Plasma proteomic associations with Alzheimer's disease endophenotypes

Abstract

Clinical Alzheimer's disease is currently characterized by cerebral β-amyloidosis associated with cognitive impairment. However, most cases of Alzheimer's disease are associated with multiple neuropathologies at autopsy. The peripheral protein changes associated with these disease endophenotypes are poorly understood. In this study, we analyzed the plasma proteomes of individuals from four cohorts (n = 2,139 participants) to identify proteins and pathways associated with cerebral β-amyloidosis and other neuropathologies, including tau, Lewy bodies, TDP43, cerebral amyloid angiopathy, atherosclerosis, arteriolosclerosis and infarcts as well as cognitive function. Analyses in a cohort with paired brain data showed that known neuropathologies could account for only half of proteins associated with cognitive function and that many plasma proteins associated with these neuropathologies are not strongly correlated to levels in brain, suggesting a potential contribution of peripheral factors to the development of Alzheimer's disease endophenotypes. Targeting pathways represented by these peripheral proteins may modify Alzheimer's disease risk or disease progression.

Fig. 1. Study overview.

a, Plasma samples from participants in the Bio-Hermes study, Emory GADRC, ROSMAP and Emory Other were analyzed on the SomaScan 7K proteomic platform. Assessment of cerebral amyloidosis was performed using different methods in each cohort. The resulting protein quantitative data were used in multiple downstream analyses. b, Protein associations with cerebral Aβ with and without adjustment for APOE ε4, and protein associations with cognitive function, were assessed separately in each cohort and then combined in respective meta-analyses. c, Plasma proteins associated with cerebral Aβ were compared to brain proteins associated with Aβ plaques from multiple studies as described in Askenazi et al., and plasma proteins associated with cognitive function in ROSMAP were assessed for overlap with the ROSMAP brain proteome as described in Johnson et al.. d, Plasma proteins associated with different neuropathologies in the ROSMAP cohort were identified and assessed for overlap with proteins associated with cognitive function. e, Proteins associated with risk of incident cognitive impairment in ROSMAP were identified. f, Proteins associated with different neuropathologies in ROSMAP were tested for module overlap in a four-cohort consensus plasma protein co-expression network. The plasma network was also tested for overlap with a consensus AD brain co-expression network as described in Johnson et al. and an AD CSF co-expression network as described in Dammer et al.. g, Plasma proteins associated with cerebral Aβ and cognitive function from the meta-analyses were compared to CSF proteins associated with cerebral Aβ and cognitive function as described in Dammer et al. to assess overlap of associations between compartments. HR, hazard ratio.

Fig. 2. Plasma protein associations with cerebral β-amyloidosis.

Proteins were assessed for association with cerebral Aβ in Bio-Hermes, Emory GADRC, ROSMAP and Emory Other cohorts as assessed by amyloid PET SUVR, plasma pTau217, direct silver stain for Aβ plaques and CSF Aβ42/tau, respectively. a, Associations with cerebral Aβ after meta-analysis, without (left) or with (right) adjustment for APOE ε4. Positive meta z-scores indicate higher plasma protein levels with Aβ deposition; negative z-scores indicate lower protein levels with Aβ deposition. Proteins significant at meta q < 0.05 after Benjamini–Hochberg correction are colored in red and magenta. b, Overlap of total number of proteins significantly associated with cerebral Aβ with and without APOE ε4 adjustment. c, Top 30 cerebral Aβ-associated proteins most dependent upon APOE ε4. Meta q values are shown before and after adjustment for ε4. d, Gene Ontology analysis of proteins positively (top) and negatively (bottom) associated with cerebral Aβ deposition, highlighting biological pathway (BP), molecular function (MF) and cellular compartment (CC) terms. A summary of the terms is provided to the left in red. Significance was determined by one-sided Fisher’s exact test. ROS, reactive oxygen species. e, Overlap of plasma APOE ε4-dependent cerebral Aβ-associated proteins with serum ε4-dependent AD-associated proteins from Frick et al.. f, Overlap of cerebral Aβ-associated plasma proteins with proteins significantly enriched or depleted in cerebral Aβ plaques as assessed by direct mass spectrometry methods. Proteins are colored based on pattern of association between plasma and brain plaque.

