Platelet biomarkers for a descending cognitive function: A proteomic approach

Abstract Memory loss is the most common clinical sign in Alzheimer's disease (AD); thus, searching for peripheral biomarkers to predict cognitive decline is promising for early diagnosis of AD. As platelets share similarities to neuron biology, it may serve as a peripheral matrix for biomarkers of neurological disorders. Here, we conducted a comprehensive and in‐depth platelet proteomic analysis using TMT‐LC‐MS/MS in the populations with mild cognitive impairment (MCI, MMSE = 18–23), severe cognitive impairments (AD, MMSE = 2–17), and the age‐/sex‐matched normal cognition controls (MMSE = 29–30). A total of 360 differential proteins were detected in MCI and AD patients compared with the controls. These differential proteins were involved in multiple KEGG pathways, including AD, AMP‐activated protein kinase (AMPK) pathway, telomerase RNA localization, platelet activation, and complement activation. By correlation analysis with MMSE score, three positively correlated pathways and two negatively correlated pathways were identified to be closely related to cognitive decline in MCI and AD patients. Partial least squares discriminant analysis (PLS‐DA) showed that changes of nine proteins, including PHB, UQCRH, CD63, GP1BA, FINC, RAP1A, ITPR1/2, and ADAM10 could effectively distinguish the cognitively impaired patients from the controls. Further machine learning analysis revealed that a combination of four decreased platelet proteins, that is, PHB, UQCRH, GP1BA, and FINC, was most promising for predicting cognitive decline in MCI and AD patients. Taken together, our data provide a set of platelet biomarkers for predicting cognitive decline which may be applied for the early screening of AD.


| INTRODUC TI ON
Alzheimer's disease (AD) is the most common cause of neurodegenerative disorders, and its prevalence is exacerbated by an aging population (Collaborators, 2019). It is estimated that about 47 million people are currently affected by dementia, and the number is expected to reach 131 million by 2050, with appropriate interventions and treatment leading to a reduction in prevalence (Hodson, 2018). The main clinical manifestations of AD patients are memory impairment and cognitive deficits, which make them unable to effectively carry out daily life (Q uerfurth & LaFerla, 2010). However, the underlying pathology, including amyloid plaque deposition and neurofibrillary tangles, may have occurred before symptoms appear (Hodson, 2018;Jack et al., 2010). Therefore, timely diagnosis, intervention, and treatment are particularly important. However, the diagnosis of AD has not been standardized, and the main diagnostic methods include MRI and PET brain imaging, biochemical analysis of Aβ42/40, and total tau (t-tau) and phosphorylated tau (p-tau181) levels in the cerebrospinal fluid (CSF) (Bocchetta et al., 2015;Rice & Bisdas, 2017;Ritchie et al., 2017). Although these diagnostic methods have made significant progress, they are hardly acceptable to the potential patients because these methods are either expensive or invasive. In addition, researchers have paid more attention to the periphery, such as microRNA455-3p in blood has the potential to serve as a peripheral marker for early diagnosis of AD (Kumar & Reddy, 2018Kumar, Vijayan, & Reddy, 2017). Therefore, finding blood biomarkers is of great significance for the early diagnosis of AD.
Platelet, a non-nuclear fragment from megakaryocytes (Cardigan et al., 2005;Kamath et al., 2001), shares multiple similarities with neuron biology, and it is easily affected by diseases (Akingbade et al., 2018). Once activated, platelets will release a variety of biochemically active factors including cytokines, chemokines, and neurotransmitters (Qureshi et al., 2009). In addition to participating in hemostasis, they also play an important role in the regulation of immunity and inflammation (Gawaz et al., 2005). It has been clearly documented that the specific brain pathology of AD is also reflected in platelets, including an increased membrane fluidity, abnormal cytoskeleton, cytochrome oxidase deficiency, abnormal cytoplasmic calcium flux, abnormal glutamate transporter activity, a decreased phospholipase A2 activity, an increased cytoplasmic protein kinase C level, and an increased oxidative stress level (Kawamoto et al., 2005;Vignini et al., 2007). The brain and platelets contain high concentrations of APP, and during AD, the non-amyloidogenic pathway enzyme disintegrin and metalloproteinase domain-containing protein 10 (ADAM10) are down-regulated and the amyloidogenic pathway enzyme BACE1 is up-regulated (Colciaghi et al., 2002). The activity of GSK-3β, which promotes tau hyperphosphorylation and tangle formation in the AD brains, is significantly increased in the platelet of AD and MCI patients (Veitinger et al., 2014). Mao-B, a mitochondrial protein closely related to mitochondrial damage and neuronal apoptosis, is significantly increased in the platelet of AD patients (Forlenza et al., 2011).
In addition, the platelet activation state is positively correlated with the rate of cognitive decline measured by the mini-mental state examination (MMSE) (Stellos et al., 2010). In short, platelets can reflect the AD-related pathological events and thus may serve as a perfect peripheral matrix for searching biomarkers to objectively predict AD in early stage.
Proteome has special value in studying disease-related mechanisms and diagnostic markers, which reveals disease phenotype (Lygirou et al., 2018). Compared with traditional proteomic techniques, TMT-LC-MS/MS can capture and quantify proteins in a comprehensive and efficient manner with a smaller sample requirement without offset. Recently, proteomic technology based on mass spectrometry has shown its strong power in the neurological field, such as overall analysis of protein expression level, inter-molecular correlation, and biomarker screening (Bader et al., 2020;Xiong et al., 2019).
By using TMT-LC-MS/MS, we did a comprehensive proteomic analysis in the platelets of MCI and AD patients and as well as the age/sex-matched control population. We found that multiple pathways, including AD, AMPK signaling, platelet activation, telomerase RNA localization, and complement activation, were remarkably changed in MCI and AD patients. Further PLS-DA analysis plus machine learning revealed that a combination of decreased proteins PHB, UQCRH, GP1BA, and FINC in platelets could be promising in objectively predict the cognitive decline in MCI and AD patients.

