. coli*, YedYZ forms a sulfite oxidoreductase [77], and expression of a homologous protein in *S. meliloti* 1021 is induced by taurine and thiosulfate [78]. Thus, the heavy-metal

resistance phenotype may be mediated through disrupting sulfite metabolism, which may influence antioxidant defenses against reactive oxygen species (ROS) generated by heavy metals [70].

Many scientists have used population genetics approaches to identify loci associated with heavy-metal resistance. This was achieved by performing genome-wide association studies on a population's pan-genome, considering allelic variations in the core genome (the set of genes shared by the members of the population), and gene presence/absence in the dispensable genome fraction (the set of genes present in only a fraction of the population). Genomic variants statistically associated with nickel adaptation were identified in a *Mesorhizobium* population using this approach [79]. A population of 47 *Mesorhizobium* strains, isolated from root nodules and soils with different levels of nickel contamination, was studied. Most of the variants associated with metal adaptation were found in the dispensable genome fraction. This work highlights that adaptation to heavy metal stress is likely driven predominately by horizontal gene transfer, and is not due to mutations of pre-existing genes.

Multiple studies have demonstrated that the genetic determinants of metal-resistance in rhizobia are relevant for phytoremediation purposes. Mutation of *ceuO*, *yedYZ*, and the *fixH*-like gene negatively impacted the *M. lupulina* nodulation kinetics of *S. meliloti* CCNWSX0020 in the presence of Cu and/or Zn [70], while deletion of the *cusA*-like gene had a negative effect, even in the absence of heavy metals. It was separately observed that *M. lupulina* plants inoculated with *S. meliloti* CCNWSX0020 strains with independent mutations in five Cu resistance loci were smaller than plants inoculated with the wild type, when grown in the presence of Cu [80]. Notably, *M. lupulina* plants inoculated with any of the *S. meliloti* CCNWSX0020 mutants mentioned above accumulated lower amounts of Cu and/or Ni [78]. Similarly, *Robinia pseudoacacia* plants inoculated with a *Mesorhizobium amorphae* 186 *copA* mutant accumulated 10–15% less Cu than plants inoculated with the wild type [81]; however, no effect on plant growth was observed.


**Table 2.** Genes for heavy-metal (and metalloid) tolerance in symbiotic rhizobia. A summary of the main genes whose function in tolerance was confirmed experimentally is reported.

### **4. Genomic Manipulation Strategies for Improving Legume Phytoremediation**

Various attempts have been made to increase plant growth in the presence of toxic metal concentrations through genetic modification of their rhizobial microsymbionts. One approach is to introduce new genes conferring heavy-metal resistance into the rhizobium. For example, inoculation of a genetically-modified *M. truncatula* line (which expressed a metallothionein gene from *Arabidopsis thaliana* in its roots) with wild type *S. medicae* resulted in elevated Cu tolerance [84]. Copper tolerance was further increased using a *S. medicae* strain expressing the *P. fluorescence copAB* Cu resistance genes [84]. Inoculation with the latter strain also resulted in elevated Cu accumulation in the plant roots [84]. Similarly, the introduction of an algal As(III) methyltransferase gene (*arsM*) into the chromosome of *R. leguminosarum* bv. *trifolii* produced a strain that was able to methylate and volatilize inorganic arsenic in symbiosis with red clover (with no negative impact on nitrogen fixation) [85]. A second approach is the insertion of genes in rhizobia to modulate phytohormone production, thereby reducing plant stress perception. For example, an ACC deaminase overproducing *S. meliloti* strain increased Cu tolerance and promoted plant growth of the host plant *M. lupulina* [86]. This result was probably due to reduced production of ethylene by the host plant, in turn decreasing stress perception. However, it should be kept in mind that a relatively high number of genes may contribute to the heavy-metal stress response [87–89]. Consequently, a multigenic, genome-wide approach should be considered when attempting to genetically modify competitive rhizobial symbionts to have increased heavy-metal tolerance. One possibility along these lines is the introduction of entire, large resistance plasmids from a non-symbiotic (but highly resistant) strain to a phylogenetically-related, symbiotic metal-sensitive strain. A candidate plasmid for such studies is the pSinA plasmid of the non-symbiotic *Ensifer* sp. M14, which was isolated from an As-contaminated gold mine [76,90,91]. The pSinA plasmid is a self-transmissible replicon with a broad host range. It harbors a genomic island with genes for arsenite oxidation (*aio* genes) and arsenite resistance (*ars* genes), and its transfer to other species results in increased arsenic resistance [90]. Transfer of the pSinA plasmid to closely-related rhizobia, such as *S. meliloti*, may result in the construction of As-tolerant legume symbionts for use in arsenic remediation. Subsequent acquisition of pSinA by other members of the rhizospheric microbiota may further stimulate phytoremediation of arsenic contaminated soils through reducing the arsenic toxicity (oxidizing arsenites to arsenates) and biofortification (increase of the arsenic resistance level) of the autochthonic or augmented microflora.

