Gene expression profile analysis of Drosophila melanogaster selected for resistance to environmental stressors


Department of Ecology and Genetics, University of Aarhus, Ny Munkegade, Building 540, DK-8000 Aarhus C, Denmark.
Tel.: +45 8942 3135; fax: +45 8942 2722; e-mail:


Here, we report a detailed analysis of changes in gene expression in Drosophila melanogaster selected for ecologically relevant environmental stress resistance traits. We analysed females from seven replicated selection regimes and one control regime using whole genome gene expression arrays. When compared with gene expression profiles of control lines, we were able to detect consistent selection responses at the transcript level in each specific selection regime and also found a group of differentially expressed genes that were changed among all selected lines. Replicated selection lines showed similar changes in gene expression (compared with controls) and thus showed that 10 generations of artificial selection give a clear signal with respect to the resulting gene expression profile. The changes in gene expression in lines selected for increased longevity, desiccation and starvation resistance, respectively, showed high similarities. Cold resistance-selected lines showed little differentiation from controls. Different methods of heat selection (heat survival, heat knock down and constant 30 °C) showed little similarity verifying that different mechanisms are involved in high temperature adaptation. For most individual selection regimes, and in the comparison of all selected lines and controls, the gene expression changes were exclusively in one direction, although the different selection regimes varied in the direction of response. The responses to selection restricted to individual selection regimes can be interpreted as stress specific, whereas the response shared among all selected lines can be considered as a general stress response. Here, we identified genes belonging to both types of responses to selection for stress resistance.


Environmental abiotic factors play a key role in determining the distribution and abundance of species (Cossins & Bowler, 1987; Hoffmann et al., 2003). The forces of the environmental factors have promoted adaptive evolution and general and specific responses to cope with environmental challenges. To better understand these effects, laboratory studies on Drosophila have been widely used to accumulate evidence on the physiological basis of such adaptations and responses in insects. Some candidate genes and processes have been identified, e.g. the stress-inducible heat shock proteins (HSPS), which are induced by a variety of environmental stressors (for reviews, see, e.g. Feder & Hofmann, 1999; Sørensen et al., 2003) and genes involved in the general metabolism affecting metabolic rate or efficiency (Parsons, 2005; Sørensen et al., 2005). Although a number of adaptive mechanisms are known, it is clear that other mechanisms and currently unidentified genes must be involved as the known mechanisms do not fully explain resistance. Even though both high and low temperatures can induce the heat shock response, induced HSPS seem to play little role in heat knock down, acute cold stress survival and the ability to rapidly cold harden at lowered temperatures within a few hours (Nielsen et al., 2005; Overgaard et al., 2005). Moreover, the generality of different stress responses and adaptive mechanisms to cope with stress, i.e. the occurrence of cross-resistance and correlated responses to selection is often asymmetric and not well understood (Hoffmann et al., 2003; Bubliy & Loeschcke, 2005). Evolutionary physiologists and geneticists often search for specific and universal mechanisms that promote resistance to multiple forms of stress (e.g. Hoffmann & Parsons, 1991). Some insight into the basis for stress resistance comes from investigations of genetic associations between stress resistance traits and between stress resistance and longevity by comparing correlated responses with artificial selection in Drosophila melanogaster. An analysis of correlated responses in laboratory selection experiments on Drosophila has been a traditional tool for testing evolutionary hypotheses on stress resistance and longevity (Harshman & Hoffmann, 2000; Bubliy & Loeschcke, 2005).

The availability of full genome microarrays has allowed screening of the full genome for changes in gene expression under different conditions (but see Michalak, 2006). Global gene expression studies in Drosophila have provided new insights to various biological processes, mostly based on response to treatments and interactions between genotype and the environment (e.g. Landis et al., 2004; Sørensen et al., 2005; Kristensen et al., 2006). Additionally, recent studies addressed changes in untreated flies after selection for resistance or inbreeding have been published (Toma et al., 2002; Pedra et al., 2004; Harbison et al., 2005; Kristensen et al., 2005). This method can provide a wealth of information at the level of gene expression and identify genes and pathways involved in various processes, making it an excellent tool for such exploratory studies.

Identification of new genes of importance for evolutionary adaptation and immediate responses to environmental stress has relevance for many fields of biology. Stress responsive genes are of interest not only for the study and understanding of environmental stress resistance and the stress response in general, but also for protein folding-mediated diseases in humans, immunological responses, animal breeding, genetic stress, protein quality control, developmental biology and in the study of gene regulation.

In the present study, we used the microarray technique to test for patterns of gene expression in D. melanogaster selected for increased resistance to different environmental stress factors. We tested lines selected for increased survival rates in adults after severe cold, three different heat resistance traits, desiccation and starvation stresses as well as extended life span under nonstressful conditions. Individuals selected to survive cold and heat stresses were first acclimated/hardened with nonlethal low and high temperatures respectively. This was done to focus selection on tolerance to the highest stresses, which flies can resist and still reproduce. We have investigated the responses to selection on the gene expression level in untreated female flies from replicated lines for each selection regime and corresponding controls. Differential expressed genes in each selection regime and genes generally changed by selection for all stress resistance types were investigated to find general stress genes. The effect of selection for increased heat shock survival before and after a mild heat shock was already addressed by Nielsen et al. (2006) for these lines.

We show that selection for environmental stress resistance manifests itself consistently in the gene expression pattern of unstressed flies and that the response to selection is apparently composed of a signal specific to individual selection regimes and a signal shared among all stress selection regimes, interpreted as a general stress selection response. Also, the patterns of gene expression to a large extent are consistent with phenotypic stress resistance data (see Bubliy & Loeschcke, 2005), suggesting that the effect of selection consistently can be followed from the level of phenotype to that of gene expression.

