Metabarcoding as a quantitative tool for estimating biodiversity and relative biomass of marine zooplankton

as a quantitative tool for estimating biodiversity and relative biomass of marine zooplankton. Although metabarcoding is a well-established tool for describing diversity of pelagic communities, its quantitative value is still controversial, with poor correlations previously reported between organism abundance/biomass and sequence reads. In this study, we explored an enhanced quantitative approach by metabarcoding whole zooplankton communities using a highly degenerate primer set for the mitochondrial marker cytochrome oxidase I and compared the results to biomass estimates obtained using the traditional morphological approach of processing zooplankton samples. As expected, detected species richness using the metabarcoding approach was – times higher compared to morphological processing, with the highest diﬀerences found in the meroplankton fraction. About % of the species identiﬁed using microscopy were also recovered in the metabarcoding run. Within the taxa detected using both approaches, the relative numbers of sequence counts showed a strong quantitative relationship to their relative biomass, estimated from length-weight regressions, for a wide range of metazoan taxa. The highest correlations were found for crustaceans and the lowest for meroplanktonic larvae. Our results show that the reported approach of using a metabarcoding marker with improved taxonomic resolution, universal coverage for metazoans, reduced primer bias, and availability of a com-prehensive reference database, allow for rapid and relatively inexpensive processing of hundreds of samples at a higher taxonomic resolution than traditional zooplankton sorting. The described approach can therefore be widely applied for monitoring or ecological studies.


