Single-cell transcriptomic atlas of Alzheimer's disease middle temporal gyrus reveals region, cell type, and sex specificity of gene expression with novel genetic risk for MERTK in female Authors: Le Zhang, Tianyu Liu, Chuan H. He, Anthony Russo, Sarah Coffey, Dominic Yin, I-Uen Hsu, Chi Zhang, Yixuan Ye, Chang Su, Joshua Spurrier, Lashae Nicholson, Brian W. Kunkle, Eden R. Martin, Omar G. Rodriguez, Lissette Gomez, Rui Chang, Carla V. Rothlin, Sourav Ghosh, Pallavi P. Gopal, David A. Hafler, Hongyu Zhao, Stephen M. Strittmatter Published: 2026 ABSTRACT BackgroundAlzheimer's disease (AD), the most common age-related neurodegenerative disease, is closely associated with both amyloid-β plaque and neuroinflammation. Two thirds of AD patients are female, and they have a higher disease risk; women with AD have more extensive brain histological changes than men along with more severe cognitive symptoms and neurodegeneration.ObjectiveThis study aimed to determine how sex difference induces structural brain changes and molecular cell vulnerabilities in AD, with a focus on identifying sex-specific transcriptional alterations and genetic risk factors.MethodsWe performed single nucleus RNA sequencing on postmortem brains from individuals with AD and age- and sex-matched controls, focusing on the middle temporal gyrus, a cortical brain region strongly affected by the disease, and integrated single nucleus RNA sequencing results with genome-wide association study (GWAS) data using cell type-specific enrichment and generalized gene-set analysis approaches. The analysis pipeline is provided with threshold information.ResultsWe identified a selectively vulnerable subpopulation of layer 2/3 excitatory neurons that were RORB-negative and CDH9-expressing in both males and females. Disease-associated, but sex-independent, reactive astrocyte signatures were also present. In clear contrast, the microglia signatures of AD brains differed between males and females. Integrating single cell transcriptomic data with results from GWAS, we identified MERTK genetic variation as a candidate novel risk factor for AD selectively in females.ConclusionsTaken together, our single cell atlas of middle temporal gyrus revealed a unique cellular-level view of sex-specific transcriptional changes in AD, illuminating GWAS identification of sex-specific AD genes. These data serve as a rich resource for interrogation of the molecular and cellular basis of AD. Introduction Alzheimer's disease (AD), the most common age-related neurodegenerative disease, is characterized by progressive cognitive decline and dementia. It remains a major unmet medical need with limited options for disease-modifying therapy in clinical practice today.1,2 Two thirds of AD patients are females, and women have a higher risk of developing AD.3,4 Women with AD are reported to have more extensive brain histological changes, more severe cognitive symptoms, and more severe neurodegeneration than men with AD.5–9 However, some studies suggest more rapid progression to death in men, 10 emphasizing our lack of knowledge regarding the differential effect of AD on female and male brains. Thus, research on sex differences in AD is essential to the development of effective interventions and sex-focused therapies. Although sex differences in the risk of AD, vulnerability to genetic load and severity of AD pathology burden have been established,5,11,12 the underlying mechanisms and molecular pathways that are differentially mediated in female and male AD brains remain poorly understood. Previous studies have explored certain aspects of genetic variation on sex differences in AD. Apolipoprotein E (APOE) has been shown to have a differential impact on age-at-onset between males and females.13,14 Women who have an APOE4 allele are more likely to develop AD than are male carriers. This begs the question of whether other known or novel AD risk factors have sex-differential roles in this genetically complex disease. Recent studies suggest that sex differences in genetic risk go well beyond APOE4.11,15–17 Using sex-based Alzheimer's Disease Genetic Consortium (ADGC) genome-wide association study (GWAS) data and calculation of polygenic hazard scores and polygenic risk scores, it is predicted that effect sizes for BIN1, MS4A6A, DNAJA2, and FERMT2 were higher in women, while FAM193B, C2orf47, and TYW5 variants appeared stronger in men. 17 Using Sex-Specific Family-Based Association Analysis of Whole-Genome Sequence Data, novel loci (GRID1, RIOK3, MCPH1, ZBTB7C) were identified showing sex-specific association with AD risk. 16 It is now clear that inflammation in the central nervous system is an age-dependent process that is associated with impaired memory, cognition, and other brain functions, and is a crucial mediator of neurodegenerative diseases.18–23 The important role of neuroinflammation in AD is supported by findings that genes for immune receptors, including TREM2, CR1 and CD33 are associated with AD pathology.24,25 Misfolded and aggregated proteins bind to pattern recognition receptors on microglia and astroglia, triggering an innate immune response characterized by the release of inflammatory mediators that contribute to disease progression and severity.23,26 Although the importance of myeloid populations in the central nervous system in both health and disease has been increasingly recognized,27–31 the contribution of brain myeloid cells, microglia in particular,29,32 to AD-associated neuronal injury remains incomplete. Single cell genomic technologies enable unbiased characterization of cell types and states, transitions from normal to disease, and response to therapies.33–35 Characterizing the neuroinflammatory networks will be crucial for guiding the development of therapeutic avenues targeting inflammation in AD. Recent research on microglia development identified a microglia-specific gene expression program in mice that was used to create a microglia developmental index. When compared to microglial expression profiles of healthy and AD brain human datasets, differences between males and females 36 were revealed, suggesting that male microglia are more developmentally mature than female microglia in adult brain. Thus, it is of interest to further explore sex-specific manners of microglia in neuroinflammation and AD. To define the impact of sex differences on AD risk and neuroinflammation, as well as obtain an accurate and unbiased assessment of sex-specific and