Conceived and designed the experiments: SZ TFCM RRHA. Performed the experiments: SZ TGC. Analyzed the data: SZ EAS. Wrote the paper: SZ TFCM RRHA.
The authors have declared that no competing interests exist.
Phenotypic plasticity is the ability of a single genotype to produce different phenotypes in response to changing environments. We assessed variation in genome-wide gene expression and four fitness-related phenotypes of an outbred
Unlike Mendelian traits, where the genotype allows a direct prediction of the phenotype, predicting phenotypic values is not straightforward for complex traits, which arise from multiple segregating genes and their interactions with the environment. Here, a single genotype can often express different phenotypes in different environments. Such phenotypic plasticity is the counterpoint to “environmental canalization,” whereby genotypes produce the same phenotype in different environments. Whereas phenotypic plasticity allows organisms to respond rapidly to changing environments, environmental canalization buffers phenotypes against environmental perturbations. The balance between plasticity and robustness is crucial for optimal fitness, but the genetic basis for phenotypic plasticity is poorly defined. Here, we present the most comprehensive analysis to date of variation in genome-wide gene expression of an outbred
Phenotypic plasticity is the ability of a single genotype to give rise to different phenotypes in different environments
Elucidating the genetic underpinnings of phenotypic plasticity (and its converse, environmental canalization) requires that we determine what fraction of the genome is environmentally sensitive, which genes respond to the same or different environmental perturbations and how expression of environmentally sensitive genes is correlated with plasticity of organismal phenotypes. It is also necessary to determine what the relationship is between genetic variance and phenotypic plasticity, whether the same genes affecting phenotypic plasticity for a trait also affect genetic variation for that trait, and whether environmentally plastic and environmentally robust genes evolve at different rates. Although previous studies have analyzed changes in gene expression under one or few different environmental or physiological conditions
We examined phenotypic plasticity in genome-wide gene expression and four organismal phenotypes related to reproductive fitness in a population generated by crossing 40 wild-derived inbred
To identify phenotypically plastic and environmentally canalized transcripts, we assessed genome-wide gene expression of flies exposed to 20 treatments, including a control treatment of mated flies reared under standard conditions, and different nutrient or drug supplements, exposure to different physical and social environments, and maintenance at different reproductive states. Of the 18,800 transcripts represented on the microarray, 14,400 (76.6%) generated signal intensities above background under at least one treatment, similar to the proportion of the transcriptome detected in a previous study, in which transcript profiles were obtained separately for the 40 individual genotypes that gave rise to our outbred population
We refer to the 1,249 transcripts exhibiting phenotypic plasticity as quantified by the significant treatment term in the ANOVAs as Class I transcripts. To simplify statistical analyses and maintain optimum power we excluded 166 Class I transcripts that also had significant treatment by sex interaction terms, giving 1,133 phenotypically plastic transcripts for further analyses (
To further examine the relationship between gene expression and environmental exposure, we compared transcript abundance levels of the Class I transcripts under the different treatments to the standard rearing condition with
(A) Transcripts with differential expression levels under different experimental treatments compared to their expression under the standard condition. The blue-red color scale accentuates increasing numbers of transcripts. Pair-wise comparisons indicate the number of overlapping transcripts with differential expression under two conditions. (B) Proportion of phenotypically plastic transcripts. The gray area of the pie chart indicates the proportion of the transcriptome that does not undergo altered expression under 20 different environmental conditions. The red slice indicates the proportion of genes that show differential expression compared to their expression under the standard rearing condition, and the pie-chart insert shows the proportion of those genes that are affected by each of the 19 treatments. Treatments are ordered clockwise from the largest pie slice. Transcripts are identified in
Among the 1,133 Class I transcripts, 691 are computationally predicted with unknown function, 14 probe sets correspond to intergenic regions, non-coding RNAs and transposons, and 428 are annotated. The transcripts that show altered expression after heat shock include 13
Up-regulation or down-regulation of members of the cytochrome P450, glutathione-S-transferase and UDP-glucose-glycoprotein glucosyltransferase families under different treatments compared to the standard growth condition is indicated by red and blue boxes, respectively.
