Proteomic signatures of the APOE ε4 and APOE ε2 genetic variants and Alzheimer’s disease

Proteomic signatures of the APOE ε4 and APOE ε2 genetic variants and Alzheimer’s disease | medRxiv /* */ /* */ <!-- <!-- /*! * yepnope1.5.4 * (c) WTFPL, GPLv2 */ (function(a,b,c){function d(a){return"[object Function]"==o.call(a)}function e(a){return"string"==typeof a}function f(){}function g(a){return!a||"loaded"==a||"complete"==a||"uninitialized"==a}function h(){var a=p.shift();q=1,a?a.t?m(function(){("c"==a.t?B.injectCss:B.injectJs)(a.s,0,a.a,a.x,a.e,1)},0):(a(),h()):q=0}function i(a,c,d,e,f,i,j){function k(b){if(!o&&g(l.readyState)&&(u.r=o=1,!q&&h(),l.onload=l.onreadystatechange=null,b)){"img"!=a&&m(function(){t.removeChild(l)},50);for(var d in y[c])y[c].hasOwnProperty(d)&&y[c][d].onload()}}var j=j||B.errorTimeout,l=b.createElement(a),o=0,r=0,u={t:d,s:c,e:f,a:i,x:j};1===y[c]&&(r=1,y[c]=[]),"object"==a?l.data=c:(l.src=c,l.type=a),l.width=l.height="0",l.onerror=l.onload=l.onreadystatechange=function(){k.call(this,r)},p.splice(e,0,u),"img"!=a&&(r||2===y[c]?(t.insertBefore(l,s?null:n),m(k,j)):y[c].push(l))}function j(a,b,c,d,f){return q=0,b=b||"j",e(a)?i("c"==b?v:u,a,b,this.i++,c,d,f):(p.splice(this.i++,0,a),1==p.length&&h()),this}function k(){var a=B;return a.loader={load:j,i:0},a}var l=b.documentElement,m=a.setTimeout,n=b.getElementsByTagName("script")[0],o={}.toString,p=[],q=0,r="MozAppearance"in l.style,s=r&&!!b.createRange().compareNode,t=s?l:n.parentNode,l=a.opera&&"[object Opera]"==o.call(a.opera),l=!!b.attachEvent&&!l,u=r?"object":l?"script":"img",v=l?"script":u,w=Array.isArray||function(a){return"[object Array]"==o.call(a)},x=[],y={},z={timeout:function(a,b){return b.length&&(a.timeout=b[0]),a}},A,B;B=function(a){function b(a){var a=a.split("!"),b=x.length,c=a.pop(),d=a.length,c={url:c,origUrl:c,prefixes:a},e,f,g;for(f=0;f<d;f++)g=a[f].split("="),(e=z[g.shift()])&&(c=e(c,g));for(f=0;f<b;f++)c=x[f](c);return c}function g(a,e,f,g,h){var i=b(a),j=i.autoCallback;i.url.split(".").pop().split("?").shift(),i.bypass||(e&&(e=d(e)?e:e[a]||e[g]||e[a.split("/").pop().split("?")[0]]),i.instead?i.instead(a,e,f,g,h):(y[i.url]?i.noexec=!0:y[i.url]=1,f.load(i.url,i.forceCSS||!i.forceJS&&"css"==i.url.split(".").pop().split("?").shift()?"c":c,i.noexec,i.attrs,i.timeout),(d(e)||d(j))&&f.load(function(){k(),e&&e(i.origUrl,h,g),j&&j(i.origUrl,h,g),y[i.url]=2})))}function h(a,b){function c(a,c){if(a){if(e(a))c||(j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}),g(a,j,b,0,h);else if(Object(a)===a)for(n in m=function(){var b=0,c;for(c in a)a.hasOwnProperty(c)&&b++;return b}(),a)a.hasOwnProperty(n)&&(!c&&!--m&&(d(j)?j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}:j[n]=function(a){return function(){var b=[].slice.call(arguments);a&&a.apply(this,b),l()}}(k[n])),g(a[n],j,b,n,h))}else!c&&l()}var h=!!a.test,i=a.load||a.both,j=a.callback||f,k=j,l=a.complete||f,m,n;c(h?a.yep:a.nope,!!i),i&&c(i)}var i,j,l=this.yepnope.loader;if(e(a))g(a,0,l,0);else if(w(a))for(i=0;i (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0];var j=d.createElement(s);var dl=l!='dataLayer'?'&l='+l:'';j.src='//www.googletagmanager.com/gtm.js?id='+i+dl;j.type='text/javascript';j.async=true;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-P4HH5NV'); Skip to main content Home About Submit ALERTS / RSS Search for this keyword Advanced Search Proteomic signatures of the APOE ε4 and APOE ε2 genetic variants and Alzheimer’s disease View ORCID Profile Lina Lu , View ORCID Profile Alexa Pichet Binette , Ines Hristovska , Shorena Janelidze , Bart Smets , View ORCID Profile Irene Cumplido Mayoral , Aparna Vasanthakumar , Britney Milkovich , View ORCID Profile Rik Ossenkoppele , Varsha Krish , Farhad Imam , View ORCID Profile Sebastian Palmqvist , Jacob Vogel , View ORCID Profile Erik Stomrud , View ORCID Profile Oskar Hansson , View ORCID Profile Niklas Mattsson-Carlgren doi: https://doi.org/10.1101/2025.08.04.25332945 Lina Lu 1 Clinical Memory Research Unit, Department of Clinical Sciences Malmö, Faculty of Medicine, Lund University , Lund, Sweden Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Lina Lu For correspondence: lina.lu{at}med.lu.se niklas.mattsson-carlgren{at}med.lu.se Alexa Pichet Binette 1 Clinical Memory Research Unit, Department of Clinical Sciences Malmö, Faculty of Medicine, Lund University , Lund, Sweden 2 Department of Physiology and Pharmacology, Université de Montréal , Montréal, QC, Canada 3 Centre de recherche de l’institut universitaire de gériatrie de Montréal (CRIUGM) , Montréal, QC, Canada Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Alexa Pichet Binette Ines Hristovska 1 Clinical Memory Research Unit, Department of Clinical Sciences Malmö, Faculty of Medicine, Lund University , Lund, Sweden Find this author on Google Scholar Find this author on PubMed Search for this author on this site Shorena Janelidze 1 Clinical Memory Research Unit, Department of Clinical Sciences Malmö, Faculty of Medicine, Lund University , Lund, Sweden Find this author on Google Scholar Find this author on PubMed Search for this author on this site Bart Smets 4 Janssen Pharmaceutica NV, a Johnson & Johnson company , Beerse, Belgium Find this author on Google Scholar Find this author on PubMed Search for this author on this site Irene Cumplido Mayoral 1 Clinical Memory Research Unit, Department of Clinical Sciences Malmö, Faculty of Medicine, Lund University , Lund, Sweden Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Irene Cumplido Mayoral Aparna Vasanthakumar 5 Abbvie, North Chicago , IL, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site Britney Milkovich 5 Abbvie, North Chicago , IL, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site Rik Ossenkoppele 1 Clinical Memory Research Unit, Department of Clinical Sciences Malmö, Faculty of Medicine, Lund University , Lund, Sweden 6 Amsterdam Neuroscience, Neurodegeneration, Vrije Universiteit Amsterdam , Amsterdam, Netherlands 7 Alzheimer Center Amsterdam, Neurology, Vrije Universiteit Amsterdam , Amsterdam UMC location VUmc, Amsterdam, Netherlands Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Rik Ossenkoppele Varsha Krish 8 Gates Ventures , Seattle, WA, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site Farhad Imam 8 Gates Ventures , Seattle, WA, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site Sebastian Palmqvist 1 Clinical Memory Research Unit, Department of Clinical Sciences Malmö, Faculty of Medicine, Lund University , Lund, Sweden 10 Memory Clinic, Skåne University Hospital , Malmö, Sweden Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Sebastian Palmqvist Jacob Vogel 9 Department of Clinical Sciences Malmö, SciLifeLab,Lund University , Lund, Sweden Find this author on Google Scholar Find this author on PubMed Search for this author on this site Erik Stomrud 1 Clinical Memory Research Unit, Department of Clinical Sciences Malmö, Faculty of Medicine, Lund University , Lund, Sweden 10 Memory Clinic, Skåne University Hospital , Malmö, Sweden Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Erik Stomrud Oskar Hansson 1 Clinical Memory Research Unit, Department of Clinical Sciences Malmö, Faculty of Medicine, Lund University , Lund, Sweden Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Oskar Hansson Niklas Mattsson-Carlgren 1 Clinical Memory Research Unit, Department of Clinical Sciences Malmö, Faculty of Medicine, Lund University , Lund, Sweden 10 Memory Clinic, Skåne University Hospital , Malmö, Sweden Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Niklas Mattsson-Carlgren For correspondence: lina.lu{at}med.lu.se niklas.mattsson-carlgren{at}med.lu.se Abstract Full Text Info/History Metrics Supplementary material Data/Code Preview PDF 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. Main The Apolipoprotein E ( APOE ) gene is the strongest genetic factor for sporadic Alzheimer’s disease (AD), with three main alleles: ε2, ε3, and ε4. Compared to ε3, the ε4 allele ( APOE4 ) increases AD risk in a dose-dependent manner—approximately 2–3 fold in people with one ε4 allele and up to 12-fold in ε4 homozygotes 1 , 2 and is linked to Aβ aggregation 3 – 5 . In contrast, APOE2 carriers have a reduced AD risk 6 – 8 and Aβ burden 9 – 11 , and delayed onset in high-penetrance AD mutations carriers 12 . APOE has been suggested to contribute to AD via lipid metabolism 13 , 14 , mitochondrial function 15 , 16 , Aβ regulation 17 , 18 , and neuroinflammation 19 , 20 , but most data on APOE comes from animal or cell experiments, with limited human evidence supporting specific molecular pathways linking APOE to AD. Developments in proteomics provide a new perspective for analyzing the molecular heterogeneity of AD, for example identifying CSF biomarkers in both asymptomatic and symptomatic AD 21 , and CSF biomarkers related to Aβ- and tau-related pathology 22 . Proteomic studies also suggested APOE4 -dependent pathways 23 , but were generally limited by small sample sizes 24 , 25 , or reliance on single tissue sources or single proteomics platforms 23 , 26 , 27 , limiting generalizability. To reduce bias from individual methods or tissues, there is a need for integrative studies that combine multiple cohorts, matrices (e.g., CSF and blood), and analytical platforms. A systematic, fine-grained analysis of the APOE -related protein network—considering both APOE2 and APOE4 —may reveal distinct mechanistic pathways and inform targeted intervention strategies. Therefore, we mapped APOE -related proteins across large cohorts and tissues using multi-level proteomics, revealing distinct APOE4 - and APOE2 -associated signatures linked to AD pathology. This provides a molecular basis for future stratified management and targeted therapies. Results The APOE-plasma protein signature and its role in AD: Discovery cohort The GNPC plasma SomaLogic Cohort In the Global Neurodegeneration Proteomics Consortium (GNPC) cohort (N=4,045, aged 20-90, median age: 75), APOE4 -associated proteins were identified by comparing APOE4 carriers (1,334 ε3ε4; 243 ε4ε4) to ε3ε3 individuals (N=1,969). To assess the protective effect of APOE2 in AD, ε2 carriers (13 ε2ε2; 375 ε2ε3) were compared with ε3ε3 individuals. ε2ε4 carriers (N=111) were excluded from these comparisons due to potential confounding but included in analyses of protein level changes across all APOE genotypes (Supplement Table 1). APOE4 and AD protein signatures: variances and similarity To delineate the distinct and overlapping proteomic landscapes of APOE4 and AD (clinical diagnosis), we applied three regression models across 7,285 proteins. Model 1 identified 478 proteins associated with APOE4 (unadjusted for AD diagnosis), e.g., upregulated SPC25 and LRRN1, and downregulated S100A13 and TBCA ( Fig. 2a ). Model 2 identified 4,014 proteins associated with AD (unadjusted for APOE4 ), with top hits e.g., ACHE, SPC25 and MMP8 ( Fig. 2b ). Model 3 included terms for both APOE4 and AD to isolate independent effects (Supplement Fig. 1A, B). Of the APOE4 -associated proteins, 253 (53%) remained significant after adjustment for AD diagnosis, including 78 not linked to AD—suggesting APOE4 -specific pathways ( Fig. 2c , red). Conversely, 3,783 (94%) AD-associated proteins remained significant after adjustment for APOE4 status, with 3,453 specific to AD ( Fig. 2c , blue). 