Potential Value of Plasma Amyloid-β, Total Tau, and Neurofilament

May 30, 2019 - The objective of the study was to explore the potential value of plasma indicators for identifying amnesic mild cognitive impairment (a...
0 downloads 0 Views 2MB Size
Download PDF
Research Article Cite This: ACS Chem. Neurosci. XXXX, XXX, XXX−XXX

pubs.acs.org/chemneuro

Potential Value of Plasma Amyloid-β, Total Tau, and Neurofilament Light for Identification of Early Alzheimer’s Disease Yachen Shi,†,# Xiang Lu,†,# Linhai Zhang,§ Hao Shu,† Lihua Gu,† Zan Wang,† Lijuan Gao,† Jianli Zhu,‡ Haisan Zhang,‡ Deyu Zhou,§ and Zhijun Zhang*,†,‡ †

Department of Neurology, Affiliated ZhongDa Hospital, School of Medicine, Southeast University, Nanjing, Jiangsu 210009, China Department of Psychology, Xinxiang Medical University, Xinxiang, Henan 453003, China § School of Computer Science and Engineering, Key Laboratory of Computer Network and Information Integration, Ministry of Education, Southeast University, Nanjing, Jiangsu 210009, China Downloaded via NOTTINGHAM TRENT UNIV on July 17, 2019 at 20:51:59 (UTC). See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.

‡

S Supporting Information *

ABSTRACT: The objective of the study was to explore the potential value of plasma indicators for identifying amnesic mild cognitive impairment (aMCI) and determine whether levels of plasma indicators are related to the performance of cognitive function and brain tissue volumes. In total, 155 participants (68 aMCI patients and 87 health controls) were recruited in the present cross-sectional study. The levels of plasma amyloid-β (Aβ) 40, Aβ42, total tau (t-tau), and neurofilament light (NFL) were measured using an ultrasensitive quantitative method. Machine learning algorithms were performed for establishing an optimal model of identifying aMCI. Compared with healthy controls, Aβ40 and Aβ42 levels were lower and NFL levels were higher in plasma of aMCI patients with an exception of t-tau levels. In aMCI patients, the higher plasma Aβ40 levels were correlated with the impaired episodic memory and negative correlations were observed between plasma t-tau levels and global cognitive function and gray matter (GM) volume. In addition, the higher plasma NFL levels were correlated with reduced hippocampus volume and total GM volume of the left inferior and middle temporal gyrus. An integrated model included clinical features, hippocampus volume, and plasma Aβ42 and NFL and had the highest accuracy for detecting aMCI patients (accuracy, 74.2%). We demonstrated that plasma Aβ40, Aβ42, t-tau, and NFL may be useful to identify aMCI and correlate with cognitive decline and brain atrophy. Among these plasma indicators, Aβ42 and NFL are more valuable as key members of a peripheral biomarker panel to detect aMCI. KEYWORDS: aMCI, plasma amyloid-β, plasma t-tau, plasma NFL, memory, machine learning

■

fluid (CSF) proteins and molecular positron electron tomography (PET) imaging, blood-based biomarkers would be less invasive, more cost-effective for identifying AD, and comparatively convenient for routine screening.5,6

INTRODUCTION

The hallmark pathologies of Alzheimer’s disease (AD) involve the aberrant accumulation of amyloid-β (Aβ) peptides and neurofibrillary tangles (NFT).1,2 Individuals with amnestic mild cognitive impairment (aMCI) is considered to have an increased risk of developing AD,3 and longitudinal studies indicated that aMCI patients have an 80% risk to develop AD within 6 years of diagnosis.4 Compared with cerebrospinal © XXXX American Chemical Society

Received: February 12, 2019 Accepted: May 30, 2019 Published: May 30, 2019 A

DOI: 10.1021/acschemneuro.9b00095 ACS Chem. Neurosci. XXXX, XXX, XXX−XXX

Research Article

ACS Chemical Neuroscience Previous studies showed that plasma Aβ42 levels were strongly related to those of CSF and the degree of amyloid deposition in the brain revealed by 11C-labeled Pittsburgh compound-B (PIB)-PET.7,8 Again, although the correlation between CSF and plasma total tau (t-tau) level was not observed in some studies,9 published results supported that elevated CSF t-tau would result in the change of peripheral tau levels, and there was a correlation between plasma and CSF ttau levels.10 In addition, neuron damage released neurofilament light chain (NFL) into the extracellular space, CSF, and plasma.11 It had been proved that CSF NFL was a reliable biomarker to differentiate AD patients from healthy controls, and plasma NFL levels positively correlated with CSF NFL levels in AD patients.12,13 In the present study, the levels of Aβ40, Aβ42, t-tau, and NFL in plasma were measured in aMCI patients. Furthermore, we determined the association of four plasma indicators with the performance of cognitive function and imaging markers. Finally, we used machine learning to explore and establish an integrated model as more precise guideposts for detection of aMCI patients.

