The Quarterly Review of Biology THE EVOLUTIONARY GENETICS OF CANALIZATION

Evolutionary genetics has recently made enormous progress in understanding how genetic variation maps into phenotypic variation. However, why some traits are phenotypically invariant despite apparent genetic and environmental changes has remained a major puzzle. In the 1940s, Conrad Hal Waddington coined the concept and term “canalization” to describe the robustness of phenotypes to perturbation; a similar concept was proposed by Waddington’s contemporary Ivan Ivanovich Schmalhausen. This paper reviews what has been learned about canalization since Waddington. Canalization implies that a genotype’s phenotype remains relatively invariant when individuals of a particular genotype are exposed to different environments (environmental canalization) or when individuals of the same singleor multilocus genotype differ in their genetic background (genetic canalization). Consequently, genetic canalization can be viewed as a particular kind of epistasis, and environmental canalization and phenotypic plasticity are two aspects of the same phenomenon. Canalization results in the accumulation of phenotypically cryptic genetic variation, which can be released after a “decanalizing” event. Thus, canalized genotypes maintain a cryptic potential for expressing particular phenotypes, which are only uncovered under particular decanalizing environmental or genetic conditions. Selection may then act on this newly released genetic variation. The accumulation of cryptic genetic variation by canalization may therefore increase evolvability at the population level by leading to phenotypic diversification under decanalizing conditions. On the other hand, under canalizing conditions, a major part of the segregating genetic variation may remain phenotypically cryptic; canaliza288 Volume 80 THE QUARTERLY REVIEW OF BIOLOGY tion may therefore, at least temporarily, constrain phenotypic evolution. Mechanistically, canalization can be understood in terms of transmission patterns, such as epistasis, pleiotropy, and genotype by environment interactions, and in terms of genetic redundancy, modularity, and emergent properties of gene networks and biochemical pathways. While different forms of selection can favor canalization, the requirements for its evolution are typically rather restrictive. Although there are several methods to detect canalization, there are still serious problems with unambiguously demonstrating canalization, particularly its adaptive value. T STUDY of phenotypic variation is a central theme in evolutionary biology. Natural selection results from variation in fitness among individuals; the response to selection depends on the heritable determinants of the phenotype. Phenotypes are the product of developmental processes that depend both on the genotype and environment and their interaction. It is thus important to understand how genetic variation maps to phenotypic variation, and how this genotypephenotype map is influenced by genetic and environmental change. Although evolutionary biology is to a major degree concerned with the study of variation, we have only a limited understanding of the absence of variation. Why are some traits phenotypically invariant despite apparent genetic and environmental changes? Why are some traits less phenotypically variable in some taxa, but not in others? One of the most intriguing observations in evolutionary genetics is that wild-type populations often harbor vast amounts of hidden genetic variation, this variation being phenotypically expressed only in particular environments or genetic backgrounds (e.g., Gibson et al. 1999; Gibson and Dworkin 2004). Thus, there seems to be a strong robustness of some phenotypes against genetic and nongenetic change or perturbation. More generally, the amount and quality of phenotypic variation can differ dramatically within and among populations. Some traits are highly invariant within species while simultaneously being highly variable among closely related species; other characters seem to be highly conserved among species or clades. Why is this the case? Similarly, the same trait may be more variable in some taxa than in others. Does robustness occur in all taxa and for all traits to the same extent? Some of the differences in the amount and quality of phenotypic variation may well be explained by classical scenarios, such as selection or drift. Yet, it remains a fascinating possibility that organisms may have evolved specific mechanisms that make them insensitive to genetic and environmental change, thereby decreasing their capacity for evolutionary change. Alternatively, the molecular details of the functional architecture of complex phenotypic traits may lead to robustness as a nonadaptive byproduct, an emergent property. This commonly observed robustness