Distinctive transcriptomic and epigenomic signatures of bone marrow-derived myeloid cells and microglia in CNS autoimmunity

Significance

The innate immune system plays a critical pathogenic role in multiple sclerosis (MS). We propose that this inflammation is likely perpetrated by bone marrow-derived myeloid cells (BMC), which accumulate in the central nervous system during the early relapsing phase of the disease. Our data derived from analyses of accessible chromatin indicate that these BMC are equipped with active or latent transcriptional programs that endow them with distinct functions compared to microglia in the animal model experimental autoimmune encephalomyelitis and potentially in MS. The identification of a cellular signature of activated myeloid cells, elucidated in greater detail herein, reveals promising potential therapeutic targets for which there is currently an unmet need.

Abstract

In the context of autoimmunity, myeloid cells of the central nervous system (CNS) constitute an ontogenically heterogeneous population that includes yolk sac-derived microglia and infiltrating bone marrow-derived cells (BMC). We previously identified a myeloid cell subset in the brain and spinal cord that expresses the surface markers CD88 and CD317 and is associated with the onset and persistence of clinical disease in the murine model of the human CNS autoimmune disorder, experimental autoimmune encephalomyelitis (EAE). We employed an experimental platform utilizing single-cell transcriptomic and epigenomic profiling of bone marrow-chimeric mice to categorically distinguish BMC from microglia during CNS autoimmunity. Analysis of gene expression and chromosomal accessibility identified CD88⁺CD317⁺ myeloid cells in the CNS of EAE mice as originating from BMC and microglia. Interestingly, each cell lineage exhibited overlapping and unique gene expression patterns and transcription factor motifs that allowed their segregation. Our observations will facilitate determining pathogenic contributions of BMC and microglia in CNS autoimmune disease. Ultimately, this agnostic characterization of myeloid cells will be required for devising disease stage-specific and tissue-specific interventions for CNS inflammatory and neurodegenerative disorders.

Myeloid cells of the central nervous system (CNS) are ontogenically and functionally dynamic, and their heterogeneity is perceived to correlate with their roles in inflammatory and neurodegenerative disorders (1–3). Outside of pathological contexts in quiescent brain, the myeloid cell population in the CNS comprises microglia as well as perivascular and meningeal-associated macrophages derived from precursors emanating from the extra-embryonic yolk sac (4). Disruption of the blood brain barrier (BBB), as occurs during inflammation or exposure to ionizing radiation, witnesses the arrival of bone marrow-derived myeloid cells (BMC) that have the capacity to engraft within the parenchyma and develop characteristics of tissue-resident counterparts (5, 6).

Myeloid cells represent promising therapeutic targets in multiple sclerosis (MS) (7, 8). The available arsenal of disease modifying therapies, approved on the basis of their efficacy at reducing the frequency of clinical relapses, is ineffective at preventing the transition to and the accumulation of disability associated with progressive MS (9, 10). They also primarily target bone marrow-derived cells. During later, progressive forms of MS, a CNS compartmentalization appears to occur that often leaves patients treatment-resistant. Myeloid cells figure prominently as primary constituents of chronic active, or “smoldering,” lesions that demonstrate gradual expansion in the absence of radiographic BBB compromise (7, 8). Finally, the perceived nature of progressive MS as the neurodegenerative phase of the disease (11) positions myeloid cells as candidate players given their acknowledged roles in primarily neurodegenerative conditions such as Alzheimer’s disease and amyotrophic lateral sclerosis (4, 12).

Critical to any effort to attribute a functional role for myeloid cells in CNS inflammation is their irrefutable identification. This is hampered by well-known difficulties in distinguishing among myeloid cell subsets using morphology, localization, or expression of limited cell-surface markers.

We had previously identified a signature of circulating BMC associated with inflammation in an animal model of MS, experimental autoimmune encephalomyelitis (EAE) (13). It consisted of the surface markers CD11c (encoded by the gene Itgax), CD88 (C5ar1), and CD317 (Bst2). This signature was specifically up-regulated exclusively at the onset of disease and was expressed by cells in suspensions of CNS tissue, correlating with a concomitant decrease in cells in the blood harboring a subset of the markers. Importantly, we detected this signature in “microglia-like” cells within cerebrospinal fluid (CSF) samples of human patients with demyelinating disorders, further suggesting the pathological relevance of these markers even across species. A notable aspect of this signature was the non-specificity of the CD11c, traditionally recognized as an exclusive marker for mature dendritic cells (DC). Phenotyping studies on lineage-specific animal models harboring Cre-recombinase under the control of ITGAX promoter by our group in unison with other reports painted CD11c to be merely another marker shared to different extents by myeloid subsets (14–18). In contrast, we showed that in fact the co-expression of CD88 and CD317 by the CNS myeloid pool was a rare event during quiescent states (13). However, whether this observed phenotype was confined to CNS-resident microglia or the infiltrating BMC or both remained an outstanding question. Therefore, characterizing the origin of CNS CD88⁺CD317⁺ myeloid cells would help determine their utility as disease biomarkers and therapeutic targets.

