Skip to content

Polygenic and Gut Metagenomic Risk Integration

Summary

Liu, Ritchie, Teo, et al. (2024) present a prospective multiomic risk integration study leveraging the FINRISK 2002 cohort ($N=5,676$ with linked genome-wide arrays and baseline shallow shotgun gut metagenomic sequencing, median 17.8 years follow-up via national electronic health records) [1]. Polygenic risk scores (PRSs) consistently and significantly improved incident disease prediction over conventional clinical risk factors across coronary artery disease (CAD), type 2 diabetes (T2D), Alzheimer disease (AD), and prostate cancer [1]. Standalone gut metagenomic scores (GMS) significantly predicted incident disease for all four conditions, but after adjusting for conventional clinical risk factors, gut microbiome predictive value remained independently significant for T2D ($\text{HR } 1.20, p = 9.13 \times 10^{-6}$) and prostate cancer ($\text{HR } 1.23, p = 0.020$) only, functioning as a complementary dynamic risk layer orthogonal to static germline genetics [1].

Study Cohort & Multimodal Protocol

                        FINRISK 2002 Survey Cohort (N = 8,783)
                                           │
                                           │ QC Filter (Remove low reads, pregnancy,
                                           │ extreme BMI, recent antibiotic use)
                                           ▼
                       Multiomic Subcohort (N = 5,676 Participants)
                       - Linked Genome-Wide Imputed Genotypes
                       - Shallow Shotgun Gut Metagenomic Sequencing
                       - Median 17.8-Year National EHR Follow-Up
                                           │
       ┌──────────────────┬────────────────┴──────────────────┬──────────────────┐
       ▼                  ▼                                   ▼                  ▼
  Incident CAD       Incident T2D                        Incident AD      Prostate Cancer
   (N = 333)          (N = 579)                           (N = 273)          (N = 141) [1]
  • Genomic Polygenic Risk Scores: Downloaded from the Polygenic Score Catalog (PGS Catalog): CAD (PGS000018, 6.3M variants), T2D (PGS000036, 136k variants), AD (PGS000334, 21 variants), and Prostate Cancer (PGS000662, 269 variants) [1].
  • Gut Metagenomic Risk Scores (GMS): Generated by shallow shotgun sequencing of stool samples, filtering to 235 species-level taxonomic groups after removing low-prevalence taxa ($< 1\%$ abundance) [1]. Ridge logistic regression classifiers ($10 \times 3$-fold stratified cross-validation) computed continuous disease-specific risk scores [1].
  • Clinical Endpoints: Incident cases derived from linked national registers (Hospital Discharge Register, Causes of Death Register, Drug Reimbursement Register) [1].

Standalone Predictor Performance & Independence from Family History

In sex-stratified Cox proportional hazards models adjusting for baseline age, PRSs demonstrated powerful, independent hazard ratios per standard deviation across all four endpoints [1]:

