Weak but Repeated Patterns of Co-Introgression of Nuclear OXPHOS Genes and Mitochondrial DNA in Iberian Wall Lizards

Abstract

In this study, we took advantage of the previously reported instances of mitochondrial DNA (mtDNA) capture in the Podarcis Iberian group, a speciose group of Iberian wall lizards, to test the hypothesis that nuclear genes from the OXPHOS (oxidative phosphorylation) chain can co-introgress with the mitochondria as an evolutionary response to mitigate the costs of mitonuclear incompatibilities. Using dense population sampling and transcriptome data, we generated capture-sequence datasets for nuclear OXPHOS chain genes (nucOXPHOS), random nuclear loci (nucControl) and the complete mitochondrial genome. Phylogenetic analyses of nuclear and mitochondrial genes confirmed two previously identified events of mitochondrial introgression in the Podarcis Iberian group and revealed two new cases. Three of these cases have led to complete local mtDNA replacements, where the introgressed mitotypes have replaced the native ones in several populations, and involve a currently unknown and presumably extinct donor species, so-called “ghost lineage.” Detecting introgression from ghost lineages, whose genomes are not accessible, remains challenging. To overcome this issue, we designed or adapted several tests aimed at detecting differential signals of introgression between our nucOXPHOS and nucControl gene sets. One of these tests, based on the effects of introgression on branch lengths in phylogenetic trees, uncovered a weak but consistently significant signal of partial co-introgression of nucOXPHOS genes compared to the genomic background (represented by the nucControl gene set) in three out of four cases of mtDNA introgression.

Significance.

The role of mitonuclear interactions in shaping patterns of introgression remains poorly understood. Here, we studied multiple independent cases of mitochondrial DNA capture in Iberian Podarcis lizards, and tested whether nuclear oxidative phosphorylation genes preferentially co-introgress with mitochondrial DNA allowing for the mitigation of incompatibilities. Using dense sampling and targeted sequencing, we uncovered a weak but consistent signal of preferential co-introgression in three of four events, some of which involved complete mitochondrial DNA replacements from ghost lineages. These findings reveal that mitonuclear co-introgression, though subtle, can leave detectable signatures across repeated introgression events.

Introduction

Introgression can be defined as the incorporation of genetic material from a differentiated population into the genome of another differentiated population through hybridization and subsequent backcrossing. Interspecific introgression, the transfer of genetic material from one species to another, has long been regarded as an exception to the rule of reproductive isolation between species. Progress in genomic techniques has now revealed that valid biological species can still exchange genes during a significant part of their history and that interspecific introgression is thus widespread among living organisms (Mallet 2005; Taylor and Larson 2019).

Within the genome of the recipient species, most introgressed heterospecific alleles are likely to be counter-selected or neutral, as expected from the genomic basis of speciation. Most counter-selected alleles will be removed by selection, while neutral alleles may escape linkage to counter-selected loci through recombination and behave neutrally in the foreign genetic background. More rarely, introgression can be favored by positive selection on foreign alleles when they are beneficial in the genomic background of the receiving species, a process called adaptive introgression (Hedrick 2013; Burgarella et al. 2019; Edelman and Mallet 2021). Although it remains a difficult phenomenon to demonstrate, adaptive introgression has been identified in various groups, involving for example thermal metabolism (Rocha et al. 2023; Horníková et al. 2024), seasonal camouflage (Jones et al. 2018), resistance to pesticides (Song et al. 2011; Clarkson et al. 2014; Svedberg et al. 2021), adaptation to altitude in humans (Huerta-Sánchez et al. 2014), mimicry in Heliconius butterflies (The Heliconius Genome Consortium 2012), adaptation to abiotic conditions in sunflower Helianthus (Spear et al. 2023), or the major histocompatibility complex in vertebrates (Gaczorek et al. 2024).

A special case of introgression is interspecific mitochondrial introgression, the transfer of mitochondrial DNA (mtDNA) across the species barrier. Because mtDNA is clonally transmitted without recombination, foreign mtDNA is never diluted by generations of backcrossing. The introgressed mtDNA can be lost, fixed, or maintained polymorphic in a population. Gershoni et al. (2009) hypothesized that mitochondrial genes should be over-represented in reproductive isolation. If verified, this supposition would make mitochondrial introgression highly unlikely. The basis of this hypothesis stems from the importance of mitochondria for the functioning of the organism. The main function of mitochondria is the production of energy in the cells through the oxidative phosphorylation (OXPHOS) pathway. The OXPHOS chain comprises five complexes, four of which are made of proteins encoded by both nuclear and mitochondrial genomes, with these proteins physically interacting (Rand et al. 2004). When nuclear and mitochondrial genomes evolve independently in separate populations, incompatibilities may arise that can disrupt the proper functioning of the OXPHOS chain when nuclear and mitochondrial alleles from diverging populations are interacting. Such incompatibilities can impair metabolic efficiency and may result in unviable or less fit hybrid individuals (Ellison and Burton 2006; McDiarmid et al. 2024; Moran et al. 2024), leading to reproductive isolation between diverging populations.

While these observations suggest that mitonuclear incompatibilities can contribute to the evolution of reproductive isolation, there is little definitive evidence supporting their primary role in the speciation process (Burton 2022) and, contrary to this expectation, empirical evidence suggests that mitochondrial introgression is a widespread phenomenon (reviewed in Toews and Brelsford 2012; see also Kunerth et al. 2022; Potter et al. 2024; Eriksen et al. 2025 and references therein for more recent examples). In some systems, mitochondrial introgression has been suggested to be adaptive, which seems even more surprising (Llopart et al. 2014; Hulsey et al. 2016).

One potential explanation to solve this paradox is that mtDNA introgression may be accompanied by the co-introgression of nuclear OXPHOS genes that previously coevolved with the introgressed mtDNA (reviewed in Hill 2019). This co-introgression of nuclear OXPHOS genes with foreign mtDNA may help to overcome the incompatibilities inherent to mitochondrial introgression. However, previous empirical examination of nuclear introgression in organisms with mitochondrial introgression either failed to detect preferential co-introgression of nuclear loci from the OXPHOS chain (Bailey and Stevison 2021; Mikkelsen and Weir 2023; Kato et al. 2024) or identified preferential co-introgression of only a few of the genes that interact with mitochondria (eg Beck et al. 2015; Evans et al. 2021; Wang et al. 2021; Farleigh et al. 2023; Jensen et al. 2023; Moran et al. 2024). Only a few studies have documented the massive co-introgression of nuclear genes that interact with mitochondrial components, including OXPHOS genes, with mtDNA. This process involves a whole set of genes showing preferential introgression relative to the rest of the nuclear genome (Ward et al. 2022; Forsythe et al. 2026).

The lacertid genus Podarcis has been suggested to have a reticulated evolutionary history, with numerous instances of ancient admixture events during its diversification (Yang et al. 2021). Recent or even contemporary introgression events are also widespread, especially in the Iberian clade of the P. hispanicus complex (hereafter, the Iberian group, Caeiro-Dias et al. 2021a; Gaczorek et al. 2023). The Iberian group also includes a rare case of massive mitochondrial introgression from a ghost lineage (ie a lineage that is now extinct or unsampled). Indeed, in large areas of the distribution of P. liolepis and in some populations of P. hispanicus, the original mtDNA has been replaced by introgressed mtDNA from the same ghost lineage (Renoult et al. 2009). Another potential case of massive introgression is suggested by the discordant phylogenies of mitochondrial versus nuclear DNA among populations of P. hispanicus (compare Kaliontzopoulou et al. 2011 with Yang et al. 2021). The Iberian group of wall lizards thus represents a promising system for assessing the generality of co-introgression of nuclear OXPHOS genes.

In this study, we use target capture data to test whether mtDNA introgression is correlated with preferential co-introgression of nuclear OXPHOS chain genes in lizards of the Iberian group. Using a previously generated transcriptome of Podarcis from the Iberian group (Chiari et al. 2012), we built a set of probes targeting 149 nuclear genes unrelated to mitochondrial functions (nucControl genes), 76 nuclear loci of the OXPHOS chain (nucOXPHOS genes) and the complete mitochondrial genome. We sequenced these genes in all described or candidate species of the Iberian group, including the North African lineages (see Kaliontzopoulou et al. 2011, 2012), with increased sample sizes for species where mtDNA introgression had been previously suspected. We then used these data to build independent phylogenies based on all concatenated nuclear loci and on complete mitochondrial genomes. Comparing the topologies of the trees obtained with these two datasets allowed us to confirm multiple instances of mitochondrial introgressions. Some of these mitochondrial introgressions involve mtDNA from lineages that do not correspond to any current evolutionary units (ghost lineages).

We tested whether populations or individuals carrying introgressed mtDNA exhibited an excess of nucOXPHOS introgressed alleles relative to the genome average, represented by nucControl genes in our dataset. As we generally do not have any information on the nuclear genomic background of the ghost lineages, we were unable to directly quantify the nuclear contributions from these ghost lineages into the individuals carrying their mtDNA. We thus designed a series of indirect tests (detailed in Table 1 and Materials and Methods) comparing groups of introgressed and non-introgressed individuals for statistics computed on the nucOXPHOS and nucControl genes. Briefly, in case of preferential co-introgression of OXPHOS genes compared to control genes, we would expect different topologies and branch length distributions for trees built on the two datasets, more phylogenetic discordance [as measured by quartet concordance factors (qCFs)] among OXPHOS genes, more genetic differentiation between introgressed and non-introgressed groups of individuals for OXPHOS genes and higher nucleotide diversity for OXPHOS genes in introgressed individuals. Tests based on trees (topology, qCFs, branch lengths) and comparisons of p-distances were performed on concatenated alignments of nucOXPHOS and nucControl datasets, population genetics metrics (Fst, pi) were computed gene-by-gene and their distributions compared between nucOXPHOS and nucControl datasets. Tests based on concatenated datasets were also repeated by grouping nucOXPHOS genes per OXPHOS chain complex (Appendix 1).

