Evolution of Hsp90 targeting: From stress biology to clinical translation

Abstract

Hsp90 inhibitors represent a decades-long experimental framework that has progressively uncovered how molecular chaperone systems are organized, regulated, and rewired in disease. Early natural products established that pharmacologic engagement of Hsp90 simultaneously destabilizes broad client networks and exposes a central layer of proteostasis control. Subsequent structural, biochemical, and translational studies revealed the mechanistic importance of nucleotide-driven conformational cycling, cochaperone exchange, paralog specialization, and subcellular compartmentalization, as well as the emergence of disease-specific chaperone assemblies such as the epichaperome. In parallel, adaptive responses to Hsp90 inhibition, including heat-shock factor 1 activation, unfolded-protein-response signaling, autophagy induction, and oncogenic network rewiring, illuminated the resilience built into proteostasis architecture and the conditions under which it can be overcome. Later inhibitor classes, encompassing C-terminal allosteric ligands, middle-domain modulators, isoform-selective agents, epichaperome-directed compounds, and targeted degraders, extended the field from pan-inhibition toward increasingly precise intervention in specific chaperone states. The first regulatory approval of an Hsp90 inhibitor, Pimitespib, for refractory gastrointestinal stromal tumor, marks a translational milestone that validates this framework clinically. Combination studies have further mapped where Hsp90 inhibition is most informative and most effective, demonstrating that benefit is strongest when deployment is guided by defined client dependency, proteostasis burden, immune context, or biomarker selection. Together, these advances position Hsp90 inhibitor research as a major source of mechanistic insight into molecular chaperones and as a foundation for biomarker-guided, context-aware targeting of proteostasis in oncology and beyond.

Background

Over the past several decades, research on Hsp90 inhibitors has offered fundamental insight into the organization and regulation of proteostasis. Beginning with the discovery of geldanamycin as an Hsp90 inhibitor¹ and subsequent structural elucidation of the N-terminal ATP-binding pocket,2, 3 the field revealed a remarkably dynamic chaperone system characterized by coordinated conformational cycling, cochaperone control,⁴ paralog specialization,⁵ and context-dependent assembly into higher-order network states, including the epichaperome.6, 7 As chemically distinct inhibitor classes emerged, so too did deeper biological understanding, revealing how cells maintain signaling competence, buffer proteotoxic stress, and reprogram survival pathways under selective pressure.8, 9 Hsp90 inhibitors, therefore, came to represent more than a therapeutic strategy, they provided a conceptual framework for understanding how molecular chaperones govern cellular plasticity across diverse disease states.10, 11 This review traces that trajectory and emphasizes how efforts to inhibit Hsp90 helped establish the principles that now guide selective, mechanism-informed targeting of proteostasis, with early signs of clinical translation alongside a far more nuanced view of chaperone biology (Box 1).

Box 1. Key Concepts in Hsp90 Biology.

• •Molecular Chaperone Hub: A central regulator that can account for ∼1–3% of total cellular protein and supports the folding, stability, and functional maturation of a large client repertoire.
• •Client Dependency States: Cells under oncogenic, proteotoxic, inflammatory, or degenerative stress can become disproportionately dependent on Hsp90-supported folding networks.
• •The ATPase "Molecular Switch": Hsp90 functions as a homodimer in which ATP binding promotes a closed, client-processing state and hydrolysis enables reopening and client release.
• •Disease-associated Chaperone Assemblies: The epichaperome is a stable, highly integrated network of Hsp90, Hsp70, and cochaperones that emerge in selected pathological states and create distinct vulnerabilities.
• •Co-chaperone Rheostats: Regulatory proteins such as Aha1, p23, and Cdc37 tune cycle speed, client selection, and the balance between folding, stabilization, and degradation.

Historical background

The history of Hsp90 research matters not only because it charts the rise of a therapeutic target, but because it shows how pharmacologic interest in a stress protein opened a much broader view of cellular regulation. What began as a heat-shock phenomenon gradually became a framework for understanding how metastable signaling proteins are folded, protected, and functionally controlled (Figure 1). The early history of the field, therefore, marks the point at which proteostasis started to emerge as an organized and druggable system, with Hsp90 at its center.

Fig. 1.

Milestones in Hsp90 discovery, biology, and therapeutic translation. This figure was created using BioRender.

Discovery and early characterization of a stress-induced protein

The foundations of the field were laid in 1962, when Ritossa reported that heat exposure induces discrete chromosomal “puffs” in Drosophila melanogaster, reflecting transcriptional activation of stress-responsive loci.¹² Twelve years later, Tissières and colleagues demonstrated that these puffs correspond to the selective synthesis of a defined set of proteins, thereby establishing the heat-shock response as a regulated proteomic program rather than a consequence of protein damage.¹³ Among the newly synthesized species, a prominent polypeptide migrating near 84–90 kDa was consistently observed,14, 15 although its function remained unclear.

The molecular identity of this protein began to crystallize in the early 1980s. Oppermann, Levinson, and Bishop discovered that an abundant 89 kDa heat-inducible cellular protein associates with the transforming kinase pp60v-src, providing the first indication that a heat-shock protein participates directly in oncogenic signaling complexes.16, 17 Shortly thereafter, Finkelstein and Strausberg cloned the yeast gene encoding the 90 kDa heat-shock protein and introduced the standardized nomenclature “Hsp90,” establishing it as a conserved and genetically defined member of the HSP family.¹⁸

From client interactions to essential chaperone

A key conceptual advance in the field followed the discovery that Hsp90 forms stable heterocomplexes with steroid hormone receptors. Independent studies demonstrated that the same 90 kDa protein associates with glucocorticoid, progesterone, estrogen, and androgen receptors, as well as the transforming kinase v-Src.19, 20, 21, 22 These findings revealed that Hsp90 is constitutively abundant and specifically recruited into regulatory protein complexes to maintain receptors in a ligand-competent yet transcriptionally inactive conformation.

Genetic evidence reinforced the central importance of this chaperone, demonstrating that Hsp90 is essential for eukaryotic viability and that reducing its levels in vivo directly compromises steroid receptor function.23, 24 During this period, the work of Lindquist and Craig reframed Hsp90 from an enigmatic stress protein into an ATP-dependent molecular chaperone.25, 26, 27, 28 This was supported by direct biochemical proof showing that Hsp90 prevents the aggregation of non-native substrates and facilitates productive refolding in reticulocyte lysate systems.29, 30 These milestones established Hsp90 as a core component of cellular homeostasis that stabilizes metastable signaling proteins.

The most profound expansion of Hsp90’s biological role came with the identification of its function as a "molecular capacitor" for evolution. This concept describes Hsp90’s ability to mask the phenotypic consequences of genetic variation by using its folding capacity to stabilize mutant or unstable proteins.10, 31 By buffering these variants, Hsp90 allows a reservoir of "cryptic mutations" to accumulate silently within a genome, hidden from natural selection as long as the chaperone system is functioning at full capacity.32, 33, 34 This capacitor role has critical implications for cancer therapy and drug resistance. When Hsp90 is pharmacologically inhibited, the capacitor is effectively "discharged," suddenly revealing these previously hidden genetic variations.35, 36 In the hostile environment of a tumor, this "unmasking" can accelerate the emergence of resistant clones that possess the phenotypic traits needed to bypass therapy. Consequently, Hsp90 is now recognized as a potent "enabler of cancer adaptation," whose activity not only supports current oncogenic drivers but also facilitates future evolutionary escape.³⁷

Emergence of Hsp90 as a therapeutic target

The therapeutic significance of this biology became evident in 1994 through the pioneering work of Neckers, Whitesell, and colleagues. Using a geldanamycin affinity probe, Neckers lab identified Hsp90 as the direct cellular target of this natural product and showed that inhibitor binding disrupts Hsp90 complexes with pp60v-src, leading to degradation of oncogenic kinases and reversal of transformed phenotypes.¹ This study provided the first pharmacological demonstration that cancer cells depend on Hsp90-mediated stabilization of oncogenic clients, thereby validating Hsp90 as a druggable target. Structural studies soon supplied a mechanistic framework. Crystal structures of the Hsp90 N-terminal domain, including complexes with geldanamycin, revealed an ATP-binding pocket that accommodates the inhibitor and explained how drug binding arrests the chaperone's conformational cycle.2, 3 These findings defined the canonical mode of Hsp90 inhibition and laid the groundwork for rational drug design.

Targeting Hsp90 for cancer therapy initially appeared paradoxical because Hsp90 is essential for normal proteostasis, yet cancer cells show a greater functional dependence on this chaperone system. This selectivity is partly explained by increased Hsp90 abundance, as intracellular Hsp90 levels can be 2- to 10-fold higher in cancer cells than in normal cells.³⁸ However, tumor selectivity is not driven by expression alone. Kamal and colleagues showed that tumor-derived Hsp90 is enriched in activated, high-affinity multichaperone complexes that bind 17-AAG more strongly than Hsp90 from normal cells, linking chaperone-complex state to drug sensitivity.³⁹ Pharmacological studies further supported this concept by showing that Hsp90 inhibitors can display strong cellular activity despite modest affinity for purified Hsp90, while in vivo studies with PU24FCl demonstrated preferential tumor accumulation and rapid clearance from normal tissues.40, 41 Later affinity-based proteomics showed that PU-H71 preferentially engages a tumor-enriched Hsp90 pool associated with oncogenic client networks, reinforcing the idea that tumor retention reflects the biochemical state of Hsp90 as well as drug distribution.⁴² Together, these findings indicate that Hsp90 inhibitor selectivity arises from multiple converging mechanisms, including elevated Hsp90 expression, enrichment of activated chaperone complexes, preferential intratumoral accumulation, prolonged tumor retention, and cancer-cell dependence on Hsp90-stabilized oncogenic clients such as HER2, mutant BRAF, mutant EGFR, and ALK fusion proteins.37, 43, 44, 45, 46, 47, 48, 49 This vulnerability reflects both oncogene addiction and non-oncogene addiction, in which malignant cells rely on stress-buffering proteostasis pathways to survive proteotoxic, metabolic, and microenvironmental stress.50, 51 Clinically, this relevance is supported by a 10-year Yale breast cancer cohort study showing that high Hsp90 expression was associated with adverse clinicopathologic features and decreased survival.⁵²

Mechanistic and structural insights from Hsp90 targeting

Once Hsp90 became pharmacologically druggable, structural and biochemical studies clarified the basis of its unusually pleiotropic biological effects. These analyses revealed that Hsp90 functions as a highly dynamic chaperone whose activity is governed by ATP-dependent conformational cycling, regulated client exchange, and coordinated cochaperone interactions (Figure 2). This mechanistic framework continues to underpin contemporary strategies for designing Hsp90-targeted therapeutics.

Fig. 2.

Overview of the Hsp90 ATP-dependent chaperone cycle and targeting strategies. This figure was created using BioRender. The 3D structure of full-length Hsp90α (PDB ID: 7KRJ) was refined using SWISS-MODEL and visualized in PyMOL.

