A Chemical Biology Approach to the Chaperome in Cancer—HSP90 and Beyond

Abstract

Cancer is often associated with alterations in the chaperome, a collection of chaperones, cochaperones, and other cofactors. Changes in the expression levels of components of the chaperome, in the interaction strength among chaperome components, alterations in chaperome constituency, and in the cellular location of chaperome members, are all hallmarks of cancer. Here we aim to provide an overview on how chemical biology has played a role in deciphering such complexity in the biology of the chaperome in cancer and in other diseases. The focus here is narrow and on pathologic changes in the chaperome executed by enhancing the interaction strength between components of distinct chaperome pathways, specifically between those of HSP90 and HSP70 pathways. We will review chemical tools and chemical probe-based assays, with a focus on HSP90. We will discuss how kinetic binding, not classical equilibrium binding, is most appropriate in the development of drugs and probes for the chaperome in disease. We will then present our view on how chaperome inhibitors may become potential drugs and diagnostics in cancer.

The last two decades have been exciting times in seeing a new science take shape and develop—chemical biology. Continuously defining itself (Bertozzi et al. 2015), the scope of chemical biology is broad and highly multidisciplinary. We will discuss here one approach of chemical biology, wherein chemicals are “mutated” to obtain knowledge on a protein or a cellular state in a manner akin to how proteins are mutated in structural biology or cells are engineered in cell biology. These mutated chemicals are commonly referred to as probes or chemical probes, and we often also hear the term chemical toolbox or toolset, when not one but many probes are available to study and investigate a specific biological question. Chemical probes come in many flavors, from ligands that inhibit or activate a target, solid-support attached ligands, fluorescently labeled or radiolabeled ligands, among others, and each probe may have a specific or multifaceted role in addressing biology as we will exemplify below. Our focus and deep interest in chemical biology have always transcended beyond basic biological investigation and into the realm of applying our understanding toward the treatment of human disease.

Chemical biology has played an important role in our understanding of the chaperome as it has allowed us to ask questions that are less amenable to being addressed with classical methods such as those routinely used in cell biology, genetics, and biochemistry (Lindquist and Kelly 2011; Brandvold and Morimoto 2015; Shrestha et al. 2016b; Blagg and Workman 2017; Gestwicki and Shao 2018; Zuehlke et al. 2018). Here we will expand on these seemingly bold statements by providing examples of how chemical biology has contributed to our understanding of the chaperome and its role in proteostasis in disease.

The term chaperome was introduced in 2006 to characterize an assembly of chaperones, cochaperones, and related factors (Wang et al. 2006). Compiled lists of the human chaperome came soon after (Finka and Goloubinoff 2013; Brehme et al. 2014; Brehme and Voisine 2016; Esfahani et al. 2018). These initially included 147 bioinformatically predicted members, such as the heat shock protein 90s (HSP90s), HSP70s, HSP60, HSP110s, HSP40s (also known as DNAJ proteins), HSP10, and the small HSPs (sHSPs), as well as their cochaperones and members of the folding peptidylprolyl isomerase and protein disulfide isomerase enzymes. The name of each HSP family is derived from the molecular weight of the original founding member. In eukaryotes, most families also have organelle-specific members, such as those expressed in the endoplasmic reticulum (ER) and mitochondria. For example, HSP90s, in addition to the cytosolic HSP90α and HSP90β, also have an ER and a mitochondrial paralog, glucose-regulated protein 94 (GRP94, also called endoplasmin, gp96), and tumor necrosis factor type 1 receptor-associated protein (TRAP1, also called HSP75), respectively. The chaperome list was recently expanded to 332 chaperones and cochaperones and included tetratricopeptide repeat (TPR)-domain-containing proteins, selected on the basis of their functional interactions with chaperones (Allan and Ratajczak 2011). A web-based chaperome analysis tool is now also available (Esfahani et al. 2018). We believe the inventory of the chaperome will continue to be refined as our understanding of this assembly continues to develop.

In addition to being a large group of proteins, the chaperome is also one of the most abundant protein assembly in human cells. For example, in HeLa cancer cells, the 147 chaperome members together contribute 7.6% of the total number of polypeptides and 10.3% of the total protein mass (Finka and Goloubinoff 2013). HSP90s are the most abundant members, averaging 2.8% of the total protein mass alone and together with the HSP70s up to 5.5%, whereas 1.5% of the total mass consists mostly of substoichiometric regulatory cochaperones of the HSP90 and HSP70 machineries.

The chaperome is continuously being revealed, both structurally and functionally. We have learned how disease states modify the chaperome by changing the expression levels and posttranslational modification (PTM) of chaperome members (Dunn et al. 2015; Woodford et al. 2016a), the interaction strength among chaperome components (Rodina et al. 2016; Joshi et al. 2018; Kishinevsky et al. 2018), the constituency of chaperome complexes, as well as the subcellular localization of chaperome members (Caino and Altieri 2016; Shevtsov and Multhoff 2016; Wong and Jay 2016; Boudesco et al. 2018; Calderwood 2018). We have also learned how specific chaperome functions are being altered in the context of disease (Woodford et al. 2016b; Boland et al. 2018; Hoter et al. 2018; Joshi et al. 2018; Kropski and Blackwell 2018; Rodríguez-Iturbe and Johnson 2018; Yakubu and Morano 2018). A number of excellent review articles and perspectives, including those comprising articles in this collection, are written on chaperome members and their role in cellular physiology and pathology (Gorenberg and Chandra 2017; Schopf et al. 2017; Sontag et al. 2017; Stürner and Behl 2017; Zuiderweg et al. 2017; Gomez-Pastor et al. 2018; Joshi et al. 2018; Klaips et al. 2018; Wang et al. 2018). We will provide here snippets on the contribution of chemical biology to such knowledge. The focus here is narrow. It is on pathologic changes in the chaperome executed by enhancing the interaction strength between components of distinct chaperome machineries. The probes we discuss interact with HSP90, and therefore the chaperome changes we discuss reflect those involving this chaperone.

