Experimental introgression in Drosophila: Asymmetric postzygotic isolation associated with chromosomal inversions and an incompatibility locus on the X chromosome

Abstract Interspecific gene flow (introgression) is an important source of new genetic variation, but selection against it can reinforce reproductive barriers between interbreeding species. We used an experimental approach to trace the role of chromosomal inversions and incompatibility genes in preventing introgression between two partly sympatric Drosophila virilis group species, D. flavomontana and D. montana. We backcrossed F1 hybrid females from a cross between D. flavomontana female and D. montana male with the males of the parental species for two generations and sequenced pools of parental strains and their reciprocal second generation backcross (BC2mon and BC2fla) females. Contrasting the observed amount of introgression (mean hybrid index, HI) in BC2 female pools along the genome to simulations under different scenarios allowed us to identify chromosomal regions of restricted and increased introgression. We found no deviation from the HI expected under a neutral null model for any chromosome for the BC2mon pool, suggesting no evidence for genetic incompatibilities in backcrosses towards D. montana. In contrast, the BC2fla pool showed high variation in the observed HI between different chromosomes, and massive reduction of introgression on the X chromosome (large X‐effect). This observation is compatible with reduced recombination combined with at least one dominant incompatibility locus residing within the X inversion(s). Overall, our study suggests that genetic incompatibilities arising within chromosomal inversions can play an important role in speciation.

At the same time, selection against introgression at certain loci acts to maintain barrier loci and protect species' integrity from the negative effects of hybridization (Barton & Bengtsson, 1986; formed gametes are partially avoided, since these gametes remain in the polar nuclei and do not enter the developing gametes (Hoffmann & Rieseberg, 2008;Sturtevant & Beadle, 1936). Perhaps more importantly, reduced recombination across inverted regions, particularly near inversion breakpoints and within overlapping inversions, facilitates the build-up of BDMIs via divergent selection and/or drift (Fishman et al., 2013;Khadem et al., 2011;Mcgaugh & Noor, 2012;Navarro & Barton, 2003;Noor et al., 2001). While blocks of genetic material can occasionally be exchanged through double crossovers within long inversions (Navarro et al., 1997) and smaller DNA sections (several hundred bps) though gene conversion events within any kind of inversions (Korunes & Noor, 2019), recombination within inversions generally remains lower than on colinear chromosome sections (Hoffmann & Rieseberg, 2008).
The disproportionate involvement of sex chromosomes in reproductive isolation in many systems is captured by two general observations: Haldane's rule -the increased F 1 inviability and sterility of the heterogametic sex compared to the homogametic sex (Haldane, 1922;Orr, 1997;Turelli & Orr, 2000) -and the large Xeffect -the fact that the X chromosome shows a disproportionately large effect on the sterility and inviability of backcross hybrids (Masly & Presgraves, 2007;Turelli & Orr, 2000). Explanation for both observations often presume recessivity of X-linked alleles, which can lead to more pronounced effects in hemizygous than in heterozygous hybrids ("Dominance theory"; Coyne & Orr, 2004;Turelli & Orr, 1995, 2000 and/or rapid evolution of X-linked alleles facilitating BDMIs as a byproduct ("Faster X evolution" ;Charlesworth et al., 1987Charlesworth et al., , 2018. The X chromosome has also been suggested to be enriched for genes that create postzygotic isolation in hybrids compared to autosomes (Coyne, 2018). In particular, meiotic drive loci are more frequent on the X than on autosomes, and incompatibilities between drivers and their suppressors in hybrids may generate problems in hybrid development (Courret et al., 2019;Crespi & Nosil, 2013;Crown et al., 2018).

