Given a recent time of separation and modest effective size for the ancestral population, mtDNA should have coalesced quickly and should display minimal differentiation between Mexican and U.S. flies. Last, although inverted regions can be biased toward containing haplotypes with deeper RNDs, unless population splitting was precise, one would still expect to see a subset of inversions shared in common between Mexican and U.S. flies. Haplotypes in the shared inversions should show shallow RNDs, similar to loci on chromosomes 4 and 5. Consequently, the observed gene trees are more consistent with the hypothesis of repeated isolation and secondary contact, with inversions on chromosomes 1–3 becoming increasing more recalcitrant to introgression through time relative to collinear regions of the genome.

Our data could also be explained by a series of gene duplication and deletion events within R. pomonella and the outgroup species R. electromorpha such that many of the haplotype comparisons made in the study were between paralogous rather than orthologous sequences. Four of the original 19 loci amplified in the study were found to be duplicate loci. If similar duplications were accompanied by deletions for many of the other 15 loci, then these duplications/deletions could confound our biogeographic interpretation of the gene trees. However, the deletion scenario, considered alone, suffers the same difficulties as the lineage-sorting hypotheses in explaining the tripartite distribution and deep congruence of chromosome 1–3 nuclear and mtDNA RND values. But it is possible that a composite biogeography/deletion model could account for the pattern. Under this scenario, Mexican and Northern isolates formed ≈1.57 Mya. A period of secondary contact and gene flow followed from 0.5–1 Mya. After this time, Altiplano and Northern populations have remained disjunct. The shallow RNDS observed for loci not on chromosomes 1–3 would be due reciprocal deletions of paralogous genes in R. pomonella and R. electromorpha, resulting in improper comparisons of orthologous Mexican and U.S. haplotypes within R. pomonella to a highly diverged paralogous outgroup sequence for R. electromorpha.

The deletion hypothesis would not negate the contributory roles of allopatry and secondary introgression in facilitating the sympatric radiation of the R. pomonella group by means of host shifting. However, it would call into question whether gene flow was differential for inverted vs. collinear regions of the genome. In essence, there would not have been a second period of recent contact when such a pattern could have been fully generated. Genetic crosses of flies imply that U.S. haplotypes represent allelic variation segregating at single loci (Feder et al., 2003b; Roethele et al., 2001). However, it is difficult to completely rule out the possibility that deletions at very tightly linked duplicated loci generated the observed segregation patterns. Moreover, test cross results for R. pomonella