Eradicate Neuroblastoma with 100% Response Rate

Indisulam resistance in a transgenic MYCN/ALK F1178L mouse model is associated with cell state switch to MES and Schwann cell precursor phenotypes

Human ALKF1174L (analogous to mouse ALKF1178L) is the most frequent somatic mutation in neuroblastoma and is associated with MYCN amplification, conferring a worse prognosis than MYCN amplification alone58. Our previous study showed that indisulam treatment of MYCN/ALKF1178L mice with a regular dosing (25 mg/kg, 5 days on/2 days off for two weeks) led to durable, complete responses even when tumor sizes reached over 1000 mm3 prior to treatment initiation30. When testing the dosing schedule by treating the MYCN/ALKF1178L mice with only one dose of indisulam (25 mg/kg) per week for 3 weeks in our current study, we found variable responses (Fig. 1a). Two weeks after therapy discontinuation, we resumed the regular dosing schedule for an additional two weeks. Among the 5 treated mice, three showed complete and durable responses, one responded to the first dose of therapy (tumor volume reduced from 1100 mm3 to 300 mm3) but then developed resistance to the following treatment (Fig. 1a), while another one showed a complete response but eventually relapsed after second therapy discontinuation (Fig. 1a, b). To understand the mechanism of therapy resistance, we performed RNA-seq analysis followed by gene set enrichment analysis (GSEA) to compare the naïve tumors with the relapsed and resistant tumors (Fig. 1c). In comparison with the naïve tumors that showed a high “E2F” gene signature, indicative of the high proliferation rate of naïve tumor cells, the “human 20q11 amplicon” was the top upregulated gene signature in the resistant/relapsed tumors (Fig. 1c, d). Human RBM39 is located in 20q11. This was consistent with the high levels of RNA-seq reads of Rbm39 in the relapsed and resistant tumors (log2 fold change = 0.73), which is located in Chr2qC of mouse genome (Fig. 1e). Surprisingly, the RBM39 protein was actually downregulated (Fig. 1f), which was in contrast to its mRNA expression. These data suggest that cells strived to survive by producing more mRNAs to compensate for the RBM39 protein degradation. This was further verified by the correlation of indisulam resistance with the RBM39 copy number and mRNA expression levels across 534 cell lines in DepMap (depmap.org) (Fig. 1g, h).

Additionally, we noticed that the relapsed/resistant tumor cells acquired a high “neural crest stem cells” gene signature (Fig. 1c,i), suggesting that these therapy-resistant cancer cells underwent de-differentiation. We then examined the ADRN and MES signatures46 and found that the relapsed/resistant cells exhibited a significant shift to the MES state (Fig. 1j). Particularly, we also found that the relapsed/resistant tumor cells expressed a signature of “Schwann cell precursors (SCP)”51,52, characterized by high expression of S100b, Sox10, Plp1 and other genes (Fig. 1k). Then we colored the UMAP projection by the expression of Sox10, the SCP transcription factor, into the single cell analysis of human adrenal medulla, which demonstrated high levels of SOX10 expression in SCPs but not in chromaffin and neuroblast cells (Fig.1l), the downstream progenies of SCP and origin of neuroblastoma. We further examined the SCP signature, which was highly expressed in the resistant/relapsed Th-MYCN/ALKF1178L tumors, in two human neuroblastoma cohorts (Target NBL21 and St Jude PCGP59). The SCP markers such as TLX3, SOX10, S100b, and FOXD3 appeared to be enriched in non-MYCN amplified tumors (Supplementary Fig. 1a, b), suggesting that the resistant/relapsed MYCN-driven tumors acquired gene features similar to the non-MYCN amplified tumors. Taken together, these data indicate that the resistant/relapsed neuroblastomas acquired new cell state properties of neural crest progenies.