Table 1. Comparison of Clinical Features, Cognitive Function, Brain Tissue Volumes, and Plasma Indicators between Healthy Controls and aMCI Patientsa control(n = 87) age (years) gender (M/F) education (years) MMSE raw scores Composite Z Scores episodic memory information processing speed executive function visuospatial function total GM volume (mL) total WM volume (mL) total brain volume (mL) hippocampus volume (mL) plasma Aβ40 (pg/mL) plasma Aβ42 (pg/mL) plasma t-tau (pg/mL) plasma NFL (pg/mL)

■

RESULTS AND DISCUSSION Compared Clinical Features, Cognitive Function, Brain Tissue Volumes and Levels of Plasma Indicators between aMCI Patients and Healthy Controls. Table 1 and Supplementary Table 1 list the main demographic, neuropsychological features, and brain tissue volumes of aMCI patients and healthy controls. Two groups showed no significant difference in demographic information, brain tissues volumes except education years and hippocampus volume. Among neuropsychological performance, most assessments were significantly different between aMCI patients and healthy controls apart from the ROCTF test (Supplementary Table 1). Compared with healthy controls, plasma Aβ40 and Aβ42 levels were lower significantly (Table 1) and plasma NFL levels were significantly higher (Table 1) in aMCI patients. However, there were no significant difference in plasma t-tau levels between healthy controls and aMCI patients (Table 1). However, except for plasma Aβ42, other plasma indicators did not achieve statistical significance of multiple comparisons (p < 0.0125). In addition, with the analysis of voxel-based morphometry for GM volume in the whole brain, present findings showed that aMCI patients had a reduced GM volume in the bilateral hippocampus and the right parahippocampal gyrus when compared to healthy controls (Figure 1A). Relationship of Plasma Indicators with Cognitive Assessments and Brain Tissue Volumes in aMCI Patients. As displayed in Table 2, in aMCI group, the higher plasma Aβ40 and Aβ42 levels were obviously correlated with the lower Z score of episodic memory. The higher plasma t-tau levels were correlated with the worse score of MMSE and reduced total GM volume. Finally, plasma NFL levels were negatively correlated with hippocampus volume. Furthermore, the higher plasma NFL levels significantly also correlated with the reduced the GM volume of the left inferior and middle temporal gyrus by the correlation analysis of voxel-based morphometry for GM volume in the whole brain in aMCI group (Figure 1B,C). Again, in the stepwise regression analysis, the most findings of partial correlation were included in the regression models with an exception of the association between the higher plasma

aMCI (n = 68)

64.77 ± 7.40 64.53 ± 7.68 36/51 29/39 11.82 ± 2.66 10.56 ± 2.85 28.55 ± 1.16 27.26 ± 1.67 of Each Cognitive Domain 0.51 ± 0.55 −0.66 ± 0.55 0.32 ± 0.70 −0.35 ± 0.72

p-value 0.844b 0.874c 0.005b 0.000b,d 0.000b 0.002b

0.30 ± 0.58 0.25 ± 0.45

−0.35 ± 0.54 −0.32 ± 0.99

0.000b 0.004b

606.14 ± 50.39

606.21 ± 63.03

0.994b

503.30 ± 54.30

494.99 ± 58.23

0.361b

1358.33 ± 124.77

1354.42 ± 130.26

0.973b

4.51 ± 0.39

4.34 ± 0.47

0.013b

183.76 ± 61.87

157.65 ± 64.50

0.011e

8.14 ± 3.12

5.95 ± 2.60

0.000e

3.56 ± 1.84

3.71 ± 2.30

0.865e

5.80 ± 2.27

7.00 ± 3.18

0.043e

a

Abbreviations: aMCI, amnesic mild cognitive impairment; M/F, male/female; MMSE, Mini-Mental State Examination; GM, gray matter; WM, white matter; Aβ, amyloid-β; t-tau, total tau; NFL, neurofilament light. Data are presented as the mean ± stand deviation (SD). bp-values were obtained by t test. cp-values were obtained by χ2 test. dThe performance of MMSE is presented as raw scores. ep-values were obtained by Mann−Whitney U test.

Aβ42 levels and the lower Z score of episodic memory (Table 2). Diagnostic Performance of Plasma Indicators with Machine Learning Algorithms. After 64 (16 × 4) rounds of experiments, the accuracy, sensitivity, and specificity of all models were recorded (Figure 2, Supplementary Figures 1 and 2). Among all results, SVC with plasma Aβ42 and NFL as parameters simultaneously obtained the highest accuracy of identifying individuals with aMCI (accuracy, 74.2%), which had more balanced sensitivity and specificity (sensitivity, 73.5%; specificity, 74.7%) than other models (Supplementary Table 2). In the present study, we demonstrated that (1) plasma Aβ40 and Aβ42 levels were lower and plasma NFL levels were higher in aMCI patients compared with healthy controls, but plasma t-tau levels were no different between the two groups; (2) in aMCI patients, plasma Aβ40 and t-tau levels were negatively correlated with the performance of episodic memory and global cognitive function, respectively; (3) negative correlations were observed between plasma t-tau levels and total GM volume, and higher plasma NFL levels were correlated with atrophic hippocampus volume and reduced the GM volume of left inferior and middle temporal gyrus in aMCI patients; (4) by using machine learning algorithms, the model with plasma Aβ42 and NFL, clinical feature, and hippocampus volume as input had the top accuracy for detecting aMCI patients. The B

