New Results

Evolution of phenotypic variance provides insights into the genetic basis of adaption

Wei-Yun Lai, Viola Nolte, Ana Marija Jakšić, Christian Schlötterer
doi: https://doi.org/10.1101/2021.01.19.427260

Abstract

Most traits are polygenic and the contributing loci can be identified by GWAS. Their adaptive architecture is, however, poorly characterized. Here, we propose a new approach to study the adaptive architecture, which does not depend on genomic data. Relying on experimental evolution we measure the phenotypic variance in replicated populations during adaptation to a new environment. Extensive computer simulations show that the evolution of phenotypic variance in a replicated experimental evolution setting is a powerful approach to distinguish between oligogenic and polygenic adaptive architectures. We apply this new method to gene expression variance in male Drosophila simulans before and after 100 generations of adaptation to a novel hot environment. The variance change in gene expression was indistinguishable for genes with and without a significant change in mean expression after 100 generations of evolution. We conclude that adaptive gene expression evolution is best explained by a highly polygenic adaptive architecture. We propose that the evolution of phenotypic variance provides a powerful approach to characterize the adaptive architecture, in particular when combined with genomic data.

Introduction

It is widely accepted that most complex traits have a polygenic or even infinitesimal basis (Ayroles et al., 2009; Boyle et al., 2017; Liu et al., 2019). Nevertheless, it is difficult to predict which of these loci are responding to selection when a population is exposed to a new selection regime. If pleiotropic constraints are strong, only a small subset of the genes constituting are free to respond to respond to selection. Hence, the genetic basis of the adaptive response of a complex trait (i.e. adaptive architecture (Barghi et al., 2020)) may differ substantially from the genetic architecture. Since even for large phenotypic changes the genetic basis of an adaptive response is difficult to study when more than a handful of genes are contributing, we introduce a new approach to study the complexity of the adaptive architecture. Rather than aiming to map the contributing loci, we propose to study the evolution of phenotypic variance in an experimental evolution framework.

The phenotypic variance of a quantitative trait is a key determinant for its response to selection. It can be decomposed into genetic and environmental components (Falconer and Mackay, 1963). Over the past years, mathematical models have been developed which describe the expected genetic variance of a quantitative trait under selection and its maintenance in evolving populations (Bulmer, 1972; Chevalet, 1994; Kimura and Crow, 1964; Turelli, 1984). For infinitely large populations and traits controlled by many independent loci with infinitesimal small effect, changes in trait optimum are not expected to affect the phenotypic variance (Lande, 1976). A much more complex picture is expected when the effect sizes are not equal, the population size is finite, or the traits have a simpler genetic basis (Barton and Keightley, 2002; Barton and Turelli, 1987; Franssen et al., 2017; Hayward and Sella, 2019; Jain and Stephan, 2015; Keightley and Hill, 1989). For instance, for a trait with oligogenic architectures, the genetic variance could drop dramatically during adaptation, while with polygenic architectures, only minor effects on the variance are expected (Barton et al., 2017; Franssen et al., 2017; Jain and Stephan, 2015). These studies suggest that a time-resolved analysis of phenotypic variance has the potential to shed light onto the complexity of the underlying adaptive architecture.

Despite its potential importance for the understanding of adaption, we are faced with the situation that very few empirical data are available for the evolution of phenotypic variance. The use of natural populations to study changes in phenotypes and even more so phenotypic variances is limited, as the environmental heterogeneity cannot be controlled. A complementary approach to study the evolution of phenotypic variance is experimental evolution (Kawecki et al., 2012). With replicated populations starting from the same founders and evolving under tightly controlled environmental conditions, experimental evolution provides an enormous potential to study the evolution of phenotypic variance.

Most experimental evolution studies in sexual populations focused on the evolution of phenotypic means, rather than variance (Burke et al., 2010; Chippindale et al., 1996; Jakšić et al., 2020; Mallard et al., 2018). A notable exception is a study which applied fluctuating, stabilizing and disruptive selection to a small number of traits (wing shape) and observed a change of the phenotypic variance (Pélabon et al., 2010). Instead of looking at a preselected subset of phenotypes which limits the generality, we will focus on gene expression, a set of molecular phenotypes, which can be easily quantified since microarrays and, more recently, RNA-Seq have become available. Importantly, the expression levels of genes exhibit the same properties (e.g.: continuality and normality) as other complex quantitative traits (Mackay et al., 2009). Thus, gene expression has also been widely employed to search for putative adaptive traits of locally adapted populations (Romero et al., 2012; Signor and Nuzhdin, 2018; Sork, 2017) or ancestral and evolved populations in the context of experimental evolution (Ferea et al., 1999; Huang and Agrawal, 2016; Lenski et al., 1994; Mallard et al., 2018).

In this study, we used forward simulations that match not only essential design features of typical experimental evolution studies, but also incorporate realistic parameters of the genetic architecture. We recapitulate the classic expectations that even a moderately polygenic architecture is associated with a high stability of the phenotypic variance of a selected trait across different phases of adaptation. Applying this insight to a real dataset (Barghi et al., 2019; Hsu et al., 2020; Jakšić et al., 2020), we show that a considerable set of genes changed their mean expression, but their expression variance was indistinguishable from genes without changes in mean expression. We propose that this pattern reflects that adaptive gene expression evolution generally has a polygenic basis.

Results

The central idea of this study is that the complexity of an adaptive trait can be estimated from the trajectory of the phenotypic variance during adaptation: the phenotypic variance remains relative stable for a trait with polygenic (infinitesimal) architecture while it changes across generations for a trait with oligogenic architecture. Although this prediction has been illustrated in multiple theoretical and simulation studies (e.g.: Barton et al., 2017; Franssen et al., 2017; Jain and Stephan, 2015), as the first step of this study, we explored to what extent it can be generalized to a typical E&R setting considering a broader parameter space. Assuming additivity and a negative correlation between ancestral allele frequency and effect size (Otte et al., 2020) (Figure 1b), we simulated traits adapting to a mild/distant shift in trait optimum with weak/intermediate/strong stabilizing selection (Figure 1a and Figure 1 – Figure supplement 1). With three different distributions of effect size (Figure 1c) we investigate how the number of contributing loci affects the phenotypic variance.

Figure 1.
Figure 1. Simulating polygenic adaptation to a shift in trait optimum with different parameter combinations.

a. For the computer simulations we consider a quantitative trait (in black) experiencing a sudden shift in trait optimum under stabilizing selection. The underlying fitness functions are illustrated in red. The new trait optimum is shifted from the ancestral trait mean by one/three standard deviation of the ancestral trait distribution. The strength of stabilizing selection is modified by changing the variance of the fitness function: 1.8, 3.6 and 5.4 standard deviations of the ancestral trait distribution. b. The negative correlation between the allele frequencies and the effect sizes (r = −0.7, estimated in Barghi et al., 2019). We consider such negative correlation when assigning the effect sizes to variants underlying a simulated trait. c. The distribution of effect sizes of the contributing loci is determined by the shape parameter (α) of gamma sampling process (α = 0.5, 2.5 and 100).

