Loss of LDOC1 by chromatin compaction in mesenchymal tumor cells is required for PFA1 ependymoma growth

Abstract

Background

Posterior fossa molecular subtype A (PFA) ependymoma occurs in young children and is the deadliest subtype of pediatric ependymoma. High-risk subtypes with chromosome 1q + and/or 6q- exhibit significantly poorer outcomes compared to wild-type PFA. However, 50% of wild-type PFA patients relapse and there is a high risk of gaining chromosome 1q at recurrence. We previously found constitutively active NF-κB, through loss of LDOC1, led to chronic IL-6 secretion and an overall immunosuppressive tumor microenvironment in the higher-risk wild-type PFA ependymoma subset (PFA1).

Methods

In this study, we delineate the mechanistic consequences of LDOC1 loss in PFA1, using our PFA ependymoma in vitro and in vivo models under normoxia and hypoxia conditions.

Results

We noted chromatin compaction by H3K27me3 at the LDOC1 loci results in loss of LDOC1 gene expression. Restoration of LDOC1 was sufficient to reduce proliferation, NF-κB signaling, and a significant decrease in IL-6 secretion. Furthermore, tumors implanted with LDOC1-transduced cells in vivo were out competed by non-transduced cells, suggesting loss of LDOC1 is required for PFA tumor growth.

Conclusion

These findings shed further light on the biology of PFA1 ependymoma and the role LDOC1 loss has on the tumor and immunobiology of high-risk pediatric ependymoma.

Graphical Abstract

Open in new tabDownload slide

Methylation of histones leads to chromatin compaction (CC) in PFA1 tumor. CC leads to silencing of LDOC1, which allows for constitutively active NF-κB signaling and chronic IL-6 secretion. IL-6 reprograms infiltrating myeloid cells to a myeloid-derived suppressor cell (MDSC) phenotype. IL-8 from MDSC perpetuates the inflammatory myeloid signaling by down regulating MHC and Fc-receptors and upregulating IL-10, IL-1β, and IL-13 and T-cell dysfunction through up regulation of TIM3.? denotes the remaining questions in the model.

Posterior fossa ependymoma type A (PFA) is the deadliest and most common ependymoma subgroup in children, occurring most frequently in very young patients. Despite extensive molecular characterization using genomics and transcriptomics, no specific mutational driver has been identified. High-risk chromosomal abnormalities including gain of chromosome 1q (1q+) and/or loss of chromosome 6q (6q-) are present in 25% of newly diagnosed PFA ependymomas.¹ Gross total resection followed by radiation is effective only in 50% of children, with the risk of gaining 1q+/6q- at recurrence increasing to 50%.¹ Molecularly, PFA ependymomas are divided into PFA1 and PFA2 subtypes.² In general, patients with PFA1 are more likely to recur and time to recurrence is often shorter than PFA2 patients.

We have previously reported distinct immune signature profiles between PFA1 and PFA2.³ PFA1 tumors are defined by an overall immunosuppressive gene signature and PFA2 have a more antiviral signature.³ Furthermore, we showed the immunosuppressive gene signature in PFA1 was driven by chronic secretion of IL-6 which polarizes the tumor-infiltrating immune cells.⁴ Chronic IL-6 secretion was a result of constitutively active NF-κB signaling in PFA1 tumors.⁵ Recent single-cell and spatial transcriptomics data have shown that PFA tumor biology and transcriptomic subgrouping is impacted by the relative proportion of different tumor cell subpopulations.^6,7 PFA1 tumors are enriched with mesenchymal ependymal cells (MEC) and undifferentiated ependymal cells (UEC). In contrast, PFA2 tumors have greater enrichment of differentiated tumor subpopulations such as transporter ependymal cells (TEC) and ciliated ependymal cells (CEC).^6,7 Spatially, the MECs were primarily located within and immediately surrounding regions of tumor necrosis. The presence of MECs was associated with poorer outcome and myeloid-derived suppressor cells, harboring characteristics of the IL-6 polarized myeloid cells.⁸

In a previous study, we identified LDOC1 as the only gene silenced in PFA1 compared to PFA2.⁵ LDOC1 is a tumor suppressor that negatively regulates NF-κB signaling.^9–14 Epigenetic silencing of LDOC1 by cigarette smoke exposure was shown to be involved in oral neoplastic transformation,¹⁴ while decreased LDOC1 expression is associated with poor outcomes in patients with chronic lymphocytic leukemia.¹⁵ We also found that loss of LDOC1 gene expression was associated with poor progression-free survival (PFS) with a median of 24 months compared to 48 months for PFA tumors with high expression of LDOC1 (HR: 2.7, 95%CI: 1.046–7.306).⁵ In the limited studies that have evaluated LDOC1, loss of gene expression was shown by hypermethylation of the promoter region of LDOC1.^9–14 We demonstrated that LDOC1 gene and protein expression can be restored using epigenetic modifiers such as DNA methyltransferase and histone deacetylase inhibitors, suggesting epigenetic factors contribute to LDOC1 loss in PFA ependymoma.⁵ However, preliminary DNA methylation data showed methylation beta values were similar between PFA tumors with high LDOC1 expression and PFA tumors with low LDOC1 expression.

