Lysosomal Stress in Obese Adipose Tissue Macrophages Contributes to MITF-Dependent Gpnmb Induction

Obesity is characterized by a state of low-grade inflammation, which is an essential contributing factor to insulin resistance (IR) (1). Both adaptive and innate immunity contribute to this inflammatory response. Approximately a decade ago, increased macrophage numbers were first described in obese adipose tissue (AT) (2). In recent years, virtually all cells of the immune system have been described in AT. In lean AT, a crucial role is played by eosinophils, innate lymphoid type 2 cells, invariant natural killer T cells, and regulatory T cells, which promote an anti-inflammatory environment that warrants insulin sensitivity (3–6). During obesity, AT becomes populated by proinflammatory Th1 T cells (both CD4⁺ and CD8⁺), neutrophils, mast cells, and B cells, promoting IR (7–11). During diet-induced obesity (DIO), a phenotypical change occurs in AT macrophages (ATMs) characterized by a shift toward a more proinflammatory state (12,13). One typical characteristic of inflamed obese AT is the presence of crown-like structures (CLSs), which consist, in part, of actively recruited proinflammatory ATMs, surrounding dead adipocytes (2,14). Another characteristic of inflamed obese ATMs is lipid accumulation, as these cells must buffer the fat spillover from dying adipocytes (15,16).

Glycoprotein nonmetastatic melanoma protein B (Gpnmb) is a transmembrane protein expressed by several cell types, including macrophages (17). It is also referred to as osteoactivin, hematopoietic growth factor–inducible neurokinin 1, and the dendritic cell–associated heparin sulfate proteoglycan–dependent integrin ligand (18–21). Gpnmb has been implicated in regulation of both adaptive and innate immunity (22–24). Interestingly, Gpnmb induction has also been observed in a mouse model of Gaucher disease, a macrophage lipid–storage disease characterized by deficient lysosomal acid β-glucosidase and IR in humans (25,26). Presumably, in this case, Gpnmb induction is triggered by the lysosomal accumulation of nondegradable glycosphingolipids in macrophages. Recently, the lysosome regained scientific interest as it was discovered that the transcription factor EB (TFEB) plays a crucial role in driving lysosomal biogenesis (27). TFEB belongs to the microphthalmia transcription factor (MITF) E subfamily of basic helix-loop-helix leucine zipper transcription factors to which MITF, transcription factor E3, and transcription factor EC also belong (28).

Given the insight that Gpnmb seemingly tracts with two key characteristics of obese settings, immune function and lipid exposure, we first investigated a possible link between Gpnmb expression and metabolic-related traits using an in silico approach. Next, we analyzed Gpnmb expression in AT of lean versus obese mice (genetically induced and high-fat diet [HFD]–induced models for murine obesity) as well as in human AT compartments. Finally, we addressed the mechanism of transcriptional regulation for Gpnmb expression in ATMs. We report Gpnmb as a novel marker for obesity-induced ATM infiltration and point toward a crucial role for lysosomal stress and MITF in driving part of the ATM phenotype. In addition, we provide evidence that Gpnmb potentiates interleukin (IL)-4–mediated arginase-1 induction.

We investigated two independently generated mouse F2 populations: an F2 intercross between CAST/Ei × C57BL/6J (CTB6F2; n = 408; males and females combined) and an F2 intercross between C57BL/6J × C3H/HeJ both on an apolipoprotein E–null background (BHF2; n = 135; females only) (29,30). We queried GeneNetwork (http://www.genenetwork.org/webqtl/main.py), which is a web-based program archiving these data sets (31), using the following microarray and phenotype data sets: 1) UCLA CTB6/B6CTF2 adipose (2005) mlratio, 2) UCLA BHF2 adipose female mlratio, and 3) CTB6F2 published phenotypes. Published phenotypes from the BHF2 were imported into the database ([30] and http://www.sagebase.org/index.php). The sole Gpnmb probe on each of the adipose databases (record 10024406500) was selected, and either the top ∼500 Gpnmb covariates from each experiment or the correlation to all scored phenotypes were determined. To reduce trait redundancy wherever possible only trait:gene correlations using the normalized or transformed trait expression values are shown. Heatmap transformations of trait:gene correlations were performed using the gplots package available from http://www.r-project.org/.

