Background

Water mites (Hydrachnidia) are highly diverse in aquatic habitats [1]. They have a complex life-cycle that includes a prelarva, a parasitic larval stage, an initial resting stage (protonymph), a free living deutonymph, a second resting stage (tritonymph) and the final adult stage [2]. Hydrachnidia disperse predominantly through passive rather than active pathways because water mite larvae parasitize other invertebrate taxa, generally insect hosts that fly [2,3,4,5].

Nearly all freshwater environments are inhabited by water mite species with a high degree of habitat specialization [6]. Mites in springs and other groundwater-influenced ecosystems occur in remarkably high diversity of habitats [7,8,9]. Of the 970 recorded European water mite species, 137 are found solely in or near springs and are adapted to several microhabitats, such as different substrate types and environmental conditions [2, 9]. Due to the high degree of adaptation and their influence on ecosystem functioning for other invertebrate taxa [10,11,12,13,14], these so called crenobiontic (occur exclusively in spring habitats) and crenophilous (tendency to be found in the spring brook) species play a critical role in spring species communities. Considering that springs are island-like habitats within an uninhabitable terrestrial matrix [15, 16], spring dwelling water mite populations are assumed to be rather isolated. This would promote reproductive isolation and therefore lead to an increased speciation rate [4, 17], which is among other things an explanation for the relatively high species diversity of water mites in springs. However, the degree of isolation of spring water mite populations is highly dependent on the dispersal abilities of their hosts and influenced by taxon specific host spectra and specificity [5, 18, 19]. Furthermore, the high microhabitat diversity in springs [15, 20], their relatively stable environmental conditions [21, 22] and absence of large predators, e.g. fish [9], make them exceptionally favorable habitats for insect larvae and benefit their development. Therefore, the diversity and abundance of water mite hosts is relatively high in springs compared to other freshwater habitats, which is likewise considered as precondition for the high number of crenobiont water mite species [9].

Despite their importance for freshwater species communities, the taxonomic knowledge about Hydrachnidia species is still limited today and new species are discovered frequently (e.g. [23,24,25,26]). The intensity of re-examination of European Hydrachnidia has increased over the past years with several taxonomic revisions published [9]. Most water mite species known today have been described based on morphology only and studies applying genetic methods to verify and complement these descriptions are still relatively scarce (e.g. [27,28,29,30]). Nonetheless, many studies have shown that genetic species delimitation, frequently using cox1 barcoding [31], has a large potential to reveal new species, resolve taxonomic questions and contribute to biodiversity baselines and assessments (e.g. [29, 32,33,34,35]). Species identifications and their ecological interactions are crucial for contributions to crenobiology, community ecology, developing reliable bioindicators and understanding population dynamics. Moreover, newly developed methods to monitor invertebrate assemblages in freshwater environments, such as the simultaneous identification of bulk sampled individuals (metabarcoding) [36] or the indirect community reconstruction by analyzing environmental DNA (eDNA) [37], rely on previously established genetic species reference databases. Thus, a proper taxonomical knowledge and species description is greatly needed.

Several factors are important to account for when using genetic species identification methods [38], such as the presence of endosymbionts like the alpha-proteobacteria Wolbachia sp. [39, 40] or the presence of pseudogenes and nuclear copies of mitochondrial DNA (numts) [41,42,43], which compromise the suitability of mitochondrial molecular markers to identify species. Standard barcoding methods are mainly based on sequence similarity and the relation between intra- and interspecific genetic distance, which is commonly calculated by using the Kimura 2-parameter (K2P) [44] and uncorrected (p) distances [31, 45, 46]. Nonetheless, the usability of genetic markers to identify species can vary between different taxonomic groups, geographical origin and sampling strategy [31, 38, 47], which implies a taxon-specific evaluation prior to a broad-scale application in environmental assessment and conservation.

In this study we use an integrative taxonomy approach to verify the species status and validity of the most common spring related Hydrachnidia species in Europe. We tested the reliability of techniques commonly used to identify and delimit species using fragments of the mitochondrial cytochrome c oxidase subunit 1 gene (cox1) [31, 48] and the D1-D2 region of the LSU rDNA gene (28S) [49], or using both (e.g. [28, 50,51,52]). This study aims at improving knowledge and analytical techniques for assessing Hydrachnidia diversity in springs and explores the strength and weaknesses of standardized barcoding loci to identify water mite species.

