Roderick E Hubbard Vernalis (R&D) Ltd, Cambridge YSBL & HYMS, Univ of York, UK Fragments 2013

For slides – [email protected]

Progressing fragments for challenging targets

1 Trends for new technologies

• In the beginning –lots of excitement • Which can lead to hype and over‐selling

Expectation

Technology Time Trigger 2 Analysis / terms initially proposed by Gartner group (Murcko) Trends for new technologies

• In the beginning –lots of excitement • Which can lead to hype and over‐selling Inflated expectations Expectation

Technology Time Trigger 3 Analysis / terms initially proposed by Gartner group Trends for new technologies

• Too rapid (often inexpert) deployment • It doesn’t work ‐ disillusionment Inflated expectations Expectation

Technology Trough of Time Trigger Disillusionment 4 Analysis / terms initially proposed by Gartner group Trends for new technologies

• Too rapid (often inexpert) deployment • It doesn’t work ‐ disillusionment Inflated expectations Expectation

Technology Trough of Time Trigger Disillusionment 5 Analysis / terms initially proposed by Gartner group Trends for new technologies

• Eventually, expertise grows • Begin to understand how and where to apply methods Inflated expectations Expectation

Slope of Enlightenment

Technology Trough of Time Trigger Disillusionment 6 Analysis / terms initially proposed by Gartner group Trends for new technologies

• Learn how to integrate the methods into the process • Add to productivity Inflated expectations Expectation

Slope of Plateau of Enlightenment productivity

Technology Trough of Time Trigger Disillusionment 7 Analysis / terms initially proposed by Gartner group Fragments

• The ideas established in the molecular modelling / structural biology community during the 1980s and early 1990s • First reduced to practice by Abbott in SAR by NMR approach in mid 1990s • Other pharmaceutical companies unable to replicate success • Approaches developed in small pharma companies in late 1990s / early 2000s • Astex, Vernalis, Plexxikon, SGX ….. and underground in large pharma … • Underpinning concepts developed in the 2000s –complexity, ligand efficiency • Success has led to increased use • Different aspects of FBLD are on different parts of this curve • And in different organisatiions Inflated Expectation expectations Slope of Plateau of Enlightenment productivity

Technology Trough of Time 8 Trigger Disillusionment Fragments 2013 talk

• Summary from 2009 • Where we were 4 years ago • New approaches / ideas for conventional targets • Screening methods • Off‐rate screening for fragment to hit optimisation • How to approach non‐conventional targets • What is a non‐conventional target? • Methods for determining binding mode • Issues – assays, plasticity, compound properties, 3D • Final remarks

9 Fragments 2013 talk

10 Fragments 2009 talk SeeDs process ‐ 2008 Structural Exploitation of Experimental Drug Startpoints* Target Biacore a b c Validation

X-ray Structures

NMR Competitive HSQC Binding Experiment O

2NH O H NH NH N O N N N NH NH NH

O H Wet assay N N H OH N NH O / Biacore NH N N N N NH H NH N

NH O

O OH

N O N NH Fragment Library O O Evolution 11*Hubbard et al (2007), Curr Topics Med Chem, 7, 1568 Fragments 2009 talk Chen & Hubbard (2009), JCAMD, 23, 603 Finding fragments

• Finding fragments that bind is not difficult • A good way of assessing target “ligandability”

8% Low hit rates (< 2%) 7% High hit rates (> 2%) 6%

rate 5% hit protein-protein interaction 4% Kinases (0.4-3% hit rate) 3% (3-5% hit rate) Validated Class 1 hits rates 2% Poor targets 1%

0% 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 12 LigandabilityCalculatecalculatedDscore druggability from structure1 (DScore)

Fragments 2009 talk Chen & Hubbard (2009), JCAMD, 23, 603 Finding fragments

• Finding fragments that bind is not difficult • A good way of assessing target “ligandability” • Low hit rate can indicate difficult to progress • See also Hajduk (2005) J Med Chem, 48, 2518

8% Low hit rates (< 2%) 7% High hit rates (> 2%) 6%

rate 5% hit protein-protein interaction 4% Kinases (0.4-3% hit rate) 3% (3-5% hit rate) Validated Class 1 hits rates 2% Poor targets 1%

0% 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 13 LigandabilityCalculatecalculatedDscore druggability from structure1 (DScore)

