A TRAIL-TL1A Paracrine Network Involving Adipocytes, Macrophages, and Lymphocytes Induces Adipose Tissue Dysfunction Downstream of E2F1 in Human Obesity

Obesity, defined as abnormal or excessive fat accumulation that presents a risk to health, was recognized as a disease by the American Medical Association in 2013 (1). However, with its increasing global prevalence, obesity is highly heterogeneous in the degree of health risks it imposes and presents itself differently even in patients with similarly elevated BMI. With the increasing quest for personalized health care, there is an urgent need to better subtype obesity (2). Compared to persons with cardiometabolically benign obesity (arguably termed “healthy/insulin-sensitive/metabolically normal obesity”), in persons with “high–cardiometabolic risk obesity” adipose tissues differ in distribution, morphology, cellular composition, molecular patterns, and function (3). Initial findings indicate that such differences, particularly morphological (adipocyte size, fibrosis) may be of assistance in subphenotyping of obesity, not only in cross-sectional analyses, but also in prediction of the clinical response to obesity interventions (4). Yet, molecular fingerprinting of adipose tissue that could serve to better stratify, prioritize, or even personalize clinical care is largely still lacking.

We identified a transcriptional molecular network in visceral adipose tissue (VAT) of patients with metabolically dysfunctional obesity (5,6). A central “hub” in this network is the transcription factor E2F1, which, when upregulated, drives aberrant expression and activation of a MAP kinase signaling cascade (ASK1-MKK4-p38 MAPK/JNK), and of autophagy (5,6). E2F1, like other E2F family members, is a major regulator of cell-cycle progression. In adipose tissue, however, E2F1’s functions are not well characterized. In obesity, elevated adipose tissue E2F1 expression originates mainly from the adipocytes (6), while during adipogenesis E2F1 acts as an adipogenic stimulator (7) and negatively regulates energy expenditure and mitochondrial activity (8). In myocytes, E2F1 favors glycolysis over glucose oxidation by activating pyruvate dehydrogenase kinase 4 (PDK4) (9). In pancreatic β-cells, it upregulates the expression of Kir6.2 and thus glucose-stimulated insulin secretion (10). Yet, E2F1-knockout mice are not diabetic due to increased insulin sensitivity (11), suggesting a role for E2F1 in decreasing insulin responsiveness. In the liver, E2F1 overexpression supports lipid accumulation and steatosis, evidently tied to obesity and insulin resistance (12), and also drives hyperglycemia by facilitating gluconeogenesis (13). E2F1 is also highly activated in multiple inflammatory states (14), acting upstream and/or downstream of inflammatory signals, offering a putative link between obesity, low-grade inflammation, and cardiometabolic complications.

Consistent with this notion, we found that VAT E2F1 protein and mRNA levels correlated with multiple clinical indicators of high cardiometabolic risk, an effect mediated by direct E2F1 binding to promoter regions of ASK1 and autophagy genes (5,6). Yet, multivariate analyses that adjusted for the activation of these putative effector genes of E2F1 suggested that additional pathways mediate the link between increased VAT E2F1 and metabolic dysfunction (5,6). In this study, we used unbiased transcriptomic analysis of E2F1^high versus E2F1^low human VAT (hVAT) to delineate such possible pathways. We followed E2F1^high-associated tumor necrosis factor superfamily (TNFSF) members that this analysis revealed, and we uncovered a putative intricate intercellular paracrine network within adipose tissue, involving adipocytes, macrophages, and T cells, linking high VAT E2F1 expression with adipose tissue dysfunction.

