Proinsulin-producing, hyperglycemia-induced adipose tissue macrophages underlie insulin resistance in high fat-fed diabetic mice

Abstract

Adipose tissue macrophages (ATMs) play an important role in the pathogenesis of obese type 2 diabetes. High-fat diet (HFD)-induced obesity has been shown to lead to ATM accumulation in rodents; however, the impact of hyperglycemia on ATM dynamics in HFD-fed type 2 diabetic models has not been studied. We previously showed that hyperglycemia induces the appearance of proinsulin (PI)-producing proinflammatory bone marrow (BM)-derived cells (PI-BMDCs) in rodents. We fed a 60% HFD to C57BL6/J mice to produce an obese type 2 diabetes model. Absent in chow-fed animals, PI-BMDCs account for 60% of the ATMs in the type 2 diabetic mice. The PI-ATM subset expresses TNF-α and other inflammatory markers, and is highly enriched within crownlike structures (CLSs). We found that amelioration of hyperglycemia by different hypoglycemic agents forestalled PI-producing ATM accumulation and adipose inflammation in these animals. We developed a diphtheria toxin receptor-based strategy to selectively ablate PI-BMDCs among ATMs. Application of the maneuver in HFD-fed type 2 diabetic mice was found to lead to near total disappearance of complex CLSs and reversal of insulin resistance and hepatosteatosis in these animals. In sum, we have identified a novel ATM subset in type 2 diabetic rodents that underlies systemic insulin resistance.—Buras, E. D., Yang, L., Saha, P., Kim, J., Mehta, P., Yang, Y., Hilsenbeck, S., Kojima, H., Chen, W., Smith, C. W., Chan, L. Proinsulin-producing, hyperglycemia-induced adipose tissue macrophages underlie insulin resistance in high fat-fed diabetic mice.

Obesity and type 2 diabetes are diseases of global metabolic dysregulation and systemic inflammation (1). The intersection of these phenomena is especially evident in the obese visceral adipose tissue, where weight gain is accompanied by macrophage accumulation (2–4). In high-fat diet (HFD)-fed mice, adipose tissue macrophages (ATMs) are not only more numerous than their counterparts in chow-fed controls, but broadly shifted toward an activated phenotype, with an overall increased expression of proinflammatory mediators such as TNF-α. At the same time, the number of ATMs expressing CD11c, a surface marker associated with classic macrophage activation, increases dramatically (5–7).

Proinflammatory macrophages are not randomly distributed throughout visceral adipose tissue, but cluster around dead adipocytes in aggregates called crownlike structures (4, 8). Within crownlike structures (CLSs), macrophages engulf lipids to form foam cells and/or fuse into multinucleate syncytia like those seen in inflammatory granuloma (8). The number of CLSs is correlated with the appearance of insulin resistance in HFD-fed mice (4) and obese people (9, 10).

The role of visceral ATMs in development of systemic metabolic dysfunction is not well understood. For instance, the mechanism of macrophage activation and extent to which it depends on the global glycemic environment remain unclear. Human and murine obesity is associated with different degrees of hyperglycemia and overt diabetes. Hyperglycemia has been shown to enhance myelopoiesis (11), activate myeloid cells (12, 13), and contribute to macrophage retention in atherosclerotic lesions (11, 14).

Kojima et al. first identified a leukocyte population marked by the production of insulin mRNA and proinsulin (PI) protein; they are called PI-producing bone marrow-derived cells (PI-BMDCs) (15). Though readily detectable in tissues of streptozotocin (STZ)-diabetic mice (15–18), PI-BMDCs are absent in nondiabetic controls. Their presence is dependent on hyperglycemia, as glucose injections in nondiabetic mice induce their appearance within 1–3 d (15), and reversal of STZ-diabetes by insulin causes a rapid decline in their number (16). PI-BMDCs often coexpress proinflammatory markers such as TNF-α (15–18) and are also detected in the peripheral nervous system of STZ-diabetic mice, where they play a crucial role in the pathogenesis of diabetic neuropathy (18). PI-BMDCs occur at low frequency (2–3%) in the bone marrow (BM) and liver of mildly hyperglycemic HFD-fed and ob/ob mice (15). Interestingly, PI-BMDCs are relatively abundant in visceral adipose tissue of these rodent models of type 2 diabetes (15).

In this study, we tested the hypothesis that hyperglycemia contributes to adipose tissue inflammation and insulin resistance by examining the pathogenic role of adipose tissue PI-BMDCs in response to hyperglycemia and during the development of HFD-induced type 2 diabetes.