Materials and methods

Origin of flies and selection regimes

The replicate selection and control lines were derived from a mass laboratory population of D. melanogaster established in September 2002 (for details see Bubliy & Loeschcke, 2005). The mass population was kept at 25 °C on standard oatmeal–sugar–yeast–agar Drosophila medium. There were 25 bottles in total with c. 50 pairs of parents per bottle. The genetic diversity of the starting material was ensured by mixing 600–700 flies from each of four pre-existing laboratory stocks that had all been kept at discrete generations using large numbers of breeding individuals. The stocks were: (1) Hov–Copenhagen basic strain collected in October 1997 from two Danish localities, Hov (Jutland peninsula, east coast) and Hvidovre (Zealand island, near Copenhagen). They were kept as 27 and 30 isofemale lines, respectively, until February 1998 where one large interbreeding population was founded. (2) Supermass Hov–Copenhagen founded in September 2001 by mixing a few sets of heat and longevity selection lines. The heat selection lines were started from the 17th generation of the Hov–Copenhagen basic strain. The longevity selection lines were established in April 2000 also by sampling flies from the Hov–Copenhagen basic strain. (3) Heat knock down originated from two sets of highly inbred laboratory lines described by Norry et al. (2004). The first of them (SH) originated from Australian flies collected near Melbourne in February 1994 and selected for increased heat knock-down resistance. The flies for the second set of lines (D) were sampled from the 10th generation of the Hov–Copenhagen basic strain and selected for reduced heat knock-down resistance. These two sets of lines had been mixed for several generations prior to becoming part of the base population. (4) The Leiden population originated from 30 isofemale lines founded by females collected near Leiden (the Netherlands) in October 1999. For the first five generations, it was maintained at 25 °C and then at 20 °C. The experimental lines were established by flies from the fourth generation of the mass population and from then on each replicate line was kept in five culture bottles using c. 30 pairs of parents per bottle to reduce density. The parents were allowed to oviposit in the bottles for a period of time needed to obtain the same moderately high level of larval density in all lines. This level was such that there was no delay in pre-adult development and most of the flies emerged on day 9.

Eight experimental regimes were used with three biological independent replicate lines assessed for gene expression per regime (total 24 chip experiments). These were unselected control (UC) maintained during the whole experiment without selection treatment. Emerged flies were aged for 5 days before the next generation was initiated. Cold shock resistance selection (CS) took place by ageing emerged flies for 2 days and acclimating them at 11 °C for another 5 days. The acclimated individuals were then chilled at 0.5 °C at close to 100% relative humidity (RH). The initial chilling period of 27 h was gradually increased to 50 h following changes in cold resistance of the selected lines. The cold-stressed flies were allowed to recover for 24 h at 25 °C before the next generation was established. Heat shock resistance selection (HS) was performed using 5-day-old flies that were hardened for 30 min at 36.0 °C in a preheated water bath. The hardened flies recovered for 20 h at 25 °C and were then exposed to 38.0 °C for 1 h in a water bath. The temperature and exposure time was gradually increased to maintain the mortality constant as the lines improved their resistance because of selection. Survivors recovered for 24 h at 25 °C before the next generation was established. Heat knock-down resistance selection (KD) was performed at 40.0 °C on 5-day-old flies using a preheated knock-down tube (Huey et al., 1992; Sørensen et al., 2001). For each run, 300 pairs of flies were placed in the tube. The flies succumbed to heat and rolled down a series of baffles reaching a collecting vial, which was replaced every 30 s. Flies with longer knock-down time were mixed and transferred to culture bottles to start a new generation. Desiccation resistance selection (DS) used 5-day-old flies that were transferred to a standard exsiccator containing desiccant gel. Time of exposure to desiccation stress was increased from 12 to 20 h during the course of the experiment. Starvation resistance selection (ST) took place immediately after emergence. Flies were transferred to vials with pure agar medium (2% of agar) to provide moisture and avoid desiccation while starving. The time needed to reach the desired mortality rate because of starvation stress was gradually increased from 35 to 60 h during selection. Longevity selection (LS) followed this protocol. After emergence, flies were immediately placed in vials with the standard medium. Every second day they were transferred to fresh food vials until the desired mortality rate was reached. This period was 4 weeks at the beginning of the experiment and was later increased to 6 weeks. The live individuals were used to start a new generation. Constant 30 °C (C30) lines were set up at 25 °C as the controls, but were kept at constant 30 °C from egg to emergence of adults. Hereafter, flies were moved to 25 °C to regain fertility. In all selection regimes except C30 adult flies were the selected life stage. For all regimes selection was implemented every second generation to allow the population to recover and avoid any cross-generational effects (see, e.g. Watson & Hoffmann, 1996; Hercus & Hoffmann, 2000). Selection lines were treated in the same way as the control lines (UC) during the relaxed generations. All flies were reared at 25 ± 1 °C, 12 : 12 h light : dark cycle in 100-mL bottles containing 35 mL of standard oatmeal–sugar–yeast–agar medium. For nonstressed parents in the selection lines as well as for parents in the control lines (UC) the period of egg laying was 1 day, whereas for stress-selected parents it was either 2 (KD, DS) or 3 days (CS, HS, ST) to equalize density. The selected parents in the longevity test were kept in the bottles for 3 days. In every generation, progeny collected from different bottles of the same replicate line were mixed. For the mortality assays (CS, HS, DS, ST) and longevity test (LS), flies from each replicate line were distributed among 10 shell vials (100 × 24 mm, 7 mL of medium) with 30 pairs per vial. The severity of all the stress treatments and the ageing period in the longevity test were such to give a mortality of c. 50% except for the laboratory natural selection regime of C30 where mortality has been estimated to be considerable lower (Doth Andersen, pers. obs.). The flies used for this experiment had gone through 10 selection events (20 generations of maintenance), except the LS lines that only had passed seven selection events. Before sampling all lines were maintained at control conditions for two generations at approximately equal density to avoid cross-generational effects (see, e.g. Watson & Hoffmann, 1996; Hercus & Hoffmann, 2000; Faurby et al., 2005).

