While adaptive evolution is a central tenet of modern biology, studies of its associated processes remain difficult. The primary obstacle to such investigations is the genetic complexity that underlies individual traits; the "one gene, one
Mitchell B. Cruzan1 and Jennifer M. Rhode2. Department of Biology, Portland State University, Portland, OR 97207, U.S.A.
Correct citation: Cruzan, M.B., and Rhode, J.M. 2004. Experimental analysis of adaptive landscape topographies. In Plant Adaptation: Molecular Genetics and Ecology. Proceedings of an International Workshop held December 11-13, 2002, in Vancouver, British Columbia, Canada. Edited by Q.C.B. Cronk, J. Whitton, R.H. Ree, and I.E.P. Taylor. NRC Research Press, Ottawa, Ontario. pp. 61-69.
Corresponding author e-mail address: [email protected] 2Current address: Georgia College and State University, Department of Biology and Environmental Science, CBX 081, Herty Hall 202, Milledgeville, GA 31061, U.S.A.
product" model fails to describe most living systems. Research into adaptive evolution is further confounded by phenotypic plasticity, which allows complex morphological traits to be expressed differentially across space and time. In the absence of much experimental evidence, our view of adaptive evolution has been shaped by the heuristic tools of Wright (Wright 1931; 1932; 1988) and of Fisher (1930). Both authors used the metaphor of a fitness landscape with regions of low and high fitness to describe evolution in populations. These models, while not providing specific hypotheses for testing and analysis, summarize contrasting views of the adaptive evolution process.
Models of both Wright and Fisher allow fitness optima to shift in response to environmental changes and emphasize adaptive "hill climbing", but they differ in the relative importance of epistatic (unequal expression of a mutant allele in different backgrounds) and additive (equal effects of a mutant allele in a variety of genetic backgrounds) genetic variation. In Wright's model, interactions among alleles at different loci (epistasis) explain a large proportion of the genetic variation. Populations conforming to this model inhabit a rugged fitness surface with multiple high-fitness peaks, and their evolution is characterized by periods of stasis interrupted by rapid changes (peak shifts) (Simpson 1944; Arnold et al. 2001). Unlike Wright, Fisher envisioned evolution as a gradual process, with allelic substitutions enabling populations to move steadily across a relatively smooth incline towards a single fitness optimum. These contrasting views of genetic architecture of traits, and the varied topographies that result, continue to be a subject of active debate (e.g., Coyne et al. 1997; Wade and Goodnight 1998). Though opinions on topographic specifics differ, fitness surfaces are considered a useful metaphor for understanding adaptive evolution.
Wright's and Fisher's models of adaptive evolution predict that allele substitution will have different effects on genotypes occupying the same environment; Wright's rugged landscape is the consequence of fitness effects of specific combinations of alleles at separate loci, while Fisher's model predicts that alleles contributing to fitness differences will have similar effects regardless of genetic background. For example, alleles contributing to an increase in the size of floral displays may have positive fitness effects in a broad range of genotypes, but alleles affecting expression of characters that are part of a pollination syndrome (e.g., red petals) may only have positive fitness effects in the presence of alleles that produce other specific characters (e.g., tubular flowers). This simple example serves to illustrate the salient differences between these views and highlights the fact that they need not be mutually exclusive.
Wright's vision of a rugged adaptive landscape has garnered support as it has been widely used as a contextual framework for theoretical and empirical investigations of adaptive evolution (Barton and Charlesworth 1984; Whitlock et al. 19956; Wade and Goodnight 1998). This model, which is largely verbal with graphical depictions of the fitness surface, has been modified since its introduction (Wright 1931; Wright 1932). Wright's original depiction of a multilocus genetic surface (Wright 1932; Provine 1986) was later refined to an allele frequency surface for two loci (Wright 1988; Whitlock et al. 1995a), and this genetic landscape has been interpreted as a phenotypic surface for paired traits (Pearson 1903; Simpson 1944; Lande 1976; Arnold et al. 2001; Cruzan 2001). The key feature shared by these depictions of adaptive evolution is the critical role of epistasis; only interactions among alleles at different loci can produce a rugged fitness topography when the environment is homogeneous.
