Genomic Convergence in Hibernating Mammals Elucidates the Genetics of Metabolic Regulation in the Hypothalamus

Abstract

Extreme metabolic adaptations can elucidate genetic programs governing mammalian metabolism. Here we used convergent evolutionary changes in hibernating lineages to define conserved cis-regulatory elements (CREs) and metabolic programs. We characterized mouse hypothalamus gene expression and chromatin dynamics across fed, fasted, and refed states, then used comparative genomics of hibernating versus non-hibernating lineages to identify cis-elements with convergent changes in hibernators. Multi-omics approaches pinpointed CREs, hub genes, regulatory programs, and cell types underlying lineage divergence. Hibernators accumulated loss-of-function effects for CREs regulating hypothalamic responses, and the refeeding period after fasting served as a key phase for molecular processes with convergent evolutionary changes. This work provides a genetic framework for harnessing hibernator adaptations to understand human metabolic control.

Species differences in metabolic traits offer a lens into the genetic control of metabolism. Hibernation is an extreme metabolic phenotype that evolved recurrently across mammals (1, 2). Obligate hibernators undergo seasonal behavioral and metabolic shifts (Fig. 1A), gaining up to 50% body mass (3) and entering prolonged torpor sustained by fat reserves (1, 2), whereas facultative hibernators exhibit short torpor bouts in response to stress (1, 4). In contrast, homeotherms maintain stable metabolic rates year-round (1, 2). Hibernator adaptations offer insights into metabolic control, obesity, neuroprotection (5), reversal of neurodegeneration (6-8), improved insulin sensitivity (3, 9, 10), tumor dormancy (11, 12), and extended lifespan (13).

Fig. 1. Dynamics of metabolic states, gene expression, and chromatin accessibility across stages of the Fed-Fasted-Refed (FFR) response in the mouse hypothalamus.

(A) Schematic of the Fed-Fasted-Refed (FFR) cycle in obligate hibernators, highlighting metabolic states, processes, and seasonal cycles. Facultative hibernators show acute torpor responses to food deprivation and/or cold, rather than seasonal cycles. Homeotherms show little metabolic and physiological flexibility, though mice are facultative heterotherms capable of brief stress-induced torpor bouts. (B) Internal body temperature and respiratory exchange ratio (RER) in adult female mice across Fed (24°C, yellow), Fasted + Cold (18°C, blue), and Refed (green) states over 10 days. Red and orange dashed lines represent mean RER for dark and light cycles, respectively. Black p-values are fed vs. refed contrasts and purple p-values are Day 1 vs. 9 contrasts. Data are mean ± SEM (n = 10). See fig. S1 for further data. (C) Heatmap of RNA-Seq gene expression in the adult mouse hypothalamus for differentially expressed genes across FFR states (FDR <5%, n=3-4), showing fold change scaled by row. FA: fasted; RF: refed; 12hrs RF: refed 12 hours after 72 hours fasting; 1w RF: refed 1 week after 72 hours fasting; h: hours; w: weeks; Cold: 18°C vs. 24°C ambient temperature. (D) Circular plot showing the number of significant differentially expressed genes (FDR < 5%) at each step of the FFR cycle in the mouse hypothalamus, with bar and arrow colors indicating gene expression comparisons. (E and F) The heatmap (E) shows genes with increased expression from 72-hour fasted to 1-hour refed, and the bar plot (F) highlights KEGG pathways enriched for these DEGs (FDR < 5%). (G) Signal plots of ATAC-Seq data where the highlighted area indicates a significant peak in the 12-hr refed condition (n = 8, FDR<5%). (H) Venn diagram showing significant ATAC-Seq peaks genome-wide in the hypothalamus, with increased chromatin accessibility in refeeding (RF) compared to Fed and 72-hour fasted (FA) states (Genrich R package; FDR < 5%, n = 8).

