REVIEWS The causes of evolvability and their evolution Joshua L. Payne1 and Andreas Wagner2* Abstract | Evolvability is the ability of a biological system to produce phenotypic variation that is both heritable and adaptive. It has long been the subject of anecdotal observations and theoretical work. In recent years, however, the molecular causes of evolvability have been an increasing focus of experimental work. Here, we review recent experimental progress in areas as different as the evolution of drug resistance in cancer cells and the rewiring of transcriptional regulation circuits in vertebrates. This research reveals the importance of three major themes: multiple genetic and non-genetic mechanisms to generate phenotypic diversity , robustness in genetic systems, and adaptive landscape topography. We also discuss the mounting evidence that evolvability can evolve and the question of whether it evolves adaptively. Isogenic populations Evolvability research is now entering its fourth decade. different as the evolution of antibiotic resistance in bac- Populations of individuals with Although the term was first used as early as 1932, teria and the evolutionary rescue of populations threat- the same genotype. evolvability as a scientific subdiscipline of evolution- ened by climate and other environmental change. Their ary biology is often associated with a 1989 article by insights fall into three major categories, which provide a Phenotypic plasticity 1 The ability of one genotype Richard Dawkins describing what are now called dig- scaffold for this Review. 2 to produce more than one ital organisms . Today, research on evolvability is inte- The first major category encompasses molecular phenotype in response to gral to multiple fields, including population genetics, mechanisms that create phenotypic heterogeneity and different environmental stimuli. quantitative genetics, molecular biology and develop- do so not just through DNA mutations but even in the mental biology. Not surprisingly then, this diversity of absence of such mutations. These mechanisms have Modularity 3 The extent to which a system research has led to various definitions of evolvability . become central to evolvability research because they can be partitioned into distinct We here focus on one of them because we consider it allow isogenic populations to create phenotypic varia- components. the most fundamental: evolvability is the ability of a tion, some of which may facilitate survival in new or biological system to produce phenotypic variation that rapidly changing environments and may thus provide Pleiotropy When one gene or one is both heritable and adaptive. The definition is funda- time for an advantageous phenotype to be reinforced or mutation affects multiple mental because, first, heritable phenotypic variation is stabilized via DNA mutation, gene duplication, recombi- phenotypes. the essential raw material of evolution. Second, unless nation or epigenetic modification. The second category a biological system has the potential to produce varia- of evidence revolves around robustness, which is cen- tion that is adaptive (beneficial) in some environments, tral to evolvability because it allows an evolving popu- adaptation by natural selection is impossible. Third, the lation to explore new genotypes without detrimentally definition is broad enough to apply to fields as different affecting essential phenotypes. The resulting genotypic as population genetics and molecular biology, which diversity may serve as a springboard for subsequent study evolvability in different ways3. mutations to generate novel phenotypes, or it may bring Most early evolvability research was theoretical forth new phenotypic variation when the environment or guided by few experimental studies1,3–11. This has changes. The third category of evidence regards the topo- changed. Research on evolvability is becoming increas- graphical features of an adaptive landscape, such as its 1Institute of Integrative ingly experimental and driven by advances in high- smoothness, and a population’s location within such a Biology, ETH Zurich, throughput technologies (BOX 1). The observations landscape. These factors determine the amount of adap- Zurich, Switzerland. from such experiments are providing a mechanistic tive phenotypic variation that mutation can bring forth. 2Department of Evolutionary understanding of how living systems generate herita- Adaptive landscapes provide a useful geometric frame- Biology and Environmental ble adaptive variation12. We focus this Review on such work to encapsulate genotype–phenotype (or fitness) Studies, University of Zurich, Zurich, Switzerland. experimental studies, which come from a diversity of relationships that affect evolvability. *e- mail: andreas.wagner@ fields, ranging from developmental to cancer biology. Unfortunately, space constraints prevent us from ieu.uzh.ch Many make no explicit mention of evolvability, yet they reviewing other important aspects of evolvability https://doi.org/10.1038/ all shed light on the causes of evolvability and some also research, including the roles of phenotypic plasticity, organ- s41576-018-0069-z on its evolution. They are relevant for phenomena as ismal development, modularity and pleiotropy, as well as NATURE REVIEWS | GENETICS REVIEWS Box 1 | Methodological advances they may simply ‘buy time’ for a population to adapt in other ways to an environmental challenge (FiG. 1a). Our ability to study the molecular causes of evolvability has been greatly improved by recent methodological advances. For example, our growing understanding of Stochastic gene expression. Stochastic gene expression, phenotypic heterogeneity is driven by microfluidic devices and time- lapse microscopy, or gene expression noise, has multiple causes, including which provide information about the compositions, morphologies and growth rates of single cells in dynamic environments186. Complementary information is provided by the varying efficiency of transcription and transla- 19,20 methods such as fluorescence in situ hybridization and single-cell RNA sequencing tion as well as the regulation of gene expression by (RNA- seq), which describe the location and abundance of mRNA transcripts, low- abundance molecules whose numbers fluctuate respectively187,188. Combined with whole-genome sequencing, such methods have randomly in a cell21 (FiG. 1b). Stochastic gene expression detailed the molecular causes of phenotypic heterogeneity, such as how stochastic can create non-genetic, adaptive diversity in phenotypes gene expression drives persistence in bacteria26 and rare cell variability in cancer24. as diverse as viral latency, bacterial competence, antibiotic Non- single-cell methodologies have also furthered our understanding of phenotypic resistance, as well as drug resistance in cancer22–24. heterogeneity. For example, ribosome footprint profiling, which characterizes the One example in which stochastic gene expression 189 distribution of ribosomes on mRNA transcripts , has detailed the prevalence of causes adaptive phenotypic variation is persistence, stop- codon readthrough in yeast, fly and human39. in which some cells in an isogenic population exhibit Several methodological advances have improved our understanding of mutational robustness and of adaptive landscapes. For example, approaches that characterize a a physiologically dormant phenotype called a per- 25 small region of an adaptive landscape typically rely on deep mutational scanning139, sister phenotype . This phenotype is adaptive because a method that combines systematic mutagenesis with high-throughput phenotypic a dormant subpopulation has the potential to survive assays. These assays include fluorescence- activated cell sorting, which can be used to drugs that require active growth for killing, affording measure protein functions such as fluorescence or ligand binding, as well as EMPIRIC190, the persistent subpopulation time to acquire resistance- which can measure the fitness of many cells in parallel. To capture the effects of conferring DNA mutations. This phenomenon was mutations in their native genomic context, genome-editing tools such as CRISPR–Cas9 recently demonstrated in a laboratory evolution exper- 103 can be used to introduce mutations to specific chromosomal loci . Approaches iment of Escherichia coli populations subjected to inter- that exhaustively characterize an entire (small) genotype space have profited from mittent exposures of ampicillin26, in which persistence chip- based technologies that simultaneously assay the phenotypes of all possible served as a stopgap until some individuals acquired genotypes93, as well as from high-throughput in vitro selection methods that systematically enrich an initially random library of sequences for those sequences that resistance- causing mutations. perform a particular function, such as binding a ligand147. Persistence arises in only a small fraction of a pop- To understand how these causes of evolvability have changed over long evolutionary ulation; therefore, one might think that the resulting timescales, they are often combined with maximum likelihood methods to statistically population bottleneck would hinder evolvability by reduc- infer and experimentally reconstruct the genotypes and phenotypes of ancient ing the supply of beneficial mutations. However, a recent macromolecules191. study of non- small-cell lung cancer indicates that this need not be the case27. These cells stochastically express a theoretical advances.