Convergent Neuroimaging and Molecular Signatures in Mild Cognitive Impairment and Alzheimer’s Disease: A Data-Driven Meta-Analysis with N = 3,118

The current study aimed to evaluate the susceptibility to regional brain atrophy and its biological mechanism in Alzheimer’s disease (AD). We conducted data-driven meta-analyses to combine 3,118 structural magnetic resonance images from three datasets to obtain robust atrophy patterns. Then we introduced a set of radiogenomic analyses to investigate the biological basis of the atrophy patterns in AD. Our results showed that the hippocampus and amygdala exhibit the most severe atrophy, followed by the temporal, frontal, and occipital lobes in mild cognitive impairment (MCI) and AD. The extent of atrophy in MCI was less severe than that in AD. A series of biological processes related to the glutamate signaling pathway, cellular stress response, and synapse structure and function were investigated through gene set enrichment analysis. Our study contributes to understanding the manifestations of atrophy and a deeper understanding of the pathophysiological processes that contribute to atrophy, providing new insight for further clinical research on AD. Supplementary Information The online version contains supplementary material available at 10.1007/s12264-024-01218-x.

The recruited AD patients fulfilled the following inclusion criteria: (1) diagnosed using the criteria of the National Institute of Neurological and Communicative Disorders and Stroke and the Alzheimer's Disease and Related Disorders Association for probable AD; (2) CDR = 1 or 2; (3)   currently receiving no nootropic drugs, such as cholinesterase inhibitors; and (4) able to perform the neuropsychological tests and tolerate magnetic resonance (MR) scanning.
The exclusion criteria for the participants in this study included the following: (1) metabolic conditions, such as hypothyroidism and vitamin B12/folic acid deficiencies; (2) psychiatric disorders, such as schizophrenia and depression; (3) infarction or brain hemorrhaging, as indicated by MR/computed tomography (CT) imaging; and (4) Parkinsonian syndrome, epilepsy, or other nervous system diseases that can influence cognitive function.In addition, patients with a metallic foreign body, such as a cochlear implant, heart stent, or other relevant MR scanning contraindications, were excluded from the study.

HH_Z
This dataset followed the protocol used in PL_G and PL_S and was approved by the Medical Ethics Committee of Tianjin Huanhu Hospital, Tianjin, China.The patients were recruited from the memory clinic of the neurology department of Tianjin Huanhu Hospital.The control subjects were recruited from the local community using advertisements.Written informed consent was given by each enrolled subject or his/her authorized guardian.The participants underwent general physical, psychological, and laboratory examinations prior to enrollment in the formal study.The participants did not take medications that might have influenced their cognition during the scans, and all patients received professional suggestions for further treatment.

QL_W
This dataset followed the protocol used in PL_G and PL_S and was approved by the Medical Ethics Committee of Qilu Hospital of Shandong University, Jinan, China.The patients were recruited from the memory clinic of the Department of Neurology and Radiology, Qilu Hospital of Shandong University.The control subjects were recruited from the local community using advertisements.
Written informed consent was given by each enrolled subject or his/her authorized guardian.The participants underwent general physical, psychological, and laboratory examinations prior to enrollment in the formal study.The participants did not take medications that might have influenced their cognition during the scans, and all patients received professional suggestions for further treatment.Related publications can be found elsewhere [16] .

XW_H
The study was approved by the Medical Research Ethics Committee and Institutional Review Board of Xuanwu Hospital, Beijing, China (ClinicalTrials.govidentifiers: NCT02353884 and NCT02225964).Some of the data have been used in previous studies, and detailed information can be found elsewhere [17,18] .
All subjects underwent a series of standardized clinical evaluations, including a medical history interview, a neurological examination, and a battery of neuropsychological tests.The neuropsychological tests included the Chinese version of the MMSE, the Beijing version of the Montreal Cognitive Assessment (MoCA) [19] , the CDR Scale [3] , the AVLT [20] , an ADL assessment, the Hachinski Ischemic Scale, the Hamilton Depression Rating (HAMD) Scale [21] , and the Center for Epidemiologic Studies Depression scale [22] .Confirmation of diagnosis for all subjects was made by the consensus of at least two experienced neurologists at the Neurology Department of Xuanwu Hospital.The diagnoses were based on the data available from the neuropsychological assessment evaluation, a battery of general neurological examinations, and subject symptoms as well as functional capacity reports.
The inclusion criteria for an MCI diagnosis included the following [23]  AD subjects were diagnosed according to the National Institute of Aging-Alzheimer's Association (NIA-AA) criteria for clinically probable AD as follows [24,25] : (a) meeting the criteria for dementia;

