T‐Cell Signaling Pathways, Including Exhaustion, Predominate in Unhealthy Visceral and Subcutaneous Adipose Tissues

ABSTRACT

Objective

Obesity is an imperfect correlate of metabolic health. Visceral adipose tissue (VAT) characteristics are considered determinants of poor health and subcutaneous adipose tissue (SAT) considered protective. There is a gap in knowledge regarding shared vs. unique SAT and VAT function across the metabolic syndrome (MetS) spectrum.

Methods

We quantified SAT and VAT transcriptomes in a nonhuman primate model of MetS. We calculated quantitative MetS risk scores using composite factors, applied unbiased clustering methods (weighted gene correlation network analysis) to identify transcripts that correlated with MetS risk scores, and performed pathway enrichment analysis.

Results

We found convergence in SAT and VAT on T‐cell signaling genes and pathways, with T‐cell exhaustion signaling prominent. Pathways unique to VAT highlighted interferon signaling and innate/adaptive immune cross talk in response to pathogens. Pathways unique to SAT included innate immune cell signaling, centered on vascular involvement. Although molecular signatures highlight T‐cell signaling, T‐cell abundance in VAT was unrelated to MetS scores.

Conclusions

T‐cell signaling and exhaustion predominate in metabolically unhealthy adaptations of both SAT and VAT. This novel handling of transcript data using an unbiased clustering approach and the creation of continuous MetS scores lead to new insights regarding adipose depot responses and T‐cell biology.

Study Importance

• What is already known?

  • ○
    Adipose inflammation is a driver of metabolic health in individuals who are lean and those who have obesity.
  • ○
    Macrophage activation is generally considered the predominant immune profile in unhealthy adipose tissue.
• What does this study add?

  • ○
    A new analytic approach was developed to relate metabolic health scoring with weighted gene co‐expression network analyses using a translational nonhuman primate model of metabolic syndrome.
  • ○
    In subcutaneous and visceral adipose depots from the same individual, T‐cell pathways were commonly identified as primarily important.
• How might these results change the direction of research?

  • ○
    Understanding the development and subsequent reversal and/or clearance of exhausted T‐cells is a new direction for improving adipose tissue function.
  • ○
    T‐cell biology is a new focus in adipose that is common to both visceral and subcutaneous locations, and alterations in health were observed regardless of obesity status.

1. Introduction

The heterogeneity of health status among individuals who are lean and those who have obesity indicates the importance of understanding adipose tissue (AT) function rather than adipose mass. Unifying theories regarding adipose distributions in ectopic spaces and inflammation have been proposed to connect metabolic health outcomes with AT function, regardless of total body fatness. Supportive evidence includes the lack of associations of mortality with body mass index when inflammation and/or insulin sensitivity indices are included in associative models [1, 2]. AT signatures relating to metabolic dysregulation in people have been defined in both lean people and those with obesity and implicate the stromal vascular cellular fraction [3, 4]. AT contains many different types of immune cells including T‐cells, neutrophils, and macrophages, and these can be influenced by obesity in both quantity and nature [3, 4, 5, 6, 7]. A large body of work, including ours [3, 4, 8, 9], implicates macrophage numbers and polarization/activation states as critical for differentiating healthy and unhealthy AT. However B‐cells and T‐cells are also present in substantial numbers, representing the adaptive immune system, and can accumulate in AT in pathological states [10].

Some dogma cites that visceral adipose tissue (VAT) amount and character are considered determinants of poor metabolic health, while subcutaneous adipose tissue (SAT) depots are protective. There exists a gap in knowledge regarding shared versus unique SAT and VAT function across the metabolic syndrome (MetS) spectrum. This gap has partially arisen from the reliance on rodent models of obesity and metabolic disease, models which have less relevant adipose distributions and AT immune cell populations compared to humans [11]. It is also difficult to source paired VAT and SAT from humans that represent the full range of obesity and health statuses to address this gap.

In this report we hypothesized that unhealthy AT, regardless of adipose mass and location, would have both common and depot‐specific signatures. We exploited an nonhuman primate (NHP) model of spontaneous obesity with a range of metabolic health as a highly translationally relevant model of human MetS that is uncomplicated by confounds of Westernized dietary components. Western diets are known to drive inflammation [12, 13], and so our study is able to identify cellular processes that relate to an individual's intrinsic risk for MetS. To capture the complex relationship of variation in adipose mass and function, we calculated a quantitative MetS risk score for each animal using composite factors [14]. To identify molecular variation in VAT and SAT that co‐correlated with MetS risk score variation, we used the co‐clustering approach of weighted gene correlation network analysis (WGCNA) and annotated gene clusters common and unique to VAT and SAT using pathway enrichment. We used this analytical approach in which we treated both MetS and adipose transcriptomes as continuous traits as a novel way to elucidate molecular signatures correlated with MetS, as opposed to traditional differential gene expression methods with categorization by MetS phenotype(s). Our results clearly revealed T‐cells as the primary immune cell type central to MetS signatures in AT, with functional differences across the MetS spectrum in both SAT and VAT of NHPs.

