Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2022 Jun 10.
Published in final edited form as: Nat Aging. 2021 Dec 10;1(12):1107–1116. doi: 10.1038/s43587-021-00142-3

Profiling senescent cells in human brains reveals neurons with CDKN2D/p19 and tau neuropathology

Shiva Kazempour Dehkordi 1,2,#, Jamie Walker 1,#, Eric Sah 3, Emma Bennett 3, Farzaneh Atrian 2,4, Bess Frost 1,2,4, Benjamin Woost 5, Rachel E Bennett 5, Timothy C Orr 6, Yingyue Zhou 7, Prabhakar S Andhey 7, Marco Colonna 7, Peter H Sudmant 8, Peng Xu 9,10,11, Minghui Wang 9,10,11, Bin Zhang 9,10,11,12, Habil Zare 1,2,*, Miranda E Orr 3,13,*
PMCID: PMC9075501  NIHMSID: NIHMS1797052  PMID: 35531351

Abstract

Senescent cells contribute to pathology and dysfunction in animal models1. Their sparse distribution and heterogenous phenotype have presented challenges for detecting them in human tissues. We developed a senescence eigengene approach to identify these rare cells within large, diverse populations of postmortem human brain cells. Eigengenes are useful when no single gene reliably captures a phenotype, like senescence; they also help to reduce noise, which is important in large transcriptomic datasets where subtle signals from low-expressing genes can be lost. Each of our eigengenes detected ~2% senescent cells from a population of ~140,000 single nuclei derived from 76 postmortem human brains with various levels of Alzheimer’s disease (AD) pathology. More than 97% of the senescent cells were excitatory neurons and overlapped with tau-containing neurofibrillary tangles (NFTs). Cyclin dependent kinase inhibitor 2D (CDKN2D/p19) was predicted as the most significant contributor to the primary senescence eigengene. RNAscope and immunofluorescence confirmed its elevated expression in AD brain tissue whereby p19-expressing neurons had 1.8-fold larger nuclei and significantly more cells with lipofuscin than p19-negative neurons. These hallmark senescence phenotypes were further elevated in the presence of NFTs. Collectively, CDKN2D/p19-expressing neurons with NFTs represent a unique cellular population in human AD with a senescence phenotype. The eigengenes developed may be useful in future senescence profiling studies as they accurately identified senescent cells in snRNASeq datasets and predicted biomarkers for histological investigation.

Editor summary:

Using a new computational approach to identify senescent cells from single-cell transcriptomic data, the authors find that most senescent cells in the human brain are excitatory neurons with elevated CDKN2D/p19 expression, which often display Alzheimer's disease-associated tau pathology.

Cellular senescence is a complex stress response that culminates as a change in cell fate. Senescent cells are cell cycle arrested and resistant to apoptosis1. Their survival contributes to long-term health decline as they notoriously secrete a molecular milieu, referred to as senescence-associated secretory phenotype (SASP), that negatively impacts their extracellular environment2. Data from rodent models indicate that senescent neurons3-6, astrocytes7, microglia8 and oligodendrocyte precursor cells9 contribute to neurodegeneration and cognitive dysfunction. Information on the relative proportion of senescent cells in humans has been restricted to accessible tissues (i.e., ~2-5% in skin, adipose and blood10,11). Identifying senescent cells in human AD presents challenges beyond their low abundance and inability to routinely biopsy brain tissue. For example, in vitro-based molecular profiles and senescence assays generate inconsistent results when applied to the brain, p16 and/or p21 senescence marker genes signify aberrant neuronal cycle activity and are upregulated during cell differentiation and glial activation independent of senescence12-14, and brain cells secrete molecules that overlap with SASP factors in the absence of a cell cycle arrest (i.e., glial cells become hyper-proliferative and inflammatory in many neurodegenerative diseases15,16). To overcome these obstacles, we developed unbiased bioinformatic tools, senescence eigengenes, to identify and profile senescent cells in human AD.

We created three eigengenes representing distinct features of senescence: stress response, cell cycle arrest, and inflammatory response in order to minimize the likelihood for mistaking cells with one, but not all, key senescent phenotypes. Thus, senescent cells could be distinguished from those that may be cell cycle arrested, stressed, or inflammatory independent of senescence. Each eigengene included genes commonly associated with senescence that had been reported across cell and tissue types, including those from aged and transgenic mouse models of AD pathology3-9. The gene sets reflected 1) a canonical senescence phenotype (CSP) with 22 genes including CDKN2A and CDKN1A that are upregulated in many senescent cell types; 2) 48 genes upregulated early in senescence, which we termed senescence initiating pathway (SIP); and 3) 44 genes upregulated after the stable arrest and involved in SASP production, which we termed senescence response pathway (SRP) (Supplementary Table 1). For each of these three gene sets, we performed a principal component analysis to compute a weighted average expression over all genes in the corresponding list17, i.e., an eigengene18. Weights were optimized in a way that explained variance was maximized, and thus, the loss of biological information is expected to be minimal. Cells were considered to be senescent if their level of eigengene expression was more than the mean expression over all cells plus three times the standard deviation (mean+3sd).

Two independent single nucleus RNA sequencing (snRNA-Seq) datasets were used, Mathys et. al, 201919 and Zhou et. al, 202020, referred to as cohort 1 and cohort 2, respectively. Analyses using the CSP eigengene revealed 1,526 senescent cells in the dorsolateral prefrontal cortex (2.1%) from cohort 1; the proportion differed across individuals (0-13%, Figure 1a-g). A total of 1,351 cells and 1,256 cells expressed the SIP and SRP, respectively (2% and 1.7%, Figure 1a-c). Similar results were derived from the cohort 2 whereby 1,331 (2.3%), 1,485 (2.6%), and 951 (1.6%) cells expressed the CSP, SIP, and SRP eigengenes, respectively (Supplementary Figure 1 and Supplementary Table 2). We used a dataset generated from embryonic brains as a control21; notably, the eigengenes identified few senescent cells (CSP: 16 (0.4%); SIP: 23 (0.6%); SRP: 82 (2%), Extended Data Figure 1).

Figure 1 ∣. The prominent senescent cell type in the DL-PFC were excitatory neurons.

Figure 1 ∣

Open in a new tab

Eigengenes for each gene list (a, d, h) using n=70,634 cells in canonical senescence pathway (CSP); (b, e, i) senescence initiating pathway (SIP); and (c, f, j) senescence response pathway (SRP) were computed using principal component analyses. (a-c) The proportion of cells from each brain expressing the respective eigengene were plotted. (d-f) Cell types and (g) counts represented in the senescent cell population discovered in a-c. A one-sided hypergeometric test was used in order to report the significant cell types. All the p-values are adjusted using Bonferroni correction. (h-j) The ratio of senescent excitatory neurons that expressed the respective eigengenes to total neurons within each brain, n=48 brains. (k) Scatter plot for the ratio of senescent excitatory neurons to the total number of excitatory neurons in cohort 1 with n=48 brains. Each dot represents one brain. The size of the dots depicts the ratio in SRP. The senescence excitatory ratios of CSP highly correlated with SIP (Pearson correlation: 0.96) and SRP (0.90). Also, the SIP ratio was positively correlated with the SRP ratio (0.93). The line inside each box plot in a-c and h-j shows the median. The lower and upper hinges of box plots correspond to the first and the third quartiles, respectively. The whiskers extend from the bottom or the top of the box for at most 1.5 of the interquartile range (IQR), which is the distance between first and third quartiles. Samples not between the whiskers were considered outliers, which are shown with yellow (a-c) and black (h-j) dots. Cell populations: astrocytes [Ast], endothelial cells [End], excitatory neurons [Ex], inhibitory neuron [In], microglia [Mic], oligodendrocytes [Oli], oligodendrocyte precursor cells [Opc], and pericytes [Per]) were classified in the original publication21.

