Introduction
Alzheimer’s disease (AD) is the most common cause of dementia, affecting one in 10 people aged 65 years or older [
3]. To date, the development of disease-modifying treatments for AD has largely targeted one of its pathological hallmarks, amyloid beta (Aβ), and to a lesser extent tau, with notably high failure rates in clinical trials [
12]. Molecular markers such as amyloid PET, cerebrospinal fluid (CSF) Aβ, and CSF tau are frequently employed in the clinical diagnosis of AD, and biomarkers are increasingly recognized as integral components of the drug development pipeline [
10]. However, recent evidence from human genetics [
45,
52], neuroimaging [
14], and in-vivo modeling [
9] has strongly implicated additional disease mechanisms in AD, not reflected by the established Aβ and tau biomarkers. These potential pathophysiological processes, such as inflammation [
14,
52], membrane phospholipid dysregulation [
62], and neurovascular disruption [
45], are less frequently targeted, and remain less well understood [
22].
Proteomics, the analysis of proteins in different bodily tissues and fluids, can be used to study myriad pathways putatively affected in AD in-vivo, before the onset of overt neurodegeneration, thus potentially elucidating important disease mechanisms and informing future pharmacotherapeutic trials. In CSF, certain proteomic assays can reflect underlying brain pathology [
48], and thus, may be useful for pharmacokinetic monitoring, diagnosis, and stratification [
5]. However, lumbar puncture is invasive. Blood may serve as a more accessible body fluid; however, many blood-based molecular profiling studies of AD have been restricted by small sample sizes, methodological variability, and suboptimal diagnostic sensitivity [
28,
38]. Additionally, with the exception of certain blood biomarkers, such as plasma Aβ [
39] and plasma tau [
28], few studies have directly explored the relationship between CSF proteins and their blood-based analogs in the same cohort; thus, the extent to which peripheral molecular changes accurately reflect CNS dynamics has yet to be characterized at large scale.
Here, we sought to understand the contributions of proteins downstream of, and potentially orthogonal to, Aβ and tau across the preclinical, prodromal, and dementia stages of AD. We measured a diverse panel of 270 proteins in the CSF and plasma of up to 1022 individuals, using a validated, highly sensitive and specific immunoassay [
4]. We identified, and replicated, evidence of differential regulation in proteins related to innate and adaptive immunity (CD200, CHIT1, MMP-9, MMP-10, oncostatin-M, STAMBP), membrane phospholipids (LDLR), axon guidance, cell adhesion and differentiation (ALCAM, RGMB, ROBO2), ischemic injury (uPA, tPA, SMOC2), mTOR and Wnt/β-catenin signaling (AXIN1, EIF4EBP1), and glucose metabolism (HAGH) in early and later-stage AD. Multiple proteins associated with baseline cortical thickness and cognitive performance. Approximately half of all assayed proteins in CSF showed modest correlations with their analogs in plasma, and a combination of plasma analytes could differentiate AD from non-AD with high accuracy. Thus, our findings reemphasize the importance of targeting pathways beyond Aβ and tau in AD, and identify several candidate biomarkers for early detection of brain pathology, pending additional validation.
Discussion
In one of the broadest multiplex proteomics studies of dementia conducted to date, we identified a series of significant alterations across plasma and CSF, implicating multiple proteins involved in regulation of the inflammatory response, apoptosis, endocytosis, leukocyte proliferation, and other biological processes believed to be downstream of, and potentially orthogonal to, Aβ and tau deposition. A subset of proteins collectively discriminated individuals with AD dementia from healthy controls with 95% predictive accuracy in CSF, and 94% accuracy in plasma, representing an improvement over prior multiplex panels [
13,
16,
36,
42,
43]. With the exceptions of YKL-40 [
1,
11,
18] and CHIT1 [
27] in CSF, and MMP-9 [
7] in blood, our findings highlight a set of relatively unexplored candidate biomarkers of neurodegeneration, warranting further validation in independent disease cohorts using targeted proteomic assays.