2. Methods

A 44‐animal, age‐matched NHP cohort was studied ( Chlorocebus aethiops sabaeus) at Wake Forest University School of Medicine (WFUSM). Animals were socially housed and consumed a low‐fat, low‐cholesterol, high‐protein and fiber diet with no added sugar (Lab Diet 5038; LabDiet). As with many NHP studies, selection from a breeding colony led to inclusion of females. All procedures were performed in accordance with the Guide for Care and Use of Laboratory Animals. Protocols for avoidance of pain and discomfort were adhered to and conducted in compliance with the WFUSM Institutional Animal Care and Use Committee. Under a single anesthesia event, excisional tissue biopsies of subcutaneous (SAT) and visceral omental (VAT) adipose at the level of the umbilicus were collected. Adipose was immediately frozen in liquid nitrogen until processing or fixed in 4% paraformaldehyde for 48 h, as previously described [8, 9].

2.1. Comparison of MetS Categories With Quantitative MetS Risk Scores

2.1.1. MetS Categorization

We used a modified version of the National Cholesterol Education Program Adult Treatment Panel III definition [14], as previously described [8, 9]. MetS criteria included the waist measure adjusted for NHPs (waist circumference ≥ 40 cm = obesity), HbA1c > 6, fasting blood glucose (FBG) ≥ 100 mg/dL, systolic blood pressure (SBP) > 135 mmHg, diastolic blood pressure (DBP) > 85 mmHg, triglycerides (TG) ≥ 125 mg/dL, and high‐density lipoprotein cholesterol (HDLC) ≤ 50 mg/dL to identify risk. MetS scores were calculated as the sum of each MetS criterion met. The NHP cohort was selected to ensure equal representation of lean and obese individuals at each level of MetS risk. Obesity was further substantiated by measuring body fatness by computed tomography image analyses, as previously described [8, 9]. NHPs with type 2 diabetes (n = 3) were included and managed with insulin therapy, which was withdrawn 24 h prior to tissue biopsy. These individuals remained hyperglycemic (A1c > 8%) despite insulin and are retained in the analyses to provide representation of the full range of glucoregulatory states to which adipose is exposed [15]. By including a range of body fatness, we ensured representation of metabolically healthy lean (MHL) and obese (MHO), as well as metabolically unhealthy lean (MUL) and obese (MUO) individuals. This purposeful distribution of obesity across health categories aimed to identify factors of adipose that confer healthiness, independently of fat mass or adipocyte size. All MetS data are provided in online Supporting Information and raw data files.

2.1.2. MetS Risk Score Calculation

MetS is a clinically applied categorization that has come under some controversy, but more importantly the categorical summing of binary risks to create an individual MetS score results in loss of statistical power. A quantitative and continuous MetS risk score was calculated by using seven international Diabetes Federation (IDF) risk factors [16] for data analysis. The IDF risk factors included: waist circumference, TG, HDLC, SBP, DBP, FBG, and HbA1c. First, principal component (PC) analysis (varimax rotation) was applied to the risk factors to derive two PCs (eigenvalue > 1.0) to determine which components contribute significantly to the MetS variance. Then MetS risk score was computed by summing both individual PC scores, each weighted for the relative contribution of PC1 and PC2 in the explained variance [16]. The relationships between each continuous variable included in the MetS risk definition were correlated with the MetS score and the calculated continuous MetS risk score using Spearman's and Pearson's correlational analyses respectively (Statistica V13, StatSoft Inc.).

2.2. Transcriptome Data Generation

Qiagen RNeasy Lipid Mini tissue kit (cat. no. 74804) was used according to the manufacturer's protocol. Genomic DNA was removed using the Qiagen RNase‐Free DNase set according to the manufacturer's protocol. Total RNA was used to prepare cDNA libraries using the Illumina TruSeq Stranded mRNA Library Prep and IDT for Illumina‐TruSeq RNA UD Indexes (Illumina Inc.). RIN values for the RNA samples ranged from 7.2 to 9.2. Briefly, 750 ng of total RNA was used (except for one animal; 500 ng was used due to low mass) and polyA selected using oligo‐dT magnetic beads, followed by enzymatic fragmentation, reverse‐transcription, and double‐stranded cDNA purification using AMPure XP magnetic beads. The cDNA was end‐repaired and 3′ adenylated, with Illumina sequencing adaptors ligated onto the fragment ends, and the stranded libraries were preamplified with PCR. The library size distribution was validated and quality inspected using an Agilent 2100 Bioanalyzer. The quantity of each cDNA library was measured using the Qubit 3.0 (Thermo Fisher). The libraries were pooled and sequenced to a target read depth of 25 M reads per library using single‐end NovaSeq6000 SP Reagent Kit (100 cycles) (Illumina) on the NovaSeq6000.