There is overlap between microglia and BMC based on molecular markers, and thus an exclusive categorization of either cell type has remained elusive. To address this unmet need, unsupervised annotations and profiles from previously published and curated murine cell atlases were utilized in the present study. In contrast to using limited sets of previously “validated” canonical markers, this approach takes advantage of the entirety of the transcriptome for classification. Furthermore, the epigenome—local or global, chemical and structural changes in DNA and associated proteins (19)—impacts gene expression. Epigenomic constraints may operate at different timescales than transcriptional changes. For instance, some epigenetic modifications may represent remnants of early developmental milestones with unclear significance for current cell function (20, 21). Thus, profiling epigenetic states of cells may yield distinct information independent of the corresponding transcriptomes.

Toward this, we employed state-of-the-art single-cell sequencing to objectively and unbiasedly identify the provenance of this signature. While we find that CD88 and CD317 co-expression is detected in both microglia and BMC during acute EAE, we can show that patterns of transcription, as well as the epigenetic changes that influence them, are sufficient to distinguish these cell types.

Materials and Methods

Generation of C57BL/6 CD45.1⁺ > CD45.2⁺ Bone Marrow Chimera Mice.

Bone marrow cells were isolated from the femurs and tibias of CD45.1⁺ congenic C57BL/6 mice. Bones were crushed with a mortar and pestle. Specimens were then passed through a 70 µm nylon mesh cell retainer. Cells were treated with RBC lysis buffer (Sigma-Aldrich), washed 2 times with cold phosphate-buffered saline (PBS), and re-suspended in PBS for counting with hemocytometer. A total of 15 × 10⁶ CD45.1⁺ bone marrow cells were transferred to CD45.2⁺ C57BL/6 recipient mice at 6 wk of age via tail vein injection. Recipient mice were treated with oral antibiotic (0.66 to 0.7 mg trimethoprim and sulfamethoxazole per mL) for 7 d prior to bone marrow transplant. Host bone marrow ablation was achieved using X-ray irradiation. Specifically, animals were placed in a transparent multi-chamber plexiglass container (1 animal/chamber); chambers were cleaned with 70% ethanol prior to use. The animals were dosed with X-ray, using X-Rad 320 instrument (Precision X-ray, North Branford, CT), for a cumulative dose of 950 Rads, (total of 360 s irradiation time). The mice were provided with antibiotic water following the transplant.

Peptides.

Mouse myelin oligodendrocyte glycoprotein peptide 35 to 55 (MOG_p35-55) (MEVGWYRSPFSRVVHLYRNGK), synthesized by Fmoc chemistry by Quality Controlled Biochemicals, Inc. (QCB) and CS Bio, was utilized for active immunization EAE.

Experimental Autoimmune Encephalomyelitis.

To induce active EAE, experimental mice aged 8 to 12 wk (or 6 to 8 wk post-irradiation in case of bone marrow chimera mice) were immunized subcutaneously with myelin MOG_p35-55 (100 μg/100 µL/mouse), emulsified in an equal volume of Complete Freund Adjuvant containing 4 mg/mL H37Ra M. Tuberculosis (Difco, BD, Franklin Lakes, NJ) in each flank as described. Upon immunization and 48-h later, experimental animals received intraperitoneal injection of 200 ng PTX in 200 µL PBS. All experimental animals were observed at least twice daily, and disease severity scores were recorded based on a standard EAE scoring system (22–24); briefly: 0 = no observable clinical signs, 1 = loss of tail tone or mild hind limb weakness but not both, 2 = limb tail and mild hind limb weakness, 3 = moderate hind limb weakness with or without unilateral hind limb paralysis, 4 = complete bilateral hind limb paralysis, 5 = bilateral hind limb paralysis with fore limb weakness or moribund state or death. The following cumulative interventions were performed for all experiment animals. Moist chow was provided daily once the animal reached a clinical score of 3. Regular palpation of the bladder and manual assistance with urination was performed for animals with disease score of 4; animals with EAE score of 4 were kept separated from cage mates with lower scores. Persistence of disease severity at EAE score 4, for at least for 72 h or observation of EAE score 5 regardless of initiation time, warranted euthanasia using carbon dioxide asphyxiation followed by cervical dislocation as the secondary physical method. In our current animal facility, and with our current EAE induction protocol, immunized mice on the C57BL/6 background typically exhibit peak disease severity on day 15 post-immunization. This is followed by a persistent plateau of clinical disease severity. We used this timeline to guide the biological sample collection efforts and aim for the maximum disease severity in each cohort.

Enzymatic CNS Digestion.

