Population perspectives on functional genomic variation in yeast

Abstract

Advances in high-throughput sequencing have facilitated large-scale surveys of genomic variation in the budding yeast, Saccharomyces cerevisiae. These surveys have revealed extensive sequence variation between yeast strains. However, much less is known about how such variation influences the amount and nature of variation for functional genomic traits within and between yeast lineages. We review population-level studies of functional genomic variation, with a particular focus on how population functional genomic approaches can provide insights into both genome function and the evolutionary process. Although variation in functional genomics phenotypes is pervasive, our understanding of the consequences of this variation, either in physiological or evolutionary terms, is still rudimentary and thus motivates increased attention to appropriate null models. To date, much of the focus of population functional genomic studies has been on gene expression variation, but other functional genomic data types are just as likely to reveal important insights at the population level, suggesting a pressing need for more studies that go beyond transcription. Finally, we discuss how a population functional genomic perspective can be a powerful approach for developing a mechanistic understanding of the processes that link genomic variation to organismal phenotypes through gene networks.

Introduction

For the past two decades, Saccharomyces cerevisiae has been at the forefront of the genomics revolution. Budding yeast was the first eukaryote with a sequenced genome [1], the first eukaryote subjected to genome-wide transcriptional profiling [2] (and the first in which population genetic variation in genome-wide gene expression was assessed [3]), the first organism with a gene deletion collection of nearly all nonessential genes [4, 5] and the subject of some of the earliest proteomic analyses [6–9] and metabolomic studies [10]. Most of these pioneering studies were carried out in only a handful of laboratory strains, in particular strains derived from the reference strain, S288c. However, as a species S. cerevisiae exhibits extensive genetic diversity, with pairwise genetic divergence among strains in some cases exceeding 1% [11]. The high level of genetic diversity among strains of S. cerevisiae has an impact on functional genomic variation at all levels.

We use the term functional genomics as short hand for all dynamic aspects of genome biology, including gene expression and protein abundance, the myriad interactions between complex biomolecules that are essential for signal transduction, protein complex formation and function, the physical packing of the genome and the structure of and interactions within metabolic networks. In brief, all of these aspects help the genome to ‘come alive'. The combination of functional and population genomic approaches holds promise for providing novel insights into both genome function and the evolutionary process. Perhaps most importantly, detailed studies of how natural genetic variation affects functional genomic traits, and how this functional genomic variation leads to differences in cellular and organismal phenotypes, can provide critical insights into phenomena such as pleiotropy, epistasis and robustness that are critical for understanding the mapping from genotype to phenotype.

There has been a recent resurgence of interest in the ecology and natural genetic variation of S. cerevisiae and other yeasts [reviewed in 12, 13]. Several recent reviews have focused on the advantages of exploiting natural variation to dissect quantitative trait variation [14, 15] and the molecular basis of quantitative trait variation in yeast [16]. Similarly, there have been thorough overviews of specific methodologies and resources for functional genomics in S. cerevisiae [e.g. 17–19]. In the pages that follow we synthesize current knowledge of levels and patterns of functional genomic variation from a population, quantitative and evolutionary genetic perspective. We begin by reviewing the nature and extent of genome variation within S. cerevisiae, followed by a survey of the current understanding of population variation in functional genomic traits. We then summarize knowledge of transcriptional evolution revealed by model-based and experimental evolution studies, and discuss examples of how combining functional genomics with population and quantitative genetics can provide mechanistic insights into the genotype–phenotype map.

