HSP90 identified by a proteomic approach as druggable target to reverse platinum resistance in ovarian cancer

Acquired resistance to platinum (Pt)‐based therapies is an urgent unmet need in the management of epithelial ovarian cancer (EOC) patients. Here, we characterized by an unbiased proteomics method three isogenic EOC models of acquired Pt resistance (TOV‐112D, OVSAHO, and MDAH‐2774). Using this approach, we identified several differentially expressed proteins in Pt‐resistant (Pt‐res) compared to parental cells and the chaperone HSP90 as a central hub of these protein networks. Accordingly, up‐regulation of HSP90 was observed in all Pt‐res cells and heat‐shock protein 90 alpha isoform knockout resensitizes Pt‐res cells to cisplatin (CDDP) treatment. Moreover, pharmacological HSP90 inhibition using two different inhibitors [17‐(allylamino)‐17‐demethoxygeldanamycin (17AAG) and ganetespib] synergizes with CDDP in killing Pt‐res cells in all tested models. Mechanistically, genetic or pharmacological HSP90 inhibition plus CDDP ‐induced apoptosis and increased DNA damage, particularly in Pt‐res cells. Importantly, the antitumor activities of HSP90 inhibitors (HSP90i) were confirmed both ex vivo in primary cultures derived from Pt‐res EOC patients ascites and in vivo in a xenograft model. Collectively, our data suggest an innovative antitumor strategy, based on Pt compounds plus HSP90i, to rechallenge Pt‐res EOC patients that might warrant further clinical evaluation.


Introduction
Epithelial ovarian cancer (EOC) represents the seventh most commonly diagnosed cancer among women and the most lethal gynecological disease, with a 5-year relative survival rate of 46% after the diagnosis [1,2]. EOC is not a single disease but comprises different histological entities (e.g., high-grade serous, endometrioid, mucinous, clear cell, and low-grade serous) with different biological, molecular, and clinical characteristics and high-grade serous ovarian cancers (HGSOC) are the most common type of EOC, accounting for 75% of all EOC [3]. The standard treatment of advanced EOC is based on the maximum debulking surgery, followed by platinum (Pt)-based chemotherapy, which remained the same over the past three decades. Many anticancer agents, including molecular-targeted agents and combination therapies, have been developed and validated clinically in Pt-sensitive EOC patients [3]. However, the overall survival rate has not been improved significantly in the metastatic disease. Indeed, although EOC is one of the most chemo-responsive tumors, almost invariably after an initial response, the majority of patients relapse and Pt resistance invariably emerges [4,5]. Therefore, understanding the underlying molecular mechanisms associated with the onset of Pt resistance is an urgent unmet clinical need. On this regard, we previously generated and characterized three CDDP-resistant isogenic high-grade EOC cell lines (TOV-112D Pt-res, MDAH-2774 Pt-res, and OVSAHO Pt-res) from their parental counterparts [6][7][8]. Since gene expression profiling failed to identify signaling pathways commonly altered in Pt-resistant (Pt-res) cells [6], we took advantage of a proteomic approach to investigate the possible mechanisms by which cells acquire CDDP resistance. We highlighted the heat-shock protein 90 (HSP90) as a central hub of the network of proteins differentially expressed between resistant and parental cells in all the three ovarian cancer models explored. Small-molecule inhibitors for the HSP90 have been extensively exploited in preclinical studies representing promising agents in cancer treatment and some of them, such as tanespimycin (17AAG) and ganetespib (STA9090), are in phase II/III clinical studies in cancers patients alone or in combination with conventional chemotherapy 2 . Therefore, we tested either 17AAG or ganetespib activity in Pt-res ovarian cancer cells demonstrating a strong synergistic antitumor effect when used in combination with CDDP in vitro and in vivo xenograft models as well as in primary cultures from Pt-res ovarian cancer patients.
CDDP-resistant cancer cells (referred as Pt-res cells) were generated as previously described and include pooled resistant populations (pool) and individual clones, as indicated in the text [6][7][8]. Primary cells (OV102, OV202, OV199, OV200) were established from ascites of EOC patients, collected by the CRO-Aviano National Cancer Institute Institutional Biobank with a written informed consent from patients. The CRO-Aviano Internal Review Board approved this study (#IRB-06/2011).

