Platelet biomarkers identifying mild cognitive impairment in type 2 diabetes patients

Abstract Type 2 diabetes mellitus (T2DM) is an independent risk factor of Alzheimer's disease (AD). Therefore, identifying periphery biomarkers correlated with mild cognitive impairment (MCI) is of importance for early diagnosis of AD. Here, we performed platelet proteomics in T2DM patients with MCI (T2DM‐MCI) and without MCI (T2DM‐nMCI). Pearson analysis of the omics data with MMSE (mini‐mental state examination), Aβ1‐42/Aβ1‐40 (β‐amyloid), and rGSK‐3β(T/S9) (total to Serine‐9‐phosphorylated glycogen synthase kinase‐3β) revealed that mitophagy/autophagy‐, insulin signaling‐, and glycolysis/gluconeogenesis pathways‐related proteins were most significantly involved. Among them, only the increase of optineurin, an autophagy‐related protein, was simultaneously correlated with the reduced MMSE score, and the increased Aβ1‐42/Aβ1‐40 and rGSK‐3β(T/S9), and the optineurin alone could discriminate T2DM‐MCI from T2DM‐nMCI. Combination of the elevated platelet optineurin and rGSK‐3β(T/S9) enhanced the MCI‐discriminating efficiency with AUC of 0.927, specificity of 86.7%, sensitivity of 85.3%, and accuracy of 0.859, which is promising for predicting cognitive decline in T2DM patients.


| INTRODUC TI ON
Type 2 diabetes mellitus (T2DM) and Alzheimer's disease (AD) are age-related disorders that affect millions of populations worldwide (Chornenkyy et al., 2019;Exalto et al., 2012). Increasing epidemiological data suggest that T2DM is an independent risk factor for AD (Huang et al., 2014;Janson et al., 2004;Strachan et al., 2011). It is also shown that T2DM patients have an increased risk of dementia (73%) compared to non-T2DM patients, and the cognitive decline seems to begin in the insulin resistance stage of prediabetes (Biessels et al., 2006;Koekkoek et al., 2015). Because of the lifestyle changes, such as diet, overweight and lack of exercise, the incidence of T2DM is rapidly increasing in recent years (Carracher et al., 2018;Kahn et al., 2014). T2DM and AD have many commonalities in pathophysiology, such as amyloidosis, oxidative stress, endothelial dysfunction, and abnormal enzyme activities (de Matos et al., 2018). It is believed that the increasing incidence of AD may be not only related to aging but also to the increasing diabetes (Prince et al., 2013).
Mild cognitive impairment (MCI) is another independent risk factor of AD. Populations with MCI generally developed into AD after decades, which provides a valuable window period for the intervention (Hodson, 2018). Aβ deposition and neurofibrillary tangles formed by the phosphorylated tau proteins are the main pathological features of AD (Jack et al., 2010). However, the accumulation of Aβ has already appeared 10-15 years before the appearance of the clinical phenotypes (Hodson, 2018). The cerebrospinal cord (CSF) level of Aβ1-42, a marker of amyloidosis, and the level of Aβ-PET are recognized as effective diagnostic biomarkers for AD (Hansson et al., 2019). However, these methods are invasive or expensive, so that they are hardly popularized in the clinic.
Many evidences suggest that platelets, the fragments shed by megakaryocytes, have many biological similarities with neurons (Chornenkyy et al., 2019;Veitinger et al., 2014). For instance, level of MAO-B, which is closely related to neuronal activity, is increased significantly in AD platelets (Forlenza et al., 2011). It is also reported that CD62P (P-selectin) in platelets is activated in AD patients (Sevush et al., 1998), while thrombin receptor activating peptide 6 (TRAP-6), a molecule related to platelet activation, is decreased in AD (Jaremo et al., 2013). Interestingly, like peripheral synaptic vesicles, platelets share many of the same secretory pathways and transporters as the synaptic terminals of neurons during neurotransmitter uptake and packaging (Kaneez & Saeed, 2009;Walther et al., 2003).
The amyloidosis-related protein BACE1 and tau hyperphosphorylation related protein glycogen synthase kinase-3β (GSK-3β), were significantly activated in AD platelets (Colciaghi et al., 2002;Veitinger et al., 2014). We have also reported that the platelet GSK-3β activity is increased in T2DM with MCI (T2DM-MCI) patients compared to T2DM without MCI (T2DM-nMCI) (Z. P. Xu et al., 2016). Therefore, platelets contain abundant information related to the central system and are stable in the peripheral region, which makes it a perfect model for exploring the peripheral biomarkers.
Proteomics is widely used in neuroscience (Bader et al., 2020;Xiong et al., 2019), due to its unique value in deciphering complex pathological mechanisms and screening diagnostic biomarkers. In the present study, we performed an in-depth and comprehensive proteomic analysis in T2DM-MCI and T2DM-nMCI patients. We found that mitophagy/autophagy, insulin signaling, and glycolysis/ gluconeogenesis pathways-related proteins were most significantly deregulated in T2DM-MCI patients with elevated levels of platelet rGSK-3β and Aβ1-42/Aβ1-40 ratio. The increase of optineurin (OPTN) alone can discriminate T2DM-MCI from T2DM-nMCI, and combination of the elevated platelet OPTN with rGSK-3β has greatly increased the discriminating efficiency.