XW_Z
All participants were recruited by advertisements and supported throughout the testing procedures at a specialized neuropsychological research facility at Xuanwu Hospital in Beijing, China.Patients and informants (usually a family member) were interviewed clinically by a senior psychiatrist (X.Zhang).
Written consent forms were given by all subjects or their legal guardians (usually a family member).
The study was approved by the ethics committee of Xuanwu Hospital.AD subjects were diagnosed using standard operationalized criteria (DSM-IVR [American Psychiatric Association, 1994] and NINCDS-ADRDA [24] ]).
Patients with a diagnosis of AD and a CDR score of 1 were classified as having mild AD, while patients with a CDR score of 2 or 3 were diagnosed with severe AD MCI was diagnosed according to standard criteria [4,26,27] , including subjective memory loss with objective evidence of memory impairment in the context of normal or near-normal performance on other domains of cognitive functioning, minimal impairment of daily living activities, and a CDR score All patients underwent a complete physical and neurological examination, an extensive battery of neuropsychological assessments, and standard laboratory tests.In addition, healthy volunteers underwent a brief clinical interview and the MMSE to confirm that they satisfied the exclusion criteria for cognitive deficits, psychoactive drug use, and clinical disorders.Detailed information can be found in our previous studies [28][29][30][31][32] .

ZJ_L
The subjects were recruited from the Memory Clinic of Zhejiang Provincial People's Hospital.All participants were right-handed and were asked to provide written informed consent, and the study was approved by the local Ethics Committee of Zhejiang Provincial People's Hospital [33] .MRI data were acquired using a G.E. 3-Tesla scanner (MR 750) equipped with an 8-channel head coil (GE, USA) at Zhejiang Provincial People's Hospital.Three-dimensional T1-magnetization fast spoiled gradient echo (FSPGR) sagittal images were collected using the following parameters: slice thickness/gap = 1/0 mm; in-plane resolution = 256 × 256; TR = 6.6 ms; TE = 2.9 ms; inversion time = 450 ms; flip angle = 90°; FOV = 256 mm × 256 mm; and voxel size = 1 mm × 1 mm × 1 mm, 192 sagittal slices (Note: Among all the subjects, one had 190 slices, one had 188 slices, one had 184 slices, one had 182 slices, one had 168 slices, one had 164 slices, and two had 160 slices).

Quality of the Included Images
First, we performed a visual check (Y.L, X.K) of all the T1-weighted brain images, and then all the images were processed using the same pipeline.All the MRIs were preprocessed with the standard steps of the CAT12 segmentation process to extract grey matter images and surface data [35] .All the sMRI data were bias-corrected, segmented, and registered to MNI space.The Cat12 toolkit provided a quality score for each subject according to image resolution, noise, bias, and IQR (http://www.neuro.uni-jena.de/cat12-html/cat_methods_QA.html).We removed any subject's data whose image quality was <0.6 (a total of 50 subjects were excluded, leaving 3118 subjects for analysis, Fig. S1) after we considered the reliability of the analysis results [36] .Overall image quality statistics reported by the CAT12 segmentation process of the 3118 used subjects are provided in Fig. S2.

Included ADNI PET Images
For

Part 2. Supplementary Meta-analysis
In addition to the primary meta-analysis we described in the manuscript, we also performed a series of tests to verify the robustness and reproducibility of our results.
(2) Analysis using original features We applied analyses using the original data (without controlling for covariates, Fig. S5).
(3) Analysis with Age 2 removed We applied the analyses using ROI features that controlled for age, age 2 , gender, and TIV (Fig. S6).
(4) Meta-analysis for each dataset We conducted meta-analyses that only included subjects in one dataset (Fig. S7-9).
(5) Testing using different brain atlases (AAL atlas, Schaefer 1000 atlas) We also tested the ROI results using the AAL (116 regions, Fig. S10) and Schaefer 1000 (1000 regions, Fig. S11) atlases.The mean cortical thickness in these atlases was calculated by resampling both the atlas and the individual thickness onto an HCP 32K surface using Cat12.
(6) Random sample analysis We applied the bootstrapping strategy 5000 times (Fig. S12A).Each time we randomly sampled 80% of the subjects from each cohort.Next, we calculated the effect size and cohort weights using inverse variance with random effect models.Then, we combined all the cohorts' effect sizes to calculate the total effect size.Finally, we compared the effect size of 80% of the subjects with the effect size of all the subjects.In the process of SIMPLS analysis, we conducted 5000 permutation tests to calculate the P-value to improve the reliability of our results (https://github.com/rmarkello/pyls#regression-with-simpls).
We also took 5000 randomly sampled meta-analysis results to perform a PLS analysis and evaluated the correlation between its first component weight and primary PLS analysis (Fig. S12B).PLS was also applied to the single dataset meta results, and their correlation with the primary PLS analysis was r = 0.98 for MCAD, r = 0.99 for EDSD, and r = 0.96 for ADNI.

