Deletion Mutations Keep Kinase Inhibitors in the Loop

Abstract

Effective clinical application of conformationally selective kinase inhibitors requires tailoring drug choice to the tumor's activating mutation(s). In this issue of Cancer Cell, Foster et al. (2016) describe how activating deletions in BRAF, EGFR, and HER2 cause primary resistance to common inhibitors, suggesting strategies for improved inhibitor selection.

Studies of oncogenic kinases in the era of tumor DNA sequencing illustrate the power of modern translational research. Our current understanding of EGF receptor tyrosine kinase activation emerged only from studies that integrated crystal structures with activating mutations observed in patient lung cancer tissues (Jura et al., 2011). Similarly, the clinical effects of BRAF inhibitors have revealed unexpected and important lessons about the complexity of signaling at this level of the MAP kinase pathway (Lito et al., 2013). Two avenues most likely to illuminate paths to the development of improved therapeutics are thorough mechanistic assessment of oncogenic mutations (MacArthur et al., 2014) and understanding of relationships between kinase activation mode and conformational selectivity of inhibitors (Wang et al., 2014). A paper published in this issue of Cancer Cell by Foster and colleagues combines these two considerations for poorly understood mutations in three commonly targeted kinases: EGFR, HER2/ErbB2, and BRAF (Foster et al., 2016). Their results, together with those of a related study from Chen et al. (2016), illustrate the importance—and feasibility—of tailoring the conformational selectivity of kinase inhibitors to the mutation(s) observed in tumors.

Foster et al. focused on short deletions in the loop between the third β strand and the crucial αC helix in the kinase domains of EGFR and HER2 (both receptor tyrosine kinases) and BRAF (a serine/threonine kinase). These β3/αC deletions are most frequent in EGFR, occurring in exon 19 and accounting for up to ~44% of EGFR alterations in lung adenocarcinomas (Foster et al., 2016), but are much rarer in HER2 and BRAF. For HER2, β3/αC loop deletions are found in ~2% of breast tumors with any ERBB2 aberration (Foster et al., 2016). For BRAF, β3/αC loop deletions are seen in 0.6%–1% of pancreatic and thyroid tumors (Chen et al., 2016; Foster et al., 2016).

The proximity of these deletions to the crucial αC helix of each kinase suggests an important role in activation, and each example observed in a tumor was indeed found to activate its respective kinase through cellular studies (Chen et al., 2016; Foster et al., 2016). As shown in Figure 1, helix αC plays a central role in the conformational transition from an “inactive” to an “active” kinase domain, moving from an “out” position to an “in” position that enables it to contribute (via a key glutamate side chain) to setting up the catalytic site. Deletions in the loop preceding αC would be expected to influence this transition, restraining helix αC and potentially biasing the conformational equilibrium toward the active state. In the EGFR family (which includes HER2), αC is held in the “out” position prior to activation by interactions with a short α helix in the activation loop (Figure 1), as first described for Src family and cyclindependent kinases (CDKs) (Jura et al., 2011). In normal kinase activation, these autoinhibitory interactions are disrupted by one of several influences, including asymmetric kinase domain dimerization (in EGFR family kinases), association of allosteric regulators such as cyclins (in CDKs), and loss of other inhibitory restraints (in Src family kinases). Activation of these and other kinases by oncogenic mutations can also frequently be traced to direct destabilization of these same autoinhibitory interactions. Importantly, BRAF employs a very similar mode of regulation. A short α helix in the BRAF activation loop associates with (and displaces) helix αC in the inactive state to autoinhibit the kinase, and the resulting autoinhibition is ordinarily reversed by symmetric kinase domain dimerization and activation loop phosphorylation promoted by Ras activation (Thevakumaran et al., 2015).

Figure 1. Activating Deletion Mutations Change Kinase Inhibitor Selectivity.

