Noncanonical activation of GLI signaling in SOX2⁺ cells drives medulloblastoma relapse

Abstract

SRY (sex determining region Y)–box 2 (SOX2)–labeled cells play key roles in chemoresistance and tumor relapse; thus, it is critical to elucidate the mechanisms propagating them. Single-cell transcriptomic analyses of the most common malignant pediatric brain tumor, medulloblastoma (MB), revealed the existence of astrocytic Sox2⁺ cells expressing sonic hedgehog (SHH) signaling biomarkers. Treatment with vismodegib, an SHH inhibitor that acts on Smoothened (Smo), led to increases in astrocyte-like Sox2⁺ cells. Using SOX2-enriched MB cultures, we observed that SOX2⁺ cells required SHH signaling to propagate, and unlike in the proliferative tumor bulk, the SHH pathway was activated in these cells downstream of Smo in an MYC-dependent manner. Functionally different GLI inhibitors depleted vismodegib-resistant SOX2⁺ cells from MB tissues, reduced their ability to further engraft in vivo, and increased symptom-free survival. Our results emphasize the promise of therapies targeting GLI to deplete SOX2⁺ cells and provide stable tumor remission.

Noncanonical activation of sonic hedgehog signaling drives the propagation of a SOX2⁺ population to promote tumor relapse.

INTRODUCTION

Brain tumors are the leading cause of cancer-related death in children, with medulloblastoma (MB) being the most common type (1). Genome-based classification has enabled the stratification of MB into four major distinct subgroups (2), each with its own molecular drivers and prognoses. Children within the sonic hedgehog (SHH) subgroup represent 30% of all MB cases. This subset of tumors is characterized by the constitutive activation of SHH signaling in the granular cell precursors (GCPs) of the cerebellum (3–6). Such activation is frequently due to mutations in the gene encoding for the SHH receptor PATCHED-1 (PTCH1) (7–9). Stimulation of this receptor by SHH ligand binding causes derepression of the heterotrimeric GTP-binding protein–coupled receptor Smoothened (SMO) that ultimately alters the activity, location, and stability of the three downstream glioma associated oncogene (GLI) family members: GLI1, GLI2, and GLI3 (10, 11). GLI1 and GLI2 mostly function as transcriptional activators, while GLI3 mainly plays a repressor role. Noncanonical activation of SHH signaling is also frequently observed downstream of SMO and is driven by gain of oncogenic proteins or loss of tumor suppressors (10, 12). To date, most of the compounds that block SHH signaling target the pivotal upstream activator component SMO (11). Two of these inhibitors, vismodegib and sonidegib, are undergoing clinical evaluation in patients with MB (13). However, rapid tumor recurrence is frequently observed in patients treated with these inhibitors (14–17).

Recent work suggests that SHH MB arises from a rare population of MB-propagating cells (MPC) expressing the transcription factor SRY (sex determining region Y)–box 2 (SOX2) (18, 19). These cells actively divide during tumor initiation and hierarchically contribute to the hyperplasia of the GCPs before acquiring a stem-like quiescent transcriptional profile in full-blown tumors (18–20). The versatile nature of these cells allows them to give rise to more differentiated transit-amplifying progenies, evade therapies, and initiate tumor relapses (18, 19). Accordingly, SOX2⁺ cells have enhanced abilities to engraft in vivo and to escape various chemotherapeutics (21), including SMO inhibitors (19). Despite the central role that SOX2⁺ cells play in treatment resistance and tumor relapse, the underlying mechanisms controlling their regulation remain largely unknown.

Because of the resistance to SMO inhibition exhibited by SOX2⁺ MB cells (19), we speculated that SHH signaling would be inactive in these cells. Unexpectedly, analyses of single-cell transcriptomic [single-cell RNA sequencing (scRNA-seq)] data (20) revealed that SHH effectors were expressed in a Sox2-enriched astrocyte-like population. In this particular cell cluster, the numbers of cells expressing Sox2 increased upon vismodegib exposure. Moreover, despite expressing biomarkers indicative of SHH signaling activation, expression of the pathway effectors Gli1/Gli2 was increased by SMO inhibition in this subset of Sox2⁺ cells. These data suggested that activation of SHH signaling in astrocyte-like MB cells might occur in an SMO-independent manner. Using SOX2⁺ cell–enriched MPC cultures (22), we now show that the propagation of these cells is resistant to SMO inhibitors but sensitive to compounds targeting GLI family members. We further show that GLI activation in these cells is driven, at least in part, by MYC signaling. Moreover, in vivo attenuation of SHH signaling downstream of SMO in response to administration of functionally distinct GLI inhibitors (23–26) reduced the numbers of vismodegib-resistant SOX2⁺ cells in residual tumors and impeded their propagation in naïve mouse recipients. Our current results imply that relapses observed in patients treated with SMO inhibitors (14, 16) might derive from the propagation of a vismodegib-resistant SOX2⁺ population, which is targetable by inhibiting SHH signaling at the level of GLI. Accordingly, treatment with a clinically relevant GLI inhibitor not only depleted SOX2⁺ MB cells from tumor tissues but also increased the overall survival of mice harboring MB.

RESULTS

A subset of SOX2⁺ MB cells express biomarkers indicative of functional SHH signaling

Because of their resistance to SMO inhibition (19), we hypothesized that SOX2⁺ cells display low basal expression of SHH biomarkers, suggesting a nonfunctional pathway. Differential expression analyses for SHH pathway biomarkers using transcriptional data from a Sox2-eGFP; Ptch1^+/− mouse MB model (19) were performed. Although expression of some components of the SHH signaling pathway appears to be down-regulated in SOX2⁺ sorted cells, statistically significant differences were only found in zinc finger protein of the cerebellum 2 (Zic2), whose expression was lower in SOX2⁺ cells (Fig. 1A and fig. S1). These data suggest that SOX2⁺ MB cells might retain some level of SHH signaling activation. Accordingly, 41 and 61% of SOX2⁺ MB cells found in Ptch1 and in Smo mutant tumors, respectively, were also labeled for a major SHH activator, GLI2 (Fig. 1B and figs. S2, A to D, and S3A). Similar double-labeled cells (19% in RCMB18 and 24% in SJSHHMB-14-7196 (SJSHHMB) were found in two different SHH subgroup patient-derived orthotopic xenografts (PDOXs) and in a human MB tumor microarray (TMA) (Fig. 1B and fig. S3, A and B). In this TMA, 35% of biopsies showed broad nuclear GLI1, which is consistent with the frequency of patients with SHH subgroup MB (2). Eighty-five percent of these GLI1⁺ biopsies were also positive for GLI2, and 66% of these GLI1/2⁺ tissues harbored SOX2⁺ cells also expressing GLI2 (21% of the SOX2⁺ cell count) (fig. S3, B and C). Together, these data indicate that, despite the heterogenicity found in human disease, in more than the half of GLI-expressing MB tumors, a subset of SOX2⁺ cells express biomarkers suggestive of an active SHH signaling pathway.

Fig. 1. A subset of SOX2⁺ MB cells express biomarkers indicative of functional SHH signaling.

