Multivariate Platelet Analysis Differentiates Between Patients with Alzheimer’s Disease and Healthy Controls at First Clinical Diagnosis

Abstract

Background:

Early diagnosis of Alzheimer’s disease (AD) is challenging, and easily accessible biomarkers are an unmet need. Blood platelets frequently serve as peripheral model for studying AD pathogenesis and might represent a reasonable biomarker source.

Objective:

In the present study, we investigated the potential to differentiate AD patients from healthy controls (HC) based on blood count, platelet morphology, and function as well as molecular markers at the time of first clinical diagnosis.

Methods:

Blood samples from 40 AD patients and 29 age-matched HC were included for determination of 78 parameter by blood counting, platelet morphometry, aggregometry, flow cytometry (CD62P, CD63, activated fibrinogen receptor), protein quantification of nicotinic acetylcholine receptor α7 (nAChRα7) and caveolin-1 (CAV-1), and miRNA quantification (miR-26b, miR-199a, miR-335). Group comparison between patients and controls was performed in univariate and multivariate statistical analyses.

Results:

AD patients showed significantly lower aggregation response to ADP and arachidonic acid and significantly decreased CD62P and CD63 surface expression induced by ADP and U46619 compared to HC. Relative nAChRα7 and CAV-1 expression was significantly higher AD platelets than in HC. Multivariate analysis of 63 parameter revealed significant differences between AD patients and healthy controls. The best performing feature model revealed a sensitivity of 96.6%, a specificity of 80.0%, and a positive predictive value of 89.3%. No grouping could be achieved by using single parameter groups.

Conclusion:

Significant differences between platelet characteristics from AD patients and HC at the time of first clinical diagnosis were observed. The best performing parameter can be used as a blood-based biomarker for AD diagnosis in a multivariate model in addition to the standardized mental tests.

Keywords

Aggregometry Alzheimer’s disease blood-based biomarker platelet morphology

INTRODUCTION

Alzheimer’s disease (AD) is a complex neurodegenerative disorder that develops gradually and progressively, with symptoms progressing over time from mild forgetfulness to severe mental impairment. Early diagnosis is challenging and the sole clinical definition of AD is only considered probable because it is based on the systematic exclusion of other potential etiologies in a patient with a dementia syndrome and not on positive proof of AD pathology. Historically, AD could be confirmed only through postmortem findings. Now, AD is defined based on clinical manifestations plus typical biomarker changes. It is now recognized that pathophysiological changes begin many years prior to clinical manifestations of disease such that the spectrum of AD spans from clinically asymptomatic to severely impaired [1]. Advances in biomarker development of amyloid pathology such as amyloid-β 1–42 (Aβ₄₂) as well as neurodegeneration (total and phosphorylated tau protein) [2, 3] have prompted a shift in how the disease is considered, with an appreciation that AD is a clinico-pathophysiological entity moving along a continuum of advancing and multifaceted clinical manifestations. Easily accessible biomarkers for core AD pathophysiology might address this challenge and are thus an urgent unmet need.

Many biomarkers from peripheral blood or urine have been discussed [4, 5] but results were inconsistent and lack reproducibility [5]. Very recently, plasma markers of inflammation and complement dysregulation were identified to support diagnosis in AD [6]. Platelets are widely used as peripheral model to study molecular pathomechanisms of AD and are considered as a source for peripheral biomarkers [7, 8]. Neuronal cells and platelets have many molecules and cellular mechanisms in common. The processing of amyloid-β protein precursor (AβPP) to amyloid-β (Aβ) peptides can be studied in platelets [9, 10]. An association between impaired platelet structure and function and AD is repeatedly reported, i.e., AD patients showed increased calcium mobilization, altered platelet aggregate formation, surface expression of P-selectin, elevated leukocyte/platelet complexes, and increased levels of urinary 11-dehydro-TxB₂ compared to healthy controls [11 –14]. Furthermore, increased platelet activation was correlated with fast cognitive decline of AD patients [15].

Our present study aimed at developing a model using multiple functional and molecular platelet parameter as a peripheral biomarker for the early diagnosis of AD. The diagnostic approach included blood cell counts, platelet morphometry, whole blood aggregometry, platelet degranulation, fibrinogen receptor activation as well as quantification of proteins and miRNAs. Platelet data from AD patients were compared to age-matched healthy controls (HC) in univariate and multivariate statistical models. To our knowledge, we are the first to examine platelets as AD biomarkers in a multi-dimensional way.

METHODS

Demographic and clinical data

Forty patients who fulfilled the NINDS-ADRDA criteria of probable AD [16] were recruited through the Memory Clinic of the Central Institute of Mental Health (Mannheim, Germany) from July 2014 until December 2017. Dementia was defined as a clinical diagnosis according to ICD-10 criteria with cognitive impairment in two or more cognitive domains severe enough to interfere with normal functioning in the community. By a detailed history and examination, other causes of an impaired cognition were excluded, e.g., depression of delirious states. Diagnostic procedures for AD dementia followed the rules of the German S3 dementia guideline of the Societies of Neurology and Psychiatry [2]. The control group (HC) included 29 healthy age-matched individuals, mostly the partners of the patients. The study protocol was approved by the local ethics committee, and written informed consent was obtained from each patient or his/her legal guardian, and each control individual before study start.

