Preclinical Validation Studies Support Causal Machine Learning Based Identification of Novel Drug Targets for High-Risk Multiple Myeloma

Introduction

Regardless of significant advances in the therapy of multiple myeloma (MM) there is still a lack of effective treatment options for patients with high-risk disease. In this context, we recently developed a network of high-risk disease based on more than 30 000 genomic and clinical variables from 645 patients of the CoMMpass dataset (Gruber et al., ASH 2016). Validation of these findings has been performed in the IFM/DFCI 2009 trial dataset (Furchtgott et al., ASH 2017). This comprehensive computational approach revealed a network of 17 genes driving high-risk (defined as progression or myeloma-related death within 18 months). Here, we performed preclinical validation of potential novel drug targets to confirm the utility of in silico guided target discovery in high-risk MM.

Methods

TTK (CFI402257, BAY-1217389), PLK4 (CFI400945, Centrinone), MELK (OTSSP167) and CDK1 (CPG71514) inhibitors were studied in a panel of human MM cell lines (n=11) for their activity in cell viability, cell growth, cell cycle, apoptosis, colony formation, drug combination and co-culture experiments. PKMYT1, TTK and PLK4 were targeted with doxycycline-inducible shRNAs. Analysis of gene expression (GEP) data (GSE24080) was used to link candidate genes to certain MM subgroups.

Results

The network of 17 genes driving high-risk disease contained eight kinases that serve as attractive drug targets (AURKA, NEK2, CDK1, BUB1B, MELK, TTK, PKMYT1, and PLK4), all of them involved in cell cycle regulation. Accordingly, expression levels of all kinases (except PKMYT1) were enriched in the GEP-defined proliferation associated subgroup of MM and thus linked to poor outcome. To study the interconnectedness of the individual network genes we first investigated the impact of previously reported CDK1 and MELK inhibitors on other network members. This demonstrated rapid loss of CDK1, NEK2, MELK, PKMYT1 and FOXM1 protein levels.

We then selected TTK and PLK4 as putative novel MM targets with available inhibitors undergoing clinical testing in solid tumors. Protein and mRNA expression of both genes was confirmed in all MM cell lines. Two selective compounds per gene were used for preclinical studies. All four inhibitors significantly reduced MM cell viability and single dose IC70 treatment impaired cell growth up to 10 days (60-98% reduction, P<0.01). This growth inhibitory effect was confirmed with inducible shRNAs. Mechanistically, growth impairment was linked to G2M cell cycle arrest followed by the accumulation of polyploid cells (15-90% of cells 72h post treatment) which is in line with the role of both genes in chromosome segregation. The formation of aberrant mitoses led to the induction of apoptosis 3-5 days post treatment (≤20% viable cells at day 5 post treatment with 3/4 inhibitors) and was accompanied by the presence of active caspase 3 and cleaved PARP. Importantly, the activity of these drugs persisted in the presence of BMSCs and showed potent activity in colony formation assays (DMSO: 168±15 and 131±57, BAY-1217389: 39±29 and 41±3, CFI-400945: 19±26 and 30±6 colonies in KMS12BM and OPM2 cells at day 14, P<0.05). Drug combination studies pointed to favorable activity in combination with dexamethasone and lenalidomide. Furthermore, confirmatory TTK knockdown with two independent shRNAs sensitized MM cells to dexamethasone.

Finally, PKMYT1 was chosen as putative target based on its role as major driver of high-risk in our model. We transduced MM cells with three doxycycline-inducible PKMYT1-targeting shRNAs and observed an impressive impact on myeloma cell growth upon doxycycline induction compared to non-targeting control shRNA (up to 85% reduction at day 10, P<0.01). Furthermore, PKMYT1 knockdown led to the induction of apoptosis in all MM cell lines tested. Based on these encouraging results we currently perform in-depth in silico and in vitro analyses of the underlying PKMYT1 signaling network. Detailed results of these sub-studies will be presented at the meeting.

Conclusions

Our results confirm the utility of computational based modelling of high-risk disease. This strategy not only revealed a genetic network closely associated to adverse prognosis, but also enabled the identification of so far unnoticed drug targets. Importantly, inhibitors of TTK and PLK4 are already in clinical testing and thus enable rapid clinical translation of our findings to MM patients in need of alternative therapeutic options.