To determine which brain cell type(s) were overrepresented in the senescent cell population, we used a hypergeometric test to categorize the overlap with cell types, as defined in the original studies19,20. In cohort 1, excitatory neurons were the only cell population with more than expected senescent cells based on all three senescence gene sets (Figure 1d-g). Subpopulations of astrocytes, endothelial cells and pericytes expressed gene patterns consistent with inflammation (SRP), but not with canonical senescence. Cohort 2 produced similar results where the number of cells expressing CSP and SIP eigengenes were overrepresented in excitatory neurons. Astrocytes and endothelial cells were identified to express the SRP eigengene, which may reflect an inflammatory phenotype independent of a canonical senescence stress response22,23 (Extended Data Figure 2). Endothelial cells were also identified based on the CSP eigengene, but only in cohort 2. These data may indicate vascular cell senescence in the brain, as seen in cardiovascular disease24, and recently reported in human AD22. Similarly, in the embryonic dataset, the higher expression of SRP reflected endothelial cells, 77 (49%). However, they did not express either the CSP or SIP eigengenes indicating an incomplete senescence profile that may reflect physiological senescence or developmental processes associated with these molecules25,26. Excluding endothelial cells, the number of identified senescent cells in the embryonic brains was significantly fewer than the ~2% rate in cohorts 1 and 2 (−log10 p-values from binomial tests: 16 for CSP, 12 for SIP, and 26 for SRP). Collectively, the predominant senescent cell population in both adult brain cohorts was excitatory neurons representing 97% and 92% of CSP cells in cohort 1 and cohort 2, respectively. Even in the Grubman et al. dataset27, in which neurons were undersampled due to use of single cell profiling (i.e., only 656 (5%) neurons were sequenced), neurons had the highest rate, and the most significant p-value, of CSP senescent enrichment (Supplementary Figure 2).

The relative proportion of senescent excitatory neurons to total excitatory neurons within individual brains varied among individuals and ranged from 0% to 20% . On average, 4.2%, 3.5% and 2.5% excitatory neurons in cohort 1 were senescent as determined using the CSP, SIP and SRP eigengenes, respectively (−log10 p-value: 3,143, 2,027, and 380, respectively, Figure 1h-j); Cohort 2, Supplementary Figure 1. Moreover, the senescent cells identified based on these three eigengenes overlapped significantly (−log10 p-value: >232, Extended Data Figure 3). There was no significant association between the number of excitatory neurons expressing CSP, SIP, and SRP eigengenes with sex (p-values: >0.7, >0.7, >0.6, respectively) or age (p-values: >0.5, >0.2, >0.3, respectively). In relation to cohort 1, brains from cohort 2 had similar senescent cell burden and between subject variability (Supplementary Figure 1) but a higher proportion of excitatory neurons expressing CSP and SIP (9.7% and 10.3%, respectively) with < 1% expressing SRP (− log10 p-values: 4,820, 4,663, and 0, respectively; Extended Data Figure 2 and Supplementary Figure 3).

As described above, the three CSP, SIP, and SRP eigengenes were designed to capture distinct aspects of senescence that we reasoned would be associated in senescent cells, but not necessarily associated in non-senescent cells. To test this hypothesis, we evaluated the correlation among the three senescence eigengenes within excitatory neurons. The ratio of senescent excitatory neurons, as defined by CSP, highly correlated with SIP (Pearson correlation: 0.96) and SRP (0.90). Moreover, brains with higher proportions of senescent neurons displayed higher eigengene expression levels (Figure 1k). Pathways were similarly correlated in cohort 2 (CSP and SIP: 0.94, Supplementary Figure 3). Thus the eigengenes were significantly associated, and excitatory neurons represented the predominant senescent cell type across the ~140,000 cells analyzed in these 76 brains as determined by three independent eigengenes (i.e., signatures) of senescence.

To confirm that these results were not an artifact of our gene lists, we computed an eigengene for senescence per each of the gene lists obtained from CellAge28, Gene Ontology (GO)29,30, and Kyoto Encyclopedia of Genes and Genomes (KEGG)31 databases (Supplementary Table 1). These eigengenes showed that 3.5%, 3.7% and 3.4% excitatory neurons in cohort 1 were senescent, respectively, which is significantly more than expected by chance (−log10 p-value: 2,534, 2,673, and 2,277, respectively, Extended Data Figure 4). Overall, the overlap between the identified senescent cells based on the six CSP, SIP, SRP, CellAge, GO, and KEGG eigengenes was significantly more than expected by chance (−log10 p-value >232, Extended Data Figure 5).

The accumulation of intraneuronal tau protein is a common pathology across neurodegenerative diseases including AD32. Neurofibrillary tangles (NFTs) are characterized histologically by the presence of aggregated, phosphorylated misfolded tau proteins. They accumulate preferentially in excitatory neurons in human AD33 and drive senescence in transgenic mice3,8. We hypothesized that excitatory neurons containing NFTs may be a source of senescent excitatory neurons detected by the eigengenes. To quantify the association between NFTs and senescence, we created eigengenes from two independent gene lists derived from laser capture microdissected neurons with NFTs34,35, referred to as “NFTDuckley” and “NFTGarcia”, respectively. Using the NFT eigengenes, we identified 1,050 NFTDunckley cells and 1,022 NFTGarcia cells in cohort 1 reflecting 1.5% and 1.4% of the total cellular population, respectively (Figure 2). The overlap between NFTDunckley cells and NFTGarcia cells was significant (i.e., 765 (1%) cells, −log10 p-value: 3,035). Cohort 2 had slightly higher levels of NFT-bearing neurons than cohort 1, specifically 1,523 (2.6%) NFTDunckley and 1,761 (3%) NFTGarcia cells with a significant overlap of 1,214 (2%) cells (−log10 p-value: 4,038, Supplementary Figure 4). As expected, we did not identify any NFTDunckley or NFTGarcia cells in the embryonic control brains (Supplementary Table 2). In both cohorts 1 and 2, and based on both NFTDunckley and NFTGarcia eigengenes, the predominant cell type expressing the NFT eigengenes were excitatory neurons. These data are consistent with NFTs driving neuronal senescence in transgenic mice3 and preferential accumulation in excitatory neurons in AD33.

Figure 2 ∣. Neurofibrillary tangle eigengene expression significantly correlated with senescence expression.

Figure 2 ∣

Open in a new tab

(a) Eigengenes representing neurofibrillary tangle (NFT) expression were calculated from two separate datasets, Dunckley34 and (b) Garcia35, respectively. Cell types (a, b) and counts (c) expressing each NFT eigengene were calculated and plotted using n=70,634 cells. A one-sided hypergeometric test was used in order to report the significant cell types. All the p-values are adjusted using Bonferroni correction. (d-e) The ratio of NFT-containing excitatory neurons to total neurons expressing each respective eigengene within each brain, n=48 brains. The line inside each box plot in d-e shows the median. The lower and upper hinges of box plots correspond to the first and the third quartiles, respectively. The whiskers extend from the bottom or the top of the box for at most 1.5 of the IQR. Any sample not between the whiskers is known as an outlier and is shown with a black dot. (f) Scatter plot for eigengene values for CSP genes on x axis versus Dunckley NFT marker genes on y axis. Each dot represents one neuron. Red line represents intercept, and the blue line shows the best linear fit. A linear regression model is fitted on NFT ~ Senescence with coefficient being equal to zero as the null hypothesis.