| Participants information and their platelet protein network alterations during progression of T2DM to MCI
The platelets from two cohorts of T2DM patients were collected for candidate biomarkers screening (10 cases T2DM-nMCI, 9 cases T2DM-MCI) and their validation (30 cases T2DM-nMCI, 34 cases T2DM-MCI), respectively (Table 1).
By using TMT-LC-MS/MS proteomics, a total of 2994 platelet proteins were captured, of which 46 differentially expressed proteins (DEPs) were identified in T2DM-MCI vs. T2DM-nMCI (p < 0.05) ( Figure 1a, Excel S1). To further understand the biological function of DEPs and the signaling events, PPI network analysis was performed based on KEGG database. As shown in Figure 1b, the complex network regulation of the DEPs was mainly involved in endocytosis, peroxidase, ErbB, phosphatidylinositol signaling pathways.
These data together reveal the complex platelet protein network alterations during progression of T2DM to MCI, providing a valuable resource for their systematic discovery and validation.

| Potential platelet biomarkers associated with MCI in T2DM patients
During patient selection for this proteomic analysis, we payed special attention to ApoE genetype, olfactory function, and rGSK-3β(T/ S9), because changes of these factors are correlated to the cognitive decline in T2DM patients (Michaelson, 2014;Rahayel et al., 2012;Zhang et al., 2018). Although a strong trend of increase was shown for platelet rGSK-3β(T/S9), the difference did not reach statistical significance between T2DM-MCI and T2DM-nMCI groups  (Table 1), which may be due to the relatively small sample size. We also detected plasma levels of Aβ1-40 and Aβ1-42, because the β-amyloidosis is the common pathology in both AD and diabetes patients (de Matos et al., 2018). We found that Aβ1-40 was decreased in T2DM-MCI compared with T2DM-nMCI patients while no significant difference was detected for Aβ1-42 (p = 0.400), consequently, the ratio of Aβ1-42/Aβ1-40 was increased in T2DM-MCI vs T2DM-nMCI   The rGSK-3β-correlated proteins (n = 65) were mainly related to mitophagy, insulin, and PI3K-Akt signaling pathways (Figure 4c-d).
To describe the signal transduction pathway more intuitively, cytoscape (3.7.0) software and its wiki pathway, KEGG plug-in were used to specifically display the signal regulation pathway. As shown in Figure  These data together suggest that proteins associated with deregulated mitophagy/autophagy, insulin signaling, and glycolysis/ gluconeogenesis pathways could be potential platelet biomarkers for cognitive decline in T2DM patients.

| Combined platelet OPTN and rGSK3β elevation in discriminating T2DM-MCI from T2DM-NMCI patients
By using PLS-DA analysis, we further identified the contribution of different variables in discriminating MCI in T2DM patients ( Figure 6a). The results showed that four factors, that is OPTN, rGSK-3β, olfactory score, and GSK-3β-Ser9, have the greatest contribution in distinguishing T2DM-MCI from the T2DM-nMCI patients ( Figure 6b).

| DISCUSS ION
Type 2 diabetes mellitus (T2DM) is an independent risk factor for AD (Huang et al., 2014;Janson et al., 2004;Strachan et al., 2011), therefore, predicting who, in T2DM populations, will suffer from dementia is important for early diagnosis and intervention of AD.