cell-type-specific changes associated with AD, we transcriptionally profiled AD brains and age- and sex-matched controls at the single cell level, focusing on the middle temporal gyrus (MTG), a brain region strongly affected by AD. We thoroughly characterized our single cell dataset of AD MTG, revealed reactive astrocyte signatures, and identified a subpopulation of excitatory neurons that are selectively vulnerable in AD in this brain region. Moreover, this single cell atlas of AD brain MTG revealed sex-specific changes in microglia and AD-associated female-specific gene sets. Combining single cell and GWAS data, we analyzed these gene sets and identified MERTK as a candidate novel genetic risk factor specific to female AD brains. Methods Human brain tissue samples The autopsy brain specimens were collected and dissected at Alzheimer's Disease Research Center (ADRC) at Yale University in New Haven, CT. The samples are fresh frozen brain tissues with postmortem intervals of less than 24 h in most of the cases (average 12 h). Exclusion criteria for tissue specimens included prolonged agonal state, and other confounding medical conditions. Each sample includes a de-identified full autopsy report of neuropathology evaluation and neuro/AD diagnosis based on recommendations of the National Institute on Aging-Alzheimer Association (NIA-AA), 37 including stains with antibodies to α-synuclein, Aβ, and tau as well as CERAD and Braak staging. A detailed description of the age and sex, race/ethnicity and education of the donors, histopathology and clinical diagnosis is listed in Supplemental Table 1. Human brain nuclei isolation Nuclei were isolated from postmortem fresh frozen human brain tissues of MTG. Frozen tissue (50 to 100 mg) was homogenized in 15 ml of ice-cold nuclei homogenization buffer [2 M sucrose, 10 mM Hepes (pH 7.9), 25 mM KCl, 1 mM EDTA (pH 8.0), 10% glycerol, and ribonuclease (RNase) inhibitors freshly added (80 U/ml)] with Dounce tissue grinder (10 strokes with loose pestle and 10 strokes with tight pestle). The homogenate was transferred into an ultracentrifuge tube on top of 10 ml of fresh nuclei homogenization buffer and centrifuged at 25,000 rpm for 60 min at 4°C on ultracentrifuge. The supernatant was removed, and the pellet was resuspended in 1 ml of nuclei resuspension buffer [15 mM Hepes (pH 7.4), 15 mM NaCl, 60 mM KCl, 2 mM MgCl2, 3 mM CaCl2, and RNase inhibitors freshly added (80 U/ml)] and counted on a hemocytometer. The nuclei were centrifuged at 800×g for 10 min at 4°C and resuspended at a final concentration of 700 to 1200 nuclei/µl for 10x Genomics Chromium loading. 10x single nucleus library construction and sequencing Single nucleus libraries were prepared by the Chromium Single Cell 3′ Reagent Kit v3 chemistry according to the manufacturer's instructions (10x Genomics) and sequenced using Illumina NovaSeq6000 S4 sequencer, at a sequencing depth of 300 million reads per sample, targeting an average sequencing depth of 50,000 reads per nucleus. The raw sequencing data of FASTQ files were aligned to human genome reference GRCh38 (GENCODE v32/Ensembl 98) to map both unspliced pre-mRNA and mature mRNA using Cell Ranger version 6.0.1 (10x Genomics). Single cell analysis with quality control, clustering, cell type annotation, and differential expression After quality control filtering of eliminating nuclei with less than 200 genes (low quality nuclei) and more than 7000 genes (potential doublets) per nucleus, we profiled 104,082 brain nuclei for downstream analysis. Across these brain nuclei, we detected a median of 2153 genes and 4460 transcripts per nucleus (Supplemental Table 1). Seurat (version 4.0.2) 38 single cell analysis R package was used for processing the single nucleus RNA sequencing data. Briefly, the top 2000 most variable genes across all nuclei in each sample were identified, followed by the integration of all samples. Dimensionality reduction were performed using principal components analysis (PCA, 20 latent dimensions), and Uniform Manifold Approximation and Projection for Dimension Reduction (UMAP, 2 latent dimensions) was applied to visualize all cell clusters. Cell type classification and annotation were based on canonical and known marker genes for major brain cell types and the entire single nucleus gene expression matrix (Supplemental Figure 1A). Cell types were annotated with known prefrontal cortex marker genes of different clusters, with a Seurat clustering resolution of 0.8. Differential expression analysis for single nucleus data was performed using the Wilcoxon Rank Sum test with the function FindMarkers of the Seurat package in R, with thresholds of log 2-fold change more than 0.25 and p value adjusted less than 0.01. Ingenuity Pathway Analysis (IPA, QIAGEN) was used to interpret differential gene expression and identify enriched pathways and upstream regulators. Disease lineage and trajectory inference for single cell transcriptomics were performed using Slingshot, 39 and trajectory-based differential expression analysis for single cell dataset was performed using TradeSeq. 40 All analyses and visualizations, including ggplot2, were conducted using default parameters unless otherwise specified. ADGC dataset The Alzheimer′s Disease Genetic Consortium (ADGC) datasets comprise case-control, prospective, and family-based studies focusing on individuals who develop AD after age 60. These datasets were derived from Alzheimer's Disease Centers across the USA, while healthy controls were enrolled from the general community between 1984 and 2012. In our study, we were granted access to the sex-specific (male-specific and female-specific) and the combined GWAS data from individuals of European ancestry within the ADGC cohorts. Our analysis involved a total of 29,056 participants, with 13,851 cases (5988 males and 7863 females), which were drawn from 28 cohorts, including ACT1, ACT2, ACT3, ADC1, ADC2, ADC3, ADC4, ADC5, ADC6, ADC7, ADC8, ADC9, ADC10, ADC11, ADC12, ADNI, GSK, MAYO, OHSU, ROSMAP1, TARCC1, TGEN2, UKS, UMVUMSSM, UMVUTARC2, UPITT, WASHU1, and WHICAP. GWAS analysis MAGMA 41 was used for both gene and gene-set analysis, where the gene analysis is based on a multiple regression model and has been shown to provide robust statistical performance. UTMOST 42 was used for transcriptome-wide association analysis (TWAS), which provides both single-tissue results and cross-tissue results. There are three steps in UTMOST: first, gene expression