Introduction

Mild traumatic brain injury (mTBI) is associated with short-term and long-term cognitive dysfunction with increased risk of neurodegenerative disease, especially when experienced repeatedly (Gupta and Sen, 2016). The pathophysiological mechanisms driving long-term cognitive dysfunction and neurodegeneration after mTBI are unclear, though pathological changes in the immediate period following injury have been elucidated. Indeed, what has been termed the secondary injury cascade after mTBI involves axonal damage, microglial activation, cell death, oxidative stress, and neuroinflammation (Pearn et al., 2017; Stojanovski et al., 2019b; Velayudhan et al., 2021; George et al., 2022). More recently, cellular senescence has been identified as a feature of this secondary injury cascade (Ritzel et al., 2019; Schwab et al., 2019, 2021; Tominaga et al., 2019).

Cellular senescence is defined as a state of cell cycle arrest accompanied by morphological changes, nuclear disorganization, metabolic alterations reflecting increased glycolysis, ROS generation, protein synthesis, mitochondrial dysfunction, and the secretion of pro-inflammatory factors collectively called the senescence-associated secretory phenotype (SASP) (Childs et al., 2017). Senescence can be induced by persistent DNA damage signaling through the DNA damage response (DDR), which activates cell cycle control proteins p53, p21, and p16 and their downstream effectors (Kumari and Jat, 2021). Senescent cells are characterized by anti-apoptotic features known as senescent cell anti-apoptotic pathways (SCAPs), that prevent them from apoptosis despite the accumulation of genomic and mitochondrial DNA damage (Zhu et al., 2015). Known and targetable SCAPs include the BCL-2 family of apoptotic proteins, PI3K/Akt, p53/p21/Serpine, HIF1α, and HSP90 (Kirkland et al., 2017). These pathways are often targeted in therapies aimed at reducing or eliminating senescence, as this would lead to apoptosis of SCAP-expressing cells (Kirkland and Tchkonia, 2020). Senescent cells have been shown to accumulate in the brains of patients with Alzheimer’s disease (AD) (Guerrero et al., 2021), Parkinson’s disease (PD) (Chinta et al., 2018), amyotrophic lateral sclerosis (ALS) (Vazquez-Villaseñor et al., 2020), mood disorders (Diniz et al., 2019), stroke (Torres-Querol et al., 2021), and traumatic brain injury (Ritzel et al., 2019; Schwab et al., 2019. Tominaga et al., 2019; Schwab et al., 2021).

There is emerging evidence that cellular senescence is involved in the pathogenesis of mTBI. This was first published by our lab in a cohort of post-mortem male athletes with a history of mTBI (Schwab et al., 2019) and has since been shown in mouse models by us (Schwab et al., 2021) and others (Ritzel et al., 2019; Tominaga et al., 2019). Although DNA damage-induced cellular senescence has clearly been identified as a mechanism driving brain dysfunction after mTBI, uncertainties remain. Indeed, which cell types become senescent and their respective functional repercussions remain unclear. Furthermore, senolytic drugs that target SCAP-expressing senescent cells for apoptosis require further investigation as they represent a potential treatment for individuals who remain symptomatic post-mTBI and at risk of developing a neurodegenerative disease. In recent years, senolytic drugs have been shown to reduce pathology and improve cognitive abilities in a mouse model of AD (Zhang et al., 2019), improve lifespan and measures of frailty in aged mice (Xu et al., 2018), and reduce infarct volume and improve cognitive outcomes in a mouse model of stroke (Lim et al., 2021), suggesting this treatment may also be beneficial in mTBI. To address these gaps, this manuscript uses a mouse model of repeated mTBI (rmTBI) to explore cellular senescence in the injured brain. Using various molecular techniques and single cell RNA sequencing (scRNAseq) we show that the DNA damage response and cellular senescence are early events following rmTBI, in both glial and neuronal cell types. We have also shown that administering the senolytic drug ABT263 1 week following injury improves neurobehavioral function and reduces markers of the DDR and senescence. This manuscript provides compelling evidence of a causal link between cellular senescence and neurobehavioral dysfunction after rmTBI, and highlights novel sex differences which may explain underlying sex differences in mTBI outcomes and highlight potential personalized therapeutic strategies.

Materials and methods Animals

All experiments were approved by the Centre for Phenogenomics Animal Care Committee. Adult (8–10 week old) female and male C57BL/6 mice, housed randomly in the same room, were used in this study. Mice were kept under standard laboratory conditions with access to food and water ad libitum and a 12 h light/dark schedule (all surgical and behavioral work was performed during the light cycle). Separate groups of mice were used for molecular analysis (n = 4 per group unless otherwise stated) and behavior (n = 8 per group, unless otherwise stated) to avoid potential confounding effects of behavioral testing on the brain.

Repeated mild traumatic brain injury model

Mice were randomly assigned to receive rmTBI with a closed-skull controlled cortical impact model or sham procedures. All mice received pre-operative subcutaneous injections of sustained release buprenorphine (1.2 mg/kg) then anesthetized with isoflurane (induced at 4% and maintained at 2%). Once anesthetized, mice were injected subcutaneously with lactated ringers containing 5% dextrose (0.75 ml) and under the scalp with equal parts xylocaine and bupivacaine (total injection volume of 0.1 ml). Mice were placed in a stereotactic frame (Stoelting) with a warmer ready base to maintain animal body temperature, preventing movement of the head with non-rupture ear bars. Once secured in place, a midline incision was made to expose the skull. Once the skull was leveled to the horizontal plane the desired location (2.5 mm right of Bregma, over the right somatosensory cortex) was identified and an impact of 2 m/s was administered with an electromagnetically driven Impact One Stereotaxic Impactor (Leica, Buffalo Grove, IL) using a 5 mm metal tip, to a depth of 1.2 mm, and with a dwell time of 200 ms. After suturing the midline incision closed, mice were placed in a recovery cage on their back, and righting reflex (total time taken for the mouse to flip over onto all four limbs) was recorded. Sham surgeries underwent the same procedure (including injections, incision, and isoflurane exposure), omitting the actual impact to the head. This procedure was repeated three times, 24 h apart, to give a total of three impacts or equivalent sham surgeries to the mice.

