Heme Oxygenase-1 Drives Metaflammation and Insulin Resistance in Mouse and Man

HO-1 Expression Predicts Insulin Resistance in Mouse and Man

Over the last years we collected a unique resource of matched serum, liver, and adipose biopsies from well-characterized individuals for studying “healthy” versus “diseased” obesity. Recruited subjects were required to be clinically obese (BMI > 30), to exhibit normal fasting glucose levels, to have no overt inflammation, and to be free of any medications. Critically, whereas obese insulin-resistant (obIR) individuals exhibited clearly elevated insulin resistance (HOMA-IR ≥ 5), obese insulin-sensitive (obIS) individuals were required to show no signs of systemic insulin resistance, i.e., HOMA-IR ≤ 2. We thus biopsied 17 obIS and 27 obIR individuals well-matched for age and BMI (Table 1) and complemented the samples with biopsies from 6 nonobese individuals. Consistent with their insulin resistance, the obIR group displayed minimally elevated waist circumference, visceral adiposity, triglyceride to HDL ratio, as well as GLUT4 and HSD11B1 mRNA expression by qPCR (Table 1). Thus, we assembled a clinical resource to probe early determinants of insulin resistance in obesity.

Interestingly, and in contrast to our initial expectations, we found elevated HO-1 expression in obIR individuals in both the liver and visceral fat compartments (Figures 1A and 1B). HO-1 levels correlated directly with metabolic dysregulation, increasing progressively from lean to obIS and through to obIR states (Figures 1A and 1B). Importantly, regression analyses including seven “intermediate” insulin resistance subjects (HOMA-IR > 2 and < 5) revealed that liver and adipose HO-1 expression explain 21.7% and 43.3% of HOMA-IR, relationships maintained in both males and females, and independent of waist circumference or visceral fat (Tables S1 and S2 available online). These findings were confirmed on the protein level, and importantly, placed HO-1 within both the hepatocyte and Kupffer cell compartments in liver and in macrophages in adipose (Figures 1C and 1D). Of note, HO-1 protein was elevated in both hepatic compartments in obIR individuals (Figures 1C and 1E). In adipose, HO-1 was particularly evident in inflammatory macrophage crown-like structures (Figure 1D). Thus, liver and adipose HO-1 are potent independent predictors of insulin resistance in man.

Since a proinflammatory role for HO-1 contradicted bulk literature on the enzyme, we performed an unbiased search for data sets to validate our findings. We identified five data sets, two in liver and three in adipose, comparing obese and/or insulin-resistant adult humans with controls, and mined them for any pattern of HO-1 expression. Impressively, HO-1 was increased in the metabolically compromised setting in five of five studies. HO-1 was increased in livers of obese versus lean individuals (Figure 1F) (Pihlajamäki et al., 2009) and of diabetic versus nondiabetic subjects (Figure 1G) (Misu et al., 2010). In adipose, HO-1 was elevated in insulin-resistant obese subjects (Figure 1H) (NCBI, GEO data set GDS3665). In a rare opportunity to control for genetic background in the human setting, we mined adipose expression profiles from monozygotic twins discordant for BMI and observed HO-1 again to be consistently higher in the obese sibling (Figure 1I) (Pietiläinen et al., 2008). Finally, in a landmark survey of adipose from 387 morbidly obese individuals, HO-1 scored 3^rd of all insulin-resistance risk genes (Figure 1J) (Qatanani et al., 2013). Thus, liver and adipose tissue HO-1 expression consistently predict metabolic dysfunction in humans.

