Metabolomics-Guided Genomic Comparisons Reveal Convergent Evolution of Hibernation Genes in Mammals

Abstract

Hibernation exists in several unrelated mammalian lineages, allowing animals to survive extreme 0environmental conditions through profound physiological shifts, including reduced metabolic rate, heart rate, respiration, and body temperature. These physiological shifts allow hibernators to rely solely on fat reserves, simultaneously avoiding the adverse effects of prolonged immobility seen in nonhibernating species. Although research on individual species has highlighted key aspects of these adaptations, the genetic basis of hibernation across mammals remains poorly understood. Synthesizing both single species and comparative approaches, we use metabolomic data from waking and hibernating black bears (Ursus americanus) to guide bioinformatic analyses of genes using tests of selection and evolutionary rate convergence across independent lineages of hibernating mammals. Our analyses reveal significant changes in carnitine levels between states. Using public databases, we generate candidate genes which may contribute to regulation of carnitine, and use these to test for signatures of selection across several independent lineages of hibernating mammals. We also utilize a dataset of 19k proteins across 120 mammalian genomes to identify genes evolving at convergent rates across hibernating mammals. Using both approaches, we find several novel genes likely to impact carnitine metabolism and related functions vital to hibernation such as metabolic shifts, oxidative stress, and tissue preservation. These findings provide new insights into the genetic basis of hibernation and offer promising targets for translational research, including the development of clinical therapies that mimic hibernation-like states for applications in medicine and space exploration.

Graphical Abstract

Graphical Abstract.

Introduction

Hibernation is a complex physiological adaptation that is observed across at least 11 mammalian lineages, enabling survival during prolonged periods of cold temperatures and limited food availability (Geiser and Ruf 1995; Ruf and Geiser 2015). During hibernation, animals undergo dramatic reductions in metabolic rate, body temperature, and heart rate, which return to normal during the active season. These physiological shifts allow hibernators to endure extended periods of immobility without food or water, while somehow mitigating the risks typically associated with inactivity (Geiser and Ruf 1995). Characterizing the physiological changes that occur during hibernation has been of central interest not only to evolutionary biology but also to various applied medical fields (traumatic injury, long sedation on ventilators, coagulopathy, osteoporosis, muscular atrophy, and diabetes), and even space travel (Zhou et al. 2001; Wu and Storey 2016; Perez De Lara Rodriguez et al. 2017; Wolf et al. 2018; Andrews 2019; Chazarin et al. 2019; Bertile et al. 2021; Cerri et al. 2021).

Hibernating mammals suppress mitochondrial respiration while upregulating lipid oxidation and downregulating glycolysis, allowing for the efficient use of fatty acids and ketone bodies when glucose is scarce (Jani et al. 2013; Horii et al. 2018; De Vrij et al. 2023). This metabolic reprogramming enables hibernators to use alternative fuels such as fatty acids and ketone bodies when glucose availability is low (Horii et al. 2018). Additionally, mammals have a complex regulation of hormones that influence their metabolism during hibernation. For example, iodothyronamine, leptin, and ghrelin may play a role in modulating thermogenesis, energy expenditure, and appetite (Wu and Storey 2016). Studies of metabolic shifts in hibernators have provided valuable insights into proteomic and genomic mechanisms with broad applicability, such as the identification of heat shock protein 47 (HSP47) as a critical regulator of thrombosis (Thienel et al. 2023). Hibernation research has also revealed critical insights into mammalian physiology, offering a deeper understanding of metabolic suppression, resistance to tissue damage, and muscle preservation during prolonged inactivity (Carey et al. 2003). These discoveries have led to important medical applications, including therapeutic hypothermia for ischemic injury management and strategies to prevent muscle atrophy in immobilized patients (Storey 2003; Storey 2010; Liu et al. 2014; Al-Attar and Storey 2020; Ebert et al. 2020). Besides therapeutic hypothermia, hibernation-based therapies have become an ever-expanding field as research has shown the efficacy of chemically inducing the same or similar pathways and physiologies seen during hibernation (Wolf et al. 2018; Matsuo et al. 2021; Sun et al. 2024).

Despite advances in understanding the metabolic mechanisms underlying hibernation, the genetic basis of these adaptations—and the extent to which they are convergent across species—remains unclear. Hibernation occurs in many distantly related mammalian lineages, suggesting that it may have evolved multiple times through convergent evolution (Geiser 1998). However, this interpretation remains debated, as some researchers argue that torpor and hibernation may represent retained ancestral traits, particularly given the thermolability observed in basal mammals like the echidna (Grigg et al. 2004; Ruf and Geiser 2015; Lovegrove 2017; Nowack and Turbill 2022). Recent work on the evolution of endothermy in ray-finned fishes further supports the plausibility of repeated origins for complex thermoregulatory traits: phylogenomic and comparative genomic analyses demonstrate at least four independent origins of endothermy in fishes, shaped not by shared ancestry or paleoclimate, but by ecological interactions and selective pressures such as competition with cetaceans (Melendez-Vazquez et al. 2025). Even within bats, recent molecular phylogenetic analyses suggest that hibernation likely evolved independently at least four times from a nonhibernating bat ancestor, rather than being inherited from a common ancestor (Lazzeroni et al. 2018). Recent molecular, phylogenetic, and comparative studies increasingly support the view that hibernation arose repeatedly and convergently in mammals, challenging earlier parsimony-based ancestral-retention hypotheses, we therefore operate under this assumption for the purposed of this paper (Geiser 1994; Andrews 2007; Turbill et al. 2011; Lovegrove et al. 2014; Van Breukelen and Martin 2015; Staples 2016; Heldstab et al. 2018; Lazzeroni et al. 2018; Boyles et al. 2020; Nespolo et al. 2021). However, regardless of whether hibernation in mammals reflects an ancestral trait or a derived adaptation, independently evolved hibernating species that have closely related, nonhibernating counterparts offer a powerful comparative framework. In other words, whether hibernation represents an ancestral capacity modified in some lineages or a repeatedly derived adaptation, the deeply split, independently evolving lineages studied here provide a valid comparative framework for investigating selection, regardless of ancestral state assumptions. These contrasts can be leveraged to detect evolutionary signals of divergent selective regimes and identify the genetic basis of complex physiological traits such as metabolic suppression, thermal regulation, and energy conservation. Although the hibernation phenotype varies considerably across species, including differences in the use of interbout arousals, the extent of metabolic suppression, maintenance of circadian rhythms, and reliance on food stores, this diversity strengthens rather than weakens our comparative framework. While species-specific mechanisms may be missed in broad analyses, identifying consistent signals across independently evolved and physiologically diverse hibernators highlights conserved molecular components likely to be central to the hibernation phenotype. These shared features represent promising targets for understanding the fundamental biology of hibernation and for translational applications as repeated evolutionary targets may also represent reduced risk of adverse pleiotropic effects. This significant gap in knowledge limits our ability to manipulate these key physiological processes for potential therapeutic applications. Understanding mechanisms shared by all hibernating mammals not only sheds light on the repeatability of complex adaptations but also highlights evolutionary targets that are more likely to be conserved across all mammals, making them promising candidates for human medical applications.

