Authors: Adam D. Durbin (aDivision of Molecular Oncology, Department of Oncology, St. Jude Children’s Research Hospital, Memphis, TN, USA), Rogier Versteeg (bDepartment of Pediatric Oncology and Hematology, Charité – Universitätsmedizin, Berlin, Germany)
Categories: Article, Neuroblastoma, Plasticity, Embryogenesis
Source: EJC paediatric oncology
Authors: Adam D. Durbin, Rogier Versteeg
Evolving evidence indicates that tumor cells can transdifferentiate between distinct transcriptionally-determined cell states with changes in resultant phenotypes, a phenomenon known as cellular plasticity. These transitions are not driven by genetic mutations and typically in contrast to normal developmental processes, may proceed bidirectionally. Here, we review the role of cellular plasticity in the aggressive childhood solid tumor, neuroblastoma. We discuss the detection of two cell states, termed mesenchymal (MES) and adrenergic (ADRN), their properties and the transcriptional circuitries that control them, their relation to the normal embryogenesis of the sympathetic nervous system and their potential role in drug resistance, escape from therapy and development of relapse.
One of the most devastating features of cancer is that tumors that initially respond to therapy with complete remission, may relapse as therapy-resistant disease. Therapy resistance is classically thought to result from ongoing mutagenesis of tumors, occasionally giving rise to drug-resistant subclonal cells that are selected for and drive tumor progression. In contrast to this model, in recent years, several tumor types have been demonstrated to have highly structured cellular heterogeneity. In glioblastoma, oligodendroglioma, melanoma and small cell lung carcinoma, individual tumors are composed of two or more types of tumor cells with discrete mRNA expression patterns [1–9]. Single cell RNA-sequencing (scRNA-seq) of these tumor types shows that individual tumor cells have immature or more lineage-committed phenotypes. These tumor cells seem to recapitulate aspects of normal lineage-specific differentiation stages. While during normal development, progenitor cells may differentiate into more differentiated cell types, transdifferentiation phenotypes in cancer may be more restricted within a cell type [10]. These phenotypically divergent cancer cells are, in principle, genetically identical and are thought to transdifferentiate into one another, either uni- or bi-directionally, a phenomenon called cell state plasticity. The notion that individual tumors consist of cells in distinct cell states that mimic normal development has therefore changed our view of cancer and provides new opportunities to consider therapeutic strategies. These genetically identical cell states have their own distinct epigenetic structure, expression of transcription factors, differentiation markers, growth factors, receptors and may differ in motility among other phenotypes. One critical finding is that immature-like cell states are often less sensitive to chemotherapy and/or targeted therapies. These immature cell states may therefore play a role in escape from current therapies, that is independent of the contribution of genetic mutations in this process.
These observations have been extended to the high-risk pediatric solid tumor neuroblastoma, where to date, at least two phenotypically divergent cell states have been identified. Neuroblastoma is a pediatric solid tumor with a low mutational burden, thought to be derived from cells of the neural crest lineage [11]. These tumors are clinically sub-classified into low-, intermediate and high-risk tumors, which differ in molecular features, prognosis and treatment. Low-risk tumors are typically diagnosed in children younger than 18 months and have an excellent prognosis, while high-risk tumors, making up roughly 50 % of all patients, are characterized by an older age at onset (>18 months), frequent amplification of the MYCN oncogene, and/or disseminated disease [12]. The standard treatment of high-risk neuroblastoma includes surgery, tandem high-dose chemotherapy with autologous stem cell rescue and radiotherapy. Immunotherapy using antibodies targeting the neuroblastoma antigen GD2, often in combination with retinoic acid, is included in the standard consolidation therapy [13,14]. ALK inhibitors are included for the therapy of tumors harboring ALK mutations [15]. Crucially, while many high-risk neuroblastomas respond to therapy, overall nearly half of children with high-risk neuroblastoma will ultimately relapse with lethal drug-resistant disease [11,16].
