On the effectiveness of two universal metabarcoding primer pairs in amplifying Arctic gelatinous zooplankton DNA from tissue and environmental samples

Gelatinous zooplankton (GZP) comprise a highly diverse group of marine taxa that play key ecological roles throughout the world’s oceans. However, their fragile bodies make them difficult to effectively study using traditional sampling methods such as net and optical surveys. While the recent advancements of non-invasive DNA metabarcoding techniques have become valuable tools in GZP biodiversity research, the performance of commonly used metabarcoding markers for these taxa remains poorly understood. In this study, we evaluated two universal primer pairs targeting the 18S rRNA V1–V2 and mitochondrial COI Leray gene regions for detecting key Arctic GZP species. Primer performance was assessed using in silico PCRs, then in vitro PCRs and sequencing of tissue-derived DNA, and environmental DNA collected from a controlled mock community experiment. The COI primers provided high species-level taxonomic resolution but failed to amplify several hydrozoan species and the cydippid ctenophores. In contrast, the 18S primers amplified a broader range of taxa but lacked sufficient resolution to discriminate among closely related genera and species. We also identified substantial gaps in public reference databases that limit the accurate identification of Arctic GZP and finally, we provide additional barcodes to expand the existing coverage in public databases. These findings demonstrate the complementary strengths and limitations of commonly-used COI and 18S markers and provide guidance for marker and exploration of new genetic markers to support DNA-based monitoring of Arctic GZP. This is essential in surveying these elusive organisms in light of their climate-change driven observed and projected distributional shifts.

Gelatinous zooplankton (GZP) are a highly diverse and polyphyletic group (van den Heuvel-Greve et al., 2026) which includes pelagic tunicates (Phylum: Chordata), comb jellies (Phylum: Ctenophora) and medusae (Phylum: Cnidaria). Despite being historically excluded from zooplankton surveys, recent research has demonstrated that they play major roles throughout marine ecosystems and food webs (reviewed in: Hays et al., 2018). Their gelatinous bodies and polymorphic lifecycles, which can include pelagic medusae, benthic polyps and small-sized larvae, make them susceptible to damage or being missed completely by invasive sampling techniques such as plankton nets or fishing trawls (Haddock, 2004; Hosia et al., 2017; Nogueira Júnior et al., 2015). Optical surveys represent a non-invasive sampling method that have proven valuable for estimating distribution and abundances of GZP (Pantiukhin et al., 2023, 2024a; Raskoff et al., 2010) yet rely on the presence of recognizable morphological features which can hinder high-resolution identifications (Haddock, 2004; Hosia et al., 2017). Consequently, knowledge of GZP ecology, including community composition, distribution, and abundance, remains limited. This is particularly true in remote regions with challenging environmental conditions such as the Arctic Ocean, where multiple GZP species are expected to undergo significant shifts in range distribution (Pantiukhin et al., 2024b), as a response to amplified climate change effects including increasing ocean temperatures (Rantanen et al., 2022).

In recent years, molecular barcoding of tissue as well as metabarcoding of environmental DNA (eDNA) and diet samples have become valuable tools for surveying and identifying fragile taxa like GZP (Ames et al., 2021; Bolte et al., 2021; Günther et al., 2021; Minamoto et al., 2017; Ogata et al., 2021; Rantanen et al., 2022). Molecular barcodes, or DNA barcodes, are gene regions that are used for the purpose of taxonomic identification of single-species samples. The most widely used barcode region for marine metazoans is the ~658 base pair (bp) region of the mitochondrial cytochrome c oxidase I (COI) gene (Bucklin et al., 2011). A ‘metabarcode’ is applied to multi-species samples and is often shorter than a barcode region. Several studies based on metabarcoding of eDNA and stomach contents in the Arctic have revealed the previously unknown importance of GZP as prey items for fish (Dischereit et al., 2024b), amphipods (Dischereit et al., 2024a, 2022) and decapods (Urban et al., 2022). Additionally, eDNA-based studies in the Arctic have successfully recovered GZP taxa in pelagic and deep-sea habitats (Murray et al., 2024), in the Bering Strait (Verhaegen et al., 2025), a high-Arctic fjord (Murray et al., 2025a) and the marginal ice zone (Murray et al., 2025b). When compared to morphological sampling from the same sites, eDNA has consistently recovered taxa overlooked by nets and/or optical surveys and was able to detect more GZP taxa in total (Feng et al., 2022; Govindarajan et al., 2021; Murray et al., 2025a, 2024).