MATERIALS AND METHODS

Animals and treatments

Male C57BL6/J, mouse insulin promoter-trifusion protein (MIP-TF), rat insulin promoter (RIP)-Cre and ob/ob mice were purchased from Jackson Laboratories (Bar Harbor, ME, USA). Mouse insulin promoter-green fluorescent protein (MIP-GFP) mice and inducible diphtheria toxin (DT) receptor (iDTR) mice were provided by Dr. V. Yechoor and Dr. H. Virgin, respectively. We housed mice in the Baylor College of Medicine Taub Animal Facility and maintained them on a 12-h light-dark cycle. Soy protein-free standard chow (Harlan, Houston, TX, USA) or HFD (60% of calories from fat; D12492, Research Diets, New Brunswick, NJ, USA) were used. We ensured that all procedures performed on mice were in keeping with the policies of the Baylor College of Medicine IACUC. For BM transplantation, 6-wk-old male C57BL6/J mice were irradiated (9.5 Gy divided into 2 doses, 4 h apart) and injected with 4 × 10⁶ BM cells isolated from male iDTR or iDTR/RIP-Cre mice. Experiments were performed after 4 wk of recovery. For reversal of diabetes, we injected mice with 1 U of insulin glargine using Lantus Solo Star Pen (Sanofi-Aventis, Bridgewater, NJ, USA) or 200 mg/kg of phloridzin dihydrate (Sigma-Aldrich, St. Louis, MO, USA) diluted in buffer containing 10% ethanol/15% DMSO/85% PBS subcutaneously twice a day (at 0:00 and 18:00) over the indicated time course. For PI-BMDC ablation studies, DT (Sigma-Aldrich) in normal saline was delivered to mice by tail vein injection at a dose of 30 ng/g body weight daily for 2 consecutive days.

Metabolic measurement

We measured glucose with a One Touch Ultra glucometer (LifeScan, Milpitas, CA, USA). We measured HbA1c using a A1C Now meter (Bayer, Whippany, NJ, USA). For intraperitoneal glucose tolerance tests (GTTs), we injected 1.0 g of glucose per kilogram body weight following 4 h of food withdrawal. For intraperitoneal insulin tolerance tests (ITTs), we administered human regular insulin (Eli Lilly, Indianapolis, IN, USA) at 1.5 U/kg after a 6-h food withdrawal. Blood glucose was measured at 0, 15, 30, 60, and 120 min after injection. We performed euglycemic hyperinsulinemic clamp on unrestrained mice using the insulin clamp technique in combination with HPLC purified [3-³H]glucose and [¹⁴C]2-deoxyglucose as described previously (19). Overnight unfed conscious mice received a priming dose (10 μCi) and then a constant infusion (0.1 μCi/min) of [3-³H]glucose for 3.5 h. Blood samples were collected from the tail vein at 0, 50, 55, and 60 min to measure the basal glucose production rate. After 1 h infusion, mice were primed with regular insulin (bolus 40 mU/kg of body weight) followed by a 2-h constant insulin infusion (6 mU/kg/min). Using a separate pump, 25% glucose was used to maintain the blood glucose level at 100–140 mg/dl, as determined every 10 min using a glucometer. Glucose production rate, peripheral glucose disposal rate, and glucose infusion rate were then calculated. At the end of the clamp procedure, we killed the mice and snap froze the soleus, gastrocnemius, adipose tissue, and liver in liquid nitrogen. Glucose uptake in different tissues was calculated from plasma 2-[¹⁴C] deoxyglucose profile fitted with double exponential curve and tissue content of 2-[¹⁴C]deoxyglucose-6-phosphate.

Stromal vascular cell isolation and flow cytometry

We isolated stromal vascular cells from gonadal adipose tissue using a modification of the method described by Brake et al. (20). Briefly, after carefully removing lymph nodes and large vessels, we minced and digested gonadal adipose tissue with collagenase D (1 mg/ml; Roche, Branchburg, NJ, USA) for 45 min at 37°C. Next, we filtered cell suspensions through a 100 μm filter and centrifuged at 500 g for 10 min. Pellets were then incubated with red blood cell lysis buffer (eBioscience, San Diego, CA, USA) for 5 min prior to centrifugation (300 g for 5 min) and resuspension in staining buffer (PBS with 2% fetal bovine serum). For some of experiments, we also stored stromal vascular cells at −80°C directly. Prior to flow cytometry analysis, we counted stromal vascular cells using a hemocytometer, then blocked Fc receptors by incubating in Fc Block (1:100; BD Biosciences, San Diego, CA, USA). We performed 3-color staining using F4/80-allophycocyanin, CD11b-FITC, and CD11c-phycoerythrin (1:100; BD Biosciences). We ran samples on a LSRII flow cytometer and analyzed using FACS Diva software (BD Biosciences).

