Inference of selection gradients using performance measures as fitness proxies

Abstract

Summary Selection coefficients, i.e. selection differentials and gradients, are useful for quantifying selection and for making comparisons across traits and organisms, because they appear in known equations for relating selection and genetic variation to one another and to evolutionary change. However, selection coefficients can only be estimated in organisms where traits and fitness (components) can be measured. This is probably a major contributor to taxonomic biases of selection studies. Aspects of organismal performance, i.e. quantities that are likely to be positively related to fitness components, such as body size, are sometimes used as proxies for fitness, i.e. used in place of fitness components in regression‐based selection analysis. To date, little theory exists to inform empirical studies about whether such procedures may yield selection coefficients with known relationships to genetic variation and evolution. We show that the conditions under which performance measures can be used as proxies for fitness are very limited. Such analyses require that the regression of fitness on the proxy is linear and goes through the origin. We illustrate how fitness proxies may be used in conjunction with information about the performance–fitness relationship, and clarify how this is different from substituting fitness proxies for fitness components in selection analyses. We apply proxy‐based and fitness component‐based selection analysis to a system where traits, a performance measure (size; similar to proxies that are commonly used in place of fitness), and a more proximate fitness measure, are all available on the same set of individuals. We find that proxy‐based selection gradients are poorly reflective of selection gradients estimated using fitness components, even when proxy–fitness relationships are quite strong and reasonably linear. We discuss the implications for proxy‐based selection analysis. We emphasise that measures of organismal performance, such as size, may in many cases provide useful information that can contribute to quantitative inferences about natural selection, and their use could allow quantitative inference about selection to be conducted in a wider range of taxa. However, such inferences require quantitative analysis of both trait–performance and performance–fitness relationships, rather than substitution of performance for measures of fitness or fitness components.

Natural selection for body shape in resource polymorphic Icelandic Arctic charr

“…Malmquist et al ., ; Robinson & Wilson, ; Arnegard et al ., ), morphological trade‐offs contributed little to diversifying selection via effects on trophic performance. In Thingvallavatn, the lake with greater phenotypic differentiation, morphologically intermediate individuals experienced reduced relative growth, indicating disruptive selection of shape if growth is monotonically related to fitness (Franklin & Morrissey, ). This diversifying selection is mediated by detrimental effects of Diphyllobothrium ( Copepoda‐ transmitted tapeworms) and effects of shape on growth independent of resource use variation.…”

“…Because we lack estimates of growth‐fitness or condition factor‐fitness relationships in our study populations, our estimates are performance gradients (Arnold, ) and cannot be interpreted quantitatively as selection gradients. However, under the assumption that these relationships are monotonic, performance gradients should reflect qualitative features of the true selective surface (Franklin & Morrissey, ).…”

“…The total effects of shape on relative growth can be considered ‘extended performance gradients’ in the manner of extended selection gradients ( η ; Morrissey, ). These total effects are the sum of all effects of shape on growth, via measured or unmeasured pathways, and can qualitatively reflect selection for shape (Franklin & Morrissey, ). Total effects can be divided into indirect and direct effects on growth: indirect effects involve mediating variables determined by functional hypotheses (here, trophic proxies of isotopic signatures and parasite infections; Fig.…”

A meta‐analysis of the agents of selection on floral traits

“…Records in the pollination category estimated components such as the number of pollinator visits and the amount of pollen deposited per flower. Records that estimated selection via the fitness proxy flower number were excluded from all analyses and from Table because estimates of selection via such proxies may not reflect selection via fitness components (i.e., survival and fecundity; Franklin and Morrissey ). We also excluded records that estimated fitness as fruits per flower because it is a ratio, and thus any variation could be caused by differences in either fruit or flower production.…”

How to measure natural selection

“…First, Franklin & Morrissey () consider the knotty issue of what to do when one can only obtain a proxy for fitness, such as body size, biomass, or growth rate. Their work builds on a classic framework developed by Arnold ().…”