98 proteins were independently associated with both APOE4 and AD across models ( Fig. 2c , black). Download figure Open in new tab 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 . Download figure Open in new tab 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. Mediation reveals bidirectional links between APOE4 , proteins, and AD diagnosis To disentangle directionality between APOE4- associated proteins and AD, we conducted mediation analyses on the 478 APOE4 -associated proteins identified in the previous analysis. 25 proteins, e.g., SPC25, ARL2, LRRN1, NEFL, S100A13, TBCA, and ZW10 showed stronger mediation in the APOE4 => protein => AD pathway than the reverse ( Fig. 2d ), with partial mediation effects accounting for up to ∼38% of APOE4 ’s total effect on AD risk, consistent with effects upstream of clinical onset (potentially reflecting early pathophysiological changes). In contrast, 306 proteins showed stronger mediation in the APOE4 => AD => protein pathway, suggesting downstream consequences of disease. 111 proteins, e.g. NPTXR, were affected exclusively through AD ( Fig. 2d ). For 195 proteins, e.g., RAB14, both direct and AD-mediated effects were observed, with partial AD-mediated (absolute) proportions reaching up to 67% (Supplement Table 2). Effects of APOE4 in cognitively unimpaired individuals To explore early, potentially pre-morbid effects of APOE4 , we performed analysis within CU individuals. APOE4 was associated with 224 proteins in CU, including 22 of 25 AD-mediators (e.g., SPC25, LRRN1, S100A13, TBCA, Supplement Fig. 1C) and 90 of 195 AD partly-mediated proteins (e.g. RAB14). Most (199, 89%) showed similar significant effect sizes as in the full cohort, supporting APOE4 ’s pre-symptomatic impact ( Fig. 2e ). Age stratification (median age = 73) revealed that 35% (78) of APOE4 -associated proteins in CU remained significant in both younger CU and in older CU, including SPC25, TBCA, and S100A13 (Supplement Fig. 1D, E, F). Models including an APOE4 *age interaction confirmed largely age-independent effects of APOE4 on these proteins in CU, except for PPP1R2, WNT10B and DSG2 (Supplement Fig. 1G, H). Among 93 overlapping proteins between APOE4 -associated alteration in CU and AD-associated alteration in ε3ε3 carriers, APOE4 had a greater effect on some mediators e.g. ARL2 and ZW10 ( Fig. 2f ). 31 (e.g., ARL2, SNAP23) showed concordant effects, while 62 showed opposite directions, supporting that APOE4 may induce early, possibly compensatory changes that differ from those driven by downstream AD pathology. Consistent regulation pattern of APOE4 across diagnoses To assess how disease status modifies APOE4 effects, we also identified APOE4 -associated proteins within AD cases. Of 123 APOE4 -associated proteins in AD, 103 overlapped with those in CU and the full cohort, indicating largely diagnosis-independent effects (Supplement Fig. 1I, J). However, two proteins—SPC25 and TBCA—showed significant APOE4 *AD interactions: SPC25 was more strongly upregulated by APOE4 in AD patients, while TBCA exhibited enhanced downregulation (Supplement Fig. 1K, L), potentially reflecting secondary responses to disease progression. Early dysregulated proteins highlight APOE4 -linked pathways To elucidate functional consequences of pre-clinical APOE4 -related proteomic changes, 199 proteins consistently associated with APOE4 across the full and CU cohorts were mapped to Reactome networks and used to train a deep learning classifier (accuracy: 0.68 training, 0.71 testing) in predicting AD using biologically informed neural networks (BINNs)-based pathway analysis. Key predictors such as RIPK1, AKT2, and SPC25 related cellular responses to mechanical stimuli, developmental cell lineages, mRNA capping, protein-protein interactions at synapses, cell cycle control etc. ( Fig. 2g ). Stratifying APOE4 -associated proteins reveals distinct mechanistic routes to AD dementia diagnosis To further dissect the heterogeneity of APOE4 -associated proteomic functionality, we subdivided all 478 APOE4 -associated proteins into four mechanistic categories based on mediation and diagnostic associations ( Fig. 2h ). Clustering with linear discriminant analysis (LDA) supported biological distinctiveness ( Fig. 2i ). Cell-type enrichment, pathway, and interaction analyses further supported differences between these four groups of proteins ( Fig. 2j ): AD mediator proteins (N=25, APOE4 => protein => AD): Upregulated proteins (N=8, e.g., SPC25 and LRRN1) were enriched in OPCs (nominally) and linked to neuronal development and synaptic function, cell cycle and MHC assembly. Downregulated proteins (N=17, e.g., NEFL, ZW10, TBCA, S100A13 and ARL2) were involved mostly in cytoskeletal organization and organelle localization. AD-mediated proteins (N=306, APOE4 => AD => protein): Upregulated proteins (N=137) were enriched in vascular cells (e.g., smooth muscle cells and perivascular fibroblasts, the latter with nominal significance) and involved in organelle assembly and localization, protein and hormone regulation. Downregulated proteins (N=169) were mainly linked to immune function. APOE4 -specific proteins (N=69, associated with APOE4 but not with AD): Upregulated proteins (N=24) were related to vascular function, tissue remodeling and immune response, while downregulated ones (N=45) involved ubiquitin-dependent catabolic processes and metabolic pathways. 9 proteins were reclassified due to partial mediation effect ( Fig. 2h ). Non-specific proteins (N=78, associated with both APOE4 and AD but not involved in either mediation pathway): Upregulated proteins (N=31) were enriched (nominally) in pericytes; downregulated (N=47) in T cells. These proteins are largely clustered together with proteins involved in meditation paths in LDA ( Fig. 2i ), suggesting overlapping biological patterns. The APOE2-plasma protein signature Of 211 proteins associated with APOE2 , 201 (95%) remained significant after adjusting for AD, with 99 specific to APOE2 without AD association (Supplement Fig. 2A, C). Among 3,055 AD-associated proteins, 3,016 (99%) remained after adjusting for APOE2 , with 2,911 (97%) specific to AD (Supplement Fig. 2B, C). 88 proteins consistently associated with both APOE2 and AD. UNG, VPS29 and BIRC2 were most upregulated in APOE2 carriers and were also upregulated to a similar extent in CU APOE2 carriers, this including a total of 157 proteins (74% of all 211 APOE2 -associated proteins) ( Fig. 3a, b ). Some (N=13), like VPS29 and AKT2, showed an age-dependent pattern with stronger upregulation in younger CU carriers ( Fig. 3c, d ). Mediation analyses identified 71 proteins (e.g., UNG, ARL2, PCLAF, and SNAP23) for which the APOE2 => protein => AD diagnosis pathway contributed more than the APOE2 => AD diagnosis => protein pathway. These proteins partially mediate 5–92% of APOE2 ’s effect on AD. Conversely, 28 proteins (e.g. RAB14) showed partial mediation by AD, with up to 16% effect explained (Supplement Table 2). Download figure Open in new tab 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. Cell-type and functional landscape of APOE2 -associated proteins 157 consistently identified proteins (full and CU cohorts) were investigated using BINN. The resulting deep learning model (accuracy: 0.74 training, 0.75 testing) revealed top contributors including UBB, APOB, EIF4G1, BIRC2 and AKT2 ( Fig. 3f ), enriched in DNA repair, peroxisomal protein import and autophagy etc. We further classified all 211 APOE2 -associated proteins into four mechanistic categories ( Fig. 3g ), protein clustering showed a scattered distribution in LDA space ( Fig. 3h ), indicating distinct functional and spatial features supported by cell type enrichment and subgrouped pathway enrichment results ( Fig. 3i , Supplement Table 3): AD mediator proteins (N=71, APOE2 => protein => AD): Upregulated proteins (N=20, e.g., SIRT1, KMT2C) were enriched in ependymal cells (nominally) and linked to metabolism, chromatin remodeling, and DNA repair; downregulated proteins were tied to membrane organization (SNAP23, APOB etc.), immune and NF-κB pathways (e.g., MAPK8, TRADD), suggesting suppressed inflammation. AD-mediated proteins (N=28, APOE2 => AD => protein): 7 upregulated proteins, enriched in vascular cells (notably meningeal fibroblasts), supported angiogenesis (e.g., MEOX2). APOE2 -specific proteins (N=98): Upregulated proteins (N=31, e.g., PHGDH, DLL1) were enriched in oligodendrocytes (nominally) and supported myelination and vascular homeostasis; downregulated ones involved in ubiquitin-mediated proteolysis. Non-specific proteins (N=14): Close to the APOE2 -specific cluster in LDA space ( Fig. 3h ), indicating APOE2 -related functionality. Most (N=9) were downregulated and enriched in ependymal cells (nominally), linked to endomembrane system organization (e.g., ZW10, AKT2). APOE4 and APOE2 : Specificity, Offset, and Synergy APOE4 and APOE2 associated proteins were largely non-overlapping ( Fig. 4a ). Specifically, 8 proteins —including SPC25 and LRRN1—mediated APOE4 ’s effect on AD without being associated with APOE2 . These proteins showed APOE4 gene dose–dependent patterns in CU individuals ( Fig. 4b ), with ε4ε4 showing extreme alteration, and ε2ε4 carriers resembling ε3ε4 carriers (consistent with ε4 dominance). Conversely, 25 proteins (e.g. UNG, PCLAF) mediated APOE2 ’s effect without being associated with APOE4 , and followed an ε2-dominant pattern ( Fig. 4b ). 3 proteins (ARL2, OTULIN and TMCC3) mediated both APOE4 and APOE2 ’s effect on AD diagnosis. Download figure Open in new tab Fig. 4: Different pattens 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. SNRPF mediated APOE4 ’s effect on AD but appeared downstream of pathology in APOE2 models, while 41 proteins—including SNAP23 and APOB—showed the opposite pattern (Supplement Table 2). These shifts underscore divergent molecular pathways for APOE4 and APOE2 in AD. 62 proteins were not mediated by AD in either APOE4 or APOE2 models— suggesting upstream regulation—yet were associated with both alleles. 