CHEMICAL BIOLOGY AND HSP90—A BRIEF HISTORY

HSP90 was the first chaperome member for which chemical biology has made a significant impact in terms of its targetability. In an era where cancer therapeutics are primarily driven by genetic abnormalities, it was chemical biology that added the abundant HSP90 chaperone onto the map of potential cancer targets (Neckers 2006). Indeed, HSP90, discovered in the 1980s, had drawn little interest until the early 1990s. The discovery of a chemical tool, geldanamycin (GM), in a screen of compounds able to revert the phenotype of cells transfected with the v-src oncogene, altered this state of affairs (Whitesell et al. 1992). The target of GM, like for many hits derived from phenotypic screens, remained unknown until Neckers and colleagues at the National Institutes of Health (NIH) took a chemical biology approach (Whitesell et al. 1994). Namely, they attached GM to a solid support to create another chemical probe, GM-beads, and then incubated this with cellular homogenates to isolate a protein with a molecular mass of approximately 90 kDa. Follow-up experiments, including microsequencing of protein material bound to radiolabeled affinity-tagged GM (also a chemical probe), revealed HSP90 as the target of GM. Subsequently, crystal structures of GM bound to HSP90 confirmed the ATP regulatory pocket as the binding site (Stebbins et al. 1997; Roe et al. 1999). Soon after, the toolbox of HSP90 probes started to fill, first with radicicol (Schulte et al. 1998) and other natural products (Amolins and Blagg 2009), followed by synthetic probes, such as PU3 (Chiosis et al. 2001), a ligand designed to interact with the regulatory pocket of HSP90. Other interactors soon followed: first came probes that bound to the ATP-binding pocket (Messaoudi et al. 2011; Patel et al. 2011; Jhaveri et al. 2014; Taldone et al. 2014b), but today the HSP90 toolbox has expanded to include allosteric regulators that bind in noncanonical ways, including those that target the carboxy-terminal domain of HSP90 (Koay et al. 2016; Kumar et al. 2018; Ferraro et al. 2019) and those that target protein–protein interactions between HSP90 and cochaperones (Yang et al. 2006; Cortajarena et al. 2008; Yi and Regan 2008; Zhang et al. 2008; Sreeramulu et al. 2009; Yi et al. 2009) or client proteins (Plescia et al. 2005; Meli et al. 2006; Liu et al. 2010). Following this and a deeper understanding of ligand-binding modes was a chemical toolset to selectively probe HSP90 paralogs in humans (Altieri et al. 2012; Patel et al. 2013; Gewirth 2016; Shrestha et al. 2016b; Que et al. 2018) and HSP90 in other organisms (Wang et al. 2014).

With a variety of ATP-competitive HSP90 inhibitors reported soon after 2000, investigations into HSP90 have exponentially increased in cancer biology. These probes were used extensively to understand the client proteins of HSP90 (Wayne et al. 2011; Radli and Rüdiger 2018; Sima and Richter 2018), the contribution of PTMs to HSP90 in cancer (Woodford et al. 2016a) and the mechanism by which HSP90 may execute its oncogenic properties (Taldone et al. 2014a; Calderwood and Neckers 2016; Isaacs 2016; Shrestha et al. 2016b). These probes were also used to derive information that eventually moved HSP90 inhibitors into the clinic as potential cancer therapeutics (Alarcon et al. 2012; Travers et al. 2012; Garcia-Carbonero et al. 2013; Jhaveri et al. 2014). Numerous compounds have entered into clinical evaluation over the last 10 years, and all of these have in common that they act by competing with ATP for binding to the amino-terminal-binding domain (Patel et al. 2011; Neckers and Workman 2012; Soga et al. 2013; Shrestha et al. 2016a).

CHEMICAL PROBES—EQUILIBRIUM AND KINETIC BINDING

With these ATP-competitive ligands inserting in the same pocket, there was little reason initially to believe that such inhibitors could distinctively modulate HSP90. Furthermore, with HSP90 viewed as the centerpiece of a machinery on which certain client proteins, such as kinases, are transferred over from the HSP70 pathway for folding and maturation (Schopf et al. 2017), most studies with these inhibitors addressed HSP90 and its significance as reflected by the client (Travers et al. 2012). A model was built based on the assumption that if an oncoprotein (or several oncoproteins), so-called “client protein(s),” required HSP90 for structural stabilization and/or functional modulation, then an HSP90 inhibitor will induce the degradation of such client(s) in tumors, and in turn the tumor will respond to HSP90 therapy (Trepel et al. 2010; Jhaveri et al. 2012). Indeed, a number of distinct inhibitors depleted the same clients, and this was attributed to a common mechanism of action (Wu et al. 2012).

However, differences among inhibitors, which seemingly bind to the same pocket, were observed during various studies (Chiosis et al. 2003; Caldas-Lopes et al. 2009; Pimienta et al. 2011). Intrigued by these differences, poorly explained by the prevailing mechanistic model, we created a number of chemical probes to investigate the nature of such divergences. These probes were made by immobilizing individual ATP-competitive HSP90 ligands onto agarose beads (Taldone et al. 2011b). In addition to the GM-beads reported earlier by Neckers (Whitesell et al. 1994), PU-H71 (He et al. 2006), SNX-2112 (Huang et al. 2009), and NVP-AUY922 (Brough et al. 2008) immobilized onto a solid support (Taldone et al. 2011b). In each case, careful consideration was given to the available cocrystal structure so that these probes could be designed with minimal effect on binding to HSP90 (see below).

We incubated these probes with a cancer cell homogenate to find that each isolated a distinct pool of cellular HSP90 (Moulick et al. 2011). We used K562 chronic myeloid leukemia cells, which express both the oncogenic BCR-ABL and the physiological c-ABL, both of which bind HSP90 as evidenced by immunoprecipitation with an HSP90 antibody. We observed differences among these probes. Although GM was very effective in isolating HSP90, it was poor at isolating its complexes with client proteins. Other probes isolated HSP90 bound to clients, some however, captured both the BCR-ABL and the c-ABL HSP90 complexes, whereas others, such as PU-H71, preferred the BCR-ABL-bound HSP90 pool (Moulick et al. 2011). Ligands that preferred HSP90 bound to BCR-ABL, also showed this proclivity for HSP90 bound to oncogenic v-FLIP and v-SRC over the physiological c-FLIP and c-SRC, respectively (Moulick et al. 2011; Nayar et al. 2013; Ojala 2013). An interesting observation from these early studies was that the HSP90 pool that bound BCR-ABL also tightly bound several cochaperones, such as those that function in concert with HSP70 (Moulick et al. 2011). BCR-ABL but not c-ABL was sensitive to HSP70 knockdown in these cells (Moulick et al. 2011). HSP90, therefore, binds both BCR-ABL and ABL, but there is a structural and/or dynamic difference between the two pools, and this difference is sensed by the chemical probes. Presumably, in addition to being structurally distinct, these HSP90 pools also have divergent functions, with the BCR-ABL but not the c-ABL pool supporting the oncogenic properties of K562 cells (Fig. 1A). It remains to be elucidated how the different HSP90 pools execute such divergent functions. It was recently hypothesized that the HSP90 pools that regulate oncogenic protein function, such as those of BCR-ABL, act as scaffolding platforms to enable a restructuring of protein–protein interactions and therefore the rewiring of cellular protein networks, rather than in a one-on-one, cyclic fashion, exhibited by distinct molecular chaperones and cochaperones that aid in protein folding and degradation (Joshi et al. 2018).