Participants were from the Beer-Sheva, Israel (n = 123), and Leipzig, Germany (n = 421), cohorts (Table 1) with use of coordinated procedures as previously described (5). Briefly, following approval by the ethics committees of the two centers and obtaining of written informed consent, participants (18–75 years old) were recruited before undergoing elective abdominal surgeries (bariatric or other elective procedures). After overnight fasting, blood samples were drawn and analyzed by the clinical biochemistry and endocrinology laboratories. Visceral (omental) adipose tissue biopsies were obtained during surgery and immediately delivered to the laboratory, where they were processed for mRNA or protein expression using coordinated procedures, as previously described (5). For obtaining human adipose tissue explants, tissue samples were carefully cut and cultured in MEM-Alpha containing 4.5 mmol/L glucose, 10% FBS, 2 mmol/L l-glutamine, and 100 units/mL penicillin-streptomycin for 24 h recovery. Human TRAIL (hTRAIL) (375-TEC-010; D&R Systems) 25 or 100 ng/mL was added to fresh serum-free media for 24 h treatment. Adipocyte and stromal vascular (SVF) fractions were prepared by collagenase (C6885-5G; Sigma-Aldrich) digestion as previously described (15).

Human monocyte-derived macrophages (MDM) and T cells were isolated, following approval by the Soroka Academic Medical Center Institutional Review Committee, from peripheral blood samples (100 mL) of healthy volunteers, who signed a written informed consent. Peripheral blood mononuclear cells (PBMC) were obtained from PBS–2% FBS–EDTA (2 mmol/L)–diluted plasma mounted on Ficol Lymphocytes Separation Medium (50494; MP Biomedicals), incubated for 20 min with CD14 microbeads, and transferred through magnetic columns (130-050-201 and 130-042-401, respectively; Miltenyi Biotech) to isolate monocytes. To obtain MDM, we cultured monocytes in growth media (RPMI-1640, 10% FBS, 2% l-glutamine, and 1% antibiotics) containing M-CSF 50 ng/mL (300-25; PeproTech), GM-CSF 50 ng/mL (300-03; PeproTech), or M-CSF + IL4 20 ng/mL for M0/M1/M2 polarization, respectively. Following a 6-day culture period, functional assays were conducted as further described below. The nonmonocyte (CD14⁻) cell fraction was cultured on anti-human CD3 antibody (Ab) (300314; BioLegend)–coated plates to isolate T cells, with similar growth media. Chub-S7 cells were cultured in DMEM containing 4.5 mmol/L glucose, 20% FBS, 2 mmol/L l-glutamine, and 100 units/mL penicillin-streptomycin, until 24 h postconfluence. Differentiation was induced by addition of 1 mmol/L dexamethasone, 0.5 mmol/L 3-isobutyl-1-methylxanthine (IBMX), 10 mg/mL insulin, and 10 mmol/L rosiglitazone to the medium for 14–21 days, until the appearance of adipocyte morphology, at which stage cells exhibited robust upregulation of adipocyte genes (Supplementary Fig. 1). When differentiated, cells were cultured with either control serum-free media or serum-free media containing human recombinant (hr)TL1A (1319-TL-010; D&R Systems) or hrTRAIL for 24 h. TL1A and TRAIL concentration range used was based on previous cell-culture studies in the literature (16,17). For insulin stimulation, Chubs-S7 were serum starved for 24 h, PBS washed, and stimulated with insulin at 37°C for 7 min.

Preparation of tissue/cell lysates and antibodies used has previously been described (6). For E2F1 protein determination, we used anti-E2F1 monoclonal Ab by GeneTex (GTX-70154; GeneTex, Irvine, CA).

