Population Genomics of Parallel Adaptation in Threespine Stickleback using Sequenced RAD Tags

Picture a small, armored fish in the Pacific Ocean, plated head to tail in bony lateral armor, built for saltwater life. Now picture the same species in a landlocked lake a few miles inland, stripped of most of that armor, with a smaller pelvic structure and physiology reconfigured for fresh water. That transformation has happened independently in dozens of lakes, over and over again since the glaciers retreated roughly ten thousand years ago. This raises a question that cuts to the heart of how evolution works: when the same ancestral fish colonizes a new freshwater lake repeatedly, is it reaching for the same genetic toolkit each time, or reinventing itself from scratch at every location? That is the question Paul Hohenlohe and colleagues set out to answer, and their answer, published in a landmark genome-wide study of threespine stickleback, is that evolution is far more repetitive at the genetic level than anyone had fully demonstrated before. The threespine stickleback, Gasterosteus aculeatus, is a near-perfect natural experiment. Large, panmictic oceanic populations that are genetically similar to one another because of extensive long-distance gene flow repeatedly colonized newly formed freshwater lakes after the last glacial retreat. Three of the focal freshwater populations in this study are inferred to be less than ten thousand years old. In a few documented cases, adaptation has happened in mere decades. Because the oceanic populations are so similar to one another, they serve as a reliable stand-in for the ancestral stock. And because freshwater populations have little or no ongoing gene flow among them, each one represents an independent run of the same evolutionary experiment. The phenotypic outcomes are strikingly consistent: freshwater stickleback repeatedly evolve reduced lateral armor plates, reduced pelvic structures, and reconfigured ion transport for low-mineral water. The question is whether that phenotypic convergence reflects convergent genetics. To find out, you need genome-wide data from wild fish, not just a handful of candidate genes. Hohenlohe and colleagues used RAD sequencing, or Restriction-site Associated DNA sequencing, to generate exactly that. The logic is elegant and scalable. Genomic DNA from each fish was cut with the restriction enzyme SbfI, producing a set of short flanking fragments, RAD tags, that are then sequenced. The published stickleback genome contains over twenty-two thousand SbfI sites across its twenty-one linkage groups. The team sequenced one hundred individual fish, twenty each from five populations: two oceanic populations, Rabbit Slough and Resurrection Bay, and three independently derived freshwater populations, Bear Paw, Boot Lake, and Mud Lake. The payoff was scale. From roughly two point one million nucleotide sites surveyed, Hohenlohe and colleagues identified forty-five thousand seven hundred eighty-nine single-nucleotide polymorphisms. That was the first high-density, whole-genome single-nucleotide polymorphism scan ever conducted in wild stickleback, and it made a chromosome-by-chromosome search for parallel selection signatures possible for the first time. The genome-wide portrait that emerged confirmed the biogeographic story and then went further. Oceanic populations harbor more genetic diversity than freshwater populations, consistent with freshwater founders passing through a bottleneck on colonization. The two oceanic populations are nearly identical to each other, with an FST, a measure of genetic differentiation running from zero, meaning identical, to one, meaning completely separated, of just zero point zero zero seven six. Comparisons between oceanic and freshwater populations show FST values typically in the range of zero point zero five to zero point one five, reflecting moderate divergence. But that average masks dramatic local variation. At specific genomic intervals, smoothed FST values exceed zero point three five, and individual peaks climb far higher in some pairwise comparisons. The genome is not diverging uniformly; it is diverging in concentrated, specific locations. The scan also picked up signatures of balancing selection at a handful of loci, a mode of natural selection that actively maintains multiple variants in a population, the opposite of a directional sweep. A striking region on Linkage Group III, spanning roughly fourteen point eight to sixteen point one megabases, shows elevated diversity and reduced between-population differentiation. It contains genes including ZEB1, two adjacent APOL genes, and inflammation-pathway genes. A second elevated-diversity region on Linkage Group XIII includes clusters of TRIM genes. These signals suggest that some genomic regions are being kept variable by selection even as others are being pushed toward fixation. The central finding is the peaks of parallel differentiation. Hohenlohe and colleagues identified nine of the most consistent and significant peaks across the genome, falling on six linkage groups — I, IV, VII, VIII, XI, and XXI — and covering roughly twelve point two megabases, about two point