| Common differential proteins and pathways in MCI and AD platelets by whole-proteome analysis
According to the MMSE score, platelet samples of 10 cases of mild cognitive impairment (MCI; MMSE score 18-23), 9 cases of severe cognitive impairment (AD; MMSE score 2-17), and 9 age/sexmatched healthy controls (Ctrl; MMSE score 29-30) were collected for proteomic analysis (Figure 1a). The major goals of platelet proteomics data collection and their bioinformatic analysis were set as follows: (a) to analyze the changes of platelet protein profile during the progression of cognitive decline (from normal cognition to MCI to AD); (b) to clarify the biological mechanisms of platelet during the progression of cognitive decline; (c) to find MMSE-correlated proteins; and finally (d) to identify peripheral diagnostic biomarkers for cognitive impairment. To explore the dynamic changes of platelet proteome during the progression of cognitive decline, we performed cluster and proteinprotein interaction (PPI) network analyses in MCI, AD, and the cognitively normal control populations. Proteins in cluster 1 (n = 160), including CD63, PHB, UQCRH, ANXA5, and EGF, showed a decreasing tendency from Ctrl to MCI to AD (Figure 2a (Figure 2b). By PPI network analysis using MCODE (molecular complex detection) on the differentially expressed proteins, we further defined eight protein interaction modules which supported the identified pathways in the above-mentioned clusters (Figure 2c). By biological processes analyses of the differential proteins, more comprehensive and detailed biological mechanisms were shown, including regulation of insulin secretion, platelet activation (cluster 1); protein transport, cell-cell adhesion, ER to Golgi vesicle-mediated transport (cluster 2); and complement activation, protein folding (cluster 3) ( Figure 2d).
These whole-proteome data reveal the total differential proteins and the involved pathways during the progression of cognitive decline in MCI and AD.

| Differential platelet proteins or pathways correlated to MMSE score analyzed by Pearson
Mini-mental state examination (MMSE) score has been wildly used as a subjective measure of cognitive performance. To explore the periphery molecular markers that can objectively predict cognitive impairment, we performed correlation analysis of MMSE score to the entire omics data received from normal Ctrl, MCI, and AD. A total of 173 proteins were identified to be strongly correlated to Excel S4). In addition, Aβ-related protein ADAM10 was found to be strongly correlated with complement activation-related protein

