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

The study investigates how habitual (chronic) caffeine consumption modulates molecular pathways in the hippocampus, a key brain region for learning and memory. While caffeine is widely consumed and associated epidemiologically with reduced all-cause mortality and benefits in aging and neurodegenerative conditions, most mechanistic studies focus on acute exposure. The authors hypothesize that chronic caffeine intake induces coordinated, cell type–specific changes across the epigenome, transcriptome, proteome, and metabolome that affect metabolism and synaptic plasticity, thereby influencing learning-dependent transcriptional responses.

Prior work shows inverse associations between caffeine/coffee consumption and mortality, and suggests benefits in aging, Alzheimer’s disease, and other neuropsychiatric conditions. Experimental studies report cognition-enhancing effects of caffeine across species: improved memory in honeybees after caffeine reward, enhanced memory performance in rats following acute caffeine, and improved discrimination in humans when caffeine is administered post-learning. In hippocampal slices and rodent hippocampus, caffeine enhances basal synaptic transmission and modulates LTP and sharp wave-ripple complexes, proposed mechanisms for memory consolidation. Human studies show LTP-like effects and modulation of cortical excitability following caffeine. Mechanistically, caffeine antagonizes adenosine receptors (A1R and A2AR), which have opposing effects on synaptic transmission. However, adaptive downstream molecular pathways engaged by chronic consumption have been largely overlooked, motivating a multi-omics investigation of chronic caffeine effects in the hippocampus.

Design and treatment: Male C57BL6/J mice were housed under standard conditions. Chronic caffeine was administered via drinking water at 0.3 g/L (dose reflecting moderate human intake) for 2 weeks; controls received water. Average intake was 4.83 ± 0.15 mL/mouse/day. Additional groups were used for acute exposure (24 h caffeine), and for withdrawal (2 weeks caffeine followed by 2 weeks water). Brain concentrations of caffeine and metabolites (paraxanthine, theobromine, theophylline) were quantified by LC-MS. Learning paradigm: A Morris water maze (MWM) protocol induced hippocampus-dependent learning. Mice were habituated for 2 days, then trained for 3 days (4 trials/day) to locate a hidden platform. One hour after the last trial on day 3, dorsal hippocampi were collected for RNA-Seq. Behavioral measures included distance traveled, thigmotaxis, and swim speed. Epigenomics (bulk ChIP-Seq): Dorsal hippocampi from water- and caffeine-treated mice were crosslinked, chromatin was sonicated, and ChIP performed with antibodies to H3K27ac and H3K9/14ac. Two biological replicates per condition per mark were sequenced (Illumina HiSeq 4000, SE50). Reads were aligned to mm10; peaks and differentially enriched regions were called with SICER; ENCODE blacklist filtering applied. GREAT, KEGG, STRING, IPA were used for pathway and upstream regulator analyses. Cell type–specific epigenomics (CUT&Tag-Seq): Hippocampi were dissociated; “all cells” and neuron-enriched suspensions (70,000 cells/sample) were prepared (Miltenyi kits). CUT&Tag profiled H3K27ac (active) and H3K27me3 (repressive) marks (two biological replicates per group), with Bowtie2 alignment, quality filtering (MAPQ ≥30), ENCODE blacklist removal, and SICER for differential regions. Functional enrichment used GREAT, DAVID, KEGG, STRING; IPA identified upstream regulators. Transcriptomics (RNA-Seq): Total RNA (n=4/group) from dorsal hippocampus in home cage (resting) and learning conditions was extracted (TRIzol), libraries prepared (Illumina TruSeq Stranded mRNA), sequenced (HiSeq 4000, SE50), mapped (STAR/Bowtie2), quantified (HTSeq-count), normalized (Anders et al.), and differential expression tested (DESeq2, BH adjustment). Gene set z-scores were computed for histone-depleted gene sets. Proteomics (LC-MS/MS): Dorsal hippocampal proteins (n=3/group) were processed by FASP, high-pH fractionated, separated by nanoLC, and analyzed on a Q Exactive Orbitrap. Mascot searches against SwissProt mouse (1% FDR) identified and quantified proteins. STRING and SynGO annotated functional clusters and synaptic localization. Withdrawal effects were assessed in separate groups. Metabolomics (MALDI-MSI): Coronal sections at bregma −1.7 mm (n=6/group) were matrix-coated (1,5-DAN for negative mode; 2,5-DHB for positive mode) and analyzed on a 7T MALDI-FTICR at 35 μm resolution. PCA distinguished groups; Student’s t test (P<0.05) identified discriminant m/z features; annotations used METLIN and HMDB (±10 ppm). Histological Nissl overlays aided regional interpretation. Statistics and approvals: Appropriate multiple-testing controls (BH FDR) and cutoffs (e.g., FDR <1×10⁻⁵ for ChIP-Seq, <1×10⁻⁴ for CUT&Tag) were applied. All animal procedures were ethically approved (CEEA75). Sequencing data deposited in GEO (GSE167123).