(A) Gene expression profiling from SOX2-eGFP⁺ and eGFP⁻ sorted cells isolated from tumors developed in eGFP-Sox2; Ptch1^+/− mice (RNA-seq data obtained from Dirks, 2014 dataset) was analyzed using a Kyoto Encyclopedia of Genes and Genomes (KEGG) Hedgehog signaling signature. A heatmap comparing gene expression is shown. Arrowhead indicates differentially expressed gene (P = 0.014). (B) Spontaneous MB tissues developed in Ptch1-Gfap or ND2:SmoA1 mice, two independent PDOX-derived tissues [SJSHHMB-14-7196 (SJSHHMB) and RCMB18], or a human MB TMA were stained for GLI2 and SOX2. (C) Expression of Sox2 in a t-SNE projection of cells was visualized by analyzing previously run scRNA-seq data from spontaneous SmoM2 MB (sequencing data from Ocasio, 2019 dataset). (D) A similar scRNA-seq dataset was used to show expression of SHH biomarkers across indicated nodes. (E) Spontaneous MB developed in Ptch1-Gfap mice, from mice harboring an SHH subgroup PDOX (RCMB18) or from a TMA were stained for SOX2 and GFAP. (F) Percentages of Sox2⁺ cells from SmoM2 tumors treated with vehicle or vismodegib in the indicated cell clusters were obtained using Ocasio’s transcriptomic data. The numbers of Sox2⁺ cells were normalized to total sample cell count. Data shown represent the average percentage for each sequenced tumor. (G) Percentages of Gli1/Gli2⁺ cells in vehicle or vismodegib-treated SmoM2 tumors were similarly calculated. (H) A density blot showing Gli1/Gli2/Sox2 expression in a t-SNE projection of cells from similarly treated tumors. Square details density on astrocytic cluster. Representative IHCs, with highlighted double-labeled cells, are shown. Scale bars, 50 μm. *P ≤ 0.05.

To better characterize this SOX2⁺ cell population expressing biomarkers suggestive of SHH signaling activation, we explored previously generated scRNA-seq data (20). Analyses performed in a Smo-driven mouse MB model (SmoM2-eYFP^loxP/loxP; Math1-Cre) indicated that Sox2 expression was mainly localized in two independent cell clusters with either an oligodendrocyte- or an astrocyte-like transcriptome (Fig. 1C) (20). Although some cells expressing SHH biomarkers might fall under the threshold of detection in scRNA-seq analyses, expression of SHH biomarkers was detected not only in the previously described proliferative nodes A and B (20) but also in the Sox2-enriched astrocyte compartment (Fig. 1D). Presence of Sox2⁺ astrocytic cells in SHH MB was confirmed upon staining mouse- and human-derived MB tissues for SOX2 and astrocyte markers (Fig. 1E and fig. S4A). Moreover, although, in our TMA, the presence of SOX2⁺ cells expressing the astrocyte marker glial fibrillary acidic protein (GFAP) was not unique to GLI1/GLI2⁺ biopsies, in these tissues, ~25% of SOX2⁺ cells were also expressing GFAP (Fig. 1E and fig. S3, B and C).

Because the expression of SHH biomarkers, astrocytic Sox2⁺ MB cells might be driven by SHH signaling. Analyses of similar publicly available scRNA-seq data obtained from MB harboring SmoM2 mice treated with either vehicle or vismodegib (20) showed a decrease in cells expressing Sox2 in nodes A and B, as well as in the oligodendrocytic cluster, while an increase was observed in the astrocyte-like one (Fig. 1F), suggesting that these cells are resistant to vismodegib. In these tumor tissues, a reduction in levels (fig. S4B) and percentages (fig. S4, C and D) of either Gli1⁺ or Gli2⁺ cells in nodes A and B, as well as in the Sox2-enriched oligodendrocytic cluster was observed. In contrast, examination of the astrocytic compartment revealed an increase of Gli1/Gli2-expressing cells (Fig. 1G). Further supporting their resistance to vismodegib, MB tissues exposed to this compound showed an increase in Sox2/Gli1/Gli2 expression density in the astrocytic cluster (Fig. 1H). These data reveal that vismodegib enriches MB tissues for a population of SMO inhibitor–resistant astrocyte-like Sox2⁺ cells that seems to retain SHH signaling.

GLI signaling drives the growth of SOX2-enriched MB cultures

As activation of SHH signaling downstream of SMO is a common mechanism of resistance to vismodegib in MB and basal cell carcinoma (16, 17), we hypothesized that in SOX2⁺ MB cells, SHH signaling is activated in a noncanonical manner downstream of SMO. To test this, we used previously described MPC cultures isolated from a Ptch1-LacZ; Trp53 KO murine MB model and enriched for SOX2 expression (22). Recapitulation of our scRNA-seq analyses showed that a percentage of the SOX2⁺ cells in culture also exhibited GFAP, GLI1, and GLI2 labeling (Fig. 2A). Despite expressing SHH downstream effectors, concentrations of SMO inhibitors affecting the viability of vismodegib-sensitive SHH MB cells (fig. S5A) (27, 28) had minimal effects on MPC culture proliferation and did not result in increased apoptosis (Fig. 2B and fig. S5B). These results, as well as data from other previously reported research (18, 19, 22), supported our scRNA-seq analyses, suggesting that SOX2⁺ MB cells are resistant to SMO inhibition. To determine whether activation of SHH signaling in SOX2⁺ cells occurs downstream of SMO, we exposed MPC cultures to GLI antagonist-61 (GANT-61), a hexahydropyrimidine derivative, which inhibits GLI1- and GLI2-driven transcription (26). GANT-61 attenuated the expression of the SHH target genes (Fig. 2C) and GLI1 and GLI2 levels (fig. S5C), while it reduced the viability of these cultures (Fig. 2D). Moreover, an additional GLI inhibitor, I-BET bromodomain inhibitor 151 (BET151), which blocks SHH signaling by impeding the binding of the bromodomain containing 4 (BRD4) protein to the Gli1 locus (23–25), similarly reduced SHH target gene expression (Fig. 2E) and decreased the number of viable MPCs (Fig. 2F); at the same concentrations, it was previously shown to block SHH signaling (23, 25, 30). Specifically, I-BET151 reduced the ability of these cultures to proliferate and increased apoptosis (Fig. 2G and fig. S5D), which reduced not only the overall numbers of SOX2⁺ cells in culture but also those coexpressing GLI1, GLI2, and GFAP (Fig. 2H). Consistent with I-BET151 targeting the vismodegib-resistant astrocyte-like Sox2⁺ cells identified in our scRNA-seq analyses, this compound reduced the percentage of MPCs expressing astrocyte and SHH biomarkers (Fig. 2I and fig. S5E).

Fig. 2. GLI signaling drives the growth of SOX2-enriched MB cultures.

(A) Indicated proteins were stained in MPC1 cultures, and percentages of positive cells were determined by flow cytometry analyses. (B) MPC2 cultures were exposed to cyclopamine (10 μM) or its inactive analog tomatidine (10 μM), vismodegib (100 nM), or dimethyl sulfoxide (DMSO) for 16 hours before staining for bromodeoxyuridine (BrdU) and cleaved caspase-3. (C) MPC2 cultures were exposed to GANT-61 for 16 hours, and expression of indicated genes was determined. (D) Indicated cultures were exposed to GANT-61 for 72 hours before determining cell viability by MTT [3-(4,5-dimethyl-2-thiazolyl) 2,5-diphenyl-2H-tetrazolium bromide] reduction. (E) MPC1 cultures were exposed to I-BET151 for 16 hours before determining the expression of indicated genes. (F) MPC cultures were exposed to I-BET151 for 72 hours before determining cell viability by MTT reduction. (G) MPC1 cultures were exposed to I-BET151 (500 nM) for 16 hours before assay BrdU incorporation and cleaved caspase-3 expression. (H) MPC1 cultures were exposed to I-BET151 (500 nM) for 24 hours, and percentages of positive cells were determined by flow cytometry. (I) Percentage of GFAP⁺ cells in a SOX2/GLI2⁺ pool was similarly determined. (J) MPC2 cultures were transfected with pooled Genome siRNA sequences targeting the indicated genes or a scramble siRNA control (siSC), and expression of SHH target genes was determined 72 hours later. (K) MPC2 cultures were transfected with similar siRNAs or a GFP-labeled siRNA. Cell viability was determined 5 days later by MTT reduction. (L) MPC1 cells were similarly transfected, and numbers of SOX2⁺ cells were determined by flow cytometry. (M) MPC1 cultures were transfected with indicated vectors and, 48 hours later, exposed to I-BET151 (500 nM) for additional 72 hours. Cell viability was determined by MTT reduction. (N) MPC1 cultures were transfected with indicated Genome siRNA sequences. Forty-eight hours after transfection, cells were exposed to I-BET151 (500 nM) for additional 72 hours, and cell viability was similarly determined. Scale bars, 100 μm. *P ≤ 0.05.