Neurobehavioral testing

One week following the final injury, mice were subjected to neurobehavioral testing first with the light dark test (day 1 of testing) and the Morris Water Maze (MWM) test (days 2–10 of testing). Each testing day began with cages placed inside of the testing room for at least 30 min prior to testing, to acclimate the mice to the testing environment. The light dark test was used to assess exploratory behaviors and anxiety-like behaviors. For this test an arena (43.5 cm2) was prepared such that half of the space was bright (200 LUX) and half of the arena was darkened by a black box. Mice were placed in the arena for 10 min and their activity was monitored by infrared tracking. Time spent in each of the components (light or dark) was measured. The next day, MWM test was prepared to evaluate the mice for spatial memory and learning. The maze was positioned inside a pool centrally under the tracking device and approximately 1 meter away from surrounding walls with four different visual cues. The pool was filled with water to approximately 10 cm from the edge (providing mice with a clear view of external cues), heated to 25°C, and colored with non-toxic white paint to make the platform, placed in the centre of a target quadrant, invisible. The MWM test began on testing day 2 with a familiarization day, in which the platform remained visible above the water. The familiarization procedure entailed the mice standing on the platform for 60 s, followed by a single session containing four trials from different starting positions (pseudo-randomly predetermined N, S, E, W), followed again by a 60 s period on the platform. On days 3–5 mice underwent a training phase for the learning paradigm. In this time, each mouse underwent four blocks of three trials (a total of 12 trials per day per mouse) with approximately 15 min of rest between trial blocks. Time taken to reach the hidden platform (escape latency) was measured and swim speed was assessed. The learning probe day was on day 6, where the platform was removed and the number of goal crossings and distance to the goal was measured. Days 7–9 were again training days but this time for the reversal learning paradigm, in which the platform location was changed and the time taken to reach the new platform location was measured. The final testing day was the reversal learning probe day, in which the platform was removed, and the number of goal crossings and distance to the goal was again measured.

Senolytic intervention

A group of male rmTBI and sham mice received a single intraperitoneal injection at 1-week post-injury containing either the senolytic agent ABT263 (1.5 mg/kg, dissolved in PEG-400, ddH2O, and DMSO) or a vehicle. The rationale for a single dose of the drug ABT263 was a recent publication indicating that a single systemic dose of ABT263 is sufficient to eliminate senescent cells in the brain and improve behavior (Fatt et al., 2022). One week after injection, a total of 2 weeks post-injury, mice underwent the MWM paradigm, as described above, or molecular analysis, as described below.

Animal sacrifice

Following the three impacts or sham surgeries, mice were sacrificed for tissue analysis at 1 week after the final injury, and mice used for behavioral testing (not included in tissue analyses) were sacrificed on the final day of behavioral testing. For molecular analysis, mice were sacrificed via transcardial perfusion with PBS and heparin under ketamine (150 mg/kg) and xylazine (10 mg/kg). Following perfusion, the brain was immediately removed, cut into two hemispheres, and flash frozen in liquid nitrogen in a bath of isopentane and stored at –80°C. For histological analysis, mice underwent the same perfusion protocol followed by an additional perfusion with 4% paraformaldehyde, followed by 24H post-fixation in 4% PFA. These fixed brains were then cut into two hemispheres and processed in paraffin embedded blocks.

Histology

Formalin-fixed paraffin embedded blocks from the ipsilateral side of each mouse brain were used for immunohistochemistry. Sagittal blocks were cut into six-micron sections and mounted on glass slides. Sections were stained with hematoxylin and eosin (H&E), glial fibrillary acidic protein (GFAP, Dako kit, KAC0002), ionized calcium binding adaptor molecule 1 (IBA1, FujiFilm Wako #019-19741 1:5000), and γH2AX (Millipore, 1:1,000, JBW301). Quantification of histology images was performed using QuPath software by taking three ROIs from the cortical region near impact, estimating stain vectors using automated pre-processing, followed by an automated positive pixel count analysis. Differences in staining were determined by percent of positive pixels detected.

Western blot and dot blot

For Western Blot (WB) the ipsilateral cortex and underlying hippocampus near the impact region was homogenized and sonicated in ice-cold PBS containing a protease and phosphatase inhibitor cocktail (Thermofisher). The supernatants were harvested after centrifugating at 12,000 g for 20 min. Protein concentration was measured using a protein assay kit (Biorad). Protein samples were loaded in 10% polyacrylamide gels (Biorad) and transferred to a nitrocellulose membrane (Biorad). Membranes were blocked in 2% BSA of tris buffered saline with 0.05% tween-20 (TBS-T) for 2 h and incubated with primary antibodies for 2 h at room temperature or overnight at 4°C. HRP conjugated secondary antibodies [Sigma, anti-mouse (A9044, 1:10,000) and rabbit (A0545, 1:20,000)] were employed. Membranes were visualized by the Odyssey Fc imaging system (LI-COR). Primary antibodies were 53BP1 (Novus biologicals, NB100-304, 1:10,000), DNA2 (Thermofisher, PA5-68167, 1:2,000), p21 (BD biosciences, 556430, 1:1,000), p16 (Thermofisher, PIMA517142, 1:1,000), γH2AX (Cell signaling technology, 2577S, 1:1,000), HMGB1 (Abcam, ab18256, 1:1,000), Mouse-Reactive STING Pathway Antibody Sampler Kit (Cell signaling technology, 16029T), and GAPDH (Cell signaling technology, 2118S, 1:1,000). Targets were normalized to GAPDH to control for loading.

For the Dot Blot, the same brain region as used in Western Blot analysis was used and dsDNA was isolated by first isolating the nuclear fraction followed by purification of genomic DNA containing DNA-RNA hybrids. dsDNA was measured by NanoDrop One (Thermofisher) and diluted in elution buffer for final concentrations of 200, 100, 50, 25, and 12.5 ng. Three nitrocellulose membranes were spotted with 2 uL of each sample dilution and dried for 1 h, followed by 2 h incubation in 1% BSA in TBST, then incubation in primary antibodies in 1% BSA in TBST overnight at 4°C. The primary antibodies were as follows: anti-dsDNA for detection of DNA (1:20,000 dilution, Abcam AB27156), anti-S9.6 for detection of R-Loops (1:5,000, Sigma MABE1095), and anti-DNA/RNA oxidative damage (1:4,000, Abcam ab62623). Membranes were incubated in HRP-conjugated secondaries (anti-mouse, 1:10,000 dilution, source) for 2 h followed by a 15 min wash with PBST and two 15 min PBS washes. Finally, membranes were developed with enhanced chemiluminescence (ECL) substrate (Bio-Rad) and blots were captured with the Odyssey Fc imaging system (LI-COR). Targets were normalized to dsDNA to control for loading.

