In Vivo Modeling of Chemoresistant Neuroblastoma Provides New Insights into Chemorefractory Disease and Metastasis

Neuroblastoma is an aggressive tumor of neural crest origin. At time of diagnosis, approximately half of patients have high-risk disease defined by the presence of metastatic disease, amplification of the MYCN oncogene or other clinical risk criteria. Conventional multimodal treatment is intensive and is characterized by frequent development of chemoresistant, metastatic disease (1, 2), reinforcing the need to develop mechanistically targeted treatments for chemorefractory patients.

In neuroblastoma, chemoresistance relating to increased drug efflux via MDR1 and MRP transport for MRP/MDR drug substrates, and temozolomide resistance related to upregulation of MGMT activity is well characterized (3, 4). Evidence for intratumoral clonal selection as a consequence of conventional treatment is emerging from sequencing of paired diagnostic:relapse tissue biopsies, where evolution of new mutations not present at diagnosis, and clonal enrichment of alterations in a limited number of genes such as HRAS, BRAF, and ALK is described (5–7). The recent identification of two transcriptionally defined cell identity states in neuroblastoma, a “noradrenergic” state and a “mesenchymal” state, associated with increased cell motility and acquisition of chemoresistance following therapy, provides a possible framework to explain the origin of metastatic and chemorefractory disease in patients with neuroblastoma (8, 9). However, there are only few robust preclinical models that allow the emergence of chemorefractory metastatic disease to be monitored and mechanistically probed in detail.

To study disease evolution and progression in a context more reflective of clinical therapy, we used the well-characterized Th-MYCN genetically engineered mouse (GEM), which replicates many clinical and genomic features of high-risk, MYCN-driven neuroblastoma, with the exception of clinically evident metastases or treatment resistance typical of human patients (10, 11). We asked whether any clinically relevant changes would occur in tumor behavior, microenvironmental structure, or genomic integrity in these mice, under conditions of maintained genomic stress characteristic of patients undergoing repeated cycles of chemotherapy. We administered multiple cycles of cyclophosphamide, a first-line chemotherapy drug and mainstay of neuroblastoma induction therapy, to Th-MYCN mice using MRI to guide dose administration and to limit tumor progression. Under these conditions, the majority of mice became refractory to CPM treatment, and developed native bone marrow metastases characteristic of human disease. Primary tumors developed cell-intrinsic cross-resistance to additional chemotherapeutics including vincristine and doxorubicin, and exhibited structural and microenvironmental changes, including increased collagen deposition and recruitment of tumor-associated fibroblasts. At a molecular level, we identified enhanced expression of genes associated with high-risk neuroblastoma, and changes in expression of key signaling pathways such as JAK-STAT3, activation of which correlates with increased motility, invasion and cell proliferation, and advanced neuroblastoma disease progression (12, 13). Using the clinical JAK1/JAK2 oral inhibitor CYT387, currently in phase III clinical trials in adult cancer (NCT02101021), we show that cell lines derived from cyclophosphamide-resistant tumors are sensitive to JAK1/2 inhibition and that in vivo treatment with CYT387 and cyclophosphamide significantly improved survival of Th-MYCN mice bearing cyclophosphamide-resistant tumors. Taken together, these data describe a preclinical model system and approach that is useful to model the microenvironmental and molecular alterations associated with the development of invasive, treatment-resistant neuroblastoma, and highlight an approach to identify potentially targetable alterations in signaling pathways such as JAK-STAT3 that may be useful to combat chemoresistance in neuroblastoma.

All experimental protocols were approved and monitored by The Institute of Cancer Research Animal Welfare and Ethical Review Body (PPL 70/7945, later PPL P91E52C32), in compliance with the UK Home Office Animals (Scientific Procedures) Act 1986, the United Kingdom National Cancer Research Institute guidelines for the welfare of animals in cancer research (14), and the ARRIVE guidelines (15). Th-MYCN mice (129 × 1/SvJ-Tg(Th-MYCN)41Waw/Nci) have been described previously (11). In this study, we used homozygous Th-MYCN mice in which we observe 100% tumor penetrance with mean onset of tumors at 35 ± 7.5 days, consistent with other published studies (11, 16). Two to five mice were caged together and were allowed access to sterilized food and water ad libitum.