Mice were anesthetized with 250 mg/kg i.p. injection of tribromoethanol (Avertin, Sigma-Aldrich) solution and perfused with PBS through the left ventricle as described. Brains were dissected from the skull, and spinal cords were flushed from the spinal column with PBS. Tissues were finely minced using a sterile scalpel and homogenized in cold PBS. We used a commercially available neural tissue dissociation kit for enzymatic dissociation following the manufacturer’s protocol (Miltenyi Biotec). Specimens were subsequently washed with cold PBS. 37% Percoll™ PLUS (GE Healthcare) gradient was used to remove the residual myelin. The myelin-free single-cell suspensions were counted with a hemocytometer.

Multiparameter Flow Cytometry.

A total of 1 × 10⁶ mononuclear cells suspended in PBS were stained using incubation with 1 µg Fc Block (anti-CD16/32, Tonbo Biosciences) for 15 min at 4 °C. Cells were then stained with the following antibodies: CD45 (APC/Fire 750-Biolegend Clone: 13/2.3), CD11c (APC-Biolegend, Clone: N418) MHCII (PerCp Cy5.5-Biolegend, Clone: M5/114.15.2), CD88 (PE, Biolegend, Clone: 20/70), CD317 (BV421, Biolegend, Clone: 927), CD45.1 (FITC-Biolegend Clone: A20), and CD45.2 (BV605, Biolegend, Clone: 104) for 45 min at 4 °C. Events were recorded via BD FACS LSR Fortessa (The Moody Foundation Flow Cytometry Facility, UT Southwestern), equipped with Diva acquisition software (BD Bioscience). For FACS sorting, BD FACSARIA-II SORP (The Moody Foundation Flow Cytometry Facility, UT Southwestern) was utilized. Cells were gated according to morphology based on their side scatter (SSC) vs. forward scatter (FSC) distributions. Doublets were excluded (FSC-A vs. FSC-H and SSC-A vs. SSC-H, where A corresponds to area and H corresponds to height). Gating for FACS sorting consisted of singlet CD45⁺MHCII⁺CD88⁺CD317⁺ CD45.1⁺ cells or CD45⁺MHCII⁺CD88⁺CD317⁺CD45.2⁺ cells. In each sample, a minimum of 50 × 10³ live events were recorded. FlowJo software (BD Bioscience) was used for data analysis.

Single-Cell Gene Expression and Assay for Transposase-Accessible Chromatin (ATAC) Library Preparation and Analysis.

CD45.1⁺ or CD45.2⁺ CD11c⁺CD88⁺CD317⁺ sorted cells (with a viability of >95% by Trypan blue staining) were re-suspended to a concentration of ~1,000 cells/µL for library preparation using the Chromium Single-Cell Multiome ATAC + Gene Expression workflow (10× Genomics, Pleasanton, California, USA). Libraries were then sequenced on an Illumina NovaSeq instrument (Illumina, Inc.). Sample preprocessing, which encompassed barcode demultiplexing and genome alignment (mouse reference genome mm10), and count/cell matrix generation were carried out in Cell Ranger (v3, 10× Genomics).

For gene expression data, raw gene expression matrices were input into Seurat (25) (v4.1) for further analysis. Cell doublets, identified using the R package scDblFinder (26), and low-quality cells expressing <50 genes were purged. This resulted in a total of 12,316 cells in the CD45.1⁺ group and 11,265 cells in the CD45.2⁺ group. To focus downstream analysis on protein-coding genes, we removed mitochondrial, ribosomal, and non-coding genes from all cells. Count matrices were then normalized, integrated, and scaled adhering to the integration pipeline (https://satijalab.org/seurat/articles/integration_introduction.html). We performed nonlinear dimension reduction (UMAP) and cluster identification on the top 30 principal components identified via principal component analysis. Unsupervised cell annotation was carried out using the singleR (27) package with the MouseRNASeqData reference dataset provided by the celldex (27) package. Differentially expressed genes focused on comparisons between CD45.2⁺ microglia and CD45.1⁺ BMC and were identified using MAST (28) with the parameters min.pct = −Inf, logfc.threshold = −Inf, min.cells.feature = 1, min.cells.group = 1, so as to consider all genes rather than a small subset. These genes were then ranked according to average log₂ fold-change, and only genes with a difference of 0.2 in the percentage of cells expressing a gene between BMC and microglia were considered “differentially expressed.” Accounting for the number of total molecules in each cell in the calculation of differentially expressed genes (via the parameter latent.vars) only altered the presence of 1 gene among the top 25. Gene ontology over-representation and gene set enrichment analysis (GSEA) were conducted using the clusterProfiler package (29) with a GO level of 3 and fgsea package (30) with the Reactome database, respectively.

