Proefschrift Groenendijk by Nicole Nijhuis

Cancer Diagnostics: The Future Ainâ&#x20AC;&#x2122;t What It Used to Be

Floris Groenendijk

The research described in this thesis was performed at the Division of Molecular Carcinogenesis of the Netherlands Cancer Institute (Amsterdam, The Netherlands) with financial support by a grant from the European Research Council (ERC). ISBN: 978-94-6108-882-6 Cover design: F.H. Groenendijk Cover illustration: iStock – www.istockphoto.com Layout and printing: Gildeprint – www.gildeprint.nl, with financial support from the Netherlands Cancer Institute. Printed on FSC certified paper

Cancer Diagnostics: The Future Ainâ&#x20AC;&#x2122;t What It Used to Be De toekomst van kankerdiagnostiek is niet meer wat het geweest is (met een samenvatting in het Nederlands)

Proefschrift

ter verkrijging van de graad van doctor aan de Universiteit Utrecht op gezag van de rector magnificus, prof.dr. G.J. van der Zwaan, ingevolge het besluit van het college voor promoties in het openbaar te verdedigen op woensdag 11 februari 2015 des middags te 2.30 uur

door

Floor Hendrik Groenendijk geboren op 29 juni 1986 te Alblasserdam

Promotor: Prof.dr. R. Bernards

Table of Contents Chapter 1 General Introduction Chapter 2 Drug resistance to targeted therapies: dĂŠjĂ vu all over again

Review, published in Molecular Oncology 2014

Chapter 3

Sorafenib synergizes with metformin in NSCLC through AMPK pathway activation

7 17

Article, published in International Journal of Cancer 2014

Chapter 4 Estrogen receptor splice variants as a potential source of false-positive estrogen receptor status in breast cancer diagnostics

Article, published in Breast Cancer Research and Treatment 2013

Chapter 5 ERBB2 missense mutations characterize a subgroup of muscle-invasive bladder cancers with complete response to neoadjuvant chemotherapy

Article, submitted for publication

Chapter 6

Copy number alterations and subtyping in muscle-invasive bladder cancer correlate with response to neoadjuvant chemotherapy

103

129

Article, in preparation

Chapter 7

General Discussion

Addendum Summary Samenvatting Publication list Dankwoord About the author

149

165 167 170 173 177

“All cancers are alike but they are alike in a unique way.” ― Siddhartha Mukherjee, The Emperor of All Maladies

Chapter 1 General Introduction

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39 8 | Chapter 1

Several decades of cancer research have greatly increased our understanding of cancer biology and, most importantly, extended the survival probability and improved quality of life for patients diagnosed with cancer. It would be very difficult to even enumerate all the fundamental, translational and clinical research that led to increased understanding of cancer and improvements in prevention, diagnosis and treatment of cancer. The shortest but perhaps also the most comprehensive summary of all this deeper understanding came from Douglas Hanahan and Robert Weinberg, stating in their ‘The Hallmarks of Cancer’ paper that “cancer research has generated a rich and complex body of knowledge, revealing cancer to be a disease involving dynamic changes in the genome” (Hanahan and Weinberg, 2000). This thesis is concentrated around the framework of cancer being a disease of the genome and cancer medicine moving towards personalized or precision medicine. This chapter is intended as a brief introduction to the research described in this thesis. A short history of cancer genome sequencing Advances in DNA-sequencing technology powerfully illustrate the impact and driving force of technology in current cancer research. The first papers describing practical methods of DNA sequencing were published in 1977 (Maxam and Gilbert, 1977; Sanger et al., 1977). These methods required radioisotopic labeling of DNA, handcrafting of large electrophoretic gels, and considerable expertise with biochemical and recombinant-DNA techniques. Although the impact of these early DNA-sequencing methods on biological discovery was immense, the total amount of sequence deposited in GenBank, the central depository for such data (Benson et al., 2014), did not pass one billion base pairs - which is roughly one third of the length of a single human genome - until 1997 (NCBI, 2014). This milestone was reached after the introduction of automated Sanger sequencing instruments. Since then, including data from whole-genome sequencing (WGS) studies, almost 1000 billion base pairs have been deposited, illustrating the still ongoing explosion of genomic data in the last two decades (NCBI, 2014). For the Human Genome Project, formally founded in 1990, it took about 10 years and 3 billion dollars to sequence a single human genome. Now, the newest nextgeneration sequencing (NGS) ultra high-throughput machines can do the same for a few thousand dollars and in about a day, depending on coverage and other specifications. And the cost is still dropping rapidly, with a “$1,000 genome” already being a realistic target. The $1,000 genome price tag is often seen as essential to make whole-genome sequencing cost-effective for medical testing and personalized

General Introduction | 9

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

medicine. Basically, we can now take any human tumor sample, or tumor samples from large cancer patient cohorts, and get data in the form of a very long sequence comprised of four nucleotides (A, C, G, and T), the building blocks of our genome. This allows researchers to identify the genetic alterations that drive tumor growth and metastatic capacity. Targeted therapies The identification of genetic alterations that drive the cancer cells has accelerated the development of small-molecule inhibitors directed towards these drivers, socalled targeted therapies. Although impressive responses to these inhibitors are observed, the majority of these responses are not durable since the cancer cells acquire resistance to the therapy. Anticancer drug resistance has been studied for a long time, but has gained momentum after the introduction of new technologies, such as RNA interference and next-generation sequencing, that are now widely used to study drug resistance. In chapter 2, I reviewed the current landscape of cancer drug resistance and future developments in the use of targeted therapies for cancer treatment. I argue that the way to delay or overcome drug resistance is to anticipate the next move of the cancer cell and to use a combination cocktail of inhibitors together with continuous monitoring of the dynamic changes in the cancer genome as the tumors adapt to the therapy. However, the number of possible random combinations therapies is almost unlimited and therefore we need to develop combinations with a biologic rationale. An example of such a mechanism-based combination is described in chapter 3, where I discovered a combination treatment for non-small cell lung cancer (NSCLC) using the multikinase inhibitor sorafenib and the anti-diabetic drug metformin. This combination was found after the response analysis in a phase II clinical trial with sorafenib in patients with NSCLC. We investigated the mechanism underlying the combination and found that both drugs activate the protein AMPactivated protein kinase (AMPK), a metabolic checkpoint coordinating cellular growth. The combination of these drugs strongly inhibited the mTOR-signaling pathway. From tumor heterogeneity and complexity to intrinsic molecular subtypes Despite common histopathologic observations of intra- and intertumoral heterogeneity in the clinical setting, our knowledge of the extent and causes of tumor heterogeneity has remained indistinct for a long time. This question could be addressed after the advent of next-generation sequencing. Several sequencing studies in breast and renal cell cancer have revealed that intra- and intertumoral heterogeneity is also present at the (epi)genetic level, even on single cell level

10 | Chapter 1

(Gerlinger et al., 2012; Wang et al., 2014). Genetic studies before NGS systematically underestimated the amount of genetic variation and tumor heterogeneity. The genetic and clonal diversity illustrates the complexity of cancer as a disease and has critical consequences for clinical diagnosis and treatment (Burrell and Swanton, 2014; Swanton, 2012). It represents a major challenge for the development of personalized medicine as tumors can easily escape from treatment by adaptation, which is discussed in chapter 2 of this thesis. Moreover, the complexity of tumors implies that we have to identify novel tumor classification strategies that better reflect tumor behavior and integrate molecular information. Starting more than a decade ago, clustering of tumors by differential gene expression profiling has uncovered distinct molecular subtypes of breast cancer. Four major intrinsic subtypes were identified (Luminal A and Luminal B, Basal-like and HER2-enriched) (Perou et al., 2000; Sorlie et al., 2001; The Cancer Genome Atlas Network, 2012). In some studies, a small fifth subtype (Claudin-low) has been found (Hennessy et al., 2009; Herschkowitz et al., 2007; The Cancer Genome Atlas Network, 2012). The idea behind this exercise was that each of these subtypes has unique biology and prognostic features, which has been proven by numerous studies over the years [reviewed in (Prat and Perou, 2011)]. Using a commercial breast cancer subtype classifier (BluePrintâ&#x201E;˘, Agendia), I discovered in chapter 4 a discrepancy between estrogen receptor status and estrogen receptor functionality in a small percentage of breast cancers. This could have important clinical consequences as these tumors may lack a functional response to estrogen and consequently may not respond to hormonal therapy (e.g. tamoxifen). I found that this small subgroup of estrogen receptor positive, basal-type breast cancers had a relatively high expression of a dominant negative estrogen receptor splice variant, ERÎąÎ&#x201D;7. Molecular subtyping has been successfully applied to other cancer types more recently (Alizadeh et al., 2000; Choi et al., 2014; Chung et al., 2004; Collisson et al., 2011; Damrauer et al., 2014; De Sousa et al., 2013; Gibson et al., 2010; Perez-Villamil et al., 2012; Sadanandam et al., 2013; The Cancer Genomic Atlas Network, 2014). In most of the cancer types the identification of subtypes had profound implications for research and treatment of these cancers. I studied the prognostic and predictive value of basal-like and luminal-like subtyping in bladder cancers treated with neoadjuvant chemotherapy. This work, described in chapter 6, revealed that basallike bladder cancers have a lower response rate and tumor downstaging compared to luminal-like bladder cancers. Furthermore, patients with basal-like bladders had a significantly shorter cancer-specific survival and shorter post-recurrence survival.

General Introduction | 11

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

Towards personalized medicine The phrases personalized medicine or precision medicine, which are often used interchangeably, are essentially terms for a revolutionary and new era in (cancer) medicine. They are somewhat misinterpreted, since medical doctors have been making personalized treatment plans since the early days of cancer treatment. However, what is new is that we are able to use -omic information, including genomic analysis, to understand the biology of individual tumors and have the availability of targeted compounds in order to match the deregulated signaling pathways with (cocktails of) effective, targeted compounds. However, the mainstay treatment for several cancer types is still (neo)adjuvant chemotherapy. Understanding the effectiveness of chemotherapy is therefore of high relevance and importance. I investigated in chapter 5 whether genomic aberrations can predict the response to neoadjuvant chemotherapy in muscle-invasive bladder. Using this approach, I found that ERBB2 mutations are strongly associated with complete response to neoadjuvant chemotherapy in bladder cancer. This can be used clinically as a biomarker to select bladder cancer patients that will benefit from chemotherapy. However, despite a complete response to neoadjuvant chemotherapy, some ERBB2-mutant patients still develop recurrent disease. These patients may benefit from ERBB2 tyrosine kinase inhibitors, alone or in combination with chemotherapy. Although promising, personalized cancer medicine is still a challenging multidisciplinary concept, with multiple hurdles on the road. The last chapter of this thesis (chapter 7) discusses the hurdles on the road towards wide clinical application of tumor genotyping and tumor subtyping to further personalize cancer medicine.

12 | Chapter 1

REFERENCES Alizadeh, A.A., Eisen, M.B., Davis, R.E., et al., 2000. Distinct types of diffuse large B-cell lymphoma identified by gene expression profiling. Nature 403, 503-511. Benson, D.A., Clark, K., Karsch-Mizrachi, I., et al., 2014. GenBank. Nucleic Acids Res 42, D3237. Burrell, R.A., Swanton, C., 2014. Tumour heterogeneity and the evolution of polyclonal drug resistance. Mol Oncol 8, 1095-1111. Choi, W., Porten, S., Kim, S., et al., 2014. Identification of distinct basal and luminal subtypes of muscle-invasive bladder cancer with different sensitivities to frontline chemotherapy. Cancer Cell 25, 152-165. Chung, C.H., Parker, J.S., Karaca, G., et al., 2004. Molecular classification of head and neck squamous cell carcinomas using patterns of gene expression. Cancer Cell 5, 489-500. Collisson, E.A., Sadanandam, A., Olson, P., et al., 2011. Subtypes of pancreatic ductal adenocarcinoma and their differing responses to therapy. Nat Med 17, 500-503. Damrauer, J.S., Hoadley, K.A., Chism, D.D., et al., 2014. Intrinsic subtypes of high-grade bladder cancer reflect the hallmarks of breast cancer biology. Proc Natl Acad Sci U S A 111, 3110-3115. De Sousa, E.M.F., Wang, X., Jansen, M., et al., 2013. Poor-prognosis colon cancer is defined by a molecularly distinct subtype and develops from serrated precursor lesions. Nat Med 19, 614-618. Gerlinger, M., Rowan, A.J., Horswell, S., et al., 2012. Intratumor heterogeneity and branched evolution revealed by multiregion sequencing. N Engl J Med 366, 883-892. Gibson, P., Tong, Y., Robinson, G., et al., 2010. Subtypes of medulloblastoma have distinct developmental origins. Nature 468, 1095-1099. Hanahan, D., Weinberg, R.A., 2000. The hallmarks of cancer. Cell 100, 57-70.

Hennessy, B.T., Gonzalez-Angulo, A.M., Stemke-Hale, K., et al., 2009. Characterization of a naturally occurring breast cancer subset enriched in epithelial-to-mesenchymal transition and stem cell characteristics. Cancer Res 69, 4116-4124. Herschkowitz, J.I., Simin, K., Weigman, V.J., et al., 2007. Identification of conserved gene expression features between murine mammary carcinoma models and human breast tumors. Genome Biol 8, R76. Maxam, A.M., Gilbert, W., 1977. A new method for sequencing DNA. Proc Natl Acad Sci U S A 74, 560-564. NCBI, 2014. Growth of GenBank and WGS, GenBank Statistics. National Center for Biotechnology Information., http://www.ncbi. nlm.nih.gov/genbank/statistics. Perez-Villamil, B., Romera-Lopez, A., Hernandez-Prieto, S., et al., 2012. Colon cancer molecular subtypes identified by expression profiling and associated to stroma, mucinous type and different clinical behavior. BMC Cancer 12, 260. Perou, C.M., Sorlie, T., Eisen, M.B., et al., 2000. Molecular portraits of human breast tumours. Nature 406, 747-752. Prat, A., Perou, C.M., 2011. Deconstructing the molecular portraits of breast cancer. Mol Oncol 5, 5-23. Sadanandam, A., Lyssiotis, C.A., Homicsko, K., et al., 2013. A colorectal cancer classification system that associates cellular phenotype and responses to therapy. Nat Med 19, 619-625. Sanger, F., Nicklen, S., Coulson, A.R., 1977. DNA sequencing with chain-terminating inhibitors. Proc Natl Acad Sci U S A 74, 5463-5467. Sorlie, T., Perou, C.M., Tibshirani, R., et al., 2001. Gene expression patterns of breast carcinomas distinguish tumor subclasses with clinical implications. Proc Natl Acad Sci U S A 98, 10869-10874.

General Introduction | 13

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

Swanton, C., 2012. Intratumor heterogeneity: evolution through space and time. Cancer Res 72, 4875-4882.

The Cancer Genomic Atlas Network, 2014. Comprehensive molecular characterization of gastric adenocarcinoma. Nature 513, 202-209.

The Cancer Genome Atlas Network, 2012. Comprehensive molecular portraits of human breast tumours. Nature 490, 61-70.

Wang, Y., Waters, J., Leung, M.L., et al., 2014. Clonal evolution in breast cancer revealed by single nucleus genome sequencing. Nature 512, 155-160.

14 | Chapter 1

General Introduction | 15

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

Chapter 2 Drug resistance to targeted therapies: déjà vu all over again

Floris H. Groenendijk and René Bernards

Division of Molecular Carcinogenesis, Cancer Genomics Centre, The Netherlands Cancer Institute, Plesmanlaan 121, 1066 CX Amsterdam, The Netherlands.

Published in Molecular Oncology 2014;8:1067-1083.

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

ABSTRACT A major limitation of targeted anticancer therapies is intrinsic or acquired resistance. This review emphasizes similarities in the mechanisms of resistance to endocrine therapies in breast cancer and those seen with the new generation of targeted cancer therapeutics. Resistance to single-agent cancer therapeutics is frequently the result of reactivation of the signaling pathway, indicating that a major limitation of targeted agents lies in their inability to fully block the cancer-relevant signaling pathway. The development of mechanism-based combinations of targeted therapies together with non-invasive molecular disease monitoring is a logical way forward to delay and ultimately overcome drug resistance development.

KEYWORDS Anticancer therapy  Drug resistance  Targeted therapy  Pathway reactivation  Endocrine therapy  Drug combinations

18 | Chapter 2

INTRODUCTION Resistance to therapy remains a major challenge in oncology. Resistance comes in two flavors: (1) early intrinsic resistance (also known as innate or de novo resistance) or fast adaptive tumor responses, and (2) late acquired resistance, resulting from clonal evolution of resistant variants. Anticancer drug resistance has been studied since the 1960s (Brockman, 1963), but has gained momentum after the introduction of targeted cancer therapeutics and several technological advances such as RNA interference (Brummelkamp et al., 2002) and next-generation DNA/RNA sequencing. Selective targeting of activated pathways has proven to be effective, but the observed responses are usually partial and not durable when using single agent therapies. This translates clinically in prolonged progression-free survival, but similar overall survival compared to standard of care. Examples where prolonged progression-free survival has been achieved without giving rise to improved overall survival are crizotinib (ALK-TKI) in advanced ALK-positive Non-Small Cell Lung Cancer (NSCLC) and gefitinib (EGFR-TKI) in EGFR-mutated NSCLC (Maemondo et al., 2010; Shaw et al., 2013). An exception is the case of the BRAFV600E-specific inhibitor vemurafenib in BRAFV600E-mutated metastatic melanoma. Patients with metastatic melanoma have a median survival of 6 to 10 months and activating BRAFmutation is associated with shortened survival in patients with metastatic disease (Long et al., 2011). An early interim analysis of a phase 3 trial with vemurafenib in BRAFV600E-mutated metastatic melanoma showed significant improvement in both progression-free survival and overall survival with vemurafenib compared to chemotherapy (Chapman et al., 2011). Although the median duration of follow-up in this study was too short to draw strong conclusions, long follow-up data of a phase 2 trial with vemurafenib in the same clinical setting confirmed these early results and showed increase in median overall survival to approximately 16 months (Sosman et al., 2012). Here we review the recent insights into mechanisms of resistance to targeted therapies. We focus on the reactivation of signaling pathways as a recurrent pattern of resistance development to single-agent targeted therapies. We first discuss the resistance mechanisms to endocrine therapy in breast cancer, the first targeted therapy introduced in the clinic. We will use this as an example to highlight that the mechanisms of resistance to endocrine therapy that have been identified in breast cancer are seen all over again with the new pathways-targeted therapies in other cancers. Finally, we argue that synthetic lethal combinations of targeted therapies together with non-invasive molecular disease monitoring are a promising way forward to fight drug resistance.

Drug resistance to targeted therapies: dĂŠjĂ vu all over again | 19

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

ENDOCRINE RESISTANCE IN BREAST CANCER The synthesis of competitive inhibitors of the binding of the hormone estrogen to its receptor (ERα) in the 1970s led to the development of the first targeted cancer drug: tamoxifen. Tamoxifen is a triphenylethylene derivative classified as a Selective Estrogen Receptor Modulator (SERM). It impairs the mitogenic function of ERα in breast cancer by competing with estrogen for binding to the receptor. The binding of tamoxifen to the ERα changes the receptor conformation, which is distinct from the conformational change that is induced by estrogen binding. This conformation change prevents the formation of the ERα complex with its essential transcriptional co-activators and thereby inhibits ERα-mediated transcription (Shiau et al., 1998). A second class of endocrine drugs that target estrogen synthesis has been developed subsequently: the aromatase inhibitors. Aromatase is the enzyme responsible for the estrogen synthesis from androgenic substrates (extra-ovarian synthesis) (Smith and Dowsett, 2003). Aromatase inhibitors cannot inhibit the estradiol production in the ovaries themselves and are therefore not active in premenopausal patients without ovarian suppression. Consequently, tamoxifen is typically given to premenopausal patients, whereas aromatase inhibitors are given to postmenopausal patients, although postmenopausal sequential treatment of tamoxifen and aromatase inhibitors is often prescribed as well. Almost 70% of breast cancers are classified as ERα-positive by IHC, and endocrine therapies targeting estrogen action (anti-estrogens and aromatase inhibitors) are only effective in ERα-positive breast cancers. Expression of ERα protein is strongly predictive of response to endocrine therapies (Davies et al., 2011). However, approximately one third of ERα-positive early breast cancers do not respond to endocrine therapy (intrinsic resistance) or relapse after an initial response (acquired resistance) (EBCTCG, 2005). The proportion of breast cancer patients with advanced or metastatic disease that relapses during or after endocrine therapy is even higher. It is important to note that patients who develop resistance to one kind of endocrine treatment can still respond to another type (Wang et al., 2009; Yoo et al., 2011). The various mechanisms underlying resistance to endocrine therapy that have been proposed and studied are outlined below. They can be classified in three main categories (Table 1): (1) alterations of the drug target (i.e. ESR1/ERα), (2) alterations in downstream and upstream effectors of ERα signaling, and (3) bypass mechanisms. Alterations of ESR1 and its encoded protein ERα Patients with the highest ERα protein expression benefit slightly more from tamoxifen compared to patients with low receptor expression, but the latter group

20 | Chapter 2

still have substantial benefit (Davies et al., 2011). However, response to tamoxifen is rare in ERα-negative breast cancer. A portion of ERα-positive tumors becomes independent of estrogen signaling after which they loose ERα expression and, hence, are tamoxifen resistant. Gutierrez et al. studied the ERα expression in paired clinical breast cancer samples from before the start of tamoxifen treatment and after tumor progression (Gutierrez et al., 2005). They found that loss of ERα expression at the time of tumor progression occurred in 17% of ERα-positive tumors. This was in line with earlier reports showing that ERα loss occurs in 15-30% of the tumors at the time of recurrence (Encarnacion et al., 1993; Johnston et al., 1995; Kuukasjarvi et al., 1996). Loss of ERα is associated with tamoxifen resistance (Johnston et al., 1995) and can be used as a predictor of poor response to subsequent endocrine therapy (Kuukasjarvi et al., 1996). Mutations in ESR1, the gene coding for ERα, were proposed as another possible mechanism of endocrine therapy resistance. However, ESR1 mutations were only found in a very low percentage of primary breast cancers, if at all present (Cancer Genome Atlas Network, 2012). In a recent report by Li et al., ESR1 ligand-binding domain mutations were identified in xenografts derived from primary or metastatic breast cancer patients who were resistant to hormonal treatment (Li et al., 2013). This report was followed by three independent studies showing that activating ESR1 mutations are relatively frequent in advanced ERα-positive hormone-resistant breast cancers (Jeselsohn et al., 2014; Robinson et al., 2013; Toy et al., 2013). The mutation frequency was somewhat variable in the different cohorts they analyzed (range 11 to 55%), with an average frequency of 17.3% in 167 cases. The authors also show that these ESR1 mutations were not present at detectable levels in the primary tumors, indicating that the specific mutations are either clonally selected or arise de novo under the pressure of endocrine therapy with tamoxifen or aromatase inhibitors. In addition, the study by Jeselsohn et al. showed absence of these mutations in any stage of ERα-negative disease (Jeselsohn et al., 2014). The mutations are clustered in the ligand-binding domain of ERα, similar to what was reported by Li et al. for ESR1 mutations in the patient-derived xenograft (PDX) models. Mutations in this domain result in ligand-independent constitutive activity. On the other hand, Toy et al. reported no significant difference in sensitivity to aromatase inhibitors for ESR1 wild-type or ESR1-mutant patients, but the number of patients was small. The three studies together demonstrate that mutant ERα can still bind anti-estrogens like tamoxifen, although higher doses are required. Further studies are needed to fully address the association with ESR1 mutations and clinical outcome.

Drug resistance to targeted therapies: déjà vu all over again | 21

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

22 | Chapter 2

by downstream components

by upstream components

MYC activation 11

AIB1 (SRC3) overexpression 10

PKA 9

BCAR1 overexpression 8

BRAF-/MEK-inhibitor (melanoma)

trastuzumab (breast cancer)

PTEN loss and PIK3CA mutation MEK1/2 mutations 29

cetuximab (CRC)

BRAF-inhibitor (melanoma)

BRAF-inhibitor (melanoma, CRC, thyroid); MEK inhibitor (TNBC); AKT inhibitor

BRAF-inhibitor (melanoma)

EGFR-TKI (GBM)

imatinib (CML) BRAF-inhibitor (melanoma) EGFR-TKI (NSCLC)

KRAS mutation and amplification 27

RAS mutations 26

RAS pathway activation 7

BRAF splice variant 24

RTK activation 25

ERÎą splice variants 4

Splice variant

Loss of mutant EGFR expression 23

BCR-ABL amplification 20 BRAF amplification 21 EGFR amplification 22

cetuximab (CRC)

EGFR ectodomain mutation 19

HER family activation 5 PI3K-AKT-mTOR activation 6

Loss of ERÎą expression 3

Receptor downregulation

Pathway activation

ESR1 amplification 2

Amplification

imatinib (CML) EGFR-TKIs (NSCLC) crizotinib (NSCLC) crizotinib (NSCLC)

Drug (cancer type)

ABL T315I mutation 15 EGFR T790M mutation 16 ALK F1147L mutation 17 ROS1 mutation 18

Example

Example ESR1 mutation / translocation 1

Targeted therapies in other cancers

Endocrine therapies breast cancer

Alterations of the drug target Mutation

Drug resistance mechanism

Table 1. Similarity in the mechanisms of resistance to endocrine therapies in breast cancer and to targeted therapies in other cancers.

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

Drug resistance to targeted therapies: dĂŠjĂ vu all over again | 23

Loss of CDK10 expression 14

Metabolic activity of CYP2D6

Notch activation 12 13

MET amplification

EMT 35

HGF secrection 34

PI3K pathway activation 33 Multiple TKIs

BRAF-inhibitor (melanoma)

EGFR-TKI (NSCLC); cetuximab (CRC)

BRAF-inhibitor (melanoma)

COT expression 32

BRAF-MEK-inhibitor (melanoma)

PKA activation 30

(Cancer Genome Atlas Network, 2012; Jeselsohn et al., 2014; Li et al., 2013; Robinson et al., 2013; Toy et al., 2013) (Albertson, 2012; Ooi et al., 2012) 3 (Encarnacion et al., 1993; Gutierrez et al., 2005; Johnston et al., 1995; Kuukasjarvi et al., 1996) 4 (Groenendijk et al., 2013; Herynk and Fuqua, 2004; Shi et al., 2009) 5 (Arpino et al., 2004; De Laurentiis et al., 2005; Gutierrez et al., 2005; Osborne et al., 2003) 6 (Miller et al., 2011a; Yamnik and Holz, 2010) 7 (McGlynn et al., 2009; Yamnik and Holz, 2010) 8 (van der Flier et al., 2000) 9 (Holm et al., 2009; Michalides et al., 2004) 10 (Osborne et al., 2003) 11 (Miller et al., 2011b) 12 (Magnani et al., 2013; Rizzo et al., 2008) 13 (Goetz et al., 2005; Hoskins et al., 2009; Singh et al., 2011) 14 (Iorns et al., 2008) 15 (Gorre et al., 2001; Michor et al., 2005) 16 (Kobayashi et al., 2005; Sequist et al., 2011) 17 (Choi et al., 2010; Katayama et al., 2012; Sasaki et al., 2010) 18 (Awad et al., 2013) 19 (Montagut et al., 2012) 20 (Gorre et al., 2001) 21 (Shi et al., 2012) 22 (Sequist et al., 2011) 23 (Nathanson et al., 2014) 24 (Poulikakos et al., 2011; Shi et al., 2014) 25 (Chandarlapaty et al., 2011; Corcoran et al., 2012; Duncan et al., 2012; Girotti et al., 2013; Montero-Conde et al., 2013; Nazarian et al., 2010; Prahallad et al., 2012; Sun et al., 2014a,b; Villanueva et al., 2010) 26 (Nazarian et al., 2010; Shi et al., 2014; Trunzer et al., 2013) 27 (Diaz et al., 2012; Misale et al., 2012; Valtorta et al., 2013) 28 (Berns et al., 2007; Majewski et al., 2014; Nagata et al., 2004) 29 (Emery et al., 2009; Trunzer et al., 2013; Van Allen et al., 2014; Wagle et al., 2011) 30 (Johannessen et al., 2013) 31 (Johannessen et al., 2010) 32 (Bardelli et al., 2013; Engelman et al., 2007; Turke et al., 2010; Sequist et al., 2011) 33 (Shi et al., 2014) 34 (Straussman et al., 2012) 35 (Brunen et al., 2013; Byers et al., 2013; Fuchs et al., 2008; Huang et al., 2012; Salt et al. 2014, Thomson et al., 2005; Yauch et al., 2005)

Bypass mechanism

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

The study by Li et al. on breast cancer patients-derived xenografts reported an ESR1/ YAP1 gene rearrangement that replaces the ligand binding and AF2 domains of ESR1 (Li et al., 2013). This translocation induces estradiol-independent growth and makes cells insensitive to the selective ER down-regulator fulvestrant. Similar ESR1 translocations, that preserve the DNA-binding and AF1 domains of ESR1, were detected by the authors in the TCGA breast cancer RNA-seq data (Li et al., 2013). The frequency of these in-frame ESR1 translocations might be very low, but can still be of clinical significance considering the high number of breast cancer patients. Several different ERα variant mRNAs have been described in human breast cancer. Almost all of these naturally occurring variants are mRNA splicing variants, in which one or more exons are absent from the ERα mRNA. In most ERα splicing variants, except for variants lacking exon 3 or 4, translation runs out of frame after the site of splicing, leading to a truncated protein (Herynk and Fuqua, 2004). One identified variant, the ERα36, is transcribed from an alternative promoter in the first intron of the ESR1 gene (Shi et al., 2009). ERα36 mediates in vitro membraneinitiated effects of estrogen signaling, and tamoxifen treatment fails to block the ERα36 mediated activation of the MAPK pathway. Shi et al. also found that breast cancer patients with high ERα36 expression are less likely to benefit from tamoxifen treatment. Two ERα splice variants, the ERΔ3 and the ERΔ7 variant, were described as dominant negative receptor forms in the presence of wild-type ERα (Herynk and Fuqua, 2004). The expression of the ERΔ7 is associated with a basal subtype of breast cancers expressing ERα mRNA (Groenendijk et al., 2013). Gene amplification is defined as a focal high-level copy number increase of a region of genomic DNA. Gene amplification usually goes hand in hand with gene overexpression. Overexpression of ERα in cell lines caused broad anti-estrogen resistance in vitro (Tolhurst et al., 2011). Holst et al. reported in 2007, using FISH on tissue microarrays, that ESR1 was amplified in 21% of breast cancers and that these amplifications were present in breast cancers with high ERα expression. Multiple groups have since tried to reproduce these findings, reporting a considerable difference in amplification frequencies when using in situ hybridization compared to biochemical methods (aCGH, MLPA, qPCR) (as summarized in (Albertson, 2012)). It was later found that the large clustered FISH signals, interpreted as ESR1 amplification, are sensitive to RNase treatment (Ooi et al., 2012). This indicates that the FISH probe was detecting accumulation of ESR1 mRNA transcripts in the nucleus of cells expressing high levels of ERα, providing an explanation for the controversy in ESR1 amplification frequencies between different studies. Although this finding of RNase sensitivity of ESR1 FISH signals has yet to be reproduced, it seems likely that the amplification frequency will be around 5% of breast cancers. Despite this small

24 | Chapter 2

percentage, there may still be a (modest) role of ESR1 amplification in endocrine therapy resistance. Alterations in upstream and downstream effectors of ERα signaling Both preclinical and clinical studies suggest that crosstalk between ERα and growth factor and stress kinase signaling pathways is involved in endocrine therapy resistance (as reviewed in (Musgrove and Sutherland, 2009; Osborne and Schiff, 2011)). Among those, the crosstalk between ERα and the HER family is best studied. The HER/ERBB family consists of 4 structurally related membrane receptors: HER1 (EGFR), HER2 (ERBB2), HER3 (ERBB3), and HER4 (ERBB4). The tyrosine kinase HER2 is amplified in 10-20% of ERα-positive breast cancers. It has unique receptor features compared to the other receptors of the family: it has no specific ligand, but forms homodimers with itself and heterodimers with other ERBB receptors. Tamoxifen activates HER2 via the membrane functions of ERα, and this activated HER2 phosphorylates both ERα and AIB1 (SRC3). There is clinical evidence that tamoxifen is less effective in ERα-positive breast cancer overexpressing HER2 (De Laurentiis et al., 2005; Osborne et al., 2003). Furthermore, a small percentage of tumors originally negative for HER2 showed HER2 amplification or overexpression after tumor progression with tamoxifen (Gutierrez et al., 2005). Although HER1 (EGFR) signaling downstream is not identical to HER2 signaling, there is significant overlap. In a study by Arpino et al. it was shown that patients with EGFR overexpression were less likely to respond to tamoxifen and had shorter time to treatment failure (Arpino et al., 2004). Aromatase inhibitors may be different, because ligand deprivation shuts off all nuclear and membrane ERα activity, disallowing crosstalk of ERα with other pathways. Increasing evidence supports an interaction between the mTOR pathway and ERα signaling, and hyperactivation of the PI3K-AKT-mTOR pathway promotes anti-estrogen resistance (Miller et al., 2011a). The mTOR-substrate S6 kinase 1 can phosphorylate the AF1 domain of ERα, which is responsible for ligand-independent receptor activation (Yamnik and Holz, 2010). The clinical benefit of mTOR inhibition in combination with endocrine therapy has already been demonstrated. The mTOR inhibitor everolimus combined with an aromatase inhibitor improved progressionfree survival compared with aromatase inhibitor alone in postmenopausal ERα-positive advanced breast cancer (BOLERO-2 trial) (Baselga et al., 2012b). Furthermore, the TAMRAD trial reported that tamoxifen plus everolimus increased clinical benefit, time-to-progression and overall survival compared with tamoxifen alone in postmenopausal women with aromatase-inhibitor-resistant metastatic breast cancer (Bachelot et al., 2012).

Drug resistance to targeted therapies: déjà vu all over again | 25

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

Expression and activation of the MAPK pathway is associated with poor outcome on tamoxifen (McGlynn et al., 2009). MAPK activation can phosphorylate serine-118 and serine-167 of ERα, resulting in ligand-independent activation of the receptor and transcription of estrogen-regulated genes and proliferation (Yamnik and Holz, 2010). Phosphorylation of ERα at serine-305 by protein kinase A (PKA) has been associated with tamoxifen resistance (Michalides et al., 2004). This phosphorylation induces a conformational change in the receptor, which prevents the induction to the inactive state by tamoxifen. Patients whose tumors express serine-305 phosphorylated ERα (Holm et al., 2009) or patients with downregulation of a negative regulator of PKA (Michalides et al., 2004) do not benefit from tamoxifen. Phosphorylation of ERα at other sites (serine-118 and serine-167) is associated with anti-estrogen sensitivity as well, although different effects are reported for pre- versus postmenopausal women (Beelen et al., 2012). The function of ERα is regulated by interactions with other transcription factors and through post-translational modification of the receptor itself or its interactors. Overexpression and increased phosphorylation (hence activity) of ERα coactivators leads to constitutive ERα transcriptional output. The activity of nuclear receptor coactivator 3 (NCOA3, also known as AIB1 or SRC3) is associated with reduced responsiveness to tamoxifen in patients (Osborne et al., 2003). The ERα coactivator PELP1 localizes to the cytoplasm in many breast cancers, where it functions as scaffold for the interaction of ERα with SRC that subsequently recruits other cofactors (Gururaj et al., 2006). Modeling cytoplasmic PELP1 localization in vitro makes tumor cells hypersensitive to estrogen but resistant to tamoxifen. The SRC substrate BCAR1 (also known as CAS) can induce tamoxifen resistance in vitro upon overexpression and breast cancers overexpressing BCAR1 are less responsive to tamoxifen (van der Flier et al., 2000). There is extensive evidence from in vitro studies that cell cycle regulators play a role in endocrine therapy resistance. However, clinical evidence supporting this notion is scarce, especially for the prognosis-independent predictive value of cell cycle regulators. Cyclin D1 and MYC are the best studied, although studies for the role of cyclin D1 have yielded contradictory results (Beelen et al., 2012). MYC is amplified in around 15% of breast cancers; MYC protein overexpression and a MYC activation signature were both predictive for the response to endocrine therapy in a large cohort of ERα-positive breast cancers (Miller et al., 2011b).

26 | Chapter 2

Bypass mechanisms Reprogramming of the chromatin landscape through epigenetic modifications or changes in genome accessibility may play a central role in endocrine therapy resistance. Magnani et al. showed significant differences in the transcriptional programs of endocrine therapy-resistant and -responsive breast cancer cell lines (Magnani et al., 2013). The histone H3 lysine 36 trimethylation (H3K36me3) mark is found in actively transcribed gene bodies. Only around 12% of H3K36me3 sites were shared between endocrine therapy-resistant and -responsive cells (Magnani et al., 2013). Furthermore, the authors showed that most of the open chromatin sites are cell type-specific and only a minority is shared between resistant and responsive cells. The genome-wide reprogramming of the chromatin landscape disengage the resistant cells from classical ERα signaling and promotes Notch signaling (Magnani et al., 2013; Poulikakos et al., 2011)). Crosstalk between the Notch pathway and ERα signaling in breast cancer has been reported previously (Rizzo et al., 2008). Pharmacological targeting of Notch signaling can be of interest for endocrine therapy resistant breast cancers (Haughian et al., 2012) Tamoxifen is converted in the liver into its active metabolites endoxifen and 4-hydroxytamoxifen and this conversion is catalyzed by the polymorphic enzyme P450 2D6 (CYP2D6) (Hoskins et al., 2009). CYP2D6 is located on chromosome 22q13.1, and polymorphisms in this gene can significantly affect enzymatic activity. Approximately 6-10% of Caucasian women carry less active or inactive alleles of this gene and are consequently less responsive to tamoxifen (Goetz et al., 2005; Hoskins et al., 2009; Singh et al., 2011). In addition, the metabolic activity of CYP2D6 can be lowered by certain commonly prescribed medications, like for example certain selective serotonin or noradrenaline reuptake inhibitors (Hemeryck and Belpaire, 2002). Finally, Iorns et al. showed both in vitro and in patient series that loss of CDK10 expression resulted in resistance to tamoxifen (Iorns et al., 2008). Loss of CDK10 activates RAF1 (CRAF) and the downstream MEK-ERK signaling cascade, resulting in cyclin D1 expression and subsequent independence of estrogen signaling.

DRUG RESISTANCE TO MORE RECENT TARGETED THERAPIES The recent DNA and RNA sequencing projects in several cancer types by The Cancer Genomic Atlas (cancergenome.nih.gov) and other consortia have revealed recurrent somatic alterations in several genes that are considered to be drivers of oncogenesis. They activate crucial oncoproteins such as RAS, BRAF, EGFR, PIK3CA, ALK, BCR-

Drug resistance to targeted therapies: déjà vu all over again | 27

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

ABL and many others. These driver alterations can give rise to a tumor dependency on a particular signaling pathway or module. These dependencies, called ‘oncogene addiction’, can result from activation of genes that stimulate a pathway or inactivation of genes that inhibit a signaling pathway. Tumor dependencies can also be the result of other factors, like hormone dependency or lineage dependency. The addiction of certain cancer types to specific signaling pathways or modules in those pathways creates an Achilles heel for tumor maintenance that can be exploited therapeutically (Weinstein, 2002). However, examples of pure ‘oncogene addiction’ from a single oncogenic event are rare. Most cancers contain many mutations, making such tumors potentially less dependent on a single oncogenic event. Another complication is that signaling pathways are interconnected and these interactions are often dynamic (Lemmon and Schlessinger, 2010). A third factor limiting the success of targeted single-agent therapy is the heterogeneity of tumors, especially in the advanced or metastatic setting. Similar to resistance mechanisms for endocrine therapy, the mechanisms underlying resistance to other targeted therapies can be classified in three main categories (Table 1): (1) alterations of the drug target (Fig. 1A), (2) alterations in upstream and downstream effectors resulting in pathway reactivation (Fig. 1B) and (3) bypass mechanisms (Fig. 1C). In that sense, the recently-identified mechanisms of resistance to targeted cancer drugs represent a clear case of “déjà vu all over again” (a quote from the legendary American baseball player Yogi Berra (Berra, 1998)). Alterations of the drug target One of the mechanisms of resistance is mutation of the target kinase domain, which reduces the ability of the drug to inhibit the kinase, a so-called ‘gatekeeper’ mutation. Mutations in the ABL kinase domain (T315 residue) confer resistance to the ABL-inhibitor imatinib in chronic myeloid leukemia (CML) patients with the Philadelphia-chromosome translocation (Gorre et al., 2001). These mutations were shown to be pre-existing in subpopulations of tumor cells positively selected by the treatment (Michor et al., 2005). Another well-known example is the EGFR T790M mutation, which abrogates the inhibitory activity of EGFR-TKIs and is present in half of the patients that develop resistance to EGFR-TKIs (Kobayashi et al., 2005; Sequist et al., 2011). A subset of these patients also developed EGFR amplification, where the T790M allele is selectively amplified (Sequist et al., 2011). Analogous to the T790M mutation in EGFR are mutations in the kinase domain of ALK that are found at the time of resistance to crizotinib in tumors with ALK-fusions (Choi et al., 2010; Katayama et al., 2012; Sasaki et al., 2010).

28 | Chapter 2

(A) Alterations of the drug target

(B) Upstream and downstream pathway activation RTK

RTK

RTK-2

Drug

STK2 STK3

STK1

STK1 Drug

STK2 STK3

Drug

STK2 STK3

Drug 3.