Immunohistochemical analysis and cell quantification

We subjected slides to xylene deparaffinization and subsequent rehydration through an ethanol series. We permeabilized cells with 0.5% Tween-20 and employed Reveal Decloaker (Biocare Medical, Concord, CA, USA) for antigen retrieval in a steamer. We used 3% hydrogen peroxide to quench endogenous peroxidase activity and performed on-slide blocking, first with an Avidin/Biotin Block kit (Vector, Burlingame, CA, USA) and next with 1% normal goat serum in tris buffered saline (0.05 M Tris/HCl, 0.15 M NaCl, pH 7.6). We incubated in primary antibody overnight at 4°C and in secondary antibody for 30 min at room temperature. We performed all washes with auto wash solution (Biocare Medical, Concord, CA, USA). We used the following primary-secondary antibody combinations for detection of each protein, diluting all antibodies in commercial antibody diluents (Dako, Carpinteria, CA, USA): For PI detection we used monoclonal mouse anti-PI (1:100; Fitzgerald, North Acton, MA, USA) followed by goat anti-rat Fab secondary (1:200; Vector). This PI antibody specifically recognizes the B-peptide-C-peptide junction of the PI molecule and has been used for the detection of PI in the spleen and thymus of humans and mice. We confirmed our results with polyclonal guinea pig anti-PI (1:700; Progen, Heidelberg, Germany) followed by goat anti-guinea pig Fab secondary (1:500; Vector). We employed mouse and guinea pig IgG isotypes (Rockland, Limerick, PA, USA) to ensure the specificity of the staining. For perilipin detection, we used guinea pig anti-perilipin (1:1000; Progen) followed by goat anti-guinea pig Fab secondary (1:1000; Vector). For CD11c detection, we used hamster anti-CD11c (1:20; AbD Serotec, Raleigh, NC, USA) followed by goat anti-hamster Fab secondary (1:200; Vector). For F4/80 staining, rat anti-F4/80 (1:500; AbD Serotec, Raleigh, NC, USA) followed by goat anti-rat Fab secondary (1:500; Vector). For TNF-α staining, rat anti-TNF-α (1:50; Abcam, Cambridge, MA, USA) followed by goat anti-rat Fab secondary (1:100; Vector). We visualized signal with the ABC and NovaRed kits (Vector), then counterstained with Mayer’s hematoxylin, dehydrated through an ethanol series, incubated in xylene, and coverslipped with CytoSeal 60 (Dako). For GFP staining, we incubated slides overnight in rabbit anti-GFP (1:4000; Abcam) then performed secondary antibody staining and detection with the Rabbit Envision kit (Dako) per manufacturer’s instructions. For immunofluorescence, we performed the above protocol up to the primary antibody step, leaving out the peroxidase quench step. We employed FITC-conjugated anti-GFP (1:100; Abcam) and mouse anti-PI (1:100; Fitzgerald, North Acton, MA, USA) with goat anti-mouse IgG labeled with Alexa Fluor 546 (1:1000). We collected images using an Axiovert microscope and Axiovision 4.0 software (Zeiss, Thornwood, NY, USA). For adjacent section immunohistochemistry, we took images from consecutive 5-μm sections from 3 to 6 mice in each group. We quantified cells that had a visible nucleus in consecutive sections for overlap staining.

Gene expression analysis

We homogenized adipose tissue or cells using Qiazol reagent, then isolated RNA with the RNeasy Lipid-rich Tissue Kit with on-column DNase treatment (Qiagen, Valencia, CA, USA). Totan RNA (2 μg) was reverse-transcribed using iScript cDNA Synthesis Kit (Bio-Rad Laboratories, Hercules, CA, USA). For detection of Ins-1 and Ins-2 transcripts, we employed 3-step PCR protocol (98°C, 61°C, and 72°C) for 33 cycles with the Phusion High-Fidelity PCR Kit (Finnzymes, Grand Island, NY, USA). Then, we ran reaction products on 1.5% agarose gel and detected bands via ethidium bromide staining. For quantitative RT-PCR, SYBR green-Rox mix (Quanta, Gaithersburg, MD, USA) was used following manufacturer’s instructions in a thermal cycler (Applied Biosystems, Grand Island, NY, USA). We obtained expression levels of transcripts relative to optimal sets of housekeeping genes (Eef1g and Gapdh) using the GeNorm program (Biogazelle, Zwijnaarde, Belgium). PCR products were first confirmed by direct sequencing. The sequences of individual primers are listed in Table 1. For Ins-1 expression, we also used commercial primers (PPM02964F; Qiagen).

TABLE 1.

Sequence of primers

Gene	Forward	Reverse	Annealing (°C)
mIns1	ATGGCCCTGTTGGTGCACTTCC	TTAGTTGCAGTAGTTCTCCAGCTGG	62
mIns2	ATGGCCCTGTGGATGCGCTT	CTAGTTGCAGTAGTTCTCCAGCTGG	62
mTNF-a	ACGGCATGGATCTCAAAGAC	AGATAGCAAATCGGCTGACG	60
mCCR-2	TCAGCTGCCTGCAAAGACCAG	CATACGGTGTGGTGGCCCCT	60
mMac-1 (CD11b)	TGGGCAGGTGGAGCCTTCCT	CACTGCCACCGTGCCCTCTG	60
mEmr-1 (F4/80)	CTTTGGCTATGGGCTTCCAGTC	GCAAGGAGGACAGAGTTTATCGTG	60
mCD11c	CTGGATAGCCTTTCTTCTGCTG	GCACACTGTGTCCGAACTC	60
mMCP-1	ACTGAAGCCAGCTCTCTCTTCCTC	TTCCTTCTTGGGGTCAGCACAGAC	60
mCD206	CAGGTGTGGGCTCAGGTAGT	TGTGTTGAGCTGAAAGGTGA	60
mIL-1b	CTGGTGTGTGACGTTCCCATTA	CCGACAGCACGAGGCTTT	60
mEef1g	GGCCAAACCAACCGCACC	CGATGTCACTGTCAGCAAAG	60
mGAPDH	TGAAGGTCGGTGTGACCG	CCATTCTCGGCCTTGACT	60

Statistical analysis

We presented results as means ± sem, using Student’s t test or ANOVA. For analysis of weight curves, GTT and ITT, we used 2-way (intervention × time) ANOVA with repeated measures. A statistically significant difference was defined as a P < 0.05.