To test conservation, we set out to generate a parallel data set in mice. Similar to humans, mice exhibit substantial HO-1 expression in the lung, kidney, and spleen and, interestingly, in metabolic organs including the liver and white adipose tissue (Figure S1A). To explore mouse HO-1 variation with metabolic disease, we generated a large cohort of high-fat-diet (HFD)-treated male C57BL/6J mice, stratified them according to glucose (Figure S1B) and insulin tolerance (Figure S1C) and selected subgroups exhibiting either insulin sensitivity (IS) or insulin resistance (IR) despite comparable weight gain on the HFD regimen (Figure S1D). Of note, and in complete agreement with our human findings, HO-1 expression in liver (Figure 1K) and adipose (Figure 1L) was directly proportional to severity of metabolic dysregulation. Again, measures were confirmed on the protein level, in both liver (Figure 1M) and epididymal adipose tissue (Figure 1N). Importantly, the obese-IR and -IS groups did not differ in body weight (Figure S1D), and were matched for age and developmental environment. Thus, hepatic and adipose tissue HO-1 predict insulin resistance in mouse and man.

Hepatocyte HO-1 Knockout Mice Are Insulin Hypersensitive

To directly assess the role of HO-1 in the etiology of metabolic disease, we generated mice bearing a conditional allele for HO-1 (HO-1^fl/fl) (Figures S2A and S2B). First, we crossed them to Albumin-Cre transgenic mice (Alb-Cre) to obtain liver (hepatocyte)-specific HO-1 knockouts (Lhoko). Efficient deletion was confirmed on the DNA, protein and activity levels (Figures 2A, 2B, and S2C). Functional deletion of HO-1 was confirmed in vivo by examining HO-1 expression in whole livers upon hemin injection, a heme analog known to induce HO-1 expression and activity (Figure 2C). Thus, we generated hepatocyte-specific HO-1 mutant mice.

Lhoko mice were born at Mendelian ratios, developed normally, and exhibited normal liver morphology as well as serum parameters (Figures S2D and S2E). Importantly, Lhoko animals exhibited completely normal glucose and insulin tolerance relative to their littermate controls (Figures S2F and S2G), showing no evidence of HO-1 imposed anti-inflammatory tone in the naive state. Remembering that our observations of disease predicting HO-1 were made in metabolically stressed individuals, we challenged Lhoko mice with HFD. With respect to adiposity, Lhoko mice showed no difference in HFD-induced body weight gain relative to littermates (Figure S2H). Similarly, they maintained unremarkable oral glucose tolerance (Figure 2D). Intriguingly though, insulin tolerance, fasting insulin, and HOMA-IR all indicated significantly increased insulin sensitivity in the HFD-treated Lhoko animals (Figures 2D and 2E), a clear indication of improved metabolic health. In agreement with these findings, serum measures of hepatotoxicity including alanine transaminase (ALAT), aspartate transaminase (ASAT), and ASAT/ALAT ratios were clearly improved (Figures 2F). To understand the insulin sensitivity on a cellular level, an insulin bolus was administered directly into the portal vein of HFD-treated Lhoko and control animals, and liver biopsies taken over 10 min. Intriguingly, loss of HO-1 induced stark augmentation and acceleration of key upstream insulin signaling activation events including insulin receptor (INSR) and AKT phosphorylation (Figure 2G). The effects were highly reproducible, and evident within 2 min of injection. Indicating more widespread positive effects of hepatic HO-1 deletion, examinations of muscle and adipose responses in the same animals revealed moderately elevated pAKT (Figure 2G). In keeping with these improvements in metabolic function, oil red O staining and biochemical measures revealed attenuated liver steatosis in HFD-treated Lhoko animals (Figures 2H and 2I). Gluconeogenic and lipogenic programs were largely unaltered (Figure S2I). Together, these findings comprise direct genetic evidence that loss of hepatocyte HO-1 in vivo induces proximal insulin signaling hypersensitivity and resistance to HFD-induced metabolic sequelae.

To corroborate these data, we also performed the opposite genetic experiment, acutely overexpressing HO-1 via tail vein injection of adenoviral HO-1 (Figures S2J and S2K). Critically, and in direct support of the human and Lhoko data, 1 week of ~5-fold HO-1 overexpression in unchallenged chow-fed C57BL/6J mice triggered glucose intolerance, insulin resistance, and elevated HOMA-IR relative to Adeno-LacZ injected controls (Figures 2J, 2K, and S2L). Thus, genetic gain- and loss-of-function studies in vivo indicate that hepatocyte HO-1 exacerbates insulin resistance, steatosis, liver damage and metabolic disease.