Diagnostic procedures included a detailed medical history through both patient and caregiver, a neuro-psychiatric and physical examination, and assessment of the subject’s functional level and basic cognitive functions (Mini-Mental State Examination, MMSE) [17] performed by an experienced geriatric psychiatrist. A detailed neuropsychological test battery, taken by a trained neuropsychologist, included the test battery of the Consortium to Establish a Registry for Alzheimer Disease (CERAD) [18] plus the Wechsler memory scale – logical memory (WMS) immediate and delayed recall, and the trail making test A (TMT-A) and B (TMT-B). Routine blood laboratory assessments were used to exclude any systemic illness, which might have caused secondary dementia. 3T-MRI scans were obtained in all subjects, they were quality checked and assessed visually by a trained neuroradiologist. T1-, T2-weighted and FLAIR images were used for qualitative visual rating, including medial temporal lobe atrophy, global cortical atrophy, white matter hyperintensities, and microbleeds (defined as small (<10 mm) round foci of hypointense signal in brain parenchyma). In 34 patients, cerebrospinal fluid (CSF) was obtained by lumbar puncture for exclusion of cerebral inflammation and infection and for core AD biomarkers (measured by commercially available ELISA for Aβ₄₂, Aβ₄₀, amyloid ratio, total tau, and phosphorylated tau protein). In 6 patients, CSF could not be obtained. All available information was reviewed by a multiprofessional team, who made a consensus diagnosis. Patients with mild to moderate stage of dementia and a diagnosis of AD dementia were included.

The HC group included healthy age-matched individuals and were recruited from partners of the demented subjects to ensure comparable environmental determinants and living conditions as well as a comparable age. They were subjected to a clinical evaluation similar to the procedures described above, including a general medical history, neuropsychiatric examination, and a brief assessment of cognitive function. Subjects qualified as HC if there was no functional impairment in activities of daily living and no cognitive impairment found after investigation by an experienced geriatric psychiatrist. Patients and HC were free of any medication influencing blood coagulation or known to affect platelet function such as new oral anticoagulants (NOACs), acetylsalicylic acid, nonsteroidal anti-inflammatory drugs (NSAIDs), and serotonin reuptake inhibitors. Dementia patients were untreated with any anti-dementia medication (AChE inhibitors, memantine).

Blood collection and preparation

Whole blood was collected by venipuncture with three different anti-coagulants (Sarstedt S-Monovettes^®). Hirudin-coated monovettes (525 ATE Hirudin/ml blood, 2.7 ml) were used for the whole blood aggregation assay. For preparation of platelet rich plasma (PRP) for morphometry, flow cytometry, western blot, and miRNA experiments, 3.2% (w/v) trisodium citrate with 1/10 volume was used. PRP was obtained by centrifugation of anti-coagulated blood at 150 g for 15 min. Before total RNA extraction leukocytes were depleted from PRP by adhesion filtration (Purecell PL leukocyte removal filters; Pall Medical, Dreieich, Germany) as described before [19]. EDTA anti-coagulated blood was collected for blood count analysis.

Blood counts

All blood counts were automatically measured using the hematology analyzer Cell Dyn 3200 (Abbott GmbH & Ko KG, Wiesbaden, Germany). The platelet counts included cell concentration, mean platelet volume (MPV) and platelet distribution width (PDW). The thrombocrit (Tct) was calculated from the platelet count and the MPV. For the erythrocytes cell and hemoglobin concentration was determined and for white blood cells (WBC) the total cell concentration and the differential cell counts were determined. In total, 6 measured blood count parameter and 1 calculated parameter (thrombocrit) were included for statistical analyses (Supplementary Table 1).

Platelet morphometry

Thirty minutes after centrifugation, PRP was fixated at room temperature in 0.4% HEPES-buffered formaldehyde (1 part formaldehyde 4%; Otto Fischar GmbH & Co. KG, Saarbruecken, Germany, in 9 parts HEPES buffer solution, 238 g/L; Sigma-Aldrich GmbH, Steinheim, Germany). Sample preparation, automated robotic microscopic examination and automated morphometric analysis were carried out according to the standardized protocol described earlier [20, 21]. In total, 15 morphometric parameters were included for statistical analysis (Supplementary Table 1).

Whole blood platelet aggregometry

Platelet aggregation was investigated by whole blood impedance aggregometry using the multiple platelet function analyzer (Multiplate^â Analyzer, Roche Diagnostics, Mannheim, Germany). Hirudin anti-coagulated blood of the individuals was treated either with 0.5 μM ADP (ADPtest; Roche Diagnostics), 50 mM arachidonic acid at (ASPItest; Roche Diagnostics) or with 0.75 μM U46619 (Tocris, Wiesbaden, Germany) for stimulated aggregation or with the same amount of NaCl for unstimulated spontaneous aggregation, according to Multiplate^â standard protocols. The aggregation response is given in arbitrary units (U) and area under the curve (AuC). In total, 8 measured and 3 calculated parameters from aggregometry were included for statistical analyses (Supplementary Table 1).

Flow cytometric analysis in platelets

P-selectin (CD62P) and LAMP3 (CD63) surface expression as markers for the release of α-granules and dense granules, respectively, as well as the activated platelet fibrinogen receptor complex (GPIIb/IIIa, αIIbβ3) were measured in PRP by flow cytometry (FACSCalibur^TM, BD, Heidelberg, Germany). All antibodies (mouse anti-human CD62P-FITC, Clone AK-4; mouse anti-human CD63-FITC, clone H5C6; mouse anti-human GPIIb/IIIa-FITC, Clone PAC-1) were purchased from BD Biosciences (Heidelberg, Germany). After the addition of 0.5 μM ADP (Sigma Aldrich, Taufkirchen, Germany) or 0.5 μM U46619 (agonist of the thromboxane A₂ receptor; Tocris, Wiesbaden, Germany) the PRP was incubated in a dark environment at room temperature for 20 min with saturating antibody concentrations. Based on forward and side scatters, a gate was set around the platelet population and 20,000 events were acquired from each sample. The binding of the CD62P, CD63, and PAC-1 antibodies is measured as mean fluorescence intensity of all events (MFIall) and of the events within the gate (MFIM1). The fold increase of activation markers was calculated from the ratio of gated events after stimulation to unstimulated platelets. The percentage of events within the gate from all events was also documented. In total, 27 measured and 6 calculated parameters from flow cytometry were included for statistical analyses (Supplementary Table 1).