Here, we sought to identify the mechanism driving the loss of LDOC1 and the resulting consequence in PFA tumor biology. Our findings highlight the loss of LDOC1 as an essential event in PFA tumor development, providing new insights into potential therapeutic targets for this aggressive childhood cancer.

Materials and Methods

Study Approval

Primary patient tumor and normal brain samples were obtained from Children’s Hospital Colorado and collected in accordance with local and federal human research protection guidelines and institutional review board regulations (COMIRB 95-500). Informed consent was obtained for all specimens collected. Patient information is given in Supplementary Table S1.

Visium Spatial Transcriptomics

To evaluate the spatial transcriptomics of LDOC1 in posterior fossa ependymoma, we utilized our dataset consisting of 14 pediatric PFAs (GSE195661).⁶ The data were analyzed using Loupe Browser v8 (10X Genomics) with Harmony-clustered subpopulations and in individual samples. CA9 and CAPS expression were used as markers for MECs and CECs, respectively.

Cell Lines and Hypoxia Stimulation

EPN cell lines, MAF-811 and MAF-928, were established from recurrent pediatric PFA that have been well characterized and cultured as previously described.^4,5,16 Briefly, the cells were cultured as adherent monolayers in Opti-MEM supplemented with 15% fetal bovine serum and 1% pen/strep (O15).¹⁶ For neurosphere growth assays, both cell lines were cultured in ultra-low attachment plates in O15 for 14 days. Normoxia (20% O₂) and hypoxia (2% O₂) assays were conducted for 3 days. Pharmaceutical-induced hypoxia was performed using deferoxamine (DFO, 100 μM, Cat. #D9533; Sigma-Aldrich) and cobalt chloride (CoCl2, 100 μM, Cat. #232696; Sigma-Aldrich) for 24 h. Detailed information is given in Supplementary Material and Methods. HEK 293 FT cells ([RRID:CVCL_6911], ATCC) were cultured as previously described.⁵

PFA Neoplastic Subpopulation Deconvolution

Bulk tumor tissue gene expression from 40 PFAs (24 PFA1, 16 PFA2, 2 PFB, 3 EPN-RELA, and 1 EPN-YAP1) (Affymetrix array GSE50385), was analyzed to identify genes correlated with LDOC1 expression (Pearson’s R). The top 100 positively and negatively correlated genes were deconvoluted to estimate the relative proportion of PFA neoplastic subpopulations as previously described.⁷

RNA Extraction and Quantitative Real-time PCR

Total RNA was isolated from cells using DNA/RNA All Prep Kit (Cat. #80204, Qiagen) and analyzed for gene expression using qRT-PCR performed on a StepOne Plus Real-Time PCR System with TaqMan Gene Assay Reagents (Applied Biosystems), according to the manufacturer's protocols. cDNA was synthesized from total RNA using a High-Capacity cDNA Kit (Applied Biosystems). The resulting cDNA was loaded at 25ng per well for qRT-PCR (4 replicates), with TaqMan gene-specific probe for LDOC1 (Hs00273392_s1), CA9 (Hs00154208_m1), LGR5 (Hs00969422_m1), COL9A2 (Hs00895570_m1), and 18S (Hs03003631_g1). The relative gene expression was calculated using the 2^-ΔΔCT method¹⁷

Transcriptomic Analysis

RNA of MAF-811 and MAF-928 cells, under normoxia or hypoxia for 3 days, was isolated from cells and sequenced using the Illumina Novaseq6000. Sequencing data were aligned using STAR (RRID:SCR_004463) and quantification of RNA expression as fragments per kilobase per million and derived by Cufflinks (RRID:SCR_014597). Analysis performed using Pluto (https://pluto.bio). Detailed information is given in Supplementary Material and Methods.

Immunofluorescence

MAF-811 and MAF-928 were plated on poly-D-lysine coated chamber slides (Corning, 354632). After treatment of normoxia or induced hypoxia, cells were fixed in 4% paraformaldehyde. Fluorescence intensities of LDOC1 were analyzed and images were captured at 40× magnification using Keyence Microscope. Detailed information is given in Supplementary Material and Methods.