Fragments 2009 talk Using fragments

• Growing fragments

14 Fragments 2009 talk Using fragments

• Growing fragments –CHK1 example

Chk-1 IC50 >100µM LE ~ 0.39

15 Fragments 2009 talk Using fragments

• Growing fragments –CHK1 example

Chk-1 IC50 = 5µM LE = 0.39

GI50 HCT116 >80µM pH2AX (MEC) – inactive

16 Fragments 2009 talk Using fragments

• Growing fragments –CHK1 example

Chk-1 IC50 = 0.2µM LE = 0.33

GI50 HCT116 = 4µM pH2AX (MEC) = 7µM

17 Fragments 2009 talk Using fragments

• Growing fragments –CHK1 example

Chk-1 IC50 = 0.013µM LE = 0.39

GI50 HCT116 = 1.8µM pH2AX (MEC) =0.2µM

Series members further optimised 18 to identify Candidate V158411 Fragments 2009 talk Using fragments

• Merging –HSP90 example

Known Ligands Screen Virtual Screening hits

Detailed Design

19 Fragments 2009 talk Brough et al (2009) J Med Chem 52,4794‐4809 HSP90 – BEP800 story Roughley et al (2012) Top Curr Chem 317, 61

VER-82576 Fragment Evolved fragment K58 NVP-BEP800D93 G97

FP IC50=0.058 M NH 2 KD = 0.9nM (SPR) NH2 HCT116 GI50=0.161 M N N BT474 GI =0.057 M N N H 50  N OMe S O L107 O F138 H NH N N VER-26734 VER-52959 2 S

FP IC50>5mM FP IC50=535 M N O Virtual Screening Hit Cl NH 2 N S NH N 2 O Cl O N NH2

NH2 N N Cl OEt O VER-45616 N O N FP IC50=0.9 M N OH N Cl H

VER-41113 OH N O FP IC50=1.56 M NVP-AUY922 20Virtual Screening Hit Vernalis Phase II candidate (FBLD / SBDD derived) Fragments 2009 talk Pin1 Story

• Proline isomerase – persuasive biological rationale that key oncology target • Structure available + D‐peptide tool compound

21 Fragments 2009 talk O Pin1 Story OH

H3C • Proline isomerase – persuasive biological rationale that key oncology target • Fragments identified: fragment to hit evolution • No correlation between biophysical and enzyme assays

22 Fragments 2009 talk O Pin1 Story OH

H3C • Proline isomerase – persuasive biological rationale that key oncology target • Fragments identified: fragment to hit evolution • Issue with over‐binding –multiple copies of fragment binding to the protein –SPR and Xray

23 Fragments 2009 talk HO O O Pin1 Story OH O HN N CH 3 Potter AJ et al, Bioorg Med Chem NH NH Lett. 2010; 20:586‐590 O

24 Fragments 2013 talk

25 Hubbard et al (2007), Curr Topics Med Chem, 7, 1568 Hubbard and Murray (2011), Methods Enzym, 493, 509 The Vernalis process

Target Screen by SPR, DSF, Virtual screen; literature; WAC, biochem or library screen binding assay, Xray

Hits Characterisation Competitive

2NH N S NH NMR screen 2 Fragment Library NH 2 O ~ 1200 compounds N N N NH2 OMe Ave MW 190 O Optimise fragment

OEt H NH N NH2 2 S N N N N H O N Fragment to hit : S Cl O N O SAR by catalog NH2 Drug? Cl O N N Cl N O N off‐rate screening OH H RU RU

N 30 8 Cl 25 N 6 20 4 15 N O 2 10 OH 0 Response Response 5

-2 0 -4 -5 X‐ray or NMR -100 -50 0 50 100 150 200-50 250 300 0 50 100 150 200 250 300 26 Time s Time s Design, Build & Test guided model Some comments on fragment screening Hubbard & Murray (2011), Meth Enzymology, 493, 509 • For “good” target sites (many enzymes): • If assays configured correctly • Same hits identified by ligand observed NMR and SPR • validated hits tend to give crystal structures • Careful QC of fragment library – attention to assay conditions • There can be lots of false negatives from screening by X‐ray • Requires suitable crystal system • “Wet” assays can work sometimes • But high concentrations can confound the assay • Thermal melt methods unreliable • For non‐conventional targets (such as protein‐protein): • Many issues • Overbinding, problems due to properties of compounds and target • Cross‐validate binding by different techniques

