The relationship between genotype and phenotype is a central concern of many fields, from developmental biology to human genetics to evolutionary biology. Although in most cases this relationship is poorly understood, some general properties do seem to be shared across diverse systems. Chief among these is robustness to genetic and environmental variation [1
]. That is, most species maintain abundant genetic variation and experience a wide range of environmental conditions, yet phenotypic variation is relatively low [2
]. Because of its ubiquity, phenotypic robustness, also termed canalization or buffering, is worthy of study in its own right [3
]. It also presents an apparent contradiction: if biological systems are so robust, how do they diverge and adapt through evolutionary time?
The contradiction might be resolved if the robustness itself were to be modulated by particular mutations or environmental conditions. The robust system would accumulate conditionally neutral, or “cryptic,” genetic variation. A genetic or environmental perturbation that impaired the system's robustness would then reveal the cryptic variation in the form of greater phenotypic diversity. The modulation of robustness would not only allow evolutionary divergence, but it might also accelerate it relative to the slow, step-wise fixation of fitness-increasing alleles that is normally considered within the Neodarwinian paradigm [3
]. It is therefore essential to investigate mechanisms that contribute to the robustness of biological systems, and to understand how such mechanisms determine the phenotypic effects of different sources of variation.
A model for the buffering and release of variation is provided by the molecular chaperone Hsp90, which targets a large set of signal transduction proteins. In both Drosophila
, compromised Hsp90 function results in diverse morphological changes that exhibit strong dependence on the genetic background [6
]. This implies that Hsp90 normally contributes to phenotypic robustness to genetic variation. Because Hsp90 function allows stores of genetic variation to build up, and Hsp90 impairment releases this variation to have phenotypic effects, it is termed a “phenotypic capacitor” [7
Controversy surrounds the evolutionary relevance of Hsp90-mediated capacitance and any similar mechanisms that might exist. One issue is whether any fraction of the phenotypic variation revealed by an impaired capacitor is adaptive, or instead whether the variants consist entirely of hopeful, but ultimately unfit, monsters [8
]. The major morphological defects seen originally in flies [7
] support the latter conclusion, yet selection for one such defect did not cause correlated fitness costs, suggesting that the pleiotropic effects of Hsp90 impairment are modular and not unconditionally deleterious [10
]. The variation seen in Arabidopsis
also supports the potential adaptive value of capacitance, in that this variation is considerably less monstrous than that seen in flies [6
]. A more systematic analysis of the phenotypic and fitness effects of capacitor impairment is needed to resolve this issue.
A second debate concerns the ultimate evolutionary reason that capacitance exists. One view is that the ability to modulate evolvability is itself an adaptive trait, and that natural selection has therefore favored capacitor function [11
]. This view generally meets with great skepticism, as do similar views on the evolutionary benefits of mechanisms that alter mutation or recombination rates [9
]. Nonetheless, a population-genetic model has shown that an allele that modifies the rate of revelation of cryptic genetic variation can invade a population under a realistic range of parameter values [12
]. Although adaptive evolution of capacitance therefore remains a formal possibility, many favor an alternative view that considers capacitance a side effect of other selected properties. One possibility is that natural selection favors mechanisms that buffer against environmental variation, with environmental variation taken to mean both large macro-environmental differences and stochastic fluctuations in the external micro-environment or internal cellular environment. The buffering of genetic variation then results from a hypothesized mechanistic congruence between the impacts of allelic variation and environmental variation on regulatory networks [14
]. Another possibility is that regulatory networks inherently attenuate variation because they contain thresholds and other nonlinearities that allow them to respond properly to internal or external cues [8
]. Indeed, our own simulations of evolving regulatory networks predicted that many gene products should act as phenotypic capacitors, contributing to phenotypic robustness when present and producing greater phenotypic diversity when absent [16
The above considerations motivate the development of an experimental system in which many phenotypes can be precisely measured in many individuals, multiple gene products can be screened for capacitor function, and sources of variation can be precisely controlled and partitioned. Here we present such a system, using single-cell morphological phenotypes in the yeast S. cerevisiae. We focus here on the robustness of these phenotypes to environmental variation caused by stochastic fluctuations in a constant macro-environment. While the study of robustness to environmental variation is critical to understanding the development and physiology of organisms, it also lays the foundation for future work that will rigorously test the congruence between mechanisms of environmental and genetic buffering and that will investigate the impact of capacitors on evolutionary trajectories.
To identify gene products that contribute to robustness to environmental variation, we take advantage of data from high-dimensional quantitative morphological phenotyping of 4,718 haploid yeast single-gene knockout (YKO) strains [17
]. Phenotyping was performed by growing cells in rich media to logarithmic growth phase and triply staining them for the cell surface, actin cytoskeleton, and nuclear DNA. Digital micrographs of ~200 cells per strain were processed using automated image analysis [17
], yielding means and variances for 220 diverse quantitative phenotypes for each YKO (A). The phenotypes include measures of the size and shape of mother and bud cells and their nuclei, the number and size of actin patches, the position of nuclei or actin patches in reference to other cell landmarks, and relationships between the mother and bud, such as the bud angle (for a complete list, see [17
Genome-Wide Screen for Phenotypic Capacitors in S. cerevisiae
Using these data we identify more than 300 gene products required for robustness to environmental variation. We find that these capacitors are involved in a number of critical cellular processes and that they are highly connected, in terms of both physical and genetic interaction networks. Despite this centrality, capacitor deletions result on average in decreases in growth rate that could allow these mutants to persist for many generations in the presence of wild-type cells, suggesting that capacitor impairment need not produce unfit monsters. Capacitors encoded by a member of a duplicate gene pair differ in their functional and network properties from those encoded by singletons, suggesting that these two classes of capacitors are likely to buffer environmental variation by different mechanisms.