Overall study design
As depicted Figure 1a, we conducted scRNAseq on the central nervous system (hippocampus and hypothalamus) and circulation (peripheral blood leukocytes) from mice with or without TBI treatment at acute (24hr) and chronic (7-day) phases. Sensitive cell types and DEGs within each cell type were identified, and cellular communications were derived based on ligand-receptor co-expression analysis. To connect the mouse genes with human diseases, enrichment of human GWAS signals of neurological diseases among cell-type specific DEGs affected by TBI was assessed. We further prioritized the mitochondrial gene mt-Rnr2, encoding humanin (HN), as a broad target of TBI across cell types, tissues, and time points, and tested the potential of HN to improve TBI cognitive outcome and molecular and cellular pathways.
Unbiased identification of cell identities in multiple tissues across multiple timepoints
We sequenced a total of 78,895 single cells which passed quality control from blood, hippocampus, and frontal cortex at two time points (Supplementary Table 1). A single-cell digital gene expression matrix was generated using a Snakemake15 workflow of Drop-seq Tools13 and dropEst16. Cells were projected onto two dimensions with uniform manifold approximation and projection (UMAP)17 and Louvain18 clustering was used to define cell clusters (Methods). Based on the assessment of sequencing depth per cell type cluster (Supplementary Figure 1a-b), overall library sequencing depth (Supplementary Figure 2), batch effect as inferred by kBET (Supplementary Figure 3), and the clustering of individual samples (Supplementary Figure 4), we saw no evidence of technical or batch contribution to cell clusters.
We used canonical correlation analysis (CCA)19 (Methods) to identify cell type marker genes which were consistent across the different timepoints or conditions. After each tissue was aligned using CCA, cell cluster identities were determined using previously defined cell type marker genes (Supplementary Table 2) from literature as well as from single cell transcriptome references based on hippocampus cells and frontal cortex cells from the DropViz mouse brain atlas20 (Methods). We determined the cell type identities for all cell clusters of the three tissues using known marker genes for 8 blood leukocytes clusters (Figure 1b; Supplementary Figures 5-6), 13 frontal cortex clusters (Figure 1c; Supplementary Figures 7-8), and 17 hippocampal clusters (Figure 1d; Supplementary Figures 9-10). Further subclustering of neuronal populations in hippocampus and frontal cortex revealed 7 and 13 neuronal subtypes, respectively (Supplementary Figures 11-12 & 13-14). In addition to canonical marker genes, we also identified additional highly expressed marker genes for each cell type in each tissue using our dataset (Supplementary Table 3) (details in Methods). The 24 distinct cell clusters (Figure 1e) showed observable gene expression differences among the three tissue types (Figure 1f), between the two time points (Figure 1g), and between TBI and controls (Figure 1h).
Quantification of dynamic and regional shifts in cell types in response to mTBI
Visual inspection of the UMAP two-dimensional embeddings of single cell transcriptomes from mTBI and sham animals revealed striking differences in their gene programs in each tissue at each timepoint (Figure 2a). To quantify the transcriptomic shifts, we applied various approaches and identified top ranked cell types sensitive to mTBI based on converging evidence (Table 1, Supplementary Table 4).
First, we measured how the relative cell type abundances were altered post-mTBI (Supplementary Figure 15). At 24-hrs post-TBI, there was an increase in the fractions of granulocytes in the peripheral blood, and a decrease in neurons and an increase in macrophages in the hippocampus. At 7-day post-TBI, there was a decrease in astrocyte fraction in the hippocampus and an increase in microglia in the frontal cortex. Increases in immune populations across tissues and timepoints is consistent with activated immune response following injury21,22, and neuronal loss at the acute phase of mTBI in the hippocampus is a known consequence23. However, interpretation of these results requires caution due to recent evidence demonstrating less accurate estimation of cell proportions using scRNAseq24.