A personal dose escalation protocol was developed aiming to achieve tumors resistant to 32 mg/kg cyclophosphamide. We enrolled animals in three steps to (i) identify the dose schedule (n = 15), (ii) test the individualized protocol (n = 9), and (iii) incorporate imaging into the resistance protocol (n = 9), alongside 6 untreated animals serving as controls. Th-MYCN tumor–bearing mice were enrolled on trial when a trained technician detected a palpable tumor size of approximately 5 mm. Dosing was initiated with a single dose of 16 mg/kg. Tumor response was detected with palpation. Upon regrowth (following regression in size), the tumor was dosed again with 16 mg/kg. This treatment was repeated until tumors no longer responded to the dose, at which time, the dose was increased by 50%. Studies were terminated when the tumor became resistant to 32 mg/kg cyclophosphamide and continued to grow to a palpable size of 15 mm or immediately upon showing any signs of ill health. Tumor size, animal weight, and overall animal well-being were scored daily throughout the study. No outliers were excluded. Details of in vivo imaging can be found in the Supplementary Data.

Wild-type littermates from the Th-MYCN colony were used for allograft studies. Briefly, allograft studies were repeated twice with n = 8 mice each experiment (n = 4 controls and n = 4 treated). Mice were injected subcutaneously unilaterally or bilaterally with 2 × 10⁶ primary uncultured cells. Calipers were used to measure tumor diameter on two orthogonal axes 2–3 times per week. Volume was calculated using the equation: v = 4/3πr³ (where r = radius, calculated as an average of the two axes). Dosing occurred at predetermined tumor size of 5–10 mm. Studies were terminated when the mean diameter of the tumor reached 15 mm.

The following compounds were used for in vitro and in vivo experiments: cyclophosphamide (PHR1404 Sigma-Aldrich), momelotinib (CYT387; S2219, Selleckchem), vincristine (S1241, Selleckchem), doxorubicin (S1208, Selleckchem), 4-hydroperoxy cyclophosphamide (39800-16-30, Caymanchem).

The neuroblastoma cell lines KELLY (DSMZ, ACC355), SK-N-AS (ATCC, CRL-2137), BE(2)C (Public Health England, 95011817), and IMR-5 (provided by professor Martin Eilers, University of Würzburg, Würzburg, Germany) were in house routinely tested to be Mycoplasma-free using LookOut Mycoplasma PCR Detection Kit (Sigma-Aldrich, MP0035). The identity of each line was authenticated before the study by short tandem repeat (STR) profiling (ECACC, Public Health England).

Murine STR profiling was performed by Microsynth AG, Switzerland, using highly polymorphic STR loci (17). Human cell lines were cultured in RPMI with 10% FBS and murine cell lines were cultured in DMEM/F12 medium (with B27, 40 μmol/L EGF, 40 μmol/L bFGF, 10% FCS) in a 37°C, 5% CO₂ tissue culture incubator.

To identify mechanisms associated with the development of chemoresistance in neuroblastoma, we developed an individualized and MRI-guided dose escalation schedule for cyclophosphamide treatment using the homozygous Th-MYCN model, in which 100% of homozygotes develop tumors with a mean onset of 35 ± 7.5 days. To establish the dose range and degree of sensitivity of spontaneously arising Th-MYCN tumors to cyclophosphamide, we monitored dose response via anatomic MRI and palpation. We tested initial single doses of 16 versus 32 mg/kg cyclophosphamide. We found that following the 16 mg/kg dose tumors exhibited an initial partial response (PR, 40% volumetric response, RECIST), at day 2 and progressive disease (PD) at day 7, while treatment with 32 mg/kg led to a very good partial response, (VGPR, 90% volumetric response, RECIST) for up to 7 days (Fig. 1A). Treatment with 16 mg/kg cyclophosphamide induced cell death, evident as increased pyknotic nuclei on histologic staining (Supplementary Fig. S1A).

