Regulation of De Novo Adipocyte Differentiation Through Cross Talk Between Adipocytes and Preadipocytes

Over the past several years, the view of adipose tissue has evolved from an inert energy storage to a multifunctional endocrine organ regulating whole-body energy metabolism (1,2). Adipokines play an important role in this context, for example, in the regulation of food intake, lipid metabolism, and insulin sensitivity, but also in the control of adipocyte development and the recruitment of inflammatory cells to adipose tissue (2,3).

Adipose tissue development is controlled by complex interactions of extracellular and intracellular signals. This is partly due to specific adipokine receptors expressed in adipose tissue, which modulate adipocyte differentiation (4). The secretion and plasma levels of certain adipokines are related to body fat mass and the degree of obesity (5). Thus, a proinflammatory adipokine profile is associated with obesity and is thought to promote insulin resistance (5,6), while adiponectin and other anti-inflammatory adipokines are reduced in obesity and therefore might contribute to the maintenance of insulin sensitivity (4,5,7).

The growth of adipose tissue can be achieved through hypertrophy and hyperplasia (4,8,9). In hypertrophic obese subjects, proinflammatory “classically activated” M1 macrophage infiltration is increased (10). Conversely, hyperplasia is followed by a reduction in M1-type markers and an increase in anti-inflammatory “alternatively activated” M2 macrophages (11). Thus, hypertrophy is strongly associated with the development of insulin resistance and type 2 diabetes (4,5), whereas hyperplasia is associated with improved insulin sensitivity (6,12). Since changes in adipocyte size influence the adipokine profile, it can be envisaged that the balance of hypertrophy/hyperplasia leads to alterations in adipokine profiles, and this ultimately might affect disease progression.

Adipokine secretion is dependent not only on adipocyte size, but also on adipose depots; for example, subcutaneous adipose tissue has been shown to secrete higher amounts of metabolically beneficial adipokines such as leptin and adiponectin, whereas proinflammatory adipokine secretion is higher in visceral adipose tissue (13).

Although much is known about spatial and temporal differences in adipokine secretion, their effects on adipose tissue remodeling are not well understood. By using a coculture method integrated with fractionation techniques, we here identify adipokines secreted by adipocytes or adipose tissues explants that positively or negatively regulate adipocyte differentiation.

Recombinant vimentin (Vim) was purchased from R&D, and mouse interleukin (IL)-33 AG-40B-0041 and anti-mouse IL-33 antibody (clone Bondy-1–1, AG-27B-0013) were purchased from Adipogen. Dexamethasone, isobutylmethylxanthine, and insulin were obtained from Sigma-Aldrich. Bodipy493/503, Hoechst, Syto60, TRIzol, SYBR Green, and Lipofectamine were purchased from Invitrogen.

All animal studies were approved by the Zurich Veterinary Office. Male 10- to 16-week-old C57BL/6 mice were obtained from Charles River Laboratories and housed in a pathogen-free animal facility on a 12-h dark/light cycle.

The clinical studies were approved by the ethics committee of the University Hospital, Bratislava, Slovakia; Comenius University; and the University Hospital of Heidelberg, Heidelberg, Germany. Both studies adhere to the ethical guidelines of the 2000 Declaration of Helsinki, and all study participants provided witnessed written informed consent before entering the study. Details of the study groups and the techniques for clinical parameter assessment have previously been published (14,15). Vim expression was analyzed in 24 samples, and the other targets were analyzed in a subpopulation of 16 patients.

Fat tissue from mice was dissected, minced, and washed with 100 mL PBS over a 40-µm filter. Tissue was gently shaken for 1 min and washed on the 40-µm mesh with 200 mL warm PBS and then centrifuged to remove erythrocytes and debris. One gram of subcutaneous (SE) or visceral (VE) adipose tissue explants was cultured in serum-free media at 37°C, washed every 2 h for 8 h followed by one washing step after 24 h, and then cultured on insert membranes (Supplementary Fig. 1A).

