We constructed a heterogeneous base population from isofemale lines sampled from a Raleigh natural population and used artificial selection to create replicate genetically divergent lines with increased resistance (R) or sensitivity (S) to ethanol exposure. We also generated replicate unselected control (C) lines to enable us to monitor the symmetry of the response and genetic drift. Lines had established maximum divergence after 25 generations of selection. At generation 25, the mean elution time (MET) for the replicate control lines (C1 and C2) was MET = 7.4 minutes and MET = 8.8 minutes, respectively; for the replicate sensitive lines (S1 and S2), MET = 2.9 minutes and MET = 2.7 minutes, respectively; and for the replicate resistant lines (R1 and R2), MET = 17.6 minutes and MET = 19.3 minutes, respectively (Figure 1a). Thus, the R and S replicate lines diverged from each other by an average of 15.65 minutes at generation 25. The response to selection was symmetrical. Realized heritability estimates from the divergence between R and S lines over 25 generations were h² = 0.081 ± 0.0097 (P < 0.0001) and h² = 0.069 ± 0.0096 (P < 0.0001) for the respective replicates (Figure 1b). After generation 25 there was almost no response to selection. Realized heritability estimates from the divergence between R and S lines from generation 25 to 35 were h² = -0.056 ± 0.036 (P = 0.1567) and h² = 0.0031 ± 0.027 (P = 0.91) for the respective replicates.

Exposure to alcohol affects locomotion 2223. Furthermore, in human populations excessive alcohol consumption can give rise to aggressive and violent behaviors 343536. Alcohol sensitivity also depends on metabolic and physiological state 3738394041. In addition, exposure to alcohol results in an acute down-regulation of the expression of a group of odorant receptors and odorant binding proteins 19, which raises the question whether artificial selection for alcohol sensitivity would be associated with a reduction in olfactory ability. To assess whether the response to selection was specific for alcohol sensitivity or whether other phenotypes underwent correlated selection, we tested the selection lines for locomotion, aggression, starvation resistance, and olfactory behavior.

We found no differences in locomotor behavior among the selection lines using either an assay for locomotor reactivity (F_2,3 = 3.14, p = 0.18; Figure 2a) or a climbing assay (F_2,3 = 1.48, p = 0.36; Figure 2b). The selection lines also did not differ in the number of aggressive encounters under conditions of competition for limited food (F_2,3 = 3.10, p = 0.19; Figure 2c). Selection lines also did not differ in starvation resistance (F_2,3 = 0.56, p = 0.64; Figure 2d). Finally, there was no correlation between alcohol sensitivity and olfactory avoidance behavior over a range of concentrations of the repellent odorant benzaldehyde (F_2,3 = 0.40, p = 0.70; Figure 2e) (although there were significant differences between replicates of selection lines in avoidance response (F_3,3 = 455.36, p = 0.0002), with line S2 showing reduced olfactory responsiveness). Our results, therefore, indicate that the response to selection was specific for alcohol sensitivity.

Drosophila encounters ethanol in its natural habitat, as flies feed on fermented food sources. Natural selection, at least under some environmental conditions, affects allele frequencies of the Alcohol dehydrogenase (Adh) locus, which is polymorphic for two allozymes, which differ by a single amino acid (T192K), designated Slow and Fast, based on their gel migration profile 4243. Fast homozygotes have a higher level of enzymatic activity than Slow homozygotes and a higher tolerance to alcohol in laboratory toxicity tests 444546.

To assess whether differences in alcohol sensitivity in our selection lines could be attributed in part to the Slow and Fast electrophoretic alleles of Adh 4547, we developed a single nucleotide polymorphism marker for this polymorphism and measured allele frequencies in our selection lines. Frequencies of the Fast allele in the replicate control lines were 0.79 and 0.24. The R1 and R2 replicate lines had Fast allele frequencies of 0.42 and 0.58, respectively. However, in both the sensitive selection lines the Slow allele was fixed. Previous studies have shown that flies homozygous for the Slow Adh allele are more sensitive to alcohol 46.

