Inhibition of Metabolic Enzymes As Differentiation Therapy in Acute Myeloid Leukemia

Inhibition of Metabolic Enzymes as Differentiation Therapy in Acute Myeloid Leukemia

The Harvard community has made this article openly available. Please share how this access benefits you. Your story matters

Citable link http://nrs.harvard.edu/urn-3:HUL.InstRepos:37945013

Terms of Use This article was downloaded from Harvard University’s DASH repository, and is made available under the terms and conditions applicable to Other Posted Material, as set forth at http:// nrs.harvard.edu/urn-3:HUL.InstRepos:dash.current.terms-of- use#LAA

Inhibition of Metabolic Enzymes as Differentiation Therapy in Acute Myeloid Leukemia

A dissertation presented

Jason Michael Law

The Department of Chemistry and Chemical Biology

in partial fulfillment of the requirements

for the degree of

Doctor of Philosophy

in the subject of

Chemistry

Harvard University

Cambridge, Massachusetts

October, 2016

Inhibition of Metabolic Enzymes as Differentiation Therapy in Acute Myeloid Leukemia

Abstract

Acute myeloid leukemia (AML) is among the deadliest of cancers: it carries a 27% 5-year survival rate and is responsible for over 10,000 deaths annually in the United States alone. Despite the clear medical need, new therapeutics effective in most AML patients have not been developed in forty years, and cytotoxic chemotherapy remains the typical treatment. Fortunately, advances in the understanding of AML biology have the potential to lead to new treatments. A hallmark of AML is the differentiation block: cancer cells are halted at an early phase of hematopoietic development in which they are programmed to proliferate rapidly. In a rare subset of AML—acute promyelocytic leukemia—a genetically targeted therapy is able to overcome the differentiation block, leading to complete remissions in over

90% of patients. In this work, I describe two projects, utilizing different strategies, aimed at the development of differentiation therapy agents effective in AML more broadly.

First, I describe how frequent mutations of the metabolic enzyme isocitrate dehydrogenase 1

(IDH1) found in AML tumors inspired me to design and run a target-based compound screen with the aim of developing a chemical probe selective for the mutant allele of IDH1. The result of this effort is BRD2879, a validated inhibitor of the IDH1-R132H mutant enzyme. The compound is active in cells and possesses a structure markedly different from other known inhibitors. Second, I describe a phenotypic, cell-based screen for AML differentiation, and the target identification and validation experiments leading to the

iii surprising discovery that inhibition of dihydroorotate dehydrogenase (DHODH)—another metabolic enzyme—reliably induces differentiation across many models of AML. I also describe my efforts to uncover a biochemical rationale for why DHODH inhibition leads to differentiation. Together, the outcomes of the two projects suggest a profound relationship between a cell’s metabolic state and its differentiation state, with the implication that modulation of a cell’s metabolism might be exploited to cause therapeutically beneficial differentiation in diseases like AML.

Table of Contents

Acknowledgements ...... viii

List of Abbreviations ...... xii

Chapter 1 – Approaches to Differentiation Therapy in Acute Myeloid Leukemia 1

1.1 – Treatment of acute myeloid leukemia is an unmet medical need ...... 2

1.2 – AML is a disease of impaired differentiation ...... 2

1.3 – Differentiation therapy is highly effective in APL ...... 3

1.4 – Targeted vs. phenotypic screening ...... 5

1.5 – Many AML tumors harbor mutations in IDH1 or IDH2 ...... 6

1.6 – Most AML tumors are driven by overexpression of HoxA9 ...... 9

1.7 – Two approaches to differentiation therapy ...... 10

1.8 – References ...... 11

Chapter 2 – Development of a Biologically Relevant Assay for IDH1 Inhibition 16

2.1 – The importance of a biologically relevant assay ...... 17

2.2 – Assay design for IDH1-R132H inhibition ...... 18

2.3 – The enzymatic assay functions most easily with Mn2+ as a cofactor ...... 20

2.4 – Screening identifies BRD5667 as an IDH1-R132H inhibitor ...... 21

2.5 – Synthesis and testing of analogs defines structure-activity relationships for BRD5667 ...... 22

2.6 – BRD5667 and its analogs work poorly in cells ...... 24

2.7 – Compound activity depends on which metal cofactor is used ...... 26

2.8 – Mg2+ is superior to Mn2+ as a cofactor in the IDH1 enzymatic assay...... 28

2.9 – All assay conditions were optimized before the final round of screening ...... 30

2.10 – Experimental methods: biology and biochemistry ...... 31

2.11 – Experimental methods: chemistry, with compound characterization ...... 34

2.12 – References ...... 43

Chapter 3 – Development of BRD2879, a Cell-Active Inhibitor of Mutant IDH1 44

3.1 – High-throughput screening identifies BRD2879 as a promising lead ...... 45

3.2 – Resynthesis of BRD2879 and exploration of structure-activity relationships ...... 48

3.3 – Validation of BRD2879 specificity through biophysical and cell-based assays ...... 52

3.4 – Prospects for the use of BRD2879 as a probe compound ...... 57

3.5 – Experimental methods: biology and biochemistry ...... 57

3.6 – Experimental methods: chemistry, with compound characterization ...... 62

3.7 – References ...... 67

Chapter 4 – Discovery of DHODH as a Target for AML Differentiation 68

4.1 – The role of phenotypic screening in development of differentiation therapy ...... 69

4.2 – Lysozyme-GFP-ER-HoxA9 cells establish a screenable model for AML differentiation ...... 70

4.3 – A high-throughput screen identifies small-molecule inducers of AML differentiation ...... 73

4.4 – Analysis of resistant cell lines identifies DHODH as the target ...... 74

4.5 – Dihydroorotate dehydrogenase is a key metabolic enzyme ...... 76

4.6 – Targeting of DHODH confirmed by in vitro enzyme inhibition assay ...... 77

4.7 – Medicinal chemistry leads to ML390, a more potent differentiating agent ...... 78

4.8 – X-ray crystallography of DHODH with ML390 defines binding ...... 82

4.9 – Brequinar, another DHODH inhibitor, is suitable for in vivo studies ...... 85

4.10 – Brequinar demonstrates anti-leukemia activity and differentiation in vivo ...... 86

4.11 – Differentiation is caused by lack of pyrimidine metabolites ...... 88

4.12 – The mechanism of myeloid differentiation in response to uridine deprivation ...... 90

4.13 – The differentiation effects of DHODH inhibitors are stronger than those of standard chemotherapy ...... 95

4.14 – The history of DHODH as a therapeutic target ...... 98

4.15 – The potential of brequinar as differentiation therapy in AML ...... 99

4.16 – Experimental methods ...... 101

4.17 – References ...... 102

Chapter 5 – Lessons on Probe Development and the Link Between Metabolism and Differentiation in

AML 107

5.1 – Unifying themes and lessons learned ...... 108

5.2 – A metabolism—differentiation link was discovered through two approaches ...... 108

5.3 – The advantage of chemical biology ...... 109

5.4 – Improved probes were found through literature search ...... 110

5.5 – Assay conditions are key to the identification of high-quality probes ...... 111

5.6 – High-impact innovation requires speed—or a unique approach ...... 114

5.7 – Reference ...... 115

vii

Acknowledgements

Having arrived at Harvard with limited knowledge of the ways of the laboratory, I owe a debt of gratitude to those who took time to teach me the skills, technical or otherwise, necessary to complete my

work. As my projects expanded to require different forms of expertise, there were always individuals

happy to show me the way. In teaching me the art of chemical synthesis, I would like to especially thank

Dr. Mahmud Hussain, Dr. Zarko Boskovic, and Dr. Max Majireck, the trio of experienced chemistry

postdocs whose advice was critical to getting started in the lab. Although my work did not always involve

chemical synthesis, I returned periodically to the fume hood throughout my time in the Schreiber group

and benefited from additional assistance and advice from Dr. Matthias Leiendecker, Dr. Leslie Aldrich,

Dr. Marshall Morningstar, Micah Maetani, and Shawn D. Nelson, Jr.

When my work moved to biochemistry, cell culture, and enzymology, I started knowing almost

nothing and was dependent of Dr. Ke Liu to explain techniques, help me troubleshoot, and tell me which

textbooks books to read. Ke probably did more than any other single individual to help me develop my

technical skills, for which I am grateful. I will also never forget the universally applicable advice given to

me by Dr. Daisuke Ito: “step by step.”

Of course, my advisor was key to my experience grad school. Prof. Stuart Schreiber is always

inspirational, never pressuring, and never said “no.” I know the freedom I was given to pursue my interests

to the extent I wanted and without worrying about how to pay for it all is unusual for a graduate student

and I was lucky to have such unconditional support. Stuart was especially helpful in providing nudges or

suggestions, especially in connecting me to people who could help me or provide an opportunity to work

on something new and exciting. I’m certainly glad Stuart pulled me aside in the hallway in June 2014 and

said, “Jason, you really should talk to David Sykes.” The resulting collaboration introduced me to new

viii friends and provided an opportunity to use my newly honed chemistry and enzymology skills on my

highest-impact work in graduate school.

Speaking of which, Dr. David Sykes became like a second advisor to me late in grad school, teaching me myeloid cell biology, flow cytometry, and the art of cheerfully moving forward with a large, collaborative project in spite of all technical and bureaucratic obstacles. Another key advisor was Dr. Aly

Shamji, who guided me through the IDH project on a week-to-week basis, never ran out of ideas for me to try, and never got upset when I only had time to complete one in three experiments he suggested.

Biochemistry guru Dr. Andy Stern provided essential advice on assay development.

I’d like to thank the rest of my committee for providing additional perspective. Prof. Dan Kahne asked me the big questions, “Are you sure this is the best thing you can be doing with your life?” that so often get forgotten when deep in the details of a project, and Prof. Greg Verdine reminded me there is more to therapeutics than small molecules, and that there are many ways to do good science in both the academic and industrial spheres. Outside of my official committee, I received valuable advice from Prof.

David Scadden and Prof. Eric Jacobsen.

I was fortunate to work with several great mentees during my time at the Broad. First, I have to thank my undergraduate trainee Norah Liang, for bearing with me when I unthinkingly told her adjust the pH of a Tris solution to 6.5. Norah was an asset to the IDH team, working on the assay development, screening follow-up, and medicinal chemistry around BRD2879. Later, I had the privilege of working with visiting graduate students Dr. Oscar Verho and Sebastian Stark. Oscar’s help was invaluable when I would otherwise have run a high-throughput screen by myself, and without Sebastian, the amount of medicinal chemistry around BRD2879 would have been pitifully small.

Of course, I can’t forget my undergraduate advisor Prof. Paul Wender at Stanford. His lectures and our subsequent discussions convinced me of the power of chemistry to drive innovation in human

ix health and in many ways set the course of my career. The chemistry skills I was taught while working in the Wender group by Dr. Brian Trantow and Dr. Nate Cardin have served me well ever since. I would also like to thank Prof. Steve Boxer, whose Chem 185 literature survey course was the first place I felt like I could be a real scientist rather than just a student. In high school, Dr. Joan D’Agostino, Dr. Louis Leithold,

Brian Corrigan, and Bob Perry were key in encouraging my love for science.

On a lab operations level, I want to thank Pat Mark for keeping the lab running and addressing every equipment issue that ever threatened our work, even if it meant returning to lab in the middle of the night. Cindy Hon made sure I always had the funding I needed, especially when the project’s main grant ran out and we had to get creative. Dr. Stephen Johnston was always willing to help accommodate my unusual analytical chemistry needs. Dr. Yan-Ling Zhang guided me through the high-throughput screening process. Dr. Joshiawa Paulk provided a mix of commiseration and inspiration late in the evening. Dr. Jamie Cheah organized regular Dim Sum outings, and Jamie and later Dr. Amedeo Vetere maintained some sort of order in the tissue culture room when anarchy threatened.

On a personal level, the Schreiber Lab has always been a friendly, collaborative, low-stress environment, and I would like to thank everyone in the group for maintaining that atmosphere. There are too many to thank individually, but I am grateful to everyone involved in the running club, the whitewater outing, the trivia nights, the CSofT socials, or just chatting in the lunch room. It has been great hanging out with you for these years.

Outside of lab, there was ASAASA and its irrepressible leader Dr. Noam Prywes, introducing me to new friends and new topics, even if it meant I occasionally spent way too much time reading about homing pigeons or naked mole rats at the expense of my thesis work. Playing squash with Noam, Dr.

Anders Hansen, and Dr. Ryan Babbush helped me keep my sanity, as did regular coffee breaks with Alan

Ransil.

I would like to thank my family for putting up with my excursion the other side of the continent.

My parents Russ and Denise have given me their love, support and advice on every step of this journey.

I’m grateful that Angela and I have been able to stick together no matter how many airplane flights it took to make it happen.

Finally, the projects described herein are the work of many people, and I wouldn’t have come close to finishing projects of this scope without their help. The focus of this dissertation is necessarily on the work I performed personally, but to tell a coherent scientific narrative, I often include results of my colleagues’ experiments when they relate directly to my work. Chapters begin with acknowledgement of colleagues who helped with or performed experiments I describe, as well as references to scientific papers that tell the more complete story. I would like to thank my advisor for our discussion on managing the tricky task of writing a personal dissertation about a collaborative project without taking undue credit for results or leaving out key parts of the story.

List of Abbreviations

α-KG ...... alpha-ketoglutarate Ac-5S-GlcNAc ...... 2-acetamido-1,3,4,6-tetra-O-acetyl-2-deoxy-5-thio-α-D-glucopyranose AML ...... Acute myeloid leukemia APL ...... Acute promyelocytic leukemia ATP ...... Adenosine triphosphate ATRA ...... All-trans retinoid acid BRQ ...... Brequinar sodium CMP ...... Common myeloid progenitor DCIP ...... 2,6-dichlorophenolindophenol DCM ...... Dichloromethane DHODH ...... Dihydroorotate dehydrogenase DMSO ...... Dimethyl sulfoxide DOS ...... Diversity-oriented synthesis E2 ...... 17-β-estradiol ER ...... Estrogen receptor ER-HoxA9 ...... Estrogen receptor—HoxA9 fusion protein FACS ...... Fluorescence-activated cell sorting FMN(H2) ...... Flavin mononucleotide (reduced) GFP ...... Green fluorescent protein GMP ...... Granulocyte-macrophage progenitor GPCR ...... G-protein-coupled receptor HPLC ...... High pressure liquid chromatography HRMS ...... High resolution mass spectrometry HTS ...... High throughput screening IDH1/2 ...... Isocitrate dehydrogenase 1/2 IP ...... Intraperitoneal KDM4C ...... Lysine demethylase 4C LC-MS ...... Liquid chromatography – mass spectrometry MTD ...... Maximum tolerated dose NADP(+/H) ...... Nicotinamide adenine dinucleotide 2’-phosphate (oxidized/reduced) NSAID ...... Non-steroidal anti-inflammatory drug OGT ...... O-GlcNAc transferase PMB ...... Para-methoxybenzyl PML ...... Promyelocytic leukemia (can refer to disease, gene or protein) PML/RARα ...... Promyelocytic leukemia/retinoic acid receptor alpha fusion protein Q[N]D ...... Every [N] days (dosing schedule) R-2HG ...... (R)-2-hydroxyglutarate RNA-seq ...... RNA sequencing SAR ...... Structure-activity relationship SD ...... Standard deviation THF ...... Tetrahydrofuran TLC ...... Thin layer chromatography UMP ...... Uridine 5’-monophosphate UMPS ...... Uridine 5’-monophosphate synthase WES ...... Whole-exome sequencing

xii

Chapter 1

Approaches to Differentiation Therapy in Acute Myeloid Leukemia

Portions of this chapter are reproduced with permission from the following publications:

Acute myeloid leukemia (AML) is a form of cancer characterized by the sudden extreme overproduction of immature myeloid leukocytes, leading to catastrophic failure of the hematopoietic system. The disease is most common in individuals over 55 years of age and carries a 5-year survival rate of 26.6 %, leading to an estimated 10,430 deaths annually in the United States.3 The standard of care for

90 % of AML cases is a form of cytotoxic chemotherapy in use for the past forty years: combination

treatment with the nucleoside analog cytarabine and a topoisomerase inhibitor such as daunorubicin.

Such treatment prevents successful DNA replication and is toxic to all rapidly proliferating cells in addition

to the tumor itself. The broad, systemic effects of these drugs lead to such extreme toxicity that it is usually

not possible kill all cancerous cells in this manner without also killing the patient.4

The 10 % of patients with a type of AML called acute promyelocytic leukemia (APL) undergo a different type of treatment and their clinical outlook is much brighter. These patients can take advantage of differentiation therapy, a pharmaceutical intervention designed not to kill cancer cells directly but to

reprogram them gently to a quiescent state. The result is a treatment that is better tolerated and far more

effective than cytotoxic chemotherapy. Before the advent of differentiation therapy, APL was the most

deadly form of AML, carrying a 5-year survival of 10 %. Now, APL is the most treatable form of AML, with

5-year survival of 85 %.5 The development of differentiation therapy agents effective in the 90 % of AML patients who currently have no access to this type of therapy would be transformative in the treatment of this disease.

1.2 AML is a disease of impaired differentiation

The hematopoietic system is home to a great degree of cell proliferation: on the order of 20,000

hematopoietic stem cells6 are responsible for seeding, among other cell types, the daily production of 50

billion neutrophils through a process of asymmetric division and differentiation.7 Along the way, cells pass

2 through the common myeloid progenitor (CMP) phase and the granulocyte monocyte progenitor (GMP) phase, both of which are characterized by multipotency and the ability to proliferate rapidly (Figure 1.1).

The tremendous proliferative potential of the hematopoietic system allows for a rapid immune response tailored to a specific pathogen. With such a high rate of hematopoietic cell production, most terminally differentiated hematopoietic cells must be programmed to commit apoptosis after a period of days to make room for their freshly differentiated brethren.

Figure 1.1: Selected major cell populations on the path from hematopoietic stem cells to neutrophils, the most numerous of the white blood cells.

AML occurs when immature hematopoietic cells become leukemic through a block to the differentiation process and become stuck in a phase of development resembling the granulocyte/monocyte progenitor stage.8 These cells are programmed to grow and proliferate rapidly and

not to commit apoptosis, therefore the differentiation block causes an explosive increase in leukemic cells

at the expense of other components of blood. It is the critical lack of functional blood cells which causes

rapid death in AML patients.

1.3 Differentiation therapy is highly effective in APL

By 1980, it was clear that AML was a disease of undifferentiated cells, and researchers began the first

experiments searching for compounds which could serve as differentiation therapy. In the days before

3 high-throughput screening, this work involved picking specific compounds hypothesized to effect differentiation and testing them one-by-one through functional cell-based assays. Among compounds tested in these experiments were the retinoids, a compound class that had been subjected to decades of study because it includes vitamin A. Treatment of the APL-derived cell line HL-60 with the retinoid all- trans retinoic acid (ATRA) produced spectacular results, with a 1 µM dose causing 90% of cells to differentiate with 6 days of continuous exposure.9 In most cases, PML is genetically driven by a PML/RARA

fusion oncogene, whose protein product PML/RARα homodimerizes, binds to DNA, and recruits

chromatin-modifying enzymes which cause global transcriptional repression, thus preventing granulocytic

differentiation. ATRA, as an endogenous ligand of RARα, is able to b ind the PML-RARα fusion protein,

disrupting its oncogenic activity and targeting it for degradation.10 While treatment of ATRA was able to reliably induce remission in PML, remissions were not durable. Treatment of APL was further improved by combination treatment of ATRA with cytotoxic chemotherapy agents, especially anthracyclines.

Work in the 1990s led to the discovery of another differentiation therapy for PML, arsenic trioxide.

The PML portion of the PML/RARα fusion protein contains adjacent cysteine residues in ideal configuration for the chelation of arsenic. Binding of arsenic to PML/RARα leads to oxidation of cysteine residues into disulfide bonds; these crosslink PML-RARα into an oligomerized form that is targeted for degradation. The activation of wild-type PML protein by arsenic trioxide also appears important for the activity of the drug. The latest combination treatment regimens of PML with ATRA and arsenic trioxide may prove curative in over 90% of patients.11

ATRA differentiation therapy is only effective in PML because the mechanism of action involves the

PML-specific PML/RARA oncogene. While the effective treatment of PML with ATRA and arsenic trioxide proves that myeloid leukemias can in principle be treated by differentiation therapy as part of a combination regimen, realization of differentiation therapy in AML more broadly will require the discovery of new targets.

1.4 Targeted vs. Phenotypic Screening

Broadly speaking, there are two strategies for discovering a candidate therapeutic target. The first

strategy is target-based compound discovery, in which biochemical or genetic evidence is used to form a hypothesis that a particular modulation of a specific biomolecule will prove beneficial. Among the most robust evidence for a therapeutic hypothesis is the discovery of risk and protective alleles of the same gene in humans through genome-wide association studies. This type of study was used to validate the cardiovascular target PCSK9 and the Alzheimer’s disease-associated APOE. In cancer, sequencing tumor

cells and comparing their genome to that of untransformed cells from the same patient—or in simple

cases even comparing karyotypes of these cells—can lead to the discovery of mutated oncogenes driving

proliferation of tumor cells. Once the target of interest has been defined, an assay can be set up to detect

modulation of the target in a manner amenable to high-throughput screening. Often, this assay entails

measuring the activity of purified, recombinant protein in microplates.

The second method for target discovery, phenotypic screening, is useful when well validated targets

are not available for the disease of interest—or when those targets will not readily bind small molecules—

but it is possible to develop an assay reading out the disease phenotype in a manner amenable to high-

throughput screening. The disease model, usually cell-based, is exposed to members of a compound

library and readout of disease phenotype can occur by a variety of methods, including imaging,

immunofluorescence, and luciferase-mediated luminescence. Phenotypic screening was effective in

discovering many of the early antibiotics, because the ability of chemicals to kill bacteria on a dish is both

predictive of their therapeutic efficacy and easy to measure. The method took longer to catch on for other

therapeutic classes due to the lack of suitable models, but advances in cell culture and biological

engineering have made phenotypic screening feasible for an increasing number of disease types (Figure

1.2).

Strategy Advantages Disadvantages Target- HTS-compatible assay development is The proper target for a given disease is based usually feasible often not apparent Phenotypic Ability to uncover unknown, druggable Assay development can be difficult, targets especially for complex diseases. Target ID can be difficult. Figure 1.2: Key advantages and disadvantages of target-based and phenotypic screening

In this work, I will present two therapeutic-discovery projects aimed at realizing differentiation

therapy in AML: one a target-based screen for inhibitors of mutant IDH1, and the other a phenotypic

screen taking advantage of a newly developed cell-based model of myeloid differentiation. An

understanding of the molecular drivers of AML was essential in the conception of both projects.

1.5 Many AML tumors harbor mutations in IDH1 or IDH2

The idea that the mutant alleles of IDH1 and IDH2 may be targets for differentiation therapy arises

from systematic efforts to characterize the genomes of patient tumors, which are revealing the genomic

alterations that cause and maintain different cancers. Somatic mutations in the genes encoding the

isocitrate dehydrogenases IDH1 and IDH2 have been found in >70% of grade II-III gliomas and secondary

12,13 14,15 glioblastomas, ~17% of acute myeloid leukemias (AML), ~56% of central and periosteal

chondrosarcoma,16 and sporadically in other tumor types. Mutations are nearly always heterozygous and occur frequently at codons IDH1-R132, IDH2-R172, or IDH2-R140. IDH enzymes normally catalyze the

interconversion of isocitrate and α-ketoglutarate (α-KG), but these mutations unmask an otherwise cryptic ketoreductase activity, allowing the enzyme to reduce α-KG to (R)-2-hydroxyglutarate (R-2HG).14,17

The discovery of an apparently neomorphic enzyme activity raises the question of how a single amino

acid substitution can possibly cause an enzyme to catalyze a whole different reaction. The answer is that

wild-type IDH1 indeed has the ability to reduce α-KG to R-2HG, but two caveats render this theoretical activity mostly moot in biological context (Figure 1.3).

A IDH1 WT - typical conditions NH R132 NH R132

H2N NH2 H2N NH2 NADPH NADP+ O O O O O O CO2 O O O O O O O O O O O O OH α -ketoglutarate oxalosuccinate isocitrate

B IDH1 WT - low [isocitrate], low [CO2] C Effects of R132H Mutation IDH1-R132H - typical conditions NADPH NADP+ O O O O Reduced affinity for isocitrate: Less product inhibition O O O O O OH α Increased affinity for NADPH: -ketoglutarate 2-hydroxyglutarate Reduction before carboxylation

Figure 1.3: The IDH1-R132H mutation unmasks an otherwise cryptic ketoreductase activity. (A) Wild- type IDH1 interconverts α-KG and isocitrate through an oxalosuccinate intermediate. R132 is responsible for organizing the enzyme active site and directly hydrogen bonds with the isocitrate substrate. (B) In artificial conditions of low isocitrate and low CO2, wild-type IDH1 can instead interconvert α-KG and R-2HG. (C) The R132H mutation reduces affinity for isocitrate and increases affinity for NADPH. These changes in substrate binding favor the R-2HG reaction pathway by reducing product inhibition by isocitrate and by promoting reduction before carboxylation.

First, IDH1 normally runs in the oxidative direction given typical cellular concentrations of isocitrate.

Second, the enzyme’s low affinity for NADPH, compared to its affinity for CO2, means that even when isocitrate concentrations are low and the enzyme runs in the reductive direction, carboxylation usually occurs before reduction, leaving isocitrate as the major product and R-2HG a minor byproduct. The R132H mutation simultaneously removes both impediments to IDH1-catalyzed formation of R-2HG. First, the loss of arginine-isocitrate hydrogen bonds means affinity for isocitrate is reduced 1000-fold, allowing the reaction to proceed reductively in the absence of product inhibition.18 Second, the mutant enzyme

contains a rearranged active site which binds NADPH extremely tightly, leading reduction to occur before

carboxylation has a chance to proceed.19 As a result, IDH1-R132H reduces α-KG to R-2HG under normal

7 physiological conditions, leading to R-2HG levels which are elevated >50-fold in samples from patients with IDH mutations.14

The pathogenesis of IDH-mutant tumors is thought to center on the ability of R-2HG to act as an

‘oncometabolite’. Due to its structural similarity to α-KG, R-2HG competitively inhibits several α-KG- dependent dioxygenases when present at the high concentrations observed in IDH-mutant tumors. In particular, R-2HG impairs DNA demethylation through inhibition of TET2,20 impairs histone demethylation

through inhibition of various lysine demethylases,21,22 and modulates hypoxic stress response through

activation of EGLN1.23 These molecular changes are thought to cause the enhanced proliferation and

impaired differentiation observed in IDH-mutant tumors. The mutual exclusivity of IDH and TET2

mutations in AML tumors20 and the ability of exogenous R-2HG to induce leukemogenesis in blood cells24

further implicate R-2HG as critical mediator in how mutant IDH contributes to AML. In IDH1-mutant glioma models, hypermethylation of CTCF binding sites has been shown to cause genomic insulator dysfunction leading to aberrant activation of PDGFRA, providing an explanation for how R-2HG impacts cancer initiation in the brain.25

The discovery of mutant-allele-selective small-molecule inhibitors of IDH1 and IDH2 has enabled the validation of these enzymes as therapeutic targets (Figure 1.4). In particular, studies of AGI-5198, AGI-

6780 and GSK321 in disease models indicate that these inhibitors may shift cancer cells towards a more mature and less proliferative state, suggesting such compounds might form the basis of a future differentiation therapy.26-29 Early results from clinical trials of mutant-IDH inhibitors AG-12030 and AG-22131

in IDH-mutant AML suggest that the compounds may be effective in this context.

