Investigating Multi-Omic Signatures of Ethnicity and Dysglycaemia in Asian Chinese and European Caucasian Adults: Cross-Sectional Analysis of the TOFI_Asia Study at 4-Year Follow-Up

Abstract

Background: Type 2 diabetes (T2D) is a global health epidemic with rising prevalence within Asian populations, particularly amongst individuals with high visceral adiposity and ectopic organ fat, the so-called Thin-Outside, Fat-Inside phenotype. Metabolomic and microbiome shifts may herald T2D onset, presenting potential biomarkers and mechanistic insight into metabolic dysregulation. However, multi-omics datasets across ethnicities remain limited. Methods: We performed cross-sectional multi-omics analyses on 171 adults (99 Asian Chinese, 72 European Caucasian) from the New Zealand-based TOFI_Asia cohort at 4-years follow-up. Paired plasma and faecal samples were analysed using untargeted metabolomic profiling (polar/lipid fractions) and shotgun metagenomic sequencing, respectively. Sparse multi-block partial least squares regression and discriminant analysis (DIABLO) unveiled signatures associated with ethnicity, glycaemic status, and sex. Results: Ethnicity-based DIABLO modelling achieved a balanced error rate of 0.22, correctly classifying 76.54% of test samples. Polar metabolites had the highest discriminatory power (AUC = 0.96), with trigonelline enriched in European Caucasians and carnitine in Asian Chinese. Lipid profiles highlighted ethnicity-specific signatures: Asian Chinese showed enrichment of polyunsaturated triglycerides (TG.16:0_18:2_22:6, TG.18:1_18:2_22:6) and ether-linked phospholipids, while European Caucasians exhibited higher levels of saturated species (TG.16:0_16:0_14:1, TG.15:0_15:0_17:1). The bacteria Bifidobacterium pseudocatenulatum, Erysipelatoclostridium ramosum, and Enterocloster bolteae characterised Asian Chinese participants, while Oscillibacter sp. and Clostridium innocuum characterised European Caucasians. Cross-omic correlations highlighted negative correlations of Phocaeicola vulgatus with amino acids (r = -0.84 to -0.76), while E. ramosum and C. innocuum positively correlated with long-chain triglycerides (r = 0.55-0.62). Conclusions: Ethnicity drove robust multi-omic differentiation, revealing distinctive metabolic and microbial profiles potentially underlying the differential T2D risk between Asian Chinese and European Caucasians.

Keywords: BCAA; Phocaeicola vulgatus; ethnicity; glycaemic status; gut microbiome; host–microbiome interactions; multi-omics; plasma metabolomics; shotgun metagenomics; triglycerides; type 2 diabetes.

Figure 1

Participant flowchart for the faecal metagenomics and plasma metabolomics analyses.

Figure 2

Influence of ethnicity. (A) DIABLO diagnostic plots showing multi-omics data integration according to ethnicity, with the most substantial discrimination between Asian Chinese and European Caucasian cohorts based on latent components from all the datasets. The upper right of the figure contains scatter plots, coloured by group types, with ellipses representing 95% confidence. Values to the lower left represent Pearson correlation coefficients between the first components from each dataset. (B) Diagnostic plots visualising samples projected on the latent components, showing weak discrimination by each block (data type). Colours distinguish samples from Asian Chinese (orange) and European Caucasian (blue) cohorts.

Figure 3

Multivariate analysis of ethnicity using DIABLO. Loading plots represent the top discriminating features for each dataset: (A) bacterial species (OTU), (B) KEGG pathways, (C) lipids, and (D) polar metabolites. Features are sorted according to discriminatory strength; bar colour signifies that a feature’s maximal median value is associated with either Asian Chinese (orange) or European Caucasian (blue).

Figure 4

Top cross-omic correlations from the ethnicity DIABLO model. The correlation matrix was derived from the supervised DIABLO model discriminating Asian Chinese vs. European Caucasian participants along components 1–4 and consists of four block pairs: Lipid–OTU (A), Lipid–KEGG (B), Polar–OUT (C), and Polar–KEGG (D). The top 100 feature pairs with the largest correlation are displayed as dots, sized by correlation value magnitude and coloured from blue to orange for negative-to-positive values. Full feature details and correlation values are in Supplementary Tables S6–S9.

Figure 5

Discrimination by sex using sparse partial least squares discriminant analysis (sPLS-DA) on polar metabolites. (A) Sample plots showing the separation of females (green) and males (purple), across components 1 vs. 2 and 2 vs. 3, based on latent variables derived from the polar metabolite dataset. (B) Loading plots highlight the top discriminatory polar metabolites contributing to each component, with colour indicating the sex with the higher median abundance.