The transcriptional regulator of the chaperone response HSF1 controls hepatic bioenergetics and protein homeostasis

Changes in global bioenergetics and metabolic state, particularly starvation or nutrient overload, place a high demand on energy sources and protein quality-control mechanisms of the cell and have profound effects on organismal health. Mitochondria are essential organelles for the preservation of cellular bioenergetics given their established role in energy generation via respiration, a process that requires coordinated interaction of protein subunits encoded by mitochondrial and nuclear DNA (Wallace, 2005; Liesa and Shirihai, 2013), synthesis of metabolites, and regulation of signaling pathways. Indeed, compromised mitochondrial function caused by disruption of energy production and metabolic signal integration as well as increased oxidative stress is one of the hallmarks of numerous age-related neurologic and metabolic disorders (Wallace, 2005; Koopman et al., 2012).

Protein quality-control mechanisms, including translation and protein folding, and metabolic pathways such as gluconeogenesis and lipid synthesis are the most energy-demanding processes in most organisms and present a large burden to ATP and NAD⁺ supplies. For example, in differentiating mammalian cells, translation and gluconeogenesis are estimated to account for 25–30% and 7–10% of total ATP consumption, respectively (Buttgereit and Brand, 1995; Rolfe and Brown, 1997). In general, the architectures of several signaling and metabolic pathways have been considered to be robust against variations in energy supply and protein levels, and the cell has fail-safe mechanisms to adjust its catalytic capacity to cope with this challenge. Under pathological increase of metabolic and protein synthesis demands, however, this tremendous energy cost, in turn, can compromise the mitochondrial integrity and functional capacity to maintain energy homeostasis and fulfill the biosynthetic and metabolic bioenergetic needs of the cell. The fundamental question is how cells sense energetic demands and regulate the bioenergetic networks to balance the contrasting anabolic and catabolic needs under physiological and stress/disease conditions. Although tissue-specific differences in bioenergetic efficiency (e.g., brown adipose tissues, muscle or β cells; Liesa and Shirihai, 2013) are reasonably understood, the means of complex regulatory networks by which tissue and nutrient-dependent intracellular NAD⁺ and ATP pools are monitored and their integration within the larger set of protein quality-control and metabolic reprogramming mechanisms remains poorly defined.

Here, we investigated hepatic homeostatic energy-sensing mechanisms adjusting the rate of anabolic metabolic pathways to intracellular energy state, thereby allowing the metabolic capacity to readily adapt to challenging nutritional conditions. We show that heat shock factor 1 (HSF1), a key regulator of the classical chaperone response to cope with the increased protein load and protein folding defects caused by genetic abnormalities especially in malignant cells (Dai et al., 2007; Akerfelt et al., 2010; Anckar and Sistonen, 2011; Jin et al., 2011), also has critical function as an information hub for anabolic metabolic and bioenergetics and protein synthesis pathways in the cell.

A causal link between metabolism and protein homeostasis driven by HSF1 with respect to adaptive responses of the cell to energy stress has been proposed (Hahn et al., 2004; Jin et al., 2011; Santagata et al., 2013; Su et al., 2016), but knowledge of the underlying mechanisms that integrate them is lacking. We have reported that global hsf1 deletion induces a cellular energy deficit associated with increased basal and insulin-induced AMP-activated protein kinase (AMPK) activation (Jin et al., 2011). This observation prompted us to investigate further the metabolic consequences on ATP and NAD⁺ systems, focusing on liver cell bioenergetics. In this study, we used mice with whole-body gene deletion (hsf1^−/−; Zhang et al., 2002). To investigate the function of HSF1 in liver metabolism, we also generated a hepatocyte-specific hsf1 knockout model (L-hsf1^−/− hereafter) by crossing albumin-Cre mice (Postic et al., 1999) with hsf1^f/f mice, which were functionally wild type (WT; Fig. S1). Surprisingly, we found that levels of ATP were not sustained in the livers of hsf1^−/− mice (Fig. 1 A). This ATP decline significantly increased the AMP level and [AMP/ATP] ratio, reflecting an increase in AMPK activity (Fig. 1 B). Because ATP production is coupled with NAD⁺ availability, we also examined whether hsf1 deletion affects the intracellular [NAD⁺/NADH] ratio. Indeed, HSF1 loss in mice and hepatic cells decreased the cellular NAD⁺ bioavailability with maximal effects on mitochondrial NAD⁺ levels, but it did not significantly affect NADH levels (Fig. 1, C and D). The change in NAD⁺ level translated to a decreased mitochondrial [NAD⁺/NADH] ratio, suggesting an organelle-specific HSF1-dependent effect on NAD⁺ biosynthesis and/or consumption. Notably, the effects on cellular energy charge ([ATP], [AMP/ATP], [NAD⁺], and [NAD⁺/NADH]) were particularly evident in fasted livers but also detected in primary hepatocytes.

NAD⁺ decline could be caused by an increase in NAD-consuming enzymes, such as poly(ADP-ribose) polymerases (PARPs), CD38, sirtuins, and/or a decrease in NAD-synthesizing enzymes (Dölle et al., 2013; Cantó et al., 2015). In the absence of HSF1, global PARP activity (Adamietz, 1987), and PARP-1 or CD38 contents were not significantly affected (Fig. S2, A and B), and the activities of sirtuins were decreased (see Fig. 5). These data do not support a causal role for enhanced PARP-, CD38-, or sirtuin-mediated NAD⁺ breakdown in NAD⁺ decline but rather suggest that this could be mediated by alterations in NAD⁺ biosynthesis.

In mammalian cells, NAD⁺ synthesis occurs from diverse dietary sources, including tryptophan (TRP), nicotinic acid (NA), and nicotinamide riboside (NR; Cantó et al., 2015; Fig. 2 A). In addition, NAD⁺ can be resynthesized through the NAD⁺-salvage pathway that involves the conversion of nicotinamide (NAM) to nicotinamide mononucleotide (NMN) by the rate-limiting enzyme nicotinamide phosphoribosyltransferase (NAMPT) and then into NAD⁺ by NMNAT1 or NMNAT3. Interestingly, the basal and nutrient-induced mRNA and the protein level of NAMPT were significantly decreased in livers of hsf1^−/− mice (Fig. 2 B). The subcellular levels of NAMPT were also markedly reduced (∼40%) in fed hsf1^−/− mice compared with control mice, although the mitochondria-associated NAMPT level appeared to be very low (Fig. 2 C). We validated NAMPT suppression by comparing mRNA and protein levels after treatment of hepatocytes with stress agents, including heat shock or decreasing concentration of glucose (Fig. 2, D–E), which are known to induce NAMPT expression in yeast and mammalian cells (Anderson et al., 2003; Yang et al., 2007). In addition, we tested whether HSF1 directly controls Nampt promoter activity and expression. Promoter analysis of mouse Nampt identified multiple repeats exhibiting the characteristic core motif 5′-nTTCn-3′ described as noncanonical TTC-rich binding sites for HSF1 (Guertin and Lis, 2010). Chromatin immunoprecipitation (ChIP) quantitative PCR assays confirmed specific increased HSF1 occupancy at a potential binding site at nucleotide −1,209 to −965 bp of the Nampt promoter in primary hepatocytes exposed to decreasing glucose concentrations (Fig. 2 F). Because nutrient levels influence transcriptional programs driven by HSF1, we also investigated HSF1 occupancy on the Nampt and Hsp70 promoters in livers during fed, fasting, or refeeding conditions (Fig. 2 G). Interestingly, HSF1 binding to the Nampt promoter was uniform at the fed or fasted state but significantly declined with refeeding; in contrast HSF1 binding to the Hsp70 promoter was strongly increased under the refeeding condition. In keeping with previous studies, the nuclear accumulation of the hyperphosphorylated (active) form of HSF1 fluctuated in a nutrient-dependent manner (Fig. 2 H). Consistent with a direct role of HSF1 in NAD⁺ salvage, we also determined that NAM treatment in vitro and in vivo could not reverse the NAD⁺ and ATP-decline or restore the diminished glucose production capacity of hepatocytes caused by HSF1 loss (Fig. 2, I and J). However overexpression of HSF1 in primary hsf1^−/− hepatocytes completely restored the nutrient-induced mRNA expression of Nampt gene (Fig. 2 K), demonstrating an HSF1-dependent cell-intrinsic effect. Furthermore, the levels of fed or fasting-induced mRNAs encoding Nrk1 and Nrk2, Nmnat1, Nmnat3, and Nadsyn1 did not significantly differ between genotypes (Fig. S2 D). Finally, to assess whether HSF1 loss may affect redox pathways regulating the cellular NAD⁺ pool, we profiled transcriptional levels of selected NAD⁺-dependent enzymes. The mRNA levels of several dehydrogenases that regulate the NAD⁺/NADH or NADP⁺/NADPH redox balance, however, were either not significantly affected or decreased (Fig. S2 C). Thus, these data suggest that HSF1-mediated transcriptional regulation of Nampt critically impacts the intracellular NAD⁺ and ATP pools.