RNA extraction, cDNA generation, and RT-PCR amplification were conducted as previously detailed (6). RNA sequencing (RNA-seq) was performed at the Technion Genome Centre (Israel Institute of Technology, Haifa, Israel) using Illumina technology. Bioinformatics analysis was performed using the NeatSeq-Flow platform (18) and R. Raw sequences were quality trimmed and filtered using Trim Galore!, and alignment of the reads to the human genome (assembly GRCh38.p10) was done with STAR (19). Number of reads per gene per sample was counted using RSEM (20). Raw read counts were loaded into R and analyzed with DESeq2 (21). For quality assessment diagnostic plots, counts were log2 transformed, normalized, and subjected to variance-stabilizing transformation. Principal component analysis showed that sample number 5 was an outlier, highly separable from all other samples. Therefore, this sample along with its matched pair number 6 was excluded from further analyses. The design formula for paired statistical testing included the matched pair number in addition to the E2F1 condition (“∼Matched_pair+E2F1”). For each gene, P value, false discovery rate (FDR)-adjusted P value, and fold change were computed for the E2F1 effect. Fold change values were expressed in linear scale. To expand the search for TNF/TNFSF genes associated with E2F1 level, we carried out an enrichment test versus a manually assembled data set of 54 TNFSF genes using Gene Set Enrichment Analysis (GSEA) standalone software. Analysis was done using DESeq2-normalized counts of all genes considering absolute change values between high to low E2F1.

Leptin and adiponectin levels in conditioned media of Chub-S7 and in human serum were measured as previously (6). Lactate dehydrogenase (LDH) release from Chub-S7 adipocytes was measured using the Lactate Dehydrogenase Activity Assay Kit (MAK066-1KT; Sigma-Aldrich). Released LDH was calculated as percentage of total (medium / medium + cells). hTRAIL secretion to conditioned media was measured using Immunoquantitative TRAIL ELISA (RayBio). Lipolysis was measured as previously described (22), with adenosine deaminase (ADA) at 1 µU/mL and insulin’s antilipolytic effect after 24 h insulin starvation.

MDM were differentiated as described in 96-well μClear, black, plates (Greiner-Bio One). M0 MDM were treated with control or TL1A 25 ng/mL serum-free media for 24 h. Lipid accumulation in fatty acid–enriched conditions was done by adding 10 mmol/L oleic acid for the final 2-h incubation as previously described (23). Cells were fixed with 4% formaldehyde followed by staining with BODIPY 493/503 (1 μg/mL) (D3922; Thermo Fisher Scientific, Waltham, MA) and DAPI (5 ng/mL) (62248; Thermo Fisher Scientific) for 20 min at room temperature and then washed with Ca⁺²/Mg⁺²-supplemented PBS (02-020-1A; Biological Industries). Images were acquired in a fully automated, unbiased manner using a ×40 wide angle lens–equipped microscope (Operetta; PerkinElmer, Waltham, MA). Statistical analysis was done using Columbus software (PerkinElmer).

After 3 h stimulation with or without hTRAIL, SVF cells were washed with FACS buffer and stained for membranal antigens (BioLegend, San Diego, CA) anti–CD3-FITC, CD4-BV510, CD8-AF700, and viability dye 780. Then, after 20 min paraformaldehyde fixation, cell were permeabilized and stained with anti–TL1A-PerCP-cy5.5 following the instructions of eBioscience’s FoxP3 intracellular staining kit. Samples were analyzed by a Beckman Coulter CytoFLEX flow cytometer and FlowJo software.

Differentiated M0 MDM, generated as described, were cultured on permeable inserts with 0.4-μm high-density pores (353494; Falcon) and treated with control versus TL1A 25 ng/mL serum-free media for 24 h. After thorough PBS washing, inserts were transferred into Chub-S7–containing plates and cocultured in serum-free Chub-S7 growth media. After 24 h, conditioned media was collected and leptin/adiponectin levels were measured as previously described (6).

For in vitro/ex vivo experiments, calculations were made using GraphPad. Statistically significant differences between groups were evaluated using unpaired Student t test/Mann-Whitney. Correlations were assessed by either Pearson or Spearman rank-order correlation tests as indicated. For human cohort data, statistical analysis was done using SPSS Statistics (version 20). Correlations were assessed using Pearson correlation test. Statistically significant differences in clinical parameters between groups were evaluated using unpaired Student t test.

The data sets generated in the current study are available from the corresponding authors upon request.