Figure 1.4: Structures of published IDH1 inhibitors. Notes: (a) IDH1 inhibitor. (b) IDH2 inhibitor. (c) Clinical compound.

While the clinical progress of existing IDH inhibitors is encouraging, the fact that cancers frequently develop resistant alleles to targeted therapy motivates the need to develop structurally and mechanistically diverse ‘next-in-class’ compounds targeting IDH. The co-crystallization of mutant IDH1 with probes such as VVS and SYC-435 has already uncovered distinct binding modes for inhibitors of mutant IDH1.29,32-34 Despite this progress in the field, the discovery of additional molecules with novel

properties would prove useful in the study of diseases associated with mutant IDH1.

1.6 Most AML tumors are driven by overexpression of HoxA9

Phenotypic screening for compounds that induce myeloid differentiation in AML requires a cell-

based model faithful to the in vivo biochemical state of the leukemic cells. Specifically, a cell model system

is required which possesses a differentiation block of the same kind as that found naturally in leukemia.

As AML can arise due to a wide variety of discrete genetic lesions, there is no obvious strategy to produce

a screenable model which would be predictive across all cases of AML.

Led by my collaborators at Massachusetts General Hospital, we reasoned that diverse mutagenic events that affect differentiation or self-renewal may funnel through common molecular pathways. We sought to define and target pathways of differentiation that might be shared across a range of genetic subtypes of AML, and were intrigued by the observation that the expression of the homeobox transcription factor HoxA9 is upregulated in 70% of patients with AML,35 likely reflecting that the leukemic

blasts are halted at a common stage of differentiation arrest. HoxA9 is critical to normal myelopoiesis,

and its expression must be downregulated to permit normal differentiation.36 Furthermore, HoxA9 is

essential to the maintenance of leukemias driven by MLL translocations such as MLL/AF9,37 HoxA9 is

upregulated during the transition in chronic myeloid leukemia patients to blast-phase,38 and HoxA9

expression itself is an independent risk factor in children with leukemia.39 Therefore, though it is not found to be mutated, we reasoned that the persistent expression of HoxA9 might represent a commonly dysregulated node in AML suitable for therapeutic targeting across a range of disparate subtypes.

There are no known small-molecule inhibitors of HoxA9, and as a transcription factor without known

small-molecule binding sites, the discovery of a small-molecule inhibitor through a target-based screen of

HoxA9 would have a low probability of success. While HoxA9 is not a promising target for target-based

screening, it is promising as a basis for designing a model system for phenotypic screening given its

importance across a range of AML tumors. Thus, my colleagues developed a cellular model of HoxA9-

enforced myeloid differentiation arrest that could be used in an unbiased phenotypic screen.

1.7 Two approaches to differentiation therapy

In this work, I will describe two complementary approaches to the development of differentiation

therapy agents, one a target-based screen of IDH1-R132H and the other a phenotypic screen of cells with

a HoxA9-induced differentiation block. The projects unfolded with surprising symmetry, with studies

inspired by mutations in the metabolic enzyme IDH1 leading to enzyme inhibitors that cause myeloid

10 differentiation, and the phenotypic screen for differentiation agents uncovering molecules that acted through inhibition of dihydroorotate dehydrogenase (DHODH), another metabolic enzyme. Chapter 2 describes assay development for the targeted screen against IDH1-R132H, and Chapter 3 describes discovery of and studies on the IDH1-R132H inhibitor BRD2879. Chapter 4 describes the phenotypic screen for AML differentiation agents leading to discovery of DHODH inhibition as a strategy for inducing differentiation, with a focus on my contributions in identifying the target of screening hits and determining their mechanism of action. In Chapter 5, I describe insights arising from the research described herein, especially surrounding the connections between the metabolic and differentiation states of the leukemia cell.

1.8 References

(1) Law, J. M.; Stark, S. C.; Liu, K.; Liang, N. E.; Hussain, M. M.; Leiendecker, M.; Ito, D.; Verho, O.; Stern, A. M.; Johnston, S. E.; Zhang, Y. L.; Dunn, G. P.; Shamji, A. F.; Schreiber, S. L. "Discovery of 8- Membered Ring Sulfonamides as Inhibitors of Oncogenic Mutant Isocitrate Dehydrogenase 1." ACS Med Chem Lett 2016, 7, 944. DOI: 10.1021/acsmedchemlett.6b00264

(2) Sykes, D. B.; Kfoury, Y. S.; Mercier, F. E.; Wawer, M. J.; Law, J. M.; Haynes, M. K.; Lewis, T. A.; Schajnovitz, A.; Jain, E.; Lee, D.; Meyer, H.; Pierce, K. A.; Tolliday, N. J.; Waller, A.; Ferrara, S. J.; Eheim, A. L.; Stoeckigt, D.; Maxcy, K. L.; Cobert, J. M.; Bachand, J.; Szekely, B. A.; Mukherjee, S.; Sklar, L. A.; Kotz, J. D.; Clish, C. B.; Sadreyev, R. I.; Clemons, P. A.; Janzer, A.; Schreiber, S. L.; Scadden, D. T. "Inhibition of Dihydroorotate Dehydrogenase Overcomes Differentiation Blockade in Acute Myeloid Leukemia." Cell 2016, 167, 171. DOI: 10.1016/j.cell.2016.08.057

(3) "SEER Stat Fact Sheets: Acute Myeloid Leukemia (AML)," National Cancer Institute 2016 http://seer.cancer.gov/statfacts/html/amyl.html

(4) Fernandez, H. F. "New trends in the standard of care for initial therapy of acute myeloid leukemia." Hematology Am Soc Hematol Educ Program 2010, 2010, 56. DOI: 10.1182/asheducation- 2010.1.56

(5) Lo-Coco, F.; Avvisati, G.; Vignetti, M.; Thiede, C.; Orlando, S. M.; Iacobelli, S.; Ferrara, F.; Fazi, P.; Cicconi, L.; Di Bona, E.; Specchia, G.; Sica, S.; Divona, M.; Levis, A.; Fiedler, W.; Cerqui, E.; Breccia, M.; Fioritoni, G.; Salih, H. R.; Cazzola, M.; Melillo, L.; Carella, A. M.; Brandts, C. H.; Morra, E.; von Lilienfeld- Toal, M.; Hertenstein, B.; Wattad, M.; Lubbert, M.; Hanel, M.; Schmitz, N.; Link, H.; Kropp, M. G.; Rambaldi, A.; La Nasa, G.; Luppi, M.; Ciceri, F.; Finizio, O.; Venditti, A.; Fabbiano, F.; Dohner, K.; Sauer, M.; Ganser,

A.; Amadori, S.; Mandelli, F.; Dohner, H.; Ehninger, G.; Schlenk, R. F.; Platzbecker, U.; Gruppo Italiano Malattie Ematologiche, d. A.; German-Austrian Acute Myeloid Leukemia Study, G.; Study Alliance, L. "Retinoic acid and arsenic trioxide for acute promyelocytic leukemia." N Engl J Med 2013, 369, 111. DOI: 10.1056/NEJMoa1300874

(6) Abkowitz, J. L.; Catlin, S. N.; McCallie, M. T.; Guttorp, P. "Evidence that the number of hematopoietic stem cells per animal is conserved in mammals." Blood 2002, 100, 2665.

(7) Dancey, J. T.; Deubelbeiss, K. A.; Harker, L. A.; Finch, C. A. "Neutrophil kinetics in man." J Clin Invest 1976, 58, 705. DOI: 10.1172/JCI108517

(8) Jung, N.; Dai, B.; Gentles, A. J.; Majeti, R.; Feinberg, A. P. "An LSC epigenetic signature is largely mutation independent and implicates the HOXA cluster in AML pathogenesis." Nat Commun 2015, 6, 8489. DOI: 10.1038/ncomms9489

(9) Breitman, T. R.; Selonick, S. E.; Collins, S. J. "Induction of differentiation of the human promyelocytic leukemia cell line (HL-60) by retinoic acid." Proc Natl Acad Sci U S A 1980, 77, 2936.

(10) Nowak, D.; Stewart, D.; Koeffler, H. P. "Differentiation therapy of leukemia: 3 decades of development." Blood 2009, 113, 3655. DOI: 10.1182/blood-2009-01-198911

(11) Lallemand-Breitenbach, V.; Zhu, J.; Chen, Z.; de The, H. "Curing APL through PML/RARA degradation by As2O3." Trends Mol Med 2012, 18, 36. DOI: 10.1016/j.molmed.2011.10.001

(12) Parsons, D. W.; Jones, S.; Zhang, X.; Lin, J. C.; Leary, R. J.; Angenendt, P.; Mankoo, P.; Carter, H.; Siu, I. M.; Gallia, G. L.; Olivi, A.; McLendon, R.; Rasheed, B. A.; Keir, S.; Nikolskaya, T.; Nikolsky, Y.; Busam, D. A.; Tekleab, H.; Diaz, L. A., Jr.; Hartigan, J.; Smith, D. R.; Strausberg, R. L.; Marie, S. K.; Shinjo, S. M.; Yan, H.; Riggins, G. J.; Bigner, D. D.; Karchin, R.; Papadopoulos, N.; Parmigiani, G.; Vogelstein, B.; Velculescu, V. E.; Kinzler, K. W. "An integrated genomic analysis of human glioblastoma multiforme." Science 2008, 321, 1807. DOI: 10.1126/science.1164382

(13) Yan, H.; Parsons, D. W.; Jin, G.; McLendon, R.; Rasheed, B. A.; Yuan, W.; Kos, I.; Batinic- Haberle, I.; Jones, S.; Riggins, G. J.; Friedman, H.; Friedman, A.; Reardon, D.; Herndon, J.; Kinzler, K. W.; Velculescu, V. E.; Vogelstein, B.; Bigner, D. D. "IDH1 and IDH2 mutations in gliomas." N Engl J Med 2009, 360, 765. DOI: 10.1056/NEJMoa0808710

(14) Ward, P. S.; Patel, J.; Wise, D. R.; Abdel-Wahab, O.; Bennett, B. D.; Coller, H. A.; Cross, J. R.; Fantin, V. R.; Hedvat, C. V.; Perl, A. E.; Rabinowitz, J. D.; Carroll, M.; Su, S. M.; Sharp, K. A.; Levine, R. L.; Thompson, C. B. "The common feature of leukemia-associated IDH1 and IDH2 mutations is a neomorphic enzyme activity converting alpha-ketoglutarate to 2-hydroxyglutarate." Cancer Cell 2010, 17, 225. DOI: 10.1016/j.ccr.2010.01.020

(15) Abbas, S.; Lugthart, S.; Kavelaars, F. G.; Schelen, A.; Koenders, J. E.; Zeilemaker, A.; van Putten, W. J. L.; Rijneveld, A. W.; Löwenberg, B.; Valk, P. J. M. "Acquired mutations in the genes encoding IDH1 and IDH2 both are recurrent aberrations in acute myeloid leukemia: prevalence and prognostic value." Blood 2010, 116, 2122.

(16) Amary, M. F.; Bacsi, K.; Maggiani, F.; Damato, S.; Halai, D.; Berisha, F.; Pollock, R.; O'Donnell, P.; Grigoriadis, A.; Diss, T.; Eskandarpour, M.; Presneau, N.; Hogendoorn, P. C.; Futreal, A.; Tirabosco, R.; Flanagan, A. M. "IDH1 and IDH2 mutations are frequent events in central chondrosarcoma and central and periosteal chondromas but not in other mesenchymal tumours." J Pathol 2011, 224, 334. DOI: 10.1002/path.2913

(17) Dang, L.; White, D. W.; Gross, S.; Bennett, B. D.; Bittinger, M. A.; Driggers, E. M.; Fantin, V. R.; Jang, H. G.; Jin, S.; Keenan, M. C.; Marks, K. M.; Prins, R. M.; Ward, P. S.; Yen, K. E.; Liau, L. M.; Rabinowitz, J. D.; Cantley, L. C.; Thompson, C. B.; Vander Heiden, M. G.; Su, S. M. "Cancer-associated IDH1 mutations produce 2-hydroxyglutarate." Nature 2009, 462, 739. DOI: 10.1038/nature08617

(18) Pietrak, B.; Zhao, H.; Qi, H.; Quinn, C.; Gao, E.; Boyer, J. G.; Concha, N.; Brown, K.; Duraiswami, C.; Wooster, R.; Sweitzer, S.; Schwartz, B. "A tale of two subunits: how the neomorphic R132H IDH1 mutation enhances production of alphaHG." Biochemistry 2011, 50, 4804. DOI: 10.1021/bi200499m

(19) Rendina, A. R.; Pietrak, B.; Smallwood, A.; Zhao, H.; Qi, H.; Quinn, C.; Adams, N. D.; Concha, N.; Duraiswami, C.; Thrall, S. H.; Sweitzer, S.; Schwartz, B. "Mutant IDH1 enhances the production of 2- hydroxyglutarate due to its kinetic mechanism." Biochemistry 2013, 52, 4563. DOI: 10.1021/bi400514k

(20) Figueroa, M. E.; Abdel-Wahab, O.; Lu, C.; Ward, P. S.; Patel, J.; Shih, A.; Li, Y.; Bhagwat, N.; Vasanthakumar, A.; Fernandez, H. F.; Tallman, M. S.; Sun, Z.; Wolniak, K.; Peeters, J. K.; Liu, W.; Choe, S. E.; Fantin, V. R.; Paietta, E.; Lowenberg, B.; Licht, J. D.; Godley, L. A.; Delwel, R.; Valk, P. J.; Thompson, C. B.; Levine, R. L.; Melnick, A. "Leukemic IDH1 and IDH2 mutations result in a hypermethylation phenotype, disrupt TET2 function, and impair hematopoietic differentiation." Cancer Cell 2010, 18, 553. DOI: 10.1016/j.ccr.2010.11.015

(21) Chowdhury, R.; Yeoh, K. K.; Tian, Y. M.; Hillringhaus, L.; Bagg, E. A.; Rose, N. R.; Leung, I. K.; Li, X. S.; Woon, E. C.; Yang, M.; McDonough, M. A.; King, O. N.; Clifton, I. J.; Klose, R. J.; Claridge, T. D.; Ratcliffe, P. J.; Schofield, C. J.; Kawamura, A. "The oncometabolite 2-hydroxyglutarate inhibits histone lysine demethylases." EMBO Rep 2011, 12, 463. DOI: 10.1038/embor.2011.43

(22) Lu, C.; Ward, P. S.; Kapoor, G. S.; Rohle, D.; Turcan, S.; Abdel-Wahab, O.; Edwards, C. R.; Khanin, R.; Figueroa, M. E.; Melnick, A.; Wellen, K. E.; O'Rourke, D. M.; Berger, S. L.; Chan, T. A.; Levine, R. L.; Mellinghoff, I. K.; Thompson, C. B. "IDH mutation impairs histone demethylation and results in a block to cell differentiation." Nature 2012, 483, 474. DOI: 10.1038/nature10860

(23) Koivunen, P.; Lee, S.; Duncan, C. G.; Lopez, G.; Lu, G.; Ramkissoon, S.; Losman, J. A.; Joensuu, P.; Bergmann, U.; Gross, S.; Travins, J.; Weiss, S.; Looper, R.; Ligon, K. L.; Verhaak, R. G.; Yan, H.;

Kaelin, W. G., Jr. "Transformation by the (R)-enantiomer of 2-hydroxyglutarate linked to EGLN activation." Nature 2012, 483, 484. DOI: 10.1038/nature10898

(24) Losman, J. A.; Looper, R. E.; Koivunen, P.; Lee, S.; Schneider, R. K.; McMahon, C.; Cowley, G. S.; Root, D. E.; Ebert, B. L.; Kaelin, W. G., Jr. "(R)-2-hydroxyglutarate is sufficient to promote leukemogenesis and its effects are reversible." Science 2013, 339, 1621. DOI: 10.1126/science.1231677

(25) Flavahan, W. A.; Drier, Y.; Liau, B. B.; Gillespie, S. M.; Venteicher, A. S.; Stemmer- Rachamimov, A. O.; Suva, M. L.; Bernstein, B. E. "Insulator dysfunction and oncogene activation in IDH mutant gliomas." Nature 2016, 529, 110. DOI: 10.1038/nature16490

(26) Rohle, D.; Popovici-Muller, J.; Palaskas, N.; Turcan, S.; Grommes, C.; Campos, C.; Tsoi, J.; Clark, O.; Oldrini, B.; Komisopoulou, E.; Kunii, K.; Pedraza, A.; Schalm, S.; Silverman, L.; Miller, A.; Wang, F.; Yang, H.; Chen, Y.; Kernytsky, A.; Rosenblum, M. K.; Liu, W.; Biller, S. A.; Su, S. M.; Brennan, C. W.; Chan, T. A.; Graeber, T. G.; Yen, K. E.; Mellinghoff, I. K. "An inhibitor of mutant IDH1 delays growth and promotes differentiation of glioma cells." Science 2013, 340, 626. DOI: 10.1126/science.1236062

(27) Wang, F.; Travins, J.; DeLaBarre, B.; Penard-Lacronique, V.; Schalm, S.; Hansen, E.; Straley, K.; Kernytsky, A.; Liu, W.; Gliser, C.; Yang, H.; Gross, S.; Artin, E.; Saada, V.; Mylonas, E.; Quivoron, C.; Popovici-Muller, J.; Saunders, J. O.; Salituro, F. G.; Yan, S.; Murray, S.; Wei, W.; Gao, Y.; Dang, L.; Dorsch, M.; Agresta, S.; Schenkein, D. P.; Biller, S. A.; Su, S. M.; de Botton, S.; Yen, K. E. "Targeted inhibition of mutant IDH2 in leukemia cells induces cellular differentiation." Science 2013, 340, 622. DOI: 10.1126/science.1234769

(28) Kernytsky, A.; Wang, F.; Hansen, E.; Schalm, S.; Straley, K.; Gliser, C.; Yang, H.; Travins, J.; Murray, S.; Dorsch, M.; Agresta, S.; Schenkein, D. P.; Biller, S. A.; Su, S. M.; Liu, W.; Yen, K. E. "IDH2 mutation-induced histone and DNA hypermethylation is progressively reversed by small-molecule inhibition." Blood 2014, 125, 296.

(29) Okoye-Okafor, U. C.; Bartholdy, B.; Cartier, J.; Gao, E. N.; Pietrak, B.; Rendina, A. R.; Rominger, C.; Quinn, C.; Smallwood, A.; Wiggall, K. J.; Reif, A. J.; Schmidt, S. J.; Qi, H.; Zhao, H.; Joberty, G.; Faelth-Savitski, M.; Bantscheff, M.; Drewes, G.; Duraiswami, C.; Brady, P.; Groy, A.; Narayanagari, S. R.; Antony-Debre, I.; Mitchell, K.; Wang, H. R.; Kao, Y. R.; Christopeit, M.; Carvajal, L.; Barreyro, L.; Paietta, E.; Makishima, H.; Will, B.; Concha, N.; Adams, N. D.; Schwartz, B.; McCabe, M. T.; Maciejewski, J.; Verma, A.; Steidl, U. "New IDH1 mutant inhibitors for treatment of acute myeloid leukemia." Nat Chem Biol 2015, 11, 878. DOI: 10.1038/nchembio.1930

(30) de Botton, S.; Pollyea, D. A.; Stein, E. M.; DiNardo, C.; Fathi, A. T.; Roboz, G. J.; Collins, R.; Swords, R. T.; Flinn, I.; Altman, J. K.; Tallman, M. S.; Kantarjian, H.; Colby, K.; Fan, B.; Goldwasser, M.; Prahl, M.; Agresta, S.; Stone, R. M. In European Hematology Association Learning Center 2015, p P563.

(31) Stein, E. M.; Altman, J. K.; Collins, R.; DeAngelo, D. J. In 56th ASH Annual Meeting; American Society of Hematology: San Francisco, CA, 2014.

(32) Zheng, B.; Yao, Y.; Liu, Z.; Deng, L.; Anglin, J. L.; Jiang, H.; Prasad, B. V.; Song, Y. "Crystallographic Investigation and Selective Inhibition of Mutant Isocitrate Dehydrogenase." ACS Med Chem Lett 2013, 4, 542. DOI: 10.1021/ml400036z

(33) Deng, G.; Shen, J.; Yin, M.; McManus, J.; Mathieu, M.; Gee, P.; He, T.; Shi, C.; Bedel, O.; McLean, L. R.; Le-Strat, F.; Zhang, Y.; Marquette, J. P.; Gao, Q.; Zhang, B.; Rak, A.; Hoffmann, D.; Rooney, E.; Vassort, A.; Englaro, W.; Li, Y.; Patel, V.; Adrian, F.; Gross, S.; Wiederschain, D.; Cheng, H.; Licht, S. "Selective inhibition of mutant isocitrate dehydrogenase 1 (IDH1) via disruption of a metal binding network by an allosteric small molecule." J Biol Chem 2015, 290, 762. DOI: 10.1074/jbc.M114.608497

(34) Wu, F.; Jiang, H.; Zheng, B.; Kogiso, M.; Yao, Y.; Zhou, C.; Li, X. N.; Song, Y. "Inhibition of Cancer-Associated Mutant Isocitrate Dehydrogenases by 2-Thiohydantoin Compounds." J Med Chem 2015, 58, 6899. DOI: 10.1021/acs.jmedchem.5b00684

(35) Golub, T. R.; Slonim, D. K.; Tamayo, P.; Huard, C.; Gaasenbeek, M.; Mesirov, J. P.; Coller, H.; Loh, M. L.; Downing, J. R.; Caligiuri, M. A.; Bloomfield, C. D.; Lander, E. S. "Molecular classification of cancer: class discovery and class prediction by gene expression monitoring." Science 1999, 286, 531. DOI: 10.1126/science.286.5439.531

(36) Sauvageau, G.; Lansdorp, P. M.; Eaves, C. J.; Hogge, D. E.; Dragowska, W. H.; Reid, D. S.; Largman, C.; Lawrence, H. J.; Humphries, R. K. "Differential expression of homeobox genes in functionally distinct CD34+ subpopulations of human bone marrow cells." Proc Natl Acad Sci U S A 1994, 91, 12223.

(37) Collins, C. T.; Hess, J. L. "Role of HOXA9 in leukemia: dysregulation, cofactors and essential targets." Oncogene 2016, 35, 1090. DOI: 10.1038/onc.2015.174

(38) Tedeschi, F. A.; Zalazar, F. E. "HOXA9 gene expression in the chronic myeloid leukemia progression." Leuk Res 2006, 30, 1453. DOI: 10.1016/j.leukres.2006.02.022

(39) Adamaki, M.; Lambrou, G. I.; Athanasiadou, A.; Vlahopoulos, S.; Papavassiliou, A. G.; Moschovi, M. "HOXA9 and MEIS1 gene overexpression in the diagnosis of childhood acute leukemias: Significant correlation with relapse and overall survival." Leuk Res 2015, 39, 874. DOI: 10.1016/j.leukres.2015.04.012

Chapter 2

Development of a Biologically Relevant Assay for IDH1 Inhibition

Portions of this chapter are reproduced with permission from the following publication:

Law, J. M.; Stark, S. C.; Liu, K.; Liang, N. E.; Hussain, M. M.; Leiendecker, M.; Ito, D.; Verho, O.; Stern, A. M.; Johnston, S. E.; Zhang, Y.-L.; Dunn, G. P.; Shamji, A. F.; Schreiber, S. L. "Discovery of 8-Membered Ring Sulfonamides as Inhibitors of Oncogenic Mutant Isocitrate Dehydrogenase 1." ACS Med. Chem. Lett. 2016. DOI: 10.1021/acsmedchemlett.6b0026441 Collaborator Contributions: Dr. Ke Liu and Dr. Daisuke Ito developed the first iteration of the biochemical assay. Dr. Mahmud Hussain and Dr. Matthias Leiendecker worked with me on the synthesis of BRD5667 and its analogs. The assay measuring R-2HG levels in cells was developed and run with Dr. Hussain, Dr.

Ito, Dr. Leiendecker, and Dr. Stephen Johnston. High-throughput compound screening was conducted by

Dr. Yan-Ling Zhang.

2.1 The importance of a biologically relevant assay

The sine qua non of any screening project is an assay that faithfully reproduces the biological

activity of interest in an in vitro setting. Without such an assay, even compounds displaying excellent

activity in the screening assay are likely to prove useless when applied to modulate the real biological

system. Thus, assay development is the most critical phase of a probe development project. In this

chapter, I describe the extensive, nonlinear process leading to the discovery of assay conditions

appropriate for in vitro screening for inhibition of purified IDH1-R132H. My colleagues and I conducted a

screen and performed a round of medicinal chemistry aimed at optimizing in vitro potency before we

realized we had been optimizing compounds for an assay that was not biologically relevant. While not

leading to a successful chemical probe, this early screening effort did teach us which parameters are

necessary in a biologically-relevant screening assay and gave us hints about the endogenous catalytic

mechanism of IDH1-R132H, setting the stage for the more successful phase of probe development

described in Chapter 3.

2.2 Assay design for IDH1-R123H inhibition

2.2.1 Assay design: detection technology

Utilization of the best possible readout technology is key to the success of a biochemical screen.

In an assay that will be run over 100,000 times and read automatically, the signal must be quickly

detectable by robotic equipment, the signal-to-noise ratio must be high, and false positives and negatives

must be minimized. Some sort of photometric readout is preferred.

Figure 2.1: Chemical reactions involved in the screen for IDH1-R132H activity, and detection technologies used. (A) Molecules involved in the IDH1-R132H-catalyzed ketoreductase reaction can be measured directly in low throughput. (B) NADPH can be used to reduce resazurin to the brightly- fluorescent resorufin through action of diaphorase, producing a bright signal for high-throughput detection.

