The Polycomb Repressive Complex 2.1 Links the Food Environment to a Persistent Neural State

Interactions between genes and environment sculpt the responses of cells to environmental stimuli. In neuronal cells this process can lead to long term changes in the behavioral repertoire of animals, which in turn impacts disease risk. Here we show that the Polycomb Repressive Complex 2.1 (PRC2.1) modulates the physiology of sweet gustatory neurons and the taste behavior of ​D. melanogaster​ fruit flies in response to the food environment. A high sugar diet caused a redistribution of PRC2.1 chromatin occupancy resulting in the repression of a transcriptional network required for the responsiveness of the gustatory neurons to sweet stimuli. These changes led to lower sweet sensation, which in turn promoted obesity. Nearly half of the transcriptional changes mediated by PRC2.1 on a sugar diet persisted when animals were moved back to a control diet, causing a permanent decrease in sweet taste that was dependent on the constitutive activity of PRC2.1. Thus, our results point to a novel mechanism involved in modulating neural plasticity, behavior, and disease in response to the food environment. Introduction In 1958 D. L Nanney coined the term “epigenetic control system” to explain phenotypic variation that was not dependent on the “genetic library” ​(Nanney, 1958)​. More than half a century later, we understand that e ​pigenetic mechanisms sculpt cellular responses to environmental variation via changes in gene expression ​(Allis and Jenuwein, 2016)​.​ These responses show varying degrees of stability, and some phenotypes persist even after the environmental signals disappear, forming a memory of past experiences. This process has far-reaching implications for all cells, but it is particularly critical for neurons, because their molecular properties direct the physiology and behavior of the whole organism. Indeed, environmental conditions that permanently alter neurons reshape the way in which future stimuli are perceived and processed, limiting the behavioral repertoire of animals and influencing disease risk ​(Campbell and Wood, 2019; Dulac, 2010; Sweatt, 2013)​. The effects of environmental reprogramming on behavioral plasticity have been elegantly studied in the context of maternal care, learning and memory, and addiction ​(Champagne, 2008; Dulac, 2010; Nestler and Lüscher, 2019; Sweatt, 2013)​, which has led to the identification of some of the molecular mechanisms of anxiety, depression, and cognitive ability ​(Campbell and Wood, 2019)​. However, because these studies were done in whole brain regions such as the hippocampus and striatum, we still do not know how environmental signals impact the function of specific neural circuits to persistently reprogram behavior. Further, the mechanisms through which other biomedically important environments – such as diet and pollutants – alter brain physiology and behavior remain largely undefined. Here we studied the effects of high dietary sugar on taste sensation as a model to identify the molecular mechanisms through which the food environment reprograms neural physiology and behavior. Like many behaviors, the ability of animals to detect food sources is under the control of both genes and environment. The genome shapes the sensory apparatus of each species to match their unique ecological niche, but despite this genetic constraint, this system is plastic and can be altered by diet composition ​(Beauchamp and Jiang, 2015)​. Indeed, diet has been shown to change the taste perceptions of animals from insects to humans ​(Bertino et al., 1982; Glendinning et al., 2001; May et al., 2019a; Stewart and Keast, 2012; Wang et al., 2016; Wise et al., 2016; Zhang et al., 2013)​, and across organisms, sensory alterations affect food preference and intake, influencing the development of conditions like obesity and heart disease (Bertino et al., 1982; May et al., 2019a; Stewart and Keast, 2012)​. We recently found that high dietary sugar dulls the responses of the fruit fly taste neurons to sweet stimuli, causing higher food intake and weight gain ​(May et al., 2019a, 2019b)​. Mammals fed high nutrient diets also show changes in taste, neural responses, and food preferences ​(Ahart et al., 2019; Chen et al., 2010; Crow, 2012; Kaufman et al., 2018; Maliphol et al., 2013; May et al., 2019a; McCluskey et al., 2020; Sartor et al., 2011; Weiss et al., 2019)​, arguing that the effects of diet on taste are conserved and promote obesity. However, the molecular mechanisms through which the food environment alters taste sensation, and more broadly, neural physiology, are largely unknown. Here we exploited the exquisite genetics tools of the fruit fly and the relative simplicity of its sensory system to tackle these questions. We found that the chromatin silencing Polycomb Repressive Complex 2.1 (PRC2.1) tuned the activity of the sweet sensory neurons and taste sensation to the food environment by repressing a transcriptional program that shapes the synaptic, signaling, and metabolic properties of these cells. Interestingly, this diet-dependent transcriptional remodeling persisted even when animals were returned to the control diet, leading to lasting changes in sweet taste sensation that depended on the constitutive activity of PRC2.1. Together our findings suggest that the food environment reprograms sensory responses by activating epigenetic mechanisms that persistently restrict the perception of future stimuli, alter behavior, and increase the risk for obesity and metabolic disease. Results PRC2.1 modulates sweet taste in response to the food environment Drosophila melanogaster​ fruit flies fed high dietary sugar experience lower sweet taste sensation as a result of the decreased responsiveness of the sweet sensory neurons to sugar stimuli ​(May et al., 2019a)​. Given the importance of sensory cues to control eating, and recent data that diet also impacts taste in mammals ​(Ahart et al., 2019; Chen et al., 2010; Kaufman et al., 2018; Maliphol et al., 2013; May et al., 2019a; McCluskey et al., 2020; Weiss et al., 2019)​, we set out to identify the molecular mechanisms through which the food environment shapes sensory responses. Since sweet taste deficits develop within 2-3 days upon exposure to the high sugar diet and independently of weight gain ​(May et al., 2019a)​, we reasoned that gene regulatory mechanisms may be involved in modulating the responses of the sensory neurons. To test this hypothesis we conducted a screen for gene regulatory factors necessary for sweet taste defects on a high sugar diet. To do this, we fed control (​w1118 ​) and mutant flies a control diet (CD, ~5% sucrose) or a diet supplemented with 30% sucrose (sugar diet, SD) for 7 days and then tested their ability to taste using the Proboscis Extension Response ​(Shiraiwa and Carlson, 2007)​. This behavioral assay measures taste responses by quantifying the amount of proboscis extension (0= no extension, 0.5=partial extension, 1= full extension) when the fly labellum – where the dendrites and cell bodies of the taste neurons are located (Fig. S1A) – is stimulated with three different concentrations of sucrose (30%, 10%, 5%); this generates a taste curve where flies respond more intensely to higher sugar stimuli (Fig. 1B, gray circles). Flies fed a sugar diet show a marked decrease in PER compared to control diet flies (Fig. 1B, gray squares); however, mutants for the core Polycomb Repressive Complex 2 (PRC2) – which includes the histone 3 lysine 27 (H3K27) methyltransferase ​Enhancer of Zeste ​(​E(z)​), and the obligate accessory factors ​Suppressor of zeste 12, ​(​Su(z)12 ​) and ​extra sex combs, ​(​esc​) (Fig. 1A) – had largely the same proboscis extension response (PER) on a control and sugar diet (Fig. 1B, right, red shades). To confirm the role of PRC2 in taste sensation, we supplemented the control and sugar diet with EED226, a PRC2 inhibitor (herein referred to as EEDi) that destabilizes the core complex by binding to the tri-methyl H3K27 (H3K27me3) binding pocket of EED (the homologue of ​esc​ in ​M. musculus​) ​(Qi et al., 2017)​. While animals fed a sugar diet plus vehicle (10% DMSO) experienced lower PER, those fed a SD+EEDi retained normal sweet taste responses (Fig. 1C), consistent with results from the PRC2 mutants. Thus, mutations and inhibition of PRC2 rescue the blunting of sweet taste that occurs in the high sugar food environment. In flies PRC2 forms two main subcomplexes, PRC2.1 and PRC2.2, which contain distinct accessory factors that influence the targeting of the core complex to the genome (Laugesen et al., 2019)​. Mutations in the ​Polycomb-like ​ (​Pcl ​) gene, the accessory factor to PRC2.1, phenocopied PRC2 mutants and rescued sweet taste deficits in flies fed a sugar diet (Fig. 1D). In contrast, flies with deficits in the PRC2.2-members ​Jumonji, AT rich interactive domain 2 ​ (​Jarid2 ​) and ​jing ​ still showed a blunting of sweet taste responses in flies fed a sugar diet (Fig. S1B). Interestingly, members of the Polycomb Repressive Complex 1 (PRC1) and the recruiter complex PhoRC were also not required for taste changes in responses to a sugar diet (Fig. S1C and Fig. S1D-E). Thus, the PRC2.1 complex is necessary for the sensory changes that occur in the high sugar environment. We next asked if PRC2.1 is required specifically in the sweet sensory neurons to decrease their responses to sweet stimuli on the sugar diet. To do this, we used the GAL4/UAS system to knock down ​Pcl ​ in the sweet taste neurons using the ​Gustatory receptor 5a ​GAL4 driver​, Gr5a-GAL4, ​which labels ~60 cells in the proboscis of adult flies ​(Chyb et al., 2003)​; we selected ​Pcl ​ to narrow the effect to the PRC2.1 complex. Knockdown ​ ​of​ Pcl ​ in ​Gr5a+​ neurons using two independent RNAi transgenes (50% knockdown efficiency, Fig. S2A) prevented sweet taste deficits in animals fed a sugar diet (Fig. 1E, and Fig. S2B). ​Pcl ​ knockdown, however, had no effect on a control diet (Fig. S2C), in accordance 


