FOXR2 activation is not exclusive of CNS neuroblastoma

Abstract

Background

FOXR2 activation is regarded as pathognomonic for CNS neuroblastoma (NB). However, a comprehensive understanding of the landscape for CNS tumors exhibiting FOXR2 activation is lacking.

Methods

Histopathologic, molecular, imaging, and clinical data of 42 CNS tumors with FOXR2 overexpression identified through screening institutional datasets and published institutional cases were analyzed.

Results

Among the 42 tumors, 21 (50.0%) were high-grade gliomas (HGGs), and 18 (42.9%) were embryonal tumors. The HGGs included ten H3 K27M-mutant diffuse midline gliomas (DMGs) and eight radiation-associated tumors. The embryonal tumors included 11 CNS NBs and six pineoblastomas (PBs). FOXR2 expression was similar between CNS NB and other tumor types (P = 0.82). HGGs with FOXR2 overexpression, unlike NBs and PBs, displayed diverse concomitant genetic alterations. The most common mechanisms of FOXR2 activation involved structural alterations causing promoter donation and enhancer hijacking from active genes essential for brain development, followed by alternative promoter activation or truncated LINE-1 retrotransposition. The preferential activation mechanism varied by tumor type. All but two aberrant FOXR2 transcripts incorporated non-canonical, non-coding exons. Gene set enrichment analysis demonstrated shared downstream effects of FOXR2 activation at the epigenome and transcriptome levels across tumor types. DMGs and PBs with FOXR2 overexpression were aggressive, with 0% 2-year overall survival, whereas CNS NBs responded well to combined chemotherapy and radiation.

Conclusions

CNS tumors with FOXR2 overexpression manifest significant histological, molecular, imaging, and clinical diversity. While HGGs and PBs with FOXR2 overexpression demonstrated inferior prognosis, CNS NBs showed favorable outcomes. Integrating histologic and molecular diagnostic approaches is imperative for accurate prognostication and optimal therapeutic decision-making.

FOXR2 (Forkhead Box R2) is a transcription factor belonging to a large family of highly conserved FOX proteins known for their roles in regulating gene expression and a wide array of cellular functions.¹ Normally expressed only in the testis, FOXR2 has been identified as an epigenetically activated oncogene in many cancers.² In the CNS, FOXR2 activation is considered the pathognomonic molecular signature of CNS neuroblastoma (NB)^3–5 and has been demonstrated to be causative in developing CNS embryonal tumors.⁶ However, more recent studies indicate that FOXR2 may be implicated in a broader range of CNS tumors. These studies have implicated FOXR2 in subtypes of pineoblastoma (PB),⁷ medulloblastoma (MB),⁸ and high-grade glioma (HGG).^2,9 However, a comprehensive understanding of the landscape for CNS tumors exhibiting FOXR2 activation is lacking, and the exact underlying mechanisms of FOXR2 activation in specific brain tumor types have yet to be elucidated. Previous studies have demonstrated the association of elevated FOXR2 expression with poor prognosis in some cancers.^10–12 However, whether FOXR2 overexpression has implications for therapeutic responses and clinical outcomes in specific CNS tumor types remains unknown. Moreover, since FOXR2-activated CNS NB is rare, collecting as much information as possible about their clinicopathologic profiles is essential for gaining further insights into their treatment response and prognosis.

We hypothesized that CNS tumors with FOXR2 overexpression exhibit diverse histopathologic, molecular, and clinical features. To understand the landscape of CNS tumors with FOXR2 overexpression, we screened the institutional whole-transcriptome gene expression (RNA-seq) datasets to identify tumors with FOXR2 overexpression. Previously published institutional cases were included for complete analysis.^13,14 The study cohort was subjected to comprehensive medical history, imaging, and histopathologic review to understand the association of specific CNS tumor types, therapeutic responses, and patient outcomes. Detailed genomic, transcriptomic, and epigenomic analyses were performed to elucidate the molecular mechanisms of FOXR2 activation in specific CNS tumor types and the downstream effects of FOXR2 activation.