We used Affymetrix high density oligonucleotide microarrays to assess whole genome transcript abundance in three- to five-day-old flies of the selection lines at generation 25. Raw expression data have been deposited in NCBIs Gene Expression Omnibus 48 and are accessible through GEO series number (GSE 7614).

We used a stepwise procedure to analyze the data. First, we used factorial ANOVA to quantify statistically significant differences in transcript levels for each probe set on the array. Using a stringent false discovery rate 49 of q < 0.001, we found that 9,931 probe sets were significant for the main effect of sex, 2,612 were significant for the main effect of line, and 184 were significant for the line × sex interaction term (Additional data file 1). Only two genes that were significant for the interaction term were not significant for the main effect of line: CG1751, which is involved in proteolysis, and CG12128, which encodes a transcript of unknown function.

Next, we used ANOVA contrast statements on the 2,612 probe sets with differences in transcript abundance between selection lines to detect probe sets that were consistently up- or down-regulated in replicate lines 31. We identified 2,458 probe sets (13% of the total probe sets on the microarray) that differed between the selection lines when pooled across replicates (Additional data file 2).

Among these 2,458 probe sets, 1,572 were divergent between resistant and control lines, 1,617 between sensitive and control lines, and 1,678 between resistant and sensitive selection lines. Although the transcriptional response to selection for alcohol sensitivity was widespread, the magnitudes of the changes in transcript abundance were relatively small, with the vast majority of probe sets showing less than two-fold changes in abundance (Figure 3). In fact, only 121 probe sets showed larger than two-fold differences in transcript abundance. Among these probe sets 37 have not been annotated; 14 encode genes involved in defense response and response to stress, including Defensin, Attacin-A, Lysozyme P, Immune induced molecules 1, 10, and 23, and Metchnikowin; and 12 probe sets that encode gene products involved in carbohydrate metabolism (sugar transporter 1, Mitogen-activated protein kinase phosphatase 3, CG9463, CG14959 CG10725, CG10924, Lysozyme P) (Additional data files 3 and 4).

Probe sets with altered transcript abundance between selection lines fell into all major biological process and molecular function Gene Ontology (GO) categories (Additional data files 5 and 6). We used χ² tests to determine which categories were represented more or less frequently than expected by chance, based on their representation on the microarray. One interpretation of these analyses is that over-represented GO categories contain probe sets for which transcript abundance has responded to artificial selection, whereas under-represented GO categories contain probe sets for which transcript abundance is under stabilizing natural selection 31. Highlights of the transcriptional response to artificial selection for alcohol sensitivity for probe sets differentially expressed between resistant and sensitive selection lines are given in Table 1. For example, the resistant lines are enriched for up-regulated genes affecting responses to chemical stimulus (including response to toxin and pheromone), extracellular transport, and lipid metabolism; while the sensitive lines are enriched for up-regulated genes affecting alcohol metabolism, defense response, electron transport, catabolism, and lipid and carbohydrate metabolism. Transcripts in the 'response to toxin' GO category are over-represented in both sensitive and resistant lines, but the magnitude of over-representation is higher for resistant lines (p = 4.19E-7, compared to p = 0.029 for sensitive lines. GO categories for lipid metabolism are notably over-represented in sensitive lines (p = 1.29E-11, compared to p = 1.2E-04 for resistant lines).

These GO categories correlate well with GO categories that were over-represented during the acute response to a single exposure to ethanol 19, which also resulted in extensive changes in transcript abundance for chemosensory behavior, response to chemical stimulus, and response to toxin.

Changes in expression of transcripts during artificial selection for locomotor reactivity, aggression, and alcohol sensitivity 3233 each encompass a significant percentage of the genome, implying extensive pleiotropy. We found that the transcriptional response to selection for alcohol sensitivity results in changes in expression of over 2,600 probe sets (approximately 14% of the genome) between the selection lines at a stringent false discovery rate of q < 0.001. Similarly, transcript abundance of over 1,800 probe sets evolved as a correlated response to selection for increased and decreased levels of locomotor reactivity 33 and expression of over 1,500 probe sets changed during selection for high and low levels of aggressive behavior 32. Since these studies used the same initial base population, we could assess overlap in transcripts with altered expression between our selection lines and data from previous studies with lines selected for locomotor reactivity and aggression.