Pairwise BDMIs may involve substitutions in both diverging
lineages, or derived substitutions in one lineage and preserved ancestral alleles in another lineage (Barbash et al., 2004;Cattani & Presgraves, 2009;Coyne & Orr, 2004). BDMIs can also result from cumulative effects of many small incompatibilities or from a single incompatibility between two complementary genes, and the complexity of the incompatibility interaction does not reflect the severity of the barrier (Orr, 1995;Presgraves, 2010a). Importantly,  (Hoikkala & Poikela, 2022;Patterson, 1952 ;Throckmorton, 1982). D. montana has expanded around the northern hemisphere, whereas D. flavomontana has remained in North America (Hoikkala & Poikela, 2022). D. montana lives generally in colder environments and uses different host trees than D. flavomontana (Patterson, 1952;Throckmorton, 1982).
Reproductive barriers between D. montana females and D. flavomontana males are nearly complete, with extremely strong prezygotic barriers and inviability and sterility of rarely produced F 1 hybrids (Poikela et al., 2019). However, in crosses between D. flavomontana females and D. montana males, strong postzygotic isolation is accompanied by prezygotic barriers of variable strength, and F 1 hybrid females can still be crossed with the males of both parental species to obtain backcross progenies in both directions (Poikela et al., 2019). Interspecific hybrids have also reportedly been found in nature (Patterson, 1952;Throckmorton, 1982). Our recent demographic modelling shows that the species have diverged ~3 Mya, with low levels of postdivergence gene flow from D. montana to D. flavomontana . Moreover, we found several inversions that were fixed between the species in all studied individuals across different populations in North America . These inversions were already present in species' common ancestor, and they may have contributed to the build-up and maintenance of adaptive traits and reproductive barriers by restricting gene flow between the evolving lineages .
The goal of this study was to determine which genomic regions are likely to accommodate dominant BDMIs in hybrids between D. montana and D. flavomontana, paying special attention to fixed inversions and the X chromosome. We investigated BDMIs between these species experimentally by sequencing pools of D. montana females from an allopatric population and D. flavomontana females from a (presently) parapatric population, as well as pools of second backcross generation (BC 2 ) females in both directions ( Figure 1). We identified chromosomal regions with decreased and increased introgression by quantifying the amount of introgressed genetic material (mean hybrid index, HI) along the genome in both backcross pools. We then compared the observed HI to the distribution of chromosome-wide HI in in silico replicates of this "introgress-and-resequence" experiment under contrasting assumptions about the presence and location of BDMIs. Since this experimental design involved backcross females, we were able to detect only BDMIs involving a dominant allele, while the recessive-recessive BDMIs remained masked (Table 1) prior to their use in the present study. For the crosses, the flies were sexed under light CO 2 anaesthesia within 3 days after emergence, when they were still virgins. Males and females were transferred into fresh malt vials once a week and used in the crossing experiments at age 20 ± 2 days when they were sexually mature (Salminen & Hoikkala, 2013).

| Crossing experiment
We started the crossing experiment by performing a single-pair cross between D. flavomontana female (strain MT13F11) and D. montana male (strain SE13F37), as reciprocal cross is not successful. Our crossing design (outlined in Figure 1) only involved hybrid females because F 1 males are largely sterile (Päällysaho et al., 2003;Poikela et al., 2019), and because Drosophila males lack recombination (crossing-over) in meiosis. The initial cross produced seven F 1 females, which were backcrossed towards both parental species: four were mated to D. montana males and three to D. flavomontana males. The first backcross generation females (BC 1 mon and BC 1 fla females) were backcrossed to the same paternal species as in the previous generation to obtain BC 2 mon and BC 2 fla females (82 females in both directions). BC 2 females were collected within 3 days after their emergence and stored in −20°C for DNA extractions.

| Fertility of BC 1 females
We defined the fertility of BC 1 females by checking whether they produced progeny after mating with a D. montana or D. flavomontana male (Figure 1). Each BC 1 female was placed in a malt vial with a single male of either species. Once the flies mated, the couple was kept together in the vial so that the female could remate and lay eggs until she died. BC 1 females were considered fertile, if they produced at least some larval, pupal, and/or adult-stage offspring (1 = fertile, 0 = sterile). We used a one-sample Student's t-test (t-test function) to test whether the BC 1 females from the reciprocal crosses showed reduced fertility, when the expected fertility was 1. We also compared the fertility of BC 1 females between the reciprocal crosses to define possible asymmetries (BC 1 mon vs. BC 1 fla), using a generalized linear model (GLM) with binomial distribution (1 = fertile, 0 = sterile) (glm function). All analyses were conducted in base r version 1.2.1335-1 and r studio version 3.6.1.

