Modulation of the Drosophila transcriptome by developmental exposure to alcohol

Prenatal exposure to ethanol can cause fetal alcohol spectrum disorder (FASD), a prevalent, preventable pediatric disorder. Identifying genetic risk alleles for FASD is challenging since time, dose, and frequency of exposure are often unknown, and manifestations of FASD are diverse and evident long after exposure. Drosophila melanogaster is an excellent model to study the genetic basis of the effects of developmental alcohol exposure since many individuals of the same genotype can be reared under controlled environmental conditions. We used 96 sequenced, wild-derived inbred lines from the Drosophila melanogaster Genetic Reference Panel (DGRP) to profile genome-wide transcript abundances in young adult flies that developed on ethanol-supplemented medium or standard culture medium. We found substantial genetic variation in gene expression in response to ethanol with extensive sexual dimorphism. We constructed sex-specific genetic networks associated with alcohol-dependent modulation of gene expression that include protein-coding genes, Novel Transcribed Regions (NTRs, postulated to encode long non-coding RNAs) and female-specific coordinated regulation of snoRNAs that regulate pseudouridylation of ribosomal RNA. We reared DGRP lines which showed extreme upregulation or downregulation of snoRNA expression during developmental alcohol exposure on standard or ethanol supplemented medium and demonstrated that developmental exposure to ethanol has genotype-specific effects on adult locomotor activity and sleep. There is significant and sex-specific natural genetic variation in the transcriptional response to developmental exposure to ethanol in Drosophila that comprises networks of genes affecting nervous system development and ethanol metabolism as well as networks of regulatory non-coding RNAs.

exposing their developing fetus to alcohol [9]. Although defects from prenatal alcohol exposure can be replicated in mouse models [10], identifying genetic factors that contribute to susceptibility to FASD is virtually impossible in human populations since time, dose, and frequency of exposure are generally unknown, and manifestations of FASD are diverse and become evident long after exposure.
Drosophila melanogaster presents an advantageous model for studies on the genetic underpinnings associated with symptoms of developmental alcohol exposure. The Drosophila model allows strict control over the genetic background. In addition, individuals of the same genotype can be reared in large numbers under controlled environmental conditions, without regulatory restrictions and at low cost. Following acute exposure to alcohol, flies undergo loss of postural control, sedation, and development of tolerance, physiological and behavioral changes that resemble human alcohol intoxication [11][12][13][14].
Previous studies on the effects of developmental alcohol exposure in Drosophila showed reduced viability and delayed development time [15,16], reduced adult body size [15], and disruption of neural development [17]. Developmental exposure to alcohol was associated with reduction in the expression of a subset of insulinlike peptides and the insulin receptor [15], dysregulation of lipid metabolism and concomitant increased oxidative stress [18] and reduced larval food intake due to altered neuropeptide F signaling [19].
In previous studies, we have taken advantage of natural variation in the Drosophila melanogaster Genetic Reference Panel (DGRP), a well characterized population of 205 inbred wild-derived lines with complete genome sequences [20,21], to study the genetic underpinnings of developmental alcohol exposure [16], voluntary ethanol consumption [22], acute ethanol intoxication, and induction of tolerance [23,24], a prelude to the development of alcohol dependence in people. Linkage disequilibrium in the DGRP decays within a few hundred base pairs [21], which enables identification of candidate causal single nucleotide polymorphisms (SNPs) associated with phenotypic variation in alcohol-related traits. A major unresolved question relates to the mechanism(s) by which alcohol exposure during development affects adult phenotypes. Here, we performed RNA sequencing to assess the effects of developmental alcohol exposure on genome wide genetic variation in gene expression of young adults. Evolutionary conservation of fundamental biological processes and superposition of human orthologs on transcriptional networks identified in flies provides translational potential for studies in the Drosophila model.

