The failure of rodent carcinogenesis as a model for Man

Colin Berry
Queen Mary, London, Mile End Rd, London E1 4NS, UK. E-mail:

Received 23rd October 2017 , Accepted 14th December 2017

First published on 20th December 2017

Recent advances in our understanding of the process of carcinogenesis in Man have required revision of our thinking about the classical initiation/promotion sequence; understanding must now encompass the roles of both genetic and epigenetic change, realisation of the importance of the variable genetic backgrounds of the tumour bearers in any group and an understanding of the importance of random genetic events over time. The behavior of tumours, once established, is more complex than has been thought. Current views of the processes involved are not modelled in toxicity testing programmes.


Fifteen years ago Ettlin and Prentice1 identified a number of problems that were implicit in considering the results of long-term rodent carcinogenicity studies and speculated on what might be done about the issues identified. They wrote about study design and considered dose selection, study conduct, the choice of species, the quality and care of animals, analytical certainties (is the compound tested pure, might contaminants be responsible for the results?) and asked – were the pathological examinations made appropriate and sufficient? Was there peer review of a defined component of the number of tumours examined? Were appropriate guidelines were followed for the recording of data? They emphasised the need for additional studies into mechanism of production of tumours if any are found and suggested how material stored from the study might be examined to help in this later objective.

These issues continue to be discussed by those who have the task of evaluating the results of these tests and few of the problems have disappeared – data recording is perhaps the one which has been best resolved and there is now a more sensible attitude to dose in some quarters. But new doubts have emerged. If we simply consider diet and the conditions in which animals are kept, a different gut biotome may affect results and different suppliers may provide animals with a different gut flora2 – an issue not factored into consideration in most current studies. Importantly there has been no discussion of the reproducibility of results, a major issue in current discussions of the biological experimentation, with some uncomfortable conclusions as attempts to characterise this form of uncertainty, even in simple systems, have appeared.3 Finally, the genetic background of the animals is not considered in any detail; no-one supposes that the “same” strain of animals will be identical in this regard if they are obtained from different suppliers.

Rather than considering the problems of current testing, this paper will start from a different viewpoint. If we take into account the great increase in our knowledge of the mechanisms of carcinogenesis in Man in the last half-century, we might ask – does an investigatory paradigm, established almost sixty years ago, still have value as an indicator of potential harm to Man? This is in no sense a “new” issue; Hartung4 called for change almost a decade ago and several decades ago Munro5 among others, had already questioned the usefulness of rodent studies for pharmaceuticals.

Problems implicit in the current position

How did we get to where we are? The classical pattern of a two stage model of carcinogenesis is almost 100 years old. The concept of an initiating event with a subsequent promotion step was based on the experiments of Katsusaburo Yamagiwa and Koichi Ichikawa and their work on painting the skin of rabbit ears with cyclic hydrocarbons (coal tar) with subsequent trauma to the skin as a second “event”. The experiments were reported in Japanese in 1915, and published in English in 1918.6

These early studies had clearly influenced regulatory toxicology but that influence, having been instructive in identifying fields of enquiry (the role of specific exposures and environmental interactions) ignored another important early observation. Although it had not been made explicit that the two stages envisaged necessarily depended on direct genetic change, a multi-mutation theory on cancer had been proposed by Nordling.7 He noted that in industrialized nations the frequency of cancer seemed to increase according to the sixth power of age and that this correlation could be explained by assuming that the outbreak of cancer requires the accumulation of six consecutive mutations. After many years, this type of progress to malignancy was clearly demonstrated in the colon adenoma/carcinoma sequence by the work of Vogelstein (see review by Vogelstein and Kinzler8).

The prescient observations of Nordling7 and later Armitage and Doll9 do not appear to have been built into our thinking of how tumour development might be considered in regulatory toxicology. Much recent work has shown that in most common tumours of Man complex cascades of events occur, see Katoh and Katoh10 for example. These events depend on a wide range of changes in cell-cycle behaviour on both the activation of genes that stimulate cell proliferation and the deactivation of the so-called tumour suppressor genes. Importantly, the capacity of cells to repair damage to nucleoproteins has been demonstrated to be significant in the causation of some tumours. In this decade it has become clear that knowledge of these interactions and processes may offer insights into possible therapeutic interventions.