For ATAC data analysis, fragment files were initially analyzed in ArchR (31) (v1.0.1) for quality control, retaining cells with at least 1,000 unique nuclear fragments, transcription start site (TSS) ratio of greater than 4, and only singlets (using a filterRatio parameter value of 1). Filtered cells were then analyzed in Signac (32) (v1.4), where we performed latent semantic indexing to obtain low-dimensional projections that were used to integrate datasets according to the ATAC integration pipeline (https://satijalab.org/signac/articles/integrate_atac.html). Finally, we carried out UMAP dimension reduction and cluster identification for visualization and further downstream analysis. For annotation of cell clusters, we employed label transfer from the corresponding gene expression data as described in the integration pipeline. Peak detection was performed using MACS2 (33), and peak differential analysis and motif detection were conducted in Signac using the FindMarkers and FindMotifs functions, respectively (the latter restricted to peaks in the top 25th percentile ranked by average log₂ fold-change).

Details regarding statistical tests and corrections to control false discovery rates in multiple comparisons were specific to the software packages employed.

Results

CNS Tissue of CD45.1⁺ > CD45.2⁺ Bone Marrow-Chimeric Mice contains CD88⁺ CD317⁺ BMC following Active EAE.

To track the origin of CNS myeloid cells, we generated bone marrow-chimeric mice using CD45.1⁺ donor cells and CD45.2⁺ congenic recipients (hereafter CD45.1⁺ > CD45.2⁺ mice; see Fig. 1A). Phenotyping of post-transplant CNS tissue from CD45.1⁺ > CD45.2⁺ mice during acute EAE indicated successful engraftment of donor CD45.1⁺ and subsequent presence of these cells within the CNS of recipient animals (Fig. 1 B and C). Clinical course of active EAE was similar in CD45.1⁺ > CD45.2⁺ mice compared to CD45.2⁺ wild-type (WT) controls (Fig. 1D). As previously described by our group, CD88⁺CD317⁺ myeloid cells constitute a unique subset in the context of CNS inflammation. Importantly, during active EAE, co-expression of these markers is only appreciably detected following establishment of clinical disease. This prompted us to designate this specific phenotype of myeloid cells as an inflammation-associated subset. Cellular characterization of the CNS compartment in the CD45.1⁺ > CD45.2⁺ bone marrow-chimeric mice and CD45.1⁺ or CD45.2⁺ WT controls showed a significant presence of CD88⁺CD317⁺ myeloid cells in all groups during acute EAE (Fig. 1E). In the CD45.1⁺ > CD45.2⁺ mice, this population comprised predominantly bone marrow-derived CD45.1⁺ and to a lesser extent CD45.2⁺ cells, while the CD45.1⁺ or CD45.2⁺ myeloid cells were absent in CD45.2 or CD45.1 WT controls, respectively (Fig. 1E). Taken together, these results suggested successful engraftment of donor CD45.1⁺ cells in recipient mice as well as de novo generation of CD45.1⁺ BMC that were recruited to the CNS and contributed to the CNS myeloid pool following established EAE. Furthermore, we observed that CD45.1⁺ donor-derived BMC contributed to the CD88⁺CD317⁺ myeloid cell pool within the inflamed CNS. In contrast, host-derived CD45.2⁺CD88⁺CD317⁺ myeloid cells almost certainly encompassed tissue-resident microglia along with any remaining radiation-resistant BMC. Consequently, CNS CD45.2⁺ cells in this model could not be assumed to exclusively represent microglia, and further refinement of this population was required.

Fig. 1.

Generation and characterization of CD45.1⁺ > CD45.2⁺ chimeric animals. (A) Irradiated CD45.2⁺ animals intravenously received 15 × 10⁶ bone marrow cells from CD45.1⁺ donors. EAE was induced using active immunization. CNS tissues were collected at the peak of disease activity for phenotyping and sorting via FACS. (B and C) The mean ± SD of CD45.1⁺ and CD45.2⁺ BMC relative to total CD45⁺ cells within the CNS compartment of engrafted recipients following acute EAE is presented (n = 12, data display pooled analysis of all study animals). FACS analysis confirmed the presence of donor CD45.1⁺ BMC within the CNS compartment following clinical EAE and indicated their contribution to the CD88⁺CD317⁺ “pro-inflammatory” myeloid pool. The gating strategy was employed to separate mutually exclusive CD45.1⁺ donor and CD45.2⁺ host cells for subsequent single-cell analyses. (D) The mean ± SE of EAE clinical scores following active immunization in experimental animals is presented. There was no significant difference between CD45.1⁺> CD45.2⁺ chimera and CD45.2⁺ WT controls in disease onset or severity (P-value > 0.05, n = 12, data display pooled analysis of all study animals). (E) The mean ± SD of CD88⁺DC317⁺ myeloid cells from the total CD45⁺ cells within the CNS of representative experimental animals is shown. There was no significant difference between CD45.1⁺> CD45.2⁺ chimera and CD45.1⁺ or CD45.2⁺ WT controls (P-value > 0.05; n = 3); CD88⁺CD317⁺ cells of CD45.1⁺ > CD45.2⁺ chimera animals comprised both CD45.1⁺ and CD45.2⁺ cells. Abbreviations: EAE, experimental autoimmune encephalomyelitis; BMC, bone marrow-derived myeloid cell; FACS, fluorescence-activated cell sorting.