To maximize the potential for development of resistance, we dose-adjusted and individualized treatment to maintain stable disease (SD) using cyclophosphamide doses of 16–32 mg/kg for up to 120 days, (Fig. 1B–D), rendering all initially chemosensitive tumors resistant to 32 mg/kg cyclophosphamide at study end (last dose, Fig. 1E). After cessation of therapy, chemoresistant tumors were rapidly progressive (4 days vs. 8 days to regrowth of initial tumor volume; Fig. 1F), with a high Ki-67 staining index (Supplementary Fig. S1B). Resistance to 32 mg/kg cyclophosphamide was induced within 5–18 dose cycles, over 34–120 days (Th-MYCN^CPM32; Fig. 1G; Table 1). Cyclophosphamide is activated from an inactive prodrug to the cytotoxic metabolite (phosphoramide mustard) via one or more forms of aldehyde dehydrogenases (ALDH; refs. 18, 19). To rule out altered intratumoral cyclophosphamide metabolism or distribution as a potential explanation for these findings, we measured intratumoral cyclophosphamide and levels of the key metabolites 2-dechloroethyl cyclophosphamide, carboxyphosphamide, and 4-ketocyclophosphamide at 1 hour posttreatment in treatment-naïve and resistant tumors and found no significant differences in these measurements (Supplementary Fig. S1C and S1D). Taken together, these data indicate that multicycle exposure to cyclophosphamide treatment induces chemoresistance in this model, via mechanisms not related to altered drug penetration or modulation of canonical detoxification pathways.

Stromal content and microenvironmental structure are biological factors that correlate with clinical aggressiveness of human neuroblastoma. At diagnosis, low stromal content, a high mitosis-karyorrhexis index (MKI), and undifferentiated tumor morphology correlate with poor prognosis and aggressive disease. Epithelial-to-mesenchymal transition (EMT) is thought to accompany acquisition of a migratory/metastatic phenotype and chemoresistance (20–23). In chemoresistant mice, we observed alterations in stromal structure and extracellular matrix (ECM) content, with collagen enrichment and a 2.5-fold increase in total aggregated collagen bundles as detected by picrosirius red staining (Fig. 2A and B). IHC analysis identified increased expression of collagen IV (Fig. 2C), and to a lesser extent collagen I and fibronectin, in cyclophosphamide-resistant tumors (Supplementary Fig. S2A). Furthermore, immunofluorescence analysis indicated an increase in cell-associated smooth muscle actin staining, a marker consistent with presence of cancer-associated fibroblasts (Fig. 2D). Concomitant with these ECM changes, we observed coexpression of vimentin with the neural marker Tuj1 in a subpopulation of TH-MYCN^CPM32 tumor cells (bottom panel arrows Fig. 2E), whereas in TH-MYCN mice, the mesenchymal marker vimentin was observed only in endothelial cells (top arrows). These results indicate that together with the development of chemoresistance following treatment, tumors exhibit evidence of microenvironmental changes and enrichment in a subpopulation of tumor cells with mesenchymal characteristics, changes associated with EMT.

Given that EMT is one mechanism by which highly motile and invasive cells are generated (24, 25), we examined bone marrow in the Th-MYCN^CPM32 model for the presence of metastases. Tumor cells in bone marrow were identified by FACS as live, single cells, CD45 negative, CD11b negative, and GD2 (neuroblastoma antigen) positive (Fig. 3A). We found that 6 of 9 (66%) mice with resistant tumors had detectable tumor cells in bone marrow compared with 2 of 6 (33%) from mice with treatment-naïve tumors (Fig. 3B). Next, we used IHC with the neural marker neurofilament-light (NF-L) to detect metastatic cells present in tibial sections. We identified multifocal bone marrow metastases defined by positive NF-L staining in 9 of 13 (69%) Th-MYCN^CPM32 mice and 0 of 6 untreated controls (Table 1; Fig. 3C). Furthermore, pathologic examination of the metastatic foci indicated a morphologic resemblance between human and murine neuroblastoma bone marrow (stained with the clinical human neuroblastoma marker CD56 and the equivalent murine marker NF-L, respectively; Fig. 3C). One intriguing interpretation of this data is that treatment and acquisition of chemoresistance in these tumors is associated with EMT in a subset of cells that migrate to bone marrow and generate disseminated metastases. This would be an obvious focus for further work in the model.

