Abstract
Background
Despite large genome‐wide association studies, only ∼30% of the heritability of Alzheimer’s disease is explained. The Alzheimer’s Disease Sequencing Project Whole Exome Sequencing (ADSP WES) has identified millions of genetic variants, over 97% of which are rare (MAF<1%), with 23% appearing in only one person. These rare variants could provide valuable information about new and previously identified risk loci for AD. However, current analysis strategies do not have power to detect associations for such rare variants.
Method
We have developed a protein structure‐based approach that evaluates rare missense variants based on their spatial distribution in a known protein structure rather than on their allele frequency. We hypothesize that AD cases exhibit clustering of rare variants within a protein structure relative to cognitive normal controls. We applied our approach to the ADSP WES Discovery Dataset with 5,522 AD cases and 4,919 controls on 5,969 genes with known structures from the Protein Data Bank (PDB) and 17,450 genes with Alpha Fold2 predicted structures. Only rare variants(MAF<0.05) were included in the analysis. We validated the identified genes within an independent dataset with multi‐ancestry individuals (ADSP WGS Replication) and a European‐ancestry dataset with 15,078 individuals (ADSP validation dataset).
Result
We identified three significant genes (TREM2, SORL1, and EXOC3L4) and one suggestive gene (CSF1R) with AD‐associated spatial clusterings from the ADSP WES data. For TREM2(PDB:6XDS; p‐value = 3.592E‐07) and SORL1(PDB:3WSY;p‐value = 6.701E‐05), two known AD genes, the spatial clusters are significant after excluding known AD risk variants, indicating the presence of additional low‐frequency risk variants within these genes. EXOC3L4(AlphaFold2:Q17RC7;p‐value = 2.504E‐05) is a novel AD risk gene that has a cluster of variants primarily shared by AD cases around the C terminal end of the Sec6 domain. This cluster replicated with significant associations in the ADSP WGS Replication and ADSP validation dataset.
Conclusion
Our result suggests multiple rare missense variants in TREM2, SORL1, EXOC3L4, and likely CSF1R are associated with AD risk by spatial clustering within the protein structure. These spatial clusterings have been replicated in two additional datasets. These spatial patterns may indicate potential functional regions in the protein structure associated with AD risk and are prime targets for further experimental validation.