On 2015 Feb 25, Miguel Lopez-Lazaro commented:

Conclusion not supported by the data

The authors propose that cancer is largely caused by unavoidable mutations arising during DNA replication. This proposal is based 1) on a strong correlation between the number of stem cell divisions accumulated by a tissue and the risk of being diagnosed with cancer in the tissue, and 2) on the assumption that the number of stem cell divisions is equivalent to the number of unavoidable mutations arising during DNA replication. The authors do not report any correlation between the number of mutations in a tissue and the risk of cancer in the tissue. However, since cell division can generate mutations, they assumed that the parameters “stem cell divisions” and “mutations arising during DNA replication” are interchangeable. Recent data indicate that this assumption is incorrect:

Tissue-specific mutation accumulation in human adult stem cells during life. https://www.ncbi.nlm.nih.gov/pubmed/27698416

Cancer Etiology: Variation in Cancer Risk among Tissues is Poorly Explained by the Number of Gene Mutations. https://www.preprints.org/manuscript/201708.0103/v1

The correlation reported by the authors indicates that carcinogenesis is driven by the accumulation of cell divisions in stem cells, and not by random mutations arising during DNA replication. The implications are completely different: https://www.preprints.org/manuscript/201707.0074/v1

On 2015 Feb 06, Daniel Corcos commented:

In the abstract of the paper, there is a wrong statement : « These results suggest that only a third of the variation in cancer risk among tissues is attributable to environmental factors or inherited predispositions », seemingly supported by the following sentence : « If there is a high cancer risk of that tissue type relative to its number of stem cell divisions—then one would expect that environmental or inherited factors would play a relatively more important role in that cancer’s risk. »

Basically, this type of reasoning tends to confuse mathematical normality with health, which leads to the conclusion that the higher disease risk population is always less than half of the total population.

Estimating the percentage of cancers related to genetic inheritance is impossible.
 As an example, if a minority of individuals are protected by their genetic constitution (for instance, if they have a three time less probability of cancer), then one could rightly say that the majority of cancers are related to genetic inheritance. Pure speculation?
 Then have a look:

http://www.ncbi.nlm.nih.gov/pubmed/10506723

Now, let's suppose that two cancer types have different incidence relative to their corresponding normal tissue cell divisions. One might say, with the authors, that one is due to environmental factors, whereas the other is not. Wrong. Both could be due to environmental factors, but one would be more affected.

More generally, if it is possible to say that ten per cent of cancers are attributable to tobacco, it is impossible to say that there is a defined percentage of cancers due to environmental factors, because there is no such thing as an environment free population. The only thing we can do is to give a minimal estimate of the percentage of cancers that would be prevented by removing defined environmental factors.

In conclusion, in addition to the criticism that has been made on the methodology in this place and many others, one may wonder if this paper has anything to do with science (as a knowledge enterprise) and how it has passed peer review in Science (the journal).

https://www.researchgate.net/profile/Daniel_Corcos2

On 2015 Feb 05, Jim Brody commented:

The key paragraph in this paper is:

"A linear correlation equal to 0.804 suggests that 65% (39% to 81%; 95% CI) of the differences in cancer risk among different tissues can be explained by the total number of stem cell divisions in those tissues. Thus, the stochastic effects of DNA replication appear to be the major contributor to cancer in humans."

Thus, the authors have a parameter that quantifies the "randomness" of cancer. By definition this parameter must be between 0% and 100%. The authors are 95% confident that it lies between 39% and 81%.

I look forward to more precise measurements of this parameter.

On 2015 Feb 05, Cyrille Delpierre commented:

No proof that cancer is mainly a random process C Delpierre, R Fantin, S Lamy, P Grosclaude, T Lang, M Kelly-Irving

Tomasetti and Vogelstein (1) suggest in a recent study that the majority of the variation in cancer risk in tissues (65%) is attributable to “bad luck”. This resounding fact is underscored in their paper, and has led to widespread global media interest. The unfortunate consequence of such media interest is misrepresentation and confusion surrounding the results. Three points should be raised to appropriately consider the significance of the authors’ findings. First point: the authors compare the probability of a cancer cell appearing at the tissue level, and not at the individual level. Their study does not say anything about the differences in cancer incidence observed at the population level, between countries or social groups. The risk of developing cancer is not random at the population level. The risk at the tissue level cannot be attributed to the level of the individual human, and certainly should not lead to population-level assumptions. The authors found that 21 cancers among the 32 studied (71%) represent ‘replicative cancers’ (stochastic cancers) and 9 cancers (29%) represent ‘deterministic cancers’ (linked to environmental or hereditary of cancer types). If we consider the number of cancer cases by type, the proportions are then dramatically different. Based on the same data used by the authors on cancer incidence in 2014 (2) and information provided in the supplementary materials, the proportion of total cancers represented by replicative cancers is around 20%. Cancer types not included in the study like prostate, breast, cervical, uterine and endometrial, kidney and bladder cancers, for which environmental factors have been identified or suspected, represent around 40% of cancer cases. Consequently, the significance accorded to “bad luck” is radically different in terms of cancer cases, since only a minority of cancer cases may be due to “bad luck”, ultimately changing the conclusions of the paper in particular at a public health level regarding possible prevention strategies. It is then ambiguous to write in the last paragraph that “stochastic effects associated with DNA replication contribute in a substantial way to human cancer incidence.” Second point: we have a number of questions regarding calculations presented in the paper. After careful reading of the main article and the supplementary materials it seems that figure 1 does not represent the lifetime risk of cancer according to the total number of stem cell divisions, as written in column 3 on page 79, but plots the log of the lifetime risk for cancer according to the log of the total number of stem cell divisions (as indicated on page 11 of the supplementary materials). If so the Pearson correlation of 0.804 has in fact been calculated using the log of the two values. Consequently the explained part of variance of 65% refers to the explained part of variance of the log of the lifetime risk for cancer. There are two ways to estimate the proportion of observed differences in cancer risk among tissues explained by the observed total number of stem cells divisions: 1) Using the initial values, the linear correlation is 0.53 suggesting that this proportion is 28% instead of 65%. 2) Using a log-log model (seemingly used by the authors here), the inverse function of the log10-function should be used to calculate the errors of the model. According to our calculations the proportion is 15%. Obviously, whatever the approach used, the results refer to a significantly smaller proportion of the variation in cancer risk among tissues due to ‘bad luck’ which substantially modifies the main message of the paper. Third point: as epidemiologists it seems important to underline that correlation does not mean causation. The authors highlight that cell replication is a major factor determining the appearance of tumor cells. However, mutation is a necessary – but not sufficient - condition for developing cancer. A cancer occurs when many physiological systems fail in particular the immune system which must fail to identify and destroy a cancer cell, allowing it to replicate (3,4). The “behavior” of the immune system has not been shown to be random, but linked with a number of exogenous factors. Thus, even if mutations occur at random, cancer development cannot be considered a random process. Moreover, the assumption that tumor cells are forming at different rates in different tissues on a regular basis depending on the number of stem cell divisions is questionable. Some evidence exists indeed regarding the interconnection between epigenetic processes and mutations in cancer (5). Since epigenetic processes are likely to vary according to environmental conditions, mutation rates might vary according to environmental challenges through epigenetic mechanisms.

Cancer does not occur randomly. While tumor cell production may have an inherent stochastic nature, this is one component of an interaction between complex systems at the individual level, which are not random. At the population level, cancers are not randomly distributed between groups. Medicine and public health need to persist in finding areas of cancer prevention moving above and beyond classic risk factors that take whole systems, both biological and social, into account.

References 1. C. Tomasetti, B. Vogelstein. Science 347, 78-81 (2015) 2. National Cancer Institute, Surveillance, Epidemiology, and End Results Program. http://www.seer.cancer.gov/statfacts 3. K Ryunga, E Manabu, T Kazuaki. Immunology 121, 1-14 (2007) 4. T.J. Stewart, S. Abrams. Oncogene. 2008 Oct 6;27, 5894-903 (2008). 5. J.S. You, P.A. Jones. Cancer Cell Review. 22, 9-20 (2012)

On 2015 Feb 04, Michele Ciulla commented:

The human side of randomness.

