Proteomic analyses reveal plasma EFEMP1 and CXCL12 as biomarkers and determinants of neurodegeneration

Abstract

Introduction: Plasma proteomic analyses of unique brain atrophy patterns may illuminate peripheral drivers of neurodegeneration and identify novel biomarkers for predicting clinically relevant outcomes.

Methods: We identified proteomic signatures associated with machine learning-derived aging- and Alzheimer's disease (AD) -related brain atrophy patterns in the Baltimore Longitudinal Study of Aging (n = 815). Using data from five cohorts, we examined whether candidate proteins were associated with AD endophenotypes and long-term dementia risk.

Results: Plasma proteins associated with distinct patterns of age- and AD-related atrophy were also associated with plasma/cerebrospinal fluid (CSF) AD biomarkers, cognition, AD risk, as well as mid-life (20-year) and late-life (8-year) dementia risk. EFEMP1 and CXCL12 showed the most consistent associations across cohorts and were mechanistically implicated as determinants of brain structure using genetic methods, including Mendelian randomization.

Discussion: Our findings reveal plasma proteomic signatures of unique aging- and AD-related brain atrophy patterns and implicate EFEMP1 and CXCL12 as important molecular drivers of neurodegeneration.

Highlights: Plasma proteomic signatures are associated with unique patterns of brain atrophy. Brain atrophy-related proteins predict clinically relevant outcomes across cohorts. Genetic variation underlying plasma EFEMP1 and CXCL12 influences brain structure. EFEMP1 and CXCL12 may be important molecular drivers of neurodegeneration.

Keywords: Alzheimer's disease; biomarkers; inflammation; neurodegeneration; proteomics.

FIGURE 1

Study design. (A) BLSA participants included in current study had a concurrent 3T MRI brain scan (which was used to calculate machine learning‐derived indices of aging and AD brain atrophy patterns) and a blood draw (which was used measure the abundance of 7268 proteins in plasma). (B) Proteomic analyses identified individual plasma proteins (FDR p < 0.05) and broader proteomic signatures (uncorrected p < 0.05) associated with brain atrophy patterns. (C) Plasma proteins associated with brain atrophy were related to plasma biomarkers of AD pathology (Aβ_42/40, pTau‐181), neuroinflammation (GFAP) and neurodegeneration (NfL), as well as prevalent cognitive impairment risk in the BLSA. (D) Plasma proteins associated with brain atrophy in the BLSA were related to CSF AD biomarkers (Aβ₄₂, total tau, pTau‐181), cognitive performance (MoCA), and prevalent AD risk in the Emory Goizueta ADRC, with performance across six cognitive domains, brain volumes, and white matter integrity in the GenS study, and with mid‐life (20‐year follow up) and late‐life (8‐year follow up) incident dementia risk in the ARIC study. (E) Genetic variation that influenced plasma protein levels was associated with brain atrophy, including causal relationships established with two sample MR. (F) The biological implications and functional relevance of plasma proteins associated with brain atrophy were examined using a variety of bioinformatic tools. Aβ_42/40, amyloid‐beta _42/40 ratio; AD, Alzheimer's disease; ADRC, Alzheimer's disease and related dementias; ARIC, atherosclerosis risk in communities; BLSA, Baltimore Longitudinal Study of Aging; GenS, Generation Scotland; GFAP, glial fibrillary acidic protein; MCI, mild cognitive impairment; MoCA, Montreal Cognitive Assessment; MR, Mendelian randomization; MRI, magnetic resonance imaging; NfL, neurofilament light chain; NfL, neurofilament light.

