Adipocyte Pseudohypoxia Suppresses Lipolysis and Facilitates Benign Adipose Tissue Expansion

In obesity, the gradual expansion of adipose tissue associates with local hypoxia due to an inadequate vascular response (1,2). Hypoxic adipose tissue, as found in diet-induced obesity models, or transgenic hypoxia-inducible factor (HIF)-1α overexpression in adipose tissue associates with local inflammation, fibrosis, and metabolic dysfunction (1–6). In contrast, in models of dietary obesity in which adipose tissue hypoxia is attenuated (11β-HSD1–deficient mice and transgenic adipose overexpression of vascular endothelial growth factor [VEGF]), the structural and metabolic abnormalities are ameliorated, and benign adipose tissue expansion occurs (7–10). Understanding mechanisms that allow benign expansion of adipose depots is of high importance, as it may lead to therapeutic strategies for minimizing the pathogenesis of obesity and/or lipodystrophy syndromes.

Critical for benign expansion of fat depots is efficient storage and release of fatty acids in adipocytes (11,12) and adequate expandability of the adipose–vascular network (13). Adipose tissue expansion is a dynamic process that involves adipocyte hypertrophy in combination with vascular remodeling involving endothelial cells, macrophages, and the extracellular matrix (13–16). Low oxygen tension (hypoxia) can occur due to an inability of the tissue to provide adequate compensatory vascular supply (1–5). In this context, cells respond to hypoxia by activation and stabilization of the HIFα isoforms (17). Increased HIF-1α activation may contribute to the pathological changes within adipose tissue in obesity (1–5) in part through inhibition of peroxisome proliferator-activated receptor γ-2–dependent adipogenesis (18). This concept is supported by the phenotype of transgenic mice overexpressing Hif1α in adipose tissue that exhibits insulin resistance and localized adipose tissue fibrosis (6). In contrast, HIF-2α promotes adipose differentiation in vitro and, given that HIF-2α levels are also increased after 4 weeks in high-fat–fed mice (19), may counteract pathogenic changes associated with HIF-1 at early stages of obesity development. The oxygen-sensitive signal event that regulates HIF is mediated by hydroxylase enzymes that regulate the protein stability and consequent transcriptional activity of HIFα (20). HIF–prolyl hydroxylases (PHDs; otherwise known as EGLNs) belong to the large family of Fe (II) and 2-oxoglutarate–dependent oxygenases (21–24). PHDs hydroxylate conserved prolyl residues of the HIF-1α and HIF-2α subunits, thus promoting their binding to the von Hippel Lindau (VHL) tumor suppressor protein, which targets HIFα isoforms for proteasomal degradation in normoxia (21–24). In humans, there are three isoforms of PHD enzymes (PHD1–3), with PHD2 (EGLN1) the most abundant enzyme, including in mature adipocytes (25). PHD2 is the most important for setting basal activity of the HIF system in most cells (20). Despite the growing understanding of the pathological role of HIF-1α activation in adipose tissue during obesity (1–10,26–32), direct pharmacological targeting of HIF remains challenging. In contrast, therapeutic targeting of PHDs to induce a pseudohypoxic (activation of HIFα and target genes in normoxia) state is under active clinical development in the context of anemia and other diseases involving hypoxia (33,34). In this study, we addressed the metabolic consequences and potential therapeutic impact of pseudohypoxia by genetic and pharmacological inhibition of the principal oxygen-sensing enzyme PHD2 in adipose tissue.

