Single cell profiling of CD45+ spinal cord cells reveals microglial and B cell heterogeneity and crosstalk following spinal cord injury

Single-cell sequencing of CD45+ cells after spinal cord injury

To establish the cellular and transcriptional profile of resident and infiltrating immune cells in the spinal cord, adult (10–12-week-old) female Swiss-Webster mice received a moderate thoracic spinal cord contusion injury (T9, 6.25 mm, NYU device). Moderate contusion injuries allow for the recruitment of immune cells following SCI and for some level of motor recovery in the animals [45]. Three-millimeter blocks of spinal cord tissue centered on the lesion epicenter were pooled, dissociated, and processed from injured versus uninjured control mice (n = 6 mice/condition), as summarized in Fig. 1A–C. Individual immune cells were isolated by FACS using the pan-immune cell marker CD45, focusing specifically on cells that were smaller and less granular by flow cytometry (i.e., reduced side scatter, Additional file 1). scRNA-seq was performed using the nanowell-based ICell8 system (Takara Bio USA), which generates more reads per cell than a droplet-based system. In total, 426 cells from uninjured, 102 cells from 3 dpi, 454 cells from 7 dpi, and 945 cells from 60 dpi passed quality control and were analyzed. The average total transcript reads across conditions was 645,000 reads per cell (Fig. 1C). Sample reads at each timepoint were well within the range necessary for downstream batch correction and analyses to robustly discriminate different cell populations [46]. Fewer cells and substantially fewer reads (median of 5,423 reads/cell) were detected in the 3-dpi sample compared to the others. While this did not affect the analysis in context of the whole data set, disparate read counts and sparsity from this sample introduced excess variability into downstream analysis of smaller subsets. As such, after the initial clustering analysis (Fig. 2), the 3 dpi data were excluded from subsequent analyses.

Fig. 2figure 2

Analysis of combined data across all timepoints reveals clusters containing major immune cell types. A UMAP embedding of all timepoints combined after normalization by SCTransform. Clusters clearly separated into distinct immune cell populations. Clusters 0, 1, 5 and 9 express canonical microglia markers. Cluster 2, 3 and 7 express B cell markers. Cluster 4 expresses markers of T cells, and cluster 8 expresses NK and NKT (simplified hereafter as NK-/T) cell markers. Cluster 6 expresses markers of monocytes/macrophages, and cluster 10 expresses granulocyte markers. B UMAP plots highlighting the localization of cells across timepoints. Red cells are the cells isolated on the day indicated. C Heatmap of the top five genes driving cluster separation. Macs macrophages, Monos  monocytes, NK Natural Killer cells, NKT Natural Killer T cells

Characterization of immune cell diversity after SCI

Samples from all timepoints were combined and clustered together using the shared nearest neighbor algorithm to gain an overview of the complete immune cell data set. Eleven cell clusters (Clusters 0–10) were identified and subsequently mapped to specific immune cell types (UMAP: Fig. 2A). The identified cell clusters included the major innate and adaptive immune cell types known to participate in the response to SCI [24, 47] including microglia (clusters 0, 1, 5, and 9), NK-/T cells (cluster 8), macrophages/monocytes (cluster 6), granulocytes (cluster 10), T cells (cluster 4) and B cells (clusters 2, 3, and 7) (Fig. 2A). Following SCTransform [48] to normalize the data and remove technical artifacts, samples were plotted by time to demonstrate proper normalization (Fig. 2B). The top five DE genes for each cluster are presented as a heatmap (Fig. 2C). Cell-type designations were first established by analyzing differentially expressed (DE) genes in each cluster and manually comparing them to several canonical markers of leukocytes and microglia. Cell-type designations were confirmed using both SingleR [36], a nearest neighbor, reference-based, label-transfer approach based on Spearman correlations, and by comparing the results with the cell-type reference ImmGen database (Additional file 4A, B).

Dynamic temporal changes in immune cell types occur after SCI

Microglia (Cluster 0, 1, 5 and 9) and B cells ([2, 3 and 7]) comprised the bulk of the total CD45+ cells combined across timepoints (Fig. 3A, B) (53.8% and 28.5% of the total cells, respectively). The remaining 17.7% of total cells were macrophage/monocytes (5.2%), NK-/T cells (3.5%), neutrophils (1.3%) and T cells (7.7%). Our gating strategy focused on smaller, less granular, cells enabling us to capture microglia and to limit the number of granulocytes, macrophages and monocytes sorted for analysis.

