Targeting senescent cells alleviates obesity‐induced metabolic dysfunction

Abstract Adipose tissue inflammation and dysfunction are associated with obesity‐related insulin resistance and diabetes, but mechanisms underlying this relationship are unclear. Although senescent cells accumulate in adipose tissue of obese humans and rodents, a direct pathogenic role for these cells in the development of diabetes remains to be demonstrated. Here, we show that reducing senescent cell burden in obese mice, either by activating drug‐inducible “suicide” genes driven by the p16Ink4a promoter or by treatment with senolytic agents, alleviates metabolic and adipose tissue dysfunction. These senolytic interventions improved glucose tolerance, enhanced insulin sensitivity, lowered circulating inflammatory mediators, and promoted adipogenesis in obese mice. Elimination of senescent cells also prevented the migration of transplanted monocytes into intra‐abdominal adipose tissue and reduced the number of macrophages in this tissue. In addition, microalbuminuria, renal podocyte function, and cardiac diastolic function improved with senolytic therapy. Our results implicate cellular senescence as a causal factor in obesity‐related inflammation and metabolic derangements and show that emerging senolytic agents hold promise for treating obesity‐related metabolic dysfunction and its complications.


| INTRODUC TI ON
The prevalence of type 2 diabetes has quadrupled since 1980 (World Health Organization, 2016). It is associated with multi-organ complications, including cardiovascular and renal disease. Obesity and, more specifically, dysfunctional adipose tissue are strongly associated with whole-body insulin resistance and type 2 diabetes mellitus (Kahn & Flier, 2000;Palmer et al., 2015). Senescent cells accumulate in adipose tissue of obese and diabetic humans and mice (Minamino et al., 2009;Schafer et al., 2016;Tchkonia et al., 2010), but it is unclear whether they are merely associated with diabetes or if their presence is a causal driver.
Cellular senescence is a cell fate that entails proliferative arrest and acquisition of a pro-inflammatory senescence-associated secretory phenotype (SASP;Coppe et al., 2008). Although senescent cells exist in relatively small numbers in any particular tissue, they have been associated with multiple diseases of aging and are emerging as useful therapeutic targets for age-related diseases, including cardiovascular disease, pulmonary fibrosis, neurodegeneration, and osteoporosis (Farr et al., 2017;Musi et al., 2018;Roos et al., 2016;Schafer et al., 2017). A number of stimuli, including potentially oncogenic, inflammatory, damage-related, and metabolic stimuli, can trigger a senescence response (Munoz-Espin & Serrano, 2014). Components of the SASP secreted by adipose-derived senescent cells have been postulated to confer insulin resistance upon metabolic tissues, inhibit adipogenesis, and attract immune cells that can exacerbate insulin resistance . Here, we determined whether removing senescent cells in the context of obesity improves metabolic phenotypes.
