While there are now over 30 FDA-approved selective kinase inhibitors available for the treatment of cancer, the median progression-free survival is still <1 year for a majority of these drugs. Drug resistance is responsible for >90% of deaths in patients with metastatic cancer. In many of these cases, mutations in the target of therapy drive resistance by abolishing or reducing inhibitor affinity while maintaining or increasing kinase activity.
Developing a quantitative and predictive understanding for how these inhibitors elicit resistance through mutational mechanisms would be of immense value, allowing the rapid design of second-generation inhibitors to circumvent resistance, the exploration of therapeutic combinations to minimize the emergence of resistance, and potentially the development of new therapeutics unlikely to elicit resistance. To predict resistance using physical modeling techniques, we hypothesize that resistance mutations ablate inhibitor binding affinity while maintaining some physically computable surrogate for activity, such as ATP and substrate peptide binding affinities. This allows us to compute a resistance index for each possible point mutation, and to construct enhanced sampling schemes that selectively search for point mutations that maximize the resistance index. We are currently building a framework to perform these calculations using our GPU-accelerated alchemical binding free energy code YANK, using kinases ABL and SRC as model systems. Predicted resistance mutations will be compared with clinically-identified mutations in the cBioPortal, and biophysically characterized by engineering these mutations into kinase catalytic domains that can be expressed recombinantly in bacteria. We have developed an automated fluorescence assay to perform affinity measurements against a variety of small molecule kinase inhibitors of clinical interest, and use activity assays to determine whether our hypothesis regarding physical surrogates for activity is valid for individual kinases.
FUNCTIONAL RAMIFICATIONS OF MUTATIONS OBSERVED IN TUMOR MUTATIONS
We are working with the Hsieh laboratory at MSKCC to understand the molecular mechanisms underlying clinically-identified activating mutations in mTOR, the target of several FDA approved inhibitors for kidney cancer. mTOR is responsible for regulating cell growth and metabolism, integrating signaling from a number of other genes that are commonly mutated in many other cancer types as well. While mTOR itself is most commonly mutated in kidney cancer, it is mutated in 2% of all other cancer types. In kidney cancer, where mTOR inhibitors are the standard of care, patients with some mTOR mutations respond well to therapy, while others do not. If we were able to understand the molecular mechanism underlying mTOR activating mutations and which mutations sensitize the kinase to mTOR inhibitors, mTOR inhibitor treatment could potentially be useful for millions of cancer patients across all cancer types who harbor relevant mTOR mutations.
As a first step to mapping the functional ramifications of clinically-identified mTOR mutations, we have developed an automated pipeline to model mutations onto mTOR structures, prepare them for simulation on Folding@home, and perform consistent analyses across all mutant and wild-type kinases. We have collected over 2.2 milliseconds of simulation data in aggregate from over 225,000 simulations on Folding@home donor machines, and are in the process of analyzing this data further. Initial analysis suggested immediate structural consequences underlie the activating behavior of certain mutations, as we describe in a recent Journal of Clinical Investigation paper
PREDICTING DRUG SENSITIVITY AND RESISTANCE IN PATIENT TUMORS
The decreasing cost of next-generation sequencing has spurred the deep sequencing of patient tumors with the goal of improving patient outcomes by individually tailoring treatment plans to patient-specific genetic alterations---potentially found only in small subpopulations of tumor cells. While this approach will undoubtedly revolutionize cancer therapy, there is a great deal of work to be done over the next decade to connect tumor alterations to optimal therapeutic plans. Evaluating whether missense mutations in therapeutic targets represent resistance, neutral, or sensitizing mutations could significantly shape the use of targeted kinase inhibitors for cancer therapy. Though a number of frequently-occuring mutations have been noted (and, less frequently, characterized biochemically and in cell lines), such as the T315I gatekeeper mutation in Abl, which causes resistance to first-generation targeted inhibitors, the long tail of infrequently observed mutations makes it impossible to build a catalog of mutations presented in more than a minor fraction of cancers.
We adopt a two-pronged approach to predicting the therapeutic resistance or sensitivity of patient-derived tumor mutations:
First, we utilize a computational model built to predict the emergence of mutational resistance to identify potential resistance mutations (mutations that ablate inhibitor affinity while maintaining some surrogate for activity), and test these by engineering these mutations into bacterially-expressing kinases in our laboratory. Success of this computational model in identifying mutations that cause resistance or susceptibility could eventually aid the development of individually-tailored treatment plans that maximize therapeutic success, avoiding drugs for which intrinsic resistance (even at the subclonal level) already exists, or employing combinations of therapeutics capable of overcoming inherent resistance.
Second, we are developing an automated high-speed cell-free expression pipeline that would allow mutant kinases to be engineered, expressed, and assayed against all available clinical kinase inhibitors within a day. New cost-effective cell-free expression technologies based on S12 extracts and fed by glucose allow for inexpensive and rapid scalable production of many proteins. Aided by an FGI Rapid Response grant, our laboratory is currently exploring the automated optimization of these techniques (with GFP fluorescence or mass spectrometry as a readout) using machine-learning algorithms based on Gaussian processes to improve yields for kinase domains represented in the IMPACT assay to allow the production of kinases containing engineered patient mutations in a matter of hours. Kinase production would be immediately followed by the fluorescence assay of kinase inhibitor affinity against all FDA-approved kinase inhibitors to aid the determination of whether inhibitors clinicians would like to use would be effective in light of tumor-specific mutations. We anticipate this would reduce the entire engineering and assay pipeline to turn around sensitivity/resistance results within a single day---fast enough to have substantial impact on therapeutic decisions.
Markus Seeliger (Stony Brook University): Kinase biochemistry and structural biology
Nicholas Levinson (University of Minnesota): Kinase engineering, structural biology, and spectroscopy
James Hsieh (MSKCC): Clinical mutations in MTOR in kidney cancer patients
Mechanistically distinct cancer-associated mTOR activation clusters predict sensitivity to rapamycin
Xu Jianing, Pham CG, Albanese SK, Dong Yiyu, Oyama T, Lee CH, Rodrik-Outmezguine V, Yao Z, Han S, Chen D, Parton DL, Chodera JD, Rosen N, Cheng EH, and Hsieh J.
Journal of Clinical Investigation, 126:3526, 2016. [DOI] [PDF]
An open library of human kinase domain constructs for automated bacterial expression
Daniel L. Parton, Sonya M. Hanson, Lucelenie Rodríguez-Laureano, Steven K. Albanese, Scott Gradia, Chris Jeans, Markus Seeliger, and John D. Chodera.
Manuscript prior to publication: [bioRxiv] / Interactive data browser: [github.io] / Plasmids available via AddGene