Figure 1.

Structural and functional heterogeneity of the chaperome and implications for small molecule interaction and activity. (A) K562 cancer cells have at least two structurally and functionally distinct heat shock protein 90 (HSP90) pools. One regulates the folding of c-ABL through formation of dynamic protein complexes characterized by weak protein–protein interactions. Regulation of BCR-ABL, on the other hand, is performed by stable HSP90 complexes characterized by strong interactions between the protein partners. These complexes act as scaffolding platforms to enhance and rewire oncogenic protein and protein network activity. (B) The dynamics and strength of interaction between HSP90 and its protein partners, as well as the constitution of the protein complexes in the two HSP90 pools may be kinetically differentiated by small molecules. The dissociation rate of PU-H71 is a determining factor in providing this chemical probe with selectivity for one pool over the others. (C) The heterogeneity of HSP90 observed among nontransformed cells and tumor cells, as well as among different tumor cells, can be detected by chemical probes. It is the kinetic and not the thermodynamic selectivity that play a determining role for probes such as PU-H71.

These observations were intriguing but presented a dilemma. How could ligands that are seemingly bound to the same site on HSP90 have such diverse activity? If different pools of HSP90 are present in a cancer cell, and if inhibitors preferred one pool over others, why then did they degrade a similar spectrum of clients in a given cancer cell?

To answer these questions, we must first consider that the treatment of cells is generally performed under a state of equilibrium. When cells in culture are treated with HSP90 agents for hours, more often for 24 h, the observed biological effect is a direct reflection of equilibrium binding. Under these long incubation conditions, the compound is present inside the cell at a constant concentration over the duration of the experiment, so that the ligand binds to various HSP90 pools in a thermodynamic manner. Under these conditions, the parameters measured at equilibrium, namely, affinity constants (i.e., K_d and IC₅₀/EC₅₀), are useful indicators to describe the effects of HSP90 agents in cell culture. Under equilibrium binding however, inhibitors may not necessarily show selectivity based on K_d alone.

Affinity purification experiments, on the other hand, are driven by a yet additional binding feature, and that is the kinetics of binding. Here, the rate of association (k_on) to and/or dissociation (k_off) of a ligand from a specific HSP90 pool may determine its selectivity. To appreciate this concept, we should discuss how affinity purifications are performed. Here the bait is attached onto a solid support and is incubated with a cell homogenate for a determined time. The cargo (in this case the HSP90 pools and their interactome) is then isolated. Assuming the binding step is done with excess chemical probe and for a sufficiently long time, the capture reflects a state of equilibrium binding where a majority of the HSP90 pools are bound onto the chemical probe. The next step of an affinity purification is to separate the solid support, and thus its cargo, from the supernatant, and then wash the cargo repeatedly with buffer. Over these sequential washes, there is a gradual disruption of the cargo from the chemical probe; this is directly dependent on dissociation rates. The chaperome pool that dissociates slowest from the chemical probe (i.e., has a slow k_off) is retained longest on the bait. In other words, the probe kinetically selects for a specific HSP90 pool even if its affinity (i.e., K_d) for the several intracellular HSP90 pools may (or may not) be the same.

Another way to appreciate kinetic selectivity is to refer to classic methods for evaluation of the dissociation reaction kinetics of a ligand from its receptor. One way is to compete a fluorescent or radioactive ligand from its receptor with a large excess of unlabeled ligand. Here, the labeled ligand dissociates at a rate determined by k_off and does not rebind, because the unlabeled ligand takes its place. This method was recently used to show that the dissociation rate of PU-H71 was dictated by the intrinsic nature of HSP90 pools found in one cancer cell line over another, and that it was independent of the total HSP90 levels (Rodina et al. 2016; Wang et al. 2018). This finding was reproduced when binding dissociation was then evaluated in cells and in xenografted tumors in mice (Rodina et al. 2016).

Another way is to dilute an equilibrium mixture of ligand, receptor, and ligand–receptor complex and observe the time course of the dissociation of complex to establish new equilibrium concentrations of ligand and receptor. As we have previously discussed (Patel et al. 2011; Taldone et al. 2014a; Shrestha et al. 2016b; Wang et al. 2018), monitoring pharmacokinetic (PK) properties of drugs after bolus administration may, in essence, be a surrogate of such a method. When we studied the in vivo properties of PU24FCl, an early generation probe and a precursor to the clinical agent PU-H71, we observed that distribution of the agent was rapid in all tissues, and so was its binding. Clearance, however, was fast from plasma and normal tissues but not from the tumor (Vilenchik et al. 2004). This “tumor-retention” characteristic, or long tumor residence time (T_r), was not a result of drug nonspecifically partitioning to tumor tissue or because of other nontarget-related factors, because oncogenic proteins and pathways remained suppressed in tumors for the duration the agent also remained in the tumor, and the magnitude of oncogenic protein suppression corresponded to the concentration of PU24FCl measured in the tumor. In other words, tumor PK (i.e., the concentration of PU24FCl in the tumor mass) mirrored tumor pharmacodynamics (PD) (i.e., engagement of the target in the tumor mass), and the off-rate (k_off), or the dissociation constant from the target in tumor, determined the biological activity of this agent in tumor-bearing mice.

This extended tumor-retention profile was soon reported for other HSP90 agents, which demonstrated in vivo antitumor activity (Eiseman et al. 2005; Sydor et al. 2006; Chandarlapaty et al. 2008; Eccles et al. 2008; Jensen et al. 2008; Bao et al. 2009a,b; Caldas-Lopes et al. 2009; Leow et al. 2009; Lundgren et al. 2009; Woodhead et al. 2010; Graham et al. 2012; Shi et al. 2012; Shimamura et al. 2012), albeit the off-rate and PD effects were vastly different among the many evaluated agents (Patel et al. 2011), indicating again a different selectivity profile of these many agents for the various HSP90 pools. It should be restated that in tissue culture, under conditions of equilibrium, several of these agents have similar biological activity (Wu et al. 2012; Rodina et al. 2016).

Why are nonequilibrium binding parameters of essence in vivo? While drug concentrations are kept constant in in vitro experiments (so that equilibrium binding can be reached after a sufficiently long incubation time), this is not the case in the human (or mouse) body because the body is a system in flux. The concentration of a free drug in an in vivo setting changes over time. It is determined by the drug's rate of absorption, distribution, and elimination, which combine to ensure that drug concentration, both locally at the target and in circulation in the body, is in constant fluctuation. Under these conditions, a kinetic approach must be taken, one where the rates of association and dissociation to target are considered. These kinetic factors in vivo, influence not only potency and selectivity, but can also greatly impact the safety profile of an inhibitor (Figs. 1B,C; Taldone et al. 2014a; Joshi et al. 2018).

