Is cancer just bad luck? Part II. It's a genetic, but usually unpredictable, disease

Yesterday, we discussed some history of research on the cause and predictability of cancer.  Today, we'll try to raise some questions that seem to have been overlooked in the recent Tomasetti and Vogelstein paper in Science that argues that much or most cancer, with a few notable and clear exceptions, does not arise from inherited genetic mutations, nor from lifestyle exposures, but arises just by bad luck during the countless cell divisions that occur during our lives.  Much reaction to the paper has overlooked these issues as well.

In the usual use of the term, cancer is not genetic because there are only a few types of cancer that are clearly due to inherited variations in known individual genes. Even these are usually only a subset of all instances of cancer of the particular organ in question.  Most breast cancer does not involve inherited variation in the BRCA1 or BRCA2 genes, for example.

At the same time, some cancers, most notably breast but also colorectal and some other cancers, show family correlations of risk, suggesting that multiple contributing inherited variants might be involved. By far the bulk of cancers are 'sporadic' in the sense that they arise without detectable genetic risk factors.  Even large-scale GWAS type studies find very few genome sites that contribute more than individually very small, barely detectable, risk.

Before the frenetic genome mapping era began around 20 years ago, it seemed clear that with few exceptions (those perhaps mainly due to viruses) cancer was the archetype of a lifestyle-related disease.  Smoking caused a very clear risk of lung cancer.  Some viral exposures caused cancers. Colorectal cancers were largely due to low-roughage western diets, and various things like hormone drugs, coffee, and you-name-it, were suspects.  In addition, we knew clearly that ionizing radiation such as in x-rays and in uranium miners caused cancer risk.

The genome-wielders largely took over, of course, but that was as much a sociopolitical coup as it was based on any serious level science.  DNA was fashionable, sequencers were fancy (and expensive), and we could search the whole genome to find the culprit variants.  This turned out largely to be a big low-payoff bust, though not all geneticists are candid enough to admit it. Still, to many, with the few known exceptions, cancer has been seen as not a genetic disease.

But it's 'genetic' nonetheless!
This may all be true--it certainly is so empirically.  It gives the impression cancer is not really a genetic disease, in the usual sense of the word, meaning due to inherited risk.  But another sense of the word refers to mechanism, and cancer generally does seem clearly to be genetic in that sense.  It's just that the source of the variation is among cells within the body rather than among people (really, conceptuses) in a population.  Or, more properly it's a mix.  In fact if it were really genetic in the inherited sense the fetus would not develop properly, so one should never expect a really deterministic variant to 'cause' cancer by itself.  In this sense, cancer really is, if anything, the archetype of a genetic disease.  Here's why.

Diseases all must arise in some way or other in the behavior of cells.  Usually, it will be some collective aspect of cells, say, the pancreas's cells, as a whole, just don't make enough insulin,  Or by the way diets and other factors affect them, the blood stream produces too much of the wrong kinds of fats and they clog arteries.

But cancer is a disease of a single cell that then goes awry, and its cellular descendants.  The reason is that its genes are not responding in the usually self-restrained way for their local tissue environment. The genes could be induced by viral insertions, or by somatic mutations (that is, mutations occurring in body cells but that were not in the sequences inherited by the individual at his/her conception). The mutations cause the cells to divide without the usual orderly constraints.

Somatic mutations are not in the germ line and are not transmitted from parents to offspring.  They don't generate family risk correlations.  They can't be found by GWAS or other studies based on sequencing inherited genomes.  But they are genetic changes nonetheless, and many studies have shown that tumor cells do share mutational changes not found in normal tissue from the same person, and that as a tumor grows, spreads, develops drug resistance the cells in different descendant parts of the cancer have acquired even further mutational changes.

So that most cancer is not predictable from inherited genotypes is a disappointment, at least for genetic epidemiologists, it's a genetic disease nonetheless.  It's just hard or impossible to detect individual cells with a combination of the 'wrong' changes so as to found a tumor lineage.