Hsp90 architecture and the ATPase cycle

Hsp90 is a highly conserved and abundant molecular chaperone that functions as a homodimer, with each protomer comprising three structurally and functionally distinct domains: an N-terminal domain (NTD) containing the ATP/ADP binding site, a middle domain (MD) that provides docking sites for client proteins and cochaperones, and a C-terminal domain (CTD) that mediates dimerization and contains a second drug binding region⁵³ (Figure 2). In mammalian cells, Hsp90 exists as a family of four paralogues: the cytosolic isoforms Hsp90α (encoded by HSP90AA) stress-inducible and Hsp90β (HSP90AB1) constitutively expressed, GRP94 (HSP90B1) in the endoplasmic reticulum and TRAP1 in the mitochondria.¹¹ While these isoforms share high sequence similarity and adopt common conformational states, they exhibit distinct tissue distributions and functional specializations that have important implications for therapeutic targeting.⁵⁴

The Hsp90 ATPase cycle drives client protein maturation through a series of conformational transitions. ATP binding to the open V-shaped dimer induces NTD dimerization (closed I state), followed by NTD–MD association (closed II state), generating an ATPase-competent conformation.⁵⁵ ATP hydrolysis triggers reopening and client release (Figure 2). Notably, human Hsp90 displays greater conformational flexibility and substantially lower intrinsic ATPase turnover than yeast Hsp90, with typical reported ranges of human ∼0.05-0.1 ATP·min⁻¹ vs yeast ≈0.8-1.3 ATP·min⁻¹.⁵⁶ The cellular environment profoundly modulates this cycle, as molecular crowding favors the closed active state, while the physiological ATP:ADP ratio may favor an ATPase-incompetent heterodimer as the predominant conformer under normal conditions.57, 58, 59

Regulatory layers governing Hsp90 client processing

Hsp90 recognizes and stabilizes a remarkably diverse array of client proteins, with hundreds cataloged in the Hsp90 molecular chaperone complex database, including protein kinases, steroid hormone receptors, transcription factors, and ubiquitin ligases58, 60 (Hsp90Int.db). Hsp90 acts late in the folding process; it often receives clients from Hsp70, which interacts with early folding intermediates.⁶¹ This handover is orchestrated by a sophisticated network of cochaperones that has expanded dramatically during evolution, from approximately 14 in yeast to over 40 in humans, reflecting the increased complexity of cellular proteostasis demands in higher eukaryotes.⁶²

The stoichiometry of the Hsp90 system is notable. Proteomic surveys indicate that the chaperome represents approximately 10% of the total protein mass in many human cell lines, with Hsp90 and Hsp70 contributing a large fraction of that pool. Depending on the cell type and measurement method, Hsp90 alone can account for an order of 1%-3% of total cellular protein.63, 64 Cochaperones are generally expressed at substantially lower abundance, and reported cellular molar ratios between Hsp90 and individual cochaperones vary widely; for example, proteomic data report ratios on the order of ∼1:7 for p23 up to ∼1:300 for some peptidyl-prolyl isomerases.62, 63 These abundance patterns are consistent with a model in which many cochaperones associate with Hsp90 transiently and dynamically, while others form more stable, client-specific subcomplexes. Such substoichiometric and dynamic interactions enable competition and exchange among cochaperones, adding a regulatory layer that helps fine-tune client processing.⁵⁸

Cochaperones regulate every stage of the Hsp90 cycle. Hop (Stip1 in yeast) facilitates client transfer from Hsp70 by simultaneously binding both chaperones via its TPR domains, stabilizing a semi-closed, client-binding competent state.65, 66 Cdc37 serves as a kinase-specific cochaperone, recognizing kinase clients through their catalytic domains and delivering them to Hsp90.⁶⁷ Aha1 potently accelerates ATPase activity (15- to 30-fold),⁶⁸ promoting cycle progression, while p23 stabilizes the closed ATP-bound state, prolonging client residence time.⁶⁹ Large peptidyl-prolyl isomerases such as FKBP51, FKBP52, and Cyp40 engage late in the cycle, facilitating client maturation and nuclear trafficking.⁷⁰

Hsp90 function is further governed by an extensive network of post-translational modifications (PTMs) referred to as the chaperone code, which enables the same chaperone machinery to generate distinct biochemical outputs depending on cellular state. These PTMs include phosphorylation, acetylation, SUMOylation, methylation, ubiquitination, S-nitrosylation, oxidation, and O-GlcNAcylation, each influencing core aspects of Hsp90 biology such as ATPase activity, conformational cycling, co-chaperone engagement, client selection, subcellular trafficking, proteostasis, and drug sensitivity.71, 72, 73 Phosphorylation remains the most extensively characterized layer of regulation. CK2-dependent modification of Hsp90α at Thr36, corresponding to Thr22 in yeast, weakens interactions with co-chaperones such as Cdc37 and Aha1, whereas tyrosine phosphorylation at Tyr197, Tyr313, and Tyr627 orchestrates sequential transitions through the chaperone cycle by modulating Cdc37 release, Aha1 recruitment, and late-stage client maturation.74, 75 Additional phosphorylation events also influence pharmacological response, as shown by Mps1-mediated Hsp90 phosphorylation, which contributes to renal cell carcinoma sensitivity and selectivity toward Hsp90 inhibitors.⁷⁶ Acetylation acts as a second major regulatory axis, particularly through HDAC6-mediated deacetylation; hyperacetylation, including at the Lys294 hotspot, reduces chaperone competence, destabilizes oncogenic clients, and enhances susceptibility to NTD Hsp90 inhibitors.77, 78 Further PTMs refine chaperone output and therapeutic response: asymmetric SUMOylation of Hsp90α at Lys191 promotes Aha1 recruitment and increases sensitivity to ATP-competitive inhibitors;⁷⁹ methylation near the C-terminal dimerization region modulates long-range conformational communication within the Hsp90 dimer;80, 81 ubiquitination contributes to Hsp90 turnover and quality control;⁷³ and redox-sensitive modifications, including S-nitrosylation at Cys597 and oxidative modifications, suppress ATPase activity under stress.⁸² Through this multilayered regulation, Hsp90 acts as a context-sensitive hub that directs client maturation, stabilization, remodeling, or degradation.

Structural basis for therapeutic intervention

The deep mechanistic understanding of Hsp90 biology has revealed multiple strategies for therapeutic intervention beyond ATP-competitive inhibition. The NTD ATP-binding pocket, first characterized through geldanamycin co-crystallization, remains the most extensively exploited target.2, 3 This pocket is highly conserved yet structurally distinct from other cellular ATPases, providing a basis for selective inhibition. NTD inhibitors such as geldanamycin derivatives and synthetic compounds trap Hsp90 in specific conformational states, disrupting the nucleotide-dependent dynamics essential for client stabilization and leading to client protein misfolding and proteasomal degradation.⁸³

However, the CTD also harbors a druggable site, as demonstrated by novobiocin and related compounds that disrupt Hsp90 dimerization, particularly affecting Hsp90β homodimers.84, 85 Allosteric CTD binders can modulate Hsp90 oligomerization states, offering distinct mechanisms of action.⁸⁶ Perhaps most significantly, understanding cochaperone interactions has opened new therapeutic avenues: disrupting the Hsp90–Cdc37 interface can selectively destabilize kinase clients, while targeting the Hsp90–Hop interaction affects client loading.87, 88 The recognition that Hsp90 forms distinct structural assemblies in disease states, including the pathological hetero-oligomeric epichaperomes found in cancer and neurodegeneration, has further refined targeting strategies, enabling discrimination between disease-associated and physiological Hsp90 pools.6, 42 Together, these mechanistic studies expanded the therapeutic map of Hsp90 far beyond the classical ATP pocket and defined the chaperone as a modular system whose pharmacology can be tuned with increasing precision.

First-generation N-terminal inhibitors

Natural products were pathfinder molecules that turned Hsp90 inhibition into both a therapeutic concept and a biological probe. They established that pharmacologic engagement of the chaperone can simultaneously destabilize multiple clients, and they provided the structural logic that informed every subsequent wave of inhibitor design. A summary of representative Hsp90 inhibitors arranged by binding site is provided in Table 1.

Table 1.

Preclinical and emerging Hsp90 inhibitors.

