Poster presented at: ASH Annual Meeting, Orlando, FL December 5-8, 2015 © Astex Pharmaceuticals, Inc.

Multivariate Analysis Identifies Significant Correlations Between Baseline Biomarkers, DNA Demethylation and Clinical Responses in 122 Patients treated

in a Phase 1/2, Study of Guadecitabine (SGI-110), in the Treatment of Relapsed/Refractory Acute Myeloid Leukemia (r/r AML)

13 9 4

Abst.

No. 2594 Srikanth Gottipati1, Shashank Rohatagi1, Woonbok Chung2, Pietro Taverna6, Hagop Kantarjian3, Gail J. Roboz4, Patricia Kropf6, Mohammad Azab6 and Jean Pierre Issa2

Otsuka Pharmaceutical Development & Commercialization, Inc. Princeton, NJ, Fels Institute, Temple University, Philadelphia, PA2, The Ohio State University, Columbus, OH9, University of Texas, MD Anderson Cancer Center, Houston, TX3, , Weill Cornell Medical College and ,The New York Presbyterian Hospital, Division of Hematology and

Oncology, New York, NY4, Fox Chase Cancer Center, Philadelphia, PA5, Astex Pharmaceuticals Inc., Pleasanton, CA6

In conclusion, in r/r AML patients, guadecitabine-induced global DNA demethylation was strongly associated with clinical response and our model based on baseline gene expression (high CTCF & CDA and low DNMT3B) in patients’ peripheral blood and clinical/demographic characteristics (high age) can predict response to guadecitabine with 70% accuracy.

Gene expression (∆ CT)

Response CDA P15 P21 DNMT3B DNMT3A CTCF DNMT1

Responders 1.76(1.6) -4.4(2.1) 0.3(1.2) -2.7(1.7) 0.79(0.8) 1.2(0.5) 1.2(0.7)

Non-responders 0.12(2.5) -5.4(2.9) -0.5(1.6) -1.7(1.7) 0.67(1.45) 0.84(0.97) 0.95(1.1)

1. Issa JP et al, Safety and tolerability of guadecitabine (SGI-110) in patients with myelodysplastic syndrome and acute myeloid leukemia: a multicenter, randomized, dose-escalation phase 1 study, Lancet Oncol. 2015 Sep;16(9):1099-110.

2. Kantarjian H et al , First clinical results of a randomized phase 2 study of SGI-110, a novel subcutaneous (SQ) hypomethylating agent (HMA), in adult patients with acute myeloid leukemia (AML), abst 497, ASH 2013.

3. Griffiths E et al, First results of a Phase 2 study using a 10-day subcutaneous (SC) regimen of the novel hypomethylating agent (HMA) SGI-110 for the treatment of relapsed/refractory Acute Myeloid Leukemia (r/r AML), abst 3074, ASCO 2014.

Guadecitabine (SGI-110) is a novel subcutaneous (SC) next generation hypomethylating agent (HMA) designed as a dinucleotide of decitabine (DAC) and deoxyguanosine that is resistant to degradation by cytidine deaminase (CDA) and results in prolonged in vivo exposure to its active moiety DAC. The differentiated pharmacokinetic profile offers the potential of improved biological and clinical activity and safety over currently available HMAs.

In the Phase 1 study, patients with r/r AML were treated at escalating doses of guadecitabine. In the Phase 2 study, r/r AML patients were randomized to receive guadecitabine at 60 mg/m2 or 90 mg/m2 SC daily for 5 days (dailyx5). In a separate cohort, patients were treated with 60 mg/m2 SC daily for 10 days (dailyx10: days 1-5 and 8-12) for up to 4 cycles followed by subsequent cycles of the dailyx5 regimen. All regimens were dosed with a 28 day treatment cycle.

We have reported the clinical efficacy and safety results from the Phase 1 dose-escalation study in AML and MDS (Issa et al, Lancet Oncology 2015) and the Phase 2 randomized dose-response study in r/r AML at 2 doses (60 and 90 mg/m2) in a 5-day regimen (Kantarjian et al, ASH 2013) and 60 mg/m2 in the 10-day regimen (Griffiths et al, ASCO 2014).

The relative influence of predictors of clinical responses with the goal to improve the accuracy of response classification. Baseline expression of a panel of 7 genes, treatment schedule, baseline hematological profiles, cytogenetic risk, FLT3-ITD/NMP1 mutations, demographics and global DNA demethylation assessed by LINE1 assay were used as predictors of response.

Samples from 122 patients with r/r AML were analyzed : o 27 from Phase 1 study were treated at a therapeutic dose range of guadecitabine

(36 mg/m2/d – 125 mg/m2/d) o 47 patients from Phase 2 were treated with 60 mg/m2/d (n=22) or 90 mg/m2/d (n=25)

dailyx5 o 48 patients received a dose of 60 mg/m2/d dailyx10.

Global DNA methylation at baseline and during cycle 1 were estimated using bisulphite-

pyrosequencing on the LINE-1 repetitive sequence for assessing guadecitabine pharmacodynamic (PD) effects. o LINE-1 methylation time-course profiles were available in 117 out of 122 analyzed patients. o Blood samples were drawn during screening, and on days 1, 8, 15, 22 in cycle 1 of the

treatment and on the first day of cycle 2 (day 29).