Fig. 3figure 3

Immune cell populations isolated within the spinal cord change over time in response to SCI. A Stacked bar chart showing the number of cells present within each cluster at each timepoint as a proportion of all CD45+ cells isolated. B Bar chart of the percentage of major immune cell types present in the spinal cord and how they change in response to injury. C, D bar charts depicting the percentage of microglia (C) and B cells (D) in each cluster relative to all microglia or B cells, respectively. Note the change in percentage of the different clusters within each cell type in response to injury

The profile of the immune cells within the spinal cord changed in response to injury progression. In the uninjured spinal cord, the vast majority of CD45+ cells isolated were microglia (84%). The next largest cell population was B cells (7.7%), which may have included some peripheral blood cells, since animals were not perfused before tissue harvest. Peripheral myeloid, NK-/T, and T cells made up the remainder (8.3%). After SCI, microglia predominated sub-acutely (64.1% at 7 dpi); but, in chronic SCI, they comprised just 36.3% of total CD45+ cells. In contrast, B cells that were initially 7.7% of total CD45+ cells in the uninjured tissue increased to 14.7% at 7 dpi and became the largest population (43%) in the chronic state. Other cells identified showed smaller changes in response to injury (Fig. 3A, B). T cells decreased from ~ 5% of CD45+ cells in the uninjured cord to 2.8% at 7 dpi, rebounding to 10.6% of cells isolated chronically. Peripheral myeloid and NK-/T cells were present at all timepoints post-SCI, ranging from 2 to 10% for monocytes/macrophages, 0–4% for granulocytes, and 1–5% for NK-/T cells. Collectively, this analysis revealed substantial alterations in immune cell populations isolated within the acutely and chronically injured spinal cord, particularly in microglia and B cells.

Microglia and B cells were distributed across multiple clusters, implicating different cell activation states or subtypes[30]. Analysis of the proportion of microglia (Fig. 3C) and B cells (Fig. 3D) by their respective clusters showed time-specific changes warranting further analysis.

Uninjured spinal cord microglia are primed to respond to injury

SCI induces robust inflammation and greater scar formation than a similar magnitude brain injury [49, 50]. Microglia are hypothesized to be involved, as microglia–astrocyte interactions drive glial scar formation [51, 52]. To identify potential functional differences between brain and spinal cord microglia, we compared our uninjured spinal cord microglial transcriptomes with previously published brain microglia transcriptomes [41]. Once the data sets were integrated, we performed DE analysis in Seurat. The resulting gene lists were then evaluated via gene set enrichment analysis, against GO categories, to determine enriched pathways within and across data sets. This identified several distinct gene expression profiles and GO pathway enrichments for microglia in each tissue (Fig. 4A, Additional file 5). Brain microglia were enriched for functions related to RNA processing, including translation, protein synthesis and protein processing. In contrast, spinal cord microglia were enriched for processes involved in peripheral immune cell recruitment and oxidative stress-related processes, such as "PERK-mediated unfolded protein response" and nitric oxide biosynthetic and metabolic processes. This analysis suggests that the brain and spinal cord microglia differ at rest, though more empirical evidence should confirm these identifications.

Fig. 4figure 4

Comparison of uninjured and injured SCI microglia to uninjured, injured, and diseased brain microglia. A GO pathway enrichment in uninjured spinal cord and brain microglia [41] using significantly DE genes, represented as a dot plot (See Additional file 5 for additional enrichments). The x axis is the percentage of genes represented in the pathway which overlapped with enriched genes. B Venn diagram of genes significantly upregulated in the subacute and chronic phases of SCI versus Injury Response Microglia (IRM [41]) and Disease-Associated Microglia (DAM) [53] (See Additional file 6 for lists of up and down regulated genes following SCI.) C Venn diagram of genes significantly downregulated in subacute and chronic SCI versus DAM. D Violin plots of select significantly altered genes overlapping across injury states in our SCI data