Accelerated evolution—rapid nucleotide changes in conserved DNA—can highlight cis-regulatory elements (CREs) underlying species-specific adaptations (14). Conserved elements with accelerated evolution in lineages with biomedically important traits have pinpointed relevant CREs (15, 16). These, and other studies (17-19), demonstrate that convergent evolution in different lineages can reveal the genetic mechanisms of complex traits. Here, we used comparative genomics to identify CREs linked to metabolic and hibernation-relevant phenotypes. Focusing on the hypothalamus—a central regulator of metabolism, feeding, thermogenesis, and torpor (20-22)—we profiled gene expression and chromatin accessibility in mice across fed, fasted, and refed (FFR) states. Using Zoonomia genome alignments (23), we identified CREs with convergent accelerated evolution in hibernators, linked them to hub genes driving FFR responses, and mapped their activity across hypothalamic cell types. A companion study confirms their in vivo relevance (24), providing a framework for understanding CRE-based regulation of mammalian metabolism.

Genetic Programs for Hypothalamic Responses to Fed-Fasted-Refed States

Hibernation enables survival during food scarcity. We hypothesized that obligate hibernators evolved distinct hypothalamic genetic programs influencing FFR responses (Fig. 1A). Mice, as homeotherms capable of facultative heterothermy and fasting-induced torpor (25), provide a genetic model for studying conserved pathways in metabolic suppression, thermoregulation, and energy conservation.

To characterize metabolic and behavioral responses to FFR states, adult female mice were implanted with internal body temperature monitors and placed in CLAMS (comprehensive lab animal monitoring system) metabolic cages for a FFR paradigm. Mice were fed for 48 hours (fed phase), fasted for 48 hours at a reduced ambient temperature (25°C → 18°C) to induce torpor (torpor phase), and refed for six days at 25°C (refed phase) (Fig. 1B, S1A). During fasting, body temperature dropped to ~20°C, indicative of torpor, and returned to baseline within hours of refeeding. The respiratory exchange ratio increased post-torpor, indicating excessive CO₂ production and energy consumption (Fig. 1B). Metabolic rate, locomotion, and energy expenditure declined during refeeding, remaining suppressed for at least six days (fig. S1A-C).

To examine how torpor alters hypothalamic gene expression, we performed RNA-Seq across FFR phases, conducting sequential comparisons: (1) fed vs. 24-hr fasting, (2) 24-hr vs. 72-hr fasting, (3) 72-hr fasting vs. fasting + cold (18°C), (4) 72-hr fasting vs. 1-hr refeeding, (5) 1-hr vs. 12-hr refeeding, (6) 12-hr vs. 1-week refeeding, and (7) 1-week refeeding vs. fed baseline. Whereas the transition from fed to fasted caused modest gene expression changes, prolonged fasting (24–72 hrs) affected thousands of genes (Fig. 1C, D). Cold exposure (18°C) during fasting altered an additional 304 genes. The largest changes occurred within 1 hr of refeeding, affecting over 10,000 genes (FDR < 5%) (Fig. 1D). Gene expression continued evolving throughout refeeding, with 3,261 genes changing between 1 and 12 hrs and 2,917 between 12 hrs and 1 week. Gene expression did not fully return to baseline, with thousands of differences persisting for at least a month post-fast, affecting key pathways (fig. S2A-C). Thus, extended fasting induces hypothalamic changes, whereas refeeding drives transcriptional shifts over weeks (table S1).

Gene expression analysis revealed biological processes, pathways, and disease mechanisms activated or suppressed at each FFR stage (fig. S3A-C, table S2). Fasting and hibernation influence neurodegeneration (8, 26), and we found that neurodegeneration pathways were suppressed during fasting but reactivated during refeeding, surpassing baseline values (fig. S3C). Cold exposure (18°C) during fasting activated immune response pathways, including natural killer cell-mediated cytotoxicity (fig. S3C). The longevity-regulating pathway increased between 1 and 12 hours of refeeding (fig. S3C), aligning with evidence that fasting and hibernation affect longevity (13, 27, 28). Thus, each FFR phase uniquely influences gene expression (table S2).

One hour of refeeding after a 72-hour fast activated ATP-dependent chromatin remodeling genes (Fig. 1E, F; table S2). To test for changes, we used ATAC-Seq to profile chromatin accessibility in the hypothalamus across four conditions: (1) fed baseline, (2) 72-hr fasted, (3) 12-hr refed, and (4) 1-week refed. We identified distal genomic regions with substantial chromatin accessibility changes across FFR conditions (table S3). Our data show that 42,919 sites (25% of peaks) were specific to the refed period (Fig. 1H). These findings demonstrate how fasting and refeeding reshapes hypothalamic gene expression and chromatin, highlighting refeeding as a critical window for changes.

