Linking genotypes to phenotypes and fitness: how mechanistic biology can inform molecular ecology


Anne C. Dalziel, Fax: 604-822-2416;


The accessibility of new genomic resources, high-throughput molecular technologies and analytical approaches such as genome scans have made finding genes contributing to fitness variation in natural populations an increasingly feasible task. Once candidate genes are identified, we argue that it is necessary to take a mechanistic approach and work up through the levels of biological organization to fully understand the impacts of genetic variation at these candidate genes. We demonstrate how this approach provides testable hypotheses about the causal links among levels of biological organization, and assists in designing relevant experiments to test the effects of genetic variation on phenotype, whole-organism performance capabilities and fitness. We review some of the research programs that have incorporated mechanistic approaches when examining naturally occurring genetic and phenotypic variation and use these examples to highlight the value of developing a comprehensive understanding of the relationship between genotype and fitness. We give suggestions to guide future research aimed at uncovering and understanding the genetic basis of adaptation and argue that further integration of mechanistic approaches will help molecular ecologists better understand the evolution of natural populations.


Recent advances in high-throughput molecular biology have made it possible to rapidly characterize a large number of genetic polymorphisms in virtually any natural population (reviewed by Bouck & Vision 2007; Lister et al. 2009; Mardis 2008). In addition to providing a wealth of putatively neutral markers that can be used to study ecological and evolutionary processes in natural populations, these high-throughput techniques also facilitate the search for adaptively significant genetic variation. The value of using a combination of population genomic and quantitative genetic methods to identify the genes underlying ecologically important traits in multicellular eukaryotes has been widely reviewed (e.g. Luikart et al. 2003; Storz 2005; Vasemagi & Primmer 2005; Ehrenreich & Purugganan 2006; Jensen et al. 2007; Hoffmann & Willi 2008; Naish & Hard 2008; Pavlidis et al. 2008; Schmidt et al. 2008; Stinchcombe & Hoekstra 2008; Mackay et al. 2009; Slate et al. 2009), and we will not revisit these issues here. Instead, the purpose of this review is to highlight the benefits of incorporating a mechanistic perspective when attempting to find the genetic variants associated with ecologically relevant phenotypic variation, predicting the potential impacts of this variation across levels of biological organization, and ultimately testing these predictions.

We define a mechanistic perspective as one that incorporates a priori knowledge about the function of genes, proteins, biochemical networks and pathways, and their resulting effects on phenotypic traits, whole-organism performance and fitness. Quite simply, incorporating a mechanistic perspective means thinking about how organisms ‘work’. Although the benefits of incorporating a mechanistic perspective into evolutionary studies have long been recognized (reviewed by Autumn et al. 2002; Dean & Thornton 2007; Watt 1985, 2000; Watt & Dean 2000), the ability to perform truly integrative studies has been limited by technical and analytical hurdles. We are now at a time where the widespread availability of molecular techniques, and recent advances in this field (both empirical and analytical), are making truly integrative studies more practical.

In this review, we begin by showing how mechanistic knowledge can help molecular ecologists to formulate testable predictions about the effects of genetic variation on phenotype, performance and fitness. We then review three classic research programs that successfully exploited a mechanistic approach to choose candidate genes and then test the phenotypic (and fitness) consequences of genetic variation at these loci. We conclude by providing an outline of the methods that will be required to carry this approach into the future.

Box 1 Selected online resources to identify candidate genes and assess their function

Online databases provide a rich source of information about genes and their biological functions. The list below is organized by level of biological organization, but many of these resources are useful across multiple levels. This list focuses on resources for protein coding genes. Resources for detecting and interpreting the functional consequences of variation in ncRNA genes and regulatory sites are also available, but the complexities of these analyses are beyond the scope of this review (see Mituyama et al. 2009; Nardone et al. 2004; Portales-Casamar et al. 2009; Wasserman & Sandelin 2004 for further information).

Finding genes present in a genomic region

If the results of population genomic or quantitative genetic screens highlight a particular genomic region, this region can be scanned for genes of interest using a number of genome browsers. The Ensembl ( and UCSC ( genome browsers are useful for researchers studying animals or fungi (Kent et al. 2002; Hubbard et al. 2009). For plants, NCBI’s plant genome central ( or species-specific sites such as the genome browser at The Arabidopsis Information Resource (TAIR) ( (Swarbreck et al. 2008) can be used. These browsers allow users to examine the genome of your species of interest (if available) or that of a closely related species. Levels of annotation vary among genomes, and the locations of ncRNA and regulatory sites are not normally included.

Gene function

Once a gene of interest has been identified, NCBI’s Entrez Gene ( is a very useful resource for obtaining mechanistic information about protein-coding genes (Maglott et al. 2007) because it summarizes information from multiple sources. Entrez Gene provides intron/exon structure, alternative splice variants, gene ontology information (from, a text summary of the protein’s functions, and links to other NCBI and external resources (Sayers et al. 2009). NCBI’s conserved domain database ( can help identify protein domains within protein-coding regions and is particularly helpful for identifying gene function in cases where the function of the complete protein is unknown (Marchler-Bauer et al. 2009).

Protein structure

NCBI’s molecular modeling DataBase (MMDB) (, coupled with their Cn3D structure viewer, can display the location of mutation within the 3D structure of a protein (Wang et al. 2007). These 3D crystal structures are obtained from the RCSB Protein data bank (PDB) ( (Berman et al. 2000). If a protein is part of a multi-subunit protein complex, these databases also include interactions among subunits.

Pathways and networks

If a gene is part of a well-studied pathway, resources such as WikiPathways (, Pathway commons (, the Kyoto encyclopaedia of genes and genomes (KEGG) (Kanehisa et al. 2008)(, PANTHER (PRotein ANalysis THrough Evolutionary Relationships) (, the Reactome database ( or the Plant metabolic network ( can provide a way to assess the relationship between a candidate gene and other genes within the pathway or network. BioCarta ( provides another, very different, compendium of biological pathways. Although the BioCarta compendium lacks the sophisticated functionality and broad coverage of the other pathway databases, it provides extremely accessible ‘cartoon’ versions of the pathways that it covers, accompanied by explanatory text at the level of an introductory textbook.

Effects on organismal phenotype

NCBI’s Online Mendelian Inheritance in Man (OMIM) ( is a collection of information about human genes often coupled to information about phenotype, particularly those related to Mendelian disorders in humans (Amberger et al. 2009). OMIM often provides helpful clues as to the function of a particular gene in the context of a whole organism. Online Mendelian Inheritance in Animals (OMIA) ( is a newer resource that contains some additional information (Lenffer et al. 2006) on other animal species. For plant biologists, TAIR ( contains some information about the relationship between genotype and phenotype (Swarbreck et al. 2008).

The list above is not comprehensive, and new databases are regularly developed. The journal Nucleic Acids Research publishes an annual ‘Database Issue’ which contains updated information on these and many other genomic resources (e.g. for the January 2009 listing).