imputation models were trained using GTEx v8 data, where expression levels were normalized and adjusted for possible confounding factors as described in the eQTL analysis in GTEx v8 project; next, the trained cross-tissue gene expression imputation weights were combined with the GWAS summary statistics to identify disease-associated genes in each single tissue; and finally, association test statistics from tissues were combined using the Generalized Berk-Jones test for each gene. Significant disease-associated genes across all tissues and in each single tissue were identified after Bonferroni multiple-testing correction. During our analyses, we mainly used the single tissue results from UTMOST to prioritize the tissues for AD association. Phospho-Tau immunohistochemistry and neurofibrillary tangle burden Histopathology and immunohistochemistry for phospho-Tau Ser202, Thr205 were performed by the Yale Pathology Tissue Services Histology Core on sections of MTG and middle frontal gyrus (MFG) obtained from the same cohort of brains used for single nucleus RNA sequencing. Hematoxylin and eosin (H&E) staining and immunohistochemical staining for phospho-S202, T205 Tau (clone AT8, 1:2000; Thermo Scientific MN1020) were performed according to manufacturer's protocol and product datasheet, and with appropriate controls. Quantitative scoring of neurofibrillary tangles (NFT) pathology was performed in a blinded manner. The number of NFTs was counted in 3 × 10 consecutive high-power fields (400x magnification; area of HPF 0.24 mm2) in sections of MFG and MTG. The average (mean) number of NFTs per 10 HPFs and SEM were calculated for each subject and in each region. A paired t-test was to compare NFT density in MTG versus MFG. RNA in situ hybridization, imaging, and quantification To validate gene expression differences identified by single nucleus RNA sequencing, RNA in situ hybridization was performed using the RNAscope Multiplex Fluorescent v2 Assay (Advanced Cell Diagnostics, Inc.). Human brain tissue from the MTG was flash-frozen at an average postmortem interval of 12 h and stored at −80°C. Prior to performing the assay, frozen tissue was embedded in OCT and sectioned at 10 μm using a cryostat Leica CM3050S. RNAscope in situ hybridization was performed according to the manufacturer's protocol. Briefly, fresh frozen sections were fixed in 10% NBF for 1 h at room temperature, then sequentially dehydrated in 50% EtOH, 70% EtOH, 100% EtOH, and 100% EtOH for 5 min each at room temperature. Tissue sections were pretreated with hydrogen peroxide and protease to block endogenous peroxidase activity and optimally permeabilize the sections. Target probes, including AQP4, TREM2, CHRM3 and HS3ST4, IBA1(AIF1), and MERTK were hybridized to tissue sections, signals were amplified, and fluorescent dyes were applied. Positive control probes for human POLR2S, PPIB, UBC, and HPRT-1 and a negative control probe for Bacillus subtilis DapB were used to ensure detection of target RNAs and a lack of non-specific probe binding. All tissue sections were counterstained with DAPI. TrueBlack lipofuscin autofluorescence quencher was used to remove autofluorescent signals. Images were captured using a Leica SP8 motorized staging confocal microscope with a 20x lens (Leica #506517), while confocal imaging settings were kept constant between healthy controls and AD. RNA expression was quantified using QuPath, by detecting the cells using DAPI channel, identifying and measuring probe channel signals, and calculating estimated spot count per cell for the entire cryosection. RNA expression in human brain sections from AD patients was compared to that in healthy control human brains in approximately 10,000 nuclei from each tissue section, and statistical significance was determined using Student's t-test. Immunohistochemistry, imaging, and quantification For each tissue sample from participants with AD and healthy controls, a 5-µm thick, formalin-fixed and paraffin-embedded section underwent immunofluorescence against RORB and CDH9 as described below. Sections were deparaffinized in two 15-min immersions in xylenes, then sequentially dehydrated in 100% EtOH, 100% EtOH, 90% EtOH, 80% EtOH, and 70% EtOH for 5 min at room temperature. Next, sections were incubated in 3% hydrogen peroxide in methanol for 30 min to inactivate endogenous peroxidase. Antigen retrieval was performed in 1x Tris-EDTA buffer (ab93684, Abcam, pH 9.0) in a steam bath for 20 min. Sections were blocked overnight with 5% milk/PBS-T to reduce nonspecific staining. Tissue sections were incubated overnight in primary antibodies against RORB (1:100 dilution; mouse, NBP2-45610, Novus) and CDH9 (1:100 dilution; rabbit, PA5-106548, Thermo Fisher). After washing with PBS, sections were developed with Alexa Fluor 488 conjugated anti-mouse secondary (1:150 dilution: A-21121, Thermo Fisher) and CDH9 was developed with Alexa Fluor 546 conjugated anti-rabbit secondary (1:150 dilution; A-11010, Thermo Fisher) for 1 h. TrueBlack lipofuscin autofluorescence quencher (23007, Biotium) was used to remove autofluorescent signals in human brain sections. Sections were counterstained using DAPI mounting medium (0100-20, SouthernBiotech). Images were captured using a Leica SP8 motorized staging confocal microscope with a 20x lens (Leica #506517) of the entire section, while confocal imaging settings were kept constant between healthy controls and AD. Protein expression was quantified using QuPath, by detecting the cells using DAPI channel, identifying and measuring probe channel signals, and calculating positive or negative cells in two 2700 × 2700-µm ROIs overlaying cortical layers. Protein expression in human brain sections from AD patients was compared to that in healthy control human brains in approximately 10,000 cellular nuclei from each tissue section, and statistical significance was determined using Welch's t test. Results Single nucleus transcriptomic profiling of human AD brain middle temporal gyrus To understand the cellular heterogeneity and disease associated cellular changes in the human AD brain, we performed unbiased single nucleus RNA sequencing on postmortem frozen human brain tissues of MTG, a cortical brain cortical region strongly affected by AD (Supplemental Figure 1). We isolated brain nuclei using sucrose gradient ultracentrifugation (Methods), generated single nucleus libraries with 10x Chromium platform (10x Genomics), and sequenced on NovaSeq6000 S4 sequencer (Illumina), from 18 age- and sex-matched individuals with and without AD (Figure 1A). We integrated our single cell data of human brain MTG from 9 individuals with AD (Braak Stage V/VI) and 9 healthy controls (Braak Stage 0-III) by single cell analysis using Seurat 38 (Supplemental Table 1). After quality control filtering, we profiled and analyzed 104,082 high-quality single nucleus transcriptomes, with 2153 median genes and 4460 median transcripts per nucleus (Supplemental Table 1). We clustered all the cells jointly across the 18 donors that include 8 females and 10 males, and identified and annotated the major cell types of the human brain by interrogating the expression patterns of known gene markers, including neurons (GRIN1), excitatory neurons (ExN, SLC17A7), inhibitory neurons (InN, GAD1), astrocytes (Astro, AQP4), microglia (MG, CSF1R), oligodendrocytes (Oligo, MBP), oligodendrocyte precursor cells (OPC, PDGFRA), and endothelial cells (Endo, CLDN5) (Figure 1B and 1C, Supplemental Figure 2). We next identified the top marker genes of each major brain cell type, including ST18, PLP1, and MBP for oligodendrocytes, SLC17A7, LDB2, and CLSTN2 for excitatory neurons, GAD1, GRIP1, and SYNPR for inhibitory neurons, GFAP, RYR3, and AQP4 for astrocytes, SPP1, RUNX1, and DOCK8 for microglia, and PDGFRA, VCAN, and XYLT1 for OPC, which show unique and enriched expression in each cell type (Figure 1D). Single nucleus transcriptomic analysis of human brain middle temporal gyrus revealed cell-type-specific changes in AD. (A) Schematic workflow of experimental design with the information of sample number and brain region. (B-C) UMAP plotting of human brain nuclei (n = 104,082 nuclei), colored by (B) major brain cell types, including excitatory (ExN) and inhibitory (InN) neurons, oligodendrocytes (Oligo), astrocytes (Astro), microglia (MG), oligodendrocyte precursor cells (OPC), Endothelial cells (Endo) and undefined cells (Undef), and (C) disease diagnosis of either Alzheimer's disease (AD) or healthy controls (HC). (D) Heatmap of top three marker genes in each brain cell type. (E) Differentially expressed genes (DEGs) counts of either upregulated (Up) or downregulated (Down) genes between AD brains and controls for each brain cell type (log2 fold change > 0.25, FDR corrected p-value adjusted < 0.01). (F) Ingenuity Pathway Analysis (IPA) of differentially expressed genes between AD and controls for neuronal (ExN and InN) and glial (Astro and MG) cell types (Top 10 pathways of each cell type were shown; red ink: shared pathways between neuronal and glial cell types; purple ink: unique pathway in microglia). (Color figure available online). We performed differential gene expression analysis for each brain cell type between AD and healthy controls and identified 3015 differentially expressed genes (DEGs) across all brain cell types that are either upregulated or downregulated in AD brains (Figure 1E, Supplemental Table 2, log2 fold change > 0.25, p value adjusted < 0.01). We validated our single nucleus transcriptomic dataset using targeted gene expression method of RNAscope in situ hybridization on an additional cohort of AD and control brain sections of MTG, focusing on DEGs that were identified by single nucleus RNA sequencing, including AQP4, TREM2, CHRM3, and HS3ST4 (Supplemental Figure 3A). We quantified RNAscope signals at the single RNA molecule level between AD and control brains using QuPath (Supplemental Figure 3B, C), which confirmed differential expression of these selected genes in AD. We further performed functional analysis for DEGs in each major neuronal and glial cell type, such as excitatory neurons, inhibitory neurons, astrocytes and microglia to identify changes in canonical signaling pathways in AD brains compared to healthy controls using IPA (Figure 1F, Supplemental Table 3). Interestingly, several signaling pathways were commonly affected by AD across neuronal and glial cell types, such as synaptogenesis signaling and synaptic long term depression pathways, consistent with the key feature of synapse loss and dysfunction in AD. Neuroinflammation pathways were altered in both neurons and glia, highlighting the central role of neuroinflammation in AD pathogenesis. In addition, a number of unique signaling pathways were changed in a certain brain cell type in AD, for example, FcγR-mediated phagocytosis was elevated in AD microglia, which may function as a protective role in the neurodegenerative disease.43,44 Selective vulnerability of excitatory neuron subpopulations in AD middle temporal gyrus Next, we focused on the neuronal changes in AD brain MTG and unbiasedly re-clustered all neurons into neuronal subpopulations, identifying seven subclusters of excitatory neurons (ExN1, ExN2, …, ExN7) and six subclusters of inhibitory neurons (InN1, InN2, …, InN6) (Figure 2A). We revealed a reduction specifically in excitatory neurons in AD brains, while inhibitory neuron subpopulations showed similar proportions between AD and control brains, suggesting that selectively vulnerable neurons in AD in this brain region of MTG are excitatory neurons (Figure 2B, 2C). Within excitatory neurons, the most significant decline of cell composition and relative abundance in AD was in the subpopulations of ExN1 and ExN6, which showed severe neuronal loss in AD brains compared to healthy controls (Figure 2C). This pattern was consistent in both male and female brain samples (Supplemental Figure 4). We further characterized neuronal subpopulations by assessing the top marker genes of each excitatory and inhibitory neuronal cluster, as shown in the heatmap plot (Figure 2D, Supplemental Table 4). Single cell analysis of neurons revealed selective vulnerability of excitatory neuron subpopulations in human AD brain middle temporal gyrus. (A-B) UMAP plotting of human brain neuronal nuclei (n = 43,197 neurons from 18 individuals with and without AD), colored by neuronal cell subtypes (A), such as excitatory neurons (ExN) and inhibitory neurons (InN), or disease diagnosis of either AD or controls (HC) (B). (C) Neuronal cell type composition between AD brains or healthy controls. (D) Heatmap of top three marker genes in each excitatory or inhibitory neuronal cell subpopulation in human brain middle temporal gyrus. (E) UMAP plotting of human brain excitatory neurons, colored by disease diagnosis of either AD or controls. (F) UMAP plotting of human brain excitatory neuronal subpopulations ExN1, ExN2 and