Metabolic changes associated with the long winter fast dominate the liver proteome in 13-lined ground squirrels

Abstract

Small-bodied hibernators partition the year between active homeothermy and hibernating heterothermy accompanied by fasting. To define molecular events underlying hibernation that are both dependent and independent of fasting, we analyzed the liver proteome among two active and four hibernation states in 13-lined ground squirrels. We also examined fall animals transitioning between fed homeothermy and fasting heterothermy. Significantly enriched pathways differing between activity and hibernation were biased toward metabolic enzymes, concordant with the fuel shifts accompanying fasting physiology. Although metabolic reprogramming to support fasting dominated these data, arousing (rewarming) animals had the most distinct proteome among the hibernation states. Instead of a dominant metabolic enzyme signature, torpor-arousal cycles featured differences in plasma proteins and intracellular membrane traffic and its regulation. Phosphorylated NSFL1C, a membrane regulator, exhibited this torpor-arousal cycle pattern; its role in autophagosome formation may promote utilization of local substrates upon metabolic reactivation in arousal. Fall animals transitioning to hibernation lagged in their proteomic adjustment, indicating that the liver is more responsive than preparatory to the metabolic reprogramming of hibernation. Specifically, torpor use had little impact on the fall liver proteome, consistent with a dominant role of nutritional status. In contrast to our prediction of reprogramming the transition between activity and hibernation by gene expression and then within-hibernation transitions by posttranslational modification (PTM), we found extremely limited evidence of reversible PTMs within torpor-arousal cycles. Rather, acetylation contributed to seasonal differences, being highest in winter (specifically in torpor), consistent with fasting physiology and decreased abundance of the mitochondrial deacetylase, SIRT3.

small temperate-zone hibernators are unique among mammals in their ability to survive a prolonged fast by exploiting the cold to drastically reduce energetic requirements. During hibernation, many species including the 13-lined ground squirrels studied here switch from an active homeothermic mode to one of winter heterothermy. In winter, these animals spend most of their time in an extreme physiological state known as torpor, which is characterized by dramatically reduced metabolic, heart, and respiratory rates, and core body temperature (T_b) for periods lasting between a few days and a few weeks. Torpor is periodically interrupted by metabolic activation and rapid rewarming that transiently restores more typical mammalian physiology for ∼12 h before the cycle begins anew (reviewed in Ref. 5).

The strong temporal segregation between periods of homeothermy (activity) and heterothermy (hibernation) in ground squirrels led us to propose a two-switch model for this pattern of circannual hibernation (45). Specifically, the model postulates that baseline physiology alters seasonally between activity and hibernation and that only the hibernation mode is permissive for torpor (Fig. 1B). Fluctuations of liver metabolites representing just three states from the ground squirrel's year support the predictions of that model (45), as do plasma metabolites (10) and proteomes in skeletal muscle (20), heart (15), and kidney (28), all quantified from six or more states representing the annual cycle. A mathematical model independently predicted that two switches define the T_b rhythms characterizing a hibernator's year (17). Further evidence for specific changes in baseline physiology associated with this hibernation phenotype is provided by data demonstrating cell membrane remodeling to increase polyunsaturated fatty acid content in the fall (2) and by the seasonal sensitivity of arctic ground squirrels to N(6)-cyclohexyladenosine, an adenosine A(1) receptor agonist that induces torpor only in winter (29).

Fig. 1.

Seasonal homeothermy and heterothermy in circannual hibernation. A: the different physiological states examined in this study are indicated on this representative body temperature (T_b) trace: 2 homeothermic groups, Sp (spring) and SA (summer active); 4 heterothermic, winter hibernating groups, IBA (interbout aroused), Ent (entrance into torpor), LT (late torpor), and Ar (early arousing), comprised the base states. Two fall transition groups, FT1 and FT2, were distinguished by whether the animal had previously been torpid (see methods for details). B: model depicting heterothermic states, including all 4 winter hibernation states and FT2 (the fall transition animals that had used torpor based on T_b records), constrained to a distinct baseline physiology (light blue shading).

Additional features of hibernation in ground squirrels that are important to consider in terms of gaining mechanistic insight into their strong seasonal and torpor-arousal cycles are that differential gene expression likely plays a role (48), but that role is necessarily limited by the near complete depression of transcription at the low T_bs of torpor (53). Thus, the dramatic transition from torpor to arousal is particularly unlikely to be mediated by new transcription. Translation, too, is limited by the depression of initiation as animals cool during entrance into torpor (54), although a subset of mRNAs with internal ribosome entry sites may be selectively translated during early arousal (39, 55).

Alternatively, posttranslational modification (PTM) to alter protein localization or activity provides a low-energy mechanism that could be rapidly invoked to control and stabilize cells during torpor-arousal cycles. The fact that PTMs require far less energy than modulation of protein levels via differential expression is consistent with the energy-saving role of torpor (Ref. 20 and references therein). Control of enzyme activity via phosphorylation has long been linked to hibernation (50). In the few studies with sufficient sampling to dissociate temperature effects from seasonal effects, both T_b and time of year can be shown to affect protein phosphorylation in hibernators (6, 15, 20–22). The importance of another PTM, acetylation, in metabolic control has been discovered recently (61).

The ground squirrel liver provides unique opportunities to study the extremes of mammalian physiology on two fronts. First, liver is a crucial metabolic hub. Studies of liver are therefore particularly relevant to hibernation, which is characterized by transient obesity and metabolic syndrome followed by a prolonged fast accompanied by reliance on stored fat to meet metabolic fuel requirements (34). Second, the liver from hibernating 13-lined ground squirrels, whether taken from torpid or aroused hibernators, is resistant to damage when subjected to cold storage followed by warm reperfusion, compared with livers from summer squirrels or rats (33). This observation suggests that the seasonal resetting of baseline physiology described above includes increasing resistance to damage related to ischemia-reperfusion injury. Although previous studies of hibernator liver have examined changes in mRNA, proteins, and metabolites that accompany distinct phenotypic states of hibernation, these studies have been limited to just two (9, 12, 42) or three states (45, 46, 57, 59), thereby severely limiting the ability to distinguish seasonal and torpor-arousal cycle changes.

Here, we report the results of a proteomics screen to quantify and identify protein changes that accompany the phenotypic transitions across the ground squirrel's year. Specifically we used the quantitative fluorescence two-dimensional (2D) gel method, difference gel electrophoresis (DiGE), to evaluate protein spot differences. This method was chosen because it provides a robust means to uncover both quantitative and qualitative (i.e., alternative isoforms and PTMs) protein changes. These data were used to test a key prediction of the two-switch model; protein changes occurring largely at the level of differential gene expression distinguish homeothermic and heterothermic states, whereas PTM of proteins, altering their localization and activity without altering overall abundance, dominate torpor-arousal cycles. Ground squirrels in two homeothermic states, spring (Sp) and summer active (SA), were compared with winter heterothermic animals in late torpor (LT), arousing (Ar), interbout aroused (IBA), and entrance (Ent) states of hibernation. We also considered animals undergoing the fall transition to test the hypothesis that the liver proteome is reset to a new baseline “torpor-permissive” mode to support heterothermy. This hypothesis predicts that proteins distinguishing active homeotherms from hibernating heterotherms would remain at homeothermic levels in the fall transition (FT) animals that had not yet begun to use torpor, but reset to hibernation levels in the heterothermic FT animals. Therefore two groups of FT animals were studied: 1) those that had previously used torpor based on T_b data recordings and 2) those that had not. The differences in protein abundance, phosphorylation, and acetylation revealed by this study confirm (12, 42) and substantially extend our understanding of the protein adjustments in 13-lined ground squirrel liver that support the hibernation phenotype.

METHODS

Animals.

