Towards Personalized Cancer Medicine

Journal Articles

Keynote: J. Michael Bishop

Chen GQ, Shi XG, Tang W, et al. 1997. Use of arsenic trioxide (As2O3) in the treatment of acute promyelocytic leukemia (APL): I. As2O3 exerts dose-dependent dual effects on APL cells. Blood 89: 3345-3353.

Goga A, Yang D, Tward AD, et al. 2007. Inhibition of CDK1 as a potential therapy for tumors over-expressing MYC. Nat. Med. 13: 820-827.

Hu J, Liu Y, Wu C, et al. 2009. Long-term efficacy and safety of all-trans retinoic acid/arsenic trioxide-based therapy in newly diagnosed acute promyelocytic leukemia. Proc. Natl. Acad. Sci. USA 106: 3342-3347.

Shen ZX, Chen GQ, Ni JH, et al. 1997. Use of arsenic trioxide (As2O3) in the treatment of acute promyelocytic leukemia (APL): II. Clinical efficacy and pharmacokinetics in relapsed patients. Blood 89: 3354-3360.

Yang D, Liu H, Goga A, et al. 2010. Therapeutic potential of a synthetic lethal interaction between the MYC proto-oncogene and inhibition of aurora-B kinase. Proc. Natl. Acad. Sci. USA 107: 13836-13841.

Yuneva M, Zamboni N, Oefner P, et al. 2007. Deficiency in glutamine but not glucose induces MYC-dependent apoptosis in human cells. J. Cell Biol. 178: 93-105.

Session 1

Bais C, Wu X, Yao J, et al. 2010. PlGF blockade does not inhibit angiogenesis during primary tumor growth. Cell 141: 166-177.

Beroukhim R, Brunet J, Di Napoli A, et al. 2009. Patterns of gene expression and copy-number alterations in von-hippel lindau disease-associated and sporadic clear cell carcinoma of the kidney. Cancer Res. 69: 4674-4681.

Chen L, Yang S, Jakoncic J, et al. 2010. Migrastatin analogues target fascin to block tumour metastasis. Nature 464: 1062-1066.

Ferrara N, Kerbel RS. 2005. Angiogenesis as a therapeutic target. Nature 438: 967-974.

Lallemand-Breitenbach V, Guillemin MC, Janin A, et al. 1999. Retinoic acid and arsenic synergize to eradicate leukemic cells in a mouse model of acute promyelocytic leukemia. J. Exp. Med. 189: 1043-1052.

McDonough MA, Li V, Flashman E, et al. 2006. Cellular oxygen sensing: Crystal structure of hypoxia-inducible factor prolyl hydroxylase (PHD2). Proc. Natl. Acad. Sci. USA 103: 9814-9819.

Minn AJ, Gupta GP, Siegel PM, et al. 2005. Genes that mediate breast cancer metastasis to lung. Nature 436: 518-524.

Nguyen DX, Bos PD, Massagué J. 2009. Metastasis: from dissemination to organ-specific colonization. Nat. Rev. Cancer 9: 274-284.

Padua D, Zhang XH, Wang Q, et al. 2008. TGFbeta primes breast tumors for lung metastasis seeding through angiopoietin-like 4. Cell 133: 66-77.

Rosenfeld PJ, Brown DM, Heier JS, et al. 2006. Ranibizumab for neovascular age-related macular degeneration. N. Engl. J. Med. 355: 1419-1431.

Shojaei F, Wu X, Qu X, et al. 2009. G-CSF-initiated myeloid cell mobilization and angiogenesis mediate tumor refractoriness to anti-VEGF therapy in mouse models. Proc. Natl. Acad. Sci. USA 106: 6742-6747.

Shojaei F, Wu X, Malik AK, et al. 2007. Tumor refractoriness to anti-VEGF treatment is mediated by CD11b+Gr1+ myeloid cells. Nat. Biotechnol. 25: 911-920.

Toschi A, Lee E, Gadir N, Ohh M, Foster DA. 2008. Differential dependence of hypoxia-inducible factors 1 alpha and 2 alpha on mTORC1 and mTORC2. J. Biol. Chem. 283: 34495-34499.

Zhang Q, Gu J, Li L, et al. 2009. Control of cyclin D1 and breast tumorigenesis by the EglN2 prolyl hydroxylase. Cancer Cell 16: 413-424.

Zhang XH, Wang Q, Gerald W, et al. 2009. Latent bone metastasis in breast cancer tied to Src-dependent survival signals. Cancer Cell 16: 67-78.

Session 2

A haplotype map of the human genome. 2005. Nature 437: 1299-1320.

The International HapMap Project. 2003. Nature 426: 789-796.

Brock MV, Hooker CM, Ota-Machida E, et al. 2008. DNA methylation markers and early recurrence in stage I lung cancer. N. Engl. J. Med. 358: 1118-1128.

Cedar H, Bergman Y. 2009. Linking DNA methylation and histone modification: patterns and paradigms. Nat. Rev. Genet. 10: 295-304.

Chantarangsu S, Mushiroda T, Mahasirimongkol S, et al. 2009. HLA-B*3505 allele is a strong predictor for nevirapine-induced skin adverse drug reactions in HIV-infected Thai patients. Pharmacogenet. Genomics 19: 139-146.

Herman JG, Baylin SB. 2003. Gene silencing in cancer in association with promoter hypermethylation. N. Engl. J. Med. 349: 2042-2054.

Ionov Y, Peinado MA, Malkhosyan S, et al. 1993. Ubiquitous somatic mutations in simple repeated sequences reveal a new mechanism for colonic carcinogenesis. Nature 363: 558-561.

Kiyotani K, Mushiroda T, Imamura CK, et al. 2010. Significant effect of polymorphisms in CYP2D6 and ABCC2 on clinical outcomes of adjuvant tamoxifen therapy for breast cancer patients. J. Clin. Oncol. 28: 1287-1293.

Muto M, Nakane M, Hitomi Y, et al. 2002. Association between aldehyde dehydrogenase gene polymorphisms and the phenomenon of field cancerization in patients with head and neck cancer. Carcinogenesis 23: 1759-1765.

Ohm JE, McGarvey KM, Yu X, et al. 2007. A stem cell-like chromatin pattern may predispose tumor suppressor genes to DNA hypermethylation and heritable silencing. Nat. Genet. 39: 237-242.

Ohm JE, Baylin SB. 2007. Stem cell chromatin patterns: an instructive mechanism for DNA hypermethylation? Cell Cycle 6: 1040-1043.

Sharma SV, Lee DY, Li B, et al. 2010. A chromatin-mediated reversible drug-tolerant state in cancer cell subpopulations. Cell 14: 69-80.

Suzuki K, Suzuki I, Leodolter A, et al. 2006. Global DNA demethylation in gastrointestinal cancer is age dependent and precedes genomic damage. Cancer Cell 9: 199-207.

Yamashita K, Dai T, Dai Y, et al. 2003. Genetics supersedes epigenetics in colon cancer phenotype. Cancer Cell 4: 121-131.

Session 3

Bernards R. 2010. It's diagnostics, stupid. Cell 141: 13-17.

Cairo S, Wang Y, de Reyniès A, et al. 2010. Stem cell-like micro-RNA signature driven by Myc in aggressive liver cancer. Proc. Natl. Acad. Sci. USA Nov. 8 [Epub ahead of print]

Dalgliesh GL, Furge K, Greenman C, et al. 2010. Systematic sequencing of renal carcinoma reveals inactivation of histone modifying genes. Nature 463: 360-363.

Hölzel M, Huang S, Koster J, et al. 2010. NF1 is a tumor suppressor in neuroblastoma that determines retinoic acid response and disease outcome. Cell 142: 218-229.

Kumar-Sinha C, Tomlins SA, Chinnaiyan AM. 2008. Recurrent gene fusions in prostate cancer. Nat. Rev. Cancer 8: 497-511.

Kuo W, Das D, Ziyad S, et al. 2009. A systems analysis of the chemosensitivity of breast cancer cells to the polyamine analogue PG-11047. BMC Med. 7: 77.

Mani R, Tomlins SA, Callahan K, et al. 2009. Induced chromosomal proximity and gene fusions in prostate cancer. Science 326: 1230.

Mullenders J, Bernards R. 2009. Loss-of-function genetic screens as a tool to improve the diagnosis and treatment of cancer. Oncogene 28: 4409-4420.

Neve RM, Chin K, Fridlyand J, et al. 2006. A collection of breast cancer cell lines for the study of functionally distinct cancer subtypes. Cancer Cell 10: 515-527.

Pekarsky Y, Croce CM. 2010. Is miR-29 an oncogene or tumor suppressor in CLL? Oncotarget 1: 224-227.

Rao X, Di Leva G, Li M, et al. 2010. MicroRNA-221/222 confers breast cancer fulvestrant resistance by regulating multiple signaling pathways. Oncogene Nov. 8 [Epub ahead of print]

Santanam U, Zanesi N, Efanov A, et al. 2010. Chronic lymphocytic leukemia modeled in mouse by targeted miR-29 expression. Proc. Natl. Acad. Sci. USA 107: 12210-12215.

Tomlins SA, Rhodes DR, Perner S, et al. 2005. Recurrent fusion of TMPRSS2 and ETS transcription factor genes in prostate cancer. Science 310: 644-648.

Yu J, Mani R, Cao Q, et al. 2010. An integrated network of androgen receptor, polycomb, and TMPRSS2-ERG gene fusions in prostate cancer progression. Cancer Cell 17: 443-454.

Session 4

Blows FM, Driver KE, Schmidt MK, et al. 2010. Subtyping of breast cancer by immunohistochemistry to investigate a relationship between subtype and short and long term survival: a collaborative analysis of data for 10,159 cases from 12 studies. PLoS Med. 7: e1000279.

Bueno-de-Mesquita JM, van Harten WH, Retel VP, et al. 2007. Use of 70-gene signature to predict prognosis of patients with node-negative breast cancer: a prospective community-based feasibility study (RASTER). Lancet Oncol. 8: 1079-1087.

Carey LA, Perou CM, Livasy CA, et al. 2006. Race, breast cancer subtypes, and survival in the Carolina Breast Cancer Study. JAMA 295: 2492-2502.

Carro MS, Lim WK, Alvarez MJ, et al. 2010. The transcriptional network for mesenchymal transformation of brain tumours. Nature 463: 318-325.

Huo D, Ikpatt F, Khramtsov A, et al. 2009. Population differences in breast cancer: survey in indigenous African women reveals over-representation of triple-negative breast cancer. J. Clin. Oncol. 27: 4515-4521.

Lim E, Vaillant F, Wu D, et al. 2009. Aberrant luminal progenitors as the candidate target population for basal tumor development in BRCA1 mutation carriers. Nat. Med. 15: 907-913.

Mehrabian M, Allayee H, Stockton J, et al. 2005. Integrating genotypic and expression data in a segregating mouse population to identify 5-lipoxygenase as a susceptibility gene for obesity and bone traits. Nat. Genet. 37: 1224-1233.

Parker JS, Mullins M, Cheang MCU, et al. 2009. Supervised risk predictor of breast cancer based on intrinsic subtypes. J. Clin. Oncol. 27: 1160-1167.

Pei X, Bai F, Smith MD, et al. 2009. CDK inhibitor p18(INK4c) is a downstream target of GATA3 and restrains mammary luminal progenitor cell proliferation and tumorigenesis. Cancer Cell 15: 389-401.

Prat A, Parker JS, Karginova O, et al. 2010. Phenotypic and molecular characterization of the claudin-low intrinsic subtype of breast cancer. Breast Cancer Res. 12: R68.

Raouf A, Zhao Y, To K, et al. 2008. Transcriptome analysis of the normal human mammary cell commitment and differentiation process. Cell Stem Cell 3: 109-118.

Schadt EE, Molony C, Chudin E, et al. 2008. Mapping the genetic architecture of gene expression in human liver. PLoS Biol. 6: e107.

Speliotes EK, Willer CJ, Berndt SI, et al. 2010. Association analyses of 249,796 individuals reveal 18 new loci associated with body mass index. Nat. Genet. 42: 937-948.

Teschendorff AE, Naderi A, Barbosa-Morais NL, et al. 2006. A consensus prognostic gene expression classifier for ER positive breast cancer. Genome Biol. (10): R101.

van't Veer LJ, Bernards R. 2008. Enabling personalized cancer medicine through analysis of gene-expression patterns. Nature 452: 564-570.

van 't Veer LJ, Dai H, van de Vijver MJ, et al. 2002. Gene expression profiling predicts clinical outcome of breast cancer. Nature 415: 530-536.

Watters JW, Cheng C, Majumder PK, et al. 2009. De novo discovery of a gamma-secretase inhibitor response signature using a novel in vivo breast tumor model. Cancer Res. 69: 8949-8957.

Weigelt B, Hu Z, He X, et al. 2005. Molecular portraits and 70-gene prognosis signature are preserved throughout the metastatic process of breast cancer. Cancer Res. 65: 9155-9158.

Zhu J, Zhang B, Smith EN, et al. 2008. Integrating large-scale functional genomic data to dissect the complexity of yeast regulatory networks. Nat. Genet. 40: 854-861.

Session 5

Audeh MW, Carmichael J, Penson RT, et al. 2010. Oral poly(ADP-ribose) polymerase inhibitor olaparib in patients with BRCA1 or BRCA2 mutations and recurrent ovarian cancer: a proof-of-concept trial. Lancet 376: 245-251.

Ashworth A, Bernards R. 2010. Using functional genetics to understand breast cancer biology. Cold Spring Harb Perspect Biol. 2: a003327.

Banerjee S, Kaye SB, Ashworth A. 2010. Making the best of PARP inhibitors in ovarian cancer. Nat. Rev. Clin. Oncol. 7: 508-519.

Farmer H, McCabe N, Lord CJ, et al. 2005. Targeting the DNA repair defect in BRCA mutant cells as a therapeutic strategy. Nature 434: 917-921.

Fong PC, Yap TA, Boss DS, et al. 2010. Poly(ADP)-ribose polymerase inhibition: frequent durable responses in BRCA carrier ovarian cancer correlating with platinum-free interval. J. Clin. Oncol. 28: 2512-2519.

Fong PC, Boss DS, Yap TA, et al. 2009. Inhibition of poly(ADP-ribose) polymerase in tumors from BRCA mutation carriers. N. Engl. J. Med. 361: 123-134.

Martin SA, McCarthy A, Barber LJ, et al. 2009. Methotrexate induces oxidative DNA damage and is selectively lethal to tumour cells with defects in the DNA mismatch repair gene MSH2. EMBO Mol. Med. 1: 323-337.

Mendes-Pereira AM, Martin SA, Brough R, et al. 2009. Synthetic lethal targeting of PTEN mutant cells with PARP inhibitors. EMBO Mol. Med. 1: 315-322.

Shen WH, Balajee AS, Wang J, et al. 2007. Essential role for nuclear PTEN in maintaining chromosomal integrity. Cell 128: 157-170.

Turner N, Pearson A, Sharpe R, et al. 2010. FGFR1 amplification drives endocrine therapy resistance and is a therapeutic target in breast cancer. Cancer Res. 70: 2085-2094.

