Multi-omics of human obesity and related multi-system diseases

Abstract

Context

Efforts to characterize the shared molecular risk factors that contribute to obesity and the downstream disease sequelae it triggers have been limited.

Objective

We aimed to identify functional genes with evidence for both causal and consequential effects on obesity related traits and their downstream sequalae using integrated genomic and proteomic data.

Methods:

We investigated the association of obesity related traits with 2,912 plasma proteins in 259 individuals from the Cameron County Hispanic Cohort (CCHC) with validation of results in ~45,000 participants from UK Biobank (UKBB). Through colocalization and Mendelian Randomization, we assessed evidence for the shared underpinning and the causal direction of significant proteins with respect to obesity and obesity-associated illnesses. We used gene ontology and cell- and tissue-specific protein and transcriptional activity patterns of the genes encoding target proteins to illuminate the functional relevance of implicated pathways. We additionally investigated the suitability of target proteins as potential therapeutic targets for drug development.

Results

Of the 122 significantly associated with obesity metrics at a false discovery adjusted level (FDR<0.05), 121 replicated in UKBB. Most function in adipogenesis, inflammation, glucose metabolism, and neural and appetite regulation. Eighty of 121 replicated proteins showed evidence of statistical causality for obesity or obesity-associated illnesses. Causally linked showed elevated transcript abundance in adipose and brain tissues and adipocytes. The promising weight reduction potential of several target proteins highlights their suitability for future pharmaceutical repurposing.

Conclusion

Our analyses revealed key regulatory mechanisms influenced by and influencing obesity, offering valuable targets for biomarkers and clinical interventions.

INTRODUCTION

Integrated studies of genomics and protein dysregulation in obesity can identify and statistically discern molecular features that are the cause or consequence of obesity and potent risk factors for its sequelae. Obesity exacts disproportionate metabolic consequences and impacts risk of a multitude of downstream diseases(1). Indeed, surgical(2) and pharmacologic treatments for obesity(3) can reverse clinical and biochemical effects of obesity across multiple downstream sequalae(4): including for renal, cardiovascular, metabolic, and neurologic diseases. Despite the recent availability of treatments that are effective across diverse backgrounds(5), obesity duration, and metabolic dysfunction(6), the prevalence of obesity and its sequelae are increasing globally, highlighting the need for identification of causal molecular targets and mechanisms that impact downstream disease pathogenesis.

While genomic studies implicate a growing array of targets in obesity(7), limited diversity among study participants(8), small per-locus effect size, and poor understanding of the impact of identified variants on downstream disease challenge genetic studies to identify broadly relevant mechanisms of obesity-related diseases. Recent applications of omic measures (e.g., proteome, metabolome) have enabled further characterization of genetic drivers(9,10). Innovative approaches for integration of multi-omics across multiple large data resources hold promise for illuminating targets involved in obesity-related disease. We aimed to identify genes with evidence for both causal and consequential effects on obesity related traits and their downstream sequalae. To accomplish this goal we integrated results of proteomic and genomic analyses to form a “proteogenomic” study of thirteen obesity related traits (Supplementary Table 1, Methods)(11) with discovery in 259 extensively phenotyped Hispanic/Latino study participants in Cameron County Hispanic Cohort (CCHC) who are at high metabolic risk(12). We replicated our findings and assessed the clinical significance of findings in ≈45,000 UK Biobank participants. Despite differences in genetic ancestry and self-identified race/ethnicity, sample size, and obesity distribution in our samples, our proteogenomic approach revealed shared causal mechanisms involved in obesity and its clinical consequences. Many of our identified biomarkers are transcribed and translated broadly across obesity relevant tissues and cell types and show empiric or mechanistic evidence for druggability, illustrating how integration of measures across the stages of protein synthesis can inform our understanding of obesity pathogenesis, treatment, and prevention.

MATERIALS AND METHODS

Study cohorts

Discovery cohort:

The Cameron County Hispanic Cohort (CCHC) is a community-based study that initiated in 2004 in Brownsville, Texas. Participants were sampled from a low income, minoritized Mexican American population (predominantly self-identified as first or second-generation) with a high burden of cardiometabolic risk(12). Households were randomly ascertained and recruited for CCHC via census sampling methods (N≈5,159), with continuous recruitment and surveillance. We measured demographic, lifestyle, and clinical traits in adults at an initial examination and subsequently at 5-year increments. All participants provided written informed consent, and the study procedures were approved by the Committee for the Protection of Human Subjects (CPHS) of the University of Texas Health Science Center-Houston.

Replication cohort (UK Biobank):

The study design of the UK Biobank has been previously described(13). Briefly, the UK Biobank is a rich study of ≈500,000 individuals across the United Kingdom with genetic characterization and multi-dimensional phenotyping, including lifestyle, anthropometry, and health and disease conditions. In this study, we accessed ≈45,000 individuals with circulating protein quantification(14) via the Olink Explore 3072 (ThermoFisher; Waltham, MA) method.

Anthropometric and body composition traits

Here, we studied the association of protein abundance with thirteen measures related to obesity: 1) BMI; waist circumference (WC) adjusted for BMI in 2) men, 3) women, and 4) together; waist-to-hip ratio (WHR) adjusted for BMI in 5) men, 6) women, and 7) together; dual-energy x-ray absorptiometry (DXA)-based measures of body composition including 8) total trunk fat (TFAT), 9) total body fat percentage (BFP), 10) total body fat mass (BFM), 11) total visceral fat (VFAT), 12) visceral to subcutaneous fat ratio; 13) class III obesity (“severe” obesity) defined as a BMI ≥ 40 kg/m2 as compared to 18.5≤BMI<25 group (i.e. normal weight). Our analytic measures of obesity were informed by previous studies linking select measures to obesity-related outcomes(15). Details on the measurement of anthropometric and DXA are provided in Supplementary A(11).

Quantification of the circulating proteome

For the CCHC, we used the Olink Explore 3072 platform with frozen serum (maintained at −80°C until time of proteomics). Protein concentrations were expressed in log₂ units (normalized protein expression, NPX). We used 2,921 proteins including proteins below the limit of detection (LOD) after quality control per Olink recommendation(14).

Statistical methods

Association between obesity phenotype and the human proteome:

Our first step was to identify proteomic associations with obesity traits in the CCHC to prioritize downstream genomic and tissue studies. We used proteomics quantified at a single CCHC visit for each participant visit alongside concurrent obesity phenotypes. We fit a mixed model ${\mathit{P}\mathit{r}\mathit{o}\mathit{t}\mathit{e}\mathit{i}\mathit{n}}_{\mathit{i}\mathit{j}}~{\mathit{T}\mathit{r}\mathit{a}\mathit{i}\mathit{t}}_{\mathit{i}\mathit{j}}+{\mathit{C}\mathit{o}\mathit{v}\mathit{a}\mathit{r}\mathit{i}\mathit{a}\mathit{t}\mathit{e}\mathit{s}}_{\mathit{i}\mathit{j}}+\left(1\right|{\mathit{F}\mathit{a}\mathit{m}}_{\mathit{j}})$, with protein as the dependent variable and each obesity trait as an independent variable in each individual i, with family identifier j as a random effect (to adjust for genetic relation between individuals) and covariate adjustments as fixed effects. Details on adjusted covariates, association models, multiple testing correction, and replication in UK Biobank are provided in Supplementary A(11). In summary, covariates included age, sex, probabilistic estimations of expression residual (PEER) factors, genetic ancestry principal components, and age-sex interaction terms. Associations were estimated using multiple models with variable covariate sets to assess their robustness.