To identify differentially regulated pathways associated with the dysmetabolic phenotype of obese E2F1^high VAT, we used RNA-seq. To assess the range of VAT E2F1 expression, we measured E2F1 protein levels in VAT samples (as in previously described [6]) from n = 67 patients undergoing elective abdominal surgeries. VAT E2F1 levels were divided into quintiles, and the two lower and higher quintiles (the lowest and highest 40%) were defined as E2F1^low and E2F1^high, respectively (middle E2F1 expression quintile was excluded to minimize misclassification bias) (Fig. 1A). We matched pairs of patients with BMI ≥30 kg/m² from the E2F1^low and E2F1^high subgroups for age, sex, and BMI (Fig. 1B–D). Patients with VAT E2F1^low and E2F1^high (n = 8 in each group, with a 3:5 male:female ratio) (subcohort 1 [Table 2]) were, respectively, 41.25 and 40.25 years old (average), had an average BMI of 40.5 and 41.5 kg/m², and were comparable in terms of blood pressure and diabetes status.

Matched statistical analyses of the RNA-seq data, using an FDR-corrected P value of 0.05, revealed 73 differentially expressed (DE) (>1.3-fold difference) genes between the two molecularly defined subgroups, (Fig. 1E). Among the most DE genes, E2F1^high samples displayed significant alterations in TNFSF factors including TNFSF10 (TRAIL) and TNFRSF25 (DR3) (Fig. 1F and G, respectively), previously linked to adipocyte and macrophage dysfunction (24–28). We validated the association between E2F1 and TRAIL or DR3 in an independent cohort (subcohort 2 [Table 2]) of n = 48 patients with obesity (Fig. 1H and I), of whom 11 pairs of patients could also be matched as in our initial analysis (Supplementary Fig. 2). These two analyses largely recapitulated the initial results of the RNA-seq. Moreover, hVAT explants secreted TRAIL proportionally to E2F1 protein expression (Fig. 1J). To expand the search for TNFSF that may be enriched among the DE genes based on VAT E2F1 levels, we used the GSEA algorithm (Fig. 1K). Among a predefined list of TNFSF members, 19 “leading edge” genes, i.e., TNFSF genes that “crowded” at the top of a ranked list of the entire gene set, were identified. Interestingly, in addition to TRAIL and DR3, TRAIL receptors DR4 and DR5 (TNFRSF10A/B) and TL1A (TNFSF15 [ligand for DR3]) emerged as significantly enriched—all but DR3 more highly expressed in VAT E2F1^high compared to VAT E2F1^low. Collectively, with use of an unbiased transcriptomic approach, these results uncovered a possible mechanistic link between E2F1 and TNFSF members TRAIL and TL1A in hVAT.

The association between E2F1 and TNFSF members suggests that these factors may regulate each other, either directly or indirectly. We initially hypothesized that the TNFSF cytokine TRAIL, which was already shown to alter adipocyte function (26–28), could induce E2F1 expression in adipocytes, the cell fraction in which E2F1 was shown to be upregulated (6). Yet, stimulation of hVAT explants with recombinant hTRAIL did not induce E2F1 expression (Fig. 2A). The plausibility of the reverse regulation—i.e., that E2F1 regulates TRAIL—was supported in human embryonic kidney HEK293 cells, in which overexpression or siRNA-mediated knockdown of E2F1 increased or decreased, respectively, TRAIL (Supplementary Fig. 3A and B). Moreover, possible direct regulation of TRAIL gene expression by E2F1 was proposed by promoter prediction analysis that revealed at least three strong putative E2F1 binding sights in the minimal promoter of hTRAIL (Supplementary Fig. 3C). Between these two E2F1-associated TNFSF members, TRAIL dose dependently increased TRAIL and TL1A mRNA levels in hVAT explants (Fig. 2A), suggesting a possible autoregulatory positive-feedback loop. To further delineate this loop, particularly TRAIL-induced TL1A, and its possible implications in adipose tissue, we assessed whether these TNFSF members and their respective receptors similarly originate from the adipocyte and/or the SVF. Following collagenase digestion of hVAT, the adipocyte cell fraction expectedly expressed 200-fold more adiponectin (ADIPOQ) and only 15% the CD68 level compared with SVF, confirming a satisfactory fractions separation (Fig. 2B). Quantitative (q)RT-PCR analysis revealed that while TRAIL originated from both tissue fractions, TL1A was almost exclusively expressed in SVF (Fig. 2C and D). Interestingly, SVF TL1A expression correlated only with adipocyte TRAIL (R² = 0.66, P = 0.048) (Fig. 2E), not with SVF TRAIL (R² = 0.21, P = 0.35), suggesting that TRAIL derived from adipocytes may regulate SVF TL1A. DR3 (TL1A receptor) and DR4 and DR5 (TRAIL receptors) were detectable in both fractions, though significantly more so in SVF (Fig. 2F).