The Phd2 conditional allele (35) on a congenic C57BL/6J background, the Hif1α conditional allele (36), and the Hif2α conditional allele (37) were crossed with the fatty acid binding protein 4 (Fapb4)-Cre allele (38) (The Jackson Laboratory) to achieve adipose-specific conditional knockout mice. Hif1α (stock number 007561) and Hif2α (stock number 008407) mice were purchased from The Jackson Laboratory. Genotyping and recombination efficiency PCRs were performed as previously described (35–39). In all experiments described, control littermates were used for comparisons. For diet-induced obesity experiments, mice were given the D12331 high-fat diet (58% kcal fat; Research Diets Inc.) for 12 weeks. To assess the effect on adipocyte lipolysis of pharmacologically inhibiting PHD, we used 2-(1-chloro-4-hydroxyisoquinoline-3-carboxamido) acetic acid (40,41), a potent small-molecule inhibitor of the PHD enzymes that has been shown to activate HIFα (40,41). For analysis of the effects of this PHD inhibitor (PHI) in vivo, C57BL/6J mice were used. In brief, mice were fasted overnight, blood was collected for basal nonesterified fatty acid (NEFA) quantification, mice were then divided into two groups, those receiving intraperitoneal (i.p.) PHI (30 mg/kg; a dose that was sufficient to robustly induce HIF in liver) (41) and a control group receiving vehicle (5% DMSO) alone. PHI was administered for an hour prior to CL316,243 (1 μg/kg) stimulation. Male adult mice were used in all the experiments. Animals were bred under standard conditions and fed standard chow (product 801151; Special Diet Services, Essex, U.K.) ad libitum unless stated otherwise. Animal studies were performed under licensed approval in accordance with the U.K. Home Office Animals (Scientific Procedures) Act, 1986.

For glucose tolerance tests, mice were fasted 5 h before administration of glucose (2 mg/g body weight [BW]) orally or by i.p. injection. Glucose concentration was measured by a blood glucose monitoring system (OneTouch Ultra2; LifeScan, Milpitas, CA). For β3-adrenergic receptor agonist tests, mice were fasted overnight, and blood samples were collected before and 15, 30, and 60 min after i.p. injection of 1 μg/kg CL316,243 (Sigma-Aldrich, Dorset, U.K.). In the PHI experiments, PHI was administered i.p. (30 mg/kg) and blood collected 1 h prior to the i.p. CL316,243 challenge. Insulin and leptin concentrations were measured using commercial ELISA kits (Crystal Chem Inc., Downers Grove, IL). Liver and muscle triglyceride levels were measured using a commercial kit (Abcam, Cambridge, U.K.) according to the manufacturer’s instructions.

Total RNA was extracted from cells and tissue using TRIzol (Invitrogen, Paisley, U.K.) and treated with DNase I (Invitrogen). One microgram total RNA was used for first-strand DNA synthesis using SuperScript III cDNA Synthesis system (Invitrogen), and quantitative PCR (qPCR) was performed with the LightCycler 480 (Roche), using mouse TaqMan assays (Life Technologies, Paisley, U.K.) for all genes measured. A standard curve was constructed for each gene measured using a serial dilution of cDNA pooled from all samples. Results were normalized to the expression of 18S rRNA.

Whole-cell lysates were prepared in ice-cold buffer (5 mmol/L HEPES, 137 mmol/L NaCl, 1 mmol/L MgCl₂, 1 mmol/L CaCl₂, 10 mmol/L NaF, 2 mmol/L EDTA, 10 mmol/L Na pyrophosphate, 2 mmol/L Na₃VO₄, 1% Nonidet P-40, and 10% glycerol) containing protease inhibitors (Complete Mini; Roche Diagnostics Ltd., West Sussex, U.K.). Blots were probed with HIF-1α (Cayman Chemical, Ann Arbor, MI), HIF-2α (Novus Biologicals, Littleton, CO), and HIF PHD2 (Novus Biologicals) antibodies. Lipolysis activation antibody kit was used to identify hormone-sensitive lipase (HSL), phosphorylated (p-)HSL⁶⁶⁰, p-HSL⁵⁶⁵, and perilipin (New England BioLabs, Ipswich, MA). The adipose triglyceride lipase (ATGL), AKT, and p-AKT (Ser473) antibodies were from Cell Signaling Technology (Hertfordshire, U.K). HRP-conjugated anti-rabbit (Dako, Cambridgeshire, U.K.) secondary antibody was used. Signal was detected using ECL Plus (GE Healthcare Life Sciences, Buckinghamshire, U.K.). Blots were reprobed with an HRP-conjugated anti-actin antibody (Abcam). Densitometry was performed using the ImageJ software.

