The location, host species, mite species, accession numbers of mites under test are listed in Appendix 1. Mite isolates have been sampled from wild bird nests or from farms as described in Roy et al. [10]. The distribution of samples is rather large and diverse within France, and includes a few samples from other countries (ex. The Netherlands, Poland, USA)(Appendix 1). Nests were analyzed using a method described by De Lillo [12] involving immersion of the nest followed by filtering, except that no sodium hypochlorite was added to the water solution to wash the stack of sieves and that the sieves had a somewhat different mesh width (top to bottom: 2,500 μm, 1,400 μm, 180 μm, 100 μm).

One isolate corresponds to mites of a single Dermanyssus species, isolated from an individual nest or from a group of nests closely located to each other in a bird colony (wild avifauna) or from a single building (farms). From each isolate, 1–5 individuals have been separately sequenced. Additionally, for five of these isolates (1 D. apodis: GO, 1 D. carpathicus: BER7, 2 D. gallinae s. str.: SK, IL, 1 D.gallinae special lineage L1: 9001; see Roy et al. 2009a), 18–24 individuals have been separately sequenced, in order to get an overview of the intrapopulation variation. Moreover, 21 individuals belonging to D. hirundinis collected from barn swallows distributed around France were included in the analyses and are referred to as DhirF. These six groups are used specifically for statistical analyses.

Data are composed of DNA sequences obtained following the method described in Roy et al. [10,11] and newly obtained DNA sequences following the procedures below. The previously tested DNA regions include partial 18S–28S rRNA, including complete ITS1, ITS2 and 5.8 S (ITS), partial 16S rRNA and partial coding gene for cytochrome oxidase I (COI). The newly developed markers are one intron flanked by two portions of the coding gene for Tropomyosin and a portion of the coding gene for elongation factor 1-alpha (EF-1α).

Sequences of Tropomyosin and EF-1α were obtained as in Roy et al. [11], with PCR involving annealing temperatures of 56 °C and 57 °C, respectively. Primer sequences and pairs are provided in Appendix z2.

D. gallinae has been shown to be haplodiploid with diploid females developing from fertilized eggs [13,14]. As these authors also observed similar haplodiploidy in a closely related family (Macronyssidae), we assume here that other Dermanyssus species reproduce the same way. As a result, only adult females were integrated in the study for standardization, notably to regularly detect allelic variation and to avoid potentially confounding effects of the male haploid genome.

In case of heterozygosity in the Tropomyosin targeted fragment, the two alleles were separated from each other using internal primers (cf. Appendix z2), or in some cases by cloning. As for EF-1α some sequences were cloned to separate double copies found after PCR. Vector cloning was necessary for separation of multi-copies sequences in some cases. PCR products were gel purified (Qiagen, Les Ulis France) and corresponding fragments were cloned into TOPO^® TA cloning vector according to manufacturers’ protocols. Transformant clones were checked by restriction enzyme profile and five positive clones were submitted to sequencing.

DNA and amino acid alignments were performed using MUSCLE 3.7. Without refinement, MUSCLE has been shown to achieve accuracy statistically indistinguishable from T-Coffee and MAFFT, but overall is the fastest of the tested methods for large numbers of sequences [15]. Seaview 4.0 [16] was used for DNA and amino-acid alignment handling. Six different matrices, composed of DNA sequence alignments, were generated during Step 1 and three different matrices, two of which are DNA sequence alignments and one is a digital matrix, during Step 2.

Some statistical analyses were performed between and within the six focused isolates, representing three species. They were performed using the isolate DNA alignments of COI haplotype sequences on one hand, and individual Tropomyosin alleles (phased alleles in heterozygous individuals and duplicated homozygous sequences, in such a way that sequences represent the diploid state of chromosomes) on the other hand.