of phenotypes was named “canalization” by Waddington in the 1940s to describe the mechanisms that cause the phenotype to be insensitive against genetic and nongenetic perturbations and change (Waddington 1942; see also Schmalhausen 1949). Canalization is highly relevant for evolutionary biology. For example, it implies that phenotypes may be stable around their fitness optimum despite genetic and environmental change (e.g., Rendel 1967). By keeping phenotypic variation low, canalization may constrain phenotypic evolution (e.g., Charlesworth et al. 1982; Maynard Smith et al. 1985) and provide a microevolutionary mechanism for character stasis (e.g., Stearns 1994). Canalization also allows genetic variation that is phenotypically not expressed to accumulate. This cryptic variation can lead to the appearance of new phenotypes when development is “decanalized,” for instance by environmental stress, thereby allowing evolutionary change (e.g., Rutherford and Lindquist 1998). Despite the long history of the canalization concept (Waddington 1942; Schmalhausen 1949; also see Hall 1992; Gilbert 2000; Slack 2002), the evidence for canalization is limited (e.g., Scharloo 1991; Gibson and G P Wagner 2000; De Visser 2003). The evolutionary role of canalization has remained puzzling: the September 2005 289 GENETICS OF CANALIZATION concept is difficult to define, and its predictions are often beyond the reach of experiments. Similarly, the molecular mechanisms governing canalization are poorly understood. Thus, not surprisingly, following the classical work by Waddington, Schmalhausen, and others, empirical and theoretical research on canalization declined, presumably due to the lack of suitable theoretical and genetic methods to tackle the problem of phenotypic robustness. However, recent advances in molecular developmental genetics and theoretical biology have set the stage for a comeback of the canalization concept among theoretical biologists (cf. Gibson and G P Wagner 2000), evolutionary geneticists (cf. De Visser et al. 2003), and developmental biologists (cf. Gerhart and Kirschner 1997; Hartman et al. 2001). Thus, recent theoretical work (e.g., G P Wagner et al. 1997; Rice 1998; Kawecki 2000; Siegal and Bergman 2002; Bergman and Siegal 2003; Hermisson et al. 2003; Hermisson and G P Wagner 2004; Proulx and Phillips 2005), evolutionary experiments (e.g., Stearns and Kawecki 1994; Stearns et al. 1995; Elena and Lenski 2001), and molecular studies (e.g., Gibson and Hogness 1996; Rutherford and Lindquist 1998; True and Lindquist 2000; A Wagner 2000b; Queitsch et al. 2002; Gu et al. 2003; Sollars et al. 2003; True et al. 2004) have renewed interest in canalization. This paper provides a comprehensive review of what has been learned about the evolutionary genetics of canalization since Waddington (for recent, shorter reviews, see Gibson and G P Wagner 2000; Meiklejohn and Hartl 2002; De Visser et al. 2003; Gibson and Dworkin 2004; also see the book edited by Hall and Olson 2003). I ask six questions: (1) How can canalization be defined? (2) What is the relationship between canalization, epistasis, and genotype by environment interactions? (3) At the proximate level, which molecular mechanisms may lead to canalization? (4) At the ultimate level, how does canalization originate? Is it a sideproduct or an emergent property of the genotypephenotype map, or is it shaped by natural selection? (5) What are the consequences of canalization for evolutionary processes? (6) How can canalization be measured? One of the most fundamental problems of evolutionary biology is to better understand the pathways that connect genotypes with phenotypes, the genotype-phenotype map (e.g., Lewontin 1974a; Wright 1977; Houle 1991, 2001; Schlichting and Pigliucci 1998). Canalization is but one aspect of the genotype-phenotype map; the general problem is to understand the functional architecture of complex phenotypes that depend upon manifold interactions among underlying genes and the environment. A better understanding of canalization will contribute to a clearer conceptual picture of how genes, development, and the environment interact to produce phenotypes. The Phenomenon of Canalization the stability of the wild-type Waddington based the concept of canalization on the observation that genotypes differ in their phenotypic reactions to genetic and environmental change, and that wildtype phenotypes are phenotypically much less variable than mutants or environmentallyinduced phenotypes (Waddington 1942; Gibson et al. 1999). In a similar vein, Schmalhausen (1949) argued that the “stability of the morphogenetic system is destroyed (rendered labile) due either to variation in environmental factors or to mutation” (Schmalhausen 1949/1986:79). Waddington (1957) suggested that the reason for the difference in variation between wild-type