Single cell ribonucleic acid sequencing

The ipsilateral hemisphere, excluding brainstem and cerebellum, were incubated in lysis buffer before homogenization. Isolated nuclei were washed twice with cold wash and resuspension buffer and then filtered through a 40 um filter. Nuclei were stained with DAPI and flow sorted. Sorted nuclei were used as input into the 10X Genomics single-cell 3′ v3.1 assay and processed as outlined by user guide. Briefly, following counting, the appropriate volume for each sample was calculated for a target capture of 10,000 nuclei. After droplet generation, samples were transferred onto a pre-chilled 96 well plate (Eppendorf), heat sealed and incubated overnight in a Veriti 96-well thermos cycler (Thermo Fisher). The next day, sample cDNA was recovered using Recovery Agent provided by 10x and subsequently cleaned up using a Silane DynaBead (Thermo Fisher) mix as outlined by the user guide. Purified cDNA was amplified for 11 cycles before being cleaned up using SPRIselect beads (Beckman). Samples were diluted 4:1 [elution buffer (Qiagen):cDNA] and run on a Bioanalyzer (Agilent Technologies) to determine cDNA concentration. cDNA libraries were prepared as outlined by the Single Cell 3′ Reagent Kits v3.1 user guide with modifications to the PCR cycles based on the calculated cDNA concentration. The molarity of each library was calculated based on library size as measured bioanalyzer (Agilent Technologies) and quantitative polymerase chain reaction (qPCR) amplification data (Roche). Samples were pooled and normalized to 1.5 nM. Library pool was denatured using 0.2N NaOH (Sigma) for 8 min at room temperature, neutralized with 400 mM Tris-HCL (Sigma). Library pool at a final concentration of 300 pM were loaded to sequence on Novaseq 6000 (Illumina). Samples were sequenced with the following run parameters: Read 1–28 cycles, Read 2–90, index 1–8 cycles.

After deconvolution and quantification using the 10X Genomics CellRanger software, the data were filtered to include only cells with between 500 and 2,500 genes detected, with less than 5% of transcripts contributed by mitochondrial genes. All samples were normalized together using the clustered pool and deconvolute approach available in the scran R package (Lun et al., 2016). Highly variable genes were identified by their deviation from the fitted mean-variance relationship. Principal component analysis was used to further summarize each cell’s transcriptome, and both 2D embeddings and clustering was performed from these principal components as implemented in the Seurat R package (Hao et al., 2021). Clustering parameter optimization was performed as outlined in scClustViz (Innes and Bader, 2018). Differential expression between rmTBI treatment and sham was performed using MAST (Finak et al., 2015) to fit a hurdle model with a random effect term for individuals to effectively address pseudoreplication bias (Zimmerman et al., 2021). Cell types were annotated based on a published mouse brain atlas (Zeisel et al., 2018) using scpred (Alquicira-Hernandez et al., 2019). Gene Set Enrichment Analysis (Subramanian et al., 2005) was performed using a custom gene set collection designed for Enrichment Map (Merico et al., 2010) containing both GO biological process terms as well as a compilation of pathway databases available here: http://download.baderlab.org/EM_Genesets/January_01_2020/Mouse/symbol/Mouse_GOBP_AllPathways_no_GO_iea_January_01_2020_symbol.gmt. This gene set was supplemented by custom gene sets defining senescence core and effector genes, as well as genes associated with the SASP (Coppé et al., 2010). Results from differential gene expression and gene set enrichment analysis can be found in Supplementary files 1 and 2, respectively.

Statistical analysis

Righting reflex was assessed with a two-way ANOVA using injury and sex as variables. Differences between GFAP and Iba-1 staining were assessed with unpaired t-tests. For the light dark test, raw data from Ethovision (Noldus) was obtained and the total time spent in the dark chamber was statistically compared using a two-way ANOVA with injury and sex as variables. For the MWM, raw data from Ethovision (Noldus) was obtained and analyzed in R, using the RTrack package (Overall et al., 2020). The package utilizes a machine learning algorithm to classify swim paths into strategies including goal-oriented, procedural, and non-goal-oriented types. Differences in escape latency were statistically assessed with three-way ANOVAs using injury, sex, and training day as variables and probe trials were assessed with two-way ANOVAs using injury and sex as variables. Differences in search strategy uses were assessed by performing a Chi-Square analysis on the total number of search strategy uses over the entire training period. Western blot data was assessed by normalizing targets to GAPDH followed by a two-way ANOVA with injury and sex as variables. Statistical analysis of scRNAseq data is described in the previous subsection.

Results Repeated mild traumatic brain injury is induced by closed skull impact model in mice

The injuries sustained by mice in this study were deemed mild, as injury led to attenuated righting reflex under 15 min (Namjoshi et al., 2014). Injured mice took significantly longer to right than shams on each of the three injury days (p < 0.001, main effect of injury, two-way ANOVA) (Figure 1A). Over the three injury days, injured mice took on average 267.9 s to right compared to only 54.0 s in shams. There was no significant effect of sex on righting reflex by way of two-way ANOVA.

Figure 1. Injury induced by closed skull impact model is mild. The injury induced by the presented model can be deemed mild, as indicated by righting reflex measured immediately after injury. rmTBI mice took significantly longer to right than shams (A) on each day with sex having no effect (p < 0.001, main effect of injury and main effect of day, three-way ANOVA, n = 6 per group). Also supportive of a mild injury, despite the absence of any gross lesion (B), rmTBI mice showed significantly increased GFAP staining (p = 0.04, unpaired t-test, n = 4 per group) (C,D,G), and an increase in lba-1 staining that did not reach statistical significance (n = 4 per group) (E–G) in the cortical region near impact. Error bars represent standard deviation. Indication of statistical significance is as follows: *p < 0.05.

Despite the absence of a visible lesion (Figure 1B), injured mice showed other evidence of mild injury. rmTBI mice had increased GFAP staining in the cortex and hippocampus (Figures 1C,D), which was significantly elevated compared to shams in the cortical region (p = 0.04, unpaired t-test) (Figure 1G). In addition to significantly increased positivity, GFAP-positive astrocytes in injured mice frequently displayed ramification and branching (Figure 1D inset) typical of the astrogliosis response initiated by mTBI. Positive staining for Iba-1 in the same brain regions was increased in injured mice compared to shams (Figures 1E,F), although this did not reach statistical significance in the cortical region (Figure 1G). Some Iba-1-positive microglia in the injured brain displayed altered morphologies, including ramification and elongation (Figure 1F inset). These changes suggest the presence of a mild injury induced by this model.