DOI: 10.1021/acschemneuro.9b00095 ACS Chem. Neurosci. XXXX, XXX, XXX−XXX

Research Article

ACS Chemical Neuroscience

Figure 1. Results from the analysis of voxel-based morphometry for GM volume in the whole brain. (A) Brain regions from a significant atrophy in aMCI patients compared with healthy controls: the right hippocampus and parahippocampal gyrus (left black arrow) and the left hippocampus (right black arrow). (B) The correlation of the level of plasma NFL negatively with the GM volume of left inferior/middle temporal gyrus (black arrow) in aMCI patients. (C) Scatterplot of the correlation between the level of plasma NFL and the mean values of GM volume extracted from the left inferior/middle temporal gyrus in aMCI patients. Abbreviations: SPM, statistical parametric maps; GM, gray matter; GRF, Gaussian random field; aMCI, amnesic mild cognitive impairment; NFL, neurofilament light. Note: GRF correction was used with a threshold of p < 0.01 in the analysis of voxel-based morphometry. All analyses were adjusted for age, gender, and education years.

precisely detect the indicator levels.14 To the best of our knowledge, this is the first study about machine learning that was used to analyze individual clinical features, MRI data, and plasma indicator levels together and establish models for identifying high-risk AD patients at the individual level. Plasma Aβ40 and Aβ42 levels were lower in aMCI patients, which are in line with several studies that also observed similar results in MCI15 or AD16 but not in other studies.17,18 Possible mechanisms of peripheral Aβ were presented in published studies, such as blood-derived Aβ crossing the blood-brain barrier (BBB) and may directly trigger cerebral amyloidosis19 or indirectly facilitate production of Aβ in the brain via the “seeds” hypothesis.20 In addition, previous studies revealed that the major clearance pathway of brain Aβ was into the plasma via BBB transporters, and this direct brain-to-blood transport happened prior to brain-to-CSF transport.21 In 2001, DeMattos and colleagues initially presented a peripheral sink theory that the reduction of peripheral Aβ is a dynamic mechanism to decrease Aβ levels in the brain.22 Besides, a serial measurement study showed that normal cognitive performance of elder people was associated with an agerelated increase in plasma Aβ levels23 and transgenic animal studies suggested that plasma Aβ levels reduced when deposition of Aβ began in the brain.24 Hence, low plasma Aβ42 levels in aMCI patients can be explained by Aβ42 clearance decrease from the brain to the peripheral blood when increasing Aβ accumulation in the brain.25 In addition, plasma Aβ40 decreased possibly due to the compensatory functions of peripheral organs (e.g., liver and kidneys) for clearance Aβ in early AD stage,26,27 and with the Alzheimer’s progression, high plasma levels of Aβ40 were associated with a high risk of

Table 2. Correlation Coefficients of Plasma Indicators with MMSE Raw Score, Four Cognitive Domains Z Scores, and Brain Tissue Volumes (mL) in aMCI Patientsa Aβ40

Aβ42

t-tau

NFL

plasma indicators’ levels (pg/ mL)

r

r

r

r

MMSE episodic memory information processing speed executive function visuospatial function total GM volume total WM volume total brain volume hippocampus volume

−0.113 −0.325c 0.004 −0.191 0.130 −0.059 −0.068 −0.058 −0.231

−0.202 −0.279b 0.080 −0.162 0.051 −0.027 −0.004 −0.001 −0.204

−0.281b 0.089 −0.114 −0.036 −0.145 −0.262b −0.027 −0.139 −0.162

−0.206 −0.243 0.177 −0.055 −0.076 −0.166 −0.033 −0.085 −0.259b

a

Abbreviations: AD, Alzheimer’s disease; aMCI, amnesic mild cognitive impairment; Aβ, amyloid-β; T-tau, total tau; NFL, neurofilament light; MMSE, Mini-Mental State Examination; GM, gray matter; WM, white matter. All analyses were adjusted for age, gender, and education years. Hippocampal volume was obtained after correcting for whole-brain size. bp < 0.05. cp < 0.01.

findings of this study suggested that these plasma indicators were promising as key members in a peripheral biomarker panel for discovering early AD. In our study, aMCI patients were chosen as research objects to explore plasma indicators in the early stagesof AD. Because blood-based biomarkers were always lower concentration and difficult to measure by conventional immunoassay technology, we used an ultrasensitive single-molecule array technology to C

DOI: 10.1021/acschemneuro.9b00095 ACS Chem. Neurosci. XXXX, XXX, XXX−XXX

Research Article

ACS Chemical Neuroscience

Figure 2. Scatterplot of the accuracy of 16 models using four machine learning algorithms. Abbreviations: LR, logistics regression; SVC, support vector classifier; DT, decision tree; NN, neural network. Note: Dashed box indicates the 10th group with the highest accuracy revealed by four machine learning algorithms.