Introduction
Mesozooplankton are critical components of the world oceans, providing an essential link from the ocean's microbial primary and secondary producers to the upper levels of the food web. Monitoring zooplankton communities provides us with critical insight on the state of the pelagic ecosystem, as well as the implications for consumers, including harvestable species. Compared to organisms in other marine and terrestrial habitats, most planktonic species are characterized by relatively short generation times and react very rapidly to shifts in the physical environment, acting as first sentinels of climate change (Richardson, 2008;Beaugrand et al., 2009).
Additionally, since plankton drifts passively with currents, distinct planktonic communities and individual species can serve as useful markers of currents and water mass types (Beaugrand et al., 2002). Monitoring of zooplankton communities is a common way to quantify the feeding stocks of higher trophic levels, including commercially valuable species (Orlova et al., 2002). It can also be used to assess the general state of the ecosystem, such as introduction of alien species (Couton et al., 2019;Sepulveda et al., 2020) or changes in the community due to anthropological forcing (Andújar et al., 2018). The need for such surveys globally is growing, as marine systems are placed under increasing strain through development of shipping and industry, pollution, overfishing, and C International Council for the Exploration of the Sea 2021. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited.  E.A. Ershova et al. changing climate patterns (Chiba et al., 2018;Klunder et al., 2020).
Although zooplankton is easy to sample from both small and large vessels, and the methods of collection do not require special skills or expensive equipment, the post-processing of samples continues to be a bottleneck in most zooplankton monitoring surveys. For example, a recent study examining long-term ecosystem shifts in the Norwegian Sea included more than 1500 bulk zooplankton biomass measurements, yet merely one hundred samples were analysed taxonomically due to "limited capacity at the laboratory to analyze samples" (Toresen et al., 2019). Despite significant advances in the development of optical, acoustic and molecular approaches in zooplankton ecology, these methods still often lack sufficient power to discriminate between taxa, and/or are not quantitative. Microscopic processing of preserved zooplankton samples remains therefore the main method for quantitative assessment of zooplankton communities at the species level. This time-consuming approach requires a trained taxonomist, and the processing of a single sample can take from several hours to several days. Additionally, this method is inadequate for quantifying organisms that deform or disintegrate in fixatives (i.e. gelatinous plankton) or those that lack sufficient morphological features for visual identification (many early larval stages of both planktonic and benthic animals). The data produced via manual sorting are also highly prone to human bias, with the education and background of the person performing the analysis playing a key role. As a result of the high time and expertise requirements, many months or years generally pass between sample collection and data acquisition, which may cause significant delays in management actions. This drives the need for the development of rapid, alternative methods of bulk assessment of planktonic communities.
In recent years, deoxyribonucleic acid (DNA) barcoding of zooplankton taxa has greatly reduced the time and expertise needed for zooplankton identification by taking advantage of the divergence in DNA sequences across taxa (Hebert et al., 2003). Advances in DNA-based identification have challenged some established paradigms in zooplankton ecology, such as species distribution ranges (Choquet et al., 2017), and highlighted the widespread existence of cryptic or pseudo-cryptic species complexes in the marine realm (Miyamoto et al., 2010;Burridge et al., 2015;Kolbasova et al., 2015;Cornils et al., 2017;Ershova, 2020;Ershova et al., 2021). However, most molecular studies to this day have focused on individual species or groups, and biodiversity or community analyses via metabarcoding are only now becoming widely applied in zooplankton research (Bucklin et al., 2016(Bucklin et al., , 2019Zhang et al., 2018;Santoferrara, 2019;Questel et al., 2021). Unlike optical-based technologies, metabarcoding can potentially detect all species in a community, regardless of developmental stage or preservation of distinguishing features. Metabarcoding can therefore be used for the rapid detection of shifts in biodiversity and community composition, for monitoring of rare species that are unlikely to be captured in sufficient numbers to be identified visually, and for the detection of invasive species. One challenge that has limited the application of metabarcoding is finding a suitable DNA target region and a set of primers to amplify it. Ideally, the primers need to be truly universal, binding to all taxa equally, but the amplified DNA regions themselves need to be sufficiently genetically divergent across species while conserved within species, to allow species-level identification. Finally, to ensure the identification of all taxa present, the target DNA region needs to have a complete database of reference sequences, containing all species that may be found in the community. To this day, the majority of metabarcoding plankton studies have used various regions of the 18S gene (Bucklin et al., 2016), the primer-binding sites for which are well conserved across most taxa, but also provide relatively poor taxonomic resolution for metazoan groups (Mohrbeck et al., 2015). Other commonly used markers include regions of the nuclear 28s rRNA and the mitochondrial 16S rRNA gene (Bucklin et al., 2016). Although the mitochondrial cytochrome oxidase I (COI), the commonly used "barcoding" gene for metazoans, is generally too long for next-generation sequencing, good success has been demonstrated using smaller fragments of this gene, such as the one amplified by the Leray primer set (Leray and Knowlton, 2015).
Although metabarcoding is now a well-established tool for describing diversity of marine biological communities, its quantitative value is still controversial, with poor correlations previously reported between organism abundance and sequence reads (Bucklin et al., 2016;Lamb et al., 2019;Santoferrara, 2019;van der Loos and Nijland, 2020). Since the size of planktonic organisms spans ten orders of magnitude, the amount of DNA per organism is extremely variable, with even organisms of the same species often having vastly different numbers of DNA copies at different life stages. Better correlations have been obtained for biomass within some groups of organisms, typically, copepods (Clarke et al., 2017;Yang et al., 2017). In addition to an initial discrepancy in DNA copy numbers, extraction and polymerase chain reaction (PCR) methods introduce additional biases, resulting in better DNA recovery and amplification of some organisms than others. Given the exponential nature of PCR, these errors can compound to a very high degree.
In this study, we aimed to establish a relatively inexpensive and rapid protocol that would amount in a maximum recovery of diversity in environmental zooplankton samples, as well as provide quantitative information on relative biomass of organisms in the sample. To accomplish this, we used the highly degenerate modification of the Leray primers (Leray and Knowlton, 2015), Leray-XT (Wangensteen et al., 2018), to amplify a 313-base pair (bp) region of the COI gene in whole zooplankton samples. Degeneration ensures high attachment across multiple groups of metazoans, and the availability of the Barcode of Life Database (BOLD) allows for a good rate of success in species-level identification, despite the existing gaps in databases for marine taxa (Wangensteen et al., 2018). The results of metabarcoding using the Leray-XT primer set have been shown to reflect the quantitative composition of rocky reef benthic communities (Wangensteen et al., 2018), freshwater invertebrates (Elbrecht and Leese, 2015), and trophic contents (Siegenthaler et al., 2019), but have not yet been tested on marine zooplankton. Simultaneously, we processed the sample using traditional microscope sorting and estimated the biomass of all organisms present using published length-mass relationships. We hypothesized that (i) metabarcoding will recover a higher diversity of zooplankton, including most of the species identified via microscope sorting, and (ii) the relative proportion of sequence reads will be correlated with estimated relative biomass of the respective organisms. As such, this method could allow rapid and relatively inexpensive bulk processing of samples while limiting human bias, and provide a valuable tool in zooplankton monitoring studies.

Material
We used nine zooplankton samples from five locations in the Arctic and north Atlantic (Figure 1, Supplementary Table 1)-in the northern Barents Sea (one sample), Svalbard fjords (two samples), and north Norwegian fjords (six samples). The samples were collected using a WP2 net (mesh size 64 or 180 μm) and preserved with 96% ethanol. Samples were split quantitatively into two fractions. The first half was saved for traditional sorting and the other for molecular analysis. Any large (>1 cm) organisms in the sample were counted and measured prior to splitting.