Cognitive dysfunction and behavioral disinhibition 1 week after rmTBI

At 1-week post-injury, injured mice displayed behavioral disinhibition, defined as a tendency toward risk-taking, novelty-seeking, and negligence (Iacono et al., 2003) as measured by the light dark box task. In this test, injured mice spent significantly more time in the light chamber and less time in the dark chamber compared to shams (p = 0.04, two-way ANOVA) with no significant effect of sex (Figure 2A).

Figure 2. Sex-specific cognitive dysfunction and behavioral disinhibition 1 week after rmTBI. In the light-dark box test (A), rmTBI mice spent significantly longer in the light chamber compared to shams (p = 0.04, main effect of injury, two-way ANOVA), with no main effect of sex. In the Morris Water Maze (MWM), no significant differences in swim speed were present between groups (B). In the learning and reversal learning trials, rmTBI mice spent significantly longer to reach the platform compared to shams (p < 0.001, main effect of injury, three-way ANOVA) over trial days (p < 0.001, main effect of training day, three-way ANOVA) (C). There was a small effect of sex nearing significance (p = 0.1, three-way ANOVA), and therefore a significant interaction between injury, sex, and training day (p = 0.05, three-way ANOVA) reflecting baseline differences between male and female sham mice. On the learning probe days, rmTBI mice had a significantly higher distance to the goal compared to shams (p < 0.01, main effect of injury, two-way ANOVA) (D). This measurement also showed a sex difference (p < 0.01, main effect of sex, two-way ANOVA) and a significant interaction between sex and injury (p = 0.05, two-way ANOVA). Post-hoc testing indicates a significant difference between male shams vs male rmTBI mice (p = 0.01, Tukey HSD), male shams vs female shams (p = 0.005, Tukey HSD), and male shams vs female rmTBI mice (p = 0.001, Tukey HSD). Similarly, rmTBI mice had significantly less number of goal crossings compared to shams (p = 0.02, main effect of injury, two-way ANOVA) (E) with an effect of sex (p = 0.04, main effect of sex, two-way ANOVA) and a significant interaction between sex and injury (p = 0.01, two-way ANOVA). Similar to distance to goal, post-hot testing indicates a significant difference between male and female shams (p = 0.01, Tukey HSD), male shams and male rmTBI mice (p = 0.007, Tukey HSD), and male shams and female rmTBI mice (p = 0.02, Tukey HSD). Reversal probe tests were not significantly different between shams and rmTBI mice. Using a machine learning classifier (RTrack), search strategy was assessed over training days between sham and rmTBI mice, without consideration for sex (F). Overall, rmTBI mice used more goal-oriented strategies (perserverence, direct path, corrected search, and directed search) and less procedural (chaining, scanning) and non-goal-oriented (random path, circling, thigmotaxis) strategies compared to sham mice in total (p < 0.0001, x2 = 39.95, Chi Square). Significant differences between rmTBI and sham mice were found on day 2 (p = 0.003, x2 = 11.83), day 3 (p < 0.0001, x2 = 22.51), day 6 (p = 0.02, x2 = 8.25), and day 7 (p = 0.004, x2 = 11.25). Error bars represent standard deviation with n = 8 per group. Indication of statistical significance is as follows: *p < 0.05, **p < 0.01, ***p < 0.001. ns, not significant.

Injured mice displayed significant cognitive impairment in the MWM task 1 week following injury (Figures 2B–F). No significant differences in swim speed (velocity) were found between groups (two-way ANOVA; Figure 2B). Across the training days in the learning paradigm, there was a significant main effect of injury on escape latency (p < 0.001, three-way ANOVA) with injured mice taking significantly longer to find the hidden platform compared to shams (Figure 2C), as well as a main effect of training day (p < 0.001, three-way ANOVA). A small effect of sex was seen nearing significance (p < 0.1, three-way ANOVA) resulting in a significant interaction between injury, sex, and training day (p = 0.05, three-way ANOVA) reflecting baseline differences between male and female sham mice in which female sham mice are performing worse than their male counterparts. In the learning probe test, rmTBI mice had a significantly higher distance to the goal compared to sham mice (p < 0.01, main effect of injury, two-way ANOVA) (Figure 2D). This measurement also showed a sex difference (p < 0.01, main effect of sex, two-way ANOVA) and a significant interaction between sex and injury (p = 0.05, two-way ANOVA). Post-hoc testing indicates a significant difference between male shams and male rmTBI mice (p = 0.01, Tukey HSD), male shams and female shams (p = 0.005, Tukey HSD), and male shams vs female rmTBI mice (p = 0.001, Tukey HSD). Similarly, rmTBI mice had a significantly less number of goal crossings compared to shams (p = 0.02, main effect of injury, two-way ANOVA; Figure 2E) with a main effect of sex (p = 0.04, two-way ANOVA) and a significant interaction between sex and injury (p = 0.01, two-way ANOVA). Similar to distance to the goal, post hoc testing indicates a significant difference between male and female shams (p = 0.01, Tukey HSD), male shams and male rmTBI mice (p = 0.0007, Tukey HSD), and male shams and female rmTBI mice (p = 0.02, Tukey HSD). In the reversal probe test, there were no significant differences in mean distance to the goal nor number of goal crossings, although both male and female rmTBI mice had a slightly higher distance to the goal compared to shams (Figures 2D,E).

Search strategy analysis further confirmed the presence of cognitive and executive dysfunction in injured mice compared to shams (Figure 2F). Three main categories of search strategies were assessed. Non-goal-oriented strategies include thigmotaxis, circling, and random paths, indicating that the mouse is not attempting to reach the goal but is instead motivated to escape the pool. Procedural strategies, including scanning and chaining, indicating that the mouse is aware of a goal, but is using egocentric measures to find it. Last, allocentric strategies including direct searches, corrected searches, and direct paths indicate that the mouse is orienting itself based on cues around the room to find the goal. rmTBI mice used significantly more non-goal-oriented strategies and less goal-oriented strategies compared to shams overall (p < 0.0001, χ2 = 39.95, Chi-Square analysis) (Figure 2E). For the learning paradigm on the first day of training, use of search strategies was not significantly different between rmTBI mice and shams. However, rmTBI and sham mice utilized significantly different strategies on training days 2 (χ2 = 11.83, p = 0.003) and 3 (χ2 = 22.51, p < 0.001). From training day one to training day three sham mice increased the use of goal-oriented strategies by 32.0%, while rmTBI mice only increased the use of goal-oriented strategies by 17.7%. For the reversal learning paradigm, there was no significant difference between search strategy usage on day 5, but a significant difference on days 6 and 7 (χ2 = 8.25, p = 0.016 for day 6, χ2 = 11.25, p = 0.003 for day 7). From the first day of reversal learning training on day 5 to the final day of training on day 7, sham mice increased the use of goal-oriented strategies by 21.0%, whereas rmTBI mice only increased the use of these strategies by 17.0%. Together, these data indicate that deficits in both spatial memory and executive dysfunction occur 1 week following rmTBI.