Macrophage HO-1 Knockout Mice Resist Metabolic Disease

Considering the observed increases in macrophage HO-1 in insulin-resistant liver and adipose in both mouse and man, and the central role of macrophages in metaflammation, we next tested the contribution of macrophage HO-1 toward insulin resistance. We crossed our conditional HO-1 mice to macrophage deleting LysM-Cre transgenic animals, creating macrophage HO-1 knockout (Macho) mice. Equal and efficient deletion was confirmed in naive, as well as lipopolysaccharide (LPS)/interferon (IFN)γ-driven proinflammatory (M1) and interleukin (IL)-4/IL-13-driven anti-inflammatory (M2) bone-marrow-derived macrophages (BMDMs; Figures 3A–3C). Similar to Lhoko animals, Macho mice were born at Mendelian ratios, developed normally, exhibited normal white blood cell counts, and appeared robust and healthy (Figures S3A and S3B and data not shown). Importantly, metabolic profiling was equally unremarkable in the unchallenged state with no detectable alterations in body weight, food intake, glucose tolerance, insulin sensitivity, or plasma-free fatty acid levels (Figures S3A–S3F). Thus, macrophage HO-1 is dispensable for normal growth, development, and metabolic homeostasis.

As mentioned above, bulk literature would have predicted, if anything, enhanced inflammation upon HO-1 deletion. In an attempt to unmask such a phenotype, we again turned to HFD. Intriguingly, and again in contrast to expectations, we observed marked increases in glucose clearance and insulin sensitivity in Macho animals after a 16 week HFD regimen (Figures 3D and 3E). Indeed, even body weight gain was slightly reduced (Figure 3D)—despite no measurable change in food intake (Figure S3G). Importantly, all aspects of improved glucose homeostasis were evident before body weight differences emerged (data not shown) and, independently, when comparing body-weight-matched animals at the experiment’s end (Figure 3F). Interestingly, the metabolic improvements elicited by macrophage HO-1 loss were systemic. Both liver morphology and histology revealed clear reductions in hepatic steatosis in Macho animals (Figure 3G), improvements confirmed by reduced measures of lipid content and liver toxicity, as well as lipogenic and gluconeogenic profiles (Figures 3H, 3I, and S3H). Specifically, expression of the lipogenic factors Pparg, Srebf1, Cd36, and Fasn were reduced in HFD-Macho livers as well as G6pc, which mediates hepatic glucose output (Figure 3H). Expression of inflammatory Tnf, Il6, and Il1b were equally reduced (Figure 3H), a contrast to elevated anti-inflammatory Il10 (Figure 3H). Importantly, HO-1 expression, which we show here to be proinflammatory, was not only reduced in the myeloid compartment, but importantly also in hepatocytes of the Macho mice (Figure 3G), indicating that macrophage HO-1 is necessary for effective transmission of inflammatory status to target tissues. Consistent with our previous findings, liver, muscle, and fat responses to intraportal insulin were significantly enhanced in the HFD-treated Macho mice (Figure S3I). These findings are in direct agreement with our observations that HO-1 expression in both hepatocytes and macrophages of human obese liver predicts disease severity. An aside on the observed body weight difference, Macho animals exhibited a minor increase in energy expenditure (Figure 3J) independent of any altered activity (Figure 3K) or respiratory quotient (Figure S3J). These data are consistent with the observed system-wide improvements in metabolic homeostasis. In all, the data of both mouse models and the human cohort implicate myeloid HO-1 as an effector of metabolic disease. Thus, macrophage HO-1 is necessary for the manifestation of HFD-induced metaflammation.