Because mitochondrial biogenesis and/or electron transport chain (ETC) function have significant impact on cellular ATP production and energy status, the overall effects of HSF1 depletion on ATP and NAD⁺ contents raised two nonexclusive possibilities: first, the mitochondrial dynamics and biogenesis may be affected; and second, the reduced ATP production may be caused through a defect in mitochondrial ETC function.

We first investigated the potential effects on mitochondrial morphology and biogenesis in primary hepatocytes and livers. In addition to ATP decline, lack of HSF1 in mouse hepatocytes or HepG2 cells with hsf1 silencing by shRNA significantly reduced the amount of mitochondrial proteins (normalized to total protein level), mass, and DNA content, indicating a general reduction in mitochondria biogenesis (Fig. 3, A–F). The mRNAs of both mitochondrial- and nuclear-encoded genes that are functionally involved in mitochondrial biogenesis (Pgc1α, Pgc1β, Nrf1, Nrf2, Pparα, and ERRα) and OXPHOS (Nd1, Cox2, Co5b, Atp5j, and Atp6) were present at lower levels (Fig. 3 G). Interestingly, however, changes in mRNA transcripts were seen only under fasting, which highlights a stronger contribution of HSF1 activity to cellular energy status maintenance under nutrient-limiting conditions. EM of liver tissues revealed that mitochondrial density and coverage were lower in hsf1^−/− mice than WT mice in both the fed and fasted states (Fig. 3, H and I). In the fed state, these morphological changes in hsf1^−/− mice were accompanied with a significant increase in the mean mitochondrial size (Fig. 3 J) but a decrease in aspect ratio (AR), indicating a more rounded shape of mitochondria (Fig. 3 K). In the fasted state, liver cells from hsf1^−/− mice did not present additional significant changes in relative mitochondrial parameters other than a further increase in mitochondrial size compared with the fed condition without AR changes (Fig. 3, J and K). The morphological appearance of the enlarged mitochondria in livers of hsf1^−/− mice (Fig. 3 L) is thus indicative of changes in mitochondrial fusion–fission dynamics with a trend for enhanced fusion. Because the mRNA level of genes regulating mitochondrial morphology (Drp1, Mfn1, Mfn2, and Fis1) and mitophagy (Pink1 and Bnip3) did not significantly differ between genotypes (Fig. S3 A), it is unlikely that HSF1 is solely a direct and potent mRNA transcript regulator of key mitochondrial morphogenesis genes. Validation studies confirmed that levels of proteins implicated in macroautophagy (LC3-II, BECLIN 1, p62, and BAX) and mitochondrial fusion (OPA1 and MFN1) were not affected (Fig. S3 B). However, we noted a significantly higher MFN2 level after prolonged fasting in hsf1^−/− mice, which may be critical to maintain mitochondrial fusion. Strikingly, hsf1^−/− mice exhibited a lower p-DRP1/DRP1 ratio under nutrient-available conditions, indicating impaired mitochondrial fission (Fig. S3 B). Tetramethylrhodamine ethyl ester (TMRE) staining indicated that hsf1 deletion increased the mitochondria potential and conferred mitochondria resistant to carbonyl cyanide m-chlorophenyl hydrazone–induced mitochondrial membrane depolarization and cell death. Unexpectedly, we found that HSF1 loss moderately decreased cytoplasmic and mitochondrial reactive oxygen species levels, based on dihydroethidium and MitoSOX staining of primary hepatocytes (unpublished data). This supports the notion that HSF1 loss engages mitochondria fusion to preserve the mitochondrial integrity. Thus, it is conceivable that the altered mitochondrial dynamics in hsf1^−/− mice represents an adaptive (compensatory) response to reduced NAD⁺ and ATP production.

We next assessed the impact on mitochondrial metabolic function under cellular stress. We therefore measured the oxygen consumption rate (OCR), an indicator of mitochondrial OXPHOS, and the extracellular acidification rate (ECAR), an indicator of lactate production and glycolysis. Remarkably, hsf1^−/− primary hepatocytes presented a low OCR normalized to cell number. Indeed, HSF1 loss markedly decreased both basal and ATP-coupled OCRs and maximal respiratory capacities (Fig. 4 A). However, the overall OCR profile was comparable in both genotypes, when normalized to mitochondrial protein or DNA levels (unpublished data). HSF1 loss also attenuated the OCRs induced by glucose, pyruvate, and lactate, or palmitate, when normalized to cell number (Fig. 4, B–D). In contrast, hsf1^−/− cells presented an ECAR slightly above WT levels (Fig. 4 E). To directly characterize mitochondrial function, we also examined the respiratory capacity of isolated mitochondria. In the absence of HSF1, the overall mitochondrial respiration capacity was barely affected (Fig. 4 F). Mitochondrial complex I (MC-I) activity was an exception to this pattern, as it was slightly but reproducibly reduced in hsf1^−/− mitochondria. We confirmed this result using a modified assay for measuring MC-I enzymatic activity (Spinazzi et al., 2012; Fig. 4 G). Blue native (BN) PAGE and Western blot analyses of liver mitochondria lysates for OXPHOS or representative proteins of respiratory chain complexes (C-I to C-V) showed no apparent changes, indicating that except for MC-I activity, the respiratory complex superassemblies were not significantly affected upon HSF1 loss (Fig. 4, H and I). In summary, these results suggest that the impaired respiration and ATP production capacity in hsf1^−/− cells is likely to be caused by decreased mitochondrial density and, to a lesser extent, by a modest decrease in MC-I–supported respiration as a result of changes in NAD⁺/NADH ratio and altered redox state rather than by dysfunctional respiratory complex organization.

To examine whether HSF1 loss might also impair ATP production and OCR through mechanisms affecting nutrient flux and utilization, we measured the steady-state levels of metabolites involved in major bioenergetic pathways. Targeted gas chromatography–mass spectrometry analyses of livers from fed or fasted hsf1^−/− mice compared with WT controls did not reveal an overt effect in metabolite levels associated with glycolysis, tricarboxylic acid cycle, and lipid metabolism (unpublished data). The potential influence of conversion of pyruvate to lactate or the reverse, which is catalyzed by LDH enzymes and regenerates or consumes NAD⁺, is also not supported by our data (unpublished data). Thus, although hsf1^−/− livers displayed a trend to increased [lactate]/[pyruvate] ratio, cellular lactate concentration did not significantly differ between the genotypes, suggesting that the NAD⁺ content deficit is unlikely to be primarily mediated by changes in this redox status. We further conducted stable isotope tracer analysis with [U-¹³C]-glucose to assess whether glucose metabolism was altered in hepatocytes. Comparable concentrations of ¹³C-enrichment (fractional abundance of M+2 isotopomers) were observed for the major tricarboxylic acid cycle metabolites between the genotypes (Fig. 4 J). This suggests that HSF1 ablation did not significantly impair the flux of glucose through the tricarboxylic acid cycle and glycolysis in primary hepatocytes.

