Predicting Clinical Success of ADCs using a Mechanistic Modeling & Simulation Approach

Transcription

1 Predicting Clinical Success of ADCs using a Mechanistic Modeling & Simulation Approach Alison Betts Translational Modeling & Simulation Biomedicine Design Department Pfizer Worldwide R&D, Cambridge MA 02139, U.S. alison.betts@pfizer.com

2 Introduction ADCs: an exciting class of therapeutics for the treatment of cancer, combining the selective properties of mabs with the potent cell killing activities of cytotoxic agents ~59 ADCs being tested in clinical trials across multiple cancer types, by a variety of Pharma/ Biotech companies 332 unique clinical trials for ADCs (Beacon database) Unlikely that such a large population of ADCs could make it to the market, and as such differentiation will occur in clinical trials (an expensive place for drug failure). ADCs in Clinical Development 2016 Question posed: Could we use quantitative M&S tools to analyze data on these ADCs to predict likelihood of success or failure in the clinic?

3 Subset of ADCs chosen to study ADC name Adcetris (Brentuximab vedotin; SGN-35) Kadcyla (Trastuzumab-DM1) Inotuzumab ozogamicin (CMC-544) Lifastuzumab vedotin (RG-7599) Linker- Payload MOA Target Indication Company Clinical Status vc-mmae MTI CD30 HL, salcl Seattle Genetics Approved: 1.8mpk q3w smcc-dm1 MTI HER-2 MBC Genentech Approved: 3.6mpk q3w Acbut-calich DNA CD22 ALL Pfizer Approved for ALL: 1.8mg/m 2 (0.05mpk) Qw. Discontinued for NHL. vc-mmae MTI NaPi2b NSCLC, Ovarian Pinatuzumab vedotin vc-mmae MTI CD22 NHL Lorvotuzumab maytansine (IMGN-901) Coltuximab ravtansine (SAR-3419) Vorsetuzumab mafadotin (SGN-75) Genentech Genentech Ph 1: NSLC: PR:12% + 3pts (11%) with unconfirmed PR. Ovarian: PR: 50% Ph2: Platinum resistant Ovarian Ph2: Combined With Rituximab or Polatuzumab Vedotin (DLBCL, FL) SPP-DM1 MTI CD56 SCLC Immunogen Discontinued: SCLC: Endpoints not met in PhII. SPDB-DM4 MTI CD19 NHL (DLBCL) Sanofi Discontinued (strategic reasons): mc-mmaf MTI CD70 RCC, NHL Seattle Genetics 44% ORR Discontinued (Ph1b): RCC: 2mpk PR: 2/9

4 Reasons why ADCs do not succeed in the clinic: Part 1 ADCs can fail if they do not reach their clinical efficacious dose before you get toxicity (TI). Clinical MTD is fairly well understood based on payload classes and cynomolgus monkey tox. Can we do a better job of predicting efficacious dose?

5 ADC Translational Strategy Assumptions: 1. Assume ADC plasma concentrations are driving response. 2. Assume mouse PD parameters translate directly to clinic. Outcomes: 1. Predicted efficacious dose in the clinic 2. Predict optimal regimen 3. Predict ORR Can we use this strategy to predict efficacious dose of ADCs (as a metric for clinical success)?

6 Kadcyla example: preclinical PK/PD 3 cell lines: BT474, Fo5 and N87: T-DM1 vs. Fo5 T-DM1 vs. BT474EEI PK/PD model: k k Drug kill function kk C max Non - linear k C C50 MM Jumbe, Tibbitts et al. JPKPD 2010 Haddish-Berhane, Shah and Betts. JPKPD 2013 Mouse PK/PD parameters estimated : Tumor growth: Exponential, linear and maximum tumor growth ADC effect: maximum kill rate, concentration at halfmaximum kill rate, transduction rate Tumor static concentration (TSC): BT474: 11µg/ml, Fo5: 41µg/ml & N87: 29µg/ml