Table 1.

Expectations under the hypothesis of co-introgression of nuclear OXPHOS genes (nucOXPHOS) with mitochondrial OXPHOS genes (here the mtDNA gene set)

Type of test	Metrics or patterns	Prediction if nucOXPHOS loci co-introgressed with the mtDNA
Phylogenetics	Topology	Topology of nucOXPHOS = mtDNA and nucOXPHOS ≠ nucControl
Phylogenetic discordance	Quartet concordance factor	More discordance in nucOXPHOS than in nucControl genes
Basal attraction of branches with introgression events	Branch length	nucOXPHOS branch length < nucControl for introgressed individuals
Genetic differentiation	Fst, p-distance	nucOXPHOS > nucControl
Nucleotide diversity	pi	For nucOXPHOS: introgressed ≠ ancestral mitotype; for nucControl: introgressed = ancestral

Several metrics are used to compare nucOXPHOS and mtDNA genetic information to nuclear Control genes information (nucControl). Tests based on trees (topology, quarter concordance factor, and branch lengths) and comparisons of p-distances were performed on concatenated alignments of nucOXPHOS and nucControl datasets, population genetics metrics (Fst, pi) were computed gene-by-gene and their distributions compared between nucOXPHOS and nucControl datasets.

Results

Nuclear Phylogeny

After removing genes without coverage, genes with paralogy and nucControl genes that interact with the mitochondria, we retained 67 nucOXPHOS and 144 nucControl genes out of the initial 76 and 149, respectively (Table S1). Our sampling includes a total of 288 individuals (individuals’ locations in Fig. S1 and Table S2). This filtered all-species concatenated nuclear dataset includes 187,501 base pairs (bp; 45,030 for nucOXPHOS genes and 142,471 for nucControl loci, of which 2,093 and 6,103, respectively were informative sites). As expected, the nuclear tree built on concatenated nuclear nucOXPHOS and nucControl genes groups individuals by species, confirming the validity of the current taxonomy of the group (Figs. 1 and 2). Yang et al. (2021), using whole-genome of one sample of each lineage of the species we study here, recovered a broadly similar tree, except for Podarcis sp. Maghreb (named P. vaucheri SL in Yang et al.), which is sister to P. vaucheri in their tree but is sister to a P. vaucheri plus P. hispanicus clade in our tree (Fig. 1a). A second difference between our tree and Yang et al. is that P. carbonelli does not group with P. liolepis in their tree.

Fig. 1.

Phylogenies of concatenated a) nuclear control loci (nucControl) and b) mitochondrial genes (mtDNA) inferred with IQ-TREE showing several discordances highlighted by colored bands. Individual tips are colored by species. The tree was rooted with Teira dugesii and Podarcis tiliguerta (not shown for simplicity). Node supports are shown for major branches as follows: SH-aLRT/UFBoot. The branches leading to the mitotypes of interest are annotated by their abbreviation.

Fig. 2.

Phylogenies of a) nuclear OXPHOS (nucOXPHOS) and nuclear control (nucControl) concatenated datasets, b) nucOXPHOS genes only, and c) mitochondrial genes (mtDNA) inferred with IQ-TREE. Individual tips are colored by species. The tree was rooted with Teira dugesii and Podarcis tiliguerta (not shown for simplicity). Node supports are shown for major branches as follows: SH-aLRT/UFBoot.

Within P. liolepis, three evolutionary units are recovered in the nucControl phylogeny (Fig. 3a). The earliest diverging one corresponds to populations south of Valencia and north of Benidorm, hereafter named liolepis_south (colored in blue in Fig. 3). The next group that branches off corresponds to the isolated populations of P. liolepis from the mountains in southeast Spain (Sierra de Cazorla and Sierra Espuña), which are disconnected from the rest of the species’ range and surrounded by the distribution of P. hispanicus, P. vaucheri, and P. virescens; they constitute the liolepis_sierras lineage (colored in gray in Fig. 3). The rest of the populations, from Valencia to the north of the species distribution, constitute the P. liolepis northern lineage (colored in red, green, and purple in Fig. 3). This phylogeographic structure forms the basis of our definition of groups for some tests of co-introgression (see below).

Fig. 3.

Phylogenies and geographic locations of the grouping of individuals used to test for co-introgression of nucOXPHOS genes in Podarcis liolepis. Phylogenies focusing on the P. liolepis clade within the a) concatenated nuclear control (nucControl) genes tree and b) the concatenated nuclear OXPHOS (nucOXPHOS) genes tree produced in IQ-TREE. Individual tips are colored by group, and the shape represents the mitotype carried. Node supports are shown for major branches as follows: SH-aLRT/UFBoot. The yellow star in (a) represents the root of the P. liolepis northern nuclear lineage. c) Locations of the Podarcis liolepis individuals sampled, colored by group and mitotype (respectively, inner and outer circles). The inset maps show the locations of populations where both the native and introgressed mitotypes are found.

A strong phylogeographic structure is also apparent within P. hispanicus, with two main clades exhibiting allopatric distributions. The first one, occurring mostly north of Murcia, is hereafter called AM (for Albacete-Murcia) in accordance with previous studies (colored in purple in Fig. 4). The southern lineage, south of Murcia (including the village of Galera), is called the Gal lineage as in previous studies (colored in black and green in Fig. 4). A single individual which branches out at the crown node of the P. hispanicus clade is briefly discussed below as it also carries a unique mitotype.

Fig. 4.

Phylogenies and geographic distributions of the grouping of individuals used to test for co-introgression of nucOXPHOS genes in Podarcis hispanicus. Phylogenies focusing on the P. hispanicus clade within the a) concatenated nuclear control (nucControl) genes tree and b) the concatenated nuclear OXPHOS (nucOXPHOS) genes tree produced in IQ-TREE. Individual tips are colored by group. Node supports are shown for major branches as follows: SH-aLRT/UFBoot. c) Locations of P. hispanicus with the individuals colored by group.

Mitochondrial Phylogeny and Inference of Mitochondrial Introgression Events

The final complete mtDNA alignment includes 11,400 bp with 4,724 informative sites. Our mitochondrial phylogeny is similar to the mitochondrial tree of Yang et al. (2021) built on complete mitogenomes and the tree built by Kaliontzopoulou et al. (2011) using a shorter DNA alignment. The only difference between our mitochondrial tree and their trees is that P. lusitanicus and P. bocagei are sister species in our tree (Fig. 1b) while P. lusitanicus is sister to P. guadarramae in the two other studies (these species are called 1A and 1B, respectively in Kaliontzopoulou et al. 2011). These discrepancies might be explained by the low support of our tree at these nodes (Fig. 1b).

Our more extensive dataset also uncovered two new, rare mtDNA lineages represented by single individuals from Spain with long internal branches in our tree: one P. liolepis (BEV.9863 from El Dosel south of Valencia, Table S2) that does not show any sign of nuclear differentiation (Figs. S2 and S3) and one P. hispanicus (BEV.7050 from Alcaudique near Almeria, Table S2) which appears to be the earliest diverging individual of its species clade in the nuclear tree with short branches, suggesting admixture from a neighboring species (Fig. 4a, Figs. S2 and S3). Therefore, the latter individual was removed from subsequent analyses.

After inferring phylogenies, we noticed several mitonuclear discordances in our dataset (Fig. 1). In order to test whether the mitochondrial phylogeny is significantly discordant from the nucControl phylogeny, we used the approximately unbiased test. We obtained a log-likelihood difference of 87,736.832 between the mitochondrial and nuclear topology, corresponding to a P-value <0.0001. The nuclear topology is therefore not supported by mitochondrial genetic information.

The first discordance between nucControl and mtDNA phylogenies involves some P. vaucheri that have the same mitotype as P. carbonelli (carbonelli mitotype hereafter carb), suggesting a recent introgression of the mtDNA from P. carbonelli into P. vaucheri (highlighted by pink bands in Fig. 1). These P. vaucheri originate from Matalascañas in southern Spain where the two species live in syntopy and where hybridization has previously been reported (Caeiro-Dias et al. 2021a). The other P. vaucheri individuals from this population and all other P. vaucheri carry the ancestral P. vaucheri mitotype (vauch).

The second discordance between nucControl and mtDNA phylogenies involves P. liolepis, which is monophyletic in the nuclear phylogenies but is made of two highly divergent and non-sister clades in mtDNA (highlighted by orange bands in Fig. 1). One lineage forms a highly divergent clade together with the mtDNA Gal lineage defined below and matches the P. liolepis placement in the nuclear phylogeny, we thus consider this mitotype to be the ancestral one (lio mitotype hereafter). The other mitotype is closely related to the P. hispanicus mtDNA lineage and is embedded in the clade composed of P. hispanicus, P. vaucheri, and Podarcis sp. Maghreb: we consider this lineage to result from a ghost introgression from an extinct lineage, as no source species is known for this lineage. This ghost mitotype will be called Val (for Valencia) as in Renoult et al. (2009) (it is the one called “P. hispanica s.s.” in Kaliontzopoulou et al. 2011). The Val mitotype is the only one found in P. liolepis south of Valencia, in the liolepis_south and liolepis_sierras nuclear lineages. The lio mitotype is the only one found in the northern part of the distribution of P. liolepis. In between the two areas, the two lineages have a complex limit with some overlap across the east of central Spain (Fig. 3c). The two mitotypes are thus found in the P. liolepis northern evolutionary lineage (Fig. 3a). Note that the two evolutionary lineages corresponding to liolepis_sierras and liolepis_south do not form a monophyletic clade in our trees; there might thus have been two independent introgressions of the Val mitotype into each of them separately. We refrain from concluding on this issue here because we feel our phylogenetic reconstructions are not robust enough (see Discussion).