In the case of the IDH1-R132H reaction, the substrate NADPH can itself be measured by absorbance or fluorescence spectroscopy (Figure 2.1A). While I utilized this direct measurement of

NADPH for follow-up experiments, the fluorescence is too weak for a screening experiment run on small volumes of material. I was fortunate that IDH1 is a redox enzyme, meaning that reactants or products exist in an energetically unstable redox state, rendering them chemically reactive. Thus, I could rely on the ability of NADPH to reduce nonfluorescent resazurin to the brightly fluorescent resorufin in the

presence of the enzyme diaphorase (Figure 2.1B). Not only is resorufin brightly fluorescent, but it absorbs

at a longer wavelength than most molecules found in the screening collection, making it unlikely those

molecules would interfere with the assay through autofluorescence.

Low assay volumes are beneficial, allowing large numbers of assay reactions to be run more

quickly at lower cost. With relatively simple chemistry not requiring large fluid volumes and a brightly

fluorescent readout, I chose to run the assay in 1536-well plates using a 5 µL reaction volume.

2.2.2 Assay design: compound selection

As other academic groups and pharmaceutical companies were known to be looking for mutant-

IDH1 inhibitors, I needed to build into my effort some sort of comparative advantage to avoid duplicating

the work of others. To this end, the Broad Institute’s diversity-oriented synthesis (DOS) compound library

provided me and my team exclusive access to chemical matter with increased sp3 content and myriad stereochemical features compared with that found in a standard commercial library.2,3 I hypothesized that

screening the DOS collection could lead to a probe with unique binding characteristics.

2.2.3 Assay design: reagent concentrations

Conducting a compound screen in the presence of high concentrations of enzyme substrates maximizes the speed of the reaction, but can make it difficult for competitive inhibitors to successfully compete for their binding site. To allow detection of competitive inhibitors, it is best practice to conduct compound screening with substrates present at concentrations near their Km. Thus, I measured each

enzyme-substrate Km value in the buffer conditions that would be used for the screening assay. In our case, α-KG-competitive inhibitors were considered likely and α-KG was included in the screening assay at

2+ its measured Km of 0.6 mM. The metal cofactor Mn was included at a concentration near its Km value for the same reason.

By contrast, the Km of IDH1-R132H for NADPH is too low for it to be practical to include NADPH at its Km concentration. As NADPH is involved in the detection of reaction progress, high levels of NADPH are

necessary for a robust signal. Thus, I had to include NADPH at many times its Km concentration, meaning it would be difficult for the screen to detect NADPH-competitive inhibitors unless they were extremely potent.

2.2.4 Assay design: the details

In designing the assay, I needed to optimize the reaction for the screening equipment on which it was run. The reaction needed to be slow enough to allow batch-based plate reading but fast enough that small fluid volumes would not have a chance to evaporate before the reaction was complete. The stoichiometry of the system had to be set such that the assumptions of steady-state enzyme kinetics held true. Fluid-dispensing and plate-reading equipment had to be calibrated. A standard curve was run to ensure that signal varied linearly with concentration of the detection reagent.

Even with the detection working well, I was confronted with many assay details which seem minor but, given the sensitivity of enzymes to their precise environment, can prove key to the success of the screen. Such details include the pH of the reaction, the choice of buffering agent, and the presence of detergent or nonspecific protein. In my case, the most problematic detail proved to be the choice of metal cofactor required by IDH1 to catalyze its reaction.

2.3 The enzymatic assay functions most easily with Mn2+ as a cofactor

IDH1 is reported to function with either Mn2+ or Mg2+ as a cofactor, and my colleagues and I

eventually performed screens under both conditions. The cofactor used initially was Mn2+, because

enzyme turnover was substantially faster (Figure 2.2). The faster turnover meant assays could run with

less enzyme present, leading to cheaper and easier compound screening. Unfortunately, the potency of

inhibitors in this model system eventually proved to be a poor predictor of cellular activity and I later

needed to design screening conditions using Mg2+ as the cofactor to proceed with the project.

Figure 2.2: Relative velocity of IDH1-R132H with Mg2+ and Mn2+ cofactor, under reaction conditions optimized separately for each. Data are mean ± SD of 3 technical replicates of one representative of three independent experiments.

2.4 Screening identifies BRD5667 as an IDH1-R132H inhibitor

Using the Mn2+-based enzymatic assay, my colleagues screened roughly 82,900 compounds in duplicate, consisting of roughly 63,000 novel compounds synthesized using diversity-oriented synthesis

(DOS) and 20,000 compounds sourced from commercial libraries. From the 114 active compounds showing >40 % enzyme inhibition in both replicates, we prioritized compounds by potency,

reproducibility, and the presence of structure/activity relationships (SARs) in the screening results. The

DOS screening library consists of many groups of structural analogs for a given scaffold, including nearly

all possible stereoisomers of each compound. This design enables the identification of series that display

SAR suggestive of a specific molecular interaction with the protein target. In a process that would later

serve as a template for the discovery of the improved probe BRD2879 (Chapter 3), my colleagues and I

identified BRD5667, a 9-membered amide containing 3 stereocenters in (2S, 5R, 6R) configuration which

displayed an IC50 at least 10-fold lower than those of its stereoisomers (Figure 2.3). This stereochemistry–

activity pattern was consistent among structurally related compounds in the screening collection.

a b OH Stereochemistry (C2, C5, C6) O (S) IC50 (µM) 2 N 5 (R) O 6 R,S,S S,R,R R,R,R S,S,S O (R) >60 2.1 18 >60 N N N H H N R,S,R S,R,S R,R,S S,S,R >60 17 30 >60 BRD5667

Figure 2.3: (A) Structure of BRD5667, the lead compound from the IDH1-R132H inhibition screen in Mn2+ conditions. (B) Potency of stereoisomers of BRD5667 in the enzymatic assay. The dependence of enzyme inhibition on stereochemistry is suggestive of a specific binding interaction between the small molecule and enzyme.

2.5 Synthesis and testing of analogs defines structure-activity relationships for BRD5667

To confirm the activity of BRD5667 and find analogs with improved potency, my colleagues and I

resynthesized BRD5667, along with all its stereoisomers and 71 analogs. The synthesis of analogs was

facilitated by the modular synthesis developed earlier in our research group,4 by which a core structure can be built up through a relatively straightforward 10-step synthesis which follows a build/couple/pair strategy (Figure 2.4). This core contains two differently protected nitrogen atoms that serve as attachment points for appendages R1 and R2.

Preliminary SARs were established by examining modifications of three regions of the molecule: the aniline nitrogen (R1), secondary amine (R2), and amido alcohol side chain (R3) (Figure 2.5). I focused on the

synthesis of truncated R3 analogs by incorporating alternate building blocks early in the synthesis (Figure

2.11, Methods Section), while Dr. Hussain and Dr. Leiendecker focused on coupling different appendages

in the R1 and R2 positions. Recalling the stereochemical SAR for BRD5667, we synthesized analogs retaining

the (2S, 5R, 6R) configuration.

OH ∗ PMBO O∗ NH2 OTBS ∗ ∗ ∗ N PMBO ∗ O OTBS N O ∗ ∗ ∗ H N HO Boc N H N N 2 Boc Boc

Build Couple Pair

Figure 2.4: I resynthesized DOS compounds following the build/couple/pair synthetic strategy. The final structure may be synthesized with any desired stereochemistry at the indicated three positions through appropriate choice of “build” components.

Substituting R1 with hydrogen or a range of functional groups, including ureas and amides bearing aliphatic and aromatic groups with varying electronic demands, had minimal effect on potency in the

IDH1-R132H enzymatic assay. However, substitution with certain polar groups or elimination of the urea linker led to diminished activity, implying that the R1 substitutions can modulate activity in some circumstances. Several changes to the polarity and size of the amido alcohol side chain (R3) were tolerated, although truncating the isopropyl alcohol side chain by substitution with a simple methyl or isopropyl group led to moderately reduced activity. In contrast, the IDH1-R132H inhibitory activity of BRD5667 was exquisitely sensitive to modifications at the R2 position: for example, removing the heteroatom of the pyridyl group or changing the position of its nitrogen atom nearly eliminated inhibitory activity. Electron- withdrawing groups meta to the pyridyl nitrogen also led to reduced activity, suggesting that the pyridyl

nitrogen may participate in a hydrogen bond. These SAR studies were successful in that we discovered

regions of BRD5667 that are crucial for its inhibitory function and regions that may be have been amenable

for further modification. However, we were unable to discover any analogs with high potency in cells.

OH OH O O O R3 N N N Cy O O NH O O CH2(4-Pyr) O N R2 N N N N N O N R1 N H H H H IC50 IC50 IC50 IC50 IC50 R1 R1 R2 R2 R3 (µM) (µM) (µM) (µM) (µM)

F OH 2.1 N N 60 2.1 15 2.1 N N N N N H

MeO 3.6 3.2 > 60 O N > 60 O 5.8 MeO N N N H H N H

O F N 1.9 N N 2.5 > 60 > 60 12 H N N O N H

8.0 N 21 > 60 28 30 H N

Cl O O N 40 N H 9.9 7.9 > 60 34 O N N

O CH3 H O N N N NH N 27 14 > 60 N 2.6 N 3.1 N H N H

Figure 2.5: Structure-activity relationship of BRD5667, depicting selected analogs. IDH1-R132H inhibition activity is exquisitely sensitive to modification at the R2 position, but not at the R1 or R3 positions.

2.6 BRD5667 and analogs work poorly in cells

To confirm the activity of our inhibitors in a more physiologically relevant context, we established

a cell-based model system for mutant IDH1 activity. Since it has proved difficult to establish robust cell

lines directly from IDH-mutant tumors, we developed an engineered model based on overexpression of

IDH1-R132H in HA1E-M cells (a HEK-293 derivative).5 These cells express high levels of R-2HG, which is released into the growth media. I determined the cellular activity of compounds by measuring R-2HG

present in conditioned media by liquid chromatography-mass spectrometry (LC-MS) after 48 hours of

compound treatment. After harvesting media, I also determined viability of compound-treated cells by

observing cell morphology and measuring ATP levels.

The measurements of R-2HG in cell media did convince me that the compound is active in cells,

and that activity is probably on target. Measurements indicated a dose response, lack of gross toxicity associated with treatment, and some correlation between activity of BRD5667 stereoisomers in the enzymatic and cell-based assays (Figure 2.6). Unfortunately, potency was poor, with 80 µM BRD5667 required to reduce R-2HG levels by 50%. Cell-based tests of all analogs showing high potency against purified IDH1-R132H did not reveal molecules with significantly improved cell-based activity compared to

BRD5667.

A B C

Figure 2.6: (A) Treatment with BRD5667 partially reduces R-2HG production in IDH1-R132H- expressing cells. HA1E-M cells expressing R132H were treated with various concentrations of BRD5667 and cellular 2-HG levels were then analyzed by LC-MS after 48 h. (B) No gross toxicity is associated with treatment with BRD5667, as determined by the measurement of ATP levels. (C) Cellular R-2HG levels after 48 h treatment with 80 µM BRD5667 (“SRR”) and two of its stereoisomers. Potency of the stereoisomers in the purified enzyme inhibition assay is shown for comparison. Data are mean ± SEM of duplicates in one representative of three independent experiments.

2.7 Compound activity depends on which metal cofactor is used

While I was struggling to get BRD5667 analogs to function more effectively in cells, a description of

another IDH1-R132H inhibitor, AGI-5198, was published.6 While in some sense this publication meant my team had been “beaten” in the race to publish the first IDH1-R132H inhibitor, the information in the publication helped my project in several ways. First, the failure of BRD5667 or its analogs to completely suppress R-2HG production in cells, contrasted with the ability of AGI-5198 to do the same, prompted me to compare the probes side-by-side in a number of assays (Figure 2.7). While running these comparisons,

I was unable to replicate the reported potency of AGI-5198 using my Mn2+-based assay. Considering

discrepancies between my assay and that of AGI-5198’s discoverers, I noticed that the AGI-5198 publication described Mg2+ as the metal cofactor in the enzymatic assay. These discoveries aroused my

suspicion that compound potency was somehow dependent on the identity of the metal cofactor in the

assay and prompted me to renew efforts to design a workable assay utilizing Mg2+.

On the recommendation of Dr. Andrew Stern, I added to the Mg2+-containing assay buffer 0.1 mM

1,10-phenanthroline, a chelating agent meant to prevent any transition-metal contaminants from interfering with the enzyme. With this additive in place, the Mg2+-based assay could be run at sufficient speed to test the effects of metal cofactor choice on the ability of BRD5667 and its analogs to inhibit IDH1-

R132H. To my surprise, the choice of metal cofactor led to an approximately tenfold difference in the potency of R132H inhibitors. The biggest surprise, however, was that the directionality of this potency difference was opposite between the BRD5667 and AGI-5198 series of compounds. Specifically, BRD5667 was 10x as potent in the Mn2+ conditions, while AGI-5198 was 10x as potent in the Mg2+ conditions (Figure

2.8). This meant that although the probes had similar potency under Mn2+ conditions, AGI-5198 was roughly 100x as potent under Mg2+ conditions.

Figure 2.7: BRD5667 treatment reduces R-2HG secreted by cells, but to a lesser extent than AGI-5198 analogs, even at a higher dose. (A) Treatment with 80 µM BRD5667 reduces cellular R-2HG secretion by 60%, but treatment with 5 µM of some AGI-5198 analogs reduces cellular R-2HG secretion by over 90%. (B) None of the compounds tested had a large effect on cell viability as determined by ATP levels. (C) Structures of AGI-5198 analogs used in this experiment. All measurements were taken after 48 hours of compound treatment. Error bars are mean ± SD of three replicates in one representative experiment.

Figure 2.8: Potency of IDH1-R132H inhibitors is dependent on the identity of the metal cofactor. (A) BRD5667 is more potent in the presence of Mn2+, although the Hill coefficient is closer to 1 in the presence of Mg2+. (B) In contrast, AGI-5198 is more potent in the presence of Mg2+. Data are from one representative of three experiments.

2.8 Mg2+ is superior to Mn2+ as a cofactor in the IDH1 enzymatic assay

With the discrepancies between the Mn2+ and Mg2+ enzymatic assays apparent, it was important to

determine which assay was more faithful to the true biological context. Although the enzyme had faster

turnover using Mn2+, three pieces of evidence convinced me that the Mg2+-based assay was a better

model.

First, as shown in Figure 2.8A, the Hill coefficient of the inhibition curve for BRD5667 in Mn2+

conditions is substantially less than one. This is a deviation from the prediction of simple enzyme inhibition

models, and while there could be innocuous explanations for this deviation, it indicates the assay is not

as clean or simple as it could be. In the Mg2+-based assay, the Hill coefficients of the inhibition curves of both probes are equal to one, within experimental error, as predicted by simple enzyme inhibition models.

Beyond the issues with the Hill coefficient, the Mn2+-based assay was in general less robust, with changes in the buffering agent and sulfate concentration causing substantial variation in assay performance. The lack of assay robustness and low Hill coefficient caused me to worry that the enzyme did not reside in a single, stable low-energy conformation in the presence of Mn2+.

Second, the Mg2+-based assay did a much better job of predicting the efficacy of probes in the cell-

based assay. While the cell-based assay is an artificial system, with the HA1E-M cells massively overexpressing IDH1-R132H and feeding on MEMα media, it is highly likely to be closer to the natural biological state than a purified enzyme assay. The Mg2+-based enzymatic assay was consistent with the cell-based assay in showing that AGI-5198 was approximately 100x as potent as BRD5667.

Third, biochemical calculations based on my enzymology experiments and estimates from the

literature suggested that IDH1-R132H would normally bind Mg2+ instead of Mn2+ in the body. While I

2+ 2+ measured the enzyme’s Km for Mn as 10x lower than its Km for Mg , literature estimates suggest the amount of Mn2+ in cells is much less than a tenth of the amount of Mg2+. Mg2+ levels are typically in the millimolar range,7 while Mn2+ is not found at levels above 20 µM in any tissue, and levels in serum are approximately 20 nM.8 Thus, Mg2+ would be expected to occupy the active site of the enzyme most of the

time. This is also consistent with biochemical intuition: Mg2+ is widely used as a Lewis acid catalyst for the activation of carbonyl groups—its role in IDH1—while Mn2+, as a transition metal ion, is best known for

facilitating single-electron chemistry in superoxide dismutase 2.

Metal Cofactor IDH1 Wild Type Km (mM) IDH1-R132H Km (mM) Mn2+ 0.014 0.5 Mg2+ 0.11 5

2+ 2+ Figure 2.9: Km values for Mn and Mg for wild-type and R132H-mutant IDH1. The wild-type enzyme binds metal cofactors much more tightly than the mutant form, but the relative affinity for Mn2+ vs. Mg2+ is unchanged.

While it is conceivable that the R132H mutation could rearrange the enzyme active site sufficiently to change metal-binding preferences, comparison of the Km values of wild-type and R132H- mutant IDH1 for Mn2+ and Mg2+ suggests relative metal binding is unchanged by the mutation (Figure 2.9).

2.9 All assay conditions were optimized before the final round of screening

Before screening the DOS compound library for inhibition of ketoreductase activity of IDH1-R132H in the presence of Mg2+, I undertook to reoptimize each parameter of the assay. With a change in metal cofactor from Mn2+ to Mg2+ and the use of a new batch of recombinant protein in which an unintended

A307S mutation was corrected, there was no guarantee that the original assay parameters would be ideal

for the new screen. Thus, I conducted experiments to reoptimize each parameter in turn. Optimized

values are listed in Figure 2.10.

Assay Component Optimized Value Reaction time 45 minutes Buffering agent Tris pH 7.5 [aKG] 0.6 mM [NADPH] 22 µM [IDH1-R132H] 1 µg/mL [Mg2+] 10 mM [1,10-phenanthroline] 0.1 mM [Diaphorase] 0.3 µg/mL Detergent Tween 20, 0.01 % Excitation/Emission Wavelength 535/595 nm Positive control AGI-5198 [Positive control] 15 μM Figure 2.10: Final optimized values for each assay parameter in the screen for inhibition of ketoreductase activity of IDH1-R132H in the presence of Mg2+

In summary, to identify small molecules that inhibit IDH1-R132H, I developed an in vitro assay measuring the enzyme’s ketoreductase activity in 1536-well plates. Enzyme activity was measured by detecting consumption of NADPH in a diaphorase-coupled reaction with a fluorescence readout. The screening buffer used, critically, 10 mM MgCl2 as the metal cofactor source and included 0.01% Tween 20 detergent to minimize false positives due to compound aggregation. The compound source was the Broad

Institute’s DOS library. With these painstakingly determined assay conditions in hand, I was able to

conduct the screen described in Chapter 3, from which I was finally able to validate potent, cell-active

IDH1-R132H inhibitors.

2.10 Experimental methods: biology and biochemistry

2.10.1 Cloning, expression and purification of wild type and mutant IDH1 in E. coli.

The human isocitrate dehydrogenase (IDH1, EC 1.1.1.42) cDNA clone was purchased from

Invitrogen (Cat# IOH62682). QuickChange Lightning site-directed mutagenesis kit (Aligent Cat# 210519)

was used to generate the R132H and R132C mutations; the mutation was confirmed by sequencing. Wild-

type and R132H mutant IDH1 were subcloned into pET41a (EMD Biosciences) using NDEI and XHOI

restriction sites to enable the E. coli expression of C-terminal His8-tagged proteins. Wild-type and mutant

IDH1 proteins were expressed in Rosetta E. coli in LB media using kanamycin and chloramphenicol as selection agents and IPTG to induce protein production. After overnight protein expression at 18°C, cells were pelleted and resuspended in Lysis Buffer (20 mM Tris pH 7.4, 0.1% Triton X-100, 500 mM NaCl, 5 mM BME, 10% glycerol, protease inhibitor cocktail), then lysed using a microfluidizer at 18,000 PSI. Cell debris was removed by ultracentrifugation and the supernatant was passed through a 0.45 μm filter.

Protein was purified by nickel affinity chromatography on a HisTrap 5 mL column (GE Healthcare) using a gradient of 5-100% Column Buffer B (20 mM Tris pH 7.4, 500 mM NaCl, 10% glycerol, 500 mM imidazole) in Column Buffer A (20 mM Tris pH 7.4, 500 mM NaCl, 10% glycerol). Purity was verified by SDS-PAGE with

Coomassie Blue staining. Protein concentration was determined by A280 using protein denatured in 6 M guanidinium chloride and an extinction coefficient based on the sequence of IDH1, 64080 (M cm)-1,

calculated with ExPASy ProtParam. Proteins were stored in 50% glycerol in Enzyme Storage Buffer (20

mM Tris pH 7.4, 200 mM NaCl, 5 mM BME, and 10% glycerol) at -80°C.

2.10.2 Enzyme assay development and screening for inhibitors of IDH1-R132H – Mn2+ conditions.

The final IDH1-R132H-Mn2+ assay conditions were as follows: IDH1-R132H enzyme stock (3.18

mg/ml) and substrates (NADPH & α-KG) were diluted in assay buffer containing 100 mM Tris HCl pH 7.4,

150 mM NaCl, 5 mM MnCl2 and 0.03% BSA. 2.5 μL enzyme mix (16 μg/mL IDH1-R132H) and 5 nL of 10 mM

compound per well are pre-incubated for 20 min. Reaction was initiated by addition of 2.5 μL of substrate

mix (40 μM NADPH, 0.8 mM α-KG), followed by 50 min incubation at RT, at which time the reaction is still

in the linear range. Plates were read on an Envision plate reader before the addition of detection mix to

identify fluorescent compounds. IDH1-R132H enzyme reaction was then terminated by 2.5 μL of detection mix (1 μg/mL diaphorase, 50 μM resazurin) and assay plates are measured again in an Envision plate reader (excitation 535 nm, emission 595 nm). We screened 74,437 compounds in duplicate, 49,237 of which were also screened in the Mg2+ conditions (Chapter 3). Compounds that inhibit ≥40% enzyme activity in the primary screen were selected as hits, corresponding to >4x the sum of the standard deviations of the negative (DMSO) and positive (no enzyme) controls.

2.10.3 Absorbance assay for IDH1-R132H, Mn2+ conditions

The buffer composition was as follows: 25 mM Tris-HCl, 150 mM NaCl, 0.1% PEG3350, 10 mM MnSO4, pH 7.5. To a solution of 40 μL/well IDH1-R132H enzyme in buffer was added compound in

DMSO; compound and enzyme were incubated for 20 minutes before addition of a substrate mix containing α-KG and NADPH. Compounds were plated in 8-point dose (3x dilution starting at 16.7 μM) in clear 384-well plates (Corning #3640). The final concentration of reagents was 3.3 μg/mL IDH1-R132H, 0.6 mM α-KG, 75 μM NADPH, and 0.17% DMSO. After addition of substrates, the absorbance of NADPH at

340 nm was monitored via Spectramax M5 plate reader. The slope of the linear portion of the absorbance vs. time trace was used to determine enzyme activity.

2.10.4 Cell-based assay: measurement of R-2HG in conditioned media

HA1E-M-IDH1-R132H cells were counted and plated in 96-well tissue-culture plates (Corning

3904) at a density of 10,000 cells/well. The plates were incubated to allow the cells to adhere and grow to confluence (about 36 hours). Media was aspirated and immediately replaced with 100 μL/well media containing compound (DMSO concentration 0.3%). After 48 hours of treatment, 60 μL/well media was removed and added to 800 μL aqueous 80% methanol solution containing deuterated racemic 2HG (2HG- d4) as an internal standard. The methanol solution was evaporated at 40°C under a stream of dry nitrogen and the resulting residue was resuspended in a solution of 9 mM ammonium hydroxide in 68% acetonitrile/22% methanol/10% water. This mixture was briefly vortexed and sonicated to encourage dissolution of R-2HG, then centrifuged to remove solids. Targeted LC-MS data were then acquired using a

Waters 2795 separations module and Waters 3100 mass detector. R-2HG was separated using a Luna NH2

column (50 x 2 mm, 3 μm beads, cat. no. 00B-4377-B0, Phenomenex, Torrance, CA) and eluted using a 5

min linear gradient initiated with 90% mobile phase B (10 mM ammonium hydroxide in 75%

acetonitrile/25% methanol) and concluding with 100% mobile phase A (aqueous solution of 20 mM

ammonium hydroxide and 20 mM ammonium acetate). 2HG is measured using selected-ion monitoring

in the negative ion mode at M/Z 147 (2HG) and 151 (2HG-d4). The ratio of ion counts at these M/Z is used

to determine the relative amounts of R-2HG in various conditioned media samples.

2.10.5 Determination of cell viability

Cells were observed under a light microscope (100x magnification) to observe visible changes in

cell morphology. At toxic compound concentrations, cells appeared shriveled up or were not present. For

quantitative viability measurements, intracellular ATP levels were determined via CellTiter-Glo assay

(Promega, Madison, WI).

2.11 Experimental methods: chemistry, with compound characterization

2.11.1 General analytical methods

NMR spectra were recorded on a Bruker UltraShield 300 (300 MHz 1H, 75 MHz 13C) or Bruker

Avance 400 (400 MHz 1H, 100 MHz 13C) NMR spectrometer. Spectra were referenced to the residual

solvent peak. Data are reported as follows: chemical shift in ppm (multiplicity (br = broad, s = singlet, d =

doublet, t = triplet, q = quartet, quin = quintet, sex = sextet, m = multiplet), coupling constants (Hz),

integration). Reactions were monitored by TLC and LC-MS (see below). Flash chromatography was performed on a Teledyne Isco CombiFlash Rf system using RediSep Rf columns. High-performance liquid chromatography (HPLC) was performed on a Waters HPLC system using a basic solvent system

(water/acetonitrile/0.2 % NH4OH).

2.11.2 Compound purity determination by LC-MS

Compound purity and identity were determined by LC-MS (Alliance 2795, Waters, Milford, MA).

Purity was measured by UV absorbance at 210 nm. Identity was determined on a SQ mass spectrometer by positive and negative electrospray ionization. Mobile phase A consisted of 0.01% formic acid in water, while mobile phase B consisted of 0.01% formic acid in acetonitrile. The gradient ran from 5% to 95% mobile phase B over 1.75 minutes at 1.75 mL/min. An Agilent Poroshell 120 EC-C18, 2.7 µm, 3.0x30 mm column was used with column temperature maintained at 40 °C. 2.1 µL of sample solution was injected.

2.11.3 Exact mass determination

High-resolution mass-spectra were acquired on an Agilent 1290 Infinity separations module coupled to a 6230 time-of-flight (TOF) mass detector operating in ESI+ or ESI- mode. Masses were confirmed using the "Find by Formula" feature in MassHunter Qualitative Analysis vB.06.00.