Western blot analysis

Western blotting was performed according to standard procedures on the total protein extracts (50 μg) of platelets using specific antibodies for the α7 subtype of nicotinic acetylcholine receptors (anti-human nAChRα7 goat polyclonal antibody, clone C20; Santa Cruz Biotechnology, Heidelberg, Germany) and Caveolin-1 (anti-human CAV-1 mouse monoclonal antibody, clone 7C8; R&D Systems, Wiesbaden, Germany). For relative quantification β-actin (anti-human ACTB mouse monoclonal antibody 0411; Santa Cruz Biotechnology) was detected as reference protein. Evaluation of nAChRα7:ACTB and CAV-1:ACTB ratios were calculated on the basis of the optical density (OD) of the specific protein bands. In total, 5 parameters from protein quantification were included for statistical analyses (Supplementary Table 1).

miRNA analysis

Published data [22, 23] and databases (miRBase: http://www.mirbase.org; miRTarBase: http://mirtarbase.mbc.nctu.edu.tw) were screened for candidate miRNAs altered in AD and probably expressed in platelets. Commercial real-time PCR assays (Life Technologies, Darmstadt, Germany) were used according to the manufacturer’s protocol for quantification of candidate miRNAs (miR-26b, miR-199a, miR-335) and the ubiquitously expressed miR-15b as reference. In total, 7 parameters from miRNA quantification were included for statistical analyses (Supplementary Table 1).

Statistical analyses

In total, 6 basic parameters and 78 laboratory parameters either directly measured (n = 60) or calculated from measured values (n = 18) were used for univariate and multivariate analysis (Supplementary Table 1). Univariate statistical calculations were performed with the use of the SPSS software (SPSS Vers. 25; SPSS Inc., Chicago, IL, USA). After performing descriptive statistics, all data was tested for normal distribution with Kolmogorov-Smirnov and Shapiro-Wilk test. Homogeneity of variance was assessed with Levene test. All variables were analyzed with unpaired Student’s T-test, Mann-Whitney U-test, and Kolmogorov-Smirnov Z-test. In addition, Pearson correlation between two parameters was tested.

For the multivariate analysis, a subset of the data was used including all study subjects with complete data (n = 41:26 AD patients and 15 HC) for all measured parameters and the 3 calculated parameters from platelet morphometry. A detailed description of the multivariate statistical approach is included in the Supplementary Material. Briefly, different subsets of the 63 features were formed using sequential feature selection based on a linear discriminant model (LDA-FS) and a rank-importance using the ReliefF algorithm. For each of the seven distinct feature sets a linear discriminant model (LDA) as well as a support vector machine (SVM) with linear kernel function were trained (Fig. 1). Each model was cross-validated employing leave-one-out scheme and the predicted class membership (control or patient) was stored in the matrix C. Comparison between predicted and true class membership defines the false prediction rate for each combination of learning model and feature set. The F1 score as the harmonic mean of classification precision and recall was determined for each model/feature combination.

Fig.1

Approach of the multivariate analysis. Each subject was represented by a 63-dimensional feature vector which forms the basis for supervised learning. As some of the parameters might be more relevant for the discrimination between controls and patients than others, different subsets of the 63 features were formed using two distinct feature selection approaches: sequential feature selection based on a linear discriminant model (LDA-FS) and a rank-importance using the ReliefF algorithm. M, normalized data matrix; C, predicted class membership (control or patient) matrix.

RESULTS

Univariate statistics of demographic data and laboratory parameter

In total, 40 patients (male n = 22, female n = 18) who all were diagnosed with AD according to current German S3 guidelines for dementia were included. CSF core AD biomarkers were positive for AD pathophysiology in 28 patients, while in 6 patients, the profile showed suspected non-AD pathophysiology (Aβ₄₂≥600 pg/ml and Aβ_42/40 ratio > 0.5 and total tau > 450 pg/ml OR pTau > 60 pg/ml OR MTA score > 1) and in 6 patients, the CSF core AD biomarkers were unknown. However, in all these subjects, MTA score > 1 showed hippocampal atrophy consistent with neurodegeneration (Table 1). The HC group included 29 age-matched individuals with a slightly lower male to female ratio (male n = 13, female n = 16) compared to the AD patients (p = 0.4155). HC and AD patients did not differ significantly in age and BMI. A statistical comparison of subgroups (pathophysiology, gender) was not performed due to the limited sample size of this study. As expected, AD patients had significantly lower cognitive function compared to HC as documented by the MMSE (p < 0.0001 in all univariate tests; Table 2). Blood cell counts of AD patients in comparison to HC revealed significantly lower platelet count (U-test: p = 0.0291; Z-test: p = 0.0032) and thrombocrit (T-test: p = 0.0260; U-test: 0.0128) in AD patients (Table 2). The platelet distribution width and the MPV was similar in both study groups (Supplementary Table 2).

Table 1

CSF biomarkers for the 40 patients with clinical AD diagnosis

CSF biomarker	AD amyloid positive (28)	AD amyloid negative (6)	AD amyloid unknown (6)	p ^*
Total Tau [pg/ml]	716±525	283±142	–	0.0561
pTau [pg/ml]	87±50	47±21	–	0.0656
Tau/MTA cut-off pathologic	28	6	6
Aβ₄₂ [pg/ml]	484±138	963±267	–	<0.0001
Aβ₄₀ [pg/ml]	10.3±4.3	10.0±3.1	–	0.8987
Aβ_42/40 ratio	0.53±0.24	1.04±0.38	–	0.0002
Aβ cut-off pathologic	28	0	–

p*, t-test AD amyloid positive versus AD amyloid negative; Tau/MTA cut-off pathologic, if tTau > 400 pg/ml or pTau > 60 pg/ml or Scheltens score for medio-temporal lobe atrophy (MTA) > 1; Aβ cut-off pathologic, if Aβ₄₂ < 600 pg/ml or Aβ_42/40 ratio > 0.5.

