Transcriptomic evidence connecting low aggression to disease risk in honey bees

Background: Many organisms display a generalized suite of behaviors that indicate infection or predict infection susceptibility. We apply this concept to honey bee aggression, a behavior that has been associated with positive health outcomes in previous studies. We sequenced the transcriptomes of the brain, fat body, and midgut of adult sibling worker bees who developed as pre-adults in relatively high versus low aggression colonies. Previous studies showed that this pre-adult experience impacted both aggressive behavior and resilience to pesticides. We performed enrichment analyses on differentially expressed genes to determine whether variation in aggression resembles the molecular response to infection. We further assessed whether the transcriptomic signature of aggression in the brain overlapped with that observed following acute predator threat, exposure to a high-aggression environment as an adult, or changes associated with adult behavioral maturation. Results: Across all three tissues assessed, genes that are differentially expressed as a function of aggression signicantly overlap with genes whose expression is modulated by a variety of pathogens. In the fat body, and to a lesser degree the midgut, we nd evidence of directional concordance consistent with the hypothesis that low aggression resembles a diseased or parasitized state. However, we nd little evidence of acute infection in low aggression individuals. Furthermore, we nd little evidence that the brain molecular signature of aggression in the current study is enriched for genes modulated by either ephemeral or stable social cues that induce aggression in adults. However, we do nd evidence that genes associated with adult behavioral maturation are enriched in our brain samples, with no clear directional bias. Conclusions: Results support the hypothesis that low aggression resembles a molecular state associated with infection. This pattern is most robust in the peripheral fat body, an immune responsive tissue in the honey bee. Although these results are correlative, we nd no evidence of acute infection in low aggression bees, suggesting the physiological state associated with low aggression may predispose bees to negative health outcomes. The similarity of molecular with traits suggests that these characteristics may, fact, be intimately tied.

expressed genes to determine whether variation in aggression resembles the molecular response to infection. We further assessed whether the transcriptomic signature of aggression in the brain overlapped with that observed following acute predator threat, exposure to a high-aggression environment as an adult, or changes associated with adult behavioral maturation. Results: Across all three tissues assessed, genes that are differentially expressed as a function of aggression signi cantly overlap with genes whose expression is modulated by a variety of pathogens. In the fat body, and to a lesser degree the midgut, we nd evidence of directional concordance consistent with the hypothesis that low aggression resembles a diseased or parasitized state. However, we nd little evidence of acute infection in low aggression individuals. Furthermore, we nd little evidence that the brain molecular signature of aggression in the current study is enriched for genes modulated by either ephemeral or stable social cues that induce aggression in adults. However, we do nd evidence that genes associated with adult behavioral maturation are enriched in our brain samples, with no clear directional bias. Conclusions: Results support the hypothesis that low aggression resembles a molecular state associated with infection. This pattern is most robust in the peripheral fat body, an immune responsive tissue in the honey bee. Although these results are correlative, we nd no evidence of acute infection in low aggression bees, suggesting the physiological state associated with low aggression may predispose bees to negative health outcomes.
The similarity of molecular signatures associated with the seemingly disparate traits of aggression and disease suggests that these characteristics may, in fact, be intimately tied.

