The truth about GWAS and the KRAS-variant
Although much of the world has accepted that GWAS studies, which use an outdated technology looking at large amounts of data, have failed to find any functional mutations, this approach has twice been used to try and refute the validity of the KRAS-variant. Please read more to understand the significant flaws with these two studies, and their inapplicability to the KRAS-variant, and all other functional genetic biomarkers.
Inapplicability of GWAS-based Analysis to Functional Genetic Biomarkers with Gene-Environment Interactions
Since its discovery in 2007, the KRAS-variant, an inherited gene mutation found in 6-10% of the population, has been the subject of over 50 publications, representing over 100,000 individuals tested worldwide. The body of international research has identified the KRAS-variant as a functional biological marker that predicts breast and ovarian cancer risk[i][ii][iii][iv][v] and treatment response across numerous cancer types[vi][vii][viii]. Most recently, we completed a large study, paired with biological work, showing that KRAS-variant individuals are affected by estrogen exposure. The findings—demonstrated clinically and biologically—show that abrupt estrogen withdrawal in women with the KRAS-variant increases breast cancer development[v]. In addition to the clinical utility, these findings solidify the role of the KRAS-variant as a modifiable gene mutation and help highlight the gene-environment interaction for this mutation that is now a prominent focus of study within the genetics field.
Since its discovery in 2007, there have been two studies published refuting the association of the KRAS-variant as a predictive genetic marker, specifically for ovarian and breast cancer risk. (It should be noted that neither study has tried to address the KRAS-variant’s widely proven biological function.) The first, in 2011, was written by the Ovarian Cancer Association Consortium (OCAC) and published in Clinical Cancer Research[ix]. Initially, I was a lead author on the study, and had full access to all the case controls that were included in the evaluation. I found numerous concerning shortcomings in the data sets used to evaluate the role of the KRAS-variant in ovarian cancer risk: the exclusion of fallopian cancer, inclusion of over 10% of cases that had not been proven pathologically, the use of biased ascertainment times (e.g., samples collected from women having survived 10 or 20 years post-diagnosis), and inclusion of men as controls. Additionally, the study used a surrogate (or imputed) marker for the KRAS-variant for many of the samples, which we confirmed had up to a 10% error rate in appropriately identifying the sample as positive or negative. Although the lead authors were aware of these shortcomings, they chose to ignore them, which led to my decision to withdraw my name from the study. I communicated these issues post-publication via multiple avenues, including an opinion letter[x] and in direct communications with the FDA, who appeared to agree with our conclusion that GWAS studies, because of these types of flaws, were inappropriate to evaluate clinical utility.
Over the last five years, GWAS studies have come under continued scrutiny for their significant limitations,[xi][xii]which include some of the issues we pointed out in the 2011 commentary. Summarizing criticisms of the approach, Visscher, et al. stated in the American Journal of Human Genetics, “GWAS studies have not delivered meaningful, biologically relevant knowledge or results of clinical or any other utility.”[xiii] Surprisingly, in spite of the known shortcomings in the GWAS approach, in May 2015, the Ovarian Cancer Association Consortium, in partnership with the Breast Cancer Association Consortium, and the Consortium of Modifiers of BRCA1 and BRCA1, published an additional GWAS study, once again refuting the association of the KRAS-variant as a predictor of both breast and ovarian cancer risk[xiv]. While the authors of the 2015 study claimed this new study had been revised to address all of its predecessor’s flaws, it employed almost an identical approach.
Repeated and additional flaws to point out in the new study include:
- Interestingly, the study did find that the KRAS-variant was a significant predictor of several things, including: low-grade serous ovarian cancer (p = 0.031); overall breast cancer risk from BCAC (p=0.043); and breast cancer risk for women under 40 (p=0.029). However, the authors claimed that these findings would “not be significant using Bonferroni correction” in the statistical analysis. Of course, Bonferroni correction is an inappropriate statistical correction to apply to a functional biomarker, and is rarely used outside of pure GWAS studies[xv][xvi]. It is clear that misapplication of such a method would be assured to miss functional biomarkers such as the KRAS-variant, and the suggestion is misguided.