Visual identification and biomass estimation
The subsample was washed from the ethanol with filtered seawater, after which it was split into consecutively smaller fractions using a Folsom plankton splitter until approximately 100 individuals of the most common taxa remained in the final split. Increasingly larger fractions were examined for rarer groups until a total of approximately 500 individuals were counted under a Leica stereomicroscope at 25-40x magnification. The entire sample was screened for larger organisms such as chaetognaths or jellies. Each counted organism was measured using the ZoopBiom digitizing system (Roff and Hopcroft, 1986) and its biomass in milligram dry weight was estimated using a length-weight regression known from published literature for this or a similar species (Hopcroft et al., 2010;Ershova et al., 2015). Pseudocalanus spp. and Microcalanus spp. were iden- E.A. Ershova et al. tified at the genus level. Calanus glacialis and Calanus finmarchicus were identified based on prosome length and pigmentation as described in Unstad and Tande (1991). Meroplanktonic larvae were identified at the phylum or class level.

Molecular analysis
The second half of the sample was homogenized in a 2000 W blender for 3 minutes, then the excess ethanol was removed using centrifugation and three 0.3 g subsamples of the homogenate were taken for DNA extraction. Extraction was done using the Power-Soil DNA extraction kit (Qiagen) according to the manufacturer's protocol. Strict measures to avoid sample cross-contamination were observed, and all instruments were cleaned with 10% bleach and rinsed in distilled water between samples. Since DNA extracted from plankton net samples is organismal-DNA (not extraorganismal DNA), less strict measures to avoid contamination may be taken compared to eDNA samples (Rodríguez-Ezpeleta et al., 2021). A ∼313 bp fragment from the 5' region of the COI gene was amplified using the Leray-XT primers (Wangensteen et al., 2018), which include the forward primer mlCOIintF-XT 5"-GGWACWRGWTGRACWITITAYCCYCC-3" and reverse primer jgHCO2198 5"-TAIACYTCIGGRTGICCRAARAAYCA-3" (Geller et al., 2013). Both ends of the primers contained sample tags of 8 bp and a variable number (2-4) of leading Ns for improving sequence diversity. Amplification was conducted in a single PCR using Am-pliTaq Gold DNA polymerase (Applied Biosystems), with 1 μl of each 5 μM forward and reverse tagged primers, 3 μg of bovine serum albumin, and 2 μl of extracted DNA in a total volume of 20 μl per sample. The PCR protocol consisted of a denaturing step of 10 min at 95ºC, followed by 35 cycles of 94ºC for 1 min, 45ºC for 1 min, and 72ºC for 1 min, and a final extension of 5 min at 72ºC. Two PCR blanks were added to the pool of multiplexed samples, which yielded only a few metabarcoding reads (<30 reads/blank sample). The PCR products were purified using Minelute PCR purification columns (www.qiagen.com) and pooled into a single library. The Illumina library was prepared from the DNA pool using the NextFlex PCR-free library preparation kit (Perkin-Elmer). This protocol incorporates Illumina adapters using a ligation procedure without any further PCR step, thus minimizing biases. The resulting library was sequenced in an Illumina MiSeq using 1 2 of a V3 2×250 bp kit (Illumina).

Bioinformatics
The first quality filtering steps of the bioinformatics pipeline were conducted using OBITools v.1.01.22 (Boyer et al. 2016). Paired-end reads were aligned with illuminapairedend and reads with alignment score >40 were kept. Demultiplexing and removal of primer sequences were done using ngsfilter. Reads with length between 299 and 320 and without ambiguous nucleotides were selected using obigrep and dereplicated using obiuniq. The uchime_denovo algorithm (Edgar et al., 2011)) implemented in vsearch v.1.10.1 (Rognes et al., 2016) was then used to remove chimeric sequences.
Step-by-step clustering was performed in SWARM 2.1.13 (Mahé et al., 2015) using a distance value of d = 13 to cluster individual sequences into molecular operational taxonomic units (MOTUs). This distance value has previously been used to cluster similar datasets using the same COI fragment (e.g. Bakker et al., 2019;Antich et al., 2020;Atienza et al., 2020). After removing singletons (MOTUs with abundance of one read), taxonomic assignment of the representative sequence of remaining MOTUs was then performed using ecotag (Boyer et al., 2016) against DUFA-Leray v.2020-06-10, a custom reference database (publicly available from github.com/uit-metabarcoding/DUFA), which included Leray fragment sequences extracted from BOLD and Genbank, completed with in-house generated sequences. The resulting dataset was curated for putative pseudogene sequences using the alogrithm LULU (Frøslev et al., 2017). The next refining steps consisted of removing MOTUs assigned to Prokaryotes and taxonomy check of selected MOTUs using BOLD (www.boldsystems.org). A species level identification was assigned with a minimum of 97% similarity, with the exception of Parasagitta elegans, which can have extremely high within-species divergence of the mitochondrial genome (Marlétaz et al., 2017) and for which a similarity cut-off of 90% was used. A barcoding gap analysis was performed for the most common planktonic metazoan taxa with enough intra-species information in BOLD (Supplementary File F1). This analysis confirmed similar behaviours for the whole Folmer barcode fragment (∼673 bp) and the Leray fragment (∼313 bp). Therefore, identical similarity thresholds can be used for both markers for a species-level assignment match. All MOTUs belonging to clearly non-planktonic organisms (i.e. insects and mammals) were removed. Finally, only MOTUs observed in a minimum of two sample replicates and accounting for at least 0.01% of the total reads of any sample were kept in the final dataset.