We established a coculture system using a cell culture insert (BD Falcon) with a membrane pore size of 3.0 µm to allow free exchange of media. Isolated subcutaneous (SA) or visceral (VA) adipocytes, SE, or VE were cultured on the insert, and the secretome sample was harvested every 24 h for 120 h (Supplementary Fig. 1A), centrifuged, and stored at −80°C. At the end of the experiment, samples were lysed to analyze cell viability by measuring lactate dehydrogenase activity (Supplementary Fig. 1B). 3T3-L1 cells and primary subcutaneous and visceral preadipocytes were cultured on the bottom chambers of sixwell plates at the density of 4.66 × 10⁵, 8.7 × 10⁶, and 8.7 × 10⁶ cells/well as described previously (16,17). SA, SE, VA, or VE were cultured separately on the insert for 48 h. Preadipocytes were cultured for 3 days and then SA, SE, VA, or VE were transferred to the culture plates and cocultured together (Supplementary Fig. 2A).

To determine the protein composition of SA, SE, VA, or VE secretion, the secretome sample was concentrated by ultrafiltration (3-kDa cut off; Millipore) at 4°C and silver stained according to the manufacturer's protocols (Invitrogen) (Supplementary Fig. 1C and D).

Fast protein liquid chromatography (FPLC) was used for separation of total proteins mixtures. The concentrated secretome sample was adjusted to buffer A (20 mmol/L Tris-HCl, pH 8.0) and then filtered using a 0.22-μm filter. The column was equilibrated with five column volumes of buffer A. A total of 1 mg protein for each sample was separated via a HiTrap Q HP 1-mL anion-exchange column. The samples were eluted with 5–10 column volumes by gradient elution at 4°C, the highest concentration being 1 mol/L NaCl. Each sample was fractionated into four fractions (5 mL). The fractions were desalted and concentrated by ultrafiltration at 4°C.

In brief, 60 µg SA, SE, VA, or VE fractions (F1, F2, and F3) was precipitated with 16.6 µL 100% TCA and washed with cold acetone. The samples were dried and dissolved in 40 µL buffer (10 mmol/L Tris/2 mmol/L CaCl2, pH 8.2) plus 5 µL trypsin (100 ng/µL in 10 mmol/L HCl) plus 5 µL Rapi Gest (1% in water). The samples were incubated overnight at 37°C and then incubated for 30 min with 3 µL of 1 mol/L HCl at 37°C, centrifuged, desalted, and dried. Samples were dissolved in 15 µL of 0.1% formic acid and transferred to autosampler vials for liquid chromatography–tandem mass spectrometry (MS/MS) analysis. Database searches were performed by Mascot. Proteins were measured by label-free quantification, and MS/MS tolerance was set at 0.08 Da. MS/MS-normalized spectra were acquired from Scaffold 3. Protein identifications were considered at protein threshold ≥99% and if they contained five or more peptides for analysis. The quantitative data obtained were further analyzed using the Spotfire DecisionSite program (version 9.1.1; TIBCO). For classification of secreted and nonsecreted proteins, DAVID Bioinformatics Resources 6.7 (18), SecretomeP 2.0 Server (19), and SignalP 3.0 (13) were used.

Subcutaneous, visceral, and 3T3-L1 preadipocytes were grown and induced to differentiate with induction medium containing 10 mg/mL insulin, 10 mmol/L dexamethasone, 11.5 µg/mL isobutylmethylxanthine, and 10% FBS as described previously (16,17,20,21) for a total of 10 days. For validation of the preadipocyte differentiation method, confluent cells were treated with or without rosiglitazone during induction of differentiation and adipocyte formation was quantified using automated image-based analysis (16,17,20,21).

For the RNA interference assay, 25 μL Optimem and 100 nmol/L small interfering (si)RNA were mixed on collagen-coated 96-well plates. Lipofectamine RNAiMax was used for transfection. Preadipocytes were cultured for 24 h with siRNA; afterward, the medium was changed to regular medium. After the differentiation procedure, automated differentiation quantification was performed (16,17,20,21). Custom library siRNA was purchased from Dharmacon (Wohlen, Switzerland) and Microsynth (Balgach, Switzerland) (sequences available upon request).

For expression of soluble protein candidates, cDNA sequences were cloned into the pSecTag System (Invitrogen). For transfection of human embryonic kidney (HEK)293A cells, 800 µL Optimem and 30 µL Lipofectamine 2000 were mixed and 20 µg plasmid DNA was diluted in 1 mL Optimem. Transfection efficiency was visualized using green fluorescent protein. Media was changed 24 h posttransfection and harvested after 24 h.