That the genetic background of the individual might also be of significant was first brought into focus by Alfred Knudson's statistical analysis of cases of retinoblastoma,11 a tumour that occurs both sporadically and as an inherited disease; he noted that inherited retinoblastoma occurs at a younger age than the sporadic disease and that the children with inherited retinoblastoma often developed the tumour in both eyes, suggesting an underlying predisposition. Knudson suggested, coming from a different perspective, that multiple “hits” to DNA were necessary to cause cancer. In the children with inherited retinoblastoma, the first insult was inherited in the DNA and any second insult would rapidly lead to cancer. In non-inherited retinoblastoma, two “hits” had to take place before a tumour could develop, explaining the age difference in incidence. Knudson's hypothesis referred specifically to the heterozygosity of what were subsequently identified as tumour suppressor genes; a mutation in both alleles is required to produce a neoplastic outcome as a single functional tumour suppressor gene (TSG) is sufficient to prevent tumour development. This work made clear that the genetic soil mattered as much as the environmental seed.

More recently, further work demonstrating the significance of the genetic background in human neoplasia is summarised by Couzin-Frankel12 in a feature in Science. Investigating the parents of children diagnosed with tumours in infancy reveals that there is often a history of premature death from cancer in a parent, sometimes on the basis of a P53 or RB gene abnormality. This may be found in both the child's and parent's somatic (non-tumour) cells. At the Hospital for Sick Children in Toronto, around 40% of children in the hospitals cancer clinic have a family history suggestive of an adverse genetic inheritance and a detailed analysis of 370 surviving patients showed that approximately 29% might have tumours with an inherited genetic component (this is a selected group – remember that non-survivors might also have anomalies). This work has been extended by Mody et al.13 to show that the genetic abnormality found in parental somatic cells may not be “directly” tumour related – a child with neuroblastoma may have an abnormality in a BRCA1 gene (a gene related to breast cancer). In other work, the cancer related genes BRCA1 or BRCA2 have been found in non-tumour cells in older patients in a sarcoma, a carcinoma of the stomach and in a neuroendocrine tumour.

More evidence of complexity dependent on genetic background has been revealed by the scrutiny of those treated for neoplasia later in life. Genome wide studies of those exposed to the cellularly-aggressive therapies used in treatment (many aggressive chemotherapeutic regimes are carcinogenic) show that a sub-set who develop therapy-related myeloid neoplasms do so on the basis of hereditary cancer predisposition syndromes. In a set of 828 individuals analysed, six pedigrees were suggestive of an hereditary breast and ovarian cancer syndrome, three of a Li-Fraumeni-like syndrome and three index patients had multiple primary neoplasms14

So it is reasonable to suggest that the genetic soil should be a major consideration in considering likely risks to any individual from any environmental hazard – to ignore it completely in testing for carcinogenicity is nonsensical. We do not have any real notion of what the likely effects of this loading in short-lived animals might be – somatic cells of tumour-prone strains have not been tested in this context, as far as I am able to determine. However, even if we consider that a finding might be important in mechanistic terms and indicative of risk in Man, data from human studies where well defined causal links are found does not allow assumptions to be made. In adenocarcinoma of the lung, a tumour seen in non-smokers (small-cell and squamous tumours are very rarely found in never-smokers) Alexandrov et al.15 have shown that there is a great increase in signature 4 mutations (characterised by C > A/G > T base substitutions, the type of mutation seen with bezo[a]pyrene in in vitro experiments) in smokers with this tumour; they are almost never seen in non-smokers. However, in other forms of cancer associated with smoking (including bladder cancer as an example, where an increase in DNA adducts is found) it appears that a different carcinogenic mechanism operates – perhaps indirect activation of DNA editing by APOBEC cytidine deaminases and of an endogenous clocklike mutational process, the authors suggest. Thus, even where there is a powerful environmental agent is acting in a manner where it is possible to envisage a direct linkage to a mode of action in one organ, the same mode of action does not appear to operate elsewhere.