To uncover which specific SVF cell type responded to TRAIL by increasing TL1A expression, we first used human peripheral blood mononuclear cells (PBMC). Unpolarized (M0), M1-polarized, or M2-polarized human MDM did not exhibit TRAIL-induced TL1A upregulation (Fig. 2G). Similarly, for better representation of human adipose tissue macrophages (ATM), MDM differentiated into macrophages in the presence of hVAT-conditioned media were generated, and they, too, did not significantly increase TL1A expression in response to hTRAIL stimulation (Supplementary Fig. 3D). In contrast to MDM, T cells isolated from the CD14⁻ cell population of PBMC exhibited dose-dependent hTRAIL-induced TL1A (Fig. 2G). Additionally, TRAIL treatment augmented TRAIL receptor (DR4/DR5) expression in T cells (Supplementary Fig. 4). These findings in PBMC-derived cells prompted us to test TL1A induction by TRAIL in bona fide ATM and hVAT T cells. SVF cells isolated from hVAT were stimulated with hTRAIL, stained for cell-specific markers and TL1A protein, and assessed by flow cytometry. Consistent with the PBMC-derived cells, while CD14⁺ ATM did not exhibit TRAIL-induced TL1A upregulation, adipose tissue CD3⁺ T cells did (Fig. 2H and I). Furthermore, this response was apparent in all CD4/CD8 subtypes of adipose tissue T cells—in particular, the double-positive cells (Fig. 2J), a subpopulation of T cells shown to be particularly cytokine secreting (29). Thus, adipose tissue T cells, but not ATM, likely contribute to TRAIL-induced TL1A in VAT.

Possible contribution of TRAIL to adipose tissue dysfunction in obesity has previously been reported (26). Given our results so far, we assessed whether this could be mediated by TL1A. We used Chub-S7, a human preadipocyte cell line that readily differentiates into adipocyte-like cells (Supplementary Fig. 1). First, given the proapoptotic action of many TNFSF factors, we verified that TL1A (up to 25 ng/mL) did not increase adipocyte cell death (Fig. 3A). Human Chub-S7 adipocytes treated for 24 h with hTL1A dose dependently decreased adiponectin and increased leptin secretion, resulting in significantly reduced secreted adiponectin-to-leptin ratio (Fig. 3B–D). Additionally, TL1A-treated cells exhibited a significantly attenuated insulin-stimulated Akt and GSK phosphorylation, suggesting induction of insulin resistance (Fig. 3E–G). Indeed, metabolically, TL1A pretreatment stimulated basal lipolysis and perturbed the capacity of insulin to inhibit isoproterenol-stimulated lipolysis (Fig. 3H). Finally, human adipocytes treated with TL1A exhibited increased expression of E2F1 mRNA (Fig. 3I), albeit non-dose-dependently. This suggests that TL1A, even at low doses, may contribute to elevated adipose tissue E2F1 expression in human obesity. Collectively, TL1A, possibly released by T-cells in E2F1^high VAT, can induce adipocyte dysfunction without increasing adipocyte cell death and possibly fuels a vicious cycle by contributing to a VAT milieu that upregulates adipocyte E2F1.

