DNA metabarcoding of mites from small soil samples: limited agreement with morphological identifications but improved results from long-read sequencing

Introduction

Soil mites (Acari) are among the most diverse and ecologically significant groups of microarthropods in terrestrial ecosystems, playing crucial roles in decomposition, nutrient cycling, and soil structure maintenance (Coleman, Geisen & Wall, 2024). Due to their sensitivity to environmental changes, they are also widely used as bioindicators (Breure et al., 2005). Traditionally, the identification of soil mites has relied on morphological examination, which requires considerable taxonomic expertise and is often time-consuming due to the small size and morphological similarity of many mite taxa. Morphological identification of specimens is further complicated by the presence of juvenile stages, which frequently lack the diagnostic features for reliable species-level identification (Coleman, Callaham & Crossley Jr, 2017).

Molecular approaches, particularly DNA metabarcoding, has emerged as a powerful tool for biodiversity assessment. Metabarcoding has the potential to overcome several limitations of morphological identification, offering increasing throughput, the ability to detect cryptic and juvenile forms, and the possibility of standardizing biodiversity assessments across studies (Taberlet et al., 2018). By comparing morphological with metabarcoding-based identification of soil arthropods, Oliverio et al. (2018) and Ustinova et al. (2021) have reported correlating richness and community composition patterns among methods. Similarly, Young & Hebert (2022) found congruent diversity patterns for soil arthropods between soil eDNA and bulk arthropod samples. Therefore, metabarcoding approaches have been further applied in several studies for identifying the composition of arthropods in the soil (Andujar et al., 2022; Arribas et al., 2016; Perdomo-González et al., 2025; Sahdra et al., 2025).

However, soil arthropod metabarcoding studies have utilized second-generation sequencing technologies for sequencing short reads (158–418 base pairs; Oliverio et al., 2018; Andujar et al., 2022). Second-generation sequencing technologies, such as Illumina, allow for sequencing of hundreds of millions of reads with high accuracy, but are limited to read lengths of 2 × 300 base pairs. In contrast, third-generation sequencing platforms, such as PacBio, can generate longer reads, allowing more accurate taxa classifications (Tedersoo et al., 2021). Recent advances in third-generation sequencing technologies have substantially improved their accuracy, generating growing interest in metabarcoding approaches that target full-length DNA barcodes (Doorenspleet et al., 2025; Latz et al., 2022; Srivathsan et al., 2024), which typically exceed the read length limitations of second-generation platforms. Additionally, although previous studies have shown that morphological and metabarcoding datasets of soil arthropods may yield correlated results, it remains unclear how the quantity of soil used for DNA extraction influences this relationship.

The objective of this study was to determine whether the identification of mite communities from soil samples using short-read (Illumina) and long-read (PacBio) metabarcoding aligns with results from morphological identification. To this end, we used two commonly applied COI primer pairs: mlCOIintF/jgHCO2198 (Geller et al., 2013; Leray et al., 2013) to amplify a ∼313 bp fragment for Illumina sequencing, and LCO1490/HCO2198 (Folmer et al., 1994) to amplify a ∼658 bp fragment for PacBio sequencing. We further explored how the amount of soil (0.2 g and 2 g) used for DNA extraction influences mite and overall metazoan community detection in metabarcoding approaches.

Materials & Methods

Soil samples (including litter) were collected from five sites between July and September 2022 (Table 1). Six samples were taken from each site along a north-south transect, spaced three meters apart. The size of each sample collection site was 10 × 20 cm, with a depth of five cm. The collected material per sample was placed in zip-lock bags, then homogenized (thoroughly mixed in between hands) and divided into two equal ∼0.5 L parts. One part was used to extract soil animals with Tullgren funnel method for morphological identification, and other for metabarcoding. On the day of the sample collection, one sample replicate was subjected to Tullgren funnel heat extraction (7 days; 95% ethanol in the collection tubes), and the other replicate was dried in a clean paper bag in a drying cabinet with active airflow at 38 °C for at least 24 h (for metabarcoding).