Introduction
In 1958 D. L Nanney coined the term "epigenetic control system" to explain phenotypic variation that was not dependent on the "genetic library" (Nanney, 1958) . More than half a century later, we understand that e pigenetic mechanisms sculpt cellular responses to environmental variation via changes in gene expression (Allis and Jenuwein, 2016) . These responses show varying degrees of stability, and some phenotypes persist even after the environmental signals disappear, forming a memory of past experiences. This process has far-reaching implications for all cells, but it is particularly critical for neurons, because their molecular properties direct the physiology and behavior of the whole organism. Indeed, environmental conditions that permanently alter neurons reshape the way in which future stimuli are perceived and processed, limiting the behavioral repertoire of animals and influencing disease risk (Campbell and Wood, 2019;Dulac, 2010;Sweatt, 2013) . The effects of environmental reprogramming on behavioral plasticity have been elegantly studied in the context of maternal care, learning and memory, and addiction (Champagne, 2008;Dulac, 2010;Nestler and Lüscher, 2019;Sweatt, 2013) , which has led to the identification of some of the molecular mechanisms of anxiety, depression, and cognitive ability (Campbell and Wood, 2019) . However, because these studies were done in whole brain regions such as the hippocampus and striatum, we still do not know how environmental signals impact the function of specific neural circuits to persistently reprogram behavior. Further, the mechanisms through which other biomedically important environments -such as diet and pollutants -alter brain physiology and behavior remain largely undefined. Here we studied the effects of high dietary sugar on taste sensation as a model to identify the molecular mechanisms through which the food environment reprograms neural physiology and behavior. Like many behaviors, the ability of animals to detect food sources is under the control of both genes and environment. The genome shapes the sensory apparatus of each species to match their unique ecological niche, but despite this genetic constraint, this system is plastic and can be altered by diet composition (Beauchamp and Jiang, 2015) . Indeed, diet has been shown to change the taste perceptions of animals from insects to humans (Bertino et al., 1982;Glendinning et al., 2001;May et al., 2019a;Stewart and Keast, 2012;Wang et al., 2016;Wise et al., 2016;Zhang et al., 2013) , and across organisms, sensory alterations affect food preference and intake, influencing the development of conditions like obesity and heart disease (Bertino et al., 1982;May et al., 2019a;Stewart and Keast, 2012) . We recently found that high dietary sugar dulls the responses of the fruit fly taste neurons to sweet stimuli, causing higher food intake and weight gain (May et al., 2019a(May et al., , 2019b . Mammals fed high nutrient diets also show changes in taste, neural responses, and food preferences (Ahart et al., 2019;Chen et al., 2010;Crow, 2012;Kaufman et al., 2018;Maliphol et al., 2013;May et al., 2019a;McCluskey et al., 2020;Sartor et al., 2011;Weiss et al., 2019) , arguing that the effects of diet on taste are conserved and promote obesity. However, the molecular mechanisms through which the food environment alters taste sensation, and more broadly, neural physiology, are largely unknown.
Here we exploited the exquisite genetics tools of the fruit fly and the relative simplicity of its sensory system to tackle these questions. We found that the chromatin silencing Polycomb Repressive Complex 2.1 (PRC2.1) tuned the activity of the sweet sensory neurons and taste sensation to the food environment by repressing a transcriptional program that shapes the synaptic, signaling, and metabolic properties of these cells. Interestingly, this diet-dependent transcriptional remodeling persisted even when animals were returned to the control diet, leading to lasting changes in sweet taste sensation that depended on the constitutive activity of PRC2.1. Together our findings suggest that the food environment reprograms sensory responses by activating epigenetic mechanisms that persistently restrict the perception of future stimuli, alter behavior, and increase the risk for obesity and metabolic disease.