ATM are thought to contribute to whole adipose tissue dysfunction in obesity, specifically when displaying a proinflammatory profile and becoming lipid laden (15,30). Interestingly, TL1A was previously reported to support macrophage proinflammatory activation, cholesterol accumulation, and foam cells formation (24), without increasing cell death. Consistently, human MDM exhibited increased neutral-lipid accumulation following 24 h exposure to hTL1A—interestingly, more so in the absence (∼3.5-fold) than in the presence (25% increase) of oleic acid added to the media (Fig. 4A and B, respectively). The greater effect in the absence of oleic acid suggests that TL1A supports de novo lipogenesis (Fig. 4C). This was supported by the finding that C75, a fatty-acyl synthase (FASN) inhibitor, significantly inhibited (by ∼75%) TL1A-induced lipid accumulation in MDM in the absence, but not in the presence, of oleic acid (Fig. 4D). Inhibitors of steps common to both de novo lipogenesis and fatty acid reesterification (Fig. 4C) invariably inhibited TL1A-induced lipid accumulation in MDM (Fig. 4D). Complementarily, TL1A induced the expression of glucose transporter GLUT1 and FASN and DGAT1 but not of the lipid transporter CD36 (Fig. 4E). These data suggest that TL1A’s induction of de novo lipogenesis is mediated via upregulation of both exclusive de novo lipogenic (FASN) and common enzymes to fatty acid reesterification. Moreover, TL1A may rely on these two lipogenic pathways in macrophages depending on the availability of fatty acids.

Additionally, in human MDM, TL1A induced increased mRNA levels of typical M1-polarization genes (IL6, MCP1) (Fig. 5A) and lower M2 markers (CD206, CD209) (Fig. 5B). Lastly, we assessed whether TL1A-prestimulated macrophages may induce adipocyte dysfunction: human MDM pretreated for 24 h with TL1A were washed and cocultured with human adipocytes (Fig. 5C) and induced decreased adiponectin-to-leptin ratio in the adipocyte conditioned media (Fig. 5D). Jointly, in human macrophages, TL1A induces lipid accumulation, supports a proinflammatory profile, and disrupts macrophage-adipocyte communication as evident by a dysfunctional adipocyte secretory profile.

Using expression data from our Beer-Sheva cohort of patients with obesity, we observed that VAT expression of TL1A, TRAIL, and E2F1 was significantly higher in those whose obesity was complicated by type 2 diabetes (T2D) than in those without diabetes (subcohort 3 [Fig. 6A–C]). Moreover, patients with high ATM lipid content expressed higher TL1A and TRAIL mRNA in whole VAT (subcohort 4 [Fig. 6D and E]). In addition, using the Leipzig cohort (n = 421, all with BMI ≥30 kg/m²) (Table 1), we found that the expression of most of the TNFSF members studied herein significantly intercorrelated (Fig. 6F), with the strongest correlation observed between TRAIL and TL1A (r = 0.646, P < 0.001), consistent with the proposition that TRAIL regulates TL1A expression. We then divided the cohort population into subgroups based on low (lower two quintiles, i.e., lower 40%) or high (upper two quintiles, i.e., upper 40%) TRAIL and TL1A expression, consistent with our approach for E2F1 (Fig. 1A). This generated four TRAIL/TL1A expression groups: low-low/low-high/high-low/high-high. Since the high-low and low-high groups were very small (n = 16 and n = 20, respectively), we compared the two extreme groups (high-high vs. low-low, n = 142 and n = 148, respectively). There were no significant differences in mean age, sex, or BMI between the groups (Fig. 6G). Yet, the high-high group exhibited significantly higher serum IL6 and leptin levels. Jointly, these analyses support the putative contribution of E2F1-related TNFSF members to adipose tissue dysfunction in human obesity.