Genome variation provides ample substrate for functional genomic variation

Although the genome of S. cerevisiae was sequenced nearly 20 years ago [1], little was known at the time about levels of sequence variation present within the species. Early studies based on modest fractions of the genome made it clear that there was significant genetic variation among strains derived from natural isolates [20–24]. Subsequent studies surveyed larger fractions of the genome and/or larger collections of strains [25–27] to produce far more complete catalogs of population genomic variation in S. cerevisiae. Collectively, these studies found that sequence variation is widespread, with hundreds of thousands of single nucleotide polymorphisms (SNPs) and roughly 5–10% that number of small insertions and deletions (indels), as well as substantial variation in gene content, copy number and transposable elements [25–27]. More recently, decreasing costs of high-throughput genotyping and a deluge of sequence data made possible by the widespread adoption of short-read sequencing methods have facilitated refinement and improvement of these early surveys of genetic variation in yeast. For example, Cromie et al. [28] examined a reduced but substantial portion of the genome across several hundred strains, finding that the partitioning of sequence variation was consistent with a model of geographic differentiation followed by human-associated admixture between certain populations. Both Bergstrom et al. [29] and Strope et al. [30] showed the utility of complete sequencing followed by de novo assembly, with Bergstrom et al. [29] focusing on a comparison of S. cerevisiae to its sister species S. paradoxus and Strope et al. [30] on differences between clinical and nonclinical strains.

Although many studies of sequence variation in S. cerevisiae have focused on SNPs and small indels, several studies have noted widespread structural variation (particularly in subtelomeres) and aneuploidy in strains derived from natural isolates [29–32]. Variation in gene content is common [29]; notably, gene function is highly dependent on genetic background [33]. Finally, there are numerous examples of large genomic segments introgressed from other species, perhaps illustrated most dramatically by 120 kb of introgressed sequence containing genes involved in key wine fermentation functions in a wine strain [34].

How large is the genotypic space of S. cerevisiae? Using estimates of the mutation rate and effective population size [35–37], it would take only ∼300 generations for each base in the genome to be mutated once on average. Thus, despite the abundant sequence variation documented in this species, the combined forces of genetic drift and natural selection have dramatically winnowed the amount of variation that remains polymorphic. Nevertheless, reports of highly diverged S. cerevisiae isolated from primeval forests remote from human activity [11] raise the possibility that a significant fraction of genome sequence variation present in this species remains to be described and characterized. In general, most studies that involve population genomic variation, while undoubtedly useful, are sampling only a small portion of the species’ genotypic space (e.g. analyzing the progeny of crosses between two divergent strains [38–42]).

Population-level functional genomic variation

Transcriptional variation is pervasive but often has uncertain functional significance

Although techniques for detecting biomolecules such as RNA and proteins have been available since the 1970s, these early methods were relatively laborious and low-throughput. The emergence of microarrays and serial analysis of gene expression in 1995 [43, 44] heralded the beginning of the era of functional genomics. Our understanding of functional genomic variation at the population level parallels the development of technologies for assaying molecular phenotypes; as such, this understanding is most mature for transcript abundance. An important finding apparent from the very earliest studies population functional genomic studies is that variation in gene expression among individuals within populations (or derived from a cross) is pervasive [3, 45]. Combining gene expression measurements of segregants from a cross with quantitative trait locus (QTL) mapping, an approach pioneered by Brem et al. [38], has been a particularly useful paradigm for mapping the genetic determinants of expression variation. Most gene expression variation has a complex genetic basis, with heritable variation due to multiple QTL of weak effect and widespread transgressive segregation and epistatic interactions between loci [46, 47].

In light of the abundant gene expression variation seen within yeast populations and among segregants, it is important to note that in most cases the functional significance of variability in transcript abundance remains unclear. Fraser et al. [48] argued that a minimum of ∼10% of S. cerevisiae genes with expression QTL are under lineage-specific selection, suggesting that a significant fraction of observed transcriptional variability has biological relevance. In one example of a change in gene expression with direct functional and possibly adaptive significance, Fay et al. [49] identified differential expression of the aquaporin gene AQY2 as responsible for strain differences in freeze tolerance. However, even in this relatively focused example, the authors found hundreds of differentially expressed genes with no obvious functional significance [49]. Knowledge of the specific polymorphisms underlying differences in gene expression is not necessarily informative of the functional significance of such transcriptional variation. For example, Brown et al. [50] found that a single-base insertion had highly pleiotropic effects on gene expression, accounting for 45% of gene expression divergence between progeny of a heterozygous natural S. cerevisiae isolate, but that this extensive expression divergence had undetectable effects on growth rate in heterozygous form. Similarly, a quantitative trait nucleotide (QTN) for colony coloration induces changes in expression of dozens of genes, but most are not required for production of the coloration phenotype [51]. More generally, functional significance may be difficult to detect if the change in gene expression manifests or is subject to genetic control only in particular environments [52–54]. In a detailed characterization of the LCB2 gene, Rest et al. [55] determined that the gene expression–fitness curve is nonlinear as well as background- and environment-dependent, suggesting that the fitness consequences of expression variation are likely to be highly context-dependent.