PRC2.1 modulates sweet taste in response to the food environment
Drosophila melanogaster fruit flies fed high dietary sugar experience lower sweet taste sensation as a result of the decreased responsiveness of the sweet sensory neurons to sugar stimuli (May et al., 2019a) . Given the importance of sensory cues to control eating, and recent data that diet also impacts taste in mammals (Ahart et al., 2019;Chen et al., 2010;Kaufman et al., 2018;Maliphol et al., 2013;May et al., 2019a;McCluskey et al., 2020;Weiss et al., 2019) , we set out to identify the molecular mechanisms through which the food environment shapes sensory responses. Since sweet taste deficits develop within 2-3 days upon exposure to the high sugar diet and independently of weight gain (May et al., 2019a) , we reasoned that gene regulatory mechanisms may be involved in modulating the responses of the sensory neurons.
To test this hypothesis we conducted a screen for gene regulatory factors necessary for sweet taste defects on a high sugar diet. To do this, we fed control ( w1118 CS ) and mutant flies a control diet (CD, ~5% sucrose) or a diet supplemented with 30% sucrose (sugar diet, SD) for 7 days and then tested their ability to taste using the Proboscis Extension Response (Shiraiwa and Carlson, 2007) . This behavioral assay measures taste responses by quantifying the amount of proboscis extension (0= no extension, 0.5=partial extension, 1= full extension) when the fly labellum -where the dendrites and cell bodies of the taste neurons are located (Fig. S1A) -is stimulated with three different concentrations of sucrose (30%, 10%, 5%); this generates a taste curve where flies respond more intensely to higher sugar stimuli (Fig. 1B, gray circles). Flies fed a sugar diet show a marked decrease in PER compared to control diet flies (Fig. 1B, gray squares); however, mutants for the core Polycomb Repressive Complex 2 (PRC2) -which includes the histone 3 lysine 27 (H3K27) methyltransferase Enhancer of Zeste ( E(z) ), and the obligate accessory factors Suppressor of zeste 12, ( Su(z)12 ) and extra sex combs, ( esc ) (Fig.   1A) -had largely the same proboscis extension response (PER) on a control and sugar diet (Fig. 1B,right,red shades). To confirm the role of PRC2 in taste sensation, we supplemented the control and sugar diet with EED226, a PRC2 inhibitor (herein referred to as EEDi) that destabilizes the core complex by binding to the tri-methyl H3K27 (H3K27me3) binding pocket of EED (the homologue of esc in M. musculus ) (Qi et al., 2017) . While animals fed a sugar diet plus vehicle (10% DMSO) experienced lower PER, those fed a SD+EEDi retained normal sweet taste responses (Fig. 1C), consistent with results from the PRC2 mutants. Thus, mutations and inhibition of PRC2 rescue the blunting of sweet taste that occurs in the high sugar food environment.
In flies PRC2 forms two main subcomplexes, PRC2.1 and PRC2.2, which contain distinct accessory factors that influence the targeting of the core complex to the genome (Laugesen et al., 2019) . Mutations in the Polycomb-like ( Pcl ) gene, the accessory factor to PRC2.1, phenocopied PRC2 mutants and rescued sweet taste deficits in flies fed a sugar diet ( Fig. 1D). In contrast, flies with deficits in the PRC2.2-members Jumonji, AT rich interactive domain 2 ( Jarid2 ) and jing still showed a blunting of sweet taste responses in flies fed a sugar diet (Fig. S1B). Interestingly, members of the Polycomb Repressive Complex 1 (PRC1) and the recruiter complex PhoRC were also not required for taste changes in responses to a sugar diet . Thus, the PRC2.1 complex is necessary for the sensory changes that occur in the high sugar environment.
We next asked if PRC2.1 is required specifically in the sweet sensory neurons to decrease their responses to sweet stimuli on the sugar diet. To do this, we used the GAL4/UAS system to knock down Pcl in the sweet taste neurons using the Gustatory receptor 5a GAL4 driver , Gr5a-GAL4, which labels ~60 cells in the proboscis of adult flies (Chyb et al., 2003) ; we selected Pcl to narrow the effect to the PRC2.1 complex. Knockdown of Pcl in Gr5a+ neurons using two independent RNAi transgenes (50% knockdown efficiency, Fig. S2A) prevented sweet taste deficits in animals fed a sugar diet (Fig. 1E, and Fig. S2B). Pcl knockdown, however, had no effect on a control diet (Fig. S2C), in accordance with the observation that E(z) and Pcl mutants have no effect on taste on a control diet (Fig. 1B, C) and suggesting that these phenotypes are uncovered only by the high sugar food environment.
Since Pcl is thought to target the PRC2 core complex to chromatin (Laugesen et al., 2019) , we hypothesized that its overexpression may be sufficient to induce sweet taste deficits even in the absence of a high sugar food environment. Indeed, overexpression of Pcl specifically in the Gr5a+ neurons induced sweet taste deficits in flies fed a control diet compared to transgenic controls (Fig. 1F). The effects of Pcl overexpression were abolished by treatment with the PRC2 inhibitor EEDi (Fig. 1G), arguing that Pcl causes sweet taste deficits entirely through the action of PRC2 and not through some yet unidentified mechanism. Importantly, Pcl overexpression had no effect on the number of Gr5a+ neurons in the proboscis (Fig. S2D), and so the taste deficits cannot be attributed to a decrease in the number of cells. To exclude the possibility that the effects of manipulating Pcl were developmental, we used the temperature sensitive tubulin-GAL80 ts transgene to limit expression of UAS-Pcl and Pcl RNAi-1 only in adult flies. Switching the flies to the non-permissive temperature and diet 4 days post eclosion, resulted in the same effects on sweet taste as using the Gr5a-GAL4 alone (Fig. S2E). Together, these experiments establish that PRC2.1 is required cell-autonomously in the Gr5a+ neurons to mediate the effects of a high sugar diet on sweet taste.