Finally, while P. hispanicus constitutes a monophyletic clade in nuclear phylogenies, its representatives are spread in three different places in the mitochondrial phylogeny, corresponding with three distinct mitotypes and suggesting at least two introgression events. (i) The AM mtDNA lineage (Albacete/Murcia in Kaliontzopoulou et al. 2011) is closely related to the P. vaucheri mtDNA lineage, in agreement with the nuclear phylogeny; we therefore interpret this lineage as the ancestral mitotype of P. hispanicus. It is the only mitotype in the hispanicus_AM nuclear group (Fig. 4). (ii) The Gal mtDNA lineage (Galera in Kaliontzopoulou et al. 2011) is sister to the ancestral P. liolepis mtDNA lineage and hence has a position in the tree which is strongly discordant from the nuclear phylogeny; no other species is currently known to harbor this lineage and we interpret this as a past introgression from a ghost lineage. An alternative hypothesis is an ancient bidirectional introgression between P. liolepis and P. hispanicus, as discussed below. Gal is the most common mitotype in the hispanicus_Gal nuclear cluster. (iii) A small number of P. hispanicus individuals sampled in two populations in the Sierra de María, part of the hispanicus_Gal nuclear lineage, all carry the Val mitotype, shared with southern populations of P. liolepis (Fig. S2). In a distant population of hispanicus_Gal on the northern foothill of the Sierra Nevada (La Calahorra), out of two sequenced individuals, one carries the Gal mitotype and the other the Val mitotype. As in P. liolepis, the occurrence of these mitotypes is thus strongly structured phylogeographically.

To conclude, we infer at least four independent mitochondrial introgression events in Podarcis, two of which had already been identified by Renoult et al. (2009). One of them (carb mitotype intro P. vaucheri) is limited to a single population with current hybridization and is most likely very recent while the other three are not accompanied by any current hybridization, involve a “ghost species” as donor and are likely ancient given the deep branching of discordant mitochondrial trees for these lineages.

Test of Co-Introgression of nucOXPHOS Genes With mtDNA Based on Topologies

In both the nucControl and nucOXPHOS phylogenies (Figs. 1a and 2b), species form monophyletic clades, regardless of the mitotype of the individuals or populations. If nucOXPHOS genes had massively co-introgressed with introgressing mtDNA, we would expect populations or individuals with introgressed mitotypes to have different placements in the nucOXPHOS and nucControl phylogenies, with individuals carrying introgressed mitotypes placed further away from conspecific individuals with native mitotypes. We did not find this pattern (compare nucOXPHOS and nucControl phylogenies in Figs. 1 and 2), suggesting a lack of massive co-introgression of nucOXPHOS genes. We also looked for this pattern in trees built by OXPHOS complex (Appendix 1), but their poorly supported topologies made them difficult to interpret. Of course, this very crude test does not preclude that less extensive co-introgression occurred for a few nucOXPHOS genes, we thus tested for more subtle signals of co-introgression using population genetics and other methods as detailed below.

Nonetheless, comparison of the nucControl and nucOXPHOS phylogenies reveals several topological differences between the two datasets, concerning internal nodes affecting interspecies groupings (Figs. 1a and 2b). Those differences are significant with a log likelihood difference of 4,833.268 (reference = nucControl, AU test, P-value <0.0001). Intriguingly, in two instances, the nucOXPHOS topology matches the mitochondrial phylogeny when the nucControl and complete concatenated datasets do not: P. carbonelli is sister to P. virescens and P. sp. Maghreb is sister to P. vaucheri in mtDNA and nucOXPHOS trees, but not in nucControl and total nuclear datasets (Figs. 1 and 2).

Tests of Co-Introgression of nucOXPHOS Genes With mtDNA Based on Within-Species Comparisons

Definition of the Groups for the Population Genetics Tests

In order to test for preferential co-introgression of nucOXPHOS genes together with the introgressed mtDNA, we used previous results on within-species populations relationships and distribution of native and introgressed mitotypes to define groups, allowing us to compare individuals with the introgressed mitotype to individuals with the ancestral mitotype while accounting for phylogenetic relationships and geographic origin of the individuals (Table 2). In the rest of this article, groups will be named as “species_group(mitotype abbreviation)” or “species(mitotype abbreviation).” These groups form the basis of the tests introduced in the Materials and Methods section (Table 1) and are defined below for each species. The tests were performed for each mitotype introgression event: carb into P. vaucheri, Gal into P. hispanicus, Val into P. hispanicus (hispanicus_Gal nuclear lineage), and Val into P. liolepis (Table 2).

Table 2.

Summary results of the tests classified by introgression events and pairs of groups chosen

Species	Ancestral mitotype	Introgressed mitotype (abbreviation, origin)	Group comparison	qCF	Branch length	Fst	P- distance	pi
Podarcis vaucheri	vaucheri (vauch)	carbonelli (carb, P. carbonelli)	vaucheri(carb) - vaucheri(vauch)	ns	ns	ns	ns	ns
			vaucheri_Mat(carb/vauch) - vaucheri_Ref(vauch)	ns	(−)	ns	ns	ns
Podarcis liolepis	liolepis (lio)	Valencia (Val, ghost)	liolepis(val) - liolepis(lio)	ns	(+)	…	…	…
			liolepis_sierras(Val) - liolepis_Ref(lio)	…	…	ns	ns	ns
			liolepis_south(Val) - liolepis_Ref(lio)	…	…	ns	ns	ns
			liolepis_cuenca(Val) - liolepis_Ref(lio)	…	…	ns	ns	ns
			liolepis_central(Val) - liolepis_central(lio)	…	…	ns	ns	ns
Podarcis hispanicus	Albacete-Murcia (AM)	Galera (Gal, ghost),	hispanicus_Gal(Gal) - hispanicus_AM(AM)	ns	(+)	ns	ns	ns
		Valencia (Val, ghost)	hispanicus_Gal(Val) - hispanicus_Gal(Gal)	ns	…	ns	ns	ns
			hispanicus_Gal(Val) - hispanicus_AM(AM)	…	(+)	…	…	…

Group nomenclature is as follows: “species_group(mitotype abbreviation)” or “species(mitotype abbreviation)”. “(+)” and “(−)” indicate that the metrics comparison was significant for the test respectively in the expected or opposite direction of our hypothesis. “ns” indicates that the metrics were not different between nucControl and nucOXPHOS loci. Points left means that the test was not done for this group comparison (see Materials and Methods). qCF stands for quartet concordance factor.

Introgression of the carb Mitotype into P. vaucheri

For this analysis, we discarded three individuals due to their high level of admixture, as revealed by a pre-analysis PCA (Fig. S4). We first divided Podarcis vaucheri into individuals carrying the ancestral vauch mitotype, hereafter named vaucheri(vauch) (n = 32) and individuals carrying the carb mitotype, hereafter vaucheri(carb) (n = 9) and all sampled in Matalascañas (Fig. 5). These two groups can be compared to test for potential co-introgression of nucOXPHOS genes with the introgressed mtDNA (Table 2). This type of comparison between individuals carrying either mitotype irrespective of their population would be meaningful only if incompatibilities between mismatching mitotype and OXPHOS nuclear alleles were strong enough to affect the survival of the individuals involved, as detected in Evans et al. (2021). These types of strong effects at individual level might not be the most common, however. Instead, having alien mitotypes segregating inside a given population might exert positive selection for co-adapted OXPHOS nuclear alleles within the same population. To account for this type of population-level selection, we defined a second grouping for P. vaucheri opposing all individuals from the Matalascañas locality [where the carb mitotype occurs in P. vaucheri individuals, vaucheri_Mat(carb/vauch) (n = 17)] and all other individuals of P. vaucheri [vaucheri_Ref(vauch) (n = 24)].

Fig. 5.

Phylogenies and geographic distributions of the grouping of individuals used to test for co-introgression of nucOXPHOS genes in Podarcis vaucheri. Phylogenies focusing on the P. vaucheri clade within the a) concatenated nuclear control (nucControl) genes tree and b) the concatenated nuclear OXPHOS (nucOXPHOS) genes tree produced in IQ-TREE. Individual tips are colored by group. c) Locations of P. vaucheri and P. carbonelli with the individuals colored by group. Inset map represents individuals of the Matalascañas locality where some P. vaucheri individuals carry the P. carbonelli mitotype (carb). For readability, North African individuals are not shown.

All P. vaucheri individuals formed a monophyletic group in the trees built on the nucControl and nucOXPHOS datasets, revealing no major effect of preferential co-introgression on topology (Figs. 1 and 2). qCF values were used to evaluate if introgressed and non-introgressed individuals group monophyletically in nucControl and nucOXPHOS datasets. Phylogenetic discordance was not significantly different between nucOXPHOS and nucControl genes (Table S3), again suggesting that nucOXPHOS genes were not significantly co-introgressing with the carb mitotype. For the test based on branch length and for both types of grouping, we detected a significant effect of the interaction between group and dataset, but the effect was opposite to expectation, with introgressed individuals having longer branches in the nucOXPHOS tree relative to the nucControl tree (Table S4).

For Fst among groups per locus and p-distance between groups for concatenated alignments, no significant difference was detected between the nucOXPHOS and nucControl dataset (Tables S5 and S6) for both types of grouping. In the p-distance test, carbonelli(carb) individuals were used as pop3 (see Materials and Methods). For P. vaucheri, the mitotype origin is known so we also calculated the Fst values between vaucheri(carb) and carbonelli(carb). In case of co-introgression, we would expect lower Fst between these groups (donor and recipient) in nucOXPHOS genes. No differences were found in Fst between the nucOXPHOS and nucControl gene sets (Table S5). Examination of gene-specific pi and Fst values (Figs. S6 and S7) did not reveal more outliers for nucOXPHOS genes.

Finally, we did not detect any significant interaction between gene set (nucOXPHOS or nucControl) and group for the nucleotide diversity (pi value) comparisons between the groups (Table S7).