STK1

STK4

STK2

STK5

STK3

STK6

Figure 1. Schematic representation of three recurrent mechanisms of resistance to targeted therapies. (A) Alterations of the drug target (e.g. mutation, amplification or splice variants) (B) Pathway reactivation through (1) RTK activation or (2) mutation or amplification of an upstream component or (3) by mutation or amplification of a downstream component. (C) Bypass by parallel pathway activation through (1) activation of a second RTK by itself or through feedback regulation or (2) mutation of a parallel STK. Abbreviation: RTK, receptor tyrosine kinase; STK, serine/threonine kinase. Figure adapted from (Berns and Bernards, 2012)

Resistance by mutating the drug target was also reported for ROS1. Chromosomal translocations that create fusion proteins involving the tyrosine kinase domains of ALK or ROS1 were identified in lung adenocarcinoma, and are of particular interest because of the availability of an active inhibitor of these kinases: the multi-targeted TKI crizotinib. Crizotinib has shown efficacy in the treatment of lung cancers harboring ROS1 translocations (Davies et al., 2012; Ou et al., 2013). Resistance to crizotinib developed in a patient with a CD74-ROS1 translocated metastatic lung adenocarcinoma. This resistance was caused by a mutation in the ROS1 kinase domain that made the kinase insensitive to crizotinib (Awad et al., 2013). Some colorectal cancers with KRAS or BRAF mutations are addicted to the MAPK pathway. Because MEK1/2 are the only substrates of RAF kinases, their inhibition is explored as a treatment strategy to block the MAPK signaling in KRAS and BRAF mutated tumors. However, tumors can bypass this blockade by amplifying the signal that goes through this pathway, for example by amplification of the driving oncogene. In vitro, it was shown that KRAS or BRAF mutated colorectal cancer cell lines made resistant to the MEK-inhibitor had a marked upregulation of their

Drug resistance to targeted therapies: dĂŠjĂ vu all over again | 29

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

respective driving oncogene due to amplification (Corcoran et al., 2010; Little et al., 2011). Amplification of mutant BRAF was also demonstrated in melanoma patients treated with BRAF-inhibitor (Shi et al., 2012). Similarly, resistance to imatinib was associated with progressive BCR-ABL gene amplification (Gorre et al., 2001). Alternative splicing of BRAF has been implicated in drug resistance to vemurafenib in BRAF-mutated melanoma. In vitro, in a subset of cells resistant to vemurafenib, a BRAFV600E-splicing variant was identified. The splice variant encodes a version of BRAF lacking the RAS-binding domain and is resistant to the BRAFinhibitor. This splicing variant for BRAFV600E was also found by PCR analysis in six of the 19 tumors derived from melanoma patients having developed vemurafenib resistance (Poulikakos et al., 2011). Mutations in NRAS were identified in four out of 19 progression samples and were mutually exclusive with BRAF-splicing variants. BRAF-splicing variants were not detected in two samples derived from patients with de novo resistance or in melanoma samples that were never treated with vemurafenib. However, it remains questionable whether these splice variants are in fact acquired during the treatment or were already present in a small subpopulation that was too small for the detection. In a different study analyzing tumor samples from patients whose melanomas developed acquired resistance to either vemurafenib or dabrafenib monotherapy, alternative BRAF splicing was detected in six out of 48 (13%) tumors after progression (Shi et al., 2014). Another type of drug target modification is the finding of an acquired EGFR ectodomain mutation (S492R) that prevents cetuximab binding and confers resistance to cetuximab. This mutation was identified in two out of ten patients with disease progression after cetuximab treatment (Montagut et al., 2012). The selective sensitivity of tumor cells to targeted drugs relies on the expression of and dependency on the oncogenic target. An example is the sensitivity of tumor cells that express mutant EGFR to EGFR-TKIs. As we discussed for endocrine therapy, the downregulation of the oncogenic target is one way for tumor cells to acquire resistance. Single-cell analyses of glioblastoma multiforme (GBM) PDX models and clinical samples from GBM patients treated with EGFR-TKIs have demonstrated that tumor cells reversibly up- or downregulate mutant EGFR expression to reach an optimal state for tumor growth (Nathanson et al., 2014). Resistance to EGFRTKIs occurred by elimination of mutant EGFR from extrachromosomal DNA. The withdrawal of the drug was followed by the re-emergence of EGFR mutations on extrachromosomal DNA. The re-emergence of EGFR mutations suggests that these cells regain sensitivity to EGFR-TKIs, as was indeed shown in the paper. The benefit from intermittent dosing of TKIs instead of continuous dosing is supported by the

30 | Chapter 2

clinical observation that patients can regain sensitivity to a drug after a so-called ‘drug holiday’ (Yano et al., 2005). Recently, it was reported that discontinuous dosing of vemurafenib in a PDX model of BRAFV600E-mutant melanoma delayed the development of drug resistance (Das Thakur et al., 2013). The reversible drug resistance suggests that epigenetic regulatory mechanisms play a role in acquiring this reversible-resistance state. Data supporting the existence of such an epigenetic mechanism came from an shRNA screen performed in our laboratory to address the question whether changes in expression of chromatin regulators could confer resistance to BRAF-inhibitors (Sun et al., 2014b). We found that suppression of sex determining region Y-box 10 (SOX10) in melanoma caused activation of TGFβ signaling, which led to transcriptional upregulation of EGFR and PDGFRB, causing drug resistance. In a heterogeneous population of melanoma cells having varying levels of SOX10 suppression, cells with low SOX10 and consequently high EGFR expression were rapidly enriched in the presence of the BRAF-inhibitor, but this was reversed when the drug treatment was discontinued. These data highlight that conditions that confer selective advantage during drug exposure can be a liability in the absence of drug (Sun et al., 2014b). Signaling pathway reactivation mediated by alterations in upstream or downstream signaling Activation of receptor tyrosine kinases (RTKs) represents a major mechanism for upstream signaling activation. RTKs are high-affinity cell surface receptors for many growth factors, cytokines and hormones. In humans, a total of 58 receptor tyrosine kinase proteins have been identified that can be divided in 20 subclasses or subfamilies (Lemmon and Schlessinger, 2010). All RTKs have a similar molecular architecture, with ligand-binding domains in the extracellular region, a single transmembrane helix, and a cytoplasmic region that contains the protein tyrosine kinase domain plus additional regulatory regions. The mechanisms of RTK activation and regulation have been excellently reviewed by Lemmon and Schlessinger (Lemmon and Schlessinger, 2010). The RTKs are important drug targets, although resistance will eventually develop against every inhibitor of a single RTK. It became clear that all pathways previously thought to be linear are in fact highly interconnected into a complex and dynamic signaling network, with RTKs functioning as key regulatory nodes. However, knowing the components of a signaling pathway or network is not sufficient, due to the presence of positive and negative feedback mechanisms. In recent years, the upregulation of RTKs by feedback mechanisms, following the inhibition of selective kinases has received much attention. We discuss several examples where RTKs play a role in acquired resistance to targeted therapies in the following paragraph. Drug resistance to targeted therapies: déjà vu all over again | 31

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

As discussed, the clinical activity of the BRAFV600E-specific inhibitor in BRAFmutated melanoma is well-established. In sharp contrast to this is the observation that inhibition of BRAF in BRAFV600E-mutant colorectal cancer (CRC) did not result in clinical benefit. It was found that this lack of response to BRAF inhibition in colon cancer is caused by the feedback activation of EGFR upon BRAF inhibition, suggesting a drug combination strategy to make colon cancer cells BRAF-inhibitor responsive (Corcoran et al., 2012; Prahallad et al., 2012). RTK upregulation also plays a role in the acquired resistance of melanomas to BRAFV600E-inhibition (Girotti et al., 2013; Nazarian et al., 2010; Sun et al., 2014b; Villanueva et al., 2010). Induction of PDGFRÎ˛ emerged as a dominant feature of acquired resistance in a subset of melanoma cell lines, patient-derived biopsies and short-term cultures (Nazarian et al., 2010). Furthermore, EGFR or SRC family kinase signaling activation was observed after resistance to vemurafenib in melanoma patients (Girotti et al., 2013; Sun et al., 2014b). The third malignancy with a high frequency of BRAF-mutations is thyroid cancer. Thyroid cancers, like colorectal cancers, are also relatively resistant to BRAF-inhibitors. It was reported that the majority of BRAF-mutant thyroid cancer cell lines are insensitive due to a feedback mechanism that induces HER3 transcription (Montero-Conde et al., 2013). HER3 activation was part of a generalized hyperactivation of RTKs, but HER3 together with HER2 rapidly reactivates the MAPK signaling. Another node in the MAPK pathway, MEK, has been explored as a target in triple-negative breast cancer (TNBC). The inhibition of MEK with a specific inhibitor in TNBC cell lines and in genetically engineered mouse models (GEMMs) resulted in rapid reprogramming of the kinome through induced expression and activation of multiple RTKs that bypassed the initial ERK inhibition (Duncan et al., 2012). The inhibition of MEK caused a rapid loss of ERK activity, resulting in rapid MYC degradation, which in turn induced expression and activation of several RTKs. Apart from inhibition of the MAPK pathway, inhibition of the PI3K-AKT pathway is also associated with activated RTK signaling. Inhibition of AKT induces a conserved set of RTKs, including HER3, IGF1R and insulin receptor (Chandarlapaty et al., 2011). This activation is partially due to mTORC1 inhibition (Oâ&#x20AC;&#x2122;Reilly et al., 2006) and partially due to a FOXO-dependent activation of receptor expression. Using a kinome-centered synthetic lethality screen, we found in our laboratory that suppression of HER3 sensitizes KRAS mutant lung and colon cancer cells to MEK inhibitors (Sun et al., 2014a). It was found that MEK inhibition results in MYC-dependent transcriptional upregulation of HER3, which is responsible for intrinsic drug resistance. Drugs targeting both EGFR and HER2, each capable of forming heterodimers with HER3, can reverse unresponsiveness to MEK inhibition

32 | Chapter 2

by decreasing inhibitory phosphorylation of the pro-apoptotic BCL2-associated proteins BAD and BIM (Sun et al., 2014a). The now almost classic example of a reactivated pathway by second mutation resulting from downstream pathway activation is the emergence of KRAS alterations in KRAS wild-type colorectal cancers treated with the EGFR monoclonal antibody cetuximab or panitumumab. Several studies have identified somatic mutations in KRAS, a downstream effector of EGFR, as a biomarker of intrinsic resistance to antiEGFR therapy in CRC (Amado et al., 2008; Karapetis et al., 2008). It was recently shown that a proportion of CRC patients with an initial response to anti-EGFR therapy, developed alterations (e.g. mutation or focal amplification) in the KRAS gene at the time of progression (Diaz et al., 2012; Misale et al., 2012; Valtorta et al., 2013). Importantly, these mutations can be detected non-invasively in the blood of patients, several months before radiographic detection of disease progression. Mathematical modeling indicated that the mutations were present in expanded subclones before the start of anti-EGFR therapy (Diaz et al., 2012) although new mutations can also arise (Misale et al., 2012). Downstream secondary mutations were also found as a resistance mechanism to MEK and BRAF inhibition. Mutations in MEK1 or MEK2 were found in tumor samples after disease progression (Emery et al., 2009; Trunzer et al., 2013; Van Allen et al., 2014; Wagle et al., 2011). The anti-HER2 monoclonal antibody trastuzumab, known in the clinic as herceptin®, is used in breast cancers overexpressing the RTK HER2. HER2 is overexpressed by either focal amplification or, in a small percentage, by mutation (Bose et al., 2013). Using RNA interference as a screening tool for drug resistance, it was found that loss of PTEN conferred resistance to trastuzumab and that activation of the PI3K pathway is predictive for response to trastuzumab in clinical samples (Berns et al., 2007; Nagata et al., 2004). There is also clinical evidence that women with HER2positive breast cancer who also carry activating mutations in PIK3CA are less likely to benefit from neoadjuvant HER2-targeted therapies than those without a PIK3CA mutation (Majewski et al., 2014). It has been shown that tumor regressions with vemurafenib treatment in BRAFmutant metastatic melanoma patients are dependent on the MAPK pathway blockade (Bollag et al., 2010). An inhibition of the pathway below a certain threshold was insufficient for tumor regression. Modulating the signaling output of the inhibited pathway around the threshold required for tumor inhibition can have major consequences for tumor response. In spite of the high response rates of

Drug resistance to targeted therapies: déjà vu all over again | 33

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

BRAF-mutant melanomas to vemurafenib and the improvement in overall survival, all patients eventually develop disease progression. Based on studies investigating resistance mechanisms to vemurafenib in BRAFV600E-mutated melanoma, the mitogen-activated protein kinase (MAPK) pathway is frequently reactivated in relapsed tumors (Shi et al., 2014). Melanomas can acquire resistance to the BRAFV600Einhibitor by expressing high levels of activated RAS resulting from NRAS or KRAS mutations. High levels of activated RAS upstream of RAF lead to subsequent MAPK pathway reactivation (Nazarian et al., 2010; Shi et al., 2014; Trunzer et al., 2013). Another example of this phenomenon is the identification of COT, also known as MAP3K8, as a MAPK pathway agonist. This was found through expression of all kinase and kinase-related open reading frames (ORFs) in cells that are sensitive to the BRAF-inhibitor (Johannessen et al., 2010). COT activates ERK through a MEKdependent mechanism that does not require RAF signaling. Indeed, high COT expression is associated with acquired resistance in patients that relapsed following treatment with MEK or RAF inhibitors (Johannessen et al., 2010). Bypass mechanisms Pathway reactivation is a common and recurrent mechanism of resistance to RTKinhibitors, MAPK pathway inhibitors and PI3K-AKT pathway inhibitors. However, activation of parallel pathways to bypass pathway inhibition also contributes to intrinsic and acquired resistance. Whole-exome sequencing identified upregulation of PI3K-AKT signaling as a second core resistance mechanism to BRAF inhibition (Shi et al., 2014). Mutations in the PI3K-PTEN-AKT pathway were present in 22% of the cases with acquired resistance to BRAF-inhibition, in most cases overlapping with alterations in the MAPK pathway. Engelman et al. have shown that amplification of MET occurred in 22% of lung cancer samples from patients that developed resistance to gefitinib or erlotinib (Engelman et al., 2007). Furthermore, it was shown that subpopulations of cells with MET amplification were present prior to drug exposure (Turke et al., 2010). Although the frequency might be lower than the initially reported 22% (Sequist et al., 2011), amplification of MET is driving HER3-dependent activation of the PI3KAKT pathway. This acquired activation of the PI3K-AKT pathway during EGFR-TKI is in other cases caused by PIK3CA mutation (Sequist et al., 2011). Takezawa et al. showed in vitro and in vivo the importance of HER2 in mediating the sensitivity of EGFR-mutant tumor cells to EGFR-TKIs (Takezawa et al., 2012). HER2 was amplified in 12% of tumors with acquired resistance and mutually exclusive with T790M mutation (Takezawa et al., 2012).

34 | Chapter 2

Amplification of the MET receptor is also associated with acquired resistance in CRC patients treated with anti-EGFR monoclonal antibodies that do not develop KRAS mutations (Bardelli et al., 2013). The prevalence of MET amplification was only 1% in unselected metastatic CRC samples, but increased to 12.5% in cetuximabresistant, genetically pre-selected xenopatients, indicating that MET amplification characterizes a significant fraction of cetuximab-resistant cases that are wild-type for KRAS, BRAF, NRAS, PIK3CA and HER2. In an unselected cohort, the clinical validation of MET amplification as a biomarker of resistance to EGFR therapy in metastatic CRC requires large (retrospective) studies. By expressing 15,906 human ORFs to identify genes whose upregulation conferred resistance to MAPK-pathway inhibition, it was found that, beside genes activating ERK signaling, cAMP and cAMP-dependent protein kinase (PKA) activation could confer resistance to MAPK-pathway inhibition in melanoma cells (Johannessen et al., 2013). PKA, once activated, can phosphorylate several substrates and modulate the activity of several transcription factors. Phosphorylation of two transcription factors downstream of cAMP and PKA, CREB and ATF1, was measured in tumor biopsies before or during treatment and after relapse with the BRAF-inhibitor or the BRAF-MEK-inhibitor combination. Levels of CREB and ATF1 phosphorylation were suppressed after treatment with the BRAF-inhibitor or the BRAF-MEK-inhibitor combination. In contrast, the levels of CREB and ATF1 phosphorylation in patients with a relapse were equal to the levels of phosphorylation in the pre-treatment cohort. There are many studies on the role of tumor microenvironment in tumor growth and metastasis, but only a few on a possible role in drug resistance. Nearly all studies on drug resistance have focused on cell-autonomous mechanisms of drug resistance, disregarding the effects of the tumor microenvironment. Technical challenges to properly model the microenvironment in vitro and in vivo is a major contributor of this publication bias. A co-culture model of tumor cells with stromal cells showed that the effect of targeted cancer drugs on tumor cells is greatly diminished when the tumor cells are cultured in the presence of stromal cells (Straussman et al., 2012). This stroma-mediated resistance mechanism was more striking for targeted therapies compared with cytotoxic chemotherapy. They further zoomed in on the role of stroma-mediated BRAF-inhibitor resistance and found that stromal secretion of hepatocyte growth factor (HGF) resulted in activation of MET receptor and thereby reactivation of the MAKP and PI3K-AKT pathway, causing BRAF-inhibitor resistance (Straussman et al., 2012).

Drug resistance to targeted therapies: dĂŠjĂ vu all over again | 35

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

The evidence that TGFβ-signaling plays a role in resistance against targeted and conventional cancer therapies further supports the importance of indirect mechanisms of drug resistance. TGFβ-signaling is an important player in the epithelial-to-mesenchymal transition (EMT), but also regulates the immune response of the extracellular matrix and activates direct or indirect multiple signaling pathways (Brunen et al., 2013). Activation of TGFβ-signaling was found sufficient to induce TKI resistance in multiple cancer types (Huang et al., 2012). Although the mechanism is still poorly understood, the process of EMT and the EMT-phenotype both have been linked several times to drug sensitivity and drug resistance (Byers et al., 2013; Fuchs et al., 2008; Salt et al., 2014; Thomson et al., 2005; Yauch et al., 2005).

WHAT WE HAVE LEARNED FROM DRUG RESISTANCE AND HOW TO DEAL WITH IT An understanding of the recurrent patterns of resistance, as discussed in this review, is necessary to design strategies to overcome intrinsic resistance and delay acquired resistance to targeted cancer drugs. To address this, significant efforts are now underway to (1) develop better drugs or more effective drug combinations to fully suppress the oncogenic signaling pathways and (2) development of non-invasive assays to monitor drug response and the emergence of drug resistance (Fig. 2). Developing better drugs and effective drug combinations The continuous effort to design more selective and potent inhibitors will definitely result in increased response rates to these inhibitors (Fig. 2A). An example is the development of inhibitors that can inhibit the ‘gatekeeper’ mutations that are responsible for a large part of the resistance to some TKIs. Patients with EGFRmutant NSCLC who have acquired resistance through the T790M mutation (‘gatekeeper’ mutation) have few treatment options. Second-generation EGFR inhibitors have been developed that also inhibit T790M-mutant EGFR. One of these second-generation EGFR inhibitors, afatinib, was FDA-approved last year as firstline treatment for EGFR-mutated NSCLC. However, in contrast to previous in vitro and in vivo studies with second-generation EGFR TKIs, Kim et al. reported the T790M mutation as a resistance mechanism for afatinib as well (Kim et al., 2012). This might be related to the nondiscriminatory action of this compound toward wild-type and mutant EGFR. A mutant-selective irreversible inhibitor might be more appropriate to overcome this problem. However, it should be noted that the emergence of the T790M mutation might take longer with second-generation inhibitors compared to

36 | Chapter 2

first-generation EGFR-TKIs, like erlotinib and gefitinib. Fortunately, new drugs have been designed that inhibit T790M-mutant EGFR. Several of these inhibitors have been described (Zhou et al., 2009), and one of them is currently being evaluated in phase 1 and 2 clinical trials in EGFR-mutant NSCLC after showing promising results in in vitro and in vivo validations (Walter et al., 2013). An important note to add here is that the greater potency of targeted therapies may result in increased heterogeneity of resistance mechanisms to these therapies. (A) New and better compounds Afatinib chemical structure:

(B) Mechanism-based drug combinations

polyclonal drug enhancer screen Virus-encoded shRNA

Infect

Select

Design more potent and selective inhibitors

-drug

+drug

Liquid biopsy for the detection of emerging resistance variants in tumor-derived cell-free DNA

‘Drugging the undruggable’ Identify depleted shRNA bar code sequences by deep sequencing

K. Berns, Bernards lab

Misale et al., 2012

Figure 2. Strategies to overcome intrinsic resistance and delay acquired resistance. (A) The design of more selective and potent inhibitors, also for what is currently ‘undruggable’. (B) The development of more effective drug combinations, e.g. by investigating mechanisms of drug resistance in the clinic or by in vitro and in vivo synthetic lethality screens. (C) Repeated non-invasive detection of emerging resistance variants, for example in tumor-derived cellfree DNA (cfDNA), to dynamically adapt the combination strategy.

The L1196M ‘gatekeeper’ mutation in ALK, conferring resistance to the ALKinhibitor crizotinib, can be inhibited by a newly designed potent, selective, and orally available ALK-inhibitor (Sakamoto et al., 2011). Similarly, a compound was designed that targets the T315I mutation in ABL (O’Hare et al., 2009). Based on results of the phase 2 trial, this drug named ponatinib was FDA-approved in 2012 for patients with resistant, BCR-ABL positive CML. However, on 9 October 2013, the FDA issued a partial clinical hold on new trial enrollment for ponatinib due to an increased number of blood clots observed in patients taking the drug in the phase 3 trial (EPIC trial). This trial was discontinued shortly thereafter. What we have learned from the identified resistance mechanisms? An outstanding lesson is that drug resistance to targeted single-agent therapies is frequently driven

Drug resistance to targeted therapies: déjà vu all over again | 37

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

by pathway reactivation, suggesting that a major limitation of targeted agents lies in their inability to fully block the pathway of interest. We therefore need to identify and target points of coalescence: nodes onto which multiple independent resistance mechanisms and tumor dependencies may arise. Furthermore, this points to the need for more potent and mechanism-based combination therapies (Fig. 2B). A recent example is the design of clinical trials with the dual combination of BRAF-inhibitors and EGFR inhibition, or a triple combination with adding PI3Kor MEK-inhibition in BRAF-mutant colorectal cancer (ClinicalTrials.gov identifiers NCT01719380, NCT01750918 and NCT01791309). Another example is the combination of the BRAF-inhibitor with a MEK-inhibitor in patients with metastatic BRAF-mutant melanoma, after the identification of MAPK pathway reactivation as a frequent resistance mechanism to monotherapy with the BRAF-inhibitor. Complete inhibition of the MAPK pathway could potentially be achieved by combining the BRAF-inhibitor with a MEK (MAP2K)-inhibitor. The clinical trial with the combination of the BRAFV600E-inhibitor dabrafenib and the MEKinhibitor trametinib reported an improved median progression-free survival of the combination by almost 4 months compared to dabrafenib monotherapy (Flaherty et al., 2012). Toxicity of combinations with targeted therapies has always been regarded as the limiting factor for this approach. However, the study by Flaherty et al. showed that dabrafenib and trametinib could be safely combined at full monotherapy doses (Flaherty et al., 2012). In fact, cutaneous squamous-cell carcinoma was seen in only 7% of patients receiving the combination, in comparison to 19% of patients receiving BRAF monotherapy. The development of new generation targeted therapies with alternative dosing strategies to overcome toxicities will allow previously toxic combinations to be developed. Although the progression-free survival of patients with BRAF-mutant melanoma increased by almost 4 months using the combination the BRAF-inhibitor with the MEK-inhibitor, resistance still developed in most patients. In a follow-up report of this trial, the authors investigated the resistance mechanisms for 5 patients that experienced clinical benefit before developing progressive disease (Wagle et al., 2014). Despite the dual inhibition of the MAPK pathway, they surprisingly identified in 3 out of 5 cases a resistance mechanism that engages MAPK-effectors: in one patient an acquired MEK2-mutation was found, in another patient the previously reported BRAF-splice variant and the third patient a focal BRAF amplification. The prevalence of MAPK pathway alterations in these resistant tumors underlines the reliance of BRAF-mutant melanomas on MEK/ERK signaling and the incomplete inhibition of the pathway despite using two drugs hitting this pathway. More potent MEK inhibition by increasing the dose of the MEK-inhibitor or, in the future, ERKinhibitors, may provide a solution for overcoming resistance to the combination. 38 | Chapter 2

Several combinations with endocrine therapy in breast cancer are being tested. As discussed, the combination of the mTOR inhibitor everolimus with tamoxifen or an aromatase inhibitor showed significant clinical benefit for advanced or metastatic ERα-positive breast cancer. Furthermore, dual targeting of HER2-positive tumors with trastuzumab (HER2 monoclonal antibody) and lapatinib (EGFR/HER2-TKI) was undertaken because of the synergistic interaction between the two compounds in preclinical models (Scaltriti et al., 2009; Xia et al., 2005). In the NeoALTTO study, the use of lapatinib, trastuzumab, and their combination was assessed as neo-adjuvant therapy for women with HER2-positive early breast cancer (Baselga et al., 2012a). The pathological complete response (pCR) rate was significantly higher in the group treated with the combination than in the group treated with trastuzumab alone (difference 21.1%, 9.1-34.2, P = 0.0001). Furthermore, the combination of pertuzumab (an anti-HER2 humanized monoclonal antibody that inhibits receptor dimerization) plus trastuzumab plus docetaxel in patients with HER2-positive metastatic breast cancer significantly prolonged progression-free survival and overall survival as compared to placebo plus trastuzumab plus docetaxel (Baselga et al., 2012c; Swain et al., 2013). Finally, combination of the CDK4/6 inhibitor palbociclib with the aromatase inhibitor letrozole demonstrated clinical benefit in advanced ERα-positive breast cancer. The combination improved progression-free survival in a phase 2 trial from 7.5 months for patients treated with letrozole monotherapy to 26.1 months for those in the combination arm and was well tolerated (Finn et al., 2012). These positive results are currently validated in a large phase 3 clinical trial (ClinicalTrials.gov identifier NCT01740427). Future developments Although effective drug combinations are indispensable to deal upfront with resistance, the extensive tumor heterogeneity in some tumors is likely to limit the durability of responses to these combinations. In fact, as mentioned above, even resistance to combination therapies was already observed in the clinic. Mathematical modeling showed that the presence of a single mutation conferring cross-resistance to each of the two drugs will not lead to sustained improvement for the majority of patients (Bozic et al., 2013). Sequential non-invasive detection of emerging resistance variants will be essential to dynamically adapt the combination strategy (Fig. 2C). A potential source to identify these variants is circulating, cell-free DNA (cfDNA) in the blood (reviewed in (Crowley et al., 2013)). This was used to identify the emergence of KRAS mutations or MET amplifications during treatment of CRC patients with anti-EGFR therapy (Bardelli et al., 2013; Diaz et al., 2012; Misale et al.,

Drug resistance to targeted therapies: déjà vu all over again | 39

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

2012). Currently, the sensitivity of these assays is around 0.1-1% variant frequencies, which currently limits the utility to finding hotspot mutations in predefined genes. Future improvements in sequencing accuracy will undoubtedly increase the sensitivity of cfDNA analysis. With the continuous drop in sequencing cost, it may soon become feasible to monitor apparently healthy individuals for the presence of occult cancer. When this becomes reality, stage IV metastatic disease may become rare, with associated increase in cancer survival. Until now, clinicians could often only wait until drug resistance develops and then try, often without solid scientific rationale, a second line therapy. These sequential therapies generally form a perfect recipe for certain treatment failure. However, we are now entering an era in which we should be able to anticipate the next move of the cancer cell, due to increasing understanding of the resistance mechanisms, and develop rational combination therapies. Moreover, even if we fail to predict the next move of the cancer, we will be able to monitor its progression from early stages on using non-invasive diagnostic approaches. By analogy, to win a game of chess, it is imperative to predict the next likely move of the opponent. We are now entering a new era of cancer therapy in which we will keep getting better in predicting and understanding the next move of the cancer to evade therapy. As a result, our chances of winning should increase proportionally. However, as Yogi Berra said: â&#x20AC;&#x153;It is tough to make predictions, especially about the futureâ&#x20AC;?.

40 | Chapter 2

REFERENCES Albertson, D.G., 2012. ESR1 amplification in breast cancer: controversy resolved? J Pathol 227, 1-3. Amado, R.G., Wolf, M., Peeters, M., et al., 2008. Wild-type KRAS is required for panitumumab efficacy in patients with metastatic colorectal cancer. J Clin Oncol 26, 1626-1634. Arpino, G., Green, S.J., Allred, D.C., et al., 2004. HER-2 amplification, HER-1 expression, and tamoxifen response in estrogen receptorpositive metastatic breast cancer: a southwest oncology group study. Clin Cancer Res 10, 56705676. Awad, M.M., Katayama, R., McTigue, M., et al., 2013. Acquired resistance to crizotinib from a mutation in CD74-ROS1. N Engl J Med 368, 2395-2401. Bachelot, T., Bourgier, C., Cropet, C., et al., 2012. Randomized phase II trial of everolimus in combination with tamoxifen in patients with hormone receptor-positive, human epidermal growth factor receptor 2-negative metastatic breast cancer with prior exposure to aromatase inhibitors: a GINECO study. J Clin Oncol 30, 2718-2724. Bardelli, A., Corso, S., Bertotti, A., et al., 2013. Amplification of the MET receptor drives resistance to anti-EGFR therapies in colorectal cancer. Cancer Discov 3, 658-673. Baselga, J., Bradbury, I., Eidtmann, H., et al., 2012a. Lapatinib with trastuzumab for HER2positive early breast cancer (NeoALTTO): a randomised, open-label, multicentre, phase 3 trial. Lancet 379, 633-640. Baselga, J., Campone, M., Piccart, M., et al., 2012b. Everolimus in postmenopausal hormone-receptor-positive advanced breast cancer. N Engl J Med 366, 520-529. Baselga, J., Cortes, J., Kim, S.B., et al., 2012c. Pertuzumab plus trastuzumab plus docetaxel for metastatic breast cancer. N Engl J Med 366, 109-119.

Beelen, K., Zwart, W., Linn, S.C., 2012. Can predictive biomarkers in breast cancer guide adjuvant endocrine therapy? Nat Rev Clin Oncol 9, 529-541. Berns, K., Horlings, H.M., Hennessy, B.T., et al., 2007. A functional genetic approach identifies the PI3K pathway as a major determinant of trastuzumab resistance in breast cancer. Cancer Cell 12, 395-402. Berra, Y., 1998. The Yogi Book: “I Really Didn’t Say Everything I Said!”. Workman Publishing Company, Inc. Bollag, G., Hirth, P., Tsai, J., et al., 2010. Clinical efficacy of a RAF inhibitor needs broad target blockade in BRAF-mutant melanoma. Nature 467, 596-599. Bose, R., Kavuri, S.M., Searleman, A.C., et al., 2013. Activating HER2 mutations in HER2 gene amplification negative breast cancer. Cancer Discov 3, 224-237. Bozic, I., Reiter, J.G., Allen, B., et al., 2013. Evolutionary dynamics of cancer in response to targeted combination therapy. Elife 2, e00747. Brockman, R.W., 1963. Mechanisms of Resistance to Anticancer Agents. Adv Cancer Res 7, 129-234. Brummelkamp, T.R., Bernards, R., Agami, R., 2002. A system for stable expression of short interfering RNAs in mammalian cells. Science 296, 550-553. Brunen, D., Willems, S.M., Kellner, U., et al., 2013. TGF-beta: an emerging player in drug resistance. Cell Cycle 12, 2960-2968. Byers, L.A., Diao, L., Wang, J., et al., 2013. An epithelial-mesenchymal transition gene signature predicts resistance to EGFR and PI3K inhibitors and identifies Axl as a therapeutic target for overcoming EGFR inhibitor resistance. Clin Cancer Res 19, 279-290. Cancer Genome Atlas Network, 2012. Comprehensive molecular portraits of human breast tumours. Nature 490, 61-70.

Drug resistance to targeted therapies: déjà vu all over again | 41

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

Chandarlapaty, S., Sawai, A., Scaltriti, M., et al., 2011. AKT inhibition relieves feedback suppression of receptor tyrosine kinase expression and activity. Cancer Cell 19, 58-71.

Diaz, L.A., Jr., Williams, R.T., Wu, J., et al., 2012. The molecular evolution of acquired resistance to targeted EGFR blockade in colorectal cancers. Nature 486, 537-540.

Chapman, P.B., Hauschild, A., Robert, C., et al., 2011. Improved survival with vemurafenib in melanoma with BRAF V600E mutation. N Engl J Med 364, 2507-2516.

Duncan, J.S., Whittle, M.C., Nakamura, K., et al., 2012. Dynamic reprogramming of the kinome in response to targeted MEK inhibition in triplenegative breast cancer. Cell 149, 307-321.

Choi, Y.L., Soda, M., Yamashita, Y., et al., 2010. EML4-ALK mutations in lung cancer that confer resistance to ALK inhibitors. N Engl J Med 363, 1734-1739.

EBCTCG, 2005. Effects of chemotherapy and hormonal therapy for early breast cancer on recurrence and 15-year survival: an overview of the randomised trials. Lancet 365, 1687-1717.

Corcoran, R.B., Dias-Santagata, D., Bergethon, K., et al., 2010. BRAF gene amplification can promote acquired resistance to MEK inhibitors in cancer cells harboring the BRAF V600E mutation. Sci Signal 3, ra84.

Emery, C.M., Vijayendran, K.G., Zipser, M.C., et al., 2009. MEK1 mutations confer resistance to MEK and B-RAF inhibition. Proc Natl Acad Sci U S A 106, 20411-20416.

Corcoran, R.B., Ebi, H., Turke, A.B., et al., 2012. EGFR-mediated re-activation of MAPK signaling contributes to insensitivity of BRAF mutant colorectal cancers to RAF inhibition with vemurafenib. Cancer Discov 2, 227-235. Crowley, E., Di Nicolantonio, F., Loupakis, F., et al., 2013. Liquid biopsy: monitoring cancergenetics in the blood. Nat Rev Clin Oncol 10, 472-484. Das Thakur, M., Salangsang, F., Landman, A.S., et al., 2013. Modelling vemurafenib resistance in melanoma reveals a strategy to forestall drug resistance. Nature 494, 251-255. Davies, C., Godwin, J., Gray, R., et al., 2011. Relevance of breast cancer hormone receptors and other factors to the efficacy of adjuvant tamoxifen: patient-level meta-analysis of randomised trials. Lancet 378, 771-784. Davies, K.D., Le, A.T., Theodoro, M.F., et al., 2012. Identifying and targeting ROS1 gene fusions in non-small cell lung cancer. Clin Cancer Res 18, 4570-4579. De Laurentiis, M., Arpino, G., Massarelli, E., et al., 2005. A meta-analysis on the interaction between HER-2 expression and response to endocrine treatment in advanced breast cancer. Clin Cancer Res 11, 4741-4748.

42 | Chapter 2

Encarnacion, C.A., Ciocca, D.R., McGuire, W.L., et al., 1993. Measurement of steroid hormone receptors in breast cancer patients on tamoxifen. Breast Cancer Res Treat 26, 237-246. Engelman, J.A., Zejnullahu, K., Mitsudomi, T., et al., 2007. MET amplification leads to gefitinib resistance in lung cancer by activating ERBB3 signaling. Science 316, 1039-1043. Finn, R.S., Crown, J.P., Lang, I., et al., 2012. Results of a randomized phase 2 study of PD 0332991, a cyclin-dependent kinase (CDK) 4/6 inhibitor, in combination with letrozole vs letrozole alone for first-line treatment of ER+/ HER2â&#x20AC;&#x201C; advanced breast cancer (BC). Cancer Res 72 (24 Suppl), Abstract nr S1-6. Flaherty, K.T., Infante, J.R., Daud, A., et al., 2012. Combined BRAF and MEK inhibition in melanoma with BRAF V600 mutations. N Engl J Med 367, 1694-1703. Fuchs, B.C., Fujii, T., Dorfman, J.D., et al., 2008. Epithelial-to-mesenchymal transition and integrin-linked kinase mediate sensitivity to epidermal growth factor receptor inhibition in human hepatoma cells. Cancer Res 68, 23912399. Girotti, M.R., Pedersen, M., Sanchez-Laorden, B., et al., 2013. Inhibiting EGF receptor or SRC family kinase signaling overcomes BRAF inhibitor resistance in melanoma. Cancer Discov 3, 158-167.

Goetz, M.P., Rae, J.M., Suman, V.J., et al., 2005. Pharmacogenetics of tamoxifen biotransformation is associated with clinical outcomes of efficacy and hot flashes. J Clin Oncol 23, 9312-9318. Gorre, M.E., Mohammed, M., Ellwood, K., et al., 2001. Clinical resistance to STI-571 cancer therapy caused by BCR-ABL gene mutation or amplification. Science 293, 876-880. Groenendijk, F.H., Zwart, W., Floore, A., et al., 2013. Estrogen receptor splice variants as a potential source of false-positive estrogen receptor status in breast cancer diagnostics. Breast Cancer Res Treat 140, 475-484.

Iorns, E., Turner, N.C., Elliott, R., et al., 2008. Identification of CDK10 as an important determinant of resistance to endocrine therapy for breast cancer. Cancer Cell 13, 91-104. Jeselsohn, R., Yelensky, R., Buchwalter, G., et al., 2014. Emergence of Constitutively Active Estrogen Receptor-alpha Mutations in Pretreated Advanced Estrogen ReceptorPositive Breast Cancer. Clin Cancer Res 20, 1757-1767. Johannessen, C.M., Boehm, J.S., Kim, S.Y., et al., 2010. COT drives resistance to RAF inhibition through MAP kinase pathway reactivation. Nature 468, 968-972.

Gururaj, A.E., Rayala, S.K., Vadlamudi, R.K., et al., 2006. Novel mechanisms of resistance to endocrine therapy: genomic and nongenomic considerations. Clin Cancer Res 12, 1001s-1007s.

Johannessen, C.M., Johnson, L.A., Piccioni, F., et al., 2013. A melanocyte lineage program confers resistance to MAP kinase pathway inhibition. Nature 504, 138-142.

Gutierrez, M.C., Detre, S., Johnston, S., et al., 2005. Molecular changes in tamoxifen-resistant breast cancer: relationship between estrogen receptor, HER-2, and p38 mitogen-activated protein kinase. J Clin Oncol 23, 2469-2476.

Johnston, S.R., Saccani-Jotti, G., Smith, I.E., et al., 1995. Changes in estrogen receptor, progesterone receptor, and pS2 expression in tamoxifen-resistant human breast cancer. Cancer Res 55, 3331-3338.

Haughian, J.M., Pinto, M.P., Harrell, J.C., et al., 2012. Maintenance of hormone responsiveness in luminal breast cancers by suppression of Notch. Proc Natl Acad Sci U S A 109, 2742-2747.

Karapetis, C.S., Khambata-Ford, S., Jonker, D.J., et al., 2008. K-ras mutations and benefit from cetuximab in advanced colorectal cancer. N Engl J Med 359, 1757-1765.

Hemeryck, A., Belpaire, F.M., 2002. Selective serotonin reuptake inhibitors and cytochrome P-450 mediated drug-drug interactions: an update. Curr Drug Metab 3, 13-37.

Katayama, R., Shaw, A.T., Khan, T.M., et al., 2012. Mechanisms of acquired crizotinib resistance in ALK-rearranged lung Cancers. Sci Transl Med 4, 120ra117.

Herynk, M.H., Fuqua, S.A., 2004. Estrogen receptor mutations in human disease. Endocr Rev 25, 869-898.

Kim, Y., Ko, J., Cui, Z., et al., 2012. The EGFR T790M mutation in acquired resistance to an irreversible second-generation EGFR inhibitor. Mol Cancer Ther 11, 784-791.

Holm, C., Kok, M., Michalides, R., et al., 2009. Phosphorylation of the oestrogen receptor alpha at serine 305 and prediction of tamoxifen resistance in breast cancer. J Pathol 217, 372-379. Hoskins, J.M., Carey, L.A., McLeod, H.L., 2009. CYP2D6 and tamoxifen: DNA matters in breast cancer. Nat Rev Cancer 9, 576-586. Huang, S., Holzel, M., Knijnenburg, T., et al., 2012. MED12 controls the response to multiple cancer drugs through regulation of TGF-beta receptor signaling. Cell 151, 937-950.

Kobayashi, S., Boggon, T.J., Dayaram, T., et al., 2005. EGFR mutation and resistance of nonsmall-cell lung cancer to gefitinib. N Engl J Med 352, 786-792. Kuukasjarvi, T., Kononen, J., Helin, H., et al., 1996. Loss of estrogen receptor in recurrent breast cancer is associated with poor response to endocrine therapy. J Clin Oncol 14, 2584-2589. Lemmon, M.A., Schlessinger, J., 2010. Cell signaling by receptor tyrosine kinases. Cell 141, 1117-1134.