Data analysis
All analyses were performed in R (R Development Core Team, 2011). The recovery of biodiversity via metabarcoding was estimated by rarefaction curves using the package "vegan" (Oksanen et al., 2016). Relative biomass and relative sequence counts (%) were used in analyses. These values were fourth-root transformed to account for the very large spread of the biomass values and the exponential nature of PCR in the sequencing data. Overall community structure and replication was investigated using non-metric multidimensional scaling (nMDS) on Bray-Curtis dissimilarities of fourth-root transformed relative sequence counts using "vegan" (Oksanen et al., 2016). Relationships between relative abundance, biomass, and sequence counts (%) of both broad taxonomic categories and individual species across samples and within each individual sample were examined using simple linear regressions, with a Bonferroni correction to the p-value when multiple comparisons were included (p adjust ). Significance was assumed at p < 0.05.

Sequencing summary
The MiSeq run produced 4082508 paired-end raw reads for the multiplexed library. After all quality filtering steps, the final dataset consisted of 3418499 metabarcoding reads, clustered into 477 eukaryotic MOTUs. The sequencing depth per sample ranged from 45109 to 238991 reads, with an average value of 128840 reads per sample. Rarefaction curves (Supplementary Figure 1) suggest that this was enough to recover most, although not all diversity, although some individual station replicates failed to reach an asymptote.

Replication
The nMDS ordination (Figure 2) showed the three extraction replicates of each sample to be very similar in composition, with differences between stations generally greatly exceeding any within-sample variability. Despite limited sample size, the ordination also showed grouping within each location. On a broader spatial scale, the two Svalbard fjords were similar to each other while the two Norwegian Sea fjords were more divergent.

Overall biodiversity
A total of 48 unique taxa were identified visually (Supplementary  Table 2). Among sequences that accounted for at least 0.01% of total reads of any sample, 477 MOTUs were identified, which corresponded to 189 unique taxonomic categories (Supplementary Table 3). In total, 243 of the MOTUs were identified to species, 22 to genus, and 27 to at least phylum level, while the remainder (188) were grouped into an unknown category. Other large groupings of MOTUs into a single taxonomic category included Parasagitta elegans (37 MOTUs), and unknown Calanoida (32 MOTUs). The number of unique taxa per station identified using morphological identification was 20-30; metabarcoding resulted in 80-130 unique categories per station (Figure 3a). Metabarcoding provided the largest gains in biodiversity compared to morphological identification for the meroplankton fraction, which expanded from 3-5 categories per station to 100 or more, followed by small copepods and cnidarians. Of the 48 morphologically-identified taxa, 36 were also identified via metabarcoding. The taxa that failed to appear in the metabarcoding data included the appendicularians Oikopleura vanhoeffeni and Fritellaria borealis, the pelagic polychaete Tomopteris helgolandica, the pteropod Clione limacina and several species of copepods. With the exception of C. limacina, all of the excluded taxa were missing from the reference databases. Among the top-ten dominant taxa by biomass and by sequence counts, five to eight overlapped between the two methods (mean 6.9 ± 1.1); of the top-10 dominant taxa by abundance four to seven overlapped (mean 5.4 ± 1.0) (Figure 3b, Supplementary Table 3). Some of the most common species identified by morphology that consistently did not show up in the top-ten metabarcoding results include the copepods Microsetella norvegica, Acartia spp. and Metridia spp., while metabarcoding regularly identified among the dominant taxa Ctenophora and meroplanktonic groups, which failed to show up in the top-ten dominant lists via abundance and biomass.

