Structural Studies of Proteins Involved in Nucleotide Metabolism

Home , Thiopurine

Daniel Rehling Structural Studies of Proteins involved in Nucleotide Metabolism Structural of Proteins Studies involved in Nucleotide Metabolism Studies of a ribonucleotide reductase from A. aeolicus and NUDT15

Daniel Rehling

ISBN 978-91-7911-572-2

Department of Biochemistry and Biophysics

Doctoral Thesis in Biochemistry at Stockholm University, Sweden 2021

Structural Studies of Proteins involved in Nucleotide Metabolism Studies of a ribonucleotide reductase from A. aeolicus and NUDT15 Daniel Rehling Academic dissertation for the Degree of Doctor of Philosophy in Biochemistry at Stockholm University to be publicly defended on Thursday 23 September 2021 at 10.00 online via Zoom, public link is available at the department website.

Abstract This thesis is separated into two parts. The first part concerns ribonucleotide reductase from Aquifex aeolicus. A. aeolicus is a hyperthermophilic bacterium that thrives at extremely high temperatures of 80-95 °C. We present the X-ray crystal structures of both the R1 and R2 subunits of this protein, which represents the first structure of a class Ia ribonucleotide reductase from a hyperthermophile and the first structure of an R1 from the NrdAh phylogenetic subclass. Several novel features were seen in the R1 structure such as the simultaneous binding of two ATP molecules in the ATP-cone domain as well as a novel “β-hairpin hook” feature which spans the dimer interface of the R1 protein. The gene encoding the R2 protein contains a self-cleaving intein domain. We examined two constructs of this protein, one with the sequence of the intein removed at the DNA level and the wild-type construct. Both crystal structures were found to be identical, showing the efficient cleavage of the intein domain in the wild-type construct. The second part of this thesis concerns the NUDIX hydrolase NUDT15. The physiological function of NUDT15 is still unknown, however certain mutations in this gene are associated with thiopurine intolerance in patients. Thiopurines are chemotherapeutic drugs used in the treatment of diseases such as acute lymphoblastic leukemia, the most common type of childhood leukemia, and inflammatory bowel disease. Thiopurine drugs are converted by the cell to the active metabolite 6-thio-dGTP which can then act as a substrate for DNA polymerase. Incorporation of these anti-metabolites into DNA produces the desired cytotoxic effects. We show that NUDT15 breaks down the active metabolites of these drugs which leads to a lowered effective dose. The absence of a functioning NUDT15 protein in patients that have inactivating mutations in the gene coding for NUDT15 results in a drastically increased effective dose of these compounds. A normal dose of a thiopurine drug can lead to severe and possibly life-threatening complications in these patients. The role of NUDT15 in thiopurine metabolism is established by in vitro and cellular data as well as the X-ray crystal structure of NUDT15 in complex with 6-thio-GMP. Acyclovir and ganciclovir are two antiviral drugs whose mechanism of action is similar to that of thiopurines. These drugs are also metabolized to their tri-phosphorylated forms and are then preferentially incorporated into viral DNA. Here again, we use in vitro, cellular and structural data to show that NUDT15 breaks down the active metabolites of these drugs. Two separate and structurally distinct lines of potent inhibitors for NUDT15 were developed with support of crystallographic studies. We show that cells are sensitized to both thiopurine and antiviral treatments in the presence of these inhibitors. Binding of our inhibitors to NUDT15 provided substantial thermal stabilization. The stabilizing effect of inhibitor binding enabled us to solve structures of the four most clinically relevant NUDT15 variants, thus elucidating the structural basis for the thiopurine sensitivity phenotype.

Keywords: X-ray crystallography, ribonucleotide reductase, NUDT15.

Stockholm 2021 http://urn.kb.se/resolve?urn=urn:nbn:se:su:diva-195222

ISBN 978-91-7911-572-2 ISBN 978-91-7911-573-9

Department of Biochemistry and Biophysics

Stockholm University, 106 91 Stockholm

STRUCTURAL STUDIES OF PROTEINS INVOLVED IN NUCLEOTIDE METABOLISM

Daniel Rehling

Structural Studies of Proteins involved in Nucleotide Metabolism

Studies of a ribonucleotide reductase from A. aeolicus and NUDT15

Daniel Rehling ©Daniel Rehling, Stockholm University 2021

ISBN print 978-91-7911-572-2 ISBN PDF 978-91-7911-573-9

Printed in Sweden by Universitetsservice US-AB, Stockholm 2021 List of publications

This thesis is based on the following papers, which are referred to in the thesis by their roman numerals.

I. Daniel Rehling†, Emma Rose Scaletti†, Inna Rozman Grinberg†, Dan- iel Lundin, Margareta Sahlin, Anders Hofer, Britt-Marie Sjöberg and Pål Stenmark “Structural and biochemical investigation of class I ribonucleotide reductase from the hyperthermophile Aquifex aeolicus” Manuscript

II. Nicholas C. K. Valerie†, Anna Hagenkort†, Brent D. G. Page, Geoffrey Masuyer, Daniel Rehling, Megan Carter, Luka Bevc, Patrick Herr, Evert Homan, Nina G. Sheppard, Pål Stenmark, Ann-Sofie Jemth† and Thomas Helleday “NUDT15 Hydrolyzes 6-Thio-DeoxyGTP to Me- diate the Anticancer Efficacy of 6-Thioguanine.” Cancer Research 76, 5501–5511 (2016).

III. Si Min Zhang†, Matthieu Desroses†, Anna Hagenkort†, Nicholas C. K. Valerie, Daniel Rehling, Megan Carter, Olov Wallner, Tobias Kool- meister, Adam Throup, Ann-Sofie Jemth, Ingrid Almlöf, Olga Lo- seva, Thomas Lundbäck, Hanna Axelsson, Shruti Regmi, Antonio Sarno, Andreas Krämer, Linda Pudelko, Lars Bräutigam, Azita Rasti, Mona Göttmann, Elisée Wiita, Juliane Kutzner, Torsten Schaller, Christina Kalderén, Armando Cázares-Körner, Brent D. G. Page, Rosa Krimpenfort, Saeed Eshtad, Mikael Altun, Sean G. Rudd, Stefan Knapp, Martin Scobie, Evert J. Homan, Ulrika Warpman Berglund, Pål Stenmark and Thomas Helleday “Development of a chemical probe against NUDT15.” Nature Chemical Biology 16, 1120–1128 (2020).

† Authors contributed equally

i IV. Rina Nishii, Takanori Mizuno, Daniel Rehling, Colton Smith, Brandi L. Clark, Xujie Zhao, Scott A. Brown, Brandon Smart, Takaya Mori- yama, Yuji Yamada, Tatsuo Ichinohe, Makoto Onizuka, Yoshiko Atsuta, Lei Yang, Wenjian Yang, Paul G. Thomas, Pål Stenmark, Mo- tohiro Kato and Jun J. Yang “NUDT15 polymorphism influences the metabolism and therapeutic effects of acyclovir and ganciclovir” Nature communications 12, 4181 (2021)

V. Si Min Zhang, Daniel Rehling, Ann-Sofie Jemth, Adam Throup, Na- talia Landázuri, Ingrid Almlöf, Mona Göttmann, Nicholas C.K. Va- lerie, Sanjay R. Borhade, Prasad Wakchaure, Brent D. G. Page, Mat- thieu Desroses, Evert J. Homan, Martin Scobie, Sean G. Rudd, Ulrika Warpman Berglund, Cecilia Söderberg-Nauclér, Pål Stenmark and Thomas Helleday “NUDT15-mediated hydrolysis limits the efficacy of anti-HCMV drug ganciclovir” Cell Chemical Biology, in press

VI. Daniel Rehling†, Si Min Zhang†, Ann-Sofie Jemth, Tobias Koolmeis- ter, Adam Throup, Olov Wallner, Emma Scaletti, Takaya Moriyama, Rina Nishii, Jonathan Davies, Matthieu Desroses, Sean G. Rudd, Martin Scobie, Evert Homan, Ulrika Warpman Berglund, Jun J. Yang, Thomas Helleday and Pål Stenmark “Crystal structures of NUDT15 variants enabled by a potent inhibitor reveal the structural basis for thiopurine sensitivity” Journal of Biological Chemistry 296, 100568 (2021)

† Authors contributed equally ii

Additional publications

1. Nicholas C. K. Valerie, Kumar Sanjiv, Oliver Mortusewicz, Si Min Zhang, Azita Rasti, Marie F. Langelier, Daniel Rehling, Adam Throup, Matthieu Desroses, Ingrid Almlöf, Seher Alam, Johan Bos- tröm, Luca Bevc, Pål Stenmark, John M. Pascal, Thomas Helleday, Brent D. G. Page†, and Mikael Altun† “Coupling cellular target engagement to drug-induced responses with CeTEAM” Manuscript

† Authors contributed equally

iii

List of Figures

Figure 1. Structures of nucleotides...... 1 Figure 2. X-ray diffraction from a protein crystal...... 4 Figure 3. The phase diagram of protein crystallography...... 5 Figure 4. The R2 subunit of the RNR from E. coli ...... 8 Figure 5. RNR mechanism ...... 9 Figure 6. Binding sites of the a-subunit of the RNR from E. coli ...... 10 Figure 7. Allosteric regulation...... 10 Figure 8. ATP domain and the promotion of different oligomers ...... 12 Figure 9. Intein schematic mechanism...... 14 Figure 10. AaR1 crystal structure ...... 17 Figure 11. ATP-cone domain of AaR1 showing the binding of two ATP molecules...... 18 Figure 12. Structure of the AaR2 homodimer...... 19 Figure 13. Structure of the NUDT15 homodimer...... 23 Figure 14. Structures of 6-MP, 6-TG and AZA-T...... 24 Figure 15. Schematic illustration of the role of NUDT15 in thiopurine metabolism...... 26 Figure 16. Crystal structure of NUDT15 in complex with 6-thio-GMP...... 27 Figure 17. Development of TH1760...... 30 Figure 18. Structure of the 6-thioG analog NSC56456 in complex with NUDT15...... 32 Figure 19. Structure of NUDT15 in complex with TH8321...... 33 Figure 20. NUDT15 in complex with TH7755...... 34

v Figure 21. Structures of acyclovir triphosphate and ganciclovir triphosphate ...... 35 Figure 22. Structures of NUDT15 in complex with acyclovir and ganciclovir...... 36 Figure 23. NUDT15 variant locations...... 39 Figure 24. Thermal stability assay of wild-type and mutant NUDT15 in the presence of different ligands...... 40 Figure 25. Arg139 mutation site of NUDT15...... 41 Figure 26. Val18 mutation site of NUDT15...... 42

List of abbreviations

6-MP 6-mercaptopurine 6-TG 6-Thioguanine 6-Thio-(d)GTP 6-Thio-(deoxy)guanosine triphosphate Å Ångström (1 × 10!"# �) ALL Acute lymphoblastic leukemia (d)ATP (deoxy-)Adenosine triphosphate AZA Azathioprine CETSA Cellular thermal shift assay CMV Cytomegalovirus DARTS Drug affinity responsive target stability DNA Deoxyribonucleic acid GTP Guanosine triphosphate ITDRF Isothermal dose-response fingerprint MMR Mismatch repair MRC-5 Medical Research Council cell strain 5 (d)NDP (deoxy-)Nucleoside diphosphate NRF2 Nuclear factor erythroid 2-related factor 2 NUDIX Nucleoside diphosphate linked to moiety x PDB Protein Databank PPi Pyrophosphate PPase Pyrophosphatase RNR Ribonucleotide reductase ROS Reactive oxygen species SAR Structure activity relationship SNP Single nucleotide polymorphism TPMT Thiopurine methyltransferase WT Wild type

vii

viii

Contents

Introduction ...... 1 Nucleotides ...... 1 Experimental methods ...... 3 Brief primer on X-ray Crystallography ...... 3 Ribonucleotide Reductase from the hyperthermophile Aquifex aeolicus ...... 7 Ribonucleotide Reductases ...... 7 Structure and mechanism of Class Ia RNRs ...... 7 Brief overview of other RNR classes ...... 12 Ribonucleotide reductase from the hyperthermophile Aquifex aeolicus ...... 14 NUDT15 ...... 21 NUDIX Hydrolases ...... 21 NUDT15 and MTH1 ...... 22 NUDT15 breaks down thiopurine metabolites ...... 26 The first inhibitor for NUDT15 ...... 29 Further development of inhibitors for NUDT15 ...... 32 The antiviral drugs acyclovir and ganciclovir ...... 35 NUDT15 breaks down antiviral drug metabolites ...... 36 Examining the structural basis of thiopurine intolerance ...... 39 Summary and conclusions ...... 43 Popular science summary ...... 45 Populärvetenskaplig sammanfattning ...... 47 Acknowledgements ...... 49 References ...... 51

Introduction

Nucleotides All of life, at every level, can be conceived of as information processing. A central component of every information processing system is the storage of information. In living organisms on earth, this is achieved, to a large extent, by molecules of DNA. These molecules store the information needed for a cell to produce every protein used by a cell throughout its lifecycle. The structure of DNA is a double helix comprising two antiparallel polymers of nucleotides. Each of these polymers is a long chain of deoxy-ribose molecules in which the 3’ position of one ribose is connected to the 5’ position of the next ribose with a phosphate group in between forming a phosphodiester bond. Each deoxy-ribose molecule is also connected to one of four possible nucleobases through an N-glycosidic bond. The four nucleobases found in DNA are adenine, guanine, thymine and cytidine (Figure 1). These nucleobases form the center of the DNA double helix and each of these nucleobases can form base pairing interactions with one specific other complementary nucleobase. The bulky purines adenine and guanine base pair with the smaller pyrimidines cytidine and thymine.

Figure 1. Structures of nucleotides.

1 The two proteins studied in this thesis, Ribonucleotide reductase (RNR) and NUDT15, are both involved in protecting the integrity of DNA. RNR is responsible for making deoxy-ribonucleotides and ensuring that there is a balanced supply of nucleotides. The physiological function of NUDT15 is still unknown, however this enzyme interferes with the mechanism of action of a certain kind of drug. Nucleotide analogs are drugs that do not occur naturally but are structurally similar enough to nucleotides for them to be incorporated into DNA, where they produce the desired cytotoxic effects. NUDT15 prevents this from happening by breaking down the active metabolites of some of these drugs before they can be incorporated into DNA.