Thirteen-lined ground squirrels (Ictidomys tridecemlineatus) were purchased from the University of Wisconsin at Oshkosh captive breeding program and were transported in July of 2006, 2007, or 2008 to Colorado as described in Refs. 10, 43. Animals were transferred in late September or early October to an environmental chamber that functions as a hibernaculum. The chamber was maintained at 4°C in constant darkness; food and water were removed after the animals entered torpor but were provided again in the spring when torpor bouts naturally reduced in length and the time spent euthermic increased. Ground squirrels received surgical abdominal implants of radiotelemeters (VM-FH disks, Minimitter) and dataloggers (iButton, Embedded Data Systems) in late August or early September. Dataloggers assured recovery of T_b history for all groups except the August SA animals; radiotelemetry enabled remote monitoring and collection of tissues with precision during winter hibernation (Fig. 1A). Animals were euthanized for tissue collection in specific stages of their annual cycle, n = 6/group as follows: SA animals, Aug. 4–8; all other groups were defined based on T_b captured by abdominally implanted remote telemetry or ibuttons. FT animals, sampled Sept. 21–Oct. 20, were initially divided into two groups, FT1 and FT2, based on T_b history. None of the FT1 animals had previously entered torpor (T_b < 30°C for at least 3 h), whereas all in the second group (FT2) had undergone at least one torpor bout (T_b < 30°C for at least 12 h). All FT had T_b > 32°C at the time of tissue collection except for one of the FT2 animals, which was entering torpor with T_b = 19°C. Three of the six FT1 and all of the FT2 animals were still in conventional housing, whereas three FT1 individuals had been moved to the hibernaculum where the room was constantly dark with a temperature of 4°C. Food and water were available to all FT, Sp, and SA animals. The four groups of hibernating animals were also in the hibernaculum, but food and water were withdrawn: Ent, entering into torpor with T_b 27–23°C between Jan. 3 and 31; LT, typically after 7–10 days of torpor calculated as 80–95% of previous torpor bout length with T_b near 4°C between Feb. 1 and Mar. 4; Ar, during spontaneous endogenous arousal from torpor with T_b 7–12°C between Dec. 28 and Jan. 22; IBA, ∼3 h after reaching euthermic T_b of 35–37°C following a torpor bout and spontaneous natural arousal between Nov. 19 and Mar. 26. Because torpor patterns lengthen and stabilize during midhibernation, and all four hibernation groups were collected when torpor bouts were long and reproducible in length, we call these animals winter hibernators (but note that the stable pattern, and thus “winter,” may establish prior to and extend beyond the official calendar definition of winter). When T_b traces began to show rapid torpor-arousal cycles and cessation of hibernation (e.g., March in Fig. 1A), food and water were returned to the animals. Sp squirrels were euthanized Mar. 21–Apr. 27 after spontaneous spring arousal and maintenance of euthermic T_b for 11–20 days while still housed in the dark and cold (4°C). The Sp and SA groups represent the active season while bracketing hibernation, thus, despite their common homeothermy, Sp and SA animals are physiologically distinct with regard to body weight (Fig. 2B) and, based on an extensive literature (reviewed in Ref. 5), reproductive and nutritional status.

Fig. 2.

Key parameters of animals used in this study. Individuals are represented by (●) and the mean of each group by a horizontal line; groups that differed are indicated by small letters. A: date, Aug. 1 = day 1. B: body mass at time of liver collection. C: number of males (black bar) and females (gray bar) in each group.

Ground squirrels were deeply anesthetized with isoflurane, assessed by their lack of response to toe pinch; this process took 10 ± 3 min and differed (P = 0.03) only between LT (13.5 ± 3.1 min) and Ar (7.5 ± 3.3 min). Animals were then euthanized by cardiac exsanguination followed by perfusion with ice-cold saline; this took 12.3 ± 3.2 min, with no differences among groups (P = 0.24). Livers were excised, flash-frozen in liquid nitrogen, and stored at −80°C until use. All animal work was conducted according to the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health; procedures were approved by the University of Colorado School of Medicine Institutional Animal Care and Use Committee (protocol #44309).

Protein samples.

Frozen liver (∼100 mg) was placed in ice-cold homogenization buffer (0.5 M sucrose, 100 mM phosphate, 5 mM MgCl₂, 1 mM PMSF, 10 μg/ml protease inhibitors, 2× phosphatase inhibitor cocktail HALT from Thermo Scientific, Rockford, IL) and immediately homogenized as previously described (9). The homogenate was transferred to a microfuge tube and spun at 500 g, 10 min at 4°C. Aliquots of the supernatant were snap-frozen in liquid nitrogen and stored at −80°C; each aliquot was used only once. One aliquot from each animal was used for protein quantitation by BCA protein assay (Pierce, Rockford, IL). Another aliquot from each animal was used to prepare the reference standard.

DiGE.

Protein samples were labeled and fractionated by 2D gel electrophoresis as previously described (11, 15). Gels were scanned to maintain quantitative pixel volumes for all spots. Similarly run gels with the unlabeled protein pool were stained with SyproRuby and scanned (pick gels). All pick gel and CyDye gel images were imported into DeCyder 2D 7.0 software (GE) for spot matching and quantitation. Spot pixel intensities were exported from DeCyder; Cy3 and Cy5 spot values were normalized to their corresponding Cy2 volumes and used for all subsequent analyses. Only spots with q ≤ 0.05 that were found to be reproducible and robust by individual inspection on all Cy2 images were excised for identification.

Spot picking, mass spectral analysis, and protein identification.

Two gels were used, with 417 and 382 spots picked, respectively, for a total of 584 unique spots (some were picked on both gels). Picks and digests were done in the University of Colorado School of Medicine Mass Spectrometry and Proteomics core facility using an Ettan spot picker and an Ettan spot digester (GE Healthcare). Tryptic fragments were separated in a 3–50% hydrophobicity gradient of buffer B (90% ACN, 0.1% formic acid) over 7 min, followed by a 3 min wash in 90% ACN using an HPLC Chip Cube (Chip #G4240-62001, Zorbax 300SB C18 RP; Agilent Technologies, Santa Clara, CA) for analysis by liquid chromatography-tandem mass spectrometry (LC-MS/MS) in an ESI ion trap (LC/MSD XCT Plus, Agilent Technologies) in positive ion mode in the Ultra Scan setting. A full MS scan was followed by tandem mass spectral (MS/MS) scans of the four highest peaks with a dynamic exclusion of 30 s. Spectral data were collected with 6300 Series TrapControl V6.1 Build 83 software (Agilent Technologies), and raw mass spectra were analyzed using Spectrum Mill MS Proteomics Workbench Rev A.03.02.060a ETD-65. Spectral data were initially compared with a database containing 792,655 entries, which was compiled in-house from all available mammalian sequences at the National Center for Biotechnology Information (NCBI) in April 2010 and all available ground squirrel proteins from Ref. 46. Subsequently, the recently released 13-lined ground squirrel protein database (Ensembl; Ictidomys_tridecemlineatus.spetri2.69.pep.all) was used to re-evaluate weak or unsuccessful IDs from the initial search. For all searches, the range of mass limits for the precursor ions was 600–4,000 Da, parent and fragment masses were both set to monoisotopic, precursor peptide mass tolerance was 2.5 Da, fragment ion tolerance was 0.7, the enzyme specified was trypsin, maximum number of internal missed cleavage sites was two, and cysteines were given a fixed modification of +57 for carbamidomethylation.

ExtracTags was used to collate Spectrum Mill results (35). For each peptide, the NCBI GI#, species of the organism to which that peptide was matched, predicted peptide sequence, Spectrum Mill score, scored peak intensity, parent mass and charge, and delta mass were all recorded. In “homology” mode, ExtracTags included peptides that missed up to two residues from the best protein ID in order to map the peptides from homologous proteins that had otherwise not been merged for inclusion in the overall protein coverage. For each protein ID obtained, a combined score from all peptides and the total amino acid coverage by those peptides were also calculated. Protein IDs required a minimum of two peptides and a score >30. Additionally, for spots with multiple credible identifications meeting these criteria if the average spectral intensity for one protein ID exceeded those of another protein in the same spot by fourfold or more, the protein with lower average spectral intensity was discarded (35). Trypsin and keratin were eliminated. Identifications that passed these criteria were considered valid, and spots with single valid protein IDs after these eliminations were considered unique IDs. In cases where the top ID was not from human, the name of the human RefSeq ortholog was found using BLink (NCBI). In the text, proteins are largely referred to by their unique gene identifiers to reduce ambiguity (Supplemental Tables S1, S2).¹

Data analysis.