Convergent Evolution in Hibernators Targets the Hypothalamic FFR Genetic Program

We next asked whether independently evolved hibernators share genetic changes affecting the hypothalamic FFR response observed in mice. To test this, we analyzed recent Zoonomia data (29), comparing accelerated evolution in hibernators and homeotherms (29). Examining ~1.3 million DNA sequences conserved for ~100 million years in a set of background homeotherms (conserved regions), we identified accelerated regions (ARs) in hibernators, including the Madagascar hedgehog (Echinops telfairi), little brown bat (Myotis lucifugus), thirteen-lined ground squirrel (Ictidomys tridecemlineatus), and fat-tailed dwarf lemur (Cheirogaleus medius) (Fig. 2B, C). We also defined ARs and conserved region deletions in homeotherm controls (Fig. 2B, C).

Fig. 2. Convergent genomic changes in hibernators affect gene regulatory mechanisms governing hypothalamic FFR responses.

(A) Schematic summary of hibernator-homeotherm comparative genomics analysis. (B) Phylogenetic tree from the 241-way mammalian alignment (241-mammalian-2020v2b.maf), showing obligate hibernators (blue), homeotherm controls (red), and background homeothermic species (black) used to define conserved regions (phastCons, FDR < 5%). Conserved regions were tested for accelerated evolution (ARs, phyloP, LRT; FDR < 5%) or deletions in hibernators and controls. (C) Images of each hibernator (blue) and control homeothermic species (red), with the number of significant accelerated regions (ARs, FDR 5%) and deletions (DELs) identified from ~1.3 million conserved regions shown below. Images and data correspond to species in the adjacent phylogenetic tree (A). (D and E) Bar plots showing the number of ARs (C) or deleted regions (D) shared by at least 2 of 4 hibernators (pHibARs, blue) or homeotherm controls (pHomeoARs, red). (F and G) H3K27ac+ PLAC-Seq identifies significant promoter regulatory contacts genome-wide in the mouse hypothalamus (FDR < 1%, total read count ≥ 12, observed/expected > 2; n = 10). (F) Heatmap showing the number of significant contacts on chromosome 18. (G) Contact plot for Spats2l (a significant FFR response DEG) illustrates how PLAC-Seq links pHibARs (light blue) and pHibDELs (dark blue) to contacted gene promoters. Contact loop height reflects the observed/expected read count. Tracks also include significant H3K27ac+ ChIP-Seq peaks (FDR < 5%, orange), FFR cycle ATAC-Seq peaks (FDR < 5%, black), gene models, and highlighted promoters (gray bars). (H) Pie charts of hypothalamus PLAC-Seq data show the proportions of pHibARs (light blue) and pHibDELs (dark blue) overlapping 10 kb windows with promoter-promoter (P-P, darker), enhancer-promoter (E-P, medium), or no promoter (No P, light) contacts. (I) Bar plot showing the odds ratio of pHibARs (blue) and pHomeoARs (red) in 10 kb regions contacting promoters of hypothalamus FFR response differentially expressed genes (DEGs) compared to background conserved regions. pHibARs show significant enrichment over pHomeoARs (Woolf test). ****p < 0.0001, **p < 0.01, *p < 0.05. (J) pHibARs and pHomeoARs were tested for enrichment in open chromatin sites (significant ATAC-Seq peaks, FDR < 5%, n = 8) relative to background conserved regions. pHibARs show significant enrichment compared to pHomeoARs (Woolf test). ****p < 0.0001; ns, not significant.

To enrich for hibernation-relevant elements, we identified convergent genomic changes by testing for parallel hibernator accelerated regions (pHibARs) and deletions (pHibDELs) impacting the same conserved regions in multiple hibernators. Peak numbers of pHibARs were found in ~40 bp conserved regions (fig. S4A), whereas pHibDELs peaked in ~30 bp regions (fig. S4B). Examples are shown in fig. S4C, D. Hibernators have more pHibARs and pHibDELs than expected by chance (fig. S4E, F). We identified 9,374 pHibARs and 25,108 pHibDELs across the mammalian genome (Fig. 2D, E; tables S4, S5), defining convergent hibernator changes to cis-elements conserved in homeotherms.

