Conservation of the behavioral and transcriptional response to social experience among Drosophilids

While social experience has been shown to significantly alter behaviors in a wide range of species, comparative studies that uniformly measure the impact of a single experience across multiple species have been lacking, limiting our understanding of how plastic traits evolve. To address this, we quantified variations in social feeding behaviors across 10 species of Drosophilids, tested the effect of altering rearing context on these behaviors (reared in groups or in isolation) and correlated observed behavioral shifts to accompanying transcriptional changes in the heads of these flies. We observed significant variability in the extent of aggressiveness, the utilization of social cues during food search, and social space preferences across species. The sensitivity of these behaviors to rearing experience also varied: socially naive flies were more aggressive than their socialized conspecifics in some species, and more reserved or identical in others. Despite these differences, the mechanism of socialization appeared to be conserved within the melanogaster subgroup as species could cross‐socialize each other, and the transcriptional response to social exposure was significantly conserved. The expression levels of chemosensory‐perception genes often varied between species and rearing conditions, supporting a growing body of evidence that behavioral evolution is driven by the differential regulation of this class of genes. The clear differences in behavioral responses to socialization observed in Drosophilids make this an ideal system for continued studies on the genetic basis and evolution of socialization and behavioral plasticity.

While social experience has been shown to significantly alter behaviors in a wide range of species, comparative studies that uniformly measure the impact of a single experience across multiple species have been lacking, limiting our understanding of how plastic traits evolve. To address this, we quantified variations in social feeding behaviors across 10 species of Drosophilids, tested the effect of altering rearing context on these behaviors (reared in groups or in isolation) and correlated observed behavioral shifts to accompanying transcriptional changes in the heads of these flies. We observed significant variability in the extent of aggressiveness, the utilization of social cues during food search, and social space preferences across species. The sensitivity of these behaviors to rearing experience also varied: socially naive flies were more aggressive than their socialized conspecifics in some species, and more reserved or identical in others. Despite these differences, the mechanism of socialization appeared to be conserved within the melanogaster subgroup as species could cross-socialize each other, and the transcriptional response to social exposure was significantly conserved. The expression levels of chemosensory-perception genes often varied between species and rearing conditions, supporting a growing body of evidence that behavioral evolution is driven by the differential regulation of this class of genes. The clear differences in behavioral responses to socialization observed in Drosophilids make this an ideal system for continued studies on the genetic basis and evolution of socialization and behavioral plasticity.
behavior, chemosensation, Drosophilids, evolution, experience, gene expression, interspecies variation, neurogenomics, socialization, social interactions 1 | INTRODUCTION Prolonged social exposure has been shown to significantly alter the behavior in species ranging from fruit flies to humans, [1][2][3][4][5] suggesting that the ability to modulate behavior in response to the frequency of social interactions is evolutionarily advantageous. These socially mediated behavioral shifts correlate with experience-induced alterations in gene expression in the brain, [5][6][7][8] implying that this behavioral response is partially determined by the properties of the gene regulatory network. One of the clearest examples linking social experience, gene regulation, and behavior is in the fruit fly Drosophila melanogaster, where Wang et al. showed that the expression level of a single cytochrome P450 gene, Cyp6a20, increases with the amount of time the fly is socialized and is inversely proportional to aggressiveness in males. 5 Interestingly, Cyp6a20 is also downregulated in D. melanogaster selectively bred to be hyperaggressive, 9 suggesting that evolution and experience can both modulate behavior through altering the expression levels of the same target genes, 5,6 and that phenotypic evolution may occur through the canalization of an experience-altered state. 10 While many studies have shown that behavior and gene expression are responsive to social experience, the conservation of these effects across multiple species has not been characterized in a controlled assay. Here, we quantified changes in social feeding behaviors across 10 species of Drosophilids reared in different contexts (in groups or in isolation), and correlated changes in behavior to gene expression levels in the heads of evolutionarily and experientially diverged flies. We use the term social feeding behaviors to encompass behaviors involved in food exploration by a group of flies and interactions among individuals at the food source. Comparisons of independently evolving systems have been extremely useful in the elucidation of biological mechanisms and evolutionary principles. [11][12][13] This has been especially true in Drosophilids where the genomes of more than 30 species have been sequenced allowing for phenotypic differences to be related to genetic changes. 