Hibernation state-dependent alterations in protein spot intensity were assessed by ANOVA followed by Hochberg multiple test correction (3) and then Tukey tests to identify pairwise changes in R, version 3.0.2 (51). Random Forests (RF) analyses using 50,000 trees in both supervised and unsupervised modes in R were used for sample classification and to define main classifiers (4). RF in variable selection mode (8) was repeated five times to identify top classifiers. The 280 identified protein spots were clustered into six intensity patterns using the divisive method, DIANA (7), as implemented in R, with Pearson correlations using mean normalized intensity values for each state. Genes present in each of these six clusters were subjected to functional annotation analyses compared with the human background using DAVID (25) with classification stringency set to high for functional annotation clustering; only functional groups with scores ≥ 1.3 were reported. For comparison with previously published datasets, we compared the corresponding Tukey significant subset (e.g., between SA and Ent animals) of data from the appropriate two-state comparison, after establishing significance in the six-state ANOVA, with either all significant changes (9, 12, 42) or the Qspec significant changes (46). A protein was considered concordant if the direction of change was identical in the two datasets and discordant if it was opposite.

Western blots and analyses of PTM by phosphorylation and acetylation.

Protein phosphorylation changes were assessed by comparing spot intensity obtained after staining of 2D gels with ProQ Diamond (Molecular Probes, Eugene, OR) with that of SyproRuby as described previously (20) using Sp, Ent, and Ar samples (n = 3 of each). Acetylation among nine stages in the hibernation cycle (n = 3 per group for all 8 proteomics groups plus animals in early torpor, 5–10% duration of previous bout length) was also assessed by 1D Western blot using 20 μg of each total liver protein extract and antibodies for acetyl-lysine (1:1,000, rabbit pAb #9441; Cell Signaling Technology, Danvers, MA) and β-tubulin (1:1,000, goat pAb #ab21-57; Abcam, Cambridge, MA), followed by detection with IRdye-conjugated secondary antibodies (1:25,000 680LT anti-goat and 800CW anti-rabbit; Li-Cor, Lincoln, NE). Protein bands were visualized using the Odyssey near-infrared imaging system (Li-COR) and analyzed with ImageQuant TL software (GE Healthcare). To correct for inconsistencies in protein loading, acetyl-lysine protein band intensities were normalized to the β-tubulin protein content in the same lane. A one-way ANOVA across all groups followed by pairwise Tukey post hoc tests assessed total acetylated lysine (whole lane) and individual band volume. To identify specific acetylated proteins in these bands, the acetyl-lysine antibody was covalently cross-linked to beads (Crosslink Magnetic IP Kit, Pierce). Immunoprecipitates were obtained according to the manufacturer's protocol, fractionated by SDS-PAGE, and visualized with SyproRuby staining. The proteins in excised bands were identified by LC-MS/MS as described above. Additional acetylated protein spots among the set identified in the DiGE analysis as changing across the annual cycle were identified after fractionation of CyDye-labeled SA, IBA, and LT protein extracts (n = 3 each) on 2D gels, followed by Western blotting as described for the 1D blots, above. Sirtuin 3 (SIRT3) Western blots were developed using a 1:200 dilution of rabbit pA#86671 (abcam, Cambridge, MA) followed with the IR secondary, as described above except that the ANOVA was followed by a Dunnett's multiple comparison to identify states differing from SA.

RESULTS

Protein differences among multiple physiologically distinct states of a hibernator's year were quantified using a 2D gel approach (DiGE). We examined liver homogenates from six individuals from each of six base states, plus 12 animals with variable physiology representing the fall transition (Fig. 1). The base states included two homeothermic groups, SA and Sp, and four winter hibernating groups (defined as in methods), interbout aroused, entrance, late torpor and early arousing, denoted here as IBA, Ent, LT, and Ar, respectively. Animals representing the phenotypically highly variable fall transition state (43, FT) were FT1 with no prior record of torpor, and FT2, with least one previous torpor bout. The animals examined in this study differed not only by physiological status with respect to T_b and T_b history, but also by their environmental conditions, time of year, body weight, and sex (Fig. 2). The 2D DiGE analysis provided relative intensity information for 2,472 matched protein spots. After filtering for spots present in at least four of the six gels per state, we ran two ANOVAs on the remaining 1,317 spots to identify those that differed significantly among: 1) the six base groups: Sp, SA, IBA, Ent, LT, and Ar and 2) all eight groups, i.e., six base plus two FT groups. In the six-state comparison, 525 spots (40%) differed significantly (q < 0.05). Adding the two FT groups to the analysis gave similar results; 521 (40%) of tested protein spots significantly differed by hibernation state.

There were relatively few protein differences between Sp and SA (24) or among the four hibernating groups (range from 1 to 56), but many protein changes separated Sp and SA animals from those in IBA, Ent, LT, and Ar (135 to 283; Table 1, top right of diagonal). Significantly, both FT groups were more similar to Sp and SA, particularly SA, than they were to any of the winter groups, independent of previous torpor use (Table 1, top right). The FT groups were, however, closer than either Sp or SA to the winter groups, indicating they are not only transitioning physiologically toward the winter state, but also transitioning biochemically. The majority of intrawinter differences were found in comparisons involving animals in the process of arousing from torpor (Ar). We also independently assessed the relatedness of spot intensity data among groups using RF (4); with the subset of all spots present on all gels (630) RF results supported the statistical findings by classifying Sp nearest to SA, separated from IBA, Ent, LT, and Ar, with Ar as the most distinct winter group (Fig. 3A). As with the ANOVAs, the fall animals were more similar to SA and Sp than to any of the winter states (Fig. 3A).

Table 1.

Number of protein spot differences between groups

	Sp	SA	FT1	FT2	IBA	Ent	LT	Ar
Sp		24	22	36	145	234	223	283
SA	17		12	7	135	210	203	233
FT1	16	6		4	85	143	134	188
FT2	21	4	2		61	130	137	170
IBA	83	81	53	36		1	14	56
Ent	120	109	80	68	1		3	38
LT	111	107	72	73	6	2		20
Ar	141	120	98	92	22	16	13

Number of pairwise differences (Tukey < 0.05) among 521 ANOVA significant (q < 0.05) DiGE spots (upper right), or among 280 identified spots (lower left); the full and selected datasets are highly correlated with R = 0.99.

Sp, spring; SA, summer active; FT, fall transition; IBA, interbout aroused; Ent, entrance; LT, late torpor; Ar, arousing.

Fig. 3.

Random forests classification by liver protein spot intensities. A: classification of individual samples from all 8 groups using the relative intensities of 630 protein spots present on all gels (n = 6 per state). B: as in A, except using the 186 identified protein spots that were present on all gels (n = 6 per state). Nine of the top 10 classifiers from the full analysis (A) were identified and also appeared in the top 10 named protein classifiers (B). These proteins are (ID_spot number): HIBADH_1664, DMGDH_407, DBT_1007, SND1_767, PRDX3_1839, MPST_1552, GPT_1010, APOA1_1849, and IVD_1237.

The proteins in 280 of the spots that differed among groups were successfully identified by LC-MS/MS. Of these, 224 differed whether or not the two FT groups were included, whereas 30 lost and 26 gained significance upon inclusion of FT (Supplemental Tables S1, S2). Because the spots with identified proteins represented a subset of the spots that differed, we re-evaluated statistical and RF analyses on this subset to verify that the identified proteins were representative of the changes occurring in the full dataset. Based on the counts of significant differences among groups (Table 1, lower left), the subset of identified protein spots was highly correlated with the full dataset (R = 0.99). Moreover, the RF clustering pattern using just the identified spots generally mirrored that obtained with the full dataset (Fig. 3, compare B with A). Therefore, the identified proteins faithfully recapitulated the significant changes in the liver proteome as ground squirrels cycled among physiological states, and all subsequent analyses were focused on this subset of 280 identified protein spots.

