Hsp90 C-terminal domain inhibition enhances ferroptosis by disrupting GPX4-VDAC1 interaction to increase HMOX1 release from oligomerized VDAC1 channels.

1
Introduction
HCC, the most prevalent form of primary liver cancer, poses a significant global health challenge due to the high mortality and morbidity rates [[1], [2], [3], [4]]. Recently, ferroptosis has emerged as a promising strategy for HCC treatment, as it can directly kill cancer cells and enhance drug sensitivity [[5], [6], [7], [8]]. Ferroptosis is a programmed cell death defined by iron-dependent lipid peroxidation. Dysregulation of iron metabolism and lipid balance is an important feature of HCC, making tumor cells susceptible to ferroptosis induction. Several strategies have been studied to exploit ferroptosis for HCC treatment. One approach is to modulate the expression or activity of key molecules involved in ferroptosis pathways. For instance, inhibiting GPX4, an essential enzyme in the antioxidant system, can sensitize HCC cells to ferroptosis and enhance the efficacy of chemotherapy or radiotherapy [[9], [10], [11]]. Another aspect is the regulation of iron metabolism. Intracellular iron promotes the Fenton reaction, producing cytotoxicity and leading to ferroptosis [[12], [13], [14], [15]]. In advanced HCC treatment, donafenib promotes iron accumulation and further ferroptosis by up-regulating iron metabolism-related protein HMOX1 [16].
Hsp90 has emerged as a significant factor influencing ferroptosis. Hsp90 is a powerful molecular chaperone regulating the folding, stability, and function of the client proteins, including those linked to iron metabolism and oxidative stress responses [17]. Hsp90 occupies a pivotal position in the complex regulatory network of cells, currently, Hsp90 exhibits a dual role in ferroptosis. Overexpression of Hsp90 promotes dephosphorylation of mitochondrial fission protein DRP1 at the Ser637 site, thereby upregulating ACSL4 levels, altering mitochondrial morphological changes, promoting lipid peroxidation, and further enhancing Erastin-induced ferroptosis in glioma cells [18]. Gambogic acid suppresses tumor cell tolerance to thermal stress by inhibiting Hsp90 levels, depletes glutathione (GSH), and downregulates GPX4 levels, thereby weakening cellular antioxidant capacity, disrupting intracellular redox homeostasis, and promoting ferroptosis [19]. Previous research has indicated that the Hsp90 N-terminal domain (NTD) inhibitor 17-AAG [20] can increase GPX4 levels to suppress ferroptosis induced by Timosaponin AIII [21]. However, based on the mechanism by which Hsp90 can maintain protein stability as a molecular chaperone, the different domains of Hsp90, particularly the C-terminal domain (CTD), on the stability of GPX4 and the subsequent impact on ferroptosis remain uncertain. Given the well-established significance of ferroptosis in HCC progression and the potential role of Hsp90 as a GPX4 chaperone, it is of utmost importance to investigate the specific mechanisms underlying the interaction between Hsp90 and GPX4 in HCC.
Mitochondria, as the primary energy-producing organelles, are highly susceptible to oxidative stress and play an essential role in governing ferroptosis. [[22], [23], [24]]. VDAC1 functions in transporting ions and metabolites across OMM to regulate cellular homeostasis and could exist in different oligomeric states under oxidative stress conditions [[25], [26], [27]]. However, the precise relationship between VDAC1 oligomerization and oxidative modifications such as carbonylation remains underexplored. In previous research, we reported that Hsp90α CTD inhibition reduced mono-ubiquitination at the K274 site of VDAC1, which enhances its oligomerization [28]. Building upon this, our current study offers a new perspective by demonstrating that the VDAC1-K274 mutation impairs the GPX4-VDAC1 interaction and further enhances VDAC1 oligomerization through a redox-dependent mechanism by promoting its carbonylation. Recent study illustrated that the VDAC1 oligomerization inhibitor VBIT-12 can protect mitochondria from ferroptosis [29]. CISD1 affected the VDAC1 oligomer pore opening and negatively regulated ferroptosis by inhibiting iron uptake and mitochondrial lipid peroxidation [30,31]. Yet, the exact function and mechanism of ferroptosis through VDAC1 oligomerization, especially whether its pores mediate the transport of ferroptosis-related proteins is still elusive.
Additionally, iron overload is also an important hallmark of ferroptosis. HMOX1 is responsible for cleaving heme to free ferrous iron, whereas iron released may induce ferroptosis [[32], [33], [34]]. Here, we identified that Hsp90α CTD deletion and Novobiocin (NB), a C-terminal inhibitor of Hsp90, triggered the ferroptosis of HCC cells with the degradation of GPX4 through autophagy, leading to the decreased interaction between VDAC1 and GPX4. This disruption promotes VDAC1 carbonylation and consequently oligomerization, facilitating the release of HMOX1 from mitochondria and resulting in iron overload and ferroptosis.