The interest of the article by Tomasetti and Vogelstein (1) is mainly epistemological: can science, with its current setting, help us to understand the true meaning of diseases such as cancer? When we try to justify our inferences on the machine of fate, despite knowing that this machine is driven largely by randomness, well we have a problem and this is, at least in part, our partial ignorance of the phenomenon that requires, above all, a reflection on the science of certainty and uncertainty (2). From the clinical point of view, a random event, like an unexpected disease, could have other explanations related to the history of that individual patient. Before building the clinic of randomness, it might be useful to consider patients not only as cases of a statistics but like mind-body unities with a psychosocial individuality and physicians are invited to reflect on Descartes (3). When considering the series of events leading to the neoplastic drift they are possibly non-linear showing a kind of evolution which reflects the changes of the environmental pressure on the individuals and their adaptive responses. This pressure is higher exactly where the genetic program has planned to allocate generative and re-generative resources for development and to buffer environmental changes (4). Thus, tissues that undergo the greatest environmental stress and, therefore, require a greater renewal, are the ones most exposed to the risk of developing malignancies, as the article clearly shows. The boundary between health and disease moves according the reciprocal interaction phenotype-environment and each of us, it should be remembered, is a different phenotype. Who will be next? It is not a roll of the dice to decide it, we have a genetic program that goes on and an environment in continuous change, the machine of fate is just what we call living.

References

1 C.Tomasetti, B. Vogelstein, Science 347 (6217), 78-81 (2015) http://www.sciencemag.org/content/347/6217/78.long

2 E.V. Colani, Journal of Uncertain Systems 2 (3), 202-211 (2008) http://www.worldacademicunion.com/journal/jus/jusVol02No3paper05.pdf

3 G. Duncan, Journal of Medicine and Philosophy 25 (4), 485-513 (2000) http://jmp.oxfordjournals.org/content/25/4/485.short

4 M.M. Ciulla, G.L. Perrucci and F. Magrini, in Regenerative Medicine and Tissue Engineering (InTech Press, 2013), chap 26. http://www.intechopen.com/books/regenerative-medicine-and-tissue-engineering/adaptation-and-evolution-in-a-gravitational-environment-a-theoretical-framework-for-the-limited-re-g

On 2015 Jan 20, Paolo Vineis commented:

We read Drs. Tomasetti and Vogelstein paper on the strong and positive association between the frequency of stem cell division and the risk of cancer with interest (1). However, their analytical approach was limited and their interpretation of findings was misleading. First, the study was based on a relatively small number of cancer types, most of which are rare, and excluded several common cancer types such as breast, prostate, bladder, and endometrium. Second, the frequency of stem cell division over time or across region is expected to change very little compared to changes in risk of cancer for the various cancer types. For example, during the 20th century in the US, risk for lung cancer increased by more than 50 fold but decreased by about ten-fold for cervix and stomach cancers (2). Liver cancer incidence rates in males (number of newly diagnosed cancer cases per 100,000 males per year) range from 2 in Iceland to almost 100 in Mongolia (3), with even larger variation if we were to consider incidence in high-risk vs. low-risk subgroups of populations. These data suggest that the degree of association between the frequency of stem cell division and the risk of cancer across tissues is unlikely to remain constant over time and across regions. Third, their statement on page 79, first column “we show that these stochastic influences are in fact the major contributors to cancer overall, often more important than either hereditary or external environmental factors” is not supported by the data. They can only say that variations in life time risk of cancer across cancer types could be explained by differences in frequency of stem cell divisions as stated on page 79 of the paper. Fourth, the inclusion of oesophageal and head and neck cancers in the “Replicative” category is questionable, since risk factors are well-known for a large fraction of these cancers. The overall conclusion that a large proportion of cancers would not be preventable is not supported by the analyses contained in the paper.

Paolo Vineis School of Public Health, Imperial College London, W2 1PG UK. e-mail - p.vineis@imperial.ac.uk Ahmedin Jemal American Cancer Society, Atlanta, GA 30303 USA

1. Tomasetti C, Vogelstein B. Cancer etiology. Variation in cancer risk among tissues can be explained by the number of stem cell divisions. Science. 2015 Jan 2;347(6217):78-81
2. Siegel RL, Miller KD, Jemal A. Cancer Statistics, 2015. CA Cancer J Clin. 2015 Jan 5. doi: 10.3322/caac.21254. [Epub ahead of PRINT
3. Globocan 2012, International Agency for Reaserch on Cancer. Acessed on January 9, 2015. http://globocan.iarc.fr/Default.aspx

On 2015 Jan 15, Vladimir Kuznetsov commented:

One of the main hallmarks of cancer is uncontrolled cell proliferation, ultimately leading to the death of the multicellular organism. The strong link between cell proliferation and cancer risk is well-known[1]. While studying this correlation in various human tissues, the scientists at Johns Hopkins University in the U.S. hypothesize that most of the genomic changes leading to human cancer "occur simply by chance during DNA replication rather than as a result of carcinogenic factors"[2]. According to their hypothesis, the authors observed correlation between the estimates of division rates of normal self-renewing tissue-specific cell population (termed as “normal stem cells”(NSC)) with the estimates of overall lifetime risk of cancers, studied across the USA population. In this comment, I point out that the lack of tissue-specific data quality and its incompleteness, population bias, and oversimplification of modeling can lead to ambiguity in the interpretation of the results and loss of confidence in the dichotomized classification of tissue-specific cancer risks in context of efficiency of 'primary prevention measures'. 1. The cell counts and division rates of NSC in normally slow-proliferative and non-proliferated tissues/cell subpopulation (in liver, brain, gallbladder, thyroid tissue, bone, and pancreas) might not be accurately estimated and extrapolate to CSC. 2. The correlation model does not consider directly the level of variation of the number of NSC divisions due to "random mutations arising during DNA replication in normal non-cancerous stem cells" in individuals. According to the authors, "random mutations arising during DNA replication in normal non-cancerous stem cells” leading eventually to occurrence of cancer stem cells (CSC) and tumors due to “many genomic changes occur simply by chance”, or “bad luck”. However, recent integrative genomics studies of somatic mutation spectra in different tumors defined tissue-associated non-random somatic mutation signatures, mutation clusters with specific sequence context and preferential location of the mutations in certain disease-associated chromosomal regions involved in the initiation and development tissue-origin tumor types and subtypes [3, 4]. 3. The model in [2] assumes that the proliferation rate of NSC in normal tissue and the proliferation rate of CSC are closely correlated, however the assumption may be not true [3]. 4. Based on "extra risk score"(ERS) cluster analysis, the authors provided a classification of the tumors into two classes, referred as R-tumors (occurred at random; relatively smaller ERS) and D-tumors (occurred via additional 'deterministic factors'). However, ERS did not include any additive and multiplicative factors which allow to estimate explicitly the effects of individual covariates and their interactions (for instance see [5]). Thus, clustering model does not allow in principle to estimate the significance of alternative factors due to absence of these covariates in the model risk assessment. 4.Variation of the total number of stem cell divisions during an individual’s lifetime is unknown. Therefore, the model reported in [2] says nothing about variation in the cancer risk between individuals and cannot say that ~70% of cancer cases are just "bad luck" only due to count of the rates of divisions in the limited number of normal stem cells. It was argued, that breast cancer stem cell-like cells arise de novo form non-stem-like tumor cells [3] and this could make cancer cell population more heterogeneous. Such plasticity indicates that the concept of CSC can be essentially different from that of NSC. It makes a linear correlation model proposed in [2] more difficult in context of its mechanistic interpretation.<br> 5.The authors considered the "ovarian cancer germ cell" as a representative precursor (“cell of origin”) of the cancer cells in ovaries. However, according to the literature, “ovarian cancer germ cells” (should be classified as D-tumors) and responsible for only a few percent of ovarian cancer cases. Furthermore, "ovarian cancer” can be mostly represented by some distinct secondary “cell of origin” due to migration (or metastasis) from source tissues/organs [6, 7]. Therefore, the estimations under their model assumptions become inappropriate.<br> 6.The authors used 31 normal tissues. The exclusion of the common cancers (breast, prostate and gastric cancers) and the inclusion of 5 times osteosarcomas (depended samples) could induce essential bias which further complicates result interpretation and ability to extrapolate the results into entire population in the USA. 7.Their analysis was not directly concerned with the variations of population-specific incidences or other environmental causes of cancers. For instance, according to published statistics, oral cancer (OrCa) is a heterogeneous group of cancers arising from different parts of the oral cavity, with well-defined and differentiated predisposing risk factors, prevalence, and treatment outcomes[8-10]. There is a significant difference in the incidence of OrCa in different regions of the world. In contrast with the U.S. population where oral cavity cancer represents only about 3% of occurring malignancies, it accounts for over 30% of all oral cancers in India. Due to these results, it is unlikely that these well-established observations can be explained with the prevalence of “oral cancer stem cells”[2] variations. It was estimated that 91% of OrCa cases in the U.K. are linked to lifestyle factors including smoking (57%), alcohol (30%), and infections (13%)[9]. Such knowledge provides oncologists and patients a real hope for prophylactic efforts and prevention via early detection of the OrCa in a near future[10,11]. However according to the prediction in [2], OrCa was classified as so-called R-tumors, of which “primary prevention measures are not likely to be very effective”[2]. 8. Comparison of Fig2 and Fig S1 in the main text and suppl. file, shows that so-called D class tumors includes 9 cancer types in Fig 2 whereas12 cancer types were represented in Figure S1. In Fig S1, head &neck, melanoma, and gallbladder tumors were included in the D-cluster. Also, for ovarian, testicular, and thyroid cancers which were classified by the authors[2] as R-tumors, the significance of the impact of lifestyle and diet in reducing the cancer risk has been reported[12]. For others such as pancreatic, laryngeal, lip and oral cancers which were also classified as R-tumors, the significance of smoking as significant risk factors has also been established. Therefore, there are several inconsistencies in the results of [2] when compared with current knowledge from the literature. 9.Summary: The classification of the tumors on the R (random) and D (deterministic) classes is based on indirect and unreliable measurements and to a certain extent, even inconsistent with well-established data. Risks of at least several of the R-tumor types of cancer can be significantly reduced by several environmental improvements, diets and prophylactic approaches. Therefore, the conclusion that "primary prevention measures are not likely to be effective..." for tumors arising in organs undergoing origin stem cells and their divisions could be misleading and inappropriate. The predictions of the models based on the U.S. data might not be scalable onto other countries and geographic regions. Direct detection of the NSC and CSC characteristics should be obtained and multivariate probabilistic models of cancer risk prediction should be developed and used.