Table 2

Basic data of the study cohort and laboratory parameter with significant differences between AD patients (n = 40) and HC (n = 29)

Parameter	AD patients mean±SD	Healthy controls mean±SD	T-test (p)*	U-test (p)	Z-test (p)
Age (y)	70.4±7.8	70.3±7.4	0.9166	0.7938	0.7914
BMI	25.6±3.5	25.2±2.8	0.6208	0.7568	0.6969
MMSE	22.3±4.4	29.5±0.7	<0.0001	<0.0001	<0.0001
Blood counts
PLT (K/μL)	232.5±58.7	255.3±59.7	0.1183	0.0291	0.0032
Thrombocrit (%)	0.19±0.05	0.22±0.05	0.0260	0.0128	0.1106
Platelet shape
Area (μm²)	11.0±2.4	9.5±2.9	0.0775	0.0882	0.0380
Extent (%)	65.3±2.5	63.5±2.8	0.0477	0.0581	0.2738
Solidity (%)	89.1±2.0	87.4±2.8	0.0307	0.0697	0.0380
Aggregation
AA (RUO)	140.5±44.6	160.8±41.1	0.0570	0.0121	0.0051
AA (AuC)	72.4±24.6	83.8±23.5	0.0570	0.0222	0.0353
ADP (RUO)	114.0±46.6	134.3±42.5	0.0681	0.0436	0.1428
ADP (AuC)	58.3±24.2	70.0±23.8	0.0501	0.0468	0.1597
Activation markers
PAC-1 ADP (fold increase)	10.3±3.0	8.1±4.1	0.0377	0.0431	0.0184
CD62P ADP (MFIM1)	133.8±35.2	160.1±34.6	0.0200	0.0240	0.0316
CD62P U46619 (MFIM1)	309.5±134.3	382.1±112.0	0.0238	0.0312	0.0493
CD63 U46619 (MFIM1)	239.6±118.1	305.0±125.6	0.0320	0.0191	0.0354
Protein expression
nAChRα7 (OD ratio)	1.25±0.21	1.04±0.17	<0.0001	0.0001	0.0027
CAV-1 (OD ratio)	1.21±0.18	1.05±0.15	0.0002	0.0005	0.0050

For description of the parameter, see Supplementary Table 1; p-values from Student’s T-test, Mann-Whitney U-Test and Kolmogorov-Smirnov Z-Test; for results from univariate statistics of all 78 parameter, see Supplementary Table 2.

The morphometric analysis of shape change parameters revealed that platelets in the patient group were larger and had a more even outline (Table 2). The average area was significantly higher in patients than in controls (11.0±2.4 μm² versus 9.5±2.9 μm²; Z-test: p = 0.0380). The increased extent (65.3±2.5% versus 63.5±2.8; t-test: p = 0.0477) and solidity (89.1±2.0% versus 87.4±2.8%; T-test: p = 0.0307; Z-test: p = 0.0380) in AD platelets demonstrated the decreased irregularity of the platelets outline in the patient group. The other shape parameters consistently confirmed this observed pattern. Spontaneous unstimulated platelet aggregation and U46619-stimulated aggregation was similar for both study groups, whereas, the ADP- and AA-induced response was significantly lower in AD patients compared to HC (see p-values for U- and Z-tests in Table 2). A significant correlation between platelet count as well as WBC count and ADP- or AA-stimulated platelet aggregation was found (Fig. 2A).

Fig.2

Correlation of platelet morphology, aggregation and activation parameter. A) Platelet and WBC counts showed significantly positive correlation with ADP- (upper panel) and AA- (lower panel) induced platelet aggregation response (AuC) in all study subjects. B) The ADP- (upper panel) and U46619- (lower panel) induced binding of the PAC-1 antibody to the activated fibrinogen receptor revealed a significantly positive correlation with the platelet extent and negative correlation with eccentricity.

Surface expression of CD62P on ADP- and U46619-stimulated platelets was significantly lower in AD patients compared to HC as documented by the MFIM1 values (p < 0.05 in all univariate tests; Table 2). Whereas, the fold increase of fibrinogen receptor activation (binding of the PAC-1 antibody) after ADP stimulation was significantly higher in AD patients than in HC (p < 0.05 in all univariate tests; Table 2). Both, ADP- and U46619-stimulated PAC-1 binding correlated significantly with descriptors of platelet shape (Supplementary Table 3), i.e., extent, circularity, and solidity showed positive correlation, whereas eccentricity and the predicted number of pseudopodia (PredNPseudoPods) were negatively correlated (Fig. 2B). The protein expression levels of nAChRα7 and CAV-1 relative to the reference protein β-actin were significantly higher in platelets from AD patients compared to HC (p < 0.05 in all univariate tests; Table 2; Fig. 3). No differences were seen for the expression of miR-26b, miR-199a, and miR-335.

Fig.3

Expression levels of the nAChRα7 and CAV-1 proteins. Left panel: representative Western blot results for the detection of the nAChRα7, CAV-1, and β-actin proteins. Right panel: calculation of nAChRα7 and CAV-1 relative to β-actin expression based on the optical density (OD) of the protein bands. Expression of nAChRα7 and CAV-1 was significantly higher in platelets from AD patients compared to HC (*p < 0.001).

