Sage Journals: Discover world-class research

Abstract

Background

Obesity is heterogeneous; standard metrics (BMI, waist–hip) conflate fat distribution with muscle, limiting causal inference for venous thromboembolism (VTE). This Mendelian randomization (MR) study leverages four magnetic resonance imaging (MRI)-defined adiposity axes to assess VTE causality and inflammatory cytokines mediation.

Methods

Independent genome-wide significant instruments were selected from dates of the general-obesity, lower-body fat, muscle-dominant, and peripheral fat axes. Outcomes were VTE and its subtypes—deep vein thrombosis (DVT) and pulmonary embolism (PE)—from FinnGen-R12, ensuring non-overlapping samples. The primary analysis used inverse-variance weighting with false discovery rate (FDR) control. Robustness was evaluated using eight complementary MR estimators alongside tests for horizontal pleiotropy and heterogeneity. Two-step mediation MR was used to investigate the obesity-inflammation-thrombosis pathway.

Results

Genetically proxied general-obesity axis increased risks of VTE (OR 1.431, 95% CI 1.152-1.778), DVT (OR 1.646, 95% CI 1.124-2.410), and PE (OR 1.273, 95% CI 1.150-1.410); The lower-body-fat axis also raised VTE (1.246, 1.141-1.361, 1.20 × 10⁻⁵) and DVT (1.216, 1.039-1.424, 0.036); all P_FDR< 0.05. Mediation showed CTACK/CCL27 accounted for 8.30% (0.07-16.54%) of General-obesity→ PE, while beta-nerve growth factor (Beta-NGF) and monocyte chemoattractant protein-3 (MCP-3) explained 6.77% (1.25-12.30%) and 6.44% (0.08-12.80%) of general-obesity→VTE. In lower-body-fat→ VTE, platelet-derived growth factor BB (PDGF-BB) and monokine induced by gamma interferon (MIG) mildly masked 5.79–7.89% without altering the direct effect.

Conclusions

This study indicates that inflammation partly mediates general-obesity axis effects on VTE and its subtypes, while lower-body fat axis confers VTE/DVT risk chiefly via local venous hemodynamic pathways.

Keywords

mendelian randomization obesity venous thromboembolism adiposity axes deep vein thrombosis pulmonary embolism

Introduction

Venous thromboembolism (VTE), encompassing deep vein thrombosis (DVT) of the lower extremities and pulmonary embolism (PE), represents the third leading cause of vascular-related mortality worldwide and a major contributor to death and disability.¹ This condition affects approximately 10 million individuals globally each year, with a lifetime risk exceeding 1 in 12, and these figures are projected to escalate amid population aging.² Long-term sequelae of VTE include post-thrombotic syndrome and post-pulmonary embolism syndrome, imposing substantial burdens on patients and socioeconomic systems.^3,4 Nevertheless, despite the widespread adoption of anticoagulation prophylaxis, the overall incidence of VTE has not declined substantially, with over one-third of cases being idiopathic without identifiable triggers and their pathogenesis remaining incompletely understood.⁵ Accordingly, it is essential to identify and assess VTE risk factors to inform more targeted and effective interventions and ultimately reduce population-level risk.

Obesity is recognized as one of the foremost lifestyle risk factors for VTE, accounting for an attributable risk of 10%–30%.⁵ Obesity induces a cascade of pathological changes.^6,7 In individuals with excess adiposity, hypertrophied adipocytes and infiltrating immune cells secrete pro-inflammatory cytokines and adipokines, sustaining chronic low-grade inflammation and promoting systemic insulin resistance. Lipid overflow and ectopic fat deposition in organs such as the liver and skeletal muscle further aggravate insulin resistance and dyslipidemia. At the vascular level, these inflammatory and metabolic disturbances impair endothelial nitric oxide bioavailability, increase oxidative stress, and upregulate adhesion molecules, thereby leading to endothelial dysfunction. In parallel, obesity is associated with elevated levels of coagulation factors and plasminogen activator inhibitor-1, as well as enhanced platelet activation, which collectively enhance the coagulation cascade and foster a prothrombotic milieu, thereby elevating VTE risk. However, obesity is a highly heterogeneous condition; individuals exhibit marked heterogeneity in metabolic status, complication risks, and responses to weight loss, including a transient “metabolically healthy” obese subtype.⁸ Conventional metrics, such as BMI >30 kg/m², waist circumference, hip circumference, or waist-to-hip ratio, fail to differentiate fat from muscle mass and overlook distributional variations in visceral, subcutaneous, and lower-body fat, thus inadequately addressing the demands of precision medicine for risk stratification.^8,9 Moreover, traditional observational studies are susceptible to reverse causation and confounding. For example, preclinical or undiagnosed disease may lead to changes in body weight or lifestyle, making it difficult to determine whether obesity is a cause or a consequence of the outcome. In addition, residual confounding from lifestyle, socioeconomic, and clinical factors may persist even after multivariable adjustment. As a result, such studies typically identify statistical associations rather than establishing a unidirectional causal effect.