Formalin-fixed, paraffin-embedded adipose sections (4 μm) were used. Images were acquired using a Zeiss microscope (Zeiss, Hertfordshire, U.K.) equipped with a Kodak DCS 330 camera (Eastman Kodak, Rochester, NY). Adipocyte size was calculated by measuring the diameter of adipocytes from 20 randomly selected areas per section using a Zeiss KS300 image analyzer. Antibodies to CD31 (Abcam) for endothelial cells/vessels or F4/80 (Abcam) for macrophages were used. Binding of primary antibody was visualized using diaminobenzidine, Chromogen-A (Dako). Counterstain was performed by rinsing in 70% hematoxylin. Picrosirius red (Sigma-Aldrich) staining of adipose sections was used to identify total collagen. Analysis was performed blind to the experimental grouping as previously described (7).

Primary murine or human adipocytes were isolated as previously described (7). For the analysis of lipolysis in aP2-Phd2KO, aP2-Hif1αKO, and aP2-Hif2αKO adipocytes, hypoxic stimulation was performed by replacing the media with 1% O₂ preconditioned media. Cells were left at 1% O₂ in a hypoxic chamber (1% O₂) for 4 h, after which media was collected for measurement of NEFA release, and cells were lysed in RIPA protein lysis buffer for analysis by immunoblotting. Effects of PHI on lipolysis were examined in C57BL/6J primary adipocytes. In brief, adipocytes were pretreated with PHI [0.5 mmol/L, a sufficient dose to robustly activate HIFα (41)] or vehicle (DMSO, 0.5%) for 1 h prior to β-adrenergic stimulation with CL316,243 (100 nmol/L) for 3 h. Medium was collected for measuring NEFA concentrations with a NEFA kit (Alpha Laboratories, Eastleigh, U.K.), and adipocytes were lysed for protein quantification and immunoblot analysis.

Human abdominal subcutaneous adipose tissue (SAT) was obtained, after written consent, from healthy female subjects (age range 30–50 years old) with mean BMI 25.6 (± 1.9) kg/m² undergoing cosmetic lipectomy procedures. Ethical approval for the collection of tissue and subsequent research was granted by the South East Scotland Research Ethics Committee 3, reference number 10/S1103/45. Adipocytes were isolated as above and exposed to PHI as above. In brief, human adipocytes were pretreated with PHI or vehicle 1 h prior to stimulation with isoproterenol (100 nmol/L) for an additional 2 h. Medium was collected for measuring NEFA concentrations, and adipocytes were lysed for protein quantification and immunoblot analysis.

Adipose tissue samples (91 subcutaneous and 102 omental abdominal) were obtained from the same location during elective surgical procedures (cholecystectomy [n = 10], surgery of abdominal hernia [n = 14], and gastric bypass surgery [n = 78]) from participants with BMI within 20–68 kg/m². All subjects were of Caucasian origin and reported that their BW had been stable for at least 3 months before the study. In the obese type 2 diabetic group, two subjects were treated with metformin, one with glitazones, and one with insulin. Liver and renal diseases were specifically excluded by biochemical workup, measuring biomarkers of liver and renal injury. All subjects gave written informed consent, approved by the ethics committee (Comitè d’Ètica d’Investigació Clínica). Adipose tissue samples were immediately frozen in liquid nitrogen and stored at −80°C. RNA was prepared using RNeasy Lipid Tissue Mini Kit (Qiagen, Barcelona, Spain) and integrity checked by Agilent Bioanalyzer (Agilent Technologies, Palo Alto, CA) and reverse transcribed to cDNA using High-Capacity cDNA Archive Kit (Applied Biosystems, Madrid, Spain) according to the manufacturer’s protocol. Gene expression was assessed by real-time PCR using a LightCycler 480 Real-Time PCR System (Roche Diagnostics SL, Barcelona, Spain), using commercially available TaqMan primer/probe sets (Applied Biosystems). A threshold cycle (Ct) value was obtained for each amplification curve, and a ∆Ct value was first calculated by subtracting the Ct value for human Cyclophilin A (PPIA) RNA from the Ct value for each sample. Fold changes compared with the endogenous control were then determined by calculating 2^−∆Ct. Results are expressed as ratio to PPIA gene expression.

