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)

Introduction: 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). Here, we report the associations between clinical responses and global DNA demethylation assessed by LINE1 assay, baseline expression of a panel of 7 genes (CDA, P15, P21, DNMT3B, DNMT3A, DNMT1 and CTCF) assessed by qRT-PCR, and FLT3-ITD/NMP1 mutations.

Methods: Samples from 122 patients with r/r AML were analyzed (27 from phase 1 study treated at 36 m/m²/d or higher, 47 from the dailyx5 and 48 from the dailyx10 regimen of the phase 2). Global DNA methylation at baseline and after treatment, were estimated using bisulphite-pyrosequencing on the LINE-1 repetitive sequence for assessing guadecitabine pharmacodynamic (PD) effects. LINE-1 methylation time-course profiles were available in 117 out of 122 analyzed patients. Multivariate analysis was performed using nonlinear ensemble decision tree-based classification algorithms (Random Forests and Gradient Boosting Trees); the algorithms were applied to rank demographic features, treatment schedule features, baseline expression of 7 genes, FLT3-ITD and NPM1 mutations, LINE-1 methylation profiles and baseline hematological features as predictors of clinical response and to measure their directionality. All reported percentages are a measure of the average relative contribution of each feature in predicting response across boosted trees and are normalized across features.

Results: 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). High baseline CDA gene expression in peripheral blood was consistently ranked the main predictor of clinical response (mean for responders~1.76 and non-responders~0.12) irrespective of patient gender, dose and dosing cycles, and contributed for 22% of the model accuracy in predicting responses to guadecitabine. Expression of the genes CTCF (11%), DNMT3B (7%) and P21 (6%) were other genes that contributed > 5% in predicting clinical response. Age (8%) was also a significant predictor of response. FLT3-ITD mutations and NPM1 mutations were not significant predictors of response. In an integrated analysis, where methylation levels and gene expression were ranked together on their power to predict clinical response, LINE-1 methylation percent change from baseline on days 8(6%) and 22(19%), age (16%), CTCF (9%), CDA(7%), P21(6.5%) gene expression at baseline had greater than 5% contribution in predicting clinical response. A similar analysis was then performed to rank features on their power to predict day 8 methylation; CDA (28%), DNMT1 (8.5%), P15 (7.5%), and CTCF (7%) gene expression at baseline and age (7%) had greater than 5% predictive power.

Conclusions: In conclusion, in r/r AML patients, global DNA demethylation was strongly associated with clinical responses to guadecitabine and analysis of baseline gene expression can identify trends that might enrich for r/r AML patients more likely to respond to HMA therapy with guadecitabine.