13 showed opposing allele effects in CU individuals ( Fig. 4c ), following either ε4-dominant (e.g., FOXO1, PHGDH) or ε2-dominant (e.g., VPS29, BIRC2) patterns ( Fig. 4d ). The remaining 49 proteins were similarly regulated by both alleles; key mediators such as S100A13, BCDIN3D, TBCA, and ARL2 were downregulated in both ε2 and ε4 carriers, with ε2ε4 individuals resembling ε4ε4 carriers ( Fig. 4e ). APOE4 , plasma Proteins, and Aβ pathology: BioFINDER-2 Cohort The BioFINDER-2 plasma SomaLogic Cohort BioFINDER-2 cohort is a subset of the GNPC with deeper phenotyping, where whether plasma proteins mediated the effects of APOE on Aβ pathology was tested. The same proteins as in GNPC were available for testing (Supplement Fig. 3A). Focusing on Aβ pathology, 259 AD dementia, 316 Mild Cognitive Impairment (MCI) and 846 CU participants were included, resulting in 1,421 participants in total (aged 20–93 years; median age: 72). Based on CSF Aβ42/40 ratios, 715 individuals (50.3%) were Aβ- and 706 (49.7%) were Aβ+ (Supplement Table 1). Mediators of clinical AD vs. Aβ pathology Consistent with GNPC findings, most plasma proteins altered in APOE4 carriers were independent of Aβ status but showed Aβ-mediated changes. 65 proteins (39% of APOE4 -associated proteins identified in this cohort) were altered in both the full cohort and Aβ-individuals, supporting early, preclinical effects. Among them, AKT2, HDAC8, APOB, CYP3A4, SPC25, and BCDIN3D—linked to chromatin organization, gene expression, drug ADME and cell cycle—were top features in the BINN model predicting Aβ status in CU and MCI individuals (Supplement Fig. 4A). Similarly, 43 (59% of APOE2 -associated proteins) proteins were already altered in Aβ- APOE2 carriers. Among these, key Aβ predictors like APOB, SHH, and BIRC2 were enriched in pathways like mitochondrial import, SLC transport, lipid metabolism, and adaptive immunity (Supplement Fig. 4B). Mediation analysis identified 8 proteins (e.g., SPC25, TBCA, S100A13, CTF1, BCDIN3D) partially mediated APOE4 ’s effect on Aβ pathology (4–35%; Fig. 5a ); 7 remained significant after adjusting for population stratification (except UBL3; Supplement Fig. 5A). Ten proteins mediated APOE4 ’s effect on AD diagnosis (5–83%) ( Fig. 5b ), among these, SPC25, TBCA, and S100A13 replicated from GNPC ( Fig. 2d ) and showed full mediation, 5 proteins, including S100A13, TBCA, and PHGDH, remained robust after adjusting for population stratification (Supplement Fig. 5B). Download figure Open in new tab Fig. 5: Mediators of clinical AD vs. AB 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. Most APOE2 -associated proteins were independent of Aβ and showed no mediation effect by Aβ. However, APOB, SNAP23, WARS2, and PCLAF—replicated from GNPC—fully mediated APOE2 ’s effect on Aβ pathology (Supplement Table 2). No mediators were identified for APOE2 ’s effect on clinical AD diagnosis. Alterations of these mediators were detectable in Aβ-individuals, largely age-independent (except SPC25), and showed allele-specific patterns: APOE4 mediators displayed ε4 dose-dependence and ε4-dominance in ε2ε4 carriers, while APOE2 mediators showed ε2-dominant effects ( Fig. 5c, d ). APOE4 *protein interaction effects on Aβ status Beyond mediation, we assessed whether APOE4 -associated proteins modulate genetic risk. An APOE4 *protein interaction analysis revealed that S100A13 significantly influenced the relationship between APOE4 and Aβ status: APOE4 carriers with lower S100A13 levels showed higher Aβ positivity in CU and MCI individuals, an effect diminished at higher protein levels ( Fig. 5e ). Associations between APOE -related proteins and AD features in BioFINDER-2 To evaluate functional relevance, we assessed associations between APOE -related proteins and AD features (tau-PET, cortical thickness, cognition) in Aβ- (N=715) and Aβ+ (N=706) individuals. In the Aβ-group, higher VPS29 and UBA2 levels were associated with worse cognition, lower MYSM1, CRIP1, PDHB or SEMA6C levels were associated with more severe atrophy ( Fig. 5f ). In Aβ+, lower FGFBP3, higher EIF4A2 or GBP1 levels were associated with better cognition ( Fig. 5g ). Those proteins are either APOE ( APOE4 or APOE2 )-specific or associated with both APOE and Aβ with no mediation effect. The lack of associations between mediator proteins and AD biomarkers or cognition may reflect their early role in APOE -related pathways, prior to the emergence of downstream pathology and clinical symptoms. CSF proteomic signature in SomaLogic proteomics (ADNI cohort) In ADNI CSF (N=666; SomaLogic 7k; aged 54-91, median age 74; 58% Aβ+; 325 ε4, 46 ε2, 295 ε3ε3 carriers), we validated APOE -associated pathways and mediators. APOE4 was associated with 684 proteins (124 GNPC-consistent), 48% independent of Aβ status; 122 already altered in Aβ-individuals, including BINN top features like AKT2 and ARF1 (accuracy=0.66 for training and testing), enriched in ECM, autophagy, meiosis and RNA pathways (Supplement Fig 6A). APOE2 was associated with 93 proteins (all evident in Aβ-, 35 GNPC-consistent), non-overlapping with Aβ-related proteins; BINN predictors (accuracy: 0.70 training, 0.66 testing) such as PTPN1 and MAPK8 were enriched in IGF signaling, autophagy, and DNA repair (Supplement Fig 6B). Mediation analysis identified 20 proteins linking APOE4 to Aβ (2–57% mediation; Fig. 6a ) while more proteins (251) were downstream of Aβ (Supplement Table 2). Several mediators overlapped with plasma results ( Fig. 6b ): S100A13 and TBCA were consistent across GNPC and BioFINDER-2; BCDIN3D was consistent when using Aβ as outcome, indicating Aβ-specific mediation; CSF SPC25 was not a mediator despite its plasma effect, suggesting matrix specificity. No APOE2 mediators were identified (our power was very limited here due to few ε2 carriers in ADNI). Download figure Open in new tab 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 ) CSF proteomic signature in OLINK proteomics (BioFINDER-2 cohort) To validate CSF APOE proteomic findings in a different proteomic platform, we utilized OLINK proteomics data in the BioFINDER-2 cohort (N=1,475, 1,391 proteins, several key mediators e.g., SPC25, CTF1, ARL2, ZW10, APOB, PCLAF and WARS2 were unavailable). Available key mediators e.g., S100A13, TBCA and SNAP23 were not significantly associated with APOE in CSF OLINK. Of 43 APOE4 -associated proteins, most lost significance after Aβ adjustment, except for 5 proteins e.g., SNAP25 and DMD ( Fig. 6c ). Mediation analysis showed weak APOE4 => Aβ partial mediation via CKAP4 (2%) and partial APOE4 => AD mediation via SNAP25 (7%) (Supplement Table 2). SNAP25 was more upregulated in young Aβ- APOE4 carriers, with age-dependent attenuation ( Fig. 6d ). For APOE2 , only LDLR showed significant association regardless of Aβ adjustment ( Fig. 6e ) but did not mediate AD diagnosis or Aβ pathology. Plasma proteomic signature in OLINK proteomics (UKB cohort) To validate plasma APOE proteomic signatures in a different proteomic platform, we utilized the UKB cohort (N= 4,820, 1,319 OLINK proteins, most key mediators from the SomaLogic analyses were unavailable in the OLINK assays). 10 APOE4 and 38 APOE2 associated proteins were identified, among these, BRK1 and PLA2G7 were associated with both alleles in opposing directions (Supplementary Fig. 7A, B). Interestingly, both APOE4 and APOE2 showed gene-dose effects on PLA2G7 (Supplementary Fig. 7C). LDLR was markedly elevated in ε2ε2 individuals (Supplement Fig. 7C). SNPs associated with Aβ42 measurements and AD in coding genes of mediator-proteins In external GWAS data from the the GCST002245 study 28 , of the 33 proteins identified as mediators of APOE4 ’s effect on AD in GNPC, BioFINDER-2 or ADNI, 20 had coding genes harboring AD-associated SNPs. Similarly, among 71 APOE2 => AD mediator proteins, 35 had AD-associated SNPs. Based on the GCST90129599 study 29 , 17 of 26 APOE4 => Aβ mediators and all 4 APOE2 => Aβ mediators had SNPs associated with CSF Aβ42 levels. Notably, among the 6 key proteins mediating APOE4 ’s effect on AD diagnosis or Aβ pathology in at least two datasets ( Fig. 6f ), 4 (S100A13, SPC25, TBCA and CTF1) had coding variants linked to Aβ42 or AD. All 4 key APOE2 mediators (APOB, PCLAF, SNAP23, WARS2) identified in multiple datasets were also supported by SNP associations with Aβ42 or AD ( Fig. 6f ). Spatial Overlap Analyses APOE -protein spatial co-expression was validated using Allen Human Brain Atlas transcriptomic data. Among 368 APOE4 -associated proteins (in GNPC) with available expression data, 170 (46%) showed significant spatial correlation with APOE , including 10 AD or Aβ mediators e.g. positive associations for S100A13, PHGDH, and LRRN1, and negative associations for ST8SIA1, NEFL, and ZW10 ( Fig. 4f ). Of 166 APOE2 -associated proteins, 70 (42%) showed spatial correlation with APOE , including 21 AD mediators such as HES5 (R = 0.76) and PRDX6 (R = 0.72). Several key mediators, e.g., SPC25, APOB, and PCLAF, were unavailable for transcriptomic testing. No spatial correlation was observed between Aβ-PET signal and the expression of APOE or any APOE -associated genes (Supplement Table 4). Dynamic profiling in CSF supports APOE temporal effects To investigate longitudinal APOE -protein associations, we utilized the PPMI CSF OLINK cohort (N=253), among 72 APOE4 and 12 APOE2 associated proteins from GNPC, only LGALS9, PRSS8 and NEFL showed nominally significant APOE *time interactions. Similarly, among 26 testable proteins from BioFINDER-2 CSF OLINK, only SIGLEC1 showed a nominally significant APOE4 *time interaction. Suggesting that APOE -associated proteomic signatures are largely stable over time, supporting their role as early and persistent molecular features rather than dynamic markers of progression. Full description is in Supplement text. Cross-platform and cross-tissue variability of APOE-associated proteomic signatures APOE4 -associated protein profiles showed strong consistency within the same tissue and platform (e.g., r=0.72 between plasma SomaLogic datasets), but substantially lower correlations across tissues and platforms (Supplement Fig. 8A-E). NEFL exhibited notable heterogeneity across platforms (Supplement Fig. 8F, G). No protein was consistently associated with APOE4 or APOE2 across all datasets and technologies, indicating high heterozygosity of proteomics findings across tissues or platforms (Supplement Fig. 8H, I). Full description is in Supplement text. Discussion We conducted a large-scale, cross-platform proteomic analysis of plasma and CSF from over 10,000 samples to map how APOE isoforms (ε2, ε4) modulate AD risk. APOE -associated proteins were classified into risk mediators, protective mediators, downstream reflectors, and neutral markers, revealing largely distinct pathways for APOE4 and APOE2 . Key APOE4 -associated mediators—such as SPC25 (involved in kinetochore assembly, previously linked to microglial activation in mouse models 30 and elevated in MCI patients 31 ), S100A13 (moderating Aβ positivity in APOE4 carriers, and