C57BL/6J control mice and leptin-deficient obese (Ob/Ob) mice (C57BL/6J background) were obtained from Harlan. Animals (n = 6 per group unless stated otherwise) were fed a commercial chow diet (AM-II). Studies were performed using 12-week-old male mice.

For DIO studies, 6-week-old male C57Bl/6J mice were fed either an HFD (D12492, Research Diets Inc.; consisting of 20 kcal % protein, 20% carbohydrates, and 60% fat) or low-fat diet (LFD; D12450B, Research Diets Inc.; consisting of 20 kcal % protein, 70% carbohydrates, and 10% fat) for 16 weeks. Principles of laboratory animal care were followed, and approval for the study was obtained from the local ethical committee for animal experiments.

Epididymal white AT (EWAT) of lean, Ob/Ob, and LFD- and HFD-fed mice was dissected, minced, and exposed to collagenase II (Sigma)/DNase I (Roche) treatment for 25 min. The cell suspension was passed through a 70 μm (lean) or 100 μm (obese) cell strainer (BD Falcon) and washed in Hanks’ balanced salt solution/2% FCS. The obtained pellet is referred to as the stromal vascular fraction (SVF). Erythrocytes were removed from the SVF using red cell erythrocyte lysis buffer (Sigma).

SVF were incubated with rat anti-mouse CD16/CD32 mouse Fc–blocking reagent (BD Biosciences) for 10 min and subsequently double labeled with rat anti-mouse macrophage galactose–specific lectin/CD301 Alexa 647 (Bio-Connect) and rat anti-mouse F4/80 fluorescein isothiocyanate (eBioscience). From lean and LFD SVFs, CD301⁺/F4/80⁺ cells were sorted, and from Ob/Ob and HFD-fed mice SVFs, CD301⁺/F4/80⁺ and CD301^lo/F4/80⁺ cells were sorted using a FACSAria instrument (BD Biosciences). Positive selection of ATMs from SVF was performed with anti-CD11b conjugated magnetic microbeads (Miltenyi Biotec) using MACS according to the manufacturer’s protocol.

Twenty obese (BMI ranging from 34.8 to 61.3 kg/m²) Caucasian women with a mean age of 40.5 (26–50) years who were scheduled for Roux-en-Y gastric bypass surgery were included. Patients were eligible if they had no Diagnostic and Statistical Manual of Mental Disorders IV diagnosis, were older than 18 years, understood the objective of the study, and gave informed consent. Biopsies of the subcutaneous (SC), omental (OM), and mesenteric (MES) fat compartment were taken at the beginning of the surgical procedure after an overnight fast. The control group consisted of age- and sex-matched female subjects scheduled for elective cholecystectomy. They were lean (BMI 20–24.4 kg/m²) and had a normal glucose tolerance test, HOMA-IR, and HbA_1c. The samples were snap-frozen in liquid nitrogen and thereafter stored at −80°C for subsequent analysis. The women participated in a study on the short-term metabolic effects of bariatric surgery. The study was approved by the Medical Ethical Committee of the Academic Medical Center.

Total RNA was extracted from total fat-, SVF-, and FACS-sorted ATMs using TRIzol (Invitrogen) reagent and the NucleoSpin II extraction kit, which included an RNase-free DNase step (Macherey Nagel). RNA concentrations were measured using the NanoDrop spectrophotometer (NanoDrop Technologies). Equal amounts of RNA were used to synthesize cDNA according to the manufacturer’s method (Invitrogen). Gene-specific analysis was done by real-time RT-PCR using an iCycler MyiQ system (Bio-Rad). Expression levels in mouse and human were normalized to those of acidic ribosomal phosphoprotein 36B4, also referred to as P0.