As observed with senescence, the relative proportion of NFT-containing neurons to all excitatory neurons varied across individuals in cohort 1 (Figure 2d-e) and cohort 2 (Supplementary Figure 5). Since the cell type (i.e., excitatory neurons) and proportions were consistent with other reports33,36, we interpreted that the eigengene could be used as a transcriptional surrogate to identify NFT-bearing neurons. Plotting NFT eigengene expression against senescence eigengene expression revealed a significant relationship between them (−log10 p-value >15, adjusted R2 = 0.6803, m = 0.863, Figure 2f). Hypergeometric tests were used to assess the association between senescent and NFT-bearing cells. In cohort 1, we identified 1,485 CSP-senescent excitatory neurons and 1,032 NFTDunckley-bearing excitatory neurons. We expected to identify 44 overlapping cells if senescence and NFTs were not associated (i.e., co-expression by chance); however, 598 cells co-expressed both eigengenes indicating a significant association (−log10 adjusted p-value = 1,337, Supplementary Table 3). Similar results were obtained in cohort 2 and using the NFTGarcia eigengene (Supplementary Table 3). NFTDunckley-bearing excitatory neurons also significantly overlapped with SIP and SRP senescent cells (−log10 p-values >231, Extended Data Figure 3). These data confirmed the significant association whereby the senescence and NFT eigengenes were upregulated within the same cells.

Senescent and NFT neurons constituted a minor proportion of all neurons (i.e., of the 44,172 total neurons analyzed in cohort 1, only 3.4% and 2.3% excitatory neurons met the criteria of senescent and NFTs, respectively). To visualize overlap between senescent and NFT neuron populations, we plotted their distributions within the entire neuronal population. These data revealed a continuum of senescence gene expression whereby 99% of NFT-bearing neurons displayed upregulated senescence eigengene greater than that of neurons without NFTs (Figure 3). Specifically, the density plots indicated that <1% of NFT-bearing neurons expressed the CSP eigengene lower than the mean (i.e., < 1% of neurons with NFTs could be confidently considered not senescent, Figure 3a). However, we are cautious not to label the remaining 99% of NFT-bearing neurons as senescent. Our stringent cutoff required expression levels >mean+3sd. With these criteria, 35% of NFT-bearing neurons were identified as senescent and <1% as not senescent. The remaining 64% of NFT-bearing neurons could not be considered either senescent or not-senescent, but instead with upregulated senescence eigengene expression. Similar patterns were observed across all eigengenes to confirm and validate the interpretation (Figure 3a-c). We also determined the distribution of the senescent neurons identified to contain NFTs (i.e., used the same experimental approach, but asked the question in the opposite direction). In cohort 1, all senescent cells expressed the NFT eigengene greater than the mean of all neurons (Figure 3d-f). Approximately 14% of the senescent neurons co-expressed the NFT eigengene >mean+3sd. The remaining 86% of cells had upregulated the NFT eigengene (4%, 34% and 48% mean+1sd, 2sd, 3sd, respectively), but did not reach >mean+3sd criteria. Overall, our data indicate significant overlap in cells co-expressing the senescence and NFT eigengenes; all neurons that expressed the NFT eigengene co-expressed the senescence eigengene greater than those without NFTs. Moreover, 99% of neurons that expressed the senescence eigengene co-expressed the NFT eigengene greater than those that were not senescent. Thus, senescent and NFT-bearing neurons overlap.

Figure 3 ∣. Senescent excitatory neurons contain NFTs and NFT-bearing neurons are senescent.

Figure 3 ∣

Open in a new tab

(a-c) Plots of total neuron counts, pink, against expression of the eigengene CSP, (b) SIP or (c) SRP. Cell densities of where the NFT-bearing neurons (green) lie within the plot (inset). (d-f) Plots of total neuron counts, pink, against expression of the NFTDunckley eigengene. Cell densities of where the CSP, (d) SRP or (e) SIP cell populations lie within the plot (inset). Larger plots are scaled by the number of cells and insets are scaled by cell density. Mean and standard deviation (sd) are calculated for the eigengene value of all neurons.

The association between senescence and stage of AD was also evaluated. Across eigengenes, the number of identified senescent cells per brain and AD stages were not significantly associated except for astrocytes, and only with the SRP eigengene. The number of astrocytes that expressed the SRP eigengene (i.e., inflammation/SASP) was higher in the 24 individuals that Mathys et al. reported to have early or late AD pathology compared to the 24 individuals without pathology (i.e., 107 vs. 61 cells, p-value adjusted for testing multiple cell types <0.003). Astrocytes expressing SRP were also more abundant in the 24 individuals in Braak stage IV to VI compared to the 24 individuals Braak stage I to III (i.e., 60 vs. 108 cells, adjusted p-value <0.02). Since the number of astrocytes expressing the CSP or SIP were similar between groups, we interpret that these cells were not fully senescent. Instead, they may have been in transition to becoming senescent or displayed a pro-inflammatory stress response not linked to senescence.

To gain insight into the mechanistic regulators of the senescence phenotype, we determined the weight that each gene contributed to their respective senescence eigengenes (CSP: Figure 4a; SIP and SRP: Supplementary Figure 6; and Supplementary Table 1). Cyclin dependent kinase inhibitor 2D, CDKN2D, contributed most to the CSP eigengene (Figure 4a). The average value of normalized CDKN2D expression in the three cell clusters that were enriched in senescent neurons (Supplementary Figure 7) was 0.1, which was 1.4-fold higher than the excitatory neurons in other clusters (−log10 p-value from the Wilcoxon rank sum test >15). Profiling single cells based on only elevated CDKN2D expression resulted in a similar cellular composition as with the senescence eigengenes (i.e., predominantly excitatory neurons, Extended Data Figure 6). However, the number of senescent excitatory neurons was overestimated in this analysis (CDKN2D: 2,731 versus CSP eigengene: 1,485) to indicate that not all cells with elevated CDKN2D could be considered senescent.

Figure 4 ∣. Upregulated CDKN2D and p19 deposition co-occur with tau neuropathology and morphological characteristics of cellular senescence in human Alzheimer’s disease.

Figure 4 ∣

Open in a new tab

(a) Weight of each gene in the canonical senescence pathway (CSP) eigengene based on principal component analysis; CDKN2D had the highest weight. (b) RNAscope probe CDKN2D in control and (c) AD brains was performed on n=3 control and n=3 AD cases. Scale bar 50 μm. (d-g) RNAscope co-labeled (d) nuclei, (e) CDKN2D, (f) AT8 (phosphorylated tau, NFTs) and (g) color merged images. Scale bar 10 μm. (h-i) Representative images of frontal cortex in control (h, n=2) and Alzheimer's disease neuropathologic change (ADNC, n=9) cases (i-k) stained with (h-k) AT8 and adjacent section stained with (l-o) anti-p19 antibody (i.e. corresponding AT8 stains are directly above the p19 stains). Scale bar: 80 μm. (p) Co-immunofluorescence (IF) staining with AT8, (q) nuclear membrane, lamin B1, (r) p19 and (s) Hoechst nuclear stain. (t) Overlap of all channels. (u) Color-inverted Hoechst nuclear image for purposes of better visualizing nuclear morphology. Open black arrow: nuclei without p19 or NFTs; open cyan arrow: nuclei with p19 and closed arrowhead: nuclei with p19 with NFTs. Scale bar 10 μm. (v) Quantification of cell nucleus area across cells without p19 or NFT staining (control) or expressing p19 with or without NFTs (as indicated in panel u). Characteristic lipofuscin autofluorescence (white arrowheads in p and t) was also quantified. Data presented as mean ± standard error. One way ANOVA with Tukeys multiple comparisons test. **: p=0.0031, ****p<0.0001. n (number of cells): Control (p19 negative, AT8 negative): 101, p19 only: 44, p19+NFTs: 164.