FIGURE 2

Proteomic signatures of aging and AD brain atrophy patterns. (A) Representative voxel‐wise images show machine learning‐derived brain atrophy patterns associated with aging (R1, R2, R3, R4, R5) and AD (SPARE‐AD) in BLSA participants. R1–R5 and SPARE‐AD have been trained and validated previously in large, external cohorts where they predict age‐related clinical traits, neurodegenerative disease risk, and clinical dementia progression, . (B) Volcano plots show individual plasma proteins (FDR p < 0.05) related to brain atrophy measures (R1, R3, SPARE‐AD) in the BLSA. Dot colors indicate tissue specificity of plasma proteins, with an emphasis on the top enriched tissue types for each brain atrophy measure; black dots indicate proteins that are p < 0.05 and tissue specific (but not a top enriched tissue for a given brain atrophy measure) and dark gray dots indicate proteins that are p < 0.05 and not tissue specific. Results were derived from linear regression models adjusting for age, sex, race, education, eGFR, APOEε4 status, and a comorbidity index (i.e., obesity, hypertension, diabetes, cancer, ischemic heart disease, chronic heart failure, chronic kidney disease, and chronic obstructive pulmonary disease). (C) Sagittal images show the associations of plasma proteins with voxel‐wise differences in gray matter volumes. A threshold of 50 voxels with a t‐value of 2.0 was used to define significant clusters. Results were derived from covariate adjusted linear regression models, with additional adjustment for ICV. (D) Graphical summary shows proteomic analyses of machine learning‐derived brain atrophy measures, including individual plasma proteins that survived multiple comparison correction (FDR p < 0.05). Broader proteomic signatures (uncorrected p < 0.05; upregulated ↑; downregulated ↓) of each brain atrophy measure were enriched for specific cell and tissue types (obtained from Tabula Sapiens and the Genotype‐Tissue Expression project, respectively), as well as distinct biological processes (obtained from IPA), such as PDGF signaling and ubiquitination. AD, Alzheimer's disease; APOEε4, apolipoprotein Eε4; BSLA, Baltimore Longitudinal Study of Aging; eGFR, estimated glomerular filtration rate; FDR, false discovery rate; ICV, intracranial volume; IPA, Ingenuity Pathway Analysis; PDGF, platelet derived growth factor; SPARE‐AD, spatial pattern of abnormality for recognition of early Alzheimer's disease.

FIGURE 3

Brain atrophy‐enriched biological pathways. Bar graphs show the top ten canonical pathways (as well as corresponding p‐values and activity patterns) enriched among the machine learning‐derived brain atrophy measures (A) R1 (subcortical atrophy), (B) R2 (medial‐temporal lobe), (C) R3 (parieto‐temporal lobe), (D) R4 (diffuse cortical), (E) R5 (perisylvian), and (F) SPARE‐AD (AD‐related regions). Results were derived from IPA. Proteomic signatures (i.e., uncorrected p < 0.05) used for enrichment analyses were derived from linear regression models adjusting for age, sex, race, education, APOEε4, eGFR and a comorbidity index (i.e., obesity, hypertension, diabetes, cancer, ischemic heart disease, chronic heart failure, chronic kidney disease, and chronic obstructive pulmonary disease). APOEε4, apolipoprotein Eε4; eGFR, estimated glomerular filtration rate; IPA, Ingenuity Pathway Analysis; SPARE‐AD, spatial pattern of abnormality for recognition of early.

FIGURE 4

Plasma proteins linked to brain atrophy in the BLSA are associated with plasma and CSF biomarkers, neurocognitive outcomes, and incident dementia risk in external cohorts. (A) Forest plot shows how plasma proteins linked to brain atrophy relate to differences in plasma biomarkers of AD pathology (Aβ_42/40, pTau‐181), neuroinflammation (GFAP) and neurodegeneration (NfL) among BLSA participants. Results were derived from linear regression models adjusting for age, sex, race, education, eGFR, APOEε4 status, and a comorbidity index (i.e., obesity, hypertension, diabetes, cancer, ischemic heart disease, chronic heart failure, chronic kidney disease, and chronic obstructive pulmonary disease). Filled in shapes indicate statistically significant associations (p < 0.05). (B) Forest plot shows how plasma proteins linked to brain atrophy relate to prevalent cognitive impairment (MCI + dementia), MCI, and dementia among BLSA participants. Results were derived from logistic regression models adjusted for similar covariates. Filled in shapes indicate statistically significant associations (p < 0.05). (C) Heatmap shows how plasma proteins relate to differences in CSF AD biomarkers (Aβ₄₂, total tau, pTau‐181) and cognitive performance (MoCA) in the Emory Goizueta ADRC cohort. Results were derived from linear regression models adjusted for age, sex, and APOEε4 *Indicates statistically significant associations (p < 0.05). (D) Boxplots show how plasma proteins relate to prevalent AD  in the Emory ADRC. Results were derived from logistic regression models adjusted for similar covariates. (E) Heatmap shows how plasma proteins relate to differences in six cognitive domains in the GenS cohort. NPTX2 was not measured in GenS. Results were derived from linear mixed‐effects regression models correcting for relatedness between individuals (i.e., kinship matrix) and adjusting for age, sex, depression diagnosis, clinic study site, and sample storage time. *Indicates statistically significant associations (p < 0.05). (F) Heatmap shows how plasma proteins relate to differences in brain volumes and white matter integrity in the GenS cohort. Results were derived from similar covariate‐adjusted models with additional adjustment for ICV and processing artifacts (batch, presence, or absence of manual intervention during QC). *Indicates statistically significant associations (p < 0.05). (G) Forest plot shows how plasma proteins linked to higher and lower brain atrophy in the BLSA relate to incident risk for mid‐life (20‐year) and late‐life (8‐year) dementia in the ARIC study. Results were derived from Cox proportional hazards regression models adjusting for age, sex, race‐center, education, APOEε4, eGFR‐creatinine, and cardiovascular risk factors (BMI, diabetes, hypertension, and current smoking status). Aβ, amyloid‐beta; Aβ_42/40, amyloid‐beta _42/40 ratio; AD, Alzheimer's disease; ADRC, AD research center; APOEε4, apolipoprotein Eε4; ARIC, atherosclerosis risk in communities; BSLA, Baltimore Longitudinal Study of Aging; CSF, cerebrospinal fluid; DVR, distribution volume ratio; eGFR, estimated glomerular filtration rate; ETC, entorhinal cortex; GenS, Generation Scotland; GFAP, GFAP, glial fibrillary acidic protein; HR, hazard ratio; ITG, inferior temporal gyrus; MCI, mild cognitive impairment; MoCA, Montreal Cognitive Assessment; NfL, Neurofilament light; NPTX2, neuronal pentraxin; QC, quality control; RFU, relative fluorescent units; SUVR, standardized uptake value ratio.