RAW264.7 from American Type Culture Collection were cultured in DMEM/10% FCS supplemented with penicillin-streptomycin. Palmitate and oleate (Sigma) were coupled to BSA Fraction V fatty acid–free (Roche) as described previously (32). IL-4 (R&D Systems), IL-10 (PeproTech), and interferon-γ (IFN-γ; PeproTech) were used at 50 ng/mL, and lipopolysaccharide (LPS) from Salmonella minnesota R 595 (Alexa) was used at 100 ng/mL. BSA-coupled lipids were used at 500 µmol/L; chloroquine (CQ) at 40 µmol/L; bafilomycin at 50 nmol/L; concanamycin A at 2 nmol/L; tunicamycin at 1, 5, and 10 μg/mL; thapsigargin at 1, 10, and 100 nmol/L; and Torin at 250 nmol/L. β-Hexosaminidase activity was determined in cell-free supernatant using 4MU-β-D-6-sulpho-2-acetamido-2-deoxy-glucopyranoside as substrate (Sigma). MITF knockdown in RAW264.7 cells was achieved using a lentiviral approach. The short hairpin RNA (shRNA) containing pKLO.1 puro lentiviral vectors were derived from the Department of Human Genetics of the Academic Medical Center. High-titer lentiviral stocks were generated by calcium phosphate–mediated transfection of the purified plasmids and packaging vectors pMDL/pRRE, pRSV-Rev, and pMD2.VSVG into HEK-293T cells. Supernatants were collected 48 and 72 h posttransfection and subjected to ultracentrifugation to obtain concentrated viral stocks. Titers were determined with p24 ELISA and by means of real-time PCR using primers directed at proviral DNA in genomic host-cell DNA. The following shRNA-encoding sequences targeting murine Mitf were used in this study: Mitf shRNA#1 TRCN0000095285 (5′-GCAGTACCTTTCTACCACTTT-3′); Mitf shRNA#2 TRCN0000095288 (5′-GCAAATACGTTACCCGTCTCT-3′); and Gpnmb shRNA TRCN0000011929 (5′-CCGAATAAAC AGATATGGCTA-3′). The noncoding MISSION Non-Target Control Vector (SHC002, 5′-CAACAAGATGAAGAGCACCAA-3′) served as an internal control in our experiments. Raw264.7 cells were seeded at 40% confluency and the following day exposed to optimized lentiviral titers for 4 h. After 24 h, transduced cells were selected with 5 µg/mL puromycin for 2 days, and experiments were performed immediately thereafter. Mouse plasma Gpnmb levels were determined by ELISA (R&D Systems).

EWAT was fixed in 4% formaldehyde, phosphate-buffered at pH 7.0, and embedded in paraffin. Deparaffinized tissue sections (4 μm) were blocked for endogenous peroxidase activity by immersion in 0.3% H₂O₂ in methanol for 20 min. For immunostaining, sections were incubated with primary goat IgG anti-mouse Gpnmb antigen affinity-purified polyclonal antibody (AF2330; R&D Systems) diluted in phosphate-buffered primary antibody diluent (ScyTek Laboratories) followed by rabbit anti-goat IgG (Dako) and horseradish peroxidase–conjugated swine anti-rabbit IgG (Dako). Bound horseradish peroxidase activity was visualized using diaminobenzidine as substrate. Sections were counterstained with either hematoxylin or methyl green and were mounted in Pertex. RAW264.7 were fixed in 4% formaldehyde and permeabilized in 0.2% Tween/PBS followed by endogenous peroxidase inactivation and staining with anti-MITF antibody (Exalpha Biologicals Inc.).

Cell lysates were prepared in radioimmunoprecipitation assay buffer (150 mmol/L NaCl, 50 mmol/L Tris-HCl pH 7.4, 2 mmol/L EDTA, 0.5% deoxycholaat, 1 mmol/L Na₃VO₄, 20 mmol/L NaF, and 0.5% Triton X-100) supplemented with protease inhibitor cocktail (Roche) and phenylmethylsulfonyl fluoride (PMSF). Primary antibodies used were anti-Gpnmb (R&D Systems), anti-tubulin-α (Cedarlane Laboratories Limited), anti-MITF (Exalpha Biologicals Inc.), anti-lamin A/B (Santa Cruz Biotechnology Inc.), anti-4E-BP1 (Cell Signaling Technology Inc.), anti-phospho-4E-BP-1 (thr37/46) (Cell Signaling Technology Inc.), anti-gapdh (Cell Signaling Technology Inc.), and matching secondary IRDye-conjugated antibodies (Westburg BV) were used for detection in an Odyssey version 3.0 (LI-COR Inc.).