Multiple forms of deoxyribonucleic acid damage accumulate 1 week post-injury

Accumulation of DNA damage in the form of double-strand breaks (DSBs), oxidative lesions, and R-loops was apparent as early as 1-week post-injury. Immunohistochemistry for the γH2AX, a marker of DSBs, revealed extensive staining in the cortex and hippocampus underlying the impact region in rmTBI mice (Figures 3C,D) but not in shams (Figures 3A,B). Injured mice also showed significant increased levels of R-Loops (p = 0.05, two-way ANOVA; Figure 3E), three-stranded nucleic acid structures comprised of a DNA-RNA hybrid and a displaced DNA loop which are a key source of genomic instability and DNA damage (Hegazy et al., 2020). Post hoc testing revealed that this increase was only statistically significant in female mice (p = 0.04, Bonferroni; Figure 3E). Similarly, rmTBI mice showed evidence of increased oxidative DNA damage, marked by 8-hydroxy-2′-deoxyguanosine, that trended toward significance (p = 0.08) compared to shams (Figure 3E). Again, post hoc testing revealed a significant increase in 8-oxo in female rmTBI mice only (p = 0.04, Bonferroni; Figure 3E).

Figure 3. Deoxyribonucleic acid (DNA) damage accumulates near the region of impact 1 week after rmTBI. DNA damage is evident 1 week post-injuries. Immunohistochemistry for yH2AX in female mice reveals DNA damage in the form of DSBs accumulates after rmTBI in both the cortex at the impact region (A,C, scale bar represents 200 μm) and underlying hippocampal dentate gyrus (B,D, scale bar represents 200 μm), with minimal staining in shams. Staining of dsDNA with S9.6 to assess levels of R-Loop structures revealed a significant main effect of injury (p = 0.05, two-way ANOVA), and post-hoc testing indicates that this increased expression is only significant in the female rmTBI mice compared to their sham counterparts (p = 0.04, Bonferroni). Staining of dsDNA for 8-oxo to assess oxidative base damage shows a similar trend, with a main effect of injury nearing significance (p = 0.08, two-way ANOVA) and post-hoc testing revealing the increased expression to be significant only in female mice (p = 0.04, Bonferroni). Representative images of the dot blot are shown to the right of the quantification plot (E). Using Western Blotting, protein levels of yH2AX were found to be significantly increased in rmTBI mice compared to controls (F) (p = 0.05, main effect of injury, two-way ANOVA), with a significant effect of sex (p = 0.0004, main effect of sex, two-way ANOVA) and no significant interaction. DNA2 trended toward increased expression in female rmTBI mice compared to shams, but did not reach statistical significance (F). 53BP1 was found to have a significant main effect of sex (p < 0.0001, two-way ANOVA) but no significant main effect of injury, however, it was trending toward increased expression in female rmTBI mice compared to shams. Representative blots are shown to the right of the plot (F). Error bars represent standard deviation and there are n = 4 per group. Indication of statistical significance is as follows: *p < 0.05.

Using Western Blotting, γH2AX was found to be significantly increased in rmTBI mice compared to controls (p = 0.05, main effect of injury, two-way ANOVA; Figure 3F) with a significant effect of sex (p = 0.0004, main effect of sex, two-way ANOVA) and no significant interaction. DNA2 was not statistically different between groups but trended toward increased expression in female rmTBI mice compared to shams (Figure 3F). 53BP1 also had a main effect of sex (p < 0.0001, two-way ANOVA) but no significant effect of injury, again trending toward increased expression in rmTBI female mice compared to shams.

Together, these data indicate the presence of multiple forms of DNA damage at 1-week post-injury, including DSBs, oxidative damage, and the accumulation of R-Loops, as well as the activation of a DNA damage response in a possibly sex-dependent manner.

Cellular senescence is evident 1 week post-injury

The protein expression of cell cycle control and senescence regulator proteins p16 and p21 were assessed at 1-week post-injury as markers of cellular senescence (Figure 4). rmTBI mice showed significant protein elevation of p21 compared to shams (p = 0.01, main effect of injury, two-way ANOVA) with no effect of sex. Protein expression of p16 was also significantly increased by injury (p = 0.01, main effect of injury, two-way ANOVA) but had a main effect of sex as well (p = 0.001, main effect of sex, two-way ANOVA) with no interactive effect. Post hoc testing indicates significantly higher p16 expression in male shams compared to female shams (p = 0.001, Tukey HSD), female rmTBI compared to female shams (p = 0.05, Tukey HSD), and male rmTBI compared to female rmTBI (p = 0.04, Tukey HSD).

Figure 4. Cellular senescence is evident 1 week post-injuries. Using Western Blotting, protein levels of p21 and p16 were assessed. One week post-injuries p21 was significantly elevated in rmTBI mice compared to shams (p = 0.01, main effect of injury). There was no significant effect of sex on p21 expression. p16 expression was significantly increased in rmTBI mice compared to shams (p = 0.01, main effect of injury) and significantly differed between sexes (p = 0.001, main effect of sex) with no interactive effect. Post-hoc testing revealed significantly higher p16 expression in male shams compared to female shams (p = 0.001, Tukey HSD), female rmTBI compared to female shams (p = 0.05, Tukey HSD), and male rmTBI compared to female rmTBI (p = 0.04, Tukey HSD). All statistical values are results of two-way ANOVAs with n = 4 per sex-segregated group. Error bars represent standard deviation. Indication of statistical significance is as follows: #p < 0.01, *p < 0.05, **p < 0.01, ***p < 0.001.

cGAS-STING mediated type-1 interferon signaling 1 week post-injury

We next assessed the expression of proteins associated with cGAS-STING signaling, a cytosolic DNA sensing pathway involved in initiating a type-I interferon response to genotoxic stress or viral infection which has been shown to be elevated in brain injury and neurodegenerative diseases (Barrett et al., 2021; Decout et al., 2021; Paul et al., 2021). One-week post-injury, cGAS protein level was significantly elevated in rmTBI mice compared to shams (p < 0.0001, main effect of injury, two-way ANOVA) with no sex differences (Figure 5). STING was not significantly altered by injury but instead was significantly higher in males compared to females (p < 0.0001, main effect of sex, two-way ANOVA; Figure 5). IRF3 was significantly increased in rmTBI mice compared to shams (p = 0.02, main effect of injury, two-way ANOVA) and was also higher in males compared to females (p < 0.0001, main effect of sex, two-way ANOVA; Figure 5). No statistically significant changes were found for HMGB1 (Figure 5).