Methods

Water mite sampling

The studied crenobiontic and crenophilous Hydrachnidia species were sampled between 2008 and 2018 in 87 different sampling sites across Europe during multiple sampling occasions (Additional file 1: Table S1). Most of the specimens were collected from springs located within the protected areas Berchtesgaden National Park (Germany), Black Forest National Park (Germany), Gesäuse National Park (Austria) and in the Swiss National Park (Switzerland). Samples were manually collected with a 200-µm hand net. Water mites were either sampled alive in the field or sorted out in the laboratory under a stereomicroscope from mixed samples containing bulk substrate. All specimens were subsequently stored in ethanol (100%) and kept at 4 °C until further processing.

Non-destructive DNA extraction

Total genomic DNA (gDNA) of each individual mite was extracted by using either GeneReleaser® (BioVentures, Murfreesboro, TN, USA) or the DNeasy Blood & Tissue Kit (Qiagen, Hilden, Germany). Both methods allow a non-destructive DNA extraction, which is essential when voucher specimens need to be retained for morphological identification in barcoding projects. Prior to both extraction methods, each individual was cleaned by using forceps and entomological needles in a small Petri dish filled with ethanol (100%) under a stereomicroscope. All instruments and vessels used were cleaned after processing each mite specimen by rinsing it with sodium hypochlorite (13%), molecular grade water and ethanol (80%). Afterwards, the specimens were air dried and soaked in molecular grade water for 3 min to ensure the absence of ethanol residues. The cleaned specimen was then transferred to either a 0.2-ml PCR tube containing 0.9 µl of molecular grade water and 0.1 µl of 1× PCR buffer (Qiagen) in the case of GeneReleaser® or to a 1.5-ml tube containing 180 µl of buffer ATL (Qiagen) and 20 µl (20 mg/ml) of Proteinase K (Qiagen) when using the DNeasy Blood & Tissue Kit. The GeneReleaser® method was conducted as originally described by Schizas et al. [53] and modified by Böttger-Schnack & Machida [54], see also [55]. Instead of resuspending the supernatant in TE buffer, step 6 of the modified protocol [54], approximately 12 µl of supernatant was transferred into a new 0.2-ml PCR tube and used directly as DNA template for the subsequent PCR reactions. The DNeasy Blood & Tissue extraction was performed according to the manufacturer’s protocol (Animal Tissues, Spin-Column Protocol, Qiagen) with minor changes. The specimens were incubated in buffer ATL and Proteinase K at 56 °C on a shaking thermomixer (400× rpm) overnight (step 2 in the manufacturer’s protocol) and the elution buffer (AE) volume was decreased to 100 µl in the last step to increase the gDNA concentration. The concentration of every DNA template was measured after the extraction by using a Qbit 3.0 Fluorometer (Thermo Fisher Scientific, Waltham, MA, USA) and the dsDNA HS Assay Kit (Thermo Fisher Scientific). After the first processed specimens it was evident that the mean amount of total gDNA obtained by the DNeasy procedure is higher (mean ± SD: 71.5 ± 2.3 ng in 100 µl of solution, n = 92) than when extracting gDNA by using the GeneReleaser® method (mean ± SD: 56.8 ± 4.8 ng in 12 µl of solution, n = 105). Therefore, the DNeasy method was chosen for all subsequent extractions.

Morphological examination

All water mite individuals were identified morphologically by the authors RG and LB using current Hydrachnidia identification keys [56,57,58]. After the DNA extraction, the mite specimens were dissected and mounted on slides in Hoyer’s medium or identified as whole individuals under a compound microscope when possible. The enzymatic DNA extraction method (Proteinase K) leads to a partial digestion of the specimens causing discoloration and therefore improved visibility of morphological characters, especially of sclerotized parts. This often allows the morphological identification without dissection. However, digestion is a process that affects membranous parts and therefore deteriorates the observability of integument structures such as the papillae, tubercles or lining. All voucher specimens are stored in the acarological collection of the Natural History Museum Basel (Switzerland) under the museum identifications presented in Additional file 1: Table S1.

