This study includes 40 representatives of all five subdivisions of Circumdehiscentiae (after Grandjean 35 ): Opsiopheredermata, Eupheredermata, dorsodeficient and pycnonotic Apheredermata and Poronota (Table 1). Families categorized as the sixth subdivision "Circumdehiscentiae with wrinkled nymphs" (after Wauthy 37 ) are written in bold lettering in the table. Based on Weigmann 48 we chose Hermannia gibba (Hermanniidae), a member of the Desmonomata, sister group of the Circumdehiscentiae, as outgroup. Sequences not generated in the framework of this study were obtained from GenBank (see Table 1).

Specimens were extracted from mosses, lichens or soil samples with Berlese-Tullgren funnels and preserved in absolute ethanol. Total genomic DNA was extracted from single individuals applying the CTAB (hexadecyltriethylammonium bromide) method described in Schäffer et al. 30 . After DNA extraction, the sclerotized body remnants were mounted on permanent slides and used for species identification using the criteria defined in Weigmann 33 .

Fragments of 28S rDNA, ef-1α and hsp82 genes were amplified by polymerase chain reaction (PCR) using the following primers: D3A and D3B 49 for the D3 fragment of the 28S rDNA, 40.71F and 52.RC 50 and EF-SyFwd and EF-SyRev 10 for ef-1α, hsp1.2 and hsp8.x 42 for hsp82. Polymerase Chain Reaction (PCR), purification of PCR products and DNA sequencing followed the protocol described in Schäffer et al. 51 . DNA fragments were purified with Sephadex™ G-50 (Amersham Biosciences) following the manufacturer's instruction and sequencing reaction products were analyzed on an ABI PRISM 3130xl automated sequencer (Applied Biosystems). Sequences are available from GenBank under the accession numbers listed in Table 1.

We sequenced 311-329 bp of the D3 region of the 28S rDNA, 475 bp of the ef-1α, and 467-503 bp of the hsp82 gene in 34 specimens (from Liacarus cf. subterraneus and Arthrodamaeus sp. the fragment of hsp82 could not be amplified). Sequences were verified by comparisons with known oribatid sequences from GenBank and aligned by eye in MEGA 3.1 52 . We removed poorly aligned regions from the alignments of 28S rDNA and hsp82 using the program trimAl 53 which is a tool for automated alignment trimming. Gap threshold was set to 0.8 and similarity threshold to 0.001. All sequences were combined into a single data set with a resulting length of 1,298 bp for further analyses.

Phylogenetic inference was based on Bayesian inference (BI), conducted in MrBayes 3.1.2 54 . Data were partitioned by gene and the ef-1α gene was further partitioned by codon position. Number of substitution types was set to six (GTR model) for each data partition and among-site rate variation was drawn from a gamma distribution. Posterior probabilities were obtained from a Metropolis-coupled Markov chain Monte Carlo simulation (2 independent runs; 4 chains with two million generations each; trees sampled every 100 generations), with parameters estimated from the data set. Mixing and convergence to stationary distributions were evaluated in Tracer v.1.4 by inspecting graphically the trace of the parameter against the generation numbers 55 . The first 4000 (20%) trees were discarded as burn-in prior to constructing a 50% majority-rule consensus from the remaining 16,001 trees.

Ancestral state reconstruction (ASR) is an increasingly popular method to map morphological or ecological traits onto a molecular phylogeny. However, there are still controversial opinions about the accuracy of commonly used methods (maximum parsimony (MP), maximum likelihood (ML), Bayesian) and each suffers from certain advantages and limitations 56 57 58 59 . According to Ekman et al. 60 , who showed the importance of conducting ASR with more than one method, we performed our analyses using all three above mentioned methods. Furthermore, specifying the right models and priors in a Bayesian analysis is of uttermost importance 60 61 . We employed the reversible-jump (RJ) MCMC, where models are visited in proportion to their posterior probability 62 .