Background
Behavior often re ects an organism's health status. For example, in vertebrates, illness and infection cause a distinct suite of behavioral responses known collectively as "sickness behavior" (Maes et al. 2012). These phenotypes, which include lethargy, fatigue, and changes in cognitive function, are regulated by molecules that signal systemic infection to the brain (Biesmans et al. 2013). Historically considered a by-product of illness, sickness behavior is now thought to be an adaptive response that helps an organism ght infection (Dantzer 2009).
The behavioral response to illness or infection is typically generalized to multiple different infectious pathogens, possibly due to the fact that shared mechanisms communicate peripheral infection to the brain, regardless of the infectious source (Dantzer 2009;Hennessy et al. 2014). In some organisms, even non-physical psychological or social stressors can induce sickness behaviors via these same mechanisms (Hodes et al. 2014). Thus, sickness behavior can re ect a cumulative physiological state that is the result of multiple different environmental stressors, acting alone or synergistically. As a result, behavioral predictors of infection may be particularly useful in species where multiple stressors interact to varying degrees to give rise to diseased states across populations.
Although behavior can serve as an indicator of illness, it can also re ect disease susceptibility in healthy individuals. For example, in healthy cattle, the behavioral response to management conditions, de ned as "temperament", is correlated with the strength of the immune response to infection (Burdick et al. 2011).
Stress can also result in differential activation of immune pathways in individuals with "proactive" versus "reactive" behavioral types (Proudfoot et al. 2012 McMenamin & Genersch 2015). Recent studies suggest that, from a mechanistic perspective, these stressors behave synergistically at the colony level in part because they target similar pathways involved in immune and stress response in individual worker bees (Doublet et al. 2017). This shared physiological response to health stressors raises the possibility that a common behavioral phenotype (i.e., a sickness behavior) may be associated with disease in this species. Previous studies in the honey bee have associated some behavioral responses with speci c infectious agents ( Alaux et al. 2009). Transcriptomic studies suggest that the brain molecular pro le associated with high aggression is consistent whether the source of behavioral variation is genetic or environmental (Alaux et al. 2009;Chandrasekaran et al. 2015;Rittschof et al. 2015b). Thus, high aggression could either serve as a predictor of disease resilience (e.g., if it is linked pleiotropically to immune function), but it may also be a response to infection (i.e., an environmentallyinduced sickness behavior). In the current study, we use a molecular approach to determine whether variation in aggression is correlated with a generalized immune response to infection in honey bees.
The diverse health outcomes associated with high aggression in the honey bee implicate a number of tissues including the brain as a regulator of behavior, the fat body, a metabolic tissue that is involved in immune response (Wheeler & Robinson 2014), and the midgut, which is involved in pesticide detoxi cation (Mao et al. 2013). Communication between peripheral, immune responsive tissues and the brain is characteristic of sickness behavior in vertebrates (Dantzer 2009), but in the context of honey bee aggression, no study has evaluated tissues other than the brain to determine the role of peripheral tissues in behavioral modulation. Here we sequence RNA extracted from the brain, fat body, and midgut of high and low-aggression individual worker bees. These individuals are siblings that differ in aggression as a result of their developmental experience (Rittschof et al. 2015a). Speci cally, siblings that developed in high-aggression colonies were more aggressive and more pesticide tolerant as adults compared to ones that developed in low-aggression colonies. In our analysis, we rst assess evidence of differential viral or bacterial infection in our samples, based on RNA abundance. We then determine whether genes differentially expressed as a function of aggression are signi cantly enriched for general disease-related transcripts identi ed in a recent meta-analysis (Doublet et al. 2017). We further assess overlapping genes for directional concordance based on the hypothesis that low aggression is a sickness behavior. We take a similar approach to evaluate alternative hypotheses, that differentially expressed genes are associated with social modulation of predator response, or that they are associated with behavioral maturation, the process by which adult honey bees progress through different behavioral tasks as they age (Ament et al. 2010). Although this study is correlative, it is a critical step towards determining whether low aggression predisposes bees to mortality risk, or manifests as a sickness behavior, through changes in gene expression.