Compound	Class / Scaffold	Binding site / selectivity	Cancer type / cell line	Key mechanistic findings	Ref.
NTD inhibitors
Geldanamycin	Ansamycin (natural product)	Pan-Hsp90 NTD ATP pocket	v-Src transformation model	•Disrupts the Hsp90–pp60v-src heteroprotein complex required for v-Src transformation. •Reverses transformed phenotype and establishes Src as an Hsp90-dependent client.	1
17-AAG (Tanespimycin)	Ansamycin (semi-synthetic)	Pan-Hsp90 NTD ATP pocket	HER2+ breast cancer; solid tumors	•Depletes ErbB2/HER2, Raf-1 and mutant p53 after NTD engagement. •Revealed tumor-selective activated Hsp90 complexes and NQO1-linked response biology.	71, 72
17-DMAG (Alvespimycin)	Ansamycin (semi-synthetic)	Pan-Hsp90 NTD ATP pocket	Melanoma; breast cancer (MDA-MB-231 xenograft)	•Preserves canonical ansamycin binding with improved aqueous handling / oral exposure versus 17-AAG. •Decreases Raf-1, Akt and ErbB2 and induces Hsp70 in tumor models.	81, 82
IPI-504 (Retaspimycin)	Ansamycin hydroquinone	Pan-Hsp90 NTD ATP pocket	Multiple cancers	•Hydroquinone form of 17-AAG designed to control quinone/hydroquinone redox behavior. •Maintains Hsp90/GRP94 engagement and shows stronger apparent binding under reducing conditions.	84, 85
Radicicol	Resorcylic acid lactone	Pan-Hsp90 NTD ATP pocket	Multiple cancer cell lines	•Occupies the same NTD ATP pocket as ATP and geldanamycin, blocking ATP hydrolysis. •Reported pan-Hsp90 activity includes Kd ≈ 19 nM, but in vivo development was limited by instability.	66, 70
BIIB021	Purine scaffold (fully synthetic)	Pan-Hsp90 NTD ATP pocket	Hodgkin lymphoma	•Produces low-nanomolar growth inhibition in Hodgkin lymphoma and suppresses constitutive NF-κB signaling. •Upregulates NKG2D ligands and delays xenograft growth, sensitizing tumor cells to NK-cell killing.	94
SNX-2112 / SNX-5422	Purine scaffold / oral prodrug pair	Pan-Hsp90 NTD ATP pocket	Multiple myeloma; hematologic tumors	•SNX-2112 is a potent NTD inhibitor (reported IC50 ≈ 3.0 nM); SNX-5422 is its oral prodrug. •Abrogates Akt/ERK signaling and suppresses tumor growth, angiogenesis and osteoclastogenesis in myeloma.	95
KW-2478	Non-ansamycin synthetic inhibitor	Pan-Hsp90 NTD ATP pocket	Multiple myeloma; B-cell malignancies	•Novel Hsp90 inhibitor with reported Hsp90 IC50 ≈ 3.8 nM. •Depletes FGFR3 and downstream phospho-ERK and shows bortezomib synergy in myeloma models.	97, 98
Ganetespib (STA-9090)	Resorcinol / triazolone	Pan-Hsp90 NTD ATP pocket	NSCLC, including ALK-driven disease	•Rapidly induces degradation of Hsp90 client proteins and is more potent than 17-AAG in many models. •Suppresses ALK-dependent signaling and HIF-1α-driven programs in lung/TNBC models.	91, 186
Luminespib (AUY922)	Isoxazole-resorcinol	Pan-Hsp90 NTD ATP pocket	Breast cancer (BT474 spheroids); solid tumors	•Potent isoxazole-resorcinol inhibitor with strong activity in BT474 breast-cancer spheroids and xenografts. •Inhibits proliferation and invasion/angiogenesis programs; prolonged action supported FLT-PET monitoring.	92
Onalespib (AT13387)	Resorcinol (fragment-derived)	Pan-Hsp90 NTD ATP pocket	TNBC; solid tumors	•Fragment-derived inhibitor built from a 2,4-dihydroxybenzamide lead to optimize pocket fit. •Advanced because of durable Hsp90 engagement and favorable pharmacologic behavior in tumor models.	101
Pimitespib (TAS-116)	Synthetic pyrazolo-triazine	NTD; Hsp90α/β-selective	GIST; tumor immunology models	•Selectively inhibits cytosolic Hsp90α/β while sparing GRP94 and TRAP1. •Suppresses KIT signaling in imatinib-naïve and resistant GIST and later shown to degrade STAT5 in FOXP3high Tregs.	104, 105, 106
CTD Inhibitors
Novobiocin	Coumarin antibiotic	Hsp90 CTD ATP site / allosteric CTD ligand	Breast cancer (MCF-7)	•Binds the Hsp90 CTD (Glu537–Asn686 region) and allosterically prevents NTD ATP binding. •Depletes Hsp90-dependent signaling proteins in MCF-7 cells (reported IC50 ≈ 256 μM) without canonical HSR.	60, 131
KU-32	Novobiocin analog	CTD allosteric modulator	Non-oncology comparator (peripheral neuropathy)	•CTD allosteric modulator that stimulates, rather than inhibits, Hsp90 chaperone function. •Uncouples cytoprotective from cytotoxic outcomes and is best viewed as a mechanistic comparator, not an anticancer lead.	138
Middle-Domain / Co-Chaperone Disruptors
Gedunin	Limonoid terpenoid	p23-directed / indirect Hsp90 modulation	Multiple cancer cell lines	•Directly inactivates the cochaperone p23 instead of binding the Hsp90 ATP pocket. •Destabilizes p23-dependent Hsp90 function and triggers apoptotic cancer-cell death.	140
Celastrol	Triterpenoid	Hsp90–CDC37 interface	Pancreatic cancer cells; monocytic leukemia (U937)	•Disrupts the Hsp90–CDC37 complex in a thiol-dependent, ATP-noncompetitive manner. •Downregulates kinase clients AKT and CDK; reported antiproliferative activity ≈ IC50 3.0 μM in pancreatic cancer cells.	63, 141, 142
Sansalvamide A and derivatives	Cyclic depsipeptide series	Hsp90 middle domain	Melanoma (B16)	•Bind the N-terminal segment of the Hsp90 middle domain and allosterically block C-terminal cochaperone interactions. •H-10, LY-15 and H-15 inhibit B16 melanoma proliferation and induce apoptosis or differentiation.	144, 145, 146, 147
Withaferin A (WFA)	Withanolide steroidal lactone	CTD / Hsp90–CDC37 interface	Pancreatic cancer	•Preferentially engages the Hsp90 C-terminal region and disrupts Hsp90–CDC37 complexes independently of ATP competition. •Drives proteasome-dependent degradation of Hsp90 client proteins in pancreatic cancer cells.	148, 149
Gambogic acid (GBA)	Xanthonoid	Hsp90β middle-domain pocket	Multiple cancer cell lines	•Binds an Hsp90β middle-domain pocket with >3-fold selectivity over Hsp90α. •Also acts as an electrophile, inducing a thiol-dependent heat-shock response and disrupting Hsp90–HSF1/HSF2 complexes.	150, 151
Diptoindonesin G (Dip G)	Resveratrol oligomer	Hsp90 middle domain	Endocrine-resistant breast cancer	•Middle-domain binder with reported Kd ≈ 0.13 μM. •Promotes client degradation without strong HSR and retains activity in ER-mutant endocrine-resistant breast-cancer cells.	152
SOMCL-16-171/175	Synthetic small molecules	Hsp90 middle domain (allosteric)	Multiple cancer cell lines	•First designed synthetic Hsp90 middle-domain binders in this series. •Allosterically remodel Hsp90 and destabilize clients without canonical ATP-site liabilities.	153
Enniatin A (EnnA)	Cyclic depsipeptide	Hsp90 middle domain	Triple-negative breast cancer (TNBC)	•Dissociates Hsp90 from PD-L1 and drives PD-L1 proteasomal loss without a robust HSR. •Triggers immunogenic cell death and enhances CD8+ antitumor immunity in TNBC.	154
Isoform-Selective / Organelle-Selective Inhibitors
KUNB31	NTD scaffold	Hsp90β-selective	Isoform-selectivity / multiple cancer cell lines	•Hsp90β-selective NTD inhibitor that engages Asp88 and Thr179. •Reported Hsp90β Kd ≈ 0.18 μM versus 8.5 μM (GRP94) and 9.6 μM (Hsp90α).	157
NDNB1182	NTD scaffold	Hsp90β-selective	Prostate cancer; breast cancer	•Hsp90β-selective inhibition degrades CDK4 and lowers DNMT1. •Reactivates ERV/interferon signaling and enhances immune-checkpoint blockade in murine tumors.	160
Corylin	Natural flavonoid	Hsp90β-selective	Liver cell lines; primary hepatocytes	•Selectively binds the Hsp90β middle domain and disrupts the Hsp90β–Akt axis. •Promotes ubiquitination/proteasomal degradation of mature SREBPs, lowering lipid burden in hepatocytes.	156
KUNA-115	AT13387-derived scaffold	Hsp90α-selective	Isoform-selectivity study	•Hsp90α-selective inhibitor with Hsp90α IC50 ≈ 0.84 μM and Hsp90β IC50 > 50 μM. •Uses a hydrogen-bond network involving Leu48, Asn51, Asp93 and Thr184 to achieve α bias.	162
KUNA-111	AT13387-derived scaffold	Hsp90α-selective	Isoform-selectivity study	•5-fluoroisoindoline derivative with a novel Hsp90α binding mode. •Retains ∼15-fold selectivity for Hsp90α while expanding α-selective medicinal chemistry.	163
PU-WS13	Purine scaffold	GRP94 Site-2 pocket / GRP94-selective	HER2-driven tumors; IGF-II/TLR-dependent models	•Accesses the GRP94 Site-2 pocket that is unavailable in cytosolic Hsp90α/β. •Defines GRP94-dependent HER2 regulation and blocks IGF-II secretion / TLR trafficking.	166
KUNG65 / ACO1 / PU-H39	Purine / aminocyclohexanol / benzamide	GRP94-selective	Metastatic cancer / GRP94 biology	•Orthogonal GRP94-selective chemotypes validate the ER paralog as a distinct drug target. •Reported values include PU-H39 IC50 ≈ 2.4 μM (GRP94) vs >500 μM (Hsp90α/β) and KUNG65 Kd ≈ 0.54 μM vs 39 μM (Hsp90α).	168, 169
Gamitrinibs	Mitochondria-targeted Hsp90 inhibitors	Mitochondrial Hsp90 / TRAP1 state	Multiple cancers	•Couple an Hsp90 inhibitory warhead to a mitochondria-targeting moiety. •Collapse mitochondrial integrity / permeability transition and trigger apoptosis, including in cytosolic-Hsp90-resistant cells.	129
SMTIN-P01	Structure-guided TRAP1 inhibitor	TRAP1 ATP pocket / mitochondrial targeting	TRAP1-directed tumor models	•Structure-guided inhibitor built from human TRAP1 crystal information. •Provides direct proof-of-concept for TRAP1-directed mitochondrial pharmacology.	170
Epichaperome-Directed Compounds
PU-H71	Purine scaffold	Epichaperome-selective Hsp90 conformation	Triple-negative breast cancer (TNBC)	•Preferentially binds Hsp90 within epichaperome assemblies rather than bulk cellular Hsp90. •Produces durable, including complete, TNBC regressions and maps tumor-specific chaperome networks.	174
Targeted Protein Degraders (PROTACs)
BP3	PROTAC	Hsp90-targeting degrader	Breast cancer	•First reported PROTAC in this class to degrade Hsp90 itself. •Potently suppresses breast-cancer cell growth and inhibits tumors in vivo.	179
X10g	PROTAC	Hsp90α-selective degrader	Breast cancer	•Selectively degrades Hsp90α while sparing Hsp90β. •Shows better in vivo antitumor efficacy with lower toxicity than nonselective Hsp90 degradation.	180
Geldanamycin-based PROTACs	PROTAC	Geldanamycin warhead + CRBN recruiter	Multiple cancer cell lines	•Re-purpose a classical geldanamycin Hsp90 warhead into cereblon-recruiting degraders. •Drive proteasomal degradation of both Hsp90α and Hsp90β.	181
lw13	PROTAC	Hsp90-targeting degrader	Cervical cancer	•Hsp90-targeting PROTAC that degrades Hsp90 at low concentrations. •Enhances cisplatin activity in cervical-cancer models.	182
HIM-PROTAC	PROTAC	Hsp90 interactome-mediated effector	Multiple cancer models	•Uses an Hsp90 ligand as an effector module to degrade the nonclient protein GPX4. •Triggers ferroptosis and extends Hsp90 pharmacology beyond direct Hsp90/client degradation.	183

Ansamycin derivatives

Ansamycin derivatives gave the Hsp90 inhibitor field its first clear biological and translational framework. These compounds showed that Hsp90 inhibition can trigger the coordinated loss of multiple oncogenic client proteins, identified the N-terminal nucleotide-binding pocket as the main site for inhibitor binding, and provided the earliest clinical evidence that this biology could be used therapeutically.⁸ At the same time, they revealed the main limits of the approach, including quinone-related toxicity, redox-dependent drug behavior, and induction of the heat-shock response that can reduce treatment response.89, 90

Geldanamycin established the target and the core mechanism. Whitesell and colleagues showed that benzoquinone ansamycins disrupt the Hsp90–pp60^v-src heteroprotein complex required for oncogenic transformation.¹ This study placed Hsp90 at the center of the drug response and established that the compound acts through the chaperone system rather than through direct inhibition of Src itself. Structural studies then showed that geldanamycin binds within the NTD ATP-binding pocket of Hsp90 and provided the first detailed view of the site that became the canonical binding region for this inhibitor class.³ Roe and colleagues later showed that geldanamycin and radicicol occupy the same binding pocket in the NTD and inhibit Hsp90 ATPase activity, bringing structurally distinct natural products into a shared mechanistic model.⁹¹ However, early preclinical pharmacology revealed the main translational limitation of geldanamycin. Although geldanamycin could reach plasma concentrations above those required for cellular activity, systemic exposure was short, and the liver emerged as the principal target of acute toxicity. In beagle dogs, doses of 2.0 and 4.2 mg/kg produced elevations in serum transaminases and other liver-function markers consistent with acute hepatic necrosis.⁹² These findings positioned geldanamycin as a powerful target-validation compound but a poor clinical candidate. As a result, medicinal chemistry shifted toward semi-synthetic 17-substituted analogs designed to retain Hsp90 inhibition while improving tolerability and drug-like properties.

Within a few years, semi-synthetic optimization moved the field toward more clinically tractable ansamycins. At this stage, medicinal chemistry was guided by cellular phenotypes, especially ErbB2/HER2 modulation,93, 94 together with the need to overcome the hepatotoxic and pharmacologic limitations of the parent compound. The Pfizer series reported by Schnur and colleagues systematically tested substitutions across the scaffold and helped define which positions could be modified without losing activity, creating an early structure–activity framework for ansamycin development.⁹⁵ This work supported the development of 17-AAG and later 17-DMAG, which preserved the Hsp90-targeting mechanism of geldanamycin while improving its translational potential.

17-AAG, 17-allylamino-17-demethoxygeldanamycin, (tanespimycin) carried this mechanism into translational oncology. It retained direct Hsp90 binding and reproduced the main downstream effects of geldanamycin, including depletion of important oncogenic clients such as ErbB2, Raf-1, and mutant p53.⁹⁶ Kamal and colleagues then proposed an influential explanation for tumor selectivity by showing that tumor-derived Hsp90 exists in activated multichaperone complexes with much higher affinity for 17-AAG than Hsp90 from normal tissues.³⁹ This shifted attention toward chaperone state, client load, and proteotoxic stress as determinants of response. Redox biology added another important layer. Kelland and colleagues linked NQO1 expression to 17-AAG sensitivity, and Guo et al. later showed that NQO1-dependent formation of the hydroquinone metabolite can affect effective Hsp90 inhibition.97, 98 Gooljarsingh et al further showed that geldanamycin behaves as a slow, tight-binding inhibitor, indicating that residence time and assay timing can strongly affect apparent potency.⁹⁹ The clinical development of 17-AAG established the first practical pharmacodynamic framework for Hsp90 inhibitors. Early phase I studies showed induction of Hsp70 and depletion of client proteins in peripheral blood cells and tumor samples, making these useful markers of target engagement in patients.¹⁰⁰ Disease context also proved important; In HER2-positive breast cancer, tanespimycin combined with trastuzumab showed activity in trastuzumab-refractory disease and became an early example of how Hsp90 inhibition can be effective when tumors depend on a highly sensitive Hsp90 client protein.93, 101 Hsp70 induction, however, is now understood more broadly than it was in these early clinical studies. It remains a useful marker of Hsp90 inhibition, but it also reflects activation of the HSF1-driven heat shock response, which helps cells survive proteotoxic stress.102, 103 Since elevated Hsp70 can suppress apoptosis upstream of mitochondria, its induction should be interpreted as both evidence of target engagement and activation of an adaptive resistance response.¹⁰⁴

17-DMAG extended the same mechanism into a more pharmaceutically useful analog and showed what problems remained. The co-crystal structure of 17-DMAG bound to human Hsp90α showed that a clinically oriented C17-substituted derivative could preserve the canonical ansamycin binding mode while accommodating more polar functionality.¹⁰⁵ Preclinical studies confirmed that 17-DMAG retained the expected biological effects of Hsp90 inhibition, including client depletion and Hsp70 induction,¹⁰⁶ while offering better pharmaceutical properties than 17-AAG.¹⁰⁷ Even so, this analog reinforced an important lesson: improved solubility and handling did not remove the narrow therapeutic window of quinone-containing Hsp90 inhibitors. Toxicology studies still documented gastrointestinal, bone marrow, and hepatic toxicity at higher exposure, showing that scaffold optimization alone would not fully solve the tolerability problem.¹⁰⁸