We traced the evolution of six morphological characters (scalps, centrodorsal setae, sclerits and wrinkled cuticle which are all developed in nymphs plus octotaxic system and pteromorphs both in adults; Table 1) over the molecular phylogeny using MP, ML and Bayesian approaches. Information on the studied characters was retrieved from the literature 33 35 51 63 64 65 66 67 68 69 70 71 and from the body remnants of our specimens. Ancestral character state reconstruction for the notogastral octotaxic system, employed two different data sets, one (titled porose organs-1) with porose organs (regardless of which type) as either absent (0) or present (1) and one (titled porose organs-2) coded with the different types found in oribatid mites: porose organs absent (0), porose areas (1), saccules type1 (2), saccules type2 (3). With reference to Alberti et al 36 saccules type2 differs from type1 by lacking an elaborate microvilli system, having a considerable number of mitochondria and rather characteristic lysosome-like inclusions. Parsimony and likelihood based ASR were conducted in Mesquite v.2.71 72 . To account for topological uncertainty we used the "trace character over trees" option, which summarizes the ASR over a series of trees. All reconstructions were integrated over the last 6001 post burn-in trees of the Bayesian analysis and the ancestral states were summarized on the BI consensus tree. As model of evolution for the ML reconstructions we employed the Markov k-state 1 (Mk1) parameter model, with equal probability for any particular character change. To account for phylogenetic mapping uncertainty, we further evaluated probabilities of ancestral states calculated from the same 6001 BI trees using the MCMC method in BayesMultiState 73 , implemented in the BayesTraits 1.0 package. Ancestral states were only reconstructed for 23 nodes (see Fig. 2), which were selected based on their posterior probability support values of the BI analysis (only those nodes with PP ≥ 0.95 were used). A reversible-jump (RJ) hyperprior with a gamma prior (exponential prior seeded from a uniform distribution on the interval 0 to 30) was used to reduce uncertainty and arbitrariness of choosing priors in the MCMC analysis. According to preliminary analyses, we set the ratedev value to 8, achieving an acceptance rate of proposed changes between 20 and 40%. The option "AddNode" was used to find the proportion of the likelihood associated with each of the possible states at each node. Three independent MCMC runs were performed with 5,050,000 iterations. Chains were sampled every 100 iteration after a burn-in of 50,000 iterations. Because of similar results of the three runs, we only report one of them here. The output files were analyzed using Tracer v1.4.

Bayes factors (BF) are statistical tools to compare alternative hypotheses against a null hypothesis 74 75 76 . We used BF to test ten alternative phylogenetic hypotheses against the unconstrained BI tree: Monophyly of i) Apheredermata (hypothesis 1), ii) the dorsodeficient nymphs (hypothesis 2), iii) Poronota (hypothesis 3), iv) nymphs with macrosclerits (hypothesis 4), v) "Circumdehiscentiae with wrinkled nymphs" (hypothesis 5), and vi) pteromorphs (hypothesis 6). Furthermore, we tested the monophyly of the families Scutoverticidae (hypothesis 7), Cymbaeremaeidae (hypothesis 8) and Ameronothridae (hypothesis 9), plus monophyly of each of the three families (hypothesis 10). Alternative phylogenetic trees were inferred in MrBayes by applying topological constraints. Run settings were the same as for the unconstrained BI tree. We combined the two log-files of the Bayesian analysis with the program LogCombiner v1.5.4 available in BEAST package 77 and calculated the BF in Tracer. Standard errors were assessed using 1,000 bootstrap replicates. Interpretation of BF followed Kass and Raftery 76 .

Pairwise sequence divergence (uncorrected p-distances) between the investigated species ranged from 0-11% in the D3 fragment of 28S rDNA, from 0-22% in the ef-1α gene and from 2-26% in the hsp82 gene. In the combined data set, pairwise differences ranged from 1-20%.

