Exploring the potential of the platelet membrane proteome as a source of peripheral biomarkers for Alzheimer's disease

Participant selection

Participants were selected from the Clinical Research in Neurology (CRIN) database at the Emory Alzheimer's Disease Research Center (ADRC). Institutional Review Board (IRB) approval was obtained from Emory University and all participants gave written informed consent before being included in this study. Seven controls and seven probable AD participants were enrolled in the study (Table 1), of which five of each group were pooled for proteomic analysis, matched as closely as possible for sex and age. Control subjects were selected based on a current MMSE score greater than 27 out of a total 30 points. Patients with mild to moderate AD all had a consensus diagnosis of probable AD and were selected based on having an MMSE score between 10 and 24. Participants were selected to be as similar as possible between groups with respect to age and were matched based on whether or not they were taking aspirin. Exclusion criteria included participants on clopidogrel (for example Plavix), those with conditions that could cause an increase in platelet activation level (a history of cancer within the past 5 years, autoimmune disorders, any acute illness, or chronic illnesses such as end-stage liver disease, end-stage renal disease requiring hemodialysis, and end-stage heart failure), and patients with bleeding disorders or other blood dyscrasias.

Platelet isolation from whole blood

Whole blood (40 cc) was collected in acid citrate dextrose (ACD) using a 21 gauge butterfly needle. To help prevent platelet activation and aggregation, the tourniquet was removed after the initial needle stick and the first 5 cc of blood withdrawn was discarded. Platelet isolation from the remaining 35 cc of blood was adapted from Quereshi et al. [10]. Centrifugation times were optimized to maximize platelet yield and purity based on analysis of the sample by light microscopy after each step. Blood was centrifuged at 200 × g for 20 minutes immediately after collection to separate red and white blood cells from platelet-rich plasma. The top 2/3 of the platelet-rich plasma was collected to minimize white blood cell contamination, and transferred to a 5 ml polypropylene tube. The platelet-rich plasma was kept at room temperature and centrifuged within three hours at 120 g x 6 minutes (Eppendorf 5810 centrifuge, Eppendorf AG, Hamburg, Germany) to remove additional remaining red and white blood cells. A majority (top 2/3) of the purified platelet-rich plasma was transferred to a second 5 ml polypropylene tube and centrifuged at 1500 × g for 10 minutes. The platelet-poor plasma was removed and the platelet pellet was resuspended in 1 ml of citrate wash buffer (11 mM glucose, 128 mM NaCl, 4.3 mM NaH₂PO₄, 4.8 mM sodium citrate, 2.4 mM citric acid, pH 6.5) and recentrifuged at 120 × g for 4 minutes. The washed platelets were transferred to an Eppendorf tube and pelleted at 1500 × g for 10 minutes in an Eppendorf 5417C table-top centrifuge. The platelet pellet was frozen at -80°C in citrate wash buffer to minimize in vitro platelet activation. Platelet purity was assessed using flow cytometry (Becton Dickinson LSRII digital benchtop analyzer, Becton, Dickinson and Co., Franklin Lakes, NJ, USA). Briefly, the sample of purified platelets was stained with allophycocyanin (APC)-tagged CD45 to identify white blood cells and fluorescein isothiocyanate (FITC)-tagged anti-CD41/integrin αIIβ to identify platelets. Data were analyzed using FlowJo (version 7.6.1) software (Tree Star Inc., Ashland, OR, USA).

