Single‐cell analysis reveals transcriptomic reprogramming in aging primate entorhinal cortex and the relevance with Alzheimer's disease

Abstract The entorhinal cortex is of great importance in cognition and memory, its dysfunction causes a variety of neurological diseases, particularly Alzheimer's disease (AD). Yet so far, research on entorhinal cortex is still limited. Here, we provided the first single‐nucleus transcriptomic map of primate entorhinal cortex aging. Our result revealed that synapse signaling, neurogenesis, cellular homeostasis, and inflammation‐related genes and pathways changed in a cell‐type‐specific manner with age. Moreover, among the 7 identified cell types, we highlighted the neuronal lineage that was most affected by aging. By integrating multiple datasets, we found entorhinal cortex aging was closely related to multiple neurodegenerative diseases, particularly for AD. The expression levels of APP and MAPT, which generate β‐amyloid (Aβ) and neurofibrillary tangles, respectively, were increased in most aged entorhinal cortex cell types. In addition, we found that neuronal lineage in the aged entorhinal cortex is more prone to AD and identified a subpopulation of excitatory neurons that are most highly associated with AD. Altogether, this study provides a comprehensive cellular and molecular atlas of the primate entorhinal cortex at single‐cell resolution and provides new insights into potential therapeutic targets against age‐related neurodegenerative diseases.

Buffalo, 2020). It forms circuits with different brain regions (Schultz et al., 2015), such as the hippocampus, amygdaloid nucleus, and neocortex (Gerlei et al., 2021). It processes information generated by the cerebral cortex and sends it to the hippocampus and amygdala, and vice versa. Thus, the entorhinal cortex is the "interface" for continuous information exchange between the hippocampus and neocortex (Sirota et al., 2003), and plays a crucial role in the acquisition, retrieval, and extinction of many forms of learning and memory (Coutureau & Di Scala, 2009;Eichenbaum et al., 2007). Pathological changes in the entorhinal cortex are associated with a variety of neurological diseases, particularly Alzheimer's disease (AD) (Khan et al., 2014). The entorhinal cortex is one of the first cortical brain regions to exhibit neuronal loss in AD (Braak & Braak, 1995;Leng et al., 2021).In addition, entorhinal cortex is among the first cortical fields to accumulate formation of β-amyloid (Aβ) and neurofibrillary tangles (NFTs) in AD brains (Huijbers et al., 2014;Knopman et al., 2019). Therefore, a comprehensive understanding of the mechanisms underlying aging in the entorhinal cortex could provide insight into disease mechanisms and lead to therapeutic strategies.
Non-human primates (NHPs), such as cynomolgus monkeys, are similar to humans in terms of entorhinal cortex structure, anatomical location, and function (Garcia & Buffalo, 2020), Therefore, analysis of the entorhinal cortex isolated from monkeys will help to better understand the etiology of aging-related memory loss and cognitive decline(M.L. Li et al., 2019). Given the cellular heterogeneity of the entorhinal cortex (Kim & Park, 2021), the application of single-cell/ nucleus RNA sequencing (scRNA-seq/snRNA-seq) could expand our understanding of how cell types are affected during entorhinal cortex aging (J. Li et al., 2021;W. Zhang et al., 2020).
Here, we obtained a single-nuclear transcriptome atlas of the monkey entorhinal cortex as well as clarified gene and pathway alterations in a cell-type-specific manner during entorhinal cortex aging. Moreover, we integrated multiple neurodegenerative disease datasets based on single-cell transcriptome data to clarify the correlation between disease and entorhinal cortex aging. This study advances our understanding of entorhinal cortex aging at the singlecell level and elucidates potential therapeutic targets for interventions against neurodegenerative diseases in humans.
To analyze cell populations and molecular characteristics, we performed snRNA-seq on the entorhinal cortex of the cynomolgus monkeys (Figure 1a). After cell quality control and filtering, 76,839 single cells were retained for downstream analyses. Using unbiased clustering and uniform manifold approximation and projection (UMAP) analysis, we identified seven cell types in the entorhinal cortex based on classic cell-type-specific markers (Figure 1c;  Furthermore, we identified the upstream regulators that drive cell differentiation in the entorhinal cortex. For example, the regulons of CREM and MEIS2, which are involved in cell differentiation and neurodegeneration (Mantamadiotis et al., 2002), were crucial regulators for neuronal lineage differentiation (ExN/InN) ( Figure 1f, Table S3).
Together, our results clarify the cellular characteristic in the entorhinal cortex, providing the first single-nucleus transcriptomic map of the entorhinal cortex in NHPs.