2-D DIGE
(two-dimensional differential in-gel electrophoresis), image acquisition, analysis, and processing All 2-D DIGE reagents and instruments were provided by GE Healthcare (Pittsburgh, PA, USA). Proteomic experiments were performed as described by Milone et al. [9]. In detail, four independent biological conditions are required to give statistic confidence to 2-D DIGE data. All proteomics maps were subjected to the following two analysis modules from the DeCyder ™ version 7.2 software (GE Healthcare): differential in-gel analysis (DIA) and biological variation analysis. Matched spots considered for the analysis were filtered according to the following criteria: their presence was detected in at least 80% of spot maps, the average ratio selected was ≤ 1.3 or ≥ 1.3, and unpaired Student's t-test. Protein expression changes were statistically supported by a p < 0.05. In addition, an unsupervised multivariate analysis was carried out by the extended data analysis module, and based on differential protein expression profiles, the overall 2 https://clinicaltrials.gov/ct2/results?cond=cancer&term=hsp90+inhibitor&cntry=&state=&city=&dist=. In. 2020. correlations occurring between the spot maps were visualized using a principal component analysis (PCA).

Protein identification by liquid chromatography-mass spectrometry tandem (LC-MS/MS)
Protein extracts were separated on preparative gels and recovered from the gels for identification by LC-MS/MS as previously described [9]. Mass fingerprinting searching was carried out in Swiss-Prot/TrEMBL database using Mascot (Matrix Science Ltd., London, UK).

Immunoblotting
Protein extraction after 48 h of cell culture and immunoblotting was performed as previously described [9]. Western blots were quantified using IMAGEJ software (Rasband, W.S., U.S., National Institutes of Health, Bethesda, Maryland, USA . Secondary antibodies were purchased as follows: polyclonal swine anti-rabbit immunoglobulins/horseradish peroxidase (HRP)-linked IgG secondary antibody conjugate and polyclonal rabbit anti-goat immunoglobulins/HRP conjugate from Dako Products, Agilent (Santa Clara, CA, USA); rabbit polyclonal antimouse IgG H&L (HRP) conjugate from Abcam.

Protein network analyses
The MS-identified proteins, whose expression changes, were analyzed by (ingenuity pathway analysis) IPA software (GeneGo Inc., St. Joseph, MI, USA). IPA includes a manually annotated database of protein interactions and metabolic reactions obtained from the scientific literature. The networks were graphically visualized as hubs (proteins) and edges (the relationship between proteins).

Cell proliferation assay and drug combination studies
cis-Diammine Pt (II) dichloride (CDDP) was provided by Sigma-Aldrich (St. Louis, MO, USA); carboplatin was provided by Teva (Milan, Italy). Stock solutions were prepared in PBS. 17AAG, a derivative of geldanamycin (GM), was provided by StressMarq Biosciences. Stock solutions were prepared in DMSO. Ganetespib was provided by MedChem Express (Sollentuna, Sweden), and stock solutions were prepared in DMSO.
Cell proliferation was measured in 96-well plates in cells untreated and treated with 17AAG, ganetespib and CDDP as single agent or in combination. Cell proliferation was measured using a spectrophotometric dye incorporation assay (sulforhodamine) [12]. Drug combination studies were based on concentration-effect curves generated as a plot of the fraction of unaffected (surviving) cells vs drug concentration after 72 or 96 h of treatment. Synergism, additivity, and antagonism were quantified after an evaluation of the combination index (CI), which was calculated by the Chou-Talalay equation with CALCUSYN software (Biosoft, Cambridge, UK), as described elsewhere [13]. A CI ≤ 0.8, CI ≤ 0.9, CI = 0.9-1.1, and CI > 1.1 indicated a strong synergistic, a synergistic, an additive, or an antagonistic effect, respectively. The dose reduction index (DRI) determines the magnitude of dose reduction allowed for each drug when given in combination, compared with the concentration of a single agent that is needed to achieve the same effect.
For dose-response curves, primary EOC cells were seeded in 96-well culture plates and treated with increasing doses of CDDP with or without ganetespib

Colony formation assay
Five hundred cells per well were plated in 6-well plates. Cells were treated concomitantly or with 24 h of delay with different concentrations (IC 10 -IC 25 ) of ganetespib and CDDP and were allowed to grow for about 10 days until colonies could be clearly seen. Cell culture plates containing colonies were gently washed with PBS and fixed and stained with crystal violet 0.5%, methanol 20% for 30 min. Excess stain was removed by washing repeatedly with distilled water. Crystal violet was eluted with 100% methanol. Absorbance was read at 595 nm. All the procedures were done at room temperature.