Membrane enrichment strategy

The membrane enrichment strategy employed was modified from previously published methods [33, 34]. Briefly, frozen platelets were thawed on ice, resuspended in a hypotonic solution containing 100 μl citrate wash buffer and 900 μl deionized water, and kept on ice for 1 hour. Following hypotonic lysis, the mixture was sonicated (Sonic Dismembrator, Thermo Fisher Scientific, Waltham, MA, USA) twice for five seconds at 20% amplitude (maximum intensity) to disrupt cell membranes and large cytoskeletal fragments. Following sonication, the whole platelet homogenate (W) was centrifuged at 1500 × g for 10 minutes (Eppendorf 5417C) to sediment any cellular debris. The supernatant (S1) was transferred to a polycarbonate ultracentrifuge tube and centrifuged at 180,000 × g for one hour at 4°C (Beckman Optima TLX ultracentrifuge, TLA 100.4 rotor, Beckman Coulter Inc., Brea, CA, USA). The supernatant (S2) containing the soluble protein fraction was removed and saved. The resulting pellet (P2) was resuspended in 1 ml of 0.1 M sodium carbonate, pH 11 with protease and phosphatase inhibitors and incubated on ice for 15 minutes to strip proteins only loosely associated with the membrane. The samples were recentrifuged at 180,000 × g for one hour at 4°C (Beckman Optima TLX ultracentrifuge, TLA 100.4 rotor). The supernatant (W1) was removed and saved and the resulting membrane-enriched, insoluble pellet (P3) was dissolved in 50 μl 8 M urea 10 mM Tris pH 7.8. Protein concentrations from each of the five fractions (W, S1, S2, W1, P3) were determined by the bicinchoninic acid (BCA) method (Pierce, Rockford, IL, USA). The different fractions obtained from the enrichment protocol were analyzed by silver stain. Briefly, protein (1 μg) was loaded from each fraction into a 10% acrylamide gel and separated by gel electrophoresis. The gel was fixed in a solution containing 50% methanol and 5% acetic acid for 10 minutes and washed with deionized water. After rinsing in 0.02% sodium thiosulfate for 1 minute, the gel was stained with 0.1% silver nitrate for 10 minutes and developed with 3% sodium carbonate, 0.05% formaldehyde solution until the bands were sufficiently stained.

Mass spectrometry, peptide identification and quantification

Protein (20 μg/case) from the membrane-enriched fraction (P3) was pooled for proteomic analysis. After pooling, samples were alkylated with 10 mM dithiothreitol (DTT) and 50 mM iodoacetamide (IAA). Total protein was loaded into a 10% acrylamide gel and separated by SDS-PAGE. Gels were stained with Coomassie blue overnight. After destaining, gel lanes were cut into three molecular weight regions. Individual gel regions were diced into 1 mm³ pieces and destained with 50% acetonitrile (ACN) and 50 mM ammonium bicarbonate until the pieces became clear. Gel slices were digested overnight with trypsin (12.5 ng/μL; Promega Corp., Madison, WI, USA) diluted 1:20 in 50 mM NH₄HCO₃ at 37°C. The following day, peptides were extracted with buffer (5% formic acid, 50% ACN), dried in a SpeedVac concentrator (Thermo Scientific) and stored at -20°C. Purified peptides were analyzed by reverse-phase liquid chromatography coupled with tandem mass spectrometry (LC-MS/MS) and each sample was analyzed in technical replicate [35]. Briefly, peptide mixtures were loaded onto a C₁₈ column [100 μm internal diameter (i.d.), 20 cm long, 2.7 μm HALO resin from Michrom Bioresources, Inc., Auburn, CA, USA] and eluted over a 10 to 30% gradient (Buffer A: 0.1% formic acid, 0.005% heptafluorobutyric acid, and 5% acetonitrile; Buffer B: 0.1% formic acid, 0.005% heptafluorobutyric acid, and 95% acetonitrile) for 90 minutes. Eluates were monitored in a MS survey scan followed by 10 data-dependent MS/MS scans on an LTQ-Orbitrap ion trap mass spectrometer (Thermo Finnigan, San Jose, CA, USA). The LTQ was used to acquire MS/MS spectra (3 m/z isolation width, 35% collision energy, 5,000 AGC target, 200 ms maximum ion time). The Orbitrap was used to collect MS scans (300 to 1600 m/z, 1,000,000 AGC target, 1,000 ms maximum ion time, resolution 30,000). All data were converted from raw files to the .dta format using ExtractMS version 2.0 (Thermo Electron, San Jose, CA, USA) and searched against human reference database downloaded from the National Center for Biotechnology Information (19 November 2008) using the SEQUEST Sorcerer algorithm (version 3.11, SAGE-N Research, San Jose, CA, USA). Searching parameters included mass tolerance of precursor ions (±50 ppm) and product ion (±0.5 m/z), partial tryptic restriction, with a dynamic mass shift for oxidized Met (+15.9949), two maximal modification sites and a maximum of two missed cleavages. Only b and y ions were considered during the database match. To evaluate the false discovery rate (FDR), all original protein sequences were reversed to generate a decoy database that was concatenated to the original database (77,764 entries) [36]. The FDR was estimated by the number of decoy matches (nd) and total number of assigned matches (nt). FDR = 2*nd/nt, assuming mismatches in the original database were the same as in the decoy database [36]. To remove false positive matches, assigned peptides were grouped by a combination of trypticity (fully and partial) and precursor ion-charge state (+2, +3 and +4). Each group was first filtered by mass accuracy (20 ppm) and by dynamically increasing correlation coefficient and ΔCn values to reduce theoretical protein FDR by the above measure to less than 1%. All MS/MS spectra for proteins identified by a single peptide were manually inspected as described previously [37]. If peptides were shared by multiple members of a protein family, the matched members were clustered into a single group. On the basis of the principle of parsimony, the group was represented by the protein with the greatest number of assigned peptides. All identified proteins (represented by the top homolog within a group (1,009 proteins identified with unique peptides) and ungrouped (1,957 total potentially identified proteins and isoforms)) are provided respectively in Table S1 and Table S2 in Additional File 1. Quantification of peptides and proteins was based on the comparison of paired peptides from AD and control samples. Ion current intensities for identified peptides were extracted in MS survey scans of high-resolution and a ratio of the peak intensities for the peptide precursor ion was calculated using in-house DQuan software as described previously [38].