RESULTS

PI expression marks a proinflammatory, highly CLS-enriched macrophage subset in the visceral adipose tissue of HFD-fed mice

We fed 60% HFD (2, 3, 21) to 10-wk-old male C57BL6/J mice for 16 wk. The HFD-fed mice gained weight progressively (Fig. 1A) with increased fat mass, impaired glucose tolerance, and insulin resistance as revealed by abnormal ITT (data not shown). They also developed significant hyperglycemia; the 4 h fasting glucose level diverged by wk 3, and the difference in blood glucose was ∼ 60 mg/dl at wk 16 (Fig. 1B). Immunohistochemistry revealed numerous PI⁺ cells in gonadal adipose tissue of HFD-fed mice, but not in chow-fed mice (Fig. 1C). PI⁺ cells were similarly abundant in HFD-fed mesenteric and perirenal depots (data not shown), but considerably less common in inguinal adipose tissue (Supplemental Fig. 1). PI⁺ cells were, however, rare in liver and skeletal muscle. We confirmed staining specificity with an alternate PI antibody and isotype controls (Supplemental Fig. 1). Ins-1 and Ins-2 transcripts were readily detectable in gonadal adipose tissue isolated from HFD-fed mice, but neither was present in chow-fed controls. These transcripts were found only in the stromal vascular cells and not in the adipocyte fraction (Fig. 1D). In a parallel experiment, we fed MIP-TF mice, which express GFP behind the mouse Ins-2 promoter (22), the same HFD and observed complete correspondence between GFP and PI staining (Fig. 1E). Similar results were obtained with HFD-fed MIP-GFP mice, which express green fluorescence protein (GFP) under the control of the mouse Ins-1 promoter (Supplemental Fig. 2) (23).

Figure 1.

PI-producing cells accumulate in obese visceral adipose tissue. A) Body weight and B) 4 h fasting blood glucose of HFD- and chow-fed mice (n = 16 per group) for 16 wk. C) Immunostaining for PI in gonadal adipose tissue. D) Gel analysis of PCR products for insulin transcript in whole gonadal adipose tissue of HFD- and chow-fed mice followed by insulin transcript in gonadal adipocytes vs. stromal vascular cell (SVC) fractions in HFD-fed mice. E) Immunofluorescent staining of GFP and PI in MIP-TF mice. For all such cells, there is complete overlap between GPF and PI staining in MIP-TF mice. All images show representative 5 μm thick sections from groups of 3–12 mice. Error bars represent sem. *P < 0.05. Scale bars, 75 μm.

We next finely characterized the PI-BMDCs by immunohistochemistry after 16 wk of HFD. In serial sections, we found that 95% of PI⁺ cells coexpressed F4/80 and 62% of F4/80(+) cells stained positive for PI (Fig. 2A). These data indicate that in gonadal adipose tissue of HFD-fed mice, virtually all PI⁺ cells are macrophages and the majority of macrophages produce PI. Direct quantification revealed that 59 and 68% of PI⁺ cells coexpressed CD11c (Fig. 2B) and TNF-α (Fig. 2C), respectively. Virtually all detectable TNF-α(+) cells had stromal vascular cell morphology. This is noteworthy because TNF-α mRNA has been reported in adipocytes as well (24, 25). In contrast, only 42% PI⁺ cells showed overlap with CD206 (Fig. 2D), a marker of alternatively activated M2 macrophages. Finally, we sought to determine the relationship between PI⁺ macrophages and CLSs. Employing a technique described by Cinti et al. (8)—in which dead adipocytes at the center of CLSs can be identified by absence of perilipin staining—we found that 95% of CLSs contain at least 1 PI⁺ cell. Moreover, 79% of all PI⁺ cells were detected within CLSs, leaving only 21% among non-CLS singlet ATMs (Fig. 2E). Taken together, these data demonstrate that PI expression identifies a population of macrophages with cytologic hallmarks and histologic distribution consistent with a role in visceral adipose inflammation.

Figure 2.

PI-producing cells are macrophages that coexpress proinflammatory markers and are enriched in CLSs. A–D) Consecutive gonadal adipose tissue sections stained for PI (left) and different markers (right) including F4/80, CD11c, TNF-α, and CD206 after 16 wk of HFD. E) Consecutive gonadal adipose tissue sections stained for PI and perilipin. Absence of perilipin staining indicates CLSs. Percentage of CLSs containing at least one PI⁺ cell (red bar) vs. those containing no PI⁺ cell (yellow bar). Percentage of PI⁺ cells that are found within CLSs (dark green bar) vs. those found as singlets, not associated with CLSs (light green bar). Cells with a visible nucleus in consecutive sections were counted. Five images per mouse (each with >300 cells) and 3–6 mice in each group. All sections are 5 μm. Error bars represent sem. Arrowheads show representative double-positive cells in the fields. Scale bar, 50 μm.

HFD feeding-induced hyperglycemia underlies accumulation of PI-producing macrophages in visceral adipose tissue and contributes to adipose inflammation and insulin resistance

We next aimed to determine whether presence of PI⁺ ATMs was contingent on the hyperglycemic environment. Hyperglycemia per se has been shown to synergize with fatty acids in in vitro macrophage activation assays (26). HFD-feeding induces significant but mild hyperglycemia (Fig. 1B); however, the impact of HFD-associated hyperglycemia on ATM accumulation has not been tested. We performed another 14 wk HFD feeding protocol (Fig. 3A). During the first 8 wk, mice received HFD as previously described. During the last 6 wk, they received HFD plus a glucose-lowering agent: insulin glargine or phloridzin vs. vehicle control (27). Phloridzin is an inhibitor of sodium/glucose cotransporters located in proximal nephrons and intestine (28). Phloridzin treatment lowers glycemia without instigating other metabolic effects of insulin. We selected doses of glargine and phloridzin that did not significantly impact body weight or fat content of HFD-fed mice (pilot data not shown). As expected, glargine- and phloridzin-treated mice displayed significant amelioration in hyperglycemia (Fig. 3B) and hemoglobin A1c (Fig. 3C) vs. vehicle without differences in body weight (37.9 ± 0.5 and 37.0 ± 0.5 vs. 37.6 ± 0.8 g) or composition (Fig. 3D).