Apoptosis assays
Apoptosis and necrosis were identified by flow cytometry analysis. Briefly, 3 × 10 5 cells were seeded and after 48 or 72 h of treatment were stained with Annexin V-FITC and propidium iodide, following the manufacturer's instructions (MACS; MiltenyiBiotec, Bergisch Gladbach, Germany) and analyzed by flow cytometry.

Statistical analysis
All experiments were performed at least three times. Statistical significance was determined by one-way ANOVA, or by a two-tailed unpaired Student's t-test, as indicated and a P < 0.05 was considered to be statistically significant. All statistical evaluations were performed with GRAPHPAD PRISM 8 (Ritme Informatique, Paris, France).

Proteomic analysis identified several differentially expressed proteins between Ptresistant and parental EOC cells
To investigate the mechanism by which cells acquire Pt resistance, we took advantage of 2-D DIGE followed by LC-MS/MS to compare the protein expression profile of the three CDDP-resistant cells (TOV- 112D Pt-res pool, OVSAHO Pt-res pool, and MDAH-2774 Pt-res pool) that we have previously established and characterized [6][7][8]. Biological replicates of parental cells or pooled resistant populations (two different pools for each Pt-res cell line) were analyzed by 2-D DIGE proteomic approach performed as previously described [14]. Approximately 1200 protein spots were consistently detected, quantified, normalized, and intergel-matched in each 2-D DIGE experiment. In detail, for each condition, we evaluated biological replicates (in quadruplicate) and performed reverse-labeled by the fluorescent cyanine dyes Cy3 and Cy5. The Cy2 dye was used as the internal standard, generated from an equal combination of all the samples tested in the same experiment, leading to a proper quantitative comparison of proteomic variations with statistical confidence. PCA showed a good experimental reproducibility of 2-D DIGE results as demonstrated by the close relation between the biological replicates clearly clustering, for each cell model, into two groups corresponding to Pt-res pool cells (red) vs parental cells (black; Fig. 1A). To be included in the analysis, the differentially expressed protein spots should be detected in 80% of the spot maps (replicates) and should display a statistically significant expression variation of at least 1.3-fold (P < 0.05; Fig. 1A).
Among the three EOC Pt-res/parental models, we did not identify any common differentially expressed protein. However, several identified proteins belonged to the same family or shared common functional classes (Fig. 1B), and eight of them were differentially expressed in at least two cellular models: HSP7C, HNRPL, PHB, prelamin-A/C (LMNA), ATP synthase subunit, mitochondrial (ATPA), elongation factor Tu mitochondrial (EFTU), heterogeneous nuclear ribonucleoprotein K, and D-3-phosphoglycerate dehydrogenase (highlighted in gray in Tables S1-S3). The different expression of HSP7C, HNRPL, PHB, and LMNA was reported as 3D view spot quantification and validated by western blot with a high concordance between the two analyses ( Fig. 1C).
Same results were obtained for differentially expressed proteins specifically altered in each cell model, overall confirming the quality and strength of the used proteomic approach (Fig. S1).