Figure 5. cGAS-STING mediated signaling 1 week post-rmTBI. Western blotting was used to quantify proteins involved in cGAS-STING signaling 1 week after rmTBI. cGAS was significantly increased in rmTBI mice compared to shams (p = 0.005, main effect of injury) with no sex differences. STING was significantly affected by sex (p < 0.0001, main effect of sex), and although trending toward increased expression did not reach statistical significance for effect of injury. Post-hoc analysis revealed STING was significantly increased in sham males compared to both sham and rmTBI females (both p < 0.0001, Tukey HSD), and rmTBI males compared to both sham and rmTBI females (both p < 0.0001, Tukey HSD). IRF3 was significantly increased by injury (p = 0.002, main effect of injury) and affected by sex (p < 0.0001, main effect of sex). Similar to STING, post-hoc analysis indicates that IRF3 expression was significantly higher in sham males compared to both sham females and rmTBI females (p < 0.0001, Tukey HSD). No statistical significant change for HMGB1 was observed. All statistical values are results of two-way ANOVAs with n = 4 per sex-segregated group. Error bars represent standard deviation and indication of statistical analysis is as follows: #p < 0.1, *p < 0.05, **p < 0.01, ***p < 0.001. ns, not significant.

Single-cell ribonucleic acid sequencing of the injured mouse brain reveals senescent-like neurons Identification of cell types and quality control

To further characterize the pathophysiological mechanisms activated following rmTBI, we used unbiased high-throughput scRNAseq to compare the transcriptional profiles between rmTBI and sham mice 1-week post-injury. Data was obtained from the brain tissue of 2 rmTBI and 2 sham male mice. Cells with similar transcriptomes were grouped together using a recognized clustering algorithm and quality control was performed as described in the Methods section. Following quality control, 44830 cells (20959 from sham group, 23871 from rmTBI group) were grouped into 23 clusters with distinct profiles that mapped onto 6 major cell types: neurons, astrocytes, oligodendrocytes, immune cells, vascular cells, and ependymal cells (Figure 6A). Cell counts for each cell type between experimental groups are found in Table 1. A very small number of ependymal cells were identified, and so they were excluded from our gene set enrichment analysis (GSEA) analysis and interpretation due to lack of statistical power.

Figure 6. Single-Cell RNA sequencing of the injured mouse brain. A total of 23 unique cell clusters were identified (A), with similar cellular composition and RNA feature counts between shams and rmTBI mice (B,C). In total, 5,684 pathways were significantly enriched in neuronal cell types (D), of which 636 (11.2%) were uniquely enriched in a single cluster (E). Gene set enrichment analysis (GSEA) revealed a senescence-like expression profile in neuronal clusters (F) including DNA damage and cell cycle control, senescence-associated inflammation, senescence-associated anti-apoptotic mechanisms, and cell specific changes with potential functional repercussions. Glial cells shared a total of 161 enriched pathways alongside unique features for each cell type (G). GSEA revealed a similar expression profile in glial cells to neurons, indicating senescence and cell-specific functional changes (H).

Table 1. Summary of cluster identities and cell count per cluster by condition.

Cellular composition is similar between rmTBI and sham animals

Cell identity was largely preserved in the injured brain, as indicated with unbiased clustering and a comparison of feature RNA count between rmTBI and sham mice (Figure 6B). Furthermore, the proportion of cells identified for each group per cluster confirmed a similar cellular composition between sham and injured brains (Figure 6C).

Identification of rmTBI-related pathways

We performed differential gene expression (DGE) analysis between sham and injured cells, then subsequently performed GSEA, a powerful and sensitive tool with higher sensitivity compared to the DGE analysis (Liu et al., 2018). GSEA aggregates information from broad sets of genes that are presumed to be functionally related. This analysis revealed several pathways which are significantly changed following brain injury.

Although cellular senescence is evident in the mouse brain 1-week post-injury, senescent cells likely only make up a small subset of cells in the injured brain, with their downstream repercussions having toxic effects to surrounding cells and slowly spreading the phenotype. As single-cell sequencing is inherently prone to incomplete detection of genes expressed at low levels, we do not expect senescence-related genes to be the top DEGs in this analysis. Therefore, to detect senescence, emphasis is made on pathways enriched in rmTBI using GSEA. Supplementary data files containing DEG analysis and GSEA are made available. Noteworthy, the number of cells in cluster 23 was insufficient for DEG and GSEA analysis, and so it was excluded from future results sections.

Senescent-like neurons emerge in the week following rmTBI

Whether or not neurons can become senescent is controversial and has been debated. Some studies have argued that neurons cannot truly become senescent due to their post-mitotic nature. To account for an ongoing debate on whether neurons can become senescent and/or take on a senescent-like phenotype (Fielder et al., 2017; Sah et al., 2021), we interpreted pathways enriched in glial cell and neuronal clusters separately and investigated differences in their senescence pathways. Neurons were classified by subtypes that were assigned by comparing transcriptomes to an existing mouse brain atlas (Zeisel et al., 2018). In total, 5684 pathways were enriched in neuronal cell types (Figure 6D). To first identify pervasive neuronal changes irrespective of neuronal subtype, we looked at 85 pathways which were common to all neuron-identified clusters: 0, 2, 4, 5, 7, 9, 10, 11, 12, and 14. This analysis revealed activation of the DDR in all neuron subtypes, and in particular several pathways involved in regulation of nucleotide levels were enriched. Indeed, activation of the DDR leads to an increase in nucleotide pools to reduce genotoxic stress, provide resources for repair, and escape potential senescence induction (Moretton and Loizou, 2020). In addition, neurons showed evidence of oxidative phosphorylation, mTOR signaling, selective autophagy, interaction with the innate immune system, proteasome degradation, and cell stress signals. Notably, neuronal clusters also showed enrichment for pathways involved in the initiation of proteinopathies including Parkinson’s disease and amyloid fiber formation. Collectively, the shared enriched pathways across all neuronal clusters provide evidence that events leading to proteinopathy are established early and potentially regulated by the DDR. To identify how different subtypes of neurons may respond to these common signals, we next identified enriched pathways unique to each neuronal cluster. On average, 11.2% of the enriched pathways were unique to a cluster, with cluster 11 showing the most unique signature of 44.0% (Figure 6E). A small subset of the GSEA for neuronal clusters has been visualized in Figure 6F, highlighting features of cellular senescence and features enriched in only a subset of neuronal subtypes. Pathways unique to each neuronal subcluster demonstrate how different subsets of neurons uniquely respond to senescent-like and genotoxic cues and provide a functional differentiation of cells that become hyperexcitable or pro-inflammatory (Figure 6F).

