Richard A. Ward Medicinal Chemistry , Oncology, IMED Biotech Unit, AstraZeneca, Cambridge, UK.

1. The covalent targeting of cysteine residues in drug discovery and its application to the discovery of osimertinib (TagrissoTM) Richard A. Ward Medicinal Chemistry, Oncology, IMED Biotech Unit, AstraZeneca, Cambridge, UK. BigChem Training School, Gothenburg May 2019

2. Targeted covalent inhibition • A covalent inhibitor forms a reversible or irreversible bond to the target • Potential gains in: • Potency • Selectivity • Duration of action • Cysteine targeting: • Relatively rare and often poorly conserved • Nucleophilic • Two important steps: The has been an explosion of interest around covalent inhibitors in recent years

3. Targeted covalent inhibition strategies EARLY ADDITION OF COVALENT GROUP LATE

4. Designing covalent inhibitors: optimizing k2 The right geometry Can the warhead get close enough to the target cysteine for reaction to take place? The right warhead Too reactive – toxicity issues Not reactive enough – potency issues The right cysteine Are there cysteines in the binding pocket that can react with a warhead? Conformation of cysteine Warhead vector Tune reactivity by varying substituents near warhead Assess local environment of cysteine H2O? C S S S H C C S B-? S C Cys Cys- C

5. Cysteine activation • Thiolate form of cysteine is a strong nucleophile • Deprotonation occurs before reaction with warhead • Cysteines in proteins generally less easily activated than free cysteine in solution • Thiolate form favoured in polar environment • Activation can be assisted by solvent or by interaction with neighbouring residues • e.g. deprotonation by histidine in cysteine proteases Acrylamide warhead D800 C797 • C797 in EGFR: • Solvent exposed • Nearby aspartic acid Could cysteine pKa be an indicator of ease of activation?

6. C781 pKa analysis of cysteines in EGFR • pKa of cysteine sensitive to change in local environment • More solvent exposure -> lower pKa • Large fluctuations in pKa pKa of cysteines – calculated1 over all PDB EGFR X-ray structures C939 C818 C950 C775 C781 C797 C775 C797 pKa of cysteines – calculated* from MD simulation C818 C950 C939 C818 C950 C775 C781 C797 Conformational sampling of cysteines is important Reported experimental pKa of 5.5 1 Finite difference Poisson-Boltzmann/Debye-Huckel (FD/DH) method – Jim Warwicker (Manchester) 2 Truong et al. Cell Chemical Biology (2016), 23, 837-848. C939

7. PredictedpKa alone can be challenging to use prospectively

8. Designing covalent inhibitors: optimizing k2 The right geometry Can the warhead get close enough to the target cysteine for reaction to take place? The right warhead Too reactive – toxicity issues Not reactive enough – potency issues The right cysteine Are there cysteines in the binding pocket that can react with a warhead? Conformation of cysteine Warhead vector Tune reactivity by varying substituents near warhead Assess local environment of cysteine H2O? C S S S H C C S B-? S C Cys Cys- C

9. Glutathione half-life [GSH t1/2] • Half-life to glutathione conjugation is an indicator of relative warhead reactivity • Similar reactivity towards Michael acceptors as cysteine • Useful assay, requires compound to be synthesized • SAR from Hammett parameters useful for rational modification of reactivity of model aromatic acrylamides1 • Cannot be used for saturated compounds, e.g. GSH t1/2 (mins) R2=0.92 σind parameter ARS-8532 Can we predict warhead reactivity towards GSH for a diverse range of compounds? Ward et al., J. Med. Chem. 2013,56, 7025-48 Rosen et al., Science2016, 351, 604-8.

10. QM-based predictions of GSH t1/2 • Energy barrier ΔE‡ to reaction with model compound (MeSH) previously shown to correlate with GSH t1/2 for model compounds1 • Approach employed against our test set (a) • Correlation also observed for ΔE against GSH t1/2(b) • Quicker to calculate than ΔE‡ • Can be more routinely applied to larger molecules ΔE‡ ΔE Accurate but slow and resource intensive • Flanagan et al., J. Med. Chem. 2014,57, 10072-9. • Ward et al., J. Med. Chem. 2013,56, 7025-48

11. Relationship between GSH t1/2 and pKa of parent base GSH t1/2 pKa • Acrylamide reactivity can be increased by making terminal alkene carbon more electrophilic • Internal compound collection searched for matched pairs of acrylamides and amine precursors • Calculated (and experimental) pKa of parent base in good agreement with experiment in majority of cases • Can predict GSH t1/2 for saturated acrylamides – ‘warhead corrections’ can be applied ibrutinib Can tune warhead reactivity rationally using same techniques as for varying pKa Lonsdale, Ward. JCIM, 2017, 57 (12), 3124.