Polymorphism in haplotype sequences (COI haplotypes and separated Tropomyosin alleles) within the eight “focused isolates” was examined (gaps excluded and as the fifth state in Tropomyosin) using DnaSP v5 [20]. We estimated the number of segregating sites (S), average number of nucleotide differences (k), and haplotype diversity (Hd). Pairwise genetic distances were computed using Fst (Hudson et al. 1992) and statistical significance assessed after 1000 permutations in all cases using Arlequin 3.1 [21] (haplotypic dataset for COI and genotypic dataset for Tropomyosin in order to avoid potential bias due to departure to the Hardy-Weinberg equilibrium).

The five gene portions have been obtained in six Dermanyssus species, two Ornithonyssus species and one Androlaelaps species. Within Dermanyssus, one isolate of D. hirsutus, two of D. apodis, two of D. longipes, three of D. hirundinis, three of D. carpathicus, nine of D. gallinae (of which three of D. gallinae special lineage L1, see Roy et al. [11]) have been integrated into the analyses. In some cases, two different profiles have been distinguished within some isolates of D. gallinae and D. carpathicus, due to some intra-isolate variations (mutations and/or indels) in COI and or Tropomyosin sequences. See Appendix 1 for detailed informations about individuals, isolate locations and accession numbers. All genes except Tropomyosin have been found for one additional dermanyssoid genus in EMBL database (Varroa) and all except Tropomyosin and 16S for another one (Tropilaelaps).

Both COI and Tropomyosin portions have been obtained from 211 individuals. Additionally, COI was also obtained in 41 individuals and Tropomyosin in 16 other individuals. This resulted in 56 different COI haplotypes isolated from five species of Dermanyssus (three in D. longipes, three in D. hirundinis, six in D. carpathicus, four in D. apodis, 35 in D. gallinae, one in D. hirsutus) and 62 different Tropomyosin alleles (five in D. longipes, two in D. hirundinis, seven in D. carpathicus, two in D. apodis, 36 in D. gallinae, one in D. hirsutus). See Appendix 1 for detailed informations about individuals and isolates.

The results integrating the newly developed nuclear marker Tropomyosin confirm reproductive isolation, and consequently specific status of D. carpathicus, D. hirundinis, D. apodis and D. gallinae. Dermanyssus carpathicus and D. hirsutus appear to be strongly differentiated from each other and from other Dermanyssus species in all genes except the small portions composed of Tropomyosin exon n and n + 1 (see Table 3). Dermanyssus apodis is strongly differentiated from other Dermanyssus species in both mitochondrial datasets and in the nuclear Tropomyosin datasets (both intron and exons), whereas it is very weakly differentiated based on ITS1 and 2 and not differentiated in 5.8S. On the other hand, special lineage L1, although less differentiated than D. apodis in other genes, appears more different in the ITS sequences.

Moreover, although the reproductive isolation of D. longipes isolates seems to be confirmed, this entity is revealed to be diphyletic. Two contradictory clades against D. longipes PAS + EN among the different partitioning schemes tested in Step 1 and a paraphyletic position of D. longipes’ Tpm sequences in two well supported clades in the MP analysis of Step 2 with gaps as the fifth state shows that this entity is composed of two different species. The characterization of both these lineages remains imprecise, since they are each represented here by one or two isolates from a single site. However, as noted in Roy et al. [11], sequences of ITS1 and 2 published by Brännström et al. [30] from mites sampled in Sweden on different bird species are exactly identical to ITS in D. longipes EN, whereas ITS sequences of D. longipes PAS diverge by 2%. Given the very low amount of interspecific divergence in ITS sequences within Dermanyssus, it is very likely that the Swedish isolates belong to the same specific entity as present EN isolates. Of the two D. longipes lineages, D. longipes PAS appears as closer to D. hirundinis (Figure 4b). Since this isolate has been sampled near the type locality and in the type host genus, it is to be considered the name bearing species. As a result, the lineage EN will be referred to as Dermanyssus sp. EN in the remaining text.

As for D. gallinae special lineage L1 [11], it appears as an isolated lineage with strong support. The monophyly of this lineage and the gallinae complex is also supported. Nevertheless, its position within the gallinae complex or as a sister to it remains unclear.