Cell state plasticity is a process where cells may undergo major epigenetically and transcriptionally-determined phenotypic changes that are not driven by genomic mutation, and are therefore reversible. To highlight the potential role of plasticity in neuroblastoma, we first briefly summarize below what is known of the irreversible genomic defects found in primary tumors and relapses.
Primary high-risk neuroblastomas frequently demonstrate gains of extra copies of the 17q chromosome arm, and deletions of large regions on chromosome arms 1p or 11q. The role of these lesions in pathogenesis is enigmatic, but their frequency and consistent pattern suggest that they are major drivers of tumorigenesis. Gain of the chromosome 17q region is found in almost 100 % of high-risk cases. Deletions of chromosome 1p and 11q are mutually exclusive and each have a frequency of 30–40 %. Amplification of the MYCN gene is found in roughly 1/3 of high-risk tumors and is almost always accompanied by loss of the chromosome 1p region, but only rarely by 11q losses. High-throughput sequencing of neuroblastoma has shown that other gene defects are characteristically scarce, save for TERT rearrangements (25 %), ALK mutations (10–15 %) and ATRX rearrangements (10 %). More rarely, mutations have been identified in PTPN11 and ARID1A [17,18]. Defects of MAP kinase (MAPK) pathway genes are infrequent in primary tumors [18–21].
In contrast, analyses of genomic lesions in relapsed tumors have lagged behind that of primary tumors. The diagnosis of relapsed neuroblastoma can be made by non-invasive methods like urinary catecholamine measurement and imaging using the radioisotope ^123^I-meta-iodo-benzylguanidine (MIBG) [22–24]. As a result, it is only recently that systematic biopsy of relapsed tumors has been undertaken. Early analyses of selected samples suggested a strong enrichment of ALK and MAPK pathway mutations [25,26]. The increase in ALK mutations recently turned out to be more modest. Sequencing of ALK in tumor samples from a large cohort of matched primary and relapsed neuroblastomas demonstrated a mutational frequency of ~10 % at diagnosis, and 17.7 % at relapse [27], compared with ~16–25 % in a smaller cohort [28]. In general, relapsed tumors display a higher frequency of MAPK pathway mutations [26,27,29]. It is possible that mutations of ALK and other genes were present sub-clonally, and undetectable with current sequencing technologies in the primary tumor, and that these clones expanded after treatment, resulting in enhanced detection. Most importantly, it is unknown if the additional mutations found in relapses are responsible for the treatment-resistance character of relapses. As primary and relapse samples often present with mutations in the same genes, it is unclear how they would specifically convey a resistant phenotype in relapse. Thorough evaluation of whole genome sequencing, reconstruction of the mutational evolution of primary neuroblastomas in relapse, and specific testing of these mutations alone and in combination, on chemoresistant phenotypes and cell states will be required to evaluate the contribution of both the coding and non-coding genome to tumor evolution under the selective pressures of chemotherapy, radiation and immunotherapies.
The first description of phenotypic plasticity in neuroblastoma dates from 1983, when June Biedler and Barbara Spengler identified distinct phenotypic subtypes of neuroblastoma cells, in vitro [30]. Using the MYCN-non-amplified neuroblastoma cell line SK-N-SH, Spengler and Biedler isolated a subclone called SH-SY-5Y which displayed a more epithelial, flattened morphology than the parental cell line. They also identified a separate subclone, termed “SH-EP” which, while initially appearing epithelial, could be subcloned to cells with a more “neuroblastic-like” phenotype [30,31]. Intriguingly, these cells could interconvert between epithelial and neuroblastic-like morphologies, and the tantalizing suggestion was made that interconversion may also occur in patients [30,31]. These cells were termed “N-type,” for neuroblastic-like, and “S-type” for substrate-adherent/epithelial in morphology [32]. N-type cells expressed neuronal markers, which were absent in S-type cells. In contrast, S-type cells displayed a distinct cell surface marker profile, resembling ectodermal or mesenchymally-derived cells [32]. The cells differed in cytoskeletal morphologies [33], expression of collagens [34], fibronectin [35], tyrosine hydroxylase and dopamine-beta-hydroxylase and vimentin [36]. A third type of cell, termed “intermediate” or “I-type” expressed both N-type and S-type cell properties [37], suggesting that the I-type cell may function as either the stem cell for N- and S-type cells, or, alternatively, a transitory state between the N- and S-type cells during their interconversion [38]. In subsequent years, several reports described that neuroblastoma cell lines selected for in vitro resistance to chemotherapeutics could have a more mesenchymal phenotype (see below) [39–41]. However, with the human genome project and wave of new technologies that came available at the onset of the 21st century, the main focus in neuroblastoma research shifted towards genomic analyses. The hope was that when the gene mutations that drive neuroblastoma would be identified, targeted therapy of these genes would provide a breakthrough in treatment options. Consequently, studies of non-genetically driven phenotypic changes in neuroblastoma were less pursued.