Transcriptional profiles of flies reared on ethanol supplemented medium
We performed transcriptional profiling of flies reared on regular medium or medium supplemented with 10% (v/v) ethanol from 96 DGRP lines that span the phenotypic spectrum of alcohol sensitivity following developmental exposure to alcohol (Additional File 1) [16]. For each line, we obtained duplicate samples of 30 males and 25 females, aged 3-5 days, all collected at the same time of day, and performed RNA sequencing using 125 bp single end reads. A total of 33,580 transcripts were expressed in adult flies, of which 16,165 (48.1%) are annotated and 17,415 (51.9%) are novel transcribed regions (NTRs) [25,26] (Additional file 2). We performed three-way factorial mixed model ANOVAs with the main effects of DGRP line, sex, and treatment for all expressed transcripts, and used a False Discovery Rate (FDR) threshold < 0.05 for statistical significance of each term in the ANOVA model (Additional File 2). As in previous analyses of genome-wide gene expression using whole flies [25][26][27], we find that expression of nearly all expressed transcripts is sexually dimorphic (28,343,84.4%). The expression of 16,278 transcripts (48.5%) is modulated by alcohol, and for 10,002 transcripts the transcriptional response to ethanol differs between males and females (i.e., there is a treatment by sex interaction). There is significant genetic variation in expression for 10,620 transcripts, as well as context-specific genetic variation, with 11,338 transcripts exhibiting genetic variation in sexual dimorphism (sex by line interaction), 1,222 showing genetic variation in response to developmental exposure to ethanol (treatment by line interaction) and 77 with genetic variation in sexual dimorphism in response to ethanol (sex by treatment by line interaction).
Because we found extensive interactions with sex and treatment and sex and DGRP line, we performed two-way ANOVAs partitioning gene expression variation by line, treatment, and the line by treatment interaction (L × T), separately for males and females (Additional File 3). The main effect of treatment was significant (FDR < 0.05) for 14,158 (13,827) transcripts in females (males), and the main effect of line was significant for 13,521 (20,996) transcripts in females (males). We are most interested in transcripts with a significant genotype by alcohol treatment interaction (L × T) since these transcripts show genetic variation in their response to developmental ethanol exposure. When we compared expression profiles of flies grown on ethanol supplemented medium with transcript abundance levels obtained under standard growth conditions, we found 939 significant L × T interactions in females and 823 in males (Additional File 4). Of these 1,351 transcripts that have genetically variable responses to ethanol exposure during development, 253 are in common between males and females, 499 are female-specific and 346 are male-specific. The transcripts with significant L × T terms in females were enriched for biological process gene ontology terms involved in metabolism, biosynthesis, transcription, immune/defense response, and chromatin organization; and the transcripts with significant L × T terms in males were enriched for biological process gene ontology terms involved in immune/defense response, metabolism, and development [28] (Additional File 4). A total of 65.6% of these genes have human orthologs (Additional File 4).

Co-regulated modules of transcripts that are differentially regulated after developmental alcohol exposure
We analyzed the sex-specific correlation structure of the developmental alcohol-sensitive transcriptome and identified eight highly interconnected modules in females (Fig. 1). These modules contained transcripts associated with xenobiotic detoxification and metabolism (Figs. 1 A, B), development and cell adhesion (Figs. 1B, C), cuticle formation (Fig. 1E) and neural signaling (Fig. 1G). The latter includes sNPF, which is associated with neuropeptide F signaling and has been previously implicated with reduced larval food intake when larvae were grown on alcohol-supplemented medium [19]. One module consists almost exclusively of NTRs (designated XLOCs; Fig. 2D) [25,26]. Of special interest is Ilp3, which encodes an insulin-like peptide implicated in developmental exposure to alcohol [15] and is correlated with a network of NTRs (Fig. 1F). The most highly correlated group of transcripts are the small nucleolar RNAs (snoRNAs; Fig. 1H).
Although there was overlap between alcohol-modulated transcripts in males and females, the correlation structure for differentially expressed genes in males shows an entirely different modular organization (Fig. 2). The nine male modules are generally smaller and more difficult to interpret than the female modules, as they are dominated by NTRs and genes of unknown function. Whereas the functions of this new class of noncoding transcripts remain to be established, they feature prominently in networks of alcohol-modulated transcripts ( Figs. 1 and 2). This suggests a regulatory role for non-coding elements in the genome in modulating the transcriptional response to developmental alcohol exposure.
To identify genetic associations within modules of differentially expressed genes, we mapped expression quantitative trait loci associated with the difference in expression between standard and ethanol-supplemented medium, or response expression-Quantitative Trait Loci (e-QTLs; [25,26]). We identified 53 eQTLs, including 19 NTRs, in females, and 45 eQTLs, including 11 NTRs and four long-noncoding RNAs (lncRNAs), in males (Additional File 5). All eQTLs, with the exception of the eQTL associated with the difference in NetA expression between standard and ethanol-supplemented medium in females, were in trans (greater than 1 kb from the start and end of the gene body) to the genes associated with an LxT interaction). Therefore, we determined to what genes these eQTLs were in cis (defined by within 1 kb of a gene body) (Additional File 5). We then input the genes with LxT interactions and the genes to which the eQTLs are cis into known genetic and protein-protein or RNA-protein physical interaction networks to construct sex-specific networks associated with genetic variation in response to developmental exposure to ethanol (Fig. 3). The resulting networks are composed of integrative modules that highlight considerable overlap between cellular processes in males and females despite significant sexual dimorphism in ethanol-induced differential gene expression. Genes associated with neuronal development and differentiation, response to ethanol, and reproduction are evident in networks of both sexes. The female network also incorporates modules associated with glutathione metabolism and phototransduction (Fig. 3A), whereas the male network contains modules associated with immune response, starvation and stress response, and septate junction assembly (Fig. 3B).