Session 6

Bardeesy N, DePinho RA. 2002. Pancreatic cancer biology and genetics. Nat. Rev. Cancer 2: 897-909.

Fischer T, Stone RM, Deangelo DJ, et al. 2010. Phase IIB trial of oral Midostaurin (PKC412), the FMS-like tyrosine kinase 3 receptor (FLT3) and multi-targeted kinase inhibitor, in patients with acute myeloid leukemia and high-risk myelodysplastic syndrome with either wild-type or mutated FLT3. J. Clin. Oncol. 28: 4339-4345.

Hodi FS, Friedlander P, Corless CL, et al. 2008. Major response to imatinib mesylate in KIT-mutated melanoma. J. Clin. Oncol. 26: 2046-2051.

Hodi FS, O'Day SJ, McDermott DF, et al. 2010. Improved survival with ipilimumab in patients with metastatic melanoma. N. Engl. J. Med. 363: 711-723.

Jones S, Zhang X, Parsons DW, et al. 2008. Core signaling pathways in human pancreatic cancers revealed by global genomic analyses. Science 321: 1801-1806.

Kharas MG, Lengner CJ, Al-Shahrour F, et al. 2010. Musashi-2 regulates normal hematopoiesis and promotes aggressive myeloid leukemia. Nat. Med. 16: 903-908.

Lee JY, Nakada D, Yilmaz OH, et al. 2010. mTOR activation induces tumor suppressors that inhibit leukemogenesis and deplete hematopoietic stem cells after Pten deletion. Cell Stem Cell 7: 593-605.

Paez JG, Jänne PA, Lee JC, et al. 2004. EGFR mutations in lung cancer: correlation with clinical response to gefitinib therapy. Science 304: 1497-1500.

Sebolt-Leopold JS. 2008. Advances in the development of cancer therapeutics directed against the RAS-mitogen-activated protein kinase pathway. Clin. Cancer Res. 14: 3651-3656.

Serra V, Markman B, Scaltriti M, et al. 2008. NVP-BEZ235, a dual PI3K/mTOR inhibitor, prevents PI3K signaling and inhibits the growth of cancer cells with activating PI3K mutations. Cancer Res. 68: 8022-8030.

Wang R, Ferrell LD, Faouzi S, et al. 2001. Activation of the Met receptor by cell attachment induces and sustains hepatocellular carcinomas in transgenic mice. J. Cell Biol. 153: 1023-1034.

Zhang L, Lee KC, Bhojani MS, et al. 2007. Molecular imaging of Akt kinase activity. Nat. Med. 13: 1114-1119.

Technology Workshop

Alix-Panabières C, Riethdorf S, Pantel K. 2008. Circulating tumor cells and bone marrow micrometastasis. Clin. Cancer Res. 14: 5013-5021.

Nguyen QT, Olson ES, Aguilera TA, et al. 2010. Surgery with molecular fluorescence imaging using activatable cell-penetrating peptides decreases residual cancer and improves survival. Proc. Natl. Acad. Sci. USA 107: 4317-4322.

Pantel K, Alix-Panabières C. 2010. Circulating tumour cells in cancer patients: challenges and perspectives. Trends Mol. Med. 16(9): 398-406.

Olson ES, Jiang T, Aguilera TA, et al. 2010. Activatable cell penetrating peptides linked to nanoparticles as dual probes for in vivo fluorescence and MR imaging of proteases. Proc. Natl. Acad. Sci. USA 107: 4311-4316.

Schummer M, Green A, Beatty JD, et al. 2010. Comparison of breast cancer to healthy control tissue discovers novel markers with potential for prognosis and early detection. PLoS ONE 5: e9122.

Sponsors

Presented by

Gold Sponsors

Silver Sponsors

Bronze Sponsors

Cepheid

Ferrer InCode

IkerChem S.L.

LIFE Biosystems AG

Millennium Pharmaceuticals - The Takeda Oncology Company

Novartis

Academy Friends

Affymetrix

Agilent Technologies

Bayer Schering Pharma

Biocat (The BioRegion of Catalonia)

Catalan Institute of Oncology

Cell Signaling Technology, Inc.

Laboratorio de Análisis Dr. Echevarne

Pangaea Biotech S.A.

USP Instituto Universitario Dexeus

Grant Support

This activity is supported by an educational grant from ImClone Systems, a wholly-owned subsidiary of Eli Lilly and Company.

The project described was supported by Award Number R13CA144428 from the National Cancer Institute, and the National Human Genome Research Institute.  The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Cancer Institute or the National Institutes of Health.

Accreditation

This event was accredited with 18 ESMO-MORA points Category 1. For more information about ESMO-MORA points, please visit the ESMO website.

Personalized medicine, the treatment of patients based upon their individual genetic, epigenetic, and phenotypic makeup, is the ultimate goal of many researchers and clinicians trying to find less toxic and more effective therapies for cancer. While cancer as a whole is characterized by uncontrollably proliferating cells, the disease is being subdivided into increasingly smaller units, from organ-specific to gene-specific categories. An understanding of the molecular basis of some cancers and an increasing knowledge of how those cancers differ from those in other organs and in other individuals have led to the development of drugs enormously successful at treating those types of cancer. Yet many other candidate drugs based on the same kind of analysis fail in the clinic, leaving cytotoxic chemotherapy still the standard treatment for many patients.

From May 19–21 2010, academic and industry researchers, technology developers, and clinicians from around the world gathered in Barcelona, Spain, to discuss the progress and the challenges in the field of personalized cancer medicine. Presented by the New York Academy of Sciences, Taléncia, and Fundació la Caixa, the conference brought together over 300 attendees to learn about topics ranging from basic cell biology and how it goes awry in cancer cells to massive efforts to characterize cancer genomes. A special technology workshop highlighted the important role of technological advances in driving the efforts toward personalized medicine. Conference participants also discussed their concerns about the field, in particular the failure of the U.S. clinical trial system to keep up with the changes demanded by the new approach.

Researchers have made great progress in discovering the cellular processes that have become pathological in cancer. For example, it is clear that genetic and epigenetic changes that affect cell proliferation, DNA repair, programmed cell death, stem cell maturation or self-renewal, and other fundamental cellular and molecular activities can lead to cancer formation. New information is also coming to light about more systemic problems, such as the survival of cancer cells in different microenvironments, the cells' ability to stimulate angiogenesis, and the way they promote their own survival by interacting with noncancerous cells to acquire nutrients, suppress immune response, and more. Several speakers at the conference shared their research into these and other biological events that lead to cancer.

Post-diagnostic treatments are clearly important, but the best way to deal with any disease is to prevent it from ever occurring. To that end, scientists and clinicians are working together to find ways to determine whether a person is more or less likely to develop cancer and why. To associate genetic status with cancer susceptibility, researchers need to analyze large amounts of clinical and molecular data, sometimes collected over a lengthy clinical follow-up process. Japan has taken up this mantle with its BioBank Japan project, in which researchers and clinicians are collecting DNA, serum, and clinical information from 310,000 patients with common diseases, including some types of cancer. The information will be used to investigate genetic susceptibility to disease, molecular targets for drug discovery, genetic information for personalized medicine, and more.

The processes that lead to epigenetic changes, such as methylation, acetylation, and changes in chromatin structure, are less well understood than those that lead to DNA mutations, but are likely to be just as important in determining cancer susceptibility. Gene expression is highly dependent upon the physical state of the genes' promoter regions, which in turn is regulated by epigenetic mechanisms. Overexpression of genes that lead to cell proliferation or under expression of those that promote terminal cell differentiation, as well as other alterations in gene expression, can give rise to tumors. Speakers in the second session of the conference provided insight into the specific ways genetic and epigenetic changes can be linked to cancer development.

The third session of the conference focused on oncogenomics and tumor profiling. Researchers are using whole genome sequencing of tumors to find the underlying mutations. This genomic information can reveal the signaling pathways mutated in a particular tumor type as well as the variability among tumors within that type. In addition to chromosomes with point mutations and deletions, cancer cells are characterized by a large number of genomic rearrangements that can result in gene fusions and consequent oncogenic proteins. Special techniques have been developed to identify these rearranged chromosomes and to pinpoint the fusion genes that result. Some of the rearrangements affect microRNAs, which as a molecular class went unnoticed until 2001 when the human counterparts of the C. elegans versions were located. It may be possible to design therapies targeting microRNAs in oncogenic signaling pathways, as they can be downstream targets of processes that are often dysregulated in cancer.

While basic research is revealing new targets for cancer drugs, treatment currently consists of some combination of neoadjuvant therapy, surgery, and chemotherapy or targeted therapy. Because of the variety of treatment options, many of which are expensive and quite toxic, it is critical to determine which treatment is most likely to work for each patient. Breast cancer researchers are at the forefront of efforts to use gene expression signatures and other molecular data to guide treatment, predict risk, and understand epidemiological phenomena. There are several clinical trials underway to test some of these new tools, some of which were presented in the fourth session's talks about efforts to identify biomarkers for prognosis and treatment response.

The fifth and sixth sessions of the conference delved into a variety of approaches to developing targeted therapies for cancer. Some of these efforts take advantage of technologies that allow high-throughput analyses of genomic information and drug responses in cell culture. Others use preclinical mouse models to evaluate candidate target therapies or to counteract the drug resistance that often thwarts cancer treatment. Next-generation sequencing and microarray technology have also played an important part in the advance toward personalized cancer medicine, as a technology workshop highlighted.

Although the conference participants covered a broad range of topics in cancer medicine, three ideas kept coming to the fore in their talks: first, research is moving into a new era in which very large amounts of data are being generated and efforts must be made to share this information across research groups in academia and industry; second, the current clinical trial system in the United States is not keeping pace with changes in drug discovery; and third, as treatments become more personalized, the pharmaceutical industry, regulatory agencies, and patients will have to adjust to the fact that the number of people treated with any given drug will be much smaller than in the past.

The clinical trial system in the United States is regulated by the U.S. Food and Drug Administration. There are several features of the current system that need to be changed to enable personalized medicine to reach its full potential. For example, the current system prohibits the testing of combination therapies before each drug has been proven efficacious on its own. However, a growing understanding of the molecular mechanisms of drug action suggests that some drugs may only work in combination with others, and will fail in single drug clinical trials. In other instances, a drug's efficacy may depend on its administration early in the disease process, but clinical trials often test new drugs on late stage cancer patients. Finally, for targeted therapies, the patient population will need to be stratified, making it difficult to test a new compound in the number of patients currently required for a drug efficacy to be established. The clinical trial system will need to be modernized to account for these changes in drug development.

While cancer remains a challenging disease to treat, the advent of technologies and data processing tools that allow for the gathering and analysis of large amounts of information, the proven success of some personalized therapies, and the sheer breadth of cancer research efforts suggests that, from diagnosis to treatment, clinicians will have new and powerful ways to tackle the disease in the future.

Speaker:
J. Michael Bishop, University of California, San Francisco

Highlights

• Sequencing of cancer genomes will identify all the common mutations in major cancer types.
• Bimodal attack on cancer targets reduces the likelihood of drug resistance.
• Synthetic lethal approaches identify drugs that only kill cells harboring a preexisting aberration, such as overexpression of the MYC oncogene.

Report from the front lines

Nobel laureate J. Michael Bishop opened the meeting by presenting contrasting media reports of progress in cancer research since U.S. President Richard Nixon declared war on the disease in 1971. Explaining that cancer is a far more complex disease than another prevalent disease, cardiovascular disease, and therefore cannot be expected to yield to treatment as readily, Bishop nonetheless argued that cancer research has come a long way over the years and that recent advances hold great promise for future therapies.

One major development is the advent of inexpensive, rapid sequencing techniques. Scientists are adding to the current understanding of the genetic mutations and genome alterations that can give rise to cancer with new efforts to identify all the common mutations in major cancers through complete sequencing of tumor genomes. This is a massive undertaking because identifying genes mutated in 5% of a cancer types will require more than 100 tumor genome sequences. It is difficult, but it is possible, said Bishop.

Genomic analysis has already yielded some emerging principles about the genetic landscape underlying the pathology. First, individual cancers can contain from dozens to hundreds of genetic lesions. Second, each type of cancer has a distinctive, but not unique set of malfunctioning genes, called a genetic fingerprint, which can vary appreciably among tumors of the same type. And finally, the many malfunctioning genes are responsible for a more limited number of signaling pathways.

Combination therapy and synthetic lethal approaches

The question then becomes: once scientists have identified a cancer-inducing mutation, how can they develop a therapy to specifically target the cancer-causing agent? Bishop presented one example, the cure of acute promyelocytic leukemia (APL), a cancer caused by the fusion of the retinoic acid receptor alpha gene with a previously unknown gene. Scientists in China initially discovered that all-trans retinoic acid could cause a temporary remission of the cancer when used in conjunction with cytotoxic therapy. Further investigation indicated that addition of arsenic trioxide, the active ingredient of a Chinese traditional medicine called Ai-Lin, led to complete remission of the disease in 95% of patients. The action of each of the components of the therapy explains why it works: the retinoic acid targets the retinoic acid receptor alpha component of the fusion protein, the arsenic trioxide targets the APL component, and both promote destruction of the protein. This bimodal attack on a single target has a remarkable outcome that reduces the effect of single drug resistance.

In many cases, the cancer-causing protein may not be a good target for therapies, perhaps because overexpression of the wild-type protein, rather than an aberrant one, is the cause of neoplastic transformation and tumors, as is the case for the MYC oncoprotein. Wild-type proteins may also be essential for cell survival, which would make them poor therapy targets. This is where a synthetic lethal approach may be very effective. Synthetic lethality was first described as a phenomenon whereby two mutations together cause lethality although individually they are not detrimental to cell survival. Starting from the premise that the oncogene is the first mutation, researchers realized that they could mimic the effect of a second mutation with chemical agents that, in combination with that oncogene, lead to lethality while normal cells remain viable.

Bishop presented three examples of this research approach, starting with the challenge of MYC overexpression. His postdoctoral student Andrei Goga found that the drug purvalanol, developed in Peter Schultz's lab to inhibit the cyclin-dependent kinase Cdk1, caused cell death in a variety of cells lines in which MYC is overexpressed, but not in normal cells where it caused reversible cell cycle arrest. Results of preclinical models suggest that such a therapy may be effective.

Similar results were found with a drug (VX-680) that inhibits the Aurora kinase family, which is required for chromosome segregation and cytokinesis (Figure 1). Cells overexpressing MYC were killed whereas normal cells arrested their cell cycle. Interestingly, Dun Yang found that even after the drug was removed, cells continued to die and that this death was a result of autophagy rather than of the apoptosis that characterizes the acute response. In MYC-driven B- and T-cell lymphoma mouse models there was a 3-fold extension of lifespan. In addition, while cancer cells can become resistant to death by apoptosis by mutations in the cell death pathway, no resistance to death by autophagy has appeared yet, though results are still preliminary.

Figure 1. Diagram of the effects of an aurora kinase inhibitor on normal proliferating cells (left) and on MYC-overexpressing cells. In addition to arresting the cell cycle as it does in normal cells, VX-680 also causes apoptosis and autophagy in the mutated cells.