The evolution of transcriptional variation

Directly modeling the evolution of gene expression levels

One approach to better understanding the functional significance of transcriptional variation is to characterize the evolutionary forces governing transcript abundances. A number of studies have treated gene expression levels as quantitative characters and attempted to make general statements about the average evolutionary forces affecting large numbers of genes. Hodgins-Davis et al. [56] employed detailed estimates of the genetic and mutational variance, mutation rate and effective population size in S. cerevisiae, worms and flies to demonstrate that overall patterns of expression variation are well fit by a so-called House-of-Cards model of stabilizing selection where mutations can have large fitness effects which are largely independent of the premutation state. For a single gene, it can be difficult to reject the possibility of neutral evolution using a model of this type. Lande [57, 58] and Lynch and Hill [59] developed models of phenotypic evolution incorporating mutation and drift which can be used to test for stabilizing or directional selection, although Turelli et al. [60] noted that these models are best viewed as qualitative assessments of agreement with neutrality. Rohlfs et al. [61] presented a statistical framework for gene expression evolution that can model both random drift and natural selection. However, distinguishing drift versus stabilizing selection or identifying lineage-specific directional selection is relatively underpowered even using samples of gene expression taken from large phylogenies with 10 species and tens of individuals per species [61]. A major caveat to the above approaches is that they reveal the average mode of selection acting on each gene, so positive selection can be swamped out by a larger amount of negative selection [62].

Another class of studies has utilized approximately random mutations to calibrate expectations of gene expression evolution under neutrality. Nearly random mutations can be obtained using mutation accumulation (MA) lines, where populations are subjected to repeated strong bottlenecks designed to minimize selection and permit the random fixation of all but lethal or sterile mutations. Landry et al. [63] measured genome-wide gene expression in four yeast MA lines that had evolved under minimal selective pressure for ∼4000 generations and used gene-specific estimates of mutational variance to explore transcriptional plasticity and protein expression noise. Although this approach is clearly informative, each MA line’s genome only harbored on the order of tens of mutations [36], which is insufficient to gain a detailed view of neutral expectations for single genes. In a deep study of the effects of random mutations at a single locus, Metzger et al. [64] compared transcriptional changes of a set of naturally occurring polymorphisms in the S. cerevisiae TDH3 promoter to hundreds of synthesized point mutations in the same sequence. By measuring gene expression at the single-cell level, these authors were able to provide empirical evidence for stronger selection on expression noise than on mean expression due to differences in the distributions of mutational effects [64].

Testing for directional selection using QTL

An alternative to treating gene expression levels as phenotypes and directly modeling their evolutionary trajectories is to test for selection using QTL known to underlie variation in expression levels. These QTL are characterized by both an underlying DNA sequence and a directionality of effect on the level of expression of the gene in question. Fraser [62] proposed a modified version of Orr’s sign test [65] that uses information about the directionality of gene expression QTL effects in order to identify candidate single genes and sets of genes with evidence for adaptive evolution. An attractive feature of this type of test is that it is relatively robust to assumptions about selection, mutation and population demography [62]. However, the original test [65] has been criticized as problematic under conditions of trait ascertainment or when QTL effect sizes are nearly uniform or highly skewed [66, 67]. In an examination of two strains of S. cerevisiae, Fraser et al. [48] found an excess of pairs of reinforcing cis–trans gene expression QTL, suggesting the possibility of widespread adaptive gene expression, although opposing results were reported in a hybrid of S. cerevisiae and S. paradoxus [68], perhaps due to the difference in evolutionary timescales. Making inferences about the evolutionary forces acting on single gene expression levels would require a large number of QTL (which are usually not available), but an alternative is to consider pathways or sets of coregulated genes [62]. Bullard et al. [69] and Fraser et al. [70] identified examples of putatively adaptive polygenic cis-regulatory evolution among sets of coregulated genes or physically interacting proteins by examining the directionality of cis-acting gene expression QTL. Such sign-based tests require appropriate polarization of directionality to identify adaptive evolution in pathways with both inducing and repressing components. Furthermore, these tests require additional evidence to distinguish between lineage-specific positive selection and relaxed negative selection, such as population genetic evidence for selective sweeps apparent from the DNA sequences underlying expression QTL [48]. Rice and Townsend [71] presented a test for selection not subject to this caveat that combines the empirically obtained QTL effect size distribution with data from mutation accumulation lines. Although this test requires both QTL and mutation accumulation data, it provides an elegant solution for obtaining a quantitative estimate of the strength of selection acting on a phenotype of interest [71].