Materials and Methods

Study Cohort

Institutional RNA-seq datasets comprising over 3,500 tumors were screened for CNS tumors with FOXR2 overexpression following batch correction using ComBat and normalization using variance stabilizing transformation. Tumors (n = 40) exhibiting FOXR2 normalized HTSeq counts exceeding the upper quartile (Q₃) plus 1.5 times the interquartile range (IQR) were selected for review. Samples (n = 5) with suboptimal sequencing coverage, FOXR2 expression levels resembling background noise, or lacking distinct FOXR2 exonic expression patterns or overexpression in replicates were excluded. The cohort was supplemented with published institutional cases of CNS NB-FOXR2 (n = 5) and PB-FOXR2 (n = 2).^13,14 A cohort (n = 29) of H3 K27M-mutant diffuse midline glioma (DMG) without FOXR2 overexpression was assembled for comparison by screening institutional RNA-seq datasets. Clinical data were available for 26 of the 29 cases.

Transcriptome Sequencing (RNA-seq) and Analysis

RNA was extracted from formalin-fixed paraffin-embedded (FFPE) tissue using a PureLink FFPE RNA Isolation Kit (Thermo Fisher Scientific), as previously described.^15–20 Purified RNA was quantitated on a Qubit 3 Fluorometer and subjected to RNA-sequencing following the Illumina TruSeq Stranded Total RNA protocol (500 ng RNA minimum). All sequencing data were generated after 100 cycles of paired-end run on an Illumina NovaSeq X Plus. The RNA-seq data were aligned to the human reference genome (hg19).

FOXR2 structural variants (SVs) were detected by CICERO²¹ or by manually examining sample bam files for soft-clipped reads upstream of the RefSeq FOXR2 transcript (NM_198451.4). Soft-clipped sequences were used to identify the partner genes involved in the structural alterations by BLAT.²² Expressed FOXR2 exons were identified by manually examining the bam file for peaks in coverage and spliced reads. The exon annotation utilized follows that described in Tsai et al.²

RNA feature counts for brain tumor samples from 325 patients were obtained from St. Jude Cloud datasets.^15,23–25 Normalized counts were used for gene set enrichment analysis (GSEA) using the GSEA Desktop application (version 4.3.3). The analysis parameters included 1,000 permutations and phenotype labels based on activation or non-activation of FOXR2.

Whole Genome Sequencing and Analysis

Genomic DNA was extracted from snap-frozen tumor tissue using an automated Maxwell RSC Instrument (Promega), as previously described.^15,16,26 DNA quality was assessed on a 4200 TapeStation (Agilent). Paired-end sequencing was conducted on the Illumina HiSeq platform with a 100- or 125-bp read length or NovaSeq with a 150-bp read length. Sequencing results were analyzed using an institutionally established pipeline for alignment and calling of single nucleotide variants and insertions or deletions, which were annotated and ranked by putative pathogenicity using “medal ceremony” and subsequently manually reviewed, as described previously.²⁷ DNA breakpoints for FOXR2 SVs were identified in WGS data (if available) or through intronic RNA soft clips.

Whole Exome Sequencing and Analysis

Genomic DNA was extracted from FFPE tissue using a QIAamp DNA FFPE Tissue Kit (Qiagen), as previously described.^15,16,18,26 The genomic libraries were generated using the SureSelectXT kit (Agilent Technologies), followed by exome enrichment using the SureSelectXT Human All Exon V8 bait set (Agilent Technologies). The resulting exon-enriched libraries were subjected to the paired-end, 100-cycle sequencing performed on a NovaSeq X Plus (Illumina). Sequencing results were analyzed as described above.

Genome-wide DNA methylation profiling and analysis

Genomic DNA was extracted as stated above and subjected to genome-wide methylation profiling on an Illumina Infinium MethylationEPIC platform, as previously described.^15–20 Differential methylation analysis was performed in R 4.4.0 using the minfi 1.50.0,²⁸ limma 3.60.2,²⁹ ChAMP 2.29.1,³⁰ and sesame 1.22.1³¹ packages with default parameters. The DMRs identified were then used for GSEA by querying against the MSigDB (https://www.gsea-msigdb.org/gsea/msigdb/). Tumor methylation profiles were classified using the Molecular Neuropathology brain tumor classifier (www.molecularneuropathology.org) v12.8. A matching score higher than 0.9 was used as the threshold for classifying molecular tumor type.