Cartoon of kinase domain regulation in the EGFR family or RAF. In the wild-type kinase, helix αC is normally held in the “out” position by interactions with a short helix in the activation loop (A-loop). Upon activation, the A-loop is reorganized, freeing αC to adopt the “in” position and contribute to the active site through a salt bridge between a key glutamate (red) in αC and lysine (blue) in strand β3. β3/αC deletions force αC into the “in” position regardless of other influences, promoting A-loop reorganization and kinase activation (lower panel). Whereas the wild-type kinase accommodates inhibitors that bind with αC “in” or “out,” β3/αC-deleted kinase is restricted to inhibitors that will bind when αC is “in.”

By determining crystal structures of the BRAF kinase domain harboring a 5 amino acid (aa) β3/αC deletion, Foster et al. provided a satisfying explanation for how this deletion activates the kinase by constraining αC in the expected active-like “in” position (Foster et al., 2016). Accommodating the deletion necessitates an unwinding of the most N-terminal turn of helix αC, as well as formation of a well-defined type I β turn between β3 and the shortened αC helix. With this rigid connection to the β sheet in the kinase N-lobe, it is clear that the position of αC will be highly constrained. Importantly, this constraint on the position of helix αC also explains the findings that BRAF variants harboring β3/αC deletions are resistant to inhibition by vemurafenib—an ATP-competitive inhibitor known to bind only when αC is in the “out” configuration—but can be inhibited by those that can bind the kinase when αC is “in” (Chen et al., 2016; Foster et al., 2016). Because the αC helix cannot move into the “out” position in the context of the β3/αC loop deletions, vemurafenib cannot be accommodated unless αC is distorted—as Foster et al. surprisingly observe for another inhibitor, dabrafenib. The structural analysis of β3/αC-deleted BRAF also provides a likely explanation for the impaired ability of lapatinib (which also binds only with αC “out”) to inhibit EGFR or HER2 variants with 5 aa deletions in the β33/αC loop, as previously reported for HER2 Δ755-759 mutations (Bose et al., 2013).

The precise nature by which the 5 aa deletion in the β3/αC loop is accommodated also suggests that 5 is a special number in this structural context. Indeed, by assessing β3/αC loop deletion length in EGFR across a spectrum of lung adenocarcinomas, Foster et al. found that deletions of 5 aa are by far the most common (~80%), with just 20% of cases exhibiting 6 aa deletions and only a handful occurring with 3, 4, or 7 amino acids deleted (Foster et al., 2016). Engineering deletions in EGFR and BRAF gave a similar impression, with a clear peak in activation when 5 aa was deleted from the loop. The structure makes it clear that deletions of more than 5 aa from the β3/αC loop would require more unwinding of αC (or other disruption), whereas deleting fewer than 5 aa would yield structures with different turns, likely restraining helix αC in different positions (or destabilizing the structure). Consistent with these interpretations, Foster et al. found that BRAF with a 5 aa β3/αC loop deletion is active regardless of dimerization (Foster et al., 2016), although the kinase domain dimerizes in the crystal. By contrast, the variant studied by Chen et al. (with a 6 aa deletion) appears to be dependent on dimerization for its ability to transform cells (Chen et al., 2016). Although quite variable across kinases in general, Foster et al. point out that β3/αC loop length is similar for EGFR, RAF, and Src family kinases. This is likely to reflect the similar autoinhibitory modes in these examples, requiring comparable latitude for the αC helix transition.

In a broader context, the findings reported by Foster et al. and Chen et al. boldly underline the need to understand the nature of kinase-activating mutations in cancer in order to make the appropriate treatment choice from the current palette of kinase inhibitors. As the number of known patient-derived mutations burgeons, this will amount first to understanding which are actually activating; recent studies in neuroblastoma, for example, revealed that a significant fraction are not (Bresler et al., 2014). Second, for those mutations that are found to be activating, a mechanistic understanding is essential for choosing between inhibitors that trap inactive conformations and those that bind active-like states. Structural considerations make it clear why trying to trap β3/αC-deleted BRAF, EGFR, or HER2 variants in the αC-out conformation is futile (Chen et al., 2016; Foster et al., 2016). Conversely, studies in glioblastoma indicate that certain subsets of oncogenically activated EGFR mutations are only responsive to inhibitors that bind the αC-out conformation (Vivanco et al., 2012). Each oncogenic kinase will have its own characteristics, demanding that individual mutations are interpreted in terms of molecular mechanism.