Silencing HSF1 in hepatocytes did not impair ETC formation and respiratory chain capacity as assessed at the mitochondrial protein level. We propose, therefore, a model whereby the effects of HSF1 loss on mitochondrial content can be linked to sirtuin loss of function that leads to increased levels of acetylation of proteins related to organization of the mitochondrial DNA (mtDNA) genome. Among the mammalian sirtuins, nuclear sirtuin 1 (SIRT1) and mitochondrial SIRT3 have robust deacetylase activity. Activated by a rise in NAD⁺ level and changes in the NAD⁺/NADH ratio, SIRT1 and SIRT3 play a critical role in promoting metabolic adaptations to energy stress by enhancing mitochondrial biogenesis and OXPHOS (Cantó et al., 2015). In light of the established role of PGC-1 isoforms in mitochondrial function and a recent study on the potential role of HSF1 as a transcriptional regulator of PGC-1α in fat and muscle (Ma et al., 2015), we investigated the effects of hsf1 deletion on hepatic PGC-1α/β expression. hsf1 deletion significantly reduced the fasting-induced mRNA level of Pgc-1α relative to control, although it only had a transient (confined to a 24-h fasting time point) and moderate effect on PGC1α protein levels. The mRNA and protein levels of Pgc-1β were not affected (Figs. 3 G and 5 A). In contrast, hsf1^−/− mice displayed increased protein acetylation of PGC-1α, a known target of SIRT1 (Fig. 5 B). Similar to murine liver, knockdown of HSF1 increased PGC-1α acetylation in human cell lines (HepG2 and 293T; Fig. 5, C and D). Consistent with the finding that PGC-1α activity is inhibited by acetylation, the mRNA levels of PGC-1α target genes (e.g., Nrf1, Nrf2, and Cox5) in response to nutrient deprivation were reduced in hsf1^−/− mice (Fig. 3 G). As a specific readout of SIRT3 activity loss, we profiled acetylated mitochondrial proteins from hsf1^−/− and WT liver targets. Acetylation of NDUFS3, MT-CO1, SOD2, and TFAM was increased in hsf1^−/− livers, whereas NDUFA9, UQCRC2, and ATP5A acetylation was unchanged (Fig. 5 E). These data provide evidence linking HSF1 loss to broad-scale changes in acetylation status of SIRT1 and SIRT3 nuclear and mitochondrial target proteins.

Conceivably, the relative effect of hsf1 deletion on mitochondrial content might be explained by changes in protein acetylation status accompanied by attenuation of the NAD⁺-SIRT1-PGC-1α axis. However, PGC-1α/β seems to be dispensable for mitochondrial density and mtDNA content (Rowe et al., 2013). Interestingly, recent biochemical analyses have provided corroborating evidence that changes in sirtuin activity influence the acetylation and activity of proteins in diverse metabolic pathways, including those regulating mitochondrial content integrity and genome maintenance (Hebert et al., 2013). We first profiled the mRNA expression of genes associated with the basal transcription machinery of mitochondria. Fasting-induced mRNA levels of genes involved in mtDNA replication were reduced in hsf1^−/− livers (Fig. S3 C), which is consistent with the observed marked decreased in mtDNA content. We next explored, biochemically, whether the specific mitochondrial adaptive pathways are modified by hsf1 deletion. Sirtuin-dependent mitochondrial changes associated with decreased mtDNA content are linked to induction of the mitochondrial unfolded protein response (UPR^mit), which is triggered by an imbalance in mitochondrial- versus nuclear-encoded mitochondrial proteins (Mouchiroud et al., 2013). However, no differences in mitonuclear protein ratio (SDHA/MT-CO1; Fig. S3D), mRNA expression of UPR^mit (Clpp, Hspe1, or Hspd1; Fig. S3E), or UPR^ER (GRP78, XPB1, and ATF4) protein biomarkers (Fig. S3 F) were detected between the genotypes.

Finally, we sought to alter the protein acetylation status by manipulating the cellular NAD⁺ level in vivo. Consistent with the notion that the activity of sirtuins (including SIRT1 and SIRT3) is inhibited by a lower NAD⁺ level or NAD⁺/NADH ratio, we found that nuclear and mitochondrial extracts from livers of hsf1^−/− mice exhibited increased protein acetylation (Fig. 5, F and G). Interestingly, the increased protein acetylation status of mitochondria isolated from hsf1^−/− livers could be reversed by treatment of mice with NR, which increases the NAD⁺ level and normalizes the NAD⁺/NADH ratio (Fig. 5 G). Importantly, the NR supplementation in hsf1^−/− mice increased mitochondrial mass and DNA abundance in hepatic cells (unpublished data). Collectively, these findings established links among HSF1, the NAD⁺/NADH ratio, and protein acetylation status largely regulated by SIRT1 and SIRT3, connecting mitochondrial biogenesis to respiratory function and ATP production.

The idea that cells must successfully balance metabolic demands with energetic requirements to maintain viability prompted us to investigate potential effects of hsf1 deletion on nutrient and energy-regulated anabolic pathways, including gluconeogenesis, protein translation, and lipid metabolism.

Hepatic glucose production is central to metabolic adaptation during fasting and occurs most efficiently with substrates generated by catabolism of glycogen, glycerol, or lactate/pyruvate/amino acids. In this process, mitochondria play a pivotal role through provision of carbon substrates, reducing equivalents, and ATP. Consistent with our previous study (Jin et al., 2011), we found that hsf1^−/− or L-hsf1^−/− mice exhibited a markedly improved glucose tolerance and displayed lower blood glucose levels after pyruvate or glucagon exposure (Fig. 6, A and B). Basal blood glucose levels during the fed or fasting states were reproducibly lower in both hsf1^−/− and L-hsf1^−/− mice than in WT controls (Fig. 6, C and D). These data suggest that HSF1 is required for proper regulation of glucose signals. Indeed, insulin stimulation of p-AKT, which reflects its activation, was significantly enhanced in the absence of HSF1 (Fig. 6 E). Increased AKT signaling by insulin was also evident in primary L-hsf1^−/− and hsf1^−/− hepatocytes (unpublished data). Next, we determined whether HSF1 ablation exerts its effects by inhibiting hepatic gluconeogenesis mediated by glucagon signaling (Miller et al., 2013). Binding of glucagon to its receptor on the hepatocyte plasma membrane leads to activation of adenylyl cyclase, production of cAMP, and stimulation of protein kinase A (PKA), which phosphorylates protein targets that work in concert to increase hepatic glucose output (Oh et al., 2013). Consistent with an effect on gluconeogenesis, primary hsf1^−/− hepatocytes produced significantly less glucose when stimulated with lactate and pyruvate, glucagon, or forskolin (FSK), but this inhibitory effect is reversed by an analogue of cAMP (Bt2-cAMP; Fig. 6 F). Interestingly, HSF1 loss did not affect glucagon receptor expression, but caused significant reductions in the glucagon- or FSK-stimulated increase in cAMP levels (Fig. 6, G and H). The decreased glucose production capacity paralleled attenuation of glucagon or FSK-induced signaling that was associated with reduced phosphorylation of PKA substrates; this inhibition was bypassed by Bt2-cAMP (Fig. 6 I). Specifically, hsf1 deletion reduced expression levels of glucagon or FSK-induced p-CREB or nuclear active dephosphorylated CRTC2 (TORC2), which are PKA-regulated proteins, as well as reduced mRNA levels of key gluconeogenic genes (Fig. 6, J and K); as expected Bt2-cAMP-dependent signaling induction was not affected (Fig. 6 L). To further dissect the mechanism of blunted glucagon action, glucagon-stimulated signals were assessed. Fasting or injection of glucagon stimulated hepatic p-CREB in WT mice, but this expression was significantly reduced in hsf1^−/− and L-hsf1^−/− livers (Fig. S4, A–F). In line with a previous study (Jin et al., 2011), hsf1^−/− mice also displayed increased AKT-FOXO1 signaling not only upon feeding, which stimulates insulin production, but also upon fasting when the insulin level is low. Interestingly, the combined significant increase in acetylated-FOXO1 and p-FOXO1 levels seen in the livers of hsf1^−/− mice upon fasting caused a more pronounced hepatic nuclear exclusion and inhibition of FOXO1 activity during gluconeogenesis (Fig. S4 G), which is consistent with a recent study (Qiang et al., 2010).

