Abstract
The evolution of reproductive barriers is the first step in the formation of new species and can help us understand the diversification of life on Earth. These reproductive barriers often take the form of hybrid incompatibilities, in which alleles derived from two different species no longer interact properly in hybrids1,2,3. Theory predicts that hybrid incompatibilities may be more likely to arise at rapidly evolving genes4,5,6 and that incompatibilities involving multiple genes should be common7,8, but there has been sparse empirical data to evaluate these predictions. Here we describe a mitonuclear incompatibility involving three genes whose protein products are in physical contact within respiratory complex I of naturally hybridizing swordtail fish species. Individuals homozygous for mismatched protein combinations do not complete embryonic development or die as juveniles, whereas those heterozygous for the incompatibility have reduced complex I function and unbalanced representation of parental alleles in the mitochondrial proteome. We find that the effects of different genetic interactions on survival are non-additive, highlighting subtle complexity in the genetic architecture of hybrid incompatibilities. Finally, we document the evolutionary history of the genes involved, showing signals of accelerated evolution and evidence that an incompatibility has been transferred between species via hybridization.
Similar content being viewed by others
Main
Biologists have long been fascinated by the question of how new species are formed and the mechanisms that maintain isolation between them. One key factor in the formation and maintenance of new species is the emergence of genetic incompatibilities that reduce viability or fertility in hybrids1. As originally described by the Dobzhansky–Muller model of hybrid incompatibility2,3, the unique sets of mutations that accumulate in diverging species may interact poorly when they are brought together in hybrids, given that they have never been tested against one another by selection. Owing to the technical challenges of identifying these interactions4, only around a dozen genes involved in hybrid incompatibilities have been precisely mapped5 and exploration of the functional and evolutionary causes of hybrid incompatibilities has been limited to a small number of cases in model organisms4.
This knowledge gap leaves key predictions about the evolutionary processes that drive the emergence of hybrid incompatibilities untested. For one, theory suggests that incompatibilities should be more common within dense gene networks, both because genes involved in such interactions are expected to be tightly co-evolving and because the number of potentially incompatible genotypes explodes as the complexity of the genetic interaction increases7,8. Consistent with this prediction, mutagenesis experiments have highlighted the sensitivity of multi-protein interactions to changes in any of their components8. However, genetic interactions are notoriously difficult to detect empirically9, and this problem is exacerbated with complex genetic interactions10. Such technical challenges may explain the rarity of incompatibilities involving three or more genes in the empirical literature8 (but see refs. 9,11,12,13,14).
Another open question is the degree to which the genes that become involved in hybrid incompatibilities are predictable from their molecular or evolutionary properties. Researchers have proposed that rapid molecular evolution will increase the rate at which incompatibilities accumulate between species4,5,6. Although several known incompatibilities involve genes showing signatures of positive selection, it is unclear how unusual rates of protein evolution are in these genes relative to the genomic background5,6. The mitochondrial genome, in particular, has been proposed as a hotspot for the accumulation of genetic incompatibilities15,16, owing to substitution rates up to 25 times higher than the nuclear genome in many animals17,18 and the potential for sexually antagonistic selection driven by its predominantly maternal inheritance19,20, among other factors21. At the same time, nuclear and mitochondrial proteins must interact with each other in key steps of ATP synthesis, increasing the likelihood of coevolution between these genomes22,23. These factors suggest that interactions between mitochondrial- and nuclear-encoded proteins could have an outsized role in the emergence of hybrid incompatibilities15, consistent with results from numerous species24,25,26.
As we begin to identify the individual genes underlying hybrid incompatibilities, the next frontier is evaluating the processes that drive their evolution. Over the past two decades, it has become abundantly clear that hybridization is exceptionally common in species groups where it was once thought to be rare27,28. As a result, it is now appreciated how frequently species derive genes from their relatives29,30,31. The effects of historical hybridization on the evolution of hybrid incompatibilities have been poorly investigated32, since the foundational theory in this area was developed before the ubiquity of hybridization was fully appreciated7.
Here we use an integrative approach to precisely map the genetic basis and physiological effects of a lethal mitonuclear hybrid incompatibility in swordtail fish and uncover its evolutionary history. The sister species Xiphophorus birchmanni and Xiphophorus malinche began hybridizing approximately 100 generations ago in multiple river systems33 after premating barriers were disrupted by habitat disturbance34, and are a powerful system to study the emergence of hybrid incompatibilities in young species. Despite their recent divergence35 (around 250,000 generations; 0.5% divergence per basepair), some hybrids between X. birchmanni and X. malinche experience strong selection against incompatibilities35,36. One incompatibility that causes melanoma has been previously mapped in this system and population genetic patterns suggest that dozens may be segregating in natural hybrid populations35,36,37,38. Moreover, the ability to generate controlled crosses39,40 and the development of high-quality genomic resources38,41 makes this system particularly tractable for studying hybrid incompatibilities in natural populations. Leveraging data from controlled laboratory crosses and natural hybrid populations, we pinpoint two nuclear-encoded X. birchmanni genes that are lethal when mismatched with the X. malinche mitochondria in hybrids, explore the developmental and physiological effects of this incompatibility, and trace its evolutionary history.
Mapping mitonuclear incompatibilities
To identify loci under selection in X. birchmanni × X. malinche hybrids, we generated approximately 1× low-coverage whole-genome sequence data for 943 individuals from an F2 laboratory cross and 359 wild-caught hybrid adults, and applied a hidden Markov model to data at more than 600,000 ancestry-informative sites along the genome to infer local ancestry (approximately 1 informative site per kilobase37,42; Methods and Supplementary Information 1.1.1–1.1.4). Using these results, we found evidence for a previously unknown incompatibility between the nuclear genome of X. birchmanni and the mitochondrial genome of X. malinche (Supplementary Information 1.1.5–1.1.10). Our first direct evidence for this incompatibility came from controlled laboratory crosses (Methods and Supplementary Information 1.1.1). Because the cross is largely unsuccessful in the opposite direction, all laboratory-bred hybrids were the offspring of F1 hybrids generated between X. malinche females and X. birchmanni males and harboured a mitochondrial haplotype derived from the X. malinche parent species. Offspring of F1 intercrosses are expected to derive on average 50% of their genome from each parent species. This expectation is satisfied genome wide and locally along most chromosomes in F2 hybrids (on average 50.3% X. malinche ancestry; Supplementary Fig. 1). However, we detected six segregation distorters genome wide40, with the most extreme signals falling along a 6.5 Mb block of chromosome 13 and a 4.9 Mb block of chromosome 6 (Fig. 1a,d).
Closer examination of genotypes in the chromosome 13 region showed that almost none of the surviving individuals harboured homozygous X. birchmanni ancestry in a 3.75 Mb subregion (Fig. 1c and Supplementary Fig. 2; 0.1% observed versus 25% expected). This pattern is unexpected even in the case of a lethal incompatibility involving only nuclear loci (see Supplementary Information 1.1.1), but is consistent with a lethal mitonuclear incompatibility. Using approximate Bayesian computation (ABC) approaches we inferred the strength of selection against X. birchmanni ancestry in this region that was consistent with the observed genotypes and ancestry deviations. We estimated posterior distributions of selection and dominance coefficients and inferred that selection on this genotype in F2 is largely recessive and essentially lethal (maximum a posteriori estimate h = 0.12 and s = 0.996, 95% credible interval h = 0.010–0.194 and s = 0.986–0.999; Fig. 1b, Extended Data Fig. 1, Methods and Supplementary Information 1.2.1–1.2.2).