Experimental methods

Brief primer on X-ray Crystallography Biochemical experiments are not enough to fully understand the detailed mechanisms of proteins. To get the full picture one needs a model of the three- dimensional structure of a protein at an atomic scale. Conventional light mi- croscopes are limited to a resolution of approximately 0.2 µm. This limitation is largely explained by the fact that visible light has a wavelength of 400-700 nm. Atomic scale models cannot be achieved at this wavelength due to the diffraction limit. One can use X-ray beams with wavelengths of around 1 Å, however here one encounters a different problem. X-ray beams cannot be focused by lenses like visible light; therefore, a different technique must be used. A given sample is analyzed by placing it in the path of a powerful X-ray beam. The electrons in the sample interact with this X-ray beam and scatter photons creating a diffraction pattern which one can measure using suitable detectors. The scattering from single molecules is much too weak to be analyzed, so instead one makes use of the special properties of crystals (Figure 2A). Every crystal is made up of a regular pattern of repeating elements forming the crystal lattice. The X-ray beam interacts with all the electrons in the sample crystal and scattering photons are emitted in all directions. However, most of these cancel each other out by destructive interference. Constructive interference that gives rise to an interpretable signal is only possible when the angle between the incident beam and a given crystal plane meets the Bragg condition (�� = 2��). This is only true for a small subset of possible angles which means the diffraction pattern that emerges is a series of spots (see Figure 2B).

Figure 2. X-ray diffraction from a protein crystal. A) AaR2 crystal. B) Diffraction pattern produced by the AaR2 crystal. There is a limited set of symmetric arrangements that the individual protein molecules can assume for them to produce the indefinitely repeating three- dimensional pattern of a crystal lattice. These arrangements are categorized according to different symmetry operations into space groups and each of them produces a unique diffraction pattern. There are 230 different space groups in general. However, protein crystals are special in that the enantio- meric amino acids that make up the protein only exist in one of the two possible isomeric forms (L-amino acids). This means the mirror and inversion symmetry operations are absent in protein crystals and this reduces the number of possible space groups down to 65. In principle this diffraction pattern is all that is needed to reconstruct a model of the protein that forms the repeating unit within the crystal, with one exception. We can only measure the intensity of the light that hits the detector and any information about the phase of the light waves is lost. There are a variety of methods that can be used to obtain the phase information and the method used in all structures presented in this thesis is molecular replacement. In this case one takes the phase information of an already solved structure from a similar protein and combines this information with the experimentally determined intensities of the structure on would like to solve. In a typical X-ray crystallography project, the major bottleneck is obtaining well diffracting crystals. Many conditions must be met before this is possible. First one needs a suitable protein sample. This needs to be sufficiently pure, monodisperse and of a high enough concentration (often around 10-15 mg/ml). Many crystallization projects already fail at this step since protein purification is not always a trivial task without an already established purification protocol. This is especially true for membrane proteins. The next and often the biggest stumbling block on the way to obtaining a crystal is finding a crystallization condition that works. Crystallization conditions depend on a combination of different factors such as temperature, pH, 4 ionic strength, concentration of the protein and the presence or absence of certain chemical additives. All the crystals used to produce the data shown in this thesis were obtained using vapor diffusion crystallization. In a single trial, a reservoir in a sealed volume is filled with a particular mixture of different chemicals. Within the same enclosed volume, one places a drop of protein solution mixed with a certain ratio of the reservoir solution.

Figure 3. The phase diagram of protein crystallography. At the beginning of a crystallization project when the crystallization condition is not yet known, one must screen a great number of possible conditions. What happens in the drop of protein solution can be nicely visualized using a phase diagram shown in Figure 3. The trial starts somewhere in the phase diagram, depending on the mixture of chemicals chosen, but typically in the undersaturation zone, where there is less protein in solution than can be dissolved. Water molecules will diffuse and exchange between the protein solution drop and the reservoir solution. Because the reservoir solution has a lower concentration of water, the water molecules tend to move from the protein solution into the reservoir solution. This decreases the water concentration in the protein solution drop over time. In the phase diagram this translates to a simultaneous increase of precipitant and protein concentration. Each individual crystallization trial in each screen starts off at a different position in the phase diagram but over time that position traverses the phase diagram from the

5 undersaturation zone towards the precipitation zone. If the conditions are just right, one hits the nucleation zone where protein crystals start to form. The nucleation zone for each protein is different. Some proteins that are too flexible, do not crystallize at all, so the zone can even be nonexistent. Once protein crystals have started to form, they usually grow over time. This reduces the protein concentration in the drop but not the precipitant concentration. In the right conditions, the crystals will grow large enough for data collection at a synchrotron facility.

Ribonucleotide Reductase from the hyperthermophile Aquifex aeolicus

Ribonucleotide Reductases Ribonucleotide reductases (RNRs) are enzymes that perform the rate-limiting step of replacing the 2’-OH group of nucleoside diphosphates (NDPs) directly with a hydrogen atom, thereby turning them into deoxy-nucleoside diphosphates (dNDPs). These dNDPs can then be ATP-dependently phosphorylated by nucleoside diphosphate kinase and can subsequently be used by DNA-polymerase as building blocks for DNA in replication and repair. There are several key aspects that make this enzyme truly remarkable. Firstly, it is the only known de novo source of all four dNTPs needed for DNA. This fact has led to the hypothesis that this enzyme evolved before, and is a prerequisite for the appearance of DNA itself.1 Another aspect is the unusual reaction mechanism employed by RNRs. To reduce nucleotides, RNRs make use of a free radical. This was the first protein for which it was demonstrated that it is capable of stably storing this radical within the protein structure.2,3 The last important aspect is the complex mechanism of allosteric regulation which enables this single enzyme to ensure that the cell has a balanced supply of all four dNTPs. Even 60 years after their discovery in E. coli,4,5 interesting new discoveries in the field of RNRs are still being made to this day. What follows is a brief overview of some of the most important aspects of RNRs.

Structure and mechanism of Class Ia RNRs

Overall structure and reaction mechanism The details of the structure and mechanism of RNRs vary across different organisms and RNRs are grouped into separate classes based on these details. The first RNR to be studied was the class Ia RNR from E. coli and the RNR presented in paper I also belongs to the same class. The class Ia RNR is split into two subunits a and b. Both of these subunits come together to form ho- modimers referred to as R1 and R2 respectively. The smaller b-subunit is the radical generating subunit. This subunit has a ferritin like fold and houses a µ-

7 oxo-bridged di-iron center capable of generating a radical which is then transferred to a nearby tyrosine residue where it is stored until needed (Figure 4).3,6– 10

Figure 4. The R2 subunit of the RNR from E. coli (PDB ID: 1MXR 11). A) Structure of the R2 homodimer. B) Radical generating di-iron center.

The other, larger subunit is the a-subunit which is where the reaction takes place. The two dimers can come together to form complexes in a variety of different oligomeric states. When the complex is in the active state, a pathway opens up which allows for the transfer of the radical from the tyrosyl in the R2 subunit all the way to the catalytic site (c-site) of the R1 subunit, a distance 12–14 of over 30 Å. A structure of the active a2b2 complex showing this pathway in its entirety was recently published by Kang et al.12

Reaction mechanism at the c-site Within the c-site there are three important cysteine residues (Figure 5). One of them accepts the radical, which is shuttled to it from the radical transport pathway.13,15 This generates a transient thiyl radical which is capable of ab- stracting a hydrogen from the C-3’ position of the ribose moiety of the nucleotide substrate thus creating a substrate radical. The remaining two cysteines form a pair that is able to reduce the substrate radical, forming a disulfide bridge in the process while splitting off the OH group at the C-2’ position.16– 18 The radical is then transferred back to first cysteine restoring the thiyl radical.1

The cysteine pair must be reduced again before the next reaction can occur. This is facilitated by either of two protein systems: the thioredoxin system or the glutaredoxin system.19,20

Figure 5. RNR mechanism (adapted from Greene et al. 2020 21).

Allosteric regulation at the specificity site The nucleotide pool in cells is comparably small. Mammalian cells for example only have enough nucleotides to sustain S-phase DNA replication for a few minutes.1 The supply of nucleotides must not only be continuously re- plenished but it is also crucially important that the relative abundance of the nucleotides must be tightly controlled and kept in balance. Imbalances in the nucleotide pool have been shown to lead to increased mutation rates and cell death.22–25 This balancing is achieved entirely by the a-subunit where the affinity of the c-site is modulated depending on which effector is bound to a nearby allosteric site, called the specificity site (s-site).13,26–32

Figure 6. Binding sites of the a-subunit of the RNR from E. coli (PDB ID: 4R1R 33). One monomer of the E. coli homodimer is shown in orange, the other in light orange. C-site is highlighted in pink, s-site in blue and a-site in red. The effectors that can bind in the s-site are mainly dNTPs, the end products of this metabolic pathway, but also to a lesser extent ATP. The mechanism by which effector binding influences the c-site can be explained structurally. Two flexible loops (loop 1 and loop 2) are located between the s-site and the c-site. Binding of different dNTPs to the s-site directly influences the shape of loop 2, with loop 1 acting in a supportive fashion (Figure 7).31

Figure 7. Allosteric regulation. A) close-up view of loop 2 located between the s-site and c-site in the a-subunit from the RNR from E. coli (PDB ID: 6W4X 12). B) Sche- matic overview of the effects of binding of the different allosteric effectors.

Since loop 2 also forms part of the c-site, conformational changes in this loop directly modulate the affinity of the c-site for the different NDP substrates.27,34,35 A schematic overview of substrate affinity in the presence of different effectors is shown in Figure 7B. An illustrative example of how substrate affinity modulation by the effectors leads to a balanced supply of dNTPs is given by the sequence of events that take place in the cell after removal of hydroxyurea which inhibits class I RNRs.28 With the dNTP pool depleted after RNR inhibition, dTTP is the only dNTP still in relative abundance because it can be generated via salvage reactions.28,36 As seen in the schematic, dTTP bound to the s-site increases the affinity of the c-site for GDP, stimulating dGTP production. Once enough dGTP is in the cytosol, it competes the dTTP out of the s-site and in turn increases the affinity of the c-site for ADP. At high concentrations the resulting dATP will bind in a separate allosteric site, discussed below, but at low concentrations dATP will bind in the s-site and pro- mote CDP binding and dCTP production. This sequence of events restores the dNTP levels back to optimal levels for the cell.28

Allosteric regulation at the activity-site In addition to the c-site and s-site, many RNRs also have a third nucleotide binding site called the activity-site (a-site). This site is in a separate domain (ATP-cone domain) that is typically, but not always, located at the N-terminus of the a-subunit.37 As the name implies, this site acts as a general activity regulator for the enzyme.29,38,39 Binding of either ATP or dATP in this site promotes the assembly of different oligomeric states of the protein. These oligomeric states differ among organisms but in the case of the class Ia RNR from E. coli, binding of ATP leads to the formation of the active a2b2 complex, while binding of dATP promotes the assembly of the larger, ring-shaped, inactive a4b4 complex. The a4b4 complex is not simply an assembly of two a2b2 complexes. The two a and b subunits are arranged differently relative to each other in the two complexes. In the a4b4 complex, this shift disrupts the radical transfer pathway between the two subunits, causing the inactivation of the enzyme.

Figure 8. ATP domain and the promotion of different oligomers of the RNR complex from E. coli. a-subunits colored in orange and light orange, b-subunits colored in marine and slate blue. (PDB ID of active complex: 6W4X 12. PDB ID of inactive complex: 5CNS 40.)

The recently solved structure of the a2b2 complex confirmed that it was in fact asymmetrical and not symmetrical as originally proposed in the docking model that had been the working hypothesis for many years (Figure 8).12,13

Brief overview of other RNR classes Up to this point the discussion has been focused on class Ia RNRs as was first discovered in E. coli. However, the field of RNRs is large and there are many examples of RNRs whose structures and functional mechanisms differ markedly from what has been discussed. RNRs are classified according to their relationship with oxygen.8,27,41 Class I RNRs require oxygen for the generation of the radical. Class III RNRs however, are found in anaerobic organisms and these RNRs are inactivated in the presence of oxygen.42–46 In fact in class III RNRs, the glycyl radical is inactivated by oxygen and this leads to the cleavage of the a peptide bond within the protein, thereby destroying the protein.47 Their radical generating mechanism is therefore also very different from what is found in class I RNRs and uses an 4Fe-4S cluster and S-adenosyl methio- nine.1 They also directly reduce NTPs to produce dNTPs.42,48 Many bacteria carry genes for multiple classes of RNRs. For example E. coli can express two different class I RNRs (class Ia and class Ib, described later) as well as a class III RNR. This helps the organism to thrive under many different environmental conditions.49 Class II RNRs are different again. They neither require oxygen, nor are they inactivated by it. In these enzymes the radical generation and catalytic

12 activity all occurs in the same protein and is not split into two subunits as in class I RNRs. In this case the radical site is not a tyrosyl but an adenosyl co- balamin cofactor.50 Within this class there are also examples of RNRs that use NDPs as substrates and others that use NTPs. Another important classification in RNRs is the mechanism of radical generation and storage. Many different strategies have been found, particularly within class I RNRs. As already mentioned, class Ia utilizes a di-ferric oxygen center (Fe-O-Fe) to generate a radical and a nearby tyrosine residue to store it.1,9 In class Ib it is not a di-ferric oxygen center but a di-manganese cofactor that is used to generate the radical.51,52 The class Ib RNR operon encodes a separate protein called NrdI that is not found in the other classes of RNRs. NrdI is a flavoprotein that uses flavin mononucleotide (FMN) as a cofactor to produce superoxide from molecular oxygen, which then activates the di-manganese cofactor and enables the generation of the tyrosyl radical needed for the reduction reaction.9,52,53 In class Ic the radical is not stored in a tyrosine. The tyrosine close to the metal center is replaced by a phenylalanine in these proteins and the radical is stored in the metal center directly, which in this case is a heteronuclear manganese-iron cofactor.54,55 Class Id RNRs preferentially incorporate manganese to form a di-manganese center. Other than class Ib RNRs this center is not activated by a separate NrdI protein, but it appears to be activated directly by superoxide/hydrogen peroxide. The tyrosine residue is located slightly too far away for it to interact with the metal center and so the radical is not stored as a tyrosyl radical but on the metal center itself similar to class Ic RNR.56–58 Finally, there is the recently discovered metal-free class Ie RNR. In this case the tyrosine residue that usually stores the radical is co- valently modified into a DOPA residue capable of storing the radical. No metals are seen in the active site of the b-subunit in this class of RNRs.59

13 Ribonucleotide reductase from the hyperthermophile Aquifex aeolicus Aquifex aeolicus is a hyperthermophilic bacterium that is typically found in hot springs and close to underwater volcanoes. With an optimal growth temperature of 85-95° C, this organism is one of the most thermophilic bacteria known.60,61 For this species to survive at such extreme temperatures, it must have evolved special features and adaptations that increase the thermal stability of the proteins expressed within the cell. One of the reasons why we became interested in solving the structure of the RNR from this organism is that we wanted to see how this organism solved the problem of keeping an RNR stable at extreme temperatures. Another aspect that led us to be interested in this protein is that the gene for the b-subunit contains a region coding for an intein (intervening protein) within its iron binding site.