Multiple functional copies of EF-1α have been found in various Arthropoda [31–33] including in some cases both functional and non functional copies in single species [34]. Moreover, Hedin and Maddison [25] have shown the presence of intronless non functional copies among multiple copies in a portion overlapping the region studied here within the genus Habronattus (Aranea: Salticidae) (positions 529–1005 of the H. melpomene CDS). Intronless copies seem to be evolving under relaxed functional constraints. Goetze [34] published a similar report on Crustaceans with exons of putative functional / non functional sequences almost undistinguishable from each other.

In the present datasets, it is worth noting that there is no intron in any of the EF-1α portions under test, although several other Arthropoda show introns precisely within this portion, especially some Arachnida: between positions 771–772 (836–837 of the whole published sequence), a 167 bp intron is present in Mecaphesa sp. (Arachnida:Araneae:Araneomorphae: Thomisidae, FJ590835), a 124 bp intron in Habronattus (Arachnida: Araneae: Salticidae, AF477231) and a 147 bp intron in Haplochthonius simplex (Arachnida: Acari: Oribatida: Haplochthoniidae, GQ398254). And this position also possesses an intron in some insects (Coleoptera, Hymenoptera) [35] and some crustaceans (Calanoidea: Eucalanidae) [34]. Not to mention that at least two other positions in the present region possess introns in the same crustaceans and in some Diptera [34].

The intronless nature of presently tested sequences may suggest that they are not functional and evolve as pseudogenes. Of course, there are no stop codons in any of obtained sequences, but there are a few non silent mutations either in sequences of clade EF_A or in clade EF_B. And yet EF-1α has already revealed some difficulties in the application to deeper systematic relationships in Mesostigmata [36], and present inferences show largely unsupported internal relationships within Parasitiformes, including a few incongruences if compared with their phylogenetic relationships robustly established by Klompen et al. [37] based on rRNAs. Here, EF-1α copies position in the phylogenetic reconstructions of mites based on this gene (Figure 1) demonstrates that gene duplication has occurred anteriorly to the split between Macronyssidae and Dermanyssidae. The date of the duplication may be posterior to the split between dermanyssoid and other Mesostigmata according to the position of double copies in the large phylogenetic reconstruction represented in Figure 1. But further exploration would be needed in order to check this.

All the more, it is not unlikely that other difficulties are generated by multiple copies, including once more non functional ones at a lower taxonomic level within clade EF_B (Figure 1). Indeed, the structure remains unclear, as some sequences of outgroups and of ingroups group together in a distal position. A potentially more recent duplication event may have occurred, but on which present dataset is not sufficient to get any explanation, once more likely because of some failure in amplifying all copies. In any case, this gene region does not seem to be appropriate for interspecific investigations within Dermanyssus, nor for interfamily exploration within Dermanyssoidea. And this is in accordance with results of Goetze [34] in Calanoidea (Crustacea), where paralogous gene copies are likely to be problematic for phylogenetic studies, especially at the species level.

In the present study, the biclade topology described by Roy et al. [10] based on mitochondrial DNA, opposing the group of specialist species (D. carpathicus + D. longipes + Dermanyssus sp. EN + D. hirundinis + D. apodis + hirsutus-group) to the synanthropic gallinae complex is not supported by most of results (Figures 3 and 4). It appears it is a topology retained solely based on COI alone, only with the enlarged set of isolates in Step 2 (Top5, see Appendix 5), and with rather low support values at internal nodes.

Topologies retained in present multi-gene analyses show a biclade structure with gaps as the fifth state and a semi-biclade structure with gaps as missing data (clades ξ and η, see Figure 3), but in any cases, the two clades do not oppose non synanthropic species to the synanthropic one: the generalist and synanthropic D. gallinae in both greedy consensuses groups with D. apodis (clade ξ), which is a swift specialist and is absent from farms, with correct support (max. occurrences per partitioning scheme 4.0, final repetition index 3).