RNA extraction and array hybridization

The Affymetrix array used here contains 13 966 probe sets representing approximately 13 000 unique genes. Each assayed sample consisted of 10 female flies. The RNA extraction and creation of cDNA was performed as described in Dyrskjot et al. (2003). Fifteen micrograms of cDNA was fragmented and loaded onto the Affymetrix probe array cartridge (Drosophila Genome Array, 13.966 probe sets). Incubation, washing and staining procedure was performed in the Affymetrix Fluidics Station. The probe arrays were scanned at 560 nm using a confocal laser-scanning microscope (Hewlett Packard GeneArray Scanner G2500A; Hewlett-Packard, Palo Alto, CA, USA). All steps were performed as described in Sørensen et al. (2005).

Statistical analysis

The raw data was GC-RMA normalized with the BIOCONDUCTOR application for R (Wu & Irizarry, 2004). We used the significant analysis of microarrays (SAM) method to select probe sets differentially expressed among selection and control regimes (Tusher et al., 2001). As it tests each gene separately it is less affected by the number of tests and all SAM analyses were run on the full data set; thus, no initial filtering was performed. The SAM procedure allows the identification of genes differentially expressed consistently across replicates while controlling for multiple testing by estimating the false discovery rate (FDR) (Benjamini & Hochberg, 1995). For the SAM analysis, we used 100 permutations (or all unique permutations if less than 100), which gave a stable number of significant genes and estimates of FDR among runs. Significant genes were further analysed by applying hierarchical clustering of experiments (chips) to identify patterns among selection regimes. Following this, all selection regimes were compared with the control flies separately using a two-class unpaired SAM. We also tested for a signal of a general stress response to selection for environmental stress resistance by comparing controls with all selected regimes (two-class SAM with three control and 21 selected samples/experiments). SAM and hierarchical trees were performed using the computer package TIGR MeV v. 3.0.1 (Saeed et al., 2003). The genes from each comparison were analysed further to establish functional groups of genes that increased or decreased over time and to establish functional links. Information from the gene ontology (GO) (Gene-Ontology-Consortium, 2001) database was combined with expression data, using the EASE application on the DAVID homepage ( (Hosack et al., 2003) to evaluate over-represented functional groups of genes in each gene list. In the EASE application, each probe set was assigned, when one was available, to its GO annotation. The number of probe sets for the different GO terms was computed for each gene list and a probability for recovering this number, given the number of genes in the data set, was assigned to each represented GO term (Hosack et al., 2003).

Differential expression of functional groups of genes after selection was additionally investigated by the Global test method of Goeman et al. (2004). For this analysis the data set was filtered so that only probe sets consistently detected in the experiments were retained. This was done to reduce the number of probe sets tested and, thus, to exclude data not contributing information. The criterion for retaining a probe set in the data set was that it was assigned as ‘present’ by the Affymetrix GeneChip software and had expression intensity above 50 units on at least one of the 24 experiments/arrays. This reduced the number of probe sets tested to 7753. The Goeman Global test investigates whether groups of genes are significantly differently expressed by testing the change of all genes in each GO category and calculating the influence of each gene. Thus, whereas the SAM analysis successfully detects single genes with large effect, the Goeman Global test is a powerful tool to detect smaller but consistent change in functional groups of genes. In this way the Goeman Global test complements the SAM approach and as selection is expected to affect many genes to a small extent it might be better suited to this data set. The GO categories reported from the Goeman Global test were considered significant at the 1% level.

The chromosomal positions of detected genes were examined to test if they were randomly distributed throughout the Drosophila genome. For each gene list the band positions were recorded and genes within a span of five band units were grouped and counted. The deviation from a random distribution was tested with a chi-squared test. To account for differences in gene density along the genome, we determined the distribution of all array probe sets and used this as a base distribution to calculate the expected number of genes in each class and gene list.

Results and discussion

General analysis of the selection responses

Initial analysis of the selection response with multiclass SAM (eight groups tested) showed small differences in gene expression patterns among the lines. Accepting an FDR of 20% we detected 262 genes significantly differentially expressed among selection regimes. Gene expression was changed only slightly, as fold changes between controls and the most down- or up-regulated genes in each selection regime rarely exceed twofold (Table 1). However, in our opinion it is not unexpected that selection for complex traits such as stress resistance leads to quantitatively small changes across many genes.

Table 1.   Fold change for significantly differently expressed genes for each selection regime.
 All selec. (up)ST (down)LS (down)KD (up)HS (up)HS (down)DS (up)DS (down)C30 (down)CS
  1. (All selec., all selected lines; ST, starvation selected; LS, longevity selected; KD, heat knock down selected; HS, heat survival selected; DS, desiccation selected; C30, constant 30 °C selected, CS, cold survival selected).

  2. Table shows the mean, std. error (SE), maximum (Max) and minimum (Min) observed fold change for genes identified in each SAM contrast with controls and for both up- and down-regulated genes when present. In cases where genes were down regulated (FC < 1), the corresponding reciprocal value is given in parenthesis for easier comparison.