The RF results in Fig. 3, together with the significant differences reported in Table 1, indicate a seasonal separation of individuals based on the liver proteome, with more limited separation between Sp and SA and among hibernating groups; i.e., the homeotherms (Sp and SA) are more similar to each other than they are to the heterotherms, which are also more similar to one another than to either Sp or SA. To identify the top proteins responsible for this separation and deconvolute seasonal and torpor-arousal cycle changes we ran RF in variable selection mode with either: 1) the six base states or 2) just the four hibernating states. These analyses (Fig. 4) revealed two proteins that most strongly distinguish active-period homeotherms from hibernating heterotherms, hydroxyisobutyrate dehydrogenase (HIBADH) and dihydrolipoamide branched chain transacylase E2 (DBT); three proteins that are informative for both seasonal and torpor-arousal cycle changes, PRDX3, DMGDH, and SND1; and two proteins that distinguish among the four states within the torpor-arousal cycle, FGB and ALB.

Fig. 4.

Identification of key proteins that distinguish among ground squirrels in different stages of the seasonal and torpor-arousal cycles of hibernation. A: Venn diagram separates the classifiers identified in the 6-state RF analysis (Sp, SA, IBA, Ent, LT, and Ar: red) from those identified using just the winter states (IBA, Ent, LT and Ar: blue). Each RF was run 5 times: for the 4-state winter dataset these 5 protein spots were recovered 100% of the time. For the base state analyses, the 3 spots shared with the winter analysis were recovered in all 5 attempts, but HIBADH and DBT were additionally recovered 3 and 2 times, respectively. B–H: boxplots show the distribution of relative intensity values for animals from each state for each of these top classifying proteins, identified by their gene name and spot number. Proteins vary: B–D across season and hibernation state; E and F seasonally; G and H within winter hibernation. The horizontal lines, triangles, circles, rectangles, and whiskers represent the median, mean, outliers, 25–75th percentile, and 5–95th percentile, respectively.

To gain biological insight into the protein dynamics of hibernator liver, we grouped the relative intensities of the 280 identified proteins among the six base states for gene enrichment analysis. With DIANA top-down clustering, just three patterns (clusters 1–3) accounted for most (82%) of the protein spot differences (Fig. 5, Supplemental Table S2). Specifically, 100 spots were elevated in both Sp and SA compared with all four winter groups (cluster 1), 87 spots were lowest in Sp and highest in Ar due to a within-hibernation cycle (cluster 2), and 44 spots exhibited the inverse pattern to cluster 1, i.e., elevated in the hibernating heterotherms compared with the active homeotherms (cluster 3). The genes in each of these three clusters were subjected to enrichment analysis using DAVID (Refs. 25, 26; Table 2). Cluster 1, with proteins reduced throughout hibernation, was enriched with enzymes involved with glucose, amino acid, hormone and xenobiotic catabolism. The opposite pattern, i.e., increased in hibernation (cluster 3) was enriched for proteins with roles in fatty acid catabolism in the peroxisome, ion homeostasis, and regulation of apoptosis. Distinct subsets of mitochondrial proteins were overrepresented in both clusters 1 and 3. In contrast to the strong bias toward metabolic enzymes in the seasonally changing patterns, the proteins in cluster 2, which exhibited the strongest fluctuation during the torpor-arousal cycle, were noteworthy for their lack of metabolic enzymes. Instead, cluster 2 was enriched in cytoplasmic vesicle and plasma proteins and proteins involved in protein localization, ubiquitin-dependent protein catabolism, redox homeostasis, response to oxidative stress and negative regulation of apoptosis. The protein spots comprising the remaining three clusters (clusters 4–6, Fig. 5) differed strongly between Sp and SA and by whether they were relatively low or high in winter but contained too few proteins (<20) for meaningful enrichment analysis by DAVID.

Fig. 5.

Liver protein dynamics in 6 stages of hibernation. Heat map plots relative intensity (blue, gold, and white represent least, most, and missing values) for each protein spot (rows, ordered based on divisive clustering into 6 abundance patterns) for each individual (columns, ordered by physiological state). The mean spot intensity value ± 95% confidence interval for each of the 6 patterns is plotted to the right of the heat-map. The gene enrichments for the 3 most populated clusters (1–3) are given in Table 2.

Table 2.

Functional enrichment categories in the three major patterns of protein intensity differences among the six base states

Cluster	Term	Enrichment Score	Count	P Value	Fold Enrichment
1	mitochondria	13.8	18	1.44E-15	14.44
	hexose metabolism	11.7	13	9.04E-13	19.91
	amino acid catabolism	4.0	5	7.45E-05	21.62
	triglyceride metabolism	1.8	3	8.96E-03	20.52
	hormone metabolism	1.4	3	1.64E-02	14.95
2	cytoplasmic vesicle	9.5	19	8.23E-11	6.76
	organelle lumen	4.5	22	9.60E-06	2.76
	cell redox homeostasis	4.1	5	2.07E-04	16.78
	regulation of apoptosis	2.9	12	1.10E-03	3.15
	negative regulation of apoptosis	2.9	8	1.24E-03	4.78
	ubiquitin-dependent protein catabolism	2.7	5	1.29E-03	10.36
	protein localization	2.6	12	2.30E-03	2.88
	plasma	2.2	5	2.77E-04	15.67
	response to oxidative stress	1.9	4	4.09E-02	5.16
3	peroxisome	5	6	6.11E-06	21.90
	mitochondrial membrane	2.8	7	4.51E-04	6.68
	regulation of apoptosis	2.3	8	4.68E-03	3.64
	fatty acid catabolism	1.83	3	4.10E-03	30.47
	cellular ion homeostasis	1.5	5	1.69E-02	4.89

DAVID enrichment terms (Term), Enrichment Score, Count (number of proteins associated with that term), P Value, and Fold Enrichment values are shown for each pattern of protein abundance (Cluster). The patterns for each are: cluster 1, decreased in winter; cluster 2, increased in LT and Ar; cluster 3, increased in winter (see Fig. 5).

We next examined the FT animals to gain insight into the molecular details of the transition between homeothermy and heterothermy. First, we asked whether the observed liver proteomic changes were consistent with the a priori division of the FT animals into two groups based on T_b records. Importantly, if these two groups were distinguishable and gene expression is reprogrammed to switch from a “torpor excluded” to a “torpor permissive” mode, protein differences would be expected to identify pathways linked to the switch between homeothermic and heterothermic phases of the circannual rhythm. The spot intensity data from the analyses presented in Table 1 reveals just two differences between FT1 and FT2. These were identified as hydroxymethylglutaryl-CoA synthase (HMGCS2) and NDUFS8, but neither exhibited the expected pattern of FT1 being more like the homeotherms and FT2 more like the heterotherms (Table 1; Supplemental Tables S1, S2). An intermediate seasonal status of FT individuals is revealed by the number of pairwise protein spot differences between either FT group and the base states (Table 1), and is supported by RF (Fig. 3); FT1 and FT2 individuals were typically juxtaposed and closer to SA and Sp than to the hibernating groups, but nonetheless beginning to move away from the homeotherms towards the heterotherms. The torpor-naïve FT1 animals were more variable than the FT2 individuals, all of whom had previously used torpor. Surprisingly, the closest FT individual to the hibernators (IBA, Ent, LT, or Ar animals) was an FT1 (Fig. 3). Consistent with this observation, when the spot intensity data from the six base states were used as a training set to classify the FT samples, 11 of the 12 (FT1 + FT2) were classified predominately as homeotherms (Sp and SA) rather than as heterotherms (IBA, Ent, and Ar), despite the fact that the euthermic FT2 animals could reasonably be considered equivalent to IBA hibernators. Based on these findings, the final analysis of the FT proteome was done treating these 12 animals as one group.