as opposed to mutant or environmentally-induced phenotypes is that the wild-type has been exposed to many generations of stabilizing selection, whereas the mutant has not (see also Schmalhausen 1949). Several experiments seem to support Waddington’s view by showing that stabilizing selection can reduce the variability of environmentally-induced or mutant phenotypes, and that phenotypic change becomes progressively more difficult when the trait approaches the wild-type pattern (e.g., Maynard Smith and Sondhi 1960; Waddington 1960). This has been taken as evidence for the adaptive canalization of the wild-type (cf. Scharloo 1991). However, in most experiments it is not clear whether stabilizing selection just decreased genetic variation or 290 Volume 80 THE QUARTERLY REVIEW OF BIOLOGY whether it selected for canalizing mechanisms itself. Furthermore, the wild-type concept is an ambiguous abstraction since natural wild-type populations are often highly genetically and phenotypically variable. See Scharloo (1991) for a comprehensive review of Waddington’s original view. definition of canalization Canalization is the reduced sensitivity of a phenotype to changes or perturbations in the underlying genetic and nongenetic factors that determine its expression (see also Meiklejohn and Hartl 2002; De Visser et al. 2003). Canalization is a relative term, and can thus only be defined as a matter of comparison. Thus, a phenotype P is more canalized than another phenotype P* if P remains relatively invariant when the singleor multilocus genotype G, which determines P, is exposed to different environments (environmental canalization) or located in different genetic backgrounds (genetic canalization): P is “resilient,” “robust,” or “insensitive” to genetic and/or environmental changes or perturbations. Canalization can therefore be recognized by observing that most genetic or environmental changes leave the phenotypic expression of G, and thus the phenotype P, invariant; the expression of G is changed such that specific phenotypic changes (PrP*) are induced only in some genetic backgrounds or environments (or combinations of genetic backgrounds and environments). Consequently, a canalizing allele or genotype G reduces the phenotypic variation of a trait across a range of genetic backgrounds and environments relative to a noncanalizing allele or genotype G*, and a canalized trait P exhibits a restricted range of phenotypic variation across genetic backgrounds and environments as compared to a noncanalized trait P* (Meiklejohn and Hartl 2002). The proximate (molecular) mechanisms causing canalization, as well as the nature of the perturbations, can be manifold. For example, the canalizing mechanisms can potentially be located at any level of the biological hierarchy, from gene expression, RNA stability, protein structure and folding, intermediate metabolism and physiology to morphology, behavior, and life-history traits. For instance, canalization may be observed at the level of morphology, but not at the level of gene expression. The buffering would then occur at some intermediate level between gene expression and morphology. Furthermore, to maintain a particular trait (e.g., body temperature in homeotherms) despite (e.g., thermal) perturbations, an organism may vary other traits (such as basal metabolic rate and dilation of blood vessels). Stability (homeostasis) at one level may depend upon lability or sensitivity at another level. Thus, it seems that there exists an intrinsic paradox of canalization: buffering at one level (the phenotype) may be coupled to higher variation at another level (e.g., gene expression). Variation in such buffering mechanisms can be heritable or nonheritable. Homology, when two or more structures are alike because of shared ancestry, may be another interesting aspect of the biological hierarchy related to canalization. Homology can occur between entities at different levels of the biological hierarchy; for instance some organisms may show homologies at the genetic level, but not necessarily at the morphological level (Laubichler 2000). Consequently, individuals from different species are often composed of the same kind of (structurally identical) building blocks (e.g., genes, organs, and traits). Such homologous structures may be a manifestation of canalization; canalization may potentially explain the stability of homologues, since canalization may limit or constrain variation in these structures (e.g., G P Wagner 1996; G P Wagner and Altenberg 1996; Laubichler 2000). A similar concept to homology is homoplasy, when two or more structures are alike but not due to common ancestry. It is an interesting, yet open empirical question whether and how canalization affects homology and homoplasy and whether, for example, homoplasy is more commonly observed in taxa with less canalized