Bishop's final example of synthetic lethality was the examination of glutamine metabolism in cell lines overexpressing different oncogenes. Different tumor types vary in their glutamine usage depending on the expression of certain transporters and enzymes. Some cancer cells overexpressing MYC are heavily dependent on glutamine. Omission of glutamine from the media feeding these cells results in 80% lethality, whereas in cultures of cells that do not overexpress the oncogene, only 10% of the cells die. A drug that inhibits glutaminase may be a useful treatment for cancers caused by MYC overexpression.

Investigations into individual oncogenes and tumor types have served as a proof-of-concept for the usefulness of synthetic lethality in drug development. And the promising results of these studies have prompted the use of RNA interference for genome-wide inquiries into synthetic lethal interactions.

The future of cancer research

Bishop concluded with an optimistic view of the future of cancer research. Targeted combination therapies show promise for preventing the drug resistance that often leads to cancer treatment failure. Personalized therapies on the horizon include the development of tests to determine whether a patient will respond to a drug and the ability to spot resistance early on. Clinical trials may yield better results if the trial population can be defined by the presence of the drug target. Cohorts can be smaller, less expensive, and more readily assembled. Responses may be evaluated more quickly and trials adapted along the way.

As with all paradigm changes, the shift from a non-personalized clinical trial model to a genetically defined, personalized test population brings challenges. Furthermore, fractionation of the drug market by the treatment of only those patients with certain mutations may be a deterrent to the pharmaceutical industry and lead to more expensive drugs. In addition, FDA regulation of clinical trials in the U.S. has not kept pace with the developments in personalized medicine. The FDA has a "one drug at a time" rule that prevents the testing of combination therapy involving two unapproved drugs and virtually guarantees failure of drugs known to work only in combination with another. Nevertheless both industry and government recognize that the landscape is changing, and they are working to adapt to the new and exciting drug discovery environment.

Speakers:
Hans Clevers, Hubrecht Institute in Utrecht
William G. Kaelin, Jr., Harvard Medical School
Napoleone Ferrara, Genentech
Joan Massagué, Memorial Sloan-Kettering Cancer Center; Institute for Research in Biomedicine, Barcelona (Adjunct)

Highlights

• Crypt base columnar cells give rise to large tumors when mutated and appear to comprise the intestinal stem cell population.
• The dioxygenase family of enzymes may provide a new target for drug discovery.
• G-CSF, a myeloid growth factor secreted by tumor cells, upregulates expression of the angiogenesis-promoting the Bv8 gene in myeloid cells.
• Src-pathway activation is associated with late-onset bone metastasis of breast cancer.

Research into new cancer therapies still relies heavily on an understanding of the basic mechanisms of disease development and progression. The first session of the conference featured a variety of investigations ranging from the role of stem cells in colon cancer to the genetics and cell biology of metastasis, all of them exploring in some way the basis of oncogenesis and metastasis.

Covering cancer's genesis, the proliferation of cancer cells, their acquisition of nutrients, and cancer's deadly spread from one organ to another, speakers considered their research in the broader context of research into both fundamental cancer properties and the application of that knowledge in the search for applicable clinical solutions. Hans Clevers's talk elucidated the relationship between normal processes of regeneration and pathological regeneration: between intestinal cell renewal and intestinal cancer. Following the trail of cancer's impact on bodily functions, William Kaelin and Napoleone Ferrara both talked of tumors' seemingly intractable path to access the nutrients they need. Driven to seek those nutrients, tumors might also be driven by genetic and epigenetic markers to metastasize to specific organs, even years after their first appearance, as Joan Massagué explained.

The self-renewing intestine

Hans Clevers of the Hubrecht Institute in Utrecht described his group's efforts to link the physiology of intestinal self-renewal with the biology of colon cancer. The intestine is one of the most rapidly self-renewing tissues in the both adult humans and mice.

The source of the new cells is a small compartment called a crypt that lies between the villi of the small intestine. There are approximately 10 billion crypts in the intestine, each of which has from 6–10 stem cells that generate about 200 cells, called transit amplifying cells, every day. These daughter cells proliferate for 2–3 days, undergoing 5 cell cycles before differentiating and finally, dying at the tips of the villi 2–3 days later. Before Clevers's work scientists knew of 4–6 stem cells residing at the base of crypts, but they did not know precisely which crypt cells were these stem cells.

Proliferation of cells in self-renewing intestinal tissue is regulated by the Wnt signaling pathway. In colon cancer, the Wnt pathway is locked in the ON state. Stimulation of the Wnt signaling pathway in intestinal cells ultimately activates the TCF4 transcription factor, whose targets are all cancer genes. Many of these genes are expressed in the transit-amplifying cells, and others are expressed in the Paneth cells, a group of cells derived from the stem cells that are thought to sterilize the lumina of the crypts.

Lgr5 expression marks putative stem cells

To identify the intestinal stem cells, Clevers's team looked for genes expressed only in a few cells near the bottom of the crypts. They found that the Lgr5 gene was expressed in cycling crypt base columnar (CBC) cells, a small group of cells located between the Paneth cells that were first identified by Charles Philippe Leblond in 1974 and that subsequently fell into obscurity. Lgr5 encodes an orphan, G-protein coupled receptor related to the glycoprotein hormone receptors.

A number of experiments were performed to look for stem cell properties in Lgr5-expressing cells. A lineage-tracing experiment demonstrated that they produce the right number of daughter cells. Also, the marker for putative stem cells and their daughters persisted through the lifetime of the mouse, as expected. Multicolor lineage tracing showed that 10 individual stem cells were active at the same time in the crypt. Some crypts became clonal for a while then the competition started over. Based on these findings, the Lgr5-expressing CBC cells appear to comprise the intestinal stem cell population (Figure 2). Curiously, these cells are never quiescent, a property thought to be inherent in stem cells.

Figure 2. Lineage tracing of Lgr5-expressing crypt-base columnar (CBC) cells. The CBC cells give rise to all the requisite cell types of the epithelium.

Clevers's team next showed that single CBC cells cultured in matrigel culture with three growth factors can form crypt structures with all the requisite cell types. An unusual property of this system is that the Paneth cells appear to provide the niche that supports the stem cells although the Paneth cells are their daughter cells. Genetic manipulations that cause the gradual loss of Paneth cells result in the disappearance of the stem cells.

Relating his work on intestinal stem cells to cancer, Clevers described the results of deleting the APC (Adenomatous polyposis coli) gene, which plays an inhibitory role in the Wnt pathway and is often mutated in colon cancer, either in intestinal stem cells or in other intestinal cells. Transgenic mice with APC deleted from the stem cells rapidly developed large tumors. In contrast, mice with APC deleted from anywhere but the stem cells developed tiny adenomas that failed to grow any larger. Taken together these data support the hypothesis that the columnar base crypt cells are the stem cells in the intestine and that mutations that affect proliferation of these cells may play a role in the development of colon cancer.

New cancer targets from studies of the VHL tumor suppressor protein

The growth of tumors is limited by their inability to access nutrients and oxygen. Tumors overcome these limitations both by stimulating the growth of blood vessels and by activating a host of genes, via the hypoxia-inducible factor alpha (HIFα) transcription factors, that promote survival in hypoxic conditions. William Kaelin and his team at Harvard Medical School are teasing out the intricate pathway by which the HIFα transcription factors contribute to tumor survival. HIFα is targeted for degradation by the VHL tumor suppressor protein, whose mutation is a common early event in colon cancer. In the presence of oxygen, VHL serves as an ubiquitin ligase that targets the alpha subunit of HIFα. The oxygen sensitivity depends on the fact that HIFα has to undergo an oxygen-dependent posttranslational modification to be recognized by VHL. The main enzyme that modifies HIFα for this recognition is known as EglN1 (or PHD2), a prolyl hydroxylase.

Figure 3. The druggable pocket of EglN1 (PHD2). (Adapted from McDonough et al. 2006 Proc. Natl. Acad. Sci. 103(26): 9814-9819.)

Transcription factors are notoriously difficult to target with drugs because of their structure. By contrast, enzymes are often excellent targets, and EglN1 is no exception. EglN1 crystallizes as a homotrimer with a pocket that provides a possible inhibitor target site (Figure 3). Interestingly, EglN1 is a member of a larger enzyme family that includes EglN2, which is a protein induced in estrogenic stimulation of breast cancer. Knockdown of EglN2 leads to loss of cyclin D1, part of cell cycle progression, and EglN2 shRNA inhibits cancer cell proliferation. Taken together, these results provide evidence that this enzyme family may be a good target for new cancer drugs.

Myeloid cells and angiogenesis

Although tumors can turn on genes that help them survive in hypoxic conditions, ultimately they need a source of food and oxygen to survive. To get these resources, they secrete factors that promote angiogenesis. In 1989, Napoleone Ferrara and his team at Genentech identified and cloned the vascular endothelial cell growth factor gene (VEGF), which encodes a growth factor that plays a key role in promoting new blood vessel growth. The team later developed a monoclonal antibody, now available clinically, that targets VEGF. But as is the case in so many attempts to treat cancer, the patient eventually develops resistance to the treatment and succumbs to the disease. Ferrara's team is now trying to identify the distinguishing characteristics of tumors that are resistant to VEGF inhibition.

Investigations into the tumor environment have indicated that myeloid cells in the area also promote angiogenesis. To investigate whether recruitment of myeloid cells is a factor in resistance to VEGF inhibitors, Ferrara's team tested whether refractory tumors recruited more myeloid cells than sensitive, treatment responsive, cells. They transplanted GFP-labeled bone marrow cells from transgenic mice into lethally irradiated mice, then implanted either sensitive or resistant tumor cells into the chimeric mice. The group found that refractory cells did recruit more GFP-labeled cells and that the majority of cells associated with the refractory tumors consisted of CD11b⁺Gr1⁺ myeloid cells, which encompasses a broad population of immune cells. Ferrara noted that evidence suggests that these myeloid cells are both proangiogenic and immunosuppressive.

Analysis of gene expression in bone marrow from mice with resistant or sensitive tumors revealed a large number of differentially expressed genes, including growth factors and genes involved in mobilizing myeloid cells. Ferrara's team chose to focus on one such gene called Bv8, which has tissue-specific angiogenic properties.

Figure 4. In refractory cells, those insensitive to VEGF-A inhibition, Bv8 upregulation promotes angiogenesis when the VEGF pathway is blocked.

Ferrara's group found that tumors that are highly refractory to VEGF inhibitors produce G-CSF, a myeloid growth factor, which in turn dramatically upregulates expression of Bv8 in myeloid cells. Upregulation of Bv8 is a work-around for the blocked VEGF pathway (Figure 4). "G-CSF is highly likely to mediate communication between the tumor and bone marrow because it behaves like a classic hormone," said Ferrara. It is secreted into the blood, unlike VEGF, which binds tightly to heparin and is confined to the tumor microenvironment. Preliminary experiments show that early anti-Bv8 treatment inhibits the growth of human xenografts in mice though not as effectively as anti-VEGF treatment.

Resistance to a drug usually arises when the cells manage to restore the pathway that is blocked or to find an alternate route to bypass the drug's effects. The identification of such alternate routes gives drug developers a way to head off resistance and improve treatment regimens.

Understanding metastasis

Oncologists and their patients hope that the excision of a tumor before it has a chance to metastasize will be the end of the cancer. But even a clean excision is no guarantee that the cancer will not recur. A better understanding of the process of metastasis is crucial for the development of more effective cancer treatments for before and after the excision. Joan Massagué and his colleagues at Memorial Sloan-Kettering Cancer Center are trying to understand how the cells move from the circulation into a new tissue, survive in that microenvironment, and retain the ability to reinitiate tumors after they are established.

Different cancers have very different paths to recurrence. In breast adenocarcinoma, it may be decades from the time the initial tumor was excised to when the cancer returns. The metastatic cells that are derived from the initial tumors appear to be dormant over the years. When they do reappear, they generally have taken up residence in the lung, bone, or brain. In contrast, lung or pancreatic adenocarcinomas have very short latency periods and there is not much selectivity in which tissues they colonize.

One approach to studying metastasis is to examine the metastatic end-products. Massagué's group showed that metastatic cells from breast cancer patients tend to establish themselves in the same organ when transferred from one mouse to another. Furthermore they showed that cells with a predilection for one organ over another preexisted in the initial population.

The next step was to determine what was different about the type of metastatic cells that colonized one organ versus another. In a study of metastasis to lung tissue, they identified a set of 18 genes that predict relapse from lung infiltration and are found in the primary tumors of ER negative breast cancer patients. Among the genes, the group found the expected genes promoting tumor angiogenesis, vascular permeability, cell mobility, invadopodia, and collagenase production, but they also found genes that are expressed in response to signals in the tumor microenvironment. One such gene ANGPTL4 (angiopoietin-like 4) encodes a protein that disrupts the microvascular endothelial cell junctions, allowing the cells to move from blood vessels into the lung tissue. Interestingly, ANGPTL4 expression is driven by TGF-β released by the stroma in the primary tumor, paving the way for invasion when the cells arrive in the lung.

By looking at the expression of genes in metastatic tumors, Massagué's group is mapping out the steps required to establish tumors in specific organs. But this approach does not allow them to answer the question of what keeps the cells alive for such a long time before relapse. To investigate this issue, the team took another approach, correlating the expression levels in primary tumors of genes in the major growth and survival signaling pathways (TGF-β, Src, Ras, Wnt, etc.) with late-onset relapse. The Src pathway was the only one that was associated with metastasis to bone. Furthermore, they found that stromal cells in the bone secrete two factors that promote long-term survival of tumor cells with a high level of Src activity.

So, if highly Src-active tumor cells are associated with metastasis to bone and are supported for the long term by bone cells, can Src inhibitors prevent metastasis in high risk patients? Preliminary experiments in vitro (Figure 5) and in mouse models indicate that the approach may work in cancers that have a short latency period as long as treatment is administered early on in the latent state. The group's gene expression studies pointed to another potential target, in addition to Src, for inhibition. That target is an actin bundling protein called Fascin, which is important for invasive motility and has, in every gene expression study, been associated with aggressive metastatic cancer. Fascin is also implicated in attracting circulating tumor cells back to the primary tumor, a process that might be hindered by a recently developed Fascin inhibitor.

Figure 5. Src inhibition mainly suppresses cancer cell survival, not invasion or proliferation.

Metastasis is ultimately what kills cancer patients. But scientists, propelled by research like Massagué's into the precursors of metastatic tumors that arise early on in cancer progression, are ever closer to finding new treatments to prevent this deadly development. Already begun in Massagué's lab, successful identification of the genes expressed in these cellular forerunners to metastatic tumors will be crucial to attempts to prevent metastasis before it becomes terminal.

Speakers:
Yusuke Nakamura, The University of Tokyo, Institute of Medical Science
Manuel Perucho, Institute of Predictive and Personalized Cancer Medicine (IMPCC), Barcelona; Sanford-Burnham Medical Research Institute, La Jolla
Stephen B. Baylin, The Sidney Kimmel Comprehensive Cancer Center at Johns Hopkins University School of Medicine

Highlights

• BioBank Japan has collected DNA, serum, and clinical information from 300,000 hospital cases in Japan.
• DNA demethylation and genomic instability may lead to the mutations that lead to tumor growth and progression.
• Colon tumor behavior can be categorized by the presence or absence of microsatellite instability.