Given that a number of pathways distinguished the homeotherms from the hibernators, we next examined FT animals for acquisition of proteomic attributes characteristic of heterothermy. For this, the 144 proteins in the base-state DIANA clusters 1 and 3 (Fig. 5) were clustered by abundance pattern using just SA, FT, and IBA individuals. The euthermic aroused state (IBA) was chosen to represent winter hibernation because: 1) it was most similar to SA and FT (Table 1, Fig. 3); 2) four of the six FT2 animals were effectively interbout aroused, i.e., euthermic following a prior torpor bout based on T_b; and 3) among the winter states RF prediction placed most of the FT animals (11 of 12) closest to IBA (not shown). The spot intensity differences between SA, FT, and IBA were described by eight clusters, with the six depicted in Fig. 6 containing 138 of the 144 (96%) spots analyzed. Consistent with the other analyses involving animals undergoing the FT (Fig. 3, Table 1), most spots were similar between SA and FT (Fig. 6, compare the 58 spots in clusters 1 and 4, to the 14 spots in clusters 3 and 6). FT individuals were intermediate between SA and IBA in clusters 2 and 5, containing a total of 66 protein spots. Of the seven proteins identified by RF as key for separating the base states (Fig. 4A), both of the two that were most important for differentiating homeotherms from hibernators (HIBADH_1664 and DBT_1007) were present in this FT analysis; DBT retained its high SA abundance in FT, whereas HIBADH had completed its transition to match IBA (Fig. 6, clusters 1 and 3; Supplemental Table S2). Gene enrichment analysis of these FT patterns reveals that FT ground squirrels retain relatively high glycolytic and amino acid catabolism signatures of homeothermy, as well as low regulation of apoptosis and proteins involved in ion homeostasis. FT also displays intermediate signatures of declining glucose metabolism in favor of increasing fatty acid metabolism. Protein spots that had already achieved their winter state in FT were too few for gene enrichment analysis. In addition to HIBADH_1664, CPS1_249, DMGDH_413, and the lone FABP7 spot abundances were reduced in preparation for hibernation (Fig. 6, Supplemental Table S2).

Fig. 6.

Liver proteome dynamics through the fall transition. Line plots (means ± SE) for 6 patterns of protein change occurring among SA, FT, and IBA in the proteins comprising cluster 1 (A) and cluster 3 (B) in Fig. 5, i.e., decreased and increased in winter heterothermy, respectively. Among these SA-FT-IBA patterns, A plots cluster 1 with 42 spots, 22 unique proteins, enriched for mitochondria, glucose catabolism, gluconeogenesis, and Aln, Ala, Asp, Arg, and Pro metabolism; cluster 2 with 44 spots, 23 unique proteins, enriched for mitochondria and glucose metabolism; and cluster 3 with just 10 spots and 9 unique proteins (Supplemental Table S2). In B, cluster 4 with 39 spots, 35 unique proteins, enriched for peroxisome and fatty acid metabolism; cluster 5 with 17 spots, 16 unique proteins, enriched for regulation of apoptosis, mitochondria, and ion homeostasis; and cluster 6 had only 4 spots, all unique proteins.

We also evaluated the liver proteome for state-dependent protein isoform/PTM shifts in hibernation, initially by asking whether proteins identified in more than one spot were in distinct, reciprocal abundance clusters. Of the 280 spots identified, 117 proteins were found in only one spot and 56 proteins were identified in at least two, but in one case (CPS1) as many as 22 distinct spots. Although 30 of the proteins identified in more than one spot changed concordantly with hibernation physiology (i.e., were in the same DIANA cluster), 26 proteins appeared in more than one cluster, which raises the possibility of biologically relevant differential regulation by PTM. These spots were examined for phosphorylation after staining 2D gels with a phospho-specific protein stain. None of the identified proteins with spots belonging to more than one DIANA cluster were unambiguously attributable to differential phosphorylation, although numerous phosphoproteins were detected in this experiment (Supplemental Table S2). Moreover, no clear seasonal shift in overall phosphorylation abundance was detected by comparing the ratio of stained phosphoprotein to total protein in n = 3 2D gels from Ent, Ar, and Sp (Fig. 7A). In fact, the majority of the detected phosphoproteins did not differ quantitatively among the selected physiological groups in the DiGE experiment and hence were not identified in our study. One exception was the protein NSFL1C_1144 (p47), detected as only a single, phosphorylated protein spot (Fig. 7B) with an abundance pattern linked to the torpor-arousal cycle (cluster 2).

Fig. 7.

Posttranslational modification of liver proteins: protein acetylation is increased during winter hibernation. A: overall phosphorylation level (vs. SYPRO Ruby total protein stain intensity) in 2D gels is constant among Sp, Ent, and Ar (means ± SD, n = 3) ground squirrels. B: the same 2D gel stained for phospho- and total protein, using ProQ Diamond and SyproRuby, respectively, shows that spot 1144 with NSFL1C is a phosphoprotein. C: Western blot to detect acetyl-lysine reveals several bands with increased acetylation in winter heterotherms. Subsequent IP identified proteins in those bands as indicated on the left. D: relative abundance (vs. β-tubulin) of all acetylated bands detected by Western blot (mean ± SD, n = 3 from each state, small letters indicate distinct groups). E: overall acetylation labeling (vs. background) in 2D Western blots is most intense in winter hibernators, particularly LT (n = 2 SA, IBA, n = 3 LT; mean ± SD). F: the same 2D gel labeled by acetyl-lysine antibody for acetylated proteins (Ac-Protein, top) and for total protein with Cy2 (bottom) shows differential acetylation of 2 spots containing DMGDH (spot 413 is acetylated, 407 is not). G: Western blot of CPS1 shows decreased abundance in winter heterotherms, consistent with DiGE quantification. H: representative Western blot of SIRT3, quantified in I after normalization to β-tubulin (n = 3 per state, mean ± SD), shows decrease of SIRT3 during the winter fast [ANOVA P = 0.03; IBA, LT, and Ar ≠ SA (*), P < 0.04].

State-dependent changes in hepatic protein acetylation were also evaluated; the importance of this PTM for metabolic control was recognized recently (56, 61). Western blot analysis of ground squirrel liver protein extracts from multiple physiological stages revealed increased acetylation of discrete bands during winter and especially during torpor (Fig. 7, C and D). This result was supported by an increased total acetylation signal (normalized to total protein as labeled by Cy fluors) in winter, particularly during torpor, from 2D Western blots (Fig. 7E). We used two approaches to identify the acetylated proteins: 1) immunoprecipitation with the acetyl-lysine antibody and 2) 2D Western blot to determine whether any of the protein spots identified by DiGE as different among physiological states were acetylated. Proteins recovered from early torpor extracts, but not from control reactions done in parallel but lacking primary antibody, were identified in the bands as indicated in Fig. 7C. Alternatively, protein spots that bound the acetyl-lysine antibody on 2D Western blots were recognized by labeling the entire protein sample with CyDye and aligning that total protein image to the DiGE reference gel. As with the phosphoprotein analysis, there were no clear examples of reciprocal patterns that could be attributed to changes in protein acetylation. However, we did detect acetylated protein isoforms bearing different, nonreciprocal patterns, as well as several clear examples of acetylated proteins that were the sole isoform identified. For example, acetylation modified a key classifier of the hibernation cycle, DMGDH, as identified by RF (Fig. 4). Present as two isoforms in the DiGE dataset (Fig. 7F), unacetylated DMGDH_407 increased in SA (cluster 5, Fig. 5) and was a key RF predictor (Fig. 4). Acetylated DMGDH_413, on the other hand, was elevated in both Sp and SA (cluster 1). PCK1_681 and PCK2_676 were each identified in just one spot in this dataset; both were acetylated and decreased in winter. Acetylated CPS1 was identified in the immunoprecipitates, and the abundance of the band identified as CPS1 follows the general torpor-increase of the acetylated liver proteome (Fig. 7, C and D). On 2D Western blots, several of the 22 different isoforms of CPS1 identified in the DiGE experiment were modified by acetylation (data not shown). Yet the overall abundance of CPS1 as detected by DiGE (cluster 1, Supplemental Table S2) or Western blotting decreased in winter (Fig. 7G).

