Caffeine intake exerts dual genome-wide effects on hippocampal metabolism and learning-dependent transcription

1 Caffeine intake exerts dual genome-wide effects on hippocampal metabolism 1 and learning-dependent transcription 2 3 Isabel Paiva1†, Lucrezia Cellai2,3†, Céline Meriaux2,3†, Lauranne Poncelet4†, Ouada Nebie2,3, Jean4 Michel Saliou5, Anne-Sophie-Lacoste5, Anthony Papegaey2,3, Hervé Drobecq6, Stéphanie Le Gras7, 5 Marion Schneider8, Enas M. Malik8, Christa E. Müller8, Emilie Faivre2,3, Kevin Carvalho2,3, Victoria 6 Gomez-Murcia2,3, Didier Vieau2,3, Bryan Thiroux2,3, Sabiha Eddarkaoui2,3, Thibaud Lebouvier2,3,9, 7 Estelle Schueller1, Laura Tzeplaeff1, Iris Grgurina1, Jonathan Seguin1, Jonathan Stauber4, Luisa V. 8 Lopes10, Luc Buée2,3, Valérie Buée-Scherrer2,3, Rodrigo A. Cunha11,12, Rima Ait-Belkacem4‡, Nicolas 9 Sergeant2,3‡, Jean-Sébastien Annicotte13,14‡, Anne-Laurence Boutillier1‡*, David Blum2,3‡* 10 11 12 † Equal contributions 13 ‡ Equal contributions 14 15 1. University of Strasbourg, CNRS, UMR7364 - Laboratoire de Neuroscience Cognitives et 16 Adaptatives (LNCA), F-67000 Strasbourg, France. 17 2. University of Lille, Inserm, CHU Lille, UMR-S1172 LilNCog - Lille Neuroscience & Cognition, Lille, 18 France. 19 3. Alzheimer and Tauopathies, LabEx DISTALZ, France. 20 4. ImaBiotech SAS, Parc Eurasanté, F-59120 Loos, France. 21 5. Univ. Lille, CNRS, Inserm, CHU Lille, Institut Pasteur de Lille, UAR CNRS 2014 - US Inserm 41 - 22 PLBS, F-59000 Lille, France 23 6. CIIL - Centre d’Infection et d’Immunité de Lille (CIIL) - INSERM U1019 - UMR 9017 24 7. Univ. Strasbourg, CNRS UMR7104, Inserm U1258 - GenomEast Platform – IGBMC - Institut de 25 Génétique et de Biologie Moléculaire et Cellulaire, F-67404 Illkirch, France. 26 8. PharmaCenter Bonn, Pharmaceutical Institute, Pharmaceutical & Medicinal Chemistry, University 27 of Bonn, D-53121 Bonn, Germany. 28 9. CHU Lille, Memory Clinic, Lille France. 29 10. Instituto de Medicina Molecular, Faculdade de Medicina de Lisboa, Universidade de Lisboa, 30 Lisbon, Portugal. 31 11. CNC - Center for Neuroscience and Cell Biology, University of Coimbra, 3004-504 Coimbra, 32 Portugal. 33 12. Faculty of Medicine, University of Coimbra, 3004-504 Coimbra, Portugal. 34 13. Univ. Lille, INSERM, CNRS, CHU Lille, Institut Pasteur de Lille, Inserm U1283 / CNRS UMR8199 35 - EGID, 59000 Lille, France. 36 14. Univ. Lille, INSERM, CHU Lille, Institut Pasteur de Lille, U1167 – RID-AGE-Facteurs de risque 37 et déterminants moléculaires des maladies liées au vieillissement, 59000 Lille, France. 38 39 40 # Correspondence to: 41 David Blum, Inserm UMR-S1172, “Alzheimer & Tauopathies”, Place de Verdun, 59045, Lille Cedex, 42 France. Orcid Number: 0000-0001-5691-431X. Tel: +33320298850, Fax: +33320538562. 43 [email protected] 44 Anne-Laurence Boutillier, Laboratoire de Neuroscience Cognitives et Adaptatives (LNCA), 45 UMR7364 Cnrs Unistra, 67000 Strasbourg, France. Orcid Number: 0000-0002-2317-928.0 46 laure[email protected] 47 48 49 Conflict of interest. The authors have declared that no conflict of interest exists. 50 2 Abstract 51 52 Caffeine is the most consumed psychoactive substance worldwide. Strikingly, molecular pathways 53 engaged by its regular consumption remain unclear. We herein addressed the mechanisms 54 associated with habitual (chronic) caffeine consumption in the mouse hippocampus using untargeted 55 orthogonal-omics techniques. Our results revealed that chronic caffeine exerts concerted pleiotropic 56 effects in the hippocampus, at the epigenomic, proteomic and metabolomic levels. Caffeine lowers 57 metabolic-related processes in the bulk tissue, while it induces neuronal-specific epigenetic changes 58 at synaptic transmission/plasticity-related genes and increased experience-driven transcriptional 59 activity. Altogether, these findings suggest that regular caffeine intake improves the signal-to-noise 60 ratio during information encoding, in part through a fine-tuning of metabolic genes while boosting the 61 salience of information processing during learning in neuronal circuits. 62 63 64 65 66 3 Introduction 67 68 Caffeine is the most consumed psychoactive substance worldwide (about 80% of the population) via 69 dietary intake from coffee, tea and soda beverages. Its popularity derives from its ability to enhance 70 well-being and some central-related functions such as attention and alertness (1). Large 71 epidemiological studies point out an inverse association between coffee/caffeine consumption and 72 all-cause mortality (2–4). In general, the impact of caffeine on human health follows an inverted bell73 shaped dose-response curve with benefits observable at doses of 200-400 mg per day, that can be 74 recapitulated by 0.3 g/L p.o. in rodents. 75 Compelling epidemiological and experimental evidence support that habitual/chronic caffeine 76 consumption normalizes synaptic plasticity and cognitive decline in altered allostatic situations such 77 as ageing, Alzheimer’s disease or other neuro-psychiatric conditions (5–7). A more limited number 78 of studies however also support that, independently of its ability to favor arousal and attention, 79 caffeine may exhibit cognitive-enhancing properties. After being rewarded with caffeine, honeybees 80 are able to remember a previously learned floral scent (8). Also, acute caffeine administration in rats 81 can enhance memory test performance (9, 10). In Humans, caffeine intake immediately following 82 learning improves discrimination performance 24 hours later (11). These results are in line with 83 observations supporting the ability of caffeine to modulate hippocampal/cortical excitability in 84 homeostatic conditions. Indeed, caffeine treatment in hippocampal slices enhances basal synaptic 85 transmission (12–14) and modulates long-term potentiation (LTP) in rodents’ hippocampus (12, 15, 86 16) and sharp wave ripple complexes, that are proposed to underlie memory consolidation (17). 87 Caffeine also controls neuronal excitability and LTP-like effects in the human cortex (18, 19). Most 88 of these studies however rely on acute administrations with limited relevance towards 89 habitual/chronic consumption. 90 Despite caffeine’s popularity, brain molecular changes associated with its chronic intake remain ill91 defined. Caffeine is known to essentially interfere with the adenosinergic system where it acts as an 92 4 antagonist (20). However, adaptive downstream pathways engaged by habitual/chronic caffeine 93 consumption have been largely overlooked. In the present study, we used a combination of unbiased 94 orthogonal-omics techniques to analyze the epigenome, transcriptome, proteome and metabolome 95 of the mouse hippocampus in order to uncover the molecular pathways impacted by chronic caffeine 96 consumption in neuronal processing during learning. 