That complex cellular cascades exist in human tumours is clear, see ref. 10 for example, but the possibility that similar cascades might be demonstrated in rodents is difficult to imagine; they depend on the passage of time. In the sequence elegantly demonstrated by Vogelstein for human colonic cancer, this time scale is measured in decades. Factors that increase the size of a stem cell pool from which tumours might be generated can be identified in rodent gut by Beyaz et al.16 but this is far from a reconstruction of the Vogelstein paradigm, it is significant only in indicating a larger cell pool in which change can occur.

There appears to be a mind-set which is reluctant to accept the significance of the genetic background. The observations of Vogelstein and Tomasetti;17,18 suggested to the authors that a significant number of cancers are not preventable since they depend on a genetic predisposition (see Couzin-Frankel) and although this view has been attacked19 it has also been defended20 The hypothesis is clearly true to some extent, even if quantitative arguments continue but this set of arguments, like those of earlier models, are flawed in the assumption that stem cell numbers and their proliferation rates remain constant within individual tissues over a lifetime and are not susceptible to environmental interactions. Many have pointed out that this assumption is flawed – and the experimental verification of this is seen in the recent paper of Beyaz et al.

But all of this type of argument suggests that we are trying to find “phenocopies” – seeing if we can map a “route to tumour”; essentially what is being explored in Mode of Action analyses (MoA, see below). However, the testing program is, in as far as it is made explicit, meant to be indicative of a putative risk of producing a form of damage to the process of cell replication that predisposes to tumour formation. If that is a reasonable assumption, are there better ways of doing it?

It is important not to overemphasise our present capability to identify “causative” factors. What we must ensure is that rigorous science be used to identify and evaluate their significance, if they are to be found. Creating anxiety based on imperfect science is to be deplored.


Would formal testing of specific genetic activity in human cell lines be helpful? – untargeted genotoxicity testing is an important part of the current regime. The use human cell lines in compound screening has been explored in depth but whilst some “read-across” is justified many of the extrapolations made are excessive. Damage to genetic mechanisms, when identified, is only indicative of a capability to do harm and even when tests are carefully standardised most will agree that the information on derived from isolated cells must be regarded with caution, in mechanistic terms. Using human tumour cells to test the likely efficacy of anti-tumour treatments has not been a success and recently the NCI-60 panel of 60 human cancer cell lines has been abandoned in favour of patient-derived xenographs (pieces of human tumour grown in mice21). These grafts have been valuable in identifying potential therapeutic targets (see comments by Dienstmann and Tabernero22) but the behaviour of tumours in these systems is not readily characterised.

Miller et al.23 reported on the intractable tumour, glioblastoma, having grown human tumour cells both in vivo and in vitro (in this instance, cells grown in culture or as xenografts in mouse brain) made analyses of epigenetic modifiers. They introduced RNA molecules that inhibited the expression of specific genes and looked for genes whose expression caused reduction in cell numbers, comparing these with controls in which the inhibitory genes were inactive. There was almost no congruence between the results in the two systems; what was found was a profound difference in gene expression in the nutrient-rich environment of culture and in the more “testing” environment of the xenograft; genetic interventions that affected cell survival in one milieu were not effective in the other. Other reasons for caution about using superficially attractive testing systems for genetic activity have been described by Meacham et al.24

Nevertheless, the continuing collection of data on human tumours reveals that genes involved in DNA replication, cell division and programmed cell death are expressed at higher levels in tumours than in somatic cells25 – should we devise mechanisms for looking at these genes, or their operating mechanisms? As Cohen points out MoA investigations have sought to establish the quantitative or qualitative human relevance of specific events at molecular level and subsequent events (in cell division rates, say) that might account for effects observed in toxicity studies. Where increasingly complex MoA datasets are established, consideration of dose and dose–response relationships are critical to meaningful evaluations.

