Developmental origins and emerging therapeutic opportunities for childhood cancer

Abstract

Cancer is the leading disease-related cause of death in children in developed countries. Arising in the context of actively growing tissues, childhood cancers are fundamentally diseases of dysregulated development. Childhood cancers exhibit a lower overall mutational burden than adult cancers, and recent sequencing studies have revealed that the genomic events central to childhood oncogenesis include mutations resulting in broad epigenetic changes or translocations that result in fusion oncoproteins. Here, we will review the developmental origins of childhood cancers, epigenetic dysregulation in tissue stem/precursor cells in numerous examples of childhood cancer oncogenesis and emerging therapeutic opportunities aimed at both cell-intrinsic and microenvironmental targets together with new insights into the mechanisms underlying long-term sequelae of childhood cancer therapy.

Childhood cancers represent the leading cause of disease-related morbidity and mortality in childhood, second only to accidents as a cause of pediatric death¹ in the United States and other developed countries. Childhood cancers encompass leukemias, lymphomas, central nervous system tumors, sarcomas of bone and soft tissue, neuroblastoma, retinoblastoma, rhabdoid tumors, liver tumors, renal tumors, germ cell tumors and additional rare cancers. Revolutionary advances in next-generation sequencing technology together with rapidly increasing progress in chromatin and stem cell biology have ushered in a new molecular understanding of childhood cancer. Recent landmark sequencing studies have demonstrated that the mutational burden in most childhood cancers is substantially lower than that in adult cancers^2,3. Fusion genes are more common than in adult cancers, and certain specific mutations found in pediatric cancers are rare in adult malignancies. Rather than numerous mutational ‘hits’ frequently observed in adult cancers, the emerging theme is that epigenetic dysregulation is central to many forms of childhood cancer. Furthermore, evidence from genetically engineered mouse models of childhood cancer suggest that many pediatric tumors originate from stem or progenitor cells during particular developmental time windows^4–11. It is clear that these cells of origin must provide a transcriptional program permissive for the tumorigenic effects of a first genetic or epigenetic hit, which, as a consequence, distorts further cell divisions toward favoring self-renewal over differentiation. Thus, the developmental context of cells in which tumorigenic mutations occur and the microenvironment in which they form both underscore the unique biology of childhood cancer and the challenges of therapeutically targeting these cancers in actively developing tissues, requiring unique therapeutic approaches. This Review will discuss the principles of childhood cancer pathobiology and therapy, chiefly focusing on childhood leukemias, brain tumors, rhabdoid tumors and sarcomas to illustrate these principles.

Mechanisms of deregulation in childhood cancers

Common types of genetic alterations.

Not only is the rate of mutations and structural variants lower in childhood malignancies compared with adult cancer, but also the types of alterations and particular genes affected differ from adult cancers. Pediatric-cancer-driving point mutations are enriched in genes that encode epigenetic machinery and are largely specific to the diseases in which they arise^2,3. Additionally, chromosomal fusion events that juxtapose oncogenes with gene partners that deregulate their proper activation or function are particularly prevalent among many types of childhood cancers (Supplementary Table 1). These fusion genes often activate genes crucial to development, such as the neurotrophic growth factor receptor family (NTRK) genes^12,13. Another feature of pediatric cancers is that a rather high percentage of pediatric cancer patients (~8%)² carry an unambiguous germline mutation that predisposes them to developing cancer. The extent to which cancer surveillance efforts may identify those at greater risk for childhood malignancy and whether any interventions could be used at the pre-malignant state remain under investigation.

Histone mutations.

Rather than the multiple genomic events that characterize oncogenesis in many adult cancers, a prominent feature of childhood cancers is epigenetic dysregulation, with malignancy attributed to broad dysregulation of gene expression. One striking example of this principle is the discovery of point mutations in histone genes that dysregulate histone post-translational modifications modulating gene expression^14–16. The best-studied example of this is the histone-3 (H3)-K27M mutation that occurs in diffuse midline gliomas, such as diffuse intrinsic pontine glioma (DIPG) and thalamic and spinal cord gliomas of childhood¹⁷. In DIPG, the H3-K27M (rarely H3-K27I) mutation occurs in both canonical and non-canonical H3 genes encoding H3.3 and H3.1 and is found in ~80% of DIPG cases. While the mutation is always heterozygous, and so affects less than 15% of the total histone protein pool in a cell, a profound decrease in overall H3-K27 methylation is observed with the mutation^18,19. The proposed mechanistic model is that the H3-K27M mutation leads to a global decrease in H3-K27 trimethylation by causing dysfunction of the methyltransferase activity of the catalytic enhancer of zeste homolog 2 (EZH2) subunit of polycomb repressive complex 2 (PRC2), preventing methylation of wild-type (WT) H3 at residue 27. The resulting derepression of gene expression then promotes gliomagenesis^4,5,18,20 (Fig. 1). Oncogenesis is only thought to happen if the mutation occurs in the context of a cell in a susceptible transcriptional state, such as early neural or glial precursor cells^4,5,20–22.