2
Materials and methods
2.1
Cell culture
HepG2, Huh7, MHCC97H were purchased from the American Type Culture Collection (Manassas, VA, USA), and has been authenticated for mycoplasma contamination and short tandem repeat profile analysis. In an incubator set at 37 °C with 5 % CO2, cells were grown in DMEM combined with 10 % fetal bovine serum and 100 U/ml penicillin-streptomycin combination (Gibco, USA).

2.2
Antibodies and reagents
Antibodies against HMOX1 (70081S), SLC7A11 (12691S), FTH1 (4393S), COXⅣ (4850S), β-Tubulin (2128s), LC3 (4108S) and Hsp90α (8165s) were acquired from Cell Signaling Technology (Danvers, MA, USA). Antibodies against GPX4 (ab125066), DNP (ab178020), and VDAC1 (ab154856) were purchased from Abcam (Cambridge, MA, USA). Antibodies against β-Actin (66009-1-Ig), GPX4 (67763-1-Ig) and P62/SQSTMI (18420-1-AP) were bought from Proteintech (Rosemont, IL, USA). Antibodies against HA (H36630) Sigma (St Louis, MO, USA). Antibodies against 4-HNE (HY–P81208) MCE (Monmouth Junction, NJ, USA). Antibodies against LAMP1 (sc-20011) Santa Cruz Biotechnology, (CA, USA). STA9090 (S1159), NB (S2492), Erastin (S7242), ferrostatin-1 (S7243), deferoxamine (S5742), MG132 (S2619), CQ (S6999) were obtained from Selleck Chemicals (Houston, TX, USA). The EGS (21565) were acquired from Thermo Fisher Scientific (Waltham, MA, USA).

2.3
Western blot
Cell Lysis Assay kit (KGP2100) was purchased from Keygen Biotech (Nanjing, China). Cells were added the lysis buffer and lysed on ice for 5 min, scrape the cells with a cell hanger and transferred to a 1.5 ml centrifuge at 4 °C at 15 min × 13000 rpm for centrifugation, and the supernatant was through Detergent Compatible Bradford Protein Assay Kit (P0006C) to detects protein concentration. For protein extraction from liver cancer tissues and xenograft tumor tissues, add 300 μl whole protein lysate for every 10 mg of tissues, thoroughly grind in a protein homogenizer, centrifuge under the same conditions as above, and take the supernatant to test the protein concentration. The protein was isolated using sodium dodecyl sulfate-polyacrylamide gel electrophoresis and then moved to a polyvinylidene fluoride (PVDF) membrane. Prepare the PVDF membrane by blocking it with 5 % BSA, followed by overnight addition of the primary antibody at 4 °C; add the secondary antibody for 1 h at room temperature, scan with the Li COR Odyssey infrared imaging system, and finally analyze with the Image J software.