implicated in cellular senescence 32 ), and ZW10 (a spindle checkpoint protein), some were enriched in OPCs—point to dysregulation of cell cycle control and mitotic processes as potential contributors to AD risk and Aβ accumulation 33 – 36 , especially in glial cells. TBCA and ARL2 further highlight cytoskeletal disruption as an APOE4 -related feature contributing to neurodegeneration 37 , 38 . In contrast, APOE2 -associated mediators—including SNAP23 (endocytic trafficking), APOB (suggesting altered lipid transport and metabolism), WARS2 (mitochondrial translation), and PCLAF (DNA repair)—suggest a protective effect of APOE2 via maintenance of cellular homeostasis, mitochondrial function, and genomic stability 39 – 42 . These mechanistic insights are supported by consistent signals across datasets, reinforced by GWAS and gene-dosage associations. Together, they nominate distinct, APOE genotype-specific biological processes—particularly those involving cell cycle regulation, cytoskeletal integrity, and mitochondrial health—as promising therapeutic targets for AD. While individual mediators reveal discrete mechanisms, our broader analysis reveals a temporally dynamic APOE4 -driven program: asymptomatic APOE4 carriers exhibit early proteomic changes implicating cell cycle, stress responses, and cytoskeletal disruption, preceding amyloid or symptoms. As pathology advances, more APOE4 -associated proteins become AD-mediated, including vascular proteins involved in organelle organization and adhesion, alongside downregulated immune-related proteins 43 – 45 , reflecting a staged progression from genetic priming to pathology-driven cascades. Beyond merely countering APOE4 46 – 48 , APOE2 confers protection through early molecular programs—such as DNA repair, vesicular trafficking, and mitochondrial function— detectable even in CU or Aβ-individuals and supported by prior studies 49 – 51 . Most APOE2 proteins act as upstream of AD with non-Aβ mediated effect and minimal mediation by AD, suggesting early buffering of pathology (may explain better baseline cognitive ability in APOE2 carriers 52 ) that may wane over time. Functional classification delineates four categories: (1) AD mediators promoting metabolism, DNA repair, and reduced inflammation; (2) AD-mediated proteins involved in vascular remodeling and angiogenesis aligning with AD-mediated proteins in APOE4 carriers; (3) APOE2 -specific proteins supporting oligodendrocyte function, myelination, and proteostasis; and (4) non-specific proteins linked to endomembrane and ependymal function—together defining a comprehensive, system-level functional classification of APOE2 -associated pathways, addressing a critical gap in the current understanding of APOE isoform biology. Though largely distinct, some proteins are dysregulated by both alleles, showing allele-specific dominance patterns—e.g., FOXO1 and PHGDH ( APOE4 dominant), VPS29 and BIRC2 ( APOE2 dominant)—indicating genotype-specific molecular switches. Some proteins (ARL2, TBCA, S100A13) are similarly affected by both, with APOE2 occasionally mimicking ε4-like effects, which may partly explain reports linking APOE2 to neurodegenerative risk under certain conditions 52 . Strengths of our study include the integration of multiple platforms and tissues across >10,000 participants, enabling a comprehensive APOE molecular map. Mediation modeling separates causal mediators from downstream cascades, providing mechanistic insight beyond correlations. Key results were robustly replicated and supported by genetic and transcriptomic data, although variability across datasets highlights the need for context-aware analyses and validation 53 , 54 . Limitations include platform-related protein coverage differences, limited ε2 carrier numbers reducing power especially in CSF and longitudinal analyses, and a largely cross-sectional design limiting temporal resolution. Causality inferred by statistical mediation requires functional validation. In summary, our integrative proteomic analysis reveals that APOE4 and APOE2 regulate AD risk through distinct, largely non-overlapping pathways: APOE4 induces early cell cycle dysfunction, especially in glial cells, while APOE2 enhances resilience through DNA repair, inflammation reduction and mitochondrial function. These allele-specific changes appear before symptoms or Aβ pathology, highlighting a critical early intervention window. The consistency of these findings across diverse platforms and tissues highlights robust, genotype-specific biomarkers and intervention targets, based on APOE -linked proteomic changes. Methods GNPC SomaLogic 7K participants The GNPC is a multi-center, international proteomics initiative. The cohort composition, sample processing, and proteomics pipeline have been described in detail previously 55 . The present study represents a focused deep dive into the GNPC dataset to investigate the role of APOE4 and APOE2 in AD, using the same inclusion criteria and preprocessing steps as in Imam et al. (2025). In total, 4,045 participants were included (aged 20–90 years; median age 75), comprising 2,566 CU individuals and 1,479 patients with AD dementia. Plasma proteomics were profiled using the SomaScan® 7K platform (SomaLogic Inc.), which quantifies proteins with DNA-based SOMAmer® reagents 56 . Proteomics data underwent the standard SomaScan® pipeline, including quality control and adaptive normalization by maximum likelihood (ANML). APOE -targeting aptamers were excluded, resulting in 7,285 aptamers corresponding to 6,358 unique proteins. All aptamer intensities were log2-transformed, and values deviating more than 5 standard deviations from the mean were defined as outliers and excluded. This preprocessing was applied to all SomaLogic proteomics datasets used in this study. For primary analyses, ε2ε4 carriers were excluded to avoid confounding 55 . APOE4 analyses compared ε3ε4 and ε4ε4 carriers to ε3ε3 (binary variable), and APOE2 analyses compared ε2ε2 and ε2ε3 carriers to ε3ε3. ε2ε4 carriers were only included in descriptive comparisons of all six genotype groups (ε2ε2, ε2ε3, ε2ε4, ε3ε3, ε3ε4, ε4ε4), consistent with GNPC, BioFINDER-2, and UK Biobank analyses. BioFINDER-2 cohort The Swedish BioFINDER-2 study ( NCT03174938 ; https://biofinder.se ) was one of the data contribution sites in GNPC with available detailed AD phenotyping data, e.g., Aβ data, including CSF Aβ42/Aβ40 ratio and Aβ PET. We used BioFINDER-2 plasma SomaLogic data to validate the plasma proteomics of APOE ( APOE4 and APOE2 ) and explore the role of these proteins in APOE -Aβ associations. Focused on Aβ-related pathological changes across AD progression, we included 846 CU, 316 MCI, and 259 AD individuals with genotyping and SomaLogic proteomics measurements in plasma, resulting in 1,421 individuals (aged 20-93, median age 72) in total. Other dementia types that might introduce confounding neurodegenerative pathways were excluded. The details of these sub-cohorts have been described previously, including the diagnostic criteria 57 and cognitive staging 58 . Aβ status (positive or negative) were defined using CSF Aβ42/Aβ40 ratio (CSF Aβ42 and Aβ40 were measured using the Roche Diagnostics Elecsys® CSF electrochemiluminescence immunoassay and Roche NeuroToolKit respectively) as described before 59 – 61 . The initial processing of proteomics data was the same as the GNPC dataset. We also explored and compared APOE -associated CSF proteins and Aβ effects on them using CSF proteomics data from OLINK Explore 3072 proximity extension assay. A total of 1,567 individuals (aged 20-93, median age 72) were included. Like the plasma SomaLogic dataset, we only included CU and subjects diagnosed with MCI and AD, excluding individuals with a suspected neurodegenerative disease other than AD. Protein quality control was conducted as described 22 . Specifically, the normalized protein expression NPX value of each protein was compared to its Limit Of Detection (LOD), and only proteins with levels above the LOD in at least 70% of participants were retained. APOE-proteins were also excluded. Resulting in 1,391 measurements (tested and reported as independent proteins as in all cohorts. 1,382 unique proteins, Supplement Text and Figures) in total. The mean protein level for 1,391 proteins was calculated for each included subject and used as a covariate in all models 62 . In order to investigate the heterozygosity of protein measurements across multiple platforms, in the BioFINDER-2 cohort, plasma Nfl measured using the Quanterix Simoa® NfL assay, CSF NfL level measured using Roche Elecsys NeuroToolKit (NTK) assay were utilized and were compared to plasma SomaLogic and CSF OLINK measurements. Ethically, written informed consent was obtained from all participants prior to entering the study. Ethical approval was obtained from the Regional Ethical Committee in Lund, Sweden. ADNI APOE -protein profiles in CSF and proteins mediate APOE ’s effect on Aβ were validated in ADNI. ADNI is a multisite, longitudinal study launched in 2003 as a public-private partnership, with the primary aim of determining whether serial MRI, PET imaging, fluid biomarkers, and clinical/neuropsychological assessments can be integrated to track the progression of MCI and early AD. Further details about the study are available at www.adni-info.org . A total of 666 CU, MCI, AD dementia individuals with available APOE genotyping and CSF proteomic profiles measured using the SomaScan 7K platform (v4.1) were included. Aβ status was defined using Aβ-PET, as previously described 63 . The APOE protein itself was excluded from downstream analyses, resulting in a final dataset comprising 7,001 quantified proteins. SomaLogic proteomics data were processed similarly as described above. UK Biobank In order to validate and compare APOE -associated protein profiles across tissues and proteomics platforms, APOE -associated proteins were identified in data from the UK Biobank dataset 64 . This large-scale, multicenter prospective cohort study included approximately 500,000 participants aged 40 to 69 years, who were enrolled between 2006 and 2010. For more information on study methods and data collection, see the UKB online protocol ( www.ukbiobank.ac.uk ). 