Multivariate analysis

Three principal components (#3, #4, and #8) explaining 39.1% of the total variance were identified for which the contrast between control and patient group was largest. A good separation between both cohorts was evident (Fig. 4A). This was also confirmed by the group-specific distribution of the scaled Euclidean distance (dE) with relation to the center of the control group (Fig. 4B). The patient distribution is shifted to positive values whereas the dE of the control subjects is centered around zero as expected. The relative overlap between both curves was 38.3%. This can be regarded as the typical error of a classification of individual subjects based on their specific signed dE. The dE distributions in Fig. 4B were calculated using the 10 best discriminating PCs, which together explained 76.2% of the total data variance present in the matrix M (Supplementary Figure 2).

Fig.4

Multivariate analysis for differentiation between AD patients and HC. A) Three principal components (PCs #3, #4, and #8) were identified for which the contrast between control and patient group was largest. B) This is also confirmed by the group-specific distribution of the scaled Euclidean distance with relation to the center of the control group. With the 10 best PCs the relative overlap between both curves was 38.3%.

The false prediction rate from the different supervised models using the SVM learner ranged between 9.8% (F1:0.93) for the LDA-FS-Weak feature set and 34.2% (F1:0.70) for the top 5 ranked parameters (RelieF-5) (Table 3). Feature sets with smaller number of features performed quite poor as opposed to feature sets with≥15 parameters (LDA-FS-Weak, AllData), apart from the LDA classification based on the best performing subset (LDA-FS-Strong). This appears to be a general trend as the linear discriminant model was clearly outperformed by SVM in dataset with≥15 features, whereas it performed better when using only a small subset of feature space (Table 3).

Table 3

Classification results from the different supervised models using different parameter

	Parameter (n)	LDA false rate (F1)	SVM false rate (F1)
All data	63	0.3659 (0.69)	0.1951 (0.85)
RelieF-3	3	0.2439 (0.81)	0.2927 (0.76)
RelieF-5	5	0.3171 (0.73)	0.3415 (0.70)
RelieF-10	10	0.2927 (0.75)	0.2439 (0.80)
RelieF-15	15	0.2927 (0.77)	0.3415 (0.71)
LDA-FS-Weak	24	0.3659 (0.68)	0.0976 (0.93)
LDA-FS-Strong	5	0.2195 (0.82)	0.2439 (0.81)

A more detailed analysis of the best performing feature-model-combination (LDA-FS-Weak with the SVM learner) revealed that only 3.4% (1/26) of dementia patients were falsely classified as control subjects, whereas 20% (3/15) of control subjects where classified as dementia patients. This results in a sensitivity of 96.6%, a specificity of 80.0%, and a positive predictive value (PPV) of 89.3% (95% CI: 75.1%–95.8%) for the LDA-FS-Weak SVM model. We found a slightly but not significantly higher PPV for females (92.9%) compared to males (85.7%). The LDA-FS-Weak set contains a total of 24 features, from which 4 were based on geometrical platelet shape, 2 were from aggregation analysis, 16 were from flow cytometric analysis of activation markers, and 2 were from protein and miR expression analysis (Supplementary Table 4).

DISCUSSION

We measured a broad spectrum of platelet parameter in AD patients compared to age-matched HC, including platelet morphometry, aggregation, and degranulation. Significant differences between patients and controls for several platelet function parameter were observed; however, none of the differences were clinically relevant. Neither AD patients nor controls showed a tendency for increased bleeding or thromboembolism. The platelet counts of all study individuals were within the normal range (150,000 to 450,000 per μL). The descriptors of platelet shape indicated an increased activation state of unstimulated AD platelets. In contrast, after stimulation with AA, ADP or U46619 platelets from AD patients consistently showed a significantly decreased aggregation response compared to HC. Such effect has been described before as an inhibitory effect of AβPP [24, 25]. Because the platelet aggregation response correlates with platelet count [26, 27], we evaluated the effect of platelet count on the observed decreased aggregation by multiple regression analysis. The platelet count could explain 20.3% of the aggregation variance in HC, whereas in AD patients, only 8.9% of the aggregation variance could be explained. Therefore, the platelet count influenced aggregation but it was only a minor factor concerning the decreased platelet aggregation response in AD patients.

In contrast to the decreased aggregation response we showed a higher ADP-induced GPIIb/IIIa activation (by PAC-1 antibody binding) in AD patients than in HC. A higher platelet activation in AD patients was also found in other studies [12, 28]. Stellos et al. [15] made a similar observation when grouping AD patients into slow and fast cognitive decline groups. However, for the other activation markers, CD62P and CD63, we found a significantly lower expression in AD platelets than in HC. This could indicate that fibrinogen binding of platelets is increased in AD patients, but platelet degranulation is impaired or granule number or content is decreased as previously reported [29]. The altered platelet morphology and function observed in AD patients may reflect a general degenerative effect of AD on the formation of platelets in the bone marrow since a neural regulation of hematopoiesis is well known (for review, see [30]).

The nAChRα7 and CAV-1 were interesting candidates for the analysis in platelets from AD patients. Both proteins are expressed in neuronal cells and in platelets and both proteins were found upregulated in AD brain [31 –36]. It is widely accepted that nAChRα7 plays important roles in brain development and AD pathogenesis (for review, see [37]), whereas, the role of CAV-1 in the progression of AD is not fully understood. Gauderault et al. [34] showed increased CAV-1 levels in the hippocampal area and in the frontal cortex of the brain in AD patients compared to healthy controls. Whereas, Head et al. [38] stated, that low CAV-1 levels in mice brains led to premature neuronal aging and degeneration. In our study, we found both proteins nAChRα7 and CAV-1 at significantly higher expression levels in AD platelets than in HC. We like to propose the use of platelet nAChRα7 and CAV-1 levels as blood-based biomarkers for AD diagnosis.