We monitored the change in phenotypic variance over 200 generations, which was sufficient to reach the trait optimum for most parameter combinations (Figure 2 – Figure supplement 4 and 5). We compared the change in variance relative to the start of the experiment in populations with and without selection. First, we studied a mild (one standard deviation of the ancestral phenotypic distribution) shift in trait optimum (Figure 2 and Figure 2 – Figure supplement 2). As expected for a founder population derived from a substantially larger natural population, we find that even under neutrality the phenotypic variance does not remain constant, but gradually decreases during 200 generations of experimental evolution (Figure 2). This loss of variance is best explained by the fixation of variants segregating in the founder population and we did not simulate new mutations, as they do not contribute to adaptation in such short time scales (Burke et al., 2010). Our simulations show that even experimental evolution studies with moderate population sizes and linkage very nicely recapitulate the patterns from previous studies (Barton et al., 2017; Franssen et al., 2017; Jain and Stephan, 2015). A pronounced drop in phenotypic variance is observed while a trait is approaching a shifted optimum with few contributing loci (Figure 2). When more loci are contributing to the selected phenotype, the difference to neutrality becomes very small (Figure 2 and Figure 2 – Figure supplement 2). In addition to the number of contributing loci, also the heterogeneity in effect size among loci and the shape of the fitness function have a major impact. The larger the difference in effect size is, the more pronounced was the influence of the number of contributing loci (Figure 2). The opposite effect was seen for the width of the fitness function – a larger variance decreased the influence of the number of contributing loci (Figure 2). Importantly, these patterns were not affected by the duration of the experiment-qualitatively identical patterns were seen at different time points until generation 200.

Figure 2.
Figure 2. The trajectory of changes in phenotypic variance during adaptation to a mild optimum shift.

The changes in phenotypic variance within 200 generations adapting to a moderate optimum shift (orange) are compared to the changes under neutrality (grey) on y axis. The change in variance (F) is calculated as the ratio of phenotypic variance between each evolved time point (generation x) and the ancestral state Embedded Image. The simulations cover traits controlled by varying numbers of loci underlying the adaptation with three different distributions of effect sizes (columns) under different strength of stabilizing selection (rows). For each scenario, 1000 runs of simulations have been performed. Only traits with the most (dotted lines, 1000 loci), intermediate (dash lines, 50 loci) and the least (solid lines, 5 loci) polygenic architectures are shown. In all scenarios, the variance of the trait decreases drastically when the adaptation is controlled by a small number of loci (orange solid lines; 5 loci). While, for traits with extremely polygenic basis, the phenotypic variance stays stable over time (orange dotted lines).

For a more distant trait optimum (three standard deviations of the ancestral phenotypic distribution), we noticed some interesting dynamics that were not apparent for a closer trait optimum (Figure 2 – Figure supplement 1 and 3). The most striking one was the temporal heterogeneity of the phenotypic variance for few contributing loci of unequal effects. During the early stage of adaptation, the variance increased and dropped later below the variance in the founder population. With an increasing number of contributing loci, this pattern disappeared and closely matched the neutral case (Figure 2 – Figure supplement 1). Modifying dominance did not change the overall patterns-with a large number of contributing loci the variance fitted the neutral pattern best (Figure 2 – Figure supplement 6). Overall, our simulations indicate that only for a small set of parameters, the variance will increase during the early stage of adaptation – in particular scenarios based on a few contributing loci with very different effect sizes. The large influence of key parameters of the adaptive architecture, in particular the number of contributing loci and their effect sizes on the temporal phenotypic variance dynamics, suggest that it should be possible to exploit this for a test of polygenic adaptation, which is independent from genomic data.

As a proof of principle, we studied the evolution of gene expression variance in replicated populations evolving in a new hot temperature regime (Barghi et al., 2019; Hsu et al., 2020; Jakšić et al., 2020). The evolved populations were derived from the same ancestral population, but evolved independently for 100 generations in a novel temperature regime with daily temperature fluctuations between 18 and 28°C (Figure 3a). Rather than relying on pooled samples that allow only to estimate means, we quantified gene expression of individuals from reconstituted ancestral populations and two evolved populations in a common garden setup. Similar to previous studies (Ayroles et al., 2009; Huang et al., 2015), we estimated a heritability of around 60% across the transcriptome among individual flies from different families, which demonstrates the robustness of our experimental setup to estimate expression variances (Method – Figure supplement 1; See Materials and methods). Principle Component Analysis (PCA) indicated that 11.9% of the variation in gene expression can be explained by the first PC which separates evolved and ancestral populations, reflecting clear adaptive gene expression changes in response to the novel, hot temperature regime (Figure 3b). The means and variances of the expression of each gene were estimated and compared between the reconstituted ancestral populations and the two evolved populations (Method – Figure supplement 2 and 3; See Materials and methods). Due to the usage of different lot numbers for the RNA-Seq library preparation (Supplementary file 1), we only contrasted ancestral and evolved samples generated with the same lot number (See Materials and methods) to avoid any unnecessary confounding results.

Figure 3.
Figure 3. Schematic overview of the experimental procedures (a) and the divergence in gene expression during experimental evolution (b).

a. Experimental evolution: starting from one common founder population, two replicate populations evolved for 100 generations in a hot laboratory environment fluctuating between 18 and 28°C. Common garden Experiment: after 100 generations, the two evolved replicate populations were maintained together with the reconstituted ancestral population for two generations in a hot laboratory environment fluctuating between 18 and 28°C. After this common garden procedure, about 30 males from each population were subjected to RNA-Seq. b. Principle Component Analysis (PCA) of the transcriptomic profiles of individuals from the ancestral population (blue) and the hot-evolved population (red). Circles indicate individuals of the first replicate (Anc. No. 27 and Evo. No. 4). Triangles represent individuals of the second replicate (Anc. No. 28 and Evo. No. 9). The two replicates were made with two different batches of library cards for RNA-Seq library preparation.

The comparison of ancestral and evolved populations identified 2,812 genes in the first replicate and 2,704 genes in the second replicate which significantly changed mean expression in the evolved flies (FDR<0.05, Supplementary file 2). With about 20% of the genes changing mean expression, it is apparent that both populations evolved during 100 generations of exposure to a novel environment. 93.8% of the genes with a significant mean expression change in both populations changed in the same direction, more than expected by chance (Figure 4a, χ2 = 896.34, p-value < 2.2 x 10−16). This concordance suggests that most of the altered expression means are mainly driven by selection, rather than by drift. We quantified the expression change by relating the change in gene expression to the standard deviation in the ancestral population. The differentially expressed genes in both replicates showed a broad distribution of expression change, but the mean expression changed by one standard deviation (Figure 4b). Assuming that all expression phenotypes reached trait optimum, this reflects on average a moderate shift in trait optimum.