A critical factor in metabarcoding studies is the selection of appropriate genetic markers and primer pairs, as these choices directly influence detection success and taxonomic recovery (Clarke et al., 2017; Taberlet et al., 2018; Elbrecht et al., 2019). Mismatches between the primers and the DNA of the target species and other PCR biases as well as gaps in reference databases can lead to false negatives and underestimation of species richness (Taberlet et al., 2018). When targeting a polyphyletic group such as zooplankton, “universal” markers are usually employed in order to ensure as much diversity is captured as possible. Two such genes are the 18S nuclear small-subunit ribosomal RNA gene (henceforth 18S) and the mitochondrial COI gene, with the former being the most frequently used in earlier zooplankton metabarcoding studies (reviewed in Bucklin et al., 2016). Due to its conserved nature, 18S primer binding regions are shared among many eukaryotic taxa, enabling broad amplification of eukaryotes. However, this conservative nature also limits its discriminatory power between species and restricts taxonomic identification to only order and family level for many metazoans, including zooplankton (Bucklin et al., 2016; Questel et al., 2021; Tang et al., 2012; Wu et al., 2015). More recently, the Leray fragment of the COI gene (within the Folmer region) has become a popular choice for species discrimination in marine metazoans (Bucklin et al., 2016; Hebert et al., 2003b). Its rapid evolutionary rate provides greater discriminatory power for fine-scale taxonomic assignments, including species-level identification when suitable reference sequences are available, and it has also proved to be an effective tool for detecting levels of intraspecific diversity (Adams et al., 2019; Antich et al., 2022). There are several primer pairs used to amplify this fragment, with the highly degenerated Leray-XT pair being a recent yet popular addition (Albaina et al., 2024; Geller et al., 2013; Wangensteen et al., 2018). Extensive curated public databases for marine flora and fauna exist for the COI marker (Andújar et al., 2018), including the Barcode of Life Data System (BOLD; Ratnasingham and Hebert, 2007) and the zooplankton-specialized MetaZooGene database and barcode atlas (MZGdb; O’Brien et al., 2024).

Both the 18S V1-V2 and COI Leray markers have been employed in diet and eDNA metabarcoding studies with the aim of increasing the coverage of GZP taxa, and each have shown benefits and limitations depending on the taxonomic group. In a study on Greenland fish diets, a higher number of hydrozoan and scyphozoan species were recovered with COI, while 18S recovered more ctenophores (Dischereit et al., 2024b). In the Southern Ocean, 18S recovered salps in the diets of fish, which were overlooked by COI (Ruiz et al., 2024a, b). In an Arctic eDNA study, more GZP taxa overall were found with COI in the sub-ice water column in the Arctic marginal ice zone, but appendicularians were only detected with 18S (Murray et al., 2025b). However, it remains unclear whether the reason these GZP taxa are overlooked by COI is due to primer biases or gaps in reference databases. Given the importance of GZP in Arctic marine ecosystems and the increased costs of multi-marker studies compared to single-marker ones, it is important to determine whether the inclusion of a ribosomal marker such as 18S will actually lead to an increased GZP species detection in the region.

Here, we aim to address these knowledge gaps by investigating the performance of two common primer pairs used for amplifying COI and 18S metabarcoding fragments on a subset of Arctic GZP species. We approached this by assembling an in-field mock community of Arctic GZP species regularly detected in the region with net and optical sampling. We (i) assessed the coverage of the target species on two major public reference databases, (ii) evaluated primer binding efficiency in silico on sequences from these databases, (iii) tested the primers on DNA derived from the tissue of net-caught specimens and finally (iv) tested the primers on a mock community with metabarcoding.

Specimen collection

Gelatinous zooplankton specimens used in the mock community were collected during the oceanographic expedition PS126 aboard the R/V Polarstern (Soltwedel, 2021). The samples were collected at the station called HG-IX (79°07.967’N 002°45.607’E) in the HAUSGARTEN observatory in Fram Strait, which is between Svalbard and Greenland (Figure 1). The sampling region is influenced by both Arctic and boreal (Atlantic) water masses, resulting in a zooplankton community of mixed biogeographic origin, including arctic, arctic-boreal, and boreal species (Weydmann et al., 2014). The composition of the net catch and the subsequent mock community reflected this natural assemblage. The individuals were collected using a Maxi-Multinet (Hydrobios, Kiel) with a 330 µm net size, in a vertical haul from a maximum sampling depth of 2500 m to the surface (Soltwedel, 2021). The specimens were morphologically identified to the highest possible taxonomic resolution onboard using taxonomy keys (Bouillon et al., 2006; Licandro et al., 2017a, b; Licandro and Lindsay, 2017).

Mock community preparation

A mock community comprising of 18 individuals representing five cnidarian families and the ctenophore class Tentaculata was assembled from net-caught specimens (Table 1). As the experiment was conducted in the field, the mock community reflected the naturally occurring GZP assemblage at the sampling location and time rather than an exhaustive representation of Arctic GZP diversity. The least damaged individuals were selected from the net catch, of which the majority were confirmed to be still alive. The individuals were incubated in a non-circulating tank containing 14 L of seawater, which was collected from the onboard CTD rosette water sampler and filtered through polycarbonate filters to remove as much DNA as possible. The seawater filtration was performed sequentially through a 3 µM and then a 0.22 µM filter, using a peristaltic pump. A control sample of 500 ml of the filtered seawater was collected before the addition of the GZP specimens and filtered over a 0.22 µM Sterivex-GP filter (Merck Millipore) as a control. The assemblage was kept in the tank for approximately 40 hours, after which triplicates of approximately 2 L were filtered over 0.22 µM Sterivex-GP filters. The mock community tank experiment and all filtration procedures were conducted in an onboard laboratory. All work surfaces were decontaminated with a 10% bleach solution followed by a Milli-Q water rinse, and sterilized tubing and fittings were used throughout to minimize the risk of cross-contamination. Filters were subsequently preserved at -80°C until further processing in the home lab. After retrieval from the experiment, the specimens were stored at -80°C for DNA extraction in the home lab, approximately 2 years later.