Based on receiver operating characteristic (ROC) analysis it was shown that a combination of biomarkers, such as hippocampal volume and CSF markers (Aβ₄₀/Aβ₄₂, total and phosphorylated tau), may improve AD diagnosis [39]. In our study, we developed a multivariate statistical model that distinguished AD patients and HC based on 24 platelet parameters with a sensitivity of 96.6% and a specificity of 80%. Since biomarkers of AD had not been obtained in cognitively unimpaired control subjects, it cannot be excluded that the 3 control subjects falsely classified as AD, were actually preclinical AD [40]. All of the controls were family members/caregivers of AD patients, who were fully functioning in the community and did not report any subjective cognitive decline (a risk factor for progressive preclinical AD) [41]. Among this population, roughly 10% might be diagnosed with preclinical AD, if investigated for brain amyloid load (calculated from incidence of AD and duration of preclinical AD) [42, 43]. It remains to be shown by future studies, if our model of platelet dysfunction proves to be a state (stages of disease) or trait (AD or not) marker. At any rate, the inclusion of false negative subjects among controls would work against the hypothesis and only underestimate the PPV of our current model.

Only one AD patient was misclassified as control in our sample. Since at least 6 AD subjects of the original AD group were amyloid-negative by CSF biomarkers, the classification algorithm must be independent of amyloid pathophysiology. It remains to be determined by future studies, if relevant platelet parameters can be related to mechanisms of neurodegeneration or other aspects of AD pathophysiology. However, some care has to be taken as different models using different feature sets were tested. The results presented refer only to the best performing model. Obviously, the question what is best can only be determined after the results are known. Such an approach is always susceptible to the introduction of some amount of selection bias, especially in a study with relatively small number of subjects as ours. One should therefore conservatively estimate the true prediction rate ranging between 80% to 90% if a proper set of features is used for classification. Moreover, the number of control subjects investigated is smaller than the corresponding number of patients. Even though this asymmetry was considered by assigning different prior probabilities during model training, it cannot be fully excluded that the difference between sensitivity and specificity is due to the different group sizes. A more detailed analysis requires increasing the number of subjects investigated, including other dementia forms, and a broader range of cognitive impairment, e.g., subjective cognitive decline [44] and mild cognitive impairment due to AD [39].

We matched HC and AD patients with similar socio-economic and environmental background by recruiting the partners of AD patients. This could minimize external confounders concerning the numerous factors influencing platelet function [45]. However, conclusions from our study are limited by sample size. Our multifactoral model based on platelet parameter was developed to distinguish AD patients from HC. Since the pathophysiological process of AD is expected to begin decades before first symptoms of cognitive decline appear, further longitudinal and differential studies are required to show platelet function over normal aging and over AD onset and progression. This is crucial for fully establishing a platelet-based prediction model for AD.

Footnotes

ACKNOWLEDGMENTS

We thank Gabi Rink for expert technical assistance.

Authors’ disclosures available online ().

The supplementary material is available in the electronic version of this article: .

References

Aisen

, Cummings

, Jack

Jr , Morris

, Sperling

, Frölich

, Jones

, Dowsett

, Matthews

, Raskin

, Scheltens

, Dubois

(2017) On the path to 2025: Understanding the Alzheimer’s disease continuum. Alzheimers Res Ther 9, 60.

Deuschl

, Maier

, Jessen

, Spottke

(2016) S3-Leitlinie Demenzen. Deutsche Gesellschaft für Neurologie, Hrsg Leitlinien für Diagnostik und Therapie in der Neurologie Online: http://www.dgn.org/leitlinien.

McKhann

, Knopman

, Chertkow

, Hyman

, Jack

Jr , Kawas

, Klunk

, Koroshetz

, Manly

, Mayeux

, Mohs

, Morris

, Rossor

, Scheltens

, Carrillo

, Thies

, Weintraub

, Phelps

(2011) The diagnosis of dementia due to Alzheimer’s disease: Recommendations from the National Institute on Aging-Alzheimer’s Association workgroups on diagnostic guidelines for Alzheimer’s disease. Alzheimers Dement 7, 263–269.

Akingbade

OES

, Gibson

, Kalaria

, Mukaetova-Ladinska

(2018) Platelets: Peripheral biomarkers of dementia? J Alzheimers Dis 63, 1235–1259.

O’Bryant

, Mielke

, Rissman

, Lista

, Vanderstichele

, Zetterberg

, Lewczuk

, Posner

, Hall

, Johnson

, Fong

, Luthman

, Jeromin

, Batrla-Utermann

, Villarreal

, Britton

, Snyder

, Henriksen

, Grammas

, Gupta

, Martins

, Hampel

; Biofluid Based Biomarker Professional Interest Area (2017) Blood-based biomarkers in Alzheimer disease: Current state of the science and a novel collaborative paradigm for advancing from discovery to clinic. Alzheimers Dement 13, 45–58.

Morgan

, Touchard

, Leckey

, O’Hagan

, Nevado-Holgado

; NIMA Consortium, Barkhof

, Bertram

, Blin

, Bos

, Dobricic

, Engelborghs

, Frisoni

, Frölich

, Gabel

, Johannsen

, Kettunen

, Kłoszewska

, Legido-Quigley

, Lleó

, Martinez-Lage

, Mecocci

, Meersmans

, Molinuevo

, Peyratout

, Popp

, Richardson

, Sala

, Scheltens

, Streffer

, Soininen

, Tainta-Cuezva

, Teunissen

, Tsolaki

, Vandenberghe

, Visser

, Vos

, Wahlund

, Wallin

, Westwood

, Zetterberg

, Lovestone

, Morgan

; Annex: NIMA–Wellcome Trust Consortium for Neuroimmunology of Mood Disorders and Alzheimer’s Disease (2019) Inflammatory biomarkers in Alzheimer’s disease plasma. Alzheimers Dement 15, 776–787.

