Proteomic signatures of the APOE ε 4 and APOE ε 2 genetic variants and Alzheimer's disease

Abstract

The ε4 and ε2 alleles of the Apolipoprotein E (APOE) gene confer opposite genetic risks for Alzheimer's disease (AD), but their underlying molecular mechanisms remain poorly characterized in humans. To resolve this, we systematically profiled APOE-associated proteomic alterations across five cohorts-including the Global Neurodegeneration Proteomics Consortium (GNPC), BioFINDER-2, the Alzheimer's Disease Neuroimaging Initiative (ADNI), the Parkinson's Progression Markers Initiative (PPMI), and UK Biobank (UKB)-using SomaLogic and OLINK platforms in plasma and cerebrospinal fluid (CSF) from over 10,000 individuals. Using GNPC (plasma SomaLogic, N=4,045), we mapped a comprehensive APOE-protein network and applied mediation modeling to classify genotype-related signals as upstream mediators, downstream consequences, or APOE-specific changes. We then leveraged CSF beta-amyloid (Aβ) biomarker data from BioFINDER-2 (plasma SomaLogic, N=1,421) to improve temporal resolution and isolate early, Aβ-independent proteomic programs. In the Aβ- individuals, APOE4 was linked to cell cycle and chromatin remodeling, while APOE2 was associated with mitochondrial regulation and DNA repair. Mediation analyses nominated proteins such as S100A13, TBCA, SPC25 for APOE4, and APOB, SNAP23 for APOE2 as candidate upstream effectors, supported by CSF validation (ADNI, SomaLogic, N=666), brain transcriptomic co-expression, and AD GWAS colocalization. Longitudinal CSF data from PPMI confirmed the temporal stability of several APOE-associated proteins. Cross-platform comparisons (UKB plasma OLINK, N=4,820, and BF2 CSF OLINK, N=1,475) revealed matrix- and assay-specific heterogeneity, underscoring challenges in reproducibility. Together, our results delineate allele-specific, temporally structured proteomic signatures that precede AD pathology, offering insight into APOE-driven molecular pathways and potential therapeutic targets for early intervention.

Fig. 1:. Study design

Overview of datasets: The Global Neurodegeneration Proteomics Consortium (GNPC) cohort (N=4,045, plasma, SomaLogic 7k) was used to identify proteins associated with APOE4 or APOE2 and clinical diagnosis (CU vs. AD dementia). To investigate APOE-related proteomic changes in relation to Aβ status (Aβ− vs. Aβ+), we utilized plasma data from BioFINDER-2 (SomaLogic 7k, a subset of GNPC), CSF data from Alzheimer’s Disease Neuroimaging Initiative (ADNI) (SomaLogic 7k), and CSF data from BioFINDER-2 (OLINK). The UK BioBank (UKB) (plasma, OLINK) was used exclusively to define APOE-driven protein signatures, as clinical diagnosis and Aβ status were unavailable in the dataset. Parkinson’s Progression Markers Initiative (PPMI) cohort (CSF, OLINK) is utilized to investigate the longitudinal APOE-protein associations. Differential abundance analysis identified APOE (APOE4 or APOE2) associated proteins and AD diagnosis (or Aβ status) associated proteins, resulting in different groups of proteins specifically associated with APOE or AD, or jointly associated with both. Proteins associated with APOE were further tested and categorized based on its upstreaming or downstreaming role in AD using 2 mediation models: APOE => protein => AD diagnosis or Aβ status (Upstream mediation model) and APOE => AD diagnosis or Aβ status => protein (Downstream mediation model). Diagnosis or Aβ status stratification analysis were conducted to investigate in-depth the changes of APOE-protein associations. An age stratification analysis only in CU or Aβ− to investigate how early APOE-protein association changes with age. APOE4 and APOE2 analysis were conducted separately and the results were compared together with the inclusion of ε2ε4 carriers to investigate protein level change in detailed APOE genotypes (6 genotypes: ε2ε2, ε2ε3, ε2ε4, ε3ε3, ε3ε4 and ε4ε4). APOE associated proteins were annotated using cell type enrichment analysis, Gene Ontology (GO) term enrichment analysis or biologically informed neural networks (BINNs)-enriched Reactome pathway analysis, protein-protein interaction (PPI) analysis and spatial co-expression analysis (APOE and associated genes). Those are done based on the GNPC cohort as it has a larger sample size than other cohorts. For the BioFINDER-2 and ADNI cohort, BINN-enriched reactome analysis was used to globally validate the pathways or processes found in the GNPC cohort. In the BioFINDER-2 cohort, we evaluated the association between APOE-associated proteins and available 5 AD biomarkers or cognitive function: tau-PET, Aβ-PET, atrophy: cortical thickness in the temporal lobe, cognitive performance measurements Mini-Mental State Examination (MMSE) and modified Preclinical Alzheimer Cognitive Composite (mPACC) scores. A separate cohort (PPMI) was used to evaluate longitudinal protein-APOE associations. Figure created with BioRender.com.