Epigenetic and genetic markers can tell scientists an enormous amount about the regulation of gene expression, and ultimately, about how tumors behave. But scientists need to employ an ever-increasing range of techniques to find the significant markers amongst the insignificant and to relate those markers to clinical and molecular phenotypic data before investigating them as possible therapeutic targets. In this session, Yusuke Nakamura detailed for the audience BioBank Japan's project to analyze vast amounts of patient data to gain insight into a range of clinical problems, such as the mechanisms of drug non-responsiveness and the environmental antecedents of cancer. Manuel Perucho and Stephen Baylin both presented on their investigations into the relationship between epigenetic alterations and cancer: Perucho on fundamental ways mutations give rise to cancer, some of which run contrary to cancer research orthodoxy; and Baylin on the role of epigenetics in drug sensitivity and oncogenesis with a view to prompting the re-expression of silenced genes.

The power of numbers

Japan has been an important player in large scale genomic projects right from the start. In addition to its well known SNP-typing facility in the Center for Genomic Research at RIKEN, it has started a very ambitious project called BioBank Japan, which has several goals, said Yusuke Nakamura of the University of Tokyo. It aims to discover genes that confer susceptibility to various diseases or that promote favorable or adverse reactions to drugs. Additionally, the project endeavors to identify molecular targets for "evidence-based" drug development, to provide important medical information useful to personalized medicine, and to perform genetic and environmental epidemiology for the prevention of disease dissemination.

From 2003–2008, DNA, serum, and clinical information were collected from 300,000 hospital cases in Japan. 66 hospitals participated, representing 2% of all hospital beds in the country. DNA was collected from each patient once and serum and clinical information were obtained each subsequent year. A 2-D barcode system was used to protect patient privacy.

Nakamura provided several examples of the types of discoveries that can be made using this rich source of data. In one investigation, the data have been used to investigate gene-environment interactions in esophageal cancer. Studies have shown that some alleles of ALDH1B and ALDH2, which code for two enzymes required for alcohol metabolism, are associated with an increased risk of esophageal cancer (Figure 6). These alleles code for enzymes with reduced function, resulting in the accumulation of acetaldehyde, an intermediate in alcohol metabolism and a known carcinogen. The BioBank Japan data indicated that when people with these alleles smoke and consume alcohol, they have a 190x increase in risk of esophageal cancer.

Figure 6. Certain alleles of ALDH1 B and ALDH2 are associated with a decreased capacity to eliminate acetaldehyde, a byproduct of alcohol metabolism and known carcinogen, from the body.

BioBank Japan has data from 750,000 cases of cancer. These data can be used to investigate adverse drug reactions, such as the reaction to docetaxel, to which patients may have a bad reaction if they have alleles of particular transporters that reduce clearance of the drug in the liver. It can also be used to identify causes drug inefficacy, such as the lack of response to tamoxifen that occurs when a patient has a missense mutation in an enzyme needed to metabolize the prodrug to the active form.

Nakamura concluded by describing the development of inexpensive and rapid diagnostic tests, the optimization of which will render advances in personalized medicine more widely practicable.

Genetics and epigenetics of colon cancer

It has been known for quite some time that a sequence of mutations leads to cancer. But hypotheses about how these mutations arise have been evolving as researchers learn more about the processes of DNA repair and genetic instability. Mutations in mismatch repair (MMR) genes, also called "mutator genes" to reflect the consequences of their ineffective expression, lead to mutations in hundreds of thousands of certain genomic sequences. These are especially repetitive sequences or microsatellites. Most of these mutations are innocuous, but some will affect oncogenes and tumor suppressors. As these mutations accumulate, the cells acquire what is known as Microsatellite Instability (MSI), or instability in sequences of small repeating units, and the combination of MSI and its distal cause, mutations to MMR genes, is called the "mutator phenotype." The mutator phenotype is the hallmark of sporadic as well as hereditary cancers, epitomized by the hereditary non-polyposis colon cancer syndrome (HNPCC). Manuel Perucho of the Institute of Predictive and Personalized Cancer Medicine (IMPCC), Barcelona and Sanford-Burnham Medical Research Institute, La Jolla, focused his talk on the genetic and epigenetic changes that give rise to colorectal cancer.

Perucho clarified that the mutator genes are not a cause of cancer the way oncogenes or tumor suppressor genes are: they do not confer a growth or survival advantage to cells; they only lead to a mutator phenotype which greatly increases the mutation rate. As he explained, MSI proves a useful way to classify colorectal tumors, since tumors exhibiting MSI differ significantly in both genotype and phenotype from their non-MSI counterparts, though they do represent only a minority of all colorectal cancers. Surprisingly, MSI colon cancers had fewer mutations of APC, RAS, p53 and other genes typically associated with colorectal oncogenesis. On the other hand, they exhibited a large number of mutations to genes in the same oncogenic networks, such as those involved in apoptosis, which are not normally mutated in non-MSI colon cancers.

MSI and non-MSI colon tumors differed phenotypically across a number of metrics, including location, stage of progression, degree of differentiation, and many others. Perhaps the most important of these differences, however, was in the respective survival rate of patients with each of the tumor types. In fact, patients with MSI tumors had improved survival data over their non-MSI counterparts, providing a clear example of the value of these studies in personalized medicine.

Figure 7. The combined effect of epigenetic and genetic alterations on oncogenesis.

In addition to genetic instability, epigenetic instability is also implicated in oncogenesis. Methylation is a key source of changes to gene expression that do not involve mutations in genes or associated regulatory sequences. Hypermethylation can silence mutator genes, leading to failures in DNA repair, which made researchers expect to uncover methylator genes to control the hypermethylation. No such methylator phenotype could be found, however, and scientists do not yet fully understand how unmethylated sequences become methylated in tumors. The reverse process, the demethylation of previously methylated genes, is conceptually much easier to explain, by spontaneous errors of replication of the methylation patterns by the constitutive methyl transferases. This is not surprising taking into account the huge size of the human genome, with high abundance of methylated sites, and the absence of a repair system to correct the methylation replication errors.

Hypomethylation provides a mechanistic explanation for the correlation between increasing age and increasing incidence of most types of cancer. Although tumor specific genetic alterations do not escalate with age in colon cancer, DNA demethylation is gradual and does increase with age. The consequent hypomethylation may lead to mitotic errors, alterations in cancer genes, and ultimately the emergence of some colorectal cancers.

Key aspects of the cancer epigenome

Stephen Baylin of the Sidney Kimmel Comprehensive Cancer Center at Johns Hopkins University School of Medicine is also looking at the role of epigenetics in the development of cancer. He pointed out that while DNA damage by stress and the environment is a known source of mutations leading to cancer, the opening and closing of chromatin to make DNA repairs causes its own damage to the epigenetic landscape and contributes to subsequent problems.

Two major epigenetic changes are often seen in tumors, usually simultaneously. The first is loss of DNA methylation, resulting in a more open configuration and more acetylation in some regions. The second is the acquisition of abnormal DNA methylation in CpG islands in promoter regions, resulting in closing of chromatin in these regions and repressive marks replacing those of active marks. Thus transcription is altered and chromatin integrity compromised in these tumors.

Microarray analysis of promoter-biased regions of CpG methylation in cancer cells revealed that hundreds of genes take on hypermethylation. Not a random selection gene by gene but rather a program of some sort appears to make these genes vulnerable to hypermethylation, said Baylin. A therapeutic approach to target epigenetic changes would likely have to account for this selective hypermethylation.

These epigenetic changes have been directly associated with drug resistance. A recent paper from Jeffrey Settleman's group at Massachusetts General Hospital has found that a heterogeneous population of tumor cells that are insensitive to a range of drug treatments will, by selective pressure, end up being characterized by a population of drug-tolerant persisters. These cells have stem cell-like characteristics and have very elevated levels of a protein called KDM5A. This protein is a histone demethylase that downregulates an active histone mark. When the protein is knocked down, cells insensitive to drug treatment become sensitive. The same effect can be seen by using histone deacetylase inhibitors or IGFR-1 inhibitors. The finding that resistance is reversible is quite encouraging.

Baylin next turned to the role of epigenetics in promoting stem cell and progenitor build-up and in blocking stem cell maturation. In a preinvasive colon polyp a number of genes that would counter these developments are hypermethylated. This epigenetic environment would be a candidate mechanism for the early clonal expansion and self-renewal of abnormal cells. A very high percentage of genes that get hypermethylated in colon cancer are marked by the Polycomb group of proteins. This complex initiates a series of steps that lead to gene silencing during normal development. Interestingly, when cells are stressed a very large complex of silencing proteins amasses at Polycomb-marked sites that may help maintain the cell in a stem-like state, with low DNA methylation and a low level of transcription.

A postdoc in the lab, Vijay Tiwari, looked at GATA-4, a gene that is Polycomb-marked and non-DNA methylated in embryonic stem cells but frequently DNA methylated in colon cancer. Polycomb marking and DNA methylation are two important epigenetic ways gene expression is regulated, and the presence of Polycomb group proteins around the GATA-4 gene has been shown, as Baylin explained, to repress transcription of that gene by "maintaining a series of long-range chromatin interactions." Although the precise mechanism is unclear, the Polycomb marking and methylation appear to act in concert to reduce transcription. When DNA methylation is largely absent, as is the case in embryonic cells, GATA-4 transcription is somewhat impediment, but that impediment can be reversed. The reversal to upregulated transcription is possible because the DNA methylation that would ordinarily sustain interactions between chromatin loops is not present, so the loops themselves can be eliminated, and with them the interactions that make transcription so difficult. By contrast, when DNA is hypermethylated, as in colon cancer cells, transcription is completely repressed, and removing the DNA methylation only slightly restores expression.

Thus DNA methylation and Polycomb marking do more than regulate transcription on a nucleotide by nucleotide basis; they sustain higher order interactions between chromatin.

Speakers:
Todd R. Golub, The Broad Institute of Harvard and MIT
Arul Chinnaiyan, University of Michigan Medical School
Joe W. Gray, Lawrence Berkeley Laboratory (talk presented by Laura Vant Veer)
René Bernards, The Netherlands Cancer Institute, Amsterdam
Carlo M. Croce, The Ohio State University Medical Center

Highlights

• Whole cancer genome sequencing can reveal previously unidentified oncogenes.
• Translocations that fuse androgen-driven regulatory sequences to ETS family transcription factors can be found in 60%–70% of prostate cancers.
• Putative cancer drugs can be tested in cell lines classified by breast cancer subtypes.
• The Ras signaling pathway confers resistance to retinoic acid therapy for neuroblastoma by blocking retinoic acid-induced transcription.
• MicroRNAs are downstream targets of pathways that are commonly dysregulated in cancer.

Finding mutations is itself a cumbersome task, but sorting out the mutations relevant to cancer is even more of a difficult process. Scientists, however, have successfully employed a variety of new methods to do just that. Todd Golub and his team categorized cancers by their genetic signature, which required them first to map mutations and then to determine their statistical significance. His group's work yielded some very interesting results about the development of cirrhosis into hepatocellular carcinoma. The need to sort through great quantities of genetic data for meaningful mutations inspired Arul Chinnaiyan and his group to develop and apply an algorithm specifically designed to analyze tumor microarray data to find cancer-associated chimeric DNA sequences and the genes that regulate them.

Interested in the efficacy of cancer drugs, Joe Gray and René Bernards investigated, respectively, particular drugs' ability to inhibit breast cancer cell growth and the signaling pathways that affect retinoic acid sensitivity in neuroblastoma. Carlo Croce wrapped up the session by discussing his use of microRNA signatures to characterize tumor behavior. MicroRNAs, he said, make great therapeutic targets because they are the post-transcriptional regulators for genes that affect everything from epigenetic dysfunction to DNA repair.

Statistically significant mutations in multiple myeloma

Todd Golub of the Broad Institute of Harvard and MIT and his group are using genome sequencing and expression profiling to find gene signatures that define different cancer types. In one approach, the team is using whole genome sequencing to find a gene signature for multiple myeloma, a relatively rare but incurable hematological disorder. They began by sequencing tumor/normal pairs using a combination of whole genome and whole exome shotgun sequencing on an Illumina platform. They identified the nonsilent mutations in each tumor and confirmed the existence of the mutations by genotyping. Determining which mutations are statistically significant was challenging, and required correcting for gene size and base composition. Nevertheless they have identified several statistically significant genes, some of which are new to multiple myeloma and to cancer in general.

In an effort to understand the role of some of the genes whose function in cancer is unclear, Golub's group correlated their expression patterns with that of all other genes in the transcriptome in a collection of multiple myeloma samples. The group was able to link these genes with certain cellular processes, expanding our understanding of their role in oncogenesis.

Golub's team was also interested in determining whether rare mutations, which individually weren't statistically significant, in aggregate identified pathways that were hitherto unlinked to the disease. They identified several such pathways and work is now underway to determine how these pathways are contributing to multiple myeloma progression.

Development of a gene signature

Whereas large scale sequencing efforts can be used to find new oncogenes, large scale gene expression studies can be used to classify patients' cancers and perhaps determine prognosis. To identify a gene signature for hepatocellular carcinoma, Golub's postdoc Yujin Hoshida began by looking at gene expression in tumors that had been excised from liver tissue.

Once he had identified a signature, he asked if it was possible to predict risk of hepatocellular carcinoma in patients with cirrhosis, the condition that often precedes development of hepatocellular carcinoma. Initial results of that investigation are indeed promising. Golub's lab is also encouraged by preliminary testing of a way to use gene expression profiles to predict patients with high risk and treat them specifically rather than treating a large, heterogeneous group of patients who may have very different outcomes and very different treatment needs.

Translocations in prostate cancer

Chromosomal translocations can, much like point mutations, create new gene products, contribute to the overall dysregulation of genes, genetic instability, and ultimately to cancer formation. Translocation can result in, among many other possibilities, the fusion of a promoter sequence from one gene to the coding sequence of another or the fusion of two gene sequences together, which can lead to dysfunctional or fusion proteins, some of which are immune to typical post-translational regulation. The range of outcomes to these translocation events is vast, and counted among them are a few mutations known specifically for their association with oncogenesis.

One of the most successful drugs developed for cancer therapy is Gleevec, a drug used to target the Bcr-Abl fusion oncoprotein that causes chronic myelogenous leukemia. Bcr-Abl is produced by a translocation between chromosomes 9 and 22, and the resulting chimeric chromosome is known as the Philadelphia chromosome. Encouraged by the successful targeting of an oncogenic fusion product, Arul Chinnaiyan and his colleagues at the University of Michigan Medical School are taking a bioinformatics approach to look for gene fusions in common solid tumors, such as those found in prostate cancer. They developed an algorithm called COPA that identifies genes that are expressed at very high levels, as might occur if, for example, a foreign promoter had been driving expression of a gene it normally doesn't regulate. Application of the algorithm to a compendium of tumor microarray profiling data identified the ETS transcription factor as highly upregulated in prostate cancer. Further analysis revealed that the regulatory region of an androgen-driven gene had been fused to the ETS gene.

