sign in sign up
<<
Home , Elesclomol

Development and Application of High-Throughput Chemical Genomic Screens for Functional Studies of Therapeutics

by

Kahlin Cheung-Ong

A thesis submitted in conformity with the requirements for the degree of Doctor of Philosophy Department of Molecular Genetics University of Toronto

© Copyright by Kahlin Cheung-Ong 2013

Development and Application of High-Throughput Chemical Genomic Screens for Functional Studies of Cancer Therapeutics

Kahlin Cheung-Ong

Doctor of Philosophy

Department of Molecular Genetics University of Toronto

2013 Abstract

Chemotherapeutic agents act by targeting rapidly dividing cancer cells. The full extent of their cellular mechanisms, which is essential to balance efficacy and toxicity, is often unclear. In addition, the use of many anticancer drugs is limited by dose-limiting toxicities as well as the development of drug resistance. The work presented in this thesis aims to address the basic biology that underlies these issues through the development and application of chemical genomic tools to probe mechanisms of current and novel anticancer compounds. Chemical genomic screens in the yeast Saccharomyces cerevisiae have been used to successfully identify targets and pathways related to a compound‟s mode of action. I applied these screens to examine the mode of action of potential anticancer drugs: a class of platinum-acridine compounds and the -inducing compound elesclomol. By analogy to the yeast screens, I developed an

RNAi-mediated chemical genomic screen in human cells which has the potential to reveal novel targets and drug mechanisms. This screen was applied to further understand ‟s mode of action. In parallel with the loss-of-function assays, our lab developed a human ORF overexpression screen in human cells. I applied this gain-of-function screen to identify those genes that, when overexpressed, are toxic to cells. Characterization of such genes that cause toxicity can provide insight into human diseases where gene amplification is prevalent.

ii

Acknowledgements

First and foremost, I would like to thank my supervisors, Corey Nislow and Guri Giaever, for their guidance and support throughout my Ph.D. Thank for you all your encouragement, wise words, and contagious enthusiasm for science. I could not have asked for better mentors. I would also like to thank my committee members Jason Moffat and Dan Durocher for excellent advice and guidance on my research.

Thank you to everyone in the lab for your support and friendship over the years. To Elke Ericson, thanks for taking me under your wing during the early years of my graduate career. To Anthony Arnoldo and Jing Kittanakom, thanks for your friendship, support, and magic protocols. To Ron Ammar, Simon Alfred, and Nikko Torres, thanks for the helpful discussions and constant laughs. Thanks to Anna Lee, Daniel Shabtai, Kevin Song, and Larry Heisler working with me on the platinum project. To Marinella Gebbia, Andrew Smith, Malene Urbanus, and Elena Lissina, thanks for answering all my questions about yeast. Also thanks to the Moffat, Brown, and Stagljar labs for sharing advice, protocols, and reagents.

I would like to thank my friends for putting up with me and helping me keep my sanity. Finally, I would like to thank my family for their unconditional love and support during these years, even though you never really understood what I was doing.

iii

Table of Contents

Acknowledgements ...... iii

Table of Contents ...... iv

List of Tables ...... ix

List of Figures ...... x

List of Abbreviations ...... xii

Chapter 1 - Introduction ...... 1

1.1 Cancer ...... 1

1.1.1 DNA-damaging agents as cancer chemotherapeutics ...... 1

1.1.1.1 DNA-reactive agents ...... 3

1.1.1.2 ...... 5

1.1.1.3 Topoisomerase poisons ...... 5

1.1.2 Mitochondria as a target for cancer chemotherapy ...... 7

1.1.3 Limitations of current chemotherapeutics ...... 8

1.1.4 New designs for chemotherapeutics ...... 10

1.2 Chemical genomics in model organisms ...... 11

1.2.1 Importance of screening to understand drug mechanisms of action ...... 11

1.2.2 Chemical genomic screens in yeast ...... 13

1.2.2.1 Haploinsufficiency profiling ...... 15

1.2.2.2 Homozygous profiling ...... 16

1.2.2.3 Multicopy suppression profiling ...... 16

1.3 Functional genomics in mammalian cells ...... 17

1.3.1 Loss-of-function screens ...... 18

1.3.1.1 RNA interference ...... 18

1.3.1.2 Using RNAi for genome-scale loss-of-function screens ...... 20

iv

1.3.2 Gain-of-function screens ...... 22

1.3.2.1 Examples of systematic gain-of-function mammalian screens ...... 22

1.3.2.2 Downstream applications of overexpression studies ...... 24

1.4 Project rationale ...... 24

Chapter 2 - Comparative chemogenomics to examine the mechanism of action of DNA- targeted platinum-acridine anticancer agents ...... 26

2.1 Introduction ...... 26

2.2 Results ...... 29

2.2.1 Global analysis of fitness profiles ...... 29

2.2.2 Specific genes involved in DNA-damage response to these novel platinum compounds ...... 31

2.2.3 Genome-wide profiling of DNA-damaging platinum-acridines in S. pombe ...... 39

2.2.4 Effect of platinum-acridines on DNA-replication ...... 39

2.2.5 Effect of platinum-acridines on mitochondria ...... 40

2.3 Discussion ...... 45

2.4 Methods ...... 47

2.4.1 Reagents ...... 47

2.4.2 Yeast strains and media ...... 47

2.4.3 Deletion pool growth and chip experiments ...... 48

2.4.4 Data analysis ...... 48

2.4.5 Comparing genome-wide profiles ...... 48

2.4.6 Gene set enrichment analysis (GSEA) ...... 49

2.4.7 DNA content analysis ...... 50

2.4.8 Molecular combing ...... 50

2.4.9 Microscopy ...... 51

Chapter 3 - Mitochondrial electron transport is the cellular target of the oncology drug elesclomol ...... 53

v

3.1 Introduction ...... 53

3.2 Results ...... 56

3.2.1 Yeast are sensitive to elesclomol treatment only in the presence of copper ...... 56

3.2.2 There is no cellular protein target for elesclomol ...... 56

3.2.3 Elesclomol targets electron transport activity in the mitochondrion ...... 57

3.2.4 Elesclomol works by a distinct mechanism of action ...... 61

3.2.5 Elesclomol interacts similarly with the ETC in human cells ...... 61

3.2.6 Human cells lacking mitochondrial DNA are sensitive to elesclomol ...... 62

3.3 Discussion ...... 62

3.4 Materials and Methods ...... 68

3.4.1 Reagents ...... 68

3.4.2 Yeast Strains ...... 68

3.4.3 Minimum Inhibitory Concentration (MIC) and Cidality Determination ...... 69

3.4.4 Deletion Pool Growth and Chip Experiments ...... 69

3.4.5 Analysis of Elesclomol Sensitivity in Yeast Deletion Mutants ...... 70

3.4.6 Gene Set Enrichment Analysis (GSEA) ...... 70

3.4.7 Multiple Drug Effect Analysis ...... 70

3.4.8 Analysis of ρ0 cell lines ...... 71

Chapter 4 - A high-throughput chemogenomic loss-of-function screen to examine doxorubicin mode-of-action in human cells ...... 72

4.1 Introduction ...... 72

4.1.1 Development of a large-scale shRNA-based chemogenomic assay ...... 73

4.1.2 Doxorubicin ...... 75

4.2 Results ...... 76

4.2.1 Optimization of large-scale shRNA-based chemogenomic screens ...... 76

4.2.2 Screen for potential doxorubicin targets and analysis of results ...... 79

vi

4.2.3 Validation of candidates ...... 81

4.2.4 Doxorubicin modulates histone H3 acetylation via MYST4 ...... 87

4.3 Discussion ...... 87

4.4 Methods ...... 91

4.4.1 TRC library and pools ...... 91

4.4.2 Cell culture ...... 91

4.4.3 Individual shRNA knockdowns ...... 92

4.4.4 Screens ...... 92

4.4.5 Individual hit confirmations ...... 94

4.4.6 Immunoblots ...... 94

Chapter 5 - A systematic analysis of human genes that adversely affect fitness when overexpressed ...... 95

5.1 Introduction ...... 95

5.2 Results ...... 97

5.2.1 A systematic genome-wide human ORF overexpression screen ...... 97

5.2.2 Screen for toxic overexpressed genes in HEK293M2 ...... 100

5.2.3 Analysis of the screen results ...... 100

5.2.4 Confirmed hits ...... 101

5.3 Discussion ...... 101

5.4 Methods ...... 106

5.4.1 Cell culture ...... 106

5.4.2 Production of hORFeome overexpression pool ...... 106

5.4.3 Screen for toxic mutants ...... 107

5.4.4 Analysis of screen results ...... 108

5.4.5 Confirmation of toxic hits ...... 108

Chapter 6 - Summary and Perspectives ...... 109

vii

6.1 Summary ...... 109

6.2 Perspectives ...... 110

6.2.1 Chemical genomics ...... 110

6.2.2 Genome-wide screens in mammalian systems ...... 112

7 References ...... 114

viii

List of Tables

Table 2-1. Growth responses of strains deleted for DDR genes in the top 20 individual mutant strains to compound treatment...... 35

Table 2-2. Confirmations of individual mutant strain sensitivity to compound treatment using on solid media dilution spotting...... 36

Table 4-3. TRC lentiviral shRNA clones used in this study...... 93

Table 5-1. Lethal gene set from the overexpression toxicity screen in HEK293-M2s...... 104

Table 5-2. Results from confirmation studies of toxic genes when overexpressed in HEK293-M2 cells...... 105

ix

List of Figures

Figure 1-1. Structures of selected DNA-damaging anticancer compounds...... 4

Figure 1-2. Yeast chemical genomic screens...... 14

Figure 1-3. RNAi pathway in mammalian cells...... 19

Figure 2-1. Chemical structures of the platinum-acridine compounds and carriers screened...... 28

Figure 2-2. Two-dimensional hierarchical clustering of all 14 compounds based on the profile similarity scores of their genome-wide S. cerevisiae profiles...... 30

Figure 2-3. S. cerevisiae and S. pombe profiles for the platinum-acridine compounds that require DNA-damage response pathways...... 32

Figure 2-4. Relative importance of DNA-damage repair modules in the resistance to the DNA- damaging platinum compounds in S. cerevisiae and S. pombe...... 33

Figure 2-5. Gene set enrichment analysis for DNA-damaging platinum-acridines...... 37

Figure 2-6. DNA content analysis profiles...... 41

Figure 2-7. Microscopy confirms G2/M arrest of cells treated with PT-AMIDINE(EN)...... 42

Figure 2-8. PT-AMIDINE(EN) treated cells show defects in replication fork progression by DNA combing...... 43

Figure 2-9. Mitochondria are disrupted by platinum-acridine compounds...... 44

Figure 2-10. Growth responses of the yeast deletion pools to compound...... 52

Figure 3-1. Elesclomol-induced ROS generation and cytoxicity in yeast is dependent on the presence of copper...... 55

Figure 3-2. Sensitivity of S. cerevisiae mutant strains to elesclomol-Cu...... 58

x

Figure 3-3. Biological processes and protein complexes associated with sensitivity to elesclomol...... 60

Figure 3-4. Combinations of elesclomol-Cu and ETC complex inhibitors in cells .... 63

Figure 3-5. Mitochondrial DNA depleted (ρ0) cells are hypersensitive to elesclomol...... 64

Figure 4-1. The RNAi-mediated chemogenomic assay...... 74

Figure 4-2. The representation of hairpins in the pool does not decrease after freezing and thawing cells...... 78

Figure 4-3. A549 cells containing shRNAs against RAD50 exhibit increased sensitivity to doxorubicin...... 78

Figure 4-4. Chemical genomic profile of sodium fluoride (NaF)...... 80

Figure 4-5. Functional enrichment of the top 2% hits from doxorubicin screen...... 82

Figure 4-6. Viability of individual knockdown cells after treatment with doxorubicin...... 86

Figure 4-7. Doxorubicin treatment increases acetylation of histone H3 in A549 cells ...... 88

Figure 4-8. MYST4 knockdown alters the effect of doxorubicin on histone H3 acetylation...... 88

Figure 5-1. The hORFeome overexpression assay...... 98

Figure 5-2. Lentivirus vector map for pLD-T-IRES-Venus-WPRE-STOP...... 99

Figure 5-3. Schematic of confirmation experiments ...... 103

xi

List of Abbreviations

5-FU 5- cDNA complementary DNA CGH comparative genome hybridization DDR DNA-damage response DMSO dimethyl sulfoxide DNA deoxyribonucleic acid DSB DNA double-strand break esiRNA endoribonuclease-prepared siRNAs ETC electron transport chain FACS fluorescence-activated cell sorting HIP haploinsufficiency profiling HOP homozygous profiling HRR homologous recombination repair LDH lactate dehydrogenase MGC Mammalian Gene Collection MIC minimum inhibitory concentration miRNA microRNA MOMP mitochondrial outer membrane permeabilization MSP multicopy suppression profiling NER nucleotide excision repair NSCLC non-small cell lung cancer ORF open reading frame PCR polymerase chain reaction PRR post-replication repair RNA ribonucleic acid RNAi RNA interference ROS shRNA short hairpin RNA siRNA short interfering RNA SRB sulforhodamine B TLS translesion synthesis TRC The RNAi Consortium TRE tetracycline-responsive element YKO yeast knockout

xii

Chapter 1 Introduction

The material presented in this chapter has been adapted from Cheung-Ong K, Giaever G, and Nislow C. (2013) DNA-Damaging Agents in Cancer Chemotherapy: Serendipity and Chemical Biology. Chem Biol. In press.

1.1 Cancer chemotherapy

1.1.1 DNA-damaging agents as cancer chemotherapeutics

Although considered by some to be a modern disease, cancer in humans has been documented for millennia (for review, see David and Zimmerman, 20101). Currently, cancer accounts for 7-8 million deaths (13% of all deaths) worldwide2. Despite repeated campaigns to defeat cancer, such as Nixon‟s War on Cancer3, all have failed because cancer is not a single disease. In fact, it is a collection of highly complex diseases characterized by unregulated cell proliferation that can arise from contributions from numerous different factors including genetic and environmental.

The treatment of cancer is still largely based on the use of chemotherapeutic drugs to eliminate cancer cells, reduce tumour growth, and alleviate pain. The first widely used cancer drugs were discovered in the 1940s as a result of studying victims of chemical warfare during World Wars I and II (for review, see Chabner and Roberts, 20054). Soldiers exposed to sulfur mustard gas were found to have depleted bone marrow and reduced lymph nodes5. Alfred Gilman and Louis Goodman began testing more stable compounds, such as bis and tris β- chloroethyl amines, and found that they caused tumour regressions in mice with transplanted lymphoid tumours6,7. Next, they treated a patient with late-stage non-Hodgkin‟s lymphoma with tris β-chloroethyl amine and found that the tumour subsided6. Subsequent testing of β- chloroethyl amines in 67 patients with non-Hodgkin‟s lymphoma and leukaemia revealed marked tumour regression8. It was later noted that these remissions were short-lived, with resistance to the compounds developing rapidly; however, the idea that tumours could be cured, 1

if only temporarily, ushered in an era of widespread research into discovering and characterizing cancer therapeutics.

Around the same time (1946-1948), Sidney Farber was investigating the effects of folic acid in leukaemia patients. He discovered that when folic acid was administered to these patients, it led to increased proliferation of acute lymphoblastic leukaemia cells9. Folic acid deficiencies were identified in patients with megaloblastic anaemia and its administration was found to stimulate bone marrow maturation and growth10. Farber‟s observation led to collaborations with Yellapragada Subbarao to develop folate analogues that could chemically block folic acid and hence inhibit the production of abnormal bone marrow associated with leukaemia. This was one of the first examples of rational drug design, as opposed to serendipitous discovery. The first folate analogue, was administered to children with acute lymphoblastic leukaemia and led to successful remissions11. Though the remissions were brief, it was clear that had potential as anticancer compounds. Another folate analogue with less toxicity than aminopterin, (amethopterin), was one of the first drugs to cure a solid tumour (choriocarcinoma) in the 1950s12,13.

It took a decade to identify what these two compounds had in common: both nitrogen mustards and folate antagonists are effective at killing cancer cells due to their DNA-active properties. Nitrogen mustards directly alkylate DNA on purine bases leading to stalled replication fork progression and subsequent cell death via apoptosis. Aminopterin and methotrexate act further upstream in the DNA synthesis pathway: by binding to dihydrofolate reductase (DHFR), these antifolates block the synthesis of purine bases thereby preventing DNA synthesis and causing arrest and cell death. DNA integrity is critical for proper cellular function and proliferation. High levels of damage to DNA are detected by cell cycle checkpoint proteins which induce cell cycle arrest to prevent the transmission of damaged DNA during mitosis14. DNA lesions that occur during the S phase of the cell cycle block replication fork progression and can lead to replication-associated DNA double-strand breaks (DSBs) which are among the most toxic of all DNA lesions. If the damaged DNA cannot be properly repaired, cell death may result. Cancer cells typically have relaxed DNA-damage sensing/repair capabilities which allows them to ignore normal checkpoints and achieve high proliferation rates15; this also makes them more susceptible to DNA damage since replicating damaged DNA increases the likelihood of 2

cell death. The concept of aiming at DNA as a target for anticancer drugs inspired the development of numerous anticancer compounds such as , doxorubicin, 5- fluorouracil, and cisplatin16.

1.1.1.1 DNA-reactive agents

In the 1960s and 1970s, there was a surge of interest in developing anticancer compounds which react chemically with DNA. These included compounds that directly modify DNA bases, or intercalate between bases, or form crosslinks in DNA. Derivatives of nitrogen mustards were developed, including the DNA alkylators , , and , all of which are currently used clinical therapeutics. Other examples of DNA-alkylating agents used in cancer treatment include (e.g. , , ) and (e.g. , )17,18. Natural products which alkylate DNA bases were also discovered around this time, such as and streptozotocin19,20. These compounds and several of the alkylators mentioned above crosslink DNA on opposite strands of the double helix (interstrand crosslinks), resulting in a more potent effect against cancer cells compared to monofunctional alkylation. For example, carmustine (N,N′-bis(2-chloroethyl)-) binds to the N1 of guanine on one DNA strand and the N3 of cytosine of the opposite strand to form interstrand crosslinks which block DNA replication and can cause cell death if not repaired21.

The discovery of the alkylating-like platinum agents had a significant positive impact on anticancer drug research. The first platinum-based compound, cis-diamminedichloroplatinum(II) (), was discovered by accident in the 1960s when a magnetic field generated by platinum electrodes was shown to block E. coli cell division22. Cisplatin contains a platinum core with two chloride leaving groups and two amine non-leaving groups (Figure 1-1). After cell entry, aquation of the chloride groups allows the platinum to bind guanine residues, and to a lesser extent adenine residues, to form adducts on DNA. When two platinum adducts are formed on adjacent bases on the same DNA strand, they form intrastrand crosslinks23,24. The structures of these platinum-DNA adducts have been solved at atomic resolution using X-ray crystallography and nuclear magnetic resonance25,26. Inspired by the efficacy of cisplatin, platinum-based analogues have been developed, including and , which also act by forming DNA crosslinks but have different pharmacological properties, decreased side-effects, and

3

Figure 2-1. Structures of selected DNA-damaging anticancer compounds.

4

increased efficacy against different tumours27. In particular, platinum compounds have been very successful in the treatment of solid tumours. Indeed, cisplatin therapy can cure over 90% of all testicular cancer cases and also has good efficacy in the treatment of ovarian, bladder, head and neck, and cervical cancers24. Current efforts to develop cisplatin analogues are aimed at reducing the toxicity of this compound to non-targeted tissues, which results in dose-limiting toxicities such as nephrotoxicity and neurotoxicity. The spectrum of different platinum compounds under development is broad and platinum compounds have also encouraged the synthesis and testing of other metal-containing compounds for use in chemotherapy28-30.

1.1.1.2 Antimetabolites

Antimetabolites represent a class of anticancer drugs that mimic normal cellular molecules and consequently interfere with DNA replication. Many of these compounds are DNA antagonists that exert their activity by blocking nucleotide metabolism pathways. Examples of widely used anticancer compounds include the pyrimidine analogues 5-fluorouracil (5-FU), , , and gemcitabine, and the purine analogues 6-, 8- azaguanine, , , and cladribine31. The incorporation of purine and pyrimidine analogues into DNA during the S phase of the cell cycle prevents proper nucleobase addition, causing DNA replication to fail. For example, 5-FU can be incorporated into DNA and RNA in place of thymine or uracil, respectively. Because 5-FU contains a fluoride atom at the 5- carbon position on the ring, it prevents the addition of the next nucleobase on the strand, therefore terminating chain elongation which induces apoptosis32. In addition to nucleobase analogues, other antimetabolites inhibit enzymes important for DNA synthesis. Methotrexate and newer antifolates such as inhibit the dihydrofolate reductase enzyme to block the synthesis of nucleotides. Another , , directly inhibits thymidylate synthase33. Methotrexate, the primary antifolate used in chemotherapy, displays a broad range of antitumour activities against breast, ovarian, bladder, and head and neck cancers34.

1.1.1.3 Topoisomerase poisons

A mechanistically distinct way to interfere with normal DNA function is to target essential protein-DNA complexes. The assembly of proteins onto DNA is crucial for many DNA processes including transcription, replication, recombination, and repair. Therefore, it is not

5

surprising that many DNA-active drugs act by interfering with DNA-protein binding. Topoisomerases are a class of enzymes responsible for releasing the torsional strain of the DNA double helix. Topoisomerase I allows the passage of a single DNA strand through a transient single-strand break created in the complementary strand of the double helix. Topoisomerase II cuts both strands of the double helix to allow the passage of an intact helix to unwind supercoiled DNA. Topoisomerase poisons trap the DNA-enzyme intermediate as a complex, preventing re- ligation of the break, inhibiting replication forks, and causing toxic DSBs (for review, see Froelich-Ammon and Osheroff, 199535). Initial insights into how these inhibitors work came from plant analogues developed from podophyllotoxin and its derivatives, such as etoposide and , which were found to have antineoplastic effects36. Interestingly, it was found that DNA strand breaks caused by etoposide were not present when etoposide was incubated with purified DNA37. It was soon discovered that etoposide binds to the topoisomerase II-DNA complex38. Cellular levels of topoisomerase II determine the efficacy of etoposide as a cytotoxic agent, with higher levels leading to a greater efficacy of the drug39. This correlation can be used to inform topoisomerase II chemotherapy. Another plant-produced product, , was found to be a topoisomerase I poison40. As with etoposide, camptothecin does not bind the enzyme or DNA alone but rather binds to the DNA-topoisomerase complex to inhibit strand re- ligation41.

The antibiotics are a group of antineoplastic agents which, like etoposide, poison topoisomerase II but they also have additional antitumour mechanisms such as the ability to intercalate into DNA42. Doxorubicin and (Figure 1-1), the first of these compounds (derived from Streptomyces peucetius), are currently used to treat breast cancer, small-cell lung tumours, soft tissue sarcomas and lymphomas, and acute lymphoblastic or myeloblastic leukaemias43. These compounds and the newer , , and have become mainstays of cancer chemotherapy. Anthracyclines are extremely toxic, likely because of their multiple mechanisms of action in addition to binding the DNA-topoisomerase complex. Anthracyclines are also able to intercalate into DNA, generate free radicals, bind and alkylate DNA, crosslink DNA, interfere with helicase activity, and induce apoptosis (for review, see Minotti et al., 200543). Despite their key roles in cancer treatment, anthracyclines are also

6

associated with toxic side effects, especially cardiotoxicity (for review, see Olson and Mushlin, 199044).

1.1.2 Mitochondria as a target for cancer chemotherapy

Mitochondria have recently emerged as potential targets for cancer chemotherapy. As the cell‟s main energy producers, mitochondria are vital for cellular bioenergetics, ion transport, protein import, and energy conservation. Mitochondria also play a key role in cell death; they control the activation of the intrinsic pathway of apoptosis and are involved in non-apoptotic cell death pathways45.

A role for mitochondria in cancer biology was first described in the 1920s when it was observed that cancer cells, in the presence of oxygen, synthesize ATP through aerobic glycolysis, which requires high glucose uptake and local acidification46. This phenomenon, known as the Warburg effect, was the first indication that cancer cells have different cellular energetics and metabolism than normal cells. These changes in metabolism include unregulated glucose uptake, defects in mitochondrial respiration, and, importantly, the ability of the cell to evade apoptosis which is one of the hallmarks of cancer cells47,48. This extensive metabolic reprogramming of tumour cells also makes them more susceptible to mitochondrial perturbations. In addition, polymorphisms in mitochondrial DNA have been described as influencing the risk of developing certain such as breast and prostate cancer49,50.

Cancer-specific alterations to mitochondrial function represent attractive targets for the development of novel anticancer drugs. Current approaches used for targeting mitochondria in cancer include: stimulating mitochondrial outer membrane permeabilization (MOMP) which is often decreased in cancer cells but leads to apoptosis when activated, targeting mitochondrial metabolism by inhibiting proteins involved in glycolysis, and targeting the mitochondrial permeability transition (MPT) by increasing cytosolic calcium and/or ROS levels which results in deregulation of the mitochondrial membrane potential (for review, see Galluzzi et al., 200651 and Fulda et al., 201052). Many of the most promising targeted anticancer compounds are still in pre-clinical testing.

7

1.1.3 Limitations of current chemotherapeutics

In describing the limitations of anticancer treatments, I will focus on two of the most successful antineoplastic compounds, cisplatin and doxorubicin, as exemplar agents.

A primary cause of failure of anticancer treatments is the intrinsic or acquired resistance of a tumour to the drug which often leads to disease re-occurrence53. This was initially characterized in the early studies of nitrogen mustards: after tumours receded they would recur and subsequently become resistant to further treatment. Resistance to anticancer compounds can arise in various ways and understanding these mechanisms can help inform new strategies of cancer treatment. In many cases, cells manifest multidrug resistance by reducing drug uptake and/or increasing drug efflux through modulation of the expression or activity of drug pumps such as P- glycoprotein and other multi-drug resistance transporters in the ATP-binding cassette (ABC) family54. In cases where the drug has a specific target, such as with the antifolates, loss of a cell surface receptor or mutation of the specific drug target (e.g. by gene amplification in the case of DHFR) can cause resistance. Resistance to the successful anticancer drug cisplatin can occur as a result of increased levels of drug detoxification by boosting of the production of cellular thiols, enhanced replication bypass of platinum-DNA adducts, changes in levels of regulatory proteins, increased DNA repair capacity, increased DNA damage tolerance, and the failure of cell death pathways55-57. Doxorubicin resistance can arise from alterations in DNA-damage sensing and repair capacities, and can also arise from decreased topoisomerase II expression and/or its catalytic activity. Increased expression of antioxidants that increase glutathione peroxidase activity also decreases doxorubicin efficacy53. Specific drug resistance can, in some cases, be addressed by combination treatments with compounds that act through different mechanisms of action.

Another limitation in the use of anticancer compounds arises from adverse toxicity to non- targeted tissues. Because most anticancer drugs were discovered based on their efficacy against cancer cells, little attention was initially given to their effects on normal tissues. One major side- effect of cisplatin is nephrotoxicity58. Although the proximal tubule cells of the kidney are quiescent, they are selectively damaged by cisplatin. Mechanisms that have been suggested to explain toxicity to these cells include: activation of mitogen-activated protein kinases (MAPK),

8

reactive oxygen species (ROS), and stimulation of inflammation and fibrogenesis58. Another significant dose-limiting side effect of cisplatin is toxicity to the brain where cisplatin use can lead to tinnitus, high-frequency hearing loss, and peripheral neuropathies including loss of vibration sense, paraesthesia, and weakness57,59. Platinum-based compounds preferentially enter the dorsal root ganglia and peripheral nerves and do not readily penetrate through the blood-brain barrier60. Cisplatin will bind to DNA and form adducts in dorsal root ganglia neurons leading to apoptosis of the neurons. The mechanism by which platinum-DNA adducts lead to neuronal apoptosis is currently not fully understood. Interestingly, because of their inability to traverse the blood-brain-barrier, platinum-based compounds can be delivered directly to the CNS to treat brain tumours61.

The main limitation to doxorubicin use is cardiotoxicity including cardiomyopathy and congestive heart failure43,44. Swain et al.62 analyzed 630 patients with breast carcinoma or small- cell lung carcinoma and found that ~26% of patients experienced doxorubicin-related heart failure at a dose of 550 mg/m2. Several mechanisms have been proposed to explain the particular sensitivity of the heart to doxorubicin-mediated toxicity. A widely accepted explanation is that induced by intramyocardial production of ROS following doxorubicin treatment produces cardiotoxic effects63. Another is that the heart is susceptible to the anthracyclines owing to its elevated levels of mitochondria activity and because the compounds bind to cardiolipin within the mitochondrial inner membrane. Additional explanations for doxorubicin toxicity include binding to cardiolipin, inhibition of nucleic acid and protein synthesis, release of vasoactive amines, alterations in adrenergic function and adenylate cyclase activity, changes in calcium transport, and alterations in cellular iron metabolism64,65. Interestingly, a recent study by Zhang et al. suggests that doxorubicin-mediated cardiotoxicity is not an off-target effect of the drug, but is instead caused by interactions with a form of topoisomerase (Top2b) which is ubiquitously expressed in cardiomyocytes66. Another organ targeted by doxorubicin is the brain, despite the drug not being able to cross the blood-brain barrier. Studies have suggested that tumour necrosis factor-α (TNF-α) is responsible for this toxicity67. Doxorubicin increases the production of TNF-α which in turn increases the production of inflammatory cytokines by microglial cells in the brain.

9

While the mechanisms that underlie these side effects have been studied for years, our understanding remains incomplete. The severity of these side effects can be reduced by using combination therapies that have the effect of minimizing the overall dose of each single agent. In addition, synergistic combinations with non-overlapping toxicities can reduce side-effects. Timing of combinations can also be exploited; for example, patients undergoing doxorubicin therapy are often pretreated with tamoxifen to reduce the level of toxic metabolites derived from doxorubicin53. While these and other clinical strategies can certainly improve outcomes, new therapies and a better understanding of traditional therapies will be invaluable.

1.1.4 New designs for chemotherapeutics

A common theme in drug discovery and development is to address limitations in current anticancer therapies by designing novel compounds with mechanisms that are based on successful drugs. Thousands of cisplatin analogues have been synthesized with the goals of 1) reducing toxicity to patients, 2) overcoming tumour resistance, and 3) increasing the range of antitumour activity. Early work in the design of novel platinum-based anticancer drugs focused on developing compounds through the modification of substituents surrounding the cisplatin core68 (Figure 1-1). An early, clinically successful cisplatin analogue, carboplatin, was developed by replacing the chloride leaving groups with a more stable bidentate dicarboxylate ligand. Carboplatin treatment is less nephrotoxic and less neurotoxic, however it can lead to myelosuppression24. It also requires a higher dosage for efficacy compared to cisplatin. Another cisplatin analogue, oxaliplatin, contains a diaminocyclohexane carrier as the non-leaving group. Oxaliplatin was shown to exhibit a different pattern of sensitivity against the NCI-60 panel of tumour cell lines which may be related to its ability to form different crosslink patterns than cisplatin69,70. Interestingly, oxaliplatin damage does not induce expression of genes involved in mismatch repair71. Picoplatin was designed to increase the steric bulk around the platinum core in an effort to reduce thiol-mediated inactivation. Picoplatin was found to have activity in cells resistant to cisplatin and to have antitumour activity in vivo72. Although picoplatin did not meet its primary endpoint of overall survival in Phase III trials for cisplatin-resistant small cell lung cancer, it is currently in Phase II trials for metastatic colorectal cancer73.

10

While these approaches have led to several compounds currently in clinical trials to address dose-limiting toxicities, because of their structural similarities to cisplatin, it is unlikely that these compounds will function through alternative mechanisms of action74. Approaches to platinum-based compound design which address the issue of drug resistance by maintaining antitumour activity, but which are not constrained to adhering to all of the structure-activity features of cisplatin and its analogues, have significant potential. The design of compounds with biologically active carrier ligands has paved the way for platinum-intercalator complexes in which the carrier group functions independent of the platinum. To date, compounds have been generated in which cisplatin is attached to DNA-intercalators such as acridine orange, chloroquine, and ethidium bromide, essentially bringing the platinum to its site of action75-78. In addition to DNA-intercalators, “hybrid drugs” in which the platinum moiety is attached to doxorubicin or to estrogen analogues have been tested and found to have increased efficacy against cisplatin-resistant tumours79,80.