It is evident that extrapolation from animal models to humans requires an understanding of the putative mode of action and that models that include screening for DNA reactivity; evaluation of possible immunosuppression and the evaluation of estrogenic activity can be used. But as Cohen26 points out, evaluation for increased cell proliferation in other tissues can be readily accomplished in short term models (13 weeks or less), with a high degree of dependability and can be readily examined for reproducibility.27 If a positive result is detected in any of these short term screens, the mode of action and detailed dose response can be determined so that a risk evaluation can be made for human beings – whatever reservations you may have about this format, it is more readily repeatable.

Non-laboratory alternatives

It is easy to suggest that human epidemiological studies are the answer; they are considered to be the gold standard for risk assessment and public health decision-making. They are valued over other types of information, including current animal toxicology studies, mainly because they are conducted in the species of interest so we do not have to address the challenge of extrapolating from one species to another. They also provide information on human variability and susceptibility and can now be altered to investigate genetic predisposition – a step which needs to be made. Importantly, epidemiological studies of environmental exposures examine more realistic durations of exposure (often decades longer than those observed in experimental studies) and may identify health effects that generally occur at lower and more human-relevant, levels of exposure. Because of these advantages, EPA's risk assessment guidelines are clear that human data are generally preferable over animal data and should be given greater weight in hazard characterization and dose–response assessment. But, and it is a big but, epidemiology may identify hazards and may establish risks only in a population that has been exposed for some time – this is not testing for toxicity.

In a suggested alternative to current testing Nel and Malloy28 invoke the use of conceptual pathways linking a molecular initiating event [MIE] to an adverse outcome pathway through a series of causally linked events and suggest that high output screening might provide a link to the likelihood of an MIE for a particular compound or chemical structure. In one sense the recent report of Tan et al.29 follows this path in relating aldehyde carcinogenicity to BRCA 2 gene abnormalities in Man. This is clearly an advance in the rationality of its approach but it's generally applicability must be tested.


We must not forget the ritual aspect of toxicology testing; many are happy that a compound is tested without thinking too much about what has been discovered. The late Gerhard Zbinden used to say, when asked how to devise a system that would ensure that no adverse effect from a xenobiotic would be discovered after it had been in use, “you do what you can”. It must be remembered that the commonest reason for the withdrawal of a medicine that has been licensed for use in Man is an adverse effect.

No system can be perfect in this sense – but it should be rational. A life-time rodent carcinogenicity test is clearly not a perfect detector but more importantly, it is no longer a rational approach. It is possible to do better without radical change in laboratory practice. Work on tumour behaviour in Man will no doubt continue to demonstrate that “the proper study of Mankind is Man”.30 This is not nihilism, we can improve outcomes and reduce animal use with the knowledge we have.31


Advances in the understanding of tumour biology in Man have altered our thinking about classical models of the process of carcinogenicity. This casts doubt on the value of a basic current “predictive” test, the rodent lifetime study. Reproducibility of results might be better explored, the applicability of the data obtained more readily evaluated and the numbers of animals used be reduced with sensible application of the information we have today.

Conflicts of interest

There are no conflicts to declare.