Establishing candidate biomarkers with statistical and pathway analysis

Statistical analysis to evaluate the significance of the protein changes was performed as previously described with modifications [38, 39]. Relative differences in protein levels were derived from extracted ion intensities for all identified peptides and expressed as signal-to-noise ratios. A ratio of ion intensities for the peptide precursor ions from AD and control pools were calculated, log₂ transformed, and averaged to obtain a protein ratio across samples (AD/control). A null experiment was represented by a comparison of log₂-transformed protein ratios for control replicates (replicate 1/replicate 2). As predicted by the central limit theorem, the histogram of the differences (AD/control) and the null experiment (replicate 1/replicate 2) between protein log₂ ratios fit Gaussian distributions, which enabled evaluation of systematic bias according to the mean and biological variation based on standard deviation (SD). A Gaussian curve for the binned frequencies of log₂ ratios was determined using Igor Pro v6.1 (WaveMetrics, Inc., Lake Oswego, OR, USA) and the mean was subtracted from all log₂ ratios to center the population of average AD/control log2 ratio and control 1/control 2 at zero. The SD was determined via the Gaussian width as 0.72 log₂ ratio units for AD/control and 0.30 for the null experiment (control 1/control 2). A total of 144 proteins were considered significantly changed and met the following criteria, i) fell outside the null experiment distribution or beyond 99.9% confidence interval (3.29-fold SD), ii) had an absolute value ≥1.17, (3.55 times the standard deviation of the null experiment), iii) had a coefficient of variation (CV) of less than 100% and iv) had a signal-to-noise ratio greater than 10 in both control measurements. True biomarker FDR was estimated by counting the false positives surviving the above filters in the null experiment and calculating this number as a percentage of the total number of biomarkers proposed in the list for the AD/control comparison, as described in results. The list of gene symbols for these proteins was input into DAVID pathway analysis v6.7 [40], and significantly changing ontological classes of proteins were further considered in the context of available references which related each class to AD.

Antibodies

Primary antibodies used in these studies were as follows: FITC-conjugated CD41/integrin αIIb (1:1000, mouse monoclonal (SZ.22); Abcam, Cambridge, MA, USA); APC-conjugated CD45 (1:1000, mouse monoclonal; BD Pharmingen, Franklin Lakes, NJ, USA); THBS1 (1:1000, mouse monoclonal; Thermo Fischer Scientific, Waltham, MA, USA); beta-actin (1:1000, mouse monoclonal; Abcam, Cambridge, MA, USA); the antibody dilutions for THBS1 and beta-actin reflect prior dilution of each antibody (1:1) with glycerol.

Immunoblotting

Equal concentrations of protein from each sample were loaded into a 10% acrylamide gel and separated by SDS-PAGE. Proteins were transferred onto polyvinylidene fluoride (PVDF) Immobilon-P membranes (Millipore, Billerica, MA, USA) overnight at 4°C. Immunoblots were blocked for 2 hours at room temperature with Tris-buffered saline (TBS)/Tween and blocking buffer (5x blocking buffer ultrapure from US Biological, Salem, MA, USA) and probed for the protein of interest with a primary antibody overnight at 4°C. The following day, blots were incubated with fluorophore-conjugated secondary antibodies (1:20,000) for 1 hour in the dark. All blots were scanned and quantified using the Odyssey Infrared Imaging System (Li-Cor Biosciences, Lincoln, NE, USA). Statistical analysis was performed using a two-tailed Student's t-test.