The first high throughput mRNA profiling studies of neuroblastoma revealed an extremely high expression of Delta-like 1 (DLK1), which is a ligand of NOTCH receptors [42,43]. DLK1 expression correlated with that of the differentiation markers DBH and TH in a neuroblastoma cell line panel, but inversely correlated with NOTCH expression, suggesting two divergent differentiation states in the cell lines [43,44]. Notch signaling plays a key role in embryogenesis and can mediate the development of one precursor cell into two divergent daughter cells. Induction of NOTCH signaling in neuroblastoma cell lines was subsequently found to induce a motile and more mesenchymal phenotype [45]. A 36 gene-signature for this mesenchymal phenotype nominated NOTCH as a molecular master switch in 24 neuroblastoma cell lines, which contrasted to cell lines with a lineage committed phenotype and expression of the more adrenergic marker genes TH and DBH. The analyses of both cell line types, called “MES-type” for mesenchymal and “ADRN-type” for the adrenergic phenotype, were extended to genome-wide chromatin modification studies [46]. MES and ADRN cell line pairs derived from the same patient showed two divergent patterns of chromatin modification at enhancers of hundreds of genes. In parallel, Boeva et al. showed similar epigenetic differences in neuroblastoma cell lines [47]. These distinct patterns were reminiscent of the concept of core regulatory circuitries (CRCs), initially identified in embryonic stem cells [48,49], which are networks of positive feed-back loops of master transcription factors whose gene loci are controlled by super-enhancers. CRC’s are thought to impose cell state specification, by establishing the expressed transcriptome. In neuroblastoma, distinct highly expressed, super-enhancer regulated transcription factors were identified in MES-type and ADRN-type cells [46,47,50]. Strikingly, these differences aligned with the conventional SH-SY-5Y and SH-EP cell lines, suggesting that they were controlled by the same epigenetic and transcription factor mechanisms. By K-means expression clustering, SH-EP and other MES cell lines aligned with five neural crest cell lines, suggesting that MES cell lines were more closely related to neural crest progenitors [46]. Expression analyses of MES and ADRN cell lines identified signatures of 369 and 454 genes that mark these two cell states, which are now widely implemented [46].