The fact that cancer chemotherapy is limited by drug toxicity and resistance highlights the need to better understand drug mode of action within the cell. In addition, there is a clear need for novel compounds that act through different and/or complementary mechanisms which can be combined with existing agents to overcome resistance. These novel compounds also need to be evaluated to understand their mechanisms of action. Unbiased methods to examine drug function can lead to the development of better anticancer drugs and treatment regimens.

1.2 Chemical genomics in model organisms

1.2.1 Importance of screening to understand drug mechanisms of action

Since the completion of the human genome sequence in 2003, we have amassed a wealth of structural and functional information about the human genome and proteome. Until quite recently, much of this information had not been utilized in the development of new therapies. This is now changing, with recent advances in molecular biology, genetic engineering, and genome-scale screening providing powerful new technologies for identifying drug targets and understanding drug mechanism of action. Within the human genome, it is estimated that

11

approximately 3,000 human genes are “druggable,” defined by the potential ability of their protein products to bind drug-like chemical entities81. This information is based on sequence databases, structural information about the proteins, biological information on protein function, and homology to conserved protein families where a specific drug target has been characterized. However, less than half of the proteins expressed by the human genome are functionally characterized, suggesting that this number is an underestimate82. Furthermore, this characterization is often just a starting point. It is also possible that the number of potential targets could be larger than the number of genes in the genome; for example, post-translationally modified proteins or splice variants may be specifically druggable. In addition to identifying druggable genes, the Human Genome Project also led to the discovery of over 1,800 genes linked to human diseases including Alzheimer‟s, cancer and diabetes83. These genes were identified through alterations in the DNA sequence or changes in copy number to produce a specific phenotype. The intersection between druggable and disease-linked genes, estimated to be ~600-1500 genes, represents potential drug targets that are relevant for drug development81. Most drugs were discovered based on their efficacy and elucidation of their targets followed much later, if at all. In the past 10 years, research into potential drug targets has led to an increased proportion of small molecules that have been identified through target-based approaches82.

To compile a comprehensive understanding of a drug‟s cellular actions, it would be ideal to identify all primary and secondary targets of a drug. To this end, tools need to be developed that are rapid, cost-efficient, and can be used to study all cellular proteins (and other macromolecules) with different types of small molecule drugs or probes. Chemical genomics is one such approach which employs small molecules to explore gene function and to identify potential drug targets. An early example of the power of chemical genetics was the characterization of the protein tubulin as “colchicine-binding protein”; this discovery was made a decade before the tubulin gene was sequenced and 50 years before the term “chemical biology” was coined84. The complete sequence of the genome of human as well as other organisms has provided an invaluable “parts list” of potential targets. Indeed, genomic tools can be used to compile a comprehensive list of all proteins affected by a specific drug, providing the potential to greatly enhance our knowledge of a drug‟s mechanism of action. The identification of specific

12

drug targets, off-target effects, and proteins that modulate drug activity would lead to more rationalized drug design to optimize lead drugs for increased activity and potency while decreasing unwanted side effects. Despite these early promises, the insights provided by the human genome sequence are only now being effectively reduced to practice in drug discovery and development.

1.2.2 Chemical genomic screens in yeast

The budding yeast Saccharomyces cerevisiae has been the benchmark organism for the development, testing, and application of genomic technologies. It is also an ideal model organism for the development of high-throughput genomic screens. The first eukaryote to have its genome sequence completed, S. cerevisiae has a well-characterized genome and proteome, a rapid generation time, is inexpensive to cultivate, and is highly amenable to genetic manipulations such as gene deletion and dosage level variation85. A great resource for the development of systematic screening technology in S. cerevisiae is the yeast knockout collection (YKO). This is a complete deletion set of haploid strains and heterozygous and homozygous diploid strains in which each open reading frame (ORF) in the yeast genome has been precisely deleted from start to stop codon and replaced with a kanMX dominant drug resistance cassette86,87. The cassette contains two unique 20-nucleotide sequences which act as barcodes for identifying each deletion strain. These barcodes are flanked by common primer sequences, allowing for PCR amplification of barcodes and subsequent hybridization to a DNA barcode microarray88 or next-generation sequencing89 to identify strains in a mixed population of deletion strains. Therefore, the YKO collection can be pooled and grown in parallel and the relative abundance of each strain can be determined by the abundance of each barcode.

Indeed, the YKO collection presents an ideal resource for competitive growth assays that allow the systematic evaluation of growth of the deletion mutants in different conditions. Here, I discuss the application of this collection to chemical genomic screens. By growing the strains in the presence of drug, one can identify strains that confer growth advantages or disadvantages to the drug (Figure 1-2). This pooled approach allows for a rapid method of identifying growth effects using an unbiased, miniaturized approach90.

13

Figure 2-2. Yeast chemical genomic screens. a) HIPHOP assay. The YKO collection is pooled (1) and grown in the presence of a compound (2). Genomic DNA is extracted from the pool (3) and DNA barcodes are PCR amplified (4). The barcodes are hybridized to an Affymetrix TAG4 microarray (5). The signal intensity from the microarray is compared to an untreated control and the relative intensity represents relative abundance of the strain in the pool. b) Multi-copy suppression profiling. Yeast cells are transformed with a high-copy yeast ORF library (1) and grown in the presence of a high dose of compound (2). Plasmid DNA is isolated from resistant cells (3) and ORF sequences are PCR amplified (4) and hybridized to a DNA TAG4 microarray (5). High relative signal intensities represent strains containing ORFs that confer resistance to the compound. Adapted from Smith et al., 201091.

14

1.2.2.1 Haploinsufficiency profiling

One of the first chemical genomic screens developed using the YKO collection is based on the concept of haploinsufficiency, where a diploid cell bearing one single copy of a given gene grows indistinguishably from wild-type, except in conditions that require full protein function or activity. The assay, known as drug-induced HaploInsufficiency Profiling (HIP) is based on the observation that a strain containing a heterozygous deletion in an essential gene encoding the drug target results in sensitization of that strain to the drug92-94. Therefore, in a pooled culture, that strain would have a reduced fitness that can be quantified. The HIP assay has the ability to identify direct targets of compounds and proteins that may act in the same pathway as the target. In numerous studies, this assay has proven its ability to identify targets in well known and novel compounds89,95,96. The disadvantages to using HIP to identify drug targets are that it is limited to drug targets that are essential for growth and that the targets of human-specific drugs may not exist in yeast.

For example, Giaever et al.93 demonstrated the power of the HIP assay for drug target identification through screens of 10 diverse compounds which included several anticancer compounds. In the screen for methotrexate-sensitive mutants, the most highly sensitive strain was that containing a heterozygous deletion for the yeast dihydrofolate reductase gene, DFR1. Other heterozygous strains that displayed sensitivity to methotrexate were deleted for one copy of FOL1 or FOL2 which encode enzymes in the folic biosynthesis pathway that act upstream of DFR1. The screen also identified YBT1 and YOR072w mutant strains as hypersensitive to methotrexate. YBT1 is a known methotrexate transporter while the function of YOR072w remains uncharacterized. The HIP screen is also able to uncover additional pathways through which some compounds act. The antimetabolite anticancer agent 5-FU is known to act through inhibition of thymidylate synthase. In HIP screens of this compound, the thymidylate synthase gene deletion strain did not exhibit sensitivity. Instead, genes that conferred sensitivity were those involved in ribosome biogenesis and rRNA processing93,94, an observation that was novel and subsequently confirmed in follow-up studies showing that 5-FU blocks rRNA processing by the exosome94.

15

1.2.2.2 Homozygous profiling

Homozygous profiling (HOP) is similar to HIP but instead uses complete homozygous deletions to identify genes that confer resistance to a drug. An advantage of using homozygous deletions is that there is no chance for compensation of the gene deletion by increased transcription of the remaining allele. The HOP assay is useful in studies where there is no specific protein target or where the target is known and identification of genes that interact with the target is required. While it is possible to use this assay to identify direct targets, as demonstrated by its ability to identify FKBP12 and TOR as targets of rapamycin97, in most cases, this assay is used to identify genes in pathways that buffer the effects of the compound. An important application of HOP is to identify resistance genes in pathways that may be functionally redundant or have high transcriptional compensation in the cell, for example those in DNA-damage response (DDR)98- 100.

Examples of the use of HOP to study DDR include a study by Birrell et al. where the assay was used to identify genes involved in UV radiation sensitivity98. The authors were able to identify genes known to be involved in DNA repair pathways such as nucleotide excision repair, cell cycle checkpoints, homologous recombination, and post-replication repair. This study led to the identification of three genes (THR1, LSM1, YAF9) not previously known to be involved in DNA- damage repair pathways. Two of these genes have human orthologues associated with cancer. Lee et al.101 used the HOP assay to identify genes required for resistance to DNA-damaging agents with diverse mechanisms of action. In this study, 12 compounds that damage DNA were tested to uncover genes involved in distinct DDR pathways that are important to repair damage by each compound. The authors found that relative importance of different DDR pathways was able to distinguish between compound mechanisms. Specifically, they identified genetic determinants required for resistance against DNA interstrand crosslinking agents. In addition, new genes involved in the DDR were uncovered in this study.

1.2.2.3 Multicopy suppression profiling

As a complement to the HIP and HOP screens, the multicopy suppression profiling (MSP) assay uses overexpression of genes to identify drug targets. This assay relies on the concept that an increased dose of a drug target will render that strain resistant to a high concentration of drug

16

that would otherwise lead to growth inhibition. The idea of MSP was used to identify kinase targets of uncharacterized compounds102. To develop MSP into a systematic, miniaturized assay, mutant strains are pooled and grown competitively with strains conferring resistance exhibiting a growth advantage90. In this assay, the ORF sequences are PCR amplified and hybridized to a microarray containing probes against each gene in the yeast genome or quantified by next- generation sequencing. A disadvantage of MSP is that, similar to HIP, the drug should have a single molecular target and that the target cannot be part of complexes that may have stoichiometric constraints.

Luesch et al.102 performed MSP in an arrayed format to identify the targets of a kinase-directed library of growth-inhibitory small molecules. Pkc1 overexpression was found to suppress growth inhibition by a phenylaminopyrimidine (PAP) compound. Also, overexpression of kinases downstream to Pkc1, Bck1, and Mkk2, displayed multicopy suppression, indicating the pathway through which the PAP compound acts. Affinity chromatography and biochemical assays confirmed Pkc1 as the target.

HIP, HOP, and MSP are complementary assays that can be integrated to reveal a comprehensive understanding of a drug‟s mode of action in the cell. Taken together, these screens have the potential to greatly impact future drug discovery strategies. (For review, see Smith et al., 201091)

1.3 Functional genomics in mammalian cells

Model organism research has demonstrated the importance of employing genomics and chemical genomics to understand biological function and human disease. However, although model organisms are excellent test-beds for understanding well-conserved processes, there are processes that can only be studied in genetically related cell types. Indeed, adapting loss-of- function and gain-of-function screens performed in model organisms to mammalian systems has long been a goal, and recently great strides have been made towards this goal. While adapting genomic studies to mammalian systems is more challenging on several fronts, recent developments of RNA interference (RNAi) libraries (to decrease gene dosage) and human ORF

17

libraries (to increase gene dosage) have enabled large-scale, genome-wide studies to be applied to mammalian cells.

1.3.1 Loss-of-function screens

1.3.1.1 RNA interference

RNA interference is the process by which non-coding double-stranded RNA (dsRNA) molecules mediate target-specific degradation of mRNA. It was first observed in C. elegans when antisense RNA molecules were tested for their ability to inhibit gene expression103. Fire and Mello found that dsRNA was more potent in triggering gene silencing than single-stranded RNA; this work was recognized with a Nobel Prize in 2006. Unlike in C. elegans and D. melanogaster, where the first RNAi studies were performed, long dsRNA molecules trigger an interferon response in mammalian systems and are degraded before they can bind their target104. Since dsRNA is processed to 21-23 nucleotide fragments to induce the RNAi machinery105, synthetic short sequences of 21-25 nucleotides, known as short interfering RNAs (siRNAs)106, are used in mammalian systems to circumvent the interferon response. siRNAs can be used directly or introduced on plasmids that express short hairpin RNAs (shRNAs) which can be processed in the cell by the Dicer enzyme to produce the siRNA duplex107,108. No endogenous siRNAs have been identified in mammalian systems to date. Instead, the endogenous RNAs that utilize the RNAi machinery are microRNAs (miRNAs)109, which are produced as long transcriptions (called pri- miRNAs) that can contain multiple hairpin structures110. Pri-miRNAs are cleaved by the RNase III enzyme Drosha to produce pre-miRNA hairpin structures in the nucleus111. The pre-miRNAs are exported to the cytoplasm and processed similar to shRNAs. Another small RNA molecule class that is often used to induce RNAi are endoribonuclease-produced siRNAs (esiRNAs)112,113. In order to produce gene silencing, the short siRNA duplex produced after processing by Dicer is exported into the cytoplasm and incorporated into the RNA-induced silencing complex (RISC) which contains proteins in the Argonaute family114 (Figure 1-3). The duplex is unwound, with the antisense strand remaining in the complex and the other strand quickly degraded. The antisense strand is used to guide the RISC complex to bind to its target mRNA for subsequent endonucleolytic cleavage. (For review of RNAi, see Hannon, 2002114.)

18

Figure 2-3. RNAi pathway in mammalian cells. shRNA molecules on plasmids can be introduced into the cell by transfection or virus infection. The shRNA is transcribed in the nucleus to produce the hairpin structure which is exported to the cytoplasm. Dicer cleaves the hairpin structure to produce the siRNA duplex. siRNAs can be directly transfected into cells. siRNA duplexes are incorporated into the RNA-induced silencing complex (RISC). The duplex is unwound and the antisense strand is used to guide the RISC complex to the target mRNA for cleavage and degradation.

19

1.3.1.2 Using RNAi for genome-scale loss-of-function screens

The ease of genetic manipulation through RNAi has produced a shift in mammalian molecular biology research by accelerating the development of genome-scale functional studies in human cells. RNAi-based loss-of-function studies allow us to examine human genes directly in cultured cells. Both commercial and academic laboratories have generated a number of large genome- scale siRNA, shRNA, and esiRNA libraries in plasmids and viral vectors that enable large- format screening of human genes. To illustrate the diversity of these resources, The RNAi Consortium has a library of 90,000 shRNA constructs that target ~18,000 human genes in lentiviral vectors115, the Netherlands Cancer Institute has a library of 24,000 19-mer shRNAs in retroviral vectors116, and the Hannon and Elledge labs generated a library of microRNA-adapted shRNAs (shRNA-miRs) which contains 395,830 shRNA-miRs against 57,293 human transcripts117.

The primary advantages to using RNAi for functional genomics are that RNAi molecules can be easily introduced into cells, they can be introduced transiently (siRNA, esiRNA) or stably (shRNA, shRNA-miR) into cells, and screens can be done in arrayed (well-by-well) or pooled formats. A disadvantage to using RNAi is its potential for off-target effects which can occur when RNAi sequences bind to non-targeted mRNAs. Another limitation is that the efficacy of mRNA knockdown varies (in unpredictable ways) depending on the cell type and the sequence target of the RNAi molecule.

In an arrayed screen, each RNAi molecule is placed into a single well of a micro-titre plate. The advantage to this type of screen is that there are no confounding effects from other infected cells, since only one knockdown or one set of knockdowns/gene are assayed. In addition, this format is suitable for multiple types of readouts, including colorimetric, fluorescence, and luminescence assays and high-content morphological phenotyping. The main disadvantages of array-based screens, especially with the large number of molecules in each of these libraries, are that costly infrastructure is required, the volume of reagents is high, and a minimum level of automation is necessary. For pooled approaches, the library is typically introduced into cells en masse and, based on the statistics of infection, each cell, on average, contains a single RNAi molecule. Pooled approaches are more feasible for smaller laboratories but a disadvantage to pooled

20

screens is that analysis of results requires costly deconvolution through microarrays or next- generation sequencing, and sufficient bioinformatic expertise to identify the RNAi molecules that are enriched or depleted in a pool. Our lab recently developed a new microarray-based platform that allows the deconvolution of pools of up to 90,000 shRNA molecules118. The technical challenge of identifying specific molecules in highly complex RNAi libraries can be addressed by dividing genome-scale collections into smaller pools119,120.

Several RNAi-based genome-scale studies have been used to identify novel molecular targets of cancer. An example of an arrayed RNAi screen was done by Moffat et al.115 to identify genes involved in mitotic progression in human cancer cells. This study examined 5,000 shRNAs targeting ~1,000 human genes including 476 protein kinases and 180 phosphatases. HT29 colon cancer cells were infected with shRNA-containing lentiviruses such that each well contained a distinct shRNA-expressing cell population. Automated cell imaging was used to detect cells in mitosis by visualizing histone H3 phosphoryation, a marker for mitotic cells. This screen identified known mitotic regulators as well as over 100 candidate genes previously unlinked to mitosis. The authors note that genes that are identified as mitotic regulators in malignant cells and not in non-malignant cells may be potential cancer targets.

In a proof-of-principle chemical screen, Luo et al.121 performed a pool-based screen on H82 small-cell lung cancer cells infected with ~45,000 shRNAs to identify the target of the DNA- damaging anticancer agent, etoposide. As described above, etoposide is a topoisomerase II poison and exhibits increased toxicity with increased cellular levels of the topoisomerase II protein. This positive selection screen, which used a high dose of etoposide, correctly identified TOPIIA as a suppressor gene. Cells containing TOPIIA knockdowns exhibited up-to-50 fold enrichment in etoposide treatment compared to untreated cells.

RNAi-based chemical screens (i.e. synthetic lethal screens) have also been used to identify novel members in DNA-damage related pathways. In a study by O‟Connell et al., HeLa cells were infected with a pool of 75,000 shRNAs and treated with the topoisomerase I poison, camptothecin (CPT)122. The screen uncovered shRNAs targeting TOP1 as enriched in CPT- treated cells. On the opposite side, 331 hairpins were found to confer sensitivity to CPT treatment. Among the genes with multiple shRNA hits, the group discovered a protein related to

21

yeast Mms22p/Mus7p, MMS22L, which interacts with another hit from the screen, NFKBIL2. An independent study also identified MMS22L and NFKBIL2 (TONSL) in an RNAi-based screen to identify siRNAs that lead to increased 53BP1 subnuclear foci formation in response to ionizing radiation123. Subsequent experiments determined that both these proteins promote homologous recombination repair, increase DNA-damage foci when knocked down, and accumulate at sites of replication stress and DNA damage122,123. These studies highlight the power and potential of chemical genomic screens to identify novel genes in known biological pathways.

1.3.2 Gain-of-function screens

A complementary approach to RNAi is to investigate gene function by overexpressing genes. Until recently, the number of gain-of-function studies in mammalian cells lagged behind loss-of- function studies mainly due to the lack of availability of fully-sequenced complete-ORF- containing cDNAs cloned into mammalian expression vectors. Previously, fragmented genomes or cDNA libraries produced from specific cell lines or tissues were used in gain-of-function studies. However, cDNAs derived from pooled reverse transcription of mRNA may have an unequal representation of genes due to differences in expression in the cell line or tissue. Now, there are several full-length ORF collections in development124-126. Currently, the largest full- length cDNA collection is the human ORFeome collection developed by the Center for Cancer Systems Biology at the Dana-Farber Cancer Institute. This hORFeome is derived from the Mammalian Gene Collection (MGC) of full-length cDNAs for human, rat, and mouse genes127. The MGC was produced in an arrayed format allowing for equal representation of each ORF. The human ORFeome v8.1 contains 14,524 fully-sequenced ORFs representing 12,940 human genes which have been cloned into Gateway Entry vectors allowing easy transfer into Gateway- compatible expression vectors of choice128. Though this list is not comprehensive, this collection will greatly enhanced the ability to perform systematic genome-wide gain-of-function screens in mammalian cells.

1.3.2.1 Examples of systematic gain-of-function mammalian screens

To date, examples of systematic gain-of-functions screens using ORF libraries have been done on a limited scale. Boehm et al.129 screened a set of 354 kinases and kinase-related ORFs to

22

identify human oncogenes. They first confirmed that the activation of the PI3K pathway by myr- AKT along with activated MEK (MEKDD) in the MAPK pathway could stimulate anchorage- independent growth in immortalized HEK cells. Pools of 10-12 validated kinase ORFs were screened to identify kinases that could replace PI3K signaling in a transformation assay along with MEKDD. The authors identified four kinases that could substitute for myr-AKT. One of these, IKBKE, is amplified and overexpressed in one-third of all human primary breast tumours. Further analyses revealed that knockdown of IKBKE leads to apoptosis in breast cancer cells. This study exemplifies integrated loss- and gain-of-function studies to understand cancer biology.

A larger-scale screen examined ~600 kinase and kinase-related open reading frames (ORFs) to identify genes that cause resistance to a RAF kinase inhibitor in a B-RAF malignant melanoma cell line130. From this array-based screen, nine ORFs, including three tyrosine kinases, were identified as conferring resistance to the inhibitor. MAP3K8 (which encodes the COT protein) was identified as a MAPK pathway activator that induces resistance to RAF inhibition. This study shows how identifying gene overexpression that leads to drug resistance can be used to identify new drug targets and pathways for future combination anticancer strategies.

Transposon-based technologies represent another method to introduce gain-of-function mutations into cells. Transposons allow the translocation of DNA sequences from one DNA site to another, such as from a vector into genomic DNA, using a “cut-and-paste” mechanism. The advantages to using transposons for insertional mutagenesis are that transponsons can be maintained on simple non-viral plasmids, transposon insertions can be remobilized, and that they often show target-specific insertion131,132. Disadvantages to using transposon-based mechanisms are that the integration may be unstable and that there may be multiple insertions in a single genome. De et al.133 used a transposon-based promoter insertion strategy to identify mechanisms of resistance to , a inhibitory anticancer agent, in breast cancer cells. Using validation-based insertional mutagenesis, they inserted cytomegalovirus (CMV) promoters randomly into the genomes of docetaxel-sensitive MDA-MB-231 breast cancer cells. These cells were selected with docetaxel resistant clones were sequenced to identify any full-length ORFs that were overexpressed. This screen identified the KIFC3 gene as responsible for resistance to

23

docetaxel. Further studies revealed that KIFC3 and additional kinesins KIFC1, KIF1A, KIF5A increased resistance to docetaxel in a naïve cell line when overexpressed.

1.3.2.2 Downstream applications of overexpression studies

In addition to identifying phenotypes generated from gene overexpression in a wild-type background, there are many additional applications for gain-of-function screens. For example, they can be the starting platform for the identification of suppressor or enhancer mutations, which are mutation phenotypes that are suppressed or enhanced by overexpression of another gene. Both of types of screens can be useful to delineate molecular pathways and add additional functions to genes. Gain-of-function screens can also be a powerful method of drug target identification, as demonstrated in yeast by MSP. The examples of overexpression studies in the previous section represent positive-selection screens. Negative-selection screens can be used to identify gain-of-function mutations that are lethal to cell lines in various conditions. These types of screens can be the starting point to identify mutations that rescue such effects.

1.4 Project rationale

In this introduction, I have described the effectiveness of DNA-targeted compounds as anticancer agents but also emphasized that there are limitations to their use. Therefore, there is a need for ways to better understand drug mechanism of action of known and novel anticancer drugs to allow for better drug design and combination therapies. I have also described the success of chemical genomic screens in model organisms and novel resources allowing for functional genomics in mammalian cells. Together this indicates that the development of chemical genomics screens in mammalian cells is the next step for directly studying drug function against the human genome.

My first objective is to evaluate the potential of novel platinum-based compounds as anticancer agents using chemical genomic screens in yeast. My second objective also employs yeast chemical genomics to study a novel mitochondria-targeted anticancer drug, elesclomol. As the development of chemogenomic screens in yeast has greatly improved our understanding of drug- gene interactions, the second half of my thesis focuses on the generation of parallel assays for 24

human cell culture that will allow the direct study of human genes that respond to drugs. Here, the results can be more readily applied to studying human diseases and therapeutics. Therefore, my third objective is to develop and apply a genome-scale loss-of-function approach using RNAi to study human genes that confer resistance to the anthracycline anticancer agent doxorubicin. The identification of these genes may have implications in doxorubicin toxicity and resistance. My last objective is to identify genes that are toxic when overexpressed in human cells using a gain-of-function screen developed in our lab. By characterizing their role in cell death, this study may reveal novel functions of the cytotoxic genes and allow further investigation of their roles in diseases which are caused by gene hyperactivation.

25

Chapter 2 Comparative chemogenomics to examine the mechanism of action of DNA-targeted platinum-acridine anticancer agents

The material presented in this chapter is reproduced with permission from: Cheung-Ong, K. et al. Comparative Chemogenomics To Examine the Mechanism of Action of DNA-Targeted Platinum-Acridine Anticancer Agents. ACS Chem Biol 7, 1892-901. doi: 10.1021/cb300320d Copyright 2012 American Chemical Society.

Contributions: I performed the analysis of HIPHOP screen data, individual strain confirmations, cell cycle experiments and analysis, mitochondrial experiments and analysis, and sample preparation for molecular combing. Kevin Song performed the HIPHOP screens, Daniel Shabtai clustered the screen results, Anna Lee performed GSEA, David Gallo performed the molecular combing experiment and analysis. Uli Bierbach and Zhidong Ma generated the compounds. This study was supervised by Corey Nislow, Guri Giaever and Uli Bierbach.

2 2.1 Introduction

Platinum-based drugs are extremely successful cancer chemotherapeutics. The widely used agent cisplatin (cis-diamminedichloroplatinum(II)) acts by forming cross-links in the major groove of DNA which inhibit DNA synthesis and lead to apoptosis23. Cisplatin is effective against testicular, ovarian, bladder and cervical tumours134 but it manifests dose-limiting toxicities on the nervous and renal systems. Equally troubling is the frequency at which resistance develops during treatment135. In light of these limitations, attention has focused on developing platinum- based agents with improved activity in notoriously cisplatin-resistant cancers.

Diverse designs for new platinum-based anticancer drugs have focused on modifying substituents surrounding the cisplatin core68. Clinical data shows that these compounds offer certain advantages over cisplatin, yet because their mechanism of action is similar to that of cisplatin at the nuclear level, they represent an incremental improvement74. Alternative 26

approaches to platinum-based drug design, in contrast, do not adhere to all of the structural dictates of cisplatin136,137. For example, a non-classical trinuclear platinum agent, BBR3464 (Novuspharma/Cell Therapeutics), damages DNA in a radically different manner than cisplatin by forming a range of unique flexible, long-range DNA adducts138,139. In preclinical trials, BBR3464 was cytotoxic at doses up to 1000-fold lower than cisplatin and showed efficacy against cisplatin-resistant lung, colorectal, pancreatic and ovarian cancers27. Phase II clinical trials of BBR3464 in gastric and small cell lung cancer ultimately failed due to lack of activity140,141. While there were some promising results from two other Phase II studies against non-small cell lung and ovarian cancers, the compound has not moved into Phase III trials27.

Recently, a class of promising platinum-acridine conjugates was developed by combining the DNA-damaging features of a platinum complex with the DNA-intercalating properties of acridine-based chromophores. The prototype of this class of hybrid agents, PT-ACRAMTU(EN) ([PtCl(en)(ACRAMTU)] dinitrate salt, ACRAMTU = 1-[2-(acridin-9-ylamino)ethyl]-1,3- dimethylthiourea, en = ethane-1,2-diamine), shows a dose-response similar or superior to cisplatin in several solid tumour cell lines142-145. In the DNA adducts produced by PT- ACRAMTU(EN), the metal forms a single coordinative bond with the nucleobase nitrogen while the ACRAMTU moiety intercalates into the base-pair step adjacent to the site of platination146. This results in a novel mechanism of DNA damage distinct from the DNA cross-linking characteristic of cisplatin. Based on these promising initial results, additional platinum-acridine compounds were synthesized with a focus on modifications to the prototypical PT- ACRAMTU(EN) structure (Fig. 2-1)147-149. Structure-activity relationship studies revealed PT- ACRAMTU(EN) analogues with up to 500-fold higher potency in non-small cell lung cancer (NSCLC) when compared to cisplatin150. The promise offered by the platinum-acridines is clear, yet, the cellular mechanism of these dual-mode agents remains largely uncharacterized.

Here we use a validated cell-based chemical genomic assay to examine a library of these platinum-acridine hybrid agents to characterize their cellular mechanism(s) of action and

27

Figure 2-1. Chemical structures of the platinum-acridine compounds and carriers screened.

28

evaluate their potential as anti-cancer agents. Our lab has previously shown that the Saccharomyces cerevisiae gene deletion collection is a powerful means to dissect the detailed mechanisms of action of functionally diverse DNA-damaging agents101 by screening these agents against the complete pool of ~4,800 barcoded non-essential deletion strains. In this study we apply a similar approach in both S. cerevisiae and the distantly related fission yeast Schizosaccharomyces pombe to dissect the mechanism(s) of action of these novel platinum- acridine conjugates. The results offer insight into the relative role of each structural modification on the genome-wide response and provide mechanistic insight into the relationship between compound structure and biological activity.

2.2 Results

2.2.1 Global analysis of fitness profiles

Platinum-acridines derived from the prototype PT-ACRAMTU(EN) were developed using a modular synthetic approach. The compounds tested in this study can be grouped into three categories: platinum-free carriers (2 derivatives), platinating-intercalating hybrid agents (7 derivatives) and bisintercalators (4 derivatives) (Fig. 2-1). Each genome-wide screen produces a fitness profile that consists of a numerical value for each deletion strain‟s fitness (relative growth with and without drug treatment) (Appendix Dataset 2-1). To compare genome-wide profiles between compounds, Daniel Shabtai developed a rank-based similarity scoring algorithm to account for non-biological “batch”-based systematic effects151. With this algorithm, every gene deletion strain within a profile is distributed into sections, or “buckets,” where strains that are most influenced by the chemical compound (and therefore have the highest fitness defect scores) comprise one bucket, the next set of strains are distributed into a second bucket, and so forth. Once all strains are divided into buckets, a scoring matrix is constructed and used to evaluate the similarity between profiles (see Methods). This approach effectively minimizes the influence of “batch” effects that can overwhelm the biological signal from any one experiment152. Using the resulting profile similarity scores, compounds with similar mechanisms cluster together, with experimental replicates being the most similar (Fig. 2-2).

Clustering of all S. cerevisiae profiles revealed that, of all the platinum-acridines, PT- ACRAMTU(EN) induces a cellular response most similar to cisplatin, further suggesting that the

29

Figure 2-2. Two-dimensional hierarchical clustering of all 14 compounds based on the profile similarity scores of their genome-wide S. cerevisiae profiles. Red indicates high similarity between compound profiles and grey indicates low similarity. Profiles from experiments repeated under the same conditions cluster together with few outliers, indicating the reproducibility and resolution of the genome-wide assay. The dendrogram reveals structure- function relationships between compounds.

30

most significantly sensitive strains represent those deleted for genes involved in DNA-damage repair. PT-ACRAMTU(PN), PT-ATUCA and PT-AMIDINE(EN) also grouped together with cisplatin and PT-ACRAMTU(EN). The platinum-free carriers and the 7 other platinum-acridine compounds clustered separately. Notably, PT-ACRAMTU(BPY), which was previously examined for its DNA-damaging properties148, clusters with the platinum-free carriers, suggesting that it does not generate fitness defects in S. cerevisiae. The individual genome-wide strain sensitivities for each compound are discussed in detail below.

2.2.2 Specific genes involved in DNA-damage response to these novel platinum compounds

Cisplatin was profiled to serve as a benchmark for the platinum-acridines. We previously demonstrated that the cellular response to cisplatin requires genes with roles in nucleotide excision repair (NER), homologous recombination repair (HRR), post-replication repair (PRR) and translesion synthesis (TLS)153,154. In this study, strains deleted for genes in NER (RAD1, RAD2, RAD4, RAD10 and RAD14), HRR (RAD51, RAD52, RAD54, RAD55 and RAD57) and PRR (RAD5 and RAD18) were among the top 30 most cisplatin-sensitive strains. Strains deleted for PSO2, an NER gene involved in cross-link repair155, and genes involved in TLS (REV1, REV3 and REV7) also ranked highly; these were previously found to be the principal differences in the response to well-characterized cross-linking and non-cross-linking agents156.