Chinnaiyan's team went on to identify a host of other gene fusions in prostate cancer that can be used to classify the disease into subtypes. Similar to their first find, these fusion genes largely contain androgen-driven regulatory sequences fused to the genes of ETS family transcription factors. Chinnaiyan's group and others have found that 60%–70% of prostate cancers from prostatectomy cohorts bear these ETS gene fusions. They occur early in prostate cancer and can be used as a diagnostic marker of the disease. Several different kinds of genetic experiments in model systems have been used to verify that the fusion genes are oncogenic.

Figure 8. Diagram showing the many steps in the pathway from TMPRSS2-ERG gene fusion to the transcription and translation of the dysfunctional ETS protein and then to its blocking of prostate cell differentiation, some of which processes present possible targets for drug therapies.

After investigating the translocations' importance to cancer development, the team was eager to understand how they arose. In androgen-responsive cultured prostate cells, the androgen-sensitive regulatory gene region (TMPRSS2) and the ETS gene (ERG) were brought in close physical proximity by the addition of androgen, and irradiating cells induced the translocation required to have TMPRSS2 and ERG on the same chromosome. The induced translocations are cell type specific; for example the Bcr-Abl fusion can't be induced in prostate cells.

To understand the role of the fusion genes in prostate cancer the group developed a compendium of ChIP (chromatin immunoprecipitation) sequencing analyses in prostate cancer cell lines to map the genomic landscape of the master regulators of prostate cancer. The results suggested that the androgen receptor signaling that ordinarily drives differentiation of prostate cells is blocked in several ways by the actions of the fusion gene (TMPRSS2-ERG), leading to epigenetic silencing and oncogenesis.

Scientists are developing diagnostic tests to identify the fusion gene in the blood or urine of prostate cancer patients. At the same time, a number of therapeutic approaches can be explored for when the gene is present, such as blocking androgen receptor signaling, degrading the fusion gene transcript with siRNA, using small molecule inhibitors that directly engage the ETS gene, or finding synthetic lethal interactions to kill cells bearing the fusion mutation.

Chinnaiyan's group is now using next-generation sequencing to analyze the transcriptome of prostate cancer cell lines and identify other chimeric genes, which might also play a role in the cancer's behavior. The approach has been validated and a robust pipeline has been developed. In recent work, for instance, the team has identified fusions of B-RAF and C-RAF in the group of prostate cancers that do not bear the ETS fusions, as well as in other types of cancer. Patients bearing these RAF fusions may respond to treatment with RAF and MEK inhibitors, treatment avenues made available only through investigation of chimeric genes and their effects on oncogenesis.

Subsets, specific therapies in breast cancer

Joe Gray of the University of California, San Francisco and his group are surveying the range of breast cancers by looking at a large number of cell lines established from breast cancer tumors. Gray was unable to attend the event but Laura van't Veer presented his work in his stead. Gray's team began by characterizing 50 breast cancer cell lines to ensure that they were representative of human tumors. Using gene, protein, and phosphoprotein expression profiling, they found that the cell lines could be assigned to the different breast cancer subtypes defined by Charles Perou and others (Figure 9).

Figure 9. Cell lines model luminal, basal, and claudin low transcriptional subtypes. Expression profiling allowed the assignment of cell lines to breast cancer subtypes.

Looking for growth inhibition, the group tested 77 compounds on each of 50 cell lines and graded the cells from sensitive to resistant according to their response. Gray's team found that 18 compounds were most effective on the slower-growing cell lines, 17 compounds were associated with copy number abnormalities, and 31 were specific for cells belonging one or another breast cancer subtype defined by their transcription profile.

The study also shed light on the mechanism of cells' response to compounds that target the mitotic apparatus. Work done in collaboration with Carlos Caldas showed that mitotic apparatus protein inhibitors tend to work on the basal subtype of breast cancer. To put this information to practical use, they developed a mitotic network activity index (MNAI) that is the sum of the expression levels of genes in the network. MNAI was associated with subtype and outcome in the 1000 primary breast tumors sampled. The signature overlaps with a chromosome instability signature developed by Carter and others. Interestingly, this network is only found in tumors and has yielded therapeutic targets that can be validated in these tumors.

Finding mechanisms and biomarkers of drug resistance in cancer

Many cancer patients do not respond to current targeted therapeutic approaches but those who do, respond very well, stipulated René Bernards of the Netherlands Cancer Institute. Thus, it is crucial to develop better ways to identify patients who will respond to the drug and to stop overtreating patients who will not respond. But identifying likely responsive patients requires moving beyond known biomarkers and known therapeutic targets. Bernards framed his talk with the question: "How can biomarkers of a response be found when there are no likely suspects [such as EGFR or Bcr-Abl] to consider?"

Such is the case with neuroblastoma, a frequent childhood tumor with heterogeneous clinical behavior. While N-MYC amplification is found in approximately 20% of neuroblastoma cases and loss-of-heterozygosity of 1p36 is found in approximately 30% of cases, not much else is known about the genetics of the disease, which limits what can be used to narrow down potential drug treatment populations. Treatment relies on induction chemotherapy followed by autologous bone marrow transplantation or continued chemotherapy. Retinoic acid, which differentiates neuronal cells, is used for maintenance therapy. Unfortunately, retinoic acid treatment has limited therapeutic benefit.

Classical genetic approaches, which are unbiased, are well suited for a search for causal genes. For example, a gain-of-function screen can be used to look for cases where overexpression of a gene results in altered response to a drug. The opposite approach is also useful—a loss-of-function screen can identify cases where the suppression of a gene results in a drug response.

Bernards described one such genetic screen: a loss-of-function screen looking for responsiveness to drugs in neuroblastoma. While the genetic technique may be classical, its implementation is decidedly modern. Bernards' team used a large library of short-hairpin RNA (shRNA) vectors that target mRNA for destruction to test the effect of individually suppressing thousands of genes. Because each shRNA contains a sequence unique to its target, it serves as a molecular bar code that can be used to identify target genes.

Figure 10. A schematic of the process of shRNA screening to determine the shRNAs that confer resistance to cancer drugs.

In the shRNA bar code screen Bernards described, neuroblastoma cells that were very responsive to retinoic acid were infected with the shRNA library. These were divided into two replicate cultures, a control and a dish treated with retinoic acid. Cells that are responsive to the drug will stop dividing, whereas those that have acquired an shRNA that confers drug resistance will continue to grow, increasing the abundance of that shRNA in the culture. After a few weeks, a comparison of the levels of shRNA by PCR amplification and hybridization to a microarray identifies both shRNAs that confer resistance and those that contribute to cell death, thereby illuminating mechanisms of drug resistance and susceptibility.

The analysis identified two independent shRNAs that knocked down the expression of the NF-1 neurofibromatosis tumor suppressor gene (Figure 11), conferring drug resistance upon the cell culture. NF-1 encodes a GTPase-activating protein that inhibits Ras signaling but that had not previously been implicated in retinoic acid signaling. Further experiments showed that loss of NF-1 conferred retinoic acid resistance in a neuroblastoma cell line. An activated Ras protein that cannot be rendered inactive by NF-1 also confers resistance to retinoic acid on the cells. Furthermore, Bernard's team determined that the Ras signaling pathway was making cells resistant to retinoic acid by blocking retinoic acid-induced transcription.

Figure 11. Red boxed plates show restoration of sensitivity to Retinoic acid (RA) with MEK inhibition. Retinoic-acid induced transcription can then continue, resulting in apoptosis.

Previously, the group had discovered that such transcription in neuronal cells requires a co-activator called ZNF423, which binds to the retinoic acid receptor complex at promoters. Microarray analysis of NF-1 knockdown cells showed that ZNF423 transcription was repressed and that expression of the gene in these cells restores retinoic acid sensitivity. These findings have clinical implications as well; patients with low levels of NF-1 or ZNF423 have a poor prognosis compared to those with high levels. Approximately 6% of neuroblastoma patients show a genetic aberration in the NF-1 gene.

Genetic experiments yielded an important, clinically relevant finding: Of the three pathways that mediate Ras signaling, only the MEK pathway affects the retinoic acid sensitivity of neuroblastoma cells. Small molecule inhibitors of MEK restore retinoic acid sensitivity to resistant cells. As his work on retinoic acid drug resistance demonstrates, and as Bernards stressed, it is critically important to understand not just potential perturbations in signaling pathways, but also the crosstalk between those pathways, the many ways they affect one another.

MicroRNAs and cancer

Carlo Croce of the Ohio State University Medical Center introduced another genetic contributor to cancer formation: microRNA genes. He and his team made use chromosomal alterations such as deletions and translocations to identify genes involved in cancer as did many of the conference presenters. But they encountered difficulty with a deletion found in some cases of chronic lymphocytic leukemia (CLL), which prompted them to try to unravel the complex genetic mechanisms involved in the disease's pathogenesis. Believing initially that CLL-inducing mutations would be in genes that code for proteins, they sequenced 2MB of DNA in the region of the deletion and characterized every protein-coding gene. Disappointingly, they failed to find a protein-coding gene that was specifically altered in CLL. However, the presence of a very rare translocation in the same region allowed them to map the chromosomal breakpoint to the nucleotide level and a small deletion in another patient gave them the evidence they needed to locate the relevant gene precisely on the chromosome.

Figure 12. Histology of spleen cells (a, b, & c) and bone marrow (d, e, & f) in mice with and without the dysregulation of mIR155. Image c shows the disruption over time (from a to b to c) of cell structure caused by leukemic cells in a transgenic mouse with mIR155 dysregulated. Image f shows the change to bone marrow histology (compared to the wild type in image d) induced by mIR155 dysregulation. Dysregulation of a single microRNA can cause cancer, but it takes time.

Though there was no protein-coding gene there, the discovery of human microRNA genes led them to ask whether such genes might be found at that spot. In fact, two microRNA genes, called mIR15 and mIR16 were affected by the CLL-related translocation. Further analysis revealed that approximately 70% of CLL cases have disruptions in mIR15 and mIR16, and most if not all cases of the indolent form of the disease have these aberrations. In mouse models loss of the mIR15 and mIR16 resulted in the appearance of the indolent form of CLL, providing further evidence for the microRNA's contribution to oncogenesis. Croce's research shows that a small gene signature of 13 microRNAs can be used to diagnosis whether CLL is the indolent or aggressive form of the disease.

After determining that microRNA genes played a role in oncogenesis, the team still needed to uncover the nature of that role and, more specifically, the target of microRNA genes' regulation. Target prediction for microRNAs is difficult but Croce's group was able to find that one of the top targets for mIR15 and mIR16 was the oncogene the group had cloned in 1984, Bcl2. Bcl2 levels were shown to be increased in leukemia cells lacking mIR15 and mIR16. Adding back mIR15 and mIR16 reduced Bcl2 levels, as expected if it were a target of the microRNA genes, and eventually caused the cells to die. For this reason, Croce proposed that replacing mIR15 and mIR16 in cancer cells lacking these genes is a promising therapeutic option since the end result is apoptosis of the cancer cells.

Two different microRNAs were found to be critical in the regulation of another oncogene, TCL1, and the researchers were able to prove that TCL1 was the oncogene that caused the aggressive form of CLL. Deregulated expression of TCL1 caused T-cell leukemia in mice, but it also, somewhat surprisingly gave rise, with 100% penetrance, to B-cell leukemia in mice. In each of these studies and in the group's work on microRNA genes in lung tissue adenocarcinoma, the transformation of mutated cells to malignancy took time, sometimes up to 15 months, indicating to Croce that some additional genetic or epigenetic changes might be required for pathogenesis (Figure 12).

Croce also noted that the same microRNA can act as a tumor suppressor or an oncogene depending on the cell type in which they are functioning, and this tissue-specificity has allowed Croce's team to establish microRNA gene signatures for several other solid tumors, all of which depend on the tumors' location. Since microRNAs are downstream targets of pathways that are commonly dysregulated in cancer, they are excellent candidates for therapeutics, he said. They are targetable, if they're lost you can replace them, and if they're overexpressed you can knock them down. And, because they regulate processes such as epigenetic dysfunction and DNA repair, microRNAs can tackle mutations that are very difficult to target in the DNA itself.

Speakers:
Stephen H. Friend, Sage Bionetworks
Carlos Caldas, University of Cambridge
Charles M. Perou, University of North Carolina at Chapel Hill
Laura van't Veer, University of California, San Francisco (formerly at The Netherlands Cancer Institute)
James Wilson Watters, Sanofi-Aventis Oncology (formerly at Merck & Co. Inc.)

Highlights

• Retrospective analyses can be valuable in diseases that play out over a time span of decades.
• Genomic instability can be used as objective measures of tumor proliferation.
• Patients' outcomes were better if they based their treatment choices on MammaPrint technology rather than on existing protocols.
• Whole genome sequencing approaches reveal complex signaling substructure and potential drivers of disease.

Ideally, analyzing tumor type data in concert with epidemiological results will enhance clinicians' ability to treat their patients according to individualized risk rather than generalized breast cancer prognosis. Speakers in the fourth session demonstrated just how valuable large scale analyses can be for doctors seeking better predictors of metastatic risk and treatment efficacy. Stephen Friend shed light on current projects to integrate molecular and clinical datasets computationally. Carlos Caldas discussed his team's retrospective analysis of more than 10,000 cases of invasive breast cancer in which study they were able to follow patients with many different levels of treatment for 10 years to see how cancer and treatment played out over time. Also focusing his talk on breast cancer typing, Chuck Perou conducted expression array analyses to categorize tumors, and correlated those categories with population-level epidemiological phenotypes. The clinical applications of these efforts to subtype breast cancer came through clearly in Laura van't Veer's talk as she discussed improved treatment results for patients whose treatment was based on the 70-gene signature her group developed. And James Watters concluded the session by highlighting the importance of whole exome studies for rational drug development and combination therapies.

Computational dataset integration

Stephen Friend of Sage Bionetworks honed in on a topic touched on by many of the investigators at the conference: the need for a new way of conducting biological research. Despite the expenditure of a lot of effort and money, most candidate drugs fail clinical trials and many of those that are approved are not very effective. Friend argued that analyzing individual signaling pathways and choosing targets from its components is not the most effective way to find treatments for diseases such as cancer or heart disease. A better approach is to use "computational methods for integrating molecular and clinical datasets obtained across sizeable populations into predictive disease models," he said (Figure 13).

Figure 13. A schematic of a predictive network-aided drug discovery process.

Friend described a $150 million effort of Merck and Co., Inc., carried out from 2003–2008, to generate a predictive model for understanding diabetes. Eric Schadt and other colleagues at Rosetta (which was acquired by Merck) devised a "top-down" approach, first generating experimental data and collecting clinical findings, then building models that could be tested experimentally and used to develop clinical treatments and biomarkers. They found that DNA variation in populations could be used to find the drivers of the disease. Bayesian, coexpression, or causality networks could be used to decipher biological systems. Their analysis indicated that components that lack redundancy or buffering, and thus are vulnerable to perturbations in the system, drive the disease process.