To validate our GLI inhibitor data, we attenuated the expression of Smo and the downstream SHH effectors, Gli1 and Gli2, in MPC cultures using pooled siRNA sequences. Silencing Smo (fig. S5F) had minimal effects on the expression of SHH target genes (Fig. 2J). However, silencing Gli2 (fig. S5G) attenuated expression of Gli1 and Ptch2 (Fig. 2J), while Gli1 silencing down-regulated Ptch2 expression (Fig. 2J) without affecting Gli2 (fig. S5H). Consistent with the notion that GLI signaling controls the proliferation of SOX2-enriched MPC cultures, attenuation of Gli1 and Gli2, but not Smo, partially reduced the number of viable MPC cells (Fig. 2K). Incomplete reduction of numbers of viable cells in these experiments might be due to a partial gene knockdown (Fig. 2J and fig. S5, F to H), likely resulting from a suboptimal transfection rate in sphere cultures or by the presence of a subset of non–GLI-expressing cells in MPC cultures as shown in Fig. 2A. Validating the on-target effect of our small interfering RNA (siRNA) data, an independent set of siRNA pooled sequences targeting Gli1 and Gli2 similarly reduced SHH target gene expression (fig. S5, I and J) and the viability of MPC cultures (fig. S5K). Last, flow cytometry analyses showed that, like I-BET151, siRNA targeting either Gli1 or Gli2 reduced the numbers of SOX2⁺ cells in culture (Fig. 2L). Besides Gli1, BET inhibition regulates the promoter activity of other well-known oncogenes (30–32). Thus, to determine whether I-BET151 acts on GLI in our system, we either overexpressed or silenced the positive SHH regulators Gli1 and Gli2 in MPC cultures before quantifying the effect of I-BET151 on cell viability. Overexpression of any of these GLI family members rescued the effect of this BET inhibitor in cell viability (Fig. 2M). In contrast, silencing of either Gli1 or Gli2 on MPC cultures did not have an additive effect to that of I-BET151, suggesting that this drug is targeting a GLI-dependent cell population (Fig. 2N). Together, these data suggest that the proliferation of SOX2-enriched MB cultures might require noncanonical SHH signaling involving GLI transcription factors.

Activation of GLI signaling in SOX2-enriched MB cultures is dependent on MYC

Previous evidence suggests that an increase in the tumor oncogenic load, elicited by the gain of oncogenes or the loss of a tumor suppressor such as P53, triggers noncanonical GLI signaling (10, 12, 33–35). Our previous data in MPC cultures were performed in cells isolated from Ptch1-LacZ; Trp53 KO mice (MPC1 and MPC2). We therefore studied whether activation of SHH signaling downstream of SMO was unique to cultures lacking Trp53. GLI inhibition using GANT-61 or I-BET151 reduced the proliferation (Fig. 3A) and SHH target gene expression (Fig. 3, B and C) of MPC cultures isolated from an MB developed in Ptch1-LacZ mice that are wild type (WT) for Trp53 (MPC47). Moreover, siRNA-mediated knockdown of Gli1 or Gli2, but not that of Smo, also reduced SHH biomarkers (Fig. 3D and fig. S6A) and proliferation (Fig. 3E) on similar WT Trp53 cultures. Thus, the loss of Trp53 does not seem to be required to trigger noncanonical GLI signaling in SOX2-enriched MPC cultures.

Fig. 3. Activation of GLI signaling in SOX2-enriched MB cultures is dependent on MYC.

(A) MPC47 cultures were exposed to GANT-61 or I-BET151 for 72 hours, and cell viability was determined by MTT reduction. (B) MPC47 cultures were exposed to GANT-61 for 16 hours, and expression of indicated genes was determined. (C) MPC47 cultures were exposed to I-BET151 for 16 hours, and SHH target gene expression was determined by RT-qPCR. (D) MPC47 cultures were transfected with indicated pooled Genome siRNA sequences, and expression of indicated genes was similarly determined 72 hours later. (E) MPC47 cultures were transfected with similar siRNA sequences or a GFP-siRNA. Cell viability was determined 5 days later by MTT reduction. (F) MPC cultures were exposed to 10058-F4 for 16 hours, and SHH target gene expression was determined. (G) Indicated cultures were exposed to 10058-F4 for 48 hours before determining cell viability by MTT reduction. (H) MPC2 cultures were transfected with pooled Genome siRNA sequences, and expression of Myc was determined 72 hours later. (I) MPC2 cultures were transfected with similar siRNA sequences, and expression of indicated genes was determined 72 hours later. (J) MPC cultures were similarly transfected, and cell viability was determined 5 days later by MTT reduction. (K) MPC1 cultures were transfected with pooled Genome siRNA sequences for 5 days before quantify percentages of SOX2⁺ cells by flow cytometry. (L) Indicated proteins were stained in ND2:SmoA1 brains. Square details double-positive cell in a representative image. Scale bars, 50 μm. (M) scRNA-seq data from SmoM2 mice exposed to vehicle or vismodegib (Ocasio’s dataset) were used to study the density of Sox2/Gli1/Gli2/Myc expression in a t-SNE plot. (N) Schematic of canonical and MYC-driven noncanonical SHH signaling in SOX2⁺ MB cells. *P ≤ 0.05.

In addition to loss of P53, noncanonical activation of GLI signaling might be triggered by oncogenes including MYC (10, 35, 36). Thus, we studied whether MYC could be regulating GLI signaling in SOX2-enriched MB cultures. Inhibition of MYC transcriptional activity using the small-molecule inhibitor 10048-F4, which blocks the binding of MYC to its binding partner MAX (37), reduced expression of SHH target genes in our SOX2-enriched MPC cultures (Fig. 3F). This molecule also reduced MPC culture proliferation (Fig. 3G) at concentrations similar to those described before to block MYC signaling (37–39). Validating 10048-F4 data, attenuation of Myc expression using two independent siRNA smart pools (Fig. 3H and fig. S6B) reduced the expression of SHH target genes in these cells (Fig. 3I and fig. S6B), as well as their overall viability (Fig. 3J and fig. S6C) and specific numbers of SOX2 expressing cells (Fig. 3K). Similar reduction in SHH target genes (fig. S6D) was observed by using a CRISPR-Cas9–mediated approach to silence Myc (fig. S6E). In line with a vismodegib-resistant astrocyte-like SOX2⁺ population in which noncanonical SHH signaling is activated by MYC, in spontaneous SmoA1 mutant MB tissues, expression of MYC was found in a small number of cells labeled for the astrocyte biomarker GFAP and in SOX2⁺ cells (Fig. 3L). Moreover, while our analyses of scRNA-seq data performed in SmoM2 mice showed a reduction in Myc expression on proliferative nodes A and B in tumors exposed to vismodegib (fig. S6F), a specific enrichment in Sox2/Gli1/Gli2/Myc density in the astrocytic cluster was observed (Fig. 3M). As illustrated in Fig. 3N, these results show that, although additional signaling mechanisms might be involved, MYC seems to activate SHH signaling downstream of SMO in SOX2-enriched MPC cultures.