The Bayesian 50% majority rule consensus tree is shown in Fig. 2. Most basal nodes were statistically well supported whereas some more recent splits were poorly resolved. Compared to the traditional classification, our BI tree revealed some discrepancies. None of the five or six, respectively, major subdivisions seems to be monophyletic. In particular, the "Eupheredermata", "dorsodeficient Apheredermata" and "Poronota" appeared as para- or polyphyletic. Wauthy's 37 subdivision "Circumdehiscentiae with wrinkled nymphs" clusters together in a major clade, but also includes three species of Poronota (Trichoribates trimaculatus, Euzetes globulus and Galumna cf. obvia), though with low statistical support. The two "circumdehiscent" species Cymbaeremaeus cymba and Ametroproctus lamellatus form a well supported more basal clade, rendering the "Circumdehiscentiae with wrinkled nymphs" paraphyletic. Likewise, the family Scutoverticidae was not resolved as monophylum, but rather constitutes two distinct clusters: one includes exclusively species of the genus Scutovertex, whereas the other one comprises members of the three other genera - Provertex kuehnelti, Lamellovertex caelatus, Exochocepheus hungaricus - plus "Scutovertex pictus" and furthermore Ameronothrus maculatus and Scapheremaeus cf. palustris, thus rendering the family paraphyletic. BF of the tested alternative hypothesis rejected a monophyletic family Scutoverticidae (hypothesis 7, Table 2). The aforementioned results also imply paraphyly of the families Cymbaeremaeidae and Ameronothridae supported by the BF which decisively discriminated against a monophyly of the two families (hypotheses 8 and 9, Table 2). Also the monophyly of each of the three families was rejected by the BF (hypothesis 10, Table 2). Moreover, Achipteria coleoptrata clusters with Parachipteria punctata and Achipteria quadridentata as sister taxon, also rendering the genus Achipteria paraphyletic; alternatively, Parachipteria might represent a synonym of Achipteria.

The results from the ASR of scalps, centrodorsal setae, porose organs, sclerits in nymphs, nymphal cuticle structure and pteromorphs are shown in Figs. 3, 4 and Table 3. The reconstruction yielded no conflicts between parsimony, likelihood and Bayesian analyses except for some nodes, depending on the studied character (Table 3), but the likelihood approach reconstructed them with greater uncertainty (equivocal) than the Bayesian approach. For example, nodes 4 and 5 in scalp evolution (Fig. 3A) or node 17 in data set porose organs-1 (Fig. 3C) were ambiguously reconstructed in the likelihood analysis as compared to the Bayesian analysis (Table 3). On the other hand, the likelihood analysis reconstructed some nodes with greater certainty for a particular character state than the Bayesian approach, as for example nodes 1 and 3 in scalp evolution (Fig. 3A, Table 3) or node 20 in the reconstruction of gastronotic cuticle structure evolution (Fig. 4B, Table 3).

The reconstruction of character evolution of the nymphal scalps (Fig. 3A) and centrodorsal setae (Fig. 3B) revealed a more or less single origin of the different character states, with the exceptions of the two apheredermous species Ceratoppia quadridentata and Liacarus cf. subterraneus, which clustered together with eupheredermous species, and the eupheredermous Arthrodamaeus sp., which grouped with apheredermous species. The phylogenetic placement of the latter species implies that dorsodeficient nymphs evolved twice. BF of alternative hypotheses reject both the monophyly of Apheredermata (hypothesis 1, Table 2) and a monophyletic origin of dorsodeficient nymphs (hypothesis 2, Table 2). The two-state data set of the porose octotaxic organs revealed multiple evolution of this character (Fig. 3C). This agrees with the ASR of the second data set (Fig. 3D; coded with the different types of porose organs) in which porose areas, saccules type1 and 2 do not share one most recent common ancestor (MRCA). More precisely, saccules type2 evolved two times and porose areas also at least twice. Particular types of porose organs are typically restricted to one phylogenetic lineage, but with some exceptions. In one family (Achipteriidae), two different types of porose organs are present. Whereas species of the genus Achipteria possess porose areas, the genus Parachipteria has saccules type1, which has been proposed to be the derived form 78 . Results of the ASR of the octotaxic porose organs are supported by BF comparison, which decisively rejected a monophylum Poronota (hypothesis 3, Table 2). Ancestral state reconstructions of the development of sclerits in poronotic nymphs indicated multiple independent origin of this trait (Fig. 4A). With Phauloppia cf. lucorum we only had one representative of the character state "nymphs with microsclerits", such that we focused specifically on species showing "nymphs with macrosclerits". According to the reconstruction of Mesquite (Fig. 4A), this trait could have evolved at least two times which is supported by the BF of the tested alternative hypothesis (hypothesis 4, Table 2). The ASR of the gastronotic cuticle structure of juveniles suggested that wrinkles in nymphs evolved independently at least two to maximum four times (Fig. 4B). Consistent with this observation, BF comparison rejected the hypothesis enforcing a monophyletic subgroup "Circumdehiscentiae with wrinkled nymphs" (hypothesis 5, Table 2). The results of the ASR of pteromorphs (Fig. 4C, Table 3) indicated that this structure evolved twice within the Apheredermata and the BF of the alternative hypothesis strongly discriminated against a monophyletic clade of species with pteromorphs.