Mean1.310.86 (1.16)0.85 (1.17)1.361.290.79 (1.27)1.220.87 (1.14)0.86 (1.16)
Max1.840.57 (1.73)0.65 (1.54)2.211.830.34 (2.96)1.270.55 (1.82)0.77 (1.30)
Min1.070.96 (1.04)0.96 (1.04) (1.09)1.170.95 (1.05)0.92 (1.09)

Hierarchical clustering of experiments

The 262 genes differentially expressed among all lines were used as the basis for hierarchical clustering of experiments/chips. The resulting tree showed structure with respect to selection regimes, i.e. independent selection replicates clustered together (Fig. 1) suggesting that the differences in gene expression observed are the result of the selection procedure. We were thus able to trace the selection to consistent changes at the gene expression level across independent biological replicates. The hierarchical tree suggests that lines selected for increased heat shock survival are most distant from the other lines. Selection for increased tolerance to DS, ST and LS make up a group separated from, but closer to controls than to HS-selected lines (Fig. 1). Only two regimes do not cluster with all three biological independent lines within a single branch (CS and KD) suggesting that the selection for these traits produced little differentiation from other lines or relatively high variance among replicated selection lines. However, in both cases, two of the three lines branch together and in the remaining six selection regimes all replicate lines cluster within a single branch. Thus, despite some outliers a consistent change in gene expression within each selection regime was observed after only 10 generations of selection. We detected a gene expression pattern that separated the different selection regimes (Fig. 1) and also found a group of differentially expressed genes that were generally shared by all selected lines.

Figure 1.

 Result of hierarchical clustering of experiments based on 262 significant genes identified by multiclass significant analysis of microarrays on the full data set. Lines used were unselected controls (UC), cold shock resistance (CS), heat shock resistance (HS), heat knock-down resistance (KD), desiccation resistance (DS), starvation resistance (ST), longevity (LS) and constant 30 °C (C30) selection respectively. For six selection regimes (all except KD and CS) the three arrays on independent biological replicate lines clustered together, suggesting a distinct and uniform effect of selection on gene expression. For the two remaining selection regimes two of the three replicate arrays formed a distinct group, whereas the remaining arrays in both cases clustered with the HS branch outside the main cluster. This does not necessarily mean that the gene expression resembled the expression of HS lines, but more likely that variation in these two outlier arrays was larger than in the rest of the data set.

Contrasts of individual selection regimes with controls by SAM

Subsequently, each selection regime was contrasted to the controls to investigate the genes specifically changed by each selection procedure. We also contrasted the controls to all selected samples pooled, to look for a general response to selection for stress resistance (denoted all selected).

Because of different degrees of response in gene expression in different selection regimes, the analysis led to large differences in the number of genes and certainty of the test among the different contrasts performed. Thus, the accepted FDR was individually set for each contrast (10% for the contrasts with HS, DS, ST and to 20% for contrasts with C30, KD, LS, CS, all selected). We considered this acceptable, as we did not interpret individual genes, but base an important part of the following analysis on screening for groups of functionally related genes for a list of the number of genes detected at different FDR). Thus, the probability that gene expression change is related to selection for stress resistance is much higher when several genes from a certain molecular pathway or process are simultaneously detected.

The number of genes detected by individual SAM contrasts and the overlap in genes among contrasts are given in Table 2 and the functional GO categories detected by EASE are given in Table 3. Individual gene lists for each contrast are available in. Some interesting patterns were observed in these analyses suggesting that relevant genes were detected even though relatively few genes at a high FDR threshold were found. First of all, the number of genes found to differ in each contrast supports the hierarchical cluster. Controls and cold selected clustered together and we found no genes differentially expressed between these selection regimes. Desiccation and starvation selection differ most from control in terms of the number of genes and are also distant to controls in the hierarchical tree. Another interesting feature is that the different measures of heat selection (HS, KD and C30) showed little similarity suggesting that these traits share different underlying mechanisms. This is also supported by the hierarchical tree, where each heat resistance selection regime is found in a separate main branch (Fig. 1).

Table 2.   Number of genes detected as significantly differentially expressed between controls and individual selection regimes by SAM analysis are given along the diagonal in bold. Arrows indicate up- and down regulation, respectively. The number of individual genes found to be significant in multiple comparisons is given in the table.
  1. (All selec., all selected lines; ST, starvation selected; LS, longevity selected; KD, heat knock down selected; HS, heat survival selected; DS, desiccation selected; C30, constant 30 °C selected; CS, Cold survival selected).