In vivo inhibition of SHH signaling downstream of SMO depletes SOX2⁺ MB cells

We have shown that SHH MB harbors a SOX2⁺ subpopulation that is resistant to SMO inhibitors but relies on GLI signaling for ex vivo growth. We therefore studied whether attenuation of SHH signaling downstream of SMO would similarly target this SOX2⁺ cell population in vivo. Freshly isolated MB cells from Ptch1-LacZ and Ptch1-LacZ; Trp53 KO mice devoid of any expansion in culture were subcutaneously implanted in mice. Because of the mixed genetic background of these cells, immunocompromised mice were used. SHH signaling was inhibited at the level of GLI in these mice by administrating I-BET151 (Fig. 4A). As proliferation of these tumors is driven by lack of PTCH1 activity (9), MB growth was significantly reduced by this drug (Fig. 4B). In I-BET151–treated mice, the overall SOX2 levels (Fig. 4A) and the number of SOX2⁺ cells (Fig. 4C) were reduced in residual tumors. These data suggest that inhibition of SHH signaling downstream of SMO depletes a SOX2⁺ population in vivo. Because I-BET151 can regulate the activity of several promoters (30–32) and therefore might be acting independently of GLI in vivo, we treated similar Ptch1-LacZ mutant tumors with the functionally different GLI inhibitor Gant-61 (26). Similar to I-BET151, Gant-61 attenuated tumor growth (Fig. 4D) and reduced the levels of the final SHH effectors GLI1 and GLI2, as well as overall SOX2 levels (Fig. 4E) and cell numbers (Fig. 4F).

Fig. 4. In vivo inhibition of SHH signaling downstream of SMO depletes SOX2⁺ MB cells.

(A) Spontaneous tumors developed in Ptch1-LacZ; Trp53-KO (MB1) or Ptch1-LacZ (MB47) mice were subcutaneously implanted into mice. Once the tumor size reached approximately 100 mm³, mice were treated with I-BET151 [30 mg/kg, intraperitoneally (i.p.), daily (q.d.)] or vehicle for 5 days, and levels of indicated proteins were determined by immunoblotting. (B) Tumor volume was measured in similarly treated mice. (C) MB1 and MB47 tumor tissues were similarly exposed to I-BET151, and residual tumors were stained for SOX2 and number of positive cells per field quantified. (D) Mice harboring MB1 subcutaneous tumors were treated with Gant-61 (40 mg/kg, subcutaneously, every other day) or vehicle for 8 days. Tumor volume was measured every other day. (E) Mice harboring MB1 tumors were similarly exposed to Gant-61, and levels of indicated proteins were determined by immunoblotting. (F) MB1 tumor tissues were similarly exposed to Gant-61; residual tumors were stained for SOX2, and the number of positive cells per field was quantified. Representative images are shown. Scale bars, 50 μm. *P ≤ 0.05.

GLI inhibition attenuates SHH MB progression

Given that neither I-BET151 (40) nor Gant-61 (41) are likely to cross the blood-brain barrier, limiting their efficacy to subcutaneous MB models, we turned to the drug JQ-1 as an alternative BET inhibitor, which could demonstrate efficacy in the brain (42). Similar to I-BET151 or Gant-61, JQ-1 attenuated the proliferation (Fig. 5A) and inhibited SHH signaling (Fig. 5B) of SOX2⁺ cell–enriched MPC cultures and did this, at least in part, via a GLI-dependent manner (fig. S7A). Consistent with an SHH inhibitor, in a Ptch1-LacZ in vivo MB model, JQ-1 reduced the expression of SHH biomarkers (Fig. 5C) and target genes (Fig. 5D). When comparing the activity of JQ-1 to that of vismodegib, both decreased the size of similar SHH-driven tumors, either when implanted subcutaneously (Fig. 5E) or orthotopically in the brain (Fig. 5F). Consistent with our scRNA-seq analyses, a significant enrichment in cells expressing SOX2 was observed in vismodegib treated tumors (Fig. 5G). On the contrary, tumor tissues exposed to JQ-1 showed a reduction in number (Fig. 5G) and overall protein levels (Fig. 5C) of SOX2. Flow cytometry analyses performed in mice exposed to either vehicle or JQ-1 showed similar decrease in numbers of SOX2⁺ cells, including those expressing either GLI1 or GLI2 (Fig. 5H). Last, consistent with JQ-1 targeting the previously described pool of astrocyte-like SOX2⁺ cells, this BET inhibitor also depleted from tumors those SOX2⁺ cells labeled with astrocyte and SHH biomarkers (Fig. 5I). These results further suggest that SHH MB harbors a pool of vismodegib-resistant SOX2⁺ MB cells that can be targeted by inhibiting SHH signaling downstream of SMO.

Fig. 5. GLI inhibition attenuates SHH MB progression.

(A) MPC cultures were exposed to JQ-1 for 72 hours before determining cell viability by MTT reduction. (B) MPC47 cells were exposed to JQ-1, and expression of indicated genes was determined 16 hours later. (C) Mice harboring Ptch1-LacZ (MB47) subcutaneous tumors were treated with JQ-1 (30 mg/kg, i.p., q.d.) or vehicle for 8 days, before determining levels of the indicated proteins. (D) Expression of indicated genes was determined in similarly dosed mice. (E) Mice harboring similar tumors were treated with JQ-1 (30 mg/kg, i.p., q.d.), vismodegib (25 mg/kg, i.p., q.d.), or vehicle for 8 days, and the tumor volume was measured. (F) Mice harboring orthotopic MB47 tumors were similarly dosed, and tumor area was quantified. Scale bars, 200 μm. (G) Number of SOX2⁺ cells in brain tumors from similarly treated mice were quantified. (H) Mice harboring subcutaneous MB47 tumors were treated with vehicle or JQ-1 (30 mg/kg, i.p., q.d.) for 8 days, before determining percentage of positive cells by flow cytometry. (I) Percentages of GFAP⁺ in SOX2/GLI2⁺ cells were similarly determined. (J) Mice harboring subcutaneous MB47 were treated with JQ-1 (30 mg/kg, i.p., q.d.), vismodegib (25 mg/kg, i.p., q.d.), or vehicle for 8 days. Equal numbers of viable cells from residual tumors were then allowed to form spheres ex vivo or to orthotopically engraft in vivo. A schematic of the procedure is shown. (K) Numbers of spheres grown from treated tumors described in (J) were quantified. (L) Tumor engraftment capability from similar residual tumors was determined. (M) PDOX harboring mice were dosed with vehicle, vismodegib (25 mg/kg, i.p., q.d.), or JQ-1 (30 mg/kg, i.p., q.d.) for 8 days, before staining brains for SOX2. Unless otherwise indicated, scale bars, 50 μm. *P ≤ 0.05.

Given that vismodegib-treated patients tend to relapse (14–17), we questioned whether the vismodegib-resistant SOX2/GLI⁺ cell population was the underlying reason for tumor regrowth and whether attenuation of SHH signaling downstream of SMO would impede it. To test this, mice harboring subcutaneous Ptch1-LacZ MB were dosed with vismodegib, JQ-1, or vehicle control for 8 days before the harvest of residual tumor tissues. An equal number of viable cells from these tumors were subsequently either plated or orthotopically implanted (Fig. 5J). Treatment with JQ-1 reduced primary sphere formation, while vismodegib-treated residual tumors promoted a greater number of spheres compared to vehicle-treated tissues (Fig. 5K). Similar to this ex vivo observation, JQ-1 treatment partly prevented secondary tumor engraftment, while mice implanted with residual vismodegib-treated tumors had a worse outcome compared to their vehicle-treated counterparts (Fig. 5L). As high expression of SOX2 correlates with poor outcomes in patients with MB (19), we studied whether tumors grown from vismodegib preexposed tissues in Fig. 5L overexpressed SOX2. Such reengrafted tumors were indeed enriched in cells expressing SOX2 compared to those treated with JQ-1 or vehicle (fig. S7B). Collectively, our data show that vismodegib-exposed residual MB gives rise to aggressive SOX2⁺ cell–enriched tumors, whose propagation could be prevented by inhibiting SHH signaling downstream of SMO. Moreover, according to our observations in mouse MB tissues, treatment of mice harboring an SHH subgroup PDOX with JQ-1 similarly reduced the number of SOX2⁺ cells while vismodegib failed to do so (Fig. 5M).