97 98 5 Results 99 100 Mouse monitoring and caffeine concentrations. In our experimental conditions, neither mortality 101 nor signs of animal suffering in caffeine-treated animals were encountered. Average consumption of 102 0.3 g/L caffeinated water was 4.83 ± 0.15 mL/mouse/day resulting in brain caffeine concentrations 103 of 3.6 ± 1.1 µM, corresponding to a moderate intake in Humans (20). Caffeine metabolites 104 (paraxanthine, theobromine and theophylline) were also detected in the brain of treated mice with 105 respective concentrations of 1.9 ± 0.4 µM, 1.8 ± 0.3 µM, and 0.10 ± 0.03 µM (n=5). 106 107 Chronic caffeine consumption decreases histone acetylation of metabolic-related genes in 108 the hippocampus. We hypothesized that chronic caffeine consumption could affect hippocampal 109 epigenome of mice. As caffeine is a psychostimulant, we focused on two chromatin marks 110 associated with “active chromatin” and specific transcriptional states. Histone H3 acetylation at lysine 111 27 (H3K27ac) is preferentially enriched at active enhancers (21), also forming large clusters of 112 H3K27ac-enriched enhancers known as “super-enhancers” on highly transcribed genes that are cell113 or tissue-specific (22, 23). Histone H3 lysines K9 and K14 (H3K9/K14ac), on which acetylation co114 occurs at many gene regulatory elements, allows to differentiate active enhancers from inactive ones 115 and thus represents a dynamic mark accounting for stimuli dependent activation (24). Locus specific 116 acetylation was evaluated by chromatin immunoprecipitation followed by sequencing (ChIP-seq) 117 experiments in dorsal hippocampus of control (water) and caffeine-treated mice. A total of 2 biological 118 replicates were performed and Principal Component Analysis (PCA) of the two histone marks was 119 generated (Supplemental Figure 1A,B). Chronic caffeine intake significantly decreased the 120 acetylation of both histone marks at many genomic loci. H3K9/14ac was depleted in 778 genomic 121 regions (768 genes) while only 3 were rarely identified as significantly enriched in caffeine-treated 122 animals (FDR<1E-5) (Figure 1A, Supplemental Table 1). Gene ontology analysis using Genomic 123 6 Regions Enrichment of Annotations Tool (GREAT) revealed that these acetylation-depleted regions 124 were associated with genes involved in the regulation of metabolic processes (amide, lipids), mRNA 125 transport, regulation of translation and dendritic spine morphogenesis and development (Figure 1B). 126 A more robust effect was observed in H3K27ac whose peaks were found decreased in 2105 genomic 127 regions (1766 genes) and increased in only 4 genomic regions in caffeine vs. control mice (FDR<1E128 5) (Figure 1C, Supplemental Table 2). Metabolic-related pathways, such as lipid catabolic or amide 129 metabolic processes were among the decreased peaks of both histone marks (Figure 1B,D). 130 Additionally, H3K27ac-depleted regions were significantly associated with myelin-related processes, 131 MAP kinase, negative regulation of calcium-mediated signaling pathways, as well as 132 heterochromatin organization (Figure 1D). We also performed Kyoto Encyclopedia of Genes and 133 Genomes (KEGG) pathway analyses and identified many processes, some of which related to 134 cAMP-, MAP kinase, Rap1-signaling pathways and circadian entrainment for both H3K9/14 and 135 H3K27ac depleted regions (Figure 1E). Of note, the KEGG pathway database pointed out metabolic136 related pathways, such as “insulin signaling”, for genes depleted in acetylation of both histone marks 137 (Figure 1E) and “glucagon signaling pathway” for those associated with H3K9/14ac depleted regions 138 (Figure 1E, blue bars). Those genes associated with insulin and glucagon signaling pathways were 139 represented by protein-protein interaction network analysis (STRING), showing strong 140 interconnectivity (Figure 1F, yellow and pink dots, respectively). As examples, genomic region 141 representation of the Insulin Receptor Substrate 1 (Irs1) gene, which is required for insulin signaling 142 and related spine maturation and synaptic plasticity (25) and the Glycogen Synthase Kinase 3 Beta 143 (Gsk3b) gene are shown (Figure 1G), with significant acetylation depletion for both marks in the 144 caffeine-treated group versus control (respectively left, H3K9/14ac, FDR=7.75E-05 and H3K27ac, 145 FDR=1.82E-12; right, H3K9/14ac, FDR=2.58E-11 and H3K27ac, FDR=4.83E-05). Other regions, 146 such as those associated with Dusp3, Psme3 and Mlh3 genes, did not exhibit such histone 147 acetylation changes upon caffeine treatment, attesting for selectivity of the caffeine effect for both 148 histone marks (Supplemental Figure 1C). In addition, integrated pathway analysis (IPA) applied to 149 7 common ChIP-seq data of both marks confirmed that metabolic pathways, such as insulin or IGF-1 150 signaling, were canonical pathways downregulated upon caffeine treatment (Supplemental Table 151 3). Potential contributors to the caffeine effects on the epigenome were further assessed using the 152 “upstream regulator analysis” function of IPA (Supplemental Table 4). We identified in the 153 acetylation-depleted genes, TCF7L2 (Transcription factor 7-like 2) as the most significant upstream 154 regulator inhibited upon caffeine consumption for both marks. Furthermore, ADORA2A (A2AR) was 155 identified as another upstream regulator in the epigenomic data, in striking accordance with the 156 primary ability of caffeine to antagonize adenosine receptors (20). Altogether, these data show that 157 in the bulk hippocampus, chronic caffeine treatment induces an overall deacetylation of two active 158 transcription marks, H3K27ac and H3K9/14ac, on genes related to translation, lipid and 159 glucose/insulin-related metabolisms. 160 To assess whether this histone acetylation depletion exerts an effect on gene transcription, we 161 performed RNA-sequencing (RNA-seq) of both water and caffeine-treated mice. Although differential 162 expression analysis revealed no statistically significant changes of gene expression between groups 163 (Supplemental Figure 2A), relative quantification of the gene expression (z-score) corresponding 164 to all H3K27ac-depleted loci showed an overall decrease in expression (Supplemental Figure 2B) 165 over the same number of randomly chosen genes. Furthermore, we also checked by RT-qPCR (n=5166 6/group) expression levels of several genes chosen amongst the most depleted ones in H3K27ac 167 and observed a decreased expression following chronic caffeine treatment (Supplemental Figure 168 2D, red columns). Importantly, we found that some of these genes, such as PBX Homeobox 1 169 (Pbx1), NAD Kinase 2 (Nadk2) and Spindle And Centriole Associated Protein 1 (Spice1), displayed 170 decreased expression not only upon chronic (2 weeks) but also following an acute (24h) caffeine 171 treatment (Supplemental Figure 2D, green columns). However, the Cytochrome P450 Family 51 172 Subfamily A Member 1 (Cyp51) gene, that plays a central role in cholesterol and lipid metabolisms, 173 showed decreased expression solely upon chronic caffeine treatment. Moreover, a persistent effect 174 of caffeine on gene expression was observed for Pbx1 and Nadk2 genes, as their expression 175 8 remained decreased even after a 2-week caffeine withdrawal following chronic administration 176 (Supplemental Figure 2D, blue columns). 