Fig. 1 |. Epigenetic dysregulation and therapeutic opportunities in midline gliomas with histone mutations.

a, H3.3- or H3.1-K27M mutations (green histones) influence transcriptional regulation through global reduction of methylation (shown by red stars, which are absent in a) at position H3-K27 and global increases in acetylation (yellow circles) at H3-K27 (refs. ^18,19). This alters local as well as distal gene regulation through super-enhancers and leads to de-repression of target oncogenes. b, Epigenetic modifiers panobinostat (HDAC inhibitor)⁸⁷, GSKJ4 (histone lysine demethylase inhibitor)⁸⁸ and JQ1 (bromodomain inhibitor)^89,90 as well as transcriptional inhibitors like THZ1 (cyclin-dependent kinase 7 inhibitor)⁸⁹ partially counteract the downstream effects of the histone mutation, including restoration of H3-K27 tri-methylation (red stars) by panobinostat and subsequent repression of target oncogenes⁸⁷. POL II, RNA polymerase II. BRD4, bromodomain-containing protein 4. CDK7, cyclin-dependent kinase 7. PRC2, polycomb repressive complex II.

Aberrant patterns of H3-K27 methylation also occur in subgroups of pediatric ependymomas and medulloblastoma, although at a low frequency for the latter^24,25. This convergence of genetic and epigenetic alterations of K27 in H3 variants in tumors of the hindbrain may point toward an important role of H3-K27 during normal development and tumorigenesis in the hindbrain. This point is further exemplified by posterior fossa ependymomas, discussed in more detail below.

Extending the role of ‘oncohistones’ to tumors outside of the hindbrain, the oncogenic histone mutations H3-K27M and H3-K27L are also found occasionally in acute myeloid leukemia (AML) in adolescents and adults²³. Other histone mutations further illustrate spatio-temporal and tumor-type specificity: exclusive H3.3- H3.3-G34R and H3.3-G34V mutations occur in hemispheric pediatric glioblastomas, while H3-K36M mutations occur in chondroblastomas of young adulthood and human papillomavirus (HPV)-negative head and neck squamous cell carcinomas in older adults, and H3-G34W and H3-G34L mutations occur in giant cell tumors of the bone, chiefly affecting young adults^26–28.

Epigenetic modifiers.

In addition to oncohistones, mutations in epigenetic modifiers such as histone modifying enzymes (for example, SETD2 and KDM6A) and chromatin complexes (for example, the BRG1-associated factors (BAF) complex and polycomb repressive complex 2 (PRC2)) have emerged as the most common and largest group of mutated genes in a recent pan-cancer analysis of pediatric cancers^2,3. Notably, all types of pediatric leukemias (T cell acute lymphoblastic leukemia (T-ALL), B cell ALL and AML) exhibited prominent somatic mutations in genes with roles in epigenetic modification (present in 53.7%, 38.7% and 34%, respectively). Frequently mutated genes include those encoding histone methyl transferases SETD2 and MLL1, histone demethylase KDM6A, histone deacetylase HDAC2, chromatin complex BAF member ARID1A, histone deacetylase interactors BCORL1 and BCOR, and isocitrate dehydrogenase 1 and 2 (IDH½—when mutated, this produces abnormal metabolite 2-hydroxyglutarate, which inhibits histone and DNA demethylases)^2,3,29–31.

CpG methylation.

In two large studies of pediatric tumors representing two dozen molecularly defined tumor types across over 2,500 cases^2,3 analyzed by whole-genome sequencing (WGS), 10% of all pediatric tumors did not have any underlying mutation or structural/copy-number variant^2,3. This suggests an additional, non-genetic layer of epigenetic dysregulation at the root of tumorigenesis. One example is a subtype of ependymoma occurring in young children with a very low rate of recurring DNA mutations³². These ‘genetically quiet’ ependymomas of the posterior fossa (posterior fossa group A ependymomas (PFAs)) exhibit CpG-island hypermethylation compared with other forms of ependymoma, implicating broad epigenetic alterations in their pathogenesis^32,33. Intriguingly, PFAs also have global histone hypomethylation at H3-K27 (ref. ³³), analogous to the abovementioned pediatric midline gliomas with a H3-K27M mutation. While the cause and mechanisms of this largely non-genetic regulation of DNA and histone methylation are not yet known, a recent study reported H3-K27M mutations in 4.2% of PFAs²⁴, suggesting the convergence of pathogenic mechanisms involving H3-K27-regulated gene expression in hindbrain tumors.

PFA ependymomas might be an extreme illustration of locked-in developmental stages at the center of tumorigenesis. Recent work has demonstrated that altered chromatin landscapes and super-enhancers in ependymomas activate associated genes on which tumor cells depend for proliferation³⁴ and underscores the need for new epigenetic drugs targeting these molecular alterations. The absence of genomic alterations in most cases of ependymomas also exemplifies the need for including novel molecular platforms (see Box 1), like DNA methylation array testing, in the diagnoses and stratification of these highly diverse subtypes of pediatric brain tumors³⁵. DNA methylation arrays can also be used to precisely classify undifferentiated tumors, especially sarcomas³⁶.

