Discussion

In this large-scale, prospective analysis of the UK Biobank cohort, we have demonstrated that metabolic vulnerability, as comprehensively assessed through MVX and its component indices (IVX and MMX), represents a significant and independent risk factor for incident hypertension. These findings extend beyond conventional clinical risk factors and provide novel insights into the metabolic underpinnings of hypertension pathogenesis, with important implications for risk stratification and preventive strategies.

The robust association between MVX and hypertension risk underscores the complex interplay between multidimensional metabolic dysregulation and blood pressure homeostasis. Mechanistically, this relationship likely reflects multiple pathophysiological pathways. Elevated branched-chain amino acids (BCAA) – leucine, isoleucine and valine – integrated within the MMX component, have been implicated in insulin resistance through activation of the mammalian target of rapamycin signalling pathway^{(Reference Lynch and Adams13)}. Chronic mammalian target of rapamycin activation may subsequently promote endothelial dysfunction and vascular remodelling through increased oxidative stress and impaired nitric oxide bioavailability^{(Reference Peng, Fu and Wu14)}. Concurrently, reduced citrate levels may reflect mitochondrial dysfunction and disrupted energy metabolism in vascular tissues, potentially compromising the ability of vascular smooth muscle cells to maintain appropriate vasomotor tone^{(Reference Yap, Wanga and Wüst15)}.

The inflammatory component of metabolic vulnerability, captured by IVX, likely contributes to hypertension through distinct yet complementary mechanisms. GlycA, a composite nuclear magnetic resonance spectroscopy signal reflecting acute-phase glycoproteins, serves as a stable marker of systemic inflammation and has been previously associated with cardiometabolic disorders^{(Reference Fung, Chan and Ng16)}. Mechanistically, chronic low-grade inflammation may promote hypertension through endothelial activation, increased vascular permeability and enhanced reactivity to vasoconstrictors^{(Reference Caillon and Schiffrin17)}. Furthermore, the inverse relationship between small HDL particles and hypertension risk identified in our study aligns with evidence suggesting that specific HDL subclasses possess vasoprotective properties, including promotion of endothelial nitric oxide synthase activity and reduction of oxidative stress^{(Reference Holmes, Millwood and Kartsonaki5)}.

Particularly noteworthy was our observation of effect modification by obesity status, with stronger associations between metabolic indices and hypertension among non-obese individuals. This seemingly paradoxical finding might be explained by the concept of ‘metabolically unhealthy normal weight’ phenotype, wherein normal-weight individuals with multidimensional metabolic dysregulation may harbour disproportionate visceral adiposity and ectopic fat deposition, contributing to heightened cardiometabolic risk despite normal BMI^{(Reference Stefan, Schick and Häring18)}. Additionally, in obese individuals, the strong influence of excess adiposity on hypertension risk might partially obscure the relative contribution of more subtle metabolic perturbations captured by MVX^{(Reference Leggio, Lombardi and Caldarone19)}.

The amplified association between MVX and hypertension risk among those with elevated polygenic risk scores (PRS) suggests a gene–environment interaction that warrants further investigation. Genetic variants associated with hypertension susceptibility may sensitise individuals to the adverse effects of multidimensional metabolic dysregulation through shared pathways involving renin–angiotensin–aldosterone system function, renal Na handling and vascular tone regulation^{(Reference Padmanabhan and Dominiczak20)}. This synergistic interaction underscores the potential utility of integrating genetic and metabolomic data for personalised risk assessment.

Our findings also revealed a dose-response relationship between MVX quartiles and hypertension risk, with hazard ratios increasing monotonically across quartiles. The persistence of these associations after comprehensive adjustment for traditional risk factors indicates that metabolic vulnerability provides additional associative information beyond established clinical parameters. Furthermore, the stronger associations observed for isolated systolic hypertension compared to diastolic hypertension align with the understanding that arterial stiffness, which predominantly affects systolic pressure, may be particularly sensitive to metabolic perturbations and inflammatory processes^{(Reference Scuteri, Cunha and Agabiti Rosei21)}.

The stratified analyses revealing stronger associations among younger individuals suggest that metabolic vulnerability may represent an early marker of vascular dysfunction that precedes clinically detectable hypertension. This finding has important implications for life course approaches to hypertension prevention, highlighting the potential benefit of early metabolic health assessment and intervention^{(Reference Pavlou, Paschou and Anagnostis22)}.

From a clinical perspective, our results suggest that MVX could enhance risk stratification for hypertension beyond traditional factors. The integration of metabolomic profiling into clinical risk assessment might identify individuals who would benefit from more intensive lifestyle interventions or earlier pharmacological therapy. Moreover, the distinct components of MVX could inform targeted interventions addressing specific metabolic or inflammatory pathways^{(Reference Zhang, Wei and Wang23)}.