2.11.4 Representative synthesis of BRD5667 analogs varied at the R3 position

Si 1. PyBOP, DIPEA Si Si BH - Me S O O Boc 2. isopropylamine O O Boc 3 2 O Boc N N N HO N N H H 2-1 2-2 2-3

O Si Cl O Boc O F N N N a. NH TBAF 4F, O O N 2 O b. NaH N F O N 2 Boc 2-4 Et3N 2-5 O2N

O O O C N N N H2, Pd/C O O O N N N N H N Boc 2 Boc H H 2-6 2-7

O O

TFA N 1. 4-pyridinecarboxaldehyde, N O 2% AcOH in DMF O O O N 2. Na(OAc)3BH N N N H N N H H H H 2-8 2-9 N

Figure 2.11: Scheme for the synthesis of BRD5667 analogs varying at the R3 (amide substituent) position. Varying analogs can be synthesized based on the choice of coupling partner in step 1. As an example, synthesis of the isopropyl analog 2-9 is depicted here.

Si 1. PyBOP, DIPEA Si O O Boc 2. isopropylamine O O Boc N N HO N H 2-1 2-2

Compound 2-1 was synthesized according to published procedure.4 The remaining reagents for the following reaction are commercially available. To an oven-dried round-bottom flask equipped with a magnetic stirrer was added 2-1 (945 mg, 2.61 mmol, 1.0 equiv) and 10 mL dichloromethane. PyBOP (1.50

g, 2.88 mmol, 1.1 equiv) and diisopropyl ethylamine (1.01 g, 7.84 mmol, 3.0 equiv) were added. The

resulting mixture was cooled in an ice bath before isopropylamine (185 mg, 3.14 mmol, 1.2 equiv) was

added as a solution in dichloromethane (3 mL) dropwise over 20 minutes. The mixture was stirred for

three days at room temperature. The reaction was quenched with water and extracted with

dichloromethane. The combined organic extracts were dried over magnesium sulfate, filtered, and

concentrated to yield a white solid. The soluble portions of the mixture were taken up in diethyl ether

and the insoluble phosphoramide byproducts were removed via filtration. The solvent was removed in

vacuo and the crude product was isolated. Flash chromatography on silica gel (0-50% ethyl acetate in

hexanes) gave the product (540 mg, 51%).

Si Si BH - Me S O O Boc 3 2 O Boc N N N N H H

2-2 2-3

To an oven-dried 2-necked round-bottom flask equipped with a magnetic stirbar and condenser was added 2-2 (530 mg, 1.32 mmol, 1 equiv) in 15 mL tetrahydrofuran (final concentration 0.05 M).

Borane-dimethylsulfide complex (500 mg, 5.68 mmol, 5.0 equiv) was added dropwise via syringe and the reaction was heated at 65 °C for 5 hr. The flask was placed on an ice bath and excess hydride was quenched by addition of methanol until bubbling stopped. The mixture was concentrated to an oil and evaporated three times with methanol to remove excess B(OMe)3. The oil was redissolved in a mixture

of aqueous sodium potassium tartrate (10 mL, 0.5 M) and methanol (10 mL) and the resulting slurry was

heated at reflux (85 °C) for 12 hours to disrupt the boron-nitrogen complex. Volatiles were removed under reduced pressure and the resulting aqueous mixture was extracted with ethyl acetate. The combined organic extracts were washed once with brine, dried over magnesium sulfate, filtered, and concentrated to provide the desired amine c as a colorless oil (527.6 mg, crude yield 103%). The material was taken on without purification.

O Si Cl O Boc Si F N N O Boc N O2N O N H F 2-4 2-3 Et3N O2N

To a flame-dried round-bottom flask equipped with stirbar and purged with nitrogen was added

2-3 (427 mg, 1.10 mmol, 1.0 equiv) in dichloromethane (final volume 15 mL, concentration 0.04 M). The solution cooled in an ice bath and triethylamine (556 mg, 5.49 mmol, 5.0 equiv) and 2-fluoro-5- nitrophenylacetic acid chloride (598 mg, 2.75 mmol, 2.5 equiv, prepared as described4) were added via

syringe. The reaction mixture took on a dark red-orange color. The vessel was warmed to room

temperature and allowed to stir overnight. The reaction was quenched with water and extracted with

dichloromethane. The combined organic extracts were dried over magnesium sulfate and concentrated

to yield the crude product. The material was purified by flash chromatography on silica gel (0-50% EtOAc

in hexanes) to give the product 2-4 (526 mg, 84%).

Si O Boc O N N N a. NH TBAF 4F, O O b. NaH N F O N 2 Boc 2-4 2-5 O2N

To a flame-dried 50 mL flask that had been purged with nitrogen was added ammonium fluoride

(171 mg, 4.61 mmol, 5.0 equiv) followed by 2-4 (526 mg, 0.922 mmol, 1.0 equiv) in tetrahydrofuran (14 mL, final concentration 0.07 M) and tetrabutylammonium fluoride (1.21 g, 4.61 mmol, 5.0 equiv). The mixture was stirred at room temperature overnight. The reaction was quenched with saturated aqueous ammonium chloride. The organic and aqueous layers were separated and the aqueous layer was extracted with ethyl acetate. The combined organic layers were washed with acetic acid solution (1.0 M in water) and brine and dried over magnesium sulfate. The mixture was filtered and the filtrate was concentrated. Flash chromatography on silica gel (20-60% EtOAc in hexanes) yielded the deprotected alcohol (215 mg). This intermediate was dissolved in THF (15 mL, final concentration 0.03 M) and placed in a flame-dried round-bottom flask with stirbar that had been purged with nitrogen. Sodium hydride (94 mg, 2.36 mmol, 5.0 equiv) was added in one portion and the resulting mixture was allowed to stir overnight at room temperature. The reaction was quenched with not-quite-saturated aqueous ammonium chloride solution. The organic and aqueous layers were separated and the aqueous layer was extracted with ethyl acetate. The combined organic extracts were dried over magnesium sulfate and

filtered. The filtrate was concentrated to yield the crude product. Flash chromatography on silica gel (50-

70% EtOAc in hexanes) afforded the product as a white crystalline solid (122 mg, 30%).

1 H NMR (300 MHz, CDCl3, 22 °C) δ 8.18 (d, J = 2.6 Hz, 1H), 8.06 (dd, J = 2.5, 8.8 Hz, 1H), 7.09 (d, J

= 8.8 Hz, 1H), 4.68-4.41 (m, 1H), 4.29-3.95 (m, 1H), 3.78-3.29 (m, 5H), 2.98-2.85 (m, 4H), 1.74-1.60 (m,

13 1H), 1.40 (s, 9H), 1.30-1.09 (m, 6H), 0.98 (d, J = 6.9 Hz, 3H). C NMR (300 MHz, CDCl3, 22 °C) δ 162.8,

144.0, 131.4, 127.7, 123.9, 123.2, 80.2, 52.3, 51.5, 37.4, 36.2, 28.3, 21.2, 19.8, 15.8. HRMS (ESI) calc’d for

+ C22H33N3O6 [M+Na] : 458.2262. Found: 458.2269.

O O

N N H2, Pd/C O O

N N O N H N 2 Boc 2 Boc 2-5 2-6

To a flame-dried flask equipped with a stirbar was added 10% palladium on carbon (28.2 mg, 26

µmol, 0.1 equiv), followed by purging of the flask with nitrogen. 2-5 (115 mg, 256 µmol, 1.0 equiv) was added as a solution in ethanol (11 mL, final concentration 0.025 M). The mixture was heated to 35 °C and a hydrogen atmosphere was applied via balloon. The reaction was allowed to progress for two hours, after which the mixture was removed from heat and filtered through Celite and the solvent was removed in vacuo to afford the crude product (95 mg, crude yield 88%). The material was taken on crude.

O O O C N N N O O O N N N N H N Boc 2 Boc H H 2-6 2-7

To a flame-dried 5 mL flask with stirbar that had been purged with nitrogen was added 2-6 (9.2 mg, 23 µmol, 1.0 equiv) and 500 µL dichloromethane. Cyclohexyl isocyanate (12.8 mg, 102 µmol, 4.5 equiv) was added and the mixture was stirred for four days. Extra dichloromethane was periodically added to replace that which evaporated over the course of the reaction. The reaction was quenched with water and extracted with methylene chloride. The combined organic extracts were dried over sodium sulfate; the solution was decanted and concentrated in vacuo to afford crude product. Flash chromatography on silica gel (0-10% MeOH in dichloromethane) provided the product (10.4 mg, 86%).

O O N TFA N O O O O N N N N H H Boc N N H H H 2-7 2-8

To a small vial equipped with a stirbar was added 2-7 (10.4 mg, 20 µmol, 1.0 equiv) in dichloromethane (600 µL, 0.02 M). Trifluoroacetic acid (444 mg, 3.89 mmol, 200 equiv) was added and the mixture was stirred for 30 minutes. The mixture was then concentrated in vacuo and mixed with saturated aqueous sodium bicarbonate. The resulting aqueous mixture was extracted with dichloromethane; the combined organic extracts were dried over sodium sulfate. The solution was

decanted and the solvent was removed by evaporation under a stream of dry nitrogen. The material was

taken on crude.

O O

N 1. 4-pyridinecarboxaldehyde, N O 2% AcOH in DMF O O O N 2. Na(OAc)3BH N N N H N N H H H H 2-8 2-9 N

To a vial equipped with a stirbar was added 2-8 (9.6 mg, 22 µmol, 1.0 equiv) dissolved in 2% acetic acid in dimethylformamide (612 µL, 0.036 M). Isonicotinaldehyde (5.7 mg, 53 µmol, 2.4 equiv) was added via syringe and the reaction was allowed to proceed for 90 minutes. Sodium triacetoxyborohydride (9.0 mg, 42 µmol, 1.9 equiv) was added and the mixture was stirred for two days. The solvent was removed in vacuo at 50°C and the residue was taken up in saturated aqueous sodium bicarbonate solution. The aqueous mixture was extracted with dichloromethane; the combined organic extracts were dried over sodium sulfate. The solution was decanted and concentrated in vacuo. The residue was purified by high- pressure liquid chromatography to yield the product (4.9 mg, 42%)

1 H NMR (300 MHz, CDCl3, 22 °C) δ 7.47-7.39 (m, 2H), 7.12-6.89 (m, 4H), 6.84-6.73 (m, 1H), 5.45 (s,

1H), 4.78 (s, 1H), 4.26-4.04 (m, 1H), 3.82-3.65 (m, 2H), 3.61-3.44 (m, 3H), 3.41-3.28 (m, 1H), 3.21-2.99 (m,

1H), 2.85-2.70 (m, 1H), 2.63-2.50 (m, 1H), 2.47-2.31 (m, 1H), 2.22 (s, 3H), 1.97-1.74 (m, 5H), 1.69-1.41 (m,

+ 6H), 1.36-1.13 (m, 6H), 1.10-0.89 (m, 3H). HRMS (ESI) calc’d for C30H43N5O3 [M+Na] : 544.3258. Found:

544.3267.

2.11.5 Characterization of additional BRD5667 analogs

Compound 2-11 was synthesized via intermediate 2-10 analogously to the synthesis of 2-9

through intermediate 2-5, with the exception that methyl amine instead of isopropyl amine was coupled

to 2-1.

N O

N N O2 Boc 2-10

1 H NMR (500 MHz, CDCl3, 22°C) δ 8.17 (s, 1H), 8.03 (d, J = 8.8 Hz, 1H), 7.07 (s, 1H), 4.11-4.01 (m,

1H), 3.89-3.74 (m, 2H), 3.67-3.52 (m, 2H), 2.99-2.92 (m, 6H), 2.17 (s, 2H), 1.42 (s, 9H), 1.11-0.86 (m, 4H).

13 C NMR (500 MHz, CDCl3, 22°C) δ 162.8, 143.9, 131.3, 127.2, 124.3, 122.1, 80.3, 64.3, 36.4, 30.6, 28.4,

+ 20.4, 19.1, 17.3, 13.7. HRMS (ESI) calc’d for C20H29N3O6 [M+Na] : 430.1949. Found: 430.1953.

N O O N N N H H N 2-11

1 H NMR (300 MHz, CD3OD, 22°C) δ 8.47 (s, 2H), 8.20 (s, 1H), 7.46 (s, 2H), 7.19 (m, 2H), 4.39 (m,

1H), 4.16-4.01 (m, 2H), 3.67-3.55 (m, 3H), 2.97 (s, 3H), 2.82-2.68 (m, 7H), 2.30 (s, 3H), 1.94-1.89 (m, 2H),

13 1.77-1.72 (m, 2H), 1.65-1.61 (m, 1H), 1.41-0.99 (m, 8H). C NMR (300 MHz, CD3OD, 22°C) δ 157.6, 151.9,

149.8, 137.5, 131.9, 125.6, 123.8, 63.0, 44.2, 40.4, 36.8, 34.5, 26.7, 26.0. HRMS (ESI) calc’d for C28H39N5O3

[M+Na]+: 516.2945. Found: 516.2954.

2.12 References

(1) Law, J. M.; Stark, S. C.; Liu, K.; Liang, N. E.; Hussain, M. M.; Leiendecker, M.; Ito, D.; Verho, O.; Stern, A. M.; Johnston, S. E.; Zhang, Y.-L.; Dunn, G. P.; Shamji, A. F.; Schreiber, S. L. "Discovery of 8- Membered Ring Sulfonamides as Inhibitors of Oncogenic Mutant Isocitrate Dehydrogenase 1." ACS Med. Chem. Lett. 2016. DOI: 10.1021/acsmedchemlett.6b00264

(2) Clemons, P. A.; Bodycombe, N. E.; Carrinski, H. A.; Wilson, J. A.; Shamji, A. F.; Wagner, B. K.; Koehler, A. N.; Schreiber, S. L. "Small molecules of different origins have distinct distributions of structural complexity that correlate with protein-binding profiles." Proc. Natl. Acad. Sci. U. S. A. 2010, 107, 18787. DOI: 10.1073/pnas.1012741107

(3) Clemons, P. A.; Wilson, J. A.; Dancik, V.; Muller, S.; Carrinski, H. A.; Wagner, B. K.; Koehler, A. N.; Schreiber, S. L. "Quantifying structure and performance diversity for sets of small molecules comprising small-molecule screening collections." Proc. Natl. Acad. Sci. U. S. A. 2011, 108, 6817. DOI: 10.1073/pnas.1015024108

(4) Marcaurelle, L. A.; Comer, E.; Dandapani, S.; Duvall, J. R.; Gerard, B.; Kesavan, S.; Lee, M. D. t.; Liu, H.; Lowe, J. T.; Marie, J. C.; Mulrooney, C. A.; Pandya, B. A.; Rowley, A.; Ryba, T. D.; Suh, B. C.; Wei, J.; Young, D. W.; Akella, L. B.; Ross, N. T.; Zhang, Y. L.; Fass, D. M.; Reis, S. A.; Zhao, W. N.; Haggarty, S. J.; Palmer, M.; Foley, M. A. "An aldol-based build/couple/pair strategy for the synthesis of medium- and large-sized rings: discovery of macrocyclic histone deacetylase inhibitors." J. Am. Chem. Soc. 2010, 132, 16962. DOI: 10.1021/ja105119r

(5) Boehm, J. S.; Zhao, J. J.; Yao, J.; Kim, S. Y.; Firestein, R.; Dunn, I. F.; Sjostrom, S. K.; Garraway, L. A.; Weremowicz, S.; Richardson, A. L.; Greulich, H.; Stewart, C. J.; Mulvey, L. A.; Shen, R. R.; Ambrogio, L.; Hirozane-Kishikawa, T.; Hill, D. E.; Vidal, M.; Meyerson, M.; Grenier, J. K.; Hinkle, G.; Root, D. E.; Roberts, T. M.; Lander, E. S.; Polyak, K.; Hahn, W. C. "Integrative genomic approaches identify IKBKE as a breast cancer oncogene." Cell 2007, 129, 1065. DOI: 10.1016/j.cell.2007.03.052

(6) Popovici-Muller, J.; Saunders, J. O.; Salituro, F. G.; Travins, J. M.; Yan, S.; Zhao, F.; Gross, S.; Dang, L.; Yen, K. E.; Yang, H.; Straley, K. S.; Jin, S.; Kunii, K.; Fantin, V. R.; Zhang, S.; Pan, Q.; Shi, D.; Biller, S. A.; Su, S. M. "Discovery of the First Potent Inhibitors of Mutant IDH1 That Lower Tumor 2-HG in Vivo." ACS Med. Chem. Lett. 2012, 3, 850. DOI: 10.1021/ml300225h

(7) Berg, J.; Tymoczko, J. L.; Stryer, L. Biochemistry; 6 ed., 2006.

(8) Manganese and its Role in Biological Systems; Sigel, A.; Sigel, H., Eds.; CRC Press, 2000; Vol. 37.

Chapter 3

Development of BRD2879, a Cell-Active Inhibitor of

Mutant IDH1

This chapter is adapted from the following publication, reused with permission:

consultation with Dr. Yan-Ling Zhang and conducted with the assistance of Dr. Oscar Verho. Follow-up assays were designed in consultation with Dr. Zhang and conducted with the assistance of Norah Liang.

BRD2879 and most of its analogs were synthesized by Sebastian Stark under my supervision; other analogs were synthesized by Ms. Liang or by me. The cell-based assay was originally developed by Dr.

Daisuke Ito and Dr. Stephen Johnston, was optimized by Dr. Mahmud Hussain, Dr. Matthias Leiendecker, and me; and all final measurements were performed by me.

3.1 High-throughput screening identifies BRD2879 as a promising lead

In this chapter, I report the use of optimized screening conditions to discover BRD2879, a potent and cell-active inhibitor of IDH1-R132H with a markedly different structure from previously reported probes. The IDH1-R132H-Mg2+ complex was screened in duplicate against 89,093 compounds from the

Broad Institute’s DOS screening library using assay conditions optimized as described in Chapter 2.

Primary screening at 15 μM yielded 551 positives with ≥ 60% inhibition in both replicates (Figure 3.1).

I retested positives in 8-point dose in the primary screening assay and in an orthogonal enzymatic

assay detecting NADPH by absorbance to confirm compound activity and exclude detection-specific

artifacts. I then supervised Ms. Liang in the testing of compounds for selectivity with respect to wild-type

IDH1. Wild-type IDH1 inhibition was measured using an assay analogous to that used for the primary screen, measuring the production of NADPH from NADP+ and isocitrate in a diaphorase-coupled reaction.

Notably, only 15 of the positives from this screen inhibited wild-type IDH1 with an IC50 below 50 μM, and

none of these showed an IC50 below 20 µM. This allele-selectivity is consistent with that seen for most previously published mutant IDH1 inhibitors and is likely due to the substantial differences in tertiary structure between wild-type and R132H-mutant IDH1.2 45

Figure 3.1: Replicate plot for the screen for IDH1-R132H inhibition. Compounds in blue caused >60% enzyme inhibition in both replicates and were selected for follow-up analysis.

To prioritize the 103 confirmed hits for follow-up investigation, I examined preliminary structure- activity relationships (SARs) present in the screening data as well as biological activity annotations in

PubChem as a readout of compound selectivity. The DOS screening library consists of many groups of structural analogs for a given scaffold, including nearly all possible stereoisomers of each compound. This design enables the identification of series that display SAR suggestive of a specific molecular interaction with the protein target. Through this process, I identified BRD2879, an 8-membered sulfonamide containing 3 stereocenters in (2S, 5R, 6R) configuration which displayed an IC50 10- to 1000-fold lower

than those of its stereoisomers (Figure 3.2A, B). This stereochemistry–activity pattern was consistent

among structurally related compounds in the screening collection.

Figure 3.2: BRD2879, but not its stereoisomers, potently inhibits IDH1-R132H in vitro. (a) Structure of BRD2879 (b) The (2S, 5R, 6R) configuration shows >10x lower IC50 than its stereoisomers. Values are geometric means of three independent experiments. (c) Activity of IDH1-R132H after incubation with compounds. BRD2879 inhibits IDH1-R132H with comparable potency and Hill slope to previously disclosed AGI-5198. Values are mean ± SD of three independent experiments.

BRD2879 and other DOS-derived compounds have been tested in a variety of screens conducted through the NIH Molecular Libraries Program, and the data can be analyzed for evidence of promiscuous activity.3 BRD2879 was inactive in all 40 screening assays for which results are available on PubChem, including a screen for inhibitors of KDM4C, which like IDH1 uses α-KG as a cofactor. Overall, the potency

of BRD2879, its encouraging preliminary SAR, and its lack of activity in other screens led me to prioritize

this scaffold for follow-up experiments.

3.2 Resynthesis of BRD2879 and exploration of structure-activity relationships

To confirm the activity of BRD2879 and explore SAR of the scaffold, we resynthesized the probe

along with 36 analogs and tested their activity in an enzymatic assay. The core structure 3-1 was

synthesized according to published procedure,4,5 and elaboration to the final compounds was accomplished by N-capping and Sonogashira reactions in either order, followed where appropriate by functionalization of the primary alcohol (exemplified in Figure 3.3). A two-step boc deprotection avoiding strong acid was used to avoid concurrent removal of the PMB group. All analogs were synthesized via this route with the exception of the ‘reverse amide’ 3-36, which required a modified synthesis of the structural core. I confirmed the activity of resynthesized BRD2879 and analogs in dose by monitoring the kinetic decrease in fluorescence of NADPH when incubated with IDH1-R132H and α-KG (Figure 3.2c). In comparison to the enzymatic assays used in screening, this direct fluorescence detection method allowed me to evaluate compounds at lower enzyme concentrations. The assay modification was necessary to maintain [inhibitor] >> [enzyme] and ensure that IC50 is a good approximation of binding affinity.

Our SAR studies focused on the alkyne (R1) and urea (R2) side chains, as these are facile points of

diversification designed into the DOS library (Figure 3.4). Recalling the stereochemical SAR for BRD2879,

we synthesized analogs retaining the (2S, 5R, 6R) configuration. Activity is maintained when the R1 side

chain is any of a variety of large hydrophobic groups, with the 4-fluorophenyl substituent (3-3) showing a slight improvement over BRD2879. Compounds with smaller or more polar substituents in this position show reduced activity. Several analogs were synthesized with the 3-pyridyl substituent in this position in order to improve solubility, but this modification significantly reduced potency.

OPMB OPMB O O O O S N S N a,b,c

O Br O N N F N 3-1 Boc 3-2 O H

OH N O O O O S N S N d e,f,g O O N N F N F N O H O H 3-3 3-4

Figure 3.3: Elaboration of the sulfonamide scaffold. Reagents and conditions: (a) 1) TBSOTf, 2,6- lutidine, DCM, rt, 1 h. 2) HF-pyr, THF, rt, 0.5 h. (b) cyclohexyl isocyanate, TEA, DCM, rt, 0.5 h, 86% (2 steps). (c) 1-ethynyl-4-fluorobenzene, TEA, XPhos-Pd-G3, MeCN, 70 °C, 14 h, 94%. (d) DDQ, DCM, pH 7 buffer, rt, 5 h, 66%. (e) PPh3, DIAD, phthalimide, THF, 0-23 °C, 18 h. (f) methylhydrazine, EtOH, 80 °C, 1 h, 36% (2 steps). (g) CH2O, Na(OAc)3BH, DCM, rt, 6 h, 51%.

OH Cmp R1 R2 IC50 (µM) Cmp R1 R2 IC50 (µM) O O S N NH NH F O 0.057 3-3 O 0.037 BRDD2879 F R1 O F N " NH " 3-5 O 0.080 3-20 0.091 R2

" NH " 3-6 O 1.0 3-21 0.056 Cl OH O O " NH Cl " S N 3-7 O 0.55 3-22 0.060

O N " 3-8 NH 1.4 F " NH O 3-23 0.028 O F 3-9 " NH 0.16 3-36 O N " IC50 = 9.1 µM 3-24 0.26 F " NH 3-10 O 3.1 3-25 " 0.14 N NH OPMB " 3-11 O 11 O " O O 3-26 N 0.32 S N N N " O O 3-12 N 19 N O " N NH 3-27 0.091 O " MeO 3-13 O 0.36 3-37 3-28 " 0.075 IC50 = 0.096 µM 3-14 " 0.44 O 3-29 " 0.19 NH 3-15 S 0.16 N " N 3-30 N 1.6 O N O S N " N 3-16 O 5.7 " O 3-31 0.090 N F NH " O 3-17 O 7.9 " O 3-32 6.4 N " 3-4 3-18 " 7.4 3-33 0.14 Me3Si IC50 = 0.032 µM N H 3-34 " 7.7 3-19 " O 1.6 " O 3-35 Br >20

Figure 3.4: Biochemical potency of BRD2879 analogs. Values are geometric means of at least 3 independent experiments; typical CV is 35%

Variations of the R2 substituent indicate a clear preference for the cyclohexyl urea group, and decreasing the size of the substituent leads to decreases in potency (3-5—3-8). Aromatic or hydrophilic residues are poorly tolerated. The urea linkage is essential, as replacement with an amide in either orientation (3-13, 3-36) reduces activity, as does N-methylation (3-16). These data suggest the urea may be a site for specific interaction of the compounds with mutant IDH1. Varying the alcohol in the R3 position did not significantly alter potency in the small set of analogs tested (3-4, 3-37), suggesting an opportunity for future modification or conjugation at this site. The SAR of BRD2879 is summarized in Figure 3.5.

Figure 3.5: Summary of SARs uncovered for BRD2879.

3.3 Validation of BRD2879 specificity through biophysical and cell-based assays

To provide further evidence that BRD2879 inhibits IDH1-R132H through direct binding of the protein, I examined the effect of the compound on the melting temperature of recombinant IDH1-R132H as measured by differential scanning fluorimetry. The 3.7 °C melting point shift observed in the presence of BRD2879 is consistent with a direct binding interaction and is equivalent to the shift seen in the presence of AGI-5198. I then performed steady-state kinetics studies to determine the mechanism of action of BRD2879, finding that the compound is competitive with both the α-KG substrate and the Mg2+

cofactor (Figure 3.6), and that binding of BRD2879 and AGI-5198 is mutually exclusive (Figure 3.7).

Figure 3.6: Mechanism of action determination for BRD2879 with respect to (A) the α-KG substrate and (B) the Mg2+ cofactor, using a method appropriate for tight-binding inhibitors as described by 6 Copeland. Rising IC50 values with increasing substrate or cofactor concentration indicate competitive inhibition. One representative experiment shown of three independent experiments.

Figure 3.7: Evidence that binding of IDH1-R132H by BRD2879 and AGI-5198 is mutually exclusive. Vij is normalized enzyme velocity in the presence of both inhibitors. The equal slopes of the [BRD2879] 6 vs 1/vij trend lines at various concentrations of AGI-5198 indicate mutually exclusive binding. One experiment is shown of two independent experiments.