Genetic studies of proteins associated with obesity

A major limitation in many human molecular studies of obesity is the use of cross-sectional molecular markers and phenotypes(16), which limits the ability to decipher temporality (causation). We used both colocalization and Mendelian randomization (MR) approaches to address this concern. Olink proteomics-based protein trait quantitative trait loci (pQTL) were extracted from the UK Biobank Pharma Proteomics Project (UKB-PPP) online repository (synapse.org). Study details and summary statistics have been published previously(14).

Obesity traits:

We obtained GWAS summary statistics on obesity traits from the GWAS Catalogue (ebi.ac.uk/gwas) and OpenGWAS (gwas.mrcieu.ac.uk). For each obesity trait, we selected GWAS studies with either the largest sample sizes or those reporting a higher number of independent GWAS-significant associations when sample sizes were comparable (Table A2 in Supplementary A)(11). We found no sufficiently powered sex-combined GWAS for WC or visceral-to-subcutaneous fat ratio (VSR), so these traits were excluded from further downstream genetic analyses.

Obesity-related disorders:

For obesity-related health conditions, we obtained GWAS summary statistics from the GWAS Catalogue (Table A3 in Supplementary A)(11). We prioritized disorders with broad epidemiologic support for obesity as a driver of the condition: (1) cardiovascular disease (coronary artery disease [CAD], heart failure [HF], atrial fibrillation [AF]); (2) dysglycemia (type 2 diabetes [T2D]); (3) renal disease (chronic kidney disease [CKD]); (4) malignancy (breast cancer [BrCan], colorectal cancer [ColrCan]); (5) hypertension (systolic blood pressure [SBP]); (6) neurocognitive disease (Alzheimer’s disease or history of [AD]); (7) dyslipidemia (modeled using LDL GWAS). We chose GWASs with either the largest sample sizes or those reporting a higher number of independent GWAS-significant associations when sample sizes were comparable.

Colocalization:

As a first step to implicate obesity-associated proteins causally in its pathogenesis, we performed colocalization analyses, which identifies joint causal variant effects influencing both obesity (and related traits) and protein abundance [coloc in R(17)]. For each protein associated with an obesity-related trait (FDR <0.05), we performed pQTL-GWAS colocalization with all obesity traits and obesity-associated disorders. Details on colocalization steps are provided in Supplementary A(11).

Mendelian randomization:

We performed two-sample Mendelian randomization (2SMR) in independent GWAS samples to investigate potential causal associations and directionality between traits or outcomes with the underlying obesity-associated proteome. Of note, a partial 2SMR/1SMR was used when sample overlap existed between pQTL and GWAS, respectively. We performed MR analyses in both “forward” (protein affects obesity trait) and “reverse” (obesity trait affects protein) directions to ascertain directionality of effect in the absence of longitudinal assessment. Details on QC and instrumental variables (IV) selection steps are provided in Supplementary A(11).

To minimize bias in causal inferences when using MR framework, and consistent with STROBE guidelines for observational studies(18), forward MR was performed under multiple approaches including the inverse variance weighted (IVW) method, as well as IVW robust penalized (IVW-RP), median penalized, and Egger robust penalized methods. For IVW method, we used the “default effect” model when there was no heterogeneity between IVs (P>0.05), or the “random effect” model when heterogeneity was present. We primarily used IVW-RP effect estimates for concordance analyses with phenotypic and PheWAS effect sizes.

If the number of IVs for a protein was greater than 2, we considered the protein potentially causal for obesity and/or obesity-associated conditions if it was at least nominally significant under all MR methods and there was no significant evidence for pleiotropy (via MR-Egger intercept P>0.05), and robustly causal if additionally passing FDR <0.05 under IVW-RP. If a protein had fewer than 3 IVs, we relied on results from IVW and IVW-RP for potential causal association inference.

Reverse MR was conducted using identical methods as forward MR. We considered BMI as a significant causal factor when the association was nominally significant across all MR methods and there was no evidence for pleiotropy (via MR Egger intercept). Finally, we considered a protein to be both a cause and consequence of obesity if both forward and reverse MR associations were concurrently nominally significant. MR analyses were performed using the MendelianRandomization in R.

Functional context of the obesity proteome

Given the system-wide impact of obesity, we sought to map identified proteins to pathways and into tissues with organ and cellular resolution. First, we utilized the Protein Analysis Through Evolutionary Relationships (PANTHER) database (pantherdb.org) gene ontology tools to identify relevant molecular functions, cellular components, and biological processes associated with significantly associated proteins. Given that curated pathway annotations may not completely reflect updated experimental and human studies, we supplemented in-silico pathway identification with an extensive literature review to explore potential functional annotations.

Expression of the obesity proteome at the protein and transcriptional level at organ and cellular resolution

To understand the tissue source of targets identified through proteogenomic studies, we next analyzed tissue and cell-specific expression patterns of genes encoding proteins that demonstrated (1) evidence of association in proteomic studies; and (2) evidence by colocalization or MR studies of an effect on obesity (either forward or reverse MR, at nominal significance). We used GTEx (gtexportal.org)(19) and cell-level gene expression data (proteinatlas.org) (20) for this analysis. We restricted analyses to tissues and cells with non-zero expression levels for target genes. Following seminal work annotating “tissue specificity” of a proteome via RNA expression(21), we conservatively considered a protein tissue- or cell-specific, respectively, if scaled tissue or cell-level expression exceeded 10 across tested tissues, or 4 across all tested single-cells.

As another approach to characterize tissue- and cell-specific gene expression, we used the Tabula sapiens atlas (cellxgene.cziscience.com), a molecular reference atlas of more than 400 cell types of the human body, excluding brain tissue. The cell types with fewer than 500 cells and the genes with zero counts in all cells were removed. The list of genes with potential causal association with obesity that were present in the Tabula atlas were used to calculate a gene activity score (via AddModuleScore in Seurat). Given the focus of most obesity related discovery on fat tissue, we finally explored RNA expression at single cell resolution in adipose tissue in a published reference (SCP1376)(22). We calculated gene activity score similarly as above and visualized genes across cell types.