References: PMID: 1: 2174724; 2: 25554788; 3: 21854987; 4: 24132290;5: ISBN 978-0-205-45938-4; 6: 24879340; 7: 24265397; 8: 24408568; 9: http://www.cancerresearchuk.org/cancer-info/cancerstats/keyfacts/Allcancerscombined/;10: 16629526; 11: 15936419; 12: 24379012

On 2015 Jan 13, Mark Burkitt commented:

The powerful correlation between the rate of stem cell proliferation and the incidence of cancer across a range of tissues, reported recently by Tomasetti and Vogelstein (1), provides important insight into the origins of cancer.

From the findings of their study, the authors appear to make the case that random mutations, occurring during stem cell divisions, are responsible for the majority of cancers. Although many environmental agents (including dietary carcinogens and various forms of radiation) are able to cause mutation, Tomasetti and Vogelstein argue that the variation in the incidence of cancer across different tissues can be explained largely in terms of the differences in the rates of cell division between tissues: the greater the number of cell divisions, the greater the number of random mutations, hence the greater the incidence of cancer. There is, however, an alternative explanation for the statistical correlation revealed by Tomasetti and Vogelstein – an explanation in which environmental factors play a more important, underlying role.

Consider the scenario in which an environmental agent (let’s say, a chemical carcinogen in the diet) causes a mutation in a stem cell. Whether or not this develops into a tumour depends on several factors, including whether the mutated (precancerous) cell undergoes proliferation before it can be destroyed by apoptosis and/or the immune system. There is also a window of opportunity for DNA repair enzymes to detect and repair the lesion before the cell divides. Therefore, from the moment a mutation arises, a race is on: can the cell be repaired or deleted before it divides? If the mutation occurs in a cell type having a high rate of cell division, it is more likely to lead to a clone of modified cells and, consequently, a tumour.

Such a view of the origins of cancer is consistent with the statistical trend reported by Tomasetti and Vogelstein: an environmental ‘factor’ acting on a stem cell in, for example, the colon is more likely to result in a tumour than the same factor acting on a cell in the leg. The rate of cell proliferation correlates with the risk of cancer so convincingly precisely because cell division is needed to ‘fix’ the mutation caused by the environmental agent – to convert the mutated cell into a tumour clone before it is repaired or deleted by the various surveillance systems.

The situation is not unlike that encountered in chemical kinetics, a central tenet of which is that the overall rate of a chemical reaction cannot be greater than the rate of its slowest step, the so-called rate-limiting step. No matter how many environmental mutagens you throw at the DNA of a cell, its rate of conversion to a tumour is determined by the rate-limiting step of the whole process – the rate of cell proliferation. If so, then environmental factors may play a greater (albeit ‘hidden’) role in the risk of cancer than might be suggested by Tomasetti and Vogelstein. It could be argued that the environmental component of cancer risk has evaded detection by the statistician’s radar.

References 1. Tomasetti C, Vogelstein B. Variation in cancer risk among tissues can be explained by the number of stem cell divisions. Science 2015; 347:78-81. doi: 10.1126/science.1260825

On 2015 Jan 09, GEORGE BLANCK commented:

Stochastic aspects of cancer development:

The article by Tomasetti and Vogelstein (1) has helped add another dimension to the study of cancer development, most often driven by researching intracellular signaling pathway malfunctions and microenvironment effects, including inflammation. In fact, there are many molecular aspects cancer development that are governed by a random chance component of biological processes. First, genes that form cancer fusion genes are comparatively large (2, 3), presumably due to large introns providing for many opportunities for a productive recombination that leads to a chimeric protein, requiring of course intact exons, but in most cases, no particular retention of intron sequences. Interestingly, one of the smallest cancer fusion genes, EWSR1 (2), occurs in Ewing’s sarcoma, among the rarest cancers.

The sequential order of mutations, originally thought to represent requirements for an ordered process of acquisition of cancer hallmarks, could also be (at least partially) explained on the basis of gene size (4), with metastasis suppressor genes, which cannot be readily distinguished from classical tumor suppressor genes via signal pathway exclusivity, being comparatively small (4) and thus potentially less likely to be mutated early in cancer development. It is not inconceivable that certain functions are required to precede others in cancer development. For example, lack of apoptosis remains a good candidate for preceding metastasis (4). However, stochastic functions are likely to be a major basis for sequential mutations, and this underlying factor is consistent with the more recent appreciation of signal pathway degeneracy in cancer, reflected in the many alternative pathways discovered in drug-resistant cells and by many other aspects of cancer research indicating pathway degeneracy. In short, there is less opportunity to argue for an ordered acquisition of cancer hallmarks if the underlying mechanisms do not credibly distinguish such hallmarks (4, 5).

Interestingly, cytoskeletal related proteins represent some of the largest coding regions in the human genome, and not surprisingly cytoskeletal related protein coding regions are very commonly found mutated in the cancer genome atlas (6). Thus, mutant cytoskeletal proteins may be stochastically inevitable, which raises several interesting questions related to the mutational basis of cancer development and cell shape. First, do mutant cytoskeletal related proteins have a high propensity for a dominant negative impact on the cytoskeleton, as do mutant forms of collagen polypeptides in Osteogenesis imperfecta, where cartilage polymer formation is disrupted by a mutant collagen molecule from just one allele? Second, is the oft-reported, and decades-old connection between metastasis and spherical cells (7, 8) due to mutant forms of the cytoskeleton, which provide for cell rounding and detachment and thus distant circulation? And finally, is the oft-reported, but much more recent connection between spherical cells and drug resistance (9-11) due to common, mutant cytoskeletons that essentially lead to a decreased surface area to volume ratio, in turn leading to reduced intracellular drug concentrations?

As noted (1), the more cell divisions, the more errors, due to intrinsic DNA replication error rates. It remains to be seen to what extent this conclusion is relevant to cancer development in specific settings. Cell division rates may vary with circumstance, particularly over a lifetime, due to such events as wound healing, surgery or radiation or accidents that remove replicative tissue, lymphocyte replication in infections and vaccinations, etc.