Pcl mutant flies have normal sensory responses and are resistant to diet-induced obesity
Flies on a high sugar diet have lower sweet taste because the neural responses of the taste neurons to sweet stimuli are decreased (May et al., 2019a) . Since Pcl mutants have identical taste on a control and sugar diet, we hypothesized that the responses of the sensory neurons to sucrose stimulation should also be similar. To test this, we expressed the genetically encoded presynaptic calcium indicator UAS -GCaMP6s- Brp-mCherry (Kiragasi et al., 2017) in the sweet sensing neurons and measured their in vivo responses to stimulation of the proboscis with 20% sucrose in Pcl mutant animals ( Fig. 2A ). Indeed, the responses to sucrose stimulation were identical in Pcl mutant flies fed a control diet and sugar diet (Fig. 2B), matching the behavioral data ( Fig. 1); importantly, this rescue was not due to an increase in the number of sweet taste cells (Fig. 2C).
We previously showed that restoration of sweet taste neuron activity in flies fed high dietary sugar protected them from diet-induced obesity (May et al., 2019a) . Since Pcl mutants abolished the deficits in neural and behavioral responses to sweetness in animals fed a high sugar diet, we anticipated that they should also prevent a diet-dependent increase in triglycerides. Indeed, sugar-diet flies with knockdown of Pcl in the Gr5a+ neurons remained as lean as animals on a control diet (Fig. 2D), while triglycerides increased in control flies fed a sugar diet (Fig. 2D). Importantly, there was no difference in the levels of triglycerides between control and Pcl knockdown flies fed a control diet (Fig. 2D). Together, these data suggest that, in response to the food environment, Pcl modulates the responsiveness of the sweet gustatory neurons to promote diet-induced obesity.

Pcl chromatin occupancy is redistributed in the high sugar environment
Our experiments show that PRC2.1 plays a critical role in the neural activity, behavior, and the metabolic state of animals exposed to the high sugar food environment. To identify the molecular mechanisms underlying these phenotypes, we measured the chromatin occupancy of 20 hours after they had been exposed to a control or sugar diet for 3 days (Fig. 3A). We selected this time point because we previously showed that sweet taste defects developed within 3 days of exposure to the sugar diet (May et al., 2019a) .
Most of the variation in the biological replicates of Dam::Pcl normalized to Dam alone (see Methods) was due to diet ( Fig. S3A), consistent with the high Pearson correlations within each dietary condition (Fig. S3B). Further, the accessibility profile of Dam at the Gr5a sweet taste receptor gene promoter was high, while that at the Gr66a bitter taste receptor promoter -which is only expressed in bitter cells, closely located near the sweet cells-was low (Fig. 3B), suggesting that the transgenes were appropriately targeted to the sweet taste neurons and that the limited induction controlled for background DNA methylation.
We first analyzed Pcl chromatin occupancy in the Gr5a+ neurons of flies on a control diet by comparing our data to a previous study that annotated five major chromatin types in D.  S4). While the large majority of genes differentially bound by Pcl were in the gene regulation category (80%), the pathway enrichment analysis also uncovered a few metabolism Gene Ontology (GO) terms ( Fig. 3H and Fig. S4). In summary, we found that PRC2.1 targeted transcription factors involved in neuronal processes and development in the Gr5a+ neurons, and its chromatin occupancy was redistributed at these loci in the high sugar diet environment. This redistribution could result in changes in the expression of these transcription factors and their targets, and in turn, affect the sensitivity of the sensory neurons and taste sensation.