To capture the metabolic shifts that are crucial for facilitating and terminating hibernation we focus on comparative analyses of blood plasma between American black bears (Ursus americanus) that are fully hibernating in their early-denning phase (hereafter “hibernating”) and bears that are late denning in the transition to waking (hereafter “waking”). We find that a key metabolite (carnitine) shifts during the transition from hibernating to waking periods. To identify genes underlying shifts in this metabolite we use bioinformatic resources to create a dataset of 374 candidate genes in carnitine metabolism pathways for hundreds of mammals, including several convergent lineages of hibernating species. Using this dataset we employ several nested tests of convergent selection (Fig. 1). We also integrate a physiology-agnostic dataset of 19 thousand genes (protein sequences) from 120 mammalian genomes, to identify genes evolving at convergent evolutionary rates in hibernators (Fig. 2). Together, our data reveals 38 of 374 genes evolving under convergent selection in hibernators, and 33 of 19 thousand genes evolving at convergent evolutionary rates across hibernating mammals (two of which were also under selection) (Fig. 2). This studly highlights the utility of using a metabolomics-guided comparative genomics approach to understanding the genetic basis of complex adaptation. Metabolomics-guided tests of selection provided specific insights into convergent genetic pathways involved in an empirically verified metabolic shift and resulted in many more genes with strong evolutionary signatures of convergent selection. Whereas our top-down physiologically agnostic dataset yielded fewer strongly supported insights, it revealed convergent genetic mechanisms for a larger diversity of functions related to hibernation biology and yielded insight on physiologies that warrant future investigation (Fig. 2). Integrating both top-down and bottom-up comparative genomics, our data elucidates new genetic targets of selection in hibernating mammals, as well as confirm the importance of genes identified in prior studies of expression, proteomics, and predictive genomics. This work highlights the value of integrating both top-down and bottom-up methods and contributes vital insights into the genetic basis of physiologies that facilitate hibernation, providing promising avenues to translational medicine aiming to both understand and manipulate the same pathways and physiological processes.

Fig. 1.

Rooted topology showing phylogenetic relationships of the mammals used for genomic comparative analyses. Lineages known to hibernate are indicated in red (appears as a lighter shade text for those with color-vision differences).

Fig. 2.

Infographic workflow of the two datasets used in this study. Bottom-up analysis was informed by metabolic shifts observed between waking and hibernating black bears. Rather than analyzing raw metabolite abundances individually, we used predefined metabolic indicators (MIs)—combinations of metabolites such as ratios or sums—that capture coordinated biochemical changes and provide more interpretable, biologically meaningful signals. Multivariate analysis (sPLS-DA) identified latent components (MIs) that best distinguished physiological states, and individual metabolites contributing most strongly to those components (loadings) were examined to pinpoint candidate pathways. Of these, carnitine was the only loading component showing significant differential abundance, underscoring its importance in mediating metabolic transitions associated with hibernation. Genes for comparative analysis were identified as part of the carnitine pathway in KEGG and present in the NCBI ortholog database, yielding 371 genes suitable for multispecies comparisons. Top-down analysis utilized 19,600 protein-coding genes from 120 mammalian genomes to generate gene trees for convergent rate analysis, identifying 33 genes evolving significantly faster or slower across all hibernating species. Both gene sets (371 + 33) were tested for positive and relaxed selection using the HyPhy software suite, incorporating several nested models to account for potential sources of type I and type III errors (see Materials and Methods).

Materials and Methods

Bear Sample Acquisition

Free-ranging black bears are studied by the Minnesota Department of Natural Resources (MNDNR) to better understand the population dynamics and track the health and biometrics of the population. For this study, we collaborated with the MNDNR to collect blood samples to research black bear adaptations and physiology. A total of 75 venous blood samples were collected from 50 unique bears were they were anesthetized during den visits at three seasonal time points: (i) the active time during the summer months (n = 12, June and July), (ii) Hibernating (early-denning), their most immobile time (hibernations, November, December, January, n = 35); and (iii) waking (late denning) as they are emerging from hibernation (February, March, n = 28). Though some bears were sampled in multiple time points, the dataset as a whole was unpaired. Bears were anesthetized using standard wildlife handling procedures (Kreeger and Arnemo 2016; Sikes 2016). Biometric data were gathered, including height, weight, and the presence of cubs/yearlings (supplementary table S6, Supplementary Material online) Whole blood was collected in citrated plasma tubes and spun for plasma and stored at −80°C.

Identifying Key Metabolic Indicators

All metabolites were identified using isotopically labeled internal standards and multiple reaction monitoring. Multiple reaction monitoring is a targeted mass spectrometry technique that selectively detects specific ion transitions for precise quantification. This method allows for the precise identification and quantification of metabolites under optimized conditions provided by Biocrates (Innsbruck, Austria), ensuring consistent and reliable results for metabolomic analysis. Isotopically labeled internal standards are added to ensure accurate quantification and account for sample handing and preparation differences. For quantification either a 7-point calibration curve or one point calibration was used depending on the metabolite class. Data were captured using the Analyst (Sciex) software suite and transferred to the MetID software (version Oxygen; Biocrates Life Sciences AG, Innsbruck, Austria) which was used for further data processing and analysis. Metabolites with less than 80% detection between all experimental samples were removed from our dataset, leaving 385 metabolites for our analysis. Any missing values were imputed using a K-nearest neighbor approach (k = 7) (Anderson et al. 2023). The entirety of the dataset was normalized using quantum normalization methods in the preprocessCore R package (Ben Bolstad 2017). The samples were categorized into three time periods: summer, hibernating, and waking. Data were subset by liquid chromatography methods and flow injection analysis for broad analysis of proteins.