Using mechanistic biology to predict the effects of genetic variation

Biological systems can be divided into a series of hierarchical levels, with genetic variation (at the base of this hierarchy) affecting processes at higher levels (Fig. 1). Thus, variation at the genetic level can affect the function and/or amount of proteins, which may then alter protein–protein interactions, influence biochemical pathways and networks and eventually modify cellular function, organismal phenotype, whole-organism performance capabilities and fitness. The value of examining multiple levels of biological organization has been clearly summarized by the prominent ecological physiologist George Bartholomew, who wrote ‘there are a number of levels of biological integration and… each level finds its explanations of mechanism in the levels below and significance in the levels above it’ (Bartholomew 1966). Here we show how incorporating available mechanistic information can help to generate predictions about the impacts of variation at each level of biological organization and guide experiments at higher levels of organization. Decades of mechanistic investigation have provided a wealth of information that can be used to guide these predictions. Box 1 summarizes some of the online resources that can be used to access this information. We argue that paying attention to as many levels of biological organization as possible should enhance our ability to understand the consequences of genetic variation, and to detect the genes that are subject to natural selection in the wild.

Figure 1.

 Connections across levels of biological organization. Genetic variation may affect phenotype at a number of levels of biological organization to ultimately influence whole organism performance capabilities and fitness. *Genetic variation may result from a variety of types of polymorphisms [e.g. single nucleotide polymorphisms (SNPs), small insertions or deletions (indels), or copy number variation (CNV)] in a variety of gene classes (e.g. protein-coding genes, ncRNAs, or regulatory elements). †‘Environment’ includes all of the biotic and abiotic factors that can influence traits at any level of this hierarchy.


There are many types of genetic variation that can affect the function and/or amount of proteins, including single nucleotide polymorphisms (SNPs), small insertion–deletion polymorphisms (indels), microsatellites and larger copy number variants (Feder 2007). All of these types of genetic variation have been shown to contribute to variation in ecologically relevant traits (e.g. Hammock & Young 2004; Perry et al. 2007; Aminetzach et al. 2005; Schlenke & Begun 2004; and supporting information Table S1). To make a prediction about the functional consequences of any of these types of genetic variation from sequence information alone, it is first necessary to determine whether the variant occurs in a protein-coding or non-protein-coding sequence region. If the genetic variant occurs in a protein-coding gene, it is often possible to make predictions about its function (see Proteins). However, less than 2% of a typical mammalian genome codes for proteins (Amaral et al. 2008), so many genetic variants will occur in non-coding regions of the genome. For example, variants may fall within a regulatory element (e.g. transcription factor binding site) that does not need to be transcribed to perform its function. Many non-transcribed regulatory elements have been functionally characterized and there are several examples of variation in regulatory elements that contribute to ecologically important variation (reviewed by Wray 2007). However, there are still no general rules that can be used to predict how sequence changes in regulatory regions affect function (e.g. Wray et al. 2003; Segal & Widom 2009), so the effects of sequence variation in regulatory elements often must be determined experimentally in the absence of a priori predictions. Alternatively, a genetic variant may fall within a non-coding RNA (ncRNA), which is transcribed but not translated into a protein (e.g. tRNAs, rRNAs snoRNAs, piRNAs and microRNAs). Variation in ncRNA genes is likely to have important ecological consequences, given that almost 70% of the genome is transcribed (Amaral et al. 2008; Mattick 2009). Predicting the effects of variation in ncRNAs will unfortunately remain difficult until mechanistic information about their functions improves (Mattick 2009). However, the difficulty of making mechanistic predictions about some of these ncRNAs from sequence data alone does not reduce the importance of taking a mechanistic approach to study the impacts of genetic variation in these genes; it simply increases the difficulty of designing refined experimental tests of functional impacts at the next level of biological organization.


Genetic variation must ultimately affect the amount or function of a protein to have consequences at higher levels of biological organization. As discussed above, when examining protein-coding genes, it is often possible to make predictions about the effects of genetic variation from primary sequence information. There are many online resources through which available mechanistic information about protein structure and function can be accessed (highlighted in Box 1). These resources can help to identify the probable function of a protein-coding gene and the specific functional domains within this protein. If further information is available about the roles of particular amino acids within a domain, more refined predictions about how a genetic variant effects biochemical phenotype can be made.

For example, Johns & Somero (2004) used mechanistic knowledge about the links between lactate dehydrogenase-A (LDH-A) primary sequence, tertiary structure and enzyme function to predict which genetic polymorphisms were responsible for observed differences in LDH-A kinetics among temperate and tropical damselfish (genus Chromis) that were hypothesized to be important for thermal adaptation. They predicted that one polymorphic site (T219A), would be of particular importance because of its location in a key ‘hinge’ region of the protein and because this site was correlated with low temperature adaptation in Antarctic nototheniod fishes (Fields & Somero 1998). They tested this prediction using site-directed mutagenesis followed by in vitro expression and biochemical tests of protein function. As predicted, all fixed differences between tropical and temperate fishes had some effect, but only the T219A mutation was sufficient to produce the biochemical changes that occur among species. Other excellent examples in which mechanistic predictions have been used to design relevant experiments to test the impacts of genetic variation have been comprehensively reviewed by Dean & Thornton (2007).

Biochemical pathways and networks

Information about the function of a protein in isolation can provide some insights into the impact of protein sequence variation on phenotype. However, most proteins perform their functions through interactions with other proteins, or as part of biochemical pathways and networks. Understanding these higher-order interactions is critical for predicting the linkages from genotype to phenotype to fitness. For example, the enzyme cytochrome c oxidase (COX) is made up of 13 protein subunits in animals. Mutations in any one of these subunits can affect interactions among subunits and thus change the functional properties of the enzyme. The multi-subunit COX enzyme is itself part of the mitochondrial electron transport chain, which is made up of four multi-subunit protein complexes and the protein cytochrome c, which interacts directly with COX (reviewed by Rand et al. 2004). The importance of the interactions between COX and cytochrome c can be clearly seen in the intertidal copepod, Tigriopus californicus. When divergent populations of this species are crossed, there is an incompatibility between the nuclear encoded cytochrome c gene and a mitochondrially encoded COX subunit that results in mitochondrial dysfunction in hybrids (Rawson & Burton 2002; Ellison & Burton 2008). Electron transport chain function is also dependent on the proper functioning of other biochemical pathways, such as the Citric acid cycle, which produces the reducing equivalents needed for electron transport chain function, and ultimately those pathways that produce the substrates for the citric acid cycle, such as glycolysis and the β-oxidation pathway (Hochachka & Somero 2001). Thus, mutations that directly affect the function or amount of a protein and mutations that affect a protein’s interactions with other pathway and network members have the potential to affect the phenotype. There are now a number of databases that summarize the roles of specific genes in biochemical pathways and networks so that this knowledge can be more easily incorporated into predictions about the effects of genetic variation (Box 1).

As the proteins within a network play a collaborative role in generating the phenotype, changes in several different genes within a network could result in similar changes in ecologically relevant phenotypes (e.g. coat colour in mice; Steiner et al. 2009). Alternatively, specific genes within a network may be repeatedly targeted by evolution in multiple taxa because of their role and location in the network (Derome & Bernatchez 2006; Carroll 2008; Stern & Orgogozo 2008, 2009; Erwin & Davidson 2009; Streisfeld & Rausher 2009). Functional and physical interactions among proteins in multi-protein complexes, biochemical pathways and networks are the underlying cause of the genetic phenomenon of epistasis (Phillips 2008; Tyler et al. 2009). A network perspective is therefore necessary to understand why some genes and not others are involved in generating adaptively significant phenotypic variation (Carroll 2008; Stern & Orgogozo 2008, 2009; Erwin & Davidson 2009). This perspective also helps to illuminate the distinction, if any, between convergent and parallel evolution (see Abouheif 2008; Arendt & Reznick 2008). Viewed narrowly, true parallelism only occurs when there are independent origins of the same mutation, causing the same amino change, in the same protein in two taxa that have the same underlying biochemical networks. It is more useful, however, to mechanistically determine the level of biological organization at which parallelism ends and convergence begins. For example, similar traits in different taxa could arise via changes in the same gene, in different genes within the same pathway, or changes in different pathways that interact within a network; a trait might have evolved in parallel when viewed at one level of biological organization, but by convergence at an underlying level.