- Although the authors claimed they did not use prevalent cases, the same studies from OCAC that were previously confirmed to contain prevalent cases were included in this study[xvii], indicating that the authors’ statement was false.
- Although the authors claimed they no longer used imputed data, it is clear that KRAS-variant genotyping data was not available from CIMBA ovarian cancer cases[xviii] and is not included in the Illumina platforms stated to be used for several OCAC genome-wide association studies. Again, this indicates that the authors’ statement was false.
- Although the authors claim that they did not find an association between hormone replacement therapy use and breast tumor histology as has been independently shown in a previous study[xix], and which has now been shown clinically and biologically in McVeigh et al., it is the discontinuation of hormone replacement therapy that influences breast tumor biology and increases risk for women with the KRAS-variantv. This information, like most clinical information, was either not available, or simply not evaluated in their current study.
While there was once some epidemiological utility for GWASs, this approach is not a meaningful, nor relevant platform for assessing the clinical utility of a functional biomarker such as the KRAS-variant. It has been known for years that GWAS studies are unable to detect gene-environment interactions[xx]. Given that the KRAS-variant is one of the first mutations shown to have a gene-environment interaction in women’s cancer risk, it is therefore both puzzling and concerning that these GWAS-based groups continue to discount its utility. One could speculate about their motivations, but we will refrain, as we believe that the obvious flaws in their analyses speak for themselves.
Our data, replicated by numerous groups since our original findings, confirms that the KRAS-variant is an important, functional, inherited biomarker with implications for both risk assessment and risk reduction strategies for breast and ovarian cancer as well as to direct therapy broadly across all cancer types.
[i] Ratner, E., L. Lu, M. Boeke, R. Barnett, S. Nallur, L. Chin, C. Pelletier, R. Blitzblau, R. Tassi, T. Paranjape, P. Hui, A. Godwin, H. Yu, H. Risch, T. Rutherford, P. Schwartz, A. Santin, E. Matloff, D. Zelterman, F. Slack and J. Weidhaas (2010). “A KRAS-variant in Ovarian Cancer Acts as a Genetic Marker of Cancer Risk.” Cancer Research 15: 6509-6515.
[ii]Paranjape, T., H. Heneghan, R. Lindner, F. Keane, A. Hoffman, A. Hollestelle, J. Dorairaj, K. Geyda, C. Pelletier, S. Nallur, J. Martens, M. Hooning, M. Kerin, D. Zelterman, Y. Zhu, D. Tuck, L. Harris, N. Miller, F. Slack and J. Weidhaas (2011). “A 3′-untranslated region KRAS variant and triple-negative breast cancer: a case-control and genetic analysis.” Lancet Oncology 12(4): 377-386.
[iii] Ratner, E., F. Keane, R. Lindner, R. Tassi, T. Paranjape, M. Glasgow, S. Nallur, Y. Deng, L. Lu, L. Steele, S. Sand, R. Muller, E. Bignotti, S. Bellone, M. Boeke, X. Yao, S. Pecorelli, A. Ravaggi, D. Katsaros, D. Zelterman, M. Cristea, H. Yu, T. Rutherford, J. Weitzel, S. Neuhausen, P. Schwartz, F. Slack, A. Santin and J. Weidhaas (2011). “A KRAS-variant is a Biomarker of Poor Outcome, Platinum Chemotherapy Resistance and a Potential Target for Therapy in Ovarian Cancer.” Oncogene Dec 5.
[iv] Pilarski, R., D. Patel, J. Weitzel, T. McVeigh, J. Dorairaj, H. Heneghan, N. Miller, J. Weidhaas, M. Kerin, M. McKenna, X. Wu, M. Hildebrandt, D. Zelterman, S. Sand and S. LP (2012). “A KRAS-variant is associated with risk of developing double primary breast and ovarian cancer.” PLos ONE 7(5): e37891.