Remarkably, in spite of the increased nutrient-stimulated AKT-mediated signals caused by HSF1 ablation, a marked decrease in the activities of mTOR and its downstream target proteins (p70S6K, S6R, and 4E-BP), which stimulate mitochondrial biogenesis and protein translation (Morita et al., 2013), was evident (Fig. 7 A). This effect, which can be explained by the concomitant ATP decline combined with a powerful increase in hepatic AMPK activity, caused a significant decrease in global translation relative to control as measured by ³H-Leu labeling in primary hepatocytes or polysome profiling performed on liver extracts from fed or fasted hsf1^−/− mice (Fig. 7, B and C). These in vitro and in vivo findings further support the conclusion that HSF1 activity and hepatic metabolic activity may be integrated with nutritional state and protein translation.

In cells, most metabolic pathways contain multiple levels of feedback control to maintain cellular ATP. Fatty acid oxidation (FAO) intertwined with the tricarboxylic acid cycle and ETC activity is a major supplier of ATP during fasting. This prompted examination as to whether the decreased hepatic ATP production in the absence of HSF1 is linked to changes in hepatic FAO or lipid synthesis. We observed no changes in basal FAO during the fed state but a slightly lower FAO level during the fasting state in the hsf1^−/− livers (Fig. 7 D). However, liver cells from hsf1^−/− mice exhibited significantly reduced lipid production (Fig. 7 E).

Collectively, these data confirmed that the hepatic energy deficit caused by HSF1 loss triggers compensatory pathways to restore intracellular energy levels through inhibition of ATP-consuming biosynthetic pathways. Moreover, our data suggest that these effects are likely due, at least in part, to the effects of AMPK activation and lower ATP levels.

Several pharmacological and genetic approaches have been explored to increase NAD⁺ bioavailability and enhance sirtuin activity as a strategy to ameliorate and protect organisms from metabolic diseases induced by excessive nutrient consumption, aging, or genetic disorders (Cantó et al., 2015). The ability of NAD precursors to increase NAD⁺ levels in vivo and in vitro (Yoshino et al., 2011; Cantó et al., 2012) prompted us to test whether the reduced NAD⁺ levels in hsf1^−/− hepatic cells can be reversed by NAD⁺ precursors. Indeed, supplementation with NA, NR, or TRP raised the NAD⁺ and ATP levels and restored glucose production capacity in hsf1^−/− hepatocytes (Fig. 8, A–C). Likewise, NMN increased NAD⁺ and ATP levels and improved impaired glucagon-induced mRNA levels of gluconeogenic genes and glucose output (Fig. 8, D–G). In contrast, WT hepatocytes treated with NMN displayed a slight but reproducible decrease in ATP and glucose production levels as well as a marked repression of gluconeogenic genes (Fig. 8, E–G). Consistent with these findings, we found that NMN ameliorated the metabolic dysfunction of mitochondria in hsf1^−/− hepatocytes, increasing the basal, ATP-linked, and maximal OCRs (Fig. 8 H). Importantly, the effects on NAD⁺, ATP, glucose, and OCR values could effectively be reversed by overexpression of HSF1 (Fig. 8, I–L). These observations suggest that the energetic defects caused by HSF1 loss sensitize livers and possibly other tissues to NAD precursors and activators of mitochondrial bioenergetics.

Based on these findings, we predicted that the metabolic defects in HSF1-mediated NAD⁺ biosynthesis could be ameliorated by administration of NR in mice. We observed enhanced NAD⁺ levels at both the fed and fasting states in NR-treated mice (Fig. 9 A). NR treatment also improved the mtDNA content in hsf1^−/− mice (unpublished data). In concordance with previous studies (Yoshino et al., 2011; Cantó et al., 2012; Trammell et al., 2016), NR-treated WT mice displayed a lower blood glucose level in both the fed and fasted states (Fig. 9 B). Conversely, by comparing NR with vehicle-treated hsf1^−/− mice, we found that NR slightly increased glucose levels in the fed state but considerably increased glucose levels in the fasting (Fig. 9 B). Consistent with this result, NR-treated WT mice displayed lower blood glucose concentrations in pyruvate tolerance test (PTT), confirming significantly lower gluconeogenic conversion of pyruvate to glucose (Fig. 9 C). In contrast, NR treatment of hsf1^−/− mice manifested an increase in PTT, confirming restoration of hepatic gluconeogenesis. The contrasting results of NR treatment on glucose production in WT versus hsf1^−/− mice prompt speculation that the effects of NAD-precursor supplementation might stem from its ability to cause differential changes in NAD⁺/NADH ratios that would affect the ATP pools. One would also suspect that the effects of acute NR administration, particularly those on the physiological function of nuclear and mitochondrial sirtuins and their multiple targets, might not be the same between genotypes. Therefore, the detailed molecular mechanisms responsible for this effect need to be investigated further.

A common notion holds that the level of ATP is universally homeostatic and, therefore, it is unlikely to be the growth-limiting resource for most cell types (Hochachka and McClelland, 1997; Lunt and Vander Heiden, 2011). Although this may be the case in steady-state conditions, it may not be sufficient for extensive energy-demanding processes. For instance, in several cancer models, the NAD⁺/NADH and NADP⁺/NADPH redox function coupled to ATP production levels plays a significant role in dictating survival or death of the tumor cells (Zong et al., 2016). Considering the strong evidence directly linking NAD⁺/NADH redox imbalance and ATP insufficiency to diverse human pathologies (Verdin, 2015), a key challenge in deciphering metabolic adaptation is the identification of the regulators that maintain cellular energetic efficiency. HSF1 is a central transcriptional regulator of eukaryotic protein quality control, and evidence suggests that it plays a role as a metabolic sensor. Although HSF1 exerts an impact on a large array of cellular functions that are altered during nutrient fluctuation, particularly in malignancy, including translation, ribosome biogenesis, and glucose metabolism (Hahn et al., 2004), the specific mechanisms by which HSF1 may preserve optimal cell function, especially under challenging nutrient and energy-deprived conditions, are not well understood. Here, we validated a mechanism whereby HSF1 monitors the metabolic and bioenergetics and protein synthesis pathways in the cell, acting as a pivotal determinant of key NAD⁺- and ATP-dependent metabolic hepatic functions during different nutrient states. Although diminished redox function and mitochondrial ATP production capacity are considered sources of impaired cellular function and pathology, surprisingly, HSF1 loss increases cellular fitness and preserves cell viability by triggering an adaptive compensatory response to balance the energy deficit, a response that stems from attenuation of nutrient-regulated anabolic metabolic processes in the liver.