Fig. 2:. APOE4-associated proteins in GNPC (plasma Somalogic)

a. Volcano plot for proteins associated with APOE4 without adjusting for AD diagnosis, with red representing significant association after FDR correction. At y-axis, -log10(FDR) above 300 was set to 300 for a better visualization. b. Volcano plot shows proteins associated with AD diagnosis without adjusting for APOE4, with red representing significant association after FDR correction. c. UpSet plot shows the number of proteins associated with APOE4 or AD with or without adjusting for each other, blue indicating AD-specific associated proteins, red indicates APOE4-specific associated proteins, black indicates the number of proteins independently associated with both. d. Heatmap shows the mediation effects and significance of APOE4-associated proteins that are involved in the 2 mediation paths. Color indicates the strength of the mediation effect, the darker the color, the bigger the mediation proportion at each path. Proteins at x-axis are categorized and labeled according to their significance in 2 paths (protein-mediated pathway: APOE4 => protein => AD diagnosis, AD-mediated pathway: APOE4 => AD diagnosis => protein), with those exhibiting significance in both paths further classified by the strength of the mediation effect. Proteins were marked in red if their protein-mediated pathway was significant alone (after FDR correction) or if the protein-mediated pathway exhibited a stronger mediating effect (when indirect effects in both mediation paths are significant after FDR correction). Proteins were marked in blue if their AD-mediated pathway was significant alone (after FDR correction) or if the AD-mediated pathway exhibited a stronger mediating effect (when indirect effects in both mediation paths are significant after FDR correction). Additionally, bold font indicates a total mediation effect and non-bold font indicates a partial mediation effect in the preferred (same color) path. For heatmap annotations, *represents nominal significant, *+ represents significant after FDR correction. e. Scatter plot shows APOE4’s effect size on protein without adjusting AD diagnosis in whole cohort (x-axis) vs. in CU individuals (y-axis) for each protein. The Spearman correlation coefficient (R = 0.622, p < 1e-16) indicates a moderate positive correlation between the effect sizes. Red represents proteins associated with APOE4 in both the whole cohort and in the CU group. f. The scatter plot shows the effect of APOE4 on proteins in CU (x-axis) vs. the effect of AD on proteins in ε3ε3 carriers (y-axis); Only proteins associated with APOE4 in CU individuals and with AD diagnosis in ε3ε3 carriers were visualized. Red indicates the same effect direction while blue indicates an opposite effect direction. g. BINN-enriched reactome pathway analysis for proteins associated with APOE4 in both the whole cohort and in CU. The darker the dot, the more important the protein and the pathway in the deep learning model predicting AD dementia diagnosis. More features are hidden in the sink for a better visualization. h. Venn plot shows the number of proteins in each category. Note that 78 proteins were specifically associated proteins in panel c, 7 proteins partially mediated the effect of APOE4 on AD diagnosis, and were therefore subdivided into mediator-proteins group; 2 proteins were partially mediated by AD pathology, and were therefore reassigned into AD-mediated proteins. i. The linear discriminant score plot shows the projection score of all tested proteins in the discriminant direction. Proteins were colored by assigned groups. j. The integrative matrix summarizes differential regulation (red for upregulated and blue for downregulated proteins in APOE4 carriers), cell-type enrichment based on scaled RNA expression, and functional characterization of each protein. Cell types from the ROSMAP atlas are labeled in black on the x-axis, while those from the Human Brain Vascular atlas are labeled in red. Gray boxes indicate nominal significance (p<0.05), and black boxes indicate FDR-corrected significance (pFDR<0.05) in cell-type enrichment analysis. Gene Ontology (GO) biological process terms associated with each protein were grouped into broader representative categories; small red boxes indicate the involvement of a given protein in the corresponding process. Protein–protein interactions (PPIs) are annotated using STRING database interactions with a confidence score≥0.7. The number of interactions per protein is shown as a heatmap, and direct interactions between proteins are represented by lines, color-coded according to their assigned cluster.