Single-Cell RNA-seq Distinguishes between BMC and Microglia.

We leveraged the ability of single-cell transcriptome sequencing to reveal the identity and phenotype of the cell populations in question. Cells expressing CD88⁺CD317⁺ and either CD45.1 or CD45.2 were sorted from brain and spinal cord tissue of CD45.1⁺ > CD45.2⁺ bone marrow-chimeric mice at peak disease following EAE induction. On dimension-reduced projections of gene expression, cells largely segregated into two main groups defined by their compartment of origin. Specifically, the larger group included the entirety of bone marrow-derived cells expressing CD45.1, while the smaller group exclusively contained CD45.2⁺ cells, suggesting possible enrichment for host microglia (Fig. 2A). Ultimately, transcriptomic analyses gleaned from publicly available annotation and agnostic to any single canonical microglia-associated marker unequivocally identified cells of the smaller group as microglia, comprising of only CD45.2⁺ cells (Fig. 2A). In contrast, using the same approach, myeloid cells within the larger group, containing CD45.1⁺ cells, were made up of macrophages, monocytes, and DC (Fig. 2A). In downstream analysis, we constrained the BMC population to those cells expressing CD45.1, as these represent the cells that infiltrate the CNS during inflammation. CD45.2⁺ non-microglia cells would include yolk sac-derived CNS-associated macrophages (e.g., perivascular macrophages) as well as other cells that entered the CNS through a BBB rendered leaky by irradiation.

Fig. 2.

Transcriptomic and chromosome accessibility landscape of CD88⁺CD317⁺ cells within distinct myeloid cell pools. (A) CD88⁺CD317⁺ cells expressing either CD45.1⁺ or CD45.2⁺ form distinct clusters on UMAP plots defined by their patterns of gene expression; a clear segregation between yolk sac-derived microglia, with cells of largely CD45.2⁺ origin, and BMC is appreciated. Note that additional labels on the UMAP plot refer to other cell types identified using unsupervised cell annotation rather than unique clusters. (B) UMAP plot of CD88⁺CD317⁺ cells based on patterns of chromosomal accessibility and stratified by expression of CD45.1⁺ or CD45.2⁺ recapitulating the segregation of CD45.2⁺ microglia from other myeloid cells. (C) Heat map of the 28 shared genes (rows) among the top 50 highly expressed genes of microglia and BMC shows heterogeneous expression of genes within cells belonging to the two groups (columns). (D) Overrepresented gene ontology pathways within the top 50 highly expressed genes of microglia and BMC reflect pathways involving antigen presentation, immune regulation, and synaptic plasticity, among others. GeneRatio refers to the fraction of the top genes in a particular process with adjusted P-values, depicted colorimetrically computed using the clusterProfiler package. Abbreviations: UMAP, uniform manifold approximation and projection.

Single-Cell ATAC-seq Distinguishes between BMC and Microglia.

Based on the hierarchy of information flow in cells, we proposed that an assessment of the epigenetic state would provide a complementary analysis of BMC and microglia. Toward this, single-cell chromosome accessibility profiling was conducted simultaneously with single-cell transcriptomics on the same sorted cell populations (CD88⁺CD317⁺ myeloid cells that expressed either CD45.1⁺ or CD45.2⁺). Clustering of cells solely on the patterns of genomic accessibility (agnostic to gene expression data) recapitulated a similar anisotropic distribution, with microglia representing one group segregated from a more densely populated group consisting of other cells largely of bone marrow origin. Taken together, we have identified two discrete populations of CD88⁺CD317⁺-expressing myeloid cells present in the CNS of mice during inflammation using distinct modalities. Unique patterns of epigenetic states and transcriptomic activity permitted their accurate identification as microglia and CNS BMC (Fig. 2B).

Highly Expressed Genes Depict Significant Overlap between BMC and Microglia.

The revelation of two ontogenically distinct cell populations emerging from the data was surprising, as sorting of cells using the markers CD88 and CD317 was expected to enrich a phenotypically uniform myeloid cell type. In fact, similarities between BMC and microglia extended beyond these sorted markers: comparison of the top 50 expressed genes within each group identified 28 shared genes (Fig. 2C and Dataset S1). All had been previously reported to be expressed in mouse or human microglia. These included genes involved in the biological processes of antigen processing and presentation (B2m, Ctss, Psap, H2-K1, Cd74, H2-D1), maintenance of cellular location or storage (Fth1, Tmsb4x, Apoe, Ftl1, Hexb), cellular destruction (B2m, Lyz2, H2-K1, H2-D1), regulation of immune processes (C1qb, Cd74), coagulation (Apoe, Plek), leukocyte migration (Camk1d, Cd74, Lgmn, Spp1), respiratory burst production (Camk1d), and learning or memory (B2m, Apoe, Lgmn, Picalm) (Fig. 2D).