Box 1 |. Epigenetic diagnostics in the clinical lab

• DNA sequencing and fusion testing of genes involved in epigenetic regulation (widely used in academic centers)

• H3-K27M immunohistochemistry (widely used)

• Histone posttranslational modification testing; e.g., K27me3 IHC³³ (widely used)

• CpG island methylation³² (less commonly used)

• Large-scale DNA-methylation-based classification³⁵ (currently used in select academic centers in Europe and the United States; becoming more common)

Box 2 |. Challenges in clinical trials in childhood cancer

• Relatively rare diseases make enrollment of a sufficient number of subjects difficult at any single institution and geographical location, requiring multi-institutional collaboration

• Randomized controlled trials are ideal, but given the small numbers of patients, this can be difficult; defining small but significant incremental improvements in outcome requires a prohibitively large number of patients

• Trial designs often do not take heterogeneous molecular characteristics into account; cutting-edge molecular diagnostics are increasingly required for appropriate classification and subtype stratification

• Detailed pharmacokinetic and pharmacodynamic assessments are often missing owing to small sample size (or not done as in blood-brain barrier penetrability assessment)

• Clinical trials in pediatrics are generally limited to age 3–18 years or 3–21 years; however, pharmacokinetics, tolerability, metabolism, immune system and microenvironment can differ dramatically within this age range

• Therapeutics are often developed for adult indications and re-purposed for pediatric diseases; therapeutics developed for the often unique pathophysiology of childhood cancers are in need

• Monitoring cognitive and other long-term sequelae of childhood cancer therapies through development and into adulthood are particularly important for childhood cancer clinical studies

Developmental origins of childhood cancers

Leukemias.

Evidence indicates that the cellular context in which oncogenic mutations occur are developmental-stage-specific. The first demonstration that the initial genetic hit for a childhood cancer can arise in a cell of fetal origin was in childhood B-ALL. This cancer of the B lymphoid line of blood cells is the most common cancer of childhood. Chimeric fusion genes (lysine lethyltransferase 2A (KMT2A, also known as mixed-lineage leukemia 1 (MLL)) fusions to variety of partner genes and TEL-AML1 fusions) were found in monozygotic twins with concordant leukemia, while additional chromosomal abnormalities were subclonal and distinct in each twin^37,38. Direct evidence for the prenatal origin of childhood leukemia was then provided by retrospectively examining clonal fusion gene sequences in archived neonatal blood spots—previously used for testing of metabolic disease and known as Guthrie cards—from individuals who later developed leukemia during childhood³⁹. At the time of diagnosis, the corresponding archived neonatal blood spots had the same genetic fusion years before leukemia emerged. This was demonstrated in subjects who developed childhood ALL and carried the chimeric transcription factor fusion gene ETV6-RUNX1 or the chimeric chromatin modifier fusion gene MLL-AF4, and in about half of patients who developed childhood AML with chimeric transcription factor fusion gene AML1-ETO^39–43. In an independent study of unselected cord blood of healthy children, the incidence of the ETV6-RUNX1 fusion was about 100-fold higher than the incidence of ETV6-RUNX1 fusion-positive B-ALL in children, suggesting that additional genetic events or a specific micro-environmental context are necessary to turn preleukemic cells into overt leukemia⁴⁴.

Indeed, when the ETV6-RUNX1 fusion gene was introduced into human cord blood cells and transplanted into immune-compromised mice, a pool of pre-leukemic stem cells emerged that self-renewed and survived⁴⁵, but this fusion gene alone was not sufficient to cause full leukemic transformation. All these findings have led to the development of a multihit model for ALL, in which gene fusions like ETV6-RUNX1 create a pre-leukemic pool when introduced in hematopoetic stem cells, but secondary cooperating events are required to develop overt leukemia. Recent work shows that this pre-leukemic cell pool can also be generated by introducing ETV6-RUNX1 into a developmentally restricted B cell progenitor unique to early embryonic life⁴⁶.

Another disease for which a prenatal cell of mutation has been demonstrated is transient myeloproliferative disease (TMD) and the related acute megakaryoblastic leukemia in patients with Down syndrome/trisomy 21 (Down syndrome-AMKL or Down syndrome-AML D7). TMD is a disorder of the megakaryocyte lineage and occurs in about 5–10% of neonates with Down syndrome/trisomy 21. In most children, TMD resolves spontaneously; however, about 20–30% of these affected individuals present a few years later with DS-AMKL, an acute leukemia of the megakaryocyte linage^47,48. Both of these diseases are thought to arise because of mutations in the hematopoetic transcription factor GATA binding protein-1 (GATA1) cooperating with as yet unknown genes on chromosome 21. One mouse study elegantly demonstrated that a mutated GATA1 allele leads to hyperproliferation of embryonic megakaryocyte precursor cells that first appear in the yolk sac and then move to the fetal liver. This hyperproliferation was only seen when the GATA1 mutation was introduced into prenatal cells, but the same mutation did not affect the proliferation of adult bone marrow megakaryocyte progenitors⁶. These findings might account for the restriction of both TMD and DS-AMKL to infants and young children, as mutations and cooperating overexpression of trisomy 21 genes might only be oncogenic during a short window of developmentally regulated transcriptional states.

