Divergent sorting of a balanced ancestral polymorphism underlies the establishment of gene-flow barriers in Capsella

In the Bateson–Dobzhansky–Muller model of genetic incompatibilities post-zygotic gene-flow barriers arise by fixation of novel alleles at interacting loci in separated populations. Many such incompatibilities are polymorphic in plants, implying an important role for genetic drift or balancing selection in their origin and evolution. Here we show that NPR1 and RPP5 loci cause a genetic incompatibility between the incipient species Capsella grandiflora and C. rubella, and the more distantly related C. rubella and C. orientalis. The incompatible RPP5 allele results from a mutation in C. rubella, while the incompatible NPR1 allele is frequent in the ancestral C. grandiflora. Compatible and incompatible NPR1 haplotypes are maintained by balancing selection in C. grandiflora, and were divergently sorted into the derived C. rubella and C. orientalis. Thus, by maintaining differentiated alleles at high frequencies, balancing selection on ancestral polymorphisms can facilitate establishing gene-flow barriers between derived populations through lineage sorting of the alternative alleles.

a Table of additional crosses performed and proportion of incompatible phenotypes in resulting F2 populations. Blue font indicates crosses in which no RPP5 was segregating. Note that the absence of incompatible phenotypes in crosses 4 and 12 is not informative, as the C. grandiflora parental plants of these crosses did not pass on an NPR1 go allele to the respective F1 plants. This is due to the C. grandiflora accession Cg926 being heterozygous for NPR1 go /NPR1 rub (see Fig. 3a).  The additive effect was calculated as half of the difference of the phenotypic averages between Cr and Cg homozygote genotypes derived from the coding scheme +1 for Cr and -1 for Cg.

Genetic incompatibilities in the genus Capsella.
The genetic incompatibility in the Cr1504 x Cg926 RIL population suggested that the presence within the same cells of RPP5 and NPR1 go induces an autoimmune response. If true, any crosses combining a functional RPP5 protein with NPR1 go should recreate the stunted growth phenotype. We tested this hypothesis by performing random crosses between seven C. rubella and nine C. grandiflora accessions (Supplementary Fig. 7). A phenotype similar to the Cr1504 x Cg926 incompatible hybrids was segregating in 13 of the resulting F 2 populations.
The RPP5 alleles from three of the C. rubella accessions used for the crosses (Cr4.23, Cr86IT1-C and Cr1377/5) belong to different haplotype groups than Cr1504RPP5 ( Supplementary Fig. 5b), while no RPP5 could be detected in a fourth C. rubella accession (Cr 1GR1-TS1). Among the C. grandiflora x C. rubella F 2 progenies descending from these C. rubella accessions, we did not observe any correlation between the RPP5 genotype (as determined by a tightly linked molecular marker HiG2 to be able to assess the genotype at the RPP5 locus also in progeny of Cr 1GR1-TS1) and the stunted growth. Similarly, no correlation was observed in the F 2 population descending from Cr22.5 in cross 21, whose RPP5 allele belongs to the same haplotype group as Cr1504RPP5. We note that the absence of incompatible phenotypes in cross 12, also involving Cr22.5, and in cross 4 is not informative, as the C. grandiflora parental plants of these crosses did not pass on an NPR1 go allele to the respective F1 plants. This is due to the C. grandiflora accession Cg926 being heterozygous for NPR1 go /NPR1 rub (see also Fig. 3a). In fact, a robust correlation between incompatible phenotype and RPP5 genotype was only seen in F 2 s descending from crosses between Cr1504 and diverse C. grandiflora individuals. These results indicate that only a specific RPP5 haplotype is incompatible with the NPR1 go alleles. It is therefore likely that after the divergence of C. rubella from C. grandiflora, a novel mutation in Cr1504RPP5 has rendered it incompatible with NPR1 go . This hypothesis is supported by the observation that no correlation between stunted growth phenotypes and RPP5/NPR1 genotypes was observed within natural C. grandiflora populations.
In addition, modifier alleles of the NPR1/RPP5 incompatibility appear to be segregating in C.
grandiflora. This is based on the observation that the strength and frequency of the stunted growth varies between the different F 2 s descending from Cr1504. For example, although the same NPR1 go and RPP5 haplotype are segregating in the crosses 6 and 7, only 10 % of the F 2 s show an incompatible phenotype in cross 7 versus 30% in cross 6 ( Supplementary Fig.   6A,S7), and the phenotypic severity is reduced in the cross 7 relative to the cross 6 F 2 progenies ( Supplementary Fig. 7B). Since Cr1504 is highly inbred, this indicates that modifiers affecting the penetrance and expressivity of the genetic incompatibility are segregating within C. grandiflora.
Importantly, we observed that NPR1 go haplotypes are also associated with genetic incompatibilities that do not involve RPP5 (Supplementary Fig. 7; crosses 3 and 9). In the progeny of cross 9, all plants homozygous for the NPR1 go allele showed the incompatible phenotype, while none of the plants homozygous for NPR1 rub did so, yet there was no difference in the RPP5 genotype frequency between plants with incompatible and compatible phenotypes. Similarly, in the F 2 of cross 3, in which no RPP5 presence could be detected, all but one of the 39 incompatible hybrids had at least one copy of the NPR1 go allele, and NPR1 go homozygotes were underrepresented amongst compatible plants. Individual NPR1 rub homozygotes amongst the incompatible plants were also observed in two of the crosses involving Cr1504, which is likely due to mis-scoring of the phenotype or to unrelated deleterious mutations from the C. grandiflora parent that became homozygous in the F 2 .
Given also the molecular evidence for a constitutive auto-immune response in the incompatible plants in the F 2 from cross 3 ( Supplementary Fig. 8a,b), we consider it highly likely that this incompatibility is also caused by the NPR1 go allele. In the F 2 of cross 10, the stunted growth phenotype co-segregated with the NPR1 rub allele, but not with the RPP5 genotype ( Supplementary Fig. 7). While suggestive of an involvement of NPR1 rub in the incompatibility, several incompatible plants were homozygous for NPR1 go , and further experiments will be needed to test a causal role for NPR1 rub . In any case, these results indicate that alleles at additional loci are incompatible with NPR1 go and possibly NPR1 rub ; these alleles are likely to segregate in C. grandiflora, as the incompatibilities were only observed in some, but not all crosses involving a given C. rubella accession (compare for example crosses 3, 20 and 24). Together, these results support the notion that the two NPR1 haplotype groups with their strong sequence divergence resulting from long-term balancing selection facilitate the establishment of genetic incompatibilities by mutations to interacting loci. If any such strongly divergent haplotypes were sorted into derived populations in a mutually exclusive manner, for example following genetic bottlenecks such as during independent transitions to selfing from an ancestral outbreeding species, the first step to the formation of a gene-flow barrier would already have been achieved, and only one further mutation to an interacting locus in one of the derived populations would be required for the establishment of a BDMI.