Sex-Stratified Single-Cell RNA-Seq Analysis Identifies Sex-Specific and Cell Type-Specific Transcriptional Responses in Alzheimer’s Disease Across Two Brain Regions

Alzheimer’s disease (AD) is a pervasive neurodegenerative disorder that disproportionately affects women. Since neural anatomy and disease pathophysiology differ by sex, investigating sex-specific mechanisms in AD pathophysiology can inform new therapeutic approaches for both sexes. Previous bulk human brain RNA sequencing studies have revealed sex differences in dysregulated molecular pathways related to energy production, neuronal function, and immune response; however, the sex differences in disease mechanisms are yet to be examined comprehensively on a single-cell level. We leveraged nearly 74,000 cells from human prefrontal and entorhinal cortex samples from the first two publicly available single-cell RNA sequencing AD datasets to perform a case versus control sex-stratified differential gene expression analysis and pathway network enrichment in a cell type-specific manner for each brain region. Our examination at the single-cell level revealed sex differences in AD prominently in glial cells of the prefrontal cortex. In the entorhinal cortex, we observed the same genes and networks to be perturbed in opposing directions between sexes in AD relative to healthy state. Our findings contribute to growing evidence of sex differences in AD-related transcriptomic changes, which can fuel the development of therapies that may prove more effective at reversing AD pathophysiology. Supplementary Information The online version contains supplementary material available at 10.1007/s12035-021-02591-8.


Introduction
Alzheimer's disease (AD) is an irreversible neurodegenerative disorder that causes progressive memory decline, cognitive deficits, and behavioral changes [1][2][3] . It is the most common form of dementia and is reaching epidemic proportion as a result of extended life expectancies and increased elderly populations worldwide 4,5 . It is of high priority to find disease-modifying treatments for AD, as more than five million people are diagnosed with AD currently in the United States, a number estimated to triple by 2050 6,7 .
Although first described more than a century ago 8 , the underlying molecular mechanisms of AD remain elusive 9 . Extensive research efforts reveal that AD is histologically characterized by pathological brain aggregates including extracellular amyloid-β (Aβ) plaques, and intracellular tau protein neurofibrillary tangles (NFTs) 10,11 . Increasing evidence suggests that neuroinflammation and brain dysfunction led by neuronal supporting cells, which include microglia, astrocytes, and oligodendrocytes, could contribute to AD pathophysiology 12,13 . These pathological features are accompanied by impaired neurotransmitter signaling, dysregulated neuronal metabolism, neuronal loss, and cerebral atrophy [14][15][16] . Overall, the exact pathogenesis of AD remains uncertain, which hinders the development of effective therapies.
Sex differences have been clinically documented in AD 17,18 , yet the underlying cause for these differences are not well understood. Approximately two thirds of AD diagnoses are in women 19 .
In addition to greater longevity in females 20 , other biological differences may be responsible for the higher prevalence and accelerated cognitive decline observed in women during disease progression 18,21,22 . For instance, a longitudinal study examining a postmortem cohort of about 1,500 individuals observed that in the presence of similarly high Aβ burden, females exhibited faster cognitive decline than males 22 , suggesting females might be more susceptible to Aβ 4 toxicity. Furthermore, after adjusting for age and education, women had a higher tau tangle density 22,23 . Among genetic risk factors implicated in AD, the apolipoprotein E (APOE) ε4 risk allele has been observed to have a differential influence and increased risk for AD in women compared to men 24,25 . Sex hormones, especially the decline in hormone levels post-menopause, could also contribute to sex differences in AD progression. For example, after menopause, women experience an abrupt loss of progesterone 26 , which was previously shown to be neuroprotective by promoting myelin repair and reducing inflammation 27,28 . In fact, compared to men, women experience more inflammation-driven symptoms and have an increased risk for autoimmune diseases [29][30][31] . These findings suggest that investigating sex differences in AD will not only provide insight into deciphering the fundamental biological and mechanistic causes of AD pathogenesis, but also highlight the necessity of developing personalized therapeutic strategies.
Previous studies suggest that cellular and molecular heterogeneity in AD pathogenesis 32,33 and brain immune cell dysfunction contribute to sex-specific AD pathophysiology 34 ; however, sexspecific disease complexity at single-cell resolution is masked in bulk brain sequencing analysis.
Recent advances in single-cell RNA sequencing technology and the increasing availability of human transcriptomic datasets present a novel opportunity to examine cell type-specific transcriptional alterations in AD brain pathology. In recent years, two single-nucleus RNA-Seq (snRNA-Seq) datasets were generated from the prefrontal 35 and entorhinal 36 cortices of age and sex-matched human AD patients and cognitively normal controls. For the prefrontal cortex dataset, Mathys and colleagues performed differential expression analysis on transcriptomic results of 80,660 droplet-based nuclei within six major cell types across 48 individuals of varying degrees of AD pathology. They identified cell type-specific differentially expressed genes (DEGs) in AD in comparison to control individuals and sex-specific cell subpopulations associated with AD pathology. While the authors reported on the general sexual dimorphic 5 transcriptional response to AD pathology, they did not extensively examine sex-specific DEGs in the individual brain cell types or delineate any subsequent sex-specific molecular pathway enrichments in AD. Similar to the Mathys analysis, Grubman and colleagues analyzed 13,214 droplet-based nuclei with postmortem tissue from the entorhinal cortex of 12 age and sexmatched human AD patients and controls. Besides investigating the likelihood of sex as a covariate factor for DEG variance observed, no sex difference analysis was performed in this study.
Understanding gene expression changes unique to each sex provides opportunities to decipher molecular underpinnings that differentially contribute to AD in males and females. In this study, we leveraged the two snRNA-Seq datasets to characterize sex-stratified cell type-specific gene expression perturbations in AD and to identify sex-specific disease-associated cellular pathways as potential precision therapeutic targets. In both brain regions, we identified sex-specific disease changes primarily in glial cells and observed samples to cluster by sex when examining gene expression changes in AD compared to controls. Our findings will be of fervent interest to the field in studying differing vulnerabilities between sexes in AD.