Cytosolic and nuclear fractions were prepared as follows. Cells pellets were washed in PBS and taken up in 500 μL buffer (25 mmol/L HEPES, 5 mmol/L KCl, 0.5 mmol/L MgCl₂, 1 mmol/L dithiothreitol, and 1 mmol/L PMSF) and allowed to swell on ice for 15 min. Next, 200 μL buffer containing 0.35% Triton X-100 was added and after 15 min rotation centrifuged for 30 s at maximum speed. Supernatant contains cytosolic fraction. Pellets were resuspended in 25 mmol/L HEPES, 350 mmol/L NaCl, 10% sucrose, 0.1% Triton X-100, 1 mmol/L dithiothreitol, and 1 mmol/L PMSF and rotated for 1 h; centrifuge and supernatant contains nuclear proteins.

Values presented in figures represent means ± SEM. Statistical analysis of the two groups was assessed by Student t test (two tailed). Level of significance is depicted in the figures with the actual P values. P values <0.05 were considered significant unless otherwise stated.

Inbred mouse strains display significant inherent genetic diversity affecting their susceptibility to various diseases, including aspects of metabolic syndrome. F2 intercrosses of such inbred mice consequently produce genotypically distinct offspring randomly segregating for these genetic and varying disease susceptibilities. As a result, these populations of mice provide an experimental model for correlating phenotypic trait expression to generate biological insight. In our in silico analysis, we used data from two independent F2 intercrosses, namely, the CTB6F2 and the BHF2 (29,30,33).

We determined the relationship between the adipose Gpnmb gene expression levels and that of various metabolic traits scored in the F2 individuals. A Heatmap transformation of the correlation coefficients is shown in Fig. 1A. The most striking observation was the positive correlation of Gpnmb to body weight (r = 0.66 and ρ = 0.43; P value <5.0 × 10⁻¹⁰) in both the BHF2 and CTB6F2 crosses, respectively. In contrast, Gpnmb expression was most significantly inversely correlated to the ratio of glucose to insulin (r = −0.47 and ρ = −0.25; P value <5.0 × 10⁻⁵) in both the BHF2 and CTB6F2 crosses, respectively. An example of the correlation plot for the BHF2 cross is shown in Fig. 1B. The results of this correlation analysis strongly suggest a link between the expression of Gpnmb in AT and markers of insulin sensitivity.

To extend and validate the observations of Gpnmb in the mouse F2 populations, we investigated Gpnmb expression in two independent models for murine obesity, namely, the high-fat DIO model and the genetically obese mouse model (Ob/Ob). By using quantitative PCR (qPCR), a dramatic increase of Gpnmb mRNA was observed in EWAT of 12-week-old Ob/Ob mice and 16-week-old HFD-fed DIO mice when compared with lean or LFD control groups (all n = 6 per group) (Fig. 2A). Active recruitment of inflammatory macrophages into AT results in the formation of CLSs, a hallmark of AT inflammation (2,14). As expected, the macrophage content, which was analyzed by qPCR as F4/80 expression, increased in DIO and Ob/Ob AT when compared with lean and LFD-fed AT (Fig. 2B). Interestingly, Gpnmb expression correlated well with F4/80 expression in total AT (r² = 0.94; P < 0.0001) (Fig. 2C).

Next, we prepared SVF of EWAT to narrow down the cells possibly expressing Gpnmb. A clear enrichment of Gpnmb and F4/80 expression in the SVF of both obese mouse models was observed (n = 6 per group) (Fig. 2D and E). By using anti-Gpnmb antibodies, we demonstrated highly Gpnmb-positive CLSs (Fig. 2F). In male mice, highest expression of both F4/80 and Gpnmb was observed in the EWAT compartment (Fig. 2G and H). We also analyzed other macrophage-rich tissues and only found a modest 10-fold induction in liver on HFD feeding, but not in lung and spleen (Supplementary Fig. 1). In cultured 3T3-L1 adipocytes, Gpnmb was virtually absent and not induced during maturation, whereas both peroxisome proliferator–activated receptor γ2 and adiponectin were induced as anticipated (Fig. 2I). In plasma, Gpnmb levels were not different between the LFD (13 ± 4.6 ng/mL) and HFD (12 ± 3.7 ng/mL), despite the striking increase in EWAT.

Analysis of human lean and obese AT samples, including SC, MES, and OM AT compartments, revealed a significant increase of both CD68 (as a marker of macrophages) and Gpnmb expression in the SC and MES fat of the obese subjects compared with the lean controls (Fig. 2J and K). When combining all AT compartments, we observed a strong correlation between CD68 and Gpnmb (r = 0.68; P < 0.0001). Interestingly, in both SC AT and MES AT, Gpnmb significantly correlated with BMI (r = 0.46 [P = 0.021] and r = 0.44 [P = 0.044], respectively).

