Study design

Shenzhen Bay Laboratory launched the community-based longitudinal cohort study GHABS (ClinicalTrials.gov ID: NCT06183658) in May 2021. The GHABS project was approved by the Shenzhen Bay Laboratory’s and the collaborated hospitals’ Ethical Committees. The flow chart of participant engagement in the GHABS is delineated in Supplementary Fig. 1. The GHABS participants were recruited via posters and lectures in the community and nursing homes. Each participant signed the written informed consent of the GHABS project before enrollment. The participants who met the inclusion and exclusion requirements were informed about the baseline and follow-up examinations. From 2021 to 2026, the GHABS cohort will recruit 1400 individuals aged 55 and older, including 1100 CU older adults, 200 MCI patients, and 100 dementia patients. The scheme for recruiting GHABS participants is illustrated in Supplemental Fig. 1.

All the GHABS participants will undergo cognitive assessments, genetic screening, and blood sample collection. Some will have CSF collection, stool sample collection, MRI scanning, Aβ PET scanning, and tau PET scanning. All baseline examinations will be completed within three months. At follow-up, clinical assessments and blood sample collection will be conducted annually. CSF sample collection, MRI scan, Aβ PET scan, and tau PET scan are evaluated every two years. Six hundred twenty-seven participants have completed cognitive assessments and blood sample collection in the GHABS cohort by Sep 08, 2023 (Fig. 1). Among them, 26%, 44%, 19%, and 11% of the cohort were NC, SCD, MCI, and dementia, respectively. Additionally, 369 and 93 participants completed stool and CSF sample collection. So far, 469, 370, and 105 participants had MRI, Aβ PET, and tau PET imaging scans, respectively. This study identified 470 GHABS participants who simultaneously completed cognitive assessments, plasma Aβ42/Aβ40, p-Tau181, NfL, and GFAP data measured by the Simoa platform. Among them, 119, 207, 87, and 57 were NC, SCD, MCI, and dementia. Moreover, 301, 195, and 70 individuals had concurrent MRI, 18F-D3-FSP Aβ PET, and FTP tau PET scans.

Fig. 1

The current sample sizes of neurocognitive assessments, blood samples, CSF samples, stool samples, multimodal MRI scans, Aβ PET scans, and tau PET scans of the Greater-Bay-Area Healthy Aging Brain Study (By Sep 2023)

Inclusion and exclusion criteria

Briefly, the inclusion criteria of GHABS are as follows: (1) adults between the ages of 55 and 90 and speak Mandarin fluently [31]). Notably, individuals below 60 years old are required to have a family dementia history and meet the criteria of subjective cognitive decline (SCD), while people with family history of autosomal dominant or other familial AD are not limited by age; (2) the score on the Geriatric Depression Scale (GDS) is less than 6 points; (3) visual and auditory acuity is sufficient for neuropsychological testing (Including normal corrected vision and hearing); (4) female participants are not pregnant, lactating, or of childbearing potential (that is, women must be two years postmenopausal or surgically sterile); (5) a modified version of the Hachinski Ischemic scores less than or equal to 4; (6) have completed primary school (6 years of education) or have good work experience (sufficient to rule out mental retardation). Individuals with an infection, infarction, or other focal lesions or multiple lacunes or lacunes in critical memory structures and who do not meet the MRI scanning requirements are excluded from the GHABS study. More details of the inclusion and exclusion criteria can be found in the supplemental materials.

Cognitive assessments

The cognitive profiles are assessed via a series of cognitive ability tests, including the Alzheimer’s Disease Assessment Scale–Cognitive Subscale (ADAS-Cog), Logical Memory Test I & II (the Chinese version), Mini-mental State Examination (MMSE, the Chinese version), Montreal Cognitive Assessment Basic (MoCA-Basic), Shape Trail Test (STT), Clock Drawing Test (CDT), Auditory Verbal Learning Test (AVLT), Symbol Digit Modalities Test (SDMT), Digit Span Test (DST), Animal Verbal Fluency Test (AFT), Cognitive Change Index (CCI), SCD. Besides, functional and behavioral tests were also executed, including the Hachinski Ischemic Score (HIS), Clinical Dementia Rating (CDR), Neuropsychiatric Inventory (NPI), the GDS, Functional Activities Questionnaire (FAQ), Activity of Daily Living Scale (ADL) [32,33,34], measurement of everyday cognition (Ecog), Pittsburgh sleep quality index (PSQI), REM sleep behavior disorder screening questionnaire (RBDSQ), Epworth Sleepiness Scale (ESS).