Primer selection

All primer pairs selected for analysis target widely used (meta)barcoding regions of the 18S rRNA and COI genes. To contribute reference barcodes to existing public databases, tissue-derived DNA extracts of the specimens included in the mock community were amplified using primers targeting the full-length 18S rRNA gene (~1,800 bp) and the ~658 bp Folmer fragment of the COI gene. For metabarcoding, the V1–V2 region (approximately 356 bp long) of the 18S rRNA gene was amplified using the primer pair described by Blaxter et al. (1998) and Sinniger et al. (2016). This primer set has been applied in recent polar metabarcoding GZP studies (Dischereit et al., 2024b; Murray et al., 2025b; Ruiz et al., 2024b, 2024a). For COI metabarcoding, the highly degenerate Leray-XT primer pair (Geller et al., 2013; Wangensteen et al., 2018) was used to amplify the 313 bp Leray fragment. This primer pair is widely employed in metabarcoding studies of marine metazoans (e.g., Antich et al., 2022; van den Heuvel-Greve et al., 2026; Wangensteen et al., 2018a) and has been successfully applied to GZP communities (Dischereit et al., 2024a, b; Murray et al., 2024, 2025a, b; Ruiz et al., 2024a, b; Verhaegen et al., 2025). The location of the primer binding sites can be found in Figure 2 and the sequence details of each primer pair can be found in Supplementary Table S1.

In silico PCR

Primer binding success was assessed in silico using the “Test Primers” function in Geneious software v2024.0.5 (https://www.geneious.com/), which is based on the software Primer3 2.3.7 (Rozen and Skaletsky, 1999). A local sequence database was created by downloading 18S and COI sequences from the MZGdb and BOLD databases. The primers were tested on each individual sequence using the settings: (a) minimum amplicon length of 250bp (excluding primers) and (b) a maximum of five mismatches in each direction. This high threshold was selected as a permissive criterion to account for sequence variation among taxa, the degeneracy of the COI Leray-XT primers, and lower annealing temperature in the second COI PCR protocol, while allowing potential primer-binding regions to be identified. Congeneric sequences were included when the target taxa (i.e., mock community individuals) were identified with a lower than species-level resolution (e.g., Botrynema spp. and Atolla spp.). For the unidentified ctenophore juveniles, we included sequences from the most commonly occurring Tentaculata in the region (genera Pleurobrachia, Mertensia, and Euplokamis) where possible.

Tissue DNA extraction, amplification and sequencing

Intra-organismal DNA was isolated from the tissue of all individuals used in the mock community experiment. It was extracted using the QIAGEN DNeasy Blood & Tissue kit following (Murray et al., 2023), with a final DNA elution volume in AE buffer of 100uL. The DNA concentration was then checked using a NanoDrop One (Thermo Scientific) and subsequently diluted to a concentration of 50 ng/µl where necessary. The primer sequences, PCR reaction master mixes and conditions for each primer pair can be found in Supplementary Table S1. After amplification, PCR products were verified on 2% agarose gel and then bidirectionally sequenced on a Sanger Sequencing platform at EUROFINS (Germany). All work surfaces and pipettes were sterilized with a 10% bleach and Milli-Q solution and subsequent UV light prior to extraction and PCRs.

After sequencing, primer sequences were subsequently removed and sequences trimmed using Geneious v2024.0.5. Consensus sequences were assembled using the forward and reverse reads for each sample. Then they were manually checked for ambiguous base calls and stop codons, before being assigned using the MegaBlast function in Geneious. The COI sequences were further checked with the BOLD “Identification V4” function.

Mock community eDNA extraction, library preparation and sequencing

Environmental DNA was isolated from the filter membranes using the DNeasy Blood and Tissue kit (QIAGEN) following the protocol detailed in Murray et al. (2024). All laboratory surfaces and extraction equipment were sterilized between extractions using a 1:10 bleach solution diluted with MilliQ water, followed by a MilliQ rinse and dried using 70% ethanol. Finally, all benches and equipment were treated with a minimum of 1 hour of UV light. RNA/DNA Exitus-plus (Applichem) was used for sterilization of gloves and pipettes. The DNA extract was stored at -20 0 C for downstream processing. Finally, an extraction blank was included to detect any potential contamination during these steps.

Three different datasets were sequenced from this DNA: 18S rRNA V1-V2, COI Leray-XT with 2-step PCR, and COI Leray XT with 1-step PCR and ligation. These will be henceforth referred to as the 18S, COI-A and COI-B datasets respectively. The PCR amplification, metabarcoding library preparation and sequencing for the 18S and COI-A datasets were carried out by AllGenetics & Biology SL, Spain. A 2-step PCR protocol was used for these two datasets, amplifying the respective target fragment using the metabarcoding primers in the first PCR and the attachment of the oligonucleotide indices in the second PCR (see Supplementary Table S1 for PCR details). The annealing temperature was (54.7°C) was optimized through testing by the sequencing company. The PCR products were verified on 2% agarose gel before purification with Mag-Bind RXNPure Plus magnetic beads (Omega Bio-tek), following the manufacturer protocol. The PCR amplification and library preparation for COI-B were conducted in the home lab, following the original protocol created for the primers set (Wangensteen et al., 2018), which has a lower annealing temperature of 45°C. A negative control was included in all PCRs in order to detect any potential contamination. The ligation of the oligonucleotide indices and sequencing were conducted by Novogene, Germany. All libraries were quantified using the Qubit dsDNA HS Assay (Thermo Fisher Scientific) before being pooled with equimolar amounts and sequenced using NovaSeq PE250 flow cells (Illumina).