Differentially Expressed Genes by BMC and Microglia Exhibit Activation of Distinct Biological Pathways.

We performed differential gene expression testing to identify unique genes that distinguished microglia from BMC. Focusing on the top 25 most differentially expressed genes within microglia (Fig. 3A), this analysis highlighted a well-known and conserved microglial marker (P2ry12) (1), as well as other “microglia-signature” genes established in other transcriptomic studies, including Cd81 (34), Sparc(2), Siglech (1), Mef2c(1), Pde3b (35), Arhgap5 (36), and Capn3 (35). Additionally, there was a preponderance of genes involved in synaptic transmission, including Tanc2, Nav2, Frmd4a, Cadm1, and Ophn1, in keeping with the critical role of microglia in the regulation of synaptic development and plasticity (1). More rigorous investigation into the representation of biological processes among these differentially expressed genes revealed TGF-beta signaling and cell migration and neuronal polarity (regulated by a member of the Rnd subgroup of the Rho family of GTPases) as statistically significant enriched pathways (Fig. 3B).

Fig. 3.

Heat maps of the top 25 differentially expressed genes (ranked according to average log₂ fold-change) in microglia (A) and BMC (B) with expression levels depicted colorimetrically. Genes are sorted based on ascending p-value produced by the MAST algorithm. Notice the relative paucity of signal within BMC for genes more highly expressed in microglia (A), whereas the more differentially expressed genes within BMC exhibit notable expression within microglia (B). The top ten Reactome gene sets (arranged by the normalized enrichment scores [NES]) computed by GSEA of microglia (C) and BMC (D) reflect distinct active processes, with the statistically significant ones depicted in blue according to their associated adjusted P-value. Abbreviations: BMC, bone marrow-derived myeloid cell.

Compared to microglia, only three genes were preferentially expressed within BMC (Fig. 3C). Cdk8 is a member of the cyclin-dependent kinase family that promotes pro-inflammatory functions such as expression of chemokines, suppression of anti-inflammatory cytokines, and co-activation of NFκB-induced genes (37). Cmss1 is involved in translation as a component of the small ribosomal subunit. Finally, Lars2 encodes the mitochondrial leucyl-tRNA synthetase and supports increased metabolic states (38). Significantly enriched cellular processes represented in the gene expression data of BMC included pathways involved in cellular and mitochondrial metabolism (electron transport chain, gluconeogenesis), cellular response to hypoxia, actin and tubulin folding, reactive oxygen species detoxification, polyamine metabolism, and translation initiation (Fig. 3D).

Taken together, differentially expressed genes characterizing microglia and CNS-infiltrating BMC reflect established roles in the maintenance of local environments to support neuronal communication networks and modulation of the immune system, respectively. When examined in the context of the most highly expressed genes, however, definite pro- or anti-inflammatory roles for these cell subsets are difficult to discern. Of note, the relative paucity of differentially expressed genes within BMC could be the consequence of decreased numbers and levels of expressed genes in comparison with microglia (SI Appendix, Fig. S1A) together with the apparent overlap in gene expression patterns. Lastly, while our analyses with BMC were restricted to the CD45.1⁺ population for reasons we provide above, this decision is not felt to alter our results as comparison of the transcriptomes of CD45.1⁺ and CD45.2⁺ BMC reflected a remarkable degree of correlation (R² = 0.933, p = 2.2 × 10⁻¹⁶, SI Appendix, Fig. S1B).

Chromatin Signatures of Myeloid Cells Reinforce Unique Functional Identities of BMC and Microglia.

We exploited the availability of chromatin accessibility information on BMC and microglia to identify differentially available epigenetic signatures between them. We discerned 1,043 loci in microglia and 40 in BMC that were differentially detected at a log₂ fold-change ratio of 0.5 after correction for sequencing throughput between the two libraries. In microglia, the top-ranked locus occurred within an untranslated region (UTR) of Sall1 (Fig. 4A), a transcription factor (TF) that subserves microglial homeostatic functions (39). Importantly, the openness of the Sall1 locus in microglia relative to macrophages also correlated with higher average expression in the former (Fig. 4A). Other highly ranked loci were positioned within the coding sequences (cds) of Magi1, a scaffolding protein (40); Nova1, whose protein product is involved in RNA processing (41); Map3k4, a kinase implicated in pro-inflammatory responses (42); and a UTR in close proximity to Npnt, which encodes an extracellular matrix protein (43) (Fig. 4A). The enrichment of genomic fragments at these loci was also accompanied by corresponding increases in gene expression.

