Our results showed that the amount of sampled sediment varied greatly amongst studies and was often not reported clearly. The amount of sampled sediment may be important especially when DNA is extracted from isolated meiofaunal specimens or from a subsample of homogenized sediment and should thus be reported consistently in addition to the amount of sediment used for DNA extraction. In 92% of studies in which the sampling depth of the sediment layer was reported, the upper 10 cm of sediment or less was sampled, corresponding to the most biologically active part of the sediment in which meiofauna generally tends to concentrate (Thiel, 1983; Simpson and Batley, 2016).

Sample preservation is a critical step to avoid DNA degradation and was consistently reported in all but three studies. Our survey highlights that freezing or freeze-drying is the predominant method used, without fixatives such as ethanol or DESS. However, when a fixative was used this mostly comprised isolated meiofauna samples rather than sediment. Consistently, van der Loos and Nijland (2021) report that freezing is often used for sediment samples, and recommend the use of DESS over ethanol for bulk samples as it preserves both DNA and morphology with higher quality.

DNA extraction was most often performed directly from sediment samples compared to specimens isolated from sediment prior to DNA extraction. Compared to isolation of meiofauna, using sediment samples may be easier, it limits processing time, it avoids contamination, and it does not introduce any bias due to different methods of specimen isolation (Fonseca et al., 2018; Fais et al., 2020b). Our survey highlights that sediment samples are dominant in studies targeting environments hard to sample, such as the deep sea (e.g., Sinniger et al., 2016; Atienza et al., 2020; Kitahashi et al., 2020), and muddy and fine sediments difficult to elutriate (i.e., separating meiofauna from sediment by specific gravity using a liquid stream, Somerfield et al., 2005; viz., Lanzén et al., 2017; Faria et al., 2018; Salonen et al., 2019; Brandt et al., 2020). In addition, studies including a wider size range of benthic organisms such as micro- and macrofauna assessed through metabarcoding of sediment eDNA (e.g., Xie et al., 2017, 2018; Fais et al., 2020a,b; Klunder et al., 2020a,b) might have artificially raised the number of studies in which DNA was directly extracted from sediment samples. This method may lead to partially misleading results on the species composition present, as DNA in sediment can be recent or older extracellular DNA (Thomsen and Willerslev, 2015). This is especially relevant in deep sea environments, where extracellular DNA can constitute for 50–90% of the DNA, and ancient DNA can be very well conserved because of adsorption onto the sediment matrix (Torti et al., 2015). On the contrary, isolation of meiofaunal specimens from the sediment allows researchers to process a larger volume of sediment per sample, thereby possibly reducing sample heterogeneity and increasing recovery of rare species (Fonseca et al., 2010; Brannock and Halanych, 2015). Sieving reduces the dominance of large taxa (Cowart et al., 2015; Fonseca et al., 2018; He et al., 2020). Furthermore, elutriating or sieving samples was reported to result in a relatively high abundance of metazoan taxa while reducing the amplification of non-target bacterial or non-benthic taxa (Brannock and Halanych, 2015; He et al., 2020). When isolating meiofauna from the sediment, careful consideration is needed as soft-bodied organisms are easily damaged, whereas relatively heavy hard-bodied taxa such as mollusks may be retained in the sediment during elutriation, possibly leading to an underrepresentation of certain taxonomic groups (van der Loos and Nijland, 2021).

Recently, Castro et al. (2021) compared the recovered community composition from DNA extracted directly from sediment cores with two meiofauna extraction methods commonly used in metabarcoding studies, namely (1) separation of meiofaunal specimens from sediment by flotation and centrifugation with Ludox and (2) separation of meiofaunal specimens from sediment by decantation of specimens with MgCl₂. The community composition recovered through direct DNA extraction from sediment differed strongly from that obtained with the isolation techniques (Castro et al., 2021). Isolation of specimens with Ludox or MgCl₂ led to more taxa and meiofaunal species compared to DNA extraction directly from sediment (Castro et al., 2021). Differences were also found amongst isolation techniques, as some genera belonging to the phyla of platyhelminthes, annelids, mollusks, xenacoelomorpha, gastrotrichs, and nematodes were found in a lower abundance in samples obtained through MgCl₂ decantation relative to Ludox flotation (Castro et al., 2021).