To confirm the activity of BRD2879 in a more physiologically relevant context, I used an improved

version of the cell-based model system for mutant IDH1 activity described in Chapter 2. To recap, I use

HA1E-M cells engineered to express high levels of R-2HG, which is released into the growth media. I

determined the cellular activity of compounds by measuring R-2HG present in conditioned media by liquid

chromatography-mass spectrometry (LC-MS) after 72 hours of compound treatment. In this version of the

assay, I change the media after 24 hours of treatment to wash out residual R-2HG generated before the

compounds have a chance to take effect. After harvesting media, I also determined viability of compound-

treated cells by observing cell morphology and measuring ATP levels. In this model, AGI-5198 causes dose-

dependent reduction in R-2HG levels with an EC50 of 0.06 µM, consistent with previous reports of its

7 efficacy in cells. I found that BRD2879 also reduced R-2HG levels, though at lower potency (EC 50 = 0.3

μM) as compared to its biochemical activity (Figure 3.8A).

Figure 3.8: BRD2879 is effective in HA1E-M cells. (A) BRD2879 treatment causes dose-dependent reduction in R-2HG secreted by cells, though at lower potency than AGI-5198. (B) Cell viability, as measured by ATP levels, is maintained at doses up to 10 µM but drops off at higher doses. Measurements were taken after 72 hours of treatment. Values represent percentages normalized to DMSO-treated control samples, mean ± SD from three independent experiments, each run in triplicate.

To determine the toxicity of BRD2879, I measured intracellular ATP levels and observed cell morphology of our engineered HA1E-M cells and the AML-derived cell lines U937 and THP1 after three days of compound treatment (Figure 3.8B, Figure 3.9). By these measures, BRD2879 began to show inhibitory effects on cell viability at the 10 μM concentration required for near-complete suppression of

R-2HG production. U937 and THP1 are wild-type for IDH1, and IDH1-R132H inhibition would not be expected to affect the viability of engineered HA1E-M cells because the cell line was already capable of growth and proliferation before the IDH1-R132H enzyme was introduced. Thus, the observed toxicity is likely due to off-target effects and is possibly related to the compound’s poor solubility.

Figure 3.9: Effects of BRD2879 on viability of AML cell lines wild-type for IDH1 (U937, THP1), as a measure of off-target toxicity. Viability is determined by measuring ATP levels after 72 hours of treatment. The ability of BRD2879 to suppress R-2HG production in HA1E-M cells is shown for comparison. Values represent percentages normalized to DMSO-treated control samples, mean ± SD from three independent experiments, each run in triplicate.

To assess further the suitability of BRD2879 for in vivo use, I obtained data concerning several relevant physical properties of the probe, either through my own measurements or by outsourcing measurements to a contract research organization (Figure 3.10). The rapid degradation of the compound by mouse and human liver microsomes indicates optimization of the compound for metabolic stability will be required before in vivo use is possible. Additionally, the compound’s low solubility and high logD are liabilities even in cell-based model systems, as the solubility is barely sufficient to allow an efficacious dose in solution. I synthesized a small number of analogs in an attempt to improve solubility of the probe,

but these modifications either reduced potency (3-25) or failed to improve solubility as expected (3-4),

indicating the need for further effort in this area.

a In vitro enzyme inhibition, IC50 IDH1-R132H 0.05 μM IDH1-R132C 2.5 μM IDH1-WT >20 μM IDH2-R140Q >20 μM Thermal stabilizationb IDH1-R132H 3.7 °C IDH1-R132C 1.4 °C c Cell-based R-2HG suppression, EC50 HA1E-M-IDH1-R132H 0.3 μM Mechanism of inhibition w.r.t. α-KG competitive w.r.t. Mg2+ competitive Solubility/Lipophilicity, pH 7.4 Thermodynamic solubility 5 μMd LogD 5.19e f Microsomal stability, t1/2 Mouse 0.7 min Human 0.7 min Plasma protein bindingd Mouse 96.3% Human 99.5%

Figure 3.10: Key properties of BRD2879. Notes: aGeometric mean of at least three independent experiments. bMean of 3 independent experiments, each with 7 replicates. cMean of 3 independent experiments, each with 3 replicates. dMean of 3 replicates in one experiment. eSingle experiment, includes 1% DMSO. fSingle experiment, calculated from a 6-point curve over 1 hour.

Although the high molecular weight, lipophilicity, and low solubility of BRD2879 raise concerns that it may inhibit IDH1-R132H by nonspecific aggregation, I found that its activity in vitro is unaffected by increasing concentrations of Tween 20 detergent (Figure 3.11). Furthemore, the low activity of BRD2879’s enantiomer suggests that the compound’s activity may rely on specific interactions with the target rather than simply its physical properties. The thermal stabilization of purified enzyme, lack of activity against wild-type IDH1 and across many other assays, and ability to suppress R-2HG production in cells provide further evidence for this hypothesis.

Figure 3.11: Effects of the detergent Tween 20 on IDH1-R132H enzyme activity and on enzyme inhibition by BRD2879. The lack of an effect of Tween 20 on the IC50 of BRD2879 at concentrations below its critical micelle concentration (CMC) is evidence against compound aggregation as a mechanism of enzyme inhibition.8 Values are mean ± SD of three independent experiments.

3.4 Prospects for the use of BRD2879 as a probe compound

While BRD2879 is of limited utility in its present form, exploration of the SAR has revealed sites

which are not critical for compound potency and could be modified to improve solubility, selectivity, and

susceptibility to metabolism. BRD2879 represents a new structural class of mutant IDH1 inhibitors which,

with optimization, may prove useful in the study of this enzyme and its role in cancer.

3.5 Experimental methods: biology and biochemistry

3.5.1 Assay development and screening for inhibitors of IDH1-R132H – Mg2+ conditions. (“Primary

screen”)

The ability of IDH1-R132H to convert α-KG to R-2HG using NADPH is assayed by measuring consumption of NADPH in a diaphorase-coupled assay. Briefly, resazurin is a dark blue reagent that has little intrinsic fluorescence. In the presence of NADPH, resazurin is reduced by diaphorase to resorufin,

which is highly fluorescent with an excitation peak at 579 nm and an emission peak at 584 nm. The specific

assay conditions are as follows: IDH1-R132H enzyme stock (10 mg/ml) and substrates (NADPH & α-KG) are diluted in assay buffer containing 25 mM Tris-HCl pH 7.5, 150 mM NaCl, 10 mM MgCl2, 0.1 mM 1,10-

phenanthroline, 0.1% PEG3350, and 0.01% Tween 20. To a 1536-well plate (Aurora Biotechnologies cat#

00019180BX) was added 5 nL of 10 mM compound in DMSO per well via acoustic transfer. To each well

was added 2.5 μL buffer solution containing enzyme (20 μg/mL IDH1 -R132H) and 44 μM NADPH, then plates were incubated for 60 min at RT. The reaction was initiated by addition of 2.5 μL buffer solution containing 1.2 mM α-KG, followed by 50 min incubation at RT, at which time the reaction is still in the linear range. Plates were read on an Envision plate reader before the addition of detection mix to identify fluorescent compounds that could be read as false positives in the assay. The reaction was then terminated by the addition of 2.5 μL of detection mix (1 μg/mL diaphorase, 60 μM resazurin) and fluorescence (excitation 535 nm, emission 595 nm) was measured again on the Envision plate reader. AGI-

5198 was used as a positive control on each plate to validate the assay. We screened 89,093 compounds in duplicate, of which 551 compounds that inhibit ≥60% enzyme activity in both replicates in the primary screen were selected for further analysis.

3.5.2 Absorbance assay for IDH1-R132H (used to confirm screening positives), Mg2+ conditions

The buffer composition was identical to that of the primary screen and used Mg2+ as the metal

cofactor. To a solution of 40 μL/well IDH1-R132H enzyme in buffer was added compound in DMSO; compound and enzyme were incubated for 20 minutes before addition of a substrate mix containing α-

KG and NADPH. Compounds were plated in 8-point dose (3x dilution starting at 16.7 μM) in clear 384-well plates (Corning #3640). The final concentration of reagents was 4.4 μg/mL IDH1-R132H, 0.6 mM α-KG, 20

μM NADPH, and 0.17% DMSO. After addition of substrates, the absorbance of NADPH at 340 nm was

monitored via Spectramax M5 plate reader. The slope of the linear portion of the absorbance vs. time

trace was used to determine enzyme activity.

3.5.3 Fluorescence assay for IDH1-R132H (used for SAR studies)

Direct fluorescence of NADPH was used to measure IDH1-R132H inhibition by tight-binding compounds in order to reduce the amount of enzyme required such that the IC50 determined in the assay provided a more accurate measure of compound potency. The buffer consisted of 25 mM Tris-HCl, 150 mM NaCl, 10 mM MgCl2, 0.1 mM 1,10-phenanthroline, and 0.1% PEG3350, in aqueous solution with pH

7.5. No detergent was used in the assay in order to maximize enzyme activity. To a solution of 40 μL

enzyme in black-walled 384-well plates (Corning #3711 or #3575) was added 100 nL of compound in DMSO

by pin transfer; compound and enzyme were incubated for about 10 minutes before the reaction was

started by addition of substrates NADPH and α-KG in 20 μL buffer. The final concentration of reagents was

0.5 μg/mL IDH1-R132H, 0.6 mM α-KG, 7.5 μM NADPH, and 0.17% DMSO. After addition of substrates, the

progress of the reaction was monitored by the fluorescence of NADPH (ex. 340 nm, em. 465 nm) on a

Spectramax M5 plate reader. The slope of the linear portion of the fluorescence vs. time trace was used

to determine enzyme activity.

3.5.4 Fluorescence assay for other IDH alleles (IDH1-R132C and IDH2-R140Q)

Inhibition of IDH1-R132C and IDH2-R140Q by compounds was determined by a direct NADPH fluorescence assay analogous to that used for IDH1-R132H SAR studies (see above), with the following modifications. For the R132C assay, the reaction mixture contained 2 mM MgCl2 and 0.34 mM α-KG due

to the lower Km for these cofactors of the R132C allele compared to the R132H allele. The IDH2-R140Q

assay used the same cofactor concentrations as the IDH1-R132H assay. IDH1-R132C was tested at 2.8

µg/mL enzyme concentration and IDH2-R140Q was tested at 1.7 μg/mL enzyme concentration.

3.5.5 Use of PubChem to assess compound selectivity

The public PubChem database (pubchem.ncbi.nlm.nih.gov) was queried for CID 54619248, the

record code for BRD2879. Biological test results were last observed in section 6.1 on 6/14/2016. The

KDM4C (aka GASC1) assay is AID 720574. Note that while the compound is listed as inactive in AID 624101

“Development of IDH1/2 inhibitors,” this in vitro assay was conducted by us using a Mn2+ cofactor and proved not to be predictive of in vitro activity using a Mg2+ cofactor nor of activity in cells.

3.5.6 Cell-based assay: measurement of R-2HG in conditioned media

This method is improved from that described in Chapter 2. HA1E-M-IDH1-R132H cells were counted and plated in 96-well tissue-culture plates (Corning 3904) at a density of 10,000 cells/well. The plates were incubated to allow the cells to adhere and grow to confluence (about 36 hours). Media was aspirated and immediately replaced with 100 μL/well media containing compound (DMSO concentration

0.3%). Compound-treated media was aspirated and replaced after 24 hours of treatment to wash out any

R-2HG which was secreted before compounds took effect. After 72 hours of treatment (i.e. 48 hours after last media change), 60 μL/well media was removed and added to 800 μL aqueous 80% methanol solution containing deuterated racemic 2HG (2HG-d4) as an internal standard. The methanol solution was evaporated at 40°C under reduced pressure or a stream of dry nitrogen and the resulting residue was resuspended in a solution of 9 mM ammonium hydroxide in 68% acetonitrile/22% methanol/10% water.

This mixture was briefly vortexed and sonicated to encourage dissolution of R-2HG, then centrifuged to remove solids. Targeted LC-MS data were then acquired using a Waters 2795 separations module and

Waters 3100 mass detector. R-2HG was separated using a Luna NH2 column (50 x 2 mm, 3 μm beads, cat. no. 00B-4377-B0, Phenomenex, Torrance, CA) and eluted using a 5-min linear gradient initiated with 90% mobile phase B (10 mM ammonium hydroxide in 75% acetonitrile/25% methanol) and concluding with

100% mobile phase A (aqueous solution of 20 mM ammonium hydroxide and 20 mM ammonium acetate).

2HG is measured using selected-ion monitoring in the negative ion mode at M/Z 147 (2HG) and 151 (2HG- d4). The ratio of ion counts at these M/Z is used to determine the relative amounts of R-2HG in various conditioned media samples.

3.5.7 Determination of cell viability

Refer to Section 2.10.5

3.5.8 Determination of U937, THP1 cell viability

U937 or THP1 cells were obtained from ATCC and seeded at a density of 10,000 cells/well in 96- well opaque-wall plates in 100 µL RPMI media, supplemented with 10% fetal bovine serum and penicillin/streptomycin. These conditions leave the cells ample space and nutrients to continue growth.

Cells were incubated at 37 °C for 72 hours, after which viability was determined as described above.

3.5.9 Thermal shift assay by differential scanning fluorimetry (DSF)

In a 20 μL volume on a clear LightCycler 384-well plate (Roche cat# 05102430001), 5 μM compound was incubated with 125 μg/mL (3 μM) IDH protein of the specified allele and 1:600 dilution of

SYPRO orange (Molecular Probes cat# S6650) in aqueous buffer containing 25 mM Tris-HCl, 150 mM NaCl, and 0.1% PEG3350 at pH 7.5. The plate was heated from 25 °C to 95 °C over 20 minutes while SYPRO orange fluorescence was measured on a Roche LightCycler 480 II. The protein melting point was determined using LightCycler 480 Protein Melting software.

3.5.10 Mechanism of action studies

BRD2879’s mechanism of enzyme inhibition was investigated using steady-state kinetics using the method described by Copeland for tight-binding inhibitors.6 Assay conditions were identical to the IDH1-

R132H fluorescence assay used for SAR studies (see above), except the substrate or cofactor under investigation was varied in concentration. Enzyme was treated with BRD2879 in dose in the presence of 61

varying concentrations of substrate/cofactor ranging from 0.2 – 10x the substrate’s/cofactor’s Km. The change in the inhibitor’s IC50 with increasing substrate concentration is diagnostic of the mechanism of action.

3.5.11 Calculation of dose-response curves

All dose-response curves and IC50/EC50 values were generated/calculated with GraphPad Prism version 6 or 7 using four-parameter nonlinear regression, with curve top and bottom fixed at values associated with positive and negative control treatments. Curves were then inspected by eye for plausibility.

3.6 Experimental methods: chemistry, with compound characterization

3.6.1 General analytical methods, compound purity determination, and HRMS

Refer to Sections 2.11.1—2.11.3

3.6.2 Synthesis of 4 (Scheme 1)

OPMB OPMB O O O O S N 1) TBSOTf, 2,6-lutidine, DCM, rt, 1 h S N

Br 2) HF-pyr, THF, rt, 0.5 h O Br O TEA, DCM, rt, 0.5 h N 3) CyNCO, N 3-1 Boc 3-1a N O H

To an oven-dried, argon-filled flask containing 3-1 (715 mg) in dry DCM (11.4 mL, 0.1 M) was added 2,6-lutidine (537 mg, 584 μL, 4.4 eq) and tert-butyldimethylsilyl trifluoromethanesulfonate (843 mg, 733 μL, 2.8 eq). The mixture was stirred for 1 hour at room temperature, then quenched with saturated aqueous ammonium chloride. The organic and aqueous phases were separated and the aqueous phase was extracted 3x with ethyl acetate. The combined organic extracts were dried over

magnesium sulfate, filtered, and concentrated to a yellow oil. This oil was dissolved in THF (5.7 mL, 0.2 M)

and transferred to a polypropylene tube, and hydrogen fluoride-pyridine (23 mg, 1.02 eq) was added. The mixture was stirred until gas evolution ceased (5 min). The reaction was quenched with saturated aqueous sodium bicarbonate, extracted 3x with ethyl acetate, dried over magnesium sulfate, filtered, and concentrated. The residue was dissolved in DCM and transferred to a dry flask under nitrogen. Cyclohexyl isocyanate (185 mg, 189 μL, 1.3 eq) and triethylamine (184 mg, 254 μL, 1.6 eq) were added and the mixture was stirred for 25 minutes at room temperature. The reaction was quenched with saturated aqueous ammonium chloride, the aqueous and organic layers were separated, and the aqueous layer was extracted 3x with DCM. The combined organic layers were dried over magnesium sulfate. The crude material was purified by silica flash column chromatography using a solvent gradient of 0-5% methanol in

DCM, with 1% triethylamine additive. 640 mg of 3-1a was isolated in 86% yield and identified by LC-MS.

OPMB OPMB O TEA, O O 1-ethynyl-4-fluorobenzene, O S N XPhos-Pd-G3, MeCN, 70°C, 14 h S N

Br O O N N 3-1a N F N O H 3-2 O H

To a vial containing 3-1a (158 mg) in acetonitrile (2.4 mL, 0.1 M, degassed by bubbling argon) was

added 1-ethynyl-4-fluorobenzene (100 mg, 3.4 eq). This mixture was transferred via syringe to a vial containing XPhos Pd G3 mesylate (20.4 mg, 0.1 eq, Sigma-Aldrich, cat# 7633814) under nitrogen and the reaction was started by addition of triethylamine (489 mg, 674 µL, 20 eq., degassed by bubbling argon).

The reaction was stirred overnight at 70 °C, then quenched with pH 7 aqueous phosphate buffer. Organic and aqueous phases were separated and the aqueous phase was extracted 3x with ethyl acetate. The combined organic extracts were dried over anhydrous sodium sulfate, decanted, and concentrated. The 63

material was purified by flash silica chromatography using a solvent gradient of 0-5% methanol in DCM.

157 mg 3-2 was recovered in 94% yield.

1H NMR (300 MHz, CDCl3, 27 °C) δ 7.84 (d, J = 8.2 Hz, 1H), 7.56-7.47 (m, 2H), 7.29 (dd, J = 8.2, 1.6

Hz, 1H), 7.12-7.00 (m, 5H), 6.87-6.79 (m, 2H), 4.39 (d, J = 7.7 Hz, 1H), 4.33-4.22 (m, 2H), 4.15-4.03 (m, 2H),

3.91 (dd, J = 15.6, 5.2 Hz, 1H), 3.77 (s, 3H), 3.82-3.59 (m, 3H), 3.56-3.43 (m, 2H), 3.03 (dd, J = 14.7, 2.5 Hz,

1H), 2.72 (s, 3H), 2.26-2.10 (m, 1H), 2.09- 1.90 (m, 2H), 1.70-1.49 (m, 3H), 1.33 (d, J = 6.6 Hz, 3H), 1.39-

1.20 (m, 2H), 1.20-0.98 (m, 3H), 0.96 (d, J = 7.1 Hz, 3H). 13C NMR (75 MHz, CDCl3, 27 °C) δ 164.71, 161.39,

159.39, 158.02, 156.24, 135.87, 133.89, 133.78, 130.14, 129.39, 128.79, 128.53, 126.91, 126.53, 116.11,

115.82, 113.89, 91.36, 87.66, 86.57, 72.94, 72.53, 56.99, 55.40, 51.50, 51.34, 49.82, 36.72, 34.80, 34.28,

25.77, 25.31, 25.24, 16.86, 16.06.

OPMB OH O O O O S N S N DDQ, DCM, pH 7 buffer, rt, 5 h

O O N N N F N F H 3-2 O H 3-3 O

To a flask containing 3-2 (157 mg) in DCM (4.2 mL, 0.045 M) and phosphate-buffered water (0.84 mL, pH 7) was added DDQ (82 mg, 1.6 eq). The reaction was stirred vigorously under nitrogen at room temperature for 1 hour. The reaction was quenched with saturated aqueous sodium bicarbonate and the mixture was filtered through Celite. The mixture was partitioned between water and DCM, the phases were separated, and the aqueous layer was extracted 3x with DCM. The combined organic layers were dried over anhydrous magnesium sulfate, filtered, and evaporated. The material was purified by flash

silica chromatography using a solvent gradient of 0-5% methanol in DCM. 86 mg of 3-3 was isolated in

66% yield.

1H NMR (300 MHz, CDCl3, 27 °C) δ 7.87 (d, J = 8.2 Hz, 1H), 7.53-7.43 (m, 2H), 7.31 (dd, J = 8.3, 1.6

Hz, 1H), 7.17 (d, J = 1.6 Hz, 1H), 7.11-7.00 (m, 2H), 4.57 (td, J = 9.6, 2.5 Hz, 1H), 4.36-4.22 (m, 2H), 3.97-

3.78 (m, 2H), 3.74-3.59 (m, 2H), 3.51 (ddd, J = 12.2, 9.9, 4.0 Hz, 1H), 3.40 (dd, J = 15.8, 5.1 Hz, 1H), 3.19

(dd, J = 10.0, 3.0 Hz, 1H), 3.11 (dd, J = 14.4, 2.6 Hz, 1H), 2.66 (s, 3H), 2.30-2.15 (m, 1H), 2.11-2.01 (m, 1H),

2.01-1.90 (m, 1H), 1.70-1.58 (m, 1H), 1.58-1.43 (m, 2H), 1.22 (d, J = 6.7 Hz, 3H), 1.40-0.96 (m, 5H), 0.94 (d,

J = 6.9 Hz, 3H). 13C NMR (75 MHz, CDCl3, 27 °C) δ 157.75, 154.78, 134.30, 133.91, 133.79, 129.51, 129.00,

127.59, 126.74, 116.10, 115.81, 91.73, 87.38, 85.57, 64.86, 58.01, 51.63, 49.86, 48.48, 36.72, 34.50, 34.31,

+ 25.74, 25.30, 25.23, 15.77, 15.05. HRMS (ESI) calc’d for C30H38FN3O5S [M+H] : 572.2589. Found: 572.2588.

OH NH O 2 O O S N 1) PPh3, DIAD, phthalimide, THF, 0-23°C, 18 h O S N 2) methylhydrazine, EtOH, 80°C, 1 h O O N N F N O H F N O H 3-3 3-3a

To a solution of triphenylphosphine (42 mg, 3.5 eq) in THF (0.5 mL) at 0 °C was added dropwise diisopropyl azodicarboxylate (32 mg, 31 µL, 3.5 eq) and a solution of 3-3 (26 mg) in THF (0.5 mL, 0.05 M final). The reaction was allowed stirred overnight and allowed to warm to room temperature. The solvent was removed by evaporation and the mixture was dissolved in DCM and partially purified by silica flash chromatography using a solvent gradient of 0-5% methanol in DCM. The resulting material was dissolved in ethanol (0.57 mL, 0.1 M), methylhydrazine (44 mg, 50 µL, 17 eq) was added, and the mixture was stirred at 80 °C for 1 hr. Solvent was removed by evaporation under reduced pressure and the material was

purified by silica flash chromatography using a solvent gradient of 0-10% methanol in DCM. 33 mg of 3 -

3a was isolated in 36% yield and identified by LC-MS.

NH2 N O O O O S N S N CH2O, Na(OAc)3BH, DCM, rt, 6 h

O O N N F N F N O H O H 3-3a 3-4

To a solution of 3-3a (12 mg) in DCM (0.45 mL, 0.045 M) was added anhydrous magnesium sulfate

(25 mg, 10 eq) and formaldehyde 30% aqueous solution (3.7 mg formaldehyde, 11.4 µL solution, 6 eq).

The mixture was stirred for 1 hour at room temperature, followed by addition of sodium

triacetoxyborohydride (53 mg, 12 eq) and an additional 6 hours of stirring. The solvent was evaporated

and the material was purified by silica flash chromatography using a solvent gradient of 0-10% methanol in DCM. The material was further purified by HPLC to yield 6.35 mg of 3-4 (51% after HPLC).

1 H NMR (400 MHz, CDCl3, 25°C) δ 7.84 (d, J = 8.2 Hz, 1H), 7.55 – 7.44 (m, 2H), 7.28 (d, J = 1.6 Hz,

1H), 7.15 (d, J = 1.6 Hz, 1H), 7.11 – 7.01 (m, 2H), 4.43 (td, J = 9.2, 2.6 Hz, 1H), 4.35 (d, J = 7.6 Hz, 1H), 4.20

(dd, J = 14.5, 9.6 Hz, 1H), 3.76 – 3.52 (m, 3H), 3.16 (dd, J = 14.5, 2.6 Hz, 1H), 2.73 (s, 2H), 2.53 – 2.47 (m,

1H), 2.42 – 2.35 (m, 1H), 2.27 – 2.17 (m, 1H), 2.09 (s, 6H), 2.10 – 1.93 (m, 1H), 1.75 — 1.50 (m, 5H), 1.40 –

13 0.63 (m, 14H). C NMR (101 MHz, CDCl3, 25°C) δ 157.73, 155.78, 135.06, 133.69, 133.61, 129.02, 128.34,

126.68, 126.22, 115.91, 115.68, 91.17, 85.54, 77.32, 77.21, 77.01, 76.69, 64.16, 54.81, 51.52, 50.48, 49.68,

45.82, 36.61, 34.40, 34.17, 34.14, 25.57, 25.17, 25.10, 17.53, 15.50. HRMS (ESI) calc’d for C32H43FN4O4S

[M+H]+: 599.3062. Found: 599.3056.

3.7 References

(2) Yang, B.; Zhong, C.; Peng, Y.; Lai, Z.; Ding, J. "Molecular mechanisms of "off-on switch" of activities of human IDH1 by tumor-associated mutation R132H." Cell Res. 2010, 20, 1188. DOI: 10.1038/cr.2010.145

(3) Wang, Y.; Xiao, J.; Suzek, T. O.; Zhang, J.; Wang, J.; Bryant, S. H. "PubChem: a public information system for analyzing bioactivities of small molecules." Nucleic Acids Res. 2009, 37, W623. DOI: 10.1093/nar/gkp456

(5) Gerard, B.; Duvall, J. R.; Lowe, J. T.; Murillo, T.; Wei, J.; Akella, L. B.; Marcaurelle, L. A. "Synthesis of a stereochemically diverse library of medium-sized lactams and sultams via S(N)Ar cycloetherification." ACS Comb. Sci. 2011, 13, 365. DOI: 10.1021/co2000218

(6) Copeland, R. A. Evaluation of Enzyme Inhibitors in Drug Discovery; John Wiley & Sons: Hoboken, NJ, 2005.