SIRT3 is a major regulator of metabolism via its activity as a protein deacetylase in the mitochondrion (19). Because a large number of the proteins identified in this study were mitochondrial proteins (Table 2) and because the increased acetylation observed during torpor was reminiscent of that observed during caloric restriction (44), we examined SIRT3 content in liver protein extracts from seven states by Western blotting (Fig. 7, H and I). The abundance of SIRT3 was reciprocal to that of liver protein acetylation (compare Fig. 7I with 7D); reduced SIRT3 and hence its deacetylase activity in winter is consistent with the observed increase of protein acetylation.

DISCUSSION

Hibernation in ground squirrels comprises dynamic cycles wherein physiological parameters alternate between those of a typical mammal and the dramatically depressed state of torpor. The deep torpor of hibernation occurs concomitant with a long period of fasting. The extremes encountered during torpor are lethal to nonhibernators yet are evoked by hibernators numerous times each winter (5). The phylogenetic distribution of hibernating species implies that all mammals share the genes needed to orchestrate this extraordinary physiological phenotype (48). Because phenotype results from the expression and activity of proteins, we hypothesized that a proteomics screen for changes in protein abundance and PTM, both of which are revealed by the DiGE technique employed here, would identify key components of the hibernation phenotype in liver. Given the large contribution of this organ to metabolic rate and its importance to whole body metabolism, particularly metabolic homeostasis, liver is expected to be a key target organ for metabolic suppression. We also anticipated that liver would best reflect protein changes related to the winter fast, enabling a consideration of fasting-dependent and fasting-independent adjustments in hibernation. Earlier work led us to hypothesize that baseline physiology of the animals and their underlying biochemistry is modified in winter, such that it is only upon this altered “winter state” that the extremes of the torpor-arousal cycle are evoked (45). The predictions of this hypothesis are supported by proteomics data collected from at least six physiological states in ground squirrels from heart (15), skeletal muscle (20), and kidney (27) but less so in brain and brown adipose tissue (21, 22). Liver has not been explored using a similarly broad sampling strategy.

Previous studies examined changes in the liver proteome for a limited number of states in golden-mantled, 13-lined, and arctic ground squirrels. Of the 91 unique proteins that differentiate SA from Ent hibernators in the present study (Tukey SA-Ent ≤ 0.05, Supplemental Table S1), just 21 were shared with the golden-mantled ground squirrel dataset (9), and all but five (16 of 21, 76%) changed concordantly. Interestingly, these five were all metabolic enzymes; in the present study AASS, ALDH7A1, PC, and PCK2 were increased in SA compared with Ent, whereas NIT2 was elevated in Ent, opposite to what was observed with the golden-mantled ground squirrels. While these few discordances could reflect species-specific differences in the metabolism of hibernation, we suspect it is more likely that they reflect differences in the timing or treatment of the SA animals; in the golden-mantled ground squirrel study, SA animals were sampled earlier in the summer (late June) shortly after being trapped in the wild. Another two reports identified protein differences between SA and Ent livers in 13-lined ground squirrels. One of these also used DiGE (12), whereas the other used a metabolically labeled internal reference standard for quantification (42), thereby avoiding the use of 2D gels. Of the 33 proteins that were recovered in the present and the earlier DiGE study as altered between SA and Ent, all but one (ACO2), or 97%, changed concordantly. These concordant proteins were dominated by mitochondrial proteins, consistent with the mitochondrial-based metabolic reprogramming of the hibernator's liver that was described previously (12). Likewise, in the comparison to the Rose et al. (42) dataset, all but one of 15 proteins, HAO1, changed concordantly. It is worth noting that at least two isoforms of both ACO2 and HAO1 were identified here, raising the possibility that protein isoforms with distinct kinetics during hibernation may have differentially recovered in the various studies. Nevertheless, the strong agreement observed between our present findings and those of earlier work using the same species, albeit entirely different animals, with the same (12) or different (42) methodologies strongly corroborates the findings reported by all three.

The findings that emerge from this work can also be compared with those obtained using a shotgun proteomics method to define pairwise changes among three states of the annual cycle in arctic ground squirrels: late torpor, early aroused, and nonhibernating, postreproductive animals, most comparable to our LT, IBA, and Sp, respectively (46). This comparison reveals a surprisingly small overlap among the significantly different proteins recovered between the two datasets; of the 102 unique protein IDs that differ in these three pairwise comparisons in the 13-lined ground squirrels, just 12 were shared with the lists of proteins that differed in the arctic ground squirrels. Again, these differences may be due to the different species, proteomics methods, and animal treatments used. Specifically the postreproductive arctic ground squirrels were later in the homeothermic period than our Sp 13-lined ground squirrels but preceded our SA animals. The early aroused arctic ground squirrels also were sampled at a different (earlier) time in the interbout euthermic period than our IBA animals. However, despite differences among the specific proteins identified among the various studies, it is noteworthy that the key findings that emerge from the gene lists and gene enrichment profiles are consistent; the heterothermic season is characterized by increased abundance of proteins involved in fatty acid catabolism, protein synthesis, folding, and trafficking, whereas the increased proteins in the homeothermic season are those involved with glucose and amino acid catabolism, and detoxification.

Defining the seasonal transition.

Inclusion of additional physiologically distinct stages including Sp, Ent, Ar, and FT animals in this study provides new insights into classes of protein changes and their kinetics in both seasonal and torpor-arousal cycles. For example, proteins increased in SA relative to one of the hibernation states may be increased in homeothermy (i.e., elevated in both Sp and SA, cluster 1, Fig. 5) or be increased in only SA compared with all of the other states (cluster 5). Despite the small number of proteins (just 16) that are elevated in SA alone, this cluster is dominated by enzymes involved in glucose metabolism (glycerol-3-phosphate dehydrogenase 1, aldolase B, dicarbonyl/l-xylulose reductase and enolase) and others (acetyl-CoA acetyltransferase and transketolase) consistent with this time being relatively important for cell and tissue growth.

Half of the identified protein abundance changes (51%) supported a shift between homeothermy and heterothermy (clusters 1 and 3, Fig. 5). Pathway enrichments imply that feeding and fasting are a key difference between homeotherms and heterotherms in ground squirrel liver. Although this feature of hibernation in sciurid rodents is well known (5), the interrelationship between fasting status and expression of heterothermy at the molecular level had not been explored by this type of broad screening approach. We therefore examined this liver dataset with the intent to identify components of the hibernation-capable phenotype of the heterotherms that were consistent with fasting physiology and those that were not. The enrichment patterns that distinguished the homeothermic groups (Sp and SA) from the winter heterotherms (IBA, Ent, LT, and Ar) were largely metabolic, revealing winter elevation of lipid catabolism and protection of amino acids from breakdown (Table 2). The two most important proteins defining this seasonal switch were HIBADH and DBT (Fig. 4). Both support branched-chain amino acid catabolism, and their strong winter depression is consistent with a protein sparing response to fasting. The two fatty acid binding proteins (FABP) in these data linked to fasting but also to other features of the annual cycle. FABP7 is decreased throughout winter, whereas FABP1 is increased, particularly in Ent, LT, and Ar compared with IBA. These observed abundance patterns are consistent with the known role of FABP7 in cell proliferation and that of FABP1 in boosting lipid catabolism in the fasted state (reviewed in Ref. 49).

Analysis of FT can begin to separate effects of torpor use and fasting. Ground squirrels using torpor in standard animal housing where food and water were continuously available differ from the hibernators using torpor in the hibernaculum where food and water were withdrawn to mimic natural conditions. Previous work demonstrates widespread behavioral variation during FT in 13-lined ground squirrels; affected parameters include the timing of the first use of torpor, as well as the length and frequency of torpor bouts thereafter (43). From individual body weight records, we infer that food intake and metabolic expenditure varied widely during FT, although neither was measured directly. We did not uncover a significant relationship between weight loss prior to sampling (as a fasting metric) and the proximity of the liver proteome to homeotherms or winter heterotherms. Interestingly, the lone FT animal that was most like the hibernators (the FT1 nearest to an IBA animal in Fig. 3A) exhibited the greatest weight loss just prior to tissue collection compared with all of the other FT animals.