Values are reported as mean ± SEM. All statistical analysis was performed in GraphPad Prism 5 software. Simple comparisons were analyzed using Student t test. For multiple comparisons, differences between genotypes and the effect of treatment(s) were analyzed by two-way ANOVA with subsequent Tukey post hoc test or Mann–Whitney if data were not normally distributed. Significance was set at P < 0.05.

We studied intercross progeny of the littermate genotype Phd2^lox/lox mice (control, designated Phd2; described in Refs. 35,39) with the Fapb-4-Cre (aP2-Cre) mice (38) to make an adipose tissue-selective Phd2 deficiency model designated aP2-Phd2KO. Efficient recombination was detected in epididymal, mesenteric, and brown adipose tissue and in isolated epididymal adipocytes (Supplementary Fig. 1A). Recombination was observed in the whole stromal vascular fraction, but there was no evidence for recombination in macrophages (Supplementary Fig. 1A). PHD2 mRNA (Fig. 1A) and protein were efficiently knocked down (∼70%) in aP2-Phd2KO adipose tissue (Fig. 1B) and in primary adipocytes (Fig. 1C) cultured in vitro. No significant Phd2 recombination was found in other tissues such as heart, muscle, spleen, or liver (Supplementary Fig. 1B). Adipose Phd1 and Phd3 mRNA levels were unaffected (Supplementary Fig. 1C). Hypoxia markedly upregulated Phd2 mRNA levels in vitro (threefold; P < 0.0001) in control adipocytes, but this induction was absent in aP2-Phd2KO adipocytes (Fig. 1D).

Homozygous deletion of Phd2 in adipocytes led to stabilization of HIF-1α and HIF-2α protein levels in adipose under normoxic conditions (Fig. 1E and F). HIF target genes encoding proteins for nutrient metabolism (Glut1, Glut4, Pdk1, and Pdk4) and angiogenesis (Vegfα, Angptl4, and Tgfb) were higher in white adipose tissue of aP2-Phd2KO compared with control littermates (Fig. 2A). Hypoxia led to significantly higher mRNA levels of Glut1 and Vegfα in isolated aP2-Phd2KO than control adipocytes (Fig. 2B). In order to evaluate which Hifα isoform dominantly regulates target genes in the aP2-Phd2KO model, we measured mRNA levels of the same genes in mice with adipose-specific deletion (aP2-Cre–driven deletion) of Hif1α (aP2-Hif1αKO) or Hif2α (aP2-Hif2αKO), which blunts the hypoxia-induced stabilization of HIF-1α in adipocytes (Supplementary Fig. 2A). Deletion of Hif1α in adipose tissue reduced the mRNA levels of genes that control nutrient metabolism (Glut1, Pdk1, and Pdk4) and the profibrotic response (Tgfb), with pronounced effects in high-fat–fed animals (Fig. 2C). This effect was localized to mature adipocytes (Supplementary Fig. 2B). In contrast, adipose Hif2α deletion reduced the mRNA levels of angiogenic genes (Vegfα and Angptl4) (Fig. 2D), indicating a predominant role for HIF-2α in adipose vascularization.

Chow diet (low-fat)–fed aP2-Phd2KO mice exhibited increased adiposity by 19 weeks of age compared with control littermates (Fig. 3A), with approximately twofold higher cumulative weight gain from 9–30 weeks (Fig. 3B). Adult aP2-Phd2KO mice had more white adipose tissue mass (>50%) in all depots, but brown adipose and liver masses were unaltered (Fig. 3C). Body and adipose tissue weight differences were not due to increased food intake (Fig. 3D). As expected for their higher adiposity, aP2-Phd2KO mice showed a parallel increase in plasma leptin levels (Table 1).