Sample classification and analytic workflow
Samples were categorized into cases and controls based on tau tangle and Aβ plaque burdens, using Braak clinical staging and Consortium to Establish a Registry for Alzheimer's Disease (CERAD) scores 37 , respectively (AD: Braak stage ≥ IV, CERAD score ≤ 2; Control: Braak stage I-III, CERAD score ≥ 3). This resulted in single-nucleus RNA-Seq datasets containing 17,723 genes expressed by 62,741 cells from the prefrontal cortex, and 10,846 genes expressed by 11,284 cells from the entorhinal cortex, which were acquired from different sets of individuals ( Figure 1). In both brain regions, a sex-stratified differential gene expression (DGE) analysis was performed comparing AD cases to controls, with APOE genotype as a covariate, in astrocytes (Ast), microglia (Mic), excitatory neurons (Ex), inhibitory neurons (In), undifferentiated neurons (Neu), oligodendrocytes (Oli), and oligodendrocyte progenitor cells (OPCs) (Supplementary Tables 1 and 2). For the entorhinal cortex cohort, data integration was performed and APOE genotype was included as a covariate in our DGE analysis to account for batch effects and avoid collinearity in our model. Differentially expressed genes (DEGs) were determined using a Benjamini-Hochberg adjusted p-value < 0.05 and absolute log2 fold change (LFC) > 0.25 as cutoffs. DEGs were passed as inputs for pathway enrichment analysis, which provided pathways to be used as inputs for subsequent network analysis. We examined gene expression and pathway network differences in AD versus neurotypical cells to identify cell typeand brain region-specific and non-specific differences based on sex.