DNA Methylation

Following incubation in normoxia and hypoxia conditions, DNA was isolated from cells using DNA/RNA All Prep Kit (Cat. #80204, Qiagen). DNA methylation was assessed using the Infinium® MethylationEPIC BeadChip (Illumina). Analysis was performed as previously described and detailed in Supplementary Material and Methods.⁵

Western Blot

Western blots were performed as previously described.^4,5 Primary antibodies used were anti-LDOC1 (Abcam, ab86126), anti-RELA (Cell Signaling, 8242, clone D14E12), anti-NF-κB2 p100/p52 (Cell Signaling, 4882), anti-H3K27Ac (Abcam, ab4729), and anti-H3K27me3 (Abcam, ab6002) variants. Densitometry measurements were performed using ImageJ software. Detailed information is given in Supplementary Material and Methods.

Cut&Run Protocol

A total of 500,000 cells per reaction were harvested and captured using pre-activated ConA beads (EpiCypher). Beads with attached cells were incubated overnight on a nutator at 4°C with 50 µL of antibody buffer (wash buffer supplemented with 0.1% digitonin and 2 mM EDTA) H3K27me3 (MA5–11198, Thermo Fisher), and a negative control, IgG (EpiCypher, 13–0042K). Detailed methods for DNA isolation are located in Supplementary Materials and Methods. Libraries were prepared with the NEBNext Ultra II DNA kit, according to the manufacturer's instructions (New England Biolabs). Final libraries were size-selected (200–600 bp) and analyzed on Bioanalyzer High Sensitivity DNA chips (Agilent) to confirm 200–400 bp fragment size range. Paired-end 150 bp sequencing of pooled barcoded libraries was carried out on the Illumina Novaseq 6000 platform by the Novogene (CA) facility or Functional Genomics Core (CU Anschutz).

Cut&Run Analysis

Paired-end FASTQ files were processed using the nf-core-cutandrun pipeline (v3.1)^18–20 performed using Pluto (https://pluto.bio). Full analysis details are in Supplementary Materials and Methods. PFA H3K27 acetylation and trimethylation data was generously provided by Dr. Lukas Chavez, under the accession code: EGAS00001002696.²¹

Chromatin Accessibility Assay

PCR primers were designed using the NCBI Primer Blast program (https://www.ncbi.nlm.nih.gov/tools/primer-blast/). LDOC1 coding sequence was used as the PCR template. Primers from IDT (https://www.idtdna.com/pages) were designed to flank the transcription start site (TSS) of LDOC1 (F: AACTAGGAGCTAGGGGAGCC, R: GGTGTAGACAAAGAGGGCGT). PFA patients with low and high expressing LDOC1 were selected from our dataset of 40 pediatric ependymoma. Low LDOC1 expression was defined as log2 gene expression below a threshold of 4 and high LDOC1 expression was defined as log2 gene expression above 4. Chromatin was isolated per manufacture’s protocol (Abcam ab185901) on n = 7 low expressing (average log2 LDOC1 = 2.96 SD = 0.88) and n = 8 high expressing (average log2 LDOC1 = 7.14, SD = 1.06) snap tumor specimens. Full protocol and analysis can be found in Supplementary Materials and Methods. Chi2 contingency analysis was used to compare open and closed between the low and high-expressing LDOC1 tumor samples.

NF-κB Proteome profiler

Changes in NF-κB pathway activation between normoxia and hypoxia-treated PFA cell lines were evaluated using the Proteome Profiler Human NF-κB Pathway Array Kit (R&D Systems: ARY029) per manufacturer's protocol. Quantification was performed as previously described^22,23 (Supplementary Table S3). Detailed information is found in Supplementary Material and Methods.

Cytokine Release Assay

Media supernatant was collected from PFA cell lines following 3 days of incubation in normoxia or hypoxia conditions. Media was loaded onto CodePlex Innate Immune Secretome Panel (Bruker Cellular Analysis, CODEPLEX-2L03) and run on IsoSpark per the manufacturer's protocol. Cytokine capture was analyzed using IsoSpeak software (https://brukercellularanalysis.com/products/isospeak-software/).