2.3. Transcriptome Quantification

After sequencing, data were transformed to fastq files and two samples were removed because of poor quality of sequencing reads. Data were aligned, quantified, and corrected for batch effects prior to normalization (reads per kilobase per million mapped reads). The adipose depot samples were available for 42 SAT and 41 VAT samples for WGCNA. Transcript counts were transformed into expression using LIMMA software (https://bioconductor.org/packages/release/bioc/html/limma.html) prior to WGCNA. Counts were further filtered to remove transcripts with ≤ 30 counts across all samples for both SAT and VAT separately. A total of 13,914 SAT and 14,142 VAT transcripts that passed quality filters were included in WGCNA.

2.4. Correlation of MetS Risk Scores With Co‐Correlated Gene Clusters—WGCNA

Genes were analyzed by WGCNA [17] to identify modules of genes correlated with MetS and MetS‐related traits using the R software package (https://horvath.genetics.ucla.edu/html/CoexpressionNetwork/Rpackages/WGCNA/). In brief, Pearson correlation was used to generate a correlation matrix for all pairwise genes. The threshold function was used to obtain soft threshold (power‐18) to construct an adjacency matrix in accordance with a scale‐free network [18]. The adjacency matrix was transformed into a topological overlap matrix (TOM) to measure relative gene interconnectedness and proximity. The TOM was used to calculate the corresponding dissimilarity (1–TOM). Average linkage hierarchical clustering coupled with the TOM‐based dissimilarity was used to group correlated gene expression data into modules [18]. More specifically, modules were generated from the Dynamic Tree Cut method for branch cutting [19]. The major parameters were set with a deep‐split value of 2 to branch splitting and a minimum size cutoff of 100 (minimum cluster size = 100) to avoid abnormal modules in the dendrogram; highly similar modules were merged together with a height cutoff of 0.25. Modules were considered significant if the correlation was ≥ 0.30 and p value < 0.05. In the resulting network, as neighbors in a cluster share high topological overlap, the resulting modules likely indicate a common functional class.

2.5. Construction of Module‐Trait Relationships

The gene modules obtained from WGCNA summarize the main pattern of variation. The first PC represents the summary of the module and is referred to as the module eigengene [20]. The relationship between module eigengenes and clinical traits including MetS risk was assessed by Pearson correlation. The modules correlating with MetS and overlapping with MetS‐related traits that showed significant correlations (p < 0.05 and ρ ≥ 0.30) were selected for further investigation. A heat map was used for visualization of the correlations of each module‐trait relationship.

2.6. Pathway Enrichment Analysis

Pathway enrichment analysis was performed using Ingenuity Pathway Analysis (IPA) software. All gene modules were positively correlated with MetS risk scores. Genes from SAT and VAT modules that correlated with MetS were imported separately into IPA, and enriched pathways were identified using core analyses. To visualize directionality of enriched pathways for each module, genes were assigned positive fold changes in IPA to reflect the positive correlations. Canonical pathway p values were calculated in IPA using two‐tailed Fisher's exact test with p ≤ 0.05 considered significant.

2.7. Immunohistochemistry

Histological sections were created from paraffin embedded tissue. All VAT sections were stained for CD3 as a pan‐T‐cell marker (CD3 Rabbit Polyclonal at 1:1000, Agilent, #A045229‐2). A selection of 10 VAT sections were stained for programmed cell death 1 (PD‐1; Mouse Monoclonal at 1:100, Abcam, #NAT105). Stained sections were digitally analyzed for the tissue area, total number of nuclei detected, and the numbers of nuclei with overlapping positive staining for CD3 and/or PD‐1, corresponding to an immunopositive cell (Visiopharm, V2023.09x64) (Table S1; Figure S1). We used a deep learning, artificial intelligence (AI) algorithm to identify regions of interest (ROIs) composed of adipocytes while excluding interlobular connective tissue, lymphoid tissue, and vasculature [21]. We checked accuracy by comparing hand‐drawn ROIs within adipose, with a goal of 10,000 nuclei per image analysis. We found acceptable agreement between the hand‐drawn ROIs and AI‐drawn ROIs (Figure S2); therefore, AI‐drawn ROIs were used in analyses. Immunopositive cells were normalized for the adipose density (all nuclei and tissue area).