The appreciation of probabilistic functions governing cancer development should lead to fresh research avenues, as did the understanding of the roles of signal pathway malfunctions and inflammation. Can screening be organized with more refined purposes and more cost-efficiency, when accounting for the stochastic aspect of mutation occurrence and cancer development? For example, it is likely that point mutations leading to an activating oncoprotein are relatively rare and will have a relatively high probability of being followed by a mutation in a large tumor suppressor gene. On the other hand, many (apparent) cancers that represent large gene mutations may not require aggressive treatment, because the chance of a single base change leading to an activating oncoprotein, needed for aggressive cancer, may be minimal (12).

And, how do stochastic events compare to differences in DNA repair polymorphisms or other rate-limiting aspects of mutation occurrence, such as the differences in mutation rates between heterochromatin and transcriptionally active regions (13)? For example, are there long-term, disease-free smokers because of luck or because of highly efficient repair mechanisms?<br>   References to PubMed comment, January 9, 2015: 1. PMID: 25554788. 2. PMID: 19446742. 3. PMID: 23162078. 4. PMID: 22701759. 5. PMID: 25450826. 6. PMID: 25451318. 7. PMID: 6256751. 8. PMID: 7000337. 9. PMID: 24409314. 10. PMID: 24112388. 11. PMID: 24821384. 12. PMID: 25294886. 13. PMID: 25456125.

On 2015 Jan 06, Andrea Coletta commented:

As other commentators before me, I find hard to accept the authors conclusions.

Although the clustering shows a larger number of cancer types (22 vs 9) for which random mutations seems to be the major causing factor, this cannot directly lead to the conclusion that the majority of absolute number of cancers are caused by random mutations. If we compare Fig1 and Fig2 in the article we can see that the so called "minority" (9 cancer types) have an approximate average lifetime risk ~10E-1 while this is surely lower in the case of the "majority" (22 cancer types). I think that a simple weighted average would re-balance the presented figures of 1/3 vs 2/3 in "favor" of the genetic+environment causes, though I'm not sure this would be the most accurate way to follow (from a statistical point of view).

On 2015 Jan 05, Vincent Detours commented:

No one denies that cancer initiation has a stochastic component, but the conclusion that "prevention measure are not likely to be effective" for tumors arising in organs undergoing many stem cell divisions could be dangerously broad if not misleading.

The paper's investigation is limited to the variation of cancer incidence among organs within a single population. But the incidence of many cancers varies enormously among populations. For example, the incidence of esophagus cancer is 20-30 time higher in China than in the USA and 50-100 higher in subject with a history of Barett esophagus (http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3769895). Yet, this cancer is considered the result of 'bad luck' and unlikely to benefit from prevention measures (R-group) according to Tomasetti and Vogelstein who consider only the overall incidence in the USA. Their analysis is blind to the population-specific incidences and consequently to many environmental and genetics causes of cancer.

The variation of cancer incidence among organs spans five orders of magnitudes. Hence, a one or two orders of magnitude difference due to, say, Barett esophagus, would presumably not affect drastically the overall correlation between organ-related incidence and stem cell division. The classification in the 'deterministic' vs. 'replicative' framework proposed in the paper, however, could change dramatically. This is in fact illustrated by a few cancers for which the authors stratify incidence according to etiology, e.g. virus-associated liver and head & neck cancers vs. their non virus-related counterparts and lung cancers of smokers vs. non smokers. Likely the same could have occurred with many other cancers provided a more detailed population stratification.

Another possibility limiting the authors conclusions is that the total number stem cells divisions in an organ could itself vary from person to person under the influence of non-random genetic and/or environmental factors. A trivial example is age. It is hardly a preventable phenomenon, but other preventable factors could also play a role, tissue injury for example.

A fully developped and extended version of this comment can be found here: http://biorxiv.org/content/early/2015/08/12/024497.

On 2015 Jan 03, William Grant commented:

The paper by Tomasetti and Vogelstein reported that variation in cancer risk among tissues can be explained by the number of stem cell divisions (1). In addition, they divided cancer types into two categories: those more susceptible to environmental factors (D-tumors) and those more susceptible to stochastic effects associated with DNA replication of the tissues' stem cells (R-tumors). The authors concluded that primary prevention measures were not likely to be very effective for R-tumor types of cancer. In this comment, I point out that risk of cancers in this category can be reduced through primary prevention measures.

One way to reduce risk of many types of cancer is through higher solar UVB exposure. Many types of cancers have been found inversely correlated with indices of solar UVB doses in geographical ecological studies in the United States and several other countries including these types of R-tumor types of cancer:esophageal, gallbladder, ovarian, and pancreatic cancer (2). Also, these R-tumor types of cancer have been found inversely correlated with 25-hydroxyvitamin D [25(OH)D] concentrations in prospective observational studies: chronic lymphocytic leukemia (3), head and neck cancer (4), hepatocellular carcinoma (5), and pancreatic cancer (4). One D-tumor type of cancer has the strongest evidence for beneficial effects of UVB exposure and vitamin D: colorectal cancer.

The findings with respect to UVB exposure are generally attributed to production of vitamin D. However, they may also include some effects from mechanisms other than vitamin D. For example, in a mouse model experiment on intestinal tumors, considered an R-tumor type of cancer, both UVB exposure and oral vitamin D reduced the progression of the tumors, with UVB being more effective than oral vitamin D at approximately the same 25(OH)D concentrations (6). However, neither approach reduced the incidence rate of intestinal tumors in this mouse model with a genetic propensity for intestinal tumors. The mechanisms whereby vitamin D reduces the risk of cancer include effects on cellular differentiation, progression, and apoptosis (2). Vitamin D also reduces progression of tumors by reducing angiogenesis around tumors, and reduces metastasis as well.

Another very important risk factor for cancer is diet. In a multi-country ecological study of cancer incidence rates with respect to diet, smoking, latitude, gross domestic product, alcohol consumption, and life expectancy, high animal product consumption was a significant risk factor for three types of R-tumor cancers: ovarian, testicular, and thyroid cancer (7). Smoking was a significant risk factor for these types of R-tumor cancers: laryngeal, lip and oral, and pancreatic cancer. Animal product consumption is a risk factor for cancer in part through increasing insulin-like growth factor-I (IGF-I) (8), which increases growth tumors as shown for small cell lung cancer (9).

Thus, risk of several of the R-tumor types of cancer can be reduced by several environmental approaches including higher UVB exposure and 25(OH)D concentrations, eating fewer animal products, and not smoking. So "bad luck" can be overcome by "healthy choices".

References 1. Tomasetti C, Vogelstein B. Variation in cancer risk among tissues can be explained by the number of stem cell divisions. Science 2015;347:78-81. 2. Moukayed M, Grant WB. Molecular link between vitamin D and cancer prevention. Nutrients. 2013;5:3993-4023. http://www.mdpi.com/2072-6643/5/10/3993 3. Luczynska A, Kaaks R, Rohrmann S, et al. Plasma 25-hydroxyvitamin D concentration and lymphoma risk: results of the European Prospective Investigation into Cancer and Nutrition. Am J Clin Nutr. 2013;98:827-38 4. Afzal S, Bojesen SE, Nordestgaard BG. Low plasma 25-hydroxyvitamin D and risk of tobacco-related cancer. Clin Chem. 2013;59:771-80. 5. Fedirko V, Duarte-Salles T, Bamia C, et al. Pre-diagnostic circulating vitamin D levels and risk of hepatocellular carcinoma in European populations: A nested case-control study. Hepatology. 2014;60:1222-30. 6. Rebel H, der Spek CD, Salvatori D, et al. UV exposure inhibits intestinal tumour growth and progression to malignancy in intestine-specific Apc mutant mice kept on low vitamin D diet. Int J Cancer. 2015;136:271-7. 7. Grant WB. A multicountry ecological study of cancer incidence rates in 2008 with respect to various risk-modifying factors, Nutrients. 2014;6:163-189. http://www.mdpi.com/2072-6643/6/1/163 8. Larsson SC, Wolk K, Brismar K, Wolk A. Association of diet with serum insulin-like growth factor I in middle-aged and elderly men. Am J Clin Nutr. 2005;81:1163-7. 9. Warshamana-Greene GS, Litz J, Buchdunger E, et al. The insulin-like growth factor-I (IGF-I) receptor kinase inhibitor NVP-ADW742, in combination with STI571, delineates a spectrum of dependence of small cell lung cancer on IGF-I and stem cell factor signaling. Mol Cancer Ther. 2004;3:527-35.

Disclosure I receive funding from Bio-Tech Pharmacal (Fayetteville, AR) and MediSun Technology (Highland Park, IL).