LDOC1 Genetic Knock-in

LDOC1 expression vector was designed and purchased through GeneCopia (Supplementary Figure S1A) for in vitro studies and Vector Builder (Supplementary Figure S1B) for in vivo. The LDOC1 vector (hLDOC1) and corresponding control vector (Ctl) were transfected into HEK293FT cells with psPax and P.M2DG packaging plasmids and TransIT-LT1 Reagent (Mirus Bio). For in vitro assays, viral supernatant was collected in O15, and for in vivo experiments in Opti-MEM with 5% FBS and 1% pen/strep. Cells used in in vitro experiments were transduced with viral supernatant using 8 μg/mL Polybrene and selected with 0.8 μg/mL puromycin and/or GFP expression. LDOC1 expression was validated with qRT-PCR and Western blot.

Cell Viability Assays

MAF-811 and MAF-928 Ctl and hLDOC1 cells were cultured in normoxia and hypoxia conditions for 5 days and cell viability was measured using CellTiter-Glo (Promega) following the manufacturer's protocol. The optical density of each well was measured using CellTiter-Glo protocol template on GloMax Explorer (RRID: SCR_015575, Promega).

3H-thymidine Uptake

Cell proliferation in Ctl and hLDOC1 cells as determined by the rate of DNA synthesis was measured by 3H-thymidine incorporation after 5 days in normoxia or hypoxia. Detailed information in Supplementary Materials and Methods.

Neurospheres Growth Assay

Cells were plated at 1000 cells per well in O15 on a round bottom ultra-low attachment plate (Corning, 7007). Neurosphere growth was monitored on the IncuCyte S3 with real-time images of wells every 12 h for 10 days. The area was calculated using a quantification processing definition on the IncuCyte S3 (Sartorius).

In Vivo Tumor Growth Assays

We have previously published the characteristics of MAF-811 and MAF-928 patient-derived xenograft models.²⁴ All mouse work was approved by the Colorado University Anschutz Institutional Animal Care and Use Committee (IACUC 00152). For this study, we utilized subcutaneous flank injections as these models are easier to monitor tumor growth. Xenograft cells were centrifuged with hLDOC1-Luc or control (Ctl-Luc) viral supernatant (Supplementary Figure S1B). Cells were washed twice in serum-free Opti-Mem before being resuspended at 250,000 cells per 100 μL and injected into the flanks of NSG mice. The equal number of males and females were used for each construct. Mice were monitored daily, and weights were documented twice weekly. Bioluminescence imaging was performed as previously described.²⁴ Once any tumor reached a size of 2 cm², the entire experiment was euthanized, and tumors were harvested. LDOC1 and Ki67 expression was measured by qRT-PCR and immunohistochemistry (IHC). Blinded IHC scoring was performed (Supplementary Table S4). Detailed methods are in Supplementary Materials and Methods.

Statistical Analyses

Statistical analyses were performed using R bioinformatics, Pluto Bio, and Prism GraphPad v10. Analysis included the Student’s t-test or 1-way ANOVA test, followed by Bonferroni´s test. Differences in gene expression were shown by Fold Change (FC). P < 0.05 was statistically significant. Functional assays were reported as mean ± SE of 3 independent experiments in triplicate. All experiments were done in triplicate.

Data Availability

Transcriptomic data have been deposited in the National Center for Biotechnology Information Gene Expression Omnibus (GEO) database and are publicly accessible through GEO accession numbers GSE274900 and GSE275007. https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE274900

https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE275007

Results

LDOC1 Loss Is Correlated with Mesenchymal Ependymal Cells

Spatial transcriptomics allows us to visualize gene expression within tumor architecture for a deeper understanding of how critical genes associate with the structure of tumors. We and others have shown regions of epithelial to mesenchymal transitions (EMT) that correlate with worse prognosis of posterior fossa ependymoma.^6,7,25 The mesenchymal regions contain multiple subtypes of cells (MEC-A, MEC-B, MEC-C, and MEC-D) that have distinct functions. MEC-A for example has more genes that predict interactions with immune cells to induce proliferation of undifferentiated cells (UEC) and in contrast, MEC-D appears to be more associated with the hypoxic nature of necrotic regions. We previously reported the MEC-B subpopulation is defined by genes related to NF-κB signaling and represents late EMT stages. Given LDOC1 is a negative regulator of NF-κB signaling, we hypothesized the loss of transcription would be associated with the MEC-B subpopulation. LDOC1 has the highest expression in the epithelial cells and normal cerebellum granular cells (Figure 1A). When these subclusters are aligned with tumor architecture, we can see a band of LDOC1-expressing cells along the border of EMT zone (Figure 1B). Additionally, there is a significant difference in LDOC1 expression between the epithelial subpopulations and mesenchymal subpopulations (P = .031), with the lowest expression as predicted in MEC-B subpopulation (Figure 1C). We curated a gene set associated with the loss of LDOC1 and analyzed the correlation to the main ependymal cell subtypes. The loss of LDOC1 was significantly associated with MEC cells, UEC-A, and to some extent UEC-B. In contrast, genes associated with LDOC1 expression were significantly correlated with ciliated ependymal cells (CEC) (Figure 1D).

