Gut disorders forecast Alzheimer’s and Parkinson’s risk years before diagnosis

By Vijay Kumar MalesuReviewed by Susha Cheriyedath, M.Sc.Aug 28 2025

Common, treatable gut-related conditions signal Alzheimer’s and Parkinson’s disease risk long before symptoms, paving the way for earlier, personalized brain health strategies.

Gut-brain nexus: Mapping multimodal links to neurodegeneration at biobank scale. Image Credit: Inkoly / Shutterstock

In a recent study published in the journal Science Advances, a group of researchers mapped temporal, genetic, proteomic, and clinical links between gut-brain axis disorders and the risk and classification of Alzheimer’s disease (AD) and Parkinson’s disease (PD) across population biobanks.

Background

AD and PD cumulatively affect more than 400 million people worldwide, straining families, workplaces, and health systems. The gut-brain axis is a bidirectional network that links the gastrointestinal tract and the central nervous system through hormones, metabolism, and immunity.

Signals traverse neural, cytokine, and endocrine pathways, so disturbances in digestion, nutritional status, or glucose regulation can reverberate in the brain. Because common conditions, such as diabetes mellitus, vitamin D deficiency, and functional bowel disorders, are widespread and treatable, they may be useful for risk stratification and early detection, while causal effects require further study.  

About the study

The study leveraged three population resources, such as the United Kingdom Biobank (UKB), Secure Anonymised Information Linkage (SAIL), and FinnGen. Investigators curated 155 diagnoses from the International Classification of Diseases, 10th Revision (ICD-10), spanning endocrine, nutritional, metabolic, and digestive disorders. Cox proportional hazards models estimated hazard ratios (HRs) and 95% confidence intervals (CIs) for subsequent AD or PD, with multiple testing controlled using the false discovery rate (FDR) by the Benjamini-Hochberg method. To test whether timing mattered, models were repeated in three prediagnostic windows: 1–5, 5–10, and 10–15 years before diagnosis.

Fine-Gray subdistribution hazards were used to account for the competing risk of death. These analyses preserved effect directions for most codes, with some attenuation, and Kaplan–Meier curves showed higher cumulative incidence in individuals with significant codes.

Polygenic risk scores (PRS) were computed from genome-wide association study (GWAS) summary statistics, with AD models adjusted for apolipoprotein E (APOE) and PD models adjusted for leucine-rich repeat kinase 2 (LRRK2) and glucosylceramidase beta 1 (GBA1) status. Proteomic specificity was examined using the Olink platform in the UKB, spanning 1,463 proteins in 52,705 participants. 

Finally, generalized linear models (GLMs) integrated clinical, genetic, and proteomic features to classify cases and controls, developed and evaluated in the UK Biobank. Replication was applied only to the epidemiologic associations across SAIL and FinnGen, as proteomics and the classifier were UKB-only. An interactive Streamlit tool was also provided to explore results.

Study results

Across cohorts, several endocrine, nutritional, metabolic, and digestive disorders predicted later AD. Replicated ICD-10 codes with HR above 1 included insulin-dependent diabetes mellitus (E10), noninsulin-dependent diabetes mellitus (E11), unspecified diabetes mellitus (E14), disorders of lipoprotein metabolism (E78), vitamin D deficiency (E55), other disorders of electrolyte, fluid, and acid-base balance (E87), other functional intestinal disorders (K59), and gastrointestinal inflammation codes such as other noninfective gastroenteritis and colitis (K52), gastritis and duodenitis (K29), esophagitis (K20), and other bacterial intestinal infections (A04).

Hemorrhoids and perianal venous thrombosis (K64) showed HR below 1. For PD, replicated risks included dyspepsia (K30), E10, E11, and K59, while diverticular disease (K57), other diseases of the intestine (K63), and other disorders of the peritoneum (K66) were associated with lower risk. In total, 14 ICD-10 codes replicated for AD and 7 for PD across cohorts.

Timing effects were pronounced. In AD, E11 and E14 conferred a greater risk when recorded 10 to 15 years before diagnosis, whereas E10 was elevated in every window. E55 and E87 were associated with both near diagnosis and across the full observation period. In PD, E14 was strongly associated at 1–5, 5–10, and 10–15 years; E10 peaked 5 to 10 years prior to diagnosis; deficiency of other B group vitamins (E53) was most predictive 1 to 5 years prior to diagnosis; and K30 conferred risk throughout. UKB Kaplan–Meier curves showed lower probabilities of remaining free of AD or PD in individuals with significant codes such as E10, E11, and E14.

PRS analyses revealed a lower average genetic burden in cases with co-occurring disorders compared to isolated cases for both diseases, after adjusting for APOE in AD and LRRK2 and GBA1 in PD. No synergistic interaction between PRS and ICD-10 diagnoses was detected; the interaction odds ratio (OR) did not exceed 1 at the nominal significance level.

Proteomics using the Olink platform identified 22 biomarkers that differed in AD versus controls and 156 that differed in PD. 37 proteins were higher in AD when specific co-occurring codes were present, and five were higher in PD, indicating that gut-brain axis comorbidities shape plasma signatures. GLM that combined diagnoses, PRS, and proteomics outperformed single-modality models, supporting multimodal classification in UKB only.

Conclusions

This biobank-scale mapping of gut-brain axis conditions reveals that common, treatable disorders can forecast AD and PD years in advance, with a timing that can inform prevention strategies. Stronger or earlier links for diabetes, vitamin D deficiency, electrolyte imbalance, and functional bowel disorders point to screening and risk modification opportunities in primary care.

Lower PRS in comorbid cases and distinct proteomic profiles suggest overlapping yet partly environmental pathways to neurodegeneration. Multimodal models that fuse diagnoses, genetics, and proteomics add practical accuracy for classification and clinical triage. 

Together, these insights support earlier, more personalized brain health strategies. Limitations include the proteomics being restricted to the UKB, reliance on ICD-10 diagnostic codes, differences across cohorts, and the restriction to European ancestry, which constrains generalizability.