14,15 For our behavioral comparison, we chose 10 Drosophilid species that satisfied the following criteria: (1) have sequenced genomes, (2) cover a large range of evolutionary divergences, (3) are found in contrasting ecologies, and (4) can be reared easily on the same laboratory food source. Eight of these species, Drosophila melanogaster, Drosophila simulans, Drosophila sechellia, Drosophila yakuba, Drosophila erecta, Drosophila ananassae, Drosophila pseudoobscura, and Drosophila willistoni, belong to the subgenus Sophophora and represent both dietary generalists and species that are specialized upon a particular fruit: D. erecta on the seasonally restricted Pandanus fruit 16 and D. sechellia on the toxic Morinda fruit. 17 The other 2 species, Drosophila arizonae and Drosophila virilis, belong to the subgenus Drosophila, and are separated from the others by approximately 40 million years ago. 14 Both of these species are specialists: D. arizonae breeds in necrotic cactus, 18 and D. virilis breeds in the slime flux exudates of deciduous trees. 19 The phylogenetic relationships of all 10 species are shown in Figure 1E, and additional ecological information is described in Table S1 21 This effect was decreased in their assay for vision and smell mutants, suggesting that these social aggregation cues are transmitted through both sensory modalities. Clearly pheromone detection plays an important role in food search as the social pheromone 11-cis vaccenyl acetate (CVA) has been shown to significantly increase aggregation rates when mixed with food odors. 22,23 Interestingly, CVA by itself is not sufficient to attract freely searching D. melanogaster, but does trigger aggregation in D. virilis, 22 suggesting a divergence in how this social cue is processed, and potential interspecies variation in the utilization of social cues during food search.
Once flies aggregate on a food source, they exhibit a range of social interactions including courtship, mating, and aggression. [24][25][26] These behaviors have been shown to have a wide range of intraspecific variation, but studies that quantify interspecific variation have been limited to a few species, and focused on mating behaviors where large behavioral shifts are thought to drive speciation. [27][28][29] Interspecific variation in aggression has not been systematically characterized, despite the fact that extensive studies of aggression in D. melanogaster have showed that it is a highly variable trait. 9 These aggression studies have primarily focused on male-male interactions, but female D. melanogaster aggression has been observed and occurs less frequently. 4,30 Social experience has been shown to alter these interactions in D. melanogaster: male courtship and aggression frequencies are reduced after unsuccessful attempts at mating and fighting, respectively, 31,32 male and female aggression decreases with increased socialization, 4,5 and social space also decreases with increased socialization. 33 Conservation of plasticity in these behaviors has not been characterized, but 4 species of Drosophilids have been shown to similarly alter their oviposition site preference in the presence of parasitic wasps, 34 suggesting that specific behavioral responses to experiences can be conserved between species. Furthermore, gene expression levels have also been shown to alter with aggressive and sexual experiences in D. melanogaster, 8,35-37 but again conservation of experience-driven gene expression changes and their relationship to behavioral shifts have not been characterized.

| Food aggregation assay
We adapted the flight chamber described in Tinette et al. for our food aggregation assay 21 ( Figure 1A). The chamber is a Plexiglass box with interior dimensions 12.75 00 wide X 8.875 00 deep X 10.75 00 high. Two flush inset slides in the solid white base of the chamber allow food to be easily introduced and removed in the presence of flies. The box is surrounded on 3 sides by a white plastic shield, and covered by a solid black plastic lid with an access port above each food source that fits a camera lens, or may be closed if no camera is in use. This shielding prevents external visual distractions. A 17 00 X 24 00 LED lightbox (Artograph A950) was placed 5.75 00 from the exposed side of the chamber to provide uniform light during the assay. To further eliminate external distractions, all assays were performed in an enclosed cabinet.
Fly stocks were reared in plastic vials (Genesee Scientific Cat. # 32-109, El Cajon, California) at 23C on standard lab food made of dark corn syrup (30 mL/L), sucrose (15 g/L), yeast (35 g/L), and agar (10 g/ L). A total of 100 virgin females of each species were collected within 20 hours after eclosion and housed in groups of 25 flies. Flies were aged 5 to 6 days, with a target ratio of 50 flies of each age. Flies were maintained on a 12:12 light/dark cycle, and loaded into the chamber at 1 hour after lights on through a load port in the center of the unshielded wall of the chamber ( Figure 1A). We acclimated the flies to the chamber for 2 hours prior to food introduction to allow them to recover from the transfer, and to increase their attraction to the food source after a short starvation period. After 2 hours, a white cap

