**4. Discussion**

### *4.1. Congruence of Taxonomic Information between eDNA Metabarcoding and CMI*

Many of recent studies have strived to profile and quantify taxonomic composition of plankton communities using either eDNA metabarcoding or CMI [14,24,56]. Among them, a few studies have reported a degree of disagreement between the two pronged identification methods [32,57]. In this respect, our study also presented some disagreement, between eDNA and CMI samples, in community composition (Figure 3a,b as well as in relationships of biotic information (Figure 3e,f). Some pieces of literature on eDNA monitoring have enumerated possible reasons to explain the discrepancy between the two identification methods. It is reported that the capacity of identification between molecular and morphological datasets could have mainly caused the disagreement [24,58]. That is, specimen identification can vary along accuracy of molecular reference databases [59]. Therefore, the establishment of well-curated databases of reference DNA sequences for identified specimens is essential in the field of eDNA metabarcoding to make the taxonomic information congruen<sup>t</sup> with CMI. Additionally, there is another concern with the drawback of eDNA metabarcoding associated with technical biases/difficulties, such as copy number variation in the process of polymerase chain reaction (PCR) [60]. Related to a primer, its amplification and binding affinity are critical factors to bring about taxonomic biases in eDNA detection [61–65]. In terms of sensitivity of species detection, CMI-based assessment is also subject to an unpredictable, but probably significant, bias due to the presence of cryptic species [66]. Particularly in our study, marine calanoid copepods, *Candacia*, were only detected by eDNA metabarcoding in a very low proportion of <5%. However, we also admit that taxonomic misclassification due to lack of expertise and difficult to impossible taxonomic determination rather than just cryptic species also causes bias.

With these concerns in mind, our results on the community composition might be influenced by the primer amplification effects (Figure 3a,b). The previous related research reported some technical biases against low-abundant taxa in delineating microbial diversity [63]. In fact, while Cnidaria comprised 3% in CMI, they were 14% in eDNA samples. Likewise, Mollusks occupied 5% in CMI, but did 10% in eDNA samples (Figure 3a,b). In contrast to these differences, the compositional changes between the two samples were not significant for the rest low-abundant taxa containing Annelida, Chaetognatha, Echinoderm, and fish (Figure 3a,b). Namely, our results showed that low-abundant taxa could always be overestimated in eDNA metabarcoding. These results of difference and variation might be associated with several reasons. Firstly, eDNA metabarcoding is highly sensitive to detecting species. This high sensitivity is advantageous in identifying low-abundant/cryptic species. However, it can also lead to variations originating not only from organisms that are a few miles away from the sampling site but also from food items hidden in organisms. In addition, abundance estimates are possibly erroneous because many small organisms could generate the same number of sequence reads as a few large organisms. Secondly, although it is relatively unexplored, the copy number variation derived from the technical bias during the PCR process is another factor leading to inaccurate estimation [60,63]. Lastly, CMI is also error-prone depending on expertise/experience and specimen size. Therefore, we notice that eDNA may not be able to fully present diversity yet.

Despite some discrepancy between the eDNA and CMI samples, one highlighting point is the relational consistency in richness and diversity. Traditionally, plankton community assessment on richness and diversity has been complicated and time-consuming. However, compared to CMI, the eDNA metabarcoding also presented a positive relationship between richness and diversity (Figure 3c,d). Furthermore, while CMI exhibited a shorter range of richness and diversity (Figure 3d), the eDNA metabarcoding displayed a wider range (Figure 3c). Although its accuracy is another issue as previously mentioned, therefore, our study explicitly accounts for better capability of detection and identification by means of metabarcoding skills.