| Pool-sequencing, mapping, and variant calling
We made DNA extractions from four pools, one pool of each parental strain (D. montana SE13F37 and D. flavomontana MT13F11) and pools for the two second generation backcrosses (BC 2 mon and BC 2 fla). Each pool consisted of 82 females. We used cetyltrimethylammonium bromide (CTAB) solution with RNAse treatment, phenolchloroform-isoamyl alcohol (25:24:1) and chloroform-isoamyl alcohol (24:1) washing steps and ethanol precipitation. Nextera library preparation and 150 bp Illumina paired-end sequencing were performed on two lanes using HiSeq4000 Illumina instrument at Edinburgh Genomics. Illumina paired-end reads of all four samples were quality-checked with fastqc version 0.11.8 (Andrews, 2010) and trimmed for adapter contamination and low-quality bases using fastp version 0.20.0 (using settings --detect_adapter_for_pe, --cut_ front, --cut_tail, --cut_window_size 4, --cut_mean_quality 20; Chen F I G U R E 1 Illustration of the crossing experiment showing the inheritance of sex chromosomes (inheritance of autosomes is similar to that of female X chromosomes). F 1 females, produced in a single-pair cross between Drosophila flavomontana (fla) female and Drosophila montana (mon) male, were backcrossed to either D. flavomontana or D. montana male. In the next generation, each BC 1 female was mated with a male of its paternal species. In every generation, the expected amount of genetic material that is transferred from the gene pool of one species into the gene pool of another one (introgression) is halved (red percentages). Thus, under a null neutral model, we expect a mean HI of 12.5% for the BC 2 pools that were sequenced. Note that recombination occurring in the gametes produced by F 1 and BC 1 females creates variation in the expected amount of HI. For simplicity, the figure shows products of only one crossover event that has occurred in each backcross direction.

Dominant-dominant incompatibility (both loci act dominantly)
A 1 _B 2 _ hybrids are affected in the F 1 generation Recessive-recessive incompatibility (both loci act recessively) A 1 A 1 B 2 B 2 hybrids are affected in the F 2 generation Dominant-recessive incompatibility (A 1 acts dominantly, B 2 B 2 recessively) A 1 _B 2 B 2 hybrids are affected in backcross generations Note: Here, gene A 1 of one species interacts negatively with gene B 2 of another species. Underscore represents any allele, and it does not change the outcome. Note that dominance refers to an allele's effect on fitness on a hybrid genetic background, and it does not necessarily assume dominance of alleles on their normal background within species. et al., 2018). After filtering, the total number of reads per pool varied from 153 to 174 million, the mean length and insert size peak being 141-143 and 150 bp, respectively (Table S1).
To consider potential effects of reference bias on the results, we performed the analyses using both D. flavomontana and D. montana chromosome-level reference genomes . The genomes cover most regions for all the chromosomes, except for the with read group information ). The alignments were sorted with samtools version 1.10 ) and PCR duplicates marked with sambamba version 0.7.0 (Tarasov et al., 2015).
The separate BAM-files of each sample were merged and filtered for mapping quality of >20 using SAMtools. The mean coverage of the pools varied from 163 to 193 based on D. flavomontana reference, and 151-204 based on D. montana reference (Table S1). Allele counts for each sample at each genomic position were obtained with SAMtools mpileup using options to exclude indels and to keep reads with a mapping quality of >20 and sites with a base quality of >15.
The resulting BAM-files were used for variant calling with the unmasked version of the reference genomes using heuristic SNP calling software poolsnp (Kapun et al., 2020). In poolsnp, we specified a minimum count of 5 to call a SNP, and a minimum coverage of 80 to reliably calculate allele frequencies and to minimize potential reference bias. For a maximum coverage, we considered positions within the 95% coverage percentile for a given sample and chromosome.
Variant calling detected a total of 4,489,437 biallelic SNPs when using D. flavomontana reference genome, and 4,407,029 biallelic SNPs when using D. montana reference genome.