Fig. 3. Plasma protein associations with cognitive function and other brain pathologies.

a, Meta-analysis of protein associations with composite measures of cognitive function in Bio-Hermes, Emory GADRC, ROSMAP and Emory Other cohorts, without (left) or with (right) adjustment for cerebral Aβ. Positive meta z-scores indicate higher plasma protein levels are associated with better cognitive function; negative z-scores indicate higher protein levels are associated with worse cognitive function. Proteins significant at meta q < 0.05 after Benjamini–Hochberg correction are colored in red and gold. b, Overlap of proteins associated with cerebral Aβ and cognitive function. The list of overlapping proteins is provided in Supplementary Table 7. c, Proteins associated with cognitive function most dependent upon APOE ε4. Meta q values are shown before and after adjustment for ε4. d, Gene Ontology analysis of proteins positively (top) and negatively (bottom) associated with cognitive function after adjustment for Aβ levels, highlighting biological pathway, molecular function and cellular compartment terms. A summary of the terms is provided to the left in gold. Significance was determined by one-sided Fisher’s exact test. e, Overlap of proteins associated with cognitive function with proteins associated with levels in brain. The direction of association with cognitive function is shown as positive (+) or negative (−) for each overlapping protein. f, Overlap of proteins associated with different neuropathologies with proteins associated with cognitive function in ROSMAP. The gray circle in the middle represents the sum of all unique proteins associated with cognitive function and at least one neuropathology. Proteins associated with each neuropathology are provided in Supplementary Table 8, with BPs for each neuropathology association provided in Supplementary Fig. 1. g, Overlap of proteins associated with cognitive function and proteins associated with levels in brain (top) and overlap of proteins associated with cognitive function and no other neuropathology with proteins associated with levels in brain (bottom). The direction of association with cognitive function is shown as positive or negative for each overlapping protein. Proteins highlighted in red indicate significant association with cognitive function in the meta-analysis after Benjamini–Hochberg correction. ESCRT, endosomal sorting complex required for transport.

Fig. 4. Proteins associated with risk of incident cognitive impairment.

Proteins associated with risk of conversion to a cognitively impaired state in ROSMAP were determined using two-sided Cox proportional hazard analysis. No proteins survived Benjamini–Hochberg correction, and, therefore, results are reported at a nominal statistical level. a,d, Proteins associated with risk of cognitive impairment regardless of underlying etiology (a, n = 172 converters out of 310 total), with Gene Ontology analysis of proteins associated with increased (left) and decreased (right) risk (d), highlighting biological pathway, molecular function and cellular compartment terms. A summary of the terms is provided to the left. b,e, Proteins associated with risk of cognitive impairment with confirmed high Aβ and tau pathology at autopsy in those who converted (b, n = 131 out of 269 total), with Gene Ontology analysis of proteins associated with increased (left) and decreased (right) risk (e). c,f, Proteins associated with risk of cognitive impairment with confirmed high Aβ and tau pathology in converters and low Aβ and tau pathology in non-converters (c, n = 131 out of 187 total), with Gene Ontology analysis of proteins associated with increased (left) and decreased (right) risk (f). Proteins associated with risk of conversion in each analysis are provided in Supplementary Table 10. Significance of ontology enrichment was determined by one-sided Fisher’s exact test. g, Overlap of proteins associated with risk of conversion in each analysis. h, Overlap of proteins associated with risk of cognitive impairment regardless of underlying etiology with proteins associated with risk of conversion to dementia in the ARIC cohort as previously described by Walker et al. in 2021 (ref. ) and in 2023 (ref. ). i, Overlap of proteins associated with risk of cognitive impairment regardless of underlying etiology with proteins associated with risk of conversion to AD in the AGES-Reykjavik cohort as previously described by Frick et al.. Proteins in ARIC and AGES-Reykjavik were controlled for multiple comparisons. Most overlapping proteins were associated with increased (+) rather than decreased (−) risk. GPCR, G-protein coupled receptor; PRR, pattern recognition receptor; RTK, receptor tyrosine kinase.