Results

DNA extraction from one sample from site2 (site2_6) failed; thus, is excluded from all datasets. The average sequencing depth (from 29 samples, across five sites; Table 1) per sample was 161,472 for Illumina and 2,142 for PacBio. In total, the Illumina data yielded 11,883 OTUs, of which 1,517 (12.8%) were classified as Metazoa (with bootstrap on ≥0.8; Table S1). The rest of the OTUs were classified as fungi, bacteria, plants, or remained unclassified at kingdom level. The PacBio dataset yielded 12,357 OTUs with 2,118 (17.1%) metazoan OTUs (Table S2). The number of mite OTUs was 59 and 61 for the whole (0.2 g + 2 g treatments) Illumina and PacBio datasets, respectively. In the Illumina dataset, the number of mite OTUs was 44 in the 0.2 g treatment and 38 in the 2 g treatment, whereas in the PacBio dataset, 36 and 40 OTUs in the 0.2 g and 2 g treatments, respectively. However, a pairwise Wilcoxon rank-sum test demonstrated no significant differences between any pair of treatments (all p > 0.3; Fig. S1). The same was found when analyzing all metazoan OTUs in the metabarcoding datasets (all p = 1; Fig. S2).

From the 17 samples subjected to Tullgren funnel extraction followed by morphological identification of mites (across three sites; Table 1), the Illumina dataset revealed 42 mite OTUs in the 0.2 g treatment and 25 OTUs in the 2 g treatment samples. PacBio dataset hosted 20 and 33 mite OTUs in 0.2 g and 2 g treatments, respectively. However, a significantly higher mite taxon richness (morphotypes) was identified in the morphological dataset (p < 0.001, Fig. 1). A total of 105 morphotypes, were identified based on morphological examination (Table S3).

A total of 51 mite families were identified morphologically. In comparison, the Illumina dataset contained 22 families in total, with 13 families recovered from the 2 g and 21 from the 0.2 g treatments. PacBio dataset hosted 15 families overall, with 12 families detected in the 2 g treatment and 10 in the 0.2 g treatment. The Euler diagram (Fig. 2) illustrates the overlap and uniqueness of family-level detections among the different identification methods. The majority of families (32) were unique to the morphological dataset and not detected by either metabarcoding approach (Fig. 2).

Tullgren funnel extraction, followed by morphological identification showed that all samples contained mites (Fig. 1). However, metabarcoding missed mites from 1 to 6 samples (depending on the identification method; Fig. 1). For comparisons of mite community structure ordinations, we included only those samples in which mites were detected in all treatments being compared. Despite the substantially higher mite richness observed in the morphological dataset, Procrustes analysis revealed a strong concordance in the relationships among samples, as represented in ordination space, between morphological and PacBio datasets (Procrustes R = 0.807, p < 0.001 for 2 g; Procrustes R = 0.763, p = 0.007 for 0.2 g; Fig. 3A). In contrast, the correspondence between morphological and Illumina datasets was much weaker and not statistically significant (Procrustes R = 0.399, p = 0.289 for 2 g; Procrustes R = 0.334, p = 0.496 for 0.2 g; Fig. 3B).

Furthermore, we compared the relationships among samples from the PacBio and Illumina mite OTUs datasets. Procrustes analysis showed high and significant concordance between 0.2 g and 2 g treatments only within a metabarcoding dataset (Procrustes R = 0.887, p < 0.001 for PacBio 0.2 g vs. PacBio 2 g and Procrustes R = 0.859, p = 0.041 for Illumina 0.2 g vs. Illumina 2 g). There was a strong correlation also between PacBio 0.2 g and Illumina 0.2 g mite datasets (Procrustes R = 0.862), but this was marginally significant (p = 0.054). All other comparisons among the metabarcoding dataset revealed non-significant relationships among samples, as represented in ordination space (all p > 0.290 in Procrustes analyses). However, the analysis with all metazoan OTUs in the 0.2 g and 2 g PacBio and Illumina datasets (Tables S1 and S2) demonstrated strong concordance in the relationships among samples for all pairwise cases (Procrustes R > 0.923, p < 0.001). Similarly, when comparing metazoan OTUs datasets with the morphological data (only mites), strong concordance was also observed (Fig. 4; Procrustes R = 0.902, p < 0.001 between PacBio and morphological dataset; Procrustes R = 0.894, p < 0.001 between Illumina and morphological dataset).