The degree of segregation distortion observed in F2 individuals on chromosome 6 is also surprising (Fig. 1d). Only 3% of individuals harbour homozygous X. birchmanni ancestry in this region (compared with 0.1% in the chromosome 13 region and 25% on average at other loci across the genome; Fig. 1f), which is again lower than expected for a nuclear–nuclear hybrid incompatibility (Supplementary Information 1.1.1). ABC approaches indicate that selection on homozygous X. birchmanni ancestry on chromosome 6 is also severe (maximum a posteriori estimate h = 0.09 and s = 0.91, 95% credible interval interval h = 0.01–0.21 and s = 0.87–0.94; Fig. 1e, Extended Data Fig. 1 and Supplementary Information 1.2.2). Thus, our F2 data show that homozygous X. birchmanni ancestry in regions of either chromosome 13 or chromosome 6 is almost completely lethal in hybrids with X. malinche mitochondria (Fig. 1h).
To formally test for the presence of a mitonuclear incompatibility involving chromosome 13 and chromosome 6, or elsewhere in the genome, we leveraged data from natural hybrid populations. Most naturally occurring X. birchmanni × X. malinche hybrid populations are fixed for mitochondrial haplotypes from one parental species (Supplementary Information 1.1.2 and 1.1.6). However, a few populations segregate for the mitochondrial genomes of both parental types, and we focused on one such population (the ‘Calnali low’ population, hereafter referred to as the admixture mapping population). Admixture mapping for associations between nuclear genotype and mitochondrial ancestry (after adjusting for expected covariance due to genome-wide ancestry36) revealed two genome-wide significant peaks and one peak that approached genome-wide significance (Fig. 1g and Supplementary Tables 1–3). The strongest peak of association spanned approximately 77 kb and fell within the region of chromosome 13 identified using F2 crosses (Fig. 1g). This peak was also replicated in another hybrid population (Methods, Supplementary Fig. 3 and Supplementary Information 1.1.5) and contains only three genes: the NADH dehydrogenase ubiquinone iron–sulfur protein 5 (ndufs5), E3 ubiquitin–protein ligase, and microtubule–actin cross-linking factor 1. Of these three genes, only ndufs5 forms a protein complex with mitochondrially encoded proteins, which along with other evidence implicates it as one of the interacting partners in the mitonuclear incompatibility (Fig. 1c, Extended Data Fig. 2 and Supplementary Fig. 4; see Supplementary Information 1.1.6–1.1.9).
We also identified a peak on chromosome 6 that approached genome-wide significance (Fig. 1g, Supplementary Fig. 5, Supplementary Table 2 and Supplementary Information 1.1.10) and fell precisely within the segregation distortion region previously mapped in F2 hybrids (Fig. 1d and Supplementary Information 1.1.1). This peak contained 20 genes, including the mitochondrial complex I gene ndufa13 (Extended Data Fig. 2, Supplementary Fig. 5, Supplementary Information 1.1.10 and Methods). Depletion of non-mitochondrial parent ancestry at ndufa13 was unidirectional (Fig. 1f), consistent with selection acting only against the combination of the X. malinche mitochondria with homozygous X. birchmanni ancestry at ndufa13 (see Supplementary Information 1.2.3–1.2.4). Genomic analyses in natural hybrid populations confirmed this asymmetry (Extended Data Fig. 2).
Together, these results indicate that at least two X. birchmanni nuclear genes cause incompatibility when they are mismatched in ancestry with the X. malinche mitochondria (Fig. 1h and Supplementary Information 1.2.5). These genes, ndufs5 and ndufa13, belong to a group of proteins and assembly factors that form respiratory complex I (ref. 43) (see Supplementary Table 1 for locations of the 51 annotated complex I genes in the Xiphophorus genome). Complex I is the first component of the mitochondrial electron transport chain that ultimately enables the cell to generate ATP. Both nuclear proteins interface with several mitochondrially derived proteins at the core of the complex I structure, hinting at the possibility that physical interactions could underlie this multi-gene mitonuclear incompatibility.
Interactions with X. birchmanni mitochondrial DNA
Admixture mapping analysis also identified a strong peak of mitonuclear association on chromosome 15, which we briefly discuss here and in Supplementary Information 1.1.10 and 1.2.1. This peak was associated with X. birchmanni mitochondrial ancestry (Extended Data Fig. 3), indicating that it has a distinct genetic architecture from the incompatibility involving the X. malinche mitochondria and X. birchmanni ndufs5 and ndufa13. Specifically, analysis of genotypes at the admixture mapping peak indicates that X. birchmanni mitochondrial ancestry is incompatible with homozygous X. malinche ancestry on chromosome 15 (Fig. 1c and Extended Data Fig. 3). This region did not contain any members of complex I, but dozens of genes in this interval interact with known mitonuclear genes (see Supplementary Table 3 and Supplementary Information 1.1.10). The fact that we detect incompatible interactions with both the X. malinche mitochondria (at ndufs5 and ndufa13) and the X. birchmanni mitochondria (ndufs5 and chromosome 15) in our admixture mapping results supports the idea that mitonuclear interactions can act as ‘hotspots’ for the evolution of hybrid incompatibilities15.
Lethal effects in early development
The combination of X. birchmanni ndufs5 or ndufa13 with the X. malinche mitochondria appears to be lethal by the time individuals reach adulthood. To investigate the developmental timing of the incompatibility, we genotyped pregnant females from the admixture mapping population and recorded the developmental stages of their embryos44 (swordtails are livebearing fish; Methods). We found a significant interaction between developmental stage and ndufs5 genotype, whereas ndufa13 genotype did not affect developmental stage (Fig. 2a,b, Supplementary Figs. 6–9 and Supplementary Information 1.3.1). Genotyping results revealed that embryos with homozygous X. birchmanni ancestry at ndufs5 and X. malinche mitochondria are present at early developmental stages, but that these embryos did not develop beyond a phenotype typical of the first seven days of gestation (the full length of gestation is 21–28 days in Xiphophorus; Fig. 2a,b,d,e). Individuals with mismatched ancestry at ndufs5 whose siblings were fully developed still had a detectable heartbeat but had consumed less yolk than their siblings and remained morphologically underdeveloped (Fig. 2d, Extended Data Fig. 4 and Supplementary Figs. 10–14). Unlike other species, in Xiphophorus this developmental lag could itself cause mortality, since embryos that do not complete embryonic development inside the mother do not survive more than a few days after birth (Supplementary Information 1.3.1 and Supplementary Table 4). Given that complex I inhibition lethally arrests development in zebrafish embryos45,46, we also tested the effects of complex I inhibition on X. birchmanni and X. malinche fry, and found a similar level of sensitivity (Supplementary Information 1.3.2).
In contrast to individuals with mismatched ancestry at ndufs5, those with ndufa13 mismatch survived embryonic development but suffered mortality in the early post-natal period (Fig. 2c). We tracked 74 F2 fry from 24 h post birth to adulthood (Supplementary Information 1.3.3). We found that most fry with incompatible genotypes at ndufa13 had already suffered mortality by the time tracking began, with only 7 individuals found 24 h post birth that were homozygous X. birchmanni at ndufa13 (versus 19 expected; binomial P = 0.0005). No natural mortality was observed between 1 day and 3 months post birth (Supplementary Information 1.3.3).