DGE analysis in the prefrontal cortex reveals modest sex-specific disease related changes specifically in glial cell types
Leveraging data from Mathys et al., from our sex-stratified DGE analysis, we identified DEGs meeting significance and LFC thresholds (Table 3) in all cell types except male inhibitory neurons when comparing AD to non-AD (Supplementary Table 3). We identified 73 DEGs across all cell types in the prefrontal cortex (Table 3, Supplementary Table 3). Of these DEGs, 36 were shared in both sexes, while 8 and 29 were specific to AD compared to control males and females, respectively. We also observed more shared DEGs in AD case versus control female signatures versus male signatures across the cell types (Fig. 2a), which is consistent with previous bulk tissue analysis 34 . Some of the differentially expressed genes include LINGO1, a negative regulator of myelination 38,39 , which we found upregulated in all AD compared to control female cell types; SLC1A3, which encodes excitatory amino acid transporter 1 that transports glutamate in the synaptic cleft 40 and was perturbed in all female AD compared to control cell types except oligodendrocytes and OPCs; and SPP1, a protein involved in neuroinflammation also known as Osteopontin 41 that we observed to be upregulated in AD versus control samples of both female and male excitatory neurons and microglia, as well as female astrocytes and inhibitory neurons. Also, clustering samples by AD compared to control pseudo-bulk cell type gene expression (Fig. 2b) showed samples to be clustered primarily by sex before cell type identity for all cell types except excitatory neurons.
In addition to identifying shared DEGs across cell types and sexes, we also observed a larger  were shared in both sexes, while 20 and 1 were specific to AD compared to control males and females, respectively. We observed shared DEGs across cell types when comparing AD versus control samples in both sexes (Fig. 3a). Some of these globally shared genes include CLU 9,46 ,

Comparative analysis across brain regions reveals more shared transcriptomic sex differences in the entorhinal cortex
We compared DEG results from the prefrontal and entorhinal cortices to determine whether changes in each sex were consistent across brain regions. Overall, we observed more overlaps in across sex DEGs to be in the entorhinal cortex (Fig. 4a). Additionally, clustering samples by AD compared to control pseudo-bulk cell type gene expression ( Fig. 4b) showed samples to be clustered primarily by brain region and sex, and not by cell type.

Pathway and network analysis reveals sex-specific transcriptomic perturbations in glial cells in the prefrontal cortex and sex-shared, but flipped AD-enriched pathways in the entorhinal cortex
Beyond identifying sex-dimorphic disease-associated genes, we performed a gene set enrichment analysis to elucidate potential biological mechanisms implicated in disease progression that are either shared or unique to each sex and to reveal the interconnections between disease-linked pathways within AD. The pathway enrichment was performed in g:Profiler 50 , a web tool that performs functional enrichment analysis from a given gene list, using separate lists of upregulated and downregulated DEGs with an adjusted p-value <0.05 and relaxed absolute LFC above 0.1 in cell types of each sex as inputs. Significantly enriched biological pathways with an adjusted p-value < 0.05 were applied to EnrichmentMap 51 , a functional category grouping method from the Cytoscape software, to identify pathway network clusters annotated by associated biological processes (Fig. 5, Supplementary Figures 3 and 4).
Female and male AD compared to control excitatory neurons of the prefrontal cortex shared six common enriched clusters of pathways (Fig. 5a), which were all perturbed in the same direction for both sexes. Two of these clusters (neurotransmitter glutamate/aspartate transmembrane activity and carboxylic acid biosynthetic process) were upregulated in disease in both sexes. Of the four downregulated pathway clusters, three were related to synaptic activity (modulation of the synaptic membrane, neurotransmitter release, and synapse assembly/cell junction organization), indicating a dysregulation of synaptic plasticity in AD excitatory neurons. The other downregulated pathway cluster was plasma membrane morphogenesis, which consisted of pathways including axonogenesis, cellular projection, and plasma membrane organization. Tables 5 and 6).