  1. R. A. Ettlin and D. E. Prentice, Unexpected tumour findings in lifetime rodent bioassay studies-what to do?, Toxicol. Lett., 2002, 128, 17–33 CrossRef CAS PubMed .
  2. A. C. Ericsson, J. W. Davis, W. Spollen, N. Bivens, S. Givan, C. E. Hagan, M. McIntosh and C. L. Franklin, Effects of Vendor and Genetic Background on the Composition of the Fecal Microbiota of Inbred Mice, PLoS One, 2015, 10, e0116704 Search PubMed .
  3. G. L. Lithgow, M. Driscoll and P. Phillips, A long journey to reproducible results, Nature, 2117, 548, 387–388 CrossRef PubMed .
  4. T. Hartung, Toxicology for the Twenty-First Century, Nature, 2009, 460, 208–2012 CrossRef CAS PubMed .
  5. A. Munro, Are lifespan rodent carcinogenicity studies defensible for pharmaceutical agents?, Exp. Toxicol. Pathol., 1996, 48, 155–166 CrossRef .
  6. K. Yamagiwa and K. Ichikawa, Experimental Study of the Pathogenesis of Carcinoma, J. Cancer Res., 1918, 3, 1–21 Search PubMed .
  7. C. O. Nordling, A new theory on cancer-inducing mechanisms, Br. J. Cancer, 1953, 7, 68 CrossRef CAS PubMed .
  8. B. Vogelstein and K. W. Kinzler, Cancer genes and the pathways they control, Nat. Med., 2004, 10, 789–799 CrossRef CAS PubMed .
  9. P. Armitage and R. Doll, The age distribution of cancer and a multi-stage theory of carcinogenesis, Br. J. Cancer, 1954, 8, 1–12 CrossRef CAS PubMed .
  10. Y. Katoh and M. Katoh, Curr. Mol. Med., 2009, 9, 873–886 CrossRef CAS PubMed .
  11. A. G. KnudsonJr, Mutation and cancer: statistical study of retinoblastoma, Proc. Natl. Acad. Sci. U. S. A., 1971, 68, 820–823 CrossRef CAS .
  12. J. Couzin-Frankel, The bad luck of cancer, Science, 2015, 347, 12 CrossRef CAS PubMed .
  13. R. J. Mody, Y.-M. Wu, R. J. Lonigro, X. Cao, S. Roychowdhury, P. Vats, K. M. Frank, J. R. Prensner, N. P. Asangani, J. R. Dillman, R. M. Rabah, L. P. Kunju, J. Everett, V. M. Raymond, Y. Ning, F. Su, E. M. Stoffel, J. W. Innis, J. S. Roberts, P. L. Robertson, G. Yanik, A. Chamdin, J. A. Connelly, S. Choi, A. C. Harris, C. Kitko, R. J. Rao, J. E. Levine, V. P. Castle, R. J. Hutchinson, M. Talpaz, D. R. Robinson and A. M. Chinnaiyan, Integrative Clinical Sequencing in the Management of Refractory or Relapsed Cancer in Youth, J. Am. Med. Assoc., 2015, 314, 913–925,  DOI:10.1001/jama.2015.10080 .
  14. E. Schulz, A. Valentin, P. Ulz, C. Beham-Schmid, K. Lind, V. Rupp, H. Lackner, A. Wölfler, A. Zebisch, W. Olipitz, J. Geigl, A. Berghold, M. R. Speicher and H. Sill, Germline mutations in the DNA damage response genes BRCA1, BRCA2, BARD1 and TP53 in patients with therapy related myeloid neoplasms, J. Med. Genet., 2012, 49, 422–428,  DOI:10.1136/jmedgenet-2011-100674 .
  15. L. B. Alexandrov, Y. S. Ju, K. Haase, P. Van Loo, I. Martincorena, S. Nik-Zainal, Y. Totoki, A. Fujimoto, H. Nakagawa, T. Shibata, P. J. Campbell, P. Vineis, D. H. Phillips and M. R. Stratton, Mutational signatures associated with tobacco smoking in human cancer, Science, 2016, 354, 618–622 CrossRef CAS PubMed .
  16. S. Beyaz, M. D. Mana, J. Roper, D. Kedrin, A. Saadatpour, S. J. Hong, K. E. Bauer-Rowe, M. E. Xifaras, A. Akkad, E. Arias, L. Pinello, Y. Katz, S. Shinagare, M. Abu-Remaileh, M. M. Mihaylova, D. W. Lamming, R. Dogum, G. Guo, G. W. Bell, M. Selig, G. P. Nielsen, N. Gupta, C. R. Ferrone, V. Deshpande, G. C. Yuan, S. H. Orkin, D. M. Sabatini and Ö.H Yilmaz, High-fat diet enhances stemness and tumorigenicity of intestinal progenitors, Nature, 2016, 531, 53–58 CrossRef CAS PubMed .