PCR amplification and sequencing

The approximate 650 bp standard barcoding fragment of the cytochrome c oxidase subunit 1 (cox1) [31] mitochondrial gene was first amplified by using universal primers LCO1490 and HCO2198 [59] of a subset of Hydrachnidia species belonging to several genera (Atractides, Feltria, Hygrobates, Lebertia, Partnunia, Protzia and Sperchon). PCR reactions contained 0.25 µl of Phusion High Fidelity DNA Polymerase (2 U/µl) (Thermo Fisher Scientific), 5 µl of 5× Phusion HF Buffer (Thermo Fisher Scientific), 0.5 µl of dNTP mix (10 mM) (Sigma-Aldrich, Buchs, SG, Switzerland), 1.25 µl of forward and reverse primers (10 µM each), 5 µl of template DNA and ultrapure water to a total reaction volume of 25 µl. The PCR conditions were as follows: initial denaturation for 30 s at 98 °C; 35 cycles of 10 s at 98 °C, 30 s at 50 °C and 30 s at 72 °C; final elongation for 2 min at 72 °C. PCR products were then stored at 4 °C. As this primer combination failed to amplify several samples we decided to design new genus-specific cox1 barcoding primers on the basis of the previously amplified water mite specimens. This was possible due to a low sequence variability at the 5’ and 3’ ends of the barcoding fragment.

Additionally, a new set of universal barcoding primers was designed by degenerating and modifying several positions of the original LCO1490/HCO2198 primers to enable a higher amplification performance when barcoding Hydrachnidia species. These new primer sets (Table 1) were used to amplify all remaining samples. The subsequent PCRs were performed by using 0.25 of µl Q5® High-Fidelity DNA Polymerase (2U/µl) (NEB, Ipswich, USA), 5 µl of 5× Q5® Reaction Buffer (NEB), 0.5 µl of dNTP mix (10 mM) (Sigma-Aldrich), 1.25 µl of forward and reverse primers (10 µM each), 5 µl of template DNA and ultrapure water to a total reaction volume of 25 µl. The PCR conditions were the same for all newly designed cox1 primer sets and were as follows: initial denaturation for 30 s at 98 °C; 35 cycles of 10 s at 98 °C, 30 s at 51 °C and 20 s at 72 °C; and a final elongation step for 2 min at 72 °C. The PCR products were then kept at 4 °C until further processing. To amplify the D1-D2 domain of the LSU rRNA 28S gene we designed new water mite-specific primer sets on the basis of the D1D2fw2 forward primer [49] and by aligning different 28S Hydrachnidia sequences downloaded from GenBank. The new primers 28SHy_F and 28SHy_R (Table 1) reliably amplified an approximately 1.2 kbp long fragment of the 28S D1-D2 domain. The 28S PCR reactions were done by using the same reaction components and conditions as used when amplifying with the new cox1 primer sets. The only difference was a higher annealing temperature at 68 °C instead of 51 °C. All PCR primers (cox1 and 28S) were tailed with modified M13 sequences (M13: 5′-TGT AAA ACG ACG GCC AG-3′ and M13r: 5′-CAG GAA ACA GCT ATG AC-3′) [60], which has shown to improve the amplification and sequencing reactions in this and previous studies [61, 62]. Prior to sequencing, PCR products where examined on an agarose gel electrophoresis and purified using the QIAquick PCR Purification Kit (Qiagen) according to the manufacturer’s protocol on a QIAcube (Qiagen). The purified PCR products were Sanger sequenced with the above mentioned M13 primers by Mycrosynth AG (Balgach, Switzerland).