Tumor necrosis is associated with hypoxic tumor conditions. In our previous single-cell RNA-seq analysis of PFA, the MEC subpopulation were defined by genes associated with hypoxia.⁷ To determine whether LDOC1 expression is associated with hypoxia, we exposed PFA tumor cells to 2% oxygen conditions for 3 days and measured LDOC1 gene and protein expression. LDOC1 gene expression (Figure 1E) was significantly reduced which also led to reduction of LDOC1 protein (Figure 1F). Chemically inducing hypoxia with CoCl₂ or desferrioxamine (DFO) also resulted in a significant loss of LDOC1 gene expression (Figure 1G).

Chromatin Compaction Results in Loss of LDOC1 Gene Expression

LDOC1 silencing has been reported through DNA methylation.^9–11,14 However, our previous studies showed that DNA methylation was not the mechanism for the loss of LDOC1 gene expression in PFA.⁵ Consistent with this, hypoxia-induced reduction of LDOC1 gene expression was not associated with significant changes in DNA methylation (Figure 2A). We did see global changes in DNA methylation when PFA cells were exposed to hypoxic conditions (Supplementary Figure S2A). We performed geneset enrichment analysis with top differentially changed methylation beta values. We found there was an enrichment of methylation beta values in normoxia conditions that were associated with hallmark signaling of EMT, hypoxia, and glycolysis (Supplementary Figure S2B). Consistent with mesenchymal phenotype, hypoxia increased DNA methylation of apical junctions, a hallmark signaling pathway of epithelial cells (Supplementary Figure S2C).

One of the salient features of PFA ependymoma is the increased EZHIP (CxORF67) expression that leads to global loss in histone 3 lysine 27 trimethylation (H3K27me3).^26–28 PFA cell lines treated with the histone deacetylase inhibitor TSA restored the LDOC1 expression.⁵ Therefore, we hypothesized the expression of LDOC1 is regulated by histone methylation and is in a region of chromatin with increased H3K27me3. We isolated histones from cells after incubation in hypoxia and measured global changes in H3K27me3 and H3K27 acetylation (H3K27Ac). Interestingly, we found that globally, H3K27Ac increased with hypoxia while H3K27me3 was unchanged (Figure 2B). Looking specifically at the LDOC1 loci, we were only able to detect consensus peaks with H3K27me3 antibody (Figure 2C–D) suggesting that histone methylation may lead to loss of LDOC1. We analyzed the LDOC1 loci from H3K27Ac ChIP sequencing of PFA tumors²¹ to validate these findings. In both PFA1 tumors, there were reduced H3K27Ac peaks in LDOC1 compared to PFA2 tumors (Supplementary Figure S3A). Additionally, the PFA1 tumor had an enrichment of H3K27me3 peaks (Supplementary Figure S3B), consistent with our hypothesis that histone methylation results in LDOC1 loss.

EZHIP overexpression (Supplementary Figure S4A) mimics the H3K27M mutation found in diffuse midline gliomas (DMG) by inhibiting EZH2, leading to a global loss of H3K27me3.^26–28 To determine whether EZH2 regulates LDOC1 silencencing, we treated MAF-811 the EZH2 inhibitor, Tazemetostat. We did not find a significant difference in LDOC1 expression at either dose of Tazemetostat (Supplementary Figure S4B). Additionally, we evaluated whether EZHIP regulated the loss of LDOC1 by treating both cell lines with Panobinostat, which is known to reverse the effects of H3K27M mutations. Interestingly, we found a further reduction of LDOC1 gene expression with Panobinostat treatment (Supplementary Figure S4C), suggesting that LDOC1 loss is independent of EZHIP overexpression.

To further explore the role of histones in LDOC1 silencing, we performed an open chromatin PCR for the transcription start site (TSS) of LDOC1 in PFA snap-frozen tumors. This assay uses PCR probes designed to amplify specific DNA regions; we designed primers that would span the TSS of LDOC1. If there is amplification of the DNA region, this would suggest that the gene is located within a region of open chromatin and that DNA transcription machinery has access to the gene. However, if the site is not amplified, then the site is located within compacted chromatin, which prevents DNA transcription machinery from accessing that gene. For patients with high levels of LDOC1 gene expression (n = 8), only 2 samples had undetectable amplification of LDOC1 TSS. In contrast, 6 of 7 PFA tumors with low expression of LDOC1 had no amplification of LDOC1 TSS (Figure 2E, chi2 = 0.0187). This work is consistent with single-nuclei ATAC-sequencing performed by Aubin, R. G. et.al, showing the LDOC1 locus was inaccessible in the mesenchymal-like cells.²⁹ This would indicate that in PFA1, LDOC1 is in a region of condensed chromatin and thus not accessible for gene transcription.