7 Kadcyla example: clinical translation Simulated Tumor Volume (mm^3) T-DM1 clinical PK taken from Krop et al, T-DM1 PK/PD Modeling Mouse TSC (CV% ) µg/ml Predicted T-DM1 Stasis Dose (mg/kg Q3weeks) Cell Line (Her2) BT474EEI (Jumbe et al, 2010) 11 (30) 2.4 Fo5 (Jumbe et al, 2010) 41 (30) 4.8 N87 (In-house data) 29 [ ] 4.6 Actual Clinical Data 14* 3.6 *C average at 3.6mg/kg dose in clinic (Krop et al, 2010). PK/PD Modeling Approach from 3 cell lines Successfully Predicted Clinical Dose of T-DM1 Predicted T-DM1 clinical dose = mg/kg Q3 weeks T-DM1 Phase 2 dose= 3.6mg/kg Q3 weeks T-DM1 Simulated Tumor Volume (from BT474EEI cell line) Time (Days) Stasis at ~2.4mg/kg Q3weeks 0.3mg/kg 0.6mg/kg 1.2mg/kg 2.4mg/kg 3.6mg/kg 4.8mg/kg Dose 1 Dose 2 Dose 3 7

8 Predicted vs. observed efficacious doses Highlighted in RED ADCs which did not meet their clinical endpoints. For Ino (NHL) & Lorvatuzumab predicted efficacious doses were NOT reached in clinical trials.

9 Reasons why ADCs do not succeed in the clinic: Part 2 Why is Vorsetuzumab vedotin (SGN-75) not efficacious in the clinic? Need to dive deeper into the mechanism of action of ADCs:

10 A complex mechanism: Many factors contributing to success of ADCs Binding affinities: ADC to membrane target (pm to nm) PL to intracellular target Pharmacokinetics: Clinical half-life 1-10 days SGN-75=10 days Soluble target HER-2, CD56: yes CD22, CD70: no Receptor expression HER2: 1x10 6 CD70: 1x x10 3 Internalization rate mins to hours CD22: 5mins CD70: 1.5-2hours Recycling rate CD56 recycles CD22 denovo (re)synthesis Payload exocytosis: Bystander effect MMAE yes Calicheamicin yes Cys-mc-MMAF no Payload release Chemical: MMAE Acid hydrolysis: calicheamicin Catabolism of mab: mc-mmaf Can integrate these parameters into a mechanistic modeling framework

11 SGN-75 Mechanistic Model Structure PK X2 ADC V2 ADC Bolus Dose ADC CLD ADC CD70 Receptor Expression: 2K- 420K/ cell (RCC) 1K-120K/cell (NHL) X1 ADC V1 ADC Surface Exchange ADC Vascular Exchange Tumor kon ADC koff ADC kint Ag ADC Free ADC Bound Receptor Internalization: 1.5-2hrs kon PL Drug to Antibody Ratio: 4 X2 PL V2 PL CL ADC CLD PL DAR k dis X1 PL V1 PL Vascular Exchange Payload Surface Exchange PL Free k dis kout PL kin PL PL Free koff PL Cell PL Bound ADC Kd: 1nM & PL Kd: 18nM CL PL PD TV kg 1 V1 Ex V Max 1 ψ ψ kg 1 Ex TV kg L kkill PL Max Tumor kc PL 50 Tumor Tau V1 V2 V3 Tau V4 Tau Cell Death No Bystander effect Shah, Haddish-Berhane and Betts, ADC PK and PD (tumor growth inhibition)

12 Translation to the clinic Mouse Model Measured and modeled: ADC Mouse PK mcmmaf PK: internal data Tumor growth (doubling time of 12 days) and inhibition: 786-O mouse xenograft PD data CD70 expression in 786-O: 190K (RCC cell line) Initial tumor volume: ~100mm 3 Human Model for RCC Measured: ADC Human PK (Tannir 2014) mcmmaf PK: internal data Literature: CD70 expression: 190K (2K-420K) Initial tumor volume: cm 3 Tumor growth: day -1 Doubling time of 90 days Predicted/ simulated: Tumor growth inhibition: using drug effect parameters/ variability from mouse model. A dose of 1.5mpk was predicted for tumor regression: predicted efficacious dose consistent with previous modeling

13 Why does SGN-75 not work at its predicted efficacious dose? A parameter that may be designed during discovery Antigen Heterogeneity In the clinical trial, patients had range of cells expressing CD70 Literature sources give a range of 2, ,000 CD70/cell Binding affinity Antigen Expression Maximal Killing Tumor doubling time Uncertain due to limited mouse TGI doses; correlated with cell line growth rate Uncertain due to lack of data, particularly no intracellular data Payload degradation Variable across patients; observed clinical growth is much lower /variable than xenograft cell line used in mouse studies