Discussion

This study reveals insights into the effectiveness and limitations of DNA metabarcoding approaches when characterizing soil mite communities by using universal COI primers for generating amplicon libraries. Regardless of soil quantity input for DNA extraction, morphological identification of mite specimens revealed significantly higher richness across taxonomic levels compared with either metabarcoding approach. Despite that, our results demonstrated more consistent mite community composition patterns (in terms of sample similarities in an ordination space) between PacBio and morphological datasets, compared with Illumina and morphological datasets.

Although the morphological dataset contained higher mite richness, the metabarcoding approach detected additional taxa not identified by morphology, indicating complementary detection rather than a strict subset relationship. For example, species such as Steneotarsonemus laticeps (Halbert, 1923) and Sierraphytoptus ambulans (Chetverikov & Sukhareva, 2009) were only found in the PacBio dataset. Those species have never been recorded before in Estonia, but the closest records of S. ambulans are from Finland (Chetverikov & Sukhareva, 2009) and S. laticeps from Poland (Labanowski, Labanowska & Suski, 1990), suggesting rather under-sampling than erroneous metabarcoding records. It is known that some taxa may respond poorly to Tullgren funnel heat extraction (Coleman, Callaham & CrossleyJr, 2017), thus a complementary set of taxa from soil eDNA is expected (Young & Hebert, 2022).

Surprisingly, there were also substantial discrepancies in family-level diversity detected by different methods. Morphological identification revealed a total of 51 mite families, while metabarcoding approaches detected less than half this number (24 families for Illumina and 15 for PacBio; Fig. 2). Although the used COI primers aim to target metazoans, our metabarcoding datasets demonstrated that only 12–17% of all the OTUs were classified as Metazoa—an observation common in studies where universal COI primers are used to amplify soil eDNA (Anslan et al., 2021; Kirse et al., 2021; Watts et al., 2019; Young & Hebert, 2022). This is likely attributed to the diluted signal from soil mites, despite they are one of the most abundant groups of soil arthropods (Rosenberg et al., 2023). To increase the proportion of arthropod DNA for amplicon library generation, many studies extract the arthropods from soil samples prior to DNA extraction (Andujar et al., 2022; Arribas et al., 2021; Perdomo-González et al., 2025). However, when comparing results from morphological identifications of soil arthropods with metabarcoding extracted specimens, Basset et al. (2022) reported non-significant correlations between datasets, whereas other studies reported overall correlating richness and composition patterns between molecular and morphological approaches (Ustinova et al., 2021; Young & Hebert, 2022). Given the patchydistributions of soil arthropods (Bahram et al., 2016; Bardgett, 2002) the contrasting results from latter examples may stem from the different sampling methods: Basset et al. (2022) analyzed discrete paired samples (separated by ∼10 cm in the field), while Young & Hebert (2022) and Ustinova et al. (2021) homogenized the soil sample before extracting specimens and DNA. In our study, even with homogenized soil samples, morphological identification yielded considerably higher mite richness than either metabarcoding approach. While Oliverio et al. (2018) found similar results for mites, their study showed more consistent patterns between metabarcoding and morphological datasets for the entire arthropod community. This indicates that when a focus is on a specific soil arthropod group, such as mites, rather than the broader community patterns, methodological biases may become more pronounced. A similar observation has been noted, for example, in metabarcoding ostracods directly from lake sediment eDNA samples using universal COI primers (Echeverría-Galindo et al., 2021).

Despite detecting fewer mite taxa overall, the mite OTUs communities in PacBio dataset showed relatively strong concordance with morphological data in terms of the relationships among samples (Fig. 3A). Although the Illumina dataset had substantially higher sequencing depth and slightly higher mite Family richness, the Illumina dataset failed to align with the morphological data. This suggests that while both molecular methods are missing mite taxa, the longer reads generated may provide more reliable information for determining the community patterns. The better performance of long reads may not be attributed directly to the better performance of LCO1490/HCO2189 primers compared with mlCOIintF/jgHCO2198 as the overall mite OTUs richness was not significantly different between metabarcoding datasets. However, longer reads are less susceptible to amplifying relic, fragmented DNA (Taberlet et al., 2018). Consequently, the community patterns from long reads may reveal better representation of the active fauna, which may explain the stronger alignment with the morphological data derived from active mite extraction.