FIGURE 5

Genetic variation underlying EFEMP1 and CXCL12 protein levels is associated with brain structure. (A) Forest plot shows cis EFEMP1 and CXCL12 SNPs associated with plasma protein levels and neuroimaging traits in external cohorts. *rs7596872 [C] is also a pQTL for higher protein expression in CSF and higher RNA expression in the hippocampus, cortex, and the anterior cingulate cortex. pQTLs were obtained from deCODE Genetics. Supplementary information on individual SNPs was obtained with the OpenTargets, Genecards, OnTime, and GTEx platforms. (B) Boxplots show plasma levels of EFEMP1 and CXCL12 stratified by rs573431210 and rs77542162 genotypes, respectively, among BLSA participants. Results were derived from linear regression models adjusted for covariates used in the pQTL discovery cohort, namely age and sex. (C) Boxplot shows parieto‐temporal atrophy (R3) stratified by rs10793514 genotype among BLSA participants. Results were derived from linear regression models adjusted for covariates used in the pQTL discovery cohort, namely age and sex. (D) Scatter plot shows two sample MR results that examined the relationship of genetically determined EFEMP1 plasma protein levels with hippocampal volume. pQTLs were obtained from deCODE Genetics. GWAS summary statistics of hippocampal volume were obtained from the ENIGMA Consortium. (E) Scatter plot shows two sample MR results that examined the relationship of genetically determined CXCL12 plasma protein levels with parieto‐temporal atrophy (R3). pQTLs were obtained from deCODE Genetics. GWAS of R3 was obtained from the UKBB. BLSA, Baltimore Longitudinal Study of Aging; ENIGMA, Enhancing Neuroimaging Genetics through Meta‐analyses Consortium; GWAS, genome wide association study; MR, Mendelian randomization; pQTLs, protein quantitative trait loci; UKBB, UK Biobank.

FIGURE 6

Cell‐specific expression, downstream molecules, and a summary of evidence for plasma proteins linked to brain atrophy. (A) Heatmap shows expression levels of cognate genes encoding plasma proteins across 76 available cell types based on single‐cell transcriptomics data sourced from the Human Protein Atlas. Dendrograms reflect hierarchical clustering using Euclidean distances calculated from nTPM. nTPMs used to generate the heatmap were standardized within each cognate gene to improve interpretability. (B) Interaction network displays the downstream molecules of CXCL12 and EFEMP1, their direct (solid lines) and indirect (dashed lines) associations, as well as this network's top enriched canonical pathways, subcellar localizations, and disease annotations. Results were derived from IPA Path Builder. (C) Graphical summary of evidence for each plasma protein linked to brain atrophy. AD, Alzheimer's disease; ARIC, atherosclerosis risk in communities study; BLSA, Baltimore Longitudinal study of Aging; Emory ADRC, Emory Goizueta AD Research Center; ENIGMA, Enhancing Neuroimaging Genetics through Meta‐analyses Consortium; GenS, Generation Scotland study; IPA, Ingenuity Pathway Analysis; Mayo, Mayo RNAseq; MR, two‐sample Mendelian randomization; MSBB, Mount Sinai Brain Bank; nTPM, normalized transcripts per million; pQTL, protein quantitative trait loci; ROSMAP, Religious Orders Study and Rush Memory and Aging Project; Stanford ACRC, Stanford Aging Clinical Research Center; UKBB, UK biobank.