Organismal phenotypes, whole-organism performance and fitness

Changes in biochemical pathways and networks result in cellular-level changes that can influence the structure and function of tissues, organs and organ systems, and thus alter a wide variety of complex organismal traits, including morphology, behaviour and physiology (Fig. 1). Predicting the effects of genetic variation on morphology, behaviour and physiology requires an understanding of the interactions among the many underlying levels of biological organization. Most of the databases relating genotype to organismal phenotype contain information compiled from naturally occurring (e.g. human disease phenotypes) and laboratory-produced mutants (Box 1). In natural populations, organism-level traits interact to influence the performance capacity of an organism. Whole organism performance capacity can be defined ‘as the ability of an animal to conduct an ecologically relevant task’ (Irschick et al. 2008), and is a metric of how well a task is done. Such tasks may include foraging ability, dispersal or predator avoidance/resistance. Knowledge about the interactions among organism-level traits can often be used to make predictions about their individual impacts upon whole-organism performance.

Finally, the ability of an organism to perform all the various tasks necessary for it to survive and reproduce can ultimately influence fitness (Irschick 2003). It is only in those cases in which genetic variants have differential effects on whole-organism performance that fitness effects will occur. The set of potential relationships between phenotypic traits, performance and fitness has been clearly outlined by Arnold (1983), using path analysis to statistically model these associations. Similar associations could, in principle, be drawn between processes at any level of biological organization from genetic variation through the intervening levels to fitness.

Empirically, determining if a gene or trait has evolved by natural selection is best accomplished by directly measuring the contribution of alternate alleles to the next generation (e.g. Endler 1986; Barrett et al. 2008). However, indirect methods, such as examining molecular data for signatures of selection (reviewed by Jensen et al. 2007; Nielsen 2005) and comparative phylogenetic analyses (reviewed by Garland et al. 2005; Leroi et al. 1994) may also be used to provide evidence for selection. These methods are complementary, so a combination of these approaches, with mechanistic evidence of a connection between genetic and phenotypic variation, provides an accumulation of evidence that strongly supports the hypothesis of adaptive evolution of a trait.

Classic examples of mechanistic approaches

There are many research programs that have made tremendous progress in characterizing genes of large effect associated with ecologically important traits [e.g. flowering time in Arabidopsis thaliana (reviewed by Ehrenreich & Purugganan 2006; Roux et al. 2006; Shindo et al. 2007); body armour and colouration in threespine stickleback (e.g. Shapiro et al. 2004; Colosimo et al. 2005; Miller et al. 2007); for additional examples, see supporting information Table S1]. In addition, several groups have already identified, or are on the cusp of identifying, candidate genes for ecologically important traits in a variety of plant and animal species [e.g. whitefish (Rogers & Bernatchez 2007; Whiteley et al. 2008; Jeukens et al. 2009; Nolte et al. 2009); sunflowers (e.g. Kane & Rieseberg 2007; Sapir et al. 2007; Lai et al. 2008); Bochera stricta, a close relative of Arabidopsis (e.g. Schranz et al. 2009); wild tomatoes (e.g. Moyle 2008); marine snails (e.g. Wood et al. 2008; Galindo et al. 2009); additional examples reviewed in Karrenberg & Widmer (2008)]. Given this, we expect that many more candidate genes associated with ecologically important traits will be characterized in the near future, although this will remain difficult for genes of small effect. Below, we discuss in more detail three classic pre-genomics research programs that have integrated across levels of biological organization to study the effects of genetic variation at a selected locus (Fig. 2). Each of these research programs used knowledge of mechanism to provide clear, testable hypotheses about the links between genotype, phenotype and fitness.

Figure 2.

 Three integrative research programs linking genotype to phenotype, performance and fitness. Connections across levels of biological organization follow the structure of Fig. 1. Links that have not yet been studied are denoted with question marks. If the link from performance to fitness has not yet been studied directly, the arrows are excluded. See the accompanying text for further details and references. (a) Lactate dehydrogenase-B (LDH-B) in Fundulus heteroclitus. (b) Phosphoglucose isomerase (PGI) in Colias spp. Similar evidence for the effects of PGI genotype on phenotype and performance has been found in the Glanville fritillary butterfly (Melitaea cinxia); this evidence is listed in blue and traits which are linked to PGI genotype in both Colias spp. and M. cinxia are noted with a blue asterisk. (Photo provided by W. B. Watt). (c) The voltage-gated sodium channel (Nav1.4) in Thamnophis sirtalis. (Photo provided by E. D. Brodie III).

Lactate dehydrogenase and thermal adaptation of metabolism in killifish

The common killifish (Fundulus heteroclitus) is a small teleost fish that lives in marshes and estuaries along the Atlantic coast of North America. There is a steep latitudinal thermal cline over this species’ range such that northern populations experience temperatures that are, on average, 13 °C colder than those experienced by southern populations at the same time of year (reviewed by Powers & Schulte 1998). Dennis Powers and colleagues initiated the search for genes that are differentially selected between northern and southern populations of this species over 30 years ago, using allozyme screening; the ‘genome scans’ of the pre-genomic era (Place & Powers 1978; Powers & Place 1978). Analyses of allozyme frequencies detected clines at a number of loci, including LDH-B, an enzyme that catalyzes the interconversion of pyruvate (a fuel for aerobic respiration) and lactate (the end product of anaerobic glycolysis). When Place & Powers (1979, 1984a, b) examined the kinetics of purified LDH-B enzymes in vitro they found that the northern LDH-B enzyme (LDH-BNN) had a higher catalytic efficiency at low temperatures, as would be predicted if local adaptation to low temperatures had occurred in northern populations, but did not find evidence for local adaptation of the southern LDH-B (LDH-BSS) genotype to warmer temperatures. Sequence analyses of LDH-B alleles suggested that a particular amino acid variant at site 311 (Ala → Asp) was responsible for functional differences among LDH-B alleles (reviewed by Powers & Schulte 1998).

Powers et al. (1979) also discovered that ATP concentrations ([ATP]) in red blood cells are correlated with LDH-B genotype. Since ATP decreases haemoglobin-oxygen binding affinity, DiMichele & Powers (1982b) predicted that LDH-BNN fish, which have higher [ATP], would have a lower haemoglobin-oxygen binding affinity, allowing for more efficient unloading of oxygen at the working muscles and an improvement in endurance swimming performance (a trait that is highly dependent upon oxygen availability, transport and use). As expected, LDH-BNN fish had higher [ATP], lower haemoglobin-oxygen affinity and superior endurance swimming performance at 10 °C, a temperature rarely experienced by southern fish. However, there were no differences in haemoglobin-oxygen affinity, [ATP] or swimming performance at 25 °C, which is consistent with in vitro studies of LDH-B kinetics that find no differences among genotypes at warm temperatures (DiMichele & Powers 1982b).