The liver proteome of FT ground squirrels largely reflects an early phase of the transition between homeothermy and heterothermy based on the relative similarity of FT animals to SA (Table 1, Fig. 3) and the small fraction of protein spots (10%) that had attained IBA abundance levels compared with those that remained similar to SA; fully 40% of clustered protein spots remained in their SA condition in the FT animals (Fig. 6). Even the 46% of analyzed spots that exhibited an intermediate abundance in FT tended to be more similar to SA than to IBA (Fig. 6, Clusters 2 and 5). The few proteins that had fully decreased to their winter abundance during FT suggest that slowing or ceasing the conversion of dietary carbohydrates and amino acids into fatty acid for storage occurs early in the fall transition. Fatty acid synthase (FASN), HIBADH, sarcosine dehydrogenase (SARDH), galactokinase (GALK), and one isoform each dimethylglycine dehydrogenase (DMGDH) and carbamoyl-phosphate synthase (CPS1) were among the few spots with this pattern; FASN is the key enzyme for fatty acid biosynthesis; HIBADH, SARDH, and DMGDH degrade dietary amino acids and choline; GALK is important for recovery of the energy residing in galactose from seed mucilage; and CPS1 catalyzes the first committed step in the urea cycle. However, many other enzymes involved in amino acid and carbohydrate catabolism, the majority of CPS1 isoforms, and another key urea cycle enzyme, argininosuccinate synthase (ASS1), had not yet decreased to their winter abundance in the FT animals. Additionally, the FT liver had not fully shifted to elevated levels of most of the proteins involved in fatty acid catabolism and the peroxisome that typify the winter hibernation samples. Two proteins relevant to fasting had fully increased to their winter abundance in the FT animals, however, aldolase B (ALDOB), and one of two isoforms of HMGCS2, involved in gluconeogenesis and ketone biosynthesis, respectively. In terms of proteins other than metabolic enzymes, those involved in ion homeostasis and regulation of apoptosis retained their low SA values in FT; only copine III (CPNE3, a Ca²⁺-dependent membrane-binding protein) and glutathione S-transferase mu 3 (GSTM3, involved in protection against oxidative stress) had transitioned to their high winter abundance.

Taken together, these data highlight the initial protein-level steps toward adopting a fasting metabolism, enhancing tissue protection, and suppressing growth associated with heterothermy despite continued food availability. Rather than fully losing their homeothermic metabolic capacity as they first begin to exploit heterothermy, the observed pathway enrichments suggest a gradual transition during which time FT animals retain the ability to catabolize carbohydrates and amino acids as fuel while turning away from fatty acid synthesis and toward enhancing the protein infrastructure required for winter fatty acid catabolism. Based on the correlation of the FT position at the very start of the seasonal switch (Fig. 1) and the representation of the onset of shifts in the abundance of metabolic enzymes that would support long term fasting (31), we propose the notion that nutritional status is a significant sensor initiating the transition from homeothermy and may in fact provide a necessary foundation for the successful execution of heterothermy.

Significantly, the largely SA liver proteome of FT ground squirrels does not prevent their use of torpor. Among the liver proteins surveyed here there is no evidence supporting the requirement for a specific protein change prior to the onset of torpor. Although this may mean that the two-switch model is incorrect, there are many alternative explanations including: some of the FT1 group had completed the switch to winter mode and were on the verge of torpor, thereby creating so much variability in the data that a significant change was not recovered; none of the key switch components are among the subset of abundant soluble liver proteins captured in our study; and the winter liver proteome is not a cause of the torpor-permissive state but rather a response to multiple bouts of cold torpor (33).

Regulation of proteomic shifts.

The protein abundance changes we have detected may be the result of modified gene expression, altered isoform abundance due to PTM, or sequestration within or on the surface of cells. We examined two types of PTM in this study. Both provide critical regulatory control to proteins and enzymes (40, 61) and are natural candidate mechanisms to rapidly alter protein activity with low overall energetic cost, a strategy that could be exploited across torpor-arousal cycles where transcription and translation are transiently depressed at the low T_b of torpor (53, 54). Phosphorylation control of metabolic enzymes and protein synthesis has been linked to hibernation for decades (50), but their lysine acetylation has not previously been examined in hibernators.

Just one abundance pattern cycled with torpor and arousal in 13-lined ground squirrel livers, regardless of PTM state; proteins in this group (cluster 2, Fig. 5) were generally elevated in winter heterotherms but were lowest in IBA and highest in Ar. Despite the requirement for rapid adjustments in liver metabolic capacity across torpor-arousal cycles, this cluster had relatively few metabolic enzymes and was instead dominated by proteins involved with protein folding, localization, turnover, trafficking, and plasma proteins. In fact, the largest fold changes in this dataset were increased 14- and 12.9-fold in albumin and apolipoprotein A-I in Ar compared with Sp. A large increase of these proteins, as well as transferrin and α₂-macroglobulin, during Ar was previously observed in kidney during hibernation in 13-lined ground squirrels, where it was attributed to a slowing of receptor-mediated endocytosis at low T_b (28). All four of these proteins, plus β-fibrinogen, are made in the liver and secreted into the bloodstream, making it difficult to distinguish between protein secretion and endocytosis, both likely to be affected by low T_b.

This dataset had a surprising paucity of reciprocal abundance patterns among protein isoforms, which would have suggested a protein pool divided among PTM modified and unmodified variants. Instead, we documented a limited number of modified protein spots bearing unique but nonreciprocal patterns compared with other spots of the same identity, as well as uniquely identified, modified spots with differential abundance in hibernation. One example of a protein undergoing torpor-arousal cycle abundance changes in our dataset is NSFL1 (p97) cofactor (p47). This protein occurs in a single phosphorylated spot representing cluster 2 (Figs. 5, 7F), which peaks in Ar and is lowest in homeothermy. Localized primarily in the nucleus, a cytosolic subset of the p47 pool is associated with the Golgi complex, where it interacts with another cluster 2 protein in our dataset, valosin-containing protein, VCP (p97), to mediate membrane fusion together with N-ethylmaleimide-sensitive fusion protein (NSF, 52). The role of p47 in Golgi membrane dis/assembly is best known in mitosis, when animal cells disassemble, disperse, and then reassemble Golgi in daughter cells. Phosphorylated p47 is specifically associated with disassembly as it is unable to bind the Golgi membrane in this state and disrupts the membrane fusion activities of p97 and NSF in early mitosis. Indeed, in vitro assays demonstrate that p47 phosphorylation is required for Golgi membrane disassembly during the cell cycle (52). Low abundance of phosphorylated p47, the form associated with early mitosis and growth, in the homeothermic groups, taken together with its peak in the cold-T_b heterothermic states of torpor and early arousal suggest that the observed dynamics of p47 in ground squirrel liver are functionally important for a process distinct from cell division. The Golgi apparatus has recently been identified as a critical element in starvation by regulating trafficking and providing membrane components for the formation of autophagosomes (13, 36). Macroautophagy is a feature of the starvation response; cytoplasmic proteins and organelles are sequestered into double-membrane vesicles called autophagosomes that then fuse with lysosomes where their contents are degraded and recycled (41). In starved yeast, the p47 homolog, Shp1, is needed to generate autophagic bodies in the cytosol and, along with VCP-p97 (yeast Cdc48), regulates autophagosome biogenesis (30). Thus, we speculate that the intrawinter dynamics of phosphorylated p47 are a feature of autophagy regulation in the 13-lined ground squirrel liver, rather than indicative of mitosis in during winter heterothermy and particularly in the cold. This process may be important to support metabolic reactivation despite reduced perfusion during early arousal, which limits access to circulating substrates, e.g., from white adipose tissue.