The rooting of the common ancestor is problematic. When using the enlarged set of isolates (Step 2, multi-isolates analyses), pairwise specific relationships appear similar to multi-gene topologies, but a noticeable paraphyly between mitochondrial and nuclear analyses arises (see Table 2, Figures 3 and 4). In multi-gene greedy consensuses, the position of D. hirsutus remains unclear (basal with gaps as missing data, a sister to the clade ξ, i.e., D. gallinae + D. apodis) with gaps as the fifth state. At the specific level, in multi-isolate analyses, scale-like topologies are retained, which are mutually and symmetrically paraphyletic, with D. gallinae as the basal entity in mt-COI analyses and D. longipes, D. sp. EN and D. hirundinis as the basal entity in nuclear Tpm analyses (Figure 4). In Roy et al. [11], a slight incongruence was vaguely suggested between mitochondrial and nuclear topologies, but the nuclear gene region used in this study was not variable enough to establish it firmly (ITS1–5.8S-ITS2).

The species tree is often not identical to the gene tree [38], due to several potential causes. Reduced effective population size (Ne) in mitochondrial DNA compared to nuclear DNA (a quarter in most sexually reproductive organisms, a third in the present case, due to haplodiploidy) often causes a high mutation rate in mitochondrial DNA resulting in less resolved internal relationship in mitochondrial topologies than in nuclear topologies. As a result, in mitochondrial gene trees, the most recurrent bias is due to homoplasy, and inconsistencies in nuclear gene trees are due to the stochastic effects of lineage sorting. Additionally, interspecific hybridization in some cases may induce reticulation [39].

The choice of a more appropriate gene tree for relationships between species requires some attention. Some authors considered more appropriate mitochondrial markers for inferring phylogenies at the specific level [40,41] but they were dealing with organisms with smaller numbers of generations per year (birds, rodents; D. gallinae in natura, around 15 gen/y, in farms, >200 gen/y), and thus with likely reduced mutation rates in both mitochondrial and nuclear genomes. McCracken et al. [39] recommended a balanced approach, taking into account both advantages and flaws due to different effective population size Ne and considering first whether independent gene trees are adequately resolved and then whether those trees are congruent with the species history.

In the present case, a bias may come from the outgroups: no very close outgroup has been integrated into the analyses, because no individual of Liponyssoides, the only other genus within Dermanyssidae, has been found despite an intense screening of potential hosts. As a result, only other Dermanyssoid families are represented as outgroups (Macronyssidae, Laelapidae, Varroidae). Bayesian topologies reveal important distances between in- and outgroups, which might explain some difficulties in trying to clearly locate the ancestral rooting point. The different genera integrated here as outgroups are rather diverse, and provide good coverage across Dermanyssoidea. Since there are currently no published phylogenetic hypotheses for Dermanyssoidea, it is unknown what is the most appropriate outgroup, which is why it was important to have a diverse and a priori paraphyletic set of outgroups. Additionally, the results indicate that some clades are better, or more reliable than others, and consequently may help for the choice of the most reliable topology: clade η (recovered in both greedy summaries in multi-gene analyses, see Table 2, Figure 3) is not recovered in any multi-isolate analyses, and yet its maximum number of occurrence in partitioning schemes is 3.0 and its repetition index is 1 in multi-gene analyses (both gaps as missing data and as the fifth state). On the other hand, clade ξ (D. gallinae + D. apodis) receives higher support values in both the multi-gene MP analyses (4.0/3). And yet this clade is present in Tpm multi-isolate analyses, whereas absent in mt-COI multi-isolate analysis. Moreover, nuclear results show much less homoplasy (cf. CI and RI in Table 6) than mitochondrial results. And the most supported internal nodes in multi-gene analyses also occur in Tropomyosin-based multi-isolate analyses. Finally, a comparable topology was already suggested based on ITS sequences by Roy et al. [11], but the low amount of DNA divergence in this sequence did not provide enough resolution. It was only vaguely suggested and the few resolved internal nodes were not supported.