Inteins Inteins are regions of DNA embedded within a gene which codes for a protein. This DNA region is translated along with the rest of the gene and the resulting protein domain is then able to carry out an auto-catalytic processing step, known as protein splicing, in which it cleaves itself out from the precursor protein post-translationally. In the same step it fuses together the two extein (external protein) segments of the host protein in which it was embedded.62 This process produces two proteins: the functioning product of the host gene as well as the excised intein (Figure 9). 63

Figure 9. Intein schematic mechanism.

No energy source or external factors are needed for the mechanism of protein splicing to take place. It relies exclusively on the conservation of certain key residues and the correct folding of the intein domain.64 The most important residues are two cysteine residues, one of them is located at the N-terminus of the intein segment and the other (which can also be a serine/threonine) is located at the first N-terminal position of the C-extein. Through a series of acyl- transfer reactions involving these residues as well as other conserved motifs within the intein segment, the two peptide bonds flanking the intein are cleaved and a new peptide bond is formed between the two exteins. It is not obvious what the evolutionary advantage is for the host to harbor such inteins within its genome. They are thought to be parasitic DNA elements with no discernable benefit for the host.63 Two factors are important for the conservation of inteins in their host genome. Firstly, they are often located precisely in conserved regions of proteins that are critical for host survival. This makes the removal of the intein DNA sequence from the host very chal- lenging because imprecise excision would lead to inactivation of the host protein as well. Another factor is that many inteins contain a homing endonucle- ase domain. When the intein is spliced out of the host protein and this domain is properly folded, it can cause a double strand break at a specific recognition site. This recognition site is usually only found in a version of its own host gene that does not already contain a copy of the intein coding DNA segment. Occurrence of inteins in viral DNA seems to support the hypothesis that these are parasitic elements. If this were the case, one would expect inteins to be mostly evenly distributed among highly conserved gene regions. However, intein locations are significantly biased to being within genes that code for proteins of specific functions, mainly DNA replication and repair. Even more puzzling is that this is true even among proteins that have a similar function but are non-orthologous and genetically divergent.65 The protein splicing reaction of some inteins has been shown to be dependent on environmental factors such as pH and temperature and others. It is an interesting hypothesis that inteins could serve as a switch that could turn on proteins in response to certain stimuli. Since these proteins are already expressed and possibly even partially folded, this could be a much more rapid way to activate proteins rather than initiating transcription through a promotor for example. However direct evidence that these inteins benefit the host is lacking.64 In this context it is also notable that inteins are primarily found in bacteria and archaea and less so in eukaryotes and they are not found in multicellular organisms.64 It seems that more complex organisms are better able to purge these elements from their genome.

15 The topic of paper I is the structural and biochemical analysis of the RNR from Aquifex aeolicus. Phylogenetically, this RNR belongs to the NrdAh/NrdBh* subclass. A structure of the R2 from Clostridium botulinum was recently published.66 This R2 also belongs to the NrdBh subclass. How- ever, no structure had yet been solved of a corresponding R1 belonging to the NrdAh subclass. In the paper we present a 2.7 Å resolution crystal structure of the R1 from A. aeolicus (AaR1) in complex with ATP. As the first structure from the NrdAh subclass, a novel structural feature was observed that is present in all members of the NrdAh subclass but absent from all other R1 proteins. This feature is a stretch of approximately 20 amino acids that form a b-hairpin ‘hook’ that is located at the dimer interface. The b-hairpin ‘hook’ from one monomer interacts with the other b-hairpin ‘hook’ on the second monomer within the homodimer to create a four stranded anti-parallel b-sheet across the dimer interface (Figure 10).

* The naming convention of these phylogenetic subclasses follows that of the genes coding for these proteins. The genes coding for the a- and b-subunits of class I RNRs are named nrdA and nrdB, with the exceptions of class Ib RNRs, where these genes are named nrdE and nrdF respectively. Class II RNR genes are named nrdJ and class III genes are named nrdD. In the literature, the proteins themselves can be referred to by these names instead of a/b or R1/R2. 16

Figure 10. AaR1 crystal structure with novel ‘b-hairpin hook’ highlighted. One monomer shown in red the second in blue. (Figure adapted from paper I.) While this structural feature is common to all members of the NrdAh phylogenetic subclass, there is one aspect that is unique to AaR1 in particular. A tyrosine residue (Tyr511) located at the center of the b-sheet on one monomer forms a pi-stacking interaction with the same tyrosine residue on the second monomer. Sequence analysis revealed that this tyrosine is unique to AaR1, and the pi-stacking interaction cannot take place in other NrdAh proteins. It seems plausible that this tyrosine residue might be important for the extreme thermal stability of this protein. Another interesting aspect of this structure is that there are two ATP molecules bound in the a-site (Figure 11). Normally only one nucleotide is bound

17 in the ATP-cone domain of most RNRs and ATP-cone domains that bind more than one nucleotide are rare. Two structures have been published that show two nucleotides bound in the ATP-cone domain. These two structures are the R1 from Pseudomonas aeruginosa (PDB ID: 5IM3)67 and the R2 from Leeu- wenhoekiella blandensis (PDB ID: 5OLK).57 In both cases the nucleotide bound is dATP and the AaR1 structure therefore represents the first example of an ATP-cone domain shown to bind two ATP molecules.

Figure 11. ATP-cone domain of AaR1 showing the binding of two ATP molecules. A) Overview of the ATP-cone domain. B) Binding interactions of the ATP1 molecule. C) Binding interactions of the ATP2 molecule. (Figure adapted from paper I.) The binding mode of these nucleotides is also interesting. One of the two ATP molecules binds in a similar fashion as seen in other RNRs. The position and conformation of this molecule overlaps well with that seen in other structures, both those structures that bind one nucleotide and the two structures mentioned that bind two dATPs. The residues binding this molecule are also highly conserved between AaR1 and other RNRs. However, the second ATP molecule binds in a very different way. In both structures that bind two dATPs, the two dATPs are in a symmetrical arrangement around the central magnesium ion that coordinates the phosphate groups. In the case of AaR1 the second ATP molecule is arranged asymmetrically with the plane of the 18 phosphate groups being perpendicular to that of the nucleotide bases. Interest- ingly, some of the residues that coordinate the second ATP molecule (Tyr105 and Arg99) are actually not part of the ATP-cone domain but belong to the loop that connects the ATP-cone domain to the rest of the protein. Both the s- and c-sites are empty in the AaR1 structure. Because of this, loops 1 and 2 are also disordered in the structure. We also examined the R2 subunit from A. aeolicus (AaR2). As previously mentioned, the DNA sequence for this subunit contains an intein. The intein is located between the critical residues Glu228 and His231 which take part in the coordination of the di-iron center. To examine the cleavage of the intein, we engineered a construct of AaR2 where the intein was removed at the DNA level. We then solved crystal structures of this construct and the WT AaR2 which still had the intein DNA sequence in the plasmid DNA. The intein cleaved itself out successfully in the WT structure. Figure 12 shows the peptide bond between Cys230 and Leu229. In the WT protein, this bond was formed not at the ribosome but in the process of intein mediated protein splicing. No difference is seen between the two structures, indicating that the intein cleaved itself out successfully and it had no effect on the protein folding of AaR2.

Figure 12. Structure of the AaR2 homodimer. Insert: Alignment of the two AaR2 structures. The genomic AaR2 is shown in blue and the AaR2 construct lacking the intein DNA sequence is shown in grey. (Figure adapted from paper I.) The metal site is poorly occupied in both structures. This is also reflected in the EPR analysis which shows that only approximately 5% of the protein sample contains a radical. A reason for this could be the recombinant expression in E. coli. However even with such a small fraction of the protein in the sample being functional, we were still able to carry out activity measurements successfully. These measurements revealed that the protein is inactive at temperatures below 50° C and that the temperature optimum for catalysis was 79° C,

19 which is to be expected since the protein is from a hyperthermophilic organism. Finally, we also looked at the specificity regulation and the oligomerization states of this RNR. The specificity regulation for the A. aeolicus RNR is similar to that found in other RNRs, such as E. coli. This is also to be expected since s-site regulation is highly conserved across RNRs from different spe- 28 cies. The oligomerization studies show that the RNR forms an a2b2 active heterotetramer in the absence of any effectors as well as in the presence of ATP. The inactive a4b4 complex was also formed in the presence of dATP; however this required a higher protein concentration.

NUDT15

NUDIX Hydrolases NUDT15 is a member of the NUDIX (Nucleoside diphosphate linked to moiety x) hydrolase family of proteins.68 These proteins are housekeeping enzymes with important functions in nucleotide metabolism because they are typically involved in hydrolyzing a variety of different phosphorylated nucle- osides. By breaking down harmful oxidized or otherwise modified nucleotides before these can be incorporated into DNA, these proteins help to maintain the integrity of the genome.69 The first member of the NUDIX family of proteins to be discovered and studied in detail was the nucleoside-triphosphate pyrophosphohydrolase MutT from E. coli.70–72 The protein was originally discovered because a mutation in the gene coding for this protein caused a phenotype characterized by a 100- 10000 fold increased mutation rate.73 NUDIX enzymes are found in all domains of life including viruses.74 Many organisms have a collection of different proteins from this family. For bacteria it has been shown that the number of NUDIX genes and the genome size of an organism is linearly correlated.75 Humans have 24 genes coding for NUDIX hydrolases, all carrying out slightly different functions. The main substrates of NUDIX hydrolases are the already mentioned phosphorylated nu- cleosides (canonical as well as modified), however some NUDIX enzymes can also use other substrates such as capped RNA, nucleotide sugars, diphos- phoinositol polyphosphates and dinucleotide coenzymes and polyphosphates.75 These enzymes are small proteins, typically less than 20 kDa in size and are characterized by a sequence motif spanning 23 amino acids forming a loop-helix-loop feature called theUDIX box motifN (Gx5Ex5[UA]xREx2EExGU; x: any residue; U, aliphatic or hydrophobic residue)75,76. Among the most important residues within this motif are the glu- tamic acids in the REx2EE sequence which are responsible for the binding of divalent cations (Mg2+ in most cases). In the case of MutT there are two important Mg2+ ions involved in the catalysis reaction, both of which are located close to the phosphate groups of the substrate. One of them directly interacts with the b- and g-phosphates of the substrate. The other is located further away from the substrate phosphates, leaving enough space for water molecules to coordinate the Mg2+ ion. One of these water molecules bound to the Mg2+ ion

21 is able to perform a nucleophilic attack on the b-phosphate of the substrate. This leads to the breakage of the phosphodiester bond linking the a- and b- phosphates. The reaction then yields the monophosphate nucleotide and pyrophosphate.77

Reaction mechanism of MutT: & 8-oxo-dGTP + H$O → 8-oxo-dGMP + PP% + H

8-oxo-dGTP is a dangerous oxidized nucleotide species that can be incorporated into DNA. Incorporation of 8-oxo-dGTP into DNA leads to a drastic increase in AT -> GC transversion mutations78 and ultimately resulting in cell death. This explains the MutT mutator phenotype already mentioned. If MutT breaks down 8-oxo-dGTP before it can be used as a substrate by DNA-polymerase, then loss of MutT increases the amount of 8-oxo-dG being incorporated into DNA, causing excess mutations.79

NUDT15 and MTH1

MTH1 NUDIX hydrolases have recently risen to prominence with the discovery that MTH1, a homolog of MutT in humans, is a promising anticancer drug target.80,81 This enzyme was shown to carry out the same reaction in humans as MutT in E. coli. In addition to 8-oxo-dGTP it can also use other substrates such as 8-oxo-GTP, 2-OH-(d)ATP, 8-oxo-(d)ATP, O6-methyl(d)GTP and N6-methyl-dATP.82–88 The idea behind choosing MTH1 as a drug target was that increased levels of reactive oxygen species (ROS), due to malfunctioning redox regulation, is a phenotype common to many different forms of cancer.89,90 Since DNA is somewhat protected by the tight packing around the nucleosomes, it is the dNTP pool that is especially vulnerable to the elevated levels of ROS.91 This leads to increases in oxidized nucleotides which, as discussed, can then be incorporated into DNA, causing increased mutations and eventually cell death. Cancer cells have an increased need for dNTPs due to their rapid growth and they rely on proteins such as MTH1 for their survival because of their role as sanitizers of the oxidized nucleotide pool. At the same time, knockdown of MTH1 in normal cells has been shown not to be particularly toxic.80,92–94

NUDT15 As a homologue of MTH1, NUDT15 was initially named MTH2 (Figure 13). It was at first reported that mouse NUDT15 also had 8-oxo-dGTPase activity.95,96 After the initial results showing the potential for MTH1 inhibition in cancer treatment, NUDT15 was investigated because of the possibility that it 22 might compensate for the loss of MTH1. It was shown that NUDT15 knockdown has little to no effect on the survival of cancer cells, that the activity of NUDT15 towards 8-oxo-dGTP was much lower than towards the canonical dGTP, and that NUDT15 did not show any influence on the amount of 8-oxo- dG incorporation into DNA.92 NUDT15 was also screened for a variety of different oxidized nucleotides and no significant activity was observed. NUDT15 must have a different function in the cell. To this day the true function of this enzyme remains unknown, however NUDT15 has come to atten- tion in another context, which is also related to cancer and that is thiopurine therapy.

