A large-scale study sheds light on the extraordinary molecular-recognition skills of the chaperone HSP90, which allow this protein to interact selectively with hundreds of other proteins of diverse function.
The protein HSP90 plays an essential part in a plethora of cellular processes, in evolution and in disease. It is a molecular chaperone, a type of protein required for the activity and stability of other proteins, which are known as its 'clients'. Many of HSP90's clients are oncogenic protein kinases — when made overactive, they can lead to cancer1. That is why 20 small-molecule HSP90 inhibitors are currently in clinical trials for antitumour therapy2. Despite all this interest, the fundamental question of how HSP90 chooses its clients remains unanswered. Writing in Cell, Taipale et al.3 report that another protein (a co-chaperone) facilitates client-family recognition by HSP90, whereas the thermal stability of the client determines the strength of its interaction with the chaperone itself.
In the absence of functional HSP90, its clients form aggregates or are degraded. So previous large-scale efforts to identify clients were based on either isolation of HSP90-binding proteins4 or demonstration of client depletion in cells in which HSP90 function had been perturbed (for example, by pharmacological inhibition)5. These studies have so far revealed around 350 highly diverse clients for HSP90.
Taipale et al. carried out their own large-scale survey of HSP90 clients among selected protein classes, including the protein kinases. They used a modification of the LUMIER assay6, a luminescence-based measure of association between a 'prey' protein (in this case, HSP90) and 'baits' (putative clients) in mammalian cells. Compared with other techniques used to detect binding, such as mass spectrometry, the modified LUMIER assay is much more sensitive, detects shorter-lived interactions and gives a quantitative rather than a binary (binding/no binding) readout. Of the 314 kinases investigated, 193 (61%) interacted to some extent with HSP90. Strikingly, the strength of these interactions varied across a continuous 100-fold range. This finding supports the growing view that the binary categorization of kinases into clients or non-clients should be replaced by a continuum of HSP90 dependence.
In agreement with an earlier analysis of binary data7, Taipale and colleagues confirmed and extended previous findings that even very closely related kinases (such as ARAF and BRAF) exhibit extreme differences in the strength of their interactions with HSP90 (refs 1,8). Seeking a structural explanation for this, Taipale et al. constructed and analysed variants of ARAF and BRAF containing single-point mutations, and also analysed chimaeric proteins made of different pieces taken from the two kinases. Notably, no single amino-acid change within the catalytic kinase domain (the part of the protein that defines a protein kinase) altered the protein's affinity for HSP90. Rather, the authors found that determinants for HSP90 association were widely distributed across the kinase domain.
Yet, surprisingly, they also showed that association of a client with HSP90 could be modulated by changes to certain highly localized regions in the kinase domain, regardless of their amino-acid sequences. The changes mimicked those produced by a natural process known as alternative splicing, by which a protein-encoding messenger RNA can be processed in different ways to generate various 'isoforms' of the protein. That finding might explain Taipale and colleagues' observation that isoforms of the same kinase can have distinct affinities for HSP90.
The authors noticed that fewer crystal structures had been reported for the kinase domains of HSP90 clients than for those of non-clients, which suggested that clients might be less soluble and stable than non-clients. On the basis of this and their previous findings, Taipale et al. hypothesized that affinity for HSP90 might be related to a client's instability. Indeed, they showed that the strength of the interactions between HSP90 and 56 kinases was correlated with the thermal instability of the kinase domain. The thermal-instability data used had been obtained previously by measuring the progressive unfolding of the proteins at increasing temperatures9, and probably reflect the presence of flexible, unstructured regions of the kinase domain that HSP90 can identify and bind to.
To delve deeper into their thermal-instability hypothesis, the authors focused on the oncogenic kinase BCR-ABL, because it is known9 that certain small-molecule inhibitors and activators of this kinase can lock the protein in a stable (active or inactive) conformation — which should increase its thermal stability. Indeed, treatment of cells with such molecules decreased the association of BCR-ABL with HSP90. Moreover, mutations that reduced or enhanced the thermal stability of BCR-ABL resulted in increased or decreased HSP90 affinity, respectively. Several lines of evidence therefore back up the view that the thermal stability of a kinase determines its association with HSP90 (Fig. 1).
But how does HSP90 'know' whether a protein is a kinase? There was previous evidence that the co-chaperone CDC37 mediated the interaction between HSP90 and certain kinases10. Taipale et al. now provide additional, systematic evidence that HSP90 requires CDC37 to recognize kinases — but not to recognize clients that are transcription-factor proteins. Future large-scale analyses incorporating other co-chaperones may reveal additional recognition determinants specific to various client classes. Such analyses could also be valuable in defining the multi-protein interactions established by HSP90 and how they are affected by small-molecule inhibitors of HSP90 and of client function.
In addition, the researchers report that HSP90 interacted with 117 proteins belonging to the E3 ubiquitin ligase class, which targets proteins for degradation. The recognition rules for choosing clients among these degradation partners must now be determined. Moreover, the mechanisms by which inhibition of HSP90 leads to the destruction of some clients but not others (which then form insoluble aggregates) remain unclear.
Taipale and colleagues' systematic and quantitative study indicates that the co-chaperone CDC37 helps HSP90 to recognize the kinase client class, and that the subsequent HSP90–kinase interaction is determined by an as-yet-unknown feature that can be measured in terms of thermal instability. Exactly how HSP90 recognizes thermally unstable kinase domains is a key question for future research. To answer it, we probably need to solve the three-dimensional structures of HSP90 multi-protein complexes containing clients. However, it is tempting to speculate that thermal instability is related to the flexibility of the kinase 'hinge' — a segment of the protein that connects the amino- and carboxy-terminal parts of a kinase domain — or other mobile regions.
Such emerging recognition mechanisms should eventually provide a molecular explanation for the various roles of HSP90. They are already beginning to reveal precisely how the chaperone acts as a driving force in kinase evolution: by stabilizing potentially advantageous but unstable protein forms, HSP90 buffers the effects of genetic variation that could be detrimental to the organism but that provide improved survival under stress conditions where HSP90 is partially inhibited11. In addition, these mechanisms can also explain how HSP90 exhibits its 'dark side': HSP90 protects kinases that are abnormally activated by mutations in cancer cells and that would otherwise be prone to aggregation or degradation12. Nevertheless, there is still much to be discovered about how HSP90 chooses its clients, co-chaperones and degradation partners in its highly complex, choreographed dance — knowledge that will have profound implications for evolutionary and cell biology, as well as for the treatment of disease.
About this article
Chemical Communications (2013)