Spinal cord and brain microglia respond differently to injury

Similarities between spinal cord microglia and disease associated microglia (DAM) [4] have recently been identified. Activated brain and spinal cord microglia were compared using up- (Fig. 4B) and down- (Fig. 4C) regulated microglial genes in subacute and chronic SCI to those previously identified in brain injury and disease. For brain microglia, we used signature genes previously established for DAM from Alzheimer’s Disease (AD) model mouse brains [53], and injury responsive microglia (IRM [41]), generated by injection of lysolecithin into brain white matter. DAM were collected from 6-month-old 5XFAD mice, which are on a C57/BL6-SJL background, and IRM were isolated from 100-day-old C57/BL6-JL mice 7 days following lysolecithin injection. Both studies were performed using droplet-based approaches. For down-regulated genes, the comparison was restricted to SCI-microglia and DAM, because downregulated IRM genes were not available for analysis.

Tissue- and temporal-specific changes in gene expression were identified for spinal cord and brain microglia in response to injury (Fig. 4B, C). Six overlapping genes were identified across all conditions. SCI–microglia shared four upregulated genes with DAM and IRM, Apoe, Cd63, Lyz2, and Spp1 (Fig. 4D), and two downregulated genes with DAM, the canonical microglia markers [41, 53, 54] P2ry12 and Selplg (Fig. 4D). Temporal evaluation post-SCI identified eleven upregulated genes in acute SCI shared with DAMs and IRMs. These genes were related to metabolism (Cst7, Ctsb, Npc2, and Fth1a), and cell migration/adhesion (Ccl3, Ccl4, Cxcl16, Cd63, Lgals3bp and Axl). They also shared one downregulated genes with DAM, Malat1. A subset of the common genes is shown in Fig. 4D. The DAM and IRM microglia genes overlapping with acute SCI included Axl, a gene upregulated in brain microglia in almost all disease states [55]. Chronic SCI–microglia had no upregulated genes in common with DAM and IRM, but eighteen downregulated genes were shared with DAM, including canonical microglia genes Cx3cr1, P2ry13, and Csfr1. This indicated that while some SCI–microglia transcriptional responses resemble those in the injured brain [4], most were unique to SCI microglia isolated in our study. This could contribute to altered injury magnitudes between different CNS sites, which should be tested empirically.

Chronic response of B cells following SCI

The composition of B cells present after SCI, their functional profile, and the mechanisms leading to their accumulation are not yet known. The prevalence of B cells in our data set—25% of the total CD45+ cells across timepoints and 43% of cells isolated at 60 dpi—indicated roles for B cells in spinal cord pathophysiology beyond those currently described [22, 56].

First, we examined mouse spinal cord sections by immunohistochemistry for B220, a general marker of B cells. These sections came from the B57/BL6 strain, treated with a moderate thoracic contusion using the Infinite Horizon Impactor (for more details, see methods) [44]. B220+ cells were found clustered in large discrete foci within the spinal cord at 42 dpi, (Fig. 5A-A’’). B cells were found in the spinal cord at 3 dpi; however, they were present as individual cells, and no clusters were evident. Quantification of the number and area of clusters showed they formed between 3 and 28 dpi (Fig. 5B) and increased in size between 28 and 42 dpi (Fig. 5C, Additional file 7A, B). Similar accumulation of B cells with time post-SCI has been reported and hypothesized to contain of a mixed cell population [22, 56].

Fig. 5figure 5

B cells increase in prevalence over time following SCI and contain multiple subtypes. A Immunohistochemical staining of B cells in spinal cord tissue 42 dpi, stained with B220 (Green), Tuj1 (Red), and DAPI (Blue). Large and small dashed boxes indicate the regions shown in high power in A’ and A,” respectively. Scale bars: A 500 μm, A’ and A” 100 μm. B Quantification of number of clusters identified in SCI animals over time following injury. Each dot represents an individual animal. Error bars represent SD. C Quantification of the area of clusters per section, represented in pixels. D Isolated and re-clustered UMAP of clusters 2, 3, and 7 identifies four subclusters of B cells. E UMAP plot highlighting cells isolated from each timepoint, top is 0 dpi, middle is 7 dpi, and bottom is 60 dpi, demonstrating most B cells were isolated at 60 dpi. F Dot plot demonstrating relative abundance and expression of the Ig class detected, of selected markers of early B cells, and selected MHC-I and MHC-II genes. G Schematic showing the developmental lineage of B cells including important events along their development. H Heatmap of re-clustered data, and genes identifying them as a subtype identified by Jensen [57]. I UMAP of B cell clusters, with cells identified by developmental stage