To conclude, for the carb introgression into P. vaucheri, only branch length revealed significant differences between the two gene sets but in a direction opposite to the predictions based on preferential co-introgression of nucOXPHOS genes. Other tests show no pattern of favored co-introgression of nucOXPHOS genes.

Introgression of the Val Mitotype into P. liolepis

For the qCF and branch length tests, P. liolepis was split into two groups corresponding to (i) individuals with the Val introgressed mitotype (n = 56) and (ii) individuals with the lio ancestral mitotype (n = 64), regardless of geography or phylogeny. No signal of co-introgression was detected when comparing the qCFs of the branch supporting the grouping of these two groups between the nucControl and nucOXPHOS datasets (Table S3). However, branch lengths were significantly lower for the introgressed individuals in the nucOXPHOS dataset, as expected under the preferential co-introgression hypothesis, and this conclusion remained after correction for multiple testing (Table S4).

For the other tests, P. liolepis individuals were divided into six groups based on the species’ phylogenetic structure (phylogeography) and the geographic distribution of the native (lio) and introgressed (Val) mitotypes. The first two groups correspond to the two evolutionary units from the south of the species distribution that are fixed for the Val introgressed mitotype: the populations from the southern sierras, liolepis_sierras(Val) (n = 8), and from the south of Valencia, liolepis_south(Val) (n = 25), colored respectively in gray and blue in Fig. 3. The rest of the individuals form a monophyletic evolutionary unit inhabiting the main distribution of the species from central Spain to southern France where the native and introgressed mitotypes segregate geographically with some limited overlap. Within this lineage, we defined the group liolepis_cuenca(Val) (n = 15) for individuals from the west of the Sierra de Cuenca that are fixed for the Val mitotype and form a monophyletic group nested within the P. liolepis northern lineage (colored in purple in Fig. 3). The rest of P. liolepis individuals are not monophyletic and were grouped based on their geographical origin and mitotype: in the central region, both the introgressed Val and the ancestral lio mitotypes are present, defining a liolepis_central(Val) (n = 8) and a liolepis_central(lio) (n = 24) group (both colored in green in Fig. 3). In the north of the distribution, the native mitotype is fixed and all individuals were assigned to the liolepis_Ref(lio) (n = 40) group (in red in Fig. 3). For most tests, we compared groups where the introgressed Val mitotype is fixed [liolepis_sierras(Val), liolepis_cuenca(Val) and liolepis_south(Val)] to the group where the ancestral lio mitotype is fixed [liolepis_Ref(lio), Table 2] which represents a robust unadmixed reference. We also compared central P. liolepis with and without mtDNA introgression [liolepis_central(Val) vs liolepis_central(lio), Table 2].

For the p-distance test, we used P. vaucheri as pop3 because it is a closely related species to the ghost lineage from which the Val introgressed mitotype originated. However, we excluded the Matalascañas individuals of P. vaucheri from this analysis to avoid interference from the introgressed carb mitotype. For the p-distance test and the diversity test, the four groups carrying the Val mitotype were compared to the liolepis_Ref(lio) group. We did not find any differences in p-distance between nucOXPHOS and nucControl loci (Table S6), and we did not detect any significant interaction between gene set (nucOXPHOS or nucControl) and group for the pi values comparisons, suggesting an absence of co-introgression of nucOXPHOS genes with introgressed mtDNA (Table S7).

For Fst, we did not find any significant difference between the nucOXPHOS and nucControl datasets between groups carrying introgressed or native mitotypes (Table S5), again suggesting an absence of co-introgression of nucOXPHOS genes together with introgressed mtDNA.

To conclude, the test based on branch lengths detected a significant signal of preferential co-introgression of nucOXPHOS genes in P. liolepis but all other tests failed to detect any sign of difference between the nucControl and nucOXPHOS genes (Table 2).

Introgression of AM and Val Mitotypes into P. hispanicus

We divided P. hispanicus individuals into three groups corresponding to the three mitotypes present in this species: (i) hispanicus_AM(AM) (n = 17) for the individuals from the AM nuclear cluster carrying the ancestral AM mitotype; (ii) hispanicus_Gal(Gal) (n = 19) for the individuals from the Galera nuclear lineage carrying the introgressed Gal mitotype; and (iii) hispanicus_Gal(Val) (n = 7) for the few individuals from the Galera nuclear lineage carrying the introgressed Val mitotype (respectively, colored in purple, black and green in Fig. 4).

For the qCF test, discordance had to be measured separately for individuals carrying the Gal introgressed mitotype [branch leading to the (hispanicus_AM(AM), hispanicus_Gal(Gal)) clade] and individuals carrying the Val mitotype [branch leading to the (hispanicus_Gal(Val), hispanicus_Gal(Gal)) clade] to avoid biases in the quartets formed. In addition, when studying potential discordance linked with the Gal mitotype, hispanicus_Gal(Val) individuals were excluded from the analyses to focus on the branch leading to the (hispanicus_AM(AM), hispanicus_Gal(Gal)) clade. When studying discordance following the introgression of the Val mitotype, all individuals were kept in order to be more conservative and we focused on the branch leading to (hispanicus_Gal(Gal), hispanicus_Gal(Val)).

We found that nucOXPHOS genes had significantly more discordance than nucControl genes in the Gal introgression event, suggesting some preferential co-introgression of nucOXPHOS genes in individuals carrying the introgressed Gal haplotype, however that conclusion did not hold after correction for multiple testing (Table S3). No co-introgression signal was detected for the Val introgression event in P. hispanicus (Table S3).

For the branch length test, we compared separately hispanicus_AM(AM) with hispanicus_Gal(Gal) and hispanicus_AM(AM) with hispanicus_Gal(Val). We did not compare hispanicus_Gal(Val) and hispanicus_Gal(Gal) because the shortening of the branches of individuals of one group would blur any signal of other group (which is not a problem for tests of genetic distances or topologies as they do not follow the same logic). The linear mixed models (LMMs) revealed that individuals carrying the introgressed haplotypes had shorter branches in the nucOXPHOS tree relative to the nucControl tree, for both comparisons (groupings), and this was robust to correction for multiple testing (Table S4).

For the p-distance test, we used liolepis_Ref(lio) as pop3 for hispanicus_Gal(Gal) group, because the sister lineage of the Galera mitotype is the liolepis ancestral mitotype. For the hispanicus_Gal(Val) group, we chose liolepis_sierras(Val) as pop3 because the Valencia mitotype carried by hispanicus_Gal(Val) is closest to liolepis_sierras(Val) individuals (see Fig. S2). We did not find any differences of p-distance delta between nucOXPHOS and nucControl loci (Table S6).

For population genetics tests, hispanicus_Gal(Gal) were compared to hispanicus_AM(AM) and, since P. hispanicus individuals carrying the Val introgressed mitotype are part of the hispanicus_Gal evolutionary lineage (Fig. 4), hispanicus_Gal(Val) were compared to hispanicus_Gal(Gal) to minimize background differentiation predating the Val mitotype introgression (Table 2). We did not find any significant difference between the nucOXPHOS loci and nucControl loci in Fst values (Table S5) and we did not detect any significant interaction between gene sets (nucOXPHOS or nucControl) and group for the pi value comparisons, suggesting an absence of co-introgression of nucOXPHOS genes together with introgressed mtDNA (Table S7).

Finally, as the donor of the Gal mitotype in P. hispanicus is supposed to be the sympatric lineage of P. liolepis, we compared hispanicus_Gal(Val) and liolepis_sierras(Val) for Fst. No differences were found in Fst between the nucOXPHOS and nucControl gene sets (Table S5).

To conclude, the test based on branch length detected a significant signal of preferential co-introgression of nucOXPHOS genes in P. hispanicus for both introgression events, but all other tests failed to detect any significant difference between the nucControl and nucOXPHOS genes.

Discussion

Characterization of Four Introgression Events

By comparing nucControl and mitochondrial phylogenies (Fig. 1), we were able to detect four cases of mtDNA introgression in the Podarcis Iberian group and inferred the most likely ancestral mitotypes of each species, as well as the origins of introgressed mtDNA either from ghost (extinct) lineages or extant species. Two of these introgressions (of the Val mitotype in P. liolepis and in P. hispanicus) had been reported previously, while two (Gal in P. hispanicus and carb in P. vaucheri) are new and reinforce the relevance of the Podarcis Iberian group as a model for comparative analyses of mitochondrial introgressions. Even if we acknowledge that a strict validation of this conclusion would require explicit tests of alternative scenarios using simulation-based approaches, we feel confident that the topological discordances that we uncovered represent instances of “true” discordance (= cytonuclear dissonance sensu Larson et al. 2026) due to introgression rather than mere products of the random nature of lineage sorting (see Funk and Omland 2003), because of the very different positions of the relevant lineages in the nuclear and mitochondrial trees, and sometimes of the very close relationships between mitochondrial haplotypes sampled in different species.

Previous work on the Podarcis Iberian group had already reported two non-monophyletic mitotypes in P. liolepis and interpreted this pattern as the result of past introgression (Renoult et al. 2009). The ancestral (native) mitotype was suggested by Renoult et al. to be the one occurring in the north of the range of the species (the liolepis = lio mitotype) because (i) the lio mitotype occurred in most of the species’ range and (ii) the second P. liolepis mitotype (Valencia = Val) was shared with another species, P. hispanicus. Our multilocus nuclear phylogeny confirms this conclusion: the lio mitotype has a placement in the mtDNA tree of the Iberian group that is fully concordant with the placement of P. liolepis in the nucControl phylogeny; the Val mitotype falls into a different section of the mtDNA tree and is thus confirmed as the introgressed mitotype by our phylogenies. Our sampling of P. liolepis is more extensive geographically than in Renoult et al., we confirmed that the Val mitotype is fixed in the south of the distribution of the species (see below) and identified a narrow band across the east of Spain where both Val and lio mitotypes co-occur, sometimes in the same populations (Fig. 3c). However, our sampling is not dense enough within localities to assess whether this contact zone between the two mitotypes is a mosaic of populations where one or the other mitotype is usually fixed or is mostly composed of polymorphic populations.

