The Missing Response to Selection in the Wild

Predicting the Response to Selection, in Theory

The additive genetic variance (V_A) of traits is central to quantitative genetic studies of surveyed populations because V_A is used as a predictor of the evolutionary potential for trait change under directional selection. From V_A, we derive other parameters for comparing genetic variabilities [8]. Evolvability, for example, can be estimated by the coefficient of variation ($\text{C}{V}_A=\sqrt{V_{\text{A}}}/{\displaystyle \overline{z}}$) to evaluate the magnitude of the expected directional response to selection. Narrow-sense heritability (h² = V_A/V_P, where V_P is the phenotypic variance) is also frequently used, and evaluates the potential of a population to respond to selection. In theory, the univariate and multivariate breeder equations predict the expected change in mean trait value(s) as a result of selection ($\Delta {\displaystyle \overline{z}}$). In the univariate case, this change is proportional to trait heritability (h²) and the strength of selection (or selection differential, S), i.e., $\Delta {\displaystyle \overline{z}}=h^{2}S$ [9]. These approaches have been successfully applied in plant and animal breeding. However, precisely measuring V_A and its derived statistics, and measuring selection, is challenging, as detailed below.

Oversimplifying Genetic and Environmental Effects Might Bias Predictions

Quantitative genetic studies in surveyed populations often assume an oversimplified view of the genetic basis of traits [10]. Non-additive sources of genetic variation (e.g., epistasis [11] and dominance [12]) are rarely accounted for, and can be confounded with V_A, which leads to inflated estimates. Environmental causes of phenotypic similarity between related individuals have similar effects, but they are more frequently approximated. This is generally done by including environmental factors [13] or indirect genetic effects (e.g., maternal effects [14]) in the mixed model. Failing to account for these effects can bias estimates of V_A and therefore the prediction of the response to selection [15].

Owing to Precision Issues with Statistical Estimates, Qualitative Predictions Might Be More Realistic

The precision of quantitative genetic and selection estimates are reduced by low statistical power. For example, V_A estimates often have large confidence intervals [7]. Consequently, these approaches provide an approximate magnitude for the genetic basis of traits (e.g., small, medium, and large h²), and caution must therefore be taken with quantitative predictions or comparisons. The same caution must be taken when measuring selection. In addition, skewed phenotypic distributions can generate false estimates of directional selection [16] because the estimation of selection gradients assumes normality. Nevertheless, our opinion is that discriminating between evolutionary scenarios on the basis of qualitative differences remains possible (e.g., low vs high h²) provided that confidence intervals are discussed and power analyses are conducted.

Extrapolating Predictions Can Be Misleading as a Result of Covarying Factors

Extrapolating findings based on V_A and selection to different contexts (e.g., multiple populations, years, similar environments) is unreliable [7]. This is because these parameters are not independent from covarying factors such as environmental conditions and trait means [8]. They also vary with the method being used (e.g., sibship similarity vs animal models for V_A). Estimates are therefore associated with a unique set of conditions and cannot be generalized to other populations. Comparisons between multiple populations can still be achieved but require pedigree connections to quantify the effect of shared genes in different environments [17], or controlled conditions to homogenize environmental effects [18].

The Problem of Assigning Causality Between Selection and Genetic Change

Another limit to evaluating the expected response to selection by using h² is the rarely verified assumption that selection acts on the underlying genetic variation of traits 19, 20. Any environmental source of covariance between trait and fitness violates this assumption [20], causing biased predictions [19]. An alternative method which relaxes the assumption of causation is the Robertson–Price equation, also known as the secondary theorem of natural selection (STS [19]), $\Delta {\displaystyle \overline{z}}={\sigma }_A\left(w,z\right)$, where phenotypic change $\Delta {\displaystyle \overline{z}}$ is equal to the additive genetic covariance between the trait (z) and relative fitness (w). Although theoretically correct, its application is nevertheless subject to two limitations. First, examples of genetic variance for fitness are rare (but see [21]). Second, the STS might not always quantify a change caused by selection [19]. Using the STS to estimate phenotypic change caused by selection is misleading because the change in a trait of interest that is not under selection can be driven by its genetic correlation with any unmeasured trait under selection. This will cause genetic covariance between measured traits and fitness [19]. A possible solution is to adopt a multivariate approach to selection (genotypic selection gradient [21]). Such an approach might lead to a more accurate prediction of evolutionary change by selection on phenotypic traits.

Our interpretation of evolutionary processes and biological mechanisms is based on statistical measures (e.g., V_A). The limitations of these measures imply that it might be statistical failures, rather than biological mechanisms, that cause a discrepancy between predictions and observations of microevolutionary change. However, qualitative predictions for evolutionary scenarios are still possible. The conclusion that there is a missing response to selection might therefore be misleading if the expected response to selection is misestimated.

The Mechanisms We Know, and the Mechanisms We Do Not Know

The hypothesis that biological mechanisms might be responsible for the missing response to selection in surveyed populations is widely acknowledged [2]. Many biological mechanisms, such as phenotypic plasticity, genetic correlations, indirect genetic effects, and age-specific responses, are well known to interfere with the connection between genetic variation and the response to selection, but are not always accounted for (Box 2). In addition, fluctuating selection affects the spatial and temporal scale at which we detect selection (Box 2). Figure 1 graphically depicts the impact of each of these mechanisms on the response to selection that was predicted from the breeder’s equation (R, represented on the horizontal axis). The vertical axis depicts the departure from the baseline expectation that V_A will decrease following selection. For example, in the case of negative genetic correlations between traits under positive selection, the response to selection is expected to be constrained, and genetic variation maintained (Figure 1Bii). We present below other mechanisms to incorporate into the list of biological explanations.