Mendelian randomization (MR) leverages the “natural randomization” of genetic variants to infer causality while mitigating confounding and reverse causation.¹⁰ Recent MR studies have confirmed causal effects of adult and childhood BMI, waist circumference, and hip circumference on VTE risk^11–14; however, these aggregate measures do not capture the heterogeneity in body composition or underlying mechanisms. Recently, a genome-wide association study (GWAS) based on whole-body magnetic resonance imaging (MRI) from the UK Biobank decomposed body composition into four mutually independent adiposity axes, offering novel insights into obesity heterogeneity.¹⁵ The general obesity axis reflects synchronous increases in visceral, subcutaneous, and ectopic fat; the muscle-dominant axis is characterized by relatively higher muscle mass; the peripheral adiposity axis focuses on abdominal and thigh subcutaneous fat accumulation; and the lower-body fat axis manifests as increased gluteofemoral fat with reduced ectopic fat. These axes exhibit distinct genetic backgrounds, metabolic profiles, and disease associations, providing a new phenotypic framework for dissecting obesity-related pathologies. Accordingly, this study employs two-sample MR to systematically evaluate the causal impacts of these adiposity axes on VTE and its subtypes (DVT and PE). Additionally, two-step mediation MR is used to quantify the mediating roles of inflammatory cytokines in the “adiposity axis → VTE” pathway. We hypothesize that different adiposity axes exert distinct causal effects on VTE and its subtypes through axis-specific inflammatory and hemodynamic pathways, and that clarifying these patterns may help refine risk stratification, screening, and preventive strategies beyond traditional measures such as BMI.

Methods

Study Design

The analytical workflow is shown in Figure 1. We followed STROBE-MR reporting guidelines to ensure transparency.¹⁶ GWAS summary statistics for the four adiposity axes served as exposures, and independent variants passing stringent instrument-selection criteria were used as instrumental variables (IVs) in two-sample MR analyses. Outcomes were VTE, PE, and DVT; data sources were chosen to minimise sample overlap. Positive associations from the primary MR stage were examined in two-step mediation MR to construct adiposity-inflammation-thrombosis pathways. All GWAS datasets are publicly available and had received the requisite ethical approvals; as this study is a secondary analysis of de-identified data, no additional approval was required.

Figure 1.

Overview of the study design and analytical workflow. The upper panel depicts the identification of four MRI-derived obesity axes (general obesity, muscle-dominant, peripheral fat, and lower-body fat) via principal component analysis of 24 fat and muscle volume traits from UK Biobank imaging data. Genome-wide association studies (GWAS) identified 41 genetic loci reaching genome-wide significance. The middle panel illustrates the selection of genetic instrumental variables (IVs) and the application of two-sample Mendelian randomization (MR) to evaluate causal effects on venous thromboembolism (VTE) outcomes. Inflammatory cytokines (n = 41, categorized into interleukins, chemokines, growth factors, TNF superfamily, colony-stimulating factors, and other mediators) were incorporated for mediation analyses. The lower panel outlines the two-step mediation MR framework, including estimation of total, direct, and indirect effects via mediators, alongside sensitivity analyses (eg, F-statistic > 10, MR-Steiger filtering, MR-Egger regression intercept, MR-PRESSO global test, penalized weighted median, RAPs, dIVW, BWMR, and Cochrane's Q test for heterogeneity).

Mendelian Randomization Assumptions

MR inference rests on three core assumptions¹⁷: (i) relevance—genetic variants are strongly associated with the exposure; (ii) independence—variants are independent of confounders; and (iii) exclusion restriction—variants affect the outcome exclusively through the exposure. Relevance was assessed by instrument-selection thresholds and F-statistics. Independence and exclusion-restriction assumptions, which cannot be tested directly, were evaluated through complementary sensitivity analyses (see below).

Data Sources

Adiposity Axes Sources and Genetic Instrument Selection

Data for the four adiposity axes were derived from the UK Biobank imaging subcohort.¹⁵ The research team extracted 24 fat/muscle volume metrics from neck-to-knee Dixon and abdominal multi-echo MRI scans of 33,122 participants, performing principal component analysis stratified by sex to yield four uncorrelated axes (general adiposity, muscle-dominant, peripheral subcutaneous fat, and lower-body fat). GWAS was conducted in 25,637 White British individuals who passed genotyping and imaging quality control (using BOLT-LMM, with covariates including age, age², scanning center, genotyping array, and the first 40 genetic principal components), identifying 41 genome-wide significant loci: 2 for the general obesity axis, 9 for the muscle-dominant axis, 15 for the peripheral subcutaneous fat axis, and 15 for the lower-body fat axis (Table S1). The original GWAS reported no sex-specific effects of the obesity axes, and sex-combined summary statistics were therefore used in the present analysis. All axis scores underwent rank-based inverse normal transformation; thus, subsequent MR effect sizes represent changes per 1 standard deviation (SD) unit on the Z-score scale. For MR analyses, SNPs were initially selected under genome-wide significance and independence criteria (P < 5 × 10^–8; r² < 0.001 within 10 Mb), followed by matching with outcome genetic variants. Palindromic SNPs and those with intermediate allele frequencies were excluded after allele harmonization. Notably, effect allele frequencies (EAFs) for each SNP were imputed based on European ancestry data from Phase 3 of the 1000 Genomes Project, using dbSNP151 (Build 37) and the GWAS Catalog. To ensure precision, missing matches with outcomes were not imputed using proxy SNPs. Instrument strength was further evaluated by calculating R² and F-statistics, excluding weak variants (F-statistic < 10).^18,19 Steiger filtering was then applied based on corresponding sample sizes and R² to confirm that genetic variants influenced the outcome solely through the exposure and not in the reverse direction.