2.4
Lipid peroxidation assessment
C11-BODIPY581/591(D3861) was bought from Thermo Fisher Scientific (Waltham, MA, USA). HCC cells (1x106) were seeded on 6-well dishes. After reaching the end of the treatment, cells were collected and C11-BODIPY581/591 was diluted with basal medium, treating the cells at a concentration of 5 μM for 30 min in the dark. The level of lipid peroxidation was assessed using flow cytometry (Accuri 6 cytometer, BD Biosciences) with 488 nm laser excitation and 510 nm laser emission.

2.5
Malondialdehyde (MDA) assay
The lipid peroxidation byproduct malondialdehyde (MDA) can react with thiobarbituric acid to produce a red compound that exhibits a peak absorption at 532 nm. A lipid peroxidation MDA assay kit (Keygen Biotech, Nangjing, China) was utilized to identify the accumulation of lipid peroxidation products in cells. Count 1x 106 cells in 1 ml precool IP lysis buffer, lyse on ice for 20 min, and shake vigorously for 30 s every 5 min during this period. After centrifugation, add the reagents to the reaction according to the protocol and take the supernatant to detect its absorbance at 532 nm wavelength.

2.6
Protein carbonylation
Protein carbonylation following the Western blot assay above, after blocking for 1 h with 5 % BSA blocking solution, the PVDF membranes were incubated 30mins in 2 mM HCL. Further, the protein reacted with 2,4-dinitrophenylhydrazine (20 mmol/L in 2 mM HCL) to derivatize into 2,4-dinitrophenylhydrazone, then probed with anti-DNP antibody overnight at 4 °C and the protein carbonylation was detected by densitometric analysis using Odyssey software.

2.7
Cross-linking assay
Cells were collected and washed with PBS (pH 8.0). The cells were then incubated in PBS (pH 8.0) containing the crosslinking reagent EGS (500 μM) at 37 °C for 15 min. Subsequently, crosslinking was terminated by treating with Tris (pH 7.5) at a concentration of 10 mM for 15 min at room temperature. After the cells were collected by centrifugation, 2X SDS loading buffer was added to lyse the cells. Western blot was used to detect the level of VDAC1 oligomerization.

2.8
Co-immunoprecipitation
Cells were lysed in NP-40 buffer (Keygen Biotech, Nangjing, China) for 10 min on ice; then the antibody was incubated in the cell lysate overnight at 4 °C. The Protein A/G Magnetic Beads were mixed with immunoprecipitation complex for 2 h. Collect immunoprecipitants and wash it with NP-40 lysis buffer three times and complexed with SDS loading buffer, then the sample was detected by Western blotting.
In the Co-IP process, the MOCK group refers to the experimental group in which no specific antibody is added during the immunoprecipitation step to assess non-specific binding. In the Input process, the MOCK group refers to the sample that has not undergone the IP process but is derived from the same cell lysate as other experimental groups.

2.9
Cell proliferation
5x104 cells were cultured in 96-well cluster plates. CCK-8 reagents were used to dilute at a ratio of 1:9; add CCK-8 reagent 100 μl per well in the dark for 2 h in a 37 °C, finally read the absorbance at 450 nm wavelength.

2.10
Animal studies
5-week-old BALB/cu mice were acquired from the Animal Center of Southern Medical University to construct xenograft tumor models. The xenograft tumor model was constructed by injecting HepG2 cells, the GPX4 stably overexpressing cells with a density of 5x106 cells into the back of limb of nude mice. For the intrahepatic tumor model, The GPX4 stably knockdown cells and the sh NC cells with GFP label (5x106) were injected into the left liver. The tumor size was observed by fluorescence imaging and the tumors were further processed for Immunohistochemistry and Western blot. Every animal study complies with the Institutional Animal Care and Use Committee's guidelines and passes the Southern Medical University Ethics Committee's review (SMUL2021053).