(Supplementary
In prefrontal cortex excitatory neurons, we also identified uniquely enriched disease pathway clusters for each sex (Fig. 5a). Female excitatory neurons showed upregulation of the HOXA5 factor, a DNA-binding transcription factor that regulates cell morphogenesis and tumor suppressor that inhibits proliferation and induces apoptosis 52 , and downregulation of inflammatory-mediated cell to cell interaction through adhesion and molecule binding.
Interestingly, a recent epigenome-wide association study examining samples in the prefrontal cortex and superior temporal gyrus observed elevated DNA methylation of the HOXA gene cluster to be associated with neuropathology in AD 53 . In male excitatory neurons, we observed upregulation of axon regeneration, and downregulation of distal axonal growth cone polarization. Interestingly, we also observed downregulation of tetrahydrobiopterin (BH4) synthesis, which is important for the production of essential neurotransmitters 54 , and Rho GTPase activities in male AD compared to control excitatory neurons. Overall, excitatory neurons of the prefrontal cortex shared most case vs control differentially enriched pathways between male and females, the majority of which were downregulated in AD.
Similar to enriched pathways in disease observed in excitatory neurons, the inhibitory neurons of the prefrontal cortex showed upregulation for glutamate/aspartate activities in both female and male AD inhibitory neurons compared to controls (Fig. 5b). Like male AD excitatory neurons, male AD inhibitory neurons also showed downregulation of axonal growth cone polarization and BH4 activities compared to controls. In addition, males specifically demonstrated upregulation in anterograde synaptic transmission and downregulation of nitric synthase, heat shock protein 90 (HSP90) complex, voltage potassium transporter, and kainite calcium-permeable receptors activities in AD. The ITGAV-ITGB-SPP1 complex, with known function in cell adhesion 55 and without previous links to AD, was uniquely upregulated in male inhibitory neurons. Of note, the pathway cluster neuronal projection was upregulated in females and downregulated in males, consistent with the enriched upregulated pathways clusters uniquely observed in females, which were modulation of spine morphogenesis and synaptic membranes. Lastly, the transcription factors, nuclear receptor TLX (essential for the regulation of self-renewal, neurogenesis and maintenance in neuron stem cell) 56 and nuclear protein HOXB2 (involved in cellular development) 57 , were upregulated only in AD female inhibitory neurons.
Unlike in neurons in the prefrontal cortex, we identified a variety of commonly enriched disease pathway networks in entorhinal cortex neurons that were regulated in opposite directions for the sexes (Fig. 5c). For instance, amyloid-beta binding/fibril formation, mitochondrial abnormality, coupled electron ATP metabolic process, demyelination/remyelination, cellular metabolism, extracellular organelle exosome vesicle and cation transmembrane transport were among the clusters downregulated in females and upregulated in males. We did not observe any pathway networks unique to female neurons; however, for the AD male neurons in the entorhinal cortex, we identified pathways in maintaining cellular metabolism and homeostasis, through the upregulation of genes involved in axon myelination, regulation of the metabolic process, cell component locomotion, cytoskeleton organization, and intracellular ferritin complex (iron storage). In male neurons, we also observed synaptic activity deficiency, indicated by the downregulation of pathways in synaptic vesicle transport, presynaptic assembly at cell junction, synaptic membrane clustering, postsynaptic membrane morphogenesis, chemical regulation at the synapse, neuroligin family protein binding, and ionotropic receptor signaling. Additionally, male AD neurons compared to controls also showed downregulation in plasma membrane regulation, cell projection, and developmental process in differentiation. As a whole, while sex differences are minimal in the neurons of the prefrontal cortex, we observed overwhelmingly shared but inversely regulated enrichment pathways in the neurons of the entorhinal cortex.
Microglia, the resident immune cells of the brain, have gained growing recognition as being critically involved in AD pathogenesis due to their key role contributing to neuroinflammation, a prominent feature of AD 58 . Only a few significantly enriched disease pathways were observed in microglial cells of the prefrontal cortex and none were shared across sexes (Fig. 5d). We observed upregulation of axon sprouting in response to injury in males, as well as an enriched upregulated pathway in axonogenesis regulation in females (Supplementary Table 6).