The BET inhibitor BMS-986158 attenuates SHH signaling and Ptch1 driven MB growth

Our data suggest that, because of the depletion of the SOX2 cell population from tumors, inhibition of SHH signaling at the level of GLI would provide a more stable disease remission than targeting SMO. Limiting the translational relevance of our data, none of the GLI inhibitors used thus far have a pharmacokinetic profile suitable for brain tumor patients, as neither I-BET151 (40) nor Gant-61 (41) are brain penetrant and JQ-1 has a short half-life (43). Therefore, we studied whether the BET inhibitor, BMS-986158, currently undergoing clinical evaluation on pediatric and adult malignancies (NCT02419417, NCT03936465, and NCT04817007) could similarly block SHH signaling at the level of GLI and therefore provide long-lasting MB remission. Low nanomolar concentrations of this compound attenuated Gli1 promoter activity in an SHH reporter cell line (Fig. 6A), as well as the expression of the SHH biomarker GLI1 (Fig. 6B) and SHH target genes (Fig. 6C) in immortalized mouse embryonic fibroblasts (MEFs) (44) in which SHH signaling was induced by using an SMO agonist (SAG). Consistent with its predicted mechanism of action on GLI, BMS-986158 also attenuated SHH signaling in immortalized MEFs lacking Sufu expression (Sufu^−/− MEFs), which therefore harbor SHH signaling constitutively active downstream of SMO (Fig. 6D) (45), and reduced the enrichment of BRD4 in the Gli1 locus (Fig. 6E). This BMS compound also attenuated the proliferation of cerebellar GCPs, which multiply in response to SHH signaling activation without inducing cell death in either induced (Fig. 6F) or uninduced (fig. S7C) GCPs.

Fig. 6. The BET inhibitor BMS-986158 attenuates SHH signaling and Ptch1 driven MB growth.

(A) Light2 cells were exposed to 100 nM SAG for 24 h, before exposing them to BMS-986158. GLI- driven luciferase activity was determined 16 hours later. (B) WT MEFs were similarly treated, before immunoblot protein lysates for indicated proteins. (C) Expression of the indicated genes in similarly treated MEFs was determined. (D) Sufu^−/− MEFs were treated with BMS-986158 for 16 hours, and Gli1 expression was determined. (E) WT MEFs were treated with SAG (100 nM) for 24 hours before their exposure to BMS-986158 for additional 16 hours. BRD4 occupancy in Gli1 locus was analyzed by ChIP-PCR and normalized to a ChIP performed using rabbit IgG. Primer set 3 (PS3) was used as a negative control, while PS6 to PS8 aligned with Gli1 promoter. (F) GCPs were induced with 100 nM SAG for 24 hours and exposed to BMS-986158 for additional 16 hours, before assay BrdU incorporation and stain-cleaved caspase-3. Scale bar, 100 μm. (G) Vismodegib-sensitive SHH-47 cells were exposed to BMS-986158 for 16 hours, and expression of indicated proteins was determined. (H) Expression of indicated genes was assayed in similarly treated cells. (I) Indicated cells were exposed to BMS-986158 for 72 hours, and cell viability was determined by MTT reduction. (J) Mice harboring subcutaneous MB47 (Ptch1-LacZ) and MB1 (Ptch1-LacZ, Trp53-KO) tumors were dosed with BMS-986158 (3 mg/kg, i.p., q.d.) or vehicle for 8 days, before determining levels of indicated proteins. (K) Tumor size in similarly treated mice was determined. (L) Ten-day-old Ptch1-Gfap mice were dosed for 8 days with vehicle or BMS-986158 (3 mg/kg, i.p., q.d.), and the size of tumors was quantified. Scale bar, 200 μm. (M) Brains from similarly treated Ptch1-Gfap mice were stained for the indicated proteins. Scale bars, 50 μm. *P ≤ 0.05.

In tumor cells, BMS-986158 attenuated SHH signaling (Fig. 6, G and H) and the ex vivo growth (Fig. 6I) of previously described vismodegib-sensitive MB cultures (27, 28). In vivo, this BMS compound similarly reduced levels of SHH signaling biomarkers (Fig. 6J) and abrogated subcutaneous tumor growth (Fig. 6K). In a spontaneous Ptch1-driven mouse MB model (Ptch1-Gfap), BMS-986158 treatment similarly reduced the size of residual tumors (Fig. 6L), cutting off tumor proliferation, and increased the expression of apoptotic biomarkers (Fig. 6M). Together, these data show that use of a clinically relevant BET inhibitor leads to SHH signaling inhibition at the level of GLI, which, in turn, attenuates in vivo MB growth.

BMS-986158 exhausts a SOX2⁺ cell population and prevents MB relapse

Because BMS-986158 inhibits SHH signaling at the level of GLI, we speculated that this novel BET inhibitor would prevent SOX2⁺-driven MB relapse. Similar to the other GLI inhibitors tested, BMS-986158 attenuated SHH signaling (Fig. 7, A and B) and the proliferation (Fig. 7C) of vismodegib-resistant SOX2⁺ cell–enriched MB cultures, at least in part, via a GLI-dependent manner (fig. S7A). Consistent with our previous in vivo observations, GLI inhibition by BMS-986158 reduced the levels (Fig. 7D) and the number of SOX2 expressing cells (Fig. 7E). Similar to JQ-1, and as illustrated in Fig. 5J, residual tumors isolated from BMS-986158–treated mice made fewer primary spheres compared to those derived from vehicle-treated mice (Fig. 7F) and mostly failed to engraft in vivo (Fig. 7G).

Fig. 7. BMS-986158 exhausts a SOX2⁺ cell pool and prevents MB relapse.

(A) MPC47 cultures were exposed to BMS-986158 for 16 hours before determining expression of SHH target genes. (B) MPC1 cultures were similarly exposed to BMS-986158, and expression of indicated genes was determined. (C) MPC cultures were exposed to BMS-986158 for 72 hours before determining cell viability by MTT reduction. (D) Mice harboring subcutaneous MB47 (Ptch1-LacZ) and MB1 (Ptch1-LacZ, Trp53-KO) tumors were exposed to BMS-986158 (3 mg/kg, i.p., q.d.) or vehicle for 8 days, and levels of indicated proteins were determined. (E) Ten-day-old Ptch1-Gfap mice were treated with vehicle or BMS-986158 (3 mg/kg, i.p., q.d.) for 8 days, and SOX2 was detected by IHC staining. (F) Mice harboring subcutaneous MB47 were treated with BMS-986158 (3 mg/kg, i.p., q.d.) or vehicle for 8 days. As shown in Fig. 5J, equal number of viable cells from residual tumors were then allowed to form spheres ex vivo. (G) As represented in Fig. 5J tumor engraftment capability from similarly treated tumors was determined. (H) Ten-day-old Ptch1-Gfap mice were treated with BMS-986158 (3 mg/kg, i.p., q.d.) or vehicle for 8 days before labeling brain tissues with indicated metal conjugated antibodies. Representative multiplexing mass cytometry images and corresponding H&E staining are shown. Scale bars, 200 μm. (I) Ptch1-Gfap mice were similarly treated before staining brains for SOX2 and GFAP. Unless otherwise indicated, scale bars, 50 μm. *P ≤ 0.05.

Our scRNA-seq analyses suggested that SHH MB harbors a population of vismodegib-resistant SOX2⁺ cells with an astrocytic transcriptome. Thus, we wanted to know whether BMS-986158 efficacy correlated with a reduction of the astrocytic SOX2⁺ cell pool. As we described before, mass cytometry imaging of spontaneous Ptch1-driven tumors exposed to BMS-986158 showed a reduction in the expression of GLI2, the proliferation biomarker Ki67, and SOX2 labeling (Fig. 7H). Similar to our previous results in JQ-1–exposed tumors, a reduction of GFAP⁺ cells that also expressed SOX2 was observed in tissues treated with BMS-986158 (Fig. 7I). Thus, BMS-986158 drains the overall SOX2⁺ cell population, including astrocytic SOX2⁺ cells that were previously resistant to vismodegib treatment.