Fig. 5. Enrichment of phenotypes in a plasma protein co-expression network.

All plasma samples (n = 2,657) across the four cohorts were used to construct a protein co-expression network that identified 27 modules associated with diverse biological pathways and processes. Modules that were ambiguous in their primary ontologies are not labeled. Proteins associated with each phenotype in meta-analyses across the cohorts or in ROSMAP-only analyses were tested for enrichment in network modules. Significance of enrichment after Benjamini–Hochberg correction is indicated by color, with a threshold for color at FDR < 0.05 (−log₁₀(FDR) > 1.30). Additional information on phenotype enrichments is provided in Extended Data Fig. 8 and Supplementary Table 13.

Fig. 6. Overlap of plasma proteins in brain and CSF compartments.

a,b, Plasma protein co-expression modules were tested for overlap with brain protein co-expression modules as described in Johnson et al. (a) and CSF protein co-expression modules as described in Dammer et al. (b). Numbers indicate −log₁₀ Benjamini–Hochberg-corrected P values for the one-tailed overlap where significant at FDR < 0.05. Modules with particularly strong overlap or interest in AD are highlighted in red. c, Enrichment of plasma proteins positively (POS), negatively (NEG) or regardless of direction (ALL) associated with cerebral β-amyloidosis, cognitive function and cognitive function adjusted for cerebral Aβ in the meta-analysis across cohorts in brain protein co-expression modules. d, Enrichment of plasma proteins as described in c in CSF protein co-expression modules. The proteins associated with cognitive function after adjustment for Aβ used for overlap testing in c and d were nominally associated at P < 0.05 to increase power of enrichment. Significance of enrichment was determined using one-sided Fisher’s exact test. Benjamini–Hochberg correction was applied for overlap tests encompassing positive, negative and all protein groups for each analysis in c and d. Numbers in c and d indicate Benjamini–Hochberg-corrected P values (FDR) for the enrichments significant at FDR < 0.05.

Fig. 7. Direction of protein associations with cerebral Aβ and cognitive function in plasma and CSF.

Plasma protein associations were compared to those in CSF as described in Dammer et al.. a, Overlap of plasma proteins positively (left) and negatively (right) associated with cerebral Aβ with CSF proteins positively and negatively associated with cerebral Aβ. Gene Ontology analysis of proteins in each overlap is provided below each Venn diagram, with a summary of terms provided in red. Significance was determined by one-sided Fisher’s exact test. There were no significant terms for the plasma (−)–CSF (+) overlap (N/A). b, The same analysis as shown in a for cognitive function.

Extended Data Fig. 1. Plasma Protein Associations with Cerebral β-Amyloidosis.

(A) (Left) Overlap of nominally significant protein associations with cerebral Aβ in each cohort. (Right) Overlap of the top 50 proteins by FDR rank after meta-analysis across the cohorts. Proteins had to be nominally significant (p < 0.05) in at least one cohort. (B) Overlap of proteins associated with cerebral Aβ deposition using a linear model or a binary cutoff approach and logistic regression without (top) and with (bottom) adjustment for APOE ε4. (C) Gene ontology (GO) analysis for proteins positively (top) and negatively (bottom) associated with cerebral Aβ levels after adjustment for APOE ε4 (BP, biological process; MF, molecular function; CC, cellular compartment). (D) GO analysis for Aβ-associated proteins dependent upon APOE ε4. (E) Associations with cerebral Aβ after meta-analysis and adjustment for APOE ε4 and cognitive function. Positive meta z-scores indicate higher plasma protein levels with Aβ deposition; negative z-scores indicate lower protein levels with Aβ deposition. Proteins significant at meta q < 0.05 after Benjamini-Hochberg (BH) correction are colored in cyan. (F) Overlap of total number of proteins significantly associated with cerebral Aβ after APOE ε4 adjustment only and after both ε4 and cognitive function adjustment. (G) Association of plasma protein levels with cerebral Aβ for those proteins in the M42 matrisome brain protein co-expression module highly associated with Aβ plaque, as previously described. Aptamers are distinguished by their aptamer code if multiple aptamers exist to the same gene product. (H) Association of plasma protein levels as measured by SomaScan with the corresponding protein in brain as measured by mass spectrometry in the ROSMAP cohort. (I) GO analysis of proteins positively (top) and negatively (bottom) associated with levels in brain. (J) Overlap of proteins associated with cerebral Aβ from the meta-analysis and proteins significantly positively or negatively associated with levels in brain in the ROSMAP cohort. The direction of association with cerebral Aβ is shown as positive (+) or negative (–) for each overlapping protein.