| Food choice and density assays
To assay if food type affects fly aggregation behavior ( Figure S1), we measured D. melanogaster, D. arizonae, and D. sechellia on 3 food sources: standard lab food, lab food mixed with cactus rot liquid (natural food of D. arizonae 18 ), or with noni fruit juice (natural food of D. sechellia 17 ). To make the cactus food, we added water to frozen necrotic Carnegiea gigantea cactus, incubated for 5 days at 37 C, and added 1 mL of rot liquid to 5 mL of melted lab food. This mixture was then added to a food dish (3 mL) to solidify. To make the noni fruit food, we dissolved 1 g of dried noni fruit leather (Hawaiian Organic Noni, Anahola, Hawaii) in 8.75 mL water at 37 C for 1.5 hours, and added 1 mL of noni juice to 5 mL of melted lab food. This mixture was then added to food dish (3 mL) to solidify. In addition, 3 replicates of the food choice assays were performed with D. melanogaster, D. sechellia, and D. arizonae against all 3 food sources. Fly density experiments were conducted using the standard aggregation assay except that 25, 50, 75, or 100 flies were loaded into the chamber. We performed 4 replicates for each density, and assays ran for at least 2 hours. We were not able to obtain a T-40 measurement for D. melanogaster at a density of 25 flies, because they did not aggregate to a level of 40% over the course of the assay. The regression line for D. pseudoobscura was only fit to the first 3 densities. Painted flies were not included for either food choice or dilution assays.

| Denatonium assay
To determine how the presence of flies impacts exploration of a food source, we performed a food choice assay similar to the one described by Tinette et al. 21 We poured two 3 mL food dishes containing lab food. We added 150 μL of 10 mg/mL denatonium to the center of 1 food dish (after solidified), and let sit for 10 minutes. The food aggregation assay was performed as described above, except we slid in 2 food dishes (with and without denatonium) for comparison. The number of new flies was counted that went to each food dish within a 21 minutes window. Four replicates were carried out for D. melanogaster and D. erecta.

| Sensory subtraction assay
To assay the relative importance of vision in socially mediated food search, we poured two 1.5 mL food dishes containing lab food, added 50 virgin females to one, and covered both with a piece of cheesecloth. Both food dishes were loaded into the chamber, and we counted the number of accumulated flies on both dishes after 2 hours. To test the importance of smell, we conducted the same experiment except that both food dishes were covered with a glass coverslip. Three replicates were performed for all conditions for D. melanogaster and D. erecta.

| Cooperative search simulations
Food search simulations were carried out using the script "FlyAggre-

| Obp49a mutant assay
To generate a D. melanogaster Obp49a heterozygous mutant, we crossed the Obp49a deletion strain created by Jeong et al. 38 (Bloomington Stock #55037, Bloomington, Indiana) with the D. melanogaster Canton S strain we used in the species comparisons.
As described above, 4 replicates of 100 virgin females aged 5 to 6 days were performed.