Endpoint Baseline Primary Clinical Predictor ($C$-statistic) Standalone PRS Hazard Ratio (per SD) Standalone PRS $p$-value Standalone PRS $C$-statistic PRS Incremental Gain ($\Delta C$ over Clinical Factors)
CAD Baseline Age ($C = 0.719$) $\text{HR } 1.68$ ($95\%\text{ CI } 1.50\text{--}1.88$) $2.25 \times 10^{-19}$ $C = 0.626$ $+0.023$ ($95\%\text{ CI } 0.013\text{--}0.034$) [1]
T2D Body Mass Index ($C = 0.745$) $\text{HR } 1.42$ ($95\%\text{ CI } 1.30\text{--}1.55$) $6.48 \times 10^{-15}$ $C = 0.612$ $+0.010$ ($95\%\text{ CI } 0.004\text{--}0.016$) [1]
AD Baseline Age ($C = 0.880$) $\text{HR } 1.92$ ($95\%\text{ CI } 1.73\text{--}2.15$) $4.27 \times 10^{-32}$ $C = 0.650$ $+0.017$ ($95\%\text{ CI } 0.010\text{--}0.024$) [1]
Prostate Cancer Baseline Age ($C = 0.769$) $\text{HR } 1.73$ ($95\%\text{ CI } 1.47\text{--}2.04$) $5.50 \times 10^{-11}$ $C = 0.641$ $+0.027$ ($95\%\text{ CI } 0.009\text{--}0.047$) [1]
  • Independence from Family History: For CAD, T2D, and prostate cancer, PRSs and reported family history were mutually independent in joint Cox models ($p < 10^{-5}$ for both), proving that genomic risk scores capture unshared, additive risk unaccounted for by pedigree [1].
  • Alzheimer Disease $\ge 60$ Subanalysis: In older adults ($\ge 60$ years), the AD PRS achieved a standalone $C$-statistic of $0.667$, outperforming any single clinical factor and the full clinical model combined [1]. Adding the PRS increased the C-statistic by $\Delta C = +0.064$ (reaching total $C = 0.722$), with the PRS HR ($1.87$ per SD, $p = 8.95 \times 10^{-23}$) exceeding that of baseline age ($1.73$ per SD, $p = 4.50 \times 10^{-15}$) [1].
  • T2D NMR Glucose Subanalysis: When incorporating nuclear magnetic resonance (NMR)-determined fasting glucose as an additional clinical risk factor, BMI retained the highest individual $C$-statistic ($0.743$), while PRS ($C = 0.612$) and glucose ($C = 0.656$) displayed comparable independent hazard ratios ($\text{HR } 1.40, p = 1.85 \times 10^{-12}$ vs $\text{HR } 1.38, p = 5.95 \times 10^{-19}$) [1].

Gut Metagenomic Diversity & Risk Score Attenuation

Microbial Alpha and Beta Diversity

  • Alpha-Diversity: Lower Shannon index and Chao-Shannon diversity significantly predicted incident T2D ($\text{HR } 0.89$ per SD, $p = 0.004$; $\text{HR } 0.90$ per SD, $p = 0.014$), confirming that gut dysbiosis precedes clinical diabetes onset [1]. Species richness was positively associated with incident prostate cancer ($\text{HR } 1.23, p = 4.20 \times 10^{-4}$) [1].
  • Beta-Diversity: Principal component analysis of Aitchison distances revealed PC2 was strongly associated with incident T2D ($\text{HR } 0.94, p = 1.31 \times 10^{-5}$) and PC5 ($\text{HR } 1.04, p = 0.030$) [1]. Bray-Curtis dissimilarity principal coordinate analysis linked PCoA1 ($\text{HR } 1.78, p = 0.024$) and PCoA5 ($\text{HR } 3.26, p = 0.005$) to incident T2D [1].
  Unadjusted Metagenomic Risk Score
  - CAD:  HR 1.28 (p < 0.001)
  - T2D:  HR 1.40 (p < 0.001)
  - AD:   HR 1.34 (p < 0.001)
  - PCa:  HR 1.50 (p < 0.001)
                │
                │ Adjust for Clinical Factors (Age, BMI, Blood Pressure, Lipids)
                ▼
  Multivariable Clinical Adjustment
  - CAD:  HR 1.05 (p = 0.42)  --> Attenuated to Non-Significance
  - T2D:  HR 1.20 (p = 9.1e-6) --> REMAINS SIGNIFICANT [1]
  - AD:   HR 1.08 (p = 0.28)  --> Attenuated to Non-Significance
  - PCa:  HR 1.23 (p = 0.020) --> REMAINS SIGNIFICANT [1]

Attenuation Pattern Under Clinical Adjustment

While standalone species-level microbiome scores predicted all four diseases in unadjusted models ($\text{AUROC } 0.564\text{--}0.613$), adjusting for conventional clinical risk factors resulted in complete attenuation for CAD ($\text{HR } 1.05, p = 0.42$) and AD ($\text{HR } 1.08, p = 0.28$) [1]. Microbiome risk maintained independent prospective value strictly for T2D ($\text{HR } 1.20, p = 9.13 \times 10^{-6}$) and prostate cancer ($\text{HR } 1.23, p = 0.020$) [1].