Coordinated sex-specific modulation of an ensemble of snoRNAs
Altered co-regulation of 38 snoRNAs was observed only in females exposed to alcohol during development. The direction of changes in snoRNA expression was strongly genetic background-dependent, with some DGRP lines showing coordinated up-regulation, some exhibiting coordinated down-regulation, and others showing no change in snoRNA expression ( Fig. 4; Additional File 6). The snoRNAs that exhibit genetic variation in their response to developmental ethanol exposure belong primarily to the H/ACA class, which are associated with pseudouridylation of ribosomal RNAs [29,30]; many are in introns of RpS4, RpL5, RpS5a, RpS7, RpL11, RpS16, RpL17, and RpL22 ribosomal protein encoding genes. The number of snoRNAs within each gene varies from 1 to 15 (Additional File 6). When multiple snoRNAs are present in a gene, only some show altered expression in response to ethanol exposure, although in some cases clusters of genes, likely expressed as polycistronic transcripts, are regulated together. Other snoRNAs with altered expression in response to chronic ethanol exposure are in introns in dom, which contributes to histone acetyl transferase activity associated with epigenetic modification of gene expression [31]; CG13900 (Sf3b3), inferred to form part of a spliceosome complex [32]; SC35, which encodes a splicing factor [33]; kra, which is annotated as a translation initiation factor binding protein [34]; Aladin, predicted to form part of the nucleopore complex [35]; and Nop60B, which encodes pseudouridine synthase [36,37]. Examination of variation in transcript abundances of snoRNAs and host genes other than those encoding ribosomal proteins show that the expression of these snoRNAs is regulated independently from the host genes (Fig. S1). Interestingly, transcript abundance levels of all alcohol-sensitive snoRNAs were correlated with variation in expression of Uhg4, which is highly expressed in ovaries and encodes a lncRNA (Fig. S1) [38]. A previous genome-wide association study in the DGRP identified Cyclin E (CycE) as a highly interconnected hub gene in a genetic network associated with alcohol-induced variation in viability and development time [16]. Variation in CycE transcript abundance was not correlated with variation in transcript abundance levels of snoRNAs but was highly correlated with variation in transcript abundance levels of their host genes (Fig. S2). Expression levels of stet, dom, Aladin, kra and Nop60B were also highly correlated (Fig. S2).

Effects of developmental alcohol exposure on activity and sleep
The transcriptome is a proximal determinant of organismal phenotypes. FAS/FASD symptoms include hyperactivity and sleep disorders. Since activity and sleep are universal measures of nervous system function which can be modeled in Drosophila, we used activity and sleep parameters as a read-out of the behavioral effects of alcohol exposure during development (Fig. 5). We selected three DGRP lines which exhibited the highest degree of coordinated up-regulation (DGRP_177, DGRP_208, DGRP_367) and down-regulation (DGRP_555, DGRP_705, DGRP_730) of a subset of snoRNAs in response to developmental alcohol exposure and reared them either on regular food or food supplemented with ethanol. We measured their activity and sleep phenotypes using the Drosophila Activity Monitor (DAM) system, in which single flies are introduced into narrow tubes and a movement is recorded any time the fly disrupts an infrared beam. We performed factorial mixed model ANOVA analyses with the main effects of sex, treatment and DGRP line for activity, night and day sleep proportion, and night and day bout count. There was significant genetic variation for all traits among the DGRP lines, and the L × T interaction was significant for activity, night bout count, and the proportion of day and night sleep (Additional File 7). Thus, exposing Drosophila to alcohol during development affects activity and sleep phenotypes relevant to patients with FASD, but the effects are dependent on genetic context.

