Ever since the identification1,2 in 1982 of mutated human genes that drive cancer, a massive body of evidence has established the paradigm that cancer results from mutations in specific genes. The recent wave of genome sequencing of human tumours has confirmed this concept, but it has also identified some tumours with surprisingly few cancer-associated gene mutations. The results presented in this issue further urge us to revisit the role of gene mutations in cancer. In their genomic analyses of three subtypes of ependymoma brain tumour, Parker et al.3 (page 451) and Mack et al.4 (page 445) find that one subtype carries an intrachromosomal translocation that creates a new tumour-driving gene, another lacks tumour-driving mutations but has aberrant epigenetic modifications, and a third shows neither gene mutations nor epigenetic aberrations.

Ependymomas are tumours of the central nervous system that originate from the wall of the ventricular system along the entire cranio-spinal axis (Fig. 1). They can be treated by surgery and radiation, but chemotherapy is ineffective, and survival rates for patients with these cancers have not appreciably improved over the past decade. Locality and molecular data define the four subtypes of ependymoma — supratentorial, posterior fossa type A, posterior fossa type B and spinal-cord ependymoma — which also differ in age of onset and prognosis5.

Figure 1: Genomic characterization of ependymomas.
figure 1

Parker et al.3 and Mack et al.4 performed whole-genome analysis of three subtypes of ependymoma, which differ in their location in the central nervous system. The age of onset, patient prognosis and previously known molecular characteristics5 of the tumour types are listed in black; new molecular findings from the sequence analyses are listed in red.

Full size image

Parker et al. applied whole-genome sequencing to supratentorial and posterior fossa ependymoma cells. Among supratentorial tumours, they found frequent cases in which translocation of a region of chromosome 11 caused the fusion of two genes, RELA and C11orf95. These translocations resulted from chromothripsis — a mechanism that pulverizes a genomic region and randomly stitches the fragments together. RELA is a transcription factor in the NF-kB signalling pathway, which regulates many cellular processes. Whereas wild-type RELA is located in the cytoplasm unless it is activated, the C11orf95–RELA hybrid protein spontaneously translocates to the nucleus, where it activates the expression of target genes. The authors also show that transplantation of neural stem cells expressing this fusion gene into mouse brains induces ependymoma-like tumours, thus adding a new tumour-driving gene to the list of those associated with cancer.

But equally remarkable is what Parker et al. did not find: their sequencing of posterior fossa tumours did not reveal any recurrently mutated gene or translocation. Mack et al. also sequenced posterior fossa ependymomas, with the same puzzling result. Recent sequencing6 of many adult tumour types has typically detected 25 to 200 amino-acid-changing mutations per tumour; most of these affect 'bystander' genes, but cancer-driving gene mutations expose themselves by their recurrence in multiple tumours. However, there are certain tumour types in which very few recurrently mutated genes — often only one or even none per tumour — have been detected, namely in childhood tumours such as medulloblastoma7, neuroblastoma8 and rhabdoid tumours9. All current sequencing technologies sometimes miss mutations6, but the fact that this paucity of mutations was observed with diverse technologies, and only in childhood tumours, makes it unlikely that it results from technical artefacts. Now, these two papers report that posterior fossa ependymomas also seemingly lack recurrently mutated genes.

If not gene mutations, what else could cause cancer? It has long been suspected that defective epigenetic modifications — that is, non-sequence-changing alterations, such as the methylation or acetylation of DNA or DNA-associated chromatin proteins — might also be oncogenic. Several genes encoding enzymes that apply or remove these modifications have been shown to be mutated in tumours, confirming a role for epigenetics in cancer10, but such mutations were not detected in the ependymoma studies. However, Mack et al. did find increased DNA methylation of specific genes, as well as silencing of their expression, in type A, but not type B, posterior fossa ependymomas.

Expression of these same genes was previously found to be silenced in embryonic stem cells by the protein complex PRC2 (ref. 11), which mediates a common epigenetic modification: trimethylation of the amino acid lysine at position 27 in the histone protein H3 (H3K27me3). Indeed, Mack and colleagues found H3K27me3 marks on many of the genes with DNA methylation in posterior fossa type A tumours. They therefore hypothesize that the repression of these genes by PRC2 keeps these tumour cells in an embryonic and proliferative state.

A lack of model systems for experimental studies of posterior fossa ependymoma did not allow crucial testing of this hypothesis, and these data remain correlative, but preliminary testing of drugs targeting DNA methylation and H3K27me3 inhibited the proliferation of type A tumour cells in vitro. The key question emerging from these findings is whether and how a cell can escape the normal regulatory mechanisms governing epigenetic modifications such that oncogenic gene-expression patterns can persist, without having DNA-sequence mutations.

What remains are posterior fossa type B ependymomas, for which neither tumour-driving gene mutations nor epigenetic changes have yet been found. Both groups of authors refrain from any interpretation of this. Type B tumour cells differ from type A tumour cells by typically containing gains and losses of entire chromosomes or large chromosomal fragments5. Chromosomal deletions occur frequently in almost all tumour types and are usually interpreted as part of a two-hit inactivation of a tumour-suppressor gene, in which one copy of the gene is destroyed by a mutation and the other copy by a deletion. However, genome-sequencing data have shattered the expectation that deletions will always involve a tumour-suppressor gene. Neuroblastoma tumours, for example, frequently show deletions of regions of chromosomes 1 and 11, but no recurrently mutated tumour-suppressor genes have been found on the corresponding section of the paired chromosome8,12,13.

Thus, it may be that some such deletions have an entirely different role in cancer. The deleted areas usually encompass hundreds of genes, and changes in the expression of large numbers of genes can be highly pathogenic. For example, one extra copy of chromosome 21 causes Down's syndrome, and third copies of any other chromosome are mostly lethal. Speculatively, cancer initiation could occur when a deleted region encompasses several inhibitory genes of a particular cell-signalling pathway, and a gained region contains several positive regulators of that pathway. The resulting modest changes in expression of each individual gene could together exponentially activate the pathway, and may drive cancer.

It will be challenging to test whether chromosomal gains and losses, or epigenetic modifications without gene mutations, can indeed drive cancer development. The clinical implications of such alternative oncogenic routes would, however, be far reaching. Much research focuses on drugs that target gene mutations. The C11orf95–RELA fusion protein identified by Parker and colleagues provides a new target for drugs against supratentorial ependymoma, but the treatment of posterior fossa tumours might require a fundamentally different approach.