| Behavioral analysis
Video was automatically analyzed using custom scripts written for R with the EBImage library 39 (Table S7). All videos were scaled so that the size and position of the food dish were identical for all runs. The number of flies on and proximal to the food dish were counted for every frame over a 10 second window and averaged. This was repeated for every 45 second increment. Social space was calculated as the average minimum distance between all flies within a frame. This was performed automatically by: (1) fitting an ellipse to each fly, (2) finding the nearest neighbor for each ellipse as measured as the distance between ellipse centers, and (3) calculating the mean distance between flies. As social space varied as a function of fly density at the food (more flies trying to eat increased crowding and decreased social space, Figure S6), these means were averaged over all frames that had a specified range of fly densities for interspecies comparison.
If 2 flies were the nearest neighbors to each other, that distance was only counted once. To normalize social space to body size for each species, we calculated the mean of the major and minor axes of all fit ellipses in a single frame and subtracted it from the average minimum distance. By doing this correction, social space is a measure of the distance between ellipse edges rather than centers, and is independent of fly size. This independence is confirmed by the lack of correlation between social space measured at a fly density of 15 to 25 flies ( Figure 1) and fly body size (R 2 = 0.19).
To quantify aggressiveness, we manually counted the number of aggressive lunges exhibited by individual flies within their first 150 seconds of arriving to the food source. A lunge is considered aggressive if it displaces another fly, is preceded by display of wingthreat or reorientation of the fly toward its target, or is accompanied by chasing or multiple rapid contacts. 4 We scored aggression in all painted flies that went to the food and at least 20 random socially reared flies for each species. Random flies were picked around times that naive flies arrived at the food to compare flies at similar densities.
We only scored 9 socially reared D. arizonae because they did not aggregate at sufficiently high levels. We were able to score at least

| RNA-Seq library preparation and sequencing
To homogenize fly heads, 200 μL of TRI-Reagent (Ambion, Foster City, California) was added to 50 to 100 heads, and then homogenized with a motorized pestal for 60 to 90 seconds. An additional 800 μL of TRI-Reagent was added to this homogenate, centrifuged for 1 minute at 12 000g, and the supernatant was divided between 2 RNAse free 1.8 mL tubes. Total RNA was then purified using the Direct-zol RNA

| RNA-Seq processing
Reads were mapped to their respective genomes using CLC Genomics Workbench Version 5.1.5. Up to 2 mismatches were allowed per read.
For each species, we used the all_chromosome.fasta files reported on Reads per gene were calculated as both raw reads (for use with DESeq 42 ) and as RPKMs. RPKM values were subsequently converted to transcripts per million (TPM bases sequenced). To do this, we normalized the RPKM values for all genes, so that their summed value was 1 million.

| RNA-Seq analysis
To identify orthologous genes between species, we used the gene_ orthologs_fb_2014_05.tsv gene ortholog assignment reported on FlyBase. For subsequent comparative analysis, we only considered those genes that had an ortholog to a gene in D. melanogaster. Table S4 contains the number of genes for each species that had D. melanogaster orthologs, the number of reads that mapped to exons in these genes, and the correlation coefficients (R 2 ) between all biological replicates. We generated RNA-Seq data for at least 3 biological replicates for 10 of the 14 conditions assayed (10 species socially reared, 4 species non-socially reared), and only 2 replicates for the remaining 4 conditions (socialized D. ananassae, socialized D. sechellia, socialized D. yakuba and non-socialized D. erecta). Table S5 has raw read counts, DESeq-normalized raw read counts (estimateSizeFactors function), RPKM and TPM values for all D. melanogaster orthologs in all sequenced samples. To cluster the species based on relatedness in their global gene expression levels, we used the R function cor to determine the Spearman rank correlation coefficient for all pairwise species comparisons. We subtracted these values from 1 and used the as.dist function to generate a dissimilarity matrix which was then clustered using the hclust function with the "Ward" method option. 43 Differential expression analysis was carried out using the raw read values reported in Table S5 and the DESeq version 1.4.1 functions newCountDataSet, estimateSizeFactors, sizeFactors, esti-mateVarianceFunctions and nbinomTest with default parameters. 42 Differential expression was considered significant if the DESeq reported adjusted P value (padj) was ≤0.05. Expression in 1 species was considered significantly differentially expressed relative to a set of other species (i.e. the melanogaster subgroup) if all pairwise comparisons had a padj ≤ 0.05. If 1 species of the set did not have an ortholog assignment for a given gene, that gene was not considered.
Significant enrichment of gene ontology terms was carried out using the Princeton Generic Gene Ontology Term Finder found at go. princeton.edu. 44 To determine if the overlap of differentially expressed genes between samples was significant, we randomly selected the same number of differentially expressed genes for each sample from all D. melanogaster orthologs, and calculated the fraction of overlap. We repeated this 10,000 times and determined significance of a given overlap based on the distribution of overlap from the randomly sampled lists.