• Exposure and brain levels: Chronic 0.3 g/L caffeine for 2 weeks yielded brain caffeine 3.6 ± 1.1 μM; metabolites detected: paraxanthine 1.9 ± 0.4 μM, theobromine 1.8 ± 0.3 μM, theophylline 0.10 ± 0.03 μM (n=5).
• Bulk hippocampal epigenomics: Strong global decreases in active chromatin marks at metabolism-related loci with chronic caffeine. • H3K9/14ac: 778 regions decreased, 3 increased (FDR < 1×10⁻⁵). • H3K27ac: 2105 regions decreased, 4 increased (FDR < 1×10⁻⁵). • Enrichments among decreased peaks: regulation of amide/lipid metabolic processes, translation, mRNA transport; KEGG: insulin signaling (both marks), glucagon signaling (H3K9/14ac), circadian entrainment, cAMP/MAPK/Rap1 pathways. • Example loci: Irs1 and Gsk3b showed significant H3K27ac/H3K9/14ac depletion (e.g., Irs1: H3K9/14ac FDR = 7.75×10⁻⁶; H3K27ac FDR = 1.82×10⁻¹²; Gsk3b: H3K9/14ac FDR = 2.58×10⁻¹¹; H3K27ac FDR = 4.83×10⁻⁷). • IPA upstream regulators among depleted genes included TCF7L2 and ADORA2A (A2AR).
• Bulk transcriptomics linkage: While individual H3K27ac-depleted genes were not significantly changed at FDR threshold, the set of 1776 H3K27ac-depleted genes showed a significant reduction in expression (z score) in caffeine vs water; qPCR validated decreased expression for several targets (e.g., Pbx1, Nadk2, Spic1), with Cyp51 reduced only after chronic exposure and some effects persisting post-withdrawal.
• Metabolomics (MALDI-MSI): PCA cleanly separated groups. Overall, 59% of features were assigned (27% metabolites, 32% lipids). A striking 92% of discriminant metabolites/lipids decreased (8% increased) in caffeine vs water in hippocampus, indicating reduced metabolic tone.
• Proteomics: 179 proteins altered (49 decreased, 130 increased). • Decreased proteins linked to peptide/cellular amide metabolism and mitochondria (e.g., NDUFA3, MPC1, ACSL4) suggesting reduced energy metabolism; 35/49 normalized after withdrawal, 14 remained decreased (e.g., IGF2R). • Increased proteins formed three clusters: RNA binding/spliceosome; autophagosome/ER protein processing; glutamatergic synapse/phosphatase activity. SynGO annotated many as synaptic signaling/chemical synaptic transmission (e.g., SHANK3, SYNPO, CRTC1). Post-withdrawal, 57/130 reverted, 73 persisted (e.g., ATAD1).
• Neuron-specific epigenomics (CUT&Tag): In neuron-enriched populations, caffeine increased H3K27ac at synaptic genes and decreased H3K27me3 at ion transport/learning genes. • H3K27ac: 7127 enriched vs 4343 depleted regions (FDR < 1×10⁻⁵); enriched peaks mapped to synapse, postsynaptic density, axon/dendrite; processes: synaptic plasticity, LTP, memory, action potential. • H3K27me3: 2734 enriched vs 1712 depleted regions; depleted regions associated with ion transmembrane transport (Ca2+, K+), chemical synaptic transmission, learning. • Overlap showed 352 regions enriched in acetylation and depleted in methylation tied to ion transport functions. Integration with proteomics identified 28 caffeine-increased proteins with H3K27ac enrichment at their genes, mostly glutamatergic synapse-related (e.g., PITPNM3, TANC1, CRTC1).
• Learning-dependent transcriptomics: Caffeine amplified the hippocampal transcriptional response to learning (MWM training). • Water: 209 genes changed (47 down, 162 up) in learning vs home cage. • Caffeine: 1139 genes changed (419 down, 720 up) in learning vs home cage (~5-fold increase). • Genes downregulated by caffeine during learning enriched for KEGG “ribosome,” consistent with prior translation-related epigenomic signatures (and upstream regulator MKNK1). • Genes upregulated by learning under caffeine showed increased significance for transcription-related processes; immediate early genes (e.g., Fosb) and Xbp1 were more activated. Among 607 genes specifically upregulated with caffeine+learning were Vegfa and Acss1. • Notably, 121 genes upregulated by learning under caffeine had been H3K27ac-depleted and showed reduced expression at rest, and were enriched for metabolic process regulation, indicating basal fine-tuning enabling strong inducibility during learning.