Part 3. t-test Based on Harmonized Data
To further validate and strengthen our results, we used an alternative statistical analysis method on the gray matter data.We first used neuroCombat [37] to harmonize the GMV and CT to reduce the impact of site effects.When performing harmonization, we use the label, age, gender, TIV, or average CT as covariates.Then we applied a t-test to evaluate the atrophy based on harmonized data.The results are shown in Fig. S13.

Part 5. Tissue and Cell-Type Enrichment Analysis
We investigated whether the high-ranking genes associated with AD atrophy patterns are enriched in specific tissues and cell types (Fig. S17).

Part 6. Relationship Between Gray Matter Features and PET Features
We collected subjects with both gray matter images and PET images from the ADNI dataset.We used the Brainnetome atlas to extract the gray matter and PET features of brain regions.And then we applied correlation analysis to each brain region in these subjects, and obtained a correlation value for each region (Fig. S18).
: (a) memory complaints confirmed by an informant; (b) objectively impaired memory confirmed by neuropsychological tests; (c) a definitive history of cognitive decline; (d) not meeting the criteria for dementia according to the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition, Revised (DSM-IV-R); and (e) a CDR score of 0.5.
(b) insidious and gradual onset (not sudden) over >6 months; (c) definitive history of declining cognition; (d) initial and most prominent cognitive deficits evident in amnestic or nonamnestic performance; and (e) hippocampal atrophy confirmed by structural magnetic resonance imaging (MRI).The NC subjects were required to meet the following research criteria: (a) no memory concerns; (b) MMSE and MoCA scores within the normal ranges (adjusted for age, sex, and education); and (c) a CDR score of 0. The exclusion criteria applied to all subjects included the following: (a) vascular cognitive impairment (Hachinski Ischemic Scale score >4 points) or a history of stroke; (b) severe depression (HAMD score >24 points or Center for Epidemiological Studies Depression Scale score >21 points); (c) other central nervous system diseases that could cause cognitive decline (e.g., epilepsy, brain tumors, Parkinson's disease, or encephalitis); (d) systemic diseases that could cause cognitive impairments (e.g., anthracemia, syphilis, thyroid dysfunctions, severe anemia, or HIV); (e) history of psychosis or congenital mental growth retardation; (f) severe hypopsia or dysacusis; (g) cognitive decline caused by traumatic brain injury; (h) severe end-stage disease or severe diseases in acute stages; or (i) unable to complete neuropsychological tests or contraindication with MRI.MR images were acquired on a 3.0 T MR scanner (Magnetom Trio, Siemens, Germany).T1weighted MR images were acquired using an MP-RAGE sequence (TR = 1900 ms; TE = 2.2 ms; FA = 9°; FOV = 256 mm × 224 mm; matrix = 512 × 448; inversion time = 900 ms; slice thickness = 1 mm, no gap; 176 slices).
of 0.5.Normal volunteers had a CDR score of 0. All participants satisfied the following inclusion criteria: (1) no history of an affective disorder within one month prior to assessment; (2) normal vision and audition; (3) able to cooperate with cognitive testing; (4) aged between 50 and 90 years; (5) no clinical history of stroke or other severe cerebrovascular diseases; and (6) no more than one lacunar infarction, without patchy or diffuse leukoaraiosis, on neuroradiological assessment of conventional MR images.The exclusion criteria included the following: (1) severe general medical disorders of the cardiovascular, endocrine, renal, or hepatic systems; neurological disorders associated with potential cognitive dysfunction, including local brain lesions, traumatic brain injury with loss of consciousness or confusion, and dementia associated with neurosyphilis, Parkinsonism, or Lewy body disease; psychiatric disorders, including depression, alcohol, or drug abuse; (2) concomitant use of psychotropic medication; or (3) insufficient cognitive capacity to understand and cooperate with the study procedures.