To profile the B cells further, a secondary cluster analysis was performed using the three original clusters (Clusters 2, 3, 7). Four distinct sub-clusters, denoted A–D, were delineated (Fig. 5D; Additional file 8; heatmap of top DE genes in the clusters)—including one, Cluster D, which emerged post-SCI (cf., Fig. 5D, E). B cell development involves progression from Pro-B1 cells through Mature B, which are activated to either produce antibodies (Plasma cells) or act as antigen presenting cells (APCs) (Fig. 5F). To establish the activation and maturation states of cells present, several analyses were performed.

Expression of components of the B cell receptor (BCR) were examined to identify maturation state and whether any B cells were acting as APCs. Components analyzed included immunoglobulin heavy (Igh) and light chains (Igl, Igk), signaling components Ig-a (Cd79a) and Ig-b (Cd79b), Ig subclass type (i.e., IgM, IgD, IgG, IgA, IgE), and expression of MHC-I and MHC-II (Fig. 5G). Cluster D, which appeared after SCI, had the most immature BCR. It contained cells expressing Igl components and enzymes needed for heavy chain V(D)J recombination, typically found in pro/pre B cells. Cluster A cells expressed intermediate levels of genes suggesting development of BCR expression. Cluster C cells expressed high levels of IgM and IgG2b indicative of activated B cells post-class switching. Cluster B cells expressed MHC-II components found in APCs. These results indicate the presence of B cells in multiple states of development and activation.

To identify their developmental state more precisely, our B cell data were compared to bulk RNA sequencing data from defined B cell linage states[57] made into a reference in SingleR[36]. This analysis identified different B cell developmental stages from Pro-B1 cells through Mature B cells (Fig. 5H). UMAP plotting of B cell identities showed Cluster D contained pro-B1, pro-B2, and pre-B1 cells, Cluster A was a mixture of pre-B2, pre-B3 and immature B cells, Cluster C included both immature and mature B cells and Cluster B was comprised of mature B cells (Fig. 5I). Thus, a range of the B cell lineage existed within the spinal cord (Fig. 5H, I), including a population of pro/pre-B cells that emerges after SCI.

Localization of pre-B cells to the spinal cord meninges was recently reported [42]. Comparison of our B cells with those isolated from the meninges of SCI and SOD1 mice [42], a model of amyotrophic lateral sclerosis (ALS), showed the B cells we isolated from the spinal cord were most similar to those found in the meninges under inflammatory conditions (Additional file 3A,B). This demonstrated immature B cells are not restricted to the meninges after SCI. Understanding the functions of the B cells within the spinal cord and factors driving their entry could lead to new interventions to counteract B cell-mediated spinal cord pathology [22, 56].

Immune cell pathway responses after spinal cord injury

SCI induces complex immune-mediated changes [58]. Using our scRNA-seq data set, functional analysis of the GO biological pathways and semantic similarity measurements were used to identify shared or divergent pathways across time in response to injury (Top 10 terms, Fig. 6A, B; complete data set, Additional file 9). We focused our analysis on the cell-type specific functional enrichments occurring in response to SCI for microglia and B cells. The functions of other immune cells are included in Additional file 10. Top enrichments unique for each cell type and timepoint analyzed indicates specific, evolving roles of the various immune cells after SCI.

Fig. 6figure 6

Identification of major pathways associated with timepoint and cell identity. A, B Per timepoint, the Wilcoxon rank sum test was used on all cells isolated, followed by GO enrichment using the hypeR package. Semantic similarities of enriched GO terms were generated using rrvgo package [90], producing new categories displayed as a dot plot. The size of the dot indicates the False Discovery Rate (FDR) for the enriched category, and the x axis indicates the percentage of genes overlapping that pathway (See also Additional file 9 for enrichments). A Enriched pathways with minimal change between timepoints. B Enriched pathways either changed with time or were not represented at all timepoints. C, D Wilcoxon rank sum test was used for microglia (C), or B cells (D) at each timepoint, followed by GO enrichment using the EnrichR package and the “GO Biological Pathway 2021 library” to generate bar plots of significantly enriched pathways at each timepoint for the individual cell types (See Additional file 11 for additional enrichments not included in bar charts).