Physiology and complex fitness effects
Our analysis of developing embryos indicates that individuals with the ndufs5 incompatibility exhibited abnormal embryonic development, whereas those with the ndufa13 incompatibility did not. This suggests that these genes may drive lethality through partially distinct mechanisms. Thus, we chose to further investigate the effects of ndufs5 and ndufa13, and their possible interactions. We sampled 235 F2 embryos at a range of developmental stages and measured their overall rates of respiration (Supplementary Information 1.3.4–1.3.5). We also used imaging of these embryos to track cardiovascular phenotypes as these have been associated with ndufa13 defects in mammals47. We found that incompatible genotypes at ndufs5 and ndufa13 affected a range of phenotypes, including heart rate, length relative to compatible siblings, and length-corrected head size (Extended Data Figs. 4 and 5, Supplementary Figs. 11–17 and Supplementary Tables 5–7). ndufa13 mismatch has a large effect on cardiovascular phenotypes, including heart rate and the size of the sinu-atrium (an embryo-specific heart chamber; Fig. 2g,h, Supplementary Figs. 14 and 16 and Supplementary Tables 6 and 8), whereas ndufs5 affects only heart rate (Supplementary Fig. 14 and Supplementary Table 6). We find initial evidence that cardiac defects persist into adulthood in surviving individuals with ndufa13 mismatch (Extended Data Fig. 5 and Supplementary Information 1.3.3). By contrast, ndufs5 mismatch has a major effect on rates of respiration and yolk consumption during development (Fig. 2f, Supplementary Figs. 18–20 and Supplementary Tables 9–11).
Naively, the separable impacts of incompatible genotypes at ndufs5 and ndufa13 could indicate that even though these proteins are in physical contact in complex I (see below), they represent two distinct hybrid incompatibilities. We investigated this question by taking advantage of rare survivors of the ndufa13 incompatibility in an expanded dataset of 1,010 F2 hybrids. Using this dataset, we were able to identify dozens of survivors of the ndufa13 incompatibility (3.4% of individuals) and found that genotypes at ndufa13 and ndufs5 were not independent (χ2 association test P = 0.032; Supplementary Information 1.2.5). Upon further investigation, we found that the majority of survivors of the ndufa13 incompatibility had homozygous X. malinche ancestry at ndufs5, suggesting that harbouring even one X. birchmanni allele at ndufs5 may sensitize fry to the ndufa13 incompatibility. Indeed, we found that individuals that had heterozygous ancestry at ndufs5 were significantly under-represented among surviving ndufa13 incompatible individuals (Permutation test P = 0.015; Fig. 2i and Supplementary Information 1.2.5). These findings highlight a subtle but significant non-additive effect of ndufs5 and ndufa13 on survival.
Mitochondrial biology in heterozygotes
Because few individuals homozygous for incompatible genotypes at ndufs5 or ndufa13 survive past birth, our previous experiments focused on embryos. However, the small size of Xiphophorus embryos prevents us from using assays that directly target complex I. To further explore the effects of the hybrid incompatibility on complex I function in vivo, we turned to adult F1 hybrids (Fig. 3a). Since F1 hybrids that derive their mitochondria from X. malinche and are heterozygous for ancestry at ndufs5 and ndufa13 are fully viable, we tested whether there was evidence for compensatory regulation that might be protective in F1 hybrids. We found no evidence for significant differences in expression of ndufs5 or ndufa13 (Supplementary Information 1.3.6 and Supplementary Figs. 21 and 22) or in mitochondrial copy number (Fig. 3b and Supplementary Information 1.3.7) between F1 hybrids and parental species.
With no indication of a compensatory regulatory response, we reasoned that we might be able to detect reduced mitochondrial complex I function in hybrids heterozygous for ancestry at ndufs5 and ndufa13. We quantified respiratory phenotypes in isolated mitochondria using an Oroboros O2K respirometer in adult hybrids and parental species (Methods, Supplementary Fig. 23 and Supplementary Information 1.3.8). We found that complex I efficiency was lower in hybrids (Fig. 3c and Supplementary Fig. 24, orthogonal contrast t = −2.53, P = 0.023, n = 7 per genotype), and that the time required for hybrids to reach maximum complex I-driven respiration was around 2.5 times longer (orthogonal contrast t = 4.303, P < 0.001; Fig. 3d and Supplementary Fig. 25). Conversely, overall levels of mitochondrial respiration were unaffected by genotype (Fig. 3e, orthogonal contrast t = 0.078, P = 0.94, n = 7 per genotype; Supplementary Information 1.3.8) as were other measures of mitochondrial integrity and function (Supplementary Figs. 26 and 27 and Supplementary Information 1.3.8 and 1.3.9). Together, these data point to reduced function of complex I without broader phenotypic consequences in individuals that are heterozygous for incompatible alleles48.
Given the physiological evidence for some reduction in complex I function in hybrids heterozygous at ndufs5 and ndufa13, we predicted that there might be an altered frequency of protein complexes incorporating both X. malinche mitochondrial proteins and X. birchmanni proteins at ndufs5 and ndufa13 in F1 hybrids. To test this prediction, we took a mass spectrometry-based quantitative proteomics approach. We used stable isotope-labelled peptides to distinguish between the X. birchmanni and X. malinche ndufs5 and ndufa13 peptides in mitochondrial proteomes extracted from F1 hybrids (n = 5; see Methods and Supplementary Information 1.4.1–1.4.4). Although endogenous ndufa13 peptides were not observed frequently enough to quantify accurately, we found consistent deviations from the expected 50:50 ratio of X. birchmanni to X. malinche peptides for ndufs5 in F1 hybrids, with a significant overrepresentation of matched ancestry at ndufs5 in the mitochondrial proteome (t = 3.96, P = 0.016; Fig. 3f, Supplementary Fig. 28 and Supplementary Information 1.4.5). Since we did not observe allele-specific expression of ndufs5 (Fig. 3f and Supplementary Information 1.3.6), this result is consistent with disproportionate degradation of X. birchmanni-derived ndufs5 peptides in the mitochondrial proteome or differences in translation of ndufs5 transcripts derived from the two species.
Mitonuclear substitutions in complex I
To begin to explore the possible mitochondrial partners of ndufs5 and ndufa13 among the 37 non-recombining genes in the swordtail mitochondrial genome, we turned to protein modelling, relying on high-quality cryo-electron microscopy (cryo-EM)-based structures49,50,51. Although these structures are only available for distant relatives of swordtails, the presence of the same set of supernumerary complex I subunits and high sequence similarity suggest that using these structures is appropriate (Supplementary Tables 12 and 13, Supplementary Figs. 29–31 and Supplementary Information 1.4.6).
Barring a hybrid incompatibility generated by regulatory divergence (see Supplementary Information 1.3.6), our expectation is that hybrid incompatibilities will be driven by amino acid changes in interacting proteins52. We used the program RaptorX53 to generate predicted structures of X. birchmanni and X. malinche Ndufs5, Ndufa13 and nearby complex I proteins encoded by mitochondrial and nuclear genes, which we aligned to a mouse cryo-EM complex I structure49 (Fig. 4a, Supplementary Figs. 29–31 and Methods). Using these structures, we visualized amino acid substitutions between X. birchmanni and X. malinche at the interfaces of Ndufs5, Ndufa13 and mitochondrial-encoded proteins (Extended Data Fig. 6 and Supplementary Figs. 32 and 33). Whereas there are dozens of substitutions in the four mitochondrial-encoded proteins that are in close physical proximity to Ndufs5 or Ndufa13 (Supplementary Fig. 29; Nd2, Nd3, Nd4l and Nd6), there are only five cases where amino acid substitutions in either nuclear-encoded protein are predicted to be close enough to contact substitutions in any mitochondrial-encoded protein, all of which involve Nd2 or Nd6 (Fig. 4a and Extended Data Table 1; see Supplementary Fig. 33 for pairwise visualizations of interacting proteins). These paired substitutions in regions of close proximity between mitochondrial- and nuclear-encoded proteins suggest that nd2 and nd6 are the genes most likely to be involved in the mitochondrial component of the hybrid incompatibility (Fig. 4a,b Extended Data Fig. 6 and Supplementary Figs. 33–35), and will be promising candidates for functional validation when such approaches become possible in swordtails.