To identify candidate genes and pathways that are altered in resistant tumors, and that could represent potential therapeutic targets, we performed bulk gene expression analysis of treatment-naïve and resistant tumors (Supplementary Tables S1–S4). We found that expression of only a small subset of genes (n = 22) was significantly altered (P < 0.05) in resistant as compared with treatment-naïve tumors (Fig. 4A). Among the upregulated genes, we identified the cell-cycle regulator cyclin B1–interacting protein 1 (Ccnb1ip1) and thymidine kinase (TK1), a biomarker for aggressiveness in many types of cancer (26, 27). We confirmed these findings in an independent sample set using quantitative real-time PCR (Fig. 4B and C). High expression levels of both genes are significantly associated with poor survival in human neuroblastoma (Fig. 4D and E; https://r2.amc.nl). However, the upregulation of TK1 was not observed at the protein level (Supplementary Fig. S2B) and quantitation of CCNB1IP1 protein was not possible due to lack of a reliable antibody. To identify pathways associated with resistance in the Th-MYCN^CPM32 model, we performed gene-set enrichment analysis (GSEA), identifying upregulated expression of genes associated with gain of human chromosome 17q (Fig. 4F), which, together with loss of chromosome 1p36, defines an extremely poor prognosis group in neuroblastoma with a 15.6% 5-year relapse-free survival (28).

We performed copy number (CN) analysis on whole-exome sequencing, comparing tissue obtained from sensitive tumors to resistant tumors with matched bone marrow metastases (Fig. 4G; Supplementary Fig. S3A). In 6 of 7 (85%) of resistant tumors and in 4 of 7 (57%) matched bone marrow metastases, we found gains on mouse chromosome 11 in a region spanning 100 Mbp to 120 Mbp. This locus is syntenic with human 17q gain (Supplementary Fig. S3B; Supplementary Table S5). No corresponding changes were evident in treatment-naïve tumors. Although gains on chromosome 11 have been reported in 40% of hemizygous Th-MYCN tumors with long latency (11), they have not previously been identified in homozygous Th-MYCN tumors. Three genes on the gained region (TBX4, TK1, and Tnrc6c) were within a set of 22 genes with upregulated RNA expression in the Th-MYCN^CPM32 tumors (Fig. 4A). Levels of TNRC6 protein were increased in resistant tumors (Supplementary Fig. S2B). We were unable to quantitate TBX4 protein due to lack of a reliable antibody. We next evaluated tumor samples for somatic variants in genes frequently mutated in relapsed neuroblastoma: Alk, Atrx, Arid1a, Arid1b, PHOX2B, Ptpn11, Kras, Nras, Hras1 (29). On the basis of our sequencing coverage, in which approximately 97.4% of exonic positions have at least five reads and approximately 70.4% of exonic positions have at least 50 reads (Supplementary Table S6), no mutations in these genes were identified. These results indicate that following prolonged treatment with cyclophosphamide, tumors in the Th-MYCN murine model exhibit one of the commonest segmental chromosomal abnormalities associated with high-risk neuroblastoma. We did not observe an accumulation of SNVs that are described as enriched in human patients at time of relapse, at least within the treatment conditions used here.

One potential use of this model is to highlight altered expression of pathways that may contribute to development of chemoresistance and metastasis. Consistent with other assays suggestive of EMT in these tumors, GSEA detected upregulation of genes within the mesenchymal core-regulatory state and relative underexpression of genes associated with the adrenergic state (Supplementary Fig. S3C–S3E; ref. 9).