Discussion
Although previous studies have identified a plethora of genetic risk factors that contribute to alcohol related phenotypes in human populations or rodent models [11,39], our understanding of the interaction between environmental alcohol exposure and allelic variants remains incomplete. Evolutionary conservation of fundamental biological processes and similarity of the effects of alcohol exposure between flies and people have established D. melanogaster as a useful translational gene discovery system [11,23,40]. Here, we identified transcripts that undergo altered regulation when flies are reared on ethanol. Changes in transcript abundance patterns are sexually dimorphic and reveal regulatory networks in which NTRs feature prominently. It is of special interest that females exposed to developmental ethanol exposure show altered regulation of a large ensemble of H/ACA class snoRNA genes, which may mediate widespread changes in protein synthesis upon chronic alcohol exposure. The complex relationship between expression of host genes and embedded snoRNAs, which we observe in our studies, has also been documented across seven human tissues, including brain [41]. The mechanisms by which alcohol triggers changes in gene expression remain unknown and could include direct effects on the genome, indirect effects on the genome mediated via alcoholinduced metabolic changes, and/or epigenetic modifications of DNA. Pseudouridylation of mRNAs, tRNAs and other small RNAs in response to environmental stress can consolidate or destabilize interactions between RNAs and proteins [42]. The intimate relationship between snoRNAs and ribosomal function may represent a conduit between the genome and the proteome that can adaptively modulate the composition of the proteome in response to ethanol exposure. The D. melanogaster transcriptome is highly intercorrelated [25] and changes in gene expression due to an environmental disturbance result in modulation of transcriptional niches (i.e., coregulated ensembles) of focal genes [40,43]. This raises a central "cause versus effect" question as it is not evident a priori which gene is the focal gene that directly responds to the environmental perturbation. Insights can be derived from eQTL analyses that can delineate cis-eQTLs and trans-eQTLs [25,26]. Interestingly, all but one of the eQTLs associated with transcripts with genetic variation in the difference in expression between standard rearing conditions and developmental exposure to ethanol were in trans to the focal transcripts. However, we were able to incorporate the genes to which these eQTLs are cis into known interaction networks to derive sex-specific networks associated with genetic variation in response to developmental exposure to ethanol in which these genes are candidate regulatory drivers.
The data we present were obtained from flies that were continuously exposed to alcohol from egg to adult. Exposures that are restricted to different developmental stages will provide a finer grained picture of the dynamics of the alcohol-sensitive genome. Similarly, we analyzed transcriptional responses in whole flies. Single cell RNA sequencing experiments with defined tissues, such as the brain, can provide tissue-specific resolution of the transcriptional response to developmental alcohol exposure [44]. However, the data we obtained in this study underscore extensive sexual dimorphism and emphasize the importance of non-coding elements in regulating the transcriptional response to alcohol exposure during development. Not all aspects of FASD (e.g. cognitive impairment) can be readily modeled in flies. Nevertheless, results from this study illustrate the power of the Drosophila model as a gene discovery system to gain insights into human disorders, such as FASD, that can only be addressed through comparative genomics approaches.

Conclusion
Because of conservation of fundamental biological processes and homologies between fly gene products and their human counterparts, studies on the transcriptional response to developmental exposure to alcohol in Drosophila melanogaster can provide insights in the genetic underpinnings that may predispose to FASD.

Drosophila lines
The DGRP lines have been generated and are maintained in our laboratories. We selected 96 DGRP lines [20,21] across the range of phenotypic variation of effects of alcohol exposure on viability and developmental time (Additional File 1) [16,24] and reared them on cornmealmolasses-agar medium supplemented with10% (v/v) ethanol at 25 °C, 60-75% relative humidity and a 12-h light-dark cycle at equal larval densities. We collected two replicates of mated 3-5-day old flies (25 females and 30 males per line) for a total of 384 samples, following procedures described previously for baseline sample collection [26]. We used a randomized experimental design for sample collection that was done strictly between 1-3 pm and froze collected flies over ice supplemented with liquid nitrogen. The flies were sexed and stored in 2.0 ml nuclease-free microcentrifuge tubes (Ambion) at -80 °C until processing.