Reference data base coverage

We used two popular databases to estimate the coverage of Arctic GZP in species in publicly available reference databases: the MetaZooGene database (MZGdb; O’Brien et al., 2024) and the Barcode of Life Data System (BOLD; Ratnasingham and Hebert, 2007). First, we used the coverage and completeness summary function in the MZGdb and second, the BOLD v4 Taxonomy Browser. The MZGdb contains a combination of sequences from GenBank and BOLD, with a focus on zooplankton, while BOLD contains sequences from GenBank as well as those directly submitted by registered users. Sequence counts were based on publicly available sequences only. Both databases have strict curation protocols of the sequences therein.

Bioinformatic processing, blank correction and data curation of mock community eDNA

The COI-A raw sequence reads were demultiplexed by the sequencing company using Cutadapt v5.1 (Martin, 2011), while for COI-B demultiplexing was performed as part of the bioinformatic processing. The bioinformatic pipeline used to process the two COI datasets was based on OBITOOLS 3 (Boyer et al., 2016), following (Murray et al., 2025a). Briefly, data was demultiplexed (only COI-B), paired-end reads were merged, length-filtered and sequences dereplicated. Chimeric sequences were removed with the uchime denovo algorithm in VSEARCH v2.28.1 (Rognes et al., 2016). Sequences were denoised using DnoisE v1.4 (Antich González et al., 2022) (alpha = 4) and singletons removed. A relative threshold abundance of 0.002% for each sample was applied to filter out sequencing artefacts, and a blank correction was applied following Antich et al., 2021. Next, clustering denoised sequences into molecular operational taxonomic units (MOTUs) (d=13) with SWARM v.3.1.5 (Mahé et al., 2021). Any remaining MOTUs with a total read count of <5 were removed. The ecotag algorithm was used for taxonomic assignment against a custom reference database (https://github.com/adriantich/NJORDR-MJOLNIR3; DUFA v202305). This database consists of sequences from NCBI GenBank, BOLD and in-house sequences. Post-clustering filtration was done with LULU v1.0 (Frøslev et al., 2017). Finally, bacteria and nuclear mitochondrial DNA segments (NUMTs) were filtered out. Taxonomic assignments were further checked and improved where possible using the BOLDigger3 v2.1 software (Buchner and Leese, 2020) and the BOLD database (Ratnasingham and Hebert, 2007). The following identity thresholds were used: species (97%), genus (95%), family (90%), order (85%), phylum (80%) following Murray et al. (2024). All MOTUs with taxonomic assignments <70% were removed from the metabarcoding datasets completely.

The 18S raw sequence reads were processed using DADA2 v1.18 (Callahan et al., 2016) following Murray et al. (2025). Briefly, forward reads were trimmed at 220 bp and reverse reads at 210 bp and filtered with a maximum expected error threshold of 2.2 and 2.1, for forward and reverse reads respectively. Sequences were denoised and paired ends merged using a minimum overlap of 25bp and zero mismatches. Singletons were discarded and chimeras filtered out and the same blank correction used in the COI dataset was applied here. Taxonomic assignment of the resulting amplicon sequence variants (ASVs) was done using the MZGdb (O’Brien et al., 2024). The version “MZGdb 3.0” containing 68,484 18S rRNA sequences containing genus and species data from any ocean was used (Mode C). A minimum identification bootstrap support value of >75% was set for taxonomic assignment, and lower values were set as unassigned.

For both datasets, reads assigned to bacteria, non-eukaryotes and obvious terrestrial taxa were removed. Finally, reads present in the control sample (filtered seawater used in mock community before the addition of the GZP individuals) were removed from the main samples on a read-by-read basis (i.e., if 50 reads of a MOTU/ASV were present in the control then 50 reads of this MOTU/ASV were removed from the samples). In the case of a MOTU/ASV having less reads in the sample than in the control, the number of reads was set to zero.

Data analysis

The program R in version 4.1.0 was used to visualize and conduct all analysis unless otherwise stated. The phyloseq package (McMurdie and Holmes, 2013) was used for initial metabarcoding data exploration. The packages ggplot2 (Wickham, 2016), ggOceanMaps (Vihtakari, 2026) and Fantaxtic (Teunisse, 2022) were used for data visualization.

Coverage of Arctic GZP on public databases

According to MZGdb (v2023-m07-15), 122 Hydrozoa, 10 Scyphozoa, and seven Ctenophora species are commonly reported in the Arctic Ocean. Sequence coverage was lowest for Hydrozoa (COI: 50%; 18S: 40%), followed by Ctenophora (COI: 71%; 18S: 85%) and Scyphozoa (COI: 100%; 18S: 80%). A BOLD database search (18 December 2025) using the taxa Ctenophora, Hydrozoa, Scyphozoa, and Appendicularia and the location filter “Arctic Ocean” returned 174 publicly available barcodes from 15 species across four genes. Barcode representation was highest for 18S, 16S, and COI (44 sequences each), followed by 28S (42 sequences).