2.11
Patient tumor specimen
10 cases of hepatocellular carcinoma human tissue and corresponding 10 cases of adjacent tissue were collected from the Hepatobiliary Surgery Department of Nanfang Hospital. The tissues 1–2 cm away from the lesion were identified as adjacent tissues. Immediately separate a portion of the tissue, freeze, and store it in at −80 °C for subsequent Western blot analysis. Another part of the tissue was fixed and embedded in paraffin for subsequent immunohistochemical (IHC) detection. Pathological specimens are obtained without affecting their treatment and prognosis. All organizations have been reviewed by the Ethics Committee of Nanfang Hospital, and patients have also approved the acquisition and use of specimens (NFEC-2022-184).

2.12
Iron assay
FerroOrange (Dojindo, Japan) was utilized to ascertain the intracellular Fe2+ content. FerroOrange was diluted with PBS, and the cells were treated at a concentration of 1 μM and placed at 37 °C for 30 min, and then the fluorescence intensity was acquired by flow cytometry using 543 nm excitation wavelength and 580 nm emission wavelength.

2.13
Transfection and stable cell line generation
Professor Matthias P. Mayer of Heidelberg University presented the plasmids of Hsp90α-FL, HA-ΔC, HA-ΔN, and HA-(ΔN + ΔC) [35]. The cDNA encoding for the Hsp90α-FL (encoding amino acids 1–727) was then cloned into pIRESII vector, with the HA-tag at the C-terminal of the Hsp90α-FL sequence. Hsp90α-ΔC contains the sequence (encoding amino acids 1–548). Hsp90α-ΔN contains the sequence (encoding amino acids 284–727). Hsp90α-(ΔC + ΔN) includes the sequence (encoding amino acids 284–548). The HA-tagged sequence was cloned into pIRESII vector, with the tag at the M-domain for the Hsp90α-ΔC, Hsp90α-ΔN, and Hsp90α-(ΔC + ΔN) sequences. The plasmids of VDAC1-FL, K274R, GPX4-FL, U73A were purchased from Hanheng Bioengineering (Shanghai, China). GPX4-specific siRNA and GPX4-shRNA were chemically synthesized by Gene Pharma (Suzhou, China). Transfection was performed with the transfection reagent (Invitrogen, Carlsbad, CA, USA) for 48 h, the small interfering RNA oligonucleotides with sequence targeting GPX4 (Sense: 5′-GTGGATGAAGATCCAACCCAATT-3′, antisense: 5′-TTGGGTTGGATCTTCATCCACTT-3′) were acquired from Gene pharma (Shanghai, China).
To establish stable GPX4 knockdown cell lines, LV3 vectors (H1/GFP&Puro) containing the human GPX4-shRNA sequence (5′- CCGGGTGGATGAAGATCCAACCCAACTCGAGTTGGGTTGGATCTTCATCCACTTTTTG -3′) and the non-targeting control shRNA (sh NC) were transfected into HepG2 cells for 72 h. Following transfection, cells were selected with 2 μg/ml puromycin for 4 consecutive weeks to eliminate non-transfected cells, generating stable GPX4-knockdown and sh NC control cell lines. GPX4 knockdown (sh GPX4) efficiency was validated by Western blot and RT-qPCR analyses.

2.14
EdU assays
Analyze the cell proliferation capacity based on DNA synthesis through BeyoClick™ EdU Cell cell proliferation assay kit (Beyond Biotechnology, Shanghai, China). Confocal dish inoculated with cells was incubated for 2 h with 10 μM EdU in DMEM; After fixation and permeation treatment with 4 % paraformaldehyde and methanol, the Click Additive Solution was applied to each well for reacting for 30 min. DAPI staining was used for labeling the cell's nucleus. Subsequently, Images were acquired with Olympus FV1000 Confocal Laser Scanning Microscopy.