F I G U R E 2 Differential proteins and biological pathways identified in MCI and AD patients compared with normal cognition controls. (a)
The protein changes were divided into three clusters according to trends from Ctrl to MCI to AD (each line represents a protein). (b) Pathway enrichment analysis of three cluster proteins with Metascape online analysis (the significantly enriched pathway has been defined as overlap proteins ≥3, p < 0.01). (c) Detected PPI modules in clusters. (d) Differential protein enrichment analysis of biological function. Differential protein enrichment analysis of biological function. The red modules, green modules, and green module represent top 3 biological process with -log 10 (p-value) in cluster 1, 2, and 3, respectively

| Selecting the best combination of platelet biomarkers to predict cognitive decline by machine learning
Further machine learning was applied to select the best combination of the biomarkers. Considering the effectiveness of the biomarkers, seven proteins involved in the complement pathway were excluded, and 19 candidate proteins were selected from 26 ( Figure 5) for subsequent sample discrimination. Partial least squares discrimination analysis (PLS-DA) to the selected 19 candidate proteins could nicely distinguish MCI and AD from the Ctrls, though it could not distinguish MCI from AD ( Figure 6A). Nine of them, including PHB, RAP1A, ITPR1, UQCRH, CD63, ADAM10, GP1BA, ITPR2, and FINC, were identified as the core contributors to distinguishing normal cognition from the cognitively impaired individuals, and PHB showed the highest differentiation with the predictive variable importance in projection larger than 1 (VIPpred >1) among the nine core candidates ( Figure 6B).
To receive the best combination of the platelet biomarkers for predicting cognitive decline, we further analyzed 9 core candidate proteins using leave-one-out (LOO) method. This method leaves out one sample at a time as validation set and uses the rest samples as the training set, so that all samples were trained n times and validated n times. By LOO analysis, various specificity and accuracy were observed using different combinations of the 9 biomarkers, and the combination of PHB, UQCRH, GP1BA, and FINC showed the highest specificity with a maximum receiver operating charac-  By using Western blotting to verify the above-mentioned nine target proteins (PHB, RAP1A, ITPR1, UQCRH, CD63, ADAM10, GP1BA, ITPR2, and FINC), we observed a decreasing trend of the levels of PHB, CD63, GP1BA, and FINC in MCI or/and AD, which was consistent with the proteomic results, but no significant decrease for the other five molecules ( Figure S1). The discrepancy might be caused by the method or/and the limited sample size.
Together, the machine leaning further selects out the combination of PHB, UQCRH, GP1BA, and FINC as the best platelet biomarkers for evaluating the cognitive decline in MCI and AD patients.