(7) Popovici-Muller, J.; Saunders, J. O.; Salituro, F. G.; Travins, J. M.; Yan, S.; Zhao, F.; Gross, S.; Dang, L.; Yen, K. E.; Yang, H.; Straley, K. S.; Jin, S.; Kunii, K.; Fantin, V. R.; Zhang, S.; Pan, Q.; Shi, D.; Biller, S. A.; Su, S. M. "Discovery of the First Potent Inhibitors of Mutant IDH1 That Lower Tumor 2-HG in Vivo." ACS Med. Chem. Lett. 2012, 3, 850. DOI: 10.1021/ml300225h

(8) Ryan, A. J.; Gray, N. M.; Lowe, P. N.; Chung, C.-w. "Effect of Detergent on “Promiscuous” Inhibitors." J. of Med. Chem. 2003, 46, 3448. DOI: 10.1021/jm0340896

Chapter 4

Discovery of DHODH as a Target for AML Differentiation

This chapter is adapted from the following publications, reused with permission:

Sykes, D. B.; Kfoury, Y. S.; Mercier, F. E.; Wawer, M. J.; Law, J. M.; Haynes, M. K.; Lewis, T. A.; Schajnovitz, A.; Jain, E.; Lee, D.; Meyer, H.; Pierce, K. A.; Tolliday, N. J.; Waller, A.; Ferrara, S. J.; Eheim, A. L.; Stoeckigt, D.; Maxcy, K. L.; Cobert, J. M.; Bachand, J.; Szekely, B. A.; Mukherjee, S.; Sklar, L. A.; Kotz, J. D.; Clish, C. B.; Sadreyev, R. I.; Clemons, P. A.; Janzer, A.; Schreiber, S. L.; Scadden, D. T. "Inhibition of Dihydroorotate Dehydrogenase Overcomes Differentiation Blockade in Acute Myeloid Leukemia." Cell 2016, 167, 171. DOI: 10.1016/j.cell.2016.08.0571 Lewis, T. A.; Sykes, D. B.; Law, J. M.; Munoz, B.; Rustiguel, J. K.; Nonato, M. C.; Scadden, D. T.; Schreiber, S. L. "Development of ML390: A Human DHODH Inhibitor that Induces Differentiation in Acute Myeloid Leukemia." ACS Med. Chem. Lett. 2016. DOI: 10.1021/acsmedchemlett.6b003162 Collaborator Contributions: The AML differentiation cell-based assay was developed by Dr. David B. Sykes and performed with the assistance of Dr. Sykes and Katrina Maxcy. Screening and culturing of resistant

cell lines was performed by Dr. Sykes. Probes were synthesized with the assistance of Dr. Timothy A.

Lewis and Dr. Steven Ferrara. Crystallography studies of ML390 and DHODH were conducted by Dr. Joane

Rustiguel and Prof. Maria Cristina Nonato. Metabolomics was performed by Dr. Kerry Pierce using samples prepared with the assistance of Dr. Sykes or Ms. Maxcy. Gene expression analysis was performed by Dr. Mathias Wawer. In vivo experiments were performed by Dr. Sykes, Ms. Maxcy, Dr. Youmna S.

Kfoury, and Dr. François E. Mercier. The publications from which this chapter is adapted were drafted by

Dr. Lewis, Dr. Sykes, and Prof. David T. Scadden.

4.1 The role of phenotypic screening in development of differentiation therapy

In this chapter, I describe a phenotypic screen for AML differentiation leading to the discovery of dihydroorotate dehydrogenase (DHODH) as a potential therapeutic target. This phenotypic approach contrasts with that of the last two chapters, where I took advantage of human genetic evidence to launch a target-based project against IDH1. While finding evidence from human genetics is a fantastic way to discover therapeutic targets, such evidence is not always available. In AML specifically, mutations in IDH1,

IDH2, or PML-RARα provide a rationale for targeting these proteins in a subset of patients, but in the majority of cases, no such readily targetable protein can be identified through genetic studies. Therefore, a phenotypic screening approach was required to find a new therapeutic target applicable to the majority of AML cases.

Phenotypic screening comes with its own set of challenges, of which two are paramount. First is the challenge of developing a disease model system that is both simple enough to be amenable to screening

69 and complex enough to be biologically-relevant. The second challenge is identifying the target of molecules that are active in the screen. In this case, we successfully addressed both challenges, leading

to the preclinical development of a first-in-class DHODH inhibitor for the treatment of AML. This chapter focuses on the assay development and target identification strategies leading to validation of DHODH as a therapeutic target, as well as exploration of the biological mechanisms by which DHODH inhibition leads to myeloid differentiation.

4.2 Lysozyme-GFP-ER-HoxA9 cells establish a screenable model for AML differentiation

Development of a robust and accurate in vitro model of myeloid differentiation was key to the success of the phenotypic screen. In the past, myeloid differentiation has typically been assessed in AML cell lines where the mechanism of differentiation arrest is not defined.3 With such undefined models, it is unclear

whether results of the screen would be broadly applicable to the treatment of patient tumors. In addition,

high-throughput myeloid differentiation assays are challenging, as specific measures of myeloid

differentiation are cumbersome and have historically relied on morphologic or enzymatic assays4 or more

recently on changes in gene expression.5 Thus, the key challenges to solve in assay development were (a) creating a model of differentiation arrest representative of real disease and (b) developing a robust readout of differentiation suitable for high-throughput screening.

To address the first point, we used an estrogen receptor–HoxA9 (ER–HoxA9) fusion protein to immortalize cultures of primary murine bone marrow conditionally. The persistent expression of the wild- type HoxA9 protein is sufficient to enforce myeloid differentiation arrest in cultures of murine bone marrow, and injection of these cells into recipient mice leads to acute myeloid leukemia (AML), albeit with a long latency.6 Fusion of the hormone-binding domain of the human estrogen receptor (ER) to the N- terminus of HoxA9 results in a protein that is constitutively translated and sequestered in an inactive form in the cytoplasm in the absence of beta-estradiol (E2). Upon binding its beta-estradiol ligand, the ER- 70

HoxA9 protein translocates to the nucleus where it retains its wild-type activity as a transcription factor.

We used the G400V variant of the human ER which renders it insensitive to physiologic concentrations of estrogen or to the trace estrogens that are found in fetal bovine serum.7

Primary murine bone marrow cells transduced with the ER–HoxA9 construct proliferate as stem cell factor-dependent myeloblast cell lines, with the cell-surface receptor profile consistent with that of granulocyte-macrophage progenitors (GMPs). In the presence of β-estradiol and active ER–HoxA9 protein these cells proliferate indefinitely as immature myeloblasts, while upon withdrawal of β-estradiol, the cells undergo synchronous and terminal neutrophil differentiation over the course of 4-5 days, demonstrating the expected changes in cell-surface CD11b and Gr-1 expression (Figure 4.1A). This normal and terminal granulopoietic differentiation was confirmed by assaying changes in cell cycle (Figure 4.1B) and morphology (Figure 4.1C), as well as by assays of functional neutrophil effector function including phagocytosis (Figure 4.1D) and superoxide production.

In order to facilitate a small-molecule differentiation screen, an ER–HoxA9 GMP cell line was derived from the bone marrow of the lysozyme–GFP (green fluorescent protein) knock-in mouse, in which the expression of GFP is limited to mature myeloid cells.8 The step-wise pattern of gene expression during the differentiation of ER–HoxA9 cells was measured by RNA-sequencing and shows remarkable similarity with that of murine and human primary myeloblasts.

In the Lys–GFP–ER–HoxA9 cell line, GFP expression accompanied the normal process of myeloid

differentiation (Figure 4.2A) and paralleled the expression of the myeloid markers CD11b and Gr-1 (Figure

4.2B). Imaging flow cytometry showed that the single-cell morphologic changes associated with differentiation were accompanied by an increase in cell-surface CD11b staining, a decrease in cell-surface

CD117 (CKIT) staining, and an increase in cytoplasmic GFP expression (Figure 4.2C).

Figure 4.1: ER–HoxA9 cells faithfully represent a model of conditional myeloid differentiation. (A) Primary murine bone marrow cells transduced with MSCVneo–ER–HoxA9 grow as lineage-negative cells in the presence of beta-estradiol ((+) E2). Removal of E2 and inactivation of ER-HoxA9 results in the synchronous upregulation of the myeloid differentiation markers CD11b and Gr-1 as demonstrated by flow cytometry. (B) Terminal differentiation of the ER-HoxA9 cells is accompanied by exit from the cell cycle. (C) The morphologic changes that accompany myeloid differentiation are confirmed by Wright-Giemsa staining of cells in the presence and absence of E2. (D) Terminally differentiated cells, but not the undifferentiated cells, are capable of phagocytosis of fluorescently labeled E. coli.

Figure 4.2: Lysozyme-GFP-ER-HoxA9 cells produce a robust signal upon differentiation which is amenable to high-throughput screening. (A) Following differentiation upon the removal of estradiol, Lys–GFP–ER–HoxA9 GMPs upregulate GFP fluorescence, and (B) the cell-surface markers CD11b and Gr-1. (C) Imaging flow cytometry demonstrates upregulation of Lys-GFP and CD11b and downregulation of KIT expression in response to differentiation.

4.3 A high-throughput screen identifies small-molecule inducers of AML differentiation

Using the Lys–GFP–ER–HoxA9 GMPs, we performed a high-throughput small-molecule phenotypic screen to identify compounds that could trigger myeloid differentiation in the presence of active HoxA9.

After four days of compound treatment, cells were assessed by high-throughput flow cytometry9 for

viability based on forward and side-scatter properties, and for differentiation as determined by the endogenous expression of GFP and cell-surface expression of CD11b. The assay achieved a Z-factor of 0.9, indicating an excellent signal-to-noise ratio.10 We assessed the differentiation potential of more than

330,000 small molecules within the NIH Molecular Library Program’s Molecular Library Small-Molecule

Repository library over approximately 25 screening days.11

Active compounds were re-screened by flow cytometry in concentration-response experiments to eliminate toxic compounds (<10% viable cells), autofluorescent (green and/or far-red fluorescent) compounds, and estrogen antagonists. We confirmed that 12 compounds demonstrated reproducible

73 myeloid differentiation in multiple clones of ER–HoxA9 and wild-type HoxA9 murine GMP cell lines. The

12 active compounds were screened for cross-species differentiation activity in four human cell-line models of AML (HL60, NB4, THP1, and U937). Compounds 4-1 and 4-2 (Figure 4.3A, B) promoted differentiation (as assayed by upregulation of CD11b expression) in the U937 and THP1 human leukemia cell lines and were selected as starting points for compound optimization.

Figure 4.3: (A, B) Hits arising from the AML differentiation screen. (C) Both 4-1 and 4-2 show dose- response in the cell-based assay. The enantiomer 4-2* shows no activity, suggesting specific binding by 4-2.

4.4 Analysis of resistant cell lines identifies DHODH as the target

Target identification often presents a difficulty in phenotypic screening. Given that we had screened a largely un-annotated library, the protein targets of the active molecules were unknown. We used an approach of generating cell lines with acquired resistance to the differentiation effects of our small molecules in order to determine the mechanism of action, an approach that was recently successful by another lab member in the identification of a kinesin inhibitor.12 To generate resistant cell lines, the

murine Lys–GFP–ER–HoxA9 and human U937 leukemia cells were cultured in slowly escalating (5 µM to

50 µM) concentrations of DMSO, 4-1, or 4-2. Initially the treated cells grew very slowly and appeared more differentiated (based on morphology and CD11b staining) than the parental cells.

Over a period of six months, with passage every 3-4 days, resistant undifferentiated cells emerged that proliferated at the same rate and were indistinguishable from cells cultured in DMSO vehicle (Figure

4.4A). Resistant cell lines emerged in a similar time frame in both the 4-2 and 4-1-treated cultures, as well as in both murine and human cell lines. Despite their seemingly unrelated chemical structures, cells resistant to 4-2 exhibited cross-resistance to 4-1, and vice versa, suggesting a similar resistance mechanism. Furthermore, cells retained their resistance for more than six weeks after discontinuing treatment with either compound, suggesting a stable genetic alteration as the mechanism of resistance.

Figure 4.4: Resistant cell lines identify DHODH as the target of ML390. (A) ER-HoxA9 and U937 cell lines resistant to 4-1 and 4-2 were generated by continuous culture in slowly increasing concentration of compound. (B) Analysis of the whole-exome sequencing data from resistant cells revealed an increased coverage over a narrow region of chromosome 16, consistent with chromosomal amplification as the mechanism underlying increased gene expression.

We analyzed the gene expression of our four resistant cell lines by RNA-Seq, and compared the

overlap of expression changes between cells resistant to 4-1 and 4-2 as well as the overlap between

75 murine and human cells. The analysis revealed that only 8 shared genes were highly (>4-fold) upregulated among the 4-1-resistant and 4-2-resistant populations in both Lys–GFP–ER–HoxA9 and U937 cell lines.

Interestingly, these 8 transcripts were gene neighbors within the same 100 kb region of the long-arm of chromosome 16 (human) or chromosome 8 (mouse), suggesting chromosomal amplification as the mechanism of resistance. Analysis of the whole-exome sequencing (WES) data confirmed this hypothesis, demonstrating a higher degree of coverage within this region (Figure 4.4B). WES did not identify any consistent point mutations.

One of the amplified genes encoded the enzyme dihydroorotate dehydrogenase (DHODH), a critical enzyme in the intracellular de novo synthesis of pyrimidines. The enzyme is highly conserved between human and mouse (90% amino acid homology), consistent with the ability of 4-1 and 4-2 to trigger differentiation in both murine and human AML models.

4.5 Dihydroorotate dehydrogenase is a key metabolic enzyme

DHODH catalyzes the fourth step in pyrimidine synthesis, the conversion of dihydroorotate to orotate.

The enzyme localizes to the inner mitochondrial membrane and transfers electrons between dihydroorotate and complex III of the electron-transport chain via the reduction of its ubiquinone cofactor. DHODH is critical during development, and a complete lack of enzyme activity is not compatible with life (murine or human). The Miller syndrome is a rare autosomal recessive disorder in which patients have inherited hypomorphic mutations in both alleles of DHODH, resulting in multi-organ dysfunction.13

Though ubiquitously expressed, mutations in DHODH in the context of malignancy have not been

reported.

4.6 Targeting of DHODH confirmed by in vitro enzyme inhibition assay

To confirm that 4-1 and 4-2 were inhibitors of DHODH, I developed an in vitro enzyme inhibition assay

using recombinant human DHODH protein. The assay was closely based on technology previously used to

evaluate DHODH inhibitors and provides a rapid, colorimetric readout of enzyme activity (Figure 4.5).14

Figure 4.5: Chemical reactions involved in the activity assay of purified DHODH protein. As in the natural enzyme mechanism, DHODH oxidizes dihydroorotate by reducing flavin mononucleotide (FMN). Reduced FMN (FMNH2) is used to reduce decylubiquinone, an analog of Co-Q10 with superior water solubility. Reduced decylubiquinone then reductively bleaches the blue DCIP, allowing a colorimetric readout of DHODH activity.

The enzyme inhibitory activity (IC50) closely paralleled the biological differentiation effect (ED50).

Likewise, known inhibitors of DHODH including leflunomide, its active metabolite teriflunomide, and

brequinar sodium were also active in both our enzyme-inhibition and cellular differentiation assays

(Figure 4.6). In addition, 11 of the 12 top hits from the 330,000-compound MLPCN library were inhibitors of DHODH, highlighting the exquisite small-molecule binding properties of this enzyme.

Figure 4.6: The enzymatic DHODH inhibition assay reveals screening hits 4-1 and 4-2 inhibit the enzyme with IC50 near 1 µM. This potency level lies between that of validated DHODH inhibitors brequinar and teriflunomide. The enantiomer of 4-2 (4-2*) is inactive in the enzymatic assay, just as in the cell-based assay, suggesting specific, on-target binding.

4.7 Medicinal chemistry leads to ML390, a more potent differentiating agent

Two compounds with distinct chemical scaffolds (4-1 and 4-2) were chosen for further study based

on their cross-species activity in both murine and human AML models. Compound 4-1 was structurally

similar to known non-steroidal anti-inflammatory compounds (NSAIDs). However, treatment of the Lys–

GFP–ER–HoxA9 cells with a variety of confirmed NSAIDs did not result in myeloid differentiation (Figure

4.7), suggesting that this was not the mechanism of action.

Compou ER-HOX-GFP U937 THP-1 DHODH Structure a a a b nd (EC50 uM) (EC50 uM) (EC50 uM) (IC50, uM) COOH H N 4-1 3.0 ± 0.2 6.1 ± 0.4 10.7 ± 0.2 0.40 ± 0.07 N Cl COOH H N 4-1A >20 >20 >20 17.4 ± 0.7 N COOH H N 4-1B >20 >20 >20 2.8 ± 0.5 N F COOH H N

4-1C N 3.7 ± 0.3 5.6 ± 0.9 7.7 ± 0.2 0.50 ± 0.07 F3CO

COOH H N 4-1D 10 17 ± 1 >20 0.42 ± 0.09 Cl COOH O 4-1E >20 >20 >20 >20 N Cl COOH H N 4-1F >20 >20 >20 >20 Cl COOH H N 4-1G >20 >20 >20 10.0 ± 0.2 MeO COOH H F3C N flunixin N 15.9 ± 0.6 >20 >20 8 ± 1

COOH H F C N niflumic 3 >20 >20 >20 17 ± 2 acid N

COOH H Cl N clonixin >20 >20 >20 13 ± 2 N

Figure 4.7: Analogs of 4-1, including NSAIDs, showed weaker differentiation and DHODH-inhibition activity than the parent compound. Notes: (a) Calculated from combined data from 3 experiments; value ± std. error. (b) Geometric mean ± s.e.m. of at least 3 experiments.

Next, Dr. Tim Lewis initiated a medicinal chemistry study where he synthesized several structural

analogs of 4-1 and 4-2 in order to define SAR and find analogs with improved potency and properties.

Following his synthesis, I resynthesized a key compound, tested all analogs for DHODH inhibition, and tested analogs for differentiation in the Lys–GFP–ER–HoxA9, THP1, and U937 cell lines with the assistance of David Sykes.

In the compound 4-1 series, the original HTS hit compound 4-1 was the most active in both cell

differentiation and enzyme inhibition assays (Figure 4.7). Replacing the chlorine atom with hydrogen (4-

1A) or fluorine (4-1B) diminished activity, while replacement with a CF3O group (4-1C) retained activity.

Analogs where the pyridine nitrogen was replaced with a carbon (4-1D) or the anilino nitrogen of 4-1 was replaced with an oxygen (4-1E) displayed less activity. All the NSAIDs tested were inactive. The ER-HOX-

GFP cell line was always the most sensitive of the three cell lines tested whenever activity was observed.

Testing analogs of 4-2 derivatives demonstrated that the biological activity was specific to the (R)– enantiomer (Figure 4.3, 4.6). For the compound 4-2 series, removing the chlorine atom from 4-2 (4-2A) diminished activity. Replacement of chlorine by electron-donating methyl (4-2B) or methoxy (4-2C) groups gave compounds with measurable but misleading EC50 values: while the EC50s obtained differ little from those of 4-2, the maximal increase in GFP fluorescence (ER-HOX) or number of differentiated cells (U937 and THP-1) was marginal.15 Replacement of chlorine by cyano (4-2D) or fluoro (4-2E) groups led to inactive

compounds, whereas replacement by trifluoromethyl (4-2F) or trifluoromethoxy (4-2G) groups improved

activity relative to chlorine. Moving the chlorine atom from the para to the meta position (4-2H) gave a slight decrease in activity while the 3,4-dichloro analog (4-2I) improved activity relative to 4-2 in all three cell lines. The difluorodioxolane analog (4-2J) was active as well. Increasing or decreasing the chain length between the two amides by one methylene unit with the more active substituents led to a complete loss of activity in all cases tested (4-2K—4-2R). When treated with analogs of 4-2, the three cell lines tested showed roughly equal sensitivity (Figure 4.8)

The most potent compound against our engineered Lys–GFP–ER–HoxA9 cell line was compound 4-

2G. Compound 4-2G has moderate solubility (2.5 µM in phosphate buffered saline (PBS), pH 7.4), high plasma protein binding (99% murine and human) and is stable in human plasma for 5 hours and PBS for

24 hours. 4-2G was designated ML390, as part of the NIH Molecular Libraries Program, as a small-molecule probe active in overcoming differentiation arrest in a model of AML.

O O Y N N H n H X

ER-HOX- U937 THP-1 DHODH Compo X Y n GFP (EC50 (EC50 (EC50 (IC50, und uM)a uM)a uM)a uM)b 4-2 Cl H 1 4.5 5.6 3.2 1.3 ± 0.2 4-2A H H 1 NA NA NA 12.4 ± 0.4 4-2B Me H 1 8.1* 4.2* 2.7* 1.9 ± 0.2 4-2C OMe H 1 5.0* 12* 5.1* 3.6 ± 0.1 4-2D CN H 1 NA NA NA 8.9 ± 0.7 4-2E F H 1 NA NA NA 9.5 ± 0.8

4-2F CF3 H 1 3.6 2.7 2.5 0.91 ± 0.2

4-2G OCF3 H 1 1.8 8.8 6.5 0.56 ± 0.1 4-2H H Cl 1 7.9 8.8 6.5 7.4 ± 0.4 4-2I Cl Cl 1 3.2 2.4 2.5 1.5 ± 0.05

4-2J -OCF2O- 4.7 4.4 4.3 2.7 ± 0.5 4-2K Cl H 0 NA NA NA >20 4-2L Cl Cl 0 NA NA NA >20

4-2M CF3 H 0 NA NA NA >20

4-2N OCF3 H 0 NA NA NA >20 4-2O Cl H 2 NA NA NA >20 4-2P Cl Cl 2 NA NA NA >20

4-2Q CF3 H 2 NA NA NA >20

4-2R OCF3 H 2 NA NA NA >20 Brequi - - - 0.004 ± nar 0.507 0.044 0.094 0.001

Figure 4.8: Analogs of 4-2, including 4-2G, which was designated ML390 and used in further experiments. Notes: NA = No activity detected at 10 µM. * = Curve height minimal. (a) Calculated from combined data from 3 experiments; typical s.e.m. 0.03 log units. (b) Geometric mean ± s.e.m. of at least 3 experiments.

ML390 was active with an ED50 (effective concentration triggering 50% of its maximal differentiation activity) of approximately 2 µM in murine and human AML cell lines (Figure 4.9A). The differentiation triggered by incubation with ML390 was similar to the normal differentiation accompanying ER–HoxA9 inactivation as measured by imaging flow cytometry to simultaneously compare cell morphology, intracellular GFP fluorescence, and cell-surface staining (Figure 4.9B).

Figure 4.9: ML390 induces differentiation in myeloid cell models. (A) ML390 is capable of causing myeloid differentiation in murine (ER-HoxA9) and human (U937 and THP1) AML models. (B) Imaging flow cytometry demonstrates upregulation of GFP and CD11b expression as well as the downregulation of KIT expression in Lys–GFP–ER–HoxA9 cells during differentiation in the absence of estradiol (-E2) or differentiation as the result of treatment with ML390.

4.8 X-Ray crystallography of DHODH with ML390 defines binding

To further confirm DHODH as a target and elucidate detailed structural interaction of ML390 with human DHODH, I sent a sample of ML390 to my collaborator Prof. Maria Cristina Nonato, whose lab crystallized the recombinant protein both with and without ML390. Complete data sets were collected and processed to 1.70 and 1.66 Å resolution for DHODH and DHODH-ML390, respectively. The atomic coordinates of DHODH and DHODH-ML390 have been deposited in the Protein Data Bank under the accession codes 5K9D and 5K9C, respectively. Overall, the crystal structure is similar to several other reported structures of DHODH bound to small-molecule inhibitors, the interaction with Arg136 being the most important interaction (Figure 4.10).15,16

Figure 4.10: Crystal structure of DHODH-ML390. 2mFo-DFc electron density for ML390 (carbons in yellow) contoured at 1.0 σ. Dashed in blue are the hydrogen bonds formed between ML390 and Arg136; the red dashes represent the hydrogen bonds between ML390 and Thr360.

With the crystal structure in hand, I could computationally predict binding of hypothetical ML390 analogs using Glide ligand docking software by Schrodinger. A common method for increasing activity of ligands is to lock them in their binding conformation; such conformational locking reduces the entropic penalty of binding. The crystal structure indicates that ML390 binds to the enzyme in a U-shaped configuration, suggesting affinity may be increased by modifying ML390 with a ring in its central portion such that the amide substituents are cis about the ring. With this goal in mind, I computationally docked all ML390 analogs containing 3-, 4-, 5-, and 6-membered rings fused to the central linker portion of the molecule. Of all the molecules tested, only one configuration of the cyclopropyl analog showed superior

83 predicted binding affinity to ML390. In this analog, ML390’s substituents are locked in the proper

orientation by the minimal addition of bulk to the molecule (Figure 4.11).

Figure 4.11: Modifications to ML390 are predicted to improve binding to DHODH. (A) Co-crystal structure of ML390 and DHODH showing binding orientation. (B) Docking prediction for cyclopropyl- ML390 in DHODH, showing a conformational lock in the preferred orientation. The protein residues above ML390 have been removed for clarity.

I attempted to synthesize the two cis diastereomers of cyclopropyl ML390 but was unsuccessful as the compounds appeared to be unstable. A careful look at the 1H NMR spectrum of the purchased starting material, Boc-protected cis-2-aminocyclopropanecarboxylic acid, showed a lack of characteristic cyclopropyl peaks at δ < 2.0 ppm, suggesting the starting material may have rearranged even before synthesis began. With ML390 itself performing well enough for in vitro studies and other structural classes of DHODH inhibitor available for use in vivo, I deprioritized improvements to ML390 and conducted most subsequent work with the well-validated DHODH inhibitor brequinar.

4.9 Brequinar, another DHODH inhibitor, is suitable for in vivo studies

Because ML390’s low solubility and bioavailability limited its potential as an in vivo tool compound,

we evaluated the suitability of other DHODH inhibitors for in vivo studies. Brequinar sodium is a potent

inhibitor of DHODH originally developed by DuPont Pharmaceuticals (DUP 785; NSC 368390) as an anti-

proliferative agent. Brequinar inhibits DHODH activity in vitro with an IC50 of < 10 nM (Figure 4.6) and

triggers differentiation in the ER–HoxA9, U937, and THP1 cells with an EC50 of < 1 µM (Figure 4.12A, B).

The potency of brequinar is dependent on extracellular concentrations of uridine; cells cultured in 50%

FBS (to better approximate the extracellular plasma concentrations of uridine in vivo) showed a ~2-fold increase in their EC50 (Figure 4.12C).

