Identification of plasma proteins relating to brain neurodegeneration and vascular pathology in cognitively normal individuals

Abstract Introduction This study aims to first discover plasma proteomic biomarkers relating to neurodegeneration (N) and vascular (V) damage in cognitively normal individuals and second to discover proteins mediating sex‐related difference in N and V pathology. Methods Five thousand and thirty‐two plasma proteins were measured in 1061 cognitively normal individuals (628 females and 433 males), nearly 90% of whom had magnetic resonance imaging measures of hippocampal volume (as N) and white matter hyperintensities (as V). Results Differential protein expression analysis and co‐expression network analysis revealed different proteins and modules associated with N and V, respectively. Furthermore, causal mediation analysis revealed four proteins mediated sex‐related difference in N and one protein mediated such difference in V damage. Discussion Once validated, the identified proteins could help to select cognitively normal individuals with N and V pathology for Alzheimer's disease clinical trials and provide targets for further mechanistic studies on brain sex differences, leading to sex‐specific therapeutic strategies.


INTRODUCTION
The National Institute on Aging and Alzheimer's Association (NIA-AA) have proposed classifying Alzheimer's disease (AD) based on biomarkers of amyloid pathology (A), tau pathology (T), and neurodegeneration (N). 1 The flexibility of the AT(N) system could be expanded to incorporate new biomarkers that track brain vascular (V) damage, leading to ATV(N). 1 Both neurodegeneration and vascular pathology can be measured by several different magnetic resonance imaging (MRI) measures. For example, neurodegeneration can be measured by brain atrophy. Multiple lines of evidence showed that hippocampal atrophy is closely associated with AD. [2][3][4] Vascular pathology can be measured by white matter hyperintensities (WMHs), which have been associated with an increased risk for developing AD and dementia. [5][6][7] There is increasing evidence to suggest an influence of biological sex on neuroimaging biomarkers in AD pathogenesis. For example, brain atrophy rates for those with mild cognitive impairment (MCI) and AD dementia were faster in females compared to males. [8][9][10][11] In contrast, increased WMHs led to faster progression to MCI or AD only among males. 12,13 Although the differences are likely due to sex hormones, 14 the exact mechanisms that underlie these sex-related differences are still unclear. Gaining a more detailed picture of the causes of sexrelated differences could yield important clues about the pathophysiology of AD and eventually lead to sex-specific preventative or therapeutic strategies.
Compared to MRI measures, blood-based biomarkers show promise as a simple and potentially cost-effective option for the early detection, classification, and monitoring of AD pathology. With this in mind, the present study had two main objectives: first, to identify plasma biomarkers related to neurodegeneration and vascular pathology in cognitively normal individuals; second, to identify proteins mediating sex-related differences in neurodegeneration and vascular damage.
To do this, we used SomaLogic's Somascan assay to measure 5032 proteins in plasma from 1061 cognitively healthy individuals. Using a range of statistical approaches, we identified different proteins associated with markers of neurodegeneration and vascular damage. Furthermore, using causal mediation analysis, we found evidence for sev-eral proteins that may mediate sex-related differences in neurodegeneration and vascular pathology ( Figure 1).

Participants
The participants in this study were recruited as part of the Stratifying Resilience and Depression Longitudinally (STRADL) study, which re-contacted participants from the Generation Scotland: Scottish Family Health Study (GS). The GS study is a large, family-structured, population-based cohort study of more than 24,000 individuals from across Scotland. Recruitment took place between 2006 and 2011 with a clinical visit during which detailed health and cognitive test data were collected along with biological samples (blood, urine, saliva).
Full details of the STRADL cohort and GS protocol are published elsewhere. [15][16][17] Briefly, blood was collected by venepuncture into standard polypropylene ethylenediaminetetraacetic acid (EDTA) test tubes followed by centrifugation; the obtained plasma was aliquoted in 500 μL aliquots, and stored at −20 • C for future analyses. 17

F I G U R E 1
Overview of the study design: 5032 proteins were measured in 1061 cognitively normal individuals who had magnetic resonance imaging (MRI) measurement. Proteome-wide association study of MRI and protein co-expression network analysis revealed potential proteins for causal mediation analysis, leading to the finding of proteins mediating sex-related differences in brain neurodegeneration and vascular damage

Plasma analyses
Plasma proteins were measured using the SOMAscan assay platform (SomaLogic Inc.). SOMAscan is an aptamer-based assay allowing for the simultaneous measurement and quantification of, in the version used here, 5032 proteins. The assay uses chemically modified nucleotides to transform a protein signal into a nucleotide signal that can be quantified using relative fluorescence on microarrays. 18 Raw data processing and initial quality control (QC) led to 4235 proteins for final analysis.
The abundance of each protein was log-transformed, then the effects of sample collection site and plasma storage time on proteins were removed by linear regression and the residuals were used for all subsequent analyses.