We used χ² tests to assess whether we observed more common differentially regulated probe sets than expected by chance. We found 727 probe sets in common between lines selected for alcohol sensitivity and locomotor reactivity, (χ₁² = 883, p << 0.0001); 474 probe sets in common between lines selected for aggressive behavior and locomotor reactivity (χ₁² = 731, p << 0.0001); and 674 probe sets in common between lines selected for alcohol sensitivity and aggressive behavior (χ₁² = 986.1, p << 0.0001). The transcript abundance of 307 genes was altered as a correlated response to selection for all three behaviors (χ₁² = 3928.87, p << 0.0001).

GO categories that were significantly over-represented among these 307 genes include lipid metabolism (p = 2.2E-16), electron transport (p = 1.2E-7), response to chemical stimulus (p = 6.1E-5), carbohydrate metabolism (p = 9.4E-5) and generation of precursor metabolites and energy (p = 8.4E-7). These genes included 17 members of the cytochrome P450 family and additional genes involved in defense response and/or response to toxin (Glutathione S transferases D9, E1 and E5; Immune induced molecule 10, Cbl, UDP-glycosyltransferase 35b, Juvenile hormone epoxide hydrolase 1 and 2, Lysozyme P and Peroxiredoxin 2540; Additional data files 7 and 8). Members of this group of 307 genes appear to represent a common group of environmental response genes.

To validate our premise that transcriptional profiling of artificial selection lines can identify candidate genes that contribute to the trait that responds to selection, we measured alcohol sensitivity of 45 independent P[GT1]-element insertion lines corresponding to 35 candidate genes 5051. These candidate genes are involved in diverse biological processes, including carbohydrate metabolism (Malic enzyme, Poly(ADP-ribose)glycohydrolase, CG9674), regulation of transcription (little imaginal discs, pipsqueak, lilliputian, longitudinals lacking, CG9650), nervous system development (Beadex, Laminin A, longitudinals lacking, muscleblind, smell impaired 35A), lipid metabolism (retinal degeneration B, sugarless, CG17646) and signal transduction (βν integrin, Laminin A, sugarless, wing blister, CG32560). Five of the candidate genes encode predicted transcripts of unknown function (lamina ancestor, CG11133, CG30015, CG14591 and CG6175). Overall, 33 (73%) of the P[GT1]-element insertion lines exhibited significant differences in alcohol sensitivity compared to co-isogenic Canton S (B) control at p < 0.05, and for 19 of these lines (58%) statistically significant differences from the control survived Bonferroni correction for multiple tests (Table 2, Figure 4). Remarkably, P-element insertions implicate 32 out of 35 genes in alcohol sensitivity. P-element mutants in Beadex, corto, Glutamate oxaloacetate transaminase 1, Kinesin-73, Laminin A, lethal (1) G0007, little imaginal discs, longitudinals lacking, Poly(ADP-ribose) glycohydrolase, Malic enzyme, muscleblind, nuclear fallout, retinal degeneration B, sugarless, visgun, wing blister, CG6175, CG14591, CG7832, CG17646, CG5946 and CG30015 were more resistant to ethanol exposure than the control. In contrast, mutants for βν integrin, lamina ancestor, Lipid storage droplet-2, pipsqueak, Toll, CG9650, CG32560, CG12505 and CG9674 were more sensitive to ethanol exposure than the control. Three of these P-element insertion lines with transposon insertions at Malic enzyme, nuclear fallout and longitudinals lacking were previously implicated in alcohol sensitivity and/or tolerance in Drosophila 19.

Our results demonstrate that transcriptional profiling of artificial selection lines is a powerful strategy for identifying genes that contribute to the selected trait, in our case sensitivity to alcohol.

We observed differences in Fast and Slow Adh allele frequencies between sensitive and resistant lines. However, the probe set for the Adh gene was not differentially expressed between the selection lines. This is perhaps not surprising, as previous studies showed that there is no correlation between ethanol tolerance and ADH activity in lines homozygous for the Fast and Slow Adh alleles 52 and the increase in tolerance to ethanol in adult flies was not accompanied by an increase in overall ADH activity 425354.