Interestingly, a cluster of the PDE4B-DISC1-complex, with important functions in cAMP-regulated signal transduction and synaptic plasticity, 59 was downregulated in females.
The phosphodiesterase 4B (PDE4B) enzyme was previously shown to be pro-inflammatory in microglia and is currently under study as a therapeutic target for neuroinflammation and cognitive function impairment 59 .
Microglia in the entorhinal cortex had mostly downregulated pathway clusters in females and upregulated pathway clusters in males (Fig. 5e). Amyloid fibril formation, chaperone-mediated autophagy, protein folding, protein stability regulation, cell junction synapse, neurogenesis structure development, and cell body assembly were among the clusters shared by both sexes 13 but downregulated in females and upregulated in males. Protein homeostasis was altered in disease for females, as shown by downregulation of tau protein kinase activity, tau protein binding, protein folding chaperone, and histone deacetylase binding. Protein degradation and secretion were also downregulated in females with AD compared to controls, as indicated through downregulation of lytic vacuole lysosome and secretory granule vesicle exocytosis respectively. Interestingly, nitric oxide synthase 3 (NOS3), which is involved in a complex cascade of events in oxidative stress that may induce cellular injury and accelerate neurodegenerative changes 60 , and its chaperone, HSP90 61 , were downregulated in AD females compared to controls. In males, myelination in axon ensheathment, synaptic signaling transmission, and energy coupled proton transport were upregulated. We also identified downregulation of two microRNA clusters, hsa-miR-190a and hsa-miR-3605, in AD males compared to healthy controls. These are potentially important findings because epigenetic modulation by microRNAs has the capacity to modify microglial behavior in physiological conditions, and dysregulation of microRNAs could mediate microglial hyper-activation and persistent neuroinflammation in neurological diseases 62 . Overall, we observed extensive sexspecific pathway enrichments in microglial populations of AD compared to controls for both brain regions, but especially pronounced in entorhinal cortex.
Furthermore, astrocytes, oligodendrocytes, and OPCs also demonstrated sex-specific pathway perturbations in both prefrontal and entorhinal cortices ( Supplementary Fig. 3 and 4). In astrocytes, which normally function to maintain overall brain homeostasis, we observed downregulated plasma and presynaptic membrane components and upregulated postsynaptic asymmetric synapse density in the prefrontal cortex of AD compared to controls in both sexes.
In female AD astrocytes, we observed downregulation in pathways related to amino acid transport and vascular transport across the blood brain barrier. Although the downregulation of these pathways was not observed in males, a related pathway cluster, presynaptic filopodia 14 activities, was downregulated. These observed pathway networks suggest that the same biological process, regulation of synaptic activities, was disrupted in both sexes but via different mechanisms.
In oligodendrocytes, which provide support and insulation to axons in the brain, we observed downregulation in pathways related to regulation of synaptic activity in both female and male AD compared to controls, indicated by the downregulated clusters of cleft regulation, presynaptic assembly, and transmembrane transport channel in females, and neurotransmitter secretion, transmembrane ion transporter, and postsynaptic membrane potential regulation in males.
Interestingly, pathways related to cell morphological changes and energy production were upregulated in males and downregulated in females, such as pathway clusters of neuron projection organization, cell migration/locomotion, cellular component organization, ATP coupled electron transport, mitochondrial NADH dehydrogenase, suggesting oligodendrocyte responses were sex-specific when challenged by disease.
Lastly, we observed upregulation of membrane morphogenesis in female OPCs in the prefrontal cortex, as well as related pathway cluster, TROY-NGR-LINGO1-NGFR complex, which plays essential roles in the inhibition of axonal regeneration 63 . In the entorhinal cortex, a few pathways were downregulated in female and male OPCs, including cell junction synapse assembly, glutamatergic synapse, and plasma membrane intrinsic component. The male OPCs of the entorhinal cortex were overwhelmingly enriched with upregulation in neuronal development, axon ensheathment, neuron myelination, metabolic protein regulation, as well as ion and vesicle transport, with the exception that synaptic membrane adhesion molecules were downregulated.
Although inconclusive due to the unbalanced numbers of significantly enriched pathways obtained in OPCs from both sexes, our observations suggest that AD female OPCs in the