Turning toward the problem of gathering and analyzing massive amounts of data, Friend suggested that a new paradigm is needed that allows investigators to make data sets publicly available without worrying that they won't be able to get credit for their work via the traditional route of publishing a journal article. He recommended following the physics model whereby the data is shared among a large number of researchers and they receive credit for their analysis of the information, not its generation and publication.

To this end, Friend and others have formed Sage Bionetworks, a nonprofit organization located within the Fred Hutchinson Cancer Research Center to perform research, train scientists, and build a platform for the sharing of data sets. The goal is to host a large number of global coherent data sets, defined as a data set containing genome-wide DNA variation and intermediate traits, as well as physiological phenotype data across a population of individuals large enough to power association or linkage studies, typically 50 or more individuals. To be coherent, the data need to be matched with consistent identifiers. Intermediate traits are typically gene expression, but may also include proteomic, metabolomic, and other molecular data. The models generated from this data will also be made available by the participants. Sage Bionetworks is working with foundations, publishers, and research institutions in an attempt to revolutionize the biological research process.

A new breast cancer typology

Carlos Caldas of the University of Cambridge joined the conference discussion by highlighting the importance of analyzing existent data in order to take cancer research in productive new directions. Lamenting what has gone unlearnt over the past 25 years, he cautions the cancer-researching community to do better over the next 25 by harvesting more information from clinical trials and from non-trial patients' clinical follow-up. Since cancer development and patient recovery both take time, scientists need long-term data to predict outcomes for particular patients with particular cancers.

Caldas' research focuses on breast cancer, which has one of the higher survival rates overall when compared with other cancers such as pancreatic cancer. Nonetheless, some types of breast cancer, historically characterized by tumor size, tumor grade and lymph node involvement, have survival rates significantly under the average, and the disease is still the most common epithelial malignancy diagnosed in women. But in order to associate the molecular phenotypes, or barcodes, of these more deadly diseases with metastatic and other behaviors and to determine more appropriate levels and kinds of treatment for individual patients, scientists need to examine why some patients respond to treatments and some do not.

Caldas suggests that the cancer research community, if it wants to avoid the inefficiency of the last few decades, might need to abandon the paradigm that biomarkers must be prospectively validated in causal studies in order to be considered validated at all. Statistical analyses of retrospective data provide a promising alternative, according to Caldas.

His own research used "stone-age arrays" of cDNA clones to demonstrate first that tumors' molecular barcodes divide them into recognizable subtypes, and second that these subtypes had prognostic significance. Immunohistochemical criteria (ICH), estrogen receptor (ER), progesterone receptor (PR), and human epidermal growth factor receptor (HER) 2 expression were used as biomarkers to determine tumor subtypes. Caldas also tested for expression of cytokeratin (CK) and EGFR (HER1) basal markers to further divide what he called the "triple negative" category of tumors, those that are non-luminal (neither ER nor PR expression) and HER negative. After profiling nearly 10,300 breast cancers, the researchers determined that triple negative breast cancers are not equivalent—their behavior depends heavily on their basal expression, which is not the case for HER2 positive tumors.

For ER positive cancers, driven by the estrogen receptor, patients' prognosis is determined by how many oncogenes are amplified and therefore how much of the cancer is proliferating. It therefore becomes very important to find clear measures of tumor proliferation. Caldas and his collaborators have been developing "objective" measures of the fraction of tumors proliferating, based on genomic instability, and cross-referencing those results with survival numbers to see if such a fraction holds prognostic value. One result from Caldas's and others' early experimentation with these methods is that basal-like tumors do not always express genomic instability despite the predicted association of the core basal phenotype's poor prognosis and greater genomic aberration. Further research and specifically further retrospective analysis are needed to determine the ramifications of this result.

Caldas explained that a similar level of retrospectively acquired data could be obtained from ongoing trials. For instance, his research group contributed to a clinical trial (Neo-tAnGo) testing the efficacy of gemcitabine added before or after anthracycline as a neo-adjuvant therapy. He was particularly interested in developing a predictive signature of response to certain drugs. Moreover, he contends that this signature could be used to design new drug trials, without its being validated prospectively.

Caldas has used the vast amount of clinical follow-up data available through the Molecular Taxonomy of Breast Cancer International Consortium (METABRIC) to demonstrate the utility of these kinds of statistical analyses. By analyzing tumor and treatment types, his group has been able to uncover 50% more new copy number variations (CNVs) and to locate very narrow amplicons, which many scientists believe are key to finding driving oncogenes.

He closed his talk by highlighting the general importance of retrospective analyses of large amounts of patient data. These data and the gene-environment-disease correlations they illuminate can and should be used to design clever clinical trials. They can, he contends, open up a world of options for determining markers for risk assessment, markers that are diagnostic, prognostic, and predictive, markers for determining drug sensitivity or resistance, and markers to monitor treatment progress along the way and head off relapse before it takes hold.

Molecular portraits of breast tumors

Charles Perou of the University of North Carolina at Chapel Hill has been crucial to the successful effort to divide breast cancers into five subtypes according to their expression profiles. Those subtypes help clinicians predict patient prognosis and tumor behavior. After comparing 22 tumor pairs (20 "before and after" doxorubicin treatment), Perou found that the "before" and "after" samples almost always had more similar expression profiles to each other, than to other tumors. This result indicated that every tumor has a unique and reproducible expression profile that persists over time, despite the evolution of the tumor as it acquires new mutations. Furthermore, this expression profile is a reflection of the dominant biology of breast cancer and is highly correlated with response to therapy, said Perou.

The five molecular subtypes were named Luminal A, Luminal B, Claudin-low, Basal-like, and HER2-enriched. Interestingly, each of these subtypes can be found within the clinically defined estrogen receptor positive (ER⁺) group, indicating that there is significant heterogeneity within the ER⁺ group. Thus, subtyping tumors can yield additional information to guide treatment options. The Luminal A subtype is the only one with a relatively good prognosis. The Claudin-low subtype is the most newly defined group and is characterized by having some stem cell and early breast cell developmental features. Perou suggested that the subtypes reflect different stages of mammary cell development, with each type arresting at a given stage.

The five categories are also capable of sorting the majority of tumor samples, but there are some tumors that have an expression profile that falls between two subtypes. To account for these ambiguous cases, Perou's group developed a Cox model (called the Risk of Relapse/ROR) where the distance of a given tumor to each of 4 tumor subtypes (Luminal A, Luminal B, Basal-like, and HER2-enriched) is recorded as a classifying variable. These variables can include many relevant data types (e.g., tumor size, node status, genomic alterations), not just tumor genomics. But since not all criteria are equally important, each characteristic is weighted, according to the information from a training data set, and then the weighted criteria are used to calculate the "ROR" score for a given tumor, which are described as a risk of relapse score, from 0–100. The data can be plotted so that the risk of relapse score is proportional to the probability of relapse at 5 years. In this way the subtypes can be used to predict patient outcomes even for patients whose tumors don't fit precisely within one subtype. This information can be valuable for patients deciding whether to have chemotherapy or other treatments. To bring the ability to determine tumor subtype to the clinic, Perou's team transitioned from large-scale microarray analysis to a quantitative RT-PCR assay of just 50 genes. They also designed probes that work with RNA coming from paraffin-preserved tissue, which can be difficult to use.

Understanding mechanistic differences between subtypes

Perou next described research into the triple-negative subtype, also known as the Basal-like subtype. Triple-negative tumors do not express the estrogen, progesterone, and HER2/ERBB2 protein, rendering them insensitive to therapies that target these receptors. The type has a poor baseline prognosis and is characterized by expression of basal cytokeratins, mutation in the p53 tumor suppressor a majority of the time, and often mutation in BRCA1.

Expression patterns of basal-like tumor cells indicated that levels of the Rb tumor suppressor are low, due to mutations in the Rb gene and loss of heterozygosity. Based on this pattern and the expression pattern of other cell cycle genes, Perou developed a model for the basal-like phenotype in which the cells are cycling quickly and the cells try to compensate by expressing high levels of cyclin-dependent kinase (CDK) inhibitors. But since these inhibitors need Rb to be effective, they fail to slow down the cell cycle. In contrast, Luminal A expression patterns show that CDK inhibitor levels are very low in this subtype, which indicates that a different mechanism is controlling tumor behavior. These models can help explain the differences in response to chemotherapy drugs in human trials (Figure 14).

Figure 14. Subtypes show over and under expression of different components of the cell cycle.

An understanding of the different subtypes of breast cancer can also yield insights into results of population-based studies of the disease. African-Americans and young women have a lower incidence of breast cancer but a higher rate of mortality from the disease. One study, called the Carolina Breast Cancer Study, aimed to understand how subtype may be related to these incidence levels. The study found that the occurrence of the basal-like subtype correlated with the poor outcome of African-Americans and young women; occurrence was highest in young African-Americans and lowest in post-menopausal Caucasian women. At the same time, the frequency of the Luminal A subtype was lower in African-Americans and young women. Perou has extended the analysis to other populations as well. Somewhat counter intuitively, epidemiological studies showed that factors that were associated with being protective for breast cancer as a whole, such as having several children, were associated with a predisposition to basal-like cancers. As these studies proliferate, scientists will have more and more information, from individual tumor expression profiles to population-wide life habit data, to inform the increasingly personalized diagnosis and treatment of breast cancer patients.

Breast cancer biomarkers to guide treatment decisions

Laura van't Veer and her group at the Netherlands Cancer Institute (now at the University of California, San Francisco) has developed one of the first commercially available prognostic tests for breast cancer, called MammaPrint, which is based on a 70-gene signature. For 30% of the cases in one study, the gene signature her group developed yielded a different prognosis than that predicted by the classical definitions of clinical risk as defined by Dutch CBO (Dutch Institute for Healthcare Improvement) guidelines. Approximately 20% of those cases adapted their treatment based on the gene signature, and follow-up indicated that the outcomes were better for the group who based treatment on the gene signature rather than on CBO guidelines. Similar discordancy was seen when MammaPrint was compared to another set of criteria provided by Adjuvant Online.

Clinical trials are underway to test a number of treatment options for breast cancer patients. One such trial, MIND-ACT, is randomly assigning treatments in the 13% of cases that have discordant predictions when analyzed by MammaPrint and Adjuvant Online. Clarifying whether this group is high risk or low risk can potentially spare 10%–15% of patients from chemotherapy without jeopardizing their survival. In addition to separating the groups based on the two diagnostic tools, the groups are being randomized to two types of chemotherapy. In a pilot trial of 800 women, van't Veer said the randomization protocol was actually being followed, which allayed her concerns that doctors and patients ultimately might not be willing to follow the protocol. Copious molecular data is being accumulated to help assess the efficacy of the different approaches.

Van't Veer turned to a description of the neoadjuvant I-SPY trial set up by Laura Esserman and Nola Hilton of UCSF. The trial was designed to assess the utility of serial MRI scans and serial core biopsies in determining the therapeutic response to the first exposure to chemotherapy (neoadjuvant treatment that usually precedes surgery). The biopsy samples were subjected to a large number of molecular assays to characterize gene, protein, and cell biological status. Ninety-one percent of the tumors collected in the trial had gene expression profiles indicative of poor prognosis according to the 70-gene signature analysis.

The data indicate that patients whose prognosis is poor according to their tumors' biological profile will nevertheless have a good outcome if they show a good early response to neoadjuvant chemotherapy. Patients whose tumors have a good biological profile will have a good outcome whether or not they respond well to neoadjuvant chemotherapy, thus such treatment is not useful for them.

Another trial, called I-SPY2, is designed to screen phase 2 agents in combination with standard chemotherapy as neoadjuvant treatments. Ordinarily phase 2 agents are used in metastatic patients not in pre-surgery patients. The researchers hope to accelerate the process of identifying drugs that are effective against specific breast cancer subtypes. The trial will also be useful for identifying the utility of markers revealed by Joe Gray's in vitro analyses. So, after the biomarker profile of a minimum of 20 patients in this trial show response a particular drug, new patients with that biomarker profile will be put on the drug. Together these approaches hold the promise of getting drugs to the market faster, using fewer patients, and at a lower cost.

Expression signature for whole Ras signaling pathway

James Watters of Merck (now at Sanofi-Aventis Oncology) discussed his company's efforts to use molecular profiling to understand disease subpopulations and drivers in general, and specifically to find gene expression signatures in lung cancer. Expression profiling in non-small cell lung carcinoma (NSCLC) cell lines indicated that genes involved in the epithelial-mesenchymal transition (EMT) and those in Ras signaling are the key drivers of variability in those cancer cells. Aberrations in EMT are thought to play a role in cancer as cells lose their epithelial cell characteristics and gain mesenchymal cell abilities to migrate and survive in distant environments. The researchers identified an EMT signature by comparing mesenchymal-like and epithelial-like cell lines and selecting the genes that differentiate these groups by ANOVA. Known EMT drivers such as ZEB1 and ZEB2 were contained in the signature. Further analysis indicated that FGFR1 and Notch signaling are possible drivers of “mesenchymal” cells in NSCLC.

Watters next explained the rationale for finding a RAS whole-pathway signature: testing for mutations in genes that code for RAS subfamily proteins will only identify a subset of the tumors that actually have RAS signaling activated, since the signaling pathway can be activated at multiple points. The RAS whole-pathway signature the team identified doubles the population of tumors classified as “Ras active,” a label that also correlates with drug sensitivity. Further experiments indicated that cMET and EGFR signaling are the likely drivers of the RAS signature pathway in NSCLC. RAS pathway signatures are highly correlated with MET expression, and MET inhibition downregulates the RAS signature

Moving beyond gene expression, the Merck team also analyzed copy number variation and mutation data to create a systems view of NSCLC adenocarcinoma and NSCLC cell lines. The data suggest that FGFR1 and the GS complex are likely druggable targets of mesenchymal tumors.(Figure 15).

Figure 15. Classification of NSCLC adenocarcinoma. "Ras active" (in pink) classification correlates with drug response.

Watters agreed with other meeting participants that testing combination therapies is very difficult with current industry and regulatory practices. One hindrance is that often the different compounds that might work well together were developed by competing companies. To combat this problem, Merck and Astra-Zeneca signed the first-ever collaborative pre-licensing agreement to test an Akt inhibitor designed by Merck with a Mek inhibitor designed by Astra-Zeneca.

Watters concluded with a conundrum: if researchers start with available drugs and try to determine the drugs' ideal target population then they risk ignoring key differences in patients' biology. If, on the other hand, they begin by typing patients according to their individual biology, then drugs for those targets may simply not be available. Gene signatures like the one developed by Watters and his team for non-small cell lung carcinoma (NSCLC) can help researchers assess multiple aspects of biology in conjunction with multiple therapeutic options.

Speakers:
Alan Ashworth, The Institute of Cancer Research, London
D. Gary Gilliland, Merck and Co. Inc.
José Baselga, Massachusetts General Hospital (formerly at Vall d'Hebron Institute of Oncology (VHIO) and Vall d'Hebron University Hospital, Barcelona)

Highlights

• Genome-scale siRNA screen reveals that the CHK1 gene might enhance cytotoxicity of gemcitabine.
• PARP inhibitors provide second, and lethal hit, to cells with BRCA2 mutations.
• mTor inhibitors in combination with IGF1 receptor antibodies show promising results against cancers with PI3K mutations.