177 178 Impact of chronic caffeine consumption on hippocampal metabolome. Considering that 179 caffeine decreased histone acetylation of metabolic-related genes, we further assessed the impact 180 of the decreased histone acetylation on the hippocampal metabolome. To do so, tissue spatial 181 distribution of molecules was visualized by MALDI (matrix assisted laser desorption ionization) mass 182 spectrometry imaging analysis, acquired from the dorsal hippocampus (Bregma -1.7mm; Figure 2A) 183 of water and caffeine-treated mice (n=6/group). PCA analysis was then performed on the recorded 184 mass spectrometry images from both mouse groups (water and caffeine-treated), in order to highlight 185 differences in their hippocampal molecular distribution profiles (Figure 2B). This revealed lipidomic 186 and metabolomic signatures related to chronic caffeine intake, resulting in two distinctly separated 187 clusters. The identification of metabolites and lipids was based on the measurement of their m/z and 188 subsequent comparison with different databanks. In total, 59% of the metabolome was assigned to 189 the biochemical class of metabolites (27%) and lipids (32%) (Figure 2C). The m/z value of the 190 remaining 41% did not allow for a univocal assignment to a specific biochemical class. Ultimately, 191 statistical analysis of the molecular datasets revealed that chronic caffeine consumption induced a 192 major decrease in metabolites and lipid levels (92% decreased vs. 8% increased; Figure 2D). The 193 identified species between water and caffeine groups (p < 0.05), detected in positive and negative 194 ionization mode, are listed in Supplemental Table 5. Related molecular images taken from 195 hippocampi of water and caffeine-treated mice, showing their different levels and distribution are 196 displayed in Figure 2E. 197 198 Proteomic hippocampal signature associated with chronic caffeine consumption. To gain 199 insights into the potential effect of chronic caffeine intake at the protein level, we performed mass 200 spectrometry proteomic analysis of the bulk dorsal hippocampus of water (control) and chronic 201 9 caffeine-treated mice (n=3/group). Caffeine induced alterations of 179 proteins, of which 49 202 displayed decreased and 130 increased expression levels (Figure 3A, Supplemental Table 6). In 203 line with the two previous datasets (epigenomics and metabolomics), gene ontology and protein 204 network analysis revealed that decreased proteins were again associated with peptide and cellular 205 amide metabolic processes as well as with mitochondria, with reduction of NADH:Ubiquinone 206 Oxidoreductase Subunit A3 (NDUFA3) involved in mitochondrial respiratory chain complex I 207 assembly, of Mitochondrial Pyruvate Carrier 1 (MPC1) responsible for transporting pyruvate into 208 mitochondria or of Long-Chain-Fatty-Acid-CoA Ligase 4 (ACSL4) involved in lipid metabolism 209 (Figure 3B). Together, these three approaches suggest a robust decrease in metabolic processes 210 induced by chronic caffeine intake in the bulk hippocampal tissue. 35 out of the 49 proteins 211 decreased by caffeine, including Insulin Degrading Enzyme (IDE) and NDUFA3, were reversed by 212 caffeine withdrawal. Only 14 proteins, such as Insulin Like Growth Factor 2 Receptor (IGF2R) 213 remained decreased following caffeine withdrawal (Supplemental Table 6). 214 Gene Ontology analysis of the increased proteins revealed three main protein clusters: one related 215 with RNA-binding and spliceosome, a second linked to autophagosome and protein processing to 216 endoplasmic reticulum, and a last one associated with glutamatergic synapse and phosphatase 217 activity. Considering that caffeine induced expression of some synaptic proteins, and controls 218 glutamatergic synaptic transmission (e.g. (19), we further assessed their predicted role in the 219 synaptic compartment using the Synaptic Gene Ontologies and annotations (SynGO) (26). We 220 observed that most of the synaptic proteins annotated were related to synaptic organization and 221 signaling, more particularly, to chemical synaptic transmission, such as SH3 And Multiple Ankyrin 222 Repeat Domains 3 (SHANK3) that encodes critical scaffolding proteins for glutamatergic 223 neurotransmission in the post-synaptic densities (27), Synaptopodin (SYNPO) a part of the actin 224 cytoskeleton of postsynaptic densities (28) or CREB–Regulated Transcription Coactivator 1 225 (CRTC1) involved in hippocampal plasticity and memory (29). Overall, proteomic analysis revealed 226 a decrease in metabolism-related proteins, concomitant with an increase of neuronal/synapse227 16 in mitochondrial activity (e.g. NDUFA3 and MPC1). These chronic changes were to some point 369 related to acute caffeine treatment as a few genes were similarly impacted following a 24h and a 2370 week caffeine treatment, in line with Yu et al., 2009 (41), but the main changes were associated with 371 long-term exposure to caffeine, as found for e.g. the Cyp51 gene, encoding a protein involved in 372 cholesterol and lipid metabolism. In accordance, we found that 14 over 49 downregulated 373 hippocampal proteins were still altered despite 2 weeks of caffeine withdrawal, indicating a 374 persistence of chronic caffeine effects, as previously suggested (42). Among these long-lasting 375 impacted proteins by chronic caffeine intake, we found ACSL4 and GNA14, which are involved in 376 the cellular synthesis of fatty acids/lipids, or IGF2 receptor and ITPR3, involved in insulin-dependent 377 regulations. Importantly, these data are in line with and bring molecular support to recent functional 378 magnetic resonance imaging data showing that habitual coffee drinkers exhibit decreased brain 379 functional connectivity at rest (43). As bulk hippocampal tissue was investigated, a question lies in 380 understanding the cellular types underlying such metabolic decrease. Independent IPA analysis of 381 our two sets of epigenomic data (ChIP-seq on bulk hippocampal tissue and CUT&Tag-seq on 382 dissociated hippocampal cells, “all cells”) particularly pointed at three common upstream regulators: 383 TCF7L2 (transcription factor 7 like 2), MKNK1 (MAPK interacting serine/threonine kinase 1) and 384 NFASC (neurofascin). In the mouse brain, these genes are predominantly expressed by non385 neuronal cells: TCF7L2 is preferentially expressed by newly formed oligodendrocytes and 386 astrocytes, NFASC in newly formed oligodendrocytes, while MKNK1 is particularly enriched in 387 microglia (see https://www.brainrnaseq.org/). IPA analysis of “all cells” CUT&Tag-seq data further 388 highlighted the involvement of GLI1 and SOX2, that are both particularly enriched in astrocytes. 389 These observations strongly support that the basal/resting signatures elicited by chronic caffeine 390 intake may rely on non-neuronal, likely glial, responses. 