Orthogonality & Lack of Gene-Microbiome Interaction

Statistical interaction tests between GMS and PRS revealed no significant multiplicative interaction terms ($p > 0.05$ across CAD, T2D, AD, and prostate cancer) [1]. Germline polygenic predisposition and gut metagenomic risk operate as mutually orthogonal, additive risk vectors [1].

Synthesis: The attenuation of gut metagenomic scores for CAD and AD under clinical adjustment indicates that gut dysbiosis signals in these conditions are largely mediated by, or correlated with, upstream cardiometabolic risk factors (e.g. SBP, lipids, systemic inflammation). In contrast, the retention of independent predictive value for T2D ($\text{HR } 1.20$) and prostate cancer ($\text{HR } 1.23$) indicates that gut microbial metabolites (e.g. short-chain fatty acids, secondary bile acids) provide uncaptured metabolic and immune-modulatory signals that bypass standard clinical laboratory measurements.

Integrated Multiomic Model Performance

Full Cox proportional hazards models incorporating Baseline Age + Sex + Conventional Risk Factors + PRS + Microbiome Risk Score achieved the absolute highest predictive discrimination across all traits studied [1]:

Disease Model Baseline Clinical Model ($C$-statistic) Integrated Model ($C$-statistic) Net Improvement ($\Delta C$-statistic vs Clinical)
Coronary Artery Disease $0.748$ ($95\%\text{ CI } 0.725\text{--}0.771$) $0.772$ ($95\%\text{ CI } 0.749\text{--}0.795$) $+0.024$ ($95\%\text{ CI } 0.013\text{--}0.035$) [1]
Type 2 Diabetes $0.791$ ($95\%\text{ CI } 0.774\text{--}0.808$) $0.805$ ($95\%\text{ CI } 0.789\text{--}0.821$) $+0.014$ ($95\%\text{ CI } 0.007\text{--}0.021$) [1]
Alzheimer Disease $0.884$ ($95\%\text{ CI } 0.868\text{--}0.900$) $0.901$ ($95\%\text{ CI } 0.887\text{--}0.915$) $+0.017$ ($95\%\text{ CI } 0.009\text{--}0.024$) [1]
Prostate Cancer $0.769$ ($95\%\text{ CI } 0.739\text{--}0.798$) $0.800$ ($95\%\text{ CI } 0.769\text{--}0.831$) $+0.031$ ($95\%\text{ CI } 0.011\text{--}0.050$) [1]

Significance

This study establishes the clinical hierarchy of multiomic profiling in population screening: germline polygenic risk scores provide robust, static, lifetime risk stratification across cardiovascular, metabolic, neurodegenerative, and oncological domains [1]. Shotgun metagenomics provides complementary, dynamic physiological risk information for metabolic (T2D) and specific oncological (prostate cancer) conditions, though its incremental discrimination gain over standard clinical markers is modest [1].

See Also

Citations

[1] Liu, Y., Ritchie, S.C., Teo, S.M., Ruuskanen, M.O., Kambur, O., Zhu, Q., Sanders, J., Vázquez-Baeza, Y., Verspoor, K., Jousilahti, P., Lahti, L., Niiranen, T., Salomaa, V., Havulinna, A.S., Knight, R., Méric, G., & Inouye, M. (2024). Integration of polygenic and gut metagenomic risk prediction for common diseases. Nature Aging, 4, 584--594. DOI: 10.1038/s43587-024-00590-7. Source: s43587-024-00590-7.pdf. Supports: FINRISK 2002 cohort size (N=5,676, median 17.8 yrs follow-up), incident case counts (333 CAD, 579 T2D, 273 AD, 141 prostate cancer), PRS HRs per SD, AD age >=60 subanalysis (PRS HR 1.87 vs age 1.73), gut microbiome attenuation under clinical adjustment (significant for T2D/prostate only), Aitchison/Bray-Curtis beta-diversity metrics, GMS x PRS interaction testing, and integrated C-statistics. Location: Full text -- Abstract, Results, Table 1-2, Figures 1-4. Verified 2026-07-30.