Participant selection

Characteristics of participants with clinically diagnosed AD and controls are presented in Table 1. Controls were selected to be as similar as possible to AD patients (matched as closely as possible with regard to age and sex). As expected, there was a significant difference between the groups in MMSE scores (P = 0.01). Aspirin status was matched between groups to help control for any effect of aspirin on the platelet proteome. By matching for aspirin usage, we were able to obtain a sample more representative of the general population affected by AD. Furthermore, ideal biomarkers will change in disease independent of factors such as medications. Apolipoprotein (Apo)E genotype of selected cases was not considered, since subsequent pooling of samples has the effect of cancelling specific differences in protein abundance due to individual case variation. Quantitative proteomic analysis found no significant difference in ApoE levels between control and clinically diagnosed AD platelet membrane fractions, as described below.

Platelet isolation and membrane protein enrichment strategy

Platelets from whole blood were isolated through centrifugation as previously published [10] using a citrate buffer, which significantly minimizes in vitro platelet activation (Figure 1). To assess the purity of the isolated platelets, flow cytometry was performed after double-labeling with antibodies against the platelet-specific marker, CD41 (integrin αIIβ), and a marker for white blood cells, CD45. Results demonstrate that the platelet-enriched samples contained greater than 90% CD41 positive cells, whereas CD45 positive cells made up 1.3% of the cells (Figure 2A-C). Platelets contain an extensive intracellular membrane, an open canalicular system that serves as a reservoir for plasma membrane proteins and membrane receptors and provides a passage for granule release after activation [41]. There are also numerous membrane-bound granules in platelets, the contents of which could be more easily identified in a membrane-enriched sample. Differential centrifugation fractions obtained during enrichment of the membrane proteome prior to LC-MS/MS analysis (Figure 2D) were first visualized by silver stain of a representative sample indicating altered protein complexity in the whole (W), soluble (S2), wash (W) and membrane fraction (P3) (Figure 2E). Immunoblotting with antibodies against CD41, a platelet-specific transmembrane protein, demonstrated an average of 2.2-fold enrichment in the membrane fraction (Figure 2F). Conversely, immunoblotting with antibodies against actin, a cytoskeleton protein, demonstrated approximately 20-fold depletion in the membrane fraction compared to whole platelet lysate (averaged from three independent experiments). To assess the global enrichment of proteins in the membrane fraction, LC-MS/MS analysis was performed from equal amounts of whole platelet lysate (W) and membrane-rich fraction (P3). TMHMM 2.0 [38, 42] was used to predict the number of transmembrane domains (TMD) for each protein. In the membrane-enriched fraction, 40% (389/966) of proteins were predicted to have a TMD. This was a 2.5-fold increase from an analysis of the whole platelet proteome performed in parallel in which 17% (225/1290) of proteins identified contained a predicted TMD (Table S2 in Additional file 1). The results are consistent with a similar previously published enrichment strategy for membrane proteins from platelets [30]. The most significantly enriched and depleted proteins in the membrane fraction were determined by relative quantification using spectral counting (Table S2 in Additional file 1). These included CD41 and beta-actin, which were significantly enriched and depleted, respectively, in the membrane fraction consistent with immunoblot analysis. Together, these results indicate that the differential centrifugation approach was effective at both enriching membrane proteins and depleting soluble cytoplasmic proteins.

Label-free quantification of membrane-enriched proteome differences in AD

To determine differences between AD and control membrane samples, pooled control or AD cases were analyzed by LC-MS/MS (Table 1). Pooling samples prior to LC-MS/MS analysis has been shown to decrease intersubject variability and enhance the likelihood that any changes detected would be universal to disease [43]. Prior to pooling, each control and probable AD membrane-rich protein fraction was visualized by silver staining following 1D gel electrophoresis to confirm equal protein contributions and to demonstrate comparable purity and integrity (Figure 3A). Peptides were extracted from the samples following an in-gel tryptic digest and analyzed in technical replicate using LC-MS/MS in a data-dependent manner as described in the methods. After database searching, we identified and quantified 7,910 peptides representing 1,957 proteins (Table S2 in Additional file 1, organized by homology groups with highest to lowest total number of peptide identifications). A total of 1,009 homologous protein groups (each with unique peptides that could be used for quantification) were identified across control and AD samples (Table S1 in Additional file 1, sorted by log₂ (AD/control) fold difference from decreasing in AD to increasing). Of these, 38% (378/1009) contain at least one transmembrane domain predicted by TMHMM 2.0 [38, 42], and DAVID ontology analysis reported that 55% (559/1009) had Protein Information Resource annotations relating to 'membrane'.