During obesity, a shift toward a more proinflammatory phenotype is observed in ATMs (13). Murine ATM populations can for instance be defined by differences in macrophage galactose N-acetyl-galactosamine–specific lectin 1 (MGL1) expression (12). By FACS sorting, F4/80⁺MGL1⁺ fractions were isolated from lean and LFD SVF, representing alternatively activated resident M2 macrophages (Fig. 3A and B). From Ob/Ob and HFD-fed SVF, F4/80⁺MGL1⁺ (M2) macrophages were sorted as well as inflammatory-recruited (M1) macrophages F4/80⁺MGL1^lo (Fig. 3C and D). Using qPCR, we reconfirmed that the sorted cells were indeed F4/80⁺, either positive or negative for MGL1 expression (Fig. 3E and F) and that the inflammatory-recruited (M1) macrophages F4/80⁺MGL1^lo expressed high levels of CD11c (Fig. 3G).

A striking induction of Gpnmb was observed in obese ATMs, both the F4/80⁺MGL1⁺ and F4/80⁺MGL1^lo sorted population (Fig. 3H). Of interest, no correlation was observed with FIZZ1, an alternatively activated macrophage marker, which was highly present in lean ATMs but reduced to a large extent in both ATM populations isolated from Ob/Ob and HFD mice (Fig. 3I). The latter is in line with the lean Th2 adipose environment (eosinophil-derived IL-4), which is lost in obese AT (3). The Gpnmb expression pattern in obese ATMs showed a remarkable similarity with osteopontin expression, a marker previously reported to be induced in ATMs in obese AT (Fig. 3J) (34).

It is known that the development of obesity is associated with both immunological and metabolic changes such as hypoxia and lipotoxicity. To get more insight in the regulation of Gpnmb expression in obese ATMs, several experiments were performed using macrophage-like RAW264.7 cells. During obesity, the inflammatory status of AT switches from a Th2-prone environment (eosinophil-derived IL-4 and regulatory T cell–derived IL-10) toward a Th1 environment (T cell–derived IFN-γ). It was observed that Gpnmb was neither induced by the Th2 cytokines IL-4 and IL-10 nor the Th1-driving factors IFN-γ, LPS, or a combination of IFN-γ with LPS. Analysis of macrophage polarization markers such as arginase-1 (Th2) and nitric oxide synthase (Th1) confirmed that the stimuli worked (Fig. 4A–C). The role of hypoxia as a possible inducer of Gpnmb expression was studied by using the hypoxia mimetic cobalt chloride (CoCl₂) at 100 μmol/L. As a reflection of the hypoxic status of the cells, an increase of GLUT-1 (fourfold) was observed. In contrast, we observed that Gpnmb expression was reduced, suggesting hypoxia does not control Gpnmb expression (Fig. 4D). Acute endoplasmic reticulum (ER) stress as a possible Gpnmb inducer was explored by using thapsigargin (1–100 nmol/L) and tunicamycin (1–10 μg/mL), but neither induced Gpnmb, whereas the ER stress marker CHOP was clearly dose-dependently induced (Fig. 4E).

During the development of obesity, adipocyte dysfunction occurs and macrophages, among others, increasingly scavenge the adipocyte-derived lipids. This lipotoxic environment was mimicked by loading RAW cells with BSA-coupled palmitate. Palmitate condensation to serine via the action of serine palmitoyltransferase initiates sphingolipid synthesis. Indeed after 7 h of lipid challenge, increased formation of dihydroceramide (10-fold) was detected when compared with BSA alone or oleate (Fig. 5A). Interestingly, Gpnmb was strongly induced by the palmitate treatment, whereas oleate did not affect Gpnmb expression. At the level of protein, the effect of palmitate was confirmed by immunoblot analysis (Fig. 5B and C). To address if the effect was caused by palmitate itself or a newly synthesized sphingolipid, we used myriocin, which inhibits serine palmitoyltransferase. We observed that myriocin reduced de novo synthesis of sphingolipids, whereas it did not lower Gpnmb protein, suggesting that Gpnmb induction was not caused by a newly derived sphingolipid (Fig. 5D and E). Given the overexpression of Gpnmb in mouse models of Gaucher disease, a lysosomal storage disorder, we studied the potential impact of lysosomal stress on Gpnmb expression (25). To address if lysosomal dysfunction could account for adipose Gpnmb induction, RAW cells were treated for 24 h with the lysosomotropic agent CQ. Targeting of lysosomes was confirmed by analysis of β-hexosaminidase activity in the medium of cells, which increased fourfold. Interestingly, we observed a striking induction of Gpnmb at both the mRNA and the protein level (Fig. 5F–H). When using the lysosomal vacuolar-type H⁺-ATPase inhibitors bafilomycin and concanamycin A, we also observed a potent induction of Gpnmb (Supplementary Fig. 2).