Venous Thromboembolism

GWAS data for VTE (finngen_R12_I9_VTE; 26,333 cases and 474,015 controls), PE (finngen_R12_I9_PULMEMB; 12,762 cases and 486,319 controls), and DVT (finngen_R12_I9_PHLETHROMBDVTLOW; 8134 cases and 432 223 controls) were obtained from the latest R12 release of the FinnGen consortium.²⁰ Diseases were defined using ICD-10/9/8 codes, with detailed descriptions available in FinnGen Risteys. All GWAS underwent quality control adjusted for up to 20 genetic principal components, including age and sex. No sample overlap was detected between exposure and outcome cohorts, ensuring causal estimates were not biased by winner's curse.

Inflammatory Cytokines

Given the adiposity-inflammation-thrombosis pathway, genetic data for inflammatory cytokines were sourced from the GWAS by Ahola-Olli et al, comprising three population cohorts: the Cardiovascular Risk in Young Finns Study, FINRISK 1997, and FINRISK 2002, totaling 8293 European-ancestry participants.²¹ All individuals were genotyped using the Illumina OmniExpress array and imputed genome-wide using the 1000 Genomes Phase 3 reference panel; cytokine concentrations were measured via Bio-Rad Bio-Plex 27-plex and 21-plex magnetic bead arrays. Raw values were log-transformed and residualized via linear regression on age, sex, BMI, and the first 10 genetic principal components. Forty-one inflammatory cytokines were included in the two-step mediation analyses, with detailed phenotypic descriptions in Table S2.

Statistical Analysis

The primary method employed inverse-variance weighted (IVW) estimation,²² with false discovery rate (FDR) correction applied to significance thresholds. Associations were deemed significant when P < .05 and P_FDR < 0.05, and suggestive when P < .05 but P_FDR> 0.05. Validation analyses incorporated MR-Egger, weighted median, penalized weighted median, contamination mixture (ConMix), robust adjusted profile score (RAPS), debiased IVW (dIVW), constrained maximum likelihood (cML), and Bayesian weighted MR (BWMR). Robustness was affirmed when over half of the validation methods yielded consistent direction and evidence with the primary analysis. Statistical power for binary outcomes was assessed using an online tool. Sensitivity analyses included Cochrane's Q tests for heterogeneity using IVW and MR-Egger models, followed by I² calculation to select the primary effect model. Horizontal pleiotropy was evaluated via the MR-Egger intercept²³ and Mendelian Randomization Pleiotropy RESidual Sum and Outlier (MR-PRESSO)²⁴ for outlier detection and removal. Leave-one-out analyses were additionally performed to assess whether causal associations were driven by single SNPs.

In two-step mediation MR, the genetic effect of the exposure on the mediator was first estimated, followed by the mediator's effect on the outcome; their product yielded the indirect (mediated) effect. The total effect (exposure SNPs on outcome) minus the indirect effect provided the direct effect. The mediation proportion was quantified as (indirect effect ÷ total effect), with 95% confidence intervals (CIs) propagated using the delta method.²⁵ All statistical analyses were conducted in R version 4.2.3, primarily using the TwoSampleMR, mr.raps, BWMR, MVMR, MendelianRandomization, and MRPRESSO packages.

Results

Genetic Instrument Selection, Strength, and MR Assumption Evaluation

The number of instrumental SNPs included in the analyses ranged from 2 to 11, collectively explaining 0.34%–2.09% of the phenotypic variance. All individual SNP F-statistics exceeded 10, with mean F-values per analysis ranging from 27 to 51, indicating the absence of weak instrument bias. MR-PRESSO outlier detection removed a total of 9 aberrant SNPs; Steiger directionality tests yielded “TRUE” for all, confirming the causal direction from exposure to outcome. Overall, all instruments satisfied the MR assumptions of relevance, independence, and exclusion restriction; detailed information is provided in Table S3.

Causal Inference of Four Adiposity Axes on Venous Thromboembolism

Following FDR multiple correction (Figure 2), robust evidence supported a significant causal association between a genetically predicted 1-SD increase in the general obesity axis and elevated risks of VTE (OR = 1.431, 95% CI: 1.152-1.778, P = .001, P_FDR = 0.005), DVT (OR = 1.646, 95% CI: 1.124-2.410, P = .010, P_FDR = 0.031), and PE (OR = 1.273, 95% CI: 1.150-1.410, P = 3.30 × 10⁻⁶, P_FDR = 1.98 × 10⁻⁵). Additionally, a genetically predicted increase in the lower-body fat axis was causally associated with heightened risks of VTE (OR = 1.246, 95% CI: 1.141-1.361, P = 1.00 × 10⁻⁶, P_FDR = 1.20 × 10⁻⁵) and DVT (OR = 1.216, 95% CI: 1.039-1.424, P = .015, P_FDR = 0.036). Except for the general obesity axis-PE association, all positive findings were replicated in ≥50% of supplementary methods, with statistical power ≥80% (Table S4). Although the general obesity axis-PE association had insufficient power (power = 42%), it was replicated consistently in an independent cohort (ebi-a-GCST90013887; OR = 1.258, 95% CI: 1.112-1.424, P = 2.54 × 10⁻⁴), thus considered robust. Neither IVW nor MR-Egger Cochrane's Q tests detected heterogeneity (Q-Pval > 0.05), and I² values (<50%) corroborated consistency, supporting the use of fixed-effects models in primary analyses. MR-Egger intercepts (intercept-Pval > 0.05) and MR-PRESSO global tests (global test Pval > 0.05) revealed no evidence of horizontal pleiotropy (Table 1). Leave-one-out analyses confirmed that causal estimates were not driven by single SNPs.