All selec.262  0 0 6 5  0 0
ST 23021 0 6 50 8
LS  64 0 3 16 2
KD   21 4  0 0
HS    33  2 0
DS    61260 6
C30       224
CS       0
Table 3.   Significant GO annotations among the significant genes from the SAM analysis of all selection regimes. The GO annotations are divided into three systems; Molecular Function (MF), Cellular Component (CC) and Biological Process (BP). In many cases several related GO categories are significant and in some of these cases the more general categories are left out and thus not reported here.
AnalysisSystemCategoryNo. # genesEASE Score
All regimes multi class (260 genes)MFIon transporter activity191.86E-05
MFHydrogen Ion transporter activity120.000166
CCMitochondrial inner membrane130.000225
CCRespiratory chain complex I60.00335
CCCytosolic ribosome (sensu Eukarya)70.0228
All selec (262 genes)BPDetection of light112.49E-11
CCPlasma membrane270.000000026
BPResponse to abiotic stimulus150.000000113
BPDeactivation of rhodopsin mediated signaling60.000000372
BPRhodopsin mediated phototransduction60.00000463
MFG-protein coupled photoreceptor activity50.0000113
MFPhotoreceptor activity50.0000113
BPResponse to external stimulus190.0000267
MFCytoskeletal protein binding140.0000681
MFCalcium ion binding90.000139
MFPorter activity120.000449
MFElectrochemical potential-driven transp. act.120.000449
MFMetal ion binding90.000758
BPPhototransduction, UV30.00241
MFCarrier activity190.00265
BPG-protein coupled recep prot. sign. pathway90.00274
MFSolute:sodium symporter activity50.00395
BPNeurotransmitter metabolism30.00832
MFEnzyme inhibitor activity60.0198
BPBiogenic amine metabolism30.0228
ST (230 genes)BPNucleo-base, -side, -tide and nucleic acid met.220.00189
MFProtein binding150.00404
BPPhysiological process610.00438
BPTranscription, DNA-dependent110.00911
MFTranscription cofactor activity40.0239
MFActin binding60.025
MFTranscription corepressor activity30.0328
MFEnzyme regulator activity90.0331
BPRegulation of transcription90.0435
CCSpliceosome complex60.0466
BPCarbohydrate catabolism30.0467
BPTranscription from Pol II promoter70.0479
LS (64 genes)MFGTP binding30.0318
MFGuanyl nucleotide binding30.0367
MFPurine nucleotide binding30.0486
KD (21 genes)MFGlutathione transferase activity20.0576
HS (all) (94 genes)CCRibonucleoprotein complex100.011
MFStructural constituent of ribosome60.0168
CCCytosolic ribosome (sensu Eukarya)50.0232
MFProtein binding90.0257
BPMorphogenesis of an epithelium50.0293
HS (up)CCCytosolic ribosome (sensu Eukarya)50.000678
(33 genes)MFStructural constituent of ribosome50.00208
BPProtein biosynthesis70.0113
CCRibonucleoprotein complex60.0119
BPMacromolecule biosynthesis70.0207
CCCyt. large rib. subunit (s. Eu.)30.0226
MFStructural molecule activity50.0339
HS (down)BPCellular physiological process170.000925
(61 genes)BPMorphogenesis120.00138
BPCell differentiation70.00386
BPMorphogenesis of an epithelium50.00438
MFProtein binding80.00532
BPEye morphogenesis50.0111
BPIntracellular signaling cascade40.0168
MFMicrotubule binding30.0309
BPProtein kinase cascade30.0309
BPImaginal disc development50.0357
BPCell migration40.04
DS (262 genes)MFPhosphotransf. act., alcohol group as acceptor110.0126
MFKinase activity110.0259
MFEnzyme regulator activity80.0364
MFProtein phosphatase regulator activity30.05
C30 (24 genes)   
CS (0 genes)   

Direction of gene expression change after selection

The direction of change in the identified genes shows a very consistent pattern. For the majority of the individual selection regime contrasts, the identified genes are all changed in one direction only (see Table 2). A selection response by down-regulation was found for the control vs. DS, LS, C30 and ST (except two genes), whereas the response in the KD lines was by up-regulation. Only heat shock-selected lines had a significant proportion of genes both up- and down-regulated. When comparing the control vs. all selected lines, the response was in one direction with all detected genes up-regulated. Thus, there are strong differences in the way selection affected gene expression in the comparison between controls and all selected lines (up-regulation) compared with the gene expression responses seen in specific selection regimes (down-regulation for most traits). This apparent discrepancy may rely on the following: first, it seems that the genes down-regulated in each selection regime, at least to some degree, are specific to that particular regime (i.e. they are not general stress genes). Second, the power of detection is too low to detect these up-regulated genes in the individual contrasts. This indicates that the down-regulated genes are more changed than the up-regulated ones, but the latter are consistently up-regulated across many selection regimes. Thus, genes that are identified in the all selected contrasts could be considered as general stress genes.

The consistent trend in altered gene expression was unexpected. Only the contrast between UC and HS detected both up- and down-regulation. The remaining selection lines (except KD) showed almost entirely down-regulated genes (Table 2). One possible explanation for the general decreased gene expression in selected lines could be the effect selection had on metabolism. It has been argued that a decrease in general metabolism is beneficial as this conserves energy and allows more resources to be used for increased resistance (see Parsons, 2005), especially for traits like starvation resistance and desiccation resistance, increased longevity and resistance to other types of environmental stress. However, when we considered all selected lines together, the general stress response was exclusively up-regulation, which does not support this hypothesis at the gene expression level. Likewise, the KD lines showed entirely up-regulation when compared with controls. As the only individual selection regime showing this pattern, it suggests that knock down is fundamentally different from all the other stress types. This is supported by the results of Bubliy & Loeschcke (2005) who investigated the phenotypic responses to selection in these lines. In this study, the authors have shown that only the KD lines had increased knock-down resistance and no other selection regimes showed correlated responses in the form of changed resistance because of selection for other stress resistance traits. Adding to the complexity of this, the responses were asymmetrical as the KD lines showed correlated responses for some of the other stress types (Bubliy & Loeschcke, 2005). It is surprising that the general patterns of gene expression changes (i.e. the direction of change) are reflected so clearly at the phenotypic level, which gives promise as to the relevance and robustness of gene expression data.