3. Results

3.1. Comparison of MetS Categories Versus MetS Quantitative Risk Scores

The majority of prior studies of adipose biology across the MetS spectrum have categorized individuals based on scores that artificially characterize continuous traits. Our strategy was to leverage the variation in our NHP cohort by calculating a quantitative MetS risk score and use analytical methods that allowed us to identify molecular variation that correlates with metabolic variation. The study included 44 NHPs that range from young adult to geriatric adults, with 42 SAT samples and 41 VAT samples, and Table 1 highlights the variation among animals in our cohort. The absolute values for each metabolic measure are human‐relevant and denote we captured the full range of metabolic health with glycemic values spanning the normoglycemic to poorly controlled diabetic range; similarly, lipids and lipoproteins range from healthy to severely dyslipidemic and blood pressure from normo‐ to hypertensive. Body fatness corroborated the weight and waist circumferences as encompassing very lean to extremely obese individuals. Figure 1 shows the superior performance of the continuous quantitative MetS score for the majority of risk criteria. In our NHPs, SBP (Figure 1D) was positively associated with cumulative MetS scores but the relationship was not statistically significant. Adiposity was unrelated to MetS scores (Figure 1E, waist circumference data not shown but results were comparable), as by design we had equal numbers of lean and obese individuals represented at each level of MetS.

TABLE 1.

Demographics of nonhuman primate cohort.

Variables	N	Mean	Standard deviation	Minimum	Maximum
A1C (%)	44	4.61	1.45	4.00	10.10
Age (years)	44	15.19	4.49	6.65	23.50
BW (kg)	44	5.95	1.16	4.04	9.12
DBP (mmHg)	44	70	15	46	110
Body fat (%)	44	27.36	12.97	8.21	52.00
FBG (mg/dL)	44	88	39	49	239
HDLC (mg/dL)	44	95.75	49.99	38.54	205.80
HOMA (AU)	41	6.09	8.11	0.098	39.61
SBP (mmHg)	44	119	21	82	164
TG (mg/dL)	44	70.99	34.44	33.40	181.54
WC (cm)	44	38.98	5.26	28.00	50.50

Abbreviations: A1C, glycosylated hemoglobin; BW, body weight; DBP, diastolic blood pressure; FBG, fasting blood glucose; HDLC, high‐density lipoprotein cholesterol; HOMA, homeostasis model assessment; SBP, systolic blood pressure; TG, triglycerides; WC, waist circumference.

FIGURE 1.

Metabolic syndrome (MetS) scores displayed as categorical and continuous quantitative variables and their correlations with key metabolic syndrome criteria. Plots are shown to be able to visualize the health groupings (metabolically healthy lean and obese [MHL, MHO] and metabolically unhealthy lean and obese [MUL, MUO] nonhuman primates). The calculated continuous quantitative variable for the MetS score was superior in associations with most risk criteria. The lack of relationships with body fatness (panel E) was a deliberate feature of the study, such that the selection of nonhuman primates enabled evaluation of adipose features that conferred lower MetS scores independent of adipose mass.

3.2. Identification of Gene Clusters Correlated With MetS

WGCNA identified 19 modules of co‐correlated genes for SAT and 19 modules of co‐correlated genes for VAT (Figure 2). Module colors are arbitrary in each analysis and do not indicate the same gene composition for the same color in VAT and SAT. In SAT (Figure 2A), two modules were significantly correlated with quantitative MetS risk score (brown, 1749 genes; p = 0.02, correlation = 0.35; and dark turquoise, 109 genes, p = 0.003, correlation = 0.44, Table S4). The brown module was also significantly correlated with body weight, waist circumference, body fat, HDLC, and total cholesterol (TC). The dark turquoise module was also correlated with HDLC and TG. In VAT (Figure 2B), two modules were correlated with MetS risk score (sky blue, 102 genes, p = 0.003, correlation = 0.45; and brown, 731 genes, p = 0.04, correlation = 0.32, Table S5). FBG, HDLC, TC, and TG also correlated with the sky blue module. A positive relationship with MetS‐related modules and HDLC is counterintuitive; however, in the NHP model consuming a cholesterol‐free diet, higher circulatory cholesterol is distributed across both low‐density lipoproteins (LDL) and HDL [22]. In our dataset TC and HDLC were significantly positively correlated (R = 0.40, p = 0.02). Therefore, the high MetS‐associated gene modules do relate to higher LDLC and dyslipidemia, including TG in SAT. Age was included as a covariate in the WGCNA. Age showed significant and positive association with a gene module (white) as well as a significant negative association with gene module (pink) in VAT; however, neither gene module was related to MetS or the risk criteria. Age did not show any significant associations with gene modules in SAT. Additionally, age did not overlap with MetS risk score in both VAT and SAT (Figure 2A,B). The top modules correlated with MetS scores in VAT were surprisingly unrelated to adiposity, which is a central MetS feature, but were more highly related to FBG and cholesterol‐related risk factors. The top modules correlated within SAT similarly correlated with cholesterol and TG and variably with adiposity. Gene clustering further substantiates that fat mass may not be the driving factor behind MetS. Only a few weak negative associations were observed between gene modules and MetS in SAT and VAT. There were more modules correlating negatively with obesity measures, but they represented genes that have been well described in the literature and separated themselves from those relating to MetS [3, 23].