27 Leads generated for many Targets

• Disclosed targets include: • Kinases: CDK2, Chk1, PDPK1, PDHK1, Pim1, STK33, Pak4 • ATPases: DNA gyrase, Grp78, HSP70 and HSP90 • protein ‐protein interaction targets: Pin1 and Bcl‐2 • FAAH and tankyrase • Undisclosed targets include: • kinases with unusual binding sites • protein ‐protein interaction targets • novel classes of enzymes in large multi‐domain complexes

28 Summary –fragments and conventional targets

• Fragment screen to assess targets • Fragments alongside HTS and knowledge‐based • You always find something new • Fragments to hits for conventional targets is relatively straightforward • (when crystal structures / good models available) • The main issues are organisational and cultural • Discuss !!

Inflated Expectation expectations Slope of Plateau of Enlightenment productivity

Technology Trough of Time 29 Trigger Disillusionment Fragments 2013 talk

30 James Murray, Steve Roughley, Natalia Matassova Exploiting the dissociation rate constant

-1 koff (s ) KD = -1 -1 kon (M s )

koff PL P + L kon

31 James Murray, Steve Roughley, Natalia Matassova Exploiting the dissociation rate constant

-1 koff (s ) KD = -1 -1 kon (M s ) • Cannot improve association rate constant above ~10‐8 M‐1s‐1 • No point in drug discovery, too many membranes in the way! • Dissociation rate constant can be infinite (ie covalent!) • We have seen that it is the key driver of potency • As have others (review Copeland, Future Med. Chem. 3(12), 2011)

32 James Murray, Steve Roughley, Natalia Matassova Exploiting the dissociation rate constant

n Dissociation - k o off k

- Resonance n o Kinetics i Signal (RU) t a i c o s s

A Concentration Time (s) • Surface plasmon resonance –a way to measure kinetics

33 • FOCUS ON THE Off‐RATE ‐ koff

James Murray, Steve Roughley, Natalia Matassova Exploiting the dissociation rate constant

-1 koff (s ) KD = -1 -1 kon (M s ) • Cannot improve association rate constant above ~10‐8 M‐1s‐1 • No point in drug discovery, too many membranes in the way • Dissociation rate constant can be infinite (ie covalent) • We have seen that it is the key driver of potency • As have others (review Copeland, Future Med. Chem. 3(12), 2011) • Independent of concentration • Exploit this to assess unpurified reactions, off‐rate screening (ORS)

34 James Murray, Steve Roughley, Natalia Matassova Off rate screening (ORS): Example

• Historical Hsp90: thienopyrimidines

• 200 nM–5 MIC50 by FP assay

• Re‐prepared compounds by Suzuki reaction

R R Cl B(OH)2

O DMF / H2O N O N NH N S O NaHCO 2 3 S O 2NH N Pd(Ph3P)2Cl2 100 ºC Microwave 10 min • Minimal work‐up • Evaporate, partition • Purity 50 –80 % (LCMS) • Screened by ORS 35 James Murray, Steve Roughley, Natalia Matassova Off rate screening (ORS): Example

• Historical Hsp90: thienopyrimidines

• 200 nM–5 MIC50 by FP assay

• Re‐prepared compounds by Suzuki reaction

R R Cl B(OH)2

O DMF / H2O N O N NH N S O NaHCO 2 3 S O 2NH N Pd(Ph3P)2Cl2 100 ºC Microwave A: Pure starting material 10 min B: Faux reaction • Minimal work‐up • Evaporate, partition • Purity 50 –80 % (LCMS) • Screened by ORS 36 James Murray, Steve Roughley, Natalia Matassova Off rate screening (ORS): Example

• Historical Hsp90: thienopyrimidines

• 200 nM–5 MIC50 by FP assay

• Re‐prepared compounds by Suzuki reaction

R R Cl B(OH)2

O DMF / H2O N O N NH N S O NaHCO 2 3 S O 2NH N Pd(Ph3P)2Cl2 100 ºC Microwave A: Pure starting material 10 min B: Faux reaction C: Purified product • Minimal work‐up D: Crude reaction • Evaporate, partition • Purity 50 –80 % (LCMS) • Screened by ORS 37 Tankyrase ‐ Introduction