There are a number of ways that cancer therapy can fail. Some oncogenic proteins are not inhibited by currently available drug compounds. In other cases, resistance arises to drugs that were once effective. Speakers in the fifth session discussed some of the ways their research tackles drug inefficacy or resistance. Alan Ashworth spoke of using the synthetic lethal approach to kill cells with mutations in the homologous recombination repair system. Gary Gilliland considered Merck's efforts to find possible combination therapies by using siRNA in both synthetic lethality and enhancer screens. José Baselga presented his work on drug resistance, highlighting the importance of understanding how signaling pathways function and interact with each other in devising more effective treatments.

Synthetic lethality and DNA repair defects

Alan Ashworth of the Institute of Cancer Research discussed synthetic lethal approaches that exploit defects in DNA repair. A network of repair pathways works to repair an average of 10,000 DNA lesions in every cell every day. The rationale for targeting DNA repair defects in tumors is threefold: germline mutations in DNA repair genes, such as in BRCA1 and BRCA2, predispose individuals to cancer; many tumors carry signs of genomic/genetic instability as a result of unchecked mutations; and most chemotherapeutic agents work by damaging DNA, thus, defects in the repair pathways confer sensitivity to these drugs.

Carriers of BRCA1 or BRCA2 mutations have one wild-type copy of the gene, and the loss of this copy over time predisposes carriers to cancer. In homozygous mutant BRCA1 or BRCA2 cells, the normally efficient homologous recombination repair pathway is inactivated so an alternative pathway that haphazardly puts together broken DNA ends takes over. The result is a propensity to develop cancer. Ashworth's team looked for ways to target only the cells that had lost both copies of BRCA1 or BRCA2, not the heterozygous but phenotypically healthy cells.

Historically, researchers have been unable to develop drugs that work on cells that are dysfunctional due to the lack a critical protein, such as a DNA repair protein. But a classical genetic approach, called synthetic lethality, can take advantage of the cell's altered state, even in the case of a missing component, and thus is a promising answer to this problem. Synthetic lethality, originally described by Theodor Dobzhansky in 1946, is a situation where two mutations are not lethal when each occurs singly in a cell, but are lethal when combined. In the modern interpretation of this phenomenon, a drug stands in for the second mutation required to kill the cell.

Ashworth's group used PARP (Poly ADP ribose polymerase) inhibitors to inhibit base-excision repair in BRCA2 mutant cells, hoping to create a situation where inhibition of two different DNA repair pathways leads to cell death. In cells lacking only 1 copy of BRCA2, the homologous recombination pathway should still be functional so those cells should remain viable. Indeed they found that BRCA2-deficient mouse embryonic stem cells were extremely sensitive to PARP inhibition, whereas heterozygous BRCA2 cells behave exactly like normal cells (Figure 16).

Figure 16. Genome instability is clearly visible even on the chromosome level for BRCA2 mutant cells with inhibited PARP.

But, both for clinical implications and to show that the proposed mechanism of action was correct, Ashworth's team wanted to reveal the limits of this inhibition by finding cells that developed resistance to the synthetic lethal PARP treatment. BRCA2 mutant cells were treated with PARP inhibitor for several months until resistance developed. Analysis of the BRCA2 genes in the resistant cells showed that a functional BRCA2 gene had been reconstituted in these cells. The clinical implication for this work is that the best results to treatments are likely to be achieved early in the course of the disease when there are fewer cells to kill before resistance can arise.

The finding that PARP inhibitors show synthetic lethality in BRCA2 mutant cells allowed the researchers to redirect the route through the clinical trial system so that after finding the highest nontoxic dose in an unselected patient population, they could test the drug in cancer patients with the BRCA2 mutation. The results of the trial were very encouraging with a total clinical benefit rate of 59% in a very heavily pretreated patient population. The phase 2 results are also encouraging.

But what about all the cancers that aren't caused by BRCA mutations? The same approach can be used in cases where a second hit is required for lethality. Ashworth's team turned to the PTEN (phosphatase and tensin homolog) gene, mutations or deletions in which are common across a broad spectrum of cancers. Studies have shown that PTEN plays a role in regulating Rad51, a key component of the homologous recombination pathway. After showing that homologous recombination efficiency was reduced in PTEN mutant cells, Ashworth's group treated the cells with PARP inhibitors. While the sensitivity of these cells was not as great as in the BRCA mutant cells, the result was still very good and they are trying to start clinical trials in patients with PTEN mutations.

Ashworth's group is also developing assays for pharmacodynamic analysis of the consequence of PARP inhibitor treatment and for a Rad51 homologous recombination biomarker to aid in optimization of this and related treatments.

High throughput screening & combination therapies

A major problem with translational research is how to take all the information being generated and translate it into clinical benefit for patients. All too often in drug development, treatment results may show statistical significance but have very little effect on the patient's condition when put to the test clinically, commented Gary Gilliland of Merck, & Co. Not only is this a problem for the patient, it is a problem for the pharmaceutical industry because companies are required to test new drugs against the latest standard of care in the control population. This standard may be an expensive combination therapy that is barely more effective than the drug that had preceded it. Gilliland suggested that we need new therapeutic paradigms that will match the right drug to the right patient, permit the development of cocktails of targeted agents, and improve the value to discrete patient populations, without requiring pharmaceutical companies to demonstrate drug efficacy across a large population exhibiting a broad range of treatment needs.

As mounting evidence suggests that combination therapies hold the most promise for getting better treatment outcomes and avoiding resistance, Merck is using RNAi screens to inform combination chemotherapy regimens. Short interfering RNAs (siRNAs) that block the expression of their targets can be used in synthetic lethality screens and in enhancer screens to identify targets that may work in such combinations, by finding pathways that are upregulated or downregulated in tumors. Merck is taking advantage of the power of automation with ultra-highthroughput screens and multiparametric assays to identify biologically relevant siRNA hits.

Gilliland described the results of one such screen, a genome-scale siRNA screen for enhancers of gemcitabine (a common chemotherapy drug) cytotoxicity that yielded the CHK1 gene as a hit. Since gemcitabine is a drug that will be used as the standard-of-care arm of clinical trials, CHK1 inhibitors can be tested to see if they improve gemcitabine's ability to kill cancer cells in those trials.

The length and expense of clinical trials is no guarantee that drugs will make it through, and indeed, many do not. Gilliland suggested a number of changes that could be made during each trial phase to improve the odds of success. In phase 1, researchers need to show that the drug affects cell function. In addition, they must develop a test to select patients and validate the test in the clinic. In phase 2, researchers must at least see tumor shrinkage in a defined subpopulation. They must follow CLIA (Clinical Laboratory Improvement Amendments) regulations of clinical lab research performed on samples from humans. In phase 3, a new drug must be tested in all patients and shown to work only in subpopulations. Alternatively, an in vitro diagnostic can be developed to select patients who fall into the targeted population, so that the drugs never have to be tested on people who are unlikely to respond significantly.

To increase scientists' ability to find the most appropriate test populations, Merck has set up a very large scale effort to collect tumors on every patient that comes into participating hospitals. These tumors will be analyzed to identify signatures of tumors and treatments will be selected based upon the specific tumor phenotypes. The profiles of these patients will be stored at the hospitals, and the plan is to collect 25,000 clinically annotated tumor specimens over five years, of which 75% will be primary tumors and 25% will be metastatic tumors. A small subset of those collected will be biopsy samples. Gilliland noted that in 2008, the hospitals, to increase the samples' utility, began collecting large enough samples to split for DNA and RNA analysis (Figure 17).

Figure 17. Moffitt/M2GEN collaboration — pre-selective, biomarker-driven enrollment. Schematic of biomarker-driven, branched, subpopulation-based trials.

As a consensus emerges that companies and academia need to be much more open with their data, Merck has decided to share the mRNA profiles with the participating hospital system and ultimately with the public. In the recognition of the new paradigm, Lilly, Pfizer, and Merck have teamed up to form the Asian Research Cancer group. The power of new techniques such as next-generation sequencing and other high-throughput technologies, and a new openness to sharing data are beginning to change the landscape of cancer drug development.

Combating nonresponsiveness: countering PI3K mutations

Even after seemingly successful drugs have been developed to treat cancer, they may show less success in certain subsets of the treatment population. Expanding the treatable population for existing drugs is one of the reasons José Baselga of the Vall d'Hebron Institute of Oncology (now at Massachusetts General Hospital), investigates mutations in the PI3K pathway. Activating mutations or deletions in the PI3K/Akt/TOR pathway are extremely common in cancer.

As others have noted during this conference, breast cancers can be divided into subgroups based on their expression of receptors for hormones that stimulate cell proliferation. About 33% of hormone-responsive breast cancers and about 22% of HER2⁺ (human epidermal growth factor receptor) cancers have PI3K mutations. The incidence of the mutation is lower in triple negative tumors. Researchers have found that PI3K mutations interfere with response to trastuzumab, which is a monoclonal antibody that targets HER2 receptors and therefore HER2⁺ tumors.

The PI3K pathway can be targeted at many points by agents that inhibit pathway components. Baselga's group hypothesized that mTOR (mammalian target of rapamycin) inhibitors may be useful with anti-estrogen therapy against estrogen receptor positive (ER⁺) breast cancers. In a phase 2 neoadjuvant study of Letrozole (an anti-estrogen agent) with or without Everolimus (an mTor inhibitor), the group saw statistically significant results with the combination therapy, that is with Letrozole and Everolimus together. The results held up when another determinant of efficacy, a cell proliferation marker called Ki67, was employed. Based on these findings, Novartis has started a large phase 3 trial of this combination therapy on patients with advanced disease.

New PI3K inhibitors, as well as a dual inhibitor of PI3K and mTor, have been developed since Everolimus and some of them are entering clinical trials as well. Pleased with the direction of this research, Baselga generalized the trials' results: "When we achieve a good dose of these inhibitors, we do block the pathway in patients."

Applying more broadly the lessons learned in the search for PI3K and mTor inhibitors, Baselga outlined the questions that must be answered when trying to develop drugs for these pathways: What is the best target in the pathway? Will toxicities differ? Will the drugs' efficacy depend on where in the pathway a patient has a mutation, as is the case with the B-RAF inhibitor? Will specificity to patients' phenotypes make the drugs safer overall? Which combinations will be best? And what alternative pathways will cells turn to when one is successfully blocked?

Figure 18. Combination therapies can be an effective strategy to resuscitate treatment efficacy by knocking out compensatory ("work around") pathways in addition to the primary pathways.

Combination therapy is thought to be the best way to combat the development of resistance to treatment, but challenges arise even with this approach. Baselga's group found that in a study of Everolimus, phospho-Akt levels rose, indicating that a compensatory pathway (Figure 18) had been activated, and that treatment efficacy was likely reduced. This can be averted by using a combination of an antibody against the IGF1 receptor, a part of the compensatory pathway, and an mTOR inhibitor. Further analysis showed that activated Erk levels were also raised with mTOR inhibitor treatment. This result is context specific so the pathway driving Erk activation must first be determined before that compensatory pathway can also be blocked.

Baselga highlighted the newest direction his research is taking: incorporating molecular/dynamic imaging in early clinical trials, developing quantitative tissue-based biomarkers (fluorescent immunohistochemistry and protein arrays), and understanding mechanisms of primary and acquired resistance. Garnering this new knowledge will be enormously helpful in developing new, more effective, and more personalized cancer treatments.

Speakers:
William R. Sellers, Novartis Institutes for BioMedical Research
Judith Sebolt-Leopold, University of Michigan Medical School
James Allison, Memorial Sloan Kettering Cancer Center

Highlights

• Combination therapy with a PI3K inhibitor and a Smo inhibitor decreased relapse in Ptch-mutant mouse models.
• MEK inhibitors unlike BRAF inhibitors are effective against a significant fraction of tumors with KRAS mutations.
• Treatment with an antibody to CTLA-4 has resulted in tumor regression in many instances and shows promise in increasing the immune response to cancer.

While much progress has been made in cancer research, some forms of the disease have proved less tractable than others. Presenters in this session shared their efforts to move drug discovery forward in these areas. William R. Sellers discussed the development of an "encyclopedia" of cultured cells using automated, mechanical handling and bioinformatic analysis techniques, and to study processes that won't manifest in cultured cells, an analogous collection of human tumors transplanted into a nude mouse model. Judith Sebolt-Leopold presented her work on one of the most difficult cancers to treat, pancreatic cancer, and specifically efforts to develop MEK inhibitors for cancer therapies. James Allison closed the session by providing a novel tactic for getting the body to fight even the most "untreatable" cancers on its own.

Encylopediae of cultured cells and transplanted tumors

William R. Sellers of Novartis Institutes for BioMedical Research (NIBR) began by emphasizing the importance of understanding the pathogenesis of a disease so that it can be most effectively treated. He raised the example of the identification of the Bcr-Abl fusion product as the underlying cause of chronic myelogenous leukemia, and the unprecedented success of Gleevec, the drug designed to inhibit it. Other similar examples are gefitinib (an EGFR inhibitor) treatment of EGFR-mutant lung cancers, and, for a rare subtype of melanoma caused by KIT mutations, the successful response to treatment with imatinib. Targeting of oncogenic FGFR mutations is on the horizon.

To get results like those described above, it is necessary to be able to match clinical findings with preclinical cancer models. Sellers pointed out that using only a few cancer cell lines to test a drug is the equivalent of doing clinical trials in two patients and deciding whether they were reflective of lung cancer or prostate cancer more generally.

To avoid the problems caused by small sample sizes, Novartis is collaborating with the Broad Institute to develop a cell line "encyclopedia" in which the cell lines are profiled for genetic alterations as well as their expression profiles. They are analyzing approximately 1000 cell lines, with the choice of which lines to include driven by unmet medical need. Affirming the need for openness within the research community, Novartis will make the results publicly available.

A major question that arises in such an effort is how to functionally interrogate the cell line data: what questions to probe and how to do so. If the cells can be analyzed more efficiently, the kinds of questions one can ask is less limited. Sellers showed a video of a robotic system that removes cell lines from an incubator, moves it to a compound distribution platform, and returns it to the incubator. After 72 hours the number of cells in the dish is determined in an automated way. With this system, they were able to run 1800 compounds with 9 point IC50 (half maximal inhibitory concentration) curves across 500 cell lines in 3 months. If performed in a typical lab, without this equipment, this analysis would take years.

The cell line information was used to investigate the activity of BKM120, a pan-Type 1 PI3K inhibitor. Analysis of 230 cell lines showed that mutation in PI3Ka is correlated with sensitivity to the inhibitor across a number of cell lines, except for in colorectal cancer cell lines, which also have a Ras mutation. This compound (BKM120) has served as proof-of-principle and is now in phase 1 clinical trials for breast cancer treatment.