What regulates Gpnmb expression in ATMs is unknown. The association with lysosomal stress—which we define as a dysfunction of lysosomes, which can be reached by impairment of lysosomal pH regulation or through subtle overloading with lipids—and suggested connections in literature prompted us to address if MITF controlled Gpnmb induction in obese ATMs. Interestingly, MITF gene expression was increased in total EWAT of obese mice when compared with lean EWAT (Fig. 6A). Its macrophage-derived origin in total EWAT was strengthened, as MITF expression correlated significantly with the macrophage marker F4/80 (r² = 0.90; P < 0.0001) and with Gpnmb (r² = 0.88; P < 0.0001), hence reflecting ATM load. In obese SVF, MITF was increased at the mRNA and protein level (Fig. 6B and C). In sorted ATMs, MITF mRNA expression levels were not different between studied populations (Fig. 6D). In other words, per macrophage, MITF is not different, but it reflects increased macrophage load in total EWAT during obese conditions. Increased Gpnmb protein in obese ATMs seems not to be accompanied by increased MITF, suggesting that posttranscriptional regulation of MITF can occur during obesity. Recent evidence suggests active regulation of MITF localization from cytosol to the nucleus (35,36). This prompted us to study if palmitate and CQ-mediated induction of Gpnmb also required translocation of MITF to the nucleus. Therefore, RAW264.7 cells were incubated with CQ (40 μmol/L), BSA, or BSA-coupled palmitate. Next, MITF protein was analyzed by immunoblot in whole-cell extracts, cytosol, or nuclear-enriched fractions. A clear shift from the cytosol to nuclei of MITF protein was observed in CQ-pulsed and palmitate-loaded cells (Fig. 6E). This was further confirmed by immunocytochemistry (Fig. 6F).

To ensure that Gpnmb indeed was solely induced in a MITF-dependent manner, we used a lentiviral knockdown strategy targeting MITF. We confirmed efficient MITF knockdown on protein level compared with cells exposed to control virus. Importantly, MITF depletion completely abrogated the induction of Gpnmb following palmitate loading and CQ stimulation (Fig. 6G and H). Interestingly, MITF knockdown also prevented CQ-mediated induction of Ccl3 and osteopontin gene expression (Fig. 6I). Osteopontin was already found to behave as Gpnmb in sorted ATMs (Fig. 3J). The important ATM defining marker CD11c, but also tumor necrosis factor, was not induced by CQ in a MITF-dependent manner (data not shown).

Very recently, it became clear that active mammalian target of rapamycin complex 1 (mTORC1) at the site of the lysosomal membrane sequesters MITF in the cytosol by means of phosphorylation and that inhibition of the kinase activity of mTORC1 resulted in the nuclear localization of MITF (35). mTORC1 is widely considered as the master regulator of metabolism (37). To address if mTORC1 activity regulated Gpnmb induction in RAW cells, we made use of the inhibitor Torin1. Inhibition of the kinase activity of mTORC1 was confirmed by analyzing the phosphorylation of its substrate 4E binding protein-1 (Fig. 7A). At the level of gene expression, a clear induction of Gpnmb was detected, which also occurred at the level of protein (Fig. 7B and C). In agreement, MITF rapidly translocated from the cytosol to the nucleus upon incubation with Torin1 (Fig. 7D and E). Importantly, MITF knockdown also prevented induction of Gpnmb by Torin1 (Fig. 7F). Of note, rapamycin failed to induce Gpnmb, suggesting autophagy is not the driving factor of Gpnmb induction.