For the species included in the mock community, the highest number of publicly available COI reference sequences were for A. digitale, with nine sequences from the Arctic Ocean, while S. arctica had only two sequences. There were no publicly available sequences for the ctenophore species Euplokamis dunlapae, and only three sequences for Mertensia ovum (non-Arctic locations). For 18S, every targeted species had at least one reference sequence available. The Botrynema genus was the best represented with 12 sequences and the ctenophore E. dunlapae the least with a single 18S barcode (Supplementary Table S2).

Assessment of metabarcoding primers using i n silico PCR

COI Leray-XT primers

A total of 74 reference sequences from eight species were harvested from the MZGdb and included in the in silico testing of the Leray-XT primer pair (COI gene). Binding of the forward primer (mlCOIintF-XT) was successful for all species with available reference sequences (Supplementary Table S3). Successful binding of the reverse primer (jgHCO2198) occurred for at least one sequence of every species, except for Solmundella bitentaculata and S. arctica. However, the binding site of this primer is the same as the reverse Folmer primer and is thus cut off from most sequences of this fragment that are available on public databases. As a result, many available sequences were cut too short and did not include the binding site, thus preventing more thorough testing. Most mismatches occurred with forward primer (4), while only one occurred with the reverse primer. The majority of these mismatches occurred in the binding site of the forward primer for the hydrozoan A. digitale and ctenophore Pleurobrachia pileus (Supplementary Table S3). There were no reference sequences available for the ctenophore E. dunlapae making its inclusion in the analysis impossible.

18S V1-V2 primers

A total of 28 sequences from 12 species were used for in silico testing of the 18S V1-V2 primers. Primer binding failed in both directions for S. arctica. The binding of the forward primer (SSU_F04) was successful for at least one sequence of all species, with the exception of Atolla tenella (Supplementary Table S4). However, this sequence was shorter than the majority of other Atolla sequences (<1800 bp), and an alignment with the congeneric Atolla wyvillei confirmed that the primer binding site was excluded from the sequence. No primer mismatches occurred for either of the primers. There were no reference sequences available for E. dunlapae and it was thus excluded from this analysis.

Tissue-derived DNA sequencing

COI

When sequencing the Folmer fragment (~658 bp), successful sequences were acquired for only three out of 16 specimens (Table 2). Of these, only A. digitale DNA was assigned to species level. The narcomedusae specimen was assigned to the family Aeginidae and the ctenophore juvenile did not have any matches on BOLD nor when using the BLAST function in Geneious, which searches the NCBI nucleotide database. More successful sequences were produced from the shorter Leray fragment (~313 bp), with 12 out of 16 individuals resulting in successful sequences. Of these, six were identified to species level and one to family level. The remaining five did not match any reference sequences.

18S

Successful sequences were produced for all 16 individuals for both the full 18S gene and the V1-V2 fragment (Table 2). The only 100% match for the full longer fragment with a match to a single species was for A. digitale. All other assignments showed very high matches (>99.5%) with multiple species or genera, preventing accurate genus- and species-level delimitation.

Metabarcoding of the mock community eDNA

Sequencing summary

Three samples from the mock community experiment, as well as one seawater control, one extraction blank, and PCR controls, were successfully sequenced across three sequencing runs (18S, COI-A, and COI-B). Following bioinformatic filtering, blank correction and data curation, the COI-A dataset comprised 153,334 reads assigned to 89 MOTUs, while the COI-B dataset contained 199,441 reads assigned to 90 MOTUs. The 18S dataset contained 1,193,934 reads assigned to 483 ASVs. Sequencing summaries for each processing step are available in Supplementary Table S5.

Within the COI datasets, 115,999 reads were assigned to 10 GZP MOTUs (phyla Cnidaria and Ctenophora) in COI-A, and 141,384 reads were assigned to 12 GZP MOTUs in COI-B (Supplementary Table S6). In the 18S dataset, 54 ASVs were assigned to GZP, representing a total of 804,142 reads (Supplementary Table S7).

Leray-XT primers

A total of 10 GZP MOTUs were recovered in the COI-A dataset and 12 in the COI-B (Supplementary Table S6). Except for ctenophores and the siphonophore D. arctica, the detected MOTUs were assigned to all the target taxa included in the mock community (Figures 3; Supplementary Table S6). The main difference between COI-A and COI-B was the positive detection of A. digitale in the latter dataset only. A total of eight MOTUs were assigned to genus and species level in COI-A and nine for COI-B. The most abundant MOTU in both datasets was assigned to the species Atolla tenella, followed by a MOTU assigned to the order Narcomedusae (Figure 4).

V1-V2 primers

A total of 54 GZP ASVs from 21 different taxa were detected in the 18S dataset (Supplementary Table S7). All taxa included in the mock community were detected, with the exception of the hydrozoan Sminthea arctica, for which only the family was detected (Rhopalonematidae). The detection of D. arctica was only a genus-level detection (Figure 3). The family Diphydae was the most abundant in terms of relative abundance, followed by Rhopalonematidae and the ctenophore family Mertensiidae (Figure 4). A total of 20 ASVs were assigned to genus level with high bootstrap support, and species-level assignments were not considered due to the high likelihood of assignment errors for this marker.