• Axin is targeted for turnover through poly‐D‐ribose labelling by Tankyrase. This removes axin, which has a role in stabilising ‐catenin • Inhibition of Tankyrase has been proposed to enhance the degradation of ‐catenin, an indirect way of affecting the WNT‐pathway • Crystal structure of tankyrase available in 2007

2RF5 3UH4 Structural Genomics Novartis (2012) Consortium (2007)

38 Alba Macias, Chris Graham and team Tankyrase ‐ Introduction

• Axin is targeted for turnover through poly‐D‐ribose labelling by Tankyrase. This removes axin, which has a role in stabilising ‐catenin • Inhibition of Tankyrase is therefore proposed to enhance the degradation of ‐catenin, an indirect way of affecting the WNT‐pathway • Crystal structure of tankyrase available in 2007

• At Vernalis: • Initially –low levels of protein production –insufficient material for fragment screen by NMR • Able to produce large numbers of apo‐crystals that preliminary trials showed were suitable for ligand soaking 39 Alba Macias, Chris Graham and team Tankyrase ‐ Hit identification

Crystals

Soaked fragments in pools of 8

Data collection (up to 1.6Å resolution)

Streamlined structure solution

62 hits from 1563 fragments

Characterised by SPR and TSA

40 Alba Macias, Chris Graham and team Tankyrase ‐ Fragment to hit evolution

• The most attractive fragments were not suitable for rapid chemistry • Designed a modified fragment • Off‐rate screening libraries identified vectors and substituents • Crystals soaked directly with reaction mixtures • Lead Series driven using combinations of tools • Computational chemistry • X‐ray crystallography • SPR (ORS), DSF • Medicinal Chemistry • Properties • 5 nM vs TNKS2, high ligand efficiency (0.60) • Affects PD markers in cells (stabilises Axin2, inhibits WNT pathway) • Tools to probe the biology

41 PDHK –an unusal kinase

• GHKL family protein (like HSP90) • Other companies have targetted for diabetes • AZ and Novartis in the 1990s –various allosteric sites • Vernalis investigated as a cancer metabolism target • Off ‐rate screening to selectively evolve a fragment

DCA –2q8h

Pfizer –2bu7

E2 / L2 site AZD‐7545 –2q8g Novartis – 2bu5

42 ATP site –2q8i PDHK project

• Initial fragment hit for ATP site –also binds to HSP90 • Structure shows which vector(s) to explore Fragment bound to PDHK1 • Parallel libraries synthesised and screened by SPR • Using the off‐rate screening method –changes in k off VER‐00236029 • One SPR channel with HSP90, one with PDHK

• Can identify PDHK selective compounds PDHK1 • And HSP90 selective compounds HSP90

VER‐00236030

PDHK1 HSP90

43 Fragments 2013 talk

44 Fragments 2013 talk

45 Non‐conventional targets

46 Non‐conventional targets

• Can find fragments that bind • Orthogonal biophysical methods can validate and characterise fragment binding

47 Non‐conventional targets

• Can find fragments that bind • Evolution requires robust model of fragment binding • Best model is from X‐ray structure • But sometimes high affinity ligand required for structure

48 Non‐conventional targets

Unfiltered NOESY - All NOEs

49 Non‐conventional targets

• Can find fragments that bind • Evolution requires robust model of fragment binding • Best model is from X‐ray structure • NMR methods can provide sufficient quality of model • Experiments can be filtered to reveal just the interactions between protein and ligand Protein/ligand

Filtered NOESY Unfiltered - Intermolecular NOESY NOEs - All NOEs

50 Non‐conventional targets

51 Video of Bcl‐2 plasticity

52 Bcl‐2 ‐ ligandability

8% Low hit rates (< 2%) 7% High hit rates (> 2%) 6%

5% rate

hit protein-protein interaction 4% (0.4-3% hit rate) 3% Validated Class 1 hits rates 2% 1% 0% 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 DscoreLigandability1 Calculatecalculated druggability from structure (DScore) • changes dramatically as ligands explore available pockets / flexibility

53 Geneste, Murray, Davidson, leDiguarher et al Selective Bcl‐2 inhibitor program • Structure‐guided optimisation has generated selective Bcl‐2 inhibitors • MW < 780; > 100‐fold selective over other BH3 domains • Sub‐10 nM efficacy in cell models of AML • In vivo , rapid and strong apoptotic response in RS4;11 xenograft models, both iv and oral • Platelet sparing (cf ABT‐263) • Key was biophysical assays to assess cell penetration and compound aggregation 1400