ExN3, with colors from dark blue to light yellow showing the disease lineage along the simulation disease trajectory from HC to AD. (G) Differential expression of significantly altered genes along the disease trajectory (left panels) and their expression levels along the trajectory (right panels). (H) Violin plotting of the expression of RORB and CDH9 in the subpopulations of excitatory neurons in the middle temporal gyrus. (I) The alignment of highly vulnerable excitatory neuronal subpopulation ExN1 using its top marker genes to Allen Brain single cell atlas of middle temporal gyrus, grouped by brain cortical layers and clusters. (J) Immunofluorescence of the middle temporal gyrus with representative RORB and CDH9 staining. (DAPI: blue, RORB: green, CDH9: red). (K) Proportion of CDH9 + RORB- cells among all cells (DAPI+) in either healthy control (HC) brains or AD brains (n = 12 sections in each group, p = 0.024 by unpaired Welch's test). (Color figure available online). Based on transcriptional profiles, three major excitatory neuronal subpopulations of ExN1, ExN3 and ExN2 were unbiasedly clustered together in UMAP space and formed a lineage continuum pattern, suggesting a spectrum of cell lineage influencing susceptibility to AD across these major excitatory neuron subclusters. We subsetted and re-clustered the excitatory neurons (Figure 2E), performed cellular linage and pseudotime inference using Slingshot 39 on ExN1, ExN2, and ExN3, which presented a smooth trajectory for the simulation of disease lineage from healthy control to AD, and calculated differential expression genes along this trajectory using TradeSeq 40 (Figure 2F, Supplemental Table 5). Along this disease progression trajectory, a set of genes, including TSHZ2, DCC, and TOX, were upregulated from control to AD, while others, such as diacylglycerol kinase DGKB for neurite spine formation, were downregulated in the AD excitatory neurons, which may contribute to impaired neuronal function and cognitive decline (Figure 2G). Previous studies reported that RORB-expressing excitatory neurons were selectively vulnerable in AD specifically in the entorhinal cortex, but not in superior frontal cortex or prefrontal cortex.45,46 Surprisingly, in our MTG single cell atlas, the most vulnerable excitatory neurons in AD, such as ExN1 and ExN6 (Figure 2C), were RORB-negative and CDH9-expressing (Figure 2H). Notably, RORB-expressing excitatory neuronal subpopulations of ExN2-ExN5 remained abundant in the MTG of AD brains. This demonstrates that RORB-expressing excitatory neurons are differently affected in distinct brain regions in AD pathogenesis. We next performed marker gene alignment for top marker gene sets of the most vulnerable excitatory neuron subpopulation ExN1 to an established human MTG brain cell atlas from the Allen Brain Institute, generated by Smart-seq2 full length cDNA single cell sequencing. In the MTG, the highly vulnerable RORB-negative and CDH9-expressing excitatory neurons in AD are mostly in the cortical layers 2 and 3 (Figure 2I), consistent with a recent study of MTG. 47 To validate our findings that RORB-negative and CDH9-expressing neurons in the MTG were highly vulnerable in AD, while RORB-positive neurons are abundant in this brain region, we performed immunohistochemistry on postmortem brains of 12 individuals with or without AD. We quantified RORB-negative and CDH9-positive neurons in the cortical layers of MTG (Supplemental Figure 5A) and revealed a significant reduction in the proportion of CDH9 + RORB- neurons among all brain cells (Figure 2J-2K). In contrast, the proportions of CDH9+/RORB + neurons or CDH9-/RORB + neurons show no difference between AD and healthy control brains (Supplemental Figure 5B-D). Pan reactivation of astrocytes in AD brain middle temporal gyrus To investigate the role of astrocytes in AD brains, we subsetted and re-clustered astrocytes into 10 subpopulations of AS0, AS1, …, AS9 (Figure 3A, 3B), with no notable change in the cell type proportion of astrocytes between AD and healthy controls (Figure 3C). Neuroinflammation can induce two different types of reactive astrocytes termed as A1 and A2, which were previously defined in mice, whereas A1 astrocytes are neuroinflammatory and A2 are proliferative and anti-inflammatory. 48 We investigated whether these reactive astrocytes are induced in AD using C3 as A1 marker gene and S100A10 as A2 marker gene. We found that both A1 and A2 astrocytes are highly induced in human AD brain MTG (Figure 3D), across most of the astrocytes subclusters, with a dramatic increase in the subclusters of AS2, AS3, and AS4 (Figure 3E), indicating both harmful A1 astrocytes and potentially protective A2 astrocytes contribute to AD pathogenesis. Single cell analysis of astrocytes revealed pan reactivation in human AD brain middle temporal gyrus. (A-B) UMAP plotting of human brain astrocytes (n = 9000 astrocytes from 18 individuals with and without AD), colored by astrocyte subpopulations (AS0-AS9) (A), or disease diagnosis of either AD or healthy controls (HC) (B). (C) The percentage of astrocytes in all brain cells in each individual of AD or HC. ns, no significant difference. (D) UMAP plotting of the expression levels of A1 astrocytes marker gene C3 or A2 astrocytes marker gene S100A10 in AD and HC. (E) Dot plotting of the expression levels of A1 astrocytes marker gene C3 or A2 astrocytes marker gene S100A10 in each astrocyte subpopulation between AD and HC. (F) Dot plotting of the expression levels of pan reactive astrocytes marker genes as listed between AD and HC. The expression of C3, S100A10, GFAP, CP, CD44, VIM, and HSPB1 in astrocytic AD nuclei differed significantly from that in HC, adjusted p < 1E-07, by Wilcoxon rank-sum test. Since reactive astrocytes can be induced by central nervous system diseases, including neurogenerative disorders such as AD, we further examined astrocyte features in the AD MTG using 12 known pan reactive markers 48 for differential expression analysis (Figure 3F). We identified a robust induction of reactive astrocytes in AD brain MTG with all 12 marker genes being significantly upregulated in AD compared to healthy controls. These findings reveal a pan reactivation of astrocytes induced by AD, without obvious sex-specific variation (Supplemental Figure 6). AD-associated and female-specific cellular subpopulations and gene sets in microglia Based on the cellular composition, we observed a specific increase of microglia