A different approach is to examine the DNA sequences of QTL known to underlie variation in gene expression and to leverage the well-developed body of theory on DNA sequence evolution to test for patterns indicative of the action of selection. The primary challenge of this approach is that we have not yet identified large compendia of fully resolved QTL (i.e. QTN) known to affect gene expression. Ronald and Akey [72] examined sequences known to harbor regulatory polymorphisms (promoter and $3^{\prime }$ UTR) in a large set of genes with cis-regulatory expression QTL and used the allele frequency distribution to quantitatively estimate the global strength of purifying selection governing cis-regulatory gene expression evolution. This approach should become more informative as we obtain increasingly large catalogs of cis- and trans-regulatory QTN.

Exploiting experimental evolution to understand functional genomic variation

As discussed above, drawing inferences about the adaptive significance of differences in functional genomic traits is difficult. However, by combining the power of experimental evolution and functional genomics it is possible to gain detailed insights into how functional genomic traits evolve in experimental lineages experiencing specific selective environments and with defined demographic regimes. A typical yeast experimental evolution study will start with clonal populations, and propagate these populations for several hundred generations, either in chemostats or by serial transfer, under the selective regime of interest. Traits of interest are then compared between the initial population and the evolved lines, and genomic sequencing of isolates or whole populations may be used to identify mutations that have arisen and increased in frequency over the course of the experiment.

Among the first investigations to combine experimental evolution and functional genomics in yeast was a study by Ferea et al. [73], in which replicate populations were evolved for >250 generations in glucose-limited chemostats. Whole genome expression was measured using spotted DNA microarrays, and the authors identified ∼180 genes that exhibited similar divergence from the founding population in at least two of the three replicate lines. Many of the genes that showed changes in gene expression could be rationalized in terms of their role in glycolysis, oxidative phosphorylation and the TCA cycle.

Subsequent studies have largely followed the same basic approach laid out in [73], incorporating additional factors of interest. For example, Gresham et al. [74] evolved two different strain backgrounds in each of three different environments, and included both haploid and diploid populations. Comparison of lines within environments suggested that glucose and phosphate limitation were more ‘multi-peaked' relative to the sulfur-limited environment, as the diversity of transcriptional changes was greater between lines in the first two environments. Large-scale structural variants were predominant across populations, suggesting a central role for gene amplification during adaptation in such environments. The study of Dhar et al. [75], in which populations were evolved under hyperosmotic salt stress for several hundred generations, is notable because the authors distinguished between changes in basal expression (gene expression differences between founding and evolved populations under all growth conditions) and changes in regulation (expression changes specifically associated with the selective environment).

A shared feature of most yeast experimental evolution studies to date is that the fitness landscape is relatively static over the time scales that have been examined. This may be an important contributing factor to the predominance of large-scale variation and aneuploidy that has been observed in such experiments, and may lead to biases in the number, type and magnitude of adaptive changes in functional genomic traits. For example, a study by Kvitek and Sherlock [76], using whole genome population sequencing, showed that lineages evolving in a constant glucose-limited environment tended to accumulate loss-of-function mutations in environmentally responsive signaling pathways. These mutations, while beneficial in the constant environment of the experiments, were maladaptive in nonconstant environments [76]. An area ripe for future investigations is to explore the evolution of functional genomic traits in environments that better reflect the complexity of the natural niches that yeast inhabit.