To further explore this, we examined protein levels of the canonical adrenergic marker PHOX2B and the mesenchymal marker vimentin (9) using IHC in tumor sections derived from the Th-MYCN and the Th-MYCN^CPM32 mice (Supplementary Fig. S4A). Consistent with the detection of vimentin-positive neuroblastoma cells in the resistant tumors shown in Fig. 2E, we found a greater proportion of cells that were PHOX2B negative and an increase in vimentin-positive cells in Th-MYCN^CPM32 compared with Th-MYCN tumors (Supplementary Fig. S4B). We next evaluated the mesenchymal status of cell lines derived from Th-MYCN^CPM32 tumors. We established a panel of cell lines from a Th-MYCN^CPM32 tumor derived either by directly culturing the primary tumor in vitro (68557), or by passaging it in vivo as subcutaneous tumor (83984) or as an intratibial tumor (96459) prior to subsequent in vitro culture. From the intratibial tumor, two cell lines were generated; 96459A and 96459B (Supplementary Fig. S4C and S4D). Via immunoblot, all Th-MYCN^CPM32 cell lines showed a reduction in levels of the adrenergic marker PHOX2B and an increase in the levels of the mesenchymal markers vimentin and FMO3, compared with untreated Th-MYCN tumors and the human neuroblastoma cell lines IMR5 and KELLY (Supplementary Fig. S4E). Gene expression in these cell lines mapped very closely to established core mesenchymal, but not adrenergic, gene signatures (Supplementary Fig. S4F; refs. 8, 9). Of interest, we found reduced or negligible levels of MYCN protein (68557, 83984, and 96459B) and a reduction in MYCN mRNA levels (83984 and 96459B; Supplementary Fig. S4D and S4G). Taken together, these results suggest that a transition to a mesenchymal state occurs in Th-MYCN^CPM32 tumors following treatment.

To identify aberrantly regulated and potentially druggable events, we further used GSEA to investigate alterations in gene expression within these samples. We identified 15 significantly enriched Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways (Supplementary Table S7) and focused on the antiapoptotic and prometastatic IL6-JAK-STAT3 signaling pathway (Fig. 4H–J). We therefore compared levels of total STAT3 and activated p-STAT3 (Y705) in sensitive and resistant tumors. We found a significant enhancement of total STAT3 nuclear staining in resistant tumors compared with treatment naïve tumors (Fig. 4K and Supplementary Fig. S5A). The majority (6 of 8) of primary tumors from mice with detectable bone marrow metastases exhibited p-STAT3 staining (Table 1; Fig. 4L), as is observed in human neuroblastoma (13). Given that JAK-STAT3 pathway activation is present in these mice and is a feature associated with high-risk neuroblastoma (30), we investigated whether JAK–STAT3 pathway inhibition would alter disease progression in Th-MYCN^CPM32 mice with treatment-resistant tumors.

We used the clinical candidate JAK1/JAK2 ATP–competitive inhibitor CYT387 to assess efficacy of JAK–STAT3 pathway inhibition. While several other JAK inhibitors are available, CYT387 is currently being used in a phase III randomized clinical trial for metastatic pancreatic ductal adenocarcinoma (NCT02101021). Its efficacy in neuroblastoma has not yet been tested. We tested sensitivity to CYT387 in our Th-MYCN^CPM32 tumor-derived cell line panel (68557, 83984, and 96459B; Supplementary Fig. S4C), in addition to the well-characterized human neuroblastoma cell lines, BE2C, and SK-N-AS (31, 32), derived from heavily pretreated patients, IMR5 (33, 34), derived from a patient at time of diagnosis, and KELLY cells (treatment status unknown). To test the efficacy of CYT387, we assayed levels of p-STAT3, an indicator of JAK1/2 inhibition and of c-Fos, a known transcriptional target of STAT3 (35). Cytotoxicity was evident at 0.37 μmol/L (Supplementary Fig. S5B) and dose-dependent inhibition of STAT3 phosphorylation on immunoblots was consistent with on-target activity (range 1 μmol/L–3 μmol/L; Fig. 5A and B). CYT387 led to a reduction in the mRNA and protein levels of c-Fos in cyclophosphamide-resistant cell lines (Fig. 5C). Although CYT387 treatment did not lead to robust cell death, it profoundly decreased cell density (Supplementary Fig. S5C). BRDU incorporation and cell-cycle analysis indicated a significant (>50%) reduction in cell proliferation at 3 μmol/L CYT387 in Th-MYCN^CPM32cell lines but not in KELLY cells (Fig. 5D). Following pretreatment with CYT387 in IMR5 and 96459B, both showed a reduction in colony formation with or without continued CYT387 treatment (Fig. 5E). To confirm that the effect of CYT387 is related to the inhibition of STAT3, we genetically manipulated STAT3 levels using an siRNA and a STAT3 overexpression vector. Abrogation of STAT3 phenocopied the effect of CYT387 treatment, causing a 25% reduction in cell proliferation as compared with transfection with a nontargeted control siRNA (Supplementary Fig. S6A and S6B). Reexpression of STAT3 partially rescued cells following CYT387 treatment, and increased cell viability from 49% ± 1.575% in control cells to 61% ± 2.24% in STAT3-overexpressing cells (Supplementary Fig. S6C and S6D).