Figure 2.

Forest plot illustrating causal effects of four adiposity axes on VTE, DVT, and PE using inverse-variance weighted Mendelian randomization. Columns include number of SNPs, F-statistic, statistical power (%), odds ratio (OR) with 95% CI, p-value, and FDR-adjusted p-value. Significant associations (FDR-p < .05) are evident for the general obesity axis with all outcomes and the lower-body fat axis with VTE and DVT.

Table 1.

Summary of Sensitivity-Analysis Results.

Exposure	Outcome	MR-Egger Intercept			MR-PRESSO Global Test			Cochrane's Q-IVW			Cochrane's Q-MR-Egger			I² (%)
Exposure	Outcome	Intercept	SE	Pval	RSS_obs	Pval	Outlier	Q	Q_df	Q_pval	Q	Q_df	Q_pval	I² (%)
General obesity	DVT	NA	NA	NA	NA	NA	NA	1.322	1	0.250	NA	NA	NA	24
General obesity	PE	NA	NA	NA	NA	NA	NA	0.112	1	0.738	NA	NA	NA	0
General obesity	VTE	NA	NA	NA	NA	NA	NA	0.253	1	0.615	NA	NA	NA	0
Lower-body fat	DVT	−0.019	0.021	0.382	16.645	0.164	rs2287922, rs6888037	13.088	9	0.159	11.823	8	0.159	31
Lower-body fat	PE	−0.007	0.017	0.671	11.825	0.411	rs2287922, rs3747207	9.652	9	0.379	9.423	8	0.308	7
Lower-body fat	VTE	−0.010	0.011	0.368	17.506	0.213	rs2287922	13.846	10	0.180	12.588	9	0.182	28
Muscle dominant	DVT	0.004	0.027	0.873	5.945	0.724	NA	4.490	7	0.722	4.462	6	0.614	0
Muscle dominant	PE	0.002	0.025	0.945	10.580	0.345	NA	7.830	7	0.348	7.824	6	0.251	11
Muscle dominant	VTE	−0.003	0.020	0.883	14.550	0.163	NA	10.771	7	0.149	10.729	6	0.097	35
Peripheral fat	DVT	0.032	0.034	0.363	18.294	0.142	rs2267373	14.832	10	0.138	13.457	9	0.143	33
Peripheral fat	PE	0.027	0.027	0.344	18.124	0.121	rs2267373, rs9565581	14.873	10	0.137	13.388	9	0.146	33
Peripheral fat	VTE	−0.009	0.016	0.581	12.600	0.402	rs2267373	10.074	10	0.434	9.719	9	0.374	1

Two-Step Mediation Mendelian Randomization Analysis

In the two-step mediation MR analyses, an “intersection” strategy was employed to identify potential mediators. We first assessed the genetic effects of the four adiposity axes on 41 inflammatory cytokines, followed by examining the causal effects of these cytokines on VTE and its subtypes (due to the limited scale of cytokine GWAS, the selection threshold was relaxed to 5 × 10⁻⁶). A cytokine was deemed a potential mediator only if all three paths—“axis → cytokine,” “cytokine → outcome,” and “axis → outcome”—reached statistical significance (P < .05), whereupon it was incorporated into the mediation model with the corresponding adiposity axis. Results identified 19 positive associations from adiposity axes to cytokines (9 for the general obesity axis and 10 for the lower-body fat axis) (Figure 3 and Table S5); 24 positive associations from cytokines to VTE subtypes (9 for VTE, 6 for DVT, and 9 for PE) (Figure 3 and Table S6); ultimately yielding 9 intersecting cytokines for mediation analysis (Table 2).

Figure 3.

Circos plot depicting causal associations from two-step Mendelian randomization analyses. Inner links show effects of general obesity and lower-body fat axes on 41 inflammatory cytokines (OR values), while outer links illustrate cytokine effects on VTE, DVT, and PE (p-values and ORs).

Table 2.

Summary of Two-Step Mediation Mendelian Randomization Analysis Results.

