My views on a saltationist theory of cancer evolution

One easy way to spot who reviewed a paper is to observe who is writing a News and Views afterwards.

So for example, Nick Navin just published a paper in Nature Genetics describing “Punctuated copy number evolution and clonal stasis in triple-negative breast cancer” and, looky-look, someone wrote a News and Views about it.

And given that only a single reviewer hectored them about the details of a 40 year old paper on punctuated equilibria (in the last round of revision! when everyone else said It’s fine!), the authors might now even have reasons to suspect who was evil reviewer 3.

Apparently, this link let’s you see the News and Views without a subscription.

I discussed Nick Navin’s paper with several people while writing this piece, but Nat Gen is not allowing acknowledgments in News and Views. So let me just use this space here to acknowledge input by Peter van Loo, Andrea Sottoriva, Geoff Macintyre, Edith Ross and the rest of my team.

They all had very interesting things to say, but in the end I had to condense all of it, including my own ideas, into 900 words (well, more like 1000 in this case).

A recurrent piece of feedback on early drafts was that no-one knew what ‘saltationist’ meant. Well, I always answered smugly, let Wiki be your friend. (Because that’s where I look up smart and long words.)

One idea I am touching on in the article, but don’t discuss in great depth, is the fact that classical evolutionary theories (like crisis and stasis in Eldredge and Gould’s theory of punctuated equilibria) is about phenotypes, while in Gao et al (and previous papers on chromothripsis and chromoplexy) these concepts are applied to genomes.

I do understand how a phenotype can be stable over time, even though there are ongoing mutations (either neutral or with very small fitness effect), but I find it much harder to understand how the genomes of the clones can be stable – who tells the genomes to stop getting messed up? It’s cancer after all … genome instability is one of the things it’s really good at.

Any ideas anyone?




6 thoughts on “My views on a saltationist theory of cancer evolution

  1. Hi Florian, I also read the paper and was disappointed about how some concepts are rapidly used. I already mentioned briefly my thoughts about punctuated evolution in cancer ( Enough to say Even Gould later acknowledged that punctuated and gradual evolution are not exclusive ideas. Indeed, not even Darwin knew that different species would not evolve under a constant, universal rate. Take a look at the Wikipedia entry for punctuated evolution ( It is very informative

    In any case, whatever name we use to describe the observed pattern, the important point would be the timing of the driver mutations. Do they all occur at the very beginning as Navin suggests? Well, the trees on that paper have not time in them. These trees really look to me like recent selective sweeps, and the fact that in their “gradual” simulations could not repeat this pattern does not mean much to me, as this is a question of how big are the mutations simulated (i.e., how much increase in fitness they provide). And moreover, these trees do not look at all compatible with the big bang model, which implies multiple old clones, as they claim. I am not convinced by the conclusions of this paper, but this does not mean that variation in mutation rates do not actually happen. It is easy to imagine that in cancer driver mutation rates would vary in time and space in a complex manner, and in a different way in distinct occasions.

    Best, D


    1. Thanks, David.

      I believe the history of these names and concepts is a very useful context to current debates. The wikipedia page on Punctuated evolution is indeed very good. The section on Saltationism motivated the title of my piece. Nick Navin called his model ‘punctuated evolution’ to set it apart from ‘punctuated equilibria’ (because evolution is a different word than equilibrium), but that was a bit too subtle for me.

      PE vs Big-Bang:
      You say something very interesting about the trees in the simulation. How would a Big Bang tree look different from the ones they show? To me, Nick Navin’s description of the model sounded very much like the Big Bang. The only difference I could really see is that Nick Navin emphasises stasis (a lot!) whereas the Big Bang folks like Andrea Sottoriva would always assume that there is ongoing (mostly neutral) evolution in each clone. But if you actually look at Figure S1 in Gao et al you will see that their views might not be that different (whatever the main text says) because the picture shows an ongoing accumulation throughout stasis.

      Mutation rates:
      How would you measure variations in mutations rates? The bulk-sequencing people I hang out with use the number of mutations appearing in a clone (= cluster in their VAF or CCF plots) as a measure of age, which I think means that mutation rates are assumed constant (over time and between clones).



  2. oops, meant to write that Darwin indeed did know (and said) that different species would not evolve under a constant, universal rate. And yes, saltationism sounds better, still I bet the observed topologies in that paper can obtained with multiple selective sweeps –let’s see if we find the time to prove this 😉


      1. Hi Florian,

        Yes, theories and concepts are indeed useful (how could it be otherwise) but when people is well informed and knows what is talking about. But I understand it is tempting to have “new models” named after us, even if they represent common patterns seen multiple times in multiple situations in different contexts.

        But the important point, how punctuated equilibrium (PE) trees are expected to look compared to Big-Bang (BB) trees, *when the data is single tumor cells*? Well, first we need to compare these two models (at least in the way I understand them). They can overlap in some parts, but they focus on different aspects of the evolutionary process. The key for PE is whether mutations accumulate continuously or not, while BB’s main idea is neutral evolution. The PE model does not imply neutrality or lack of clonal competition (in fact it is compatible with selection), but the BB does. The PE model implies one dominant big mutant (the hopeful monster if you want; or few similar dominant clones) that increase its frequency through selection and that then almost stop evolving. In any case, many clonal mutations. On the other hand, the BB model implies after transformation an radiation of multiple clones with many private mutations that increase their frequency by neutral drift.

        So *in general* PE trees will have an extremely long internal branch (branch length in phylogenetics means amount of change, i.e., time x substitution rate) that separates the healthy cells from the sampled tumor cells (what some people calls the “trunk”…while indeed this is not the trunk of the tree…). Then, the branches emerging from the tumor most recent common ancestor or MRCA will be short. That is, the sampled tumor cells will share many changes and will be quite similar. That is, few tumor clones, *very alike to each other*, that represent minor variations during the stasis period of the initial big mutant. Like the trees in Figure 6 in Gao et al. Indeed, gradual evolution coupled with *hard* selective sweeps would look exactly in the same way. My take is that Gao gradual simulations probably did not implicitly allowed for strong sweeps to take place.

        On the other hand, BB trees should have a comparatively smaller branch separating healthy cells from the sampled tumor cells (still reasonably long as we need to have some initial driver mutations) and then multiple long internal branches arising from the tumor MRCA (i.e., star-like trees; a well-known pattern typical of a growing population). That is, we should see multiple tumor clones that are *quite different among themselves*. We see this in several trees in Figure S2 in Sottoriva et al BB paper.

        Importantly, in both PE and BB situations the way sample (for example single vs multiregion) and when we sample, the actual population structure, errors in phylogenetic reconstruction, the type of data (e.g., CNV vs SNVs) and/or the power to detect NGS variants at low frequencies can distort to a different extent what we actually observe …

        Regarding your second question, how would I measure variations in mutations rates? With single cell data we could use, after some modifications, available relaxed clock phylogenetic models (see; or this, for fun, Obviously if we had longitudinal samples then out power to detect this changes in rate would be maximized.

        Uf, what a rant, hope I was clear and did not miss many things … 😉



  3. Thanks!
    I agree with you David. When I was introduced to this paper, I thought it could be a result of recent sweep too. And I happen to have the evidence from my simulations, which is part of a modelling paper I am working on. I am also concerned about the lack of longitudinal data in current cancer studies, which may lead to many misleading/false interpretations.


You gotta talk to me!

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s