Methods
Honey bee tissue samples Samples for sequencing were a subset of specimens from a previously published study in which we showed that sibling workers housed in high-aggression hives as pre-adults (through their egg, larval, and pupal stages) are more aggressive as adults compared to siblings housed in low-aggression hives. The more aggressive individuals also show increased pesticide tolerance. We demonstrated that behavioral effects were robust across 18 unique colonies (9 high and 9 low aggression) using sibling workers derived from 15 queens. This sample re ects three different experiments conducted across two years and two geographic locations, Illinois and Pennsylvania, at three times during the summer (Rittschof et al. 2015a).
During summer 2014, sibling worker bee eggs were introduced to either a high or low aggression hive when they were 0-24 h old. The workers remained in these hives during their egg, larval, and pupal stages.
The samples used in the current study were siblings from a single queen kept in one high and one low aggression hive. The two hives had equivalent mite loads (5 mites per colony, measured on a sticky board; Rittschof et al. 2015a). Because the queen mother of these siblings was outbred and naturallymated (honey bee queens mate with 17-20 males; Tarpy et al. 2015), individuals were a mixture of full and half-siblings, and thus results are generalized to more than one genetic background. Workers were removed from the hives one day prior to adult emergence and allowed to emerge in a laboratory incubator kept at 34ºC. Once workers emerged, they were kept in large groups (~70 individuals) and fed 50% sucrose until they were 8 days old. On this day, they were killed in a -20ºC freezer and transferred to a -80ºC freezer for long-term storage.
We dissected brains and midguts by submerging heads and abdominal tissues in chilled RNAlater ICE (Thermo Fisher Scienti c Waltham, MA, USA) Galbraith et al. 2015). Additional tissues (e.g., the sting apparatus) were removed from the abdomen, and fat body RNA was extracted directly from the tissue that remained adhered to the abdominal cuticle. We extracted RNA using the Aurum Fatty and Fibrous RNA kit (Bio-Rad, Hercules, CA, USA, includes on-column DNA digestion). Brains were homogenized using a handheld motorized pestle, while midgut and fat body were homogenized with a bead homogenizer (MP Biomedicals, Santa Ana, CA, USA). RNA was quanti ed on a plate reader (ClarioStar, BMG Labtech, Ortenberg, Germany) and Bioanalyzer instrument (Agilent Technologies, Santa Clara, CA, USA). Where possible, we retained samples for sequencing for which we had all three tissues from a single individual, and where the RNA Integrity Number was greater than 7. The nal sequencing results include N=11 individuals from each colony with all three tissues sequenced, and N=1 low aggression and N=2 high aggression individuals with the brain and midgut only sequenced (72 samples total).
Sequencing, mapping, and differential expression analysis Samples averaged 89.6% alignment success (~30 million reads per sample). Reads were also assessed for the presence of common honey bee pathogens (see "Pathogen assessment" below). We used the estimateDisp, glmQLFit, and glmQLFTest functions in EdgeR (v.3.24.3) to evaluate differential expression as a function of hive aggression on a per-tissue basis.  2017), BeeBase IDs identi ed in our current study were converted to RefSeq IDs using NCBI Batch Entrez. Differences in gene identities and methods across studies decreased the size of the gene universe for enrichment analyses, and all analyses accounted for this change. We performed hypergeometric tests for enrichment using the phyper function in R (Wheeler & Robinson 2014). Tests for signi cant bias in direction of differential expression were performed using the binom.test function in R.

Pathogen assessment
We evaluated the relationship between pathogen presence and aggression by estimating the abundance of previously identi ed honey bee pathogens with our RNAseq data. Reads from each specimen were mapped to a database of known honey bee pathogens with sequenced genomes. This database consisted of the ve bacterial pathogens Melissococcus plutonius (GCF_000747585. and a single RPKM value was calculated for each pathogen genome for each bee specimen. Wilcoxon rank-sum tests were then used to calculate differences in RPKM estimates in each tissue type between high and low aggression hives. Results were corrected for multiple testing (18 total tests) using the Benjamini-Hochberg approach. We also performed χ2 tests for each pathogen to determine if their presence, rather than abundance, was associated with aggressive behavior. The pathogen was counted as present if its RPKM value was greater than the 10 th percentile of the RPKM's across all samples for that pathogen. Again, the resulting p-values were corrected using Benjamini-Hochberg.

Differential expression analysis
We analyzed differential gene expression as a function of aggression on a per-tissue basis. 85, 1571, and 312 genes were differentially expressed in the brain, fat body, and midgut tissues, respectively (Tables S1-S3). Genes in the brain were signi cantly biased towards up-regulation in low aggression bees (81%, binomial test, P<0.0001), while direction of expression was not signi cantly biased in the fat body (49% upregulated, binomial test, P=0.27) or midgut (55%, binomial test, P=0.07).
To describe the function of genes differentially expressed as a function of aggression, we performed a Gene Ontology (GO) analysis followed by a REViGO analysis of signi cant GO terms (Benjamini-Hochberg corrected P < 0.05). REViGO clusters GO terms on the basis of semantic similarity to identify major patterns in long GO term lists (Supek et al. 2011). Differentially expressed genes in the brain were signi cantly enriched for 23 GO terms (Table S4). The REViGO clustering analysis showed clusters of processes and functions related to chaeta morphogenesis, disaccharide transport, and RNA polymerase II regulatory region sequence-speci c DNA binding. These results suggest strong roles for transcriptional regulation, sensory development, and carbohydrate metabolism in differentiating brain gene expression pro les for high versus low aggression bees. Differentially expressed fat body genes were signi cantly enriched for 188 terms (Table S5), including processes and functions associated with nucleotide and energy metabolism, and transporter activity. Only one GO category, toxin activity, was signi cantly enriched among differentially expressed midgut genes.
All pairwise tissue comparisons showed some overlap in terms of genes differentially expressed as a function of aggression, with the strongest similarities between the midgut and fat body. Eight genes were differentially expressed in both the fat body and brain (enrichment test for signi cant overlap, P=0.79), and seven of eight genes showed the same direction of change as a function of aggression (binomial test, P=0.07). For the brain and midgut, six genes overlapped (P=0.006) with ve of six genes showing the same direction of change (binomial test, P=0.22). 76 genes overlapped between the fat body and midgut (hypergeometric test, P<0.0001), with 71 showing the same direction of regulation across these two tissues (binomial test, P<0.0001), suggesting robust tissue expression similarity across these tissues.
Only a single gene, a homeobox transcription factor (GB51409) was differentially expressed across all three tissues.