On 2015 Jan 03, William Grant commented:

The paper by Tomasetti and Vogelstein reported that variation in cancer risk among tissues can be explained by the number of stem cell divisions (1). In addition, they divided cancer types into two categories: those more susceptible to environmental factors (D-tumors) and those more susceptible to stochastic effects associated with DNA replication of the tissues' stem cells (R-tumors). The authors concluded that primary prevention measures were not likely to be very effective for R-tumor types of cancer. In this comment, I point out that risk of cancers in this category can be reduced through primary prevention measures.

One way to reduce risk of many types of cancer is through higher solar UVB exposure. Many types of cancers have been found inversely correlated with indices of solar UVB doses in geographical ecological studies in the United States and several other countries including these types of R-tumor types of cancer:esophageal, gallbladder, ovarian, and pancreatic cancer (2). Also, these R-tumor types of cancer have been found inversely correlated with 25-hydroxyvitamin D [25(OH)D] concentrations in prospective observational studies: chronic lymphocytic leukemia (3), head and neck cancer (4), hepatocellular carcinoma (5), and pancreatic cancer (4). One D-tumor type of cancer has the strongest evidence for beneficial effects of UVB exposure and vitamin D: colorectal cancer.

The findings with respect to UVB exposure are generally attributed to production of vitamin D. However, they may also include some effects from mechanisms other than vitamin D. For example, in a mouse model experiment on intestinal tumors, considered an R-tumor type of cancer, both UVB exposure and oral vitamin D reduced the progression of the tumors, with UVB being more effective than oral vitamin D at approximately the same 25(OH)D concentrations (6). However, neither approach reduced the incidence rate of intestinal tumors in this mouse model with a genetic propensity for intestinal tumors. The mechanisms whereby vitamin D reduces the risk of cancer include effects on cellular differentiation, progression, and apoptosis (2). Vitamin D also reduces progression of tumors by reducing angiogenesis around tumors, and reduces metastasis as well.

Another very important risk factor for cancer is diet. In a multi-country ecological study of cancer incidence rates with respect to diet, smoking, latitude, gross domestic product, alcohol consumption, and life expectancy, high animal product consumption was a significant risk factor for three types of R-tumor cancers: ovarian, testicular, and thyroid cancer (7). Smoking was a significant risk factor for these types of R-tumor cancers: laryngeal, lip and oral, and pancreatic cancer. Animal product consumption is a risk factor for cancer in part through increasing insulin-like growth factor-I (IGF-I) (8), which increases growth tumors as shown for small cell lung cancer (9).

Thus, risk of several of the R-tumor types of cancer can be reduced by several environmental approaches including higher UVB exposure and 25(OH)D concentrations, eating fewer animal products, and not smoking. So "bad luck" can be overcome by "healthy choices".

References 1. Tomasetti C, Vogelstein B. Variation in cancer risk among tissues can be explained by the number of stem cell divisions. Science 2015;347:78-81. 2. Moukayed M, Grant WB. Molecular link between vitamin D and cancer prevention. Nutrients. 2013;5:3993-4023. http://www.mdpi.com/2072-6643/5/10/3993 3. Luczynska A, Kaaks R, Rohrmann S, et al. Plasma 25-hydroxyvitamin D concentration and lymphoma risk: results of the European Prospective Investigation into Cancer and Nutrition. Am J Clin Nutr. 2013;98:827-38 4. Afzal S, Bojesen SE, Nordestgaard BG. Low plasma 25-hydroxyvitamin D and risk of tobacco-related cancer. Clin Chem. 2013;59:771-80. 5. Fedirko V, Duarte-Salles T, Bamia C, et al. Pre-diagnostic circulating vitamin D levels and risk of hepatocellular carcinoma in European populations: A nested case-control study. Hepatology. 2014;60:1222-30. 6. Rebel H, der Spek CD, Salvatori D, et al. UV exposure inhibits intestinal tumour growth and progression to malignancy in intestine-specific Apc mutant mice kept on low vitamin D diet. Int J Cancer. 2015;136:271-7. 7. Grant WB. A multicountry ecological study of cancer incidence rates in 2008 with respect to various risk-modifying factors, Nutrients. 2014;6:163-189. http://www.mdpi.com/2072-6643/6/1/163 8. Larsson SC, Wolk K, Brismar K, Wolk A. Association of diet with serum insulin-like growth factor I in middle-aged and elderly men. Am J Clin Nutr. 2005;81:1163-7. 9. Warshamana-Greene GS, Litz J, Buchdunger E, et al. The insulin-like growth factor-I (IGF-I) receptor kinase inhibitor NVP-ADW742, in combination with STI571, delineates a spectrum of dependence of small cell lung cancer on IGF-I and stem cell factor signaling. Mol Cancer Ther. 2004;3:527-35.

Disclosure I receive funding from Bio-Tech Pharmacal (Fayetteville, AR) and MediSun Technology (Highland Park, IL).

On 2015 Jan 05, Vincent Detours commented:

No one denies that cancer initiation has a stochastic component, but the conclusion that "prevention measure are not likely to be effective" for tumors arising in organs undergoing many stem cell divisions could be dangerously broad if not misleading.

The paper's investigation is limited to the variation of cancer incidence among organs within a single population. But the incidence of many cancers varies enormously among populations. For example, the incidence of esophagus cancer is 20-30 time higher in China than in the USA and 50-100 higher in subject with a history of Barett esophagus (http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3769895). Yet, this cancer is considered the result of 'bad luck' and unlikely to benefit from prevention measures (R-group) according to Tomasetti and Vogelstein who consider only the overall incidence in the USA. Their analysis is blind to the population-specific incidences and consequently to many environmental and genetics causes of cancer.

The variation of cancer incidence among organs spans five orders of magnitudes. Hence, a one or two orders of magnitude difference due to, say, Barett esophagus, would presumably not affect drastically the overall correlation between organ-related incidence and stem cell division. The classification in the 'deterministic' vs. 'replicative' framework proposed in the paper, however, could change dramatically. This is in fact illustrated by a few cancers for which the authors stratify incidence according to etiology, e.g. virus-associated liver and head & neck cancers vs. their non virus-related counterparts and lung cancers of smokers vs. non smokers. Likely the same could have occurred with many other cancers provided a more detailed population stratification.

Another possibility limiting the authors conclusions is that the total number stem cells divisions in an organ could itself vary from person to person under the influence of non-random genetic and/or environmental factors. A trivial example is age. It is hardly a preventable phenomenon, but other preventable factors could also play a role, tissue injury for example.

A fully developped and extended version of this comment can be found here: http://biorxiv.org/content/early/2015/08/12/024497.

On 2015 Jan 06, Andrea Coletta commented:

As other commentators before me, I find hard to accept the authors conclusions.

Although the clustering shows a larger number of cancer types (22 vs 9) for which random mutations seems to be the major causing factor, this cannot directly lead to the conclusion that the majority of absolute number of cancers are caused by random mutations. If we compare Fig1 and Fig2 in the article we can see that the so called "minority" (9 cancer types) have an approximate average lifetime risk ~10E-1 while this is surely lower in the case of the "majority" (22 cancer types). I think that a simple weighted average would re-balance the presented figures of 1/3 vs 2/3 in "favor" of the genetic+environment causes, though I'm not sure this would be the most accurate way to follow (from a statistical point of view).

On 2015 Jan 09, GEORGE BLANCK commented:

Stochastic aspects of cancer development:

The article by Tomasetti and Vogelstein (1) has helped add another dimension to the study of cancer development, most often driven by researching intracellular signaling pathway malfunctions and microenvironment effects, including inflammation. In fact, there are many molecular aspects cancer development that are governed by a random chance component of biological processes. First, genes that form cancer fusion genes are comparatively large (2, 3), presumably due to large introns providing for many opportunities for a productive recombination that leads to a chimeric protein, requiring of course intact exons, but in most cases, no particular retention of intron sequences. Interestingly, one of the smallest cancer fusion genes, EWSR1 (2), occurs in Ewing’s sarcoma, among the rarest cancers.