| Comparative behavioral assay
We quantified changes in social feeding behaviors between evolutionarily and experientially diverged flies using a novel, highly automatable assay (see Section 2). Briefly, 100 virgin females were loaded into a Plexiglass chamber, given a single food source via a slide on the chamber floor, and filmed interacting on the food for at least 2 hours (schematic in Figure 1A). We only used virgin females to limit potential behavioral variability from comparing mated to virgin females (sexual maturation times vary between species 45 Figure 1B are painted). Flies were raised in complete isolation, as opposed to a more "natural" rearing context (small group size or limited social exposure), to minimize the handling of flies prior to behavioral analysis, and to maximize the induced behavioral and transcriptional differences between rearing conditions.

| Interspecific variations in group-reared behaviors
Interspecific differences in social feeding behaviors among groupreared flies were immediately apparent by examining the density and positioning of each species on the food after 2 hours ( Figure 1E). As we only tested a single strain for each species, observed behavioral traits for that strain may not be representive of the species as a whole.
Regardless, we provide ecological information for each species to support observed interspecies differences, and to identify behavioral trends across species that evolved under similar environmental pressures. Aggregation kinetics were highly variable, even among closely related species ( Figure 1D). D. erecta, a dietary specialist whose primary food source (Pandanus fruit) is only available 3 months of the year, 16 aggregated much faster than any other species. This is consistent with previous observations that D. erecta aggregate at high densities in the wild, 16  were performed using the Wilcoxon rank sum test unless otherwise noted), suggesting that variations in aggregation levels are not due to differences in food affinity ( Figure S1).
Differences in aggregation may be explained by variability in physiological properties including sensitivity to odors, visual acuity, dehydration rates, and metabolism. Alternatively, these differences may be caused by variations in the use of social cues during food search between species. 21,22 As previously mentioned, D. melanogaster are more likely to go to a food source that already has flies accumulated on it, suggesting that food search kinetics show positive cooperativity. 21,51 To quantify the degree of social interaction during food search, we measured the amount of time necessary for 40% of the flies in the chamber to go to the food (T-40) as a function of fly density 51 ( Figure 1C). Of the 5 species that aggregated at high levels in Figure 1D 20 Below each species is a snapshot of the flies on the food after 120 minutes, and the average minimum distance (social space) AE1 SD calculated from frames that had only 15 to 25 flies (i.e. low-fly density) during the course of the assay. Social space is measured in pixels (px); Drosophila melanogaster average body length = 55.4 px food search simulation suggests that there is a limit to which an increase in fly number increases the attractiveness of the food ( Figure S2). This limit may be lower for D. pseudoobscura, which is larger than the other species assayed.
To further assay cooperativity, we performed a food choice assay similar to the one perdomed by Tinette et al. 21 for D. melanogaster (cooperative) and D. erecta (non-cooperative). We presented flies with 2 food sources: one containing the odorless, but taste-aversive  Figure S9. (B) The average number of aggressive events observed in socially reared (social: Dark bars) and isolated (naive: Light bars) flies during the first 150 seconds after they arrive at the food source. The error bars show standard error of the mean. Asterisks designate significant differences. D. erecta is significantly more aggressive than all other species. (C) Same as (B) except includes number of aggressive events for flies reared with a different species chemical denatonium that prevents aggregation at the food, and one without. As previously observed, the initial kinetics of D. melanogaster to either food source was equivalent, but the relative rate of aggregation to the non-denatonium food increased as more flies accumulated there suggesting cooperative food search ( Figure S3). 21 Like D. melanogaster, D. erecta showed similar initial kinetics to both food sources, but remained clustered in a small group next to the food containing denatonium, suggesting that even though D. erecta may not exhibit cooperativity during food search, they may still be highly social ( Figure S4). To determine the importance of vision and smell in our assay, we placed either flies and food, or food alone under either a piece of cheesecloth (no visual cue) or a glass coverslip (no odor cue).
No condition was sufficient to drive the aggregation of D. melanogaster suggesting a search dependence on both smell and vision ( Figure S5), as previously observed by Tinette et al. 21 D. erecta aggregated on either food, or food and flies under cheesecloth at similar rates, suggesting that food odors are the primary attractant for these flies, which is consistent with their apparent loss of cooperativity during food search. D. erecta did not accumulate on flies and food covered by a coverslip ( Figure S5).
Once aggregated, the species varied significantly in how they interacted on the food. D. melanogaster, D. pseudoobscura, and D. sechellia had a significantly larger social space than all other species at low fly densities (P < .01 for all pairwise comparisons), and D. erecta, D. yakuba, and D. simulans consistently had a smaller social space across densities ( Figure 1E; Figure S6). D. erecta was significantly more aggressive than any other species (P ≤ .05, Figure 2B; Video S1), including its close relative D. yakuba which exhibited slow aggregation kinetics and minimal aggressive contact ( Figure 1D and 2B). Both species exhibited a strong preference to be on the food ( Figure S7) and a tight social space suggesting that aggressiveness and social space can evolve independently. While aggression levels have been shown to be highly variable in D. melanogaster males, 9 the hyperaggressive behavior observed in D. erecta has not been observed previously in females. D. pseudoobscura was significantly more aggressive than the remaining species except for D. melanogaster (P ≤ 0.05), which in turn was significantly more aggressive than all other species except for its close relative, D. simulans (P ≤ 0.05, Figure 2B). As species that aggregated at higher levels in Figure 1D tended to be more aggressive, we tested whether the proportional instance of aggression FIGURE 3 Gene expression levels are moderately conserved across species. Scatter plots on the right show the mean expression level (TPM) for each gene in Drosophila melanogaster (x-axis) vs the mean expression level for the corresponding ortholog in Drosophila erecta, Drosophila pseudoobscura, or Drosophila virilis (y-axis). A regression line was fit to these plots for all pairwise species comparisons, and the R 2 was used to color the heat map on the left (color scale to the right of the heat map, all values below R 2 = 0.68 are colored white). The green dendrograms show the predicted relatedness of the species to each other as determined by hierarchical cluster analysis (see Section 2) increased with fly density. It did not for any species (Figure S8), suggesting that like higher aggregation levels, increased aggression may be selected for in food-competitive species.

