Untargeted metabolomics reveals plasma metabolites predictive of ectopic fat in pancreas and liver as assessed by magnetic resonance imaging: the TOFI_Asia study

Obesity is a known risk factor for cardiometabolic disease, but individuals within the same BMI category show heterogeneity in metabolic risk. South and East Asians develop type 2 diabetes (T2D) at lower BMI and younger age, potentially due in part to greater propensity for visceral and ectopic fat deposition. Visceral and ectopic fat are linked to insulin resistance, dyslipidaemia, metabolic syndrome, T2D, and cardiovascular disease independent of BMI. Accurate quantification of ectopic fat typically requires advanced imaging or biopsy, which are costly or invasive, and circulating biomarkers for early detection of ectopic fat in liver and pancreas are lacking. Recent position statements highlight the importance of identifying such biomarkers. Advances in metabolomics combined with machine learning may enable detection of plasma markers reflecting VAT and organ fat burden and underlying metabolic perturbations. Prior metabolomic studies have identified markers for NAFLD progression, but early asymptomatic liver fat markers and especially circulating markers of pancreatic fat remain largely unidentified. The present cross-sectional study in Caucasian and Chinese women (TOFI_Asia) aimed to identify plasma metabolite markers predictive of pancreatic fat, liver fat, and VAT/SAT ratio measured by MRI/MRS, and to compare their predictive performance to clinical and anthropometric measures.

Systemic metabolomic profiling in NAFLD has identified candidate plasma markers including taurocholate, glutamyl dipeptides, mannose, lactate, carnitine and acylcarnitines, free fatty acids, lysophosphatidylcholines, and glycerolipids associated with NAFLD progression. However, most were discovered in diagnosed NAFLD cohorts, leaving a gap for biomarkers of early, asymptomatic liver fat accumulation. Notably, no circulating biomarkers of pancreatic fat have been established; a prior targeted metabolomics study (Jaghutriz et al.) in prediabetic European Caucasians with impaired glucose tolerance did not find metabolites distinguishing high vs low pancreatic fat. These gaps motivate untargeted approaches to discover markers of visceral and pancreatic fat.

Design and participants: Cross-sectional analysis within the TOFI_Asia study. Sixty-eight female participants (34 Chinese, 34 Caucasian), aged 20–70 years, BMI 20–45 kg/m², fasting plasma glucose ≤ 6.9 mmol/L, with both parents of the same ethnicity, were included. Exclusions: >10% weight change in prior 3 months, bariatric surgery, glucose-related medications, known disease including T2D, pregnancy, breastfeeding. Clinical measures: Height, weight, waist, hip, SBP, DBP were recorded. Fasting plasma glucose (hexokinase), HbA1c (capillary electrophoresis), liver function tests, and lipid profile were measured by standard methods. Fasting glucoregulatory peptides (insulin, C-peptide, glucagon, amylin, GIP, total GLP-1) were measured using a multiplex bead-based assay. Body composition and imaging: Total body fat (TBF) and total body lean mass (TBL) were obtained by DXA; %TBF = TBF*100/(TBL+TBF). Abdominal fat depots VAT and SAT, pancreatic fat, and liver fat were quantified by MRI/MRS on a 3T Siemens Skyra. A 2-point Dixon technique acquired three blocks of forty 5-mm axial slices during 11-s breath-holds for fat–water separation; a fat fraction (FF) map at L4–L5 was constructed and segmented (ImageJ) into VAT and SAT to compute VAT/SAT ratio. Pancreatic fat was estimated using an MR-opsy approach (mean of two candidate pancreas FF maps across head, body, tail ROIs). Liver MRS used a 2×2×2 cm³ voxel in the right lobe avoiding vessels/biliary tree; spectra acquired across planes with and without water suppression; liver fat expressed as percent fat-to-water volume using AUC of peaks (SIVIC). Due to artifacts, 65 pancreatic fat, 67 liver fat, and 68 VAT/SAT measures were analyzed. Metabolomics: Plasma metabolites were extracted via biphasic extraction; aqueous and organic phases analyzed on two LC-MS platforms. Raw data were converted to mzXML (ProteoWizard MSconvert). Preprocessing included normalization (LOESS in W4M Galaxy), QC-based feature filtering (%CV<30 in QCs), and annotation; details in SI. Statistical analysis: Outliers were removed using PLS residual analysis; normality of response residuals assessed with Kurtosis test. Multivariate PLS regression with MUVR (R v3.5.1) was applied to the full metabolome (polar + lipids) to select variables associated with pancreatic fat, liver fat, or VAT/SAT (continuous outcomes). To avoid overfitting, 100 permutation tests with repeated double cross-validation (rdCV) were performed on PLS models using either the full metabolome or post-selection variables. Redundant chemometric features (isotopes, adducts) were removed post-selection. Final PLS models (SIMCA 16) used an optimal number of components minimizing RMSEcv; performance summarized by R2Y, Q2, and Pearson r between predicted vs measured values. Clinical correlates for each fat depot were identified by linear regression adjusted for ethnicity and combined into panels for PLS modeling. For individual metabolites, linear regression with Benjamini–Hochberg correction assessed associations in models adjusted sequentially for: ethnicity (M1), total adiposity (BMI and %TBF; M2), VAT/SAT for pancreatic and liver fat (M3), and age (M4).