Histopathology Review and Immunohistochemistry

Hematoxylin and eosin-stained 5 μm sections of FFPE tissue specimens of tumor samples were centrally reviewed by a board-certified neuropathologist specialized in pediatric CNS tumors (JC) to confirm the diagnosis. The following antibodies were used for immunohistochemistry: GFAP (DAKO, M0761, clone 6F2, 1:400), Olig2 (Cell Marque, 387M-15, clone 211F1.1, 1:50), and synaptophysin (Leica Microsystems, NCL-L-SYNAO-299, clone 27G12, 1:400).

Imaging Review

A retrospective review of all baseline and follow-up magnetic resonance imaging (MRI) was performed by a pediatric neuroradiologist (SNP) blinded to the final diagnosis. Baseline tumor location, margins, cellularity, enhancement, and leptomeningeal metastases were recorded. On follow-up, tumors were assessed for pseudoprogression, response to therapy, or progressive disease. Pseudoprogression was defined as the transient increase in tumor size within the first three months after completion of radiation or immunotherapy, which resolved on follow-up imaging.^32,33 Response to therapy was graded as complete, partial, or minor following the Response Assessment in Pediatric Neuro-Oncology (RAPNO) criteria or, in the case of immunotherapy, the Immunotherapy Response Assessment in Neuro-oncology (iRano) criteria. Progressive disease was defined as a greater than 25% increase in the product of the biperpendicular dimensions of the primary tumor or the presence of new leptomeningeal metastases, which persisted/worsened on follow-up imaging.^34,35

Statistical Analysis

Clinical data, including demographic information, initial metastatic staging, and treatment details, was extracted from medical records. The surgery date to obtain diagnostic tissue samples was designated as the date of diagnosis. The date of progression was determined by imaging. Progression was defined by imaging as described above or the conversion of CSF cytology to positive in two consecutive evaluations. Event-free survival (EFS) was defined as the interval from the date of primary tumor diagnosis to the earliest date of disease progression, second malignancy, or death due to any cause for patients who experienced events or to the date of last follow-up for patients without events. Overall survival (OS) was defined as the time interval from the date of diagnosis to the date of death from any cause or to the date of the last follow-up for survivors. Survival outcomes were estimated with the survival package (version 3.3.1) in R 4.2.1 using the Kaplan–Meier method with 95% confidence intervals. Differences in survival distributions were examined by log-rank test.

Results

The Landscape of CNS Tumors with FOXR2 Overexpression

A total of 42 CNS tumors with FOXR2 overexpression from 41 patients were included in this study, including one patient with CNS NB and a secondary radiation-associated tumor (Figs. 1 and 2, Supplementary Table S1). Among them, 21 (50.0%) were high-grade gliomas (HGGs), 18 (42.9%) were embryonal tumors, two were difficult-to-classify high-grade neuroepithelial tumors (HGNET), and one was pineal parenchymal tumor of intermediate differentiation (PPTID). This finding confirms our hypothesis that CNS tumors with FOXR2 overexpression comprise tumors of diverse histology. While rare, a PPTID in this cohort suggests that CNS tumors with FOXR2 overexpression are not necessarily histologically high-grade. There were no significant differences in FOXR2 expression levels among tumors in different categories (CNS NB-FOXR2 vs. all others, P = 0.82).

The HGGs with FOXR2 overexpression included ten H3 K27M-mutant DMGs, eight radiation-associated secondary HGGs, one with MN1::PATZ1 fusion, and two not-elsewhere-classified HGGs (Fig. 2A). HGGs with FOXR2 overexpression harbored diverse concomitant genetic alterations frequently encountered in pediatric HGGs, including DMGs. The two thalamic DMGs harbored concomitant histone H3 K27M and BRAF V600E mutations, an infrequent combination in DMGs. One pontine DMG harbored a KIAA1549::BRAF fusion, a molecular alteration associated with pilocytic astrocytoma and diffuse leptomeningeal glioneuronal tumor and exceedingly rare in DMGs. Chromosome arm 1q gain (1q+), a characteristic molecular alteration of CNS NB-FOXR2, was seen in 70.6% (12/17) of HGGs with FOXR2 overexpression tested, demonstrating that the combination of 1q+ and FOXR2 overexpression is not specific to CNS NB-FOXR2.