### *4.2. Potential Values of an eDNA Approach for Biological Monitoring and Assessment*

Most conventional approaches for biological monitoring and assessment were based primarily on microscopy. Due to the time consumption and expertise requirement for identification in species level, the current environmental monitoring and assessment of community composition highly demand new alternative technologies in terms of cost efficiency. In this regard, eDNA metabarcoding has been deemed as a promising tool for species detection and identification [58]. Particularly in plankton research, the eDNA approach helped reveal a previously hidden taxonomic richness for diverse meroplankton, such as Bivalvia, Gastropoda, and Polychaeta, which are relatively hard to identify in CMI [67]. Our study also advocates that a wider variety of species, including the aforementioned meroplankton, were detected in the eDNA samples (Table 1).

At the same time, however, we recognize that some discrepancies of abundance between metabarcoding and CMI have been contentious [62,68,69]. This discrepancy may limit the scope of eDNA research, which is also associated with the varying lengths of time to eDNA degradation in response to ambient environmental conditions [11,59,70,71]. Nevertheless, several studies have found a significant relationship between determining relative or rank abundance, highlighting the potential value of eDNA, though the variation inherent in environmental samples makes it difficult to quantify [12,32].

In our study, we found some clear patterns of coastal plankton communities in time (early vs. late summer) and space (inner vs. outer bay). From our analysis using eDNA and CMI samples, the main features of Gwangyang Bay could be characterized more clearly: (*i*) inner bay in early summer; (*ii*) outer bay in early summer; (*iii*) inner bay in late summer; and (*iv*) outer bay in late summer (Figure 5). Each characteristic was explicitly delineated by the prominent species. For example, in Gwangyang Bay, Asterozoa were predominant in early summer, *Sagitta* spp. were abundant in the inner bay, and zooplankton *Centrophages* spp. were in late summer. Dinoflagellates were separately characterized by *Noctilluca* spp. in early summer and by *Hematodinium* spp. in late summer. Although we did not use the eDNA samples solely, our spatiotemporal analysis presented the main plankton community features based on both eDNA and CMI samples. The CMI samples in addition to the eDNA make our pattern analysis more robust and reliable, because the predominant plankton would be separately presented if the eDNA and CMI samples differed significantly from each other. Thus, the information from the eDNA and CMI samples was highly similar given the subtle discrepancy of richness, diversity and their relationships (Figure 3c,d). However, we stress that the eDNA samples were good enough to delineate spatial and temporal characteristics of coastal plankton communities in Gwangyang Bay (Figure 5).


**Figure 5.** Main characteristics of marine plankton communities in Gwangyang Bay. The groups with % present relative abundance (derived from eDNA), and those without % present absolute abundance (derived from CMI).

In sum, we learn from our study that eDNA metabarcoding can be an effective alternative to monitor and assess entire communities from even a single sample. In addition, the eDNA metabarcoding is highly beneficial in terms of sensitivity for cryptic species and cost-efficiency for morphological identification. At the same time, however, our study also put emphasis on bio-assessment that can be affected by some information discrepancy of richness and diversity between eDNA and CMI samples. Hence, eDNA-based research should be further investigated to make the derived results become more stable. The current limited capacity of eDNA-based research is probably subject to a grea<sup>t</sup> deal of uncertainties associated with amplification, reference database, NGS-sequencing, and eDNA degradation [57,71]. To this end, we stress that eDNA research should be more active in order to shed light on ecosystem monitoring and assessment in future.

**Author Contributions:** Conceptualization: D.-K.K., K.P., H.J. and I.-S.K.; Methodology: D.-K.K. and K.P.; Formal Analysis: D.-K.K. and H.J.; Investigation: D.-K.K., K.P. and H.J.; Resources: I.-S.K.; Writing—Original Draft Preparation: D.-K.K.; Writing—Review and Editing: D.-K.K. and I.-S.K.; Supervision: I.-S.K.; Project Administration: I.-S.K.; Funding Acquisition: I.-S.K.

**Funding:** This research was funded by the National Research Foundation of Korea, gran<sup>t</sup> number NRF-2018R1A6A1A03024314.

**Conflicts of Interest:** The authors declare that they have no competing interests.

### **Appendix A Rarefaction Curves of the 18S rDNA V9 Samples in May (A) and September (B)**