RNA sequencing
We extracted total RNA as described previously [26] with Trizol using the RNeasy Mini Kit (Qiagen, Inc.), and depleted ribosomal RNA from 5 μg of total RNA using the Ribo-Zero ™ Gold Kit (Illumina, Inc.). Depleted mRNA was fragmented and converted to first strand cDNA. During the synthesis of second strand cDNA, we used dUTP instead of dTTP to label the second strand cDNA. We used cDNA from each RNA sample to produce barcoded cDNA libraries using NEXTflex ™ DNA Barcodes (Bioo Scientific, Inc.) with an Illumina TruSeq compatible protocol. Libraries were size selected for 250 bp (insert size ~ 130 bp) using Agencourt Ampure XP Beads (Beckman Coulter, Inc.). Second strand DNA was digested with Uracil-DNA Glycosylase before amplification to produce directional cDNA libraries. We quantified the libraries using Qubit dsDNA HS Kits (Life Technologies, Inc.) and Bioanalyzer (Agilent Technologies, Inc.) to calculate molarity. The libraries were subsequently diluted to equal molarity and re-quantified.
Samples were processed in batches of 48 and 16 libraries were pooled randomly into 25 pools. We quantified the pooled library samples again to calculate final molarity and after denaturation diluted them to 14 pM. Pooled library samples were clustered on an Illumina cBot; each pool was sequenced on one lane of an Illumina Hiseq2500 using 125 bp single-read v4 chemistry.

RNA sequence analysis
Sequences were analyzed exactly as described previously [26]. We demultiplexed barcoded sequence reads using the Illumina pipeline v1.9 and trimmed adapter sequences using cutadapt v1.6 [45]. The trimmed sequences were aligned to multiple target sequence databases, using BWA v0.7.10 (MEM algorithm with parameters '-v 2 -t 4') [46]. First, we aligned all trimmed sequences against a ribosomal RNA database to filter out residual rRNA that escaped depletion during library preparation. Next, we aligned the remaining sequences against a custom database of potential microbiome component species using BWA. We then aligned sequences that did not align to either the rRNA or microbiome databases to all D. melanogaster sequences in RepBase [47]. Finally, we aligned the remaining sequences that did not align to any of the databases above to the D. melanogaster genome (BDGP5) and known transcriptome (FlyBase v5.57) using STAR v2.4.0e [48] .

Gene expression estimation
We followed the analysis described previously [26] to compute read counts for known and novel gene models using HTSeq-count [49] with the 'intersection-nonempty' assignment method. Tabulated read counts for each endogenous gene present in both Baseline and ethanol-treated lines were combined and normalized across all samples using EdgeR [50]. We used the normalized gene expression in all following analyses.

Genetics of gene expression
For each expression feature (known and novel transcripts) we fit mixed-effect models to the normalized gene expression data corresponding to: where Y is the observed log2 (normalized read count), S is sex, W is Wolbachia infection status, W × S is Wolbachia by sex interaction, L is DGRP line, T is treatment (ethanol-supplemented vs standard medium), L × S is the line by sex interaction, L × T is the line by treatment interaction, T × S × L is the treatment by line by sex interaction and ε is the residual error. We also performed reduced analyses for sexes separately We identified genetically variable transcripts as those that passed a 5% FDR threshold (based on Benjamini-Hochberg corrected P-values [51]) for the L, T and L × T terms. We computed the broad sense heritabilities (H 2 ) for each gene expression trait separately for males and females as TxL , σ 2 SxTxL and σ 2 ε are, respectively, the among line, sex by line, treatment by line, sex by treatment by line, and within line variance components.
In addition, for all expression features that were significant for the L × T interaction term we re-analyzed data for each Treatment condition separately to identify transcripts with significant changes in expression in one or another condition, or both. We performed reduced analyses for sexes separately (Y = W + L + ε). We also calculated the Line means differences for the matching transcripts that were significant for the L × T term (i.e., Line. ETOH -Line.Baseline) for males and females separately. These line means differences in expression values were used for hierarchical clustering analysis for the subset of the significant gene expression features using the JMP12 package (SAS, Cary, USA).