Druggability of prioritized protein targets

Human genetic evidence for drug targets has been shown to substantially improve the success rate from clinical development to drug approval(23). To determine how identified proteins may be targeted therapeutically, we utilized DGIdb (version 5.0), a large-scale “druggable genome” resource consisting of known and predicted interactions of drugs with genes or gene products. We supplemented identified drug-gene interactions for the identified proteogenomic targets with extensive literature review to investigate potential mechanistic evidence.

Ethics approval and consent to participate

The CCHC portion of this study was approved by the Committee for the Protection of Human Subjects of the University of Texas Health Science Center, Houston.

RESULTS

Figure 1A shows our study design. The descriptive statistics of obesity traits are listed in Methods and shown in Supplementary Table 1(11). In CCHC, the mean age was 51±14.5 years (66% women), mean BMI was 32.5±7.1 kg/m² and we observed strong, expected correlations across regional obesity phenotypes (Supplementary Figure 1A)(11). BMI was modestly correlated with Dual Xray Absorptiometry (DXA) derived visceral fat or subcutaneous fat (r≈0.7), consistent with prior reports from large cohorts across a wider range of BMI and populations(24). The CCHC participants demonstrated high prevalence of cardiometabolic diseases with significantly elevated rates of dyslipidemia, and dysglycemia compared to national averages (Supplementary Table 1)(11). In contrast to CCHC, UK Biobank participants had lower BMI and had lower cardiometabolic disease burden (mean BMI 27.5±4.8 kg/m², Supplementary Figure 1B)(11).

Figure 1.

A: graphical summary of the study design. B: Upshot plot of proteins associated with obesity traits in CCHC at multiple testing corrected significance level. C: functional characteristics of 122 proteins associated with obesity in CCHC (MF: molecular function, BP: biological process, CC: cellular component). D: Graphical illustration of comparative effect sizes and directions of 122 proteins’ associations with BMI, and their pathophysiological roles in obesity. Inner circular track demonstrates scaled phenotypic effect size (scaled ES) in CCHC (Beta.CCHC), phenotypic scaled ES size in UKBB (Beta. UKBB), scaled EF in forward Mendelian Randomization (FMR), and reverse MR (RM). Out circular track represents pathophysiologic roles attributed to each protein. Correlation scatterplot illustrates Pearson correlation between protein-BMI effect size in CCHC vs protein-BMI effect size in UKBB. White color in circular plot denotes non-significant effects. Tracks are arranged by pathophysiologic grouping. CCHC: Cameron County Hispanic Cohort, UKBB: UK Biobank.

Across 2,921 proteins, we identified 122 unique proteins in CCHC associated with one or more obesity metrics at false discovery adjusted level (FDR<0.05, Supplementary Table 2; Supplementary Figure 2)(11). The largest proportion of proteins were identified in the analysis of BMI (103/122; among which 80 were exclusively identified in the BMI analysis; Figure 1B). Significant proteins have been implicated in a variety of biological processes in obesity, including cell development and apoptosis, immunity, lipid and glucose metabolism, and cellular responses (Figure 1C, Table 1). Moreover, identified proteins include both canonical and emerging therapeutic targets in human obesity (Table 1). A large fraction of these targets spanned functional pathways of adipose tissue metabolism, inflammation, and fibrosis (LEP, FABP4, IGFBP1/2, IL1RN, PLIN1, CTGF/CCN2; see Table 1 for references), fat metabolism (LPL), body weight regulation (GFRAL), appetite (GHRL), and adipogenesis and browning (AHNAK, ADAMTS15; Figure 1D). Additional pathways implicated by identified proteins were biologically plausible yet not widely reported in obesity in humans (Table 1), including energy expenditure and adipocyte biology (CTHRC1), oxidative stress (PRDX5), and a miscellaneous group (e.g., ADAMTSL5, BLVRB, C9orf40, CDHR1, etc.; Table 1). Importantly, we observed broad replication of significant proteins (except for SLK) and largely consistent direction of effect in UK Biobank participants (Pearson R²~0.81; Figure 1D, Supplementary Table 3)(11), supporting a similarity in the plasma obesity proteome across diverse populations.

Table 1.

Role of target proteins in obesity mechanism as reported in the literature, and corroborating evidence in replication, and causal inferencing analyses. At least 20 proteins were previously reported for obesity. Symbol: ↑ represents positive effect direction while ↓ is negative effect direction, and ✓ demonstrates colocalization between pQTL and obesity traits. FMR: Forward Mendelian Randomization, RMR: Reverse Mendelian Randomization. pQTL: protein Quantitative Trait Loci.