4,820 participants were included, all of whom had genotype data and protein levels in plasma measured using OLINK Explore 3072. The same protein retention criteria as for BioFINDER-2 OLINK dataset were applied and 1,319 proteins were ultimately included. PPMI To explore longitudinal APOE -protein association, we utilized data from the Parkinson’s Progression Markers Initiative (PPMI), which is an ongoing, international, observational study conducted across the United States, Europe, Israel, and Australia. To date, the study has enrolled approximately 4,000 participants, including healthy controls (HC), de novo Parkinson’s Disease (PD) patients, prodromal individuals (aged 60 or older with a dopamine transporter [DAT] deficit and either REM sleep behavior disorder [RBD] or hyposmia), and non-manifesting carriers of LRRK2 and GBA gene mutations. Participants undergo in-depth clinical assessments, neuroimaging, and molecular phenotyping. The study was approved by the Institutional Review Board at each site, and participants provided written informed consent. For this study, we utilized the curated dataset from the PPMI sub-study Project 9000. Briefly, Project 9000 includes PD patients, prodromal individuals, and healthy control participants. Longitudinal proteomics profiling was performed on CSF samples using the OLINK Explore platform. A full study protocol is available at: https://ida.loni.usc.edu/download/files . 260 participants with available CSF proteomics data were included, including 74 PD patients, 56 prodromal individuals, and 130 healthy controls. Only proteins for which at least 70% of participants had levels above the LOD were retained as described above, resulting in 826 proteins. Additionally, the mean overall protein levels were also calculated and included as a nuisance covariate in all models. Main statistical models Proteins associated with APOE4 allele carrier status (binary, carriers [ε3ε4 and ε4ε4] vs. ε3ε3 carriers) were analyzed using linear model 1. Proteins associated with AD diagnosis (binary, CU vs. AD) in GNPC or with Aβ status (binary, Aβ- vs. Aβ+) defined by CSF Aβ42/Aβ40 ratio in the BioFINDER-2 cohort and by Aβ-PET in the ADNI cohort were analyzed using linear model 2. A third model incorporated both APOE4 carrier and AD diagnosis (or Aβ status) as independent variables to account for the confounding effect of the APOE4 -AD or APOE4 -Aβ association: Model 1: Protein = β1 + β2 APOE4 + covariates + ɛ Model 2: Protein = β3 + β4 AD diagnosis (or Aβ) + covariates + ɛ Model 3: Protein = β5 + β6 APOE4 + β7 AD diagnosis (or Aβ) + covariates + ɛ Covariates included age, sex, and mean overall protein level. Data contributor sites were included as an extra covariate for the GNPC cohort. In order to investigate the protein profile of APOE2 , we used the same 3 models on individuals containing only ε2 carriers (ε2ε2 and ε2ε3 carriers) compared to ε3ε3 carriers. In the 3 models, APOE4 (ε3ε4+ε4ε4 vs. ε3ε3) was changed to APOE2 (ε2ε2+ε2ε3 vs. ε3ε3), resulting in: Model 1: Protein = β1 + β2 APOE2 + covariates + ɛ Model 3: Protein = β5 + β6 APOE 2 + β7 AD diagnosis (or Aβ) + covariates + ɛ Covariates are the same as APOE4 analysis. To investigate early APOE effects, Model 1 was applied within the CU and Aβ-subgroups. To assess potential age-dependent variation in these associations, participants were stratified by median age into younger and older subgroups. Additionally, an APOE *age interaction term was added to Model 1 within CU and Aβ-groups to statistically evaluate age-related modification of APOE -protein associations. To investigate APOE -proteins associations change across different diagnostic or Aβ status groups, we also conducted separate analyses using Model 1 for the AD dementia (or Aβ+) groups. Similarly, an APOE *AD dementia diagnosis (or APOE *Aβ status) interaction term in Model 3 for the whole cohort was utilized to statistically assess whether the effects of APOE on proteins differ significantly between CU and AD individuals or between Aβ- and Aβ+ individuals. To evaluate whether protein levels in CU and MCI APOE4 carriers would alter their Aβ positivity rates, in BioFINDER-2 plasma SomaLogic cohort, we fit interaction models of the form: Aβ status = β1 + β2 APOE4 + β3 protein + β4 ( APOE4 *protein) + covariates + ɛ. Covariates included age, sex, and mean overall protein level, full statistical analysis available in Supplement Table 2. For proteins showing significant interaction with APOE4 , we performed simple slopes analysis and calculated Johnson-Neyman intervals using the interactions R package (v1.2.0). This allowed us to identify the range of protein levels over which the effect of APOE4 on Aβ status was statistically significant (α = 0.05), providing a more nuanced understanding of effect modification. In the simple slopes plot, the effect of APOE4 on Aβ positivity was estimated at three protein expression levels: the mean, and ±1 standard deviation, with the slope representing the strength and direction of the association at each level. All regression analyses were performed in R V.4.4.2. Graphs of individual protein levels grouped by APOE genotypes are the residuals after eliminating the effect of covariates on proteins using a regression model, statistical details can be found at Supplement Table 6. All statistical tests were two-sided, p values were adjusted using the Benjamini–Hochberg (FDR) correction method (for all multi-testing analysis), and significance was reported using FDR-corrected p values. Data visualizations were generated using the R package ggplot2 or the seaborn Python package V 0.13.2. Mediation analysis To further explore the role of APOE -related proteins in AD clinical diagnosis or Aβ pathology, and to identify APOE ( APOE4 or APOE2 ) -related proteins affected by AD pathology or Aβ pathology, we tested them in different statistical mediation models. The mediation models were used to systematically classify APOE -related proteins into two protein groups: proteins that mediate the effects of APOE on AD clinical diagnosis or Aβ pathology (upstream mediation path: APOE => protein => AD diagnosis or Aβ status) and proteins that are affected by APOE through an AD-mediated pathway (downstream mediation path: APOE => AD diagnosis or Aβ status => protein). Our mediation study follows the Baron and Kenny’s Causal-Steps Test requirement, e.g. there is a significant correlation between the independent and dependent variables, i.e. only proteins associated with APOE in model 1 in the whole cohort were further tested for mediation models. Specifically, in the upstream mediation path, APOE was the exposure, AD diagnosis or Aβ status was the outcome, and APOE -associated proteins in model 1 (after FDR correction) were potential mediators. In the downstream mediation path, APOE was the exposure, APOE -associated proteins in model 1 (after FDR correction) were outcomes, and AD diagnosis or Aβ status was a potential mediator. Mediation analysis was performed using the “mediation” (V 4.5.0) R package where mediation proportions were estimated using causal mediation analysis with 95% confidence intervals based on 1,000 bootstrapping replicates. Direct effect, indirect effect and mediation proportion were extracted directly from model results and p values for these metrics were FDR corrected. BINN-enriched reactome pathway analysis Biologically informed neural networks(BINNs)-enriched pathway analysis 65 was used to globally annotate APOE -associated proteins that are altered already in CU or Aβ and its associated functionality. Specifically, based on the Reactome database (directed graph), the tool extracts the pathway hierarchy associated with the input protein and converts it into a neural network-like hierarchy (input APOE associated proteins → pathway layer → biological process layer → AD diagnosis or Aβ status). Finally, the Deep SHAP algorithm was used to calculate node importance and quantify the contribution of proteins/pathways to classification. The analysis is implemented in the binn Python package (V0.1.1). Proteins associated with APOE4 or APOE2 in both the whole cohort and CU (in GNPC, or Aβ- in BioFINDER-2 and ADNI) were conducted as input nodes separately. For proteins measured using multiple aptamers, we only kept the one with the smallest adjusted p value for APOE in model 1. In GNPC, the BINN model was trained to predict AD diagnosis. In BioFINDER-2 and ADNI, the model was trained to predict Aβ positivity in CU and MCI individuals to focus on early pathological changes; in ADNI, AD patients were included due to limited sample size. Data were split into training (80%) and validation (20%) sets, using a four-layer architecture trained over 100 epochs (500 epochs in ADNI due to smaller sample size). Subdivision of APOE associated proteins To functionally stratify APOE -associated proteins, we categorized them into four groups based on their associations with AD diagnosis and mediation relationships: Mediator proteins: Proteins showing significant indirect effects in the upstream mediation pathway ( APOE => protein => AD diagnosis), but not in the reverse ( APOE => AD diagnosis => protein), or for which the mediation proportion was stronger in the upstream than in the downstream direction (when both were significant after FDR correction). These proteins likely reflect early, genotype-driven molecular changes. Pathology-mediated proteins: Proteins meeting the reverse pattern—i.e., stronger or exclusive mediation in the downstream pathway—suggesting they are regulated secondarily by AD. APOE -specific proteins: Proteins significantly associated with APOE but not with AD diagnosis, potentially representing APOE -driven molecular changes independent of canonical AD diagnosis. Proteins falling into both this and mediation-related groups were classified according to their mediation status. Non-specific proteins: Proteins associated with both APOE and AD, but not meeting mediation criteria in either direction. For reference, we also defined a set of AD-specific proteins—those associated with AD diagnosis but not with APOE genotype—as putative markers of APOE -independent disease processes (Supplementary Table 2). In analyses focused on Aβ pathology, classification of APOE -associated proteins into the four functional categories was performed analogously, based on their associations with Aβ status and respective mediation pathways. Linear Discriminant Analysis To assess whether the identified functional categories of APOE -associated proteins correspond to distinct molecular profiles, we evaluated their separability using Linear Discriminant Analysis (LDA) in the GNPC cohort. Specifically, an LDA model was trained using the five predefined categories—mediator proteins, pathology-mediated proteins, APOE -specific proteins, non-specific proteins, and AD-specific proteins—as class labels. Feature vectors were constructed from protein expression residuals after adjusting for age, sex, mean protein level, and data collection site, to isolate APOE and AD related variance. The model, implemented using the scikit-learn Python package (v1.3.1), was trained on this labeled subset and used to derive two linear discriminants (LD1 and LD2) capturing maximal between-class variance for visualization. After training, the entire proteome (7,285 proteins) was projected into