Fig. 3: Disease-dependent and disease-independent of APOE ε4 on the human proteome.

a, Volcano plot shows the protein association profile of APOE ε4 after adjusting for AD dementia diagnosis, with red representing significant associations (after FDR correction). At the y axis, the −log₁₀(FDR-adjusted P values) > 300 were set to 300 for better visualization. This was done for S100A13, TBCA, NEFL, LRRN1 and SPC25. b,c, Box plots show plasma protein level changes of the proteins with the strongest APOE ε4 associations (b) and for APOE ε4-associated proteins strongly tied to AD dementia diagnosis (c). For b and c, the y axis represents residual protein levels after adjusting for age, sex, mean protein level and contribution site. The center line of each box indicates the median, with lower and upper edges representing the 25th and 75th percentiles. Whiskers extend to the most extreme values within 1.5 times the interquartile range; data points beyond this range were excluded as outliers. The x axis represents AD diagnosis. The color indicates APOE ε4 carrier status; ‘−/−’ indicates APOE ε4 non-carriers; ‘±’ indicates ε3/ε4; and ‘+/+’ indicates ε4/ε4. Welch’s t-test was used to compare residual protein levels between groups. Two-sided P values are reported. ****P < 0.0001 and *P < 0.05. P values were not adjusted for multiple comparisons, as only prespecified group contrasts are shown. Results marked with **** remain significant (pFDR < 0.0001) even after adjustment for multiple comparisons with the Benjamini–Hochberg method, whereas those marked with * do not. d, Receiver operating characteristic area under the curve (ROC-AUC) showing the performance of a machine learning model using only five proteins to predict APOE ε4 status across different diagnostic groups, in a held-out sample. e, Protein interaction network including four of those five proteins (red). f, Neural cell type expression of RNA transcripts encoding the five APOE ε4-predictive proteins. Plot shows mix-max scaling of protein-coding transcripts per million for each identified APOE ε4 protein. g, Correlation of effect sizes for proteins associated with APOE ε4 in cognitively unimpaired samples (x axis) and AD associated with AD diagnosis in APOE ε3/ε3 homozygotes. Limma t-statistic is shown for both contrasts; only proteins associated with both APOE ε4 and AD (adjusted P < 0.05 for both analyses) with the same direction of effect are visualized. For visibility purposes, t-statistic values higher than 10 were capped. PDD, Parkinsonʼs disease dementia; pFDR, FDR-corrected P value.