HSP90 is a central hub of the differential expressed proteins network in Pt-resistant EOC cells
Next, to recognize the regulatory pathways in which the identified proteins were involved and hence to elucidate their biological functions and relationship with the Pt resistance, we interrogated the IPA software. We first searched for direct interactions and restricted the analysis to the eight proteins shared by at least two cellular models. Significantly, all eight proteins are associated in a single pathway related with cancer, and the chaperone HSP90 emerged as a relevant central hub in the network ( Fig. 2A). Moreover, the IPA analysis also showed a relevant relationship between almost all the identified proteins in each EOC Pt-res model with 21/22 of the identified proteins closely clustering together in TOV-112D (Fig. S2A), 20/20 in MDAH (Fig. S2B), and 19/19 in OVSAHO model (Fig. S2C). Interestingly, HSP90 emerged again as a main hub in all the three different networks. Furthermore, when we included all the proteins identified in the three cellular models, HSP90 maintained the role of a central hub in a single network (Fig. S2D).
Based on these results, we next evaluated the expression of total HSP90α in parental and Pt-res models showing their increased expression in all tested pools (Fig. 2B). Moreover, using two additional TOV-112D Pt-res clones (clones #2 and #7) recently generated [7,8] we further confirmed HSP90 overexpression (Fig. 2B).
Conversely, the mRNA expression of both HSP90α and HSP90β isoforms was not consistently changed in Pt-res cells, suggesting that HSP90 expression is mainly regulated at post-transcriptional level (Fig. S3). Moreover, HSP90 expression is not modulated by acute CDDP treatment neither in parental nor in Ptres cells (Fig. 2C).

HSP90 inhibitors potentiate CDDP antitumor effect and reverse Pt resistance
The high levels of HSP90, and particularly of HSP90α, in Pt-res relative to parental cells suggested that this chaperone protein might represent a potential therapeutic target in Pt-res EOC setting. In order to test this hypothesis, taking advantage of CRISPR-Cas9 system, we generated knockout HSP90α cells (Pt-res KO#2) from TOV-112D Pt-res clone #7 (Fig. 3A). HSP90α knockout completely reversed CDDP resistance, as demonstrated by colorimetric cytotoxicity (Fig. 3B) or colony assay (Fig. 3C). Next, we investigated two HSP90i, 17AAG and ganetespib, both in advanced clinical studies. As single agents, ganetespib showed a greater antiproliferative potency than 17AAG, being effective on the majority of the examined cells with IC 50 values in the low nanomolar range (Table S4). Interestingly, in the majority of the cases, Pt-res cancer cells were also more resistant to both the HSP90i compared to parental controls.
We then investigated the combination of CDDP with 17AAG (Table S5) or ganetespib (Tables 1 and S6) using equipotent doses (50 : 50 cytotoxic ratio) or   Tables S1-S3. β-Actin and GAPDH were used as loading controls. Western blot quantification was performed by IMAGEJ software.
an excess of either agent (25 : 75 or 75 : 25 cytotoxic ratio) and evaluated the CI values calculated at 50% (CI 50 ) or 75% (CI 75 ) of cell lethality, in TOV-112D and MDAH parental cells and in Pt-res pools or single clones. By considering strong synergism CI ≤ 0.8, synergism CI ≤ 0.9, additivity CI > 0.9 and ≤ 1.1, and antagonism CI > 1.1 [15,16], we obtained consistent synergistic antiproliferative effects at equipotent doses and with an excess of HSP90i, in both parental and resistant cells, with slight better results observed by combining CDDP with ganetespib, the latter being effective even at lower doses in combination treatments.
The synergistic interaction between 17AAG or ganetespib and CDDP was also confirmed by the evaluation of the DRIs, which represent the order of magnitude (fold) of dose reduction obtained for the IC 50 (DRI 50 ) of each agent in combination vs single drug treatments (Tables 1 and S5-S6). Significantly, in hTERT-immortalized human foreskin fibroblast cell line BJ-hTERT, we observed antagonist effects (Table S6), suggesting a selective synergistic effect of the double combination on tumor cells.
For ganetespib/CDDP combination, we also compared simultaneous vs sequential (with 24 h delay between the two agents) treatment, demonstrating better effect in concomitant treatment schedule or when ganetespib preceded CDDP. Notably, in these setting in Pt-res cells we observed a clear potentiation of CDDP effect with DRI ranging from a minimum of   1.4 up to almost 5 (Table 1). Indeed, concomitant treatment of TOV-112D Pt-res cells with ganetespib and CDDP, demonstrated a complete reversion of the CDDP resistance (Fig. 4A) and a significant synergistic reduction of cell viability, compared to either agent alone (Fig. 4B). Moreover, we demonstrated that concomitant ganetespib/CDDP or sequential treatment of ganetespib 24 h before CDDP, using IC 72 h 25 of either agent, completely suppressed colony formation in parental (Fig. 4C) and Pt-res TOV-112D cells (Fig. 4D,E), compared with single agents, or CDDP treatment preceding ganetespib. This effect was also confirmed at IC 72 h 10 dosages, particularly in parental cells (Fig. S4A,C).
Furthermore, since TOV-112D Pt-res cells were cross-resistant to doxorubicin [6] (Table S4), another chemotherapeutic drug commonly used to treat EOC patients, we evaluated concomitant treatment of this agent plus ganetespib, also demonstrating synergistic interaction with CI 50 lower than 0.8 and DRI for doxorubicin up to 5 (Table S7), suggesting that HSP90i may be able to overcome a common mechanism of acquired drug resistance.
Finally, we observed that the potentiation of CDDP antitumor effect and the reversion of Pt resistance by HSP90α knockout or pharmacological inhibitors such as ganetespib was due to an increased apoptosis as demonstrated by the evaluation of Annexin V positivity and of cleaved PARP1 and caspase-3 expression both in TOV-112D and in MDAH models (Figs 5A-E and S5A-C). The observed increased apoptosis observed was accompanied by a clear induction of γH2AX, indicating increased DNA damage (Figs 5C,E and S5A). These data support the possibility that the intrinsic apoptotic pathway was activated for an increased DNA damage. Accordingly, with respect to control or single-agent treatments, the ganetespib plus CDDP combination induced a clear up-regulation of the pro-apoptotic BAX with a down-regulation of the antiapoptotic Bcl-2 protein and an overall increase of the BAX/Bcl-2 ratio (Fig. 5C). Interestingly, we also observed in TOV-112D parental and Pt-res cells a decrease of HSP90α protein expression induced by ganetespib/CDDP combination compared with untreated or single agents-treated cells (Fig. S6A).
The collected data suggested that Pt-res EOC cells were particularly sensitive to the ganetespib/CDDP. We thus tested this possibility using primary cells obtained from ascites of two Pt-sensitive EOC patients and from one patient that developed Pt resistance over the time (Fig. 6A). Cultured cells responded to CDDP treatment as predicted by the Pt sensitivity of the corresponding patients (Fig. 6B). The use of ganetespib effectively increased the activity of CDDP in both Pt-sensitive and Pt-res primary cultures with some differences. In Pt-sensitive cells (i.e., OV102 and OV202), ganetespib increased the activity of low doses of CDDP (1 µM) while it had no effects when high doses of CDDP were used (25 µM). In these primary cultures at the dose used (25 nM), ganetespib had no or minor effects. (Fig. 6C left graphs). In Pt-res cells (i.e., OV199 and OV200), ganetespib used alone significantly reduced cell survival and this activity prevailed on the one of low doses of CDDP (1 µM). Importantly, the addition of ganetespib to a high dose of CDDP (25 µM) significantly improved CDDP efficacy (Fig. 6C  right graphs), confirming that the combination therapy is effective also on ex vivo cultures of patient-derived tumor cells.