The embryonal tumors with FOXR2 overexpression included 11 CNS NB-FOXR2, six pineoblastoma (PB), including five PB-FOXR2 and one pineal anlage tumor (PAT), and one medulloblastoma (MB). Therefore, FOXR2 overexpression is not limited to PB-FOXR2 in the pineal region. Except for the one SHH-activated and TP53-mutant MB, all other embryonal tumors with FOXR2 overexpression did not harbor other pathogenic or likely pathogenic genetic alterations.

All CNS NB-FOXR2 were cerebral cortical tumors, while other CNS tumors with FOXR2 overexpression show regional predilections depending on the specific tumor type. Our results support our hypothesis that CNS tumors with FOXR2 overexpression exhibit diverse histopathologic, molecular, and clinical features.

Patient Outcomes

Forty patients had detailed clinical information for analysis. Demographic information, histological diagnosis, the extent of resection, metastatic staging at diagnosis, and treatment regimens are summarized in SupplementaryTable S2 and SupplementaryFig. S1. The median age at diagnosis of the entire cohort was 4.5 years (0.50–22.28 years) and varied with underlying diagnosis. Eighteen (44%) were females. Age, metastatic status at diagnosis, and tumor location were associated with histological diagnosis (SupplementaryTable S2). Treatment varied significantly based on the age of diagnosis and histological diagnosis, with the majority receiving multimodal therapy upfront. Event-free (EFS) and overall survival (OS) based on diagnosis are depicted in Fig. 2B and C.

CNS NB-FOXR2

Ten patients had CNS NB-FOXR2 with clinical data available. The median age at diagnosis was 5.3 (range: 1.6–13.6 years). All tumors were supratentorial/hemispheric and non-metastatic at diagnosis. Two patients remained in upfront therapy at the time of analysis. Overall outcomes were significantly better than all other tumor types with FOXR2 overexpression. Treatment with multimodal therapy appears to be crucial, as one patient with no therapy had rapid progression and died. One other patient died due to a secondary HGG with no recurrence of CNS NB. (Fig. 2D)

DMG

Ten patients had DMG with FOXR2 overexpression. Their demographics and clinical parameters were compared with an institutional cohort of 26 DMGs without FOXR2 activation (Suppl Table 3). Patients with DMG with FOXR2 overexpression were younger at diagnosis (6.0 vs 9.1 years, P = 0.007), but rates of metastatic disease at diagnosis, gender, and extent of resection were similar. DMGs with FOXR2 overexpression included eight pontine and two thalamic tumors, with 1-year EFS at 0% and OS at 20%. The DMGs without FOXR2 activation included 18 pontine, five thalamic, one spinal, and two other locations, with 1-year EFS and OS at 3.9% and 46.2%, with no significant difference between those with and without FOXR2 activation (SupplementaryFig. S2). All patients received radiation, and adjuvant or targeted therapies showed no survival benefit.

HGG

Eleven patients had non-DMG HGG, including eight radiation-induced and three primary tumors. All received radiation and predominantly temozolomide-based chemotherapy. Two patients with radiation-induced HGGs and one with MN1::PATZ1 fusion are long-term survivors. Two are recently diagnosed and undergoing treatment. The remaining patients experienced rapid disease progression and poor outcomes despite treatment.

PB

Six patients with PB with FOXR2 overexpression (five classic PB and one pineal anlage tumor) had a mean age of diagnosis of 1.7 years (0.5–5.7 years). All received upfront multimodal chemotherapy, with radiation administered to older patients. Survival outcomes were poor, with 0% OS at two years and frequent progression during upfront therapy.