While theoretical analyses that assume a rugged fitness surface have become common (Barton and Charlesworth 1984; Wade and Goodnight 1998), empirical assessments of adaptive landscape topography are relatively scarce. Pheno-typic landscapes are the most amenable to empirical analysis and have become popular as tools for making direct assessments of fitness topographies (Kingsolver 1988; Armbruster 1990; Cresswell and Galen 1991; Schluter 2000). Recent analyses of gene frequency landscapes have focused on the relative importance of epistasis (Whitlock et al. 1995a; Armbruster et al. 1997; Kim and Rieseberg 2001) and on evidence for the existence of specific gene combinations representing separate adaptive peaks (Wade and Goodnight 1991; Korona et al. 1994; Lenski and Travisano 1994; Cluster and Allard 1995; Burch and Chao 1999; Kaltz and Bell 2002). Most of these analyses indicate that adaptive landscapes are somewhat rugged, but the resolution of many of these studies are too low to assess the depth of valleys (i.e., the relative importance of epistasis) and the presence of ridges that may connect separate adaptive peaks.
Since many loci probably contribute to fitness, and numbers of potential interactions among them are exponentially higher, it has been suggested that Wright's vision of a two-locus, two-dimensional landscape may be inadequate to accurately represent fitness relationships among multilocus genotypes (Whitlock et al. 1995a; Gavrilets 1997a). Gavrilets and colleagues (Gavrilets 19976; Gavrilets 1997a; Gavrilets 1999) approached this problem from a theoretical perspective and determined that epistasis among a large number of loci affecting fitness will result in a multidimensional genetic landscape that retains many of the characteristics of two-dimensional rugged landscapes. However, a critical difference develops at high levels of dimensionality. Here, clusters of adaptive genotypes become interconnected by "ridges" of high fitness, which are the direct consequence of epistasis among a large number of loci contributing to fitness. These ridges represent paths of high fitness that form a network of interconnected regions rather than the isolated peaks that are characteristic of systems with few dimensions. Based on these results, Gavrilets concluded that, with a sufficient number of interacting loci, populations could easily
Fig. 1. Graphical depiction of divergent evolution along the rim of a fitness "hole" conforming to Gavrilets' (1997a) metaphor of a holey adaptive landscape. The arrows indicate the expected relative success of crosses between the progenitor population (A) and populations at different stages of divergence (B-E). Hybridization between newly derived populations (A to B) results in a large proportion of second-generation hybrid genotypes falling along the fitness "ridge". For crosses to more divergent populations, an increasing number of second-generation hybrids display low fitness recombinant genotypes (i.e., they fall into the fitness "hole").
move between "peaks" by traversing "ridges" of high fitness through mutation-drift processes. Speciation would result when populations came to occupy opposite sides of a low fitness "hole" (i.e., such that crosses between them only produced inviable hybrids) without ever having to pass through an adaptive valley (Fig. 1).
The metaphor of a holey adaptive landscape provides insights into speciation via accumulation of mutations in divergent lineages (Dobzhansky 1936; Muller 1942; Orr 1995; Coyne and Orr 1998; Orr 2001). In the Dobzhansky-Muller model for evolution of reproductive isolation, a lack of fitness effects of mutations that accumulate in separate lineages is contingent upon the presence of previously fixed alleles at different loci in the same genome (Orr 1995). However, when mutations from two different lineages are combined through hybridization, reduced fitness can result from interaction between mutant, divergent alleles (Orr 2001). Hence, speciation can proceed without loss of fitness via a chain of intermediate genotypes separated by single mutational steps. The Dobzhansky-Muller model is consistent with conditions expected under a holey adaptive landscape because it is dependent on the presence of ridges that allow population divergence and reproductive isolation to develop without severe fitness losses.
This view of a "holey" adaptive landscape is compelling because the resulting topology is a direct consequence of extrapolating from the simple two-dimensional case of Wright to a multidimensional level (Gavrilets 1997a). Moreover, there is empirical evidence that it is common to have ridges of high fitness that connect adaptive regions. For example, experiments have demonstrated that separate microbial populations derived from the same ancestral population will tend to diverge over time without losing fitness compared to progenitors (Korona et al. 1994; Lenski and Travisano 1994). Indeed, evolution of these populations was accompanied by fitness increases reminiscent of a mutation/selection process expected under the type of strictly additive genetic model considered by Fisher (1930). However, the microbial populations in these experiments can also undergo morphological divergence (Korona et al. 1994) as they occupy dif ferent regions of the adaptive landscape (rather than a single region as expected under Fisher's model). The resulting populations can also be characterized by different average fitnesses (adaptive peaks of differing heights) that are retained over many thousands of generations (Lenski and Travisano 1994), suggesting that they may have become trapped in regions of the adaptive landscape that are characterized by low connectivity (i.e., fewer ridges). While these experiments are intriguing, more analyses in a wider variety of organisms are needed before any general conclusions can be drawn concerning the ruggedness of adaptive landscapes and levels of connectivity among high fitness regions.
Was this article helpful?