Rapid evolution of complex I proteins
Theory predicts that hybrid incompatibilities are more likely to arise in rapidly evolving genes4,5,6,7. Consistent with this hypothesis, ndufs5 is among the most rapidly evolving genes genome-wide between X. birchmanni and X. malinche (Fig. 4c,d). Aligning the ndufs5 coding sequences of X. birchmanni, X. malinche and 12 other swordtail species revealed that all 4 amino acid substitutions that differentiate X. birchmanni and X. malinche at ndufs5 were derived on the X. birchmanni branch (Fig. 4c). Phylogenetic tests indicate that there has been accelerated evolution of ndufs5 on this branch (inferred ratio of non-synonymous substitutions per non-synonymous site to synonymous substitutions per synonymous site (dN/dS) > 99, N = 4, S = 0, codeml branch test P = 0.005; Fig. 4c). Similar patterns of rapid evolution are observed at ndufa13, which also showed evidence for accelerated evolution in X. birchmanni (Fig. 4e; dN/dS = 1.2, N = 3, S = 1, codeml branch test P = 0.002). Although explicit tests for adaptive evolution at ndufs5 and ndufa13 could not exclude a scenario of relaxed selection (Extended Data Table 2 and Supplementary Information 1.5.1 and 1.5.2), our comparisons across phylogenetic scales highlight strong conservation in some regions of the proteins and rapid turnover in others, complicating our interpretation of this test (Supplementary Fig. 36).
Rapid evolution of ndufs5 and ndufa13 could be driven by coevolution with mitochondrial substitutions, a mechanism that has been proposed to explain the outsized role of the mitochondria in hybrid incompatibilities15,54. Indeed, there is an excess of derived substitutions in the X. birchmanni mitochondrial protein Nd6, one of the proteins that physically contacts Ndufs5 and Ndufa13 (Extended Data Fig. 7 and Extended Data Table 2; codeml branch test P = 0.005). Moreover, several of the substitutions observed in both mitochondrial and nuclear genes are predicted to have functional consequences (Extended Data Table 3 and Supplementary Information 1.5.1), including ones predicted to be in contact between Ndufs5, Ndufa13, Nd2 and Nd6 (Fig. 4a,b and Extended Data Fig. 6).
Introgression of incompatibility genes
The presence of a mitonuclear incompatibility in Xiphophorus is especially intriguing, given previous reports that mitochondrial genomes may have introgressed between species29. While X. malinche and X. birchmanni are sister species based on the nuclear genome, they are mitochondrially divergent, with X. malinche and Xiphophorus cortezi grouped as sister species based on the mitochondrial phylogeny29 (Fig. 5a,b). As we show, all X. cortezi mitochondria sequenced to date are nested within X. malinche mitochondrial diversity (Fig. 5b, Supplementary Fig. 37 and Supplementary Information 1.5.3 and 1.5.4). Simulations indicate that gene flow, rather than incomplete lineage sorting, drove replacement of the X. cortezi mitochondria with the X. malinche sequence (P < 0.002 by simulation; Fig. 5c and Supplementary Information 1.5.4).
The introgression of the mitochondrial genome from X. malinche into X. cortezi raises the possibility that other complex I genes may have co-introgressed55. Indeed, the nucleotide sequence for ndufs5 is identical between X. malinche and X. cortezi, and the sequence of ndufa13 differs by a single synonymous mutation (although conservation of both genes is high throughout Xiphophorus; Supplementary Figs. 38 and 39). The identical amino acid sequences of the proteins suggest that hybrids between X. cortezi and X. birchmanni are likely to harbour the same mitonuclear incompatibility we observe between X. malinche and X. birchmanni, as a result of ancient introgression between X. malinche and X. cortezi (Fig. 5d and Supplementary Information 1.5.3–1.5.5).
This inference is supported by analysis of three contemporary X. birchmanni × X. cortezi hybrid populations40 (Supplementary Fig. 40). We find that all known X. birchmanni × X. cortezi hybrid populations are fixed for the mitochondrial genome from X. cortezi (that originated in X. malinche) and show a striking depletion of X. birchmanni ancestry at ndufs5 and ndufa13 (Fig. 5e and Supplementary Fig. 41). This replicated depletion is not expected by chance (Fig. 5e and Supplementary Information 1.5.6, P = 0.0001) and instead indicates that selection has acted on these regions. These results suggest that the mitonuclear incompatibility observed in X. birchmanni × X. malinche is also active in hybridizing X. birchmanni × X. cortezi populations. This exciting finding shows that genes underlying hybrid incompatibilities can introgress together, transferring incompatibilities between related species.
Discussion
Here we investigate the genetic and evolutionary forces that drive the emergence of hybrid incompatibilities. Theory predicts that hybrid incompatibilities involving multiple genes should be common7,8, but with few exceptions9,11,12,13, they remain almost uncharacterized at the genic level8. We have identified incompatible interactions in mitochondrial complex I that cause hybrid lethality in laboratory and wild populations. Our findings in naturally hybridizing species echo predictions from theory and studies in laboratory models9,11,12,13 suggesting that protein complexes may be a critical site of hybrid breakdown.
Researchers have proposed mitonuclear interactions as hotspots for the emergence of hybrid incompatibilities, given that mitochondrial genomes often experience higher substitution rates between species17,18,56, yet must intimately interact with nuclear proteins to perform essential cellular functions22,23. Our findings support this prediction, identifying incompatible interactions with both the X. malinche and X. birchmanni mitochondria. We also show that there has been exceptionally rapid evolution in both mitochondrial and interacting nuclear genes in X. birchmanni (Fig. 4). Whether driven by adaptation or some other mechanism, our findings support the hypothesis that the coevolution of mitochondrial and nuclear genes could drive the overrepresentation of mitonuclear interactions in hybrid incompatibilities22,23,54. More broadly, our results are consistent with predictions that rapidly evolving proteins are more likely to become involved in hybrid incompatibilities than their slowly evolving counterparts4,5,6.
Characterizing the incompatibility across multiple scales of organization enabled us to explore the mechanisms through which it acts57,58,59. Our results suggest that in the case of the X. malinche mitochondria hybrid lethality is mediated through arrested development in utero of individuals with mismatched ancestry at ndufs5, whereas individuals with ndufa13 mismatch have vascular defects and typically die shortly after birth. Intriguingly, individuals with ndufa13 mismatch that do survive are much less likely to harbour any X. birchmanni alleles at ndufs5 (Fig. 2i). Together, our results indicate that a subtle three-way interaction overlays two strong pairwise mitonuclear incompatibilities at ndufs5 and ndufa13. Evolutionary biologists have been fascinated by the idea that hybrid incompatibilities may commonly involve three or more genes following theoretical work by Orr7 nearly 30 years ago, but this question has been challenging to address empirically. Our results highlight how the nuances of actual fitness landscapes may defy simplifying assumptions.
Finally, this mitonuclear incompatibility provides a new case in which the same genes are involved in incompatibilities across multiple species30,38,60. However, tracing the evolutionary history of the genes that underlie it adds further complexity to this phenomenon: we found that introgression has resulted in the transfer of genes underlying the incompatibility from X. malinche to X. cortezi, and evidence from X. birchmanni × X. cortezi hybrid populations indicates that the incompatibility is probably under selection in these populations as well. The possibility that hybridization could transfer incompatibilities between species has not been previously recognized, perhaps due to an underappreciation of the frequency of hybridization. The impact of past hybridization on the structure of present-day reproductive barriers between species is an exciting area for future inquiry.
Methods
Biological materials
Wild parental and hybrid individuals used in this study were collected from natural populations in Hidalgo, Mexico (permit no. PPF/DGOPA-002/19). Artificial F1 and F2 hybrids were generated using mesocosm tanks as described previously39. Caudal fin clips were used as the source for all DNA isolation and for flow cytometry, and liver tissue for RNA-seq, respirometry, and proteomic assays were collected following Stanford Administrative Panel on Laboratory Animal Care (APLAC) protocol no. 33071.