Our findings highlight the roles of several functionally pleiotropic proteins in AD, such as CASP8, which was increased in the plasma and CSF of AD-dementia patients, and in the plasma of Aβ+ MCI patients. CASP8 is involved in synaptic plasticity, amyloid processing, memory, and regulation of microglial pro-inflammatory activation [
60]. The
CASP8 gene has shown a significant rare variant burden association with AD susceptibility [
46], and prior immunohistochemical analyses have revealed activation of CASP8 in AD hippocampal tissue [
47]. Caspase inhibition has been proposed as a mechanism to promote neuronal survival [
49], and may be a beneficial therapeutic strategy for AD.
Similarly, JAM-B, a protein involved in synaptic adhesion, lymphocyte transendothelial migration [
21], and vascular inflammation [
59], was down-regulated in the plasma and CSF of AD patients and the plasma of MCI patients compared with cognitively normal Aβ- controls. In plasma, JAM-B levels negatively associated with baseline CDR-SB scores. Polymorphisms in or near the
JAM2 gene have shown suggestive associations with cognitive performance [
34], postmortem Aβ load [
51], and longitudinal changes in amyloid plaque burden on PET [
44].
CSF concentrations of MMP-10 were elevated in Aβ+ MCI and AD patients, positively correlated with cognitive performance, and negatively associated with AD-signature cortical thickness. Other matrix metalloproteinase-related proteins, including MMP-9, TIMP-4, and ADAM22, were also up-regulated in AD. These findings highlight the complex pathophysiological role of endopeptidases in AD, potentially via direct degradation of Aβ deposits [
32] or indirect mediation of Aβ-induced blood-CSF barrier dysfunction [
6]. Transferrin receptor, which facilitates brain iron uptake, was also significantly increased in the plasma of AD patients, implicating elevated brain iron levels, potentially via blood-brain barrier (BBB) breakdown, in disease pathophysiology [
54]. Further investigation of the role of metalloproteinases, endothelial proteins and adhesion molecules, particularly as they relate to meningeal lymphatics, BBB function and iron levels in AD [
54], may facilitate novel therapeutic target discovery.
In the CSF and plasma of AD patients, we found decreased concentrations of tPA, a blood clotting enzyme which co-localizes with amyloid-rich regions and phosphorylated tau in post-mortem AD brains [
30], and has been linked to accelerated Aβ degradation in-vivo [
37]. In AD plasma, we found decreased levels of another plasminogen activator, uPA, which has been associated with inhibition of Aβ neurotoxicity and fibrillogenesis [
56] and recovery after axonal injury [
31].
Other notable proteins that showed evidence of differential regulation in AD compared to healthy elderly individuals included CD200, a glycoprotein previously associated with enhanced in-vivo and in-vitro amyloid phagocytosis [
57], UNC5, an axonal guidance protein receptor which has previously shown genetic links to familial, late-onset AD via increased susceptibility to neuronal cell death [
61], and SMOC2, a member of the SPARC protein family, which is involved in microgliosis and functional recovery after cortical ischemia [
24], and was significantly associated with cognitive performance in our study. We also observed profound increases of HAGH, also known as glyoxalase II, in plasma, reemphasizing the potential role of advanced glycation end products in neurodegeneration [
23,
63].
Our study highlights the changes and potential contributions of several chemokines (CXCL1, CCL3, CXCL5, CXCL6, CCL3, CCL19), interleukins (IL8, IL6-RA, IL18-BP), and other immune markers (OSM, ALCAM, OPG, CHIT1, YKL-40, STAMBP, MMP-9, MMP-10) in the prodromal and dementia phases of AD. YKL-40, a chitinase expressed by astrocytes and microglia in the CNS [
8], was increased in the CSF of Aβ+ CN individuals and patients with AD, and negatively associated with thickness measures in AD-vulnerable brain regions, replicating prior findings [
2] and reemphasizing similar findings from the same cohort, using a standard ELISA assay [
19]. However, elevations of YKL-40 in CSF were modest compared with prior investigations [
1,
11,
18] and were negligible in plasma, indicating limited diagnostic utility of this biomarker in isolation [
27]. CHIT1, another putative marker of microglial activation, showed more pronounced elevations in the CSF of Aβ+ MCI and AD patients, supporting prior observations of up-regulated CHIT1 in progressive brain illnesses [
53,
58], and warranting further investigation of the role of chitinases in immune response and neurodegeneration.