Indisulam resistance is associated with a cell state switch from ADRN to MES and melanocytic state in human MYCN-amplified neuroblastomas

To further understand the cell state alterations in neuroblastoma resistance, we treated MYCN-amplified patient-derived xenografts (SJNB14) implanted subcutaneously into CB17/SCID mice with indisulam (25 mg/kg, 5 days on/2 days off for two weeks). In comparison to the vehicle control group, all five tumors treated with indisulam had complete responses, but eventually they relapsed. However, these tumors remained responsive to indisulam treatment until the eighth cycle of therapy when tumors developed full resistance (Fig. 2a). RNA-seq followed by GSEA showed that the resistant tumors had a significant reduction of the “ADRN” signature, followed by a “hypoxia” signature (Fig. 2b, c). However, the resistant tumors acquired a high “interferon alpha and beta” signature (Fig. 2d), in line with the feature of MES neuroblastoma cells, which were reported to express high levels of interferon pathway genes60,61. The master transcriptional factors of MES state cells, such as c-MYC, PRRX1 and NOTCH1 were highly upregulated in the resistant tumor cells, together with the SCP marker S100B and stem cell markers KIT and SALL4 (Fig. 2e). Wnt ligands were also highly upregulated in the resistant tumors. A recent study based on network analysis of neuroblastoma expression data proposed that Wnt signaling is a major determinant of regulatory networks that underlie mesenchymal/neural crest cell-like cell identities through PRRX1 and YAP/TAZ transcription factors62. Surprisingly, melanocytic markers, including MITF, which encodes the master transcriptional factor of melanocytes and melanomas, were also highly upregulated. It is known that SCPs are melanocyte progenitors63. The scRNA-seq analysis in the developmental trajectories of SCP to chromaffin cells and neuroblasts showed no expression of MITF (Supplementary Fig. 2a), supporting the hypothesis that neuroblastoma cells either directly transdifferentiated to a melanocyte-like state or dedifferentiated to SCPs which then differentiated to a state with melanocytic features. We further examined the expression of Wnt ligands and melanocytic gene signatures in human neuroblastoma cohorts. The Wnt ligand signature was enriched in neuroblastomas without MYCN amplification in both the TARGET NBL and St Jude PCGP cohorts21,59 (Fig. 2f; Supplementary 2b). Interestingly, the melanocytic gene signature was enriched in TARGET neuroblastomas without MYCN amplification but not in St Jude PCGP data (Fig. 2f; Supplementary 2b), probably due to the confounding factors such as different stages, disease risk classification, and pre-treatment and post-treatment from both cohorts. It was known that the TARGET NBL dataset was enriched with high-stage and high-risk neuroblastomas21. Distinct from the MYCN/ALKF1178L resistant murine tumors, the RBM39 expression showed no great induction in the resistant SJNB14 tumors (Fig. 2g), and Sanger DNA sequencing showed no mutations in the indisulam binding motif (Supplementary Fig. 2c), suggesting that the cell state switch plays a major role in mediating the indisulam therapy resistance in this MYCN-amplified PDX model.

Similarly, mice implanted with SIMA (with MYCN amplification) cell line–based xenografts also showed excellent outcomes after the completion of a 10-day treatment, but the neuroblastoma eventually relapsed (Supplementary Fig. 3a). To test if the relapsed SIMA tumors remained sensitive to indisulam, we resumed treatment after tumor recurrence. Like SJNB14 PDX, SIMA xenografts had 100% complete response after two additional repeated treatment cycles (Supplementary Fig. 3a). Western blot analysis of xenografts harvested at 24-hours after the final dosing from the third cycle of treatment confirmed that RBM39 expression was nearly completely abrogated in the treatment group (Supplementary Fig. 3b). RNA splicing analysis of tumor xenografts showed that indisulam induced very similar events to those induced by RBM39 knockdown and in vitro indisulam treatment, including genome-wide splicing anomalies in skipped exons30 (Supplementary Fig. 3c). RNA-seq and GSEA analysis revealed that tumor cells displayed upregulated gene signatures of neural crest migration and type I interferons (Supplementary Fig. 3d), although the changes in ADRN and MES signatures were not significant. Nevertheless, manual examination of master transcriptional factors of ADRN and MES revealed that the tumors treated with indisulam expressed higher levels of C-MYC, PRRX1, and FOSL2 but reduced levels of PHOX2A, PHOX2B, and MYCN (Supplementary Fig. 3e), indicating that the cell state underwent a transition from ADRN to MES during the indisulam therapy. Like SJNB14, a significant induction of melanocytic markers and Wnt ligands was observed in the SIMA tumors treated with indisulam (Supplementary Fig. 3d, e).