Evidence that low aggression is a sickness behavior
Are low aggression bees infected with a pathogen?
We detected ve bacterial pathogens, four fungal pathogens, deformed wing virus, and acute bee paralysis virus in all three tissues in at least one individual in our study (Table 1). No pathogen was detected in every individual in any tissue, but most pathogens were present in at least one tissue in one individual. No pathogen was signi cantly more abundant or more likely to be present in low aggression samples (Table S6-S8), suggesting molecular differences as a function of aggression were not caused by acute pathogen infection.

Do low aggression bees show evidence of heightened immune activity?
To evaluate whether the molecular patterns associated with low aggression resemble a diseased state, we compared our differentially expressed gene lists with a recently published meta-analysis that identi ed genes for which expression changed in response to immune challenge across a variety of tissue types and combinations of tissues, including the whole bee, whole abdomen, fat body, midgut, and brain (Doublet et al. 2017). This meta-analysis identi ed 57 genes consistently upregulated and 110 genes consistently downregulated in response to immune challenge (across a range of tissue types, see METHODS), whether the source was parasitic mite feeding, viral or fungal infection, or some combination. We performed two enrichment tests per tissue type in our study, evaluating signi cance in overlap between our differentially expressed gene lists and the up and down-regulated genes from Doublet et al. (2017). We also evaluated directional concordance, with the hypothesis that genes upregulated with infection would be upregulated in low aggression bees if it is a phenotype associated with disease.
In the brain, only one differentially expressed gene overlapped with the Doublet et al. (2017) upregulated gene list, signi cant overlap due to the relatively small number of differentially expressed genes in this tissue (particularly after list conversion, see METHODS, hypergeometric test, P=0.03). This single gene, GB42523 (an uncharacterized non-coding RNA), was upregulated in low aggression bees, consistent with a hypothesis of a sickness phenotype. Two genes overlapped with the downregulated Doublet et al. list (P=0.01). GB45913 (lethal (2) essential for life, related to heat-shock proteins) was downregulated in low aggression bees, while the second, GB50116 (chymotrypsin inhibitor) was upregulated in low aggression bees.
In the fat body, 13 genes overlapped with the 56 upregulated genes in the Doublet et al. list (Table 2). This overlap was statistically signi cant (hypergeometric test, P=0.04). Moreover, 10 of the 13 genes were upregulated in low aggression bees, 77% directional concordance with the hypothesis that the fat body molecular signature of low aggression resembles sickness (a signi cant directional bias, binomial test, P<0.05). 17 genes overlapped with the downregulated Doublet et al. list (out of 110), but this was not statistically signi cant (P=0.39), nor was the degree of directional concordance (Table 3,

Do low aggression bees show differences in predator-responsive genes?
The pre-adult developmental environment could cause low aggression by modulating the baseline expression of genes that are responsive to alarm and predator cues. To test this possibility, we compared our list of genes differentially expressed in the brain as a function of aggression to genes differentially expressed following alarm pheromone exposure (Alaux et al. 2009), which induces a rapid, aggressive anti-predator response. Two genes (GB40074, hormone-like receptor in 38 and GB45913, protein lethal(2)essential for life) overlapped, a non-signi cant result (P=0.09).
Do pre-adult and adult colony environment effects on aggression share a molecular signature? High-aggression "Africanized" honey bees are genetically distinct from relatively docile honey bees of European origin. Using a series of experiments that involved housing adult worker bees of Africanized or European origin in either Africanized or European colonies for several weeks, Alaux et al. (2009) found that certain genes in the brain are differentially expressed as a consequence of colony environment, irrespective of individual genotype. This social treatment also affected expression of aggression (Hunt et al. 2003;Alaux et al. 2009). We compared genes differentially expressed as a function of adult colony environment to those differentially expressed as a function of aggression in our study to determine if similar genes are regulated by the adult and pre-adult social environment. Four genes were shared across these lists (GB54316, cardioacceleratory peptide receptor, GB43805, membrane metallo-endopeptidase-like 1, GB41643, blue sensitive opsin, GB54675, uncharacterized), but this degree of overlap was not signi cant (P=0.19).