Total RNA was extracted using TRIzol reagent according to the manufacturer's protocol (Invitrogen). Total RNA (1 µg) was converted into first-strand cDNA using the High Capacity cDNA Reverse Transcription kit (Invitrogen). Real-time PCR quantification was performed using fast SYBR Green master mix and gene specific primer sets (sequences are available upon request).

Secretome samples (1 mL) were added to 20 mg dextran-coated charcoal and then mixed overnight with an overhead mixer at 4°C. Charcoals were removed from the suspension by centrifugation at 4°C.

All data are expressed as mean ± SEM. The significance was determined using a two-tailed Student t test. Association between clinical parameters and adipokine expression was calculated using either Pearson or Spearman rank correlation.

To study the role of adipocyte secretion on adipocyte differentiation, we used primary adipocytes isolated from SA and VA as well as SE and VE (Supplementary Fig. 1A). We quantified the effect of secreted factors from SA (SAFs), SE (SEFs), VA (VAFs), or VE (VEFs) on adipocyte differentiation, using 3T3-L1 fibroblast, subcutaneous, and visceral preadipocytes (Supplementary Fig. 2A). Coculture of SA, SE, VA, or VE inhibited differentiation of 3T3-L1, subcutaneous, or visceral preadipocytes (Supplementary Fig. 2B–D). A significant reduction of adipocyte differentiation was also observed by using conditioned media, verifying the coculture approach (Fig. 1A–C). To assess different markers of adipocyte differentiation, we measured gene expression of known adipocyte markers, as well as adiponectin secretion, which was reduced upon coculture or addition of conditioned media (Supplementary Fig. 2E–H [data not shown]). The inhibitory effects of SAFs, SEFs, VAFs, or VEFs on adipocyte differentiation could be completely reversed by heat inactivation (Fig. 2A–F), while charcoal extraction reduced adipocyte differentiation to a lesser extent (Fig. 2A–F). These data indicate that adipose tissue secretes a spectrum of molecules, including heat-sensitive factors, which may regulate adipocyte differentiation.

For identification of specific factors present in SAFs, SEFs, VAFs, or VEFs that influence adipocyte differentiation, secretome samples were fractionated by FPLC. Proteins were detected in F1–F3, but not in F4 (Supplementary Fig. 3). For assessment of the activity of the isolated fractions on adipocyte formation, cultured 3T3-L1 cells were treated with these fractions as depicted in Fig. 3. Treatment with F1 and F3 from SA caused an increase in adipocyte formation, whereas F2 decreased differentiation (Fig. 3A). In contrast, F1–F3 from SE reduced adipocyte differentiation. Treatment with F1 and F3 from VA led to significant induction of adipocyte differentiation, whereas F2 from VA reduced adipocyte differentiation. F3 from VE showed a trend toward increased adipocyte differentiation, while F1 and F2 resulted in reduced differentiation. Notably, heat inactivation completely reversed the stimulatory or inhibitory effects of all fractions tested (Fig. 3B–E). Taken together, these data demonstrate that the profile of factors secreted from SA, SE, VA, or VE is more complex than anticipated and is composed of a variety of different signals, which ultimately affect the cross talk between adipose tissue and adipocyte precursors.

For identification of potential candidate proteins secreted by SA, SE, VA, or VE, which regulate adipocyte differentiation, the fractions were analyzed by mass spectrometry. Based on a fold change of normalized quantitative total spectral counts, a protein threshold of ≥99%, and five or more peptides, proteins with a ratio (SA/SE, VA/VE) of at least 1.5-fold were considered for further analysis. From all secretome fractions, a total of 164 proteins were identified (Fig. 4). Of these, 79, 59, and 29 proteins were identified in F1, F2, and F3, respectively (Fig. 4A–C and Supplementary Table 1). Proteins were classified based on four main enriched characterizations, namely, extracellular matrix (41 proteins), glycoprotein (82 proteins), cell adhesion (25 proteins), and secreted (84 proteins) (Fig. 4A–F and Supplementary Table 2). In addition, secreted proteins were classified by using SecretomeP 2.0 (19) and SignalP 3.0 (18) at a signal peptide NN ≥0.5. In total, 98 proteins from all fractions were classified as secreted (Supplementary Fig. 4A–C and Supplementary Tables 3 and 5).