Overlap of genes detected among selection regimes

The number of significant genes identified in several contrasts supports results from the hierarchical tree and is given in Table 2. DS and ST contrasts have many overlapping genes suggesting some shared mechanisms among these stress types. Also LS, as suggested by the hierarchical tree, leads to change in gene expression in a relatively high number of genes shared with both DS- and ST-selected regimes. A connection between resistance to starvation, desiccation and increased longevity has also been found at the phenotypic level (Hoffmann & Harshman, 1999 and references therein; see Bubliy & Loeschcke, 2005). HS, KD and C30 contrasts only shared few genes with each other and with other regimes suggesting that these different measures of heat stress resistance are rather unrelated and unlikely to share the general mechanisms of tolerance between them. This supports findings at the phenotypic level for the same line studies here (Bubliy & Loeschcke, 2005) and other studies where high resistance for one type of heat resistance seems unrelated to the resistance level of other types (Berrigan, 2000; Sørensen et al., 2003; Nielsen et al., 2005).

Chromosomal position of detected genes

We found the distribution of chromosomal positions of the genes detected for the contrast between control and DS (inline image = 34.4, P < 0.025) to deviate significantly from the expected distribution because of several regions having a higher than expected number of genes, whereas the distribution of genes in the remaining contrasts were all not significant. The significant test results were because of the occurrence of single or few classes with relatively many of the detected genes. This is a rough test of random distribution of significant genes across the chromosomes and should be interpreted with caution. The distribution of genes on autosomes and the X chromosome was also considered. Here two contrasts showed over representation on the X chromosome. These were DS (inline image = 5.4, P < 0.025) and ST (inline image = 4.0, P < 0.05). The general trend of random distribution of genes across the genome indicates that linkage does not affect the chance to identify significantly differentially genes to a large degree. The distribution of genes can also be used to address the issue of distribution in relation to common inversions that might occur in D. melanogaster populations and might affect whole chromosomal areas. The positions identified above with many genes include breakpoints of common inversions; however, in all cases the genes were distributed evenly on each side of the breakpoint and the density of genes was in no cases different inside areas spanning known inversions compared with outside. Thus, in this study inversions seem to play no particular role for the distribution of significantly differently expressed genes.

The biological meaning of the distributions is unknown. As such, the identified positions of peak gene expression change are unexplained and can be considered as areas of particular interest, like a QTL, that can serve to identify candidates for further investigations. No obvious known candidate genes or group of genes are found among the significant genes occurring in the respective chromosomal areas, but some of the areas do correspond to areas identified in QTL studies. For DS the position 86–90 was picked up and this area was also found to be a QTL for heat resistance (Norry et al., 2004; Morgan & Mackay, 2006) and this position includes the inducible Hsp70 genes known to be involved in responses to and important for resistance for many types of stress (Sørensen et al., 2003, 2005). Hsp70 was not found to be differently expressed in this study and the reason for the many genes occurring in this area for particularly heat and desiccation remains unknown.

Gene list overlaps with other studies of gene expression

The genes and functional groups of genes found here were compared with the results of several other array studies addressing stress responses (Table 4). These studies investigated gene expression change induced by exposure to oxidative stress and old age (Girardot et al., 2004; Landis et al., 2004), starvation (Zinke et al., 2002) and heat stress (Sørensen et al., 2005). Furthermore, two studies addressing changes in basal gene expression were included. In these studies flies had been selected for dichlorodiphenyltrichloroethane (DDT) (Pedra et al., 2004) or showed differences in starvation resistance (Harbison et al., 2005) respectively. These and other studies have found at least some degree of overlap between genes induced by different types of stress exposure suggesting some degree of overlap among the responses to different environmental challenges (e.g. Girardot et al., 2004; Kristensen et al., 2005; Sørensen et al., 2005). Here, the expected and observed number of overlapping genes was too low to be tested with any power in several cases. However, when tests could be performed the overlap between genes identified among contrasts in this study and the above-mentioned studies was very small (Table 4). Only five of 27 tests were significant when considering individual selection regimes of this study and only one of these showed an over-representation of observed genes (ST vs. oxidative stress of Girardot et al., 2004). Most notably of the four tests with significantly fewer than expected genes is the lack of overlap between selection for starvation resistance in this study and the study by Harbison et al. (2005). It is possible that differences in experimental procedure, selection history (the flies used by Harbison showed differences in starvation tolerance, but were not selected for this trait) or alternative mechanisms of adaptation are the reason for the lack of similar changes in gene expression in the starvation resistant flies. In any case, it underlines the difficulty in predicting and interpreting overlaps in gene expression studies. It has been suggested that the acute response to stress and the long term adaptation to stress are separate and based on separate mechanisms. This has earlier been suggested to be the case for heat resistance (Sørensen et al., 2003, 2005; Nielsen et al., 2006). In these studies the heat stress response including Hsp expression did not seem to play a major role for evolutionary adaptation to high temperatures. The low overlap between the genes found to respond to heat and other stressors and the genes changed by selection for the same types of stress seems to confirm this hypothesis for heat and to apply for environmental stress in general (Table 4).

Table 4.   Overlap in significant genes detected in this (horizontally) and other stress related gene expression studies (vertically).
Treatment All selec.STLSKDHSDSC30
  1. *P < 0.05; **P < 0.01, ***P < 0.001, ns, nonsignificant.

  2. Treatment represents stress condition used in the experiment and No. # genes represent the number of genes found in each individual experiment. Number of genes found to be differentially expressed in two studies is given in the table. Below in italics the number of genes expected to overlap by chance is given. The expected numbers are calculated as the product of the frequency of significant genes in each experiment times the total number of gene. In cases where the expected number was three or above the deviation from random overlap was tested with a chi-square test. Studies used for comparisons were 1: Landis et al. (2004), 2: Sørensen et al. (2005), 3: Girardot et al. (2004), 4: Zinke et al. (2002), 5: Harbison et al. (2005), 6: Pedra et al. (2004).