Do low aggression bees show a change in rate of adult behavioral maturation?
Adult workers shift tasks as they age, a process called behavioral maturation. This process is in uenced by social and environmental cues (Schulz et al. 1999 foraging tasks are typically more aggressive than younger hive bees, and accelerated transition to foraging is associated with higher aggression (Giray et al. 2000). Juvenile hormone regulates this behavioral progression as well as larval development, suggesting these processes could be mechanistically linked. Alternatively, stress tends to accelerate the transition to foraging, which may manifest as accelerated development in low aggression individuals in our study. To assess whether bees show differences in adult behavioral maturation as a function of pre-adult environment, we compared differentially expressed genes in the brain to those differentially expressed between foragers (older adult workers) and nurses (younger adult workers) (Alaux et al. 2009). We found that seven genes (Table 4) overlapped between these lists, a statistically signi cant result (P=0.01). Five out of seven genes showed directional concordance between low aggression bees and younger nurse bees, suggesting low aggression bees are developmentally delayed, however this relationship was not statistically signi cant (P=0.23).

Discussion
Our results show that variation in aggression in honey bees is associated with a molecular phenotype that broadly characterizes infection and parasitic feeding (Fig 1). We found signi cant enrichment for infection-responsive genes in all three tissues, and in the fat body, and to a lesser degree the midgut, we nd evidence of directional concordance consistent with the hypothesis that low aggression resembles a diseased or parasitized state. However, we found little evidence of acute infection in low aggression individuals; the abundance of infectious agents, as measured by the presence of pathogen-derived sequence reads, was not higher in these bees. We also used enrichment analyses to assess alternative hypotheses for the patterns in our molecular data. In doing so, we found little evidence that the brain molecular signature in the current study is enriched for genes modulated by either ephemeral or stable social cues that induce aggression in adults. Interestingly, we do see a signature of carbohydrate metabolism among genes differentially expressed in the brain in our study, consistent with studies linking glycolysis and oxidative phosphorylation to social and environmental modulation of aggression ( . Finally, enrichment analyses provide some support for the hypothesis that variation in aggression in our study re ects variation in the pacing of behavioral maturation in adults. Although our method for assessing pathogen infection is indirect and limited to a transcriptional signature in speci c tissues, at least some bacterial, fungal, and viral pathogens were found in every individual examined, suggesting that these data can be used to estimate infection load. Using these estimates, we nd no signi cant differences in the abundance of any pathogen between high and low aggression bees, indicating that the between colony variation in aggression is not the result of differences in infection rates. The set of pathogens we considered includes those that are known to commonly infect honey bees (Brutscher et  including Deformed Wing Virus, a version of which has been associated with aggression in a previous study (Fujiyuki et al. 2004; see also Rortais et al. 2006). This approach for estimating infection rates may be useful for studies of honey bee behavior moving forward; despite the use of polyA-enrichment for extracting mRNA, substantial numbers of both bacterial and viral reads were present in our RNAseq datasets.
In the brain, we found evidence that genes differentially expressed between high and low aggression siblings are signi cantly enriched for genes differentially expressed between nurse and forager worker bees (Whit eld et al. 2002; Alaux et al. 2009). Worker bees change tasks as they age, a process known as behavioral maturation. Young workers perform tasks inside the hive including nursing, while older bees perform tasks outside of the hive including energetically-demanding foraging and defensive behaviors (Winston 1987). Thus, our results suggest that the pre-adult developmental environment, and resulting variation in aggression and pesticide tolerance, is associated with variation in adult developmental pacing. Older bees are typically more aggressive, and in keeping with this, a majority of overlapping genes support the hypothesis that high aggression bees show accelerated behavioral maturation, although this directional bias was not signi cant. Aggression is modulated by the social environment experienced throughout adulthood, but we found little overlap with the molecular signature of this effect. In adults, genes rapidly modulated by alarm pheromone, an aggression inducing social cue, and genes modulated by long-term residence in a highly aggressive colony show signi cant overlap (Alaux et al. 2009), but neither of these sets of genes overlapped with those modulated by aggression experienced during pre-adult development. This could be related to the difference in stability of social effects experienced at these two different life stages. Socially-induced changes in aggression during adulthood are reversible (Alaux & Robinson 2007;Rittschof 2017;Shpigler et al. 2017b), while effects induced during the pre-adult stages are relatively stable, present 1 week into adulthood, even when individuals were kept in a common laboratory environment (Rittschof et al. 2015a). Alternatively, variation in aggression associated with sickness may have a fundamentally different molecular signature, for example, if sickness affects only a subset of neuronal populations that regulate aggression (Kabelik et al. 2009).
In the current results, changes in brain molecular state are accompanied by shifts in gene expression in both the fat body and midgut. This result is consistent with patterns of sickness behavior in other animals, where molecular signals of peripheral infection impact aggression-relevant signaling in the brain (Nelson & Chiavegatto 2001). In the honey bee, no previous study of aggression has assessed molecular variation in peripheral tissues, although recent work suggests there may be some common master regulatory genes associated with age-related behavioral changes across diverse tissues in the honey bee (Ament et al. 2012;Johnson & Jasper 2016). In our results, brain gene expression changes were modest relative to the fat body and midgut, and perhaps as a result, we found only a single gene that was differentially expressed across all three tissues. Because this gene, GB51409, is a homeobox transcription factor (Nkx-6.1), it may indeed serve as a master regulator of molecular state. However, it was not identi ed as such across the age-related tissue comparison in Johnson and Jasper (2016). Particularly comparing the fat body and midgut, genes that were differentially expressed as a function of aggression across more than one tissue showed concordance in expression direction, consistent with the possibility that a systemic signal is regulating tissue molecular state similarly across tissues. Future work can investigate correlated expression across tissue types, the factors that coordinate the infection-like molecular state across tissues, and the relationship between baseline aggression and susceptibility to infection.
Aggression is easy to rapidly assess at the colony level (Rittschof et al. 2015a); future work should consider how it is related to other phenotypes that impact hive success. Aggression is an energy-intensive high-performance phenotype often correlated with foraging activity at the colony level, suggesting foraging effort may shift concurrently with changes in aggression. Foraging behavior is impacted by individual health, but it is also modulated by social cues, raising the possibility that social responsiveness is altered in low-aggression or diseased individuals. A recent study in honey bees showed that individuals exhibit different levels of social responsiveness, showing high or low levels of response to cues, whether these cues are characteristic of young adult (nursing) or older adult (nest defense) behavioral tasks (Shpigler et al. 2017a). Similarly, honey bees who have experienced chronic stress are less likely to exhibit aggressive behaviors in certain social contexts (Rittschof 2017). Behavioral variation could re ect individual variation in response threshold to, e.g., olfactory stimuli, in social contexts. In keeping with this idea, we nd that differentially expressed genes as a function of aggression in the current study are enriched for processes related to sensory development.