Previous studies established a non-cell-cycle role for E2F1 in VAT, driving the expression of autophagy and a MAPK cascade, two stress pathways through which E2F1 contributes to adipose tissue dysfunction in obesity (5,6). In the current study, we used an unbiased approach to uncover additional mechanisms/mediators that could connect high expression of VAT E2F1 in human obesity with increased cardiometabolic risk. Indeed, E2F1^high VAT expressed a distinct transcriptome compared with E2F1^low VAT, independently of age, sex, and the degree of obesity, including differential expression of multiple TNFSF members. Expression of TRAIL (TNFSF10) and TL1A (TNFSF15) was enriched in patients with obesity and E2F1^high VAT, TRAIL secretion by hVAT was proportional to E2F1 expression, and tissue fractionation suggested that while TRAIL originates from both adipocytes and SVF, TL1A is predominantly expressed in SVF. TRAIL treatment of whole adipose tissue induced upregulation of both TRAIL itself and TL1A, the latter originating from VAT T cells. Since the expression of TL1A in SVF correlated only with TRAIL expression in VAT adipocyte fraction (but not SVF), we propose the following model (Fig. 7): Upregulated adipocyte E2F1 induces TRAIL, which in turn upregulates TL1A in adipose tissue T cells. Elevated VAT TL1A induces ATM inflammatory polarization and lipid accumulation and in adipocytes disrupt insulin responsiveness, lipolysis, and adipokine secretory profile (the latter both directly and via TL1A’s effect on ATM). Furthermore, elevated VAT TL1A may further contribute to upregulate adipocyte E2F1 expression, thereby completing a vicious-cycle regulatory loop. This putative E2F1-associated TRAIL-TL1A loop is likely clinically relevant, as VAT TL1A strongly correlated with that of TRAIL and the TNFSF receptors in a large human cohort of n = 421 patients. TL1A also significantly associated with circulating IL-6 and leptin levels and was elevated in those with T2D and in patients with increased ATM lipid content.

TNFα, the best-characterized TNFSF cytokine, was the first inflammatory mediator implicated in the development of adipose tissue and systemic insulin resistance in obesity (31), followed by Fas ligand and additional members (32,33). Interestingly, despite structural and functional resemblance, TNFSF members activate distinct molecular mechanisms. TRAIL was previously identified as a major effector of adipocyte function and systemic metabolic status (27,28). In preadipocytes, TRAIL exerts a proliferative effect and may thus support hyperplastic adipose tissue expansion when upregulated in obesity (34). Yet, TRAIL was also shown to inhibit early adipogenesis (28), which may contribute to hypertrophic adipose tissue expansion, and to facilitate an inflammatory transcriptome in preadipocytes (16). In mature adipocytes, TRAIL also induced an inflammatory profile (16), while impairing insulin response (27). These effects were all mediated by the proapoptotic DR4 and DR5 receptors but were not accompanied by increased apoptosis (28), demonstrating that “death receptor” signaling activates, at least in adipocytes, nonapoptotic pathways. In macrophages, TRAIL promotes cytokine production (35) and regulates macrophage autophagy (36), a process known to be altered in diseased adipose tissue and specifically in ATM (23,37). TRAIL also induces the formation of lipid-laden foam cells (25), though these were cholesterol-accumulating macrophages rather than the triglyceride (TG)-filled ones found in obese adipose tissue. Jointly, at the cellular level, current literature supports a putative role for TRAIL in driving dysfunctional cell phenotypes, which could promote adipose tissue dysfunction in obesity. Systemically, the effects of TRAIL are more complex. Serum levels of TRAIL in human patients correlate with total body fat, weight gain, and circulating cholesterol levels (reviewed in 26), and mouse models of genetic obesity show elevated expression of TRAIL and its receptors in adipose tissue (27). Serum TRAIL is interestingly lower in patients newly diagnosed with T2D but is elevated in patients with severe end-organ complications (26). Within the subpopulation of T2D patients, TRAIL presents significant correlations with BMI, TGs, and HOMA of insulin resistance, suggesting it as a marker for disease severity. Conversely, in different animal models it has been suggested that administering TRAIL can decrease lipid accumulation and proinflammatory activation in adipose tissue and even improve insulin sensitivity (38). Curiously, similar discrepancies in pathogenic versus protective roles of TRAIL were demonstrated in atherosclerosis (reviewed in 39) and in cancer (40). Altogether, though signaling via death receptors, TRAIL serves as a mediator of multiple biological nonapoptotic processes. Its pathogenic effect in adipose tissue is complex and might be imposed on adipocytes both directly and indirectly via other cell types, as suggested herein.