A classic example of the importance of kinetic selectivity in drug safety is the development of drugs that target muscarinic receptors (MRs) to treat asthma (Moulton and Fryer 2011). Inhibitors of MRs, such as atropine and scopolamine, are known to humans since ancient times as they are naturally occurring in several nightshade family plants such as in Atropa belladonna, Datura stramonium (Jimsonweed), and Hyoscyamus niger (henbane). Yet the pharmacologically effective dose range for atropine and others is very close to the toxic range, limiting their use in asthma treatment. There are several MR receptors, with the M₃ being responsible for the antiasthma activity and the M₂ for toxicity. The several muscarinic antagonists developed so far, such as atropium, ipatropium, clidinium, and tiotropium, are characterized by similar K_ds (i.e., equal affinity for several of the MR receptors). Yet, tiotropium, despite having equal affinity for the M₁, M₂, and M₃ receptors, is functionally selective for M₃ receptors. This selectivity is provided by the ability of tiotropium to dissociate from M₂ receptors 10 times faster than it does from M₃ receptors (T_1/2 = 3.6 h for M₂ vs. T_1/2 = 34.7 h for M₃) (Moulton and Fryer 2011). This kinetic selectivity for the M₃ receptor provides tiotropium with a superior safety profile over the other inhibitors and supports its clinical application for chronic obstructive pulmonary disease (Disse et al. 1999).

Similarities can be drawn to HSP90 agents. For example, HSP90 agents that lingered in the eye or in the gastrointestinal (GI) tract, resulted in visual disturbances and GI toxicity presumably limiting the dose such agents could be administered at in clinic for maximal target engagement (Kanamaru et al. 2014; Stjepanovic et al. 2016; Shah et al. 2018). By the same token, attempts for daily dosing in HSP90 therapy have also been limited by toxicity, and not paradoxically, better responses were seen in the clinic when the agent was given a dose every other day rather than when given daily (Bauer et al. 2013).

If we assume that the tumor-retention profile of a particular HSP90 probe is a result of significantly diminished k_off, and thus enhanced tumor residence time (T_r = 1/k_off), this suggests that there is something distinct between HSP90 pools found in normal tissue versus those found in tumor tissue, and among the HSP90 pools of distinct tumors. Combined with the affinity purification experiments described above, where it was found that HSP90 bound both BCR-ABL and c-ABL, but probes could isolate one pool over the other, one concludes that heterogeneity in the structural and functional makeup of HSP90 exists not only between normal cells and cancer cells, and between distinct cancer cells, but also inside a given cancer cell (Fig. 1).

Application of kinetic binding analysis, rather than classical equilibrium binding, toward the behavior of HSP90-targeted agents could therefore be more revealing to explain inhibitor selectivity and in better understanding inhibitor behavior under nonequilibrium experimental conditions (Copeland et al. 2006; Tonge 2018). The kinetic approach utilizes rate constants (k_on and k_off), which better characterize the heterogeneity and the dynamics of a system (Schuetz et al. 2017; Tonge 2018). There is some debate as to their relative importance (Folmer 2018); however, the general consensus is that k_off is more significant and more desirable to optimize for a ligand (Copeland et al. 2006; Copeland 2016; Tonge 2018). Because the dissociation rate (k_off) is determined by the difference in energy between the ground-state inhibitor-bound complex and the transition state leading to its formation, the rate can be favorably modified (low k_off) either by stabilizing the ground-state inhibitor-bound complex or by destabilizing the transition state leading to its formation. Although it is appreciated that favorable interactions between the inhibitor and protein stabilize the ground state, factors leading to transition state destabilization are less understood (Lu and Tonge 2010).

For HSP90, one can envision a multitude of complexes that a ligand can potentially bind to, each one leading to a different transition state complex that would then lead to a different inhibitor-bound complex. Alternatively, and not necessarily an exclusive mechanism, is one where the ligand induces a conformational change to HSP90 following binding (Immormino et al. 2006; Moulick et al. 2011; Amaral et al. 2017). Recent studies shed light on these possible mechanisms (Taldone et al. 2013a; Amaral et al. 2017). Although seemingly binding to the same pocket in HSP90, a closer look at the many available HSP90-ligand crystal structures indicated that several differences exist in the ligand-binding mode and the conformation of the pocket upon ligand binding (Taldone et al. 2013a). Based on these considerations, the ATP-binding pocket could be subdivided into three major subpockets (referred to as A, B, and C); inhibitors occupied some or all of these subpockets. The binding of GM and several resorcinols was restricted to subpockets A and C. Subpocket B was unveiled by PU-H71 by inducing a conformational change in the flexible lid of HSP90, creating a new binding channel (Immormino et al. 2006; Taldone et al. 2013a). Ligands that bound HSP90 in a “helical” conformation (the pocket configuration unveiled by PU-H71) displayed much slower dissociation rates when compared to similar ligands that bound in a “loop” conformation (the configuration of GM-bound HSP90) (Amaral et al. 2017). Conformational adaptation of HSP90 can, therefore, greatly influence the residence time (i.e., the dissociation rate) of an inhibitor on the target. The above analyses were shown with recombinant HSP90, so one may envision a more complicated situation in the cell. Here HSP90 can be further modified by a variety of factors, such as PTMs, the binding strength between partners and the nature of binding partners, cellular location and others, in creating a heterogeneous pool of HSP90s (Graner 2016; Rodina et al. 2016; Zuehlke et al. 2017; Bachman et al. 2018). These factors may further influence HSP90 conformations, and the binding of probes to the HSP90 protein (Beebe et al. 2013; Mollapour et al. 2014; Rodina et al. 2016). These cell-specific HSP90 conformations might be intrinsically present (Rodina et al. 2016) or might be inhibitor induced (Moulick et al. 2011), but as stated above, these two mechanisms need not be exclusive.

CHEMICAL PROBES IN ADDRESSING CHAPEROME ESSENTIALITY

Thus, chemical biology, by providing probes that differentiate, and selectively identify the native HSP90 pools of tumors, has provided valuable insights into both the biochemical and functional heterogeneity of the chaperome in cancer with implications to inhibitor design. We will next discuss how chemical probes have helped in the clinical development of HSP90 drugs, specifically in the identification of tumors vulnerable to HSP90 therapy.