391 Concomitant with this de-acetylation process observed in the bulk hippocampus, we showed that 392 chronic caffeine was able to induce a neuron-autonomous epigenomic response using both active 393 (H3K27ac) and repressive (H3K27me3) marks: acetylation of H3K27 was enriched while its tri394 17 methylation was depleted at genes related to membrane potential, potassium ion regulation and 395 learning and memory processes. This suggests that the overall chronic caffeine effect positively 396 regulates neuronal activity and synaptic transmission. Proteomic studies supported this argument as 397 a series of identified upregulated proteins were related to the glutamatergic synapse. It is interesting 398 to note that 73 out of 130 upregulated proteins -some of them related to the synapseremained 399 elevated even after a 2-weeks caffeine withdrawal, revealing a long-lasting impact of chronic caffeine 400 intake on neurons. Integration of epigenomic and proteomic data particularly pointed towards 401 CRTC1, known to act as a coincidence sensor of calcium and cAMP signals in neurons triggering a 402 transcriptional response involved in late-phase LTP maintenance at hippocampal synapses (44). We 403 further observed that chronic caffeine intake impacts the learning/training-induced transcriptome by 404 significantly enhancing the number of differentially regulated genes. Integration of the learning405 induced genes with epigenomic data identified a group of 121 genes related to metabolic processes 406 that, besides being over-activated in caffeine-treated mice in learning conditions, were also de407 acetylated with decreased overall expression in resting conditions (z-score). This suggests that the 408 resting-state effect of caffeine in non-neuronal/glial cells might be a pre-requisite to the robust 409 activation of metabolic pathways then improving quality and precision of learning-associated 410 processes, in line with its cognitive enhancing function. 411 Thus, a major overall conclusion of the present study is the ability of regular caffeine intake to exert 412 a long-term effect on neuronal activity/plasticity in the adult brain, through concerted actions on the 413 epigenome, transcriptome, proteome and metabolome, ultimately lowering metabolic-related 414 processes; and to simultaneously finely tuning activity-dependent regulations for a more efficient 415 response to experience. In other words, in non-neuronal cells caffeine decreases -omic activities 416 under basal conditions and improves the signal-to-noise ratio during information encoding in brain 417 circuits, thus contributing to bolster the salience of information in brain circuits. Remarkably, this dual 418 and opposite impact of caffeine under resting conditions and upon brain activation is in line with 419 human brain imaging studies: under basal conditions caffeine increased brain entropy (45) and 420 18 decreased functional connectivity (46), whereas it increases BOLD activation in the frontopolar and 421 cingulate cortex in a verbal working memory task (47) reflecting an increased processing potential. 422 Additionally, neurophysiological studies on the putative targets of caffeine - adenosine receptors – 423 are in line with this dual role of caffeine, as shown by the opposite effects of A2AR to enhance 424 glutamate release contrasting with the A1R-mediated inhibition of basal synaptic transmission (48), 425 which is also controlled by A2AR (49). Finally, our data also show that the amplitude of the 426 transcriptomic effects of caffeine was far greater when neuronal networks were activated during the 427 learning process rather than in basal conditions, as noted by others when studying the impact of 428 caffeine on gene expression in the basal ganglia (50). This might particularly relate to a “priming” of 429 neuronal activity which would favor the rise of activity-dependent response, as it has been suggested 430 for the mechanism of action of HDAC inhibitors (51). How caffeine coordinates these epigenomic 431 responses in the different cell types is an interesting question that we are currently pursuing. 432 Finally, the present study highlights the molecular impact of caffeine in the homeostatic brain, that 433 will deserve further investigations, namely regarding the differential mechanisms operating at the 434 cell-specific level to modulate physiological brain activity in resting and activity settings. Our data 435 have additional far-reaching implications. While it is recognized that caffeine exhibits normalizing 436 properties in models of synaptic dysfunction, as in Alzheimer’s disease (52–54), the cell-specific 437 molecular mechanisms remains to be uncovered. In the opposite side of the allostatic brain spectrum 438 (55), caffeine has been suggested to impact synaptic fate in brain development (56, 57) but the 439 involvement of neuronal vs. non-neuronal mechanisms remains ill-defined. It is therefore particularly 440 relevant and important to address, at a larger scale, the integrated actions of caffeine in neuronal vs. 441 non-neuronal cells in the immature, homeostatic and ageing brain. 442 443 19 Materials and Methods 444 445 Animals. Male C57Bl6/J mice (Charles River Laboratories, France) were housed in a pathogen-free 446 facility (University of Lille, France). Mice were 5-6 per cage (GM500, Tecniplast) and maintained 447 under controlled housing conditions for temperature (22°C) and light (12-hour light/dark cycle), with 448 ad libitum access to food and water. 449 450 Caffeine treatment. Two-three-months-old mice were randomly assigned to the two following 451 experimental groups: water (control) and caffeine. Caffeine solutions were kept in dark bottles thus 452 protected from light and changed weekly. Treatment started at 8-9 weeks of age and lasted for two 453 weeks. The chronic caffeine treatment in mice has been set in order to mimic the usual dose range 454 of caffeine consumption in Humans. The selected caffeine dose of 0.3 g/L p.o., administered through 455 drinking water at 0.3 g/L, has been previously shown to provide a significant benefit in 456 neurodegenerative contexts (54, 58, 59). Regarding the comparison of caffeine exposure for 2 457 weeks vs. 24 hours vs. caffeine removal, we proceed as follows: 6 animals were kept under water 458 and other 6 animals were treated with caffeine for 2 weeks and returned to water for 2 additional 459 weeks (caffeine withdrawal group). When the later group of animals returned to water, an additional 460 group that was under water for 2 weeks was then treated with caffeine. A last group was kept under 461 water for 2 weeks and treated with caffeine for only 24 hours. All animals were then sacrificed the 462 same day, the dorsal hippocampus was sampled and stored as indicated below and used for 463 proteomics and RT-qPCR analysis. 