The CRC proposed for the ADRN cell state included PHOX2B and ASCL1 [47,50,51]. The transcription factor PHOX2B, which is mutated in sporadic and familiar neuroblastoma [52–54], was highly expressed in ADRN neuroblastoma cells and more weakly expressed or absent in MES cells. PHOX2B is a master transcription factor for the adrenergic lineage and PHOX2B knock-out mice fail to develop adrenaline-producing cells [55]. PHOX2B induces ASCL1 expression in early adrenergic development, suggesting that both genes are early activators of the adrenergic CRC. The role of these transcription factors in adrenergic neuroblastoma cells would ultimately be confirmed and extended, harnessing exome-wide CRISPR-cas9 knockout studies in combination with ChIP-sequencing to demonstrate that only a subset of these super-enhancer regulated, highly expressed transcription factors, including HAND2, ISL1, GATA3, PHOX2B, TBX2, ASCL1 and TFAP2B, physically bind to their own and each others super-enhancers to form a core-regulatory circuit that establishes the adrenergic cell transcriptome in MYCN-amplified cells [50,51,56,57]. Specific disruption of any of these factors results in loss of expression of the others, confirming the autoregulatory nature of the circuit [50,56]. Importantly, these transcription factors play a role in maintaining cell identity, as their loss caused reduced cellular fitness [50]. Further highlighting the importance of these transcription factors were studies identifying the “hijacking” of some of their loci, including HAND2, ISL1 and GATA3 to drive high-level expression of telomerase or MYC oncogenes [21,58]. Subsequently, additional factors have been implicated in the adrenergic core regulatory circuitry, including SOX11 [59], MEIS2 [60] and the non-DNA-binding linker protein LMO1 [51,61,62]. Additional, specific roles for individual members of the CRC have been suggested (reviewed in [63]), including binding and recruiting the histone acetyltransferase EP300 by TFAP2B [56], similar binding and recruitment of the BAF chromatin remodeler by SOX11 [59] and potential pioneer transcription factor activity by ASCL1 [64]. To date, a CRC for MES cells is undefined, as MES cell lines are more heterogeneous in their transcription factor expression than ADRN cell lines.
As demonstrated by Spengler and Biedler nearly forty years earlier, MES and ADRN cell states were also able to switch between each other. Spontaneous transitions in both directions occur in vitro [46]. A quantitative study of in vitro transitions between the MES and ADRN states of the cell line SK-N-SH shows swift transitions from the ADRN to the MES state, and slower transition in the opposite direction [65]. Further, transcription factors can induce a transition of ADRN cells to a MES phenotype, as shown for PRRX1 and NOTCH3 [45,46,66]. Activation of NOTCH3 starts a feed-forward loop in which other NOTCH paralogs as well as NOTCH ligands were also activated, thus shedding light on how upstream signaling can break through the constraints of a cell state to induce rewiring [66]. NOTCH-induced MES cells show loss of ADRN core-regulatory circuitry gene expression and remodeling of genome-wide super-enhancer localization. Induction of transitions in the opposite direction (MES to ADRN) by transcription factors have not been reported to date.
Tumor-cell intrinsic genetic and environmental factors may also influence cell state, resulting in intermediate states, subtypes of the ADRN and MES state, or entirely new or different states. For example, in addition to NOTCH3 and PRRX1, Shi et al. demonstrated that loss of ARID1A, which is found mutated in a fraction of neuroblastomas [17, 18], in ADRN cells results in a partial switch to a more MES-like gene expression pattern [67]. Further subdividing the known cell states, Gartlgruber et al. demonstrated that the ADRN super-enhancer landscape has three subtypes, found in MYCN-amplified (MA), MYCN-non-amplified high-risk (MNA-HR) and MYCN-non-amplified low risk (MNA-LR) samples respectively. These subtypes are marked by different, but overlapping, super-enhancer profiles [68]. Exogenous treatments may also affect cell state. For example, olfactomedin, which is produced by embryonic sympathetic ganglia, contributes to an ADRN to MES shift [69]. Further, retinoic acid treatment, used clinically to induce neuronal differentiation of neuroblasts, causes the production of yet an additional cell state, termed the “retino-sympathetic” cell state [70]. In this state, neuroblastoma cells form a more differentiated, neuronal-like state, marked by expression of SOX4 and MEIS1, with reduced expression of conventional adrenergic core-regulatory circuitry members, such as GATA3, PHOX2B and ASCL1 [70]. Importantly, several members of the adrenergic core-regulatory circuitry, including HAND2, ISL1 and TBX2 remain expressed and active members of the retino-sympathetic CRC, indicating that modular sets of TFs are required to establish distinct states, and that some TFs may be shared between them [70]. Together, these results highlight the observation that there are probably more than two distinct states in neuroblastoma, and that existing methodologies to label individual states require refinement to capture the potential diversity of cell states and their corresponding malignant properties occurring in any given neuroblastoma (Fig. 1).