the learned discriminant space to evaluate whether proteins in different functional categories formed separable clusters. Cluster separation and within-group compactness were qualitatively assessed in the resulting 2D projection. Cell type and functional enrichment analysis To further investigate the potential cellular and function origins of APOE -associated proteomic changes, we performed cell-type enrichment analysis for genes corresponding to proteins in each of the four functional categories defined above. For cell-type enrichment analysis, we used two independent single-cell RNA sequencing (scRNA-seq) datasets: 1) 2.3 million single-cell transcriptomes from the aged human prefrontal cortex, collected from 427 participants in the Religious Orders Study (ROS) and Rush Memory and Aging Project (MAP) 66 , spanning neuronal, glial, and vascular populations; and 2) the Human Brain Vascular Atlas 67 , focused specifically on vascular and perivascular cell types, including 143,793 single-cell transcriptomes from the hippocampus and cortex of eight post-mortem human brains. For both datasets, we obtained the Seurat objects and processed them using the R package Seurat (version 4.3.0) 68 . We applied the AverageExpression function to compute the average gene expression levels for each cell type. Subsequently, we calculated the proportion of gene expression distributed across major neuronal and non-neuronal cell populations. To assess whether specific protein sets showed preferential expression in particular cell types, we computed the mean expression levels of the significant protein within each protein group across all cell types. To evaluate whether the observed enrichment exceeded what would be expected by chance, we generated 10,000 random gene lists sampled from the SomaLogic background set, each matched in size to the significant list. Using these random sets, we constructed a probability distribution of average expression levels across cell types. This approach allowed us to estimate the probability that the observed enrichment was greater than random expectation, accounting for background expression variability. Cell types were considered significantly enriched if the false discovery rate (pFDR), corrected using the Benjamini–Hochberg method, was below 0.05. Functional enrichment analyses were performed using Gene Ontology (GO) Biological Process (BP) terms. We conducted an overrepresentation analysis with the enrichGO() function from the clusterProfiler package 69 (v4.6.2) and org.Hs.eg.db annotation 70 database (v3.16.0) in R. As individual significant protein groups did not yield enrichment at a false discovery rate (pFDR) threshold of 0.05, and to maximize biological interpretability by capturing the broader functional context of the identified proteins, we conducted functional enrichment analysis on two complementary protein sets: (1) Set 1, consisting of proteins significantly upregulated or downregulated by APOE within each protein group; and (2) Set 2, consisting of proteins significantly upregulated or downregulated by APOE within each protein group together with their first-degree protein-protein interaction (PPI) partners. First-degree PPI partners were identified using the InWeb database 71 of measured and inferred physical interactions, restricting analyses to high-confidence (’gold-standard’) interactions. The background set consisted of all proteins quantified by the SomaLogic platform combined with the first-degree protein-protein interaction partners of the significant proteins identified in our study. To increase the robustness of the results and minimize the likelihood of spurious findings, we focused on biological processes that were enriched in both Set 1 and Set 2. We selected nominally significant terms (nominal p<0.05) from Set 1 and FDR-corrected significant terms (pFDR<0.05) from Set 2, identified overlapping biological processes between the two sets, reduced redundancy among GO terms, and manually summarized the top processes into major biological processes. All significant overlapping biological processes are reported in Supplementary Table 4. Finally, protein-protein interaction networks were constructed by querying the STRING database separately for the full list of significant proteins associated with APOE4 or APOE2 , using a minimum combined score threshold of 0.7 to ensure high-confidence interactions. Sensitivity analyses While primary analyses of the GNPC and BioFINDER-2 cohorts were conducted independently—with GNPC focused on clinical AD diagnosis and BioFINDER-2 on Aβ pathology—we noted partial sample overlap between the two cohorts. To ensure fair cross-cohort comparisons and avoid artificial inflation of concordance, we performed a sensitivity analysis in which overlapping individuals were excluded from the GNPC dataset prior to comparison with BioFINDER-2 results. Association and mediation analyses were then re-conducted in the reduced GNPC sample, and the results were compared to those from BioFINDER-2 to assess the robustness and consistency of findings across non-overlapping populations (Supplement Fig. 8A, B; Supplement Table 2). As for the BioFINDER-2 cohort, we utilized available SNP data and performed principal component analysis (PCA) using PLINK v2.0 to account for potential population stratification. The top 10 principal components (PCs) were calculated based on genome-wide SNP data (with LD filtering, MAF threshold and other pre-processing steps). Among them, the first five PCs (PC1–PC5), which captured the major axes of genetic variation, were included as covariates for BioFINDER-2 data in a separate analysis to correct for population structure. Results for sensitive analysis are available in Supplement text and figures. Regional gene expression analysis To further investigate the relation between APOE and related proteins, we performed region-wise association between gene expression of the proteins related to APOE4 or APOE2 , and regional gene expression of APOE or regional Aβ-PET, based on the Schaefer atlas with 200 parcels. Gene expression was generated using the regional microarray expression data obtained from 6 post-mortem brains provided by the Allen Human Brain Atlas 72 ( https://human.brain-map.org ). Data were processed with the abagen toolbox version 0.1.3 72 – 74 ( https://abagen.readthedocs.io/en/stable/index.html ), a Python-based package to access and work with the Allen Brain data microarray expression data. To ensure analyses were not driven by statistical autocorrelations, we performed null modeling with BrainSMASH ( https://brainsmash.readthedocs.io/en/latest/index.html ), a Python-based package for statistical testing of spatially autocorrelated brain maps 75 . BrainSMASH simulates surrogate brain maps with spatial autocorrelation that matches the spatial autocorrelation of the original brain map of interest, in this case the APOE or the Aβ-PET map. For both software, we used the default parameters and followed the same steps as described previously 22 . Non-parametric p -values were computed, corresponding to the frequency that correlation with the surrogate maps exceeded the observed correlation with the original gene expression map. APOE4 - or APOE2 -related proteins were considered to have gene expression profiles significantly related to APOE when BrainSMASH-corrected p-values after FDR correction was smaller than 0.05. Longitudinal analysis To explore how proteomic trajectories change over time in APOE4 or APOE2 carriers. We examined the interaction between time and APOE status in the PPMI cohort (CSF OLINK), adjusting for age, sex, mean overall protein levels, and the diagnostic group using linear mixed-effects (LME) models. Separate models were fitted to assess the interactive effects of time with APOE4 (comparing ε4 carriers to ε3ε3 carriers) or APOE2 (comparing ε2 carriers to ε3ε3 carriers) excluding ε2ε4 carriers. These analyses were conducted across all 826 proteins, full results are available in Supplement Table 5. In the main text, we focused on a subset of proteins associated with APOE in GNPC or in BioFINDER-2 CSF OLINK baseline analyses. Data availability GNPC ( https://www.neuroproteome.org/ ) will be open-access upon the publication of this study. ADNI, UKB and PPMI data used in this manuscript are publicly available from the ADNI database ( adni.loni.usc.edu ), UKB database ( https://www.ukbiobank.ac.uk/ ) and PPMI database ( https://www.ppmi-info.org/ ) on registration and compliance with the data use agreement. GWAS summary data was publicly available in https://www.ebi.ac.uk/gwas/ . Pseudonymized data from the BioFINDER study (Principal Investigator: O.H.) are available to qualified academic researchers upon request, specifically for the purpose of replicating the analyses reported in this study. In accordance with the EU General Data Protection Regulation (GDPR), access to these data requires a data transfer agreement with Skåne University Hospital (Region Skåne), which outlines provisions for secure data handling, storage, and usage. All proposed analyses must comply with the ethical approvals granted by the Swedish Ethical Review Authority. This controlled access ensures the confidentiality of study participants, who did not consent to unrestricted public data sharing, and ensures that data use remains consistent with the scope of the original ethical approvals. Code Availability Essential code for the manuscript can be found at https://github.com/Lina0125/APOE_proteomics . Data Availability GNPC ( https://www.neuroproteome.org/ ) will be open-access upon the publication of this study. ADNI, UKB and PPMI data used in this manuscript are publicly available from the ADNI database (adni.loni.usc.edu), UKB database ( https://www.ukbiobank.ac.uk/ ) and PPMI database ( https://www.ppmi-info.org/ ) on registration and compliance with the data use agreement. GWAS summary data was publicly available in https://www.ebi.ac.uk/gwas/ . Pseudonymized data from the BioFINDER study (Principal Investigator: O.H.) are available to qualified academic researchers upon request, specifically for the purpose of replicating the analyses reported in this study. In accordance with the EU General Data Protection Regulation (GDPR), access to these data requires a data transfer agreement with Skane University Hospital (Region Skane), which outlines provisions for secure data handling, storage, and usage. All proposed analyses must comply with the ethical approvals granted by the Swedish Ethical Review Authority. This controlled access ensures the confidentiality of study participants, who did not consent to unrestricted public data sharing, and ensures that data use remains consistent with the scope of the original ethical approvals. https://github.com/Lina0125/APOE_proteomics Ethics