Extended Data Fig. 2. Classification of Aβ Positive and Negative Cases Using the Plasma Proteome.

Optimal 1/1, 2/2, 3/3, and 4/4 protein ratios to classify cerebral Aβ positive and negative individuals were identified from the top positively (numerator) and negatively (denominator) Aβ-associated proteins. Ratio membership is provided in Supplementary Table 6. (A) The best classifier for each ratio derived from the meta-analysis results in the Bio-Hermes cohort. Amyloid positivity was assessed by examiner interpretation of amyloid PET scan. Plasma pTau217 (Lilly) in Bio-Hermes is shown for comparison. n = 933. (B) Ratios derived from the meta-analysis results in the Emory GADRC, Emory Other, and ROSMAP cohorts. Amyloid positivity was determined using a binary plasma pTau217 (AlzPath) cutoff for the majority of Emory GADRC cases; by CSF Aβ42/tTau ratio gaussian mixture model classification in Emory Other; and by the presence or absence of cerebral Aβ deposition by direct microscopic assessment in ROSMAP. In ROSMAP, both composite neuritic+diffuse plaque and CERAD plaque measures were tested, including a cutoff that included CERAD 1 as negative. n = 475 (Emory GADRC), n = 221 (Emory Other), and n = 216 (ROSMAP). (C) Ratios derived from optimal Aβ-associated proteins in each cohort were identified and tested in each cohort separately. Total numbers were the same as in (B). (D) Ratios derived from the meta-analysis results for proteins associated with cerebral Aβ independent of APOE ε4, and tested in APOE ε4-negative individuals. n = 583 (Bio-Hermes), n = 230 (Emory GADRC), n = 94 (Emory Other), and n = 174 (ROSMAP).

Extended Data Fig. 3. Plasma Protein Associations with Cognitive Function and Other Brain Pathologies.

(A) (Left) Overlap of proteins nominally associated with cognitive function in each cohort. The neuronal pentraxin receptor (NPTXR) was the only protein significantly associated with cognitive function across all cohorts at a nominal p < 0.05 level. (Right) Overlap of the top 50 proteins by FDR rank after meta-analysis across the cohorts. Proteins had to be nominally significant (p < 0.05) in at least one cohort. Two aptamers (2x) for NPTXR were significantly associated with cognitive function across all cohorts. (B) Gene ontology (GO) analysis of proteins positively (left) and negatively (right) associated with cognitive function highlighting biological pathway (BP), molecular function (MF), and cellular compartment (CC) terms. A summary of the terms is provided to the left in red. (C) GO analysis of proteins associated with cognitive function and not with other neuropathologies in ROSMAP regardless of direction of association (top), positively associated (bottom left), and negatively associated (bottom right).

Extended Data Fig. 4. Overlap of Brain-linked Proteins and Proteins Associated with Different Neuropathologies in ROSMAP.

Proteins significantly linked to brain levels were assessed for proteins associated with AT8-positive tangles (A), neurofibrillary tangles assessed by silver-stain (B), diffuse plaques (C), neuritic plaques (D), arteriolosclerosis (E), cerebral amyloid angiopathy (CAA, F), atherosclerosis (G), microinfarcts (H), gross infarcts (I), Lewy bodies (J), and TAR DNA-binding protein 43-positive inclusions (TDP-43, K). Proteins associated with brain levels and each neuropathology are provided in Supplementary Tables 5 and 8.