Reduced Metaflammation and Improved Adipose Tissue Architecture in Macho Mice

Steatosis results in part from fatty acid spill-over resulting from inflammation, insulin resistance, and improper lipolysis. In line with the reduced steatosis, HFD-Macho mice exhibited reduced fasted and fed serum-free fatty acids (Figure 4A). Further, examination of HFD-Macho epididymal fat revealed dramatic improvements in metaflammation. Included were near-absence of crown-like structures (Figure 4B), reduced adipocyte size (Figures 4B and 4C), and a substantial reduction in macrophage infiltration (Figure 4D). Importantly, these healthy metabolic attributes were manifest in the face of clear obesity. Consistent with this “healthy” obesity profile, HFD-Macho animals exhibited clear reductions in the systemic indicators of metaflammation, namely TNF-α, IL-1β, MCP-1, and RANTES (Figure 4E). Adiponectin (Adipoq) and retinol binding protein-4 (Rbp4) mRNA levels were increased in HFD-Macho mouse adipose, whereas leptin (Lep) levels were not affected (Figure 4F). Notably, the Rbp4 mRNA changes were not manifest on the protein level (Figure 4G). Thus, Macho mice exhibit reduced systemic and adipose tissue metaflammation.

The human and mouse data to this point suggested a predominantly myeloid and hepatic role for HO-1 in metaflammation and insulin resistance. We also largely ruled out the contribution of HO-1 in muscle, adipocyte, and pancreatic β-cell compartments, as targeted deletion in these depots did not affect HFD-challenged body weight accumulation, glucose, or insulin tolerance (Figures S4A–S4C).

Next, we asked how HO-1 elicits metaflammation. We purified and FACS analyzed the nonadipocyte stromal vascular fraction (SVF) from epididymal adipose of HFD-treated Macho and control mice. In line with the reduced evidence for metaflammation, HFD-Macho adipose tissue contained a 2-fold reduction in proinflammatory CD11c⁺ and CCR2⁺ macrophage infiltrates, while protective M2-like MGL⁺ macrophage numbers were increased (Figures 4H–4J and Figure S4D). These observations were equally reflected in qPCR measures of Cd68, Itgax/Cd11c and the chemokine receptors Ccr2 and Ccr5 in whole-adipose RNA preparations (Figures S4E and S4F), as well as confocal evidence of improved adipose tissue architecture (Figure 4K). In support of a conserved system in humans, reexamination of our human adipose biopsies revealed an increase in total macrophage number and M1-like marker expression (Table S3). Importantly, HO-1 exhibited a tight positive correlation with M1-like marker expression (TNF, ITGAX/CD11c), whereas M2-like markers showed no correlation (MRC1/CD206) or an inverse relationship (CLEC10A/CD301, CCL18) (Table S4). Importantly, these findings were independent of waist circumference and again maintained after correcting for visceral fat. Thus, macrophage HO-1 is necessary to induce proinflammatory macrophage skewing and disruption of adipose tissue architecture.

HO-1 Loss Induces an Anti-Inflammatory Metabolic Signature and Blunts NF-κB Signaling

Next, to gain insight into the attenuated metaflammatory phenotype, we harvested and in vitro differentiated bone marrow progenitors from unchallenged, chow-fed Macho animals and controls into macrophages (BMDMs). Using TNF-α, we activated BMDMs and examined NF-κB signaling. Using phospho-IkBα as a surrogate for activation, we found a clearly diminished NF-κB response in Macho BMDMs upon induction (Figure 5A). Importantly, this finding was confirmed using more direct assays as well: RelA/p65 nuclear translocation was reduced (Figure 5B), its target-sequence binding by EMSA was reduced (Figure 5C), and NF-κB target gene expression including Nfkbia/IkBα, Tnf, and Nos2/iNos was reduced (Figure 5D). Further, chromatin immunoprecipitation measures of endogenous NF-κB promoter binding at known target genes Nos2 and Tnf also showed clear reductions (Figure 5E). Thus HO-1-deficient BMDMs have blunted NF-κB signaling.