Non-target GZP

Both markers amplified sequences from non-target taxa (i.e., GZP taxa not included in the mock community), albeit at low relative read abundances. One notable example is the large scyphozoan Cyanea capillata (species-level in COI and family-level in 18S) was found in low abundances in all datasets. Additionally, the COI dataset recovered the siphonophores Rudjakovia plicata, Homonema platygon (COI-B only) and Crystallophyes amygdalina (Supplementary Table S5) and the 18S dataset an ASV assigned to the ctenophore family Beroidae (Supplementary Table S7).

Recent advances in molecular techniques have led to the effective implementation of (meta)barcoding for biodiversity surveys across many marine environments (Bucklin et al., 2016; Leray and Knowlton, 2016). This is especially advantageous when targeting fragile taxa such as GZP, and in remote areas such as the Arctic Ocean. However, species recovery can be impeded and downstream data interpretation biased by several factors including primer mismatches and gaps in sequence reference databases. Although previous studies have revealed these issues when targeting GZP (Dischereit et al., 2024b; Murray et al., 2024; Ruiz et al., 2024a), there has been a lack of evaluation of why some species or groups are consistently overlooked in eDNA surveys in the Arctic. In the present study, we aimed to address these knowledge gaps by evaluating commonly used primer pairs of the COI and 18S genes on a mock community comprised of species commonly found in the Arctic Ocean.

Database coverage and in silico evaluation of metabarcoding primers

The coverage, accuracy and availability of reference databases are fundamental to the effectiveness of metabarcoding studies (Porter and Hajibabaei, 2018a; Taberlet et al., 2018). Curated databases such as BOLD and MZGdb, have been successfully applied in many marine (meta)barcoding studies and can improve taxonomic assignment by prioritizing sequence quality, taxonomic accuracy, and associated metadata. However, despite a rapid increase in barcoding effort world-wide some groups remain under-represented, impeding accurate taxonomic assignments (Porter and Hajibabaei, 2018b).

Our results showed that reference sequence coverage of Arctic GZP species remains limited for both COI and 18S. This was particularly evident for the class Hydrozoa, where only 40% of the species known to occur in the Arctic had 18S barcodes available on MZGdb, and only 50% had COI barcodes. Moreover, our results highlight that not only are there species and genera lacking in reference sequences in general, but there is a paucity of specimens caught and barcoded from the Arctic Ocean itself. Relying on reference sequences from other regions could result in endemic congenerics and cryptic species being overlooked (Porter and Hajibabaei, 2018b). For 18S, although reference sequences were available for most mock-community taxa, many lacked sufficient sequence coverage of the targeted V1–V2 region. This may reflect the lower sequence quality often associated with the beginning of Sanger sequences (Crossley et al., 2020), suggesting that targeting a more central fragment (e.g., V4 or V7) might help to circumvent this issue.

In silico primer analysis showed that both of the metabarcoding primers pairs had high binding rates. The Leray-XT primer pair successfully matched at least one primer-binding site for all tested species, with the exception of the reverse primer in the narcomedusan Solmundella bidenticulata and the hydrozoan S. arctica. However, several taxa, including cydippid ctenophores, were absent from the reference databases and therefore could not be evaluated. The 18S primers showed successful bindings for all species tested, in both directions, supporting the prior assumption that both primer sets are “universal” and that 18S has a higher coverage for ctenophores of the class Tentaculata (Dischereit et al., 2024b).

Nevertheless, while in silico assessments of primer coverage can give a good indication of how appropriate a primer pair is for targeting specific groups, some species may show successful amplification in silico but not in vitro and vice versa (Ficetola et al., 2010). Accuracy of in silico analyses is also affected by the quality and coverage of available sequences. Expanding reference databases with additional specimens, particularly from the Arctic Ocean and adjacent seas, would therefore improve both primer evaluation and subsequent taxonomic assignments. Overall, our results support the broad applicability of both primer sets while highlighting that database completeness remains an important limitation for GZP metabarcoding.