| Genetic differentiation, hybrid index and the types of genetic incompatibilities
The expected amount of genetic material transferred from one species into the other halves with every backcross generation (Figure 1).
Given species-specific alleles, we can measure introgression via the hybrid index (HI), which can be defined simply as the heterospecific fraction of genome in an individual (or a pool of individuals).
Thus, in the pool of second backcross generation hybrid females, the genome-wide HI is expected to be 12.5% in the absence of BDMIs ( Figure 1). However, given the random inheritance of chromatids in gametes and the randomness of crossover locations, we expect substantial variation around the expected mean HI, even in the absence of BDMIs.
To estimate the amount of introgression in the BC 2 pools, we computed the HI in both pools along the genome based on speciesdiagnostic SNPs (variants that are differentially fixed between the parental pools). Differentially fixed SNPs were defined as SNPs with allele frequency 1 in one parental pool and 0 in the other one (1 = all reads supporting the alternate allele, 0 = all reads supporting the reference allele). The total number of SNPs that were differentially fixed between the parental species was 1,668,294 when using D. flavomontana reference genome, and 1,570,556 when using D. montana reference genome. For each differentially fixed SNP between the species, allele frequencies were calculated by dividing "alternate read depth (AD)" by "the total read depth (DP)". To enable comparison between backcross directions, the allele frequencies for nonreference alleles were calculated with the formula "1 -allele frequency" (e.g., allele frequency of 87.5% would become 12.5%). Finally, given that a maximum allele frequency for a SNP in a hybrid is 0.5, any SNPs with an allele frequency over 0.5 were discarded (78 out  Using the diagnostic SNPs, we calculated the mean HI and its standard deviation separately for different chromosomes for BC 2 fla and BC 2 mon pools. We also calculated the number of SNPs without any introgressed material (HI = 0%) separately for each chromosome for both pools. Finally, we plotted HI in nonoverlapping windows of 400 SNPs for each chromosome and BC 2 pool using a custom script (https://github.com/vihoi kka/SNP_mappe r/blob/main/datas mooth er2.py). In principle, crossover (CO) events involving the two ancestral backgrounds (Fisher junctions;Fisher, 1954) should be visible as step changes in the HI of each pool. Assuming on average one CO per chromosome and female meiosis, the expected number of CO events per chromosome generated during the experiment is given by the total number of females (nBC 1 + nBC 2 ; Table S3)  In practice, however, the resolution especially for the junctions that are unique to a single BC 2 individual (which correspond to a change in allele frequency of 1/82) is limited by the randomness in sequencing coverage of the pool.
Given that this experiment was started with a single-pair cross between the parental species and continued with repeated backcrosses between hybrid females and parental males, all backcross individuals inherited a maximum of one allele per locus from the donor species (Figure 1). Thus, the genomes of BC individuals are a mosaic of two types of tracts: (i) homozygous for the genetic background of the recipient species or (ii) heterozygous between species. This limits the types of BDMIs that can be expressed ( Table 1). Dominantdominant pairwise BDMIs arise already in the F 1 generation and, if severe, can cause sterility/inviability in both sexes. Recessiverecessive pairwise BDMIs cannot be detected in our experiment even if they were X-linked since (i) all BC individuals involved in the experiment were females (no hemizygosity), and (ii) the expression of these incompatibilities would require homozygous tracts for both species (Figure 1). Hence, dominant-recessive BDMIs are the only strong postzygotic barriers that we expect to detect in this study.