Encoding Gene	Role in Obesity Mechanism	PubMed	Reported For Obesity	Direction in CCHC	Direction in UKBB	FMR direction	RMR direction	Colocalization evidence
ADAMTS15	Promote thermogenesis and adipose browning via β3AR-PKA-CREB axis	28702327	Yes	↑	↑	↑	↑
ADAMTSL5	Upstream to receptor tyrosine kinase (RTK) pathway which is an energy homeostasis regulator. Strong linked to cancer. Also reported to be playing regulatory role in extra cellular matrix modeling (promoting fibrosis) via TGF-Beta pathway.	33197513	No	↓	↓
AGER	mediates the activation of nuclear factor κB (NF-κB) and other pro-inflammatory pathways leading to obesity-induced inflammation	38139276	Yes	↓	↓
AGR3	promote insulin regulation and lipid uptake via playing role in insulin signaling pathway	22969776	Yes	↑	↑	↑
AHNAK	plays role in thermogenesis and adipose tissue browning via β-adrenergic signaling pathway. Also, it is suggested to promote adipocyte differentiation via Bmp4/Smad1 signaling pathway.	30154465; 26987950	Yes	↑	↑		↑
AMPD3	Plays role in energy balance via activating PGC-1α signaling pathway (particularly in brown adipose tissue).	24066180	Yes	↑	↑
ARG1	Compete with endothelial nitric oxide synthase (eNOS) which results in vascular remodeling (endothelial dysfunction).	33197620	Yes	↑	↓	↓		✓
BAG3	regulate glucogenesis and fatty acid oxidation via multiple pathways	37308077	Yes	↑	↑
BLVRB	Plays a role in insulin regulation via insulin signaling pathway.	38056462	No	↑	↑
C9orf40	largely unexplored for obesity, but connected to malignancy via multiple signaling pathways	37373333	No	↑	↑
CA5A	Plays a role in glucogenesis, but precise mechanism is not yet explored.	23589845	Yes	↓	↑		↑
CCL7	Promote adipose tissue macrophage M1 via CCR1/JAK2/STAT1 pathway, which increase inflammation in adipose tissue.	36109744	Yes	↑	↑		↑
CCN2	inhibit adipocyte maturation via Wnt signaling dependent; TGF-beta pathway and CCEBP-beta and-delta dependent mechanisms	25354561	Yes	↑	↑
CCN3	negative regulation of glucose oxidation; also impairs β-cell proliferation concomitantly with a reduction in cAMP levels	29411334	Yes	↑	↑
CD248	Plays a role in ECM remodeling via ERK1–2/Src/PI3K/c-fos pathway. ECM remodeling promotes either inflammation or adipocyte enlargement or both.	31221584	Yes	↑	↑	↑
CD300LG	Mechanism unknown but maybe associated with lipid metabolism and impaired glucose metabolism. Also linked to angiogenesis but mechanism not fully explored.	26336608; doi.org/10.7554/eLife.96535.1	Yes	↑	↑
CD55	Enriched in response to TGF-beta signaling and regulate glucose metabolism	38844451	Yes	↑	↑
CD5L	Plays a role in lipid metabolism in Th17 immunity cells via RORγ lipid pathway. This pathway is a master lipid regulator.	26607794	Yes	↓	↑
CDHR1	A photoreceptor protein and significant regulator of visual sensor in retina. Exact association with obesity is not known but the protein is suggested to be associated with lipid metabolism via AMPK/Wnt-β catenin regulation (downregulation of AMPK and upregulation of the Wnt-β catenin). Plays a role in epithelial-mesenchymal transition and progression toward fibrosis, cell migration. Also suggested to play a role in insuling signaling pathway and insulin regulation.	31547193; 26986842	No	↑	↑
CDON	Constricting adipocyte growth by inhibiting adipogenesis via hedgehog signaling pathway.	25576054	Yes	↑	↑		↑
CELSR2	Affect lipid metabolism via ROS signaling, and glucose metabolism via insulin signaling pathway. Supress lipid accumulation	34478580; 26986842	Yes	↓	↑
CFB	Promoter of PPARγ which in turn play a regulatory rule in mesenchymal differentiation to adipocyte and adipocyte maturation (adipogenesis) via AMPK/ERK signaling (probably).	29137982	Yes	↑	↑		↑
CFD	Promoter of PPARγ which in turn play a regulatory rule in mesenchymal differentiation to adipocyte and adipocyte maturation (adipogenesis) via AMPK/ERK signaling (probably).	27611793	Yes	↑	↑		↑
CFH	Plays a role in lipid metabolism and adipocyte inflammation via alternative complement pathway.	19833879	Yes	↑	↑		↑
CHGB	Involved in biogenesis, regulation and function of pancreatic β-cell insulin granules	33711921	Yes	↓	↓
CKB	Promote thermogenesis and adipose tissue browning via β3-adrenergic receptor pathway.	33597756	Yes	↓	↓		↓
CLEC3B	Promote angiogenesis in extracellular matrix and fatty acid oxidation in adipocyte via activating AMPK signaling pathway. Lipid metabolism regulator.	28546359; 32265595	Yes	↑	↑
CLMP	regulate adipocyte maturation and lipid accumulation via multiple pathways	15563274	Yes	↑	↑
CNTN3	Promote myocyte and neural development via Wnt/β-catenin pathway, therefore with probable with a rule in lipid metabolism.	30919020; 25300137	No	↑	↑		↑
COL15A1	adipocyte maturation (adipogenesis, probable) and extracellular matrix (ECM) remodeling (promotion of collagenesis) and promotion of smooth muscle cell migration via AMPK signaling.	37239083	Yes	↑	↑	↓		✓
COL4A1	Vascular remodeling (inflammation promotion) via TGF-β signaling pathway.	27049236	Yes	↓	↓
COL6A3	Two mechanisms suggested: Vascular constriction via TGF-β signaling pathway, and appetite regulation via leptin signaling pathway (and fibrosis promotion).	37884762; 37884762	Yes	↑	↑		↑
COLEC12	Play a role in lipid metabolism via scavenging oxidized-LDL.	34252459	Yes	↑	↑		↑
COMP	Promote adipocyte maturation (adipogenesis) via remodeling (collagen deposition) of extracellular matrix (ECM) via AMPK signaling. But at a high level it can negative affect adipogenesis.	30100245	Yes	↑	↑	↓	↑
CPM	Promoter of PPARγ which in turn play a regulatory rule in mesenchymal differentiation to adipocyte and adipocyte maturation (adipogenesis) via AMPK/ERK signaling.	23294303	Yes	↑	↑		↑
CST6	Lipid metabolism via AMPK signaling pathway (probably adaptogenic).	28234994	No	↓	↓
CTHRC1	Plays a role in thermogenesis and adipose tissue browning via TGF-β/CTHRC1/GPR180 signaling.	34880217	Yes	↑	↑	↑	↑
CYTL1	Precise mechanism is not known. Associated with adipocyte inflammation.	30342560	Yes	↓	↓	↓		✓
ECI2	Play a role in lipolysis, lipid metabolism, and energy homeostasis via fatty acid oxidation process.	25972572	Yes	↓	↑	↓
EDIL3	regulation of angiogenesis and vascular remodeling via mTOR signaling pathway	34538531	Yes	↓	↓
ENO3	Plays a role in glucose metabolism via miR-34a/ENO3 pathway, and glycolytic pathway.	37960269	Yes	↑	↑	↑	↑
F11	Precise mechanism not yet precisely characterized, but encoding gene is in close proximity of USF1 gene which is involved in lipid transport into cellular environment and intracellular cholesterol metabolism (lipid metabolism).	18067551	Yes	↑	↑		↑