Distal CREs regulate gene expression through long-range 3D interactions. To identify genes contacted by pHibARs or pHibDELs in the hypothalamus, we used H3K27ac targeted proximity-assisted ligation ChIP (PLAC)-Seq (30, 31) (Fig. 2F, G). This generated a significant contact map, allowing us to assign pHibARs, pHibDELs, and ATAC-Seq peaks to gene promoters (Fig. 2F, G; tables S6, S7). In the hypothalamus, 14% of pHibARs formed enhancer-promoter (E-P) contacts, 58% formed enhancer-enhancer (E-E) contacts, and 28% were in conserved regions without significant regulatory contacts (Fig. 2H). Similar results were observed for pHibDELs (Fig. 2H). Hypothalamic regulatory contacts remained stable under fasting conditions, showing no metabolism-induced changes (fig. S5).

Using this integrated dataset, we examined interactions between pHibARs, pHibDELs, and metabolic gene expression responses. pHibARs strongly and disproportionately interact with FFR response gene promoters compared to conserved regions, particularly genes differentially expressed during 24–72 hr fasting and refeeding (fig. S6A). Additionally, pHibARs preferentially form regulatory contacts with fasting- and refeeding-affected genes, whereas parallel ARs from homeothermic lineages (pHomeoARs) are depleted for these interactions (Fig. 2I). These differences persist from the 72-hr fasting stage through 1-week refeeding (Fig. 2I). We analyzed ATAC-Seq peaks across fed, fasting, and refed states and found pHibARs were enriched in accessible hypothalamic CREs relative to conserved regions (Fig. S6B) and pHomeoARs (Fig. 2J). pHibDELs were not enriched for contacts with FFR genes but were overrepresented in ATAC-Seq peaks (Fig. S6C-F). Overall, convergent evolution in hibernators targets cis-regulatory mechanisms controlling hypothalamic FFR responses.

pHibAR Cis-Elements Target Central Hub Genes Controlling Metabolic Response Gene Modules

Hub genes act as central regulators in gene co-expression networks (32). We tested whether hibernator genomic changes preferentially targeted hub genes in hypothalamic FFR response networks (Fig. 3A). Using weighted gene correlation network analysis on RNA-Seq data from seven FFR stages (Fig. 1D), we identified 41 gene co-expression modules (fig. S7A), each involving distinct patterns and pathways across FFR phases (fig. S7B, S8, S9). Hibernator changes disproportionately impacted five modules enriched for genes with pHibAR regulatory contacts (fig. S7C). These modules regulate neuron differentiation, neurotransmitter and opioid signaling, inflammatory responses, lysosome biogenesis, fatty acid metabolism, circadian rhythm, and dopaminergic function (fig. S8A-B, S9B). Thus, convergent evolution in hibernators selectively impacts key FFR gene co-expression modules.

Fig. 3. pHibARs regulate hub genes driving FFR response gene co-expression modules and ARs indicate CRE loss-of-function.