In vivo synergistic antitumor effect of CDDP and ganetespib combination
Finally, we used the CDDP/ganetespib combination to treat mice-bearing xenograft tumors formed by TOV-112D Pt-res pool 2 cells. When tumors reached~500 mm 3 of volume, mice were randomized in four treatment groups: control, CDDP, ganetespib, and CDDP plus ganetespib. Significantly, CDDP/ganetespib-treated group was the only one that produced a statistically significant TV reduction compared with all other groups (Fig. 7A,B). Tumors in vehicle-treated, CDDP or ganetespib alone treated mice, grew rapidly and reached the endpoint size within 20 days, when the average TV of single agents-treated groups was even slightly increased compared with controls. We observed a modest decrease of mice body weight in the combination treatment; however, it was less than 10% and was not paralleled by other signs of acute or delayed toxicity (Fig. 7C).
The synergistic effect of the CDDP/ganetespib combination treatment was confirmed by the resulting TGD that reached a peak of about 35% indicating that the mean rate of tumor growth was more than three-fold higher in the controls or single agents-treated groups, than the combination setting (Fig. 7D).
Western blot analyses on tumor lysates demonstrated that the combination treatment induced a clear reduction in HSP90α and of total HSP90 levels that was accompanied by a prominent increase of PARP1 and caspase 3 cleavage and of γH2AX expression, compared with the other groups, confirming in vivo what observed in vitro (Fig. 7E). Conversely, in xenograft model of parental TOV-112D cells, by using carboplatin and ganetespib, we showed that both agents   were effective in inhibiting tumor growth in monotherapy and that their combination did not produce any clear potentiation of the antitumor effect (Fig. S6B,C).