Genotyping and local ancestry calling
Genomic DNA was extracted from fin clips and individually barcoded tagmentation-based libraries were generated (Supplementary Information 1.1.3). Hybrids were genotyped with low-coverage whole-genome sequencing followed by local ancestry inference across the 24 Xiphophorus chromosomes and the mitochondrial genome using the ancestryinfer pipeline38,39,42,61 (Supplementary Information 1.1.3 and 1.1.4). We converted posterior probabilities for each ancestry state to hard calls for downstream analysis, using a posterior probability threshold of 0.9, and analysed ancestry variation across the genome.
QTL and admixture mapping
The regions interacting with the mitochondrial genome were first identified based on analysis of segregation distortion in 943 F2 hybrids generated from F1 crosses between X. malinche females and X. birchmanni males (Supplementary Information 1.1.1 and Langdon et al.40). Since all hybrids in this artificial cross harboured the X. malinche mitochondria, we scanned for regions of exceptionally high X. malinche ancestry along the genome (>60% X. malinche ancestry), identifying one such region on chromosome 13 and one on chromosome 6 (Fig. 1; see also ref. 40). Evidence for interactions between these regions and the mitochondrial genome were confirmed using admixture mapping in two hybrid populations that segregated for the mitochondrial haplotype of both species (Supplementary Information 1.1.2): the Calnali Low hybrid population (n = 359) and the Chahuaco falls hybrid population (n = 244). In brief, we used a partial correlation analysis to identify regions of the genome strongly associated with mitochondrial ancestry, after regressing out genome-wide ancestry to account for covariance in ancestry due to population structure (see ref. 36 and Supplementary Information 1.1.5 and 1.1.9). Significance thresholds were determined using simulations (Supplementary Information 1.1.5; see also Supplementary Information 1.1.6).
Estimating selection on the incompatibility
We used an ABC approach to estimate the strength of selection against the incompatible interaction between the X. malinche mitochondrial haplotype and X. birchmanni ancestry at the two nuclear genes involved in the hybrid incompatibility: ndufs5 and ndufa13 (Supplementary Information 1.2.2). For these simulations, we asked what selection coefficients (0–1) and dominance coefficients (0–1) could generate the observed deviations from the expectation of 50:50 X. birchmanni–X. malinche ancestry in F2 hybrids at ndufs5 and ndufa13 after two generations of selection. We performed 500,000 simulations for each interaction and accepted or rejected simulations based on comparisons to the real data using a 5% tolerance threshold (Supplementary Information 1.2.2). We also evaluated evidence for incompatible interactions with the X. birchmanni mitochondrial haplotype (Supplementary Information 1.2.1–1.2.4).
Embryo staging and genotyping
To pinpoint when in development the incompatibility between the X. malinche mitochondria and X. birchmanni nuclear genotypes causes lethality, we collected a dataset on the developmental stages of embryos with different genotype combinations. Whole ovaries were removed from pregnant females and embryos were individually dissected. Each embryo was assigned a developmental stage ranging from 1–11 based on established protocols for poeciliid embryos44. Unfertilized eggs were excluded from analysis. Following staging, individual embryos (n = 296) were genotyped as described above and in Supplementary Information 1.3.1. We tested for significant differences in developmental stage between siblings with compatible and incompatible genotype combinations using a two-sided two-sample t-test (Supplementary Information 1.3.1) and examined differences in ancestry between large groups of siblings that varied in their developmental stages (Supplementary Information 1.1.7). We also collected data on embryonic stage and variability between siblings in embryonic stage from both pure parental species (Supplementary Information 1.3.1). We used a different approach to pinpoint the timing of ndufa13 lethality given that it appeared to act postnatally (Supplementary Information 1.3.3).
Embryo respirometry and morphometrics
To study the mechanisms of ndufs5- and ndufa13-driven lethality, we performed oxygen consumption measurements on F2 embryos in a Loligo plate respirometer (Supplementary Information 1.3.4). Embryos were dissected from mothers and transferred to wells of a 24-well plate, where their oxygen consumption was measured over 60 min. The measurement was then repeated in media dosed with 5 μM rotenone to test sensitivity to complex I inhibition, after which the embryos were video recorded and photographed for morphometrics in ImageJ. We used linear models to test the effect of ndufs5 genotype, ndufa13 genotype, and individual standard length on a number of variables, controlling for batch effects (Supplementary Information 1.3.4 and 1.3.5).
Mitochondrial respirometry
To further evaluate mitochondrial function in individuals heterozygous for the mitonuclear incompatibility (Supplementary Information 1.3.6 and 1.3.7), we conducted respirometry assays on X. birchmanni, X. malinche, and hybrid individuals that had the X. malinche mitochondria and were heterozygous for the nuclear components of the hybrid incompatibility (n = 7 of each genotype). Mitochondria were isolated from whole liver tissue and mitochondrial respiration was quantified using the Oroboros O2K respirometry system62 (Supplementary Fig. 23). A step-by-step description of this protocol and methods used to calculate respiratory flux control factors is outlined in Supplementary Information 1.3.8. We complemented the results of these respirometry experiments with measures of mitochondrial membrane potential using a flow cytometry-based approach (Supplementary Information 1.3.9).
Parallel reaction monitoring proteomics
For parallel reaction monitoring (PRM) with mass spectrometry, we used a similar approach to that used for respirometry to isolate whole mitochondria from five F1 hybrids (which harboured X. malinche mitochondria). This approach is described in detail in Supplementary Information 1.4.1. In brief, we designed heavy isotope-labelled peptides to distinguish between the X. birchmanni and X. malinche copies of Ndufs5 and Ndufa13, facilitating quantification of the peptides of interest in the mitochondrial proteome (Supplementary Information 1.4.2). Mitochondrial isolates were prepared for mass spectrometry and combined with heavy isotope-labelled peptides in known quantities (see Supplementary Information 1.4.3), then submitted to Orbitrap mass spectrometry with separation with ultra performance liquid chromatography and PRM for ion selection. The protocol for mass spectrometry and PRM is described in detail in Supplementary Information 1.4.4.
To analyse the results, the focal peptide’s spectral peak was identified based on the peak of the heavy isotope-labelled spike-in peptide. We focused analysis on the Ndufs5 peptide WLL[L/P]QSGEQPYK, since other endogenous peptides were below the expected sensitivity limits of our PRM protocol (Supplementary Information 1.4.5). We normalized the intensity of the Ndufs5 peptide based on the known spike-in quantity, and quantified the proportion of Ndufs5 in each F1 individual derived from X. malinche versus X. birchmanni (Supplementary Information 1.4.5). We tested whether these ratios significantly deviated from the 50:50 expectation for F1 hybrids using a two-sided one-sample t-test.
Complex I protein modelling
Mapping results allowed us to identify ndufs5 and ndufa13 as X. birchmanni genes that interact negatively with X. malinche mitochondrial genes. We used a protein modelling-based approach with RaptorX (http://raptorx.uchicago.edu) to identify the mitochondrial genes most likely to interact with ndufs5 and ndufa13 (see Supplementary Information 1.4.6). Using the mouse cryo-EM structure (Protein Data Bank (PDB) ID 6G2J) of complex I, we identified proteins in contact with Ndufs5 and Ndufa13, which included several mitochondrial (Nd2, Nd3, Nd4l and Nd6) and nuclear (Ndufa1, Ndufa8, Ndufb5 and Ndufc2) proteins. We then used RaptorX to predict structures for both the X. birchmanni and X. malinche versions of the proteins. In addition, we evaluated the robustness of these predictions to choice of cryo-EM template; see Supplementary Information 1.4.6.