GLI inhibition provides stable tumor remission in SMO inhibitor resistant MB

Because of the mechanism of inhibition of SHH signaling by BMS-986158, we wondered whether it would show efficacy not only in vismodegib-sensitive Ptch1-driven MB but also in those that are vismodegib resistant in the clinic (46). In a spontaneous Smo mutant MB model, BMS-986158 reduced tumor size (Fig. 8A) and proliferation index, while it increased the number of apoptotic cells (Fig. 8B). In these tumors, we observed a reduction in the number of cells expressing the SHH effector GLI2 (Fig. 8B), which is consistent with an attenuation of SHH signaling. Moreover, residual SmoA1 MB tissues exposed to BMS-986158 had fewer SOX2⁺ cells than those exposed to vehicle control (Fig. 8B) and showed significantly longer symptom-free survival (Fig. 8C).

Fig. 8. GLI inhibition provides stable tumor remission in SMO inhibitor-resistant MB.

(A) Six-month-old ND2:SmoA1 mice were exposed to BMS-986158 (3 mg/kg, i.p., q.d.) for 8 days. The size of residual tumors was quantified. Scale bars, 200 μm. (B) Brains from similarly treated ND2:SmoA1 mice were stained for indicated proteins, and numbers of positive cells were quantified. (C) Two-month-old ND2:SmoA1 mice were exposed to BMS-986158 (3 mg/kg, i.p., q.d.) or vehicle, and symptom-free survival was determined. (D) Three weeks after orthotopically implanting SJSHHMB PDOX cells, mice were exposed to BMS-986158 (3 mg/kg, i.p., q.d.) or vehicle for 8 days before harvesting their brains, and the size residual tumors was measured. Scale bars, 200 μm. (E) Mice harboring SJSHHMB PDOX tissues were similarly treated, and brains were stained for indicated proteins. (F) Similarly implanted mice were treated with BMS-986158 (3 mg/kg, i.p., q.d.) or vehicle for 8 days before staining tissues for GLI1 and GLI2. (G) Mice implanted with either SJSHHMB or RCMB18 PDOX cells were similarly dosed before staining brain tissues for SOX2. (H) Mice orthotopically harboring a RCMB18 PDOX were dosed with BMS-986158 (3 mg/kg, i.p., q.d.) or vehicle, and symptom-free survival was determined. (I) Model illustrating cell diversity in SHH MB. Unless otherwise indicated, scale bars, 50 μm. *P ≤ 0.05.

Last, we studied the efficacy of BMS-986158 in treatment-resistant TP53 mutant and MYCN-amplified MB tissues, which is considered a very high-risk disease and has the worst outcome within the SHH subgroup of patients with MB (48). In these tumors, BMS-986158 reduced tumor size (Fig. 8D) and proliferation index while slightly increasing caspase-3 cleavage (Fig. 8E). Consistent with an SHH inhibitor, this BMS compound also reduced the numbers of cells expressing the SHH biomarkers GLI1 and GLI2 in residual tumor tissues (Fig. 8F). Similar to JQ-1, BMS-986158 reduced the number of SOX2⁺ MB cells in two independent PDOX models (Fig. 8G). Ultimately, BMS-986158 increased the life span of mice harboring these high-risk PDOX tumors (Fig. 8H), which further emphasized the clinical advantage and translational promise of using compounds that target GLI for the treatment of patients with MB.

DISCUSSION

Here, we show that SHH MB harbors a pool of SOX2-labeled MB cells in which expression of the final SHH signaling effectors Gli1 and Gli2 is increased by vismodegib. Consistent with SMO-independent activation of SHH signaling, the proliferation of SOX2-enriched MB cultures is not affected by SMO inhibition but rather by those compounds acting on GLI. According to this observation, in vivo inhibition of SHH signaling downstream of SMO reduces the number of SOX2⁺ cells, while SMO inhibitors exhibit an opposite effect. In accordance with the key role of SOX2⁺ cells in propagating MB (18, 19), GLI inhibition reduces the chances of tumor reengraftment, while vismodegib-exposed residual tumors become even more aggressive than their vehicle-treated counterparts. Combined, our results uncover a key driver of the SOX2⁺ cancer–propagating cell characterized by a noncanonical activation of GLI signaling, which might facilitate tumor propagation and likely underlies treatment resistance and subsequent cancer relapse seen in patients with SHH subgroup MB (14, 16).

Clinical observation of frequent tumor recurrence upon administration of SMO inhibitors (14–17) suggests that compensatory signaling pathways drive relapse mechanisms. Such relapses could be driven by treatment-resistant tumor-propagating cells labeled by stemness biomarkers. Similar to previous reports (19, 20), we observed that treatment with vismodegib increases the numbers of cells expressing SOX2 in residual tumor tissues. These data suggest that SOX2⁺ cells are resistant to SMO inhibition. In contrast, our current findings suggest that attenuation of GLI signaling using a clinically relevant BET inhibitor (BMS-986158) not only attenuates vismodegib-sensitive and vismodegib-resistant primary tumor growth but also partially reduces the chances of tumor recurrence. The fact that some tumors preexposed to BET inhibitors did engraft suggests either a partial depletion of the SOX2⁺ MB pool or additional signaling mechanisms allowing cancer relapse.

Our results in mouse- and human-derived MB models reveal that only a select number of SOX2⁺ MB cells express biomarkers indicative of SHH signaling activation. Similar GLI expressing SOX2⁺ cells were found in more than the half of GLI1/GLI2-labeled biopsies in an MB TMA. The partial presence of these cells in GLI⁺ biopsies might indicate their link to one or more of the four SHH MB subtypes (47) or could be used to predict sensitivities to BET inhibitors. Therefore, future analyses should be performed to address whether relapses of GLI⁺ tissues lacking SOX2/GLI⁺ cells could be prevented by BET inhibition. According to our scRNA-seq analyses, this subset of SOX2/GLI⁺ cells harbors an astrocytic signature. Although additional studies are required to better characterize these astrocyte-like SOX2⁺ MB cells, a similar population of SOX2⁺ astrocyte progenitors has been shown to promote malignancy in glioblastoma models (48) or can be transformed to trigger group 3 MB development (49). Further suggesting the key role of astrocyte-like cells in brain neoplasia, in SHH MB, GCPs transdifferentiate into tumor-associated astrocytes, which have been shown to then go on to coordinate with microglia to promote tumor growth (50).

The drivers of proliferation of SOX2⁺ cell–enriched MB cultures may be altered by the presence of trophic factors in growth media. Accordingly, fibroblast growth factor (FGF) blocks SHH signaling in MB cells (51), which may explain why SOX2⁺ cell–enriched MB cultures grown in FGF-containing media are resistant to SMO inhibitors. Nevertheless, this observation does not explain why in vivo exposure to vismodegib enriches tumor tissues that were never expanded ex vivo for SOX2-expressing cells or why these cells rely on GLI signaling to proliferate. An increase in the tumor oncogenic load might activate GLI signaling in an FGF-independent manner through the activity of oncogenic proteins (10). Fittingly, our data suggest, that in SOX2⁺ MB cells GLI signaling is activated in an MYC-dependent manner. It was previously shown that MYC interacts with the 5′-regulatory region of GLI1 to control GLI-driven transcription (35, 36) and that this regulation occurs through the interaction of the MYC-interacting zinc finger protein 1 with GLI2 (52). Our data reveal that, similar to GLI, attenuation of MYC signaling using an MYC inhibitor blocks the proliferation of SOX2⁺ cell–enriched MB cultures, signifying that SHH MB patients might benefit from the use of MYC inhibitors depleting aggressive SOX2⁺ cells. Here, we use BET inhibitors to disrupt SHH signaling downstream of SMO. As BET activity is similarly needed to sustain MYC signaling in MB tissues (53), the efficacy displayed by BET inhibitors on depleting SOX2⁺ cells might be due not only to their effect on GLI but also to that on MYC. Although our data suggest that GLI signaling is activated in an MYC-dependent manner in MPC cultures, the mechanism by which a small subset of MYC-expressing SOX2⁺ cells could be orchestrating MB relapse, as well as the roles that additional signaling mechanisms play, including Zic2, which was down-regulated in SOX2⁺ MB cells and shown to cooperate with GLI (54, 55), and control neurogenesis of the GCPs (56) should be further explored.