FIGURE 2.

(A) Heat map of gene modules that correlated with metabolic syndrome (MetS) calculated continuous quantitative scores, and its components, in 42 subcutaneous adipose tissues. The two most significant modules (brown and dark turquoise) that correlated with MetS scores were selected for subsequent analyses and are shown boxed. It was noted that HDLC was a component that had uniquely strong relationships with the strongest MetS‐related gene clusters in this depot (boxed examples shown). (B) Heat map of gene modules that correlate with MetS calculated continuous quantitative scores, and its components, in 41 visceral adipose tissues. The two strongest modules (sky blue 3 and brown) that correlated with MetS scores were selected for subsequent analyses and are shown boxed. It was noted that FBG was a component that had a uniquely strong relationship with the top gene cluster in this depot (skyblue3, FBG correlation boxed to highlight), and these genes also were related to adiposity in the opposite direction. (C) Venn diagram demonstrating the number of genes in subcutaneous adipose (SAT) and visceral adipose (VAT) tissues, the selected modules these genes represented from panels A and B, and the number of genes within these that were common to both depots. Genes are listed in Tables S1–S3. BF, body fatness as %; BW, body weight; DBP, diastolic blood pressure; FBG, fasting blood glucose; HDLC, high‐density lipoprotein cholesterol; SBP, systolic blood pressure; TG, triglycerides; WC, waist circumference.

3.3. Pathway Enrichment Analysis of SAT and VAT Common Genes

We used pathway enrichment analysis to annotate genes in modules that were significantly correlated with MetS risk score. For the 1858 SAT module genes, 1604 unique genes were mappable for pathway enrichment analysis in IPA. For the 833 VAT module genes, 721 unique genes were mappable for pathway enrichment analysis in IPA (Figure 2C and Tables S1–S3 for numbers and identities of genes). Comparison of SAT and VAT genes in the modules revealed 399 common genes that were significantly correlated with MetS and highlights the greater number of genes associated with MetS in SAT versus VAT, 1205 and 322, respectively. Pathway enrichment analysis of SAT and VAT genes revealed T‐cell signaling to be of primary importance as 7 of the top 10 pathways in both adipose depots included T‐cell function pathways (Table 2). Figure 3A,B depicts genes mapped to the T‐cell exhaustion pathway in SAT (p = 4.22E⁻⁰⁹) and VAT (p = 1.55E⁻²²) depots. T‐cell exhaustion was a higher ranked and more significant pathway in VAT compared to SAT, which may relate to the higher load of microbial and lipid‐derived antigens processed in VAT [24]. A key T‐cell exhaustion marker, programmed cell death 1 (PD‐1), was highly associated between VAT and SAT (r = 0.472, p = 0.002; Figure S3), supportive of common immunological processes.

TABLE 2.

Top common pathways mapped to genes contained in top two modules correlated with metabolic syndrome scores in both subcutaneous (SAT) and visceral (VAT) adipose tissue.

Common pathways between SAT and VAT	p value for enrichment	Number of genes
Pathogen‐induced cytokine storm signaling pathway	7.65E−27	48
T‐cell exhaustion signaling pathway	2.47E−19	31
T‐cell receptor signaling	2.23E−18	32
Th1 pathway	1.13E−17	25
CD28 signaling in T helper cells	1.94E−17	26
Th1 and Th2 activation pathway	4.24E−17	26
Co‐stimulation by the CD28 family	4.51E−14	18
Th2 pathway	1.36E−13	22
Role of pattern recognition receptors in recognition of bacteria and viruses	1.76E−12	21
Multiple sclerosis signaling pathway	1.97E−12	25
TEC kinase signaling	2.23E−12	25
Phagosome formation	6.35E−12	43

FIGURE 3.

T‐cell exhaustion pathway genes identified from mapping of metabolic syndrome (MetS)‐correlated module genes. Genes identified in subcutaneous adipose tissue (SAT) and in visceral adipose tissue (VAT). Genes with red fill are overlap of genes between VAT and SAT. Genes with green fill are SAT genes and genes with purple fill are VAT. All genes were positively correlated with MetS scores and predicted to be activated. Both the overlap of genes between VAT and SAT and the number of genes within the T‐cell exhaustion pathway are high. Depot‐specific differences in genes within this pathway are noted in Table 5.