Discussion
Despite numerous efforts to develop novel therapeutic approaches, the treatment of refractory/resistant EOC patients after first-line Pt treatment remains a clinical challenge. Different mechanisms of Pt resistance have been described, including alterations in intracellular concentration due to influx or efflux defects as well as increased DNA repair or survival pathways [17]. Indeed, several attempts have been undertaken to bypass these mechanisms with relative scarce success, mainly due to the multimodal nature of the resistance [17]. The complex and heterogeneous genetic profile of EOC, particularly of HGSOC, limited the stratification of patients subtypes also preventing the identification of distinct genetic features predicting Pt resistance [5].
In the present study, by using a proteomic approach followed by a bioinformatic analysis using IPA software, we demonstrated that HSP90 emerged as a central hub in a protein network, associated with 'cancer', connecting the identified differentially expressed proteins in all three Pt-res models, compared with parental controls. HSP90 was overexpressed in Pt-res compared to parental cells, mostly due to HSP90α    isoform protein overexpression, further suggesting a potential role in the mechanism of resistance. Indeed, we demonstrated that HSP90α knockout sensitizes Ptres cells to CDDP treatment. Most importantly, by targeting HSP90 with specific HSP90i, we demonstrated, in vitro, ex vivo in primary cultures derived from the ascites of Pt-res EOC patients and in vivo in Pt-res EOC xenograft models, synergistic antitumor effect in combination with CDDP. Stress-associated phenotype, including proteotoxic stress, recently emerged as an additional hallmark of cancer, also highlighting the critical role in cancer cells of chaperons such as HSP90 that is involved in protein homeostasis [18]. HSP90 is a family of ATP-dependent molecular chaperones that with the help of co-chaperones modulate signal transduction and stress responses by regulating the folding, stability, and activity of client proteins, including several oncogenic molecules (i.e., HER2, estrogen receptors, AKT, MET, VEGFR, BRCA-1, MMP2) [19]. Thereby, HSP90 is also involved in many other hallmarks of cancer, including cell growth, survival, angiogenesis, metastases, and immune response [19,20]. Indeed, high HSP90 expression levels have been reported in many tumors and correlated with metastasis status and chemo-resistance as well as recently also with immune refractory status [19][20][21]. Overexpression of HSP90 has been also reported in ovarian carcinoma where it was linked to the FIGO stage of the disease [22]. Moreover, by using an ovarian cancer expression library for a serological screening, HSP90 emerged as a specific tumor antigen for this disease [22].
Notably, the two largest groups of identified proteins in all three Pt-res EOC models were those under the functional clusters 'stress response, chaperones and protein fate' and 'nucleic acid processing and DNA damage'. An additional functional cluster of identified proteins is EMT, cancer stem cells, and cytoskeleton organization", and included common biological features we have previously attributed to our Pt-res EOC models [6]. Interestingly, numerous identified proteins and particularly the eight deregulated in at least two Pt-res models have been directly connected with Pt resistance and/or ovarian cancer aggressive features [23][24][25][26][27]. For example, among those proteins overexpressed in our PT-res EOC models: molecular chaperones, including HSP90 or proteins involved in drug detoxification such as peroxiredoxins, (PRDX) were reported up-regulated in PT-res cancer cells and/or induced by CDDP treatment [23]; similarly, PRDX2, the chaperone GRP75, and the protein involved in stress response EFTU were proposed as candidate biomarkers of drug-resistant disease in ovarian cancers [24]; furthermore, PHB, involved in cancer proliferation and metastasis, and PRDX4 were significantly found increased in ovarian cancer tissues from chemotherapy nonresponders patients [27]. Conversely, among those proteins whose expression we have found reduced in Pt-res models: loss of nuclear lamina protein LMNA in EOC was associated with metastasis and poor prognosis [25], while reduced secreted expression of the dehydrogenase 3-PGDH was identified among candidate circulating CDDP-resistant biomarkers from EOC cell secretomes [26].
In mammalian cells, there are two major HSP90 isoforms that are the constitutively expressed: HSP90β and the stress-inducible HSP90α, sharing a conserved ATP-binding domain targeted by most HSP90i [28,29].
HSP90α protein is expressed at differing levels in a tissue-specific manner and is specifically up-regulated in several cancers, while the corresponding HSP90AA1 gene is not altered in the majority of tumors [28]. In line with our data, it was previously demonstrated that overexpression of HSP90α in ovarian cancer cells decreased the chemosensitivity to CDDP [30].