brain regions.39 Hence, plasma NFL should be regarded as a promising indicator for identifying early AD. As well as previous studies,40,41 atrophic hippocampus volume also was observed in aMCI patients in our study. In fact, individual age, gender, and education years were important factors for the occurrence of AD,42−44 which were not ignored in individual screening of early AD. Hence, in the present study, we made clinical features and hippocampus volume as basic elements of the models and discovered that one model included plasma Aβ42 and NFL and had maximal accuracy of identifying aMCI patients. In addition, our results also suggested that measuring multifactorial biomarkers, and utilizing machine learning, may have value in distinguishing potential AD patients from normal persons. The present study has certain limitations, discussed below. (1) This study lacked amyloid-PET imaging or CSF protein as diagnostic standards. Because of a relatively small number of aMCI patients receiving lumbar puncture examination, data were not analyzed. (2) Plasma p-tau levels were not measured due to no commercialized Simoa kit for sale. We had conferred with Quanterix to optimize an existing kit for measurement of plasma p-tau. In conclusion, we showed that plasma Aβ40, Aβ42, and NFL levels may play an important role to identify aMCI patients. In addition, plasma Aβ42 and NFL had a more significant performance as key members of a blood-based biomarker panel for the screening of early AD. However, it is critical to emphasize that a longitudinal study with strong quality control procedures should be carried out to confirm continued reliability and predictive value of these plasma indicators.

dementia, especially for people who have low levels of plasma Aβ42.28 However, because of the definite mechanism of Aβ production and clearance during the AD’s process is not fully clear, in our study, the association between plasma Aβ40 levels and the performance of episodic memory in aMCI patients was not be explained adequately. In this study, plasma t-tau levels were not significantly different between the healthy controls and aMCI patients, which is consistent with published studies.9,29 Due to the proportion of plasma t-tau increase being less than CSF t-tau and obvious neuronal damage occurring in the AD stage, there was no difference in plasma t-tau levels between individuals who were a healthy control or MCI but plasma t-tau levels were significantly elevated in AD patients.9,30 Several studies suggested that plasma t-tau may represent an indicator for brain injury assessment, such as acute stroke and traumatic brain injury.31−33 In addition, our findings showed that high plasma t-tau is correlated with both worse global cognitive function and decreased total GM volume in aMCI patients, even after controlling for potentially confounding factors. These results are consistent with recent studies that also discovered an association between the higher plasma t-tau levels and memory decline or cortical thinning in MCI or AD patients.29,30,34 Taken together, further studies are needed to evaluate plasma t-tau as a preferred screening tool of early AD through follow up investigation. In accordance with previously published findings,13 we observed higher plasma NFL levels in aMCI patients compared with healthy controls. A recent study suggested that plasma NFL levels increased gradually with the AD’s progress and had good performance to distinguish AD from normal controls same as the core CSF biomarkers.35 Indeed, plasma NFL is a reliable indicator for many neurodegenerative diseases, for instance AD, progressive supranuclear palsy (PSP), frontotemporal dementia (FTD), and Parkinson’s disease (PD).36−38 Furthermore, we found an inverse correlation of plasma levels with the hippocampus volume in aMCI patients, which was similar to a previous study.26 Interestingly, our findings first revealed that in aMCI patients, plasma NFL levels were also correlated with the GM volume of the left inferior and middle temporal gyrus, which may be related to synaptic loss in these

■

METHODS

Study Participants. The present study was approved and documented by the Research Ethics Committee of the Affiliated ZhongDa Hospital, Southeast University, Nanjing, China. All participants or their relatives were informed about all details of the study and signed written informed consent. A total of 87 healthy controls and 68 aMCI patients were recruited through community health screening events and media advertisements. Each participant underwent a standardized clinical interview that included details on demographic inventory, medical history, and examination of their physical and mental health status. D