population in AD brains compared to healthy controls, while the percentage of neuronal and other glial cell populations remained unchanged (Figure 4A). We subsetted microglia in the MTG through re-clustering at a finer resolution into 10 subpopulations of MG0, MG1, …, MG9, and then ran top marker gene analysis for their marker gene set of each subcluster (Figure 4B-4D). The top marker genes of MG4 showed AD- and female-specific upregulation, such as CD163, TPRG1 and MYO1E (Figure 4D). Interestingly, although complement genes of C1QA and C1QC showed higher expression in male microglia than in female healthy controls, their expression increased to a substantially greater degree specifically in AD female microglia, revealing a female-specific change in AD pathogenesis (Figure 4D). We further assessed the AD-association and female-specificity in the cellular composition of these microglial subclusters. Six of the subclusters showed statistically significant association between AD diagnosis and sex, and three of these subclusters (MG4, MG6, and MG7) were enriched for female nuclei (Figure 4E). We focused on the transcriptional features of marker gene sets from MG4 and MG7 as they are AD-associated and female-specific microglia subclusters (Supplemental Table 6). While our single nucleus dataset included a limited number of donors, the enrichment of MG4 microglia subcluster was consistently detected at the individual donor level (Supplemental Figure 7A, B). This enrichment pattern remained evident after unsupervised hierarchical clustering of maker gene expression profiles at the donor level, which revealed grouping by sex and disease diagnosis (Supplemental Figure 7C). Integrated single cell and GWAS analysis revealed AD-associated and female-specific microglial subpopulations and gene sets and identified novel female-specific AD risk gene MERTK in human AD brain middle temporal gyrus. (A) The percentage of microglia in all brain cells in each individual of AD or healthy control (HC). *p-value = 0.0391, Wilcoxon test. (B-C) UMAP plotting of human brain microglia (n = 7874 microglia from 18 individuals with and without AD) in the middle temporal gyrus, colored by microglial subclusters (MG0-MG9) (B), or sex and disease diagnosis (C). (D) Dot plotting of the gene expression of AD-associated and female-specific marker genes as listed between HC males, AD males, HC females, and AD females. The expression of each of these marker genes in microglial AD nuclei differed significantly from that in HC nuclei, adjusted p < 1E-10, by Wilcoxon rank-sum test. (E) Microglial cell subtype composition between four groups of AD female (red), AD male (green), HC female (blue), and HC male (purple) brains, with their information of AD and sex association as well as female specificity. (F-G) MAGMA analysis of AD and female-specific microglial subcluster MG4 (F) or MG7 (G) marker gene sets based on ADGC GWAS dataset. The lower dashed line represents a p-value of 0.05 and the upper dashed line indicates a p-value threshold after Bonferroni correction (0.05 divided by the number of genes in the corresponding cluster). Top panel: all population with both male and female; middle panel: female only; bottom panel: male only, in the ADGC dataset. (Color figure available online). We compared our microglial subclusters MG4 and MG7 from MTG with two other microglial transcriptome datasets,49,50 including one single nucleus dataset 49 and one single cell microglia dataset 50 of human AD brains. We re-analyzed these published microglial datasets and identified their subclusters, including AD- and female-specific subclusters (Supplemental Figure 8). To assess the correlation of sex-specific AD microglial features between our dataset and these published datasets, we used gene set enrichment analysis (GSEA) to evaluate enriched expression of our AD-associated and female-specific microglial gene sets (MG4 and MG7) in the single nucleus microglia-derived subclusters (Supplemental Figure 8A-C). Marker gene sets of MG4 and MG7 showed high enrichment in subclusters 3 and 5 from published single nucleus microglia dataset, and these two subclusters from the published dataset also showed statistically significant AD association and female predominance, supporting a shared AD-associated and female-specific GSEA pattern across these datasets (Supplemental Figure 8B, C). Similarly, in the published single cell microglial dataset, our MG4 and MG7 marker gene sets aligned most closely with AD/female specific subclusters, such as subcluster 2, which showed statistically significant AD association and female predominance (Supplemental Figure 8D-F). Together, these findings indicate that AD-associated and female-specific enrichment of microglial gene sets is highly consistent and robust across datasets. Identification of novel female-specific AD risk gene MERTK from GWAS data To extend our assessment of AD-associated and female-specific microglial gene sets, we integrated our single nucleus transcriptomic findings with ADGC GWAS data to identify candidate sex-biased or sex-specific AD genetic risk factors. We used ADGC sex-specific GWAS datasets, including the first three principal components as covariates, as well as combined GWAS dataset involving both females and males with sex and its first three principal components as covariates. The ADGC GWAS dataset were generated from 29,056 participants (AD cases =13,851), of which 12,079 were male (AD cases =5988) and 16,977 were female (AD cases =7863). With the sex-specific GWAS summary statistics, we analyzed the AD-associated and female-specific microglial gene sets of MG4 and MG7 using MAGMA, 41 which is a tool for generalized gene-set analysis of GWAS data. We calculated the association of MG4/MG7 top marker gene sets with AD among the males, females, and the overall population, and identified a set of known AD risk genes including APOE, APOC1, MS4A4A, and MS4A4E 51 in all populations that includes both male and female individuals, confirming the ability of our integrated transcriptomic-genomic approach for identifying AD risk genes in GWAS data (Figure 4F, 4G). Strikingly, we identified a novel AD risk gene MERTK specifically in the female AD population, but not in the combined population or male population after Bonferroni correction for multiple testing within each gene set (Figure 4F, 4G). We next leveraged a recent published human AD brain microglial atlas generated from 443 human subjects with six brain regions of prefrontal cortex, hippocampus, middle temporal cortex, angular gyrus, entorhinal cortex, and thalamus prefrontal cortex. 