| DISCUSS ION
AD is most common neurodegenerative disorder affecting an increasing number of the populations with old age. As there is no efficient cure for this devastating disease, finding objective periphery biomarkers is extremely important for early diagnosis and drug development of AD. Integrating the existing brain/CSF proteomics (Bai et al., 2020; Wang et al., 2020), we found some interesting changes in the central and peripheral systems. Consistent with human brain and CSF proteomics, the levels of mitochondrial proteins were decreased and complement-associated proteins were increased in patients with AD (Bai et al., 2020;Wang et al., 2020). In addition, lipid metabolismrelated proteins were increased in the brain and decreased in platelets of AD (Bai et al., 2020). Importantly, we found that platelet activation, telomerase RNA localization pathway dysregulation was specific in platelets. Platelet and complement activation, calcium imbalance pathways were reported in another platelet proteomics (Donovan et al., 2013).
In addition to Aβ deposition and abnormal tau-related neurofibrillary tangle formation, AD also includes a variety of pathological changes involving calcium imbalance, autophagy defects, mitochondrial abnormalities, and synaptic damage (Grontvedt et al., 2018).
In the current study, we detected multiple enriched proteins in the AD pathways, including mitochondrial dysregulated proteins (PHB, SLC25A5, UQCRH, MPPB), Ca 2+ flow imbalance (ITPR1, ITPR2), nonamyloid protein production related proteins ADAM10, endoplasmic reticulum protein RTN4. PHB (inhibin) plays a key role in the regulation of mitochondrial protein homeostasis through the proteolytic machinery m-AAA protease in the inner mitochondrial membrane (Steglich et al., 1999). PHB also serves as a mitochondrial respiratory chain chaperone protein and the decrease of PHB induces mitochondrial dysfunction and ROS overproduction (Kathiria et al., 2012). MPPB is related to mitochondrial biogenesis (Nagayama et al., 2008), and UQCRH, as a subunit of mitochondrial respiratory chain complex III (Liu et al., 2016), cooperates with SLC25A5 (ADP/ATP transport enzyme 2) to regulate F I G U R E 5 Correlation ranking of candidate protein levels to MMSE score. (a) Negatively correlated (blue) and positively correlated (red) candidate biomarkers (p < 0.05) were ranked according to their Pearson correlation coefficients. The ratio of the color shade and the circle represent the degree of correlation. (b, c) The relative abundance of the negatively correlated (b) and positively correlated (c) proteins (brick red represents increased proteins and dark blue represents the decreased proteins)   C4BPB  C4BPA  C8A  C8G  SERPINF2   RUVB2  CD63  RAP1A  ITPR1  PHB  GP1BA  UQCRH  ITPR2  CPT1A  EGF  RTN4  ADAM10   FINC  MPPB  DCMC  SLC25A5  GNAI3  Reticulon family members can reduce Aβ generation through negative regulation of β-secretase (BACE1) (He et al., 2004;Murayama et al., 2006). Several studies have shown that the level of platelet BACE1 increases from the early stage to the late stage of AD (Colciaghi et al., 2004;Marksteiner & Humpel, 2013). Interestingly, the activity of BACE1 in platelets only increases in AD, but does not change in MCI (Bermejo-Bescos et al., 2013). Moreover, our omics data showed that ADAM10 (α-secretase) was significantly decreased in MCI, which was consistent with a previous study (Colciaghi et al., 2002). In a cohort study of the elderly in Brazil, it was found that the level of ADAM10 was continually decreased with the degree of cognitive impairment, which has the potential as a diagnostic biomarker for AD (Manzine et al., 2013). Therefore, considering the pathological connection with Aβ deposition and the significant correlation with the clinical symptoms of dementia, ADAM10 and BACE1 could serve as peripheral platelet biomarkers for early diagnosis of AD.
It is well known that patients with AD have significant energy imbalance (Yin et al., 2016), and AMPK signaling pathways play a central role in energy balance (Carling, 2017). We found here that DCMC and CPT1A, involved in lipid metabolism-related processes regulated by AMPK (Derdak et al., 2013;Xie et al., 2019), were significantly decreased in MCI and AD platelets. In a fatty liver study, pifithrinα p-nitro (PFT) can promote the expression of DCMC by regulating the SIRT1/LKB1/AMPK pathway, and the activity of CPT1A could be stimulated by reducing malonyl-CoA (mCoA) (Derdak et al., 2013). Studies showed that abnormal lipid metabolism was closely related to AD pathology (Liu et al., 2013;Wong et al., 2017). Cholesterol is an important part of axonal growth, formation, and remodeling (Liu et al., 2013). Therefore, the decreased expression of DCMC and CPT1A in peripheral platelets may be related to the abnormal lipid metabolism in MCI and AD patients. The production of bioactive products of lipid peroxidation leads to continuous platelet activation, which may contribute to amyloid deposition and complications of atherosclerotic thrombosis (Ciabattoni et al., 2007).
Consistent with the previous reports, patients with AD have significant dysregulation in the platelet activation pathway (Akingbade et al., 2018;Veitinger et al., 2014). Epidemiological data show that the increased levels of platelet activation biomarkers, activation of glycoprotein IIb-IIIa complex and P-selectin, are significantly related to cognitive decline in AD patients (Stellos et al., 2010). CD63, a member of the four-transmembrane family, is easily located in the plasma membrane from lysosome during platelet activation (Maduskuie et al., 1998), and cooperates with P-selectin to promote thrombosis in atherosclerosis (Cha et al., 2003;Yamazaki et al., 2001). Interestingly, we also found that the expression of several proteins (GP1BA, FINC, RAP1A, and VWF) involved in platelet function related to hemostasis and thrombogenesis were decreased in MCI and AD patients. RAP1A and RAP1B, important components of RAP GTPase, identify injured sites and as important switches for platelet adhesion and activation to ensure vascular integrity (Stefanini & Bergmeier, 2019). The collective reduction of the platelet activation-related proteins may affect hemostasis and maintain normal vascular function, which is consistent with the vascular risk factors of AD patients, such as diabetes, hypertension, atherosclerosis (Casserly & Topol, 2004;Helzner et al., 2009;Huo et al., 2003).
Our proteome data showed that the proteins negatively correlated with MMSE scores had strong enrichment in the complement activation pathway, suggesting a strong complement inflammatory response in the peripheral system; however, only slight increase of the complement activation pathway proteins (SERPINA1, C4BPA, C8A, C8G, SERPINF2, C1S, and C4BPB) were detected in MCI and AD patients. Recently, the complement pathway has attracted great attention, which is involved in the regulation of microglial synaptic pruning in the early stage of AD (Hong et al., 2016), and C1q-blocking antibody reverses synaptic damage in Tau-301S mice (Dejanovic et al., 2018). Brain proteomics also revealed that the complement pathway (C1R, C1S, C3, C4A, and C4B) was activated during progression of MCI into AD (Bai et al., 2020). Our data may be a good addition to illustrate the synchronous activation of the complement pathway in the peripheral and central systems.
Based on the nine candidate proteins identified from wholeproteome and MMSE correlation, we conducted further machine learning. After twenty-eight rounds of training and testing, the strict LOO algorithm revealed that combination of platelet PHB, UQCRH, GP1BA, FINC could most accurately identify the cognitive decline in MCI and AD patients. Interestingly, the four molecules identified by the machine learning algorithm represent two important pathological processes, that is, the mitochondrial dysfunction (PHB, UQCRH) and platelet activation (GP1BA, FINC).
In summary, such in-depth and comprehensive analysis of peripheral platelet protein expression profiles of MCI and AD patients has given us new understanding of the role of platelets in AD.
Bioinformatics analysis revealed that the linkage effect between peripheral and AD reflected by platelet omics involved platelet activation, complement pathway activation, mitochondrial dysfunction, calcium ion imbalance, and APP metabolic abnormality. Machine learning identified distinctive cognitive impairment-platelet combination biomarkers (PHB, UQCRH, GP1BA, and FINC). Altogether, the exploration of platelet proteomics is novel and a great supplement to understanding the peripheral changes of AD, and platelet combination biomarkers have great application potential in precision medicine for AD.