The sequential order of mutations, originally thought to represent requirements for an ordered process of acquisition of cancer hallmarks, could also be (at least partially) explained on the basis of gene size (4), with metastasis suppressor genes, which cannot be readily distinguished from classical tumor suppressor genes via signal pathway exclusivity, being comparatively small (4) and thus potentially less likely to be mutated early in cancer development. It is not inconceivable that certain functions are required to precede others in cancer development. For example, lack of apoptosis remains a good candidate for preceding metastasis (4). However, stochastic functions are likely to be a major basis for sequential mutations, and this underlying factor is consistent with the more recent appreciation of signal pathway degeneracy in cancer, reflected in the many alternative pathways discovered in drug-resistant cells and by many other aspects of cancer research indicating pathway degeneracy. In short, there is less opportunity to argue for an ordered acquisition of cancer hallmarks if the underlying mechanisms do not credibly distinguish such hallmarks (4, 5).

Interestingly, cytoskeletal related proteins represent some of the largest coding regions in the human genome, and not surprisingly cytoskeletal related protein coding regions are very commonly found mutated in the cancer genome atlas (6). Thus, mutant cytoskeletal proteins may be stochastically inevitable, which raises several interesting questions related to the mutational basis of cancer development and cell shape. First, do mutant cytoskeletal related proteins have a high propensity for a dominant negative impact on the cytoskeleton, as do mutant forms of collagen polypeptides in Osteogenesis imperfecta, where cartilage polymer formation is disrupted by a mutant collagen molecule from just one allele? Second, is the oft-reported, and decades-old connection between metastasis and spherical cells (7, 8) due to mutant forms of the cytoskeleton, which provide for cell rounding and detachment and thus distant circulation? And finally, is the oft-reported, but much more recent connection between spherical cells and drug resistance (9-11) due to common, mutant cytoskeletons that essentially lead to a decreased surface area to volume ratio, in turn leading to reduced intracellular drug concentrations?

As noted (1), the more cell divisions, the more errors, due to intrinsic DNA replication error rates. It remains to be seen to what extent this conclusion is relevant to cancer development in specific settings. Cell division rates may vary with circumstance, particularly over a lifetime, due to such events as wound healing, surgery or radiation or accidents that remove replicative tissue, lymphocyte replication in infections and vaccinations, etc.

The appreciation of probabilistic functions governing cancer development should lead to fresh research avenues, as did the understanding of the roles of signal pathway malfunctions and inflammation. Can screening be organized with more refined purposes and more cost-efficiency, when accounting for the stochastic aspect of mutation occurrence and cancer development? For example, it is likely that point mutations leading to an activating oncoprotein are relatively rare and will have a relatively high probability of being followed by a mutation in a large tumor suppressor gene. On the other hand, many (apparent) cancers that represent large gene mutations may not require aggressive treatment, because the chance of a single base change leading to an activating oncoprotein, needed for aggressive cancer, may be minimal (12).

And, how do stochastic events compare to differences in DNA repair polymorphisms or other rate-limiting aspects of mutation occurrence, such as the differences in mutation rates between heterochromatin and transcriptionally active regions (13)? For example, are there long-term, disease-free smokers because of luck or because of highly efficient repair mechanisms?<br>   References to PubMed comment, January 9, 2015: 1. PMID: 25554788. 2. PMID: 19446742. 3. PMID: 23162078. 4. PMID: 22701759. 5. PMID: 25450826. 6. PMID: 25451318. 7. PMID: 6256751. 8. PMID: 7000337. 9. PMID: 24409314. 10. PMID: 24112388. 11. PMID: 24821384. 12. PMID: 25294886. 13. PMID: 25456125.

On 2015 Jan 13, Mark Burkitt commented:

The powerful correlation between the rate of stem cell proliferation and the incidence of cancer across a range of tissues, reported recently by Tomasetti and Vogelstein (1), provides important insight into the origins of cancer.

From the findings of their study, the authors appear to make the case that random mutations, occurring during stem cell divisions, are responsible for the majority of cancers. Although many environmental agents (including dietary carcinogens and various forms of radiation) are able to cause mutation, Tomasetti and Vogelstein argue that the variation in the incidence of cancer across different tissues can be explained largely in terms of the differences in the rates of cell division between tissues: the greater the number of cell divisions, the greater the number of random mutations, hence the greater the incidence of cancer. There is, however, an alternative explanation for the statistical correlation revealed by Tomasetti and Vogelstein – an explanation in which environmental factors play a more important, underlying role.

Consider the scenario in which an environmental agent (let’s say, a chemical carcinogen in the diet) causes a mutation in a stem cell. Whether or not this develops into a tumour depends on several factors, including whether the mutated (precancerous) cell undergoes proliferation before it can be destroyed by apoptosis and/or the immune system. There is also a window of opportunity for DNA repair enzymes to detect and repair the lesion before the cell divides. Therefore, from the moment a mutation arises, a race is on: can the cell be repaired or deleted before it divides? If the mutation occurs in a cell type having a high rate of cell division, it is more likely to lead to a clone of modified cells and, consequently, a tumour.

Such a view of the origins of cancer is consistent with the statistical trend reported by Tomasetti and Vogelstein: an environmental ‘factor’ acting on a stem cell in, for example, the colon is more likely to result in a tumour than the same factor acting on a cell in the leg. The rate of cell proliferation correlates with the risk of cancer so convincingly precisely because cell division is needed to ‘fix’ the mutation caused by the environmental agent – to convert the mutated cell into a tumour clone before it is repaired or deleted by the various surveillance systems.

The situation is not unlike that encountered in chemical kinetics, a central tenet of which is that the overall rate of a chemical reaction cannot be greater than the rate of its slowest step, the so-called rate-limiting step. No matter how many environmental mutagens you throw at the DNA of a cell, its rate of conversion to a tumour is determined by the rate-limiting step of the whole process – the rate of cell proliferation. If so, then environmental factors may play a greater (albeit ‘hidden’) role in the risk of cancer than might be suggested by Tomasetti and Vogelstein. It could be argued that the environmental component of cancer risk has evaded detection by the statistician’s radar.

References 1. Tomasetti C, Vogelstein B. Variation in cancer risk among tissues can be explained by the number of stem cell divisions. Science 2015; 347:78-81. doi: 10.1126/science.1260825

On 2015 Jan 15, Vladimir Kuznetsov commented:

One of the main hallmarks of cancer is uncontrolled cell proliferation, ultimately leading to the death of the multicellular organism. The strong link between cell proliferation and cancer risk is well-known[1]. While studying this correlation in various human tissues, the scientists at Johns Hopkins University in the U.S. hypothesize that most of the genomic changes leading to human cancer "occur simply by chance during DNA replication rather than as a result of carcinogenic factors"[2]. According to their hypothesis, the authors observed correlation between the estimates of division rates of normal self-renewing tissue-specific cell population (termed as “normal stem cells”(NSC)) with the estimates of overall lifetime risk of cancers, studied across the USA population. In this comment, I point out that the lack of tissue-specific data quality and its incompleteness, population bias, and oversimplification of modeling can lead to ambiguity in the interpretation of the results and loss of confidence in the dichotomized classification of tissue-specific cancer risks in context of efficiency of 'primary prevention measures'. 1. The cell counts and division rates of NSC in normally slow-proliferative and non-proliferated tissues/cell subpopulation (in liver, brain, gallbladder, thyroid tissue, bone, and pancreas) might not be accurately estimated and extrapolate to CSC. 2. The correlation model does not consider directly the level of variation of the number of NSC divisions due to "random mutations arising during DNA replication in normal non-cancerous stem cells" in individuals. According to the authors, "random mutations arising during DNA replication in normal non-cancerous stem cells” leading eventually to occurrence of cancer stem cells (CSC) and tumors due to “many genomic changes occur simply by chance”, or “bad luck”. However, recent integrative genomics studies of somatic mutation spectra in different tumors defined tissue-associated non-random somatic mutation signatures, mutation clusters with specific sequence context and preferential location of the mutations in certain disease-associated chromosomal regions involved in the initiation and development tissue-origin tumor types and subtypes [3, 4]. 3. The model in [2] assumes that the proliferation rate of NSC in normal tissue and the proliferation rate of CSC are closely correlated, however the assumption may be not true [3]. 4. Based on "extra risk score"(ERS) cluster analysis, the authors provided a classification of the tumors into two classes, referred as R-tumors (occurred at random; relatively smaller ERS) and D-tumors (occurred via additional 'deterministic factors'). However, ERS did not include any additive and multiplicative factors which allow to estimate explicitly the effects of individual covariates and their interactions (for instance see [5]). Thus, clustering model does not allow in principle to estimate the significance of alternative factors due to absence of these covariates in the model risk assessment. 4.Variation of the total number of stem cell divisions during an individual’s lifetime is unknown. Therefore, the model reported in [2] says nothing about variation in the cancer risk between individuals and cannot say that ~70% of cancer cases are just "bad luck" only due to count of the rates of divisions in the limited number of normal stem cells. It was argued, that breast cancer stem cell-like cells arise de novo form non-stem-like tumor cells [3] and this could make cancer cell population more heterogeneous. Such plasticity indicates that the concept of CSC can be essentially different from that of NSC. It makes a linear correlation model proposed in [2] more difficult in context of its mechanistic interpretation.<br> 5.The authors considered the "ovarian cancer germ cell" as a representative precursor (“cell of origin”) of the cancer cells in ovaries. However, according to the literature, “ovarian cancer germ cells” (should be classified as D-tumors) and responsible for only a few percent of ovarian cancer cases. Furthermore, "ovarian cancer” can be mostly represented by some distinct secondary “cell of origin” due to migration (or metastasis) from source tissues/organs [6, 7]. Therefore, the estimations under their model assumptions become inappropriate.<br> 6.The authors used 31 normal tissues. The exclusion of the common cancers (breast, prostate and gastric cancers) and the inclusion of 5 times osteosarcomas (depended samples) could induce essential bias which further complicates result interpretation and ability to extrapolate the results into entire population in the USA. 7.Their analysis was not directly concerned with the variations of population-specific incidences or other environmental causes of cancers. For instance, according to published statistics, oral cancer (OrCa) is a heterogeneous group of cancers arising from different parts of the oral cavity, with well-defined and differentiated predisposing risk factors, prevalence, and treatment outcomes[8-10]. There is a significant difference in the incidence of OrCa in different regions of the world. In contrast with the U.S. population where oral cavity cancer represents only about 3% of occurring malignancies, it accounts for over 30% of all oral cancers in India. Due to these results, it is unlikely that these well-established observations can be explained with the prevalence of “oral cancer stem cells”[2] variations. It was estimated that 91% of OrCa cases in the U.K. are linked to lifestyle factors including smoking (57%), alcohol (30%), and infections (13%)[9]. Such knowledge provides oncologists and patients a real hope for prophylactic efforts and prevention via early detection of the OrCa in a near future[10,11]. However according to the prediction in [2], OrCa was classified as so-called R-tumors, of which “primary prevention measures are not likely to be very effective”[2]. 8. Comparison of Fig2 and Fig S1 in the main text and suppl. file, shows that so-called D class tumors includes 9 cancer types in Fig 2 whereas12 cancer types were represented in Figure S1. In Fig S1, head &neck, melanoma, and gallbladder tumors were included in the D-cluster. Also, for ovarian, testicular, and thyroid cancers which were classified by the authors[2] as R-tumors, the significance of the impact of lifestyle and diet in reducing the cancer risk has been reported[12]. For others such as pancreatic, laryngeal, lip and oral cancers which were also classified as R-tumors, the significance of smoking as significant risk factors has also been established. Therefore, there are several inconsistencies in the results of [2] when compared with current knowledge from the literature. 9.Summary: The classification of the tumors on the R (random) and D (deterministic) classes is based on indirect and unreliable measurements and to a certain extent, even inconsistent with well-established data. Risks of at least several of the R-tumor types of cancer can be significantly reduced by several environmental improvements, diets and prophylactic approaches. Therefore, the conclusion that "primary prevention measures are not likely to be effective..." for tumors arising in organs undergoing origin stem cells and their divisions could be misleading and inappropriate. The predictions of the models based on the U.S. data might not be scalable onto other countries and geographic regions. Direct detection of the NSC and CSC characteristics should be obtained and multivariate probabilistic models of cancer risk prediction should be developed and used.

References: PMID: 1: 2174724; 2: 25554788; 3: 21854987; 4: 24132290;5: ISBN 978-0-205-45938-4; 6: 24879340; 7: 24265397; 8: 24408568; 9: http://www.cancerresearchuk.org/cancer-info/cancerstats/keyfacts/Allcancerscombined/;10: 16629526; 11: 15936419; 12: 24379012

On 2015 Jan 20, Paolo Vineis commented:

We read Drs. Tomasetti and Vogelstein paper on the strong and positive association between the frequency of stem cell division and the risk of cancer with interest (1). However, their analytical approach was limited and their interpretation of findings was misleading. First, the study was based on a relatively small number of cancer types, most of which are rare, and excluded several common cancer types such as breast, prostate, bladder, and endometrium. Second, the frequency of stem cell division over time or across region is expected to change very little compared to changes in risk of cancer for the various cancer types. For example, during the 20th century in the US, risk for lung cancer increased by more than 50 fold but decreased by about ten-fold for cervix and stomach cancers (2). Liver cancer incidence rates in males (number of newly diagnosed cancer cases per 100,000 males per year) range from 2 in Iceland to almost 100 in Mongolia (3), with even larger variation if we were to consider incidence in high-risk vs. low-risk subgroups of populations. These data suggest that the degree of association between the frequency of stem cell division and the risk of cancer across tissues is unlikely to remain constant over time and across regions. Third, their statement on page 79, first column “we show that these stochastic influences are in fact the major contributors to cancer overall, often more important than either hereditary or external environmental factors” is not supported by the data. They can only say that variations in life time risk of cancer across cancer types could be explained by differences in frequency of stem cell divisions as stated on page 79 of the paper. Fourth, the inclusion of oesophageal and head and neck cancers in the “Replicative” category is questionable, since risk factors are well-known for a large fraction of these cancers. The overall conclusion that a large proportion of cancers would not be preventable is not supported by the analyses contained in the paper.

Paolo Vineis School of Public Health, Imperial College London, W2 1PG UK. e-mail - p.vineis@imperial.ac.uk Ahmedin Jemal American Cancer Society, Atlanta, GA 30303 USA

1. Tomasetti C, Vogelstein B. Cancer etiology. Variation in cancer risk among tissues can be explained by the number of stem cell divisions. Science. 2015 Jan 2;347(6217):78-81
2. Siegel RL, Miller KD, Jemal A. Cancer Statistics, 2015. CA Cancer J Clin. 2015 Jan 5. doi: 10.3322/caac.21254. [Epub ahead of PRINT
3. Globocan 2012, International Agency for Reaserch on Cancer. Acessed on January 9, 2015. http://globocan.iarc.fr/Default.aspx

On 2015 Feb 04, Michele Ciulla commented:

The human side of randomness.

The interest of the article by Tomasetti and Vogelstein (1) is mainly epistemological: can science, with its current setting, help us to understand the true meaning of diseases such as cancer? When we try to justify our inferences on the machine of fate, despite knowing that this machine is driven largely by randomness, well we have a problem and this is, at least in part, our partial ignorance of the phenomenon that requires, above all, a reflection on the science of certainty and uncertainty (2). From the clinical point of view, a random event, like an unexpected disease, could have other explanations related to the history of that individual patient. Before building the clinic of randomness, it might be useful to consider patients not only as cases of a statistics but like mind-body unities with a psychosocial individuality and physicians are invited to reflect on Descartes (3). When considering the series of events leading to the neoplastic drift they are possibly non-linear showing a kind of evolution which reflects the changes of the environmental pressure on the individuals and their adaptive responses. This pressure is higher exactly where the genetic program has planned to allocate generative and re-generative resources for development and to buffer environmental changes (4). Thus, tissues that undergo the greatest environmental stress and, therefore, require a greater renewal, are the ones most exposed to the risk of developing malignancies, as the article clearly shows. The boundary between health and disease moves according the reciprocal interaction phenotype-environment and each of us, it should be remembered, is a different phenotype. Who will be next? It is not a roll of the dice to decide it, we have a genetic program that goes on and an environment in continuous change, the machine of fate is just what we call living.

