Multi-omic approach identifies hypoxic tumor-associated myeloid cells that drive immunobiology of high-risk pediatric ependymoma

Summary Ependymoma (EPN) is a devastating childhood brain tumor. Single-cell analyses have illustrated the cellular heterogeneity of EPN tumors, identifying multiple neoplastic cell states including a mesenchymal-differentiated subpopulation which characterizes the PFA1 subtype. Here, we characterize the EPN immune environment, in the context of both tumor subtypes and tumor cell subpopulations using single-cell sequencing (scRNAseq, n = 27), deconvolution of bulk tumor gene expression (n = 299), spatial proteomics (n = 54), and single-cell cytokine release assays (n = 12). We identify eight distinct myeloid-derived subpopulations from which a group of cells, termed hypoxia myeloid cells, demonstrate features of myeloid-derived suppressor cells, including IL6/STAT3 pathway activation and wound healing ontologies. In PFA tumors, hypoxia myeloid cells colocalize with mesenchymal-differentiated cells in necrotic and perivascular niches and secrete IL-8, which we hypothesize amplifies the EPN immunosuppressive microenvironment. This myeloid cell-driven immunosuppression will need to be targeted for immunotherapy to be effective in this difficult-to-cure childhood brain tumor.


Single-cell RNA sequencing identified 8 subpopulations of EPN tumor-infiltrating myeloid cells
The immune microenvironment in ependymoma has historically been inferred through gene signatures identified in bulk transcriptome datasets. 9,14Using a single-cell approach to this problem, we identified 13 transcriptionally unique immune cell subpopulations (Figure S3).These clusters include 8 of myelocytic lineage, 3 of lymphocytic lineage (2 T cell and one B cell), and two unidentifiable clusters (Figure S3).In this study, we focused on myeloid lineages (Figure 1A) as previous studies have suggested a role for the PFA myeloid immune response. 12We hypothesized that targeting myeloid cell interactions will be key to relieving microenvironmental immunosuppressive effects on T cells in PFA tumors.
Using the top 50 genes associated with these myeloid cell subpopulations (Table S2), we classified them as alternative-M2, classical-M1, hypoxia, dendritic cells, microglial, mitotic, neutrophil, and undefined-M (Figure 1A).Similar to our previous findings with neoplastic cells, 17 all five ependymoma subgroups (PFA1, PFA2, PFB, ZFTA-fused, and ST-YAP) contained variable proportions of each subpopulation (Figures 1C and 1D).Gene ontologies from DAVID were used to further characterize each subpopulation (Table S5).Characterizing tumor-infiltrating immune cells in reference to peripheral blood immune cell expression patterns is challenging, as gene expression is heavily skewed by tumor microenvironmental factors.Because of this, many of the markers used to define immune cell lineage are broadly expressed across tumor-infiltrating immune cell subpopulations.For example, CD86, commonly used to define macrophages, is highly expressed in both classical-M1 cells and hypoxia myeloid cells (Figure S4A).Similarly, CSF3R, a gene associated with neutrophils, is highly expressed by the population we defined as neutrophils but also highly expressed in classic-M1 and microglial subpopulations (Figure S4B).Therefore, we classified the myeloid cell subpopulations based on their gene expression characteristics rather than using published immune cell atlases from other anatomic sites.Where lineage was difficult to ascertain (Hypoxia and undefined-M), we defined them under the general classification of myeloid cells.

Microglial
While this subpopulation expressed similar genes to the classical-M1 subpopulation, these cells were distinguished by TMEM119 expression, a signature gene expressed in microglial cells.These cells were abundant in PFA1 tumors but had slightly higher infiltration in PFA2 (Figure 1F).

Classical-M1
The gene signature most enriched within this subpopulation of myeloid cells was antigen processing and presentation (Table S4).These myeloid cells exhibited high expression of MHC class II molecules and Fc-receptor signaling genes (Table S4).Though not significant, these cells were seen in greater abundance in PFA2.