Interestingly, recently it was pointed out that tumor addiction to a chaperon like HSP90 is due to its level of connectivity with other chaperons and/or to the hub role played within a network of proteins crucial for cancer cells homeostasis, rather than to the overexpression of the protein itself [31]. In other words, due to the dynamic nature of the binding with client proteins in order to modulate the plasticity of signaling pathways, only when HSP90 is hyper-connected and fully integrated with client proteins and chaperone machinery, then its targeting with specific inhibitors is therapeutically effective [31]. Indeed, a correlation between HSP90 isoform levels and HSP90i sensitivity in cancer cells is debated [32]. On this regard, although we found in all tested Pt-res cells HSP90α and total HSP90 protein overexpression compared with controls, HSP90 was not identified by the proteomic study as differentially expressed, probably because the magnitude of level change was not prominent and/or homogenous. Yet, we propose that the hub role of HSP90 highlighted in Pt-res cells was more biologically relevant than the mere protein overexpression and might explain the clear synergistic interaction we observed between HSP90i and CDDP in Pt-res EOC cells. Interestingly, data-mining meta-analyses integrating the results from multiple siRNA screening in EOC also identify HSP90 as a central hub and an effective therapeutic target [33]. Notably, we have also discovered HSP90 as a critical hub within a network of differentially expressed proteins identified in a Pt-res NSCLC isogenic model [9], supporting the possibility that the same mechanism is present in different tumor types. Indeed preliminary results showed also in this Pt-res NSCLC model a synergistic antitumor interaction between HSP90i and CDDP 1 .
Classical HSP90i act by competitively binding the ATP-binding site of HSP90, consequently halting the progression of the protein folding machinery, leading to the degradation of most client proteins [34]. Geldanamycin (GA), a large naturally occurring compound and its ansamycin derivatives 17AAG and 17DMAG were the first HSP90i to be tested [35,36]. In accord with our observation previous in vitro studies reported that the combination of GA, 17DMAG, or 17AAG with CDDP remarkably inhibited cell proliferation and partially reverse the drug resistance of CDDP-resistant SKOV-3 ovarian cancer cells [37]. Here we add the new notions that ganetespib is more effective than 17AAG in improving CDDP efficacy and prove for the first time that the combination therapy is active in vivo in Pt-res ovarian cancer models and in primary ovarian cancer cells derived from the ascites of Pt-res ovarian cancer patients. Preclinical studies have shown that ganetespib has greater potency than first-generation inhibitors such as 17AAG in several cancers [38][39][40] also in combination therapy (i.e., carboplatin, taxanes, and etoposide), with a better therapeutic index and no evidence of the cardiac, ocular or liver toxicity observed with other HSP90i, thus emerging as a promising anticancer agent being tested in several clinical trials [39,[42][43][44]. These data are in accord with our observation that ganetespib was able to revert the cross-resistance of Pt-res cells to doxorubicin, confirming the critical role of HSP90 in conferring a complex prosurvival chemo-resistant phenotype, likely by regulating tumor apoptosis. Of note, the lack of synergistic interaction between ganetespib and CDDP observed in normal human fibroblasts suggested a selective action on tumor cells and thus a good therapeutic index of the combined approach.
Our data support the possibility that the multimodal mechanisms of Pt resistance acquisition in our three EOC cell models, that differ in histology and genetic backgrounds, are commonly dependent on the addiction to HSP90 hub function, thereby rendering resistant cells particularly vulnerable to the combined HSP90i/CDDP approach. Interestingly, the lack of a clear synergistic interaction in vivo xenograft parental EOC models of ganetespib plus Pt compound as compared with results obtained in Pt-res cells, confirmed, at least in part, this hypothesis. Anyhow to our knowledge, this study is the first to show the reversion of CDDP resistance and the synergistic interaction of ganetespib with CDDP, both in vitro and in vivo in Ptres ovarian cancer models, providing a rationale to clinically explore this combination to rechallenge refractory/resistant EOC patients.
In unselected cancer patients, the clinical benefit of HSP90i has been modest so far, however in selected populations, with distinct molecular features and/or in refractory/resistant subgroups a promising activity has been shown [44,45]. Ganetespib has been evaluated in combination with weekly paclitaxel in Pt-res ovarian cancer patients and was generally well-tolerated with no DLTs observed [46]; however, no survival benefit was demonstrated in a randomized phase 2 trial evaluating this combination regimen vs paclitaxel alone [47]. Notably, a recent designed randomized multiarms phase 2 study, not yet recruiting, will evaluate the potential of ganetespib in combination with carboplatin in Pt-sensitive ovarian cancer patients (relapse > 6 months after previous Pt-based treatment) vs carboplatin plus paclitaxel or plus gemcitabine 3 .
In our opinion, strategies to identify HSP90 addiction within tumor cells could help in patient's stratification/selection. In this regard, changes in the expression levels, the composition of chaperone complex, or the level of interaction with other chaperones/ clients, as well the cellular location of HSP90, could be considered. Interestingly, tumor cells constitutively secrete HSP90 and recent studies showed that the plasma level of HSP90α correlates with the pathologic stage of cancer in patients, suggesting the potential of circulating HSP90 as predictive biomarker [29,48]. Similarly, a gene expression classifier that predicts sensitivity to HSP90i in a subgroup ovarian and breast cultured cells, cell line-derived xenografts, and PDX models, was recently generated and could be explored in the frame of translational clinical studies aiming at testing the feasibility and efficacy of HSP90i plus Pt in Pt-res EOC patients [32].