A large number of HSP90 inhibitors have entered clinical evaluation in cancer. These studies were based on a specific model developed for HSP90: If an HSP90 client protein is important for the function of a particular tumor cell, then the tumor will respond to HSP90 therapy (Alarcon et al. 2012; Jhaveri et al. 2012, 2014; Travers et al. 2012; Garcia-Carbonero et al. 2013). This model did well to some extent, for example in HER2- or ALK-driven cancers, and these clinical successes have been reviewed elsewhere (Butler et al. 2015; Shrestha et al. 2016a; Yuno et al. 2018). Several review articles were also written on potential limitations of HSP90 drugs, such as stemming from a feedback heat shock activation, and we direct the reader to these publications (Butler et al. 2015; Shrestha et al. 2016a; Kijima et al. 2018; Yuno et al. 2018).

With the “client-dependence” model being of limited value, the medical community was left with little means for patient selection (Scaltriti et al. 2012). Responses seen in the clinic (Garcia-Carbonero et al. 2013; Jhaveri et al. 2014) could not be built into the large clinical study needed for drug approval. The disconnect between the potential of the target and its limited success in clinic based on the “client protein” selection criteria alone, has led to a state of disillusion, and overt negativity toward HSP90, and the chaperome field in general.

Identification of biomarkers that may predict patient response to HSP90 drugs therefore remains an important and a highly sought-after goal. A chemical biology approach has recently provided answers to this puzzle and, most importantly, possible therapeutic solutions as we detail below. Rodina et al. observed that in cancer cells and in primary specimens, inhibition of HSP90 resulted in depletion of HSP90 “client proteins” irrespective of whether they were sensitive to HSP90 drugs or not. HSP70 induction, considered to be a marker of heat shock activation and often linked to reduced sensitivity to HSP90 inhibition, was observed equally in sensitive and resistant cancer cells (Rodina et al. 2016; Joshi et al. 2018). Client protein depletion and HSF-1 activation are, therefore, insufficient to explain tumor vulnerability to HSP90 inhibition.

A distinguishing factor emerged when the biochemical signature of chaperome members was analyzed under nondenaturing conditions. Cancer cells that were highly vulnerable to HSP90 inhibition contained heterooligomeric species of HSP90, HSC70, HOP, HSP110, CDC37, AHA1, and other chaperome members, which remained stable under native polyacrylamide gel electrophoresis (PAGE) and isoelectric focusing conditions (Rodina et al. 2016). In addition to being sensitive to HSP90 inhibition, cells that contained the hetero-oligomeric species were sensitive to inhibition of several other chaperome members (siRNA knockdown of HOP, HSP110, and AHA-1 were tested) (Rodina et al. 2016). Not all cancer cells formed these stable chaperome complexes, and these chaperome pools were absent in cancer cells that recovered from HSP90 inhibition. These stable chaperome pools were also absent in nontransformed cells.

The formation of the stable hetero-oligomeric chaperome complexes was independent of the expression level of HSP90 and other chaperome members, HSP90 client proteins, antiapoptotic molecules, tissue of origin or causal genetic mutations. To understand their nature, Rodina et al. applied an unbiased functional proteomics approach to both inhibitor-sensitive and -resistant cancer cells. The method uses chemical bait coupled with bioinformatics (see further) to identify the chaperome and its interactome, and to assess the connectivity between these proteins. This hypothesis-generating method, coupled with biochemical and functional validation, demonstrated that a biological outcome of stable oligomeric chaperome pools was the rewiring of chaperome connectivity (Fig. 2; Rodina et al. 2016). Specifically, formation of these stable chaperome complexes connected individual chaperome pathways into a large cellular network where connectivity among participants extended beyond that seen in the resistant cancer cells (Fig. 2). Thus, HSP90 essentiality was measured by its connectivity—HSP90 became essential when its network connections increased through engagement in protein complexes with chaperome members of other chaperome pathways (such as HSC70, the constitutive HSP70 paralog, and others). The increased interactions allowed the previously nonessential HSP90 to become a member of global (as opposed to insular) protein pathways. We coined the term epichaperome to describe these hyperconnected chaperome networks. Hyperconnecting the chaperome by increasing the interaction strength between individual chaperome members of several distinct chaperome pathways, also resulted in a functional integration of these distinct chaperome networks. Indeed, in the epichaperome expressors, but not in cancer cells with insular networks, knockdown of HSP110 (an HSP70 cochaperone) and HOP was sufficient to diminish the activity or expression of HSP90 kinases, such as p-S6K, p-ERK, and EGFR, in a fashion similar to knockdown of HSP90α/β and AHA1 (a direct HSP90 activator).

Figure 2.

Chaperome network connectivity determines the sensitivity of cancer cells to heat shock protein 90 (HSP90) inhibition. Tumors could be divided in two categories based on the properties of their cellular networks, insular and hyperconnected. In the insular network, HSP90 and HSP70 are hubs of protein networks, but they functioned separately, each with its cochaperone subset and each as a hub of its own protein network. Limited connectivity exists between the two machineries. In tumors characterized by such “insular” chaperome networks, inhibition of components of the HSP90 or the HSP70 network is not lethal. HSP90 becomes essential in cancer when its network connections increased through engagement in protein complexes with chaperome members of other chaperome machineries (such as those of HSC70, the constitutive HSP70 paralog). The increased interactions allow the previously nonessential HSP90 to become a member of global (as opposed to insular) protein pathways. We coined the term epichaperome to describe these hyperconnected chaperome networks. Epichaperome formation is executed by increasing the interaction strength between HSP90 and other chaperome members, resulting in protein complexes of enhanced stability. Proteome imbalances, such as induced by MYC hyperactivation, are partly responsible for driving the “rewiring” of HSP90 into the epichaperome in cancer.

In the HSP90 inhibitor-resistant cancer cells, HSP90 and HSP70 were hubs of protein networks, but they functioned separately, each with its cochaperone subset and each as a hub of its own protein network (Fig. 2; Rodina et al. 2016). This is similar to the case in HEK293T cells, where HSP90 and HSP70 regulate their own protein network, with limited connectivity between the two machineries, mainly executed by the ubiquitous HOP and CHIP cochaperones (Taipale et al. 2014). In tumors characterized by such “insular” chaperome networks, inhibition of components of the HSP90 or the HSP70 network was not lethal (Rodina et al. 2016). Importantly, nontransformed cells are also characterized by insular chaperome networks, and overcome acute chaperome inhibition (Rodina et al. 2016; Kishinevsky et al. 2018).