The developmental origins of pediatric AML have also been described, especially for AMLs with fusions of histone methyl-transferase MLL. Initial work in mice has demonstrated that both self-renewing hematopoetic stem cells (HSCs) as well as short-lived myeloid progenitors could be successfully transduced with the highly conserved MLL fusion gene to result in rapid onset AML, supporting the evidence that cancer stem cells do not necessarily need to overlap with multipotent stem cells⁸. Other oncogenic fusions, like the MOZ-TIF2 fusion between the histone acetyltransferase MOZ and the nuclear receptor transcriptional activator TIF2, also induced overt leukemia when introduced into either HSCs or more committed progenitor cells in mice⁷.

The exact cell type in which a mutation occurs can influence the phenotype of the resulting malignancy. Gene expression analyses have demonstrated that leukemia stem cells generated from HSCs are more chemotherapy resistant and are more similar to poor-prognostic human AMLs as compared with leukemia stem cells generated from more committed progenitors, demonstrating that the exact cell of mutation can influence the phenotype of the resulting leukemia⁴⁹. This suggests that the transcriptional/epigenetic state of the cell in which the initial mutation occurs can impact the overall clinical behavior of the resulting leukemia⁴⁹, a principle that likely extends to other pediatric malignancies. Moreover, DNA methylation patterns differed in preclinical models of leukemias derived from HSCs versus committed progenitors and matched different human AML subtypes, again highlighting that the epigenetic state of the cell of mutation contributes to the final gene expression signature and phenotype of the resulting leukemia⁴⁹. Additional work has also demonstrated that the MLL gene fusion partner results in cell type–preferential effects on the resulting lymphoid and myeloid malignancies⁵⁰. This again suggests that different oncogenic hits require specific developmental and transcriptional cellular backgrounds to induce malignancies.

Another study extended this line of inquiry and demonstrated that HSC and myeloid progenitors can only be genetically transformed to leukemic blasts if the cells are able to transition to a more committed stage of differentiation after the introduction of the driver mutations in mouse models. If this differentiation step is genetically ablated, leukemogenic mutations no longer transform HSCs and myeloid progenitors. This elegantly demonstrates that the cells with the initial mutation might be either HSCs or myeloid progenitor cells, but full transformation appears to require progression to a more permissive, committed stage of differentiation, at least for the MLL-AF9 and MOZ-TIF2 fusion oncogenic programs⁵¹. Thus, the cell of tumorigenesis can be distinct from the cell in which the mutation occurs, i.e., the mutation occurs in a developmental state pre-tumorigenesis, a principle that appears to apply in many childhood cancers (Fig. 2). Like in ALL, the sequential order of mutations in defined cellular states appear to have an important role in AML: founder mutations in genes such as DNMT3A provide a clonal advantage to pre-leukemic HSCs, but need subsequent mutations, such as in NPM1, in downstream progenitor cells to drive progression to overt AML⁵².

Fig. 2 |. Developmental origins of childhood cancer.

a, Progression of human development from embryo to childhood. Oncogenic mutations may occur in the prenatal period, but cancer only emerges at a later developmental time point. This principle has been well described in childhood leukemias. b, Myeloid differentiation, from HSC, to common myeloid progenitor (CMP), granulocyte monocyte precursor (GMP) and differentiated myeloid cells. The cell of initial mutation for AML is an earlier stem cell, while the AML cell of origin is a more differentiated precursor cell. c, Oligodendrocyte maturation, from neural stem cell (NSC) to OPC and mature myelinating oligodendrocyte (OL). Childhood gliomas may arise from precursors in the oligodendroglial lineage (pre-OPC or OPC). While the timing and cell of mutation for each pediatric glioma subtype is not yet clear, some evidence suggests that the key oncogenic mutations may occur at an earlier developmental time point .

Malignant rhabdoid tumors

Malignant rhabdoid tumors (MRTs) are highly aggressive tumors of the kidneys, soft tissues and central nervous system (CNS) that occur in young children and are largely without mutations—genetically quiet—except for germline or somatic mutations of SNF5 (also known as Ini1/BAF47/Smarcb1)^53–55. SNF5 is part of the SWI-SNF chromatin remodeling complex and functions as a tumor suppressor in a broad range of developing tissues. While the homozygous deletion of SNF5 is embryonically lethal in mice, heterozygous deletions lead to early onset sarcomas, but not CNS tumors^56–58. Consequently, the generation of rhabdoid tumors in the developing brain (also termed atypical teratoid/rhabdoid tumors (AT/RT)) seems to require deletion of both copies of SNF5 in more specific cell types and developmental contexts: when both copies of Snf5 were ablated in the population of mouse neural progenitor cells that express nestin, all mice died prenatally. However, when Snf5 was ablated in the population of neural progenitor cells marked by expression of GFAP instead, mice developed neurodegeneration, migration defects and seizures. Only when Snf5 deletion was combined with TP53 loss in the former population of neural precursor cells did mice develop highly aggressive brain tumors displaying many histopathological and clinical features of AT/RTs in this mouse model⁹.