Demography and Connectivity Shape the Potential of Populations to Respond to Selection

Population demography can affect V_A and selection, but this is neglected in most quantitative genetic studies of surveyed populations. For instance, a sharp decrease in population size (e.g., population founding event or collapse) reduces genetic diversity (genetic bottleneck [22]), V_A, and the response to selection (Figure 2A) 23, 24. Conversely, genetic connectivity between populations with different adaptive optima can limit local adaptation but increase or restore genetic variability and adaptive potential 25, 26. In addition, a negative correlation is expected between the population growth rate and the variation in relative fitness (the opportunity for selection) [27]. Thus, declining populations are expected to have a higher opportunity for selection, and might therefore experience stronger selection [28]. Although there have been some empirical studies on the effects of demography and connectivity, these effects have mostly been explored through theoretical work. Their applicability and prevalence in the wild remain to be addressed.

The Unknown Impact of Coevolution on Evolutionary Predictions

Interacting species that exert reciprocal selective pressures upon one another are widely documented in host–pathogen, predator–prey, and mutualistic interactions [29]. However, coevolution is largely ignored in studies of surveyed populations (but see 30, 31). We argue that neglecting these aspects can lead to the misestimation of V_A and the response to selection (Figure 2B). The classic illustration of this is the promotion of genetic variation in vertebrate major histocompatibility genes through coevolution with pathogens [32]. In addition, pathogens can maintain genetic variation in other traits of host species [33]. However, these studies generally estimate parameters in only one of the interacting species, while a full demonstration of coevolution requires that these are shown in both partner species [33]. Such simultaneous estimates remain virtually absent. Filling this gap should confirm predictions that host–pathogen coevolution promotes V_A [34]. Taking into account coevolution would improve the detection of selection pressures and responses that otherwise would not have been detected, or would have been overestimated.

The Potential Significance of Nongenetic Inheritance

Where adaptive phenotypic changes were detected in surveyed populations, only one third of these were associated with genetic change in response to selection [35]. This raises questions about the mechanisms underlying the response to selection, in particular whether there is a contribution of nongenetic inheritance. Theory predicts that inherited nongenetic variation can respond to natural selection 36, 37 (Figure 2C). Nongenetic inheritance might therefore account for a missing fraction of the genetic response to selection in the wild 39, 40. Equally, nongenetic inheritance can lead to inflated estimates of h², contributing to the mismatch between the predicted and observed phenotypic response to selection. Apart from parental environmental effects, nongenetic components of heritability cannot yet be estimated in surveyed populations. Nevertheless, two studies disentangled genetic and nongenetic components of phenotypic similarity 41, 42, which is an important first step [43]. There is a need to improve both theoretical and empirical understanding of nongenetic inheritance, particularly to (i) assess the long-term stability of nongenetic inheritance mechanisms, (ii) build a theoretical framework taking into account their key properties, and (iii) test for their role in the response to selection.

How to Refine Our Baseline Expectation of the Response to Selection

The basis of this approach takes the predicted response to selection (R) from the breeder’s equation, and the consequent reduction in V_A as a baseline expectation. Figures 1,2 illustrate the deviation from these baseline expectations caused by individual mechanisms (indicated by the shaded regions). Assuming an equal weighting of these mechanisms, we can qualitatively combine the effects of individual mechanisms the most likely evolutionary scenario in a given population. It is then possible to adjust baseline expectations for a particular population.

Implementing this graphical tool should be possible in any population, and give testable predictions to refine our expectations of evolutionary change. Figure 3A illustrates how the impact of three mechanisms can be qualitatively combined for an imaginary population. In our example, an imaginary population is introduced to a new habitat, leading to a founding event that increases the loss of V_A and decelerates the response to selection (Figure 2A). In this new environment, different genotypes have different levels of plasticity, which does not affect the baseline predictions for V_A but can lead to a more uncertain prediction of R (genotype-by-environment interactions; Figure 1Aiii). Furthermore, two traits under positive selection in this population have a negative genetic correlation that is not aligned with the direction of selection, thereby constraining the response to selection and maintaining V_A (Figure 1Bii). Thus, combining these mechanisms gives us the most likely evolutionary scenario in this imaginary population (illustrated by the darkest sector on the combined graph in Figure 3A): a reduced response to selection compared to baseline prediction, and a change in V_A that is likely to conform with the baseline prediction. An emergent property of this framework is that it can identify when disparate species and populations are expected to respond similarly to selection owing to parallels in the effects of biological mechanisms.

Taking Steps Towards a Global Understanding

This integrative approach could ultimately be applied to many mechanisms, populations, and species to refine the predictions of the response to selection and changes in genetic variation. Figure 3B illustrates the result of an integration of all the mechanisms outlined here. The darkest sector in the combined plot indicates a reduced response to selection compared to our baseline expectation, and the maintenance of a slightly higher than expected level of genetic variation. This integrative perspective therefore appears to match the general observation of evolutionary stasis, thereby corroborating the biological explanation for stasis. However, a range of scenarios other than evolutionary stasis remains possible, indicated by the broad shading across Figure 3B, although empirical evidence for these scenarios is scarce.

It is important to note that the statistical explanations are not refuted by this approach, and the imprecision of quantitative genetic estimates therefore still needs to be taken into account. There is also the caveat that our list of mechanisms is not exhaustive, but other mechanisms might be of interest, such as sexual selection and those that are yet to be documented, and can be incorporated into this framework. The frequency with which different mechanisms occur in natural populations remains unclear because that reported in published studies might be an artefact of research motivated to find explanations for evolutionary stasis. This could be clarified by broadening the range of study systems. Future work could integrate additional mechanisms and weight all mechanisms by their frequency of occurrence and the relative strength of their effects.