Drug resistance to targeted therapies: dĂŠjĂ vu all over again | 43

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

Li, S., Shen, D., Shao, J., et al., 2013. Endocrinetherapy-resistant ESR1 variants revealed by genomic characterization of breast-cancerderived xenografts. Cell Rep 4, 1116-1130. Little, A.S., Balmanno, K., Sale, M.J., et al., 2011. Amplification of the driving oncogene, KRAS or BRAF, underpins acquired resistance to MEK1/2 inhibitors in colorectal cancer cells. Sci Signal 4, ra17. Long, G.V., Menzies, A.M., Nagrial, A.M., et al., 2011. Prognostic and clinicopathologic associations of oncogenic BRAF in metastatic melanoma. J Clin Oncol 29, 1239-1246. Maemondo, M., Inoue, A., Kobayashi, K., et al., 2010. Gefitinib or chemotherapy for non-smallcell lung cancer with mutated EGFR. N Engl J Med 362, 2380-2388. Magnani, L., Stoeck, A., Zhang, X., et al., 2013. Genome-wide reprogramming of the chromatin landscape underlies endocrine therapy resistance in breast cancer. Proc Natl Acad Sci U S A 110, E1490-1499. Majewski, I.J., Nuciforo, P., Mittempergher, L., et al., 2014. PIK3CA mutations are associated with resistance to neoadjuvant HER2-targeted therapies in breast cancer (submitted). McGlynn, L.M., Kirkegaard, T., Edwards, J., et al., 2009. Ras/Raf-1/MAPK pathway mediates response to tamoxifen but not chemotherapy in breast cancer patients. Clin Cancer Res 15, 14871495. Michalides, R., Griekspoor, A., Balkenende, A., et al., 2004. Tamoxifen resistance by a conformational arrest of the estrogen receptor alpha after PKA activation in breast cancer. Cancer Cell 5, 597-605. Michor, F., Hughes, T.P., Iwasa, Y., et al., 2005. Dynamics of chronic myeloid leukaemia. Nature 435, 1267-1270. Miller, T.W., Balko, J.M., Arteaga, C.L., 2011a. Phosphatidylinositol 3-kinase and antiestrogen resistance in breast cancer. J Clin Oncol 29, 44524461.

44 | Chapter 2

Miller, T.W., Balko, J.M., Ghazoui, Z., et al., 2011b. A gene expression signature from human breast cancer cells with acquired hormone independence identifies MYC as a mediator of antiestrogen resistance. Clin Cancer Res 17, 2024-2034. Misale, S., Yaeger, R., Hobor, S., et al., 2012. Emergence of KRAS mutations and acquired resistance to anti-EGFR therapy in colorectal cancer. Nature 486, 532-536. Montagut, C., Dalmases, A., Bellosillo, B., et al., 2012. Identification of a mutation in the extracellular domain of the Epidermal Growth Factor Receptor conferring cetuximab resistance in colorectal cancer. Nat Med 18, 221-223. Montero-Conde, C., Ruiz-Llorente, S., Dominguez, J.M., et al., 2013. Relief of feedback inhibition of HER3 transcription by RAF and MEK inhibitors attenuates their antitumor effects in BRAF-mutant thyroid carcinomas. Cancer Discov 3, 520-533. Musgrove, E.A., Sutherland, R.L., 2009. Biological determinants of endocrine resistance in breast cancer. Nat Rev Cancer 9, 631-643. Nagata, Y., Lan, K.H., Zhou, X., et al., 2004. PTEN activation contributes to tumor inhibition by trastuzumab, and loss of PTEN predicts trastuzumab resistance in patients. Cancer Cell 6, 117-127. Nathanson, D.A., Gini, B., Mottahedeh, J., et al., 2014. Targeted therapy resistance mediated by dynamic regulation of extrachromosomal mutant EGFR DNA. Science 343, 72-76. Nazarian, R., Shi, H., Wang, Q., et al., 2010. Melanomas acquire resistance to B-RAF(V600E) inhibition by RTK or N-RAS upregulation. Nature 468, 973-977. Oâ&#x20AC;&#x2122;Hare, T., Shakespeare, W.C., Zhu, X., et al., 2009. AP24534, a pan-BCR-ABL inhibitor for chronic myeloid leukemia, potently inhibits the T315I mutant and overcomes mutation-based resistance. Cancer Cell 16, 401-412. Oâ&#x20AC;&#x2122;Reilly, K.E., Rojo, F., She, Q.B., et al., 2006. mTOR inhibition induces upstream receptor tyrosine kinase signaling and activates Akt. Cancer Res 66, 1500-1508.

Ooi, A., Inokuchi, M., Harada, S., et al., 2012. Gene amplification of ESR1 in breast cancers--fact or fiction? A fluorescence in situ hybridization and multiplex ligation-dependent probe amplification study. J Pathol 227, 8-16. Osborne, C.K., Bardou, V., Hopp, T.A., et al., 2003. Role of the estrogen receptor coactivator AIB1 (SRC-3) and HER-2/neu in tamoxifen resistance in breast cancer. J Natl Cancer Inst 95, 353-361. Osborne, C.K., Schiff, R., 2011. Mechanisms of endocrine resistance in breast cancer. Annu Rev Med 62, 233-247. Ou, S.H.I., Bang Y.J., Camidge, D.R., et al., 2013. Efficacy and safety of crizotinib in patients with advanced ROS1-rearranged non-small cell lung cancer (NSCLC). J Clin Oncol 31, (suppl; abstr 8032). Poulikakos, P.I., Persaud, Y., Janakiraman, M., et al., 2011. RAF inhibitor resistance is mediated by dimerization of aberrantly spliced BRAF(V600E). Nature 480, 387-390. Prahallad, A., Sun, C., Huang, S., et al., 2012. Unresponsiveness of colon cancer to BRAF(V600E) inhibition through feedback activation of EGFR. Nature 483, 100-103. Rizzo, P., Miao, H., D’Souza, G., et al., 2008. Cross-talk between notch and the estrogen receptor in breast cancer suggests novel therapeutic approaches. Cancer Res 68, 52265235. Robinson, D.R., Wu, Y.M., Vats, P., et al., 2013. Activating ESR1 mutations in hormoneresistant metastatic breast cancer. Nat Genet 45, 1446-1451. Sakamoto, H., Tsukaguchi, T., Hiroshima, S., et al., 2011. CH5424802, a selective ALK inhibitor capable of blocking the resistant gatekeeper mutant. Cancer Cell 19, 679-690. Salt, M.B., Bandyopadhyay, S., McCormick, F., 2014. Epithelial-to-Mesenchymal Transition Rewires the Molecular Path to PI3K-Dependent Proliferation. Cancer Discov 4, 186-199. Sasaki, T., Okuda, K., Zheng, W., et al., 2010. The neuroblastoma-associated F1174L ALK

mutation causes resistance to an ALK kinase inhibitor in ALK-translocated cancers. Cancer Res 70, 10038-10043. Scaltriti, M., Verma, C., Guzman, M., et al., 2009. Lapatinib, a HER2 tyrosine kinase inhibitor, induces stabilization and accumulation of HER2 and potentiates trastuzumab-dependent cell cytotoxicity. Oncogene 28, 803-814. Sequist, L.V., Waltman, B.A., Dias-Santagata, D., et al., 2011. Genotypic and histological evolution of lung cancers acquiring resistance to EGFR inhibitors. Sci Transl Med 3, 75ra26. Shaw, A.T., Kim, D.W., Nakagawa, K., et al., 2013. Crizotinib versus chemotherapy in advanced ALK-positive lung cancer. N Engl J Med 368, 2385-2394. Shi, H., Hugo, W., Kong, X., et al., 2014. Acquired resistance and clonal evolution in melanoma during BRAF inhibitor therapy. Cancer Discov 4, 80-93. Shi, H., Moriceau, G., Kong, X., et al., 2012. Melanoma whole-exome sequencing identifies (V600E)B-RAF amplification-mediated acquired B-RAF inhibitor resistance. Nat Commun 3, 724. Shi, L., Dong, B., Li, Z., et al., 2009. Expression of ER-{alpha}36, a novel variant of estrogen receptor {alpha}, and resistance to tamoxifen treatment in breast cancer. J Clin Oncol 27, 34233429. Shiau, A.K., Barstad, D., Loria, P.M., et al., 1998. The structural basis of estrogen receptor/ coactivator recognition and the antagonism of this interaction by tamoxifen. Cell 95, 927-937. Singh, M.S., Francis, P.A., Michael, M., 2011. Tamoxifen, cytochrome P450 genes and breast cancer clinical outcomes. Breast 20, 111-118. Smith, I.E., Dowsett, M., 2003. Aromatase inhibitors in breast cancer. N Engl J Med 348, 2431-2442. Sosman, J.A., Kim, K.B., Schuchter, L., et al., 2012. Survival in BRAF V600-mutant advanced melanoma treated with vemurafenib. N Engl J Med 366, 707-714.

Drug resistance to targeted therapies: déjà vu all over again | 45

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

Straussman, R., Morikawa, T., Shee, K., et al., 2012. Tumour micro-environment elicits innate resistance to RAF inhibitors through HGF secretion. Nature 487, 500-504.

Turke, A.B., Zejnullahu, K., Wu, Y.L., et al., 2010. Preexistence and clonal selection of MET amplification in EGFR mutant NSCLC. Cancer Cell 17, 77-88.

Sun, C., Hobor, S., Bertotti, A., et al., 2014a. Intrinsic Resistance to MEK Inhibition in KRAS Mutant Lung and Colon Cancer through Transcriptional Induction of ERBB3. Cell Rep 7, 86-93.

Valtorta, E., Misale, S., Sartore-Bianchi, A., et al., 2013. KRAS gene amplification in colorectal cancer and impact on response to EGFRtargeted therapy. Int J Cancer 133, 1259-1265.

Sun, C., Wang, L., Huang, S., et al., 2014b. Reversible and adaptive resistance to BRAF(V600E) inhibition in melanoma. Nature 508, 118-122. Swain, S.M., Kim, S.B., Cortes, J., et al., 2013. Pertuzumab, trastuzumab, and docetaxel for HER2-positive metastatic breast cancer (CLEOPATRA study): overall survival results from a randomised, double-blind, placebocontrolled, phase 3 study. Lancet Oncol 14, 461471. Takezawa, K., Pirazzoli, V., Arcila, M.E., et al., 2012. HER2 amplification: a potential mechanism of acquired resistance to EGFR inhibition in EGFR-mutant lung cancers that lack the second-site EGFRT790M mutation. Cancer Discov 2, 922-933. Thomson, S., Buck, E., Petti, F., et al., 2005. Epithelial to mesenchymal transition is a determinant of sensitivity of non-small-cell lung carcinoma cell lines and xenografts to epidermal growth factor receptor inhibition. Cancer Res 65, 9455-9462.

Van Allen, E.M., Wagle, N., Sucker, A., et al., 2014. The genetic landscape of clinical resistance to RAF inhibition in metastatic melanoma. Cancer Discov 4, 94-109. van der Flier, S., Brinkman, A., Look, M.P., et al., 2000. Bcar1/p130Cas protein and primary breast cancer: prognosis and response to tamoxifen treatment. J Natl Cancer Inst 92, 120127. Villanueva, J., Vultur, A., Lee, J.T., et al., 2010. Acquired resistance to BRAF inhibitors mediated by a RAF kinase switch in melanoma can be overcome by cotargeting MEK and IGF1R/PI3K. Cancer Cell 18, 683-695. Wagle, N., Emery, C., Berger, M.F., et al., 2011. Dissecting therapeutic resistance to RAF inhibition in melanoma by tumor genomic profiling. J Clin Oncol 29, 3085-3096. Wagle, N., Van Allen, E.M., Treacy, D.J., et al., 2014. MAP kinase pathway alterations in BRAFmutant melanoma patients with acquired resistance to combined RAF/MEK inhibition. Cancer Discov 4, 61-68.

Tolhurst, R.S., Thomas, R.S., Kyle, F.J., et al., 2011. Transient over-expression of estrogen receptor-alpha in breast cancer cells promotes cell survival and estrogen-independent growth. Breast Cancer Res Treat 128, 357-368.

Walter, A.O., Sjin, R.T., Haringsma, H.J., et al., 2013. Discovery of a Mutant-Selective Covalent Inhibitor of EGFR that Overcomes T790MMediated Resistance in NSCLC. Cancer Discov 3, 1404-1415.

Toy, W., Shen, Y., Won, H., et al., 2013. ESR1 ligand-binding domain mutations in hormoneresistant breast cancer. Nat Genet 45, 1439-1445.

Wang, J., Jain, S., Coombes, C.R., et al., 2009. Fulvestrant in advanced breast cancer following tamoxifen and aromatase inhibition: a single center experience. Breast J 15, 247-253.

Trunzer, K., Pavlick, A.C., Schuchter, L., et al., 2013. Pharmacodynamic effects and mechanisms of resistance to vemurafenib in patients with metastatic melanoma. J Clin Oncol 31, 1767-1774.

46 | Chapter 2

Weinstein, I.B., 2002. Cancer. Addiction to oncogenes--the Achilles heal of cancer. Science 297, 63-64.

Xia, W., Gerard, C.M., Liu, L., et al., 2005. Combining lapatinib (GW572016), a small molecule inhibitor of ErbB1 and ErbB2 tyrosine kinases, with therapeutic anti-ErbB2 antibodies enhances apoptosis of ErbB2-overexpressing breast cancer cells. Oncogene 24, 6213-6221. Yamnik, R.L., Holz, M.K., 2010. mTOR/ S6K1 and MAPK/RSK signaling pathways coordinately regulate estrogen receptor alpha serine 167 phosphorylation. FEBS Lett 584, 124128. Yano, S., Nakataki, E., Ohtsuka, S., et al., 2005. Retreatment of lung adenocarcinoma patients with gefitinib who had experienced favorable results from their initial treatment with this selective epidermal growth factor receptor inhibitor: a report of three cases. Oncol Res 15, 107-111.

Yauch, R.L., Januario, T., Eberhard, D.A., et al., 2005. Epithelial versus mesenchymal phenotype determines in vitro sensitivity and predicts clinical activity of erlotinib in lung cancer patients. Clin Cancer Res 11, 8686-8698. Yoo, C., Kim, S.B., Ahn, J.H., et al., 2011. Efficacy of fulvestrant in heavily pretreated postmenopausal women with advanced breast cancer: a preliminary report. J Breast Cancer 14, 135-139. Zhou, W., Ercan, D., Chen, L., et al., 2009. Novel mutant-selective EGFR kinase inhibitors against EGFR T790M. Nature 462, 1070-1074.

Drug resistance to targeted therapies: dĂŠjĂ vu all over again | 47

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

Chapter 3 Sorafenib synergizes with metformin in NSCLC through AMPK pathway activation

Floris H. Groenendijk1, Wouter W. Mellema6, Eline van der Burg2, Eva Schut2, Michael Hauptmann3, Hugo M. Horlings1, Stefan M. Willems1,5, Michel M. van den Heuvel4, Jos Jonkers2, Egbert F. Smit6 and RenĂŠ Bernards1

Division of Molecular Carcinogenesis, Cancer Genomics Centre; 2 Division of Molecular Pathology;

Division of Epidemiology and Biostatistics,

Division of Thoracic Oncology; The Netherlands Cancer

Institute, Plesmanlaan 121, 1066 CX Amsterdam, The Netherlands. 5

Department of Pathology, University Medical Center Utrecht, P.O. Box 85500, 3508 GA Utrecht, The

Netherlands 6

Department of Pulmonary Diseases, VU University Medical Center, P.O. Box 7057, 1007 MB Amsterdam,

The Netherlands

Published in the International Journal of Cancer 2014 Aug 1. doi:10.1002/ijc.29113 [Epub ahead of print]

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

ABSTRACT The multikinase inhibitor sorafenib is under clinical investigation for the treatment of many solid tumors, but in most cases, the molecular target responsible for the clinical effect is unknown. Furthermore, enhancing the effectiveness of sorafenib using combination strategies is a major clinical challenge. Here we identify sorafenib as an activator of AMP-activated protein kinase (AMPK), in a manner that involves either upstream LKB1 or CAMKK2. We further show in a phase II clinical trial with single agent sorafenib in KRAS-mutant advanced non-small cell lung cancer (NSCLC), an improved disease control rate in patients using the antidiabetic drug metformin. Consistent with this, sorafenib and metformin act synergistically in inhibiting cellular proliferation in NSCLC in vitro and in vivo. A synergistic effect of both drugs is also seen on phosphorylation of the AMPKα activation site. Our results provide a rationale for the synergistic antiproliferative effects, given that AMPK inhibits downstream mTOR signaling. These data suggest that the combination of sorafenib with AMPK activators could have beneficial effects on tumor regression by AMPK pathway activation. The combination of metformin or other AMPK activators and sorafenib could be tested in prospective clinical trials.

KEYWORDS AMP-activated protein kinase (AMPK)  Metformin  Non-small cell lung cancer (NSCLC)  Salicylate  Sorafenib

50 | Chapter 3

INTRODUCTION AMP-activated protein kinase (AMPK) signaling has increasingly attracted interest in carcinogenesis, because AMPK acts as a metabolic checkpoint coordinating cellular growth. AMPK acts through inhibition of cell cycle progression and repression of the mTOR pathway, a pathway that is frequently activated in cancer. AMPK is activated in response to AMP binding or by phosphorylation at threonine-172 (Thr172) on the alpha-subunit by the tumor suppressor protein kinase LKB1 (also called STK11) or the calcium activated kinase CAMKK2 (Mihaylova and Shaw, 2011). LKB1 is of particular interest because mutations in the LKB1 gene are responsible for PeutzJeghers syndrome, a rare autosomal dominant disorder with a strong tendency of developing cancer (Hemminki et al., 1998). In addition, cumulative evidence suggests that somatic mutations or inactivation of the LKB1 gene is involved in lung and cervical cancers (Ding et al., 2008). Gill et al. found in NSCLCs a generally low rate of somatic mutations in the LKB1 gene (11%), while they observed that either loss of heterozygosity or homozygous deletion of the LKB1 gene occurred in nearly 90% of the NSCLCs tested (Gill et al., 2011). The oral antidiabetic drug metformin belongs to the biguanide class and is the first-line drug of choice in the treatment of type II diabetes. Several epidemiological and case-controlled studies found diabetics using metformin have an up to 30% lower lifetime cancer risk in comparison to those using other antidiabetic medications (Dowling et al., 2012; Evans et al., 2005; Pierotti et al., 2013). In line with this reduced cancer risk, metformin is activating AMPK and exhibits an anti-proliferative effect on several cancer cell lines (Ben Sahra et al., 2010). Furthermore, metformin and the allosteric AMPK activator A-769662 both delay spontaneous tumor development in Pten+/- mice (Huang et al., 2008), but it has been unclear whether the reduced cancer incidence in diabetics is also explained by AMPK activity. Other mechanisms can be proposed to explain the effects of metformin on cancer, like the insulin-sensitizing and anti-hyperglycemic effects (Hardie, 2013; Pierotti et al., 2013). The activation of AMPK by metformin can be direct through an increase in AMP/ATP ratio or can be indirect through its activity on upstream kinases (Pierotti et al., 2013). Mechanistic studies have shown that AMPK plays an important role as well in the mechanism of action of thiazolidinedioneâ&#x20AC;&#x2122;s (Fryer et al., 2002) and statins (Sun et al., 2006). More recently, salicylate - the active metabolite of aspirin - was also found to act as an AMPK activator (Hawley et al., 2012). Aspirinâ&#x20AC;&#x2122;s effect on cancer has been widely studied, particularly its effect on colorectal cancer incidence and mortality. Strikingly, it was reported that only in patients with PIK3CA-mutant colorectal cancer the regular use of aspirin after diagnosis was associated with longer survival, and not in patients with PIK3CA wild-type colorectal cancer (Liao et al., 2012). Sorafenib synergizes with metformin in NSCLC through AMPK pathway activation | 51

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

Metformin might be an attractive and safe anti-cancer drug in monotherapy or in combination with chemotherapeutic or targeted agents. To date, no randomized controlled trials of metformin as a cancer therapy have been reported, but (pre-) clinical evidence suggests that metformin may enhance chemotherapy response (Chen et al., 2012; Hirsch et al., 2009; Iliopoulos et al., 2011; Jiralerspong et al., 2009; Rocha et al., 2011). However, it is also reported that metformin can drive angiogenesis and accelerate the in vivo growth of BRAFV600E-driven melanoma by upregulating VEGFA mRNA and protein levels (Martin et al., 2012). Multiple combinations of metformin and targeted agents are currently under clinical investigation (http:// www.clinicaltrials.gov/). Here, we report on post hoc analysis of a phase II clinical trial with the multikinase inhibitor sorafenib in KRAS mutant NSCLC (Dingemans et al., 2013) that patients receiving metformin during sorafenib treatment showed improved disease control rate. We observed synergistic growth inhibition of NSCLC cells in vitro and in vivo with the combination of sorafenib and metformin and describe synergistic AMPK activation and downstream mTOR pathway inhibition as the mechanism explaining the effects of the combination. In this study, we identify sorafenib as an activator of AMP-activated protein kinase (AMPK), in a manner that involved either upstream LKB1 or CAMKK2.

RESULTS Improved disease control rate in KRAS mutant advanced NSCLC patients receiving metformin during sorafenib treatment In the context of a phase II clinical trial (see Methods section), a total of 57 advanced, heavily pretreated KRAS-mutant NSCLC patients were treated with monotherapy sorafenib (Dingemans et al., 2013). Five patients were using metformin during the treatment because of type II diabetes. No other patients with diabetes participated in the study. Basic characteristics of the patients in the metformin group and the nonmetformin group are given in Supplementary Table S1. The overall DCR was 52.6% and thereby sorafenib was found active. In the five patients using the combination of sorafenib and metformin the DCR was 100%; two patients with partial response (PR) and three patients with stable disease (SD). In the 52 remaining patients, three patients had a PR, 22 patients had SD and 27 patients had progressive disease (P = 0.01; Table 1). The two patients with PR in the metformin group had duration of response of 10 and 13 months. The three patients with PR in the non-metformin group had duration of response of 1, 5 and 5 months. Median

52 | Chapter 3

PFS was 2.2 months (95% CI 1.1-3.3) in the non-metformin group and 2.8 months (95% CI 2.4-3.1) in the metformin group (P = 0.06). It is interesting to note that the two patients with PR in the metformin group had the longest PFS of all patients in this study. Median OS was 4.8 months (95% CI 1.5-8.2) in the non-metformin group and 9.0 months (95% CI 0.1-17.9) in the metformin group (P = 0.13). The two patients with PR in the metformin group also had one of the longest survivals; one patient was censored at 12 months and the other patient had a survival of 14 months. Table 1. Responses according to the RECIST criteria after six weeks of sorafenib treatment in a phase II trial in patients with locally advanced or metastatic non-squamous NSCLC harboring a KRAS mutation (n = 57)

Partial Response Stable Disease

Progressive Disease Total

No 3

27 52

Metformin

Yes 2

0 5

Total

P-value

0.01

27 57

Synergistic effect of sorafenib in combination with metformin or other AMPK activators in NSCLC cell lines To study a potential interaction between sorafenib and metformin, we tested this drug combination in a 10-day colony formation assay using the KRAS mutant NSCLC cell lines A549 and H460. As shown in Figure 1A and 1B, combination treatment of sorafenib and metformin caused a synergistic inhibition in proliferation in both A549 and H460 cells. To test whether AMPK activation by metformin underlies the synergistic inhibition of proliferation we treated cells with sorafenib in combination with the allosteric AMPK activator A-769662 or the AMPK activator salicylate. Figure 1C-F show that these compounds also act synergistically with sorafenib in inhibiting proliferation of A549 and H460 cells. Quantifications of the colony formation assays in Figure 1 are shown in Supplementary Table S4. Besides a KRAS mutation, A549 and H460 cells also harbor an inactivating mutation in the LKB1 gene. We therefore tested whether the combination of sorafenib and an AMPK activator would also synergistically inhibit the proliferation of LKB1 wild-type NSCLC cells. Colony formation assays of sorafenib in combination with salicylate in LKB1 wild-type KRAS mutant H358 cells, LKB1 mutant KRAS wild-type H838 cells and double wild-type H522 NSCLC cells showed that the effect of the combination is independent of LKB1 or KRAS mutation status (Supplementary Fig. S1).

Sorafenib synergizes with metformin in NSCLC through AMPK pathway activation | 53

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

A549 .25

METFORMIN (mM)

0 METFORMIN (mM)

H460

SORAFENIB (µM) 0

.5 1.0 2.0

.25

0 .5 1.0

H460

SORAFENIB (µM) 0

SORAFENIB (µM) 1

A-769662 (µM)

50 100

F A549

H460

SORAFENIB (µM) 0

.25

.5 1.0 2.0

SORAFENIB (µM) 0

SALICYLATE (mM)

A-769662 (µM)

SORAFENIB (µM)

D A549

SALICYLATE (mM)

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

Figure 1 A

.5 1.0 2.0

Figure 1. Sorafenib synergizes with metformin and other AMPK activators in NSCLC cell lines. (A-F) Colony formation assay of A549 (A, C, E) and H460 (B, D, F) NSCLC cells with increasing concentrations of sorafenib (0-4 μM) in the absence or presence of increasing concentrations of metformin (0-2 mM) (A, B); A-769662 (0-100 μM) (C, D); or salicylate (0-2 mM) (E, F). Cells were grown in 6-well plates and refreshed every 3 days. The cells were fixed, stained and photographed after 10 days.

54 | Chapter 3

Synergistic anti-proliferative effect through AMPK activation and downstream mTOR inhibition It has been shown that the anti-proliferative effect of metformin is mediated by AMPK activation (Ben Sahra et al., 2010; Rocha et al., 2011). We have therefore measured phosphorylation of AMPKÎą at its activation site Thr172 following treatment of cells with sorafenib, metformin, or the combination thereof. Figure 2A and 2B show an unexpected activation of AMPKÎą, as judged by Thr172 phosphorylation, by sorafenib treatment alone. Importantly, AMPK activation was increased significantly when sorafenib was combined with metformin, independent of LKB1 or KRAS mutation status (Fig. 2A and 2B). Increased AMPK activation was also observed when cells were treated with sorafenib in combination with the AMPK activators salicylate (Fig. 2C) or A-769662 (Supplementary Fig. S2A). The anti-proliferative effect of AMPK activation is mediated, at least in part, by suppression of mTOR signaling. Indeed, combination treatment showed a synergistic effect on inhibition of phosphorylation of the downstream mTOR targets 4E-BP1 and ribosomal protein S6 (Fig. 3A and 3B). Beside mTOR, AMPK has many more substrates, one of them being AcetylCoA Carboxylase (ACC) (Mihaylova and Shaw, 2011). ACC is the key enzyme in the biosynthesis and oxidation of fatty acids. Phosphorylation by AMPK at Ser79 inhibits the enzymatic activity of ACC. Treatment of cells with either sorafenib, metformin or salicylate indeed resulted in increased phosphorylation of ACC at Ser79 (Fig. 3A and 3B). Phosphorylation was increased most when cells were treated with the combination of sorafenib and metformin or salicylate (Fig. 3A and 3B).

Sorafenib synergizes with metformin in NSCLC through AMPK pathway activation | 55

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

Figure 2 A

A549 cells LKB1mt KRASmt

sorafenib metformin p-AMPKα

+ -

H460 cells LKB1mt KRASmt

+ +

+ -

+ +

+ -

H358 cells LKB1wt KRASmt -

+ +

+ -

+ +

H838 cells LKB1mt KRASwt -

+ -

+ +

(Thr172)

AMPKα HSP90

H460

A549 sorafenib salicylate

H838 + +

+ -

+ +

p-AMPKα

(Thr172)

AMPKα HSP90

Figure 2. Sorafenib in combination with metformin or the AMPK activator salicylate enhances AMPK activation. (A, B) AMPK activation with the combination of sorafenib and metformin in LKB1-mutant KRAS-mutant NSCLC cells (A549 and H460) (A), LKB1 wild-type KRAS-mutant NSCLC cells (H358) (B, left panel) or LKB1-mutant KRAS wild-type NSCLC cells (H838) (B, right panel). (C) AMPK activation with the combination of sorafenib and salicylate in LKB1-mutant KRASmutant NSCLC cells (A549 and H460) or LKB1-mutant KRAS wild-type NSCLC cells (H838). Cells were treated for 48hrs with sorafenib (1-3 μM), metformin (1-1.5 mM), salicylate (1-1.5 mM) or with the combination of sorafenib/metformin or sorafenib/salicylate using the same concentrations as were used for the individual treatments. Cell lysates were harvested for western blot analysis and probed with the indicated antibodies.

56 | Chapter 3

Figure 3 A

A549 sorafenib metformin

+ -

H358 + +

+ -

H838 + +

+ -

+ +

p-ACC

(Ser79)

ACC p-S6 RP

(Ser240/244)

p-S6 RP

(Ser235/236)

S6 RP p-4E-BP1

(Ser65)

4E-BP1 HSP90

A549 sorafenib salicylate p-ACC

(Ser79)

ACC p-S6 RP

(Ser240/244)

p-S6 RP

(Ser235/236)

S6 RP p-4E-BP1

(Ser65)

4E-BP1 HSP90

+ -

H460 + +

+ -

+ +

Figure 3. Sorafenib in combination with metformin or the AMPK activator salicylate represses the mTOR targets p-S6 and p-4E-BP1. (A) Western blot analysis with the combination of sorafenib and metformin in LKB1-mutant KRASmutant (A549), LKB1 wild-type KRAS-mutant (H358) or LKB1mutant KRAS wild-type (H838) NSCLC cells. (B) Western blot analysis with the combination of sorafenib and salicylate in LKB1mutant KRAS-mutant NSCLC cells. Cells were treated for 48hrs with either sorafenib (1-3 ÎźM), metformin (1-1.5 mM) (A) or salicylate (1-1.5 mM) (B) alone or with the combination of sorafenib/ metformin or sorafenib/salicylate using the same concentrations as were used for the individual treatments. Cell lysates were harvested for western blot analysis and probed with the indicated antibodies.

Sorafenib synergizes with metformin in NSCLC through AMPK pathway activation | 57

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

Sorafenib activates AMPK in both an LKB1-dependent and LKB1-independent, but CAMKK2-dependent manner Activation of AMPK can be brought about through two independent upstream signaling routes: one involving LKB1 and a second calcium-dependent signaling route involving CAMKK2 (Mihaylova and Shaw, 2011). To gain more insight into how sorafenib activates AMPK, we used a number of genetic and chemical inhibitors of these signaling routes. Figure 4A shows that treatment with sorafenib leads to activation of AMPK in both LKB1 wild-type and mutant NSCLC cells in a concentration dependent manner. To extend this finding, we asked whether sorafenib also activates AMPK in hepatocellular carcinoma (HCC) cells, because sorafenib is already approved for clinical use in HCC. Figure 4E shows that treatment of LKB1 wild-type Huh-7 HCC cells with sorafenib also activates AMPK. Furthermore, we show that treatment of cells with fluoro-sorafenib (known as regorafenib) activates AMPK to a similar extent as sorafenib (Supplementary Fig. S2B). Sorafenib is unique in activation of AMPK as other multikinase inhibitors, such as sunitinib (Sutent) or tivozanib, failed to increase p-AMPKÎą levels in cells (data not shown). Next, we examined whether in LKB1 mutant A549 and H460 cells the LKB1independent route is mediated by the protein kinase CAMKK2. As shown in Figure 4B and 4C, inactivation of CAMKK2 in these cells, either by shRNA knockdown or by chemical inhibition, blocked AMPK activation by sorafenib. In contrast, knockdown of CAMKK2 or chemical CAMKK inhibition failed to block the activation of AMPK by sorafenib in LKB1 wild-type H358 cells (Fig. 4B and 4C). Consistent with this, we found that CAMKK2 knockdown in LKB1 wild-type Huh-7 HCC cells is also not sufficient to inhibit the activation of AMPK by sorafenib (Fig. 4E). We next examined the effect of LKB1 knockdown in LKB1 wild-type cells on AMPK activation. As shown in Figure 4D, LKB1 knockdown reduced the basal AMPKÎą phosphorylation and thereby AMPK activity in H358 cells, although sorafenib could still activate AMPK in these cells. It is possible that this residual AMPK activation is mediated by CAMKK2. Consistent with this, we found that treatment with the CAMKK inhibitor STO-609 completely blocked the AMPK activation by sorafenib in these H358 LKB1 knockdown cells (Fig. 4D). Together, these data indicate that sorafenib activates AMPK through two redundant signaling routes that involve LKB1 and CAMKK2, respectively. Only when both kinases are blocked, AMPK activation by sorafenib is compromised. To further address the mechanism underlying the AMPK activation by sorafenib, we tested whether the LKB1-independent AMPK activation mediated by CAMKK2 is calcium-dependent. We therefore co-treated LKB1 mutant A549 and H460 cells with sorafenib and increasing concentrations of the cell permeable calcium chelator

58 | Chapter 3

BAPTA-AM. AMPK activation by sorafenib was reduced in the presence of BAPTAAM in LKB1 mutant A549 and H460 cells (Supplementary Fig. S2C), indicating that the CAMKK2-mediated AMPK activation by sorafenib is dependent on cytosolic calcium release. It has been reported that sorafenib can rapidly provoke the production of Reactive Oxygen Species (ROS) and induce apoptotic death of cells through a mitochondriadependent oxidative stress mechanism (Chiou et al., 2009). The significance of this finding was further highlighted by Coriat et al. (Coriat et al., 2012), who found that the effectiveness of sorafenib in HCC cell lines in vitro is mediated by ROS production. More interestingly, they showed using sera from HCC patients that in response to sorafenib HCC cancer cells produce massive amounts of ROS, which in turn induce tumor cell death. Their data indicate that no or weak ROS production predicts a lack of sorafenib effectiveness. We accordingly tested whether activation of AMPK in our cells was mediated via ROS production, as reported previously (Alexander et al., 2010; Mungai et al., 2011). We found that one-hour treatment with 0.4mM hydrogen peroxide (H2O2, which produces ROS) resulted in highly increased p-AMPKα levels in LKB1 wildtype H358 cells, whereas in LKB1 mutant H460 cells the increase in p-AMPKα after treatment with H2O2 was less abundant (Fig. 4F). AMPK activation by H2O2 could be blocked by co-treatment of H358 cells with the ROS scavenger N-Acetyl-L-Cysteine (NAC) (Fig. 4H). We also found that AMPK activation by H2O2 in LKB1 mutant H460 cells was mediated via CAMKK2, as co-treatment of cells with H2O2 and the CAMKK inhibitor STO-609 effectively blocked AMPK activation (Fig. 4F). In contrast, we still observed a high amount of phosphorylated AMPKα when we treated LKB1 wildtype H358 cells with a combination of H2O2 and STO-609 (Fig. 4F). Consistent with the notion that LKB1 and CAMKK2 can independently activate AMPK, we found that p-AMPKα levels in H358 LKB1 knockdown cells were greatly reduced when cells were treated with the combination of H2O2 and STO-609 (Fig. 4G). Together, these data indicate that the activation of AMPK by ROS, like the activation by sorafenib, is dependent on either LKB1 or CAMKK2. Our proposed model how sorafenib treatment leads to AMPK activation in LKB1 wild-type and LKB1 mutant cells is shown in Supplementary Figure S2D.

Sorafenib synergizes with metformin in NSCLC through AMPK pathway activation | 59

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

Figure 4 Figure 4 A Figure 4 A Figure 4 (uM) sorafenib A sorafenib (uM) p-AMPKα A B B B B

C C C C

D D D D

F F F F

sorafenib (uM) p-AMPKα AMPKα sorafenib (uM) p-AMPKα AMPKα HSP90 p-AMPKα AMPKα HSP90 AMPKα HSP90 HSP90 sorafenib

sorafenib shCAMKK2 sorafenib shCAMKK2 p-AMPKα sorafenib shCAMKK2 p-AMPKα LKB1 shCAMKK2 p-AMPKα LKB1 AMPKα p-AMPKα LKB1 AMPKα HSP90 LKB1 AMPKα HSP90 AMPKα HSP90 sorafenib HSP90 sorafenib STO-609 (uM) 0sorafenib STO-609 (uM) 0p-AMPKα sorafenib STO-609 (uM) 0p-AMPKα AMPKα STO-609 (uM) 0 p-AMPKα AMPKα HSP90 p-AMPKα AMPKα HSP90 AMPKα HSP90

LKB1mt A549 0 LKB1mt 2 4 A549 6 8 0 LKB1mt 2 4 A549 6 8 0 LKB1mt 2 4 A549 6 8 0 2 4 6 8

----

LKB1mt A549 LKB1mt + - +A549 LKB1mt +- #1 - #1 +A549 #2 LKB1mt +- #1 - #1 +A549 #2 +- #1 - #1 + #2 -

+ + #2 + #2 + #2

----

LKB1mt H460 LKB1mt + - +H460 - + LKB1mt H460 + + - #2 + - #1 #1 #2 LKB1mt + - #1 +H460 + - #1 #2 #2 + - #1 + #2 - #2 + - #1

- #1 #1 #2 #2

LKB1mt A549 +LKB1mt - + -A549 + - + LKB1mt + - 15 + 25 -A549 + 50 - 50 + 0 15 25 LKB1mt A549 + + + + 0 15 15 25 25 50 50 - 15 + 25 - 25 + 50 - 50 + 0+ 15 0 15 15 25 25 50 50

HSP90 H358 shctrl sorafenib H358 - + shctrl - + +- + sorafenib - +- shctrl STO-609 H358 +- + sorafenib - +- shctrl STO-609 H358 p-AMPKα - +- + sorafenib + STO-609 p-AMPKα LKB1 STO-609 - - + + p-AMPKα LKB1 HSP90 p-AMPKα LKB1 HSP90 LKB1 HSP90 HSP90 LKB1wt H358 LKB1wt H358 H2O2 - + - + LKB1wt H358 + H2O2 -- +- +- + STO-609 LKB1wt H358 + H2O2 -- +- +- + STO-609 p-AMPKα + H2O2 -- +- +- + STO-609 p-AMPKα AMPKα STO-609 - - + + p-AMPKα AMPKα LKB1 p-AMPKα AMPKα LKB1 HSP90 AMPKα LKB1 HSP90 LKB1 HSP90 HSP90

60 | Chapter 3

LKB1mt H460 0 LKB1mt 2 4 H460 6 8 0 LKB1mt 2 4 H460 6 8 0 LKB1mt 2 4 H460 6 8 0 2 4 6 8

H358 shLKB1 H358 - + - +shLKB1 H358 +- + - +- shLKB1 H358 +- + - +- shLKB1 -

+- + + +

LKB1mt H460 LKB1mt H460 - + - + LKB1mt H460 + - +- +- + LKB1mt H460 + -- +- +- + + -- +- +- + - - + +

LKB1mt H460 + LKB1mt - + -H460 + - + LKB1mt - 50 + + - 15 + 25 -H460 + 50 0 15 25 LKB1mt H460 + + + + 0 15 15 25 25 50 50

00- 50 + + 15 - 15 + 25 - 25 + 50 0- 0 0 0 15 15 25 25 50 50

E E E E

LKB1wt H358 0 LKB1wt 2 4 H358 6 8 0 LKB1wt 2 4 H358 6 8 0 LKB1wt 2 4 H358 6 8 0 2 4 6 8

----

LKB1wt H358 LKB1wt H358 + - + LKB1wt H358 +- #1 - #1 + #2 LKB1wt H358 +- #1 - #1 + #2 +- #1 - #1 + #2 -

+ + #2 + #2 + #2

- #1 #1 #2 #2

LKB1wt H358 + LKB1wt - + -H358 + - + LKB1wt - 50 + + - 15 + 25 -H358 + 50 0 15 25 LKB1wt H358 + + + + 0 15 15 25 25 50 50

00- 50 + - 15 + 25 - 25 + 50 0- + 0 15 0 0 15 15 25 25 50 50

LKB1wt Huh-7 HCC cells LKB1wt Huh-7 HCC cells sorafenib - + - + - + LKB1wt Huh-7 HCC cells sorafenib - +- #1 - #1 + #2 - #2 + shCAMKK2 LKB1wt Huh-7 HCC cells sorafenib - +- #1 - #1 + #2 - #2 + shCAMKK2 p-AMPKα sorafenib - +- #1 - #1 + #2 - #2 + shCAMKK2 p-AMPKα LKB1 - - #1 #1 #2 #2 shCAMKK2 p-AMPKα LKB1 HSP90 p-AMPKα LKB1 HSP90 LKB1 HSP90 H G HSP90 G H358 shLKB1 H LKB1wt H358 G H358 shLKB1 H LKB1wt H358 - + - + H2O2 - + - + G H2O2 H358 shLKB1 H LKB1wt H358 H2O2 -- +- +- + H2O2 + + -- +- +- + NAC LKB1wt STO-609 H358 shLKB1 H358 + + H2O2 - - + + H2O2 + -- +- +- + NAC STO-609 p-AMPKα p-AMPKα H2O2 -- +- +- + H2O2 + + -- +- +- + NAC STO-609 p-AMPKα p-AMPKα AMPKα AMPKα - - + + - - + + NAC STO-609 p-AMPKα p-AMPKα AMPKα AMPKα LKB1 HSP90 p-AMPKα p-AMPKα AMPKα AMPKα LKB1 HSP90 HSP90 AMPKα AMPKα LKB1 HSP90 HSP90 LKB1 HSP90 HSP90 HSP90

E H358 shctrl sorafenib

STO-609

+ -

LKB1wt Huh-7 HCC cells

H358 shLKB1 -

- + + +

sorafenib

- + + +

+ -

p-AMPKα

LKB1

HSP90

shCAMKK2

+ - + - + - #1 #1 #2 #2

G LKB1wt H358

LKB1mt H460

H2O2

STO-609

+ +

H358 shLKB1

LKB1wt H358

H2O2

STO-609 -

NAC

H2O2

p-AMPKα

AMPKα

LKB1

HSP90

Figure 4. Activation of AMPK by sorafenib is dependent on the AMPK kinases LKB1 and CAMKK2. (A) Concentration dependent AMPK activation by sorafenib in LKB1-mutant (A549 and H460) and LKB1 wild-type (H358) NSCLC cells. Cells were treated for 6hrs with increasing concentrations of sorafenib (0-8 μM). (B, C) AMPK activation by sorafenib is CAMKK2dependent in LKB1-mutant (A549 and H460) and CAMKK2-independent LKB1 wild-type (H358) NSCLC cells. (B) Cells expressing shctrl or shCAMKK2 were treated for 6hrs with 6μM sorafenib. (C) Cells were treated for 6hrs with increasing concentrations of the CAMKK inhibitor STO-609 (0-50 μM) in absence or presence of 6 μM sorafenib. (D) AMPK activation by sorafenib is mainly LKB1-dependent in LKB1 wild-type H358 cells. H358 cells expressing shctrl (left panel) or shLKB1 (right panel) were treated with the CAMKK inhibitor STO-609 (0-50 μM) in absence or presence of 6μM sorafenib for 6hrs. The p-AMPK blot in Fig. 4D is relatively high exposed to highlight to differences in AMPK phosphorylation between shctrl cells and shLKB1 H358 cells. (E) AMPK activation by sorafenib is also seen in LKB1 wild-type hepatocellular carcinoma (HCC) Huh-7 cells, independent of CAMKK2. Huh-7 cells expressing shctrl or shCAMKK2 were treated for 6hrs with 6μM sorafenib. (F) AMPK activation by induced oxidative stress is mainly LKB1-dependent in LKB1 wild-type (H358) and completely CAMKK2-dependent in LKB1-mutant (H460) NSCLC cells. Cells were treated for 1hr with 400μM H2O2 in absence or presence of 50μM of the CAMKK inhibitor STO-609. (G) H358 cells expressing shLKB1 were treated for 1hr with 400μM H2O2 in absence or presence of 50μM of the CAMKK inhibitor STO-609. Cells treated with the combination were pre-treated for 30 minutes with the CAMKK inhibitor before adding H2O2. (H) AMPK activation by induced oxidative stress is blocked with the ROS-scavenger N-acetyl-L-cysteine. H358 cells were treated for 1hr with 400μM H2O2 in absence or presence of 20mM of the ROSscavenger N-acetyl-L-cysteine (NAC). Cell lysates were harvested for western blot analysis and probed with the indicated antibodies.