Recently, drugs that preferentially decrease senescent cell burden, termed senolytics, have been identified (Kirkland, Tchkonia, Zhu, Niedernhofer, & Robbins, 2017). We discovered the first senolytics based on our observation that senescent cells rely on several survival pathways, including those regulated by PI3K/AKT-, p53/ p21/serpine-, HIF-1α-, and BCL-2/BCL-X L -family components, to confer resistance to their pro-apoptotic SASP and intracellular cell damage signals (Zhu et al., 2016(Zhu et al., ,2015. Knowing this, we identified dasatinib (D) and quercetin (Q) as orally bioactive drugs that transiently target these survival pathways to induce apoptosis preferentially in senescent cells (Zhu et al., 2015). Subsequently, we and others found that navitoclax, (ABT263, which targets Bcl-xL, Bcl-2, and Bcl-w but not Mcl-1), is also senolytic (Chang et al., 2016;Zhu et al., 2016). We showed that D and Q, alone and in combination, cause apoptosis in senescent cells without significant effects in quiescent or proliferating cells (Xu et al., 2018;Zhu et al., 2015). Senolytics do not prevent the generation of senescent cells and they are effective when administered intermittently, which could help to mitigate any potential negative effects of senescent cell removal, such as delayed wound healing (Demaria et al., 2017;Zhu et al., 2015).
We employed the combination of D plus Q (D + Q) in our studies for the following reasons. (a) In our hands, no senolytic investigated thus far targets all types of senescent cells . For example, unlike navitoclax (ABT263), fisetin, A1331852, A1155463 (Zhu et al., ,2016(Zhu et al., ,2015, or Q on its own, D selectively targets senescent adipose progenitors (Zhu et al., 2015), a key cell type for adipose tissue and metabolic function (Tchkonia et al., 2013). (b) On the other hand, Q, unlike D, is effective against senescent endothelial cells (Zhu et al., 2015), a cell type implicated in vascular complications of diabetes (Caballero, 2003). (c) D + Q is effective in alleviating multiple age-and senescence-associated disorders, including many that are frequent complications or comorbidities of diabetes in preclinical animal models; these comorbidities include arteriosclerosis, vascular hyporeactivity, osteoporosis, hepatic steatosis, physical dysfunction, neurodegeneration, and neuropsychiatric dysfunction (Farr et al., 2017;Musi et al., 2018;Ogrodnik et al., 2017Ogrodnik et al., ,2019Roos et al., 2016;Xu et al., 2018).
(d) Navitoclax and other BCL-2 family member inhibitors can be toxic, for example, causing severe thrombocytopenia, which can occur even with intermittent dosing (Wilson et al., 2010). Navitoclax can also cause neutropenia, complicating interpretation of whether its effects are due to senolytic activity or immune system suppression. For these reasons, we elected to focus on D + Q.
Here, we characterize the effects of eliminating senescent cells on obesity-induced derangements in adipose tissue function and glucose homeostasis. To do this, we used both transgenic mouse models and treatment with the senolytics, D + Q. Our findings support the idea that senescent cells could be a novel therapeutic target for treating obesity-induced metabolic dysfunction.