| Neuronal lineage is most affected by entorhinal cortex aging
We next examined cell type-specific transcriptional changes in the entorhinal cortex during aging. Comparing the relative cell proportions between young and aged NHP entorhinal cortices by multivariate test (Smillie et al., 2019), we found no significant changes in any cell types (Figure 2a; Figure S2). Next, we analyzed differentially ex-

| Transcriptomic reprogramming in aging primate entorhinal cortex
DEGs analysis revealed significant changes in several key genes during entorhinal cortex aging ( Figure 2d, Table S4). For instance, OLFM, which regulates neural progenitor maintenance and axon growth (Nakaya et al., 2012), was the most significantly up-regulated gene in the ExNs, suggesting abnormal neurogenesis in the aged entorhinal cortex. APOE, which plays a role in lipid metabolism, Aβ aggregation, and tau damage (Yin & Wang, 2018) Table S5).
Thus, these pathways are proposed as mediators of abnormal crosstalk between cell types in the aged entorhinal cortex.
Given the similar functions and frequent information exchange between the entorhinal cortex and hippocampus (Ku, Ku et al., 2021), we next asked whether similar aging mechanisms exist between these two brain regions. By performing comparative analy-

| Transcriptomic reprogramming in aging primate entorhinal cortex is associated with neurological diseases
Entorhinal cortex aging is a major risk factor for cognitive and memory deficits (Hou et al., 2019). Given the key role of entorhinal cortex aging in AD, we performed an integrated analysis of AD-associated DEGs (AD DEGs) from the human entorhinal cortex (obtained from previous snRNAseq data [Grubman et al., 2019]) and aging-related DEGs from the entorhinal cortex in our study. We identified 166 shared genes between the AD DEGs and aging-related DEGs (Table S6). These overlapping DEGs were primarily enriched in neurons (Figure 4c), suggesting that the neuronal lineage in the aged entorhinal cortex is more prone to AD. GO analysis showed that the up-regulated overlapping DEGs were primarily related to synaptic signaling, whereas the down-regulated overlapping DEGs were mainly associated with neurogenesis ( Figure 4d).

| The expression of Aβ and NFT increased across multiple cell types in aged entorhinal cortex
The entorhinal cortex is one of the brain regions in which Aβ and NFTs are first detected in old age, both with and without mild cognitive impairment (Thaker et al., 2017).

| ExNs subpopulations in aged entorhinal cortex are prone to AD pathology
To determine correlations between cell types and AD phenotypes and identify key cell types relevant to AD, we used Single-

Cell Identification of Subpopulations with Bulk Sample Phenotype
Correlation (Scissor) , which can identify cell subpopulations associated with a given phenotype from single-cell data.
Scissor integrates phenotype-associated bulk expression and singlecell data by quantifying similarity between each single cell and each bulk sample, then optimizes a regression model on the correlation matrix with the sample phenotype to identify relevant subpopulations . We applied Scissor to the scRNA-seq data from the aged entorhinal cortex with bulk transcriptomes from AD and non-AD entorhinal cortices (Jia et al., 2021) (Figure 5a). Results show that aged entorhinal cortical ExNs were more prone to AD than the other cell types (Figure 5b, c).
Selective vulnerability is a fundamental feature of neurodegenerative diseases, in which different neuronal populations show a gradient of susceptibility to degeneration. ExNs are heterogeneous and include multiple subpopulations with distinct molecular and projection properties (Erwin et al., 2021). Therefore, we applied Scissor, guided by bulk samples with AD, to identify aggressive ExNs cell subpopulations within 31,617 ExNs from the scRNA-seq dataset of the aged entorhinal cortex (Jia et al., 2021). These cells were separated into 9 clusters (Figure 5d (Table S8). Notably, functional enrichment analysis also confirmed that the synaptic signaling and adenosine triphosphate (ATP) metabolic processes were activated in Scissor_AD ExNs (Figure 5h). To further demonstrate the phenotypic associations of the cell subpopulations identified by Scissor, we constructed molecular signatures based on the DEGs in Scissor-identified cell subpopulations and used independent AD datasets to evaluate the functions of these signatures (Jia et al., 2021). As a result, the enrichment scores of the corresponding molecular signatures in A Scissor_AD ExNs were significantly higher in patients with AD than in normal controls ( Figure 5i). Thus, this Scissor_AD ExNs subpopulation could play a vital role in AD progress.
Taken together, Scissor analysis identified ExNs subpopulations that are most highly associated with AD, which could contribute to comprehending the underlying pathogenesis of AD and might facilitate disease diagnosis and therapy.