Table 1 Primers designed and used in this study
Full size table

Molecular analysis

Raw sequences were analyzed, edited and aligned in Geneious Prime v.2019.1.1 [63]. Low-quality base calls, ambiguous sites and primer binding sites at the 5′- and 3′-ends were trimmed prior to further processing. Alignments were done by using MAFFT v.7.388 [64] implemented in Geneious Prime. Potentially poorly aligned positions and divergent regions of the alignments were eliminated with Gblocks v.0.91b [65, 66]. All sequences were tested for the presence of contaminants by blasting with the Nucleotide Blast Tool (BLASTn) implemented on the NCBI website [67]. Because misleading numts can be amplified in PCRs targeting cox1 mtDNA, we translated the sequences into amino acids to check for the presence of stop codons, which is commonly seen as a suitable way to detect erroneous amplification [41]. The concatenated alignment containing cox1 and 28S sequences was generated by Sequence Matrix v.1.8 [68]. The suitable nucleotide substitution model for each marker (cox1: TPM2uf+I+G4 and 28S: TVM+I+G4) was selected according to the Bayesian information criterion (BIC) as implemented in ModelTest-NG v.0.1.5 [69], a novel software, which combines features of jModelTest2 [70] and ProtTest3 [71] on the CIPRES Science Gateway v.3.3 [72]. All sequences generated in this study are deposited in NCBI GenBank under the accession numbers MK889511–MK889751 (cox1) and MK889752–MK889992 (28S) and on BOLDsystems under the IDs LBCWS001-19 to LBCWS245-19 (cox1).

Distance-based species delimitation

Intra- and interspecific Kimura 2-parameter (K2P) [44] and uncorrected (p) distances were calculated in MEGA X [73]. The species delimitation threshold was investigated by using the threshold optimization method of the SPIDER (Species Identity and Evolution in R) v.1.5.0 package [74] implemented in R [75] as described in the tutorial (available at: http://spider.r-forge.r-project.org). Additionally, we used the Automated Barcode Gap Discovery (ABGD) procedure [76] to assign the sequences to hypothetical species based on the gap between intra- and interspecific sequence diversity, the so-called “barcoding gap”. ABGD was performed on the ABGD web interface [77] by using the MEGA distance files with default parameters, 20 steps and a modified relative gap width of 1. Additionally, the sequences were analyzed by using the Bold Systems v.4 [78, 79] tools available on the Barcode of Life webpage [80].

Phylogenetic species delimitation

Phylogenetic relationships between the sampled Hydrachnidia species were examined with maximum likelihood (ML) and Bayesian inference (BI). RAxML-NG [81], which is a new improved version of RAxML [82], was used to infer the best fitting ML trees of the single markers (28S and cox1) and the concatenated alignment, respectively. Similar sequences were treated as duplicates and removed automatically by RAxML-NG at the beginning of the tree calculation. ML branch support values were generated by the bootstrap method [83] with 1000 replicates and bipartition support for the best ML tree. Bootstrapping trees were computed directly in RAxML-NG. The BI trees were generated by using the parallel MPI version of MrBayes v.3.2.6 [84, 85]. Bayesian inferences were run for 15 × 106 MCMC generations, sampled every 5000th generation after the exclusion of 25% ‘burn-in’ by using 4 independent chains. Branches showing bootstrap values below 70% and Bayesian posterior probabilities below 0.95 were interpreted as resolved but not statistically supported [86]. The concatenated alignment was treated as partitioned dataset with unlinked base frequencies, nucleotide substitution rates, gamma shapes and proportions of invariant sites. The rates and frequencies were set according to the ModelTest-NG results. For each marker (cox1 and 28S), the appropriate nucleotide substitution model was used when running RAxML-NG and MrBayes for the single and partitioned analysis, respectively. To resolve the basal nodes and ensure a reliable rooting, several outgroup taxa were added to the tree inferences. The most distant taxon included was the terrestrial mite Labidostomma luteum Kramer, 1879 (Labidostommatoidea) (GenBank 28S/cox1: KM100974/GQ864390). Additionally, Dactylothrombium pulcherrimum (Haller, 1882) (Trombidioidea) (GenBank: KM100939/KM100985), Valgoperuvia paradoxa (Robaux, 1970) (Trombidioidea) (GenBank: KM100943/KM100988) and Stygothrombium sp. (Stygotrombidioidea) (GenBank: KM100938/ KM100995) sequences were used as closely related terrestrial Acariformes. The Halacaridae species Halacarus omului (Pepato & Da Silveira, 2013) (GenBank: MG751425/MG696236) and Rhombognathus areolatus (Abé & Fernandes, 2011) (GenBank: MG751437/MG696244) were chosen as aquatic relatives to the monophyletic Hydrachnidia [87]. The final trees were analyzed and edited in FigTree v.1.4.4 [88], Geneious Prime v.2019.1.1, Dendroscope v3.5.10 [89] and Affinity Designer v.1.6.1 (Serif Europe Ltd., Nottingham, UK).