Bayesian inference of the phylogeny of the Circumdehiscentiae based on a combined data set of fragments of three nuclear genes revealed a tree topology similar to previous molecular studies 27 28 , especially to the most recent phylogeny published by Maraun et al. 31 which is based on the 18S rDNA gene. In the latter study the statistical support for most important nodes is higher than in our case, but we note that taxon sampling differs slightly between Maraun et al.'s and our study. Splits within the Apheredermata are statistically not well supported in our phylogeny, which can be interpreted as a strong indication for a period of rapid cladogenesis at a certain time in the past. Comparing the 18S tree with our results in detail revealed a different placement of Scapheremaeus cf. palustris in both phylogenies. Whereas it is grouped with Ameronothrus maculatus and four specimens of the family Scutoverticidae in our phylogeny, it is placed with Eremaeozetes sp. as sister taxon to Tectocepheus velatus in the 18S tree, but we emphasize that Maraun et al.'s 31 data set did not include any representatives of the Scutoverticidae. Furthermore, our results show that there is no close relationship of A. maculatus and the genus Podacarus auberti cf. occidentalis as already supposed by Grandjean 79 . However, Weigmann and Schulte 80 unified the seven genera of originally three families (Ameronothridae, Podacaridae, Aquanothridae) into one single family Ameronothridae, a system accepted by Norton and Behan-Pelletier 34 . The placement of S. cf. palustris and A. maculatus within the Scutoverticidae renders not only this family but also the Cymbaeremaeidae and Ameronothridae paraphyletic. These results could certainly be caused by a strong bias of one single gene. However, the paraphyly of these three families is also supported in our single gene analyses (data not shown). The non-monophyly of Cymbaeremaeidae is also supported by the 18S rDNA gene 31 with S. palustris and Cymbaeremaeus cymba not clustering together. Whether the family Scutoverticidae should be split up or extended with additional taxa (S. cf. palustris, A. maculatus and maybe others) remains unclear until more species of the remaining genera and closely related families (e.g. Ameronothridae, Passalozetidae, Licneremaeidae) are included in a comprehensive phylogenetic study. However, our data, which are certainly not comprehensive enough to allow for a full revision of these problematic families but are sufficient to hint at some taxonomic inconsistencies, indicate that the family Scutoverticidae might be split up in "Scutoverticidae s.s.", so far including only species of Scutovertex and in "Scutoverticidae s.l." with the remaining taxa. The paraphyletic resolution of Ameronothridae, Cymbaeremaeidae and the genus Achipteria appears to be the result of a lack of adequate synapomorphic morphological characters, calling for further detailed investigations not only on a strictly morphological basis, but also including alternative approaches in a framework of integrative taxonomy 11 .