Alternative-M2
These myeloid cells were characterized by top genes MRC1 (CD206) and CD163 (Table S4).Both are signature genes of alternatively polarized myeloid cells.These cells were most abundant in ZFTA-fused tumors and predominating within only two patient samples (UPN 1158 and UPN 1329, Figure 1D).In the bulk tumor gene expression dataset (GSE64415), ZFTA-fused tumors had higher MRC1 expression than either PFA or PFB (Figure S4).

Dendritic cells
The genes associated with this subpopulation were those of dendritic cells, with TCR signaling, antigen processing, T cell activation, and response to interferon-gamma as the predominant signatures (Table S4).

Hypoxia myeloid
This was the only subpopulation that we identified, by single-cell RNA sequencing (scRNAseq), to have differential infiltration between PFA1 and PFA2 molecular subgroups with a 2-fold increase of hypoxia myeloid cells in PFA1 compared to PFA2 tumors (Figure 1F).These cells expressed genes associated with stress response, response to oxygen, angiogenesis, wound healing, cell migration, and neutrophil activation.A characteristic gene of this subpopulation is TREM1.(Table S4).

Neutrophil
While expressing similar genes as the hypoxia subpopulation, this population of cells are clearly neutrophils with high expression of S100 genes as well as neutrophil gene signatures LYZ and FCN1 (Table S4).PFA1 samples had non-significantly greater infiltration of neutrophils than PFA2.ST-ZFTA-fused on average had the most infiltration but this was due primarily to UPN 1158 and UPN 870 (Figure 1D).

Mitotic
This represented a group of highly mitotic cells without a clear lineage distinction (Table S4).

Undefined-M
This cluster of cells did not have a clear lineage distinction.The population expressed genes associated with both classical-M1 and hypoxia subpopulations (Table S4) and was evenly associated with all EPN subgroups in the single-cell analysis (Figure 1E).

Hypoxia myeloid cells are enriched in PFA1 EPN in independent bulk analyses and present in previous spatial transcriptomic analyses
To validate our single-cell findings, we performed Cibersort analyses on publicly available pediatric ependymoma bulk gene expression datasets with associated DNA methylation profiles supporting the diagnosis of PFA, PFB, or ZFTA-fusion-positive EPN (n = 299).This combined dataset contained 199 PFAs, 41 PFBs, and 59 ZFTA-fusion-positive ependymomas.Where PFA subclassification was known, there were 104 PFA1 and 52 PFA2 tumors.The distribution of each of the scRNAseq-derived immune cell subpopulations was similar across the bulk Cibersort datasets (Table S2).Consistent with single-cell data, there was low expression of markers associated with T lymphocyte subpopulations (Figure S6).
The hypoxia myeloid cell phenotype was significantly overexpressed in ZFTA-fused and PFA tumors compared to PFB (p < 0.0001) (Figure 2A).PFA1 tumors were enriched for both microglial and alternative-M2 subpopulations compared to PFA2 (p = 0.0005 and p = 0.016, respectively).The Cibersort bulk gene expression dataset confirmed the enrichment of the hypoxia myeloid cellular subpopulation in PFA1 compared to PFA2 tumors (p < 0.0001), reflective of the pattern seen in the initial scRNAseq analysis.In addition to the identification of hypoxia myeloid and classic-M subpopulations in bulk cibersort analyses, these subpopulations were also seen in our recent PFA ependymoma spatial transcriptomic analyses. 19When compared to their spatial transcriptomic equivalents, the hypoxia myeloid and classic-M single-cell subpopulations had Jaccard indices of 0.22 and 0.16, respectively.Overlap was also seen between the gene ontologies for these subpopulations.