To test whether CYT387 is effective in the setting of chemoresistance and upregulation of STAT3 signaling, we used Th-MYCN^CPM32 cell lines, which are relatively resistant to 4-hydroperoxy cyclophosphamide (4-OOH-CY), vincristine, and doxorubicin, as compared with the human MYCN-amplified neuroblastoma-sensitive cell line IMR5, (Supplementary Fig. S6E–S6G). In all three Th-MYCN^CPM32 cell lines, addition of CYT387 partially restored sensitivity to 4-OOH-CY (Fig. 5F and G) and vincristine (Fig. 5H and I) as measured by AUC reduction. These results suggest that CYT387 may have therapeutic potential in chemoresistant neuroblastoma.

In order to more precisely establish the in vivo efficacy of CYT387 in a treatment resistant setting, we used subcutaneous allografts of cells excised from Th-MYCN or Th-MYCN^CPM32 tumors in 129 × 1/SvJ (immunocompetent, strain-matched) mice. As expected, Th-MYCN^CPM32 allografts were refractory to CPM treatment with a 160% mean growth at 7 days after treatment (Fig. 6A), while Th-MYCN allografts underwent complete regression at a dose of 32 mg/kg CPM. In contrast, treatment with 32 mg/kg CPM (once per week) together with 50 mg/kg CYT387 (administered 5 days on 2 days off) significantly reduced tumor volume at day 5 in the treatment resistant Th-MYCN^CPM32 allografts from a mean of 152% ± 21% to 82% ± 20% (Fig. 6B). Furthermore, we observed an increase in survival of 6 days (representing of 37%) from a median of 16 days to a median of 22 days (Fig. 6C). These results establish that the acquisition of chemoresistance following treatment with CPM is transplantable, therefore cell-intrinsic, and that in vivo treatment with the JAK STAT inhibitor CYT387 reduces tumor growth and extends survival in CPM chemoresistant neuroblastoma.

There is a clear need to develop novel therapeutic strategies for children with high-risk neuroblastoma who become resistant to conventional treatment. The difficulty of obtaining surgical biopsies at time of relapse, and a lack of model systems that faithfully recapitulate chemoresistant neuroblastoma together with clinically relevant distant metastases, has hindered further mechanistic understanding of the biology that underpins the emergence of chemoresistant, metastatic disease.

Here, we have optimized the well-established Th-MYCN transgenic model of neuroblastoma, which develops many features of aggressive neuroblastoma but lacks distant metastases. We report that Th-MYCN tumors acquire cell-intrinsic resistance to cyclophosphamide and cross-resistance to other chemotherapeutics following prolonged genotoxic treatment with cyclophosphamide alone. Concurrently, the majority of mice develop metastases to bone marrow, gain of the chromosomal region equivalent to the human 17q and expression of 17q-equivalent localized genes. Furthermore, resistant tumors present microenvironmental–stromal changes and mesenchymal markers suggestive of transition to a migratory-invasive and mesenchymal phenotype. Taken together, these features suggest a model of chemorefractory neuroblastoma that has many of the hallmarks of high-risk chemoresistant metastatic disease.

Tumor–stromal interactions and ECM alterations, together with development of an EMT promigratory signature, play a role in drug resistance and in the development of a metastatic phenotype (25, 36, 37). We demonstrated an increase in collagen IV content and bundle thickness in the chemoresistant model of neuroblastoma (Th-MYCN^CPM32), consistent with similar observations in lung cancer cells, where increased collagen IV deposition stimulates integrin β1–PI3K pathway activity, and supports a prosurvival, drug-resistant state (38). In neuroblastoma, analysis of 102 primary high-risk tumors demonstrated large reticulin fiber networks (20). Taking this data into account, we suggest that the alterations in ECM we observe in this experimental model of chemoresistant neuroblastoma are likely to be associated with development of resistance, although further work will be required to establish the mechanism for this association.