  3. (All selec., all selected lines; ST, starvation selected; LS, longevity selected; KD, heat knock down selected; HS, heat survival selected; DS, desiccation selected; C30, constant 30 °C selected; CS, Cold survival selected).

 No. # genes26223064219426224
Old age1 92729** 8ns2ns38ns 6**1
- exp 1715416172
O2 stress1 60737***14ns3ns11ns 6ns1
- exp 1110314111
Heat–early up2 265 2ns 5ns312 4ns1
- exp 5 4102 50
Heat–early down2 50839*** 6ns254ns 2*0
- exp 10 8213101
Heat – late up2 22619*** 1ns010 2ns2
- exp 4 4102 40
O2 stress3136855***34 *8ns19ns17ns5
- exp 2623629262
Starvation4 44926*** 3ns135ns 2*1
- exp 8 7213 81
Select. for starv. resist.5 47735*** 1*231ns 5ns1
- exp 9 8213 91
Select. for DDT resist.6 15821*** 2ns013 0ns0
- exp 3 3101 30

Although little consistent overlap was found for individual selection regimes, it was opposite for the overlap between all selected and the gene expression studies used for comparison. Here, we found a high over-representation of shared genes in all cases of both stress treated and stress resistance selected except for the genes found to be associated with the immediate response to heat stress by Sørensen et al. (2005) (Table 4). This is yet another indication of a general signal of stress adaptation and stress response. It is surprising that the pattern is so consistent among the diverse stress treatments this study is compared with. Nevertheless, it supports the hypothesis that the genes found in the all selected contrast are generally involved in stress adaptation and resistance.

Contrasts of individual selection regimes with controls by Goeman Global test

The results of the Global tests supported the general pattern detected by the hierarchical clustering. The contrast with heat selection showed most differentiation from controls with several significant GO categories followed by desiccation (number of significant GO categories is presented in Table 5 and full list in), starvation and longevity and the contrast with cold the fewest. However, some categories differed between control and cold-selected lines indicating that the resolution of the Global test was higher than that of the SAM analysis. As with the SAM analysis the contrast of controls vs. all selected yielded a number of significant categories suggesting that a general stress response exists that is shared among the selection types investigated here. A number of categories were shared among contrasts (Table 5 in Supplementary Material). However, that pattern was less clear compared with the pattern of shared genes found by the SAM analysis. All significant GO categories are given in Supplemental

Table 5.   Number of specific GO categories shared among contrasts. Total number of GO categories significant at 1% level for each contrast given in bold along the diagonal.
  1. (All selec., all selected lines; ST, starvation selected; LS, longevity selected; KD, heat knock down selected; HS, heat survival selected; DS, desiccation selected; C30, constant 30 °C selected; CS, Cold survival selected).

(A) Biological process
 All selec.191135200
 ST 11124221
 LS  29712921
 KD   159621
 HS    1121081
 DS     2811
 C30      151
 CS       6
(B) Cellular component
 All selec.50020000
 ST 3102000
 LS  1311231
 KD   90010
 HS    21422
 DS     711
 C30      71
 CS       3
(C) Molecular function
 All selec.132021111
 ST 7212211
 LS  1608642
 KD   104231
 HS    501283
 DS     1744
 C30      143
 CS       4

Functional groups of genes affected by selection for stress resistance

Functional groups of genes over-represented were detected by two methods. First, by using the EASE application on the gene lists provided by the SAM analysis, and second, by applying Goeman Global tests that report significant GO categories directly. In Table 2, the EASE results are represented. A summary of the number of significant GO categories and the overlap among contrasts detected by Goeman Global tests is presented in Table 5. The full list of GO categories is given in and influence plots of all significant categories are given in.

The genes and functional groups found to differ between controls and all selected lines showed a remarkable over-representation of genes involved in perception and phototransduction (see Table 2). This is not the first time genes involved in phototransduction have been identified in connection to stress-related gene expression, as this has been found in previous analyses of the heat-selected and control flies used here (Sørensen et al., 2005; Nielsen et al., 2006). Although the connection of phototransduction to stress response and resistance is unknown, one hypothesis is that the sensory system could be used for transmitting and sensing of stress. The results of this study suggest that this does not relate to heat treatment or selection alone but is a general phenomenon among several types of environmental stresses. Other groups of genes affected in the stress-selected group are to a high extent related to transcription and energy production (see Table 2).

For ST-selected lines, GO groups like transcription, glycolysis, spliceosome and metabolism of nucleic molecules are down-regulated. Similar groups were found for LS-selected lines (e.g. transcription, glycolysis, amino acid biosynthesis, mitochondrial ribosomal subunits, spliceosome complex) and DS-selected lines (e.g. transcription and RNA processing). For KD-selected lines spliceosome, transcription, RNA processing and proteasome were significant. It should be remembered that KD lines are the only selected group with increased gene expression. Therefore, their response might be very different even though the same GO categories appear to be significantly differentially expressed. The flies selected for HS showed significant GO categories similar to ST, LS and DS and some additional categories (translation, transcription, DNA repair, ribosome, morphogenesis, spliceosome, lipid metabolism, MAPKKK cascade and proteasome complex). For flies selected by maintaining them at C30 no functional groups of genes were detected among the genes identified by SAM. This is not surprising as only 24 genes were detected which reduces the chance of picking up a larger number of genes belonging to the same functional category. However, the Global test picked up a number of GO categories, including some general metabolic groups and transcription. As indicated by the SAM analysis genes contributing to the difference between controls and flies selected at C30 were generally down-regulated. For the analysis of CS lines no genes were found to differ from the controls and no EASE analysis could be performed. A few functional groups were found by the Global test, primarily related to transcription and generally with the majority of contributing genes with decreased expression. Thus, even though selection has resulted in a marked increase in cold resistance this was not detectable at the gene expression level in this study. There is evidence for the existence and ecological relevance of cold adaptations in temperate D. melanogaster (Gibert et al., 2001; Hoffmann et al., 2003). Thus, it would be expected that some differences were found here; however, it is possible that physiological changes and mechanisms are more important here (e.g. changes in the biochemistry of membranes) that take place down stream of gene regulation.