Changes in neuroblastoma cell states are associated with epigenetic reprogramming45,46,48. To test if this occurred in the indisulam-resistant tumors, we performed Cleavage Under Targets and Tagmentation (CUT&Tag) to map the genome-wide H3K27Ac in SJNB14 naïve versus resistant tumors. H3K27Ac is an epigenetic mark indicating active promoters and enhancers in gene transcription. We identified 8580 differential H3K27Ac peaks (Fig. 2h). The enrichment of H3K27Ac at PHOX2B (a master transcription factor in ADRN) locus was greatly reduced in the resistant tumors, while the H3K27Ac levels were upregulated at the locus of S100B (a marker of SCP and MES) (Fig. 2i). S100B is also a melanoma marker64. We also found that H3K27Ac peaks at the TWIST1 locus were among the top downregulated in resistant tumors (Fig. 2h, i), in line with its reduction in mRNA levels (Fig. 2e). In neuroblastoma, TWIST1 co-occupies enhancers with MYCN and is required for MYCN–dependent proliferation65. SJNB14 is a MYCN-amplified tumor. Motif analysis for the H3K27Ac peaks revealed that transcription factors, including those determining MES state (JUN-AP1, FOSL2) could bind at the H3K27Ac loci in both naïve and resistant tumors (Supplementary Fig. 4a). Then, we performed GSEA analysis for the genes with altered H3K27Ac peaks. Consistent with the RNA-seq results, the indisulam-resistant tumor cells showed a significant enrichment of H3K27Ac at gene loci of MES transcriptional factors (i.e., SMAD3, TEAD4, MYC, Fig. 2j; Supplementary Fig. 4d), mesenchymal genes such as those involved in muscle contraction, HIPPO (YAP/TAZ) signaling pathway, and interferon alpha and beta signaling (Supplementary Fig. 4b). However, there was a significant downregulation of H3K27Ac peaks at the genomic loci of ADRN genes, and those involved in G2/M phase of cell cycle (Fig. 2k, Supplementary Fig. 4c). These data support that the epigenetic and transcriptional landscapes in the naïve tumors have been reprogrammed when tumor cells developed therapy resistance.

We investigated if both Th-MYCN/ALKF1178L and SJNB14 models share common gene signatures, including the SCP gene signature, after tumors developed resistance to indisulam. However, the upregulated gene signatures in both models were barely overlapping. This could be due to multiple reasons. First, neuroblastoma is a very heterogeneous disease and could occur at different developmental stages of neural crest cell lineage. Neural crest cells are a transient stem cell population that develops into different cell lineages under different development cues. Second, the species specificity (mouse vs. human) could also make a difference. Third, the tumor drivers are different (MYCN/ALKF1178L in murine GEMM vs. the PDX model with MYCN amplification derived from a relapsed patient that received intensive chemotherapy). Nevertheless, both models exhibited gene signatures that can be projected to some stages of neural crest lineage during development, as characterized by Schwann cell precursor and melanocytic markers. These data support the lineage plasticity of neuroblastoma cells in therapy resistance, and this may be associated with differentiation and de-differentiation of neural crest lineages. One recent study characterized chemotherapy-resistant high-risk neuroblastoma persister cells and found that these persisters were not a uniform population66. Rather, these cells were composed of 4 different groups and exhibited distinct gene signatures. Interestingly, this study identified some persister tumors that had elevated Schwann cell signatures66. Taken together, these data provided additional evidence that tumor cells not only switched their state from ADRN to MES, but also acquired additional molecular features such as those expressed in melanomas once they developed resistance, further emphasizing the cellular plasticity of neuroblastoma cells, which is reminiscent of cellular pliancy of neural crest or SCP cells.