RNAscope indicated elevated CDKN2D expression in AD cases compared to controls (Figure 4b-c, Extended Data Figure 7), localization in neurons (Extended Data Figure 8) and overlap with NFTs (Figure 4d-g and Supplementary Figure 12). Immunohistochemistry confirmed expression of the CDKN2D protein product, p19, in postmortem human AD with NFTs (Figure 4h-o). The proportion of p19-positive cells ranged from 4% to 22% (Supplementary Table 4), consistent with estimates of the eigengene (Figure 1). Staining revealed p19 expression in the nucleus, neuropil and in neuritic plaques of some AD cases (Figure 4l-o). Co-immunofluorescence with Map2 (i.e., neurons, Supplementary Figure 9) or AT8 (i.e., NFTs, Figure 4t) with p19 indicated 30% of total neurons and 72% of those with NFTs neurons co-expressed p19. Karyomegaly3,37 and lipofuscin38 represent characteristic senescence morphologies and lysosomal degradation byproducts, respectively. They are not linked to specific marker genes, thus allow for an unbiased validation of senescence. These analyses indicated that p19-expressing neurons had 1.8-fold larger nuclei than p19-negative neurons, and nuclei size was further increased by the presence of NFTs (p < 0.0001 and p = 0.0031, respectively; Figure 4v). Similarly, twice as many p19-expressing neurons contained lipofuscin deposits than p19-negative neurons; this proportion increased to 98% in the presence of NFTs (Figure 4v, Supplementary Figure 13). Co-staining with anti-lamin b1 to visualize the nuclear membrane further highlighted aberrant nuclear morphology in p19-positive cells (Figure 4q, s-u; Supplementary Figure 14), consistent with previous reports in AD39, senescence40 and our eigengene prediction (Supplementary Figure 6b).

In summary, we developed, tested, and validated the eigengene approach to identify senescent cells in transcriptomic datasets. Data from five separate human brain cohorts, six senescence transcriptomic profiles, two neuropathology profiles, and RNA and protein histological methods indicated that excitatory neurons with upregulated CDKN2D/p19 and NFT neuropathology represent a unique cell population in human AD brains with morphological features consistent with senescence. Future studies are needed to validate and identify additional biomarkers, and to elucidate the interaction between p19 and tau pathology in senescence. While our data provide early insight into the heterogeneity of senescence marker genes and cell types in the brain, they also caution against the reliance of SASP, alone, for identifying senescence in the brain. Specifically, astrocytes and endothelial cells upregulated SASP genes in the absence of other senescence hallmarks. Overall the findings emphasize the utility and importance of applying multi-analyte approaches when studying complex, dynamic cellular stress responses. Employing these validated eigengenes in future studies may help guide senescence profiling across human tissues.

Methods

Ethics Oversight.

Our research complies with all relevant ethical regulations. Postmortem tissues used for immunohistochemistry and immunofluorescence were provided by the Biggs Institute Brain Bank which collected donor tissue in accordance with the UTHSA Institutional Review Board. Postmortem tissue used for RNAScope was obtained from the Massachusetts Alzheimer’s Disease Research Center, which collected donor tissue in accordance with the MGH Institutional Review Board.

Single nucleus and single cell RNA-Seq datasets.

In this study, we refer to two snRNA-Seq datasets generated by Mathys et. al, 201919 and Zhou et. al, 202020 as cohort 1 and cohort 2, respectively. Datasets were accessed through Accelerating Medicines Partnership - AD (AMP-AD41) with Synapse IDs syn18485175 and syn21126462, respectively. The data included ~80,00019 and ~70,00020 single nuclei derived from the dorsolateral prefrontal cortex of 48 and 32 postmortem human brain samples, respectively. These data were provided by the longitudinal cohort studies of aging and dementia: The Religious Order Study (ROS) and the Rush Memory and Aging Project (MAP)42. Inclusion criteria were the same as in the data generating studies. Specifically, in cohort 1, 24 control individuals were selected with no or very little pathology in addition to 24 age-matched individuals with a spectrum of mild to severe β-amyloid and other pathologies19. The mean and median of age were 86 and 87 years old, respectively, with a standard deviation (sd) of 5 years. In cohort 2, 11 people with AD carrying TREM2-CV, 10 people with AD carrying TREM2-R62H, and 11 age-matched controls were included20. The mean and median of age were 89 years old with a sd of 6 years. Four subjects (10248033, 20207013, 10290265, and 11072071) were represented in both datasets. We included them only in cohort 1 but not cohort 2; in total ~140,000 cells were analyzed from 76 brains (Supplementary Table 2). Embryonic brains are not expected to have a significant senescence burden, and thus, can be appropriate controls for our study. We used the single cell (sc) RNA-Seq data of ~4,000 cells that Fan et. al, 2018 generated from cerebral cortex of two female twin embryos of age 22 and 23 weeks21. Neurons are often excluded from scRNA-Seq datasets due to their large size; similarly senescent cells undergo excess growth3,37 and may also be excluded from scRNA-Seq profiling. Nevertheless, to assess the extensibility of our approach, we analyzed the scRNA-Seq data of ~13,000 cells that Grubman et al., 2019 generated from the entorhinal cortex of 12 individuals27 (Supplementary Figures 2, 10, and 11).

Eigengene analysis.

We downloaded single nucleus (sn) RNA-Seq data from Mathys et. al, 201919 and Zhou et. al, 202020 studies, which were available from the AMP-AD41 website, using the synapser (https://r-docs.synapse.org/articles/synapser.html) R package43 (Version 0.6.61) and custom R scripts44 (Version 3.6.1). We downloaded clinical data from the corresponding publication pages. For each of the three gene sets in the Supplementary Table 13,34, we used the compute.pigengene() function from the Pigengene package (Version 1.13.4) to compute an eigengene18, which is a weighted average expression over all genes in the corresponding list17. Following our previous approach on computing eigengenes18,45-47 we balanced the number of cells in each cell type using oversampling, so that all cell types had comparable representatives in the analysis. Specifically, we repeated the data of each astrocyte 104, endothelial cell 2,919, excitatory neuron 10, inhibitory neuron 38, microglia 184, oligodendrocyte 19, oligodendrocyte precursor cell 134, and pericytes 2115 times, and obtained 352,768; 353,199; 349,760; 349,448; 353,280; 346,465; 352,018 and 353,205 samples from each cell type, respectively. Weights were optimized using a principal component analysis. We computed the mean expression of each eigengene over all analyzed cells. Cells were considered senescent if their level of eigengene expression was more than the mean expression over all cells plus three times the standard deviation. Hypergeometric tests were used to identify the cell types in which senescent cells were overrepresented. We used the project.eigen() function from the Pigengene package to infer the eigengenes values in the validation datasets based on the same weights that we had obtained from our analysis on cohort 1 as the training dataset. In order to see how much senescence and NFT expressing eigengenes overlap, we visualized their expression in density plots using the ggplot2 package. A kernel density estimate was used to represent the probability density function of eigengene values (Figure 3). We tested the significance of overlap between NFT cells and senescent cells with hypergeometric tests using the phyper function in R. We set the log.p parameter to TRUE to increase the reporting accuracy. Throughout this paper, we replaced the log10 of any p-value between 0.1 and 1 with ~0.

Cell clustering.