Analysis of evolutionary rates
Comparison of predicted protein sequences encoded by ndufs5, ndufa13 and mitochondrial genes of interest (nd2 and nd6) revealed a large number of substitutions between X. birchmanni and X. malinche. We calculated dN/dS between X. birchmanni and X. malinche for all annotated protein coding genes throughout the genome and found that both ndufs5 and ndufa13 have rapid protein evolution (Fig. 4d and Supplementary Information 1.5.1). Examining these mutations in a phylogenetic context revealed that many substitutions in ndufs5, nudfa13 and nd6 were derived in X. birchmanni. We tested for significant differences in evolutionary rates on the X. birchmanni lineage and for predicted functional impacts of these substitutions; these analyses are described in Supplementary Information 1.5.1.
Tests for ancient introgression
Previous work had indicated that the mitochondrial phylogeny in Xiphophorus is discordant with the whole-genome species tree29. Specifically, although X. birchmanni and X. malinche are sister species based on the nuclear genome, X. malinche and X. cortezi are sister species based on the mitochondrial genome. We used a combination of PacBio amplicon sequencing of 10 individuals (2 or more per species, Supplementary Information 1.5.3) and newly available whole-genome resequencing data to confirm this result and polarize the direction of the discordance by constructing maximum likelihood mitochondrial phylogenies with the program RAxML63. We performed similar phylogenetic analyses of the nuclear genes that interact with the X. malinche mitochondria (ndufs5 and ndufa13; Supplementary Information 1.5.3). Combined with phylogenetic results, simulation results suggest that gene flow from X. malinche into X. cortezi is the most likely cause of the discordance we observe between the mitochondrial and nuclear phylogenies (Supplementary Information 1.5.3 and 1.5.4). Since X. malinche and X. cortezi are not currently sympatric, this suggests ancient gene flow between them (Supplementary Information 1.5.5).
X. birchmanni × X. cortezi hybridization
To investigate the possibility that hybrids between X. birchmanni and X. cortezi share the same mitonuclear incompatibility as observed in hybrids between X. birchmanni and X. malinche (Supplementary Information 1.5.6), we took advantage of genomic data from recently discovered hybrid populations between X. birchmanni and X. cortezi64. Using data from three different X. birchmanni × X. cortezi populations and a permutation-based approach, we tested whether ancestry at ndufs5 and ndufa13 showed lower mismatch with mitochondrial ancestry than expected given the genome-wide ancestry distribution. This analysis is described in detail in Supplementary Information 1.5.6.
Animal care and use
All methods were performed in compliance with Stanford APLAC protocol no. 33071.
Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
Data availability
Raw sequencing reads used in this project are available under NCBI SRA Bioprojects PRJNA744894, PRJNA746324, PRJNA610049, PRJNA361133 and PRJNA745218. Mass spectrometry data are available on PRIDE with identifier PXD046217, and other datasets necessary to recreate the results of the publication are available on Dryad (https://doi.org/10.5061/dryad.j3tx95xmx). Templates for complex I protein structural modelling were accessed from the Protein Data Bank (PDB) with accession numbers 6G2J, 6G72, 5LDW, 5LNK and 5XTC.
Code availability
All custom scripts used to generate results are available on Github at https://github.com/Schumerlab/mitonuc_DMI and https://github.com/Schumerlab/Lab_shared_scripts.
References
Dagilis, A. J., Kirkpatrick, M. & Bolnick, D. I. The evolution of hybrid fitness during speciation. PLoS Genet. 15, e1008125 (2019).
Dobzhansky, T. Genetic nature of species differences. Am. Nat. 71, 404–420 (1937).
Muller, H. Isolating mechanisms, evolution, and temperature. Biol. Symp. 6, 71–125 (1942).
Maheshwari, S. & Barbash, D. A. The genetics of hybrid incompatibilities. Annu. Rev. Genet. 45, 331–355 (2011).
Presgraves, D. C. The molecular evolutionary basis of species formation. Nat. Rev. Genet. 11, 175–180 (2010).
Johnson, N. A. Hybrid incompatibility genes: remnants of a genomic battlefield? Trends Genet. 26, 317–325 (2010).
Orr, H. A. The population genetics of speciation: the evolution of hybrid incompatibilities. Genetics 139, 1805–1813 (1995).
Swamy, K. B. S., Schuyler, S. C. & Leu, J.-Y. Protein complexes form a basis for complex hybrid incompatibility. Front. Genet. 12, 609766 (2021).
Boocock, J., Sadhu, M. J., Durvasula, A., Bloom, J. S. & Kruglyak, L. Ancient balancing selection maintains incompatible versions of the galactose pathway in yeast. Science 371, 415–419 (2021).
Wei, W.-H., Hemani, G. & Haley, C. S. Detecting epistasis in human complex traits. Nat. Rev. Genet. 15, 722–733 (2014).
Phadnis, N. et al. An essential cell cycle regulation gene causes hybrid inviability in Drosophila. Science 350, 1552–1555 (2015).
Tang, S. & Presgraves, D. C. Evolution of the Drosophila nuclear pore complex results in multiple hybrid incompatibilities. Science 323, 779–782 (2009).
Cooper, J. C., Guo, P., Bladen, J. & Phadnis, N. A triple-hybrid cross reveals a new hybrid incompatibility locus between D. melanogaster and D. sechellia. Preprint at bioRxiv https://doi.org/10.1101/590588 (2019).
Kuzmin, E. et al. Systematic analysis of complex genetic interactions. Science 360, eaao1729 (2018).
Burton, R. S. & Barreto, F. S. A disproportionate role for mtDNA in Dobzhansky–Muller incompatibilities? Mol. Ecol. 21, 4942–4957 (2012).
Hill, G. E. Mitonuclear coevolution as the genesis of speciation and the mitochondrial DNA barcode gap. Ecol. Evol. 6, 5831–5842 (2016).
Ballard, J. W. O. & Whitlock, M. C. The incomplete natural history of mitochondria. Mol. Ecol. 13, 729–744 (2004).
Allio, R., Donega, S., Galtier, N. & Nabholz, B. Large variation in the ratio of mitochondrial to nuclear mutation rate across animals: implications for genetic diversity and the use of mitochondrial DNA as a molecular marker. Mol. Biol. Evol. 34, 2762–2772 (2017).
Barr, C. M., Neiman, M. & Taylor, D. R. Inheritance and recombination of mitochondrial genomes in plants, fungi and animals. New Phytol. 168, 39–50 (2005).
Case, A. L., Finseth, F. R., Barr, C. M. & Fishman, L. Selfish evolution of cytonuclear hybrid incompatibility in Mimulus. Proc. R. Soc. Lond. B 283, 20161493 (2016).
Lane, N. & Martin, W. The energetics of genome complexity. Nature 467, 929–934 (2010).
Osada, N. & Akashi, H. Mitochondrial–nuclear interactions and accelerated compensatory evolution: evidence from the primate cytochrome c oxidase complex. Mol. Biol. Evol. 29, 337–346 (2012).
Barreto, F. S. & Burton, R. S. Evidence for compensatory evolution of ribosomal proteins in response to rapid divergence of mitochondrial rRNA. Mol. Biol. Evol. 30, 310–314 (2013).
Bolnick, D. I., Turelli, M., López-Fernández, H., Wainwright, P. C. & Near, T. J. Accelerated mitochondrial evolution and “Darwin’s corollary”: asymmetric viability of reciprocal F1 hybrids in centrarchid fishes. Genetics 178, 1037–1048 (2008).
Turelli, M. & Moyle, L. C. Asymmetric postmating isolation: Darwin’s corollary to Haldane’s rule. Genetics 176, 1059–1088 (2007).
Tiffin, P., Olson, S. & Moyle, L. C. Asymmetrical crossing barriers in angiosperms. Proc. R. Soc. Lond. B 268, 861–867 (2001).