Within a set of 17 samples from three sites (used for the morphological comparison; Table 1), the Illumina 0.2 g dataset detected significantly more mite OTUs than the 2 g dataset. This is counterintuitive, as one would typically expect larger soil volumes to yield greater taxonomic richness, especially of metazoans, as demonstrated in previous studies (Kirse et al., 2021). While herein used PowerMax Soil Kit is designed for larger starting volumes, its protocol suggests eluting the DNA into a relatively large volume (up to 10 mL) compared to the PowerSoil Pro Kit, which uses a much smaller elution volume (up to 100 µl). This large elution volume in the PowerMax kit may result in a final DNA extract that is more diluted. Indeed, in our study, the average DNA concentration from the 2 g samples (PowerMax Kit) was 4.85 ng/µl, whereas the 0.2 g samples (PowerSoil Pro Kit) yielded an average concentration of 8.48 ng/µl. We hypothesize that this lower DNA concentration from the 2 g samples served as a less effective template during PCR amplification, leading to the detection of fewer mite OTUs. However, varying the initial soil amount for DNA extraction (0.2 g vs. 2 g) did not result in a statistically significant difference in mite OTU richness for either the Illumina or PacBio metabarcoding approaches across the 29 samples from 5 sites (Fig. S1). Despite that, the mite OTUs composition as represented in the ordination space significantly correlated only within each metabarcoding dataset (i.e., 0.2 g and 2 g treatments significantly correlated within PacBio or Illumina datasets) but not between them. This indicates that the choice of metabarcoding method (herein PacBio vs. Illumina) had a greater influence on the detected mite community composition than the amount of starting material. However, the importance of sample quantity may be critical at larger scales, as other studies have found that input material substantially greater than 2 g yields higher metazoan richness and significantly different community compositions compared with low inputs such as 0.2 g (Nascimento et al., 2018) or 0.5 g (Kirse et al., 2021).

In contrast to the results for mites alone, the analysis of the entire metazoan community showed high concordance across all metabarcoding treatments. Both community composition and total OTU richness were consistent across all treatments. A particularly interesting finding was the strong Procrustes correlation between the morphological mite data and the entire metazoan community from both metabarcoding datasets (Fig. 4). This indicates that soil mites serve as a powerful ecological indicator group (Gulvik, 2007), where the community structure of mites alone mirrors the broader patterns of the entire soil metazoan community. The environmental parameters shaping the mite assemblages appear to be similar to those for other soil animal groups, driving their structure. It also demonstrates that even very small quantities of soil used generally for soil microbial analyses can yield meaningful data about soil faunal communities. This validates the use of broad-spectrum metabarcoding for biomonitoring, as it suggests that whole community metabarcoding data can effectively reveal key ecological patterns across the landscape, much like those captured using traditional indicator species.

Conclusions

In conclusion, our study demonstrates that while soil eDNA metabarcoding with universal COI primers is a powerful tool, it has significant limitations for characterizing a specific taxonomic group like soil mites, capturing less than identified through morphological analysis. This suggests that for studies focusing on a specific faunal group, a preliminary extraction of the organisms from the soil is advisable which allows for the processing of a much larger soil volume than is typically feasible for direct DNA extraction, thereby increasing the likelihood of capturing a more comprehensive profile of the specific community of interest. But when working with soil eDNA, our results indicate that methodological choices (PacBio vs. Illumina) have a greater impact on mite community composition than simply input soil quantities (0.2 g, 2 g) for DNA extraction. The long-read PacBio dataset provided a more ecologically reliable community profile that aligned strongly with the morphological data. Finally, the patterns of the entire metazoan community strongly mirrored those of the morphologically identified mite communities alone. This validates the role of mites as powerful ecological indicators and demonstrates that broad-spectrum metabarcoding is a highly effective tool for large-scale biomonitoring, capable of revealing key ecological patterns even without perfect taxonomic resolution of every single group.

Supplemental Information