We then used three alternative transcriptome-based methods to quantify gene expression shifts within individual cell types to rank cell type response to mTBI. The first method used a Euclidean distance which measures the global transcriptomic shift due to mTBI14 (Methods), revealing a general stronger cellular response at 24hr compared to 7-day, particularly in leukocyte populations (Figure 2b). The second method quantified the number of statistically significant DEGs between sham and TBI cells for each cell type using subsampled cells with equivalent number of cells across cell types under the assumption that cell types which are more perturbed by mTBI will have more DEGs. We calculated DEGs on subsampled cell clusters (Supplementary Table 5, Supplementary Figure 16a-b) to give all cell types equivalent cell number hence statistical power. The final method used was a machine learning-based method25 (Supplementary Figure 16c-d), based on the premise that cell types with large differences in their transcriptomes between conditions will achieve high classification accuracy. Given that each method possesses inherent strengths and weaknesses, we considered the consistency across all the methods to rank top sensitive cell types (Table 1, Supplementary Table 4).
In peripheral blood, Cd8+ T cells and Ly6c+ monocytes were among the top cell types in both acute and chronic phases (Table 1). Ly6c+ monocytes are known to increase in number following TBI26 and Cd8+ T cells are known to infiltrate the brain following injury27. B cells, which demonstrated temporal specificity to the 7-day post-TBI timepoint (Table 1) are poorly studied with respect to mTBI and may serve as a candidate for future study of the chronic phase. Cd4+ T cells demonstrated specificity to the 24-hrs timepoint, which aligns with immunosuppression at the acute phase following mTBI, which specifically affects Cd4+ T cells28 (Table 1).
In the CNS, astrocytes were the top ranked cell type to demonstrate global transcriptional sensitivity in both the hippocampus and frontal cortex at both the acute and chronic phases, highlighting its central role in mTBI (Table 1). Endothelial cells showed transcriptomic alterations across both timepoints specifically in the frontal cortex. The hippocampus experienced a strong immediate immune response with large transcriptomic changes in the microglia and activated microglia in the acute phase. In contrast, the frontal cortex had a more delayed immune response from activated microglia at 7-day post-TBI (Table 1). Within the frontal cortex, layer 2/3 neurons was sensitive at the acute phase of mTBI, agreeing with the known neuronal hypoexcitation at this timepoint29. Oligodendrocytes and choroid plexus epithelial cells both had strong transcriptomic alterations that were specific to the chronic phase in the hippocampus. Both cell types can facilitate the repair process where oligodendrocytes can repair myelin on damaged axons30 and choroid plexus epithelial cells release growth factors31 and recruit immune cells32.
Across tissues and timepoints, our cellular sensitivity analyses based on multiple complementary methods revealed that astrocytes and activated microglia were consistently perturbed across brain regions and timepoints, whereas monocytes, T cells, B cells, neurons, endothelial cells, oligodendrocytes, and choroid plexus epithelia cells are sensitive to mTBI with spatiotemporal specificity and dynamics.
mTBI alters cell-cell ligand-receptor coexpression with regional and dynamic specificity
To investigate how mTBI influences the coordinated gene expression between cell types, we used the ligand-receptor based method CellPhoneDB33 to infer cell-cell gene expression coordination (Figure 3a). We found a consistent increase in coordinated gene expression patterns across cell types at the acute phase of mTBI across the peripheral blood (Figure 3b), hippocampus (Figure 3c), and frontal cortex (Figure 3d), which mostly subsided at the 7-day phase.
Astrocytes were found at the center of the increased coordinated gene expression patterns across cell types, especially at the acute phase post-TBI in both the hippocampus and frontal cortex (Figure 3c-d). This is consistent with the high global transcriptional sensitivity of astrocytes across tissues as revealed in the analysis above (Table 1). There was tight coordination across astrocytes, neuronal populations, and oligodendrocytes at the acute phase post-TBI in both the hippocampus and the frontal cortex. Axonal injury and demyelination are hallmarks of TBI which require oligodendrocytes for repair with signs of myelination known to begin as early as 6 hours post-TBI34. Additionally, vascular populations, including endothelial cells, pericytes, and smooth muscle cells, were tightly coordinated in the frontal cortex at the acute phase (Figure 3c-d), whereas the coordination of vascular populations in the hippocampus occurred at the chronic phase, which potentially indicates differential timing of vascular remodeling between brain regions. These results highlight divergent regional cellular responses to the pathophysiology.