Our findings align with recent metabolomic studies that have identified distinct metabolic signatures associated with cardiovascular risk. For instance, Wang et al. reported that plasma metabolite profiles predict incident cardiovascular events independent of traditional risk factors^{(Reference García-Gavilán, Babio and Toledo24)}. Similarly, Tzoulaki et al. demonstrated that metabolic profiles provide complementary information to conventional risk factors for CVD prediction^{(Reference Tzoulaki, Castagné and Boulangé25)}. Compared with previous studies examining BCAA in isolation, prospective cohort and Mendelian randomisation studies have consistently demonstrated a positive causal association between elevated circulating BCAA levels and increased hypertension risk. These studies uniformly report a monotonically linear dose-response relationship^{(Reference Flores-Guerrero, Groothof and Connelly26,Reference Cai, Fu and Chen27)} . The added value of this study’s composite metric lies in MMX’s simultaneous integration of two opposing metabolic signals—excess BCAA (elevating blood pressure via mammalian target of rapamycin-mediated insulin resistance) and reduced citrate (reflecting impaired mitochondrial function). Their antagonistic relationship generates a U–shaped nonlinear risk curve structurally undetectable by single-BCAA analysis. This aligns strongly with evidence that composite amino acid markers integrating opposing metabolic axes (e.g. tBCAA/glycine ratio) outperform single BCAA markers^{(Reference Dziedzic, Józefczuk and Guzik28)}. IVX similarly demonstrates synergistic benefits by simultaneously integrating pro-inflammatory (GlycA) and vascular protective (small-particle HDL) axes. Regarding modeling approaches, staged Cox regression (Model 1→3) transparently quantified the attenuation of effect sizes through progressive covariate adjustment. Restricted cubic spline (RCS) analysis was the critical methodological choice for revealing the nonlinear association between MMX and hypertension (P-nonlinear = 0·002)—a relationship that would have been entirely obscured using traditional linear models alone. The non–linear relationship observed between MMX and hypertension risk in our study suggests complex interactions between energy metabolism pathways and blood pressure regulation, consistent with findings from Menni et al. who identified hexadecanedioate and a panel of lipid and amino acid metabolites as plasma signatures of blood pressure regulation in multiple independent cohorts, demonstrating that specific metabolite levels correlate with both systolic and diastolic blood pressure in a pathway-specific manner. The present study extends this framework by demonstrating that a composite index integrating amino acid and energy metabolism signals (MMX) captures a non-linear hypertension risk gradient not apparent in any single-metabolite analysis^{(Reference Menni, Graham and Kastenmüller29)}.

The stronger association between MVX and hypertension in non-obese individuals parallels recent findings from Neeland et al., who demonstrated that ectopic fat distribution, rather than overall adiposity, contributes significantly to cardiometabolic risk^{(Reference Neeland, Ross and Després30)}. This emphasises the importance of metabolic phenotyping beyond traditional anthropometric measures. Furthermore, the interaction between genetic risk and metabolic vulnerability observed in our study is consistent with emerging evidence suggesting that genetic susceptibility can modify the relationship between metabolic factors and cardiovascular outcomes, as reported by Said et al. ^{(Reference Said, Verweij and van der Harst31)}.

Several limitations merit consideration. First, despite comprehensive adjustment for confounders, residual confounding cannot be entirely ruled out. Second, only approximately 30 % of the original UK Biobank participants were included in the final analytical cohort, primarily due to the requirement for complete NMR metabolomics data necessary to calculate MVX. Participants excluded were more likely to have incomplete follow-up or missing covariate data, which may introduce selection bias. A comparison of key baseline characteristics between included and excluded participants is provided in online Supplementary Table 2. Furthermore, the cohort is predominantly of European ancestry and displays a well-documented healthy volunteer bias, limiting generalisability to ethnically diverse and lower socio-economic populations^{(Reference Fry, Littlejohns and Sudlow32)}. Third, metabolites were measured at a single time point, precluding assessment of temporal changes in metabolic profiles. Fourth, since baseline blood pressure was not included as a covariate in the Cox model, we cannot entirely rule out the possibility that participants with subclinical hypertension at baseline may have already exhibited altered metabolic characteristics. This could represent a potential source of residual confounding. Future studies will consider incorporating baseline blood pressure as a covariate. Certain covariates included in model 3 – specifically BMI, C-reactive protein and diabetes status – may lie on the causal pathway between metabolic susceptibility and incident hypertension (i.e. they may act as mediators rather than mere confounders). Therefore, the risk ratios in model 3 should be interpreted as reflecting the direct association between metabolic indicators and hypertension risk. Finally, while our study establishes robust associations, causal inference requires complementary approaches, including Mendelian randomisation studies and interventional trials^{(Reference Holmes, Ala-Korpela and Smith33)}.

Future research should validate these findings in diverse populations and investigate whether interventions targeting specific metabolic pathways reflected in MVX can modify hypertension risk. Longitudinal assessment of metabolic profiles could elucidate the temporal dynamics between metabolic changes and blood pressure progression. Additionally, integration of MVX with other emerging biomarkers, including proteomics and microbiome profiles, may further improve risk stratification^{(Reference Astarita, Kelly and Lasky-Su34,Reference Zhang, Yu and Qi35)} .