Despite increased adiposity, aP2-Phd2KO mice exhibited normal glucose tolerance (Fig. 3E) and similar fasted and fed plasma insulin levels (Table 1) to their control littermates. aP2-Phd2KO mice had comparable ad libitum–fed but reduced fasting NEFA levels (after 5 h or overnight fast, Table 1 and Fig. 3F). Notably, aP2-Phd2KO mice exhibited blunted β-adrenergic agonist CL316,243 responsiveness in vivo (Fig. 3F), with significantly reduced NEFA release. Liver and muscle triglyceride levels were lower in aP2-Phd2KO mice (Fig. 3G).

Greater fat deposition in aP2-Phd2KO mice was associated with larger adipocytes (diameter, 230 ± 12 and 177 ± 18 μm in controls; P = 0.034) (Fig. 4A). However, adipose Phd2 deficiency did not affect adipose macrophage density (Fig. 4B) or fibrosis (Fig. 4C, Supplementary Fig. 3), markers of inflamed and hypoxic adipose tissue, respectively. Intriguingly, adipose Phd2-deficient mice manifested approximately twofold (P = 0.001) greater density of immunostaining for the endothelial marker CD31 in adipose tissue, indicating an enhanced vascular expansion (Fig. 4D and E).

Because Phd2 deficiency in the model presented in this study stabilizes both HIFα isoforms, we attempted to dissect the role of different HIFα isoforms in adipocyte function in vivo. Therefore, aP2-Hif1αKO or aP2-Hif2αKO mice were analyzed. Chow-fed aP2-Hif1αKO or aP2-Hif2αKO were comparable to control littermates (data not shown); therefore, these mice were analyzed after 12 weeks exposure to high-fat feeding. Notably, Hif1α-deleted mice manifested reduced BW gain and improved glucose tolerance on high-fat diet (Supplementary Fig. 4A–C). In contrast, adipose tissue Hif2-deleted mice fed a high-fat diet gained similar BW (Supplementary Fig. 4D) but showed worsening of glucose tolerance (Supplementary Fig. 4E) and elevated fasting insulin (Supplementary Fig. 4F), suggesting insulin resistance. There was no difference in plasma fasting NEFA or liver triglyceride levels in aP2-Hif2αKO or aP2-Hif2αKO mice compared with their littermate controls (Supplementary Table 1).

To investigate the mechanism by which adipose Phd2 deficiency drives adipocyte hypertrophy, we assessed lipolytic responses of the adipocytes in vitro under normoxic (21% O₂) and hypoxic (1% O₂) conditions. In normoxia, basal and β-adrenergic–stimulated NEFA release into the medium of aP2-Phd2KO adipocytes was significantly lower than control (Fig. 5A). The reduced lipolytic response of aP2-Phd2KO adipocytes was also apparent under hypoxic conditions; hypoxia itself attenuated stimulated but not basal lipolysis (Fig. 5A). The reduced NEFA release from Phd2-deficient adipocytes was associated with reduced levels of the key lipolytic proteins HSL and perilipin (Fig. 5B, C, and E). Lower HSL mRNA was also observed in aP2-Phd2KO adipose tissue, whereas perilipin mRNA levels were unaltered (Fig. 5G). In contrast, the mRNA of the triglyceride synthesis gene diacyglycerol acylotransferase-1 (Dgat1) was higher in aP2-Phd2KO adipose tissue. Supporting the HIF-1α dependence of this effect, adipocytes from aP2-Hif1αKO and not aP2-Hif2αKO mice showed a higher basal and β-adrenergic–stimulated NEFA release in hypoxia (Fig. 5H) as well as higher total HSL and perilipin protein levels (Fig. 5I and J and quantification in Supplementary Fig. 5A–C). Consistent with this, Hif2α-deficient adipocytes showed normal NEFA release and HSL levels (Supplementary Fig. 6A–C). Short-term activation of HSL is mediated by phosphorylation at serine 660 (protein kinase A mediated) and serine 565 (AMPK mediated) (42). Immunoblotting revealed that PHD/HIF-1α affected total HSL protein levels but not the p-HSL/total HSL protein ratio (Fig. 5D and F and Supplementary Fig. 5C). In addition, Phd2 deficiency increased expression of the main antilipolytic α2-adrenergic receptor and tended to decrease levels of the lipolytic β3-adrenergic receptor, consistent with greater antilipolytic tone in aP2-Phd2KO adipose tissue (Fig. 5G).

