(This post relates to work done at Weill Cornell University, where OneThree’s core technology was developed)
Proper function of microtubules is essential to any cell, especially cancer cells which rely on high growth rates. Because of this, one of the most-used classes of cancer therapeutics are drugs that inhibit the function of microtubule (MT) spindles. However, one of the largest causes of mortality in patients treated with MT-inhibitors is drug resistance, and, because of this, there is a need to discover new compounds that can be used in these resistant patients.
Working with the lab of Evi Giannakakou at Weill Cornell we set out to use our platform to see if we could identify any new MT-inhibitors that had not tested before. Our platform integrated data from multiple different sources (including genomics, structures, cell-based screens, bioassays, etc.) to computationally screen millions of orphan molecules (molecules without any known targets) to pinpoint any that were predicted to target microtubule spindles. From our top 40 predictions we found that 24 were easily obtainable from providers and the NCI DTP.
Each of these 24 compounds were tested against MDA-MB-231 breast cancer cells and we assessed any effect on the microtubule cytoskeleton by confocal microscopy and immunofluorescence. Additionally we quantified the level of microtubule engagement by each of the predicted small molecules by measuring the amount of soluble tubulin left over after treatment. We found that 67% of the tested small molecules (16/24) depolymerized microtubules as predicted.
This was incredibly exciting for us because these 16 molecules could represent new potential therapeutics for cancer patients and were discovered in under a month using an AI based approach. However the real challenge was whether any of the new candidates we identified could be viable option for patients who become resistant to currently used treatments (approximately 50% of patients). To answer this we first worked with the Giannakakou lab to develop a 1A9 ovarian cancer cell line that was resistant to Eribulin and other approved microtubule inhibitors such as colchicine and vinorelbine (we’ll call this resistant cell line 1A9-ERB). We then took the 4 candidates that had the strongest levels of microtubule inhibition in our earlier experiments (compounds #2, 15, 16, & 24) and tested each of them on this resistant cell line.
Based on the results of the immunoflourescence assay we found that 3 of the 4 predicted compounds were able to inhibit proper microtubule function in the 1A9-ERB cell line while none of the approved drugs could. To further evaluate these candidates, we took candidate #15 (our top performing candidate) and tested it and approved MT inhibitors in a cytotoxicity assay to see how well it could kill 1A9-ERB cancer cells. We found that our predicted compound had over a 1000x lower fold resistance than Eribulin and over 10x lower fold resistance than any other approved MT-inhibitors (measured by IC50 ratios). We’re planning on testing this candidate in animal studies going onward, but this case study highlighted how a computational approach that integrates multiple different forms of data could quickly and accurately identify new candidates for a given indication or patient population.
This is an ongoing collaboration so I’ll edit this post as time goes on with new results and analyses! Also check out our post on predicting synergistic combination therapies to see how we used AI to combat MT-inhibitors by identifying a new drug we could pair with docetaxol to break resistance.
Till next time!