Powers and colleagues next examined a suite of other performance traits that are influenced by environmental temperatures, including embryonic metabolic rate, growth rate and hatching time. Since killifish lay eggs at the peak of the highest spring tide, their eggs must develop in air and hatch at the next high tide 2 weeks later to survive (reviewed by Dimichele et al. 1986). Thus, these traits were expected to be under strong divergent selection between northern and southern Fundulus populations that are developing at different temperatures in the wild. They found that LDH-BNN embryos had lower metabolic rates, slower development, later hatching times, decreased lactate metabolism and decreased glucose production (reviewed by Dimichele & Powers 1982a, 1984a,b; Dimichele et al. 1986; Paynter et al. 1991). Note that these differences in metabolism, hatching and growth are opposite to what would be expected for local adaptation to a colder environment (where faster development would be expected to evolve to counter a slowing of metabolism due to colder temperatures). These correlations between genotype and cellular and organismal phenotype were then directly tested by exchanging the native LDH-B enzymes of an egg (e.g. LDH-BNN) with the alternate LDH-B enzyme (e.g. LDH-BSS). Dimichele et al. (1991) found that the injected LDH-B enzyme determined the metabolic rate and glucose use of the egg, showing that it was LDH-B, and not a linked locus, that caused the observed differences in cellular metabolism and embryonic development.

All of the in vivo experiments described above were performed on Fundulus from the hybrid zone between northern and southern genotypes, near the middle of the LDH-B cline [In Delaware, which is also near the centre of other allozyme clines (Powers & Place 1978)]. Therefore, LDH-BSS and LDH-BNN genotypes were tested in individuals with a mixture of northern and southern alleles at other loci and observed phenotypic variation between these two genotypes could be attributed to their LDH-B genotype (or to closely linked loci) rather than to correlated variation at other (unknown) loci. In addition, collecting fish from a single locality controlled for prior thermal history. However, more recent work on Fundulus from extreme northern and southern populations found either no differences in performance associated with LDH-B genotype [swimming performance (Fangue et al. 2008)], or differences in the opposite direction to the effects of LDH-B genotype alone [development rate to hatching (DiMichele & Westerman 1997), growth rate following hatching (Schultz et al. 1996) and adult metabolic rate (Podrabsky et al. 2000; Fangue et al. 2009)]. These observations are consistent with experiments on fish from the centre of the LDH-B cline in Delaware that examined multilocus genotypes at several allozymes, as opposed to LDH-B in isolation. Dimichele & Powers (1991) found that fish bearing the most common northern multilocus genotype had faster development, despite the fact that at the single locus level fish bearing the LDH-BNN genotype developed more slowly (Dimichele & Powers 1991). Thus, for at least growth rate post-hatch, examining LDH-B alone does not give a true picture of the differences among populations.

The search for the other loci influencing metabolism, hatching and growth is now underway (e.g. Whitehead & Crawford 2006). Interestingly, there are also differences in the amounts of LDH-B enzyme among northern and southern genotypes that are mediated by differences in transcriptional regulation (Crawford & Powers 1992). A combination of comparative sequence analyses, in vitro experiments and in vivo tests of promoter action found that differences in transcription are largely because of sequence variation in a cis-regulatory region upstream of the Ldh-B gene (Schulte et al. 1997, 2000) and SP1 sites in the proximal promoter (Segal et al. 1999). Analyses of molecular signatures of selection (Schulte et al. 1997) and phylogenetic comparative studies (Pierce & Crawford 1997) suggest that natural selection shaped these transcriptional differences (reviewed by Schulte 2001). In addition, Whitehead & Crawford (2006) have used comparative phylogenetic methods to identify 13 other metabolic genes that show evidence of selection for changes in expression in response to habitat temperature (or environmental factors correlated with temperature).

Phosphoglucose isomerase, flight and thermal adaptation in Colias butterflies

Ward Watt and colleagues began their research on phosphoglucose isomerase (PGI) polymorphisms in Colias butterflies by selecting this gene as a candidate underlying local adaptation to environmental temperature (Watt 1968; Sherman & Watt 1973). Colias butterflies eat nectar, a mixture of simple sugars, to fuel their flight. Based on this observation, Watt (1977) hypothesized that selection for optimal flight performance could act to fine tune glycolysis, a pathway involved in sugar metabolism, to environmental temperature. More specifically, he hypothesized that PGI would be the target of selection in response to environmental temperature as this homodimeric enzyme catalyzes the reversible conversion of fructose-6-phosphate to glucose-6-phosphate, and sits at a key branch point in glycolysis that links substrates into other pathways such as gluconeogenesis.

Watt (1977) surveyed populations of four species of Colias butterflies (Colias meadii, Colias alexandra, Colias philodice eriphyle and Colias eurytheme) for allozyme variation at PGI, and found a number of allelic variants or electromorphs (EM). Interestingly, there was an excess of heterozygotes in older butterflies when compared with younger butterflies, suggesting differential survival of genotypes (Watt 1977). Identical-by-descent laboratory-raised populations for each of the four most common allelic variants for C. eurytherme were produced, so that genotypes could be tested for differences in in vitro biochemical functioning. There were a number of biochemical differences among the alleles, including thermal stability and substrate-binding affinity (Km) (Watt 1977, 1983). The most striking result from these biochemical measurements was that homodimeric enzymes showed a trade-off between thermal stability and enzyme kinetics, whereas heterodimeric enzymes did not (Watt 1977, 1983). Thus, heterodimeric enzymes, with one allele optimized for stability and the other for catalytic efficiency, functioned better than homozygotes over a wide range of temperatures. Watt et al. (1996) found that PGI enzymes in C. meadii, although unique in origin and sequence, also show similar trade-offs between thermal stability and kinetics as in C. eurytheme enzymes.

These remarkable differences at the biochemical level generated clear predictions about how these alleles might affect whole animal physiology and performance in the wild. Watt predicted that heterozygotes should be able to fly at a wider range of environmental temperatures because their metabolic pathways would be able to function well across a range of temperatures. Indeed, C. p. eriphyle, C. eurytheme and C. meadii heterozygotes were able to fly earlier in the day (when it is cold), and fly for a longer overall time each day (Watt 1983; Watt et al. 1983). These differences in performance were also hypothesized to affect fitness components that depend on capacity for flight or thermal tolerance, such as mating success and/or fecundity. As predicted, survival during heat stress in C. p. eriphyle, male mating success in C. p. eriphyle, C. eurytheme and C. meadii, and female fecundity in C. p. eriphyle were all highest for heterozygous butterflies; thus, heterozygotes had a greater net fitness (Watt 1983, 1992; Watt et al. 1983, 1985, 1996, 2003; Carter & Watt 1988).

Phosphoglucose isomerase alleles have now been sequenced from C. eurytherme and C. meadii, and there are multiple amino acid changes among and within EM classes and among species that display evidence of evolution via natural selection (Wheat et al. 2006). While the exact mutation(s) underlying differences in thermal adaptation is still unknown for Colias, the most promising candidates lie in the region of PGI’s tertiary protein structure that links the two monomers to form a functional enzyme. Interestingly, the exact sites mutated vary from species to species, but occur in the same protein region (Wheat et al. 2006; Wang et al. 2009).