Myeloid cell infiltration is intensified close to areas of altered tumor architecture such as necrosis and vasculature
Single-cell analysis by gene expression profiling or flow cytometry usefully identifies unique cellular subtypes not previously detectable with bulk tumor transcriptomics and methylomics.However, single-cell analyses are not spatially informed.Conversely, spatial transcriptomics provides both spatial and gene expression data, but does not obtain single-cell resolution gene expression data.
The previously published spatial transcriptomic analysis provided initial evidence that ependymoma myeloid cells are located close to borders between epithelial and mesenchymal tumor areas and near to areas of tumor necrosis. 19Given this restricted distribution of myeloid cells at the gene expression level, we sought to establish whether this was recapitulated at the protein level.
The myeloid multiplex immunofluorescence panel, consisting of CD14, CD64, CD3, CD206, HLADR, and TREM1, was applied across 52 PFA EPN sections.Myeloid cells constituted on average between 0.5% and 10% of all cells (Figure 3A).This proportion correlated well with the 5% of cells estimated to be myeloid cells in the Cibersort signature published by Gillen et al. 17 The most highly expressed cell marker was CD14 (median 3.43% of cells) followed by CD64 (median 3.00% of cells).CD3 + cells were rare and less variably expressed (median 0.19% of all cells, range 0.03%-0.94%),consistent with our gene expression data showing low T lymphocyte infiltration and activity in PFA ependymoma (Figure 3A).
All immunophenotypes were present in higher proportions in necrotic and perivascular niches and at regions of transition between densely packed tumor cells and necrosis, validating, at the protein level, our recent spatial transcriptomic analysis as transitions between mesenchymal and epithelial regions 19 (Figures 3B-3E).For each immune cell phenotype, epithelial regions contained significantly smaller proportions of myeloid cell markers (p < 0.0001 for each immune cell category).However, the large number of cells analyzed means that small variations in the sizes of cellular populations could result in highly significant p values.Therefore, fold changes between epithelial and necrotic and perivascular areas for each immune cell marker were calculated to better understand the potential biological significance of these changes (Figure 3B).Myeloid phenotypes with the highest fold changes between densely packed tumor cells and necrotic and perivascular areas were CD206 + /CD14 + /CD64 + (FC.404),CD206 + /CD14 + /CD64 -(FC3.09),and TREM1 + (FC3.62).All phenotypes other than CD14 + , HLADR + , and CD14 + /HLADR + at least doubled in necrotic and perivascular areas, compared to epithelial tumor areas.