IPI-504 improved the field’s understanding of ansamycin pharmacology by making redox state part of the development strategy. Developed as the hydroquinone hydrochloride form of 17-AAG, it preserved the same core mechanism while improving formulation and delivery.¹⁰⁹ Under assay conditions designed to limit oxidative artifacts, IPI-504 showed stronger apparent binding to purified human Hsp90 than 17-AAG and also bound GRP94, highlighting both the importance of quinone–hydroquinone equilibrium and the broader Hsp90-family activity of the scaffold.¹¹⁰ This shifted the discussion away from simple potency ranking and toward a more realistic view in which redox chemistry, paralog engagement, and assay conditions all influence observed drug behavior. Clinically, IPI-504 showed the clearest activity in molecularly defined settings such as ALK-rearranged NSCLC, where tumor growth depends on a particularly Hsp90-sensitive driver.¹¹¹ Later studies in GIST and soft-tissue sarcoma reinforced the same point: the strongest clinical activity appears when Hsp90 inhibition is matched to a biologically defined dependency, but long-term development still depends on achieving adequate exposure without unacceptable toxicity.¹¹²

Radicicol and the resorcylic lactones

Radicicol is a macrocyclic resorcylic acid lactone originally isolated from Monosporium bonorden, it was identified as a potent inhibitor of Hsp90 that binds competitively to the NTD ATP-binding pocket.91, 96 Structural studies demonstrated that radicicol occupies the same nucleotide-binding cavity as ATP and geldanamycin, thereby blocking ATP hydrolysis and chaperone cycling.⁹¹ Co-crystal structures of the Hsp90 NTD revealed that the resorcinol moiety of radicicol forms critical hydrogen bonds with conserved residues lining the ATP pocket (for example, interactions with the conserved catalytic Asp93 and Thr184 in human Hsp90α).⁹¹ Unlike geldanamycin, radicicol lacks a benzoquinone ansamycin scaffold but achieves comparable low-nanomolar affinity through a distinct macrocyclic resorcylic acid framework, establishing that chemically diverse scaffolds can inhibit Hsp90 via a shared ATP-competitive mechanism.91, 96

Despite potent in vitro activity, radicicol proved chemically and metabolically unstable in vivo. Its electrophilic enone and epoxide functionalities render it susceptible to rapid metabolic degradation and chemical inactivation under physiological conditions, precluding clinical development.113, 114 Medicinal chemistry efforts, therefore, focused on stabilizing the resorcinol pharmacophore while eliminating reactive liabilities. This work led to the development of oxime derivatives and synthetic resorcinol-based inhibitors that retained high-affinity binding to the NTD pocket with improved pharmacokinetic properties.114, 115 Structural insights from radicicol–Hsp90 complexes directly informed the rational design of second-generation, fully synthetic resorcinol inhibitors, including ganetespib (STA-9090),¹¹⁶ luminespib (NVP-AUY922),¹¹⁷ and onalespib (AT13387),¹¹⁸ which preserve the key hydrogen-bonding network within the ATP-binding site while demonstrating superior metabolic stability and clinical advancement. Radicicol thus provided the foundational pharmacophore and structural blueprint for the development of modern small-molecule Hsp90 inhibitors.

Second-generation synthetic inhibitors

The second generation of Hsp90 inhibitors reflects the point at which medicinal chemistry became an instrument for refining both the molecule and the biological question. These agents were designed to improve exposure, tolerability, and pharmacologic control, but they also allowed the field to test how scaffold, binding kinetics, schedule, and isoform preference influence the consequences of chaperone perturbation.

Purine-scaffold inhibitors

This class of fully synthetic, orally bio-available inhibitors was designed to mimic the adenine core of ATP. BIIB021 was the first orally bioavailable, fully synthetic Hsp90 inhibitor to enter clinical trials (CNF2024), it also demonstrated nanomolar affinity and robust client depletion in many cancer types in preclinical models.¹¹⁹ SNX-5422, a prodrug of SNX-2112, showed broad preclinical activity, while clinical studies helped define the tolerability boundaries of this scaffold, including visual disturbances and QTc prolongation.120, 121 KW-2478, especially in combination with bortezomib for relapsed or refractory multiple myeloma, produced an objective response rate of ∼39% and median progression-free survival of ∼6.7 months, highlighting how Hsp90 inhibition can amplify proteasome-directed stress in a biologically primed setting.122, 123

Resorcinol derivatives

Resorcinol derivatives became the dominant second‑generation NTD class because they retained the radicicol pocket‑binding logic while avoiding quinone liabilities. Ganetespib exemplified this shift: preclinical studies showed broad client depletion, potent antitumor activity, and an improved preclinical safety profile relative to earlier inhibitors.¹¹⁶ Ganetespib’s activity in ALK‑driven non‑small‑cell lung cancer and in models of ALK inhibitor resistance reinforced the lesson that Hsp90 inhibition is most effective when matched to tumors driven by highly Hsp90‑dependent kinases or fusion oncoproteins.¹²⁴ However, the GALAXY‑2 phase III trial showed that ganetespib did not improve outcomes when added to docetaxel in an unselected advanced lung adenocarcinoma population, underscoring the need for biomarker‑guided deployment.¹²⁵ Luminespib (AUY922) provided consistent evidence of pharmacodynamic target engagement in early clinical testing but also highlighted a recurrent class liability: ocular adverse events and gastrointestinal toxicity limited chronic dosing and constrained program expansion.¹¹⁷ Onalespib (AT13387), discovered by fragment‑based design, emphasized pocket fit and pharmacologic behavior; subsequent clinical studies suggested that schedule and partner selection are central determinants of clinical performance for this class.118, 126

The clinical intensity of second-generation development is best illustrated by the trial volumes for specific agents (Figure 4(b)). Ganetespib has been the most heavily investigated molecule in the field with 25 total trials (15 monotherapy and 10 combinations), followed by luminespib with 17 trials. Notably, while luminespib shows a balanced distribution between mono and combination regimens, other agents like SNX-5422 have been evaluated exclusively as monotherapies in this dataset. Earlier natural product derivatives like 17-AAG and newer synthetic scaffolds like AT13387 maintain smaller, more balanced trial footprints, highlighting the shift in focus toward the resorcinol class as the primary translational engine.

Fig. 4.

Clinical trial landscape of Hsp90 inhibitors. (a) Summary of study count, status, phase, and regimen type across the Hsp90 inhibitor trial dataset. (b) Distribution of monotherapy and combination studies by individual Hsp90 inhibitor. (c) The most common drugs used in combination with Hsp90 inhibitors. (d) Disease-area distribution of Hsp90 trials and tumor-type composition within oncology. Data source: https://clinicaltrials.gov, the data analytics and visualization of this figure used Bricks.

Pimitespib (TAS-116): First-in-class approved Hsp90 inhibitor

Pimitespib (TAS-116) represents a major inflection point in Hsp90-targeted therapy. Japan approved it in 2022 for patients with gastrointestinal stromal tumors (GIST) no longer responding to imatinib, sunitinib, or regorafenib,127, 128 providing the first regulatory validation of the field. Pimitespib is an oral ATP-competitive inhibitor that selectively targets the cytosolic Hsp90α and Hsp90β isoforms, a profile expected to reduce some toxicities associated with earlier pan-inhibitors.¹²⁹ In the CHAPTER-GIST-301 phase III trial, pimitespib doubled progression-free survival versus placebo (2.8 vs. 1.4 months; HR 0.51),¹²⁷ consistent with degradation of KIT, which established the clinical benefit of the compound against heavily pretreated advanced GIST.¹³⁰ More recently, preclinical work has suggested additional immune-modulating effects through degradation of STAT5 and selective reduction of FOXP3+ regulatory T cells, findings that extend the significance of pimitespib beyond approval.¹³¹

Although pimitespib marks a genuine milestone for the field, the broader clinical landscape remains more cautionary than definitive. Across the ClinicalTrials.gov dataset (listed in Table 2), 88 Hsp90 inhibitor trials were identified, of which 55 were completed, whereas 32 were terminated, withdrawn, or remained of unknown status (Figure 4(a)). Pimitespib therefore shows that regulatory success is possible under selected conditions, but it does not alter the broader translational picture: Hsp90 inhibition has not yet achieved wide global clinical adoption, no Hsp90 inhibitor has received FDA approval, and pimitespib remains the only approved agent in this class, with approval currently limited to Japan for refractory GIST. Consistent with the historical direction of the field, the trial landscape is dominated by oncology, which accounted for the largest weighted share (84.67), far exceeding hematology (14.33), respiratory disease (∼7.83), and gastrointestinal disease (∼5.83). Within oncology, lung (∼21.5%), breast (∼19.3%), and lymphoma (∼11.1%) represented the largest subgroups, whereas melanoma, colorectal cancer, sarcoma, and other tumor types were each represented more sparsely (Figure 4(d)).

Table 2.

Hsp90 inhibitors in clinical trials.