Other

Our cohort included two HGNETs, one MB, and one PPTID. Both HGNET patients were under two years old, non-metastatic at diagnosis, and treated with upfront chemotherapy alone; one experienced disease progression and died, while the other remained in upfront therapy. The medulloblastoma and PPTID patients received radiation and chemotherapy and are alive without evidence of disease.

Histologic spectrum of CNS NB-FOXR2

While DMGs, HGGs, PBs, MB, and PPTID with FOXR2 overexpression in our cohort exhibited typical histopathologic features of their specific tumor type, FOXR2-activated CNS NB demonstrated substantial variability in histology and immunophenotype (Fig. 3). All tumors contained areas exhibiting embryonal cytology, with poorly differentiated cells with abundant mitotic activity, high nuclear-to-cytoplasmic ratios, hyperchromatic nuclei, and inconspicuous amounts of cytoplasm. None showed ganglion cell differentiation. The architecture of the embryonal areas varied (Fig. 3A–C). Two samples of small biopsy displayed only embryonal tumor appearance, and the rest with large resections contained regions showing variable glial characteristics, including more prominent eosinophilic cytoplasm (Fig. 3D), infiltrative growth (Fig. 3E), fibrillary background (Fig. 3F), and mixed high-grade and low-grade appearance (Fig. 3G). Immunohistochemistry detected various extents of genuine glial fibrillary acidic protein (GFAP) expression by the tumor cells (Fig. 3H). While typical CNS NB demonstrates extensive oligodendrocyte transcription factor 2 (Olig2) and synaptophysin expression, there was substantial variability (Fig. 3I–L).

Imaging Features of CNS Tumors with FOXR2 Overexpression

A blinded review was performed for 35 tumors from 34 patients with baseline imaging available, including one patient who initially presented with CNS NB and then developed a secondary post-RT HGG. A total of 21 tumors had imaging features suggestive of HGGs, seven were suggestive of CNS NB, six were suggestive of PB, and one was suggestive of either PB or PPTID. Key baseline imaging features are summarized in Supplementary Table S4. Imaging diagnosis was based on typical characteristics of CNS NB, HGG, PB, or PPTID without any relevance to their FOXR2 activation status. There was 100% concordance between the suspected diagnosis from the blinded imaging review and the final diagnosis.

The seven CNS NBs were all in the frontal lobe and universally well-circumscribed, showed diffusion restriction, and were cortically based with minimal white matter extension. (Fig. 4). In contrast, the seven hemispheric HGGs exhibited infiltrative margins, extensive white matter involvement, and variable hypercellularity and enhancement.

On follow-up imaging, pseudoprogression was noted in four pontine DMGs, either following radiation (three) or immunotherapy (one), and was absent in other tumors. These four cases of pontine DMGs eventually exhibited progressive disease approximately one to four months after pseudoprogression resolved.

Mechanisms of FOXR2 Activation

In our cohort, the most common mechanisms of FOXR2 activation involved structural alterations causing promoter donation from other genes and incorporating upstream non-canonical, non-coding exons of FOXR2, followed by alternative promoter activation or truncated LINE-1 retrotransposition (Fig. 5, Supplementary Figs. 3–7). Structural alterations resulting in promoter donation occurred as fusions to FOXR2 exon-6 (n = 5), -2 (n = 14), or 1 (n = 2). Alternative promoter activation events primarily occurred at the exon-3 promoter (n = 13), resulting in the expression of FOXR2 transcript 3. Only one case had apparent activation of the exon-7 promoter, resulting in the expression of FOXR2 transcript 1. Two exon-2 alterations resulted from recurrent truncated LINE-1 retrotransposon insertions.

The preferential mechanisms of FOXR2 activation varied by tumor type (Fig. 5). DMGs (n = 9) and pineal tumors (n = 4) preferred activation of the exon-3 promoter, with only one DMG and one pineal tumor having a structural event (exon-2 and -6, respectively). The pineal tumor with structural alteration was of a different type (pineal anlage tumor) than the others (PB-FOXR2, PPTID), which could explain the mechanistic difference. Structural alterations were exclusively observed for non-DMG HGGs (exon-2, n = 9; exon 1, n = 2), HGNETs (exon-2, n = 2), and MB (exon-2, n = 1). CNS NBs also prefer structural events (exon-6, n = 4; exon-2, n = 1), with one exceptional case having exon-7 promoter activation.