Exposure Axis	Mediator Phenotype	Outcome	Confidence Interval of Total Effect (95% CI)	Confidence Interval of Direct Effect (95% CI)	Confidence Interval of Mediation Effect (95% CI)	Confidence Interval of the Proportion of Mediation Effect	Pval
General obesity	CTACK	PE	0.242 (0.140, 0.344)	0.222 (0.118, 0.325)	0.020 (0.000, 0.040)	8.30% (0.07%, 16.54%)	0.047
General obesity	beta-nerve growth factor	PE	0.242 (0.140, 0.344)	0.216 (0.111, 0.321)	0.026 (−0.001, 0.053)	10.69% (−0.36%, 21.74%)	0.054
General obesity	beta-nerve growth factor	VTE	0.359 (0.249, 0.468)	0.334 (0.223, 0.445)	0.024 (0.004, 0.044)	6.77% (1.25%, 12.30%)	0.015
General obesity	Monocyte chemoattractant protein-3	VTE	0.359 (0.249, 0.468)	0.335 (0.224, 0.447)	0.023 (0.000, 0.046)	6.44% (0.08%, 12.80%)	0.047
Lower-body fat	TRAIL	DVT	0.212 (0.058, 0.366)	0.220 (0.066, 0.374)	-0.008 (-0.019, 0.003)	-3.77% (-8.74%, 1.19%)	0.118
Lower-body fat	Platelet-derived growth factor BB	DVT	0.212 (0.058, 0.366)	0.237 (0.081, 0.393)	-0.025 (-0.051, 0.001)	-11.77% (-23.96%, 0.43%)	0.053
Lower-body fat	Macrophage inflammatory protein 1b	DVT	0.212 (0.058, 0.366)	0.219 (0.064, 0.373)	-0.007 (-0.015, 0.002)	-3.08% (-7.13%, 0.98%)	0.119
Lower-body fat	Platelet-derived growth factor BB	VTE	0.177 (0.094, 0.261)	0.192 (0.107, 0.277)	-0.014 (-0.028, 0.000)	-7.89% (-15.71%, -0.08%)	0.044
Lower-body fat	Monokine induced by gamma interferon	VTE	0.177 (0.094, 0.261)	0.188 (0.103, 0.272)	-0.010 (-0.021, 0.000)	-5.79% (-11.59%, 0.01%)	0.045

Mediation analyses revealed (Table 2) that the general obesity axis mediated 8.30% (95% CI: 0.07%-16.54%) of its effect on PE through CTACK, and 6.77% (1.25%-12.30%) and 6.44% (0.08%-12.80%) of its effects on VTE through beta-nerve growth factor (beta-NGF) and monocyte chemoattractant protein-3 (MCP-3), respectively. In the lower-body fat axis-VTE pathway, mild masking effects were observed for platelet-derived growth factor BB (PDGF-BB) and monokine induced by gamma interferon (MIG, also known as CXCL9; mediation proportions: 5.79%-7.89%), yet the direct effect on VTE risk persisted (β_total = 0.178, β_direct = 0.188-0.191).

Discussion

This study used MR to investigate the causal effects of four genetically defined adiposity axes on VTE and its subtypes and to quantify the mediating roles of inflammatory cytokines. Key findings indicate that a genetically predicted increase in the general obesity axis causally elevates risks of VTE, DVT, and PE, whereas the lower-body fat axis increases risks of VTE and DVT. Mediation effects by inflammatory cytokines reached up to 8.30%, highlighting the heterogeneous contributions of distinct adiposity phenotypes to thrombotic diseases.

The general obesity axis is characterized by synchronous increases in visceral, subcutaneous, and ectopic fat across the body. Compared to singular anthropometric measures (eg, BMI, waist circumference [WC], or waist-to-hip ratio [WHR]), it better captures the “generalized” obesity phenotype of coordinated fat depot expansion at equivalent total fat levels, thereby mitigating heterogeneity dilution from conflating diverse fat distribution patterns in prior studies (with consistent genetic and metabolic associations across sexes).¹⁵ Our MR results demonstrate robust risk effects of this axis on VTE and its subtypes (DVT and PE), aligning with previous MR investigations linking genetically determined overall obesity to heightened VTE risk.¹¹ Furthermore, integrated prospective cohort and MR studies suggest that WC and BMI-adjusted WC outperform BMI in predicting VTE, underscoring abdominal fat burden as a critical risk component.¹³ Recent MR analyses further emphasize that visceral adipose tissue (VAT) volume exhibits a stronger association with VTE than BMI, highlighting the disproportionate contribution of “central fat” to thrombosis.²⁶ Relative to traditional phenotypes, the general obesity axis represents an MRI-derived composite phenotype that captures covariation between total and central fat within a single dimension, enhancing signal-to-noise ratio methodologically and aligning biologically with the true prothrombotic exposure (concurrent increments in total and central fat).