Transcriptional profiling studies showed that a large fraction of the genome undergoes altered transcriptional regulation in response to artificial selection, in line with previous selection studies on locomotion, aggression and starvation resistance 323355. The magnitudes of changes in transcript abundance, although significant at q < 0.001, were generally modest. Small (1.3- to 1.4-fold) changes in transcript abundance in response to ethanol exposure have also been reported for other animal models 17. Similarly, changes in gene expression of as little as 1.4-fold have been detected reproducibly by expression microarray analysis in the brains of human alcoholics 6.

Previously, we observed changes in transcript levels for 582 probe sets after isogenic Canton S B flies were exposed to ethanol in an inebriometer 19. The expression of 195 of these probe sets was also altered between our artificial selection lines (χ₁² = 152.1, p < 0.0001), including Adh transcription factor 1, Adenylyl cyclase 35C, 6 cytochrome P450 family members, Glutathione S transferases D5, E4, E5 and E7, Heat-shock-protein-70, Malic enzyme, Neural Lazarillo, Pheromone-binding protein-related protein 1, 2 and 5, Phosphoenolpyruvate carboxykinase, Pyruvate dehydrogenase kinase, and UDP-glycosyltransferase 35b (Additional data file 9). One likely reason that we did not detect more of the 582 probe sets previously identified is the difference in genetic background between the two studies (isogenic Canton S B versus lines derived from a genetically heterogeneous natural base population).

Regardless of whether or not the observed changes in gene expression are causally associated with genetic divergence in alcohol sensitivity between the selection lines, the genes exhibiting altered expression levels are candidate genes affecting alcohol sensitivity. We measured the response to ethanol exposure for 45 mutations in candidate genes that were generated in a common co-isogenic Canton S B background, and identified 32 genes with mutational effects on alcohol sensitivity. Three of these genes, Malic enzyme, nuclear fallout and longitudinals lacking, have been previously implicated in alcohol sensitivity and/or tolerance 19 and 23 of them have human orthologues, many of which have been implicated in diseases (Table 2).

The high success rate (73%) of these functional tests supports the hypothesis that expression profiling of genetically divergent lines can identify candidate genes that affect complex traits in Drosophila and that comparative genomic approaches can infer human candidate genes from their Drosophila orthologues. However, we could not detect genes that are differentially expressed at different developmental times. Similarly, genes affecting the trait that are not regulated at the level of transcription, but may be regulated through posttranslational modifications, will also not be detected by our transcriptional profiling approach.

We determined how many genes that have been already implicated in alcohol sensitivity and/or tolerance in Drosophila are significantly differentially expressed between selection lines, and found 38 genes previously implicated in responses to alcohol or alcohol-related metabolism (Additional data file 10). The probe set for Aldehyde oxidase 1 5657 was not present on the array. Probe sets for the cheapdate allele of amnesiac 26, the dopamine D1 receptor 58 and neuropeptide F 59 had absent calls, possibly due to low expression levels, and were consequently not included in the analysis. Of the 34 remaining genes, 10 (approximately 30%) showed altered transcript abundance between our selection lines at q < 0.001, including: Adh transcriptional factor 1 60; Acetaldehyde dehydrogenase 61; Aldolase 62; fasciclin II, which is required for the formation of odor memories and for normal sensitivity to alcohol in flies 25; Formaldehyde dehydrogenase 565763; geko 64; Glycerol 3 phosphate dehydrogenase 566263; and the cell adhesion receptor slowpoke, which encodes a large-conductance calcium-activated potassium channel 6566.

For 14 previously implicated genes (approximately 40%) the magnitude of the differences in expression after selection for alcohol sensitivity and resistance were not great enough to satisfy our stringent false discovery rate threshold of q < 0.001 even if the p value was < 0.05. Such genes include dunce (q = 0.07, p = 0.03), which encodes a cAMP-phosphodiesterase 2667; GABA receptors (Rdl and Lcch36869); lush (q = 0.0031, p = 0.009), which encodes an odorant binding protein that interacts with short chain alcohols 70; the gene that encodes the neuropeptide F receptor (q = 0.018, p = 0.005) 59; period (q = 0.007, p = 0.03), a regulator of circadian activity that has been associated with alcohol consumption in mice and humans 28; Pka-R1 (q = 0.002, p = 0.02) and Pka-C1 (q = 0.002, p = 0.005), which encode a cyclic AMP-dependent protein kinase 2771; the calcium/calmodulin-dependent adenylate cyclase encoded by the rutabaga gene (q = 0.008, p = 0.03 26); and sluggish A, a glutamate biosynthesis enzyme 72.