| DISCUSS ION
The entorhinal cortex plays a key role in cognition and memory and is an information exchange center for multiple brain areas (Gerlei et al., 2021). Abnormal entorhinal cortex function is implicated in multiple neurodegenerative diseases (Reagh et al., 2018). However, this brain region has received less attention than other regions such as the hippocampus and prefrontal cortex. In the current study, we used cynomolgus monkeys to construct a single-cell map of the entorhinal cortex and identify age-associated transcriptional changes. Our findings suggested widespread transcriptional changes across multiple cell types during entorhinal cortex aging, thus highlighting potential therapeutic targets for aging-related neurodegenerative disorders.
Our results showed that the synapse signaling-related pathway was widely down-regulated across cell types in the aged entorhinal We also observed acute changes in the neuronal lineage during entorhinal cortex aging. Specifically, we found the highest number of aging-related DEGs was observed in the neuronal lineage; genes associated with aging, AD, PD, MD, and LD were significantly more active in the neuronal lineages; and the overlap in aging-DEGs and AD DEGs was most notable in the neuronal lineage. Therefore, our results confirmed that the neuronal lineage was more vulnerable to aging in the entorhinal cortex and more susceptible to neurological disease.
We systematically explored the association between entorhinal cortex aging and AD (Reagh et al., 2018). Integrative analysis reveals a huge overlap between aging DEGs and AD DEGs across cell types. A hallmark of AD pathology is the accumulation of Aβ and phosphorylated tau (Iwata et al., 2014). In our study, APP and metabolic pathways are most highly associated with AD, which provide potential therapies for the diagnosis and treatment of AD. Nuclei were defined as DAPI-positive singlets.

| Single-nucleus RNA-sequencing library preparation
The concentration of the single-nucleus suspension was adjusted to 3 ~ 4 × 10 5 nuclei/mL in PBS and then loaded onto a microfluidic chip (GEXSCOPE® Single Nucleus RNA-seq Kit, Singleron Biotechnologies). The snRNA-seq libraries were constructed according to the manufacturer's instructions (Singleron Biotechnologies).
The resulting snRNA-seq libraries were sequenced on an Illumina HiSeq X10 instrument to a sequencing depth of at least 50,000 reads per cell with 150-bp paired-end (PE150) reads.

| Generation of single-cell gene expression matrices
Raw reads were processed to generate gene expression matrices with scopetools (https://anaco nda.org/singl eronb io/scope tools).
First, reads without polyT tails were filtered; then, cell barcodes and unique molecular identifiers (UMIs) were extracted. Adapters and polyA tails were trimmed before aligning reads to the pre-mRNA reference (Ensemble, Macaca_fascicularis_6.0). Second, reads with the same cell barcode, UMI, and gene were grouped together to count the number of UMIs per gene per cell. Cell number was then determined according to the "knee" method, a standard single-cell RNA-seq quality control approach used to determine the threshold at which cells are considered valid for experimental analysis. Highquality barcodes are located to the left of the inflection ("knee") point and retained for further analysis, while low-quality barcodes (i.e., relatively low numbers of reads) are located to the right and excluded from further analysis.

| Quality control, cell-type clustering, and major cell-type identification
We removed cells that had either <200 or >4000 expressed genes.
Low-quality/dying cells often exhibit extensive mitochondrial contamination. Therefore, we applied the "PercentageFeatureSet" for each cell type were identified using the "FindAllMarkers" function with an adjusted p < 0.05 and |logFC| > 1 cutoff.

| Age-related DEG analysis
DEGs for every cell type between young and old samples were identified with the "FindMarkers" function in Seurat R package using the Wilcoxon test (adjusted p < 0.05 and |logFC| > 0.25 threshold).

| Identifying statistically significant differences in cell proportions
we used the method reported by Smillie et al. (Smillie et al., 2019), to identify changes in cell proportions between young and aged NHP entorhinal cortex. We applied Dirichlet-multinomial regression model, which tests for differences in cell composition between young and aged NHP entorhinal cortex. This regression model and its associated p values were calculated using the "DirichReg" function in the DirichletReg R package.

| Cell-cell communication analysis
Cell-cell communication analysis was performed using Cell-PhoneDB (v1.1.0) (Efremova et al., 2020). Only receptors and ligands expressed in more than 10% of cells of any type from either young or old samples were further evaluated. Only those with p < 0.01 were used for cell-cell communication prediction between any two cell types.

| Gene set score analysis
Gene sets related to aging-related diseases were previously reported (Aging Atlas, 2021). Gene set scores were acquired by analyzing the transcriptome of each input cell against the aforementioned gene sets using the Seurat function "AddModuleScore." Changes in scores between young and old samples were analyzed using the ggpubr R package via the Wilcoxon test (https://github.com/kassa mbara/ ggpubr) (v0.2.4).

| Scissor analysis for each cell type
Three data sources are used for Scissor input: that is, single-cell expression matrix, bulk expression matrix, and phenotype of interest.
Given the above inputs, we used Scissor to select the phenotypeassociated cell subpopulations, which were fit by a binomial regression model (family = "binomial"). We set the parameter alpha (α) = 0.01 to choose AD-related cells.

CO N FLI C T O F I NTE R E S T
The authors declare no competing financial interest.

DATA AVA I L A B I L I T Y S TAT E M E N T
All the sequencing data are deposited in Genome Sequence Archive (GSA) (https://bigd.big.ac.cn/gsa/) with the accession number of CRA006617.