Moving downstream: protein and metabolite variability

Although technologies for measuring gene expression are most mature, the past decade has seen considerable advancements in other technologies, such as mass spectrometry, NMR spectroscopy and fluorescence-activated cell sorting, that allow investigators to quantify protein and metabolite levels. A large fraction of both proteins and metabolites varies in abundance among segregants in a cross [77–81] and strains derived from natural isolates [82, 83]. Quantitative variation of protein and metabolite abundance is common, indicating a complex underlying genetic basis [77, 80]. Several studies have attempted to identify regions of the genome that act in cis or in trans to modulate protein or metabolite abundance. Cost or sample collection considerations have generally limited the sample sizes of these studies, resulting in the detection of modest numbers of QTL [77–83]. Albert et al. [84] and Parts et al. [81] both took a complementary approach by using fluorescent tags and flow cytometry to measure single-cell protein abundance in very large populations of genetically variable cells of S. cerevisiae. These studies had highly concordant results, demonstrating detection of cis-acting protein QTL for about half of the dozens of assayed genes, and finding a median five trans-acting protein QTL per gene [81, 84], although their general approach would be more difficult to implement in other yeasts without preexisting GFP-tagged strain collections.

Variation in ‘higher-level' functional genomics phenotypes

New technologies are allowing investigators to focus not only on measuring the abundance of biomolecules of interest but also on ‘higher-level' functional genomics phenotypes involving interactions between biomolecules or properties such as their structure or dynamic behavior. Lee et al. [85] used a sequencing-based assay to map chromatin accessibility in 96 segregants from a cross, and identified both cis- and trans-regulatory QTL contributing to variation in accessibility, in many cases associated with polymorphic transcription factor binding sites. Nagarajan et al. [86] and Filleton et al. [87] studied histone marks (five different types of histone methylations and acetylations) in two to three genetically diverse strains of S. cerevisiae and found abundant variation, particularly near genes with previously defined high levels of transcriptional variability. Albert et al. [88] used ribosomal profiling to measure translational efficiency, finding that most instances of genetic effects on translation subtly moderate differences in mRNA abundance. In addition to documenting variation in functional genomics phenotypes and clarifying the genetic architecture underlying this variation, population-level functional genomic data can be used to propose novel biological functions. For example, Zheng et al. [42] used global binding profiles of the Ste12 transcription factor in segregants from a cross to propose two new trans-factors as likely involved in modulating Ste12p binding in the presence of mating pheromone.

However, despite these examples, we have only begun to scratch the surface in terms of exploring higher-level functional genomics phenotypes at the population level. Studies of nucleosome occupancy [89] and mRNA decay rates [90] in the closely related Saccharomyces species S. cerevisiae, S. paradoxus and S. bayanus suggest the potential for widespread intraspecies variation and point to the importance of these factors in the evolution of gene expression. Protein phosphorylation is a particularly important posttranslational modification [91] but as far as we are aware this phenomenon has been examined only among distantly related yeast species [92]. Many other phenotypes that have been studied only in single (or few closely related) genetic backgrounds are ripe for future analysis, such as RNA secondary structure [93], RNA subcellular localization [94], genetic interaction profiles [95], protein–protein interactions [96] and the three-dimensional architecture of the genome [97, 98]. Challenges to scaling these technologies to populations will include improving the cost, resolution, precision and reliability of complex experimental assays in order to reproducibly document often subtle interindividual differences.

From DNA to organismal phenotypes

Due largely to the manifold advantages of quantitative trait mapping in S. cerevisiae—small genomes, a powerful set of tools for genetic manipulation and well-developed methods for mapping QTL—there are a number of quantitatively varying organismal phenotypes for which we have mapped multiple QTL and, in some cases, fully resolved these QTL to QTN [e.g. 39, 41, 99–105]. However, there are no cases in which we have a complete understanding of how variation in DNA sequence mediates variation in the organismal phenotype. One approach for gaining insight into the genetic mechanisms underlying organismal phenotypes is to take an explicit network perspective in mapping studies. For example, in a study of the genetic architecture of biofilm formation in a clinical isolate, Granek et al. [106] found functional variation clustered in the cAMP-protein kinase A pathway. Using cAMP and targeted gene expression measurements, studies of gene knockout strains and prior knowledge about network structure, the authors proposed a mechanistic model of how genetic variation acting within the network mediates diversity in biofilm phenotypes. In another example, Lorenz and Cohen [103] found that QTL underlying variation in sporulation efficiency were clustered in genes near a bottleneck in the signal transduction pathway known to underpin this phenotype. Treusch et al. [105] employed a diverse set of strains to map QTL underlying the MAPK-mediated traits salt and caffeine tolerance, finding many loci with genes involved in the initial (environmental sensing), regulatory and downstream portions of the MAPK pathway. Although these approaches do not offer direct evidence for the complete mechanistic details underpinning the translation of sequence variation into organismal phenotypic variation, they do provide plausible mechanisms for follow-up, including clear candidate proteins and metabolites.