Gene expression was examined on a panel of genes (CDA, P15, P21, DNMT3B, DNMT3A, DNMT1, and CTCF) at baseline by quantitative qRT-PCR.

Multivariate analysis was performed using nonlinear ensemble decision tree-based classification algorithm Gradient Boosting Tree (GBT); o the algorithm was applied to rank demographic features, treatment schedule features,

baseline expression of 7 genes, FLT3-ITD and NPM1 mutations, LINE-1 methylation profiles, cytogenetic risk levels and baseline hematological features as predictors of clinical response and to measure their directionality.

o A grid based evaluation was conducted to compute optimal parameters for GBT. For each parameter set, samples are randomly held-out and a model is fit on the remaining data. The samples held out are then predicted using the estimated model. This is repeated several times to compute the average performance across the samples. Optimal parameters are chosen based upon balanced accuracy of response classification. Average classification performance measures are reported in Figure 2.

o All reported percentages (relative influence) are a measure of the average relative contribution of each predictor in predicting response across boosted trees and are normalized across predictors.

LINE-1 demethylation (%)

Response Baseline LINE-1

methylation Cycle 1 Day 8 Cycle 1 Day 15 Cycle 1 Day 22 Cycle 2 Day 1

Max LINE-1 % Demethylation

Responders 69(3.5) -28.4(10.1) -27(12.2) -16.55(11.5) -8.2(8.1) -32.21(10.11)

Non-responders 70(4.5) -21.2(12.4) -18.5(13.4) -11.4(7.7) -6.5(6.3) -21.35(13.33)

Clinical characteristics

Response Age Male (%) Platelet counts Screening PB blast

(%) Screening BM

Blast (%) Screening WBC (k/uL)

Responders 64(13) 10.2 88(94) 7.01(10.5) 25.5(17.6) 3.5 ( 3.2)

Non-responders 58(14) 51.3 58(58) 28(32) 42(28) 7.3 (14.7)

Figure 1. The above plot summarizes the clinical data. Composite complete response (CRc) category includes CR (complete remission), CRi (CR with incomplete platelet recovery) and CRp (CR with incomplete hematologic recovery) while non-responders (NR) includes PR. Gene expression = z score for each gene. BM= % bone marrow blasts at screening . PB= % peripheral blasts at screening . Platelet: Platelet Count at Screening (x109/L). WBC: WBC at Screening (K/µl). Mutations: FLT-3-ITD/NPM1 status. Demethylation = maximum LINE-1 demethylation as percentage change from baseline during cycle 1. In the 122 patients analyzed, the median age was 59.6 (range, 23-86), 75 were males (61.5%). Overall, peak LINE-1 demethylation generally occurred on day 8 after daily x 5 treatment or on day 8 or 15 after daily x 10 treatment. Overall, the maximum peak LINE-1 demethylation was -23.8 % ± 1.24. In 122 r/r AML patients, 28 showed Complete Response (23%, 14 CR and 14 CRi/CRp).

Figure 2. Relative influence(%) of each predictor to predict response/no-response as shown in subfigures (a), (b) and (d) or to predict LINE-1 demethylation profiles as shown in (c).

(a) (b)

(c) (d)

Baseline predictors of response Baseline & LINE-1 demethylation as predictors of

response

Baseline predictors of LINE-1 demethylation Baseline and maximum LINE-1 demethylation as

predictors of response

Table 1. Mean and their standard deviations (in parenthesis) of expression of 7 gene panel, global LINE-1 demethylation and clinical characteristics in responders (n=28) vs. non-responders (n=90).

Baseline expression of genes associated with DNA methylation and HMA mechanism of action were consistently ranked as main predictor of clinical response irrespective of patient gender, dose and dosing schedule (Figure 2a). High expression of CTCF and CDA were highest predictors of response (> 10% predictive

power) Age (11%) was also a significant predictor of response.

Methylation levels did not improve response classification accuracy when integrated with baseline gene expression (Figure 2b). High LINE-1 methylation percent change from baseline on days 8(14.5%) and 22(15%) were

good predictors response maximum LINE-1 demethylation does not improve classification accuracy when ranked

along with gene expression (Figure 2d), but remains a significant predictor while keeping the order of influence of baseline predictors the same.

Low DNMT3B expression along with high CTCF and CDA expression are significant predictors when integrated with methylation predictors

Time from start of treatment cycle, dosing schedule and expression of CDA(high) at baseline were good predictors of LINE-1 demethylation overall (Figure 2c).

FLT3-ITD & NPM1 mutations and cytogenetic risk were not significant predictors of response.

Srikanth Gottipati and Shashank Rohatagi are employees of Otuska Pharmaceutical Development & Commercialization. Woonbok Chung has Research Funding; Pietro Taverna and Mohammad Azab are employees of Astex Pharmaceuticals., Inc.; Hagop Kantarjian has research funding from Novartis, BMS, Pfizer, and Amgen; Gail Roboz has consultancy: Agios, Amphivena, Astellas, Astex, Astra Zeneca, Boehringer Ingelheim, Celator, Celgene, Genoptix, GlaxoSmithKline, Janssen, MEI Pharma, Roche, Seattle Genetics, Shire, Spectrum, Sunesis, & Teva Karyopharm, Incyte, TG Therapeutics, Astex Pharmaceuticals, Inc.