Across the brain and periphery, at the acute phase we observed an increase in coordinated immune cell gene expression patterns across many blood leukocyte cell types (Figure 3b) as well as between activated microglia and macrophages in the hippocampus (Figure 3c) and frontal cortex (Figure 3d). The immune cell coordination was sustained in peripheral blood at the chronic phase (Figure 3b) but not in the CNS (Figure 3c-d).
Dynamic and regional alterations in genes and pathways in mTBI
To determine the specific genes and pathways that may contribute to the mTBI pathogenesis in a regional or dynamic fashion, we identified DEGs in individual cell types (Supplementary Table 6) and annotated them with curated biological pathways (Supplementary Table 7). The immune cells displayed dramatic changes in acute phase in peripheral blood (Figure 4a). Activated microglia, microglia, astrocyte, and endothelial cells have more DEGs in acute phase in both brain regions (Figure 4b-c). New oligodendrocytes and subtypes of interneurons such as Dlk1+ and Calb2+ populations have higher number of DEGs at 7-day post-TBI in frontal cortex (Figure 4b).
The enrichment of numerous pathways among the DEGs in various cell types allowed us to attribute fundamental aspects of the mTBI pathophysiology to specific cell types, regions, and timepoints and identify both consistent and unique pathways with spatiotemporal specificity (Figure 4d, Supplementary Table 7). In the acute phase following mTBI, the apoptosis pathway was one of the consistently enriched pathways among DEGs in glial cells across the hippocampus and frontal cortex in addition to several immune cell types in the peripheral blood. The mTOR signaling pathway was also enriched among DEGs in many cell types across the hippocampus and frontal cortex, which is consistent with the role of mTOR signaling in metabolism, growth, proliferation, and survival35. There is also an enrichment of immune response pathways in macrophages, microglia, and activated microglia in the hippocampus and frontal cortex as well as in many cell types in the peripheral blood. In the chronic phase, the immune response pathways were no longer enriched in the peripheral blood and the frontal cortex, but remained enriched in the microglia populations in the hippocampus. Across both the acute and chronic phases, DEGs from Layer 2/3 neurons in the frontal cortex were enriched for decreased long-term potentiation and neurotransmission.
We also identified pathways that showed both regional and dynamic specificity. Downregulation of genes involved in oxidative phosphorylation and the electron transport chain was observed across many cell types in the hippocampus in the acute phase, agreeing with the known depression in metabolic activity caused by mTBI and pointing to the hippocampus as the main site for this known consequence of acute mTBI. In the chronic phase, however, there was an increase in oxidative phosphorylation and the electron transport chain gene expression in hippocampal microglia, smooth muscle cells, and dentate gyrus granule cells, demonstrating the dynamic shift in hippocampal cell metabolism between mTBI stages. In the frontal cortex, the hypoxia pathway was enriched in primarily glial cells and vascular cells in the acute phase following mTBI.
The pathways along with their specific cell type, tissue, and injury time context revealed by our analysis portrait the complex and dynamic molecular processes underlying mTBI pathogenesis.
Enrichment of human neurological disease genes in cell types following mTBI
To assess the association of the cell-type-specific DEGs for each timepoint and tissue with human diseases, we intersected the DEGs with full summary statistics of human GWAS for 4 neurological diseases, including AD, amyotrophic lateral sclerosis (ALS), epilepsy, and multiple sclerosis (MS), which have been associated with TBI36-39 (Methods). We found significant enrichment of DEGs for GWAS association with neurological diseases, but cellular and gene specificity of disease association differed between tissues and timepoints (Figure 4e).