Hypoxia Induces Canonical NF-κB Signaling

Unchecked NF-κB signaling is a hallmark of pediatric ependymoma both in the posterior fossa and supratentorial regions of the brain.^5,30 However, NF-κB signaling is driven by different mechanisms in the 2 compartments: supratentorial ependymoma is driven by a ZFTA fusion protein with RELA, which leads to NF-KB pathway activation³¹ whereas, in PFA ependymoma, we have shown NF-κB signaling is mediated through loss of LDOC1.⁵ In MAF-811, which grows in culture with more CEC subpopulation phenotype, hypoxia-induced enrichment of canonical NF-κB signaling genesets (Figure 3A). Hypoxia-induced NF-κB signaling is important in preventing apoptosis, initiating angiogenesis, and promoting tumor cell mobility. Given the up regulation of NF-κB with hypoxia and the known connection between LDOC1 and NF-κB regulation, we evaluated NF-κB signaling between normoxia and hypoxia growth conditions. We found down regulation of the Toll-Like Receptor signaling and anti-tumor immune gene signaling with hypoxia (Figure 3B). These gene signatures are the same immune gene signatures identified in PFA2 tumors.³ In both cell lines, we see a significant change with hypoxia in canonical NF-κB signaling (Figure 3B).

For cells to rapidly activate NF-κB signaling rather than inducing gene transcription, RELA and its DNA binding partners, NF-κB1 and NF-κB2, are restricted in the cytoplasm by the IKK/NEMO complex. Upon activation of the pathway, IKK/NEMO is phosphorylated releasing the RELA dimer to translocate to the nucleus. NF-κB2 will also go through a cleavage from its inactive form p100 to activated form p52 before the complex translocate. Once in the nucleus, transcription of NF-κB target genes is initiated.³² To further investigate NF-κB activation, we isolated nuclear and cytoplasmic fractions from cells grown under normoxia and hypoxia. Under hypoxia, we see an increase in nuclear RELA as well as the p52 isoform of NF-κB2 consistent with pathway upregulation (Figure 3C). Furthermore, we found a significant increase in cytokines associated with pro-tumor microenvironment phenotype (IL-6, VEGF, and IL-7³³) (Figure 3D). These data further support the critical role NF-κB signaling has on PFA tumor and immunobiology.

Introduction of LDOC1 Decreases PFA Viability and Growth

We next evaluated the effect of reintroducing LDOC1 in PFA cell lines. We designed a lentiviral vector containing LDOC1 open reading frame on a CMV promoter with puromycin resistance and GFP signal and transduced cells with lentivirus (Supplementary Figure S1A). In both cell lines, we see increased LDOC1 protein within similar amounts of normal cerebellum (Figure 4A). Following puromycin selection, we seeded transduced cells for cell survival and neurospheres growth under normoxia and hypoxia conditions. After 5 days of growth, cell viability was significantly reduced in hLDOC1 cells compared to the Ctl control in both MAF-811 and MAF-928 under both oxygen conditions (Figure 4B). We also observed a significant reduction in cell cycling with reduced incorporation of tritiated thymidine (Figure 4C). This loss in viability and proliferation was correlated with an increase in accumulated cleaved caspase 3/7 and cell death (Supplementary Figure S5A). The introduction of LDOC1 also resulted in significantly smaller neurospheres (Figure 4D). MAF-928 cells transduced with hLDOC1 were unable to form complete neurospheres with most of GFP signal as debris around the periphery of each well (Figure 4D). To determine whether LDOC1 has a role in clonal expansion, we performed a limiting diltution assay following puromycin selection of the cells (Supplementary Figure S6). Consistent with the other growth assays, LDOC1 knock-in cells required more input cells to form neurospheres compared to control vector cells. For MAF-928 we were able to detect and quantify the area of the neurosphere, for the control vector with the input of a minimum of 8 cells compared to hLDOC1 62 cells per well were required (Supplementary Figure S6). For MAF-811, we could measure neurospheres for control with 16 input cells but did not see replicate wells for hLDOC1 until 125 input cells. These data suggest that LDOC1 is required for PFA tumor survival and growth.