| Senescent cells accumulate in visceral fat in obesity
We used two transgenic mouse models from which senescent cells can be selectively cleared: (a) p16-3MR mice, in which a long p16 Ink4a -promoter sequence drives expression of a trimodal reporter-killer fusion protein (3MR), allowing senescent cell killing F I G U R E 1 Removal of obesity-induced senescent cells from adipose tissue. (a, b) Renilla luciferase activity in DIO p16-3MR mice (a, representative image), quantified in b (n = 6-9 per group). (c) Renilla luciferase activity localization in DIO p16-3MR dissected tissues. (d, e) Senescence-associated beta-galactosidase (SA-β-gal) activity as whole tissue activity (d; representative image) and % positive cells of total DAPI + cells in p16-3MR VAT (e; chow n = 3, DIO n = 7-9 per group). (f) Expression of p16-3MR transgene (eGFP) components (Renilla luciferase and mRFP) and p16 Ink4a (chow n = 3, DIO n = 11 per group) in p16-3MR VAT. (g, h) Senescence-associated beta-galactosidase (SA-β-gal) activity as whole tissue activity (g; representative image) and % positive cells of total DAPI + cells in p16-3MR VAT (h; n = 3-4 per group). (i) p16 Ink4a mRNA levels (c; chow n = 8, DIO n = 19-21 per group) in VAT of D + Q-treated DIO mice. (j) Percent of VAT stromal vascular fraction (SVF) cells highly expressing FLAG (a component of the p16 Ink4a promoter-driven ATTAC fusion protein), CENP-B, and p21 Cip1 after a single course of D + Q (o, n = 6 per group) in DIO INK-ATTAC mice. Means ± SEM are shown. Box and whisker plot show minimum, mean, maximum, 25th and 75th percentiles. *p < 0.05, **p < 0.005, ***p < 0.0005; one-way ANOVA with Bonferroni correction or two-tailed Student's t test when comparing two groups by ganciclovir (GCV) and identification by whole-body luminescence (Demaria et al., 2014) (Hall et al., 2017;Okuma, Hanyu, Watanabe, & Hara, 2017), limiting the sensitivity and specificity for senescent cell killing in these mouse models. Therefore, we used an additional distinct, potentially translatable approach for eliminating senescent cells, that is, senolytic drugs, which target anti-apoptotic pathways in senescent cells and do not depend on p16 Ink4a F I G U R E 2 Eliminating senescent cells enhances glucose homeostasis and insulin sensitivity. (a, b) Intraperitoneal glucose tolerance test in DIO p16-3MR (a; chow n = 4, DIO n = 6-7 per group) and DIO wild-type mice treated with D + Q (b, chow n = 4, DIO n = 11 per group) following senescent cell clearance. (c) Hemoglobin A1c in DIO p16-3MR (chow n = 4, DIO n = 15-18 per group) and DIO wild-type mice treated with D + Q (chow n = 6, DIO n = 11-12 per group). (d, e) ITT following ganciclovir treatment in p16-3MR mice (d, n = 4 chow, n = 18-19 DIO groups), or D + Q treatment in DIO wild-type mice (e, chow n = 6, DIO n = 11-12 per group). (f) Fold change in AKT serine-473 phosphorylation after 5-min ex vivo 5 nM insulin stimulation in freshly isolated p16-3MR VAT (n = 3 per group). For each mouse, p-AKT was normalized to total AKT and expressed as a ratio of p-AKT in insulin-exposed tissue to p-AKT in noninsulin-exposed tissue. (g) Glucose infusion rate (GIR) during hyperinsulinemic-euglycemic clamp in DIO mice treated with vehicle or D + Q (n = 8 per group). (h) Plasma insulin concentration at baseline and during hyperglycemic clamping (HGC) in DIO mice treated with D + Q (n = 8 per group). (i) Glucose appearance rate (Ra) during basal, hyperglycemic, and hyperinsulinemic-euglycemic clamping in DIO mice treated with D + Q (n = 8 per group). Means ± SEM are shown. *p < 0.05, **p < 0.005, ***p < 0.0005; one-way ANOVA with Bonferroni correction for multiple comparisons. # p < 0.05, two-tailed Student's t test comparing DIO vehicle-treated group to DIO ganciclovir-treated, or D + Q-treated group. Groups of interest were compared at each time point for GTTs and ITTs expression to reduce senescent cell burden Zhu et al., 2015).
Obesity was induced either by high-fat diet (diet-induced obesity, DIO, with ad libitum chow-fed controls) or by genetic means in leptin receptor knockout (db/db) mice (Supporting Information Figure S1).
Senescent cell abundance, as measured by luciferase activity in p16-3MR mice, increased in response to DIO compared to chowfed mice (Figure 1a-c and Supporting Information Figure S2a,b).
Bioluminescent imaging (BLI) in ex vivo tissues showed that the most prominent signal was present in visceral (specifically perigonadal) adipose tissue (VAT; Figure 1c and Supporting Information Figure   S2b). Although senescent cells are known to accumulate in other organs including the liver after high-fat feeding (Ogrodnik et al., 2017;Yoshimoto et al., 2013), senescent cell abundance detectible in excised tissues by BLI was not as pronounced in other adipose tissue depots, skeletal muscle, liver, and pancreas ( Figure 1c). That said, limitations of BLI, including a high threshold to visualize senescent cells and limited tissue penetration of the Renilla luciferase signal, could have contributed to our inability to detect bioluminescence in these other tissues. Therefore, we used additional, more sensitive methods including gene expression, senescence-associated betagalactosidase staining, and mass cytometry to assess senescent cell abundance in the studies reported below.