Hypoxia myeloid cells localize to necrotic and perivascular niches and are positively correlated with MEC tumor cells
At the mRNA level, the myeloid subpopulations identified by our recent spatial transcriptomic analysis located to specific ''mesenchymal'' regions characterized by the presence of MEC tumor cells. 19Through the myeloid multiplex immunofluorescence panel in this study, we have established that myeloid populations are enriched in areas of necrosis and vasculature and at transition points away from epithelial zones.Additionally, based on our single-cell analysis, hypoxia myeloid cells share several genes with the MEC tumor cell subpopulation.We then hypothesized that the hypoxia myeloid cells gene expression profile would therefore correlate with the MEC tumor profile.All four bulk gene expression datasets analyzed using Cibersort, demonstrated a high correlation coefficient linking hypoxia myeloid and MEC subpopulations (Denver, r = 0.92, 95% CI 0.85-0.95,p < 0.0001, Nottingham, r = 0.83, 95% CI 0.72-0.89,p < 0.0001, Heidelberg, r = 0.82, 95% CI 0.76-0.86,p < 0.0001, St. Jude, r = 0.91, 95% CI 0.84-0.95,p < 0.0001) (Figure S7A).Of note, while the hypoxia myeloid gene expression profile correlates with MEC subpopulations, only 11% of the genes are shared between the two subpopulations indicating they are different subpopulations of cells (Table S6; Figure S7B).
Given the correlation of gene expression by Cibersort and spatial transcriptomics, we hypothesized that the hypoxia myeloid cells would also colocalize at the protein level with MEC tumor cells.We used TREM1 to identify the hypoxia myeloid population.TREM1 is one of the top enriched genes identified in the hypoxia subpopulation with 60.3% of hypoxia myeloid cell expression and only 15% of all other cell expression (logFC 0.768, p value <0.001, pct_in 60.28, pct_out 14.94) (Figure 4A).Furthermore, TREM1 expression was rarely seen in EPN neoplastic subpopulations (https://www.pneuroonccellatlas.org).
The TREM1-positive cells have an ambiguous cytomorphology reminiscent of a monocyte with modest cytoplasm and a mono-lobated nucleus.Immunohistocompatibility (IHC) also showed that TREM1+ myeloid cells are largely localized to the interface of necrosis and viable tissue, most frequently in a perivascular and intravascular distribution (Table S7; Figure 4B).This finding suggests that the TREM1+ cells may be associated with the MEC tumor population, which we have previously described as being enriched in PFA1 tumors and localized to perinecrotic zones.This is supported by parallel IHC analysis of subpopulation-specific markers in the same cohort of PFA EPN which showed the highest TREM1 correlation was with CA9, a marker of MEC (r2 = 0.92, p < 0.001, n = 49)(Figure S7).Given the correlation between TREM1+ cells and MEC subpopulation marker, we performed multi-analyte immunofluorescent imaging of FFPE slides to examine cellular spatial relationships.We used previously described EPN tumor subpopulation markers, CA9 (MEC), CAPS (CEC), Fos (UEC), and VIPR2 (TEC), along with TREM1 and DAPI.We scanned 5 histologically distinct regions for each patient sample (Denver cohort n = 30, Nottingham cohort n = 24) on the whole slide image and segmented each region into tumor, necrosis, or blood vessel.The training algorithm gave a 93% accuracy for identifying tissue segmentation in the 15-image training set.For the tissue segmentation, we aligned each region with the corresponding H&E region and a pathologist (N.W.) identified each feature within that region (Figure S1).VIPR2 was excluded from the cell phenotype analysis as it was ubiquitously stained throughout the slide such that it was difficult to identify true staining.The antibodies selected for this panel are specific for each cell type and therefore cell phenotyping was performed on the single antibody-positive cells (Figure S1).Consistent with our IHC results, TREM1+ cells were enriched in necrosis and blood vessel regions (30.8% and 32.9%, respectively) (Figures 4C and 4D).CA9+ MEC cells were the most abundant cell population in the necrosis regions (36.6%) and were also enriched in the tumor regions on the borders of necrosis.Most of the CA9+ MEC cells were within 50 mm of TREM1+ cells in the necrosis regions (Figure 4E).Additionally, there was an increase in CD14 + myeloid cells in the blood vessel regions.Unsurprisingly, CAPS+ cells were enriched in epithelial regions of tumor (29.9%).Interestingly, the CAPS+, CEC, cells were similar distances from the TREM1+ cells as were the MEC cells suggesting the hypoxia myeloid cells may facilitate transforming the CEC cells to MEC phenotype.C-Fos+ cells were sparse and equally distributed between the different spatial regions, consistent with the role of UEC as a progenitor cell of both epithelial and mesenchymal regions in our spatial transcriptomic study. 19poxia myeloid cells have an immune-suppressive phenotype We have previously shown that PFA1 EPN harbors an immune-suppressive phenotype which we hypothesize leads to multiple recurrences and poor overall survival. 9Given the hypoxia myeloid cells are enriched in PFA1 tumors, we next focused on describing these cells in greater detail.In our prior bulk microarray analysis, we showed enrichment of the IL-6/STAT3 pathway in EPN tumor flow-sorted myeloid cells. 