Figure 4.
Figure 4. Evolution of phenotypic mean and variance over 100 generations of adaptation in empirical data.

a. The evolution of gene expression mean during adaptation in the two replicates. For the genes with significant changes (DE, in orange), the changes are correlated between replicates (Spearman’s rho = 0.53). For the genes without significant changes (non-DE, in grey), the correlation between replicates is much lower (Spearman’s rho = 0.2). b. The evolution of gene expression mean scaled by the ancestral variation. For the DE genes (in orange), the median change falls around 1 standard deviation of the ancestral expression value, suggesting mild shift in trait optimum in the novel environment. For the non-DE genes (in grey), the changes in expression are mostly negligible. The same pattern is seen in the second replicate. c. The change in expression variance during adaptation for DE and non-DE genes. In both replicates, the distribution of variance changes is indistinguishable between DE genes (orange) and non-DE genes (grey) (Wilcoxon’s rank sum test, p > 0.1 for both replicates).

Consistent with computer simulations, the analysis of gene expression variance showed a slight decrease in evolved populations relative to the ancestral ones (median F-value = 0.84 in both replicates). Only a small subset of gene (n=125 and 97 in each replicate) experienced a substantial reduction in variance. Because this reduction in variance is probably driven by a different evolutionary force, we discuss them elsewhere (Lai et al., 2021). For the remaining genes, we related the changes in gene expression variance of selected genes to non-selected ones, we tested whether the variance changes in expression differ between the genes with significant mean expression changes and those without. Hence, we assume that genes with significant mean expression changes are under selection and the rest of the transcriptome has no large effect on the fitness (neutral). Remarkably, the changes in variance of putative adaptive genes with significant mean expression changes are indistinguishable from the genes that do not change their mean expression (Figure 4c). This suggests that the selection on mean expression is independent from the change in variance during adaptation in the focal populations. While variance estimates from two time points do not provide sufficient power to estimate the number of contributing loci in absence of more information about the adaptive architecture, our computer simulations (Figure 2), suggest in line with previous theoretical work (Barton et al., 2017; Jain and Stephan, 2015) that the observed stability in variance evolution reflects a polygenic architecture underlying the adaptive gene expression evolution.

Since we only explored two time points rather than a full time series, it may be possible that an oligogenic basis could also result in a similar phenotypic variance change as a polygenic architecture (Figure 2 – Figure supplement 1 and 3). This can be seen in an intuitive case when a single/few major effect allele(s) starts at a low frequency and becomes fixed (Yoo et al., 1980). Because an oligogenic basis results in a highly parallel genomic selection response (Figure 2 – Figure supplement 7), it is possible to distinguish polygenic and oligogenic architectures with phenotypes from two time points only, when genomic data are available. Because the genomic signature in the same experiment uncovered a highly heterogeneous selection response (Barghi et al., 2019), we can exclude the unlikely explanation of an oligogenic architecture resulting in a similar expression variance as non-selected genes. Rather, we conclude that the adaptive response in gene expression is best explained by a highly polygenic architecture.

Discussion

Population genetics has a long tradition to characterize adaptation based on the genomic signature of selected loci (Nielsen, 2005). Nevertheless, for selected phenotypes with a polygenic architecture, the contribution of individual loci to phenotypic change may be too subtle to be detected with classic population genetic methods (Pritchard et al., 2010). Even with an oligogenic basis, the identification of the selection targets with classic population genetic tests can be challenging.

Here, we used a conceptually different approach, which does not build on the genomic signature, to infer the adaptive architecture. Reasoning that experimental evolution is probably the best approach to obtain phenotypic time series, we performed computer simulations specifically tailored to typical experimental evolution studies with Drosophila. We showed that the temporal dynamics of the phenotypic variance is strongly affected by the number of contributing loci and other parameters of the adaptive architecture, such as the distribution of effect size and the underlying fitness function. Similar to the classic Castle-Wright estimator (Castle, 1921) that estimates the number of loci contributing to a quantitative trait from the phenotypic variance of the F2, we propose that the temporal heterogeneity of the phenotypic variance can be used to estimate the number of loci contributing to the adaptive response of a phenotype as well as other parameters of the adaptive architecture. Hence, unlike other approaches to characterize polygenic adaptation, the proposed estimator does not require genetic data when phenotypic time series data are available.

Because gene expression changes are constituting a major component of adaptation to a novel environment (Romero et al., 2012; Signor and Nuzhdin, 2018; Sork, 2017), it provides an excellent model to evaluate a variance-based test for polygenicity. Gene expression is modified by many trans-acting factors and some cis-regulatory variation. Adaptive gene expression changes can be either driven by polymorphism in cis-regulatory elements or by trans-acting variants. While interspecific differences in gene expression are predominantly caused by cis-regulation (Wittkopp et al., 2004), intraspecific variation is mostly driven by trans regulatory changes (Suvorov et al., 2013; Wittkopp et al., 2008). Adaptive gene expression changes which are well-characterized on the molecular level typically have a cis-regulatory basis that is not only frequently associated with the insertion of a transposable element (e.g.: (Daborn et al., 2002)) but also sometimes with multiple regulatory variants (Endler et al., 2018). Two lines of evidence suggest that cis-regulatory variation cannot be the driver of adaptive gene expression changes observed in this study. First, the mutational target size is too small to harbor a sufficiently large number of alleles segregating in the founder population. Second, too few recombination events occur during the experiment to uncouple regulatory variants located on a given haplotype such that they could generate a signal of polygenic adaptation. More likely, the polygenic adaptive architecture of gene expression change reflects the joint effects of many trans-acting variants.

Because we could only analyze phenotypic data from two time points, the founder population and replicate populations evolved for 100 generations, we were not able to obtain a more quantitative estimate of the number of contributing loci, in particular as other parameters of the adaptive architecture are not known and need to be co-estimated. For the distinction between an oligogenic and polygenic basis, we additionally relied on the heterogeneity of genomic selection signatures among replicates, because for some parameter combinations the oligogenic response can also result in a similar phenotypic variance as a polygenic one, but with a much higher parallel response of genomic markers (Figure 2 – Figure supplement 7).

Hence, not only more time points describing the phenotypic trajectory, but also some genomic data could contribute to infer the adaptive architecture in experimental evolution studies. The extension of this approach to natural populations faces several challenges. First, phenotypic time series over evolutionary relevant time scales are rare (but see (Clutton-Brock and Pemberton, 2004)) and second, the distinction of environmental heterogeneity from genetic changes is considerably more challenging than under controlled laboratory conditions.

Materials and Methods

Computer simulations

We performed forward simulations with MimicrEE2 (Vlachos and Kofler, 2018) using the qff mode to illustrate the influence of the genetic architecture on the evolution of phenotypic variance during the adaptation to a new trait optimum. With 189 founder haplotypes (Barghi et al., 2019), we simulated quantitative traits under the control of different numbers of loci (5, 25, 50, 100, 200 and 1000) with an effective population size of 300. For each trait, we assume an additive model and a negative correlation (r = −0.7, estimated in Barghi et al., 2019) between the effect size and starting frequency (Figure 1b). The effect sizes of each locus can disperse in different levels which depend on the shape parameter of gamma sampling process (shape = 0.5, 2.5 and 100, Figure 1c). We used correlate() function implemented in “fabricatr” R package (Blair et al., 2019) to generate the effect sizes. The sum of effect sizes of each trait was normalized to 1. We assumed a heritability h2 = 0.6 (from a family-based estimation in this study). To simulate stabilizing selection with trait optimum shift, we provided the Guassian fitness functions with mean of Embedded Image. and standard deviation of Embedded Image, where Embedded Image is the ancestral phenotypic mean and Vanc. is the ancestral genetic variance (Figure 1a). Parameter “a” determines the distance of optimum shift, which is set to one (similar to the empirical case, Figure 4b) or three (Adopted from Sella et al., 2019). Parameter “b” indicates how strong the phenotypic constrain would be when the trait optimum is reached. The value 3.6 is adopted from Sella et al., 2019. In this study, we increase and decrease it by 50% to explore its impact (1.8 or 5.4). For the neutrality case, we assumed uniform fitness for all individuals. For each trait under each scenario, the phenotypic variance was estimated at different generations and compared to the ancestral phenotypic variance at generation 1 to illustrate the dynamic of phenotypic variance during the evolution. We note that we do not assume that the ancestral population has reached an equilibrium, because the ancestral population in this study was phenotyped in the new environment.