Analyses of MES and ADRN gene expression signatures showed that ADRN cells display differentiation markers of the adrenergic lineage, such as the enzymes of the (nor)adrenaline synthesis pathway. Absence of these markers in MES cells suggested that ADRN cells were found in an adrenergic lineage, while MES cells had a more immature character. Soon after the elucidation of the enhancer and expression differences of MES and ADRN cells, two studies established expression profiles of the main embryonal cell types of the sympatho-adrenergic lineage [71,72]. These studies permitted a comparison of these tumor cell states with the embryonal cell types found in normal differentiation.
The sympatho-adrenergic lineage emerges from the neural crest (NC). In early embryogenesis, neural crest cells delaminate and migrate ventrally in several waves to give rise to numerous anatomical structures and cell types (reviewed in [73–75]). In the first wave, trunk NC-derived cells migrate ventromedially and form ganglia around the dorsal aorta, called the primary sympathetic ganglia. From there, a subset of cells migrate ventrally from the dorsal aorta to form the adrenal medulla. In the primary sympathetic ganglia and adrenal medulla, these migratory cells differentiate to chromaffin cell or sympathetic neuroblasts. Chromaffin cells function to secrete noradrenaline, while sympathetic neuroblasts will form neurons, which do not innervate the chromaffin cells. Innervation to the adrenal medulla is provided by another subset of cells migrating from the dorsal aorta that form the secondary, postganglionic sympathetic neurons of the celiac and suprarenal plexi.
The cells in this first migratory wave have been analyzed by classical technologies like immunohistochemical markers and tracing of dye-injected neural crest cells. However, the first wave migration has not yet been studied by genetic tracing and/or scRNA-seq studies. In contrast, more details are known of the second migratory wave. In this wave, neural crest derived cells migrate more ventrolaterally and use the visceral motor nerve as a guide towards the adrenal medulla. The cells migrating along the visceral motor nerve were identified as Schwann Cell Precursors (SCP cells), which are marked by SOX10 expression. When SCP cells arrive at the adrenal medulla, some remain outside of the medulla to populate the suprarenal ganglia that innervate the adrenal medulla. Other SCPs penetrate the medulla to differentiate into chromaffin cells and sympathetic neurons. Two landmark studies have analysed the murine and human adrenal medulla during migration of SCP cells. First, the Adameyko group performed a scRNA-seq analysis of mouse E13.5-E17.5 stages and human week 6–14 post-conception (PC) stages [71,76]. Second, the Westermann group isolated adrenal medulla cells from human week 7–17 PC embryos [72]. Together, these studies showed that migratory SCP cells first transit through a bridge cell stage and subsequently a ‘connecting progenitor cell’ stage. Following this, they split into either a chromaffin or sympathetic neuron fate. These studies thereby depict the transition from the mesenchymal, migratory SCP state towards the neuro-epithelial chromaffin and sympathetic neuron states during normal development.
The identification of these normal embryonic states using scRNA-seq has allowed a comparison with the signatures of ADRN and MES tumor cell states. These comparisons identified that the MES cell signature was highly aligned with the SCP state, which is in line with the migratory phenotype of SCP cells and the motile phenotype of cultured MES cells. The ADRN cell signature, however, corresponded to the sympathetic neuron state, in line with the expression of genes for the noradrenaline synthesis pathway in both cell types [72,77]. Signatures of the ADRN tumor cell state were far less prominent in chromaffin cells. These data suggest that the two fundamentally different and interconverting tumor cell states in neuroblastoma recapitulate two different stages of normal sympatho-adrenal development. This provides a lead towards understanding how and why neuroblastoma cells can adopt two distinct transcriptionally-programmed phenotypes. A major question that arises from these observations is why in normal development do cells differentiate uni-directionally from immature towards lineage-committed cells, while in neuroblastoma cell lines, tumor cells can bi-directionally switch between the MES and ADRN state?