PT-ACRAMTU(EN) (Fig. 2-1) forms monofunctional-intercalative adducts in both grooves of double-stranded DNA but does not cross-link nucleobases157,158, therefore we anticipated its genome-wide profile would be distinct from cisplatin. Contrary to this expectation, the DNA repair modules important for resistance to PT-ACRAMTU(EN) are most similar to those of cisplatin. The top 30 strains are deleted for genes in NER (RAD1, RAD2, RAD4, RAD10 and RAD14), HRR (RAD51, RAD54 and XRS2) and PRR (RAD5 and RAD18) (Fig. 2-3) with NER and PRR being the most sensitive modules (Fig. 2-4). One major difference from cisplatin is that deletion of PSO2 did not sensitize this strain to PT-ACRAMTU(EN), suggesting that the monoadducts formed are not recognized by cross-link-specific repair mechanisms. In addition, genes essential for TLS were not found in the PT-ACRAMTU(EN) profile, as would be predicted for a non-cross-linking compound. The relative ranking of strains from these five DNA-damage response (DDR) modules are shown in Figure 2-4a for those platinum-acridines 31

Figure 2-3. S. cerevisiae and S. pombe profiles for the platinum-acridine compounds that require DNA-damage response pathways. The top 30 sensitive strains are marked and labeled. Each graph represents the average of at least 3 replicate sensitivity screens in S. cerevisiae or 2 replicates in S. pombe. The x-axis represents gene names in alphabetical order and the y-axis is the log2 ratio of barcode intensity for drug vs control. Red indicates essential genes, blue indicates non-essential genes and purple indicates marked genes.

32

Figure 2-4. Relative importance of DNA-damage repair modules in the resistance to the DNA- damaging platinum compounds in S. cerevisiae and S. pombe. Each bar represents the median rank for genes in each of the DNA repair modules listed in the top 30 or top 250 most sensitive strains. The DNA-repair modules in S. cerevisiae were defined as follows: cross-linking genes (PSO2), NER (RAD1, RAD2, RAD4, RAD10, RAD14), PRR (RAD5, RAD6, RAD18), TLS (REV1, REV3, REV7), HRR (RAD51, RAD52, RAD54, RAD55, RAD57, RAD59), stalled replication fork repair (MUS81, MMS4). The DNA-repair modules in S. pombe were defined as follows: NER (RHP14, RHP41, RHP42, RAD13, RAD16, SWI10), PRR (RHP6, RHP18, RAD8), TLS (REV1, REV3, REV7), HRR (RHP51, RHP54, RHP55, RHP57, RAD22, RTI1), stalled replication fork repair (MUS81).

33

whose profiles suggest they exert their cellular effects via damaging DNA. Significantly sensitive strains were tested in monoculture for compound sensitivity using fitness assays and spot dilutions with a 77% hit confirmation rate (Table 2-1, 2-2). To summarize the cellular responses captured by S. cerevisiae profiles, Anna Lee performed gene set enrichment analysis (GSEA) to identify biological processes significantly enriched amongst genes whose deletions induce compound-sensitivity, and visualized the results as a network (Fig. 2-5).

Three of the seven platinum-intercalating hybrid derivatives of PT-ACRAMTU(EN) elicit responses from DDR genes. The bidentate nonleaving group in PT-ACRAMTU(EN) was replaced with various diamines to yield PT-ACRAMTU(PN), PT-ACRAMTU(BPY) and PT- ACRAMTU(TMEDA) (Fig. 2-1). In previously reported in vitro DNA polymerase stop assays, PT-ACRAMTU(PN) and PT-ACRAMTU(BPY) were shown to produce DNA adducts while the sterically-hindered PT-ACRAMTU(TMEDA) was unable to platinate DNA148. Clustering of genome-wide profiles revealed that PT-ACRAMTU(PN) groups with PT-ACRAMTU(EN) and cisplatin while PT-ACRAMTU(BPY) and PT-ACRAMTU(TMEDA) cluster separately (Fig. 2- 2). The genome-wide response to PT-ACRAMTU(PN) differed from that of PT- ACRAMTU(EN); strains deleted for genes involved in NER ranked lower in sensitivity, while strains deleted for genes involved in HRR and PRR were more sensitive. Although PT- ACRAMTU(BPY) damages DNA148,159, we did not uncover strains deleted for DDR genes as sensitive in our screens, consistent with previously reported model studies, which showed that the trans-labilizing effect of the bipyridine (BPY) ligand renders the DNA adducts formed by this compound kinetically labile148. Instead, PT-ACRAMTU(BPY) elicited responses for genes involved in RNA metabolic processes. PT-ACRAMTU(TMEDA), which does not bind to DNA, did not elicit responses from any known DDR genes.

To tune the DNA sequence specificity of the monofunctional platinum moiety, the carrier group of PT-ACRAMTU(EN) was modified to produce PT-ATUCA, which features a carboxamide group attached at the 4-position of the acridine chromophore. Despite this drastic modification, the fitness profile for PT-ATUCA was similar to that of PT-ACRAMTU(EN); strains deleted for genes involved in NER, HRR and PRR were the most sensitive (Fig. 2-3), with NER being the most important pathway for response to this compound (Fig. 2-4). To generate PT-

34

Table 2-1. Growth responses of strains deleted for DDR genes in the top 20 individual mutant strains to compound treatment. Individual mutant strains were grown with and without 200 µM of PT-ACRAMTU(EN), PT-ACRAMTU(PN), PT-ATUCA and PT-AMIDINE(EN) (S. cerevisiae) or 25 µM of PT-ACRAMTU(EN), PT-ACRAMTU(PN), PT-ATUCA and 12.5 µM PT-AMIDINE(EN) (S. pombe). The average growth-rates relative to control are listed. If the inhibition of the strain after compound treatment is >10% compared to the no-drug control, then it is confirmed as sensitive. Growth Growth compared Confirmed? compared to to control (with (>10% organism gene orf::batch compound control (no drug) drug) inhibition) S. cerevisiae RAD5 YLR032W::chr00_8 PT-ACRAMTU(EN) 0.938 0.801 Y S. cerevisiae RAD14 YMR201C::chr13_4 PT-ACRAMTU(EN) 0.99 0.903 N S. cerevisiae RAD4 YER162C::chr00_5 PT-ACRAMTU(EN) 0.998 0.855 Y S. cerevisiae RAD2 YGR258C::chr00_17a PT-ACRAMTU(EN) 0.984 0.815 Y S. cerevisiae RAD10 YML095C::chr00_9 PT-ACRAMTU(EN) 1.004 0.827 Y S. cerevisiae RAD18 YCR066W::chr00_1 PT-ACRAMTU(EN) 0.908 0.518 Y S. cerevisiae RAD1 YPL022W::chr16_3 PT-ACRAMTU(EN) 0.982 0.785 Y S. cerevisiae RAD54 YGL163C::chr7_2 PT-ACRAMTU(EN) 0.78 0.729 N S. cerevisiae HPR5 YJL092W::chr10_2 PT-ACRAMTU(EN) 0.806 0.694 Y S. cerevisiae RAD51 YER095W::chr00_8 PT-ACRAMTU(EN) 0.743 0.627 Y S. cerevisiae XRS2 YDR369C::chr4_7 PT-ACRAMTU(EN) 0.685 0.545 Y S. cerevisiae RAD5 YLR032W::chr00_8 PT-ACRAMTU(PN) 0.864 0.64 Y S. cerevisiae RAD4 YER162C::chr00_5 PT-ACRAMTU(PN) 0.947 0.839 Y S. cerevisiae RAD2 YGR258C::chr00_17a PT-ACRAMTU(PN) 0.971 0.871 N S. cerevisiae RAD10 YML095C::chr00_9 PT-ACRAMTU(PN) 0.944 0.878 N S. cerevisiae RAD18 YCR066W::chr00_1 PT-ACRAMTU(PN) 0.904 0.613 Y S. cerevisiae RAD1 YPL022W::chr16_3 PT-ACRAMTU(PN) 0.991 0.987 N S. cerevisiae RAD54 YGL163C::chr7_2 PT-ACRAMTU(PN) 0.788 0.743 N S. cerevisiae HPR5 YJL092W::chr10_2 PT-ACRAMTU(PN) 0.835 0.769 N S. cerevisiae RAD51 YER095W::chr00_8 PT-ACRAMTU(PN) 0.707 0.616 N S. cerevisiae RAD55 YDR076W::chr4_4 PT-ACRAMTU(PN) 0.855 0.858 N S. cerevisiae RAD5 YLR032W::chr00_8 PT-ATUCA 0.889 0.397 Y S. cerevisiae RAD14 YMR201C::chr13_4 PT-ATUCA 0.951 0.669 Y S. cerevisiae RAD4 YER162C::chr00_5 PT-ATUCA 0.966 0.777 Y S. cerevisiae RAD2 YGR258C::chr00_17a PT-ATUCA 0.941 0.662 Y S. cerevisiae RAD10 YML095C::chr00_9 PT-ATUCA 0.981 0.666 Y S. cerevisiae RAD1 YPL022W::chr16_3 PT-ATUCA 0.898 0.595 Y S. cerevisiae RAD54 YGL163C::chr7_2 PT-ATUCA 0.759 0.508 Y S. cerevisiae IMP2' YIL154C::chr9_2 PT-ATUCA 0.756 0 Y S. cerevisiae RAD5 YLR032W::chr00_8 PT-AMIDINE(EN) 0.917 0.448 Y S. cerevisiae RAD14 YMR201C::chr13_4 PT-AMIDINE(EN) 0.954 0.236 Y S. cerevisiae RAD4 YER162C::chr00_5 PT-AMIDINE(EN) 0.938 0.274 Y S. cerevisiae RAD2 YGR258C::chr00_17a PT-AMIDINE(EN) 0.963 0.241 Y S. cerevisiae RAD10 YML095C::chr00_9 PT-AMIDINE(EN) 0.985 0.207 Y S. cerevisiae RAD18 YCR066W::chr00_1 PT-AMIDINE(EN) 0.91 0.315 Y S. cerevisiae RAD1 YPL022W::chr16_3 PT-AMIDINE(EN) 0.823 0.243 Y S. cerevisiae RAD54 YGL163C::chr7_2 PT-AMIDINE(EN) 0.886 0.584 Y S. cerevisiae HPR5 YJL092W::chr10_2 PT-AMIDINE(EN) 0.855 0.38 Y S. cerevisiae RAD51 YER095W::chr00_8 PT-AMIDINE(EN) 0.753 0.352 Y S. cerevisiae IMP2' YIL154C::chr9_2 PT-AMIDINE(EN) 0.754 0.124 Y S. cerevisiae RAD7 YJR052W::chr00_13 PT-AMIDINE(EN) 1.014 0.384 Y S. cerevisiae RAD23 YEL037C::chr5_2 PT-AMIDINE(EN) 0.999 0.376 Y S. cerevisiae MMS4 YBR098W::chr2_3 PT-AMIDINE(EN) 0.976 0.515 Y S. cerevisiae MUS81 YDR386W::chr4_7 PT-AMIDINE(EN) 1.026 1.183 N S. pombe rad13 SPBC3E7.08c PT-ACRAMTU(EN) 0.885 0.491 Y S. pombe rad1 SPAC1952.07 PT-ACRAMTU(EN) 0.69525 0.6875 N S. pombe rad26 SPAC9E9.08 PT-ACRAMTU(EN) 0.684 0.56 Y S. pombe rad9 SPAC664.07c PT-ACRAMTU(EN) 0.86225 0.6485 Y S. pombe rad22 SPAC30D11.10 PT-ACRAMTU(EN) 0.80975 0.772 N S. pombe rad13 SPBC3E7.08c PT-ACRAMTU(PN) 0.885 0.7435 Y S. pombe rad1 SPAC1952.07 PT-ACRAMTU(PN) 0.69525 0.7865 N S. pombe rad26 SPAC9E9.08 PT-ACRAMTU(PN) 0.684 0.751 N S. pombe rad9 SPAC664.07c PT-ACRAMTU(PN) 0.86225 0.934 N S. pombe rad22 SPAC30D11.10 PT-ACRAMTU(PN) 0.80975 0.9225 N S. pombe rad13 SPBC3E7.08c PT-AMIDINE(EN) 0.885 0.7105 Y S. pombe rad1 SPAC1952.07 PT-AMIDINE(EN) 0.69525 0.8455 N S. pombe rad26 SPAC9E9.08 PT-AMIDINE(EN) 0.684 0.8315 N S. pombe rad9 SPAC664.07c PT-AMIDINE(EN) 0.86225 0.8205 N S. pombe rad22 SPAC30D11.10 PT-AMIDINE(EN) 0.80975 1.0435 N

35

Table 2-2. Confirmations of individual mutant strain sensitivity to compound treatment using on solid media dilution spotting. Individual mutant strains in S. cerevisiae were grown in 200 µM of PT-ACRAMTU(EN), PT-ACRAMTU(PN), PT-ATUCA and PT-AMIDINE(EN) for 24 h and spotted onto YPD media. Grey indicates mutants with low growth, blue indicates mutants with no growth, and a strikethrough indicates that the mutant strain did not grow in the absence of compound.

PT-ACRAMTU(EN) PT-ACRAMTU(PN) PT-ATUCA PT-AMIDINE RAD5 THI3 RAD5 RAD54 RAD5 IMP2' RAD5 IMP2' TRP5 VPS15 TRP5 HPR5 RAD14 GCN3 RAD14 RAD7 RAD14 RAD54 RAD4 RAD51 RAD4 TRP3 RAD4 RAD23 RAD4 HPR5 RAD2 VPS16 RAD2 ARO2 RAD2 MMS4 RAD2 RAD51 RAD10 VPS33 RAD10 YPK1 RAD10 MUS81 RAD10 MSS11 CHC1 PEP3 RAD1 TRP1 RAD18 YBR099C CHC1 YDR154C RAD18 CTF4 THI3 TRP2 RAD1 MED7 KIP3 XRS2 RAD1 PDX3 RAD54 PHO85 RAD54 TFG1 RAD18 VPS16 THI3 RAD55 SAC1 ERG2 HPR5 VAS1 RAD1 YMR166C VPS15 TRP4 SWI3 CTK3 RAD51

36

Figure 2-5. Gene set enrichment analysis for DNA-damaging platinum-acridines. Biological processes associated with sensitivity to four platinum-acridine compounds: a) PT- ACRAMTU(EN), b) PT-ACRAMTU(PN), c) PT-ATUCA and d) PT-AMIDINE(EN). Each node represents a biological process significantly enriched amongst non-essential genes associated with sensitivity to a platinum-acridine compound (FDR ≤ 0.1). The size of a node is proportional to the level of significance at which the process is enriched [i.e. proportional to – log10(FDR)]. The width of an edge is proportional to the level of gene overlap between the two connected processes. Edges are not shown where the overlap coefficient is less than 0.5. The color of a node shows cluster membership, where clustering is based on the level of overlap between processes and thus groups together related processes.

37

AMIDINE(EN) (the first member of this class able to slow NSCLC tumour growth in a murine xenograft149) the thiourea linkage was replaced with an amidine group. This seemingly minor modification resulted in greatly accelerated platinum-DNA binding160 and reduced reactivity with cysteine sulphur, a contributor to cisplatin toxicity and tumour resistance161. The fitness profile for this promising lead compound featured sensitive strains deleted for genes involved in NER, HRR and PRR. A distinct feature of this profile is that two genes essential for stalled-fork repair, MUS81 and MMS4, appeared in the top 15 most sensitive strains (Fig. 2-3). Stalled-fork repair is known to be important for the cellular response to most platinum agents, yet only for PT-AMIDINE(EN) do these deletion strains rank so highly (Fig. 2-4). The PT-AMIDINE(EN) and PT-ATUCA fitness profiles were most similar to each other and are found in the same cluster as cisplatin, PT-ACRAMTU(EN) and PT-ACRAMTU(PN) (Fig. 2-2) - all of these compounds elicit responses from DDR genes. PT-ACR49NME2, where the intercalator group is modified with a DNA threading 4-carboxamide group, showed no significant gene enrichment, consistent with its low levels of DNA-binding and modest activity against cultured cells159.

Platinum bisintercalators were generated by replacing the chloro leaving group in PT- ACRAMTU(EN) or PT-ACR49NME2 with an ACRAMTU or ACR49NME2 moiety, respectively, yielding C/T-PT-BIS(ACRAMTU) and C/T-PT-BIS(ACR49NME2) (Fig. 2-1). These four derivatives bind to DNA in a reversible manner by simultaneously inserting their two acridine moieties into the base stack147,162. Likely because these compounds lack a suitable leaving group and therefore do not form permanent adducts with DNA, we did not identify genes involved in DNA repair pathways in our screens. Instead, for C-PT-BIS(ACRAMTU) and C-PT- BIS(ACR49NME2) the most sensitive strains are those deleted for genes involved in vesicle- mediated transport (AKR1, ARF1, ARL3, CHC1, MON2, PEP3, PMR1, RVS161, SUR4, SYS1, VPS15, VPS16, VPS33, VPS35, YPK1, YPT31) suggesting that transport and/or detoxification of these compounds is required for cellular resistance. The trans-isomers of these compounds, T- PT-BIS(ACRAMTU) and T-PT-BIS(ACR49NME2), did not yield any significantly sensitive hits in the screen, which may indicate limited uptake of the bulky compounds by the cells.

38

2.2.3 Genome-wide profiling of DNA-damaging platinum-acridines in S. pombe

The recent availability of a genome-wide deletion set of S. pombe mutants allowed us to perform, for the first time, comparative chemogenomic characterization of a set of compounds in pooled culture. We used S. pombe to examine the mechanisms of cisplatin and the four platinum- acridine compounds (PT-ACRAMTU(EN), PT-ACRAMTU(PN), PT-ATUCA, and PT- AMIDINE(EN)) that reveal DNA-damaging effects in S. cerevisiae. In genome-wide screens of S. pombe, resistance to cisplatin showed a requirement for NER, HRR and PRR genes (Fig. 2-4), similar to that seen in S. cerevisiae. We observed that the unique requirement for TLS (rev3) for cisplatin resistance was also conserved in S. pombe (Fig. 2-4). In contrast, cisplatin was the only compound with PRR genes in the top 30 most-sensitive strains. The S. pombe deletion collection is missing 20% of genes as deletion strains, including the Δpso2 strain, so its role in cisplatin resistance remains to be tested. The top 250 sensitive strains in the PT-ACRAMTU(EN) and PT- ACRAMTU(PN) profiles included those deleted for genes in NER, HRR and PRR (Fig. 2-3, 2- 4). In contrast to S. cerevisiae, HRR (rad22) ranked higher than other repair modules in response to PT-ACRAMTU(EN) (Fig. 2-4a). Furthermore, the requirement for intact PRR was less prominent than for S. cerevisiae, while the HRR requirement for resistance to PT- ACRAMTU(PN) treatment was conserved. Similar to S. cerevisiae, tolerance to PT-ATUCA required NER, HRR and PRR and the ranking of DDR genes was not as high as for the other compounds. PT-AMIDINE(EN) induced distinct profiles in S. pombe and S. cerevisiae; in S. pombe, only those strains deleted for genes involved in NER and PRR were required for resistance. The most striking difference between the profiles for these two organisms is that S. pombe does not seem to require the stalled-fork repair gene mus81 for resistance to either cisplatin or the four DNA-damaging platinum-acridine agents. Selected strains from our genome- wide S. pombe screens were tested for their sensitivity in monoculture with a 35% confirmation rate (Table 2-1). The current state of S. pombe gene annotation is too sparse to permit GSEA analysis.

2.2.4 Effect of platinum-acridines on DNA-replication

In light of the results from both yeast, we asked if these agents perturb cell cycle progression. I used flow cytometry to measure DNA content in synchronized wild-type haploid S. cerevisiae

39

cells that were released into each of PT-ACRAMTU(EN), PT-ACRAMTU(PN), PT-ATUCA and PT-AMIDINE(EN). The compounds PT-ACRAMTU(EN), PT-ACRAMTU(PN) and PT- ATUCA treatment slowed progression through the cell cycle (Fig. 2-6). Cells treated with PT- ATUCA accumulated in G1 and S phases, with only a small population of cells in G2/M at 60 minutes. Cells grown in PT-AMIDINE(EN) initially accumulated in S phase and only progress through to G2/M after 90 minutes. Cells remained arrested in G2/M and were unable to divide and re-enter the G1 phase of the subsequent cell cycle. These cells show a distinct metaphase arrest as a result of PT-AMIDINE(EN) treatment (Fig. 2-7).

The observation that the S. cerevisiae profile of PT-AMIDINE(EN) shows an enrichment of genes involved in the replication of stalled DNA replication forks, and that the cell cycle data shows slowed progression through S-phase and then accumulation at G2/M, motivated us to further examine the effect of this compound on DNA replication. To examine DNA replication fork progression directly, we performed molecular combing163 using synchronized S. cerevisiae cells. These cells, which are genetically modified to contain several copies of the Herpes Simplex Virus thymidine kinase gene, were released from G1/S synchrony into medium containing the thymidine analogue BrdU and PT-AMIDINE(EN) at an IC50 or into vehicle (PBS) for 30 min to allow BrdU incorporation into newly-synthesized DNA. David Gallo prepared genomic DNA and combed the DNA fibres on silanized coverslips163. Replication fork progression was examined by measuring the length of BrdU tracks in the fibers. In cells treated with PT-AMIDINE(EN) the nascent DNA tracks, as indicated by BrdU labeling, were significantly shorter compared to cells treated with the vehicle (Fig. 2-8). This observation is consistent with PT-AMIDINE(EN) interfering with replication fork progression in vivo.

2.2.5 Effect of platinum-acridines on mitochondria

I further explored the cellular effects of these compounds by examining mitochondrial function and integrity following treatment. Mitochondria are often subjected to high levels of DNA damage when treated with DNA-targeting drugs. Furthermore, mitochondria are an attractive target for cancer therapy due to their role in cellular energetics and apoptosis, and their derangement in different cancers. To examine whether the four DNA-damaging platinum- acridines interfere with mitochondrial function, wild-type S. cerevisiae cells were grown in

40

Figure 2-6. DNA content analysis profiles. Haploid BY4741 cells were synchronized at G1 prior to the addition of compound at IC50. Samples were taken every 30 min after compound addition. The positions of the 1N and 2N DNA content peaks are indicated.

41

Figure 2-7. Microscopy confirms G2/M arrest of cells treated with PT-AMIDINE(EN). Cells treated with either PBS or 200 µM PT-AMIDINE(EN) for 120 min. Cells were stained with DAPI to visualize the nuclei and imaged at 63x magnification.

42

Figure 2-8. PT-AMIDINE(EN) treated cells show defects in replication fork progression by DNA combing. Logarithmically growing cultures of S. cerevisiae cells (log) were arrested in G1 with -factor ( F) and released in the presence of 400 µg/mL BrdU and either PBS or PT- AMIDINE(EN) (PT-AM) at IC50. Samples were collected at 30 min and used in subsequent analysis. a) DNA content analysis. Samples were fixed and DNA contents were analyzed using flow cytometry. The positions of cells with 1C and 2C DNA contents are indicated. b) DNA combing. Representative chromosome fibers used for replication fork progression analysis. The image is assembled from fibers on different micrographs following extraction of fibers from the non-fiber background using Adobe Photoshop. A 50 kbp scale bar is indicated in the upper right corner. c) Distributions of BrdU track lengths in PBS or PT-AMIDINE(EN) treated cells, presented as a boxplot. Median BrdU track lengths are shown. The p-value was determined using a two-tailed Mann-Whitney U test.

43

a b

Figure 2-9. Mitochondria are disrupted by platinum-acridine compounds. (a) Wild-type BY4743 cells were grown in the presence of four platinum-acridine compounds under fermenting (blue) or respiring (red) conditions. The concentration to inhibit growth by 20% (IC20) is plotted. (b) Platinum-acridine compounds disrupt mitochondrial morphology. Yeast cells treated with PBS or PT-ACRAMTU(EN) for 6 h were stained with Mitotracker CMXRos to visualize the mitochondria. Cells were fixed and imaged at 100x magnification.

44

medium that requires mitochondrial respiration (YPGE) or in a fermenting medium (YPD). In all cases, the inhibitory dose of each platinum-acridine on the cells was much lower in YPGE than YPD (Fig. 2-9a) indicating that lack of functional mitochondria sensitizes cells to these agents. To examine whether these compounds affect mitochondrial morphology, compound-treated cells were stained with Mitotracker Red CMXRos which is an intrinsically fluorescent dye that accumulates in active mitochondria because of their negative membrane potential164. CMXRos contains a thiolreactive chloromethyl group which allows the mitochondria to retain the dye after aldehyde-fixation. For each compound tested, mitochondrial morphology changed from a filamentous network to clumped vesicles (Fig. 2-9b). Together, these results indicate that the four DNA-damaging platinum-acridines affect mitochondrial function.

2.3 Discussion

In this study, we use a validated genome-wide screen to characterize novel anti-cancer therapeutics and explore their structure-activity relationships in an unbiased manner. The utility of S. cerevisiae gene deletion studies for DNA-damaging agents has been exemplified by several published studies that provide a comprehensive view of DNA repair mechanisms95,156,165. The results from our screens add to these reference datasets by providing a global view of the cellular response to novel agents and new insight into the activities previously attributed to these platinum-acridines including their ability to bind to DNA and their lack of a classical cross- linking mechanism for producing DNA-damage145,148,149,158. By classifying the DNA repair modules that were required for resistance to four of the platinum compounds, we revealed new details about their potential mechanisms at the nuclear level. The response from genes involved in NER strongly suggests that adducts generated by PT-ACRAMTU(EN) primarily lead to helical distortions in the DNA166, while modification of the non-leaving group to produce PT- ACRAMTU(PN) yielded a compound that appears to damage DNA via DNA double-strand breaks, a more severe form of DNA damage produced by either cisplatin or PT- ACRAMTU(EN)167. A distinctive pattern of resistance was seen in the profile of PT- AMIDINE(EN). Specifically, we found a requirement for genes involved in stalled-fork replication repair, a process which can be inhibited by DNA-damaging compounds with very different modes of action, such as alkylation168, topoisomerase poisoning169 and ribonucleotide reductase inhibition170. In addition, PT-AMIDINE(EN), the most DNA-reactive compound in

45

this series, causes cells to initially accumulate in S phase, a delay consistent with PT- AMIDINE(EN) preventing or disrupting DNA synthesis, as demonstrated in NCI-H460 lung cancer cells171. While the three other DNA-damaging hybrid compounds featured here also slow progression of cells through S phase, the cells are able to complete the cell cycle. Our molecular combing observations suggest that the S phase delay observed with PT-AMIDINE(EN) treatment is due, at least in part, to replication fork stalling at DNA adducts formed by this compound. In this scenario, the resumption of DNA synthesis without repair would lead to the observed cell cycle arrest at G2/M due to metaphase arrest and eventually lead to cell death. The hypothesis that severe and irreparable lesions caused by PT-AMIDINE(EN) are bypassed by DNA repair machinery frames future mechanistic studies. In addition, the mitochondrial disruption caused by each of these compounds present them as interesting candidate chemotherapeutics51. For example, a minor modification to the PT-ACRAMTU(EN) structure produced a compound with a cytotoxicity profile far superior to that of current platinum drugs149. Such pharmacophores should be of considerable interest for future clinical development.

In addition to revealing differential requirements for diverse DNA repair genes, our screens also highlight that for several platinum-acridines, the mechanism by which they interfere with cellular fitness does not (primarily) involve DDR pathways. For example, the moderately cytotoxic PT- ACRAMTU(BPY) has shown some degree of coordinative binding with DNA in vitro, yet its fitness profile did not suggest that DNA damage results. Similar results were observed for the bis(ACRAMTU) compounds, which enter cells and inhibit cancer cell proliferation at micromolar-to-submicromolar concentrations172. However, the genome-wide profiles of C-PT- BIS(ACRAMTU) and C-PT-BIS(ACR49NME2) suggest that these highly charged compounds are actively detoxified by the cell and that this response is the primary means of cellular resistance. We also confirmed that bisintercalators likely do not form permanent DNA adducts based on our observation that DNA repair deficient strains are not sensitive to these analogues.

The genomes of S. pombe and S. cerevisiae share many features, however their evolutionary divergence allows for complementary functional genomic studies, particularly for examining conserved biological processes such as DNA-damage response pathways and the cell cycle, which are well-characterized in both yeast species173. Our genome-wide interrogation of S. pombe showed an array of shared and distinct cellular responses to DNA-reactive agents. The 46

hypersensitivity of translesion repair mutants to cisplatin treatment in both yeasts emphasizes the importance of this pathway to repair DNA cross-links, while its absence in the platinum-acridine profiles further highlights the different mechanism by which they function. The PT- AMIDINE(EN) profiles in S. cerevisiae and S. pombe were similar, however, the absence of HRR deletion strains as hits in the S. pombe screen, alludes to the presence of alternative mechanisms for resistance against PT-AMIDINE(EN)-induced double-strand breaks. Additional studies using the fission yeast deletion collection will initiate a productive cycle of gene annotation, in the same manner in the accumulated genome-wide data have greatly increased our understanding of gene function in budding yeast (for review see Hillenmeyer et al.174). The current study demonstrates the power of chemogenomic screening in delineating structure– activity relationships in a new class of DNA-targeted anticancer agents. The requirement for DNA-repair modules for survival of evolutionary divergent yeast cells treated with the platinating-intercalation pharmacophore unequivocally demonstrates that DNA is the principal target of these agents and gives confidence that these results can be applied in human studies. Unlike NSCLC, several cancers may be relatively insensitive to platinum-acridines. To overcome resistance in these tissues, structural designs are currently pursued which feature platinum-acridines as cytotoxic “warheads” conjugated to molecules that modulate DNA repair and/or elicit pro-apoptotic responses. Comparative chemogenomics will be an invaluable tool for studying these novel multifunctional entities.

2.4 Methods

2.4.1 Reagents

Cisplatin was purchased from Sigma-Aldrich. All other compounds screened were generated according to published procedures: ACRAMTU, PT-ACRAMTU(EN)145, ACRAMGU175, PT- ACRAMTU(PN/TMEDA/BPY)148, PT-AMIDINE(EN)149, PT-ACR49NME2159, PT-ATUCA176, C/T-PTBISACR, C/T-PTBISACR49NME2172.

2.4.2 Yeast strains and media

The S. cerevisiae deletion collection87 and the S. pombe deletion collection177 were used for this study. S. cerevisiae BY4741 and BY4743 strains were used for small-scale analyses. S.

47

cerevisiae cells were grown in YPD (yeast extract/peptone/dextrose) and S. pombe cells were grown in YES (yeast extract with supplements) at 30oC.

2.4.3 Deletion pool growth and chip experiments

Chemical genomic screens of the homozygous and heterozygous S. cerevisiae deletions pools were performed as described178. The same protocol was used for the S. pombe deletion pool. The concentrations of the compounds screened were determined from using dose-response analysis on a wild-type strain. The platinum-acridine compounds were applied at a concentration that inhibits wild-type growth by 15% (200 µM for most compounds; Fig. 2-10). Genomic DNA preparation, PCR and chip hybridization were performed for both yeast as described previously86 with the following change: S. pombe samples were hybridized to the UT-GMAP microarray118.

2.4.4 Data analysis

All microarray data have been deposited with Array Express, accession no. E-MTAB-1267. Using the GeneChip Operating Software (Affymetrix), intensity values for the probes on the chip were extracted. For each strain, fitness defect ratios were calculated as the log2 of the ratio between the mean signal intensities of the control and the drug chips. The larger ratio means higher sensitivity of the strain as compared to control condition without the drug.

2.4.5 Comparing genome-wide profiles

The algorithm used for comparing genome-wide profiles179 is based on ranking of strain sensitivities and comparing a large number of full-genome profiles. Because the scale of fitness defect values of a profile varies between profiles, a rank-based correlation of these values is needed. We address this challenge by scoring each profile as a series of sections, dividing each profile‟s gene scores into sections, or “buckets” and bucket sizes are modified according to significance. The smallest size bucket (0.05% of all genes) contains the most significant genes (higher fitness defect scores) and the bucket sizes increase exponentially such that the larger size buckets contain the least significant genes, meaning those with the lower fitness defect score. The bucket number and sizes are the same for all the screens performed.

