Population-specific polygenic risk for Alzheimer's disease is associated with Mini-Mental State Examination-based cognitive decline in a Japanese cohort

In a Japanese cohort, a population-specific polygenic risk score derived from Japanese genome-wide association studies demonstrated a stronger and more consistent association with Mini-Mental State Examination-based cognitive decline than European-derived scores, highlighting the importance of ancestry-specific genetic risk assessment for identifying cognitive impairment in community settings.

The Gist

The Big Picture: A Genetic "Weather Forecast" for Memory Loss

Imagine your brain is like a garden. Over time, some people's gardens are more prone to getting overrun by weeds (Alzheimer's disease) than others. Scientists have long known that some of this risk is written in your genetic blueprint (your DNA).

For a long time, scientists tried to create a "risk score" (a Polygenic Risk Score, or PRS) to predict who might get Alzheimer's. Think of this score like a weather forecast. If the forecast says "90% chance of rain," you bring an umbrella. If the score says "high genetic risk," doctors might want to watch that person's memory more closely.

The Problem: Until now, these "weather forecasts" were mostly built using data from people of European descent (like people from the UK or US). It's like using a weather model built for London to predict rain in Tokyo. The climate is different, the wind patterns are different, and the model might not be accurate.

What This Study Did: Checking the Forecast in Japan

The researchers in this paper wanted to see if the "European weather forecast" works for Japanese people. They gathered data from 1,301 older adults living in a community in Japan (Arao city). They didn't just look at people who were already sick; they looked at regular, healthy seniors to see if their genes could predict subtle changes in their memory.

They tested three different "forecasts":

1. Forecast A: Based on old European data.
2. Forecast B: Based on newer, larger European data.
3. Forecast C: Based on Japanese data.

They checked these scores against the MMSE, which is a simple, 30-minute "cognitive check-up" doctors use everywhere (asking questions like "What year is it?" or "Can you count backward from 100 by 7s?").

The Results: The Local Forecast Wins!

Here is what they found, using our garden analogy:

• The European Forecasts (A & B): These were like trying to use a London rain gauge in Tokyo. They didn't match the reality of the Japanese participants very well. They barely predicted who had lower memory scores.
• The Japanese Forecast (C): This was the winner. Because it was built using data from Japanese people, it accurately identified which individuals were more likely to have lower scores on the memory test.

The Key Takeaway: If you want to predict the weather in Japan, you need a model built for Japan. The same goes for genetics. A "one-size-fits-all" genetic risk score doesn't work well across different populations.

The "Smoking Gun" Discovery

The researchers noticed something very interesting when they dug deeper:

• The "Sick" vs. The "Healthy": The genetic risk score was very good at spotting people who were already showing signs of significant memory decline or dementia.
• The "Normal" Range: However, if they took the people with dementia out of the group and looked only at the healthy people, the genetic score stopped being a good predictor.

The Analogy: Imagine a smoke detector. It works great when there is a big fire (dementia). But it might not be sensitive enough to smell a tiny, smoldering ember (very early, subtle memory slips) in a healthy house. This suggests that these genetic scores are best at identifying people who are already on a path toward serious cognitive trouble, rather than catching the very first, tiny whispers of memory loss.

Why This Matters

1. Precision Medicine: We can't just copy-paste medical research from one country to another. To help Japanese (and other non-European) populations, we need to build our own genetic tools using our own data.
2. Simple Tools Work: You don't always need a super-complex brain scan or a 5-hour neuropsychological test to see a link between genes and memory. A simple, 10-minute doctor's office test (MMSE) was enough to see the connection.
3. Stratification: This helps doctors "stratify" or sort patients. If a Japanese senior has a high score on this specific Japanese genetic test, the doctor knows to pay extra attention to their memory, even before they have full-blown dementia.

The Bottom Line

This study is a reminder that genetics is local. Just as a map of New York isn't helpful for driving in Tokyo, a genetic risk map built for Europeans isn't perfect for Japanese people. By building a map specifically for the Japanese population, the researchers found a clearer picture of who is at risk for memory decline, offering a better tool for future prevention and care.

Technical Summary

1. Problem Statement

Alzheimer's Disease (AD) is a leading cause of dementia, with a complex genetic architecture involving numerous small-effect variants. Polygenic Risk Scores (PRS) have been developed to aggregate these genetic risks. However, significant gaps remain in the current literature:

• Population Bias: Most PRS models are derived from European Genome-Wide Association Studies (GWAS) and may not translate effectively to non-European populations due to differences in allele frequencies and Linkage Disequilibrium (LD) structures.
• Clinical Utility: While PRS is known to correlate with cognitive decline, its association with clinically accessible, routine measures (like the Mini-Mental State Examination or MMSE) in general community-dwelling populations (rather than just clinical cohorts of MCI or dementia patients) is not fully established, particularly in Asian populations.
• Gap in Knowledge: There is insufficient evidence regarding whether population-specific PRS (derived from Japanese GWAS) outperforms European-derived PRS in predicting cognitive decline in a Japanese cohort.