Primer performance: taxonomic coverage

When applied to tissue-derived DNA, the 18S primers pairs showed broader coverage than the COI primer pairs. For COI, the Folmer primers (~658 bp fragment) had the least success at amplifying tissue DNA, with only two out of 16 specimens successfully sequenced. The low amplification success of the Folmer primers is likely related to their limited degeneracy, which reduces primer universality across diverse zooplankton taxa and can lead to lower amplification success (Geller et al., 2013). However, it must be noted that we did not conduct any PCR optimization nor replicates for the Folmer primers in the present experiment; doing so could improve the amplification success on the target species. Further optimization, such as adjusting annealing temperatures, using different Taq polymerases, or reducing PCR inhibition through the inclusion of bovine albumin serum, may further improve amplification success (Albers et al., 2013; Taberlet et al., 2018). Amplifications using the metabarcoding primers (Leray-XT) were more effective, producing 12 out of 16 successful sequences from the tissue extracts. Amplifications using the Leray-XT primers successfully detected all families of the specimens included in the mock community experiment except for the Diphyidae siphonophores (morphologically identified as D. arctica) and the ctenophore class Tentaculata. Although reads were assigned to both siphonophores and ctenophore taxa (18S only) in the metabarcoding output, these were in very low read abundances (<0.01%), with the exception of Mertensiidae in the 18S dataset, and are therefore more likely to represent low-level background DNA, sequencing artifacts or false positives. Nevertheless, the detection of a broad range of GZP taxa is consistent with previous Arctic metabarcoding studies in Fram Strait, where the Leray-XT primers recovered numerous GZP MOTUs from open-water and marginal ice zone communities (Murray et al., 2024, 2025b). Together, these results demonstrate the broad applicability of the Leray-XT primers for Arctic GZP metabarcoding while highlighting that primer performance can vary among taxa. In contrast, the 18S primer pairs generated successful sequences from the tissue of all specimens, for both fragment lengths. In the mock community experiment, all of the families included were detected in the metabarcoding output. The highest relative read abundances were assigned to ASVs of the Diphyidae family, followed by the hydrozoan family Rhopalonematidae. The lowest relative abundances were observed for the narcomedusan family Aeginidae and the hydrozoan family Halicreatidae. Notably, the 18S metabarcoding primers were the only ones to amplify the ctenophore taxa included in the experiment (family: Tentaculata). This incongruency between the two markers may be due to a number of factors including the degradation rates of mitochondrial and 18S nuclear DNA, primer suitability and PCR-associated biases (Elbrecht and Leese, 2015; Nichols et al., 2018; Shaffer et al., 2025). However, given the high number of read copies of mitochondrial DNA compared to nucleic DNA, it is likely due to technical biases.

Primer performance: taxonomic resolution

The ability to recover taxa at an ecologically relevant resolution is an important consideration when selecting markers for community metabarcoding studies. In the present study, the COI markers showed a higher taxonomic resolution than the 18S primers, allowing for high-confidence species-level assignments (>97% match). The COI fragment has sufficient genetic divergence to distinguish between congeneric species. This was especially true for the Leray fragment, which detected Atolla tenella and Botrynema brucei to species level with high accuracy (>99%) in the tissue samples. Four ctenophore specimens were successfully sequenced for the Leray fragment but were categorized as “no match” in the taxonomic assignment. This is not surprising considering that cydippids (especially M. ovum) were some of the most poorly represented taxa in the databases for the COI gene. In the metabarcoding samples, nine out of the 12 GZP MOTUs were identified to species-level. The target groups that were not identified to genus or species level in the mock community were the least represented in the reference databases (e.g., Rhopalonematidae), except for the hydrozoan A. digitale (discussed below). These results further highlight the importance of well curated and comprehensive reference databases to allow for meaningful identifications using the COI fragment.

In contrast, the 18S primer pairs showed limited taxonomic resolution. For all specimens except Aglantha digitale, at least one additional genus or higher taxonomic level showed identical sequence matches (100% similarity and coverage), and several specimens had high similarity (>99%) with closely related genera. Consequently, species-level discrimination was generally not possible, and assignments were often limited to family level. The successful species-level assignment of A. digitale likely reflects the distinctiveness of this monospecific genus rather than the discriminatory power of the marker, particularly as only a single publicly available reference sequence was available. Overall, our findings for GZP are in line with previous findings that show accurate assignments with 18S (both full and short fragments) are restricted to family level for many eukaryotes (Lara et al., 2022) and for zooplankton taxa in particular (Tang et al., 2012; Wu et al., 2015).

False negatives in environmental samples: Aglantha digitale

In the Arctic, Aglantha digitale (Cnidaria: Rhopalonematidae) is one of the most frequently observed GZP species in net and optical surveys. This arctic-boreal species is predicted to undergo significant poleward shifts in distribution as the region continues to warm (Pantiukhin et al., 2024b). In the present study, A. digitale was well represented on COI reference databases and was successfully amplified with the primers for the 658 bp Folmer fragment. However, the Leray-XT primers failed to amplify it in the tissue samples, and it was largely absent from the COI metabarcoding datasets.

In silico analysis revealed that all tested A. digitale sequences had at least one mismatch on the Leray-XT forward primer binding region, with most sequences having four mismatches. Furthermore, the species has been absent from previous eDNA metabarcoding studies in the Arctic, despite being found in high abundances in nets at the same locations (Murray et al., 2025a, 2024) or with 18S markers (Murray et al., 2025b). In the present study, A. digitale was only detected in the COI-B metabarcoding dataset, which had a lower PCR annealing temperature to facilitate flexible primer binding. However, it occurred in very low read abundances (1% of GZP reads), despite being a significant component of the mock community in terms of biomass. This suggests that primer mismatches strongly reduced amplification efficiency, even under the less stringent PCR conditions of the COI-B dataset. A similar number of primer mismatches were found in silico for this species in a North Sea study with primer pairs targeting a shorter COI fragment, however they were still able to successfully amplify it in vitro (Günther et al., 2018). Based on our findings, we hypothesize that Leray-XT primer mismatches contribute substantially to false-negative detections of A. digitale in Arctic metabarcoding studies. This effect may be less pronounced in lower-latitude studies, where available locally collected reference sequences are more numerous, thus allowing for primers to be evaluated against sequences of more diverse populations.