LDOC1 Expression Reduces UEC-A and MEC Phenotype

We hypothesize that MEC signals to UEC-A to induce proliferation during the re-epithelization cycle.⁶ COL9A2, a marker of UEC-A, colocalized with Ki67 staining in regions bordering necrosis in patient PFA tissue.⁶ We found hLDOC1 reduced COL9A2 by 75% compared to control (Figure 4E). LGR5, another marker of UEC subpopulations, was also significantly reduced with hLDOC1 (Figure 4D). Additionally, we found a 50% reduction in CA9, a marker used to identify MEC cells, with hLDOC1 (Figure 4E). This suggests that loss of LDOC1, in part, may regulate the transition of epithelial to mesenchymal phenotype in PFA.

LDOC1 Suppresses NF-κB Signaling and Reduces IL-6 Secretion

In addition to changes in the UEC and MEC phenotype, we found some changes in NF-κB activity. Gene expression of RELA and NF-κB1 was significantly reduced with hLDOC1 (Figure 4F). This correlated with a decrease in nuclear RELA under normoxia conditions (Figure 4G and S5B). In general, proteins that activate the NF-κB signaling pathway were decreased with hLDOC1 expression (Supplementary Figure S5C). There was a significant decrease in the activated form of RELA (pS529). We also observed a general increase in the proteins that suppress or regulate NF-κB signaling. However, these same results were not observed under hypoxia conditions. Additionally, hLDOC1 led to a significant decrease IL-6 secretion under both oxygen conditions (Figure 4H). These data confirm that loss of LDOC1 in part leads to the constitutively active NF-κB signaling and chronic IL-6 secretion observed in PFA1 tumors.

Loss of LDOC1 Is Required for In Vivo Tumor Growth

To determine whether LDOC1 loss is required for tumor growth, we added a LUC2 luciferase coding sequence to our hLDOC1 and control (Ctl) constructs (Supplementary Figure S1B) to allow us to monitor tumor growth in vivo using bioluminescence imaging. Our PFA patient-derived xenograft tumor cells did not tolerate puromycin selection or flow sorting prior to injection. Therefore, tumors with measurable luminescence indicated engrafted tumors have taken in the lentiviral constructs. Bioluminescence imaging was performed 8 months after injecting PFA xenograft cells in flanks of NSG mice. For MAF-811 control cells, 10 of the 12 mice grew tumors and all 10 mice had measurable luciferase signal with an average of 1.25 × 10^8 radiance. For MAF-811 hLDOC1, 10 of the 15 mice grew tumors, and while there were some mice whose tumors visually showed luminescence, only 1 of these mice had radiance above the background (Figure 5A). Similar to MAF-928 control cell, all 11 mice grew tumors but only 8 mice had measurable luminescence. For MAF-928 hLDOC1, 12 of 15 mice had tumors but none of the mice had measurable bioluminescence. For MAF-811, control tumors were significantly larger in size and weight compared to hLDOC1 tumors (Figure 5B). For MAF-928, there was no difference in the size and weight of the tumors. To determine whether LDOC1 was being expressed in the hLDOC1-luc tumors, we harvested tumors for snap-frozen and fixed tissue. Using snap-frozen tissue, we isolated RNA and measured LDOC1 gene expression by qRT-PCR. Given the results of the spatial transcriptomics (Figure 1A), we normalized the data to endogenous LDOC1 levels from normal cerebellum samples. We also included 2 ZFTA-fused supratentorial patient tumors for normal endogenous levels of LDOC1 gene expression. For both PDX models, control tumors and hLDOC1 transduced tumor cells had nearly undetectable levels of LDOC1 compared to ZFTA-ST and normal cerebellum (Figure 5C). All tumors have visual features of ependymoma tumors (Figure 5D, H&E), with regions of ependymal rosettes, necrosis, pseudo-rosettes, and perivascular zones. LDOC1 IHC staining had positive cells in the blood vessels but most of the LDOC1 staining was lightly and diffusely expressed throughout the tumor (Supplementary Table S4). Blinded scoring of LDOC1 staining demonstrated no significant difference in the IHC scores between control and hLDOC1 transduced tumors (Figure 5D, middle panel). In addition, Ki67 was not different between groups (Figure 5D, right panel). As the cells were not able to be selected prior to injecting in the mice, a mixed population of transduced and untransduced cells were injected. Given the results of the bioluminescence imaging, combined with the lack of LDOC1 gene expression or LDOC1 protein, the cells that were untransduced (LDOC1 negative) out competed the LDOC1 positive cells to form the tumors in the hLDOC1 mice. These data indicate that loss of LDOC1 is required for PFA ependymoma tumorigenesis.