Compared with Aβ- CNs, MCI patients without evidence of pathological CSF Aβ showed significant differential regulation of OPG, RGMB and ROBO2 protein in CSF, and differential regulation of multiple proteins in plasma, including brevican. OPG, RGMB, and ROBO2 have been implicated in CNS development, axonal growth, and recovery after nerve injury [
15,
25,
50], whereas upregulation of brevican has been observed in the CNS during the consolidation of long-term memories [
35]. Thus, dysregulation of these proteins in Aβ- MCI may reflect broader mechanisms involved in neurodegeneration, potentially unrelated to Aβ pathology.
Conversely, certain proteins, dysregulated when Aβ+ patients were compared to Aβ- controls, may reflect the presence of Aβ deposits rather than a specific diagnosis of AD or MCI. To address this possibility, we conducted a series of supplementary analyses, as detailed in Additional file
1: Note 1. These analyses revealed that, with the exceptions of CHIT1, YKL-40 and SMOC2 in CSF, most differentially regulated proteins identified in Aβ+ patients may represent markers of disease diagnosis, rather than Aβ deposition alone, as they showed similar patterns of differential regulation when Aβ+ groups were compared with one another directly (see Additional file
1: Tables S23-S32).
Some differentially regulated proteins may reflect the influences of coexisting pathologies in AD, such as Lewy body inclusions or TDP-43 aggregates. To test this hypothesis, we conducted supplementary analyses in a small, independent cohort of Aβ- healthy elderly individuals compared with Aβ- patients with Parkinson’s disease (PD), multiple system atrophy (MSA), and progressive supranuclear palsy (PSP; see Additional file
1: Note 1). This analysis indicated that MMP-10, STAMBP, LDL receptor, and EIF4EBP1 likely represent disease-specific biomarkers of AD, whereas CHIT1, ALCAM, CD200, OPG, ROBO2, and RGMB may serve as broader neurodegeneration biomarkers, potentially influenced by coexisting pathologies beyond Aβ- and tau (Additional file
1: Tables S33-S38). Further studies of patients with Lewy body dementia, frontotemporal dementia, and other neurodegenerative illnesses are strongly recommended to help validate these findings.
Despite its scale, our study has limitations. Many differentially regulated proteins yielded small-to-modest effect sizes (0.05 < d < 0.4), and the functions of some proteins demonstrating AD-related changes are not fully understood. Also, given the study’s cross-sectional design, we could not investigate the precise temporal dynamics of important biological processes such as innate immunity in AD. Future investigations will combine longitudinal plasma and CSF measurements with genome-wide genotyping to disentangle causal from consequential biomarkers, and to map the development of protein biomarkers throughout the progression of disease. Additionally, the Olink™ PEA technology takes advantage of unique tags and polymerase chain reaction to measure up to 92 proteins in parallel, including many low-expressing analytes typically undetected using traditional multiplex assays. The technology therefore represents a relatively focused screen, whereby protein levels are reported as normalized protein concentration (NPX). Prospective experiments should combine more unbiased proteomics approaches (such as mass spectrometry) with targeted assays (such as enzyme-linked immunoabsorbance assays, single molecule arrays, or customized Olink™ panels), reporting results as absolute concentrations.
Finally, while we identified subsets of proteins that accurately discriminated MCI and AD patients from elderly controls, further work is required to identify a suitable selection of biomarkers for very early detection of cognitive impairment, prior to the presentation of overt clinical symptoms. Replication studies in independent cohorts, complemented by larger multiplex proteomics panels, genetic risk scores, advanced neuroimaging, neuropsychological assessments, and post-mortem expression studies will help characterize early contributors towards longitudinal cognitive decline in AD [
33].
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.