In conclusion, here, we demonstrate the existence of a SOX2⁺ cell reservoir with an astrocytic transcriptome that contributes to treatment resistance and cancer relapse. In these cells, SHH signaling is activated downstream of SMO. Thus, SMO inhibition causes the enrichment of residual MB tissues for SOX2⁺ cells, which increases the chances of tumor relapse and dictates poor patient outcome. Conversely, inhibition of SHH signaling at the level of GLI depletes the vismodegib-resistant SOX2⁺ reservoir and therefore improves patient outcome (Fig. 8I). Thus, inhibiting SHH signaling at the level of GLI has promising potential to improve survival in this challenging to treat MB patient population.

MATERIALS AND METHODS

Mouse studies

All mouse work was conducted in accordance with protocols approved by the Institutional Animal Care and Use Committee at the Medical University of South Carolina and the University of Miami. Ptch1-LacZ (Ptch1^1MPS/J), Trp53 KO (B6.129S2-Trp53^tm1Tyj/⁷), ND2:SmoA1 (C57BL/6-Tg(Neurod2-Smo*A1)199Jols/J), Ptch1LoxP (B6;129T2-Ptch1tm1Bjw/WreyJ), and Gfap-Cre (FVB-Tg(GFAP-cre)25Mes/J) strains were all obtained from the Jackson Laboratory and bred to generate the following colonies: Ptch1-LacZ, Ptch1-LacZ; Trp53 KO, ND2:SmoA1, and Ptch1-Gfap. Spontaneous tumors from Ptch1-LacZ (MB47: Ptch1^+/−) or Ptch1-LacZ; Trp53 KO (MB1: Ptch1^+/−; Trp53^+/−) mice were expanded and maintained as allografts in 6- to 8-week-old CD1-Foxn1^nu (Charles River Laboratories).

For in vivo subcutaneous tumor treatments, 1 × 10⁶ viable cells were implanted into the flanks of 6- to 8-week-old CD1-Foxn1^nu (CD-1) mice. Subcutaneous tumor volume was calculated using the formula (length × width²)/2. For orthotopic allograft experiments, 1 × 10⁵ viable cells were resuspended in 3 μl of Neurobasal-A serum-free media and implanted into the cerebella of 6- to 8-week-old CD1 mice as before (27). Expansion and treatment of PDOX: RCMB18 (57), a courtesy of Dr. R. Wechsler-Reya (Sandford Burnham Prebys, La Jolla, CA), and SJMBSHH-14-7196 (SJMBSHH) (58), a courtesy of Roussel before publication (St. Jude Children Research Hospital, Memphis, TN), was performed by similarly orthotopically implanting 1 × 10⁶ viable cells into the cerebella of 6- to 8-week-old NSG mice (NOD.Cg-Prkdcscid Il2rgtm1Wjl/SzJ mice). Two additional PDOX models, RCMB32, an SHH subtype from a Gorlin syndrome patient, and RCMB28, group 3 MB with an MYC amplification were also a courtesy of Wechsler-Reya (59) and used as controls for our immunohistochemistry (IHC) stainings.

All mice were dosed intraperitoneally and daily with indicated drugs: vismodegib (25 mg/kg), I-BET151 (30 mg/kg), JQ-1 (30 mg/kg), or BMS-986158 (3 mg/kg) dissolved in dimethyl sulfoxide to a final volume of 50 μl. Gant-61 was administrated subcutaneously every other day and dissolved in 50% ethanol and 50% corn oil. All compounds used for in vivo experiments were from Selleckchem, except BMS-986158, which was from BioVision.

For primary sphere formation and secondary tumor engraftment studies, mice harboring Ptch1-LacZ subcutaneous allografts (MB47) were treated for 8 days with vehicle, vismodegib (25 mg/kg), JQ-1 (30 mg/kg), or BMS-986158 (3 mg/kg). Residual tumors were harvested and then disaggregated mechanically and enzymatically using Accutase (Invitrogen) for 20 min. The viability of the resulting cell suspensions was evaluated by trypan blue exclusion (Invitrogen). For primary sphere formation evaluations, 3 × 10³ viable cells/μl from residual MB tissues were resuspended in MPC media containing Neurobasal-A, GlutaMAX, B27, N2, epidermal growth factor (20 ng/ml), FGF (20 ng/ml), and penicillin-streptomycin (all from Invitrogen) (22), and the ability of these cells to form spheres ex vivo was determined 1 week later. For secondary tumor engraftment experiments, 1 × 10⁵ viable cells obtained from similar residual tumors were resuspended in 3 μl of Neurobasal-A serum-free media and implanted into the cerebella of 6- to 8-week-old CD1 mice as before (27). Symptom-free survival was used to determine implantation rate. Mice were sacrificed once they developed signs of brain tumor, including dome-shaped cranium indicative of hydrocephalus, tilted head, hunching, circling, loss of weight, or ataxia. The presence of brain tumors was confirmed by hematoxylin and eosin (H&E) staining. The tumor area was quantified in H&E-stained slides imaged on a Nikon Eclipse Ni or an Olympus IX7I microscope and measured using NIS or CellSens software, respectively. The number of positive cells in tumor tissues was assessed using a blinded visual score of stained tissues. Visual quantification was validated by processing images in ImageJ in batch mode via macro script. Particle size threshold was adjusted respectively, and automated particle count was performed as before (60).

Immunohistochemistry

After in vivo treatments, tumors were harvested and fixed for 48 hours in 10% formalin. Antigen retrieval on mouse-derived tissues or an MB tissue array (Biomax, BC17012c) was performed by steaming samples in citrate buffer for 30 min before staining for GLI2 (Aviva Systems Biology), GFAP, Ki67, SOX2 (Abcam), or cleaved caspase-3 (Cell Signaling Technology) according to the manufacturer’s recommendations. For GLI1 (Invitrogen) IHC staining, samples were boiled in EDTA (pH 8.5) for 10 min instead. SOX2, GLI1, and GLI2 antibodies were validated using respective positive and negative tissues (fig. S1, A to C). Diaminobenzidine (DAB) reagent (DAKO) and a Vina Green chromogen kit (Biocare Medical) were applied for antigen detection. IHC stainings were counterstained with hematoxylin, except for double stainings using DAB and Vina Green reagents to detect nuclear proteins, in which no counterstaining was performed. DAB and Vina Green background staining for double IHC staining was determined after incubation of samples in the presence of the primary and secondary antibodies for one of the antigens and only the secondary for the other (fig. S1D). GLI1 and GLI2 staining in TMA was scored by a board-certified pathologist (K. G. Lindsey, Medical University of South Carolina, SC). For quantification of double-positive cells in TMA, slides were scanned on a Vectra Polaris instrument (PerkinElmer), and the resulting images were processed via Phenochart and analyzed using inForm software (Akoya Biosciences) as before (61). Quantifications were validated in a Nikon Eclipse Ni microscope.

Cell culture

MB tumors from Ptch1-LacZ (MB47) or Ptch1-LacZ; Trp53 KO (MB1, MB2) mice were enzymatically and mechanically digested using Accutase (Invitrogen), and the resulting cell suspensions were maintained ex vivo in MPC media to generate vismodegib-resistant cultures, one Ptch1-LacZ with Ptch1^−/− genotype (MPC-47) and two independent Ptch1-LacZ; Trp53 KO with Ptch1^−/−; Trp53^−/− genotype (MPC-1 and MPC-2) cultures (22), or in Dulbecco’s modified Eagle’s medium (DMEM)/F12 (Gibco), B27, and penicillin-streptomycin to generate matching vismodegib-sensitive SHH-MB cultures (28, 29). Similar cell culture media were used to expand Ptch1-Gfap tumors ex vivo. The isolation and culture of GCPs from P4-6 C57BL/6 mice was performed using a papain dissociation system (Worthington Biochemical Corporation). Cells were plated on poly-l-lysine (Millipore)–coated plates in Neurobasal-A, 1% GlutaMAX, 2% B27, penicillin-streptomycin, and KCl (250 mmol/liter) (27).