Indisulam resistance is associated with RBM39 upregulation and cell state switch from MES to ADRN in a C-MYC–overexpressing xenograft model

Both Th-MYCN/ALKF1178L and the MYCN-amplified SJNB14 PDX models demonstrated the cell state switch from ADRN to MES when they developed therapy resistance, although a more complex and multi-directional transdifferentiation underlies the de facto mechanism (Figs. 1, 2). To some degree, our data are in line with the hypothesis that the conversion of ADRN to MES may account for the chemoresistance of neuroblastoma cells. However, the conversion of MES to ADRN of neuroblastoma cells in therapy resistance is less clear. To explore whether the neuroblastoma cells in an MES state could convert to an ADRN state under therapy selection, we treated SK-N-AS xenografts (C-MYC overexpression by a translocated super-enhancer13, known as MES dominant45,46) with indisulam. Again, SK-N-AS xenografts also showed 100% tumor regression after a 10-day dosing with 25 mg/kg of indisulam (Supplementary Fig. 5a) and remained sensitive to indisulam until the 4th repeated treatment cycle. RNA-seq and GSEA analysis revealed that indeed the resistant tumor cells acquired ADRN and neuronal gene signatures with a significant reduction of MES and interferon signatures (Fig. 3a–e), supporting the hypothesis of interconversion of ADRN and MES states of neuroblastoma cells. Like the MYCN/ALKF1178L resistant tumors, RBM39 mRNA was increased by about 4-fold in the resistant SK-N-AS xenografts (Fig. 3f), but not its paralog, RBM23 (Fig. 3f), which has previously been shown to have no effect on RNA splicing31. These data further verified that RBM39 was the specific target of indisulam that was responsible for the therapeutic efficacy, leading to a selective pressure on RBM39 in resistant tumors. At the same time, cells underwent cell state alterations from MES to ADRN.

Epigenetic reprogramming of indisulam–resistant c-MYC–overexpressing neuroblastoma cells

We hypothesized that the resistant SK-N-AS neuroblastoma cells had undergone an epigenetic reprogramming to alter their MES cell state and thus enhancing the transcription of RBM39. To test this hypothesis, we derived cell lines from three independent SK-N-AS indisulam-resistant tumors. These cell lines indeed expressed high levels of RBM39 and were resistant to indisulam treatment in vitro (Supplementary Fig. 5b, c). To test if the resistant cells were still dependent on RBM39, we introduced exogenous DCAF15 into the indisulam-resistant SK-N-AS cells, which led to a remarkable degradation of RBM39 and cell death in comparison with the parental cells (Supplementary Fig. 5d–g). These data indicate that the indisulam-resistant cells are still dependent on RBM39. To understand the mechanism of cell state switching, we performed Assay for Transposase-Accessible Chromatin with high-throughput sequencing (ATAC-seq) for mapping genome-wide chromatin accessibility of resistant versus naïve cells. Many ADRN genes such as PHOX2A and MYCN as well as the RBM39 locus showed increased chromatin accessibility in the resistant cells (Fig. 3g), in line with the GSEA results from RNA-seq analysis. Motif analysis for the predicted transcriptional factor binding at the genes with altered chromatin accessibility demonstrated that the MES transcriptional factors, such as FOSL1, FOSL2, JUNB, JUND, were enriched in naïve SK-N-AS cells, but not in the resistant cells (Fig. 3h, i). However, the ADRN transcriptional factors such as ASCL1 and NF1A were enriched in the resistant cells. Interestingly, we found that the activity of SP transcriptional factor family members SP2 and SP3 was also enriched in the resistant SK-N-AS cells, together with MYCN (Fig. 3h, i). While it is unclear what the role of SP transcription factors is in neuroblastoma identity, one early study indicates that they are involved in driving expression of MYCN in neuroblastoma cells67. To further corroborate our ATAC-seq study, we carried out CUT&Tag to assess the global alterations of H3K27Ac in naïve and resistant cells (Fig. 3j), an epigenetic mark indicating active promoters and enhancers in gene transcription. GSEA of the differential peaks near the annotated genes revealed that the 20q locus (where RBM39 resides) ranked at the top in the resistant cells, while the c-MYC targets ranked at the top in the naive SK-N-AS cells (Fig. 3k). Again, motif analysis of the altered H3K27Ac peaks showed that the binding of MES transcriptional factors, such as in the FOS and JUN families, was specifically reduced in the resistant cells (Fig. 3l), similar to the results from ATAC-seq. The RBM39 locus showed a greater increase in H3K27Ac than the RBM23 in the resistant cells, in line with the elevated ATAC-seq peak (Fig. 3m), which explains why the resistant cells expressed higher levels of RBM39. These data indicate that the resistant SK-N-AS cells acquired a new cellular state under the selection of repeated indisulam treatment through epigenetic reprogramming.