References

1 C.Tomasetti, B. Vogelstein, Science 347 (6217), 78-81 (2015) http://www.sciencemag.org/content/347/6217/78.long

2 E.V. Colani, Journal of Uncertain Systems 2 (3), 202-211 (2008) http://www.worldacademicunion.com/journal/jus/jusVol02No3paper05.pdf

3 G. Duncan, Journal of Medicine and Philosophy 25 (4), 485-513 (2000) http://jmp.oxfordjournals.org/content/25/4/485.short

4 M.M. Ciulla, G.L. Perrucci and F. Magrini, in Regenerative Medicine and Tissue Engineering (InTech Press, 2013), chap 26. http://www.intechopen.com/books/regenerative-medicine-and-tissue-engineering/adaptation-and-evolution-in-a-gravitational-environment-a-theoretical-framework-for-the-limited-re-g

On 2015 Feb 05, Cyrille Delpierre commented:

No proof that cancer is mainly a random process C Delpierre, R Fantin, S Lamy, P Grosclaude, T Lang, M Kelly-Irving

Tomasetti and Vogelstein (1) suggest in a recent study that the majority of the variation in cancer risk in tissues (65%) is attributable to “bad luck”. This resounding fact is underscored in their paper, and has led to widespread global media interest. The unfortunate consequence of such media interest is misrepresentation and confusion surrounding the results. Three points should be raised to appropriately consider the significance of the authors’ findings. First point: the authors compare the probability of a cancer cell appearing at the tissue level, and not at the individual level. Their study does not say anything about the differences in cancer incidence observed at the population level, between countries or social groups. The risk of developing cancer is not random at the population level. The risk at the tissue level cannot be attributed to the level of the individual human, and certainly should not lead to population-level assumptions. The authors found that 21 cancers among the 32 studied (71%) represent ‘replicative cancers’ (stochastic cancers) and 9 cancers (29%) represent ‘deterministic cancers’ (linked to environmental or hereditary of cancer types). If we consider the number of cancer cases by type, the proportions are then dramatically different. Based on the same data used by the authors on cancer incidence in 2014 (2) and information provided in the supplementary materials, the proportion of total cancers represented by replicative cancers is around 20%. Cancer types not included in the study like prostate, breast, cervical, uterine and endometrial, kidney and bladder cancers, for which environmental factors have been identified or suspected, represent around 40% of cancer cases. Consequently, the significance accorded to “bad luck” is radically different in terms of cancer cases, since only a minority of cancer cases may be due to “bad luck”, ultimately changing the conclusions of the paper in particular at a public health level regarding possible prevention strategies. It is then ambiguous to write in the last paragraph that “stochastic effects associated with DNA replication contribute in a substantial way to human cancer incidence.” Second point: we have a number of questions regarding calculations presented in the paper. After careful reading of the main article and the supplementary materials it seems that figure 1 does not represent the lifetime risk of cancer according to the total number of stem cell divisions, as written in column 3 on page 79, but plots the log of the lifetime risk for cancer according to the log of the total number of stem cell divisions (as indicated on page 11 of the supplementary materials). If so the Pearson correlation of 0.804 has in fact been calculated using the log of the two values. Consequently the explained part of variance of 65% refers to the explained part of variance of the log of the lifetime risk for cancer. There are two ways to estimate the proportion of observed differences in cancer risk among tissues explained by the observed total number of stem cells divisions: 1) Using the initial values, the linear correlation is 0.53 suggesting that this proportion is 28% instead of 65%. 2) Using a log-log model (seemingly used by the authors here), the inverse function of the log10-function should be used to calculate the errors of the model. According to our calculations the proportion is 15%. Obviously, whatever the approach used, the results refer to a significantly smaller proportion of the variation in cancer risk among tissues due to ‘bad luck’ which substantially modifies the main message of the paper. Third point: as epidemiologists it seems important to underline that correlation does not mean causation. The authors highlight that cell replication is a major factor determining the appearance of tumor cells. However, mutation is a necessary – but not sufficient - condition for developing cancer. A cancer occurs when many physiological systems fail in particular the immune system which must fail to identify and destroy a cancer cell, allowing it to replicate (3,4). The “behavior” of the immune system has not been shown to be random, but linked with a number of exogenous factors. Thus, even if mutations occur at random, cancer development cannot be considered a random process. Moreover, the assumption that tumor cells are forming at different rates in different tissues on a regular basis depending on the number of stem cell divisions is questionable. Some evidence exists indeed regarding the interconnection between epigenetic processes and mutations in cancer (5). Since epigenetic processes are likely to vary according to environmental conditions, mutation rates might vary according to environmental challenges through epigenetic mechanisms.

Cancer does not occur randomly. While tumor cell production may have an inherent stochastic nature, this is one component of an interaction between complex systems at the individual level, which are not random. At the population level, cancers are not randomly distributed between groups. Medicine and public health need to persist in finding areas of cancer prevention moving above and beyond classic risk factors that take whole systems, both biological and social, into account.

References 1. C. Tomasetti, B. Vogelstein. Science 347, 78-81 (2015) 2. National Cancer Institute, Surveillance, Epidemiology, and End Results Program. http://www.seer.cancer.gov/statfacts 3. K Ryunga, E Manabu, T Kazuaki. Immunology 121, 1-14 (2007) 4. T.J. Stewart, S. Abrams. Oncogene. 2008 Oct 6;27, 5894-903 (2008). 5. J.S. You, P.A. Jones. Cancer Cell Review. 22, 9-20 (2012)

On 2015 Feb 05, Jim Brody commented:

The key paragraph in this paper is:

"A linear correlation equal to 0.804 suggests that 65% (39% to 81%; 95% CI) of the differences in cancer risk among different tissues can be explained by the total number of stem cell divisions in those tissues. Thus, the stochastic effects of DNA replication appear to be the major contributor to cancer in humans."

Thus, the authors have a parameter that quantifies the "randomness" of cancer. By definition this parameter must be between 0% and 100%. The authors are 95% confident that it lies between 39% and 81%.

I look forward to more precise measurements of this parameter.

On 2015 Feb 06, Daniel Corcos commented:

In the abstract of the paper, there is a wrong statement : « These results suggest that only a third of the variation in cancer risk among tissues is attributable to environmental factors or inherited predispositions », seemingly supported by the following sentence : « If there is a high cancer risk of that tissue type relative to its number of stem cell divisions—then one would expect that environmental or inherited factors would play a relatively more important role in that cancer’s risk. »

Basically, this type of reasoning tends to confuse mathematical normality with health, which leads to the conclusion that the higher disease risk population is always less than half of the total population.

Estimating the percentage of cancers related to genetic inheritance is impossible.
 As an example, if a minority of individuals are protected by their genetic constitution (for instance, if they have a three time less probability of cancer), then one could rightly say that the majority of cancers are related to genetic inheritance. Pure speculation?
 Then have a look:

http://www.ncbi.nlm.nih.gov/pubmed/10506723

Now, let's suppose that two cancer types have different incidence relative to their corresponding normal tissue cell divisions. One might say, with the authors, that one is due to environmental factors, whereas the other is not. Wrong. Both could be due to environmental factors, but one would be more affected.

More generally, if it is possible to say that ten per cent of cancers are attributable to tobacco, it is impossible to say that there is a defined percentage of cancers due to environmental factors, because there is no such thing as an environment free population. The only thing we can do is to give a minimal estimate of the percentage of cancers that would be prevented by removing defined environmental factors.

In conclusion, in addition to the criticism that has been made on the methodology in this place and many others, one may wonder if this paper has anything to do with science (as a knowledge enterprise) and how it has passed peer review in Science (the journal).

https://www.researchgate.net/profile/Daniel_Corcos2

On 2015 Feb 25, Miguel Lopez-Lazaro commented:

Conclusion not supported by the data

The authors propose that cancer is largely caused by unavoidable mutations arising during DNA replication. This proposal is based 1) on a strong correlation between the number of stem cell divisions accumulated by a tissue and the risk of being diagnosed with cancer in the tissue, and 2) on the assumption that the number of stem cell divisions is equivalent to the number of unavoidable mutations arising during DNA replication. The authors do not report any correlation between the number of mutations in a tissue and the risk of cancer in the tissue. However, since cell division can generate mutations, they assumed that the parameters “stem cell divisions” and “mutations arising during DNA replication” are interchangeable. Recent data indicate that this assumption is incorrect:

Tissue-specific mutation accumulation in human adult stem cells during life. https://www.ncbi.nlm.nih.gov/pubmed/27698416

Cancer Etiology: Variation in Cancer Risk among Tissues is Poorly Explained by the Number of Gene Mutations. https://www.preprints.org/manuscript/201708.0103/v1

The correlation reported by the authors indicates that carcinogenesis is driven by the accumulation of cell divisions in stem cells, and not by random mutations arising during DNA replication. The implications are completely different: https://www.preprints.org/manuscript/201707.0074/v1