Hence, as sample type significantly affects both the OTU richness and taxonomic composition of meiofaunal communities (Brannock and Halanych, 2015; Haenel et al., 2017; Nascimento et al., 2018; Castro et al., 2021), a greater consideration should be given to the choice of sample and DNA extraction workflow. Indeed, any extraction method will lead to a bias toward taxonomic groups that are more easily isolated with the chosen technique. Therefore, if the goal is to maximize recovery of taxa, a combination of isolation methods is the best option (Castro et al., 2021).

We found that when DNA was extracted directly from sediment, researchers used kits specifically designed for soil samples, while these kits were used in only two-thirds of studies that extract DNA from meiofauna isolated from the sediment (Figure 1). Humic substances often present in sediments can be retained during DNA extraction and inhibit enzymes that amplify DNA during PCR, which can lead to false negatives (Matheson et al., 2010). Soil DNA extraction kits specifically remove these substances, thereby alleviating PCR biases caused by enzyme inhibition. Nonetheless, in earlier studies, clean isolated meiofauna samples and good amplification results were obtained using non-soil kits (e.g., Creer et al., 2010; Fonseca et al., 2010). More recently, it was recommended to use a soil-specific kit in samples that may contain remains of sediment to obtain high-purity DNA (van der Loos and Nijland, 2021), which also works well for marine and coastal sediments (Lekang et al., 2015; Taberlet et al., 2018; Fais et al., 2020b). The two most used kits in our analysis (i.e., MoBio/QIAGEN PowerSoil and PowerMax kit) differ in the amount of sediment that can be processed (a maximum of 0.25 and 10 g, respectively). Indeed, we found that in the analyzed studies DNA is mostly extracted from 5 to 10 g or 0.2 to 0.35 g, corresponding to the processing maxima of these kits. As some studies have shown that the number of meiofaunal taxa and OTUs increase with sample size where no elutriation steps are performed (Nascimento et al., 2018; Brandt et al., 2020; Fais et al., 2020b), 10 g of sediment is recommended for eDNA extraction. However, we point out that the QIAGEN PowerMax is costly (about 25€/USD per extraction), thereby limiting the number of samples that can be processed in many projects, as reflected in the number of extraction replicates found in this study. None of the studies that used the PowerMax kit used multiple extraction replicates, while these were reported using the PowerSoil kit (QIAGEN). Ultimately, several replicate DNA extractions and a lysis step are strongly suggested to improve OTU and taxa richness in studies on marine meiobenthos through DNA metabarcoding approaches (Lanzén et al., 2017). Additionally, it is recommended to perform at least three replicate PCR reactions to correct for the effects of PCR stochasticity and to potentially increase the number of OTUs and species detected (Leray and Knowlton, 2015; Bourlat et al., 2016; Alberdi et al., 2018), which was performed in less than half of the studies and not reported in 36% of studies analyzed here. Again, whether PCR replicates improve the results and outweigh the costs depends on the research goal (van der Loos and Nijland, 2021).

Even though negative extraction and PCR controls are essential control measures within metabarcoding, their use was reported in less than half of the studies analyzed here. Whether these controls were truly not implemented in the analyses, or this result only reflects a gap in reporting remains unknown, but either way this indicates an important point of improvement (van der Loos and Nijland, 2021). Field extraction controls (i.e., opening sample storing material at the sampling site without sample collection), no-template DNA extraction controls, no-template or water-only PCR controls all aim to identify potential contamination introduced in any step of the workflow and should therefore be processed and sequenced along the field samples and controlled for during bioinformatic quality filtering steps (e.g., Brandt et al., 2021a).

We found that the nuclear 18S rRNA molecular marker was the most commonly used, especially when only one marker region was amplified (n = 44 for 18S and n = 1 for COI). Most studies did not mention the reason behind marker region choice. Amongst the reasons for the choice of marker, sequencing length limitations of the Illumina platform was sometimes mentioned as a reason for selecting the 18S rRNA V9 region (Brannock et al., 2014, 2018; Brannock and Halanych, 2015). In other studies, 18S V4 was chosen over COI because this region is shorter and easier to PCR amplify and sequence using Illumina, and amplifies only eukaryotic sequences (Lejzerowicz et al., 2015; He et al., 2021). Furthermore, poor amplification results for marine nematodes with the COI Folmer primers in early barcoding studies (Derycke et al., 2010) have stimulated the use of 18S rRNA gene regions as genetic markers. The more preserved nature of 18S rRNA primer binding sites compared to COI allows for the amplification of a broader range of taxa (Schenk and Fontaneto, 2020). However, the relatively high conservation of 18S priming sites comes at the cost of resolution, often lacking the ability to discriminate at the species level (Tang et al., 2012).