To identify factors among these secreted proteins that could regulate adipocyte differentiation, we used siRNA SMARTpools. In a first comparison, we analyzed differentially secreted factors from F1 of SA versus SE and VA versus VE based on their fold change–normalized spectral count (Fig. 5A and Table 1). This comparison was used because SA- and VA-secreted adipokines induced differentiation, while SE- and VE-secreted adipokines inhibited adipocyte differentiation (Fig. 3A). As a threshold, we used a 2-fold change in expression values and a 1.3-fold change in the regulation of adipocyte differentiation (Fig. 5A and Table 1). Candidates derived from this comparison were long-chain fatty acid transport protein 1 (Slc27a1) and Vim, which decreased adipocyte differentiation upon knockdown. In contrast, ablation of aspartate aminotransferase, mitochondrial (Got2), carboxypeptidase Q (Cpq), and IL-1 receptor-like 1 (Il1rl1/ST2) promoted differentiation (Fig. 5A and Table 1).

Since F2 from all four conditions tested inhibited adipocyte formation (Fig. 3A), we focused on common factors in all four depots (Fig. 5B and Table 1). Knockdown of Sparc, carboxylesterase 1D (Ces1d), fibulin-2 (Fbln2), or Cpq expressions induced adipocyte differentiation (Fig. 5B and Table 1).

Similar to F1, F3 from SA and SE exerted opposite effects on adipocyte differentiation (Fig. 3A). Candidates derived from this comparison (Fig. 5C and Table 1) were ceruloplasmin (Cp), which inhibits adipocyte differentiation when ablated, while Galectin-3–binding protein (Lgals3bp) or Serotransferrin (Tf) induced differentiation upon knockdown.

As both VA and VE from F3 induced adipocyte differentiation (Fig. 3A), we plotted the absolute spectral counts against the effect of the ablated proteins on adipocyte differentiation (Fig. 5D and Table 1) and identified Cp and extracellular matrix protein-1 (Ecm-1) whose knockdown reduced adipocyte differentiation.

Taken together, our results identify several different factors secreted either from adipose tissue or from mature adipocytes that positively or negatively regulate adipocyte differentiation via a paracrine feedback loop.

To elucidate whether these candidates are expressed in adipocytes or other cell types, we analyzed the expression during adipocyte differentiation and in a tissue panel (Supplementary Figs. 5 and 6). Slc27a1, Cpq, Got2, Vim, Il1rl1, Sparc, Ces1d, Fbln2, Cp, Lgals3bp, Tf, and Ecm1 were found to be expressed both in undifferentiated and differentiated 3T3-L1 cells, while only Vim, Cpq, Got2, Sparc, Lgals3bp, and Ecm1 were increased upon adipocyte differentiation. Cpq, Slc27a1, Got2, Vim, Il1rl1, Sparc, Ces1d, and Tf were expressed in different organs including white and brown adipose tissue.

To obtain functional proof that these factors act via cross talk between adipocytes and preadipocytes, we ablated the candidates identified above through reverse transfection in mature adipocyte by siRNA-mediated knockdown (Fig. 6A). Cultured preadipocytes were treated with conditioned media from these cells during induction of differentiation, and the effects on adipocyte differentiation were quantified. Knockdown of Cpq, Got2, Il1rl1, Sparc, or Lgals3bp in mature adipocytes enhanced the proadipogenic capacity of the secreted fraction (Fig. 6B), suggesting that these factors indeed represent inhibitory molecules derived from mature adipocytes. In contrast, treatment of preadipocytes with conditioned media obtained after knockdown of Slc27a1, Vim, Cp, or Ecm1 showed a reduction of adipocyte formation, suggesting that these are stimulatory factors, which enhance adipocyte differentiation. In contrast to the data obtained from knockdown in differentiating cultures, we did not observe any effects for Ces1d, Fbln2, or Tf (Fig. 6B).