Similarly, elite and metal-resistant rhizobia may be obtained through combining within one strain genomic elements from the species pangenome. The genomes of most rhizobia are extremely diverse, and many rhizobia have a divided genome structure consisting of at least two large DNA replicons [92]. Although there can be numerous inter-replicon functional, regulatory, and genetic interactions [93–95], in some ways, each replicon in a divided genome could be considered as an independent evolutionary and functional element [94,96–98]. Recently, it was shown that the genome and metabolism of *S. meliloti* is robust to the replacement of the symbiotic megaplasmid with the symbiotic megaplasmid of a different wild-type isolate [99]. Therefore, it may be possible to construct "hybrid" strains (Figure 2) with a collection of replicons derived from various wild-type isolates, potentially allowing for the development of elite strains with improved multifactorial phenotypes (e.g., resistance to heavy-metals, high symbiotic efficiency, and competition toward the indigenous soil microbiota).

**Figure 2.** A synthetic biology-based proposal to increase rhizobial-mediated heavy-metal tolerance. Surveys of rhizobial phenotypic and genetic diversity in heavy-metal (HM) rich areas facilitates the discovery of strains (strain 1) with high levels of heavy-metal resistance. However, such strains may not be competitive or good nitrogen-fixers in the crops to be used for phytoremediation. The simultaneous transfer of a large collection of genomic determinants that contribute to HM tolerance, good PGP, and/or nitrogen fixation (N-fix) abilities between two or more strains (strain 2) could create hybrid strains (*cis-*genic strain) with improved features for application in the field for phytoremediation.

### **5. Conclusions**

In recent years, the number of studies related to the potential exploitation of rhizobium–legume symbioses for phytoremediation practices have increased enormously as a result of environmental emergencies. In this brief review, we have presented state-of-the-art studies on heavy-metal tolerant rhizobia, and on their applications in phytoremediation as legume symbionts. A large number of investigations have indicated that rhizobia, and especially heavy-metal resistant rhizobia, can increase legume heavy-metal tolerance and promote improved legume growth in metal-rich soils, thereby resulting in greater removal of heavy-metals from the soil. Heavy-metal resistant rhizobia have been isolated from the nodules of legumes grown in soils that are rich in heavy-metals as a result of geological (e.g., serpentine outcrops) or anthropic causes (e.g., mine deposits). Genetic and genomic studies of heavy-metal resistant rhizobia have shown that although relatively few genes act as the main player in tolerance, a much larger set of genes may be involved in maximizing fitness in heavy metal rich growth conditions. Some of these genes, such as the systems for Ni2+ efflux in *S. meliloti*, may also contribute to a linkage between metal homeostasis and nitrogen-fixation efficiency. As such, systems-biology approaches are required to develop an overall picture of heavy-metal resistance and the ways that we can increase and exploit it in biotechnology. It will also be important to keep in mind that the engineering of rhizobia should consider several additional aspects, including the rhizobial genotype, the host plant genotype, and the interactions between the rhizobium with the soil and root microbiota [100].

Going forward, we suggest that large-scale genome-manipulation approaches may be considered in developing rhizobial strains with elite phenotypes (e.g., high heavy-metal resistance, high nitrogen-fixation ability, high competitiveness, etc.) for use in phytoremediation applications. As a pre-requisite to such studies, it will be necessary to increase efforts at creating culture collections of rhizobial strains from contaminated areas, since the strains isolated from these environments is quite limited in number and in terms of host plant (see also [23]). Such efforts would benefit from exploring areas that have evolved peculiar flora, such as serpentine outcrops, maximizing the chance to find well-adapted strains. Whole genome sequencing, genome-scale mutagenesis (such as Tn-seq or INseq [101]), and metabolic modeling of these strains could then be employed to fully characterize the genomic basis for tolerance against the contaminants.

**Author Contributions:** Conceptualization, A.M., C.F., A.C. and G.C.D.; Writing-Original Draft Preparation, A.M., C.F., A.C. and G.C.D.; Writing-Review & Editing, A.M., C.F., A.C., G.C.D., L.D. and K.D.A.; Funding Acquisition, A.M.

**Funding:** This research was partially funded by Ente Cassa di Risparmio di Firenze, grant "2017-0719" and by the University of Florence, project "Dinamiche dell'evoluzione dei genomi batterici: l'evoluzione del genoma multipartito e la suddivisione in moduli funzionali", call "PROGETTI STRATEGICI DI ATENEO ANNO 2014" to A.M. A.C. was supported by a grant from Fondazione Adriano Buzzati-Traverso. G.C.D. was supported by the Natural Sciences and Engineering Research Council of Canada (NSERC) through a PDF fellowship. K.D. was supported by the European Molecular Biology Organization in the frame of the EMBO Short-Term Fellowship program [grant number 7376] and the National Science Center (Poland) in the frame of the Preludium grant No. 2016/23/N/NZ9/01655.

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

### **References**


© 2018 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 (http://creativecommons.org/licenses/by/4.0/).