Figure 13. Structure of the NUDT15 homodimer. One monomer shown in marine blue, the other in light blue. NUDIX box residues are highlighted in salmon and raspberry red. Mg2+ ions are colored in grey and water molecules in red.(PDB ID: 5BON 92)

Thiopurine anticancer drugs Thiopurine drugs were first developed by the pioneers of rational drug design Gertrude Elion and George H. Hitchings in the 1950s.97 Elion and Hitchings would later share the Nobel Prize in Physiology or Medicine with Sir James W. Black in 1988 for their “discoveries of important principles for drug treatment”.98 Thiopurines are derivatives of the purine guanine. These drugs are very effective in the treatment of acute lymphoblastic leukemia (ALL), which is the most prevalent form of childhood leukemia. ALL alone accounts for approximately 25% of all cancer diagnoses and of those, 60% occur in children.99 To this day, nearly 70 years after the first thiopurine drug was ap- proved, thiopurines are still among the most important treatment options for ALL.100 These drugs are also powerful immunosuppressants which makes

23 them useful in the treatment of diseases such as Inflammatory bowel disease (IBS) and Crohn’s disease.101 There are three thiopurine drugs currently used in the clinic: 6-mercaptopurine (6-MP), 6-thioguanine (6-TG) and azathioprine (AZA-T), all of which, through many intermediate steps, are ultimately converted into the active metabolites 6-ThioGTP and 6-Thio-dGTP (Figure 14).102

Figure 14. Structures of 6-MP, 6-TG and AZA-T. 6-Thio-dGTP is a highly toxic substance that is used by DNA polymerase as a substrate and is incorporated into DNA leading to cell death.103–106 One issue that arises in the treatment of patients with these drugs is that the effects are not immediate. After these antimetabolites are incorporated into DNA, they first have to be methylated107 and only after a second round of replication does this process lead to a Me-6-thio-dG:T mispair which cannot be repaired by the cells mismatch repair (MMR) machinery.108 Some patients are extremely sensitive to thiopurines and experience severe and potentially life-threatening side-effects such as bone marrow suppression and subsequent decrease of the white blood cell count in the blood. An added complication arises because these side effects appear with a marked delay from the onset of thiopurine therapy.109 One sensitivity factor is the protein thiopurine methyltransferase (TPMT). In normal patients, TPMT methylates thiopurines before they can be metabolized into the therapeutically active 6-thio-dGTP.110–113 In some patients this protein is inactivated by a mutation and these patients experience increased toxicity from thiopurine treatment. Recommendations for pre-emptive genotyping of TPMT prior to thiopurine therapy have been in place for some years.114–116 This actually represents one of the first examples of precision medicine based on genomics.117–119 TPMT genotype is not the only predictor of thiopurine sensitivity.120,121 Only 25% of patients that are thiopurine sensitive have abnormal TPMT activity.122–124 Inactivating TPMT variants are also rare in people of east Asian descent,125 however thiopurine toxicity is still observed in these patients. In fact, the frequency of cases of thiopurine intolerance is even higher in east Asians than in patients of European descent.122 To discover other thiopurine 24 sensitivity factors, genome wide association studies for the thiopurine intolerance phenotype were carried out. It is in this context that the connection to NUDT15 was discovered. A missense variant in NUDT15 c.415C>T (rs116855232) was discovered by several groups.126 This single nucleotide polymorphism (SNP) causes the arginine at position 139 to be replaced by a cysteine residue (Arg139Cys). This mutation is significantly correlated with thiopurine sensitivity in patients being treated for ALL and inflammatory bowel disease (IBD).100,101,127–131 This mutation is common in patients of Asian (>10% of East Asian patients, 6.7% of South Asian patients) and Hispanic descent (6%), but is less frequent in Europeans (0.35%) and Africans (0.1%).122,132

25 NUDT15 breaks down thiopurine metabolites After it was established that NUDT15 is a thiopurine sensitivity factor, a substrate screen identified 6-thio-dGTP and 6-thio-GTP as substrates for NUDT15.92 In paper II we conducted a study to further characterize this relationship. We showed that the activity for the thionylated substrates was much higher than for the canonical dGTP and GTP. In fact, the catalytic effi- ciency (kcat/Km) was 13-15x higher for 6-thio-dGTP over dGTP and as high as 243-290x higher for 6-thio-GTP over GTP. This was primarily due to NUDT15 having a much higher affinity for these compounds. The fact that NUDT15 breaks down these metabolites explains why deleterious NUDT15 variants cause thiopurine sensitivity. The situation is much the same as with TPMT, if NUDT15 activity is reduced because of an inactivating mutation, then a normal dose of thiopurines will lead to a greatly increased effective dose of 6-thio-dG being incorporated into DNA (Figure 15).

Figure 15. Schematic illustration of the role of NUDT15 in thiopurine metabolism. (Figure adapted from paper II.) We expected to see that the R139C mutation would inactivate the protein. However, our in vitro data showed that this was not the case. Not only was the mutant still active in vitro, but it even showed a slightly higher activity towards 6-thio-dGTP than the WT enzyme. However, the thermal stability of this mutant is severely decreased with a DT of approximately −10° C relative to the WT enzyme and our in-cell data shows this mutant is rapidly degraded in cells.

We also co-crystallized NUDT15 together with 6-thio-GTP and solved a 1.7 Å X-ray crystal structure of NUDT15 with the hydrolysis product 6-thio-GMP bound in one of the two monomers in the homodimer (Figure 16).

Figure 16. Crystal structure of NUDT15 in complex with 6-thio-GMP. A) Closeup view of the binding site of NUDT15 B) Zoom in on Arg139. (Figure adapted from paper II.) Our structure revealed several important residues that are involved in the binding of the substrate. The ligand binds to the protein via direct hydrogen bonds to the residues Leu45, Gly137 and His49. Perhaps the most important interaction is the tight fit of the sulfur atom within the hydrophobic pocket, which is formed by Leu138, Phe135 and Leu45. It is this interaction that causes NUDT15 to have a higher affinity for the thionylated compounds than GTP or dGTP. Arg139, the mutation site associated with thiopurine intolerance, is far enough away from the substrate binding site that a mutation at this site would not cause significant distortions to the binding site itself. However, this arginine residue forms important stabilizing interactions between two helical segments (helix a2 and the short 310 helical segment h3). Presumably loss of these interactions destabilizes the protein enough to be marked for pro- teasomal degradation.133 The amino acid at position 140, directly adjacent to Arg139 happens to be a cysteine as well. Therefore, it is possible that in the R139C mutant these two residues, which both reside on helix a2, could form a disulphide bridge, which could disrupt the folding of this helix. Finally, we showed that silencing of NUDT15 mRNA by NUDT15-specific shRNA (shNUDT15) made the human colon cancer cells HCT116 and HCT116 3-6 more sensitive to treatment with 6-TG. HCT116 is an MMR de- ficient cell line and HCT116 3-6 is MMR proficient. The NUDT15 dependent thiopurine sensitivity phenotype was much more pronounced in the MMR proficient cell line. This is expected since the DNA lesions caused by thiopurine treatment cannot be repaired by MMR, leading to futile repair attempts and

27 cell death.107,109,134–136 Certain cancers can develop MMR deficiency which leads these cancers to be resistant to thiopurine therapy.137

The first inhibitor for NUDT15 Paper III describes the development of the first inhibitor for NUDT15. The study began with a small molecule screening campaign monitored by a malachite green assay92 where each screening condition contained an individual potential inhibitor, NUDT15, dGTP and pyrophosphatase (PPase). As NUDT15 hydrolyzes dGTP to dGMP, pyrophosphate (PPi) is produced and broken down by PPase into inorganic phosphate, which can then be detected by malachite green reagent. 17946 compounds were screened and 37 of these were identified as having enough of an inhibitory effect to be examined more in detail. Finally, the compound TH884, which had an IC50 of 7 µM, was chosen as starting point for further development. Structure activity relationship (SAR) studies, in which modifications of structural features of the inhibitor were again examined using a malachite green assay, quickly identified the fluorophenyl functional group as a suitable target for modification. After many cycles of iterative improvement, the lead compound TH1760 was developed, whose IC50 of 25 nM was more than 200x lower than that of TH884. An increase in the melting temperature of NUDT15 in the presence of the inhibitor also confirmed in vitro binding. We also solved a 1.6 Å resolution X-ray crystal structure of TH1760 in complex with NUDT15. The structure showed that the inhibitor bound in a similar fashion as 6-thio-GMP (Figure 17). The benzoxazolone moiety of the inhibitor is positioned deep in the binding site, similar to the guanine of 6- thio-GMP. This part of the inhibitor also makes the same hydrogen bond to Gly137, however the hydrogen bond to Leu45 in the 6-thio-GMP structure is replaced by a hydrogen bond to Leu138. Additionally, the interaction between the ligand and Thr94 is stronger in the case of the inhibitor due to it forming a direct hydrogen bond to the sulfonamide group, compared to the interaction mediated by a coordinated water molecule between Thr94 and 6-thio-GMP. The structural data explained the tighter binding of the inhibitor and having this data made it easy to identify which position of the inhibitor would need to be modified to make binding to NUDT15 impossible. A molecule suitable as a negative control was synthesized by N-methylating the benzoxazolone moiety of TH1760.

Figure 17. Development of TH1760. A) Structures of TH884 and TH1760. B) Close- up view of TH1760 binding to NUDT15. The protein is colored blue and TH1760 is colored red. The structure of NUDT15 in complex with 6-thio-GMP is aligned and shown in transparent grey. (Figure adapted from paper III.) The inhibitor was then further examined and was shown to be highly selective with little off-target effects in a screen of other NUDIX proteins and kinases performing similar functions. Cellular target engagement was confirmed by Cellular thermal shift assay (CETSA).138 In this assay cells expressing NUDT15 are heated to specific temperatures. The cells are then broken and spun down and the amount of NUDT15 still present in the soluble fraction is measured via western blot. In this way, the aggregation temperature Tagg can be determined in the presence and absence of the inhibitor. For NUDT15 in the presence of TH1760, the DTagg relative to NUDT15 without any inhibitor was measured at 6.5° C. A variation of the experiment called Isothermal dose- response fingerprint (ITDRF) CETSA was also performed. In this case a screening temperature is determined at which roughly 80% of soluble NUDT15 is aggregated. A screen is then performed where it is not the temperature but the inhibitor concentration that is varied. In the presence of TH1760, the fraction of soluble NUDT15 started to increase substantially above a 10 µM concentration of TH1760. The third assay used for examining cellular target engagement was Drug affinity responsive target stability (DARTS).139 This assay exploits the tendency of enzymes to be more resistant to protease digestion when tightly bound to and stabilized by a ligand. All these assays confirmed that TH1760 was able to bind NUDT15 in the cell. Finally, it was also shown in the paper that TH1760 inhibits NUDT15 in vivo and it is capable of potentiating thiopurine drugs. The EC50 of 6-TG was decreased ten-fold in the presence of TH1760. An assay using 14C-labeled 6-MP and later measuring the radioactivity of DNA and RNA showed that the increased toxicity of the TH1760/thiopurine synergy was due to increased ac- cumulation of 6-thio-(d)G in DNA/RNA. Interestingly, the thiopurine

30 potentiating effect was sufficient to resensitize HCT 166 cells, which are resistant to thiopurine therapy due to their MMR deficiency.

31 Further development of inhibitors for NUDT15 A second line of inhibitors was also explored. In this case a more targeted search was performed and the structure of the substrate 6-thio-dGTP was used as a starting point for the screening of 62 structurally similar compounds. The best hit was NSC56456 (IC50 619.5 nM), a patented inhibitor of the transcription factor NF-E2-related factor 2 (NRF2). The 1.8 Å resolution crystal structure of NUDT15 in complex with NSC56456 is to be published in additional paper 1. The position of the inhibitor is highly similar to that of 6-thio-GMP, with the thio-guanine moiety occupying the same space in both structures and the cyclohexane ring of NSC56456 being in the same position as the ribose of 6-thio-GMP (Figure 18).

Figure 18. Structure of the 6-thioG analog NSC56456 in complex with NUDT15. A) Structure of NSC56456. B) Close-up view of the binding pocket of NUDT15 in complex with NSC56456. NSC56456 is shown in magenta and NUDT15 in blue. The structure of NUDT15 in complex with 6-thio-GMP is aligned and shown in transparent grey. (Figure adapted from additional publication 1.) Interestingly, the NSC56456 bound structure shows significant rearrangements of bulky aromatic residues in the binding pocket, not seen in other NUDT15 structures (shown in Figure 18). The lack of hydroxyl groups on the cyclohexane ring allows for Trp136 to move in closer to the inhibitor thereby pushing out Tyr90.

NSC56456 was then iteratively improved, resulting in the lead compound TH8321 which is presented in paper V. This compound differs from NSC56456 structurally by the addition of an anisole ring at the 8’ position of the guanine moiety, leading to a 20x improved potency of this inhibitor (IC50 35.2 nM) versus NSC56456. The 2.35 Å crystal structure shows similar rearrangements of the aforemen- tioned aromatic residues, with Tyr90 oriented in a similar fashion as seen in the NSC56456 structure. However, the aromatic ring system of Trp136 is flipped 180° in the opposite direction due to the anisole ring system of TH8321 (Figure 19).

Figure 19. Structure of NUDT15 in complex with TH8321. A) Structure of TH8321. B) Close-up view of the binding pocket of NUDT15 in complex with TH8321 with the structure of NUDT15 in complex with 6-thio-GMP aligned. TH8321 is shown in grey, NUDT15 in blue. The structure of NUDT15 in complex with 6-thio-GMP shown in transparent grey. (Figure adapted from paper V.) Even though the IC50 of this 6-thio-guanine analog inhibitor is slightly worse than that of the lead compound of the first line of inhibitors TH1760, we show the cellular target engagement of TH8321 is markedly improved over TH1760. ITDRF CETSA data shows significant stabilization occurring at a concentration of 0.8 µM versus 20 µM in the case of TH1760. The inhibitor was also shown to be sufficiently selective, as well as non-toxic to cells in vitro by itself. More significantly, the in-cell potentiation of the toxic effects of thiopurine was far superior to that of TH1760, as confirmed by the Resaz- urin viability assay.