Lastly, we analyzed if CD11b⁺-selected obese ATMs also showed alterations in the mTORC1 pathway. We observed that 4E binding protein-1 phosphorylation was attenuated in obese versus lean ATMs (Fig. 7G and H). Importantly, Gpnmb was strongly induced in obese ATMs, whereas MITF protein levels were unchanged, reflecting a similar amount of both lean and obese macrophages in our analysis.

We earlier (Fig. 4) demonstrated that Gpnmb was not induced by cytokines or LPS. Here we addressed the question if Gpnmb itself could play a functional role in polarizing RAW264.7 cells. To this end, Gpnmb-directed shRNA was used, which efficiently knocked down Gpnmb in RAW cells, both at the level of RNA and protein (Fig. 8A). When basal levels of Gpnmb were knocked down, no difference was observed in the capacity of LPS (6-h stimulation) to induce iNOS when compared with scrambled control cells (Fig. 8A). Interestingly, the response to IL-4, analyzed as arginase-1 expression, was reduced by ∼80% when Gpnmb was knocked down. To address if Gpnmb protein induction could potentiate the IL-4 response, RAW cells were first cultured with CQ or Torin to induce Gpnmb (Fig. 8B) and subsequently pulsed with IL-4 for 6 h. A striking sevenfold to eightfold induction of arginase-1 was observed, which again was suppressed when Gpnmb was targeted by shRNA.

It is well established that the low-grade inflammation that develops during obesity can be attributed to both innate and adaptive immune cells. In lean AT, a crucial role is played by eosinophils, innate lymphoid type 2 cells, invariant natural killer T cells, and regulatory T cells all involved in promoting an anti-inflammatory environment and insulin sensitivity (3–6). Obese AT becomes invaded by inflammation-promoting Th1 T cells (both CD4⁺ and CD8⁺), neutrophils, mast cells, and B cells, which leads to IR (7–11). A hallmark of inflamed obese AT is the increased recruitment of macrophages, which are organized in so-called CLSs, consisting of inflammatory macrophages surrounding dysfunctional adipocytes (2,14). Given their pathophysiological role in metabolic disease, characterization of these macrophages and identification of markers is important. Here we report the exclusive and massive expression of Gpnmb in obese ATMs. Gpnmb expression was shown to reflect MITF activation as a response to lipid and lysosomal stress.

The evidence for this was obtained by experiments with cultured RAW264.7 cells. Increased lipid pressure on macrophages, caused by spillover from adipocytes occurring during the development of obesity, could be mimicked by loading RAW cells with palmitate. Importantly, we found that Gpnmb was induced by palmitate. Increased lipid load in lysosomes has already been connected to Gpnmb induction in a mouse model of Gaucher disease (25). In these studies, it has been postulated that Gpnmb may serve as a biomarker for Gaucher disease. In obese mice, we were not able to demonstrate increased levels of Gpnmb in the circulation, despite the several hundred–fold induction in EWAT. Additional studies in human obese subjects are still worthwhile, as in our human cohort, only a modest induction of Gpnmb was detected and CLSs were only found in limited numbers. A possible role for lysosomal dysfunction was further demonstrated, as it was found that CQ very potently triggered Gpnmb induction. Our lysosome-centered view was further strengthened, as neither mimicking of a hypoxic environment nor culture with cytokines nor acute induction of ER stress induced Gpnmb. We realize that chronic lysosomal stress also impacts on other cellular compartments. It therefore cannot be excluded that such later events sustain Gpnmb expression. We postulate a role for the lysosome in driving part of the ATM phenotype. In this context, it is important to note that obesity activates a program of lysosomal-dependent lipid metabolism in ATMs independently of classic activation (38). Interestingly, in our in silico KEGG pathway enrichment analysis, it was also found that Gpnmb expression was linked to the lysosome in both the BHF2 and the CTB6F2 crosses (data not shown).

Recently, it was demonstrated that lysosomal genes display coordinated transcriptional regulation with an important role for TFEB in lysosomal biogeneses (27,39,40). MITF belongs to the same family of transcription factors as TFEB, and we found MITF to be essential for Gpnmb induction in the macrophage cell line RAW264.7, as MITF knockdown blunted induction by the tested stimuli. Importantly, osteopontin and Ccl3, but not tumor necrosis factor and CD11c, were induced by CQ in an MITF-dependent manner, pointing toward a crucial role of MITF in driving the ATM phenotype under obese conditions. In FACS-sorted ATMs, Gpnmb followed osteopontin expression, which has been described to be involved in potentiating Mcp1/Ccl2-mediated monocyte chemoattraction (34).