Composition of biomass vs. sequence counts
The proportional contribution of each broad taxonomic group was very different among the metabarcoding, abundance, and biomass data ( Figure 4). Abundance was heavily dominated by small copepods, with other groups contributing 25% or less, whereas large copepods dominated the biomass, composing 20-95% of the sample. The sequence reads were more evenly distributed between taxa, with small and large copepods, chaetognaths, and meroplankton being significant contributors. Despite the differences in absolute values, strong significant correlations (p ≤ 0.02, Figure 5) were found between relative proportions of biomass (fourth-root transformed) and sequence reads for most taxonomic groups present in sufficient numbers (with the exception of Pteropoda). However, the ratios between sequence reads and biomass were generally not following 1:1 ( Figure 5). The large copepods were consistently underrepresented in the sequencing data relative to the biomass estimates, while small copepods and meroplankton were consistently overrepresented. Abundance numbers were poorly correlated to metabarcoding data, with only euphausiids showing a significant correlation between transformed % abundance and % sequence reads (p = 0.013)

Species-level correlations
We observed strong correlations between estimated % biomass and % sequence reads of all taxa that were registered using both  methods (examined at the highest common taxonomic resolution), both within each individual sample and when all samples were pooled together (R 2 = 0.59, p adjust < 0.0001, Figure 6). The correlations were strongest for copepods (R 2 = 0.70, p adjust < 0.0001) and euphausiids (R 2 = 0.78, p adjust < 0.0001), and weakest for meroplankton (R 2 = 0.52, p adjust < 0.0001). Once again, some species/taxonomic categories were consistently overrepresented in the biomass estimate vs. the metabarcoding (i.e. Calanus spp.), while others were consistently underrepresented (i.e. Polychaete larvae) in the biomass.

Case study: pseudo-cryptic taxa
We examined the relative composition of three common pseudocryptic copepod genera, Calanus spp., Pseudocalanus spp., and Microcalanus spp. (Figures 7 and 8) in the metabarcoding data. Calanus was among the dominant groups both in the metabarcoding and biomass data, and was represented by the Arctic C. glacialis and Calanus hyperboreus and the boreal C. finmarchicus and Calanus helgolandicus. Overall, the results of visual and molecular identification revealed remarkably similar patterns (Figure 7). Calanus finmarchicus dominated at most locations both in terms of estimated biomass and sequence counts, composing 35-90% of all Calanus DNA, followed by C. glacialis. Calanus hyperboreus was detected visually at only the three Arctic stations and one Balsfjord station in very low numbers, and was also less abundant in the sequence counts (with the exception of B34 in the Barents Sea). Surprisingly, its highest relative contribution was not in the Arctic locations, but in Ramfjord in November, where it was not detected visually. Calanus helgolandicus was the rarest of all Calanus, and was only observed via metabarcoding in Balsfjord and Ramfjord.
Although we did not identify the species visually within Pseudocalanus and Microcalanus genera, we anticipated more than one species to be present. We found Pseudocalanus to consist of four species: the Arctic Pseudocalanus acuspes and Pseudocalanus minutus, and boreal Pseudocalanus moultoni and Pseudocalanus elongatus (Figure 8a). In the Norwegian fjords, all four species were present, with the relative proportions changing seasonally. Pseudocalanus acuspes was the dominant species in all samples except the three Balsfjord stations and Ramfjord in December, where P. moultoni dominated. Pseudocalanus elongatus was observed only in the two Norwegian fjords, where its contribution was comparable with that of the other species. Pseudocalanus minutus was the least common species, with its highest contribution observed in the Barents Sea. Within the Microcalanus genus, only M. pusillus was observed at all examined locations.

Case study: meroplankton
Above we already showed that estimated biomass of meroplankton using visual identification was correlated with numbers of sequence reads, but the taxonomic resolution between the two methods was vastly different (Figure 3a and b). Seven categories of meroplankton were identified using microscopic sorting (Supplementary Table 1). A total of 118 benthic species belonging to 12 taxonomic groups were observed in the DNA sequences, making meroplankton the group dominating the metabarcoding dataset in terms of biodiversity. Together, meroplankton accounted for 12% of all sequence reads and 65% of the total unique taxa observed. Benthic taxa not detected using morphological identification but present in the DNA pool included members of Ascidiacea, Porifera, Cnidaria, and Caudofoveata. The most abundant groups were polychaetes, echinoderms, and bivalves, with polychaetes being the     dominant group both in terms of diversity and sequence numbers, accounting for half of the total meroplankton species and 7.5% of total sequence reads (65% of meroplankton sequences). The highest abundance and diversity of meroplankton was observed in the Balsfjord samples. An example of the species composition of echinoderm larvae is shown on Figure 8b. For example, the echinoderm larvae in Billefjorden were heavily dominated by Ophiocten sericeum, whereas this species was absent or rare in all other samples. The differences between the three Balsfjord samples (BF1-3) are notable, with Ophiacantha bidentata present in only one, and Asterias rubens in only two samples, despite the fact that these samples were collected in the same location and time. In the Ramfjorden samples (RF), the two winter samples were dominated by the holothurian Labidoplax buskii, but in the March sample it is replaced by Strongylocentrotus pallidus.