| Senescent cell clearance improves glucose homeostasis and insulin sensitivity
Senescent cell abundance declined after intermittent administration of ganciclovir to p16-3MR DIO mice (Figure 1a,b,d-f,), AP20187 to DIO INK-ATTAC mice (Supporting Information Figure S2c,d) and INK-ATTAC; db/db mice (Supporting Information Figure S2e), and D + Q to wild-type DIO mice (Figure 1g-i). CyTOF analyses conducted after a single course of AP20187 or D + Q in INK-ATTAC mice showed that adipose tissue senescent cells highly expressing p16 Ink4a -promoter-induced FLAG, CENP-B, or p21 Cip1 were decreased significantly (Figure 1j and Supporting Information Figure   S2f). In these experiments, adipose progenitor cells, rather than endothelial cells, macrophages, or T cells, were the main cell type targeted (Supporting Information Figure S2g,h).
Clearing senescent cells improved glucose tolerance (Figure 2a,b and Supporting Information Figure S3a,b) and reduced hemoglobin A1c (HbA1c), a marker of long-term glucose control (Figure 2c and Supporting Information Figure S3c). These effects were not seen in lean, chow-fed control mice (Supporting Information Figure S3d-g), wild-type (i.e., not p16-3MR) DIO mice treated with ganciclovir (Supporting Information Figure S3h,i), or DIO mice treated with navitoclax, which does not target senescent adipocyte progenitors (Zhu et al., 2016) (Supporting Information Figure S3j). The time course of metabolic improvement following initiation of D + Q treatment paralleled that of clearance of high p16 Ink4a -expressing cells by engaging death mechanisms due to transgenes (Supporting Information Figure   S3k,l). Eliminating senescent cells did not affect body weight, activity, or food intake, consistent with improved glucose homeostasis being due principally to increased insulin sensitivity (Supporting Information Figure S3m-r).
After senescent cell reduction, DIO mice became more insulin sensitive, as indicated by insulin tolerance testing (ITT; Figure 2d,e and Supporting Information Figure S4a) and an increased glucose infusion rate during hyperinsulinemic clamping ( Figure 2g).
Furthermore, AKT Ser473 phosphorylation increased in response to ex vivo insulin stimulation of adipose tissue freshly harvested from animals that had undergone senescent cell clearance ( Figure 2f and Supporting Information Figure S4b). The insulin-positive pancreatic islet area remained unchanged after depleting senescent cells (Supporting Information Figure S4c-e). Plasma insulin concentrations were lower in response to a glucose challenge in both p16-3MR mice treated with ganciclovir and DIO mice treated with D + Q (Supporting Information Figure S4f,g). Pancreatic insulin secretion was unchanged in D + Q-treated mice during hyperglycemic clamping experiments ( Figure 2h). Insulin sensitivity was not affected by genetic interventions that cause elimination of senescent cells in age-matched lean mice (Supporting Information Figure S4h) or obese WT mice (Supporting Information Figure S4i), indicating that off-target effects of AP20187 or ganciclovir are unlikely to have contributed substantially to the observed metabolic improvements.
These results indicate that, at least using these methods for senescent cell ablation in mice with DIO-induced metabolic dysfunction, metabolic benefits were primarily due to improved peripheral insulin sensitivity rather than β-cell compensation. Hepatic glucose production was also found to be unchanged upon D + Q treatment in DIO mice under basal, hyperglycemic, and hyperinsulinemic conditions ( Figure 2i). Our findings do not preclude the possibility that other approaches for clearing senescent cells or at different stages during development of obesity-induced metabolic dysfunction might affect hepatic glucose production or pancreatic insulin secretion.