disclosures Conflicts of interest NMC has received speaker/consultancy fees from Biogen, Eli Lilly, Owkin and Merck. OH is an employee of Lund University and Eli Lilly. SP has acquired research support (for the institution) from Avid and ki elements through ADDF. In the past 2 years, he has received consultancy/speaker fees from Bioartic, Biogen, Eisai, Eli Lilly, Novo Nordisk, and Roche. BM and AV are employees of Abbvie. JWV has received consultancy fees from Manifest Technologies. Acknowledgments This research has been conducted using the UK Biobank Resource under Application Number 105777. The BioFINDER study was supported by the National Institute of Aging (R01AG083740), European Research Council (ADG-101096455), Alzheimer’s Association (ZEN24-1069572, SG-23-1061717, ALZSI-26-1523522), GHR Foundation, Swedish Research Council (2022-00775, 2021-02219, 2019-2024), ERA PerMed (ERAPERMED2021-184), Knut and Alice Wallenberg foundation (2022-0231), Strategic Research Area MultiPark (Multidisciplinary Research in Parkinson’s disease) at Lund University, Swedish Alzheimer Foundation (AF-980907, AF-994229, AF-1011949, AF-1011799), Swedish Brain Foundation (FO2021-0293, FO2023-0163, FO2024-0284, FO2025-0055), WASP and DDLS Joint call for research projects (WASP/DDLS22-066), Parkinson foundation of Sweden (1412/22), Cure Alzheimer’s fund, Rönström Family Foundation, Berg Family, Bundy Academy, Konung Gustaf V:s och Drottning Victorias Frimurarestiftelse, Skåne University Hospital Foundation (2020-O000028), Regionalt Forskningsstöd (2022-1259), Michael J Fox Foundation (MJFF-025507), Lilly Research Award Program and Swedish federal government under the ALF agreement (2022-Projekt0080, 2022-Projekt0107). Author JWV was supported by the SciLifeLab & Wallenberg Data Driven Life Science Program (grant: KAW 2020.0239) and the Swedish Research Council (2024-03642). The precursor of 18F-flutemetamol was sponsored by GE Healthcare. The precursor of 18F-RO948 was provided by Roche. The funding sources had no role in the design and conduct of the study; in the collection, analysis, interpretation of the data; or in the preparation, review, or approval of the manuscript. Footnotes Section on abstract updated to clarify usage and findings of each cohort; Acknowledgement updated; Supplemental figures and texts were updated. References 1. ↵ Corder , E. H. et al. Gene dose of apolipoprotein E type 4 allele and the risk of Alzheimer’s disease in late onset families . Science 261 , 921 – 923 ( 1993 ). OpenUrl Abstract / FREE Full Text 2. ↵ Genin , E. et al. APOE and Alzheimer disease: a major gene with semi-dominant inheritance . Mol. Psychiatry 16 , 903 – 907 ( 2011 ). OpenUrl CrossRef PubMed Web of Science 3. ↵ Liu , C.-C. et al. ApoE4 accelerates early seeding of amyloid pathology . Neuron 96 , 1024 – 1032 .e3 ( 2017 ). OpenUrl CrossRef PubMed 4. Holtzman , D. M. et al. Apolipoprotein E isoform-dependent amyloid deposition and neuritic degeneration in a mouse model of Alzheimer’s disease . Proc. Natl. Acad. Sci. U. S. A . 97 , 2892 – 2897 ( 2000 ). OpenUrl Abstract / FREE Full Text 5. ↵ Mattsson , N. et al. Prevalence of the apolipoprotein E ε4 allele in amyloid β positive subjects across the spectrum of Alzheimer’s disease . Alzheimers. Dement . 14 , 913 – 924 ( 2018 ). OpenUrl PubMed 6. ↵ Reiman , E. M. et al. Exceptionally low likelihood of Alzheimer’s dementia in APOE2 homozygotes from a 5,000-person neuropathological study . Nat. Commun . 11 , 667 ( 2020 ). OpenUrl CrossRef PubMed 7. Serrano-Pozo , A. , Das , S. & Hyman , B. T . APOE and Alzheimer’s disease: advances in genetics, pathophysiology, and therapeutic approaches . Lancet Neurol . 20 , 68 – 80 ( 2021 ). OpenUrl CrossRef PubMed 8. ↵ Rebeck , G. W. , Reiter , J. S. , Strickland , D. K. & Hyman , B. T . Apolipoprotein E in sporadic Alzheimer’s disease: allelic variation and receptor interactions . Neuron 11 , 575 – 580 ( 1993 ). OpenUrl CrossRef PubMed Web of Science 9. ↵ Goldberg , T. E. , Huey , E. D. & Devanand , D. P . Association of APOE e2 genotype with Alzheimer’s and non-Alzheimer’s neurodegenerative pathologies . Nat. Commun . 11 , 4727 ( 2020 ). OpenUrl CrossRef PubMed 10. Insel , P. S. , Hansson , O. & Mattsson-Carlgren , N . Association between apolipoprotein E ε2 vs ε4, age, and β-amyloid in adults without cognitive impairment . JAMA Neurol . 78 , 229 – 235 ( 2021 ). OpenUrl PubMed 11. ↵ Cicognola , C. et al. APOE4 impact on soluble and insoluble tau pathology is mostly influenced by amyloid-beta . Brain ( 2025 ) doi: 10.1093/brain/awaf016 . OpenUrl CrossRef 12. ↵ Vélez , J. I. et al. APOE*E2 allele delays age of onset in PSEN1 E280A Alzheimer’s disease . Mol. Psychiatry 21 , 916 – 924 ( 2016 ). OpenUrl CrossRef PubMed 13. ↵ Faraji , P. , Kühn , H. & Ahmadian , S . Multiple roles of apolipoprotein E4 in oxidative lipid metabolism and ferroptosis during the pathogenesis of Alzheimer’s disease . J. Mol. Neurosci . 74 , 62 ( 2024 ). OpenUrl PubMed 14. ↵ Yang , L. G. , March , Z. M. , Stephenson , R. A. & Narayan , P. S . Apolipoprotein E in lipid metabolism and neurodegenerative disease . Trends Endocrinol. Metab . 34 , 430 – 445 ( 2023 ). OpenUrl CrossRef PubMed 15. ↵ Area-Gomez , E. et al. APOE4 is associated with differential regional vulnerability to bioenergetic deficits in aged APOE mice . Sci. Rep . 10 , 4277 ( 2020 ). OpenUrl CrossRef PubMed 16. ↵ Pires , M. & Rego , A. C . Apoe4 and Alzheimer’s disease pathogenesis-mitochondrial deregulation and targeted therapeutic strategies . Int. J. Mol. Sci . 24 , 778 ( 2023 ). OpenUrl PubMed 17. ↵ Castellano , J. M. et al. Human apoE isoforms differentially regulate brain amyloid-β peptide clearance . Sci. Transl. Med . 3 , 89ra57 ( 2011 ). OpenUrl Abstract / FREE Full Text 18. ↵ Tai , L. M. et al. Soluble apoE/Aβ complex: mechanism and therapeutic target for APOE4-induced AD risk . Mol. Neurodegener . 9 , 2 ( 2014 ). 19. ↵ Tai , L. M. et al. APOE-modulated Aβ-induced neuroinflammation in Alzheimer’s disease: current landscape, novel data, and future perspective . J. Neurochem . 133 , 465 – 488 ( 2015 ). OpenUrl CrossRef PubMed 20. ↵ Parhizkar , S. & Holtzman , D. M . APOE mediated neuroinflammation and neurodegeneration in Alzheimer’s disease . Semin. Immunol . 59 , 101594 ( 2022 ). OpenUrl CrossRef PubMed 21. ↵ Higginbotham , L. et al. Integrated proteomics reveals brain-based cerebrospinal fluid biomarkers in asymptomatic and symptomatic Alzheimer’s disease . Sci. Adv . 6 , eaaz9360 ( 2020 ). OpenUrl FREE Full Text 22. ↵ Pichet Binette , A., et al. Proteomic changes in Alzheimer’s disease associated with progressive Aβ plaque and tau tangle pathologies . Nat. Neurosci . 27 , 1880 – 1891 ( 2024 ). OpenUrl CrossRef PubMed 23. ↵ Frick , E. A. et al. Serum proteomics reveal APOE-ε4-dependent and APOE-ε4-independent protein signatures in Alzheimer’s disease. Nat . Aging 4 , 1446 – 1464 ( 2024 ). OpenUrl PubMed 24. ↵ Kaur , G. , Poljak , A. , Masters , C. L. , Fowler , C. & Sachdev , P . Impact of APOE ε3 and ε4 genotypes on plasma proteome signatures in Alzheimer’s disease . Neuroscience ( 2022 ). 25. ↵ Shvetcov , A. et al. Proteome profiling of cerebrospinal fluid using machine learning shows a unique protein signature associated with APOE4 genotype . Aging Cell 24 , e14439 ( 2025 ). OpenUrl PubMed 26. ↵ Oveisgharan , S. et al. Proteins linking APOE ɛ4 with Alzheimer’s disease . Alzheimers. Dement . 20 , 4499 – 4511 ( 2024 ). OpenUrl PubMed 27. ↵ Dai , J. et al. Effects of APOE genotype on brain proteomic network and cell type changes in Alzheimer’s disease . Front. Mol. Neurosci . 11 , 454 ( 2018 ). OpenUrl CrossRef PubMed 28. ↵ Lambert , J. C. et al. Meta-analysis of 74,046 individuals identifies 11 new susceptibility loci for Alzheimer’s disease . Nat. Genet . 45 , 1452 – 1458 ( 2013 ). OpenUrl CrossRef PubMed 29. ↵ Jansen , I. E. et al. Genome-wide meta-analysis for Alzheimer’s disease cerebrospinal fluid biomarkers . Acta Neuropathol . 144 , 821 – 842 ( 2022 ). OpenUrl CrossRef PubMed 30. ↵ Cui , F. , Xu , Z. , Lv , Y. & Hu , J . Role of spindle pole body component 25 in neurodegeneration . Ann. Transl. Med . 9 , 1432 ( 2021 ). OpenUrl PubMed 31. ↵ Shim , S.-M. , Kim , J.-H. & Jeon , J.-P . Effective litmus gene test for monitoring the quality of blood samples: Application to Alzheimer’s disease diagnostics . Sci. Rep . 7 , 16848 ( 2017 ). OpenUrl PubMed 32. ↵ Su , Y. et al. S100A13 promotes senescence-associated secretory phenotype and cellular senescence via modulation of non-classical secretion of IL-1α . Aging (Albany NY) 11 , 549 – 572 ( 2019 ). OpenUrl PubMed 33. ↵ Granic , A. & Potter , H . Mitotic spindle defects and chromosome mis-segregation induced by LDL/cholesterol-implications for Niemann-Pick C1, Alzheimer’s disease, and atherosclerosis . PLoS One 8 , e60718 ( 2013 ). OpenUrl PubMed 34. Moh , C. et al. Cell cycle deregulation in the neurons of Alzheimer’s disease . Results Probl. Cell Differ . 53 , 565 – 576 ( 2011 ). OpenUrl CrossRef PubMed Web of Science 35. Judge , M. , Hornbeck , L. , Potter , H. & Padmanabhan , J . Mitosis-specific phosphorylation of amyloid precursor protein at threonine 668 leads to its altered processing and association with centrosomes . Mol. Neurodegener . 6 , 80 ( 2011 ). OpenUrl CrossRef PubMed 36. ↵ Joseph , C. et al. Cell cycle deficits in neurodegenerative disorders: Uncovering molecular mechanisms to drive innovative therapeutic development . Aging Dis . 11 , 946 – 966 ( 2020 ). OpenUrl CrossRef PubMed 37. ↵ McMurray , C. T . Neurodegeneration: diseases of the cytoskeleton? Cell Death Differ . 7 , 861 – 865 ( 2000 ). OpenUrl CrossRef PubMed Web of Science 38. ↵ Gutiérrez-Vargas , J. A. , Castro-Álvarez , J. F. , Zapata-Berruecos , J. F. , Abdul-Rahim , K. & Arteaga-Noriega , A . Neurodegeneration and convergent factors contributing to the deterioration of the cytoskeleton in Alzheimer’s disease, cerebral ischemia and multiple sclerosis (Review) . Biomed. Rep . 16 , 27 ( 2022 ). OpenUrl PubMed 39. ↵ Li , W. , Kui , L. , Demetrios , T. , Gong , X. & Tang , M . A glimmer of hope: Maintain mitochondrial homeostasis to mitigate Alzheimer’s disease . Aging Dis . 11 , 1260 – 1275 ( 2020 ). OpenUrl CrossRef PubMed 40. Yuan , Y. , Zhao , G. & Zhao , Y . Dysregulation of energy metabolism in Alzheimer’s disease . J. Neurol . 272 , 2 ( 2024 ). OpenUrl CrossRef PubMed 41. Selfridge , J. E. , E , L ., Lu , J. & Swerdlow , R. H . Role of mitochondrial homeostasis and dynamics in Alzheimer’s disease . Neurobiol. Dis . 51 , 3 – 12 ( 2013 ). OpenUrl CrossRef PubMed 42. ↵ Tao , R. et al. Mendelian Randomization combined with single-cell transcriptome analysis reveals the role of the key gene PCLAF in the pathogenesis of atopic dermatitis . Clin. Cosmet. Investig. Dermatol . 