A subsequent study expanded these findings by conditionally inactivating SNF5 at various time points in mice from early embryonic stages of development into adulthood. The timing of the inactivation had dramatic effects on the type and location of resulting cancers. Intra- and extra-cranial sarcomas and lymphomas developed in a manner dependent on the developmental timing of SNF5 inactivation. Furthermore, only early embryonic SNF5 inactivation between E6 and E10 induced fully penetrant, mainly intracranial rhabdoid tumors that closely resembled all transcriptional subclasses of human AT/RTs¹⁰, whereas SNF5 inactivation during slightly later embryonal stages (embryonic day 12 (E12)) did not lead to tumor formation¹⁰. TP53 inactivation was not required in this model and is not prevalent in the human disease. These findings suggest that the SNF5 mutation is tumorigenic only during a rather limited window during early embryonal brain development. These findings modeling rhabdoid tumors were further confirmed in a recent study¹¹, in which SNF5 loss only induced rhabdoid tumor formation if deleted in neural crest cells during early embryogenesis (E9.5), but not when inactivated at slightly later time points on E12.5 or E13.5. As with the childhood leukemias discussed above, the mutational event and tumorigenesis appear to be temporally distinct in malignant rhabdoid tumors, with mutation occurring in the early prenatal stage and delayed tumor manifestation in the early postnatal period, at least as evident in mouse models. For further review of the literature on the prenatal origins of pediatric cancers, please see refs. ^59,60.

Gliomas.

High-grade gliomas (HGG) of childhood and adolescence exhibit a striking spatio-temporal and spatio-molecular pattern of incidence. As discussed above, midline gliomas of the thalamus, pons and spinal cord exhibit a highly recurrent and specific mutation in H3 genes and occur chiefly in younger children. In contrast, high-grade gliomas of the cerebral lobes occur in adolescents and young adults, and, intriguingly, a subset exhibit a different specific mutation in the same H3 gene^26,61. Broadly, these spatio-temporal patterns of gliomagenesis map onto developmental waves of myelination in the human central nervous system^62,63; for detailed review, see Gibson et al.⁶⁴. This correlation is concordant with the mounting evidence that high-grade gliomas of childhood may arise from precursor cells in the oligodendroglial lineage^20,21,65. While midline tumors emerge during this period of active midline structure myelin development, it is not yet fully clear when the mutation(s) occur, or in what cell type.

Like other childhood cancers discussed above, the cell of mutation and the developmental period in which that mutation occurs may be distinct from the cell of cancer origin. Indeed, recent evidence from genetic mouse models suggests that the classic genomic alterations associated with midline gliomas only induce tumors in the postnatal period if introduced during prenatal brain development⁵, and, while neural stem cells can be transformed by introduction of these mutations in vitro, the resultant tumors do not recapitulate the typical histological appearance of diffuse midline gliomas when transplanted into the mouse brain⁴. A precedent exists in the adult glioblastoma (GBM) literature to support this model of tumorigensis in which the mutation occurs in the neural stem cell resulting in cancer originating from the more restricted neural precursor cell. In one genetic mouse model, GBM-associated mutations introduced into a neural stem cell give rise to a tumor only when the stem cell differentiates into an oligodendrocyte precursor cell (OPC), the neural precursor cell that gives rise to myelinating oligodendrocytes⁶⁶. Further work suggests that GBMs can arise from both NSCs and OPCs, but the molecular subtype is determined by the cell of origin⁶⁷. To fully elucidate the cell of mutation and cell of origin for diffuse midline glioma, future studies in which the mutations are introduced that characterize diffuse midline gliomas at various developmental time points and in various cellular contexts, from neural stem cell to pre-OPC to oligodendrocyte precursor cell, will be required.

Another interesting spatiotemporal aspect of pediatric gliomas are non-brainstem HGGs arising in infants and young children less than 3 years old. A large percentage of these tumors have a lower mutational burden than other pediatric HGGs; however, these infant HGGs frequently harbor fusions of different genes with the kinase domains of neurotrophic tyrosine kinase receptors 1–3 (NTRK 1–3). These fusions, though not unique to infant HGGs, showed strong oncogenic activity in experimental settings¹². Underscoring their unique biology, the resulting infant tumors are associated with a significantly better prognosis than HGGs in older children⁶⁸.

Neither spatio-temporal and spatio-molecular patterns in pediatric gliomas are yet fully understood. Specificity to the midline is also observed in other childhood cancers. In germ cell tumors, midline location is thought to be due to the midline route of germ cell progenitors during embryogenesis, and in chordoma due to the midline location of the embryonic notochord. While the developmental origins of germ cell tumors and chordoma are well explained by the anatomical position of their developmental origins, no such explanation is yet apparent for midline gliomas. Insights into the heterogeneity of oligodendroglial and other neural precursor cells may shed some light on these patterns of gliomagenesis. Alternatively, the patterns may be explained by an as-of-yet undetermined intersection of region-specific micro-environmental influences with these cell-intrinsic vulnerabilities. Future work in both normal neurodevelopment and glioma pathobiology will be required to elucidate this enigmatic pattern of childhood glioma incidence.