  17. C. Tomasetti and B. Vogelstein, Cancer etiology. Variation in cancer risk among tissues can be explained by the number of stem cell divisions, Science, 2015, 347, 78–81 CrossRef CAS PubMed .
  18. C. Tomasetti, L. Li and B. Vogelstein, Stem cell divisions, somatic mutations, cancer etiology, and cancer prevention, Science, 2017, 355, 1330–1334 CrossRef CAS PubMed .
  19. S. Wu, S. Powers, W. Zhu and Y. A. Hannun, Substantial contribution of extrinsic risk factors to cancer development, Nature, 2015, 592, 43–47 Search PubMed .
  20. C. Tomasetti and B. Vogelstein, Cancer risk: role of the environment - response, Science, 2015, 347, 729–730 CrossRef CAS PubMed .
  21. H. Ledford, US Cancer Institute overhauls cell lines, Science, 2016, 530, 391 Search PubMed .
  22. R. Dienstmann and J. Tabernero, A precision approach to tumour treatment, Nature, 2017, 548, 40–41 CrossRef CAS PubMed .
  23. T. E. Miller, B. B. Liau, L. C. Wallace, A. R. Morton, Q. Xie, D. Dixit, D. C. Factor, L. J. Y. Kim, J. J. Morrow, Q. Wu, S. C. Mack, C. G. Hubert, S. M. Gillespie, W. A. Flavahan, T. Hoffmann, R. Thummalapalli, M. T. Hemann, P. J. Paddison, C. M. Horbinski, J. Zuber, P. C. Scacheri, B. E. Bernstein, P. J. Tesar and J. N. Rich, Transcription elongation factors represent in vivo cancer dependencies in glioblastoma, Nature, 2017, 547, 355–359 CrossRef CAS PubMed .
  24. C. E. Meacham, L. N. Lawton, Y. M. Soto-Feliciano, J. R. Pritchard, B. A. Joughin and T. Ehrenberger, A genome-scale in vivo loss-of-function screen identifies Phf6 as a lineage-specific regulator of leukemia cell growth, Genes Dev., 2015, 29, 483–488,  DOI:10.1101/gad.254151.114 .
  25. M. Uhlen, C. Zhang, S. Lee, E. Sjöstedt, L. Fagerberg, G. Bidkhori, R. Benfeitas, M. Arif, Z. Liu, F. Edfors, K. Sanli, K. von Feilitzen, P. Oksvold, E. Lundberg, S. Hober, P. Nilsson, J. Mattsson, J. M. Schwenk, H. Brunnström, B. Glimelius, T. Sjöblom, P. H. Edqvist, D. Djureinovic, P. Micke, C. Lindskog, A. Mardinoglu and F. Ponten, A pathology atlas of the human cancer transcriptome, Science, 2017, 357, 660 CrossRef CAS PubMed .
  26. S. M. Cohen, The relevance of experimental carcinogenicity studies to human safety, Curr. Opin. Toxicol., 2017, 3, 6–11 CrossRef .
  27. S. M. Cohen and L. L. Arnold, Critical role of toxicologic pathology in a short-term screen for carcinogenicity, J. Toxicol. Pathol., 2016, 29, 215–227 CrossRef PubMed .
  28. A. E. Nel and T. F. Malloy, Policy reforms to update chemical safety testing, Science, 2017, 355, 1016–1018 CrossRef CAS PubMed .
  29. S. L. W. Tan, S. Chadha, Y. Liu, E. Gabasova, D. Perera, K. Ahmed, S. Constantinou, X. Renaudin, M. Lee, R. Aebersold and A. R. Venkitaraman, A Class of Environmental and Endogenous Toxins Induces BRCA2 Haploinsufficiency and Genome Instability, Cell, 2017, 169, 1105–1118.e15,  DOI:10.1016/j.cell.2017.05.010 .
  30. A. Pope, An Essay on Man, in Epistles to a Friend (Epistle II), Printed for J. Wilford, London, 1st edn, 1733 Search PubMed .
  31. C. Berry, Hazardous products: Cut animal wastage in toxicology testing, Nature, 2015, 523, 410 CrossRef CAS PubMed .

This journal is © The Royal Society of Chemistry 2018