| Interspecific differences in behavioral response to socialization
Social space and aggression have both been shown to vary with rearing context in D. melanogaster, 4,5,33 but conservation of these experience-modulated behavioral shifts had not been characterized.
We measured the impact of socialization on these 2 behaviors in all 10 species. As previously observed, D. melanogaster flies were significantly more aggressive when raised in isolation (p = 5.7e −5 , Figure 2A, B, Video S2). 4,5 This trend was consistent but not significant for D. erecta and D. pseudoobscura ( Figure 2B). Socialization did not affect aggression levels in the other species, clearly showing interspecific variation in the plasticity of this behavior. Socially naive D. melanogaster did not exhibit a larger social space as previously observed in a pyramid assay, 33 but D. erecta did; naive D. erecta went to the food but tended to stay on the outside of the group (Figure 2A, Figure S9, Video S1). This spatial separation was not observed in any other species. Differences in the behavioral response to rearing context could be the result of interspecific differences in the transmission of socialization cues (pheromone profiles vary across species 52 ), or interspecific differences in how the received cues feed into downstream behavioral processes. To try to separate these 2 possibilities, we reared individual D. melanogaster and D. erecta flies in vials with 25 flies of the other species, and tested them in our assay. Crossspecies reared D. melanogaster and D. erecta were both significantly less aggressive than conspecifics raised in isolation (P ≤ 0.005, Figure 2C), indicating that the mechanism of socialization is conserved between these species. D. erecta reared with D. melanogaster no longer stayed on the outside of the group as seen in socially naive flies further supporting cross-species socialization (Figure 2A). Interestingly, D. erecta raised with the less aggressive D. melanogaster were themselves significantly less aggressive than D. erecta raised with conspecifics (P = 0.03), suggesting socialization cues may be expressed at higher levels in the larger D. melanogaster.