Experimental evolution

The setup of populations and evolution experiment have been described by (Barghi et al., 2019). Briefly, ten outbred populations seeded from 202 isofemale lines were exposed to a laboratory experiment at 28/18 °C with 12hr light/12hr dark photoperiod for more than 100 generations. Each replicate consisted of 1000 to 1250 adults at each generation.

Common garden experiment

The collection of samples from the evolution experiment for RNA-Seq was preceded by two generations of common garden (CGE). The common garden experiment was performed at generation 103 of the evolution in the hot environment and this CGE has been described in (Barghi et al., 2019; Hsu et al., 2020, 2019; Jakšić et al., 2020). In brief, an ancestral population was reconstituted by pooling five mated females from 184 founder isofemale lines (Nouhaud et al., 2016). No significant allele frequency differences are expected between the reconstituted ancestral populations and the original ancestral populations initiating the experiment (Nouhaud et al., 2016). Because we evaluated phenotypes on the population level, deleterious mutations will have a very limited impact. The reason is that they occur only in a single isofemale line, which represents a very small fraction of the total population. Two replicates of the reconstituted ancestral population and two independently evolved populations at generation 103 were reared for two generations with egg-density control (400 eggs/bottle) at the same temperature regime as in the evolution experiment. Freshly eclosed flies were transferred onto new food for mating. Sexes were separated under CO2 anesthesia at day 3 after eclosure, left to recover from CO2 for two days, and at the age of five days whole-body mated flies of each sex were snap-frozen at 2pm in liquid nitrogen and stored at −80°C until RNA extraction. In this study, more than 30 individual male flies from two reconstituted ancestral populations (replicate no. 27 and no. 28) and two evolved populations (replicate no. 4 and no. 9) were subjected to RNA-Seq.

RNA extraction and library preparation

Whole bodies of individual male flies were removed from the −80°C freezer and immediately homogenized in Qiazol lysis reagent (Qiagen, Hilden, Germany). The homogenate was treated with DNase I followed by addition of chloroform, centrifugation and mixture of the upper phase with 70% ethanol as described for the Qiagen RNeasy Universal Plus Mini Kit. The mixture was subsequently loaded onto a RNeasy MinElute Spin column as provided by the RNeasy Plus Micro Kit (Qiagen, Hilden, Germany), and all washing steps were performed according to the instructions for that kit. All resulting total RNA was used to prepare stranded mRNA libraries on the Neoprep Library Prep System (Illumina, San Diego, USA) following the manufacturer’s protocol: Neoprep runs were performed using software version 1.1.0.8 and protocol version 1.1.7.6 with default settings for 15 PCR cycles and an insert size of 200bp. All libraries for individuals of ancestral replicate no. 27 and evolved replicate no. 4 were prepared with library cards of lot no. 20180170; all libraries for individuals of ancestral replicate no. 28 and evolved replicate no. 9 were prepared with library cards of lot no. 20178099. 50bp single-end reads were sequenced on an Illumina HiSeq 2500. All sequencing data will be available in European Nucleotide Archive (ENA) under the accession number PRJEB37011 upon publication.

RNA-Seq data processing and quality control

All RNA-Seq reads were trimmed using ReadTools (Gómez-Sánchez and Schlötterer, 2018) with quality score of 20 and aligned to Drosoph/la s/mulans reference genome (Palmieri et al., 2015) using GSNAP (Wu et al., 2016) with parameter setting -k 15 -N 1 -m 0.08. Exon-aligned reads were piped into Rsubread (Liao et al., 2019) to calculate read counts of each gene, and raw read counts of each gene were normalized with the TMM method implemented in edgeR (Robinson et al., 2010). Samples with severe 3’-bias were removed based on visual inspection of the gene-body coverage plot (Jakšić and Schlötterer, 2016; Wang et al., 2012).

Genetic variance in gene expression across F1 families

We evaluated how much of the expression variance can be explained by genetic variation by performing RNA-Seq on individual flies, with 3-4 individuals each from three isofemale lines maintained at the same density and culturing conditions.

Six out of 184 founder isofemale lines from the evolution experiment and were maintained for one generation with controlled egg density (400 eggs/bottle) in the same environment as the main experiment (12h 28°C with light followed by 12h 18 °C with dark conditions). Using the offspring, we generated three crosses between two of the six lines each: FL 138 x FL 137, FL 157 x FL 112, FL 123 x FL 127: we combined 50 virgin females from one of the lines with 50 males from the other line, let them lay eggs under density control as above and maintained and froze their F1 offspring in the same way as in the main CGE: sexes were separated after mating at the age of three days and snap-frozen at the age of five days at 2pm. From each cross, we used four F1 males to prepare individual RNA-Seq libraries as described above.

Assuming no environmental heterogeneity, we decomposed the total variance of the expression of each gene measured in these individuals into the genetic difference among three different F1 families and random error. The data were analyzed as follows:

Natural log-transformation was applied to CPMs of all genes to improve data normality (Rocke and Durbin, 2003). Principal component and principal variance component analysis (Bushel, 2019) was performed to the whole transcriptome to decompose the variance components. We found that around 60% of the gene expression variance can be explained by the genetic difference among the three F1 families (method – Figure supplement 1a and b). This implies that the within-population gene expression variance is largely contributed by the genetic components. Because we only used offspring from single vials, we may have overestimated the heritability if the environment in the vials differs. Nevertheless, since our heritability estimates are very similar to previous ones (Ayroles et al., 2009), we consider our estimates reliable.

In addition to general analysis across all genes, we also tested for the genetic variance of each gene separately using analysis of variance (ANOVA): Embedded Image

Where i=1, 2, 3, n (the ith genes); j =1, 2, 3, 4 (the jth individuals in each cross). yij is the observed expression level of a gene in a given sample, μ is the overall mean; τi is the effect of genetic background and εij is the random noise. We calculated the proportion of total variance explained by random error using the following equation: Embedded Image

Genes were binned based on their average expression value (lnCPM) which ranged from −0.8 to 4.1, by bin size of 0.1. The average proportion of variance explained by random error of each bin was calculated.