Extended Data Fig. 5. Protein Associations with Cognitive Impairment in the Absence of Cerebral Aβ.

Aβ-negative participants with cognitive impairment were identified in Bio-Hermes (254 participants), Emory GADRC (60 participants), and Emory Other (16 participants) cohorts and proteins associated with this diagnostic state (CI.other) were determined using a logistic model with Aβ-negative cognitively intact participants as the reference state. (A) Proteins associated with CI.other after meta-analysis. Positive meta z-scores indicate higher plasma protein levels are associated with CI.other; negative z-scores indicate lower protein levels are associated with CI.other. Proteins significant at meta q < 0.05 after Benjamini-Hochberg (BH) correction are colored in red. (B) Gene ontology (GO) analysis of proteins positively (top) and negatively (bottom) associated with CI.other highlighting biological pathway (BP), molecular function (MF), and cellular compartment (CC) terms. A summary of the terms is provided to the left in red. (C) Overlap of proteins associated with different neuropathologies in ROSMAP with proteins associated with CI.other after meta-analysis (CAA, cerebral amyloid angiopathy; TDP-43, TAR DNA-binding protein 43 inclusions; NFT, neurofibrillary tangles). The gray circle in the middle represents the sum of all unique proteins associated with CI.other and at least one neuropathology. Proteins associated with CI.other and each neuropathology are provided in Supplementary Table 9.

Extended Data Fig. 6. Proteins Associated with Risk of Incident Cognitive Impairment.

(A) GO analysis of proteins associated with risk of conversion across all three analyses (top), and separated by increased risk (bottom left) and decreased risk (bottom right), highlighting biological pathway (BP), molecular function (MF), and cellular compartment (CC) terms. A summary of the terms is provided to the left in blue. (B) Overlap of proteins associated with risk of cognitive impairment regardless of underlying etiology after five-fold cross validation (CV) with proteins associated with risk of conversion to dementia in the Atherosclerosis Risk in Communities (ARIC) cohort as previously described by Walker et al. in 2021 and 2023. (C) Overlap of proteins associated with risk of cognitive impairment regardless of underlying etiology after five-fold cross validation with proteins associated with risk of conversion to AD in the AGES-Reykjavik cohort as previously described by Frick et al..

Extended Data Fig. 7. Overlap of Plasma and Serum Protein Co-expression Networks.

The plasma protein co-expression network was tested for overlap with a serum protein co-expression network as previously described by Emilsson et al. using overrepresentation analysis. Numbers indicate –log₁₀ Benjamini-Hochberg-corrected p values for the overlap where significant at FDR < 0.05. Modules with particularly strong overlap or interest in AD are highlighted in red.

Extended Data Fig. 8. Enrichment of Phenotypes in a Plasma Protein Co-Expression Network.

Proteins associated with each phenotype in meta-analyses across the cohorts or in ROSMAP-only analyses were tested for enrichment in network modules. Proteins positively (POS) and negatively (NEG) associated with each phenotype were tested separately for enrichment, in addition to all significantly associated proteins (ALL). Numbers indicate Benjamini-Hochberg-corrected p values (FDR) for the enrichments significant at FDR < 0.05.

Extended Data Fig. 9. Enrichment of Proteins Associated with Risk of Incident Cognitive Impairment in a Plasma Protein Co-Expression Network.

Proteins associated with increased risk (positive HR) or decreased risk (negative HR) of incident cognitive impairment regardless of underlying etiology (All), restricted to high degree of AD pathology in those who converted (AD Path only), or restricted to high degree of AD pathology in those who converted and low AD pathology in the non-converters (AD path vs. non-path cntl) were tested for enrichment in plasma network modules. Numbers indicate Benjamini-Hochberg-corrected p values (FDR) for the enrichments significant at FDR < 0.05.