12To determine whether this held in the scRNAseq, we performed a regulatory network inference analysis that infers the transcription factor activity at the single-cell level.Consistent with our prior findings, STAT3 was among the top transcription factor pathways upregulated in the hypoxia myeloid compartment (Table S8).We also found that retinoic acid receptor signaling was upregulated.This is of interest as we found the tretinoins to be highly effective at promoting PFA1 tumor cell death in vitro and are actively pursuing this class of chemotherapy agents as a potential maintenance regimen for high-risk EPN patients. 20poxia myeloid cells secrete immune-suppressive cytokine IL-8 We have previously reported the significance of cytokine signaling in PFA EPN immunobiology.Tumor-secreted interleukin-6 (IL-6) induces STAT3 signaling in infiltrating monocytes which results in secretion of IL-8. 12Furthermore, the IL-8-producing myeloid cells can further amplify the immune-suppressive response by polarizing naive monocytes to a pro-tumor phenotype.We therefore hypothesized that hypoxia myeloid cells are one of the myeloid subpopulations involved in this process.
To determine the cytokine profiles of infiltrating EPN myeloid cells, we utilized the IsoLight and IsoSpark platforms that can measure cytokine and chemokines secretion at the single-cell level.We isolated CD45 + cells from PFA single-cell suspensions (6 PFA1, 5 PFA2) using magnetic bead isolation.CD45 + cells were incubated with lipopolysaccharide for 24 h to enhance immune functional characteristics, and then loaded onto the Innate Immune Single-cell Secretome chip (IsoPlexis).Similar to scRNAseq data, we detected different subpopulations of cells based on cytokine/chemokine secretion profiles.Samples were pooled based on methylation phenotype and analysis results were clustered based on the polyfunctional strength index (Figure 5A), annotated from IsoSpeak software.PFA1 samples were enriched for a distinct cluster of IL-8-producing cells and an additional cluster secreting IL-8 in combination with macrophage inflammatory protein-1 alpha (MIP-1a) and MIP-1b (Figures 5A and 5B).To make certain these cytokine release profiles were not dependent of LPS stimulation, we validated these findings in a multiplex cytokine release assay using media supernatant collected from unstimulated CD45 + CD11b+ myeloid cells flow sorted from single-cell suspensions (Figure S8A).This experiment showed that cytokine secretion observed by the LPS-stimulated single-cell data is not an LPS-mediated response.
Interestingly, samples that had high infiltration of hypoxia myeloid cells by scRNAseq had subsets of myeloid cells secreting MIP-1b, MIP-1a, and IL-8 either individually or in combination (Figure 5C).MIP-1a and MIP-1b are both chemoattractants and are encoded by CCL3 and CCL4 genes.Both genes are significantly enriched in a subset of the microglial subpopulation, being detected in 81% and 54% of cells, respectively (p < 0.001).CCL3 was also enriched in the M1 subpopulation with 60% of cells expressing CCL3 (log FC 0.64, p < 0.001) (Figure 5C).Additionally, CCL3 and CCL4 were highly enriched in the chemotactic-M spot clusters identified in our prior spatial transcriptomics publication. 19These chemotactic-M spots were located along the borders of necrosis where epithelial-to-mesenchymal transition was occurring.
IL-8 (CXCL8) was found in 42.5% of hypoxia myeloid subpopulation and 31.4% of the microglial subpopulation (p < 0.001).Evaluating the spatial orientation of IL-8 expression within the tumor microenvironment, we utilized the spatial transcriptomics dataset from our previous publication. 19IL-8 (CXCL8) gene expression was enriched in the hypoxia myeloid subpopulation and MEC-D spots (Figure 5D).MEC-D is denoted as being mesenchymal ependymal tumor cells; however, when we overlayed TREM1 expression, we found most of the gene expression located in the MEC spot clusters with similar expression pattern as IL-8, suggesting intermixing of hypoxia_M with MEC-D cells.This is a limitation of spatial transcriptomics; the capture spots are 50 mm in diameter and contain 20-30 cells.Nearest neighbor analysis from spatial proteomics found the MEC cells (CA9+) were an average of 45 mm from TREM1+ cells (Figure 4E).This suggests the IL-8 and TREM1 gene expression seen in the MEC spots is from the proximity of hypoxia myeloid cells to MEC tumor cells in the regions of necrosis.Collectively, spatial proteomics and transcriptomics lead to the hypothesis that microglial cells are responding to tumor necrosis by recruiting additional myeloid cells to the regions between densely packed tumor cells and necrosis.
In PFA tumors with low hypoxia myeloid infiltration, cytokine profiles were more antitumor, characterized by higher production of GM-CSF and IL-18 (Figure 5A).These cells are functionally distinct from the IL-8, MIP-1a, and MIP-1b-producing cells and are enriched in PFA2 (Figure 5B).IL18 gene expression was enriched in the microglial and M1 subpopulations (Figure 5B) which are also both slightly higher in PFA2.
Taken together with our previous findings in PFA immunobiology, these data provide evidence that hypoxia myeloid cells significantly contribute to the generation of the EPN immune-suppressive environment.Our studies on the PFA ependymoma myeloid environment give a new insight into key tumor and immune interactions and a hypothetical mechanism of immune evasion in PFA.