The prevalence of MES-type neuroblastoma cells in human tumors in vivo is less clear. The presence of MES cells can currently only be demonstrated by scRNA-seq, as bulk RNA will also include RNA from normal infiltrating stroma and immune cells which also may have a mesenchymal signature. Thus, it is important to consider this caveat when considering bulk RNA analysis of human primary tumors. Three scRNA-seq analyses of neuroblastoma series did not identify MES-type cells in primary tumors [72,78,79]. Westermann and co-workers identified rare tumor cells with a mixed ADRN-MES phenotype [72]. In contrast, a recent manuscript deposited in BioRxiv describes the presence of MES cells in a much larger neuroblastoma series [80]. A few technical aspects are important in considering the absence of MES cells in tumor series. Firstly, MES and ADRN tumor cells are in principle detected by their signatures, but this requires sufficient mRNA read counts in single cells. The quality of the tumor RNA in these studies was often limiting, resulting in inadequate detection of a cell state in part of the cells. Secondly, when a cell with a mesenchymal signature is detected, it can either be a normal infiltrating stroma cell or a MES-type tumor cell, as both have the same mesenchymal signature [81]. Identification of the tumor cells among them is usually indirect and based on calculation of an inferred copy number, to identify cells with chromosomal copy number aberrations. This bioinformatic approach is also only valid with high mRNA read counts per cell. Finally, choice of tumor region for performing scRNAseq may also be important. Evidence indicates that MES cells display enhanced inflammatory signaling [82,83], suggesting that they may be present in more inflamed and perhaps hypoxic regions of the tumor. Thus, spatial resolution of whole tumor sections, rather than preferentially viable segments of tumor, may also be required. A definitive conclusion about presence of perhaps minor fractions of MES-type tumor cell in neuroblastoma remains therefore challenging and an important area for enquiry.
Besides the open question of whether MES-type cells exist in neuroblastoma at diagnosis, a major question is whether neuroblastoma cells display plasticity in vivo and can adopt a MES-like state during stresses such as metastasis or therapy. If so, they may survive treatment and the minimal residual disease stage. Limited analyses of relapsed tumors suggest that they are predominantly ADRN, though given the potential for cell state plasticity, this does not exclude that MES cells may have predominated at an MRD state. To this end, animal models have given additional insight in the cellular dynamics in tumors. Inoculation of SK-N-SH cells consisting of MES and ADRN subpopulations gives rise to entirely ADRN tumors, suggesting the in vivo environment provides a selection for the ADRN phenotype [84]. Similarly, neuroblastoma that spontaneously develop in the transgenic TH-MYCN mouse have an ADRN phenotype in vivo, but acquire a MES phenotype upon culturing of tumor explants [85]. The murine neuroblastoma cell line NB-9464, derived from tumors arising in the TH-MYCN mouse model, retains MES and ADRN subpopulations [83]. Although MES cells thus seem to be selected against in vivo, Westerhout et al. reported that neuroblastoma cells with an NOTCH3-induced MES-phenotype are oncogenic in mice and maintain their MES phenotype, albeit with a slower engraftment [77]. More recently, short term treatment with cisplatin was demonstrated to be sufficient to enrich the detection of MES-like cells in highly ADRN neuroblastoma genetic models, in vivo [81]. Thus, MES cells may thus be counter selected in vivo, but neuroblastoma cells seem to maintain their plasticity and the potential to adopt a MES phenotype both in vitro and in vivo.