Talib

, Joaquim

, Forlenza

(2012) Platelet biomarkers in Alzheimer’s disease. World J Psychiatry 2, 95–101.

Selkoe

(2001) Alzheimer’s disease: Genes, proteins, and therapy. Physiol Rev 81, 741–766.

Colciaghi

, Marcello

, Borroni

, Zimmermann

, Caltagirone

, Cattabeni

, Padovani

, Di Luca

(2004) Platelet APP, ADAM 10 and BACE alterations in the early stages of Alzheimer disease. Neurology 62, 498–501.

10.

Tang

, Hynan

, Baskin

, Rosenberg

(2006) Platelet amyloid precursor protein processing: A bio-marker for Alzheimer’s disease. J Neurol Sci 240, 53–58.

11.

Ripovi

, Platilova

, Strunecka

, Jirak

, Hoschl

(2000) Cytosolic calcium alterations in platelets of patients with early stages of Alzheimer’s disease. Neurobiol Aging 21, 729–734.

12.

Sevush

, Jy

, Horstman

, Mao

, Kolodny

, Ahn

(1998) Platelet activation in Alzheimer disease. Arch Neurol 55, 530–536.

13.

Ciabattoni

, Porreca

, Di Febbo

, Di Iorio

, Paganelli

, Bucciarelli

, Pescara

, Del Re

, Giusti

, Falco

, Sau

, Patrono

, Daví

(2007) Determinants of platelet activation in Alzheimer’s disease. Neurobiol Aging 28, 336–342.

14.

Järemo

, Milovanovic

, Buller

, Nilsson

, Winblad

(2013) P-selectin paradox and dementia of the Alzheimer type: Circulating P-selectin is increased but platelet-bound P-selectin after agonist provocation is compromised. Scand J Clin Lab Invest 73, 170–174.

15.

Stellos

, Panagiota

, Kogel

, Leyhe

, Gawaz

, Laske

(2010) Predictive value of platelet activation for the rate of cognitive decline in Alzheimer’s disease patients. J Cereb Blood Flow Metab 30, 1817–1820.

16.

McKhann

, Drachman

, Folstein

, Katzman

, Price

, Stadlan

(1984) Clinical diagnosis of Alzheimer’s disease: Report of the NINCDS-ADRDA Work Group under the auspices of Department of Health and Human Services Task Force on Alzheimer’s Disease. Neurology 34, 939–944.

17.

Folstein

, Folstein

, McHugh

(1975) “Mini-mental state”. A practical method for grading the cognitive state of patients for the clinician. J Psychiatr Res 12, 189–198.

18.

Morris

, Mohs

, Rogers

, Fillenbaum

, Heyman

(1988) Consortium to establish a registry for Alzheimer’s disease (CERAD) clinical and neuropsychological assessment of Alzheimer’s disease. Psychopharmacol Bull 24, 641–652.

19.

Bugert

, Dugrillon

, Günaydin

, Eichler

, Klüter

(2003) Messenger RNA profiling of human platelets by microarray hybridization. Thromb Haemost 90, 738–748.

20.

Kraus

, Neeb

, Strasser

(2014) Fractal and Euclidean descriptors of platelet shape. Platelets 25, 488–498.

21.

Kraus

, Seifert

, Strasser

, Gawaz

, Schäffer

, Rheinlaender

(2016) Comparative morphology analysis of live blood platelets using scanning ion conductance and robotic dark-field microscopy. Platelets 27, 541–546.

22.

Martinez

, Peplow

(2019) MicroRNAs as diagnostic and therapeutic tools for Alzheimer’s disease: Advances and limitations. Neural Regen Res 14, 242–255.

23.

Swarbrick

, Wragg

, Ghosh

, Stolzing

(2019) Systematic review of miRNA as biomarkers in Alzheimer’s disease. Mol Neurobiol 56, 6156–6167.

24.

Henry

, Li

, Galatis

, Hesse

, Multhaup

, Beyreuther

, Masters

, Cappai

(1998) Inhibition of platelet activation by the Alzheimer’s disease amyloid precursor protein. Brit J Haematol 103, 402–415.

25.

Rao

, Peller

, Knopman

, White

(1996) Physiology and function of platelets from patients with Alzheimer’s disease. Indian J Physiol Pharmacol 40, 5–14.

26.

Seyfert

, Haubelt

, Vogt

, Hellstern

(2007) Variables influencing Multiplate(TM) whole blood impedance platelet aggregometry and turbidimetric platelet aggregation in healthy individuals. Platelets 18, 199–206.

27.

Femia

, Scavone

, Lecchi

, Cattaneo

(2013) Effect of platelet count on platelet aggregation measured with impedance aggregometry (Multiplate analyzer) and with light transmission aggregometry. J Thromb Haemost 11, 2193–2196.

28.

Laske

, Sopova

, Stellos

(2012) Platelet activation in Alzheimer’s disease: From pathophysiology to clinical value. Curr Vasc Pharmacol 10, 626–630.

29.

Kessler

, Shinitzky

, Kessler

(1995) Number of platelet dense granules varies with age, schizophrenia and dementia. Dementia 6, 330–333.

30.

Hanoun

, Maryanovich

, Arnal-Estapé

, Frenette

(2015) Neural regulation of hematopoiesis, inflammation, and cancer. Neuron 86, 360–373.

31.