Despite the extensive use of the 18S rRNA marker to study meiofauna, the mitochondrial COI gene is becoming more popular for meiofauna metabarcoding studies (28% of the studies analyzed in this review). Compared to 18S rRNA, the main advantage of COI is its high taxonomic resolution, often allowing for identification of species and even intraspecific variability (Tang et al., 2012; Cowart et al., 2015; Leray and Knowlton, 2016; Turon et al., 2020). Notwithstanding, meiofaunal species are often underrepresented in molecular reference databases (Sinniger et al., 2016; Curry et al., 2018; Wangensteen et al., 2018), leading to high percentages of uncharacterized OTUs. For example, 70.9% of the eukaryotic clusters remained unassignable at the phylum level in a deep-sea study by Atienza et al. (2020). However, it is important to consider that a high number of unassigned reads in COI datasets can be a result of non-target amplification of bacterial and other non-target species (Weigand and Macher, 2018). Therefore, both commonly used markers have their pitfalls and the choice of the most appropriate one depends on methodology, study system and research objective, as is true for all steps in the metabarcoding workflow. Using multiple molecular markers within a single study is advisable to maximize the recovery of meiofaunal taxa (Fais et al., 2020b and references therein), which is now implemented by less than a third of the studies analyzed. Furthermore, efforts should be undertaken to barcode individual species and place molecular references in barcode reference libraries to improve our ability to taxonomically assign metabarcoding data (Wangensteen et al., 2018; Weigand et al., 2019).

Optimal primer pairs for amplification of whole meiofaunal communities ideally amplify a broad range of meiofaunal animals while minimizing off-target amplification. Ideally, extensive reference databases containing sequences for this marker should be available to allow for high taxonomic resolution. Our data showed that a total of 22 different primer combinations were used in the studies analyzed in this review. This high number can be partly explained by studies that test multiple primer pairs, but it also reflects that a consensus on the optimal primers for amplification of marine meiofaunal communities has not been reached. Primer choice is important as primer bias due to mismatches affects the relative abundance of amplicons and species recovery (Elbrecht and Leese, 2015). Thus, conserved sequences facilitate universal primer design. As such, the SSU_F04 forward and SSU_R22 reverse primer set (Blaxter et al., 1998; Fonseca et al., 2010) targeting the 18S rDNA V1-V2 region is highly conserved and amplifies a broad range of meiofaunal organisms (Creer et al., 2010). In meiofaunal taxa, the V1–V2 region amplified by these primers is the most variable nSSU region and its length of 450 bp favors amplification of DNA from living organisms (Lallias et al., 2015), benefiting recovery of present meiofaunal communities. Further modifications, such as shortening and including an ambiguity (R) into the SSU_R22 reverse primer to accommodate the inclusion of more metazoans (called SSU_R22mod, Sinniger et al., 2016), can be used to target specific groups of taxa within meiofauna.

After 18S rRNA V1–V2, the V4 and V9 were commonly used to amplify eukaryotes. The 18S rRNA V4 region is the most variable region in eukaryotes (Nickrent and Sargent, 1991). Indeed, the 18S V4 primer pairs are designed to amplify a broad range of eukaryotic taxa [F566 and R1200 by Hadziavdic et al. (2014); Uni18S and Uni18R by Zhan et al. (2013); E572F and 897R by Comeau et al. (2011) and Hugerth et al. (2014)] or eukaryotic microbes [TAReuk454FWD1 and TAReukREV3 by Stoeck et al. (2010)] instead of meiofauna taxa specifically. The same goes for the eukaryotic primer set 1380F and 1510R targeting the V9 region (Amaral-Zettler et al., 2009). The short (100–110 bp) regions that are amplified using the 18S_allshorts primer set targeting 18S rRNA V7 are suitable for amplifying extracellular DNA due to its length, however limiting meiofaunal identification at species level (Guardiola et al., 2015, 2016).