FABP4	Play major roles in metabolism. Promoter of PPARγ which in turn play a regulatory rule in mesenchymal differentiation to adipocyte and adipocyte maturation (adipogenesis) via AMPK/ERK signaling. Also suggested to be involved in glucose metabolism and lipid transportation via insulin signaling pathway and SCD-1 pathway.	16054052; 24319114	Yes	↑	↑		↑
FGF5	member of fibroblast growth factor family, and regulate angiogenesis, probably via PPARγ pathway	15831820	Yes	↓	↓
FGL1	regulate adipogenesis via ERK1/2-C/EBPβ-dependent pathway in adipocytes.	31908014	Yes	↓	↓
FSTL3	Negatively regulate lipid accumulation via NF-kB, TLR signaling/ TNF signaling pathway. Negative regulator of lipid metabolism. Suppress inflammation.	36502285	Yes	↑	↑		↑	✓
FURIN	Promoter of lipolysis, thermogenesis and adipose tissue browning via PKG and P38 MAPK pathways. Also interact with B-natriuretic peptide, a regulator of ghrelin with implication for appetite regulation.	32114491; 22698919	Yes	↑	↑	↑
GAL	Affect adipogenesis and inflammation via regulation PPARγ agonists	23401338	Yes	↓	↓
GFRAL	Mechanism is not yet characterized but interact with TGF-β, ERK/PI3K pathways that may indicate a role in inflammation. Also interact with GDNF signaling pathway which is a regulator of appetite.	33172749; 28846099	Yes	↑	↑	↑
GH1	Product of growth hormone (GH) and regulator of energy homeostasis	25295535	Yes	↓	↓
GHRL	ghrelin is an appetite regulator and energy homeostasis via multiple pathways	25263830	Yes	↓	↓
GPD1	Lipid metabolism and lipid biosynthesis via GPD-1 signaling and insulin signaling pathways.	16849634	Yes	↑	↑	↑	↑
GSN	promote lipogenesis via multiple pathways	22445754	Yes	↓	↓
HAGH	Play a role in glucose metabolism via glyoxalase pathway.	36964401	Yes	↑	↑	↑
HBQ1	Promote oxidative stress, apoptosis, and inflammation via ROS signaling	38067210	No	↑	↓
HS6ST2	energy metabolism regulator in brown adipose tissue via Wnt and FGF signaling pathways	23690091	Yes	↓	↓
HSPB1	regulator of ER stress via multiple pathways	34051341	Yes	↑	↑
HSPB6	Promote oxidative stress, apoptosis, and inflammation via ROS signaling	19248813	Yes	↑	↑	↑	↑
HSPG2	Vascular remodeling (inflammation promotion) via TGF-β signaling pathway.	27049236	Yes	↑	↑	↓	↑
IGFBP1	Promoter of glucose metabolism via INSR/MAPK signaling and INSR/Akt/PI3K signaling. Also promoting apoptosis via p53/BAK-dependent pathway.	18056423; 27049236	Yes	↓	↓		↓
IGFBP2	Promoting glucose metabolism and apoptosis via PI3K/mTOR pathway. And cell proliferation via AMPK signaling. Also, leptin signaling mediator with a rule in appetite regulation.	22410287	Yes	↓	↓	↑	↓
IGSF3	An immunoglobulin. Mechanism not yet known but an up-regulator of TGF-β with a role in inflammation (speculative).	32355822	No	↑	↑	↑	↑
IGSF8	participate in regulation of hypothalamic melanocortin pathway and thus an energy regulator	36175420	Yes	↓	↑
IL1R1	Promoter of mesenchymal differentiation to adipocyte and adipocyte maturation (adipogenesis) via AMPK/PI3K/Akt/TGF-β signaling. Reduce inflammation.	34834553	Yes	↓	↓	↑
IL1RN	Promoter of mesenchymal differentiation to adipocyte and adipocyte maturation (adipogenesis) via AMPK/PI3K/Akt/TGF-β signaling. Reduce inflammation. Also, appetite regulation via leptin signaling.	34834553	Yes	↑	↑		↑
INSR	Participate in glucose metabolism via PI3K-AKT/PKB and Ras-MAPK	12138094	Yes	↓	↑	↑		✓
ITGAV	Promote adipogenesis from adipocyte stem cell via Hippo and PI3K-AKT/PKB pathway.	35106921	Yes	↑	↓	↑↓	↓
LEP	Via leptin signaling which regulates appetite and energy expenditure.	11466583	Yes	↑	↑		↑
LGALS4	Binds to CD14 and induce macrophage differentiation via MAPK signaling, promoting inflammation.	27017379	Yes	↓	↑		↑
LGALS9	Binds to PRDX2 antioxidant and attenuate oxidative stress. Attenuate inflammation.	33727589	Yes	↑	↑		↑
LPL	Play a role in lipolysis and lipid metabolism via insuling signaling pathway.	21966368	Yes	↑	↓		↓
LSP1	plays a role in adipocyte inflammation. Mechanisms are largely unexplored in the context of obesity.	16424110	Yes	↑	↑
MEP1B	Promote adipogenesis via binding to mineralocorticoids. Involved pathways include RAAS signaling, mineralocorticoid receptor signaling, mTOR/S6K1,PPARγ,C/EBPγ.	26674805	Yes	↑	↑		↑
MIA	melanoma growth regulator and regulator of adipocyte possibly via adipokines possibly multiple pathways	34068679	No	↓	↓
MMP12	Mechanism yet unknown but is reported to restrict adipose tissue expansion, promote insulin resistance, nitric oxide induced inflammation	24914938	Yes	↓	↑		↑
NCAN	Precise mechanism is yet unknown but promote triglyceride accumulation and lipid metabolism. Mostly reported for non-alcoholic fatty liver disease (NAFLD).	23594525	Yes	↓	↓	↑	↓
NECTIN2	Promote angiogenesis and cholesterol metabolism but mechanism unknown. Also associated with grey matter volume of intracalcarine cortex in brain via 5-HT4	35136728; 34872017	Yes	↓	↑
NPTXR	Mechanism in obesity unknown. But mediating the synaptic clustering of AMPA glutamate receptors at a subset of excitatory synapses (probable implication for appetite)	36504237	Yes	↓	↓	↓	↓
NTRK2	regulator of NTRK2 signaling pathways with downstream effect on cell growth, differntiation, survival and apoptosis via MAPK signaling	38935636	Yes	↑	↑
OBP2B	Direct rule unknown, but lipocalin-14 (50% ortholog with OBP2B) in mice promotes glucogenesis via AKT signaling/AQP7	26592241	No	↓	↓		↓
OCLN	Involved in regulation of cell growth, malignancy and inflammation pathways but not studied in the context of obesity	37809090	Yes	↓	↑
PALM	Promote cholesterol metabolism via cholesterom synthesis process.	38531951	No	↑	↑		↑
PDZD2	suggested to regulated adipose tissue insulin sensitivity via interaction with melatonin	DOI: 10.5353/th_991044069407003414	Yes	↓	↑
PLIN1	Interact with PPARG, NFkappaB, and LXRA and regulate lipolysis	28860604	Yes	↑	↑
PMVK	Encodes an enzyme that functions in cholesterol biosynthesis and convert mevalonic acid-5P to mevalonic acid 5-pyrophosphate which is critical in the regulation of the secretion of insulin in pancreatic β cells.	14585928	Yes	↑	↑	↑
PODXL2	Primarily reported for cancer. PODXL associated with invadopodia formation and to promote the epithelial-mesenchymal transition, tumor migration and invasion	28004467	Yes	↓	↓		↓
PRDX5	Antioxidant enzyme that promotes adipogenesis via suppressing peroxide accumulation, prolonging cell survival. Promote adipogenesis via ROS signaling suppression	16324708	Yes	↑	↑	↑
PRKAR1A	plays a crucial role in cAMP signaling pathway which has downstream effect on lipid and glucose metabolism	32318528	Yes	↓	↑
PRRT3	Function uncharacterized, but suggested to be associated with reduced growth hormone regulation from the pituitary	32803092	No	↓	↓	↑