To determine candidate AD platelet membrane protein biomarkers from our list of 1,009 quantified proteins, we employed an approach to estimate true FDR that fully utilizes the power of technical replicates and a null experimental comparison to quantify false positives under any given filtering criteria [39]. Relative differences in protein levels, ion intensities for identified peptides, expressed as signal-to-noise ratios, were extracted in MS survey scans of high-resolution. A ratio of ion intensities for the peptide precursor ions from AD and control LC-MS runs were calculated, log₂ transformed, and averaged to obtain a protein ratio across samples (AD/control), and a null experiment log₂-transformed ratio for control replicates (replicate 1/replicate 2). As predicted by the null hypothesis, the histogram of the differences (AD/control) and null experiment between protein log₂ ratios fit Gaussian distributions, which enabled us to evaluate systematic bias according to the mean and biological variation based on SD (Figure 3B). The null experiment has a much smaller SD (log₂ = 0.30) than the average log₂ (AD/control) population (SD = 0.72). This is consistent with high reproducibility across replicates and indicates that our quantitative bioinformatics approach has sufficient precision to detect the biological variance, which manifests as a much wider SD for the latter population. As a filtering criterion, proteins with potentially increased or decreased abundance in AD that fell outside the 99.9% two-tailed confidence interval were considered as a subgroup of interest. Increased confidence in the average of two technical replicates was obtained by restricting proteins considered significantly changed to those with a coefficient of variation (standard deviation as a percentage of the mean) of less than 100%, where this filtering criterion alone reduced false positives surviving filtering in the null experiment from 74 to 24 (Figure 3C). Further applying an additional filter for minimum signal-to-noise resulted in false positives dropping to 10 when a minimum signal-to-noise ratio of 10 was required. This translates to a FDR of 6.9%. The list of 144 significantly changed proteins corresponding to this FDR in AD relative to the control platelet membrane fraction is given in Table S3 in Additional file 1.

Changes in platelet secretion and activation observed in patients with AD

Ontologies (categorization of the list into pathways, molecular functions, keywords, cellular compartments, and so on) significantly overrepresented within the list of 144 significantly changing proteins were determined using DAVID [40] (Table 2). Fifteen proteins, or about 10% of the list, represent factors likely specific to platelets that fall into the following six overlapping categories (A to F): platelet activation (Group A; P = 0.0029), platelet alpha granules (Group B; P = 5.1 × 10^-9), secretory granules (Group C; P = 1.7 × 10^-5), the complement control module (Group D; P = 0.012), complement and coagulation cascades (Group E; P = 2.3 × 10^-5), and platelet alpha granule lumen (Group F; P = 1.8 × 10^-7). All but one of the proteins in these six categories were significantly decreased, rather than increased, in AD relative to the control pool, including α-, β-, and γ-chains of fibrinogen. Fibrinogen is involved in the coagulation cascade and is secreted by alpha granules after platelet activation. It has also been included in several panels of biomarkers for AD. According to Thambisetty et al., decreased fibrinogen in association with other changes in plasma has been associated with lower brain volumes in AD [44]. Craig-Shapiro et al. have included fibrinogen in a multiplex immunoassay panel to analyze CSF biomarkers for AD. They reported that a finding of increased fibrinogen levels in the CSF in association with changes in other proteins increases the ability of the CSF tau/Aβ₄₂ ratio to discriminate between patients with very mild to mild dementia and those who are cognitively normal [45]. Platelets release alpha granules when activated. As this study looked at a membrane-enriched fraction, this finding suggests that AD platelets have a generally decreased or exhausted reserve of alpha granules consistent with having undergone activation. We speculate that low levels of fibrinogen observed in platelets from patients with AD is complementary to the reported increase in fibrinogen infiltration into AD central nervous system (CNS) tissue associated with Aβ depositions and microglial activation [46]. Contact of platelets with amyloid aggregates has been shown to result in their activation [47], and Aβ stimulates abnormal clots of cleaved fibrinogen (fibrin) resistant to clearance [48]. These findings in combination suggest widespread AD-specific platelet activation, supported by previous studies that have reported platelet activation in individuals with AD [19–21].