DOI: 10.1021/acschemneuro.9b00095 ACS Chem. Neurosci. XXXX, XXX, XXX−XXX

Research Article

ACS Chemical Neuroscience

176, thickness = 1.0 mm, gap = 0 mm, matrix = 256 × 256, field of view = 250 mm × 250 mm, number of excitations = 2. Image data preprocessing was performed using voxel-based morphometry 8 (VBM8; http://dbm.neuro.uni-jena.de/vbm8/) in the Statistical Parametric Maps 8 (SPM8) package (Institute of Neurology, London, U.K.) running under MATLAB (The MathWorks, Inc., Natick, MA). First, the anatomical images of all subjects were segmented into gray matter (GM), white matter (WM), and cerebrospinal fluid (CSF), whose volume values were recorded for subsequent analysis. In addition, the sum of aforesaid volume values was recorded as total brain volume of each subject to be analyzed. Next, they were normalized to the tissue probability maps by affine registration. Third, these affine-registered GM and WM segments created a customized DARTEL template and were spatially normalized in the same stereotactic space based on the Montreal Neurological Institute (MNI) template using 12 affine transformations. Fourth, on the basis of the DARTEL template, normalized GM maps of the nonlinear modulation were obtained to calculate the volume of the hippocampus, which corrected for individual whole-brain size. Fifth, the absolute GM volume of each subject was modulated by Jacobian determinants derived from the spatial normalization. Finally, all the segmented GM images were smoothed by a 4 mm Gaussian kernel. Each subject’s hippocampal volume was calculated with the following steps. First, bilateral hippocampus regions of interest (ROIs) were extracted from auto anatomical labeling (AAL) template by the WFU Pickatlas software. Second, bilateral hippocampus ROIs were coregistered to the GM and WM volume maps. Then, the GM and WM volume values of all voxels in bilateral coregistered ROIs were summarized by MATLAB. Lastly, the final hippocampal volume of each subject was recorded after summing across hemispheres for subsequent analysis. Data Analysis. Two-sample t tests (3dttest++) were performed between aMCI patients and healthy controls for differences in voxelwise way comparisons of whole-brain GM volume. The threshold was set at a corrected P < 0.01 determined by Gaussian Random Field (GRF) for multiple comparisons in the whole brain (voxel-wise P < 0.01, cluster size >631 mm3). Data Processing & Analysis for Brain Imaging (DPABI V3.1) was used for voxel-wise way correlation analysis.46 Then, partial correlation was performed for the relationship between GM volume in the whole brain and plasma NFL levels in aMCI patients, which was were adjusted for age, gender, and education years. The threshold was set at a corrected P < 0.01 determined by GRF for multiple comparisons in the whole brain (voxel-wise P < 0.01, cluster size >1738 mm3). Brain tissue volumes (total GM volume, total WM volume, total brain volume, and hippocampal volume) were analyzed in the Statistical Analysis section. Statistical Analysis. Data analyses were performed using SPSS version 22.0 (SPSS, Inc., Chicago, IL) or GraphPad Prism version 5 (GraphPad Software, San Diego, CA). The Kolmogorov−Smirnov test was used to assess the normal distribution of the data. Based on the variables’ normality, either independent sample t test or Mann− Whitney U test was employed to analyze clinical features, scores of cognitive assessments, brain tissue volumes, and levels of plasma Aβ40, Aβ42, t-tau, and NFL. In addition, an χ2 test was performed to analyze the gender ratio in compared groups. Bonferroni correction was applied to correct for multiple comparisons. The partial correlations and linear regression models with a stepwise analysis were used to investigate for a possible relation between each plasma indicator and the Z score of each cognitive domain and each brain tissue volume. Both partial correlations and linear regression models were adjusted for age, gender, and education years. Significance was defined as p < 0.05 (two tailed). Classification Analysis with Machine Learning Algorithms. Clinical features (age, gender, and education years), hippocampal volume, and four plasma indicators (Aβ40, Aβ42, t-tau, and NFL) were used as the input data for machine learning algorithms. To explore the effect of these four plasma indicators, the present study combined all possible combinations of plasma indicators with clinical

Inclusion and Exclusion Criteria. Two professional neurological physicians were responsible for the diagnosis of participants involved. Only participants with a consistent diagnosis were chosen to attend in the present study. General inclusion criteria of participants were as follows: (1) age between 55 and 80 years old, (2) education years more than eight, (3) biologically right-handed, and (4) generally in good health with adequate visual and auditory acuity. For aMCI patients, inclusion criteria also involved a certain degree of memory impairment as confirmed by the patient or his/her family members; decreased objective memory capability as recorded by an Auditory Verbal Learning Test-20 min Delayed Recall (AVLT-20 min-DR) score of less than or equal to 1.5 standard deviation (SD) of age and education-adjusted norms (cutoff of ≤4 correct responses on 12 items for participants with ≥8 years of education); Mini-Mental State Examination (MMSE) score ≥24; no or minimal impairment in activities of daily living; insufficient dementia to meet the National Institute of Neurological and Communicative Disorders and Stroke and the Alzheimer’s Disease and Related Disorders Association (NINCDS-ADRDA) Alzheimer’s Criteria.45 In addition, healthy controls were included with no cognitive complaints and had an AVLT-20 min-DR score >4 and MMSE score ≥26 for subjects with 8 or more years of education. Subjects that met any of following characteristics would be excluded from the present study: (1) clinically significant cerebrovascular disorders; (2) a history of brain trauma or other neurologic diseases (e.g., Parkinson’s disease, MS); (3) major depressive disorder or psychiatric illness; (4) abnormalities in levels of thyroid hormone, serum folate, or vitamin B12; or (5) significant medical problems (e.g., cancer or impaired function of the liver or kidneys); (6) MRI scan contraindications such as ferrous or electronic implants; (7) evidence of infarction, infection, or focal lesions in MRI images Neuropsychological Assessments. Participants’ cognitive function was assessed by two experienced investigators. The neuropsychological battery covered the global cognitive function (1 test, MMSE) and four major cognitive domains, including episodic memory (3 tests), information processing speed (4 tests), executive function (5 tests), and visuospatial function (2 tests) (see Supplementary Table 1 for a list of cognitive tests). For each participant, the raw scores of each test were initially transformed to Z scores with reference to the mean and SD of each test. Then, composite Z scores for each cognitive domain were acquired from average values of relevant cognitive tests. Measurement of Plasma Aβ40, Aβ42, t-Tau, and NFL by Simoa Assays. For each participant, peripheral venous blood was drawn after overnight fasting in EDTA-coated tubes between 8:00 and 9:00 am. Within 30 min after collection, blood was centrifuged at 1500g for 10 min at 4 °C to obtain plasma. Next, the plasma was aliquoted and stored at −80 °C. The concentration of plasma Aβ40, Aβ42, tau, and NFL were measured on the Quanterix Simoa-HD1 Platform (Simoa; Quanterix, Lexington, MA). Neurology 3-Plex A kits (included Aβ40, Aβ42, and t-tau; catalog number 101995) and NFL kits (NF-light; catalog number 102258) were purchased from Quanterix and used according to the kit protocols with minor modifications. To eliminate interassay variability as a confounding factor, all samples belonging to the same subject were run in triplicate on the same day with the same standard. The few samples with intraassay coefficients of variation >20% were measured again. All measurements were performed by professionally trained technicians who were blind to the participant’s state and clinical data. Magnetic Resonance Imaging Data Acquisition and Analysis. Data Acquisition. The magnetic resonance imaging (MRI) scan was performed using a 3.0 T (T) Trio Siemens scanner (Siemens, Erlangen, Germany). All subjects were fixed with a belt and a foam pad to minimize head movement. Headphones were used to reduce scanner noise. A rapid gradient echo sequence was prepared using three-dimensional magnetization to obtain high-resolution T1weighted anatomical images as follows: repetition time = 1900 ms, echo time = 2.48 ms, inversion time = 900 ms, flip angle = 9°, slices = E