PRXD5	Antioxidant enzyme that promotes adipogenesis via suppressing peroxide accumulation, prolonging cell survival. Promote adipogenesis via ROS signaling suppression	16324708	Yes	↓	↑
PSTPIP2	M2 macrophage polarization via activation of PPARγ which reduce adipose tissue inflammation via STAT6 pathway	37150118	Yes	↑	↑	↓
RBM17	regulator of mRNA splicing. Role in obesity unexplored but RBM45 from the protein family is implicated in de novo lipogenesis in malignancies	38040804	No	↓	↑
RGS10	Regulator G-protein signaling with critical role in modulation of macrophage M1/M2 activation via NF-kB signaling	24278459	Yes	↓	↑
RTN4R	RTN4 (Nogo) suppress thermogenesis in white adipose tissue via upregulating NF-kB. And RTN4B inhibit what adipose tissue adipogenesis via AKT2/GSK3β/β-catenin pathway.	34998825; 36931495	No	↑	↑		↑
S100A11	ositive regulator of AKT/mTOR signaling to induce lipid synthesis and lipid accumulation. Also reported as cancer promoter	33571540	Yes	↑	↑	↓	↑
SCARA5	positively associated with adipocyte differentiation via focal adhesion kinase (FAK) and ERK signaling pathways. And, glucocorticoids induced the expression of SCARA5 through glucocorticoids response elements (GRE) in the SCARA5 promoter. Positive regulator in adipocyte lineage commitment and early adipogenesis in mesenchymal stem cells	29093466	No	↑	↑		↑
SCG3	Induce angiogenesis, which results in endothelial proliferation, migration and tube formation through MEK/ERK signaling. Also, forms secretory granules with orexin, melanin-concentrating hormone (MCH), neuropeptide Y (NPY), and POMC in the hypothalamus. A potential regulator of food intake based on its capacity to accumulate appetite-related hormones into secretory granules	29154827; 19357184	Yes	↓	↓		↓	✓
SCGB3A1	plays row in regulation of growth factors with anti-inflammatory properties, but effect mechanism in obesity unexplored	35016921	Yes	↓	↓
SCGB3A2	plays row in regulation of growth factors with anti-inflammatory properties, but effect mechanism in obesity unexplored	35016921	Yes	↓	↓
SEMA3F	SEMA-3 mediated signaling drives hypothalamic-melanocortin development which is a regulator of appetite. Closely related SEMA3G promote adipogenesis via activation of PI3K/Akt/GSK3β signaling in the adipose tissue and the AMPK/SREBP-1c pathway in liver	30710099; 31648186	Yes	↑	↑
SEZ6L	Precise mechanism not reported. Could be via USF1 (intracellular cholesterol metabolism), and play role in lipid metabolism. May in the same network as F11 encoded gene. And mediate long-chain fatty acid (LCFA) and ver long-chain fatty acid (VLFCFA) transport across cell membrane.	35523305; 35583196	Yes	↓	↓		↓
SLC27A4	An acyl-CoA ligase catalyzing the ATP-dependent formation of fatty acyl-CoA using LCFA and very-long-chain fatty acids (VLCFA) that fatty acid efflux from cells and might drive more fatty acid uptake (lipid metabolism). Involved in EGFR signaling pathway.	23506886	Yes	↓	↑	↑
SLITRK2	implicated in neurodevelopment but role in obesity is unexplored	38283150	No	↑	↑
SLK	implicated in embryonic development via multiple signaling pathways including MAPK. Mechanism in obesity unexplored	24868594	No	↓	↓
SLURP1	Upregulate SFRP5 which in turn suppress triglyceride synthesis by suppressing fatty acid synthesis pathway	30879770	No	↑	↑		↑
SMPDL3A	downstream to lipid/cholesterol metabolism, and regulate inflammation via cGAS-STING pathway	37967525	Yes	↓	↓
SNCG	highly expressed in white adipose tissue. Affect adipogenesis, adipocyte differentiation and enlargement via PPAR gamma. Interact with INSR encoded protein via FABP4 encoded protein (both with causal association with obesity in this study)	25756178	Yes	↑	↑		↑
SORBS1	SORBS1 is CBL-adaptor protein. When insulin binds to insulin receptor (IR), CBL is recruited by interaction with SORBS1 and, upon phosphorylation, dissociates from IR and migrates to plasma membrane. This complex is involved in GLUT4 translocation to cell membrane, which may interfere with glucose uptake	32596667	Yes	↑	↑	↑	↑
SPRR3	plays role in neurodevelopment and keratinization but mechanism linking to obesity is unexplored	38279589	No	↑	↑
SRPX	differentially expressed in adipose tissue in weight loss but mechanism unexplored	22648723	Yes	↓	↓
SSC4D	Mechanism in obesity unknown. Act as scavenger and maybe involved in lipid accumulation given association with fat mass.	37550405	Yes	↑	↑		↑
TCL1A	Direct evidence does not exist. But association with obesity via PI3k/AKT signaling pathway and NF-κB, as evidenced by interplay with myocardial diseases	35499687	No	↑	↑	↓	↑
TCN2	Vitamin B12 transporter. Low B12 induces uptake of methyl malonyl-CoA leading to accumulation of methylmalonic acid (MMA). MMA is an inhibitor of the rate limiting enzyme carnitine palmitoyl transferase 1 (CPT1), critical for the breakdown of long chain fatty acids in the beta oxidation pathway. This results in accumulation of fatty acids and triglyceride.	32610503	Yes	↓	↑			✓
TGFBR2	Regulator of TGF-β signaling pathway and therefore a mediator of inflammation	29051557	Yes	↑	↑
THBS4	Mechanism unknown. Promote adaptive UPR24 by binding to the luminal domain of the endoplasmic reticulum (ER) stress transducer ATF6α and promoting ATF6α nuclear shuttling. Therefore, THBS4 may upregulates protective chaperones and improves endoplasmic reticulom stress. Also reported for reducing angiogenesis role.	34462518	Yes	↑	↑	↑
THY1	Promote Wnt/B-catenin signaling and conversely dampens PPAR gamma protein, which induces osteogenesis and suppress adipogenesis. Protective against obesity. Interaction of Wnt/B-catenin with Fyn signaling is promoted by THY1 with a causal role in T2D.	30089635	Yes	↑	↑	↓	↑
TIMP1	negative regulator of adipogenesis, and regulator of extracellular matrix	21437772	Yes	↓	↑
TNFRSF17	Activates MAPK and stimulates anti-apoptotic proteins, including Bcl-2 and Bcl-XL23 with a role in cell survival and proliferation. Role in obesity and cancer mediated by the same pathway.	36203424	Yes	↓	↑	↓
TNFRSF1B	Receptor with high affinity for TNFSF2/TNF-alpha, is promoter of apoptosis via NK-kB signaling. Long established with causal link to coronary artery diseases.	18758745	Yes	↓	↑	↓	↑
TNFRSF8	Promote cell survival and proliferation via NF-kB	34572446	Yes	↓	↑	↑↓
VCAM1	Activated by MAPK/PI3K/JAK2 signaling, promote interaction with leukocytes which in turn activate RAC1/ROS/MMP signaling leading to inflammation regulation	37442359	Yes	↓	↑	↑↓
YAP1	Mediate Hippo pathway effect via YAP/TAZ pathway and is involved in regulation of cholesterol in adipocytes.	36165813	Yes	↑	↑	↑↓	↑

