A few years back, Marius Roesti and I started to work extensively on the genomics of adaptive divergence between lake and stream stickleback population pairs from Canada, using genome-wide marker data sets generated by restriction site-associated DNA sequencing (RADseq). Based on this first genomics experience (see http://onlinelibrary.wiley.com/doi/10.1111/j.1365-294X.2012.05509.x/abstract), we agreed that there were two main aspects we wanted to improve in subsequent population genomic investigations. First, we felt that higher marker resolution was needed, because our initial RADseq resolution (based on the standard Sbf1 restriction enzyme) seemed to capture the molecular consequences of divergent selection only on a relatively crude scale. Second, we imagined that insights into parallelism in the genomic basis of adaptive divergence would be easier to obtain by investigating a study system exhibiting parallel evolution at a smaller and thus more clear-cut geographic scale. With these ideas in mind, we decided to start an investigation on lake and stream stickleback within a single watershed, the Lake Constance basin in Central Europe (Fig. A), using higher-resolution RAD methodology. We considered three populations from well-separated inlet creeks to Lake Constance (one of Europe's largest lakes), as well as the lake population itself sampled at two distant sites. The latter proved to be panmictic, so in the end we believed we were dealing with three stream populations, each diverged independently and in parallel from a shared lake ancestor.
Fig. A. Stream stickleback from the Lake Constance basin in their natural habitat. Photo credit Marius Roesti.
Based on this perspective, my group and I started to do population genomic analyses, but somehow the results did not seem to make sense and came with many surprises. For instance, we observed that genetic variation was lower in the lake than in the stream populations, despite the huge number of stickleback that must be living in the large lake. Also, the highest genome-wide differentiation emerged from a lake-stream contrast and not from a comparison of the geographically isolated streams. This was unexpected because the independent colonization of the streams by founders from the lake should have promoted differentiation among the stream populations at neutral markers. Moreover, and in a phylogenetic tree, the lake population was nested within the stream samples. Finally, inspecting genetic linkage on a genome-wide scale and haplotype structure around single genomic loci under selection revealed that the lake population has been influenced by selection more severely than the stream populations. We thus ended up with an evolutionary scenario we had completely overlooked in the beginning: the lake population must have adapted to its environment after the stream populations formed, and variation among the stream populations in the magnitude of divergence from the lake population primarily reflects to what extent genetic material from the lake population manages to introgress into the streams. We feel this scenario is well captured by the idea of ‘ecological vicariance’, that is, the ecological (as opposed to purely geographical) fragmentation of an initially widespread population (Fig. B).
Fig. B. Ecological vicariance leading to apparent parallel evolution. This process is initiated by multiple habitats becoming colonized by a shared ancestor (in our case a stream-adapted population) (top panel). Next, the connectivity among populations becomes constrained as the core population adapts to its ecologically distinct habitat (in our case the lake; the peripheral circles are stream habitats) (middle). Nevertheless, this ecologically-based reproductive isolation is not complete, allowing for introgression across habitat boundaries (bottom). Depending on asymmetries in population sizes, this introgression might primarily affect the peripheral populations. In our case, the result is variation among multiple stream populations in the magnitude of erosion of the ancestral state (shown by gray shades), mimicking variable progress in parallel evolution among the stream populations.
Hence, the Lake Constance system is appropriate for investigating divergent selection, but inappropriate for studying parallel evolution, because the stream fish (initially considered derived) reflect, to a greater or lesser extent, an ancestral state pre-dating the emergence of the derived lake population. What we learned from this work is that caution is warranted when developing evolutionary narratives in genomics; assumptions should be tested, requiring the combination of extensive analyses including those of haplotype structure around selected loci. If you are interested (there is additional stuff on adaptive chromosomal inversions), check out The genomics of ecological vicariance in threespine stickleback fish.
Roesti M, Kueng B, Moser D, Berner D (2015) The genomics of ecological vicariance in threespine stickleback fish. Nature Communications, DOI: 10.1038/ncomms9767.
Roesti M, Hendry AP, Salzburger W, Berner D (2012) Genome divergence during evolutionary diversification as revealed in replicate lake-stream stickleback population pairs. Molecular Ecology, 21: 2852-2862. DOI: 10.1111/j.1365-294X.2012.05509.x.