Once the genes of each profile are divided into buckets, we use a scoring matrix for scoring similarity between profiles. The scoring matrix is formulated in decreasing scores while taking 48

into account the bucket location of each gene. When comparing profiles, the score matrix yields the score of Si, j to a gene located in bucket i and bucket j in each of the profiles compared. For a score of , the scoring matrix follows these guidelines:

i, j | i j S S i,i j, j

i, j,k | i j k S S i, j i,k

Runs on a variety of datasets shows that the algorithm surpasses batch effects that are common in large-scale array-based experiments.

Two-dimensional hierarchical clustering was conducted using the R statistical software package with Euclidean distance metric and the complete linkage agglomerative method.

2.4.6 Gene set enrichment analysis (GSEA)

GSEA180 was used to identify biological processes enriched amongst genes associated with sensitivity to platinum-acridine compounds, when individually deleted in homozygous deletion strains. Strains in the chemogenomic profile of each compound were mapped to genes using chromosomal feature data downloaded from the Saccharomyces Genome Database (http://www.yeastgenome.org) on June 8, 2012, and the resulting profiles were analyzed by GSEA v2.07 in pre-rank mode (Java implementation). All default parameters were used except that the minimum and maximum gene set sizes were restricted to 5 and 300, respectively. Our gene sets were defined with Gene Ontology biological process gene annotations were obtained from Saccharomyces Genome Database on May 26, 2012.

The enrichment maps in Figure 2-5 were generated with the Enrichment Map Plugin v1.2181 developed for Cytoscape182. All default parameters were used. For each node (i.e. enriched gene set) in each map, we computed significance = -log10(FDR) where FDR was estimated by GSEA. For nodes where significance equals infinity, significance was changed to equal 2 + the maximum non-infinite significance value in the given map. Node sizes were changed to be proportional to significance. In addition, the nodes in the map were clustered with the Markov clustering algorithm183, using the overlap coefficient computed by the plugin as the similarity

49

metric (coefficients less than 0.5 were set to zero) and an inflation of 2. Node colors were changed to indicate cluster membership.

2.4.7 DNA content analysis

-1 Wild-type S. cerevisiae BY4741 cells at an OD600 0.1 were arrested in G1/S using 2 µg mL α- factor. After incubation for 2.5 h at 23 oC, cells were released from G1/S or G2/M and grown for o an additional 2 h in the presence of compound at an IC10 at 30 C. Samples were collected at half-hour time points, fixed in 70% (v/v) ethanol and resuspended in 0.2 mg mL-1 RNase A in 50 mM Tris pH 8.0 for 2 h at 37 oC. Cells were resuspended in 2 mg mL-1 proteinase K in 50 mM Tris pH 7.5 and incubated for 30 min at 50 oC before resuspension in FACs buffer (200 mM Tris pH 7.5, 200 mM NaCl and 78 mM MgCl2). The cell suspension was diluted 100x into 1x SYBR Green (Invitrogen) in 50 mM Tris pH 7.5 and sonicated briefly. DNA content analysis was performed on a FACSCalibur flow cytometer (BD Biosciences, USA) and analysis was performed using FlowJo software.

2.4.8 Molecular combing

S. cerevisiae E1670 (MATa ade2-1 trp1-1 can1-100 his3-11,15 leu2-3,112 RAD5+ GAL psi+ ura3::URA3/GPD-TK(7x)) cultures were grown to OD600 = 0.25 and arrested in G1 by addition of 2.5 µM -factor for 75 min at 23°C, followed by an additional aliquot of 1 µM -factor for 75 min. 400ug/mL BrdU was added to each culture 15 minutes prior to compound addition. Cultures were divided in two and released from G1 by addition of 100 µg mL-1 pronase (Sigma) and either PBS or PT-AMIDINE(EN) (350nM). Samples were collected after 30 min, made 0.1% (w/v) sodium azide, and incubated for 10 min on ice. Parallel samples were fixed in ethanol and analysed by flow cytometry, as described above. The OD600 was taken and cell 7 -1 concentration was calculated using the formula 1 OD600 = 1 x 10 cells mL . Cultures were pelleted by centrifugation at 3000 g for 3 min, washed once in 1 mL ice-cold wash buffer (10 mM Tris-HCl, 50 mM EDTA pH = 8.0), and pelleted at 13 000 g for 1 min in a microfuge tube. Pellets were re-suspended in SCE buffer (1 M sorbitol, 100 mM sodium citrate pH = 8.5, 10 mM EDTA pH = 8.0, 0.126% (v/v) -mercaptoethanol, 10 U mL-1 zymolase (BioShop)) to a final concentration of 6 x 108 cells mL-1. The cell suspension was mixed with equal volume of 1% low melt agarose (BioShop) and 100 µL was added to plug molds (BioRad) and incubated at 4 ˚C for

50

45 min to allow the agarose to solidify. Plugs were ejected and incubated in 0.5 mL per plug of SCE buffer at 37 ˚C for 48 hrs, with fresh SCE buffer added after 24 hrs. Plugs were rinsed 3x in 1 mL 0.5 M EDTA and incubated in 0.5 mL per plug of ProK buffer (0.5 M EDTA, 1% (w/v) N- lauryl sarcosyl, 1 mg mL-1 proteinase K (BioShop)) for 72 hrs, with fresh ProK buffer added every 24 hrs. Plugs were rinsed 3x in 1 mL 0.5 M EDTA and washed 3x 30 min in 10 mL TE with 0.2 M PMSF, then with 10 mL TE alone overnight. The following day one plug per sample was transferred to a new tube and washed 2x 30 min in 10 mL TE followed by incubation in 6.7 µM YOYO-1 (Invitrogen) for 30 min at room temperature. Plugs were rinsed 3x in 1 mL 0.5 M EDTA and melted in 150 µL TE for 20 min at 68°C. 2 mL of 100 mM MES (pH = 6.0, pre- warmed to 68°C) was added and plugs were incubated at 68°C for another 20 min. The agarose solution was then treated with 3 µL -agarase (NEB) at 42 ˚C overnight. The following day solutions were warmed to 68°C for 20 min and cooled back to room temperature. DNA combing and detection with anti-BrdU and anti-DNA antibodies was performed as described184. DNA fibers were imaged using an Axiovert inverted microscope (Carl Zeiss, USA) with a 63x objective. Individual coverslips were blinded before image acquisition to avoid bias in the analysis. Images were processed to maximize signal intensity and fluorescent tracks were measured in Adobe Photoshop. Approximately 175 tracks were measured per sample and track lengths were converted from pixels to kbp using a conversion factor based on combing λ DNA163. Experiments were repeated twice and data from independent experiments pooled. The distribution of track lengths was plotted as a boxplot and the two-tailed Mann-Whitney U test was used to compare the distributions of track lengths.

2.4.9 Microscopy

To examine the mitotic index of cells treated with PT-AMIDINE(EN), wild-type BY4743 cells were treated with either PBS or 200 µM PT-AMIDINE(EN) for 120 min. Cells were resuspended in VECTASHIELD mounting media containing DAPI (Vector Laboratories, CA) and imaged at 63x magnification on an Axiovert inverted microscope. To examine mitochondrial morphology, cells treated with PBS or 200 µM PT-ACRAMTU(EN) for 6 h were stained with 500 nM Mitotracker Red CMXRos (Invitrogen, USA) for 1 h, fixed with 3.7% (v/v) paraformaldehyde and imaged at 100x magnification on a Zeiss Axiovert inverted microscope.

51

Figure 2-10. Growth responses of the yeast deletion pools to compound. The S. cerevisiae and S. pombe deletion collections were grown for 24 h in a) all 14 platinum-acridines and carriers (S. cerevisiae) or b) 4 DNA-active platinum-acridines (S. pombe). Compound name and concentrations used are labelled for each plot. The x-axis represents time and the y-axis represents optical density at 600nm (OD600). The black lines indicate pool growth in compound and the red lines indicate pool growth without compound. Ratio of average growth rate with and without drug is listed in the bottom right corner of each plot.

52

Chapter 3 Mitochondrial electron transport is the cellular target of the oncology drug elesclomol

The material presented in this chapter is published in Blackman RK, Cheung-Ong K et al. (2012) Mitochondrial Electron Transport Is the Cellular Target of the Oncology Drug Elesclomol. PLoS ONE 7(1): e29798. doi:10.1371/journal.pone.0029798.

Contributions: I performed the analysis and confirmations of hits from the comparative growth assays and tests of elesclomol dosage against 143B cells. Marinella Gebbia performed the screens and copper tests. Anna Lee performed the GSEA analysis. Our collaborators at Synta performed the ETC inhibitor and flow cytometry experiments.

3 3.1 Introduction

Elesclomol is a novel small molecule drug originally identified in a cell-based screen for its potent proapoptotic activity in cancer cells. More recently, in vitro studies indicated that it strongly induces reactive oxygen species (ROS) within tumor cells185, leading to unmanageable levels of oxidative stress and consequent apoptosis. However, the cellular target of elesclomol remained unknown, as did the molecular mechanism by which it generated ROS.

Increased levels of ROS and an altered redox status have long been observed in cancer cells186, where constitutively elevated oxidative stress and dependence on antiapoptotic ROS signaling represent potential vulnerabilities of tumors that can be exploited by small molecule drugs187. To test its usefulness as a therapeutic, elesclomol has been evaluated in human clinical studies188,189 and combined data from three randomized Phase 2 and 3 trials have demonstrated therapeutic benefit, including prolonged progression-free survival, in a subset of the patients treated190. Interestingly, this subset was distinguished by the prevalence of subjects with low baseline levels of serum lactate dehydrogenase (LDH). LDH level is known as a prognostic marker for outcome in cancer, but for elesclomol, it also appears also to be a predictive marker of efficacy190. 53

Understanding the relationship between clinical benefit and serum LDH level at a molecular level would greatly assist the continued clinical development of the drug, including the identification of patients most likely to benefit.

As part of the continuing effort to better understand elesclomol‟s cellular mechanism of action (MoA), Nagai et al. found that the compound binds strongly to copper (Figure 3-1A) and that this binding is required for its cell killing activity191. When administered to humans, elesclomol acquires the needed copper ions (in the form of Cu2+) while in the bloodstream. Copper binding changes the conformation of the drug192, facilitates its uptake into cells, and allows it to undergo redox cycling [Cu(II) to Cu(I)] to generate ROS inside the cell191. In the absence of bound copper, the compound has no discernible activity.

To further refine elesclomol‟s MoA, we made use of powerful comparative growth assays available in the yeast Saccharomyces cerevisiae, specifically the drug-induced haploinsuffiency and homozygous profiling assays which are described in the introduction section of this thesis. While performed in yeast, this strategy has in previous studies (particularly for cancer drugs) identified relevant mechanistic details that are indicative of the drugs‟ actions in humans93,94,96,193,194. Briefly, the assay uses a set of diploid yeast deletion strains comprised of ~1100 essential heterozygous deletion strains and ~4800 non-essential homozygous deletion strains. These deletion strains are grown in the presence of the drug of choice, here elesclomol, at sub-lethal doses. Deletion mutations that render the cell more sensitive to treatment serve to elucidate the MoA of the drug (for review, see Smith et al., 201091).

Using this strategy, we found that elesclomol does not work through a specific cellular protein target, but rather works primarily by influencing redox reactions associated with the mitochondrial electron transport chain (ETC). This disruption leads to increased levels of ROS in the organelle and subsequently to cell death. We also show that this mechanism is conserved between yeast and human cancer cells. Taken together, these results provide a convincing explanation for the efficacy of elesclomol in low LDH patients, and support a strategy of patient stratification in future clinical work with the drug.

54

Figure 3-1. Elesclomol-induced ROS generation and cytoxicity in yeast is dependent on the presence of copper. (A) Chemical structure of the elesclomol-Cu complex. (B) The parental BY4743 yeast strain was grown in the presence of the indicated concentrations of elesclomol, preformed elesclomol-Cu, and/or copper for 21.5 h at 30°C. Absorbance at 600 nm was used to determine cell density. (C) Flow cytometric analysis measuring ROS in S. cerevisiae. ROS induction, as measured by Dihydrorhodamine 123 fluorescence, was only observed with preformed elesclomol-Cu complex at a concentration (500 nM) above the MIC but not below (100 nM), nor with free elesclomol at either concentration (upper panel). The addition of supplementary copper (via CuCl2) to elesclomol was sufficient to induce ROS, again only at the higher concentration (lower panel). (D) Elesclomol is cidal to yeast cells within an hour of treatment. Logarithmically growing cells were incubated with the indicated doses of elesclomol for 1, 2 or 4 h and then plated onto media without elesclomol. 5 µM and 22.5 µM elesclomol rendered cells unviable within 1 h, and lower doses (1.25 µM) killed cells within 4 h.

55

3.2 Results

3.2.1 Yeast are sensitive to elesclomol treatment only in the presence of copper

We initially examined whether elesclomol treatment affected the growth of yeast. Because in mammalian cells elesclomol is only active when bound to copper191, we performed the growth analysis of the wild-type BY4743 strain (the parent of the deletion strains) in the presence of elesclomol alone, elesclomol plus varying concentrations of copper chloride (CuCl2), or with preformed elesclomol-Cu complex (Figure 3-1B). The cultures with elesclomol plus copper or elesclomol-Cu showed potent growth inhibition with a minimum inhibitory concentration (MIC) in the 250–500 nM range. In contrast, elesclomol without added copper had no effect on yeast growth at concentrations up to 200 µM, while CuCl2 on its own had no effect at concentrations up to 2 mM, the latter consistent with a previous report195.

Given the requirement for copper to induce ROS in cultured cells, we also tested whether elesclomol plus copper could induce ROS in yeast (Figure 3-1C). As with mammalian cells, ROS was strongly induced only in the presence of copper (either by copper supplement or as part of a preformed complex), but only at concentrations above the MIC. No ROS was observed at 100 nM elesclomol plus copper or elesclomol-Cu, consistent with ROS induction being required for growth inhibition.

Finally, we tested whether elesclomol treatment led to cell death (cidality) or simply growth arrest. Specifically, we grew wild type cells with elesclomol-Cu at varying concentrations for 1 to 4 hours with constant shaking at 30°C. Following treatment, cells were washed in fresh media and spotted onto solid growth medium without drug (Figure 3-1D). Elesclomol-Cu at 1.25 µM prevented colony formation after 3 h of treatment, while 2.5 µM was cidal after 1 h exposure. Taken together, these results show that elesclomol in the presence of copper, but not alone, potently kills cells after strongly inducing ROS.

3.2.2 There is no cellular protein target for elesclomol

Given the conserved nature of the responses to elesclomol treatment, we reasoned that a comprehensive screen of the yeast deletion collection could reveal detailed insights into

56

elesclomol‟s mechanism of action in both yeast and mammalian cells. For our assays, we used sub-MIC doses of 300 and 400 nM elesclomol-Cu, concentrations that were determined empirically to inhibit growth of the BY4743 strain by 10–20% under the conditions of the library screening. Triplicate biological replicates were then performed at both doses. An analysis of the data (the log2 ratio of the normalized strain signal in treatment vs. DMSO control) showed that the results were similar at both drug concentrations, so we combined all six datasets by averaging the log2 ratios to yield a single set of responses (Appendix Table 3-1). The combined values were used for all analyses presented here.

As we showed previously for other drugs, if elesclomol affects cell growth by interacting or interfering with a specific protein in the cell (i.e., its protein target), we expect that the heterozygous strain deleted for that gene would be highly sensitive to the drug treatment93,94. In the plot shown in Figure 3-2A, the further up the y-axis, the more sensitive the strain, with the most sensitive strain representing the likely drug target93. In our experiments, the nus1 deletion strain was the only heterozygous deletion strain that scored as sensitive (log2 ratio >2). NUS1 encodes a putative prenyltransferase implicated in intracellular trafficking. However, the magnitude of the strain‟s sensitivity to elesclomol was modest compared to other drug-target combinations we have analyzed in the past96,196. Moreover, the nus1 heterozygote manifests a general sensitivity to a variety of compounds and conditions96, suggesting that its sensitivity to elesclomol-Cu is not target related. Taken together, these data support the idea that yeast lack a specific protein target through which elesclomol exerts its cytotoxic activity.

3.2.3 Elesclomol targets electron transport activity in the mitochondrion

In contrast, the sensitive strains among the homozygous deletions formed a biologically coherent set, consistent with elesclomol interacting with a specific target pathway (Figure 3-2B). Nearly all of the 48 strains with a log2 ratio >2 have a role in mitochondrial function (Appendix Table 3-

1). At a less stringent log2 ratio cut-off of 1.1, ~80% of the 190 genes are involved in mitochondrial activities. Closer inspection of these 190 reveals several interesting classes of genes. Notably, genes that are involved in diverse functions of electron transport, including structural components of the ETC (PET309, COB1, SHY1, COQ10, YER077C, COX12, COX9,

57

Figure 3-2. Sensitivity of S. cerevisiae mutant strains to elesclomol-Cu. (A) A genome-wide readout of heterozygous strain sensitivity. In the plot, the X-axis orders all genes by their systematic name (hence, chromosome position) while the Y-axis is a measure of the „fitness‟ of the strain deleted for the indicated gene grown in sub-lethal doses of elesclomol-Cu. The value on the Y-axis corresponds to −log2 (ratio of normalized strain signal in treatment to DMSO control). Hence, the zero line represents equivalent growth in both conditions, while each unit above the line represents a 2-fold reduction in strain fitness. (B) Genome-wide profile of homozygous strain sensitivity. The data are presented as in (A). Among the 150 most sensitive strains, the 137 having mitochondrial roles are highlighted color-coded: orange (7 genes, mitochondrial genome maintenance), yellow (8 genes, metal ion homeostasis), bright green (26 genes, mitochondrial localization), aqua (4 genes, mitochondrial, uncharacterized), blue (36 genes, ox-phos and respiration) mauve (5 genes, mitochondrial splicing), light gray (9 genes, response to stress), dark red (26 genes, mitochondrial translation), dark gray (7 genes, mitochondrial import/export) and purple (9 genes, mitochondrial tRNA). 58

QCR2, CYT1, BCS1, COQ9, QCR7) or involvement in the translation, modification or assembly of cytochrome components (OXA1, CBT1, COX20, PET54, MNE1, CYT2, COQ9, COX16, CBP2, PET117, COX18) showed sensitivity. In addition, 7 structural components of the F1F0 ATPase were sensitive as deletion alleles (ATP1,4,7,10,11,12,17). Several genes that comprise the mitochondrial ribosome or are directly involved in translation were identified, including: PET123, MRPL28,23,17,31,51,27,11,38,33 RSM22,19, IFM1, MTG2, MRPS16,17, and MEF2 as well as 7 of the 24 mitochondrial tRNA genes (MSM1, MSE1, MSY1, MSD1, MSF1, HER2, TPT1). The loss of these components would likely impair the production of the mitochondrially- encoded components of the ETC. Genes involved in copper-related functions were also identified, including COX23, an inner mitochondrial membrane gene required for copper homeostasis and cytochrome oxidase expression; COX11 which is required for delivery of copper to the Cox1p subunit of cytochrome oxidase; COX17, a mitochondrial copper metallochaperone; and CUP2, a nuclear transcription factor that is activated in the presence of copper. Finally, genes involved in the oxidative stress response were also sensitive as deletion strains, including several whose proteins are located within mitochondria: SOD2 (the mitochondrial superoxide dismutase), POS5, MGM101, PRE6, UTH1, and FMP46; as well as several reported to be present in the cytoplasm: APD1, OCA1, RIM11, CNB1, YFR039C, HSP150 and CCH1. Interestingly, deletion of SOD1, the cytoplasmic superoxide dismutase, was not sensitizing, but rather conferred a degree of resistance (see discussion below).

To examine the yeast data more systematically, the entire dataset was analyzed by gene set enrichment analysis (GSEA)180. GSEA provides an algorithmic tool to identify pathways and processes whose components are over-represented among the sensitive strains. This analysis evaluates all processes of the cell, and the only “supergroups” (major nodes) identified as significantly enriched are depicted in Figure 3-3A. The supergroup clusters in the figure highlight the connections between enriched functions within its biological category (with the connection strength indicated by the width of the lines connecting the functions). For each supergroup, the top 10 genes (or fewer in cases where fewer than 10 genes comprise that function) of the core gene set that is responsible for the enrichment score are shown in the histograms (Figure 3-3B) (see legend for details). The enriched clusters include: respiration

59

Figure 3-3. Biological processes and protein complexes associated with sensitivity to elesclomol. (A) Each node represents a significantly enriched biological process/protein complex in the elesclomol chemogenomic profile as determined by GSEA. The size of a node corresponds to the number of genes annotated to the functional category. The width of an edge corresponds to the level of gene overlap between two interconnected categories (i.e. gene sets). Edges are not shown where the overlap coefficient is less than 0.5 (see Materials and Methods). The color of a node shows the cluster membership where clustering is based on the level of overlap between categories. (B) Each bar plot corresponds to the cluster indicated by its border color, and shows the individual sensitivity scores (X-axis) of the genes that contributed to the functional enrichments of the cluster. For clusters with more than 10 genes contributing to the enrichments, only the top 10 associated with the most categories are shown. The percentage of times the particular genes occur in that gene set is indicated by a color key, with black indicating the gene is present every time that function appears enriched, graded to white for those genes in the leading edge that appear less frequently within the gene set.

60

complexes and respiration, mitochondrial membrane localization, mitochondrial tRNAs, mitochondrial genome maintenance, transport into/out of the mitochondria, splicing of mitochondrial transcripts, and metal ion homeostasis (Figure 3-3). Given that this analysis encompasses all biological processes and pathways described for the cell, it is striking that the only ones identified as significantly over-represented in our dataset involve mitochondrial processes or copper-related activities.

3.2.4 Elesclomol works by a distinct mechanism of action

The cellular response to elesclomol identified by the subset of sensitive strains provides a unique fingerprint of the system-level response to the drug. To quantify this observation, we compared the elesclomol profile to over 3,000 drug profiles (including 300 FDA approved drugs197) in our database (publicly available at (http://chemogenomics.med.utoronto.ca/hiplab/fitdb.php) and confirmed that the elesclomol profile was unique. In fact the most correlated profile showed only a modest similarity, r2 = 0.55. These heretofore uncharacterized compounds were derived from a synthetic compound library from ChemDiv Laboratories (San Diego, CA; compound IDs: 0352- 0636, 0141-0289, 0269-0018). In contrast to the elesclomol profile, these 3 compounds show enrichment for genes primarily involved in mitochondrial protein synthesis (see http://chemogenomics.med.utoronto.ca/supplemental/elesclomol/). These data strongly suggest that elesclomol acts by a novel mechanism not shared by any previously tested compound.

Notably, we previously examined the profile produced by overloading the cell with copper, i.e., 198 growth in the presence of 10 mM CuCl2 . The only overlap in the most sensitive strains from this treatment and that produced by elesclomol-Cu was the hypersensitivity of the cup2 homozygous deletion strain. This rather limited overlap presumably reflects the fact that copper overload in yeast is only seen at concentrations 25,000 fold greater than that used here for elesclomol-Cu (10 mM vs. 400 nM). This strongly suggests that the cellular activity of the elesclomol-Cu complex is entirely distinct from that produced solely by copper toxicity.

3.2.5 Elesclomol interacts similarly with the ETC in human cells

Among our sensitive yeast strains, combinations of elesclomol with specific gene deletion mutations led to greater growth impairment than would result from either insult alone. A similar approach is often undertaken with drug characterization in mammalian cells, whereby a second 61

drug is used to examine whether co-treatment leads to synergistic effects on activity. The yeast data identified mitochondrial activities, and the ETC in particular, as processes affected by elesclomol. To determine whether this requirement for mitochondrial function and an intact ETC is conserved from yeast to human, we tested elesclomol-Cu in combination with the known ETC inhibitors antimycin A (a complex III inhibitor) or rotenone (a complex I inhibitor) in human melanoma cells (Hs294T). As single agents, both ETC inhibitors were cytotoxic, although their dose response curves were broad in comparison to the steep curve exhibited by elesclomol-Cu

(Figure 3-4A). Concurrent administration of IC20 doses of elesclomol-Cu with antimycin A or rotenone resulted in substantial increases in cell death for both combination pairs and combinatorial benefit was also seen at IC50 doses (Figure 3-4B, 3-4C). Although a precise quantitation of synergy is confounded by the differences in the degrees of dose response (as reflected in the shapes of the dose response curves), these results show that direct modulation of the ETC and mitochondrial respiration in human cells enhances the cytotoxic activity of elesclomol.

3.2.6 Human cells lacking mitochondrial DNA are sensitive to elesclomol

Further evidence of a conserved MoA is provided by experiments using cells either containing their mitochondrial DNA (parent cells) or devoid of it (ρ0). Without the mitochondrial DNA, ρ0 cells cannot perform electron transport or oxidative phosphorylation, although other mitochondrial functions are maintained. I prepared dose-response curves by treating human osteosarcoma 143B and 143-ρ0 cells with elesclomol-Cu and elesclomol plus 5 μM CuCl2 for a period of 48 hours. The cells lacking mitochondrial DNA were more sensitive to elesclomol than the normal cells (Figure 3-5).

3.3 Discussion

In a previous study on elesclomol‟s mechanism of action185, the drug was shown to accomplish its cancer killing activity via the induction of untenable levels of intracellular ROS followed by apoptosis. The cellular mechanism and target by which this occurred, however, remained unknown. In those studies, mitochondrial involvement was specifically ruled out on the basis of using isolated mitochondria. However these experiments were performed in the absence of

62

Figure 3-4. Combinations of elesclomol-Cu and ETC complex inhibitors in melanoma cells (A) Single agent viability assays in Hs249T human melanoma cells using graded concentrations of elesclomol-Cu, antimycin A or rotenone for 72 h. The IC50 values obtained were 11 nM, 240 nM and 77 nM, respectively. Combination treatment of elesclomol-Cu with antimycin A (B) or rotenone (C) at IC20 or IC50 doses resulted in significantly enhanced cytotoxicity, showing that direct modulation of the ETC and mitochondrial respiration in mammalian cells enhances the cellular activity of elesclomol.

63

Figure 3-5. Mitochondrial DNA depleted (ρ0) cells are hypersensitive to elesclomol.143B and 143B-ρ0 cells were treated with elesclomol-Cu and elesclomol + 5 μM CuCl2 for 48h. Cell viability was determined using the SRB assay.

64

copper rendering them uninformative. In this study, with the importance of copper newly realized191, we have used an in vivo yeast system and additional in vitro human cell studies to identify the mitochondrion as the source of elesclomol-induced ROS and strongly implicate the process of the electron transport as the “target” of the drug.

As we have shown for other cancer drugs93,96,101,193, the approach of using the yeast deletion collection yielded an accurate indication of elesclomol‟s mechanism of action in mammalian cancer cells. This is predicated on the similarities of the cellular responses in these different eukaryotic cells, which we found to be the case. Our results showed that both yeast and human cells require copper for elesclomol activity, induce ROS to high levels when sufficient drug is present, and succumb to cell death upon relatively short elesclomol treatment. We also reveal the importance of an active ETC in both systems.

Our data indicate the lack of a unique protein “target” of elesclomol. While the analysis of the heterozygous deletions identified a single sensitive strain, nus1, its sensitivity was modest compared to other drug-target strain combinations we have analyzed in the past96,196, and it is therefore unlikely that the interaction of elesclomol and NUS1p, if any, is responsible for the primary cytotoxic activity of elesclomol in the cell.

In contrast, the analysis of the homozygous deletion set identified a robust and biologically coherent set of activities associated with mitochondrial activities. Both manual and computational (GSEA) analyses identified overlapping classes of genes involved in various elements of electron transport, mitochondrial translation (including mitochondrial ribosome subunits, translation factors, tRNAs, and mRNA splicing enzymes), mitochondrial copper availability and homeostasis, and genes involved in stress responses, particularly oxidative stress. Importantly, equally sensitizing mutations were found distributed throughout the ETC or its associated processes. Sensitive strains containing mutations affecting individual subunits of the various ETC complexes were identified, as were components required for the modification or assembly of the complexes. Given that each of the complexes contain numerous subunits, it is likely that most of these mutations would not completely abrogate ETC function, but more likely, only partially interfere with its activity. A similar argument can be made for the sensitizing mutations affecting mitochondrial translation, which produces a minority of the

65

proteins required for ETC function. Individually, these mutations would likely only have a small effect on the overall translation capability in the organelle, again perturbing but not eliminating ETC activity. Thus, it appears that modulating the electron flow at any of numerous points along the ETC can lead to enhanced elesclomol impact on the cell and therefore, we conclude that it is the disruption of the process of electron flow down the ETC, rather than disruption of a particular protein or activity, that is of relevance.

This effect on cell viability via ETC disruption appears to operate similarly in human cells. In the data presented in Figure 3-4, we used a second drug in combination with elesclomol rather than a potentially sensitizing gene disruption and obtained analogous results. Co-treatment of melanoma cells with elesclomol-Cu and either of two ETC inhibitors enhanced cytotoxicity. Again, the modulation of the process, rather than the inhibition of a particular protein, seems paramount for increased elesclomol activity.

The primary cytotoxic effect of elesclomol appears to be confined to the mitochondria and not to involve a cytoplasmic component. For example, none of the genes normally involved in the response to cytoplasmic oxidative stress were identified in the screen. This includes the YAP1 gene, which encodes the transcription factor that is the primary responder to oxidative stress in the cytoplasm and drives the up-regulation of a battery of stress response genes. Deletion of another prominent cytoplasmic stress response protein, superoxide dismutase 1 (SOD1p), actually provided slight resistance to the elesclomol treatment. Deletion of the copper chaperone protein CCS1p, required for SOD1p activity, also provided mild resistance, thereby confirming the result. In striking contrast, deletion of SOD2, the mitochondrial superoxide dismutase, was highly sensitizing.

The mitochondrion is the major site for ROS production in normal cells as well. Complexes I and III are prone to electron leakage, leading to the production of highly toxic superoxide or hydroxyl radicals in the vicinity of the ETC199. Under most conditions these free radicals are kept in check by the anti-oxidant systems in the organelle. However, this basal level of electron leakage can be amplified by inhibitors of electron chain complexes, such as rotenone (complex I), antimycin A (complex III) or cyanide (complex IV), leading to decreased viability200. Similarly, the impact of elesclomol-Cu appears to overwhelm the oxidative stress response

66

systems, allowing cytotoxic levels of ROS to accumulate. When the ETC inhibitors and elesclomol are used together during treatment, a degree of synergy is observed indicating that the combination of these drugs augments their effects in the cell as compared to monotherapy.

How might elesclomol-Cu instigate this lethal increase of ROS via its interaction with the ETC? A major clue comes from elesclomol‟s requirement for copper for its activity. Copper binds to elesclomol in the Cu(II) state. In the cell, elesclomol-Cu can undergo a redox reaction with copper being reduced to the Cu(I) state. By itself, this reaction could produce free radicals by a Fenton reaction. The redox potential for this reaction is −330 mV191 and this potential appears critical for elesclomol activity. Analysis of analogs with similar structures but with different potentials has shown that only those compounds with potentials similar to elesclomol-Cu are cytotoxic191. Very interestingly, this potential is well aligned with the potential drops along the ETC201.

Considering all of these features, there appears to be at least three major avenues by which elesclomol-Cu could lead to heightened levels of ROS. The drug could generate ROS on its own via its copper-based redox chemistry (perhaps using electrons or redox potential “stolen” from the ETC). Alternatively, the drug could interfere with the electron flow along the ETC, leading to elevated levels of electron leakage and free radical formation normally seen in cells, but here at levels that overwhelm the cell‟s defense systems. Finally, elesclomol-Cu could specifically interfere with copper-requiring events associated with ETC function. Some of the complexes are comprised of subunit proteins that require Cu for their activity and their assembly depends on specific copper chaperone proteins. Elesclomol could compete for or interfere with these processes, thereby impacting electron flow down the chain. These mechanisms are not mutually exclusive and, in fact, more than one may come into play sequentially: the initial impact of elesclomol-Cu could alter the subsequent dynamics of the ETC allowing additional mechanisms to take place that ultimately result in apoptosis. Whichever mechanism is used, we expect that the match of the redox potentials within the ETC to that of elesclomol-Cu is an important driving force.