Metabolic indicators are derived from formulas, such as ratios or sums of metabolite abundances, transforming raw metabolomic data into interpretable metrics that provide insights into biological processes. Rather than analyzing raw metabolite levels individually—which can be noisy, redundant, and hard to interpret—grouping metabolites into biologically meaningful indicators helps reduce data complexity, highlight coordinated metabolic shifts and improves statistical power. Raw metabolic abundance data was used to calculate 159 predefined metabolic indicators using the Biocrates MetaboINDICATOR™ software tool (Griffin 2020; Limonciel et al. 2021). To identify differences in metabolic indicators by denning period (linked to hibernation), we performed an sPLS-DA using the MixOmics R package (Rohart et al. 2017). To evaluate sPLS-DA model performance and feature stability, we performed a 3-fold cross-validation 10 times for each type of prediction distance (max distance, centroids distance, and Mahalanobis distance). Slight differences between the overall error rate and the balanced error rate (BER) (with BER being higher) were observed, indicating potential misclassification of minority classes. To assess the robustness of the model, we performed a feature stability analysis, leading to the selection of Components 1 and 2 for the final PLS-DA model. This analysis was repeated with only early and late denning clusters to verify the metabolic indicators which showed the largest differences between these two phases (Fig. 3a and b).

Fig. 3.

a) Sample plot generated using sPLS-DA with the mixOmics package in R. The samples are colored according to their hibernation status and ellipses represent a 95% confidence interval. Components 1 and 2 explain 13% and 9% of the variation, respectively. b) Circle plot visualizing the correlation between metabolic indicators and the latent variables (Component 1, Component 2) derived from sPLS-DA. Each point represents a metabolic indicator, with the angle indicating its contribution to the direction of cluster separation and the distance from the center representing the strength of its correlation with the latent variables. Metabolites further from the center have a stronger influence on cluster separation, guiding the clusters in the directions shown. c) Metabolic indicators between hibernating and waking identified to have the strongest correlation with latent variables. All metabolic indicators are unitless quantities derived from Biocrates MetaboINDICATOR™: MxP® Quant 500 formulas. d) Raw metabolite abundance for Carnitine and Acetylcarnitine, the metabolites with <80% valid quantitation that contribute to the three metabolic indicators found using sPLS-DA.

Identifying Key Metabolites

To identify metabolites underlying differences in metabolic indicators between hibernating and waking, we further scrutinized the metabolic indicators identified in sPLS-DA which appeared to have the starkest differences between these two states (Fig. 3b) (beta oxidation, ratio of acylcarnitine to carnitine, and carnitine esterification). Although summer data were available, the sample size was smaller for this group, and our primary research question centers on the physiological transition from hibernation to waking. While summer bears are also active, their metabolic profiles reflect additional processes such as feeding, mating, and elevated physical activity, which can obscure or confound the specific signals related to arousal from hibernation. Including summer samples in the sPLS-DA model effectively conflates distinct physiological states (summer and waking bears), reducing the model's ability to cleanly separate processes involved specifically in the transition out of hibernation. To avoid this conflation and maintain analytical clarity, we constrained the sPLS-DA model to compare only hibernating and waking bears. The summer data were instead used independently to test whether key metabolic shifts observed during arousal persist into the summer active period (Fig. 3d; supplementary fig. S1, Supplementary Material online).

sPLS-DA is a multivariate statistical method designed to identify features (in this case, metabolites) that drive group separation—hibernation versus waking. It was used because of its ability to handle high-dimensional data while selecting the most relevant variables for class distinction while minimizing overfitting. Significance was assessed using the model's variable importance in projection scores, with higher scores indicating greater contributions to the separation of groups. Additionally, model weights and loadings were examined to confirm the importance of specific metabolites.

We confirmed these indicators were significantly different between hibernating and waking periods using a Kruskal–Wallis test of significance followed by a Dunn's pairwise test with Bonferroni multiple-comparison correction (P < 0.05) (Fig. 2c). For all three metabolic indicators, we used the provided MetaboINDICATOR™: MxP® Quant 500 formulas to identify raw metabolites used to derive the indicator variables. We used the same method (Kruskal–Wallis test of significance followed by a Dunn's pairwise test with Bonferroni multiple-comparison correction) to find which metabolites were significantly different between hibernating and waking (P < 0.05). Of these, only acetylcarnitine (C2) and carnitine (C0) were above the 80% valid quantitation cut-off. When examining raw abundance only carnitine (C0) was identified as significantly different between hibernating and waking after multiple test correction (Fig. 3d; supplementary fig. S1, Supplementary Material online).

Genetic Data Curation

To identify a list of genes potentially underlying changes in carnitine levels during hibernation, genes related to pathways that involve carnitine were selected through the KEGG (Kanehisa and Goto 2000). Through KEGG, carnitine metabolism was found to be present in the following pathways: thermogenesis, fatty acid degradation, insulin resistance, diabetic cardiomyopathy, and bile acid secretion. Using the ortholog table available for each pathway on KEGG, we searched NCBI's ortholog database for multispecies ortholog sequence data. We identified 394 genes that were available for mammals, and the associated RefSeq transcripts and protein alignments were downloaded through NCBI's ortholog database.

Alignment and Tree Curation for Carnitine Pathway Genes

To perform accurate selection tests for 394 genes of interest, sequences were filtered and checked for naming convention discrepancies via a set of custom scripts, and if they did not have matching protein sequences, they were removed from the dataset. Protein alignments and Refseq transcript files were then run in the PERL script PAL2NALto generate protein-informed codon alignments (Suyama et al. 2006). Codon alignments were subject to further scrutiny and edited with a custom cleanup pipeline which: (i) Removes stop codons and all sequence downstream of stop codons, (ii) Begins the alignment with a codon that is present for at least 70% of species, and (iii) Removes sequences (species) that are >50% gaps. After filtering steps 378 gene alignments remained, and a phylogenetic tree was generated for all species included in those alignments (206 species) using TimeTree.org (Kumar et al. 2022). Species missing from TimeTree, but included in our sequence data, were added manually, resulting in a 206-species master topology (available on Dryad “206Species_MasterTree”). As each codon alignment requires a corresponding species tree to run selection analyses, a custom python script was used to generate a tree file for each codon alignment, pruned from the master topology using only the species in each alignment file. A full pipeline with all associated scripts is available on Github (https://github.com/drabe004/BearProject_Scripts).