HSP90 Inhibitor	Combination Drug	Disease	Phase	Study Status	Study start	Study completion	NCT Identifier
Alvespimycin (17-DMAG)	\	Lymphoma	Phase I	COMPLETED	2004-06	2010-12	NCT00088868
Alvespimycin (17-DMAG)	\	Metastatic solid tumors or tumors that cannot be removed by surgery	Phase I	UNKNOWN	2005-10	-	NCT00248521
Alvespimycin (17-DMAG)	\	B-cell Chronic Lymphocytic Leukemia	Phase I	TERMINATED	2010-05	2012-06	NCT01126502
Alvespimycin (17-DMAG)	Trastuzumab; Paclitaxel	Solid Tumor	Phase I	COMPLETED	2006-01	2009-08	NCT00803556
BIIB028	\	Advanced Solid Tumors	Phase I	COMPLETED	2008-06	2011-10	NCT00725933
Debio 0932 (CUDC-305)	\	Solid Tumor or Lymphoma	Phase I	COMPLETED	2010-04	2013-04	NCT01168752
Debio 0932 (CUDC-305)	\	Psoriasis Vulgaris	Phase I	UNKNOWN	2018-11		NCT03675542
Debio 0932 (CUDC-305)	Cisplatin; Pemetrexed; Gemcitabine; Docetaxel	Non-small Cell Lung Cancer	Phase I	TERMINATED	2012-08	2014-11	NCT01714037
Ganetespib (STA-9090)	\	Solid Tumors	Phase I	COMPLETED	2007-10	2011-10	NCT00687934
Ganetespib (STA-9090)	\	Solid Tumors	Phase I	UNKNOWN	2007-10	2014-04	NCT00688116
Ganetespib (STA-9090)	\	Leukemia	Phase I	COMPLETED	2009-03	2013-02	NCT00858572
Ganetespib (STA-9090)	\	Leukemia	Phase I	COMPLETED	2009-08	2011-08	NCT00964873
Ganetespib (STA-9090)	\	Colon Cancer; Rectal cancer	Phase II	COMPLETED	2010-04	2012-07	NCT01111838
Ganetespib (STA-9090)	\	Esophagogastric Cancer	Phase II	COMPLETED	2010-08	2017-04	NCT01167114
Ganetespib (STA-9090)	\	Small Cell Lung Cancer	Phase II	COMPLETED	2010-07	2016-11	NCT01173523
Ganetespib (STA-9090)	\	Ocular Melanoma	Phase II	COMPLETED	2010-09	2016-11	NCT01200238
Ganetespib (STA-9090)	\	Prostate cancer	Phase II	COMPLETED	2011-01	2014-07	NCT01270880
Ganetespib (STA-9090)	\	Melanoma	Phase II	TERMINATED	2011-09	2012-09	NCT01551693
Ganetespib (STA-9090)	\	Hepatocellular Carcinoma	Phase I	COMPLETED	2010-08	2014-03	NCT01665937
Ganetespib (STA-9090)	\	Breast Cancer	Phase II	COMPLETED	2012-07	2015-08	NCT01677455
Ganetespib (STA-9090)	\	Head and neck cancers	Phase I	TERMINATED	2014-12	2016-02	NCT02334319
Ganetespib (STA-9090)	Carboplatin; Paclitaxel; Radiation	Esophageal Cancer	Phase I	COMPLETED	2015-04	2019-07	NCT02389751
Ganetespib (STA-9090)	Docetaxel	Solid Tumors	Phase I	COMPLETED	2010-07	2013-05	NCT01183364
Ganetespib (STA-9090)	Fulvestrant	Breast Cancer	Phase II	COMPLETED	2012-06	2016-06	NCT01560416
Ganetespib (STA-9090)	Ganetespib; Paclitaxel; Doxorubicin; Cyclophosphamide	Triple Negative Breast Cancer		WITHDRAWN	2016-05	2018-12	NCT02637375
Ganetespib (STA-9090)	Niraparib; Carboplatin; Paclitaxel; Gemcitabine	Ovarian Cancer	Phase II	COMPLETED	2018-11	2023-09	NCT03783949
Ganetespib (STA-9090)	Paclitaxel	Recurrent Fallopian Tube Cancer; Ovarian cancer	Phase I + Phase II	TERMINATED	2013-10	2018-07	NCT01962948
Ganetespib (STA-9090)	Paclitaxel	Epithelial Ovarian Cancer	Phase I + Phase II	TERMINATED	2014-07	2017-12	NCT02012192
Ganetespib (STA-9090)	Paclitaxel; Trastuzumab; Pertuzumab	Metastatic HER2-positive Breast Cancer	Phase I	COMPLETED	2014-04	2018-06	NCT02060253
Ganetespib (STA-9090)	Plerixafor; AC220	Acute Myeloid Leukaemia	Phase I + Phase II	COMPLETED	2011-04	2014-01	NCT01236144
Ganetespib (STA-9090)	Radiation	Pancreatic cancer	Phase II	TERMINATED	2010-12	2013-05	NCT01227018
Ganetespib (STA-9090)	Sirolimus	Malignant Peripheral Nerve Sheath Tumors (MPNST)	Phase I + Phase II	COMPLETED	2013-12	2018-07	NCT02008877
Ganetespib (STA-9090)	Ziv Aflibercept	Neoplasms	Phase I	TERMINATED	2014-12	2016-02	NCT02192541
HSP990	\	Advanced Solid Malignancies	Phase I	COMPLETED	2009-05	2012-07	NCT00879905
HSP990	\	Advanced Solid Tumors	Phase I	TERMINATED	2010-02	2012-01	NCT01064089
HSP990	\	Parkinson's Disease (PD)		RECRUITING	2024-09		NCT07380204
KW-2478	Bortezomib	Multiple Myeloma	Phase I + Phase II	COMPLETED	2010-03	2013-11	NCT01063907
Luminespib (AUY922)	\	Breast Cancer	Phase I + Phase II	COMPLETED	2007-07	2012-04	NCT00526045
Luminespib (AUY922)	\	Non-small-cell Lung Cancer	Phase II	COMPLETED	2010-10	2014-08	NCT01124864
Luminespib (AUY922)	\	Advanced Solid Tumors	Phase I	COMPLETED	2008-11	2012-05	NCT01132625
Luminespib (AUY922)	\	Gastrointestinal Stromal Tumor	Phase II	COMPLETED	2011-12	2015-02	NCT01404650
Luminespib (AUY922)	\	Pancreatic cancer	Phase II	TERMINATED	2012-01	2013-08	NCT01484860
Luminespib (AUY922)	\	Lymphoma	Phase II	TERMINATED	2012-08	2015-11	NCT01485536
Luminespib (AUY922)		Unspecified Adult Solid Tumor, Protocol Specific	Phase I	TERMINATED	2011-09	2012-02	NCT01602627
Luminespib (AUY922)		Myeloproliferative Neoplasms	Phase II	TERMINATED	2012-08	2015-05	NCT01668173
Luminespib (AUY922)	\	Non Small Cell Lung Cancer	Phase II	COMPLETED	2013-01	2017-11	NCT01752400
Luminespib (AUY922)	Bortezomib; Dexamethasone	Relapsed or Refractory Multiple Myeloma	Phase I + Phase II	COMPLETED	2008-07	2011-01	NCT00708292
Luminespib (AUY922)	BYL719	Advanced or Metastatic Gastric Cancer	Phase I	COMPLETED	2012-12	2014-03	NCT01613950
Luminespib (AUY922)	Cetuximab	Recurrent Colon Cancer	Phase I	COMPLETED	2011-02	2015-05	NCT01294826
Luminespib (AUY922)	Docetaxel; Pemetrexed	Advanced Non Small Cell Lung Cancer (NSCLC)	Phase II	TERMINATED	2012-11	2015-11	NCT01646125
Luminespib (AUY922)	Erlotinib Hydrochloride	Non-small-cell Lung Cancer	Phase I + Phase II	COMPLETED	2011-04	2014-09	NCT01259089
Luminespib (AUY922)	Pemetrexed Disodium	Recurrent Non-small Cell Lung Cancer	Phase I	COMPLETED	2014-01	2018-10	NCT01784640
Luminespib (AUY922)	Trastuzumab	Advanced HER2-positive Breast Cancer	Phase I + Phase II	COMPLETED	2010-09	2013-10	NCT01271920
Luminespib (AUY922)	Trastuzumab	Advanced Gastric Cancer	Phase II	TERMINATED	2011-11	2013-06	NCT01402401
Onalespib (AT13387)		Metastatic Solid Tumors	Phase I	COMPLETED	2008-05	2014-03	NCT00878423
Onalespib (AT13387)	\	Solid Tumors; Breast Cancer	Phase I	COMPLETED	2010-11	2017-10	NCT01246102
Onalespib (AT13387)	\	Lymphoma	Phase II	TERMINATED	2016-04	2021-03	NCT02572453
Onalespib (AT13387)	Abiraterone Acetate; Prednisone	Prostate Cancer	Phase I + Phase II	COMPLETED	2012-09	2014-12	NCT01685268
Onalespib (AT13387)	Cdk Inhibitor AtT519	Advanced Malignant Solid Neoplasm	Phase I	COMPLETED	2016-09	2020-10	NCT02503709
Onalespib (AT13387)	Cisplatin; Onalespib; Radiation	Head and neck cancers	Phase I	COMPLETED	2015-10	2020-04	NCT02381535
Onalespib (AT13387)	Crizotinib	Non-small Cell Lung Cancer(NSCLC)	Phase I + Phase II	COMPLETED	2012-10	2017-05	NCT01712217
Onalespib (AT13387)	Olaparib	Advanced Malignant Solid Neoplasm	Phase I	COMPLETED	2017-05	2022-01	NCT02898207
Onalespib (AT13387)	Onalespib; Paclitaxel	Triple Negative Breast Cancer	Phase I	TERMINATED	2016-01	2022-10	NCT02474173
Onalespib (AT13387)	Talazoparib	Adult Solid Tumors	Phase I	WITHDRAWN	2016-03	2019-03	NCT02627430
Pimitespib (TAS-116)	\	Advanced Solid Tumors	Phase I	COMPLETED	2017-07	2019-05	NCT02965885
Pimitespib (TAS-116)	Palbociclib Oral Product	Advanced Breast Cancer; Rb-null Cancer	Phase I	WITHDRAWN	2023-09	2024-08	NCT05655598
PU-H71	\	Adult Solid Tumors	Phase I	COMPLETED	2011-07	2023-03	NCT01393509
PU-H71	\	Solid Tumors	Phase I	TERMINATED	2011-04	2014-09	NCT01581541
PU-H71	\	Primary Myelofibrosis (PMF)	Phase I	TERMINATED	2019-08	2022-11	NCT03935555
Retaspimycin (IPI-504)	\	Lung Cancer	Phase I + Phase II	COMPLETED	2007-01	2011-12	NCT00431015
Retaspimycin (IPI-504)	\	Metastatic Melanoma	Phase II	TERMINATED	2008-02	2009-10	NCT00627419
Retaspimycin (IPI-504)	\	Breast Cancer	Phase I + Phase II	WITHDRAWN	2008-04		NCT00627627
Retaspimycin (IPI-504)	\	Gastrointestinal Stromal Tumors	Phase III	TERMINATED	2008-08	2009-05	NCT00688766
Retaspimycin (IPI-504)	\	Lung Cancer	Phase II	TERMINATED	2010-10	2012-02	NCT01228435
SNX-5422	\	Solid Tumors	Phase I	COMPLETED	2007-06	2010-03	NCT00506805
SNX-5422	\	Hematologic Neoplasms	Phase I	COMPLETED	2008-01	2010-05	NCT00595686
SNX-5422	\	Lymphoma	Phase I	COMPLETED	2008-03	2011-06	NCT00644072
SNX-5422	\	Solid Tumor Malignancy; Leukemia; Lymphoma	Phase I	COMPLETED	2008-03	2010-09	NCT00647764
SNX-5422	\	Solid Tumors	Phase I	COMPLETED	2012-05	2013-03	NCT01611623
SNX-5422	\	Hematological Malignancies	Phase I	COMPLETED	2014-02	2017-02	NCT01635712
SNX-5422	\	HER2+ cancers	Phase I + Phase II	TERMINATED	2013-04	2015-06	NCT01848756
SNX-5422	\	Lung cancer	Phase I	COMPLETED	2013-03	2016-08	NCT01851096
SNX-5422	\	TP53 Null Cancers	Phase II	TERMINATED	2016-03	2016-10	NCT02612285
SNX-5422	Carboplatin; Paclitaxel	Solid Tumors	Phase I	COMPLETED	2013-11	2017-07	NCT01892046
SNX-5422	Everolimus	Neuroendocrine Tumors	Phase I	COMPLETED	2014-02	2018-03	NCT02063958
SNX-5422	Ibrutinib	Leukemia	Phase I	WITHDRAWN	2017-02	2018-07	NCT02914327
SNX-5422	Ibrutinib	Leukemia	Phase I	TERMINATED	2017-02	2019-02	NCT02973399
Tanespimycin (17-AAG)	\	Unspecified Adult Solid Tumor, Protocol Specific	Phase I	COMPLETED	1998-08	2007-01	NCT00003969
Tanespimycin (17-AAG)	\	Solid Tumors; Leukemia	Phase I	COMPLETED	2004-09	2007-08	NCT00093821
Tanespimycin (17-AAG)	\	Chronic Lymphocytic Leukemia	Phase I	TERMINATED	2005-05	2007-05	NCT00319930
Tanespimycin (17-AAG)	Bortezomib	Multiple Myeloma	Phase II + Phase III	COMPLETED	2007-08	2009-02	NCT00514371
Tanespimycin (17-AAG)	Bortezomib	Multiple Myeloma	Phase III	COMPLETED	2008-02	2010-03	NCT00546780
Tanespimycin (17-AAG)	Gemcitabine Hydrochloride; Cisplatin	Unspecified Adult Solid Tumor, Protocol Specific	Phase I	COMPLETED	2002-08	2007-09	NCT00047047
XL888	\	Solid Tumors	Phase I	TERMINATED	2008-11	2010-11	NCT00796484
XL888	Pembrolizumab	Advanced Gastrointestinal Cancer	Phase I	COMPLETED	2017-07	2021-06	NCT03095781
XL888	Vemurafenib	Melanoma	Phase I	COMPLETED	2012-07	2021-09	NCT01657591
XL888	Vemurafenib; Cobimetinib	Melanoma; Skin Cancer	Phase I	ACTIVE_NOT_RECRUITING	2016-09	-	NCT02721459

Adaptive responses reveal the resilience of the Hsp90 system

Preclinical and clinical studies showed that Hsp90 inhibition engages far more than a single drug target. Once the chaperone is perturbed, cells activate a layered adaptive program that spans transcriptional stress responses, proteostasis remodeling, signaling rewiring, metabolic compensation, and microenvironmental feedback (Figure 3). This body of work became one of the field’s most important sources of mechanistic insight because it revealed how deeply resilience is built into the architecture of chaperone-supported systems.

Fig. 3.

Cellular adaptive programs driving resistance to Hsp90 inhibition. This figure was created using BioRender.

The earliest and most immediate resistance mechanism is the activation of heat shock factor 1 (HSF1).¹³² Under basal conditions, HSF1 is retained in the cytoplasm through direct interaction with Hsp90 and Hsp70.102, 103 Pharmacologic disruption of Hsp90 releases monomeric HSF1, enabling trimerization,133, 134 nuclear translocation,¹³⁵ and binding to heat-shock elements (HSEs) within promoters of HSPA1A (Hsp70), HSPB1 (Hsp27), and DNAJB1 (Hsp40).136, 137 This transcriptional cascade induces a stimulus-dependent increase in cytoprotective chaperones, often several-fold depending on cell type and stress conditions, within hours, restoring folding capacity and buffering proteotoxic stress. Elevated Hsp70 antagonizes apoptosis by sequestering Bax and Apaf-1,104, 138 while Hsp27 stabilizes actin filaments and inhibits caspase-3 activation.138, 139 Importantly, HSF1 also drives a tumor-specific transcriptional program distinct from the canonical stress response, promoting cell-cycle progression, anabolic metabolism, invasion, and stromal remodeling.140, 141 Thus, paradoxically, pharmacologic inhibition of Hsp90 can unleash an oncogenic HSF1 program that enhances tumor fitness while buffering proteotoxic stress.