3.4. Pathway Enrichment Analysis of SAT and VAT Unique Genes

Enrichment analysis of genes unique to each depot point to subtle tissue‐location differences. Table 3 shows depot‐specific pathways with SAT are enriched for multiple pathways related to neutrophils (and other innate immune cell types) and interactions with the vasculature, while VAT is enriched for more classical inflammatory pathways (interferons) and gene signatures (Table 4). Investigation of genes unique to SAT and VAT that mapped to T‐cell exhaustion signaling show 10 genes in SAT and 12 genes in VAT (Table 5). In SAT, only 30% of the unique genes in the T‐cell exhaustion pathway have previously been shown to play roles in AT function, with the majority nonspecific, such as canonical IL6 production. Interestingly, IL4 and IL13 signaling—the sixth highest SAT pathway—has a role in white AT fibrosis, which correlates with metabolic health [25]. In VAT, 50% of the unique genes in the T‐cell exhaustion pathway are also novel to AT function, comprising a mixture of proinflammation (CD28 and interferons) and inflammation‐resolving (i.e., IL10, TGFβ) effectors (Table 5). Behind the robust T‐cell exhaustion‐associated genes seen in both depots, overall T‐cell signaling, with both Th1 and Th2 activation noted (pathways shown in Figures S4 and S5), was also highly ranked. Based on the molecular findings, we quantitated the total T‐cell population in VAT to explore if these gene signatures were related to overrepresentation (Figure S6), but we did not find overrepresentation of T‐cells in unhealthy AT.

TABLE 3.

Unique pathways in subcutaneous adipose tissue (SAT) identified from mapping of genes contained in top modules correlated to metabolic syndrome scores (1604 genes).

SAT	p value for enrichment	Number of genes
Neutrophil degranulation	3.36E−14	92
Granulocyte adhesion and diapedesis	8.15E−14	47
FAK signaling	1.87E−13	104
IL10 signaling	2.69E−12	22
Molecular mechanisms of cancer	3.98E−12	125
IL4 and IL13 signaling	9.96E−12	35
Cell surface interactions at the vascular wall	1.60E−10	37
Role of NFAT in regulation of the immune response	5.21E−10	49
Agranulocyte adhesion and diapedesis	6.40E−10	43
G‐protein coupled receptor signaling	1.17E−09	99
Atherosclerosis signaling	2.01E−09	31
G alpha (i) signaling events	3.63E−09	45
S100 family signaling pathway	1.14E−08	99
Phospholipase C signaling	2.10E−08	56
Role of chondrocytes in rheumatoid arthritis signaling pathway	4.24E−08	33
Leukocyte extravasation signaling	2.51E−08	41
Class A/1 (rhodopsin‐like receptors)	3.38E−08	50

TABLE 4.

Unique pathways in visceral adipose tissue (VAT) identified from mapping of genes contained in top modules correlated to metabolic syndrome scores (721 genes).

VAT	p value for enrichment	Number of genes
Systemic lupus erythematosus in B‐cell signaling pathway	1.90E−17	46
Interferon gamma signaling	8.61E−16	21
Neuroinflammation signaling pathway	9.82E−16	45
Macrophage classical activation signaling pathway	2.08E−15	33
Communication between innate and adaptive immune cells	9.12E−15	26
Interferon alpha/beta signaling	9.91E−15	21
Role of hypercytokinemia/hyperchemokinemia in the pathogenesis of influenza	5.91E−14	20
Immunoregulatory interactions between a lymphoid and a non‐lymphoid cell	1.68E−13	27
Dendritic cell maturation	1.68E−13	33
Natural killer cell signaling	7.04E−13	33
NOD1/2 signaling pathway	1.99E−12	29
PD‐1, PD‐L1 cancer immunotherapy pathway	2.84E−12	24

TABLE 5.

T‐cell exhaustion pathway genes that were uniquely identified by adipose depot location.

Subcutaneous adipose	Visceral adipose
HAVCR2	CD28
IL12RB2	IL10
IL6	IRF4
JUN	IRF9
MAPK3	PIK3CD
NFATC1	PPM1J
PIK3R5	PPP2CA
PIK3R6	PRKCQ
RALA	RAP2B
RALB	STAT1
	STAT2
	TGFB1

Note: In subcutaneous adipose, overall, these uniquely identified genes were nonspecific to T‐cell functions, whereas in visceral adipose, genes noted as uniquely present were a mix of inflammatory activation and dampening responses.