Similar evidence for the effects of PGI genotype on phenotype and performance has been found in the Glanville fritillary butterfly (Melitaea cinxia). M. cinxia heterozygote individuals have a higher body temperature, flight metabolic rate and dispersal distance at colder temperatures, which results in higher fecundity (e.g. Haag et al. 2005; Niitepõld et al. 2009; Saastamoinen & Hanski 2008) and population growth (Hanski & Saccheri 2006). Heterozyotes also have increased survival (Orsini et al. 2009) and a longer lifespan (Saastamoinen et al. 2009). The impacts of PGI genotype on thermal adaptation are not limited to butterflies. For example, there is evidence for local adaptation of PGI alleles to temperature in the willow beetle (Chrysomela aeneicollis) (Dahlhoff & Rank 2000) and the sea anemone Metridium senile (Zamer & Hoffmann 1989). These studies, in combination with strong empirical evidence (i.e. measuring genotype frequencies across life history stages, measuring fitness components, and testing for genetic signatures of selection) from Colias butterflies, support the hypothesis that PGI evolves by natural selection in Colias spp., and is a gene with major effects.

Voltage-gated sodium channel (Nav1.4), poison resistance and locomotion in garter snakes 

Garter snakes (Thamnophis siralis) feed on rough-skinned newts (Taricha granulose) in the regions of western North America where these two species overlap. To defend themselves from predators the newts contain a toxin, tetrodotoxin (TTX), in their skin (Wakely et al. 1966; Brodie et al. 1974). TTX is a very potent neurotoxin, which binds to, and blocks, the outer pore of voltage-gated sodium channels (Nav) in neurons and muscles. At the cellular level, blocking these channels inhibits the initiation of action potentials, which are necessary for nerve and muscle function. When even minute amounts of TTX are ingested, muscles become paralyzed and poisoned animals usually die by suffocation (reviewed by Soong & Venkatesh 2006). These devastating consequences of ingesting TTX are expected to strongly select for the evolution of TTX resistance, and as predicted, garter snakes from newt-eating populations have been shown to have greater resistance to TTX (Brodie & Brodie 1990; Brodie et al. 2002). As well, variation in TTX levels in newts is geographically correlated with levels of resistance in snake populations (Brodie & Brodie 1991; Hanifin et al. 1999), making this system a classic example of a co-evolutionary ‘arms-race’ (Brodie & Brodie 1999; Brodie et al. 2002).

Resistance to TTX was originally measured by injecting snakes with TTX and then testing muscle contraction ability via crawling, a performance trait important for escape from predators and prey capture (Brodie & Brodie 1990). TTX resistance, measured using this performance trait, did not vary when young snakes were repeatedly injected with TTX (Ridenhour et al. 1999) or among laboratory-reared and field-caught snakes (Ridenhour et al. 2004), which suggested that TTX resistance had a genetic basis and was not dependent on environmental factors. Crawling performance, a whole-animal performance measure of TTX resistance, was strongly correlated with a cellular measure of resistance: the ability for action potentials to propagate when an animal was exposed to TTX (Geffeney et al. 2002). A priori knowledge of the mechanism of action of TTX (i.e. that it binds to sodium channels) suggested that this resistance might be mediated at the biochemical level by the presence of TTX-resistant sodium channels (reviewed by Soong & Venkatesh 2006). Geffeney et al. (2005) tested this hypothesis by looking for sequence variation in the voltage-gated sodium channel gene, Nav1.4 (the isoform expressed in muscle), between resistant and susceptible garter snakes and assessing the impacts of variants on protein function in vitro. Knowledge of the structural interaction between TTX and the outer pore of the Nav1.4 enzyme allowed for clear predictions about the location of these mutations in the protein sequence. In vitro assays of TTX binding to the sodium channel (Nav1.4) demonstrated that a single mutation, found in all resistant populations, could decrease TTX binding to Nav1.4, and that the 1–3 additional non-synonymous changes found in the most resistant populations further decreased TTX binding and increased resistance (Geffeney et al. 2005). These data suggest that a great deal of the variation in TTX resistance in garter snakes can be explained by the four amino acid changes in the outer pore of the Nav1.4 enzyme, and were consistent with biochemical knowledge of this sodium channel (Fig. 2).

Feldman et al. (2009) have recently expanded this work to examine two congeners of T. siralis, Thamnophis atratus and Thamnophis couchii, that also contain TTX-resistant populations. They found that SNPs in the protein regions which form the outer pore of the Nav1.4 channel also correlate with TTX resistance in these species. However, the specific mutations that confer resistance varied among species, suggesting that resistant alleles have evolved independently (Feldman et al. 2009), and thus represent a case of convergent evolution at the nucleotide level with parallel evolution at higher levels of organization. Polymorphisms in the sodium channel gene also underlie resistance to a structurally and functionally similar neurotoxin, saxitoxin, in a wild-clam population (Bricelj et al. 2005).

Lessons learned from these examples

The examples discussed above, and listed in Table S1, provide a number of valuable lessons. The first and over-riding lesson is that mechanistic knowledge can be used to generate testable hypotheses about the effects of a particular genetic polymorphism at higher levels of biological organization. Two of the examples (i.e. LDH and PGI) discussed in detail above started at the biochemical level and worked ‘up’ to cellular phenotypes, organismal phenotypes and fitness, but in principle, such hypotheses could be generated beginning at any level in the biological hierarchy. The second important lesson is that a purely mechanistic approach has its limits. In particular, this approach has the potential to bias the search for genes underlying ecologically important traits towards well-understood biochemical pathways and miss other genes-affecting fitness. The incorporation of a top-down approach, first elucidating phenotype-environment associations followed by complementary marker-based approaches (e.g. quantitative trait locus (QTL) mapping, linkage disequilibrium mapping and/or genome scans) is likely to reduce the impacts of this ascertainment bias (e.g. Rogers & Bernatchez 2007; Whiteley et al. 2008), and detect other loci underlying a trait of interest. Once these loci are detected, available mechanistic knowledge can give insight into the molecular basis of genetic interactions, if present. For example, knowledge of the genes that underlie ecologically important differences in flowering time in A. thaliana (e.g. Ehrenreich et al. 2009; Flowers et al. 2009) coupled with extensive knowledge about the biochemical pathways underlying flowering time has guided studies on the epistatic interactions among loci, such as the interactions between flowering locus C (FLC) and FRIGIDA (FRI) genotype (Caicedo et al. 2004; Michaels & Amasino 2001; reviewed by Ehrenreich & Purugganan 2006; Mitchell-Olds & Schmitt 2006; see Table S1 for further references).

The third important lesson is that it is critical to ensure that experimental conditions are as ecologically relevant as possible when assessing the effects of genetic variation on organismal phenotypes and fitness (discussed by Ungerer et al. 2008). For example, differences in PGI and LDH function were only seen when these enzymes, and animals, were studied at certain temperatures (e.g. Watt 1977, 1983; DiMichele & Powers 1982b).

The final and perhaps most critical lesson from these examples is that isolating the impacts of a single gene in natural populations with high background genetic variation (i.e. variation at other loci throughout the genome) is necessary to firmly establish the causal link between gene, phenotype and fitness. This is clearly seen in the case of LDH-B in Fundulus, in which the effects of LDH-B genotype vary widely depending on the genetic background in which the alleles are tested. Performing experiments without controlling for genetic background can reduce the power to infer causal relationships between genotype and phenotype, or even result in false conclusions (discussed by Dean & Thornton 2007). Thus, in the section below, we explore some of the available methods for controlling for background genetic variation in studies attempting to link genetic variation at a candidate gene to effects on phenotypes, performance and fitness.