The participants were classified as NC, MCI, or dementia due to AD following the standard protocol of ADNI cohort [35]. We further define some CU individuals as SCD following the research criteria proposed by Jessen and colleagues in 2014 [36]. Several cognitive assessments were used for clinical diagnosis. The normal performance of these assessments was defined based on education or age. The MMSE used cutoff scores: >17 for participants without education, > 20 for 1–6 years of education, and > 24 for more than six years of education. The delayed recall of logical Memory test used cutoff scores as follows: ≥3 for 0–7 years, ≥5 for 8–15 years, and ≥9 for 16 years of education [37]. As for the ADL, < 23 is normal for participants under 75 years old, while < 25 is for 75 years and older.

CU participants were normal in MMSE, logical Memory recall, and ADL, and their CDR score was 0. Among CU individuals, the presence of the following symptoms as having SCD: (1) Self-experienced persistent decline in cognitive capacity in comparison with a previously normal status and unrelated to an acute event; (2) Normal age-adjusted, gender-adjusted and education-adjusted performance on standardized cognitive tests. CU individuals without SCD were defined as the NC group. Participants with MCI had unimpaired MMSE, while they showed impairment in the logical Memory test [38]. Their CDR score was 0.5, with a mandatory requirement of the memory box score being 0.5 or greater, but ADL was normal. Dementia due to AD was abnormal in MMSE, logical memory, and ADL. The CDR score was 0.5 or greater.

Biospecimen collection

Volunteers fasted for one night the day before (not less than 6 h), and blood was drawn in the morning of the next day. The venous blood of the volunteers is drawn into two 10 ml EDTA blood collection tubes and gently inverted and mixed 10–12 times to ensure that the blood and anticoagulant are thoroughly mixed. The mixed blood was placed in an incubator at 4 °C and shipped back to the laboratory within 4 h for subsequent analysis. The blood is centrifuged at 1600 g for 10 min in a refrigerated centrifuge at 4 °C. The upper plasma layer is transferred to several 2 ml centrifuge tubes using a sterile RNase-free pipette tip. To obtain more pure plasma, the separated plasma is centrifuged again at 16,000 g for 10 min at 4 °C, and then the supernatant is aliquoted into several 0.5 ml centrifuge tubes with labels, each with either 100–200 µl blood plasma. The aliquots are stored in a -80 °C refrigerator for subsequent analysis. After the whole blood is centrifuged in the first step, the buffy coat in the middle layer is gently transferred to the 2640 medium. Then, after density gradient centrifugation, erythrocyte lysis, and centrifugation steps, the isolated peripheral blood mononuclear cell sample is transferred to a 2 ml RNase-free centrifuge tubes and stored in a gradient-cooled freezer box at -80 °C for subsequent analysis. Samples will be used for genomic analysis (including whole genome sequencing and other analyses).

Before collecting the CSF sample, the volunteer must fast for one night (at least 6 h), and the lumbar spinal fluid is performed on an empty stomach the following day. Lumbar puncture is performed strictly with clinical standards, and about 5 ml of CSF is collected. The collected CSF is sent to Shenzhen Bay Laboratory within 2 h for biomarker analysis. CSF sample is quickly divided into 1.5 ml low protein adsorption centrifuge tubes. Afterward, they will be stored in a -80 °C refrigerator for subsequent analysis.

Fecal samples are collected on-site or at home after the volunteers’ consent is obtained. The volunteer stool samples are labeled, sub-packaged, and frozen in a -80 °C refrigerator. Fecal samples are used for 16 S rDNA, metagenomic, metagenome, and metabolome detection of intestinal microorganisms.