To complement our data on ablation of candidate proteins, we next assessed how these factors influenced adipocyte differentiation when their levels were increased during differentiation. We excluded Ces1d and Fbln2, as we did not observe any effect in the mature adipocyte coculture assay. We used either recombinant proteins (where available) or HEK293A cells, which expressed the secreted candidates as measured by immunoblot using the V5 tag (data not shown). In accordance with the results presented above, Vim treatment induced adipocyte formation (Fig. 6C), while addition of Sparc recombinant protein did not show any effect (data not shown). Furthermore, Slc27a1- and Cp-enriched supernatant induced adipocyte formation (Fig. 6D). In contrast to the data obtained from mature adipocytes, we did not observe any significant changes in adipocyte differentiation when using conditioned media from HEK293A cells enriched for Ecm1, Lgals3bp, Cpq, Got2, Lgals3bp, or Tf expression, although Cpq and Got2 showed a trend toward inhibition (Fig. 6D), which might be due to the fact that all proteins tested are already abundant in the tested cells. To quantify the effect of Il1r1 signaling, we used its ligand IL-33, which inhibited adipocyte differentiation, while addition of an anti–IL-33 antibody induced adipocyte differentiation (Fig. 6E–F).

Since we could show that several adipokines released from mature adipocytes regulate de novo preadipocyte formation, we quantified their expression in human subcutaneous adipocytes of lean subjects, obese subjects, and obese subjects with diabetes. Subcutaneous fat biopsies were obtained from two independent cohorts of male obese and lean subjects. To assess the relation of each factor to different metabolic parameters, we analyzed the correlation with adipocyte size, BMI, and insulin sensitivity as measured by clamp. We could show that expression of Vim, Sparc, and Tf showed an inverse correlation with mean adipocyte size and BMI and a positive correlation with insulin sensitivity (Fig. 7A, E, and F). In contrast, we observed that Slc27a1 expression correlated with mean adipocyte size and BMI but not with insulin sensitivity (Fig. 7B). Got2 was inversely correlated with insulin sensitivity (Fig. 7C), while Il1rl1 expression positively correlated with BMI (Fig. 7D). The other three candidates did not show any correlation with either value measured (data not shown). Furthermore, we extended our study to a cohort of obese male and female patients that underwent weight loss through dietary restriction. Expression analysis of the identified factors in subcutaneous adipose tissue biopsies showed no sex-specific differences (Supplementary Fig. 7A); interestingly, however, Slc27a1 expression was significantly increased, while Sparc expression was reduced upon weight loss (Supplementary Fig. 7B).

It is well-known that factors secreted from adipocytes and adipose tissue affect the function of various organs such as muscle, liver, and vasculature. Consistent with our results, coculture of primary human adipocytes has been reported to inhibit differentiation of preadipocytes (22). In contrast, coculture of rat primary adipocytes or adipose tissue led to an induction of adipocyte differentiation (23) and conditioned medium collected from differentiating human adipose tissue stroma cells showed a positive effect on adipocyte differentiation (24). These conflicting results demonstrate that the reciprocal regulation of adipocyte differentiation by secreted factors from adipose tissue is more complex and warrants further investigation.

Given that the in vitro adipocyte differentiation model we used here used chemical induction, it is possible that a bias toward the identification of inhibitory factors was introduced. Nevertheless, modulatory factors enhancing differentiation can be identified in such a context (25), which is also supported by our reciprocal findings from knockdown and overexpression studies.

The human studies were carried out with two sets of individuals who underwent clinical phenotyping including adipose tissue biopsies. While such an approach allows us to directly assess gene expression in individual patients, it precludes the recruitment of an extensive cohort, which is reflected in the mild associations observed here.

Among the 98 proteins identified by mass spectrometry, adiponectin, cystatin C, resistin, adipsin, metalloproteinase inhibitor 1, chemerin, Vim, Sparc, and Vitamin D-binding protein are known adipokines (13,26–29). Taking into account which proteins have been reported to be secreted by adipocytes and adipose tissue, we identified Cpq, Got2, and Ecm1 as potential novel adipokines.

Got2, a known cell-surface fatty acid transporter (30), is secreted from adipose tissue, and ablation of Got2 in mature adipocytes enhanced adipocyte differentiation. We could not confirm this finding in our assay using overexpression of Got2, which could be due to the fact that the adipocytes expressed this protein in sufficient amounts. Our data are in line with a recent report that demonstrated that Got2 is actively secreted by cancer cells (31). Since Got2 acts as a negative regulator of adipocyte differentiation and correlated with adipocyte size, it is possible that patients with high Got2 levels have impaired adipocyte formation.