Figure 4.12: Brequinar (BRQ) is potent in cell-based assays of myeloid differentiation. (A) BRQ is more potent than other molecules tested against ER-HoxA9 cells. (B) BRQ is more potent against human cell lines than the mouse cell line. (C) Brequinar is less potent in the presence of high serum concentration, presumably due to higher concentrations of uridine.

Brequinar has a half-life of approximately 12 hours in vivo and is highly protein-bound (98-99%), consistent with published literature.17 To help exclude the possibility that brequinar was inhibiting kinases

in addition to DHODH, we profiled brequinar against a panel of >400 known kinases (DiscoverX

KinomeScan). Brequinar showed a near-complete absence of kinase inhibitory activity at 100 nM and 1

µM concentrations.

The maximum tolerated dose (MTD) of brequinar was evaluated in wild-type C57Bl/6 mice. When

administered daily, brequinar was tolerated at doses up to 15 mg/kg. Mice receiving doses higher than 15 mg/kg exhibited both weight loss and thrombocytopenia after 6 days of daily dosing. This toxicity was reversible, and mice recovered fully following discontinuation of treatment. Measurements of plasma brequinar concentration after a single intraperitoneal (IP) dose of 15 mg/kg or 25 mg/kg suggested that an intermittent dosing schedule could also maintain concentrations above the in vitro cellular EC50 of

approximately 1 µM. Furthermore, this intermittent schedule was better tolerated, and mice given

25 mg/kg once every three days (Q3D) for 72 days showed normal weight gain and only a mild anemia

without leukopenia or thrombocytopenia.

4.10 Brequinar demonstrates anti-leukemia activity and differentiation in vivo

THP1 cells were implanted subcutaneously into the flank of SCID mice, and allowed 10 days to engraft

to tumor size of ~40 mm2. Mice were treated with vehicle control or brequinar given by IP injection at a

dosage of 5 mg/kg daily or 15 mg/kg every third day (Q3D). Brequinar slowed tumor growth at the 15

mg/kg Q3D dosage and arrested tumor growth at the 5 mg/kg daily dosage (Figure 4.13A). THP1 tumors

were explanted for differentiation analysis; THP1 cells from mice treated with brequinar exhibited marked

differentiation as evidenced by their increase in CD11b expression (depicted graphically in Figure 4.13B

and by geometric mean fluorescence intensities in Figure 4.13C).

Following these initial positive in vivo results with brequinar, my collaborators tested the compound

in a variety of additional in vivo models in order to validate DHODH as a target suitable for clinical development. Briefly, brequinar arrested tumor growth in HL60 and MOLM13 subcutaneous xenograft models of AML engrafted in NOD.SCID mice, prolonged survival in disseminated (intravenous) HL60 and

OCI/AML3 models of AML, and was active in a FLT3-ITD mutant patient-derived xenograft model of acute myeloid leukemia engrafted in recipient NSG mice.1 As my colleagues pursued in vivo validation of 86 brequinar, I focused on determining the biochemical mechanism by which DHODH inhibition leads to myeloid differentiation.

Figure 4.13: Brequinar causes differentiation and shows anti-tumor activity in an in vivo xenotransplant model of AML. (A) THP1 cells were implanted subcutaneously in the flank of SCID mice, and the mice were treated with vehicle or brequinar IP over the course of 10 days. (B) Tumors were explanted at the end of the 10 days of treatment and analyzed for the differentiation marker CD11b by flow cytometry. (C) The geometric mean fluorescence intensity of CD11b-APC expression was compared for the explanted tumors from three mice per group. (D) Metabolites from explanted tumors were extracted into methanol, and levels of intracellular uridine measured by mass spectroscopy. Data in (A) and (C) are represented as the mean ± SD.

4.11 Differentiation is caused by lack of pyrimidine metabolites

The first step in determining the mechanism of cell differentiation in response to DHODH inhibition is to determine which function of DHODH is relevant. For example, one could imagine downstream effects mediated by an overabundance of substrate dihydroorotate, a lack of pathway product uridine monophosphate (UMP), disruption of the electron transport chain in which DHODH resides, or modification of protein-protein interactions, among other possibilities. The work described in this section proves that the lack of cellular UMP and downstream metabolites is solely responsible for the differentiation effects of DHODH inhibition.

While cells depend on DHODH for intracellular UMP synthesis, they can also salvage extracellular uridine through nucleoside transporters. Supplementing media with increasing concentrations of uridine abrogated the differentiation effect of ML390, brequinar, or pyrazofurin in the Lys–GFP–ER–HoxA9, U937, and THP1 cell lines (Figure 4.14). This “uridine rescue” demonstrated that the myeloid differentiation effect was completely due to interference with UMP synthesis.

Figure 4.14: Uridine rescue provides further evidence that DHODH is the target of 4-1 and 4-2. The effects of treatment of Lys–GFP–ER–HoxA9 cells with 4-2 or its improved analog ML390 can be completely reversed by addition of uridine, the product of the pyrimidine synthesis pathway.

To establish further that the differentiation-promoting effects of DHODH inhibitors were entirely a result of uridine suppression, my colleagues and I wanted to inhibit another enzyme in the uridine

88 synthesis pathway. While DHODH catalyzes the fourth step of uridine biosynthesis, the enzyme uridine 5’- monophosphate synthase (UMPS) catalyzes the fifth and sixth steps. Pyrazofurin is a potent small- molecule inhibitor of the decarboxylase activity of UMPS,18 and treatment of the Lys–GFP–ER–HoxA9 cells

with pyrazofurin phenocopied the differentiation effect of DHODH inhibition.

To gather data on biological effects downstream of DHODH inhibition, I prepared cell samples for metabolomic analysis at the Broad Institute’s Metabolite Profiling Platform. The analysis focused on intermediates of the de novo uridine synthesis pathway as well as downstream metabolites. In vitro,

DHODH inhibition in Lys–GFP–ER–HoxA9 cells with ML390 for 48 hours led to the dramatic (>500-fold) accumulation of the immediate upstream metabolite dihydroorotate and to the depletion of uridine and

other downstream metabolites (Figure 4.15). These results provide strong evidence that DHODH is fully

inhibited in the cellular context, and that the cells have no alternative method of obtaining the original

levels of uridine metabolites.

Uridine supplementation also reversed the depletion of downstream metabolites, but did not reverse

the accumulation of DHO. Together, these findings demonstrate that inhibition of UMP synthesis at two

points along the pathway, but not the accumulation of dihydroorotate, leads to myeloid differentiation.

Thus, while dihydroorotate is a marker of enzyme inhibition, it is not an oncometabolite such as in the

case of (R)-2-hydroxyglutarate (R-2HG) in patients with IDH-mutant leukemias.19

To confirm in vivo DHODH inhibition, we performed cellular metabolite analysis of subcutaneous THP1 cells as well as HoxA9-Meis1 bone marrow leukemia cells. THP1 cells isolated from BRQ-treated mice showed a significant (90 % AUC) reduction in cellular uridine levels compared to vehicle-treated controls

(Figure 4.13D). In similar fashion, HoxA9-Meis1 leukemia cells isolated from the bone marrow of BRQ- treated mice also showed a significant reduction in cellular uridine in addition to UDP and UDP– glycoconjugates (e.g. UDP–GlcNAc, UDP–GalNAc).

Figure 4.15: Metabolite profiling of Lys–GFP–ER–HoxA9 cells shows DHODH inhibition leads to decreases in uridine and UDP-sugar conjugates and a corresponding increase in free sugars. Levels of TCA cycle metabolites citrate and isocitrate are also greatly reduced. All of these metabolic changes are reversed by addition of 100 µM uridine to the culture medium. Values are mean ± SD of 3 technical replicates of one experiment.

4.12 The mechanism of myeloid differentiation in response to uridine deprivation

With DHODH validated as a potential therapeutic target and metabolic data in hand, I became interested in the exact mechanism through which a reduction in de novo pyrimidine biosynthesis causes myeloid differentiation. DHODH is an essential metabolic enzyme, and while our experiments with

intermittent dosing suggest there could be a way to achieve therapeutic effect without major side effects,

targeting nucleic acid synthesis still entails the potential for side effects such as those seen with the DNA- damaging agents used in traditional chemotherapy. It was my hope that by determining the mechanism of differentiation in response to uridine depletion, my colleagues and I would be able to discover a therapeutic target as effective as DHODH but with less potential for toxic side effects.

Our experiments had established that depletion of uridine completely explains the differentiation

effects of DHODH inhibition. Thus, the search for the mechanism of differentiation necessarily centered

on the biological roles of uridine. The major roles of uridine are threefold: as a carrier for glycosides, as a

signaling molecule, and as raw material for DNA and RNA synthesis (Figure 4.16).

Figure 4.16: The biochemistry of uridine. Uridine is synthesized de novo from aspartate and glutamine by DHODH, among other enzymes, and can also be imported through SLC28 transporters and imported/exported through SLC29 transporters. Uridine’s major biological roles are as a ligand for P2Y receptors, as a glycoside carrier, and as raw material for RNA and DNA synthesis.

Of the many hypotheses for how uridine deprivation causes myeloid differentiation, I was most excited by those involving the glycosylating enzyme O–GlcNAc transferase (OGT) and the GPCR P2Y14, as these are small-molecule-binding proteins that could feasibly be targeted by a new drug. The enzyme OGT is a ubiquitous enzyme that transfers GlcNAc from UDP–GlcNAc to serine and threonine residues, and this

91 modification can compete with other modifications including phosphorylation in the regulation of protein function. Particularly interesting proteins that undergo GlcNAc post-translational modification include

Akt, the TET family of proteins, and c-Myc, among others noted in Figure 4.16.20,21 My colleagues

demonstrated by Western blot that inhibition of DHODH leads to a global decrease in protein N– acetylglycosylation, raising the possibility that changes in this protein modification are critical in inducing the differentiation phenotype (Figure 4.17).

Figure 4.17: Treatment of cells with ML390 or brequinar, followed by immunoblotting, demonstrates a global decrease in the degree of protein N-acetyl glycosylation (GlcNAc).

To interrogate whether decreasing OGT function is sufficient to induce differentiation in AML, I could choose among several OGT inhibitors reported in the literature. In selecting probes from the literature, 92 knowledge of a probe’s liabilities is necessary to allow the researcher to have confidence the probe will be an effective tool in the system at hand. Unfortunately, the reported OGT probes all have major

liabilities, including low potency, off-target toxicity, and poor pharmacokinetic properties. After reviewing

the probes, I chose to use 2-acetamido-1,3,4,6-tetra-O-acetyl-2-deoxy-5-thio-α-D-glucopyranose (Ac-5S-

GlcNAc), a peracetylated sulfur analog of N-acetylglucosamine.22 This probe is not ideal: it is a substrate

mimetic and must be dosed at high concentrations (50 µM), raising concerns about off-target effects on

other enzymes involved in glycosylation. However, because the probe has been validated to be stable,

cell-permeable, and not broadly toxic, I deemed it worth testing on our differentiation assay. As the

molecule is not commercially available, I synthesized the probe through an 8-step route as reported in the

literature.22 Treatment of cell-line model systems of differentiation with Ac-5S-GlcNAc resulted in no effect at low concentrations, and cytotoxicity without differentiation at the high concentrations where the compound would be expected to be effective at inhibiting OGT.

Another potential differentiation mechanism is via the autocrine signaling effects of uridine or UDP-

sugar conjugates on members of the P2Y family of nucleoside-sensing cell-surface receptors. I was

23 especially intrigued by the receptor P2Y14 due to its specificity for uridine versus other nucleosides and the because antagonizing the receptor has been shown to affect neutrophil chemotaxis, a behavior

24 specific to myeloid cells. The chemical probe PPTN is a well validated antagonist of P2Y14 active at low

nanomolar concentrations,24 and I was able obtain a sample as a gift from Dr. Kenneth A. Jacobson. Using this probe, I determined that ER-HoxA9, U937, or THP1 cells treated with PPTN were indistinguishable from control cells in viability and degree of differentiation, suggesting that uridine-induced differentiation is not mediated by P2Y14 (Figure 4.18).

To confirm the insufficiency of OGT inhibition and P2Y14 antagonism to cause differentiation, my colleagues targeted the relevant genes for CRISPR-Cas9-based knock-out in our cell-line model of

93 differentiation. Like the small-molecule treatments, these genetic perturbations were unable to cause differentiation. With two different strategies for abrogating protein function unable to produce a result, it is unlikely that OGT inhibition or P2Y14 antagonism play a key role in differentiation unless the

simultaneous modulation of other factors is required.

Figure 4.18: The P2Y14 antagonist PPTN shows no effect on three cell-line models of myeloid differentiation.

In ruling out reduced OGT-mediated protein glycosylation and reduced P2Y14 signaling as sole causes of differentiation in response to uridine depletion, hindrance of DNA/RNA synthesis was left as the most likely mechanism. This is puzzling, because while case reports suggest that low dose cytarabine in the treatment of patients with AML can induce differentiation in a limited number of cases,25 traditional DNA-

damaging chemotherapy is not thought to cause differentiation. Indeed, in our model systems, the

observation that the differentiation effect can be phenocopied by the UMPS inhibitor pyrazofurin—

another inhibitor of uridine synthesis—but not by methotrexate, cytarabine, daunorubicin, or

hydroxyurea, implicates upstream depletion of uridine as being of specific importance in the

differentiation effect, as opposed to more general DNA damage.

4.13 The differentiation effects of DHODH inhibitors are stronger than those of standard

chemotherapy

In our model systems, DHODH and UMPS inhibitors lead to depletion of uridine and to myeloid

differentiation. However, downstream inhibitors of DNA and RNA biosynthesis (methotrexate,

hydroxyurea) or DNA-damaging agents (cytarabine, daunorubicin) caused cytotoxicity without differentiation (Figure 4.19, 4.20).

Figure 4.19: Viability/differentiation plot allows comparison of compound effects on differentiation and viability in the U937 cell line. Note that DHODH inhibitors ML390 (A) and brequinar (B) greatly raise the number and proportion of differentiated cells while causing modest viability loss. Traditional chemotherapy agents methotrexate (C) and cytarabine (D) cause greater viability loss and do not increase differentiation at high doses.

Figure 4.20: The mouse ER-HoxA9 cells (A, B) provide a more stringent model of differentiation than the U937 cell line (C, D). The ER-HoxA9 cells show differentiation in response to the DHODH inhibitor ML390 but not the DNA-damage-inducing cytarabine, while the U937 cells are differentiated by cytarabine at some dose points.

A solution to the mechanistic puzzle may lie in the varied cellular responses to DNA damage: whereas severe damage caused by traditional chemotherapy provokes apoptosis, more mild damage caused by, say, a scarcity of pyrimidine nucleotides would trigger DNA repair mechanisms. Indeed, pyrimidine starvation has been shown to activate the p53 pathway,26 and AML differentiation has been associated with p53 activation in cases of APL.27

In other work, the DHODH inhibitor leflunomide was used to determine that stress from pyrimidine

nucleotide deprivation triggers expression of the stress-response transcription factor SP1 in a melanoma cell line. The resulting upregulation of HEXIM1 leads to inhibition of transcriptional elongation and suppression of cell growth, and results from zebrafish models suggest that the effect is mediated through

cell differentiation.28,29

It is intriguing to think that mild, reversible nucleotide stress could trigger differentiation-inducing

transcription factors such as p53 or SP1 without the toxicity associated with traditional chemotherapy.

Future mechanistic work will be required to determine which elements of the nucleotide stress or DNA-

damage-response machinery are involved in triggering AML differentiation in response to depletion of pyrimidine nucleotides.

Figure 4.21: Summary of the investigation of possible biological effects of uridine deprivation.

4.14 The history of DHODH as a therapeutic target

Because of its status as an essential metabolic enzyme with a large small-molecule-binding pocket, there has been past interest in pharmaceutical inhibition of DHODH as antiproliferative therapy. Much recent activity has been in the realm of anti-infectives; for example, DHODH is recognized as an anti- malarial drug target given the ability to design small molecules that specifically bind and inhibit

Plasmodium falciparum DHODH without inhibiting the human enzyme.30 Interest in targeting human

DHODH arose due to the potential for anti-inflammatory and anti-cancer effects. Thus far, two inhibitors

have been approved for anti-inflammatory indications, but DHODH inhibition as anti-cancer therapy has not been realized.

Leflunomide is a pro-drug that is effective as a disease-modifying and anti-inflammatory agent in the treatment of patients with rheumatoid arthritis.31 Its active form teriflunomide was also approved for the treatment of patients with multiple sclerosis.32 Leflunomide and teriflunomide are weak inhibitors of

DHODH (IC50 ~5 µM), are readily bioavailable, have long half-lives, and are well tolerated. They are likely to have off-target anti-kinase effects33 or possibly effects as aryl hydrocarbon antagonists.34

The potential of leflunomide as an anti-cancer agent was suggested because of its effect on the erythroid differentiation of K562 cells in vitro, which was demonstrated to be dependent on the depletion of UTP and CTP ribonucleotides.35 More recently, leflunomide was shown to be active in a zebrafish model

of melanoma where the proposed mechanism of action was one of inhibition of transcriptional

elongation, including inhibition of Myc target genes.36 The combination of leflunomide and the BRAF

inhibitor vemurafenib was proposed for a phase I/II clinical trial of patients with BRAFV600 mutant

metastatic melanoma, though the results of this trial have not been reported (Clinical trial identifier

NCT01611675).

Given these reported data, we had initially treated leukemic (HoxA9 and Meis1-expressing) mice with leflunomide. At the highest dosage, leflunomide (25 mg/kg daily) treatment resulted in a very mild increase in the expression of the differentiation marker CD11b, but did not result in a reduction in leukemic burden. Furthermore, it was poorly tolerated at this dosage, causing weight loss and lethargy in recipient mice.

Brequinar is a potent (IC50 ~20 nM) and specific inhibitor of DHODH that was developed by DuPont

Pharmaceuticals but never gained approval. Given encouraging pre-clinical activity in vitro against multiple cancer cell lines and in vivo in murine xenotransplant models, brequinar was evaluated in phase

1 and 2 trials of patients with advanced solid tumor malignancies.37-40 Brequinar was tolerated as a single agent but was not widely effective at the doses and schedules evaluated in these trials. Notably, brequinar

was not studied in the context of patients with leukemia or with other hematologic malignancies.

The lack of clinical efficacy of brequinar in previous clinical trials should be interpreted with caution.

In our model systems, sustained exposure to brequinar was required for its myeloid differentiation effect

in vitro; brequinar pulses that lasted fewer than 48 hours had almost no effect. During the human trials,

brequinar was administered in most instances as a single infusion given once every two weeks or once

every three weeks. One trial did evaluate daily dosing for 5 days but repeated only once every 4 weeks,

and this trial of 54 patients did not include any with hematologic malignancies.39 We hypothesize that these schedules would be unlikely to lead to the prolonged suppression of uridine production that would be required to kill cancer cells, or to induce AML differentiation, in vivo.

4.15 The potential of brequinar as differentiation therapy in AML

In this study, we have described how brequinar and other DHODH inhibitors triggered myeloid differentiation in vitro and in vivo, and led to the depletion of functional leukemia-initiating cells in vivo.

Wild-type mice bearing syngeneic leukemia and immunocompromised mice implanted with human xenografts (THP1) tolerated extended doses of brequinar, suggesting that there appears to be a differential sensitivity to DHODH inhibition between normal and malignant cells in vivo. This observation points to a therapeutic window in the treatment of patients with acute myeloid leukemia.

What could be the biological basis of a therapeutic window in the context of an enzyme that is ubiquitously expressed in normal and malignant cells? DHODH inhibition leads to the depletion of pyrimidine precursors and therefore inhibits nucleic acid synthesis. Unlike traditional anti-metabolite purine and pyrimidine analog chemotherapies that lead to cumulative DNA damage, DHODH inhibition results in periods of nucleotide depletion, driving a dependency on salvage pathways or autophagy. We hypothesize that the efficacy in our model systems of extended pulsatile (Q2D or Q3D) brequinar exposure results from the differential sensitivity of malignant cells as compared to normal cells to intermittent periods of nucleotide ‘starvation’. This hypothesis would also be consistent with the importance of dose schedule in our mouse model, where a high dose administered every three days demonstrates a potent anti-leukemia effect without the weight loss and thrombocytopenia that we observed with daily dosing.

The efficacy of differentiation therapy in the treatment of leukemia has been proven by the overwhelming benefits of the small molecules all-trans retinoic acid (ATRA) and arsenic trioxide (As2O3) in

those patients with PML/RAR-α translocations and acute promyelocytic leukemia. The lack of differentiation therapy for other forms of non-APL AML has been hampered by imperfect model systems.

In this study, we have described a novel phenotypic screening system that led to the unexpected identification of DHODH as a potential therapeutic target in AML, and have described the efficacy of

DHODH-inhibition in vitro and in animal models of syngeneic and xenotransplant AML.

Our work highlights the importance of phenotypic screens in identifying previously unrecognized molecular pathways relevant for normal and malignant cell biology. The in vivo efficacy of brequinar raises

100 the possibility that a better understanding of its mechanism of action will allow for a more rational and less toxic dosing schedule in future clinical trials using novel, optimized DHODH inhibitors in the treatment of patients with acute myeloid leukemia. Word on understanding the mechanism of differentiation in response to DHODH inhibition has the potential to uncover new biology, which will teach us about the differentiation process and perhaps reveal other targets for differentiation therapy.

4.16 Experimental Methods

4.16.1 Activity of purified DHODH

The enzymatic assay couples DHODH activity with bleaching of the dye 2,6-Dichlorophenolindophenol

(DCIP).41 The assay was conducted in aqueous buffer containing 50 mM Tris, 0.10% Triton X-100, 150 mM

KCl, 0.4 μg/mL DHODH, 1.0 mM dihydroorotate, 0.10 mM decylubiquinone, 0.060 mM DCIP, and 0.17%

DMSO at pH 8.0 at room temperature. Compounds were added via pin transfer and the reaction was initiated by addition of substrates. Enzyme activity was monitored kinetically by the reduction in DCIP absorbance at 600 nm. Purified recombinant human DHODH (full-length, C-terminal MYC/DDK-tag) enzyme was purchased from Origene (cat. no. TP039034). Other chemicals were purchased from Sigma-

Aldrich. Absorbance measurements were obtained using a Molecular Devices Spectramax M5 plate- reading spectrophotometer.

4.16.2 Cell culture

The ER-HoxA9 cells were maintained in RPMI supplemented with 10% fetal bovine serum, penicillin/streptomycin, and stem cell factor (SCF). The source of stem cell factor was conditioned media generated from a Chinese hamster ovary (CHO) cell line that stably secretes SCF. The conditioned media was added at a final concentration of 2% (final concentration of SCF approximately 100 ng/ml as measured

101 by ELISA). Beta-estradiol (Sigma, E2758) was added to a final 0.5 µM from a 10 mM stock dissolved in

100% ethanol. The media was stable for at least 4 weeks when maintained at 4-degrees. U937 and THP1 human AML cell lines were maintained in in RPMI supplemented with 10% fetal bovine serum and penicillin/streptomycin.

4.16.3 In vitro differentiation assay.

Lys-GFP-ER-HoxA9, U937, and THP1 cells were used in the in vitro differentiation assay. Cells (2500 to

5000) were plated in 100 µL of media in 96-well round-bottom plates. Compounds were added from

DMSO stocks using the D300 digital dispenser (Hewlett Packard / Tecan), and the cells were incubated with compound for 5 days. Cells were washed in the 96-well plate, resuspended in FACS buffer (PBS + 2%

FBS + 1 mM EDTA), and stained with CD11b-APC and propidium iodide. Samples were analyzed in 96-well format using an iQue Screener flow cytometer(IntelliCyt) and analyzed using instrument software.

4.17 References

(1) Sykes, D. B.; Kfoury, Y. S.; Mercier, F. E.; Wawer, M. J.; Law, J. M.; Haynes, M. K.; Lewis, T. A.; Schajnovitz, A.; Jain, E.; Lee, D.; Meyer, H.; Pierce, K. A.; Tolliday, N. J.; Waller, A.; Ferrara, S. J.; Eheim, A. L.; Stoeckigt, D.; Maxcy, K. L.; Cobert, J. M.; Bachand, J.; Szekely, B. A.; Mukherjee, S.; Sklar, L. A.; Kotz, J. D.; Clish, C. B.; Sadreyev, R. I.; Clemons, P. A.; Janzer, A.; Schreiber, S. L.; Scadden, D. T. "Inhibition of Dihydroorotate Dehydrogenase Overcomes Differentiation Blockade in Acute Myeloid Leukemia." Cell 2016, 167, 171. DOI: 10.1016/j.cell.2016.08.057

(2) Lewis, T. A.; Sykes, D. B.; Law, J. M.; Munoz, B.; Rustiguel, J. K.; Nonato, M. C.; Scadden, D. T.; Schreiber, S. L. "Development of ML390: A Human DHODH Inhibitor That Induces Differentiation in Acute Myeloid Leukemia." ACS Med Chem Lett 2016, 7, 1112. DOI: 10.1021/acsmedchemlett.6b00316

(3) Radomska, H. S.; Jernigan, F.; Nakayama, S.; Jorge, S. E.; Sun, L.; Tenen, D. G.; Kobayashi, S. S. "A Cell-Based High-Throughput Screening for Inducers of Myeloid Differentiation." J Biomol Screen 2015, 20, 1150. DOI: 10.1177/1087057115592220

(4) Glasser, L.; Fiederlein, R. L. "Functional differentiation of normal human neutrophils." Blood 1987, 69, 937.

102

(5) Stegmaier, K.; Ross, K. N.; Colavito, S. A.; O'Malley, S.; Stockwell, B. R.; Golub, T. R. "Gene expression-based high-throughput screening(GE-HTS) and application to leukemia differentiation." Nat Genet 2004, 36, 257. DOI: 10.1038/ng1305

(6) Kroon, E.; Krosl, J.; Thorsteinsdottir, U.; Baban, S.; Buchberg, A. M.; Sauvageau, G. "Hoxa9 transforms primary bone marrow cells through specific collaboration with Meis1a but not Pbx1b." EMBO J 1998, 17, 3714. DOI: 10.1093/emboj/17.13.3714

(7) Tora, L.; Mullick, A.; Metzger, D. "The cloned human oestrogen receptor contains a mutation which alters its hormone binding properties." The EMBO Journal 1989.

(8) Faust, N.; Varas, F.; Kelly, L. M.; Heck, S.; Graf, T. "Insertion of enhanced green fluorescent protein into the lysozyme gene creates mice with green fluorescent granulocytes and macrophages." Blood 2000.

(9) Haynes, M. K.; Strouse, J. J.; Waller, A.; Leitao, A. "Detection of intracellular granularity induction in prostate cancer cell lines by small molecules using the HyperCyt® high-throughput flow cytometry system." J Biomol Screening 2009.