The 658-bp fragment amplified by the original primer pair for the mitochondrial COI gene (LCO1490 and HCO2198, Folmer et al., 1994) is too long for commonly used Illumina sequencing platforms. The shorter mlCOIintF and dgHCO2198 (Meyer, 2003; Leray et al., 2013) and mlCOIintF and jgHCO2198 (Geller et al., 2013; Leray et al., 2013) are most commonly used for amplification of marine meiobenthos. However, the degenerate LoboR1 primer (Lobo et al., 2013) is a promising reverse primer in combination with the mlCOIintF forward primer, as it uncovered a higher diversity compared to the degenerate reverse primer dgHCO2198 (Haenel et al., 2017). This combination was used and recommended in several meiofaunal community studies (Haenel et al., 2017; Fais et al., 2020a,b; Castro et al., 2021). Additionally, inosine bases present in the widely used jgHCO2198 reverse primer may impair the performance of high-fidelity polymerases commonly used for PCR amplification in metabarcoding studies (Jungbluth et al., 2020). For analyses of meiobenthic communities extracted from the sediment, using highly degenerate primers that amplify the maximum number of taxa, such as the Leray and Lobo primers (Meyer, 2003; Leray et al., 2013; Lobo et al., 2013), may be beneficial. On the other hand, these primers can be suboptimal for amplification and sequencing of DNA extracted directly from sediment, as the primers can amplify a wide range of non-target microbial taxa abundantly present in the sediment preventing detection of target taxa (Alberdi et al., 2018; Weigand and Macher, 2018). Alternative methods such as elutriation or sieving, as suggested before, could reduce non-target amplification. Alternatively, less degenerate, more specific primers can be beneficial, at the cost of potentially losing some target taxa.

Consequently, no optimal primer pair is currently available for marine benthic meiofauna. To increase comparability between studies, researchers should aim to narrow down the amount of different primer combinations that are used in meiofaunal community studies, while considering the best primer pairs for their research objectives.

Primer pair selection eventually affects the choice of sequencing platform, and vice versa. Illumina MiSeq is currently the main sequencing platform used in meiofaunal community DNA metabarcoding studies and can produce 2 × 300 bp long reads, rendering the use of primer pairs that generate longer amplicons unfeasible. The 454 pyrosequencing platform was identified as the main sequencing platform in Carugati et al. (2015). In our review, we found that this platform was mainly used in older papers, as Roche discontinued 454 platform production in 2013 and support by 2016. It was recently shown that Illumina’s NovaSeq 6000 outcompetes the MiSeq platform in a seawater eDNA metabarcoding study, where NovaSeq detected a higher diversity at the same sequencing depth. Moreover, where the number of new sequence variants detected by the MiSeq platform leveled off at approximately one million reads, data from NovaSeq did not show such a plateau (Singer et al., 2019). At ten times the costs of MiSeq sequencing, however, the NovaSeq 6000 platform may be inaccessible for most studies (Singer et al., 2019) and was indeed not used in the studies analyzed in this review. However, costs might decrease and the platform, which can produce 2 × 250 bp long reads, might become a cost efficient option for future metabarcoding studies. In contrast, the popular Illumina MiSeq platform has relatively low costs per sequenced base pair. Sequencing techniques such as PacBio and Nanopore allow sequencing of longer amplicons (Callahan et al., 2019) and have been successfully used for metabarcoding of environmental samples (Heeger et al., 2018; Karst et al., 2018; Davidov et al., 2020; Okazaki et al., 2021). Even though trade-offs regarding the higher error rate and read coverage compared to Illumina short-read sequencing exist, longer reads comprising several genes may provide better taxonomic resolution due to the high informational content per read (Weirather et al., 2017; Baloğlu et al., 2021). To our knowledge, these sequencing platforms have not been used in marine meiofaunal metabarcoding studies and might thus be valuable options for future studies (Hebert et al., 2018).

Diversity detection increases with sequencing depth in metabarcoding studies, as shown for different target taxa (e.g., Smith and Peay, 2014; Lanzén et al., 2017; Alberdi et al., 2018). Sequencing depth increases the number of unique OTUs per PCR replicate, and thereby decreases similarity between PCR replicates (Alberdi et al., 2018). If a study aims to identify the whole meiofaunal diversity, a high sequencing depth is crucial. However, for general monitoring purposes deep sequencing depth may not be required. Nonetheless, a sufficient sequencing depth should be considered fitting the purpose of the study. Singer et al. (2019) inferred sequencing depth of 10 metabarcoding studies using DNA from various sources and found that an average sequencing depth of 55,000–60,000 reads per sample is common. Indeed, they also found that sequencing depth is often not reported clearly. Similar to the use of highly degenerate primers, detection of more species comes at the cost of an increasing number of artifactual or non-target sequences and should thus be considered during bioinformatic processing (Alberdi et al., 2018; Weigand and Macher, 2018).