Tests of Selection

To test all genes for convergent positive selection in hibernating species, sequence alignments and trees were used as input for the HyPhy suite test BUSTED (Branch-Site Unrestricted Statistical Test for Episodic Diversification), which identifies branches in a phylogenetic tree that have experienced episodic positive selection at specific sites across a gene. Briefly, BUSTED uses a codon-based model to compare the fit of a null model, where no sites are under positive selection on any branches, to an alternative model, where some sites may experience episodic positive selection (dN/dS > 1). First, to check for signal of positive selection in hibernating species only, a list of hibernating species (supplementary table S7, Supplementary Material online) was used to assign tree tips to a “foreground” with remaining tips assigned to “background.” A custom R script assigned internal branches to the Foreground, using simple maximum parsimony in the Castor package (Louca and Doebeli 2018; R Core Team 2023). Second, to check for signal of pervasive selection across the tree, a second round of tests was run designating no branches to the foreground (hereafter referred to as “hypothesis-free” test). Lastly, to verify that the main source of signal for selection was attributable to hibernating species, a drop test was performed, in which we removed all foreground species from the tree/alignments, and subsequently ran a hypothesis-free BUSTED test (Kowalczyk et al. 2021; Drabeck et al. 2024). To test for positive selection in hibernating bears only (from which our physiological signal was derived), we also ran tests of selection designating bears (genus Ursus) only as foreground taxa. All BUSTED tests were run using the “-srv Yes” flag to account for synonymous rate variation. Results of tests were assessed using a likelihood ratio test, and tests with bears and hibernators in the foreground were compared to hypothesis-free tests and drop tests to compile a list of genes under convergent positive selection which are robust to Type 1 error caused by selection in hibernators and nonhibernators.

To test for signals of relaxed selection, we used the RELAX test implemented in the HyPhy suite (Wertheim et al. 2015). Briefly, this method compares the fit of models in which both negatively selected sites and positively selected sites are trending toward an omega value of 1 (neutrality), to a more standard model of positive selection where omega is increasing across all sites. We use the same foreground and background tests as in BUSTED tests, with the exception that RELAX does not allow for a “hypothesis-free” option.

Because bats made up a large portion of the hibernating taxa, there was a risk that genes associated with other bat physiologies unrelated to hibernation would bias the dataset by driving signal for selection disproportionately. To account for this, we also reran BUSTED and RELAX (including drop tests) with only a single (hibernating) bat representative (Myotis myotis). A table outlining each nested test, including the dataset, foreground species, and interpretation is available in supplementary table S8, Supplementary Material online.

Alignment and Gene Tree Curation for RERconverge

Trees for 120 mammal species were generated from protein sequences downloaded from Hecker and Hiller (2020). When multiple isoforms of a protein were included, only the version with the greatest number of species was retained. All “*,” “?,” and “X” symbols were converted to gaps (“-”), species in which the gene was missing (i.e. all “-”) were removed, proteins with fewer than three species were removed, and protein sequences were aligned using the msa package in R (Bodenhofer et al. 2015). After aligning sequences, the RERconverge “estimatePhangornTreeAll” function (Kowalczyk et al. 2019) with default parameters was used to calculate branch lengths. Briefly, the function uses phangorn (Schliep 2011) functions to generate maximum likelihood branch lengths assuming a supplied topology (“RERConverge_120MasterTree” available on Zenodo) using the LG model with k = 4. This approach intentionally constrains all gene trees to the same species tree topology, minimizing the effects of gene tree–species tree discordance. The supplied topology was based on one provided with the alignments and adjusted to match the high-confidence tree originally reported in Meyer et al. (2018) based on Meredith et al. (2011) and Bininda-Emonds et al. (2007). After generating trees, the RERconverge readTrees function was used with default parameters to convert the trees to the proper format for further analyses and generate a master tree with branch lengths representing average evolutionary rate across many regions. In total, we produced 19,610 gene trees across 120 mammalian and used these as input for convergent rate analysis.

RERconverge

To identify genes with convergent evolutionary rates in hibernating mammals we ran RERconverge with 19,610 gene trees derived from 120 mammalian genomes that included 14 hibernating species (supplementary table S9, Supplementary Material online). Convergent rate analysis is a method of examining whole-genome data by transforming single-copy gene trees to trees where branches represent relative evolutionary rates of protein evolution, constrained by a known species topology. Foreground species which have a common, convergently evolved phenotype (e.g. hibernation) are noted and RERconverge identifies genes which are evolving with slower or faster relative evolutionary rates when compared to background species (e.g. nonhibernators). For these data, we used the 14 hibernating species as foreground taxa. All analyses were run using the “permulations” (permutated simulations) option, which evaluates the significance of evolutionary rate shifts by permuting trait values across the phylogenetic tree and recalculating relative evolutionary rates to create a null distribution for comparison. We used Benjamin–Hochberg method to correct for multiple test correction in R.

Similar to selection analyses, the overabundance of bats representing hibernators was considered as a potential source of bias. To explore the influence of this bias on the dataset, the following nested RERconverge analyses were run: (i) Removing bats from the dataset entirely and (ii) Setting bats as the only foreground. We report the RERconverge results for all hibernators alongside the associated significance data for the same dataset using only bats as foreground species and report the rank of each gene by adjusted P-value. This allows the interpretation of this data to be viewed unaltered but within a context of how strongly each gene might be influenced by signal from the bat lineages. This methodology improves on past work which has used Bayes factor analysis to account for overabundance of bat signal, as it allows us to see convergent rates of hibernating genes with a continuous weighted measure of the influence of one lineage without having to assign cut-off values (Christmas et al. 2023).

Genes significant in RERconverge analysis were further tested for positive and relaxed selection. Because RERconverge data is derived from protein alignments, not DNA, each significant RER gene (Geiser 1994) was checked for availability on NCBI ortholog database and, if available, were downloaded and filtered as detailed above (the same as KEGG pathway genes). These genes (27 total available) were also tested for positive and relaxed selection using the same methods detailed above.

Enrichment Analysis

To assess whether genes identified as significant (either by RERconverge or HyPhy tests of selection) were significantly enriched for specific function, enrichment analyses were performed on all datasets using the DAVID bioinformatics resource (https://david.ncifcrf.gov/). Because genes under selection would already be enriched for certain pathways by virtue of being chosen for their association with carnitine metabolism, and only a small number of genes were recovered both under selection and as evolving under convergent rates (RERconverge), we used the following approach to reduce statistical noise and dataset bias. Each dataset (19,610 genes for RERconverge genes and 378 genes associated with carnitine metabolism) was tested for enrichment by comparing it to all annotated human genes in DAVID. Subsequently, a subset of genes identified as significant in RERconverge analyses, positive selection analyses, and relaxed selection analyses (separately), were similarly tested for enrichment by comparing to all human genes in DAVID. Enrichment terms reported as significant in both the total set and the subset were removed, as enrichment of these terms would represent dataset-specific bias. Terms that were significantly enriched in the subsets only were kept and used to generate figures and final enrichment tables (supplementary fig. S3, Supplementary Material online).