Expression of only nine alcohol sensitive genes was not significant on our microarray, including: the gene encoding tyramine β-hydroxylase (p = 0.23), an enzyme required for the synthesis of octopamine 2471; the gene encoding GABA-B receptor-1 (p = 0.53) 73; hangover (p = 0.52), which encodes a nucleic acid binding zinc finger protein and has been implicated in both the response to heat stress and the induction of ethanol tolerance 24; and homer (p = 0.18), which is required for behavioral plasticity 74 - mutant flies exhibit both increased sensitivity to the sedative effects of ethanol and failure to develop normal levels of rapid tolerance 75. Taken together, around 70% of already implicated genes in alcohol sensitivity were found to be differentially expressed on our microarray.

Other notable probe sets with altered transcriptional regulation include Sorbitol dehydrogenase 2, CG3523, CG16935 and v(2)k05816, all of which encode products with alcohol dehydrogenase activity. A previous study reported that mutants in white rabbit (p = 0.23), which encodes RhoGAP18B (q = 0.009, p = 0.04), are resistant to the sedating effects of ethanol 29. In our study six probe sets that encode RhoGap gene family members (RhoGAP19D, 54D, 16F, 100F and 71E) showed changes in expression levels in response to ethanol selection (q < 0.001). Furthermore, 22 of the genes with changes in transcript levels on our microarrays corresponded to genes differentially expressed in the frontal cortex 9 and 10 genes from prefrontal cortex and nucleus accumbens of alcoholics 7 (Additional data file 11).

In addition to human orthologues associated with alcoholism, 246 genes with altered transcript abundance on our microarrays correspond to murine orthologues implicated in altered transcriptional regulation in a meta-analysis study of alcohol drinking preference in mice 3. Acetyl Coenzyme A synthase, Aldh, Beadex, CG16935, Fdh, Laminin B2, lethal (2) essential for life and lethal(1)G0007 were among those genes (Additional data file 12). In addition, several Drosophila transcripts that are differentially expressed in response to artificial selection have murine orthologues associated with alcohol related phenotypes (Additional data file 12), including Aldehyde dehydrogenase family 6 member, which maps to a region on 14q24.23 implicated in alcoholism 76, Carnitine palmitoyltranferse 1, Cathepsin B, Distal-less homeobox 1, Glutamate oxaloacetate transaminase 2, Dorsal switch protein 1 and synapsin 177778.

Flies can readily be grown in large numbers in defined genetic backgrounds under controlled environmental conditions and alcohol sensitivity can be quantified precisely. Our results consolidate the notion that Drosophila melanogaster can serve as a gene discovery tool for candidate genes that predispose to alcohol related phenotypes in the human population, and demonstrate the power of transcriptional profiling of selection lines derived from a common base population as a complementary approach for identifying candidate genes for complex traits.

Flies were reared on cornmeal/molasses/agar medium under standard culture conditions (25°C, 12:12 hour light/dark cycle). Behavioral assays were conducted in a behavioral chamber (25°C, 70% humidity) between 9 am and 12 am, with the exception of the alcohol selection assays, which were done between 9 am and 2 pm at room temperature. Flies were not exposed to CO₂ anesthesia for at least 24 hours prior to the assay. Homozygous P-element insertion lines containing P[GT1]-elements in or near candidate genes in the co-isogenic Canton S B background were generated by Dr Hugo Bellen (Baylor College of Medicine, Houston, TX, USA) as part of the Berkeley Drosophila Genome Project 50.

To quantify alcohol sensitivity, we placed flies (N = 60-70) in an inebriometer pre-equilibrated with ethanol vapor from which they were eluted at one minute intervals. The MET is a measure of alcohol sensitivity 79.