Figure 3.

Reversal of diabetes mitigates accumulation of PI-producing macrophages and expression of proinflammatory transcripts in visceral adipose tissue with concomitant improvements in systemic insulin sensitivity in HFD-fed mice. A) Scheme for treatment of HFD-induced diabetes. VEH, vehicle; GLAR, glargine; PHZ, phloridzin. B) Blood glucose 4 h post-VEH, GLAR, or PHZ treatment. C) Hemoglobin A1c (%). D) Body composition as assessed by ECHO MRI. E) Immunostaining for F4/80 and PI in gonadal adipose tissue quantified using at least 25 fields from 5 mice in each group. Arrowheads show representative positive cells in the field. F) Gel analysis of PCR products for Ins1 and Ins2 mRNA in gonadal adipose tissue of VEH-, PHZ-, and GLAR-treated mice. Panc, pancreas. G) Quantitative PCR for multiple macrophage and inflammatory genes in gonadal adipose tissue of VEH-, GLAR-, and PHZ-treated mice. H) Intraperitoneal GTT and I) ITT, ^#P < 0.05 between VEH and PHZ, and ^†P < 0.05 between VEH and GLAR. J) Hematoxylin-eosin (H&E) stained liver sections from VEH-, GLAR-, and PHZ-treated mice. n = 3–7 mice per group. Error bars represent sem. *P < 0.05. Scale bar, 50 μm.

Immunohistochemistry revealed numerous F4/80(+) and PI⁺ cells in gonadal adipose tissue of vehicle-treated mice. In glargine and phloridzin-treated mice, F4/80⁺ cells went down significantly and PI⁺ cells were markedly reduced (Fig. 3E). In agreement, Ins-1 and Ins-2 mRNA was present in vehicle-treated gonadal adipose tissue but markedly decreased with glargine or phloridzin treatment (Fig. 3F). CD11b mRNA was significantly reduced in glargine- and phloridzin-treated mice, and F4/80 down-trended. Importantly, although CD11c significantly declined in both glargine and phloridzin-treated mice, the level of CD206 mRNA did not change with treatment of hyperglycemia. Interestingly, levels of IL-1β and TNF-α were unaltered by amelioration of hyperglycemia (Fig. 3G).

To determine the effect of partial amelioration of hyperglycemia on glucose dynamics, we performed intraperitoneal GTT and ITT. Notably, mice did not receive glucose-lowering drugs during the 24 h before testing. There was generally no difference in GTT except for slightly but significantly lower baseline glucose level after phloridzin treatment compared with vehicle (Fig. 3H). Glargine administration caused significant improvements in ITT at the 45- and 60-min time points, and phloridzin yielded significant improvement at the 15-min time point (Fig. 3I).

Genetic knockout and cellular ablation experiments that reduce adipose inflammation often concomitantly ameliorate hepatic steatosis (21, 29, 30). Indeed, hepatic triglyceride content was significantly attenuated in the phloridzin treated group when compared with vehicle (glargin and phloridzin vs. vehicle: 15.4 ± 2.4 and 12.9 ± 1.9 vs. 20.9 ± 2.9 mg/g of liver, P < 0.05 for phloridzin vs. vehicle) (Fig. 3J). Taken together, these data suggest that hyperglycemia per se promotes accumulation of PI⁺ ATMs. Inhibiting this process by treating HFD-fed mice with glargine or phloridzin causes an overall diminution of ATMs, as well as specific proinflammatory transcripts. These local changes parallel improvements in insulin resistance and reduction in hepatosteatosis. We note, however, that glargine and phloridzin are known to have effects beyond glucose lowering that may well have contributed to some of these improvements (31, 32). Next, we sought to test the impact of PI⁺ macrophages on adipose inflammation, glucose homeostasis, and hepatic lipid accumulation.

PI-producing macrophages underlie visceral adipose inflammation

To ascertain the contribution of PI⁺ macrophages to adipose inflammation and whole-body insulin resistance, we devised a strategy to ablate PI⁺ macrophages in a targeted and time-dependent manner. We employed the iDTR mouse, an animal with a stop-floxed primate DTR knocked into the ROSA locus (33). DTR is driven by the ubiquitous β-actin promoter so that when the iDTR mouse is crossed with a Cre-containing strain, DTR is constitutively expressed in all Cre(+) cells of the progeny. DTR-expressing cells can then be specifically ablated when these animals are treated with DT. We crossed iDTR with RIP-Cre mice (the latter harboring Cre driven by the rat insulin promoter) to generate RIP-DTR animals, in which all insulin-expressing cells can be deleted with systemic DT administration. We found that administration of DT via tail vein injection to RIP-DTR mice effectively deletes pancreatic β-cells, yielding diabetes (data not shown). To ablate only hematopoietic lineages, we performed the protocol outlined in Fig. 4A. We harvested BM from RIP-DTR or iDTR control mice and transplanted into 6-wk-old, lethally irradiated wild-type recipients. After a 4 wk posttransplant recovery period, we placed recipient mice on 16 wk HFD. Then, we treated both groups with intravenous DT injections daily for 2 d. Twenty-four hours after the second DT injection, we assessed gonadal adipose tissue from both groups for presence of PI⁺ macrophages and inflammatory markers. Importantly, 1 d prior to the first DT injection, body weight, body composition, and blood glucose as well as the number of total ATMs and PI⁺ macrophages (Supplemental Fig. 3) were equivalent in iDTR and RIP-DTR recipients.