Discussion
Men and women show differing vulnerabilities to AD, with increased longevity and prevalence in women, and decreased tau and possibly cognitive decline in men [17][18][19]21,22 . To understand how AD presents in each sex on a cell type-specific level, we performed a sex-stratified differential gene expression (DGE) and pathway network analysis on the five main brain cell types using the first two publicly available human single nucleus RNA-Seq datasets. The two datasets target two separate brain regions, the entorhinal and prefrontal cortices, and we analyzed each in a sex-stratified manner, then compared findings across sexes and brain regions to highlight both general and cell type-, region-, and sex-specific transcriptional phenotypes of AD (Fig. 1).
Our gene level analysis in the prefrontal cortex showed more disease-related changes in females with AD than in males in comparison to their respective control cohorts (Fig. 2). There were also more DEGs shared among cell types in females versus males (e.g. LINGO1 38,39 , SLC1A3 40 , SPP1 41 ). While we observed a larger range of fold change in our female DGE analysis, an overall comparison across sex within each cell type showed modest differences in disease related gene expression changes. Additionally, through clustering prefrontal cortex samples based on AD compared to control pseudo-bulk gene expression, we observed samples to cluster first by sex in all cell types except excitatory neurons. In the entorhinal cortex however, we observed a higher magnitude of change in males with AD than in females in comparison to their respective control cohorts. Compared to the prefrontal cortex, we observed more overall DEGs and many global changes across cell types (Fig. 3a). Through clustering entorhinal cortex samples by AD compared to control pseudo-bulk gene expression, we observed samples to cluster by sex for all cell types, and also observed opposing expression patterns across sex (Fig. 3b), which we visualized in a handful of DEGs and examined in a pairwise manner (Fig. 3c-d). Moreover, our comparative analysis across brain regions showed more DEG overlaps across sex in the entorhinal cortex, of which we observed flipped directionality in disease related gene expression changes (Fig. 3, Fig. 4).
From the gene-set enrichment and pathway clustering network analysis, we identified sexspecific pathway network changes, which are potentially involved in AD pathogenesis through mechanisms unique to each sex (Fig. 5, Supplementary Figure 3  activity was downregulated, as well as its regulating factors, the HSP90 complex and co-factor BH4. BH4 has been extensively studied in its role of regulating nitric oxide production from nitric oxide synthases and superoxide anion radical (O2 *-) release in the endothelium 64 . Our pathway enrichment analysis suggests that perhaps excessive O2 *in diseased male neurons due to dysregulated NOS activities and BH4 levels could lead to neuronal stress and death. Therefore, resolving the chronic BH4 deficiency and change in redox state of neurons pharmacologically could be a beneficial therapy for AD male patients. In males, we observed a decrease in presynaptic intrinsic component filopodia activities. These pathways were interconnected, indicating that they belong to related biological processes, which suggests that similar resulting synaptic deficiencies were observed in both sexes but resulted from different pathway mechanisms. These are compelling evidence for focusing on glial cell pathophysiological changes in studying sex-difference in AD pathogenesis.
In the entorhinal cortex, while similar to in the prefrontal cortex, we identified sex-specific perturbed pathway networks in all cell types, where the pathways shared across sexes were overwhelmingly of opposite direction, with most pathways downregulated in female and upregulated in males (Fig. 5, Supplementary Figure 3 and Supplementary Figure 4). Out of the five cell types investigated, two were dominated by enriched pathways detected in males (neurons and OPCs), one was dominated by enriched pathways detected in females (oligodendrocytes), and two were more evenly distributed (microglia and astrocytes). The diseased female microglia demonstrated deficiency in tau protein processing uniquely, by downregulation of tau kinase activity and tau protein binding. Additionally, disruption of cellular protein homeostasis was also observed in female microglia, indicated by downregulation of protein folding chaperone, histone deacetylase binding, lysosomal activity, and exocytosis vesicle secretion. The female microglia were perceived as deficient in dealing with the degradation of the debris and cellular waste that they phagocytosed while the male microglia were active at combating the disease environment by upregulating axonal myelination, synaptic transmission signaling, cellular component assembly and energy production through energy coupled proton transport. As immune cells are critical for repair after injury, this may indicate that female AD risk relates to decreased ability to properly recover after deleterious events over time.
While we observed evidence of sex-dimorphic disease changes in glial cells in AD, it is important to note some limitations in the study. First, the data sets were limited in sample size.
The entorhinal cohort consisted of six cases (two female, four male) and five controls (two female, three male) (Fig. 1, Table 2), while the prefrontal cohort consisted of 20 cases (10 female, 10 male) and 22 controls (10 female, 12 male) (Fig. 1, Table 1). Second, there were batch effects in the entorhinal cortex data introduced by the study design. This was overcome by performing data integration and including APOE genotype as a covariate in our DGE analysis to account for batch and avoid collinearity in our model. Next, literal biological sex could be a misleading classifier for trans* individuals. A properly powered study of differences between male versus female versus recipients of testosterone-versus estrogen-focused hormone replacement therapy might help narrow down a genetic versus hormonal basis of DEGs deemed sexually dimorphic. Finally, although both datasets were age-matched, they were not APOE genotype matched. APOE4 is the largest risk factor in AD, and as a result, we would expect some transcriptional differences based on the APOE genotype of a sample 66 . In the prefrontal cortex cohort, female samples had cases but not controls with the ε4 allele of APOE, and male samples had cases and only one control sample with ε4 allele of APOE (Table1). In the entorhinal cortex cohort, female samples included one of two cases and no controls with an ε4 allele of APOE, and all male cases had at least one ε4 allele of APOE, and one of three control samples had an ε4 allele of APOE (Table 2). While we accounted for APOE genotype as a covariate in the DGE analysis, the interactions of sex and APOE genotype may still explain trends that we observe. We hope that future explorations of sex-specific transcriptomic changes in AD will include larger datasets from more brain regions with individuals of diverse age groups, racial and ethnic backgrounds, and APOE genotypes.
In general, our findings suggest that AD signatures in neurons in the prefrontal cortex were more similar in females and males compared to glial cells, as indicated by the proportions of sex-shared genes and pathways with directionally similar regulation in each cell type (  67 . Perhaps future studies could also explore the specific association between the gene changes in the entorhinal region with tau burden. Collectively, these observed sex-specific transcriptomic changes provide a valuable resource to study sex-specific cell type-specific pathophysiology of AD. Although expression differences in all cell types may be relevant to disease mechanisms in AD, we focused on discussing the cell types with the most compelling findings in our study: neurons, astrocytes and microglia. We hope this work serves as a resource for follow-up studies that will examine more deeply all the cell types and their specific roles leading to sex-specific AD pathophysiology.