Fig. 3:. APOE2 plasma protein signature in GNPC

a. Volcano plot for proteins associated with APOE2 without adjusting for AD diagnosis, with red representing significant association after FDR correction. At y-axis, -log10(FDR) above 300 was set to 300 for a better visualization. b. Scatter plot shows APOE2’s effect size on protein without adjusting AD diagnosis in whole cohort (x-axis) vs. in CU individuals (y-axis) for each protein. The Spearman correlation coefficient (R = 0.852, p < 1e-16) indicates a strong positive correlation between the effect sizes. Proteins highlighted in red are significantly associated with APOE2 in both the whole cohort and the CU subgroup. c. Volcano plot for the effect of APOE2*age on proteins that are associated with APOE2 in the CU group with red indicates a significant APOE2*age effect on proteins. d. Interaction effect of APOE2 status and age on protein levels in the CU group. Protein levels of VPS29 and AKT2_1 (“_1” represent one aptamers of AKT2) are plotted against age, with separate regression lines for individuals carrying the ε2 allele (ε2+, red) and those without it (ε2−, blue). Shaded areas represent 95% confidence intervals. e. Heatmap shows the mediation effects and significance of APOE2-associated proteins that are involved in the 2 mediation paths. Color indicates the strength of the mediation effect, the darker the color, the bigger the mediation proportion at each path. Proteins at x-axis are categorized and colored according to their significance in 2 paths (protein-mediated pathway: APOE2 => protein => AD diagnosis, AD-mediated pathway: APOE2 => AD diagnosis => protein), with those exhibiting significance in both paths further classified by the strength of the mediation effect. Proteins were marked in red if their protein-mediated pathway was significant alone (after FDR correction) or if the protein-mediated pathway exhibited a stronger mediating effect (when indirect effects of both mediation paths are significant after FDR correction). Proteins were marked in blue if their AD-mediated pathway was significant alone (after FDR correction) or if the AD-mediated pathway exhibited a stronger mediating effect (when indirect effects of both mediation paths are significant after FDR correction). Additionally, bold font indicates a total mediation effect and non-bold font indicates a partial mediation effect in the preferred (same color) path. For heatmap annotations, *represents nominal significant, *+ represents significant after FDR correction. f. BINN-enriched pathway analysis for early dysregulated proteins (associated with APOE2 in both whole cohort and in CU) in APOE2 carriers, the darker the color, the more important the protein or pathway in predicting AD dementia diagnosis. More features are hidden in the sink for a better visualization. g. Subdivision of APOE2 associated proteins based on the association between proteins, APOE2 and AD. Noting that one protein (EMC2) overlaps between APOE2-specific and mediator-proteins and was only categorized into mediator proteins for grouped GO biological process enrichment analysis. h. The linear discriminant score plot shows the projection score of each group of subdivided proteins in the discriminant direction. Proteins were colored by assigned groups. i. The integrative matrix summarizes differential regulation (red for upregulated and blue for downregulated proteins in APOE2 carriers), cell-type enrichment based on scaled RNA expression, and functional characterization of each protein. Cell types from the ROSMAP atlas are labeled in black on the x-axis, while those from the Human Brain Vascular atlas are labeled in red. Gray boxes indicate nominal significance (p<0.05), and black boxes indicate FDR-corrected significance (pFDR<0.05) in cell-type enrichment analysis. Gene Ontology (GO) biological process terms associated with each protein were grouped into broader representative categories; small red boxes indicate the involvement of a given protein in the corresponding process. Protein–protein interactions (PPIs) are annotated using STRING database interactions with a confidence score≥0.7. The number of interactions per protein is shown as a heatmap, and direct interactions between proteins are represented by lines, color-coded according to their assigned cluster.

