All the mutations that are fit to die

Abstract

In this issue of Cell Chemical Biology, Chakraborty et al. ¹ employ a deep mutational screening analysis of 3,500 single point mutations in every residue in Src kinase’s catalytic domain to determine which residues are critical for conferring ATP-competitive inhibitor resistance. They identify a dynamically controlled resistance site.

Protein kinases are key drug targets with more than 70 drugs approved by the Food and Drug Administration (FDA). Most of these drugs are employed in cancer therapy where activation of kinase signaling pathways often promote the disease. The poster child and extreme example of kinase activity causing disease is the formation of the constitutively active BCR-Abl fusion protein in chronic myelogenic leukemia (CML). Inhibition of BCR-Abl kinase by an ATP-competitive inhibitor, imatinib, reduced mortality from CML by almost 80%. Unfortunately, kinase inhibition in the context of cancer poses an extreme selection pressure for cancer cells and various mechanisms of resistance to kinase inhibitors have been described ². In the mechanistically most straightforward way, this includes mutations in the target kinase that weaken the affinity for the inhibitor. Since all but one FDA-approved small molecule kinase inhibitors are ATP competitive, mutations can also cause inhibitor resistance by strengthening the affinity for ATP. Mutations in the kinase may promote kinase signaling in the presence of an inhibitor by changing the activation state of the kinase, altering the substrate peptide preference, or affecting the pharmacodynamics of drug binding and dissociation. In addition, mutations can thermodynamically stabilize or destabilize the kinase, which can alter the interactions of the kinase with cellular factors such as chaperones ³. Apart from mutations that result in an altered protein sequence, drug resistance can arise from many other mechanisms such as overexpression of the target kinase or other kinases, downregulation of phosphatases that would provide checks and balances to kinase signaling under normal circumstances ⁴.

Currently, the process of identifying resistance mutations in the clinic is slow and relies on patients experiencing failure of potentially lifesaving therapies. In a study reported in this issue of Cell Chemical Biology ¹, the Maly group at the University of Washington utilized a saturation mutagenesis approach to overcome this limitation and to gain a comprehensive view on possible resistance mutations. They screened all possible single amino acid substitutions in the catalytic domain of Src kinase for resistance towards several inhibitors ^1,5. Src kinase was the first identified viral proto-oncogene and the first described tyrosine kinase. The kinase domain, also named Src homology 1 (SH1) domain, catalyzes the transfer of the γ-phosphate from ATP to the hydroxyl group of tyrosine residues in substrate peptides. The enzymatic activity of Src is tightly regulated by two autoinhibitory domains, the Src homology 2 (SH2) and 3 (SH3) domains, which bind to a phosphotyrosine in the carboxy-terminal tail and a proline motif of Src, respectively.

Previously, the authors had studied the effects of mutations on the regulation of Src activity in the yeast S. cerevisiae, where Src activity causes toxicity ⁵. This represents an important difference from mammalian cancer cells where Src activity can be a transforming factor and promote viability. Here, they use this yeast assay to study the effects of mutation on Src resistance towards kinase inhibitors. They generate a barcoded plasmid library of 3500 mutations, representing 70% of all possible mutations in Src kinase domain while leaving the regulatory domains outside the kinase domain unchanged. Transformed yeast cells were harvested at different time points and the DNA bar codes sequenced to determine the identity and frequency of mutant and wild type (wt) Src DNAs (Figure 1). The logarithm of the frequency ratio of mutant and wt DNA was regressed over culture time, and the negative slope was used to determine an activity score. Active Src mutants resulted in more cell death/less yeast growth, and a lower frequency of mutant plasmid DNA relative to wt, which in turn results in a shallower growth curve for yeast cells harboring Src activity and conversely a higher activity score.

Figure 1.

Overview of the workflow

This screen was then repeated in the presence of the kinase inhibitor dasatinib and four ATP-competitive and conformation-selective inhibitors. Surprisingly, the authors found a cluster of mutations in the amino-terminal region of Src kinase domain that retained activity in the presence of dasatinib. While some sites were consistent with known resistance mutations in other tyrosine kinases, overall, the high density of resistance mutations in this cluster was surprising, especially since many of the resistance-prone amino acid side chains face away from dasatinib. Many mutations in this cluster also increased kinase activity by displacing the autoinhibitory SH3/SH2 domains. This was particularly puzzling since the SH3/SH2 domains do not directly bind to the identified cluster (Figure 2). Instead, as shown by the authors, a highly resistant and activating mutation (W285T) led to an overall increase in the conformational dynamics of the kinase domain and resulted in the release of the SH3/SH2 domains.

Figure 2.

Regulation of Src kinase domain activity by autoinhibitory SH3/SH2 domains. Top: In wild-type Src kinase, the SH3 and SH2 domains bind to the enzymatic SH1 domain and downregulate kinase activity. In particular, binding of the phosphorylated C-terminal tail to the SH2 domain shifts the equilibrium towards the assembled and inactive state of Src, as indicated by the bold arrow. Bottom: Resistance mutations in the β1/ β2 cluster (green) increase the conformational dynamics of the phosphate binding P-loop (red), which leads to disassembly of the SH3/SH2 domain and shifts the equilibrium towards the open and active state of the kinase as indicated by the bold arrow.

The study provides several important advances. First of all, the authors identify most of the expected mutations in direct contact with the inhibitors. However, the authors carefully point out the limitations of their approach for describing clinically relevant drug resistance mutations. Human cancer cells and yeast cells differ in many ways that may affect resistance of Src to inhibitors: (i) Src kinase activity is downregulated in mammalian cells by C-terminal Src kinase (Csk) and Csk is absent in yeast cells, (ii) expression levels of Src kinase vary among different mammalian cells, mutations can affect the expression level relative to wt further and it is hard to tell what the adequate expression level would be in yeast cells, (iii) the presence of cell wall in yeast that can hinder the uptake of different kinase inhibitors. Additionally, cancer cells and yeast cells respond differently to drug resistance: while Src activity is toxic to yeast cells, cancer cells can be addicted to tyrosine kinase activity, which has been utilized by previous screens for drug resistance mutations ⁶. The real strength of the deep mutational screening, when paired with inhibitor binding, is the unbiased and comprehensive probing of protein architecture and function. It allows studying mutations that are rarely observed in the clinic because they may require multiple nucleotide substitutions. The yeast proliferation assay and the deep sequencing of barcodes are practical and provide a vast amount of data. Importantly, the use of different conformation-selective inhibitors allows probing the impact of each mutation on different conformational states, which adds additional dimensions to this already very data rich project.

As demonstrated so elegantly here, the approach led to the identification of the β1/ β2 cluster that controls the kinase activity by regulating the binding of the regulatory domains to the kinase domain through dampening its conformational dynamics. Mutations in this cluster can increase the conformational dynamics of the P-loop, which in turn leads to the displacement of the regulatory domains, kinase activation and resistance to drug binding.

Similar approaches on other kinases or ever larger combinations of saturation mutagenesis will improve our understanding of protein architecture, fuel AI-aided protein design and help predict resistance mutations before they arise in the clinic.