[v] McVeigh, T., S.-Y. Jung, M. Kerin, D. Salzman, S. Nallur, A. Nemec, M. Dookwah, J. Sadofsky, T. Paranjape, O. Kelly, E. Chan, N. Miller, K. Sweeney, D. Zelterman, J. Sweasy, R. Pilarski, D. Telesca, F. Slack and J. Weidhaas (2015). “Estrogen Withdrawal, increased Breast Cancer Risk and the KRAS-variant.” Cell Cycle in press.
[vi] Graziano, F., E. Canestrari, F. Loupakis, A. Ruzzo, N. Galluccio, D. Santini, M. Rocchi, B. Vincenzi, L. Salvatore, C. Cremolini, C. Spoto, V. Catalano, S. D’Emidio, P. Giordani, G. Tonini, A. Falcone and M. Magnani (2010). “Genetic modulation of the Let-7 microRNA binding to KRAS 3′-untranslated region and survival of metastatic colorectal cancer patients treated with salvage cetuximab-irinotecan.” Pharmacogenomics J 10(5): 458-464.
[vii] Chung, C., J. Lee, R. Slebos, J. Howard, J. M. Perez, H. Kang, E. Fertig, M. Considine, J. Gilbert, B. Murphy, S. Nallur, T. Paranjape, R. JOrdan, J. Garcia, B. Burtness, A. Forastiere and J. Weidhaas (2014). “A 3’UTR KRAS variant is associated with cisplatin resistance in patients with recurrent and/or metastatic head and neck squamous cell carcinoma.” Ann Oncol July 31. [Epub ahead of printing].
[viii] Saridaki, Z., J. Weidhaas, H.-J. Lenz, P. Laurent-Puig, B. Jacobs, J. De Schutter, W. De Roock, D. W. Salzman, W. Zhang, D. Yang, C. Pilati, O. Bouche, H. Piessevaux and S. Tejpar (2014). “A let-7 microRNA-binding site polymophism in KRAS predicts improved outcome in metastatic colorectal cancer (mCRC) patients treated with salvage cetuximab/panitumumab monotherapy.” Clin Cancer Res 20(17): 4499-4510.
[ix] Pharoah, P., et al. (2011). “The Role of KRAS rs61764370 in Invasive Epithelial Ovarian Cancer: Implications for Clinical Testing.” Clin Cancer Res 17(11):3742-50.
[x] Weidhaas, J. and F. Slack (2011), “KRAS rs61764370 in Epithelial Ovarian Cancer – Letter.”Clin Cancer Res, 17: p. 6600.
[xi] Goldstein, D.B. (2009), “Common Genetic Variation and Human Traits.” New England Journal of Medicine, 360(17): p. 1696-1698.
[xii] Faye, L., Machiela, M., Kraft, P., Bull, S., Sun, L. (2013), “Re-Ranking Sequencing Variants in the Post-GWAS Era for Accurate Causal Variant Identification.” PLoS Genet, 9: p. e1003609.
[xiii] Visscher, P., Brown, M., McCarthy, J. (2012), “Five Years of GWAS Discovery”, American Journal of Human Genetics, 90(1): 7–24.
[xv] Benjamini, Y. and Y. Hochberg (1995). “Controlling the False Discovery Rate: A Practical and Powerful Approach to Multiple Testing.” Journal of the Royal Statistical Society 57: 289-300.
[xvi] Perneger, T. (1998). “What’s wrong with Bonferroni adjustments.” BMJ 316: 1236-1238.
[xvii] Hollestelle A., et al., Supplemental Table S1.
[xviii] Hollestelle A., et al., Supplemental Table S7.
[xix] Cerne, J., V. Stegel, K. Gersak and S. Novakovic (2012). “KRAS rs61764370 is associated with HER2-overexpressed and poorly-differentiated breast cancer in hormone replacement therapy users: a case control study.” BMC Cancer 12(105).
[xx] Engelman, Corinne D., Baurley, James W., Chiu, Yen-Feng, Jourbert, Bonnie R., Lewinger, Juan P., Maenner, Matthew J., Murcray, Cassandra E., Shi, Gang, and Gauderman, W. James. (2009), “Detecting Gene-Environment Interactions in Genome-Wide Association Data.” Genet Epidemiol 33 (Suppl 1): S68-S73.