Fig. 4.

Chromosomal “openness” is displayed as the distribution of normalized accessibility along a chromosomal region for the top-ranked microglia loci (A) near the Sall1Magi1Nova1Map3k4, and Npnt genes as well as the top-ranked BMC loci (B) near the FXFVIIPdpnTacstd2Nrg4 genes; the top track corresponds to microglia while the lower to BMC. Associated gene structure (dark blue line and marks) and location of the differentially accessible peak between microglia and BMC (red line) are depicted below the bar plots, while the log-expression level of each gene within either BMC (green) or microglia (blue) is shown to the right as violin plots. Abbreviations: BMC, bone marrow-derived cell.

In contrast, the two most enriched loci in BMC were located within cds of the genes encoding factor X and factor VII on chromosome 8 (Fig. 4B). Notably, the literature ascribes immunomodulatory roles for coagulation factors released by macrophages in antitumor immunity and bacterial infection (44, 45). Chromatin accessibility at loci overlapping with the cds of genes Pdpn [a transmembrane glycoprotein promoting angiogenesis and motility (46, 47)], Tacstd2 [a transmembrane glycoprotein linked to tissue regeneration and cancer proliferation (48, 49)], and Nrg4 [which encodes for the ErbB4 ligand neuregulin-4 that appears to function in feedback control of activated macrophages (50)] were also enriched in BMC. However, differential gene expression of these genes was only modestly increased in this cell population relative to microglia.

To gain insight into the potential transcriptional regulation across the open chromatin regions within BMC and microglia, we identified candidate TF motifs differentially enriched using the ATAC data (Dataset S1). This analysis showed motifs for TF previously linked to microglia, including Hinfp (51), Egr2 (51), Hes1 (52), and Plagl2 (53). It also highlighted motifs with known functions within neurons and support cells such as astrocytes and oligodendrocytes, including Tcfl5 (54), Nrf1 (55), Zbtb14 (55), Zbtb33 (56), and Glis2 (57). Similarly, open chromatin regions in BMC appeared enriched in motifs for Jun/Fos, members of the Activator Protein 1 (AP-1) family of TFs that induce expression of pro-inflammatory cytokines (58); Cebpg, Cebpb, Cebpe, members of the C/EBP family of TFs known to regulate myeloid cell development and function (59); and Nfe2 and Bach2, TF that repress C/EBP-regulated genes (60). Overall, these analyses indicated noteworthy variability in gene accessibility among CD88⁺CD317⁺ myeloid cells.

Canonical Myeloid Cell Markers are shared between BMC and Microglia during CNS Inflammation.

Efforts to disentangle the heterogeneity of myeloid cells in the CNS have relied on the identification of markers to distinguish among them. The most widely utilized proteins to isolate microglia include the calcium binding adapter protein Iba1, the transmembrane protein Tmem119, the β-subunit of hexosaminidase Hexb, the P2Y purine receptor 12 P2ry12, and the scavenger receptor Fcrls (1, 61). Their utility as “canonical” markers is hampered by variability and context-dependent changes in expression level (61, 62) as well as appreciable expression in other myeloid cell lineages (1, 62, 63). We queried the expression of a broader set of microglial-associated genes (61, 63, 64) within BMC and microglia populations in our dataset. We confirmed that Iba1, encoded by the gene Aif1, is expressed in BMC (Fig. 5), consistent with its known expression within other macrophage subsets (1, 65). Interestingly, Hexb, described as a stable marker restricted to microglia (61), was robustly expressed within BMC (Fig. 5). Other analyzed markers were found to be specific to microglia (Fig. 5).

Fig. 5.

Expression levels of canonical microglia markers in CD88⁺CD317⁺ microglia and BMC are depicted in a dot plot, with the size of each dot representing the percentage of cells within BMC or microglia populations showing nonzero expression of a gene and its average log-scaled expression represented colorimetrically. Abbreviations: BMC, bone marrow-derived cell.

Discussion

Single-cell profiling of transcriptome and chromosome availability demonstrates unequivocally that the signature comprising CD88⁺CD317⁺ myeloid cells is shared by microglia and infiltrating BMC. Its upregulation following the onset of disease in EAE (13) suggests that this signature likely represents a universal activated myeloid state independent of ontogeny and pathology.

Similarities in gene expression between the “activated” CD88⁺CD317⁺ BMC and microglia belie distinct sets of cellular processes operating within them. For BMC, these processes suggested a metabolic phenotype more akin to anti-inflammatory functions, which rely on oxidative phosphorylation for energy production (66), whereas for microglia, they reflected a homeostatic or neuroprotective phenotype (67, 68). However, these statistical annotations ought to be interpreted in light of overlapping pathways enriched in the most highly expressed genes within both BMC and microglia, which primarily convey, in both subsets, pro-inflammatory functions including antigen presentation and reactive oxygen species production within the larger CD88⁺CD317⁺ myeloid pool.