Astrocyte DEGs were strongly enriched for GWAS associations with neurological diseases across timepoints and tissues (Figure 4e). For instance, in the acute phase, hippocampal astrocyte DEGs showed enrichment for ALS GWAS signals, whereas astrocyte DEGs in the frontal cortex showed enrichment for AD, epilepsy, and MS. At the 7-day timepoint, hippocampal astrocyte DEGs were enriched for genetic signals of MS, AD, and epilepsy, whereas no such enrichment was observed for astrocyte DEGs in the frontal cortex. Ventral CA1 pyramidal neuron DEGs in the hippocampus had an enrichment for ALS and AD in the chronic phase but no enrichment in the acute phase. Likewise, layer 2/3 neuron DEGs in the chronic phase in the frontal cortex had a specific enrichment for ALS associated genetic signals. B cell DEGs in the peripheral blood had consistent enrichment across timepoints for genetic signals associated with ALS and AD. These results suggest that tissue- and stage-specific gene alterations in a vulnerable cell type to mTBI may contribute to development of neurological diseases and future experimental testing is necessary to confirm disease causality.
Cell-type specific DEGs
Interrogating cell-type specific genes perturbed by mTBI can reveal fine dysregulation of microcircuits underlying the pathophysiology which can be leveraged for cell-type specific therapeutic interventions. There are many cell-type specific DEGs with unique or consistent spatiotemporal specificity (Figure 5a). Many of these genes have been implicated in the pathophysiology of TBI and related disorders or affect pathways integral to TBI.
We replicated a number of previously reported cell-type specific DEGs, including Tnf and Il1b upregulation in microglia 40,41 and Ctla2a and Adamts9 upregulation in vascular cells in both the frontal cortex and hippocampus in the acute phase42. We also found a greater fold change in Gfap expression in the frontal cortex compared to the hippocampus in the acute phase as the frontal cortex is closer to the injury location in our experiments, which is consistent with the reported Gfap increase relative to severity and proximity to the injury43. Increased expression of Il33 in oligodendrocytes44 and increased expression of Ly86 in microglia45 24-hrs post-TBI were also consistent with previous reports and observed in both the frontal cortex and hippocampus.
Novel cell-type specific DEGs identified from our study include Etnpll, which is linked to schizophrenia and bipolar disorder46, and downregulated in astrocytes across both brain regions and timepoints. In the acute phase, downregulation of Gpr88, which is implicated in spatial learning and anxiety47,48, was specific to Nrip3+ interneurons in the frontal cortex; Tfrc, which modulates ferroptosis sensitivity49 - a mechanism of cell death, was downregulated in endothelial cells in the hippocampus and frontal cortex; Mrps6, which was specifically upregulated in astrocytes in both the cortex and the hippocampus, has been linked to PD50. In the chronic phase, Timp3, which aids in neuroprotection51, was downregulated in endothelial cells in the frontal cortex and hippocampus. Ncan, which suppresses axonal regeneration after neural injury52, demonstrated unique regional and dynamic specificity as it was upregulated in astrocytes in the frontal cortex at 24hrs post-TBI and in astrocytes in the hippocampus 7days post-TBI, thereby suggesting different regional timelines for changes in axonal regeneration.
Robust mTBI DEGs across spatiotemporal domains
In addition to the above cell-type specific DEGs, we also focused on DEGs altered across cell types, tissues and timepoints. These DEGs may underlie the broad symptomology of mTBI due to their ultra-sensitivity to mTBI across spatiotemporal domains, and may serve as biomarkers that can link mTBI brain pathology with peripheral blood cells.
In our previous study, we identified Ttr as a gene which was differentially upregulated in a majority of cell types in the hippocampus at 24hrs. This guided our selection of T4 thyroid hormone to test as a protective agent against the cognitive consequences post-TBI14. The pan-hippocampal upregulation of Ttr post-TBI in the acute phase was confirmed with this independent dataset (Figure 5b). By expanding the tissues and dynamics in the current study, we found that Ttr regulation was specific to the hippocampus and was sustained from 24hr to 7-day post-TBI. The hippocampal specificity of Ttr, a main transporter of the T4 thyroid hormone in the brain, is consistent with the regional specificity of the metabolic pathway depression to the hippocampus as discussed earlier (Figure 4d).