4. Discussion

AT is host to its own immune system; however, overreliance on rodent models of obesity has resulted in decades of macrophage‐centric research [7, 11]. These differences are stark and well chronicled in single cell mapping between the humans and rodents. In these adipose cell maps, T‐cells make up a moderate fraction (4%–10% total cells and < 40% of immune cells) but are the second most abundant immune cell type after the macrophage, and they have recently been increasingly implicated in AT function [4, 26, 27]. T‐cell functions in adipose include traditional roles relating to cytotoxicity, but also large numbers of regulatory T‐cells persist [26, 28] to maintain an anti‐inflammatory environment. T‐cells have been identified as a driver in metabolic disease development [29], preceding insulin resistance and obesity, and part of the mechanism for macrophage influx that augments inflammatory signaling. T‐cell exhaustion describes a differentiation state occurring in response to sustained T‐cell receptor activation from chronic antigen exposure. Terminally exhausted T‐cells are characterized by loss of effector functions, diminished cytokine production, reduced proliferative capacity, high inhibitory receptor expression (e.g., PD‐1, CTLA‐4, TIM‐3, LAG‐3, TIGIT), and altered cellular metabolism [30]. The exhausted state is thought to be a compensatory mechanism to prevent immunopathology in settings of chronic inflammation. However, exhaustion can become deleterious when it prevents T‐cells from effectively clearing pathogens or malignant cells. Recent observations indicate that obesity in mice and humans with diabetes results in the decreased inflammatory potential of T‐cells [31, 32]. Our findings are corroborated with human clinical data performed by Porsche and colleagues, which similarly identified VAT T‐cell ineffectiveness as a key feature seen in metabolic disease [31]. In that report, small numbers of nondiabetic patients with obesity (MHO) were compared to patients with diabetes and obesity (MUO), which supports the idea that it is dysregulated metabolism, not adipose expansion, that favors T‐cell activation and exhaustion. Additionally, gene signatures were coupled with T‐cell functional assays that confirmed MUO‐sourced T‐cells were deficient in responding to cytokines and antigens. Another study compared VAT and SAT from lean humans and those with obesity and found that adipose expansion is associated with increases in the T‐cell exhaustion marker programmed cell death 1 (PD‐1) [27]. WGCNA approaches have identified in SAT T‐cell activation pathways [33] in a module that was highly positively correlated with insulin resistance, with insulin resistance being a common gene signature across SAT and VAT depots [3]. A WGCNA approach in human SAT [34, 35] identified modules of genes correlated with aging and age‐related immunosenescence including reduced immune function that includes accumulation of exhausted effector T‐cells [36].

The identification of T‐cell activation and exhaustion pathways is of high interest as there is a growing body of work documenting unique microbial species isolated from AT sourced from unhealthy humans and NHPs with diabetes [8, 24]. Originally it had been thought that T‐cell exhaustion arose from self‐antigen responses [7, 37]; however, the new understanding of local AT microbiomes may shift that assumption. Higher exposure to pathogenic microbial species in adipose would result in chronic immune stimulation and the expansion of memory and CD8+ T‐cells. Pathogen‐driven T‐cell exhaustion also better explains the MUL phenotype and the failure of T‐cell restoration of function, after improving metabolism in diet‐induced obese mice through dietary reversal [38]. WGCNA approaches in diabetic AT and peripheral immune cells [33, 39] have found that hub genes in correlated modules include TLR9 and IL2‐STAT5 which similarly suggest microbial signaling and T‐cell activation. In a study of insulin resistant people [33], VAT modules did identify some infection‐related pathways (Salmonella, HTLV‐1) within modules that were highest and positively correlated with LDL cholesterol levels, which is phenotypically related to MetS. Our data corroborate this, as we identified pathogen‐sensing pathways, including the role of pattern recognition receptors in both VAT and SAT and NOD1/2 signaling in VAT, to be highly correlated with MetS. Memory T‐cells manage the balance between antimicrobial responses and lipid metabolism [28], and if these cells move to become exhausted, neither microbial control nor metabolism is effective [40]. This is comprehensively documented in human VAT where T‐cell signatures were related to glucose metabolism, which was in turn related to pathogenic bacteria present in VAT [41].

Limitations of our study include the use of female NHPs, such that sex differences are unable to be commented on. We also are using bulk RNAseq which includes mature adipocytes and the mixed cell populations in the stromal vascular fraction. Even so, in both adipose depots T‐cell signaling pathways were identified indicative of how dominant the activities of this cell type are in metabolic diseases. NHPs are the closest animal model to humans; however, unlike humans, their adipose is not challenged by dietary conditions of high fat, sugar, and cholesterol. These results need to be replicated in human adipose as Westernized dietary factors are generally proinflammatory [31] and drive accumulation of quiescent and potentially exhausted T‐cells [13]; therefore, our animal model‐based outcomes may actually be a conservative estimation of the T‐cell related activation and exhaustion present in the human condition.