Controlling for background genetic variation

Genetic variation at candidate genes can be feasibly tested in controlled genetic backgrounds using (i) carefully selected naturally occurring genetic variants, including the use of clonal or asexual lines when possible, (ii) forward genetics or controlled cross approaches, and (iii) reverse genetics. The most appropriate method of controlling for background genetic variation (using naturally occurring variants; forward genetics; or reverse genetics) will ultimately depend upon the characteristics of the species being examined, the available genetic resources, and the question being addressed. Where possible, a combination of approaches will be extremely fruitful (e.g. Kammenga et al. 2007). In addition, because forward and reverse genetic methods of controlling for background genetic variation might underestimate the complexity found in natural populations (see Jensen et al. 2007; Ungerer et al. 2008), it is advisable to complement these approaches with a wide sampling from naturally occurring populations to achieve the best estimate of the effects of epistatic interactions.

Controlling for background genetic variation: naturally occurring variants

To control for background genetic variation using naturally occurring genetic variants, one must identify populations in which the alleles of interest are segregating within a largely homogenous genetic background. One possible approach is to take advantage of naturally occurring hybrid zones, as individuals at the centre of the hybrid zone may be segregating for the multiple alleles at the candidate gene of interest (e.g. LDH-B in Fundulus heteroclitus; Powers et al. 1991). However, strong linkage disequilibrium at the centres of hybrid zones can limit this approach as the variant of interest may be linked to many other loci also differing among populations. Collecting individuals at the edges of hybrid zones where a particular allele has introgressed into an otherwise pure genetic background might be an alternate strategy, but the steep clines expected for selected genes might also make this approach difficult (Barton & Gale 1993).

In other cases, it may be possible to identify populations in which all alleles of interest are segregating independent of a hybrid zone. For example, if the genetic variant of interest in a locally adapted population was selected from standing genetic variation, then this variant may still be segregating in the ancestral population as well. If so, alternate alleles can be tested in the genetic background of the ancestral and/or novel population. One study that used this strategy investigated the gene underlying defensive bony lateral plates in the threespine stickleback (Barrett et al. 2008). The threespine stickleback is a small fish that colonized and has adapted to hundreds of freshwater lakes from an ancestral marine environment. This transition into a freshwater environment is associated with a reduction in lateral plates and up to 70% of the phenotypic variation in plates is explained by an allele at the Ectodysplasin-A (Eda) locus (Colosimo et al. 2005). The ‘low’-plated allele at Eda is a standing genetic variant found at low levels in the ancestral marine population amidst the more frequent ‘full’ allele (Colosimo et al. 2005). Using a marine population, Barrett et al. (2008) were able to isolate the ‘low’- and ‘complete’-plated Eda alleles by collecting phenotypically partially armoured marine fish, predicted to be heterozygotes. Using these fish helped to account for background genetic variation because heterozygous marine fish were genetically ancestral with the exception of their Eda genotype and tightly linked loci (Barrett et al. 2008).

Although effective, this strategy can be quite time-consuming; Barrett et al. (2008) examined over 35 000 fish to find 354 that were partially plated. Of these 354 fish, only 182 were confirmed Eda heterozygotes. This discrepancy between phenotype and predicted genotype highlights the importance of modifier loci (i.e. loci which alter the effects of the candidate locus). Clearly, the utility of using phenotypes to screen the large numbers of individuals from wild populations needed for these studies will be dependent on the percentage of phenotypic variation explained by the candidate gene, and how much modifier loci and interactions with other genes or the environment alter phenotypic expression of the trait of interest. Thus, studies of the impacts of variation at a candidate gene such as the one described above are likely to be most successful when examining semi-dominant alleles of large effect, because low-frequency alleles can be found in the heterozygous state by screening phenotypes instead of genotypes. De novo mutations, which are also an important sources of ecologically relevant variation (e.g. Linnen et al. 2009) can also be studied in a controlled genetic background as long as the ancestral alleles at the candidate loci remain present in the population containing the new alleles, there is sufficient migration among populations, or crosses can be made between these populations.

An alternative strategy to control for background genetic variation when testing the effects of variation at candidate gene(s) is to compare naturally occurring clonal lines of plants or animals (e.g. Collier & Rogstad 2004). The primary challenge of comparing clonal lineages is the possibility that additional mutations (other than those in the candidate gene of interest) may be present between the lines. Extensive genomic surveys are thus required to rule out this possibility. In some species, it may also be possible to take advantage of existing information on relationships among individuals (e.g. pedigree information) to statistically test the effects of a candidate allelic variant alleles in variety of genetic backgrounds or test the effects of alternate versions of an allele among family members (e.g. Gratten et al. 2007).

Controlling for background genetic variation: forward genetics

The second approach to controlling for background genetic variation when testing associations between genotype and phenotype involves the use of forward genetics, or controlled crosses. Depending on the type of crossing design, the gene of interest can be tested in either a randomized [e.g. F2 and advanced generation intercrosses, recombinant inbred lines (RILs), or identical-by-descent lines (IBD)] or uniform [e.g. near isogenic lines (NILs)] genetic background (see supporting information Fig. S1 for the crossing designs used to make these types of lines). For model organisms, such as Arabidopsis, mice, maize, Drosophila and Caenorhabditis elegans, RILs are readily available from a number of parental populations (reviewed by Kammenga et al. 2008; Shindo et al. 2007; Peters et al. 2007) and have been used to test the effects of ecologically relevant variation (e.g. Callahan et al. 2005; Kammenga et al. 2007). Identical By Descent (IBD) lines (Fig. S1b) are used to break down linkage disequilibrium surronding a gene of interest and are another option for testing a gene in a randomized genetic background. Successful examples of this approach include studies of haemoglobin alleles in low- and high-altitude populations of deer mice (Chappell & Snyder 1984) and PGI variants in sulphur butterflies (e.g. Watt 1977). Introgression lines, which include NILs (Fig. S1c) and chromosome substitution lines, can be used to test the effects of a focal gene against a uniform genetic background. This strategy was successfully used by Bradshaw & Schemske (2003) to assess the effects of the yellow upper locus on flower colouration and pollinator preference, in Mimulus lewisii and Minnulus cardinalis (Bradshaw & Schemske 2003).

The decision on whether to test a candidate gene’s function against a randomized (e.g. RILs) or uniform (e.g. NILs) genetic background depends the expected number of genes involved and their predicted effect size. For example, NILs are preferable to RILs when epistatic effects are not expected to be important (e.g. Bradshaw & Schemske 2003). NILs are also preferable when the candidate gene is thought to have a small effect on phenotype because its effects are less likely to be masked when another major QTL is also present as might occur in RILs (Keurentjes et al. 2007). On the other hand, RILs are preferable to NILs when epistatic interactions are expected to be important, because RILs maintain many combinations of alleles from both source populations (Keurentjes et al. 2007). Both RILs and NILs can be used to detect genotype by environment (G × E) interaction because each line is essentially clonal (e.g. Ungerer et al. 2003; Callahan et al. 2005).

However, most forward genetic approaches are limited in their ability to capture the genetic variation present in wild populations, as even most cutting edge designs are limited to a few sets of outbred parents (reviewed by Cavanagh et al. 2008). The time needed for many generations required to establish RILs and NILs will be a major limiting factor in the application of these approaches for most systems. Thus, for species with longer generation times, F2 crosses (although possibly less informative than RILs or NILs) are likely to be more tractable. Of course, all forward genetic techniques have the limitation that they can only be applied to organisms that can be easily bred and raised in the laboratory.