1200

1000 ) 3

800

Compound 3 50 mg/kg 600 Compound 3 100 mg/kg Compound 4 50 mg/kg

Tumour volumes (mm Tumour volumes Compound 4 100 mg/kg 400 Median +/- InterQuartile Range ABT-263 50 mg/kg ABT-263 100 mg/kg

200 Control (untreated)

Treatment schedule per( os) 0 (Twice a week) 54 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 Time after tumour inoculation (days) Summary ‐ non‐conventional targets

• For non‐conventional targets • And where structures are hard to obtain • It takes time – and not yet clear how often fragments can be successful • Integration with biophysical methods can be key • Also –a commitment to the long haul

Inflated Expectation expectations Slope of Plateau of Enlightenment productivity

Technology Trough of Time 55 Trigger Disillusionment Fragments 2013 talk

56 Updated from Chen & Hubbard (2009), JCAMD, 23, 603 Finding fragments ‐ issues

• For some targets, it can take time to configure a robust assay • For fragment screening but also hit progression • Protein construct, assay conditions, etc, etc

8% Low hit rates (< 2%) 7% High hit rates (> 2%) 6%

rate 5% hit protein-protein interaction 4% Kinases targets – varying hit rates 3% high hit rate Validated Class 1 hits rates 2% Poor targets 1%

0% 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 57 LigandabilityCalculatecalculatedDscore druggability from structure1 (DScore)

3D fragments

• Most of the compounds in fragment libraries are commercially available small molecules • Medicinal chemist emphasis on chemical tractability • Some use privileged fragments from existing drugs • Most of the fragments are flat heterocycles • This is fine for some targets (kinases, ATPases) • Perhaps limiting for other (new) target classes • Some initiatives underway to introduce more 3D fragments • The challenge will be synthetic tractability • A York initiative

58 York 3D fragments

• Peter O’Brien has developed chemistry for adding lithiated N‐Boc heterocycles 2 (formed from 1) to heterocyclic, symmetrical ketones • The resulting fragments have distinctive 3D shapes, presenting useful looking pharmacophores • Shape measured by principle moments

rod Vernalis sphere rod York fragments sphere 1 1 0.95 0.95 0.9 0.9 0.85 0.85 0.8 0.8 0.75 0.75 0.7 0.7 0.65 0.65 0.6 0.6 0.55 0.55 0.5 0.5 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 disc 0 0.1 0.2 0.3 0.4disc 0.5 0.6 0.7 0.8 0.9 1

59 York 3D fragments

• Peter O’Brien has developed chemistry for adding lithiated N‐Boc heterocycles 2 (formed from 1) to heterocyclic, symmetrical ketones • The resulting fragments and lead‐like compounds have distinctive 3D shapes, presenting useful looking pharmacophores • Project underway to explore the chemistry • Generate 500 member library • See how it performs against various targets

60 Concluding remarks

• Straight forward to find fragments for most sites on most proteins • Opportunities for new “3D” fragments? • The challenge is knowing what to do with the fragments • Off ‐rate screening allows exploration of vectors • Evolving fragments in absence of structure? • For conventional targets • Lots of starting points; opportunity for “good” medicinal chemistry • Issue in some organisations is integration with medicinal chemistry • For non‐conventional targets • Provides starting points when other techniques fail • Close integration with biophysics is crucial; takes time and commitment • Not necessarily faster –patience required

61 • But hopefully better

End

• References in the slides acknowledge who did the work

• FBLD conference • 2008 –San Diego • 2009 –York • 2010 – Philadelphia • 2012 –Sa n Francisco • 2014 –Basl e • http://www.fbldconference.org

62 Vernalis Research Overview • Approximately 60 staff in research • Based in Cambridge, UK (Granta Park) • Recognised for innovation and delivery in structure and fragment ‐based drug discovery • Structure‐based drug discovery since 1997 • Distinctive expertise combining X‐ray, NMR, ITC and SPR to enable drug discovery against established and novel, challenging targets • Portfolio of discovery projects • Six development candidates generated in the past six years • Protein structure, fragments and modelling integrated with medicinal chemistry • Internal Vernalis projects in oncology • Collaborations across all therapeutic areas • e.g. oncology, neurodegeneration, anti‐infectives 63 • Aim to establish additional collaborations during 2013 / 14