| Interspecific differences in brain transcriptome
To identify potential genetic effectors of these behaviors and conserved transcriptome changes related to socialization, we measured gene expression levels in the heads of all 10 species reared in groups, and in socially naive D. melanogaster, D. erecta, D. simulans and D. yakuba. We focused on experience-driven transcriptome changes in these 4 species in the melanogaster subgroup because they significantly varied in their behavioral response to socialization, and diverged relatively recently ( Figure 1E). Gene expression levels were fairly well conserved between species (Figure 3), and clustered roughly as expected based on the previously reported genomicsequence derived phylogeny 20 ( Figure 1E). As D. erecta exhibited extremes of all behaviors assayed (fast aggregation kinetics, hyperaggression, noncooperative food searching and tight social space), we determined which genes were differentially regulated in this species relative to the rest of the melanogaster subgroup (Table S2). Again, we limited our analysis to these species because they significantly varied in behavior and diverged relatively recently. Cytochrome P450 genes, which are involved in pheromone synthesis and degradation, 53 were significantly enriched among differentially express transcripts in D. erecta (p = 4.09e −5 , GO analysis 44 ). This included Cyp6a20, which is thought to be involved in the degradation of aggression-triggering pheromones, 9 and whose expression is inversely related to aggression levels in D. melanogaster males as previously mentioned. 5 Interestingly, Cyp6a20 was significantly upregulated in the hyperaggressive D. erecta, which is counter to the expression-behavior trend observed in D. melanogaster. An evolved upregulation in expression may be advantageous to deal with increased pheromone exposure in the densely packed D. erecta. As in D. melanogaster males, 5 Cyp6a20 levels did increase with socialization in the 4 species where we have expression data for both rearing contexts, but these expression differences were not significant ( Figure 4A). However, we did observe a greater relative change in Cyp6a20 expression in the species that exhibited FIGURE 4 Behavioral variation correlates with gene expression changes. (A) Red circles mark the mean Cyp6a20 expression level (TPM reads) of biological replicates (black circles) for that species and rearing context. Aggression and Cyp6a20 expression levels are significantly higher in Drosophila erecta relative to all other species. The red numbers under each species name signifies the-fold change in Cyp6a20 expression between rearing conditions. (B) Larger circles signify the mean Obp49a expression level and smaller circles signify values from each biological replicate. Circles are colored based on the species name next to the large circle. Names that end in "N" signify data for flies reared in isolation (socially naive). Horizontal error bars are the SE of the mean for social space measurement (pixels) behavioral sensitivity to rearing condition (D. erecta and D. melanogaster, Figure 4A). Interestingly, the autism-linked Wac gene, which has been shown to affect nonassociative conditioning in D. melanogaster, was also significantly upregulated in D. erecta relative to the rest of the melanogaster subgroup (Table S2). 54 Genes involved in chemoreception and processing in the peripheral nervous system (PNS), like Cyp6a20, are generally thought to be under strong selection during behavioral evolution. [55][56][57] The PNSexpressed odorant-binding protein (Obp) genes, which specifically shuttle odorants and pheromones to odorant receptor neurons, 58 were also enriched among differentially expressed transcripts in D. erecta (P = 0.007). This is consistent with the observation that this class of genes is evolving at a faster rate in D. erecta relative to the rest of melanogaster subgroup, likely due to selection for a specialized diet. 56 As food search and social interactions are both largely mediated by the perception of chemical stimuli, 55