(A) Schematic summary of FFR response hub gene discovery and hibernator-homeotherm comparative analysis. (B) Bar plot showing the number of hub genes for each of the 41 FFR response gene co-expression modules in the mouse hypothalamus. (C) Bar plot showing the -log10(p-value) for pHibAR and pHomeoAR enrichment at FFR response hub genes compared to background conserved regions (LOLA, R). Red line indicates p = 0.05. (D) The Venn diagram shows the overlap between FFR response hub genes and genes regulated by pHibARs (Hypergeometric test, p = 2.7 × 10⁻⁵). (E) Comparison of the number of FFR response co-expression modules with and without Hibernation-Hub genes. (F) H3K27ac+ PLAC-Seq promoter contacts for the blue2 FFR co-expression module Hibernation-Hub gene, En1. Red contacts overlap pHibARs (light orange highlight) and blue contacts do not. The kWithin measures En1 intramodular connectivity. pHibARs are in orange, with gene models below (En1 highlighted in blue). (G) Bar plot showing significantly enriched Gene Ontology molecular function terms for hibernation-hub genes compared to all genes expressed in the adult mouse hypothalamus (FDR < 5%, ClusterProfiler R; bar colors correspond to BH-adjusted p-values). (H) Schematic summarizing potential effects of ARs on H3K27ac+ active CREs in the 13-lined ground squirrel compared to mice: (1) gain of CRE activity (new H3K27ac+ peaks in squirrels), (2) neutral effects (squirrel-mouse shared H3K27ac+ peaks with sequence divergence), or (3) loss of CRE activity (H3K27ac+ peaks present in mice but absent in squirrels). (I) Bar plots showing the odds ratio for squirrel ARs and mouse ARs overlapping H3K27ac+ peaks: squirrel-specific (blue), mouse-squirrel shared (purple), and mouse-specific (red), compared to conserved regions. Woolf test. ***p < 0.0001. (J) Plots showing H3K27ac+ ChIP-Seq results for the 13-lined ground squirrel (blue) and mouse (red) hypothalamus (normalized read counts for two replicates). Significant peaks in one or both species are highlighted.

We identified hub genes within each co-expression module by ranking the top 10% of genes by connection weight and selecting those with module membership >0.8 (33) (fig. 3B, table S8). Reactome pathway analysis of FFR response hub genes revealed 73 significant pathways, highlighting roles in core metabolic processes (table S8). Hub genes were significantly enriched for regulatory contacts involving pHibARs compared to conserved regions, whereas control pHomeoARs showed no such enrichment (Fig. 3C). We found 38% of FFR response hub genes were contacted by pHibARs, defining them as "Hibernation-Hub" genes (table S8). Nearly all FFR response modules (37/41) contained at least one Hibernation-Hub gene (Fig. 3E). One example Hibernation-Hub gene is En1, a top hub gene in the blue2 module, which is enriched for dopamine neuron development pathways (Fig. 3E, fig. S7B). PLAC-Seq data revealed that En1 forms two regulatory contacts with a distal pHibAR (Fig. 3F). Thus, convergent evolution in hibernators selectively impacts cis-elements regulating FFR response hub genes.

Gene ontology analysis revealed that Hibernation-Hub genes are enriched for regulatory functions, including DNA-binding transcription factor activity, nuclear receptor binding, transcription coactivator/corepressor activity, and nuclear glucocorticoid receptor binding (Fig. 3G). Glucocorticoid signaling and the hypothalamic-pituitary-adrenal axis, which regulates adrenal glucocorticoid release, are central to fasting responses (34). These findings point to Hibernation-Hub genes as regulators of FFR gene expression responses.

ARs are Associated with CRE Loss of Function Effects

Accelerated evolution can indicate selection for lineage-specific traits or result from GC-biased gene conversion. After controlling for GC content, ~76% of pHibARs showed selection effects consistent with either positive selection (gain of function) or relaxed purifying selection (loss of function) in hibernators (table S9). To distinguish these effects, we analyzed how ARs correspond to changes in hypothalamic cis-regulatory elements in the 13-lined ground squirrel (hibernator) versus mice (homeotherm). We performed H3K27ac ChIP-Seq in squirrel and mouse hypothalami to identify active CREs and assess whether squirrel ARs affect (i) mouse-CREs (loss of function), (ii) squirrel-CREs (gain of function), or (iii) shared CREs (Fig. 3H). Comparative profiling revealed species differences in CRE activity (fig. S10A,B). Whereas27% of significant H3K27ac+ peaks overlapped between species, 73% were species-specific. Squirrel ARs were enriched in mouse-CREs but depleted in squirrel and shared CREs, supporting a loss-of-function effect in squirrel (Fig. 3I,J). We conducted the same analysis in mice. Similar to squirrel, mouse ARs were depleted in mouse-CREs and shared CREs but enriched in squirrel-CREs, consistent with mouse ARs contributing to CRE loss-of-function in mice (Fig. 3I). Deleted regions, known loss-of-function events, showed the same pattern as ARs—squirrel deletions were depleted in squirrel-CREs and shared CREs but enriched in mouse-CREs, with the reciprocal result for mouse deletions (fig. S10C). Thus, ARs and deletions exhibit parallel loss-of-function effects.