We applied the Seurat pipeline48 to cluster the cells in cohort 1 based on their gene expression profile in an unbiased way, agnostic to senescence markers. First, we performed quality control and removed cells with less than 200, or more than 6,000, detected genes. We also removed genes expressed in less than three cells. Then, we normalized the raw counts using the sctransform method in Seurat, which applied a regularized negative binomial regression modeling approach for the normalization and variance stabilization of molecular counts49. We used the RunPCA() function to identify the top 3,000 most variable genes, the FindNeighbors() function to construct a K-nearest neighbor (KNN) graph based on the Euclidean distance calculated from the top 30 principal components50, and the FindClusters() function to identify cell clusters based on the Louvain algorithm51. We set the resolution parameter for the clustering granularity as 0.8. To visualize the cell clusters, we used the uniform manifold approximation and projection (UMAP) method for non-linear dimensional reduction52. For each cell cluster, we performed Fisher's exact test to assess its enrichment in the cells that we had previously identified senescent based on our eigengene approach.

Immunohistochemical (IHC) staining.

IHC stains were performed using a Thermo Scientific™ Lab Vision™ Autostainer 480 following deparaffinization of formalin-fixed paraffin-embedded sections (FFPE) and 30 minutes of heat-induced antigen retrieval in citrate buffer. Endogenous peroxidase was blocked by immersion in 3% hydrogen peroxide for 10 minutes and rinsed. A protein block for 15 minutes with 2.5% normal goat serum (Sigma) was then performed. After rinsing, sections were incubated with either mouse anti-human monoclonal AT8 antibody (Thermo Scientific) at 1:2,000 or rabbit polyclonal anti-p19 antibody (Abcam) at 1:100 for 45 minutes, washed and incubated with undiluted secondary antibody (goat anti-mouse or goat anti-rabbit, respectively, IgG (HRP), VisUCyte) for 45 minutes followed by rinsing. Diaminobenzidine (DAB) chromagen (BD Pharmigen) was used to visualize the immunoreactivity. IHC staining for p19 was performed on FFPE sections of the middle frontal gyrus from 6 Alzheimer disease (AD) cases, all of which demonstrated a high level of Alzheimer disease neuropathologic change (ADNC) with a Braak stage of VI, 3 intermediate ADNC level cases and 2 primary age-related tauopathy (PART) control cases (Braak stages I-II) with no neocortical neurofibrillary tangles.

Immunofluorescence for co-labeling.

Dual Labeling IF: Human brain sections were deparaffinized via xylene and hydrated in a series of graded alcohol. Heat-induced antigen retrieval performed using a pressure cooker and 10 mM sodium citrate (pH 6.6) with 0.2% tween. Sections were then exposed to LED light overnight in 10 mM sodium citrate (pH 6.6) with 0.2% tween and 0.05% sodium azide in 4°C and blocked using 0.25% Bovine Serum Albumin (BSA) and PBS + 0.2% Triton-X 100 (PBSTr) for 2hr at room temperature. Primary antibodies were added at the following concentrations, p19 (1:100, ab26287, Abcam, Cambridge, MA, USA) and MAP2 (1:200, PA5-17646, Invitrogen, Carlsbad, CA) and incubated overnight at 4°C. Sections were washed three times in PBSTr each for 10 minutes at room temperature. Alexa Fluor-conjugated secondary antibodies (Thermo Fisher Scientific, Waltham, MA, USA) diluted in 0.25% BSA and PBSTr (1:200) were then added and the sections incubated for 2 hr at room temperature. Sections washed three times with PBSTr and incubated with 0.3% Sudan Black in 70% ethanol for 10 minutes and washed ten times with PBS at room temperature. Slides were mounted using DAPI Fluromount-G®(0100-20, SouthernBiotech, Birmingham, AL). For AT8 and p19 co-staining: Phospho-Tau (Ser202,Thr205) (ThermoFisher Scientific, MN1020, conjugated with AF594 using Alexa Fluor 594 Antibody Labeling Kit A20185) and P19 INK4d (ab262871, indirect labeling with Alexa Fluor 647 Tyramide Reagent B40958). Syto83 was used as a nuclear counterstain as described previously53. Images were acquired using an Olympus FluoView FV1200 confocal laser scanning microscope. Quad labeling p19, lamin B1 AT8 and Hoechst: Following the p19 Tyramide boost step (anti-CDKN2D Sigma cat no. HPA043546; same clone as ab262871), slides were microwave treated in citrate buffer until boiling (100% power for 127sec) followed by 20% power for 15 minutes, then cooled for 25 minutes at room temperature. Slides were placed in the humidity chamber, blocked with 10% normal goat serum for one hour, then incubated in primary antibodies: mouse monoclonal lamin b1, clone 10H34L18, 1:500 (ThermoFisher Scientific, cat no. 702972) and AT8 1:1000 (ThermoFisher Scientific, cat no. MN1020) in CST antibody diluent overnight (18 hours) at 4C. The following day, slides were washed in TBS-T and incubated with secondary antibodies (IgG (H+L) Highly Cross-Absorbed Goat anti-Rabbit, Alexa Fluor Plus 647 (cat no. A32733) and IgG (H+L) Highly Cross-Absorbed Goat anti-Mouse, Alexa Fluor Plus 555 (cat no. A32727)) diluted 1:1500 in TBS-T for 2 hours at room temperature. Slides were rinsed 2x TBST and incubated in Hoechst 33342 (ThermoFisher Scientific, EdU Imaging Kit Component G, cat no. C10337) 1:2000 in TBS-T for 30 minutes at room temperature. Slides were rinsed 2x TBST and cover slipped using ProLong Gold Antifade Reagent (ThermoFisher Scientific, cat no. P36934), sealed with nail polish and imaged with an Olympus FluoView FV1200 confocal laser scanning microscope.

RNAscope:

Fresh frozen temporal cortex (Brodmann area 20) was obtained from the Massachusetts Alzheimer’s Disease Research Center, which collected donor tissue in accordance with the MGH Institutional Review Board. AD subjects (n=3) met clinical and neuropathologic criteria for Alzheimer’s disease and control subjects (n=3) did not have clinical or neuropathologic evidence of neurodegenerative disease (Supplementary Table 5). Cryostat sections were sliced at 8 microns and placed on SuperFrost plus slides. Sections were fixed for 15 minutes in chilled 4% paraformaldehyde in PBS and then rinsed in PBS and dehydrated in an ascending ethanol series (50%, 70%, 100% x2 for 5 minutes each). RNAscope was then performed using the RNAScope® Multiplex Fluorescent V2 Assay (ACD Bio) and HybEZ™ II Hybridization System (ACD Bio) according to the manufacturer’s recommendations with the following modification: treatment with protease IV (ACD Bio) was carried out for 5 minutes at room temperature. A custom 18 ZZ probe was designed to target 315-1378 bp region of CDKN2d (GenBank Accession #: NM_001800.4; cat no. 1098101-c1; ACD Bio). Following in situ hybridization, sections were blocked in 5% bovine serum albumin (Sigma Aldrich) for one hour and then mouse anti-HuD (E-1; cat no. sc-28299; Santa Cruz Biotechnology) was applied overnight. The following day, donkey anti-mouse Alexa750 (cat no. ab175738; Abcam) was applied and then mouse anti-tau-biotin conjugated (AT8; cat no. MN1020b; ThermoFisher Scientific) was applied and incubated overnight. AT8 was detected with streptavidin-Alexa555. All antibodies were applied at a 1:500 concentration. Sections were then coverslipped with Fluoromount G with DAPI (Southern Biotech) and sealed with nail polish. Images were acquired using an Olympus Confocal FV3000 and an Olympus VS120 slide scanner.