Finally, we found that mTORC1 inhibition also induced Gpnmb expression in an MITF-dependent manner. mTORC1 is a central regulator of cellular metabolism, and it has been shown recently that its activation occurs at the site of the lysosomal membrane (37,41). Our data propose a unifying model in which lysosomal stress (palmitate and CQ treatment) and perturbed signaling cascades (mTORC1 inhibition) could lead to MITF translocation to the nucleus with subsequent induction of Gpnmb. The observations that in RAW264.7 cells, endogenous MITF shifts from the cytosol to nucleus upon mTORC1 inhibition and CQ treatment is in agreement with recent MITF overexpression studies (35,36). Importantly, our in vitro studies on MITF and Gpnmb induction were confirmed in vivo. It was found that during obesity, phosphorylation of the downstream target of mTORC1, 4E-BP-1, was reduced. This causes MITF to shift from the cytosol to nuclei in ATMs, and consequently, Gpnmb induction occurred. Similar MITF-dependent regulation of Gpnmb was recently reported in other cell types like melanoblasts and osteoclasts (42–45).

Thus far, it is unknown what the function of Gpnmb is in obesity. It has been suggested that Gpnmb is involved in controlling immune responses. In the context of adaptive immunity, Gpnmb has been described to repress T-cell activation (22). Additional evidence comes from studies in mice homozygous for a spontaneous mutation within the Gpnmb locus causing a truncated form of Gpnmb. These mice exhibit autoimmune pigmentary glaucoma and compromising ocular immunosuppression, which is manifested by deficient anterior chamber–associated immune deviation (18,46). In the context of innate immunity, Gpnmb was demonstrated to be involved in recognition of dermatophytic fungi (24). Overexpression of Gpnmb dampens LPS responses, suggesting that Gpnmb negatively regulates inflammatory macrophage responses (17). We now provide evidence that Gpnmb potentiates IL-4–mediated induction of the alternative macrophage marker arginase-1. Ablation of Gpnmb drastically reduces IL-4–driven arginase-1 induction. Possibly, adipose Gpnmb is involved in preventing inflammation from derailing completely, thus contributing to the counterinflammation phase, which was recently proposed (47). Alternatively, Gpnmb is involved in enhancing the phagocytic/remodeling capacity of local ATMs. Recently, a connection was made with repair after kidney tissue damage and regulation of required phagocytosis (48). It is important to note in this context that arginase-1 is an important mediator of tissue repair and wound healing (49). In follow-up experiments, we plan to study in mice with macrophage-specific inactivation of Gpnmb the physiological impact in these macrophages. At present, these appropriate animals are not available. These are required to elucidate whether Gpnmb has a causal role in lipid handling in macrophages and whether this protein controls the inflammatory status of macrophages upon lysosomal overload.

Gpnmb has been implicated in several disease processes, including cancer, glaucoma, and renal damage and repair (18,44,46,48,50). Here we have provided evidence that Gpnmb is linked to obesity-driven AT inflammation as well.

In conclusion, we propose that adipose lipid spillover provokes continuous pressure on the lysosomal compartment of resident macrophages, with consequential aberrations in signaling cascades closely linked to lysosome function, such as mTORC1. Active mTORC1 normally sequesters MITF in the cytosol, but as a result of mTORC1 inhibition, MITF translocates to the nucleus where it initiates a transcriptional program including Gpnmb and hence impacts on the ATM phenotype.

Acknowledgments. The authors thank Dr. N. Zelcer from the Department of Medical Biochemistry, University of Amsterdam, for critical reading of the manuscript.

Funding. This work was supported by the Dutch Diabetes Foundation (grant number 2009.80.016).

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. T.L.G. and M.J.T. researched data and reviewed/edited the manuscript. R.O., C.v.R., J.A., N.C., B.H., B.d.W., M.J.S., C.A., and L.v.E. researched data. J.M.F.G.A. contributed to discussion and reviewed/edited the manuscript. M.v.E. wrote the manuscript and researched data. M.v.E. 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.