Sorafenib synergizes with metformin in NSCLC through AMPK pathway activation | 61

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

The effect of sorafenib in combination with metformin in a tumor xenograft model To further validate the effectiveness of the combination of sorafenib and metformin for the treatment of NSCLC, we treated nude mice bearing A549 xenografts with vehicle, sorafenib (30 mg/kg/day), metformin (400 mg/kg/day) or the combination of sorafenib and metformin. Surprisingly, the tumor growth rate was higher for the metformin treated mice than for the vehicle treated mice, although this difference was not statistically significant (P = 0.097; Supplementary Table S2). Mean tumor volume after 40 days treatment was significantly lower in the group treated with the combination compared to the group treated with sorafenib monotherapy (P = 0.046; Fig. 5B), with six of the eight mice in the combination group having a smaller tumor volume than the mice in the sorafenib group (Fig. 5B). Also, the mean tumor growth rate was significantly lower in the group treated with the combination compared to sorafenib monotherapy (P = 0.023; Fig. 5C). Based on a linear regression model, the mean tumor growth rate for the vehicle treated group was 7.2 mm3/ day (Supplementary Table S2). Separately, sorafenib and metformin lead to a nonsignificantly increased mean tumor growth rate (increased by 1 mm3/day and 3.9 mm3/day over vehicle, respectively), whereas the combination of both treatments does result in a decrease of the mean tumor growth rate by 2.9 mm3/day compared to vehicle. The interaction coefficient (-7.8) was calculated by the difference between the expected mean growth rate of the combination (1 mm3/day plus 3.9 mm3/day) and the observed mean growth rate of the combination compared to vehicle (-2.9 mm3/day). This interaction was statistically significant (P = 0.026; Supplementary Table S2).

Figure 5. Effect of the combination of sorafenib with metformin in a tumor xenograft model. (A) The growth curves of A549 cells as tumor xenografts in nude mice treated with vehicle (black curve), sorafenib (blue curve), metformin (yellow curve) or the combination (purple curve). Error bars represent SD; n=5-8. The black curve (vehicle) is shown until 25 days after start treatment, because of a bacterial infection outbreak in this group. (B) Mean tumor volume (mm3) after 40 days treatment with sorafenib (blue dots; n=5) or the combination of sorafenib with metformin (purple triangles; n=8). (C) Mean tumor growth rate (mm3/day), calculated by linear regression modeling of the individual growth curves, for the sorafenib treated tumor xenografts (blue dots; n=5) and the tumor xenografts treated with the combination of sorafenib with metformin (purple triangles; n=8). (D) H-score of phospho-4E-BP1 (Thr37/46) immunostaining of A549 xenograft tumor sections of the different treatment groups. (E) Images of A549 xenograft tumor sections of vehicle (top) and combination-treated (bottom) tumors immunostained for phospho-4E-BP1 (400x magnification). P-values <0.05 were considered as statistically significant (*); ns = non-significant.

62 | Chapter 3

Figure 5 A

B *

D *

combination

vehicle

Sorafenib synergizes with metformin in NSCLC through AMPK pathway activation | 63

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

Tumor xenograft sections from the different treatment groups were immunostained for phospho-4E-BP1 and scored using the semi-quantitative H-score (Fig. 5D). Tumors treated with the combination had a significant lower H-score compared to vehicle (P = 0.048) or sorafenib (P = 0.022), although the H-score was not significantly different from tumors treated with metformin. Figure 5E shows representative images of phospho-4E-BP1 stained tumor sections from vehicle and combinationtreated tumors.

DISCUSSION We report here the ability of sorafenib to activate AMPK in NSCLC and HCC cells. This effect of sorafenib is greatly enhanced in combination with the known AMPK activator metformin, providing a rationale for synergism of these drugs to inhibit proliferation of NSCLC cells in vitro and in vivo. The anti-cancer activity of sorafenib has been linked to its multikinase inhibitory action on several major signaling pathways. However, the role of additional pathways involved in the cytostatic and cytotoxic effects of sorafenib has obtained much attention in the last years. It has been reported that mitochondrial and endoplasmic reticulum stress induced by sorafenib is relevant for its effect on cell death (Chiou et al., 2009; Coriat et al., 2012; Rahmani et al., 2007). The mitochondria-dependent oxidative stress leads to the production of ROS and intracellular glutathione (an antioxidant) depletion. It was also shown that sorafenib can trigger cytosolic calcium mobilization and mitochondrial calcium overload (Chiou et al., 2009). Rahmani et al. showed that the increased cytosolic calcium concentrations after treatment with sorafenib promote ROS production in the mitochondria (Rahmani et al., 2007). The importance of ROS for sorafenib induced cell death, was also shown by Coriat et al. who found that sorafenib effectiveness in HCC cell lines in vitro is mediated by ROS production. Even more interestingly, they showed in sera from HCC patients that in response to sorafenib HCC cancer cells produce massive amounts of ROS, which induce tumor cell death. No or weak ROS production predicts a lack of sorafenib effectiveness (Coriat et al., 2012). Interestingly, we could link both calcium mobilization and ROS production by sorafenib to a single molecular effector mechanism, namely the activation of AMPK by either CAMKK2 or LKB1. It was recently reported by Fumarola et al. that sorafenib promotes an early perturbation of mitochondrial function in breast cancer cells (Fumarola et al., 2013). As a response to this stress condition, AMPK was rapidly activated in the cell lines analyzed. Their data also suggest a key role of AMPK-mediated mTORC1 inhibition in the anti-

64 | Chapter 3

tumor activity of sorafenib in breast cancer. The modulation of AMPK by sorafenib in combination with everolimus has been reported in osteosarcoma cells, where the activation is mediated by ROS (Pignochino et al., 2013). Eum et al. described the sustained activation of AMPK by sorafenib in v-Ha-ras-transformed 3T3 cells (Eum et al., 2013). We show that the AMPK activation is also seen with fluoro-sorafenib (regorafenib). Our observation that sorafenib in combination with metformin improves disease control rate in NSCLC patients was unexpected. However, the number of patients using metformin in this study was small and the observation therefore needs further clinical validation. Although additional mechanisms may be involved, we show that increasing AMPK activity by combining sorafenib with AMPK activators can increase the anti-proliferative effects of sorafenib. The anti-proliferative effect from combination of metformin and sorafenib in vitro was recently reported for intrahepatic cholangiocarcinoma cells (Ling et al., 2014). One of the AMPK activators used in our in vitro studies was salicylate, the active metabolite of aspirin. The anti-proliferative effect that we have seen with the combinations of sorafenib and salicylate is stronger than the effect of the combinations of sorafenib and metformin. A possible explanation for this could be the different mechanisms of AMPK activation by salicylate and metformin. Salicylate binds directly to AMPKÎą at the same site as the synthetic AMPK activator A-769662 to cause allosteric activation and inhibition of dephosphorylation of the activating phosphorylation site Thr172 (Hawley et al., 2012). This inhibition of Thr172 dephosphorylation by salicylate will further support sorafenib-mediated AMPK activation. The effects of metformin on AMPK activation are indirect and it is therefore unclear if metformin can also inhibit Thr172 dephosporylation. The plasma salicylate concentrations in humans treated with oral salsalate (Fleischman et al., 2008) or high-dose aspirin (around 7 g/day) (Hundal et al., 2002) are 1-2 mM and accordingly in the concentration range that we used in our in vitro experiments with salicylate. The concentrations of metformin used in our in vitro and in vivo experiments are equal or lower compared to concentrations used in previous studies (Ben Sahra et al., 2010; Chen et al., 2012; Martin et al., 2012; Rocha et al., 2011; Shackelford et al., 2013), but higher than the recommended therapeutic doses in humans. It is, therefore, difficult to extrapolate our results to the potential effects of metformin in a clinical trial with standard doses metformin. For this reason, the related and more potent - but toxic - drug phenformin is used in some mouse studies (Shackelford et al., 2013). However, it has been reported that AMPK-mediated effects of metformin can be seen in vivo with doses comparable to the dose we used in vivo (Martin et al., 2012). Furthermore, our finding that metformin users in a phase

Sorafenib synergizes with metformin in NSCLC through AMPK pathway activation | 65

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

II clinical trial with sorafenib have an improved disease control rate compared to non-metformin users suggests that the effects can be seen with therapeutic doses of metformin, as was shown before in breast cancer patients (Jiralerspong et al., 2009). However, the diabetic patients in our study were using metformin for a long period before administration of sorafenib and the effect can therefore be different for nondiabetic patients starting with metformin in combination with sorafenib. This study also highlights the importance of carefully monitoring co-medications that are used in clinical trials with anti-cancer agents, because these medications may influence the response to the drug of interest. The possible tumor growth acceleration by metformin monotherapy, as observed in our xenograft study, could have clinical implications, as patients with NSCLC treated with metformin for their type II diabetes are at risk for this acceleration. This detrimental effect of metformin on lung cancer has already been described in a large medical report study by Mazzone et al. (Mazzone et al., 2012). They first conclude that the use of metformin is associated with a lower likelihood of developing lung cancer in diabetic patients. However, diabetics who developed lung cancer while receiving metformin were more likely to present with metastatic disease and a shorter survival from the time of diagnosis (Hazard ratio 1.47; 95% CI 1.12-1.92; P = .005). Acceleration of tumor growth by metformin has also been described for BRAFV600E-driven melanomas, but not for NRAS-mutant melanoma. (Martin et al., 2012). Martin et al. found that transcriptional upregulation of VEGFA via AMPK was responsible for the tumor growth acceleration. We tested in our NSCLC cell lines whether transcriptional VEGFA upregulation occurred in vitro. We observed a slight (up to 2-fold) upregulation of VEGFA mRNA after 48hrs treatment with 1-3 mM metformin, independent of LKB1 or KRAS mutation status (Supplementary Fig. S3A-C). We also analyzed the number of CD31-positive vessels in the A549 xenograft tumor sections from the different treatment groups (Supplementary Fig. S3D). Tumors treated with metformin monotherapy had a significant increase in the number of vessels compared to vehicle (P = 0.021), sorafenib monotherapy (P < 0.001) or the combination (P < 0.001), indicating that the VEGFA upregulation leads to an increase in tumor vessel density in vivo. A study by Shackelford et al. demonstrated that phenformin was selectively effective in LKB1-deficient NSCLC animal models (Shackelford et al., 2013). We showed that the synergy between metformin and sorafenib is also seen in LKB1mutant NSCLC cells, which use CAMKK2 to activate AMPK. We did not test the effect of phenformin in combination with sorafenib, but it could well be that the effect in LKB1-deficient cells is larger when phenformin is used. The sensitivity of cells to biguanides might also be dependent on the glucose concentration. Birsoy

66 | Chapter 3

et al. showed that cells with defective mitochondrial oxidative phosphorylation, as a result of mitochondrial DNA mutations or impaired glucose utilization, were selectively sensitive to biguanides under metabolic stress conditions (Birsoy et al., 2014). For this reason, it is important to state that all the cell lines used in our experiments were grown in high-glucose medium. Our data provide a rationale for a clinical trial combining sorafenib and metformin in NSCLC or other cancer types in which sorafenib is either approved (HCC, renal cell carcinoma) or under clinical investigation. The fact that metformin is already used extensively in the clinic with minimal side effects together with the favorable therapeutic range of metformin makes it relative easy to start these clinical trials.

MATERIALS AND METHODS Phase II clinical trial of sorafenib in KRAS-mutant advanced NSCLC Previously, Dingemans et al. reported the activity of sorafenib monotherapy in a single arm phase II trial in patients with locally advanced or metastatic nonsquamous NSCLC harboring a KRAS mutation (Dutch trial register NL30000.029.09) (Dingemans et al., 2013). We used their study to perform a post hoc analysis on the group of patients receiving metformin for their type II diabetes. The primary endpoint of their study was disease control rate (DCR), defined as no progression after 6 weeks of treatment (RECIST 1.0 criteria). Secondary endpoints were Overall Response Rate (ORR), duration of response, Progression Free Survival (PFS) and Overall Survival (OS). Reagents Sorafenib (cat. no. S1040) was purchased from Selleck Chemicals. Metformin hydrochloride (cat. no. PHR1084), sodium salicylate (cat. no. S2679), Hydrogen Peroxide solution (cat. no. 216763) and N-Acetyl-L-cysteine (cat. no. A9165) were purchased from Sigma Aldrich. STO-609 (cat. no. sc-202820) was obtained from Santa Cruz Biotechnology. Cell Culture and Viral Transduction A549, H358, H460, H522 and H838 cells were purchased from American Type Culture Collection (ATCC) and Huh-7 cells from the Japanese Collection of Research Bioresources (JCRB) Cell Bank. An overview of the KRAS and LKB1 status of these cell lines is shown in Supplementary Table S3. The cells were cultured in DMEM (Huh-7) or RPMI (other cell lines) supplemented with 8% heat-inactivated fetal

Sorafenib synergizes with metformin in NSCLC through AMPK pathway activation | 67

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

calf serum and 1% penicillin/streptomycin at 5% CO2. HEK293T cells were used as producers of lentiviral supernatants as described (http://www.broadinstitute. org/rnai/public/resources/protocols). The calcium phosphate method was used for the transfection of 293T cells. Infected cells were selected for successful lentiviral integration using 2 μg/ml of puromycin. Plasmids All lentiviral shRNA vectors were retrieved from the arrayed TRC human genome-wide shRNA collection. The following RNAi target sequences were used: shCAMKK2#1, CGAGCGGATCATGTGTTTACA; shCAMKK2#2, CCGTTTCTACTTCCAGGATCT; shLKB1, GATCCTCAAGAAGAAGAAGTT. Control infections were performed with the empty pLKO.1 vector. Colony formation assays Cells were seeded in 6-well plates (5-10 × 103 cells per well) and cultured both in the absence and presence of drugs as indicated for 10 days. Cells were washed with phosphate buffered saline (PBS), fixed with 4% formaldehyde in PBS and stained with 0.1% crystal violet. Protein lysate preparation and immunoblots The biochemical responses of cells treated with drugs were analyzed by western blot. Cells were seeded in 6 well plates in medium containing 8% fetal calf serum. After 12h, cells were treated with drugs for the indicated time without refreshing the medium. The lysates were collected using radioimmunoprecipitation assay (RIPA) buffer containing 150 mM NaCl, 50 mM Tris pH 8.0, 1% NP-40, 0.5% sodium deoxycholate and 0.1% SDS supplemented with protease inhibitors (Complete, Roche) and Phosphatase Inhibitor Cocktails II and III (Sigma). All lysates were freshly prepared, normalized using bicinchoninic acid (BCA) protein assay (Thermo Scientific) and resolved by SDS gel electrophoresis and followed by western blotting. Primary antibodies against p-AMPKα Thr172 (40H9; #2535), AMPKα (#2532), LKB1 (26D10; #3050), p-ACC Ser79 (#3661), ACC (#3662), p-S6 RP Ser240/244 (#2215), p-S6 RP Ser235/236 (#2211), S6 RP (5G10; #2217) p-4E-BP1 Ser65 (174A9; #9456) and 4E-BP1 (#9452) were from Cell Signaling Technology and HSP 90α/β (H-114; sc7947) from Santa Cruz Biotechnology. Secondary antibody was obtained from BioRad Laboratories. Quantitative RT-PCR The 7500 Fast Real-Time PCR System from Applied Biosystems was used to

68 | Chapter 3

measure mRNA levels. mRNA expression levels were normalized to expression of GAPDH. The following primer sequences were used in the SYBR Green master mix (Roche): GAPDH_forward, 5’-AAGGTGAAGGTCGGAGTCAA-3’; GAPDH_reverse, AATGAAGGGGTCATTGATGG; VEGFA_ forward, 5’-CCCACTGAGGAGTCCAACAT-3’; VEGFA_reverse, 5’-TTTCTTGCGCTTTCGTTTTT-3’. Mouse xenografts and in vivo drug study All experimental procedures on animals were approved by the Animal Ethics Committee of the Netherlands Cancer Institute in accordance with the Dutch Act on Animal Experimentation. A549 cells (1.8 × 106 cells per mouse) were injected subcutaneously into the right posterior flanks of 7-week-old immunodeficient Balb/c female nude mice (5-8 mice per group) (Charles River). Tumor formation was monitored every other day, and tumor volume based on caliper measurements was calculated by the modified ellipsoidal formula: tumor volume = 0.5 x length × width2. When tumors reached a volume of approximately 50 mm3, mice were randomly assigned to treatment with vehicle (PEG400 in sterile PBS (1:1) by daily gavage), sorafenib (30 mg/kg of body weight by daily gavage), metformin (400 mg/kg/day dissolved in the drinking water, assuming an average water consumption per day per mouse as was calculated before the start of the study) or to the drug combination (sorafenib plus metformin), in which each compound was administered at the same dose and schedule as single agents. Sorafenib for in vivo study was dissolved in DMSO, stored in aliquots at −80 °C and diluted in vehicle before administration Immunohistochemistry Tumor sections from FFPE A549 xenograft tumors were stained for p-4E-BP1 [Thr37/46 (#2855), dilution 1:5000; Cell Signaling]. Detection was performed using CSA II signal amplification system (Dako). Sections were scored by two pathologists (H.M.H. and S.M.W.) using the semi-quantitative H-score that takes into consideration the staining intensity (0-3+) in conjunction with the percentage of viable tumor cells staining positively. H-score = (% at 0) * 0 + (% at 1+) * 1 + (% at 2+) * 2 + (% at 3+) * 3. Thus, this score produces a continuous variable that ranges from 0 to 300. Tumor vessel density was analyzed in FFPE A549 xenograft tumors. Sections were stained for CD31 [Anti-CD31 (#ab28364), dilution 1:50; Abcam]. Detection was performed using the EnVisionTM+ system (Dako). Two pathologists (H.M.H. and S.M.W.) counted the number of CD31-positive vessels in 10 randomly clockwise selected fields (200x magnification) containing viable tumor. Vessel density was calculated as the sum of CD31-positive vessels in the 10 fields. Sorafenib synergizes with metformin in NSCLC through AMPK pathway activation | 69

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

Statistical analysis The differences in response rate in the Phase II clinical trial between patients not using metformin and patients using metformin were tested using a Chi-square test with 3 categories (two-tailed P-value). Differences in tumor volume, tumor growth rate, phospho-4E-BP1 immunostaining and number of CD31-positive vessels between treatment groups from the xenograft study were tested for significance using an unpaired T-test (two-tailed P-value) in GraphPad Prism 6 software. Combination treatment was assessed using a linear regression of tumor growth rate on treatment groups including an interaction term using SPSS software. P-values <0.05 were considered as statistically significant.

ACKNOWLEDGEMENTS The authors thank Ute Boon for help with the tumor xenograft experiments. We also thank the members of the Bernards lab for their helpful support and discussions.

70 | Chapter 3

REFERENCES Alexander, A., Cai, S.L., Kim, J., et al., 2010. ATM signals to TSC2 in the cytoplasm to regulate mTORC1 in response to ROS. Proc Natl Acad Sci U S A 107, 4153-4158. Ben Sahra, I., Le Marchand-Brustel, Y., Tanti, J.F., et al., 2010. Metformin in cancer therapy: a new perspective for an old antidiabetic drug? Mol Cancer Ther 9, 1092-1099. Birsoy, K., Possemato, R., Lorbeer, F.K., et al., 2014. Metabolic determinants of cancer cell sensitivity to glucose limitation and biguanides. Nature 508, 108-112. Chen, G., Xu, S., Renko, K., et al., 2012. Metformin inhibits growth of thyroid carcinoma cells, suppresses self-renewal of derived cancer stem cells, and potentiates the effect of chemotherapeutic agents. J Clin Endocrinol Metab 97, E510-520. Chiou, J.F., Tai, C.J., Wang, Y.H., et al., 2009. Sorafenib induces preferential apoptotic killing of a drug- and radio-resistant Hep G2 cells through a mitochondria-dependent oxidative stress mechanism. Cancer Biol Ther 8, 19041913. Coriat, R., Nicco, C., Chereau, C., et al., 2012. Sorafenib-induced hepatocellular carcinoma cell death depends on reactive oxygen species production in vitro and in vivo. Mol Cancer Ther 11, 2284-2293. Ding, L., Getz, G., Wheeler, D.A., et al., 2008. Somatic mutations affect key pathways in lung adenocarcinoma. Nature 455, 1069-1075. Dingemans, A.M., Mellema, W.W., Groen, H.J., et al., 2013. A Phase II Study of Sorafenib in Patients with Platinum-Pretreated, Advanced (Stage IIIb or IV) Non-Small Cell Lung Cancer with a KRAS Mutation. Clin Cancer Res 19, 743751. Dowling, R.J., Niraula, S., Stambolic, V., et al., 2012. Metformin in cancer: translational challenges. J Mol Endocrinol 48, R31-43. Eum, K.H., Ahn, S.K., Kang, H., et al., 2013. Differential inhibitory effects of two Raf-

targeting drugs, sorafenib and PLX4720, on the growth of multidrug-resistant cells. Mol Cell Biochem 372, 65-74. Evans, J.M., Donnelly, L.A., Emslie-Smith, A.M., et al., 2005. Metformin and reduced risk of cancer in diabetic patients. BMJ 330, 1304-1305. Fleischman, A., Shoelson, S.E., Bernier, R., et al., 2008. Salsalate improves glycemia and inflammatory parameters in obese young adults. Diabetes Care 31, 289-294. Fryer, L.G., Parbu-Patel, A., Carling, D., 2002. The Anti-diabetic drugs rosiglitazone and metformin stimulate AMP-activated protein kinase through distinct signaling pathways. J Biol Chem 277, 25226-25232. Fumarola, C., Caffarra, C., La Monica, S., et al., 2013. Effects of sorafenib on energy metabolism in breast cancer cells: role of AMPK-mTORC1 signaling. Breast Cancer Res Treat 141, 67-78. Gill, R.K., Yang, S.H., Meerzaman, D., et al., 2011. Frequent homozygous deletion of the LKB1/STK11 gene in non-small cell lung cancer. Oncogene 30, 3784-3791. Hardie, D.G., 2013. The LKB1-AMPK PathwayFriend or Foe in Cancer? Cancer Cell 23, 131132. Hawley, S.A., Fullerton, M.D., Ross, F.A., et al., 2012. The ancient drug salicylate directly activates AMP-activated protein kinase. Science 336, 918-922. Hemminki, A., Markie, D., Tomlinson, I., et al., 1998. A serine/threonine kinase gene defective in Peutz-Jeghers syndrome. Nature 391, 184187. Hirsch, H.A., Iliopoulos, D., Tsichlis, P.N., et al., 2009. Metformin selectively targets cancer stem cells, and acts together with chemotherapy to block tumor growth and prolong remission. Cancer Res 69, 7507-7511. Huang, X., Wullschleger, S., Shpiro, N., et al., 2008. Important role of the LKB1-AMPK pathway in suppressing tumorigenesis in PTEN-deficient mice. Biochem J 412, 211-221.

Sorafenib synergizes with metformin in NSCLC through AMPK pathway activation | 71

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

Hundal, R.S., Petersen, K.F., Mayerson, A.B., et al., 2002. Mechanism by which high-dose aspirin improves glucose metabolism in type 2 diabetes. J Clin Invest 109, 1321-1326. Iliopoulos, D., Hirsch, H.A., Struhl, K., 2011. Metformin decreases the dose of chemotherapy for prolonging tumor remission in mouse xenografts involving multiple cancer cell types. Cancer Res 71, 3196-3201.

Mungai, P.T., Waypa, G.B., Jairaman, A., et al., 2011. Hypoxia triggers AMPK activation through reactive oxygen species-mediated activation of calcium release-activated calcium channels. Mol Cell Biol 31, 3531-3545. Pierotti, M.A., Berrino, F., Gariboldi, M., et al., 2013. Targeting metabolism for cancer treatment and prevention: metformin, an old drug with multi-faceted effects. Oncogene 32, 1475-1487.

Jiralerspong, S., Palla, S.L., Giordano, S.H., et al., 2009. Metformin and pathologic complete responses to neoadjuvant chemotherapy in diabetic patients with breast cancer. J Clin Oncol 27, 3297-3302.

Pignochino, Y., Dellâ&#x20AC;&#x2122;Aglio, C., Basirico, M., et al., 2013. The Combination of Sorafenib and Everolimus Abrogates mTORC1 and mTORC2 upregulation in osteosarcoma preclinical models. Clin Cancer Res 19, 2117-2131.

Liao, X., Lochhead, P., Nishihara, R., et al., 2012. Aspirin use, tumor PIK3CA mutation, and colorectal-cancer survival. N Engl J Med 367, 1596-1606.

Rahmani, M., Davis, E.M., Crabtree, T.R., et al., 2007. The kinase inhibitor sorafenib induces cell death through a process involving induction of endoplasmic reticulum stress. Mol Cell Biol 27, 5499-5513.

Ling, S., Feng, T., Ke, Q., et al., 2014. Metformin inhibits proliferation and enhances chemosensitivity of intrahepatic cholangiocarcinoma cell lines. Oncol Rep 31, 2611-2618. Martin, M.J., Hayward, R., Viros, A., et al., 2012. Metformin accelerates the growth of BRAF V600E-driven melanoma by upregulating VEGF-A. Cancer Discov 2, 344-355. Mazzone, P.J., Rai, H., Beukemann, M., et al., 2012. The effect of metformin and thiazolidinedione use on lung cancer in diabetics. BMC Cancer 12, 410. Mihaylova, M.M., Shaw, R.J., 2011. The AMPK signalling pathway coordinates cell growth, autophagy and metabolism. Nat Cell Biol 13, 1016-1023.

72 | Chapter 3

Rocha, G.Z., Dias, M.M., Ropelle, E.R., et al., 2011. Metformin amplifies chemotherapyinduced AMPK activation and antitumoral growth. Clin Cancer Res 17, 3993-4005. Shackelford, D.B., Abt, E., Gerken, L., et al., 2013. LKB1 Inactivation Dictates Therapeutic Response of Non-Small Cell Lung Cancer to the Metabolism Drug Phenformin. Cancer Cell 23, 143-158. Sun, W., Lee, T.S., Zhu, M., et al., 2006. Statins activate AMP-activated protein kinase in vitro and in vivo. Circulation 114, 2655-2662.

SUPPLEMENTARY DATA Supplementary Figure S1 A H358

H838 4

.25

1.0 1.5

H522

SORAFENIB (µM) 0

SALICYLATE (mM)

0 .5 1.0 2.0

.5 1.0 2.0

2.0

SORAFENIB (µM) 0

SALICYLATE (mM)

SORAFENIB (µM)

Supplementary Figure S1. Effect of the combination of sorafenib and the AMPK activator salicylate is independent of LKB1 and KRAS status. (A-C) Colony formation assay of LKB1 wildtype KRAS mutant H358 cells (A), LKB1 mutant KRAS wild-type H838 cells (B), or double wild-type cells H522 cells (C) with increasing concentrations of sorafenib (0-4 μM) in the absence or presence of increasing concentrations of salicylate (0-2 mM). Cells were grown in 6-well plates and refreshed every 3 days. The cells were fixed, stained and photographed after 10 days.

Sorafenib synergizes with metformin in NSCLC through AMPK pathway activation | 73

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

Supplementary Figure S2 A A549

H838

sorafenib

A-769662

p-AMPKα

(Thr172)

AMPKα HSP90

LKB1mt A549 regorafenib (uM)

LKB1mt H460

LKB1wt H358 8

p-AMPKα AMPKα

A549 -

H460 -

sorafenib

BAPTA-AM (uM)

0 10 10 20 20 0

0 10 10 20 20

p-AMPK HSP90

D Supplementary Figure S2. (A) AMPK activation with the combination of sorafenib and the allosteric AMPK activator Sorafenib Oxidative stress / ER stress A-769662 in LKB1 mutant KRAS mutant (A549) and LKB1 mutant KRAS wild-type (H838) NSCLC cells. (B) AMPK activation after treatment with regorafenib (fluoro-sorafenib). LKB1 mutant KRAS mutant (A549 and H460) and LKB1 wild-type KRAS mutant (H358) NSCLC cells were treated for 6hrs with increasing concentrations of the compound. (C) The CAMKK2mediated AMPK activation by Calcium sorafenib is dependent on cytosolic calcium release. A549 (Ca2+) Reactive Oxygen Species (ROS) and H460 cells were treated for 6hrs with increasing concentration of the calcium-chelator release BAPTA-AM (0-20 μM) in absence or presence of 6μM sorafenib. Cell lysates were harvested for western blot analysis and probed with the indicated antibodies. AMP STO-609

MO25

CAMKK2

STRAD AMPK activation is amplified by: Metformin A-769662 Salicylate

74 | Chapter 3

AMPK

LKB1 P

AMPKÎą

A549 -

H460 -

sorafenib

BAPTA-AM (uM)

0 10 10 20 20 0

0 10 10 20 20

p-AMPK HSP90

D Sorafenib

Oxidative stress / ER stress

Calcium (Ca2+) release

Reactive Oxygen Species (ROS)

3 AMP STO-609

MO25

CAMKK2

LKB1 P

STRAD AMPK activation is amplified by: Metformin A-769662 Salicylate

AMPK

Supplementary Figure S2D. Proposed model how sorafenib treatment leads to AMPK activation in LKB1 wild-type and LKB1 mutant cells. Sorafenib induces oxidative stress and endoplasmic reticulum (ER) stress in the cells leading to the production of reactive oxygen species (ROS) (23, 26) and the release of calcium (22). The release of calcium can activate CAMKK2 upstream of AMPK, which subsequently phosphorylate AMPK in both LKB1 wild-type and LKB1 mutant cells. The production of ROS can activate AMPK in an LKB1-dependent mechanism in LKB1 wildtype cells (Fig. 4F-H). The increase in ROS will also trigger increases in cytosolic calcium that activate AMPK in a CAMKK2-dependent manner (24). This explains the AMPK activation by H2O2 in LKB1 mutant cells (Fig. 4F).

Sorafenib synergizes with metformin in NSCLC through AMPK pathway activation | 75

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

Supplementary Figure S3 A

***

Supplementary Figure S3. VEGFA mRNA upregulation and increased tumor vessel density in response to treatment with metformin. (A-C) Relative VEGFA mRNA expression in A549, H358, H522 cells treated for 48hrs with the indicated concentrations of metformin. RNA was isolated from these cells and VEGFA mRNA expression was detected by qRT-PCR and corrected for GAPDH mRNA expression. Error bars denote technical SD. (D) Tumor vessel density in A549 xenograft tumors of the different treatment groups. The vessel density was calculated as the sum of the number of CD31-positive vessels in 10 randomly selected fields (200x magnification) per tumor. P <0.05 *; P <0.001 ***; ns = non-significant. 76 | Chapter 3

Supplementary Table S1. Basic characteristics of the non-metformin users (n = 52) and the metformin users (n = 5) in the phase II clinical trial in patients with locally advanced or metastatic non-squamous NSCLC harboring a KRAS mutation.

No (n=52) n (%)

Metformin

Yes (n=5) n (%)

Median age (SD)

58 (± 7,7)

65 (± 10,8)

Male Female

14 (26,9%)

2 (40,0%)

Former 40 (77,0%) Current 11 (21,1%)

4 (80,0%)

Sex

Smoking status

38 (73,1%)

1 (20,0%)

missing

1 (1,9%)

Grade 0

21 (40,4%)

3 (60,0%)

Grade 2

3 (5,8%)

0 (0,0%)

ECOG performance status (grade 0-5)

Grade 1 28 (53,8%) Histology

3 (60,0%)

adeno

42 (80,8%)

large cell 6 (11,5%) squamous 1 (1,9%) broncho-alveolar

Stage

IIIb

Previous lines of treatment

3 (5,8%) 4 (7,7%)

48 (92,3%)

1 31 (59,6%) >1 21 (40,4%)

2 (40,0%)

4 (80,0%) 0 (0,0%)

0 (0,0%)

1 (20,0%) 2 (40,0%)

3 (60,0%)

1 (20,0%)

4 (80,0%)

Abbreviations: SD, standard deviation; ECOG, Eastern Cooperative Oncology Group

Sorafenib synergizes with metformin in NSCLC through AMPK pathway activation | 77

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

Supplementary Table S2. SPSS output table. Linear regression of tumor growth rate on treatment groups including an interaction term.

Unstandardized Standardized 95% Confidence Coefficients coefficients Interval for B B1 Std. Error Beta t-value Sig. 1 Lower Bound Upper Bound

(Constant) Vehicle 7.168

1.670

Sorafenib

2.477

.983

.107

4.293

.000

3.705

10.632

.397

.695

-4.154

6.120

Metformin

3.942

2.276

.426

1.732

.097

-.778

8.661

Combination

-7.799

3.258

-.787

-2.394

.026

-14.557

-1.042

This column contains the values for the regression equation. The mean growth rate of the vehicle treated group was 7.168 mm3/day. The growth rates for the sorafenib (.983) and metformin (3.942) treated groups in this table are relative to the vehicle treated group (constant). The interaction coefficient (-7.799) was calculated by the difference between the expected mean growth rate of the combination (0.983 mm3/day + 3.942 mm3/day) and the observed mean growth rate for the combination compared to vehicle (-2.874 mm3/day). 2 This columns provide two-tailed P-value used in testing the null hypothesis. Â 1

Abbreviations: Std. error, standard error; Sig., significance.

Supplementary Table S3. KRAS and LKB1 status of the cell lines. Cell Line

Origin

KRAS status

LKB1 status

NCI-H358

NSCLC

G12C

wild-type

A549

NCI-H460

NCI-H522

NCI-H838 Huh-7

NSCLC

NSCLC HCC

G12S

Q61H

wild-type

Q37*

wild-type T212fs*75

wild-type

Abbreviations: NSCLC, non-small cell lung cancer; HCC, hepatocellular carcinoma.

78 | Chapter 3

Supplementary Table S4. Quantification of the colony formation assays shown in Figure 1. Colony formations were quantified using the Envision plate reader after dissolving the absorbed crystal violet with 10% acetic acid. The dose effects of the compounds individually and in the combinations are relative to the untreated control. Figure 1A. Metformin + sorafenib in A549 cells

Metformin

Sorafenib

0 mM

0.5 mM

1 mM

2 mM

0 μM

0.25 μM

0.5 μM

1 μM

2 μM

4 μM

0,16

0,35

0,48

0,54

0,70

0,92

0,00

0,33

0,56

0,00

0,59

0,79

Figure 1B. Metformin + sorafenib in H460 cells Metformin

Sorafenib

0 mM

0.5 mM

1 mM

Sorafenib

0 μM

50 μM

100 μM

Sorafenib

0 μM

50 μM

100 μM

0,88

0,89

0,95

1 μM

2 μM

4 μM

0,04

0,06

0,09

0,13

0,40

0,89

0,19

0,31

0,00

0,14

0,47

0,62

0,87

0 μM

0.5 μM

1 μM

2 μM

4 μM

0,02

0,37

0,64

0,83

0,82

0,00

0,33

0,01

0,67

0,03

0,31

0,84

0,87

0,88

0 μM

0.5 μM

1 μM

2 μM

4 μM

0,03

0,04

0,06

0,49

0,94

0,00

0,07

0,00

0,48

0,04

0,29

0,64

0,82

0,96

1 μM

2 μM

4 μM

0.25 mM

0,01

0,09

0,36

0,71

0,89

2 mM

0,22

0,28

0,35

0,01

0,60

0,68

0,92

Figure 1F. Salicylate + sorafenib in H460 cells

0,03

0,76

0,87

0,31

0,86

0,91

0,68

0,94

0,97

0,94

0,96

Sorafenib

0 μM

0.5 μM

1 μM

2 μM

4 μM

0.25 mM

0,00

0,16

0,38

0,67

0,97

0 mM

0.5 mM

1 mM

2 mM

0,00

0,10

0,57

0,00

0,53

0,76

0,94

0,04

0,64

0,80

0,96

0,29

0,93

0,96

0,97

0,96

0,85

0.5 μM

0,00

0,80

0,68

0 μM

0.5 mM

0,95

0,84

Sorafenib

0 mM

1 mM

Salicylate

0,89

0,76

0.5 μM

0,00

Figure 1E. Salicylate + sorafenib in A549 cells Salicylate

0,83

0,46

0.25 μM

0,00

Figure 1D. A-769662 + sorafenib in H460 cells A-769662

0,76

0,27

0 μM

Figure 1C. A-769662 + sorafenib in A549 cells A-769662

0,10

0,85

0,97

Sorafenib synergizes with metformin in NSCLC through AMPK pathway activation | 79

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

Chapter 4 Estrogen receptor splice variants as a potential source of false-positive estrogen receptor status in breast cancer diagnostics

Floris H. Groenendijk1, Wilbert Zwart2, Arno Floore3, Stephanie Akbari4 and RenĂŠ Bernards1,3

Division of Molecular Carcinogenesis, Cancer Genomics Centre; 2 Division of Molecular Pathology; The

Netherlands Cancer Institute, Plesmanlaan 121, 1066 CX Amsterdam, The Netherlands. 3

Department of Research and Development, Agendia NV, Amsterdam, The Netherlands.

Center for Breast Health, Virginia Hospital Center, Arlington, VA, USA.

Published in Breast Cancer Research and Treatment 2013;140:475-484.

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

ABSTRACT It is well established that only estrogen receptor (ER)-positive tumors benefit from hormonal therapies. We hypothesized that a subgroup of breast cancer patients expresses estrogen receptor α (ERα), but do not respond to hormonal therapy due to the expression of a non-functional receptor. We analyzed a series of 2,658 ERαpositive HER2-negative breast tumors for ERα and progesterone receptor (PR) status, as determined by mRNA expression, and for their molecular subtypes (Luminaltype vs Basal-type, assessed by BluePrint™ molecular subtyping assay). In addition, we assessed the recurrence risk (low vs high) using the 70-gene MammaPrint™ signature. We found that 55 out of 2,658 tumors (2.1%) that are ERα positive by mRNA analysis also demonstrate a basal molecular subtype, indicating that they lack expression of estrogen-responsive genes. These ERα-positive basal-type tumors express significantly lower levels of both ERα and PR mRNA as compared to luminaltype tumors (P < 0.0001) and almost invariably (94.5%) have a high-risk MammaPrint profile. Twelve of the MammaPrint genes are directly ERα responsive, indicating that MammaPrint assesses ERα function in breast cancer without considering ERα mRNA levels. We find a relatively high expression of the dominant negative ERα splice variant ERΔ7 in ERα-positive basal-type tumors as compared to ERα-positive luminal-type tumors (P < 0.0001). Expression of the dominant negative ERα variant ERΔ7 provides a rationale as to why tumors are of the basal molecular subtype while staining ERα positive by immunohistochemistry. These tumors may lack a functional response to estrogen and consequently may not respond to hormonal therapy. Our data indicate that such patients are of MammaPrint high recurrence risk and might benefit from adjuvant chemotherapy.