Introduction of the bona fide HSP90 client proteins v-Src or mutant MET kinase into NIH-3T3 cells failed to rewire the chaperome into epichaperomes, yet resulted in an increase in the cellular levels of several chaperones and cochaperones, such as HSP70, HSC70, HOP, HSP110, HSP40, and AHA1 (Rodina et al. 2016). Larger proteome stresses (Harper and Bennett 2016), as opposed to single “client protein” misfolding, were necessary for chaperome connectivity rewiring to occur (Zong et al. 2015; Rodina et al. 2016; Kishinevsky et al. 2018; Kourtis et al. 2018). MYC hyperactivation is such a proteome stressor. Knockdown of MYC was sufficient to reduce chaperome connectivity, and turn a hyperconnected network into an insular one (Fig. 2). The introduction of a functional MYC gene was sufficient to rewire chaperome's connectivity from an insular to a hyperconnected character, and sensitize cells to PU-H71 (Rodina et al. 2016). NOTCH, which acts as an upstream MYC activator in T-cell acute lymphoblastic leukemia (T-ALL), is an epichaperome inducer in this context (Kourtis et al. 2018). Blocking NOTCH activity, by preventing its cleavage at the cell surface with γ-secretase inhibitors, negatively modulated MYC, and in turn chaperome network connectivity and cancer cell's sensitivity to PU-H71. In the context of AML, a hyperactive signalosome may be sufficient to rewire chaperome connectivity (Zong et al. 2015). Here a direct and quantitative link was found between the hyperactivity of signaling pathways and the apoptotic sensitivity of AML to PU-H71 (Zong et al. 2015).

TAKING ADVANTAGE OF KINETIC SELECTIVITY FOR DIAGNOSTICS

Chemical biology studies, therefore, propose a novel means for patient selection where properties of protein–protein interaction networks, not genetics or client proteins, may drive HSP90 therapy implementation. Concordantly, because the epichaperome is a tumor mechanism portending therapeutic vulnerability, the epichaperome may be a biomarker (Rodina et al. 2016; Joshi et al. 2018). Chemical probes able to identify these HSP90 pools incorporated into the epichaperome in live patients and/or clinical specimens are, therefore, useful for such patient-selection diagnostic purposes (Fig. 3). The ability of PU-H71 to kinetically select for HSP90 incorporated into epichaperomes offered a valuable starting point. For PU-H71, we have shown through a variety of methods using both cell homogenates and live cells, that kinetic binding, with a slow off-rate for specific tumor HSP90 pools, provides the selectivity of PU-H71 for HSP90 pools incorporated into epichaperomes (Rodina et al. 2016). We prepared a number of modified chemical tool molecules such as fluorescently labeled or radiolabeled PU-H71 for this purpose (Taldone et al. 2011a,b, 2013b, 2016), and each of these tools has specific uses for which they are ideally suited. One of the most important factors to consider when designing these probes is to ensure that modifications do not affect binding to target. The crystal structure of PU-H71 bound to HSP90α (PDB ID: 2FWZ) shows that the purine ring as well as the 8-aryl ring make essential contacts within the nucleotide-binding pocket, whereas the N9-alkyl chain was oriented toward solvent (Immormino et al. 2006). Based on this as well as docking analysis, the N9-alkyl chain is the ideal site for modification of PU-H71, and several probes were designed modifying PU-H71 at this location. Introduction of such probes into clinical assays is enabled by technical advances in both flow cytometry and molecular imaging.

Figure 3.

Chemical probes and methods to detect heat shock protein 90 (HSP90) restructuring into epichaperomes in clinical samples. (A) A flow cytometry-based assay that utilizes a fluorescently labeled PU-H71 (PU-FITC) is used to detect epichaperome levels in liquid tumors (for example, leukemias). Here the intensity of the FITC signal in the malignant cells is compared to standard and control samples (such as the FITC signal in normal cells), and the ratio of the signals denotes the amount of the epichaperome (i.e., the HSP90 incorporated into epichaperome networks). (B) For solid tumors, a positron emission tomography (PET)-based assay utilizes iodine124 labeled PU-H71 (124I-PU-H71) to identify tumors that express HSP90-incorporating epichaperome networks. A positive PET signal detected in the tumor at 24 h post-124I-labeled PU-H71 injection of a small, nonbiological tracer amount, denotes epichaperome presence in the tumor—the intensity of the signal denotes the amount of the epichaperome in the tumor.

Although flow cytometry was introduced in the 1960s as an analytical technique able to measure various characteristics of single cells in suspension following excitation with a light source, in clinical oncology, it was initially used to study the DNA content for cell ploidy and proliferation activity determination (Danova et al. 2018). The discovery of monoclonal antibodies and the introduction of new fluorescent dyes with narrow excitation and emission spectra, expanded the application of this technology, especially in the context of liquid tumors, or hematological oncology. Having an appropriately fluorescently labeled PU-H71 could, therefore, quickly enable its introduction into the clinic in the context of liquid tumors. Such a fluorescently labeled tool, PU-FITC, was prepared by attachment of FITC to PU-H71 directly onto the basic amine of the N9-alkyl chain (Taldone et al. 2011a).

This chemical probe was validated in a number of preclinical studies and is currently used in a basket trial approach where patient selection is determined by the intensity of the PU-FITC signal in the appropriate malignant cell populations. Being coupled with flow cytometry, the assay evaluates epichaperome levels at the single-cell level (Roboz et al. 2018). It successfully identified a patient predicted to respond to treatment with PU-H71. This patient, initially diagnosed with myeloproliferative neoplasm, had progressed to acute myeloid leukemia (AML) with fibrosis, and her disease was refractory to currently available therapies. The patient was sent for precision medicine evaluation but no somatic alterations in clinically relevant genes were found. However, a novel fusion protein, PML-SYK, as well as constitutive activation of SYK, STAT5, ERK kinases were identified. Elevated epichaperome levels were found in the cell populations bearing the translocation, supporting our previously reported data in AML demonstrating a relationship between a hyperactive signalosome and sensitivity to PU-H71 (Zong et al. 2015). Based on the poor prognosis, lack of effective therapies, and laboratory data suggesting sensitivity to PU-H71, the patient was granted compassionate access to this medication by the FDA. After 16 doses of PU-H71 at 300 mg/m² over 3 months, the patient has attained complete remission, with normalization of peripheral blood counts and <5% marrow blasts. Splenomegaly and all constitutional symptoms have completely resolved (Roboz et al. 2018). This poor prognosis AML patient who had relapsed after allogeneic stem-cell transplantation is presently in remission at 9 months into treatment, a response to PU-H71 therapy obtained and maintained with little to no toxicity to the patient.

Imaging modalities are also rapidly progressing for clinical applications (Mahajan et al. 2015). Among these, ultrasound and computed tomography (CT), provide anatomical information that is dependent on anatomical changes associated with the disease, whereas positron emission tomography (PET), single-photon emission computed tomography (SPECT), optical bioluminescence/fluorescence imaging, molecular magnetic resonance imaging, and magnetic resonance spectroscopy may provide functional information, as they track biochemical processes or molecules in vivo. PET and SPECT offer an additional advantage, and that is the possibility of quantitative information. PET, SPECT, and optical imaging require the injection of molecular probes in the tested subject to acquire the imaging signal. Therefore, an appropriate imaging probe must reach the target of interest in vivo, while also specifically being retained in the target to be detected (Chen and Chen 2010). For PU-H71, PET is optimal as it takes advantage of the iodine already present in PU-H71. This can be replaced by I-124, which is a long-lived PET tracer ideally suited to monitor the difference in off-rates between the distinct HSP90 pools, and therefore, to identify those tumors expressing the epichaperome (Rodina et al. 2016; Taldone et al. 2016; Joshi et al. 2018).