In addition to gathering the data in an automated fashion, Novartis and the Broad developed a bioinformatic approach to analyze the results. In this approach, the cell lines are classified as either sensitive or insensitive and features, such as gene expression, mutations, copy number, are combined into feature sets that are analyzed using computational biological methods.

A different system is required to study signaling pathways and tissue interactions that will not manifest in cells grown in culture. For these cases, Novartis is using a genetically engineered mouse model in which tumors are harvested from mutant mice and transplanted into nude mice. Sellers detailed how the model is being used to test an inhibitor of the Smoothened (Smo) receptor, whose inhibition has a synthetic lethal effect in cases where its binding partner, Patched (Ptch), is mutated. This Ptch mutation occurs in some instances of medulloblastoma and basal cell carcinoma.

The Smoothened receptor is part of the Hedgehog signaling pathway, an important pathway in embryonic development. Binding of the Hedgehog ligand to the Patched receptor causes its dissociation from the associated Smo receptor, and activation of a cascade that leads to gene expression driven by the Gli transcription factor.

Treatment of nude mice bearing Ptch-deficient tumors with the LDE225 Smo-inhibitor led to a dose-dependent decrease in tumor volume, with complete regression at 20 mg/kg. However, continued treatment led to relapse, and the researchers found that relapse was delayed at doses higher than 20 mg/kg. This finding indicates that using complete regression as an end point may not be the best way to determine the optimal dose of a drug.

Tumors were harvested and analyzed during the sensitive period and the resistant period to understand the change in their drug response over time. In addition to finding that the Hedgehog pathway was reactivated in resistant cells, the researchers found that the PI3K/Akt pathway played a role in resistance to the Smo inhibitor. This result was supported by the finding that combination therapy with a PI3K inhibitor and the Smo inhibitor decreased relapse in the mouse model.

To duplicate the breadth of the cell line encyclopedia in mouse models, the Novartis team developed a new "encyclopedia" by implanting primary human tumors into nude mice and establishing tumors that can be removed, cut into pieces, frozen, and reimplanted. Thus far they have implanted more than 1000 human tumors and about 200 have become established. Interestingly, over time mouse stroma replaces human stroma as the tumors are passaged through the mice. This creates a situation where signaling in the stroma can be differentiated from that in the epithelial cells. In a primary pancreatic xenograft sample, there is no evidence of the human Gli transcription factor, but mouse Gli is turned on. LDE225 is able to turn off Gli signaling in the stroma, which indicates that there must be paracrine signaling between the stroma and the human epithelial cells.

Figure 19. Pancreatic tumor growth post implantation in nude mice.

Thus, Novartis has established two methods to comprehensively analyze cancerous cells, one for cultured cell lines and another for transplantable tumors, and has shown that they are both useful for testing new compounds and for finding combination treatments that may improve cancer therapy.

MEK inhibitors for pancreatic cancer

Pancreatic cancer is one of the more aggressive and difficult to treat cancers. Judith Sebolt-Leopold of the University of Michigan described the comprehensive genomic analysis of human primary pancreatic cancers carried out by Kinzler and his colleagues at Johns Hopkins to determine the underlying genomic defects in the disease. They found that there were an average of 63 genetic alterations in these tumors that fell into 12 core signaling pathways. Sebolt-Leopold discussed efforts to target the KRAS pathway, which was mutated in 100% of the tumors studied, and which has proven to be a difficult target in drug discovery.

While Sebolt-Leopold was at the pharmaceutical company Parke Davis/Pfizer, her team screened a chemical library for small molecule inhibitors of MEK, a downstream kinase in the RAS-MAP kinase pathway. Early MEK inhibitors that came out of this work, such as PD98059, did not possess the requisite pharmaceutical properties to be developed into drug candidates. However, these early compounds proved to be invaluable biological tools for probing the role of the RAS-MAPK pathway in driving tumor growth and survival. Interestingly, work by Neal Rosen's group showed that though BRAF and MEK lie in the same signaling pathway, inhibitors of these two kinases exhibit distinctly different profiles. Rosen's laboratory showed that MEK inhibitors are effective in inhibiting a significant franction of RAS mutated tumors, where selective BRAF inhibitors, e.g. PLX4032, are ineffective. Further analysis by this group as well as others showed that selective BRAF inhibition paradoxically activates ERK signaling in non-BRAF tumors.

Xenograft models have provided additional clues on which patients to treat and which patients not to treat with MEK inhibitors. Current data suggest that clinical trials for MEK inhibitors should be stratified to focus on patients whose tumors harbor RAS and BRAF mutations, excluding patients whose tumors are wild type. Nevertheless, early clinical trials of MEK inhibitors on melanoma patients were not stratified, said Sebolt-Leopold. In addition, it is unclear whether the toxicities of the MEK inhibitors that led to termination of early clinical trials were mechanistic-based. Currently there are nearly a dozen MEK inhibitor candidates that have entered clinical development. It is likely that many of these agents will not be hampered by the toxicities seen with the earliest MEK inhibitor clinical candidates. However, it remains unclear whether chronic daily dosing is ultimately the optimal regimen for MEK inhibitor-based therapies.

Figure 20. MRI of three patients with glioblastoma multiforme. Red regions show increased diffusion; green and blue regions indicate decreased or unchanged diffusion. This study shows a potential clinical role for MRI in detecting early response to treatment. (Adapted from Hamstra et al., (2005) Proc. Natl. Acad. Sci. 102:16759-64)

Sebolt-Leopold is taking an imaging approach to optimizing regimens for MEK inhibitors in the treatment of pancreatic cancer. She is starting with a transgenic mouse model characterized by the spontaneous development of KRAS driven pancreatic tumors that histologically recapitulate the human disease. She is incorporating apoptosis and MEK reporters in this model to non-invasively monitor treatment efficacy and target inhibition. Anatomical MRI will be used to assess tumor burden and the apoptosis reporter will be used to optimize the dosing schedule. Sebolt-Leopold plans to test MEK inhibitors in combination with radiation or gemcitabine treatment. Her system will be able to correlate therapeutic outcomes with the dynamics of MEK target modulation. This optimization of mouse model treatment with MEK inhibitors should give a better picture of the promise of these compounds in clinical pancreatic cancer therapy.

Enhancing immune response to cancer

Scientists have been trying for many years to harness the power of the immune system to kill tumor cells, but they have had little success. James Allison of Memorial Sloan-Kettering Cancer Center described a huge breakthrough in the field—the development of an antibody that prevents downregulation of the immune response to tumor antigens.

To activate T cells, the T-cell receptor and a co-stimulatory molecule, CD28, must be engaged simultaneously by antigen-presenting cells. After engagement, T cell proliferation is stimulated. However, to keep the immune system under control, after T cell activation the immune cells produce the CTLA-4 molecule, which begins to downregulate the response by displacing CD28 from the immunological synapse. This regulatory mechanism may explain why attempts to get an enduring response from the immune system have been unsuccessful thus far, and suggests that blocking the CTLA-4 molecule could enhance tumor-specific immune responses. An advantage of this strategy is that it could be used against all tumor types and in combination with other treatments.

In a mouse model of transplantable colon carcinoma, though cancer progression was initially observed, treatment with an antibody to CTLA-4 ultimately resulted in complete tumor regression and lifetime immunity to that tumor. In a melanoma model, combination of the anti-CTLA-4 treatment with a GM-CSF tumor cell vaccine (GVAX) resulted in a complete response though neither treatment was effective on its own.

The anti-CTLA-4/GVAX therapy activates the tumor vasculature and increases infiltration of tumors by CD4 and CD8 effector cells (Figure 21). Activation of the immune system requires presentation of tumor antigens; thus any treatment that cause cell death, releasing tumor antigens, or primes T cells serves as a good co-therapy.

Figure 21. aCTLA-4/GVax increases the number of tumor infiltrating T cells.

A phase 3 clinical trial of a human antibody against CTLA-4, ipilimumab, was conducted with 676 patients with unresectable stage III or stage IV melanoma. The results, which were published in the New England Journal of Medicine two months after the meeting, showed a significant improvement in the median overall survival of these patients. There was no standard of care to compare the treatment with because no treatment before this one had been shown to improve the outcome in these patients.

Efforts to distinguish potential responders from non-responders showed that prior existing immunity or rapid induction of immunity might be a predictor of response to ipilimumab. Scientists can determine whether prior immunity exists by looking for the presence of antibodies to a molecule that is expressed in many tumors of metastatic melanoma patients. In addition, clinical benefit was correlated with the stimulation of a pathway that promotes T cell survival.

Therapies that stimulate the immune system to attack tumors are a powerful new addition to the battle to treat cancer. Patients with cancer types that were thought to be untreatable, such as metastatic melanoma, have for the first time had a positive response to the immunostimulatory therapy.

Presenters:
Brian Pollok, Life Technologies
Klaus Pantel, University Medical Center, Hamburg-Eppendorf
Quyen Nguyen, University of California, San Diego
John McPherson, Ontario Institute for Cancer Research, Toronto (presented by Brian Pollok)
Michèl Schummer, The Fred Hutchinson Cancer Research Center

Highlights

• Circulating tumor cells and disseminated tumor cells can be detected using a novel cell sorting process.
• Fluorescent labeling of tumors and nerves may help surgeons remove tumors completely without damaging nearby nerves.
• Next-generation sequencing techniques enable researchers to perform many analyses that were cost prohibitive or unfeasible in the past.
• Publicly available data can be mined for information about possible cancer biomarkers.

The development of personalized medicine requires technologies that can process samples and analyze large amounts of data rapidly and inexpensively. Basic research, drug discovery, and clinical care all rely heavily on such technologies. Brian Pollok of Life Technologies emphasized the need to develop technologies that work together, from the determination of sequences to gaining an understanding of the function of biological molecules and systems, covering the entire workflow.

Detection of circulating tumor cells

Useful technologies have to advance with, and sometimes ahead of, the biological understanding of the system under investigation. For instance, scientists know that early tumor cell dissemination is a key step in tumor progression, so it is clearly important to be able to detect cancer cells as they spread. However, detection of circulating tumor cells (CTCs) and disseminated tumor cells (DTCs) requires the ability to pick out one tumor cell in the background of 10⁵–10⁷ normal cells. The enrichment methods that are currently available for positive selection of CTCs/DTCs are all flawed in some way because tumor cells are heterogeneous. Klaus Pantel of University Medical Center Hamburg-Eppendorf and his team decided to approach the problem from a different angle, by weeding out the leukocytes using an antibody to the CD45⁺ marker expressed on those cells (negative selection) instead of by searching for the few CTCs and DTCs among the leukocytes. For tumor cell identification cells enriched by positive or negative selection are stained for cytokeratin, which is expressed on epithelial cells and serves as a marker for CTCs and DTCs. By using an automated microscope system to look for the stained cells, they found that 36% of primary tumor breast cancer patients and 44.6% of prostate cancer patients with a primary tumor had these keratin-stained DTCs in bone marrow, as the most prominent site of full blown metastasis, whereas only 1% of people with nonmalignant disease had positive staining.

Since it is easier to screen blood than bone marrow, and blood would therefore be the desired way to test patients clinically, Pantel's group validated an automated system called CellSearch™ based on the technique described above, that has been FDA approved for detection of CTCs in the blood. The group was able to show that the number of CTCs in the blood correlated with prognosis in patients with advanced stages of cancer. The ability to detect CTCs can also be very useful in assessing the efficacy of treatment, as was the case in recent clinical trial called Geparquattro in Germany.

In addition to counting the number of CTCs in the blood, it may be useful to characterize the cells according to their expression profiles. Doing just that, Pantel's colleague Sabine Riethdorf found that while cells taken from the primary tumor of a breast cancer patient did not express HER2, CTCs from that patient were HER2⁺. It is not clear whether there was selection for this trait or whether a minor clone in the primary tumor was HER2⁺. The ability to detect HER2⁺ cells after Herceptin treatment will allow the researchers to examine resistance to the treatment.

In collaboration with the company Eppendorf, Pantel's group devised a method of performing molecular characterization of a single CTC with PCR analysis or whole genome amplification. Comparative genome hybridization and array hybridization can identify amplifications and deletions in the sample.

To perform functional characterization of CTCs, Pantel's group, in collaboration with Alix-Panabières' team, developed the EPISPOT assay, where cells are cultured in a dish with antibodies to secreted tumor proteins, the cells are washed off, then a set of fluorescent-conjugated antibodies is used to detect the bound antibodies. This assay can be used in several perturbations for functional studies. Cells lines of DTCs have also been established for functional analyses.

The advances in detecting and characterizing circulating/disseminated tumor cells will help researchers estimate risk for metastatic relapse or progression, stratify patients and monitor therapy, identify therapeutic targets and resistance mechanisms, and gain greater insights into the biology of metastatic development, Pantel noted.

Novel techniques to enhance surgery

Figure 23. Fluorescent nerve labeling could help surgeons avoid dangerous incisions during tumor excision operations. Lower panel shows the efficacy of combining fluorescent labeling techniques for highlighting nerves.

Although drug discovery gets most of the attention in discussions of cancer treatment, Quyen Nguyen of the University of California, San Diego, reminded the audience that surgery is still the primary way to treat the disease when it is first detected. Nguyen and her colleagues are developing new technologies to aid surgeons in their efforts to remove tumors completely without damaging nearby nerves. To do this, they have constructed activatable cell penetrating peptides, fluorescently labeled polycationic peptides coupled via a cleavable linker to a neutralized polyanionic peptide. Proteases expressed by diseased tissues cleave the linker and the activated peptide binds to the diseased tissue. This enables surgeons to visualize the tumor more effectively and ensure that they are not leaving behind cancerous tissue. Using a related technique, they have been able to fluorescently label nerves (Figure 23), which will help surgeons avoid severing them during tumor removal.

"Next generation sequencing"

Technology that facilitates inexpensive sequencing of individual genomes will be required to realize the goal of personalized medicine fully. Brian Pollok, speaking for John McPherson of the Ontario Institute for Cancer Research in Toronto, reviewed the three main platforms for "next generation sequencing" and presented examples of the kinds of information about genetic alterations these techniques can produce. Innovations in genomic sequencing techniques have allowed researchers to identify chromosomal rearrangements, compare the genome of tumor cells with normal cells from the same patient, and find personalized tumor biomarkers, such as chromosomal rearrangements, to study the progression of disease and the efficacy of therapy. Going forward, there is intense interest in single molecule sequencing that requires no pre-amplification and that would allow researchers to establish haplotypes (a set of alleles that are located on the same chromosome).

Looking for biomarkers in public databases

There is a lot of gene expression information available in public databases that can be mined for breast cancer research. Michèl Schummer of the Fred Hutchinson Cancer Research Center described his work looking for early markers of poor-prognosis breast cancer, comparing tumor and adjacent tissue samples from poor-outcome breast cancer patients and tissue from healthy women who had undergone breast reduction surgery. Using the public information to guide their choice of potential targets, he used PCR to screen for differential expression among the three groups. He found three genes that showed elevated expression in cancers with poor outcome measures and low expression in all other tissues. These might be useful as blood markers for early detection of breast cancers with poor prognoses.

All of the techniques presented in the technology workshop highlight the important contribution of technological advances in moving research forward.