One of the HGNET cases in our cohort was previously described, with a LINE-1 insertion event that resulted in promoter donation and FOXR2 activation (SJHGG030242).⁹ We found another case in our cohort (SJBT032161_D1) with the same RNA soft clips at FOXR2 exon-2, supporting a LINE-1 insertion event (Supplementary Fig. S7). This was the only CNS NB case with a structural alteration resulting in promoter donation to FOXR2 exon-2, as the rest preferred FOXR2 exon-6.

Analysis of RNA coverage and spliced FOXR2 reads suggested that there are at least two transcripts not currently available in RefSeq that are utilized in the activation of FOXR2 (transcripts 1 and 3, Fig. 5; Supplementary Fig. S3). All exons upstream of exon 1 appear to be non-coding, as there was no larger open reading frame present in the transcript sequences. All non-coding exon boundaries for these transcripts had the canonical GT/AG sequence at their splice sites, aside from the exon-7/exon-6 boundary, which had a GC/AG. This non-canonical junction sequence has previously been observed in humans at a low frequency.³⁶  FOXR2 exon 1 in the activated transcripts has a splice acceptor > 200 bp upstream of the RefSeq transcript (Fig. 5). All FOXR2-activated tumors in our cohort utilize this splice site.

The overexpression of FOXR2, either through structural variation or alternative promoter activation, was associated with increased CpG methylation levels at the five CpG sites covered by the Infinium MethylationEPIC v2.0 BeadChip at the “gene body” region of the aberrant, non-canonical FOXR2 transcripts (Supplementary Fig. S8), which has been associated with elevated gene expression throughout the human genome.^37,38 Among the five CpG sites at the FOXR2 locus, three (cg09113017, cg17779979, and cg26217484) showed significantly increased methylation levels in FOXR2-activated tumors, while two (cg01704778 and cg25478472) retained their high basal methylation levels. Other profiled CpG sites at nearby genes (USP51 and RRAGB) showed no changes in methylation levels, suggesting that changes at the FOXR2 locus are specific to FOXR2 activation.

Expressions Patterns of Genes That Donated Their Promoters

Expression analysis using GTEx data (accessed Aug 1, 2024) showed that most partner genes identified in our study, except for FAM156A, had low-moderate to highly elevated expression in different tissues and were overall enriched in the brain (Supplementary Fig. S9). We also analyzed H3K27ac and H3K4me3 marks across normal brain samples from Encode (ENCFF190JDV; ENCFF717CEO), which showed the SV breakpoints in the partner genes to be associated with enhancers or promoters in almost all cases, as defined by a peak within 25 kb (Supplemental Table 5).

FOXR2 partner genes that donate their promoters or enhancer elements perform a diverse array of functions (Suppl Table 6), including in RNA processing and regulation (RBM10, ELAVL2), transcriptional regulation and epigenetic modification (BCOR, FTX, KMT5B, MED13L, FAM156A), the ubiquitin-proteosome system (MARCHF6, RNF144A, USP51, TBL1XR1), neural development and synaptic function (CASK, NLGN4X, ZC4H2), and cell signaling transduction (TAB2, ITFG1, CHM). Most partner genes have roles in various stages of brain development and, when mutated in the germline, cause a spectrum of syndromic and non-syndromic neurodevelopmental disorders.