| Simulating the backcross and resequence experiment
Given the stochastic nature of inheritance of chromatids in gametes and the randomness of crossover locations in meiosis, we expected   (Table S3). We also assumed one crossover per female per chromosome in meiosis (a map length of 50 cM).
Given that the experiment involves two generations of crosses between hybrid females and pure parental males, our simulation only tracks the haplotype of female gametes contributing to BC 1 and BC 2 individuals. All in silico backcross experiments were simulated, separately for each chromosome, 10,000 times to obtain 5% and 95% quantiles for the mean HI.  Table S2), that is, we did not attempt to include interchromosomal effects (SIM2, Figure 2b). Third, we simulated the experiment under a model that assumes a single BDMI at a random position within the inverted part of the chromosome (SIM3, Figure 2c). This single locus cannot be introgressed beyond the F 1 generation, that is, BC 1 and BC 2 females that are heterozygous for this locus are not produced. Note that while we refer to this as a BDMI for simplicity, we did not explicitly simulate pairwise incompatibilities. Thus, this locus can be regarded as a BDMI involving a dominant allele on the introgressing background (donor species) that is incompatible with one or more recessive alleles in the recipient background.

| BC 1 females from the backcrosses towards D. flavomontana showed stronger genetic incompatibilities/postzygotic isolation than the ones from the backcrosses towards D. montana
In BC 1 generation, the proportion of fertile females was 75% and

| Genetic divergence between D. montana and D. flavomontana has accumulated within inverted chromosome regions especially on the X chromosome
We performed all genomic analyses using both D. flavomontana and D. montana reference genomes to be able to evaluate the potential effect of reference bias on the results. Here, we focus mainly on analyses that use D. flavomontana as a reference genome, since the backcrosses towards D. flavomontana showed more evidence for incompatibilities than the ones towards D. montana. Results based on the D. montana reference genome are also discussed here, but the corresponding figures and tables are given in Appendix S1.
Irrespective of which species was used as a reference genome, the density of SNPs that were differentially fixed between D. montana and D. flavomontana parental pools was higher on the X chromosome than on any of the autosomes (p < .001; Figure 3; Figure S3; Table S4). Moreover, the density of fixed differences was higher in inverted compared to the colinear regions within each chromosome containing inversions (p < .001; Figure 3; Figure S3; Table S5), as expected due to the reduction in recombination within inverted regions (note that chromosomes 2R and 3 have no inversions).

| Large differences in HI between chromosomes -Evidence for BDMIs located within X chromosomal inversions
The mean amount of introgression (hybrid index, HI) of hybrids backcrossed to D. montana (BC 2 mon) did not deviate significantly from the neutral expectation of 12.5% for any chromosome (SIM1). This was true irrespective of whether the reference genome of D. flavomontana (Figures 4 and 5a, Figure S4; Table S6) or D. montana ( Figures S5, S6, S7A, Table S6) was used. Moreover, in both analyses, the fraction of diagnostic SNPs that showed no introgression (HI = 0 in the BC 2 mon pool) was low (0.02%-0.20% and 0.03%-0.29% depending on whether the D. flavomontana or D. montana genome was used as a reference), across the entire genome (Table S6).
In contrast, BC 2 fla hybrids showed a significant reduction in mean HI compared to the neutral scenario (SIM1) for the fourth and the X chromosome, and these results were again robust to the choice of reference genome (D. flavomontana genome: Figures 4 and 5b, Figure S4, Table S6; D. montana genome: Figures S5, S6, Table S6). Interestingly, and irrespective of which reference genome was used, the reduced introgression on the fourth chro-  (Table S6). For the autosomes, the fraction of diagnostic SNPs that showed no introgression into D. flavomontana varied from 0.14% to 2.58%, depending on the chromosome and the choice of reference genome (Table S6).  Figure S4). However, when we used D. montana as a reference genome (which probably overestimates introgression of D. montana alleles into the BC 2 fla pool), BC 2 fla hybrids showed a significant increase in mean HI relative to the neutral scenario (SIM1) for both chromosomes ( Figures S5, S6, S7b). In this case, we find that the increase in introgression on the fifth chromosome was compatible with a reduction in crossover rate due to the inversion present on this chromosome, without invoking any selection acting on incompatibilities (SIM2; Figure S7e). In contrast, the mean estimated HI in BC 2 fla hybrids for chromosome 3 (which has no known inversion differences between the two species) was not compatible with any of the simple scenarios we simulated. Given that we have either assumed neutrality or a single dominant incompatibility locus, which is maximally deleterious, this is perhaps unsurprising (Section 4).

