Genomic signatures of reproductive isolation are decoupled from floral divergence in a long-standing hybrid zone

A central goal in evolutionary biology is to understand how species boundaries are maintained in the face of gene flow. While gene flow between species often accompanies the formation of hybrids, the genome-wide effects of hybridization depend on the presence and nature of reproductive isolating barriers. Two North American wildflower species with divergent floral syndromes, Penstemon davidsonii (bee syndrome) and P. newberryi (bird syndrome), have formed hybrid zones in the eastern Sierra Nevada for at least 85 years. Despite decades of hybridization, each species appears to have maintained phenotypic integrity while forming phenotypically intermediate hybrids in narrow elevational bands. We combined quantitative trait analysis, pollinator visual modeling, genome-wide association studies (GWAS), and hierarchical Bayesian genomic cline analysis to ask whether divergent floral syndromes enforce reproductive isolation in this species pair as predicted by classic models of pollinator-driven ecological speciation. We found that the two parent species exhibit strong divergence across multivariate trait space and have maintained genomic differentiation despite persistent hybridization. The genetic architecture of floral hue, a key component of pollination syndrome, is concentrated in a single genomic region containing two strong candidate genes with large effects on anthocyanin pigment composition. Pollinator visual models indicated that genetic variation at this region has a large effect on detectability to hummingbirds, but no effect on detectability to bees. Genomic cline analyses identified many significantly steep clines across the genome, suggesting a polygenic basis to reproductive isolation. Surprisingly, these barriers appear unrelated to floral isolation; in particular, the major floral hue locus exhibits a strikingly shallow genomic cline and elevated heterozygosity, suggesting pervasive gene flow across hybrid classes. Our findings highlight the complicated relationship between multivariate phenotypes and reproductive isolation and emphasize the need to map both trait loci and putative reproductive barriers separately to identify true barrier loci.