Several studies have attempted to take a more direct approach to understand complex traits by explicitly mapping the genetic basis of variation in molecular intermediates with the goal of understanding the flow of biological information from DNA to organismal phenotype [107]. This approach, also known as systems genetics, assumes that molecular intermediates will be more useful for a predictive understanding of phenotypic variation than QTL alone, as they are farther downstream along the path from genotype to phenotype. For example, Lewis et al. [108] identified gene expression QTL underlying response to ethanol stress, which pointed the way toward a better understanding of a link between Mkt1p and P-bodies upon stress. Gagneur et al. [109] used gene expression QTL along with extensive growth profiling across a range of environments to pinpoint causal intermediate genes with heritable expression shifts that were persistent across environments but had effects on growth rate that were specific to particular environments. While not directly examining organismal phenotypes, Zhu et al. [78] integrated six types of data on abundance and interactions between biomolecules in a network reconstruction approach to gain insight into mechanisms of gene expression QTL ‘hot spots' driving differences in expression at many genes in segregants from a cross. Finally, Gupta et al. [110] took a different approach and used functional genomic techniques to study the effects of a sporulation efficiency QTN in the gene MKT1 as the sole genetic difference between strains. By combining phenotyping to determine the temporal effect of the QTN with gene expression measurements and deletions of candidate genes, Gupta et al. [110] identified novel pathways mediating the causal effect of the QTN. Despite these successes, there are many complicating biological factors that may hinder our ability to use functional genomics to increase our understanding of the genotype–phenotype map. For example, molecular intermediates downstream of QTL that explain a large portion of phenotypic variation may not be fully predictive of organismal phenotype for many reasons such as post-transcriptional or post-translational modifications, network buffering, genetic background effects and environmental variation [111, 112].

Insights from other yeast species

Despite the amazing ecological and genetic diversity of yeasts [e.g. 113], the vast majority of population-level functional genomic studies have been conducted in S. cerevisiae (but see [114]). In addition to having the largest community of researchers, this organism has historically served as a proving ground for new genomic technologies, due at least in part to its status as the first yeast with a fully sequenced genome [1]. Due largely to our foundational knowledge of S. cerevisiae combined with new taxonomic, ecological and genomic studies of its close relatives, it has been proposed that Saccharomyces is becoming a model genus for evolutionary genomics [115]. Nevertheless, it will be important to study other genetically, phenotypically and ecologically diverse yeasts outside of the genus Saccharomyces in order to gain complementary perspectives on the mapping of genotype to phenotype and also to study important cellular processes such as RNAi, which is absent in the Saccharomyces sensu stricto yeasts [116]. A better understanding of the functional genomics of other yeasts will be accelerated by a critical mass of work ensuring high-quality genome sequences, reliable genetic tools and thorough annotations. For example, in an effort to facilitate functional genomic analysis in S. bayanus, Caudy et al. [117] systematically developed a suite of genetic tools and used gene expression measurements taken under a variety of carefully chosen conditions to determine which S. cerevisiae annotations could be reliably transferred and which genes had diverged in function.