(10) Zhang, J. H. "A Simple Statistical Parameter for Use in Evaluation and Validation of High Throughput Screening Assays." Journal of Biomolecular Screening 1999, 4, 67. DOI: 10.1177/108705719900400206

(11) Schreiber, S. L.; Kotz, J. D.; Li, M.; Aube, J.; Austin, C. P.; Reed, J. C.; Rosen, H.; White, E. L.; Sklar, L. A.; Lindsley, C. W.; Alexander, B. R.; Bittker, J. A.; Clemons, P. A.; de Souza, A.; Foley, M. A.; Palmer, M.; Shamji, A. F.; Wawer, M. J.; McManus, O.; Wu, M.; Zou, B.; Yu, H.; Golden, J. E.; Schoenen, F. J.; Simeonov, A.; Jadhav, A.; Jackson, M. R.; Pinkerton, A. B.; Chung, T. D.; Griffin, P. R.; Cravatt, B. F.; Hodder, P. S.; Roush, W. R.; Roberts, E.; Chung, D. H.; Jonsson, C. B.; Noah, J. W.; Severson, W. E.; Ananthan, S.; Edwards, B.; Oprea, T. I.; Conn, P. J.; Hopkins, C. R.; Wood, M. R.; Stauffer, S. R.; Emmitte, K. A.; Team, N. I. H. M. L. P. "Advancing Biological Understanding and Therapeutics Discovery with Small-Molecule Probes." Cell 2015, 161, 1252. DOI: 10.1016/j.cell.2015.05.023

(12) Chattopadhyay, S.; Stewart, A. L.; Mukherjee, S.; Huang, C.; Hartwell, K. A.; Miller, P. G.; Subramanian, R.; Carmody, L. C.; Yusuf, R. Z.; Sykes, D. B.; Paulk, J.; Vetere, A.; Vallet, S.; Santo, L.; Cirstea, D. D.; Hideshima, T.; Dancik, V.; Majireck, M. M.; Hussain, M. M.; Singh, S.; Quiroz, R.; Iaconelli, J.; Karmacharya, R.; Tolliday, N. J.; Clemons, P. A.; Moore, M. A.; Stern, A. M.; Shamji, A. F.; Ebert, B. L.; Golub, T. R.; Raje, N. S.; Scadden, D. T.; Schreiber, S. L. "Niche-Based Screening in Multiple Myeloma Identifies a Kinesin-5 Inhibitor with Improved Selectivity over Hematopoietic Progenitors." Cell Rep 2015. DOI: 10.1016/j.celrep.2015.01.017

(13) Ng, S. B.; Buckingham, K. J.; Lee, C.; Bigham, A. W.; Tabor, H. K.; Dent, K. M.; Huff, C. D.; Shannon, P. T.; Jabs, E. W.; Nickerson, D. A.; Shendure, J.; Bamshad, M. J. "Exome sequencing identifies the cause of a mendelian disorder." Nat Genet 2010, 42, 30. DOI: 10.1038/ng.499 103

(14) Knecht, W.; Loffler, M. "Species-related inhibition of human and rat dihydroorotate dehydrogenase by immunosuppressive isoxazol and cinchoninic acid derivatives." Biochem Pharmacol 1998, 56, 1259.

(15) Liu, S.; Neidhardt, E. A.; Grossman, T. H.; Ocain, T.; Clardy, J. "Structures of human dihydroorotate dehydrogenase in complex with antiproliferative agents." Structure 2000, 8, 25. DOI: 10.1016/S0969-2126(00)00077-0

(16) Baumgartner, R.; Walloschek, M.; Kralik, M.; Gotschlich, A.; Tasler, S.; Mies, J.; Leban, J. "Dual binding mode of a novel series of DHODH inhibitors." J Med Chem 2006, 49, 1239. DOI: 10.1021/jm0506975

(17) Cramer, D. V. "Brequinar sodium." Pediatr Nephrol 1995, 9 Suppl, S52. DOI: 10.1007/BF00867685

(18) Dix, D. E.; Lehman, C. P.; Jakubowski, A.; Moyer, J. D.; Handschumacher, R. E. "Pyrazofurin metabolism, enzyme inhibition, and resistance in L5178Y cells." Cancer Res 1979, 39, 4485.

(19) Ward, P. S.; Patel, J.; Wise, D. R.; Abdel-Wahab, O.; Bennett, B. D.; Coller, H. A.; Cross, J. R.; Fantin, V. R.; Hedvat, C. V.; Perl, A. E.; Rabinowitz, J. D.; Carroll, M.; Su, S. M.; Sharp, K. A.; Levine, R. L.; Thompson, C. B. "The common feature of leukemia-associated IDH1 and IDH2 mutations is a neomorphic enzyme activity converting alpha-ketoglutarate to 2-hydroxyglutarate." Cancer Cell 2010, 17, 225. DOI: 10.1016/j.ccr.2010.01.020

(20) Hanover, J. A.; Krause, M. W.; Love, D. C. "Bittersweet memories: linking metabolism to epigenetics through O-GlcNAcylation." Nat Rev Mol Cell Biol 2012, 13, 312. DOI: 10.1038/nrm3334

(21) Józwiak, P.; Forma, E.; Brys, M. "O-GlcNAcylation and metabolic reprograming in cancer." Front Endocrinol (Lausanne) 2015, 5, 145.

(22) Gloster, T. M.; Zandberg, W. F.; Heinonen, J. E.; Shen, D. L.; Deng, L.; Vocadlo, D. J. "Hijacking a biosynthetic pathway yields a glycosyltransferase inhibitor within cells." Nat Chem Biol 2011, 7, 174. DOI: 10.1038/nchembio.520

(23) Jacobson, K. A.; Balasubramanian, R.; Deflorian, F.; Gao, Z. G. "G protein-coupled adenosine (P1) and P2Y receptors: ligand design and receptor interactions." Purinergic Signal 2012, 8, 419. DOI: 10.1007/s11302-012-9294-7

(24) Barrett, M. O.; Sesma, J. I.; Ball, C. B.; Jayasekara, P. S.; Jacobson, K. A.; Lazarowski, E. R.; Harden, T. K. "A selective high-affinity antagonist of the P2Y14 receptor inhibits UDP-glucose-stimulated chemotaxis of human neutrophils." Mol Pharmacol 2013, 84, 41. DOI: 10.1124/mol.113.085654 104

(25) Wisch, J. S.; Griffin, J. D.; Kufe, D. W. "Response of preleukemic syndromes to continuous infusion of low-dose cytarabine." N Engl J Med 1983, 309, 1599. DOI: 10.1056/NEJM198312293092602

(26) Khutornenko, A. A.; Roudko, V. V.; Chernyak, B. V.; Vartapetian, A. B.; Chumakov, P. M.; Evstafieva, A. G. "Pyrimidine biosynthesis links mitochondrial respiration to the p53 pathway." Proc Natl Acad Sci U S A 2010, 107, 12828. DOI: 10.1073/pnas.0910885107

(27) Ablain, J.; Rice, K.; Soilihi, H.; de Reynies, A.; Minucci, S.; de The, H. "Activation of a promyelocytic leukemia-tumor protein 53 axis underlies acute promyelocytic leukemia cure." Nat Med 2014, 20, 167. DOI: 10.1038/nm.3441

(28) White, R.; Cech, J.; Ratanasirintrawoot, S.; Lin, C. Y.; Rahl, P. B.; Burke, C. J.; Langdon, E.; Tomlinson, M. L.; Mosher, J.; Kaufman, C.; Chen, F.; Long, H. K.; Kramer, M.; Datta, S.; Neuberg, D.; Granter, S.; Young, R. A.; Morrison, S.; Wheeler, G. N.; Zon, L. I. "DHODH modulates transcriptional elongation in the neural crest and melanoma." Nature 2011, 471, 518. DOI: 10.1038/nature09882

(29) Tan, J. L.; Fogley, R. D.; Flynn, R. A.; Ablain, J.; Yang, S.; Saint-Andre, V.; Fan, Z. P.; Do, B. T.; Laga, A. C.; Fujinaga, K.; Santoriello, C.; Greer, C. B.; Kim, Y. J.; Clohessy, J. G.; Bothmer, A.; Pandell, N.; Avagyan, S.; Brogie, J. E.; van Rooijen, E.; Hagedorn, E. J.; Shyh-Chang, N.; White, R. M.; Price, D. H.; Pandolfi, P. P.; Peterlin, B. M.; Zhou, Y.; Kim, T. H.; Asara, J. M.; Chang, H. Y.; Young, R. A.; Zon, L. I. "Stress from Nucleotide Depletion Activates the Transcriptional Regulator HEXIM1 to Suppress Melanoma." Mol Cell 2016, 62, 34. DOI: 10.1016/j.molcel.2016.03.013

(30) Phillips, M. A.; Lotharius, J.; Marsh, K.; White, J.; Dayan, A.; White, K. L.; Njoroge, J. W.; El Mazouni, F.; Lao, Y.; Kokkonda, S.; Tomchick, D. R.; Deng, X.; Laird, T.; Bhatia, S. N.; March, S.; Ng, C. L.; Fidock, D. A.; Wittlin, S.; Lafuente-Monasterio, M.; Benito, F. J.; Alonso, L. M.; Martinez, M. S.; Jimenez- Diaz, M. B.; Bazaga, S. F.; Angulo-Barturen, I.; Haselden, J. N.; Louttit, J.; Cui, Y.; Sridhar, A.; Zeeman, A. M.; Kocken, C.; Sauerwein, R.; Dechering, K.; Avery, V. M.; Duffy, S.; Delves, M.; Sinden, R.; Ruecker, A.; Wickham, K. S.; Rochford, R.; Gahagen, J.; Iyer, L.; Riccio, E.; Mirsalis, J.; Bathhurst, I.; Rueckle, T.; Ding, X.; Campo, B.; Leroy, D.; Rogers, M. J.; Rathod, P. K.; Burrows, J. N.; Charman, S. A. "A long-duration dihydroorotate dehydrogenase inhibitor (DSM265) for prevention and treatment of malaria." Sci Transl Med 2015, 7, 296ra111. DOI: 10.1126/scitranslmed.aaa6645

(31) Breedveld, F. C.; Dayer, J. M. "Leflunomide: mode of action in the treatment of rheumatoid arthritis." Ann Rheum Dis 2000, 59, 841. DOI: 10.1136/ard.59.11.841

(32) O'Connor, P.; Wolinsky, J. S.; Confavreux, C.; Comi, G.; Kappos, L.; Olsson, T. P.; Benzerdjeb, H.; Truffinet, P.; Wang, L.; Miller, A.; Freedman, M. S.; Group, T. T. "Randomized trial of oral teriflunomide for relapsing multiple sclerosis." N Engl J Med 2011, 365, 1293. DOI: 10.1056/NEJMoa1014656

105

(33) Doscas, M. E.; Williamson, A. J.; Usha, L.; Bogachkov, Y.; Rao, G. S.; Xiao, F.; Wang, Y.; Ruby, C.; Kaufman, H.; Zhou, J.; Williams, J. W.; Li, Y.; Xu, X. "Inhibition of p70 S6 kinase (S6K1) activity by A77 1726 and its effect on cell proliferation and cell cycle progress." Neoplasia 2014, 16, 824. DOI: 10.1016/j.neo.2014.08.006

(34) O'Donnell, E. F.; Saili, K. S.; Koch, D. C.; Kopparapu, P. R.; Farrer, D.; Bisson, W. H.; Mathew, L. K.; Sengupta, S.; Kerkvliet, N. I.; Tanguay, R. L.; Kolluri, S. K. "The anti-inflammatory drug leflunomide is an agonist of the aryl hydrocarbon receptor." PLoS One 2010, 5. DOI: 10.1371/journal.pone.0013128

(35) Huang, M.; Wang, Y.; Collins, M.; Mitchell, B. S.; Graves, L. M. "A77 1726 induces differentiation of human myeloid leukemia K562 cells by depletion of intracellular CTP pools." Mol Pharmacol 2002, 62, 463. DOI: 10.1124/mol.62.3.463

(36) White, R. M.; Cech, J.; Ratanasirintrawoot, S.; Lin, C. Y.; Rahl, P. B.; Burke, C. J.; Langdon, E.; Tomlinson, M. L.; Mosher, J.; Kaufman, C.; Chen, F.; Long, H. K.; Kramer, M.; Datta, S.; Neuberg, D.; Granter, S.; Young, R. A.; Morrison, S.; Wheeler, G. N.; Zon, L. I. "DHODH modulates transcriptional elongation in the neural crest and melanoma." Nature 2011, 471, 518. DOI: 10.1038/nature09882

(37) Arteaga, C. L.; Brown, T. D.; Kuhn, J. G.; Shen, H. S.; O'Rourke, T. J.; Beougher, K.; Brentzel, H. J.; Von Hoff, D. D.; Weiss, G. R. "Phase I clinical and pharmacokinetic trial of Brequinar sodium (DuP 785; NSC 368390)." Cancer Res 1989, 49, 4648.

(38) Burris, H. A., 3rd; Raymond, E.; Awada, A.; Kuhn, J. G.; O'Rourke, T. J.; Brentzel, J.; Lynch, W.; King, S. Y.; Brown, T. D.; Von Hoff, D. D. "Pharmacokinetic and phase I studies of brequinar (DUP 785; NSC 368390) in combination with cisplatin in patients with advanced malignancies." Invest New Drugs 1998, 16, 19. DOI: 10.1023/A:1016066529642

(39) Noe, D. A.; Rowinsky, E. K.; Shen, H. S.; Clarke, B. V.; Grochow, L. B.; McGuire, W. B.; Hantel, A.; Adams, D. B.; Abeloff, M. D.; Ettinger, D. S.; et al. "Phase I and pharmacokinetic study of brequinar sodium (NSC 368390)." Cancer Res 1990, 50, 4595.

(40) Schwartsmann, G.; Dodion, P.; Vermorken, J. B.; ten Bokkel Huinink, W. W.; Joggi, J.; Winograd, B.; Gall, H.; Simonetti, G.; van der Vijgh, W. J.; van Hennik, M. B.; et al. "Phase I study of Brequinar sodium (NSC 368390) in patients with solid malignancies." Cancer Chemother Pharmacol 1990, 25, 345. DOI: 10.1007/BF00686235

(41) Miller, R. W.; Kerr, C. T.; Curry, J. R. "Mammalian dihydroorotate – ubiquinone reductase complex." Canadian Journal of Biochemistry 1968, 46, 1099. DOI: 10.1139/o68-164

106

Chapter 5

Lessons on Probe Development and the Link Between

Metabolism and Differentiation in AML

Portions of this chapter are reproduced with permission from the following publication: Law, J. M.; Stark, S. C.; Liu, K.; Liang, N. E.; Hussain, M. M.; Leiendecker, M.; Ito, D.; Verho, O.; Stern, A. M.; Johnston, S. E.; Zhang, Y.-L.; Dunn, G. P.; Shamji, A. F.; Schreiber, S. L. "Discovery of 8-Membered Ring Sulfonamides as Inhibitors of Oncogenic Mutant Isocitrate Dehydrogenase 1." ACS Med. Chem. Lett. 2016. DOI: 10.1021/acsmedchemlett.6b002641

5.1 Unifying themes and lessons learned

In this last chapter, I discuss unifying themes emerging from results described throughout this

work. Most importantly, the discovery that inhibition of two unrelated metabolic enzymes can each cause

differentiation in AML cells suggests that metabolic modulation may be a new strategy for the treatment

of this disease. On a smaller scale, other lessons emerged that touch on how to run a chemical screening

project, how to choose the best probe, and how to achieve novel, therapeutically important results in a

crowded and fast-moving field.

5.2 A metabolism—differentiation link was discovered through two approaches

While the two approaches to differentiation therapy were conceptually and strategically distinct,

as the projects went on, they began to mirror each other in such a way as to reveal lessons about the biology of acute myeloid leukemia and effective approaches to its treatment. The mutant-IDH1-targeted approach described in Chapters 2-3 began with the surprising observation that mutations in the metabolic enzymes IDH1/2 seemed to cause leukemia. Later, work by others in the field showed that the presence of mutant IDH1/2 leads to a differentiation block which can be reversed by small-molecule inhibition of the enzyme. The phenotypic screening arm of the project described in Chapter 4 began with the goal of overcoming a differentiation block, and led to the discovery of a compound which overcomes this block through inhibition of DHODH, another metabolic enzyme. Together, the outcomes of the two projects suggest a profound relationship between a cell’s metabolic state and its differentiation state, with the implication that modulation of a cell’s metabolism might be exploited to cause therapeutically beneficial differentiation in diseases like AML (Figure 5.1).

This connection between metabolism and differentiation is important because metabolic enzymes, as professional small molecule binders, should be relatively easy to modulate at will with

108

artificially designed small molecules. The pharmaceutical accessibility of metabolism and its connection

to differentiation bodes well for the introduction of metabolism-targeted differentiation therapy agents to the clinic.

Figure 5.1: Concept map depicting how the four key chemical probes described in this work were used in linking metabolic enzyme inhibition to AML differentiation. Black lines indicate work done by me; grey lines indicate work done by others and are included to provide context. For both the IDH1 and DHODH projects, identification of the MOA of initial probes (BRD2879, ML390) allowed lead hopping to superior molecules found in the literature that share the same target (AGI-5198, brequinar; dashed arrows). Note the symmetry of the two projects, whereby a metabolism- differentiation link was discovered twice from opposite starting points.

5.3 The advantage of chemical biology

Developing small-molecule probes is hard work, and people sometimes question whether it’s

worth the effort. Why develop chemical inhibitors in a world with RNAi and CRISPR, genetic tools of

increasing utility which aren’t limited by the need for a protein target to have a small-molecule binding

site? There are many answers to this question: small molecules are easier to turn into therapeutics, they

109

can be given in dose, they give temporal control, they act on the protein directly rather than on nucleic

acid, and they can have effects more subtle than simple inhibition or activation. In fact, both projects

described in this thesis would not have been possible without the use of small molecules.

In developing IDH1 inhibitors, it was important to develop a probe selective against the mutant allele of the protein: indications are that inhibiting the wild-type protein will lead to toxicity, and any biological studies would be confused without the ability to distinguish between wild-type and mutant enzyme activities. With only a single base pair mutation involved, it would be difficult for genetic tools to select for wild-type vs. mutant DNA or RNA. On the other hand, with the mutation leading to a rearranged binding site, developing an allele-selective small-molecule probe was relatively straightforward.

In discovering that DHODH inhibition leads to AML differentiation, the use of small molecules was also key. As it turns out, complete knock-out of DHODH, or complete inhibition for a long period of time, leads to cell death rather than differentiation. Cells cannot survive without pyrimidine nucleotides. The molecules in our screen inhibited DHODH mostly—but not completely—leading to differentiation without massive toxicity. This conclusion was reinforced by the in vivo mouse experiments performed by my colleagues, in which daily dosing of a potent DHODH inhibitor was unacceptably toxic, but dosing every three days had the desired therapeutic effect. Here, we observed the dose-control and temporal-control advantages of small molecules in action.

5.4 Improved probes were discovered through literature search

In studying both IDH1 and DHODH, the most useful chemical inhibitor turned out to be a molecule other than the one developed by me and my coworkers. In the case of IDH1, AGI-5198 has more potency in cells and better metabolic stability than BRD2879, making it a superior choice for use in biological experiments. In the case of DHODH, brequinar is more potent than ML390 and had already been shown

110

to be tolerated in humans. In such cases, it was important to evaluate probes objectively and without bias

favoring those developed ‘in house’ (Figure 5.1).

In the case of the IDH1 project, the continuing realization that AGI-5198 was either equivalent or

superior to BRD2879 in every property I measured—potency, cell-based activity, off-target toxicity,

mechanism of action, solubility, metabolic stability, ease of synthesis—was a frustrating experience which

led me to realize it no longer made sense to continue working on the project. The differentiating factor of

BRD2879 is its DOS-based structure, but as IDH1-R132H appears to be perfectly druggable using traditional pharmaceutical compounds the DOS structure was not a significant advantage in this case.

Without access to unique compounds or biological assays, my lab no longer held a strategic role in this area. That being said, if future problems arise with IDH1-R132H-inhibiting drugs which are specific to the chemical series, BRD2879 stands ready for further optimization and development.

By contrast, the realization that a DHODH inhibitor existed that was superior to ML390 was an unqualified boon for the project. Access to brequinar, which was 100 times as potent and had already been chemically optimized for use in humans, allowed us to confirm DHODH as a differentiation therapy target and initiate in vivo studies on a much shorter timeline. With access to a unique differentiation assay through my collaboration with David Sykes and David Scadden, we could make full use of the best

available DHODH inhibitor, whatever the source.

5.5 Assay conditions are key to the identification of high-quality probes

The difference between the successful screen of IDH1-R132H leading to cell-active probes described in Chapter 3 and the less successful screen described in Chapter 2 was essentially a matter of one metal cofactor: Mn2+ versus Mg2+. That such a small change can have such a large effect on the outcome of a screen is one of the frustrations of probe development, and to some extent, I was simply a

111

victim of bad luck. But there are principles of assay design that, in hindsight, could have helped me make a better choice.

A key principle of assay design is that to maximize chances of success, assay conditions must hew as closely as possible to the native biological state. On the other hand, the assay must proceed quickly,

easily, and reproducibly in order to enable high-throughput screening. The tension between an assay

simple enough for screening and complex enough to inform true biology is present throughout the assay

development process. In choosing Mn2+ over Mg2+, my colleagues and I chose ease of screening over

faithfulness to native biology. In hindsight, we should have further explored how to get the assay to work

with Mg2+ before taking the drastic step of switching to an unnatural cofactor.

Figure 5.2: Number of active compounds in screens using Mg2+ or Mn2+ cofactors, out of 49,237 compounds tested in both screens. Only 31 compounds were active in both screens. The most potent hits from each screen, BRD2879 and BRD5667, were not detected in the other screen. Active compounds are those causing >60% inhibition of IDH1-R132H in both replicates of the Mg2+-based assay, or causing >40% inhibition in both replicates of the Mn2+-based assay.

112

The difference in results induced by a cofactor change is indeed large. Of the almost 50,000

compounds tested in both screens, most active compounds only showed up in one condition or the other,

including the best compounds from each screen (Figure 5.2). Moreover, the leads from each screen have

substantially different properties: most notably, BRD2879 from the Mg2+ screen is over 100 times as

potent in cells as BRD5667 from the Mn2+ screen (Figure 5.3).

Registration BRD5667 BRD2879

Structure

2+a b IDH1 R132H IC50 (µM), Mn 0.6 1.6

2+ c d IDH1 R132H IC50 (µM), Mg 5.5 0.05

e f HA1E-M Cells EC50 (µM) 80 0.3

a IDH1 wt IC50 (µM) >20 >20

Competition with α-KG Noncompetitive Competitive

Figure 5.3: Comparison of Leads from Mn2+ and Mg2+ screens. Notes: aGeometric mean of 2 independent experiments. bAlthough BRD2879 showed detectable activity against IDH1-R132H-Mn2+ during retesting, the compound was not discovered in the primary screen. cGeometric mean of 8 independent experiments. dGeometric mean of 13 independent experiments. eMean of 3 independent experiments, each in duplicate. fMean of 3 independent experiments, each in triplicate.

The lesson here is clear: you get what you screen for. Given that there appears to be a substantial

change in enzyme conformation and/or activity between the Mn2+- and Mg2+-bound states, it is not surprising that hits bind preferentially to the enzyme complex against which they were screened. In

Chapter 2, I described how BRD5667, discovered in a Mn2+-based screen, was more potent against the

Mn2+-containing complex, whereas AGI-5198, discovered in a Mg2+-based screen, was more potent against the Mg2+-containing complex. The metal preference is not simply a function of the chemical series, as

113

BRD2879, discovered in a Mg2+-based screen, was more potent against the Mg2+-containing complex, even though the chemical structure is closely related to that of BRD5667 (Figure 5.4).

Figure 5.4: The direction of the potency shift between Mn2+ and Mg2+ conditions depends on how the IDH1-R132H inhibitors were discovered. (A) BRD5667, discovered in a Mn2+-based screen, is more potent against the Mn2+-containing enzyme complex. (B) AGI-5198, discovered in a Mg2+-based screen, is more potent against the Mg2+-containing complex. (C) BRD2879, discovered in a Mg2+-based screen, is more potent against the Mg2+-containing complex, despite its structural similarity to BRD5667. Data shown is from one and representative of at least three experiments.

5.6 High-impact innovation requires speed—or a unique approach.

Near the beginning of my work developing an IDH1-R132H inhibitor, I was presenting my progress at the poster session of the Broad Institute Retreat when Ed Scolnick, former president of Merck Research

Laboratories, stopped by. He looked at my poster and said “If you’re going to do this, you need to do it seriously.” Scolnick knew that many pharmaceutical companies and therapeutically-oriented academic chemistry groups would be intrigued by the high-profile discovery of IDH1/2 mutations and would immediately go to work on the obvious first step: developing small-molecule inhibitors of these enzymes.

To be successful, a group developing an IDH1-R132H inhibitor would need to be first, have a strategy for developing a superior molecule, or at least have a strategy to produce a mechanistically distinct molecule.

In industry parlance, these goals are, respectively, “first-in-class,” “best-in-class,” or “next-in-class.”

114

In the case of my IDH1-R132H project, chances of developing a first-in-class molecule were low initially given that other groups had a head start, and any chance we had was ruined when the metal cofactor debacle forced me to start the project over using new assay conditions after two years of work.

My team had a better chance of discovering a best-in-class or next-in-class molecule due to our exclusive access to the DOS screening library. But as described in Chapter 3 and Section 5.5, it turned out that

BRD2879 was neither superior nor mechanistically distinct compared to AGI-5198, and with our limited medicinal chemistry resources we were unable to improve the compound significantly. Thus, BRD2879 stands as a low-impact innovation: at best a starting point for an improved probe, but more likely one of the many dead ends left behind in the march of technological progress.

While time will tell if DHODH inhibitors make a clinical impact, prospects are higher than they are for BRD2879. In this project, my colleagues had developed a unique approach to the problem of differentiation therapy: a cell-based assay which includes the key genetic feature of AML—HoxA9 upregulation—and a lysozyme-GFP reporter to aid in screening, but which is otherwise genetically clean.

Possession of this well engineered cell model and ready access to the Broad Institute’s compound screening, RNA sequencing, and metabolomics capabilities put us in a unique situation to make a high- impact discovery, even if we at times encountered problems that slowed us down. The satisfaction involved in successfully completing such is project is more than that of winning a race. Instead of making a discovery a little better or a little faster, it is a discovery that may not have been found at all without the approach my colleagues and I could pursue.

5.7 Reference

115