Overlaps in functional GO categories among contrasts within this study

Some trends of commonly identified processes can be seen from the overview above. Categories related to transcription, translation, energy production, metabolism and biosynthesis were repeatedly found to be differentially expressed between the individual stress-selected lines and controls. Also RNA splicing was clearly affected by selection, which is interesting as this process usually is shut down following exposure to (heat) stress (Yost & Lindquist, 1986). The number of genes detected by SAM and GO categories detected by Global test found to be overlapping among contrasts are summarized in Tables 2 and 5 respectively. These tables show two interesting results. First, a relatively low number of genes overlapping (Table 2) and a relatively higher number of functional GO categories overlapping (Table 5). Second, the relatively high overlap for genes for desiccation, starvation and longevity supporting the hypothesis of shared mechanisms of resistance among these traits. Interesting and surprisingly, the Global test showed a larger overlap in the contrast between heat, starvation and desiccation respectively. The question remains how so relatively many functionally groups of genes are repeatedly detected as changed, when so few genes are found to be overlapping among contrasts. One possibility is the same change in biological function can be reached by changes in different genes belonging to the same functional group, e.g. the turn over of a metabolic process can be increased by up-regulating a few of all involved genes. This way the individual genes are not the same whereas the process is. If this is a general pattern it questions whether it is relevant to expect and look for gene overlaps in these types of studies or whether the functional groups are the more informative unit of interest when identifying the mechanisms of biological functions at the gene expression level. However, incomplete and inaccurate genome annotation might also affect this relation. As the annotation gets better coverage the question could be addressed in more detail.

The functional groups of genes found to be significantly differentially expressed based on the SAM/EASE and Goeman Global tests were to some degree overlapping. That the overlap was not very large is not surprising as the two methods use different approaches for identifying significant groups. Furthermore, because of the subtle differences in the majority of the contrasts only few genes were detected, which decreases the possibility for finding many genes related to the same functional groups. Thus, the functional groups that are detected by both methods seem to be very consistently changed between selection lines and controls. For all contrasts, there are many groups that are too general to interpret; however, these significantly changed functional groups of genes can still serve as a background for understanding the effects of selection and adaptation and for hypothesis generation in future studies of stress responses and adaptation.


Selection for resistance to environmental stresses produced changes that are detectable at the gene expression level. Hierarchical clustering of experiments based on differentially expressed genes, comparisons of individual selection regimes with controls by SAM and Global tests revealed the same patterns, but with different resolution and power. The results presented here have not been validated by RT-PCR or similar methods (Kristensen et al., 2006). The use of replicate arrays and the improvement in statistical handling and technology renders this step of little additional information. A discrepancy between an RT-PCR run using a single primer set and replicates of validated multiple probe arrays may as well be caused by the RT-PCR as by the arrays. Still, results obtained from an RT-PCR validation of Affymetrix arrays show a high correlation (r = 0.93) suggesting that Affymetrix arrays provide reliable results (Park et al., 2004) and might even underestimate the expression change compared with RT-PCR results making the array results a conservative estimate of expression change (Yuen et al., 2002; Park et al., 2004). Furthermore, we do not base our analysis and conclusion on single genes but look at general patterns in gene expression and at groups of functionally related genes. The lack of array replication within lines limits the interpretations of this study to among replicate line effects. Each independent line might have gene expression changes that we cannot detect with the design used.

The responses to individual selection regimes were distinct for each regime, but a group of genes were commonly changed by selection across all types of selection regimes. This suggests that stress specific responses as well as general stress responses contributed to the gene expression patterns observed here. The overlap in the detection of genes among this and other gene expression studies was generally quite low, even in cases where similar selection procedures were performed in both studies. However, in the contrast of all selected with controls the overlap with all compared studies was higher suggesting the existence of a very general response to stress at the gene expression level. The changes brought about by selection for stress resistance are found among genes belonging to general categories of biological processes, with very small changes in many genes and might reflect general adjustments in translation/transcription and metabolism, at least at the gene expression level. This might be interpreted as adjustments to physiology maintaining general homeostasis or homeodynamics under different conditions. The gene expression changes found in Drosophila after heat hardening and in lines selected for thermal stress resistance seems to be rather independent (Sørensen et al., 2005; Nielsen et al., 2006). To understand responses and adaptation to environmental stress, it is, thus, important to investigate both responses and changes after selection. This study provides new insight into evolutionary adaptations and the effects of selection on the transcriptional level and the results can be used to generate and test many hypotheses regarding responses and adaptation to environmental challenges.

The data has been submitted to the Gene Expression Omnibus database with accession number G-SE6558.


We are grateful to Doth Andersen, Mia Skov Jensen and Bente Devantié for excellent technical assistance, to Peter Sørensen for helpful discussion of statistics and to Torsten N. Kristensen and Pawel Michalak for helpful suggestions to the manuscript. We thank Hinnerk Boriss, Just Justesen and Mogens Kruhøffer for discussions during the planning phase of the experiment and to Mogens Kruhøffer for support in running the microarrays. Also thanks are due to many members of the ‘Aarhus stress group’ for giving a hand during peak times of the experimental part of this work. The work was supported by the Danish Natural Sciences Research Council by a Centre grant to VL.