Page 13/23
The relationship between social behaviors and sickness is complex: social organisms have high levels of conspeci c contact, and as a result, many have evolved forms of social immunity, where social interactions are used to prevent or respond to the presence of infectious agents in a social group (Hennessy et al. 2014). Conversely, because social interactions also transmit disease, individuals may avoid or otherwise reject the presence of infected individuals ( Author contributions: CCR designed the study, collected specimens, conducted bioinformatics analyses associated with differential gene expression and enrichment analyses, and wrote the manuscript. BER conducted the informatics assessment of pathogen presence as well as the gene ontology analysis, and participated in manuscript writing. JHP conducted molecular sample preparation. of brain and behavior in the honey bee. Genome Res 12, 555-66.   Figure 1 Results of enrichment analyses for genes differentially expressed as a function of aggression. Results of analyses comparing genes differentially expressed as a function of aggression to genes differentially expressed as a result of infection (in the brain, midgut, and/or fat body). Further analyses for the brain only (comparison with Alaux et al. 2009 microarray study), evaluated overlap with genes differentially expressed following exposure to aggression-inducing alarm cues (predator threat), adult behavioral changes with aging (behavioral maturation), and exposure to a high versus low aggression environment as an adult (adult environment). Signi cant enrichment is indicated by a dotted circle. Gene numbers listed for each tissue sum to the total differentially expressed genes in the current study, not the total genes incorporated in the enrichment analyses; gene conversions across studies, spanning multiple genome versions, gene sets, and gene expression analysis methods, decreased the universe of genes used for enrichment analyses.