Image quantification:

IHC color images were manually scored by a technician blinded to cases. Total cells and p19 positive cells were counted on a minimum of 2 representative images. The images were pre-selected by a neuropathologist to ensure similar brain regions were represented across cases. To determine co-localization of p19, confocal images were analyzed by a technician blinded to cases. Cell nuclei and lipofuscin measures were quantified from co-immunofluorescent images using Adobe Photoshop 22.3.0. Neurons were chosen at random and categorized by the presence or absence of p19 and AT8 staining. Each neuron’s nuclei and lipofuscin area were measured using the Histogram tool. Cells for lamin B1 quantification were selected from confocal z-stack images opened in ImageJ. To ensure cells were profiled from a mid-cell plane (i.e., not on edges) morphology parameters were applied (i.e., size: 85um2 – infinity, with gray scale thresholding: 475-65535, and default circularity: 0.00-1.00). The ROI Manager was used to acquire data from individual cells. The corrected total cell fluorescence (CTCF) was calculated (CTCF = Integrated Density – Area of selected cell), plotted and analyzed with the aid of GraphPad Prism Software Version 9.1.0.

Statistics and Reproducibility.

No statistical method was used to predetermine sample size. No data were excluded from the analyses. Exact n’s are reported in respective figure legends. Reproducibility measures included analyses of initial and validation datasets (i.e., senescence: Cohort 1, Cohort 2, embryo control and NFTs: Dunckley and Garcia); hypothesis testing by applying multiple, distinct bioinformatic approaches on each dataset; evaluating distinct datasets generated by independent laboratories; comparing results between distinct transcriptomic technologies (i.e., snRNASeq and scRNASeq); confirming senescence eigengene results from multiple gene lists generated by our group (i.e., CSP, SIP, SRP) and those derived from publicly available senescence gene lists (i.e., Kegg, GO, CellAge); reproducing histology experiments using multiple biological replicates (i.e., postmortem human brains n=5 control and n=12 AD) derived from separate brain biorepositories (e.g., Biggs Institute Brain Bank and Massachusetts Alzheimer’s Disease Research Center), across four laboratories (e.g. Walker, Frost, Bennett and Orr) located in three separate institutions (e.g., UTHealth San Antonio, Massachusetts General Hospital and Wake Forest School of Medicine); applying multiple, complementary, histological techniques to confirm bioinformatic findings (i.e., RNAscope, immunohistochemistry and immunofluorescence). The investigators that analyzed/quantified the tissues were blinded to disease diagnosis (i.e., AD or control).

Extended Data

Extended Data Fig. 1. Prominent senescent cell types in prefrontal cortex of the embryonic control.

Extended Data Fig. 1

Open in a new tab

Cell types and counts represented in the senescent cell population discovered in (A) CSP, (B) SIP and (C) SRP. The cutoff and statistical test definitions are the same as in Figure 1. Cell populations: astrocytes [Ast], blood cells [Blood], Cajal-Retzius cells [Cajal], endothelial cells [Endo], excitatory neurons [Ext], immune cells [Immune], inhibitory neuron [Inh], microglia [Micro], neural stem cells [NSC], and oligodendrocyte precursor cells [Oligo] were classified in the original publication.

Extended Data Fig. 2. Prominent senescent cell types in the dorsal lateral prefrontal cortex in Cohort 2.

Extended Data Fig. 2

Open in a new tab

Cell types and counts represented in the senescent cell population discovered in (A) CSP, (B) SIP and (C) SRP with n=57,857. The cutoff, statistical test and abbreviations definitions are the same as in Figure 1.

Extended Data Fig. 3. Overlap between senescent and NFT neurons.

Extended Data Fig. 3

Open in a new tab

Each vertical bar represents the number of neurons in Cohort 1 that express the eigengenes marked by green circles below the bar. Each row at the bottom corresponds to an eigengene, and the number of neurons expressing that eigengene is shown in the right end on each row. The probability distributions of multi-set intersections have been calculated and the significance was tested using a hypergeometric test. The scale bar at top right shows the level of significance for each intersection. The largest p-value is −232 in log10 scale, which corresponds to the intersection between SRP and CSP expressing cells.

Extended Data Fig. 4. Prominent senescent cell types using CellAge, GO and KEGG gene lists in Cohort 1.

Extended Data Fig. 4

Open in a new tab

Cell types and counts represented in the senescent cell population discovered in (A) CellAge, (B) GO and (C) KEGG. The cutoff, statistical test and abbreviations definitions are the same as in Figure 1.

Extended Data Fig. 5. Overlap between senescent cell populations.

Extended Data Fig. 5

Open in a new tab

Each vertical bar represents the number of senescent cells in Cohort 1 that express the senescence eigengenes, marked by green circles below the bar. Each row at the bottom corresponds to a senescence eigengene, and the number of senescent cells expressing that eigengene is shown at the end of each row. The probability distributions of multi-set intersections have been calculated and the significance was tested using a hypergeometric test. The scale bar at top right shows the level of significance for each intersection. The largest p-value is-260 in log10 scale corresponding to the intersection of SRP and CSP.

Extended Data Fig. 6. Excitatory neurons are the prominent senescent cell types based on CDKN2D in (A) Cohort 1 and (B) Cohort 2.

Extended Data Fig. 6

Open in a new tab

Cell types and counts represented in the senescent cell population using only CDKN2D. The cutoff, statistical tests and abbreviations definitions are the same as in Figure 1.

Extended Data Fig. 7. RNAscope reveals higher CDKN2D expression in postmortem brains from cases with AD than age-matched control brains.

Extended Data Fig. 7

Open in a new tab

A. CDKN2D negative and positive control probe signal. B. CDKN2D RNAscope on three separate AD cases (n=3) compared to a representative age-matched non-demented control (n=3) (refer to Supplementary Table 5 for case characteristics. Scale bar 50 μm.

Extended Data Fig. 8. CDKN2D RNAscope colocalized with neuronal marker, HuD.

Extended Data Fig. 8

Open in a new tab

Postmortem AD tissue was processed for RNAscope with CDKN2D (green) and co-labeled for total nuclei (DAPI, gray) and neurons (HuD, cyan)/ Merged image display strong overlap between CDKN2D and neurons, but not other cell types (i.e., blue and green co-localization with infrequent green co-localization in nuclei without HuD staining). Scale bar 10 μm. Representative images from postmortem human brains (n=3 control and n=3 AD cases).

Supplementary Material

1797052_Sup_Info
1797052_Sup_code
1797052_Sup_Tables

Supplementary Table 1. This file contains Supplemental Tables 1-3 combined into a single excel workbook file with multiple Tabs. Supplemental Table 1 contains the gene lists used to create each of the eigengene; Supplemental Table 2 (multiple tabs) contains eigengene expression data from every cell analyzed across datasets. Supplemental Table 3 (multiple tabs) contains data describing the overlap between cells expressing the senescence eigengenes and those expressing the NFT eigengenes for both cohorts (1 and 2) using both NFT eigengenes (Dunckley and Garcia).

Acknowledgements

This work is supported by NIH/NIA (R01AG068293, R01AG057896, U01AG046170, RF1AG057440, R01AG057907, K99AG061259; P30AG062421; RF1AG051485, R21AG059176, and RF1AG059082 and T32AG021890), Cure Alzheimer’s Fund and Veterans Affairs (K2BX003804). We obtained ROSMAP data from the AD Knowledge Portal (https://adknowledgeportal.synapse.org). Study data were provided by the Rush Alzheimer’s Disease Center, Rush University Medical Center, Chicago. Data generation was supported by National Institute on Aging (NIA) and grants RF1AG57473, P30AG10161, R01AG15819, R01AG17917, U01G46152, U01AG61356, and RF1AG059082. Additional phenotypic ROSMAP data can be requested at https://www.radc.rush.edu. We acknowledge the Texas Advanced Computing Center (TACC) at The University of Texas at Austin for providing high-performance computing (HPC) resources: http://www.tacc.utexas.edu. We acknowledge the Biggs Institute Brain Bank and Massachusetts Alzheimer’s Disease Research Center for providing postmortem human tissue for analyses. The funders had no role in study design, data collection and analysis, decision to publish or preparation of the manuscript.