Results

We successfully amplified and sequenced both target loci (cox1 and 28S, respectively) of 241 individual water mite specimens representing 22 genera and 65 morphologically identified crenobiontic and crenophilous species with 1 to 19 individuals per species (Additional file 1: Table S1). Three individuals belonging to the genus Atractides sp. (H450, H528 and H531), all representatives of the loricatus species group, were not identifiable to species level. As also observed in other populations of this group collected in various parts of Europe, important diagnostic features, i.e. large vs small dorsal muscle attachment sclerites, size of acetabula in the genital field and sclerotized or smooth excretory pore, as well as character state combinations are in disagreement with the identification key in Gerecke et al. [58].

cox1 final alignment length was 650 bp, 398 sites were polymorphic (389 parsimony informative) and no alignment gaps were present. The 999 bp 28S alignment showed 466 polymorphic sites (358 parsimony informative) and 137 gap positions. The translation of the cox1 sequences into amino acids did not contain any stop codon positions and blasting the sequences confirmed the absence of contaminations. In a few cases, when using the universal primer pair (LCO1490/HCO2198), we amplified Chironomidae DNA instead of water mite DNA and discovered Wolbachia sp. infestation. These specimens were excluded from further analysis.

Distance-based species delimitation and discovery

The mean overall pairwise distances were larger between the cox1 sequences (K2P ± SD: 0.29 ± 0.10; p-distance ± SD: 0.24 ± 0.07) compared to 28S (K2P ± SD: 0.15 ± 0.10; p-distance ± SD: 0.14 ± 0.05). Out of the 65 morphologically identified taxa, 11 were singletons, 7 taxa (Bandakia concreta Thor, 1913, Hygrobates norvegicus (Thor, 1897), Ljania bipapillata Thor, 1898, Partnunia steinmanni Walter, 1906, Wandesia racovitzai Gledhill, 1970, Wandesia thori Schechtel, 1912 and Zschokkea oblonga Koenike, 1892) showed exceptionally high intraspecific cox1 K2P distances (> 0.05) and the majority (47 species) had within species K2P distances between 0 and 0.03 (Fig. 1). Hygrobates norvegicus exhibited the largest mean genetic cox1 distances within morphologically identified species (K2P ± SD: 0.12 ± 0.08; p-distance ± SD: 0.10 ± 0.07), whereas several species showed mean intraspecific K2P distanced below 0.01 (< 1%) (Fig. 1).

Fig. 1
figure 1

Intraspecific cox1 distances. K2P and p-distances within morphologically identified species represented by more than one individual

Full size image

The SPIDER threshold optimization procedure analysis, which was conducted after the removal of singleton species and sequences of the seven taxa showing exceptionally high intraspecific variation, revealed an optimal K2P distance threshold at 0.056 (5.6%) and a p-distance threshold at 0.053 (5.3%) for species identification purposes with no false positive and low false negative identifications (9 out of 225 sequences). Assuming that cox1 species identification threshold, the individuals belonging to the above-mentioned taxa, which show high intraspecific variation, are likely to represent more than one species. This threshold was confirmed by the ABGD method that indicated a barcoding gap between K2P and p-distances of 0.06 and 0.09 (Additional file 2: Figure S1). ABGD initial partition revealed 69 and the recursive partition 70 groups, which can be seen as equivalent to species. Each of the seven taxa showing high intraspecific distances were split into two separate groups. Additionally, Lebertia schechteli Thor, 1913 showed clade separation in the recursive but not the initial partition causing the disparate number of groups between the partitions (Additional files 3 and 4: Figures S2 and S3). This generally confirms the SPIDER results and indicates the presence of more species than the a priori identified morphospecies. Contrary to that, Lebertia crenophila Viets, 1920, Lebertia holsatica Viets, 1920 and Lebertia lativentris Viets, 1922 as well as Atractides macrolaminatus/A. loricatus and Atractides brendle Gerecke, 2003 were grouped together as the same species in both ABGD partitions. The same analyses were conducted for the 28S dataset. However, neither the ABGD nor the SPIDER method revealed a threshold suitable for species identification purposes. ABGD indicated the absence of a clear barcoding gap for the 28S locus (Additional file 5: Figure S4) and the threshold optimization implemented in the SPIDER R package indicated high levels of false negative and false positive identifications at different thresholds.