In principle, changes in cellular NAD⁺ content after depletion of HSF1 could stem from perturbations in cellular reductive capacity (NAD⁺/NADH ratio), the rate of NAD⁺ synthesis, and NAD⁺ consumption. Although our findings support the direct HSF1-mediated control of stress-induced NAMPT expression, perturbation in NAD⁺ recycling from NADH, a process primarily dependent on MC-I of ETC activity, potentially may contribute to a greater effect on cellular NAD⁺ deficit. The data also provide strong evidence linking perturbations in the HSF1–NAD⁺ axis to changes in sirtuin deacetylase activity, global protein acetylation, and hepatic ATP regenerative capacity (Fig. S5). An important question deserving of further investigation is how a widespread upward shift in protein acetylation impacts ATP production capacity, especially considering the contrasting effects of hyperacetylation on protein function (Nogueiras et al., 2012). Several studies linking heightened protein acetylation to metabolic dysregulation posit that at least some of these acetyl modifications alter the activity, mostly in a negative direction, of proteins required to maintain mitochondrial integrity (Hebert et al., 2013). Fitting with this prediction, the effects of HSF1 loss on mitochondrial content could simply be explained by loss-of-function acetylation of proteins that regulate mitochondrial biogenesis, including PGC-1α or TFAM, proteins playing a role in maintenance, expression, and organization of the mitochondrial DNA genome. Although the findings do not prove causation, the aforementioned conclusion is supported by restoration of the NAD⁺/NADH ratio by NAD precursors reducing protein acetylation and correcting the hsf1^−/− metabolic phenotype. Also noteworthy is that although NAD⁺ is a main substrate for class III (sirtuin) deacetylase activity, changes in the NAD⁺/NADH ratio have also been shown to modulate sirtuin activity (Karamanlidis et al., 2013). Competitive inhibition of sirtuin activity by a lower NAD⁺/NADH ratio thus might further stimulate protein acetylation.

Mitochondrial dynamics (fusion and fission) hold a prominent position in mtDNA stability, respiratory capacity, response to nutrient availability, and mitophagy. For example, defective mitophagy caused by excessive cleavage of PINK1 represents a feature of DNA repair disorders (Fang et al., 2014). Thus, in theory, the mitochondrial phenotype of hsf1^−/− mice was predicted to involve increased mitophagic elimination of altered mitochondria. However, our findings of intact autophagy and inability of mitochondria to engage fission in livers of hsf1^−/− mice did not support this notion. As HSF1 has no discernible adverse effects on cell viability, these data predict that the energetic performance of stressed mitochondria in hsf1^−/− mice will be retained through fusion reactions that maximize the energy output rather undergo mitophagy.

Based on this information, we propose a model (Fig. S5) whereby loss of HSF1 triggers an energy-saving adaptive response associated with inhibition of nutrient-driven hepatic anabolic metabolic processes. Preservation of fasting energetics through attenuation of gluconeogenesis, an energetically costly anabolic process requiring six ATP equivalents per molecule of glucose synthesized, is likely to be an important mechanism to counteract systemic hyperglycemia, for example, in diabetic patients. This process is regulated by a set of well-described signaling pathways, including the glucagon–PKA–CREB and CRTC2 axis, and FOXO1 priming is also important for maintaining fasting blood glucose levels. Mechanistically, suppression of glucose production by HSF1 loss may involve several NAD- and ATP-dependent regulatory points. First, consistent with a previous study (Miller et al., 2013), hsf1 deletion reduces glucose output by increasing the [AMP]/[ATP] ratio to attenuate cAMP–PKA and glucagon-induced signaling. Second, the decrease in ATP levels caused by HSF1 loss can predominately impact the rate of hepatic glucose production by attenuating the activity of ATP-driven gluconeogenic enzymes (Foretz et al., 2010). Another possibility supported by our data is the activation of AMPK through cellular energy status changes that inhibit gluconeogenic gene expression. Finally, the predicted effect of hsf1 deletion on loss-of-function acetylation of PGC-1α and FOXO1 can contribute to a greater suppression of glucose output (Qiang et al., 2010).

Arguably, changes in ATP supply have been linked to numerous disease conditions, particularly those that influence translation and protein folding. In addition to the high-energy–demanding systems for protein synthesis and folding, the cell needs to tightly adjust synthesis and usage to ATP levels to survive, especially under energy-limiting conditions. Our finding that HSF1 ablation provokes a systemic cellular response to adapt mTORC1-mediated protein translation to energetic stress supports a model by which mTORC1 activity is controlled directly by ATP levels (Dennis et al., 2001) or indirectly by AMPK activation (Hardie et al., 2012). However, perturbations in ATP production might have a direct negative impact on energy-sensitive steps of the protein synthesis machinery, including charging aminoacyl-tRNAs and recycling nucleotides for mRNA synthesis (Jewett et al., 2009). Another feature associated with loss of HSF1 is JNK activation and suppression of mTORC1-mediated translation, a process proposed to ensure optimal cellular survival under proteostatic stress (Su et al., 2016). However, HSF1 might not regulate mTORC1 signals predominantly through JNK in our model, because our data did not reveal a profound impact of HSF1 loss on nutrient-induced p-JNK levels (unpublished data). The intriguing possibility that the inhibition of mTORC1 activity represents a bona fide regulatory mechanism for HSF1 to optimize bioenergetic functions by selective translation of metabolic-related mRNAs (Morita et al., 2013) deserves further investigation.

In the current investigation, analysis of fatty acid metabolism revealed a marked attenuation of lipid synthesis but a slight decrease in FAO in the livers of hsf1^−/− mice. Because diminished sirtuin activity causes alterations in lipid homeostasis associated with reduced FAO and increased hepatic lipid content, our contrasting findings suggest that additional mechanisms may exist in the adaptive energy strategy associated with HSF1 ablation to balance ATP demand with ATP supply. For example, AMPK-mediated reprogramming of lipid metabolism could act to convey the energy-saving signal to an enzymatic reaction regulating this response. A diminished intracellular ATP level might also be an inhibitory signal for ATP-requiring steps of lipid synthesis.

In summary, the findings suggest coordination of energy-consuming and producing processes for glucose, protein, and lipid metabolism that underpin important biological effects of HSF1-mediated energy control programs. The results prompt speculation that targeting HSF1 activity may offer a strategy for the treatment of metabolic diseases such as diabetes as well as interfere with cancer progression without compromising normal cell viability and function.

Hsf1^−/− mice were previously described (Zhang et al., 2002). Both hsf1^−/− and hsf1^+/+ control littermates were of mixed C57BL/6-129SvPas F1 genetic background. Hsf1^f/f and hepatocyte-specific hsf1-deletion mice on a C57BL/6J genetic background for at least nine generations are described in the following paragraph. C57BL/6J-Alb-cre mice were purchased from The Jackson Laboratory. Mice were fed ad libitum and kept in 12-h light and dark cycles. Unless otherwise indicated, mice at 2–3 mo of age were used. Animal care and experiments were performed in accordance with the guidelines of the Institutional Animal Care and Use Committee and National Institutes of Health guidelines.

To generate mice conditionally expressing hsf1 (hsf1^f/f), a targeting vector was designed in which the neomycin gene with one loxP site at the distal end was inserted into intron 3 of the hsf1 allele. The proximal loxP site was inserted into intron 1. Expression of cre recombinase deletes the targeted exons 2 and 3 and generates a defective gene allele, leading to complete disruption of HSF1 protein expression. For the construction of the targeting vector, a 4.2-kb proximal fragment, an 811-bp floxed fragment, and a 3.2-kb distal fragment were amplified by PCR from a bacterial artificial chromosome clone (PR23-266H9; BACPAC Resources Center, Oakland Research Institute) containing the hsf1 gene. The neomycin gene was flanked by two Frt sites, allowing the removal of the drug selection marker from the targeted locus through a flippase-mediated recombination event. The targeting DNA construct, introduced into the XhoI site of the phage DNA vector λDASHII-254-2TK, was flanked by two thymidine kinase genes for drug negative selection (Zhang et al., 2002). The final vector was linearized with NotI, sequenced, and transfected into 129Sv/Ev embryonic stem cells. Embryonic stem cell clones with the floxed hsf1-targeted locus verified by Southern blot analysis were microinjected into blastocysts, and chimeric mice were crossed with C57BL/6J mice to generate hsf1^f/w mice. To remove the neomycin cassette, hsf1^f/w mice were crossed with an Flp-deleter strain (The Jackson Laboratory).