DOI: 10.1021/acschemneuro.9b00095 ACS Chem. Neurosci. XXXX, XXX, XXX−XXX

Research Article

ACS Chemical Neuroscience features and hippocampal volume. Then, 16 kinds of models were tested. Logistics regression (LR),47 support vector classifier (SVC),48 decision tree (DT),49 and neural network (NN)50 were performed to train models. In addition, leave-one-out cross-validation was employed to measure the performance of models, where successively one subject was chosen to test the model and the rest of the subjects were chosen to train the model. After testing, accuracy, sensitivity, and specificity of each model were recorded. To maximize the performance of the algorithms, a grid search was performed to tune the parameters in every test and select the highest classification accuracy. Python version 3.7.0 (https://www.python.org/) was used to implement all the machine learning algorithms. The “sklearn” package (version 0.19.2) of software was download from http://scikitlearn.org/stable/.

■

Aging-Alzheimer’s Association workgroups on diagnostic guidelines for Alzheimer’s disease. Alzheimer's Dementia 7 (3), 257−262. (6) Jack, C. R., Jr., Bennett, D. A., Blennow, K., et al. (2018) NIAAA Research Framework: Toward a biological definition of Alzheimer’s disease. Alzheimer's Dementia 14 (4), 535−562. (7) Nakamura, A., Kaneko, N., Villemagne, V. L., et al. (2018) High performance plasma amyloid-beta biomarkers for Alzheimer’s disease. Nature 554 (7691), 249−254. (8) Tzen, K. Y., Yang, S. Y., Chen, T. F., et al. (2014) Plasma Abeta but not tau is related to brain PiB retention in early Alzheimer’s disease. ACS Chem. Neurosci. 5 (9), 830−836. (9) Zetterberg, H., Wilson, D., Andreasson, U., et al. (2013) Plasma tau levels in Alzheimer’s disease. Alzheimer's Res. Ther. 5 (2), 9. (10) Chen, Z., Mengel, D., Keshavan, A., et al. (2019) Learnings about the complexity of extracellular tau aid development of a bloodbased screen for Alzheimer’s disease. Alzheimer's Dementia 15 (3), 487−496. (11) Steinacker, P., Semler, E., Anderl-Straub, S., et al. (2017) Neurofilament as a blood marker for diagnosis and monitoring of primary progressive aphasias. Neurology 88 (10), 961−969. (12) Olsson, B., Lautner, R., Andreasson, U., et al. (2016) CSF and blood biomarkers for the diagnosis of Alzheimer’s disease: a systematic review and meta-analysis. Lancet Neurol. 15 (7), 673−684. (13) Lewczuk, P., Ermann, N., Andreasson, U., et al. (2018) Plasma neurofilament light as a potential biomarker of neurodegeneration in Alzheimer’s disease. Alzheimer's Res. Ther. 10 (1), 71. (14) Wilson, D. H., Rissin, D. M., Kan, C. W., et al. (2016) The Simoa HD-1 Analyzer: A Novel Fully Automated Digital Immunoassay Analyzer with Single-Molecule Sensitivity and Multiplexing. J. Lab. Autom. 21 (4), 533−547. (15) Hanon, O., Vidal, J. S., Lehmann, S., et al. (2018) Plasma amyloid levels within the Alzheimer’s process and correlations with central biomarkers. Alzheimer's Dementia 14 (7), 858−868. (16) Janelidze, S., Stomrud, E., Palmqvist, S., et al. (2016) Plasma beta-amyloid in Alzheimer’s disease and vascular disease. Sci. Rep. 6, 26801. (17) Pesaresi, M., Lovati, C., Bertora, P., et al. (2006) Plasma levels of beta-amyloid (1−42) in Alzheimer’s disease and mild cognitive impairment. Neurobiol. Aging 27 (6), 904−905. (18) Mayeux, R., Honig, L. S., Tang, M. X., et al. (2003) Plasma A[beta]40 and A[beta]42 and Alzheimer’s disease: relation to age, mortality, and risk. Neurology 61 (9), 1185−1190. (19) Bu, X. L., Xiang, Y., Jin, W. S., et al. (2018) Blood-derived amyloid-beta protein induces Alzheimer’s disease pathologies. Mol. Psychiatry 23 (9), 1−9. (20) Jucker, M., and Walker, L. C. (2013) Self-propagation of pathogenic protein aggregates in neurodegenerative diseases. Nature 501 (7465), 45−51. (21) Ovod, V., Ramsey, K. N., Mawuenyega, K. G., et al. (2017) Amyloid beta concentrations and stable isotope labeling kinetics of human plasma specific to central nervous system amyloidosis. Alzheimer's Dementia 13 (8), 841−849. (22) DeMattos, R. B., Bales, K. R., Cummins, D. J., Dodart, J. C., Paul, S. M., and Holtzman, D. M. (2001) Peripheral anti-A beta antibody alters CNS and plasma A beta clearance and decreases brain A beta burden in a mouse model of Alzheimer’s disease. Proc. Natl. Acad. Sci. U. S. A. 98 (15), 8850−8855. (23) Seppala, T. T., Herukka, S. K., Hanninen, T., et al. (2010) Plasma A beta42 and Abeta40 as markers of cognitive change in follow-up: a prospective, longitudinal, population-based cohort study. J. Neurol., Neurosurg. Psychiatry 81 (10), 1123−1127. (24) Kawarabayashi, T., Younkin, L. H., Saido, T. C., Shoji, M., Ashe, K. H., and Younkin, S. G. (2001) Age-dependent changes in brain, CSF, and plasma amyloid (beta) protein in the Tg2576 transgenic mouse model of Alzheimer’s disease. J. Neurosci. 21 (2), 372−381. (25) Ramanathan, A., Nelson, A. R., Sagare, A. P., and Zlokovic, B. V. (2015) Impaired vascular-mediated clearance of brain amyloid beta