Chronic caffeine induces a dual, cell type–differential molecular program in the hippocampus. In bulk tissue at rest, caffeine reduces active histone acetylation and expression of metabolism-related genes, decreases lipid and metabolite abundance, and lowers mitochondrial/peptide metabolism proteins, suggesting a downshift of metabolic processes. Conversely, in neurons, caffeine enriches active chromatin (H3K27ac) and reduces repressive marks (H3K27me3) at synaptic transmission, plasticity, and ion transport genes, paralleling increased levels of glutamatergic synapse proteins. Functionally, this “primed” neuronal state manifests as an amplified transcriptomic response to learning, with greater induction of activity-dependent and transcriptional programs. Together, the findings support a model wherein regular caffeine intake improves the signal-to-noise ratio during information encoding by dampening basal metabolic activity (likely in non-neuronal/glial compartments) while enhancing neuronal readiness for activity-dependent plasticity. This aligns with human imaging showing reduced resting connectivity and increased task-evoked activation with caffeine, and with adenosine receptor–mediated bidirectional control of synaptic transmission.

The study demonstrates that habitual caffeine intake exerts coordinated, genome-wide effects across omic layers in the adult hippocampus: it downregulates metabolism-related pathways in bulk tissue at rest while enhancing neuron-specific epigenetic readiness and synaptic protein abundance, leading to amplified learning-dependent transcription. These dual effects suggest that caffeine fine-tunes basal metabolism to increase processing efficiency and salience during learning. Future work should delineate the precise cellular contributors (astrocytes, oligodendrocytes, microglia vs neurons), causal roles of adenosine receptor subtypes and upstream regulators (e.g., TCF7L2, MKNK1), persistence and reversibility of changes, dose–response relationships, and translational relevance to human cognition and neurodegenerative disease contexts.

• Several analyses were performed on bulk hippocampal tissue, limiting cell-type resolution for metabolism-related decreases; neuron-specific effects were profiled epigenetically but not across all omic layers at single-cell resolution.
• ChIP-Seq and CUT&Tag used two biological replicates per condition, which may limit sensitivity for subtle effects.
• Behavioral assessments focused on training-induced transcription rather than comprehensive cognitive outcome measures; direct links between molecular changes and memory performance were not established.
• Findings are in mice; extrapolation to humans requires caution regarding dose, exposure duration, and interspecies differences.
• Some gene/protein identifications and pathway annotations rely on database mapping of m/z and network inferences, which may include ambiguities.