| Adipogenic potential is improved after senescent cell reduction
Circulating and adipose tissue inflammatory mediators, some of which are components of the SASP (Coppe et al., 2008;Tchkonia et al., 2010), increase in obesity and can impede adipogenesis and contribute to insulin resistance (Kahn & Flier, 2000;Xu, Tchkonia et al., 2015). In p16-3MR mice but not WT mice treated with ganciclovir, plasma IFN-γ and IL-1β concentrations were decreased (Supporting Information Figure S5a,b). Cells expressing high concentrations of TNF-α isolated from the visceral adipose tissue (VAT) stromal vascular fraction declined upon AP20187 treatment of DIO INK-ATTAC mice (Supporting Information Figure   S5c). Plasma adiponectin, which is associated with improved insulin sensitivity (Kadowaki et al., 2006), also increased after senescent cell reduction in p16-3MR mice (Supporting Information Figure S5d,e), and adipose tissue IFN-γ expression was reduced (Supporting Information Figure S5f). F I G U R E 3 Adipogenesis is enhanced by senescent cell reduction. (a, b) Plasma activin A in p16-3MR mice (a; chow n = 3, DIO n = 9-11) and D + Q-treated DIO mice (b, chow n = 3, DIO n = 11-12). (c, d) Adipogenic gene expression in cells isolated from subcutaneous adipose tissue stromal vascular fraction (SVF) of p16-3MR (c; n = 5-7 per group) and D + Q-treated mice (d, n = 3 per group). (e) Representative images of lipid droplet formation during differentiation of adipocyte progenitors isolated from vehicle-or ganciclovir-treated p16-3MR mice after 5-day exposure to differentiation medium (scale bars indicate 50 μm). (f) VAT cell size in p16-3MR mice and D + Q-treated mice (n = 3-7 per group). (g, h) Subcutaneous:intra-abdominal adipose ratio in DIO p16-3MR mice (g, n = 5-6 per group) and D + Q-treated mice (h, n = 18-21 per group). Means ± SEM are shown. *p < 0.05; one-way ANOVA with Bonferroni correction or two-tailed Student's t test when comparing two groups Representative image of luminescence signal in AP20187-treated INK-ATTAC;db/db mice 24 hr following i.v. injection of 1 × 10 6 luciferase + monocytes (g), normalized to vehicle-treated mice and quantified in h (n = 6 per group). (i) Quantification of luminescence in D + Q-treated db/db mice 24 hr following i.v. injection of 1 × 10 6 luciferase + monocytes isolated from CAG-luc mice, normalized to vehicle-treated mice (n = 5 per group). Means ± SEM are shown. *p < 0.05, **p < 0.005, ***p < 0.0005; one-way ANOVA with Bonferroni correction or two-tailed Student's t test when comparing two groups 2015), was increased in DIO mice (Figure 3a,b) and correlated with highly p16 Ink4a -expressing senescent cell burden, as manifested by p16 Ink4a -promoter-driven luciferase expression in p16-3MR mice (Supporting Information Figure S5g). Activin A impedes expression of the insulin-sensitizing adipogenic transcription factors PPARγ and C/EBPα, thereby contributing to insulin resistance (Hamm, Jack, Pilch, & Farmer, 1999;Xu, Palmer et al., 2015;Zaragosi et al., 2010). Senescent cell ablation abrogated the DIO-induced increase in activin A (Figure 3a,b). This was associated with higher expression of adipogenic transcription factors and their targets in adipocyte progenitors from obese mice (Figure 3c,d) and enhanced adipogenic differentiation in culture (Figure 3e). These findings are consistent with previous data showing that senescent adipocyte progenitors develop a SASP that inhibits adipogenesis . It was also previously found that BrdU incorporation increases in adipose tissue following AP20187 treatment in INK-ATTAC mice (Baker et al., 2011), suggesting that cleared senescent cells can be replaced by nonsenescent, proliferation-competent adipocyte progenitors that can then differentiate into insulin-responsive fat cells.
We found that senescent cell clearance decreases adipocyte hypertrophy ( Figure 3f and Supporting Information Figure S5h) and increases the ratio of subcutaneous to intra-abdominal adipose tissue ( Figure 3g,h), reflective of better insulin sensitivity (Gustafson, Hedjazifar, Gogg, Hammarstedt, & Smith, 2015). These changes in adipose tissue distribution were mainly due to expansion of subcutaneous depots (Supporting Information Figure S5i,j). Furthermore, DIO INK-ATTAC mice treated with AP20187 had fewer lipid droplets in muscle, measured by oil red O staining (Supporting Information Figure S5k), and less severe hepatic steatosis (Ogrodnik et al., 2017).
Collectively, these findings indicate that decreasing the burden of senescent cells may enhance insulin sensitivity in part by improving the proliferative and differentiation potential of adipocyte progenitors, contributing to healthier adipose tissue distribution and limiting ectopic lipid deposition (Gustafson et al., 2015).
Activated macrophage contamination is a potentially significant confounder in these studies, since macrophages can express p16 Ink4a , β-galactosidase, and SASP-like factors (Hall et al., 2017). Unlike truly senescent cells, macrophages can revert to a nonactivated state, potentially even into monocytes that replicate. Therefore, we conducted CyTOF studies and specifically assayed the macrophage markers, F4/80 and Cdllb, and verified that AP20187 and D + Q do not specifically and immediately target activated macrophages, as this population did not decrease after one round of treatment with either agent (Supporting Information Figure S2g,h). In addition, we have previously shown that D + Q treatment sufficient to reduce senescent preadipocyte abundance does not directly reduce macrophage abundance in freshly isolated adipose tissue explants from obese human subjects (see Supporting Information Figure