The expression variance of genes with less than 1 count per million bases (CPM) is dominated by technical errors (method – Figure supplement 1c). Thus, genes with less than 1 count per million base (CPM) were excluded for subsequent analysis.

RNA-Seq data analysis

We observed some outlier individuals and suspected that the freezing process may have led to detachment of body parts, such as eyes or heads, in these individuals. We compared gene expression between such outliers and all other samples and performed tissue enrichment analysis for genes with at least 2-fold lower expression in the outlier samples. Samples with evidence of tissue detachment were excluded. After filtering, each population remained approximately 20 individuals (Supplementary file 1). Only genes with at least 1 count per million base (CPM) were included in the analyses to avoid extremely lowly expressed genes.

For all RNA-Seq data we only compared samples which were prepared with library cards from the same lot number to avoid batch effects (Replicate 1: evolved replicate 4 vs. reconstituted ancestral population replicate 27; Replicate 2: evolved replicate 9 ss. reconstituted ancestral population replicate 28).

For differential expression analysis on mean expression, we used the generalized linear modeling function implemented in edgeR (Robinson et al., 2010) to fit the expression to the model (Y = E + s) in which Y stands for gene expression, E is the effect of evolution and s is the random error. The likelihood ratio test was performed to test the effect of evolution. P-value adjustment was performed using the Benjamini-Hochberg false discovery rate (FDR) correction.

For the analysis of expression variance evolution, we applied natural log transformation (Rocke and Durbin, 2003) to eliminate the strong mean-variance dependency in RNA-Seq data due to the nature of the negative binomial distribution (method – Figure supplement 2). The variance of the expression of each gene (lnCPM) was estimated in each population. With the moderate sample size, we needed to estimate the uncertainty of variance estimates. Jackknifing was applied to measure the uncertainty of estimator (Fukunaga and Hummels, 1989). The procedure was conducted independently on four replicates and we calculated the 95% confidence interval of the estimated variance (method – Figure supplement 3). The change of gene expression variance was determined by the F statistics calculated as the ratio between the variance within the ancestral population and the variance within the evolved population of each gene. To test whether selection alter the expression variance, a comparison was made between the F statistics of genes with significant changes in mean expression and the ones without.

Author contribution

W.Y.L and C.S. conceived the study. V.N. prepared all RNA-Seq and supervised the maintenance of the evolution experiment. A.M.J supervised the common garden experiment. W.Y.L performed the simulation and data analysis. W.Y.L. and C.S. wrote the manuscript.

Competing interests

The authors declare no competing interests.

Correspondence and requests for materials should be addressed to C.S.

Figure supplements

Figure 1 – Figure supplement 1.
Figure 1 – Figure supplement 1. The evolutionary scenario for distant optimum shift.

We consider the case when a quantitative trait (in black) experiences a sudden shift in trait optimum under stabilizing selection. The imposed fitness functions (F.F.) are illustrated in red. The new trait optimum is set away from the ancestral trait mean by three standard deviation of the ancestral trait distribution for distal shift. To vary the strength of stabilizing selection, the variance of the fitness function is set as 1.8, 3.6 and 5.4 standard deviation of the ancestral trait distribution.

Figure 2 – Figure supplement 1.
Figure 2 – Figure supplement 1. The trajectory of expected changes in phenotypic variance when adapting to a distant optimum shift.

The changes in phenotypic variance within 200 generations adapting to a distant optimum shift (orange) are compared to the changes under neutrality (grey) on y axis. The change in variance (F) is calculated as the ratio of phenotypic variance between each evolved time point (generation x) and the ancestral state Embedded Image. The simulations cover traits controlled by varying numbers of loci underlying the adaptation with three different distributions of effect sizes (columns) and under different strength of stabilizing selection (rows). For each scenario, 1000 runs of simulations have been performed. Only traits with the most (dotted lines, 1000 loci), intermediate (dash lines, 50 loci) and the least (solid lines, 5 loci) polygenic architectures are shown. Unlike the continuous decreasing pattern in the cases with moderate optimum shifts, the variance of traits controlled by a few loci (5 loci) with largely dispersed effects would increase first and then decrease when the effect sizes of contributing loci are dispersed (orange solid lines). Nevertheless, for traits with extremely polygenic basis, the phenotypic variance always stays stable over time (orange dotted lines).

Figure 2 – Figure supplement 2.
Figure 2 – Figure supplement 2. The expected changes in phenotypic variance and the number of contributing loci.

The changes in phenotypic variance after 200 generations adapting to a mild optimum shift (orange) are compared to the changes under neutrality (grey) on y axis. The change in variance (F) is calculated as the ratio between the evolved and ancestral phenotypic variance Embedded Image. The simulations cover traits controlled by varying numbers of loci underlying the adaptation (x-axes) with three different distributions of effect sizes (columns) and under different strength of stabilizing selection (rows). For each scenario, 1000 runs of simulations have been performed. In all scenarios, the variance of the trait decreases drastically when the adaptation is controlled by a small number of loci. As the number of contributing loci increases, the phenotypic variance becomes more stable.

Figure 2 – Figure supplement 3.
Figure 2 – Figure supplement 3. The expected changes in phenotypic variance and the number of contributing loci.

The changes in phenotypic variance after 200 generations adapting to a distant optimum shift (orange) are compared to the changes under neutrality (grey) on y axis. The change in variance (F) is calculated as the ratio between the evolved and ancestral phenotypic variance Embedded Image. The simulations cover traits controlled by varying numbers of loci underlying the adaptation (x-axes) with three different distributions of effect sizes (columns) and under different strength of stabilizing selection (rows). For each scenario, 1000 runs of simulations have been performed. In most scenarios, the variance of the trait decreases drastically when the adaptation is controlled by a small number of loci. As the number of contributing loci increases, the phenotypic variance becomes more stable. However, exceptions can be observed when the effect sizes of the contributing loci are largely dispersed. In these cases, shift in optimum does not have the strong impact on the variance of traits under simple genetic control (5 contributing loci). In the extreme case, highly dispersed effect sizes in combination with a relaxed phenotypic constraint removes the relationship between the number of loci and the changes in phenotypic variance.

Figure 2 – Figure supplement 4.
Figure 2 – Figure supplement 4. The changes in phenotypic mean when adapting to a mild optimum shift.

The changes in phenotypic mean after 200 generations adapting to a mild optimum shift (orange) are compared to the changes under neutrality (grey) on y axis. The changes in phenotypic mean are scaled by the standard deviation of the ancestral trait distribution. The simulations cover traits controlled by varying numbers of loci underlying the adaptation (x axes) with three different distributions of effect sizes (columns) and under different strength of stabilizing selection (rows). For each scenario, 1000 runs of simulations have been performed. In most cases, the traits under selection (orange) shift their means by one standard deviation of the ancestral trait distribution (i.e. reaching the new trait optimum) while the neutral traits (grey) stay unchanged.

Figure 2 – Figure supplement 5.
Figure 2 – Figure supplement 5. The changes in phenotypic mean when adapting to a distant optimum shift.