Phylogenetic species delimitation and discovery

The results obtained by the distance-based species delimitations were generally confirmed by the phylogenetic approach. However, in contrast to the combined 28S and cox1 analysis, the single marker datasets alone did not allow to correctly reveal phylogenetic relationships at different taxonomic ranks and the BI trees (Additional files 6 and 7: Trees S1 and S2) showed several unresolved nodes and polytomies compared to the better resolved ML trees (Additional files 8 and 9: Figures S5 and S6). cox1 analyses incorrectly clustered higher taxonomic levels (e.g. genus, family and superfamily). For example, the genus Protzia Piersing, 1896 was clustered together with Sperchon Kramer, 1877 instead of the more closely related Partnunia Piersing, 1896 (Additional file 8: Figure S5). 28S correctly reconstructed higher taxonomic levels but did not allow resolving species relationships in several cases, e.g. Feltria cornuta Walter, 1927 and Feltria longispina Motas & C. Angelier, 1927 or Lebertia holsatica Viets, 1920 and Lebertia lativentris Viets, 1922 (Additional file 9: Figure S6). Compared to the single marker analysis, the overall taxonomic relatedness was depicted correctly by the combined dataset (Fig. 2). Furthermore, both phylogenetic methods (ML and BI) showed largely congruent and stable tree topologies when applied to the combined cox1 and 28S alignment (Additional file 10: Alignment S1). Due to these findings, we will focus on the ML tree with combined branch support data (Fig. 2).

Fig. 2
figure 2

Maximum likelihood tree of the combined cox1 and 28S datasets. Support values are shown as bootstrap (BS) and posterior probability (PP) values (PP/BS). Nodes fully supported by PP = 1 and BS = 100 are shown as +. In the case of clear monophyletic clades, tip nodes with more than one individual were collapsed with the number of individuals indicated as (n = X). In the case of clade separation within morphologically identified species we named the corresponding specimens by using the species name, cf. and A & B. Clades containing more molecular species than the a priori morphospecies are marked in red. Specimen IDs and sampling data of the individuals belonging to species are provided in Additional file 1: Table S1

Full size image

Results showed that the superfamilies (Hydrovolzioidea, Hydryphantoidea, Lebertioidea, Arrenuroidea and Hygrobatoidea) are monophyletic and clearly separated from each other. A relatively distinct clade affiliation of individuals belonging to the morphologically identified species is evident (Fig. 2). The previously mentioned specimens showing high intraspecific distances are also clearly separated into different clades. All corresponding branches showed high support values, indicating high probabilities of these splits. In the case of Partnunia steinmanni, individuals are grouped in two distinct clades, Partnunia cf. steinmanni A that shares a common ancestor with brook inhabiting (rhithrobiont) Partnunia angusta (Koenike, 1893) and Partnunia cf. steinmanni B. Both morphologically unidentifiable Atractides sp. (A & B) individuals represent two genetic species and also the individual belonging to the Atractides gr. macrolaminatus/loricatus is clearly separated from all other Atractides sp. specimens (Fig. 2).