33 Further work was also carried out on TH1760, the initial lead compound of the first line of inhibitors. The compound was improved by addition of two methyl groups, one to the piperazine ring system and the other to the nitrogen of the indole group. This resulted in the new inhibitor TH7755, which is presented in paper VI. TH7755 has a slightly worse IC50 compared to TH1760, however the compound has drastically improved in cell NUDT15 binding characteristics compared to TH1760, presumably because of the increased lipophilicity conferred by the additional methyl groups. CETSA data showed that NUDT15 in the presence of 10 µM TH7755 had a DTagg of » +4° C relative to NUDT15 without inhibitor. TH1760 at the same concentration only produced 1° C of cellular thermal stabilization. Similarly, ITDRF CETSA showed a 10x improvement in cellular target engagement by TH7755 over TH1760. The 1.6 Å resolution crystal structure of NUDT15 in complex with TH7755 shows that this structure is very similar to that of NUDT15 in complex with TH1760 presented in paper III, with a low RMSD value of 0.51 Å. The two ligands make the same interactions with the protein and the position and conformation of the ligand within the binding pocket is very similar between the two structures. A key difference however, is the indole ring system of TH7755, which is flipped 180° compared to that of TH1760, presumably due to the added methyl group on the indole group of TH7755 (Figure 20).

Figure 20. NUDT15 in complex with TH7755. A) Structure of TH7755. B) Close-up view of the binding pocket of NUDT15 in complex with TH7755. TH7755 shown in orange and NUDT15 shown in blue. The structure of NUDT15 in complex with 6- thio-GMP is aligned and shown in transparent grey. (Figure adapted from paper VI.)

The antiviral drugs acyclovir and ganciclovir Similar to thiopurines, the antiviral drugs acyclovir and ganciclovir are both analogs of the nucleoside guanosine. Like thiopurines, acyclovir was also the result of the work of Gertrude Elion.140 This compound actually was the first successful small molecule antiviral drug ever discovered.141 Ganciclovir is an analog of acyclovir,142 the only structural difference between the two molecules is a primary alcohol shown in Figure 21.

Figure 21. Structures of acyclovir triphosphate and ganciclovir triphosphate, the active metabolites of acyclovir and ganciclovir. These drugs are used to treat infections of various members of the herpes family of viruses. Chief among them is cytomegalovirus (CMV), a particularly prevalent virus that affects 60-90% of the worldwide population that can cause severe complications in immunocompromised patients. This viral infection is also especially relevant for hematopoietic stem cell transplantation.143 Acy- clovir is also effective against the varicella-zoster and Epstein-Barr viruses among others.98 The mechanism of action is similar for both of these compounds. They are both non-toxic to normal cells in their unphosphorylated prodrug form. Cer- tain viral thymidine kinases (TK) however are capable of performing the first phosphorylation step to convert these compounds into their monophosphate form.144 Once they are phosphorylated, cellular kinases can take over and convert the prodrugs into their active tri-phosphorylated form.145 Ganciclovir/acyclovir-TP are then used preferentially as a substrate by viral DNA polymerase and are incorporated into viral DNA, thereby disrupting virion production.146 However, these compounds are also cytotoxic. This is due the fact that in their tri-phosphorylated form, they can be incorporated into cellular DNA, disrupting cell division.147,148 The unique mechanism of action of these drugs also makes them useful for suicide gene therapy. In fact, the herpes simplex virus thymidine kinase (HSV-

35 TK)/ganciclovir system is one of the prime examples of suicide gene therapy. The main idea here is that the HSV-TK gene can be modified in such a way that it is only expressed in the presence of a tumor specific promotor. It is then administered via a viral or non-viral vector. Ideally this causes the HSV-TK protein to only be expressed in the targeted cancer cells and not in normal cells, thereby making only the cancer cells susceptible to ganciclovir.149,150

NUDT15 breaks down antiviral drug metabolites Papers IV and V are about the connection between NUDT15 and the two antiviral drugs acyclovir and ganciclovir. Among all the nucleobase and nucleoside analog (NNA) drugs, currently used in the clinic to treat various cancers and viral infections, there are 12 compounds that are only active in their triphosphorylated form. All of these 12 compounds were screened, and we show that the active metabolites of the drugs acyclovir triphosphate and ganciclovir triphosphate are also high affinity substrates for NUDT15. In this case again, NUDT15 hydrolyzes the triphosphate forms of these compounds to the corresponding monophosphates, thereby releasing pyrophosphate. Structures of NUDT15 in complex with acyclovir and ganciclovir were solved to resolutions of 1.6 Å and 1.95 Å respectively.

Figure 22. Structures of NUDT15 in complex with acyclovir and ganciclovir. Closeup view of the binding pocket of NUDT15 shown in blue in complex with (A) acyclovir, shown in orange and (B) ganciclovir, shown in green. The aligned structure of NUDT15 in complex with 6-thio-GMP is shown in grey. (Panel A adapted from paper IV and panel B adapted from paper V.)

The acyclovir structure (Figure 22A) shows the same rearrangement of the aromatic residues Trp 136 and Tyr90 as discussed in the NSC56456 complex (Figure 18). However, the same is not true in the case of ganciclovir. The polar primary alcohol of ganciclovir, which occupies the same space as the ribose moiety in the 6-thio-GMP structure, appears to stabilize Tyr90 in the same position as seen in the 6-thio-GMP structure (Figure 22B). It appears the rearrangements of these residues are likely significant for NUDT15 to be able to bind both of these compounds with similar affinity. In these two publications, we go on to show that these anti-viral drugs are potentiated in the absence of NUDT15, as one would expect. The mouse bone marrow stroma cell line M2-10B4 was gene edited using CRISPR-Cas9, to remove the NUDT15 gene, and then infected with murine CMV. Both acyclovir and ganciclovir showed increased anti-viral effectivity in NUDT15-/- cells compared to WT, as measured by virus copy number quantification by immediate-early protein (IE1) staining. The same thing was also shown for the cytotoxic effects of the acyclovir and ganciclovir triphosphates in the presence of viral TK. NUDT15-/- and NUDT15+/+ versions of the human hematopoietic cell line Nalm6 were engineered. As expected, these cells were all resistant to acyclovir and ganciclovir, however upon introduction of the HSV-TK gene, these compounds became toxic and the NUDT15-/- cells were more sensitive to both acyclovir and ganciclovir. It was also studied whether inhibiting NUDT15 with one of our inhibitors would potentiate anti-viral treatment in the same way that it potentiates thiopurines. The presence of 5 µM of the inhibitor TH8321 in the infectivity assay was shown to improve the EC50 of ganciclovir by » 0.5 log. Lastly the influence of NUDT15 diplotypes on clinical outcomes of hematopoietic stem cell transplantation were studied in vivo in a cohort of 248 patients. CMV infection prevention through treatment with acyclovir was more effective in patients whose NUDT15 diplotype identified them as having low NUDT15 activity. Unlike thiopurine potentiation by NUDT15 depletion, in the case of the antivirals, the effects of the different NUDT15 risk alleles are not additive. When only one risk allele was present, the effectiveness of Acy- clovir/ganciclovir was similar to that of the WT. Acyclovir/ganciclovir potentiation only occurred when NUDT15 activity was completely lost. NUDT15 diplotype was also a good predictor of graft failures. The probability of graft failure was highest when the stem cell donor was seropositive for CMV and when they had a NUDT15 diplotype associated with low activity. These find- ings represent the first report of pharmacogenetic factors associated with these drugs. It might prove beneficial to adjust acyclovir and ganciclovir dosing based on NUDT15 pre-emptive genotyping, as is already implemented for thiopurines, however this issue requires more in-depth study with increased sample sizes.

37 NUDT15 inhibition in combination with acyclovir/ganciclovir treatment also warrants further study. It seems plausible that NUDT15 inhibitors could ex- pand the therapeutic window of these drugs. NUDT15 inhibition might also prove useful in the context of suicide gene therapy by potentiating the effects of ganciclovir.

Examining the structural basis of thiopurine intolerance After the NUDT15 Arg139Cys variant had been discovered to be a determin- ing factor in thiopurine sensitivity, more work was carried out to find other potentially sensitizing NUDT15 variants. In 2016, in a trial involving 270 children being treated for ALL, three new NUDT15 variants were discovered.151 These variants are c.416G>A (Arg139His), c.52G>A (Val18Ile) and c.36_37insGGAGTC (Val18_Val19insGlyVal).

Figure 23. NUDT15 variant locations. A) Locations of NUDT15 variants highlighted in the protein sequence. The locations of variants with a frequency >1% are highlighted in red, locations of less frequently encountered NUDT15 variants are highlighted in yellow.122 (B) Mutation sites of the four most clinically relevant NUDT15 variants (Arg139C, Arg139His, Val18Ile and Val18_Val19insGlyVal) highlighted in the NUDT15 crystal structure (PDB ID: 5LPG). NUDT15 is shown in marine blue, 6-thio-GMP is shown in orange and relevant mutation sites are highlighted in red.

39 Several other NUDT15 variants have since been discovered in the clinic (shown in Figure 23A)122,152–154 and in a recent study a massively parallel assay examined nearly every possible amino acid substitution of the 163 residues that make up NUDT15, which identified 1152 deleterious variants.155 How- ever, the four mutations mentioned earlier (highlighted in Figure 23B) are still the most clinically relevant NUDT15 variants with a minor allele frequency of >1%.122 It was a longstanding goal of ours to obtain any structural data on the clinically relevant NUDT15 variants. However, the destabilization of the protein due to these mutations was enough to hamper our attempts to crystallize these variants. We later realized we could use the significant thermal stabilization of NUDT15 upon inhibitor binding to our advantage.

Figure 24. Thermal stability assay of wildtype and mutant NUDT15 in the presence of different ligands. (Figure adapted from paper VI.) As shown in Figure 24, binding of the inhibitors TH1760 or TH7755 increased the melting temperatures of the NUDT15 variants by an average of + 11° C. The thermal stability of the NUDT15 variants in the presence of the inhibitors was even superior to that of WT NUDT15 by itself. We were able to use this stabilization effect of the inhibitor to our advantage and proceeded to successfully crystallize and solve high resolution structures of all the four clinically relevant NUDT15 variants. The differences between the mutant structures and WT NUDT15 are quite subtle. The mutated cysteine in the R139C structure (Figure 25A), shows that this cysteine is in a dual conformation and the hypothesized disulphide bond between Cys140 and the mutated Cys139 is not seen in the structure. The stabilizing interactions, between the a-helix a2 and the adjacent 310 helix h3, are not completely disrupted in either of the two mutant variants R139C and R139H. The cysteine in the R139C structure is still partially capable of inter- acting with residues on and close to h3 (Leu131 and Leu134). The same is 40 true for the histidine in the R139H structure (Figure 25B). Here the interaction is mediated by a coordinated water molecule. In both cases the h3 segment is shifted further away from the a2 helix than is seen in the WT structure.

Figure 25. Arg139 mutation site of NUDT15. A) NUDT15 Arg139Cys mutant structure. B) NUDT15 Arg139His mutant structure. (Figure adapted from paper VI.)

The V18I mutant structure is remarkably similar to the WT structure (Fig- ure 26A). The only difference between valine and isoleucine is one additional methyl group on isoleucine. In the mutant structure this methyl group appears to be well accommodated in the immediate surroundings, causing little distortions in the residues in the vicinity of Ile18. It appears the additional methyl group of the isoleucine, which is situated on the b-strand b1 and directly un- derneath the NUDIX box motif, puts the structure under enough strain to reduce the melting temperature of the mutant by 4° C. Lastly, we solved the V18_V19insGV mutant (Figure 26B and Figure 26C). This variant has an insertion of the residues glycine and valine in a region of the protein that already contains a glycine-valine triple repeat. The insertion extends this segment by one more repetition (Figure 26B). This leaves most of the region structurally unchanged, however the N-terminal region of the protein is effectively pushed out, away from the core of the protein, by two residues (Figure 26C).

Figure 26. Val18 mutation site of NUDT15. A) Val18Ile mutant structure. B) Com- parison between the sequences of the V18_V19insGV mutant and WT NUDT15. C) V18_V19insGV mutant structure shown in blue, the N-terminal region is highlighted in red. The aligned WT NUDT15 structure is shown in transparent grey. (Figure adapted from paper VI.) A detailed structural analysis in the N-terminal region is made difficult by poor resolution due to the inherent flexibility of this region. However, the point of divergence between the WT and the V18_V19insGV mutant can clearly be seen. In the mutant structure Val14 is in the same position as Pro12 is in the WT structure and Pro12 is shifted out further in the direction of the N-terminus. This disrupts important stabilizing interactions that Pro12 makes in the WT structure, such as the CH/p interaction with His91 and the hydrophobic interaction with Phe52. Taken together, these four crystal structures of clinically relevant NUDT15 variants provide valuable insights into the structural basis for the observed thiopurine and ACV/GCV sensitivity phenotypes.

Summary and conclusions

In paper I, we examined RNR from A. aeolicus and solved the first structures of the R1 and R2 subunits of a class Ia RNR from a hyperthermophilic organism. The AaR1 structure is particularly interesting because of a novel structural feature, a ‘b-hairpin hook’ located at the dimer interface, that is a defining feature of the NrdAh phylogenetic subclass. Additionally, the ATP-cone domain of AaR1 is shown to bind two ATP molecules simultaneously. This is unusual because RNR structures with more than one nucleotide bound in the ATP-cone domain are rare. The AaR1 structure is not only the first example of an R1 with two ATPs bound, but one of the two ATP molecules also has a unique binding mode not seen before in other structures. It is currently not known what the evolutionary advantage could be for the organism to bind two molecules in the a-site. The R1 and R2 structures of this RNR from the hyperthermophilic bacterium A. aeolicus represent an interesting new addition to the growing number of RNR structures from different organisms. Paper II examines the connection between NUDT15 and thiopurines. Af- ter work by ourselves and others it has now been firmly established that NUDT15 sequence information is an important factor predicting thiopurine intolerance; pre-emptive genotyping prior to thiopurine therapy is now recommended and implemented in the clinic.110 The development of two separate lines of structurally dissimilar inhibitors for NUDT15, which was supported by X-ray crystallographic data, is described in papers III, V and VI. We show that NUDT15 inactivation by inhibition potentiates thiopurine therapy. These inhibitors are a very useful addition to the available toolkit for the study of NUDT15, whose physiological function is still unknown. They might even prove useful in the treatment of certain cancers that are resistant to thiopurines due to their MMR-deficiency. Papers IV and V show, for the first time, that NUDT15 also breaks down the active metabolites of the antiviral drugs acyclovir and ganciclovir. This is especially relevant in the treatment of immunocompromised patients and in the context of hematopoietic stem cell therapy. Cytotoxicity and stem cell graft failure can now be explained in cases where the patient’s NUDT15 diplotype categorizes them as being poor metabolisers. NUDT15 genotyping could prove to be useful in these treatments, similar to what is already implemented in thiopurine treatment. Here we also show that ganciclovir treatment

43 is potentiated in the presence of TH8321, the lead inhibitor in the 6-thioG analog inhibitor series. Finally in paper VI, we use the stabilizing effect of the already mentioned inhibitors to our advantage because they allowed us to crystallize the four most clinically relevant NUDT15 variants. We present a detailed description of the effects of all of the four mutations on the NUDT15 protein structure and how they could lead to the thiopurine intolerance phenotype.