In addition to Ttr, we identified numerous additional consistent DEGs across cell types. Rimklb, which encodes a glutamate ligase, was decreased in cell types in the hippocampus and frontal cortex only in the acute phase (Figure 5b). Rimklb couples glutamate to the acceptor molecule N-acetylaspartate (NAA) which directly controls the availability of N-acetylaspartyl-glutamate (NAAG)53,54, the most prevalent neuroactive peptide in the mammalian CNS. This decrease of Rimklb, potentially limiting NAAG, is consistent with the suppression of neurotransmission observed across the hippocampus and frontal cortex at the acute phase post-TBI (Figure 5b). Malat1, which encodes a lncRNA, was consistently upregulated across all tissues in the acute phase and downregulated across the 2 brain regions in the chronic phase, demonstrating dynamic temporal specificity. Gene mt-Cytb, part of the electron transport chain, was downregulated in the acute and chronic timepoints in the hippocampus.
Our current study also confirmed our previous results that mt-Rnr2 is a pan-hippocampally upregulated gene in the acute phase post-TBI 14 (Figure 5b). The expanded tissues and dynamics of the current study also uncovered that mt-Rnr2 was upregulated consistently across the frontal cortex, hippocampus and peripheral blood in the acute phase post-TBI, but downregulated across the hippocampus and peripheral blood cell types in the chronic phase. The changing directionality of mt-Rnr2 between timepoints indicates a potentially different regulatory role of this gene along the time course of the mTBI response. On the other hand, the fact that the expression patterns of this gene are consistent between peripheral blood cells and brain cells points to the possibility of using this gene as a biomarker of mTBI. mt-Rnr2 encodes the mitochondrial peptide humanin, which has diverse intracellular and extracellular functions and plays an important role in neuroprotection and metabolism55-65, thus serving as an interesting candidate for mTBI intervention.
Targeting mt-Rnr2 with humanin treatment reversed cognitive impairment
We postulate that humanin modulates the metabolic crisis in the acute phase and protect from neuronal death in the chronic phase following mTBI. To test this hypothesis, we introduced humanin post-TBI and evaluated cognitive behaviors as determined with a Barnes Maze test, followed by scRNAseq analysis to understand the molecular mechanisms (Figure 6a). Acute intraperitoneal injection of humanin post-mTBI prevented learning and memory impairment at one-week post-mTBI (Figure 6b).
To tease apart the underlying mechanisms, we conducted scRNAseq on the frontal cortex and hippocampus of TBI mice with and without humanin treatment (Figure 6c and Supplementary Figure 17). We found that humanin treatment reversed the expression of hundreds of DEGs (Supplementary Table 8) and pathways (Supplementary Table 9 and Figure 6d-e) induced by mTBI across many cell types in both the hippocampus and frontal cortex at 24hr post-mTBI.
In the hippocampus, humanin treatment reversed the metabolic depression (Figure 6d) observed in astrocytes, oligodendrocyte populations, endothelial cells, and smooth muscle cells under mTBI (Figure 4d). Astrocytes, known for their role in metabolic support of neurons, showed a strong upregulation in genes in the oxidative phosphorylation pathway after humanin treatment compared to TBI animals (Figure 6f). Humanin also increased genes involved in neurotransmission in layer 2/3 supragranular cortical neurons in the frontal cortex and multiple neuronal populations in the hippocampus and mitigated the disruption of the vascular system in the frontal cortex (Figure 6d-e).
We also further validated select DEGs in oligodendrocytes using RNAscope. We found the mt-Rnr1 and mt-Rnr2 expression levels were enhanced in response to TBI injury, but were normalized by humanin treatment. The expression of another mitochondrial gene mt-Cytb was elevated by humanin treatment (Figure 7).
Overall, the phenotypic and molecular reversals by humanin treatment strongly support that humanin is a master regulator that corrects diverse processes in numerous cell types involved in mTBI.