Materials Availability
This study did not generate new unique reagents.

Data and Code Availability
We accessed single nuclei RNA-Seq counts data from the prefrontal cortex via the Accelerating

Study Cohorts
The prefrontal cortex cohort comprised age and sex matched samples from 24 males and 24 females with varying degrees of AD pathology. We reclassified samples based on tau and amyloid β (Aβ) plaque burden, using Braak clinical staging and Consortium to Establish a Registry for Alzheimer's Disease (CERAD) scores, respectively. We defined cases as individuals with severe tau deposition (Braak ≥ IV), and high Aβ load (CERAD ≤ 2), and non-AD controls as individuals with low tau (Braak ≤ III) and low Aβ load (CERAD ≥ 3). For our sexstratified analysis, we focused on 20 cases (10 female, 10 male) and 22 controls (10 female, 12 male) (Fig. 1, Table 1).

22
The entorhinal cortex cohort consisted of age matched 6 (2 female, 4 male) AD patients and 6 (2 female, 4 male) control subjects, which were classified based on pathological analysis of amyloid β plaques, Braak clinical staging, and cognitive impairment records as done in the prefrontal cohort. Note, all cases in this cohort have numerous diffuse and neuritic amyloid beta plaques, and a Braak staging score of VI. We excluded one control male sample with the APOE2/4 genotype. For our sex-stratified analysis, we focused on 6 cases (2 female, 4 male) and 5 controls (2 female, 3 male) (Fig. 1, Table 2).