DISCUSSION
Enhanced understanding of the biology of posterior fossa EPN is critical for the development of novel rational treatments.We utilized scRNAseq and spatial proteomics to generate deeper insight into the tumor immune microenvironment (TME).We have identified eight transcriptionally discrete subpopulations of infiltrating myeloid cells and have described the spatial distribution of myeloid cells in PFA ependymoma.A subpopulation of hypoxia myeloid cells enriched in PFA1, with characteristics of MDSC, was associated with an immune suppressive TME.][23][24][25][26] Previous work from our group 17 and others 18 identified multiple unique neoplastic EPN subpopulations.Of these, MEC were most strongly associated with aggressive PFA1 tumors.We hypothesized a pro-tumor, immunosuppressive link between MEC and hypoxia myeloid cells within the TME.This hypothesis is supported by the strong correlation between MEC and hypoxia myeloid phenotypes across four independent ependymoma gene expression cohorts, spatial proteomic analyses using both single-stain immunohistochemistry and multiplex immunofluorescence in our study, alongside previously reported spatial transcriptomic data. 19Further in vitro and in vivo studies are now required to test these findings in the process of translation to the clinic.
27,28 Our single-cell approach improves the resolution of the previous gene expression findings.We previously showed the same cellular subtypes appear across all molecular entities of EPN and assignment to specific molecular subgroups is directly related to the proportion of cellular subpopulations in the entire tumor. 17Our study is consistent with these findings, showing both PFA subgroups exhibit all the myeloid subpopulations in variable proportions.The infiltration of the hypoxia myeloid cells provides an explanation for the immunobiology previously identified in PFA1. 9,17Our findings are consistent with a scRNAseq study on adult spinal EPN that identified high levels of intratumor heterogeneity and highlighted the importance of tumor-associated macrophages, particularly related to driving inflammation and angiogenesis, 29 both functions strongly associated with PFA 3,9,27,28 Hypoxia myeloid cells exhibit an MDSC phenotype with features including activation of STAT3 pathway, likely in response to tumorsecreted IL-6.Hypoxia-M cells also have a strong IL-8 (CXCL8) gene signature which provides support for the hypothesis that IL-8-secreting cells, identified using single-cell cytokine release assays, are hypoxia myeloid cells.We previously described that IL-8 was the only cytokine significantly upregulated in monocytes cultured in PFA1 conditioned media and could be attenuated by blocking IL-6.Furthermore, the IL-8 secretion, from polarized monocytes, downregulated HLA-DR and CD64 while upregulating immune-suppressive cytokines. 12These data would suggest the polarized monocytes are the hypoxia myeloid cells and further studies are needed for validation of this theory.
Interestingly, IL6 gene expression is exclusively enriched in the MEC subpopulation, suggesting IL-6 is facilitating the transition of the infiltrating myeloid cells to the hypoxia phenotype.This hypothesis is consistent with known patterns of monocytic-MDSC recruitment and development where expansion of the MDSC population is achieved through chronic inflammation triggering with STAT3 responses followed by activation of recruited cells via inflammatory cytokines including IL-6. 30The transcriptional profile of the hypoxia myeloid cells is consistent with this explanation, as is the corralling of the immune cells away from epithelial tumor regions into the mesenchymal zones seen in our proteomic analyses.Our spatial profiling confirms that the hypoxia myeloid cells co-localize with MEC in regions of necrosis and perivascular niches. 19Hypoxia is critical for the maintenance of PFA EPN, likely because of its developmental origins. 31However, our study is the first to propose a mechanistic link between hypoxia and development of an immunosuppressive TME driven through areas of necrosis.While the hypoxia myeloid cells are critical cellular components of PFA1 EPN, other myeloid phenotypes may exert other functions as they corral around necrotic and vascular areas and this is an area requiring more detailed investigation.
We have identified a process in PFA EPN, which, if targeted through therapeutic modulation of the immune environment, may potentiate the efficacy of promising cellular therapies such as CAR-T cells. 32These cellular therapies are urgently needed to deliver better outcomes for children with this devastating disease but, without designs incorporating detailed knowledge of the underlying immunobiologic processes, are at risk of failure.Children with particularly high-risk ependymoma, for example those with tumors harboring chromosome 1q gain and 6q loss for which conventional therapies will provide little benefit, need to be considered as early candidates for new agents based on our evolving biological knowledge. 33Considering this study, and our prior work, putative agents include the IL-6 receptor antagonist tocilizumab, IL-8 antagonists, and all-trans retinoic acid.
The interactions we have proposed between hypoxia myeloid cells and MECs provides both a rational pathway to target and a route for further investigation.Addressing the immunosuppressive environment generated by tumor-hypoxia-myeloid cell interactions will contribute to better immunotherapeutic approaches and may also serve as an adjunct to potentiate the efficacy of current standard-of-care treatments. 34