This biology has important implications for dosing. High-dose, intermittent “pulse” schedules provoke intense proteotoxic stress and robust HSF1 activation, often used as a pharmacodynamic marker. However, such acute stress may amplify cytoprotective pathways. Preclinical studies instead suggest that lower, sustained inhibition can chronically suppress oncogenic clients while avoiding maximal heat shock induction.142, 143 Notably, sub-threshold Hsp90 inhibition enhances proteasomal processing of clients and broadens the MHC class I immunopeptidome, thereby improving tumor antigen presentation without compromising T-cell function.¹⁴⁴ These findings indicate that prolonged, moderate inhibition may be more compatible with immune-mediated tumor control than short bursts of maximal stress.

With persistent exposure, tumor cells further reorganize the chaperone network itself. Many cancers increase expression of stress-inducible Hsp90α and assemble higher-order, tightly interconnected chaperone super complexes termed the epichaperome. These structures integrate Hsp90, Hsp70, and multiple co-chaperones into a stable scaffold that coordinates oncogenic signaling under stress. The epichaperome has been detected in a substantial fraction of human tumors and correlates with heightened chaperone dependence and survival capacity.⁶ Rather than simply restoring folding capacity, this network rewiring is thought to create a new, stress-adapted proteostasis architecture that stabilizes malignant signaling.

Proteostasis adaptation also extends beyond the cytosol. Hsp90 inhibition disrupts folding of secretory and membrane proteins, provoking endoplasmic reticulum stress and activation of the unfolded protein response (UPR). All three canonical UPR arms: IRE1/XBP1, PERK/eIF2α/ATF4, and ATF6 are engaged, expanding ER folding capacity through induction of BiP and GRP94 and attenuating global translation.¹⁴⁵ Over time, these acute responses can become hard-wired dependencies. For example, drug-resistant KRAS-mutant lung cancers exploit an AXL–eIF4E–UPR axis that is normally buffered by Hsp90, rendering them paradoxically reliant on sustained chaperone activity for survival.¹⁴⁶ Thus, transient proteotoxic stress is converted into a durable, rewired state that supports resistance.

When folding capacity remains insufficient, cells activate autophagy as an auxiliary clearance pathway. By degrading aggregated proteins and damaged organelles, autophagy alleviates the proteotoxic burden imposed by Hsp90 inhibition. In non-small cell lung cancer models, Hsp90 inhibition induces a caspase-9–dependent autophagic program that limits cell death, whereas pharmacologic or genetic suppression of autophagy restores drug sensitivity.¹⁴⁷ These observations provide a mechanistic rationale for combining Hsp90 inhibitors with autophagy blockade.

Resistance also arises from signaling plasticity among client oncoproteins. Many oncogenic kinases and transcription factors are unusually dependent on the Hsp90–CDC37 complex for stability. Mutant BRAF,^V600E BCR–ABL, and conformationally unstable mutant p53 variants all display strong chaperone dependence and are efficiently degraded upon Hsp90 inhibition.142, 148 However, elimination of one dominant driver often promotes compensation through alternate pathways. Examples include bypass of EGFR inhibition in lung cancer, integrin αvβ3–FAK signaling in KRAS-driven tumors, and JAK–STAT3 activation in triple-negative breast cancer, each of which attenuates Hsp90 inhibitor efficacy unless co-targeted.149, 150, 151, 152 Thus, client degradation frequently redistributes, rather than extinguishes, oncogenic signaling.

Additional robustness is provided by co-chaperone redundancy and compartmentalization. CDC37 selectively protects kinase clients, and its suppression sensitizes tumors to Hsp90 inhibition.¹⁵³ Distinct mitochondrial pools of Hsp90 and the paralog TRAP1 safeguard mitochondrial integrity; inhibitors targeting these compartments (gamitrinibs, for example) induce mitochondrial permeability transition and apoptosis even in cells resistant to cytosolic Hsp90 inhibitors.¹⁵⁴ Complete disruption of tumor proteostasis may therefore require simultaneous targeting of multiple chaperone modules. Beyond cell-intrinsic mechanisms, the tumor–immune interface also influences outcomes. High-level HSP90 inhibition can impair T-cell and NK-cell function by destabilizing key signaling intermediates,¹⁵⁵ whereas partial inhibition may enhance tumor antigen presentation.¹⁴² These dose-dependent immunologic effects support careful integration of Hsp90 inhibitors with immunotherapy.

Together, these studies show that the cellular response to Hsp90 inhibition is an active systems-level adaptation. Heat-shock response (HSR), unfolded-protein-response signaling, autophagy, client rewiring, compartmental buffering, and microenvironmental remodeling collectively revealed how proteostasis networks preserve function under pressure. In doing so, they transformed the field’s understanding of Hsp90 from a broad client stabilizer into a central organizer of cellular resilience (Box 2).

Box 2. What Hsp90 Inhibition Revealed about Cellular Adaptation.

• •Heat-shock activation: Hsp90 inhibition releases HSF1, inducing HSP70, HSP27, and other stress programs that rapidly expand folding capacity.
• •Proteostasis reinforcement: Tumor cells strengthen buffering through the unfolded protein response, autophagy, and assembly of the epichaperome.
• •Signaling plasticity: Loss of one client can uncover bypass circuits such as FAK, JAK–STAT, integrin, or receptor-tyrosine-kinase compensation.
• •Distributed robustness: Co-chaperones such as CDC37 and compartmentalized pools such as mitochondrial TRAP1 preserve selected client functions even when cytosolic Hsp90 is engaged.
• •Exposure matters: High-intensity pulse dosing and sustained low-dose inhibition can produce different biological states and different adaptive outputs.
• •Pharmacology shapes biology: Bioactivation, pharmacokinetics, and efflux determine how much of the Hsp90 system is perturbed in vivo.
• •Immune context matters: Intensive Hsp90 inhibition can either support or suppress antitumor immunity depending on dose, schedule, and cellular target.
• •Dependency must be defined: Clinical benefit is strongest when tumors are selected by client addiction, proteostasis burden, epichaperome abundance, or other biomarkers of Hsp90 reliance.

New strategies beyond N-terminal inhibition

The biology exposed by classical NTD inhibitors was richer than any single pocket. Alternative strategies are therefore sought to engage other structural surfaces, regulatory interfaces, paralog-specific features, and disease-associated chaperone states, as summarized in Table 1. This shift turned Hsp90 drug discovery from ATP-competitive inhibition toward a more modular and biologically precise view of the system.

C-terminal allosteric inhibitors

Parallel to N-terminal inhibitor development, the coumarin antibiotic novobiocin was identified as binding to an ATP-binding site in the CTD of Hsp90, establishing an alternative inhibitory mechanism.84, 156 Novobiocin’s binding to the CTD disrupts Hsp90 homodimerization through the CTD interface and interferes with cochaperone interactions mediated by the MEEVD motif, including binding to tetratricopeptide repeat (TPR)-containing immunophilins such as FKBP51, FKBP52 and Cyp40.85, 157 Importantly, CTD inhibitors do not induce HSR.157, 158, 159 Structure-activity relationship studies on novobiocin identified that removal of the 4-hydroxy and 3′-carbamate groups enhanced Hsp90 selectivity over DNA gyrase, while modifications to the noviose sugar and coumarin core improved antiproliferative potency from approximately 700 μM to low micromolar and nanomolar ranges.160, 161, 162 The novobiocin analog KU-32 demonstrated neuroprotective rather than cytotoxic activity, stimulating Hsp90 chaperone function through an allosteric mechanism that uncouples the cytotoxic and cytoprotective outcomes of Hsp90 modulation.163, 164

Middle-domain modulators and co-chaperone disruptors

A complementary strategy targets the regulatory interfaces of the Hsp90 system, either by direct engagement of the MD, which governs client recognition and ATPase regulation, or by disrupting cochaperone interactions that depend on MD integrity.

The earliest evidence for cochaperone-directed disruption came from gedunin, which was shown to directly bind and inactivate the cochaperone p23, thereby impairing p23-dependent stabilization of the Hsp90 closed state and triggering apoptotic cancer cell death.¹⁶⁵ These studies established that cochaperone-directed disruption could phenocopy Hsp90 inhibition independently of the ATP-binding domain. Shortly thereafter, the natural product celastrol was identified as another modulator of MD-associated functions: biochemical reconstitution assays demonstrated that celastrol disrupts the Hsp90–CDC37 complex, inhibits ATPase cycling without competing with ATP, and operates through a thiol-dependent mechanism in cells.87, 166, 167 This showed that celastrol acts like gedunin through inactivation of p23.¹⁶⁸

The first compound shown to engage the Hsp90 MD directly was sansalvamide A (San A), a cyclic depsipeptide that binds the N-terminal segment of the MD and allosterically interrupts interactions with CTD, engaging cochaperones while leaving the ATP-binding pocket unoccupied.¹⁶⁹ This MD-targeted mechanism motivated the synthesis of multiple San A derivatives, including H-10, LY-15, and H-15, which inhibit B16 melanoma proliferation and induce apoptosis or differentiation in preclinical models,170, 171, 172 providing early proof that the MD surface is pharmacologically druggable.

The mechanistic importance of MD targeting was further supported by the identification of withaferin A (WFA), which preferentially engages the CTD of Hsp90, disrupts Hsp90–CDC37 complexes, inhibits chaperone activity independently of ATP binding, and induces proteasome‑dependent client degradation. SAR studies further validated electrophilic withanolides as authentic Hsp90‑targeting agents.173, 174

The clearest validation of MD druggability came next with gambogic acid (GBA), which was shown to bind an isoform-selective, well-defined pocket in the Hsp90β MD, establishing the first high-resolution ligand–MD interface.¹⁷⁵ However, subsequent work revealed that GBA also acts as an electrophile capable of inducing a thiol-dependent heat-shock response and disrupting Hsp90–HSF1/HSF2 complexes.¹⁷⁶

A breakthrough in potency and selectivity was achieved with diptoindonesin G (dip G), identified as an Hsp90 MD binder (Kₐ ≈ 0.13 μM) that promotes client degradation without inducing the heat-shock response. Importantly, dip G retains full activity against endocrine-resistant breast cancer cells harboring ER-LBD mutations, demonstrating a unique therapeutic profile.¹⁷⁷

More recent work has yielded synthetic allosteric modulators such as SOMCL-16–171/175, the first designed small-molecule MD binders shown to allosterically remodel Hsp90, destabilize clients, and avoid the liabilities of ATP-site inhibition.¹⁷⁸ This was followed by the discovery of enniatin A, a cyclic depsipeptide that inhibits Hsp90 with MD-consistent biochemical signatures, triggers client degradation without a robust heat-shock response, and, in syngeneic TNBC models, induces immunogenic cell death, lowers PD-L1, and enhances CD8⁺ T-cell–mediated antitumor activity.¹⁷⁹

Structural work on human Hsp90α middle/CTDs further confirmed a druggable allosteric pocket within the MD,⁸⁶ while integrative NMR/SAXS studies revealed long-range allosteric communication across the chaperone, providing a mechanistic basis for MD-directed pharmacology.¹⁸⁰ Despite these advances, no MD Hsp90 inhibitor has entered clinical trials yet. However, the growing body of literature suggests that MD-directed modulation may overcome key limitations of NTD inhibitors, including HSR activation and lack of client selectivity.181, 182

Isoform-selective inhibitors

A major advance in Hsp90 drug discovery was the recognition that the four mammalian isoforms, the cytosolic chaperones Hsp90α and Hsp90β, the endoplasmic-reticulum isoform GRP94, and the mitochondrial isoform TRAP1 are not functionally interchangeable but instead support distinct physiological and pathological roles.183, 184 This insight reframed selectivity as a biological question as much as a pharmacological challenge, and if different paralogs govern different client sets, tissues, and stress programs, then isoform-selective ligands could reveal which fraction of the Hsp90 system drives a given phenotype.¹¹

Hsp90β-selective inhibitors

Among the isoforms, Hsp90β has emerged as the most advanced selective target. Early observations came from ligands whose selectivity was discovered empirically. Gambogic acid was shown to bind an isoform-specific pocket in the Hsp90β MD, expanding the field beyond the canonical NTD ATP site.¹⁷⁵ Corylin provided a particularly compelling example of biologically meaningful selectivity: hepatic Hsp90β, but not Hsp90α, correlated with lipid dysregulation in NAFLD and obesity models, and corylin disrupted the Hsp90β–Akt axis to promote degradation of mature SREBPs and improve metabolic phenotypes in vivo.¹⁸⁵

Structure-guided medicinal chemistry transformed these observations into a systematic strategy. KUNB31 demonstrated that strong Hsp90β preference over Hsp90α, GRP94, and TRAP1 could be engineered from an NTD scaffold.¹⁸⁶ This lineage expanded through NDNB1182 and later analogs, establishing Hsp90β selectivity as a reproducible design.187, 188 Functionally, Hsp90β-selective inhibition enhances interferon signaling and improves responses to immune checkpoint blockade in murine tumors.¹⁸⁹ Multi-omics profiling of NDNB-25 further revealed resistance mechanisms, including Rho GTPase rewiring, cytoskeletal remodeling, metabolic and redox adaptation, kynurenine–AHR signaling, and compensatory Hsp90α upregulation and identified synergy with carboplatin.¹⁹⁰ Collectively, these studies define Hsp90β-selective inhibition as a distinct therapeutic state.