To discuss whether MES cells play a clinically relevant role, we will first consider the typical clinical course of high-risk neuroblastoma. In many cases of high-risk neuroblastoma, tumors are already metastatic at diagnosis. In many tumor types and models, it has been established that metastatic spread is mediated by tumor cells that underwent epithelial-to-mesenchymal transition (EMT) which enabled them to invade basement membranes, enter the circulation and seed other organs. It has not been studied whether MES-type cells in neuroblastoma are responsible for the metastatic behavior, but may be possible since the ADRN cell state appears less migratory [45,46]. Prevention of metastases is not relevant in high-risk neuroblastoma, as tumors are commonly disseminated at diagnosis. Upon diagnosis, patients are treated with extensive cytotoxic therapy. A minority of patients display either stable or progressive disease under therapy and they have an especially poor prognosis. However, many patients with high-risk neuroblastoma have a favorable response and go into complete or near-complete clinical remission. Unfortunately, roughly half of patients in clinical remission will progress to develop a relapse, which is nearly uniformly fatal [16]. Thus, the clinical course of relapsed neuroblastoma often involves an initial stage of response, followed by therapy-resistant relapses. Thus, given the sometimes prolonged time-periods between end of initial therapy and relapse, it is clear that a very small number of tumor cells must have survived therapy and are able to generate a relapse, even years later.
Early studies of drug resistance in neuroblastoma yielded several instances of an association with a mesenchymal phenotype. Induced cisplatin resistance in the neuroblastoma cell lines SK-N-AS, Kelly and CHP-212 was associated with a migratory and invasive phenotype and changes in expression of proteins involved in epithelial adherens junctions, suggesting a more mesenchymal phenotype [39]. These cells were also cross-resistant to temozolamide, etoposide and irinotecan. Similarly, doxorubicin-resistant cell lines derived from SK-N-SH and SK-N-BE2(C) cells demonstrated markedly distinct gene expression patterns with upregulation of epithelial-to-mesenchymal (EMT)-like and TGF-β gene signatures [40]. Supporting these findings, serial in vivo passage of SH-SY-5Y cells repeatedly isolated from metastatic sites resulted in highly aggressive cell lines with high expression of the drug-resistance protein ABCG2, in addition to altered NF-KB signaling, expression of cadherins, MYC and NOTCH-1 proteins [41]. Together, these observations indicated that chemo-resistant neuroblastoma cells may adopt a more mesenchymal state. In each study, these passaged or selected cells were not profiled for mutational burden. However, these resistant states showed plasticity depending on in vitro culture conditions, suggesting that these phenotypes were unlinked to genetic mutation.
When isogenic MES and ADRN neuroblastoma cell lines, derived from the same patient, became available, it was possible to compare drug resistance in MES and ADRN cells with a genetically similar background. In three such pairs, MES cells were consistently more resistant to cisplatin, doxorubicin and etoposide than their ADRN counterparts [46]. Correspondingly, treatment of SK-N-SH cells with doxorubicin resulted in enrichment of the MES signature [47]. Cisplatin treatment of SK-N-SH cells enriched for drug-tolerant persister populations with MES-like properties, but other drugs did not change the balance between both cell states [65]. Importantly, cisplatin treatment killed all ADRN cells, but MES cells partly survived and subsequently gave rise to a heterogeneous MES-ADRN population. While these studies suggest that MES cells can escape treatment, patient studies lag behind. Small numbers of matched primary and relapsed samples suggested that patients with initially adrenergic-dominant tumors may display an enrichment of mesenchymal signatures at relapse [46,68,83]. Consistently, neuroblastoma xenograft studies have reported that after chemotherapy, the remaining tumors have a more mesenchymal mRNA expression pattern which may suggest selection [86]. These studies are echoed by recent observations demonstrating the rapid induction of MES-like cell states in cisplatin-treated tumors in vivo, in otherwise ADRN-dominant genetic models of neuroblastoma [81]. While these in vivo findings await confirmation by scRNA-seq analyses in larger series, this drug resistance data shows the potential clinical relevance of these two cell states.