Construction of correlated expression networks
We used the differences between the conditional means with and without ethanol for each line to calculate pairwise Pearson correlation coefficients for all genes that had statistically significant Benjamini-Hochberg False Discovery Rate (BH-FDR) [51] adjusted p-values (BH-FDR < 0.05) for the line-by-ethanol treatment interaction term in the linear mixed effects model (Y = W + L + T + L xT + ε run for each sex separately). The delta-expression correlation matrix was bi-clustered using hierarchical clustering with complete linkage agglomeration in Genesis statistical software [52]. The bi-clustered matrix was used for the center panel in each composite figure (Figs. 2  and 3). To generate the delta-expression correlation networks, the correlation matrix was filtered for associations with BH-FDR adjusted p-value < 0.05 and the top 10% of correlations based on the absolute value of the Pearson coefficient. Associations that survive the stringent filtering criteria were input into Cytoscape and clustered using the MCODE algorithm with default parameters, but with the 'Fluff ' setting activated to capture relationships outside of auto-correlated modules [53]. The resulting modules with significantly strong intra-connectivity (cumulative MCODE score > 4) were mapped back to the correlation matrix panel based on the identity of the gene membership of each module. We highlighted genes with a statistically significant (BH-FDR < 0.05) eQTL association calculated from the expression differences in each module within the composite figure. eQTL mapping eQTLs were mapped to differences in expression between baseline and ethanol treatment of genes with a statistically significant line-by-treatment (LxT) term from the linear mixed effects model run for each sex separately as previous described [25,26]. Briefly, we adjusted normalized FPKM values for Wolbachia infection status, chromosomal inversions, population structures organized based on top 10 principal components using Best Linear Unbiased Predictor (BLUP) using the R package lmerTest. We used covariate adjusted expression differences as phenotype for eQTL mapping using PLINK (v1.90). We compared association P-values generated by the PLINK t-tests to the empirical FDR threshold calculated by dividing the number of expected associations under the null hypothesis generated from 100 permutations at a false discovery rate of 0.05 by the observed number of associations at the same threshold to determine statistical significance. We further filtered associations filtered for independence using forward model selection, as previously described [25,26], by iteratively adding single eQTLs, starting with the smallest P-values, to an additive association model such that the conditional P-value for the last added eQTL is no more than 1E-5.

Association networks
To build association networks using the variants identified from eQTL mapping, we added pairwise eQTL associations between genes that either contain the variant or are within 1000 bp up-or down-stream of the variant and genes with statistically significant line-by-treatment (LxT) to the most recent version (fb_2021_05) of the database of known genetic and protein-protein or RNAprotein physical interactions from the FlyBase repository and visualized in Cytoscape. The resulting networks of associations were filtered to contain (i) genes that are part of the eQTL associations, and (ii) genes that have at least 5 genetic or physical interactions to genes that are part of the eQTL associations within one interaction distance (one edge). The filtered interaction network was modularized using MCODE algorithm with default settings but with 'fluff ' activated [53]. We input genes that are part of the individual modules for Gene Ontology Enrichment analysis. Statistically significant (FDR < 0.05 in the Overrepresentation Test), highly specialized terms containing the largest number of genes from the input from each module were used for functionally labeling each module. For the glutathione metabolism inlet in the female interaction network in Fig. 3, GstE family of genes were added to the GstD tri-gene cluster based on semantic similarity.

Analysis of activity and sleep phenotypes
Flies were reared on standard medium and medium supplemented with 10% (v/v) ethanol and placed in standard food collection vials overnight. The next day, mated females and males from both treatment conditions were placed in Drosophila Activity Monitor (DAM) tubes (TriKinetics, Waltham, MA) that contained agar supplemented with 5% sucrose at one end, and a small piece of yarn at the other end. Flies were placed in a 25 °C incubator on a 12-h light-dark cycle and their activity and sleep data were recorded using the DAMSystem (Trikinetics). Raw data from the DAMSystem (TriKinetics) were uploaded to ShinyR-DAM [54] and resulting output data were parsed by sleep/activity phenotype for analysis. Sleep was defined as at least 5 min of inactivity and only data from flies that survived the entire testing period (2-9 days of the fly lifespan) were retained for analysis. Sleep and activity data were analyzed using the PROC MIXED command (Type III Analysis of Variance (ANOVA)) within SAS version 9.04 (Cary, NC) according to the model Y = μ + T + L + S + TxL + TxS + LxS + TxL xS + Rep(TxLxS) + ε, where T is the fixed effect of treatment (ethanol medium, standard medium), L is the fixed effect of line (RAL_177, RAL_208, RAL_367, RAL_555, RAL_705, RAL_730), S is the fixed effect of Sex (male, female), Rep(TxLxS) is the random effect of replicate, and ε is the residual variance. Reduced Type III ANOVAs (Y = μ + T + L + TxL + Rep(TxL) + ε) were also performed by Sex.