The microenvironment of actively developing tissues

Microenvironmental influences critically regulate cancer growth, and this principle is applicable to the rich growth environment of actively developing tissues. Intercellular interactions governing tissue development and growth are subverted to promote malignancy in childhood cancers that arise from the very tissue stem/precursor cell populations these mechanisms are meant to support. A clear example of this principle is found in the influence of neuronal activity on the proliferation of both oligodendrocyte precursor cells and pediatric glioma cells. Neuronal activity promotes robust proliferation of OPCs and pre-OPCs⁶⁹ as part of an adaptive response of myelin-forming cells that influences myelin development^70–72 and represents a mechanism of experience-dependent neural plasticity. Myelin plasticity in the healthy nervous system has implications for neurological functions such as motor performance⁶⁹ and learning⁷³ (for review, see ref. ⁷⁴). In glial cancers, such as pediatric glioblastoma and H3-K27M-driven diffuse midline glioma, neuronal activity similarly promotes glioma proliferation, growth and progression through activity-regulated secretion of both known (brain-derived neurotrophic factor (BDNF)) and novel (neuroligin-3 (NLGN3)) growth factors^75,76. Glial cancers exhibit a striking growth dependency on one of these activity-regulated mechanisms of glioma growth—activity-regulated cleavage and shedding of neuroligin-3 (NLGN3) from synapses⁷⁶. Neuroligin-3 is a potent mitogen for glioma cells owing to stimulation of numerous oncogenic signaling pathways upon binding, but the reasons for this dramatic dependency⁷⁶ remain to be fully clarified.

Neuronal activity is a robust regulator of normal and neoplastic stem cell niche in numerous tissues, important for organogenesis⁷⁷, homeostasis, regeneration and malignancy (for in-depth discussion, see ref. ⁷⁸). Nervous system activity influences a number of adult cancers arising in diverse tissue types such as the brain⁷⁵ and also prostate, pancreas, colon, gastric and skin cancers^79–82. The extent to which innervation regulates non–central nervous system childhood cancers remains a largely unexplored and rich area for future study.

Therapeutic approach to childhood cancers

Among the challenges (see Box 2) in developing better therapies for childhood cancers are clinical trial execution for relatively rare diseases. Successful establishment of cooperative childhood cancer clinical trial consortia, such as the international Children’s Oncology Group (COG), the international Societe Internationale D’Oncologie Pediatrique (SIOP) and more specialized groups like the US-based Pediatric Brain Tumor Consortium (PBTC), have facilitated efficient trial execution, allowing for enrollment of sufficient subjects in a reasonable timeframe. Incentivizing legislation in the United States, such as the Creating Hope Act (21 U.S.C. 360ff) and the RACE for Children Act (21 US Code 355c), have increased interest from and collaboration with the pharmaceutical industry to advance new therapies for childhood cancers.

Distinct mechanisms of disease require dedicated therapeutic discovery effects.

Successes in treating childhood malignancies such as ALL with a combination of multiple chemotherapeutic drugs and Wilms tumor with a combination of surgery, chemotherapy and sometimes radiation therapy illustrate the power of combination approaches. Effective therapy for presently intractable childhood cancers will likely require a combination of targeting cell-intrinsic vulnerabilities, microenvironmental dependencies and tumor cell–enriched or tumor cell–specific antigens for immunotherapy. Appreciating the molecularly defined and functionally important subtypes of each disease will enable stratification of therapies to best target such vulnerabilities and dependencies. For each of these therapeutic approaches, childhood cancer targets may be non-overlapping with those in adult malignancies because of differences in the cellular and molecular context in which these cancers arise and the unique or differentially robust growth programs in the microenvironment of actively developing tissues.

There are many examples of differences in the response of childhood cancers that appear to histologically mirror their adult cancer counterparts to therapies that are effective in these counterparts. For example, temozolomide, an alkylating agent that has demonstrated survival benefit for adult high-grade gliomas⁸³, exhibits no clinical benefit for H3K27M⁺ diffuse midline gliomas of childhood^84–86. However, the pathophysiology of the H3-K27M oncohistone mutation creates disease-specific therapeutic opportunities. As discussed above, these cells have H3-K27M-induced H3-K27 hypomethylation and resulting transcriptional dysregulation, meaning that H3-K27 hypomethylation^87,88 or transcriptional dependencies^89,90 may be targeted for glioma therapy (Fig. 1). Alternatively, an immunotherapeutic approach may be taken in which aberrantly high levels of particular cell surface antigens can be targeted whose overexpression is driven by aberrant epigenetic regulation⁹¹.

Epigenetic approaches.

It is sometimes the case that drugs developed for adult indications can be repurposed for childhood cancers, but in these cases the mechanism of action may be disease-specific. For example, the histone deacetylase (HDAC) inhibitor panobinostat, a Food and Drug administration (FDA)-approved drug for the adult hematological malignancy multiple myeloma, exhibits preclinical activity against diffuse midline glioma cells with mutant H3-K27M because, at least in part, of the restoration of H3-K27 methylation and subsequent normalization of gene expression⁸⁷. This is unexpected as histone H3-K27 methylation represses gene expression, and histone deacetylation does also. The normalization of H3-K27 methylation is thought to occur possibly through steric hindrance of the interactions between the mutant methionine of H3-K27M-mutant histones and the PRC2 complex attributable to polyacetylation of nearby residues in the histone tail⁹², thereby preventing PRC2 dysfunction and enabling H3K27 methylation of wild-type histones in the cell⁸⁷. This mechanism of action, unique to cancer cells with the H3-K27M mutation, is serendipitously achieved by a drug developed for an adult indication. Panobinostat is presently in phase I clinical trials for DIPG (NCT02717455, NCT03566199, NCT03632317) (Fig. 1). As our knowledge of childhood cancer biology increases, therapeutic development efforts can be increasingly focused on the unique pathophysiologies of childhood cancers.