Footnotes

Competing Interests Statement

Patent applicant: Wake Forest University Health Sciences. Name of inventor: Miranda Orr. Application number: 63/199,927 & 63/261,630. Status of application: Pending. The specific aspect of manuscript covered in patent application: Data from this manuscript was used to file a patent, "Biosignature and therapeutic approach for neuronal senescence".

Code availability

Our R scripts, which are available as supplementary material, can be used to fully reproduce our results. Our code is also publicly available at https://bitbucket.org/habilzare/alzheimer/src/master/code/senescence/Shiva/.

Data availability

The snRNA-Seq data that were analyzed in this study are available from www.synapse.org with synapse IDs: syn18485175 and syn21126462 for cohorts 1 and 2, respectively. Accessing these data requires submitting a Data Use Certificate through AMP-AD website. Clinical data were available in the corresponding publications. The scRNA-Seq data from embryonic cortex and the scRNA-Seq data from entorhinal cortex are also available from Gene Expression Omnibus (GEO)54 with accession numbers GSE103723 and GSE138852.

References

  • 1.Tchkonia T, Zhu Y, van Deursen J, Campisi J & Kirkland JL Cellular senescence and the senescent secretory phenotype: therapeutic opportunities. J Clin Invest 123, 966–972, doi: 10.1172/JCI64098 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 2.Acosta JC et al. A complex secretory program orchestrated by the inflammasome controls paracrine senescence. Nat Cell Biol 15, 978–990, doi: 10.1038/ncb2784 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 3.Musi N et al. Tau protein aggregation is associated with cellular senescence in the brain. Aging Cell 17, e12840, doi: 10.1111/acel.12840 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 4.Chow HM et al. Age-related hyperinsulinemia leads to insulin resistance in neurons and cell-cycle-induced senescence. Nat Neurosci 22, 1806–1819, doi: 10.1038/s41593-019-0505-1 (2019). [DOI] [PubMed] [Google Scholar]
  • 5.Ogrodnik M et al. Obesity-Induced Cellular Senescence Drives Anxiety and Impairs Neurogenesis. Cell Metab 29, 1061–1077 e1068, doi: 10.1016/j.cmet.2018.12.008 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 6.Riessland M et al. Loss of SATB1 Induces p21-Dependent Cellular Senescence in Post-mitotic Dopaminergic Neurons. Cell Stem Cell 25, 514–530 e518, doi: 10.1016/j.stem.2019.08.013 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 7.Chinta SJ et al. Cellular Senescence Is Induced by the Environmental Neurotoxin Paraquat and Contributes to Neuropathology Linked to Parkinson's Disease. Cell Rep 22, 930–940, doi: 10.1016/j.celrep.2017.12.092 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8.Bussian TJ et al. Clearance of senescent glial cells prevents tau-dependent pathology and cognitive decline. Nature 562, 578–582, doi: 10.1038/s41586-018-0543-y (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 9.Zhang P et al. Senolytic therapy alleviates Abeta-associated oligodendrocyte progenitor cell senescence and cognitive deficits in an Alzheimer's disease model. Nat Neurosci 22, 719–728, doi: 10.1038/s41593-019-0372-9 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 10.Justice JN et al. Cellular Senescence Biomarker p16INK4a+ Cell Burden in Thigh Adipose is Associated With Poor Physical Function in Older Women. J Gerontol A Biol Sci Med Sci 73, 939–945, doi: 10.1093/gerona/glx134 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 11.Hickson LJ et al. Senolytics decrease senescent cells in humans: Preliminary report from a clinical trial of Dasatinib plus Quercetin in individuals with diabetic kidney disease. EBioMedicine 47, 446–456, doi: 10.1016/j.ebiom.2019.08.069 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 12.Gillispie GJ et al. Evidence of the Cellular Senescence Stress Response in Mitotically Active Brain Cells—Implications for Cancer and Neurodegeneration. Life 11, doi:doi.org/ 10.3390/life11020153 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 13.Sah E, Krishnamurthy S, Gillispie GJ, Milligan C & Orr ME The cellular senescence stress response in post-mitotic brain cells – cell survival at the expense of tissue degeneration. Life 11 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 14.Arendt T, Rodel L, Gartner U & Holzer M Expression of the cyclin-dependent kinase inhibitor p16 in Alzheimer's disease. Neuroreport 7, 3047–3049, doi: 10.1097/00001756-199611250-00050 (1996). [DOI] [PubMed] [Google Scholar]
  • 15.Katsouri L et al. Ablation of reactive astrocytes exacerbates disease pathology in a model of Alzheimer's disease. Glia 68, 1017–1030, doi: 10.1002/glia.23759 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 16.Hansen DV, Hanson JE & Sheng M Microglia in Alzheimer's disease. J Cell Biol 217, 459–472, doi: 10.1083/jcb.201709069 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 17.Oldham MC, Horvath S & Geschwind DH Conservation and evolution of gene coexpression networks in human and chimpanzee brains. Proc Natl Acad Sci U S A 103, 17973–17978, doi: 10.1073/pnas.0605938103 (2006). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 18.Foroushani A et al. Large-scale gene network analysis reveals the significance of extracellular matrix pathway and homeobox genes in acute myeloid leukemia: an introduction to the Pigengene package and its applications. BMC Med Genomics 10, 16, doi: 10.1186/s12920-017-0253-6 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 19.Mathys H et al. Single-cell transcriptomic analysis of Alzheimer's disease. Nature 570, 332–337, doi: 10.1038/s41586-019-1195-2 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 20.Zhou Y et al. Human and mouse single-nucleus transcriptomics reveal TREM2-dependent and TREM2-independent cellular responses in Alzheimer's disease. Nat Med 26, 131–142, doi: 10.1038/s41591-019-0695-9 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 21.Fan X et al. Spatial transcriptomic survey of human embryonic cerebral cortex by single-cell RNA-seq analysis. Cell Res 28, 730–745, doi: 10.1038/s41422-018-0053-3 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 22.Bryant AG et al. Cerebrovascular Senescence Is Associated With Tau Pathology in Alzheimer's Disease. Front Neurol 11, 575953, doi: 10.3389/fneur.2020.575953 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23.Bhat R et al. Astrocyte senescence as a component of Alzheimer's disease. PLoS One 7, e45069, doi: 10.1371/journal.pone.0045069 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 24.Tchkonia T & Kirkland JL Aging, Cell Senescence, and Chronic Disease: Emerging Therapeutic Strategies. JAMA 320, 1319–1320, doi: 10.1001/jama.2018.12440 (2018). [DOI] [PubMed] [Google Scholar]
  • 25.Hooper AT et al. Angiomodulin is a specific marker of vasculature and regulates vascular endothelial growth factor-A-dependent neoangiogenesis. Circ Res 105, 201–208, doi: 10.1161/CIRCRESAHA.109.196790 (2009). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 26.Boraas LC & Ahsan T Lack of vimentin impairs endothelial differentiation of embryonic stem cells. Sci Rep 6, 30814, doi: 10.1038/srep30814 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 27.Grubman A et al. A single-cell atlas of entorhinal cortex from individuals with Alzheimer's disease reveals cell-type-specific gene expression regulation. Nat Neurosci 22, 2087–2097, doi: 10.1038/s41593-019-0539-4 (2019). [DOI] [PubMed] [Google Scholar]