An alternative possibility is that alterations in tumor–stromal interactions and changes in ECM structure can directly impair drug efficacy by acting as a physical barrier to drug penetration (39). In this model, pharmacokinetic studies did not detect any significant changes in intratumoral drug levels of cyclophosphamide or its metabolites in resistant compared with treatment-naïve tumors. This, together with the observation that cells removed from resistant Th-MYCN^CPM32 tumors maintained resistance to cyclophosphamide and acquired cross-resistance to additional drugs in derivative cell culture, suggests that development of chemoresistance in this model is cell intrinsic.

Consistent with prior reports, the development of chemoresistance in the Th-MYCN^CPM32 model may be linked with transition toward a mesenchymal state. In neuroblastoma, expression of genes associated with the mesenchymal core-regulatory state correlates with enhanced in vitro resistance to standard neuroblastoma chemotherapy drugs such as cisplatin, doxorubicin, and etoposide, and is thought to be one feature associated with development of aggressive disease (9).

We detected a proportion of cells in Th-MYCN^CPM32–resistant tumors that exhibited changes in genes and protein expression, consistent with transition toward a mesenchymal state. In keeping with this, bulk RNA-seq of Th-MYCN^CPM32 and Th-MYCN tumor specimens only demonstrated a nonsignificant trend toward the mesenchymal phenotype in the Th-MYCN^CPM32. However, in Th-MYCN^CPM32–derived ex vivo culture, we observed a more distinct mesenchymal signature. This may be a consequence of expansion of these cells related to a preferential cell culture microenvironment, or alternatively enhanced growth properties that are intrinsic to mesenchymal cells. Further mechanistic studies will shed light on these observations. The reduction in MYCN levels seen in our Th-MYCN^CPM32-derived cell lines may also be associated with their mesenchymal type; it has previously been reported that MYCN-amplified lines are consistently adrenergic, whereas cell lines without MYCN amplification may display either signature (40).

Tumors from mice with refractory and metastatic disease were characterized by genomic changes and altered expression patterns characteristic of aggressive disease in human patients; gain of a chromosomal region syntenic to human 17q, Ccnbip1 and TK1 upregulation, and alterations in the JAK–STAT3 pathway. In contrast to neuroblastoma, upregulation of Ccnb1ip1, a ubiquitin ligase, in lung and breast cancer is correlated with prolonged survival (41). Although some of the known molecular functions of CCNB1IP1 would support a tumor suppressor role, its mechanism of action and its full interactive network is not yet completely characterized. TK1 levels are a diagnostic marker useful for clinical decision making in treatment of several malignancies, such as pediatric acute lymphoblastic leukemia (26, 42).

The JAK–STAT3 pathway is upregulated in aggressive neuroblastoma and in the Th-MYCN^CPM32 mice. In previously published in vivo models of neuroblastoma, Withaferin A, a nonexclusive STAT3 inhibitor, and AZD1480, a JAK1/2 inhibitor, both increased tumor cell death through induction of apoptosis (43, 44). We found that the clinical candidate JAK1/JAK2 inhibitor CYT387 counteracts in vitro resistance to both the CPM-active metabolite 4-OOH-CY and to vincristine. Furthermore, in vivo combination with CPM led to substantial growth arrest of these highly aggressive chemorefractory tumors, resulting in an increase in overall survival. In neuroblastoma xenografts, STAT3 modulates drug response through activation of the prosurvival, antiapoptotic protein Bcl-xL, and increases metastatic potential (45). The JAK–STAT3 pathway is a major regulator of tumor cell invasion and immunomodulatory pathways that are co-opted by tumorigenesis (46), implying a complex interaction between tumor cells and the microenvironment. We therefore cannot exclude that the increased signature of the JAK–STAT3 pathway, and the in vivo therapeutic benefit of CYT387 in our model, might arise both from the tumor cells and/or from the tumor microenvironment.