We asked to what extent expression patterns of phenotypically plastic transcripts are co-regulated across environmental treatments. A previous study on the 40 inbred lines from which our outbred population is derived demonstrated that the genetically variable transcriptome (10,096 transcripts) is highly inter-correlated and can be subdivided into 241 co-regulated modules
(A) Partitioning of the 1,133 Class I phenotypically plastic , shown in
What are the cellular mechanisms that regulate transcriptional responses to environmental changes? As a first step to investigating how environmental stimuli may influence transcriptional regulation, we asked which transcription factors show altered expression under the different environmental conditions. Among the Class I transcripts, we identified 26 transcripts that encode transcriptional regulators, of which 25 were differentially expressed relative to the standard growth condition (
(A) Diagram of the relationship between transcription factor regulation and rearing conditions. The 25 environmentally sensitive transcription factor transcripts are shown in circles and 16 treatment conditions are shown in magenta font in rectangular boxes. Red and blue lines designate up- and down- regulation of the transcription factor, respectively, under different treatments. Designations are: HSK, heatshock; STARV, starvation; DOP, dopamine (DOP); HY, high yeast; HT, high temperature; LT, low temperature; HS, high sugar; LC, larval crowding; AG, aging; CC, chill coma; NIC, nicotine: AC, adult crowding; VG, virgin; CL, constant light; MEN, menadione; HSHY, high sugar-high yeast. (B) Interaction networks of phenotypically plastic transcription factors. Interaction networks of the 25 transcription factors (red nodes) were analyzed through the
Each transcription factor can exert wide-ranging effects on networks of interacting genes that include other regulatory genes, non-regulatory genes and miRNAs
We next asked to what extent the phenotypic plasticity in gene expression is associated with phenotypic plasticity of organismal phenotypes. We assessed phenotypic plasticity of four fitness-related phenotypes: development time, lifespan, starvation stress resistance, and chill coma recovery time.
Development is exquisitely sensitive to environmental conditions
(A) Development latency. Development time was assessed under 14 conditions (adult stage treatments were excluded). The X-axis indicates eclosion times after egg collection for sexes pooled. (B) Lifespan. Average survival times were measured for flies reared under 19 different experimental treatments. (C) Starvation stress resistance. The number of dead flies was counted at different times following food deprivation under 19 different treatment conditions. (D) Chill coma recovery time. Average recovery times from chill coma were measured for flies reared under 19 different experimental treatments. Blue and red bars indicate males and females, respectively. Error bars, s.e.m.
In addition to prolonging development, growth at 18°C results in a two-fold increase in lifespan (
Whereas caloric restriction extends lifespan
We used regression to identify Class I transcripts associated with variation in organismal phenotypes across the 20 environmental conditions, and MMC to construct environmentally correlated modules
(A) Clustering of 426 genes significantly associated with variation in developmental latency into 116 modules. (B) Clustering of 186 genes significantly associated with variation in lifespan into 16 modules. (C) Clustering of 320 genes significantly associated with variation in starvation resistance into 32 modules. (D) Clustering of 440 genes significantly associated with variation in chill coma recovery into 23 modules. The modules populate the diagonal and are ordered by decreasing strength from the upper left to the lower right. Transcripts associated with the four phenotypes are indicated in
Some modules associated with different organismal phenotypes are enriched for common transcripts, indicating pleiotropy for phenotypic plasticity (
In summary, clustering analysis of Class I transcripts reveals a fragmented modular organization and distinct environmentally-responsive transcriptional signatures for the four fitness-related traits.