464 465 Quantitative determination of caffeine and metabolites in brain samples. Brain tissues from 466 water and caffeine groups were used to assess concentrations of caffeine and its metabolites 467 (paraxanthine, theobromine and theophylline). Samples were weighed and 1 mL of 1% formic acid 468 (FA) solution was added to each sample. To determine the recovery rate, control samples were 469 20 spiked with a mixture of caffeine, paraxanthine, theobromine and theophylline (10 µM each). The 470 tissues were lysed using 7 mm stainless steel beads and Tissue Lyser LT (Qiagen) for 8 min at 50 471 strokes/minute, then treated with an ultrasonic bath for 5 minutes and subsequently centrifuged for 472 15 minutes at 23000xg and 4°C. The supernatants were transferred to Amicon® Ultra 2 ml 3K 473 centrifugal filter units (Merck). The remaining pellets were subjected to the same protocol of tissue 474 disruption and centrifugation using 1 mL of acidified water (FA 1%). Amicon® filters containing the 475 combined supernatants from the two-fold extraction process were centrifuged for 140 minutes at 476 7500xg and 23°C. Filtrates were used for liquid chromatography-mass spectrometry analysis. 477 Samples were separated by using a Dionex UltiMate 3000 HPLC system with an integrated variable 478 wavelength detector, set at 280 nm, and equipped with a C18 column (EC Nucleodur® C18 Gravity 479 column, 2 mm ID x 50 mm, 3 µm, Macherey & Nagel). Samples (5 µL) were injected at flow rate of 480 300 µL/minutes. A solvent gradient was run from 90% A (water containing 0.2% FA and 2 mM 481 ammonium acetate) and 10% B (methanol containing 2 mM ammonium acetate) to 50% A and 50% 482 B over 10 minutes. 483 The eluate was analyzed with a coupled mass spectrometer ESI-micrOTOF-Q (Bruker Daltonics). 484 Data were acquired in positive full scan MS mode with a scan range m/z 50-1000. Identification and 485 quantification of the xanthine derivatives were performed using Data Analysis software (Bruker 486 Daltonics). The limit of detection was 5 nM for caffeine and 10 nM for its metabolites (paraxanthine, 487 theobromine and theophylline). 488 489 Learning activation in the Morris water maze. An Atlantis Morris Water Maze (MWM) tank was 490 placed in a room with several visual extra-maze cues. Water opacified with powdered chalk (Blanc 491 de meudon) was maintained at a temperature of 21°C. Mice from water (control) and caffeine groups 492 were habituated to the set-up for two consecutive days (habituation 1 and 2). During habituation 1, 493 mice were allowed to discover the pool filled with 5 cm height of water and a visible platform during 494 60 seconds. During habituation 2, mice were allowed to swim in the pool filled with water in absence 495 21 of the platform for 60 seconds. The following 3 days (acquisition day 1–3), mice were trained to 496 localize the platform hidden underneath the opacified water using the spatial cues present in the 497 room. In each acquisition day, mice performed four trials each of 60 seconds maximal duration. Each 498 trial was terminated when the mouse reached the platform or after the 60 seconds. Mice failing to 499 find the platform were gently guided to the platform and allowed to stay for 8–10 seconds. During the 500 training days, mice were subjected to MWM in a random order, so that they were tested at different 501 times of the day. All MWM evaluations of caffeineor water-treated mice were performed by 502 experimenter blind to mouse treatments. 503 504 Sacrifice and brain tissue preparation. For transcriptomic analysis, mice from Learning group were 505 killed by cervical dislocation, one hour after the last training section, while mice from the Home cage 506 group were killed at the same time. Freshly dissected tissues were immediately frozen in liquid 507 nitrogen and kept at -80°C until RNA extraction. Similar sacrifice procedures were used for animals 508 used for proteomic and RTqPCR analyses. For molecular MALDI imaging experiments, mice were 509 deeply anesthetized with sodium pentobarbital (50 mg/kg, i.p.), and then transcardially perfused with 510 cold NaCl (0.9%). Brains were collected, frozen on dry ice and stored at -80°C until use. 511 512 RNA-seq analysis. Total RNA was extracted from dorsal hippocampal tissues using TRIzol reagent 513 (Invitrogen) (n=4/group). Freshly dissected tissue was chopped, homogenized in 300 μL of TRIzol 514 reagent, and frozen (20 minutes at -80°C), followed by 3-minutes centrifugation at 14000xg before 515 chloroform/isoamyl extraction. The supernatant was used to precipitate RNA with isopropanol and 516 RNase-free glycogen (30 minutes at 4°C). The pellet was washed once with 70% ethanol and 517 resuspended in Milli-Q water. A new RNA precipitation was performed with 100% ethanol and 3 M 518 sodium acetate (overnight at -20°C). After two further 70% ethanol washes, the pellet was air-dried 519 and resuspended in 30 μL nuclease-free Milli-Q water, heated 6 minutes at 50°C, and RNA 520 quantification was performed. RNA-seq libraries (n=4/group) were generated from 500 ng of total 521 22 RNA using Illumina® TruSeq® Stranded mRNA Library Prep Kit v2. Briefly, following purification with 522 poly-T oligo attached magnetic beads, the mRNA was fragmented using divalent cations at 94°C for 523 2 minutes. The cleaved RNA fragments were copied into first-strand cDNA using reverse 524 transcriptase and random primers. Strand specificity was achieved by replacing dTTP with dUTP 525 during the second-strand cDNA synthesis by DNA Polymerase I and RNase H. Following the addition 526 of a single “A” base and the subsequent ligation of the adapter on double-stranded cDNA fragments, 527 the products were purified and enriched with PCR [30 s at 98°C; (10 seconds at 98°C, 30 seconds 528 at 60°C, 30 seconds at 72°C) × 12 cycles; 5 minutes at 72°C] to create the cDNA library. Surplus 529 PCR primers were further removed by purification using AMPure XP beads (Beckman Coulter), and 530 the final cDNA libraries were checked for quality and quantified using capillary electrophoresis. 531 Sequencing was performed on the Illumina® Genome Hiseq4000 as single-end 50 base reads 532 following Illumina’s instructions. Reads were mapped onto the mm10 assembly of Mus musculus 533 genome using STAR v2.5.3a (60) and the Bowtie 2 aligner v2.2.8 (61). Only uniquely aligned reads 534 were kept for further analyses. Quantification of gene expression was performed using HTSeq-count 535 v0.6.1p1 (62) and gene annotations from Ensembl release 90 and “union” mode. Read counts were 536 normalized across libraries with the method proposed by Ander et al. (2010) (63). Comparisons of 537 interest were performed using the test for differential expression proposed by Love (64) and 538 implemented in the DESeq2 Bioconductor library (v1.16.1). Resulting p-values were adjusted for 539 multiple testing using the Benjamini and Hochberg method (65). 540 541 Chromatin Immunoprecipitation (ChIP). Freshly dissected tissue was chopped by a razor blade 542 and rapidly incubated in 1.5 mL phosphate-buffered saline (PBS) containing 1% formaldehyde for 543 10 minutes at room temperature. To stop fixation, glycine was added (0.125 M final concentration). 