For hepatocyte-specific disruption of HSF1, hsf1^f/f mice on a C57BL/6J genetic background (at least four generations) were crossed with alb-cre mice (Postic et al., 1999; The Jackson Laboratory) to generate hsf1^f/f-Alb-Cre⁺ (designated as L-hsf1^−/− mice) or hsf1^f/f (WT) littermates, which were used as controls.

Tissues that were snap frozen in liquid N₂ were homogenized in RIPA buffer containing protease and phosphatase inhibitors cocktail (Roche and Thermo Fisher Scientific), protein extracted, and then subjected to Western blotting. Equal amounts (30 µg) of protein were loaded onto minigels (Bio-Rad Laboratories) and transferred onto a polyvinylidene fluoride membrane using established protocols. Samples were probed for the indicated proteins as described previously (Jin et al., 2011) using the antibodies listed in the following section.

The following primary antibodies were used in this study: mouse HSP70 (mAb; SMC-100A/B; StressMarq Biosciences Inc.); rabbit CREB2 (sc-200), rabbit P62 (sc-25523), rabbit Ac-FRKH (FOXO1; sc-49437), goat ATF4 (sc-7583), rabbit XBP1 (sc-7160), rabbit PGC1α (sc-13067), rabbit PGC1β (sc-67286), goat ATP5α (C15; sc-49162), rabbit UQCRC2 (sc-292924), goat CD38 (sc-7049), mouse OPA1 (mAb; sc-393296), rabbit NAMPT (sc-67020), mouse (mAb) α-tubulin (sc-5286), mouse (mAb) β-actin (sc-47778), and goat γ-tubulin (sc-7396; Santa Cruz Biotechnology, Inc.); mouse (mAb) NDUFA9 (ab147130) and mouse (mAb) MTCO1 (ab14705; Abcam); rabbit β-tubulin (2376-1; Epitomics and Abcam); mouse (mAb) PARylation (AG-20T-0001; AdipoGen); rabbit CRTC2 (ST1099; EMD Millipore); rabbit AMPK (2603), rabbit phospho-AMPK(Thr172) (2535), rabbit SDHA (5839), rabbit HSP60 (12165), rabbit Ac-K (acetylated antibody; 9441), rabbit HSF1 (4356), rabbit FOXO1 (2880), rabbit p-FOXO1 (Ser256; 9461), rabbit FOXO3a (12829), rabbit p-AKT (Ser473; 4060), rabbit p-AKT (Thr308; 2965), rabbit AKT (9272), mouse (mAb) p-CREB (Ser133; 9196), rabbit PKA substrate (RRXS/T, 9624; 9621), rabbit PARP1 (9532), rabbit GRP78 (3177), rabbit LC3BI/II (2775), rabbit beclin 1 (3738), rabbit BAX (2772), rabbit pDRP1 (S616; 3455), rabbit DRP1 (8570), rabbit p-mTOR (S2448; 2971), rabbit mTOR (2972), rabbit p-Raptor (S792; 2083), rabbit Raptor (2280), mouse (mAb) p-p70SK (T389; 9206), rabbit p70SK (9202), rabbit p-S6R (S240/244; 5364), rabbit S6R (2217), rabbit p-4E-BP1(T37/46) (2855), rabbit 4E-BP1(S65) (9451), rabbit 4E-BP1(T70) (9455), rabbit 4E-BP1 (9452; Cell Signaling Technology), rabbit TFAM (GTX112760), and rabbit NDUFS3 (GTX105835; GeneTex); rabbit SOD2 (SOD-110; Assay Designs/Stressgen), mouse (mAb) MFN1 (831101), and mouse (mAb) MFN2 (820301; BioLegend).

Primary hepatocytes were isolated via two-step collagenase perfusion and isolation techniques (Seglen, 1976) using liver perfusion media and liver digest buffer. Cells were allowed to attach for 2–3 h on collagen-coated plates in DMEM containing 4.5 g/liter glucose and supplemented with 10% FBS, 2 mM sodium pyruvate, and 2% penicillin and streptomycin. After attachment, the medium was replaced, and the cells were incubated overnight before used for treatment with compounds as specified in the analyses.

Livers were fixed in 4% paraformaldehyde in phosphate-buffered saline for at least 7 d and embedded in paraffin. Hematoxylin and eosin staining was performed on 7-µm paraffin sections. For immunohistochemistry analysis, tissues were deparaffinized in xylene and rehydrated in a series of alcohol and water mixtures. Antigen retrieval was performed in 10 mM sodium citrate, pH 6.0, at 94°C for 30 min. Tissue sections were blocked in 3% BSA and then incubated with primary antibody overnight at 4°C, biotin-conjugated secondary antibody for 1 h at 25°C, and ExtrAvidin and alkaline phosphatase substrate for 30 min at 25°C. Sections were then developed with FAST Red (Roche) and counterstained with hematoxylin. Image acquisition was performed on an AxioVision Imager fluorescent microscope (ZEISS) supported with AxioCam HRC camera (ZEISS) using a 40× objective lens (NA 0.75). AxioVision REL 4.8 software was used for data analysis.

EM analyses were performed at the High Resolution Electron Microscopy Facility of Medical College of Georgia at Augusta University. Livers were in situ fixed by perfusion with 2% glutaraldehyde in 0.1 M sodium cacodylate buffer, pH 7.4, postfixed with 2% mOsO₄ (also in sodium cacodylate buffer) for 1 h. After dehydration, tissues were embedded in epon-Araldite resin. 70-nM thin sections were cut on a MT-7000 ultramicrotome and collected on copper grids. Sections were stained with uranyl acetate and lead citrate and examined under a JEM-1200 EXII transmission electron microscope (JEOL USA Inc.). Images were captured with an UltraScan 4000 CCD camera and First Light Digital Camera Controller (Gatan Inc.).

Glucose and pyruvate tolerance tests were performed after overnight fasting (16 h) and glucagon test after 6 h fasting. Animals were injected i.p. with 2 g glucose per kilogram body weight, 2 g/kg sodium pyruvate, or 10 µg/kg glucagon, and blood glucose levels were measured using an automated glucose monitor (Bayer HealthCare LLC). For assays showing effects of glucagon on p-CREB levels in liver, 4-h-fasted mice were injected with 30 µg glucagon per kilogram body weight via the inferior cava, and livers were collected for immunostaining or Western blotting analysis after 10 min. For in vivo insulin signaling studies, mice were injected i.p. with insulin at 0.75 U/g body weight, and livers were removed 10 min after injection.

In vitro glucose production studies were performed in primary hepatocytes cultured in DMEM with 10% FBS overnight. Cells were incubated for 4 h in glucose and phenol-red-free DMEM without serum, supplemented with 2 mM pyruvate and 20 mM lactate, 100 nM glucagon, 10 µM forskolin, or 100 mM Bt2-cAMP. Glucose output was determined using a colorimetric assay kit (Eton Bioscience Inc.).