| Postzygotic barriers between D. montana and D. flavomontana show asymmetry in their strength
We have previously shown that pre-and postzygotic barriers be- fixed inversions, the inversions suppress recombination, and this suppression of recombination facilitates reproductive isolation (Faria & Navarro, 2010). D. montana populations on different continents are known to have a high number of fixed and polymorphic inversions (Morales-Hojas et al., 2007;Throckmorton, 1982), while there is less data on D. flavomontana inversions (Throckmorton, 1982). Using long-and short-read genomic data, we have recently identified several alternatively fixed inversions in D. montana and D. flavomontana across species' distribution in North America, and shown that these inversions have increased genetic divergence and lower historical introgression compared to colinear chromosome regions . In the present study, we show that these inversions have an increased number of alternatively fixed SNPs compared to colinear regions, which is in agreement with their increased genetic divergence shown in Poikela et al. (2022). We have also shown that large swathes of species-specific ancestry are retained within inverted chromosome regions (Figure 4), which suggests that inversions effectively suppress recombination in early backcross hybrids.
Finally, we find that the drastic reduction in introgression on the X chromosome can be explained by inversions that are associated with at least one dominant X chromosomal D. montana incompatibility allele interacting negatively with recessive autosomal D. flavomontana alleles. This negative epistatic interaction could cause the observed low hybrid fertility, and supports the idea that inversions act as strong barriers to gene flow by facilitating the establishment of BDMIs (Hoffmann & Rieseberg, 2008;Navarro & Barton, 2003;Noor et al., 2001).
While the involvement of the X chromosome in hybrid problems may not be surprising (see e.g., Masly & Presgraves, 2007;Tao et al., 2003), the fact that it involves a dominant incompatibility locus is. The "dominance theory" (Turelli & Orr, 1995, 2000, which aims to explain the disproportionate role of the X chromosome in hybrid incompatibilities, relies on the presence of recessive incompatibilities on the X and therefore cannot explain our result. However, the "dominance theory", as well as the "faster-male theory" and dosage compensation (reviewed in Coyne, 2018;Presgraves, 2008), can still explain the hybrid male sterility previously observed in crosses between D. flavomontana and D. montana (Poikela et al., 2019).
Accumulation of meiotic drive elements on the X chromosome could be another plausible explanation for the large X-effect in general (reviewed in Patten, 2018), but this is unlikely in our system as the meiotic drive systems described in Drosophila are typically involved in sperm killing and not in female sterility (Courret et al., 2019). Although cytoplasmic incompatibilities have been detected in other montana complex species of the Drosophila virilis group (Patterson, 1952;Throckmorton, 1982), they are not likely to play a major role in these crosses since all hybrids had D. flavomontana cytoplasm (and crosses were more unsuccessful in this direction). Finally, the large X-effect we detected in the present study could potentially be explained by "faster X evolution", based on the idea that selection increases the frequency of advantageous recessive alleles more effectively on the X chromosome than on autosomes, irrespectively of whether the incompatibilities themselves are recessive (Charlesworth et al., 1987(Charlesworth et al., , 2018. Also, the X chromosome could simply contain more genes that are prone to create postzygotic isolation than those on the autosomes (Coyne, 2018).
Several autosomes showed deviations from the expected hybrid indices in the BC 2 fla pool. Based on our simulations, the reduced introgression on the fourth chromosome could be explained by inversions' ability to restrict recombination which increases the variance in chromosome-wide HI. However, if we calculate the expected allele frequencies for a dominant-recessive BDMI by hand for the first two backcross generations, the allele frequencies (i.e., HI) after selection would be 1/22 (4.5%) for the dominant and 2/11 (18.2%) for the recessive D. montana allele in the BC 2 fla pool ( Figure S8).
These frequencies are close to the observed frequencies for example, on chromosomes 4 (4.6%) and 5 (17.5%), respectively. It is therefore tempting to speculate that pairwise BDMI loci could exist on these chromosomes. Finally, chromosomes 3 and 5 showed increased introgression in the BC 2 fla pool, but only in analyses using D. montana as a reference. This effect may be due to an overestimation of D. montana alleles in the BC 2 fla pool (i.e., reference bias).
Alternatively, the increased introgression on fifth chromosome could be explained by inversions' ability to restrict recombination, increasing the variance in chromosome-wide HI. However, the drastic increase in introgression on the third chromosome, which lacks species-specific inversions, was not explained by any of our simulations. We note that our simulations did not consider an interchromosomal effect, where inversions may trigger an increase in recombination on other freely recombining chromosomes (Crown et al., 2018;Stevison et al., 2011). However, this would only decrease the variance in HI on chromosomes lacking fixed inversions and, and thus it cannot explain the increase in HI for chromosome 3 in the BC 2 fla pool.
In future research, combining the crosses with quantitative trait loci (QTL) analyses might help to link BDMIs to for example, specific genes (Johnson, 2010), gene duplicates or transposons (Bikard et al., 2009;Masly et al., 2006). BDMI genes could also be searched by tracing whole-genome gene expression data in interspecific hybrids (Satokangas et al., 2020). However, recombination suppression of inversions presents a challenge for mapping BDMIs, and would in theory require a complex reversion of the X chromosomal inversions with genome editing tools, and repeating the current experiment to narrow down the regions of reduced introgression (Hopkins et al., 2020). Overall, finding the exact loci driving species' isolation may be difficult, as BDMIs are often complex and coevolve with rapidly evolving heterochromatic DNA (Satyaki et al., 2014).