Cell state alterations are coupled with a switch of dependency on epigenetics and cell state–specific transcription factors

We surmised that the distinct epigenetic landscapes of naïve and resistant tumor cells may result in dependency on specific epigenetic modifiers and cell state–specific transcription factors in these cells, as cell differentiation and de-differentiation is regulated by a cascade of cell state–specific transcriptional factors (TFs) in cooperation with epigenetic modifiers. To test this hypothesis, we performed CRISPR-Cas9 screening (human epigenetic library with 8 gRNAs/gene and transcription factor library with 4 gRNAs/gene) to knock out these genes for a dropout screen with MAGeCK analysis in naïve versus resistant cells (Fig. 4a, Supplementary data 1, 2, 4, 5), in which reduction of gRNA reads (dropout) indicates that cells are dependent on it for survival. In the transcription factor library screening, we identified more pre-mRNA splicing factors in the resistant cells, particularly when they were cultured under indisulam selection (Fig. 4b), which is consistent with the function of RBM39 being critical to splicing in neuroblastoma30, which also verified the screening robustness. Then we specifically examined the ADRN and MES transcriptional factors (Fig. 4c). While a general transcriptional regulator CDK9 was essential to both naïve and resistant cells (Fig. 4d), c-MYC and FOSL2 of MES TFs in the naïve cells were ranked above the ADRN TF HAND2, TBX2 and PHOX2B (Fig. 4c, left). However, in the resistant cells with or without indisulam selection, c-MYC and FOSL2 dropped down while HAND2, PHOX2B, and TBX2 became more essential to the survival of the resistant cells (Fig. 4c, middle and right, 4e, 4f). These data demonstrate that naïve cells and indisulam-resistant cells have specific dependencies on transcriptional factors that determine the cell state of neuroblastoma.

Then, we compared the epigenetic dependency of naïve and resistant cells, in which the epigenetic modifiers formed unique protein-protein interaction network modules (Fig. 4g), consistent with the fact that epigenetic modifiers usually form protein complexes to function. We identified chromatin remodeling INO80 complex was more critical to the survival of naïve SK-N-AS cells, while the SWI/SNF complex components (PBRM1, SMARCE1) and SIN3A complex components (PHF12, ARID4B) were more important in the resistant cells (Fig. 4g–i). Notably, we observed that INO80C was one of the top genes with active H3K27Ac in naïve SK-N-AS cells (Fig. 3j), providing additional evidence that the INO80 complex might be specifically important to the naïve SK-N-AS cells dominated by an MES state, while the SWI/SNF complex could be more essential to resistant SK-N-AS cells that acquired an ADRN state.

Selective and shared dependency of naïve and resistant neuroblastoma cells on targetable kinases

The dependency switch of neuroblastoma cells on epigenetic modifiers and cell state–specific TFs suggests that additional dependencies may correspondingly change once neuroblastoma cells convert to another state. As most epigenetic regulators, such as ARID4B and transcription factors, are difficult to target, we tested if we could identify targetable vulnerabilities of indisulam-resistant neuroblastoma tumor cells by focusing on kinases. Again, with a similar approach, we screened naïve SK-N-AS and indisulam–resistant SK-N-AS cells cultured with or without 250 nM indisulam, using the human kinase CRISPR library (Fig. 5a; Supplementary data 3, 6 gRNAs/gene). We identified 43, 31, and 25 kinases that were essential for either naïve, or indisulam-resistant cells under selective pressure or no selective pressure, respectively; and 13 of them were commonly shared (Fig. 5b; Supplementary data 6). We found that CLK3, DYRK1A and DYRK1B, members of the dual-specificity serine/threonine and tyrosine kinase that play an important role in pre-mRNA splicing by phosphorylating serine- and arginine-rich splicing factors in the spliceosomal complex, are essential to naïve SK-N-AS cells but not for the indisulam-resistant cells (Supplementary data 6). This is in line with the mechanism of indisulam that targets splicing. However, the changes in expression levels of these kinases in resistant vs naïve cells were not correlated with the dependency switch (Supplementary data 7).We then specifically examined the targetable kinases with inhibitors available in clinical trials and found 9 kinases are essential to SK-N-AS cells regardless of indisulam resistance, and 2 kinases (FGFR4 and CDK2) are specifically essential to indisulam-resistant SK-N-AS cells (Table 1). These kinases play critical roles in the G2/M phase cell cycle (i.e., AURKA, AURKB, PLK1), DNA repair (ATR, CHEK1) and gene transcription (CDK7, CDK9). We hypothesize that the combination of indisulam with any of these kinase inhibitors may enhance the efficacy and blunt disease relapse. To test this, we chose gartisertib, a selective ATR inhibitor currently in clinical trials68, in combination with indisulam at 10 mg/kg, for a two-week treatment. Indeed, gartisertib in combination with indisulam significantly extended mouse survival (Fig. 5c) and showed very limited toxicity based on the results of body weight gain over time (Fig. 5d), suggesting that this combination is safe and effective. Recent studies have shown that ATR inhibition is able to reverse the chemoresistance of ALT (alternative lengthening of telomere) neuroblastoma due to telomere dysfunction–induced ATM activation69, and significantly enhances the efficacy of ALK inhibition in transgenic neuroblastoma models driven by MYCN and ALK70. These promising pre-clinical data, together with our preliminary results, provide a rationale to continue testing the combination of indisulam with ATR inhibitors in more mouse models in the future.