Discussion

Despite extensive molecular characterization of pediatric ependymoma, no mutational driver has been identified for posterior fossa subgroups. We were the first to identify immune gene signature differences between PFA subtypes and these changes correlate with outcome. PFA2 tumors harbor an immune-activated, anti-viral phenotype and have better outcomes with surgery and radiation. In contrast, PFA1 develops an immunosuppressive signature and do poorly.³ Our prior studies have shown these immune differences are driven by constitutively active NF-κB signaling in PFA1 tumor cells leading to chronic IL-6 secretion.^4,5 Tumor-secreted IL-6 initiates the downstream immunosuppressive cascade⁴ (see graphical abstract). In a separate study, we identified LDOC1 as the only gene completely lost between PFA1 and PFA2 tumors. LDOC1 is a negative regulator of NF-κB activity and in a variety of cancers, expression is repressed by DNA methylation.^9–14

DNA methylation, which is the standard protocol used to assign PFA subgroups, showed no differences in beta values of LDOC1 loci CpGs between molecular subtypes of PFA.⁵ However, using single-cell analysis we found the cellular subpopulations of PFA that were associated with the loss of LDOC1 near regions of necrosis. In this study, we were unable to find a difference between LDOC1 DNA methylation beta values between normoxia and hypoxia-incubated PFA cell lines. From the single-cell transcriptomic studies, the cellular subtypes associated with perinecrotic regions are mesenchymal ependymal cells (MEC), undifferentiated ependymal cells—A (UEC-A), hypoxia myeloid cells, microglial cells and classically activated myeloid cells.^6–8,25 Interestingly, the hypoxic tumor microenvironment can induce chromatin compaction. While PFA ependymoma has global loss of histone 3 lysine 27 trimethylation, our studies suggest that compacted chromatin is the primary mechanism for loss of LDOC1. Under both normoxia and hypoxia conditions, LDOC1 was not pulled down with H3K27me3 antibody. Further studies using single-cell ATAC sequencing coupled with single RNA sequencing validated these findings.²⁹

A limitation of this study is the heterogeneity of our cell lines. While we have used these lines in several studies modeling PFA1 tumor and immunobiology, they have inherent differences. For example, MAF-811 transcriptome is more consistent with CEC phenotype of PFA2 than that of MAF-928. Additionally, MAF-928 undergoes substantial changes under hypoxia conditions, and are difficult to replicate findings in subsequent studies. Despite these limitations, they are the only fully characterized PFA ependymoma cell lines published.

It has recently been shown that IL-6 secretion by cells results in a feedback mechanism that activates DNA methyl transferases, thereby further suppressing LDOC1 expression.³⁴ Beyond the scope of this study, PFA1 tumor-secreted IL-6 is a hallmark of the MEC subpopulation, which drives the immunosuppressive microenvironment we have characterized (graphical abstract). Blocking IL-6 with tocilizumab or a neutralizing antibody was sufficient to block the myeloid cell polarization by IL-6.⁴ While we have not evaluated the effect of blocking IL-6 in our PFA1 tumor cells, we do see that restoring LDOC1 gene expression resulted in a significant reduction of IL-6 secretion. This suggests that blocking IL-6 may have a two-prong benefit on both tumor biology and immunobiology of PFA1 tumors.

Acknowledgments

This study was funded by NIH/NCI R01 CA239302–01A1 and was supported by NIH/NCATS Colorado CTSA Grant Number UM1 TR004399. Contents are the authors’ sole responsibility and do not necessarily represent official NIH views. This study also was financed, in part, by the São Paulo Research Foundation (FAPESP), Brasil. Process Number #2021/11402–9. The genomics core is funded through the University of Colorado Cancer Center (P30CA046934). IVIS spectra in the Animal Imaging Shared Resources and is funded by P30CA046934 and the HIH high-end shared instrumentation grant (S10OD023485). The PFA patient tumor H3K27Ac and trimethylation data was generously provided by Dr. Lukas Chavez.²¹ The authors would like to thank The Morgan Adams Foundation for their continued support.

Authorship Statement

G. R S., T.L., S.V., T.P., R.V. N.K.F., and A.M.G. designed and analyzed the experiments. G.R S., A.J.C., E.G., A.M.D, L. M S., K.M.J., S.V., V.A, R.V., J.M.L., and A.M.G conducted and/or interpreted the experiments. The patient specimen was provided by T.H. and M.H. The manuscript was assembled by G. R D., A.J.C, and A.M.G. and further edited by A.M.D, J.M.L., E.T V., N.K.F. G.R S., and A.M.G. was responsible for the final production of the manuscript.

Conflict of interest statement. None declared.

References