traits. Various authors have used different terminologies when referring to canalization or aspects of it (see Debat and David 2001; Meiklejohn and Hartl 2002). Autonomous development or autoregulation (Schmalhausen 1938, 1949), homeostasis, homeorhesis or bufSeptember 2005 291 GENETICS OF CANALIZATION fering (Bernard 1865; Cannon 1932; Lerner 1954; Lewontin 1956; Zakharov 1992; Kauffman 1993; Hallgrı́msson et al. 2002), developmental stability (Thoday 1955; Palmer and Strobeck 1986, 1992), epigenetic stability (A Wagner 1996), and robustness (e.g., Savageau 1971; Kauffman 1993; Little et al. 1999; De Visser et al. 2003) are essentially synonymous with canalization. Sometimes canalization is taken to mean adaptive, evolved robustness against heritable and nonheritable perturbations, whereas terms such as buffering refer to any kind of mechanism, adaptive or not, that will cause the phenotype to be resilient against perturbations (cf. DeVisser et al. 2003). See Hall and Olson (2003) for an excellent recent discussion of evolutionary developmental biology in general, including various issues bearing on canalization discussed in this review. Here I use the term canalization to mean any mechanism, structure, or process, adaptive or not, that will reduce a phenotype’s sensitivity to perturbations. an example of canalization When the function of the Hsp90 protein, a chaperone and heat shock protein, encoded by the hsp83 locus in Drosophila melanogaster is impaired by mutation or by the specific inhibitor geldanamycin, phenotypic variation increases both in laboratory and wild strains (Rutherford and Lindquist 1998). The degree of increase and the nature of the phenotypic variation depend on the genetic background and environment (e.g., temperature). A wide range of phenotypic effects is observed, including defects in bristles, eyes, halteres, legs, wings, the thorax, and abdomen. The authors showed that at least some of these phenotypes are produced by alleles that are not phenotypically expressed in the presence of the functional Hsp90 protein. This genetic variation is heritable, and the pattern of heritability is unlikely to be due to de novo mutation but to genetic variance in polygenic traits. Thus, impairment of Hsp90 uncovers previously silent genetic variation, which leads to an increase of phenotypic variation (decanalization). When decanalized lines, showing a high penetrance of a particular trait, are outcrossed with normal laboratory strains, the trait is expressed only at very low levels, implying a return to the canalized state. Thus, the functional hsp83 locus is a gene with canalizing effects, masking the effects of hidden genetic variation. Queitsch et al. (2002) confirmed the results of Rutherford and Lindquist (1998) by showing that reducing the function of Hsp90 in various Arabidopsis genotypes increases genetic variation in morphological traits. But does Hsp90 also buffer against environmental, nongenetic perturbations such as developmental noise? As work by Milton et al. (2003) demonstrates, this does not seem to be the case (but see Queitsch et al. 2002): in D. melanogaster, Hsp90 does not buffer against nongenetic perturbations as measured by fluctuating asymmetry in bristle traits. The Hsp90 system is undoubtedly the best current example of a molecular canalizing mechanism; it nicely illustrates how environmental cues (e.g., geldanamycin, temperature) or mutation can lead to the release of hidden genetic variation. Intriguingly, Sollars et al. (2003) have shown that Hsp90 can act through epigenetic mechanisms, whereby a reduced activity of Hsp90 causes a heritable change in the chromatin state (see True et al. 2004 for another example of epigenetic canalization). The importance of heat shock proteins for canalization has also been confirmed by Fares et al. (2002): the authors show that overexpression of the heat shock protein GroEL induces the recovery of fitness of Escherichia coli strains that have accumulated deleterious mutations. Yet, recent modeling work by Hermisson and G P Wagner (2004) casts some doubt on whether Hsp90 can really be seen as an example of canalization. Although canalization is an attractive explanation for the accumulation of cryptic genetic variation, theory clearly shows that canalization is not necessarily required to explain the buildup of cryptic variation. Accumulation of mutations at conditionally neutral loci for a sufficiently long time can lead to the accumulation of cryptic variation even in the absence of canalization (Hermisson and G P Wagner 2004). See Rutherford (2003) for a comprehensive review of the role of protein chaperones such 292 Volume 80 THE QUARTERLY REVIEW OF BIOLOGY as the Hsp family for canalization and buffering; see Scharloo (1991) for a review of some of the classic—but more ambiguous— examples of canalization.