At the same time, there is no reason to doubt that countless genetic variants that are inherited can affect risk, and make a cell more vulnerable to transformative somatic mutations.  It's just that, as GWAS types of research shows, the majority of these have individually very small effects--that's because they only have an effect when some other unlucky mutation(s) happen to arise in the same cell during the person's life.  But there can be uncountedly many such heritable weak-effect genome-types that simply can't be found by the current mapping techniques, and that's why such techniques don't find them.

And, yes, it's 'environmental'

Yesterday, we started this series stimulated by the Tomasetti and Vogelstein paper, in which they related the number of dividing cells in a person and the risk and age of onset of cancers of that organ.  They showed statistically that with the few known exceptions such as smoking and lung cancer, that cancer rates correlated pretty well with these considerations.  Since we ourselves were working with cancer site-specific and worldwide age-patterns of cancer, and formulating somatic-mutational models in those pre-genetic days, these ideas were already rather well-established, so the new paper uses newer data and seems very good and apt, but the idea isn't as new as the headlines and attention made it seem.  If anything, the profession at large should never have got to the point of expecting better tumor predictability than was at hand.

Still, environmental risk factors are not ruled out by that analysis.  Environmental or life-history risk factors, like diet or reproductive history and so on, stimulate cell divisions and in that way can affect the risk of mutations arising in the way Tomasetti and Vogelstein suggested: simply the normal errors in DNA copying.  Since the exposure has to affect a cell in a given tissue and in a particular relevant gene being used by that tissue, it is no surprise that the exposure's net effect, and hence predictability, is usually very small.  Still, exposure to environmental agents must contribute to mutations if the agent is known to be mutagenic or to stimulate cell-division.  So epidemiologists may be right that mutagenic or mitogenic exposures can have carcinogenic effect, but Tomasetti and Vogelstein are right that this will be essentially undetectable.  In no way does their analysis relate to the carcinogenic effect per se, just to the net magnitude.  Indeed, we know that such predictions, except relating to a few risk factors like smoking and UV light and HPV virus, haven't proven to be very powerful or reliable.  So there's nothing new here, except to the extent that genetic or environmental epidemiologists are in denial.

But actually, there are very clear environmental factors related to cancer risk.  They have to do with the subtle concept of competing causes.  If mutations arising by chance during cell division ultimately lead to transforming genotypes in some cell, the longer one lives the more likely such changes are likely to arise in at least one such cell in the person.  This is generally why most cancer rates rise with age in ways correlated with rates of cell division.

So, if we were to obtain wonderful preventive measures to eliminate heart disease and stroke, cancer rates would go dramatically up!  That is simply because those who now no longer died from the former would be alive to await the latter.  That is environmental causation, even if indirect!  Likewise, if we really want to reduce the risk of cancer, all we need do is keep eating McBurgers in greater and greater amounts, start some wars, or continue to over-use antibiotics: then we'll all die off of other causes, before we're old enough to get cancer.

Among many things that were said, unaccredited now, by many people including myself, because of the somatic mutational nature of cancer, if you were to live long enough you would get cancer of every organ you have.

Yes, luck is involved!
Indeed, even in inheritors of risk alleles, it need not be that if they get the cancer involved that their case is due to that allele--this is obvious in the sense that the same tumor can arise in people without the allele, usually the vast majority of cases.  So other factors are involved, and the natural occurrence of mutations in cell division, as well as environmental mutagenic or promoter agents doesn't change the fact that which exposed person has the wrong mutations in the vulnerable cells is simply a matter of luck.  An environmental mutagen has to hit the wrong set of genes in the wrong cell.  Naturally and fortunately, the odds are against such bad luck.

Tomasetti and Vogelstein essentially are saying that only the internal luck of mis-copying by DNA causes cancer. But environmental factors contribute to those errors, even if any individual exposure has very weak effects relative to a given type of cancer.  Relative to all cancers, it's harder to say, because through most of history few have lived long enough for there to be the kind of data needed, and since the risk per cell per cell division is small, and cell division generally slows with age, the newer evidence in an aging population will be statistically weak; cancer rates taper off, cancers grow more slowly, and the elderly have more urgent problems to deal with, as a rule.