Fig. 4:. Different patterns of APOE4− or APOE2-associated proteins

a. Venn plot shows proteins mediating APOE4’s effect of AD diagnosis and their association with APOE2, proteins mediating APOE2’s effect on AD diagnosis and their association with APOE4. b. Boxplots show protein level comparison between different APOE genotype groups, for ε4 dominated mediators (SPC25 and LRRN1) and ε2 dominated mediators (UNG and PCLAF). b. c. d. On the box plots, the y-axis represents residual protein levels after adjusting for age, sex and mean protein level. The center line of each box indicates the median (50th percentile); the lower and upper edges of the box represent the 25th and 75th percentiles, respectively. Whiskers extend to the most extreme values within 1.5 times the interquartile range (IQR); data points beyond this range are considered outliers and have been excluded from the plot display. The x-axis represents APOE genotype, ε2ε3 carriers were merged into the “22/23” group due to a small sample size of ε2ε2 carriers. Welch’s t-test was used to compare residual protein levels between groups. Two-sided p-values are tested and p-values were adjusted for multiple comparisons using Holm-Bonferroni method. Each pair of groups was compared, only pairs with significant differences after FDR correction were shown. Asterisks indicate significance levels: **** for pFDR < 0.0001, *** for pFDR < 0.001, ** for pFDR < 0.01, * for pFDR < 0.05. c. Heatmap shows non-AD mediated (non-downstreaming) proteins that are associated with both APOE4 and APOE2, color indicate the effect size of APOE4 or APOE2 in each model on those proteins, column 1 indicates APOE4’s effect on those proteins without AD diagnosis adjustment, column 2 indicates APOE4’s effect on those proteins with AD diagnosis adjustment, column 3 indicates APOE4’s effect on those proteins in CU individuals, column 4 indicates the effect of APOE4*age on those proteins in CU individuals.The remaining 4 columns correspond to APOE2’s effect: column 5 indicates APOE2’s effect on those proteins without AD diagnosis adjustment, column 6 indicates APOE2’s effect on those proteins with AD diagnosis adjustment, column 7 indicates APOE2’s effect on those proteins in CU individuals, column 8 indicates the effect of APOE2*age on those proteins in CU individuals. “☆” indicate the opposite effect direction of APOE4 and APOE2, red marked proteins are AD mediator proteins in APOE4 => protein => AD diagnosis, blue marked proteins are AD mediator proteins in APOE2 => protein => AD diagnosis, purple marked proteins are AD mediator proteins in both. Only proteins for which the average of the absolute value of the effect of APOE4 and APOE2 were greater than 0.1 are shown. For heatmap annotations, *represents nominal significant, *+ represents significant after FDR correction. d. Boxplots show protein level comparison between different APOE genotype groups for non-downstreaming proteins that are opposingly regulated by both alleles. e. Boxplots show protein level comparison between different APOE genotype groups for non-downstreaming proteins that are similarly regulated by both alleles.

Fig. 5:. Mediators of clinical AD vs. Aβ pathology (BioFINDER-2 plasma Somalogic)