Discussion
Metabarcoding has been widely applied in ecology to investigate the biodiversity of microbial, invertebrate, and vertebrate communities, as well as food web interactions between them Bucklin et al., 2016;Jakubavičiute et al., 2017;Zhang et al., 2018;Couton et al., 2019). Within the zooplankton samples examined in our study, metabarcoding yielded similar, and in many groups vastly better taxonomic resolution, than traditional microscopic sorting. Metabarcoding identified 75% of the species identified by microscopic sorting and we expect this number to continue going up as additional species are added to the sequence ref-erence databases. Since monitoring studies often focus on the most common species, assuming them to be the most ecologically relevant, it is worth noting the 50-80% (but not 100%) overlap in the top-ten most common species in the metabarcoding and morphological approaches, both when looking at abundance and biomass ( Figure 3b). It is notable that among the discrepancies between the two methods, the taxa that showed up in the top-ten of metabarcoding dataset included those that are traditionally underrepresented using morphological analysis, such as ctenophores and meroplankton.
Although metabarcoding is increasingly used as a tool in zooplankton ecology to investigate biodiversity, so far few studies have applied it as a quantitative tool in real environmental zooplankton samples. Nonetheless, such relationships have been explored previously, with most studies reporting poor to moderate correlations to zooplankton organism abundance (Elbrecht and Leese, 2015;Sun et al., 2015;Bucklin et al., 2019), which is not surprising as planktonic organisms span ten orders of magnitude in body size. Better correlations have been observed between sequence counts and biomass (Hirai et al., 2015;Thomas et al., 2016;Yang et al., 2017). A meta-analysis of 22 metabarcoding studies that targeted a wide variety of organisms and genetic markers, found a significant, albeit weak, quantitative relationship between organism biomass and number of produced sequences (Lamb et al., 2019). Similarly, our results show a poor relationship of DNA relative abundance to individual abundance, but strong, robust correlations between relative biomass and relative DNA read counts across a wide range of taxonomic groups. The correlations we observed were stronger,  E.A. Ershova et al. and covered more taxa, than have been demonstrated in previous works (Lamb et al., 2019, Hirai et al., 2015Thomas et al., 2016;Yang et al., 2017), likely due to our choice of a highly degenerate primer set. Since we used two separate subsamples for metabarcoding and microscopic sorting, and obtained biomass estimates indirectly (via length-weight regressions), it is important to note that these values also inevitably have a high error rate associated with them. This is especially true for groups where such length-weight relationships are poorly described, and a perfect relationship was not expected. Unsurprisingly, the best quantitative results were obtained for the well-studied crustaceans-copepods and euphausiids. Additionally, since our study included only nine samples, it is important to acknowledge the error rates that can be associated with such a small sample size, with a strong potential of outliers affecting the patterns in both directions. However, despite the absence of accurate regression formulas for many groups, such as benthic invertebrate larvae, and the small sample size within our study, the correlations were remarkably strong and statistically significant, suggesting that true patterns may be even stronger.
The potential quantitative value of metabarcoding has frequently been questioned (Bucklin et al., 2016) due to the difficulty of finding truly universal primers, which amplify all groups equally well, as well as the interspecies variability in the initial amount of DNA. This is also true for our data, as the relationships that we observed between relative DNA counts and relative biomass for the different taxonomic groups rarely followed a 1:1 ratio, with certain groups being consistently overestimated either in the biomass or the sequence reads numbers. This was particularly visible in large copepods, which were the dominant group within the biomass, but had a ca. ten times lower contribution to the total DNA reads. This is not entirely unexpected, as large copepods from cold regions contain large lipid stores, which can make up more than 70% of the body mass , and are thus relatively poor in DNA. Other groups, such as polychaetes, had a consistently higher contribution in the sequencing data, most likely due to inaccurate estimation of their biomass or inadequate identification. For example, "unknown larvae" and eggs from the morphological counts may have belonged to polychaetes. Knowing the individual traits of the different species, conversion factors can be developed for different taxonomic groups, or even individual species/genera. In our samples, for example, the application of a conversion factor of 10x to the number of sequence reads of large copepods brought the ratio for this group closer to 1:1 and improved correlations for the other groups, as well as the overall contributions of the individual taxa.
Many metrics are used when performing ecosystem assessments, including overall abundance and biomass, presence/absence of certain indicator species, diversity estimates and quantitative community composition, and various wellness indices that take into account age, size, body composition, and population structure of key species Canonico et al., 2019). Although metabarcoding cannot readily provide information on abundance or size/population structure of organisms, it can be used to estimate the other metrics at a comparable, or better level of accuracy than traditional zooplankton sorting. We would like to emphasize, however, that the work of skilled taxonomists is still very much needed to supply reliable identification of specimens used in barcode reference databases. With a single bulk biomass measurement of the sample, and the application of taxa-specific conversion factors, metabarcoding data can be converted from relative to absolute biomass and serve as a proxy of organism biomass in the samples (Coguiec et al., 2021). For species with a relatively constant body size, i.e. copepods, these values can be then transformed to estimate relative abundance, as is frequently done vice versa [e.g. Gluchowska et al. (2017) and references therein].
Although metabarcoding methods are increasingly coming into maturity, a series of challenges still remains to be solved before we can obtain their full potential. The most important present handicap for marine invertebrate communities is the presence of significant gaps in reference sequence databases, although significant efforts are being made to remedy this . Our samples were collected in a relatively homogenous, low diversity environment (Atlantic Arctic and sub-Arctic), with relatively good reference databases (Bluhm et al., 2011). Therefore, these results should be tested with communities from other parts of the world with higher species diversity and less complete databases. We chose a clustering-based method of taxonomic assignment as opposed to a pipeline based on sequence variants, since the latter are not recommended for COI (or any other similarly hypervariable marker), due to widespread intra-species variability leading to millions of sequence variants. However, the choice of the clustering algorithm to delineate MOTUs presents an additional challenge, as it can vastly impact the resulting diversity. In our pipelines, we have used Swarm v.2.1.13 (Mahé et al., 2015), a step-by-step