Discussion

Morphological species identification has a long tradition and is commonly used to identify species for scientific and applied (e.g. ecosystem assessment) purposes [90]. Recently, it has been shown that molecular data reliably complement morphological species identification and has many advantages, especially when used to identify multiple species at once [36, 91]. Furthermore, molecular species delimitation resolves taxonomic uncertainties. When combined with other species-defining characteristics such as morphology it produces a more complete conclusion (e.g. [34, 35, 92, 93]).

Our results generally confirm the morphological species delimitation but show that Hydrachnidia species richness is underestimated and molecular methods are essential to discover currently overlooked biodiversity. The distance-based species delimitation methods revealed an intraspecific cox1 threshold between 5.6% (SPIDER) and 6% (ABGD) K2P distance, which is relatively high compared to other taxa [31, 48] but seems to be typical in water mites [28, 29, 94]. However, species delimitation solely based on fixed genetic distance thresholds can be misleading and thresholds should be estimated individually for each dataset [38, 95, 96], especially in taxa with clade-specific intraspecific cox1 distances as demonstrated for water mites in this and previous studies (e.g. [94, 97]). As the 28S marker region did not show a distinct barcoding gap and a clear species identification threshold was not evident, we do not recommend using it as single marker for threshold-based species identification of water mites.

The distance- and morphology-based results were confirmed by applying a phylogenetic approach. A clear monophyletic clade affiliation of individuals belonging to the same species was evident. Both, the distance-based and the phylogeny-based species delimitation revealed that seven morphologically identified species (Bandakia concreta, Hygrobates norvegicus, Ljania bipapillata, Partnunia steinmanni, Wandesia racovitzai, Wandesia thori and Zschokkea oblonga) show high genetic differences and therefore are likely to represent more than one species. The cox1 ABGD results differed in two cases from the other methods. Lebertia crenophila, L. holsatica and L. lativentris were grouped together as one species as it was also the case in Atractides macrolaminatus/loricatus and A. brendle. However, these species phylogenetically belong to different clades with high support values and we therefore assume that ABGD erroneously groups the respective sequences and underestimates the number of species, a tendency that has already been shown by other studies (e.g. [98]). Furthermore, ABGD initial and the recursive partition differently grouped L. schechteli individuals either as one single or two separate species. The initial partition is considered as more stable and usually better represents the groups that are defined by taxonomists [76, 99]. In addition, the phylogenetic inferences revealed a distinct L. schechteli clade. Therefore, we assume that our L. schechteli specimens belong to a single species. However, the recursive partition results show the possibility of differentiation and indicate that further research is advisable. In the analysis of H. norvegicus and P. steinmanni we were able to include several individuals per genotype, whereas in the other taxa differentiation is based on single individuals. Therefore, further individuals of different populations will be processed prior to the final description of the potentially new species.

Results suggest the monophyly of Hydrachnidia and the previously defined superfamilies, supporting the findings of Dabert et al. [87]. In contrast to the latter authors, and possibly due to the inclusion of more taxa in this study, our data support a Hydryphantoidea monophylum. However, the corresponding node separating Hydryphantoidea and Lebertioidea is poorly supported in our results and therefore should be further questioned. Similar to the study of Dabert et al. [87], we equally observed that cox1 or 28S alone are incapable of fully resolving phylogenetic relationships. This phenomenon can probably be caused by mito-nuclear discordance, which has already been recorded for mites and other taxa [100, 101] and was reviewed intensively [102]. Besides this assumption postulating diverging nuclear and mitochondrial phylogenies, the different taxonomic resolution of the two marker regions is likely caused by the loss of phylogenetic information. This can be explained by the fast-evolving character of mitochondrial compared to nuclear DNA and therefore faster loss of ancestral polymorphisms in cox1 [102,103,104,105]. Arabi et al. [105] showed exceptionally high mitogenomic rearrangements especially in Chelicerata, which fosters this assumption. Therefore, we conclude the necessity of complementing the standard barcoding marker cox1 with at least one additional genetic marker, e.g. 28S or 18S rDNA, to investigate species relationships and fully resolve water mite taxonomy.

