Lipoprotein particles that raise the odds of heart failure lower the odds of hip and knee osteoarthritis. Not in two different studies — in one genetic analysis, across the same molecules, in the same paper.
The result sits at the end of a post-hoc analysis in PLOS Medicine that set out to answer a narrower question: does intentional weight loss leave a molecular signature, and does the route matter? The authors pooled blood profiling from two UK trials with little in common except the kilos they remove — DiRECT, a total diet replacement programme in primary care, and By-Band-Sleeve, a surgical trial of gastric bypass, sleeve gastrectomy and adjustable banding.
Two trials, one laboratory plate
The cohorts were not matched. DiRECT participants lost a mean 11.1 kg by 12 months against 1.4 kg in controls — a BMI reduction of 3.7 kg/m², and enough for 46% of the intervention group to reach diabetes remission in the original 2018 report. By-Band-Sleeve randomised 1,351 people with a mean BMI of 46.4; in the metabolomic subset, weight fell by 29.3 kg over a median 36 months, a BMI reduction of 10.5 kg/m².
Three times the weight loss, and roughly the same molecular response. Of 1,202 metabolites measured after surgery, 784 — 65% — changed at a false-discovery threshold. After the diet, 490 of 1,152 did, or 43%. Of the 932 molecules measured in both studies, 645 moved in the same direction, and 242 did so with statistical support in both.
The correlation between the two interventions’ effects was moderate — r = 0.54 on mass spectrometry, 0.63 on NMR — so surgery is not simply a larger dose of dieting.
A genetic cross-check, and three exceptions
Consistency across two interventions is suggestive, not proof, that weight itself is doing the work. A third line of evidence followed: mendelian randomisation against lifetime BMI, using genetic variants as a proxy for lifelong exposure.
Of the 242 shared metabolites, 233 had usable genetic instruments. At nominal significance, 165 showed an association with genetically predicted BMI, and 162 of those — 98% — pointed the way the weight-loss trials had. Only three contradicted: dimethylarginine, tyramine O-sulfate, and the caffeine metabolite 5-acetylamino-6-amino-3-methyluracil.
The authors call this gene–environment equivalence, and it is the strongest part of the paper: three designs with different failure modes agreeing on the same 157 to 176 molecules.
Where the signature splits
Then the analysis turns. The authors carried those metabolites into two-sample mendelian randomisation against three conditions obesity raises the risk of in quite different ways.
For all-cause heart failure, 30 metabolite traits reached significance, in a pattern that fits an ischaemic case mix: LDL and VLDL measures raised risk, HDL cholesterol enrichment lowered it, and two ratios marking dysfunctional HDL remodelling — phospholipid-to-total-lipid and triglyceride-to-total-lipid in medium HDL — went the wrong way.
For hip and knee osteoarthritis, 34 traits reached significance. The same lipoprotein family appeared, with the sign flipped: higher LDL and VLDL measures were associated with lower odds. Glycine, gamma-glutamylglycine and cinnamoylglycine ran the other direction, raising risk.
For kidney cancer, the signature all but vanished. Two metabolites survived out of 242. Higher 5-methyluridine carried an odds ratio of 1.12 per standard deviation (95% CI 1.06–1.19), instrumented by three variants; ursodeoxycholate looked protective but rested on a single variant, so no sensitivity analysis was possible. The outcome datasets differ enormously in size — 25,890 kidney cancer cases against 213,839 for osteoarthritis and 139,533 for heart failure — and the authors say sparsity may reflect power rather than biology.
Why thirty associations are fewer than thirty
The paper also walks through an arithmetic trap. Genetic instruments for metabolites overlap heavily, because metabolites are correlated and share loci. Of 211 metabolites with genome-wide instruments, the authors found 22,155 unique pairs, of which 3,303 shared at least one variant and 1,230 shared a quarter or more of their variants. Clustering on that overlap produced 17 multi-member clusters and 64 singletons; every LDL trait associated with heart failure fell into one cluster.
The most parsimonious reading, in the authors’ words, is “one signal detected multiple times” rather than thirty distinct mechanisms.
What the paper does not do is measure mediation: it cannot say what fraction of BMI’s effect on any disease travels through these molecules, and its mendelian randomisation is offered as evidence of pathway involvement, not of a causal metabolite. Add the post-hoc design, the single-arm handling of the surgical cohort, the rank-normalised scale that frustrates physiological interpretation, and the European-ancestry inputs, and the honest summary is: a well-triangulated map of where to look next.
Where that map leads is a trial question. The authors suggest the signature could serve as an intermediate biomarker in early-phase drug studies, a way to check a new agent is moving the biology and not just the scale. The obvious missing arm is incretins. The LABS study, funded by NIDDK, was built to follow outcomes over years rather than molecules over months, and nothing in a metabolomic panel substitutes for counting events — a point that has come up before in this archive’s look at cancer risk after metabolic surgery.
Weight loss is not one thing. It is one signature pointing at several diseases, and the diseases do not read it the same way.
This archive is independent of NIH and NIDDK and reports published research only; nothing here is medical advice.
Sources
- Sunderland N, Smith ML, McConnachie A, et al. A metabolomic signature of weight-loss and its associations with BMI-linked diseases: a post-hoc analysis of two intervention trials and a mendelian randomisation study. PLOS Medicine, 4 November 2026.
- By-Band-Sleeve Collaborative Group. Roux-en-Y gastric bypass, adjustable gastric banding, or sleeve gastrectomy for severe obesity (By-Band-Sleeve): a multicentre, open label, three-group, randomised controlled trial. Lancet Diabetes & Endocrinology, May 2025.
- Lean ME, Leslie WS, Barnes AC, et al. Primary care-led weight management for remission of type 2 diabetes (DiRECT): an open-label, cluster-randomised trial. The Lancet, 10 February 2018.
- NIDDK. Bariatric Surgery — health information for the public.