544 Dorsal hippocampi from 4 mice were pooled per sample and two biological replicates per condition 545 were used for the ChIP-seq. Tissue samples were then processed as described in Chatterjee et al. 546 (34) and sonicated using the Diagenode Bioruptor (30 seconds ON-30 seconds OFF at High Power 547 23 x 35 cycles). Sonicated chromatin was centrifuged 10 minutes at 14000xg, the supernatant collected 548 and diluted 1:10 in ChIP dilution buffer (0.01% SDS, 1.1% Triton X-100, 1.2 mM EDTA, 16.7 mM 549 Tris-Cl, pH 8.1, 167 mM NaCl). A fraction of the supernatant (50 µL – 10%) from each sample was 550 saved before immune-precipitation for ‘total input chromatin’. Supernatants were incubated overnight 551 (4°C) with 1/1000 primary antibodies against H3K9/14ac (Diagenode #C15410200) and H3K27ac 552 (Abcam #ab4729), followed by protein A Dynabeads (Invitrogen) for 2 hours at room temperature. 553 After several washes (low salt, high salt, LiCl and TE buffers), the resulting DNA-protein complexes 554 were eluted in 300 µL elution buffer (1% SDS, 0.1 M NaHCO3). The crosslinking was reversed 555 (overnight at 65°C) and the DNA was subsequently purified with RNAse (30 minutes at 37°C) and 556 proteinase K (2 hours at 45°C). DNA from the immunoprecipitated and input samples was isolated 557 using Diagenode MicroChIP DiaPure columns with 20 µL nuclease-free milliQ water in low binding 558 tubes. ChIP samples were further purified at the Genomeast Platform using Agencourt AMPure XP 559 beads (Beckman Coulter) and quantified using Qubit (Invitrogen). 560 561 ChIP-seq libraries and sequencing. ChIP-seq libraries were prepared from 2-10 ng of double562 stranded purified DNA using the MicroPlex Library Preparation kit v2 (C05010014, Diagenode s.a., 563 Seraing, Belgium), according to manufacturer's instructions. DNA was first repaired and yielded 564 molecules with blunt ends. Next, stem-loop adaptors with blocked 5’ ends were ligated to the 5’ end 565 of the genomic DNA (gDNA), leaving a nick at the 3’ end. The adaptors cannot ligate to each other 566 and do not have single-strand tails thus non-specific background is avoided. In the final step, the 3’ 567 ends of the gDNA were extended to complete library synthesis and Illumina compatible indexes were 568 added through a PCR amplification (4+7 cycles). Amplified libraries were purified and size-selected 569 using Agencourt AMPure XP beads (Beckman Coulter) to remove unincorporated primers and other 570 reagents. Prior to analyses, DNA libraries were checked for quality and quantified using a 2100 571 Bioanalyzer (Agilent). The libraries were loaded in the flowcell at 8 pM concentration, and clusters 572 24 were generated using the Cbot and sequenced using the Illumina HiSeq 4000 technology as single573 end 50 base reads following Illumina’s instructions. Image analysis and base calling were performed 574 using RTA and CASAVA. 575 576 ChIP-seq analyses. Sequenced reads were mapped to the Mus musculus genome assembly mm10 577 using Bowtie v1.0.0 with the following parameters «-m1-strata-best-y-l40». Samtools merge v1.3.1 578 (66) was used to combine biological replicates by condition. Then, BEDtools intersect v2.26.0 (67) 579 was used to remove reads located within ENCODE blacklisted regions. SICER (SICER-df.sh) v1.1 580 (68) was used to detect differentially bound regions on the pools of biological replicates using the 581 following parameters: «Species: mm10, Effective genome size as a fraction of reference genome: 582 0.74, Threshold for redundancy allowed for treated reads: 1, Threshold for redundancy allowed for 583 WT reads: 1, Window size: 200 bps, Fragment size: 200 bps, Gap size: 600 bps, FDR for 584 identification of enriched islands: 1E-2, FDR for identification of significant changes: 1E-2. Finally, 585 differentially bound regions were annotated with respect to the closest gene using Homer 586 annotatePeaks.pl v4.11.1 (69). An FDR of 1E-5 was used in differential analyses (caffeine vs. 587 control). 588 589 Neuronal and all cells isolation. Neuronal and all cells suspensions were obtained from mouse 590 hippocampus chronically treated with caffeine or water (control). For that, we used Neural Tissue 591 Dissociation (Miltenyi, #130-092-628) and Neuron Isolation Kits (Miltenyi, #130-115-389), following 592 manufacturer's instructions with some adaptations. Briefly, two mouse hippocampi were pooled per 593 sample and harvested in a pre-heated buffer solution containing papain. This was followed by series 594 of manual mechanical dissociations, using scissors and fire polished Pasteur pipettes of descending 595 diameter, and incubations at 37°C under slow rotation. The solution was then filtered (50 µm) and 596 centrifuged (10 minutes, 300xg, at room temperature) and myelin was removed using Myelin 597 Removal Beads II kit (Miltenyi, #130-096-733), incubating for 15 minutes at 4°C, centrifuging (10 598 25 minutes, 300xg at 4°C) and filtering the sample through MS columns (Miltenyi, #130-042-201) placed 599 in MiniMACS™ Separator (Miltenyi, #130-042-102) to collect the myelin depleted flow-through, free 600 of cell debris. The ‘all cells’ suspension was collected at this point and counted using the TC20 601 Automated Cell Counter (Bio-Rad, #1450102) to obtain a total of 70,000 cells per sample. With the 602 remaining of the samples, we proceeded with neuronal isolation according to manufacturer's 603 instructions, finally depleting the samples through MS columns to collect the flow-through enriched 604 in neurons. The samples were counted and 70,000 cells per sample were taken for CUT&Tag 605 experiments. 606 607 Cleavage Under Targets and Tagmentation (CUT&Tag). Having isolated all cells and neuronal 608 populations we proceeded with CUT&Tag method to assess their genome-wide H3K27ac and 609 H3K27me3 chromatin state. The protocol was adapted from that described by Kaya-Okur et al., 2019 610 (31) The method is based on digitonin-induced cell permeabilization (Sigma, #300410-250MG) and 611 concanavalin A-coated magnetic beads (Cell signaling, #93569S) immobilization. This is followed by 612 over-night incubation at 4°C with primary antibodies against H3K27ac (Abcam, #ab4729) and 613 H3K27me3 (Diagenode, #C15410195), followed by 1 hour incubation with the secondary antibody 614 (Antibodies online, #ABIN101961). The loaded-Tn5 is then added (Diagenode, #C01070001) and 615 the cleaved DNA is extracted using MinElute PCR Purification Kit (Quiagen, # 28004). Library 616 preparation was conducted using Nextera primers (Illumina, #FC-131-2001) and post-PCR clean-up 617 using SPRI bead slurry (Beckmann Coulter, #B23317). Concentration of the collected DNA was 618 achieved by Qubit (Invitrogen, #Q32851). Two biological replicates were used per group and Rabbit 619 IgG (Diagenode #C15410206) was used as control. 620 621 CUT&Tag analyses. Reads (paired-end) were mapped to Mus musculus genome (assembly mm10) 622 using Bowtie2 (61) v2.2.8 with default parameters except for “–end-to-end-very-sensitive-no-mixed 623 –no-discordant-I10-X700”. Prior to peak calling, reads with mapping quality below 30 were removed 624 32 References 763 764 1. Smith A. Effects of caffeine on human behavior.. Food Chem Toxicol. 2002;40(9):1243–1255. 765 2. Kim Y, Je Y, Giovannucci E. Coffee consumption and all-cause and cause-specific mortality: a 766 meta-analysis by potential modifiers.. Eur J Epidemiol. 2019;34(8):731–752. 