This chemical tool is now also used in the clinic to detect epichaperome positivity in the context of solid tumors (ClinicalTrials.gov identifier: NCT01269593). It was recently linked to an open label phase 1b study of PU-H71 with nab-paclitaxel in patients with HER2 negative metastatic breast cancer (NCT03166085). On this therapeutic study, patients received nab-paclitaxel at a standard dose of 260 mg/m² intravenously every 3 weeks. PU-H71 was administered intravenously 6 h (±1 h) post-nab-paclitaxel in two escalating dose levels (225 and 300 mg/m²). The regimen was well tolerated with promising clinical activity in this heavily pretreated cohort (i.e., median lines of therapy in the metastatic setting of 6 [range 1–11] including prior taxanes in 75% of patients). Additionally, patients had the option to enroll in the separate diagnostic PU-PET protocol (NCT01269593) to measure epichaperome expression prior to initiating treatment on the phase 1b study, wherein they received a single microdose of ¹²⁴I-PU-H71 intravenously and underwent imaging for epichaperome expression. Eight out of the 12 patients enrolled in the phase 1b study participated in the PU-PET imaging. Even in this small patient group, and in accord with preclinical findings (Rodina et al. 2016), a higher tumor epichaperome, measured by a longer retention of ¹²⁴I-PU-H71 (i.e., slower k_off; Fig. 1), correlated with a longer duration of response (Jhaveri et al. 2019). With tumor epichaperome expression at the baseline having the potential to serve as a predictive biomarker of response, the phase 2 trial of this combination is currently planned along with baseline PU-PET.

TAKING ADVANTAGE OF KINETIC SELECTIVITY FOR PROTEOMICS

Considering the central role of the chaperome in the context of large proteome networks, several proteomics approaches were developed and implemented to dissect both the chaperome and its interactome, as well as the topology of such chaperome–proteome networks. Excellent articles are written on the topic (Voisine et al. 2010; Taipale et al. 2014; Freilich et al. 2018; Rizzolo and Houry 2018), and we limit our analysis to the use of chemical probes in the development of such proteomic methods. An advantage in the use of chemical probes is that they enable the identification and study of endogenous proteomes native to a specific disease state. As mentioned above, this is because it is the probe, and not the cell, that requires engineering for this method to work. Here a solid-support immobilized probe is typically incubated with a cell homogenate, and upon pull-down and washing of the proteins isolated on the solid support, this mixture is applied to sodium dodecyl sulfate (SDS)-PAGE for protein mixture separation, and in-gel digestion, or is directly digested on the solid support. The ensuing peptide mix is analyzed by mass spectrometry for protein identification. Alternatively, the chemical probe is a biotin-linked small molecule, and if this probe has the advantage of being cell permeable, it is a powerful tool to more closely capture the native interactome.

Although initial attempts have employed a variety of antibodies to capture the HSP90 interactome in cancer cells, these studies had little success as there are several limitations to these probes (Hartson and Matts 2012; Weidenauer et al. 2017). The first attempt to use a chemical probe was by using biotinylated GM and streptavidin beads to capture HSP90 and its binders (Tsaytler et al. 2009). The pull-down efficiency was, however, low and, in turn, the identified interactome, was limited to abundant proteins.

We took advantage of the properties of PU-H71 to design an unbiased proteomics approach to investigate HSP90 networks in cancer (Darby and Workman 2011; Moulick et al. 2011). It employs a solid-support immobilized PU-H71, either by direct chemical linking to agarose beads or indirect linking via biotin-streptavidin (Fig. 4A). Whereas the creation of such tools is often deemed a simple endeavor, it is not—the nature of the linker, its length, site of attachment, all must be carefully considered, and experimentally investigated, until a probe with optimal features is developed (Taldone et al. 2013b). We took a similar approach for the development of chemical tools to investigate the interactome of HSP70 in cancer (Rodina et al. 2014). An important factor in the use of such probes is understanding how they work. An affinity probe is not just a chemical, it is a chemical attached to a resin. The quality of the cell homogenate used in the assay, the incubation buffer and the time of incubation, and the washing conditions, all influence whether a real interactome, or nonspecific material, is isolated onto the resin. As mentioned above, these characteristics also greatly influence the selectivity of the probe for an HSP90 pool over others.

Figure 4.

Chemical probes and methods for unbiased large-scale investigation of the epichaperome and its interactome. (A) For unbiased analyses, a solid support immobilized PU-H71 (PU-beads) is coupled with an affinity-purification method, followed by mass spectrometry identification of the interactome. Subsequent computational methods are used to dissect the topology and function of the epichaperome networks. (B,C) Panel B shows the influence of oxyanions and mild detergents on heat shock protein 90 (HSP90) complexation in nontransformed cells as compared to HSP90 in epichaperome-expressing cancer cells. Panel C shows how the nature of HSP90 networks (from hyperconnected, 1 to insular, 3) can be detected and characterized by affinity-purification chemical probes. These are representative examples and experiments were performed as previously described (Rodina et al. 2016). SDS-PAGE, sodium dodecyl sulfate-polyacrylamide gel electrophoresis.

Over the last 7 years, several studies have employed this method to investigate tumor-specific HSP90 interactomes, and to dissect mechanisms associated with HSP90 and HSP70 in cancer (Moulick et al. 2011; Nayar et al. 2013; Goldstein et al. 2015; Zong et al. 2015; Culjkovic-Kraljacic et al. 2016; Rodina et al. 2016; Guo et al. 2017). In such affinity purification experiments, the experimental design will determine the HSP90 pools that are enriched onto the probe. Because the PU-H71-based chemical bait select for, enrich in, and, moreover, trap the pathologic chaperome-proteome complexes, these characteristics provide for robust identification of proteins, and both low copy proteins and weak interactions are detected.