To summarize, we have generated a model of neuroblastoma that spontaneously recapitulates the chemoresistance and bone marrow metastases typical of patients with neuroblastoma at relapse. In the context of genotoxic stress resembling clinical therapy, we observe concomitant evidence of genomic and microenvironmental alterations consistent with development of high-risk, metastatic neuroblastoma in patients. The emergence of mesenchymal-like cells within these tumors, that are a potential source for the metastases that we observe, is an intriguing finding for further study. Finally, this fully immunocompetent, chemorefractory, and metastatic tumor model of aggressive, relapsed/refractory neuroblastoma provides an essential tool for further development of targeted treatment approaches, such as the suggested STAT3 inhibition, in a clinically relevant context.

J. Anderson is a medical and scientific director (employment or paid consulting) at TC Biopharm and Roche, reports receiving a commercial research grant from TC Biopharm, has received speakers bureau honoraria from Eusa Pharma, and has ownership interest (including patents) in TC Biopharm and Autolus Ltd. No potential conflicts of interest were disclosed by the other authors.

Conception and design: O. Yogev, G.S. Almeida, M. Zarowiecki, Y. Jamin, J. Anderson, L. Chesler

Development of methodology: O. Yogev, G.S. Almeid, M. Zarowiecki, A. Hallsworth, Y. Jamin

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): O. Yogev, G.S. Almeida, C. Kwok, L.M. Smith, A. Hallsworth, P. Berry, H.T. Webber, L.S. Danielson, B. Buttery, E.A. Calton, B.M. da Costa, Y. Jamin, G.J. Veal

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): O. Yogev, G.S. Almeida, S.L. George, C. Kwok, J. Campbell, M. Zarowiecki, D. Kleftogiannis, P. Berry, B. Buttery, Y. Jamin, S. Lise, G.J. Veal, N. Sebire, S.P. Robinson, J. Anderson

Writing, review, and/or revision of the manuscript: O. Yogev, G.S. Almeida, K.T. Barker, S.L. George, C. Kwok, M. Zarowiecki, D. Kleftogiannis, E.A. Calton, E. Poon, Y. Jamin, G.J. Veal, N. Sebire, S.P. Robinson, J. Anderson

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): O. Yogev, C. Kwok, M. Zarowiecki, L.M. Smith, A. Hallsworth, T. Möcklinghoff, H.T. Webber, L.S. Danielson

Study supervision: O. Yogev, J. Anderson

Other (pathology): N. Sebire

We would like to thank Gary Box for assisting with in vivo work and Igor Vivanco for critical reading of the manuscript. This study was supported by grants from The Neuroblastoma Society (NES004), Children with Cancer UK (16-215 to O. Yogev), Cancer Research UK, and EPSRC to the Cancer Imaging Centre at ICR and RMH, in association with the MRC and Department of Health (England; C1060/A10334 and C1060/A16464), The Wellcome Trust (091763Z/10/Z), an EPSRC Platform Grant (EP/H046526/1), NHS funding to the NIHR Biomedical Research Centre at The Royal Marsden and the ICR (to G.S. Almeida, Y. Jamin, and S.P. Robinson), Cancer Research UK Programme (C18339), EU FP7, IMI2 (11064 to K.T. Barker). Christopher's Smile Clinical Fellowship; NIHR Biomedical Research Centre (NHR155X to S.L. George), The Brain Tumor Charity, INSTINCT (16-193 to L.M. Smith); The Felix White Cancer Charity (to A. Hallsworth); Cancer Research UK Programme (C18339 to C. Kwok); Children with Cancer UK (16-218 to M. Zarowiecki); Royal Marsden Children's Department Fund (to H.T. Webber); The Neuroblastoma Society (NES003 to L.S. Danielson); ICR HEFCE; Cancer Research UK Programme (C18339 to L. Chesler); GOSH NIHR BRC and NIHR SI award (to N. Sebire); The Wellcome Trust (105104/Z/14/Z to D. Kleftogiannis and S. Lise); Children with Cancer UK Research Fellowship (to Y. Jamin); CRUK, ECMC (to P. Berry, T. Möcklinghoff, G.J. Veal); CRUK C2739/A22897 (to J. Campbell); GOSHCC leadership award and GOSH NIHR Biomedical Research Centre (to J. Anderson), CwCUK 16-033 (to E.A. Calton).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.