For some time before we had the benefit of cancer genomics, it was generally believed that for a cancer to disseminate and become potentially lethal, it would have had to accrue several mutations that, collectively, would provide a kind of ‘full house’ for malignancy.
It was further assumed that, in the absence of rampant genetic instability, the critical set of mutations would arise one at a time and that it would, therefore, take time to assemble a ‘full house’ set. The linear relationship of cancer incidence to age (in log-log plots) was taken to indicate that four to six rate-limiting mutational events might be involved 1,2. However this inference rested on some questionable biological assumptions 3.
Thanks to genome sequencing, we now have comprehensive audits of the mutational changes in cancer cells and these come in all possible flavours – fusion genes, copy number or ploidy changes, and sequence variants 4-6. What is particularly striking is the huge range of the numbers of mutations in different patients, from zero to tens of thousands, depending upon the tumour type.
The very high numbers of mutations in some cancers may reflect a sea of genetic noise engendered by genotoxic exposure, such as cigarette carcinogens, or UVB. They could result from a persisting acquired genetic instability in cells. Or when catastrophic genomic events occur, for example, chromothripsis, many thousands of rearrangements can result from the fragmentation of a chromosomal region, followed by imperfect DNA repair.
The number of functionally relevant – or ‘driver’ mutations, within this mosaic will be difficult to determine. It is likely to be modest, but still variable from cancer to cancer. It is probably reasonable to assume that critical or rate-limiting ‘driver’ mutations are those that impact on the key 6-8 ‘hallmark’ phenotypic features of cancer 8.
The number of mutations required, or that are both essential and sufficient for malignancy, is further complicated by the fact the cancer clones evolve via non-linear, branching evolutionary trajectories 9-11 in which side branches have reiterated mutations 9,10 and independent malignant potential 9. In effect the patient has several cancers, albeit linked by common descent.
If the patient lives long enough, progression of their disease, will be associated with additional mutations. The number of ‘driver’ mutations ‘required’ is therefore likely to be significantly less than the number that exist in any case.
So, based upon the result of recent genome-wide sequencing 3, let’s hazard a guess at the minimal numbers required.
|Cancer type||Number of mutations||Latency|
|Rare tumours in infants||0 or 1||1 year|
|Most leukaemias and paediatric cancers||2||2 – 5 years|
|Most adult epithelial cancers||5 – 10||1 – 2 decades|
Any other offers?
Well, there are some cancers that appear to need fewer mutations to be able to evolve to disseminated malignancy in a relatively shorter time frame, which is perhaps is to be expected. But why should the mutational requirements be so different in the various tumour types?
For example, in epithelial carcinomas, a relatively simple but plausible explanation for the need to acquire multiple mutations is as follows. The very high turnover of cells (-1111 day in the intestinal epithelia) up to and throughout the reproductively active period of life has required multiple microenvironmental adaptations that restrain inherent potential for clonal escape and malignancy 12.
These adaptations would include constraining the architectural organisation of epithelia – cell symmetry and stem cell localisation, adhesive contacts and physical barriers, space and nutrient limitations and limits of 02 diffusion. Several of these features, encoded in ‘tumour suppressor’ genes, will have arisen early on in the evolution of multi-cellularity, some 600 million years ago 13 as essential restraints on opportunistic cheating by replicating ‘selfish’ cells 14,15. To break free in such a stringently controlled habitat requires that several hurdles be jumped over. And maybe that takes of the order of five (or a few more) independent mutations – and a long time?
But then why should the leukaemias and paediatric cancers appear to require fewer mutations and less time?
A striking feature of the acute leukaemias and most paediatric cancers is the imposition of differentiation arrest in cancer cells. One perspective on these cancers, that sets them apart from common adult carcinomas, has been that they reflect developmental abnormalities in which continued proliferation, coupled with a differentiation block that holds the lineage in an immature state, is the essential ‘hallmark’ feature 16.
The ‘target’ progenitor or stem cells for these cancers in early development may also be inherently migratory. Modest proliferative activity of cells ‘trapped’ in the normally transitory precursor cellular state would enable clonal expansion, perhaps in the absence of hypoxic stress, to progress relatively quickly to the point of overt malignancy with only modest mutational complexity.
More than 40 years ago, and long before cancer genomes were visible at the DNA sequence level, Alfred Knudson used the age-associated incidence curve to predict a minimal ‘two hit’ mutational hypothesis for retinoblastoma 3. He could well be right.
An additional difference between the paediatric cancers and adult carcinomas could lay in their aetiology. If adult cancers but not those of childhood commonly involve genotoxic exposures, this would tend to select for the additional necessary mutations – in p53 for example, in DNA repair or genetic instability, that facilitate cell survival in the face of genotoxic assault.
An interesting potential corollary of this distinction between cancer types would be the intrinsic chemo-sensitivity (and radiation sensitivity) and curability of otherwise lethal cancers with ‘quiet’ genomes – childhood B cell acute lymphoblastic leukaemia in particular 17,18. Some adult cancers could fall into this group of ‘developmental’ abnormalities with intrinsic migratory or invasive activity and inherent sensitivity to chemotherapy – choriocarcinoma and testicular cancer are obvious candidates. We can predict that they too will have minimally deviant genomes.
Philadelphia chromosome-positive chronic myeloid leukaemia (CML) is almost certainly driven by a single lesion – the BCR-ABL1 fusion gene encoding active ABL kinase. But I don’t count this as a ‘one-hit’ only cancer because CML is essentially a pre-malignant condition; additional genetic errors are involved when the disease evolves into its acute phase.
There are however rare paediatric cancers that are malignant but appear to have just one or even no recurrent somatic mutations 19. Inherited gene variants could be playing a role, as could stable epigenetic alterations. However, it is interesting that the single mutations that are described, encode global chromatin modifiers or pivotal signalling molecules. Their impact could be pleiotypic, a single mutation impacting on several genes or pathways – a joker in the pack.
So, how many mutations does it take? The answer appears to be that the number varies according to the cell types, developmental staging and mutations involved. There are many games to play.