Similar to Got2, we identified Lgals3bp, Cpq, and Sparc as negative regulators of adipocyte differentiation. We could not confirm their negative regulation when using an overexpression paradigm, which could be due to the aforementioned effects. Lgals3bp has previously been reported to be secreted from human visceral adipose tissue (32); however, its depot-specific effects, as well as its function in metabolic control, will need to be analyzed in more detail. Cpq has not been reported to be a secreted protein, and as no reports exist on the role of Cpq in metabolism, further experiments will be needed to extend our findings. Our data on Sparc are consistent with previous reports (33,34). Interestingly, Sparc-null mice exhibit increased adiposity due to an increase in adipocyte number (35), which is consistent with our findings that Sparc in patients correlates with adipocyte size and BMI.

Il1rl1/ST2 exists both as a membrane form that binds and signals IL-33 and as a soluble form that inhibits IL-33 by functioning as a decoy receptor (36). Knockdown of Il1rl1/ST2 expression induced differentiation, and conversely, treatment of 3T3-L1 cells by IL-33, the ligand of Il1rl1/ST2, markedly reduced adipocyte differentiation, which is in line with recent findings (37). The observed activating effects using supernatants of adipocytes in which Il1rl1/ST2 was ablated may be due to the secreted soluble form of Il1rl1/ST2; however, further studies will be required to quantify the exact contributions. Conflicting data exist with regard to the role of the IL-33/Il1rl1/ST2 axis in metabolism. On one hand, IL-33 has been considered to prevent adipose tissue inflammation and thus reduce metabolic complications (38,41). However, other studies have linked increased Il-33 levels to morbid obesity and the development of diabetes (40,41). Since IL-33 has been mainly implicated in immune cell function, our findings might provide a link between inflammation and adipocyte differentiation.

We identified Slc27a1, Cp, Vim, and Ecm1 as possible positive regulators of adipocyte differentiation. Slc27a1 is highly enriched in SA and VA F1 and induces differentiation as shown by both ablation and overexpression studies. In accordance with our data, Slc27a1 has been previously reported to be secreted from adipocytes and, furthermore, has been shown to increase lipid droplet accumulation (42,43). Interestingly, similar to our findings, a negative correlation between Slc27a1 expression and BMI in obese women has been reported (44), while Slc27a1 knockout mice are protected from high-fat diet–induced lipid accumulation in skeletal muscle (45).

Recent studies have reported Cp as a novel adipokine with increased expression in adipose tissue of obese subjects (46). We here demonstrate a differentially increased secretion of Cp by adipocytes and a positive-feedback loop effect on adipocyte differentiation both when Cp is ablated in mature adipocytes and when extracellular Cp levels are induced. Thus, it is possible that Cp contributes to whole-body energy homeostasis by modulating de novo adipocyte formation.

Vim has been reported to be secreted by visceral adipose tissue (27), and our data are in agreement with previous results demonstrating that Vim treatment induced de novo adipocyte formation. Since a larger mean adipocyte size is correlated with insulin resistance (47), whereas a greater number of small adipocytes is associated with insulin sensitivity (48), it can be speculated that circulating Vim levels regulate adipocyte formation and thereby contribute to whole-body energy homeostasis. Vim expression is induced during adipocyte formation, suggesting that it might be a counterregulatory effect that would regulate the expansion of adipose tissue by controlling adipocyte differentiation.

In conclusion, our integrative technique identified novel differentially secreted adipokines that positively or negatively regulate adipocyte differentiation. Since de novo adipocyte formation in vivo is influenced by different factors and stimuli (49), future studies on the in vivo contribution of the identified factors to the maintenance of adipose tissue homeostasis are needed, which may, in turn, lead to new insights on their paracrine and endocrine effects on lipid storage and energy homeostasis.

Acknowledgments. The authors thank Ina Stützer (ETH Zurich) for expert statistical advice.

Funding. This work was supported by the European Research Council (AdipoDif) and by the Schweizer Nationalfonds (SNF) (31003A_140926).

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

Author Contributions. T.D.C. and C.W. were responsible for the concept and design of the study, data analysis, and performing the experiments and writing of the manuscript. L.G.S. performed mRNA expression analyses. M.B., G.R., J.U., and B.U. recruited human subjects and provided the patient material of the lean subjects, obese subjects, and obese subjects with diabetes. E.K. performed the overexpression experiments on candidate factors. O.D. produced adipokines and helped write the manuscript. C.W. 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.