KEYWORDS Breast cancer • Estrogen receptor variants • Intrinsic subtypes • Molecular subtypes • Tamoxifen

82 | Chapter 4

INTRODUCTION The female hormone estradiol (E2) is a potent mitogen for estrogen receptor α (ERα)-positive breast cancers. Hence, ERα protein levels, as determined by immunohistochemistry (IHC), are strongly predictive for response to endocrine therapies (Davies et al., 2011). Seventy-five percent of all breast cancers express ERα, but not all tumors that express this steroid receptor respond to hormonal therapies. ERα is a member of the nuclear hormone receptor gene family that regulates transcription in a hormone-dependent fashion through sequence-specific DNA binding (Sommer and Fuqua, 2001). Indeed, ERα binding sites are found proximal to many genes and consequently estrogen stimulation of breast cancer cells leads to significant changes in cellular gene expression (Carroll et al., 2006; Hah et al., 2011). These responsive genes include the progesterone receptor (PR), one of the best-characterized ERα target genes. Hence, the PR is often co-expressed with ERα in breast cancers and PR testing is commonly performed in conjunction with ERα testing to assess hormone receptor status of a breast tumor. However, PR status is not a strong predictor of response to endocrine therapy, indicating that PR expression is not solely controlled by ERα activity (Dowsett et al., 2010). Over a decade ago, the first large-scale gene expression profiling studies in breast cancer demonstrated that breast cancers consist of a number of “intrinsic” or “molecular” subtypes that are characterized by similarities in gene expression patterns (Perou et al., 2000). Among these intrinsic subtypes are the “Luminal” and “Basal” tumors, which are thought to represent primarily ER-positive and ERnegative tumors, respectively. Consistent with this view, it was demonstrated that BluePrint™, an 80-gene mRNA expression signature that identifies luminal and basal tumors, is significantly enriched in bona fide ERα target genes (Krijgsman et al., 2012). These data suggest that this intrinsic subtype signature primarily measures the functionality of the ER, as judged by expression of its downstream target genes. As such, this signature also has the potential to identify a subgroup of breast cancer patients who are ERα positive by IHC and/or mRNA expression, but fail to elicit the hormone-induced transcriptional responses that normally result from ER stimulation (ERα target genes “off”; basal-type). Such a scenario would imply that breast cancers having this phenotype express a dysfunctional ERα protein that can nevertheless be detected by IHC. Several different ERα variant mRNAs have been described in human breast cancer. Almost all of these naturally occurring variants are mRNA splicing variants, in which one or more exons are absent from the ERα mRNA. In most ERα splicing variants, except for variants lacking exon 3 or 4, translation runs out of frame after

Estrogen receptor splice variants as a potential source of false-positive estrogen receptor status | 83

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

the site of the splicing variation, leading to a truncated protein (Fuqua et al., 1992; Fuqua et al., 1991; Herynk and Fuqua, 2004; McGuire et al., 1991; Zhang et al., 1993). Since the antibodies for ERα used in IHC often include those that recognize an epitope encoded by the first exon of the ERα gene (Hammond et al., 2010), such splice variants are likely detected as IHC positive for ERα, even though their function may be different from the normal ERα protein. The functional activity of these variant ERα proteins can be negative, dominant negative, or dominant active on ERα target genes. Dominant negative variants are not only inactive themselves but also inactivate wild-type ERα through heterodimerization. Two variants, the ERΔ3 and the ERΔ7 variants, have been described as dominant negative receptor forms in the presence of wild-type ERα (Fuqua et al., 1992; Fuqua et al., 1991; Herynk and Fuqua, 2004; McGuire et al., 1991; Zhang et al., 1993). The ERΔ7 mRNA has been reported to be the major alternatively spliced form in most human breast tumors and cancer cell lines (Garcia Pedrero et al., 2003). The ERΔ7 is especially interesting because the hormone-binding domain, the transcription activation function-2 domain, and the dimerization domain are all partially located in exon 7 (Fig. 1). It has been shown that the ERΔ7 variant has the ability to suppress the E2-dependent transcriptional activation by both wild-type ERα and ERβ (Garcia Pedrero et al., 2003).

Figure 1

Nuclear localization TAF-2

TAF-1

Functional Domains bp 1

DNA binding

Hormone binding Dimerization

Dimerization 293

exon

2078

1 aa 1

5 6

8 aa 595

Figure 1. Organization of the ERα mRNA and functional domains. Abbreviations: TAF-1, transcription activation function 1; TAF-2, transcription activation function 2; aa, amino acid; bp, base pair.

According to the guideline recommendations from the American Society of Clinical Oncology (ASCO) and the College of American Pathologists (CAP) for IHC testing of ERα and PR in breast cancer, it is recommended that ERα assays should be considered positive if there are at least 1% (weakly) positive tumor nuclei in the sample (Hammond et al., 2010). This threshold is based on a cut-point analysis correlating IHC scores with outcome in patients treated with adjuvant endocrine therapy alone, where patients with a score correlating to 1–10% weakly positive cells

84 | Chapter 4

had a statistically significant better prognosis than patients with scores correlating with <1% positive cells (Harvey et al., 1999). However, Iwamoto et al. have shown recently that only a minority of the borderline (1–9% positive nuclei) IHC ERαpositive tumors are of the luminal subtype (as identified by the PAM50 classifier (Chia et al., 2012)) and that most of these borderline ERα-positive samples are of the basal molecular subtype (Iwamoto et al., 2012). Here we identify in a large cohort of molecular profiled breast cancers a subgroup of around 2% of breast tumors that are ERα-positive by mRNA expression analysis, but are of the basal molecular subtype. These tumors express significantly lower levels of both ERα and PR mRNA compared to luminal-type tumors and have almost invariably (94.5%) a high-risk MammaPrint profile. Furthermore, we show that these tumors have relatively high levels of the dominant negative ERΔ7 splice variant, in agreement with the notion that they may lack a functional response to estrogen and consequently may not respond to hormonal therapy.

RESULTS ERΔ7 splice variant expressed in an ERα-positive basal-type breast cancer We have recently developed an 80-gene signature (BluePrint™) that identifies the three major intrinsic subtypes (Basal, Luminal, and HER2) of breast cancer (Krijgsman et al., 2012). Of these 80 genes, 58 are used to identify the luminal subtype. Importantly, 32 out of these 58 luminal subtype reporter genes have ERαbinding sites adjacent to the transcription start site (TSS) (Krijgsman et al., 2012). This indicates that those genes that identify luminal-type breast cancer are significantly enriched for bona fide ERα target genes and suggests that the luminal subtype is characterized by tumors that have a functional ERα pathway. Conversely, BluePrint basal-type tumors would be expected to have either no significant ERα expression or a non-functional ERα pathway; these same bona fide ERα target genes show an inverse expression pattern in Basal-type tumors (Krijgsman et al., 2012). Following argumentation as outlined above, one would expect that breast tumors that are ERα positive but basal-type by BluePrint analysis, would either have a very low level of ERα protein or harbor a defective ERα protein. To test this hypothesis directly, we mined the Agendia database for patients who are ERα-positive by TargetPrint, but basal-type by BluePrint molecular subtype analysis. We initially identified a patient (patient 1, Table 1; 60-year-old woman with 9 mm, moderately differentiated, HER2 negative, ER/PR > 90% by IHC, invasive ductal carcinoma), who had undergone MammaPrint, TargetPrint, and BluePrint tests. She had

Estrogen receptor splice variants as a potential source of false-positive estrogen receptor status | 85

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

MammaPrint high-risk result, was ER/PR positive by TargetPrint, but basal subtype by BluePrint, suggesting that the ERα was present both at the protein (IHC > 90%) and mRNA levels, but that ERα target genes were not expressed in this tumor (hence basal-type). The tumor was also analyzed using the OncotypeDX™ breast cancer assay (Genomic Health Inc.), classifying the tumor as low risk for distant recurrence (Recurrence Score 8, Table 1). We used the same tumor mRNA sample as was used to perform the MammaPrint, TargetPrint, and BluePrint assays for detailed analysis of the ERα mRNA transcript in this patient. We first PCR amplified the coding sequence of ERα with specific oligonucleotides that span the start codon of ERα at the 5′ end and the stop codon at the 3′ end. Agarose gel electrophoresis of the PCR product revealed a smaller DNA fragment next to the expected DNA fragment coding for the open reading frame of ERα. Inspection of the DNA sequence of the smaller product revealed an ERαsequence lacking exon 7 of the coding sequence (data not shown). This transcript corresponds to the previously reported ERΔ7, an ERα splice variant that inhibits the function of wild type ERα in a dominant fashion (Garcia Pedrero et al., 2003).

86 | Chapter 4

Estrogen receptor splice variants as a potential source of false-positive estrogen receptor status | 87

T1bN0M0 pT1bN0Mx NA pT1cN0Mx NA pT1cN0Mx NA T1N0Mx NA pT2N0Mx T1cN0M0 pT2N0M0

1b 2 3 4 5 6 7 8 9 10 11 12

>90% 2+ NA 60-70% NA 80% NA negative NA negative positive 3+

IHC ERα >90% negative NA 40-50% NA <5% NA negative NA <5% positive 2-3+

IHC PR NAc negative NA negative NA negative NA negative NA negative negative negative

FISH HER2 0.33 0.18 0.26 0.41 0.25 0.03 0.04 0.15 0.03 0.01 0.28 0.55

TargetPrint ERα indexa 0.25 -0.16 0.22 0.16 -0.19 -0.35 -0.28 -0.28 -0.32 -0.24 0.01 0.21

TargetPrint PR indexa -0.77 -0.53 -0.52 -0.73 -0.39 -0.78 -0.51 -0.62 -0.57 -0.64 -0.59 -0.56

TargetPrint HER2 indexa

Oncotype Recurrence score 8 (low-risk) NA NA NA NA NA NA NA NA NA 31 (intermediate risk) NA

MammaPrint classification High-risk High-risk High-risk High-risk High-risk High-risk High-risk High-risk High-risk High-risk High-risk High-risk

Abbreviations: IHC, immunohistochemistry; ERα, estrogen receptor alpha; PR, progesterone receptor; FISH, Fluorescence In Situ Hybridization; NA, not available. a A TargetPrint index > 0.00 is considered as positive, a index ≤ 0.00 is considered as negative (described in the Patients and Methods section). b Patient in which we initially identified the ERΔ7 variant by cDNA sequencing as is described in the Results section. c FISH for HER2 not available, but tumor scored negative for HER2 by immunohistochemistry.

Stage

Pt.

Table 1. Characteristics of ERα-positive basal-type tumors for which the ERΔ7 expression was determined (N=12).

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

Frequency of ERα-positive basal-type tumors To determine the frequency at which ERα-positive basal-type breast tumors occur, we searched the Agendia database for additional cases. Out of 3,527 cases, we identified 2,658 ERα-positive, HER2-negative breast tumors, as judged by TargetPrint mRNA expression, for which BluePrint intrinsic subtyping data were available. From these 2,658 tumors, 2,603 (97.9%) were classified as luminal-type and 55 (2.1%) were classified as basal-type (Table 2). The mean ERα and PR TargetPrint indices for the ERα-positive basal-type tumors were significantly lower than for the ERα-positive luminal-type tumors (P < 0.0001; Table 2). Table 2. TargetPrint ERα/PR index, PR classification and MammaPrint classification of 2658 ERα-positive, HER2-negative tumors according to their BluePrint molecular subtype (Basal-type vs Luminal-type).

ERα index (Mean ± SD) PR index (Mean ± SD) PR classification PR positive

PR negative

MammaPrint classification Low-risk

High-risk

BluePrint classification Basal-type Luminal-type (n=55, 2.1%) (n=2603, 97.9%) 0.20 (± 0.15)

-0.04 (± 0.27) 24 (43.6%) 31 (56.4%) 3 (5.5%)

52 (94.5%)

0.57 (± 0.17)

0.28 (± 0.31)

2047 (78.6%) 556 (21.4%)

1434 (55.1%)

P-value < 0.0001a < 0.0001a

< 0.0001b

1169 (44.9%)

Abbreviations: ERα, estrogen receptor alpha; PR, progesterone receptor; SD, standard deviation. a Unpaired t-test, two-tailed b Fisher’s exact test, two-tailed

ERΔ7 splice variant expression in ERα-positive basal-type breast cancers We further analyzed an additional 11 of these 55 ERα-positive basal-type tumors for expression of total ERα as well as the ERΔ7 variant by qRT-PCR. The specificity of the primer pairs was tested with cDNA from MCF7 breast cancer cells overexpressing either wild-type ERα or ERΔ7 and the calculated ERΔ7/total ERα ratio was correlated with ERα protein expression in these cells. The ERα antibody clone 1D5 (Dako) was used for western blot analysis, for which the epitope is located in the N-terminal domain of ERα and therefore recognizes both wild-type ERα and ERΔ7. We show in these cells that the relative ERΔ7 levels as measured by qRT-PCR are highly concordant with protein expression (Fig. 2).

88 | Chapter 4

Figure 2 A pMX-control

70 kDa

50 kDa

pMX-ERα wild-type

pMX-ERΔ7

wild-type ERα ~ 67 kDa

ERΔ7 ~ 52 kDa

loading B

Figure 2. Sensitivity and specificity of the qRT-PCR primer pairs for the quantification of ERΔ7. (A) The ERα antibody clone 1D5 (Dako) was used for western blot analysis, for which the epitope is located in the N-terminal domain of ERα and therefore recognizes both wild-type ERα and ERΔ7 in MCF7 breast cancer cells with stable infection of pMX-vector control, pMXERα wild type, or pMX-ERΔ7. (B) ERα protein expression in these cells was correlated with the calculated ERΔ7/total ERα ratio by qRT-PCR using specific primer pairs for ERΔ7 and total ERα. Overexpression of wild-type ERα leads to a decrease of the ERΔ7/total ERα ratio, whereas overexpression of ERΔ7 increased the ERΔ7/total ERα ratio comparable to the ratio of protein expression.

The average total ERα mRNA expression by qRT-PCR was significantly lower for the 12 analyzed ERα-positive basal-type tumors compared to 15 randomly chosen ERα-positive luminal-type tumors (P = 0.002; Fig. 3A), consistent with the TargetPrint results (Table 2). There was no significant difference in average ERΔ7 mRNA expression between the ERα-positive basal-type and luminal-type samples

Estrogen receptor splice variants as a potential source of false-positive estrogen receptor status | 89

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

(P = 0.409; Fig. 3B). However, the relative ERΔ7 mRNA expression was significantly higher for the ERα-positive basal-type group compared to the ERα-positive luminaltype group (P < 0.0001; Fig. 3C), due to the lower overall ERα mRNA expression in the basal-type tumors. P = 0.002

P = 0.409

ERΔ

Figure 3 A

P < 0.0001

ERΔ

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

Figure 3. ERα-positive basal-type tumors have a relatively high ERΔ7 expression compared to ERα-positive luminal-type tumors. (A) Scatterplot of total ERα mRNA expression analysis by qRT-PCR in ER-positive basaltype (n=12) and ER-positive luminal-type (n=15) tumors. The qRT-PCR primers are located in exon 1 and exon 2. (B) Scatterplot of specific ERΔ7 mRNA expression analysis by qRT-PCR in ER-positive basal-type (n=12) and ER-positive luminal-type (n=15) tumors. The qRT-PCR Primers are located in exon 6 (forward) and over the exon 7 splice site (reverse). (C) Scatterplot of relative ERΔ7 expression calculated by dividing the ERΔ7 mRNA expression with the total ERα mRNA expression in ER-positive basal-type (n=12) and ER-positive luminal-type (n=15) tumors. Points indicate individual tumors; lines indicate mean with SD. P-values are calculated by unpaired T-test’s with Welch’s correction and are two-tailed.

90 | Chapter 4

The characteristics of the 12 ERα-positive basal-type tumors, for which ERΔ7 splice variant expression was determined, are shown in Table 1. For eight of the 12 patients, we were able to retrieve the ERα and PR IHC scoring. Based on the ERα IHC, six out of eight patients (75%) were classified as ERα-positive. In two patients, we found a discrepancy between TargetPrint and ERα IHC classification; in one of these patients, the TargetPrint ERα index was just above the ERα-positive threshold (patient 10). The PR IHC was in concordance with the PR classification based on TargetPrint in six of eight patients, and for two patients (patient 6 and 8), a small percentage of PR-positive cells was detected by IHC where the TargetPrint PR index was negative. The HER2-negative status was confirmed by fluorescence in situ hybridization (FISH) in all available cases. All patients (12/12) were stratified as high risk of distant recurrence by the MammaPrint prognostic gene signature. MammaPrint measures ERα function independent of ERα expression MammaPrint measures 70 genes that were selected from the entire complement of human genes, but ERα is not among the MammaPrint genes (van ‘t Veer et al., 2002). Nevertheless, we observed that 52 of the 55 (94.5%) ERα-positive basal-type tumors were MammaPrint high risk, while only 44.9% of the ERα-positive luminaltype tumors were classified as high risk of recurrence (P < 0.0001; Table 2). Since the MammaPrint assay identifies nearly all these ERα-positive basal-type tumors as high risk, it suggests that the test measures ERα activity independent of the ERα mRNA expression level itself. To investigate this further, we determined how many of the 70 MammaPrint prognosis genes are directly responsive to E2 treatment. For this, a publically available dataset was used that assessed gene expression changes after 10, 40, and 180 min of E2 treatment (Hah et al., 2011). We found that 16 MammaPrint reporter genes annotated in the most recent build of the human reference genome sequence are E2 regulated (Fig. 4A). Next, we tested whether these E2-responsive MammaPrint genes can be classified as direct ERα target genes. Using a publically available ChIP-seq dataset (Ross-Innes et al., 2012), the genome-wide chromatinbinding landscape of ERα in MCF7 cells was analyzed for the occurrence of an ERα binding event within 20,000 bp from the TSS of any of the MammaPrint genes. This window was chosen since most ERα-mediated gene regulation is found within this distance from a TSS (Fullwood et al., 2009). Ten out of 16 genes had an ERα binding event within 20,000 bp from the TSS (Fig. 4A), as exemplified for the LPCAT1 locus (Fig. 4B). Importantly, the essential ERα coactivators AIB1 (also known as SRC3) and p300 were also present at this specific binding site, indicating that ERα is likely to be functional here (Zwart et al., 2011). Furthermore, we confirmed that ERα binding events in E2-regulated MammaPrint genes are also found in two ER-positive luminal human breast tumor samples, for which ERα ChIP-seq data are available (Fig. 4A). Estrogen receptor splice variants as a potential source of false-positive estrogen receptor status | 91

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

E2 responsive E2 responsive

MammaPrint genes MammaPrint genes

FBXO31 PRC1 FBXO31 QSOX2 PRC1 SCUBE2 QSOX2 CDCA7 SCUBE2 DIAPH3 CDCA7 LPCAT1 DIAPH3 MCM6 LPCAT1 ALDH4A1 MCM6 BBC3 ALDH4A1 C9orf30 BBC3 GPR126 C9orf30 GSTM3 GPR126 NUSAP1 GSTM3 PECI NUSAP1 TGFB3 PECI TGFB3

tumor event tumor ER binding MCF7 #1 #2 tumor tumor MCF7 #1 #2

100 100 0 0

E2 affected genes E2 affected 40’ 160’ 10’ genes 10’

40’

160’

relative relative expression expression ( 2 log) ( 2 log)

ER binding event

tag count tag count

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

Figure 4 Figure 4 A A

B B ERα ERα

AIB1 AIB1

p300 p300

TSS TSS

Figure 4. Functional ERα target genes in MammaPrint 70-gene set. (A) Pie-chart, depicting the proportion of MammaPrint genes, which are affected by E2 treatment. Heatmap (right panel) depicts proximal ERα ChIP-seq signal by tag count (blue) as well as relative gene expression values as measured by GRO-seq, after 10, 40 and 160 minutes of E2 treatment (green-black-red heatmap). (B) Genome browser snapshot, depicting a shared binding site of ERα (red), AIB1 (green) and p300 (blue) proximal to the LPCAT1 transcription start site. Chromosome number, genomic coordinates and tag count are indicated. 92 | Chapter 4

In total, 12 out of 16 E2-regulated genes had an ERα-binding site in either MCF7 cells or in the two studied tumors (Fig. 4A). Cumulatively, these data indicate that bona fide ERα target genes are enriched in the MammaPrint gene signature, providing a plausible explanation for why the MammaPrint can measure ERα functionality rather than its mere presence, in contrast to other available assays.

DISCUSSION The present study identifies approximately 1 in 50 ER-positive breast cancer patients as basal molecular subtype. Basal-type breast tumors are characterized by an absence of expression of ERα target genes, which is generally thought to result from the absence of ERα expression (Nielsen et al., 2004). However, the group of tumors identified here is ERα positive on the mRNA level, suggesting that their basal phenotype is the result of a lack of ERα protein expression or a lack of functionality of the ERα protein present in these tumors. Indeed, we find that these tumors not only express relatively low levels of ERα mRNA but also express a splice variant of ERα missing exon 7 (ERΔ7; Fig. 3A and 3B). This ERα variant has been shown previously to act in a dominant negative fashion, meaning that this variant can inhibit the function of the wild-type ERα protein when co-expressed in the same cell (Garcia Pedrero et al., 2003). We note that the absolute levels of ERΔ7 are comparable in ERαpositive basal-type versus ERα-positive luminal-type tumors, but that the relative abundance of ERΔ7 is higher in the ERα-positive basal-type tumors (Fig. 3C). We interpret these data as follows: When the levels of wild-type ERα in a breast tumor are high, the inhibitory effects of dominant negative ERΔ7 are by comparison minor, leaving the cell with considerable ERα activity and thus with a luminal phenotype (Fig. 5, right). In contrast, lower levels of wild-type ERα in the weakly ERα-positive breast tumors are inhibited to a greater extent by the presence of ERΔ7, leaving the tumor cells with insufficient ERα activity to regulate ERα target gene expression and thus with a basal phenotype (Fig. 5, left). It remains to be explained why lower levels of ERα result in a relative increase in abundance of the ERΔ7 splice variant. It is possible that ERα also controls the expression of certain components of the splicing machinery and that low ERα activity therefore results in a different processing of the ERα (and potentially also other) precursor mRNAs.

Estrogen receptor splice variants as a potential source of false-positive estrogen receptor status | 93

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

Luminal-type

Dominant-negative effect ERΔ7 variant

Basal-type

ERα activity

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

Figure 5

Tumor 1 Low wild-type ERα mRNA expression

Tumor 2 High wild-type ERα mRNA expression

Figure 5. Proposed model by which ERΔ7 mRNA expression can affect ERα activity in low ERα wild-type expressing tumors (left) and in high ERα wild-type expressing tumors (right).

A clinically relevant question is whether this identified group of ERα-positive basaltype tumors is likely to respond to hormonal therapy. The finding that ERα target genes are not expressed suggests that the mitogenic responses in such tumors are not driven by E2 and that such tumors would be unlikely to derive significant benefit from hormonal therapy. It was reported by Ellis et al. (Ellis et al., 2011) in a cohort of postmenopausal women with clinical stage II to III ER-positive breast cancer that the single patient in their study with a basal-like intrinsic subtype was resistant to endocrine therapy. While it remains to be formally proven, there are other suggestions in the literature that the presence of ERΔ7 is associated with a lack of response to tamoxifen. Van Dijk (van Dijk, 2001) analyzed the relative ERΔ7 mRNA expression in a group of 21 primary breast tumors from postmenopausal early breast cancer patients treated with adjuvant tamoxifen. It was found that out of eleven ERα mRNA variants tested, only the ERΔ7 mRNA was significantly differentially expressed between primary breast tumors of patients who developed a tumor recurrence (13/21) and tumors of patients without recurrence (8/21). Tumors from patients with a recurrence expressed on average 24% ERΔ7 mRNA (relative to wild-type ERα mRNA expression), while tumors from patients without recurrence expressed on average 9% ERΔ7 mRNA (van Dijk, 2001). While it may be premature to withhold hormonal therapy from this group of ERα-positive breast

94 | Chapter 4

cancer patients, as this would require a large randomized outcome study, there are reasons to consider adding chemotherapy to the treatment regimen for these patients. We find that 94.5% of the ERα-positive basal-type breast cancer patients are high risk by the MammaPrint assay, making them potential candidates to benefit from chemotherapy based on their high recurrence risk. Moreover, basal-type breast cancers have been shown to be significantly more responsive to neoadjuvant chemotherapy as compared to luminal breast cancers, again indicating that addition of chemotherapy could be effective in this patient group (Krijgsman et al., 2012). The St. Gallen consensus guidelines state that patients with an (borderline) ERαpositive basal-type tumor are classified as incompletely endocrine responsive (Goldhirsch et al., 2005). This relative lack of endocrine responsiveness together with a designation of “high risk” of relapse by MammaPrint will contribute to a clinician’s recommendation of whether endocrine therapy alone may be sufficient or supplementary chemotherapy may be beneficial for these patients. Our finding that ERα-positive basal-type tumors are in general borderline ERα positive on mRNA level is in agreement with the conclusions of Iwamoto et al. who found that most of the 1–9% IHC ERα-positive tumors show molecular features similar to ERα-negative basal-like tumors (Iwamoto et al., 2012). The strength of our study is the high number of cases and therefore the better estimate we can make of the frequency of ERα-positive basal-type tumors. In addition, we show that a majority of these tumors have a high-risk prognostic profile. One limitation of our study is that we do not have all the clinical information for the entire group of patients that was studied here. For instance, we did not have access to the IHC data for all the patients in this study and had to rely on TargetPrint to assess ERα levels. However, IHC data were available for eight of the twelve ERα-positive basal-type tumors for which ERΔ7 expression was determined (Table 1) and showed that six out of eight tumors scored clearly positive for ERα protein by IHC. ERα-positive breast tumors have in general a better prognosis than ERα-negative tumors (McGuire, 1991). In spite of this, the group of ERα-positive basal-type breast tumors consists nearly exclusively of high-risk patients as judged by the MammaPrint assay (Table 2). Our present data also provide a possible explanation for this finding. In contrast to the OncotypeDX™ prognostic signature, the 70-gene MammaPrint™ signature does not include ERα (Paik et al., 2004; van ‘t Veer et al., 2002). We find that 16 MammaPrint genes are responsive to E2 treatment and that 12 of these are classified as direct ERα targets based on ERα/DNA associations in close proximity to the TSS, indicating that MammaPrint determines ERα activity rather than merely its expression. We believe that this likely explains why the first patient (patient 1, Table 1) having the ERα-positive basal phenotype was characterized by

Estrogen receptor splice variants as a potential source of false-positive estrogen receptor status | 95

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

the OncotypeDX assay as “low risk”, but “high risk” by MammaPrint and patient 11 also had a discordant risk assessment in these two assays (Table 1). The ERα mRNA is expressed at a relatively high level in these patients, which is a “good prognosis” factor in the OncotypeDX assay. However, MammaPrint identified this tumor as lacking a functional ERα and came to a “high risk” reading. In conclusion, by combining TargetPrint and BluePrint molecular subtyping analysis, we have identified a subgroup of some 2% of breast cancer patients who lack ERα function while expressing ERα at the mRNA and protein level. Our data indicate that such patients are frequently at high recurrence risk and may benefit from adjuvant chemotherapy.

PATIENTS AND METHODS Patient samples and molecular profiling A total of 3,527 breast cancer patient specimens were retrospectively analyzed. This selection was based on the availability of MammaPrint™, TargetPrint™, and BluePrint™ molecular profiling results as performed in the Agendia testing laboratories. The ERα status on mRNA levels was determined by TargetPrint, a microarray-based gene expression test, which offers a quantitative assessment of the patient’s level of ERα, PR, and HER2 expression (Roepman et al., 2009). The TargetPrint probe for ERα mRNA detection is located in the 3′ UTR region. The ERα, PR, and HER2 TargetPrint score is a value between −1 and 1, where the null cutoff value is calibrated to 1% IHC ERα-positive cells, as identified in a reference laboratory according to ASCO/CAP guidelines. Tumors are reported ERα or PR positive when the TargetPrint score is above 0, corresponding to >1% IHC-positive cells (Roepman et al., 2009). Molecular subtyping was performed using the 80-gene BluePrint™ molecular subtyping profile for the classification of breast cancer into basal-type, luminal-type, and ERBB2-type (HER2-positive) molecular subclasses (Krijgsman et al., 2012). In addition, the tumors were classified as low risk or high risk for distant recurrence using the 70-gene MammaPrint™ signature, a FDA-cleared breast cancer recurrence assay, performed by Agendia Inc. (Glas et al., 2006). ERΔ7 variant analysis We obtained RNA from 15 ERα-positive luminal-type tumors and from 12 ERαpositive basal-type tumors to analyze the relative ERΔ7 mRNA expression. cDNA was synthesized from 500ng RNA using SuperScript II Reverse Transcriptase (Invitrogen) with random hexamer primers. The total ERα and ERΔ7 mRNA

96 | Chapter 4

expression was determined by qRT-PCR. For total ERα expression, the forward primer was located in exon 1 and the reverse primer in exon 2. For ERΔ7 expression, the forward primer was located in exon 6 and the reverse primer was designed to specifically detect ERΔ7 and located partially in exon 6 (12 nucleotides) and partially in exon 8 (14 nucleotides). All qRT-PCR reactions were performed in duplicates using SYBR Green reaction mix containing 5µl cDNA. The expression levels were quantified using a reference standard dilution curve. The relative expression of the ERΔ7 variant was calculated by dividing the ERΔ7 mRNA expression by the total ERα mRNA expression. PCR primers The following primer sequences were used to PCR the coding sequence of ERα: Forward, 5’-CCGGTTTCTGAGCCTTCTG-3’, Reverse, 5’-TGGTGCATGATGAGGGTAAAA-3’. For qRT-PCR the following primer sequences were used to amplify total ER: Forward (exon 1), 5’-CAGCTGCCCTACTACCTGGAGAA-3’; Reverse (exon 2), 5’-CCCTGGCGTCGATTATCTGAATTTGG-3’. For amplification of specifically ERΔ7 the following primers were used in qRT-PCR reaction: Forward (exon 6), 5’-TGCTGGCTACATCATCTCGGTT-3’; Reverse (exon 6 [12nt] and exon 8 [14nt]), 5’-CCATGCCTTTGTTACAGAATTAAGCA-3’. Identification of ERα target genes in the 70-gene MammaPrint™ breast cancer signature The 70 MammaPrint genes were analyzed for ERα binding events within 20kb from the transcription start site, representing the most commonly detected window for ERmediated gene regulation (Fullwood et al., 2009). ERα-binding sites were identified by ChIP-seq analyses (Schmidt et al., 2009), using available datasets for the luminal breast cancer cell line MCF-7 (Ross-Innes et al., 2012) and two ER-positive luminal breast tumor samples (paper in submission; GSE40867). Publically available data on E2-stimulated gene expression were used from (Hah et al., 2011), where Global RunOn sequencing was applied to assess gene transcription after 0-, 10-, 40-, and 160min E2 treatment. Only genes with a differential expression as compared to control conditions with a false discovery rate of ~0.1% were considered as E2-regulated. Statistical analysis The two-tailed unpaired T-test with Welch’s correction for unequal variances was used to determine statistical significance of differences in ERα and PR TargetPrint indexes across the basal-type and luminal-type groups. This test was also used to

Estrogen receptor splice variants as a potential source of false-positive estrogen receptor status | 97

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

compare the groups for total ERα and ERΔ7 expression. Fisher’s exact test (two-tailed) was used to test for significance of the differences in PR levels and MammaPrint classification. All tests were performed using GraphPad Prism 6 software. P-values <0.05 were considered as statistically significant.

ACKNOWLEDGEMENTS We are grateful to Dr. Allen (Morton Plant Hospital), Dr. Chung (Joyne Wayne Cancer Institute Breast Center), Dr. Cox (Tampa Bay Breast Care Specialists), Dr. Greif (Alta Bates Summit Medical Center), Dr. Hunter (Comanche Country Hospital), Dr. Sachedina (Central Florida Breast Center), Dr. Schwartz (Thomas Jefferson University Hospital), Dr. Sinha (Rockwood Clinic), Dr. Terschluse (SSM Depaul Health Center), Dr. Weintritt (Surgical Specialists of Northern Virginia), Dr. West (The Breast Care & Imaging Center of Orange Country), and Dr. Yao (NorthShore Hospital) for assisting in obtaining informed consent and providing clinical information. The authors thank Femke de Snoo and Lisette Stork (Agendia) for administrative support and critical reading of the manuscript. We thank Neil Barth (Agendia) for helpful suggestions on the manuscript. This work was supported by a Grant from the European Research Council (to René Bernards).

CONFLICT OF INTEREST Arno Floore and René Bernards are employees of Agendia NV.

98 | Chapter 4

REFERENCES Carroll, J.S., Meyer, C.A., Song, J., et al., 2006. Genome-wide analysis of estrogen receptor binding sites. Nat Genet 38, 1289-1297. Chia, S.K., Bramwell, V.H., Tu, D., et al., 2012. A 50-gene intrinsic subtype classifier for prognosis and prediction of benefit from adjuvant tamoxifen. Clin Cancer Res 18, 4465-4472. Davies, C., Godwin, J., Gray, R., et al., 2011. Relevance of breast cancer hormone receptors and other factors to the efficacy of adjuvant tamoxifen: patient-level meta-analysis of randomised trials. Lancet 378, 771-784. Dowsett, M., Cuzick, J., Ingle, J., et al., 2010. Meta-analysis of breast cancer outcomes in adjuvant trials of aromatase inhibitors versus tamoxifen. J Clin Oncol 28, 509-518. Ellis, M.J., Suman, V.J., Hoog, J., et al., 2011. Randomized phase II neoadjuvant comparison between letrozole, anastrozole, and exemestane for postmenopausal women with estrogen receptor-rich stage 2 to 3 breast cancer: clinical and biomarker outcomes and predictive value of the baseline PAM50-based intrinsic subtype-ACOSOG Z1031. J Clin Oncol 29, 2342-2349. Fullwood, M.J., Liu, M.H., Pan, Y.F., et al., 2009. An oestrogen-receptor-alpha-bound human chromatin interactome. Nature 462, 58-64. Fuqua, S.A., Fitzgerald, S.D., Allred, D.C., et al., 1992. Inhibition of estrogen receptor action by a naturally occurring variant in human breast tumors. Cancer Res 52, 483-486. Fuqua, S.A., Fitzgerald, S.D., Chamness, G.C., et al., 1991. Variant human breast tumor estrogen receptor with constitutive transcriptional activity. Cancer Res 51, 105-109. Garcia Pedrero, J.M., Zuazua, P., MartinezCampa, C., et al., 2003. The naturally occurring variant of estrogen receptor (ER) ERDeltaE7 suppresses estrogen-dependent transcriptional activation by both wild-type ERalpha and ERbeta. Endocrinology 144, 2967-2976. Glas, A.M., Floore, A., Delahaye, L.J., et al., 2006. Converting a breast cancer microarray

signature into a high-throughput diagnostic test. BMC Genomics 7, 278. Goldhirsch, A., Glick, J.H., Gelber, R.D., et al., 2005. Meeting highlights: international expert consensus on the primary therapy of early breast cancer 2005. Ann Oncol 16, 1569-1583. Hah, N., Danko, C.G., Core, L., et al., 2011. A rapid, extensive, and transient transcriptional response to estrogen signaling in breast cancer cells. Cell 145, 622-634. Hammond, M.E., Hayes, D.F., Dowsett, M., et al., 2010. American Society of Clinical Oncology/ College of American Pathologists guideline recommendations for immunohistochemical testing of estrogen and progesterone receptors in breast cancer. Arch Pathol Lab Med 134, 907922. Harvey, J.M., Clark, G.M., Osborne, C.K., et al., 1999. Estrogen receptor status by immunohistochemistry is superior to the ligand-binding assay for predicting response to adjuvant endocrine therapy in breast cancer. J Clin Oncol 17, 1474-1481. Herynk, M.H., Fuqua, S.A., 2004. Estrogen receptor mutations in human disease. Endocr Rev 25, 869-898. Iwamoto, T., Booser, D., Valero, V., et al., 2012. Estrogen receptor (ER) mRNA and ER-related gene expression in breast cancers that are 1% to 10% ER-positive by immunohistochemistry. J Clin Oncol 30, 729-734. Krijgsman, O., Roepman, P., Zwart, W., et al., 2012. A diagnostic gene profile for molecular subtyping of breast cancer associated with treatment response. Breast Cancer Res Treat 133, 37-47. McGuire, W.L., 1991. Breast cancer prognostic factors: evaluation guidelines. J Natl Cancer Inst 83, 154-155. McGuire, W.L., Chamness, G.C., Fuqua, S.A., 1991. Estrogen receptor variants in clinical breast cancer. Mol Endocrinol 5, 1571-1577.

Estrogen receptor splice variants as a potential source of false-positive estrogen receptor status | 99

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

Nielsen, T.O., Hsu, F.D., Jensen, K., et al., 2004. Immunohistochemical and clinical characterization of the basal-like subtype of invasive breast carcinoma. Clin Cancer Res 10, 5367-5374. Paik, S., Shak, S., Tang, G., et al., 2004. A multigene assay to predict recurrence of tamoxifen-treated, node-negative breast cancer. N Engl J Med 351, 2817-2826. Perou, C.M., Sorlie, T., Eisen, M.B., et al., 2000. Molecular portraits of human breast tumours. Nature 406, 747-752. Roepman, P., Horlings, H.M., Krijgsman, O., et al., 2009. Microarray-based determination of estrogen receptor, progesterone receptor, and HER2 receptor status in breast cancer. Clin Cancer Res 15, 7003-7011. Ross-Innes, C.S., Stark, R., Teschendorff, A.E., et al., 2012. Differential oestrogen receptor binding is associated with clinical outcome in breast cancer. Nature 481, 389-393. Schmidt, D., Wilson, M.D., Spyrou, C., et al., 2009. ChIP-seq: using high-throughput sequencing to

100 | Chapter 4

discover protein-DNA interactions. Methods 48, 240-248. Sommer, S., Fuqua, S.A., 2001. Estrogen receptor and breast cancer. Semin Cancer Biol 11, 339352. van â&#x20AC;&#x2DC;t Veer, L.J., Dai, H., van de Vijver, M.J., et al., 2002. Gene expression profiling predicts clinical outcome of breast cancer. Nature 415, 530-536. van Dijk, M., 2001. Functional activity of human estrogen receptor alpha in normal breast tissue and breast cancer. Dissertation, Free University, Amsterdam, The Netherlands. Zhang, Q.X., Borg, A., Fuqua, S.A., 1993. An exon 5 deletion variant of the estrogen receptor frequently coexpressed with wild-type estrogen receptor in human breast cancer. Cancer Res 53, 5882-5884. Zwart, W., Theodorou, V., Kok, M., et al., 2011. Oestrogen receptor-co-factor-chromatin specificity in the transcriptional regulation of breast cancer. EMBO J 30, 4764-4776.

Estrogen receptor splice variants as a potential source of false-positive estrogen receptor status | 101

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

Chapter 5 ERBB2 mutations characterize a subgroup of muscleinvasive bladder cancers with complete response to neoadjuvant chemotherapy

Floris H. Groenendijk1, Jeroen de Jong2, Elisabeth E. Fransen van de Putte3, Magali Michaut1, Andreas Schlicker1, Dennis Peters4, Arno Velds5, Marja Nieuwland5, Michel M. van den Heuvel6, Ron M. Kerkhoven5, Lodewijk F. Wessels1, Annegien Broeks4, Bas W.G. van Rhijn3, RenĂŠ Bernards1 and Michiel S. van der Heijden1,7

Division of Molecular Carcinogenesis, Cancer Genomics Netherlands; 2 Department of Pathology; 3

Department of Urology; 4 Core Facility for Molecular Pathology & Biobanking, Division of Molecular Pathology; 5 Genomics Core Facility; 6 Division of Thoracic Oncology; 7 Department of Medical Oncology, The Netherlands Cancer Institute, Antoni van Leeuwenhoek Hospital, Plesmanlaan 121, 1066 CX Amsterdam, The Netherlands.

Submitted

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

ABSTRACT Pathologic complete response to neoadjuvant platinum-containing chemotherapy (NAC) is a strong prognostic determinant for patients with muscle-invasive bladder cancer (MIBC). Despite comprehensive molecular characterization of bladder cancer, associations of molecular alterations with treatment response are still largely unknown. We selected pathologic complete responders (ypT0N0; n=40) and nonresponders (ypT>2; n=33) from a cohort of 110 high-grade MIBC patients treated with neoadjuvant chemotherapy. DNA was isolated from pre-chemotherapy tumor tissue and used for next-generation sequencing of 178 cancer-associated genes (discovery cohort) or targeted sequencing (validation cohort). We found that ten out of 40 complete responders had ERBB2 missense mutations, whereas none of 33 non-responders had ERBB2 mutations (p = 0.002). ERBB2 missense mutations in responders were mostly confirmed activating mutations. ERCC2 missense mutations, recently found associated with response to NAC, were enriched in responders, however this association did not reach statistical significance in our cohort. We conclude that ERBB2 missense mutations characterize a subgroup of muscle-invasive bladder cancer patients with complete response to NAC.

KEYWORDS ERBB2  ERCC2  Neoadjuvant chemotherapy  Muscle-invasive bladder cancer  Response

STATEMENT OF CLINICAL SIGNIFICANCE Bladder cancer contains the highest frequency of ERBB2 missense mutations among all cancer types. We discovered ERBB2 missense mutation as a novel genomic biomarker of complete response to neoadjuvant platinum-containing chemotherapy. In addition, we showed that the presence of ERCC2 mutations does not always confer sensitivity to platinum-based chemotherapy.

104 | Chapter 5

INTRODUCTION Bladder cancer can be classified in non-muscle-invasive bladder cancer and muscleinvasive bladder cancer (MIBC) on the basis of invasion into the muscularis propria of the bladder wall. This classification has significant implications for prognosis, treatment and follow-up. The standard-of-care treatment for patients with MIBC is radical cystectomy including extended pelvic lymph node dissection. Neoadjuvant platinum-containing chemotherapy (NAC) is often used and improves overall survival in patients with MIBC compared to cystectomy alone (Advanced Bladder Cancer (ABC) Meta-analysis Collaboration 2005; Meeks et al., 2012). Patients with a pathologic complete response (pCR) to NAC have a superior clinical outcome compared to patients with residual disease after NAC (Grossman et al., 2003; Meijer et al., 2013; Sonpavde et al., 2009). However, no molecular markers or baseline clinical characteristics that can predict the response to NAC are clinically validated. Identification of clinical and molecular factors associated with response to NAC is required to reduce the number of patients needed to treat and improve the clinical acceptance of NAC. Despite comprehensive molecular characterization of bladder cancer in the past few years (Guo et al., 2013; Iyer et al., 2013; The Cancer Genome Atlas Network, 2014), associations of molecular alterations with clinical outcome and treatment response are still largely unknown. Recently, Van Allen et al. reported that missense mutations in ERCC2, a nucleotide excision repair gene, were selectively present in nine out of 25 MIBC patients with complete response to cisplatin-containing NAC, while ERCC2 missense mutations were absent in 25 nonresponders (Van Allen et al., 2014). We have used next-generation sequencing (NGS) of a panel of 178 cancer-associated genes on a series of MIBCs to find novel biomarkers of response to neoadjuvant platinum-containing chemotherapy. We found an unexpected association between mutations in the ERBB2 gene (also known as HER2) and complete response to chemotherapy.