To assess the relationship between genetic variation and phenotypic plasticity, we compared the previously reported genetic variance and micro-environmental variation (within-line variation) across the 40 inbred lines
(A, B) Relationships between coefficients of genetic variance of inbred lines (CVL) and coefficients of macroenvironmental variance across treatments (CVME) in males (A) and females (B). (C, D) Distributions of coefficients of genetic variance of inbred lines with respect to mean transcript expression levels over all environments in males (C) and females (D). (E, F) Correlation structures between coefficients of macroenvironmental variance and mean transcript expression levels over all environments in males (E) and females (F). Class II transcripts explain the majority of the correlation structure. Red dots indicate Class I transcripts, green and purple dots indicate Class II transcripts in males (A, C, E) and females (B, D, F), and grey dots indicate robust transcripts. (G,H) Average coefficients of genetic variance across inbred lines (light shades) and macroenvironmental variance across treatments (dark shades) of each transcript class in males (G) and females (H).
This comparison revealed an additional group of 982 environmentally sensitive transcripts with high macroenvironmental variation, but low genetic variance (
There is little correlation between the level of genetic (
Since Class II transcripts exhibited sexual dimorphism in phenotypic plasticity, we evaluated the relationship between sexual dimorphism in mean gene expression across all 20 environments, and sexual dimorphism for phenotypic plasticity, for Class I and Class II transcripts as well as a sample of robust transcripts (
Class II phenotypically plastic transcripts can be further classified into high and low expression categories. Highly expressed transcripts in females overlap transcripts with low expression in males, and GO analysis shows that these 19 transcripts encode yolk proteins and chorion proteins and are enriched for oogenesis and sexual reproduction (
To assess to what extent phenotypically plastic genes are evolutionarily conserved compared to the rest of the genome, we looked at the percentage of homologues across 12
(A) Percentage of homologues across 12
Class II genes show an even faster rate of evolution compared to robust transcripts with a significantly higher proportion of positively selected sites, as evident from the distributions of ω and α (
Genome-wide transcriptional analysis of flies reared under 20 environmental conditions shows that ∼15% of the transcriptome exhibits phenotypic plasticity, while the rest is environmentally canalized. Logistical and economic constraints have limited this initial investigation to whole flies. We surveyed the FlyAtlas database
Class I transcripts are not only phenotypically plastic, but are more genetically variable and evolve more rapidly than the rest of the transcriptome. Class I transcripts are enriched in functions of detoxification, metabolism, proteolysis and heat shock proteins. Class I transcripts also encode gene products of unknown function, including non-coding RNAs, which may contribute to modulation of chromatin structure and transcriptional regulation. The coupling of high genetic variation within a population and rapid evolution suggests interesting evolutionary forces acting on these genes.
Class II transcripts have low genetic variance for mean expression levels, but greater environmental variation in transcript abundance, and are even more rapidly evolving than Class I transcripts. It is tempting to speculate that reduced genetic variation for these transcripts within a population is the consequence of selection favoring genotypes with high phenotypic plasticity within each species, but with variable selection pressures across species
Two models of the genetic basis of phenotypic plasticity have been postulated
We generated a synthetic outbred population by round-robin crossing of 40 wild derived inbred lines of the
For nutritional and pharmacological treatments, flies were reared on standard medium supplemented with 225 ml/L molasses (‘high sugar’), 65 g/L yeast (‘high yeast’), 225 ml/L molasses and 65 g/L yeast (‘high sugar-high yeast’), 10% (v/v) ethanol, 200 µM fluoxetine hydrochloride, 47 mM dopamine, 1 mM nicotine, 2 mM caffeine or 4 mM menadione sodium bisulfite. Different physical environments included constant light, 28°C (‘high temperature’), 18°C (‘low temperature’), and exposure to different stresses, including heat shock (37°C for 1 h; 1 h recovery prior to RNA extraction), chill coma (3 h on ice; 1 h recovery prior to RNA extraction), and 24 h starvation. Different social environments included larval crowding (300 eggs/vial) and adult crowding (80 females and 80 males were pooled in each vial immediately after eclosion). To compare mated with non-mated flies, 50 single sex virgins were reared separately. Flies reared under standard conditions were mated. Aged flies were 30 days old.