52 After re-analyzing this single nucleus microglial atlas, we identified 12 clusters, including AD- and female-specific clusters 52 (Figure 5A, 5B). We further assessed AD association and female specificity across these brainwide microglial clusters and found two clusters 4 and 5 with statistically significant association between AD diagnosis and sex (Figure 5C). To determine whether these brainwide clusters carried sex-biased genetic signatures, we surveyed these brainwide cluster 4 and 5 marker gene sets with ADGC sex specific GWAS data using MAGMA analysis and calculated their associations with AD in males, females, and overall population. Consistent with our findings from microglial MG4 and MG7 in the MTG, MERTK again emerged as an AD risk gene specifically in the female AD population, but not in the combined or male population. In addition, we consistently detected the set of known AD risk genes of APOE, MS4A4A, and MS4A4E 51 across all population with both male and female individuals (Figure 5D, 5E). Integrated single cell and GWAS analysis revealed AD-associated and female-specific microglial subpopulations and gene sets and identified novel female-specific AD risk gene MERTK in brainwide AD microglia atlas. (A-B) UMAP plotting of human brain microglial atlas across six brain regions from Sun et al., 2023, colored by sex and disease diagnosis (A) or microglial subclusters (0–12) (B). (C) Microglial cell subtype composition between four groups of AD female (red), AD male (green), HC female (blue), and HC male (purple) brains, with their information of AD and sex association as well as female specificity. (D-E) MAGMA analysis of AD and female-specific brainwide microglial cluster 4 (D) or 7 (E) marker gene sets based on ADGC GWAS dataset. The lower dashed line represents a p-value of 0.05 and the upper dashed line indicates a p-value threshold after Bonferroni correction (0.05 divided by the number of genes in the corresponding cluster). Top panel: all population with both male and female; middle panel: female only; bottom panel: male only, in the ADGC dataset. (Color figure available online). To further investigate the role of MERTK in AD risk, we generated LocusZoom plots for MERTK locus in both female and male ADGC GWAS data for direct visual comparison, which revealed several SNPs with r2 > 0.8 in the locus of MERTK in females but not in males, including rs4264571, supporting MERTK as a female-specific AD risk gene (Figure 6A, 6B). To determine which human tissues may involve in the AD risk role of MERTK, we conducted tissue-specific gene-level association assessment using UTMOST v8, 42 which showed significant associations between MERTK and AD in females, but not in males, across most tissues, including brain related tissues (p-value < 0.05, Supplemental Figure 9). MERTK genetic variation as a risk factor for Alzheimer's disease selectively in females. (A-B) LocusZoom plots of MERTK in ADGC for either female (left) or male (right) GWAS dataset revealed SNPs with r2 > 0.8 in the locus of MERTK in only the female population but not the males. (C-D) MERTK eQTL analysis in all tissues in female (C) or male (D), where each point indicates one SNP in a specific tissue, the x-axis is the SNP's effect (logOR from GWAS) on AD in female (C) or male (D), and the y-axis is the SNP's effect (NES from GTEx) on the expression level of MERTK in the specific tissue (points in the same vertical line are usually the same SNP in different tissues). (E-F) MERTK eQTL analysis in brain tissue (red circles) compared to all the other tissues (gray circles) in female (E) or male (F), where each point indicates one SNP in a specific brain sub-tissue, the x-axis is the SNP's effect (logOR from GWAS) on AD in female (E) or male (F), and the y-axis is the SNP's effect (NES from GTEx) on the expression level of MERTK. (Color figure available online). We next performed an expression quantitative trait loci (eQTL) analysis to further investigate the role of MERTK in AD risk. We extracted MERTK significant eQTLs across all tissues from GTEx v8 database and tested the correlation between these eQTLs’ Z-score (logOR/SE from AD GWAS) on AD and their effects (normalized effect size, NES from GTEx) on MERTK expression in either females or males (Figure 6C, 6D). We found that in most tissues, these two sets of effect sizes are positively correlated, which suggests from the genetic perspective, higher MERTK expression corresponds to a higher risk of AD. Importantly, in brain tissues, higher MERTK expression correlated with higher AD risk specifically in females, but not in males (Figure 6E, 6F), indicating that MERTK may play a risk role in AD pathogenesis specifically in females. To validate our findings of specific increase of MERTK expression in microglia single nucleus sequencing data in the MTG of AD females (Supplemental Figure 10), we performed RNAscope in situ hybridization with co-labeling of MERTK and microglial marker IBA1 (AIF1) probes in AD and control female brains (Supplemental Figure 11A). At single molecular RNA level, we quantified both the proportions of MERTK expressing microglia and the transcript numbers in each cell and found a substantial increase of MERTK expressing microglia in female AD brains compared to female controls (Supplemental Figure 11B). We further quantified the percentage of MERTK expressing microglia among all microglia, which was similarly elevated in AD females (Supplemental Figure 11C). Thus, our results substantiate that MERTK expression is specifically increased in the microglia of female AD brains. Discussion Integrating single cell transcriptomics with GWAS data, we analyzed AD brains and identified sex-specific microglial gene signatures, leading to the identification of MERTK as a novel candidate female-specific AD genetic risk gene. These findings revealed sex-specific transcriptional changes and highlighted sex-specific risk genes in AD. Both TAM receptor tyrosine kinases Mertk (MERTK) and Axl (AXL) are important regulators of microglial physiology 53 and have previously been implicated in AD, 54 although their precise roles in AD pathogenesis remain to be further investigated. It has been suggested that Axl has a protective role in regulating neuroinflammation and underlying AD, by the longitudinal study of elevated Axl in cerebrospinal fluid. 