Intriguingly, although gluteofemoral (lower-body) fat is metabolically regarded as relatively “beneficial”—associated with higher adiponectin levels, lower proinflammatory cytokine profiles, and protection against ectopic fat deposition via prolonged fatty acid sequestration²⁷—our findings underscore that metabolic protection does not equate to venous hemodynamic benefits (OR > 1). From an anatomic and biomechanical perspective, a larger gluteofemoral fat depot increases lower-limb girth and soft-tissue mass around the deep venous system. This may augment the hydrostatic pressure column, exert extrinsic compression on the femoral and popliteal veins, and increase valve leaflet stress, all of which favor venous pooling, reflux, and chronic venous insufficiency. Within Virchow's triad, these changes predominantly affect the “stasis” component: lower-extremity venous return is highly sensitive to body habitus and gravitational forces, such that any factor that reduces venous flow velocity or impairs venous emptying can amplify DVT risk.²⁸ Specifically, the gastrocnemius muscle pump serves as the “engine” for lower-limb venous return, its efficiency influenced by limb girth, local compliance, and physical activity; increased lower-body subcutaneous fat burden may promote flow deceleration and valvular stress through elevated hydrostatic pressure, diminished muscle pump efficacy, and extrinsic “compression” on femoral/popliteal veins, thereby inducing reflux and chronic venous insufficiency.^29–31 Concurrently, obesity-related immobility scenarios (eg, prolonged sitting, standing, hip flexion postures, and reduced activity) exacerbate lower-limb venous “relative stasis,” compounded by elevated intra-abdominal pressure to drive venous stagnation.³² In line with this hemodynamic rationale, we observed a stronger effect of the lower-body fat axis on DVT than PE (OR = 1.216 vs OR = 1.061), with only minimal and directionally inconsistent mediation by inflammatory cytokines (PDGF-BB and CXCL9/MIG). This pattern suggests that, for the lower-body adiposity axis, venous stasis and local mechanical loading on the deep venous system are likely to be the predominant drivers of thrombosis, whereas systemic inflammation and hypercoagulability play a comparatively smaller role within the measured cytokine network. Clinically, these results imply that individuals with pronounced gluteofemoral fat accumulation may carry an underappreciated venous risk despite a relatively favorable metabolic profile, reinforcing the need for early mobilization, muscle pump–targeted strategies, and careful management of immobility in perioperative settings, long-haul travel, and other high-stasis contexts.

When considered within the full Virchow triad, our findings suggest that different adiposity axes engage distinct components of thrombogenic risk.^33,34 For the general obesity axis, the partial mediation by chemokine and growth factor pathways (CTACK/CCL27, β-NGF, and MCP-3/CCL7) points toward a contribution from systemic inflammation–driven hypercoagulability and endothelial perturbation: these mediators are implicated in leukocyte recruitment, immune–coagulation crosstalk, platelet–endothelial interactions, and vascular remodeling, which collectively can enhance thrombin generation and impair endothelial antithrombotic function. In contrast, for the lower-body fat axis, the weak and inconsistent mediation by the same cytokine panel, combined with a relatively stronger effect on DVT than on PE, is more consistent with a stasis-dominated mechanism, whereby mechanical and hemodynamic alterations in the lower extremities play the leading role and systemic inflammatory/hypercoagulable changes are less prominent within the measured cytokine spectrum. Thus, obesity-related VTE risk in our study appears to arise from an interplay between Virchow's triad components: generalized adiposity mainly amplifies hypercoagulability and endothelial dysfunction via inflammatory signaling, while gluteofemoral adiposity predominantly exacerbates venous stasis, with both pathways ultimately converging on an increased propensity for thrombosis.

Two-step mediation MR identified that the general obesity axis's risks for VTE/PE are partially mediated by chemokine and growth factor pathways—CTACK/CCL27, β-NGF, and MCP-3/CCL7 were pinpointed as mediators. CCL7 (MCP-3), a potent chemoattractant for monocytes/neutrophils, may amplify venous thrombotic propensity via enhanced monocyte-macrophage recruitment, tissue factor expression, and NETosis interactions, consistent with its pathological role in cardiovascular inflammation.³⁵ β-NGF contributes to angiogenesis and immune activation, suggesting that obesity-associated tissue remodeling and immune-coagulation coupling could heighten embolic susceptibility through the NGF axis (albeit with limited reports in VTE contexts).³⁶ CTACK/CCL27, traditionally linked to skin-mucosal immune homing, implies a potential role in systemic immune activation-coagulation crosstalk per our results—a previously underreported VTE risk link warranting further validation. Additionally, we noted mild masking by PDGF-BB and CXCL9/MIG in the lower-body fat axis pathway, aligning with their roles in platelet activation/endothelial remodeling and Th1 immunity; independent studies also suggest PDGF-BB's genetic-causal links to VTE risk or involvement in upstream metabolic-coagulation pathways.³⁷ Distinct from prior MR studies focusing on obesity^11–14^,26 or inflammatory factors³⁸ within thrombosis frameworks, this investigation explores VTE subtype risks using MRI-derived composite adiposity phenotypes and incorporates inflammatory cytokines into mediation analyses. It pioneers MR examination of the obesity-cytokine-thrombosis axis. Our two-step mediation MR expands the “obesity-inflammation-thrombosis” mediation spectrum from conventional cytokines to specific chemokine/growth factor networks, offering actionable candidate molecules for targeted interventions and stratified warnings.

This study has several strengths. First, it advances beyond BMI-based risk stratification, fostering exploration of axis-specific genetic mechanisms. Moreover, the robustness of causal evidence from two-sample MR hinges on core assumptions; beyond routine sensitivity analyses, we incorporated nearly 10 supplementary methods (eg, BWMR) for validation. Compared to predecessors, this is the first to integrate multiple adiposity axes for obesity heterogeneity and probe inflammatory cytokine mediation. By delineating causal graphs linking adiposity axes, cytokines, and thrombosis, the study theoretically deepens understanding of obesity's pleiotropic impacts and provides novel insights for precision medicine.

