Genomics-driven clustering of disease-related biomarkers identifies therapeutic options in myelodysplastic syndromes (MDS)

Personalized treatment of MDS

Presented at AACR 2018 Annual Meeting

  • Session Category: Bioinformatics and Systems Biology
  • Session Title: Statistical Methods, Mathematical Modeling, and Molecular Modeling
  • Session Date and Time: Tuesday Apr 17, 2018 1:00 PM - 5:00 PM
  • Location: McCormick Place South, Exhibit Hall A, Poster Section 13
  • Poster Board Number: 26
  • Permanent Abstract Number: 4285

Author Block Leylah M. Drusbosky1 , Kimberly E. Hawkins1 , Amy Meacham1 , Elizabeth Wise1 , Neeraj Kumar Singh2 , Chandan Kumar2 , Sumanth M. Vasista2 , Rakhi P. Suseela2 , Taher Abbasi3 , Shireen Vali3 , Kaoru Tohyama4 , Maher Albitar5 , Peter P. Sayeski1 , Christopher R. Cogle1 . 1University of Florida, Gainesville, FL; 2Cellworks Research India Pvt. Ltd, Bangalore, India; 3Cellworks Group Inc, San Jose, CA; 4Kawasaki Medical School, Kurashiki, Japan; 5Neogenomics, Aliso Viejo, CA

ABSTRACT: Hypomethylating agents (HMA) and lenalidomide (LEN) are approved and used in the treatment of patients (pts) with MDS, though these drugs fail in most pts. No method exists to predict drug response beyond associating single actionable mutations with a single drug’s response. We hypothesized that MDS pts can be clustered by similarities in genomic/molecular profiles, & that each cluster may be assigned combos of FDA-approved drugs to target their unique biomarker profile. Bone marrow cells from 88 MDS pts & the MDS-L cell line were analyzed by cytogenetics & for mutations in 14 myeloid genes using NGS. 31 pts had sufficient data for analysis. 20 profiles had similar aberrations & were grouped. Genomic data from pts and MDS-L were entered into a computational biology modeling (CBM) software, which generates a disease-specific protein network map using PubMed to create digital models and identify characteristic biomarkers unique to each pt. An algorithm was created to cluster the models based on overlapping disease-specific biomarkers. Digital drug simulations (DDS) were conducted both on MDS-L & pt simulation models by quantitatively measuring drug effect on a cell growth score (CGS), a composite of cell proliferation, viability & apoptosis. DDS identified drugs by assessing their impact on disease-specific biomarkers and calculated CGSs. Predictions were validated using MTT. 14/31 MDS pt profiles, including the MDS-L cell line, clustered into 4 groups based on biomarker characteristics.MDS-L cells harbor NRAS (G12A) mutation,-7,-12,+1,+8,+19,+20 and +21. Genes associated with increased copy number (CN) include AURKA, IGFR, PAR5, MTOR, IL6, JAK3, MDM4, MYC, MCL1, COX2, PDE4A, and RCE1; genes associated with decreased CN include DUSP1, RASA1, NR3C1, IRF1, ETV6, and SHH. CBM identified active biomarkers in MDS-L cells (90RSK, MAPK7, AKT and BTK), validated by western blot. DDS predicted nelfinavir+celecoxib to be effective in MDS-L. MDS-L cells were treated with nelfinavir, celecoxib, and nelfinavir+celecoxib with increasing doses. Nelfinavir & celecoxib reduced MDS-L viability in a dose-dependent manner, while nelfinavir+celecoxib showed additive reduction of MDS-L viability. DDS was performed on each pt to predict response to HMA and LEN. Two of the clusters (n=2/cluster) were predicted to be non-responders to any SOC therapy. A third cluster (n=3) was predicted to respond to HMA, but not LEN, and the fourth cluster (n=6) showed varying or no response to either HMA/LEN. These results use a novel concept of using genomics & CBM to cluster profiles with overlapping disease-relevant biomarkers & similar drug response predictions. CBM can identify pt populations who may benefit from certain therapeutic regimens, improving response rates, & give insights into the mechanisms by which each drug impacts the MDS-specific biomarkers


ResearchMatthew Matassa