MRI acquisition and analyses
Participants were scanned at two centers: the Ninewells Hospital in  Figure S1A1 in supporting information) or punctuate foci (deep WMHs, Figure S1B1); a score of 2 is defined as a smooth halo (periventricular WMHs, Figure S1A2) or beginning to confluence (deep WMHs, Figure S1B2) and a score of 3 is defined as irregular periventricular signal extending into the deep white matter (periventricular WMHs, Figure S1A3) or large confluent areas (deep WMHs, Figure S1B3). We used the total scores in this study by summing the periventricular and deep white matter scores.

Statistical analysis
Statistical analyses were completed using R (version 3.3.2). To compare baseline cohort characteristics between males and females, we used

Weighted gene correlation network analysis (WGCNA)
We first used the R package WGCNA 23 to construct a co-expression network from the proteins. The effects of age and sex on proteins were adjusted for by linear regression and the resulting residuals were used for analysis. WGCNA clustering is based on calculating correlations between paired variables, soft-threshold transforming them with a power function (cor β ), and using the result as adjacency matrix between variables. The final step applies hierarchical clustering to this adjacency matrix. We applied this algorithm with default parameters, except for the following settings: soft threshold power beta = 4, minimum module size = 10 proteins, merge cut height = 0.2. The resulting modules or groups of co-expressed proteins were used to calculate module eigenproteins. The eigenprotein-based connectivity (kME) was used to represent the strength of a protein's correlation with other protein module members. Proteins with high intramodular kME in the TA B L E 1 Demographics of participants included in the analysis by sex The AD risk was decided by both family history and APOE ε4 genotype.
In detail, individuals without family history (self-reported as no parent with AD) and without a copy of the APOE ε4 allele were categorized as low risk; individuals with either family history or APOE ε4 were defined as being at medium risk; individuals with both were classified as high risk.

Causal mediation analysis
We used R package regmedint 24  There was a significant difference between females and males in terms of hippocampal volume and WMHs after adjusting for covariates ( Figure S2B and C; criteria 1). For mediators, we selected proteins significantly associated with hippocampal volume or WMHs (criteria 2). We also checked that these proteins are significantly expressed between females and males (criteria 3). We reported those proteins whose natural indirect effect is significant and the direction is consistent with natural direct effect.

Subject demographics
Demographic information of subjects is shown in Table 1. The male group was slightly older than the female group. No significant difference was observed in the distribution of APOE ε4 carriers, education, and self-reported family history (mother or father reported as having AD). In terms of MRI measures, females had a smaller hippocampal volume than males on average, while no difference was observed in WMHs.

Proteins significantly associated with hippocampal volume and WMHs
Using partial Spearman correlation, we found nominally significant associations (P < .05) between 377 proteins and hippocampal volume in all individuals. Seventeen of these were significant after FDR correction for multiple testing (FDR P < .05; Figure 2A, Table S1 in supporting information). KEGG pathway analysis of the 228 proteins with significantly negative associations (P < .05) revealed three pathways including cytokine-cytokine receptor interaction, axon guidance, and metabolic pathways ( Figure 2B). Conversely, the 149 proteins with significantly positive associations (P < .05) indicated two pathways that were cell adhesion molecules (CAMs) and complement and coagulation cascades ( Figure 2B

Plasma protein co-expression network analysis reveals modules linked to hippocampal volume and WMHs
We first performed a network-based analysis of the plasma proteome using WGCNA. We found eight modules (M) of co-expressed proteins and ranked them based on size from largest (M1 turquoise module; n = 2694 proteins) to smallest (M8 pink module; n = 13 proteins; Table S2 in supporting information). Figure 3A shows the clustering of these modules' concordance according to similarities in expression patterns. We further investigated the biological significance of proteins in each module and found that the modules were enriched with various pathways after FDR correction (Table S3 in

Correlation of protein networks with cognitive test scores and AD risk
We further investigated the module correlations to cognitive test measures and AD risk. We found that the M4 yellow and M8 pink modules had negative correlations and M5 green module had a positive correlation with general cognitive score after FDR correction (Figure 4A-C). These results are in concordance with MRI correlations as the M4 yellow and M5 green modules had negative and positive correlations with hippocampal volume, respectively. The AD risk was decided by both family history and APOE ε4 genotype as described previously, leading to 662 low risk, 315 moderate risk, and 32 high risk individuals. We found that the M2 blue and M6 red modules showed significant increase and decrease in higher risk compared to low-risk individuals, respectively ( Figure 4D and E). Of these, the M6 red module is in concordance with MRI correlations as it had posi-tive and negative correlations with hippocampal volume and WMHs, respectively.

Causal mediation analysis reveals proteins mediating brain sex differences
We selected proteins obtained from DEA for further causal mediation analysis. Overall, we considered 17 hippocampal volume-related and two WMH-related proteins (FDR P < .05). Of the 17 proteins, 4 of them showed significant natural indirect effect with hippocampal volume (Table 2), indicating they mediated sex-related differences in brain neurodegeneration. Furthermore, two of them were in the M5 green module, which had positive correlations with hippocampal volume only in males ( Figure 3B) and NCAM-1 was the hub protein in the M5 green module (Figure 3C), further indicating the consistency  Figure 3B).