Controlling for background genetic variation: reverse genetics

Reverse genetics is a collective term for methods in which a gene’s sequence or function is altered by the investigator either by altering the DNA sequence of the candidate gene or by manipulating its expressed product. These approaches hold exceptional promise for isolating the effects of a single gene, but as with forward genetic approaches, reverse genetics are largely limited to organisms that can be bred, or at least raised, in the laboratory. There are two main approaches to directly altering the DNA sequence of a candidate gene: untargeted and targeted. The classic untargeted approach is random mutagenesis, but the resulting artificial mutants may not mimic the naturally occurring variants of a gene. Thus, mutagenesis can provide information on the link from genotype to phenotype, but this information may be limited in its ecological relevance. Targeted approaches for directly altering the sequence of a candidate gene include complete ablation of an allele (i.e. knockouts) or targeted gene replacement. Knockout approaches may suffer from problems similar to those of untargeted mutagenesis; knockouts may not mimic naturally occurring phenotypes. On the other hand, targeted gene replacement allows the investigator to alter DNA sequence at the candidate gene from one ecologically relevant allele to another, allowing these alleles to be tested in a constant genetic background. As such, targeted gene replacement provides an exceptionally clear picture of the relationship between genotype and phenotype with the potential to test variants equivalent to those observed in natural populations. The ability to perform targeted gene replacement is currently limited to a few model species, but techniques are developing rapidly (e.g. Choi et al. 2009; Shukla et al. 2009; Townsend et al. 2009; Yan et al. 2009).

An alternative to directly altering the sequence of a candidate gene of interest is to manipulate the amount of functional gene product that is present. RNA interference (RNAi) or morpholinos are commonly used techniques for knocking down RNA levels (e.g. Moczek & Rose 2009). Increasing the expression of a candidate gene is most often accomplished by inserting an extra copy of a gene into a cell (e.g. Feder et al. 1996; Abzhanov et al. 2004; Colosimo et al. 2005). Both knockouts and insertions are needed for transgenic complementation experiments, during which an allele of interest is inserted into the genetic background of a knockout (often heterologous) to see if a candidate gene restores the phenotype (e.g. Maloof et al. 2001; Zufall & Rausher 2003; Reeves et al. 2007).

Conceptually similar approaches to reverse genetics techniques such as RNAi and gene insertion involve the use of pharmacological agents to increase or decrease the amount of functional protein that is present (reviewed by Skromne & Prince 2008). Using drugs to alter the amount or function of gene product is often the easiest way to look at the links from genotype to phenotype because it is simple, inexpensive, and can be used in most organisms and at most life history stages. However, pharmacological agents have the potential to affect genes other than those targeted and vary widely in their specificity. Thus, only highly specific agents should be used when attempting to make a strong connection between a particular candidate gene and a phenotype.

Controlling for background genetic variation: combined approaches

The most convincing evidence for the effects of genetic variation on phenotype, performance and fitness come from studies that have combined different strategies to control for background genetic variation (Hoballah et al. 2007; Kammenga et al. 2007; Chiang et al. 2009). For example, Chiang et al. (2009) used all three methods of controlling for background genetic variation to investigate potential pleiotropic effects of FLC, a gene known to be involved in flowering time, on temperature-dependent germination in Arabidopsis. Naturally occurring FLC variants, overexpression of FLC in transgenic plants and NILs were examined in parallel and provided strong evidence for FLC’s effects on germination time (Chiang et al. 2009). In addition, Chiang et al. (2009) incorporated mechanistic knowledge of the hormonal regulation of both flowering and germination to design experiments to examine the mechanisms underlying the pleiotropic effects of FLC. Studies such these demonstrate the power of combining a mechanistic perspective with population genomics and quantitative genetics when attempting to understand the phenotypic consequences of naturally occurring genetic variation.

A roadmap for future research

In this section, we provide a roadmap for research programs aiming to test the importance of an identified candidate gene. Such a research program is likely to include three major components, that span the levels of biological organization depicted in Fig. 1: (i) a molecular biology component that can characterize the alternate alleles of a candidate gene in vitro (i.e. genotype → biochemical phenotype); (ii) a whole-organism component that can determine the impacts of candidate genes on cellular function, morphology, behaviour, physiology and performance in vivo (i.e. genotype → performance); and (iii) a component that assesses the fitness consequences of genetic variants either directly, with experimental tests in relevant environments, or indirectly by using comparative phylogenetic approaches and analyses looking for molecular signatures of selection (i.e. genotype → fitness).

Genotype → biochemical phenotype (in vitro tests of function)

The first step of most biochemical studies is to fully sequence the alternate alleles of a candidate gene. If the gene of interest is a protein, it can be expressed, purified and characterized in vitro (e.g. Watt et al. 1977; Place & Powers 1979; Maloof et al. 2001). Alternatively, it is possible to express the protein in a suitable cellular system and examine its function(s) within the cell (e.g. Protas et al. 2006). If a genetic variant is in a regulatory element, in vitro tests can also be used to determine whether this variant has functional consequences for gene expression (e.g. Schulte et al. 1997, 2000; Tung et al. 2009). Hoekstra et al. (2006) provide an example of the utility of linking genotype to biochemical phenotype as part of a research programme aimed at understanding the genetic basis of adaptation. Beach mice found along the light sandy dunes of Florida’s gulf coast have a lighter coat colour than do inland populations of this species and are thus better camouflaged when on these dunes – presumably as a result of predation pressure (see Table S1). Hoekstra et al. (2006) used quantitative genetics to demonstrate that alternate alleles at a candidate protein coding gene, the melanocortin-1 receptor gene (Mc1r), were responsible for up to 35% of variation in coat colour. As Mc1r is an important part of the signal transduction pathway regulating the production of melanin (which causes dark coat colour), they predicted that the allele associated with light coat colour would have reduced function, thus reducing the activity of the signal transduction pathway and resulting in less melanin formation. They tested this prediction by expressing the alternate Mc1r alleles in cultured cells and recording melanocortin receptor activity (Hoekstra et al. 2006). As predicted, the light coat colour allele of Mc1r generated less of the signal that induces melanin production.

Unfortunately, mechanistic knowledge is not available for all genes, making approaches that work ‘up’ through the hierarchy shown in Fig. 1 from gene to phenotype difficult in some cases. If necessary, an alternate approach that works in the opposite direction may be productive. For example, if a genome scan reveals the signature of strong selection in a gene with no known function, it may be possible to generate hypotheses about the impacts of this genetic variation based on correlated ecological and environmental factors and their relationship to organismal performance, which can then suggest physiological processes or morphological traits that might be of importance. In fact, exploration of the function of adaptively significant genetic variation in natural populations holds the promise of contributing to mechanistic biology by revealing potentially subtle functions of poorly understood genes, pathways and networks (Landry & Aubin-Horth 2007; Benfey & Mitchell-Olds 2008; Rockman 2008).

If there are multiple SNPs or other types of genetic variation in a candidate gene, each can be tested separately by constructing the alternate alleles by site-directed mutagenesis to determine which site represents the functional variant, or funSNP (e.g. Newcomb et al. 1997; Lozovsky et al. 2009). Site-directed mutagenesis can also be used to reconstruct ‘extinct’ alleles to examine the function of ancestral alleles that are no longer segregating in the population (reviewed by Thornton 2004).

Genotype → whole-organism performance (in vivo tests of function)

The specific experiments that must be performed to test the functional consequences of a genetic variant at higher levels of biological organization largely depend upon the type of gene to be tested and the species in which it is tested, but in all cases incorporating a priori knowledge about gene function will aid in hypothesis formulation and the design of relevant experiments. Ideally, experiments will also be informed by prior in vitro assessment of the alleles of interest (e.g. Genotype → biochemical phenotype’) and should be conducted under ecologically relevant environmental conditions.