Despite validation of MR assumptions via multiple sensitivity analyses, limitations persist. As a secondary analysis of publicly available summary-level GWAS data, we could not conduct subgroup analyses by sex or age (original GWAS already controlled for relevant principal components). Future work using larger individual-level cohorts with sex- and age-stratified GWAS, potentially combined with advanced computational or machine-learning approaches, may enable more refined investigation of effect heterogeneity across demographic subgroups. Although the primary studies noted consistent significant genetic loci and effect directions across sexes, potentially minimizing sex differences. Furthermore, our results reflect lifelong genetic susceptibilities of adiposity axes to VTE and subtypes. MR assumes lifelong exposure differences; it may not fully capture intervention effects (eg, acute cytokine reduction via drugs). Thus, while we implicate certain cytokines in causal chains, interventional studies are needed for confirmation. Second, due to limited cytokine GWAS scale, we relaxed selection thresholds in mediation analyses; future targeted, larger-scale, broader-phenotype GWAS are required for validation. In particular, we adopted a lenient instrument threshold (P < 5 × 10^–6) and a two-step, nominal P-value–based screening strategy across 41 cytokines, which is consistent with prior two-step MR mediation studies but inevitably increases the multiple-testing burden and the risk of weak mediator instruments. Although standard sensitivity analyses were applied, violation of MR assumptions for the mediator layer (eg, horizontal pleiotropy or residual confounding) cannot be fully excluded, so the cytokine-mediated pathways identified here should be regarded as exploratory and interpreted with caution. Finally, exposure and outcome data primarily derive from European-ancestry GWAS, averting allele frequency biases from other populations but limiting generalizability across ancestries and potentially underestimating true effects of some adiposity axes due to allelic heterogeneity. Cross-ancestry GWAS in diverse populations are warranted.

Conclusion

To encapsulate, this MR study furnishes a comprehensive causal framework linking multidimensional adiposity axes, inflammation, and venous thromboembolism. Two-step mediation analyses reveal that the general obesity axis mediates PE risk via CTACK and VTE risk via beta-NGF and MCP-3. It also uncovers the lower-body fat axis's effects on VTE and DVT, alongside mediation/masking by PDGF-BB and MIG, indicating primary mediation by local hemodynamic imbalances rather than systemic inflammation. Future research should validate these mediators in clinical and experimental settings and explore interventions targeting the inflammation-coagulation interface.

Supplemental Material

sj-docx-1-cat-10.1177_10760296261422488 - Supplemental material for Mapping the Four Adiposity Axes–Inflammatory Cytokine–Venous Thromboembolism Risk Landscape: A Two-Step Mediation Mendelian Randomization Analysis

Supplemental material, sj-docx-1-cat-10.1177_10760296261422488 for Mapping the Four Adiposity Axes–Inflammatory Cytokine–Venous Thromboembolism Risk Landscape: A Two-Step Mediation Mendelian Randomization Analysis by Rongrong Li, Hongping Luo, Ye Tian and Ting Wang in Clinical and Applied Thrombosis/Hemostasis

Footnotes

Acknowledgments

We thank all GWAS participants and researchers for making summary statistics publicly available.

ORCID iD

Ting Wang

Ethical Considerations

This study is a secondary analysis of publicly available data and does not require ethical approval or clinical registration.

Consent to Participate

Not applicable

Funding

The authors received no financial support for the research, authorship, and/or publication of this article.

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Data Availability

All data in this study are from publicly available GWAS data, which can be downloaded for free from https://www.ebi.ac.uk/gwas/ based on the IDs provided in of this paper

Supplemental Material

Supplemental material for this article is available online.

References

Wendelboe

Raskob

. Global burden of thrombosis: Epidemiologic aspects. Circ Res. 2016;118(9):1340-1347.

Khan

Tritschler

Kahn

Rodger

. Venous thromboembolism. Lancet. 2021;398(10294):64-77.

Lutsey

Zakai

. Epidemiology and prevention of venous thromboembolism. Nat Rev Cardiol. 2023;20(4):248-262.

Ortel

Neumann

Ageno

, et al. American Society of hematology 2020 guidelines for management of venous thromboembolism: Treatment of deep vein thrombosis and pulmonary embolism. Blood Adv. 2020;4(19):4693-4738.

Brækkan

Hansen

. VTE epidemiology and challenges for VTE prevention at the population level: VTE epidemiology. Thromb Update. 2023;10(14):100132.

Broni

Ogunmoroti

Quispe

, et al. Adipokines and incident venous thromboembolism: The multi-ethnic study of atherosclerosis. J Thromb Haemost. 2023;21(2):303-310.

La Rosa

Montecucco

Liberale

Sessarego

Carbone

. Venous thrombosis and obesity: From clinical needs to therapeutic challenges. Intern Emerg Med. 2025;20(1):47-64.

Elmaleh-Sachs

Schwartz

Bramante

Nicklas

Gudzune

Jay

. Obesity management in adults: A review. JAMA. 2023;330(20):2000-2015.

Abraham

Yaghootkar

. Identifying obesity subtypes: A review of studies utilising clinical biomarkers and genetic data. Diabet Med. 2023;40(12):e15226.

10.

Katan

. Apolipoprotein E isoforms, serum cholesterol, and cancer. Lancet. 1986;1(8479):507-508.

11.

Lindström

Germain

Crous-Bou

, et al. Assessing the causal relationship between obesity and venous thromboembolism through a Mendelian randomization study. Hum Genet. 2017;136(7):897-902.