PRC2.1 sculpts the transcriptional responses of the Gr5a+ neurons in response to diet
To test the hypothesis that redistribution of PRC2.1 chromatin occupancy alters the physiology of the sweet sensing neurons by remodeling transcription, we used Translating mRNA Affinity Purification (TRAP) (Chen and Dickman, 2017) ) to isolate mRNAs associated with the ribosomes of these cells. (Fig. 4A). To capture the dynamics of this process, we collected samples from age-matched Gr5a>Rpl3-3XFLAG flies fed a sugar diet for 3 and 7 days ( Fig. S5A). We first verified that this technique selected for mRNAs in the Gr5a+ neurons alone by quantifying the normalized read counts ( Gr5a+ /input) for three sweet taste receptor genes Gr5a , Gr64f , and Gr64a, which are expressed in cells labeled by the Gr5a -GAL4 . Indeed, these transcripts were enriched in the Gr5a+ fraction compared to the input (Fig. S5B), while the opposite was true for the bitter receptor gene Gr66a , which is expressed only in the bitter sensing neurons the taste sensilla (Scott, 2018) (Fig. S5B).
Notably, we observed a large negative skew in gene expression in the Gr5a + neurons of flies fed a sugar diet for 3 (SD3, mint; compared to the control diet) and 7 days (SD7, teal; compared to the control diet) ( Fig. 4B S6 and Fig. S7). GO terms for neuron-specific processes, such as dendritic membrane, sensory perception of chemical stimulus, and presynaptic/vesicle transport, were enriched at both timepoints ( Fig. S6 and Fig. S7), suggesting that a high sugar diet may alter the physiology of the sensory neurons through these pathways. Interestingly, flies fed a sugar diet for 7 days also showed changes in GO terms linked to neurodevelopmental processes, such as asymmetric neuroblast division and neuron projection morphogenesis, indicating that longer exposure to the diet led to additional alterations in neural function. GO terms associated with metabolic changes were present in higher numbers in flies fed a sugar diet for 7 days, consistent with the findings that longer exposure to the high sugar diet leads to higher fat accumulation (May et al., 2019a) .
Finally, we observed changes in "regulatory" GO terms such as transcription factor and corepressor, consistent with our chromatin binding experiments (TaDa). Thus, consumption of a high sugar diet altered neural, regulatory, and metabolic genes in the Gr5a+ cells.
To determine the role of PRC2.1 in these changes, we performed the transcriptional profiling experiments in the Gr5a+ neurons of Pcl c429 mutant animals fed a control diet and sugar diet for 7 days (CD and SD7) (Fig. S5C). Strikingly, the Pcl mutation abolished the negative skew ( Fig. S5D) and largely nullified the effects of the high sugar diet environment on gene expression. Specifically, of the genes repressed by a sugar diet (Fig. 4D, heatmap) 32% had a positive log 2 fold change ( Wald test , q < 0.1) and 76% were unchanged ( q <0.1, practical equivalence test using a null hypothesis of a change of at least 1.5-fold; see Methods for details ) between Pcl mutants fed a control and sugar diet. This effect was reflected in the GO analysis where terms changed by a high sugar diet in Pcl wild-type animals, such as dendritic membrane, sensory perception of chemical stimuli, synapse, and carbohydrate metabolic process, showed opposite trends in log 2 fold changes in Pcl mutants (Fig. S8). Thus, Pcl mutations abolished nearly all the gene expression changes induced by a high sugar diet consistent with their effects on behavior (PER, Fig.1 ), neural function ( in vivo calcium imaging, Fig. 2), and metabolism (triglycerides, Fig. 2). Together, these findings support the hypothesis that PRC2.1 tunes taste sensation to the food environment by influencing the expression of genes involved in the physiology of the sensory neurons.

PRC2.1 represses a transcriptional program required for sweet taste
We discovered that a high sugar diet environment repressed gene expression in the sweet sensory neurons, and that Pcl mutations almost entirely abolished its effect. This, together with the discovery that Pcl binding primarily changed at the enhancers of transcription factor genes on a sugar diet, suggests that Pcl redistribution may affect the expression of transcription factors that, in turn, control genes responsible for the overall responsiveness of these sensory neurons to sweetness. This idea was supported by the observation that Pcl -bound genes had lower expression levels than those not bound by it in the Gr5a+ neurons ( Fig. 4E), with many genes showing high binding and low expression (log 2 tpm <2, dark purple), while others having higher RNA read counts (log 2 tpm >5, light purple) (Fig. S5E). To further test this hypothesis, we looked for transcription factors that were directly bound by Pcl and that showed changes in gene expression on a sugar diet (Fig. S5F). This analysis revealed 5 transcription factors: four of these were activators -GATAe (Zn finger) , nubbin/pdm ( nub , POU homeobox) , Ptx1 (paired-domain homeobox), and caudal ( cad , hox-like homeobox)-which had higher Pcl binding (Fig. 5A) and lower mRNA levels on a sugar diet (Fig. 4F). The fifth transcription factor was the suppressor scarecrow ( Scro , NK-like homeobox), which had lower Pcl binding (Fig. 5A) and higher mRNAs levels on a sugar diet (Fig. 4F). Notably, mutations in Pcl reversed the effects of a high sugar diet on the expression of these 5 genes, suggesting that the binding of Pcl modulates their mRNA levels ( Given that the 4 activators are required for sweet taste sensation, we reasoned that they may control the expression of genes important for the proper function of the Gr5a+ neurons and normal sweet taste. To identify candidate target genes, we tiled the entire fruit fly genome using the motifs for each of these 5 transcription factors, converted the hits for each transcription factor to robust z-scores, and determined candidate regulatory targets based on estimates of the z-score threshold for binding in each case ( Fig. S10A; see Methods for details). We then flagged as "targets" the set of genes that had a putative binding site (exceeding our transcription factors-specific z-score cutoff) within a 2 kb region upstream of the annotated ORF start ( where they regulate one another and themselves to ensure stability of gene expression. Indeed we found that GATAe had binding sites in the promoters of all four regulators considered here ( cad , Scro , Ptx1 , nub ), in addition to binding its own promoter in an auto-regulatory loop ( Fig.   5E). Furthermore, Cad also targeted itself, Nub was one of Ptx1 targets, and Scro regulated both Cad and GATAe , forming a negative feedback loop with the latter (Fig. 5D). Thus, the 5 transcription factors differentially bound by PRC2.1 on a high sugar diet, form a hub that seems to regulate the properties of the Gr5a+ neurons. To gain a deeper understanding of these properties, we used pathway enrichment analysis on the regulons for each transcription factor.
GATAe targets, which comprise a large number of the genes regulated by the 4 other transcription factors, were enriched for GO terms involved in synaptic assembly and growth, terminal bouton, neural projection morphogenesis, and protein kinase regulation (summarized in To test the possibility that these targets form a functional network, we used STRING (Szklarczyk et al., 2019) and found a significant number of edges above the expected number ( protein-protein interaction enrichment of p < 1.0e-16) suggesting that the targets are indeed part of a functional and biologically connected network in the Gr5a+ neurons. We then used a subset of the neural targets to build a second network to identify candidate target genes that may play a direct role in neural physiology and sweet taste. This network showed strong interactions between genes involved in synaptic organization and signal transduction and their connection with the upstream regulators ( Fig. S14A). We chose two genes at the edges of the network, which are less likely to have redundant functions, the Adenylyl Cyclase X D ( ACXD ) gene (Ueno and Kidokoro, 2008)  To understand how this phenotype compares to that of the control diet flies at the molecular level, we conducted TRAP of the Gr5a+ neurons of flies in the SD>CD and CD>CD conditions. mRNAs from flies on a SD>CD showed a strong negative skew in log 2 fold changes compared to the control diet group (Fig. 6C, -2.02), reminiscent of the skew we observed in flies fed a sugar diet (Fig. 4C). Furthermore, we observed that 47% (310/658) of genes in the transcriptional network repressed by PRC2.1 on a sugar diet were still decreased in SD>CD flies (Fig. 6D). Interestingly, the SD>CD animals clustered with the sugar diet 7 (SD7) group compared to sugar diet 3 (SD3) and Pcl mutants fed a sugar diet (Fig. 6D). Thus, at the molecular level, half of the neural state established by dietary sugar via PRC2.1 persisted. In addition to the GO term categories changed in the sugar diet condition (like signal transduction, cilium assembly, detection of chemical stimulus, carbohydrate metabolic process), we also found new GO term categories present (Fig. S15), suggesting that there may be aspects of this persistent state that are also novel. To test the hypothesis that PRC2.1 plays an active role in maintaining this neural state, we inhibited PRC2 activity during the "recovery" diet using the EEDi inhibitor (SD>CD+EEDi). Remarkably, these animals showed a restoration of wild-type sweet taste (Fig. 6E, green triangles). Together, these data indicate that the sensory neurons retain a phenotypic memory of the sugar diet environment, and that PRC2.1 is constitutively required for its persistence.