| Conservation of experience-driven changes in gene expression
To identify genes that vary with rearing context, we focused on those transcripts that were significantly altered between socialized and nonsocialized flies in multiple species, or overlapped with previously reported gene expression changes in a similar experiment with D. melanogaster males 5 ( Figure 5). Approximately, the same number of transcripts were differentially regulated in D. melanogaster females as previously reported for D. melanogaster males 5   "melF", "ereF", "simF", "yakF" are the log2 expression change of that gene in socially reared virgin females/socially naive virgin females (color scale on right, values outside of range are set to −2 or 2 depending on sign). If the expression change for a gene in a given condition is not significant, it is black. The remaining columns indicate if that gene was previously identified as significantly altered in Drosophila melanogaster after some social experiences. Green is an upregulation and red is a down-regulation of the first social condition relative to the second in all cases. melM: males raised in groups vs males raised in isolation, 5 courtM1: males after courting females vs non-courting males, 35 courtM2: males after courting females vs noncourting males, 36 matedM: mated males vs non courting males, 36 matedF: mated females vs virgin females. 60 We show those genes that satisfy at least 1 of the following criteria: significantly altered in: (1) both melF and ereF, (2) melM and either melF or ereF and (3) at least 3 of melF, ereF, simF, and yakF. Orange names are involved in sensory perception, behavior, neuro-signaling, and neurodevelopment. Blue names are involved in immune defense response. "*" signify genes expressed in the extracellular region respectively, with an approximately 3-fold greater number of downregulated vs upregulated genes after socialization for both sexes. Of these 201 genes, 81 were differentially regulated in at least 1 other species tested in our assay (Table S3), which is significantly higher than expected by random chance (p < 1 × 10 −4 ) and supports conservation of experience-driven transcriptome changes in the brain.
Among these conserved changes, immune response genes were significantly enriched (p = 5.5e −4 , GO analysis 44  As the behavioral response to socialization varied across species, we were also interested in identifying non-conserved expression changes that may account for these behavioral differences. Interestingly, the odorant-binding protein gene Obp49a was significantly upregulated in D. erecta females after socialization, but not in females of any other species. This differential expression in D. erecta may contribute to the experience-mediated change in social space only observed in this species. Social space measurements at higher densities are strongly correlated with Obp49a expression levels across species and rearing conditions, further suggesting Obp49a is an effector of this phenotype (R 2 = 0.89 D. simulans excluded from regression, Figure 4B). However, when we tested a D. melanogaster Obp49a heterozygous mutant in our assay, we did not observe a significant difference in social space. The effect may be specific to D. erecta or only realized with more severe alterations in expression, like the observed significant down-regulation of Obp49a in D. sechellia relative to the rest of the melanogaster subgroup, which exhibits the largest social space at low densities ( Figure 1E). Obp49a has been shown to mediate the suppression of sugar-activated gustatory receptor neurons in the presence of bitter chemicals. 38 As previously mentioned, D. erecta responded very differently to food-containing denatonium than D. melanogaster; D. erecta clustered in small groups next to bitter food and D. melanogaster left immediately after tasting ( Figures S3A and   S4A). The variation in social space observed with Obp49a expression differences in D. erecta may mirror changes in aversiveness to chemical stimuli in this species.

| CONCLUSION
While it is clear that accumulated social experience influences behavioral output, 62 the conservation of this influence has not been systematically studied in a controlled assay. We have perfomed that here by exposing 10 different species of Drosophilids to 2 different rearing experiences, and by quantifying the impact of that experience on adult behavior and gene expression levels. Significant shifts in social feeding behaviors and their sensitivity to rearing context, clearly show that these behaviors and their plasticity are highly variable. Determining the mechanism of these interspecific changes is a more difficult task as these differences may be the result of interspecific variations in genic sequences, 14 anatomy, 45 pheromone profiles, 52 and/or neuronal circuitry. The cross-socialization of D. melanogaster and D. erecta suggests that at least within the melanogaster subgroup, the mechanism of socialization is conserved, but which behaviors are influenced and to what extent is not (Figure 2). The lack of socially mediated plasticity in aggression and social space in the majority of the species does not mean that socialization does not affect any behaviors in these species, just not these behaviors in these contexts. D. melanogaster males are much more aggressive than females, 30 and a significant change in male aggression levels with socialization may be more conserved across these species.
Clearly, there is some conserved response to socialization within the melanogaster subgroup, as there is significant overlap in differentially expressed genes after prolonged social exposure ( Figure 5). A growing body of evidence points to evolved differences in the sequence and expression of chemosensory genes as a driving force in behavioral evolution. [55][56][57] This is likely because variations in odorant/ pheromone detection at the most peripheral level of the nervous system can immediately alter the degree of influence of a given stimuli, or which neurons are responsive to that stimuli. Our observation that these genes are enriched among those differentially regulated in D. erecta and D. sechellia, the 2 species that display extremes of all behaviors assayed, supports this idea. Observed differences in the regulation of these genes in response to social experience, likely contributes to the observed differences in behavioral plasticity. Further experiments are needed to directly link expression changes observed here to behavioral shift, but our data will be instrumental in guiding those efforts.

ACKNOWLEDGMENTS
We would like to thank Sarah Munchel, Kenyon Applebee, Romana