The ability of elesclomol treatment to quickly lead to cell death, and not just cell arrest, is an important feature of the drug. Drugs that cause cidality are relatively uncommon in yeast, with

67

fewer than 10% of 10,000 drugs that inhibit growth inducing cidality197. Both yeast and human cells exposed to elesclomol-Cu for a few hours or less (Figures 3-1 and 3-5) are destined to die. The ability to kill a cell exposed briefly to the drug is a valuable property for an anticancer agent.

Finally, the improved mechanistic understanding of elesclomol‟s activity provided by this report has important implications for its therapeutic application in oncology. Specifically, lactate dehydrogenase (LDH) has been identified as a potential biomarker predictive of response in the clinical evaluation of elesclomol. In a Phase 3 trial of elesclomol in combination with , the primary endpoint of progression free survival was achieved in metastatic melanoma patients exhibiting low and normal LDH levels in their bloodstream, with a significant improvement in median progression free survival time. Conversely there was no benefit in the elevated LDH population190. High serum levels of LDH are thought to reflect a tumor burden with increased reliance on glycolysis for its metabolic needs202,203. Conversely, patients with lower LDH levels should have tumor burdens that are more reliant on oxidative phosphorylation, a situation we have shown here to be more sensitive to elesclomol treatment. Thus, the insights established here by our studies on yeast and human cells provide critical understanding into the clinical activity of the drug. It also offers a compelling rationale for a biomarker-based prioritization of patients likely to respond to elesclomol treatment.

3.4 Materials and Methods

3.4.1 Reagents

Elesclomol and elesclomol-copper complexes were synthesized at Synta Pharmaceuticals Corporation. Copper chloride, antimycin A and rotenone were all purchased from Sigma-Aldrich (St. Louis, MO).

3.4.2 Yeast Strains

All strains used in this study are diploid and congenic with the reference strain BY4743 (MATa/α his3Δ1/his3Δ1 leu2Δ0/leu2Δ0 lys2Δ0/LYS2 MET15/met15Δ0 ura3Δ0/ura3Δ0).

68

3.4.3 Minimum Inhibitory Concentration (MIC) and Cidality Determination

MIC determination was performed as previously described193. To test for cidality, wild type yeast (BY4743) were inoculated into YPD at an OD600 of 0.5 such that they were in the mid-log phase of growth. 100 µl of cells in media were aliquoted into wells of a 96 well plate, and a titration of elesclomol-Cu was added at a final concentration ranging from 5 nM to 5 µM in DMSO or with 2% DMSO to serve as a vehicle control. Cells were removed at hourly intervals using a pin tool to transfer 5 µl of cells in media onto agar dishes without drug. This transfer effectively dilutes the drug below inhibitory concentrations. Plated cells were incubated at 30°C for 48 h and photographed. In this assay each single viable cell transferred is able to form a visible colony. Concentrations and doses that produced no viable colonies after 48 h were scored as cidal.

3.4.4 Deletion Pool Growth and Chip Experiments

Screens were performed essentially as described by Ericson et al.204. The BY4743 strain was used to determine the dose of compound that resulted in 15% growth inhibition. Cells were inoculated at an OD600 of 0.0625 in serial dilutions of drug and grown in a Tecan GENios microplate reader (Tecan Systems Inc., San Jose, CA) at 30°C with orbital shaking. Optical density measurements (OD600) were taken every 15 minutes until the cultures were saturated, and the doubling time (D) was calculated as described205,206.

For genome-wide fitness profiles, ~4800 homozygous deletion strains and ~1200 essential heterozygous deletion strains were assayed as previously reported204 prior to genomic DNA extraction. 200 ng of genomic DNA were added to 2 separate PCR reactions, one each with primers designed to amplify all UPTAGs or all DOWNTAGS. One primer in each reaction was biotinylated such that it could be detected following hybridization to the chips using streptavidin- phycoerythrin. Intensity values for the probes on the chip were extracted using the GeneChip Operating Software (Affymetrix). Quantile normalization, outlier omission, and fitness defect ratio calculations were performed as previously described205. The larger the ratio, the more depleted (sensitive) is the strain as compared to control condition without the drug.

69

3.4.5 Analysis of Elesclomol Sensitivity in Yeast Deletion Mutants

The homozygous and heterozygous deletion strains were treated en masse with 300 or 400 nM elesclomol-Cu, producing ~15% growth inhibition. Each screen was performed in triplicate at both drug concentrations. Because all six sets of data yielded similar results, the log2 ratios from all of them were averaged. This averaged data were used for subsequent analysis.

3.4.6 Gene Set Enrichment Analysis (GSEA)

Strains in the chemogenomic profile of elesclomol were mapped to genes using chromosomal feature data downloaded from the Saccharomyces Genome Database (http://www.yeastgenome.org) and the resulting profile was analyzed by GSEA v2.07 in pre-rank mode (Java implementation). All default parameters were used except that the minimum and maximum gene set sizes were restricted to 5 and 300, respectively. Biological process and protein complex gene annotations were obtained from Gene Ontology (http://berkeleybop.org/goose). Additional protein complex annotations based on consensus across different studies were obtained from Benschop et al.207. The enrichment map was generated with the Enrichment Map Plugin v1.1181 developed for Cytoscape182. All default parameters were used. The nodes in the map were clustered with the Markov clustering algorithm208, using the overlap coefficient computed by the plugin as the similarity metric (coefficients less than 0.5 were set to zero) and an inflation of 2. For each cluster, the leading edge genes were computed as in Subramanian et al.180 for each member node.

3.4.7 Multiple Drug Effect Analysis

Hs294T melanoma cells were purchased from American Type Culture Collection (Manassas,

VA) and cultured according to standard techniques at 37°C in 5% (v/v) CO2 in DMEM plus 10%

FBS. The half maximal inhibitory concentration (IC50) values for elesclomol, rotenone and antimycin A in Hs294T cells were determined using a 1.5-fold serial dilution series of each compound. For combinatorial analysis, cells were plated in triplicate for 24 h prior to the addition of drug or vehicle (0.3% DMSO) to the culture medium. Combinations of elesclomol with rotenone or antimycin A were performed concurrently based on the IC50 and IC20 values for each agent. At 72 h, metabolic activity was monitored by alamarBlue (Invitrogen, Carlsbad, CA)

70

fluorescence (560EX/590EM nM) with a SpectraMax microplate reader (Molecular Devices). The resulting data were used to calculate cell viability, normalized to vehicle control.

3.4.8 Analysis of ρ0 cell lines

143B and 143B-ρ0 cells (a gift from Eric Shoubridge) were maintained in DMEM medium supplemented with 10% FBS, 100 μg/mL uridine and 1 mM pyruvate. Cells were seeded in 96- well tissue culture plates and treated with a 2-fold dilution series of elesclomol-Cu or elesclomol plus 5 μM CuCl2 48 h. Cell viability was determined using the sulforhodamine B (SRB) colorimetric assay for cytotoxicity screening as described209.

71

Chapter 4 A high-throughput chemogenomic loss-of-function screen to examine doxorubicin mode-of-action in human cells

Contributions: I performed all experiments presented in this chapter. The assay development work was shared between myself and Elke Ericson. The shRNA pools were provided by Kim Blakely and Jason Moffat. Larry Heisler provided bioinformatics assistance for the microarray data extraction. Kevin Brown generated the GARP scores. Corey Nislow, Guri Giaever and Jason Moffat supervised the project.

4 4.1 Introduction

Chemical genomic screening has proven to be an effective method for identifying drug targets and characterizing mechanisms of action of compounds with previously unknown functions. In these screens, each gene in the cell is evaluated simultaneously, thereby providing an unbiased method for identifying potential drug targets. To date, diverse chemical genomic screens have been developed and applied successfully in the yeast Saccharomyces cerevisiae due to its well- characterized genome and ease of genetic manipulation91,210,211. An excellent example of a loss- of-function chemical genomic screening tool in yeast is the haploinsufficiency profiling assay (HIP), which takes advantage of the observation that a lowered dose of a drug target is hypersensitive to the drug93. This assay has successfully identified and confirmed targets of known drugs, such as Hmg1 as the target of different statins and Dfr1 as the target of methotrexate89,90,93. However, there are limitations to performing compound screens in yeast: first, the drug concentration required for effect is often higher than human physiological levels due to the yeast cell wall and efflux mechanisms. More importantly, there are many genes and processes involved in mammalian cell function that are not conserved in yeast and although core processes can be studied, key targets and pathways are unique to human cells. It would therefore be useful to perform such screens to identify human gene targets in a mammalian system.

72

The application of RNA interference (RNAi) to large-scale screens is a powerful tool for performing loss-of-function studies in mammalian systems212,213. Several strategies have been developed to use genome-scale RNAi libraries to study human genes in human cell culture and to identify requirements of cancer phenotypes115,212,213. We have, in collaboration with the Moffat lab, developed, by analogy to yeast HIP, a high-throughput platform for RNAi-mediated chemical genomic screening in cultured human cells (Figure 4-1) using the RNAi Consortium shRNA library115. Several groups have already demonstrated the ability to identify targets using smaller pool formats121,212-214, however we are using a genome-scale pool for our screens. In this chapter, I will discuss the development and application of these screens to study novel targets and/or mechanisms of the anticancer agent doxorubicin.

4.1.1 Development of a large-scale shRNA-based chemogenomic assay

In order to adapt the idea of haploinsufficiency profiling to mammalian cell culture we employ RNAi to produce the lowered gene dosage. For an unbiased view of drug action in the cell, we use a genome-scale library of shRNAs. In addition, for this to be a high-throughput approach to studying potential drug targets, this assay will be pool-based, similar to yeast HIP. The deconvolution of pool results will be done using Affymetrix microarrays. Therefore, to perform this assay, a population of cells infected with a TRC sub-library containing 54,000 distinct hairpins, in which each cell contains a single hairpin, is grown in the presence or absence of compound over a period of two weeks. In the presence of compound, those knockdowns which cause sensitivity will drop out of the population. Cells will be harvested at different timepoints, genomic DNA extracted, and half-hairpin barcodes are PCR-amplified. The barcodes are then hybridized to an Affymetrix UT-GMAP microarray of our design and signal intensity of the barcodes over time enables the identification of hairpins that lead to loss of cell viability.

73

Figure 4-1. The RNAi-mediated chemogenomic assay. The pooled TRC lentiviral shRNA library is propagated in E. coli and used to produce lentivirus. The lentivirus pool is used to infect human cells (MOI=0.3) which are grown in the presence of drug. Genomic DNA is then extracted from the cells and hairpin sequences are amplified by PCR. Digestion of the PCR products by restriction enzyme produces the half-hairpin barcodes for microarray detection. The barcodes are hybridized to a UT-GMAP Affymetrix microarray and the signal intensity obtained on each probe is analyzed to find the relative abundance (compared to a no-drug control) of each shRNA in the population.

74

4.1.2 Doxorubicin

Doxorubicin is a member of the anthracycline class of widely used anticancer compounds that are effective against a wide range of tumours. In clinical use since the 1970s, doxorubicin is used in the treatment of breast cancer, esophageal cancer, osteosarcoma, Kaposi‟s sarcoma, Hodgkin‟s and non-Hodgkin‟s lymphoma215. Despite its long-term use as a chemotherapeutic, the precise mechanism through which doxorubicin acts in cells remains controversial and several mechanisms are attributed to its mode(s) of action. The primary cause of cytotoxicity by doxorubicin is considered to be through topoisomerase II poisoning; doxorubicin stabilizes the complex of topoisomerase II and DNA prior to strand re-ligation, leading to DNA double-strand breaks which inhibit DNA replication216. However, gene expression analysis of anthracyclines using the NCI panel of 60 cancer cell lines indicated that topoisomerase II may not be the only target217. A second major mechanism attributed to doxorubicin is the ability to enter the nucleus and intercalate into DNA42. Other modes of action include the ability to bind to DNA and form adducts, interference with DNA helicase activity, lipid peroxidation, and generation of free radicals, all of which lead to DNA damage218.

The use of doxorubicin and other anthracyclines in the clinic is limited by accumulated toxicity to non-targeted tissues. A major side effect with chronic administration of anthracyclines is their association with dose-dependent cardiotoxicity215. Prolonged treatment with doxorubicin can lead to cardiomyopathy and congestive heart failure. While the causal mechanism of doxorubicin-induced cardiomyopathy remains unclear, some evidence indicates that free radicals produced by anthracyclines damage cardiac myocytes43. Other hypothesized mechanisms for doxorubicin-induced cardiotoxicity have been described in the introduction of this thesis. The cardiotoxic side effects coupled with the multiple mechanisms ascribed to doxorubicin indicate that further study into the cellular mode of action of doxorubicin is required.

75

4.2 Results

4.2.1 Optimization of large-scale shRNA-based chemogenomic screens

At the start of this project, in early 2008, we obtained a pooled pLKO.1-plasmid115 sub-library of the TRC collection containing ~45,000 hairpins (45K pool) targeting ~9,000 human genes from the Broad Institute. Kim Blakely used this pool to generate a lentivirus pool using a packaging cell line and the lentivirus pool was used to infect Jurkat (human T lymphocyte) and BT474 (human breast carcinoma) cell lines. These drop-out screens require a high degree of sensitivity to detect the loss of hairpins in a pool where most hairpins will remain unaffected; therefore, the first goal was to obtain a high signal intensity and low background signal on the microarray after hybridization of barcodes corresponding to the initial pool. Using plasmid DNA from the sub- library and genomic DNA from Jurkat and BT474 cells infected with the 45K shRNA pool, we evaluated different conditions for PCR amplification, amplicon purification, digestion and hybridization.

In this section, I will describe some results from our optimization of the assay in detail. The basic protocol for preparing half-hairpin barcodes from harvested cells is to extract genomic DNA, PCR amplify the hairpins, digest the hairpins using the XhoI restriction endonuclease, purify the half-hairpin barcodes by gel-extraction, and hybridize the barcodes to a microarray. Here, “signal” refers to the median signal intensity from probes with complementary barcodes in the 45K pool and “background” refers to median background signal intensity. First, we found that using a Qiagen kit for genomic DNA extraction instead of phenol/chloroform-based DNA extraction resulted in an increased DNA yield. Next, we tested conditions for the PCR reaction. We tested different polymerases: TaKaRa Taq (Millipore), Phusion (Finnzymes) and Platinum Pfx (Invitrogen), and found that with Platinum Pfx we obtained the highest signal with lowest background. We modified the PCR reaction by adding 1% DMSO which doubled the signal. The original forward primer used in the PCR reaction was designed 51 bp away from the hairpin structures leaving a long product that is susceptible to forming secondary structures. Therefore we designed 5 new forward primers closer to the hairpin sequence. Counter to our expectations, we found that PCR amplification with these primers resulted in 3-fold lower signal intensity. Similar to our yeast TAG4 barcode microarray protocol, we added blocking primers (see Methods) that bind to regions of the barcode PCR product that do not hybridize to the chip (and 76

could potentially hybridize to other hairpins), increasing the signal. By serendipity, we found that pre-conditioning the microarrays in 10 mM NaOH for 10 min prior to hybridizing dramatically increased the signal intensity on the chips.

During the course of our assay development, we 1) increased the signal intensity more than 30- fold, 2) obtained a high signal-to-noise ratio comparable to our yeast TAG4 arrays88 (ratio ≈ 30) and 3) verified the reproducibility of the results (r>0.99 for biological replicates). One critical optimization step in developing a robust protocol was to standardize the screens by starting each experiment with a frozen aliquot of the same pool of infected cells. We found no significant decrease in the representation of the hairpins after freezing and thawing the cells (Figure 4-2).

To test whether the assay was sensitive enough to identify loss of viability in the large-scale assay, I first generated individual knockdowns of genes known to be involved in response to specific compounds to examine viability changes in response to drug. To this end, I generated lentivirus containing shRNAs against the DNA-repair gene RAD50. I then infected A549 human non-small cell lung carcinoma cells with these individual hairpins and the infected cells were grown in the presence of doxorubicin at an IC25 or IC50 for 48 and 96 hours which represent 2 and 4 generations of cell growth. Viability of the cells after compound treatment was assessed using the SRB assay209 (Figure 4-3). There was a greater decrease in the viability of the infected cells than the control (uninfected cells), indicating that the gene knockdown successfully sensitized the cell to the drug. By testing different drug concentrations and treatment lengths, we determined that an IC25 at 48 h would give the best ratio between the drug and no-drug treatments in our large-scale assays. It is worth noting that we subsequently found, based on having accumulated additional empirical data, that longer drug treatment improves the dynamic range of the assay.

Elke Ericson performed proof-of-principle studies for the large-scale screens using compounds with known specific drug targets. One example of these compounds is sodium fluoride (NaF) which is known to target pyrophosphatase 1 (inorganic; PPA1)219. Initially, individual knockdowns of PPA1 were tested and confirmed to exhibit hypersensitivity to NaF treatment before scaling-up to a pool containing 45,000 hairpins. A549 cells were infected with the 45K lentivirus pool

77

100 94.7 94.2 91.9 93.6 90 80 70 60 50 40 30 20 10

% of hairpins chip on detected 0 plasmid DNA gDNA from gDNA 2 days gDNA 7 days cultured cells post-thaw post-thaw

Figure 4-2. The representation of hairpins in the pool does not decrease after freezing and thawing cells. Barcodes isolated from plasmid DNA, gDNA from infected cells before freezing, and gDNA from cells 2 and 7 days post-thawing were hybridized to TRCBCx520397 microarrays. Detection of the hairpins was based on having a signal intensity two times greater than the background. The percentage of hairpins detected compared to the total population of 45,000 is indicated above each bar.

100 90 80 70 * * 60 48h drug treatment 50 96h drug treatment 40

30 no-drug control (%) no-drug Viability compared to to Viability compared 20 10 0 no shRNA control shRAD50_832 shRAD50_2674

Figure 4-3. A549 cells containing shRNAs against RAD50 exhibit increased sensitivity to doxorubicin. Two shRNAs against RAD50 were individually packaged into lentivirus and infected into different plates of A549 cells. Cells were treated with doxorubicin for 48 h or 96 h. Viability was measured using the SRB assay and compared to untreated A549 cells. (Error bars indicate standard deviation; * indicates a statistically significant difference from the no shRNA control, p<0.05, n=3)

78

and amplified. Next, NaF was added to the cells at an IC25 for 48 hours. Half-hairpin barcodes were prepared and hybridized to the TRCBCx520397 microarray121 and signal intensities between treated and non-treated samples were compared. The observation that 2 PPA1 hairpins were in the top 3 sensitive hits (out of 45,000 hairpins) gave us confidence that these screens have the potential to identify known drug targets (Figure 4-4).

4.2.2 Screen for potential doxorubicin targets and analysis of results

For the large-scale shRNA screens, we chose the A549 human non-small cell lung carcinoma cell line because it is a well-characterized standard among lung cancer cell lines, it is easy to maintain in cell culture, adheres well to plastic, has a relatively short doubling time (~22h), can be used for puromycin selection, and, importantly, A549 cells are readily infected with lentivirus. At this stage in the project (late 2008), the Broad Institute had generated a larger sub- library pool of 54,020 shRNA plasmids (54K pool) that targets ~11,000 human genes. Pooled virus produced from the sub-library was used to infect A549 cells such that each cell contains a single hairpin. Our lab also developed a new Affymetrix microarray, the Gene Modulation Array Platform (GMAP), that can be applied to diverse gene modulation experiments, including the deconvolution of results from complete genome-scale shRNA pools118. The GMAP platform was able to identify hairpins that had decreased levels in a simulated population giving us confidence that it will be able to successfully interpret the results of this screen118.

I performed large-scale shRNA-based chemical genomic screens on doxorubicin, a compound that acts primarily by targeting topoisomerase II to cause DNA double-strand breaks (DSBs). We chose to increase the length of treatment time from 48 h to 2 weeks in attempt to enrich for hairpins against genes required for resistance to the compounds. In the original protocol, cells were grown for 2 doublings in the presence of drug. We hypothesized that by increasing the length of compound treatment to 15 doublings in this competition assay, the difference in the number of cells containing hypersensitive hairpins compared to unaffected hairpins would be more pronounced. Therefore, cells were grown in the presence of compound at an IC25 or in the presence of vehicle (DMSO) for 2 weeks. Cells were passaged and/or harvested every 72 h. Half-hairpin barcodes were prepared from the t=0d, t=6d and t=15d samples and hybridized onto our GMAP arrays. Each set of screens (control and drug) was performed in triplicate.

79

Figure 4-4. Chemical genomic profile of sodium fluoride (NaF). A549 cells infected with the 45k TRC RNAi library were grown in the presence of NaF at IC25 for 48 hours. Cells were harvested, genomic DNA extracted, and half-hairpin barcodes prepared. The barcodes were hybridized to the TRCBCx520397 Affymetrix microarray and the log2 ratio of signal intensity between a control (no drug) and the experiment was compared. The x-axis represents gene names in alphabetical order and the y-axis is the log2 ratio of signal intensity for drug vs control. Signal intensities representing hairpins against PPA1 are highlighted in red.

80

To identify which gene knockdowns led to drug hypersensitivity, I used a scoring system developed in the Moffat lab called shARP (shRNA Activity Ranking Profile)214. Briefly, in this analytical platform, experimental conditions are normalized based on the growth rate of the cells and each hairpin is assigned a shARP score based on its change in intensity across multiple timepoints, where a low shARP score indicates a severe dropout. Because there are approximately 5 hairpins representing each gene, the shARP scores are then converted into GARP (Gene Activity Ranking Profile) scores by taking the average of the two lowest shARP scores. The GARP scores were used to generate a ranked list of genes that are sensitive to doxorubicin.

Functional enrichment of biological pathways of the top 2% of sensitive hairpins (those with low GARP scores) revealed that the targeted genes have roles in cellular growth and proliferation, gene expression, cancer and cell cycle processes (p<0.05; Figure 4-5, Table 4-1). Next, I examined the top hits on a gene-by-gene basis and found that there were few key DDR genes in the top 500 hits. Some insight into this result can be gleaned from analogous experiments in yeast where heterozygote deletants do not show obvious enrichment for DDR genes93,98 after treatment with DNA-damaging agents, whereas homozygous deletants do101. In yeast, these observations are attributed to the fact that when DNA repair as a process is targeted, few essential heterozygote strains should manifest sensitivity whereas those genes that are involved in the pathway will show sensitivity when completely abolished because the required redundancy in the pathway is disturbed. This observation suggests that a gene knockdown in mammalian cells (similar to a heterozygote because A549 cells are roughly diploid and because few hairpins completely deplete mRNA levels) may not be sufficient to sensitize cells to these drugs. However, in our top hits we did identify several interesting candidate genes involved in cell cycle arrest and DNA replication that may be involved in responding to DNA damage in a chemically synthetic manner where the combination of the compound and mutation leads to decreased fitness relative to the other cells in the pool.

4.2.3 Validation of candidates

To validate and follow-up on interesting hits that came out of the doxorubicin screen (defined using a combination of intuition and literature observations), I selected 34 genes from the top

81

1.52E-03 Cellular Growth and Proliferation

Gene Expression 2.02E-03

Cell Signaling 2.79E-03

Cancer 7.52E-03

Cell Cycle 1.01E-02 Top 2% gene hits All genes in 54k pool Protein Degradation 1.08E-02

Cell Death 1.20E-02

Cellular Function and Maintenance 2.65E-02

0 5 10 15 20 25 30 % of total genes in functional category

Figure 4-5. Functional enrichment of the top 2% hits from doxorubicin screen. Functional enrichment of the top 2% of significant hits was performed using of IPA (Ingenuity Systems, www.ingenuity.com).The percent of genes in each category sensitive to doxorubicin is plotted on the y-axis in blue. The percent of genes in the 54k TRC pool in each category is in red. P-values are listed to the right of each category.

82

Table 4-1. Functional enrichment of genes in the top 2% of significantly sensitive strains from the doxorubicin shRNA screen.

Function Annotation p-value Genes

ADAM15, B4GALT2, CCNG1, CD320, CDK11B, CDKN2B, GPX4, HGFAC, IKBKG, IL23A, IL2RB, IP6K2, LEF1, LEP, MBD3 (includes EG:53615), MMP2, NOTCH1, NRG1, PRMT1, RELA, Cellular Growth and Proliferation 1.52E-03 RHOB, SMAD4, TENC1, UBE2L6, USP3, VDAC1, ZBTB17

CDK11B, EHF, LEF1, LEP, MAD2L2, NOTCH1, PRDX1, RELA, Gene Expression 2.02E-03 SMAD4, STUB1, TAF4B, ZBTB17

Cell Signaling 2.79E-03 DUOX1, IL17C, IL2RB, MKNK2, NTRK3, PRMT1, RELA

Cancer 7.52E-03 GATA5, PGC, RELA, RPS19, SMAD4, TLK1

AKAP9, CCNG1, CD320, CDC20, CDK11B, CDKN2B, E4F1, LEP, MAD2L2, MLH1, NOTCH1, NRG1, PHB (includes EG:5245), Cell cycle 1.01E-02 POLH, RASSF2, SMAD4, STRN3, ZBTB17

Protein Degradation 1.08E-02 AURKAIP1, CDC20, HGFAC, MMP2, PGC, PMPCA, USP3

CASR, CCNG1, CDC20, CDKN2B, HOXA1, HSF1, IKBKG, IL28A, IP6K2, KSR2, LEP, LETM1, MAOA, MAPKAP1, MC1R, MLH1, NOTCH1, NRG1, OAS1, PHB (includes EG:5245), POLH, PRDX1, Cell Death 1.20E-02 RASSF2, RELA, RHOB, RPS19, STUB1, TENC1, VDAC1

Cellular Function and Maintenance 2.65E-02 E4F1, NOTCH1, PRDX1, PRMT1

83

Table 4-2. Significantly sensitive hairpins from doxorubicin shRNA screen chosen for validation.

Rank Gene Description GARP score ratio Symbol (ctrl/drug) 3 GRM3 glutamate receptor, metabotropic 3 -2.15523 9 CCNG1 cyclin G1 -1.80645 10 CASR calcium-sensing receptor -1.76194 12 TLK1 tousled-like kinase 1 -1.7059 16 GRM2 glutamate receptor, metabotropic 2 -1.69138 17 STRN3 striatin, calmodulin binding protein 3 -1.6662 19 GMIP GEM interacting protein -1.65726 20 MAPKAP1 mitogen-activated protein kinase associated protein 1 -1.64208 21 GUCY1B2 guanylate cyclase 1, soluble, beta 2 -1.63474 22 GATA5 GATA binding protein 5 -1.58454 30 POLG polymerase (DNA directed), gamma -1.49398 33 MKNK2 MAP kinase interacting serine/threonine kinase 2 -1.47356 39 CKM creatine kinase, muscle -1.42774 43 ARHGDIG Rho GDP dissociation inhibitor (GDI) gamma -1.39955 46 ZNF623 zinc finger protein 623 -1.36178 48 RASSF2 Ras association (RalGDS/AF-6) domain family member 2 -1.35349 54 RFC4 replication factor C (activator 1) 4, 37kDa -1.32203 58 MAD2L2 MAD2 mitotic arrest deficient-like 2 (yeast) -1.30707 69 DSTYK dual serine/threonine and tyrosine protein kinase -1.27662 79 RAB5A RAB5A, member RAS oncogene family -1.25438 100 RHOB ras homolog gene family, member B -1.18059 110 CDKN2B cyclin-dependent kinase inhibitor 2B (p15, inhibits CDK4) -1.16905 115 POLH polymerase (DNA directed), eta -1.15725 120 H2AFZ H2A histone family, member Z -1.15109 123 CDC20 cell division cycle 20 homolog (S. cerevisiae) -1.14953 145 POLS polymerase (DNA directed) sigma -1.09038 152 CLK4 CDC-like kinase 4 -1.08232 155 MLH1 mutL homolog 1, colon cancer, nonpolyposis type 2 (E. coli) -1.08017 164 CDC2L1 cell division cycle 2-like 1 (PITSLRE proteins) -1.05898 197 CASP8 caspase 8, apoptosis-related cysteine peptidase -0.99388 224 REM2 RAS (RAD and GEM)-like GTP binding 2 -0.95657 236 BBC3 BCL2 binding component 3 -0.94298 240 MTUS1 mitochondrial tumor suppressor 1 -0.93775 249 MYST4 MYST histone acetyltransferase (monocytic leukemia) 4 -0.92948

84

250 sensitive knockdowns (<2% of the genome) for validation (Table 4-2). These genes were chosen due to their involvement in processes that are often compromised in cancer, such as cell cycle, DNA replication, DNA repair, the RAS/MAPK pathway or if the genes represent potential contributors to off-target effects of doxorubicin, particularly genes involved in cardiac muscle function43,220,221. To confirm the results, I examined individual gene knockdowns for sensitivity to doxorubicin. In order to have an independent validation method, I used endoribonuclease- prepared siRNAs (esiRNAs; courtesy of Laurence Pelletier) to produce the gene knockdowns instead of shRNAs. esiRNAs are produced by cleaving long double-stranded RNA molecules with an endoribonuclease, here RNase III, to produce diverse pools of siRNA-like oligonucleotides to minimize off-target effects. Wildtype A549 cells were transfected with esiRNAs against the target gene and validation was performed by assessing cell viability after growing the cells in a concentration of drug that inhibits wild-type growth by 25%. The viability of the knockdowns in drug compared to their respective no-drug controls are shown in Figure 4- 6. Each knockdown was evaluated in triplicate and 6/34 of the genes selected were confirmed to lead to hypersensitivity to doxorubicin when knocked down (p<0.1, Student‟s t-test).

Several of the validated hits are potentially interesting for follow-up studies. For example, knockdown of CCNG1 in hepatocarcinoma cells by the liver-specific microRNA MiR-122 has been shown to lead to increased sensitivity to doxorubicin222. In addition, CCNG1 is a target of the tumour suppressor p53 and plays a role in cell cycle arrest in response to DNA damage223. Another hit, GRM3, is interesting because metabotropic glutamate receptors have been implicated in the development and proliferation of melanoma, glioma, medulloblastoma, and colorectal carcinoma224,225. Also, expression of the metabotropic glutamate receptor 4 leads to resistance to the antitumour compound 5-fluorouracil226. MYST4 was of particular interest since it is part of the MYST family of histone acetyltransferases which are associated with DNA replication, DDR, and transcriptional regulation and have been implicated in human cancer227.

85

140 120

100 * * *

80 * * 60 * 40 esiFLUC(%) 20

0

Viability Viability drug in compared to

CKM

FMIP

TLK1

CLK4

BBC3

RFC4

POLS

POLH

REM2

POLG

GRM2 GRM3

H2AFZ

RAB5A CDC20 CASP8

GATA5 STRN3

MYST4 DSTYK

MTUS1

MKNK2

CCNG1

ZNF623

CDC2L2

RASSF2 MAD2L2

CDKN2B

GUCY1B2

ARHGDIG MAPKAP1 Gene targeted by esiRNA

Figure 4-6. Viability of individual knockdown cells after treatment with doxorubicin. A549 cells were transiently transfected with esiRNAs targeting potential hits from the genome-wide screen. The transfected cells were grown the presence of 300 nM doxorubicin for 48 h. Each bar represents the viability after target gene-knockdown compared to the esiFLUC-transfected control in the presence of doxorubicin. * indicates a significant decrease in viability compared to the ctrl esiRNA (p<0.1, Student‟s t-test).

86

4.2.4 Doxorubicin modulates histone H3 acetylation via MYST4

I followed-up on MYST4 as a potential target of doxorubicin. Previous studies revealed that doxorubicin treatment specifically alters levels of acetylation of histone H3228 but not histone H4229. Since MYST4 specifically acetylates histone H3 and not histone H4230, I hypothesized that this specific effect on acetylation is due to doxorubicin acting on MYST4. To examine whether doxorubicin affects histone H3 acetylation in A549 cells, I treated wildtype A549 cells with doxorubicin at varying concentrations and performed immunoblotting of cell lysates using an antibody against acetylated histone H3 at lysine 9, the MYST4 acetylation site. I found that increasing doxorubicin concentrations increases histone H3 acetylation to 1 μM, after which H3 acetylation is dramatically reduced (Figure 4-7). Next, I prepared individual knockdowns of MYST4 in A549 cells, treated these cells with doxorubicin and probed for changes in histone H3 acetylation. In the cells where MYST4 was knocked down, the acetylation of histone H3 in the cells is decreased, as previously shown230. When these knockdowns are treated with doxorubicin, histone acetylation decreases below no-drug levels (Figure 4-8). These results suggest that doxorubicin‟s effect on histone H3 acetylation is mediated through MYST4.