  • 28.Avelar RA et al. A multidimensional systems biology analysis of cellular senescence in aging and disease. Genome Biol 21, 91, doi: 10.1186/s13059-020-01990-9 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 29.Ashburner M et al. Gene ontology: tool for the unification of biology. The Gene Ontology Consortium. Nat Genet 25, 25–29, doi: 10.1038/75556 (2000). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 30.Gene Ontology C The Gene Ontology resource: enriching a GOld mine. Nucleic Acids Res 49, D325–D334, doi: 10.1093/nar/gkaa1113 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 31.Kanehisa M & Goto S KEGG: kyoto encyclopedia of genes and genomes. Nucleic Acids Res 28, 27–30, doi: 10.1093/nar/28.1.27 (2000). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 32.Orr ME, Sullivan AC & Frost B A Brief Overview of Tauopathy: Causes, Consequences, and Therapeutic Strategies. Trends Pharmacol Sci 38, 637–648, doi: 10.1016/j.tips.2017.03.011 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 33.Fu H et al. A tau homeostasis signature is linked with the cellular and regional vulnerability of excitatory neurons to tau pathology. Nat Neurosci 22, 47–56, doi: 10.1038/s41593-018-0298-7 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 34.Dunckley T et al. Gene expression correlates of neurofibrillary tangles in Alzheimer's disease. Neurobiol Aging 27, 1359–1371, doi: 10.1016/j.neurobiolaging.2005.08.013 (2006). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 35.Otero-Garcia T et al. Single-soma transcriptomics of tangle-bearing neurons in Alzheimer’s disease reveals the signatures of tau-associated synaptic dysfunction. bioRxiv, doi: 10.1101/2020.05.11.088591 (2020). [DOI] [Google Scholar]
  • 36.Furcila D, Dominguez-Alvaro M, DeFelipe J & Alonso-Nanclares L Subregional Density of Neurons, Neurofibrillary Tangles and Amyloid Plaques in the Hippocampus of Patients With Alzheimer's Disease. Front Neuroanat 13, 99, doi: 10.3389/fnana.2019.00099 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 37.Neurohr GE et al. Excessive Cell Growth Causes Cytoplasm Dilution And Contributes to Senescence. Cell 176, 1083–1097 e1018, doi: 10.1016/j.cell.2019.01.018 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 38.Georgakopoulou EA et al. Specific lipofuscin staining as a novel biomarker to detect replicative and stress-induced senescence. A method applicable in cryo-preserved and archival tissues. Aging (Albany NY) 5, 37–50, doi: 10.18632/aging.100527 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 39.Frost B, Bardai FH & Feany MB Lamin Dysfunction Mediates Neurodegeneration in Tauopathies. Curr Biol 26, 129–136, doi: 10.1016/j.cub.2015.11.039 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 40.Barascu A et al. Oxidative stress induces an ATM-independent senescence pathway through p38 MAPK-mediated lamin B1 accumulation. EMBO J 31, 1080–1094, doi: 10.1038/emboj.2011.492 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 41.Hodes RJ & Buckholtz N Accelerating Medicines Partnership: Alzheimer's Disease (AMP-AD) Knowledge Portal Aids Alzheimer's Drug Discovery through Open Data Sharing. Expert Opin Ther Targets 20, 389–391, doi: 10.1517/14728222.2016.1135132 (2016). [DOI] [PubMed] [Google Scholar]
  • 42.Bennett DA et al. Religious Orders Study and Rush Memory and Aging Project. J Alzheimers Dis 64, S161–S189, doi: 10.3233/JAD-179939 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 43.Team, R. C. R: A language and environment for statistical computing, <https://www.R-project.org/.> (2013).
  • 44.R: A Language and Environment for Statistical Computing. (Vienna, Austria, 2017). [Google Scholar]
  • 45.Zainulabadeen A, Yao P & Zare H Underexpression of Specific Interferon Genes Is Associated with Poor Prognosis of Melanoma. PLoS One 12, e0170025, doi: 10.1371/journal.pone.0170025 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 46.Agrahari R et al. Applications of Bayesian network models in predicting types of hematological malignancies. Sci Rep 8, 6951, doi: 10.1038/s41598-018-24758-5 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 47.Samimi H et al. DNA methylation analysis improves the prognostication of acute myeloid leukemia. eJHaem, doi: 10.1002/jha2.187 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 48.Butler A, Hoffman P, Smibert P, Papalexi E & Satija R Integrating single-cell transcriptomic data across different conditions, technologies, and species. Nat Biotechnol 36, 411–420, doi: 10.1038/nbt.4096 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 49.Hafemeister C & Satija R Normalization and variance stabilization of single-cell RNA-seq data using regularized negative binomial regression. Genome Biol 20, 296, doi: 10.1186/s13059-019-1874-1 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 50.Altman NS An Introduction to Kernel and Nearest-Neighbor Nonparametric Regression. The American Statistician 175–185, doi: 10.1080/00031305.1992.10475879 (1992). [DOI] [Google Scholar]
  • 51.Blondel VD, Guillaume J-L, Lambiotte R & Lefebvre E Fast unfolding of communities in large networks. . Journal of Statistical Mechanics: Theory and Experiment doi: 10.1088/1742-5468/2008/10/p10008 (2008). [DOI] [Google Scholar]
  • 52.McInnes L, Healy J & Melville J Umap: Uniform manifold approximation and projection for dimension reduction. arXiv (2018). [Google Scholar]
  • 53.Merritt CR et al. Multiplex digital spatial profiling of proteins and RNA in fixed tissue. Nat Biotechnol 38, 586–599, doi: 10.1038/s41587-020-0472-9 (2020). [DOI] [PubMed] [Google Scholar]
  • 54.Edgar R, Domrachev M & Lash AE Gene Expression Omnibus: NCBI gene expression and hybridization array data repository. Nucleic Acids Res 30, 207–210, doi: 10.1093/nar/30.1.207 (2002). [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

1797052_Sup_Info
NIHMS1797052-supplement-1797052_Sup_Info.pdf (5.8MB, pdf)
1797052_Sup_code
NIHMS1797052-supplement-1797052_Sup_code.zip (73.9KB, zip)
1797052_Sup_Tables

Supplementary Table 1. This file contains Supplemental Tables 1-3 combined into a single excel workbook file with multiple Tabs. Supplemental Table 1 contains the gene lists used to create each of the eigengene; Supplemental Table 2 (multiple tabs) contains eigengene expression data from every cell analyzed across datasets. Supplemental Table 3 (multiple tabs) contains data describing the overlap between cells expressing the senescence eigengenes and those expressing the NFT eigengenes for both cohorts (1 and 2) using both NFT eigengenes (Dunckley and Garcia).

NIHMS1797052-supplement-1797052_Sup_Tables.xlsx (22.9MB, xlsx)

Data Availability Statement

The snRNA-Seq data that were analyzed in this study are available from www.synapse.org with synapse IDs: syn18485175 and syn21126462 for cohorts 1 and 2, respectively. Accessing these data requires submitting a Data Use Certificate through AMP-AD website. Clinical data were available in the corresponding publications. The scRNA-Seq data from embryonic cortex and the scRNA-Seq data from entorhinal cortex are also available from Gene Expression Omnibus (GEO)54 with accession numbers GSE103723 and GSE138852.

RESOURCES