By employing a highly sensitive TMT-LC-MS/MS, bioinformatics
and machine learning, we carried out a comprehensive proteomic analysis in T2DM-MCI (n = 9) and T2DM-nMCI (n = 10) patients.
A total of 4165 proteins were identified, of which 2994 were captured in each group. Further analysis demonstrated that the significantly altered platelet proteins were mainly involved in endocytosis, phosphatidylinositol signaling system, amyloidosis and peripheral nervous system, which could be the target pathways for the cognitive decline in T2DM patients. These data provide a valuable resource for exploring potential periphery platelet biomarkers and the molecular mechanisms underlying the cognitive impairments in T2DM patients.
We have recently demonstrated that platelet rGSK-3β elevation, olfactory dysfunction and APOE ε4 genetype were positively correlated to the cognitive decline in T2DM patients (Z. P. Xu et al., 2016). Here, we did not find statistical difference in these factors in proteomics screening cohort, which may be due to the relatively small sample size used for the proteomics. In the validation cohort, we found a significantly increased rGSK-3β in T2DM-MCI group with a negative correlation to MMSE score. As β-amyloidosis is comorbidity for both T2DM and AD, we measured plasma Aβ level. A significantly elevated Aβ1-42/Aβ1-40 was detected in the proteomics cohort with a negative correlation to the reduced MMSE score.
Therefore, we brought MMSE, Aβ1-42/Aβ1-40 and rGSK-3β into the following new biomarker studies. By which, we discovered that the significantly changed proteins were mainly enriched in the deregu- OPTN is an autophagy receptor, which links ubiquitinated substrates to autophagy membrane, and thereby mediates PINK1driven clearance of the damaged mitochondria (Richter et al., 2016).
OPTN also mediates clearance of Aβ (Du et al., 2017) and soluble tau through autophagy pathway, while SQSTM1, another autophagy receptor, targets clearance of insoluble tau proteins (Y. . We found in the current study that OPTN was significantly increased in platelets of T2DM-MCI patients and the hippocampus of aged 5xFAD, which was also observed in the brain of AD patients (Cho et al., 2014). PRKAA1 can mediate the binding of ubiquitin substrate linked with OPTN to MAP1LC3, by which it promotes autophagy degradation (Cho et al., 2014). It is well recognized that dysfunction of autophagy pathway leading to Aβ and tau accumulation plays a pivotal role in the chronic progression of AD pathologies (Nixon & Cataldo, 2006) (Zare-Shahabadi et al., 2015 (Fang et al., 2019). Tau accumulation can in turn aggravate autophagy deficit which forms a vicious cycle, and the autophagosome-lysosome fusion deficit caused by tau accumulation induces autophagy flow inhibition (Feng et al., 2020). Based on these observations, we speculate that inhibition of autophagy flow may be involved in OPTN-related cognitive decline, though the detailed mechanisms need further investigation.
In addition to OPTN, changes of FEMT2, RAB21, DNM1 were also observed in T2DM-MCI group. FERMT2 is a high-risk gene for AD (Karch & Goate, 2015), and epidemiological data show that it is stage-dependently associated with brain amyloidosis, and most significant in MCI (Apostolova et al., 2018). RAB21 is mainly involved in the process of endocytosis and autophagy, and it can promote γ-secretase internalization and translocation to the endosome/lysosome, and thus exacerbate Aβ production in AD (Sun et al., 2018).
Increased levels of mitochondrial fission-associated protein DNM1 promotes mitochondrial fragmentation, mitochondrial dynamics disorder, and thus exacerbating Aβ clearance disorder (Manczak et al., 2011). Therefore, further validation of these proteins in larger populations will confirm their role to be a periphery biomarker for predicting cognitive decline in T2DM patients.
It is well known that T2DM patients always show peripheral nervous damages, such as microangiopathy (Zochodne, 2007). We also observed that proteins enriched in GO term of peripheral nervous system disease, including MAP4, MTM1, MYO5A and GDAP1, were significantly increased in T2DM-MCI patients. As a family member of microtubule-associated proteins, MAP4 plays a role in stabilizing microtubules, but it is not expressed in neurons (Nguyen et al., 1997).
MTM1 is primarily involved with congenital myopathies through phosphatidylinositol signaling (Blondeau et al., 2000). MYO5A is highly expressed in the brain mainly at synapses, where it promotes the transport of AMPA glutamate receptors to the synapse and participating in the development of the synapse (Ultanir et al., 2014). transported between the brain and the periphery platelet deserves further investigation. According to the previous report (Reinhold & Rittner, 2017), we speculate that destruction of nerve barrier or brain blood barrier during T2DM progression may be involved.
Consistent with our previous findings that rGSK-3β has the highest efficiency in identifying MCI from T2DM patients, compared with other characteristic factors, such as aging, ApoE genetype, and olfaction (Z. P. Xu et al., 2016), we confirmed the role of rGSK-3β in the current study. We further identified that the rGSK- In summary, recent brain/CSF proteomic data show that autophagy pathways, glucose metabolism, and amyloidosis-related proteins are significantly dysregulated in AD patients Johnson et al., 2020;Wang et al., 2020). By an in-depth and compre- Guidelines (Albert et al., 2011), and received mini-mental State Examination (MMSE) test scores (Folstein et al., 1975)