To further investigate AR-associated loss-of-function effects in hibernators, we analyzed transcription factor binding sites (TFBSs) within pHibARs, comparing hibernators and closely related homeotherm sequences (fig. S10D,E). Homeotherm sequences contained more TFBS motifs than hibernator sequences, indicating that ARs in hibernators predominantly eliminate rather than create TFBSs (fig. S10D,E). This pattern was consistent across all four hibernator-homeotherm pairs. Human sequences corresponding to pHibAR and pHibDEL sites were enriched for 35 and 146 regulatory protein motifs, respectively (fig. S11, table S10), highlighting their potential roles in the regulatory divergence between hibernators and homeotherms.

A key example was the enrichment of HIF1A (Hypoxia-Inducible Factor 1 Alpha) binding motifs, with a 2.6-fold increase in pHibARs and 34-fold in pHibDELs (fig. S11). HIF1A regulates metabolic shifts under low oxygen, and its role in hibernation is linked to apneic breathing and hypothermia during torpor entry (35, 36). These findings show evolutionary changes in Hif1a signaling and its associated cis-elements. Although some ARs may confer gain-of-function, our results show that hibernator ARs predominantly cause selective CRE and TFBS loss-of-function. These changes shape the genomic specialization of hibernators.

Hibernation-Linked Gene Regulatory Programs Converge on Specific Hypothalamic Cells

We examined whether convergent genomic changes in hibernators disproportionately affect specific hypothalamic cell types under different FFR conditions (Fig. 4A). We collected snRNA-seq and snATAC-seq data from 89,036 hypothalamic cells in fed, 72-hr fasted, and 12-hr refed adult mice. From snRNA-seq, we identified eight major cell types (Fig. 4B) and 42 cell clusters (Fig. 4C). Seven clusters increased during refeeding, including GABAergic, glutamatergic, and other neuronal cells, along with oligodendrocyte precursor cells (OPCs) and microglia (Fig. 4D). Marker genes define each cluster (Table S11). These findings reveal metabolic state-dependent shifts in cell populations.

Fig. 4. Single-nucleus multi-omics links FFR responses and Hibernation-Hub genes to distinct hypothalamic cell types.

(A) Schematic summary of cellular multi-omics analysis of mouse hypothalamus FFR responses and the identification of hibernator-homeotherm genomic divergence in cellular FFR response regulatory mechanisms. (B and C) UMAP plots showing cell type (A) and cluster (B) classifications (Seurat and Harmony) from hypothalamus snRNA-Seq data for fed (n = 3), 72-hr fasted (n = 3), and 12-hr refed (n = 3) adult female mice, identifying 8 major cell types (A) and 42 clusters (B) from 89,063 cells. (D) The bar plot shows mean ± SEM cell counts per cluster across Fed, 72-hr fasted, and 12-hr refed conditions (n = 3), normalized by total cells per replicate. P-value is shown for cell cluster by FFR condition interaction effect with post-tests (*p < 0.05, **p < 0.01). (E) The heatmap shows the -log10 q-values for FFR co-expression module Hibernation-Hub gene (rows) enrichment among marker genes of cell clusters (columns), compared to all hypothalamic genes. Only significant modules are displayed. Red text highlighted cell clusters are significantly affected by FFR condition (see D). Cell types and clusters are color-coded by the legend. (F) UMAP heatmaps display snRNA-Seq expression for two example darkolivegreen module Hibernation-Hub genes. (G) Violin plots show snRNA-Seq expression level of darkolivegreen4 Hibernation-Hub genes across all cell clusters. Red text highlighted cell clusters are significantly affected by FFR condition (see D).

Hibernation-Hub genes from seven FFR response co-expression modules were disproportionately in specific cell clusters (Fig. 4E, table S12). We found that 92% of these clusters were neuronal, including nine GABAergic, 12 glutamatergic, and four other neuron clusters, alongside two OPC and one microglia cluster. Hibernation-Hub genes from the darkred module were enriched in 22 clusters, suggesting a broad role in hypothalamic neuron function, whereas darkolivegreen4 module Hibernation-Hub genes were enriched in five neuronal clusters, three of which increased during refeeding (Fig. 4E). Hibernation-Hub gene expression is elevated in these clusters (Fig. 4F, G). Given their fasting-suppressed and refeeding-induced expression (fig. S7B), the darkred and darkolivegreen4 modules likely mediate specific neuronal responses to these metabolic shifts. Thus, specific hypothalamic cell populations express Hibernation-Hub genes from specific FFR response modules, underscoring specialized roles in metabolic adaptation.