- PLS variable selection yielded 56 metabolites for pancreatic fat (91% identified), 64 for liver fat (95% identified), and 31 for VAT/SAT (100% identified). - Model performance improved after selection versus full metabolome: • Pancreatic fat: post-selection R2Y=0.81, Q2=0.69, r=0.90 (vs full R2Y=0.39, Q2=0.25, r=0.63). • Liver fat: post-selection R2Y=0.80, Q2=0.66, r=0.89 (vs full R2Y=0.52, Q2=0.33, r=0.72). • VAT/SAT: post-selection R2Y=0.70, Q2=0.62, r=0.83 (vs full R2Y=0.47, Q2=0.30, r=0.69). - Metabolite classes associated in PLS: • Pancreatic fat: sulfolithocholic acid, CE(20:3), FA(16:1), glucose (LC-MS), urea, phosphorylcholine, kynurenic acid, 5 amino acids, and numerous lipids (26 glycerolipids, 9 glycerophospholipids, 3 sphingolipids). • Liver fat: homocitrulline, lactate, LacCer(d34:1), dhSM(d36:0), 5 fatty acids, 47 glycerolipids, 5 glycerophospholipids. Among 27 MS2-annotated TGs, 75% contained palmitic acid (C16:0) and 63% contained oleic acid (C18:1); TGs were highly saturated (≤3 double bonds). • VAT/SAT: L-cystine, Cer(d41:1), PC(O-38:6), 3 fatty acids, 25 glycerolipids. Among 11 MS2-annotated TGs, 82% contained linoleic acid (C18:2) and 73% contained oleic acid (C18:1). - Clinical-marker-only PLS models performed worse than metabolite panels: • Pancreatic fat (clinical panel: FPG, HbA1c, BMI, %TBF, age, SBP, DBP, TC, TG, LDL-C): R2Y=0.51, Q2=0.46. • Liver fat (clinical panel: BMI, %TBF, age, SBP, DBP, ALT, ALP, GGT, TC, TG): R2Y=0.48, Q2=0.40. • VAT/SAT (clinical panel: FPG, HbA1c, BMI, age, SBP, DBP, GGT, TC, TG, LDL-C): R2Y=0.56, Q2=0.44. • Combining clinical and metabolite markers gave similar performance to metabolite markers alone. - Individual metabolite associations (BH-corrected): • Pancreatic fat: 44/56 independent of ethnicity. After adjusting for total adiposity, significant lipids (PC, DG, TG), CE(20:3), methionyl-methionine, and sulfolithocholic acid remained; after further VAT/SAT adjustment only sulfolithocholic acid remained; after additional age adjustment, no metabolites remained significant. • Liver fat: 54/64 independent of ethnicity and total adiposity. After VAT/SAT adjustment, lactic acid, PC(16:0/18:2), TG(16:0/18:1/18:1), and TG(58:2) became non-significant; none of the remaining associations were affected by age. • VAT/SAT: 27/31 independent of ethnicity; adjusting for total adiposity preserved most associations except FA(18:0) and TG(53:5); after age adjustment, 15 metabolites (3 DGs and 12 TGs) remained significant. - Overall, metabolite markers outperformed clinical markers in predicting ectopic and visceral fat levels.

This study demonstrates that untargeted plasma metabolomics combined with multivariate modeling can robustly estimate pancreatic fat, liver fat, and VAT/SAT ratio, outperforming panels of conventional clinical and anthropometric markers. The lipidomic signature of liver fat was characterized by predominantly saturated and monounsaturated triacylglycerols enriched in palmitic and oleic acid residues, consistent with lipogenesis and altered lipid handling in hepatic steatosis. VAT/SAT-associated triacylglycerols were enriched in linoleic and oleic acids, suggesting depot-specific lipid profiles. Importantly, numerous metabolites were associated with liver fat and VAT/SAT independent of overall adiposity and age, indicating potential utility as early indicators beyond traditional risk markers. In contrast, pancreatic fat showed limited independent associations: sulfolithocholic acid remained associated after adjustment for adiposity and VAT/SAT but not after accounting for age, suggesting that circulating markers of pancreatic steatosis may be subtler or more age-dependent. The findings support the value of metabolomics for non-invasive assessment of ectopic and visceral fat burden and highlight specific metabolite classes that could inform mechanisms of metabolic risk.

Untargeted LC-MS metabolomics identified plasma metabolites that predict ectopic organ fat and visceral fat distribution more accurately than conventional clinical measures in a cohort of Caucasian and Chinese women. Liver fat and VAT/SAT were associated with distinct sets of glycerolipids, glycerophospholipids, and sphingolipids independent of adiposity and age, whereas pancreatic fat showed a weaker independent plasma signature, with sulfolithocholic acid the only marker persisting after adiposity adjustment but not age. These results suggest potential for developing blood-based assays to estimate ectopic fat burden and to improve risk stratification prior to overt metabolic disease. Further studies are required to validate these markers in larger, diverse cohorts, to assess longitudinal predictive value for disease progression, and to elucidate biological mechanisms linking these metabolites to depot-specific fat accumulation.