Should we expect that patterns of functional genomic variation seen in S. cerevisiae will be similar in other fungi? One starting point is to consider genetic variation, the ultimate causal driver of heritable variation in functional genomic phenotypes. Overriding similarities such as large population sizes and short generation times in all yeasts will contribute to the broad contours of the landscape of genomic variation, but specific details of each species’ life cycle, ecology and association with humans can also strongly influence overall patterns of variation. Like in S. cerevisiae, there will be no single genome for any one yeast species, but rather a ‘pangenome' that encompasses a diverse assemblage of genes and ensures that results based on any single strain are unlikely to be authoritative [31]. Details of the life cycle, including frequencies of sexual reproduction, inbreeding and outcrossing, have been estimated in S. cerevisiae and its close relative S. paradoxus [35, 118, 119], but there are few quantitative studies of these factors in other yeasts. It is likely that association with human activity is correlated with complex population structure that is only partially driven by geography, a pattern borne out in S. cerevisiae, S. uvarum and Schizosaccharomyces pombe and absent from the non-human-associated S. paradoxus [20, 26, 83, 120]. Finally, basic mutational mechanisms in concert with contingent events can strongly influence patterns of variation. For example, a large (1 Mb) introgression of GC-rich sequence that is fixed in Lachancea kluyveri harbors polymorphisms characteristic of a locally increased rate of GC-biased gene conversion due to the GC-rich nature of this sequence [121].

Conclusions

We are entering an exciting time for population genomic studies in S. cerevisiae. There are >100 genetically, phenotypically and geographically diverse yeast strains with fully sequenced genomes available in fungal stock centers [26, 29, 30], and additional sequenced collections of strains are being generated by multiple laboratories. These strains represent a rich resource for studies of functional genomic variation.

As is apparent from the discussion above, the vast majority of information about population variation in functional genomic traits comes from studies of gene expression. The finding that most population-level variability in functional genomics phenotypes is subtle (e.g. Townsend et al. [3] found most changes in gene expression to be less than 2-fold; Albert et al. [88] found over 90% of differences in the number of RNA fragments bound to ribosomes to be less than 2-fold) underscores the importance of experimental design and power considerations, especially when comparing results from different studies [122]. Given the fact that large changes in molecular phenotypes will likely often be deleterious (much like missense mutations in DNA) and may be strongly selected against in natural populations, we find it likely that these issues will be important for all functional genomics phenotypes studied at the population level.

While rapid reductions in the cost of sequencing have helped to make genome and transcriptome sequencing relatively inexpensive, the relatively high per-sample costs of many other functional genomics assays, such as proteomic and metabolomic methods, continue to represent an impediment for population studies. Many of the challenges present in studies of gene expression hold for other functional genomics phenotypes as well. In general, we have limited understanding of the proportion of variation in molecular phenotypes with functional significance. For all functional genomics phenotypes, it will be important to have well-understood null models to ensure that neutrally arising variation in abundance, structure or interaction partners is not mistaken as having functional relevance [123].

Finally, an important type of information that is currently missing from most population functional genomic studies is data gathered along environmental and/or temporal axes. The vast majority of current studies have measured population variation in functional genomics phenotypes in either a single environment or using a control-treatment condition setup and at a single point in time (e.g. steady-state mRNA or protein levels or chromatin accessibility at a single point in time). Since environmental dependence and time-varying behavior is a key component of most biological processes, genetic variation that affects environmental and temporal dynamics is likely to be just as important for genome function as variation that affects abundance in a particular environment at a single point in time.

Key Points

• There is extensive genomic variation among strains of Saccharomyces cerevisiae.

• Functional genomic traits are highly variable between S. cerevisiae strains or in genetic crosses.

• There is a lack of robust null models for the evolution of most types of functional genomic traits; therefore, caution is warranted regarding inferences of adaption and selection.

• Further population-focused studies of functional genomics phenotypes besides gene expression are needed, particularly those that focus on temporal dynamics and environmental variation.

• A population perspective on functional genomic variation can inform both genome function and the evolutionary process.

Funding

This work was supported in part by grants from the National Institutes of Health [grants F32GM110997, R01GM098287] and the National Science Foundation [grant DEB-10-19753].

Biographies

Daniel A. Skelly is a postdoctoral scholar in Paul Magwene’s lab. His research focuses on mapping and understanding the genetic basis of phenotypic variation in yeast.

Paul M. Magwene is an Associate Professor of Biology at Duke University. His research interests include evolutionary genetics, systems biology and computational biology.