Discussion
In this study we set out to understand how the food environment alters the gustatory system as a model to define how interactions between genes and environment persistently reprogram complex behavior. Specifically, we took advantage of the ability of diet to change taste sensation to identify the molecular mechanisms through which the food environment changes neural state, physiology, and behavior in a circuit-specific way. Here we show that the decrease in sweet taste sensation that flies experience after chronic exposure to a high sugar diet is caused by the cell-autonomous action of the Polycomb Repressive Complex 2.1 in the sweet gustatory neurons. Mutations and pharmacological inhibition of PRC2.1 blocked the effects of the food environment on neural activity, behavior, and obesity. While we do not exclude the possibility that PRC1 and PhoRC may also be involved, we found that mutations or knockdown in these complexes had no effect on taste. In the high sugar food environment, PRC2.1 occupancy was redistributed, leading to the repression of transcription factors, neural, signaling, and metabolic genes that decreased the responsiveness of the Gr5a+ neurons and the fly's sensory experience. However, we discovered that PRC2.1 did not directly bind to neuronal genes in these cells and that, instead, it targeted transcription factors involved in sensory neuron development, synaptic function, and axon targeting. Specifically, on a high sugar diet Pcl binding was increased at the cad , GATAe , nub/pdm , Ptx1 loci and decreased at the Scro locus, with corresponding changes in the mRNA levels of these genes ( Interestingly, several of the transcription factors we identified -Ptx1 , Scro , and nub/pdm -have been shown to control the proper branching, synaptic connectivity, and function of sensory neurons (Corty et al., 2016;Iyer et al., 2013;Neumann and Cohen, 1998;Parrish et al., 2006Parrish et al., , 2007Vorbrüggen et al., 1997;Zaffran et al., 2000) , while others ( cad , nub/pdm ) play a role in neuroblast development (Doe, 2017;Kohwi and Doe, 2013) ; PRC2 also functions as a competence factor in neural proliferation, differentiation and sensory neurons (Bahrampour et al., 2019;Doe, 2017;Parrish et al., 2007) . Importantly, we found that the 4 activators that are repressed by Pcl in the high sugar condition are enriched in the Gr5a+ cells on a control diet, while Scro is depleted. Thus, the transcriptional network of ~658 genes controlled by these transcription factors may define the intrinsic properties of the sweet sensing neurons.
Interestingly, in addition to there being substantial overlap in the gene batteries regulated by the 5 transcription factors, we also found that these target genes were functionally interconnected in a network, especially those involved in signaling, synaptic function, and cell adhesion such as the kinase haspin , the adenylate cyclase ACXD , sytalpha , Arc1 , tetraspanin, jonan, and innexin proteins among others, which may affect the circuit both at the functional and connectivity levels. Since we did not observe a change in the expression of the sweet taste receptors, or the misexpression of other taste receptors (Scott, 2018) , our data are not consistent with a complete loss of identity of the Gr5a+ neurons. Instead, we propose that PRC2.1 tunes the sweet sensory neurons to the environment by altering a transcriptional network that controls the intrinsic properties of these cells, especially those involved in signal transduction, connectivity, synaptic function, and metabolism. Studies that test the effects of this network on the connectivity, morphology, and signal transduction of the sweet sensory neurons will shed light on how exactly the transcriptional remodeling caused by PRC2.1 impacts the Gr5a+ cells.
A previous study showed that PRC2 was required to maintain the properties of medium spiny neurons in post-mitotic cells (von Schimmelmann et al., 2016) , but here we found that it was its active redistribution in response to the environment -rather than its loss -that altered the properties of these cells. Thus, our work opens up the exciting possibility that PRC2 may modulate neural plasticity in response to environmental conditions. In other post-mitotic cells, PcG proteins alter transcriptional programs according to environmental stressors, such as oxidative stress, injury response, temperature, and light (Kolybaba and Classen, 2014;Marasca et al., 2018) . Our findings are in line with these and contend that, in addition to establishing cell fates, PRC2 regulates "neural states" depending on the environment.
Using neuroepigenetic mechanisms to tune neural states to external conditions could provide several advantages compared to the medley of other cellular, receptor, or synaptic plasticity based mechanisms. Specifically, it would allow cells to 1) orchestrate a coordinated response, 2) create a memory of the environment, and 3) buffer small fluctuations until a substantial challenge is perceived. It is particularly fascinating to think about the molecular mechanisms through which these neural states may be established. The need of neurons to constantly maintain their identity may mean that environmental signals like the extent of sensory stimulation could alter the expression of developmental gene batteries and affect neural states (Deneris and Hobert, 2014) . I ndeed, it has been speculated that some forms of plasticity may re-engage developmental programs that specify the intrinsic properties of neurons (Marder and Prinz, 2002;Parrish et al., 2014) . As mentioned above, we observed that the regulators of the transcriptional network we uncovered function in sensory neuron development and are enriched in the Gr5a+ cells. Thus, it could be a hallmark of neuroepigenetic plasticity to exploit developmental programs, linking the known role of PRC2 in establishing cell fates with this newly discovered function in modulating cell states. Incidentally, engaging developmental programs could be the reason why some environments and experiences leave a memory that leads to the persistent expression of the phenotype beyond the presence of the triggering stimulus, as these could target neural connectivity and set synaptic weights thresholds. Defining the circuit-specific changes of the pioneering studies on maternal care, addiction, and learning and memory (Champagne, 2008;Dulac, 2010;Nestler and Lüscher, 2019;Sweatt, 2013) would put this hypothesis to test. Here we found that the changes in taste sensation and half of the sugar diet neural state set by PRC2.1 remained even after animals were moved back to the control diet for a week. A limitation of our study is that due to the small number of Gr5a+ neurons and their anatomically inaccessible location, we were not able to measure the identity of the molecular memory in these cells alone. However, we saw that the phenotypic memory of the high sugar food environment was dependent on the constitutive action of PRC2.1. Based on other studies showing that the H3K27 methyl mark acts as a molecular memory during development (Coleman and Struhl, 2017;Laprell et al., 2017) , we speculate that this is likely to be the molecular signal in the Gr5a+ cells too. It is interesting to note that a long term (4 weeks) increase in H3K27 methylation at the BDNF exon III and IV promoter was measured in response to social defeat, although the role of this mark or PRC2 in the behavioral phenotype or its persistence was not examined (Tsankova et al., 2006) . In conclusion, we show that PRC2.1 mediates the effects of dietary sugar on sweet taste by establishing persistent alterations in the taste neurons that remain as a phenotypic and neural memory of the previous food environment. Given the importance of taste in modulating food intake, we speculate that this memory may lock animals into patterns of feeding behavior that become maladaptive and promote obesity, and, indeed, we found that Pcl mutant flies are protected from diet-induced obesity. Thus, the food environment, like other experiences, can induce lasting alterations that restrict the behavioral plasticity of animals and impact disease risk. Since the content of sugar in processed foods is similar or higher than that we fed flies and the function of PcG proteins is conserved from plants to humans, our work is broadly relevant to understanding the effects of the food environment on a whole range of diet-related conditions and diseases that affect billions of people worldwide.