It has been shown by multiple studies that bioinformatic techniques affect recovered diversity (Brannock and Halanych, 2015; Leasi et al., 2018; Antich et al., 2021). We found a total of 23 different bioinformatic approaches used in studies on marine benthic meiofauna, which can potentially render comparison of results between studies unfeasible. The clustering methods (i.e., combining sequences into representative units such as OTUs or MOTUs) analyzed in this study are mostly based on either Bayesian methods (CROP; Hao et al., 2011), single linkage clustering (SWARM; Mahé et al., 2015), or similarity thresholds. The latter was most commonly applied in the reviewed studies, often with a cut-off value of 97% similarity. Denoising approaches without clustering (i.e., rectifying sequencing errors by merging sequences containing an error with the true sequence) as done in DADA2 (Callahan et al., 2016) or UNOISE (Edgar, 2016) pipelines, yields ASVs or zero-radius OTUs (ZOTUs), which allow studying intraspecific diversity. Similarly, but not widely implemented in meiofauna research, the Deblur pipeline (Amir et al., 2017) obtains single-nucleotide resolution and produces comparable weighted results to DADA2 and UNOISE2 (Nearing et al., 2018) and could be valuable for identification at (sub)species level.

Currently, the debate whether OTUs or ASVs are more appropriate for analyzing metabarcoding data is ongoing. It is argued that ASVs, due to their higher resolution and reusability across studies, should be the standard reporting unit (Callahan et al., 2017), but other researchers argue that clustering is crucial in more variable markers such as COI to conform to the biological species concept (Antich et al., 2021; Brandt et al., 2021b). Furthermore, denoising and clustering programs that were originally developed for rRNA (e.g., CROP, SWARM, DADA2, UNOISE) are now commonly used for the mitochondrial COI gene (Antich et al., 2021). Using these programs on sequences from variable markers such as COI demands for a critical consideration of parameter settings as done for SWARM2 (Turon et al., 2020; Brandt et al., 2021b) and CROP (Leray et al., 2013; Leray and Knowlton, 2015; Wangensteen et al., 2018). However, not all studies provide the used parameter settings, implying that the default settings were used without further consideration (Antich et al., 2021).

It has been pointed out in many studies that taxonomic assignment of recovered sequences is hampered by significant gaps in reference databases (Sinniger et al., 2016; Curry et al., 2018; Wangensteen et al., 2018; Weigand and Macher, 2018), leading to difficulties especially for the interpretation of metabarcoding data at lower taxonomic levels. The COI gene tends to be more variable than rRNA genes like 18S, meaning that for reliable identification on a low taxonomic level, representatives of the species or at least closely related species have to be present in a reference database (Leray and Knowlton, 2016). Moreover, databases are often limited to few genetic markers (e.g., SILVA, which contains rRNA sequences, BOLD, which is focused on COI), contain many sequences that are not openly available and may contain erroneous or wrongly assigned sequences (e.g., Radulovici et al., 2021). Combining multiple reference databases could therefore improve the number and accuracy of OTUs assigned within a study (Macher et al., 2017). However, this was only implemented by a limited number of studies analyzed here. Almost two-third of studies use the ribosomal RNA sequence database SILVA (Pruesse et al., 2007) for taxonomic assignments, corresponding to 18S being the most used marker in meiobenthic metabarcoding studies. Surprisingly, BOLD was only used in n = 2 studies. Studies targeting COI tend to rely more on GenBank, as some taxonomic groups, including annelids and mollusks, are better represented in GenBank compared to BOLD (Weigand et al., 2019).

Overall, implementation of a mock sample with known species composition could confirm the reliability and accuracy of the metabarcoding workflow, including bioinformatic processing, and can be used not only to verify the detection of mock species but also control for overestimation of OTUs. Of the studies reviewed, only n = 6 studies (e.g., Chariton et al., 2015; Klunder et al., 2019; Brandt et al., 2020) included a mock community, and routinely sequencing such a control sample might be beneficial for future studies.