60% of myelodysplastic syndromes (MDS) patients fail to achieve clinical improvement with hypomethylating agents (HMAs). 26% of patients with non-del5q MDS respond to lenalidomide (LEN), but the mechanism of response is unknown. Treatment of non-del5q patients who fail HMA is challenging and predicting response to LEN would address this unmet clinical need.

Methods: A retrospective analysis was performed on non-del5q MDS patient cytogenetics, gene mutations, treatments, and clinical response data. Using Cellworks Group software, genomic data was used to generate disease-specific protein network maps. Digital drug simulations were done by quantitatively measuring drug effect and calculating a MDS inhibition score. Each patient-specific map was digitally screened for the extent by which LEN and HMAs inhibited MDS in a dose-dependent manner and compared to the patient's clinical outcome.

Results: Patient 1 is a non-del5q patient with trisomy 8 (+8) and an ASXL1 frameshift mutation. Simulation modeling revealed overexpression (OE) of MYC leading to IRF4 and IKZF expression, key pathways for LEN sensitivity. MYC OE also leads to CCND1 expression, aiding proliferation of MDS cells. Simulated LEN treatment targets MYC-induced upregulation of IKZF1, STUB1, CRBN, and CSNK1A1 to reduce MDS cell survival. Upregulated TP53 inhibits MYC expression, augmenting the patient's response to LEN. To compare response to other SOC drugs, azacitidine (AZA), was screened. The ASXL1 mutation led to loss of PRC2-mediated DNMT1 function. +8 results in OE of GGH, reducing methylation of CpG regions. Given loss of DNMT1 function, patient 1 was predicted be a non-responder to AZA. Patient 2 is a non-del5q patient with +8, add(14q)(11.1-11.2), add(8)(p23.3-q24.3), and an ASXL1 mutation. Modeling revealed the same dysregulated pathways as patient 1, predicting sensitivity to LEN and resistance to AZA. Patient 3 is a non-del5q patient with complex karyotype and mutations in IDH2, TET2, and SRSF2. Modeling revealed OE of CARD11 yielding sensitivity to LEN. OE of CDH1 and AXIN1 decreased beta-catenin activity, a known resistance mechanism to LEN. OE of apoptotic TP53, BAX, and BBC3 sensitize this patient's MDS cells to LEN. Modeling revealed reduced EZH2 mRNA levels due to the SRSF2 mutation, directly impacting the PRC2 complex needed for DNMT1-mediated CpG methylation. EZH2 is further down regulated by L3MBTL2 OE. TET2 is functionally challenged from the lack of CpG methylation, diminishing IDH activity, resulting in AZA resistance.

Conclusions: Computer modeling of non-del5q MDS patient biology can deduce the abnormal protein networks and consequent drug effects. This method could be used to understand drug failure and highlight resistance pathways that can be targeted to recover chemosensitivity. This technology can also be used to predict MDS patient response prior to treatment, increasing drug effectiveness and reducing side effects and treatment costs.

Citation Format: Leylah Drusbosky, Neeraj Kumar Singh, Shireen Vali, Taher Abbasi, Christopher R. Cogle. A computational biology method to predict HMA or lenalidomide treatment response in non-Del(5q) MDS [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2017; 2017 Apr 1-5; Washington, DC. Philadelphia (PA): AACR; Cancer Res 2017;77(13 Suppl):Abstract nr 4545. doi:10.1158/1538-7445.AM2017-4545