Ancestral state reconstructions and phylogenetic hypothesis testing indicate that none of Grandjean's 35 main traits is a good tool to classify the investigated taxa. However, regarding scalps and centrodorsal setae, the only disagreement to traditional taxonomic classifications is caused by the placement of Arthrodamaeus sp. which clusters within the Apheredermata (Figs. 3A-B), supported by BF comparison of alternative phylogenetic hypotheses (Table 2). Reconstruction of the character history of the octotaxic system in adults which is eponymous for the Poronota clearly suggested multiple evolution of this diagnostic character (Figs. 3C-D). The general model of the origin of saccules proposed by Grandjean 78 81 is that porose areas invaginated and formed a saccule having a lumen encircled by porose walls. Thus, porose areas in the octotaxic system would represent the plesiomorphic character state. Concerning the question whether the small porose areas or minute saccules in various Licneremaeoidea (e.g. Licneremaeus, Scutovertex) do either represent a numerical and size regression or the plesiomorphic state of the typical octotaxic system, Norton and Alberti 82 argued that the early evolution of porose organs started small because they are small, even minute in some Licneremaeoidea (a probably paraphyletic assemblage according to 82 ) - e.g. Scutoverticidae, Dendroeremaeidae 83 (in our study called saccules type2) -, in their opinion the most early-derived group of Poronota. According to the numerical regression, it should be mentioned that the octotaxic system is often reduced to one, two or three pairs of organs. Among the Scutoverticidae, for example, no species is known to have the full complement of four pairs, leading to the hypothesis that four pairs do not represent the ancestral state for this family. Norton and Alberti 82 noted that the earliest homologue of the octotaxic system might be a single pair of dermal glands that evolved by gene-duplication or by modification of developmental controls to four pairs. Our results of the ancestral state reconstructions now suggested, that, whatever character coding [either two-stated (Fig. 3C) or organ-type specific (Fig. 3D)] is used, the octotaxic system evolved independently many times in parallel within the Poronota. Furthermore, there are no indications that either porose areas or saccules type2 are the plesiomorphic state of the porose organs as proposed by Grandjean 78 81 and Norton and Alberti 82 , respectively. Concerning the weakly supported nodes within the Poronota (Fig. 2), one could hypothesize that porose areas still can be traced back to one MRCA, a hypothesis clearly rejected by our BF comparison of alternative phylogenetic hypotheses (hypothesis 3, Table 2). Saccules type2, present in Scutovertex and Exochocepheus hungaricus, was inferred to have evolved twice from an ancestor lacking porose organs (Fig. 3D). These results reject the general hypothesis that Poronota represent a natural, monophyletic subdivision, as already supposed by Grandjean 38 ; he also stated that the presence or absence of these organs alone is not sufficient for a grouping into Pycno- or Poronota. A recent study 34 avoids the terms Pycnonota and Poronota, though it differentiates between pycnonotic and poronotic taxa without implying that these represent monophyletic groupings. Woas 84 suggested that secretory porose organs probably represent functional adaptations, potentially leading to a multiple independent evolution of this morphological character complex. Additionally, it must be noted that in many poronotic families and genera species can have various types of porose organs (see table 1 in 82 ). For example, both porose areas and saccules (type1) are found among species of Achipteriidae (Table 1) or Trichoribates 85 . In this regard a potential parallel or convergent evolution of the octotaxic system should already have been a point of discussion in former time. However, considering the complexity of these paired organs located at the more or less same notogastral positions, it appears unlikely that these structures evolved independently multiple times, pointing to the need of further detailed investigations on these structures. Recently, Weigmann 86 reported on a different formation of octotaxic organs on the left and right side of the body of one single specimen and attributed this phenomenon to the differential action of regulatory genes.

In addition to the three main characters, Grandjean 35 used the morphology of nymphs and larvae for a classification within Circumdehiscentiae. He divided the Poronota into three types: 1) species with wrinkled nymphs; 2) species having nymphs with macrosclerits, and 3) species having nymphs with microsclerits. Furthermore, Grandjean 35 38 postulated that those taxa with wrinkled nymphs should form a monophyletic group regardless whether they are pycno- or poronotic (see Background) and that they might represent an intermediate group (he formulated it as "à cheval sur la limite" meaning "at the frontiers") between the pycnonotic Apheredermata and Poronota. This concerns taxa of the following 11 families: Podacaridae, Charassobatidae, Ameronothridae, Scutoverticidae, Tectocepheidae, Passalozetidae, Cymbaeremaeidae, Licneremaeidae, Achipteriidae, Tegoribatidae and Phenopelopidae (formerly Pelopsidae). Wauthy 37 followed this proposal and named the group "Circumdehiscentiae with wrinkled nymphs". However, our results revealed that the wrinkled nymphal cuticle structure evolved (and got lost) in parallel multiple times among Circumdehiscentiae. This clearly rejects the hypotheses of Grandjean 35 38 and Wauthy 37 that taxa with wrinkled nymphs are monophyletic. Ancestral state reconstructions also showed that the MRCA of Circumdehiscentiae had unwrinkled nymphs (Fig. 4B), thus rejecting the assumption of Norton and Behan-Pelletier 34 that wrinkles in nymphs - because of their occurrence in apheredermous and eupheredermous taxa - seem to be the plesiomorphic or even ancestral state in Circumdehiscentiae.