RESULTS Patient characteristics To search for biomarkers of response to neoadjuvant platinum-containing chemotherapy, we collected pre- and post-chemotherapy specimens from 110 prospectively registered MIBC patients treated with NAC. From this cohort, we identified 40 complete responders and 33 non-responders to NAC and isolated DNA

ERBB2-mutant MIBC and response to neoadjuvant chemotherapy | 105

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

from the pre-treatment transurethral resection (TUR) specimens of these patients (Methods). Detailed clinical characteristics are listed in Table 1. No significant differences in baseline clinical characteristics were identified between responders and non-responders. As expected, responders to NAC had a superior recurrencefree survival (Supplementary Fig. S1A) and cancer-specific survival (Supplementary Fig. S1B) compared to non-responders (P < 0.001). Table 1. Clinical characteristics stratified by response to neoadjuvant chemotherapy Responders (N=40) Median age, years (IQR)

Non-responders P-value1 (N=33)

59 (14.5)

63 (15.6)

0.94

7 (17.5)

8 (24.2)

0.48

12 (30.0) 28 (70.0)

8 (24.2) 25 (75.8)

0.61

N0 N1 N2 N3

16 7 13 4

11 10 9 3

(33.3) (30.3) (27.9) (9.1)

0.64

M0 M12

36 (90.0) 4 (10.0)

31 (93.9) 2 (6.1)

0.68

TUR histology, N (%) Urothelial cell Other

39 (97.5) 1 (2.5)

32 (97.0) 1 (3.0)

Neoadjuvant chemotherapy regimen, N (%) MVAC Gemcitabine/cisplatin Gemcitabine/carboplatin

25 (62.5) 9 (22.5) 6 (15.0)

18 (54.5) 7 (21.2) 8 (24.4)

0.65

4 (1.0)

4 (0.0)

0.98

10 (25.0)

26 (78.8)

< 0.001

Female sex, N (%) Clinical TNM staging, N (%) T2 Tâ&#x2030;Ľ3

Median no. of chemotherapy cycles (IQR) Patients with recurrence, N (%) Mean length of follow-up, years (95%CI)

(40.0) (17.5) (32.5) (10.0)

13.2 (10.4-16.0)

1.9 (1.1-2.7)

1 Age and number of chemotherapy cycles were compared using the Mann-Whitney U test. Gender, clinical staging, histology, neoadjuvant chemotherapy regimen and recurrence were compared using the Fisherâ&#x20AC;&#x2122;s exact test. 2 M1: nodal metastases above the common iliac arteries, but below the diaphragm Abbreviations: IQR, interquartile range; TUR, transurethral resection; MVAC, methotrexate, vinblastine, doxorubicin and cisplatin.

106 | Chapter 5

Target enrichment NGS of 178 cancer-associated genes Target enrichment NGS for mutations in 178 cancer-associated genes was performed on pre-treatment tumor DNA from 16 responders and 16 non-responders (“discovery cohort”). The mean target coverage was 200X. Two hundred fifty-nine variants, predicted to be somatic, were called in 144 of the 178 genes. We performed a contrasting analysis to identify genes with a different mutation frequency in responders compared to non-responders (Fig. 1A). ERBB2 and ATR were found exclusively mutated in tumor samples from responders (Fig. 1B). ERBB2 was the only gene significantly enriched for mutations in responders (P = 0.04; Fisher’s exact test; Fig. 1B), although this association did not reach statistical significance after correction for multiple testing. The five ERBB2 missense mutations that we identified were S310F (n=3), V777M and V842I and have all been identified previously by multiple TCGA studies (http://cbioportal.org). We found six missense mutations in ERCC2 in our discovery cohort. These six ERCC2 mutations were present in four responders and in two non-responders (Fig. 1B). All ERCC2 mutations were identified previously in other studies and were confirmed to be somatic mutations by sequencing matched germline DNA isolated from peripheral blood. The model proposed by Van Allen et al. (Van Allen et al., 2014) predicts that the ERCC2-mutant clones in the two non-responding tumors would be selectively eliminated by the NAC, whereas ERCC2-wild type clones would survive. To investigate this possibility, we isolated DNA from the postchemotherapy resistant tumors of the two ERCC2 mutant non-responders. Sanger sequencing demonstrated in both cases that the ERCC2 missense mutation was still present in the post-chemotherapy resistant tumor, indicating that the mutation was not counter-selected during chemotherapy (Fig. 2A; Supplementary Fig. S2A). This result is not consistent with an increased sensitivity to NAC of ERCC2-mutant tumor cells.

ERBB2-mutant MIBC and response to neoadjuvant chemotherapy | 107

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

1.5 ERBB2

Significance (-log10 P-value)

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

Figure 1 A

ATR 1.0

CDKN2A

MLL3 MLL2

0.5

MLL TP53 PTEN ERCC2 RB1

0.0 10

0 Non-responders Responders Effect (log-odds-ratio)

B Non-responders

Responders

ERBB2 ATR TGFBR2 KDM6A ERCC2 RB1 PIK3CA ATM ARID1A ERBB3 TP53 AR CDKN2A 14 12 10 8 6 4 2 no. of mutated samples

2 4 6 8 10 12 14 no. of mutated samples

Figure 1. Gene enrichment analysis of mutated genes in responders and non-responders. (A) volcano plot of effect size (log-odds-ratio) and significance (-log10 P-value) of the 25 genes mutated in more than two samples. Mutated genes enriched in responders are labeled green and mutated genes enriched in non-responders are labeled red. (B) pyramid plot showing the number of mutated samples in 16 responders in green and the number of mutated samples in 16 non-responders in red (*, P < 0.05). 108 | Chapter 5

Figure 2 A

Germline (serum)

L87

E86Q

E85

Pre-treatment TUR

NH2

653

675

Extracellular domain

COOH

TKD

TM 22

6 V7 9H 77 M V8 42 I

Y 10 S3

Post-treatment cystectomy

720

976

1255

Intracellular domain

Identified by target enrichment NGS Identified by Sanger sequencing

Figure 2. (A) sequencing results from an ERCC2-mutant non-responder (patientID CF1000) showing the ERCC2 missense mutation (E86Q) in DNA isolated from the pre-treatment TUR and post-treatment cystectomy, but not in the germline DNA. This demonstrates that this somatic ERCC2 mutation was not counter-selected during chemotherapy. (B) plot showing the distribution of ERBB2 missense mutations identified in this study by target enrichment NGS (dark blue circles) or Sanger sequencing (orange circles). ERBB2 missense mutations cluster at the S310 position in the extracellular domain and in the tyrosine kinase domain (TKD). TM, transmembrane domain. (C) graph showing that ERBB2 missense mutations are significantly enriched in responders and significantly depleted in non-responders to NAC, compared to the unselected TCGA urothelial bladder cancer cohort (*, P < 0.05). ERBB2-mutant MIBC and response to neoadjuvant chemotherapy | 109

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

Validation of the association of ERBB2 missense mutations with platinum-therapy response in the validation cohort We used the recently published MIBC dataset of Van Allen et al. (exome sequencing in 25 responders and 25 non-responders to neoadjuvant chemotherapy), to test the association between ERBB2 missense mutations and responsiveness to platinumcontaining chemotherapy as an initial validation cohort for our finding (Van Allen et al., 2014). We found that in this cohort, all three patients with an ERBB2 missense mutation responded to neoadjuvant chemotherapy (Van Allen et al., 2014). A fourth complete responder had the ERBB2 S310F hotspot mutation with an allelic fraction <0.10. Based on these findings, we tested the association between ERBB2 and chemotherapy sensitivity in a further validation cohort consisting of the remaining 24 responders and 17 non-responders to NAC in our patient series. Using Sanger sequencing, we identified another six ERBB2 missense mutations in five responders and none in the non-responders from this validation cohort. Taken together, we identified in this study ERBB2 missense mutations in ten of the 40 complete responders and in none of the 33 non-responders to NAC (P = 0.002; Fig. 3). All identified somatic ERBB2 missense mutations are reported in Supplementary Table S1. The distribution of ERBB2 missense mutations across the gene is shown in Figure 2B. Next, we compared the ERBB2 mutation frequency in our responder cohort and the responder cohort of Van Allen et al. to the ERBB2 missense mutation frequency in the unselected TCGA bladder cancer cohort (8.4%). ERBB2 missense mutations were significantly enriched in the chemotherapy responders compared to the unselected TCGA cohort (P = 0.01; Fig. 2C). Conversely, we found that ERBB2 missense mutations were significantly depleted in the non-responder cohorts compared to the unselected TCGA cohort (P = 0.02; Fig. 2C).

110 | Chapter 5

ERBB2-mutant MIBC and response to neoadjuvant chemotherapy | 111

Figure 3. OncoPrint showing ERBB2 missense mutations, ERBB2 amplifications and ERCC2 missense mutations in the 40 responders and 33 non-responders to NAC in this study. Individual patients are represented as columns.

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

Association of ERBB2 amplification with platinum-response Having found an association between ERBB2 missense mutations and platinumresponse, we next tested the association of ERBB2 amplification with platinumresponse. ERBB2 amplification was determined by silver-enhanced in situ hybridization (SISH, assessable for 56 patients) and/or by low-pass whole genome sequencing (low-pass WGS, available for 46 patients). ERBB2 amplification status was not available for two responders and three non-responders. Eight tumors samples from 38 responders were ERBB2 amplified, whereas four tumors from 30 non-responders had ERBB2 amplifications, resulting in an ERBB2 amplification frequency of 21% in the responders and 13% in the non-responders (P = 0.53; Fig. 3). All amplifications identified by low-pass WGS were confirmed by SISH. All ERBB2 amplified tumors showed overexpression of ERBB2 protein by immunohistochemistry (score 2-3+), whereas non-amplified ERBB2-mutant tumors did not overexpress ERBB2 protein. We found ERBB2 amplification in four ERBB2-mutant tumors with a complete response to NAC. In all of these cases, the ERBB2-mutant allele was found amplified. For two patients, where target enrichment NGS was available, the fraction of mutant reads was >0.5 with the percentage of tumor cells estimated as 60-80% (Supplementary Table S1). For the other two patients, where Sanger sequencing detected the ERBB2 mutation, we found that the peak from the mutant nucleotide was higher than the peak from the reference nucleotide (Supplementary Fig. S2B and S2C). Association of ERCC2 missense mutations with platinum-therapy response in the validation cohort To test the association between ERCC2 missense mutations and platinum response, we sequenced in our validation cohort the exons around the helicase domains of ERCC2, where all missense mutations cluster. We identified ERCC2 missense mutations in three tumor samples from 24 responders and in none of the 17 nonresponders. This brings the total number of somatic ERCC2 missense mutations identified in this study to seven in 40 responders (18%) and two in 33 non-responders (6%) (P = 0.17; Fig. 3). Six of the seven ERCC2 mutations in responders were present in patients with wild type, non-amplified ERBB2 (Fig. 3). All identified somatic ERCC2 missense mutations are reported in Supplementary Table S2.

112 | Chapter 5

DISCUSSION Bladder cancer contains the highest frequency of ERBB2 missense mutations (~10%) among all cancer types (http://cbioportal.org; accessed Sept 22, 2014). In this study, we identified ten ERBB2 missense mutations in 40 complete responders, but none in 33 non-responders to NAC. Our findings demonstrate that ERBB2 missense mutations can be used to select complete responders to neoadjuvant chemotherapy. Furthermore, it suggests that HER2-directed therapies for ERBB2 mutant bladder cancers are unlikely to replace chemotherapy in the neoadjuvant setting, since these tumors have highly favorable responses to neoadjuvant platinum-containing chemotherapy. ERBB2 is a member of the ERBB/HER receptor tyrosine kinase family; further comprising EGFR (ERBB1/HER1), ERBB3/HER3 and ERBB4/HER4. The ERBB proteins transmit proliferation and cell survival signals upon binding of a ligand to the extracellular domains of the receptor. ERBB2 is unique among its family members for two reasons: (1) no direct high-affinity extracellular ligand has been identified for the ERBB2 receptor and (2) the ERBB2 receptor is permanently fixed in an open and active conformation, and therefore the preferred dimerization partner for other ERBB family members (reviewed in (Herter-Sprie et al., 2013)). Mutational activation of ERBB2 can result from missense mutations in the extracellular domain (e.g. S310F) or mutations in the tyrosine kinase domain. Previous functional studies on ERBB2 mutations have shown that the S310F, S310Y, D769H and V842I variants identified here are activating mutations that support cellular transformation (Bose et al., 2013; Greulich et al., 2012). The R678L and V777M mutations have, to our knowledge, not been functionally characterized. However, a different amino acid substitution at the V777 same position was found to be activating (Bose et al., 2013). Six of the ten identified ERBB2 missense mutations cluster at amino acid 310 in the extracellular domain II. ERBB2 S310F and S310Y mutations behave similarly to ERBB2 kinase domain mutants. They cause elevated C-terminal tail phosphorylation without evidence of covalent dimerization (Greulich et al., 2012). The S310 position is also a mutational hotspot in the TCGA urothelial bladder cancer cohort: ~40% of all ERBB2 missense mutations cluster at this position. Of note, the S310 position is not covered in the frequently used TruSeq Amplicon Cancer-Panel (Illumina) for somatic mutation detection. This demonstrates the advantage of using target enrichment or whole-exome NGS for somatic mutation detection in the clinical setting.

ERBB2-mutant MIBC and response to neoadjuvant chemotherapy | 113

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

The strong association between activating ERBB2 mutations and chemotherapy response seems counterintuitive. Signaling through the ERBB2 receptor activates a variety of signaling pathways that promote cell proliferation and oppose apoptosis. An increased proliferation rate of tumors with activated ERBB2 signaling provides a possible explanation for the increased sensitivity of ERBB2 mutant MIBCs to DNA damaging chemotherapy. However, we did not find a similar association between ERBB2 amplification and response to NAC. The level of signaling activation and the dimerization partners induced by ERBB2 mutation could be different compared to ERBB2 amplification, something that has not been investigated so far. Our findings, combined with the observed differences in the frequency of ERBB2 mutation and amplification across several tumor types, suggest that these genetic alterations may not have completely overlapping biological functions in carcinogenesis. Based on our findings we cannot conclude whether the association between ERBB2 mutations and response to NAC is a causal or correlational relationship. ERBB2 mutant tumors may lack a specific genetic alteration that is associated with poor response to NAC or, conversely, contain another (unidentified) alteration that is causally related with favorable response to NAC. Towards that end, we showed that ERBB2 mutations are not correlated with ERCC2 mutations. Patients in the TCGA urothelial bladder cancer cohort with ERBB2-mutant tumors have an excellent disease-specific survival, although the numbers are too small to draw strong conclusions (Supplementary Fig. S3). Information regarding the use of adjuvant treatment for these patients was incomplete. It is therefore not clear from the TCGA data whether ERBB2 mutations are prognostic for outcome or predictive for therapy response. Our study was not designed to address the prognostic value of genomic alterations in MIBC, but focused on the association with response to neoadjuvant chemotherapy. ERBB2 missense mutations could identify a subgroup of bladder cancers with good prognosis and exceptional response to platinumcontaining neoadjuvant chemotherapy. Further studies are needed to confirm the predictive value of ERBB2 mutations for a clinical benefit of NAC. Associations between ERBB2 amplification and clinical outcome or treatment response are not available or limited to case reports. We found no significant difference in ERBB2 amplification frequency between responders and nonresponders to platinum-containing chemotherapy. In four patients, amplification of ERBB2 was found in combination with a missense mutation in ERBB2. We showed that these tumors amplified the mutant ERBB2 allele, stressing once more the relevance of ERBB2 mutations for MIBC oncogenesis. Also, these findings could have implications for the current development of therapies directed at ERBB2

114 | Chapter 5

amplified tumors, as amplification of mutant ERBB2 may not be susceptible to HER2 antibodies such as trastuzumab. ERCC2 missense mutations were enriched in the responders, although not significantly, since ERCC2 missense mutations were also identified in a small percentage of the non-responders. The presence of ERCC2 mutations in non-responders to NAC could be explained by intratumoral heterogeneity with outgrowth of resistant tumor subclones lacking ERCC2 mutations. However, we clearly showed in two cases that the ERCC2 mutation was still present in the post-chemotherapy resistant tumors, demonstrating that ERCC2-mutant tumors can be resistant to platinum-based therapy. The ERCC2 mutation frequency in our responder cohort (18%) is lower than the 36% reported by Van Allen et al. in their responder cohort (Van Allen et al., 2014), which can be explained by the relatively small number of patients in both studies. In the present study, we sequenced only the exons around the helicase domains of ERCC2 in the validation cohort. We may therefore have missed mutations in other exons. Validation on additional patient cohorts is needed to definitively establish the association between ERCC2 mutations and chemotherapy response. A possible limitation of this study is the heterogeneity of our cohort. We acknowledge that patients were treated with different platinum-containing chemotherapy regimens. MVAC and the gemcitabine/cisplatin combination are generally accepted neo-adjuvant regimens, but we also included patients who received gemcitabine in combination with carboplatin. Of note: the two ERCC2-mutant non-responders were treated with gemcitabine/carboplatin. Although evidence for the benefit of this regimen in the neo-adjuvant setting in terms of cancer-specific or overall survival is lacking, pathologic complete response rates appear to be similar (Mertens et al., 2012). Furthermore, the mechanism of action for cisplatin and carboplatin is largely the same. Our cohort also contained more advanced cases than most neo-adjuvant studies in bladder cancer. However, these patients therefore reflect common clinical practice, as many clinics would specifically treat this high-risk patient group with chemotherapy, followed by resection if possible. In conclusion, activating ERBB2 missense mutations are enriched in MIBCs with complete response to platinum-containing NAC and absent in MIBCs that do not respond to NAC. ERBB2 missense mutation can therefore be used as a biomarker to select complete responders to neoadjuvant chemotherapy. Based on our findings, we conclude that it is unlikely that HER2-directed therapies will play an important role in the peri-operative treatment of ERBB2-mutant bladder cancers, since these patients respond very well to neoadjuvant chemotherapy and have an excellent

ERBB2-mutant MIBC and response to neoadjuvant chemotherapy | 115

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

prognosis. However, despite a complete response to NAC, some patients still developed recurrent disease. These patients may benefit from ERBB2 tyrosine kinase inhibitors, alone or in combination with chemotherapy.

PATIENTS AND METHODS Patients and samples The study population consisted of 110 prospectively registered patients who received neoadjuvant, platinum-containing combination chemotherapy (NAC) between 1992-2014 for cT2N1-3M0-1 or cT3-4N0-3M0-1, high-grade, muscle-invasive bladder cancer and who had pre-chemotherapy tissue available. Patients with M1disease had supraregional lymph nodes below the diaphragm that were considered surgically resectable, as has been published before (Meijer et al., 2013). All patients received at least two cycles of a platinum-containing chemotherapy combination: MVAC (methotrexate/vinblastine/adriamycin/cisplatin), gemcitabine/cisplatin or gemcitabine/carboplatin. After two cycles of chemotherapy, patients were restaged. In cases of clear progressive disease, chemotherapy was discontinued. Following NAC, eligible patients underwent radical cystectomy; other patients received palliative or curative radiotherapy. Formalin-fixed paraffin-embedded (FFPE) tumor blocks from the pre-chemotherapy diagnostic transurethral resection (TUR) were collected from referring hospitals for tissue microarray (TMA) assembly and DNA isolation. Clinical data were obtained from the prospectively maintained NKIAVL Genitourinary Clinical Database. Collection and use of tissue specimens was approved by the translational research board of the institute (study registration CFMPB104). Identification of “responders” and “non-responders” to NAC From this cohort, we identified 40 patients with a pathologic complete response to NAC (“responders”) at cystectomy and 33 patients with pT>2 after NAC (“nonresponders”). Pathologic complete response (pCR) was defined as no residual invasive cancer in both cystectomy and lymph nodes after NAC. Patients with carcinoma in situ, in the absence of an invasive component and no positive lymph nodes, were considered pCR. Target enrichment NGS of 178 cancer-associated genes Target enrichment DNA next-generation sequencing was performed with a custom SureSelect XT2 bait library (Agilent Technologies) covering a selected panel of

116 | Chapter 5

178 genes, consisting of (indirect or direct) clinically relevant genes including all frequently mutated genes in bladder cancer (Supplementary Table S3). Slides were cut from FFPE blocks (10x10μM) and DNA was isolated from pre-selected areas with >50% tumor cells using the QIAamp DNA mini kit (Qiagen). Concentration of double-stranded DNA (dsDNA) was measured using a fluorometric-based assay and 100ng-1ug dsDNA was used as input for SureSelect XT2 library preparation (Agilent Technologies). Captured libraries were sequenced on the Illumina HiSeq2000 platform with a paired-end 51-basepair read protocol. Sequence reads were aligned to the human reference genome (hg19) using the BWA backtrack algorithm (Li and Durbin, 2009. Variants were called using GATK Unified Genotyper (DePristo et al., 2011). The list of variants was then filtered to focus on variants affecting protein function using the impact prediction of SnpEff (e.g. low impact variants are not considered). To further filter out likely germline variants, all variants not reported in COSMIC but with annotation in dbSNP or EVS or exome sequencing data of 2000 healthy controls from the Netherlands were filtered from the list. Finally, the list of variants was manually curated for obvious false positives (e.g. variants not reported previously that occurred with high allele frequency). ERBB2 and ERCC2 Sanger sequencing in the validation cohort Pre-treatment tumor DNA from responders and non-responders was analyzed for missense mutations in ERBB2 and ERCC2. PCR primers were designed to amplify regions containing common mutation sites in 7 exons of ERBB2 and 9 exons around the helicase domains of ERCC2 (Supplementary Table S4). PCR products were generated in a 25μl reaction mixture containing 30ng gDNA and 1U Platinum Taq polymerase (Life Technologies). Agarose gel electrophoresis confirmed that the PCR products were of expected size and ExoSAP-IT (Affymetrix) was used for PCR cleanup. PCR products were capillary sequenced using the Big Dye Terminator V3.1 Sequencing Kit (Applied Biosystems). Sequences were compared with the reference sequence (NM_00448 for ERBB2 and NM_000400 for ERCC2). Sanger sequencing of germline DNA (isolated from serum, whole blood or normal tissue) verified all ERBB2 and ERCC2 mutations, identified by either target enrichment NGS or Sanger sequencing, as somatic. ERBB2 SISH Single-probe ERBB2 silver-enhanced in situ hybridization (SISH) was performed with the BenchMark Ultra autostainer (Ventana, Roche) on a TMA consisting of triplicate core samples (0.6mm) from the pre-treatment TUR of all MIBCs selected for this study. Paraffin sections (3 µm) were heated at 75 °C for 30 minutes and then

ERBB2-mutant MIBC and response to neoadjuvant chemotherapy | 117

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

deparaffinized in the instrument. Sections were treated for 8 minutes with ISH Protease 3 (cat.no. 5273331001, Ventana, Roche) for epitope retrieval. Sections were hybridized for 6 hours with the INFORM DNP-labeled HER2 DNA probe (cat.no. 5273439001, Ventana, Roche). After hybridization, appropriate stringency washsteps (three times 8 minutes) were performed followed by incubation with Rabbit anti-DNP (cat.no. 5273447001, Ventana, Roche) for 32 minutes. The HER2 probe was visualized using the UltraView SISH Detection Kit (Ventana, Roche) and sections were counterstained with haematoxylin. The detection kit contains a goat anti-rabbit antibody conjugated to horseradish peroxidase used as the chromogenic enzyme. The silver precipitation is deposited in the nuclei and a single copy of the ERBB2 gene is visualized as a black dot. Tumors were scored as amplified when the average ERBB2 copy number in >10% of the tumor cells was >6/cell which is in accordance with the ASCO/CAP guidelines for ERBB2 testing in breast cancer (Wolff et al., 2013). ERBB2 copy number determined by low-pass WGS DNA libraries, prepared with the SureSelect XT2 library preparation kit (Agilent Technologies), were sequenced on the Illumina HiSeq2000 platform with a singleread 51-basepair protocol. Sequence reads were aligned to the reference genome using the BWA backtrack algorithm (Li and Durbin, 2009). BEDTools (Quinlan and Hall, 2010) was used for counting the reads into 20kb non-overlapping bins and calculating the GC-content of those bins. The mappability value of each bin is precomputed by summarizing the alignment results of all possible 51mers from the reference sequence. Read counts for each bin are corrected for GC-bias using non-linear loess fit. Only bins with mappability >0.8 plus and bins from autosomes are used for this fit. After GC-correction, a linear model that intercepts zero is used to fit the mappabiltity data to the corrected read counts. The slope of this fit, multiplied with the mappability value for each bin, provides the binâ&#x20AC;&#x2122;s reference value that is used to calculate the final log2 copy number ratios for each sample. Bins overlapping blacklisted regions defined by ENCODE (ENCODE Project Consortium, 2012), as well as bins with a mappability <0.2 are excluded from the final dataset. Output files were imported and analyzed using Nexus Copy Number 7.5 software (BioDiscovery), which applies the FASST2 segmentation algorithm for calling copy number alterations.

118 | Chapter 5

ERBB2 immunohistochemistry Immunohistochemical ERBB2 staining was performed on the TMA with the BenchMark Ultra autostainer (Ventana, Roche). Paraffin sections (3 µm) were heated at 75 °C for 30 minutes and then deparaffinized in the instrument. Sections were treated with Cell Conditioning 1 buffer (Ventana, Roche) for 36 minutes followed by 12 minutes incubation with the primary PATHWAY anti-HER-2/neu antibody (Clone 4B5, cat.no. 5278368001, Ventana). Bound primary antibody was detected using the UltraView Universal DAB Detection Kit (Ventana, Roche) and slides were counterstained with haematoxylin. Scoring was performed by an experienced genitourinary pathologist (J.J.) using a 0-3 scoring system for membrane staining intensity, in accordance with the ASCO/CAP guidelines for ERBB2 testing in breast cancer (Wolff et al., 2013). ERBB2 mutation and amplification frequency in other cohorts ERBB2 mutation frequency in the TCGA urothelial bladder cancer dataset was determined using the TCGA data portal (http://cancergenome.nih.gov; downloaded Sept 1, 2014). ERBB2 mutation frequency in the study of Van Allen et al. was obtained from the supplemental information of their published manuscript (Van Allen et al., 2014). ERBB2 amplification frequency in the TCGA urothelial bladder cancer dataset and the MSKCC bladder cancer dataset (Iyer et al., 2013) was obtained from the cBioPortal for Cancer Genomics (http://cbioportal.org; accessed Aug 30, 2014) (Cerami et al., 2012). Statistical analysis Association between mutations in an individual gene and response to NAC was tested with a Fisher’s exact test. For the analysis of NGS on 178 genes, we corrected all tests for multiple testing with Benjamini-Hochberg method. Statistical analyses were performed with Bioconductor software (Gentleman et al., 2004) and GraphPad Prism software version 6. Differences in clinical characteristics between responders and non-responders were tested using the Mann-Whitney U test or the Fisher’s exact test. Kaplan-Meier analysis with log-rank statistics was used for survival analysis (SPSS software version 22, IBM). Cancer-specific survival was measured from the date of NAC start to the date of death-of-cancer. Recurrence-free survival was measured from the date of NAC start to the date of disease recurrence. Results were considered statistically significant with P < 0.05 (two-tailed).

ERBB2-mutant MIBC and response to neoadjuvant chemotherapy | 119

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

ACKNOWLEDGEMENTS The authors wish to acknowledge all patients for contributing tissue for research. We thank the Core Facility for Molecular Pathology & Biobanking of our institute for their assistance. We acknowledge Laura S. Mertens (department of Urology) and Joyce Sanders (department of Pathology) for the selection of patients and tissues. We have used data generated by the TCGA Research Network (http://cancergenome. nih.gov/).

120 | Chapter 5

REFERENCES Advanced Bladder Cancer (ABC) Metaanalysis Collaboration 2005. Neoadjuvant chemotherapy in invasive bladder cancer: update of a systematic review and metaanalysis of individual patient data. Eur Urol 48, 202-205; discussion 205-206. Bose, R., Kavuri, S.M., Searleman, A.C., et al., 2013. Activating HER2 mutations in HER2 gene amplification negative breast cancer. Cancer Discov 3, 224-237. Cerami, E., Gao, J., Dogrusoz, U., et al., 2012. The cBio cancer genomics portal: an open platform for exploring multidimensional cancer genomics data. Cancer Discov 2, 401-404. DePristo, M.A., Banks, E., Poplin, R., et al., 2011. A framework for variation discovery and genotyping using next-generation DNA sequencing data. Nat Genet 43, 491-498. ENCODE Project Consortium, 2012. An integrated encyclopedia of DNA elements in the human genome. Nature 489, 57-74. Gentleman, R.C., Carey, V.J., Bates, D.M., et al., 2004. Bioconductor: open software development for computational biology and bioinformatics. Genome Biol 5, R80. Greulich, H., Kaplan, B., Mertins, P., et al., 2012. Functional analysis of receptor tyrosine kinase mutations in lung cancer identifies oncogenic extracellular domain mutations of ERBB2. Proc Natl Acad Sci U S A 109, 14476-14481. Grossman, H.B., Natale, R.B., Tangen, C.M., et al., 2003. Neoadjuvant chemotherapy plus cystectomy compared with cystectomy alone for locally advanced bladder cancer. N Engl J Med 349, 859-866. Guo, G., Sun, X., Chen, C., et al., 2013. Wholegenome and whole-exome sequencing of bladder cancer identifies frequent alterations in genes involved in sister chromatid cohesion and segregation. Nat Genet 45, 1459-1463. Herter-Sprie, G.S., Greulich, H., Wong, K.K., 2013. Activating Mutations in ERBB2 and Their Impact on Diagnostics and Treatment. Front Oncol 3, 86.

Iyer, G., Al-Ahmadie, H., Schultz, N., et al., 2013. Prevalence and co-occurrence of actionable genomic alterations in high-grade bladder cancer. J Clin Oncol 31, 3133-3140. Li, H., Durbin, R., 2009. Fast and accurate short read alignment with Burrows-Wheeler transform. Bioinformatics 25, 1754-1760. Meeks, J.J., Bellmunt, J., Bochner, B.H., et al., 2012. A systematic review of neoadjuvant and adjuvant chemotherapy for muscle-invasive bladder cancer. Eur Urol 62, 523-533. Meijer, R.P., Nieuwenhuijzen, J.A., Meinhardt, W., et al., 2013. Response to induction chemotherapy and surgery in non-organ confined bladder cancer: a single institution experience. Eur J Surg Oncol 39, 365-371. Mertens, L.S., Meijer, R.P., Kerst, J.M., et al., 2012. Carboplatin based induction chemotherapy for nonorgan confined bladder cancer--a reasonable alternative for cisplatin unfit patients? J Urol 188, 1108-1113. Quinlan, A.R., Hall, I.M., 2010. BEDTools: a flexible suite of utilities for comparing genomic features. Bioinformatics 26, 841-842. Sonpavde, G., Goldman, B.H., Speights, V.O., et al., 2009. Quality of pathologic response and surgery correlate with survival for patients with completely resected bladder cancer after neoadjuvant chemotherapy. Cancer 115, 41044109. The Cancer Genome Atlas Network, 2014. Comprehensive molecular characterization of urothelial bladder carcinoma. Nature 507, 315322. Van Allen, E.M., Mouw, K.W., Kim, P., et al., 2014. Somatic ERCC2 Mutations Correlate with Cisplatin Sensitivity in Muscle-Invasive Urothelial Carcinoma. Cancer Discov 4, 11401153. Wolff, A.C., Hammond, M.E., Hicks, D.G., et al., 2013. Recommendations for human epidermal growth factor receptor 2 testing in breast cancer: American Society of Clinical Oncology/College of American Pathologists clinical practice guideline update. J Clin Oncol 31, 3997-4013.

ERBB2-mutant MIBC and response to neoadjuvant chemotherapy | 121

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

SUPPLEMENTARY DATA

Supplementary Figure S1. (A) Kaplan-Meier analysis of recurrence-free survival in responders and non-responders to neoadjuvant chemotherapy. (B) Kaplan-Meier analysis of cancer-specific survival in responders and non-responders to neoadjuvant chemotherapy. 122 | Chapter 5

Supplementary Figure S2 A Germline (serum)

G45

S44L

P43

Pre-treatment TUR

CF3097 p.D769H 60% tumor

Post-treatment TUR

5 9013 p.T306M + p.S310F 80% tumor

c.2305G>C

c.917C>T

c.929C>T

Supplementary Figure S2. (A) sequencing results from an ERCC2-mutant non-responder (patientID CF0950) showing the ERCC2 missense mutation (S44L) in DNA isolated from the pre-treatment TUR and posttreatment cystectomy, but not in the germline DNA. This demonstrates that this somatic ERCC2 mutation was not counter-selected during chemotherapy. (B) ERBB2 Sanger sequence of DNA isolated from the pre-treatment TUR (patientID CF3097) showing an amplification of the mutant ERBB2 allele. The peak from the mutant nucleotide (C, in blue) is higher than the peak form the reference nucleotide (G, in black). The tumor cell percentage was 60%. (C) ERBB2 Sanger sequence of DNA isolated from the pre-treatment TUR (patientID 9013), showing an amplification of the double-mutant ERBB2 allele. The peaks from the mutant nucleotide (T, in red) are higher than the peaks form the reference nucleotide (C, in blue). The tumor cell percentage was 80%.

ERBB2-mutant MIBC and response to neoadjuvant chemotherapy | 123

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

Supplementary Figure S3

Supplementary Figure S3. Kaplan-Meier analysis of disease-specific survival in ERBB2-mutant and ERBB2-wild type patients from the TCGA urothelial bladder cancer cohort.

124 | Chapter 5

ERBB2-mutant MIBC and response to neoadjuvant chemotherapy | 125

Responder

MVAC

G/Cisplatin

G/Carboplatin

MVAC

G/Carboplatin

MVAC

G/Carboplatin

NGS

Sanger

NGS

Sanger

c.2524G>A

c.2329G>A

c.2305G>C

c.2033G>T

c.929C>A

c.929C>T

c.917C>T

Chemotherapy Methods Mutation (CDS) regimen

p.V842I

p.V777M

p.D769H

p.R678L

p.S310Y

p.S310F

p.T306M

Protein change

CF0979

CF0959

CF3098

CF0982

CF0949

CF1000

Responder

Non-responder

Responder

CF1005

CF0963

Non-responder

CF0950

Patient ID Cohort

MVAC

G/Cisplatin

MVAC

G/Carboplatin

MVAC

G/Cisplatin

G/Carboplatin

NGS

Sanger

NGS

Sanger

NGS

c.1451C>T

c.769G>A

c.713A>G

c.256G>C

c.215A>G

c.136A>G

c.131C>T

Chemotherapy Method Mutation (CDS) regimen

p.T484M

p.D257N

p.N238S

p.E86Q

p.Y72C

p.T46A

p.S44L

Y Y

Normal tissue Serum

Serum

Normal tissue

Serum

Whole blood

Normal tissue

Serum

Whole blood 0.22

0.42

0.52

0.30

N/A

0.34

0.64

N/A

Germline DNA Confirmed Fraction MT/WT source somatic reads

Serum

Normal tissue

Serum

Y Serum

Serum

0.33

N/A

0,34

0.25

0.32

0.21

N/A

Protein Germline DNA Confirmed Fraction MT/WT change source somatic reads

Supplementary Table S2. Summary of all identified somatic ERCC2 missense mutations

CF0977

Responder

CF0940

CF3097

Responder

CF3122

CF0992

9013

CF0955

Responder

CF1005

CF0945

Responder

9013

CF0943

Cohort

Patient ID

Supplementary Table S1. Summary of all identified somatic ERBB2 missense mutations

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

Supplementary Table S3. 178 gene panel for target enrichment NGS ABCC1

ABCG2 AREG

BCL2L11 CDH1

BCR

CDK4

CSF1R

DUSP4

CTNNB1 DUSP6

EPHB6

ERBB2

ESR2

EZH2

GNAQ

GNAS

ABL1

ARID1A BIM

CDK6

CYP2C19 EGFR

ERBB3

FBXW7

GRB7

ABL2

ATM

BIRC5

CDK8

CYP2C9

ERBB4

FGFR1

GSTM1

AEG1

ATR

BRAF

AKT1

AURKA BRCA1

EML4

CDKN2A CYP2D6

EP300

ERCC1

FGFR2

GSTP1

CENPE

EPCAM

ERCC2

FGFR3

HDAC2

CYP3A4

AKT2

AURKB

BRCA2

CHEK1

DCK

EPHA3

ERCC4

FGFR4

HDAC9

AKT3

AURKC

CCND1

CHEK2

DDR1

EPHA5

EREG

FLT1

HGF

ALK

AXL

CCNE1

COMT

DDR2

EPHA7

ERG

FLT3

HRAS

APC

BCL2

CDA

CREBBP

DPYD

EPHB1

ESR1

FLT4

IDH1

IDH2

KIT

MIK67

NF2

PIK3CA

RECQL

SMAD4

TGFBR2

TYMS

IGF1R

KRAS

MLH1

NRAS

PIK3R1

RET

SMARCA4 TMPRSS2 UGT1A1

IGF2

LRP2

MLL

OCT2

PIK3R3

ROS1

SMARCB1 TNK2

VEGFA

PALB2

PMS2

RRM1

SMO

VEGFB

INPP4A MAP2K1 MLL2

TOP1

INPP4B MAP2K4 MLL3

PAX5

POLE

RSPO2

SOX2

TOP2A

VHL

JAK2

MAP3K1 MSH2

PDCD1

PSMC2

RSPO3

SRC

TP53

YES1

JAK3

MCL1

STK11

KDM6A MDM2 KDR

KEAP1

MED12

MET

126 | Chapter 5

MSH6

PDGFRA PTEN

RUNX1

MYC

PDGFRB

RUNX1T1 TCF7L2

MYCN

NF1

PDL-1

PGR

PTK2 RAF1

RB1

SIRT2

SMAD2

TERT

TGFBR1

TPMT

YWHAZ

TSC1

ZAP70

TSC2

TYMP

Supplementary Table S4. ERBB2 and ERCC2 PCR primers Primers for ERBB2 PCR amplification Primer sequence

Forward: 5’-ACGGTAATGCTGCTCATGGT Reverse: 5’-GCTTGCTGCACTTCTCACAC Forward: 5’-CTAGCCCTCAATCCCTGACC Reverse: 5’-GGCTGGGAGGACTTCACC Forward: 5’-GATGCGGATCCTGAAAGAGA Reverse: 5’-TCCCTTCTCCGCTGTAACTG Forward: 5’-CCCACGCTCTTCTCACTCAT Reverse: 5’-AGAGACCAGAGCCCAGACCT Forward: 5’-GCTGTGGTTTGTGATGGTTG Reverse: 5’-TCCCGGACATGGTCTAAGAG Forward: 5’-TACATGGGTGCTTCCCATTC Reverse: 5’-TCTGCTCCTTGGTCCTTCAC Forward: 5’-CTCCCCACAACACACAGTTG Reverse: 5’-AGCTCTCATCCTCCCTCCAG 1

Exon 1

Amino acids covered

aa. 302 - 335

aa. 650 - 695

aa. 714 - 736

aa. 737 - 769

aa. 770 - 810

aa. 832 - 887

aa. 888 - 908

Exon 2

Amino acids covered

aa. 3 - 35

aa. 36 - 61

aa. 62 - 82

aa. 83 - 120

aa. 199 - 239

aa. 240 -271

aa. 460 - 493

aa. 587 - 610

aa. 635 - 682

NM_00448

Primers for ERCC2 PCR amplification Primer sequence

Forward: 5’-GAGGGGACGGGAACTGAC Reverse: 5’-ATCCAGACGTCCTGCAATCT Forward: 5’-CCTGCTGTTCTGAGTTGGTG Reverse: 5’-TTGTCTGCCTTTACGGGTTC Forward: 5’-AGAGTGAGTGATGCGCTGAA Reverse: 5’-ACCAATAGGGCCTAGGGAAC Forward: 5’-GAAGCTGGGGAAGAGACTGG Reverse: 5’-GCAAGGAGAAGGAACAGGTG Forward: 5’-CAGTGTGGCCAGGGGTAG Reverse: 5’-GCAAGGAGAAGGAACAGGTG Forward: 5’-CCAGCCCCTCTGAGTGAG Reverse: 5’-CCTGGGGACAAGTCAGACAG Forward: 5’-GTCCCCTCCTAGTCCCTGAC Reverse: 5’-CCCTAGCCTCTCCCACTCAC Forward: 5’-AGTGTCGCCCTGGAGAAGTA Reverse: 5’-GAGCTCTGGGAAGACACCTG Forward: 5’-CCAGCTTCTCATCCTCCGTA Reverse: 5’-CTGGGAAATGAACGGGAAAC 2

NM_000400

ERBB2-mutant MIBC and response to neoadjuvant chemotherapy | 127

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

Chapter 6 Copy number alterations and subtyping in muscleinvasive bladder cancer correlate with response to neoadjuvant chemotherapy

Floris H. Groenendijk1, Elisabeth E. Fransen van de Putte2, Jeroen de Jong3, Bas W.G. van Rhijn2, RenĂŠ Bernards1 and Michiel S. van der Heijden1,4

Division of Molecular Carcinogenesis, Cancer Genomics Netherlands; 2 Department of Urology;

Department of Pathology; 4 Department of Medical Oncology, The Netherlands Cancer Institute,

Antoni van Leeuwenhoek Hospital, Plesmanlaan 121, 1066 CX Amsterdam, The Netherlands.

