Concurrent inhibition of oncogenic and wild-type RAS-GTP for cancer therapy.

RAS oncogenes (collectively NRAS, HRAS and especially KRAS) are among the most frequently mutated genes in cancer, with common driver mutations occurring at codons 12, 13 and 611. Small molecule inhibitors of the KRAS(G12C) oncoprotein have demonstrated clinical efficacy in patients with multiple cancer types and have led to regulatory approvals for the treatment of non-small cell lung cancer2,3. Nevertheless, KRASG12C mutations account for only around 15% of KRAS-mutated cancers4,5, and there are no approved KRAS inhibitors for the majority of patients with tumours containing other common KRAS mutations. Here we describe RMC-7977, a reversible, tri-complex RAS inhibitor with broad-spectrum activity for the active state of both mutant and wild-type KRAS, NRAS and HRAS variants (a RAS(ON) multi-selective inhibitor). Preclinically, RMC-7977 demonstrated potent activity against RAS-addicted tumours carrying various RAS genotypes, particularly against cancer models with KRAS codon 12 mutations (KRASG12X). Treatment with RMC-7977 led to tumour regression and was well tolerated in diverse RAS-addicted preclinical cancer models. Additionally, RMC-7977 inhibited the growth of KRASG12C cancer models that are resistant to KRAS(G12C) inhibitors owing to restoration of RAS pathway signalling. Thus, RAS(ON) multi-selective inhibitors can target multiple oncogenic and wild-type RAS isoforms and have the potential to treat a wide range of RAS-addicted cancers with high unmet clinical need. A related RAS(ON) multi-selective inhibitor, RMC-6236, is currently under clinical evaluation in patients with KRAS-mutant solid tumours (ClinicalTrials.gov identifier: NCT05379985).

Solving the RNA design problem with reinforcement learning.

We use reinforcement learning to train an agent for computational RNA design: given a target secondary structure, design a sequence that folds to that structure in silico. Our agent uses a novel graph convolutional architecture allowing a single model to be applied to arbitrary target structures of any length. After training it on randomly generated targets, we test it on the Eterna100 benchmark and find it outperforms all previous algorithms. Analysis of its solutions shows it has successfully learned some advanced strategies identified by players of the game Eterna, allowing it to solve some very difficult structures. On the other hand, it has failed to learn other strategies, possibly because they were not required for the targets in the training set. This suggests the possibility that future improvements to the training protocol may yield further gains in performance.

Retrosynthetic Reaction Prediction Using Neural Sequence-to-Sequence Models.

We describe a fully data driven model that learns to perform a retrosynthetic reaction prediction task, which is treated as a sequence-to-sequence mapping problem. The end-to-end trained model has an encoder-decoder architecture that consists of two recurrent neural networks, which has previously shown great success in solving other sequence-to-sequence prediction tasks such as machine translation. The model is trained on 50,000 experimental reaction examples from the United States patent literature, which span 10 broad reaction types that are commonly used by medicinal chemists. We find that our model performs comparably with a rule-based expert system baseline model, and also overcomes certain limitations associated with rule-based expert systems and with any machine learning approach that contains a rule-based expert system component. Our model provides an important first step toward solving the challenging problem of computational retrosynthetic analysis.

Atomistic structural ensemble refinement reveals non-native structure stabilizes a sub-millisecond folding intermediate of CheY.

The dynamics of globular proteins can be described in terms of transitions between a folded native state and less-populated intermediates, or excited states, which can play critical roles in both protein folding and function. Excited states are by definition transient species, and therefore are difficult to characterize using current experimental techniques. Here, we report an atomistic model of the excited state ensemble of a stabilized mutant of an extensively studied flavodoxin fold protein CheY. We employed a hybrid simulation and experimental approach in which an aggregate 42 milliseconds of all-atom molecular dynamics were used as an informative prior for the structure of the excited state ensemble. This prior was then refined against small-angle X-ray scattering (SAXS) data employing an established method (EROS). The most striking feature of the resulting excited state ensemble was an unstructured N-terminus stabilized by non-native contacts in a conformation that is topologically simpler than the native state. Using these results, we then predict incisive single molecule FRET experiments as a means of model validation. This study demonstrates the paradigm of uniting simulation and experiment in a statistical model to study the structure of protein excited states and rationally design validating experiments.

Molybdenum catalyzed ammonia borane dehydrogenation: oxidation state specific mechanisms.

Though numerous catalysts for the dehydrogenation of ammonia borane (AB) are known, those that release >2 equiv of H2 are uncommon. Herein, we report the synthesis of Mo complexes supported by a para-terphenyl diphosphine ligand, 1, displaying metal-arene interactions. Both a Mo(0) N2 complex, 5, and a Mo(II) bis(acetonitrile) complex, 4, exhibit high levels of AB dehydrogenation, releasing over 2.0 equiv of H2. The reaction rate, extent of dehydrogenation, and reaction mechanism vary as a function of the precatalyst oxidation state. Several Mo hydrides (Mo(II)(H)2, [Mo(II)(H)](+), and [Mo(IV)(H)3](+)) relevant to AB chemistry were characterized.

Quantitative live imaging of cancer and normal cells treated with Kinesin-5 inhibitors indicates significant differences in phenotypic responses and cell fate.

Kinesin-5 inhibitors (K5I) are promising antimitotic cancer drug candidates. They cause prolonged mitotic arrest and death of cancer cells, but their full range of phenotypic effects in different cell types has been unclear. Using time-lapse microscopy of cancer and normal cell lines, we find that a novel K5I causes several different cancer and noncancer cell types to undergo prolonged arrest in monopolar mitosis. Subsequent events, however, differed greatly between cell types. Normal diploid cells mostly slipped from mitosis and arrested in tetraploid G(1), with little cell death. Several cancer cell lines died either during mitotic arrest or following slippage. Contrary to prevailing views, mitotic slippage was not required for death, and the duration of mitotic arrest correlated poorly with the probability of death in most cell lines. We also assayed drug reversibility and long-term responses after transient drug exposure in MCF7 breast cancer cells. Although many cells divided after drug washout during mitosis, this treatment resulted in lower survival compared with washout after spontaneous slippage likely due to chromosome segregation errors in the cells that divided. Our analysis shows that K5Is cause cancer-selective cell killing, provides important kinetic information for understanding clinical responses, and elucidates mechanisms of drug sensitivity versus resistance at the level of phenotype.