Hsp90α-selective inhibitors

Selective targeting of Hsp90α emerged later than Hsp90β-directed chemistry, but it provided important proof that the two cytosolic isoforms can be pharmacologically separated in both directions. Building on the AT13387 scaffold, several highly selective Hsp90α inhibitors were developed.191, 192, 193 Among the earliest examples, KUNA-115 showed the strongest initial selectivity, with more than 60-fold preference for Hsp90α over Hsp90β (Hsp90α IC₅₀ = 0.84 μM; Hsp90β IC₅₀>50 μM).¹⁹¹ Subsequent structure–activity relationship studies identified the 5-fluoroisoindoline derivative KUNA-111, which introduced a novel binding mode while retaining approximately 15-fold selectivity for Hsp90α.¹⁹² Structural analysis indicated that these inhibitors engage Leu48, Asn51, Asp93, and Thr184 through a hydrogen-bonding network that contributes to Hsp90α selectivity.191, 192 Further optimization of KUNA-115 showed that introducing a cell-permeable dimethylamine into the solvent-exposed region substantially improved Hsp90α selectivity, producing compounds with more than 196-fold selectivity for Hsp90α over Hsp90β, whereas replacement of permanently charged moieties increased selectivity even further to more than 294-fold.¹⁹³ Together, these studies established Hsp90α-selective inhibition as a druggable target and provided a structural basis for the continued development of α-biased ligands.

GRP94-selective inhibitors

GRP94 provided one of the earliest structural demonstrations that isoform selectivity is achievable within the Hsp90 family. Early work showed that adenosine-derived ligands such as NECA interact differently with GRP94 than with cytosolic Hsp90.¹⁹⁴ The breakthrough came when PU-H54 and related purine ligands were shown to access a hydrophobic Site-2 pocket in the GRP94 NTD that is not equivalently accessible in Hsp90α/β, thereby enabling true isoform selectivity.¹⁹⁵ Compounds such as PU-WS13 were subsequently used to define tumor-specific HER2 regulation and to show that GRP94-selective inhibition can block IGF-II secretion and TLR trafficking without reproducing canonical cytosolic Hsp90 readouts.195, 196 Later chemistry, including PU-H39, the amino-cyclohexanol scaffold ACO1, and KUNG65, further validated GRP94 as a structurally and pharmacologically distinct isoform.197, 198

TRAP1 and mitochondrial-targeted inhibitors

The trajectory for TRAP1 differed in that early selectivity arose primarily through mitochondrial targeting. The gamitrinib family coupled an Hsp90-binding scaffold to a mitochondrial-targeting moiety, concentrating inhibitory activity in tumor mitochondria and demonstrating that compartment-directed pharmacology can create isoform-selective therapeutic states.¹⁵⁴ Structure-guided work then showed that TRAP1 can also support more direct ligand development. In particular, Lee et al. used TRAP1 crystal structures to develop the mitochondria-targeted inhibitor SMTIN-P01, providing a structural basis for TRAP1-directed pharmacology.¹⁹⁹ TRAP1 also differs mechanistically from other Hsp90 family members since it has been linked to mitochondrial metabolic regulation, adopts an asymmetric closed state, and undergoes a sequential ATP hydrolysis cycle with asymmetry flipping during catalysis.200, 201, 202 These features probably contribute to their distinct draggability.

Epichaperome inhibitors

The foundation for epichaperome-targeted therapy was established by Moulick et al, who showed that Hsp90 isn’t only overexpressed in cancer, but it exists in a functionally distinct, high-affinity state. Using the inhibitor PU-H71 as an affinity probe, they captured stable, multimeric Hsp90 complexes enriched in oncogenic client proteins and co-chaperones directly from tumor cells. This work revealed that Hsp90 functions as a central scaffold for cancer-specific signaling networks and that tumor dependency is driven by its network organization rather than abundance.⁴² This discovery led to the definition of the epichaperome: a stable, stress-induced, highly integrated chaperome network composed of Hsp90, Hsp70/HSC70, and co-chaperones that stabilizes aberrant proteome-wide connectivity to support malignant survival.⁶ Unlike the dynamic chaperone assemblies found in non-transformed cells, epichaperomes act as pathological scaffolds, creating disease-specific vulnerabilities across cancer and neurodegeneration.6, 203

The purine-scaffold inhibitor PU-H71 was the first compound shown to preferentially engage Hsp90 within epichaperome assemblies, exploiting conformational states unique to these rigid networks.6, 42 In triple-negative breast cancer models, PU-H71 induced potent and durable antitumor responses, including complete tumor regressions.²⁰⁴ Affinity-based proteomics further demonstrated that PU-H71 can isolate tumor-enriched Hsp90 complexes and map the oncogenic networks coordinated by the epichaperome.⁴² These findings reframed the therapeutic logic: rather than inhibiting individual clients, epichaperome disruption collapses a pathological proteostasis scaffold and simultaneously dismantles multiple signaling dependencies.

Epichaperome-targeted imaging enabled biomarker-guided clinical development. A first-in-human study of ¹²⁴I-PU-H71 PET established the feasibility of non-invasive epichaperome imaging and validated tumor tracer retention as a pharmacometric readout of target engagement.²⁰⁵ In HER2-negative metastatic breast cancer, companion PU-PET imaging was incorporated into a phase Ib trial of PU-H71 plus nab-paclitaxel, where higher baseline epichaperome signal correlated with more durable disease control.²⁰⁶

In Alzheimer’s disease models, maladaptive epichaperomes were shown to drive hippocampal connectivity dysfunction, and pharmacologic disruption with the brain-penetrant analog PU-AD reversed key network abnormalities.²⁰³ CNS-directed probe development demonstrated on-target activity and kinetic selectivity of radiolabeled epichaperome ligands in cells and mice, together with exploratory human imaging.²⁰⁷ The oral epichaperome inhibitor icapamespib (PU-AD) showed acceptable safety, tolerability, and pharmacokinetics in healthy adults.²⁰⁸

Targeted protein degradation strategies

Targeted protein degradation (TPD) has added a new dimension to Hsp90 pharmacology by moving beyond reversible, occupancy-driven inhibition toward direct elimination of the chaperone or Hsp90 client proteins. The first reported Hsp90-targeting PROTAC, BP3, demonstrated that Hsp90 itself can be degraded, resulting in potent suppression of breast cancer cell growth and tumor inhibition in vivo.²⁰⁹ This established that Hsp90 is not only druggable but degradable.

Subsequent efforts refined this strategy toward isoform selectivity. X10g selectively degraded Hsp90α while sparing Hsp90β, providing proof that isoform-specific degradation is achievable through degrader architecture.²¹⁰ In parallel, classical Hsp90 inhibitor scaffolds were repurposed into degraders: the first geldanamycin-based PROTACs showed that cereblon-recruiting analogs can degrade both Hsp90α and Hsp90β via the ubiquitin–proteasome pathway.²¹¹ More recent chemistry expanded the degrader landscape further. For example, lw13 degraded Hsp90 at low concentrations and synergized with cisplatin in cervical cancer models.²¹²

A conceptual extension of this field is to use Hsp90 not as the degradation target, but as the effector module. This strategy has been described using related terms, including Hsp90 interactome-mediated PROTACs (HIM-PROTACs), Hsp90-mediated proteolysis-targeting chimeras (HSPTACs), and chaperone-mediated protein degradation (CHAMP). These approaches use heterobifunctional molecules to induce chemical proximity between a protein of interest and the Hsp90 chaperone complex, thereby promoting target ubiquitination and proteasome-dependent degradation. In the HIM-PROTAC approach, GPX4 was chemically bridged to the Hsp90 interactome to induce ubiquitin–proteasome-dependent GPX4 degradation and trigger ferroptosis. The optimized degraders GDCNF-2 and GDCNF-11 depleted GPX4, increased lipid peroxidation and ferroptosis, and suppressed tumor growth in vivo with lower toxicity than combined GPX4 and Hsp90 inhibitor treatment. Mechanistically, these degraders showed improved ferroptosis selectivity compared with conventional GPX4 inhibition and implicated Hsp90-associated E3 ligases, including CHIP and CUL5, in GPX4 degradation.²¹³ The scope of this strategy was further broadened by HSPTACs, which showed that the protein of interest does not need to be a client of Hsp90. In a PARP1-focused study, PARP1 was validated as a nonclient protein of Hsp90, and the lead molecule DDO3602 was designed to force the Hsp90–PARP1 ternary interaction. DDO3602 induced PARP1 ubiquitination and proteasome-dependent degradation through recruitment of multiple E3 ubiquitin ligases, including TRIM25, UBR5, MYCBP2, MARCHF5, and TRIM50. Functionally, DDO3602 promoted G2/M cell-cycle arrest and DNA damage in breast cancer cells and showed tumor-selective pharmacokinetics in vivo.²¹⁴ The CHAMP platform applied the same principle to ERK5, a kinase for which conventional inhibitors can produce paradoxical activation through nuclear translocation. ERK5-directed CHAMP degraders, including RNK06138 and the orally available RNK06395, induced proximity between ERK5 and the Hsp90 chaperone complex to promote proteasome-dependent ERK5 degradation. These compounds prevented ERK5 nuclear accumulation, reduced ERK5-dependent transcriptional activity, decreased cancer cell viability, induced cell-cycle arrest and apoptosis, and suppressed tumor growth in xenograft models. Their preferential tumor accumulation further supports the idea that the activated Hsp90 state in cancer can be exploited for both target degradation and tumor retention.²¹⁵ These Hsp90-mediated degrader strategies broaden the use of chaperone biology in TPD by shifting Hsp90 from a direct drug target to a programmable degradation platform capable of engaging client, interactome-associated, and nonclient proteins (Figure 4).