Using similar topological arguments from the comparisons of the mtDNA and nuclear DNA trees, we can also confirm that the occurrence of the Val mitotype in P. hispanicus results from past introgression, as already suggested by Renoult et al. (2009). We uncovered a larger distribution of the Val mitotype and, as for P. liolepis, identified one population where the Val mitotype coexists with another mitotype (Gal). To our knowledge, this is the first clear identification of the Gal mitotype as the result of another introgression into P. hispanicus; the results of Yang et al. (2021) already suggested this, but their limited sampling (one individual per mitotype) impedes conclusions. It therefore appears that the most likely ancestral mitotype of P. hispanicus is AM. This question could further be investigated with genome sequences using NUMTs (Nuclear Mitochondrial DNA Sequences), DNA that was originally from the mitochondria and was copied and pasted in the nucleus. NUMTs could provide information about the identity of previous mitochondria, before mtDNA introgression, for each species (Baião et al. 2023). For instance, if a NUMT is shared among all P. hispanicus lineages and if this NUMT sequence groups with one mitotype in a phylogeny, then this mitotype is the ancestral one and the others are introgressed.

Finally, we found that the carb mitotype (the native mitotype found in P. carbonelli) has unidirectionally introgressed into the P. vaucheri nuclear background in their syntopic population in southern Spain (Matalascañas). The Matalascañas population had already been identified as a hybrid zone between both species (Caeiro-Dias et al. 2021a), but the patterns of mtDNA ancestry had not been examined in this study. We found several individuals of mostly P. vaucheri ancestry carrying the carb mitotype but no instance where the vauch mitotype has introgressed into the P. carbonelli background. The carb mitotype occurrence into P. vaucheri is geographically restricted to the population where hybridization is currently taking place; this hybrid zone is supposed to be recent and linked to human-mediated habitat changes in the last decades (Caeiro-Dias et al. 2021a). This can thus be safely interpreted as a recent introgression.

The histories of the other introgression events are more difficult to interpret. Remarkably, both the Val mitotype found in P. liolepis and the Gal mitotype found in P. hispanicus seem to represent introgression from ghost lineages, as no currently known species can be identified as the source of these mitotypes (Fig. 1). We cannot exclude that a relict population of an undescribed species awaits discovery on the top of a southern sierra, but this seems increasingly unlikely. However, the occurrence of the Val mitotype in P. hispanicus is most likely an example of “secondary” introgression. In Renoult et al. (2009), the Val mitotype carried by both P. hispanicus and P. liolepis was divided into two monophyletic groups. The present wider sampling clearly reveals that the Val mitotypes carried by P. hispanicus are embedded inside P. liolepis Val mitotypes (Fig. S2). This phylogenetic placement favors the hypothesis of a secondary transmission of the Val mitotype from P. liolepis to P. hispanicus, rather than two independent introgression events from the ghost taxon to each species. The P. hispanicus individuals carrying the Val mitotype do not form a monophyletic group in the nuclear tree (Fig. 4a), but we do not know if this results from several introgression events or from nuclear gene flow post-introgression.

The number and timing of independent introgression events of the Val and Gal mitotypes into P. liolepis and P. hispanicus are more difficult to establish. (i) One possible explanation is that the Gal and Val mitotypes originate from at least two recent independent introgression events involving two ghost lineages (two extinct species). Under this scenario, the most parsimonious explanation for the Gal occurrence in P. hispanicus is a single introgression event, possibly during a protracted period of admixture and involving multiple haplotypes. For P. liolepis, the Val mitotype is present in three independent evolutionary lineages (the P. liolepis northern, liolepis_south and liolepis_sierras lineages, see Results) so the history of introgression is more complex. Whole-genome sequences would be better suited to determine the relationships between these three evolutionary lineages and assess whether the Val mitotype entered the southern P. liolepis populations once or (at least) twice. (ii) An alternative scenario is suggested by the mitochondrial phylogeny, which shows that the P. liolepis introgressed Val mitotype is sister to the ancestral AM mitotype of P. hispanicus, while the ancestral lio mitotype is sister to the introgressed Gal mitotype of P. hispanicus (Fig. 1). This mirror-like topology between introgressed and ancestral mitotypes could result from an ancient bidirectional introgression event between the ancestor of P. hispanicus and the ancestor of P. liolepis, followed by divergence between the introgressed mtDNA lineages and the ones that remained in their original species. This is in line also with the geographic distribution of introgressed mtDNA haplotypes: the Val mtDNA lineage occurs in the south of the distribution of P. liolepis and hence next to the current distribution of P. hispanicus, while the non-native Gal mitotype in P. hispanicus is found in areas where P. liolepis currently persists as isolated high-altitude populations in mountain tops, a geographical pattern suggestive of a more widespread distribution in this region during the recent past (last glacial period).

Under the first scenario, the Val and Gal mtDNA lineages originate from ancient speciation events followed by more recent introgression and subsequent extinction of the two donor (ghost) species. Under the second scenario, there is no extinct species, the Gal and Val lineages originated directly from divergence after an ancient bidirectional introgression and the ghost lineages are the ancestors of P. liolepis and of P. hispanicus. It is difficult with our data to disentangle these two hypotheses. The “single bidirectional introgression event” hypothesis might seem more parsimonious as it does not involve two extinction events, but it requires the long co-existence of two distinct mtDNA lineages in both P. hispanicus and P. liolepis through multiple glacial cycles, a rather unlikely event in itself. To resolve these questions, whole-genome sequencing of these populations would allow formal tests of alternative hypotheses (two ghost introgressions or bidirectional introgressions between the ancestors of the two species). Methods such as full-likelihood (eg Flouri et al. 2020; Pang and Zhang 2024) or demographic modeling approaches (eg Csilléry et al. 2012; Excoffier et al. 2021) have already proved effective in identifying ghost introgression events in nuclear genomes that are similar to the discordances observed in the mitochondrial phylogeny (eg Kato et al. 2024; Shen et al. 2025).

Whether these mitochondrial introgression events were driven by positive selection on introgressed mtDNA or were the result of neutral processes remains an elusive question. As demonstrated by Bonnet et al. (2017) and Seixas et al. (2018), distinguishing selective and demographic explanations for mitochondrial introgression requires extensive genomic data to estimate precisely the fraction of the nuclear genome affected by introgression and simulations to explore the demographic parameters influencing genomic and mitochondrial patterns of introgression. The data generated in this work do not allow us to delve into this issue.

Co-Introgression of nucOXPHOS Genes With mtDNA

Our tests based on individual branch lengths uncovered a clear signal of preferential co-introgression of OXPHOS genes, with shorter branches in the nucOXPHOS tree for introgressed individuals, in three of the four instances of mitochondrial introgressions (Val in P. liolepis, Gal in P. hispanicus, and Val in P. hispanicus). The co-introgression is not massive: if co-introgression of nucOXPHOS genes with the mtDNA had occurred for all or most nucOXPHOS genes, we would observe concordant mtDNA and nucOXPHOS phylogenies; on the contrary, the three species with mitochondrial introgression were still recovered as monophyletic clades in the nucOXPHOS tree (Fig. 2b), irrespective of the mitotype. Grouping OXPHOS genes per OXPHOS complex did not affect this conclusion: trees built per OXPHOS complex often did not recover species as monophyletic, but they were generally poorly supported, as a result of the smaller number of genes per complex (Appendix 1; see also Table S1).

Within-species topologies, on the other hand, are visibly different for the nucOXPHOS and nucControl trees, with less balanced trees for the nucOXPHOS trees, a signal captured by the branch length tests. By unbalanced topologies, we mean cases where internal nodes split into two lineages, one of which carries only one individual, giving rise to staircase-like topologies (Fig. 2b). This effect is visible here: species with mitochondrial introgression events have staircase-like topologies in the nucOXPHOS phylogeny but not in the nucControl phylogeny, while species without mitochondrial introgressions have no such patterns (Fig. 2b). Admixed individuals are known to create ladder-like topologies in phylogenomics (Pyron et al. 2022; Gippner et al. 2024), which suggests that nucOXPHOS genes are enriched in alien sequences. This staircase-like topology is observed in the nucOXPHOS phylogeny of liolepis(Val) individuals (Fig. S5b), as expected since these lineages carry introgressed mitotypes. The significant effects of our branch length tests are not simply a consequence of different evolutionary rates (and hence different branch lengths) for a set of markers compared to the other one: our test is based on a significant interaction between set of markers (nucOXPHOS compared to nucControl) and set of individuals (introgressed compared to non-introgressed individuals). A significant result means that branches are significantly shorter in the trees for the introgressed group and for the nucOXPHOS genes, not for the introgressed group overall nor for the nucOXPHOS genes overall. Differences in evolutionary rates between nucControl and nucOXPHOS would affect both groups of individuals (introgressed and non-introgressed) equally and would not be picked-up by our test.

Contrary to our expectations, staircase-like topologies are also observed in hispanicus_AM(AM), which carry the ancestral mitotype, a pattern opposite to expectation under co-introgression of nucOXPHOS genes with mtDNA (Fig. S5b). These patterns in hispanicus_AM(AM) may be due to admixture with surrounding lineages, independent of mitochondrial introgression. Indeed, unpublished RADseq data generated for another study suggests extensive admixture with hispanicus_Gal in a large part of the range of hispanicus_AM (G. Caeiro-Dias, pers. com.).