Targeting fusion gene products.

Early clinical trials have demonstrated a promising response to targeted inhibition of fusion gene products that involve dysregulation of a kinase. Larotrectinib, a kinase inhibitor targeting gene fusions involving NTRK1, NTRK2 or NTRK3, has demonstrated safety and efficacy in treating NTRK-fusion-positive childhood cancers in a phase I/II clinical trial (NCT02637687) and has recently received FDA approval for use in NTRK-fusion-positive malignancies, regardless of tissue type or patient age. Targeting BRAF fusions using MEK inhibitors appears similarly promising in pediatric clinical trials^93,94. More challenging to therapeutically target are oncogenic gene fusions involving transcription factors. For example, ~95% of Ewing’s sarcoma cases exhibit a fusion gene involving an RNA binding protein EWS and the ETS transcription factor FLI1 (ref. ⁹⁵). Recent advances in the preclinical literature have identified a promising compound to target the EWS-FLI1 fusion gene product⁹⁶, which may advance to the clinic in the future.

Immunotherapeutic approaches.

A shining example of successful development of a new and effective therapeutic targeting the unique biology of a childhood malignancy is the CD19-directed chimeric antigen receptor (CAR) T cell therapy, which has recently been approved by the FDA for refractory childhood leukemia⁹⁷. This potentially transformational therapy is the result of focused development efforts for childhood leukemias, which, unlike adult leukemias, express high levels of CD19, and represents only the fourth agent to gain FDA approval for a pediatric cancer indication in US history. This clinical success, together with the exponentially increasing understanding of childhood cancer-specific pathophysiology, will hopefully encourage and enable similar development efforts in the biotechnology industry.

Microenvironmental therapies.

As noted above, nervous system activity can exert a robust growth-promoting influence on pediatric gliomas and other forms of cancer. Blocking the cleavage and release of the synaptic adhesion protein neuroligin-3 (NLGN3) into the tumor microenvironment causes a profound growth inhibition in preclinical models of a range of glioma types⁷⁶, and this represents a promising new strategy to explore for therapy of these lethal childhood brain cancers. A clinical trial targeting NLGN3 release into the pediatric glioma microenvironment will be forthcoming. Similarly, rhabdomyosarcoma metastases in lung require receptive perivascular stromal cells and extracellular matrix factors for establishment⁹⁸, highlighting microenvironmental targets to limit the spread of this childhood sarcoma and other cancers. A deeper understanding of tumor-stromal cell interactions that enable and promote cancer progression will uncover additional microenvironmental targets for childhood cancer therapy. Parallel studies of microenvironmental interactions mediating normal tissue development and plasticity will elucidate potential toxicities of such strategies and potentially ways to overcome unintended effects on healthy tissue development and homeostasis.

The effects of cancer therapy in developing tissues.

Both traditional and targeted cancer therapies induce important, lasting toxicities in normal tissues, and this is particularly true when administered during periods of active development and growth. Many cancer therapeutic modalities, such as radiation and cytotoxic chemotherapy, are designed to kill proliferating cells and hence induce lasting damage to diverse tissue-specific stem and progenitor cell populations that are necessary for growth, regeneration and plasticity. It is not surprising that therapies targeting cancers that originate from and molecularly resemble normal precursor cell populations would also be affected by cancer therapies. The lasting effects of therapy on normal tissues are multifactorial and include both direct effects on tissue stem/progenitor cells and indirect effects via disruption of the normal tissue microenvironment and through influences on the endocrine system. In the musculoskeletal system, long-term effects of childhood exposure to cancer therapies on bone growth and muscle development can cause short stature, reduced muscle bulk and skeletal deformities. These musculoskeletal effects are mediated both by direct effects of radiation, chemotherapy and oncological surgeries on developing bone, muscle and connective tissues and by indirect effects on these tissues through treatment-induced disruption of the endocrine system (for review, see ref. ⁹⁹).

Neurological effects of childhood cancer therapy are particularly common and debilitating, and a syndrome characterized by impaired attention, concentration, memory function, multitasking and fine motor function affects a large fraction of children treated for childhood cancer^100–102. This cancer therapy-related cognitive impairment syndrome is particularly severe following brain radiation, but occurs after systemic chemotherapy exposure as well.

The cellular and molecular underpinnings of this lasting neurological dysfunction are beginning to come to light. Radiation and chemotherapy cause an acute depletion of neural stem and progenitor cell populations^103–105, cell populations broadly important for neural development, homeostasis, regeneration and plasticity. Following radiation of the hippocampus, a site of ongoing neurogenesis throughout life^106–108, the neural stem cell compartment regenerates to baseline levels of precursor cell numbers in the subacute period, but the long-term self-renewal capacity of neural precursor cells (NPCs) is diminished and NPC function is severely impaired owing to radiation-induced perturbation of the neurogenic microenvironment¹⁰⁹. Central to this dysfunction of the neurogenic microenvironment of the hippocampus is radiation-induced microglial inflammation (Fig. 3), which inhibits proper neuronal differentiation through inflammatory cytokine secretion^107,109,110. Microglial depletion using a small molecule CSF1R inhibitor rescues memory function in a mouse model of radiation-induced cognitive impairment^111,112. As microglial inflammation has recently been shown to induce a neurotoxic activation state of astrocytes¹¹³, intercellular interactions mediating neurological dysfunction after radiation exposure may be even more complex, as has been demonstrated after methotrexate chemotherapy¹¹⁴, discussed below (Fig. 3). Both microglia and astrocytes contribute to pruning synapses^115,116, and decreased dendritic spine density—potentially indicating exuberant synapse engulfment (Fig. 3)—has been observed following radiation¹¹¹ and is rescued by microglial depletion¹¹². Adding an additional layer of complexity, recent work indicates that bone marrow-derived monocytes contribute to repair mechanisms after radiation, so craniospinal radiation and chemotherapy-induced injury to hematological precursor cell populations and the bone marrow niche may further exacerbate radiation-induced neurological injury¹¹⁷.