Results

Bear Blood Metabolic Data

To identify clear contrasts in 159 metabolic indicators, we focused on comparisons of hibernating and waking black bears. Classification performance of the sparse partial least squares discriminant analysis (sPLS-DA) model between hibernating and waking showed that these two components (latent variable) reflect 23% of the original data's variation (Component 1: 14% of the variance, Component 2: 9%) (Fig. 3a; supplementary table S1, Supplementary Material online).

The sPLS-DA model identified three metabolic indicators—carnitine esterification, beta oxidation, and the ratio of acetylcarnitine to carnitine—that strongly define the hibernating cluster (Fig. 3b; supplementary table S2, Supplementary Material online). These indicators were selected based on their high loading weights on latent variables 1 and 2, which were optimized to maximize the covariance between metabolite profiles and group membership while penalizing less informative variables. The variable weights contributing to each component are detailed in supplementary table S3, Supplementary Material online. Although metadata such as sex, age, and weight are available for each sample (supplementary table S1, Supplementary Material online), these covariates were not included in the sPLS-DA model, which was designed to maximize discrimination between hibernation states. However, exploratory analyses comparing these variables between early and late denning groups (supplementary figs. S4 to S6, Supplementary Material online) indicate no significant differences, suggesting they are unlikely to drive the observed metabolic patterns. Calculations for carnitine esterification and beta oxidation were based on established Biocrates algorithms; a link to their methodology can be found here: https://biocrates.com/metaboindicator/. Of the raw metabolites that are included in metabolic indicator calculation, carnitine and acetylcarnitine, both are significantly differentially abundant during hibernation (Fig. 3d;  supplementary fig. S1, Supplementary Material online). All three indicators showed an inverse relationship with raw carnitine abundance (and to a lesser degree acetylcarnitine), as the metabolic indicators were all significantly higher in hibernation (Fig. 3c and d; supplementary fig. S1, Supplementary Material online). Both carnitine and acetylcarnitine were significantly differentially abundant during hibernation; raw carnitine was significantly lower in hibernating bears, while acetylcarnitine was significantly higher in hibernating bears (Fig. 3c and d; supplementary fig. S1, Supplementary Material online). These findings implicate metabolic pathways related to carnitine (including fatty acid metabolism and beta oxidation) as critical to metabolic shifts from hibernation to waking. To better pinpoint genes potentially responsible for altering carnitine levels and related pathway shifts, our subsequent bioinformatic analysis focused on pathways which were related to carnitine in Kyoto Encyclopedia of Genes and Genomes (KEGG). While numerous metabolic indicators were quantified in this dataset, carnitine-related pathways—specifically carnitine esterification, beta oxidation, and the acetylcarnitine:carnitine ratio—emerged as the strongest and most consistent signals distinguishing hibernating from waking bears in the sPLS-DA. These indicators had the highest variable loadings across both latent components, and their associated raw metabolites were also significantly differentially abundant. While a small number of additional indicators showed moderate separation in sPLS-DA or borderline differences in abundance, none showed the same consistency or strength across both metrics. We therefore focused on this dominant signal to prioritize pathways with the clearest biological and statistical support. Future analyses could expand on other metabolic signatures identified in the broader dataset (see supplementary table S2, Supplementary Material online) to examine additional dimensions of seasonal physiological change.

Positive Selection in Genes Associated with Carnitine Metabolism

To assess the signal of positive selection across 371 genes identified as potential contributors to carnitine metabolism (plus 33 genes identified as significant in RER analysis for a total of 404 genes across 206 species), we performed several nested statistical tests of selection using BUSTED in the HyPhy software suite. Grouping species into a single foreground can be a useful way to identify convergent selection, however, it also runs the risk of elevated type three error, or misattribution of the origin of signal for selection (Kowalczyk et al. 2021; Drabeck et al. 2024). We, therefore, report results for all selection tests in the context of a set of nested tests designed to narrow down convergent signatures of positive selection to hibernating lineages and minimize the potential for false positive signal from pervasive selection across the tree. A total of eighteen genes were found to be evolving under positive selection (Fig. 4; supplementary table S4, Supplementary Material online). Though hibernation has evolved independently in several bat lineages, including several species from the same lineage always has the potential to bias tests of selection for nontargeted physiology (e.g. physiologies unique to bats only). To address this, we similarly ran the same tests of positive selection removing bats from the analysis. When all but one hibernating bat is removed from the dataset, half of these genes remain significant (supplementary table S4, Supplementary Material online). Though six of these remaining nine genes are also significant in a hypothesis-free test (with no foreground specified), all of them lose significance when performing a drop test (removing all hibernators from the dataset, and performing a hypothesis-free test), indicating signal of positive selection originates primarily from the lineages of hibernators, and not from bats alone or pervasive selection across mammals (supplementary table S4, Supplementary Material online). We did a similarly comprehensive literature review of genes under positive selection (which retained significance when the majority of bats were removed (supplementary table S4, Supplementary Material online) and found genes under positive selection in hibernators were related to oxidative stress and hypoxia (Zhou et al. 2001), metabolic shifts (Ruf and Geiser 2015), thermogenesis and cold tolerance (Geiser and Ruf 1995), and innate immunity (Geiser and Ruf 1995) (Fig. 4).

Fig. 4.

Summary of genes found to be significant for tests of positive selection (+), tests of relaxed selection (−), convergently faster rates of evolution (^), and convergently slower rates of evolution (⌄) among hibernators. Details of each result and its involvement in hibernation-related physiologies can be found in supplementary tables S4 and S5, Supplementary Material online, and supplementary fig. S2, Supplementary Material online, and an expanded discussion of each can be found in supplementary Discussion, Supplementary Material online. Bottom panel shows the metabolomic-guided genes under selection and their KEGG pathways.

Relaxed Selection in Genes Associated with Carnitine Metabolism

Statistical signal of increased diversifying mutations on a gene may be indicative of positive selection or relaxation of selection, and telling apart these two processes is notoriously difficult (Wertheim et al. 2015). To distinguish these two scenarios, we use the test RELAX in the HyPhy suite which tests the fit of a model in which all sites' omega values are either moving toward 1 (neutrality) or away from 1 (Wertheim et al. 2015). Similar to tests of positive selection, signal of relaxed selection may be subject to misattribution (type three error), and so we report the results of nested hypothesis-free (BUSTED) tests as well as the same tests run removing bats entirely from the dataset (Fig. 4; supplementary table S5, Supplementary Material online). Six genes were found to be evolving under relaxed selection in all hibernators and remained significant when removing bats from the data. An additional 14 genes were significant for relaxed selection when only bears (Ursus spp.) were placed in the foreground (Fig. 4; supplementary table S5, Supplementary Material online). All but two genes (of 20 total) were not significant in a hypothesis-free BUSTED analysis, indicating a potential for high rates of diversification across the tree in these 18 genes (Fig. 4; supplementary table S5, Supplementary Material online). We also placed these genes under hibernation-related phenotypic categories based on an extensive literature review and found genes related to metabolic shifts (Andrews 2019), coagulation and dehydration (Perez De Lara Rodriguez et al. 2017), oxidative stress and hypoxia (Zhou et al. 2001), skeletal and muscle preservation (Ruf and Geiser 2015), and thermogenesis and cold tolerance (Ruf and Geiser 2015) (Fig. 4; supplementary Discussion, Supplementary Material online).