Given the signals of preferential nucOXPHOS co-introgression using topologies and branch lengths, why did the other tests (p-distance, Fst, qCF and pi, see Table 2) fail to capture any signal of co-introgression? Again, analyses per OXPHOS complex did not reveal significant signal of co-introgression in any species for any complex either (Appendix 1) and Fig. S7 did not reveal more outliers for nucOXPHOS than nucControl genes. As noted earlier, each species still formed monophyletic clades in nucOXPHOS trees, so the qCFs logically did not differ between both datasets; this test would have only been able to detect more substantial co-introgression resulting in non-monophyly of species in nucOXPHOS trees. Our test based on nucleotide diversity (pi values) would only be significant if native and introgressed nuclear alleles were still segregating within individuals with introgressed mitotypes. In our data, nucOXPHOS genes always had a higher diversity than nucControl genes (Fig. S6) but that difference was not significantly more pronounced for groups carrying introgressed mitotypes, suggesting either an absence of co-introgression or a fixation of introgressed nucOXPHOS alleles (Table S5). The sensitivity of our Fst tests is obviously constrained by the amount of genetic structure existing independently of the mitochondrial introgression. For example, there is ample evidence that the evolutionary lineages carrying introgressed or native mitotypes (liolepis_south or liolepis_sierras in P. liolepis, hispanicus_Gal in P. hispanicus) have diverged long ago from the P. liolepis northern or hispanicus_AM lineages. The background genomic Fst between the lineages carrying the native and introgressed mitotypes are thus probably already very high for most nuclear loci independently of introgression. Introgression of highly differentiated alleles (as postulated in cases of co-introgression with alien mitotypes) can be expected to increase the number of substitutions between these lineages but not necessarily the mean differences in allelic frequencies at SNPs between them. In this situation, substitution of native alleles by introgressed alleles is not expected to increase Fst values substantially, reducing the power of this test. Other comparisons where the two groups are not as strongly differentiated [such as vaucheri(vauch) versus vaucheri(carb)] are undermined by small sample sizes of each group.

Similarly, the p-distance test implemented here is only a good metric if introgression is recent and if the donor (or a closely related) population has been sampled. At least one of these two premises is not fulfilled in P. liolepis and P. hispanicus, but the test is adequate for P. vaucheri where we can see recent mtDNA divergence between carb and vauch mitotypes present in P. vaucheri.

Our p-distance test aims at capturing the same signal as the branch length test, yet it did not yield consistent signals of preferential co-introgression. We hypothesize that this originates from a lower power of the permutation test we built to evaluate the significance of the differences in p-distance: this permutation approach compares the observed between-groups difference to the “mean” obtained by randomizing loci between gene sets whereas to test difference in branch length we made full use of the differences in branch length distribution in linear modeling. We note that, even if no single comparison is significant, seven out of eight comparisons result in the expected larger differences (in absolute value) in nucOXPHOS genes compared to nucControl genes.

One difficulty of our model is that most introgression events involve a ghost donor lineage, whose genome is not accessible. This precludes “simple” approaches such as quantifying the ancestry of the donor and the receiver for nucOXPHOS and nucControl genes in introgressed and non-introgressed individuals or populations, but also more complex models of gene flow estimations (IM models among others) that all require including the donor populations in the models. Similarly, modeling of introgression from a ghost population in an existing ABC framework (such as DIYABC, Collin et al. 2021) and estimating the proportion of admixture in the different datasets require having sampled the donor populations. Methods to detect outlier SNPs such as genotype–environment associations (where mitotype would be set as the explanatory environmental variable) do not require information on the donor but would require a larger number of individuals, especially for P. vaucheri and P. hispanicus introgression cases, and most importantly would be confounded by the highly structured populations in P. liolepis or P. hispanicus, where individuals carrying native or introgressed mitotypes belong to divergent lineages.

A simulation approach of ghost introgression events would allow us to assess the power of the various tests we used here to detect variations in introgression level between different sets of nuclear genes. The age of the introgression, the divergence between the donor and the receiver, the number of genes co-introgressed and the level of structuring within the receiver are likely to be determining factors for the power of these tests. An interesting point would be to test with simulations if indeed the branch length test is more sensitive to detect introgression than other tests we have used here. Although this simulation analysis is beyond the scope of this study, we encourage future investigation down this road for more informed tests of co-introgression events.

Coevolution Between the Mitochondrial and the nucOXPHOS Genes?

Although it was not the target of this study and we did not perform any formal test, we cannot help to notice some of the species relationships in the mitochondrial phylogeny are not consistent with the nucControl genes but are fully concordant with the nucOXPHOS tree (Figs. 1 and 2). For example, P. carbonelli and P. virescens are sister species (with a moderate support) in nucOXPHOS phylogeny, while P. carbonelli groups with P. liolepis (with very high support) in nucControl phylogeny. Similarly, P. vaucheri and P. sp. Maghreb are sister groups in the nucOXPHOS and mitochondrial trees (Fig. 2b and c), but this is not recovered in the nucControl tree (Fig. 1a). Similar patterns observed at deeper evolutionary time scales have been interpreted as the result of coevolution between nuclear and mitochondrial OXPHOS subunits (Formaggioni et al. 2022; Gonçalves et al. 2025; Wallnoefer et al. 2025). The discordance between the nucControl tree and the nucOXPHOS or mtDNA trees for the grouping of P. virescens and P. carbonelli suggests that an ancient mitochondrial introgression and nucOXPHOS co-introgression event might explain the close relationships of P. virescens and P. carbonelli in mtDNA and nucOXPHOS trees.

Another result from our work suggests coevolution as a response to mitochondrial introgression: in P. vaucheri, the branch length test was significant but with an effect opposite to the expectation of nucOXPHOS co-introgression (ie longer branches in the population where the carb mitotype has introgressed). A possible explanation for this pattern could be an accelerated evolutionary rate of the OXPHOS nuclear genes following the carb mitochondrial introgression. Indeed, nucOXPHOS gene sequences might have undergone rapid adaptation that compensates for the change of mitotype, in agreement with the mitonuclear compensation theory. Compensatory evolution of nucOXPHOS genes in response to change in mtDNA genes has been documented previously (Osada and Akashi 2012).

Taxonomic Consequences

This work provides the first genomic assessment of species and population relationships in the Podarcis Iberian group, shedding light on several pending issues for the distribution and systematics of this group.

First, we confirm genetically for the first time the occurrence of P. liolepis in several mountains (Sierra de Cazorla and Sierra Espuña) in the south-east of Spain where the species was not reported by Renoult et al. (2010) (but see Speybroeck et al. 2016). These populations are, on the basis of current knowledge, isolated from the main range of the species and constitute a distinct evolutionary unit [liolepis_sierra(Val) in gray in Fig. 3] that is sister to the northern lineage of the species. Another more divergent unit is formed by the P. liolepis populations from the south of the Levante region of Spain (south of Valencia). The divergence between this undescribed lineage and the rest of the species is lower than most species-level divergence in the Podarcis Iberian group, but an analysis of its contact zone with the northern lineage of P. liolepis would be necessary to settle its taxonomic rank.

Second, our nucControl phylogeny unambiguously places two samples from the Columbretes archipelago (DB5170 from La Foradada and DB5196 from El Montcolibre) within the variation of the northern P. liolepis lineage (Fig. S3). Populations from the Columbretes Islands have been suggested to warrant species status under the name Podarcis atratus by Castilla et al. (1998) on the basis of elevated mitochondrial divergence from the closest mainland populations. However, Castilla et al. used individuals from Valencia and Castellon carrying Val mitotypes as mainland reference populations while Columbretes specimens carry lio mitotypes (see Table S2 and Fig. S2); as a consequence, the species status of atratus has not been accepted to date (see Uetz et al. 2026). More recently, Yang et al. (2021) also recovered a deep genomic divergence between liolepis and atratus and implicitly granted the later specific status as P. atratus in their Fig. 2. We fail to explain the differences between our results and Yang et al.'s results, as their sample of atratus come from the same islet as one of our samples and their sample of liolepis originates from the distribution of the northern lineage of P. liolepis. Notwithstanding this discrepancy, our population-level genomic sampling should settle the issue of the species-level taxonomy of atratus and firmly allocate it to P. liolepis.

Third, we uncover a deep genomic divergence between the lineages carrying the vauch mitotype on the one hand, and the sp. Maghreb mitotype on the other hand in North Africa (Fig. S5). While the taxonomy of this group will require more detailed analyses, our results confirm that at least two species of the Podarcis Iberian group inhabit North Africa, as suggested by mitochondrial data.

Finally, Bassitta et al. (2020) described populations of P. hispanicus carrying the Gal mitotype as a new species, P. galerai, on the basis of species-delimitation methods and phylogenetic arguments mostly based on mitochondrial data. In their trees that were mostly influenced by the mitochondrial topology, the Galera and AM evolutionary lineages were not recovered as a monophyletic group. Our genomic data confirms that individuals carrying the Gal and AM mitotypes unambiguously form a monophyletic group in all nuclear datasets, are not reciprocally monophyletic in the nucOXPHOS dataset and are more closely related than any valid species in the Podarcis Iberian group. This is in agreement with the results of Yang et al. (2021), who also recovered that P. hispanicus individuals carrying the Gal and AM mitotypes form a monophyletic clade. Our sampling, which is broader and more extensive, confirms that in spite of exhibiting diverged mitotypes, all sampled P. hispanicus specimens form a well-supported monophyletic clade in nuclear DNA (Fig. 1).

We thus argue here against the validity of the species described by Bassitta et al. (2020), because the initial analyses were misled by this unrecognized case of mitochondrial introgression, because the level of genomic divergence between galerai and hispanicus is lower than between all closely related valid species and because no arguments to date have been put forward to suggest reproductive isolation between these two taxa. The nucControl dataset still recovers two distinct evolutionary units in P. hispanicus and we suggest that the species can be treated as polytypic with the two subspecies P. hispanicus hispanicus and P. hispanicus galerai.