These findings show that the choice of species identification markers must be done with caution and should be adjusted to the research question. For Hydrachnidia, cox1 serves as a useful marker if solely species identification is of interest. If the goal is to assign sequences to higher taxonomic levels, another locus should be taken into account. This is also crucial if the aim is to assess the amount of undescribed species in an environment. Currently, most metabarcoding approaches are based on cox1 alone [106,107,108,109] and few studies investigated the performance of alternative loci (e.g. [110,111,112]). Apart from the taxon assignment limitations when using cox1 alone, our findings show that the primer bias problem [36, 111] needs to be considered when water mites are targeted in metabarcoding studies as universal cox1 primers show unsatisfactory amplification performance. This could potentially be circumvented by using a combination of more specific cox1 primer sets as has been done in this study, a better matching universal one or the establishment of a new metabarcoding marker for this taxon. We were able to show that molecular methods have a great potential to reveal new water mite species and more studies are needed to complete barcoding databases and refine biodiversity estimates. Fundamental knowledge of species diversity is an essential precondition for implementing water mites in recent monitoring approaches and use them as powerful bioindicators [113, 114] in freshwater assessments as for example required by the European Water Framework Directive (WFD) [115], and may also offer an opportunity for a more nuanced understanding of environmental change impacts on springs systems.

Looking at spring inhabiting Hydrachnidia species, our work contributes to the accumulation of species barcoding data. Our data covers 47.5% of the currently described 137 spring water mite species in Europe [9] and covers the most abundant taxa, especially in the central Alps. Compared to other studies, which investigate Hydrachnidia diversity by applying morphological and molecular techniques in other aquatic habitats than springs (e.g. [28, 87, 94, 116, 117]), we were able to include a large subset of different species belonging to different taxonomic groups including the Proto-, Eu- and Neohydrachnidia [87]. A relatively high proportion of morphologically identified species (10.8%) showed to be more diversified than had been assumed, which indicates an overall underestimation of Hydrachnidia species richness in springs and other aquatic habitats. This indicates that species diversity related research questions such as host specificity of different water mite species need to be reconsidered. Hygrobates norvegicus, Partnunia steinmanni and Ljania bipapillata were shown to have a rather wide host species spectrum [18, 19]. Considering our results, which indicate that these three morphologically described taxa consist of several novel species, the number of hosts per water mite species could be lower and reveal a tendency towards high host specificity.

The dispersal abilities of Hydrachnidia highly depend on the parasitic larval stage that attaches to an insect host that can carry it to a different habitat and therefore governs water mite presence or absence in an environment [2, 13, 118]. Especially in rather isolated freshwater habitats like springs, water mite species dispersal is directly linked with their specific hosts leading to the conclusion that phylogeographic patterns are shared between hosts and parasites. Combined with our assumption of increased host specificity due to the unexpected high degree of Hydrachnidia species diversification, future studies on gene flow patterns between water mite populations can verify the hypothesis that springs are isolated island-like habitats for the mites as well as their insect hosts.

In Europe, 970 water mite species have been recorded to date [9] and, applying our findings, at least 105 additional species potentially exist. Due to the fact that water mite species diversity increases towards southern parts of Europe [9] we expect an even higher degree of undiscovered species as our dataset mainly consists of specimens collected in central Europe. On this basis, future water mite barcoding projects will be able to gradually fill the gaps of taxonomic knowledge. This is an important prerequisite to incorporating modern species identification and monitoring techniques (e.g. metabarcoding) in future water mite-related freshwater and spring assessment studies.

Conclusions

Our aim was to verify and complement the mainly morphology-based species delimitation of an often neglected, highly diverse taxon in freshwater ecosystems. Our results show that water mite diversity in springs seems to be higher than expected. Molecular methods are largely congruent with morphology and serve as a species delimitation and identification tool. They are particularly powerful if species discovery is the main goal. cox1 as a standard barcoding marker is useful for identifying Hydrachnidia species but is not suitable for assigning them to higher taxonomic levels (e.g. genera, families or superfamilies). This limitation can be overcome by using distance- and phylogeny-based multi marker approaches. Our data contributes to genetic species identification databases by adding crenobiontic water mite sequences, which is a precondition for implementing modern methods of freshwater ecosystem assessment such as metabarcoding and eDNA species community monitoring in spring ecology.