Figure 4.

Selective ablation of PI-producing macrophages ameliorates gonadal adipose tissue inflammation in HFD-fed mice. A) Scheme for selective ablation of PI-producing macrophages. Wild-type mice were irradiated (XRT) and transplanted with BM (BMT) from RIP-DTR mice (as shown) or iDTR controls. After 4 wk of recovery, mice were placed on 60% HFD for 16 wk, and then inflammation in gonadal adipose tissue (GWAT) was evaluated after DT injections. Immunostaining for (B) PI and (C) F4/80 in iDTR and RIP-DTR recipients quantified at least 25 fields from 5 mice in each group. D) Gel analysis of PCR products for Ins1 and Ins2 mRNA. Panc, pancreas. E) Quantitative PCR analysis showing relative expression of inflammation-related transcripts. F) Quantification of cellular subsets by flow cytometry. The macrophage fraction (Mac) contains cells that are CD11b(+)/F4/80(+). The monocyte fraction (Mono) indicates CD11b(+)/F4/80(−) cells. The CD11b(−) contains all stromal vascular cells that are neither macrophages nor monocytes. CD11c(+) and CD11c(−) macrophages quantified within the population of CD11b(+)/F4/80(+) cells (n = 4 mice per group). Error bars indicate sem. *P < 0.05. All scale bars, 50 μm.

One day after the second DT injection, we found that control iDTR BM recipients maintained numbers of PI⁺ cells commensurate with our first diet study but the number of PI⁺ cells was reduced by 75% in gonadal adipose tissue of RIP-DTR BM recipients (Fig. 4B). F4/80(+) ATMs were also substantially reduced in RIP-DTR vs. iDTR recipients. In addition, as opposed to ATMs that were arranged in complex, multilayered CLSs in iDTR recipients, the residual ATMs detected in RIP-DTR recipients were associated with considerably smaller, single-layered CLSs, (Fig. 4C, confirmed by absence of perilipin staining, data not shown). Paralleling this change, Ins-1 and Ins-2 mRNA were present in iDTR recipients but markedly decreased/absent in RIP-DTR recipients (Fig. 4D). Furthermore, RIP-DTR recipients had significantly lower levels of CD11b and CD11c transcripts vs. iDTR controls, and down-trended F4/80 transcripts. CD206 mRNA levels, however, were not different between groups (Fig. 4E). These changes were consistent with the loss of a proinflammatory macrophage population and similar to those observed with glargine- or phloridzin-mediated amelioration of hyperglycemia. Also in keeping with this prior experiment, classic proinflammatory gene transcripts IL-1β and TNF-α were not different between the groups (Fig. 4E).

We further analyzed the impact of ablation on cellular composition of stromal vascular fractions of gonadal adipose tissue by flow cytometry. We found that CD11b(+)/F4/80(+) macrophages were 50% lower in RIP-DTR vs. iDTR. In contrast, nonmacrophage cellular populations, either CD11b(+)/F4/80(−) (Mono) or CD11b(−), were unchanged between groups (Fig. 4F, Supplemental Fig. 4), reinforcing the finding that ablation of PI⁺ cells targets an ATM subset. RIP-DTR vs. iDTR BM recipients had similar decrements in both CD11c(+) and CD11c(−) macrophage subsets (Fig. 4F, Supplemental Fig. 4). This finding corroborates our immunohistochemistry results (Fig. 2B) and again demonstrates that PI⁺ macrophages represent an ATM population that partially overlaps with the CD11c(+) ATM subfraction that has been implicated in CLS formation (5) and systemic insulin resistance (4). We observed no differences between posttreatment iDTR and RIP-DTR mice in number of splenocytes or peripheral blood monocyte subsets by flow cytometry (data not shown).

Ablation of PI-producing macrophages improves glucose homeostasis in obese, hyperglycemic mice

To determine the role of PI⁺ ATMs on systemic glucose homeostasis, we devised the experimental scheme outlined in Fig. 5A. Following 15 wk HFD feeding, we performed intraperitoneal GTT or ITT on RIP-DTR and iDTR BM recipients. We allowed mice to recover from the test for 48 h then injected DT daily for 2 consecutive days. Twenty-four hours after the second injection, we repeated intraperitoneal GTT or ITT on the same animals. As in the last experiment, RIP-DTR and iDTR recipients had equivalent body weights and blood glucose prior to DT treatment. Though both groups maintained similar body weights following DT treatment, blood glucose levels fell significantly in RIP-DTR, but not iDTR, mice (Fig. 5B). Strikingly, intraperitoneal GTT revealed significantly improved glucose response in DT-treated RIP-DTR (Fig. 5C, right), but not iDTR recipients (Fig. 5C, left). In addition, intraperitoneal ITT similarly showed an improved response in DT-treated RIP-DTR (Fig. 5D, right), but not iDTR mice (Fig. 5D, left). In parallel experiments, we fed RIP-DTR and iDTR mice with a chow diet for the same duration as the HFD groups and treated the mice with 2 DT injections 24 h apart. A day after the second DT treatment, we performed intraperitoneal ITT in these animals and found no difference in their serum glucose response (Supplemental Fig. 5).