Examining functions at these higher levels adds the complexity of potential epistatic interactions, which, while ecologically important, render testing mechanistic predictions difficult. It is thus imperative to control for background genetic variation, using the approaches outlined above, when first assessing the consequences of genetic variation in a candidate gene. However, once the effects of a genetic variant are established, it is then necessary to determine whether the candidate gene’s effects are context specific and dependent on epistatic interactions that may alter the links between genotype, phenotype and fitness.

Genotype → fitness

The links between genotype, phenotype and fitness can be tested directly, through laboratory- and field-based experiments, or indirectly with phylogenetic comparative methods (reviewed by Garland et al. 2005; Freckleton 2009) and methods that identify molecular signals of natural selection (reviewed by Nielsen 2005; Jensen et al. 2007). Indirect methods to test for evidence of selection can be used in systems where direct tests of fitness are not feasible, as a precursor to subsequent direct tests of fitness, or to provide a complementary line of evidence for the adaptive significance of genetic variation. Many studies have successfully integrated these indirect methods to provide evidence of selection in the context of a mechanistic research paradigm (e.g. Schulte et al. 1997; Wheat et al. 2006; Whitehead & Crawford 2006; Miller et al. 2007; Flowers et al. 2009). Laboratory-based experiments can isolate and control for environmental variables that may be correlated in the field and thus can help clarify the selective pressures acting upon a locus. Laboratory-based studies of the fitness consequences of genetic variation have been performed in a range of multicellular eukaryotes (e.g. Arabidopsis FRI (Callahan et al. 2005) and resistance genes (Korves & Bergelson 2004); Drosophila heat shock proteins (reviewed by Hoffmann 2008)]. However, laboratory-based measures of phenotypes, performance and fitness may not translate directly to field measures of fitness (e.g. Irschick 2003; Irschick et al. 2005), so they should, whenever possible, be complemented with experiments performed under natural or semi-natural conditions.

Although measuring fitness in the wild is a difficult task, it is one for which evolutionary ecologists have laid excellent groundwork (examples and techniques reviewed by Arnold 1983; Brodie et al. 1995; Ellegren & Sheldon 2008; Endler 1986; Hoekstra et al. 2001; Irschick et al. 2007; Irschick 2003; Kingsolver et al. 2001; Mitchell-Olds & Schmitt 2006; Pigliucci 2003; Primack & Hyesoon 1989; Schluter 2000). For example, common garden and reciprocal transplant experiments can be used to test for differences in fitness among populations. The fitness of different ecotypes in these experiments is assessed through the quantification of fitness components, measuring population growth or directly competing individuals from the different ecotypes and measuring the contribution of each ecotype to following generations (see Kawecki & Ebert 2004). To link genotype to fitness, these classic techniques need only be modified such that the genotypes of individuals used in fitness experiments are known for the candidate gene of interest, and variation at other loci is controlled.

As yet, few studies have controlled for background genetic variation when experimentally testing the effects of genotype on fitness in the wild, but some studies have successfully used reverse genetics [e.g. Arabidopsis (Tian et al. 2003; Frenkel et al. 2008) and Drosophila (Sorensen et al. 2009)] or forward genetic approaches [e.g. Helianthus (Lexer et al. 2003), Arabidopsis (Ungerer & Rieseberg 2003), Avena barbata (Latta 2009) and barley (Verhoeven et al. 2004)]. Note that most experiments using forward genetics have looked at the effects of a large genomic region and not a single gene (often by mapping fitness components and phenotypes of interest in combination), but once the genes underlying the traits of interest are identified in these systems it will be possible to make direct linkages between genotype and fitness. Naturally occurring variation has also been used to test the links from genotype to fitness components (e.g. Watt et al. 1983; Korves et al. 2007), but to our knowledge, only one study has succeeded in directly measuring the fitness of alleles associated with a candidate gene in the wild while controlling for background genetic variation: a field experiment measuring selection on the alleles at the Eda gene in stickleback (Barrett et al. 2008).


Understanding the functional, ecological and evolutionary consequences of genetic variation requires adopting an integrative approach that combines molecular biology, physiology, quantitative genetics, ecology and evolutionary biology and is informed by a mechanistic perspective. To effectively establish the links between genetic variants and in vivo phenotypes, whole-organism performance and fitness, it is key that experimenters control for background genetic variation. Only then can clear causal relationships be established. Although many of the key experimental methods to control for genetic background are most easily applied in certain systems (e.g. those with short generation times and that are easily bred in the laboratory), the development of controlled crosses (including RILs and NILs) and transgenic technologies in well chosen, ecologically informative species will pay enormous dividends (reviewed by Abzhanov et al. 2008; Ellegren & Sheldon 2008; Ungerer et al. 2008). Most studies to date have examined the effects of a single gene, but many ecologically relevant traits are expected to be influenced by multiple genes and their interactions. In this context, a mechanistic perspective may be even more illuminating, because it can provide information about the potential interactions among genes.

We predict that first identifying phenotypes of interest (ideally under field conditions) and then working down through the levels of biological organization to genotype and up to fitness may be a particularly fruitful approach to forging the links between genotypes, phenotypes and fitness. In particular, combining this mechanistically informed perspective with population genomic and quantitative genetic approaches (e.g. Rogers & Bernatchez 2007) should enhance the chances of successfully identifying and understanding the consequences of ecologically relevant genetic variation. In addition, examining whole-organism performance or morphological traits for which the underlying mechanistic basis is well understood (e.g. growth, metabolism, locomotion, flowering time, colouration) will aid in the generation of clear predictions about the physiological systems, morphological traits, metabolic networks, pathways and genes involved in the phenotype of interest. This approach facilitates the identification of ecologically relevant genetic variation and helps to generate hypotheses about the potential pleiotropic effects (e.g. Chiang et al. 2009) of variation at the identified candidate gene and potential epistatic interactors (e.g. Caicedo et al. 2004).

By testing hypotheses about the links between genotype, phenotype and fitness under ecologically relevant conditions while controlling for background genetic variation, we can take into account the potential ‘frailties of adaptive hypotheses’ (Lynch 2007) and the pitfalls of the adaptationist programme (Gould & Lewontin 1979). Although the effort required to apply this approach is substantial, the potential benefits of further integrating mechanistic and quantitative genetic approaches into molecular ecology are enormous.


This work was financially supported by the Natural Sciences and Engineering Research Council of Canada (NSERC) through discovery grants to P.M.S and S.M. R. and a Canada Graduate Scholarship to A. C. D. A. C. D was also supported by a UBC University Graduate pre-doctoral fellowship. Special thanks to R. D. Barrett, L. Bernatchez, H. Collin, A. C. Gerstein, D. Irschick, T. H. Vines and two anonymous reviewers for their perceptive comments on earlier drafts of this manuscript that substantially improved this review. We thank E. D. Brodie III and Ward B. Watt for providing us with the photographs used in Fig. 2. We apologize to colleagues working on relevant projects that we could not cite because of space limitations.

ACD is completing her PhD thesis in PMS’s laboratory. Her dissertation focuses on the genetic, physiological, and morphological underpinnings of intra-specific variation in exercise performance. SMR’s laboratory studies molecular evolutionary mechanisms for coping with environmental change by integrating ecological genomics and quantitative genetics with field studies of natural selection. In her research, PMS takes an integrative approach to understand the evolution and mechanistic basis of inter- and intra-specific variation in physiological traits.