The changes in phenotypic mean after 200 generations adapting to a distant optimum shift (orange) are compared to the changes under neutrality (grey) on y axis. The changes in phenotypic mean are scaled by the standard deviation of the ancestral trait distribution. The simulations cover traits controlled by varying numbers of loci underlying the adaptation (x axes) with three different distributions of effect sizes (columns) and under different strength of stabilizing selection (rows). For each scenario, 1000 runs of simulations have been performed. In most cases, the traits under selection (orange) shift their means by three standard deviation of the ancestral trait distribution (i.e. reaching the new trait optimum) while the neutral traits (grey) stay unchanged.

Figure 2 – Figure supplement 6.
Figure 2 – Figure supplement 6. The expected changes in phenotypic variance for traits controlled by dominant and recessive alleles.

The changes in phenotypic variance after 200 generations adapting to a mild optimum shift (orange) are compared to the changes under neutrality (grey) on y axis. The change in variance (F) is calculated as the ratio between the evolved and ancestral phenotypic variance Embedded Image. This simulation covers traits controlled by varying numbers of loci underlying the adaptation (x axes) with recessive, additive and dominant effects. For each scenario, 1000 runs of simulations have been performed. No matter how the dominance varies, the variance of the trait decreases drastically when the adaptation is controlled by a small number of loci. As the number of contributing loci increases, the phenotypic variance becomes more stable.

Figure 2 – Figure supplement 7.
Figure 2 – Figure supplement 7. Parallelism in genomic response across 10 evolved replicates for traits under the control of five loci and 100 loci.

For the loci with allele frequency change of at least 10% in 200 generations, the average Pearson’s correlation coefficient of the frequency change between all pairs of two evolved replicates was calculated to describe the parallelism of the evolution at these loci. An average across loci is used to obtain a general parallelism. 100 runs of simulations with 10 evolution replicates have been performed for each scenario. With five contributing loci, the genomic evolution of the contributing loci is more parallel across the 10 evolved replicates compared to the case when with 100 contributing loci.

Method – Figure supplement 1.
Method – Figure supplement 1. Genetic variance in gene expression across F1 families.

(A) Principal component analysis (PCA) on the transcriptomes of F1 individuals from three different crosses between the founder iso-female lines. Individuals from different families clustered nicely based on the first two PCs. (B) Principal variance component analysis (PVCA) on the transcriptomes of F1 individuals. 67% of the total variance in gene expression was explained by the genetic difference between the individuals. (C) Gene-wise analysis of variance (ANOVA) in gene expression. Genes were binned based on their average expression value (lnCPM) which ranged from −0.8 to 4.1, by bin size of 0.1. The average proportion of variance explained by random error of each bin was visualized. The expression variance of genes with less than 1 count per million bases (CPM) is dominated by residuals.

Method – Figure supplement 2.
Method – Figure supplement 2. Log-transformation eliminates the positive relationship between the changes in mean and variance of gene expression.

In each panel, the changes in mean expression, log(FC) Embedded Image and in variance before (A) and after (B) the natural log-transformation of each gene were visualized (Embedded Image and Embedded Image). The positive correlation (r = 0.45) due to the positive mean-variance dependency of negative binomial distribution is removed by the log-transformation (r = −0.05) on gene expression level.

Method – Figure supplement 3.
Method – Figure supplement 3. Robustness of the variance estimation using individual sequencing data.

Jackknife method was applied to measure the uncertainty of variance estimation. Given a sample size of K, the procedure is to estimate the variance of each gene for K times, each time leaving one sample out. The procedure was conducted independently on 4 populations (anc.27, anc.28, evo.4 and evo.9). In each panel, we visualize Jackknife approximated 95% confidence interval for the variance estimates of each gene. The genes are ordered based on the average variance estimates (black dash line) on the x-axis. The upper and lower limits of the 95% confidence interval are indicated with grey curves. The salmon line denotes the observed value of the variance estimates. In most cases, the estimates lie in the confidence interval, suggesting robust estimation.

Titles and legend for supplementary files

Supplementary file 1. Library information of the sample in this study. This file provides a list of all sequenced samples and the library information.

Supplementary file 2. Differential gene expression analysis of two contrasts between ancestral and evolved populations. This file reports the results of DE analysis between anc. 27 and evo. 4 (Table S2A) and between anc. 28 and evo. 9 (Table S2B).

Supplementary file 3. F value on the gene expression of two contrasts between ancestral and evolved populations. This file reports the results of gene expression variance comparisons between anc. 27 and evo. 4 (Table S3A) and between anc. 28 and evo. 9 (Table S3B).

Acknowledgments

Special thanks to David Houle, who provided fantastic support during the collection and establishment of the isofemale lines in Florida. We thank all member of the Institut für Populationsgenetik for discussion. We are grateful to Reinhard Bürger and David Houle for helpful comments on earlier versions of the manuscript. Neda Barghi, François Mallard and Kathrin Otte contributed to the common garden experiment. Illumina sequencing was performed at the VBCF NGS Unit (www.vbcf.ac.at). This work was support by the Austrian Science Funds (FWF, W1225) and the European Research Council (ERC, ArchAdapt).