We next explored the effect on lipolysis of a small-molecule PHI that activates HIFα robustly (40,41). Pharmacological inhibition of PHDs in mouse adipocytes by PHI led to a significant stabilization of HIF-1α protein levels (10-fold; P = 0.026) and, to a lesser extent, HIF-2α (Fig. 6A). In both murine (Fig. 6B) and human (Fig. 6C) primary adipocytes, PHI elicited a profound suppression of basal and adrenergic agonist–stimulated NEFA release in normoxia. As was observed in aP2-Phd2KO, total ATGL was not affected, but HSL and perilipin protein levels were lower in the PHI-treated adipocytes (Fig. 6D and E) with no effect on the ratio of p-HSL to total HSL (Fig. 6D) after adrenergic stimulation. PHI treatment reduced basal ATGL levels (Fig. 6F).

Administration of PHI to mice in vivo suppressed basal plasma NEFA and β-adrenergic–stimulated NEFA release (Fig. 6G). This effect was consistent with a selective downregulation of HSL mRNA levels in adipose tissues from PHI-treated C57BL/6J mice (Fig. 6H). In human adipocytes in vitro, isoproterenol increased the phosphorylation of AKT (Ser473) in both vehicle and PHI-treated adipocytes (Fig. 6I). However, the ratio of p-AKT (Ser473) to total AKT was higher in the PHI-treated human adipocytes (Fig. 6J), suggesting selective enhancement of insulin sensitivity with PHD inhibition.

We further extended our findings on the role of PHD2 in adipose tissue in a cohort of patients with type 2 diabetes. Adipose PHD2 (EGLN1) mRNA levels did not correlate with crude measures of obesity (BMI; Table 2). However, PHD2 mRNA levels positively correlated with mRNA for caveolin (CAV1), a coactivator of the lipolytic protein ATGL, ABHD5, and another lipolytic cascade-modulating gene, the a-kinase anchoring protein (AKAP). Notably, type 2 patients with diabetes had lower SAT PHD2 mRNA levels than nondiabetic participants (Table 2).

In this study, we describe a novel biological role for pseudohypoxia (Fig. 7) in the suppression of adipocyte lipolysis. Genetic Phd2 deficiency resulted in an increased adipose mass, but with normal glucose tolerance and reduced circulating levels of fatty acids. This apparently protective adipose expansion was associated with increased vascularity and no increase in adipose tissue macrophage burden or fibrosis. Mechanistic dissection revealed that in isolated mouse adipocytes in vitro, basal and β-adrenergic–stimulated release of NEFA was reduced by genetic inactivation of Phd2. Importantly, these findings were recapitulated by the acute exposure of mouse and human adipocytes to a selective PHI that targets PHD2.