Using 121 proteins with replicated BMI association in UK Biobank and CCHC, we next tested the relationships between proteins and obesity-related diseases using genomic approaches (Figure 2). Colocalization analyses demonstrated shared causal genetic variants affecting candidate proteins and susceptibility to obesity or obesity-associated diseases (Figure 2A). Seven proteins colocalized with at least one obesity trait (posterior probability >70%), several of which have exhibited roles in obesity and related pathophysiology in prior studies, including secretogranin III (SCG3; appetite regulation, metabolism), arginase 1 (ARG1; lipoprotein metabolism, obesity-related vascular dysfunction), follistatin-like 3 (FSTL3; insulin sensitivity), collagen XV (COL15A1; adipocyte survival), transcobalamin II (TCN2; fat metabolism), cytokine-like 1 (CYTL1; endothelial pro-angiogenic factor) and insulin receptor (INSR; Figure 2A).

Figure 2.

A: Bubble plot demonstrates proteins colocalized with obesity and/or obesity-related complications. Bubble sizes represent posterior probability (PP) of shared genetic variants. Teal blue show colocalized proteins-traits (with PP>0.7) or not (light red). B: Venn diagrams show the number of proteins potentially causal to obesity (red circle), dysregulated by obesity (blue circle), and potentially causal to obesity-related complications (red) and overlap between different sets based on forward mendelian randomization (FMR), and reverse MR (RMR) analyses. C: Arrowhead plot demonstrate protein causal to obesity-related diseases that are also: bidirectionally associated with obesity (diamond shape), potentially causal to obesity (arrowhead direction down), and potentially dysregulated by obesity (arrowhead direction up). Arrowhead sizes represent standardized effect sizes (in Mendelian Randomization results), and colors represent negative (blue) and positive effect directions (red

Subsequent bi-directional Mendelian randomization (MR) suggested a larger share of the obesity trait-associated proteome showed statistical evidence of being consequential of the obese state itself (51/117 nominally significant proteins studied of which 49/51 pass FDR <0.05 compared to 20/117 nominally significant for causal effects on BMI of which 9/20 pass FDR<0.05; Figure 2B) across well-established pathways of inflammation (HSPB6, HSPG2, IGSF3, TCL1A), dyslipidemia (GDP1, NCAN, S100A11, YAP1), and adipose tissue homeostasis (COMP, ITGAV, THY1, FABP4, IL1RN; Table 1). For this subset of proteins impacted by BMI, the effect sizes from our differential abundance analyses in CCHC and UK Biobank were largely concordant with MR estimated effects (CCHC: Pearson R²=0.63, UK Biobank R²=0.79, both P<0.001; Supplementary Figure 3)(11). Eighteen proteins displayed bidirectional effects with BMI in MR including pro-inflammatory heat shock proteins (HSPB6, HSPG2), proteins involved in thermogenesis pathways (ADAMTS15, CTHRC1), and lipid and glucose metabolism (ENO3, SORBS1). Our MR analyses further identified 20 protein candidates for therapeutic intervention, with nominally significant causal effects of protein on BMI which 9/20 proteins passed FDR >0.05), including clinically targetable pathways with roles in weight regulation (GFRAL), vasodilation (ARG1), adipose-tissue macrophage states (THBS4, CD248), and fatty acid metabolism (SLC27A4, PMVK), among others (Table 1).

We next explored colocalization and statistical causality of identified proteins on sequelae of obesity. We observed colocalization of several proteins with downstream obesity-related disease that have not been previously implicated in genetic obesity studies and which, to our knowledge, are not currently studied pharmacologic targets for human obesity (Figure 2C). As an example, in our study fibroblast growth factor 5 (FGF5; implicated in hepatic steatosis/fibrosis) colocalized with significant GWAS results of obesity complications (e.g., atrial fibrillation, coronary, and renal disease) but not with BMI. Similarly, the cadherin CELSR2 (involved in hepatic lipid metabolism and oxidative stress) colocalized with coronary disease and low-density lipoprotein level but not BMI. Additionally, by MR, we found 39 proteins with nominally significant causal effects on obesity-related illnesses (Figure 2B–C, Supplementary Table 4)(11), many of which are implicated in human metabolism (Figure 1D, Table 1).

This subset of 72 directionally consistent proteins includes known mediators of obesity and related diseases (e.g., LEP, FABP4, IGFBP1/2, FSTL3) as well as newer targets not widely described in humans, including CNTN3 (adipogenesis), IGSF3 (immunoglobulin), OBP2B (glucogenesis), PALM (cholesterol metabolism), PRRT3 (growth hormone regulation), RTN4R (adipogenesis), SLURP1 (triglyceride synthesis), and TCL1A (cell development). Transcriptionally, these 72 gene targets are broadly expressed across brain (13/72), adipose (10/72), liver (8/72), and arteries (8/72; Figure 3A). Consistent with this observation, fat, vasculature and the muscle exhibited the highest transcriptional activity of this gene set in the Tabula sapiens reference (with fibroblasts exhibiting the highest transcriptional activity; Figure 3B). These results were consistent at a tissue proteomic level, with greatest proteomic enrichment in adipose tissue (Figure 3C). Within adipose tissue at a single cell resolution, we observed the highest transcriptional activity in adipocyte, adipocyte progenitor stem cells (APSC) and pericytes (Figure 3C–D; larger range of cell types in Supplementary Figure 4)(11), all of which have been implicated in obesity-related complications(25,26). Lastly, using tissue proteomics, we observed the highest proteomic enrichment in adipose tissue, and to a lesser extent in placenta, urinary bladder, liver and smooth muscle (Figure 3E).

Figure 3.

A: Gene expression pattern of target genes whose encoded proteins potentially causally associate with obesity (prioritized genes) across diverse tissues. To increase specificity, plot is restricted to tissues where genes are expressed 10-folds higher than average across all measured tissues after scaling expression levels. B: Tissue and single-cell transcriptional activity scores of prioritized genes in Tabula Sapiens reference tissues (plots 1–3) and cells (plots 4–5). C: Within adipose tissue transcriptional activity scores of prioritized genes (plots 1–3). More intense colors represent higher transcriptional activity in plot 1. Highest activity level observed in adipocyte, then adipose stem and progenitor cells (ASPC) (plot 3). D: Gene expression enrichment patterns across diverse adipose tissues for prioritized genes in Single Nucleus Adipose Tissues Atlas. E: Tissue-specific protein enrichment patterns for prioritized genes in GTEX. Highest protein expression levels observed in adipose tissue.

Finally, to explore on- and off-target pharmacologic effects on obesity, we surveyed the 72 proteins for druggability (targets with empiric or mechanistic evidence in Supplementary Table 5)(11). Included proteins had both on- an off-target mechanisms/effects on adiposity at a class or specific pharmaceutical level, including Ceritinib (class effect, ALK inhibitor, increased weight gain), Exemestane (class effect, aromatase inhibitors, potentially reduced adiposity), Triiodothyronine (increased metabolic rates), MORAB-004 (CD248 inhibition, protection from obesity and insulin resistance), among others.

