The Health for Life in Singapore (HELIOS) Study: delivering precision medicine research for Asian populations

Figures

• 

  Fig. 1. Overview of participant recruitment, data and biospecimen collection, and linkage. DEXA dual energy X-Ray absorptiometry, ECG electrocardiogram, OCT optical coherence tomography, OCTA optical coherence tomography angiography. This figure was created in BioRender https://BioRender.com/gtyffu3.

• 

  Fig. 2. Environment exposure across three populations. a The Singapore population density across planning area overlayed with HELIOS study participant postcode; b The distribution of diabetes cases; c The distribution of food amenities, including hawker centres, shopping malls, restaurants, and convenience stores; and d the ratio of public/ private residence, overlayed with public parks, connectors, and nature reserves. The built environment has impact health outcomes, as illustrated by the correlations of diabetes case proportions and e the density of food amenities/ area, and f the public/private residence ratio, weighted by population numbers.

• 

  Fig. 3. Lifestyle factors across three populations. a Longitudinal changes in macronutrient trends nationally from 1998 to inception of HELIOS study in 2022 (n = 10,004). Ethnic variations in b macronutrients (proportion to total energy intake, n = 8119), c diet quality represented by a modified DASH score (range from 0 for low quality to 35 for high quality, data points represent mean DASH score; error bars represent standard errors, n = 8119) and d top 20 FFQ foods (servings/day) significantly different across ethnicities within HELIOS study. e Physical activity (data points represent mean total MET-minutes per week; error bars represent standard errors, n = 8119) and f Accelerometer-based physical activity according to the levels of self-reported physical activity (R = 0.23, two-sided p = 1.7 × 10-11 from Kruskal-Wallis rank sum test). g The proportion of people who meet the WHO guideline of physical activity by ethnicity and the proportion of physical activity across different domains (n = 7853). h The relationships between lifestyle factors and cardiometabolic phenotypes are heterogeneous across ethnic groups. Data points represent the coefficients derived from linear regression…

• 

  Fig. 4. Genomic variation in Asian Populations. a Number of variants annotated in each functional coding mutation category, by ancestry. The darker shades indicate unique variants observed only in the specific ancestry. b 2-dimensional PCA genomic variants by ancestry group. c Distribution of PRS scores for six complex traits in the three major ancestry groups. Comparison across the three ancestries was done using two-sided ANOVA test. d Forest plot displaying association PRS scores with hypertension (n = 8237), obesity (n = 8269), type 2 diabetes (n = 8244), coronary artery disease (n = 8261), depression (n = 8361) and osteoporosis (n = 8261) by ancestry group, and overall [C: Chinese, M: Malay, I: Indian]. Data points represent odds ratios; error bars represent 95% confidence intervals. Two-sided P-value was calculated using logistic regression. ANOVA Analysis of Variance, PCA Principal Component Analysis, PRS Polygenic Risk Scores.

• 

  Fig. 5. PCA plots of DNA methylation levels at 16,444 CpG sites that are highly stratified between our Chinese, Indian and Malay Asians, and ‘exposomic’ assessment of the associations of these stratified markers with measured and genetically inferred exposures. a–c Display the methylation diversity for individuals from the three ethnic groups (PC1 on x axis, and PC2 on y axis). Overlaid are the effect sizes of the measured and genetically inferred exposures that are associated with PC1 and PC2 of the DNA methylation (two-sided P < 0.0015 multiple testing correction for 32 independent tests; all exposures are Z-standardised to enable comparison of effect sizes); a clinical traits; b dietary exposures assessed objectively by circulating metabolites. Specifically, fermented rice/lentil items refer to idli and thosai. c Polygenic Risk Scores (PRS). The beta weights for PC1 and PC2 are scaled along the top x-axis and the right y-axis respectively. The results identify the directly measured and genetically inferred exposures that may relate to population level epigenetic variation between Asian ethnic groups. d The effect sizes and the directions for the beta coefficients derived from…

• 

  Fig. 6. Metabolic variation across three populations. a PCA plot of 153 significantly differentiated metabolites across three age-sex matched ethnic cohorts of 1146 individuals each (The first two PCs explain 23% of variation). Selection criteria for 153 metabolites include: 1) significantly associated with ethnicity in the discovery cohort (70% participants, two-sided linear regression analysis P < 1 × 10-5 after multiple-test correction), and 2) significantly associated with ethnicity in the test cohort (30% participants, two-sided test P < 0.05 and same direction of estimates as in the discovery cohort). b Circos plot of 128 well-characterized and known metabolites (in sequence from outermost to innermost layer): 1) metabolite super-pathways, 2) significant associations with HT, Obs, T2D, and CVD, denoted by a black dot, 3) estimates of regression coefficient for association with ethnicities (CI: Indian compared to Chinese, CM: Malay compared to Chinese, IM: Malay compared to Indian). Curved lines at the centre highlight significant pairwise correlation between metabolites. Grey lines represent pairwise correlations within the same super-pathway; blue lines represent pairwise…

• 

  Fig. 7. Transcriptomic variation across three Asian populations. Volcano plot of differential gene expression across the three Asian populations in a pairwise comparison between Chinese and Indians (a), Chinese and Malays (b), and Indians and Malays (c) after adjusting for Age, Sex, RIN and top six PEER factors. The x-axis is the effect size estimate of association per standard deviation change in gene expression and the y-axis is the two-sided log10P value of the association from the regression model. d Gene set enrichment analysis of the 2972 unique differentially expressed genes in at least one pairwise comparison across populations. Enrichment was performed using gprofiler2 tool in three pathway databases (Kyoto Encyclopedia of Genes and Genomes (KEGG), Reactome (REAC) and Wiki Pathways (WP). Each dot is an independent gene set and the y-axis is the one-sided hypergeometric P-value of enrichment per gene set. e Regional plots of three cis-eQTLs identified in the HELIOS cis-eQTL analyses in 1228 individuals. The first two plots are the top two strongest associations (NBPF8 and TMEM126A) and the last one is for the TXNIP gene identified as an important gene linked to diabetes via DNA…

• 

  Fig. 8. Genetic Architecture of Molecular Traits. a Manhattan plot for summary of all associations between plasma metabolite levels and genetic loci (N = 8234). Only genetic variants with P < 5 × 10-8 (two-sided test in linear mixed regression model) are coloured based on the strongest associated metabolite group for the specific variant. Genes for top 50 loci identified through SMR are annotated. b Distribution of number of associated metabolites per locus, demonstrating the pleiotropy of genetic effects on metabolites. The loci with at least 5 associated metabolites are annotated with the SMR associated gene. c Regional plot highlighting the shared causal variant between Dopamine 3-O-sulfate and SMAD5. d Regional plot highlighting the shared causal variant between X-11381 and NPHP4. The y-axis in the regional plots represents the –log₁₀ of the two-sided P-value from the GWAS analysis. GWAS Genome-Wide Association Study, SMR Summary-data-based Mendelian Randomisation.

• 

  Fig. 9. Infographics of national health data linkage. CI confidence interval, HDL High-density lipoproteins, NHD national health data, OR odds ratio, WHR Waist hip ratio.

• 

  Fig. 10. Reproducibility of measurements between baseline and repeated measures in HELIOS. Spearman correlation analyses were used to calculate the correlation coefficients between z-scored baseline and retest measurements (N = 398). Correlations are presented as box plots, where the upper bound marks the 75th percentile, and the lower bound marks the 25th percentile. BP blood pressure, DEXA dual-energy X-ray absorptiometry, ECG electrocardiogram, FFQ food frequency questionnaire, HR heart rate, IPAQ International Physical Activity Questionnaire.