767 3. Loftfield E et al. Association of Coffee Drinking With Mortality by Genetic Variation in Caffeine 768 Metabolism: Findings From the UK Biobank. JAMA Intern Med. 2018;178(8):1086–1097. 769 4. Freedman ND, Park Y, Abnet CC, Hollenbeck AR, Sinha R. Association of coffee drinking with 770 total and cause-specific mortality. N Engl J Med. 2012;366(20):1891–1904. 771 5. Flaten V et al. From epidemiology to pathophysiology: what about caffeine in Alzheimer’s disease? 772 Biochem Soc Trans. 2014;42(2):587–592. 773 6. Cunha RA. How does adenosine control neuronal dysfunction and neurodegeneration?. J 774 Neurochem. 2016;139(6):1019–1055. 775 7. Cellai L et al. The Adenosinergic Signaling: A Complex but Promising Therapeutic Target for 776 Alzheimer’s Disease. Front Neurosci. 2018;12:520. 777 8. Wright GA et al. Caffeine in Floral Nectar Enhances a Pollinator's Memory of Reward. Science. 778 2013;339(6124):1202– 1204. 779 9. Marques S, Batalha VL, Lopes LV, Outeiro TF. Modulating Alzheimer’s disease through caffeine: 780 a putative link to epigenetics.. J Alzheimers Dis. 2011;24 Suppl 2:161–171. 781 10. Angelucci MEM, Cesário C, Hiroi RH, Rosalen PL, Da Cunha C. Effects of caffeine on learning 782 and memory in rats tested in the Morris water maze. Brazilian J Med Biol Res. 2002;35(10):1201– 783 1208. 784 11. Borota D et al. Post-study caffeine administration enhances memory consolidation in humans. 785 Nat Neurosci. 2014;17(2):201–203. 786 12. Lopes JP, Pliássova A, Cunha RA. The physiological effects of caffeine on synaptic transmission 787 and plasticity in the mouse hippocampus selectively depend on adenosine A(1) and A(2A) receptors. 788 Biochem Pharmacol. 2019;166:313–321. 789 13. Costenla AR, Cunha RA, de Mendonça A. Caffeine, adenosine receptors, and synaptic plasticity. 790 J Alzheimers Dis. 2010;20 Suppl 1:S25-34. 791 14. Simons SB, Caruana DA, Zhao M, Dudek SM. Caffeine-induced synaptic potentiation in 792 hippocampal CA2 neurons. Nat Neurosci. 2011;15(1):23–25. 793 15. Lao-Peregrín C et al. Caffeine-mediated BDNF release regulates long-term synaptic plasticity 794 through activation of IRS2 signaling. Addict Biol. 2017;22(6):1706–1718. 795 33 16. Blaise JH, Park JE, Bellas NJ, Gitchell TM, Phan V. Caffeine consumption disrupts hippocampal 796 long-term potentiation in freely behaving rats. Physiol Rep. 2018;6(5):e13632. 797 17. Watanabe Y, Ikegaya Y. Caffeine Increases Hippocampal Sharp Waves in Vitro. Biol Pharm Bull. 798 2017;40(7):1111–1115. 799 18. Hanajima R et al. Effect of caffeine on long-term potentiation-like effects induced by quadripulse 800 transcranial magnetic stimulation. Exp Brain Res. 2019;237(3):647–651. 801 19. Kerkhofs A et al. Caffeine Controls Glutamatergic Synaptic Transmission and Pyramidal Neuron 802 Excitability in Human Neocortex. Front Pharmacol. 2017;8:899. 803 20. Fredholm BB, Bättig K, Holmén J, Nehlig A, Zvartau EE. Actions of caffeine in the brain with 804 special reference to factors that contribute to its widespread use. Pharmacol Rev. 1999;51(1):83– 805 133. 806 21. Heintzman ND et al. Distinct and predictive chromatin signatures of transcriptional promoters and 807 enhancers in the human genome. Nat Genet. 2007;39(3):311–318. 808 22. Hnisz D et al. Super-enhancers in the control of cell identity and disease. Cell 2013;155(4):934– 809 947. 810 23. J. PSC et al. Chromatin stretch enhancer states drive cell-specific gene regulation and harbor 811 human disease risk variants. Proc Natl Acad Sci. 2013;110(44):17921–17926. 812 24. Karmodiya K, Krebs AR, Oulad-Abdelghani M, Kimura H, Tora L. H3K9 and H3K14 acetylation 813 co-occur at many gene regulatory elements, while H3K14ac marks a subset of inactive inducible 814 promoters in mouse embryonic stem cells. BMC Genomics 2012;13:424. 815 25. Sánchez-Sarasúa S et al. IRS1 expression in hippocampus is age-dependent and is required for 816 mature spine maintenance and neuritogenesis. Mol Cell Neurosci. 2022;118:103693. 817 26. Koopmans F et al. SynGO: An Evidence-Based, Expert-Curated Knowledge Base for the 818 Synapse. Neuron 2019;103(2):217-234.e4. 819 27. Sheng M, Kim E. The Shank family of scaffold proteins. J Cell Sci. 2000;113(11):1851–1856. 820 28. Mundel P et al. Synaptopodin: an actin-associated protein in telencephalic dendrites and renal 821 podocytes. J Cell Biol. 1997;139(1):193–204. 822 29. Parra-Damas A et al. CRTC1 Function During Memory Encoding Is Disrupted in 823 Neurodegeneration. Biol Psychiatry 2017;81(2):111–123. 824 30. Wang L, Walter P. Msp1/ATAD1 in Protein Quality Control and Regulation of Synaptic Activities. 825 Annu. Rev. Cell Dev Biol. 2020;36(1):141–164. 826 31. Kaya-Okur HS et al. CUT&Tag for efficient epigenomic profiling of small samples and single cells. 827 Nat Commun. 2019;10(1):1930. 828 32. Han S et al. Regulation of dendritic spines, spatial memory, and embryonic development by the 829 34 TANC family of PSD-95-interacting proteins. J Neurosci. 2010;30(45):15102–15112. 830 33. Parra-Damas A, Rubió-Ferrarons L, Shen J, Saura CA. CRTC1 mediates preferential 831 transcription at neuronal activity-regulated CRE/TATA promoters. Sci Rep. 2017;7(1):18004. 832 34. Chatterjee S et al. Reinstating plasticity and memory in a tauopathy mouse model with an 833 acetyltransferase activator. EMBO Mol Med. 2018;10(11). 834 35. Martínez G et al. Regulation of Memory Formation by the Transcription Factor XBP1. Cell Rep. 835 2016;14(6):1382–1394. 836 36. Cao L et al. VEGF links hippocampal activity with neurogenesis, learning and memory. Nat 837 Genet. 2004;36(8):827–835. 838 37. Mews P et al. Acetyl-CoA synthetase regulates histone acetylation and hippocampal memory. 839 Nature 2017;546(7658):381–386. 840 38. Jacobson KA, von Lubitz DK, Daly JW, Fredholm BB. Adenosine receptor ligands: differences 841 with acute versus chronic treatment. Trends Pharmacol Sci. 1996;17(3):108–113. 842 39. Ferré S. An update on the mechanisms of the psychostimulant effects of caffeine. J Neurochem. 843 2008;105(4):1067–1079. 844 40. Doepker C et al. Caffeine: Friend or Foe? Annu Rev Food Sci Technol. 2016;7:117–137. 845 41. Yu L et al. Uncovering multiple molecular targets for caffeine using a drug target validation 846 strategy combining A2A receptor knockout mice with microarray profiling. Physiol Genomics 847 2009;37(3):199–210. 848 42. Svenningsson P, Nomikos GG, Fredholm BB. The stimulatory action and the development of 849 tolerance to caffeine is associated with alterations in gene expression in specific brain regions. J 850 Neurosci. 1999;19(10):4011–4022. 851 43. Magalhães R et al. Habitual coffee drinkers display a distinct pattern of brain functional 852 connectivity. Mol Psychiatry 2021;26(11):6589–6598. 853 44. A. KK et al. TORC1 is a calciumand cAMP-sensitive coincidence detector involved in 854 hippocampal long-term synaptic plasticity. Proc Natl Acad Sci. 2007;104(11):4700–4705. 855 45. Chang D et al. Caffeine Caused a Widespread Increase of Resting Brain Entropy. Sci Rep. 856 2018;8(1):2700. 857 46. Tal O et al. Caffeine-Induced Global Reductions in Resting-State BOLD Connectivity Reflect 858 Widespread Decreases in MEG Connectivity. Front Hum Neurosci. 