Figure 5.

Selective ablation of PI-producing macrophages reverses insulin resistance in HFD-fed mice. A) Scheme for determining impact of ablation of PI-producing macrophages on glucose metabolism. B) Body weight and 4 h fasting blood glucose, (C) intraperitoneal GTT, and (D) ITT of iDTR and RIP-DTR recipients before and after DT treatment. Euglycemic/hyperinsulinemic clamp data showing (E) basal glucose production (GP), GP after clamp, glucose infusion rate (GIR), and glucose disposal rate (GDR) and (F) glucose uptake into gonadal adipose tissue (GWAT), inguinal white adipose tissue (IWAT), and soleus muscle. n = 5–7 mice per group. Error bars represent sem. *P < 0.05.

To corroborate the GTT and ITT findings, we subjected the HFD-fed and DT-treated RIP-DTR and iDTR mice to euglycemic/ hyperinsulinemic clamp. First, we found that glucose infusion rate during clamp was 47% higher in RIP-DTR mice vs. iDTR mice mainly driven by improved glucose disposal rate by 40% in RIP-DTR recipients (Fig. 5E). Glucose uptake into gonadal adipose tissue increased by 65% in RIP-DTR mice, and it was unchanged in inguinal adipose tissue, which contains relatively few PI⁺ macrophages. Interestingly, RIP-DTR mice had a small but statistically significant increase in soleus glucose uptake vs. iDTR controls (Fig. 5F). Taken together, these results demonstrate that ablation of PI⁺ macrophages significantly improves glucose homeostasis both in visceral adipose tissue and systemically in hyperglycemic HFD-fed animals. Strikingly, these effects occur within only 48 h of the first, and 24 h of the second DT injection.

Ablation of PI-producing macrophages yields improvement in hepatic steatosis without impacting liver inflammation markers

Given the concomitant reduction in hepatic steatosis observed in other models that mitigate visceral adipose inflammation in HFD-fed mice (21, 29, 30), we opted to evaluate livers from post-DT treatment RIP-DTR and iDTR mice. RIP-DTR recipients displayed greatly reduced intrahepatocellular fat droplets vs. iDTR controls (Fig. 6A) on hematoxylin-eosin staining. Similarly, thin layer chromatography showed marked decrement in triglycerides in RIP-DTR recipients, which was corroborated by enzymatic assays (Fig. 6B). There was no difference between groups in plasma free fatty acid (0.323 ± 0.056 vs. 0.310 ± 0.011 mM, P = 0.83), triglyceride (55.66 ± 4.96 vs. 64.24 ± 4.97 mg/dl, P = 0.29), or cholesterol levels (185.18 ± 13.26 vs. 158.04 ± 7.97 mg/dl, P = 0.15).

Figure 6.

Selective ablation of PI-producing macrophages reverses hepatic steatosis without impacting local inflammation. A) Hematoxylin-eosin (H&E) staining of liver sections from iDTR and RIP-DTR recipients. B) Image of a thin layer chromatography plate showing lipid species in livers: triglycerides (TG), nonesterified fatty acids (NEFA), cholesterol (CHOL), and phospholipid (PL) as well as enzymatic assays for TG. C) Immunostaining for F4/80 in livers. D) Gene expression profiling of macrophage and inflammation markers in livers. n = 3 mice per group. Error bars represent sem. *P < 0.05. Scale bar, 50μm.

Next, we evaluated the impact of PI⁺ macrophage ablation on hepatic inflammation markers. As predicted from the low intrahepatic PI⁺ macrophage number, we did not detect a significant difference in the number or distribution of F4/80(+) cells between iDTR and RIP-DTR recipients after DT treatment (Fig. 6C). Furthermore, gene expression profiling of livers from each group revealed that ablation yielded no difference in levels of any proinflammatory mRNA assessed (Fig. 6D). Taken together, these data indicate that deletion of PI⁺ macrophages, a diabetes-specific leukocyte population largely restricted to visceral adipose tissue, rapidly and significantly ameliorates HFD-induced hepatic steatosis.

Discussion

There is emerging evidence that the diabetic state compromises the function of cells in the BM. Furthermore, hyperglycemia causes dysfunctional BM-derived hematopoietic cells that travel to various peripheral organs where they contribute to macrovascular as well as microvascular complications (34). PI-BMDCs represent a dysfunctional set of BM-derived hematopoietic cells that have been shown to underlie diabetic neuropathy (17, 18, 35, 36). These PI-producing cells were noted to be quite abundant in the adipose depot of diabetic rodents (15). In this study, we asked whether PI-BMDCs are part of an ATM subset, and if so, whether they contribute to insulin resistance in the HFD-induced type 2 diabetes mouse model.