But even if these findings are true but not revolutionary, not so fast!
The idea that risks per at-risk cell per cell-division that Tomasetti and Vogelstein based their analysis on makes sense, even if it's something that was essentially known decades ago.  We ourselves built multi-hit mutation-accumulation models that seemed to provide reasonably good fits to the known age-onset patterns of specific cancers.  These were based on somatic mutations.  But the T and V paper's analysis actually raises some issues that suggest maybe the authors have given too 'pat' of an explanation.

Even in the mid-20th century it was known that different species of animal also got a similar array of cancers, but that their accelerating age-specific risks, in principle related to the relative number of cells, were correlated with the species' typical lifespan.  And this had little if anything to do with environmental exposures, since the animals involved were typically those we managed or that had rather uniform environments.  This is not a trivial observation!

For example, inbred animals tell the tale as to tissues with a particular life-history of mitosis.  Mice housed in essentially identical conditions, develop an array of tumors at age-specific rates.  But mice get them in months, while we get them in decades.  This problem was raised around 1970 by prominent epidemiologist Richard Peto, but seems to have basically just been (conveniently) ignored. There are also strain-specific cancer risks in mice and other animals (including dogs and cats) that suggest that inherited vulnerability genotypes may be involved, but not single-gene variants.  If the number of cells at risk, or their division rates, are responsible for the just-bad-luck theory, then tiny mice should never get cancer!  And elephants or cows should be dropping over with huge tumors very early in life.

This raises another interesting issue about theory vs data in understanding cancer.  Among the transformative ideas in the late 1900s was that cancer is a 'multistage' disorder, that arises only after several events have occurred in some unlucky cell lineage in the body (or are inherited).  Early results suggested that only 2 events might be responsible.  A number of biostatistical epidemiologists began fitting, or I'd say 'forcing', 2 or 3-stage models to the data.  That is, they had their a specific theory, based on the fragmentary evidence then available, and fit the data to it, to estimate, for example, the rates at which the events occurred.  Then they had to explain what those events were, say, a cell-division inducer and a mutation.  But there was very little substantial evidence that that was the general story of cancer, and the evidence was far weaker than the commitment to the model.

Ranajit Chakraborty and I took a different approach.  We applied a more open multiple hit model and let the data speak for themselves; that is, we estimated (rather than pre-specified) the number of hits required.  We got, I think, better fits and better explanations.  The number of hits was higher, though at the time nothing was known about what they were.  Around 1990 Adam Connor and I suggested that the age pattern of cancer could be accounted for by the age-related probability that some individual cell would acquire some critical set of changes as a function of age, here we didn't specify the number.  This, too, seemed to fit the age-patterns and both approaches suggested that cancers were as a group due to similar genetic processes (whether or not they affected different genes in each instance--there was no useful data at the time), but left open the number of events involved. Since then, it has become clear that many different genes, and different combinations in different instances of cancer in the same organ (lung, stomach, etc.) are involved.  In all, these facts and findings account for the complexity of cancer (and, indeed, many other common normal or abnormal traits).

But if 'luck' means that some individual cell has, for whatever reason, acquired an initiating set of mutations or growth stimuli, then we can expect that to a great extent, each transformed cell is transformed for a different genotypic reason, and no one gene need be involved, or is sufficient.  You just get a bad roll of the mutational dice in one of your cells, regardless of whether the mutation is only due to DNA copying or has been affected by external agents.  The difference would be rather slight, and the main correlation (as in Tomasetti and Vogelstein) related to how many cell-turnovers are at risk.

But the species differences show that something other than just 'luck', or luck affected by lifestyle factors, is involved, and what that is, is basically not known.  That suggests that the Tomasetti and Vogelstein interpretation is itself missing something important (though it won't change the empirical fact that neither inherited genotypes nor most environmental exposures do not have highly predictive effects).

In sum
Cancer is more, and less, than pure luck.  And its causes are still poorly understood.  We think as we've said that the Tomasetti and Vogelstein paper points to many things that are shown by new data--but little if anything that wasn't known, shown, and understood for the right reasons a generation ago.  The love affair with inherited genotypes, enabled, encouraged, and funded by a variety of enthusiasms, opportunities, and vested interests, has distracted attention from working from what we knew.  The problem is that the somatic mutational nature of cancer doesn't lead to tidy prediction, prevention or interventions, at least not with current thinking.  But that's where future thinking should be going.