Cluster 0 was identified as excitatory telencephalon projecting neurons. These cells were uniquely enriched for 48 pathways compared to all other neuronal cell clusters. These pathways were involved in cell stress and DNA damage signaling, senescence, the SASP, proteinopathy, and activation of the immune system. In particular, signal transduction stimulated by angiotensin II was enriched in these cells, which has recently been shown to cause DNA damage and senescence (Herbert et al., 2008). Several features of the SASP were enriched such as IL1-β production, Senescence_SASP, production of tumor necrosis factor (TNF), and the PI3K pathway. A key feature of this cluster was enrichment of pathways involved in proteinopathy including pathways for the regulation of tau-protein kinase activity, the unfolded protein response, and Alzheimer’s disease. The unfolded protein response is another stimulus leading to DNA damage induced senescence, conceivably representing a positive feedback loop. Pathways involved in activation of the innate immune system were also evident in this cluster.

Cluster 2 was identified as multiple types of neurons including di- and mesencephalon neurons, cholinergic neurons, and telencephalon interneurons and was enriched for 32 pathways compared to all other neuronal clusters. These unique pathways predominantly reflected the DDR, specifically the global genomic nucleotide excision repair, dysregulation of the postsynaptic density structure and neurotransmission. In addition, unique pathways reflected endoplasmic reticulum stress and altered metabolism of rRNA, insulin, and ATP production.

Cluster 4 was identified as inhibitory telencephalon projecting neurons and was enriched for 74 pathways compared to all other neuronal clusters. Of these unique pathways, 15 involved altered metabolism including positive regulation of ATP production and positive regulation of nucleotides. This alteration in metabolism may reflect increased energy expenditure associated with increased production of pro-inflammatory factors in the SASP and mounting a response against genotoxic stress. Indeed, pathways uniquely enriched in this cluster were consistent with SASP signaling including terms such as “inflammatory response,” “IL-6 signaling pathway,” and “regulation of NF-kappaB signaling” and the SCAP PI3K-Akt pathway. Other enriched terms indicated a senescence-like phenotype including signaling by NOTCH2, which has been shown to induce senescence in neural stem cells via the induction of cell-cycle arrest (Hoare and Narita, 2017), and the AP-1 pathway, which has been shown to be involved in a reversible state of senescence in response to stimuli such as cytokines, growth factors, and stress signals (Martínez-Zamudio et al., 2020; Zumerle and Alimonti, 2020). The enrichment of these two pathways in particular suggest that this cluster of neurons may be receiving paracrine signals from the microenvironment, from SASP factors such as cytokines and growth factors, and acquiring a senescent-like phenotype. Dysregulation of neurotransmission was evident in this cluster with downregulation of terms involved in excitatory postsynaptic potential. Interestingly, this cluster showed enrichment of pathways involved in estrogen signaling through nuclear receptors, indicating that there may be a role of sex hormones in the senescent-like response of neurons to genotoxic stress.

Cluster 5 was identified as excitatory telencephalon projecting neurons and uniquely enriched for 64 pathways compared to all other neuronal clusters. Pathways consistent with DNA damage and the response to stress were upregulated, alongside metabolic reprogramming and presentation of SCAPs and SASP factors. This cluster showed downregulation of pathways consistent with neurotransmission, including regulation of action potential, glutamate receptor signaling pathway, and synapse assembly, indicating a decrease in neurotransmission in this cluster of cells.

Cluster 7 was identified as telencephalon interneurons and uniquely enriched for 38 pathways compared to all other neuronal clusters. Neurotransmitter signaling, particularly via acetylcholine and norepinephrine, was upregulated in this cluster, in addition to inflammatory signaling through TGF-β, secretion by cell, and transportation and localization of lipids. Similarly to cluster 4, this cluster showed enrichment of the late estrogen response and regulation of hormone levels, perhaps indicating a role of sex hormones in the neuronal response to oxidative stress.

Cluster 9 was identified as telencephalon interneurons and uniquely enriched for 30 pathways compared to all other neuronal clusters. Importantly this cluster showed upregulation of the c-Myc signaling pathway and its downstream targets, which has been previously shown in models of brain trauma, stroke, and Alzheimer’s disease (Lee et al., 2011). c-Myc activation in neurons is associated with cell death and cognitive deficits (Lee et al., 2009). This cluster also showed downregulation of several pathways involved in the regulation of gene silencing by miRNAs, suggesting increased transcriptional activation, aberrant miRNA signaling and mRNA splicing, attributes frequently reported in neurodegenerative brains (Maciotta et al., 2013; Li et al., 2021). This cluster of neurons may therefore be undergoing neurodegenerative processes in response to the cell stress signals common to all neuronal clusters as discussed above.

Cluster 10 was identified as dentate gyrus granule neurons and uniquely enriched for 10 pathways compared to all other neuronal clusters. These pathways reflected integrin signaling, metabolism of lipids and steroids, and catabolism of amyloid precursor protein.

Cluster 11 was identified as immature neurons (non-glutamatergic neuroblasts and olfactory inhibitory neurons) and uniquely enriched for 173 pathways compared to all other neuronal clusters. Many of the enriched pathways in this cluster were reflective of increased neurotransmitter signaling, particularly glutamate-mediated signaling through NMDA receptors, acetylcholine signaling through muscarinic receptors 1 and 3, release of glutamate, and alterations in ion transport. Evidence of modification of the postsynaptic structure and hyperexcitability are present, including the trafficking of new AMPA receptors. In addition to changes in cell-cell signaling pathways, the CCR3 pathway was upregulated, which has been identified as a possible driver of viral-induced dementia (van der Meer et al., 2000) and a key mediator of neuronal death (Zhang et al., 2016). This cluster may therefore be responding to the genotoxic and senescent-like signals with hyper-excitability, which has previously been shown to occur in an accelerated senescence mouse model (Nakanishi et al., 1998) and aging (Simkin et al., 2015). Indeed, neuronal hyperexcitability has been associated with cognitive impairment and may be considered an indicator of disease progression from mild cognitive impairment to neurodegenerative disease (Dickerson et al., 2005; Bakker et al., 2012).