The single increasing protein in Table 2, platelet glycoprotein IX (GP9), a surface protein on platelet and alpha granule membranes [49] is known to act as a receptor for von Willebrand factor [50]. This represents a novel platelet surface-expressed candidate marker that could be specifically increasing in a manner linked to AD. Surprisingly, other members of the GP9-containing transmembrane receptor complex, which has a reported stoichiometric configuration involving glycoproteins V and Ib alpha and beta chains [51], were well quantified and found to be unchanging in the AD platelet membrane proteome (Table S3 in Additional file 1). This suggests a change in the configuration of the multimeric receptor and potentially, a change in the responsiveness of platelets in AD individuals to von Willebrand factor. It is interesting to note that von Willebrand factor is well expressed in brain vascular endothelia [52]. Should an increase in GP9 correspond with an increase in platelet affinity for CNS vascular endothelial walls, this could be consistent with a causative role for increased surface GP9 on platelets in producing conditions whereby local von Willebrand factor and amyloid in CNS blood vessel endothelium stimulate alpha granule release and local fibrinogen invasion into the CNS of AD patients [46]. This hypothesis relies on the above findings and assumption, which await further validation in a broader cohort. In the remaining sections of this report, we discuss the broader subset of potential platelet membrane biomarkers found changing in probable AD beyond evidence for platelet activation, and possible insight they provide into disease mechanisms.

Validation of a decrease in platelet thrombospondin-1 (THBS1) and AD-associated changes detected in amyloidogenic proteins

THBS1 is a large, homomultimeric extracellular matrix glycoprotein with multiple signaling functions in different cellular contexts. It is secreted from platelets, and also from astrocytes in the CNS, where it may stimulate neuronal synaptogenesis [53]. In the context of platelet membranes, THBS1 promotes thrombosis in at least two ways: (1) it stimulates platelet aggregation through CD36 receptor-based inhibition of kinase signaling cascades [54], and (2) THBS1 acutely counteracts the promotion of blood flow by nitric oxide via binding to another receptor, CD47, on vascular smooth muscle cells [55, 56]. The platelet receptor CD36 was well quantified in the membrane proteome pools and found to be trending down (log₂ (AD/control) -0.48, Table S3 in Additional file 1), though not significantly.

To validate the potential AD-associated decrease in THBS1, the platelet membrane fraction from individual cases was immunoblotted with an antibody against THBS1. Validation of individual cases following proteomic analysis of pooled samples is important because sample pooling opens up the possibility that a large change in one individual could be driving the signal measured [38], despite the fact that interindividual variability generally is muted by pooling. In the pooled proteome quantitative analysis, THBS1 was decreased 75% in AD (log₂ (AD/control) -2.02) and immunoblotting confirmed this result (P = 0.0085, Figure 4). Notably, some of the cases used for validation were not included in the proteomics analysis. However, the confirmation of decreasing THBS1 across a number of individuals with clinically diagnosed AD increases the likelihood that the decrease in THBS1 observed by proteomics for the AD pool is disease specific.

Reduced THBS1 in AD platelet membranes could be consistent with complementary evidence for secreted THBS1 in CNS. Buée et al. found that THBS1 stained senile plaques in AD brains and suggested it may be involved in plaque formation [57]. Recently, Horn et al. examined the effect of human neutrophil alpha-defensins, components of the innate immune system, on platelet activation. They found that these defensins activated platelets and led to fibrinogen and THBS1 binding. Moreover, these fibrinogen and THBS1 complexes formed amyloid-like structures. Such a cascade could also play a role in AD pathogenesis [58].

Other significantly changed amyloidogenesis-associated proteins identified in the platelet membrane proteome included increased beta-2 microglobulin (B2M, log₂ (AD/control) 1.21) and decreased gelsolin (log₂ (AD/control) -1.40). Increased B2M binding to the surface of blood cells including granulocytes, lymphocytes and monocytes is characteristic of chronic hemodialysis, and co-occurs with vascular and renal amyloid deposits of this protein [59]. Notably, none of the patients involved in this analysis had end-stage renal disease or required dialysis. Consistent with a specific effect of AD on this protein, elevated B2M was reported as one of eight CSF biomarkers, which together made up a multianalyte profile that was able to distinguish both probable AD and Parkinson's disease individuals from controls [60]. Earlier, high B2M in probable AD patient CSF was also found via a proteomic approach [61].