Cell state switch induced by indisulam treatment can lead to upregulation of GD2 expression through epigenetic reprogramming

GD2 is a disialoganglioside that is biosynthesized from the precursor gangliosides GD3/GM3 by β−1,4-N-acetylgalactosaminyltransferase (B4GALNT1, GD2 synthase) (Fig. 6a). The expression of GD2 in normal tissues is restricted to the brain, peripheral pain fibers, and skin melanocytes, but it is abundantly expressed in neuroectodermal tumors, including neuroblastoma71. The application of anti-GD2 immunotherapy has greatly improved the survival of high-risk neuroblastoma patients when it is combined with a differentiating agent and chemotherapy72,73,74. Recent advances in the development of GD2–based CAR T and CAR NKT therapies have provided additional evidence showing the potential of anti-GD2 therapy for patients with relapsed disease75,76. While many reasons could account for the response failure of anti-GD2 immunotherapy for a fraction of patients, one of the hypotheses is that cell state switching may lead to low abundance of GD2 expression due to downregulation of GD2 synthase genes. A recent study has shown that cell lines derived from TH-MYCN transgenic tumors lost GD2 expression, accompanied by a switch of cell state from ADRN to MES77. Another study further supported that neuroblastoma state transition to MES confers resistance to anti-GD2 antibody via reduced expression of ST8SIA178. To understand the impact of cell state alterations induced by repeated indisulam treatments on GD2 expression, we examined the key enzymes of GD2 synthesis (B4GALNT1 and ST8SIA1) in our indisulam–resistant models. Our RNA-seq results showed that, in the transgenic TH-MYCN/ALKF1178L model, one out of two resistant tumors showed a higher expression of B4GALNT1 and ST8SIA1 (Fig. 6b), while the SJNB14 PDX model showed no remarkable changes (Fig. 6c). In the SIMA model, ST8SIA1 was significantly upregulated in the tumors with three rounds of indisulam treatment while the expression of B4GALNT1 showed no changes (Fig. 6d). In the SK-N-AS model, the expression of both B4GALNT1 and ST8SIA1 was significantly upregulated in the resistant tumors (Fig. 6e), which is in line with the elevated enrichment of H3K27Ac and enhanced chromatin accessibility at the promoter regions of B4GALNT1 and ST8SIA1 (Fig. 6f, g), indicating that epigenetic reprogramming of SK-N-AS cells leads to upregulation of GD2 synthesis genes. Then, we applied flow cytometry analysis to profile the GD2 expression of the three resistant cell lines derived from SK-N-AS tumors. The result showed that, in comparison with the naïve control, all three resistant clones expressed higher levels of GD2 (Fig. 6h). These data indicate that cell state alterations induced by indisulam do not lead to a reduction of GD2 expression; rather, it result in enhanced expression, at least in some neuroblastomas. These data provide a rationale for a combination therapy of indisulam in combination with anti-GD2 immunotherapy.

Complete eradication of C-MYC-driven neuroblastoma in NK cell competent mice