Downstream Effects of FOXR2 Activation in CNS Tumors

Epigenetic effect of FOXR2 activation in CNS tumors

We first looked at the downstream effects of FOXR2 activation in CNS tumors at the DNA methylation level by comparing the DNA methylome profiles of tumors with FOXR2 overexpression in our cohort (n = 30 with sufficient data quality) with those of DMGs without FOXR2 activation (n = 25). Differential methylation analysis identified 10,002 differentially methylated regions (DMRs) between these cohorts. GSEA of genes in the DMRs identified enrichment of FOXR2 target genes and downstream targets of several transcription factors (Suppl Table 7). Notably, the analysis results suggested the activation of MYC and the ETS family transcription factors ETS1, ELF2, ELF5, ELK1, GABPA, GABPB2, NRF2, CETS1P54, DIDO1, and EPC1 in CNS tumors with FOXR2 overexpression, as suggested by previous studies in various tumor types.^2,39,40

Transcriptome Analysis Further Supports the Activation of MYC and ETS Pathways

We then compared the transcriptome profiles of DMGs with FOXR2 overexpression (n = 10) with non-FOXR2-activated DMGs (n = 29), as well as the profiles of all tumors with FOXR2 overexpression (n = 32) with brain tumors lacking FOXR2 activation (n = 303). While the DMG cohorts were smaller, GESA of differentially expressed genes of both comparisons identified the enrichment of FOXR2 target genes (Fig. 6) and suggested the activation of the MYC pathway and the ETS family transcription factors (ELK1, GABPA, GABPB2, and NRF2). This result is consistent with the methylome findings and suggests the activation of MYC and ETS family transcription factors is a common feature of CNS tumors with FOXR2 overexpression.

Discussion

Our findings support the hypothesis that CNS tumors with FOXR2 overexpression exhibit diverse histopathologic, molecular, and clinical features, indicating that FOXR2 alterations are not pathognomonic of CNS NB-FOXR2. Instead, a broad spectrum of HGGs, including DMGs, CNS embryonal tumors (CNS NB, PB, PAT, and MB), and PPTID can harbor FOXR2 activation. Our study also reveals a unique enrichment of radiation-induced HGGs with FOXR2 activation, a previously unreported finding.

Clinical characteristics and outcomes differed across the various tumor types. CNS NB occurred in patients with a median age of ~6 years and overall had excellent outcomes. The favorable overall survival aligns with other published series, including a European report where PFS and OS in a cohort of 63 patients were 63% and 85%, respectively.⁴¹ Like our cohort, treatment varied, and RT appears crucial for good outcomes. However, the optimal RT dose remains unclear, given the excellent survival with therapy. Given the low incidence of leptomeningeal dissemination, the possibility of a lower CSI dose in older patients and focal RT in younger patients should be explored in future clinical trials.

The findings of our blinded imaging review were consistent with the reported imaging characteristics of FOXR2-activated CNS NB.⁴² The distinctive imaging features can help differentiate CNS NB from other embryonal tumors, such as embryonal tumors with multilayered rosettes, which have prominent intra-tumoral veins and hypo-enhancement.⁴³ On follow-up imaging, pseudoprogression was only observed in a subset of pontine DMGs and was notably absent in thalamic, spinal, and hemispheric HGGs. This unique behavior of pontine DMG with FOXR2 overexpression in our cohort may suggest an underlying susceptibility to radiation- and immunotherapy-mediated inflammation and needs further exploration in prospective studies.

Our findings reveal three mechanisms for aberrant FOXR2 activation in CNS tumors—promoter donation/enhancer hijacking through structural rearrangement, alternative promoter activation, and truncated LINE-1 retrotransposition, which showed differential preference among the various tumor types and anatomical sites. Structural rearrangement is the preferred mechanism of FOXR2 activation in cerebral tumors, including non-DMG HGGs and CNS NB. In contrast, midline (thalamic, pontine, and pineal) tumors, including DMG, PB, and PPTID, prefer utilizing alternative promoters. HGGs and DMGs with FOXR2 overexpression additionally harbor concomitant genetic alterations frequently encountered in HGGs. These findings may have diagnostic utility in cases with ambiguous histology. For example, one of the HGNETs in our cohort, while challenging to classify based on histology, harbored concomitant TP53 alteration and PDGFRA amplification, findings associated with HGGs rather than CNS NB. One RT-associated HGG with FTX::FOXR2 fusion occurred after RT for a primary CNS NB with BCOR::FOXR2 fusion, harbored NF1 and PIK3CA alterations, findings typically associated with HGGs. Both tumors progressed rapidly despite therapy, an outcome distinct from typical CNS NB.