12.

Wang

Tan

Hua

Sheng

Huang

Liu

. Genetic predisposition of both waist circumference and hip circumference increased the risk of venous thromboembolism. Thromb Haemost. 2023;123(3):347-361.

13.

Yuan

Bruzelius

Xiong

Håkansson

Åkesson

Larsson

. Overall and abdominal obesity in relation to venous thromboembolism. J Thromb Haemost. 2021;19(2):460-469.

14.

Zhou

Xie

, et al. Causal effects of childhood obesity on venous thromboembolism in adulthood: A bidirectional Mendelian randomization study. Medicine (Baltimore). 2025;104(23):e42591.

15.

Odoemelam

Naz

Thanaj

, et al. Identifying four obesity axes through integrative multiomics and imaging analysis. Diabetes. 2025;74(7):1168-1183.

16.

Skrivankova

Richmond

Woolf

BAR

, et al. Strengthening the reporting of observational studies in epidemiology using Mendelian randomization: The STROBE-MR statement. JAMA. 2021;326(16):1614-1621.

17.

Lawlor

Harbord

Sterne

JAC

Timpson

Davey Smith

. Mendelian randomization: Using genes as instruments for making causal inferences in epidemiology. Stat Med. 2008;27(8):1133-1163.

18.

Zhang

Tang

. The impact of cannabis use on erectile dysfunction and sex hormones: a Mendelian randomization analysis. Int J Impot Res. 2025;37(8):587-594.

19.

Zhang

. Genetic insights into the causal relationship between cannabis use and diabetic phenotypes: A genetic correlation and Mendelian randomization study. Drug Alcohol Depend. 2024;254:111037.

20.

Kurki

Karjalainen

Palta

, et al. Finngen provides genetic insights from a well-phenotyped isolated population. Nature. 2023;613(7944):508-518.

21.

Ahola-Olli

Würtz

Havulinna

, et al. Genome-wide association study identifies 27 loci influencing concentrations of circulating cytokines and growth factors. Am J Hum Genet. 2017;100(1):40-50.

22.

Burgess

Butterworth

Thompson

. Mendelian randomization analysis with multiple genetic variants using summarized data. Genet Epidemiol. 2013;37(7):658-665.

23.

Burgess

Thompson

. Interpreting findings from Mendelian randomization using the MR-egger method. Eur J Epidemiol. 2017;32(5):377-389.

24.

Verbanck

Chen

Neale

. Detection of widespread horizontal pleiotropy in causal relationships inferred from Mendelian randomization between complex traits and diseases. Nat Genet. 2018;50(5):693-698.

25.

Sanderson

. Multivariable Mendelian randomization and mediation. Cold Spring Harbor Perspect Med. 2021;11(2):a038984.

26.

, et al. Visceral adipose tissue demonstrates a stronger association with venous thromboembolism than body mass index. J Thromb Haemostasis. 2025;23(9):2883-2889.

27.

Manolopoulos

Karpe

Frayn

. Gluteofemoral body fat as a determinant of metabolic health. Int J Obes (Lond). 2010;34(6):949-959.

28.

Voicu

Ketfi

Stépanian

, et al. Pathophysiological processes underlying the high prevalence of deep vein thrombosis in critically ill COVID-19 patients. Front Physiol. 2021;11:608788.

29.

Houghton

Ashrani

Liedl

, et al. Reduced calf muscle pump function is a risk factor for venous thromboembolism: A population-based cohort study. Blood. 2021;137(23):3284-3290.

30.

Recek

. Calf pump activity influencing venous hemodynamics in the lower extremity. Int J Angiol. 2013;22(1):23.

31.

Willenberg

Schumacher

Amann-Vesti

, et al. Impact of obesity on venous hemodynamics of the lower limbs. J Vasc Surg. 2010;52(3):664-668.

32.

Ntinopoulou

Papathanasiou

, et al. Obesity as a risk factor for venous thromboembolism recurrence: A systematic review. Medicina (Kaunas). 2022;58(9):1290.

33.

de Athayde Soares

Matielo

Brochado Neto

Nogueira

Almeida

Sacilotto

. Comparison of the recanalization rate and postthrombotic syndrome in patients with deep venous thrombosis treated with rivaroxaban or warfarin. Surgery. 2019;166(6):1076-1083.

34.

Kushner

West

Khan Suheb

Pillarisetty

. Virchow triad. In: StatPearls. StatPearls Publishing; 2025.

35.

Chang

Chen

. CCL7 As a novel inflammatory mediator in cardiovascular disease, diabetes mellitus, and kidney disease. Cardiovasc Diabetol. 2022;21(1):185.

36.

Emanueli

Salis

Pinna

Graiani

Manni

Madeddu

. Nerve growth factor promotes angiogenesis and arteriogenesis in ischemic hindlimbs. Circulation. 2002;106(17):2257-2262.

37.

Liu

Yang

Kong

. Potential causal relationships between blood metabolites, inflammatory cytokines, and venous thromboembolism. Front Immunol. 2024;15:1445790.

38.

Zhou

Guo

Yang

Huang

. Inflammatory cytokines and venous thrombosis bidirectional causal correlations: A Mendelian randomization study. Medicine (Baltimore). 2025;104(22):e42686.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.15 MB