Limitations of the study
This is study is largely descriptive using a multi-omics approach to characterize ependymoma-infiltrating myeloid cells.This study generated new hypotheses regarding the cellular interactions and cellular function within the complex tumor microenvironment.In vitro and in vivo experiments modeling these cell-cell interactions are necessary and will be the subject of future studies.The mechanism driving the development of hypoxia myeloid cells remains to be determined although prior work would suggest that it is through mesenchymal EPN tumor cell secretion of IL-6 and hypoxic conditions within the regions of necrosis.Future in vitro co-culture studies will be needed to validate this mechanism.

STAR+METHODS
Detailed methods are provided in the online version of this paper and include the following:

Figure 1 .
Figure 1.scRNAseq analysis of immune subpopulations in EPN reveals 8 unique subpopulations of myeloid cells We sequenced 26 viably frozen, single-cell suspensions, of pediatric ependymoma with a minimum of 2000 cells per samples.(A) UMAP clustering of myeloid cells identified 8 transcriptionally unique subpopulations of cells.Populations were named based on the gene expression profiles (Alternative-M2, Classical-M1, Hypoxia, Dendritic Cells (DC), Microglial, Mitotic, Neutrophil, and Unknown M). (B) Heatmap of the top gene signatures identified in each myeloid subpopulation.Gene ontologies, from DAVID, enriched in each subpopulation listed.(C) UMAP projections of myeloid subpopulation clusters with pediatric molecular subgroups (PFA1: n = 13, PFA2: n = 7, PFB: n = 1, RELA: n = 5, YAP: n = 1).(D) Proportion of myeloid cells type in each patient sample.Quantification of proportion of myeloid cells within each subpopulation compared across ependymoma molecular subgroup.Samples with at least 50 cells were included in the analysis.Value above each comparison is p value calculated between subgroups.(E) Proportion of myeloid cells of each subpopulation identified in each molecular subgroup in scRNAseq analysis.(F) Proportion of myeloid cells of each subpopulation infiltrating PFA1 versus PFA2 ependymoma in scRNAseq analysis.

Figure 3 .
Figure 3.Immune cells corral around necrotic regions and perivascular niches (A and B) each dot is representative of a single cell identified with that phenotype.(A) Myeloid phenotypes normalized by total number of cells indicated that myeloid cells constitute approximately 0.5%-5% of all cell phenotypes in the tumor sections analyzed all though some samples express significantly more myeloid markers.(B) All myeloid panel phenotypes were associated with greater levels of expression in regions of tumor compared with necrotic or perivascular niches (p < 0.0001, chi-square test).In view of the large numbers of cells included in this analysis, the bars for each phenotype indicate the fold change to give an insight into the possible biological significance of each difference.The presence of more immune cells outside of the main tumor parenchyma provides supporting evidence for ependymoma as either an immune desert or immune-excluded type tumor.