4.3 Discussion

The development of chemical genomic screens in mammalian cells promises a direct method to study drug action on the human genome. The pooled screening approach used here is advantageous because the screens can be done on a smaller scale, which reduces costs and is more practical for standard labs than arrayed screens which require a lot of infrastructure. However, there are liabilities to pooled screening methods. First, the deconvolution of results can be costly (in terms of analysis time and effort) and there is risk that the representation of the library may be skewed. The scale of the library also impacts the results obtained. Here we used a pool of 54,000 hairpins which target ~12,000 genes with each hairpin represented by ~300 cells in the pool. Newer pools of up to 90,000 hairpins will ensure better coverage of the genome, but make the use of microarrays to identify single hairpins that drop out of the population more difficult. One way to overcome these problems is to divide up the library and screen multiple, partially overlapping pools to cover the genome. Decreasing costs of next-generation sequencing

87

Figure 4-7. Doxorubicin treatment increases acetylation of histone H3 in A549 cells. A549 cells were treated with doxorubicin at 0.05, 0.1, 0.5, 1, 2.5, and 5 μM for 24 h and Western blot was used to examine the levels of histone H3 acetylation at lysine 9/14 (Ac-H3). Ac-H3 increases with increasing doxorubicin concentration up to 1 μM. High concentrations of doxorubicin reduce histone H3 acetylation.

Figure 4-8. MYST4 knockdown alters the effect of doxorubicin on histone H3 acetylation. A549 cells containing shRNAs against MYST4 were treated with 500 nM doxorubicin for 24 h. a) Western blot was performed using antibodies against acetylated histone H3 at lysine 9/14 (Ac- H3) and α-actin (loading control). b) Levels of histone H3 acetylation compared to the vehicle- treated shGFP control were quantified based on the Western blot results. c) Quantitative PCR confirmed MYST4 knockdown by shRNAs.

88

will allow the readout of the screens to be done by barcode sequencing, thereby reducing costs and providing greater levels of sensitivity and a greater dynamic range89.

It is important to note that while these mammalian screens are designed to be similar to yeast HIPHOP, a knockdown is fundamentally different from a knockout (either heterozygous or homozygous). With RNAi, the effect of gene knockdown is variable, therefore the remaining levels of protein cannot be predicted and must be individually evaluated on the cell types used in the screens. In addition, because the gene is not completely deleted, it is unlikely that these screens will reveal genes that have high compensatory transcriptional increases, i.e. those with important roles in normal cell function and survival. From our individual knockdown tests, we have observed that when well-known targets are knocked down, the cell viability in response to drug treatment is low but the cells are not dead, indicating the possibility that some targets may be masked when performing large-scale screens. Despite this, we can be confident that those genes that appear as significantly sensitive in our screens are examples of those that are very important for compound function.

Target validation is one of the biggest challenges of large-scale chemical genomic screens. There is currently no standard, high-throughput method to determine whether a compound interacts with and/or affects the function of each potential target; as a result, choosing specific genes to validate is ultimately the bottleneck to producing meaningful results from genome-wide screens. In most large-scale RNAi screens to date (including the one presented here) individual targets are selected for validation based on prior knowledge of the drug and/or gene function. RNAi reagents can be designed against any gene and it is often more attractive to follow-up on a gene with a well-characterized function so that future experiments will be easier to design. On the other hand, the validation of uncharacterized genes has the potential for interesting discoveries. Examination of the literature to identify relationships with the potential target to the drug or cell line is also used to determine which hits to validate. It would be ideal to eventually be able to validate all potential targets from a large-scale screen, however this is rarely practical. With the increasing number of large-scale genomic screens being performed, new efforts into developing technology to facilitate target validation will be necessary.

89

In this study, I identified MYST4 as a potential target of doxorubicin. Changes in histone acetylation are known to play important roles in chromatin structure and subsequent gene transcription231,232. Acetylation of histones neutralizes the positively charged lysine residues on histone tails, leading to weakened histone-DNA and histone-histone interactions; this results in chromatin decondensation which allows easier access of transcription factors to the DNA233. Histone acetylation is also important in gene-dependent genesis or suppression of cancer phenotypes234. The effect of doxorubicin on MYST4 acetyltransferase activity leading to increased histone acetylation may explain the efficacy of the compound against many different tumours. It has been suggested that the relaxed chromatin state caused by histone acetylation allows increased interaction between doxorubicin and DNA, potentiating its DNA-damaging effects235. In addition, MYST4 alters phosphorylation levels of genes in the MAPK signaling pathway which is responsible for growth control and implicated in uncontrolled tumour growth230. Interestingly, there are many MAPK pathway-related genes that appeared in the top 250 genes in our study, and this may be a result of direct MYST4 action. Further studies will be required to examine whether the compound binds directly to the target and whether the binding leads to direct inhibition of histone acetylation. First, binding of doxorubicin to MYST4 can be examined using a colocalization experiment. A fluorescent antibody can be used to tag MYST4 and fluorescence microscopy can be used to detect whether doxorubicin (which has intrinsic fluorescence at ~595nm when excited by 479nm light236) and MYST4 co-localize in the cell, indicating binding. To identify whether a doxorubicin-MYST4 interaction inhibits histone acetylation, purified MYST4 enzyme can be incubated with a histone substrate and the effect of doxorubicin addition on changes in histone acetylation can be examined. It will be interesting to examine whether other anthracycline antibiotics, such as daunorubicin, epirubicin and idarubicin, have the same effect on this acetyltransferase. Finally, we believe the regulation of histone acetylation by doxorubicin can lead to novel combination therapies for cancer.

In summary, we have described an RNAi-mediated chemogenomic assay for screening in mammalian cells that can be applied to identify novel targets of known compounds and genetic mechanisms of resistance. By adapting screens that have been successfully performed in model organisms to a mammalian system, we can directly probe the effects of small molecules on the

90

human genome. This screen has the potential to identify targets of novel drugs and could develop into a potent drug discovery tool.

4.4 Methods

4.4.1 TRC library and pools

The RNAi Consortium (TRC) is a huge collaborative effort aimed at developing widely applicable RNAi reagents targeting human and mouse genes. The TRC‟s three-year, $18-million initiative led to development of an shRNA library that contains 160,000 shRNA constructs in lentiviral vectors that target 15,000 human and 15,000 mouse genes237. This resource is amenable to a wide variety of studies and enables the development of new technologies for functional genomics studies in human and mouse. The library was developed in the pLKO.1 puromycin- resistant vector that drives shRNA expression from a human U6 promoter115. The shRNAs in the TRC library were designed to maximize knockdown and minimized off-target effects. For more details on TRC library production, see Moffat et al.115

The 45K TRC sub-library was assembled by combining 1) purified plasmid DNA from 6 pools of 3,300 purified plasmids transformed into DH5-α cells (20,053 clones) and 2) purified plasmid DNA from 7 pools of 3,600 plasmids generated by pooling bacterial clones from 96-well library plates (25,129 clones) (see Luo et al.121 for details). This sub-library was used to generate a 45K lentivirus pool using a three-plasmid lentivirus packaging system238. The 54K pool was assembled by combining 16 normalized subpools of ~3,400 lentiviral pLKO.1 shRNA plasmids (see Cheung et al.239 for details).

4.4.2 Cell culture

HEK293T, A549, Jurkat, and BT474 cells were purchased from ATCC. HEK293T, A549 and BT474 cells were cultured in Dulbecco‟s Modified Eagle Medium (DMEM) containing 10% o fetal bovine serum and 1x penicillin/streptomycin at 37 C/5% CO2. Jurkat cells were cultured in o RPMI media containing 10% fetal bovine serum and 1x penicillin/streptomycin at 37 C/5% CO2.

91

4.4.3 Individual shRNA knockdowns

Lentivirus pools containing individual shRNAs directed against the DNA-repair gene RAD50 (Table 4-3) were generated as described in Moffat et al.115. shRAD50 lentivirus pools were used to infect A549 cells at an MOI of 0.3 and grown in the presence of doxorubicin at an IC25 or IC50 for 48 h and 96 h. Viability of the cells after compound treatment was assessed using the sulforhodamine B assay for cytotoxicity209.

4.4.4 Screens

Lentivirus pools were generated using the 54K plasmid pool as described115. Once this pool was created, A549 cells were infected at an MOI of 0.3. After 2 days of selection in 2 μg/mL puromycin, cells were frozen in liquid nitrogen. To perform each replicate of the screen, 1.6 x 107 infected cells were plated on 2-chamber Nunc Cell Factories (multi-layer tissue culture flasks with a surface area of 1264 cm2). 2 hours after seeding, 300 nM doxorubicin or DMSO control was added to the media. After 72 h, cells were harvested, 1.6 x 107 cells were reseeded into the cell factory, and the procedure was repeated for a total of 15 days. To prepare probes, genomic DNA was extracted from the harvested cells using the Qiagen Blood Maxi kit. PCR amplification was performed using Platinum Pfx polymerase (Invitrogen), 20 μg of genomic DNA as template, 600 nM primers (PCR_B-fw 5'-Biotin- AATGGACTATCATATGCTTACCGTAACTTGAA-3' and PCR_rev 5'- TGTGGATGAATACTGCCATTTGTCTCGAGGTC-3'), 1x Amplification buffer, 1x Enhancer solution, 1 mM MgSO4. The amplification reaction was performed using the following conditions: 94°C for 5 min, 30x(94°C for 15s, 55°C for 15s, 68°C for 20s), 68°C for 5 min. Samples were then electrophoresed on a 2% agarose gel and the 178bp product was excised and extracted using the Qiagen Gel Extraction kit. Additional purification was performed using the QIAquick PCR purification kit. The purified products were then digested with the XhoI restriction endonuclease for 2 hours at 37C to obtain the half-hairpin probe. Affymetrix UT- GMAP microarrays118 were conditioned by incubating with 10 mM NaOH at 40oC for 10 min, then washed with 5 mL of 6x SSPE with 0.0001% Tween 20. Then, 0.0005% Triton X-100 was

92

Table 4-1. TRC lentiviral shRNA clones used in this study.

Gene target shRNA Clone ID Clone Name

RAD50 shRAD50_832 TRCN0000040107 NM_005732.2-832s1c1

RAD50 shRAD50_2674 TRCN0000040104 NM_005732.2-2674s1c1

MYST4 shMYST4_1 TRCN0000222166 NM_012330.1-6181s1c2

MYST4 shMYST4_2 TRCN0000222164 NM_012330.1-3969s1c2

added and incubated at 40oC for 10 min. Hybridization solutions consisted of 1.2 μg of probe for the 54K shRNA pools in buffer containing 1x MES, 0.89 M NaCl, 20 mM EDTA, 0.0001% Tween 20, 0.5 mg/mL BSA, 0.1 mg/mL herring sperm DNA, 0.05 nM biotinylated B2 oligo (Affymetrix), 10% DMSO, 20 μM each blocking oligos (Block_1 5'- AATGGACTATCATATGCTTACCGTAACTTGAA-3', Block_2 5'- TTCAAGTTACGGTAAGCATATGATAGTCCATT-3', Block_3 5'- GTATTTCGATTTCTTGGCTTTATATATCTTGTGGAAAGGACGAAACACCG-3', Block_4 5'-CGGTGTTTCGTCCTTTCCACAAGATATATAAAGCCAAGAAATCGAAATAC-3'), and sterile water to a final volume of 138 μL. Samples in buffer were denatured at 95°C for 10 min, incubated at 40°C for 5 min, collected by centrifugation then applied to arrays, which were incubated for 16 h at 40°C at 60 rpm. Arrays were stained with SAPE labeling mix (1x MES staining buffer, 2 mg/mL BSA, 10 μg/mL streptavidin-phycoerythrin), washed using an Affymetrix fluidics station, and scanned.

Microarray data was extracted using Affymetrix Powertools and GARP scores were generated as described in Marcotte et al.214 Functional enrichment of the hits was generated through the use of IPA (Ingenuity Systems, www.ingenuity.com).

93

4.4.5 Individual hit confirmations

Endoribonuclease-prepared siRNAs (esiRNAs) against hits (Table 4-2) were provided by the Pelletier lab. 6 x 103 A549 cells were seeded in each well of a 48-well plate and incubated at o 37 C/5% CO2 for 24 h. 500 ng esiRNA in 100 μL of OptiMEM and 1 μL of Lipofectamine RNAiMAX in 100 μL OptiMEM were incubated separately at RT for 5 min, then mixed and incubated at RT for 30 min. 50 μL of each transfection mix was added to each well containing cells. Confirmations of individual hairpin sensitivity were performed by incubating cells with 50 nM doxorubicin for 48 h. Cells were then fixed with 3.3% TCA and cell viability was determined using the SRB assay.

4.4.6 Immunoblots

To perform histone H3 acetylation immunoblots on wildtype cells, 1 x 106 A549 cells were plated in a 10 cm2 tissue culture dish and incubated for 18 h. Doxorubicin at varying concentrations was added and cells were harvested 24 h after compound addition. To perform histone H3 acetylation tests on MYST4 knockdowns, lentivirus pools containing shMYST4 clones (Table 4-3) were generated and used to infect wild-type A549 cells. Virus-containing cells were isolated by puromycin selection and the cells were treated with 500 nM doxorubicin for 24 h. Protein extracts were prepared by collecting trypsinized cells, resuspending in 3x Lysis buffer (20 mM Hepes-KOH pH7.9, 100 nM KCl, 200 μM EDTA, 1 mM DTT, 1% Glycerol, 1 EDTA- free mini cOmplete tablet (Roche)) and sonicating 3x with resting on ice for 30 s in between. Samples were centrifuged at 14000 rpm for 20 min, lysate was collected, and samples were run on 10-20% Tris-Glycine Novex gels (Invitrogen). Proteins were transferred onto nitrocellulose membranes, blocked with 5% milk for 2 h at RT, and probed with anti-acetylhistone H3 K9/K14 rabbit antibody (Cell Signaling Technologies) and anti-β-actin mouse antibody (Santa Cruz Biotech) at 4oC overnight. The blots were washed in TBS-T and probed with anti-mouse and anti-rabbit IgG-HRP (GE Healthcare), respectively for 1 h at RT. Blots were washed in TBS-T and developed using SuperSignal West Pico Chemiluminescent Substrate (Pierce).

94

Chapter 5 A systematic analysis of human genes that adversely affect fitness when overexpressed

I performed the overexpression screens for toxicity and confirmations of the hits. Anthony Arnoldo produced the lentivirus pools and infected cells. Anthony Mak and Jason Moffat generated the lentivirus vector backbone. Anthony Arnoldo and Larry Heisler analyzed the data. Corey Nislow supervised the project.

5 5.1 Introduction

Functional genomics takes advantage of manipulating gene dosage to produce detectable phenotypes. The majority of large-scale functional screens employ loss-of-function techniques to elucidate gene function and have been quite successful, however loss-of-function mutants do not always exhibit phenotypes, for example for those genes with complementary or redundant functions. As a complement to these screens, systematic gain-of-function screens can provide a method for examining genes with subtle loss-of-function phenotypes. Increasing the dosage of genes has the potential to reveal dominant and/or alternative phenotypes, allowing the assignment of functions to previously uncharacterized genes.

Several genome-wide overexpression resources have been developed for studies in model organisms. In particular, there are a number of resources for systematic large-scale overexpression screens in S. cerevisiae which have been used to identify “overexpression lethal” genes, genes involved in the cell cycle, drug targets, and to assign gene function240-243. Examples of yeast overexpression libraries include: the GAL1-GST ORF collection240, the GAL1-driven movable ORF collection which is tagged with an His6-HA-Protein A C-terminal tag244, the GAL1-driven FLEX and barFLEX untagged ORF collections245,246, and untagged genome fragments cloned into high-copy 2μ vectors247. Similar screens have also been performed in metazoans - in Drosophila melanogaster, a transposon-based eyGAL4-driven overexpression

95

system was used to identify genes that disrupt eye development248. In Arabidopsis thaliana, full- length genes were ectopically expressed to study morphological phenotypes249. In this chapter, I introduce a genome-scale high-throughput human gene overexpression screen to characterize gene function and highlight its application to identify human genes that are toxic when overexpressed.

The balance of gene dosage is critical for normal function, therefore any changes in gene expression levels that lead to phenotypic changes in an otherwise wild-type background could point to cellular processes that are particularly sensitive to changes in protein levels250. The identification of genes that are lethal when overexpressed can be applied to assign function to dominant genes that interfere with specific biological processes. A systematic array-based yeast genome-wide overexpression study by Sopko et al.240 revealed that up to 15% of all yeast genes are toxic in a wild-type background and that this toxic gene set is enriched for genes involved in transcriptional regulation, the cell cycle, mitosis, and cell signaling. In a separate screen, Jones et al.247 found that only 1.5% of yeast genes were lethal when overexpressed. The different libraries and methods of analyses may explain the variation in results, for example, different plasmid backbones (CEN vs. 2µ), yeast strain backgrounds, and screen formats. Recently, Douglas et al.246 performed overexpression toxicity screens using their untagged barFLEX ORF collection in both pooled and arrayed formats. Both methods of screening identified, on average, 385 toxic strains (~6.5% of the genome) which is a similar number found in a previous screen of the mORF collection yet with only a 25% overlap in the results244. The identification of an analogous set of human genes that cause slow growth or toxicity to mammalian cells will provide insight into the mechanisms of toxicity caused by gene hyperactivation in human diseases; for example, neuronal cell death caused by mutant ubiquitin in Alzheimer‟s disease251, or the loss of pancreatic β-cells by calmodulin overexpression in diabetes252, or retinal degeneration caused by overexpression of rhodopsin253. Furthermore, comparison of those “overexpression lethal/sick” genes could reveal conserved functions that are hypersensitive to gene dose. Potential applications of this study include its use to identify genes or small molecules that suppress overexpression-mediated toxicity.

Here we describe the application of a pooled high-throughput genome-wide overexpression assay to systematically study cytotoxic and slow growth phenotypes resulting from an increased 96

dosage of human genes. We developed a lentiviral-based system for mammalian cell culture, using the Human ORFeome v3.1124, a library containing ~12,000 human ORF constructs, where a single human ORF can be induced to be overexpressed in each clone of a pooled culture. After growing the pool in a condition of choice, the relative growth of each clone in the pool can be assessed by microarray hybridization of the amplified ORFs, analogous to barcode hybridization (Figure 5-1). Using this screen, we can detect those human genes that, when overexpressed in cells, lead to cytotoxicity. We performed screens these screens in HEK293M2 cells, but the approach is applicable to any cell type that can be infected and support expression of the integrated ORF.

5.2 Results

5.2.1 A systematic genome-wide human ORF overexpression screen

We developed an overexpression screening platform by cloning the Human ORFeome v3.1124 into the pLD-IRES-Venus-WPRE-STOP lentiviral vector (generated by the Moffat lab) which contains a tetracycline-responsive element (TRE) in the promoter to induce overexpression and a Venus fluorescent marker following an IRES sequence for detection (Figure 5-2). To activate the TRE promoter, the tetracycline analogue doxycycline is added to the cell culture media after which it binds to the reverse-tetracycline transactivator (rtTA) protein which then binds to the promoter to turn on induction of overexpression. In the absence of doxycycline there is little or no overexpression; therefore, genes that are toxic when overexpressed should not be depleted in the population prior to the start of the screen. The TRE promoter allows for tunable expression by varying the amount of doxycycline. The pLD-IRES-Venus-WPRE-STOP does not contain an epitope tag which could interfere with the function of the exogenous protein. The Venus fluorescent marker allows for detection of exogenous ORFs in cells and, because it is linked to the ORF, induction of ORF expression must have occurred if the marker fluoresces. The disadvantages to using this vector for our overexpression screens are 1) there is no selectable marker and thus cells must be isolated by fluorescence-activated cell sorting (FACS), 2) the lack of protein tag can make follow-up studies more difficult, and 3) the TRE promoter requires a cell line that expresses the rtTA protein.

97

Figure 5-1. The hORFeome overexpression assay. The pooled hORFeome overexpression library is propagated in E. coli and packaged into lentiviruses. The lentivirus pool is used to infect human cells (MOI=0.3) which are grown in the presence of doxycycline to induce expression of the exogenous ORF products (♦). The cells can then be grown in the presence of a high dose of drug such that only those overexpression mutants that confer resistance to the drug survive. Genomic DNA is extracted from the cells and the exogenous ORF sequence is amplified by PCR. The PCR product is biotinylated and hybridized to an Affymetrix Human Gene 1.0ST microarray. The signal intensity obtained from each probe will be analyzed to find the relative abundance of each overexpressed ORF.

98

Figure 5-2. Lentivirus vector map for pLD-T-IRES-Venus-WPRE-STOP.

99

After generating the lentivirus hORF pool, we optimized infection of HEK293-M2s (human embryonic kidney cells expressing rtTA) (J. Moffat, unpublished) by transiently inducing the expression of the hORFs using doxycycline for 24 h and performing FACS to determine the percentage of cells expressing Venus (i.e. those that have been infected). This allowed us to define the amount of virus required to obtain an MOI of 0.3 for the large-scale infections, such that each cell would contain a single ORF construct integrated into its genome. No Venus fluorescence was observed in the absence of doxycycline, indicating tight regulation of expression by the TRE promoter. We performed large-scale infection of the hORFeome library into HEK293-M2 cells and 24 h after doxycycline addition, FACS was used to physically sort and collect the Venus-positive cells that contained the exogenous ORFs. We prepared frozen aliquots of hORF-infected cells for subsequent experiments.

5.2.2 Screen for toxic overexpressed genes in HEK293M2

To perform the screens, the infected cells were grown in the presence or absence of doxycycline for 4 weeks with cells harvested weekly. The screens were performed in triplicate. Genomic DNA was extracted from cells at each timepoint, the ORFs were PCR amplified using standard primers, and the PCR products were sheared, biotinylated, and hybridized to Affymetrix Human Gene 1.0ST arrays254. This array, normally used for expression analysis, has 370,000 probes against the hORFs and is a robust way to detect enrichment of ORF sequences in mixed populations. Because each clone in the pool contains a single gene overexpression construct integrated into the genome, the ORF sequences act as barcodes/tags for each clone where the abundance of each barcode correlates to signal intensity on the microarray.

5.2.3 Analysis of the screen results

The results of the overexpression toxicity screens were analyzed with the help of Larry Heisler and Anthony Arnoldo. Log2 ratios of signal intensities were compared between t0 and each of the 4 weekly timepoints. These ratios were then compared in the presence and absence of doxycycline to eliminate those clones that drop out due to random chance. These lists were then compared between all 4 timepoints to identify genes that overlapped between sets to obtain a higher confidence gene list. The resulting toxic gene set is listed in Table 5-1.

100

5.2.4 Confirmed hits

I confirmed the hits by generating individual overexpression clones in HEK293-M2 cells and assessing cell viability. To this end, I used the piggyBac transposon system255 to introduce individual doxycycline-inducible overexpression constructs into cells (Figure 5-3). Anthony Arnoldo created the PB_TGcMV_Neo vector, which contains a TRE promoter to control overexpression of ORFs. Individual human ORF hits were picked from the human ORFeome v3.1 and cloned into this vector by Gateway reaction. HEK293-M2 cells were transfected with the PB_TGcMV_Neo_hORFx clones along with a plasmid containing the transposase for insertion of the hORFx construct into the genome.These cells were then grown in the presence or absence of doxycycline and viability was assessed using a colony forming assay256. If an overexpressed clone had a viability of <75% of the non-induced clone, it was considered to be confirmed. Using these stringent criteria, 16/46 hits tested confirmed as positive toxic clones (Table 5-2). An additional 11/46 genes had 75%-90% viability when overexpressed compared to the non-induced clone.

5.3 Discussion

In this study, we find that less than 1% of human ORFs in the human ORFeome v3.1 collection exhibit overexpression-mediated toxicity, when applying a stringent threshold. Notably, in a recent screen of S. cerevisiae gene overexpression in a wild-type background only 1.5% of the genome was found to be toxic247. However, we do note that the pooled microarray-based approach and the stringency of the analysis may have masked potential toxic genes. The lethal gene set determined from our screens identifies genes with pronounced dosage sensitivity and warrant further study, as do those genes with less severe phenotypes.

It would be interesting to define the mechanism through which the toxicity is occurring. If the gene overexpression induces apoptosis, this can be confirmed by standard assays, including: examination of morphological phenotypes to detect cell shrinkage, nuclear condensation and DNA fragmentation, and FACS-based analyses of apoptosis markers, such as annexin V, which detects exposed phosphatidylserine residues on the extracellular surface of dying cells257-259.

101

Such studies would be useful in assigning additional functions for genes that are not currently linked to cell death processes. More broadly, it would be interesting to link these genes to predicted determinants of gene dosage-related toxicity. It has been suggested that many dosage sensitive genes are a result of intrinsic protein disorder and disruption of protein-protein interactions, both pairwise and in higher level complexes260.

A particularly informative follow-up study would be to use the lethal gene set as a starting point for genetic modifier screens in which overexpression-mediated toxicity is rescued either by systematically altering the dosage of other genes or by the addition of small molecules. These suppressor or “chemical-suppressor” screens combine both reverse and forward genetics with high throughput readouts. A successful example of such a suppressor approach was the use of a yeast-based screen to identify genes that modify the toxicity of FUS, a contributor to amyotrophic lateral sclerosis (ALS)261. The resulting genes can serve as potential new targets for therapeutics. In addition, genetic modifier screens can be used to determine correct levels of gene expression for gene therapy.

In addition to identifying phenotypes generated from gene overexpression in a wild-type background, our platform can be applied to other types of gain-of-function screens. For example, these screens can be the starting point for the identification of suppressor or enhancer mutations, which are mutation phenotypes that are suppressed or enhanced by overexpression of another gene. These screens can be useful to delineate molecular pathways and add additional functions to genes. Overexpression screens have also been used for drug target identification such as multicopy suppression profiling in S. cerevisiae90,262,263. We can also use overexpression screens to identify genes with similar phenotypes in heterologous hosts.

In summary, we generated an overexpression screening platform and applied it to identify genes that are toxic when overexpressed. This screen provides a resource for future screening projects using the human ORFeome collection. The toxic genes found in this screen warrant further study to identify mechanisms of action and link to human diseases.

102

Figure 5-3. Schematic of confirmation experiments. Each “lethal” ORF found in the large-scale screens was transferred into the piggyBac vector under the control of a tetracycline promoter. Each construct is co-transfected into HEK293M2 cells along with the piggyBac transposase (PBase) which will allow integration of the ORF into the host genome. Positive clones are selected, overexpression of the ORF is induced for 10 days, and cell viability is quantified using a colony forming assay.

103

Table 5-1. Lethal gene set from the overexpression toxicity screen in HEK293-M2s.

Symbol Description BMI1 B lymphoma Mo-MLV insertion region (mouse) C10orf62 C13orf16 hypothetical protein MGC35169 C18orf20 chromosome 18 open reading frame 20 C2orf47 hypothetical protein FLJ22555 C3AR1 complement component 3a receptor 1 CACNG2 calcium channel, voltage-dependent, gamma subunit 2 CBX8 chromobox homolog 8 (Pc class homolog, Drosophila) CDK6 cyclin-dependent kinase 6 CRHBP corticotropin releasing hormone binding protein DCT dopachrome tautomerase (dopachrome delta-isomerase, tyrosine-related protein 2) DCUN1D4 KIAA0276 protein DLX3 distal-less homeo box 3 DULLARD dullard homolog (Xenopus laevis) EAF1 ELL associated factor 1 ELF5 E74-like factor 5 (ets domain transcription factor) FYN FYN oncogene related to SRC, FGR, YES, transcript variant 3 GNS glucosamine (N-acetyl)-6-sulfatase (Sanfilippo disease IIID) HOXB13 homeo box B13 ILK integrin-linked kinase IRF4 interferon regulatory factor 4 LEMD1 LEM domain containing 1, mRNA (cDNA clone MGC:44374 IMAGE:5296437), complete cds. LFNG lunatic fringe homolog (Drosophila) MAP2K5 mitogen-activated protein kinase kinase 5, transcript variant A MED31 CGI-125 protein MITD1 hypothetical protein BC018453 MPDU1 mannose-P-dolichol utilization defect 1 MRPL9 mitochondrial ribosomal protein L9 NADK NAD kinase NUTF2 nuclear transport factor 2 P2RY6 pyrimidinergic receptor P2Y, G-protein coupled, 6, transcript variant 4 PAX8 paired box gene 8, transcript variant PAX8A PPAP2C phosphatidic acid phosphatase type 2C, transcript variant 1 PQLC3 chromosome 2 open reading frame 22 PSMD14 proteasome (prosome, macropain) 26S subunit, non-ATPase, 14, mRNA (cDNA clone MGC:87397 IMAGE:5296432), complete cds. PVALB parvalbumin RAN RAN, member RAS oncogene family RNF181 hypothetical protein LOC51255 RP11-35N6.1 plasticity related gene 3, transcript variant 1 SCAMP4 secretory carrier membrane protein 4 SCNN1B sodium channel, nonvoltage-gated 1, beta (Liddle syndrome) SLC7A8 solute carrier family 7 (cationic amino acid transporter, y+ system), member 8, transcript variant 1 SNAI2 snail homolog 2 (Drosophila) SUMO1P1 SUMO1 pseudogene 1, mRNA (cDNA clone MGC:71987 IMAGE:6619211), complete cds. TBX6 T-box 6, transcript variant 1 THAP1 THAP domain containing, apoptosis associated protein 1 TMEM120A transmembrane protein induced by tumor necrosis factor alpha TRAT1 T-cell receptor interacting molecule TRIB3 tribbles homolog 3 (Drosophila) XCR1 chemokine (C motif) receptor 1 ZDHHC19 zinc finger, DHHC domain containing 19

104

Table 5-2. Results from confirmation studies of toxic genes when overexpressed in HEK293-M2 cells. Average viability of induced vs. not-induced cells with exogenous ORFs stably integrated into the genome after 10 days of growth. (n=3; only results with <90% viability shown)

ORF name Average viability of induced vs. not-induced (%) MITD1 11.0 EAF1 17.3 THAP1 37.8 SCNN1B 48.2 SUMO1P1 53.7 CACNG1 55.2 PQLC3 55.8 P2RY6 55.8 SCAMP4 65.7 FYN 66.1 RAN 68.1 C10orf62 70.4 TMEM120A 72.2 IRF4 72.4 HOXB8 73.9 TRIBB3 74.7 BMI1 77.6 SLC7A2 77.6 ELF5 81.9 SNAI 82.2 PPAP2C 83.5 C3AR1 83.6 DCT 85.5 DCUND14 86.8 MRPL2 87.0 MPDUI 88.3 CRHBP 89.3

105

Looking forward, we are working on a new screen using latest version of the human ORFeome, version 8.1 (gift of David Hill and Marc Vidal), which is comprised of 16,172 distinct ORFs corresponding to 13,833 human genes128. This collection was used to generate the CCSB-Broad lentiviral expression library by cloning the ORF constructs into the pLX304-Blast-V5 vector128. This is a Gateway vector containing a CMV promoter to drive ORF expression with the selectable marker BSR for blasticidin resistance, and a V5 epitope sequence downstream of the ORF cloning site. Unlike the vector used for our hORFeome v3.1 toxicity screens, these ORFs are constitutively expressed at high levels. While there is no “on-and-off” control of overexpression of the ORF, this strong promoter may allow lethal genes to have more pronounced effects, and while ORFs with severe phenotypes may be lost during propagation, this promoter will emphasize/exaggerate phenotypes that were less pronounced in the tet-inducible screens. Because of the selectable marker in this plasmid, cells are not sorted by FACS, reducing pool manipulation and growth prior to initiating the screens. We chose to perform this screen in HeLa S3 cells which are initially grown as adherent cells but can be grown in suspension culture by changing the culture media to media lacking calcium. As adherent cells, they are easy to infect and subsequent screens can be performed in suspension, allowing for larger representation of each of the clones. I am currently performing overexpression toxicity screens in HeLa-S3 cells containing the hORFeome V8.1 in both adherent and suspension culture.

5.4 Methods

5.4.1 Cell culture

HEK293-M2 cells were provided by Jason Moffat. The cells were maintained in DMEM o supplemented with 10% fetal bovine serum and 1x penicillin/streptomycin at 37 C/5% CO2.