The basal position of D. hirsutus in the greedy consensus for multi-gene analysis treating gaps as missing data may be due to the lack of indel information present in Tropomyosin intron n. Kawakita et al. [24] have shown that inclusion of intronic gap characters consistently contribute to phylogenetic reconstruction, at least at lower taxonomic levels. Results of indels alone analysis (Appendix 6) strongly suggest that Tpm intronic indels contains important and consistent phylogenetic information and represents interesting complements to the information present in gap-free corresponding intronic portions. And yet, the basal position of D. hirsutus is less supported in gaps as missing data analyses than its median position in gaps as the fifth state (see Figure 3). This strongly suggests that the split between hirsutus group and gallinae group of Moss [42] is not valid.

Non gallinae species are more or less specialized, whereas D. gallinae has been encountered in nine different bird orders in France by Roy et al. [10]. Dermanyssus apodis and the French isolates of D. hirundinis are rather strict specialists, as exclusively encountered on the genus Apus for the former and in the family Hirundinidae for the latter in France in present study as well as in Roy et al. [10,11]. Looking like intermediate entities on the host specialization point of view, Dermanyssus sp. EN and D. carpathicus are known from two to three different bird families, but both within Passeriformes: Dermanyssus sp. EN on Paridae (in France, present data), Muscicapidae and Sylviidae (in Sweden [30]), D. carpathicus on Paridae and Muscicapidae (present data). The two incompletely isolated lineages within D. carpathicus (Lmt4 and Lmt5) do not appear to be restricted to any of the two passeriform families (see Appendix 1). As for Dermanyssus sp. EN, it is likely to be a moderate specialist, as is D. carpathicus, i.e., parasitizing various bird passeriform families, which are not as conspicuously distant between each other as are D. gallinae’s hosts (bird taxonomic considerations follow Peterson’s classification [46]), but additional samples are needed to establish it clearly. Finally, not enough isolates of D. longipes s. str. are available here to estimate the host range of this species.

In all topologies except the multi-isolate COI analyses, a derived state of generalist vs. specialist condition is noted: D. gallinae, the only generalist species in present study [10] is in a distal position. This may indicate, as already shown in the fish ectoparasite Lamellodiscus (Trematoda: Monogenea) [2], as well as in some free-living insects such as bees [3] and springtails (Collembola) [4], that, contrary to usual expectations, the specialist condition does not appear as a “dead-end”. Dermanyssus gallinae, the generalist species, is, if not the more derived species, at least one of the distal ones with D. apodis, whereas basal positions are occupied by specialist species. And it is worth noting that special lineage L1 of D. gallinae has been almost solely encountered in pigeon nests: only two individuals have been found, both dead, in other birds nests (JGC1 and GO8 in Roy et al. [11]), the one in a bird of prey’s nest, the second, in a black swift’s nest, which is known to be a frequent competitor of pigeons for nest site. Consequently, both these non-pigeon occurrences are likely fortuitous. The precise position of L1 is not absolutely clear, either basal to non L1 D. gallinae, or branching from within D. gallinae. In any case, it appears to be currently reproductively isolated and seems to represent a species recently isolated (no morphological difference with non L1 D. gallinae, low divergence based on nuclear loci Tropomyosin intron n and ITS1 and 2).

The loose relation of Dermanyssus micropredator mites to their host/prey should have led one to expect a fundamentally wider host range in Dermanyssus. And yet, the case of bees’ ancestral specialist condition as evidenced by Danforth et al. [3] is the most comparable: despite nectar and pollen collectors do not live on nor develop on their resource-plant, the placement of a paraphyletic Melittidae s.l. at the base of the phylogeny indicates that host–plant specialization is the primitive state.

Interestingly, the important difference observed in the haplotype variability between species does not seem to be due to the bird’s ecology. Apparently, an important intermingling involving a large number of Tpm haplotypes in D. gallinae is noticeable within a colony of starlings (focused isolate IL, Table 4, Figure 4a). But this does not seem to be solely correlated to the bird’s ecology, as we noted the exact contrary in D. apodis individuals from a colony of swiftlets (focused isolate GO, two Tropomyosin haplotypes) and in D. hirundinis from 6 separate French colonies of barn swallows (focused pseudo-isolate DhirF, one Tropomyosin haplotype). Thus starlings, swiftlets and swallows reuse nests of other pairs in the same colony from one year to another (O. Caparros, CRBPO, MNHN, pers. comm.). This could suggest that D. gallinae would have the opportunity to move from one nest to another by phoresy on the bird host (see above). But the ability of D. apodis to get transferred by the birds has also been shown by several adult females directly sampled on (flying) hosts [10, 11]. The conspicuous stability of both genes under test in Step 2 in D. apodis, D. carpathicus and D. hirundinis strongly contrasts with their variability in D. gallinae, inter-isolate in COI and inter- as well as intra-isolate in the nuclear Tropomyosin. This suggests that these represent two types of species that are intrinsically very different, although sister to each other.