Manuscript in preparation

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

ABSTRACT Gene expression profiling of muscle-invasive bladder cancer (MIBC) has recently identified two major MIBC subtypes, basal-like and luminal-like, with distinct genetic alterations and clinical outcome. We analyzed a series of 100 MIBCs treated with neoadjuvant platinum-based chemotherapy and found that luminal-like MIBCs have a higher response rate to chemotherapy than basal-like MIBCs. MIBC subtyping was the strongest independent prognostic factor in the multivariate cancer-specific survival analysis (P = 0.005; hazard ratio 2.75; 95%CI, 1.39-5.45). In addition, we found that patients with basal-like MIBC had a significantly shorter post-recurrence survival compared to patients with luminal-like MIBC. This indicates that basal-like MIBCs are aggressive tumors that progress rapidly after recurrence. We subsequently analyzed a group of complete responders and non-responders to neoadjuvant chemotherapy for DNA copy number alterations. Copy number gain of the E2F3 locus was enriched in responders, whereas copy number gain of the MYC locus and copy number loss of the CDKN2A locus were both associated with non-response. E2F3 overexpression in the UMUC3 bladder cancer cell line increased sensitivity of the cells to platinum-compounds. Our data suggest that copy number alterations could guide the decision to offer neoadjuvant chemotherapy to MIBC patients. However, our findings require independent clinical validation in additional cohorts of neoadjuvant treated MIBCs.

KEYWORDS Muscle-invasive bladder cancer (MIBC)  Intrinsic subtypes  Copy number alterations  Neoadjuvant chemotherapy

130 | Chapter 6

INTRODUCTION Genome-wide analyses have revealed that muscle-invasive bladder cancer (MIBC) is a biologically heterogeneous disease, which may explain the variable and unpredictable response to neoadjuvant chemotherapy (NAC). Recent gene expression profiling studies have shown that MIBCs can be divided into basal and luminal subtypes (Choi et al., 2014; Damrauer et al., 2014; Volkmer et al., 2012). These distinct subtypes correlate with tumor differentiation state and expression of different keratins (Volkmer et al., 2012). Choi et al. found that the luminal subtype can be subdivided into a third subtype, called p53-like, which was more resistant to cisplatin-based chemotherapy (Choi et al., 2014). These subtypes are enriched for specific mutations and copy number alterations (Choi et al., 2014; Damrauer et al., 2014). Genome-wide analyses have identified multiple regions of somatic copy number alterations (CNAs) in high-grade bladder cancer (Iyer et al., 2013; The Cancer Genome Atlas Network, 2014). However, the association of specific copy number alterations with treatment response is unknown. We have used keratin protein expression to subtype MIBCs into basal-like and luminal-like tumors and found that the response to neoadjuvant chemotherapy is worse for basal-like MIBCs. Copy number analysis of complete responders and nonresponders to neoadjuvant chemotherapy identified multiple alterations associated with either response or non-response.

RESULTS Patient characteristics We selected 100 patients with MIBC who received neoadjuvant chemotherapy and for whom pre-chemotherapy tissue was available for tissue microarray (TMA). Subtyping based on protein expression of keratin 20 and keratin 5/6 was available for 90 patients. Subtyping was unavailable for the other 10 patients due to insufficient tumor material in the tissue microarray (TMA). Clinical characteristics for the total patient cohort, as well as stratified by subtype, are listed in Table 1. Patients with basal-like MIBC had a higher frequency of cT3 and cT4 tumors compared to patients with luminal-like MIBC, although this difference was not statistically significant (P = 0.08; Table 1). Following NAC, 86 patients underwent radical cystectomy, 7 patients received radiotherapy with curative intent, 6 patients received palliative treatment and one patient did not receive additional treatment due to disease progression.

CNVs and subtyping in MIBC correlate with response to neoadjuvant chemotherapy | 131

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

Table 1. Clinical characteristics stratified by subtype Total (N=100)

Luminal-like Basal-like P-value1 (N=48) (N=42)

Median age, years (IQR)

63 (13.9)

63 (12.6)

64 (16.9)

0.64

Female sex, N (%)

19 (19.0)

6 (12.5)

9 (21.4)

0.27

Clinical TNM staging, N (%) T2 T3 T4

32 (32.0) 37 (37.0) 31 (31.0)

21 (43.8) 15 (31.3) 12 (25.0)

9 (21.4) 18 (42.9) 15 (35.7)

0.08

N0 N1 N2 N3

31 27 33 9

13 11 16 8

13 15 13 1

(31.0) (35.7) (31.0) (2.4)

0.11

M0 M12

88 (88.0) 12 (12.0)

42 (87.5) 6 (12.5)

39 (92.9) 3 (7.1)

0.49

TUR histology, N (%) Urothelial cell Other

98 (98.0) 2 (2.0)

47 (97.9) 1 (2.1)

41 (97.6) 1 (2.4)

Neoadjuvant chemotherapy regimen, N (%) MVAC Gemcitabine/cisplatin Gemcitabine/carboplatin

54 (54.0) 22 (22.0) 24 (24.0)

26 (54.2) 10 (20.8) 12 (25.0)

24 (57.1) 9 (21.4) 9 (21.4)

0.96

4 (0.0)

0.67

Patients with recurrence, N (%)

57 (57.0)

25 (52.1)

28 (66.7)

0.20

Mean length of follow-up, years (95%CI)

9.3 (7.2-11.3)

Median no. of chemotherapy cycles

(31.0) (27.0) (33.0) (9.0)

(27.1) (22.9) (33.3) (16.7)

11.4 (8.6-14.3)

5.0 (3.1-6.9)

132 | Chapter 6

Basal and luminal subtyping Subtyping was performed by keratin 5/6 and keratin 20 immunohistochemistry on the pre-treatment TUR tissue. Forty-eight tumors (53%) were designated as luminallike MIBC and 42 tumors (47%) were designated as basal-like MIBC. An example of differential keratin expression in both subtypes is shown in Figure 1. Basal-like cancers had a significantly lower pathologic response rate (P = 0.02; Table 2) and pT<2 downstaging frequency (P = 0.02; Table 2) compared to luminal-like cancers, and a trend towards lower chance of achieving pathologic complete response (P = 0.08; Table 2). Cancer-specific survival was significantly worse for patients with basallike MIBC compared to patients with luminal-like MIBC in univariate analysis (P = 0.01; Fig. 2A). Subtype was the strongest independent prognostic factor for cancerspecific survival in the multivariate cox regression analysis (P = 0.004; hazard ratio 2.75; 95%CI, 1.39-5.45; Table 3). We found that patients with basal-like MIBC had a significantly shorter post-recurrence survival compared to patients with luminallike MIBC (P = 0.02; Fig. 2B). This indicates that basal-like MIBCs are aggressive tumors that progress rapidly after recurrence.

Figure 1

keratin 5/6

Basal-like

Luminal-like

keratin 20

Figure 1. Example of immunohistechemical staining for keratin 20 and keratin 5/6 in luminallike and basal-like MIBCs.

CNVs and subtyping in MIBC correlate with response to neoadjuvant chemotherapy | 133

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

Table 2. Response to neoadjuvant chemotherapy stratified by subtype.

Luminal-like (n=48)

Basal-like (n=42)

P-value1

Pathologic response, N (%)

32 (67)

17 (40)

0.02

pT<2 downstaging, N (%)

29 (60)

15 (36)

0.02

Pathologic complete response (pCR), N (%)

22 (46)

11 (26)

0.08

Fisherâ&#x20AC;&#x2122;s exact test

Table 3. Multivariate Cox regression analysis for cancer-specific survival Variable

Hazard ratio

95% CI

P-value

0.99

0.95-1.02

0.523

Sex Male Female

Reference 1.15

0.48-2.74

0.751

cT stage cT2 cT3 cT4

Reference 0.36 1.11

0.14-0.91 0.49-2.52

0.032 0.809

cN stage cN0 cN1 cN2 cN3

Reference 0.93 1.35 0.92

0.37-2.37 0.55-3.34 0.25-3.44

0.878 0.510 0.904

cM stage cM0 cM1

Reference 2.48

0.92-6.68

0.074

Neoadjuvant chemotherapy regimen MVAC Gemcitabine/cisplatin Gemcitabine/carboplatin

Reference 0.91 2.46

0.38-2.16 1.10-5.48

0.822 0.028

Subtype Luminal-like Basal-like

Reference 2.75

1.39-5.45

0.004

Age (continuous)

134 | Chapter 6

Figure 2 A 100 Probability of cancer-specific survival (%)

Luminal-like MIBC Basal-like MIBC 80

P = 0.01

0 0

5 10 15 Time since start neoadjuvant chemotherapy (years)

B 100

Probability of post-recurrence survival (%)

Luminal-like MIBC Basal-like MIBC 80

P = 0.02

0 0

2 4 Time from recurrence to death (years)

Figure 2. (A) Kaplan-Meier analysis of cancer-specific survival in luminal-like (n=48) and basal-like (n=42) MIBC patients. (B) Kaplan-Meier analysis of post-recurrence survival in luminal-like (n=24) and basal-like (n=26) MIBC patients.

CNVs and subtyping in MIBC correlate with response to neoadjuvant chemotherapy | 135

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

Volkmer et al. reported that keratin 14 marks the most primitive tumor differentiation state in bladder cancer (Volkmer et al., 2012). It precedes the expression keratin 5/6 and keratin 20, and is associated with worse prognosis. In our cohort, expression of keratin 14 by immunohistochemistry (score â&#x2030;Ľ2) was selectively found in basallike MIBCs. Patients with basal-like MIBC with expression of keratin 14 had a trend towards shorter cancer-specific survival compared to patients with basal-like MIBCs that did not express keratin 14 (P = 0.08; Fig. 3).

Figure 3 100

keratin 14 (-) basal-like MIBC keratin 14 (+) basal-like MIBC Probability of cancer-specific survival (%)

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

P = 0.08

0 0

4 6 8 10 12 Time since start neoadjuvant chemotherapy (years)

Figure 3. Kaplan-Meier analysis of cancer-specific survival in keratin 14-negative basal-like MIBC patients (n=27) and keratin 14-positive basal-like MIBC patients (n=13).

136 | Chapter 6

CNAs in responders and non-responders to NAC DNA copy number alterations in pre-treatment tumor tissue were defined by using low-pass whole-genome sequencing, which was available for 26 responders and 24 non-responders to NAC. For the copy number analysis, we focused on genes with recurrent focal CNAs in the published TCGA urothelial bladder cancer dataset (Supplementary Table S1) (The Cancer Genome Atlas Network, 2014). Figure 4 shows the oncoprint for genes with a trend towards association with one of the two response-groups. We found that E2F3 locus copy number gains were significantly enriched in responders compared to non-responders (P = 0.01). Conversely, MYC locus copy number gains were significantly enriched in non-responders to NAC (P = 0.04). Copy number losses of the CDKN2A locus were also enriched in nonresponders to NAC, although this association did not reach statistical significance (P = 0.08). Copy number gains of the E2F3 locus had a strong tendency towards mutual exclusivity with copy number alterations of the MYC and CDKN2A locus (P = 0.04), which is consistent with the published results from Iyer et al. (Iyer et al., 2013) and TCGA (The Cancer Genome Atlas Network, 2014). Copy number gains of the ERBB2 locus showed a trend towards enrichment in responders (P = 0.10), although six of the seven copy number gains overlapped with E2F3 locus copy number gains. ERBB2 locus copy number gain was the only CNA significantly associated with one of the two subtypes. ERBB2 locus copy number gains were enriched in luminal-like tumors (P = 0.01). E2F3 overexpression sensitize bladder cancer cells to platinum-compounds Having found the association between E2F3 locus copy number gain and platinumresponse, we next tested the effect of E2F3 overexpression in bladder cancer cells on sensitivity to platinum-compounds. UMUC3 bladder cancer cells infected with pBP-E2F3 or pBP-control were treated for 24 hours with cisplatin or carboplatin. The clonogenic assay shows that E2F3 overexpressing UMUC3 cells are more sensitive to platinum-compounds when compared to control cells (Fig. 5), supporting the role of E2F3 in response to platinum-compounds.

CNVs and subtyping in MIBC correlate with response to neoadjuvant chemotherapy | 137

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

Figure 4. OncoPrint showing the subtype and copy number alterations in the E2F3, MYC, CDKN2A and ERBB2 locus for 26 responders and 24 non-responders to NAC. Individual patients are represented as columns.

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

138 | Chapter 6

Figure 5 A

UMUC3 cells 10day clonogenic assay pBP-ctrl

UMUC3 cells 10day clonogenic assay

pBP-E2F3

pBP-ctrl

untreated

24hrs .5 μM cisplatin

24hrs 3 μM carboplatin

24hrs 1 μM cisplatin

24hrs 4 μM carboplatin

pBP-E2F3

E2 P-

PpB

D rl

HA-E2F3 loading

Figure 5. E2F3 overexpression sensitize cells to alkylating chemotherapy. (A) Clonogenic assay of control and E2F3 overexpressing UMUC3 cells after 24hrs exposure to indicated concentrations of cisplatin. (B) Clonogenic assay of control and E2F3 overexpressing UMUC3 cells after 24hrs exposure to indicated concentrations of carboplatin. (C) Relative E2F3 mRNA expression in control and E2F3 overexpressing UMUC3 cells (D) HA-E2F3 protein expression in control and E2F3 overexpressing UMUC3 cells.

CNVs and subtyping in MIBC correlate with response to neoadjuvant chemotherapy | 139

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

DISCUSSION We observed a significant correlation between intrinsic subtypes of MIBC and response to platinum-containing neoadjuvant chemotherapy or clinical outcome. Patients with basal-like MIBC had a worse cancer-specific and post-recurrence survival and a lower response rate to neoadjuvant chemotherapy. The subtyping in our study was based on keratin protein expression by immunohistochemistry, which can be applied more easily in the clinical setting than subtyping with gene expression profiling as has been used in all previously published studies on subtyping in bladder cancer. It is therefore difficult to compare our results with the published results. Choi et al. stated that in control experiments the array-based measurements of basal and luminal marker expression correlated well with the results obtained by IHC (Choi et al., 2014). However, they did not report on the correlation for the actual subtyping using both methods. Further studies should be performed to compare the validity of subtyping by IHC with subtyping by gene expression profiling, and define the cutoffs that should be used. Towards that end, the distribution of basal-like and luminal-like in our study was comparable with the distribution in previous studies. We were unable to subclassify luminal tumors into p53-like, as Choi et al. reported, since IHC markers for this subtype do not exist (Choi et al., 2014). It could well be that there is a mixed intermediate subtype between the basal and luminal subtypes that is missed by only looking at keratin markers. We observed that in some bladder cancers, the different cores in our TMA displayed a different expression pattern of keratins, which may reflect intratumoral heterogeneity. Further studies should also address to what extent intratumor heterogeneity affect subtyping by gene expression profiling or by IHC. This is the first study that describes the association of specific copy number alterations and response to neoadjuvant chemotherapy in muscle-invasive bladder cancer. We found that copy number gain of the E2F3 locus was correlated with complete response to neoadjuvant chemotherapy in our MIBC cohort. In the TCGA urothelial bladder cohort, amplification of E2F3 was found in 18.3% of the samples and this was associated with increased E2F3 mRNA expression (http://cbioportal.org). The association between E2F3 and chemotherapy response has been observed before in breast cancer. A gene expression signature of E2F3 activation was significantly associated with the probability of pathological complete response to neoadjuvant chemotherapy in triple-negative breast cancer (Ignatiadis et al., 2012; Silver et al., 2010; Tordai et al., 2008), but not in HER2-positive breast cancer (Ignatiadis et al., 2012). This may be explained by the critical role of the E2F3 family in regulating cell cycle progression (DeGregori, 2002). 140 | Chapter 6

The E2F3 locus (6p22.3) encodes two isoforms (E2F3a and E2F3b), which differ in their N-terminus. Bladder cancer cell lines with 6p22 amplification overexpress both E2F3 isoforms (Feber et al., 2004; Hurst et al., 2008). Knockdown of E2F3a or E2F3b alone induced antiproliferative effects in bladder cancer cell lines, and the effect was increased with a combined knockdown of E2F3a and E2F3b (Hurst et al., 2008; Shen et al., 2013). This suggests that E2F3a and E2F3b co-operate in these cells. However, the two E2F3 isoforms may have partially opposing roles since E2F3a is classified as a transcriptional activator and E2F3b as a transcriptional repressor. It has been found that E2F3a is specifically induced by DNA damage, and is required for DNA damage-induced apoptosis (Martinez et al., 2010). Increased levels of E2F3 may result in a more potent apoptotic response of cells to DNA damage induced by platinum-containing chemotherapy. Our E2F3 overexpression experiment further supports this hypothesis (Fig. 5). However, more experiments are needed to address the functional consequences of E2F3 overexpression in bladder cancer cells and dissect the roles of the two isoforms in relation with response to platinum compounds. We also found a non-significant association between ERBB2 locus copy number gain and response to NAC. This association was tested in a larger cohort of responders and non-responders to NAC in chapter 5 of this thesis. We did not find a correlation between ERBB2 amplification (by SISH) and response to NAC. Choi et al. reported that basal tumors were enriched for MYC expression (Choi et al., 2014). In our cohort, most MYC locus copy number gains were also prominent in basal-like tumors, although the number of alterations was small. It may well be that the correlation between MYC locus copy number gains and non-response to NAC is caused by the enrichment of basal-like tumors in non-responders to NAC. However, our finding that copy number gain of the MYC locus, and also copy number loss of the CDKN2A locus, is associated with unfavorable response to NAC, implies that those patients may not benefit from NAC and should be treated differently. A suggested strategy for targeting MYC activated tumors is the inhibition of BET bromodomain proteins, regulatory factors for MYC, using a smallmolecular bromodomain inhibitor (e.g. JQ1) (Bandopadhayay et al., 2014; Delmore et al., 2011). We speculate that there may be a role for bromodomain inhibitors in the treatment of MYC amplified bladder cancers. CDKN2A, known as p16, acts as a tumor suppressor by binding to CDK4/6 and preventing its interaction with cyclin D (Liggett and Sidransky, 1998). Loss of p16 results in dysregulation of cell cycle progression (Liggett and Sidransky, 1998). However, a promising therapeutic strategy for bladders cancers with CDKN2A loss is not available yet.

CNVs and subtyping in MIBC correlate with response to neoadjuvant chemotherapy | 141

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

In conclusion, this work shows that keratin expression, measured by immunohistochemistry, can be used to classify MIBCs into basal-like and luminallike tumors. This classification has prognostic value for patients treated with neoadjuvant chemotherapy. We provide new insights into the correlation between copy number alterations and response to platinum-based neoadjuvant chemotherapy in MIBC. Our findings require independent clinical validation in additional cohorts of neoadjuvant treated MIBCs.

METHODS Patients and samples We selected patients who received neoadjuvant, platinum-containing combination chemotherapy (NAC) between 1992-2013 in the AVL for cT2N1-3M0-1 or cT3-4N03M0-1, high-grade, muscle-invasive bladder cancer and who had pre-chemotherapy tissue available. All patients received at least two cycles of a platinum-containing chemotherapy combination: MVAC (methotrexate/vinblastine/adriamycin/ cisplatin), gemcitabine/cisplatin or gemcitabine/carboplatin. After two cycles of chemotherapy, patients were restaged and the decision was taken if chemotherapy treatment should be continued or not. Following NAC, eligible patients underwent radical cystectomy; other patients received palliative or curative radiotherapy. Formalin-fixed paraffin-embedded (FFPE) tumor blocks from the pre-chemotherapy diagnostic transurethral resection (TUR) were collected from referring hospitals for TMA assembly and DNA isolation. Clinical data were obtained from the prospectively maintained NKI-AVL Genitourinary Clinical Database. The collection and use of tissue specimens was approved by the translational research board of the institute. Pathologic assessment Pathologic response to NAC was obtained from the surgical pathology records. Pathologic complete response (pCR) was defined as no residual invasive cancer in both cystectomy and lymph nodes after NAC. Pathologic local response was defined as downstaging to pT<2. Comparing the post-chemotherapy pathological staging with pre-treatment clinical staging identified partial responders. Pathologic response rate was calculated by combining the complete and partial responders. Subtyping based on keratin protein expression Keratin 5/6 and keratin 20 expression was analyzed on a TMA consisting of

142 | Chapter 6

triplicate core samples (0.6mm) from the pre-treatment TUR. Immunohistochemical staining was performed on the BenchMark Ultra automated staining instrument (Ventana, Roche). Paraffin sections (3µm) were heated at 75 °C for 30 min and then deparaffinized in the instrument. Sections were treated with CC1 buffer (Ventana, Roche) for 36 minutes before incubation with the primary antibody for 32 minutes. Primary antibodies used were anti-keratin 5/6 (clone D5/16 B4, cat.no. M7237, Dako) and anti-keratin 20 (clone Poly, cat.no. E16444, ImmunoLogics). For keratin 5/6 this was followed by an amplication step. Bound primary antibody was detected using the Universal DAB Detection Kit (Ventana, Roche) and slides were counterstained with haematoxylin. Scoring of the TMA IHC stainings was performed by an experienced genitourinary pathologist (J.J.) using a 0-3 scoring system for staining intensity. Tumors were classified as luminal-like when the keratin 20 score was higher than the keratin 5/6 score and basal-like when the score for keratin 5/6 score was higher than the keratin 20 score. When both markers were scored negative, tumors were classified as basallike. Keratin 14 immunohistochemistry Keratin 14 protein immunohistochemical staining was performed on the BenchMark Ultra automated staining instrument (Ventana, Roche). The staining protocol was identical to the protocol used for keratin 5/6 IHC. Primary antibody used was antikeratin 14 (clone LL002, cat.no. MS-115P, Thermo Scientific). Low-pass whole-genome sequencing for copy number alterations H&E stainings were evaluated to identify areas with >50% tumor cells for DNA extraction. Slides were cut from FFPE blocks and DNA was isolated from preselected areas with >50% tumor cells using the QIAamp DNA mini kit (Qiagen). Concentration of double-stranded DNA (dsDNA) was measured using a fluorometric-based assay and 100ng-1ug dsDNA was used as input for SureSelect XT2 library preparation (Agilent Technologies). Libraries were sequenced on the Illumina HiSeq2000 platform with a single-read 51-basepair protocol. Sequence reads were aligned to the reference genome using the BWA backtrack algorithm (Li and Durbin, 2009). BEDTools (Quinlan and Hall, 2010) was used for counting the reads into 20kb non-overlapping bins and calculating the GC-content of those bins. The mappability value of each bin is precomputed by summarizing the alignment results of all possible 51mers from the reference sequence. Read counts for each bin are corrected for GC-bias using non-linear loess fit. Only bins with mappability >0.8 plus and bins from autosomes are used for this fit. After GC-correction, a linear model

CNVs and subtyping in MIBC correlate with response to neoadjuvant chemotherapy | 143

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

that intercepts zero is used to fit the mappabiltity data to the corrected read counts. The slope of this fit, multiplied with the mappability value for each bin, provides the bin’s reference value that is used to calculate the final log2 copy number ratios for each sample. Bins overlapping blacklisted regions defined by ENCODE (ENCODE Project Consortium, 2012), as well as bins with a mappability <0.2 are excluded from the final dataset. Output files were imported and analyzed using Nexus Copy Number 7.5 software (Biodiscovery), which applies the FASST2 segmentation algorithm for calling copy number alterations. All copy number variant calls made by the software were manually inspected and only high gains (log2ratio >0.7) or big losses (log2ratio <-1.0) were used for further analysis. E2F3 overexpression The bladder cancer cell line UMUC3 was cultured in RPMI supplemented with 8% heat-inactivated fetal calf serum and 1% penicillin/streptomycin at 5% CO2. A subclone expressing the ecotropic receptor was generated to allow efficient infection with ecotropic retroviruses. Phoenix packaging cells were used as producers of high titer retroviruses. Viral supernatants were filtered through a 0.45μm filter, and infections were performed in the presence of 4µg/ml polybrene (Sigma). Infected cells were selected for successful retroviral integration using 2 μg/ml of puromycin. The pBP-HA-E2F3 overexpression construct has been described before (Xu et al., 1995). Control infections were performed with the empty pBP vector. E2F3 overexpression was analyzed by western blot and quantitative RT-PCR. For western blot analysis, cells lysates were collected using RIPA buffer containing 150 mM NaCl, 50 mM Tris pH 8.0, 1% NP-40, 0.5% sodium deoxycholate and 0.1% SDS supplemented with protease inhibitors (Complete, Roche) and Phosphatase Inhibitor Cocktails II and III (Sigma). Lysates were normalized using BCA protein assay (Thermo Scientific) and resolved by SDS gel electrophoresis and followed by western blotting. Primary antibody against HA-tag to detect HA-E2F3 was ab9110 (Abcam) and against HSP90α/β (loading control) was H-114 (sc-7947, Santa Cruz Biotechnology). Secondary antibody was obtained from Bio-Rad Laboratories. The 7500 Fast Real-Time PCR System from Applied Biosystems was used to measure mRNA levels. mRNA expression levels were normalized to expression of Actin. The following primer sequences were used in the SYBR Green master mix (Roche): Actin_forward, 5’- CCTGGCACCCAGCACAA-3’; Actin_reverse, 5’-GCCGATCCACACGGAGTACT-3’; E2F3_ forward, 5’-GCTGCAGTCTGTCTGAGGAT-3’; E2F3_reverse, 5’-AACAGTTTGAGGTCCAGGGT-3’.

144 | Chapter 6

Clonogenic assay Cells were treated for 24 hours with the indicated concentration of platinumcompounds after which they were seeded in 6-well plates (10×103 cells per well) and in normal medium for 10 days. Cells were washed with PBS, fixed with 4% Statistical analysis Differences in clinical characteristics between patients with luminal-like and basallike MIBC were tested using the Mann-Whitney U test or the Fisher’s exact test. The association between copy number alterations with response to NAC was analyzed using the Fisher’s exact test. We used a Kaplan-Meier analysis to estimate cancerspecific and post-recurrence survival, and group comparisons were made with the use of a log-rank test. Cancer-specific survival was measured from the date of NAC start to the date of death-of-cancer. Post-recurrence survival was measured from the date of disease recurrence to the date of death. Multivariate Cox regression analysis for subtypes was performed with adjustment for age, gender, clinical TNM staging and neoadjuvant chemotherapy regimen. Statistical analyses were performed using SPSS software version 22 (IBM) or GraphPad Prism software version 6. Results were considered statistically significant with P < 0.05 (two-tailed).

ACKNOWLEDGEMENTS We thank the Genomics Core Facility and the Core Facility for Molecular Pathology & Biobanking of our institute for their assistance.

CNVs and subtyping in MIBC correlate with response to neoadjuvant chemotherapy | 145

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

REFERENCES Bandopadhayay, P., Bergthold, G., Nguyen, B., et al., 2014. BET bromodomain inhibition of MYC-amplified medulloblastoma. Clin Cancer Res 20, 912-925. Choi, W., Porten, S., Kim, S., et al., 2014. Identification of distinct basal and luminal subtypes of muscle-invasive bladder cancer with different sensitivities to frontline chemotherapy. Cancer Cell 25, 152-165. Damrauer, J.S., Hoadley, K.A., Chism, D.D., et al., 2014. Intrinsic subtypes of high-grade bladder cancer reflect the hallmarks of breast cancer biology. Proc Natl Acad Sci U S A 111, 3110-3115. DeGregori, J., 2002. The genetics of the E2F family of transcription factors: shared functions and unique roles. Biochim Biophys Acta 1602, 131-150. Delmore, J.E., Issa, G.C., Lemieux, M.E., et al., 2011. BET bromodomain inhibition as a therapeutic strategy to target c-Myc. Cell 146, 904-917. ENCODE Project Consortium, 2012. An integrated encyclopedia of DNA elements in the human genome. Nature 489, 57-74. Feber, A., Clark, J., Goodwin, G., et al., 2004. Amplification and overexpression of E2F3 in human bladder cancer. Oncogene 23, 1627-1630. Hurst, C.D., Tomlinson, D.C., Williams, S.V., et al., 2008. Inactivation of the Rb pathway and overexpression of both isoforms of E2F3 are obligate events in bladder tumours with 6p22 amplification. Oncogene 27, 2716-2727. Ignatiadis, M., Singhal, S.K., Desmedt, C., et al., 2012. Gene modules and response to neoadjuvant chemotherapy in breast cancer subtypes: a pooled analysis. J Clin Oncol 30, 1996-2004. Iyer, G., Al-Ahmadie, H., Schultz, N., et al., 2013. Prevalence and co-occurrence of actionable genomic alterations in high-grade bladder cancer. J Clin Oncol 31, 3133-3140.

146 | Chapter 6

Li, H., Durbin, R., 2009. Fast and accurate short read alignment with Burrows-Wheeler transform. Bioinformatics 25, 1754-1760. Liggett, W.H., Jr., Sidransky, D., 1998. Role of the p16 tumor suppressor gene in cancer. J Clin Oncol 16, 1197-1206. Martinez, L.A., Goluszko, E., Chen, H.Z., et al., 2010. E2F3 is a mediator of DNA damageinduced apoptosis. Mol Cell Biol 30, 524-536. Quinlan, A.R., Hall, I.M., 2010. BEDTools: a flexible suite of utilities for comparing genomic features. Bioinformatics 26, 841-842. Shen, H., Morrison, C.D., Zhang, J., et al., 2013. 6p22.3 amplification as a biomarker and potential therapeutic target of advanced stage bladder cancer. Oncotarget 4, 2124-2134. Silver, D.P., Richardson, A.L., Eklund, A.C., et al., 2010. Efficacy of neoadjuvant Cisplatin in triple-negative breast cancer. J Clin Oncol 28, 1145-1153. The Cancer Genome Atlas Network, 2014. Comprehensive molecular characterization of urothelial bladder carcinoma. Nature 507, 315322. Tordai, A., Wang, J., Andre, F., et al., 2008. Evaluation of biological pathways involved in chemotherapy response in breast cancer. Breast Cancer Res 10, R37. Volkmer, J.P., Sahoo, D., Chin, R.K., et al., 2012. Three differentiation states risk-stratify bladder cancer into distinct subtypes. Proc Natl Acad Sci U S A 109, 2078-2083. Xu, G., Livingston, D.M., Krek, W., 1995. Multiple members of the E2F transcription factor family are the products of oncogenes. Proc Natl Acad Sci U S A 92, 1357-1361.

SUPPLEMENTARY DATA Supplementary Table 1. Genes with recurrent focal CNAs in the TCGA urothelial bladder cancer cohort. In red genes with copy number gains and in blue genes with copy number losses. Gene MYCL1 PVRL4 PPARG E2F3/SOX4 EGFR ZNF703 YWHAZ/PABPC1 MYC CCND1 MDM2/FRS2 PALB2 ERBB2 CCNE1 BCL2L1 LRP1B IKZF2 FHIT FAM190A FOXQ1 CDKN2A RB1 CREBBP NCOR1

Region 1p34.2 1q23.3 3p25.2 6p22.3 7p11.2 8p11.23 8q22.3 8q24.21 11q13.3 12q15 16p12.1 17q12 19q12 20q11.21 2q22.1 2q34 3p14.2 4q22.1 6p25.3 9p21.3 13q14.2 16p13.3 17p12

CNVs and subtyping in MIBC correlate with response to neoadjuvant chemotherapy | 147

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

Chapter 7 General Discussion

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39 150 | Chapter 7

“The future ain’t what it used to be”, Yogi Berra once declared (Berra, 1998). He wasn’t talking about cancer diagnostics and medicine, but he could have been. The evolution of cancer genomics and the introduction of targeted therapies have shifted the paradigms of carcinogenesis and cancer treatment. In this thesis, I have investigated and discussed several aspects of personalized cancer medicine; such as mechanisms of resistance to targeted therapies (chapter 2), combinations of targeted therapies (chapter 3) and identification of molecular subgroups or subtypes (chapter 4 to 6). This last chapter reviews the clinical applicability of tumor genotyping by next-generation sequencing and tumor subtyping by gene-expression profiling to further personalize cancer medicine. Introduction On December 5th 1996, the then-Federal Reserve Board chairman, Alan Greenspan used the words “irrational exuberance” in his speech given at the American Enterprise Institute to express his warning that the stock market was perhaps overvalued (Greenspan, 1995). Immediately after he said this, the stock markets fell sharply. The strong reaction of the markets to Greenspan’s speech was widely noted, and made the term “irrational exuberance” famous. The term is now often used to describe a state of unfounded enthusiasm. I briefly discussed in the General Introduction of this thesis (chapter 1) the rapid development and use of next-generation sequencing (NGS) technology, also known as massively parallel sequencing. Today, there is a lot of excitement about how tumor genotyping by NGS can guide personalized or precision medicine for individual cancer patients in the (near) future. An enormous number of studies and reviews have been published on this topic and several companies are established that profile cancer genotypes in patients and claim to identify clinically “actionable” mutations in up to 76% of the cases (Frampton et al., 2013). But is this excitement realistic or does it also contain an element of “irrational exuberance”? I will discuss in the first part of this chapter some of the issues that, I think, need to be addressed before tumor genotyping by NGS can fulfill its therapeutic promise. In the second part, I argue that genotyping by itself is not the “holy grail” and that we need to integrate tumor genotyping with data from additional “omics” platforms (e.g. gene-expression) and clinical information. Translating tumor genotypes into therapy decisions Next-generation sequencing has been applied successfully in the research setting to highlight and understand the complexity of cancer genomes. With the continuing acceleration in speed and efficiency of NGS technology, it has become relatively easy and affordable to genotype tumors and to identify the genetic abnormalities that

General Discussion | 151

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

are likely to drive individual tumors. Some of these genetic abnormalities can have a major impact on patient outcomes, as illustrated by HER2-amplification in breast cancer, BCR-ABL fusion in chronic myeloid leukemia, EGFR-mutation in lung cancer and BRAFV600E-mutation in melanoma. What they have in common is that, in all cases, highly selective targeted compounds directed to the protein encoded by the driving genetic aberration were available. The improved patient outcomes for cancers with these genetic drivers led to the extrapolation of genotype-directed cancer therapy becoming available for many more cancer patients harboring other genetic drivers. The benefits of this new approach can be significant, since current conventional cancer treatments are, with some exceptions, highly inefficient. On average, they only benefit 25% of the patients while most of the patients experience significant sideeffects (Spear et al., 2001). Several lists are published [e.g. (Iyer et al., 2013; Wagle et al., 2012)] with clinically actionable genetic alterations that match a specific genotype with the appropriate targeted therapy. The number of genes in these lists depends on the definition of actionable, but most of them contain a number of genes that are not directly druggable. These genes are included for their diagnostic, prognostic or therapeutic information instead of being a direct drug target. For example, a closer look at the gene list from the study by Frampton et al., that reported alterations in clinical actionable genes in 76% of the tumors, shows that the top two altered genes are TP53 and KRAS (Frampton et al., 2013). These two genes are both known as undruggable with the current repertoire of compounds, although several inhibitors are in preclinical and early clinical development and other indirect strategies for cancers with mutations in TP53 and RAS have been proposed (Corcoran et al., 2013; Khoo et al., 2014; Stephen et al., 2014; Zimmermann et al., 2013). But without these genes, the frequency of clinically actionable mutations will drop significantly, most likely in clinical practice far below 40-50% of all patients. Furthermore, the outcome of patients treated based on such tests remain largely anecdotal or unknown. And, as discussed in chapter 2, a major limitation of genotype-directed anticancer therapies is the emergence of drug resistance. Resistance to single-agent targeted cancer therapeutics is often the result of reactivation of the signaling pathway. Single-agent treatment directed to single drug targets is unlikely to be sufficient for most patients and an example of oversimplification given the complexity of a cancer genome and cell signaling. We will have to think upfront about powerful and mechanism-based combination treatments to delay or circumvent drug resistance. So despite all the excitement from a biological perspective, not every clinical indication is going to benefit from increased information made available by NGS. The frequency of clinically actionable mutations is limited and there are only a few targeted therapeutic options approved that doctors can choose from. From a clinical

152 | Chapter 7

relevance perspective, it is therefore required to fill the big gap from biological knowledge to new and personalized treatment strategies for patients. Drug target identification, biomarker discovery and drug (combination) development are therefore essential to achieve further progress with precision medicine based on tumor genotyping. In some cases, tumor genotyping can be misleading in guiding precision medicine. Experience to date suggests that the presence of a specific genetic alteration may not confer the same sensitivity to an agent across all cancers types, as exemplified by the clinical observation that BRAFV600E-mutant colorectal cancers do not respond to the BRAFV600E-mutant specific inhibitor, whereas BRAFV600E-mutant melanomas respond very well (Kopetz et al., 2010). Similarly, trastuzumab, has been shown to benefit patients with HER2-amplified breast and gastric cancer, but not those with ovarian or endometrial cancer (Bookman et al., 2003; Fleming et al., 2010). These findings support the conclusion that there are differential, tissue-specific consequences of specific gene mutations. These genotype-phenotype relationships and the underlying biology need to be studied and understood first. In the case of BRAFV600E-mutant colon cancer, the biology explaining the lack of sensitivity for the BRAFV600E-mutant specific inhibitor was elucidated by functional genetic screening (Prahallad et al., 2012). It is critical to explore further whether targeted therapies that are approved for specific cancer types also benefit patients with other types of cancer that share the same genetic aberrations. Obtaining convincing evidence to guide therapy based on genetic alterations require large genomics-based clinical trials with different tumor types. To achieve enough power, especially for driver mutations with a frequency of less than 10%, the total number of patients that needs to be screened for these trials can only be reached through large multi-institutional trial networks. In addition, functional genomics could be used as a tool for the understanding of tissue-specific biology, in order to pre-stratify patients for enrollment into clinical trials. Another very useful approach is to genotype tumors from â&#x20AC;&#x153;exceptional respondersâ&#x20AC;?: patients who experience a complete or partial response for at least 6 months in a trial in which only 1% to 10% of patients respond (definition from the National Cancer Institute). This might lead to the identification of subgroups of patients that respond to the treatment because of a specific tumor genotype. For example, this n-of-1 approach has linked loss-of-function mutations in TSC1 to response to everolimus in bladder cancer (Iyer et al., 2012), identified activating mTOR mutations in a patient with exceptional response to everolimus and pazopanib (Wagle et al., 2014) and linked RAD50 mutation to sensitivity to irinotecan in the context of CHK1 inhibition (AlAhmadie et al., 2014).