Conclusion

These results confirm the Podarcis Iberian group as an excellent model to study the genomic consequences of hybridization and the evolutionary forces shaping interspecific gene flow. The species, which diverged 2–7 million years ago (Yang et al. 2021), exhibit evidence of strong but incomplete reproductive isolation, with restricted hybridization in contact zones (Caeiro-Dias et al. 2021a, 2021b). Despite this, they have repeatedly exchanged mitochondria and nuclear genes throughout their history (Renoult et al. 2009; Gaczorek et al. 2023) and continue to do so across most contemporary contact zones studied so far. We confirmed here the pervasive nature of mitochondrial introgression by identifying four cases (two newly described) and suspecting a fifth involving P. virescens and P. carbonelli.

In all four confirmed cases, nucOXPHOS genes showed consistent branch length differences between populations carrying introgressed versus native mitotypes, whereas random nuclear loci did not three of these differences are consistent with preferential co-introgression of nucOXPHOS genes, while the last one suggests accelerated evolution of nucOXPHOS genes following mitochondrial introgression. Our results thus add to a growing body of evidence for mitonuclear coevolution and for increased introgression of mitochondrial-interacting nuclear genes relative to the genomic background (Farleigh et al. 2023; Jensen et al. 2023; Moran et al. 2024). A common feature of all previous studies that detected increased introgression in mitochondrial-interacting nuclear genes is that a small number of genes only seem to introgress more than expected, a result also apparent in our analysis where the signal of preferential co-introgression is consistent but weak.

Given the weak patterns of co-introgression that we observed in our data, the question of how an alien mitochondrion survives and functions in a foreign genetic background remains open. Several hypotheses can explain these results and weighing their relative contribution to the empirical patterns of mitochondrial and nuclear gene exchanges will require further testing. First, preferential co-introgression might affect a small number of genomic regions coding a few crucial functions, such as genic regions directly interacting with the mitochondria, that are necessary but also sufficient for efficient functioning of the respiratory chain. Second, rapid evolution of the native mitochondrial-interacting nuclear genes might alleviate the need for co-introgression of nucOXPHOS foreign alleles. Another alternative explanation as to how alien mtDNA can survive after introgression might be that they replace degenerated mtDNA that have accumulated deleterious mutations (Sloan et al. 2017). In such a scenario, no co-introgression nor compensatory evolution is necessary in nucOXPHOS genes, but foreign mtDNA would quickly reach fixation, explaining the relatively widespread occurrence of mtDNA replacements. Finally, mitochondrial introgression might often be selectively neutral.

Materials and Methods

Sampling

Samples were obtained from the tissue collections of CIBIO-InBio in Vairão, Portugal and the CEFE CNRS—EPHE collection of reptiles and amphibians of the Biogeography and Ecology of the Vertebrates team in Montpellier, France (BEV collection). The dataset includes representatives of every recognized or suspected species of the Iberian group: 17 P. bocagei, 22 P. carbonelli, 4 P. guadarramae, 44 P. hispanicus, 123 P. liolepis, 4 P. lusitanicus, 44 P. vaucheri, 6 P. virescens, and 15 Podarcis sp. Maghreb from two undescribed lineages (candidate species). Moreover, two Teira dugesii, two P. tiliguerta, and two P. muralis were included as outgroups for phylogenetic analysis, making a total of 288 individuals. The individuals’ location is shown in Fig. S1; sample names and more detailed information are available in Table S2.

Probe Design, DNA Extraction, and Sequencing

The probes for the target capture sequencing were designed from a transcriptome based on three individuals belonging to three different species of the Iberian group (one P. liolepis, one P. hispanicus, one P. virescens, see Table S8). Raw reads were generated by Chiari et al. (2012) and assembled using GS De Novo assembler (454 sequencing, Roche). The transcriptome was annotated with a BLAST search against the OXPHOS pathway of Anolis carolinensis in the KEGG database (Kanehisa et al. 2025) and further annotated using the NCBI non-redundant database (Sayers et al. 2024). To ensure correct bait design, we delineated exon boundaries, as probes spanning multiple exons would not allow proper capture. This was achieved using genome annotations from Anolis carolinensis, Gallus gallus and Homo sapiens (when Anolis data were insufficient), as well as the preliminary P. muralis genome. Various BLAST tools (blastn, blastp, tblastn; Altschul et al. 1990) were used, with additional manual alignment inspection (Appendix 2). In total, 240 nuclear genes and 13 mitochondrial genes were targeted (Table S1). Nuclear genes that were not part of the OXPHOS genes but were expressed in the mitochondria were excluded (Table S1).

DNA from each sample was extracted using a standard salt extraction protocol (Sambrook et al. 1989) modified with an overnight PBS wash of the samples before starting the protocol. The target capture enrichment and sequencing were done by RAPiD Genomics LLC company (Florida, USA) to obtain Illumina paired-end 101-bases reads.

Reads Processing

Nuclear Datasets

Illumina reads quality was checked via FastQC v0.11.9 (Andrews 2010), then the reads were trimmed via TrimGalore v0.6.6 (Krueger 2015) with options -q 25 –clip_R1 5 –clip _R2 5 to remove bad quality bases located in read ends. The reads were then mapped with bwa-mem v07.17 (Li 2013) with option -B 2 (less stringent than the default parameter because it is a multispecies dataset) and the P. muralis reference genome produced by Andrade et al. (2019). The reads were sorted with samtools v1.17 (Danecek et al. 2021) and the non-matched or multiple matched reads were eliminated thanks to samtools view -b -F 0 × 4 -f 0 × 2. The duplicates were removed with Picard v3.0.0 (http://picard.sourceforge.net/) –REMOVE_DULICATES true option. The reads were sorted again and indexed with samtools.

Variant calling was done with bcftools v1.17 (Danecek et al. 2021) mpileup option -d 1000 (for loci with more than 1000 reads only 1,000 are randomly selected for base calling) and -R to include an interval file. The interval file provided for mpileup was produced thanks to blastall (-p blastn -e 1e-10) of the probes designed for capture target sequencing against the reference genome. This interval file was then manually curated to avoid interval overlaps, because some probes covered the same positions (Appendix 3). We then used bcftools call option -m (adapted for multiallelic and rare-variant calling). In the VCF produced, the variants were normalized with bcftools norm (options -m + any to join biallelic sites into multiallelic records). Variants were filtered thanks to bcftools filter -S. (to set genotypes of failed samples to missing value) and a custom -e expression to remove variants with allelic depth below 5 and heterozygous sites with allelic balance exceeding 1/10. Four individuals were removed because they had over 60% of missing SNP data, resulting in a dataset of 284 individuals with an average of 11% missing SNP data (ranging from 0.8% to 44%, with one outlier at 58%). Finally, FASTA sequences were recovered from the VCF file using vcf2phylip v2.0 (Ortiz 2019).

In order to filter out loci that might be contaminated by paralog read mapping, we ruled out loci for which more than 80% of the individuals in the dataset shared heterozygous positions, leading to a final nuclear dataset of 67 nucOXPHOS loci and 144 nucControl loci (Table S1). The average read depth per individual was 39.4 (ranging from 13.2 to 94.5, with an outlier at 3.5). No statistical phasing was attempted as all downstream analyses of nuclear data are based on consensus, unphased alignments using IUPAC ambiguity codes for heterozygous positions.

Mitochondrial Dataset

Cleaned reads were processed with MitoFinder v1.4.2 (Allio et al. 2020) using megahit as assembler, the mitogenome of P. muralis (NCBI: NC_011607.1) as reference, –min-contig-size 10000 to retain only quite large scaffold and avoid NUMT and -o 2 as we are assembling vertebrate mtDNA. A scaffold of more than 10,000 bp was retrieved for 284 individuals (out of 288) and their gene sequences were extracted from the MitoFinder output. Sequences were aligned using the alignSequences program in MACSE v2.07 (Ranwez et al. 2018) and then concatenated using FASTCONCAT v1.11 (https://github.com/PatrickKueck/FASconCAT).

Phylogenetic Inferences

The phylogenetic trees were built in a maximum likelihood framework using IQ-TREE v2.2.0 (Minh et al. 2020). The mitochondrial dataset of the 13 protein-coding gene was partitioned by gene and codon position (giving 39 partitions), allowing each partition to have its own branch length (Chernomor et al. 2016). Evolutionary models were chosen using ModelFinder (Kalyaanamoorthy et al. 2017), and with merging of partitions to find the best scheme (-m MFP + MERGE). Node support was assessed using 1000 replicates for both ultrafast bootstraps (-b, UFBoot; Hoang et al. 2018) and SH-like approximate likelihood ratio tests (-alrt, SH-aLRT; Guindon et al. 2010). Additionally, the following parameters were used: -nstop 500 -allnni –sampling GENE.

The nuclear trees were built with the whole concatenated nuclear dataset, the concatenated nucOXPHOS dataset and the concatenated nucControl dataset, all partitioned by loci, leading to three concatenated nuclear trees with the following parameters: -m MFP + MERGE -b 1000 and -alrt 1000. Visualization was done with ggtree v3.10.1 (Yu et al. 2017) in R v4.3.2 (R Core Team 2023).

The topological differences were assessed by comparing the likelihoods of nuclear and mitochondrial topologies according to the mitochondrial genome via the approximately unbiased test with 10,000 replicates using the RELL method in IQ-TREE (Shimodaira 2002). The same procedure was performed for nucControl and nucOXPHOS topologies with regard to the nucControl genetic information.

Inference of Introgression Events

We identified mitochondrial introgression events by comparing topologies of nucControl tree versus mitochondrial tree. The ancestral mitotype was inferred as the one mirroring the nuclear phylogeny the best and the introgressed mitotype was defined as the one having the strongest discordance with the nuclear phylogeny.

Definition of Groups of Individuals

For each species showing introgression events, we defined pairs of closely related groups, one with individuals carrying the ancestral mtDNA, the other with individuals carrying the introgressed mtDNA (see Results). Intraspecific population structure and geography were considered to form paired groups of individuals that were as closely related as possible yet carried ancestral versus introgressed mtDNA. The groups are described in the Results section. We then performed a series of tests, as detailed below, on all pairs to test for co-introgression of nucOXPHOS genes with the mitochondrial introgression.

