Abstract Tubulin binding drugs are approved for the treatment of many cancer types, without the use of molecular markers to select patients likely to respond. Plinabulin binds β-tubulin in a differentiated pocket and is being tested in a Phase 3 clinical study for the treatment of NSCLC. Additional indications are being considered for plinabulin and an algorithm for selecting especially responsive cancers and patient subgroups would be of significant value. With this in mind, Affymetrix HG-U133 Plus 2.0 array mRNA expression data for 43 human breast, lung, prostate, ovarian or CNS cancer cell lines were utilized to develop mathematical models to predict in vitro plinabulin potency against the same cell lines. Cells were treated for only 24 hours with plinabulin and then cultured for another 48 hours without plinabulin. Viable cell number was then measured with a Cell Titer-Blue Assay, and the plinabulin concentration causing a 70% reduction in viable cells (IC70) versus vehicle treated controls was derived. Cell lines were clearly separable into plinabulin Active (21 cell lines with IC70<1.0 μM) and Inactive (IC70>9.5 μM) groups. Log2 transformed Affymetrix gene probeset signal values, preprocessed with the GeneChip robust multi-array average analysis algorithm, were selected and ranked as predictors of plinabulin activity with a bootstrap forest partitioning technique, utilizing JMP 14.1 statistical software. 56 HIT probesets were identified that also had significantly different expression in responding versus non-responding cell lines (p<0.01, uncorrected t-test). For probesets with gene annotation, only the probeset for each gene with the highest Jetset score was utilized (Li et al., 2011). Top HIT predictor genes include CTNNB1 (β-catenin; oncogene), CALD1 (caldesmon; inhibits myosin ATPase activity), ERI1 (RNA processing), LGR5 (adult stem cell biomarker), SECISBP2L (SLAN; prolongs mitosis), and TRAK1 (mitochondrial/endosome trafficking). Models were constructed from HIT gene probesets in JMP to identify plinabulin responding cell lines, utilizing either one-layer TanH multimode fit neural networks or binary logistic regression. Surprisingly, models incorporating approximately 4-10 probeset values were derived that perfectly predicted plinabulin activity. Importantly, the cell lines tested in the above analyses were not those known to express high levels of multi-drug resistant (MDR) transporters. The importance of MDR status was therefore evaluated separately. Plinabulin activity, unlike that of taxanes, was not significantly affected by the MDR transporter inhibitor verapamil (10 μM), in ovarian cancer cell lines with a known MDR phenotype. To conclude, our work provides novel algorithms that may be of value in selecting cancer patients with tumor cells that are particularly susceptible to the direct cytotoxic effects of plinabulin. Citation Format: James R. Tonra, Hagen Klett, Chenghao Shen, Gerhard Kelter, Ramon W. Mohanlal, G. Kenneth Lloyd, Lan Huang. Predictive models for tumor cell targeting with plinabulin, derived from in vitro screening and Affymetrix mRNA expression data [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2019; 2019 Mar 29-Apr 3; Atlanta, GA. Philadelphia (PA): AACR; Cancer Res 2019;79(13 Suppl):Abstract nr 1254.
Read full abstract