General Discussion | 153

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

NGS-guided therapy decisions already impact on the care of patients with cancer in some of the larger academic centers. However, multiple technical innovations and cultural changes are necessary for clinical implementation and decisionmaking in a community hospital setting. This includes rapid and robust DNA sequencing from formalin-fixed paraffin-embedded tumor tissue or, preferably, the switch to collect fresh frozen samples for DNA and RNA isolation to guarantee high quality. Moreover, analytical tools and decision trees need to be developed for clinical interpretation of the data for prospective use in treatment decisions. For example, in the cancer genome, passenger alterations outnumber the driver events and distinguishing the two requires sophisticated bioinformatics. Furthermore, functional prediction and validation of variants of uncertain significance (VUS) is indispensable for clinical use of NGS. Another issue is the requirement of taking multiple and repeated biopsies during the course of the disease, e.g. after tumor relapse or progression. This requires dedicated staff and high-quality facilities, like intervention radiology. This again raises a number of questions that the field will have to address. For instance, how many biopsies are required to probe the heterogeneity of the primary tumor and its metastases? However, most of these issues are related to technical and analytical infrastructures that can be solved by centralizing the wet and dry lab analysis in a specialized center or program, e.g. like the Center for Personalized Cancer Treatment (CPCT) program in The Netherlands. There are nevertheless several issues that are unresolved at the present time that preclude broad implementation of genotype-based treatment. For example, the reimbursement of this type of molecular diagnostics will be a key bottleneck that needs to be resolved. Beyond genotype: intrinsic molecular subtypes Tumor genotyping using NGS does not provide a holistic clinical perspective, since functional changes can result from events other than direct DNA alterations, and other -omics technologies will therefore continue to be relevant. One example is the profiling of tumors for intrinsic molecular subtypes. Tumor subtyping was first introduced for breast cancer where gene expression profiling was used to classify invasive breast cancers into biologically and clinically distinct subtypes that have become known as Luminal A, Luminal B, HER2-enriched and Basal-like (Perou et al., 2000). Subtypes based on gene expression profiling are providing information about the intrinsic behavior of the tumor and, therefore, these subtypes have different clinical outcomes in terms of prognosis and treatment response. Basal-like breast cancers are as different from luminal ER-positive breast cancer as they are from the lung (Hoadley et al., 2014). Unbiased molecular classification of breast cancer into

154 | Chapter 7

clinically relevant subtypes has stimulated the development of tailored treatment plans for each subtype (Higgins and Baselga, 2011). Intrinsic molecular subtypes are now defined for diffuse large B-cell lymphoma (Alizadeh et al., 2000), for head and neck squamous cell carcinoma (Chung et al., 2004), for glioblastoma multiforme (Phillips et al., 2006; Verhaak et al., 2010), for medulloblastoma (Gibson et al., 2010), for pancreatic ductal adenocarcinoma (Collisson et al., 2011), by several studies for colorectal cancer (Budinska et al., 2013; De Sousa et al., 2013; Loboda et al., 2011; Perez-Villamil et al., 2012; Roepman et al., 2014; Sadanandam et al., 2013; Salazar et al., 2011), and recently also for gastric cancer (The Cancer Genome Atlas Network, 2014) and bladder cancer (Choi et al., 2014; Damrauer et al., 2014). These are al examples of â&#x20AC;&#x153;within-a-tissueâ&#x20AC;? subtypes, but there is also evidence that subtypes are, to some extent, independent of tissue of origin: molecular subtypes in bladder cancer recapitulate different aspects of breast cancer biology (Damrauer et al., 2014) and a poor-prognosis subtype enriched for mesenchymal-associated genes is identified in colorectal cancer as well as in bladder cancer, glioblastoma multiforme and pancreatic ductal adenocarcinoma (Collisson et al., 2011; Damrauer et al., 2014; De Sousa et al., 2013; Loboda et al., 2011; Phillips et al., 2006; Roepman et al., 2014; Verhaak et al., 2010). Towards that end, it was recently found by integrative analysis of 12 cancer types from The Cancer Genome Atlas (TCGA) project, that several distinct cancer types converge into common subtypes (Hoadley et al., 2014). Subtypes can overlap with a specific genotype (e.g. HER2-type and HER2amplification in breast cancer), but can also go beyond genotypes. A good example is the identification of a BRAF-mutant-like expression profile in KRAS-mutant or double-wildtype colorectal cancers with a poor prognosis similar to BRAF-mutant colorectal cancers (Popovici et al., 2012). This suggests a shared biology between BRAF-mutant and BRAF-mutant-like tumors, which would not be predicted by the underlying genotype. Besides prognostic value, this could also yield predictive value for therapy response, something that is currently investigated. Similarly, it was found by array comparative genomic hybridization (aCGH) profiling that sporadic breast cancers with a profile similar to that of breast cancers from BRCA1- and BRCA2mutation carriers have significant benefit from platinum-based chemotherapy (Vollebergh et al., 2011). In colorectal cancer, a 64-gene signature has been developed that can predict microsatellite instability (MSI) (Tian et al., 2012). This signature identified a group of MSI-like patients that were microsatellite stable (MSS) by standard assessment. Patients with an MSI-like phenotype had, similar to MSI patients, an improved survival when compared to MSS patients. The MSI-signature

General Discussion | 155

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

is also linked to a deficient DNA mismatch repair phenotype, as MSI and MSIlike patients both showed a relatively high mutation frequency. Finally, predictive expression signatures have also been developed for p53-like tumors in breast cancer and RAS pathway dependency (“RASness”) in multiple tumor types (Guinney et al., 2014; Loboda et al., 2010; Miller et al., 2005). These examples highlight again the importance of understanding tumor biology, rather than focusing exclusively on tumor genotype. So on the one hand, molecular stratification by subtyping will lead to a better understanding of the disease; on the other hand it will help us guide therapy by providing diagnostic, prognostic and predictive information. This has been most extensively studied for intrinsic molecular subtypes in breast cancer (Chia et al., 2012; Krijgsman et al., 2012; Parker et al., 2009; Sorlie et al., 2001; Sorlie et al., 2003). For example, it was found that basal-type breast cancers are significantly more responsive to neoadjuvant chemotherapy as compared to luminal-type breast cancers (Krijgsman et al., 2012; Whitworth et al., 2014). Whithworth et al. reported that molecular subtyping, when compared to conventional immunohistochemistry and FISH, could more accurately identify breast cancer patients that will or will not respond to neoadjuvant chemotherapy (Whitworth et al., 2014). Towards that end, it was demonstrated in chapter 4 of this thesis that molecular subtyping in breast cancer can identify a subgroup of patients that would be misclassified using conventional diagnostics. This subgroup is phenotypically ERα-positive, but functionally ERαnegative (basal-type) as was determined by a molecular subtyping assay. The consequence would be that these tumors do not respond to hormonal therapy, although direct evidence for this is lacking. On the other hand they might benefit from chemotherapy, since basal type breast cancer benefit more from neoadjuvant chemotherapy as compared to luminal type breast cancer (Krijgsman et al., 2012; Whitworth et al., 2014). In colorectal cancer, the epithelial-to-mesenchymal (EMT) signature that overlaps with C-type colorectal cancers is correlated with poor prognosis and chemotherapy resistance (Huang et al., 2012; Loboda et al., 2011; Roepman et al., 2014). In bladder cancer, the luminal-like and basal-like subtypes correlate with different stages of tissue differentiation (Damrauer et al., 2014) and might also stratify patients for antiEGFR therapy, since basal-like tumors display an activation of the EGFR pathway linked to frequent EGFR gains and activation of an EGFR autocrine loop (Rebouissou et al., 2014). It was demonstrated in chapter 6 of this thesis that the basal-like subtype in bladder cancer is less responsive to neoadjuvant chemotherapy compared to the luminal-like subtype.

156 | Chapter 7

Gene signatures have been implemented in the clinic and are used for daily decision making, as illustrated by the commercial MammaPrintâ&#x201E;˘ (Agendia) and OncoType DXâ&#x201E;˘ (Genomic Health, Inc.) tests for early-stage invasive breast cancer. These tests quantify the likelihood of disease recurrence in women with early-stage breast cancer (prognostic information) and assess the likely benefit from chemotherapy (predictive information). However, clinical implementation of cancer subtypes is somewhat behind, for several reasons. First, molecular subtyping by gene expression profiling requires complex analytical tools, which are sometimes also not publicly available. Second, these tests are relatively expensive and are currently not covered by all health insurance companies. Finally, a difficult problem to solve is the construction of different gene signatures with little overlap in gene identity for the same tumor type by different groups, as is seen with breast cancer and colorectal cancer. Although some of these differences might be secondary to differences in patient cohorts, technology and bioinformatics, a consensus gene signature is necessary for clinical acceptance and implementation. It is hopeful that recently a colorectal cancer subtyping consortium was formed to identify a consensus among the different subtyping systems (Dienstmann et al., 2014). Concluding remarks The warning words of Alan Greenspan in 1996 were followed several years later by collapsing stock markets and a worldwide financial crisis. At that time, Greenspanâ&#x20AC;&#x2122;s comment was well remembered, although few heeded the warning. Will a similar scenario apply to the high expectations that cancer genotyping will lead us to precision medicine for cancer patients? Probably not, as cancer is a disease of the genome, but nevertheless we should be careful with the high expectations raised by the initial successes of cancer genotyping by NGS. We have just entered the transition phase: large-scale genomic sequencing is still largely a research tool and the number of direct actionable or predictive genomic alterations to date is rather limited. We have learned that cell type of origin can determine the functional consequence of a genomic alteration. At the molecular level, intrinsic cancer subtypes identified by gene expression profiling might turn out to be as important as the tumor genotype for predicting therapy response and prognosis, especially in tumor types where the driver events are either less understood or currently undruggable. Integration of molecular data with clinical information and treatment response will be essential to explore genotype-phenotype relationships. Understanding of rewiring cell signaling networks, adaptation mechanisms and resulting vulnerabilities will be crucial. This understanding has been and will be the key to success for personalized cancer medicine. In the exciting coming decade(s) we will learn more about how

General Discussion | 157

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

NGS as well as other technologies will contribute to the further development of personalized cancer medicine. I am confident that these technologies will eventually revolutionize cancer diagnostics and clinical cancer care just as it has been, and still is, revolutionizing cancer research. The future of cancer diagnostics ainâ&#x20AC;&#x2122;t what it used to be.

ACKNOWLEDGEMENTS I thank RenĂŠ Bernards, Roderick Beijersbergen and Loredana Vecchione for carefully reading this chapter and providing useful suggestions.

158 | Chapter 7

REFERENCES Al-Ahmadie, H., Iyer, G., Hohl, M., et al., 2014. Synthetic Lethality in ATM-Deficient RAD50Mutant Tumors Underlies Outlier Response to Cancer Therapy. Cancer Discov 4, 1014-1021. Alizadeh, A.A., Eisen, M.B., Davis, R.E., et al., 2000. Distinct types of diffuse large B-cell lymphoma identified by gene expression profiling. Nature 403, 503-511. Berra, Y., 1998. The Yogi Book: “I Really Didn’t Say Everything I Said!”. Workman Publishing Company, Inc. Bookman, M.A., Darcy, K.M., Clarke-Pearson, D., et al., 2003. Evaluation of monoclonal humanized anti-HER2 antibody, trastuzumab, in patients with recurrent or refractory ovarian or primary peritoneal carcinoma with overexpression of HER2: a phase II trial of the Gynecologic Oncology Group. J Clin Oncol 21, 283-290. Budinska, E., Popovici, V., Tejpar, S., et al., 2013. Gene expression patterns unveil a new level of molecular heterogeneity in colorectal cancer. J Pathol 231, 63-76. Chia, S.K., Bramwell, V.H., Tu, D., et al., 2012. A 50-gene intrinsic subtype classifier for prognosis and prediction of benefit from adjuvant tamoxifen. Clin Cancer Res 18, 4465-4472. Choi, W., Porten, S., Kim, S., et al., 2014. Identification of distinct basal and luminal subtypes of muscle-invasive bladder cancer with different sensitivities to frontline chemotherapy. Cancer Cell 25, 152-165.

BCL-XL and MEK inhibition promotes tumor regressions in KRAS mutant cancer models. Cancer Cell 23, 121-128. Damrauer, J.S., Hoadley, K.A., Chism, D.D., et al., 2014. Intrinsic subtypes of high-grade bladder cancer reflect the hallmarks of breast cancer biology. Proc Natl Acad Sci U S A 111, 3110-3115. De Sousa, E.M.F., Wang, X., Jansen, M., et al., 2013. Poor-prognosis colon cancer is defined by a molecularly distinct subtype and develops from serrated precursor lesions. Nat Med 19, 614-618. Dienstmann, R., Guinney, J., Delorenzi, M., et al., 2014. Colorectal Cancer Subtyping Consortium (CRCSC) identification of a consensus of molecular subtypes (ASCO Annual Meeting Abstract). J Clin Oncol 32, abstr 3511. Fleming, G.F., Sill, M.W., Darcy, K.M., et al., 2010. Phase II trial of trastuzumab in women with advanced or recurrent, HER2-positive endometrial carcinoma: a Gynecologic Oncology Group study. Gynecol Oncol 116, 1520. Frampton, G.M., Fichtenholtz, A., Otto, G.A., et al., 2013. Development and validation of a clinical cancer genomic profiling test based on massively parallel DNA sequencing. Nat Biotechnol 31, 1023-1031. Gibson, P., Tong, Y., Robinson, G., et al., 2010. Subtypes of medulloblastoma have distinct developmental origins. Nature 468, 1095-1099.

Chung, C.H., Parker, J.S., Karaca, G., et al., 2004. Molecular classification of head and neck squamous cell carcinomas using patterns of gene expression. Cancer Cell 5, 489-500.

Greenspan, A., 1995. The Challenge of Central Banking in a Democratic Society, http://www.federalreserve.gov/boarddocs/ speeches/1996/19961205.htm.

Collisson, E.A., Sadanandam, A., Olson, P., et al., 2011. Subtypes of pancreatic ductal adenocarcinoma and their differing responses to therapy. Nat Med 17, 500-503.

Guinney, J., Ferte, C., Dry, J., et al., 2014. Modeling RAS phenotype in colorectal cancer uncovers novel molecular traits of RAS dependency and improves prediction of response to targeted agents in patients. Clin Cancer Res 20, 265-272.

Corcoran, R.B., Cheng, K.A., Hata, A.N., et al., 2013. Synthetic lethal interaction of combined

Higgins, M.J., Baselga, J., 2011. Targeted therapies for breast cancer. J Clin Invest 121, 3797-3803.

General Discussion | 159

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

Hoadley, K.A., Yau, C., Wolf, D.M., et al., 2014. Multiplatform Analysis of 12 Cancer Types Reveals Molecular Classification within and across Tissues of Origin. Cell 158, 929-944. Huang, S., Holzel, M., Knijnenburg, T., et al., 2012. MED12 controls the response to multiple cancer drugs through regulation of TGF-beta receptor signaling. Cell 151, 937-950. Iyer, G., Al-Ahmadie, H., Schultz, N., et al., 2013. Prevalence and co-occurrence of actionable genomic alterations in high-grade bladder cancer. J Clin Oncol 31, 3133-3140. Iyer, G., Hanrahan, A.J., Milowsky, M.I., et al., 2012. Genome sequencing identifies a basis for everolimus sensitivity. Science 338, 221. Khoo, K.H., Verma, C.S., Lane, D.P., 2014. Drugging the p53 pathway: understanding the route to clinical efficacy. Nat Rev Drug Discov 13, 217-236. Kopetz, S., Desai, J., Chan, E., et al., 2010. PLX4032 in metastatic colon cancer patients with mutant BRAF tumors. J Clin Oncol 28, abstr 3534. Krijgsman, O., Roepman, P., Zwart, W., et al., 2012. A diagnostic gene profile for molecular subtyping of breast cancer associated with treatment response. Breast Cancer Res Treat 133, 37-47. Loboda, A., Nebozhyn, M., Klinghoffer, R., et al., 2010. A gene expression signature of RAS pathway dependence predicts response to PI3K and RAS pathway inhibitors and expands the population of RAS pathway activated tumors. BMC Med Genomics 3, 26. Loboda, A., Nebozhyn, M.V., Watters, J.W., et al., 2011. EMT is the dominant program in human colon cancer. BMC Med Genomics 4, 9. Miller, L.D., Smeds, J., George, J., et al., 2005. An expression signature for p53 status in human breast cancer predicts mutation status, transcriptional effects, and patient survival. Proc Natl Acad Sci U S A 102, 13550-13555. Parker, J.S., Mullins, M., Cheang, M.C., et al., 2009. Supervised risk predictor of breast cancer based on intrinsic subtypes. J Clin Oncol 27, 1160-1167. 160 | Chapter 7

Perez-Villamil, B., Romera-Lopez, A., Hernandez-Prieto, S., et al., 2012. Colon cancer molecular subtypes identified by expression profiling and associated to stroma, mucinous type and different clinical behavior. BMC Cancer 12, 260. Perou, C.M., Sorlie, T., Eisen, M.B., et al., 2000. Molecular portraits of human breast tumours. Nature 406, 747-752. Phillips, H.S., Kharbanda, S., Chen, R., et al., 2006. Molecular subclasses of high-grade glioma predict prognosis, delineate a pattern of disease progression, and resemble stages in neurogenesis. Cancer Cell 9, 157-173. Popovici, V., Budinska, E., Tejpar, S., et al., 2012. Identification of a poor-prognosis BRAFmutant-like population of patients with colon cancer. J Clin Oncol 30, 1288-1295. Prahallad, A., Sun, C., Huang, S., et al., 2012. Unresponsiveness of colon cancer to BRAF(V600E) inhibition through feedback activation of EGFR. Nature 483, 100-103. Rebouissou, S., Bernard-Pierrot, I., de Reynies, A., et al., 2014. EGFR as a potential therapeutic target for a subset of muscle-invasive bladder cancers presenting a basal-like phenotype. Sci Transl Med 6, 244ra291. Roepman, P., Schlicker, A., Tabernero, J., et al., 2014. Colorectal cancer intrinsic subtypes predict chemotherapy benefit, deficient mismatch repair and epithelial-to-mesenchymal transition. Int J Cancer 134, 552-562. Sadanandam, A., Lyssiotis, C.A., Homicsko, K., et al., 2013. A colorectal cancer classification system that associates cellular phenotype and responses to therapy. Nat Med 19, 619-625. Salazar, R., Roepman, P., Capella, G., et al., 2011. Gene expression signature to improve prognosis prediction of stage II and III colorectal cancer. J Clin Oncol 29, 17-24. Sorlie, T., Perou, C.M., Tibshirani, R., et al., 2001. Gene expression patterns of breast carcinomas distinguish tumor subclasses with clinical implications. Proc Natl Acad Sci U S A 98, 10869-10874.

Sorlie, T., Tibshirani, R., Parker, J., et al., 2003. Repeated observation of breast tumor subtypes in independent gene expression data sets. Proc Natl Acad Sci U S A 100, 8418-8423. Spear, B.B., Heath-Chiozzi, M., Huff, J., 2001. Clinical application of pharmacogenetics. Trends Mol Med 7, 201-204. Stephen, A.G., Esposito, D., Bagni, R.K., et al., 2014. Dragging ras back in the ring. Cancer Cell 25, 272-281. The Cancer Genome Atlas Network, 2014. Comprehensive molecular characterization of gastric adenocarcinoma. Nature 513, 202-209. Tian, S., Roepman, P., Popovici, V., et al., 2012. A robust genomic signature for the detection of colorectal cancer patients with microsatellite instability phenotype and high mutation frequency. J Pathol 228, 586-595. Verhaak, R.G., Hoadley, K.A., Purdom, E., et al., 2010. Integrated genomic analysis identifies clinically relevant subtypes of glioblastoma characterized by abnormalities in PDGFRA, IDH1, EGFR, and NF1. Cancer Cell 17, 98-110.

Vollebergh, M.A., Lips, E.H., Nederlof, P.M., et al., 2011. An aCGH classifier derived from BRCA1-mutated breast cancer and benefit of high-dose platinum-based chemotherapy in HER2-negative breast cancer patients. Ann Oncol 22, 1561-1570. Wagle, N., Berger, M.F., Davis, M.J., et al., 2012. High-throughput detection of actionable genomic alterations in clinical tumor samples by targeted, massively parallel sequencing. Cancer Discov 2, 82-93. Wagle, N., Grabiner, B.C., Van Allen, E.M., et al., 2014. Activating mTOR mutations in a patient with an extraordinary response on a phase I trial of everolimus and pazopanib. Cancer Discov 4, 546-553. Whitworth, P., Stork-Sloots, L., de Snoo, F.A., et al., 2014. Chemosensitivity Predicted by BluePrint 80-Gene Functional Subtype and MammaPrint in the Prospective Neoadjuvant Breast Registry Symphony Trial (NBRST). Ann Surg Oncol 21, 3261-3267. Zimmermann, G., Papke, B., Ismail, S., et al., 2013. Small molecule inhibition of the KRASPDEdelta interaction impairs oncogenic KRAS signalling. Nature 497, 638-642.

General Discussion | 161

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

Addendum Summary Samenvatting Publication list Dankwoord About the author

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39 164 | Summary

SUMMARY Cancer is a genomic disease. Most cancers contain multiple genetics alterations that drive their unrestrained proliferation, progression and metastatic capacity. For most cancer types, it is known what these alterations are and what their frequency is. Genomics technologies have made it possible to identify these genetic alterations on an individual patient level in a short time frame. This is a major breakthrough, as it allows the clinical implementation of genomics-driven personalized or precision medicine. This means that the genomic data are used for the selection of the best treatment strategy for each patient. Another development that fuels this implementation is the growing repertoire of effective cancer therapies targeted against these drivers or driver signaling pathways. However, we know that targeted therapies are only effective in a subgroup of patients and that observed responses are often not durable. In chapter 2, I reviewed the mechanisms of drug resistance to targeted therapies and discussed the lessons that we have learned for future developments. One of these developments is to combine multiple targeted therapies to increase effectiveness and delay or eventually overcome drug resistance. In chapter 3, I identified a novel combination of the multikinase inhibitor sorafenib with the antidiabetic drug metformin that can be used in the treatment of lung cancer. I uncovered the mechanism underlying the combinatorial effect of these drugs and found that these compounds synergistically activate the AMP-activate protein kinase (AMPK) and thereby inhibit mTOR signaling. The subsequent chapters 4 to 6 focus on the identification of molecular subgroups of breast cancer (chapter 4) and bladder cancer (chapter 5 and 6). In chapter 4, I studied a subgroup of ERÎą-positive breast cancers that are classified as basal-type instead of luminal-type. I found that those cancer express relatively high levels of the dominant-negative splice variant ERÎ&#x201D;7. Our finding that the estrogen receptor signaling in these cancers is inactive, suggests that patients with ERÎą-positive basal-type breast cancer may not benefit from estrogen receptor antagonists (e.g. tamoxifen). Furthermore, I found that those patients have a high risk of developing disease recurrence. In chapter 5, I discovered that activating ERBB2 missense mutations characterize a subgroup of muscle-invasive bladder cancer patients with complete response to neoadjuvant chemotherapy. This is important, as chemotherapy is currently the only approved drug therapy for bladder cancer. ERBB2 missense mutations can be used as a genomic biomarker to select patients who will benefit from neoadjuvant chemotherapy. Chapter 6 describes the differential response of luminal-like and basal-like muscleinvasive bladder cancers to neoadjuvant chemotherapy. In addition, I describe in this chapter the identification of specific DNA copy number alterations that correlate

Summary | 165

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

with response to neoadjuvant chemotherapy. Together, the studies described in chapter 4 to 6 contribute to a better selection of patients that will benefit from the drug therapy. This thesis is concluded with a general discussion in chapter 7 on the clinical applicability of cancer genotyping for personalized medicine. This approach has completely changed the future of cancer diagnostics and therapeutics.

166 | Summary

SAMENVATTING Kanker is een genetische ziekte. Kanker is het gevolg van veranderingen in het genetische materiaal van onze cellen, het DNA. Deze veranderingen kunnen ontstaan door externe factoren (bijvoorbeeld tabak en UV-straling), maar ontstaan meestal spontaan wanneer bij het kopiëren van het DNA voorafgaand aan de celdeling een fout wordt gemaakt. Dit is meestal zonder gevolgen, maar wanneer de veranderingen niet meer kunnen worden gerepareerd kan dit het begin zijn van een ontsporing van de cel. De cel krijgt een groeivoordeel die niet meer kan worden afgeremd en er ontstaat massa van cellen, een tumor. Door de snelle vooruitgang van technieken om de veranderingen in het DNA van tumoren te detecteren, is het nu mogelijk om voor iedere kankerpatiënt binnen enkele dagen al deze veranderingen in kaart te brengen. Dit wordt een steeds belangrijker onderdeel in de diagnostiek van kanker, omdat artsen over steeds meer geneesmiddelen beschikken die specifiek aangrijpen op deze veranderingen, de zogenaamde doelgerichte geneesmiddelen. Deze geneesmiddelen remmen de routes die kankercellen gebruiken om te overleven en ongeremd te delen. Helaas werken deze middelen niet bij alle patiënten en daarnaast kan er bij patiënten die initieel wel reageren op de behandeling na enige tijd resistentie (ongevoeligheid) ontstaan tegen deze middelen. De kankercellen hebben zich dan zo aangepast dat ze niet meer gevoelig zijn voor de behandeling en ongeremd kunnen blijven delen. In dit proefschrift heb ik allereerst onderzocht welke mechanismes de cel gebruikt om resistent te worden tegen deze doelgerichte geneesmiddelen. Voor dit literatuuronderzoek, beschreven in hoofdstuk 2, heb ik de eerste doelgerichte geneesmiddelen die op de markt kwamen (voor de behandeling van borstkanker) vergeleken met de doelgerichte geneesmiddelen die de laatste tien jaar beschikbaar zijn gekomen voor de behandeling van kanker in andere weefsels, zoals longkanker en darmkanker. Opvallend is de grote overeenkomst in de manier waarop cellen resistent zijn of worden tegen deze middelen. Dit suggereert dat cellen een aantal voorkeursmechanismes hebben; kankercellen proberen steeds de signaleringsroutes die door deze middelen geremd worden te reactiveren. Als vervolg hierop heb ik in hoofdstuk 2 ook beschreven wat we hiervan kunnen leren en hoe we de therapie zo kunnen aanpassen dat het langer duurt voordat de cellen ongevoelig worden. Dit is mogelijk door niet één doelgericht geneesmiddel, maar een specifieke en slimme combinatie van twee, drie of zelfs nog meer doelgerichte geneesmiddelen tegelijkertijd aan de patiënt te geven. Als we dit combineren met nieuwe technologieën om de tumorcellen en de specifieke genetische veranderingen

Samenvatting | 167

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

continue te monitoren, kunnen we ervoor zorgen dat we de kankercellen een stap voor blijven in plaats van reactief te handelen. Deze geneesmiddelencombinaties kunnen op verschillende manieren worden ontdekt. In hoofdstuk 3 beschrijf ik een nieuwe geneesmiddelencombinatie die ik heb ontdekt door in een klinische longkankerstudie met het doelgericht geneesmiddel sorafenib, specifiek een groep patiënten te bestuderen die goed reageerden op dit geneesmiddel. Het bleek dat dit patiënten waren die naast de longkanker ook diabetes hadden en hiervoor het geneesmiddel metformine gebruikten. Er was al bekend dat metformine een remmend effect op kankercellen kan hebben, maar er was nog niet bekend dat dit effect extra sterk is wanneer het tegelijkertijd met sorafenib wordt gebruikt. Ik heb ontdekt dat dit komt doordat zowel metformine als sorafenib een activerend effect heeft op het eiwit AMPK in de cel en dat de combinatie van deze twee middelen dit effect versterkt. Activatie van AMPK remt allerlei processen in de cel waardoor de cellen minder snel kunnen groeien. Hoofdstuk 4 van dit proefschrift beschrijft de resultaten van een studie waarin met behulp van een speciale diagnostische test (BluePrint, Agendia) een groep borstkankers is geïdentificeerd die wel een specifiek eiwit, de oestrogeen receptor, tot expressie brengen, maar waarin deze receptor niet functioneel is, dat wil zeggen niet ‘aan staat’. Ik heb ontdekt dat deze groep borstkankers, ongeveer 2% van alle borstkankers, een speciale variant van de oestrogeen receptor relatief hoog tot expressie brengt. Deze speciale variant remt juist de activiteit van de normale oestrogeen receptor. Dit is een belangrijke bevinding, omdat veel patiënten worden behandeld met een doelgericht geneesmiddel (tamoxifen) dat specifiek de activiteit van deze oestrogeen receptor remt en dus alleen werkzaam is wanneer de receptor ‘aan staat’. Mijn bevinding suggereert dat deze groep patiënten waarin de receptor niet ‘aan staat’ zeer waarschijnlijk geen baat hebben bij de behandeling met tamoxifen. Daarnaast hebben deze patiënten een hoog risico op het terugkeren van de ziekte. Mogelijk wordt dit risico verlaagd wanneer deze patiënten aanvullend met chemotherapie worden behandeld. Deze studie toont aan dat het niet alleen belangrijk is om naar de expressie van de oestrogeen receptor te kijken, maar dat daarnaast ook naar de functionaliteit van de receptor moet worden gekeken. De ontwikkeling en klinische implementatie van doelgerichte geneesmiddelen gaat snel, maar in de praktijk wordt het merendeel van de patiënten nog steeds behandeld met klassieke chemotherapeutica. Dit geldt zeker voor patiënten met blaaskanker, omdat er in de laatste decennia geen nieuwe middelen zijn geregistreerd voor de behandeling van blaaskanker. Er is een groep patiënten met blaaskanker die heel

168 | Samenvatting

goed reageert op chemotherapie en, in combinatie met chirurgische behandeling, zelfs kan worden genezen. Aan de andere kant is er ook een groep die helemaal niet reageert op chemotherapie. Er waren nog geen mogelijkheden om per patiënt te voorspellen wat de reactie op de chemotherapie zou zijn. In hoofdstuk 5 worden de resultaten beschreven van een studie waarin bij een groep patiënten die heel goed reageert op chemotherapie en bij een groep patiënten die niet reageert op chemotherapie de veranderingen (mutaties) in het DNA van 178 belangrijke genen in kaart zijn gebracht. Een gen is een specifiek stuk DNA wat instructies bevat voor de cellen om een bepaald eiwit te produceren. In deze studie heb ik ontdekt dat alle blaaskankerpatiënten met een mutatie in het gen ERBB2 een volledige respons hadden op chemotherapie voorafgaand aan de operatie. Dit is een zeer belangrijke ontdekking omdat nog niet alle centra in de wereld chemotherapie geven aan deze groep patiënten voorafgaand aan de operatie. In aanvulling hierop heb ik gekeken naar een ander soort verandering in het DNA van blaaskankers, namelijk een toename van het aantal kopieën van elk gen in het DNA. Normaal gesproken heeft elk gen twee kopieën per cel, maar kankercellen hebben vaak afwijkingen in het aantal kopieën van specifieke genen. Dit kan de kankercel bijvoorbeeld extra groeivoordeel opleveren. In hoofdstuk 6 beschrijven we dat veranderingen in het aantal kopieën van specifieke genen vaker gevonden wordt bij patiënten die goed reageren op chemotherapie en andere veranderingen in het aantal kopieën die juist vaker voorkomen in de blaastumoren van patiënten die niet reageren op de chemotherapie. Ook dit kan worden gebruikt om op basis hiervan te bepalen welke patiënten het meeste baat zullen hebben bij de behandeling met chemotherapie. In hoofdstuk 6 wordt ook beschreven dat blaaskanker in te delen is in twee groepen, die naast een verschil in overleving ook verschillend reageren op chemotherapie. Centraal in dit proefschrift staat de verandering van ons denken over de diagnostiek en de behandeling van kanker. In de toekomst zal er steeds meer gebruik worden gemaakt van nieuwe technologieën om veranderingen in het DNA van tumoren te identificeren en tumoren hiermee op een nieuwe manier te classificeren. Mijn onderzoek heeft ertoe bijgedragen dat we beter begrijpen welke patiënt voor welke behandeling in aanmerking komt. Dit zal ertoe leiden dat de behandeling van kanker steeds preciezer en nog meer op het individu gericht wordt. Het laatste hoofdstuk (hoofdstuk 7) bediscussieerd de uitdagingen voor het op grote schaal invoeren van deze veelbelovende aanpak. De toekomst van de diagnostiek en behandeling van kanker is hierdoor radicaal veranderd.

Samenvatting | 169

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

PUBLICATION LIST Groenendijk FH, de Jong J, Fransen van de Putte EE, Michaut M, Schlicker A, Peters D, Velds A, Nieuwland M, van den Heuvel MM, Kerkhoven R, Wessels LF, Broeks A, van Rhijn BWG, Bernards R, van der Heijden MS. ERBB2 mutations characterize a subgroup of muscle-invasive bladder cancers with complete response to neoadjuvant chemotherapy. Submitted for publication. Groenendijk FH, Mellema WW, van der Burg E, Schut E, Jonkers J, van den Heuvel M, Smit EF, Bernards R. Sorafenib synergizes with metformin in NSCLC through AMPK pathway activation. Int J Cancer 2014 Aug 1. doi:10.1002/ijc.29113 [Epub ahead of print]. Groenendijk FH, Bernards R. Drug resistance to targeted therapies: déjà vu all over again. Mol Oncol. 2014;8:1067-1083. Groenendijk FH, Zwart W, Floore A, Akbari S, Bernards R. Estrogen receptor splice variants as a potential source of false-positive estrogen receptor status in breast cancer diagnostics. Breast Cancer Res Treat. 2013;140:475-84. Groenendijk FH, Bernards R. Rational combinations of targeted therapies. Ned Tijdschr Oncol 2013;10:25-31. Groenendijk FH, Beijersbergen RL. KRAS mutatie als marker voor prognose en behandeling: een klinische uitdaging. Ned Tijdschr Targeted Therapy 2013;3:2-6. Huang S, Hölzel M, Knijnenburg T, Schlicker A, Roepman P, McDermott U, Garnett M, Grernrum W, Sun C, Prahallad A, Groenendijk FH, Mittempergher L, Nijkamp W, Neefjes J, Salazar R, ten Dijke P, Uramoto H, Tanaka F, Beijersbergen RL, Wessels LF, Bernards R (2011). MED12 controls the response to multiple targeted cancer drugs through regulation of TGFb receptor signaling. Cell 2012;151:937-50. Groenendijk FH, Taal W, Dubbink HJ, Haarloo CR, Kouwenhoven MC, van den Bent MJ, Kros JM, Dinjens WN. TP53 mutation is early and consistent in astrocytomas and not correlated with MGMT promoter hypermethylation – a longitudinal study including primary and recurrent glioma samples. J Neurooncol. 2011;101:405-17.

170 | Publication list

Taal W, Dubbink HJ, Zonnenberg CB, Zonnenberg BA, Postma TJ, Gijtenbeek JM, Boogerd W, Groenendijk FH, Kros JM, Kouwenhoven MC, van Marion R, van Heuvel I, van der Holt B, Bromberg JE, Sillevis Smitt PA, Dinjens WN, van den Bent MJ; Dutch Society for Neuro-Oncology. First-line temozolomide chemotherapy in progressive low-grade astrocytomas after radiotherapy: molecular characteristics in relation to response. Neuro Oncol. 2011;13:235-41. Dubbink HJ, Taal W, van Marion R, Kros JM, Van Heuvel I, Bromberg JEC, Zonnenberg BA, Zonnenberg CBL, Postma TJ, Gytenbeek JMN, Boogerd W, Groenendijk FH, Smit PS, Dinjens WNM, van den Bent MJ. IDH-1 mutations in LGA predict survival but not response to TMS. Neurology 2009;73:1792-5.

Publication list | 171

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39 172 | Dankwoord

DANKWOORD Het schrijven van een dankwoord voor je proefschrift is een mooi moment om terug te kijken op je promotietijd met de wetenschap dat je er (bijna) bent! Hoewel het alleen mijn naam is die op de kaft prijkt, had ik dit proefschrift niet kunnen schrijven zonder de bijdrage van vele anderen. Ik wil daarom beginnen met iedereen die, direct of indirect, heeft bijgedragen aan de totstandkoming van dit proefschrift te bedanken voor de inspiratie, begeleiding, enthousiasme, steun en kritische blik tijdens deze ontzettend boeiende promotietijd! Een aantal mensen wil ik in het bijzonder bedanken. Allereerst mijn promotor en begeleider prof.dr. René Bernards. Beste René, als er iemand is die kan zorgen voor een enerverend promotietraject dan ben jij het wel. Altijd vol met energie, verhalen, ideeën en creatieve oplossingen. Je geeft je mensen alle vrijheid en verantwoordelijkheid, maar toch weet je haarscherp waar iedereen mee bezig is. Ik ben nog steeds onder de indruk hoe jij bepaalde projecten ineens weet te versnellen, te verbreden of te verdiepen. Ik heb veel van je kunnen en mogen leren over de wereld binnen en buiten het lab. Ik ben je dankbaar dat je mij als arts de mogelijkheid hebt gegeven om mijn promotie te verrichten in een fantastisch, veelzijdig, translationeel onderzoekslaboratorium. Ik kan met recht iedere arts aanraden om een deel van zijn/haar carrière door te brengen in een soortgelijk lab! In mijn visie is het zelfs essentieel om als toekomstig patholoog soortgelijke ervaring te bezitten, maar dat terzijde. In de laatste fase van mijn promotie heb ik veel mogen samenwerken met Michiel van der Heijden. Michiel, bedankt voor je duidelijke supervisie en sturende rol! Ik hoop dat we onze output weten om te zetten in een mooie publicatie! Veel succes met het verder opstarten van je lab. Roderick, bedankt voor de vele discussies en interessante gezichtspunten. Niet in de laatste plaats omdat vele daarvan over fietsen gingen. Dat we nooit samen hebben gefietst heeft natuurlijk alles te maken met mijn angst om door jou uit het wiel te worden gereden. Ik wil in het bijzonder de leden van de leescommissie (prof.dr. Hans Bos, prof.dr. René Medema, prof.dr. Jan Schellens, prof.dr. Stefan Sleijfer en prof.dr. Elsken van der Wall) bedanken voor het lezen en het wetenschappelijk beoordelen van dit proefschrift. Ook speciale dank aan mijn OIO-begeleidingscommissie bestaande uit Thijn Brummelkamp, Bas van Steensel en Rob Wolthuis. Dank voor de prettige jaarlijkse voortgangsgesprekken en het vertrouwen dat jullie aan mij gaven.

Dankwoord | 173

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

Of course I would like to thank all my (ex-) collegues from the Bernards and Beijersbergen groups and from B7 in general. While writing this acknowledgment, I doubted if I had to name everyone seperately, but the list was obviously too long. I hope to thank everyone in person in the coming time. In summary, we made a lot of good fun and science together! Thanks for all the suggestions, help and support. Keep up the good science, the collaborations, the social contacts and the support of each other! Special thanks to my roommates and paranymphs Lorenza and Diede. I really enjoyed the years together and Iâ&#x20AC;&#x2122;m grateful that both of you will support me on the day of the defense. Ik wil graag Anne van Harten bedanken die ik heb mogen begeleiden tijdens een stage in ons lab. Jammer dat onze screen met metformin en salicylate niet in dit proefschrift is beland, maar dat neemt niet weg dat je super werk hebt verricht. Veel succes met je promotie, ik heb daar alle vertrouwen in! Dit proefschrift was er nooit gekomen zonder de samenwerking met alle anderen binnen en buiten het NKI. Voor het hoofdstuk over de combinatie van sorafenib en metformine waren dit Jos Jonkers en in zijn lab Eline van der Burg en Eva Schut. Jullie hebben enige tijd intensief de muizenstudie uitgevoerd. Dit was een bijzonder prettige samenwerking, dank daarvoor! Daarnaast waren Egbert Smit en Wouter Mellema van het VuMC, samen met Michel van den Heuvel van het AvL, verantwoordelijk voor de klinische observatie wat aanleiding gaf voor deze studie. Dank voor deze bed-to-bench benadering en jullie bijdrage! Voor het hoofdstuk over de ER splice variant waren dit in het bijzonder Wilbert Zwart en verschillende mensen bij Agendia. De hoofdstukken over blaaskanker zijn het resultaat van de multidisciplinaire blaaskanker onderzoeksgroep in het AvL met als voortrekkers Bas van Rhijn, Michiel van der Heijden, Jeroen de Jong en Annegien Broeks. Ik hoop en verwacht dat deze samenwerking nog veel gaat opleveren! Ik wil Elies Fransen van de Putte en Laura Mertens bedanken voor alle klinische data en de samenwerking op deze projecten. Het is geen geheim dat de hoge kwaliteit van het onderzoek binnen het NKI mogelijk wordt gemaakt door de uitstekende core facilities die de wetenschappers ondersteunen bij alle experimenten. Ik heb in het bijzonder veelvuldig een beroep gedaan op de sequencing facility, de cryobase, de genomics facility en de moleculaire pathologie & biobanking. Dank voor al jullie inzet en passie!

174 | Dankwoord

Ik wil Winand Dinjens en professor Max Kros van de afdeling Pathologie van het ErasmusMC bedanken. Het is mede dankzij hun stimulerende begeleiding tijdens mijn onderzoeksstage in het ErasmusMC (2007/2008) dat ik absoluut verder wilde met onderzoek na mijn geneeskunde studie. Ik hoop dat jullie met veel plezier het proefschrift zullen lezen en ik zie uit naar de tijd in het ErasmusMC. Lieve familie en vrienden, bedankt voor al jullie betrokkenheid en interesse! Lieve pa en ma, zonder al jullie liefde en steun was dit boekje er nooit gekomen. Jullie hebben mij altijd de vrijheid en mogelijkheden gegeven om eigen keuzes te maken en ambities waar te maken. Ik ben jullie hier voor altijd dankbaar voor. Pa, jij weet als geen ander wat het is om te strijden tegen kanker. Wat ben ik dankbaar dat we dit met elkaar mogen meemaken! Het zou fantastisch zijn als wat in dit boekje is beschreven ook maar iets kan betekenen voor al die anderen die strijden tegen kanker. Broers en zus, wat bijzonder dat we alle vijf onze eigen richting op zijn gegaan en toch dicht bij elkaar zijn gebleven. Jaap, ik hoop ook snel jouw boekwerk in ontvangst te mogen nemen. Respect dat je dit hebt gedaan! Theo, op latere leeftijd nog een tweede studie afronden, dat vraagt om een feestje! Alie, ook jij hebt het gepresteerd om twee studies af te ronden om voor de klas te kunnen staan. AndrĂŠ, ik heb altijd hoge verwachtingen van je gehad (misschien ook je eigen schuld?) en tot dusver maak je het helemaal waar. Met recht een klassiek intellectueel! Trots op jullie allemaal! En niet te vergeten mijn lieve schoonzussen, zwager en schoonfamilie. Ook al was het niet altijd even begrijpelijk wat ik toch in dat lab deed, jullie waren altijd zeer geĂŻnteresseerd en oprecht betrokken. Bedankt daarvoor! Tot slot, Eva, wil ik jou bedanken als mijn grote liefde. Als er iemand is die altijd achter me staat dan ben jij het wel! Ik hoop samen met jou nog heel veel mooie en gezonde jaren te beleven. Dit was uitdaging!

het.

Het was fantastisch, maar het is tijd voor de volgende

Floris

Dankwoord | 175

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39 176 | About the author

ABOUT THE AUTHOR Floor Hendrik (Floris) Groenendijk was born in Alblasserdam on June 29, 1986. He graduated in 2004 from the Lage Waard in Papendrecht after which he attended medical school at the University of Maastricht. Floris earned his bachelor’s degree with honor in 2007. He was selected for the Honors Research Program from the Faculty of Medicine and did a six-month research internship within the Josphine Nefkens Institute at the Erasmus Medical Center in Rotterdam, under supervision of prof.dr. J.M. Kros and dr. W.N.M. Dinjens. During that research internship, he studied molecular alterations in the progression of low-grade astrocytomas (brain tumors). Floris completed his master’s in medicine with a six-month internship within the department of Clinical Pathology at the Atrium Medical Center in Heerlen. Floris received his medical degree from the University of Maastricht in 2010. In the same year, he joined as a PhD-student the lab of prof.dr. René Bernards at the Netherlands Cancer Institute in Amsterdam. The results of his research are presented in this thesis. After obtaining his PhD, Floris will start his clinical pathology residency program at the Erasmus Medical Center in Rotterdam.

A About the author | 177

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19 R20 R21 R22 R23 R24 R25 R26 R27 R28 R29 R30 R31 R32 R33 R34 R35 R36 R37 R38 R39

Cancer Diagnostics: The Future Ain't What It Used to Be

UITNODIGING voor het bijwonen van de openbare verdediging van het proefschrift van

Floris Groenendijk Woensdag 11 februari 2015 om 14.30 uur in het Academiegebouw Domplein 29 te Utrecht Aansluitend receptie ter plaatse Paranimfen: Lorenza Mittempergher l.mittempergher@nki.nl 06-24916231 Diede Brunen d.brunen@nki.nl 06-39689388

Floris Groenendijk

Floris Groenendijk Kloosterstraat 33 2021VK Haarlem f.groenendijk@nki.nl 06-41936936