Rapid polygenic selection generates fine spatial structure among ecological niches in a well-mixed population

Evolution by natural selection may be effective enough to allow for recurrent, rapid adaptation to distinct niche environments within a well-mixed population. For this to occur, selection must act on standing genetic variation such that mortality i.e. genetic load, is minimized while polymorphism is maintained. Selection on multiple, redundant loci of small effect provides a potentially inexpensive solution. Yet, demonstrating adaptation via redundant, polygenic selection in the wild remains extremely challenging because low per-locus effect sizes and high genetic redundancy severely reduce statistical power. One approach to facilitate identification of loci underlying polygenic selection is to harness natural replicate populations experiencing similar selection pressures that harbor high within-, yet negligible among-population genetic variation. Such populations can be found among the teleost Fundulus heteroclitus. F. heteroclitus inhabits salt marsh estuaries that are characterized by high environmental heterogeneity e.g. tidal ponds, creeks, coastal basins. Here, we sample four of these heterogeneous niches (one coastal basin and three replicate tidal ponds) at two time points from among a single, panmictic F. heteroclitus population. We identify 10,861 single nucleotide polymorphisms using a genotyping-by-sequencing approach and quantify temporal allele frequency change within, as well as spatial divergence among subpopulations residing in these niches. We find a significantly elevated number of concordant allele frequency changes among all subpopulations, suggesting ecosystem-wide adaptation to a common selection pressure. Remarkably, we also find an unexpected number of temporal allele frequency changes that generate fine-scale divergence among subpopulations, suggestive of local adaptation to distinct niche environments. Both patterns are characterized by a lack of large-effect loci yet an elevated total number of significant loci. Adaptation via redundant, polygenic selection offers a likely explanation for these patterns as well as a potential mechanism for polymorphism maintenance in the F. heteroclitus system. Author Summary Evolution by adaptation to local environmental conditions may occur more rapidly than previously thought. Recent studies show that natural selection is extremely effective when acting on, not one, but multiple genetic variants that are already present in a population. Here, we show that polygenic selection can lead to adaptation within a single generation by studying a wild, well-mixed population of mud minnows inhabiting environmentally distinct locations or niches (i.e. tidal ponds and coastal basins). We monitor allele proportions at over 10,000 genetic variants over time within a single generation and find a significant number to be changing substantially in every niche, suggestive of natural selection. We further demonstrate this genetic change to be non-random, generating mild, yet significant divergence between residents inhabiting distinct niches, indicative of local adaptation. We corroborate a previous study which discovered similar genetic divergence among niches during a different year, suggesting that local adaptation via natural selection occurs every generation. We show polygenic selection on standing genetic variation to be an effective and evolutionarily inexpensive mechanism, allowing organisms to rapidly adapt to their environments even at extremely short time scales. Our study provides valuable insights into the rate of evolution and the ability of organisms to respond to environmental change.


64
Is evolution by natural selection rampant? Does natural selection lead to adaptation on 9 167 between spring and fall collections is indicative of negligible overall change, expected for 168 populations near equilibrium. Consequently, the absence of both spatial and temporal structure 169 suggests that any signal is likely limited to a minor subset of alleles.

170
Significant allele frequency changes over time 171 Resident fish are assumed to have been exposed to their niche-specific environments 172 and associated selection pressures during summer. Any selective death or deterministic 173 emigration will therefore be reflected in significant allele frequency changes relative to the spring 174 collection.

175
The significance of temporal allele frequency change was quantified by the geometric 176 mean of p values generated from three separate significance tests (Barnard's Test,177 permutations and simulations) (Fig 2). This approach yielded 611 significant SNPs in the Basin, 178 664 in Pond 1, 571 in Pond 2 and 625 in Pond 3, each undergoing allele frequency changes 179 that are unlikely due to sampling error, random death or random emigration (geometric mean p 180 < 0.05). However, these totals narrowly exceed 543 (5% of 10,861); the expected number of 181 false positives under a uniform p value distribution. In fact, only two SNPs within Pond 1 remain 182 significant after multiple test correction (red points, Fig 2). All SNPs that were significant at an 183 FDR of 10% are also significant at the Bonferroni level and are hence displayed as the latter.

184
The absence of major temporal allele frequency changes paired with a moderately elevated 185 number of significant SNPs is suggestive of widespread allele frequency changes of small 186 effect, associated with polygenic adaptation [34,35,41].

187
To further investigate this idea, the observed number of significant SNPs is compared to 188 the expected number of false positives assuming a uniform p value distribution. Fig 3 shows

303
These patterns highlight two major insights. Firstly, large, relative allele frequency 304 changes, i.e. the joint allele frequency change of a subpopulation pair due to the shift of one 305 subpopulation relative to the other, generate large, spatial allele frequency differences ( Fig 8B).

306
Essentially, antagonistic allele frequency changes (i.e. large, relative changes) produce spatially 307 divergent SNPs, whereas parallel changes (i.e. small relative changes) rarely result in 308 significant spatial differences. Secondly, and more importantly, temporal allele frequency 309 changes are more predictive of posterior (fall) allele frequency differences (i.e. fine spatial 310 structure) than prior (spring) structure. In other words, recent, temporal allele frequency 311 changes generate fine spatial differentiation among subpopulations. dispersal, yet we find significant morphological differences among fish inhabiting distinct 325 locations within the estuary (Fig 1).

326
We find an elevated number of SNPs that undergo significant temporal allele frequency 327 changes from spring to fall, i.e. changes that are unlikely due to random death, sampling effects, 328 or other neutral processes alone (Fig 3). In addition, we find an unexpectedly high proportion of 329 allele frequency changes to be concordant among subpopulations in both magnitude and 330 direction (Fig 4). Spatial data corroborates previous findings by Wagner et al. [11], showing an 331 elevated number of significantly differentiated SNPs among interbreeding, resident 332 subpopulations within the same estuary in fall (Fig 6). Finally, we show that loci undergoing 333 significant temporal changes also exhibit high spatial differentiation, suggesting that spatial 334 structure in fall is primarily determined by allele frequency changes taking place during the 335 summer months, not by prior spatial structure in spring (Fig 7). were permitted to generate allele frequency changes, and hence, apparent spatial structure.

357
The correlation between these neutral, temporal allele frequency changes and resulting spatial 358 allele frequency differences can be seen in Fig

363
This result demonstrates the inherent relationship between recent, temporal allele 364 frequency change and current spatial structure. By constructing an equation defining posterior 365 spatial allele frequency difference as a function of prior spatial difference, and relative temporal 366 allele frequency change between two populations, the inherent relationship becomes apparent: Where the super-and subscript denote time point and population respectively. In the 371 light of equation (1)  produced for every subpopulation in order to account for possible spatial heterogeneity.

678
Simulated subpopulations sizes were 1300 and 400 for the basin and ponds respectively. These 679 estimates are in agreement with the observed subpopulation sizes in the wild based on 680 exhaustive sampling at each collection site.

681
Simulations testing spatial structure were conducted in a similar fashion. Here SNP-682 specific global populations were generated under the null hypothesis of spatial homogeneity.

683
Subpopulations were considered samples of a single, large, panmictic, global population and 684 their weighted mean used to estimate the global neutral or null allele frequency. Empirical allele 685 frequency differences among sites were compared to the distribution of apparent allele 686 frequency differences under the null and p values generated as above. Spatial simulations were 687 only conducted on fall data in order to remain agnostic to possible temporal change.