In oxygenated cells, the HIF-PHD promote degradation of HIFα subunits following association of hydroxylated HIF with the VHL ubiquitin E3 ligase. PHD2 is the most abundant of these enzymes and the most important in setting levels of HIF activity in normoxic cells (20). Thus, PHD2 inactivation by genetic or pharmacological means is predicted to mimic hypoxia, creating a pseudohypoxic signal that activates the HIF transcriptional cascade in normoxic conditions. Although we cannot be certain that all of the effects we observed were mediated by this mechanism, several lines of evidence support this. First, both HIFα proteins and HIF transcriptional target genes were induced in adipose tissues and cells by these interventions. Second, genetic inactivation of Hif1α revealed effects that were consistent with induction of HIF being responsible for at least some of the observed effects of Phd2 inactivation. In particular, inactivation of Hif1α in adipose tissues resulted in reduced BW gain with high-fat feeding and enhanced lipolytic responses in adipocytes. We therefore propose a model in which Phd2 inactivation leads to coordinated activation of HIF pathways including marked upregulation of Vegfα and Angptl4 (most likely through HIF-2α activation) with concomitant suppression of the key lipolytic enzyme HSL and upregulation of lipogenesis through Dgat1 to increase fat-cell lipid storage. We extended our findings to show a positive correlation between PHD2 and ABHD5 in SAT in humans. We also found a negative correlation between PHD2 mRNA with mRNA for the lipid droplet protein-encoding genes perilipin or TIP47. These data support a role for oxygen sensing as an endogenous regulator of adipose tissue lipid mobilization in humans and suggest PHD2 inhibition could promote metabolically protective peripheral fat accumulation. In this regard, reduced adipose perilipin protein levels in the ap2-Phd2KO mouse model were an unexpected finding, as this might be predicted to promote lipolysis. The regulation of the lipolytic machinery is complex, and future work will investigate perilipin phosphorylation and indeed other lipid droplet protein levels that may reconcile the clearly antilipolytic effects of PHD inhibition at this functional level.

A number of observational studies have implicated adipose hypoxia in the pathogenesis of insulin-resistant obesity (2–4). Furthermore, several groups have generated transgenic mouse models of HIF-1α modulation in adipose tissue. Some (5,29–32), but not all (28,43,44), of these studies have reported that elevated levels of HIF-1α lead to obesity, with fibrotic scarring and insulin resistance in the adipose tissue. Consistent with this, genetic inactivation or pharmacological inhibition of the HIF system is protective against obesity and its associated metabolic abnormalities (32,33). Surprisingly, overexpression of HIF-1α specifically in adipose tissue did not induce canonical HIF-target genes (Vegfα and Glut1) or affect the local adipose tissue angiogenic phenotype (5). Instead, HIF-1α overexpression caused a profibrotic adipose gene expression profile with associated histological abnormalities (5). These contrasting findings may reflect context-dependent regulation of overlapping target genes by the pleiotropic HIFs. Underlining this complexity, a recent study showed no obvious phenotypic changes under normal dietary conditions with an adipose-selective Phd2 deletion mouse model analogous to ours (43). Indeed, this study reported obesity resistance in adipose Phd2-deleted mice (43) in contrast to previous reports showing HIF-1α activation leads to an obese phenotype (5,29–32). Moreover, obesity resistance in the adipose-selective Phd2-deleted mice became apparent only after chronic high-fat feeding (43). The reasons for the discrepancy between these studies could reflect differences in mouse strain genetic background, age, dietary intervention, and the level of Phd2 deletion achieved. Moreover, we did not observe a compensatory Phd3 upregulation in our ap2-Phd2KO model, in contrast to Matsuura et al. (43). Finally, while this manuscript was in revision, another study using a hypomorphic Phd2 mutant mouse model of whole-animal PHD2 deficiency showed an improved metabolic profile with reduced adiposity (44). Complementary findings between our study and that of Rahtu-Korpela et al. (44) are promising in terms of the therapeutic potential of PHIs to improve cholesterol as well as reduced plasma NEFA levels (our study) and differences in adiposity phenotypes likely reflect our distinct tissue-specific effects versus those of whole-animal Phd2 hypomorphicity. In any case, the focus of the work presented in this study is the effect of PHD/HIF signaling on adipocytes in mice and humans under normal dietary conditions and not under obesogenic conditions. Indeed, we have shown high-fat diets regulate the PHD system (7) and thus may potentially confound important baseline differences.

Of note, enlarged adipocytes but normal glucose tolerance test responses were reported in a distinct genetic mouse model of adipose-specific HIFα activation caused by ablation of the VHL tumor suppressor (a factor degrading HIF downstream of PHD action) (45). The adipose-specific VHL knockout model exhibited a cardiac hypertrophy phenotype resulting from severe adipose inflammation and highlighted the effect of manipulating distinct HIFα isoform levels (45). We show that targeting PHDs, as opposed to VHL, also reveals distinct effects, and our findings are key in order to fully evaluate which components of the HIF signaling pathway are relevant to metabolic pathologies.