clustering algorithm, which avoids the use of arbitrary similarity thresholds. Moreover, we used taxonomic clustering when measuring diversity, which only counted uniquely identified taxonomic categories (e.g. all unknown Calanoid MOTUs were collapsed into a single group). Taxonomic clusters thus obtained are highly dependent on the completeness of the reference database, so the richness values could increase in the future when existing reference gaps are progressively filled. The 97% similarity threshold we applied for species assignments, while the accepted standard used by BOLD and applied in most ecological studies, can be significantly higher or lower in different groups of metazoans, and could also result in an under-or an over-estimation of diversity. Richness values obtained from COI metabarcoding can also be significantly overestimated by the presence of nuclear pseudogenes (nuclear mitochondrial DNA, or NUMTs) (Song et al., 2008). Our pipeline included a curation step using the LULU algorithm, which allowed us to minimize the impact of these pseudogenes (Frøslev et al., 2017). However, the presence of these sequences in marine invertebrate genomes are poorly understood. Future studies involving analyses of whole genomes of marine metazoans would help to understand the impact of this issue in molecular biodiversity metrics. Finally, although the Leray-XT primers are nearly universal for metazoans, and the metabarcoding datasets resulting from their use undoubtedly show enhanced quantitative value compared to others obtained using less-degenerate primer sets, it still fails to amplify some planktonic organisms, which needs to be taken into account. For example, within our study pteropods and mollusc larvae were very poorly represented in the sequencing data, which could be a fault of either the DNA extraction step, or the PCR. Molluscs in general are notoriously difficult to extract and amplify DNA from (Pereira et al., 2011). Alternatively, the Leray-XT primer set may not provide optimal binding conditions for this group. Further studies including mock samples should test the extent of non-detectable planktonic taxa by this primer set. If these organisms are of particular interest, then other more specific markers can be used in conjunction with the Leray-XT set.
Despite the aforementioned limitations, our metabarcoding approach can prove invaluable to monitor cryptic/pseudocryptic species or larval communities. This is a common issue within zoo- plankton studies, since several closely related species, sharing a very similar morphology, may be present in the water column simultaneously, and very few studies attempt to distinguish between them. Examples include complexes of the copepod "species" Oithona similis (Cornils et al., 2017), Acartia tonsa (Drillet et al., 2008), Eurytemora affinis (Devreker et al., 2012) or the jellyfish Cyanea capillata (Kolbasova et al., 2015). Despite their morphological similarity, different members of such species complexes may play different roles in the ecosystem, and often have different, slightly overlapping geographical ranges, the exact extents of which were until now difficult to parse out. Within our studied zooplankton communities, examples of such pseudo-cryptic genera included Calanus (Choquet et al., 2017) and Pseudocalanus spp. (Frost, 1989;Ershova et al., 2017), which were each represented by four species in our area of study. Detection of these species in our study regions is in agreement with other studies that have examined the species composition of these groups using different molecular markers (Choquet et al., 2018;Coguiec et al., 2021). Surprisingly, only one species of Microcalanus, M. pusillus, was observed at all examined locations, although M. pygmeaus has also been previously reported in these regions (Klekowski and Węsławski, 1990;Walkusz et al., 2009). Similarly, meroplanktonic larvae are most frequently ignored in plankton studies, or are grouped into broad taxonomic categories. Within our study, the metabarcoding protocol identified DNA of over 100 benthic taxa, most of which presumably belong to planktonic larvae. Monitoring the timing when certain species are present in the water column can give important insight on the life cycles of benthic animals, as well as track the potential expansion of introduced bottom-dwelling species into the region (Couton et al., 2019;Ershova et al., 2019). Another clear advantage of metabarcoding is that the use of hypervariable markers, such as COI, has the potential to provide intra-species diversity information (population genetics) for multiple species at the same time , which can be applied to detect subtle changes in the composition of planktonic communities over both time and space.
Although molecular approaches have always been associated with a high cost, they are rapidly becoming more affordable. With metabarcoding in particular, an increasing number of simultaneously processed samples will reduce the cost dramatically ( Figure  9a; Supplementary Table 4). Similarly, the time required for process-ing per sample will be reduced substantially when using a metabarcoding approach, in contrast with traditional sorting, which follows a linear progression of effort per sample (Figure 9b; Supplementary  Table 4). For this reason, we suggest that this protocol can serve as a useful alternative to traditional sorting for monitoring studies, which generally involve hundreds of samples and require thousands of technician hours for processing. As zooplankton is inherently patchy, this would also allow for increased replication (i.e. processing more than one sample from the same location), the lack of which is a serious, but generally overlooked issue in zooplankton ecology (Skjøldal et al., 2013). The variable nature of zooplankton was apparent already in our results, where three net tows taken during the same day in the same location produced similar, but far from identical results, with several species showing up in one or two, but not all three samples. The development of sequencing technology and advances in metabarcoding approaches makes the protocol developed herein scalable and affordable. The protocol uses metabarcoding primers with unique eight-base sample-tags attached to each primer pair (Wangensteen et al., 2018), and 96 such primer pairs allow multiplexing 96 samples in a single library. A library-tag can subsequently be ligated to each library, allowing for up-scaling for simultaneous sequencing of 100-1000's of samples in a single sequencing run. For example, a project that wants to analyse 1440 samples using our protocol, will have to extract the samples in 15 96-well plates, run PCRs in 15 plates, pool the samples in each plate, purify each pool, and prepare 15 libraries by ligating Illumina sequencing adaptors and library-tags on each pool. These 15 libraries can then be pooled into a single tube and sequenced on an Illumina Novaseq using 250 bp paired-end chemistry. With an output of 400 million paired-end reads, the Novaseq is cheaper, and provides more sequencing depth per sample compared an Illumina MiSeq. These reductions in costs and time consumption will only increase within the next few years for the metabarcoding approach, as sequencing run time keeps reducing and semi-automated and fully automated pipelines are starting to emerge for markers like COI. Similarly, we expect reference databases to vastly improve in their coverage in the coming years to allow for maximum taxonomic accuracy in the sequencing data. Therefore, by reducing processing time, metabarcoding will allow analysis of more samples than was possible before with prospects for improved replication and/or spatiotemporal resolution.