NIH-3T3 cells stably expressing a GLI-dependent firefly luciferase reporter gene and a constitutive Renilla luciferase reporter gene (Light2 cells) were maintained in DMEM with 10% fetal calf serum (FBS), G418 (0.2 μg/μl), and Zeocin (0.1 μg/μl). SHH signaling was induced in these cells by treating them with SAG (100 nM) in 0.1% FCS for 24 hours. GLI-driven firefly luciferase activity was normalized to the Renilla luciferase control, and its activity was determined using a dual luciferase kit (Promega) (23, 27). Immortalized WT (44) and Sufu^−/− (45) MEFs were cultured in DMEM and 10% FBS.

Cell and molecular biology

Cell viability was monitored by the reduction of 3-(4,5-dimethyl-2-thiazolyl) 2,5-diphenyl-2H-tetrazolium bromide (MTT) to formazan (27). Total RNA was TRIzol-extracted, and expression of the indicated genes was measured by quantitative real-time polymerase chain reaction (RT-qPCR) using TaqMan probes (Invitrogen) (27). RT-qPCR data were normalized to GAPDH reference gene. Gli1 and Ptch1 were used as SHH target genes. Because the lack of Ptch1 expression, Ptch2 was used as an SHH target gene in Ptch1 mutant MB models as before (22, 23, 27). Radioimmunoprecipitation (RIPA buffer) assay buffer (Thermo Fisher Scientific) was used for protein extraction, and levels of the indicated proteins were determined by immunoblotting according to the manufacturer indications and using antibodies purchased from Cell Signaling Technology [GLI1, SOX2, MYC, HSP90, and glyceraldehyde-3-phosphate dehydrogenase (GAPDH)] or Aviva Systems Biology (GLI2) and listed in fig. S8. GAPDH was used as loading control. Chromatin immunoprecipitation analyses were performed by using the ChIP-IT Express Chromatin Immunoprecipitation Kit from Active Motif. Chromatin fragments were immunoprecipitated with BRD4 antibodies (Bethyl Laboratories) or immunoglobulin G (IgG) control. Relevant regions of the Gli1 locus were amplified by qPCR using previously published probe sets (PS) and listed in fig. S8 (62). PS3 aligns 5′ to the transcription start site and, therefore, was used as a negative control. For flow cytometry analyses, cell suspensions were stained using a LIVE/DEAD viability kit (Invitrogen) and then incubated in Cytofix/Cytoperm solution (Becton Dickinson) before staining using corresponding fluorescent-conjugated antibody: SOX2-APC (BioLegend), GLI1-APC-Cy7, GLI2-FITC, and GFAP-PE (Novus Biologicals). Flow cytometry data were acquired using a LSRFortessa X-20 flow cytometer (Becton Dickinson) and analyzed using FlowJo software. Specificity of GLI1 and GLI2 antibodies for flow cytometry analyses was determined by comparing percentages of cells expressing GLI1 and GLI2 in vismodegib-sensitive SHH-MB cell cultures exposed to vehicle or vismodegib (fig. S2E). Two different sets of siRNA (siGENOME and siOnTarget) consisting of a pool of four sequences each were purchased from Dharmacon and used at 25 nM in 3 × 10⁵ cells. Transfections were performed using Lipofectamine 2000 (Invitrogen). GLI1-expressing vector was a gift from the laboratory of Dr. D. Robbins, and pCS2-MT GLI2 FL (GLI2) was obtained from Addgene (plasmid #17648). Three small guide RNAs (sgRNAs) targeting the 5′ start site and two targeting the 3′ region of the coding domain (fig. S8) were selected for the CRISPR-Cas9–mediated silencing on Myc (fig. S8) using CHOPCHOP (https://chopchop.cbu.uib.no/) and cloned into a lentiCRISPRv2 vector (Addgene, plasmid #104990). A total of 600 ng of each sgRNA (sgMYC) or 3 μg of the same vector containing a scramble sequence (sgSC) and 1 μg of LentiCas9 vector (Addgene, plasmid #52962) were transfected into MPC cultures using Lipofectamine 2000. Four days later, cells were collected for further analyses. For cell proliferation assays, MPC cultures were plated in poly-l-lysine (Millipore)–coated glass chamber slides (Millipore) and incubated in the presence of bromodeoxyuridine (BrdU) (10 mM; Bio-Legend) for 2 hours before fixation with 16% paraformaldehyde and staining using anti-BrdU antibody (Cell Signaling Technology). Similar chamber slides were used to stain MPC cultures for cleaved caspase-3 (Cell Signaling Technology). All drugs used in our ex vivo experiments were purchased from Selleckchem, except BMS-986158, which was purchased from BioVision.

Imaging mass cytometry

All antibodies in the panel were initially tested by immunofluorescence and conjugated to metals using the Maxpar X8 Multimetal Labeling Kit (Fluidigm) according to the manufacturer’s protocol. Data acquisition was performed on a Helios time-of-flight mass cytometer coupled to the Hyperion Imaging System (Fluidigm). Before laser ablation, optical images of slides were acquired using the Hyperion software. Laser ablation was performed at a resolution of approximately 1 mm and a frequency of 200 Hz.

scRNA-seq analyses

Previously generated scRNA-seq data from T.R.G’s laboratory (University of North Carolina, Chapel Hill, NC) (20) were analyzed using R scripting. Cell populations were classified on the basis of an assignment of cell types to Louvain-Jaccard clusters and visualized by a t-distributed stochastic neighbor embedding (t-SNE) projection of the principal components analysis matrix as previously reported (20). Seurat toolkit (63) was used for the visualization of the gene expression levels in cells from vehicle- and vismodegib-treated tumors. Density plots represent the kernel density estimation of a gene expression or genes coexpression, improving the visualization of dropped-out genes (64). For the representation of coexpression with Gli1 or Gli2, the sum of products was applied.

Gene expression profiling

R2 (http://r2.amc.nl) was used to analyze RNA-seq data from enhanced green fluorescent protein (eGFP)–SOX2⁺ and eGFP⁻ MB cells previously generated at the laboratory of Dr. P. Dirks (SickKids, Toronto, Canada) (19). For differential expression analyses, gene expression was transformed to log₂, and SHH biomarkers contained in the Kyoto Encyclopedia of Genes and Genomes 48 genes Hedgehog Signaling Pathway (environmental information processing) elaborated at Kampen’s laboratory (Amsterdam UMC, Amsterdam, The Netherlands) were used to filter genes. Expression was analyzed by performing a two-sided t test, and a false discovery rate test was applied for multiple testing correction.

Statistical analysis

Results from ex vivo experiments represent means ± SEM of at least three independent experiments. Analyses of stainings done in MPC cultures were performed by quantifying numbers of positive cells per field in at least five high-magnification fields per condition, from three independent experiments. For tumor size studies, the results shown represent the mean and SEM of tumor areas found in brain slides from at least four randomly assigned mice per experimental condition, and each dot in the graph represents an individual animal. For tissue IHC quantification, numbers of positive cells per field were quantified. The results shown represent mean and SEM of positive cells located in at least four random high-magnification fields from at least three different mice, where each dot in the graph represents an individual animal. The significance for two-group analyses was determined using an unpaired Student’s t test. Multiple comparison analyses were performed using one-way analysis of variance (ANOVA), followed by a Dunnett post hoc. Kaplan-Meier methods were used to construct symptom-free survival curves using at least nine mice per experimental condition, and significance was evaluated using log-rank (Mantel-Cox) tests. Statistical significance (*) was reached when P value was <0.05.

Supplementary Materials

This PDF file includes:

Figs. S1 to S8

View/request a protocol for this paper from Bio-protocol.