Relative read abundances in mock community eDNA

Although linking read abundance to biomass was not the primary objective of this study, the COI datasets exhibited interesting patterns. Most reads were assigned to the species Atolla, the taxon with the highest biomass included in the experiment, suggesting a relationship between biomass and read abundance for this species. While this is consistent with the expectation that greater biomass yields higher read abundance, eDNA metabarcoding cannot distinguish between the effects of biomass and species-specific DNA shedding rates (Lamb et al., 2019; Shaffer et al., 2025). For example, mucus shedding, a potential source of eDNA in marine organisms (Andruszkiewicz Allan et al., 2021; Thomsen and Willerslev, 2015), can vary among taxa and increase under stress, as shown for Periphylla periphylla (Stenvers et al., 2023). Further quantitative work is therefore needed to determine how such processes influence eDNA read abundance across GZP taxa.

Another salient finding was the contrasting variability in relative read abundances between the two markers. The 18S dataset showed higher variability among triplicates, whereas COI was comparatively consistent across both datasets (COI-A and COI-B). This difference may be explained by higher copy numbers of mitochondrial genes such as COI compared to nuclear markers like 18S (Foran, 2006; Shaffer et al., 2025) and the greater susceptibility of low-copy targets to PCR-related biases (Elbrecht and Leese, 2015) Elbrecht et al., 2019. Thus, 18S may be less reliable for quantitative interpretation of GZP eDNA than COI. A similar pattern was observed in marginal ice zone samples, where 18S reads were over-dominated by calanoid copepods, while COI showed more consistent relative read abundances across replicates and sampling sites (Murray et al., 2025b). Although multi-marker approaches can improve taxonomic coverage, differences in marker amplification efficiency can complicate comparisons of read abundance and patterns related to biomass. While correlations between COI read abundance and biomass have been reported for some Arctic zooplankton (Ershova et al., 2021, 2023), limited data on GZP-specific patterns of eDNA shedding, degradation and dilution limits further quantitative interpretation here.

Non-target GZP assignments

Gelatinous zooplankton taxa not included in the mock community were recovered in both metabarcoding datasets (e.g., the large scyphozoan Cyanea capillata), even though the seawater used for the experiment was filtered several times to remove DNA, prior to starting. Furthermore, all reads detected in the seawater control were subsequently removed from the samples and the extraction and PCR controls showed no or negligible reads, none of which were assigned to GZP. The specimens were caught in plankton nets alongside other GZP species, and were not cleaned in any way before being placed in the aquarium and hence were likely carrying DNA from other species they were in contact with in the nets or in the water column. Therefore, we conclude that the source of the non-target DNA amplified here was highly likely the GZP individuals themselves (i.e., as natural samplers).

Which primers are best for metabarcoding studies on Arctic GZP?

Some studies have proposed the use of another mitochondrial marker, 16s rRNA (16S), as a more suitable alternative to COI for GZP in metabarcoding studies (Lindsay et al., 2015), particularly for hydrozoans (Zheng et al., 2014). Indeed, some of the species in the mock community that have few or no publicly available COI barcodes, are better represented with 16S barcodes in the reference databases. However, our MZGdb search revealed that substantially fewer 16S barcodes are publicly available compared with COI, despite 16S showing slightly higher species coverage for Arctic GZP taxa overall. Similarly, BOLD contained considerably more publicly available COI sequences than 16S for GZP species. The 16S marker presents several limitations, including indels that can affect sequences alignment (Hebert et al., 2003a), and limited availability of bioinformatic resources for detecting indels and pseudogenes, compared to COI (Braukmann et al., 2019). The conserved nature of 16S increases the likelihood of closely related species sharing the same sequence (Andújar et al., 2018), as observed for the hydrozoan family Pandeidae (Schuchert, 2018). Furthermore, widely used 16S primer sets and associated bioinformatic pipelines focused on metazoan metabarcoding remain limited. Thus, while 16S may be valuable for targeted species-level studies in the future, COI currently benefits from more extensive reference databases, primer options, and bioinformatic resources. For broad multi-species metabarcoding applications, we suggest that further optimization of COI markers and the expansion of reference barcode datasets for overlooked taxa and areas would be a more efficient use of resources.

Our results demonstrate that the COI Leray and 18S rRNA V1-V2 markers provide complementary information for metabarcoding Arctic GZP species. The COI Leray-XT primers enabled high-resolution species-level assignments for most of the recovered taxa, but performance was limited by primer mismatches and incomplete reference sequence coverage, resulting in the potential under-detection of some GZP taxa. In contrast, the 18S V1–V2 primers had broad taxonomic coverage across the targeted GZP groups, but the 18S gene lacked sufficient sequence variation to reliably discriminate among closely related genera and species.

While multi-marker metabarcoding can improve taxonomic coverage, as observed here at family level, it is not always economically viable and does not necessarily increase the ecologically meaningful information (i.e., species-level detections). It can also introduce further complications when interpreting relative read abundances and differences in taxonomic resolution, which can make direct comparisons between the markers impossible. Therefor the choice of primer pairs and target fragments for metabarcoding of GZP in the Arctic used should be determined by the study goal, taking the targeted taxonomic resolution and relevant reference database coverage into consideration. Finally, our findings highlight the urgent need to improve public reference databases through increased barcoding efforts, particularly for GZP taxa collected from the Arctic Ocean and its adjacent seas. Improved reference coverage globally is essential for realizing the full potential of DNA-based biodiversity assessments in polar marine ecosystems, particularly for elusive and dynamic taxa such as gelatinous zooplankton.