Fournier-Level et al. (2013) used structural equation modelling (SEM) to estimate the effects of quantitative trait loci (QTL) and traits on fitness. Closely related to multiple regression, SEM and path analysis are statistical approaches for causal modelling in complex systems, which quantify direct and indirect effects of putative causal factors (traits and QTL) on downstream phenotypes (such as survival and reproduction). SEM has been used in ecology (Grace et al. 2010) and genomics (Vogel et al. 2007; Nuzhdin et al. 2012) to quantify the consequences of causal models in biology. By incorporating both traits and QTL in SEM (Li et al. 2006), analyses can identify the direct effect of QTL on early life cycle traits, as well as direct and indirect effects of QTL on fitness in each environment.
Expression of complex traits is influenced by genetic variation, differences among environments, and genotype–environment interaction (GxE). However, understanding these interactions at the level of individual QTL has been challenging, because experiments must be large enough to achieve statistical significance at individual QTL in several environments (Anderson et al. 2011). At individual QTL, QxE might show consistent effects (Fig. 1, E2-E3), or conditional neutrality (Fig. 1, E1-E2), where a QTL has significant effects in one environment but not in another. Alternatively, a change in phenotypic ranking (Fig. 1, E3-E4) indicates genetic trade-offs, which might be due to a single locus (antagonistic pleiotropy) or several tightly linked QTL. Finally, an ecologically important trait such as flowering time might show genetic trade-offs, or it might show consistent expression, while the fitness consequences of flowering might differ among environments (Fig. 2). Conditional neutrality may result in transient local adaptation when gene flow is restricted, but long-term maintenance of genetic variation typically requires antagonistic pleiotropy (Anderson et al. 2013). Most ecological studies find conditional neutrality for components of fitness (Anderson et al. 2011), although some instances of genetic trade-offs also have been found in nature (reviewed in Anderson et al. 2013).
These field studies (Fig. 3) found that correlations between traits varied among sites, causing differences in natural selection (Fournier-Level et al. 2013). Genetic analysis showed conditionally neutral or fairly consistent QTL effects on life history traits, such as age at reproduction. However, the selective consequences of reproductive timing differed among sites: high mortality rates favoured early flowering in Norwich, England and Cologne, Germany, while late reproducing genotypes were favoured in Halle, Germany. Consequently, late flowering alleles at FRI and CRY2 were favoured in Halle, while early flowering alleles of FRI and CRY2 had higher fitness in Norwich and Cologne, respectively. This illustrates consistent effects at the level of individual phenotypes, causing genetic trade-offs in fitness due to contrasting selection in different environments (Fig. 2). In this way, ecologically mediated local selection can maintain genetic variation across a species range. Finally, despite the difficulties of modelling complex phenotypes (Travisano & Shaw 2012), this analytical approach identified the selective consequences of individual QTL—a promising result worthy of future attention.
Genotype–environment interaction also is a central problem in agriculture, where data sets and analytical resources vastly exceed those in natural systems. Plant breeders address GxE using several approaches. First, analyses of genetic correlations and reaction norms are statistical approaches reflecting similar conceptual foundations, which can predict genetic responses to selection (Chenu et al. 2011). Second, changes in QTL expression can be combined with environmental covariates (Boer et al. 2007) to improve prediction and mechanistic understanding. Third, physiologically based computer models have been parameterized for crop species and agricultural environments, mapping from many dimensions of environmental complexity to a few measures of plant performance (van Eeuwijk et al. 2010; Chenu et al. 2011). The issues of GxE in agriculture and evolutionary ecology are nearly identical, and increased communication between these fields would be productive scientifically.
The ecologically important traits studied by Fournier-Level et al. (2013) showed higher heritabilities in the low-mortality environments, consistent with meta-analyses of animal populations in stressful and nonstressful environments (Charmantier & Garant 2005). Furthermore, Fournier-Level et al. (2013) found higher heritabilities for early life cycle traits, which therefore may show stronger responses to selection. Correspondingly, the QTL that influence fitness and other late life cycle traits will be more difficult to detect and to incorporate in causal models.
Genome-wide analyses have shown that most complex traits are highly polygenic, controlled by hundreds or thousands of loci (Rockman 2012). Although some QTL show substantial effects and appear in multiple studies, species and environments (e.g., Pin & Nilsson 2012; Anderson et al. 2013; Fournier-Level et al. 2013), how can we deal with hundreds of loci that are biologically important but experimentally undetectable? One approach is genomic selection, which largely abandons the search for individual loci, and instead predicts genotypic performance from genome-wide marker data (Meuwissen et al. 2001; Jannink et al. 2010). Although genomic selection has received little attention in ecological genetics, the success of this approach emphasizes the biological importance of QTL below the significance threshold. In turn, this complicates causal modelling of QTL, which may be nonsignificant in some environments, but nevertheless biologically important for natural selection. As future studies integrate QTL effects with causal modelling of multivariate phenotypes, it may be possible to explore whether inclusion or exclusion of marginally significant QTL alters the estimates of selection on complex traits. These approaches, combined with field measurements of complex traits and fitness consequences, are poised to improve our understanding of the factors that influence genetic variation in nature.