Based on the NCBI Reference Sequence Database (RefSeq), FOXR2 is a single-exon gene in humans. However, we observed the inclusion of upstream non-canonical exons in the FOXR2 transcripts in CNS tumors with FOXR2 overexpression. A splice acceptor upstream of exon 1 is utilized in all tumors with FOXR2 overexpression. These findings must be considered when looking for aberrant FOXR2 activation by RNA-sequencing or similar techniques in CNS tumors.

No new protein variant is predicted to be generated from the aberrant FOXR2 fusion transcripts in all tumors with FOXR2 overexpression in our cohort, which drastically differs from conventional gene fusions, where a hybrid protein with functional domains from each partner is brought together, resulting in new proteins with gain of function. This finding suggests that direct interaction with MYC through its MYC-binding domain, located near the N-terminus of FOXR2, and specific DNA binding through its Forkhead domains near its C-terminus are required for FOXR2-mediated tumorigenesis. This observation and the finding that both MYC pathway and ETS family transcription factors are commonly activated in CNS tumors with FOXR2 overexpression may offer a unique therapeutic opportunity. Concomitant alterations in DMGs and HGGs with FOXR2 overexpression, including BRAF, PDGFRA, EGFR, FGFR1, PPM1D, NRAS, PIK3CA, PTEN, and ACVR1, offer additional therapeutic opportunities.

Our study identified a second example of aberrant FOXR2 activation driven by a truncated LINE-1 retrotransposition event,⁹ suggesting that such an oncogenic mechanism is a recurrent finding. The LINE-1 target site (5’-GTTGATATCTTT) located upstream of exon-2 was likely utilized in both tumors since the resulting aberrant FOXR2 transcripts share a similar structure. In both tumors, the LINE-1 promoter drives the transcription of a chimeric LINE-1::FOXR2 transcript as the only driver event present at diagnosis. Our finding again supports LINE-1 promoter donation as a novel cancer-initiating driver mechanism, as opposed to previously reported LINE-1-mediated disruption of tumor suppressors or oncogene repressors.⁴⁴

In summary, our findings demonstrate that CNS tumors with FOXR2 overexpression exhibit diverse histopathologic, molecular, and clinical features. Specific tumor types are associated with preferential mechanisms of FOXR2 activation, utilization of upstream non-canonical non-coding exons, and patient outcomes. There is an enrichment of RT-associated secondary tumors and concomitant genetic alterations in HGGs and DMGs with FOXR2 overexpression. CNS NB-FOXR2 demonstrates favorable outcomes, and therapy de-escalation should be further investigated in future clinical trials.

Funding

J.C. receives support from the St. Jude Comprehensive Cancer Center (NCI grant P30CA021765), NCI Program Project P01CA096832, the V Foundation, and American Lebanese Syrian Associated Charities. This research was funded in part through the NIH/NCI Cancer Center Support Grant P30 CA008748 to Memorial Sloan Kettering Cancer Center.

Acknowledgments

We thank Emily Walker and Sai Prabhas Konduru for their technical assistance in MethylationEPIC array and RNA-seq, respectively.

Conflict of interest statement

The authors declare no competing interests.

Authorship statement

Conceived and designed the study: M.K.E., P.R.B., D.A.W., and J.C. Performed experiments and data analysis: A.N.S., E.H., M.F.C., S.P., Q.Z., X.L., S.K.D., P.R.B., D.A.W., and J.C. Provided clinical information: E.H., S.P., C.L.T., G.R., D.L., R.K.R., A.V., M.A.K., G.W.R., and J.C. Drafted the manuscript: A.N.S., E.H., M.F.C., M.K.E., S.P., Q.Z., P.R.B., D.A.W., and J.C. All authors have read and approved the final version.

Data Availability

Once the manuscript is published, the raw and processed data will be available at the St. Jude Cloud23 (https://www.stjude.cloud/).

Ethics Approval

This study was approved by St. Jude Children’s Research Hospital’s institutional review board (IRB 24-1601).

References

Author notes