References

  1. ↵
    Ayroles JF, Carbone MA, Stone EA, Jordan KW, Lyman RF, Magwire MM, Rollmann SM, Duncan LH, Lawrence F, Anholt RRH, Mackay TFC. 2009. Systems genetics of complex traits in Drosophila melanogaster. Nat Genet 41:299–307. doi:10.1038/ng.332
    OpenUrlCrossRefPubMedWeb of Science
  2. ↵
    Barghi N, Hermisson J, Schlötterer C. 2020. Polygenic adaptation: a unifying framework to understand positive selection. Nat Rev Genet 21:769–781. doi: 10.1038/s41576-020-0250-z
    OpenUrlCrossRef
  3. ↵
    Barghi N, Tobler R, Nolte V, Jakšić AM, Mallard F, Otte KA, Dolezal M, Taus T, Kofler R, Schlötterer C. 2019. Genetic redundancy fuels polygenic adaptation in Drosophila. PLOS Biol 17:e3000128. doi:10.1371/journal.pbio.3000128
    OpenUrlCrossRef
  4. ↵
    Barton NH, Etheridge AM, Véber A. 2017. The infinitesimal model: Definition, derivation, and implications. Theor Popul Biol 118:50–73. doi:10.1016/J.TPB.2017.06.001
    OpenUrlCrossRef
  5. ↵
    Barton NH, Keightley PD. 2002. Understanding quantitative genetic variation. Nat Rev Genet 3:11–21. doi:10.1038/nrg700
    OpenUrlCrossRefPubMedWeb of Science
  6. ↵
    Barton NH, Turelli M. 1987. Adaptive landscapes, genetic distance and the evolution of quantitative characters. Genet Res 49:157–173. doi:10.1017/S0016672300026951
    OpenUrlCrossRefPubMedWeb of Science
  7. ↵
    Boyle EA, Li YI, Pritchard JK. 2017. An expanded view of complex traits: From polygenic to omnigenic. Cell 169:1177–1186. doi:10.1016/J.CELL.2017.05.038
    OpenUrlCrossRefPubMed
  8. ↵
    Bulmer MG. 1972. The genetic variability of polygenic characters under optimizing selection, mutation and drift. Genet Res 19:17–25. doi:10.1017/S0016672300014221
    OpenUrlCrossRefPubMedWeb of Science
  9. ↵
    Burke MK, Dunham JP, Shahrestani P, Thornton KR, Rose MR, Long AD. 2010. Genome-wide analysis of a long-term evolution experiment with Drosophila. Nature 467:587–590. doi:10.1038/nature09352
    OpenUrlCrossRefPubMedWeb of Science
  10. ↵
    Bushel P. 2019. pvca: Principal variance component analysis (PVCA). R Packag version 1240.
  11. ↵
    Castle WE. 1921. An improved method of estimating the number of genetic factors concerned in the cases of blending inheritance. Science 54:223. doi:10.1126/science.54.1393.223
    OpenUrlFREE Full Text
  12. ↵
    Chevalet C. 1994. An approximate theory of selection assuming a finite number of quantitative trait loci. Genet Sel Evol. 26:379. doi:10.1186/1297-9686-26-5-379
    OpenUrlCrossRef
  13. ↵
    Chippindale AK, Chu TJF, Rose MR. 1996. Complex trade-offs and the evolution of starvation resistance in Drosophila melanogaster. Evolution (N Y) 50:753–766. doi:10.1111/j.1558-5646.1996.tb03885.x
    OpenUrlCrossRef
  14. ↵
    Clutton-Brock TH, Pemberton JM (Josephine M). 2004. Soay sheep: population dynamics and selection on St. Kilda. Cambridge University Press.
  15. ↵
    Daborn PJ, Yen JL, Bogwitz MR, Le Goff G, Feil E, Jeffers S, Tijet N, Perry T, Heckel D, Batterham P, Feyereisen R, Wilson TG, ffrench-Constant RH. 2002. A single p450 allele associated with insecticide resistance in Drosophila. Science 297:2253–6. doi: 10.1126/science.1074170
    OpenUrlAbstract/FREE Full Text
  16. ↵
    Endler L, Gibert J-M, Nolte V, Schlötterer C. 2018. Pleiotropic effects of regulatory variation in tan result in correlation of two pigmentation traits in Drosophila melanogaster. Mol Ecol 27:3207–3218. doi:10.1111/mec.14781
    OpenUrlCrossRef
  17. ↵
    Falconer DS, Mackay TFC. 1963. Introduction to quantitative genetics, Poultry Science. CreateSpace Independent Publishing Platform. doi:10.3382/ps.0420547
    OpenUrlCrossRef
  18. ↵
    Ferea TL, Botstein D, Brown PO, Rosenzweig RF. 1999. Systematic changes in gene expression patterns following adaptive evolution in yeast. Proc Natl Acad Sci U S A 96:9721–6. doi:10.1073/pnas.96.17.9721
    OpenUrlAbstract/FREE Full Text
  19. ↵
    Franssen SU, Kofler R, Schlötterer C. 2017. Uncovering the genetic signature of quantitative trait evolution with replicated time series data. Heredity (Edinb) 118:42–51. doi:10.1038/hdy.2016.98
    OpenUrlCrossRef
  20. ↵
    Fukunaga K, Hummels DM. 1989. Leave-one-out procedures for nonparametric error estimates. IEEE Trans Pattern Anal Mach Intell 11:421–423. doi:10.1109/34.19039
    OpenUrlCrossRef
  21. ↵
    Gómez-Sánchez D, Schlötterer C. 2018. ReadTools: A universal toolkit for handling sequence data from different sequencing platforms. Mol Ecol Resour 18:676–680. doi:10.1111/1755-0998.12741
    OpenUrlCrossRef
  22. ↵
    Hayward LK, Sella G. 2019. Polygenic adaptation after a sudden change in environment. bioRxiv 792952. doi:10.1101/792952
    OpenUrlAbstract/FREE Full Text
  23. ↵
    Hsu S-K, Jakšić AM, Nolte V, Barghi N, Mallard F, Otte KA, Schlötterer C, Hsu S-K, Jakšić AM, Nolte V, Barghi N, Mallard F, Otte KA, Schlötterer C. 2019. A 24 h age difference causes twice as much gene expression divergence as 100 generations of adaptation to a novel environment. Genes (Basel) 10:89. doi:10.3390/genes10020089
    OpenUrlCrossRef
  24. ↵
    Hsu S-K, Jakšić AM, Nolte V, Lirakis M, Kofler R, Barghi N, Versace E, Schlötterer C. 2020. Rapid sex-specific adaptation to high temperature in Drosophila. Elife 9. doi:10.7554/eLife.53237
    OpenUrlCrossRef
  25. ↵
    Huang W, Carbone MA, Magwire MM, Peiffer JA, Lyman RF, Stone EA, Anholt RRH, Mackay TFC. 2015. Genetic basis of transcriptome diversity in Drosophila melanogaster. Proc Natl Acad Sci U S A 112:E6010–E6019. doi:10.1073/pnas.1519159112
    OpenUrlAbstract/FREE Full Text
  26. ↵
    Huang Y, Agrawal AF. 2016. Experimental evolution of gene expression and plasticity in alternative selective regimes. PLOS Genet 12:e1006336. doi:10.1371/journal.pgen.1006336
    OpenUrlCrossRef
  27. ↵
    Jain K, Stephan W. 2015. Response of polygenic traits under stabilizing selection and mutation when loci have unequal effects. G3 (Bethesda) 5:1065–74. doi:10.1534/g3.115.017970
    OpenUrlAbstract/FREE Full Text
  28. ↵
    Jakšić AM, Karner J, Nolte V, Hsu S-K, Barghi N, Mallard F, Otte KA, Svecnjak L, Senti K-A, Schlötterer C. 2020. Neuronal function and dopamine signaling evolve at high temperature in Drosophila. Mol Biol Evol 37:2630–2640. doi:10.1093/molbev/msaa116
    OpenUrlCrossRef
  29. ↵
    Jakšić AM, Schlötterer C. 2016. The interplay of temperature and genotype on patterns of alternative splicing in Drosophila melanogaster. Genetics 204:315–325. doi:10.1534/GENETICS.116.192310