| Sample preparation
The fresh blood stored in the anticoagulant tube was centrifuged at 200 g for 20 min to remove rich red and white blood cells from the plasma, and 2/3 of the platelets rich supernatant was brought into the new tube and centrifuged at 120 g for 6 min to remove remaining white blood cells, and centrifuged at 1500 g for 10 minutes to obtain relatively pure platelet precipitate. Further, the platelet The platelet samples were completely lysed by ultrasound (120s, 4 s on and 6 s off) after adding the lysis buffer (8 M urea, pH 8.0, 1 cocktail, 1 mM PMSF), and then lysed on ice for 30 min, and centrifuged at 12000 g for 10 minutes to obtain the pure protein solution.

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

| Bioinformatics analysis
Normalized data were uploaded to Perseus platform, and proteins with p < 0.05 were evaluated by t-test and considered differentially expressed (Bereczki et al., 2018

| Machine learning
SIMCA (version 14.0) software was used for partial least squares discrimination analysis (PLS-DA). The protein of predictive variable importance in projection larger than 1.5 (VIPpred >1.5) was considered to be meaningful for sample discrimination. Logistic regression (LR), a widely used machine-learning algorithm, was used to calculate the 95% confidence interval (95% CI) of the biomarker for diagnosis of MCI. The samples were trained and evaluated in a leave-one-out (LOO) cross-validation manner using scikit-learn python package, which was used for model training and parameter optimization in the life sciences was used for model training and parameter optimization (Bader et al., 2020;Shu et al., 2020). More specifically, 63 samples were randomly selected from 64 samples each time for modeling, and the remaining one was used for validation. Thus, 64 cycles are carried out to achieve the purpose of full data demonstration and cross-validation. Confusion matrix was used to assess the specificity and sensitivity of biomarkers in the diagnosis of MCI.

| Statistical analysis
The data were expressed as mean ±s.e.m. with SPSS 24.0 software (Statistical Program for Social Sciences Inc., Chicago, IL, USA). The student's t-test was used to evaluate the level of significance between the two groups, and p values <0.05 was considered to be significant.
For the workflow of the analytic procedure, please also see Figure S4.

ACK N OWLED G M ENTS
This study was supported in parts by grants from Science and Project of Medicine in Shenzhen (SZSM201611090).

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.

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 Proteome Xchange Consortium (http://www.prote omexc hange. org) via the PRIDE partner repository with the dataset identifiers PXD023316.