The modern evolutionary synthesis in the early 20th century resulted in competing models describing the relative distribution of effect sizes of the mutations fixed during adaptation . Despite major advances in the field of genetics, there is still uncertainty about the number of genes that contribute to quantitative trait variation and the effect sizes of these genes , limiting our understanding of the genetic basis of adaptation. The recent widespread use of marker-assisted genetic mapping studies provides the opportunity to compare the predictions of theoretical models with empirically estimated genetic architectures [3–6].
Fisher's infinitesimal model, one of the first and most influential theories of the genetic basis of adaptation, assumes that each locus exerts a minute effect on an organism's phenotype, and that phenotypic change occurs via the fixation of beneficial small-effect mutations at these loci , with negative pleiotropic effects constraining effect sizes of fixed mutations [5, 7, 8]. In contrast, Fisher's geometric model and subsequent work from other researchers has shown that mutations of large effect are less likely to be lost due to drift than small-effect mutations, and thus, on average, we should predict mutations of intermediate effect during an adaptive walk . In this model, mutations of large effect are more likely to be fixed early on in a bout of adaptation, when a population is far from an adaptive peak, with smaller-effect mutations that act to fine-tune the phenotype to an adaptive optimum predominating during later stages of adaptation [9–11].
These models assume directional selection moves a population up a static solitary peak in an adaptive landscape, disregarding that many adaptive landscapes harbor multiple or shifting phenotypic optima [5, 9, 12, 13]. Adaptation models predict that fixation of a large-effect mutation is more likely in populations far away from an adaptive peak. In a shifting adaptive landscape harboring multiple phenotypic optima, populations are likely to spend more time farther away from an adaptive peak, because, while selection drives them toward a phenotypic optimum, this optimum continually shifts, moving them into adaptive valleys [14–16]. Therefore, when there are continual shifts in selective pressure, mutations acting in opposite directions and mutations of large effect should be more common. Mutations that act in opposite directions include cases where one mutation acts to increase a trait value, and the other decreases the trait value. While a large number of mutations acting in one direction (for example, all mutations act to increase a trait value) indicates consistent selective pressure , temporal or spatial variation in selective pressure will not consistently favor mutations acting in one direction, and will thus result in a large number of mutations acting in opposite directions. Similarly, in the latter case, mutations of large effect should be more common because shifts in phenotypic optima will repeatedly move populations farther from an adaptive peak, thus favoring the fixation of large-effect mutations [14–16]. Similarly, if there are a variety of phenotypic optima in the adaptive landscape separated by valleys of low fitness, large jumps in phenotype will be favored because they will allow populations to quickly cross these valleys. The initial spectrum of newly- arisen mutations' fitness effects are unlikely to vary systematically across trait types or types of adaptive landscapes. However, the fixation of mutations of large effect should be more common in the types of adaptive landscapes described above, those that shift temporally or exhibit deep valleys of low fitness between peaks.
In order to test these hypotheses, we compare the genetic architecture of traits putatively under abiotic and biotic selection. The strength and spatial configuration of selection driven by biotic interactions can change rapidly because they are contingent on the distributions and evolutionary trajectories of other organisms [17–19]. A variety of systems show strongly patchy spatial distributions of interacting species as well as temporal fluctuations in interacting species' distributions or densities, which can result in high spatial and temporal variation in the form or strength of selection [20–24]. Similarly, evolution in interacting species' populations can lead to variation in selection pressures [18, 19]. To our knowledge, no studies have explicitly compared temporal or spatial variability in selection pressure by biotic and abiotic forces. However, traits under biotic selection, because they are subject to the stochastic processes that influence interacting species' distributions and abundances, are hypothesized to be more temporally and spatially variable compared to traits under abiotic selection [9, 18]. Conversely, traits under abiotic selection, such as selection imposed by environmental gradients, are not affected by spatial or temporal fluctuations in other species' densities; thus, these traits should experience an adaptive landscape with less temporal variability than biotic pressures. Additionally, traits under biotic selection may experience a landscape with more adaptive peaks than those under abiotic selection; for example, traits under this type of selection could confer adaptations to one of several potential pollinator species, or one of a series of discrete anti-herbivore compound types. We predict that the hypothesized differences in the adaptive landscape of traits under biotic selection versus abiotic selection will lead to the differences in genetic architecture outlined above between these two types of traits.
With data from quantitative trait loci (QTL) mapping studies, which have been conducted on a variety of traits and organisms, we now have the information to conduct more detailed comparative studies of the genetic architecture for different types of traits. In brief, QTL mapping locates putative genes underlying a quantitative trait within a genome using a recombinant population derived from individuals that differ in both phenotype for the trait of interest and genotype at genetic markers of known linkage. In these phenotypically and genotypically variable hybrid offspring, researchers associate genotypic differences with variation in phenotype in order to pinpoint which genomic regions are associated with the measured traits. Each detected QTL comprises a gene or a set of linked genes, and the location of each QTL corresponds to the location of this gene or set of genes. Similarly, the relative effect size of a given QTL and the direction of action of each QTL is a net sum of the action of the genes located within that QTL.
Three pieces of information from QTL mapping studies can be used to infer differences in genetic architectures: (1) the direction of a given QTL, indicated by whether the allelic effects at a QTL are in line with or oppose the difference in phenotypic trait values of the parents, (2) the percentage of the parents' phenotypic variation in the trait (percent variance explained; PVE) associated with the QTL in question; and (3) the total number of QTL detected. Unfortunately, we cannot compare the total number of QTL detected due to differences in power across studies that affect the probability of detecting small-effect QTL. In contrast, the QTL sign test, which uses QTL effect directions, may be used to evaluate whether a trait evolved under consistent directional selection , because directional selection will limit the fixation of antagonistic QTL (QTL with effects that are in the opposite direction to parental differences for those traits). A recent review validated the use of the QTL sign test for inferring historically consistent directional selection, finding fewer antagonistic QTL in traits known to be subject to strong directional selection . Finally, the PVE, or effect size, of a QTL provides an estimate of how much an allelic substitution at a QTL contributes to the parental difference for those traits. Though the vast majority of QTL mapping studies measure and report this metric for all detected QTL, effect sizes of QTL have not yet been compared across trait types.
We reviewed information from QTL studies on a variety of traits in natural and laboratory systems to evaluate differences in genetic architectures across trait types, and our study is unique in considering both QTL direction and PVE. Our particular interest is in comparing the genetic architecture of traits putatively under abiotic selection to that of traits putatively under biotic selection. We first quantify the direction of allelic affects of QTL controlling a given trait; this analysis verifies that directional selection, upon which all competing models of genetic architectures are based, is indeed occurring. This test also compares the historical consistency of directional selection in traits under abiotic versus biotic selection ; we hypothesize that directional selection is common in both types of traits, but that traits under biotic selection have a higher proportion of antagonistic QTL because in a rapidly shifting adaptive landscape experienced by traits under biotic selection, QTL operating to drive the population up one adaptive peak might be rapidly rendered maladaptive as the adaptive landscape shifts. In contrast, selection via abiotic factors may be less likely to vary so rapidly and unpredictably. Second, we compare QTL effect size distributions for different types of traits. We hypothesize that large effect QTL will be more common in traits under biotic selection, as compared to traits under abiotic selection, due to temporal variability in the adaptive landscape and many phenotypic optima. This finding would support the geometric model of adaptation, which predicts that large-effect QTL become fixed more often when populations are farther from an adaptive peak [9–11], as they are in a shifting adaptive landscape.