Gelsolin is a chaperone with multiple functions that has been shown to bind to Aβ [62] and ApoE [63] and has an independent involvement in certain amyloidoses. Although it is reportedly unchanging in AD brain, it was previously identified as a plasma AD marker that correlated positively with rapidity of cognitive decline in clinically diagnosed AD patients [64]. However, by itself, a decrease in plasma gelsolin is also associated with multiple morbidities including oxygen imbalances, major trauma, malaria, and liver injury [65]. Thus, although the changes we describe for amyloidogenic proteins including THBS1, B2M, and gelsolin in the platelet membrane proteome in AD are consistent with what is known to occur in individuals diagnosed with AD, it is also apparent that alone, these protein changes are not markers with adequate specificity for AD - obviating their inclusion into broader multianalyte profiles that consider a panel of changing proteins, be it on the membranes of platelets, or in CSF.

Co-occurrence of other pooled analyte changes consistent with previous biomarker studies

Beyond the above potential markers for clinically diagnosed AD, which confirm platelet activation plus a change in each of three amyloidosis-linked proteins THBS1 (down), B2M (up), and gelsolin (down), we asked what other changes found are consistent with previously proposed AD markers or potentially linked to proteins involved in disease mechanism, albeit not necessarily through activity in platelets. By expanding this list, the results from the current study might be applied toward the development of a future blood test that utilizes a broad multianalyte profile to aid in the confirmation or diagnosis of AD with higher specificity and accuracy. In the list of 144 significantly changing proteins in AD platelet membrane fractions, we found five additional proteins that have been identified as potential biomarkers or have a function homologous to such a protein (Table 3). Mannosyl-glycoprotein acetylglucosaminyltransferase (MGAT) 4B, elevated 5.5-fold in the AD platelet membrane pool, is involved in extended glycosylation of proteins. Comparatively low expression of a functional homolog, MGAT3, was recently reported to distinguish a fraction of AD patients from controls [66]. A vacuolar protein sorting (VPS) 13C allele specified by a single intronic SNP was recently found to significantly co-occur with AD [67], and we found that there was a significant, 67%, decrease in the AD platelet membrane pool. Synthesis of an abundant membrane lipid class called plasmalogen has been found to be defective in AD, and the rate-limiting enzyme alkylglycerone phosphate synthase (AGPS) was found to be reduced in postmortem-confirmed AD brain [68]; in the platelet membrane pool in this study, AGPS was also significantly decreased, by 68%. Ferritin heavy and light chains, usually found in a 1:1 stoichiometry, increase with age in normal, but not AD brain, and a distinguishing feature of frontal cortex in AD compared to Parkinson's disease was a large, 5-fold, increase in heavy/light ferritin ratio [69]. Ferritin light chain AD/control ratio was significantly decreased nearly 4-fold (74%) in the pooled probable AD platelet membrane proteome. Finally, insulin signaling has been linked to AD pathogenesis in multiple studies, where insulin-like growth factor 1 receptor (IGF1R) expression and signaling decreases in AD brain. IGF1R signaling has been shown to reverse amyloid beta toxicity, perhaps via regulation of amyloid precursor cleavage [70]. IGF1R also significantly decreased 74% in the AD platelet membrane pools. In conclusion, the platelet membrane proteome harbors a rich pool of analytes, a number of which are changed significantly in clinically diagnosed AD and moreover in the case of some potential AD platelet-derived markers, these proteins changed consistent with previous measurements.

Ten classes of potentially novel AD biomarkers quantified in platelet membrane pools, and the case for two additional platelet biomarker candidates

Following analysis of the 144 consistently changing proteins using DAVID bioinformatics, we manually curated 10 ontological classes of potentially novel AD markers in platelets (Table S4 in Additional file 1), where these class terms (numbered below) were found in searches of existing literature to be extensively linked to AD or CNS function, and to each other. For example, a hypothesis for calcium (1) dysregulation in AD has been reviewed [71], and related to mitochondria (2) dysfunction in AD [72]. Endocytic trafficking (8), including clathrin-mediated (7) and other forms of endocytosis (6), has been linked to amyloid beta toxicity in a recently published comprehensive yeast screen [73]. Myosin motor proteins (5) are important for neuronal vesicle transport (8) [74, 75]. N-linked glycosylation (9) mediated by the isoprenoid lipid dolichol is dysregulated in AD [76, 77], thereby implicating changes in glycoproteins (10) more generally as relevant. A loss in proteasome (3) function has been linked to various neurodegenerative conditions. While an AD-specific frontal cortex ubiquitin linkage profile did not implicate a general loss of proteasome function in AD [78], it is implicated in AD via an essential role for proteasomal degradation in modulating both inflammatory signaling outside of platelets and the degradation of tau in neurons following ubiquitination, which may be antagonized by tau phosphorylation promoted by Aβ [79, 80]. Significant decreases in two pairs of interacting proteasome subunits copurifying with the membrane fraction were reliably quantified. Finally, platelets possess the capacity to undergo apoptotic cell death, and a loss of antiapoptotic factors (4), like that seen in the membrane proteome pool from platelets, could potentially precede neuronal loss during the course of AD.