| Sample preparation
The fresh blood stored in the anticoagulant tube was centrifuged at 200 g for 20 min to remove the rich red blood cells and white blood cells from the plasma, and 2/3 of the platelet-rich supernatant was taken to the new centrifuge tube. Next, the platelet-rich plasma was centrifuged at 120 g for 6 min to remove residual white blood cells and centrifuged at 1,500 g for 10 min to obtain relatively pure platelet precipitate. Further, the platelet precipitate was washed with tyrode's solution (143.0 mM NaCl, 5.4 mM KCl, 0.25 mM NaH2PO4, 1.8 mM CaCl2, 0.5 mM MgCl2, 5.0 mM HEPES, pH 7.4; Solarbio, T1420, Beijing, China) and centrifuged at 120 g for 4 to obtain purified platelet samples and stored at −80°C.
Platelet samples were added with lysis buffer (8 M urea, pH 8.0, 1 cocktail, 1 mM PMSF) and completely lysed by ultrasound (120 s, 4 s on and 6 s off). After lysis of the ice for 30 minutes, the samples were centrifuged at 12,000 g for 10 minutes to obtain protein solution.

| Tandem mass tag (TMT) labeling
We performed a proteomic analysis of a large sample size (n = 9-10), and each sample corresponds to a TMT label (ThermoFisher 90406

| Data collection of TMT-labeled peptides using LC-MS/MS
The dried components were dissolved in 0.1% formic acid (FA),

| Bioinformatics analysis
After we normalized and filled the data on the Perseus platform, we used the t test method to calculate the p-value of the protein abundance of log2-transformed between each comparison group Characteristic Ctrls (n = 9) MCI (n = 10) AD (n = 9) p-value

| Western blot analysis
The for 50 min at room temperature. Then the membranes were treated with enhanced chemiluminescence (ECL) reagents from an ECL kit (Pierce, Thermo Scientific) for exposure.

| Statistical analysis
Statistical analysis was performed by SPSS 24.0 software (Statistical Program for Social Sciences Inc., Chicago, IL, USA). We used oneway variance analysis (ANOVA) to evaluate the statistical differences for the population information and Western blotting results, and the student's t test to compare the proteomic results of two groups. p < 0.05 was considered to be significant, and the data were expressed as mean ± SEM.

ACK N OWLED G M ENTS
This study was supported in parts by grants from Natural Science

CO N FLI C T O F I NTE R E S T
The authors declare that they have no conflict of interest to disclose.

AUTH O R S' CO NTR I B UTI O N S
Experimental design: HY, XY, and JW; recruitment of subjects and sample collection: HY, YL, BH, and TH; experimental methods: HY, YL, CC, XY, and JW; data analysis: HY, YL, CC, and JH; manuscript writing: HY, XY, and JW.

DATA AVA I L A B I L I T Y S TAT E M E N T
All data used to support the findings of this study are included within the article. Raw data used to generate the figures are available from the corresponding author upon request.