DISCUSSION

The success of surgical and pharmacological interventions for obesity which also appear to impact downstream disease sequalae underscores the clinical impact of identifying targets that limit obesity(27) as well as its downstream complications. While these interventions predate modern genomics(28), the use of human genomics to parse relevant drug targets is growing in popularity(29), owing largely to scale (≈10⁶ participants and growing) and ability to identify modifiable targets, pathways of metabolic disease and health(30), and heterogeneous drug effects(31). Nevertheless, findings from obesity GWAS have not been successfully therapeutically targeted outside of monogenic obesity forms [e.g., leptin receptor(32), POMC(33)], and novel approaches to discovery of causal modifiable targets are needed. In response, recent large efforts have demonstrated success in parsing disease phenotypes and genetic liability through proteome-wide association(7). Based on these successes, we developed a composite “proteogenomic” approach combining proteome-wide association of obesity traits in a high-risk population to integrate evidence from genetic and tissue studies and identify candidate casual biomarkers of obesity and its downstream complications.

Despite a ≈175-fold difference in sample size between our Hispanic/Latino sample and the large, ancestrally, race/ethnically and geographically distinct national UK Biobank sample, we identified proteomic associations with obesity and downstream disease sequelae, supporting the role of the proteome as a cross-population, transferrable reflection of obesity. Genomic variation underlying these clinical proteomic associations identified three overlapping categories of pathophysiologic significance: [1] a proteome dysregulated as a putative consequence of obesity; [2] proteome implicated in pathogenesis of obesity; [3] proteome causally implicated in obesity-related complications. Excluding proteins bidirectionally associated with obesity (18 proteins), the fact that more identified proteins were consequential to the obese milieu rather than being obesity-regulating corroborates previous studies in the literature and the wide-reaching taxing effects of obesity on metabolism and overall homeostasis(34).

While standard limitations to human genetics and proteomic approaches are relevant to the current report (e.g., sample size, power limitations in causal inference, unmeasured confounding), the broad reproducibility across ancestry, race/ethnicity and geography and multi-omic evidence (including tissue specificity) across obesity and its complications is notable. While our findings are supported by multiple lines of evidence strengthening the rigor of our results, key proteins involved in obesity pathogenesis and dysregulated by obesity may not have been identified by our studies, which have lower sample size relative to modern GWAS studies and in which assessments of statistical causality rely on availability of strong pQTL instruments.

Identified proteins revealed pathways involved in regulation of food intake, energy metabolism, extracellular matrix regulation, adipose tissue homeostasis, and inflammation. In addition to targets with substantial precedent in human obesity (e.g., leptin), this tiered “proteogenomic” approach pinpointed a host of proteins with biological plausibility not widely reported in large obesity genomic studies. Our prioritized proteins were associated with myriad obesity-related downstream disease sequelae, with directional consistency between genetic and clinical proteomic association as well as transcriptional and translational enrichment in fat (specifically adipocytes and pericytes) and fibroblasts, both critical in obesity-related inflammation and downstream metabolic complications.

Importantly, a survey of the druggable candidates among our directionally consistent results identified pharmaceuticals with potential on- and off-target mechanisms for disrupting obesity pathogenesis. Interestingly, there are both candidate drugs identified in our studies operate through inhibition as well as excitation of the protein (Supplementary Table 5)(11). More specifically, inhibitory candidate drugs include those affecting the master regulator of adipogenesis peroxisome proliferator-activated receptor gamma (PPAR-γ) (Certinib, Levimir, Exemstane), thermogenesis and browning of adipose tissue (Dipyridamole, IUPHAR.LIGAND:5269), lipid metabolism (BVT-24834), and those with inhibitory effects on inflammation (NNZ-2566, Succinobucol, MORAB-004). In contrast, excitatory candidate drugs potentially affect obesity through the promotion of thermogenesis (Triiodothyronine, selective beta-2-adrenoreceptor agonists), arterial smooth muscle relaxation (Taladafil, Vasopressin), and appetite regulation (Lithium). These observations underscore the interrelatedness and complexity of the human proteome. Collectively, these results highlight [1] the power of multi-omic approaches to focus discovery on biomarkers with statistically causal and consequential relationships to both obesity and its related diseases and [2] the consistency of the human proteome as a molecular barometer of obesity pathophysiology across populations.

Most discovered proteins had supportive evidence in human metabolism and adipose tissue homeostasis (e.g., CNTN3, RTN4R, SLURP1: adipogenesis; SCARA5, PALM, CDON: glucose and lipid metabolism; ENO3, OBP2B: thermogenesis and appetite regulation: ADAMTS15, FURIN). Our approach also identified proteins in pathways that have not been widely reported in obesity but that share functional similarity to those previously implicated (e.g. inflammatory pathways). Finally, integration with tissue proteomic and transcriptomic data (at single cell resolution) highlighted the multi-system involvement of the obesity proteome (brain, adipose, liver, blood/inflammatory cell), specifically indicating a role for the adipocyte and related precursors. These tissue findings are consistent with putative mechanistic roles for these cell types in extracellular matrix remodeling, adipose tissue expansion, hypoxia, and inflammation(35), and trans-differentiation of fibroblasts into eventual adipocytes(36).

The current report demonstrates the utility of a proteogenomic approach in which we have anchored our discovery in identification and replication of differentially abundant proteins and interpreted their function through a genomic lens, leading to identification of physiologically plausible biomarkers (Table 1). Thus, uniting circulating proteomics, genomics (bidirectional MR and colocalization), and tissue proteomics/transcriptomics to understand the molecular intersection of obesity and complications appears to be a successful paradigm moving forward. While we recognize that previous studies in the UK Biobank have estimated association with BMI(37), the current study further demonstrates concordance of effects across ancestry, race/ethnicity, geography, culture, social determinants of health, and obesity phenotyping, supporting transferability of the obesity proteome. While the integrated approach in this study provided strong evidence for the potential role of prioritized proteins in the pathophysiology of obesity and downstream metabolic sequelae, longitudinal follow-up in ancestrally diverse populations can further strengthen and substantiate observed causal mechanisms (once data become available).

In conclusion, our landmark study revealed key regulatory mechanisms influenced by and influencing obesity, offering valuable targets for biomarkers and clinical interventions. We argue that integration of this multi-omic prioritization approach across phenotypes and disease states in modern genomic discovery pipelines is warranted and may increase the precision for potentially viable pharmacologic targets in human disease.

Data Availability

The data supporting this study’s findings are available upon request from CCHC. The CCHC data are not publicly accessible as they contain information that could compromise participants’ privacy and consent. pQTL data utilized in MR can be found in the online repository of the UK Biobank Pharma Proteomics Project (UKB-PPP) at synapse.org. GWAS data used in MR are available from the GWAS catalog (ebi.ac.uk/gwas/). Phenotype data used in replication can be accessed from the UKBB with approval.