Combination strategies

Combination studies became especially informative once it was clear that Hsp90 sits at the junction of multiple dependency networks rather than a single pathway. By destabilizing broad client sets, Hsp90 inhibitors can intersect with cytotoxic stress, proteasome overload, receptor addiction, DNA-repair dependence, and immune-state regulation.216, 217, 218

Chemotherapy combinations

Chemotherapy was the earliest and most extensively investigated partner for Hsp90 inhibition. Quantitative data on combination partners confirm that Hsp90 inhibitors are most frequently paired with agents that target high-dependency clients or stress-sensitive pathways. Paclitaxel is the most common partner, appearing in 8 distinct studies, followed by trastuzumab in 4 studies and bortezomib in 3 studies (Figure 4(c)). Platinum-based agents (carboplatin and cisplatin) also represent frequent partners, appearing in 2 studies each. These distribution patterns underscore a strategic clinical focus on sensitizing tumors to taxane-induced mitotic stress and amplifying the effects of HER2- and proteasome-directed therapies. In preclinical non-small-cell lung cancer models, ganetespib synergized with paclitaxel and docetaxel, and the combinations produced greater tumor inhibition than either agent alone.²¹⁸ This strategy also had early support from classic 17-AAG–taxane studies showing that Hsp90 inhibition can suppress Akt-dependent survival signaling and sensitize tumors to paclitaxel.²¹⁶

In clinical settings, 17-AAG combined with gemcitabine and cisplatin showed antitumor activity in refractory solid tumors, while also highlighting the hematologic burden associated with this regimen.²¹⁹ Ganetespib with pemetrexed-platinum in malignant pleural mesothelioma and SNX-5422 with carboplatin-paclitaxel in lung cancer both demonstrated feasibility and signs of activity in early-phase settings.220, 221 Taxane-based combinations remained a major translational path, and the phase III GALAXY-2 study of ganetespib plus docetaxel ultimately clarified the boundary between strong preclinical rationale and broad clinical deployment in advanced NSCLC80. In triple-negative breast cancer, onalespib plus paclitaxel further showed that the regimen was feasible and could produce responses in a subset of patients, although the data remained early phase.¹¹⁸

One of the clearest examples of a more selective combination strategy was PU-H71 plus nab-paclitaxel in HER2-negative metastatic breast cancer. In that study, benefit was linked to epichaperome positivity measured by PU-PET, making it one of the strongest examples of biomarker-guided Hsp90 combination therapy rather than unselected combination treatment.²⁰⁶

Proteostasis-directed combinations in multiple myeloma

Multiple myeloma provided a biological setting for Hsp90 combinations because plasma cells operate close to the limit of protein-handling capacity. Their high immunoglobulin output creates chronic proteotoxic stress, so simultaneous disruption of the proteasome and the chaperone machinery can drive collapse of protein homeostasis. This rationale was supported clinically by tanespimycin plus bortezomib, which showed meaningful activity and manageable toxicity in relapsed or refractory myeloma across phase II and phase I/II studies.222, 223 The strategy remained attractive beyond geldanamycin derivatives: luminespib (AUY922) also entered early-phase clinical testing with or without bortezomib in relapsed/refractory myeloma.²²⁴ Mechanistically, this combination space is strengthened by the dependence of myeloma cells on the IRE1–XBP1 arm of the unfolded protein response, which links proteasome burden, endoplasmic-reticulum stress, and survival signaling.²²⁵

HER2-, EGFR-, and ALK-directed combinations

Some of the most coherent Hsp90 combinations have been built around tumors driven by Hsp90-dependent oncoproteins. HER2-positive breast cancer is the clearest example. In trastuzumab-refractory disease, tanespimycin plus trastuzumab produced clinically meaningful activity and helped establish the principle that HER2 can be targeted both at the receptor level and through chaperone destabilization.⁹³ This rationale was extended by combinations such as ganetespib plus paclitaxel plus trastuzumab, which showed biological and clinical activity in trastuzumab-refractory HER2-positive metastatic breast cancer.²²⁶ AUY922 plus trastuzumab was also active in patients progressing after prior anti-HER2 therapy.²²⁷ These studies illustrate the strength of Hsp90 combinations when the dominant driver is itself a highly Hsp90-dependent client.

In EGFR-mutant NSCLC, onalespib plus erlotinib was mechanistically attractive because it aimed to suppress both receptor signaling and adaptive bypass pathways. Clinical experience showed, however, that activity was modest.²²⁸ Combination studies of onalespib plus crizotinib in ALK-positive NSCLC were pursued for a similar reason: ALK fusion proteins are strongly Hsp90-dependent, and the experience remains valuable as a proof of biological rationale even though the published efficacy dataset remains exploratory (NCT01712217).

Endocrine and DNA-damage-response combinations

Hsp90 inhibitors have also been combined with endocrine and pathway-directed therapies in tumors where steroid-receptor or repair signaling remains Hsp90-dependent. In metastatic castration-resistant prostate cancer, onalespib plus abiraterone acetate showed pharmacodynamic effects, including changes in androgen-receptor signaling, but this did not translate into substantial clinical benefit.²²⁹ By contrast, combinations with PARP inhibitors are among the most mechanistically compelling because Hsp90 supports multiple proteins involved in homologous recombination, replication stress, and DNA repair. In a combined preclinical and phase I study, onalespib plus olaparib was feasible and showed durable stable disease in a subset of heavily pretreated patients with advanced solid tumors.²³⁰ This combination strategy is conceptually important because it co-targets proteostasis and genome maintenance, rather than simply layering two cytotoxic mechanisms.

Immunotherapy combinations

Hsp90 inhibitors have emerged as compelling partners for immune checkpoint blockade because they target several barriers to response at once, including checkpoint-ligand expression, interferon signaling, antigen presentation, and regulatory T-cell-mediated suppression. Proia and Kaufmann outlined the rationale for this strategy and emphasized that intermittent dosing can exploit tumor-selective drug retention to sustain intratumoral Hsp90 inhibition while limiting systemic immune suppression.²³¹ They also highlighted evidence that low-dose, non-heat-shock-inducing Hsp90 inhibition can increase tumor antigen presentation and MHC I display, thereby improving T-cell recognition of treated tumors.231, 232

This concept was strengthened by Zavareh and colleagues, who identified Hsp90 inhibitors as potent suppressors of PD-L1 and PD-L2 surface expressions in a high-throughput screen of approximately 200,000 compounds. Ganetespib reduced both ligands at sub-cytotoxic concentrations, likely through loss of the Hsp90 client transcription factors STAT3 and c-Myc, and this effect extended to primary human macrophages. In vivo, systemic ganetespib significantly reduced PD-L1 surface expression on live MC38 tumor cells, and a trend toward increased intratumoral CD69+CD8+ T-cell activation was observed, though this did not reach statistical significance.²³³ The discovery of PD-L1 as a direct Hsp90 client protein by Eisa and colleagues provides a further mechanistic basis for this axis. Using the middle-domain inhibitor Enniatin A (EnnA), they demonstrated that Hsp90 is physically required to stabilize PD-L1 protein complexes. Notably, while N-terminal inhibitors like ganetespib typically reduce PD-L1 through transcriptional suppression of STAT3 and c-Myc, EnnA induces the physical dissociation of Hsp90 from PD-L1, leading to its proteasomal degradation without decreasing mRNA levels. This targeted protein depletion was found to effectively prevent CD8+ T-cell exhaustion and was essential for the compound’s ability to reprogram the immunosuppressive tumor microenvironment in aggressive triple-negative breast cancer models.¹⁷⁹

A second axis is induction of interferon-linked tumor sensitization. Mbofung and colleagues showed that Hsp90 inhibitors were among the strongest enhancers of T-cell-mediated killing in patient-derived melanoma models. Ganetespib upregulated IFIT1, IFIT2, and IFIT3, and this response was functionally required: IFIT silencing abolished the increase in T-cell killing, whereas IFIT overexpression recapitulated it, at least in part through suppression of BCL2. In vivo, ganetespib enhanced anti-CTLA4 and anti-PD-1 therapy; in the anti-CTLA4 combination setting, it increased the CD8+ T-cell:Treg ratio, granzyme A/B production, and antigen-specific IFN-gamma responses.²³⁴ Rahmy and colleagues later supported a model in which Hsp90β-selective inhibition degrades CDK4, lowers DNMT1 expression, reactivates endogenous retroviral elements, and induces interferon-stimulated genes. In syngeneic prostate and breast cancer models, the Hsp90β-selective inhibitor NDNB1182 enhanced checkpoint blockade and was better tolerated than ganetespib, supporting Hsp90β as an attractive immunotherapy partner.189, 235

A third axis is selective Treg targeting. Tsuge and colleagues showed that pimitespib degrades STAT5, reduces FOXP3high effector Tregs, and shifts the CD8+ T-cell:Treg balance in blood and tumors, a mechanism that may be especially relevant in immunologically cold tumors.¹³¹

Clinical translation remains early but encouraging. In the phase Ib EPOC1704 study, pimitespib plus nivolumab produced a 16% objective response rate in microsatellite-stable colorectal cancer without prior checkpoint inhibitor exposure, with no dose-limiting toxicities and a manageable safety profile; exploratory analyses suggested a trend toward better outcomes in PD-L1 CPS ≥1 and TMB-high tumors.²³⁶ By contrast, XL888 plus pembrolizumab was safe and biologically active but did not produce objective responses in mismatch repair-proficient colorectal cancer, although some patients achieved stable disease and correlative studies indicated local and systemic immune modulation.²³⁷ Together, these findings indicate that Hsp90 inhibitors can function as active modulators of tumor immune state, with the greatest benefit likely in settings where the dominant immunosuppressive mechanism is Hsp90-dependent (Figure 5 and Box 3).

Fig. 5.

The Immune-Oncology Effects of Hsp90 Inhibition in the Tumor Microenvironment. This figure was created using BioRender.

Box 3. Combination Settings That Clarified Where Hsp90 Targeting Matters Most.

• 1.Combination with chemotherapy
• •Learning point: Hsp90 inhibition can lower the apoptotic threshold, suppress compensatory survival signaling, and intensify mitotic or genotoxic stress when tumors are already near a survival boundary.
• •Clinical example: Ganetespib + docetaxel in advanced lung adenocarcinoma.
• •Clinical example: 17-AAG + gemcitabine/cisplatin in refractory solid tumors.
• •Clinical example: PU-H71 + nab-paclitaxel in HER2-negative metastatic breast cancer as a biomarker-guided, epichaperome-positive strategy.
• 2.Combination with proteostasis-directed therapy
• •Learning point: Hsp90 combinations are especially coherent in tumors operating close to proteostasis collapse, where simultaneous disruption of chaperone function and protein disposal can overwhelm survival capacity.
• •Clinical example: Tanespimycin + bortezomib in relapsed/refractory multiple myeloma.
• •Clinical example: Luminespib with or without bortezomib in relapsed/refractory multiple myeloma.
• 3.Combination with targeted therapy
• •Learning point: Hsp90 inhibition is most informative when the dominant oncogenic driver is itself a strongly Hsp90-dependent client or when adaptive bypass signaling is likely to emerge.
• •Clinical example: Tanespimycin + trastuzumab in HER2-positive metastatic breast cancer.
• •Clinical example: AUY922 + trastuzumab in trastuzumab-refractory HER2-positive disease.
• •Clinical example: Onalespib + erlotinib in EGFR-mutant NSCLC.
• •Clinical example: Onalespib + crizotinib in ALK-positive NSCLC.
• 4.Combination with endocrine and DNA-damage-response therapy
• •Learning point: These combinations are most revealing when Hsp90 supports hormone-receptor signaling, homologous recombination, replication stress responses, or repair plasticity.
• •Clinical example: Onalespib + abiraterone acetate in metastatic castration-resistant prostate cancer.
• •Clinical example: Onalespib + olaparib in advanced solid tumors.
• 5.Combination with immunotherapy
• •Learning point: Hsp90 inhibition can reshape tumor immune state through effects on checkpoint-ligand expression, interferon-linked tumor sensitization, antigen presentation, and immunosuppressive regulatory T-cell programs.
• •Clinical example: Pimitespib + nivolumab in advanced solid tumors.
• •Clinical example: XL888 + pembrolizumab in mismatch repair-proficient colorectal cancer.

Conclusion

Hsp90 inhibitor research shows how sustained attempts to perturb a biologically complex system can transform both therapeutic strategy and fundamental cell biology. Across natural products, synthetic ATP-competitive inhibitors, allosteric ligands, isoform-selective compounds, epichaperome-directed agents, and degraders, the field progressively clarified how chaperone networks are organized, how they adapt under stress, and how vulnerability is distributed across paralogs, compartments, and disease-associated states. These advances established Hsp90 as a central regulator of proteome stability and cellular plasticity while providing the conceptual basis for more precise therapeutic design. The most important legacy of Hsp90 inhibition, therefore, lies not only in the compounds themselves, but in the framework the field created for viewing proteostasis as a dynamic, druggable, and highly contextual system. With biomarker-guided deployment, rational combinations, and increasingly selective modalities, that framework is now positioned to support a more mature phase of translational progress.

CRediT authorship contribution statement

Ahmed Chadli: Conceptualization, supervision, writing - review & editing. Ilham Zarguan: Conceptualization, visualization, writing - review & editing. Lamiae Belayachi: Supervision, writing - review & editing. Abdelaziz Benjouad: Supervision, writing - review & editing.

Declaration of Generative AI and AI-Assisted Technologies in the Writing Process

During the preparation of this manuscript, the author(s) used NotebookLM (Google) for literature organization and ChatGPT (OpenAI) for language editing and text refinement. All scientific content, interpretations, and conclusions were developed and critically verified by the author(s), who take(s) full responsibility for the final manuscript.

Declarations of Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Data Availability

No data was used for the research described in the article.