Fig. 3 |. Microglial inflammation is central to childhood cancer-therapy-related cognitive impairment.

Childhood cancer therapies, such as cranial radiation^107,109,110, methotrexate¹¹⁴ and cyclophosphamide¹²², induce persistent microglial reactivity that in turn induces neurotoxic astrocyte reactivity¹¹⁴, disrupts oligodendroglial lineage dynamics and impairs myelination¹¹⁴, blocks neuronal differentiation of hippocampal stem cells and thereby disrupts hippocampal neurogenesis^107,109,110 and contributes to aberrantly exuberant pruning of synapses^111,123. Strategies to modulate microglia, such as CSF1R inhibition^111,112,114, represent an important therapeutic opportunity for cancer-therapy-related cognitive impairment.

Neural precursor cells are exquisitely sensitive to multiple forms of chemotherapy, and damage to precursor cells in the oligodendroglial lineage may be particularly severe^105,118. Indeed, children suffering from chemotherapy-induced neurological dysfunction exhibit evidence of subtle white matter injury on diffusion tensor magnetic resonance imaging^119,120. Oligodendroglial precursor cell populations normally exhibit robust regenerative capacity¹²¹, but recent work has demonstrated lasting depletion of white matter oligodendrocyte precursor cells long after chemotherapy exposure in both humans and a mouse model of methotrexate-chemotherapy-related cognitive impairment¹¹⁴. Following methotrexate, the oligodendroglial lineage is dysregulated, with depletion of both precursor cells and mature, myelinating oligodendrocytes and accumulation of oligodendroglial cells in an intermediate state of differentiation. Like radiation-induced neural precursor cell dysregulation, perturbation of the brain microenvironment after remote methotrexate chemotherapy underlies this persistent dysregulation of the oligodendroglial lineage¹¹⁴. Direct activation of white matter microglia by methotrexate is central to this pathophysiology (Fig. 3) and microglial depletion immediately following methotrexate exposure using a systemically delivered small-molecule CSF1R inhibitor normalizes neurotoxic astrocyte reactivity, oligodendroglial lineage dynamics and myelination and rescues cognitive behavioral impairment following methotrexate chemotherapy¹¹⁴. Future work is required to define the therapeutic window for microglial depletion strategies, the state of repopulating microglia following such depletion strategies and complementary or alternative therapeutic strategies. In addition, the similarities and differences in microglial reactivity following exposure to different chemotherapeutic and immunotherapeutic agents remain to be explored. Microglial activation has been reported in response to other chemotherapeutic agents, including cyclophosphamide¹²², but was not observed following 5-FU exposure¹¹⁸. Our understanding of the long-term impact of cancer therapy exposure on the gliogenic and neurogenic microenvironment and its lasting influence on precursor cell population dynamics is presently incomplete and requires further study. However, an emerging concept from numerous studies in diverse models indicates that therapy-associated microglial reactivity is central to therapy-induced neuropathophysiology (Fig. 3) and may represent an important target for strategies to mitigate or even prevent cancer-therapy-related cognitive impairment.

Conclusion and future directions

In summary, there are mechanistic themes unique to childhood cancer, regardless of the tissue of origin, which warrant distinct therapeutic approaches compared to adult cancer. The genomic landscape of childhood cancer is different, as mutational burden is lower and structural variants are less common in childhood cancer than in adult cancer. Additionally, driver point mutations most frequently occur in epigenetic modifiers, and, together with oncogenic fusion events, tend to be specific to individual cancer types (and sometimes even specific to anatomical locations). Epigenetic alterations not only appear to play an important role in terms of frequently found mutations but might also reflect a locked-in epigenetic state during development, which contributes to the final gene expression and phenotype of the resulting cancers. Moreover, the developmental contexts in which mutations occur and the exact timing of these mutations appear fundamentally important: the often-observed time delay between first mutations and tumor emergence, with certain initial events even occurring prenatally, points toward an important difference between the ‘cell of mutation’ and the ‘cell of origin.’ Inherent in this concept is the need to better understand the microenvironmental factors influencing cell state during development. Future efforts toward mechanistic understanding and ultimately therapy for childhood cancers will therefore need to consider various pre- and postnatal developmental stages and their specific microenvironmental and transcriptional states. Parallel efforts to understand the complex mechanisms through which childhood cancer therapies disrupt normal tissue development, homeostasis and plasticity will elucidate strategies to restore and prevent long-term sequelae of childhood cancer therapy.