The pseudohypoxic induction of HIF caused by Phd2 deletion may effectively pre-empt the remodeling effects normally caused by local hypoxia when fat rapidly expands in obesity. This may prime adipose tissue to resist the inflammatory and fibrotic responses seen with vascular rarefaction in obesity by activating the proangiogenic factors such as Angptl4 and Vegfα and coordinating more effective vascular remodeling to cope with fat deposition (5,30,32). Inactivation of Phd2 better mimics the coordinated physiological activation of all isoforms of HIF system than overexpression of a single isoform. Thus, although in many cells, PHD2 has greater activity on HIF-1α than HIF-2α (46), we observed that inactivation of Phd2 deficiency in adipose tissue is also associated with stabilization of HIF-2α. A large body of data now indicates that HIF-1α and HIF-2α have different, and sometimes opposing, biological effects (47). In keeping with this, we observed entirely different responses to the inactivation of HIF-1α and HIF-2α in adipose tissues. Although inactivation of HIF-2α did not result in altered adipose mass under the experimental conditions used, this does not preclude concurrent activation of HIF-2α contributing to the phenotype associated with Phd2 deficiency. Indeed, HIF-2α is most likely involved in the proangiogenic response of aP2-Phd2KO mice. It is also possible that quantitative differences in the extent of HIF pathway activation underlie the differences in adipose phenotypes, as has been observed in the heart (45,48) and pancreatic β-cell (49,50), where pronounced induction of HIF is associated with organ dysfunction, but modest induction has beneficial effects on metabolic function or protection from ischemia.

Our work reveals a novel role for the oxygen sensor PHD2 in the regulation of adipocyte lipolysis. PHD2 inhibition suppresses lipolysis and promotes angiogenic responses through distinct HIF isoform activation, thereby promoting benign adipose tissue expansion. Notably, PHIs similar to those used in this study are under clinical trials for anemia and ischemia (33,34). Although the effects of PHD inhibition in metabolic disease remain to be tested, our data suggest that selective PHD2 inhibition may also be beneficial for ameliorating the detrimental metabolic consequences of elevated fatty acid levels found in insulin-resistant obesity and lipodystrophic dyslipidemia.

Acknowledgments. The authors thank Tammie Bishop and Ya-Min Tian (Ratcliffe Laboratory, Oxford, U.K.) for advice on the use of PHIs; Xantong Zou, Rhona Aird, and Karen French for excellent technical assistance; and the staff in the biomedical research resources facility (Edinburgh) for maintenance of the mouse colony. The authors also thank Prof. Nicholas Hastie, Prof. Stewart Forbes, and Prof. John Iredale (Edinburgh) for useful discussions.

Funding. This work was supported by a Sir Henry Wellcome Postdoctoral Fellowship (to Z.M., 085458/Z/08/Z) and a British Heart Foundation/University of Edinburgh Centre of Research Excellence Transition Award Fellowship (to Z.M.).

Duality of Interest. C.J.S. and P.J.R. are scientific cofounders and hold equity in ReOx Ltd., a company that is seeking to develop HIF hydroxylase inhibitors for therapeutic use. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. Z.M. designed, performed experiments, analyzed data, and wrote the manuscript. N.M.M. designed, performed experiments, contributed to discussion, and reviewed and edited the manuscript. J.M.M.N. and J.M.F.-R. designed, performed, and analyzed the human diabetic cohort gene expression study. C.C.W. and K.J.S. provided the human adipose biopsies. C.J.S. provided essential reagents and reviewed and edited the manuscript. J.R.S. contributed to discussion and reviewed and edited the manuscript. P.J.R. provided the Phd2 homofloxed mice, contributed to discussion, and reviewed and edited the manuscript. Z.M. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.