Convergent Rate Analysis

Phylogenetic analysis of 120 mammalian genomes resulted in a total of 19,610 gene trees (trees file available on Dryad “19610_mammalGeneTrees.tre”). Convergent rate analysis selecting all hibernators as the foreground identified ten genes evolving convergently at significantly slower rates (negative Rho) in all hibernators when compared to nonhibernating mammals (Table 1). Twenty-three genes were also identified as evolving at significantly faster rates convergently when compared to nonhibernating species (Table 1). No genes were identified as significant when bats were removed from the foreground, suggesting that convergence with bats genes influence these signals. This is unsurprising given that removing bats cuts the foreground lineages in half also significantly decreasing the power of the overall test. To compensate for this, we ran nested tests of evolutionary rate convergence in order to rank the influence of the bat lineages on each gene recovered as significant among all hibernators (supplementary fig. S2, Supplementary Material online). When bats were placed as the only foreground species, 573 genes were identified as evolving at significantly convergent rates. However, genes identified under convergent rate evolution for all hibernators were not a simple subset of the most significant bat-convergent genes but rather were distributed across the 573 genes ranked by their P-value (supplementary fig. S2, Supplementary Material online, Table 1). This ranking system reveals that while the significance of RER analysis for some genes is clearly mainly driven by bat lineages, (e.g. GJA8, NDST3, CNGB3), others are more so driven by the convergence of bats and other hibernators (supplementary fig. S2, Supplementary Material online, Table 1). Using an extensive literature review, we categorized each gene by the hibernation-related phenotypes to which they are most likely to contribute. We found genes evolving at convergent rate in hibernators related to tissue regeneration and repair (5 faster, 1 slower), gene regulation (2 faster, 3 slower), metabolic shifts (4 faster, 1 slower), oxidative stress and hypoxia (3 faster, 1 slower), coagulation and dehydration (2 slower), skeletal muscle preservation (2 faster, 1 slower), thermogenesis and cold tolerance (1 faster), and innate immunity (1 slower) (Fig. 4; supplementary Discussion, Supplementary Material online).

Table 1.

Results of convergent rate analysis performed with RERconverge binary trait analysis placing all hibernating mammals in the foreground, using 19,610 gene trees derived from a 120-mammalian genome (protein) alignment

Gene symbol	Rho	N	Padj	Padj bats only	Rank (1 to 573)
Genes evolving at convergent rates across all hibernators (RERconverge)
Genes evolving slower than average
IGSF22	−0.230058392	201	0.053724143	NS	NS
ZNF473	−0.218498392	221	0.054110144	0.031039	303
SLC4A9	−0.225346488	210	0.053724143	0.026846	275
CYP27A1	−0.262287744	195	0.03084742	0.01288	107
PPAN	−0.233693402	192	0.054110144	0.01288	111
HROB	−0.225983985	221	0.050117271	0.011678	92
PDZD7	−0.237011544	223	0.038115889	0.010806	84
RREB1	−0.226448056	213	0.051857967	0.00923	68
BDP1	−0.223987666	228	0.050117271	0.006494	45
SHROOM4	−0.228902052	212	0.051397465	0.004722	34
Genes evolving faster than average
FAM19A4	0.365000536	95	0.038115889	NS	NS
PDZRN3	0.252750446	169	0.053724143	0.045897	452
MNX1	0.293421504	167	0.018624689	0.024597	238
FRMD4B	0.25743257	187	0.038115889	0.021874	191
LCE1B	0.294470449	127	0.051857967	0.020725	184
PRSS12	0.243972247	201	0.045888219	0.017695	163
LRRN1*	0.273312637	158	0.046117471	0.013337	117
SARM1	0.260834114	175	0.045888219	0.013054	113
RGS4	0.245855993	173	0.054110144	0.010488	80
NECAB1	0.380097581	75	0.053724143	0.00923	65
TMEM64	0.258961508	163	0.051857967	0.008652	59
YJU2	0.242320316	179	0.054110144	0.008129	54
HOXB13	0.272711205	153	0.050117271	0.004722	28
CNTN6	0.398099728	120	0.002248326	0.004722	24
SYNPO2	0.273467722	207	0.01031322	0.004722	17
CX3CL1	0.231588736	211	0.050117271	0.003785	15
GJA8	0.284920435	195	0.01031322	0.002137	6
NDST3*	0.262815157	162	0.051397465	0.002026	3
CNGB3	0.243104716	210	0.038115889	0.002026	1

Genes evolving under significantly convergent evolutionary rates using “permulations” (phylogenetically informed simulations) and adjusted for multiple test correction are reported below. Significance for the same gene when run using only bats as foreground taxa is also reported, as well as a rank number which represents where the gene (significant for all hibernators) fell in a distribution of bat-only significant genes ordered by P-value. For example, when bats only are in the foreground CNGB3 was the number 1 most significant gene, whereas ZNF473 was the 303rd. Asterisks denote genes which were also significant in tests of selection (see supplementary tables S4 and S5, Supplementary Material online). Rho is a measure of correlation between the relative evolutionary rate and the trait over all branches (negative numbers are slower than average and positive numbers are faster than average). N is the total number of branches in the gene tree, and Padj values are P-values after correction for multiple tests using the Benjamini–Hochberg method.