5.4.2 Production of hORFeome overexpression pool

The Human ORFeome v3.1 (Lamesch et al., 2007), containing 12,212 hORFs representing 10,214 human genes, was divided into 34 minipools of 376 hORFs each (gift of Troy Moore). The minipools were cloned en masse into the pLD-IRES-Venus-WPRE-STOP lentiviral expression vector (from Anthony Mak and Jason Moffat) by Gateway LR reaction and

106

transformed into ElectroMAX DH5α-E E. coli (Invitrogen) by electroporation. Transformants were selected on LB media containing ampicillin and plasmids were isolated by maxiprep (QIAGEN Plasmid Maxi Kit). Lentivirus was produced by normalizing the amount of DNA in each of the 34 minipools and co-transfecting the plasmid DNA with the packaging plasmid psPAX2 and the envelope plasmid PMD2.G into the packaging cell line HEK293Ts using FuGENE 6 (Roche). HEK293-M2 cells were infected with the virus pool at a multiplicity of infection of 0.3. Doxycycline (2 μg/mL) was added in order to induce the expression of the Venus fluorescent protein and Venus-positive cells (representing stably integrated hORF clones) were sorted using a BD FACSAria.

5.4.3 Screen for toxic mutants

HEK293-M2 cells infected with the hORFeome overexpression library were seeded at 1.2x106 cells in one T-175 flask per replicate for each of the 3 replicates. Cells were grown in the presence of 2 μg/mL doxycycline to induce overexpression of each hORF. In parallel, cells were grown in the absence of doxycycline as a control. Cells were grown in these conditions for 4 weeks and harvested at weekly timepoints (t0, t1, t2, t3 and t4). Genomic DNA was extracted from the cells by incubating in SNET buffer (10 mM Tris pH 8.0, 0.1 M EDTA, 0.5% SDS, 100 μg/mL Proteinase K, 25 μg/mL RNase A) for 1h at 55oC. DNA was purified using phenol- chloroform extraction followed by ethanol precipitation. Exogenous ORFs were amplified by PCR from the gDNA using Platinum PCR SuperMix High Fidelity (Invitrogen), 10 μM primers (pLD-Stop-Seq-Fw - 5‟CGGTACCCGGGGATCCTCTAGTCAGCTGAC3‟; pLD-Stop-Seq- Rev - 5‟CCATTTGTCTCGAGGTCGAGAATTCTAGCTAGAATC3‟), and 500 ng genomic DNA. The amplification reaction was performed using the following conditions: 94°C for 3 min; 94°C for 30 s, 65°C for 30 s, 68°C for 10 min for 35 cycles; 68°C for 15 min. The PCR products were purified using the QIAquick PCR purification kit (Qiagen). The ORFs were biotinylated using the BioPrime DNA Labeling Kit (Invitrogen). Unincorporated biotin-14-dCTP was removed using Illustra Sephadex G-50 (GE Healthcare) in a Centri-Spin 20 column (Princeton Separations). The products were added to an Affymetrix Human Gene 1.0ST array (part # 901087) and hybridized overnight at 45oC. The arrays were washed and then stained with streptavidin-phycoerythrin (0.1 M MES, 0.9 M NaCl, 0.1% Tween 20, 2 mg/mL BSA, 10 μg/mL streptavidin-phycoerythrin).

107

5.4.4 Analysis of screen results

Log2 ratios of signal intensity were compared between t0 and each of the 4 weekly timepoints. These ratios were compared between samples grown in the presence and absence of doxycycline to eliminate those that drop out due to random chance. The toxic gene list obtained for each of the timepoints was compared to subsequent timepoints to identify genes that were in common to obtain a higher confidence in our list.

5.4.5 Confirmation of toxic hits

To confirm the hits from the toxicity assay in HEK293-M2 cells, I used the piggyBac transposon system255. These confirmations required an on/off promoter system so that gene overexpression would not exhibit toxicity before starting the screen. Anthony Arnoldo created a Gateway- compatible vector containing the Tet-ON promoter system along with the piggyBac transposon system255 so that the genes could be easily transfected into cells to generate stable cells lines. I cloned the individual ORF hits into this vector and transfected them into HEK293-M2 cells along with a plasmid containing the pBase enzyme which allows for the genes to be randomly inserted into the DNA. After selecting positive clones, confirmation studies were performed using a colony formation assay256. To this end, HEK293-M2 + PB_hORFx clones were seeded in duplicate at a density of 200 cells/well in a 6-well plate. 1 μg/mL doxycycline was added to 1 of the 2 wells the following day and cells were grown for 10 days, until colonies were visible on the plates. The cells were then fixed with TCA, stained with 0.01% crystal violet, and colonies counted.

108

Chapter 6 Summary and Perspectives 6 6.1 Summary

In this thesis, I describe the application of chemical genomic screens to probe mechanisms of action of chemotherapeutic compounds with the goal of improving current anticancer therapies. First, genome-wide HIPHOP screens in S. cerevisiae and S. pombe were employed to evaluate a library of novel platinum-acridine hybrid agents and revealed that, in the cell, four of these platinum-acridine compounds damage DNA in a manner distinct from cisplatin; these compounds may represent potential alternatives to cisplatin treatment in resistant tumours. This study also highlighted the effectiveness of employing comparative chemical genomics in evolutionarily distant organisms to define mechanisms of action of novel compounds. S. cerevisiae HIPHOP screening was also applied to examine the mechanism of the apoptosis- inducing investigational drug elesclomol. Our screen revealed that instead of targeting a specific protein, elesclomol influences redox reactions in the mitochondria to disrupt the process of electron flow through the mitochondrial electron transport chain (ETC). Increased levels of ROS caused by ETC disruption by elesclomol initiates the apoptotic cascade; this information has helped lift the hold on elesclomol in Phase II trials. The synergistic effect of elesclomol with ETC inhibitors in human cells demonstrates that chemical genomic studies can provide insight into combination therapy strategies; specifically, rather than designing combination therapies based solely on the on-target mechanism of actions, our approach allows one to sum the whole-cell effects of each agent.

To adapt chemical genomic screens to mammalian cells, I developed an RNAi-based chemical genomic screening platform. The assay development aspects of this project were, as anticipated, extremely labour intensive. Furthermore, during the course of this work, I gained an appreciation for the complexities of quantifying fitness defects in mammalian cells, especially with regard to data analysis. I used the screen as an unbiased tool to examine the clinical anticancer drug doxorubicin which has multiple mechanisms associated with its action against cancer cells and

109

causes adverse cardiotoxicity. I identified several potential doxorubicin-interacting proteins that may explain mechanisms attributed to doxorubicin mode-of-action. In one compelling example, I found a potential interaction between doxorubicin and the histone acetyltransferase MYST4 which leads to altered histone H3 acetylation. The link between doxorubicin and histone acetylation represents a potential avenue for combination therapies for cancer.

I also describe a systematic genome-wide human ORF overexpression screening platform that complements loss-of-function assays. The detection of gain-of-function mutants that are resistant to a compound can reveal potential drug targets. While this assay was initially developed to study drug target pathways, I used it to identify those genes that, when overexpressed, are toxic to cells. The screen revealed that less than 1% of ORFs in the human genome exhibit overexpression-mediated toxicity. Our overexpression screening platform demonstrates the utility of the hORFeome collection for functional genomic studies and provides a resource for future projects using this ORF library.

6.2 Perspectives

6.2.1 Chemical genomics

The availability of fully-sequenced genomes has facilitated the development of high-throughput technologies to systematically probe gene function. The work presented in this thesis highlights the utility of chemical genomic screens to identify and validate drug targets to better understand drug mechanism of action. This has important applications in the development and evaluation of novel drugs to improve on current therapeutics. In conjunction with improvements in medicinal chemistry, systematic genome-wide screens provide an effective method to evaluate structure- activity relationships, enabling the prioritization of compounds for drug development197. In addition, the identification of new “druggable” genes that arise from genome-wide compound screens can lead to the design of novel drugs to target these genes81. Results from genomic screens can be combined with other -omic strategies, such as proteomic and transcriptomic studies, to enhance drug target discovery efforts.

The S. cerevisiae HIPHOP screens employed in this thesis have been used to study drug mode- of-action for more than a decade. These chemical genomic screens are highly specific for drug 110

targets and target pathways and can also be used to identify biologically active compounds; their success has been demonstrated in numerous studies using well-characterized and previously uncharacterized compounds93,94,96,98,101. Yeast chemical genomic screens are robust, have a high positive-discovery rate, and a low false-discovery rate. A potential source of false-positive and false-negative results from our pool-based screens is the indirect growth readout using barcode microarrays that may mask true hits. The increasing use of next-generation sequencing to quantify barcodes may address this issue264. The stringency of target-identification using these screens may also contribute to false-negative results since target deletion is expected to sensitize a strain to the drug and target overexpression is expected to cause resistance, therefore, drug- gene interactions that produce altered phenotypes may be missed. In addition, because these screens are designed to identify synthetic interactions, biochemical assays will be required to identify the nature of the drug/drug-target interaction, i.e. whether the compound is directly binding to the target to produce the observed phenotype.

The findings from chemical genomic screens can be used to inform the development of combination therapy strategies for current clinical drugs. The identification of a specific gene dosage mutant that enhances the function of a compound can be used to select drugs that mimic the gene dosage effect for combination therapies, such as in Chapter 3 where we identify ETC inhibitors that are synergistic with elesclomol. Targeting the synthetic lethal partner of a drug can be used to address issues of drug resistance and also allow for decreased individual drug dose to circumvent adverse dose-related toxicity. In addition to their role in drug discovery, chemical genomic screens allow the identification of compounds that can be used as chemical probes.

The yeast Saccharomyces cerevisiae continues to be an ideal model organism to develop new tools for genomic screening due to the extensive characterization of its genome and proteome. Examples of recent chemical genomics tools developed in yeast include novel genome-scale overexpression collections246,265, differential epistasis mapping165 , and the use of next-generation sequencing to detect molecular barcodes264. Assays developed in yeast can be adapted for use in higher organisms. Alternatively, results from yeast-based screens can be directly verified in human homologues. Chemical genomics in model organisms will continue to play a major role in drug research and development. 111

6.2.2 Genome-wide screens in mammalian systems

A goal of my work is to enable chemical genomic screens to be performed directly in mammalian systems. This will allow the direct study of genes and pathways that do not have homologues in yeast. Although heterologous systems, in which human genes are introduced to model organisms, have been useful for studying human proteins266,267, examining a protein‟s response to drug in its native environment would be preferable. Also, chemical screens in mammalian systems would allow compounds to be used at physiologically-relevant doses. Once a technology is established in a “standard” cell line, genomic screens can be tailored to a specific disease-relevant system. In practice, no one cell line can be expected to model all human diseases; instead, a panel of reference lines for each disease, complemented with patient-derived cells, may be the most effective approach. Examples of these systems include primary tumours, cell lines engineered to model disease states, and patient-derived induced pluripotent stem cells that model specific diseases268.

The application of RNAi to high-throughput screening technologies has greatly accelerated functional genomics studies in human cells. Results of screens have led to a better understanding of biological processes including cancer, cell death, and host-pathogen interactions119,121,214,269- 271. However, there are still many hurdles to performing RNAi screens, including high false- positive and false-negative rates inherent to large-scale screens272. For example, in three independent screens to identify host factors required for HIV infection, there was limited overlap of hits269,270,273. False-negatives in RNAi screens can arise from a number of factors including variable gene suppression by different RNAi constructs or weak RNAi-mediated knock-down phenotypes which may be undetected or masked in large-scale screens. False-positive rates can be attributed to off-target effects of RNAi molecules; these off-target effects may also obscure true positive hits. Validation of the efficacy of RNAi molecules, independent of screen results, is required to a) evaluate the potency of an RNAi molecule to produce target-gene knockdown, and b) identify any off-target effects of an RNAi molecule. In addition, the development of methods for validation of results from RNAi screens, using statistical, bioinformatics, and experimental approaches, will be essential to confirm findings from primary screens. The false discovery rates associated with RNAi screens are expected to decrease with increased large-scale screening efforts that will lead to more standardized methodologies and validation approaches.

112

While examples of genome-wide overexpression screens in mammalian systems are currently limited, the recent generation of several human ORF libraries will resolve this issue124-126,128. Integrated genomic approaches developed in mammalian cells, similar to work in S. cerevisiae90, will greatly enhance biological research. New large-scale RNAi and ORF resources for human genome research will bring about innovative technologies to functional biology research. In addition, newer technologies are being developed that enable precise genetic manipulation in human cells and should prove scalable to the genome-wide level. One example is the generation of human gene knockout cells using haploid human cell lines274. Loss-of-function chemical screens using this technology allow the study of null alleles, emulating screens in model organisms274,275. Advances in genetic engineering in mammalian systems include the development of zinc finger nucleases (ZFN)276 and transcription activator-like effector nucleases (TALEN)277. These engineered nucleases can be used to target endogenous human genes and enable site-directed insertions, deletions, and DNA editing in human cells.

Despite significant advances enabling functional genomics studies in mammalian cells, the current state of mammalian-based genomic screening has not reached the level of maturity of model organism-based screens. Therefore, chemical genomic screens in model organisms such as S. cerevisiae will continue to be important for characterizing drug mechanism of action. With further developments in the use and understanding of RNAi, overexpression, and genome-editing technologies, the next decade holds great promise for chemical genomics in mammalian cells.

113

7 References

1. David, A.R. & Zimmerman, M.R. Cancer: an old disease, a new disease or something in between? Nat Rev Cancer 10, 728-33 (2010).

2. World Health Organization. Cancer Fact Sheet. http://www.who.int/mediacentre/factsheets/fs297/en/index.html. (2012).

3. Sporn, M.B. The war on cancer. Lancet 347, 1377-81 (1996).

4. Chabner, B.A. & Roberts, T.G., Jr. Timeline: Chemotherapy and the war on cancer. Nat Rev Cancer 5, 65-72 (2005).

5. Krumbhaar, E.B. & Krumbhaar, H.D. The Blood and Bone Marrow in Yelloe Cross Gas (Mustard Gas) Poisoning: Changes produced in the Bone Marrow of Fatal Cases. J Med Res 40, 497-508 3 (1919).

6. Gilman, A. & Philips, F.S. The Biological Actions and Therapeutic Applications of the B-Chloroethyl Amines and Sulfides. Science 103, 409-36 (1946).

7. Gilman, A. The initial clinical trial of nitrogen mustard. Am J Surg 105, 574-8 (1963).

8. Goodman, L.S., Wintrobe, M.M. & et al. Nitrogen mustard therapy; use of methyl-bis (beta-chloroethyl) amine hydrochloride and tris (beta-chloroethyl) amine hydrochloride for Hodgkin's disease, lymphosarcoma, leukemia and certain allied and miscellaneous disorders. J Am Med Assoc 132, 126-32 (1946).

9. Farber, S. Some observations on the effect of folic acid antagonists on acute leukemia and other forms of incurable cancer. Blood 4, 160-7 (1949).

10. Wills, L., Clutterbuck, P.W. & Evans, B.D. A new factor in the production and cure of macrocytic anaemias and its relation to other haemopoietic principles curative in pernicious anaemia. Biochem J 31, 2136-47 (1937).

11. Farber, S. & Diamond, L.K. Temporary remissions in acute leukemia in children produced by folic acid antagonist, 4-aminopteroyl-glutamic acid. N Engl J Med 238, 787- 93 (1948).

12. Philips, F.S., Thiersch, J.B. & Ferguson, F.C. Studies of the action of 4- aminopteroylglutamic acid and its congeners in mammals. Ann N Y Acad Sci 52, 1349-59 (1950).

114

13. Li, M.C., Hertz, R. & Bergenstal, D.M. Therapy of choriocarcinoma and related trophoblastic tumors with folic acid and purine antagonists. N Engl J Med 259, 66-74 (1958).

14. Zhou, B.B. & Elledge, S.J. The DNA damage response: putting checkpoints in perspective. Nature 408, 433-439 (2000).

15. Hanahan, D. & Weinberg, R.A. Hallmarks of cancer: the next generation. Cell 144, 646- 674 (2011).

16. Baguley, B.C. Chapter 1 - A brief history of cancer chemotherapy. in Anticancer Drug Development (eds. Bruce, C.B. & David, J.K.) 1-11 (Academic Press, San Diego, 2002).

17. Hurley, L.H. DNA and its associated processes as targets for cancer therapy. Nat Rev Cancer 2, 188-200 (2002).

18. Zhu, Y., Hu, J., Hu, Y. & Liu, W. Targeting DNA repair pathways: a novel approach to reduce cancer therapeutic resistance. Cancer Treat Rev 35, 590-6 (2009).

19. Whittington, R.M. & Close, H.P. Clinical experience with mitomycin C (NSC-26980). Cancer Chemother Rep 54, 195-8 (1970).

20. Murray-Lyon, I.M. et al. Treatment of multiple-hormone-producing malignant islet-cell tumour with . Lancet 2, 895-8 (1968).

21. Fischhaber, P.L., Gall, A.S., Duncan, J.A. & Hopkins, P.B. Direct demonstration in synthetic oligonucleotides that N,N'-bis(2-chloroethyl)-nitrosourea cross links N1 of deoxyguanosine to N3 of deoxycytidine on opposite strands of duplex DNA. Cancer Res 59, 4363-8 (1999).

22. Rosenberg, B., Vancamp, L. & Krigas, T. Inhibition of Cell Division in Escherichia Coli by Electrolysis Products from a Platinum Electrode. Nature 205, 698-9 (1965).

23. Siddik, Z.H. Cisplatin: mode of cytotoxic action and molecular basis of resistance. Oncogene 22, 7265-7279 (2003).

24. Kelland, L. The resurgence of platinum-based cancer chemotherapy. Nature reviews.Cancer 7, 573-584 (2007).

25. Takahara, P.M., Rosenzweig, A.C., Frederick, C.A. & Lippard, S.J. Crystal structure of double-stranded DNA containing the major adduct of the anticancer drug cisplatin. Nature 377, 649-52 (1995).

26. Huang, H., Zhu, L., Reid, B.R., Drobny, G.P. & Hopkins, P.B. Solution structure of a cisplatin-induced DNA interstrand cross-link. Science 270, 1842-5 (1995).

27. Wheate, N.J., Walker, S., Craig, G.E. & Oun, R. The status of platinum anticancer drugs in the clinic and in clinical trials. Dalton Trans 39, 8113-27 (2010). 115

28. Koepf-Maier, P. & Koepf, H. Non-platinum group metal antitumor agents. History, current status, and perspectives. Chemical reviews 87, 1137-1152 (1987).

29. Köpf-Maier, P. Complexes of metals other than platinum as antitumour agents. European Journal of Clinical Pharmacology 47, 1-16 (1994).

30. Bruijnincx, P.C. & Sadler, P.J. New trends for metal complexes with anticancer activity. Curr Opin Chem Biol 12, 197-206 (2008).

31. Galmarini, C.M., Mackey, J.R. & Dumontet, C. Nucleoside analogues and nucleobases in cancer treatment. Lancet Oncol 3, 415-24 (2002).

32. Parker, W.B. & Cheng, Y.C. Metabolism and mechanism of action of 5-fluorouracil. Pharmacol Ther 48, 381-95 (1990).

33. Blackledge, G. New developments in cancer treatment with the novel thymidylate synthase inhibitor raltitrexed ('Tomudex'). Br J Cancer 77 Suppl 2, 29-37 (1998).

34. Bertino, J.R. & Cronstein, B.N. Methotrexate, 139 p. (Birkhäuser Verlag, Basel ; Boston, 2000).

35. Froelich-Ammon, S.J. & Osheroff, N. Topoisomerase poisons: harnessing the dark side of enzyme mechanism. J Biol Chem 270, 21429-32 (1995).

36. Hande, K.R. Etoposide: four decades of development of a topoisomerase II inhibitor. Eur J Cancer 34, 1514-21 (1998).

37. Wozniak, A.J. & Ross, W.E. DNA damage as a basis for 4'-demethylepipodophyllotoxin- 9-(4,6-O-ethylidene-beta-D-glucopyranoside) (etoposide) cytotoxicity. Cancer Res 43, 120-4 (1983).

38. Chen, G.L. et al. Nonintercalative antitumor drugs interfere with the breakage-reunion reaction of mammalian DNA topoisomerase II. J Biol Chem 259, 13560-6 (1984).

39. Burgess, D.J. et al. Topoisomerase levels determine chemotherapy response in vitro and in vivo. Proc Natl Acad Sci U S A 105, 9053-8 (2008).

40. Hsiang, Y.H., Hertzberg, R., Hecht, S. & Liu, L.F. Camptothecin induces protein-linked DNA breaks via mammalian DNA topoisomerase I. J Biol Chem 260, 14873-8 (1985).

41. Hertzberg, R.P., Caranfa, M.J. & Hecht, S.M. On the mechanism of topoisomerase I inhibition by camptothecin: evidence for binding to an enzyme-DNA complex. Biochemistry 28, 4629-38 (1989).

42. Cutts, S.M., Nudelman, A., Rephaeli, A. & Phillips, D.R. The power and potential of doxorubicin-DNA adducts. IUBMB Life 57, 73-81 (2005).

116

43. Minotti, G., Menna, P., Salvatorelli, E., Cairo, G. & Gianni, L. Anthracyclines: molecular advances and pharmacologic developments in antitumor activity and cardiotoxicity. Pharmacological reviews 56, 185-229 (2004).

44. Olson, R.D. & Mushlin, P.S. Doxorubicin cardiotoxicity: analysis of prevailing hypotheses. FASEB J 4, 3076-86 (1990).

45. Golstein, P. & Kroemer, G. Cell death by necrosis: towards a molecular definition. Trends in Biochemical Sciences 32, 37-43 (2007).

46. Warburg, O.D., Frank M.A. The Metabolism of Tumors. American Journal of the Medical Sciences. 182, 123 (1931).

47. Kroemer, G. Mitochondria in cancer. Oncogene 25, 4630-2 (2006).

48. Hanahan, D. & Weinberg, R.A. The hallmarks of cancer. Cell 100, 57-70 (2000).

49. Canter, J.A., Kallianpur, A.R., Parl, F.F. & Millikan, R.C. Mitochondrial DNA G10398A polymorphism and invasive breast cancer in African-American women. Cancer Res 65, 8028-33 (2005).

50. Petros, J.A. et al. mtDNA mutations increase tumorigenicity in prostate cancer. Proc Natl Acad Sci U S A 102, 719-24 (2005).

51. Galluzzi, L., Larochette, N., Zamzami, N. & Kroemer, G. Mitochondria as therapeutic targets for cancer chemotherapy. Oncogene 25, 4812-30 (2006).

52. Fulda, S., Galluzzi, L. & Kroemer, G. Targeting mitochondria for cancer therapy. Nat Rev Drug Discov 9, 447-64 (2010).

53. Sawicka, M., Kalinowska, M., Skierski, J. & Lewandowski, W. A review of selected anti-tumour therapeutic agents and reasons for multidrug resistance occurrence. J Pharm Pharmacol 56, 1067-81 (2004).

54. Gottesman, M.M. Mechanisms of cancer drug resistance. Annu Rev Med 53, 615-27 (2002).

55. Chu, G. Cellular responses to cisplatin. The roles of DNA-binding proteins and DNA repair. J Biol Chem 269, 787-90 (1994).

56. Zamble, D.B. & Lippard, S.J. Cisplatin and DNA repair in cancer chemotherapy. Trends Biochem Sci 20, 435-9 (1995).

57. Wang, D. & Lippard, S.J. Cellular processing of platinum anticancer drugs. Nat Rev Drug Discov 4, 307-20 (2005).

58. Yao, X., Panichpisal, K., Kurtzman, N. & Nugent, K. Cisplatin nephrotoxicity: a review. Am J Med Sci 334, 115-24 (2007).

117

59. Amptoulach, S. & Tsavaris, N. Neurotoxicity caused by the treatment with platinum analogues. Chemother Res Pract 2011, 843019 (2011).

60. McWhinney, S.R., Goldberg, R.M. & McLeod, H.L. Platinum neurotoxicity pharmacogenetics. Mol Cancer Ther 8, 10-6 (2009).

61. Olivi, A. et al. Direct delivery of platinum-based antineoplastics to the central nervous system: a toxicity and ultrastructural study. Cancer Chemother Pharmacol 31, 449-54 (1993).

62. Swain, S.M., Whaley, F.S. & Ewer, M.S. Congestive heart failure in patients treated with doxorubicin: a retrospective analysis of three trials. Cancer 97, 2869-79 (2003).

63. Simunek, T. et al. Anthracycline-induced cardiotoxicity: overview of studies examining the roles of oxidative stress and free cellular iron. Pharmacol Rep 61, 154-71 (2009).

64. Berthiaume, J.M. & Wallace, K.B. Adriamycin-induced oxidative mitochondrial cardiotoxicity. Cell Biol Toxicol 23, 15-25 (2007).

65. Carvalho, C. et al. Doxorubicin: the good, the bad and the ugly effect. Current medicinal chemistry 16, 3267-3285 (2009).

66. Zhang, S. et al. Identification of the molecular basis of doxorubicin-induced cardiotoxicity. Nat Med 18, 1639-42 (2012).

67. Tangpong, J. et al. Adriamycin-induced, TNF-alpha-mediated central nervous system toxicity. Neurobiol Dis 23, 127-39 (2006).

68. Hambley, T.W. The influence of structure on the activity and toxicity of Pt anti-cancer drugs. Coordination Chemistry Reviews 166, 181 (1997).

69. Rixe, O. et al. Oxaliplatin, tetraplatin, cisplatin, and carboplatin: spectrum of activity in drug-resistant cell lines and in the cell lines of the National Cancer Institute's Anticancer Drug Screen panel. Biochem Pharmacol 52, 1855-65 (1996).

70. Spingler, B., Whittington, D.A. & Lippard, S.J. 2.4 A crystal structure of an oxaliplatin 1,2-d(GpG) intrastrand cross-link in a DNA dodecamer duplex. Inorg Chem 40, 5596- 602 (2001).

71. Raymond, E., Faivre, S., Chaney, S., Woynarowski, J. & Cvitkovic, E. Cellular and molecular pharmacology of oxaliplatin. Mol Cancer Ther 1, 227-35 (2002).

72. Holford, J., Sharp, S.Y., Murrer, B.A., Abrams, M. & Kelland, L.R. In vitro circumvention of cisplatin resistance by the novel sterically hindered platinum complex AMD473. Br J Cancer 77, 366-73 (1998).

73. Ciuleanu, T. et al. Randomized phase III study (SPEAR) of picoplatin plus best supportive care (BSC) or BSC alone in patients (pts) with SCLC refractory or progressive 118

within 6 months after first-line platinum-based chemotherapy. in J Clin Oncol (ASCO meeting abstracts) Vol. 28 7002 (2010).

74. Wong, E. & Giandomenico, C.M. Current status of platinum-based antitumor drugs. Chemical reviews 99, 2451-2466 (1999).

75. Baruah, H., Barry, C.G. & Bierbach, U. Platinum-intercalator conjugates: from DNA- targeted cisplatin derivatives to adenine binding complexes as potential modulators of gene regulation. Current topics in medicinal chemistry 4, 1537-1549 (2004).

76. Bowler, B.E., Hollis, L.S. & Lippard, S.J. Synthesis and DNA binding and photonicking properties of acridine orange linked by a polymethylene tether to (1,2- diaminoethane)dichloroplatinum(II). Journal of the American Chemical Society 106, 6102 (1984).

77. Sundquist, W.I., Bancroft, D.P., Chassot, L. & Lippard, S.J. DNA promotes the reaction of cis-diamminedichloroplatinum(II) with the exocyclic amino groups of ethidium bromide. Journal of the American Chemical Society 110, 8559 (1988).

78. Sundquist, W.I., Bancroft, D.P. & Lippard, S.J. Synthesis, characterization, and biological activity of cis-diammineplatinum(II) complexes of the DNA intercalators 9- aminoacridine and chloroquine. Journal of the American Chemical Society 112, 1590 (1990).

79. Zunino, F., Savi, G. & Pasini, A. Synthesis and antitumor activity of a platinum(II)- doxorubicin complex. Cancer chemotherapy and pharmacology 1118, 180 (1986).

80. Gagnon, V. et al. Biological evaluation of novel estrogen-platinum(II) hybrid molecules on uterine and ovarian cancers-molecular modeling studies. Bioorg Med Chem Lett 14, 5919-24 (2004).

81. Hopkins, A.L. & Groom, C.R. The druggable genome. Nat Rev Drug Discov 1, 727-30 (2002).

82. Orth, A.P., Batalov, S., Perrone, M. & Chanda, S.K. The promise of genomics to identify novel therapeutic targets. Expert Opin Ther Targets 8, 587-96 (2004).

83. National Institutes of Health. Fact Sheet - Human Genome Project. (October 2010).

84. Taylor, E.W. The Mechanism of Colchicine Inhibition of Mitosis. I. Kinetics of Inhibition and the Binding of H3-Colchicine. J Cell Biol 25, SUPPL:145-60 (1965).

85. Botstein, D., Chervitz, S.A. & Cherry, M. Yeast as a Model Organism. Science 277, 1259-1260 (1997).

86. Giaever, G. et al. Functional profiling of the Saccharomyces cerevisiae genome. Nature 418, 387-391 (2002).

119

87. Winzeler, E.A. et al. Functional characterization of the S. cerevisiae genome by gene deletion and parallel analysis. Science 285, 901-6 (1999).

88. Pierce, S.E. et al. A unique and universal molecular barcode array. Nature methods 3, 601-603 (2006).

89. Smith, A.M. et al. Quantitative phenotyping via deep barcode sequencing. Genome Res 19, 1836-42 (2009).

90. Hoon, S. et al. An integrated platform of genomic assays reveals small-molecule bioactivities. Nature chemical biology 4, 498-506 (2008).

91. Smith, A.M., Ammar, R., Nislow, C. & Giaever, G. A survey of yeast genomic assays for drug and target discovery. Pharmacol Ther 127, 156-64 (2010).

92. Giaever, G. et al. Genomic profiling of drug sensitivities via induced haploinsufficiency. Nature genetics 21, 278-283 (1999).

93. Giaever, G. et al. Chemogenomic profiling: identifying the functional interactions of small molecules in yeast. Proceedings of the National Academy of Sciences of the United States of America 101, 793-798 (2004).

94. Lum, P.Y. et al. Discovering modes of action for therapeutic compounds using a genome- wide screen of yeast heterozygotes. Cell 116, 121-137 (2004).

95. St Onge, R.P. et al. Systematic pathway analysis using high-resolution fitness profiling of combinatorial gene deletions. Nat Genet 39, 199-206 (2007).

96. Hillenmeyer, M.E. et al. The chemical genomic portrait of yeast: uncovering a phenotype for all genes. Science (New York, N.Y.) 320, 362-365 (2008).

97. Heitman, J., Movva, N.R. & Hall, M.N. Targets for cell cycle arrest by the immunosuppressant rapamycin in yeast. Science 253, 905-9 (1991).

98. Birrell, G.W., Giaever, G., Chu, A.M., Davis, R.W. & Brown, J.M. A genome-wide screen in Saccharomyces cerevisiae for genes affecting UV radiation sensitivity. Proceedings of the National Academy of Sciences of the United States of America 98, 12608-12613 (2001).

99. Chang, M., Bellaoui, M., Boone, C. & Brown, G.W. A genome-wide screen for methyl methanesulfonate-sensitive mutants reveals genes required for S phase progression in the presence of DNA damage. Proceedings of the National Academy of Sciences of the United States of America 99, 16934-16939 (2002).

100. Workman, C.T. et al. A systems approach to mapping DNA damage response pathways. Science (New York, N.Y.) 312, 1054-1059 (2006).

120

101. Lee, W. et al. Genome-wide requirements for resistance to functionally distinct DNA- damaging agents. PLoS genetics 1, e24 (2005).

102. Luesch, H. et al. A genome-wide overexpression screen in yeast for small-molecule target identification. Chemistry & biology 12, 55-63 (2005).

103. Fire, A. et al. Potent and specific genetic interference by double-stranded RNA in Caenorhabditis elegans. Nature 391, 806-11 (1998).

104. Bridge, A.J., Pebernard, S., Ducraux, A., Nicoulaz, A.L. & Iggo, R. Induction of an interferon response by RNAi vectors in mammalian cells. Nat Genet 34, 263-4 (2003).