Taylor, S. A. & Larson, E. L. Insights from genomes into the evolutionary importance and prevalence of hybridization in nature. Nat. Ecol. Evol. 3, 170–177 (2019).
Edelman, N. B. & Mallet, J. Prevalence and adaptive impact of introgression. Annu. Rev. Genet. 55, 265–283 (2021).
Cui, R. et al. Phylogenomics reveals extensive reticulate evolution in Xiphophorus fishes. Evolution 67, 2166–2179 (2013).
Nelson, T. C. et al. Ancient and recent introgression shape the evolutionary history of pollinator adaptation and speciation in a model monkeyflower radiation (Mimulus section Erythranthe). PLoS Genet. 17, e1009095 (2021).
Buck, R. & Flores-Rentería, L. The syngameon enigma. Plants 11, 895 (2022).
Schumer, M., Cui, R., Rosenthal, G. G. & Andolfatto, P. Reproductive isolation of hybrid populations driven by genetic incompatibilities. PLoS Genet. 11, e1005041 (2015).
Culumber, Z. W. et al. Replicated hybrid zones of Xiphophorus swordtails along an elevational gradient. Mol. Ecol. 20, 342–356 (2011).
Fisher, H. S., Wong, B. B. M. & Rosenthal, G. G. Alteration of the chemical environment disrupts communication in a freshwater fish. Proc. R. Soc. B 273, 1187–1193 (2006).
Schumer, M. et al. Natural selection interacts with recombination to shape the evolution of hybrid genomes. Science 360, 656–660 (2018).
Schumer, M. & Brandvain, Y. Determining epistatic selection in admixed populations. Mol. Ecol. 25, 2577–2591 (2016).
Schumer, M. et al. High-resolution mapping reveals hundreds of genetic incompatibilities in hybridizing fish species. eLife 3, e02535 (2014).
Powell, D. L. et al. Natural hybridization reveals incompatible alleles that cause melanoma in swordtail fish. Science 368, 731–736 (2020).
Powell, D. L. et al. The genetic architecture of variation in the sexually selected sword ornament and its evolution in hybrid populations. Curr. Biol. 31, 923–935.e11 (2021).
Langdon, Q. K. et al. Predictability and parallelism in the contemporary evolution of hybrid genomes. PLoS Genet. 18, e1009914 (2022).
Schartl, M. et al. The genome of the platyfish, Xiphophorus maculatus, provides insights into evolutionary adaptation and several complex traits. Nat. Genet. 45, 567–572 (2013).
Schumer, M., Powell, D. L. & Corbett-Detig, R. Versatile simulations of admixture and accurate local ancestry inference with mixnmatch and ancestryinfer. Mol. Ecol. Resour. 20, 1141–1151 (2020).
Sharma, L. K., Lu, J. & Bai, Y. Mitochondrial respiratory complex I: structure, function and implication in human diseases. Curr. Med. Chem. 16, 1266–1277 (2009).
Haynes, J. L. Standardized classification of poeciliid development for life-history studies. Copeia 1995, 147 (1995).
Byrnes, J. et al. Pharmacologic modeling of primary mitochondrial respiratory chain dysfunction in zebrafish. Neurochem. Int. 117, 23–34 (2018).
Pinho, B. R. et al. How mitochondrial dysfunction affects zebrafish development and cardiovascular function: an in vivo model for testing mitochondria-targeted drugs. Br. J. Pharmacol. 169, 1072–1090 (2013).
González-Quintana, A. et al. Novel NDUFA13 mutations associated with OXPHOS deficiency and Leigh syndrome: a second family report. Genes 11, 855 (2020).
McElroy, G. S. et al. Reduced expression of mitochondrial complex I subunit Ndufs2 does not impact healthspan in mice. Sci. Rep. 12, 5196 (2022).
Agip, A.-N. A. et al. Cryo-EM structures of complex I from mouse heart mitochondria in two biochemically defined states. Nat. Struct. Mol. Biol. 25, 548–556 (2018).
Kampjut, D. & Sazanov, L. A. The coupling mechanism of mammalian respiratory complex I. Science 370, eabc4209 (2020).
Fiedorczuk, K. et al. Atomic structure of the entire mammalian mitochondrial complex I. Nature 538, 406–410 (2016).
Mack, K. L. & Nachman, M. W. Gene regulation and speciation. Trends Genet. 33, 68–80 (2017).
Källberg, M. et al. Template-based protein structure modeling using the RaptorX web server. Nat. Protoc. 7, 1511–1522 (2012).
Hill, G. E. Mitonuclear compensatory coevolution. Trends Genet. 36, 403–414 (2020).
Sloan, D. B., Havird, J. C. & Sharbrough, J. The on-again, off-again relationship between mitochondrial genomes and species boundaries. Mol. Ecol. 26, 2212–2236 (2017).
Haag-Liautard, C. et al. Direct estimation of the mitochondrial DNA mutation rate in Drosophila melanogaster. PLoS Biol. 6, e204 (2008).
Ellison, C. K. & Burton, R. S. Disruption of mitochondrial function in interpopulation hybrids of Tigriopus californicus. Evolution 60, 1382–1391 (2006).
Ellison, C. K., Niehuis, O. & Gadau, J. Hybrid breakdown and mitochondrial dysfunction in hybrids of Nasonia parasitoid wasps. J. Evol. Biol. 21, 1844–1851 (2008).
Olson, J. R., Cooper, S. J., Swanson, D. L., Braun, M. J. & Williams, J. B. The relationship of metabolic performance and distribution in black‐capped and Carolina chickadees. Physiol. Biochem. Zool. 83, 263–275 (2010).
Lu, Y. et al. Oncogenic allelic interaction in Xiphophorus highlights hybrid incompatibility. Proc. Natl Acad. Sci. USA 117, 29786–29794 (2020).
Corbett-Detig, R. & Nielsen, R. A hidden Markov model approach for simultaneously estimating local ancestry and admixture time using next generation sequence data in samples of arbitrary ploidy. PLoS Genet. 13, e1006529 (2017).
Chung, D. J., Bryant, H. J. & Schulte, P. M. Thermal acclimation and subspecies-specific effects on heart and brain mitochondrial performance in a eurythermal teleost (Fundulus heteroclitus). J. Exp. Biol. 220, 1459–1471 (2017).
Stamatakis, A. RAxML-VI-HPC: Maximum likelihood-based phylogenetic analyses with thousands of taxa and mixed models. Bioinformatics 22, 2688–2690 (2006).
Powell, D. L. et al. Two new hybrid populations expand the swordtail hybridization model system. Evolution 75, 2524–2539 (2021).
Acknowledgements
The authors thank P. Andolfatto, S. Aguillon, Y. Brandvain, J. Coughlan, H. Fraser, Y. Haba, N. Phadnis, M. Przeworski, K. Thompson and members of the Schumer laboratory for helpful discussion and/or feedback on earlier versions of this manuscript, and A. Pollock for help performing rotenone trials. We thank the Federal Government of Mexico for permission to collect fish. Stanford University and the Stanford Research Computing Center provided computational support for this project. We thank the Vincent Coates Foundation Mass Spectrometry Laboratory, Stanford University Mass Spectrometry (RRID:SCR_017801) for technical and experimental support. This work was supported by a Knight–Hennessy Scholars fellowship and NSF GRFP 2019273798 to B.M.M., a CEHG fellowship and NSF PRFB (2010950) to Q.K.L., NIH P30 CA124435 in utilizing the Stanford Cancer Institute Proteomics/Mass Spectrometry Shared Resource, NIH grant 1R35GM142836 to J.C.H., and a Hanna H. Gray fellowship, Sloan Fellowship, and NIH grant 1R35GM133774 to M. Schumer.
