Parallel evolution of similar traits by independent populations in similar environments is considered strong evidence for adaptation by natural selection. Often, however, replicate populations in similar environments do not all evolve in the same way, thus deviating from any single, predominant outcome of evolution. This variation might arise from non-adaptive, population-specific effects of genetic drift, gene flow or limited genetic variation. Alternatively, these deviations from parallel evolution might also reflect predictable adaptation to cryptic environmental heterogeneity within discrete habitat categories. Here, we show that deviations from parallel evolution are the consequence of environmental variation within habitats combined with variation in gene flow. Threespine stickleback (Gasterosteus aculeatus) in adjoining lake and stream habitats (a lake-stream 'pair') diverge phenotypically, yet the direction and magnitude of this divergence is not always fully parallel among 16 replicate pairs. We found that the multivariate direction of lake-stream morphological divergence was less parallel between pairs whose environmental differences were less parallel. Thus, environmental heterogeneity among lake-stream pairs contributes to deviations from parallel evolution. Additionally, likely genomic targets of selection were more parallel between environmentally more similar pairs. In contrast, variation in the magnitude of lake-stream divergence (independent of direction) was better explained by differences in lake-stream gene flow; pairs with greater lake-stream gene flow were less morphologically diverged. Thus, both adaptive and non-adaptive processes work concurrently to generate a continuum of parallel evolution across lake-stream stickleback population pairs.

Evolutionary ecologists aim to explain and predict evolutionary change under different selective regimes. Theory suggests that such evolutionary prediction should be more difficult for biomechanical systems in which different trait combinations generate the same functional output: "many-to-one mapping." Many-to-one mapping of phenotype to function enables multiple morphological solutions to meet the same adaptive challenges. Therefore, many-to-one mapping should undermine parallel morphological evolution, and hence evolutionary predictability, even when selection pressures are shared among populations. Studying 16 replicate pairs of lake- and stream-adapted threespine stickleback (Gasterosteus aculeatus), we quantified three parts of the teleost feeding apparatus and used biomechanical models to calculate their expected functional outputs. The three feeding structures differed in their form-to-function relationship from one-to-one (lower jaw lever ratio) to increasingly many-to-one (buccal suction index, opercular 4-bar linkage). We tested for (1) weaker linear correlations between phenotype and calculated function, and (2) less parallel evolution across lake-stream pairs, in the many-to-one systems relative to the one-to-one system. We confirm both predictions, thus supporting the theoretical expectation that increasing many-to-one mapping undermines parallel evolution. Therefore, sole consideration of morphological variation within and among populations might not serve as a proxy for functional variation when multiple adaptive trait combinations exist.

The Clever Foraging Hypothesis asserts that organisms living in a more spatially complex environment will have a greater neurological capacity for cognitive processes related to spatial memory, navigation, and foraging. Because the telencephalon is often associated with spatial memory and navigation tasks, this hypothesis predicts a positive association between telencephalon size and environmental complexity. The association between habitat complexity and brain size has been supported by comparative studies across multiple species but has not been widely studied at the within‐species level. We tested for covariation between environmental complexity and neuroanatomy of threespine stickleback (Gasterosteus aculeatus) collected from 15 pairs of lakes and their parapatric streams on Vancouver Island. In most pairs, neuroanatomy differed between the adjoining lake and stream populations. However, the magnitude and direction of this difference were inconsistent between watersheds and did not covary strongly with measures of within‐site environmental heterogeneity. Overall, we find weak support for the Clever Foraging Hypothesis in our study.

No abstract