ASSOCIATED CONTENT

S Supporting Information *

The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/acschemneuro.9b00095.

■

Raw scores of the cognitive tests, the results of regression analysis and machine learning algorithms data (PDF)

AUTHOR INFORMATION

Corresponding Author

*E-mail: [email protected]. Phone: 0086-2583262241. Fax: 0086-25-832851. ORCID

Zhijun Zhang: 0000-0001-5480-0888 Author Contributions #

Yachen Shi and Xiang Lu contributed equally to this work and wrote the manuscript. Zhijun Zhang designed and supervised all aspects of this study. Yachen Shi, Hao Shu, Zan Wang, Lihua Gu, Lijuan Gao, and Jianli Zhu recruited subjects and assessed the cognitive functions of the subjects. Xiang Lu and Haisan Zhang completed the MRI scans and analyzed the MRI data. Linhai Zhang and Deyu Zhou completed the machine learning analysis. Funding

This study was partly supported by the National Natural Science Foundation of China (Grants 81671046 and 81420108012), the National Key Projects for Research and Development Program of China (Grants 2016YFC1305800 and 2016YFC1305802), and Jiangsu Provincial Medical Outstanding Talent (Grant JCRCA2016006). Notes

The authors declare no competing financial interest.

■

REFERENCES

(1) Crews, L., and Masliah, E. (2010) Molecular mechanisms of neurodegeneration in Alzheimer’s disease. Hum. Mol. Genet. 19 (R1), R12−R20. (2) Mandelkow, E. M., and Mandelkow, E. (1998) Tau in Alzheimer’s disease. Trends Cell Biol. 8 (11), 425−427. (3) Petersen, R. C., Doody, R., Kurz, A., et al. (2001) Current concepts in mild cognitive impairment. Arch. Neurol. 58 (12), 1985− 1992. (4) Petersen, R. C., Roberts, R. O., Knopman, D. S., et al. (2009) Mild cognitive impairment: ten years later. Arch. Neurol. 66 (12), 1447−1455. (5) Jack, C. R., Jr., Albert, M. S., Knopman, D. S., et al. (2011) Introduction to the recommendations from the National Institute on F