Conclusions
Our results highlight the quantitative value of a metabarcoding approach using the highly degenerate Leray-XT primer set, which provides improved taxonomic resolution, universal coverage for metazoans, reduced primer bias, and availability of a comprehensive reference database, to investigate zooplankton biodiversity, community composition, relative species biomass, and the presence of rare, poorly preserved, or cryptic species. We suggest that relative sequence counts can be used to estimate individual organism biomass via a single bulk biomass measurement, particularly within a single study, which will facilitate monitoring of spatial, seasonal, and inter-annual dynamics of zooplankton communities. Our protocol can allow rapid and relatively inexpensive processing of hundreds to thousands of samples at a higher taxonomic resolution than traditional zooplankton sorting, and can be widely applied for monitoring or ecological studies.

Data availability statement
The data underlying this article are available in the online supplementary material. The raw sequencing data and the MOTU tables are publicly available at Mendeley Data (Dataset reference 5p37bgckjf. DOI: 10.17 632/5p37bgckjf.1).

Funding
This research has been funded by UiT The Arctic University of Norway and the Tromsø Research Foundation under the project "Arctic Seasonal Ice Zone Ecology," project number 01vm/h1, and partially supported by RFBR grant 19-04-00955. This study was done within the framework of state assignment of IO RAS (theme no. 0128-2021-0007)

Supplementary Data
Supplementary material is available at the ICESJMS online version of the manuscript.