Conclusion
In conclusion in this study starting from an unbiased proteomic approach, we identified HSP90 as a central druggable target in Pt-res EOC cells. We then showed that the HSP90i ganetespib has a strong synergistic effect with CDDP in Pt-res EOC cells in vitro, ex vivo, and in vivo. Since the association between ganetespib and chemotherapy had an acceptable safety profile in cancer patients, this study has an immediate 1 Lombardi R., Pucci B, Addi L and Budillon A, unpublished observations. 3 https://clinicaltrials.gov/ct2/show/NCT03783949?term=ganetespib&cond=Ovary+Cancer&draw=2&rank=1. In. 2020. translational relevance and provides a strong rationale of an effective therapeutic strategy for the treatment of Pt-res EOC patients for whom we still do not have valid therapeutic options.

Supporting information
Additional supporting information may be found online in the Supporting Information section at the end of the article. Fig. S1. Validation by Western blot of protein identified in the cellular models as differentially expressed in the 2-D DIGE LC-MS/MS analysis. Fig. S2. Ingenuity Pathway Analysis of all identified proteins. Fig. S3. mRNA expression in parental and Pt-res EOC cell models. Fig. S4. Clonogenic assay of TOV-112D and TOV-112D Pt-res cells. Fig. S5. Pro-apoptotic effect of CDDP and/or ganetespib in Pt-MDAH Pt-res and in TOV-112D parental cells. Fig. S6. Effect of CDDP and/or ganetespib on HSP90α expression in TOV-112D and TOV-112D Ptres cells and on tumor growth of TOV-112D parental cells xenograft model. Table S1. Differentially expressed proteins identified by mass spectrometry in TOV-112D model. Table S2. Differentially expressed proteins identified by mass spectrometry in MDAH model. Table S3. Differentially expressed proteins identified by mass spectrometry in OVSAHO model. Table S4. Sensitivity of EOC cell lines to single agent treatments. Table S5. Combination index (CI) and dose reduction index (DRI) values for 17AAG and CDDP combination treatment accordingly to different cytotoxic ratio. Table S6. Combination index (CI) and dose reduction index (DRI) values for ganetespib and CDDP combination treatment accordingly to different cytotoxic ratio. Table S7. Combination index (CI) and dose reduction index (DRI) values for ganetespib and doxorubicin combination treatment at equitoxic ratio.