18 , 867 – 882 ( 2025 ). OpenUrl PubMed 43. ↵ Wojtas , A. M. et al. Proteomic changes in the human cerebrovasculature in Alzheimer’s disease and related tauopathies linked to peripheral biomarkers in plasma and cerebrospinal fluid . Alzheimers. Dement . 20 , 4043 – 4065 ( 2024 ). OpenUrl CrossRef PubMed 44. Durrant , C. S. , Ruscher , K. , Sheppard , O. , Coleman , M. P. & Özen , I . Beta secretase 1-dependent amyloid precursor protein processing promotes excessive vascular sprouting through NOTCH3 signalling . Cell Death Dis . 11 , 98 ( 2020 ). OpenUrl CrossRef PubMed 45. ↵ Badimon , A. , Torrente , D. & Norris , E. H . Vascular dysfunction in Alzheimer’s disease: Alterations in the plasma contact and fibrinolytic systems . Int. J. Mol. Sci . 24 , ( 2023 ). 46. ↵ Zhao , L. et al. Intracerebral adeno-associated virus gene delivery of apolipoprotein E2 markedly reduces brain amyloid pathology in Alzheimer’s disease mouse models . Neurobiol. Aging 44 , 159 – 172 ( 2016 ). OpenUrl CrossRef PubMed 47. Hu , J. et al. Opposing effects of viral mediated brain expression of apolipoprotein E2 (apoE2) and apoE4 on apoE lipidation and Aβ metabolism in apoE4-targeted replacement mice . Mol. Neurodegener . 10 , 6 ( 2015 ). OpenUrl CrossRef PubMed 48. ↵ Zhang , X. , Wu , L. , Swerdlow , R. H. & Zhao , L . Opposing effects of ApoE2 and ApoE4 on glycolytic metabolism in neuronal aging supports a Warburg neuroprotective cascade against Alzheimer’s disease . Cells 12 , 410 ( 2023 ). OpenUrl 49. ↵ Ranganathan , R. , Sapozhnikov , G. , Ni , W. , Li , S. & Song , Y . Recent developments in the role of DNA damage response and understanding its implications for new therapeutic approaches in Alzheimer’s disease . Transl. Med. Aging 7 , 52 – 65 ( 2023 ). OpenUrl 50. Ning , Z. et al. APOE2 protects against Aβ pathology by improving neuronal mitochondrial function through ERRα signaling . Cell. Mol. Biol. Lett . 29 , 87 ( 2024 ). OpenUrl CrossRef PubMed 51. ↵ Lee , C. Y. D. , Tse , W. , Smith , J. D. & Landreth , G. E . Apolipoprotein E promotes β-amyloid trafficking and degradation by modulating microglial cholesterol levels . J. Biol. Chem . 287 , 2032 – 2044 ( 2012 ). OpenUrl Abstract / FREE Full Text 52. ↵ Li , Z. , Shue , F. , Zhao , N. , Shinohara , M. & Bu , G . APOE2: protective mechanism and therapeutic implications for Alzheimer’s disease . Mol. Neurodegener . 15 , 63 ( 2020 ). OpenUrl CrossRef PubMed 53. ↵ Eldjarn , G. H. et al. Large-scale plasma proteomics comparisons through genetics and disease associations . Nature 622 , 348 – 358 ( 2023 ). OpenUrl CrossRef PubMed 54. ↵ Wang , B. et al. Comparative studies of 2168 plasma proteins measured by two affinity-based platforms in 4000 Chinese adults . Nat. Commun . 16 , 1869 ( 2025 ). OpenUrl CrossRef PubMed 55. ↵ Imam , F. et al. The Global Neurodegeneration Proteomics Consortium: biomarker and drug target discovery for common neurodegenerative diseases and aging . Nat. Med . 1 – 11 ( 2025 ). 56. ↵ Lollo , B. , Steele , F. & Gold , L . Beyond antibodies: new affinity reagents to unlock the proteome . Proteomics 14 , 638 – 644 ( 2014 ). OpenUrl CrossRef PubMed 57. ↵ Palmqvist , S. et al. Discriminative accuracy of plasma phospho-tau217 for Alzheimer disease vs other neurodegenerative disorders . JAMA 324 , 772 – 781 ( 2020 ). OpenUrl CrossRef PubMed 58. ↵ Palmqvist , S. et al. Cognitive effects of Lewy body pathology in clinically unimpaired individuals . Nat. Med . 29 , 1971 – 1978 ( 2023 ). OpenUrl CrossRef PubMed 59. ↵ Janelidze , S. et al. Detecting amyloid positivity in early Alzheimer’s disease using combinations of plasma Aβ42/Aβ40 and p-tau . Alzheimers. Dement . 18 , 283 – 293 ( 2022 ). OpenUrl CrossRef PubMed 60. Quadalti , C. et al. Clinical effects of Lewy body pathology in cognitively impaired individuals . Nat. Med . 29 , 1964 – 1970 ( 2023 ). OpenUrl CrossRef PubMed 61. ↵ Gobom , J. et al. Validation of the LUMIPULSE automated immunoassay for the measurement of core AD biomarkers in cerebrospinal fluid . Clin. Chem. Lab. Med . 60 , 207 – 219 ( 2022 ). OpenUrl CrossRef PubMed 62. ↵ Karlsson , L. HTML] from Nature.com Full View Cerebrospinal Fluid Reference Proteins Increase Accuracy and Interpretability of Biomarkers for Brain Diseases . 63. ↵ Sung , Y. J. et al. Proteomics of brain, CSF, and plasma identifies molecular signatures for distinguishing sporadic and genetic Alzheimer’s disease . Sci. Transl. Med . 15 , eabq5923 ( 2023 ). OpenUrl CrossRef PubMed 64. ↵ Miller , K. L. et al. Multimodal population brain imaging in the UK Biobank prospective epidemiological study . Nat. Neurosci . 19 , 1523 – 1536 ( 2016 ). OpenUrl CrossRef PubMed 65. ↵ Hartman , E. et al. Interpreting biologically informed neural networks for enhanced proteomic biomarker discovery and pathway analysis . Nat. Commun . 14 , 5359 ( 2023 ). OpenUrl PubMed 66. ↵ Mathys , H. et al. Single-cell atlas reveals correlates of high cognitive function, dementia, and resilience to Alzheimer’s disease pathology . Cell 186 , 4365 – 4385 .e27 ( 2023 ). OpenUrl CrossRef PubMed 67. ↵ Yang , A. C. et al. A human brain vascular atlas reveals diverse mediators of Alzheimer’s risk . Nature 603 , 885 – 892 ( 2022 ). OpenUrl CrossRef PubMed 68. ↵ Satija , R. , Farrell , J. A. , Gennert , D. , Schier , A. F. & Regev , A . Spatial reconstruction of single-cell gene expression data . Nat. Biotechnol . 33 , 495 – 502 ( 2015 ). OpenUrl CrossRef PubMed 69. ↵ Xu , S. et al. Using clusterProfiler to characterize multiomics data . Nat. Protoc . 19 , 3292 – 3320 ( 2024 ). OpenUrl CrossRef PubMed 70. ↵ org.Hs.eg.db . Bioconductor https://bioconductor.org/packages/release/data/annotation/html/org.Hs.eg.db.html . 71. ↵ Li , T. et al. A scored human protein-protein interaction network to catalyze genomic interpretation . Nat. Methods 14 , 61 – 64 ( 2017 ). OpenUrl CrossRef PubMed 72. ↵ Hawrylycz , M. J. et al. An anatomically comprehensive atlas of the adult human brain transcriptome . Nature 489 , 391 – 399 ( 2012 ). OpenUrl CrossRef PubMed Web of Science 73. Markello , R. D. , et al. Standardizing workflows in imaging transcriptomics with the abagen toolbox . bioRxiv ( 2021 ) doi: 10.1101/2021.07.08.451635 . OpenUrl Abstract / FREE Full Text 74. ↵ Arnatkeviciute , A. , Fulcher , B. D. & Fornito , A . A practical guide to linking brain-wide gene expression and neuroimaging data . Neuroimage 189 , 353 – 367 ( 2019 ). OpenUrl CrossRef PubMed 75. ↵ Burt , J. B. , Helmer , M. , Shinn , M. , Anticevic , A. & Murray , J. D . Generative modeling of brain maps with spatial autocorrelation . Neuroimage 220 , 117038 ( 2020 ). View the discussion thread. Back to top Previous Next Posted August 25, 2025. Download PDF Supplementary Material Data/Code Email Thank you for your interest in spreading the word about medRxiv. NOTE: Your email address is requested solely to identify you as the sender of this article. Your Email * Your Name * Send To * Enter multiple addresses on separate lines or separate them with commas. You are going to email the following Proteomic signatures of the APOE ε4 and APOE ε2 genetic variants and Alzheimer’s disease Message Subject (Your Name) has forwarded a page to you from medRxiv Message Body (Your Name) thought you would like to see this page from the medRxiv website. Your Personal Message CAPTCHA This question is for testing whether or not you are a human visitor and to prevent automated spam submissions. Share Proteomic signatures of the APOE ε4 and APOE ε2 genetic variants and Alzheimer’s disease Lina Lu , Alexa Pichet Binette , Ines Hristovska , Shorena Janelidze , Bart Smets , Irene Cumplido Mayoral , Aparna Vasanthakumar , Britney Milkovich , Rik Ossenkoppele , Varsha Krish , Farhad Imam , Sebastian Palmqvist , Jacob Vogel , Erik Stomrud , Oskar Hansson , Niklas Mattsson-Carlgren medRxiv 2025.08.04.25332945; doi: https://doi.org/10.1101/2025.08.04.25332945 Share This Article: Copy Citation Tools Proteomic signatures of the APOE ε4 and APOE ε2 genetic variants and Alzheimer’s disease Lina Lu , Alexa Pichet Binette , Ines Hristovska , Shorena Janelidze , Bart Smets , Irene Cumplido Mayoral , Aparna Vasanthakumar , Britney Milkovich , Rik Ossenkoppele , Varsha Krish , Farhad Imam , Sebastian Palmqvist , Jacob Vogel , Erik Stomrud , Oskar Hansson , Niklas Mattsson-Carlgren medRxiv 2025.08.04.25332945; doi: https://doi.org/10.1101/2025.08.04.25332945 Citation Manager Formats BibTeX Bookends EasyBib EndNote (tagged) EndNote 8 (xml) Medlars Mendeley Papers RefWorks Tagged Ref Manager RIS Zotero Tweet Widget Facebook Like Google Plus One Subject Area Neurology Subject Areas All Articles Addiction Medicine (568) Allergy and Immunology (863) Anesthesia (300) Cardiovascular Medicine (4436) Dentistry and Oral Medicine (444) Dermatology (382) Emergency Medicine (608) Endocrinology (including Diabetes Mellitus and Metabolic Disease) (1509) Epidemiology (15229) Forensic Medicine (30) Gastroenterology (1124) Genetic and Genomic Medicine (6600) Geriatric Medicine (668) Health Economics (997) Health Informatics (4538) Health Policy (1368) Health Systems and Quality Improvement (1613) Hematology (542) HIV/AIDS (1264) Infectious Diseases (except HIV/AIDS) (15916) Intensive Care and Critical Care Medicine (1103) Medical Education (623) Medical Ethics (146) Nephrology (667) Neurology (6599) Nursing (346) Nutrition (998) Obstetrics and Gynecology (1144) Occupational and Environmental Health (957) Oncology (3333) Ophthalmology (974) Orthopedics (369) Otolaryngology (420) Pain Medicine (436) Palliative Medicine (130) Pathology (663) Pediatrics (1693) Pharmacology and Therapeutics (691) Primary Care Research (711) Psychiatry and Clinical Psychology (5447) Public and Global Health (9232) Radiology and Imaging (2198) Rehabilitation Medicine and Physical Therapy (1370) Respiratory Medicine (1196) Rheumatology (593) Sexual and Reproductive Health (712) Sports Medicine (530) Surgery (712) Toxicology (99) Transplantation (289) Urology (265) (function(){function c(){var b=a.contentDocument||a.contentWindow.document;if(b){var d=b.createElement('script');d.innerHTML="window.__CF$cv$params={r:'a0103ad7dad809d6',t:'MTc3OTY2NjkxMQ=='};var a=document.createElement('script');a.src='/cdn-cgi/challenge-platform/scripts/jsd/main.js';document.getElementsByTagName('head')[0].appendChild(a);";b.getElementsByTagName('head')[0].appendChild(d)}}if(document.body){var a=document.createElement('iframe');a.height=1;a.width=1;a.style.position='absolute';a.style.top=0;a.style.left=0;a.style.border='none';a.style.visibility='hidden';document.body.appendChild(a);if('loading'!==document.readyState)c();else if(window.addEventListener)document.addEventListener('DOMContentLoaded',c);else{var e=document.onreadystatechange||function(){};document.onreadystatechange=function(b){e(b);'loading'!==document.readyState&&(document.onreadystatechange=e,c())}}}})();