12. Designing covalent inhibitors: optimizing k2 The right geometry Can the warhead get close enough to the target cysteine for reaction to take place? The right warhead Too reactive – toxicity issues Not reactive enough – potency issues The right cysteine Are there cysteines in the binding pocket that can react with a warhead? Conformation of cysteine Warhead vector Tune reactivity by varying substituents near warhead Assess local environment of cysteine H2O? C S S S H C C S B-? S C Cys Cys- C

13. The right geometry: cysteine sidechain conformation • Example project with difficult to target cysteine on edge of binding pocket • Slow covalent binding observed • MD simulations reveal two distinct conformations of targeted cysteine • Cysteine spends the majority of time in an inaccessible conformation out Dihedral angle (degrees) in Time (ps) Accessible conformation Inaccessible conformation Residue conformation may help explain lack of residue targetability

14. The discovery of osimertinib (TagrissoTM)The rational design of an irreversible EGFR T790M mutant specific inhibitor from project start to FDA approval in 6 years

15. EGFR mutant-selective inhibitor hypothesis (2009) • Exploitation of methionine gatekeeper residue to leverage a mutant-selective profile • Desired profile with activity in DM (and AM) and selectivity over WT • Set of EGFR inhibitors screened using a biochemical assay (DM & WT) • Selection of compounds from AZ Projects targeting methionine gatekeepers1 ‘Methionine-GK’ Selection Irreversible AQZ EGFR DM pIC50 AM = L858R or Exon 19 Del (EGFR+) DM = L858R/T790M or Exon 19 Del/T790M Reversible AQZ EGFR WT pIC50

16. An EGFR Mutant-Selective Template Mutant-selective examples identified based on a pyrimidine scaffold 2 1 R1=OMe SBDD 2 1 • Large enzyme-cell drop-off observed • Binding mode of Compound 1 modelled into EGFR-T790M structure • Molecular modelling deployed to target Cys-797 by covalent mechanism • Cellular activity observed in DMand AM 1. Ward, R. A et al, J. Med. Chem. 2013, 56, 7025. EGFR-WT Structure PDB Code: 4LI5

17. Piperazine modifications: selected examples • Cell potent, wild-type selective inhibitors with improved LLE and excellent oral exposure 4 5 3 3 4 5 6 • Excellent double mutant and activating mutant efficacy with margin over WT • Very low oral dose required (10 mpk) • Examples had issues with off-target activities (insulin receptor and/or hERG) Reversible and irreversible off-targets both important for optimisation Finlay et al. AACR-EORTC-NCI Molecular Targets & Cancer Therapeutics Conference, October 2013.

18. 5-Position SAR leading to the discovery of ‘AZD9291’ • Combinations explored with indole head groups and pyrimidine 5-groups • Removal of 5-Cl beneficial for properties and off-target activities • Possible to achieve good DM potency and WT selectivity • AZD9291 selected due to target potency/selectivity, mouse exposure, hERG potency, IGF profile and in vivo tolerability 5-Position SAR leading to the discovery of AZD9291 EGFR DM (Cell) Potency (log µM) IGF1R Enzyme Potency (log µM) hERG Ionworks (log µM) R2=Cl R2=H R2=Cl R2=H R2=Cl R2=H Divergent SAR identified between primary target(s) and off-targets

19. The profile of AZD9291 (Osimertinib/TagrissoTM) H1975 • AZD9291 delivers profound regression in H1975 (DM) and PC9 (AM) xenograft models Finlay & Cross et al, AACR-EORTC-NCI Molecular Targets & Cancer Therapeutics Conference, October 2013.

20. The clinical phases of osimertinib/TAGRISSOTM Rapid clinical development plan in T790M mutation positive non-small cell lung cancer (NSCLC) March 2013: First time in man (FTIM) achieved for TAGRISSOTM (AURA, NCT01802632) Adverse events consistent with EGFR pharmacology Prof M Ranson at The European Cancer Congress 2013; European Journal of Cancer Vol 49, supplement 3: LBA 33 Summer 2013: Clinical activity observed at the first dose Anti-tumour study Nov 2015: TAGRISSOTM receives U.S. FDA approval as 2L therapy 2013-4: Emerging evidence of activity in brain metastases N Engl J Med 2017; 376:629-640, 2017 April 2014: Breakthrough designation granted by the U.S. FDA Brain MRI A) Baseline Aug 2013 B) Oct 2013 April 2018: TAGRISSOTM receives U.S. FDA approval as 1L therapy Kim et al. European Society for Medical Oncology 2014 Congress, 26–30 September, Madrid, Spain; Abstract #6771

21. Summary and Acknowledgements Geometry Can the warhead get close enough to the target cysteine for reaction to take place? Warhead Too reactive – toxicity issues Not reactive enough – potency issues Cysteine Are there cysteines in the binding pocket that can react with a warhead? Use MD simulations to explore conformational flexibility of cysteines close to binding pocket • Use MD and pKa calculations to assess local environment of cysteine • Solvent accessibility • pKa • GSH t1/2 predictions – relative reactivities • Hammett parameters (aromatics) • QM calculations of ΔE and ΔE‡ • pKa of parent base • Matched molecular pairs analysis for warhead swap Thanks to Richard Lonsdale, numerous AZ colleagues and collaborators and the AZ Postdoc Programme for the cysteine targeting work presented I would like to acknowledge all our AstraZeneca colleagues involved in the discovery of osimertinib, of which there are many In addition to our collaborators, clinicians and above all the patients and their families