2. Methodology

Study Population:

• Cohort: The study utilized the Arao cohort from the Japan Prospective Studies Collaboration for Aging and Dementia (JPSC-AD).
• Participants: 1,301 community-dwelling older adults (aged 65–98) from Arao city, Kumamoto, Japan.
• Exclusions: Participants without genetic consent or missing key data (age, sex, MMSE, PRS, APOE status) were excluded.

Genetic Data Processing:

• Genotyping: Approximately 730,000 SNPs were genotyped using the Illumina Japanese Screening Array.
• Imputation: Performed using Minimac4 with reference panels from BioBank Japan (1,037 samples) and the 1000 Genomes Project (2,504 samples).
• Quality Control (QC): Strict QC was applied at both sample and SNP levels (MAF > 0.01, HWE P > 1.0 × 10⁻⁶, missingness thresholds, LD pruning).
• Ancestry Confirmation: Principal Component Analysis (PCA) confirmed all participants clustered within the East Asian group.
• APOE Genotyping: Targeted sequencing for rs429358 and rs7412 to determine APOE ε4 allele count.

Polygenic Risk Score (PRS) Construction:
Three distinct PRS models were constructed using PRSice-2, excluding the APOE genomic region (chr19:44412079-46412079) to isolate polygenic effects independent of APOE:

1. Jans2019: Based on European GWAS summary statistics (Jansen et al.).
2. Wigh2021: Based on European GWAS summary statistics (Wightman et al.).
3. Shig2021: Based on Japanese GWAS summary statistics (Shigemizu et al.).

Phenotypic and Statistical Analysis:

• Primary Outcome: Mini-Mental State Examination (MMSE) scores. Scores were log-transformed for analysis.
• Secondary Outcome: Hippocampal volume (via 1.5-T MRI).
• Statistical Models: Multiple linear regression analyses were performed, adjusting for age, sex, the first eight genetic principal components (PCs), and APOE ε4 allele count.
• Stratification: Participants were stratified by PRS deciles and extreme percentiles (top/bottom 1%) to assess differences in MMSE scores and dementia prevalence.

3. Key Contributions

• Population-Specific Validation: This is the first study to validate AD PRS against MMSE scores specifically in a large, community-dwelling Japanese cohort.
• Comparative Analysis: It directly compares the predictive power of European-derived PRS versus Japanese-derived PRS within the same target population.
• Clinical Relevance: It demonstrates the utility of PRS in conjunction with a simple, routine clinical tool (MMSE) rather than complex neuropsychological batteries.
• Insight into Severity: It elucidates that PRS associations with cognitive decline are primarily driven by individuals with clinically significant impairment (dementia/MCI) rather than subtle variations in the normal range.

4. Results

PRS Performance:

• Japanese PRS Superiority: The PRS derived from the Japanese GWAS (Shig2021) showed a significant negative correlation with MMSE scores ($R = -0.06, P = 0.02$).
• European PRS Failure: PRS derived from European GWAS (Jans2019 and Wigh2021) showed no significant correlation with MMSE scores ($P > 0.16$).
• Regression Analysis: In multivariate models, the Shig2021 PRS remained a significant predictor of lower MMSE scores. The model including Shig2021 PRS explained 11.3% of the variance in MMSE scores (a 0.4% increase over covariates alone), compared to 11.1% for the Wigh2021 model.

Stratification and Risk:

• Decile Analysis: Participants in the top 10% of the Shig2021 PRS distribution had significantly lower median MMSE scores compared to other groups.
• Extreme Groups: Comparing the top 1% vs. bottom 1% of PRS revealed a significant difference in MMSE scores ($P = 0.02$).
• Dementia Prevalence: The proportion of participants with MMSE $\le$ 23 (indicative of dementia) was 35.7% in the top 1% PRS group, compared to 7.7% in the bottom 1% group (vs. 11.6% overall).

Dementia Exclusion Analysis:

• When participants diagnosed with dementia ($N=48$) were excluded, the association between PRS and MMSE scores disappeared ($P > 0.05$). This suggests the PRS signal is driven by clinically relevant cognitive decline rather than subtle, subclinical variations.

Neuroimaging:

• No significant associations were found between any of the PRS models and hippocampal volume in this cohort.

5. Significance and Conclusion

• Necessity of Population-Specific PRS: The study provides strong evidence that PRS models trained on European data do not generalize well to Japanese populations. Developing and validating PRS using ancestry-matched GWAS data is critical for accurate risk stratification in non-European groups.
• Clinical Stratification Tool: While the effect size is modest, PRS can effectively stratify community-dwelling older adults to identify those at high risk for clinically significant cognitive impairment. This supports the potential use of PRS as a screening adjunct in primary care settings using simple tools like the MMSE.
• Limitations and Future Directions: The study is cross-sectional, precluding causal inference. The lack of association with hippocampal volume suggests that PRS effects on brain structure may require larger samples or longitudinal data to detect. Future work should focus on longitudinal cohorts to determine if PRS predicts the rate of decline and to integrate multi-modal biomarkers.

In summary, this paper establishes that a population-specific polygenic risk score is a superior predictor of cognitive decline in a Japanese cohort compared to European-derived scores, highlighting the importance of genetic ancestry in precision medicine for Alzheimer's disease.