DOI: 10.1021/acschemneuro.9b00095 ACS Chem. Neurosci. XXXX, XXX, XXX−XXX

Research Article

ACS Chemical Neuroscience in Alzheimer’s disease: the role, regulation and restoration of LRP1. Front. Aging Neurosci. 7, 136. (26) Wang, Y. R., Wang, Q. H., Zhang, T., et al. (2017) Associations Between Hepatic Functions 4 and Plasma Amyloid-Beta LevelsImplications for the Capacity of Liver in 5 Peripheral Amyloid-Beta Clearance. Mol. Neurobiol. 54 (3), 2338−2344. (27) Bu, X. L., Yao, A., Jiao, S. S., et al. (2015) A study on the association between infectious burden and Alzheimer's disease. European journal of neurology 22 (12), 1519−1525. (28) van Oijen, M., Hofman, A., Soares, H. D., Koudstaal, P. J., and Breteler, M. M. (2006) Plasma Abeta(1−40) and Abeta(1−42) and the risk of dementia: a prospective case-cohort study. Lancet Neurol. 5 (8), 655−660. (29) Dage, J. L., Wennberg, A. M. V., Airey, D. C., et al. (2016) Levels of tau protein in plasma are associated with neurodegeneration and cognitive function in a population-based elderly cohort. Alzheimer's Dementia 12 (12), 1226−1234. (30) Mattsson, N., Zetterberg, H., Janelidze, S., et al. (2016) Plasma tau in Alzheimer disease. Neurology 87 (17), 1827−1835. (31) Bielewicz, J., Kurzepa, J., Czekajska-Chehab, E., Stelmasiak, Z., and Bartosik-Psujek, H. (2011) Does serum Tau protein predict the outcome of patients with ischemic stroke? J. Mol. Neurosci. 43 (3), 241−245. (32) Liliang, P. C., Liang, C. L., Lu, K., et al. (2010) Relationship between injury severity and serum tau protein levels in traumatic brain injured rats. Resuscitation 81 (9), 1205−1208. (33) Noguchi-Shinohara, M., Hamaguchi, T., Nozaki, I., Sakai, K., and Yamada, M. (2011) Serum tau protein as a marker for the diagnosis of Creutzfeldt-Jakob disease. J. Neurol. 258 (8), 1464−1468. (34) Fan, L. Y., Tzen, K. Y., Chen, Y. F., et al. (2018) The Relation Between Brain Amyloid Deposition, Cortical Atrophy, and Plasma Biomarkers in Amnesic Mild Cognitive Impairment and Alzheimer’s Disease. Front. Aging Neurosci. 10, 175. (35) Mattsson, N., Andreasson, U., Zetterberg, H., and Blennow, K. (2017) Association of Plasma Neurofilament Light With Neurodegeneration in Patients With Alzheimer Disease. JAMA Neurol. 74 (5), 557−566. (36) Hall, S., Ohrfelt, A., Constantinescu, R., et al. (2012) Accuracy of a panel of 5 cerebrospinal fluid biomarkers in the differential diagnosis of patients with dementia and/or parkinsonian disorders. Arch. Neurol. 69 (11), 1445−1452. (37) Rohrer, J. D., Woollacott, I. O., Dick, K. M., et al. (2016) Serum neurofilament light chain protein is a measure of disease intensity in frontotemporal dementia. Neurology 87 (13), 1329−1336. (38) Rojas, J. C., Karydas, A., Bang, J., et al. (2016) Plasma neurofilament light chain predicts progression in progressive supranuclear palsy. Ann. Clin. Transl. Neurol. 3 (3), 216−225. (39) Scheff, S. W., Price, D. A., Schmitt, F. A., Scheff, M. A., and Mufson, E. J. (2011) Synaptic loss in the inferior temporal gyrus in mild cognitive impairment and Alzheimer’s disease. J. Alzheimer's Dis. 24 (3), 547−557. (40) Wang, Z. (2014) Characterizing early Alzheimer’s disease and disease progression using hippocampal volume and arterial spin labeling perfusion MRI. J. Alzheimer's Dis. 42 (s4), S495−S502. (41) Moon, S. W., Lee, B., and Choi, Y. C. (2018) Changes in the Hippocampal Volume and Shape in Early-Onset Mild Cognitive Impairment. Psychiatry Invest. 15, 531. (42) Mielke, M. M., Ferretti, M. T., Iulita, M. F., Hayden, K., and Khachaturian, A. S. (2018) Sex and gender in Alzheimer’s disease Does it matter? Alzheimer's Dementia 14 (9), 1101−1103. (43) Szajer, J., and Murphy, C. (2013) Education level predicts retrospective metamemory accuracy in healthy aging and Alzheimer’s disease. Journal of clinical and experimental neuropsychology 35 (9), 971−982. (44) Winblad, B., Amouyel, P., Andrieu, S., et al. (2016) Defeating Alzheimer’s disease and other dementias: a priority for European science and society. Lancet Neurol. 15 (5), 455−532.

(45) Petersen, R. C., Smith, G. E., Waring, S. C., Ivnik, R. J., Tangalos, E. G., and Kokmen, E. (1999) Mild cognitive impairment: clinical characterization and outcome. Arch. Neurol. 56 (3), 303−308. (46) Yan, C. G., Wang, X. D., Zuo, X. N., and Zang, Y. F. (2016) DPABI: Data Processing & Analysis for (Resting-State) Brain Imaging. Neuroinformatics 14 (3), 339−351. (47) Sainani, K. L. (2014) Logistic regression. PM&R 6 (12), 1157− 1162. (48) Larroza, A., Moratal, D., Paredes-Sanchez, A., et al. (2015) Support vector machine classification of brain metastasis and radiation necrosis based on texture analysis in MRI. J. Magn. Reson. Imaging 42 (5), 1362−1368. (49) Mohammadzadeh, F., Noorkojuri, H., Pourhoseingholi, M. A., Saadat, S., and Baghestani, A. R. (2015) Predicting the probability of mortality of gastric cancer patients using decision tree. Ir. J. Med. Sci. 184 (2), 277−284. (50) Darsey, J. A., Griffin, W. O., Joginipelli, S., and Melapu, V. K. (2015) Architecture and biological applications of artificial neural networks: a tuberculosis perspective. Methods Mol. Biol. (N. Y., NY, U. S.) 1260, 269−283.

G

DOI: 10.1021/acschemneuro.9b00095 ACS Chem. Neurosci. XXXX, XXX, XXX−XXX