| Senescent cell clearance decreases macrophage homing to adipose tissue
We found that senescent adipocyte progenitors attract monocytes in culture (Xu, Tchkonia et Figure S46m,n). Thus, senescent cells can cause macrophage migration into adipose tissue in obesity, and targeting senescent cells prevents and reduces the adipose tissue macrophage accumulation that is often associated with obesity.

| Decreasing senescent cell burden may alleviate complications of diabetes
Because our interventions target senescent cells not only in adipose, but also in other tissues, their potential effects could exceed those limited to alleviating adipose tissue dysfunction. Senolytics may affect senescence-related comorbidities associated with obesity (e.g., hepatic steatosis, osteoarthritis, and neuropsychiatric dysfunction), accelerated aging-like states associated with obesity (e.g., sarcope-  The transgenic models used in these studies are dependent on targeting high p16 Ink4a expression. Not every senescent cell has high p16 Ink4a , and there are cell types that have a transient elevation in p16 Ink4a that are not truly senescent, such as activated macrophages (Hall et al., 2017). These points suggest that interpreting studies based on targeting cells with high expression of p16 Ink4a alone requires caution, so we also used senolytics to test our hypothesis. Critically, since D + Q was given intermittently and the elimination half-life of each drug is <12 hr (Christopher et al., 2008;Graefe et al., 2001), the observed effects are consistent with reduced senescent cell burden, rather than effects on senescence-independent signaling pathways, receptors, or enzymes that require the sustained presence of drugs. Thus, intermittent administration of senolytics may delay or alleviate diabetes, complications of both obesity and diabetes, and other comorbidities associated with senescence-related chronic diseases, a possibility that, if further confirmed in pre-clinical studies, merits examination in clinical trials.
Human adipocyte progenitors, also termed adipose-derived stem cells or preadipocytes (Tchkonia et al., 2013), were isolated from abdominal subcutaneous adipose tissue biopsies from nine healthy subjects undergoing surgery to donate a kidney (age 37 ± 6 years, male, BMI 26 ± 2). These cells were used for osteopontin measurements in senescent cells and for THP-1 migration studies. These studies were approved by the Mayo Clinic Institutional Review Board.

| Metabolic testing
For DIO and lean mice, glucose tolerance and insulin tolerance testing (ipGTT, ITT, respectively) were performed by injecting 1.2 g/kg glucose or 0.8 mU/kg insulin (i.p.) in the early afternoon following a 6-hr or 4-hr fast, respectively. Glucose-stimulated insulin secretion (GSIS) in DIO and lean mice was measured after oral gavage of 1.5 g/kg glucose following a 4-hr fast. Intraperitoneal injection of glucose was used for GTT due to increased excursion of blood glucose, allowing for better differentiation between treatment groups.
Oral administration of glucose was used for insulin secretion measurements in Figure 2h to ensure the most physiologic induction of insulin secretion (including the contribution of incretins), as well as to test the overall validity of our results by testing improvements with senescent cell clearance in multiple assays. For db/db mice, glucose tolerance testing was performed by injecting 1.0 g/kg glucose i.p.
following an overnight fast. Tail vein blood glucose was measured at time 0 and at indicated time points using a handheld glucometer

| Hyperglycemic clamp experiments
Mice were fasted from 9 hr before the experiment until the end of the experiment. The study was divided into three periods, (a) a basal period in which only glucose tracer was infused, (b) a hyperglycemic period (HGC) in which, next to the glucose tracer, glucose was infused at a variable rate to maintain blood glucose levels at about 20 mM, and (c) a hyperinsulinemic-euglycemic period (HIEC) during which, in addition next to the glucose tracer, insulin was infused at a constant rate and glucose was infused at a variable rate to maintain blood glucose levels at about 6 mM. During all three periods, blood glucose levels and glucose infusion rates were monitored, from which kinetic parameters were calculated.

| Adipocyte progenitor isolation
The stromal vascular fraction of perigonadal adipose tissue was isolated by collagenase digestion as previously described (Tchkonia et al., 2005).

| Mass cytometry (CyTOF)
We designed a panel of antibodies based on surface markers, transcription factors, and cytokines (see Supporting Information Table S1).
Each antibody was tagged with a rare metal isotope and its function verified by mass cytometry according to the factory manual (Multi

| Plasma cytokine profiling
Plasma cytokines were quantified using multiplex ELISA on a Bio-Plex 200 analyzer by Eve Technologies (Calgary, Alberta, Canada).

| Quantitative real-time PCR
Quantitative real-time PCR was performed as described previously . Primer catalog numbers are in Supporting Information Table S2.

| Western blotting
Western blotting was performed as described previously .