Author information
Authors and Affiliations
Contributions
B.M.M., D.L.P. and M. Schumer designed the project. B.M.M., C.Y.P., E.N.K.I., S.M.B., A.E.D., R.A.R.-S., A.M., J.J.B., K.M.K., F.L., R.M., K.S., O.H.-P., J.C.H., A.M. and M. Schumer collected data. B.M.M., C.Y.P., Q.K.L., F.L., J.C.H., R.A.R.-S., A.M. and M. Schumer performed analyses. D.L.P., T.R.G., R.D.L., C.G., R.C.-D., J.F. and M. Schartl provided expertise and technical support.
Corresponding authors
Ethics declarations
Competing interests
The authors declare no competing interests.
Peer review
Peer review information
Nature thanks Navdeep Chandel, Lena Ho and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Peer review reports are available.
Additional information
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Extended data figures and tables
Extended Data Fig. 1 ABC inference of additional selection parameters.
(a) Posterior distributions of dominance coefficients from approximate Bayesian computation (ABC) simulations fitting observed data in F2 hybrids. Left distribution shows the results of simulations modeling the X. malinche mitochondria – X. birchmanni ndufs5 component of the hybrid incompatibility, and right distribution shows the results of simulations modeling the X. malinche mitochondria – X. birchmanni ndufa13 component of the hybrid incompatibility. Each posterior distribution shows the accepted dominance coefficients in 500 simulations and the gray dashed line indicates the maximum a posteriori estimate for the dominance coefficients of interactions involving ndufs5 and ndufa13 (b) Posterior distributions of selection coefficients from ABC simulations fitting observed data in the Calnali Low hybrid population. Left distribution shows the results of simulations modeling the X. malinche mitochondria – X. birchmanni ndufs5 incompatibility and right distribution shows the results of simulations modeling the reverse X. birchmanni mitochondria – X. malinche ndufs5 incompatibility. In the main text we focus on inferred selection coefficients from F2 hybrids for the incompatibility involving the X. malinche mitochondria (see Fig. 1) but present results from fitting the Calnali Low population data here and in Supplementary Information 1.2.2. Distribution shows the accepted selection coefficients in 500 simulations and the gray dashed line indicates the maximum a posteriori estimate. Note that we recover a much broader distribution and a lower maximum a posteriori estimate of the selection coefficient for the X. malinche mitochondria – X. birchmanni ndufs5 incompatibility here compared to F2 simulations (Fig. 1). We interpret this result to be driven by the fact that simulations fitting the Calnali Low population data modeled >20 generations of admixture. Thus, a range of selection coefficients are consistent with the observation that no individuals have X. malinche mtDNA and homozygous X. birchmanni ancestry at ndufs5 in this population.
Extended Data Fig. 2 Ancestry depletion in natural hybrid populations.
Average non-mitochondrial parent ancestry on (a) chromosome 13, (b) chromosome 6, and (c) chromosome 15 in natural hybrid populations fixed for X. birchmanni (Aguazarca, Acuapa, left column) or X. malinche (Tlatemaco, right column) mitochondrial haplotypes. Vertical dashed lines represent the position of ndufs5 and ndufa13 within the admixture mapping regions. (A) Non-mitochondrial parent ancestry is depleted around ndufs5 in all natural hybrid populations that are fixed for a particular mitochondrial haplotype. This pattern is strongly suggestive of a history of selection on this region in natural hybrid populations. In (B), ndufa13 is fixed for X. malinche ancestry only in the population with X. malinche mitochondria (Tlatemaco), mirroring the architecture of the genetic interaction between this gene and the mitochondria. Specifically, interactions with ndufa13 are only expected to be under selection in combination with the X. malinche mitochondrial haplotype (Fig. 1c).
Extended Data Fig. 3 Chromosome 15 incompatibility.
(a) Observed genotype frequencies at the peak associated marker (3.37 Mb) on chromosome 15 in the admixture mapping population. (b) Schematic of identified interactions with the X. birchmanni mitochondrial genome from our mapping data and strength of selection underlying each interaction in hybrids (gray skull – moderate, black skull – near lethal). We discuss interactions with the X. birchmanni mitochondria in more detail in Supplementary Information 1.2.1–1.2.2.
Extended Data Fig. 4 Additional F2 embryo morphometrics by ndufs5 genotype.
Relationship between (a) yolk diameter and ndufs5 genotype (n = 41 birchmanni, 96 heterozygotes, and 46 malinche) and (b) head width and ndufs5 genotype (n = 44 birchmanni, 108 heterozygotes, and 55 malinche) in F2 hybrid embryos. To control for the strong effect of length, the residuals of each variable after accounting for body length are plotted. Grey points represent individual measurements, colored points with vertical lines represent group mean ± 2 SE, and brackets with asterisks denote significant differences from Tukey’s HSD test.
Extended Data Fig. 5 Juvenile F2 heart morphology by ndufa13 genotype.
In all panels, colored points and bars show the mean ± 2 SE, and gray points show individual data. Individuals that were homozygous X. birchmanni at ndufa13 (the incompatible genotype) and homozygous X. malinche at ndufa13 (the compatible genotype) were raised in common laboratory conditions and sampling occurred at approximately 5 months of age (n = 5 juveniles per genotype). Measurements were taken from the sagittal section of largest cross-sectional area for the atrium (a-c) and ventricle (d-f), and area of occupancy for each cell class was calculated from the average of three quadrats (Supplementary Information 1.3.3). (g) Representative images of atria from incompatible (X. birchmanni) and compatible (X. malinche) individuals. Images are from the slide with maximum cross-sectional atrial area from each individual. (h) Example of histology analysis process in a representative atrium. Red square indicates randomly placed quadrat in which occupancy was calculated, and yellow borders represent areas which were manually annotated as cardiomyocytes. Note that the epicardium is visible at top left. All fish were raised and processed as one experimental group, with no independent attempts to test reproducibility at the experimental level.
Extended Data Fig. 6 Interface between ndufa13, ndufs5, and nd6 in RaptorX model.
Arrows highlight substitutions in ndufa13, ndufs5 and mitochondrial proteins in proximity in the model, with alphanumeric codes denoting the X. malinche amino acid, the residue number, and the X. birchmanni amino acid, from left to right. Colors distinguish proteins, as denoted by colored protein names at left. Asterisks denote residues with substitutions in X. birchmanni computationally predicted to affect protein function (Extended Data Table 3).
Extended Data Fig. 7 Complex I mtDNA gene trees.
Gene trees were generated with RAxML for (A) nd6 and (B) nd2, highlighting an excess of substitutions along the X. birchmanni branch in nd6. Scale bar represents number of nucleotide substitutions per site, and derived non-synonymous substitutions are indicated by red ticks along the phylogeny. Note that spacing between ticks is arbitrary and substitutions were placed on branches to maximize parsimony. In some cases, the distribution of substitutions cannot be explained by a single event, in such cases we illustrate the minimum number of events leading to observed distribution of the substitution.
Supplementary information
Supplementary Information
Supplementary text, Supplementary Figs. 1–54, Supplementary Tables 4–20 and references.
Supplementary Data 1
Multiple sequence alignments of ten complex I proteins for X. birchmanni, X. malinche and mammalian modelling templates.
Supplementary Table 1
Locations of annotated complex I genes in the X. birchmanni genome assembly.
Supplementary Table 2
Genes found in the admixture mapping QTL region on chromosome 6.
Supplementary Table 3
Genes found in the admixture mapping QTL region on chromosome 15.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Moran, B.M., Payne, C.Y., Powell, D.L. et al. A lethal mitonuclear incompatibility in complex I of natural hybrids. Nature 626, 119–127 (2024). https://doi.org/10.1038/s41586-023-06895-8
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1038/s41586-023-06895-8
This article is cited by
-
Mitochondrial haplotype and mito-nuclear matching drive somatic mutation and selection throughout ageing
Nature Ecology & Evolution (2024)
Comments
By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.