Cluster 12 was identified as excitatory telencephalon projecting neurons and uniquely enriched for 56 pathways compared to all other neuronal clusters. Signaling through p38MAPK was enriched in this cluster, including enrichment of ERK targets and nuclear events mediated by MAP kinases. Downregulation of dendritic spine organization and the excitatory postsynaptic potential were evident, suggesting dysfunctional neurotransmission in this cluster.

Cluster 14 was identified as di- and mesencephalon neurons and uniquely enriched for 63 pathways compared to all other neuronal clusters. This cluster was enriched for several pathways involved in the sensing of genotoxic stress including the p53-dependent DNA damage response, and degradation of proteins required for the cell cycle. Consistent with this, the SASP was upregulated in this cluster with terms such as “SASP,” “regulation of acute inflammatory response,” and “positive regulation of cellular response to TGFβ stimulus” enriched. Accompanied by these changes were metabolic reprogramming, activation of the innate immune system, and lamellipodium reorganization consistent with senescence-associated morphological alterations.

Cluster 15 was identified as di- and mesencephalon neurons and uniquely enriched for 46 pathways compared to all other neuronal clusters. A key feature of this cluster was the enrichment of pathways involved in neurotransmission, particularly activation of kainite receptors, response to acetylcholine, glutamate secretion, and catecholamine secretion. This was accompanied by activation of the innate immune response, oxidative stress signals, altered metabolism, and dysregulation of actin filament polymerization.

From this analysis, we have identified that all neuronal clusters show at least one core feature of senescence. p38MAPK signaling was evident in several neuronal clusters, along with high levels of SASP such as IL6, IL8, NFKB, and growth factor signaling. These clusters also tended to show elevation of SCAPs such as P13K/Akt/Metabolic, p53/p21/Serpine, BCL-2 family, HSP90, and HIF1α (Kirkland et al., 2017). Pronounced metabolic reprogramming was evident in the neuronal clusters, such as autophagy, increased glycolysis, and altered mitochondrial metabolism. These mechanisms have previously been identified as features of senescence and have been argued to reflect cells’ energetic adaption to their novel secretory phenotype (Soto-Gamez et al., 2019).

Glial and vascular cells show deoxyribonucleic acid damage response and initiation of senescence pathways

Clusters identified as glial cells, including astrocytes, oligodendrocytes, immune cells and vascular cells, were examined for enriched senescence pathways next. A total of 161 common pathways were enriched in all glial cell types (Figure 6G) with some of the unique features of each cell type noted (summary of findings in glial cells visualized in Figure 6H). Pathways involved in cell stress signals and DNA damage pathways were enriched, including cell response to stress, p21-dependent cell cycle checkpoints, and nonsense-mediated decay. Evidence of pathways involved in the response to viral infection were enriched despite absence of infection in these mice. Similar to neurons, glial cells also showed enrichment in pathways involved in energy production via glycolysis and nucleotide synthesis, indicating the cells are mounting a robust DNA damage response involving metabolic reprogramming to counter the damage. Other pathways enriched included protein translation and processing, ROBO/SLIT signaling, and pathways involved in the initiation of proteinopathy including Huntington disease, amyloid fiber formation, and Parkinson’s disease. Together this suggests that glial cells, like neurons, experience extensive genotoxic stress and exhibit a collective response involving metabolic reprogramming to synthesize nucleotides, cell cycle arrest by way of p16 and p21, and the emergence of proteinopathy-consistent pathways (Figure 6H).

In astrocyte-identified clusters, 63 pathways were uniquely enriched compared to other glial cell types encompassing the DNA damage response, astrogliosis, synaptic regulation, SASP, glucose signaling, and autophagy. Astrogliosis was indicated by enrichment of NFAT/calcineurin signaling, which is considered critical for astrocyte activation (Furman and Norris, 2014; Kipanyula et al., 2016) and Rac1 signaling that has recently been shown to control astrogliosis in the context of brain injury (Ishii et al., 2017). Dysregulation of synapses by astrocytes was evident in these clusters, indicated by enrichment of pathways involved in neurotransmitter release cycles, neurotransmitter uptake, retrograde signaling, synaptic vesicle recycling, and regulation of glutamatergic transmission. This is consistent with findings showing that senescent astrocytes lead to glutamate toxicity by means of dysregulation of neuronal synapses (Limbad et al., 2020). Indeed, astrocytes also showed features of the SASP such as interleukin signaling and G-CSF (Coppé et al., 2010). Additionally, metabolism was affected, appearing as altered metal iron homeostasis, a feature of Alzheimer’s disease (Isaev et al., 2020) and increased glucose metabolism and insulin-glucose signaling, a feature of senescence (Wang et al., 2022). Indeed, the latter is closely linked to senescence-associated autophagy (Rajendran et al., 2019) an effect seen in this cell type.

In oligodendrocyte-identified clusters 186 pathways were uniquely enriched compared to other glial cell types. A persistent DNA damage response was evident in this cell type, in particular global nucleotide excision repair, cell stress responses, and regulation of apoptosis in response to DNA damage. Several features of cellular senescence were present, including SASP features such as p38MAPK and interleukin signaling, and SCAPs HIF1α, HSP90, p53, and PI3K/Akt. Anti-apoptotic mechanisms in response to DNA damage and metabolic reprogramming were also supportive of a senescent phenotype. Activation of the innate immune system and signaling through extracellular membrane receptors was enriched in oligodendrocytes, suggesting cell-cell interactions between oligodendrocytes and immune cells and perhaps signaling from SASP factors in the extracellular environment. Dysregulation of actin dynamics was evident, suggesting alterations in myelination (Brown and Macklin, 2020).

In immune cell-identified clusters 85 pathways were uniquely enriched compared to other non-neuronal cell types. Interestingly, enrichment of muscarinic acetylcholine receptor signaling was seen in this cell population, consistent with reports of a subpopulation of microglia in AD and stroke which express functional muscarinic receptors (Pannell et al., 2016