The base population was generated from 60 isofemale lines established from flies collected in Raleigh, NC in 1999. The isofemale lines were crossed in a round robin design (line 1 ♀ × line 2 ♂, line 2 ♀ × line 3 ♂, …line 60 ♀ × line 1 ♂). Single fertilized females from each cross were placed in each of two culture bottles. In the following generation (G0), the alcohol sensitivity of 60 males and 60 virgin females of each replicate was scored using the alcohol sensitivity assay. The 20 most resistant flies (males and females) from each replicate were placed in bottles to initiate the two resistant lines (R1, R2); and the 20 sensitive flies from each replicate initiated the two sensitive lines (S1, S2). The two control lines were initiated with the remaining 20 flies (C1, C2). In the following (G1) and all subsequent generations, the same procedure was repeated: 60 males and females, separately, from each line (resistant, sensitive, and control) were scored, and the 20 highest-scoring flies from the resistant lines and the 20 lowest-scoring flies from the sensitive lines were selected as parents for the next generation. Control line flies were scored each generation and 20 random flies were used as parents.

Estimates of realized heritability (h²) were calculated by regression of the cumulative selection response (ΣR) on the cumulative selection differential (ΣS) 80.

To assess the specificity of the selection response, we tested our selected lines for a battery of other traits: locomotor, aggressive, and olfactory behavior, and starvation resistance.

Locomotor behavior was assessed using two different assays. Locomotor reactivity was assessed as described previously 33. A single three- to five-day-old fly was placed in a vial with approximately 3 ml standard medium, and subjected to gentle mechanical disturbance by tapping on the bench top. The vial was placed horizontally, and the total amount of time the fly remained mobile for a 45 second period immediately following the disturbance was the locomotor reactivity score of the individual. This assay was performed at generation 35, with 20 replicate measurements per line per sex. In the second climbing assay, individual flies were transferred without anesthesia into an empty glass vial, with the height of the vial demarcated in 5 mm intervals from 0 to 27. The fly was tapped to the bottom of the vial, which was then placed vertically. The climbing score was the maximum height reached within the eight second observation period. Twenty replicates per line per sex were tested at generation 36.

Aggressive behavior was assessed as previously described 32. Aggression of single individuals was quantified by placing one experimental male, with wild-type eye color, with three reference white-eyed isogenic w¹¹¹⁸ Canton-S males. The flies were placed in a vial without food for 90 minutes, after which they were transferred (without anesthesia) to a test arena containing a droplet of food and allowed to acclimate for two minutes. After the acclimation period, the flies were observed for two minutes. The following behaviors were scored as aggressive encounters: kicking, chasing, wing-raising and boxing 81. The score of the experimental fly was the number of encounters in which it exhibited an aggressive behavior, including interactions initiated by the experimental fly and those in which he responded aggressively to a reference fly. Thirty replicates per line per sex were tested at generation 36.

The olfactory behavior assay was performed as described previously 82. The flies were placed in a vial without food for 60 minutes and after that were screened by quantifying olfactory avoidance behavior. Five flies per replicate with 15 replicates per sex per line at a concentration of 0.1% (v/v), 0.3% (v/v) and 1% (v/v) benzaldehyde were tested between 9.00 am and 1.00 pm, in a randomized design, in which measurements on individual lines were collected over multiple days to average environmental variation.

Starvation resistance was assessed as previously described 55. Single sex groups of ten two-day-old flies were placed in vials containing non-nutritive media (1.5% agar and 5 ml water). Survival was scored every eight hours. This assay was conducted at generation 37, with five replicate measurements per line per sex.

Differences between the selection lines for the correlated traits were assessed using a nested mixed model analysis of variance (ANOVA):

Y = μ + Selection + Line (Selection) + Sex + Selection × Sex + Line (Selection) × Sex + ε

where Y is the phenotypic score, μ is the overall mean, Selection is the fixed effect of the selection treatment (resistant, control, or sensitive), Line (Selection) is the random effect of the replicate within each selection group, Sex is the fixed effect of sex, and ε is the error variance. The terms of most interest in the model are Selection and Line (Selection). A significant Selection term is indicative of a correlated response in the trait being tested to selection for alcohol sensitivity. The Line (Selection) term reveals whether replicate lines responded similarly or divergently, allowing an assessment of the effects of random genetic drift within a replicate line.