105. Zamore, P.D., Tuschl, T., Sharp, P.A. & Bartel, D.P. RNAi: double-stranded RNA directs the ATP-dependent cleavage of mRNA at 21 to 23 nucleotide intervals. Cell 101, 25-33 (2000).

106. Elbashir, S.M. et al. Duplexes of 21-nucleotide RNAs mediate RNA interference in cultured mammalian cells. Nature 411, 494-8 (2001).

107. Brummelkamp, T.R., Bernards, R. & Agami, R. A system for stable expression of short interfering RNAs in mammalian cells. Science 296, 550-3 (2002).

108. Paddison, P.J., Caudy, A.A., Bernstein, E., Hannon, G.J. & Conklin, D.S. Short hairpin RNAs (shRNAs) induce sequence-specific silencing in mammalian cells. Genes Dev 16, 948-58 (2002).

109. Bartel, D.P. MicroRNAs: genomics, biogenesis, mechanism, and function. Cell 116, 281- 97 (2004).

110. Lee, Y., Jeon, K., Lee, J.T., Kim, S. & Kim, V.N. MicroRNA maturation: stepwise processing and subcellular localization. EMBO J 21, 4663-70 (2002).

111. Lee, Y. et al. The nuclear RNase III Drosha initiates microRNA processing. Nature 425, 415-9 (2003).

112. Yang, D. et al. Short RNA duplexes produced by hydrolysis with Escherichia coli RNase III mediate effective RNA interference in mammalian cells. Proc Natl Acad Sci U S A 99, 9942-7 (2002).

113. Kittler, R., Heninger, A.K., Franke, K., Habermann, B. & Buchholz, F. Production of endoribonuclease-prepared short interfering RNAs for gene silencing in mammalian cells. Nat Methods 2, 779-84 (2005).

114. Hannon, G.J. RNA interference. Nature 418, 244-51 (2002).

115. Moffat, J. et al. A lentiviral RNAi library for human and mouse genes applied to an arrayed viral high-content screen. Cell 124, 1283-1298 (2006).

121

116. Bernards, R., Brummelkamp, T.R. & Beijersbergen, R.L. shRNA libraries and their use in cancer genetics. Nat Meth 3, 701-706 (2006).

117. Olson, A., Sheth, N., Lee, J.S., Hannon, G. & Sachidanandam, R. RNAi Codex: a portal/database for short-hairpin RNA (shRNA) gene-silencing constructs. Nucleic Acids Res 34, D153-7 (2006).

118. Ketela, T. et al. A comprehensive platform for highly multiplexed mammalian functional genetic screens. BMC Genomics 12, 213 (2011).

119. Luo, J. et al. A genome-wide RNAi screen identifies multiple synthetic lethal interactions with the Ras oncogene. Cell 137, 835-48 (2009).

120. Hurov, K.E., Cotta-Ramusino, C. & Elledge, S.J. A genetic screen identifies the Triple T complex required for DNA damage signaling and ATM and ATR stability. Genes Dev 24, 1939-50 (2010).

121. Luo, B. et al. Highly parallel identification of essential genes in cancer cells. Proceedings of the National Academy of Sciences of the United States of America 105, 20380-20385 (2008).

122. O'Connell, B.C. et al. A genome-wide camptothecin sensitivity screen identifies a mammalian MMS22L-NFKBIL2 complex required for genomic stability. Mol Cell 40, 645-57 (2010).

123. O'Donnell, L. et al. The MMS22L-TONSL complex mediates recovery from replication stress and homologous recombination. Mol Cell 40, 619-31 (2010).

124. Lamesch, P. et al. hORFeome v3.1: a resource of human open reading frames representing over 10,000 human genes. Genomics 89, 307-315 (2007).

125. Bechtel, S. et al. The full-ORF clone resource of the German cDNA Consortium. BMC Genomics 8, 399 (2007).

126. Ota, T. et al. Complete sequencing and characterization of 21,243 full-length human cDNAs. Nat Genet 36, 40-5 (2004).

127. Strausberg, R.L., Feingold, E.A., Klausner, R.D. & Collins, F.S. The mammalian gene collection. Science 286, 455-7 (1999).

128. Yang, X. et al. A public genome-scale lentiviral expression library of human ORFs. Nat Methods 8, 659-61 (2011).

129. Boehm, J.S. et al. Integrative genomic approaches identify IKBKE as a breast cancer oncogene. Cell 129, 1065-79 (2007).

130. Johannessen, C.M. et al. COT drives resistance to RAF inhibition through MAP kinase pathway reactivation. Nature 468, 968-72 (2010). 122

131. Ramachandran, S. & Sundaresan, V. Transposons as tools for functional genomics. Plant Physiology and Biochemistry 39, 243-252 (2001).

132. Miskey, C., Izsvak, Z., Kawakami, K. & Ivics, Z. DNA transposons in vertebrate functional genomics. Cell Mol Life Sci 62, 629-41 (2005).

133. De, S., Cipriano, R., Jackson, M.W. & Stark, G.R. Overexpression of kinesins mediates docetaxel resistance in breast cancer cells. Cancer Res 69, 8035-42 (2009).

134. Abrams, M.J. & Murrer, B.A. Metal compounds in therapy and diagnosis. Science (New York, N.Y.) 261, 725-730 (1993).

135. Kartalou, M. & Essigmann, J.M. Mechanisms of resistance to cisplatin. Mutat Res 478, 23-43 (2001).

136. Connors, T.A., Cleare, M.J. & Harrap, K.R. Structure-activity relationships of the antitumor platinum coordination complexes. Cancer Treat Rep 63, 1499-502 (1979).

137. Cleare, M.J. & Hoeschele, J.D. Studies on the antitumor activity of group VIII transition metal complexes. Part I. Platinum (II) complexes. Bioinorganic Chemistry 2, 187-210 (1973).

138. Manzotti, C. et al. BBR 3464: a novel triplatinum complex, exhibiting a preclinical profile of antitumor efficacy different from cisplatin. Clin Cancer Res 6, 2626-34 (2000).

139. Cosaert, J. & Quoix, E. Platinum drugs in the treatment of non-small-cell lung cancer. Br J Cancer 87, 825-33 (2002).

140. Hensing, T.A. et al. Phase II study of BBR 3464 as treatment in patients with sensitive or refractory small cell lung cancer. Anticancer Drugs 17, 697-704 (2006).

141. Jodrell, D.I. et al. Phase II studies of BBR3464, a novel tri-nuclear platinum complex, in patients with gastric or gastro-oesophageal adenocarcinoma. Eur J Cancer 40, 1872-7 (2004).

142. Guddneppanavar, R. & Bierbach, U. Adenine-N3 in the DNA minor groove - an emerging target for platinum containing anticancer pharmacophores. Anti-cancer agents in medicinal chemistry 7, 125-138 (2007).

143. Hess, S.M. et al. Platinum-acridinylthiourea conjugates show cell line-specific cytotoxic enhancement in H460 lung carcinoma cells compared to cisplatin. Cancer chemotherapy and pharmacology 56, 337-343 (2005).

144. Hess, S.M., Anderson, J.G. & Bierbach, U. A non-crosslinking platinum-acridine hybrid agent shows enhanced cytotoxicity compared to clinical BCNU and cisplatin in glioblastoma cells. Bioorganic & medicinal chemistry letters 15, 443-446 (2005).

123

145. Martins, E.T. et al. Design, synthesis, and biological activity of a novel non-cisplatin-type platinum-acridine pharmacophore. Journal of medicinal chemistry 44, 4492-4496 (2001).

146. Baruah, H., Wright, M.W. & Bierbach, U. Solution structural study of a DNA duplex containing the guanine-N7 adduct formed by a cytotoxic platinum-acridine hybrid agent. Biochemistry 44, 6059-6070 (2005).

147. Augustus, T.M., Anderson, J., Hess, S.M. & Bierbach, U. Bis(acridinylthiourea)platinum(II) complexes: synthesis, DNA affinity, and biological activity in glioblastoma cells. Bioorganic & medicinal chemistry letters 13, 855-858 (2003).

148. Guddneppanavar, R. et al. Effect of the diamine nonleaving group in platinum- acridinylthiourea conjugates on DNA damage and cytotoxicity. Journal of medicinal chemistry 50, 2259-2263 (2007).

149. Ma, Z. et al. A non-cross-linking platinum-acridine agent with potent activity in non- small-cell lung cancer. Journal of medicinal chemistry 51, 7574-7580 (2008).

150. Graham, L.A. et al. Unusual Reactivity of a Potent Platinum-Acridine Hybrid Antitumor Agent. ACS Med Chem Lett 2, 687-691 (2011).

151. Shabtai, D., Giaever, G. & Nislow, C. An algorithm for chemical genomic profiling that minimizes batch effects: Bucket evaluations. BMC Bioinformatics 13, 245 (2012).

152. Costanzo, M. et al. The genetic landscape of a cell. Science 327, 425-31 (2010).

153. Grossmann, K.F., Ward, A.M., Matkovic, M.E., Folias, A.E. & Moses, R.E. S. cerevisiae has three pathways for DNA interstrand crosslink repair. Mutation research 487, 73-83 (2001).

154. Wu, H.I., Brown, J.A., Dorie, M.J., Lazzeroni, L. & Brown, J.M. Genome-wide identification of genes conferring resistance to the anticancer agents cisplatin, oxaliplatin, and mitomycin C. Cancer Res 64, 3940-8 (2004).

155. Brendel, M. et al. Role of PSO genes in repair of DNA damage of Saccharomyces cerevisiae. Mutation research 544, 179-193 (2003).

156. Lee, W. et al. Genome-wide requirements for resistance to functionally distinct DNA- damaging agents. PLoS Genet 1, e24 (2005).

157. Barry, C.G., Baruah, H. & Bierbach, U. Unprecedented monofunctional metalation of adenine nucleobase in guanine- and thymine-containing dinucleotide sequences by a cytotoxic platinum-acridine hybrid agent. Journal of the American Chemical Society 125, 9629-9637 (2003).

124

158. Budiman, M.E., Alexander, R.W. & Bierbach, U. Unique base-step recognition by a platinum-acridinylthiourea conjugate leads to a DNA damage profile complementary to that of the anticancer drug cisplatin. Biochemistry 43, 8560-8567 (2004).

159. Guddneppanavar, R., Saluta, G., Kucera, G.L. & Bierbach, U. Synthesis, biological activity, and DNA-damage profile of platinum-threading intercalator conjugates designed to target adenine. Journal of medicinal chemistry 49, 3204-3214 (2006).

160. Choudhury, J.R., Rao, L. & Bierbach, U. Rates of intercalator-driven platination of DNA determined by a restriction enzyme cleavage inhibition assay. J Biol Inorg Chem 16, 373- 80 (2011).

161. Ma, Z., Rao, L. & Bierbach, U. Replacement of a thiourea-S with an amidine-NH donor group in a platinum-acridine antitumor compound reduces the metal's reactivity with cysteine sulfur. Journal of medicinal chemistry 52, 3424-3427 (2009).

162. Choudhury, J.R. & Bierbach, U. Characterization of the bisintercalative DNA binding mode of a bifunctional platinum-acridine agent. Nucleic acids research 33, 5622-5632 (2005).

163. Bensimon, A. et al. Alignment and sensitive detection of DNA by a moving interface. Science 265, 2096-8 (1994).

164. Poot, M. et al. Analysis of mitochondrial morphology and function with novel fixable fluorescent stains. J Histochem Cytochem 44, 1363-72 (1996).

165. Bandyopadhyay, S. et al. Rewiring of genetic networks in response to DNA damage. Science 330, 1385-9 (2010).

166. Prakash, S. & Prakash, L. Nucleotide excision repair in yeast. Mutat Res 451, 13-24 (2000).

167. Bressan, D.A., Baxter, B.K. & Petrini, J.H. The Mre11-Rad50-Xrs2 protein complex facilitates homologous recombination-based double-strand break repair in Saccharomyces cerevisiae. Molecular and cellular biology 19, 7681-7687 (1999).

168. Tercero, J.A. & Diffley, J.F. Regulation of DNA replication fork progression through damaged DNA by the Mec1/Rad53 checkpoint. Nature 412, 553-557 (2001).

169. Avemann, K., Knippers, R., Koller, T. & Sogo, J.M. Camptothecin, a specific inhibitor of type I DNA topoisomerase, induces DNA breakage at replication forks. Molecular and cellular biology 8, 3026-3034 (1988).

170. Slater, M.L. Effect of reversible inhibition of deoxyribonucleic acid synthesis on the yeast cell cycle. Journal of Bacteriology 113, 263-270 (1973).

125

171. Smyre, C.L., Saluta, G., Kute, T.E., Kucera, G.L. & Bierbach, U. Inhibition of DNA Synthesis by a Platinum–Acridine Hybrid Agent Leads to Potent Cell Kill in Nonsmall Cell Lung Cancer. ACS Medicinal Chemistry Letters 2, 870-874 (2011).

172. Choudhury, J.R., Guddneppanavar, R., Saluta, G., Kucera, G.L. & Bierbach, U. Tuning the DNA conformational perturbations induced by cytotoxic platinum-acridine bisintercalators: effect of metal cis/trans isomerism and DNA threading groups. J Med Chem 51, 3069-72 (2008).

173. Forsburg, S.L. The yeasts Saccharomyces cerevisiae and Schizosaccharomyces pombe: models for cell biology research. Gravit Space Biol Bull 18, 3-9 (2005).

174. Hillenmeyer, M.E. et al. The chemical genomic portrait of yeast: uncovering a phenotype for all genes. Science 320, 362-5 (2008).

175. Ma, Z., Day, C.S. & Bierbach, U. Unexpected reactivity of the 9-aminoacridine chromophore in guanidylation reactions. J Org Chem 72, 5387-90 (2007).

176. Ma, Z., Saluta, G., Kucera, G.L. & Bierbach, U. Effect of linkage geometry on biological activity in thiourea- and guanidine-substituted acridines and platinum-acridines. Bioorganic & medicinal chemistry letters 18, 3799-3801 (2008).

177. Kim, D.U. et al. Analysis of a genome-wide set of gene deletions in the fission yeast Schizosaccharomyces pombe. Nat Biotechnol 28, 617-23 (2010).

178. Pierce, S.E., Davis, R.W., Nislow, C. & Giaever, G. Genome-wide analysis of barcoded Saccharomyces cerevisiae gene-deletion mutants in pooled cultures. Nat Protoc 2, 2958- 74 (2007).

179. Shabtai, D., Giaever, G. & Nislow, C. An algorithm for chemical genomic profiling that minimizes batch effects: Bucket Evaluations. BMC Bioinformatics, In press. (2012).

180. Subramanian, A. et al. Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc Natl Acad Sci U S A 102, 15545-50 (2005).

181. Merico, D., Isserlin, R., Stueker, O., Emili, A. & Bader, G.D. Enrichment map: a network-based method for gene-set enrichment visualization and interpretation. PLoS One 5, e13984 (2010).

182. Smoot, M.E., Ono, K., Ruscheinski, J., Wang, P.L. & Ideker, T. Cytoscape 2.8: new features for data integration and network visualization. Bioinformatics 27, 431-2 (2011).

183. van Dongen, S. A cluster algorithm for graphs. Technical Report INS-R0010. National Research Institute for Mathematics and Computer Science in the Netherlands, Amsterdam, , 1386–3681 (2000).

126

184. Yang, J., O'Donnell, L., Durocher, D. & Brown, G.W. RMI1 Promotes DNA Replication Fork Progression and Recovery from Replication Fork Stress. Mol Cell Biol (2012).

185. Kirshner, J.R. et al. Elesclomol induces cancer cell apoptosis through oxidative stress. Mol Cancer Ther 7, 2319-27 (2008).

186. Trachootham, D., Alexandre, J. & Huang, P. Targeting cancer cells by ROS-mediated mechanisms: a radical therapeutic approach? Nat Rev Drug Discov 8, 579-91 (2009).

187. Schumacker, P.T. Reactive oxygen species in cancer cells: live by the sword, die by the sword. Cancer Cell 10, 175-6 (2006).

188. Berkenblit, A. et al. Phase I clinical trial of STA-4783 in combination with paclitaxel in patients with refractory solid tumors. Clin Cancer Res 13, 584-90 (2007).

189. O'Day, S. et al. Phase II, randomized, controlled, double-blinded trial of weekly elesclomol plus paclitaxel versus paclitaxel alone for stage IV metastatic melanoma. J Clin Oncol 27, 5452-8 (2009).

190. Vukovic, V.M., Hauschild, A., Eggermont, A.M. & O'Day, S. Phase III, randomized, double-blind study of elesclomol and paclitaxel versus paclitaxel alone in stage IV metastatic melanoma (MM): 1-year OS update. . Journal of Clinical Oncology 28, 8550 (2010).

191. Nagai, M. et al. The oxidative stress inducer elesclomol requires copper chelation for its anticancer activity. Molecular Cancer Therapeutics 8, C11 (2009).

192. Wu, L., Zhou, L., Liu, D.Q., Vogt, F.G. & Kord, A.S. LC–MS/MS and density functional theory study of copper(II) and nickel(II) chelating complexes of elesclomol (a novel anticancer agent). Journal of Pharmaceutical and Biomedical Analysis 54, 331-336 (2011).

193. Fleming, J.A. et al. Complementary whole-genome technologies reveal the cellular response to proteasome inhibition by PS-341. Proc Natl Acad Sci U S A 99, 1461-6 (2002).

194. Xu, D. et al. Chemical genetic profiling and characterization of small-molecule compounds that affect the biosynthesis of unsaturated fatty acids in Candida albicans. J Biol Chem 284, 19754-64 (2009).

195. Liang, Q. & Zhou, B. Copper and manganese induce yeast apoptosis via different pathways. Mol Biol Cell 18, 4741-9 (2007).

196. Hillenmeyer, M.E. et al. Systematic analysis of genome-wide fitness data in yeast reveals novel gene function and drug action. Genome Biol 11, R30 (2010).

197. Wallace, I.M. et al. Compound prioritization methods increase rates of chemical probe discovery in model organisms. Chem Biol 18, 1273-83 (2011). 127

198. Jo, W.J. et al. Identification of genes involved in the toxic response of Saccharomyces cerevisiae against iron and copper overload by parallel analysis of deletion mutants. Toxicol Sci 101, 140-51 (2008).

199. Lenaz, G. The mitochondrial production of reactive oxygen species: mechanisms and implications in human pathology. IUBMB Life 52, 159-64 (2001).

200. Stowe, D.F. & Camara, A.K. Mitochondrial reactive oxygen species production in excitable cells: modulators of mitochondrial and cell function. Antioxid Redox Signal 11, 1373-414 (2009).

201. Kussmaul, L. & Hirst, J. The mechanism of superoxide production by NADH:ubiquinone oxidoreductase (complex I) from bovine heart mitochondria. Proc Natl Acad Sci U S A 103, 7607-12 (2006).

202. Koukourakis, M.I. et al. Lactate dehydrogenase-5 (LDH-5) overexpression in non-small- cell lung cancer tissues is linked to tumour hypoxia, angiogenic factor production and poor prognosis. Br J Cancer 89, 877-85 (2003).

203. Hersey, P., Watts, R.N., Zhang, X.D. & Hackett, J. Metabolic approaches to treatment of melanoma. Clin Cancer Res 15, 6490-4 (2009).

204. Ericson, E. et al. Off-target effects of psychoactive drugs revealed by genome-wide assays in yeast. PLoS Genet 4, e1000151 (2008).

205. Pierce, S.E., Davis, R.W., Nislow, C. & Giaever, G. Genome-wide analysis of barcoded Saccharomyces cerevisiae gene-deletion mutants in pooled cultures. Nature protocols 2, 2958-2974 (2007).

206. Proctor, M. et al. The automated cell: compound and environment screening system (ACCESS) for chemogenomic screening. Methods Mol Biol 759, 239-69 (2011).

207. Benschop, J.J. et al. A consensus of core protein complex compositions for Saccharomyces cerevisiae. Mol Cell 38, 916-28 (2010).

208. Enright, A.J., Van Dongen, S. & Ouzounis, C.A. An efficient algorithm for large-scale detection of protein families. Nucleic Acids Res 30, 1575-84 (2002).

209. Vichai, V. & Kirtikara, K. Sulforhodamine B colorimetric assay for cytotoxicity screening. Nature protocols 1, 1112-1116 (2006).

210. Hoon, S., St Onge, R.P., Giaever, G. & Nislow, C. Yeast chemical genomics and drug discovery: an update. Trends in pharmacological sciences 29, 499-504 (2008).

211. Pena-Castillo, L. & Hughes, T.R. Why are there still over 1000 uncharacterized yeast genes? Genetics 176, 7-14 (2007).

128

212. Schlabach, M.R. et al. Cancer proliferation gene discovery through functional genomics. Science (New York, N.Y.) 319, 620-624 (2008).

213. Silva, J.M. et al. Profiling essential genes in human mammary cells by multiplex RNAi screening. Science (New York, N.Y.) 319, 617-620 (2008).

214. Marcotte, R. et al. Essential gene profiles in breast, pancreatic, and ovarian cancer cells. Cancer Discov 2, 172-89 (2012).

215. Singal, P.K. & Iliskovic, N. Doxorubicin-induced cardiomyopathy. N Engl J Med 339, 900-5 (1998).

216. Tewey, K.M., Rowe, T.C., Yang, L., Halligan, B.D. & Liu, L.F. Adriamycin-induced DNA damage mediated by mammalian DNA topoisomerase II. Science 226, 466-8 (1984).

217. Scherf, U. et al. A gene expression database for the molecular pharmacology of cancer. Nat Genet 24, 236-44 (2000).

218. Gewirtz, D.A. A critical evaluation of the mechanisms of action proposed for the antitumor effects of the anthracycline antibiotics adriamycin and daunorubicin. Biochem Pharmacol 57, 727-41 (1999).

219. Baykov, A.A., Fabrichniy, I.P., Pohjanjoki, P., Zyryanov, A.B. & Lahti, R. Fluoride effects along the reaction pathway of pyrophosphatase: evidence for a second enzyme.pyrophosphate intermediate. Biochemistry 39, 11939-47 (2000).

220. Tokarska-Schlattner, M., Wallimann, T. & Schlattner, U. Multiple interference of anthracyclines with mitochondrial creatine kinases: preferential damage of the cardiac isoenzyme and its implications for drug cardiotoxicity. Molecular pharmacology 61, 516- 523 (2002).

221. Sunavala-Dossabhoy, G. & De Benedetti, A. Tousled homolog, TLK1, binds and phosphorylates Rad9; TLK1 acts as a molecular chaperone in DNA repair. DNA Repair (Amst) 8, 87-102 (2009).

222. Fornari, F. et al. MiR-122/cyclin G1 interaction modulates p53 activity and affects doxorubicin sensitivity of human hepatocarcinoma cells. Cancer research 69, 5761-5767 (2009).

223. Kimura, S.H., Ikawa, M., Ito, A., Okabe, M. & Nojima, H. Cyclin G1 is involved in G2/M arrest in response to DNA damage and in growth control after damage recovery. Oncogene 20, 3290-3300 (2001).

224. Nicoletti, F. et al. Metabotropic glutamate receptors: new targets for the control of tumor growth? Trends in pharmacological sciences 28, 206-213 (2007).

129

225. Rzeski, W., Ikonomidou, C. & Turski, L. Glutamate antagonists limit tumor growth. Biochemical pharmacology 64, 1195-1200 (2002).

226. Yoo, B.C. et al. Metabotropic glutamate receptor 4-mediated 5-Fluorouracil resistance in a human colon cancer cell line. Clinical cancer research : an official journal of the American Association for Cancer Research 10, 4176-4184 (2004).

227. Avvakumov, N. & Cote, J. The MYST family of histone acetyltransferases and their intimate links to cancer. Oncogene 26, 5395-407 (2007).

228. Rebbaa, A., Zheng, X., Chu, F. & Mirkin, B.L. The role of histone acetylation versus DNA damage in drug-induced senescence and apoptosis. Cell Death Differ 13, 1960-7 (2006).

229. Maiso, P. et al. The synergy of plus doxorubicin in acute myeloid leukemia suggests a role for HDAC inhibitors in the control of DNA repair. Leukemia 23, 2265-74 (2009).

230. Kraft, M. et al. Disruption of the histone acetyltransferase MYST4 leads to a Noonan syndrome-like phenotype and hyperactivated MAPK signaling in humans and mice. J Clin Invest 121, 3479-91 (2011).

231. Grunstein, M. Histone acetylation in chromatin structure and transcription. Nature 389, 349-52 (1997).

232. Struhl, K. Histone acetylation and transcriptional regulatory mechanisms. Genes Dev 12, 599-606 (1998).

233. Roth, S.Y., Denu, J.M. & Allis, C.D. Histone acetyltransferases. Annu Rev Biochem 70, 81-120 (2001).

234. Archer, S.Y. & Hodin, R.A. Histone acetylation and cancer. Curr Opin Genet Dev 9, 171-4 (1999).

235. Catalano, M.G. et al. Valproic acid, a histone deacetylase inhibitor, enhances sensitivity to doxorubicin in anaplastic thyroid cancer cells. J Endocrinol 191, 465-72 (2006).

236. Karukstis, K.K., Thompson, E.H., Whiles, J.A. & Rosenfeld, R.J. Deciphering the fluorescence signature of daunomycin and doxorubicin. Biophys Chem 73, 249-63 (1998).

237. The RNAi Consortium, http://www.broadinstitute.org/rnai/trc. (2012).

238. Naldini, L. et al. In vivo gene delivery and stable transduction of nondividing cells by a lentiviral vector. Science 272, 263-7 (1996).

130

239. Cheung, H.W. et al. Systematic investigation of genetic vulnerabilities across cancer cell lines reveals lineage-specific dependencies in ovarian cancer. Proc Natl Acad Sci U S A 108, 12372-7 (2011).

240. Sopko, R. et al. Mapping pathways and phenotypes by systematic gene overexpression. Molecular cell 21, 319-330 (2006).

241. Liu, H., Krizek, J. & Bretscher, A. Construction of a GAL1-regulated yeast cDNA expression library and its application to the identification of genes whose overexpression causes lethality in yeast. Genetics 132, 665-673 (1992).

242. Niu, W., Li, Z., Zhan, W., Iyer, V.R. & Marcotte, E.M. Mechanisms of cell cycle control revealed by a systematic and quantitative overexpression screen in S. cerevisiae. PLoS genetics 4, e1000120 (2008).

243. Stevenson, L.F., Kennedy, B.K. & Harlow, E. A large-scale overexpression screen in Saccharomyces cerevisiae identifies previously uncharacterized cell cycle genes. Proceedings of the National Academy of Sciences of the United States of America 98, 3946-3951 (2001).

244. Gelperin, D.M. et al. Biochemical and genetic analysis of the yeast proteome with a movable ORF collection. Genes Dev 19, 2816-26 (2005).

245. Hu, Y. et al. Approaching a complete repository of sequence-verified protein-encoding clones for Saccharomyces cerevisiae. Genome Res 17, 536-43 (2007).

246. Douglas, A.C. et al. Functional analysis with a barcoder yeast gene overexpression system. G3 (Bethesda) 2, 1279-89 (2012).

247. Jones, G.M. et al. A systematic library for comprehensive overexpression screens in Saccharomyces cerevisiae. Nat Methods 5, 239-41 (2008).

248. Tseng, A.S. & Hariharan, I.K. An overexpression screen in Drosophila for genes that restrict growth or cell-cycle progression in the developing eye. Genetics 162, 229-43 (2002).

249. Ichikawa, T. et al. The FOX hunting system: an alternative gain-of-function gene hunting technique. Plant J 48, 974-85 (2006).

250. Birchler, J.A. & Veitia, R.A. The gene balance hypothesis: from classical genetics to modern genomics. Plant Cell 19, 395-402 (2007).

251. De Vrij, F.M. et al. Mutant ubiquitin expressed in Alzheimer's disease causes neuronal death. FASEB J 15, 2680-8 (2001).

252. Epstein, P.N., Overbeek, P.A. & Means, A.R. Calmodulin-induced early-onset diabetes in transgenic mice. Cell 58, 1067-73 (1989).

131

253. Tan, E. et al. The relationship between opsin overexpression and photoreceptor degeneration. Invest Ophthalmol Vis Sci 42, 589-600 (2001).

254. Pradervand, S. et al. Affymetrix Whole-Transcript Human Gene 1.0 ST array is highly concordant with standard 3' expression arrays. Biotechniques 44, 759-62 (2008).

255. Woltjen, K. et al. piggyBac transposition reprograms fibroblasts to induced pluripotent stem cells. Nature 458, 766-70 (2009).

256. Franken, N.A., Rodermond, H.M., Stap, J., Haveman, J. & van Bree, C. Clonogenic assay of cells in vitro. Nature protocols 1, 2315-2319 (2006).

257. Majno, G. & Joris, I. Apoptosis, oncosis, and necrosis. An overview of cell death. The American journal of pathology 146, 3-15 (1995).

258. Elmore, S. Apoptosis: a review of programmed cell death. Toxicologic pathology 35, 495-516 (2007).

259. Vermes, I., Haanen, C., Steffens-Nakken, H. & Reutelingsperger, C. A novel assay for apoptosis. Flow cytometric detection of phosphatidylserine expression on early apoptotic cells using fluorescein labelled Annexin V. Journal of immunological methods 184, 39- 51 (1995).

260. Vavouri, T., Semple, J.I., Garcia-Verdugo, R. & Lehner, B. Intrinsic protein disorder and interaction promiscuity are widely associated with dosage sensitivity. Cell 138, 198-208 (2009).

261. Couthouis, J. et al. A yeast functional screen predicts new candidate ALS disease genes. Proc Natl Acad Sci U S A 108, 20881-90 (2011).

262. Rine, J., Hansen, W., Hardeman, E. & Davis, R.W. Targeted selection of recombinant clones through gene dosage effects. Proc Natl Acad Sci U S A 80, 6750-4 (1983).

263. Li, X. et al. Multicopy suppressors for novel antibacterial compounds reveal targets and drug efflux susceptibility. Chem Biol 11, 1423-30 (2004).

264. Smith, A.M., Durbic, T., Kittanakom, S., Giaever, G. & Nislow, C. Barcode sequencing for understanding drug-gene interactions. Methods Mol Biol 910, 55-69 (2012).

265. Ho, C.H. et al. A molecular barcoded yeast ORF library enables mode-of-action analysis of bioactive compounds. Nat Biotechnol 27, 369-77 (2009).

266. Xu, Q. & Reed, J.C. Bax inhibitor-1, a mammalian apoptosis suppressor identified by functional screening in yeast. Mol Cell 1, 337-46 (1998).

267. Outeiro, T.F. & Lindquist, S. Yeast cells provide insight into alpha-synuclein biology and pathobiology. Science 302, 1772-5 (2003).

132

268. Csete, M. Translational prospects for human induced pluripotent stem cells. Regen Med 5, 509-19 (2010).

269. Brass, A.L. et al. Identification of host proteins required for HIV infection through a functional genomic screen. Science (New York, N.Y.) 319, 921-926 (2008).

270. Konig, R. et al. Global analysis of host-pathogen interactions that regulate early-stage HIV-1 replication. Cell 135, 49-60 (2008).

271. Kimura, J. et al. A functional genome-wide RNAi screen identifies TAF1 as a regulator for apoptosis in response to genotoxic stress. Nucleic Acids Res 36, 5250-9 (2008).

272. Mohr, S., Bakal, C. & Perrimon, N. Genomic screening with RNAi: results and challenges. Annu Rev Biochem 79, 37-64 (2010).

273. Yeung, M.L., Houzet, L., Yedavalli, V.S. & Jeang, K.T. A genome-wide short hairpin RNA screening of jurkat T-cells for human proteins contributing to productive HIV-1 replication. J Biol Chem 284, 19463-73 (2009).

274. Carette, J.E. et al. Haploid genetic screens in human cells identify host factors used by pathogens. Science 326, 1231-5 (2009).

275. Reiling, J.H. et al. A haploid genetic screen identifies the major facilitator domain containing 2A (MFSD2A) transporter as a key mediator in the response to tunicamycin. Proc Natl Acad Sci U S A 108, 11756-65 (2011).

276. Carroll, D. Genome engineering with zinc-finger nucleases. Genetics 188, 773-82 (2011).

277. Christian, M. et al. Targeting DNA double-strand breaks with TAL effector nucleases. Genetics 186, 757-61 (2010).

133