    OpenUrlAbstract/FREE Full Text
  30. ↵
    Kawecki TJ, Lenski RE, Ebert D, Hollis B, Olivieri I, Whitlock MC. 2012. Experimental evolution. Trends Ecol Evol 27:547–560. doi:10.1016/J.TREE.2012.06.001
    OpenUrlCrossRefPubMedWeb of Science
  31. ↵
    Keightley PD, Hill WG. 1989. Quantitative genetic variability maintained by mutation-stabilizing selection balance: sampling variation and response to subsequent directional selection. Genet Res 54:45–58. doi:10.1017/S0016672300028366
    OpenUrlCrossRefPubMedWeb of Science
  32. ↵
    Kimura M, Crow JF. 1964. The number of alleles that can be maintained in a finite population. Genetics 49:725–738.
    OpenUrlFREE Full Text
  33. ↵
    Lai W.-Y. & Schlötterer, C. 2021. Evolution of gene expression variance during adaptation to high temperature in Drosophila. BioRxiv.
  34. ↵
    Lande R. 1976. Natural selection and random genetic drift in phenotypic evoltion. Evolution (NY) 30:314–334. doi:10.1111/j.1558-5646.1976.tb00911.x
    OpenUrlCrossRefPubMed
  35. ↵
    Lenski RE, Travisano M, Larison B, Moritz C. 1994. Dynamics of adaptation and diversification: a 10,000-generation experiment with bacterial populations. Proc Natl Acad Sci 91:6808–6814. doi:10.1073/pnas.91.15.6808
    OpenUrlAbstract/FREE Full Text
  36. ↵
    Liao Y, Smyth GK, Shi W. 2019. The R package Rsubread is easier, faster, cheaper and better for alignment and quantification of RNA sequencing reads. Nucleic Acids Res 47: e47–e47. doi:10.1093/nar/gkz114
    OpenUrlCrossRefPubMed
  37. ↵
    Liu X, Li YI, Pritchard JK. 2019. Trans effects on gene expression can drive omnigenic inheritance. Cell 177:1022–1034.e6. doi:10.1016/J.CELL.2019.04.014
    OpenUrlCrossRefPubMed
  38. ↵
    Mackay TFC, Stone EA, Ayroles JF. 2009. The genetics of quantitative traits: challenges and prospects. Nat Rev Genet 10:565–577. doi:10.1038/nrg2612
    OpenUrlCrossRefPubMedWeb of Science
  39. ↵
    Mallard F, Nolte V, Tobler R, Kapun M, Schlötterer C. 2018. A simple genetic basis of adaptation to a novel thermal environment results in complex metabolic rewiring in Drosophila. Genome Biol 19:119. doi:10.1186/s13059-018-1503-4
    OpenUrlCrossRef
  40. ↵
    Nielsen R. 2005. Molecular signatures of natural selection. Annu Rev Genet 39:197–218. doi:10.1146/annurev.genet.39.073003.112420
    OpenUrlCrossRefPubMedWeb of Science
  41. ↵
    Nouhaud P, Tobler R, Nolte V, Schlötterer C. 2016. Ancestral population reconstitution from isofemale lines as a tool for experimental evolution. Ecol Evol 6:7169–7175. doi:10.1002/ece3.2402
    OpenUrlCrossRef
  42. ↵
    Otte KA, Nolte V, Mallard F, Schlötterer C. 2020. The adaptive architecture is shaped by population ancestry and not by selection regime. bioRxiv 2020.06.25.170878. doi: 10.1101/2020.06.25.170878
    OpenUrlAbstract/FREE Full Text
  43. ↵
    Palmieri N, Nolte V, Chen J, Schlötterer C. 2015. Genome assembly and annotation of a Drosophila simulans strain from Madagascar. Mol Ecol Resour 15:372–81. doi: 10.1111/1755-0998.12297
    OpenUrlCrossRefPubMed
  44. ↵
    Pélabon C, Hansen TF, Carter AJR, Houle D. 2010. Evolution of variation and variability under fluctuating, stabilizing, and disruptive selection. Evolution (N Y) 64:1912–25. doi:10.1111/j.1558-5646.2010.00979.x
    OpenUrlCrossRefPubMedWeb of Science
  45. ↵
    Pritchard JK, Pickrell JK, Coop G. 2010. The genetics of human adaptation: Hard sweeps, soft sweeps, and polygenic adaptation. Curr Biol 20:R208–R215. doi:10.1016/J.CUB.2009.11.055
    OpenUrlCrossRefPubMedWeb of Science
  46. ↵
    Robinson MD, McCarthy DJ, Smyth GK. 2010. edgeR: a Bioconductor package for differential expression analysis of digital gene expression data. Bioinformatics 26:139–140. doi:10.1093/bioinformatics/btp616
    OpenUrlCrossRefPubMedWeb of Science
  47. ↵
    Rocke DM, Durbin B. 2003. Approximate variance-stabilizing transformations for gene-expression microarray data. Bioinformatics 19:966–72.
    OpenUrlCrossRefPubMedWeb of Science
  48. ↵
    Romero IG, Ruvinsky I, Gilad Y. 2012. Comparative studies of gene expression and the evolution of gene regulation. Nat Rev Genet 13:505–516. doi:10.1038/nrg3229
    OpenUrlCrossRefPubMed
  49. ↵
    Signor SA, Nuzhdin S V. 2018. The evolution of gene expression in cis and trans. Trends Genet 34:532–544. doi:10.1016/j.tig.2018.03.007
    OpenUrlCrossRef
  50. ↵
    Sork VL. 2017. Genomic studies of local adaptation in natural plant populations. J Hered 109:3–15. doi:10.1093/jhered/esx091
    OpenUrlCrossRef
  51. ↵
    Suvorov A, Nolte V, Pandey RV, Franssen SU, Futschik A, Schlötterer C. 2013. Intra-specific regulatory variation in Drosophila pseudoobscura. PLoS One 8:e83547. doi:10.1371/journal.pone.0083547
    OpenUrlCrossRefPubMed
  52. ↵
    Turelli M. 1984. Heritable genetic variation via mutation-selection balance: Lerch’s zeta meets the abdominal bristle. Theor Popul Biol 25:138–193. doi:10.1016/0040-5809(84)90017-0
    OpenUrlCrossRefPubMedWeb of Science
  53. ↵
    Wang L, Wang S, Li W. 2012. RSeQC: quality control of RNA-seq experiments. Bioinformatics 28:2184–2185. doi:10.1093/bioinformatics/bts356
    OpenUrlCrossRefPubMedWeb of Science
  54. ↵
    Wittkopp PJ, Haerum BK, Clark AG. 2008. Regulatory changes underlying expression differences within and between Drosophila species. Nat Genet 40:346–350. doi:10.1038/ng.77
    OpenUrlCrossRefPubMedWeb of Science
  55. ↵
    Wittkopp PJ, Haerum BK, Clark AG. 2004. Evolutionary changes in cis and trans gene regulation. Nature 430:85–88. doi:10.1038/nature02698
    OpenUrlCrossRefPubMedWeb of Science
  56. ↵
    Wu TD, Reeder J, Lawrence M, Becker G, Brauer MJ. 2016. GMAP and GSNAP for genomic sequence alignment: Enhancements to speed, accuracy, and functionality. Humana Press, New York, NY. pp. 283–334. doi:10.1007/978-1-4939-3578-9_15
    OpenUrlCrossRefPubMed
  57. ↵
    Yoo BH, Nicholas FW, Rathie KA. 1980. Long-term selection for a quantitative character in large replicate populations of Drosophila melanogaster. Theor Appl Genet 57:113–117. doi:10.1007/BF00253881
    OpenUrlCrossRef
Back to top
Posted January 19, 2021.
Download PDF
Email
Evolution of phenotypic variance provides insights into the genetic basis of adaption
Wei-Yun Lai, Viola Nolte, Ana Marija Jakšić, Christian Schlötterer
bioRxiv 2021.01.19.427260; doi: https://doi.org/10.1101/2021.01.19.427260
Twitter logo Facebook logo LinkedIn logo Mendeley logo
Citation Tools

Subject Area