Although we cannot review all the evidence linking the above classes or individual proteins to AD as potential proteins of mechanistic relevance or as biomarker candidates, one protein of interest in the platelet membrane fraction is reversion-inducing cysteine-rich protein with kazal motifs (RECK), which is decreased 91% in AD patients compared to controls. RECK is an inhibitor of matrix metalloprotease (MMP) proenzyme activation, including MMP2 [81] and MMP9 [82], but most interestingly, of the presumed alpha secretase APP cleavage enzyme ADAM10 [83]. The MMP2 and 9 extracellular matrix proteases have a prominent role in angiogenesis, but were once hypothesized to function as either alpha or beta secretases [84] and MMP9 has been proposed as a biomarker for CNS inflammation in early AD [85]. In CNS, MMP2 and MMP9 may have differential activity or localization, providing different opportunities for the degradation of Aβ. MMP9 is produced by CNS neurons and degrades Aβ [86], perhaps combating amyloid plaque accumulation, albeit at the cost of increased neuroinflammation [87]. Previously reported differences in plasma MMP2 versus MMP9 activity in AD [88] might have functional implications in whole blood only in the context of decreased platelet RECK and THBS1, which has also been reported to act as an effective inhibitor of the same MMPs [89].

A second and final example of a distinguishing protein likely bound to the surface of platelet membranes is ApoB, an important component of very low-density lipoprotein (VLDL) particles and chylomicrons, which transport postprandial triglycerides from intestine to the liver. Although no significant change occurred in other platelet-associated lipoproteins, including ApoA1 (log₂ (AD/control) -0.09), ApoE (log₂ (AD/control) 0.54), ApoO-like (log₂ (AD/control) -0.68) or ApoJ (clusterin, log₂ (AD/control) -0.64), ApoB was decreased 72% (log₂ (AD/control) -1.86) in the AD platelet membrane fraction. ApoB is a highly polymorphic protein with two forms. The mRNA of the B100 form is posttranscriptionally edited at a single base to change a glutamine-encoding codon to nonsense, resulting in a shorter B48 form [90]. An artificial mutation that only produces the B100 form lowers cholesterol levels [91] while the B48 form enriches VLDL particles with high triglyceride levels [92]. The LDL receptor binding site is determined downstream of the B48 stop codon, as determined by an R3500Q mutation in B100 that decreases LDL particle affinity for its receptor [93]. The initial report of mRNA editing also demonstrated that the expression and activity of the specific RNA editase is promoted by insulin [90]; hyperinsulinemia is a major risk factor for AD [94] and has also been linked to an increase in cognitive markers of premature brain aging in individuals without AD [95]. Upon close examination, the decrease in platelet-associated ApoB measured was driven by peptides encoded exclusively by the B100 mRNA, which are encoded after the editase-dependent stop codon at residue 2180 (data not shown). This does not rule out a general decrease in ApoB binding to platelets, where THBS1 (previously described as a significantly decreasing protein) is one of a number of platelet proteins capable of binding to both VLDL and chylomicrons [96]. However, existing evidence for elevated ApoB-48 co-occurring with high Aβ in the intestinal enterocytes that serve as the normal site for ApoB RNA editing and secretion of B-48 containing chylomicrons [97] lends support for the potential usefulness of the ApoB-48/ApoB-100 ratio associated with platelets as a potential biomarker, which should be further explored, in parallel with the alternate possibility that pan-ApoB association with platelets could be decreased. Furthermore, evidence implicates that ApoB-containing lipoprotein particles can strongly influence the activity of prothrombotic proteases [98, 99].

Throughout the discussion of our results, it is notable that the platelet membrane proteome changes are often functionally linked to the process of thrombosis. To visualize the best-established functional interactions of the putative biomarkers discussed throughout these results, we built an interaction network (Figure 5). Strikingly, most of the potential biomarkers uncovered indeed do have established functional linkage to the tightly integrated multi-hubbed network of alpha granule components.