TL1A, another well-characterized TNFSF, has mostly been studied for its involvement in inflammatory and autoimmune diseases but has only recently been implicated in obesity (41). We found increased TL1A expression in adipose tissue in a TRAIL-dependent manner, consistent with TRAIL-TL1A coupling, as was previously shown in skin lesions of patients with diabetes (42). Additionally, involvement of TL1A in a TNFSF feed-forward loop has been proposed before in the pathogenesis of rheumatoid arthritis, in which TL1A induced elevated TNFα (and IL17) levels (43). Classically, TL1A is mainly known as a lymphocyte-driven cytokine that stimulates T and B cells (17), directing their polarization and thus modulating immune responses (44,45). TL1A’s expression in T cells is induced upon diverse inflammatory stimuli, though to our knowledge TRAIL was not implicated before. A putative central role for TL1A in autoimmunity is exemplified in inflammatory bowel disease (45), rheumatoid arthritis (46), psoriasis (43), primary sclerosing cholangitis (43), ankylosing spondylitis (47), and even airway hyperreactivity diseases (48), all of which are associated with elevated levels of both disease-site and circulating TL1A. Studies in animal models support a central mediatory role of TL1A by demonstrating that deletion/blockade of either TL1A or DR3 resulted in reduced disease severity (43,48). In obesity, also a chronic inflammatory disease, TL1A may serve as a previously unrecognized central pathogenic player. Herein, we suggest that TL1A may function as a paracrine mediator of lymphocyte-macrophage, lymphocyte-adipocytes, and lymphocyte-macrophage-adipocytes communication within adipose tissue (Fig. 7).

In conclusion, we describe a complex paracrine network of TNFSF factors in adipose tissue in a high-risk obesity subphenotype characterized by high E2F1 expression. Our model, tying E2F1, TRAIL, and TL1A in a single pathophysiological path, expands understanding of both E2F1 and TRAIL activity, each previously shown to induce adipose tissue dysfunction, and highlights the complexity and importance of TNFSF cytokines in the pathophysiology of adipose tissue. Also, our study proposes that when elevated in a subgroup of patients with obesity, TL1A is a mediator of adipose tissue and whole-body metabolic dysfunction. Our results suggest new therapeutic opportunities in the quest for a more precision/personalized approach to better manage the cardiometabolic complications of molecularly defined obesity subphenotypes.

Funding. This study was supported in part by grants from Deutsche Forschungsgemeinschaft (DFG) (German Research Foundation), 209933838 project number SFB1052: “Obesity mechanisms,” and the Israel Science Foundation (ISF 2176/19).

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

Author Contributions. N.M. conducted experiments and wrote the manuscript. T.P. conducted experiments and reviewed the data. Y.H. developed experimental methods. V.C.-C. conducted bioinformatic analyses. N.G. reviewed the data and manuscript. T.T. conducted experiments. I.F.L. helped establish the human Beer-Sheva cohort. B.K. helped establish the human Beer-Sheva cohort and obtained tissue biopsies. R.G. helped in epidemiological analyses of human data. O.B. and A.M. conducted flow cytometry analyses. N.B. reviewed the data and manuscript. M.B. established and provided the human Leipzig cohort, reviewed the data, and edited the manuscript. A.R. supervised the research and wrote the manuscript. A.R. and N.M. are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.