14 Heterogeneity of expression coupled with lack of bystander effect drives lack of efficacy of SGN-75 Results of sensitivity analysis: Parameter Unit Nominal Value Study Range Sensitivity Antigen expression heterogeneity % cells expressing Ag 100% 75% 100% Highest Maximal killing (k max ) 1/day /30 10x Tumor doubling time (k g0 ) days Intracellular payload half-life hours Antigen Expression Level CD70/cell 190K 1K 500K Binding affinity (K D ) nm Lowest Heterogeneity of CD70 positive and negative cells in tumors has a dramatic impact on the predicted success of treatment. Model predicts only PD events even with a very small percentage of [CD70-] cells (5% of tumor cells). This observation strongly suggests MMAF, which lacks a by-stander effect, is not an ideal payload under these conditions. Tumor doubling time is also a very sensitive parameter. However, cannot explain on it s own why SGN-75 does not work. The model predicted a very low sensitivity to the antigen expression level, even over a range of 1K 500K CD70/cell sufficient payload was delivered to cell over this entire range.

15 Conclusions Mathematical modeling and simulation can help integrate complex data to predict likelihood of success of ADCs in the clinic. Knowing your predicted efficacious dose up front helps stack your odds in favor of success. Translational strategy presented helps predict clinical efficacious dose. However, getting the dose right is not the only key to success. For SGN-75, the dose appeared correct, but heterogeneity of expression coupled with lack of bystander effect appears to be driving lack of efficacy. Patient population selected in the clinical trial is also a key factor

16 Acknowledgements Nahor Haddish-Berhane Hugh Barton Keith MacCannell Lindsay King Frank Barletta Mauricio Leal Hannah Jones Subu Chakrapani Chris O Donnell Tony Wood Puja Sapra Bradley Niesner (former RES, Needham, MA) Paul Jasper (RES, Needham MA)

17 Back-ups

18 Clinical Trial Simulations Synthetic human cohorts (n=500) were generated using parameter variability from the preclinical PK/PD modeling. These were used for Clinical Trial Simulations to calculate ORR at the predicted efficacious dose. Clinical ORR results are predicted over a range of cycle numbers and compared with observed clinical trial data. SGN-75 (CD70) SAR-3419 (CD19) RG-7599 (NaPi2b) Pinatuzumamab vedotin (CD22) Clinic (MTD) 3 mpk 4.3 mpk 2.8 mpk 2.4 mpk MTD ORR Model Prediction (Efficacious Dose) Clinic Model Dose Range Cycle Range % PR+CR ~ 3 mpk ~ 2.6 mpk ~ 4.4 mpk ~ 2 mpk 2 3 mpk mpk mpk mpk 1 16 (median 2) 1 6 (median 4) 1 28 (median 4) 1 24 (median 5) 7% 33-43% 50% 39 50% Dose 3 mpk 2.6 mpk 4.4 mpk 2 mpk Cycle # % PR+CR 0% 0% 9% 27% 30.6% 37% 38% 16% 25% 59% 78% 10% 32% 48% 60% ORR= PR ( 30% tumor regression) + CR (100% BioMedicine tumor Design regression) All PK/PD predictions in agreement with clinically observed MTD and ORR

19 Predicted vs. observed efficacious doses ADC Target LP Predicted Efficacious Dose^ (mg/kg) Clinical MTD (RP2D) (mg/kg) Regimen Kadcyla (T-DM1) HER-2 smcc-dm Q3W Adcetris (Brentuximab vedotin; SGN-35) CD30 vc-mmae Q3W Inotuzumab (CMC-544- NHL) CD22 acbut-calich Q4W Inotuzumab (CMC-544- ALL) CD22 acbut-calich Q4W Lorvatuzumab maytansine (IMGN-901 NSCLC) CD56 SPP-DM Q3W SAR-3419 CD19 SPDB-DM Q3W Vorsetuzumab mafodotin (SGN-75) CD70 mc-mmaf Q3W Pinatuzumab vedotin CD22 vc-mmae Q3W Liftastuzumab (RG-7599) Napi2b vc-mmae 4.4 >2.8 # Q3W Highlighted in RED ADCs which did not meet their clinical endpoints. For Ino (NHL) and IMGN-901 predicted efficacious doses were NOT reached in clinical trials. ^Range reflects a range of cell lines/pdx *Use of human growth parameters # MTD not reached