A second group of drugs that were tested on these isogenic cell line pairs were targeted ALK-inhibitors. These are used for treatment of neuroblastoma with activating ALK mutations and give clinically encouraging responses [15]. Remarkably, all analyzed MES cell lines lack ALK expression, while ADRN cell lines typically express ALK [77]. Even in isogenic cell line pairs derived from ALK-mutated neuroblastoma tumors, MES cells lacked ALK mRNA and protein expression. Consequently, MES cells were found to be resistant to the ALK inhibitors Lorlatinib and Alectinib. In mice, MES cells with an ALK mutation form rapidly growing tumors that lack ALK expression and are resistant to Lorlatinib, opening up the possibility that MES cells may escape ALK-inhibitor treatment in a clinical setting [77]. A third clinically used drug that is ineffective for MES cells is retinoic acid (RA). RA induces reversible differentiation of many neuroblastoma cell lines, which led to testing in clinical trials, with subsequent clinical approval [70,87,88]. MES cells were resistant to exogenous RA supplementation. MES cells were demonstrated to synthesize RA themselves and require it for proliferation [89]. These observations mirror normal development, during which RA is required to stabilize and commit to a neural crest lineage, though later in development drives differentiation [90]. Finally, it was found that MES cells largely lack expression of the GD2 modification which is a clinically-relevant target for immunotherapy of neuroblastoma [91]. Anti-GD2 antibody treatment has significantly improved outcome in high-risk patients [92]. EZH2 inhibitors upregulated GD2 in MES cells and made the cells sensitive to anti-GD2 treatment in xenografts, with simultaneous induction of a set of adrenergic genes.
These data indicate that identifying compounds that specifically target MES cells is a critical goal, not only to test in animal model systems whether they can prevent relapse, but also to prepare for clinical trials to combine with anti-adrenergic therapies and to thus improve the survival of neuroblastoma patients. One potential drug may be the lymphokine TRAIL, which specifically kills MES cells in vitro and gave a modest delay of relapse in a mouse model [77]. Interestingly, MES cells were found to have enhanced expression of transcripts associated with immune activation or inhibition [83]. This may offer targets to eliminate MES cells, but also points to a role in the immune modulation of the tumor microenvironment. Proinflammatory cytokine release from mesenchymal neuroblastoma cells was linked to enhanced T and NK cell infiltration into tumors, with enhanced tumor cell killing [83]. In addition, MES cells were noted to be highly responsive to double stranded RNA sensing through the cGAS-STING pathway, which resulted in the secretion of proinflammatory cytokines that could be harnessed to drive increased T-cell-mediated killing of tumor cells in vitro [82]. Together, these observations indicate a potential mechanism by which localized T and NK-cell infiltration, promoted by mesenchymal cells, is sufficient to enhance the anti-GD2 targeting of adrenergic cells. Moreover, these observations clearly indicate that the pro-inflammatory effects of mesenchymal cells may represent key targets for therapeutic exploitation.
The systematic escape of MES cells from many treatments that efficiently kill ADRN cells illustrates key differences between these two cell states. These results suggest a tantalizing model, whereby chemotherapeutic pressure results in a “switch” from adrenergic-dominant tumors to enriched mesenchymal tumors [93]. Equally possible, however, is the model that chemotherapy does not enforce a cell state transition, but rather depletes adrenergic cells, leaving behind mesenchymal cells that are able to repopulate the tumor with self (mesenchymal) cells, in addition to adrenergic cells through a “mesenchymal-to-adrenergic” (MAT) switch [84,93,94]. Validation of this hypothesis will be important, as while the prognosis of high-risk neuroblastoma has improved over the years with the sequential addition of increasingly toxic treatment regimens, survivors of the disease are left with enormous morbidities [95]. Moreover, current treatments are insufficient to cure all children, largely due to relapses that form after complete clinical remission or disease that is refractory to therapy in the first place [95]. Much research is therefore needed to clarify how tumor cells behave during complete and partial clinical remission, what the properties and states of the relevant residual cells are, how they are shaped by distinct therapeutic modalities, what their vulnerabilities may be, and ultimately, how they can re-enter a proliferative state and form lethal relapses.