five percent of the genome. Of nearly forty-five thousand single-nucleotide polymorphisms tested, three hundred seven were significant at the stringent threshold the authors applied, and two hundred twenty-seven of those three hundred seven fell on just those six linkage groups. What makes these peaks remarkable is not their height but their consistency: all three independently derived freshwater populations show elevated FST at largely the same chromosomal addresses. Evolution, running the experiment three separate times, is highlighting the same pages of the same book. The peak on Linkage Group XXI is the clearest example of a parallel hard sweep, a case where the same ancestral haplotype, present at low frequency in the oceanic pool, was independently driven to high frequency in each freshwater population. All three freshwater populations are strongly diverged from the oceanic populations at this region, there is little differentiation among the freshwater populations themselves, and nucleotide diversity and heterozygosity collapse in freshwater fish at this location. That combination — high oceanic-to-freshwater FST, low among-freshwater FST, reduced diversity — is the genetic fingerprint of the same variant winning the same selective contest in three different lakes. Not every peak tells the same story. On Linkage Group II, a large peak in among-freshwater FST of zero point five one, with elevated private allele density, suggests non-parallel sweeps: different alleles or loci rising to high frequency in each freshwater population independently. Linkage Group VIII appears to contain adjacent regions with both patterns side by side. Evolution is not reading from a single script, but it is working from a remarkably short list of chapters. Several of these recurrent peaks connect directly to previously identified quantitative trait loci, genomic regions linked to measurable traits in laboratory crosses. The first peak on Linkage Group IV contains the Eda gene, or Ectodysplasin A, already known to be linked to lateral plate armor loss. Other peaks overlap or lie adjacent to quantitative trait loci on Linkage Groups VII and XXI. The RAD sequencing scan is not just confirming old results; it is also identifying novel regions showing parallel differentiation, extending the map of parallel genetic evolution beyond what laboratory crosses alone could reveal. Zooming into the biology inside the nine peaks, Hohenlohe and colleagues annotated thirty-one candidate genes: twenty-three linked to skeletal patterning and homeostasis, and eight linked to osmoregulation and osmoregulatory organ development. Genes with known roles in craniofacial and branchial-arch development, including EDA, EYA1, FBLN1, and WNT5A, fall within differentiated intervals, tying the genomic signal to evolved changes in jaw and gill architecture. Genes linked to bone density disorders in humans, including LEMD3 and LEPR, also appear, consistent with selection on bone deposition and resorption pathways. Osmoregulatory candidates such as PRL2, CA4, and ATP6V1A appear in peaks on Linkage Groups IV, VII, and XXI. Two genes, CA4 and FLT1, have documented pleiotropic roles, affecting both bone biology and ion physiology simultaneously, which may help explain why the same genomic regions keep getting targeted: one gene can address two of freshwater life's core challenges at once. What this all adds up to is a picture of evolution that is more predictable than the classical view would suggest. The soft sweep model holds that adaptation can draw from standing genetic variation already present in the ancestral population, including multiple alleles or pre-existing haplotypes that rise in frequency when selection shifts, rather than waiting for new mutations to arise. Hohenlohe and colleagues' data show that much of the parallel genetic evolution in stickleback fits this model. The oceanic populations carry, in their abundant standing variation, much of the raw material for freshwater life. When a founder population is isolated in a lake, selection pulls from that reservoir repeatedly and in a repeatable direction. The genetic response is not identical every time; some peaks show parallel hard sweeps while others show non-parallel patterns. But the overall picture is one of substantial genetic parallelism at a genome-wide scale. The candidate regions these data define now form a concrete research agenda. Each of those thirty-one genes is a target for developmental genetic work, functional validation, and fitness testing in natural populations. And the RAD sequencing approach itself generalizes to any organism with a reference genome, opening the door to genome-wide selection scans in wild populations far beyond stickleback. The broader lesson is this: when replicate populations face the same environmental challenge, they often draw from a shared genetic reservoir, and the response is far more repeatable than chance mutation alone would predict. Evolution, at least in this system, has a preferred vocabulary. This lecture was created by ennepō. Go to https://ennepo.ai to Discover, Create and Follow the latest research in your field. Read when you can. Listen when you want to.