Enrichment Analysis

To identify annotated gene function which may be overly represented in genes identified by RER converge, or HyPhy selection tests, we used the DAVID bioinformatics resource to group and test functional enrichment (Sherman et al. 2022). No genes identified in the convergent rate analysis (33 genes) were overrepresented for any functional annotation term when using either the full 19,610 genes derived from 120 mammalian genomes as background or the default DAVID human genome database (all human annotations). Because the dataset of 371 genes we already derived based on relation to carnitine metabolism, we considered “unique enrichment” rather than total enrichment (as function is already bias toward carnitine metabolism). We define unique enrichment as terms enriched in the 28 genes under selection but not enriched in the original 371 gene dataset when each dataset is compared to all human genes as a background. Genes experiencing both positive and relaxed selection (28 genes) were significantly enriched for several terms when using all human annotations as a background. Terms that were uniquely enriched in the 28 genes experiencing positive or relaxed selection and not when all (371) carnitine-associated genes were used in the same analysis are reported in supplementary fig. S3, Supplementary Material online (raw data available on Dryad “RawEnrichmentData.xls”). Terms that were strongly uniquely enriched in genes under positive selection are strongly linked to hibernation-related phenotypes including: Thermogenesis, Respiratory Chain, NADH Dehydrogenation, and Mitochondrion. Genes under relaxed selection were enriched for a larger number of terms with less clear links to hibernation, with a notable exception for terms related to insulin and glucose metabolism (supplementary fig. S3, Supplementary Material online).

Discussion

While carnitine has been identified as a key metabolite in hibernating squirrels, its role in bear hibernation has not been fully characterized (Burlington and Shug 1981; Belke et al. 1998; Nelson et al. 2009; Nelson et al. 2010; Cooper et al. 2014; Luan et al. 2018; Heinis et al. 2023). However, recent work on brown bears (Ursus arctos) has shown that during hibernation, there is a metabolic shift that reduces the production of trimethylamine (a byproduct of carnitine metabolism by gut bacteria) and trimethylamine N-oxide (TMAO), whereby bears' metabolisms favor the production of betaine over TMAO (Ebert et al. 2020). Our data reveal reduced carnitine levels during black bear hibernation, potentially facilitating the metabolic shift toward betaine production observed in recent studies. Understanding the pathways and genes that contribute to these shifts help to reveal the genetic basis of this vital adaptation as well as shed light into avenue of carnitine regulation with respect to its role as a widely used biomarker for a diverse range of conditions (McCann et al. 2021). While our analyses point to specific genes with evolutionary patterns consistent with selection in hibernators, we acknowledge that these findings are correlational and do not on their own demonstrate functional involvement in hibernation. These genes should therefore be viewed as strong candidates for further functional validation in appropriate experimental systems.

Through integrated metabolomic and comparative genomic analyses, we identified convergently evolving genes across multiple hibernating lineages, potentially underlying key phenotypic changes required for hibernation (Fig. 4). Though many genes have previously been identified as having altered regulation during hibernation, signal from coding sequences suggest that these genes experience diverse selection pressures—positive, relaxed, or purifying—shaping their function and evolution at the sequence level. We also identify several novel genes, previously unknown to contribute to hibernation but related to several key hibernation-related physiologies.

Below we highlight significant genes in this study related to metabolic shifts and provide a broader review of other physiologies in the supplementary Discussion, Supplementary Material online (Fig. 4). These results generate hypotheses about potential functional roles but must be interpreted considering the limitations of comparative genomic selection analyses, which cannot alone establish causation. Enrichment analysis revealed that genes under positive selection are enriched for fewer terms than those under relaxed selection, consistent with the expectation that relaxed selection reflects a more neutral evolutionary process (supplementary fig. S3, Supplementary Material online). However, in our own literature review, we find genes under relaxed selection to be directly related to metabolic shifts and carnitine, with several known to cause beneficial effects when inhibited or knocked out in mice. While carnitine-related genes under selection are more heavily focused on metabolism and oxidative stress related functions, convergent rate analysis revealed genes that are more diverse in function (such as genes involved in gene regulation and the immune system) and may contribute to more diverse aspects of hibernation physiology (Fig. 4, supplementary Discussion, Supplementary Material online). Both approaches reveal important insights into the mechanisms that underlie shifts in metabolic indicators found in this study, as well as other aspects of hibernation physiologies found in previous studies in a wide range of hibernating species. Below, we use the following shorthand symbols to discuss genes under positive selection: GENE+, relaxed selection: GENE−, faster convergent evolutionary rates: GENE^, and slower convergent evolutionary rates: GENE⌄ in hibernating mammals (Fig. 4).

With carnitine closely linked to fat metabolism, it is unsurprising that the greatest number of genes (Thienel et al. 2023) recovered in our dataset are related to metabolic shifts (Fig. 4). PPP1R3B+ encodes a glycogen-targeting subunit of protein phosphatase 1 (PP1) that plays a crucial role in regulating hepatic glycogen synthesis, storage, and fasting glucose metabolism. PP1 has been shown to be downregulated in the liver and brain of torpid squirrels and hibernating Syrian hamsters and has been uniquely shown to be unaffected by temperature, though mechanisms for these changes are unknown (MacDonald and Storey 2007; Coussement et al. 2023). While previous work showed that profound inhibition/downregulation and temperature stability of this pathway is vital for processes involved in hibernation, our work identifies PPP1R3B as a key target gene under positive selection which may be contributing to these functional shifts (Coussement et al. 2023). While this gene's signal of selection and evidence of downregulation in prior hibernation studies are compelling, further functional work is needed to determine whether and how its activity has been modified in hibernators.

With respect to hibernation, much interest has been paid to the function of FGFR1− as a receptor that mediates the effects of FGF21 in coordination with the coreceptor β-Klotho, which can induce a torpor like state in mice, is upregulated in hibernating mammals, and is linked to the regulation of food intake, body weight, and DIO2 (a thyroid hormone enzyme) expression (Kharitonenkov et al. 2005; Nelson et al. 2013; Ni et al. 2015). The FGFR1 signaling pathways regulates glucose homeostasis, lipid metabolism, and thermogenesis, and inhibition of the FGFR1 signaling pathways has been shown to reduce adiposity, body weight, and increase insulin sensitivity (Foltz et al. 2012; Ni et al. 2015; Sun et al. 2024). Recovery of relaxed selection signal in this gene may suggest reduced evolutionary constraint, which could parallel the physiological effects observed when FGFR1 is downregulated or inhibited in experimental models—such as reduced adiposity and increased insulin sensitivity (Nelson et al. 2010; Cooper et al. 2014; Sun et al. 2024). While this interpretation is speculative, it raises the possibility that a decrease in FGFR1 function may be adaptive during hibernation, and further functional studies are warranted to test this hypothesis. Two genes identified as being under relaxed selection (ACAT2− and ACSL5−) are involved in lipid metabolism and have previously been identified as being differentially expressed in several studies of hibernating mammals and overwintering frogs (Yan et al. 2008; Nespolo et al. 2018; Srivastava et al. 2019; Niu et al. 2023). ACSL5 activates long-chain fatty acids for biosynthesis, while ACAT2 is primarily involved in cholesterol metabolism and transport, its overexpression in mice may be protective of high-fat-diet-induced weight gain (Lee et al. 2000; Rudel et al. 2005; Lopes-Marques et al. 2013; Ma et al. 2023). Both ACAT2 and ACSL5 are directly linked to carnitine as they play key roles in lipid metabolism, particularly in the activation and transport of fatty acids for β-oxidation. NR1H3− has not been previously linked to hibernation but is a nuclear receptor that regulates the expression of genes involved in cholesterol, lipid, and glucose metabolism by promoting the efflux of cholesterol from cells to prevent accumulation as well as influencing inflammatory immune response (Zhao et al. 2021). ACAA2− and GCDH− are both involved in fatty acid β-oxidation, and similarly have been previously identified as being differentially expressed across several diverse mammals as well as in hibernating frogs (Keyser et al. 2008; Reilly et al. 2013; Xu et al. 2013; Hindle et al. 2014; Ballinger et al. 2016; Houten et al. 2016; Srivastava et al. 2019).