For mitochondria analyzes, random EM sections of hepatocytes cut throughout the middle of the cell body were analyzed. Most of these sections contained the nucleus. AxioVision REL 4.8 software was used to evaluate the morphology of each individual mitochondrion in the digital images. Mitochondria cross-sectional areas were used to measure mitochondria size and mitochondria AR (major axis divided by minor axis). AR was used as a parameter of mitochondria shape description. Mitochondria density was estimated by counting the number of mitochondria per field of view area. Mitochondria coverage was estimated by dividing the total area of mitochondria by the field of view area. Probability plots were used to estimate changes in mitochondria size and shape and statistical differences were tested using the Kolmogorov–Smirnov test. Slices containing hepatocyte cytosol that were not cut across the middle of the cell were also included in this analysis. Mitochondria mass was measured by MitoTracker staining and mtDNA content by quantitative PCR. For mitochondrial potential and reactive oxygen species measurements, primary hepatocytes incubated with medium containing a 1-mM (low) or 25-mM (high) glucose concentration for 1.5 h were stained with 50 mM TMRE (Molecular Probes), 5 µM MitoSOX, or 5 µM dihydroethidium for 30 min at 37°C (protected from light) and analyzed by FACS. The susceptibility of hepatocytes to induced mitochondrial dysfunction was assayed by measuring mitochondrial membrane potential by TMRE staining of cells subjected to same conditions as above and treated with 10 µM of the protonophore carbonyl cyanide m-chlorophenyl hydrazone for 30 min.

Isolation of mitochondria from the primary hepatocytes or liver tissues was performed using a commercially available kit (catalog number 89874; Thermo Fisher Scientific). In brief, liver tissues or hepatic cells were homogenized in 2 ml mitochondria isolation buffer (70 mM sucrose, 210 mM mannitol, 5 mM Hepes, 1 mM EGTA, pH 7.2, and 0.5% BSA). The homogenates were centrifuged for 10 min at 800g at 4°C, and the supernatant was collected for additional centrifugation at 8,000g for 10 min at 4°C. The resulting pellet fraction containing the mitochondria fraction was resuspended, and the amount of protein was measured. Nuclear cell fractions were separated with commercial kit following the manufacturer’s protocol (89841; Thermo Fisher Scientific).

OCR and ECAR were measured in hepatocytes using a XF96 Extracellular Flux Analyzer (Seahorse Bioscience). Cells were seeded on collagen-coated XF96 plates at 4 × 10³ cells per well and cultured in DMEM containing 10% FBS overnight, rinsed twice, and kept in 100 µl XF medium for 1 h (nonbuffered XF base medium containing 25 mM glucose, 2 mM l-glutamine, and 1 mM sodium pyruvate). Three OCR measurements were obtained under basal conditions and upon sequential injection of 2 µM oligomycin, 2 µM fluoro-carbonyl-cyanide phenylhydrazone (FCCP), and 0.5 µM rotenone plus 0.5 µM antimycin A. OCR values were calculated from 3-min measurement cycles. The OCR measurements were adjusted to cell numbers. Glycolysis was assessed by analyzing ECAR in hepatocytes cultured in glucose-free medium after sequential addition of 10 mM glucose, 2 µM oligomycin, and 100 mM 2-deoxyglucose.

Mitochondrial FAO measurements were performed on primary hepatocytes seeded on collagen-coated XF96 plates at 4 × 10³ cells per well and allowed to attach for at least 3 h. Before the assay, cells were rinsed twice and incubated in 135 µl assay medium per well (FAO medium containing 111 mM NaCl, 4.7 mM KCl, 1.25 mM CaCl₂, 2 mM MgSO₄, 1.2 mM NaH₂PO₄, 2.5 mM glucose, 0.5 mM carnitine, and 5 mM Hepes, pH 7.4) for 45 min at 37°C. 30 µl per well 1 mM BSA-conjugated palmitate or BSA as control was added, and OCR measurements were obtained under basal conditions and upon sequential injection of 2 µM oligomycin, 2 µM FCCP, and 0.5 µM rotenone plus 0.5 µM antimycin A.

The sequential electron flow through different ETC complexes was determined by monitoring OCR in isolated liver mitochondria and sequential injections of rotenone, succinate, antimycin, or ascorbate and TMPT as described in the manufacturer’s technical manual (Seahorse Bioscience). Specifically, a buffer containing 10 mM pyruvate/2 mM malate with 4 µM FCCP was used to determine electron flow starting from complex I. Subsequent injections of 2 µM rotenone, 10 mM succinate, 4 µM antimycin A, and 100 µM TMPD/10 mM ascorbate was used to determine electron flow through complex II and complex IV, respectively. For this assay, 5 µg mitochondrial protein per well was used. OCR obtained in each step of the protocol was normalized by the mitochondrial protein content of the liver samples. Data were analyzed using XF software (Wave 2.2) and displayed as OCRs (picomoles per minute per well) following the Seahorse Bioscience instruction manual.

Complex I activity in isolated mitochondria was measured using a modified enzymatic assay (Spinazzi et al., 2012).

To study MC formation, BN-PAGE was performed exactly as recently described (Jha et al., 2016). In brief, 150 µg liver mitochondria isolated from WT and hsf1^−/− mice were solubilized in 6 g/g digitonin (digitonin/protein ratio) and separated on a 3–12% native PAGE Bis-Tris Gell System as per the manufacturer’s instructions (Thermo Fisher Scientific). After electrophoresis, gels were processed for Coomassie staining. Alternatively, 50 µg liver mitochondria solubilized in 6 g/g digitonin (protein ratio) were run on native PAGE (3–12%) gel and processed for immunoblotting with an OXPHOS antibody cocktail (457999; Invitrogen).

ATP and AMP content in liver tissues or hepatocyte cultures were measured using an ATP/ADP/AMP determination kit (A-125; State University of New York, Buffalo) following the manufacturer’s protocol. The amount was normalized to the protein content, which was determined using a protein assay (Bio-Rad Laboratories), and is presented as micromoles per gram protein. NAD⁺ and NADH levels in samples from liver and hepatocyte cultures were measured using a colorimetric NAD⁺/NADH assay kit (E2ND-100; BioAssay Systems). cAMP levels in primary hepatocytes were determined by using the Direct cAMP ELISA kit (Enzo Life Sciences) following the manufacturer’s procedures.

For mass isotopomer distribution analysis, primary hepatocytes cultured in glucose-free DMEM supplemented with dialyzed 10% FBS were uniformly labeled with 20 mM [U-¹³C]-6-glucose (Sigma-Aldrich) for 6 h. The samples were analyzed by gas chromatography/mass spectrometry to determine the contribution of glucose to tricarboxylic acid cycle metabolites. Data are presented as fraction of ¹³C-labeled species within a given sample. M + 0 corresponds to an unlabeled metabolite (for example, m + 2 represents an ion labeled with two ¹³Cs). Glucose flux analyses were performed at the Mouse Metabolic Phenotyping Center at Case Western University (Cleveland, Ohio), which is a facility sponsored by the National Institutes of Health.

For the measurement of protein synthesis in the primary hepatocytes, equal numbers of cells (2 × 10⁵ per well) were plated on six-well BioCoat plates (Costar) overnight. After three washes, cells were cultured in with leucine-free DMEM supplemented with 5% dialyzed FBS for 30 min to depleted internal stores of leucine and then pulsed with ³H-leucine in the same medium for another 30 min. Cells were washed with PBS and recultured in regular DMEM plus 10% dialyzed FBS for 30 min. After three quick washes with medium containing excess of unlabeled 2 mM leucine, 10% TCA was added to cells to precipitate the proteins, and radioactivity was measured. Protein synthesis rates were normalized to total cellular proteins prepared from parallel experiments without radioactive labeling. Experiments also were performed to measure protein synthesis after growth factor–induced signaling. To this effect, to stimulate basal translation, hepatocytes were incubated with 100 nM insulin for 30 min before metabolic labeling with ³H-leucine as described above.