Popular science summary

At the heart of life on earth lies the so-called central dogma. That is the principle that the blueprint for all the machinery that the cell needs is stored in molecules of DNA. DNA is simply a long chain of four different molecules called deoxyribonucleotides, the building blocks of DNA. Short relevant segments of the DNA are copied as needed into another similar molecule called mRNA which is made up of slightly different building blocks called ribonucleotides. This mRNA is then used as a template in a process called transla- tion, where proteins are produced. Proteins can be thought of as molecular machines that carry out all the necessary functions within the cell. For one cell to make another copy of itself, it also needs to make another copy of the DNA. DNA can also be damaged, and the cell is able to repair some forms of DNA damage. This means there always needs to be a supply of DNA and RNA building blocks ready to use. The first protein we discuss in this thesis is Ribonucleotide reductase (RNR), a protein that turns RNA building blocks into DNA building blocks, a crucially important step for all life on earth. There already is a large body of work studying this protein. Paper I examines the RNR from Aquifex aeolicus, a bacterium that lives at extremely high temperatures of 85-95° C. We examine this protein for structural modifications that could explain its extreme thermal stability. The second protein we examine is NUDT15. There are certain drugs that are so similar to normal DNA building blocks that they make their way into DNA. This is what makes these drugs toxic to the cell, however sometimes this toxic effect is exactly what one wants to achieve in medical treatment and NUDT15 interferes with this goal. One example of this is thiopurine treatment. Thiopurines are chemotherapeutic drugs used in the treatment of certain cancers such as acute lymphoblastic leukemia (ALL), the most prevalent childhood leukemia. We show in paper II that NUDT15 breaks down these drugs before they can have their desired effect. There are people that have an inactivating mutation in their NUDT15 gene. In these patients a normal dose of thiopurine treatment is effectively much too high and leads to potentially life-threatening complications. A big problem here is that the effects of thiopurines only show up after a long delay. This makes treatment especially difficult. After these discoveries have been made, it is now recommended to perform genotyping on patients before they are treated with thiopurines, to find

45 out beforehand whether their NUDT15 protein functions normally or not. In papers III, V and VI we describe the development of two separate (and structurally dissimilar) series of inhibitors; molecules that bind to NUDT15 and prevent anything else from binding, thereby inactivating the protein. We also show that the effects of thiopurine treatment are drastically increased in the presence of these inhibitors. In papers IV and V, we show that NUDT15 not only interferes with thiopurine treatment but that this is also the case with the anti-viral drugs acyclovir and ganciclovir, which are somewhat similar to thiopurines, and whose mechanism of action are also comparable because they are also incorporated into DNA, although preferentially into viral DNA. These drugs are particularly important in bone marrow transplantation where the virus cytomegalovirus (CMV) can cause severe problems. We show that NUDT15 breaks down these drugs before they can be incorporated into DNA, similar to thiopurines and also show that the effect of these drugs is enhanced by NUDT15 inhibitors. In all these studies atomic resolution models determined by X-ray crystallography provided important insights into how these drugs interact with the protein at atomic scale. It had long been attempted to obtain any structural information for clinically relevant NUDT15 variants. However, they were not stable enough to crystallize, which is a necessary step in X-ray crystallography. After seeing that the protein was much more stable when it had bound one of our inhibitors, we were able to use this to our advantage and obtained high resolution structures of four clinically relevant NUDT15 variants. We were able to show how the protein is affected by these mutations and how these changes cause some patients to be extremely sensitive to certain drugs such as thiopurines and the antiviral drugs acyclovir and ganciclovir.

Populärvetenskaplig sammanfattning

Den så kallade centrala dogmen – principen att ritningarna för cellens maskineri är lagrad i DNA-molekyler – är central för allt liv på jorden. DNA-molekylen är en lång kedja uppbyggd av fyra mindre molekyler som kallas deoxyri- bonukleotider. Kortare segment av DNA-molekylerna kopieras vid behov till en snarlik molekyl – mRNA – som består av andra, liknande byggstenar; ribonukleotider. mRNA-molekylen används sedan som mall vid translationen; produktions- processen för proteiner. Proteiner är i princip molekylära maskiner, och dessa utför nästan alla uppgifter som behövs för att cellen ska fungera. När en cell ska delas, och göra en ny kopia av sig själv, måste den också göra en ny kopia av sitt DNA. DNA-molekyen kan också bli skadad, och cellen har ett maskineri för att reparera vissa former av sådana DNA-skador. Allt detta innebär att det ständigt måste finnas ett lager av DNA- och RNA-byggstenar redo att användas i dessa processer. Det första proteinet vi diskuterar i denna avhandling är Ribonukleotidreduktas (RNR); ett protein som förvandlar RNA-byggstenar till DNA-byggstenar. Detta omvandlingssteg är viktigt för allt liv på jorden, och det har därför länge forskats mycket kring hur detta protein fungerar. I Artikel I denna avhandling undersöker vi RNR från Aquifex aeolicus, en bakterie som lever vid extremt höga temperaturer (85-95 °C), och försöker hitta strukturella modifieringar som kan förklara dess extrema värmestabilitet. Det andra proteinet vi undersöker är NUDT15. Det finns vissa läkemedel som är så lika de vanliga DNA-byggstenarna att de kan byggas in i DNA-molekylen, och dessa läkemedel är därför giftiga för cellen. Ibland är detta exakt vad man vill uppnå vid en medicinsk behandling, men NUDT15 kan förhindra att sådan inbyggnad av ”felaktiga” byggstenar sker. Ett exempel på detta är tiopurinbehandling. Tiopuriner är kemoterapeutiska läkemedel som används vid behandling av vissa cancerformer, som akut lymfoblastisk leukemi (ALL); den vanligaste formen av barnleukemi, och vi visar i artikel II att NUDT15 bryter ner dessa läkemedel innan de får sin öns- kade effekt. En del människor har dock en inaktiverande mutation i sin NUDT15-gen. Hos dessa patienter är en normal dos av tiopurin alldeles för hög och leder till po- tentiellt livshotande komplikationer. Ett stort problem för dessa patienter är att effekterna av tiopurinerna syns efter en längre tid, och sedan upptäckten av

47 denna mutation rekommenderas genotypning av patienter innan de behandlas med tiopuriner, för att redan i förväg känna till om deras NUDT15-protein fungerar normalt eller inte. I artiklarna III, V och VI beskriver vi utveckl- ingen av två separata (och strukturellt olika) serier av inhibitorer; molekyler som binder till NUDT15 och inaktiverar proteinet genom att förhindra tiopurinerna att binda till det. Vi visar också att effekterna av tiopurinbehandling ökar kraftigt i närvaro av dessa inhibitorer. I artiklarna IV och V visar vi att NUDT15 inte bara påverkar behandling med tiopurin, utan att det även påverkar behandling med de antivirala läkeme- delen acyklovir och ganciklovir. Dessa liknar tiopurinerna och har snarlik funktion eftersom de också byggs in i DNA-molekylen; men de byggs före- trädesvis in i virus-DNA. Dessa läkemedel är bland annat viktiga vid ben- märgstransplantation, där cytomegaloviruset (CMV) ofta orsakar problem. Precis som för tipourinerna visar vi att NUDT15 bryter ner dessa läkemedel innan de kan byggas in i DNA-molekylen och att effekten av dessa läkemedel förstärks av NUDT15-inhibitorer. I alla dessa studier gav modeller av proteinstrukturer, bestämda med röntgen- kristallografi, viktig insikt i hur dessa läkemedel interagerar med proteinet på atomnivå. Forskare har under lång tid arbetat med att försöka få strukturin- formation för kliniskt relevanta NUDT15-varianter. De var dock inte tillräck- ligt stabila för att kristallisera, vilket är ett nödvändigt steg i strukturbestäm- ningsprocessen. Efter att vi noterat att proteinet var mycket mer stabilt med en av våra inhibitorer bunden, kunde vi använda detta protein-inhibitorkom- plex till att bestämma högupplösta strukturer för fyra kliniskt relevanta NUDT15-varianter. Vi kunde även visa hur proteinet påverkas av de kända mutationerna, och hur detta gör att vissa patienter är extremt känsliga för vissa läkemedel; som tiopuriner och de antivirala läkemedlen acyklovir och ganciklovir.

Acknowledgements

There are many people I would like to thank for their contributions to this work. First and foremost, I would like to thank Pål for taking me in for my Master’s project and for giving me the opportunity to pursue a PhD and being able to work on these projects. You’ve been a great PI and mentor to me and the last 6 years have been a fruitful learning experience. You and Martin have built a harmonious and fun extended re-search group here at DBB and it has been a great pleasure for me to have been a part of it.

Very special thanks to Emma, you’ve been of great help with writing and proof-reading. I really appreciate all the help you’ve provided. Thank you Dan, for your help with the Swedish popular science summary.

I would like to thank all my co-authors and collaborators from the Sjöberg, Helleday and Yang groups. Special thanks to Inna, the RNR project would not have worked out the way it did if you had not jumped in. Thank you also to Margareta for all the help with the light absorption measurements.

I would like to thank all the members of the StenHög groups. Robert, Juli- ane, Soni, Esra and Terezia, it’s been great working with you all. Big thanks to Geoffrey for showing me the ropes when I was still a Master’s student. Special thanks to Markel and Jana for helping me out with the CryoEm attempts and thank you Jonathan for helping me with my data processing woes. Thank you, Riccardo, Vivek, Sara and Rohit for all the office banter. Thanks also to Hugo, Kristine and Juliane for all the helpful advice with the RNR project and otherwise. Thank you also to all the former StenHögers, Mohit, Oskar, Linda, Hen- ryk, Ben, Daniel, Karin and Camilla.

Thank you to the members of the Daley group, especially Claudio and Pat- rick for hosting me while office desks were scarce.

I also wish to thank Pia, the DBB administration staff and Matt for all the help you’ve provided over the years.

49 Thank you to my fiancé, Penelope. You have made the path to obtaining a PhD more fun and meaningful. You were my pillar and you believed in me through thick and thin. Thank you for bringing joy and laughter into my life. I look forward to spending many more fun years with you.

Thank you to my parents, without you I would not be who I am today. Mama, Papa, you have taught me to be my own person and you have always supported me all the way. Thank you for all the advice and guidance.

References

1. Jordan, A. & Reichard, P. Ribonucleotide reductases. Annu. Rev. Biochem. 67, 71–98 (1998). 2. Ehrenberg, A. & Reichard, P. Electron spin resonance of the iron- containing protein B2 from ribonucleotide reductase. J. Biol. Chem. 247, 3485–3488 (1972). 3. Sjöberg, B. M., Reichard, P., Gräslund, A. & Ehrenberg, A. The tyrosine free radical in ribonucleotide reductase from Escherichia coli. J. Biol. Chem. 253, 6863–5 (1978). 4. Reichard, P., Baldesten, A. & Rutberg, L. Formation of deoxycytidine phosphates from cytidine phosphates in extracts from Escherichia coli. J. Biol. Chem. 236, 1150–7 (1961). 5. Reichard, P. Enzymatic synthesis of deoxyribonucleotides. I. Formation of deoxycytidine diphosphate from cytidine diphosphate with enzymes from Escherichia coli. J. Biol. Chem. 237, 3513–9 (1962). 6. Nordlund, P., Sjöberg, B. M. & Eklund, H. Three-dimensional structure of the free radical protein of ribonucleotide reductase. Nature 345, 593–598 (1990). 7. Ehrenberg, A. & Reichard, P. Electron Spin Resonance of the Iron- containing Protein B2 from Ribonucleotide Reductase Electron B2 from Spin Resonance of the Iron-containing Ribonucleotide Reductase * Protein. J. Biol. Chem. 247, 3485–3488 (1972). 8. Högbom, M., Sjöberg, B. & Berggren, G. Radical Enzymes. in eLS 375–393 (Wiley, 2020). doi:10.1002/9780470015902.a0029205 9. Cotruvo, J. A. & Stubbe, J. Class I ribonucleotide reductases: metallocofactor assembly and repair in vitro and in vivo. Annu. Rev. Biochem. 80, 733–67 (2011). 10. Huang, M., Parker, M. J. & Stubbe, J. Choosing the right metal: case studies of class I ribonucleotide reductases. J. Biol. Chem. 289, 28104– 11 (2014). 11. Högbom, M. et al. Displacement of the tyrosyl radical cofactor in ribonucleotide reductase obtained by single-crystal high-field EPR and 1.4-A x-ray data. Proc. Natl. Acad. Sci. U. S. A. 100, 3209–14 (2003). 12. Kang, G., Taguchi, A. T., Stubbe, J. & Drennan, C. L. Structure of a trapped radical transfer pathway within a ribonucleotide reductase holocomplex. Science 368, 424–427 (2020).