We used Affymetrix Drosophila 2.0 arrays to assess whole genome transcriptional profiles. Males and females (3–5 days old) were collected between 1:00–3:00 pm by aspiration and immediately frozen on dry ice. RNA was extracted from three independent samples (30 flies/sex/condition), and 10 µg of biotinylated, fragmented cRNA was hybridized to each microarray. RNA extraction, labeling and hybridization were randomized across samples. Raw data were log2 transformed and normalized across sexes and conditions using a median standardization. For each probe set, we used the median log2 signal intensity as the measurement of expression. We used negative control probe sets to estimate background intensity. Probe sets with hybridization intensities below background under all different treatment conditions were removed from the analysis. We did not correct for probe mismatches due to segregating polymorphisms in the reconstituted outbred population, because (1) the average hybridization bias will be identical across all environmental conditions, and; (2) only about 3,000 single feature polymorphisms (SFPs) were identified among the original 40 inbred lines previously and their removal from the data set did not significantly influence the hybridization results
We analyzed array data using a Generalized Linear Model (GLM) in SAS to partition phenotypic variation between sexes (S, fixed), environments/treatments (E, fixed), the S×E interaction (fixed) and the error variance (ε). To identify environmentally responsive Class I transcripts we used an FDR<0.05 to correct for multiple tests.
To resolve Class II transcripts, we applied two filters. First, we selected transcripts with cross-treatment (macroenvironmental) variance >95th percentile of the macoenvironmental variance distribution of the Class I transcripts (coefficients of variation across treatments >7.06 and >7.12 for females and males, respectively). We filtered these transcripts further using an FDR>0.0001 for genetic variation among DGRP lines
The Modulated Modularity Clustering (MMC) algorithm
To measure development time, we allowed flies to lay eggs for 3 hours (10:00am–1:00pm), after which 55 eggs were collected and placed under 14 different growth conditions (300 eggs were collected to assess development time under the larval crowding condition). We counted flies, sexes separately, that eclosed every 12 hours (N = 4 vials/condition). Life span was measured by collecting three females and three males immediately 1–3 days after eclosion, transferring them to fresh vials every 2–3 days, and recording survival daily (N = 26 replicates/condition). To measure starvation stress resistance, we placed ten 3–5 days old flies in vials containing 1.5% agar, and scored survival every 8 hours (N = 4×10/sex/condition). To measure chill coma recovery, we placed 3–7 day-old flies in empty vials on ice for 3 hours, and determined their subsequent recovery time at room temperature by their ability to recover upright posture (N = 2×50 flies/sex/condition). Phenotypic data are available on the DGRP website (
We used regression to identify transcripts with variation in expression levels that associated with organismal phenotypic variation (
Variation of transcript abundance across 20 rearing conditions. The diagram represents 1,133 transcripts that show significant differences in expression levels across conditions. An additional 116 transcripts with a significant sex-by-environment interaction have been excluded from the diagram. Transcripts represented in the figure are sex centered. The color scale reflects the rank order of expression for individual genes across the 20 conditions with red and blue intensities indicating higher and lower expression levels, respectively. The 20 conditions from left to right, are sorted based on the number of genes with transcript levels higher than their median across conditions. Transcripts are identified in
(PDF)
Examples of phenotypically plastic gene families. The diagram illustrates up-regulation or down-regulation, indicated by red and blue boxes, respectively, of members of the
(PDF)
Pleiotropy between covariant transcriptional modules associated with four organismal phenotypes. Strength of connectivity within modules along the diagonals increases in a clockwise direction. Black lines connect modules with similar composition of covariant Class I phenotypically plastic transcripts, associated with variation in development latency, lifespan, starvation stress resistance and chill coma recovery time, shown in
(PDF)
Distribution of genetic and environmental variation between transcript classes. (A, B) Box plots of coefficients of genetic variation (CVL) across inbred lines of Class I, Class II and robust transcripts in males (A) and females (B). (C, D) Box plots of coefficients of macroenvironmental variation (CVME) of the three transcript classes in males (C) and females (D).