Fly Husbandry and Strains:
All flies were grown and maintained on cornmeal food (Bloomington Food B recipe) at 25°C and 45%-55% humidity under a 12:12 hour light-dark cycle (ZT0 at 9 AM). Male flies were collected under CO2 anesthesia 1-3 days after eclosion and maintained in a vial that housed 35-40 flies.
Flies were acclimated to their new vial environment for an additional 2 days. For all experiments, flies were changed to fresh food vials every other day.
For all dietary manipulations, the following compounds were mixed into standard cornmeal food (Bloomington Food B recipe) (0.58 calories per gram) by melting, mixing, and pouring new vials as in (Musselman and Kühnlein, 2018) and (Na et al., 2013) . For the 30% sugar diet (1.41 calories per gram) Domino granulated sugar w/v was added. For the EED226 inhibitor diet (AxonMedchem), EED226 was solubilized in 10% DMSO and added to control or 30% sugar diet at a total concentration of 8 uM.
For genetic manipulations the GAL4/UAS system was used to express transgenes of interest in

Proboscis Extension Response:
Male flies were food deprived for 18-24 hours in a vial with a Kimwipe dampened with 2 mL of milliQ-filtered deionized (milliQ DI) water. Proboscis extension response (PER) was carried out as described in (Shiraiwa and Carlson, 2007) .

Proboscis Immunofluorescence:
Probosces were dissected in 1xPBS and fixed in 4% PFA, mounted in FocusClear (CelExplorer) on coverslips. Cell bodies were imaged using a FV1200 Olympus confocal with a 20x objective.
Cells were counted using Imaris Image analysis software.

Calcium Imaging:
Male

RNA Extraction and Quantitative RT-PCR:
For all RNA extractions used for qPCR, fly heads from 10-20 flies were dissected into Trizol (Ambion) and homogenized with plastic pestles. RNA was extracted by phenol chloroform    (Marshall et al., 2016) , and subsequently purified. Purified DNA was digested with DpnII followed by sonication to yield fragments averaging 200-300bp. TaDa adaptors were removed from sonicated DNA by digestion (Marshall et al., 2016) .

Library Preparation for TRAP and TaDa:
For RNA sequencing libraries were generated using the Ovation SoLo RNA-Seq System for Drosophila (Nugen, 0502-96). All reactions included integrated HL-dsDNase treatment (ArcticZymes, Cat. #70800-201). For DNA sequencing libraries were generated using the Takara ThruPlex kit (cat #022818) using 3ng input and 10 cycles of PCR. All libraries were sequenced on the Illumina NextSeq platform (High-output kit v2 75 cycles) at the University of Michigan core facility.

High Throughput RNA-seq Analysis
Fastq files were assessed for quality using FastQC (Andrews and Others, 2010) . Reads with a quality score below 30 were discarded. Sequencing reads were aligned by STAR (Dobin et al., 2013) to dmel -all-chromosomes of the dm6 genome downloaded from Ensemble, and gene counts were obtained by HTseq (Anders et al., 2015) . Count files were used as input to call differential RNA abundance by DESeq2 (Love et al., 2014) . All pairwise comparisons were made to the control diet of the corresponding genotypes, such that sugar diet three days and sugar diet seven days were compared to the age matched control diet group. In Pcl mutants experiments, the pairwise comparison was made between sugar diet and control diet within the age-matched Pcl c429 genotype group. A cutoff of q val<0.1 was used to call differentially expressed genes. Skew in log 2 fold changes was measured using the R package Skewness (e1071). RNAseq data visualization was carried out in R studio using ggplot2 and the following packages, pheatmap (Kolde, 2012) , Venneuler (Wilkinson, 2012) , and EnhancedVolcano (Blighe, 2019) . To cluster columns and rows in pheatmap "Ward.D '' clustering was used.