| CON CLUS IONS
"Introgress-and-resequence" studies that combine interspecific backcrosses with genome-wide analyses and simulations are an effective approach for identifying BDMIs, in particular those involving dominant alleles. Our study supports the idea that inversions aid the accumulation of BDMIs due to reduced recombination, and shows that strong BDMIs coupled with suppressed recombination effectively restrict introgression beyond the inverted part of the genome in the first two backcross generations. We conclude that the large X-effect we observed in our experiment may result from at least one dominant incompatibility locus residing within several overlapping inversions. If the design were extended to study interspecific F 2 hybrids, assuming that the F 1 female and male hybrids are viable and fertile, one could investigate recessive-recessive BDMIs in the same way. Overall, we provide a novel framework for investigating the role of inversions and the X chromosome as genetic barriers to introgression, which we hope will encourage similar studies on a larger number of species and strains.

ACK N OWLED G EM ENTS
This work was supported by grants from the Academy of Finland project 267244 to AH and projects 268214 and 322980 to MK, as well as a grant from Emil Aaltonen to NP. KL was supported by a

Natural Environmental Research Council (NERC) UK Independent
Research fellowship (NE/L011522/1). We thank Anna-Lotta Hiillos for performing DNA extractions and Ville Hoikkala for his help with data analysis. Figure 1 was created with BioRe nder.com.

CO N FLI C T O F I NTE R E S T
The authors declare no conflict of interest.

DATA AVA I L A B I L I T Y S TAT E M E N T
Raw sequence reads have been deposited in the SRA (BioProject PRJNA895210). Other data (phenotypic and allele frequency data, reference genomes for both species, Mathematica notebooks including simulations, and Unix and R commands) are available on Dryad (https://doi.org/10.5061/dryad.4f4qr fjft).

B EN EFIT-S H A R I N G S TATEM ENT
Benefits generated: Benefits from this research accrue from the sharing of our data and results on public databases as described above.