We and others have previously shown that oxidative metabolism is a signature of activated M2-like macrophages, that activated M1-like proinflammatory cells display a more glycolytic metabolic profile, and that direct alterations in metabolism can regulate NF-κB signaling potential (Haschemi et al., 2012; Rodríguez-Prados et al., 2010; Vats et al., 2006). Intriguingly, when assessing these parameters in naive BMDMs, we found clearly elevated basal and spare respiratory capacity, as well as ATP-turnover in the HO-1-deficient cells (Figures 5F and 5G). Finding all three hallmarks of an M2 “aerobic signature” already in non-stimulated Macho BMDMs indicated that HO-1 is required for proper metabolic programming in the most basal cellular state, and that such “pre-”programming may potently influence macrophage skewing and thus whole-body inflammation. Indeed, we observed consistent reductions in TNF-α, IL-1β, and IL-6 secretion in Macho BMDMs relative to controls (Figure 5H). Thus, HO-1 is required for metabolic programming of naive macrophages, and thus, for proper TNF-α- and LPS-induced skewing and activation.

HO-1 Suppresses OxPhos and ROS Signaling

We next asked whether mitochondrial changes might constitute a common molecular link between the liver and myeloid phenotypes (Cheng et al., 2009). Indeed, when repeating our measures on both systems, parallels were evident for essentially all measures. Like Macho BMDMs, primary Lhoko hepatocytes exhibited increased basal, spare as well as maximal respiratory capacity (Figures 6A and S5A). Notably, these effects were independent of any obvious underlying changes in coupling efficiency (Figure S5B), mitochondrial DNA content (Figure S5C), mitochondrial mass (Figure S5D), and respiratory chain stoichiometry (Figure S5E). Interestingly, qPCR analysis revealed clear and widespread upregulation of the entire mitochondrial reactive oxygen species (ROS) detoxification system, from glutathione processing to the mitochondrial-specific superoxide dismutase 2 (Sod2), again in both systems (Figure 6B), indicating that the increased mitochondrial flux was coupled to the induction of ROS quenching machinery. Notably, these changes were confirmed directly, with the findings of increased SOD activity (Figure 6C) and indirectly, with an ~50% increase in intracellular H₂O₂, the product of SOD, and an increase of cytosolic ROS, again in both knockout cell types (Figures 6D and S5F). Recent studies have suggested that mild ROS elevations activate mitochondrial respiration (Mailloux et al., 2013). If ROS balance was altering oxidative capacity (Figure 6A) then mitochondrial function in the knockout cells should be particularly sensitive to antioxidants. In line with this idea, 1 hr preadministration of the potent ROS quencher N-acetyl-cysteine (NAC) reverted HO-1-deficient mitochondrial functional parameters to control levels (Figure 6E). Thus, HO-1 regulates acute ROS thresholding of mitochondrial respiration.

ROS have previously been linked to INSR activation via impairment of protein tyrosine phosphatase nonreceptor type 1 (PTPN1/PTP1B) (Tiganis, 2011) and to nuclear factor-erythroid-derived 2-like (NFE2L2/NRF2) function via impairment of the inhibitory protein kelch-like ECH-associated protein 1 (KEAP1) (Kobayashi et al., 2006). Because activation of INSR and NRF2 alone would explain the majority of phenotypes observed, we tested their activation state in our most naive HO-1 knockout models. Importantly, we found increased nuclear translocation (Figure 6F) and target-binding activity (Figure 6G) of NRF2 in Macho BMDMs and decreased PTP1B activity (Figure 6H) and acute INSR activation (Figure 6I) in naive, chow-fed Lhoko livers. Notably, in vitro reconstitution of HO-1 expression completely reverted mitochondrial respiratory changes (Figure S5H), downstream pINSR increases (Figure S5G), and TNF-α and IL-1β blunting (Figure S5I) on a timescale paralleling the 1–2 days expected for adenoviral HO-1 re-expression. This data set supports a mechanism where HO-1 continuously limits activation of key signaling systems through an acute and ROS-dependent mechanism. Again all key phenotypes were observed in naive unchallenged cells, lending further support to the notion that HO-1 contributes to the “pre”-programming of cellular function.

Human Study Population

Study subjects included 51 obese patients and 6 nonobese controls that underwent weight-reducing surgery or elective surgical procedures such as cholecystectomy. Participants were included if they had fasting plasma glucose levels <7.0 mmol/l, no history of diabetes, no medications, no weight changes >3% during the previous 2 months, C-reactive protein (CRP) levels <20 mg/l, and normal blood leukocyte counts. For further information, see the Extended Experimental Procedures.