CSF and plasma biomarkers measurement

The concentrations of Aβ40, Aβ42, NfL, GFAP, SNAP25, p-Tau181, and p-Tau217 in CSF and plasma are detected using commercial Simoa® NEUROLOGY 4-PLEX E (N4PE, cat: 103,670), SNAP-25 (cat: 103,575), pTau-181 (cat: 104,111), and pTau-217 kit in Simoa HD-X Analyzer™ (Quanterix Corp.). The concentrations of YKL40 in CSF and plasma are measured using a commercial Human YKL-40 Assay (cat: K151VLK) in MESO SECTOR S 600MM (Meso Scale Diagnostics, LLC.). The concentrations of sTREM2 in CSF and plasma are also measured by the MSD platform using an in-house immunoassay as previously described [39]. Briefly, streptavidin-coated plates are blocked overnight in PBS containing 3% bovine serum albumin (BSA) and 0.05% Tween-20 before incubating with the biotinylated capture antibody (0.25 µg/mL, cat: BAF1828) for 1 h. After washing with PBS containing 0.05% Tween-20, the plates are incubated with CSF, and plasma samples diluted in PBS containing 0.25% BSA, 0.05% Tween-20, and Protease Inhibitor Cocktail (cat: P8340) for 2 h. Recombinant human TREM2 protein (SinoBiological, cat. 11,084-H08H) is used as a standard (62.5 to 8000 pg/mL). The plates are rewashed and incubated with the detector antibody (1 µg/mL, cat: sc373828) for 1 h, followed by incubation with the MSD SULFO-tag conjugated secondary antibody (0.5 µg/mL, cat: R32AC). Finally, the electrochemical signal is developed by adding 2x MSD Read buffer T (cat: R-92TC), and the MSD SECTOR S 600MM measures the electro chemiluminescent signal.

The concentrations of PDGFR-β and GAP43 in CSF and plasma are measured by commercial Human PDGFR-β (R&D, Catalog Number: DYC385) and GAP43 (Abbexa) ELISA kits. For neurogranin, an in-house sandwich ELISA is developed combining the mouse monoclonal antibodies Ng22 and Ng2 (Abcam). Nunc maxisorp 96-well microliter plates (Thermo Fisher Scientific) are coated with the mouse anti-human neurogranin (Ng22, 3 µg/mL) in bicarbonate buffer pH 9.6 overnight. After washing with PBS containing 0.05% Tween-20, the plates are blocked in PBS containing 1% BSA and 0.05% Tween-20 for 1 h, followed by incubation with CSF and plasma samples overnight. Recombinant full-length human neurogranin protein is used as a standard (25 to 3200 pg/ml for CSF, 55 to 40,000 pg/mL for plasma). After additional washes, the plates are incubated with the biotinylated mouse anti-human neurogranin (Ng2, 2.7 µg/mL) for 1 h, followed by incubation with a Streptavidin-HRP (R&D Systems) for 30 min. After additional washes, the plates are incubated with Substrate Reagent (TMB, R&D Systems) for 10 min in the dark. The color reaction is stopped by adding 0.2 M H2SO4, and the absorbance is read at 450 nm (650 nm as a reference value).

All samples were analyzed randomly and double-blindly to avoid bias because of the effect of inter-assay variability on specific patient groups. For all biomarker measurements, the samples in the first plate were tested in duplicate, and subsequent samples were tested in single. A pooled sample was generated as a reference, aliquoted and stored at -80 °C. In each measurement, the reference sample was tested in duplicate to compare the detection variability. The mean intra-assay and inter-assay coefficient of variation (CV) was controlled within 20%.

MRI image acquisition

All the MRI scanning sequences will be conducted following the standard ADNI protocol here. The MRI image data is collected on 3.0T scanners, and the scanning parameters vary slightly depending on the specifics of scanners from various clinical centers. A series of sequences are applied for head imaging of each volunteer, including3D T1 MPRAGE, 3D T2 FLAIR (Fluid-Attenuated Inversion Recovery), High Res Hippo (High Resolution Hippocampus), SWI (Susceptibility Weighted Imaging), ASL (Arterial spin labeling), Diffusion (Axial DTI), Field Mapping, EPI-BOLD. Each sequence and parameters of MRI scanning are described as follows:

3D MPRAGE scan parameters: field of view (FOV) = 208 × 240 × 256 mm3; matrix = 208 × 240 × 256 (resolution 1 × 1 × 1 mm3); imaging plane: sagittal; TR = 2300 ms; TE = minimum TE value of the system; TI = 900 ms; FA = 7°; 2-fold acceleration in the phase-encoding direction; scan time 6 min 11 s. Purpose: (1) To provide high-resolution anatomical structure templates for low-resolution functional imaging (PET, EPI images, etc.); (2) To calculate the brain tissue volume and cortical thickness of the subjects, and to measure brain atrophy. Note that the whole brain was covered to avoid aliasing artifacts.