Prefrontal Cortex
Seurat's Read10X function was used to generate a count data matrix using the filtered count matrix of 17,296 genes and 70,634 cells, gene names, and barcodes files provided by 10X. A Seurat object was created with the count data matrix and metadata and filtered to keep genes present in at least 3 cells, and cells meeting cohort selection criteria of at least 200 genes. Log normalization was performed using Seurat's NormalizeData function with a scale factor of 10,000, and highly variable features were identified using Seurat's FindVariableFeatures, To identify cell types, following similar steps as Grubman and colleagues 36 , we applied Seurat's AddModuleScore function to lists of 200 brain cell type markers from the BRETIGEA 72 package to identify each cell type. Cell types assessed included astrocytes, neurons, microglia, oligodendrocytes, oligodendrocyte progenitor cells, pericytes, and endothelial cells. Cells with the highest score across brain cell type markers were labeled the corresponding cell type, and if the highest and second highest score were within 20%, cells were deemed hybrids and excluded from further analysis. We further confirmed successful cell type identification by assessing homogeneity and separation of clusters in UMAP plots, and by examining expression of top marker genes across cell types. While cell type identification with BRETIGEA package's cell type markers was comparable to the original paper's identification, we found the original paper's cell types more comprehensive as it distinguished excitatory from inhibitory neurons.
Thus, we used the original paper's cell type labels for the further analysis (Supplementary Table   1). Due to low cell counts, we did not analyze pericytes and endothelial cells. The final Seurat object contained 17,723 genes and 62,741 cells.

Entorhinal Cortex
We acquired a filtered raw expression matrix of 10,850 genes and 13,214 cells, which was originally composed of 33,694 genes and 14,876 cells and filtered as described by Grubman and colleagues. A Seurat object was created and consisted of genes in at least 3 cells, and cells with at least 200 genes. Normalization was performed using Seurat's SCTransform 73 method, and Seurat's integration workflow was performed to correct the confounded batches introduced by the original study's experimental design.
Dimensionality reduction was performed using values from the integrated assay to assess successful batch correction (Supplementary Figure 1). Using the method for cell type identification described for the former cohort, we identified astrocytes, endothelial cells, neurons, microglia, oligodendrocytes, and oligodendrocyte progenitor cells. We further confirmed successful cell type identification by assessing homogeneity and separation of clusters in UMAP plots. Due to limitations in the number of cells, we excluded endothelial cells from further analyses. The final Seurat object contained 10,846 genes and 11,284 cells (Supplementary Table 2).

Cell Type-Specific Sex-stratified Differential Expression Analysis
To generate molecular signatures relative to sex in each cell type, we used the Limma 74,75 package's Voom 76 pipeline for RNA-seq. For the prefrontal and entorhinal cortices, we performed a sex-stratified analysis including APOE genotype as a covariate. For the entorhinal cortex cohort, while we integrated batches in our pre-processing, we were not able to include batch as a covariate, as its collinearity did not allow for an appropriate model fit.
After the design formulas were established, the DGEList object was created from a matrix of counts extracted from the corresponding Seurat objects. To improve the accuracy of mean-variance trend modeling and lower the severity of multiple testing correction, lowly expressed genes were filtered out using edgeR's FilterByExpr function with default parameters.
Normalization was performed with Trimmed Mean of M-values with singleton pairing (TMMwsp), followed by voom, model fitting with a contrast matrix of each defined case-control comparison, and Empirical Bayes fitting of standard errors. We determined differentially expressed genes as those with a Benjamini-Hochberg corrected p-value less than 0.05, and an absolute LFC greater than 0.25.

Pathway Analysis
We performed an overrepresentation analysis of DEGs from the cell type-specific sex-stratified analysis of cells from the prefrontal and entorhinal cortex using g:Profiler 50 , a web tool that performs functional enrichment analysis from a given gene list. We queried differentially expressed genes comparison split by upregulated and down-regulated expression and selected enriched pathways with a Benjamini-Hochberg adjusted p-value cutoff of 0.05. In addition to Gene Ontology cellular components, biological processes, and molecular functions, our enrichment analysis also provided pathways from the Human Protein Atlas, Human Phenotype Ontology, KEGG, Reactome, and Wiki pathways.

Network Visualization of Enrichment Results
We followed a previously established protocol 51 for network enrichment analysis on pathway results derived from our cell type-specific DEGs. Briefly, pathway results were imported into the Cytoscape visualization application, EnrichmentMap. Then, redundant and related pathways were collapsed into single biological themes using the AutoAnnotate Cytoscape application.