Schedel

, Thornton

, Schloss

, Klüter

, Bugert

(2011) Human platelets express functional alpha7-nicotinic acetylcholine receptors. Arterioscler Thromb Vasc Biol 31, 928–934.

32.

Chu

, Ma

, Lam

, Chan

, Lee

(2005) Increased alpha 7 nicotinic acetylcholine receptor protein levels in Alzheimer’s disease patients. Dement Geriatr Cogn Disord 19, 106–112.

33.

Hernandez

, Dineley

(2012) alpha7 nicotinic acetylcholine receptors in Alzheimer’s disease: Neuroprotective, neurotrophic or both? Curr Drug Targets 13, 613–622.

34.

Gaudreault

, Dea

, Poirier

(2004) Increased caveolin-1 expression in Alzheimer’s disease brain. Neurobiol Aging 25, 753–759.

35.

Kang

, Chung

, Hwang

, Murata

, Fujimoto

, Mook-Jung

, Cha

, Park

(2006) Caveolin-1 upregulation in senescent neurons alters amyloid precursor protein processing. Exp Mol Med 38, 126–133.

36.

Jayachandran

, Miller

(2003) Human platelets contain estrogen receptor alpha, caveolin-1 and estrogen receptor associated proteins. Platelets 14, 75–81.

37.

, Qian

(2019) Alpha 7 nicotinic acetylcholine receptor and its effects on Alzheimer’s disease. Neuropeptides 73, 96–106.

38.

Head

, Peart

, Panneerselvam

, Yokoyama

, Pearn

, Niesman

, Bonds

, Schilling

, Miyanohara

, Headrick

, Ali

, Roth

, Patel

(2010) Loss of caveolin-1 accelerates neurodegeneration and aging. PLoS One 5, e15697.

39.

Frölich

, Peters

, Lewczuk

, Gruber

, Teipel

, Gertz

, Jahn

, Jessen

, Kurz

, Luckhaus

, Hüll

, Pantel

, Reischies

, Schröder

, Wagner

, Rienhoff

, Wolf

, Bauer

, Schuchhardt

, Heuser

, Rüther

, Henn

, Maier

, Wiltfang

, Kornhuber

(2017) Incremental value of biomarker combinations to predict progression of mild cognitive impairment to Alzheimer’s dementia. Alzheimers Res Ther 9, 84.

40.

Sperling

, Aisen

, Beckett

, Bennett

, Craft

, Fagan

, Iwatsubo

, Jack

Jr , Kaye

, Montine

, Park

, Reiman

, Rowe

, Siemers

, Stern

, Yaffe

, Carrillo

, Thies

, Morrison-Bogorad

, Wagster

, Phelps

(2011) Toward defining the preclinical stages of Alzheimer’s disease: Recommendations from the National Institute on Aging-Alzheimer’s Association workgroups on diagnostic guidelines for Alzheimer’s disease. Alzheimers Dement 7, 280–292.

41.

Jessen

, Amariglio

, van Boxtel

, Breteler

, Ceccaldi

, Chételat

, Dubois

, Dufouil

, Ellis

, van der Flier

, Glodzik

, van Harten

, de Leon

, McHugh

, Mielke

, Molinuevo

, Mosconi

, Osorio

, Perrotin

, Petersen

, Rabin

, Rami

, Reisberg

, Rentz

, Sachdev

, de la Sayette

, Saykin

, Scheltens

, Shulman

, Slavin

, Sperling

, Stewart

, Uspenskaya

, Vellas

, Visser

, Wagner

; Subjective Cognitive Decline Initiative (SCD-I) Working Group (2014) A conceptual framework for research on subjective cognitive decline in preclinical Alzheimer’s disease. Alzheimers Dement 10, 844–852.

42.

Fiest

, Roberts

, Maxwell

, Hogan

, Smith

, Frolkis

, Cohen

, Kirk

, Pearson

, Pringsheim

, Venegas-Torres

, Jetté

(2016) The prevalence and incidence of dementia due to Alzheimer’s disease: A systematic review and meta-analysis. Can J Neurol Sci 43(Suppl 1), S51–82.

43.

Vermunt

, Sikkes

SAM

, van den Hout

, Handels

, Bos

, van der Flier

, Kern

, Ousset

, Maruff

, Skoog

, Verhey

FRJ

, Freund-Levi

, Tsolaki

, Wallin

ÅK

, Olde Rikkert

, Soininen

, Spiru

, Zetterberg

, Blennow

, Scheltens

, Muniz-Terrera

, Visser

; Alzheimer Disease Neuroimaging Initiative; AIBL Research Group; ICTUS/DSA study groups (2019) Duration of preclinical, prodromal, and dementia stages of Alzheimer’s disease in relation to age, sex, and APOE genotype. Alzheimers Dement 15, 888–898.

44.

Slot

RER

, Sikkes

SAM

, Berkhof

, Brodaty

, Buckley

, Cavedo

, Dardiotis

, Guillo-Benarous

, Hampel

, Kochan

, Lista

, Luck

, Maruff

, Molinuevo

, Kornhuber

, Reisberg

, Riedel-Heller

, Risacher

, Roehr

, Sachdev

, Scarmeas

, Scheltens

, Shulman

, Saykin

, Verfaillie

SCJ

, Visser

, Vos

SJB

, Wagner

, Wolfsgruber

, Jessen

; Alzheimer’s Disease Neuroimaging Initiative; DESCRIPA working group; INSIGHT-preAD study group; SCD-I working group, van der Flier

(2019) Subjective cognitive decline and rates of incident Alzheimer’s disease and non-Alzheimer’s disease dementia. Alzheimers Dement 15, 465–476.

45.

McEwen

(2014) The influence of diet and nutrients on platelet function. Semin Thromb Hemost 40, 214–226.