For translational analysis by polysomal profiling in liver, tissue was gently homogenized in 50 mM Tris HCl, pH 7.8, 240 mM KCl, 10 mM MgSO₄, 5mM DTT, 250 mM sucrose, 2% Triton X-100, 90 µg/ml cycloheximide, and 30 U/ml RNasin using a glass douncer. 100 µg/ml heparin was added to the liver extract. After clarification by centrifugation 12,000 rpm for 10 min at 4°C, 5 mg protein was loaded on a 15–50% sucrose gradient dissolved in 25 mM Tris HCl, pH 7.4, 25 mM NaCl, 5 mM MgCl₂, and 1 mM DTT and run at 39,000 rpm (190,000 g) for 3 h and 30 min with a SW41Ti swing rotor (Beckman Coulter). Gradients were fractionated, and optical density at 254 nm was continuously recorded using an ISCO fractionator (Teledyne ISCO). To calculate the ratio between polysome and monosome-free subunits, the area underlying the corresponding peaks on the profile was measured using AxioVision REL 4.8 software.

For lipogenesis assay, 2 h serum starved primary hepatocytes were incubated with serum-free DMEM containing 74 KBq/ml [2-¹⁴C]-sodium acetate (MP Biomedicals) and 0.5 mM unlabeled sodium acetate. After 4-h incubation, cells were washed twice with PBS and then lysed with 1 N NaOH solution and incubated for 1.5 h at 65°C. The resulting aqueous solution was acidified with concentrated 12 N hydrochloric acid, and lipids were extracted with petroleum ether. Radioactivity incorporation in the total lipid fraction was counted with a liquid scintillation counter (Tri-Carb 2500; PerkinElmer). For insulin-stimulated lipogenesis measurements, hepatocytes were starved for 4 h and then incubated with DMEM containing 100 nM insulin, 74 KBq/ml [2-¹⁴C]-sodium acetate, and 10 µM unlabeled sodium acetate for 2 h. After lysis of the cells, lipids were extracted and radioactivity was counted.

Hepatic FAO was determined by the liberation of ³H2O from [³H]-oleic acid according to a procedure described previously (Harada et al., 2007; Sengupta et al., 2010) with modifications. The reaction mixture was Krebs–Ringer bicarbonate buffer containing 1 mM [³H]-BSA-oleic acid (final concentration of 1 µCi/mM; Sigma-Aldrich) and 4% BSA, which was aerated with 5% CO₂ for at least 30 min before use. Approximately 40-mg liver slices were incubated in individual wells of a 24-well plate with the reaction mixture (total volume of 1 ml) at 37°C and 5% oxygen for 2 h After the incubation period, the supernatant was precipitated twice with 10% TCA followed by a centrifugation at 3,000g for 5 min. 1.5 ml of the supernatant was transferred to a 1.5-ml microfuge tube with no cup. Tubes were placed in scintillation vials that contained 2 ml of water, wrapped in aluminum foil, and incubated at 55°C for 18 h. The vials were then cooled at 4°C for 30 min, the tubes were removed, and the levels of ³H₂O were measured with a liquid scintillation counter (Wallac 1450 MicroBeta Trilux; PerkinElmer).

ChIP quantitative PCR experiments were performed using a ChIP assay kit (EMD Millipore), according to manufacturer’s instructions and as described in a previous report (Mendillo et al., 2012). One million cells or 50 mg liver tissue (minced in small pieces) were cross-linked with 1% formaldehyde for 10 min at 37°C. Cells were washed twice with PBS containing protease inhibitor cocktail (1 mM PMSF, 1 µg/ml aprotinin, and 1 µg/ml pepstatin A), pelleted, resuspended in 200 µl lysis buffer (1% SDS, 10 mM EDTA, and 50 mM Tris, pH 8.1), and sonicated in a ultrasonic processor (Sonicator 3000; Misonix) to a DNA size of ∼500 bp that was confirmed by agarose gel electrophoresis. Samples were precleared with protein A agarose beads and then immunoprecipitated with HSF1 antibody (sc-9144; Santa Cruz Biotechnology, Inc.), or a rabbit IgG antibody as a control, at 4°C for 16 h. Samples from hsf1^−/− mice were used as an additional assay control. Chromatin protein–DNA complexes were eluted twice from the agarose beads by adding elution buffer (1% SDS, 0.1 M NaHCO₃, pH 8.0) at room temperature for 15 min. Cross-linking was reverse by heating at 65°C for 4 h. DNA fragment were purified by phenol/chloroform and ethanol precipitation and the putative HSE site present in the mouse nampt or hsp70 promoters were amplified by quantitative PCR with the following primers: Nampt forward, 5′-CCCGCATAAAACAGCCTAAA-3′; Nampt reverse, 5′-AACTGGCACAGACTGCATCA-3′; Hsp70 forward, 5′-TCCAGAGACAAGCGAAGACA-3′; and Hsp70 reverse, 5′-GAGTAGGTGGTGCCCAGGT-3′.

Total RNA was isolated from liver tissue using TRIzol (Invitrogen) and reverse transcribed using the Iscript cDNA synthesis kit (170-8891; Bio-Rad Laboratories). Quantitative real-time PCR reaction was performed using iQ SYBR Green Supermix (Bio-Rad Laboratories) on a Stratagene Mx 3000P QPCR system. Relative quantitative plots were constructed for quantity of RNA input and for each gene of interest. Quantitative real-time PCR was performed using gene-specific primers (primer sequences are presented in Table S1).

To measure mtDNA content (marker of mitochondrial number), liver tissue was homogenized using 1× PBS and digested in a lysis buffer containing Proteinase K overnight. Genomic and mitochondrial DNA were extracted by conventional phenol-chloroform method. Quantitative PCR was performed using mitochondrial DNA (cytochrome c) and genomic DNA (b-actin)–specific primers (Cytb forward, 5′-CGCTTTCCACTTCATCTTACC-3′; Cytb reverse, 5′-CCTGTTGGGTTGTTTGATCC-3′; Actb forward, 5′-TTTCTGTCACTCTTCTCTTAGGT-3′; and Actb reverse, 5′-AGGTCTTTACGGATGTCAAGG-3′).

All analyses were performed with at least five to ten mice. Data are presented as mean ± SD. Statistical significance between experimental groups was assessed by an independent Student’s t test, and P < 0.05 was considered significant.

Fig. S1 describes the strategy used to generate and characterize mice with a conditional hsf1 allele. Fig. S2 evaluates the effects of HSF1 loss on hepatic NAD⁺-consuming PARP and CD38 activities and profiles the mRNA levels of key enzymes in NAD⁺ biosynthesis as well as of dehydrogenases that reduce NAD⁺ to NADH. Fig. S3 describes the effects of HSF1 loss on hepatic mitochondrial dynamics and mitophagy as well as mitochondrial DNA replication and unfolded protein response. Fig. S4 describes analyses on the inhibitory effects of HSF1 loss in gluconeogenesis through attenuation of glucagon- and FOXO1-mediated signaling. Fig. S5 describes a model highlighting the HSF1-dependent effects on hepatic bioenergetics and anabolic metabolic pathways, including gluconeogenesis, translation, and lipid synthesis. Table S1 lists the primers used for quantitative PCR. Table S2 lists the genes mentioned in the text.

We thank Dr. Michelle Puchowicz at Case Western Metabolic Phenotyping Center for metabolic analyses and Chao Jiang for excellent technical support.

The research was supported by National Institutes of Health grants CA062130, CA062130S1, and CA132640 (N.F. Mivechi and D. Moskophidis).

The authors declare no competing financial interests.

Author contributions: A. Qiao, X. Jin, and J. Pang acquired the data. J.P. developed the technology. D. Moskophidis and N.F. Mivechi conceived and designed the study and wrote the manuscript.