(PDF)
Variance analysis and classification of environmentally sensitive transcripts. (A, B) Correlation structures between coefficients of within-treatment variance (CVEW) and mean transcript expression levels across all treatments for males (A) and females (B). (C, D) Correlation structures between standard deviations of the coefficients of within-treatment variance (STD_CVEW) and mean transcript expression levels across all treatments in males (C) and females (D). Class I transcripts have significant lower within-treatment variation than Class II transcripts. Transcripts associated with correlation structures that are not explained by Class II transcripts are also distinct from Class I transcripts with higher within-treatment variation (CVEW) and variance of the within-treatment variations (STD_CVEW). (E, F) Relationships between coefficients of within-line variation (CVE) and mean transcript expression levels across all treatments in males (E) and females (F). (G, H) Correlations between mean transcript expression across the original 40 inbred lines and the mean transcript expression across the 20 treatments of the reconstituted outbred population in males (G) and females (H). Red dots indicate Class I transcripts, green and purple dots indicate Class II transcripts in males (A, C, E, G) and females (B, D, F, H), respectively, and grey dots indicate robust transcripts.
(PDF)
Relationship between sexual dimorphism for the mean and coefficient of macroenvironmental variance of gene expression. (A) All Class I phenotypically plastic transcripts. (B) All Class II phenotypically plastic transcripts. (C) A random sample of 1,500 robust transcripts.
(PDF)
Distribution of ω (dN/dS), using 6 outgroup species
(PDF)
Distribution of expression patterns of modules associated with development time, life span, starvation resistance and chill coma recovery time. Tissue specific expression patterns of modules associated with (A) development time, (B) life span, (C) starvation resistance, and (D) chill coma recovery time were analyzed using the FlyAtlas database
(PDF)
Quantitative genetic analyses of variation for 14,400 expressed transcripts of the outbred population across 20 treatments. Expression is measured as the median log2 intensity of PM transcripts in each probe set. Class indicates the Class I and Class II phenotypically plastic transcripts. FDR is False Discovery Rate, CV is the coefficient of variation, CVME is the coefficient of cross treatment (macroenvironmental) variance, CVEW is the mean coefficient of within treatment variation, and STD_CVEW is the standard deviation of coefficient of within treatment variation. Std_Mean is the standard deviation of treatment mean. Inbred line means, genetic variation among lines (CVL) and micro-environmental variation within lines (CVE) are adopted from (12). MMC (13) revealed 103 modules of 1,133 Class I transcripts. |r| is the average correlation of each variable transcript with all other variable transcripts.
(XLSX)
Pair-wise comparisons of the phenotypically plastic Class I transcripts under different treatments with the standard growth condition.
(XLSX)
Gene ontology analysis of Class I phenotypically plastic transcripts using DAVID*.
(XLSX)
Class I phenotypically plastic transcripts associated with development time and clustered into modules using MMC (13).
(XLSX)
Class I phenotypically plastic transcripts associated with life span and clustered into modules using MMC (13).
(XLSX)
Class I phenotypically plastic transcripts associated with chill coma and clustered into modules using MMC (13).
(XLSX)
Class I phenotypically plastic transcripts associated with starvation resistance and clustered into modules using MMC (13).
(XLSX)
Gene ontology analysis of overlapping Class II transcripts between high expressed transcripts in females and low expressed transcripts in males using DAVID (34).
(XLSX)
Gene ontology analysis of overlapping Class II transcripts between low expressed transcripts in females and high expressed transcripts in males using DAVID (34).
(XLSX)
Gene ontology analysis of female specific Class II transcripts using DAVID (34).
(XLSX)
Gene ontology analysis of male specific Class II transcripts using DAVID (34).
(XLSX)
We thank Dr. Akihiko Yamamoto, Faye Lawrence, Laura Duncan, Elizabeth Jones, and Julien Ayroles for technical support.