High Throughput TaDa and CATaDa Analysis
Fastq files were assessed for quality using FastQC (Andrews and Others, 2010) . Reads with a quality score below 30 were discarded. The damidseq_pipeline was used to align, extend, and generate log2 ratio files ( Dam::Pcl/Dam ) in GATC resolution as described previously (Marshall and Brand, 2015) . In short, the pipeline uses Bowtie2 (Langmead and Salzberg, 2012) to align reads to dmel -all-chromosomes of the dm6 genome downloaded from Ensemble, followed by read extension to 300 bp (or to the closest GATC, whichever is first). Bam output is used to generate the ratio file in bedgraph format. Bedgraph files were converted to bigwig and visualized in the UCSD genome browser. Correlation coefficients and PCA plot between biological replicates were computed by multibigwigSummary and plotCorrelation in deepTools (Ramírez et al., 2016) . Fold Change traces for SD/CD of log 2 ( Dam::Pcl/Dam ) were generated by calculating the mean profile of all replicates for each condition and subsequently calculating fold change between the sugar diet and control diet condition with deepTools bigwigCompare (Ramírez et al., 2016) . Peaks were identified from log 2 ( Dam::Pcl/Dam ) ratio files using find_peaks (FDR<0.01) (Marshall and Brand, 2015) . To do this, the binding intensity thresholds are identified from the dataset, the dataset is then shuffled randomly, and the frequency of consecutive regions (i.e. GATC fragments or bins) with a score greater than the threshold is calculated. The FDR is the observed / expected for a number of consecutive fragments above a given threshold. Association of genes to peaks was made using the peaks2genes script (Marshall and Brand, 2015) and dm6 genome annotations. Overlapping intervals or nearby intervals were merged into a single interval using MergeBED in Bedtools (Quinlan and Hall, 2010) . Intervals common in all replicate peak files were identified by Multiple Intersect in Bedtools (Quinlan and Hall, 2010) . DiffBind was used to determine differentially bound sites on peak files in bed format based on differences in read intensities (Stark et al., 2011) .
For CATaDa experiments, all analyses were performed similar to those of TaDa with the exceptions that 1) Dam only profiles were not normalized as ratios but shown as non normalized binding profiles, 2) Dam only coverage plots were generated by converting bam files to bigwig files normalized to 1x dm6 genome, and 3) peaks were called using MACS2 call peaks on alignment files without building the shifting model with an of FDR<0.05 (Feng et al., 2012) .
To determine the proportion of genes that fit within the various chromatin domain subtypes, we first

PREdictor
Identification of predicted PRE sites was performed exactly as described in (Khabiri and Freddolino, 2019) , using the dm6 genome. As suggested in (Khabiri and Freddolino, 2019) , we use a threshold confidence score of 0.8 to identify the PREs used in the present analysis. A complete list of predicted PREs, with accompanying confidence scores, is shown in supplementary file 1. Enrichments of overlap between different PRE classes and Pcl occupancy locations were calculated by comparing the observed overlap frequency with the overlaps for 1,000 random shufflings of the binding/differential binding peak locations (calculated using bedtools 2.17.0 (Quinlan and Hall, 2010) .

Calculation of regulatory targets of transcription factors
To identify likely targets of each transcription factor of interest, we drew upon the transcription factor binding site calculations described in (Khabiri and Freddolino, 2019) , in which the motif of each transcription factor was scanned along every base pair of the D. melanogaster dm6 genome using FIMO (Grant et al., 2011) , and the base pair-wise binding results converted to robust z-scores. For each TF, we then considered its regulon to consist of all genes with at least one binding site with z-score above a TF-specific threshold within 2 kb upstream of the beginning of the gene. We identified TF-specific thresholds by manual inspection of a plot of the average expression changes between conditions vs. threshold, aiming to identify a point of maximum information content relative to noise (see supplemental figure 8 for the plots used to identify TF-specific z-score thresholds). Once the set of targets (regulon) for each factor was identified, we tested for significant enrichment or depletion of overlaps between the regulons using Fisher's exact test, reporting Benjamini-Hochberg false discovery rates (FDRs) (Benjamini and Hochberg, 1995) . All calculated odds ratios were positive, indicating enriched overlaps between the regulons.

STRING network analysis
To develop a functional network between the candidate neural targets of cad , Ptx1 , nub , GATAe  (Szklarczyk et al., 2019) . Genes were clustered by their reported protein-protein interactions and corresponding confidence scores (Szklarczyk et al., 2019) and plotted in Cytoscape (v3.7.1) (Shannon et al., 2003) . In this network edges do not represent direct protein protein interaction but rather represent a functional interaction. For network see Supplementary file 5.

Data Analysis and Statistics
Statistical tests, sample size, and p or q values are listed in each figure legend. Data were evaluated for normality and appropriate statistical tests applied if data were not normally distributed, all the tests, biological samples, and the p and q values are listed in the figure legends and specific analysis under each methods session. Because the inferential value of a failure to reject the null hypothesis in frequentist statistical approaches is limited, for all RNA-seq expression datasets, we coupled our standard differential expression with a test for whether each gene could be flagged as 'significantly not different'. Defining a region of practical equivalence (ROPE) as a change of no more than 1.5-fold in either direction, we tested the null hypothesis of at least a 1.5-fold change for each gene, using the gene-wise estimates of the standard error in log2 fold change (reported by Deseq2) and the assumption that the actual log 2 fold changes are normally distributed. Rejection of the null hypothesis in this test is taken as positive evidence that the gene's expression is not changed substantially between the conditions of interest. Python code for the practical equivalence test can be found on Github as calc_sig_unchanged.py. All data in the figures are shown as Mean ± SEM, **** p < 0.0001, *** p < 0.001, ** p < 0.01,*p< 0.05 unless otherwise indicated.

Data availability
All high throughput sequencing data files can be found on Gene Expression Omnibus GSE146245.