Pteromorphs only occur (with exception to the eupheredermous Microzetidae) in adult poronotic mites and Travé 87 postulated that the presence or absence of this trait might serve as differentiation criterion within the Apheredermata, a hypothesis rejected by our data (Fig. 4C; Table 2).

Our reconstruction of ancestral states within the Circumdehiscentiae shows that some previously used diagnostic characters are problematic for taxon classification, despite previous efforts to clarify plesio- and apomorphies 86 88 . These problems mainly arise from the difficulty to evaluate missing characters as reduced or never developed. However, specific traits could be still of appreciable value such as the presence of nymphal scalps or centrodorsal setae. Scalp retention is often correlated with the nature of dehiscence 35 , which goes up after a striking process in immature instars. If the metamorphosis fails, for example due to a genetic defect, the molting individual would stall and die. Furthermore, the absence of centrodorsal setae correlates with the scalps 35 because species retaining scalps on their notogaster are dorsodeficient (have lost setae da, dm, dp). Thus, dorsodeficient nymphs correspond to the Eupheredermata and integridorsal nymphs to the Apheredermata, respectively. As an exception the two dorsodeficient apheredermous families Ceratoppiidae and Liacaridae group with the Eupheredermata with dorsodeficient nymphs, thus suggesting a possible loss of scalps in nymphal stages and adults.

Altogether, none of Grandjean's main traits can be used as diagnostic character for classification of the Circumdehiscentiae, because of clear evidence for multiple parallel evolution (and in some cases also losses) of these traits. In the case of the octotaxic system this could be argued with a possible correlation of secretory notogastral porose organs and ecology. Norton and Alberti 82 still noted that poronotic mites with modified porose areas (saccules, multiplications) inhabit non-soil microhabitats (e.g. mosses and lichens on rocks or trees) and therefore internalization may reduce water-loss through porose surface. But they finally stated that there is no biological or ecological correlation that would help to understand frequent convergent evolution. We hypothesize that maybe there is no correlation of the secretory organs with the present ecology. Considering the old age of Oribatida - the crown radiation of Apheredermata took place at the Triassic/Jurassic boundary (c. 200 MYA) 19 89 or according to Schaefer et al. 90 in the Permian - it might be more adequate to search for congruent environmental conditions in the past, when the explosive radiation within the Apheredermata took place. A monophylum "Circumdehiscentiae with wrinkled nymphs" was clearly rejected by the results of the ancestral state reconstructions, implying that the wrinkled nymphal cuticle structure is of little value for classification. The potential function of these wrinkles was already well studied by Smrž 91 for some oribatid taxa. Thereby he recovered two different types of wrinkling (e.g. Hermannia gibba versus Scutovertex minutus) and a potential correlation between environmental conditions and type of wrinkling.

Typically, studies on character evolution focus on either the potential advantages of a trait in a given environment 92 93 94 or mechanisms that create novel phenotypes 95 96 . However, Wiens et al. 97 stated that these investigations may be necessary to elucidate why a trait has evolved in a particular instance but not why it has evolved multiple times. According to Wiens et al. 97 at least two additional factors are important in determining the number of origins, namely the biogeographic context of the selective environment and competitive interactions. The first point means that a trait which is adapted to a selective environment may evolve multiple times in geographically isolated regions with identical selective environments (e.g. 98 ). We hypothesize that this might be a possible reason for the multiple parallel evolution of the octotaxic system and wrinkled gastronotic cuticle structure in nymphs, but further detailed correlation studies including more taxa are necessary to allow for robust conclusions.