Figure 3 .
Figure 3. Continued (C) Area of transition between tumor and necrosis characterized by multiple CD206, CD14, HLADR, and TREM1-positive cells.(D) Area of central necrosis surrounded by a ring of myeloid cells staining positive for CD14, HLADR, and CD64.(E) Blood vessel surrounded by multiple myeloid related cells staining positive for multiple myeloid markers.Note that the tumor beyond the myeloid cell ring is largely devoid of positively staining cells.Images captured using 20X objective and scale bar is 50 mm.

Figure 4 .
Figure 4. Hypoxia myeloid cells are localized to regions of necrosis (A) Trem1 gene expression distribution in single-cell RNAseq myeloid subpopulations.(B) Representative images of Trem1 antibody stain on immunohistochemistry FFPE slides of PFA ependymoma.Images captured using 20X, 40X, and 100X objectives and scale bars are denoted 50 mm, 20 mm, and 10 mm, respectively.(C) Representative multi-analyte immunofluorescent images of FFPE PFA samples.Area of necrosis denoted with blue arrow and a blood vessel denoted with red arrow.TREM1+ cells are indicated with green arrows.Pink fluorophore is CA9 antibody staining.Single-stain immunohistochemistry images for the same region.DAPI is nuclear staining, CA9+ are MEC tumor cells, and TREM1+ are hypoxia myeloid cells.Images captured using 20X objective and scale bar is 50 mm.(D) Quantification of cell phenotypes stratified by tissue segmentation on 59 (n = 30 Denver and n = 29 Nottingham) FFPE slides with 5 regions per slides selected for analysis.Mean and 95% confidence interval denoted by bars.*** denotes p value <0.001.* denotes p value <0.05.(E) Nearest neighbor analysis calculating the distance between TREM1+ hypoxia myeloid cells, tumor cells, and other myeloid cells stratified by tissue segmentation.Mean and 95% confidence interval denoted by bars.*** denotes p value <0.001.* denotes p value <0.05.

Figure 5 .
Figure 5. Functional phenotyping of single-cell subpopulations Monocytes were isolated from 6 PFA1 and 5 PFA2 samples using magnetic bead separation.(A) 2D t-SNE (2D t-distributed stochastic neighbor embedding) analysis indicates greater signal intensity of multiple cytokines co-secreted by PFA1 monocytes than PFA2 monocytes.(B) PAT-PCA (polyfunctional activation topology-principal component analysis) shows that 9 subpopulations of monocytes were identified based on polyfunctional cytokine secretion of MIP-1a, MIP-1b, GM-CSF, IL-18, and IL-8.PFA1 exhibits higher polyfunctional heterogeneity at single-cell level as PFA2 mainly secretes GM-CSF or IL-18.Density and size of circle indicate number of cells in each subpopulation.Heatmaps compare the percentage of single cells secreting various monofunctional and polyfunctional groups across multiple samples.Secretions in PFA1 display as both more polyfunctional and with higher secretion frequency than PFA2.For example, the combined secretions of MIP-1a and MIP-1b, and MIP-1b and IL-8 are unique to PFA1 compared to PFA2.(C) Single-cell RNAseq UMAPs of CCL3 (MIP-1a), CCL4 (MIP-1b), and CXCL8 (IL-8) gene expression of ependymoma tumor-infiltrating myeloid cells.CCL3 (MIP-1a) expression is enriched in microglia, M1 and some M2 myeloid cells.CCL4 (MIP-1b) gene expression is localized to a portion of microglial cells.CXCL8 is enriched in hypoxia myeloid cells and CCL4 expression microglial cells.(D) Representative spatial transcriptomics slides.Left: H&E stain of PFA1 ependymoma tumor sample.Red arrow denotes region of necrosis and white arrow denotes a region of hypercellularity with prominent perivascular pseudorosettes.Middle Left: H&E image overlayed with spatial transcriptomics spot clusters.Middle Right: Spatial gene expression of CXCL8 (IL-8).Right: Spatial gene expression of TREM1.Scale bar: 5 mm.

TABLE STAR+METHODS KEY RESOURCES TABLE
Center https://www.bioconductor.org(Continued on next page)