Changes of acetylation were more common than phosphorylation among the differentially abundant proteins in this DiGE dataset (Supplemental Table S2). As with phosphorylation, differentially acetylated protein spots lacked reciprocal abundance patterns, although some nonreciprocal but distinct patterns were noted. For example, both the cytosolic and the mitochondrial forms of the key gluconeogenic enzyme phosphoenolpyruvate carboxykinase, PCK1 and PCK2, respectively, were acetylated and had decreased abundance throughout winter. Because liver glucose depletes during torpor but is restored during each IBA (45), gluconeogenesis must persist throughout winter; indeed, enhanced gluconeogenesis is a general feature of caloric restriction (16). Therefore the observed decreased abundance of PCK1 and 2 initially appeared paradoxical. However, the acetylation modification of both PCKs targets them for degradation (58). A winter reduction of acetylated PCK2 isoforms may thus lead to enhanced enzyme activity and liver gluconeogenesis. Another acetylated enzyme, dimethylglycine dehydrogenase (DMGDH), participates in one-carbon metabolism, using the dietary nutrient choline as substrate, and is not surprisingly reduced across the entire winter fast. Although the functional consequence of acetylation is not known for DMGDH, its acetylation state is linked to distinct abundance patterns in homeothermy. The acetylated form (spot 413) is increased immediately upon terminal arousal and is elevated in Sp through SA. Conversely, the nonacetylated form (spot 407) remains low in spring, reaching high abundance only in SA (Supplemental Table S2).

Counter to our expectation that PTMs would be exploited as a low-energy mechanism to regulate metabolic activity in torpor-arousal cycles, the most significant PTM changes in this dataset occurred seasonally as opposed to within winter. Acetylation increased in winter compared with the homeothermic period, although this elevation peaked during torpor, revealing some degree of within-winter variance (Fig. 7). Overall protein phosphorylation remained generally invariant between homeothermic and heterothermic states examined; however, the observed trend for increased phosphorylation in homeothermic Sp animals was consistent with a summer increase in the liver mitochondrial phosphoproteome reported previously (6).

Hyperacetylation of liver mitochondrial proteins has been linked to high-fat diet, fasting, and caloric restriction (18). Acetylation changes are significant in liver mitochondrial proteins after caloric restriction in mice, with this PTM having a larger regulatory impact than protein turnover (19). Our observation of an overall acetylation increase in the liver proteome during the hibernation season appears consistent with adopting a starvation metabolism. Moreover, several examples of key regulatory controls by acetylation were discovered in starved yeast (60). Many of these acetylations promoted autophagy, which we postulate is a necessary process to manage exploitation of local substrates across the winter fast, and particularly during torpor and early arousal when reduced perfusion limits substrate delivery via plasma.

Acetylation can either inhibit or activate enzymatic activity. Enoyl-CoA, hydratase/3-hydroxyacyl CoA dehydrogenase (EHHADH), recovered in our acetyl-lysine immunoprecipitation experiment, catalyzes two steps in FA oxidation and is activated by acetylation (61). It, along with the acetylated forms of acetyl-CoA acyltransferase 2 (ACAA2) and acyl-CoA dehydrogenase (ACADM), likely contributes to the seasonal switch to fatty acid catabolism during hibernation. We also specifically considered five proteins identified by Schwer et al. (44) as targeted for hyperacetylation during the environmental stress of caloric restriction, the alpha subunits of ATP synthase (ATP5A1), pyruvate dehydrogenase (PDHA1) and the mitochondrial trifunctional protein (HADHA), carbamoyl-phosphate synthase (CPS1), and succinate dehydrogenase (SDHA). DiGE revealed a hibernation state difference in three of these five proteins. Reaction with acetyl-lysine antibody on 2D Western blots confirmed that spots containing ATP5A1 and CPS1 were acetylated; the PTM status of HADHA was inconclusive. The pair of ATP5A1 isoforms bore slightly different abundance patterns; the leftmost spot, 995, reacted most strongly with the antibody, was most increased in winter, and remained elevated during spring emergence (cluster 6, Fig. 5). The correlation between acetylation and increased abundance in terminally aroused spring animals (common to ATP synthase and dimethylglycine dehydrogenase) may reveal the management of ongoing nutritional stress not only throughout winter hibernation, but also in the early stages of refeeding, as the ground squirrels recover from months of phase 2 starvation (14). We recovered three isoforms of HMGCS2 with differing acetylation status; both acetylated and deacetylated isoforms were elevated in heterothermy (Supplemental Table S3, Fig. 7A). The enzymatic activity of HMGCS2 is inhibited by acetylation. The apparently deacetylated isoform that oscillates in torpor-arousal cycles with an abundance peak in Ar is consistent with increased ketogenesis in support of the arousal process (spot 1099, Supplemental Table S2). We also recovered multiple acetylated isoforms of carbamoyl-phosphate synthase. While the acetylation signal from CPS1 strongly increased during winter, its overall abundance decreased (Fig. 7), indicating both genetic and PTM control of this critical urea cycle enzyme. Acetylation blocks activity (37) of the already seasonally reduced levels of CPS1, facilitating amino acid sparing during torpor yet providing a means for enhanced removal of toxic nitrogen in IBA when acetylation is lowest.

SIRT3 is the major mitochondrial deacetylase in mammals (24, 38). As a sensor of cellular energy status, SIRT3 is activated under low energy conditions such as elevated NAD⁺, with downstream effects designed to limit further cellular energy depletion and favor energy production particularly by beta-oxidation. SIRT3 abundance increases during fasting and caloric restriction in mouse liver (19, 23), suggesting a possible mechanistic link between nutrient status and protein acetylation in hibernation. Instead, SIRT3 protein content in 13-lined ground squirrel livers decreased during winter compared with the active, fed animals. While this appears to be contrary to findings in other mammals, SIRT3 has not been measured after several months of fasting in any other system. The decreased SIRT3 abundance during hibernation is consistent with the observed seasonal increase in overall liver protein acetylation (Fig. 7). Limiting SIRT3 may provide tighter control over substrate deacetylation reactions and hence metabolism to support the intense metabolic depression and then reactivation orchestrated by hibernators during each torpor-arousal cycle. SIRT3-mediated deacetylation of HMGCS2 is necessary for activation of ketogenesis (47), and the use of ketone bodies as metabolic fuel is a hallmark of hibernation physiology (1). Winter-depressed SIRT3 may also promote autophagy in liver during hibernation, as a SIRT3 knockout in mouse embryonic fibroblasts enhances autophagic flux (32).

In summary, these data advance our knowledge of the proteomic changes that support both seasonal and torpor-arousal cycles in the hibernator's liver. Our data indicate that the changes distinguishing active animals from winter hibernators are dominated by metabolic signatures of feeding and fasting. These changes are consistent with the hypothesis that nutrient deprivation forms a critical foundation for winter heterothermy. Moreover, these data suggest that substrate marshaling in preparation for caloric restriction in fall and the subsequent winter fast is a necessary precursor for surviving pronounced metabolic depression and low substrate availability during torpor. Finally, our data are consistent with the interpretation that autophagic recycling during the prolonged winter fast is tightly regulated during torpor-arousal cycles, likely permitting maximal delivery of the substrates required to support the intense metabolic reactivation during early arousal which is accompanied by delayed reperfusion and transient hypoxia.

GRANTS

This work was supported by National Institutes of Health (NIH) Grant R01HL-089049 to S. L. Martin and the Institutional Proteomics Mass Spectrometry Facility, which is supported in part by NIH Grants to the Colorado Clinical and Translational Science Institute (UL1-RR-025780) and the University of Colorado Cancer Center (P30-CA-046934).

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the author(s).

AUTHOR CONTRIBUTIONS

Author contributions: A.G.H., K.R.G., and S.L.M. conception and design of research; A.G.H., K.R.G., and L.E.E. performed experiments; A.G.H., K.R.G., L.E.E., A.K.-F., and S.L.M. analyzed data; A.G.H., K.R.G., L.E.E., A.K.-F., and S.L.M. interpreted results of experiments; A.G.H., K.R.G., and S.L.M. prepared figures; A.G.H., K.R.G., A.K.-F., and S.L.M. drafted manuscript; A.G.H., K.R.G., L.E.E., A.K.-F., and S.L.M. edited and revised manuscript; A.G.H., K.R.G., L.E.E., A.K.-F., and S.L.M. approved final version of manuscript.