To investigate Hibernation-Hub gene roles in cell type-specific FFR responses, we analyzed differential expression across three contrasts—Fed vs. Fasted, Fasted vs. Refed, and Fed vs. Refed—in eight major cell types (table S13). Fasting activated genes in GABAergic and glutamatergic neurons, whereas non-neuronal cells showed the greatest activation during refeeding (fig. S12A,B). Despite cell-type differences, Hibernation-Hub genes were disproportionately activated during refeeding across most cell types (fig. S12C). These results suggest distinct cellular responses to FFR phases, with Hibernation-Hub genes mediating refeeding adaptations that likely diverged in hibernators to support recovery after prolonged torpor.

Cellular and Metabolic Chromatin Accessibility Dynamics in the Hypothalamus

To refine our understanding of pHibAR-enriched chromatin dynamics, we used snATAC-Seq to resolve cell-type-specific changes across fed, 72-hour fasted, and 12-hour refed states. We observed strong effects of FFR state and cell type on chromatin accessibility, with refeeding driving the greatest increase in open chromatin, particularly in astrocytes, GABAergic, glutamatergic, and other neurons (Fig. 5A, B). Across all eight major cell types, we identified 91,479 differentially accessible sites, with refeeding inducing the most pronounced changes in neurons and microglia (fig. S13A, B). Many distal peaks, especially from refed conditions, likely represent CREs not annotated by ENCODE (37) (fig. S14), highlighting a dynamic regulatory landscape shaped by both cellular identity and metabolic state.

Fig. 5. FFR-driven chromatin changes reveal hibernation-linked genetic programs across hypothalamic cell types.

(A) The bar plot shows the number of significant ATAC-Seq peaks for each hypothalamic cell type and FFR condition (Genrich, FDR < 5%, n = 3). P-values are from regression tests of FFR condition and cell-type effects on peak counts. Cell types: Endothelial (Endo), oligodendrocyte precursor cells (OPC), oligodendrocytes (Oligo), microglia (Micro), astrocytes (Astro), GABAergic neurons (GABA), glutamatergic neurons (GLUT), and unclassified neurons (Neuron). (B) Mosaic plot showing the relationship between ATAC-Seq peak numbers, cell type and FFR condition (Chi-Square p-value shown). Pearson residuals indicate relative enrichment (blue) or depletion (red) for each category. (C) Bar plots show the odds ratio for pHibARs and pHomeoARs in significant ATAC-Seq peaks across hypothalamic cell types during Fed, 72-hr fasted, and 12-hr refed conditions relative to conserved regions. LOLA R package enrichment test, ***p < 0.001, **p < 0.01. See also Fig. S15. (D) Bar plot showing the number of active CREs (significant ATAC-Seq peaks) overlapping pHibARs and/or pHibDELs in each cell type, highlighting hibernation-linked CREs. (E) UMAP of snRNA-Seq showing Hspa8 expression across cells and FFR conditions, with decreased expression in the 72-hr fasted condition. (F) Violin plots showing Hspa8 expression across cell types and FFR conditions. (G) Cell type-specific hibernation-linked regulatory programs for Hspa8 in the hypothalamus. H3K27ac+ PLAC-Seq data shows Hspa8 promoter contacts (10 kb cis-bin windows), with tracks for pHibARs (red), pHibDELs (orange), and significant snATAC-Seq peaks (blue). Peaks overlapping pHibARs are highlighted (red lines). Green tracks show snATAC-Seq peaks with significant differential accessibility between Fed, 72-hr fasted, and/or 12-hr refed conditions (q < 0.1), with PLAC-Seq bins overlapping differentially accessible regions (DARs) highlighted in green. Promoter contacts are color-coded: FFR DARs (green), pHibARs in ATAC-Seq peaks (red), pHibDELs in ATAC-Seq peaks (orange), and other ATAC-Seq peaks (blue). Gene models below, with Hspa8 promoter-associated distal contact sites highlighted in purple.