The probe has a preference for the stable epichaperome complexes (Fig. 4B,C; Moulick et al. 2011; Rodina et al. 2016), and this feature can be used to investigate the topology of epichaperome networks and their function in cancer. It can be used to address the components of the epichaperome and its proteome in a context-specific manner (Rodina et al. 2016; Kishinevsky et al. 2018). The addition of oxyanions into the assay buffer may be used to stabilize the dynamic chaperome complexes (Hutchison et al. 1992), and enable their capture onto the probe (Fig. 4B). This feature was used to compare the interactome of HSP90 in tumors sensitive and resistant to HSP90 inhibitors (Rodina et al. 2016). Recently, we demonstrated how this probe can also be used to determine proteome alterations in neurodegenerative diseases induced by either toxic or genetic stress, or a combination thereof (Kishinevsky et al. 2018). Therefore, this method is applicable for detecting pathologic functional changes in the proteome induced by stresses, such as disease-related stresses (malignant and neurodegenerative disease-related).

CONCLUSIONS AND FUTURE DIRECTIONS

Here we have provided snapshots into how chemical biology has facilitated our understanding of HSP90 in cancer. Our focus here is narrow and mainly on pathologic changes because of alterations in the interaction strength and connectivity between chaperome members and between the chaperome and the proteome it regulates, because this is a topic often overlooked in the many review articles written on proteostasis in disease. Although we mostly provide information on HSP90 in cancer, many of the concepts described here likely apply to other chaperome members, as well as to other diseases.

One lesson from the chemical biology of the chaperome is the importance of the kinetics of binding. Even though the equilibrium-binding constant K_d incorporates both kinetic constants k_off and k_on, it does not provide any insight into the rate of ligand association and dissociation, which are important parameters under nonequilibrium conditions. Although parameters measured at equilibrium, namely affinity indicators, (i.e., K_d and IC₅₀/EC₅₀) have historically been the mainstay of drug discovery, and were assumed to be valid indicators of drug activity and moreover predictors of in vivo efficacy, they fall short in defining the properties of a safe and active anti-HSP90 agent. In this context, one may consider some HSP90 pools as the target and other pools of HSP90 as the off-targets. Our work on other chaperome members suggests this characteristic to be the norm, not the exception, for effective chaperome therapeutics (Taldone et al. 2014a). In the context of the human body, a system always in flux, the kinetics of the association (k_on) or dissociation (k_off) may provide selectivity, in other words, a preference for a specific protein pool. For instance, in vivo, a drug with a longer residence time in one HSP90 pool $(k_{\mathrm{off}}^1<k_{\mathrm{off}}^2)$ can kinetically select a given pool over another, even if the affinity for both is comparable at equilibrium $(K_{\mathrm{d}}^1=K_{\mathrm{d}}^2)$. Modulating the kinetic selectivity of inhibitors, therefore provides a means for the development of clinically safe and effective chaperome inhibitor agents. This strategy provides the blueprint for the discovery of HSP90 inhibitors for other diseases, and of agents that target hubs of epichaperome networks other than HSP90 for cancer.

In addition to inherent drug-target-binding properties, kinetic binding plays a significant role in decision-making for clinical dose and schedule determination. In the in vivo setting, where a drug is constantly being cleared, equilibrium in a manner observed when treating cells in a culture plate is never really achieved. The long retention of a drug in tumor tissue compared to plasma and normal tissues suggests that drug concentrations in tumors are less affected by inherent clearance mechanisms (i.e., metabolism, excretion) and are more dependent on the kinetic-binding properties of the agent. This has a tremendous impact on the way chaperome-targeted drugs could be developed, namely by moving away from plasma PK as a surrogate for tissue concentrations and toward determination of tumor PK. Optimizing the target-binding kinetics of HSP90 drugs can therefore have significant benefits, ranging from improved duration of action to enhanced efficacy to improved therapeutic index. For instance, it may be most desirable for epichaperome- targeting agents to display high C_max to ensure that the target (i.e., HSP90 incorporated into epichaperome networks) is maximally engaged, and a fast elimination to ensure minimal engagement at off-targets (i.e., HSP90 pools not participating in epichaperome formation).

Chemical biology has also highlighted the need to view the chaperome in disease as a target that is a protein–protein connectivity network, rather than a single protein acting on another protein. In this context, the epichaperome is an intrinsic hallmark of tumors that can be targeted through numerous nodes of the epichaperome networks. It may also be targeted through upstream epichaperome regulators (i.e., MYC and NOTCH). Inhibition of the epichaperome by PU-H71 is one of the many possibilities.

Last, it is chemical biology that has given us insights on how to select patients for HSP90 therapy. An important issue for the successful clinical development of HSP90 inhibitors is the ability to identify patients most amenable to inhibitor therapy and, therefore, select those most likely to benefit. In addition to helping these patients it would serve to spare those who would not likely benefit from unnecessary discomfort and cost in participating in such a trial. The HSP90 field has long struggled with an inability to pinpoint means for patient selection (Garcia-Carbonero et al. 2013; Jhaveri et al. 2016; Yuno et al. 2018). HSP90 expression is not indicative of response (Garcia-Carbonero et al. 2013); surrogate tissue measurements are certainly inappropriate considering the distinct nature of the target in tumors (Garcia-Carbonero et al. 2013), and monitoring of kinase level changes in tumor biopsies or HSP70 induction, although widely incorporated into clinical studies (Garcia-Carbonero et al. 2013; Jhaveri et al. 2016) are insufficient to predict sensitivity (Rodina et al. 2016; Joshi et al. 2018). By identifying the HSP90 pools incorporated into the epichaperome as the species portending vulnerability to HSP90 inhibition, and moreover providing noninvasive assays to detect such biochemically modified HSP90 pools, chemical biology opens the door for precision medicine (Roboz et al. 2018; Jhaveri et al. 2019).

Several fundamental questions remain unanswered regarding the epichaperome. Is this a consequence of unproductive chaperone binding to large multidomain clients or is it a toxic gain of function of the chaperome, as was recently postulated (Joshi et al. 2018)? What is the relationship between HSF-1 and the formation of epichaperome networks in disease? Are chaperones participating in epichaperome network formation adopting a conformation that reduces or blocks nucleotide binding and rebinding? Is such a characteristic influencing the slow dissociation rate observed for some inhibitors? Is inhibition of the epichaperome through nodes other than HSP90 a valid next approach to cancer? Do epichaperomes form in diseases other than cancer? Is kinetic selectivity of inhibitors opening the door for treatment in such diseases?

In summary, chemical biology has brought a new understanding of the chaperome in cancer. It helped us propose a blueprint for translation of inhibitors of hub chaperome members to the clinic based on the context-dependent vulnerability of tumors to chaperome networks. The proposed chaperome network essentiality (i.e., chaperome rewiring into epichaperomes) expands the existing concepts for therapeutic strategies to provide a framework for the discovery of cancer-specific vulnerabilities. The context-dependent nature of the chaperome essentiality can be exploited to develop more effective and more specific chaperome-targeted therapies and provides avenues for patient-tailored anticancer therapies.