Genomic DNA was extracted from 20 adult males and 20 adult females individually of each selection line using Puregene DNA purification system (Gentra systems, Minneapolis, MN, USA). PCR amplification was performed on 100 ng of DNA from each line. The primers were Adh-forward 5'CAACATTGGATCCGTCACTG' (1,355 bp) and Adh-reverse 5'GCTCAACATCCAACCAGGAG' (1,623 bp). The basic PCR conditions were 35 cycles of denaturation at 94°C for 30 s, annealing at 56°C for 30 s and extension at 72°C for 30 s with 1 unit of RedTaq DNA polymerase (Sigma-Aldrich, Carlsband, CA, USA). The 269 bp product was then digested for 3 hours using HpyCH4 IV enzyme (New England BioLabs, Iswich, MA, USA), according to the supplier's instructions. The A-C polymorphism at position 1,490 bp is responsible for the Lys - Thr substitution between Fast and Slow Adh alleles 47. The HpyCH4 IV enzyme recognized the A/CGT nucleotide sequence corresponding to the Slow allele, yielding 240 bp and 29 bp restriction fragments that could be separated by electrophoresis from the 269 bp fragment of the Fast allele.

At generation 25, two replicates of 15 three- to five-day-old virgin males and females were collected from each selection line. Total RNA was extracted from the 24 samples (six lines × two sexes × two replicates) using the Trizol reagent (Gibco BRL, Gaithersburg, MD, USA). Biotinylated cRNA probes were hybridized to high density oligonucleotide microarrays (Affymetrix, Inc. Drosophila GeneChip 2.0) and visualized with a streptavidin-phycoerythrin conjugate, as described in the Affymetrix GeneChip Expression Analysis Technical Manual (2000), using internal references for quantification. The quantitative estimate of expression of each probe set is the Signal (Sig) metric, as described in the Affymetrix Microarray Suite, Version 5.0.

The 18,800 probe sets on the Affymetrix Drosophila GeneChip 2.0 are represented by 14 perfect-match (PM) and 14 mismatch (MM) pairs. The Sig metric is computed using the weighted log(PM-MM) intensity for each probe set, and was scaled to a median intensity of 500. A detection call of Present, Absent, or Marginal is also reported for each probe set. We excluded probe sets with more than half of the samples called 'Absent' from the analysis, leaving 11,838 probe sets. This filter retained sex-specific transcripts but eliminated probe sets with very low and/or variable expression levels 31. On the remaining probe sets, we conducted two-way fixed effect ANOVAs of the Signal metric, using the following model:

Y = μ + Line + Sex + Line × Sex + ε

where Sex and Line are the fixed effects of sex and selection line and ε is the variance between replicate arrays. We corrected the p values computed in these ANOVAs for multiple tests using a stringent false discovery rate criterion of q < 0.001. We used contrast statements 31 to assess whether expression levels of probe sets with L and/or S × L terms at or below the q = 0.001 threshold were significantly different between selection groups (resistant, control, and sensitive) at the p < 0.05 level, both within each sex and pooled across sexes. All statistical analyses were performed using SAS procedures 2983. GO categories were annotated using Affymetrix 84 and FlyBase 85 compilations.

We tested whether 37 mutations in 35 of the candidate genes with altered transcript abundance between the selection lines affected alcohol sensitivity. The mutations were homozygous P[GT1] elements inserted within the candidate genes, and all were generated in a common co-isogenic background (Canton S, B background) 50. Alcohol sensitivity was assessed for all mutant lines (4-5 replicates/line, N = 60-70, 3- to 5-day-old males/replicate) using an inebriometer 79. We used analysis of variance to assess whether the sensitivity of P-element insertion lines differed significantly from the control, according to the model:

Y = μ + Line + Replicate (Line) + ε

where μ is the overall mean, Line is the fixed effect of line (P-element insertion versus Control), Replicate is the random effect of replicate, nested within line, and ε is the variance within replicates. A significant Line term suggests that the mutant is significantly different from the control.