51 13. Uhlin, U. & Eklund, H. Structure of ribonucleotide reductase protein R1. Nature 370, 533–9 (1994). 14. Uhlin, U. & Eklund, H. The ten-stranded β/α barrel in ribonucleotide reductase protein R1. J. Mol. Biol. 262, 358–369 (1996). 15. Licht, S., Gerfen, G. J. & Stubbe, J. Thiyl radicals in ribonucleotide reductases. Science 271, 477–81 (1996). 16. Thelander, L. Reaction mechanism of ribonucleoside diphosphate reductase from Escherichia coli. Oxidation-reduction-active disulfides in the B1 subunit. J. Biol. Chem. 249, 4858–62 (1974). 17. Aberg, A. et al. Evidence for two different classes of redox-active cysteines in ribonucleotide reductase of Escherichia coli. J. Biol. Chem. 264, 12249–52 (1989). 18. Mao, S. S. et al. A model for the role of multiple cysteine residues involved in ribonucleotide reduction: amazing and still confusing. Biochemistry 31, 9733–43 (1992). 19. Holmgren, A. & Björnstedt, M. Thioredoxin and thioredoxin reductase. Methods Enzymol. 252, 199–208 (1995). 20. Holmgren, A. & Aslund, F. Glutaredoxin. Methods Enzymol. 252, 283–92 (1995). 21. Greene, B. L. et al. Ribonucleotide Reductases: Structure, Chemistry, and Metabolism Suggest New Therapeutic Targets. Annu. Rev. Biochem. 89, 45–75 (2020). 22. Reichard, P. Interactions between deoxyribonucleotide and DNA synthesis. Annu. Rev. Biochem. 57, 349–374 (1988). 23. Mathews, C. K. DNA precursor metabolism and genomic stability. FASEB J. 20, 1300–1314 (2006). 24. Meuth, M. The molecular basis of mutations induced by deoxyribonucleoside triphosphate pool imbalances in mammalian cells. Exp. Cell Res. 181, 305–16 (1989). 25. Rampazzo, C. et al. Regulation by degradation, a cellular defense against deoxyribonucleotide pool imbalances. Mutat. Res. 703, 2–10 (2010). 26. Larsson, A. & Reichard, P. Allosteric effects and substrate specificity of the ribonucleoside diphosphate reductase system from Escherichia coli B. Biochim. Biophys. Acta 113, 407–8 (1966). 27. Nordlund, P. & Reichard, P. Ribonucleotide Reductases. Annu. Rev. Biochem. 75, 681–706 (2006). 28. Hofer, A., Crona, M., Logan, D. T. & Sjöberg, B.-M. DNA building blocks: keeping control of manufacture. Crit. Rev. Biochem. Mol. Biol. 47, 50–63 (2012). 29. Brown, N. C. & Reichard, P. Role of effector binding in allosteric control of ribonucleoside diphosphate reductase. J. Mol. Biol. 46, 39– 55 (1969). 30. Fairman, J. W. et al. Structural basis for allosteric regulation of human ribonucleotide reductase by nucleotide-induced oligomerization. Nat. 52

Struct. Mol. Biol. 18, 316–322 (2011). 31. Eriksson, M. et al. Binding of allosteric effectors to ribonucleotide reductase protein R1: reduction of active-site cysteines promotes substrate binding. Structure 5, 1077–92 (1997). 32. Thelander, L. & Reichard, P. Reduction of ribonucleotides. Annu. Rev. Biochem. 48, 133–158 (1979). 33. Eriksson, M. et al. Binding of allosteric effectors to ribonucleotide reductase protein R1: reduction of active-site cysteines promotes substrate binding. Structure 5, 1077–92 (1997). 34. Eklund, H., Uhlin, U., Färnegårdh, M., Logan, D. T. & Nordlund, P. Structure and function of the radical enzyme ribonucleotide reductase. Prog. Biophys. Mol. Biol. 77, 177–268 (2001). 35. Xu, H., Faber, C., Uchiki, T., Racca, J. & Dealwis, C. Structures of eukaryotic ribonucleotide reductase I define gemcitabine diphosphate binding and subunit assembly. Proc. Natl. Acad. Sci. U. S. A. 103, 4028–33 (2006). 36. Bianchi, V., Pontis, E. & Reichard, P. Changes of deoxyribonucleoside triphosphate pools induced by hydroxyurea and their relation to DNA synthesis. J. Biol. Chem. 261, 16037–42 (1986). 37. Aravind, L., Wolf, Y. I. & Koonin, E. V. The ATP-cone: an evolutionarily mobile, ATP-binding regulatory domain. J. Mol. Microbiol. Biotechnol. 2, 191–4 (2000). 38. Jonna, V. R. et al. Diversity in Overall Activity Regulation of Ribonucleotide Reductase. J. Biol. Chem. 290, 17339–48 (2015). 39. Brown, N. C. & Reichard, P. Ribonucleoside diphosphate reductase: Formation of active and inactive complexes of proteins B1 and B2. J. Mol. Biol. 46, 25–38 (1969). 40. Zimanyi, C. M., Chen, P. Y.-T., Kang, G., Funk, M. A. & Drennan, C. L. Molecular basis for allosteric specificity regulation in class Ia ribonucleotide reductase from Escherichia coli. Elife 5, e07141 (2016). 41. Lundin, D., Berggren, G., Logan, D. T. & Sjöberg, B.-M. The origin and evolution of ribonucleotide reduction. Life (Basel, Switzerland) 5, 604–36 (2015). 42. Fontecave, M., Eliasson, R. & Reichard, P. Oxygen-sensitive ribonucleoside triphosphate reductase is present in anaerobic Escherichia coli. Proc. Natl. Acad. Sci. U. S. A. 86, 2147–51 (1989). 43. Sun, X. et al. The free radical of the anaerobic ribonucleotide reductase from Escherichia coli is at glycine 681. J. Biol. Chem. 271, 6827–31 (1996). 44. Ollagnier, S. et al. The anaerobic Escherichia coli ribonucleotide reductase. Subunit structure and iron sulfur center. J. Biol. Chem. 271, 9410–6 (1996). 45. Jordan, A., Gibert, I. & Barbé, J. Cloning and sequencing of the genes from Salmonella typhimurium encoding a new bacterial ribonucleotide reductase. J. Bacteriol. 176, 3420–7 (1994).

53 46. Jordan, A. et al. A second class I ribonucleotide reductase in Enterobacteriaceae: characterization of the Salmonella typhimurium enzyme. Proc. Natl. Acad. Sci. U. S. A. 91, 12892–6 (1994). 47. King, D. S. & Reichard, P. Mass spectrometric determination of the radical scission site in the anaerobic ribonucleotide reductase of Escherichia coli. Biochem. Biophys. Res. Commun. 206, 731–5 (1995). 48. Young, P., Ohman, M., Xu, M. Q., Shub, D. A. & Sjöberg, B. M. Intron-containing T4 bacteriophage gene sunY encodes an anaerobic ribonucleotide reductase. J. Biol. Chem. 269, 20229–32 (1994). 49. Lundin, D., Gribaldo, S., Torrents, E., Sjöberg, B.-M. & Poole, A. M. Ribonucleotide reduction - horizontal transfer of a required function spans all three domains. BMC Evol. Biol. 10, 383 (2010). 50. Reichard, P. From RNA to DNA, why so many ribonucleotide reductases? Science 260, 1773–1777 (1993). 51. Högbom, M. Metal use in ribonucleotide reductase R2, di-iron, di- manganese and heterodinuclear--an intricate bioinorganic workaround to use different metals for the same reaction. Metallomics 3, 110–120 (2011). 52. Boal, A. K., Cotruvo, J. A., Stubbe, J. & Rosenzweig, A. C. Structural basis for activation of class Ib ribonucleotide reductase. Science 329, 1526–30 (2010). 53. Cotruvo, J. A. & Stubbe, J. NrdI, a flavodoxin involved in maintenance of the diferric-tyrosyl radical cofactor in Escherichia coli class Ib ribonucleotide reductase. Proc. Natl. Acad. Sci. U. S. A. 105, 14383–8 (2008). 54. Jiang, W. et al. A manganese(IV)/iron(III) cofactor in Chlamydia trachomatis ribonucleotide reductase. Science 316, 1188–91 (2007). 55. Bollinger, J. M., Jiang, W., Green, M. T. & Krebs, C. The manganese(IV)/iron(III) cofactor of Chlamydia trachomatis ribonucleotide reductase: structure, assembly, radical initiation, and evolution. Curr. Opin. Struct. Biol. 18, 650–657 (2008). 56. Grinberg, I. R. et al. A glutaredoxin domain fused to the radical- generating subunit of ribonucleotide reductase (RNR) functions as an efficient RNR reductant. J. Biol. Chem. 293, 15889–15900 (2018). 57. Rozman Grinberg, I. et al. Novel ATP-cone-driven allosteric regulation of ribonucleotide reductase via the radical-generating subunit. Elife 7, 1–26 (2018). 58. Rozman Grinberg, I. et al. Class Id ribonucleotide reductase utilizes a Mn2(IV,III) cofactor and undergoes large conformational changes on metal loading. J. Biol. Inorg. Chem. 24, 863–877 (2019). 59. Srinivas, V. et al. Metal-free ribonucleotide reduction powered by a DOPA radical in Mycoplasma pathogens. Nature 563, 416–420 (2018). 60. Deckert, G. et al. The complete genome of the hyperthermophilic bacterium Aquifex aeolicus. Nature 392, 353–358 (1998). 54

61. Guiral, M. et al. The hyperthermophilic bacterium Aquifex aeolicus: from respiratory pathways to extremely resistant enzymes and biotechnological applications. Adv. Microb. Physiol. 61, 125–94 (2012). 62. Chen, L., Benner, J. & Perler, F. B. Protein splicing in the absence of an intein penultimate histidine. J. Biol. Chem. 275, 20431–5 (2000). 63. Mills, K. V., Johnson, M. A. & Perler, F. B. Protein splicing: How Inteins escape from precursor proteins. J. Biol. Chem. 289, 14498– 14505 (2014). 64. Shah, N. H. & Muir, T. W. Inteins: Nature’s Gift to Protein Chemists. Chem. Sci. 5, 446–461 (2014). 65. Novikova, O. et al. Intein clustering suggests functional importance in different domains of life. Mol. Biol. Evol. 33, 783–799 (2016). 66. Martínez-Carranza, M. et al. A ribonucleotide reductase from Clostridium botulinum reveals distinct evolutionary pathways to regulation via the overall activity site. J. Biol. Chem. 295, 15576– 15587 (2020). 67. Johansson, R. et al. Structural Mechanism of Allosteric Activity Regulation in a Ribonucleotide Reductase with Double ATP Cones. Structure 24, 906–17 (2016). 68. Bessman, M. J., Frick, D. N. & O’Handley, S. F. The MutT Proteins or “Nudix” Hydrolases, a Family of Versatile, Widely Distributed, “Housecleaning” Enzymes. J. Biol. Chem. 271, 25059–25062 (1996). 69. Rudd, S. G., Valerie, N. C. K. & Helleday, T. Pathways controlling dNTP pools to maintain genome stability. DNA Repair (Amst). 44, 193–204 (2016). 70. Bhatnagar, S. K. & Bessman, M. J. Studies on the mutator gene, mutT of Escherichia coli. Molecular cloning of the gene, purification of the gene product, and identification of a novel nucleoside triphosphatase. J. Biol. Chem. 263, 8953–7 (1988). 71. Bhatnagar, S. K., Bullions, L. C. & Bessman, M. J. Characterization of the mutT nucleoside triphosphatase of Escherichia coli. J. Biol. Chem. 266, 9050–4 (1991). 72. Abeygunawardana, C. et al. Solution structure of the MutT enzyme, a nucleoside triphosphate pyrophosphohydrolase. Biochemistry 34, 14997–5005 (1995). 73. Treffers, H. P., Spinelli, V. & Belser, N. O. A Factor (or Mutator Gene) Influencing Mutation Rates in Escherichia Coli. Proc. Natl. Acad. Sci. U. S. A. 40, 1064–71 (1954). 74. Srouji, J. R., Xu, A., Park, A., Kirsch, J. F. & Brenner, S. E. The evolution of function within the Nudix homology clan. Proteins 85, 775–811 (2017). 75. McLennan, A. G. The Nudix hydrolase superfamily. Cell. Mol. Life Sci. 63, 123–43 (2006). 76. Koonin, E. V. A highly conserved sequence motif defining the family

55 of MutT-related proteins from eubacteria, eukaryotes and viruses. Nucleic Acids Res. 21, 4847 (1993). 77. Mildvan, A. S. et al. Structures and mechanisms of Nudix hydrolases. Arch. Biochem. Biophys. 433, 129–143 (2005). 78. Yanofsky, C., Cox, E. C. & Horn, V. The unusual mutagenic specificity of an E. Coli mutator gene. Proc. Natl. Acad. Sci. U. S. A. 55, 274–81 (1966). 79. Maki, H. & Sekiguchi, M. MutT protein specifically hydrolyses a potent mutagenic substrate for DNA synthesis. Nature 355, 273–5 (1992). 80. Gad, H. et al. MTH1 inhibition eradicates cancer by preventing sanitation of the dNTP pool. Nature 508, 215–21 (2014). 81. Furuichi, M. et al. Genomic structure and chromosome location of the human mutT homologue gene MTH1 encoding 8-oxo-dGTPase for prevention of A:T to C:G transversion. Genomics 24, 485–90 (1994). 82. Mo, J. Y., Maki, H. & Sekiguchi, M. Hydrolytic elimination of a mutagenic nucleotide, 8-oxodGTP, by human 18-kilodalton protein: sanitization of nucleotide pool. Proc. Natl. Acad. Sci. U. S. A. 89, 11021–5 (1992). 83. Sakumi, K. et al. Cloning and expression of cDNA for a human enzyme that hydrolyzes 8-oxo- dGTP, a mutagenic substrate for DNA synthesis. J. Biol. Chem. 268, 23524–23530 (1993). 84. Fujikawa, K. et al. The oxidized forms of dATP are substrates for the human MutT homologue, the hMTH1 protein. J. Biol. Chem. 274, 18201–18205 (1999). 85. Fujikawa, K., Kamiya, H., Yakushiji, H., Nakabeppu, Y. & Kasai, H. Human MTH1 protein hydrolyzes the oxidized ribonucleotide, 2- hydroxy-ATP. Nucleic Acids Res. 29, 449–54 (2001). 86. Carreras-Puigvert, J. et al. A comprehensive structural, biochemical and biological profiling of the human NUDIX hydrolase family. Nat. Commun. 8, 1541 (2017). 87. Jemth, A.-S. et al. MutT homologue 1 (MTH1) catalyzes the hydrolysis of mutagenic O6-methyl-dGTP. Nucleic Acids Res. 46, 10888–10904 (2018). 88. Scaletti, E. R. et al. MutT homologue 1 (MTH1) removes N6-methyl- dATP from the dNTP pool. J. Biol. Chem. 295, 4761–4772 (2020). 89. Rai, P. et al. Enhanced elimination of oxidized guanine nucleotides inhibits oncogenic RAS-induced DNA damage and premature senescence. Oncogene 30, 1489–96 (2011). 90. Zhang, Y. et al. Redox control of the survival of healthy and diseased cells. Antioxid. Redox Signal. 15, 2867–908 (2011). 91. Topal, M. D. & Baker, M. S. DNA precursor pool: a significant target for N-methyl-N-nitrosourea in C3H/10T1/2 clone 8 cells. Proc. Natl. Acad. Sci. U. S. A. 79, 2211–2215 (1982). 92. Carter, M. et al. Crystal structure, biochemical and cellular activities 56