3D FLAIR scanning parameters: FOV = 256 × 256 × 260 mm; matrix = 214 × 256 × 160 (resolution 1.2 × 1 × 1 mm); imaging plane: sagittal; TR = 4800 ms; TE = 119 ms; TI = 1650 ms; 3-fold acceleration in the phase-encoding direction; scan time 6 min 48 s. Purpose: To perform white matter hyperintensity (WMH) segmentation and to evaluate white matter lesions, including infarction and other pathological features.

High Res Hippo scanning parameters: FOV = 175 × 60 × 175 mm3; matrix = 449 × 30 × 449 (resolution 0.39 × 2 × 0.39 mm3); Oblique scan; layer direction perpendicular to the long axis of the hippocampus; TR = 8020 ms; TE = 50 ms; 2-fold acceleration in the phase-encoding direction; scan time 5 min 36s. Purpose: To perform hippocampal segmentation. Note: (1) the layer direction is perpendicular to the long axis of the hippocampus; (2) the imaging FOV should cover the skull, the hippocampal head, and the hippocampus tail.

SWI scanning parameters: FOV = 220 × 220 × 130 mm3; matrix = 368 × 368 × 130 (resolution 0.6 × 0.6 × 1 mm3); imaging plane: transverse; TR = 55 ms; 6 echoes, with the TE of 1st echo = 7.7 ms and the delta TE between echoes = 7.0 ms; 2-fold acceleration in the phase-encoding direction; scan time 6 min 10 s. Purpose: (1) To evaluate cerebral microbleeds; (2) To obtain the quantitative susceptibility map (QSM).

ASL scanning parameters: 3D pseudo-continuous ASL (pCASL) series; FOV = 240 × 240 × 160 mm3; matrix = 96 × 96 × 40 (resolution: 2.5 × 2.5 × 4 mm3); transverse positioning; TR = 4250 ms; TE = 9.0 ms; Label duration = 1800 ms; Post label delay (PLD) = 2000 ms; 2-fold acceleration in the phase-encoding direction; axial acquisition direction: fat chemical shift toward the posterior direction; a proton density image was also acquired within the same sequence to quantify cerebral blood flow (CBF) from the ASL series. The scanning time is 6 min 68 s. Purpose: To measure whole brain perfusion and to calculate CBF. Notes: (1) the labeling plane was placed 20 mm inferior to the lower edge of the imaging volume; (2) cover the cerebellum; (3) keep eyes open, not to think of anything in particular, and remain still during the scan.

Diffusion scanning parameters: FOV = 232 × 232 × 176 mm3; matrix = 116 × 116 × 88 (resolution: 2 × 2 × 2 mm3); transverse position; TR = 3300 ms; TE = 71 ms; multi-b value acquisition: b = 0, 1000, 2000 s/mm2 (in total 112 diffusion directions); 2-fold acceleration in the phase-encoding direction and multiband factor = 3; axial acquisition direction: fat chemical shift toward the posterior direction; scan time 11 min 16 s. 8 averages of b = 0 DTI images with fat chemical shift toward the anterior direction of the axial acquisition direction were also scanned for the geometry distortion correction. Purpose: (1) To evaluate the diffusion parameters of white matter fibers in the brain, such as anisotropy FA, AD, MD, and other parameters; (2) To track white matter fibers in the brain and evaluate the structural connection of white matter fibers. Note to cover the whole brain.

Field mapping scanning parameters: FOV = 220 × 220 × 160 mm3; matrix = 88 × 88 × 64 (resolution: 2.5 × 2.5 × 2.5 mm3); imaging plane: transverse; TR = 400 ms; TE1 = 4.92 ms, TE2 = 7.38 ms. Purpose: Used for geometry distortion correction for EPI series.