Metabolically Healthy and Unhealthy Obese Mice

Age- and body-weight-matched C57BL/6J mice were selected based on their response in glucose and insulin tolerance tests after 16 weeks on HFD.

Glucose and Insulin Tolerance Tests

Glucose and insulin tolerance tests were performed as described in the Extended Experimental Procedures.

Generation of Tissue-Specific HO-1 Knockout Mice

All mouse models are described in the Extended Experimental Procedures. Animals were kept on a 12 hr light/dark cycle with free access to food and water and housed in accordance with international guidelines. Dietary interventions started at 6 weeks of age using diets which contained 10% and 60% calories of fat (Research Diets, D12450B and D12492).

Heme Oxygenase Activity

HO activity was determined as outlined in the Extended Experimental Procedures.

In Vivo Liver Insulin Signaling

For examination of in vivo insulin signaling, mice were fasted for 2 hr, anesthetized and injected with insulin. Liver, muscle and adipose were removed at the indicated times.

Indirect Calorimetry

Indirect calorimetry and activity measurements were performed as reported elsewhere (Pospisilik et al., 2010) and as described in the Extended Experimental Procedures.

Flow Cytometry

SVF cells were obtained by collagenase digestion from epididymal adipose tissue as described (Todoric et al., 2011). FACS analysis was performed as described (Lumeng et al., 2007) and outlined in the Extended Experimental Procedures.

Nuclear Translocation Assay

Naive BMDMs were stimulated with 10 ng/ml TNF-α or medium only for 30 and 60 min, fixed in ice-cold methanol, washed, and stained with anti-RelA/p65 antibody (Santa Cruz Biotechnology). Nuclei were counterstained with DAPI.

Electrophoretic Mobility Shift Assay

Nuclear extracts were prepared from BMDMs as described in the Extended Experimental Procedures. EMSA was performed using the NF-κB EMSA kit (LI-COR Biosciences) according to the manufacturer’s instructions. For supershift assays anti-RelA/p65 (Santa Cruz Biotechnology), anti-p50 (Millipore), and anti-rabbit IgG (Vector Laboratories) was used. Unlabeled double-stranded NF-κB probe was used to test specific binding. Samples were separated on a nondenaturing polyacrylamide gel. NF-κB DNA binding was visualized using the LI-COR Odyssey Imaging System.

Chromatin Immunoprecipitation Assay

ChIP assays were performed using the MAGnify ChIP Kit (Life Technologies) as detailed by the manufacturer’s instruction and as described in the Extended Experimental Procedures.

Oxygen Consumption Assay and Bioenergetic Profile

OCR was measured using the XF24 Flux Analyzer (Seahorse Bioscience) as described (Teperino et al., 2012). For details see Extended Experimental Procedures.

PTP1B Activity Assay

PTP1B activity was determined as outlined in the Extended Experimental Procedures.

Hydrogen Peroxide Measurements

Hydrogen peroxide (H₂O₂) production was measured using the Amplex Red Hydrogen Peroxide/Peroxidase Assay Kit (Invitrogen).

Superoxide Dismutase Assay

Superoxide dismutase (SOD) activity was measured using a commercially available test kit (Cayman) according to manufacturer’s instructions.

Adenovirus Experiments

BMDM rescue experiments were performed using RGD-fiber-modified adenovirus particles (Vector Biolabs). In vivo grade viruses were produced to target livers and isolated primary Lhoko hepatocytes. For details, refer to the Extended Experimental Procedures.

NRF2 DNA-Binding Activity

The NRF2 TransAM ELISA-kit (Active Motif) was used to evaluate NRF2 DNA-binding activity according to the manufacturer’s protocol.

Statistical Analysis

Data are expressed as mean ± standard error of the mean (SEM) unless otherwise specified. Statistical analyses were performed as described in the Extended Experimental Procedures. All reported p values are two-tailed unless stated otherwise. p < 0.05 was considered to indicate statistical significance.

Other Methods

See Extended Experimental Procedures.