demonstrate separate functions of MTH1 and MTH2. Nat. Commun. 6, 7871 (2015). 93. Nagy, G. N., Leveles, I. & Vértessy, B. G. Preventive DNA repair by sanitizing the cellular (deoxy)nucleoside triphosphate pool. FEBS J. 281, 4207–4223 (2014). 94. Tsuzuki, T. et al. Spontaneous tumorigenesis in mice defective in the MTH1 gene encoding 8-oxo-dGTPase. Proc. Natl. Acad. Sci. U. S. A. 98, 11456–61 (2001). 95. Cai, J. P., Ishibashi, T., Takagi, Y., Hayakawa, H. & Sekiguchi, M. Mouse MTH2 protein which prevents mutations caused by 8- oxoguanine nucleotides. Biochem. Biophys. Res. Commun. 305, 1073– 1077 (2003). 96. Hori, M., Satou, K., Harashima, H. & Kamiya, H. Suppression of mutagenesis by 8-hydroxy-2’-deoxyguanosine 5’-triphosphate (7,8- dihydro-8-oxo-2’-deoxyguanosine 5’-triphosphate) by human MTH1, MTH2, and NUDT5. Free Radic. Biol. Med. 48, 1197–201 (2010). 97. Hitchings, G. H., Elion, G. B., Falco, E. A., Russell, P. B. & VanderWerff, H. STUDIES ON ANALOGS OF PURINES AND PYRIMIDINES. Ann. N. Y. Acad. Sci. 52, 1318–1335 (1950). 98. Elion, G. B. Nobel Lecture. The purine path to chemotherapy. Biosci. Rep. 9, 509–29 (1989). 99. Chen, Y. & Lu, G. P. Advances in the diagnosis and treatment of pediatric acute respiratory distress syndrome. Chinese J. Contemp. Pediatr. 20, 717–723 (2018). 100. Schmiegelow, K., Nielsen, S. N., Frandsen, T. L. & Nersting, J. Mercaptopurine/Methotrexate maintenance therapy of childhood acute lymphoblastic leukemia: clinical facts and fiction. J. Pediatr. Hematol. Oncol. 36, 503–17 (2014). 101. Bradford, K. & Shih, D. Q. Optimizing 6-mercaptopurine and azathioprine therapy in the management of inflammatory bowel disease. World J. Gastroenterol. 17, 4166–73 (2011). 102. Karran, P. Thiopurines, DNA damage, DNA repair and therapy-related cancer. Br. Med. Bull. 79–80, 153–170 (2006). 103. Nelson, J. A., Carpenter, J. W., Rose, L. M. & Adamson, D. J. Mechanisms of action of 6-thioguanine, 6-mercaptopurine, and 8- azaguanine. Cancer Res. 35, 2872–8 (1975). 104. Ling, Y. H., Nelson, J. A., Cheng, Y. C., Anderson, R. S. & Beattie, K. L. 2’-Deoxy-6-thioguanosine 5’-triphosphate as a substrate for purified human DNA polymerases and calf thymus terminal deoxynucleotidyltransferase in vitro. Mol. Pharmacol. 40, 508–14 (1991). 105. Warren, D. J., Andersen, A. & Slørdal, L. Quantitation of 6- thioguanine residues in peripheral blood leukocyte DNA obtained from patients receiving 6-mercaptopurine-based maintenance therapy. Cancer Res. 55, 1670–4 (1995).

57 106. Cuffari, C., Li, D. Y., Mahoney, J., Barnes, Y. & Bayless, T. M. Peripheral blood mononuclear cell DNA 6-thioguanine metabolite levels correlate with decreased interferon-gamma production in patients with Crohn’s disease on AZA therapy. Dig. Dis. Sci. 49, 133– 7 (2004). 107. Swann, P. F. et al. Role of postreplicative DNA mismatch repair in the cytotoxic action of thioguanine. Sci. (New York, NY) 273, 1109–1111 (1996). 108. Aquilina, G. et al. Tolerance to O6-methylguanine and 6-thioguanine cytotoxic effects: a cross-resistant phenotype in N-methylnitrosourea- resistant Chinese hamster ovary cells. Cancer Res. 50, 4248–53 (1990). 109. Yan, T., Berry, S. E., Desai, A. B. & Kinsella, T. J. DNA mismatch repair (MMR) mediates 6-thioguanine genotoxicity by introducing single-strand breaks to signal a G2-M arrest in MMR-proficient RKO cells. Clin. Cancer Res. 9, 2327–2334 (2003). 110. Relling, M. V. et al. Clinical Pharmacogenetics Implementation Consortium Guideline for Thiopurine Dosing Based on TPMT and NUDT15 Genotypes: 2018 Update. Clin. Pharmacol. Ther. 105, 1095–1105 (2019). 111. Relling, M. V et al. Mercaptopurine Therapy Intolerance and Heterozygosity at the Thiopurine S-Methyltransferase Gene Locus. JNCI J. Natl. Cancer Inst. 91, 2001–2008 (1999). 112. Nygaard, U., Toft, N. & Schmiegelow, K. Methylated metabolites of 6-mercaptopurine are associated with hepatotoxicity. Clin. Pharmacol. Ther. 75, 274–81 (2004). 113. Evans, W. E. Pharmacogenetics of thiopurine S-methyltransferase and thiopurine therapy. Ther. Drug Monit. 26, 186–91 (2004). 114. Relling, M. V. et al. Clinical Pharmacogenetics Implementation Consortium guidelines for thiopurine methyltransferase genotype and thiopurine dosing. Clin. Pharmacol. Ther. 89, 387–91 (2011). 115. Mowat, C. et al. Guidelines for the management of inflammatory bowel disease in adults. Gut 60, 571–607 (2011). 116. Relling, M. V, Pui, C.-H., Cheng, C. & Evans, W. E. Thiopurine methyltransferase in acute lymphoblastic leukemia. Blood 107, 843–4 (2006). 117. Evans, W. E. & Relling, M. V. Pharmacogenomics: translating functional genomics into rational therapeutics. Science 286, 487–91 (1999). 118. Evans, W. E. & Relling, M. V. Moving towards individualized medicine with pharmacogenomics. Nature 429, 464–468 (2004). 119. Giacomini, K. M. et al. Pharmacogenomics and patient care: one size does not fit all. Sci. Transl. Med. 4, 153ps18 (2012). 120. Relling, M. V et al. Mercaptopurine therapy intolerance and heterozygosity at the thiopurine S-methyltransferase gene locus. J. 58

Natl. Cancer Inst. 91, 2001–8 (1999). 121. Evans, W. E. et al. Preponderance of thiopurine S-methyltransferase deficiency and heterozygosity among patients intolerant to mercaptopurine or azathioprine. J. Clin. Oncol. 19, 2293–301 (2001). 122. Moyer, A. M. NUDT15: A bench to bedside success story. Clin. Biochem. (2021). doi:10.1016/j.clinbiochem.2021.02.007 123. Colombel, J. F. et al. Genotypic analysis of thiopurine S- methyltransferase in patients with Crohn’s disease and severe myelosuppression during azathioprine therapy. Gastroenterology 118, 1025–30 (2000). 124. Booth, R. A. et al. Assessment of thiopurine methyltransferase activity in patients prescribed azathioprine or other thiopurine-based drugs. Evid. Rep. Technol. Assess. (Full. Rep). 1–282 (2010). 125. Cooper, S. C., Ford, L. T., Berg, J. D. & Lewis, M. J. V. Ethnic variation of thiopurine S-methyltransferase activity: a large, prospective population study. Pharmacogenomics 9, 303–9 (2008). 126. Yang, S.-K. et al. A common missense variant in NUDT15 confers susceptibility to thiopurine-induced leukopenia. Nat. Genet. 46, 1017– 20 (2014). 127. Kakuta, Y. et al. NUDT15 R139C causes thiopurine-induced early severe hair loss and leukopenia in Japanese patients with IBD. Pharmacogenomics J. 16, 280–5 (2016). 128. Yang, J. J. et al. Inherited NUDT15 variant is a genetic determinant of mercaptopurine intolerance in children with acute lymphoblastic leukemia. J. Clin. Oncol. 33, 1235–1242 (2015). 129. Tanaka, Y. et al. Susceptibility to 6-MP toxicity conferred by a NUDT15 variant in Japanese children with acute lymphoblastic leukaemia. Br. J. Haematol. 171, 109–115 (2015). 130. Chiengthong, K. et al. NUDT15 c.415C>T increases risk of 6- mercaptopurine induced myelosuppression during maintenance therapy in children with acute lymphoblastic leukemia. Haematologica 101, e24-6 (2016). 131. Karran, P. & Attard, N. Thiopurines in current medical practice: molecular mechanisms and contributions to therapy-related cancer. Nat. Rev. Cancer 8, 24–36 (2008). 132. Karczewski, K. J. et al. The mutational constraint spectrum quantified from variation in 141,456 humans. Nature 581, 434–443 (2020). 133. Hartl, F. U., Bracher, A. & Hayer-Hartl, M. Molecular chaperones in protein folding and proteostasis. Nature 475, 324–32 (2011). 134. You, C., Dai, X., Yuan, B. & Wang, Y. Effects of 6-thioguanine and S6-methylthioguanine on transcription in vitro and in human cells. J. Biol. Chem. 287, 40915–40923 (2012). 135. Stojic, L. et al. Mismatch repair-dependent G 2 checkpoint induced by low doses of S N 1 type methylating agents requires the ATR kinase. Genes Dev. 18, 1331–1344 (2004).

59 136. York, S. J. & Modrich, P. Mismatch repair-dependent iterative excision at irreparable O 6-methylguanine lesions in human nuclear extracts. J. Biol. Chem. 281, 22674–22683 (2006). 137. Li, G. M. The role of mismatch repair in DNA damage-induced apoptosis. Oncol. Res. 11, 393–400 (1999). 138. Martinez Molina, D. et al. Monitoring drug target engagement in cells and tissues using the cellular thermal shift assay. Science 341, 84–7 (2013). 139. Lomenick, B. et al. Target identification using drug affinity responsive target stability (DARTS). Proc. Natl. Acad. Sci. U. S. A. 106, 21984– 9 (2009). 140. Elion, G. B. Mechanism of action and selectivity of acyclovir. Am. J. Med. 73, 7–13 (1982). 141. De Clercq, E. & Field, H. J. Antiviral prodrugs - the development of successful prodrug strategies for antiviral chemotherapy. Br. J. Pharmacol. 147, 1–11 (2006). 142. Ashton, W. T., Karkas, J. D., Field, A. K. & Tolman, R. L. Activation by thymidine kinase and potent antiherpetic activity of 2’-nor-2’- deoxyguanosine (2’NDG). Biochem. Biophys. Res. Commun. 108, 1716–21 (1982). 143. McGavin, J. K. & Goa, K. L. Ganciclovir: an update of its use in the prevention of cytomegalovirus infection and disease in transplant recipients. Drugs 61, 1153–83 (2001). 144. Sullivan, V. et al. A protein kinase homologue controls phosphorylation of ganciclovir in human cytomegalovirus-infected cells. Nature 358, 162–4 (1992). 145. Hible, G., Daalova, P., Gilles, A.-M. & Cherfils, J. Crystal structures of GMP kinase in complex with ganciclovir monophosphate and Ap5G. Biochimie 88, 1157–64 (2006). 146. Chen, H., Beardsley, G. P. & Coen, D. M. Mechanism of ganciclovir- induced chain termination revealed by resistant viral polymerase mutants with reduced exonuclease activity. Proc. Natl. Acad. Sci. U. S. A. 111, 17462–7 (2014). 147. St Clair, M. H., Miller, W. H., Miller, R. L., Lambe, C. U. & Furman, P. A. Inhibition of cellular alpha DNA polymerase and herpes simplex virus-induced DNA polymerases by the triphosphate of BW759U. Antimicrob. Agents Chemother. 25, 191–194 (1984). 148. Rubsam, L. Z., Davidson, B. L. & Shewach, D. S. Superior cytotoxicity with ganciclovir compared with acyclovir and 1-beta-D- arabinofuranosylthymine in herpes simplex virus-thymidine kinase- expressing cells: a novel paradigm for cell killing. Cancer Res. 58, 3873–82 (1998). 149. Karjoo, Z., Chen, X. & Hatefi, A. Progress and problems with the use of suicide genes for targeted cancer therapy. Adv. Drug Deliv. Rev. 99, 113–128 (2016). 60

150. Fillat, C., Carrió, M., Cascante, A. & Sangro, B. Suicide gene therapy mediated by the Herpes Simplex virus thymidine kinase gene/Ganciclovir system: fifteen years of application. Curr. Gene Ther. 3, 13–26 (2003). 151. Moriyama, T. et al. NUDT15 polymorphisms alter thiopurine metabolism and hematopoietic toxicity. Nat. Genet. 48, 367–73 (2016). 152. Yang, J. J. et al. Pharmacogene Variation Consortium Gene Introduction: NUDT15. Clin. Pharmacol. Ther. 105, 1091–1094 (2019). 153. Moriyama, T. et al. Novel variants in NUDT15 and thiopurine intolerance in children with acute lymphoblastic leukemia from diverse ancestry. Blood 130, 1209–1212 (2017). 154. Kakuta, Y. et al. NUDT15 codon 139 is the best pharmacogenetic marker for predicting thiopurine-induced severe adverse events in Japanese patients with inflammatory bowel disease: a multicenter study. J. Gastroenterol. 53, 1065–1078 (2018). 155. Suiter, C. C. et al. Massively parallel variant characterization identifies NUDT15 alleles associated with thiopurine toxicity. Proc. Natl. Acad. Sci. U. S. A. 117, 5394–5401 (2020).