This difference between Dermanyssus species might be a consequence of interspecific hybridizations within the D. gallinae complex. The monophyly of D. gallinae is not doubted here as it is recovered in all topologies. Nevertheless, this is supported by few synapomorphies and the ratio external/internal mt branch length is below (1.1) the lower limit of mt clades to which monophyletic nuclear clades correspond (<1.5). This could suggest that the date of this coalescence occurred much later than coalescences for other species. But the distance of the coalescent node for D. gallinae to the common ancestor with the closest species D. apodis (considering that the right arrangement is recovered by Tropomyosin multi-isolate topologies) is also by far the shortest (see Bayesian topology, Figure 4a). Either the two loci under test in D. apodis, D. hirsutus and D. carpathicus have evolved faster than in D. gallinae, or the apparent low rate of evolution within the latter is rather an artifact due to a radiation followed by interbreeding. The latter alternative would appear much more credible since (1) the number of nuclear segregating sites and or indel sites is significantly the highest within D.gallinae focused isolates as opposed to focused isolates or even pseudo-isolates in others species, which is suggestive of the assemblage of formerly highly divergent haplotypes (Table 4), (2) many Tpm genotypes of heterozygous individuals show high divergence percentages within D. gallinae, (3) the specialist lineage L1 with similar stability in both COI and Tpm sequences to other species’ is branching from within the widely generalist D. gallinae in multi-isolates analyses and basal species are at least moderate specialists, (4) clades Lmt3 and Ln3 share a major part of their isolates and are strongly supported in mitochondrial and nuclear analyses (indels alone and gaps not considered for the nuclear clade Ln3), whereas showing important interbreeding. These two clades could provide evidence of a vestigial completely isolated lineage L3, whose speciation has been in reversion due to subsequent interbreeding.

All that suggests that D. gallinae appears as a complex of species, which would have interbred soon after speciation, and just before pre- or post-zygotic incompatibility has been installed. Such a reversal of speciation has been shown to occur in various organisms in case of habitat defragmentation, which induce relaxed divergent selection and increased gene flow because of loss of ecological barriers [47]. Naturally, this potential scenario is no more than a hypothesis, which needs some complementary investigations before being clearly established in D. gallinae. In such a case, the idea evoked by Futuyma and Moreno [1] that specialization and rate of diversification are positively correlated would not be so clearly contradicted by the differential variability depending on specialization within Dermanyssus. The increased molecular diversity in the generalist species would not seem to be due to continuous and intrinsic evolution, but a consequence of the diversification of specialist lineages and consecutive intermingling. It is also worth noting that the only other species with allelic divergence involving some indels in Tropomyosin is D. apodis, the sister to D. gallinae. This suggests that the clade ξ might be particularly prone to hybridize, more than close groups, as it has been shown in some plant groups [48].

Of course, observed differential divergence levels between lineages within D. gallinae complex might suggest a continuum of genetic divergence from sympatric host races to species as recently evidenced in the pea aphid complex [51]. But phylogenetic inferences obtained in present studies, by integrating several sister species and so providing us with a historical scenario, strongly suggests that the gradient of genetic divergence observed here is evolving from weak variability to high variability in the tree of 6 species of Dermanyssus, not the reverse pattern. It’s likely the present markers are too slowly evolving to detect new speciation events possibly in process within the gallinae complex. But from the present data it is to be concluded that the generalist condition of D. gallinae has been arising from a specialist ancestor.