Other notable genes involved in lipid metabolism are NR1H4^, HOXB13^, and CNTN6^, all of which we find to be evolving at convergently faster rates among hibernators. NR1H4, differentially expressed in hibernating black bears, was recently identified as an upstream regulator of genes that showed altered expression after a mid-hibernation feeding of captive brown bears and was predicted to facilitate metabolic shifts to glucose metabolism (Srivastava et al. 2019; Perry et al. 2023). HOXB13 and CNTN6 have both been shown experimentally to be strongly associated with fat deposition, fat percentage, and metabolism (Smirnov et al. 2018; Xu et al. 2023).

Evolving slower than average in hibernators, PPAN⌄ (Peter Pan homolog) is a gene involved in cell growth and proliferation, playing a key role in ribosome biogenesis and protein synthesis, and has been shown to be downregulated during diapause in fruit flies (Drosophila melanogaster) (Kučerová et al. 2016). Ribosome hibernation factors (e.g. RSFS) are known to be employed to reduce energy consumption by silencing ribosomes and keeping them inactive while preserving their readiness for translation (Fatkhullin et al. 2022). Slowed evolutionary rates in PPAN may preserve its compatibility and function as a ligand for silencing factors.

In addition to shifting to fat metabolism, hibernating species also cope with prolonged starvation by inducing a state of decreased insulin sensitivity (mimicking type-2 diabetes) and reducing insulin production. Three genes identified in this study, PTPN11+, RPTOR−, and MNX1^ are all critical to these processes. RPTOR has been shown to be upregulated and hyperphosphorylated in hibernating mammals and in leeches (respectively), and RPTOR knockout mice were protected against high-fat-diet-induced obesity, insulin resistance, inflammation, and expansion of adiposity in bone marrow (Laplante and Sabatini 2012; Jiang et al. 2014; Chayama et al. 2019; Shi et al. 2020; Tangseefa et al. 2021). MNX1, not previously identified to be involved in hibernation, is a homeobox gene that encodes a transcription factor essential for the differentiation and proper functioning of insulin-producing beta cells in the pancreas (Flanagan et al. 2014). PTPN11 has been linked to glycogenesis and obesity-induced infertility and may facilitate maintenance of fertility despite the massive metabolic changes (weight loss and gain) required to weather a hibernation season (Phung et al. 1997; Zhang et al. 2023; Fonseca et al. 2024; Shuai et al. 2024).

Genes contributing to oxidative stress and hypoxia also make up a large proportion of genes recovered in this work and emphasize the interrelatedness of metabolic shifts and hypoxic stress (Fig. 4, supplementary Discussion, Supplementary Material online). Specifically, genes encoding subunits of Mitochondrial Complex 1 (NDUFA genes) are novel to hibernation biology and are indicated three separate times in our selection analyses. Although Logan and Storey (2021) predicted that NDUFA9 and NDUFA10 might be targeted by miRNAs during torpor based on computational analyses of differentially expressed miRNAs in brown adipose tissue, these subunits have not been empirically linked to hibernation in any prior studies, and no other NDUFA subunits have been previously identified in hibernation-related datasets. Genes related to coagulation and dehydration (Chazarin et al. 2019), skeletal and muscular preservation (Wolf et al. 2018), tissue representation and repair (Perez De Lara Rodriguez et al. 2017), gene regulation (Perez De Lara Rodriguez et al. 2017), thermogenesis (Wu and Storey 2016), immunity (Ruf and Geiser 2015), as well as genes that are likely to be important strictly in bat lineages are important findings of this work and are discussed in greater detail elsewhere (supplementary Discussion, Supplementary Material online). Of note are the SLC4 family genes, which regulate pH balance and ion homeostasis. These genes were independently recovered four times in our analyses and, to our knowledge, have not been previously implicated in hibernation biology.

Metabolomic data in this work indicate carnitine metabolism as a significant physiological component of hibernation. This is particularly relevant to critically ill sepsis patients as degree of mitochondrial dysfunction and increased mortality have been linked to early elevated levels of acylcarnitine (Puskarich et al. 2018; Evans et al. 2019; Jennaro et al. 2023). Measurements of the carnitine pool could lead to important advances in the diagnosis, prognosis, and treatment of these diseases (McCann et al. 2021). Although it remains unclear whether reduced carnitine levels drive or result from metabolic shifts during hibernation, manipulating carnitine levels or targeting regulatory genes identified here could inform the development of hibernation-based therapies.

Our selection and convergent rate analyses identified both known and novel genes potentially underlying the physiological adaptations of hibernation, offering a valuable framework for future research on clinically relevant therapeutic targets. Importantly, many of the genes under selection represent a confluence of signals across diverse hibernating taxa, highlighting shared mechanisms that have evolved independently in multiple lineages. While we outline potential contributions of these genes to specific physiological components, many are embedded in highly pleiotropic pathways, and their exact roles remain to be experimentally validated. Rather than definitive evidence of function, our results provide a roadmap for translational and functional studies aimed at uncovering the mechanistic basis of these adaptive shifts. Future work should focus on phenotypic characterization and validation of these candidates to fully realize their potential for informing both evolutionary biology and biomedical applications.

Supplementary Material

Supplementary material is available at Molecular Biology and Evolution online.

Data Availability

Though we are unable to publish equations for metabolic indicators which are available to Biocrates software licenses only, authors may be contacted for specific inquiries on metabolic indicator calculation. All equations and calculations are readily available via Biocrates software licenses. Raw enrichment data, species trees, species topologies, and gene trees used for selection tests and RERconverge analysis are available of Zenodo: https://doi.org/10.5281/zenodo.15529828. Code for processing data can be found here: https://github.com/drabe004/BearProject_Scripts.