DISCUSSION
In this study, we used SOMAscan to measure 5032 plasma proteins It is important to identify blood-based biomarkers relating to neurodegeneration, particularly in the preclinical stage of AD. Most clinical trials for AD have been unsuccessful to date. 26 The failure of such trials was partially caused by the fact that participants enrolled in such trials were relatively late in the disease process. Targeting treatment to earlier pre-symptomatic or prodromal stages of the disease might have more success. 27 With this in mind we sought to identify blood biomarkers relating to neurodegeneration in cognitively normal individuals collected from a population-based cohort that resembles the real-world situation of participants' recruitment in clinical trials. We identified 17 proteins that were significant after FDR correction for multiple testing in the full cohort; two proteins were FDR significant in females compared to no proteins in males. This might be because brain atrophy is small in cognitively normal individuals and that the subgroup analysis is too small to detect such differences. Nevertheless, because these proteins were associated with neurodegeneration in the preclinical stage, they could potentially help when recruiting cognitively normal individuals with neurodegeneration for clinical trials with further validation.
The importance of WMHs in AD is increasingly recognized as they are related to risk for developing AD. [5][6][7] Some studies showed that there is an association between WMHs and AD pathological hallmarks such as amyloid beta (Aβ) plaques, tau, and neurodegeneration. [28][29][30] Here, we only found two proteins that were associated with WMHs after FDR correction for multiple testing. This might be because the participants recruited in our study were relatively young whereas WMHs tend to be observed with aging. 31 Current findings on the associations of AD hallmarks in plasma with neurodegeneration and vascular pathology have generated new enthusiasm. For example, various studies found that neurofilament light chain (NfL) was closely associated with neurodegeneration. [39][40][41] Furthermore, NfL was also associated with vascular pathology in an age-dependent manner. 42 Another marker, phosphorylated tau (ptau)181 was found to be a good marker for predicting and monitoring neurodegeneration. 43,44 These two biomarkers are well recognized as they have been validated in multiple cohorts across different laboratories. In comparison, the identified biomarkers from our study need further validation in independent longitudinal cohorts. Furthermore, it is important to conduct head-to-head comparison studies in the future not only to compare the performance of our identified biomarkers to both NfL and p-tau181 but also to check whether they could add extra value on top of these two well-recognized biomarkers.
The increasing recognition of sex differences in the brain and AD risk has highlighted the urgent need for biomarkers that more comprehensively reflect the complex mechanisms underlying these differences. 45,46 Examination of these differences may shed light on the pathophysiology of AD that differs between the sexes and ultimately lead to more effective interventions and precision medicine.
Here, we reported a study that characterized plasma proteins mediating brain sex differences in cognitively normal individuals in the largest sample to date. Using a range of statistical approaches, we identified four proteins that mediated sex-related difference in brain neurodegeneration and one protein in vascular damage.
Of the five proteins, neural cell adhesion molecule (NCAM) is a part of a family of cell-surface glycoproteins that play= key roles in normal brain development, including axonal/dendritic growth and branching, and synaptic plasticity. 47 The levels of NCAM-1 have been shown to alter in AD patients' blood, cerebrospinal fluid (CSF), and brain tissue. [48][49][50] Furthermore, it interacted with amyloid precursor protein (APP) and promoted neurite outgrowth, indicating that it could be a potential therapeutic target for AD treatment. 51 cognitive score and their direction of association were in concordance with brain MRI correlations. Various studies have reported that brain atrophy and increased WMHs lead to cognitive decline. [63][64][65][66][67] Our finding further demonstrated that proteins relating to brain pathology were also associated with lower cognitive test scores.
There are three limitations for our study. First, the population in this study is of European ancestry, predominantly of Scottish ancestry, so validation in independent cohorts and particularly in other ethnic groups is needed to see if the results are generalizable. Second, our study is cross-sectional and longitudinal studies are required to determine the role of nominated proteins in sex differences in brain pathology and risk of AD. Third, although the individuals in this study are well characterized on cognition, APOE ε4 genotype, and family history, there is a lack of Aβ plaques and tau tangles to confirm the stage of AD.
Therefore, the terminology of preclinical AD needs to be interpreted with caution. Further studies on individuals defined by both neuropathology and clinical symptoms are needed to confirm the results.
Despite this, our study is the largest we are aware of to report plasma biomarkers indicative of both neurodegeneration and vascular damage in cognitively normal individuals in terms of the number of proteins assayed as well as sample size. By applying different statistical approaches, we identified individual proteins and protein networks linked to N and V in the poorly understood preclinical stage of AD. Furthermore, we demonstrated that four proteins mediated sex-related differences in brain neurodegeneration and one protein in vascular damage. These nominated proteins can not only serve as biomarkers to help recruit cognitively normal individuals with neurodegeneration and vascular damage for clinical trials but also as predictive biomarkers to monitor possible intervention outcomes. Furthermore, these proteins provide tractable targets for further mechanistic studies of sex differences in brain pathology and AD risk, with potential to lead to more effective sex-specific preventative or therapeutic strategies. Award ID: 1U19AG063744-01), Gut-liver-brain biochemical axis in

ETHICS STATEMENT
Ethical approval for the GS:SFHS study was obtained from the Tayside Committee on Medical Research Ethics (on behalf of the National Health Service).