a. The dot plot with error bars shows the mediation proportions and confidence intervals for proteins mediating the effects of APOE4 on Aβ. b. The same but for proteins mediating the effects of APOE4 on AD diagnosis. For similar plots a, b, proteins whose indirect effects in protein mediation path are significant after FDR correction alone or mediation proportions of protein-mediation pathways are greater than those of Aβ (or AD) -mediated pathways (when indirect effects of both mediation paths are significant after FDR correction) are defined as mediators and are shown. The x-axis represents the percentage of mediation proportions. The dots represent the estimated mediation proportions of each protein while red indicates a significant estimation. The horizontal lines represent the 95% confidence intervals of these estimates. Bold indicates that the direct effect of APOE4 is not significant, resulting in a total mediation effect of this protein. c. Boxplots show protein level change groups by APOE genotypes for key APOE4 => Aβ mediators in Aβ− individuals. d. The same but for key APOE2 => Aβ mediators in Aβ− individuals. On the box plots c and d, the y-axis represents residual protein levels after adjusting for age, sex and mean protein level. The center line of each box indicates the median (50th percentile); the lower and upper edges of the box represent the 25th and 75th percentiles, respectively. Whiskers extend to the most extreme values within 1.5 times the interquartile range (IQR); data points beyond this range are considered outliers and have been excluded from the plot display. The x-axis represents APOE genotype, ε2ε2 and ε2ε3 carriers were merged into the “22/23” group due to a small sample size of ε2ε2 carriers. Welch’s t-test was used to compare residual protein levels between groups. Two-sided p-values are tested, and p-values were adjusted for multiple comparisons using Holm-Bonferroni method. Each pair of groups was compared, only pairs with significant differences after FDR correction were shown. Asterisks indicate significance levels: **** for pFDR < 0.0001, *** for pFDR < 0.001, ** for pFDR < 0.01, * for pFDR < 0.05. e. The interaction plot shows the interaction between APOE4 carrier status (0 = non-carriers [ε3ε3], 1 = carriers) and S100A13 protein expression in predicting Aβ status in CU and MCI groups. Lines represent estimated Aβ status across APOE4 groups at three levels of S100A13 expression: one standard deviation below the mean (−1 SD, dashed gray), the mean (solid red), and one standard deviation above the mean (+1 SD, dashed brown). Among APOE4 carriers, higher S100A13 levels are associated with a greater likelihood of elevated Aβ status. f. The heatmap displays standardized regression coefficients (β) from models examining the associations between APOE-associated proteins (x-axis) and AD-related markers (tau-PET: tau-PET uptake in the temporal meta-ROI, Aβ-PET, Atrophy: cortical thickness in the temporal lobe, MMSE and mPACC: the modified preclinical Alzheimer’s cognitive composite) (y-axis) in Aβ− individuals. Full statistical details are available in Supplement Table 7. g. The same but in Aβ+ individuals. On heatmaps, red shades indicate positive associations; blue shades indicate negative associations. The darker the color, the stronger the associations between APOE-associated proteins and 5 AD biomarkers or cognitive performance. For heatmap annotation, * indicates a nominally significant association, *+ indicates a significant association after FDR correction.

Fig. 6:. Validation in CSF proteomics, GWAS and transcriptome

a. The dot plot with error bars shows the mediation proportions and confidence intervals for proteins mediating the effects of APOE4 on Aβ in ADNI. Proteins whose indirect effects in protein mediation path are significant after FDR alone or mediation proportions of protein-mediation pathways are greater than those of Aβ (or AD) -mediated pathways (when indirect effects of both mediation paths are significant after FDR) are defined as mediators and are shown. The x-axis represents the percentage of mediation proportions. The dots represent the estimated mediation proportions of each protein while red indicates a significant estimation. The horizontal lines represent the 95% confidence intervals of these estimates. b. Venn plot shows mediators identified in 3 SomaLogic proteomics dataset. c. Volcano plot shows APOE4 associated proteins with or without Aβ adjustment in the BioFINDER-2 CSF OLINK cohort, x-axis is the effect size of APOE4 on proteins without adjusting Aβ, red represent proteins associated with APOE4 regardless of adjusting Aβ status while blue indicate proteins associated with APOE4 only without adjusting Aβ. d. Stratified linear fit plot shows SNAP25 protein level change with age in APOE4 carriers and non-carriers in Aβ− individuals in the BioFINDER-2 CSF OLINK cohort. SNAP25 levels are plotted against age, with separate regression lines for individuals carrying the ε4 allele (ε4+, red) and those without it (ε4−, blue). Shaded areas represent 95% confidence intervals. e. Scatter plot shows proteins associated with APOE2 with or without adjusting Aβ status in the BioFINDER-2 CSF OLINK cohort. The red dot indicates a significant association regardless of adjusting Aβ. f. The table shows the number of SNPs in the coding gene of key mediators (identified as mediators in at least 2 dataset) that are associated with AD clinical diagnosis or CSF Aβ42 level in external GWAS studies. g. The plot shows brain-wide gene expression patterns, with statistics reflecting spatial correlations between APOE and APOE4 or APOE2 associated proteins. Only key proteins with significant spatial coexpression with APOE were shown. Red indicates mediators of the APOE4 => protein => AD diagnosis or Aβ pathology pathway; blue denotes mediators of the APOE2 => protein => AD diagnosis pathway. Purple marks proteins implicated in both, we note that ST8SIA1 were identified as mediators in the 2 mediation pathways by two distinct aptamers (seq.21508.7 for APOE4, seq.21663.149 for APOE2).