Fig. S2 B
Fig. S2 The CAT12 segmentation process reports overall image quality statistics.A Resolution rating.B Bias rating.C Noise rating.D IQR rating.

Fig. S3
Fig. S3 Subject counts and MMSE statistics for the Aβ and FDG PET images.A Number of subjects with Aβ for each group.B MMSE statistics for the subjects with Aβ.C Number of subjects with FDG for each group.D MMSE statistics for the subjects with FDG.

Fig. S4
Fig. S4 Comparison of main ROI-wise results with voxel/vertex results.A Main ROI GMV results.B Voxel-wise meta-analysis results.C Main ROI CT results.D Vertex-wise meta-analysis results.

Fig. S5
Fig. S5 Comparison of main results with results without controlling covariates.A Main ROI GMV results.B ROI GMV meta-analyses results based on original features.C Main ROI CT results.D ROI CT meta-analyses results based on original features.

Fig. S6
Fig. S6 Comparison of main results with results controlling age 2 .A Main ROI GMV results.B ROI GMV meta-analyses result-based features additionally controlled for age 2 .C Main ROI CT results.D ROI CT meta-analyses result-based features additionally controlled for age 2 .

Fig. S7
Fig. S7 Comparison of main results with results based on the MCAD dataset.A Main ROI GMV results.B ROI GMV meta-analyses results based on the MCAD dataset.C Main ROI CT results.D ROI CT meta-analyses results based on the MCAD dataset.

Fig. S8
Fig. S8 Comparison of the main results with results based on the ADNI dataset.A Main ROI GMV results.B ROI GMV meta-analyses results based on the ADNI dataset.C Main ROI CT results.D ROI CT meta-analyses results based on the ADNI dataset.

Fig. S9
Fig. S9 Comparison of the main results with results based on the EDSD dataset.A Main ROI GMV results.B ROI GMV meta-analyses results based on the EDSD dataset.C Main ROI CT results.D ROI CT meta-analyses results based on the EDSD dataset.

Fig. S10
Fig. S10 Comparison of the main results with results based on the AAL atlas.A Main ROI GMV results.B ROI GMV meta-analyses results based on the AAL atlas.C Main ROI CT results.D ROI CT meta-analyses results based on the AAL atlas.

Fig. S11
Fig. S11 Comparison of the main results with results based on the Schaefer atlas.A Main ROI GMV results.B ROI GMV meta-analyses results based on the Schaefer atlas.C Main ROI CT results.D ROI CT meta-analyses results based on the Schaefer atlas.

Fig. S12
Fig. S12 Robustness of the meta-analysis and gene PLSR analysis.A Random sampling of 80% of the subjects 5000 times was conducted for the meta-analysis, and the r-value of the meta-analysis is the correlation coefficient between the meta Cohen's d and the main result.B The gene PLSR analysis was based on 5000 meta-analysis results, and the r-value of the PLS1 weight is the correlation coefficient between the PLS1 weight with 80% of samples and the main PLS1 weight with the entire dataset.

Fig. S13
Fig. S13 Comparison of main meta-analysis results with t-test results based on harmonized data.A Main ROI GMV results.B The t-test results based harmonized ROI GMV.C Main ROI CT results.D The t-test results based harmonized ROI CT.

Fig. S14
Fig. S14 Analysis of surrogate maps.A Distribution of r-values between the surrogate map and the AD vs NC ROI GMV effect sizes.B Distribution of PLS1 variance explained by PLSR analysis between surrogate map and gene expression, red line indicated the PLS1 variance explained by PLSR analysis between AD vs NC ROI GMV effect sizes and gene expression.

Fig. S15 A
Fig. S15 A joining of GMV-based GSEA results and CT-based GSEA results.

Fig. S16 A
Fig. S16 A full directed acyclic graph of significant GO terms (enriched in both GMV-based GSEA and CT-based GSEA).

Fig. S17
Fig. S17 Tissue and cell-type enrichment analysis results based on the 1% of the genes with the greatest weights in the PLS result.A Significant results of FUMA GENE2FUNC (PFDR <0.001).B Significant results of CSEA (PFDR <0.05).
1003 subjects with Aβ PET images from the ADNI dataset, there are 291 AD subjects (118 female