Microglia are among the first cells to respond to injury, and have complex interactions with axons, glia and immune cells [25]. Our assessment showed that microglia in the uninjured, subacute–SCI, and chronic–SCI spinal cord had distinctly different gene expression across timepoints. The gene expression changes were suggestive of altered interactions between microglia and other immune cells (Fig. 6C). Uninjured microglia were enriched for terms reflective of a role in immune surveillance and interactions with neutrophils and other leukocytes; terms identified included “negative regulation of viral process,” “leukocyte aggregation,” and several associated with neutrophil biology. In subacute–SCI, microglia were enriched for pathways related to peripheral immune cell infiltration and stress responses including “regulation of leukocyte degranulation,” “negative regulation of cell communication,” “stress granule assembly,” and “lipoprotein transport.” In chronic–SCI, pathways linked to microglia–lymphocyte interactions predominated including “regulation of lymphocyte activation,” “B cell receptor signaling pathway,” “regulation of B cell receptor signaling pathway,” and “regulation of T cell activation.” Identification of terms in microglia related to their interactions with neutrophils and leukocytes aligns with their role as immune mediators between the CNS and periphery; currently, little is known about these interactions in the context of SCI. Detecting chronic microglia–lymphocyte interactions was particularly notable as it suggested a role for microglia in chronic B cell accumulation. Future work should further examine these results to confirm their involvement.

Unlike microglia, limited functions have been ascribed to B cells in the context of SCI [22, 56]. Our transcriptional analysis of B cells (Fig. 6D) indicated diverse roles related to antibody refinement and interactions with other immune cells. Uninjured spinal cord B cells were enriched for functions related to DNA and RNA modifications suggestive of antibody refinement, such as “regulation of DNA binding,” “regulation of DNA repair,” “regulation of DNA topoisomerase activity,” “positive regulation of isomerase activity,” and “mRNA splicing, via spliceosome.” Subacute SCI-B cells were enriched for multiple pathways regulating neutrophils, including “neutrophil degranulation,” “neutrophil activation involved in immune response,” and “neutrophil mediated immunity,” while chronic SCI-B cells were enriched for “interleukin-23 mediated signaling pathway” and several metabolism pathways. IL-23 signaling coordinates germinal center class-switching and promotes germinal B cell centers [59, 60]. Interestingly, both neutrophils and astrocytic-expression of IL-23 are linked to B cell accumulation and pathology in the experimental autoimmune encephalomyelitis (EAE) model of multiple sclerosis [60].

Interactions between microglia and B cells in the spinal cord

To identify potential cell–cell interactions between immune cells, we applied the publicly available repository of curated receptors, ligands and their interactions, CellPhoneDB [61], to our data set. As we were interested in identifying potential mechanisms contributing to the accumulation of B cells within the chronic SCI to test in future studies, and microglia were by far the majority of cells in our data set, we focused on investigating microglia-B cells interactions; additional interactions are shown in Additional file 12.

In total, 25 putative ligand–receptor interactions were identified between B cells and microglia (Fig. 7A–D). CellPhoneDB pairings are directional: 11 microglia-B cell (M–B) and 14 B cell–microglia (B–M) interactions were found. The number of predicted M–B/B–M pairings decreased with injury progression: 14, 12, and 9, were identified for uninjured, 7 dpi, and 60 dpi spinal cords, respectively. Examination of the identified receptor–ligand pairings predicts a subset with roles in microglial survival, immune cell infiltration, and neuroprotection warranting further investigation.

Fig. 7figure 7

Cell–cell interactions between microglia and B cells in the injured spinal cord. AC Microglia (clusters 0, 1 ,5 ,9) and B cell (clusters 2, 3 and 7) gene expression was analyzed at each timepoint [(A) uninjured, (B) subacute, (C) chronic] using CellPhoneDB [61] to determine genes involved in cell–cell interactions, shown as a dot plot. D Summary of the number of interactions within and across cell types and how they change over time using Venn Diagrams (See Additional file 14 for lists o

留言 (0)

沒有登入
gif