The translational relevance of our results centers on potential therapeutic options for restoring adipose health in the context of MetS. Exhausted T‐cells are characterized by the expression of inhibitory cell surface markers, such as PD‐1, CTLA‐4, and LAG‐3. These receptors are able to be blocked pharmacologically, which theoretically could allow T‐cell reactivation but in practice hasn't always been observed [35, 42]. T‐cell exhaustion is thought to be an attempt by the body to curtail overly robust immune responses and is not necessarily a “bad thing.” As such, there are concerns that reactivation approaches to reduce exhausted T‐cell burdens could lead to overstimulatory responses to resident/endemic viruses, and as most viral species have yet to be fully characterized in AT, this allows for an unpredictable immune reaction [43, 44].

T‐cells have been shown in elegant rodent studies to promote obesity and the restoration of obesity through a memory mechanism. These rodents also have documented exhausted T‐cells in expanded VAT which may persist through weight cycling [4, 45]. A notable WGCNA study [46] that diverges from our results included VAT and SAT data from individuals with obesity and with and without diabetes. However, these data were filtered to only include insulin‐responsive genes and thus missed the opportunity to detect wider tissue functions. These analyses therefore unsurprisingly centered on IL6 and IL1β as central inflammatory hubs that were consistently activated [46]. Our results do identify IL6 as a gene included in important gene modules in SAT but not VAT. It is noteworthy that VAT pathway mapping identified the TGFB1 gene within the T‐cell exhaustion pathway, which was not identified in SAT. This master regulator has been directly linked to immune cell types such as alternatively activated macrophages and regulatory T‐cell development [47], with the ultimate effect of resolving inflammation and inducing fibrosis.

In conclusion, we demonstrate a unique approach to investigating adipose cellular contributions to MetS. Our approach incorporates continuous trait modeling of the MetS risk, unbiased clustering approaches to identify pathways common to VAT and SAT depots, and an NHP animal model that allows environments and diet to be controlled and health traits to be observed with and without obesity. These novel approaches to in situ gene expression elucidated T‐cell pathways, with a specific highlighting of T‐cell exhaustion, as a key phenomenon increasing with worsening MetS.

Funding

This work was supported by NIH R01HL142930 and the Cancer Genomics Shared Resource is supported by the Wake Forest Baptist Comprehensive Cancer Center's NCI Cancer Center Support Grant P30CA012197.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Data S1: oby70133‐sup‐0001‐supinfo_1.xlsx.

Table S1: Summary data of histological features measured from visceral adipose tissue sections.

Figure S1: Demonstration of the automatic immunohistology detection of positively (red) and negatively stained nuclei (blue) for cell markers.

Figure S2:. Comparison of artificial intelligence (AI) automated image regions of interest (ROIs) versus manually drawn ROIs for adipose tissue cellular density and CD3+ cell detections in randomly selected adipose tissue immunohistologically stained sections (n = 15). The high agreement between methods provided confidence in the ability of AI to create adipose‐only ROIs, eliminating large vascular and immunological structures that may have been captured in the biopsy section. All ROIs were visually inspected prior to application of image analyses tools to mark nuclei as stained negative or positive (Figure S1).

Figure S3: T‐cell exhaustion marker programmed cell death 1 (PD‐1) has gene expressions that are positively associated in subcutaneous and visceral adipose tissues (SAT, VAT; Panel A) as expected with the convergence of T‐cell exhaustion pathways being commonly identified as one associated with worsened metabolic health. VAT PD‐1 gene expression was verified by protein detection (Panel B) in 10 samples that represented the range of gene expression values. The abundance of cells expressing PD‐1 was unrelated to the T‐cell density of VAT (Panel C), which confers some specificity to the T‐cell phenotypes in unhealthy adipose tissue.

Figure S4: Th1 activation pathway in subcutaneous (SAT) and visceral (VAT) adipose tissues. Genes with red fill are overlap of genes between VAT and SAT; Genes with green fill are SAT‐genes and genes with purple fill are VAT‐genes. All genes were positively correlated with MetS scores and predicted to be activated. There is significant overlap in the correlated genes with MetS scores between depots.

Figure S5: Th2 activation pathway in subcutaneous (SAT) and visceral (VAT) adipose tissues. Genes with red fill are overlap of genes between VAT and SAT; Genes with green fill are SAT‐genes and genes with purple fill are VAT‐genes. All genes were positively correlated with MetS scores and predicted to be activated. There is significant overlap in the correlated genes with MetS scores between depots.

Figure S6: CD3+ T‐cell abundance in visceral adipose is unrelated to metabolic health scores. (A) There was no relationship between the fraction of cells marked as T‐cells and the categorical sum of individual risk factors indicative of MetS, and (B) There was similarly no relationship between T‐cells and the calculated continuous metabolic syndrome score. Together we conclude that unhealthy adipose tissue does not have more T‐cells but have altered T‐cell functions.

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.