| Adipose tissue analyses
SA-β-gal activity was assayed as reported previously . Briefly, a small piece of adipose tissue was collected in PBS, lightly fixed with glutaraldehyde and formaldehyde for 15 min, washed 3X in PBS, and placed in SA-β-gal activity solution containing X-gal at pH 6.0 at 37°C for 14-16 hr. Tissues were rinsed in PBS, nuclei were stained with DAPI, and adipose tissue was compressed between two glass slides for light microscopy. SA-β-gal + cells as a percent of all nuclei were quantified in at least 10 images taken at random per tissue using NIS-Elements software (Nikon Instruments, Melville, NY). Neither the person who captured images for quantification nor the observer counting SA-β-gal + cells in the captured images was aware of the identity of the samples. Photographs of SA-β-gal + tissue chunks were taken after 5-8 hr of incubation at 37°C. Cell size was quantified using at least 10 images taken at random from adipose tissue chunks by light microscopy by measuring diameters of all completely visible adipocytes per field using Nikon NIS-Elements software.

| Ex vivo insulin response assay
Adipose tissue was cut into small pieces (<100 mg) and washed with PBS 3 times. The minced tissue was treated with 5 nM insulin or vehicle (PBS) for 5 min at 37°C. p-AKT (#4060) and total-AKT (#4691) antibodies for Western analyses were purchased from Cell Signaling (Danvers, MA).

| Albumin-creatinine ratio
Albumin was measured by ELISA (GenWay, San Diego, CA), and creatinine was measured by colorimetric assay (MaxDiscovery; Bio Scientific, Austin, TX) in urine collected from nonfasted mice.
Albumin-creatinine ratio (ACR) was calculated by determining microgram of albumin per mg of creatinine.

| Conditioned medium experiments
Stromal vascular collagenase digests were plated, replated within 18 hr, and serially sub-cultured for 4 population doublings using differential plating and culture media designed to select against macrophage or endothelial cell contamination. Cells were exposed to 10 Gy radia-
Medium was then replaced with DMEM/12 containing only 10% FBS and insulin, and kept in culture for an additional 5-8 days.

| Comprehensive laboratory animal monitoring system
Metabolic rate and food intake were measured using a Comprehensive Laboratory Animal Monitoring System (CLAMS) as previously described .

| THP-1 macrophage migration
Migration assays were performed using Transwell polycarbonate membrane inserts with a 5-μm pore diameter purchased from

| Monocyte infiltration
Monocytes were isolated from femur and tibial bone marrow of CAG-luc transgenic C57BL/6 mice using a commercial kit (Monocyte Isolation kit, MACS Miltenyi Biotec, San Diego, CA). 1 × 10 6 monocytes in PBS were injected i.v. via the tail vein in db/db or lean mice.
Luminescent signal was examined 24 hr after monocyte injection.

| Luminescence imaging
Both Renilla and firefly luminescence imaging were performed using a Xenogen IVIS 200 system (Caliper Life Sciences, Hopkinton, MA).
The exposure for luminescence images was 5 min.

| Echocardiography
High-resolution ultrasound imaging was used to evaluate cardiac function as previously described (Roos et al., 2016).

| Anti-Wilms tumor protein immunohistochemistry
Formalin-fixed paraffin-embedded mouse kidney tissue sections were subjected to steam heat-mediated antigen retrieval for 30 min in 10 mM sodium citrate buffer, pH 6.0. Samples were then incubated at room temperature for 1 hr with primary antibody (Anti-Wilms Tumor Protein; Abcam #ab89901) diluted 1/100, followed by a 30-min incubation in a Ready-to-Use HRP-Labelled secondary antibody (Dako K4003). Color was developed using Vector NovaRED Peroxidase Substrate Kit (SK-4800) followed by hematoxylin counterstaining.

| Quantification and statistical analyses
When comparing two groups, unpaired two-tailed Student's t tests were used. p < 0.05 was considered significant. One-way ANOVA with Bonferroni correction was used to estimate statistical significance when more than two groups were being compared. All values are expressed as mean ± SEM. Sample sizes were chosen based on the means and variation of preliminary data to achieve at least 80% power and allowing for 5% type I error.

ACK N OWLED G M ENTS
The authors are grateful to J. Armstrong for administrative as-