We examined how convergent genomic changes in hibernators impact accessible CREs across cell types and FFR states. pHibARs and pHibDELs were enriched in ATAC-Seq peaks across all eight major hypothalamic cell types and conditions, whereas pHomeoARs and pHomeoDELs were depleted (Fig. 5C, S15). Thus, hibernator genomic changes disproportionately impact hypothalamic CREs. We defined hibernation-linked CREs—ATAC peaks overlapping pHibARs or pHibDELs—in each cell type and FFR state (Fig. 5D, table S14), identifying conserved elements likely critical for hypothalamic regulation and evolutionary divergence (Data S1-3).

Cellular Genetic Programs for Hypothalamic Metabolic Responses and Hibernator Evolution

To uncover cellular regulatory programs underlying FFR responses and their modification in hibernators, we examined Hibernation-Hub genes with FFR-dependent expression changes across cell types (fig. S12C). Many genes showed state-specific expression patterns; for example, Hspa8—a heat-shock protein implicated in neuroprotection and protein aggregate clearance (38-44) —was downregulated during fasting and upregulated during refeeding in most cell types (Fig. 5E, F). Given the enhanced neuroprotective adaptations in hibernators (8), we performed a detailed analysis of Hspa8’s cell-type specific regulatory program.

Integration of PLAC-Seq, snATAC-Seq, and hibernator genomic data revealed cell type-specific regulatory programs controlling Hspa8 expression linked to hibernator evolution. Across cell types, we identified promoter contacts, open chromatin CREs, FFR-responsive CREs, and those impacted by pHibARs or pHibDELs (Fig. 5G). Hibernation-linked CREs were present in all cell types except microglia, and FFR-dependent CRE accessibility varied by cell type. These integrative analyses define cellular cis-regulatory programs for FFR responses across the genome, including CREs and TFBS motifs (Data S1-4), outlining a genetic circuitry for metabolic regulation in the hypothalamus.

Discussion

We identified hypothalamic cis-regulatory programs that underlie mammalian responses to food scarcity and revealed how genomic changes in hibernators reshape these programs. By integrating multi-omic profiling of FFR states in mice with comparative genomics across mammals, we uncovered conserved CREs showing convergent evolutionary changes in hibernators. These elements disproportionately contact hub genes central to FFR responses, revealing a gene regulatory framework for metabolic and hibernation control. Our companion study confirms that individual pHibAR-impacted CREs influence metabolism and behavior in vivo (24), offering a platform to explore how metabolic setpoints, obesity, neurodegeneration, and aging may be genetically or epigenetically reprogrammed.

Hibernators exhibit seasonal shifts in metabolism and physiology, gaining fat prior to torpor and relying on insulin resistance and lipid oxidation during winter torpor (1, 45). We show that ARs and deletions in hibernators are associated with loss-of-function at CREs conserved and active in homeotherms. This suggests that pHibARs and pHibDELs identify CREs essential for homeothermic regulation but dispensable—or even restrictive—for hibernation traits, such as extreme heterothermy and torpor. In mice, many of these CREs regulate genes activated during refeeding—a phase marked by chromatin remodeling and distinct gene expression changes in astrocytes, endothelial cells, oligodendrocytes, and neurons. Refeeding, though underexplored, is a critical window for recovery from torpor-related stresses including immobility (9, 46, 47), nutrient deprivation, insulin resistance (10, 48), accumulated wastes and toxins (49), and tauopathy (8).

Hibernators also evolved unique capabilities for longevity, neuroprotection, synapse regeneration, and cell damage prevention (5, 7, 8, 13, 50, 51). We found that Hibernation-Hub genes—those affected by convergent evolution and central to FFR gene networks—are enriched in neurodegeneration-related pathways. These include mitochondrial stress responses, protein folding, ubiquitination, and catabolism, implicating them in both fasting biology and neuroprotective processes (52). The identified CREs, hub genes, and cellular programs offer a blueprint for modulating mammalian metabolism, health, and hibernation evolution.