55 Also, in a recent study, higher levels of ApoE and soluble TAM receptors Axl and Tyro3 were detectable even before the onset of cognitive impairment and could serve as reliable indicators of the subsequent development of AD. 56 Intriguingly, in contrast to the elevated MERTK level in AD females (Figure 5, Supplemental Figure 10A, B), our data showed an upregulation of AXL at the transcriptional level in microglia specifically in AD male patients (Supplemental Figure 10C, D), indicating potentially distinct and sex-dependent roles of AXL and MERTK in regulating neuroinflammation and AD, whereas AXL may be protective and more highly expressed in males, while MERTK may be harmful and more in females, contributing to sex differences in AD. The biological basis underlying these sex-specific microglial differences in AD likely involves multiple interacting mechanisms, including sex hormones, APOE-associated genetic risk, and immune aging. 57 Estrogen signaling has been shown to regulate microglial activation states, inflammatory responses, and phagocytic activity, and age-related hormonal changes may differentially influence microglial function in females and males. 58 In parallel, APOE-dependent AD risk exhibits stronger effects in females, and APOE has been closely linked to disease-associated microglial programs. Since TAM receptors interact with ApoE-related pathways involved in lipid sensing, phagocytosis, and inflammatory regulation, sex-dependent regulation of these pathways may contribute to the differential expression of MERTK and AXL observed in our study. Furthermore, aging-associated immune remodeling may differentially affect male and female microglia, leading to distinct inflammatory and homeostatic states during AD progression. 59 Together, these findings support the concept that sex-specific microglial programs may contribute to differential AD susceptibility and progression between females and males. Limitations of the current analysis include the relatively small sample size, which may restrict statistical power to detect subtle or donor-specific effects, as well as the substantial clinical and biological heterogeneity of AD, including variability in disease stage, pathology burden, and comorbidities across individuals. Nonetheless, similar microglial subcluster marker gene set enrichment, including MERTK, with AD association and female specificity was consistently replicated in multiple other and larger AD single cell transcriptomic datasets (Figure 5, Supplemental Figure 8),49,50,52 supporting the robustness of our findings. Neuroinflammation induced A1 type reactive astrocytes were found abundant in human neurodegenerative diseases, including AD.48,60 Our data are consistent with previous observation of A1 astrocytes in AD and further showed the increase of A2 reactive astrocytes in the postmortem brains of patients with AD, which may be protective. This indicates the disease progression is complicated with the involvement of both types of reactive astrocytes for neurotoxic A1 and protective A2 astrocytes. Although the classical A1/A2 framework may not fully capture the broad spectrum of reactive astrocyte states observed in neurodegenerative diseases, it remains a useful model for understanding distinct inflammatory and protective astrocyte programs associated with disease progression.61–63 Potential blockade of A1 astrocyte conversion has been shown to be neuroprotective in models of Parkinson's disease 64 and modulation of astrocyte polarization has increasingly emerged as a potential therapeutic strategy across multiple neurodegenerative conditions. Therefore, targeting the reactive astrocyte states and further investigating the role and application of suppressing neurotoxic A1 astrocytes and/or promoting protective A2 astrocytes may represent a promising direction for future AD studies. Selective neuronal vulnerability is a fundamental hallmark of AD, which has been characterized in different brain regions of postmortem AD brains, such as entorhinal cortex, prefrontal cortex and superior frontal gyrus, with diverse findings of the most vulnerable neurons.45–47 Here, we provide the molecular signature of the brain region of MTG that is strongly affected by AD, as well as selectively vulnerable neurons in this brain region, which are RORB-negative and CDH9-expressing excitatory neurons, different from other brain regions. Our findings demonstrate the brain region specificity of AD pathogenesis and emphasize the importance of potential treatment in targeting certain neuronal populations with a cautious consideration of specific brain regions and disease progression stages. Supplemental Material sj-docx-1-alz-10.1177_13872877261480719 - Supplemental material for Single-cell transcriptomic atlas of Alzheimer's disease middle temporal gyrus reveals region, cell type, and sex specificity of gene expression with novel genetic risk for MERTK in female Supplemental material, sj-docx-1-alz-10.1177_13872877261480719 for Single-cell transcriptomic atlas of Alzheimer's disease middle temporal gyrus reveals region, cell type, and sex specificity of gene expression with novel genetic risk for MERTK in female by Le Zhang, Tianyu Liu, Chuan H. He, Anthony Russo, Sarah Coffey, Dominic Yin, I-Uen Hsu, Chi Zhang, Yixuan Ye, Chang Su, Joshua Spurrier, Lashae Nicholson, Brian W. Kunkle, Eden R. Martin, Omar G. Rodriguez, Lissette Gomez, Rui Chang, Carla V. Rothlin, Sourav Ghosh, Pallavi P. Gopal, David A. Hafler, Hongyu Zhao and Stephen M. Strittmatter in Journal of Alzheimer's Disease sj-xlsx-2-alz-10.1177_13872877261480719 - Supplemental material for Single-cell transcriptomic atlas of Alzheimer's disease middle temporal gyrus reveals region, cell type, and sex specificity of gene expression with novel genetic risk for MERTK in female Supplemental material, sj-xlsx-2-alz-10.1177_13872877261480719 for Single-cell transcriptomic atlas of Alzheimer's disease middle temporal gyrus reveals region, cell type, and sex specificity of gene expression with novel genetic risk for MERTK in female by Le Zhang, Tianyu Liu, Chuan H. He, Anthony Russo, Sarah Coffey, Dominic Yin, I-Uen Hsu, Chi Zhang, Yixuan Ye, Chang Su, Joshua Spurrier, Lashae Nicholson, Brian W. Kunkle, Eden R. Martin, Omar G. Rodriguez, Lissette Gomez, Rui Chang, Carla V. Rothlin, Sourav Ghosh, Pallavi P. Gopal, David A. Hafler, Hongyu Zhao and Stephen M. Strittmatter in Journal of Alzheimer's Disease