2013;7:63. 859 47. Koppelstaetter F et al. Does caffeine modulate verbal working memory processes? An fMRI 860 study. Neuroimage 2008;39(1):492–499. 861 48. Cunha RA. Different cellular sources and different roles of adenosine: A1 receptor-mediated 862 inhibition through astrocytic-driven volume transmission and synapse-restricted A2A receptor863 35 mediated facilitation of plasticity. Neurochem Int. 2008;52(1–2):65–72. 864 49. Lopes L V, Cunha RA, Ribeiro JA. Cross talk between A(1) and A(2A) adenosine receptors in 865 the hippocampus and cortex of young adult and old rats. J Neurophysiol. 1999;82(6):3196–3203. 866 50. Dassesse D, Ledent C, Parmentier M, Schiffmann SN. Acute and chronic caffeine administration 867 differentially alters striatal gene expression in wild-type and adenosine A(2A) receptor-deficient 868 mice. Synapse 2001;42(2):63–76. 869 51. Burns AM, Gräff J. Cognitive epigenetic priming: leveraging histone acetylation for memory 870 amelioration. Curr Opin. Neurobiol. 2021;67:75–84. 871 52. Duarte JMN, Cunha RA, Carvalho RA. Adenosine A₁ receptors control the metabolic recovery 872 after hypoxia in rat hippocampal slices. J Neurochem. 2016;136(5):947–957. 873 53. Laurent C et al. Beneficial effects of caffeine in a transgenic model of Alzheimer’s disease-like 874 tau pathology. Neurobiol Aging 2014;35(9):2079–2090. 875 54. Arendash GW et al. Caffeine protects Alzheimer’s mice against cognitive impairment and reduces 876 brain beta-amyloid production. Neuroscience 2006;142(4):941–952. 877 55. David B, V. LL. Stabilizing synapses. Science. 2021;374(6568):684–685. 878 56. G. SC et al. Adenosine Receptor Antagonists Including Caffeine Alter Fetal Brain Development 879 in Mice. Sci Transl Med. 2013;5(197):197ra104-197ra104. 880 57. Ferran G-C et al. Convergence of adenosine and GABA signaling for synapse stabilization during 881 development. Science. 2022;374(6568):eabk2055. 882 58. Arendash GW et al. Caffeine reverses cognitive impairment and decreases brain amyloid-beta 883 levels in aged Alzheimer’s disease mice. J Alzheimers Dis. 2009;17(3):661–680. 884 59. Laurent C et al. Beneficial effects of caffeine in a transgenic model of Alzheimer’s disease-like 885 tau pathology. Neurobiol Aging 2014;35(9):2079–2090. 886 60. Dobin A et al. STAR: ultrafast universal RNA-seq aligner. Bioinformatics 2013;29(1):15–21. 887 61. Langmead B, Salzberg SL. Fast gapped-read alignment with Bowtie 2. Nat Methods 888 2012;9(4):357–359. 889 62. Anders S, Pyl PT, Huber W. HTSeq--a Python framework to work with high-throughput 890 sequencing data. Bioinformatics 2015;31(2):166–169. 891 63. Anders S, Huber W. Differential expression analysis for sequence count data. Genome Biol. 892 2010;11(10):R106. 893 64. Love MI, Huber W, Anders S. Moderated estimation of fold change and dispersion for RNA-seq 894 data with DESeq2. Genome Biol. 2014;15(12):550. 895 65. Benjamini Y, Hochberg Y. Controlling the False Discovery Rate - a Practical and Powerful 896 Approach to Multiple Testing. J R Stat Soc Ser B-Methodological. 1995;57(1):289–300. 897 36 66. Li H et al. The Sequence Alignment/Map format and SAMtools. Bioinformatics 898 2009;25(16):2078–2079. 899 67. Quinlan AR, Hall IM. BEDTools: a flexible suite of utilities for comparing genomic features. 900 Bioinformatics 2010;26(6):841–842. 901 68. Xu S, Grullon S, Ge K, Peng W. Spatial clustering for identification of ChIP-enriched regions 902 (SICER) to map regions of histone methylation patterns in embryonic stem cells. Methods Mol Biol. 903 2014;1150:97–111. 904 69. Zang C et al. A clustering approach for identification of enriched domains from histone 905 modification ChIP-Seq data. Bioinformatics 2009;25(15):1952–1958. 906 70. Amemiya HM, Kundaje A, Boyle AP. The ENCODE Blacklist: Identification of Problematic 907 Regions of the Genome. Sci Rep. 2019;9(1):9354. 908 71. Heinz S et al. Simple combinations of lineage-determining transcription factors prime cis909 regulatory elements required for macrophage and B cell identities. Mol Cell 2010;38(4):576–589. 910 911 37 912 38 Figure 1. Hippocampal epigenomic alterations associated with chronic caffeine consumption. 913 (A) Volcano plot showing the differential enriched genomic regions of H3K9/14ac (ChIP-seq) upon 914 chronic caffeine treatment (778 decreased and 3 increased peaks). Red dots represent the 915 significant different regions (FDR<1E-5). (B) Genomic Regions Enrichment of Annotations Tool 916 (GREAT) analysis showing the most enriched biological processes associated with the H3K9/14ac 917 decreased peaks in caffeine-treated mice. Blue arrows point towards metabolic processes and 918 translation related terms. (C) Volcano plot representing the differentially regulated regions of 919 H3K27ac upon chronic caffeine treatment (2105 decreased and 4 increased peaks, with FDR<1E920 5). (D) GREAT analysis representing the most common biological processes associated with the 921 H3K27ac decreased peaks in the caffeine group. Regulation of metabolic processes are indicated 922 by the blue arrows. (E) KEGG pathway analyses of depleted regions of both histone marks. Dashed 923 grey line indicates the significant adjusted p-value <0.05. (F) Functional protein-protein network 924 analysis (STRING) representation of insulin and glucagon-related genes found decreased in both 925 histone acetylation marks. (G) Representation of the genomic regions (IGV) of the metabolic genes 926 Irs1 and Gsk3b showing significant decrease of H3K27ac and H3K9/14ac after caffeine treatment 927 (Irs1 H3K27ac FDR=1.82E-12; H3K9/14ac FDR=7.75E-05; Gsk3b H3K27ac FDR=4.83E-05; 928 H3K9/14ac FDR=2.58E-11). Two biological replicates per histone mark were used for ChIP-seq 929 experiments. 930 931 39 932 933 934 935 936 937 938 939 940 40 Figure 2. Hippocampal metabolomic changes induced by chronic caffeine consumption. 941 Unsupervised principal component analysis (PCA) performed in the hippocampal region of interest 942 delimitated in yellow on the Nissl staining of the brain tissue section (A). Scores from the 943 unsupervised PCA in the hippocampus of Water- (in blue) and Caffeine-treated mice (in red) are 944 presented in a plot where the differences between the molecular signatures of the two experimental 945 groups clearly emerge (B). Pie charts showing the distribution of the different classes of molecules 946 (C) and their abundance changes (D) of m/z measured in positive or negative ionization modes with 947 a significant quantitative difference after the Student’s t-test analysis in the hippocampus of Caffeine948 compared to Water-treated animals (N = 6/group). (E) Mass spectrometry images obtained at a 949 spatial resolution of 35 µm for m/z presenting a decreased (green) or increased (orange) density in 950 the hippocampus of Caffeine-treated compared to Water-treated mice. The color scale shows the 951 intensity of the m/z of interest. Cer, ceramide; PC, phosphatidylcholine; PI, phosphatidylinositol; PS, 952 phosphatidylserine. 953 954 955 956 957 958 959 960 41 961 962 963 964 965 966 967 968 969