Wild-type C57BL/6 mice fed HFD have been shown to develop obesity and mild hyperglycemia (21, 37). In this type 2 diabetes model, we found that PI-producing macrophages comprised over half of all visceral adipose macrophages and were undetectable in chow-fed controls. Almost 80% of these cells were found in CLSs, and, consistent with previous findings, many coexpressed TNF-α (15). Approximately 60% of these cells expressed CD11c, and a lower percentage (∼40%) expressed CD206. Notably, though broadly proinflammatory, the population was heterogeneous and did not simply recapitulate the expression pattern of any established marker. Interestingly, TNF-α and IL-1β transcript levels were not changed after reversal of hyperglycemia (Fig. 3G) or ablation of PI⁺ cells (Fig. 4E). TNF-α is also produced by adipocytes in the context of HFD feeding (24, 25) and, thus, may not be substantially impacted by ablation of PI⁺ macrophages. In particular, De Taeye et al. found that TNF-α knockout BM replacement did not protect against insulin resistance suggesting that ATM-derived TNF-α was not the determining factor in systemic insulin resistance (38). IL-1β is also produced by adipocytes (39) and may not be significantly impacted by PI⁺ macrophage ablation.

We previously showed that hyperglycemia induces the appearance of PI-BMDCs in adipose depots (15) and reasoned that the hyperglycemia could be the major pathogenic factor in the genesis of this PI⁺ subset of ATM. We designed an experiment to reverse the hyperglycemia without significantly affecting the obesity of these mice. Indeed, glargine therapy reduced HFD-induced visceral adipose inflammation and macrophage number—the latter appearing largely to result from a significant decline in PI⁺ macrophages (Supplemental Fig. 3). We were able to recapitulate these results in animals treated with phloridzin, a drug that has been used experimentally to divorce the glucose-lowering function of insulin from its other systemic effects (28). Though other effects of glargine or phloridzin cannot be excluded, these data strongly suggest that hyperglycemia per se can drive macrophage accumulation and adipose tissue inflammation, a process that coincides with the induction of PI-ATMs.

The RIP-DTR system allowed us to directly test the contribution of PI⁺ macrophages to adipose inflammation and systemic insulin resistance. Significantly, the extent of deletion observed with DT treatment—about 50% of all adipose macrophages—was consistent with the initial data demonstrating that ∼60% of adipose macrophages are marked by PI expression. The approximately similar reduction in CD11c(+) and CD11c(−) macrophage populations following DT injection was also expected from the ATM characterization data. It is noteworthy that recent studies suggest that the M1/M2 model of classic (CD11c⁺) and alternative (CD11c⁻) activation is a gross oversimplification that is not consistent with the data generated from comprehensive analyses of mouse and human ATMs (40–42). Nevertheless, ablation of PI⁺ macrophages did lead to significant declines in CD11c expression, whereas CD206 was unaffected. Importantly, ablation yielded an almost complete disappearance of large, complex CLSs, where PI⁺ macrophages commonly resided.

Ablation of PI-ATMs also led to significant improvements in blood glucose, GTT, and ITT within 24–48 h of DT treatment. Euglycemic/hyperinsulinemic clamp studies revealed that substantial improvements in overall glucose disposal and glucose uptake to gonadal adipose tissue and skeletal muscle contributed to these changes. Patsouris et al. previously reported that DT ablation of ∼100% of the CD11c(+) cells produced resolution of insulin resistance in HFD-fed mice (21). In our study, ablation of ∼70% of the PI⁺ ATM subpopulation, which reduced CD11c mRNA expression and CD11c⁺ cells by ∼ 50% (Fig. 4), led to rapid restoration of insulin sensitivity (Fig. 5). Multiple cytokines are involved in the inflammatory state in obese type 2 diabetes (43); it is possible that down-regulation of other cytokines may have contributed to the improvement observed. Furthermore, the autonomic nervous system plays a key role in the regulation of metabolic homeostasis (44). Adipose, liver, and skeletal muscle are innervated by autonomic nerves, which may underlie the crosstalk between adipose and other end organs involved in insulin responsiveness. Additional studies will be needed to determine if specific cytokines and/or neural pathways mediate the effect of PI-ATM ablation on insulin resistance.

Insulin resistance is a critical mediator of hyperglycemia that develops in type 2 diabetes; conversely, hyperglycemia itself may exert positive feedback to promote insulin resistance, a phenomenon known as glucotoxicity (45). Indeed, administration of insulin (46), or specific inhibitors of sodium/glucose cotransporters-2 located in the proximal nephrons such as empagliflozin (47) and dapagliflozin (48), not only decreased plasma glucose levels but also improved insulin sensitivity in type 2 diabetes patients. These results are analogous to the response of obese, hyperglycemic HFD-fed mice treated with glargine or phloridzin, which led to improved insulin sensitivity in the post-treatment period.

Taken together, we showed that hyperglycemia per se promotes adipose tissue inflammation, in part through the induction of a proinflammatory PI-ATM subset important for the genesis of systemic insulin resistance and hepatic steatosis, a process that can be forestalled by control of hyperglycemia or selective ablation of the PI-ATM subset. These findings may have implications on the molecular pathogenesis and therapy of obese type 2 diabetes.

Glossary

ATM

adipose tissue macrophage

BM

bone marrow

CLS

crownlike structure

DT

diphtheria toxin

GFP

green fluorescence protein

GTT

glucose tolerance test

HFD

high-fat diet

iDTR

inducible DT receptor

ITT

insulin tolerance test; MIP-GFP, mouse insulin promoter-green fluorescent protein; MIP-TF, mouse insulin promoter-trifusion protein

PI

proinsulin

PI-BMDC

proinsulin-producing bone marrow-derived cells; RIP-cre, rat insulin promoter-cre

STZ

streptozotocin