Echo-planar imaging (EPI) - Blood oxygen level-dependent (BOLD) scanning parameters: FOV = 220 × 220 × 160 mm3; matrix = 88 × 88 × 64 (resolution: 2.5 × 2.5 × 2.5 mm3); imaging plane: transverse; TR = 600 ms; TE = 30 ms; FA = 53° (Ernst angle with best BOLD contrast); Echo spacing = 0.49 ms; 2-fold acceleration in the phase-encoding direction and multiband factor = 4; axial acquisition direction: fat chemical shift toward the posterior direction; scan time 10 min. Purpose: to assess the functional connectivity between any pair of brain regions. Notes that participants were instructed to keep their eyes open, not to think of anything in particular, and to remain still during the scan.

PET image acquisition

The Aβ PET radiotracer [18F]-florbetapir (FBP) [40] or [18F]D3FSP (FSP) [41] and tau PET radiotracer [18F]-flortaucipir (FTP) [42] are used for PET imaging. The data acquisition is performed on either a GE Discovery™ MI Gen 2 PET/CT scanner or a Siemens Biograph™ TruePoint™ TrueV PET/CT scanner. The spatial resolution of each PET scanner is quantified with PET imaging of a Hoffman phantom. For the Aβ PET imaging, the subjects are injected with either [18F]-florbetapir or [18F]-D3FSP intravenously at 370 MBq (10 mCi ± 10%), rested for 45 min and prepared for the scanning. PET/CT imaging is performed 50 min after injection, and the PET acquisition time is 20 min. For the tau PET imaging, the participants are injected with [18F]-flortaucipir intravenously at 370 MBq (10 mCi ± 10%), rested for 75 min, and prepared for imaging. The dynamic acquisition of [18F]-flortaucipir tau PET data is completed 80–100 min after the radiotracer administration.

A dedicated head scanning procedure covering the whole brain from vertex to cerebellum is used for imaging. A diagnostic dose CT scan of the brain is acquired beforehand for attenuation correction and fusion localization of PET images. The PET scans are acquired using 3D list mode on the GE Discovery MI and Siemens BioGraph TruePoint scanners in two sites. For the GE scanner, the FOV is 256 mm×256 mm×220 mm, the scanning matrix is 336 × 336 × 109, and the voxel size is 1.02 mm×1.02 mm×2.03 mm. For the Siemens scanner, FOV is 256 mm×256 mm×198 mm, the scanning matrix is 192 × 192 × 71, and the voxel size is 1.33 mm×1.33 mm×2.79 mm. All the correction options were selected for both scanners, and no filter or smooth was used during the reconstruction. A reconstruction offset is applied to ensure that the head is entirely in the field of view within the plane. Finally, 4 frames of dynamic images are generated according to 5 min/frame segmentation, with each PET scan corresponding to a 20-minute PET image.

MRI and PET imaging analysis

The structural MRI images are segmented into different cortical and subcortical regions of interest (ROI) in Freesurfer (V7.2.0). The residual hippocampal volume (rHCV) is calculated using the hippocampal volume of both hemispheres and adjusted using the estimated total intracranial volume as we described previously [14]. In addition, the cortical thickness of AD-signature atrophy brain regions is obtained by calculating the surface area-weighted average thickness of the bilateral entorhinal, fusiform, inferior temporal, and middle temporal cortices [43].

As shown in Fig. 2, the PET images are preprocessed with the following steps before further analysis: (1) co-registering the 2nd, 3rd, and 4th frames to the 1st frame, respectively; (2) averaging the four frames into one averaged frame; (3) the averaged frame resliced into a standard AC-PC space (anterior commissure-posterior commissure) with image size = 160 × 160 × 128, voxel dimension = 1.5 mm×1.5 mm×1.5 mm; (4) this standardized image then served as a reference imaging, and each original individual frame was resliced to this reference imaging; (5) smoothing to a uniform 6 mm resolution using the Gaussian kernel function, and the full-width at half maximum (FWHM) of each site was determined from the Hoffman phantom [18F]FDG-PET imaging (As a reference for mitigating between-scanner differences in multi-center PET scans). The PET and MRI images are processed using in-house Matlab algorithms, as shown in Fig. 2. The PET images are co-registered with their corresponding structural MRI images in SPM12 (Statistical Parametric Mapping). Sixty-eight Freesurfer-defined cortical ROIs obtained from MRI segmentation extract regional FSP, FBP, and FTP measurements from the co-registered PET images.