Computation of differentially expressed genes depends naturally on their presence among the populations being interrogated as well as their levels of expression. As we demonstrate in SI Appendix, Fig. S1, CD45.1⁺ cells (BMC) have decreased numbers of genes with decreased levels of expression, which has the potential consequences of 1) emphasizing more differentially expressed genes from microglia (as any gene would, on average, be more highly expressed) and 2) de-emphasizing differentially expressed genes from BMC (even among the more abundant genes within this population). Reasons for this discrepancy are numerous. Impaired hematopoiesis of the donor (CD45.1⁺) cells or altered cell state of the host (CD45.2⁺) cells due to irradiation represent potential sources. The limited transcriptome of CD45.1⁺ cells could also be an inherent characteristic of activated BMC. Neither of these possibilities are mutually exclusive. Cell viability prior to nuclei preparation, identical library preparation, and similar sequencing statistics (Dataset S2) render other technical issues as unlikely sources for the discrepancy.

The epigenetic landscape constraining gene expression patterns also sheds light onto upstream cellular pathways governing functional phenotypes. Hes1, for which motifs were found to be enriched within accessible chromatin in microglia, is a transcriptional repressor known to down-regulate the expression of inflammatory cytokines (69). Downstream of notch signaling, Hes1 thus operates to inhibit the AP-1 transcriptional program (70), which our ATAC-seq data suggest was more functional within BMC. Sall1 is another TF which regulates microglial function and whose deletion results in the acquisition of an inflammatory phenotype (39); interestingly, Sall1 locus was virtually devoid of open chromatin peaks, in keeping with absent detectable gene expression, in BMC. Importantly, Sall1-deficient microglia have been previously shown to harbor transcriptomes that resemble those of CNS-infiltrating monocytes (6, 63), suggesting that this TF to a certain extent oversees the gene expression program that distinguishes microglia from other macrophages.

The patterns of gene expression likely reflect behavior in CD88⁺CD317⁺ myeloid cells along a spectrum of immune-regulatory to pro-inflammatory, or evolving roles during the course of disease. Active EAE in C57BL/6 mice is considered by most to be a monophasic disease, often with persistent neurological disability; therefore, it is somewhat ill-suited as a model to study dynamic molecular and cellular events that occur in chronic CNS autoimmune disease. Consequently, longitudinal experiments in other models of human CNS inflammatory disorders, rather than inference from single snapshots of transcription data, will be required to establish the precise roles of BMC and microglia.

CNS infiltration and engraftment by BMC have been documented following experimental paradigms involving genetic- or small molecule-mediated depletion of microglia (35), direct cellular transplantation (63), irradiation preceding bone marrow transplantation (6) or succeeding the parabiosis of two animals (35, 65), viral encephalitis (71), CNS malignancy (72), chemical-induced central demyelination (73), or autoimmunity (65). Our results corroborate several key findings of these studies, including A) overlapping gene signatures between infiltrating BMC and microglia (including the canonical microglia genes); B) reduced levels of gene expression within BMC; and C) unique transcriptional and epigenetic states of BMC compared to microglia as defined by sets of genes and TF motifs. Importantly, the transcriptional and epigenetic states appear to persist for months following the cell engraftment (6, 63). The preponderance of studies documenting prolonged engraftment contradicts an earlier report that infiltration of myeloid cells from the bone marrow was transient, which can be explained by the contrived use of more terminally differentiated myeloid cells (65). In fact, this phenomenon has been observed in human recipients of allogeneic bone marrow transplants, in whom graft-derived microglia-like cells are detected in the brain parenchyma (6) and CSF (74).

The innate immune system plays a critical pathogenic role in progressive MS, namely tissue atrophy, the expansion of chronic lesions, and the accrual of neurological disability resulting from compartmentalized inflammation within the brain and spinal cord (7). We propose that this inflammation is likely perpetrated by BMC during early relapsing stage of MS, which accumulate in the CNS over the course of the disease. Our data derived from analyses of accessible chromatin indicate that these BMC are equipped with active or latent transcriptional programs that appear to assign them distinct functions from microglia in EAE and potentially in MS (2, 7). The identification of a cellular signature of activated myeloid cells, elucidated in greater detail herein, reveals promising potential therapeutic targets for which there is currently an unmet need.

Supplementary Material

Appendix 01 (PDF)

Dataset S1 (XLSX)

Dataset S2 (XLSX)

Data, Materials, and Software Availability

FastQ data have been deposited in NCBI (GSE210234) (75). All study data are included in the article and/or SI Appendix.