Tests of Preferential Co-Introgression of the Nuclear OXPHOS Loci

Since most of the mtDNA introgression events that we identify here involve a ghost (presumably extinct) species as the donor lineage, we have no access to the nuclear genome of the donor lineages. We thus cannot directly quantify donor ancestry in nucOXPHOS genes and nucControl genes for populations with introgressed mtDNA, which would be a direct test of the preferential co-introgression of OXPHOS genes when mtDNA is introgressed. However, several predictions can be made for comparisons between the nucOXPHOS and nucControl gene datasets (summarized in Table 1).

If preferential co-introgression of nucOXPHOS occurred in the mtDNA-introgressed populations, we would expect that: (i) nucOXPHOS genes should have a discordant phylogenetic history compared with nucControl loci, and either concordant with the mtDNA phylogeny or influenced by the mtDNA phylogeny, and nucOXPHOS loci should have more discordant phylogenies compared to the species phylogeny, as assessed by qCFs, than nucControl genes for a species containing individuals with ancestral and introgressed mitotype; (ii) alternatively, if nucOXPHOS topology is concordant with the nucControl phylogeny, co-introgression could still be detected using branch lengths, which should be shorter for introgressed individuals (from individuals tips to the internal node of the clade of interest) in the nucOXPHOS phylogeny than in the nucControl phylogeny, because admixture tends to shorten branch lengths leading to admixed individuals or populations (Kopelman et al. 2013; Peter 2016; Forsythe et al. 2020); (iii) genetic differentiation (measured by Fst or p-distance) between groups carrying the ancestral versus introgressed mtDNA should be higher in nucOXPHOS genes than in nucControl genes; (iv) nucleotide diversity should be different for nucOXPHOS loci in introgressed individuals compared to individuals carrying the ancestral mtDNA, but no differences should be observed for nucControl loci (note that diversity can be lower in the case of adaptive introgression leading to fixation of introgressed alleles or higher if original and introgressed alleles segregate in introgressed populations).

The P-values for all the tests were adjusted for multiple testing using the Benjamini-Hochberg procedure with the p.adjust function in R (Benjamini and Hochberg 1995).

Quartet Concordance Factor

The qCF provides a way to assess discordance and evaluate support for relationships among four taxa (A, B, C, and D) in a phylogenetic tree; qCF is the proportion of gene trees supporting each of the three possible sister taxa of group A (B, C, or D), analyzing all the inter-species quartets possible for every gene trees. By analyzing qCFs across the genome, it is possible to understand the extent of phylogenetic discordance among loci (Lanfear and Hahn 2024). In our case, we tested whether individuals carrying the introgressed mitotype (taxa A) grouped more often with individuals from the same species carrying the ancestral mitotype (taxa B), or if co-introgression of nucOXPHOS genes led taxa A to be closer to individuals from other species (taxa C and D). For nucControl genes, we expect taxa A to be closest to taxa B.

Here, we compare qCFs between nucOXPHOS and nucControl genes for each internal branch of the tree, in order to test whether nucOXPHOS genes have more discordance than nucControl genes compared to the reference tree (species tree), as expected if nucOXPHOS genes co-introgressed with the mtDNA (Table 1). If nucOXPHOS genes preferentially co-introgress with the mtDNA, we expect more topological discordance compared to the species tree for the nucOXPHOS than for the nucControl data set.

qCFs were computed using ASTRAL v5.7.8 (Zhang et al. 2018; Rabiee et al. 2019) with the -t 2 option. The reference tree was built in ASTRAL by combining nucOXPHOS and nucControl tree sets, using the same principles as when building a species tree, except that in our case the clades are not always species. For each introgression event, a new reference tree was computed where we only split the species of interest as introgressed and non-introgressed and the other species were left monophyletic. We obtained two qCFs for each branch, corresponding to the nucOXPHOS and nucControl gene trees (the loci trees were built using IQ-TREE with the default parameters). For each iteration of ASTRAL, one individual is randomly selected among individuals carrying the introgressed mitotype, a second individual is randomly drawn among individuals carrying the ancestral mitotype, and two individuals belonging to two other species are randomly drawn to form a quartet. To avoid biases, in some cases individuals had to be removed (see Results). To evaluate if the observed difference in qCFs for a specific branch between the two gene sets was statistically significant, a resampling approach was used. Gene trees were randomly reassigned to two sets of the same size as the original nucOXPHOS and nucControl sets, and qCFs were recalculated for these randomized sets. This process was repeated 10,000 times to create a null distribution of qCF differences for the branch of interest. The observed qCF difference was then compared to this null distribution to calculate a P-value.

Branch Length Test

We compared the mean branch lengths (tip to crown node) between introgressed and non-introgressed individuals. If nucOXPHOS genes co-introgressed with the mtDNA, we expect shorter branch lengths in nucOXPHOS (concatenated dataset) compared with nucControl genes in introgressed individuals, due to attraction toward alien alleles (Kopelman et al. 2013).

The patristic distances (hereafter branch lengths) were calculated for each individual from tip to crown node of the species (P. liolepis, P. hispanicus, or P. vaucheri) thanks to the tree_subset in the treeio package v1.30.0 (Wang et al. 2020) and distRoot in adephylo package v1.1-16 (Jombart et al. 2010) functions in R. Each individual has two branch length values: one for the nucOXPHOS tree and one for the nucControl tree. Before statistical analysis, the distances were centered and scaled to avoid issues of dispersion or deviation in LMMs. LMMs were then used to analyze whether branch lengths differed according to gene set (nucOXPHOS or nucControl), mitotype (ancestral or introgressed), or the interaction between these two fixed effects. The identity of the individual was set as a random intercept factor. We used the lmer function in the lme4 package v1.1-36 (Bates et al. 2015) with the syntax: branch length ∼ gene set * mitotype + (1|individual). The significance of the effects was assessed using lmerTest v3.1-3 (Kuznetsova et al. 2017) with the P-values adjusted for each effect category.

Genetic Differentiation

Genetic differentiation between individuals with ancestral mtDNA and those with introgressed mtDNA is expected to be greater at nucOXPHOS loci than at nucControl loci in case of co-introgression. For each locus, the average per SNP Fst between the introgressed and non-introgressed groups was computed with pixy v1.2.7.beta1 (Korunes and Samuk 2021) with option –fst_type wc (Weir and Cockerham's estimator). Permutation tests (10,000 random permutations of loci between data sets) were conducted in R using custom script and the sample function to determine if there was any significant difference in Fst between the nucOXPHOS and the nucControl loci for a pair of populations by comparing the mean of the observed Fst with the distribution obtained from the randomized datasets.

We also estimated the raw between-groups genetic distances (p-distance) per gene set (concatenating nucOXPHOS and nucControl loci separately) with the dist.dna function in the ape package v5.8-1 (Paradis and Schliep 2019). P-distances between individuals were computed and then averaged per group to obtain p-distances between groups for each gene set. A delta of p-distances for nucOXPHOS loci and for nucControl separately were then calculated as follows:

$$\mathrm{Delta}=\mathrm{P}\text{-}\mathrm{distanc}{\mathrm{e}}_{(\mathrm{pop}1\text{-}\mathrm{pop}3)}\text{-}\mathrm{P}\text{-}\mathrm{distanc}{\mathrm{e}}_{(\mathrm{pop}2\text{-}\mathrm{pop}3)}$$ (1)

where pop1 and pop2 are the groups of the species of interest, either carrying the introgressed (pop1) or the ancestral (pop2) mitotype. The pop3 group is either the donor species of the introgressed mitotype when it is known, or a species related to the ghost donor (when the donor is extinct) as approximated from the mitochondrial phylogeny. To assess if nucOXPHOS delta and nucControl delta differ significantly, we did 1,000 permutations of the loci to reassign them to two data sets of the same sample size as the nucOXPHOS and nucControl sets (ie 67 and 144 loci respectively) and then calculated the delta for each permuted dataset. The distribution of the 1,000 permutations was compared to the observed difference to calculate a P-value.

Nucleotide Diversity

We tested whether nucleotide diversity (pi values) differed between introgressed and non-introgressed individuals, and whether there were any differences between nucOXPHOS and nucControl genes. We used LMMs to explain the variance of pi. If nucOXPHOS alleles co-introgressed and were strongly selected, we expect a lower pi (negative interaction effect in LMM) for the introgressed group. If a mix of alien and ancestral alleles segregate in the introgressed group, we expect higher pi compared to the group carrying the ancestral mitotype for the nucOXPHOS loci (positive interaction in LMM effect).

The pi values were computed with pixy per locus with default parameters. Pi values were compared between population pairs using LMMs, with gene set (nucOXPHOS or nucControl), group (introgressed or not introgressed) and their interaction as fixed effects and locus as random intercept effect. We used the lmer function in the lme4 package with the syntax: pi ∼ gene set * group + (1|locus). The significance of the effects was assessed using lmerTest with the P-values adjusted for each effect category. The loci located on the Z sex chromosome (three nucOXPHOS and five nucControl loci) were removed from the pi dataset before running LMMs to avoid biases in pi values due to differences in Z chromosome effective population size.

Supplementary Material

Supplementary material is available at Genome Biology and Evolution online.

Funding

Support was provided by national funds through Fundação para a Ciência e a Tecnologia (EXPL/BIA-EVF/1283/2012) to CP. ML was supported by a PhD scholarship from the French Ministère de l’Enseignement Supérieur et de la Recherche through GAIA doctoral school. This project was supported by the Agence Nationale de la Recherche (JCJC grant IMPRINT ANR-23-CE20-0045) to AM.

Data Availability

The raw reads have been deposited to European Nucleotide Archive (ENA) and are available at study accession number PRJEB122088. The data underlying this article are available in Zenodo, at https://dx.doi.org/10.5281/zenodo.20414855.