Fig. 2

The image processing pipelines of PET, structural MRI, functional MRI (fMRI), and diffusion MRI data of the Greater-Bay-Area Healthy Aging Brain Study

The FSP and FBP standardized uptake value ratio (SUVR) of AD summary cortical regions (posterior cingulate cortex, precuneus, frontal lobe, parietal lobe, and lateral temporal) are obtained by dividing the radiotracer uptake value of AD typical brain regions by that in the whole cerebellum [44]. In the present study, we only presented FSP Aβ PET data. For the [18F]-flortaucipir images, FTP SUVR of 68 FreeSurfer-defined ROIs are calculated by normalizing the [18F]-flortaucipir value to the value of the inferior cerebellar cortex [45]. The FTP SUVR of the AD Temporal-MetaROI [43] (entorhinal cortex, parahippocampal gyrus, fusiform, amygdala, inferior temporal and middle temporal brain regions) is used to evaluate cortical tau deposition.

The general process of resting-state fMRI data preprocessing and brain functional connectivity construction are outlined in the third column of Fig. 2 and described in detail in the Supplemental Materials. The general process of diffusion MRI processing are outlined in the fourth column of Fig. 2 and described in detail in the Supplemental Materials. Diffusion-weighted data are denoised and corrected for Gibbs ring using Mrtrix3 (V3.0.3) [46], and then corrections were applied for head motion, eddy current, and EPI susceptibility distortion using FSL (V6.0.3) [47]. 3D pCASL imaging was employed to calculate the CBF map (Supplemental Fig. 3). The WMH segmentation was processed using a custom pipeline developed by our lab based on the T2 FLAIR images (Supplemental Fig. 4). More details of fMRI, dMRI, WMH, and pCASL can be found in Supplementary Material.

Cutoffs of plasma Aβ42/Aβ40, and plasma p-Tau181, and FSP Aβ PET

We used Gaussian mixed model analysis to estimate two Gaussian distributions of low Aβ and high Aβ (Supplemental Fig. 5) for COMPOSITE FSP SUVR to define an unsupervised threshold, which corresponds to a 90% probability of belonging to the high Aβ distribution. The thresholds of FSP SUVR were defined as COMPOSITE SUVR≥0.76 (Supplemental Fig. 5). The receiver operating characteristic (ROC) curve analysis using the Youden index classifying 154 Aβ- cognitively unimpaired (CU) participants and 69 Aβ + cognitively impaired (CI) individuals (MCI or dementia) as the endpoint to define the cutoff ≤ 0.0609 for plasma Aβ42/Aβ40 ratio (Supplemental Figs. 6–7). Similarly, the ROC analysis using the Youden index classifying 143 Aβ- CU participants and 67 Aβ + CI participants as the endpoint to define the cutoff ≥ 2.418 for plasma p-Tau181 (Supplemental Figs. 8–9). The cohort was divided into A-/T-, A-/T+, A+/T-, and A+/T + according to the plasma Aβ42/Aβ40 ratio and plasma p-Tau181 thresholds.

Statistical analysis

All the statistical analyses were conducted using R (v4.3.0, The R Foundation for Statistical Computing). The normal distribution of the data in this study was determined using the Shapiro-Wilk test. Plasma p-Tau181, plasma GFAP, and plasma NfL were log10 transferred before the following analysis to meet the normal distribution. We used a two-tailed Mann-Whitney U test and Fisher’s exact test to compare the continuous and categorical characteristics at baseline between different diagnosis groups, respectively. Data were presented as median (interquartile range, IQR) or No. (%) unless otherwise noted. Generalized linear models were used to compare plasma biomarkers, COMPOSITE FSP Aβ SUVR, temporal-metaROI FTP SUVR, rHCV, and temporal-metaROI cortical thickness between different clinical groups, controlling for age, sex, and APOE-ε4. Subsequently, we investigated the frequency of different A/T profiles defined by plasma biomarkers (A: plasma Aβ42/Aβ40, T: plasma p-Tau181) and Aβ PET positivity among NC, SCD, MCI, and dementia groups.