2.15
Three-dimensional cell culture
Cells (400 cells/well) were plated into 96-well ultra-low attachment spheroid plates (Liver Biotechnology, Shenzhen, China) at 37 °C in the presence of 5 % CO2. The spheroid cells were cultured with fresh media every two days and imaged every 24 h with microscope. The area was calculated with Image J.

2.16
Molecular docking
The 3D structure of Hsp90α (AlphaFold: P07900) was obtained from the AlphaFold database (https://alphafold.ebi.ac.uk/), while the crystal structures of GPX4 (PDB ID: 2OBI) and VDAC1 (PDB ID: 5JDP) were sourced from the Protein Data Bank (PDB, https://www.rcsb.org/). To explore the interactions between these proteins, predicted protein-protein docking was performed using the GRAMM-X public web server (http://vakser.compbio.ku.edu/resources/gramm/grammx/). Subsequently, the PDBePISA website (www.ebi.ac.uk/pdbe/pisa/) was utilized to provide information on docking residues, potential hydrogen bonds, salt bridges, and to calculate the free energy (ΔiG). Finally, the docking results were visualized using PyMOL.

2.17
Immunofluorescence analysis
Cells were plated into confocal dishes(1 × 105 cells/well) and treated with drugs or other reagents. Cells were fixed in 100 % methanol at −20 °C for 15 min, followed by incubation in 0.1 % Triton X-100 for 5 min, after the treatment was completed and cells were washed with PBS. Cells were then blocked with PBS containing 5 % bovine serum albumin (BSA) at room temperature for 1 h. Subsequently, cells were incubated with the specified primary antibodies overnight at 4 °C. After washing with PBS, cells were incubated with secondary antibodies at room temperature for 2 h. Nuclei were counterstained with DAPI. Fluorescence detection of protein colocalization was performed using an Olympus FV1000 Confocal Laser Scanning Microscope (Tokyo, Japan). Image-Pro Plus software was utilized to calculate the overlap coefficient between the two proteins, representing their colocalization extent.

2.18
Mitochondrial isolation
The Cell Mitochondria Isolation Kit (C3601, Beyotime Biotechnology, Shanghai, China) was utilized to isolate cytoplasmic and mitochondrial components from cells according to the protocol. The cells were collected and treated with mitochondrial separation solution on ice for 15 min, then homogenized 80 times using a glass homogenizer. The homogenate was centrifuged at 800×g for 10 min at 4 °C to remove cell debris. The supernatant was further centrifuged at 12,000×g for 10 min. The resulting supernatant represented the cytoplasmic fraction, while the pellet was resuspended and lysed with mitochondrial lysate to obtain mitochondrial proteins. The protein concentration was determined using a BCA Protein Assay Kit (P0012, Beyotime Biotechnology, Shanghai, China).

2.19
Mitochondrial ROS determination
Cells were collected and washed with PBS. Then, the cells were incubated with fresh culture medium containing 5 μM Mito Sox Red (Thermo Fisher Scientific, M36008) at 37 °C for 30 min. Following incubation, the cells were collected by centrifugation and resuspended in PBS. The Mito ROS levels were then assessed by flow cytometry.

2.20
Statistical analysis
Using Graphpad Prism with a Student's t-test to compare two groups that satisfy the normal distribution, and the variance was uniform. All statistical analyses are two-sided tests and presented as mean values ± SD of at least triplicate biologically replicated experiments.

3
Results
3.1
Hsp90 C-terminal domain inhibitor NB induces ferroptosis in HCC cells
The inhibition of Hsp90 affects multiple carcinogenic pathways by suppressing its chaperone activity. Therefore, inhibitors of Hsp90 are widely used in clinical trials for cancer [36]; Novobiocin (NB) and Ganetespib (STA9090) are effective inhibitors of the C- and N-terminal of Hsp90, respectively. Here, two inhibitors are used to treat HCC cell lines (Fig. 1A), the CCK-8 assay confirmed that ferrostatin-1 (Fer-1) and deferoxamine (DFO), as lipid ROS scavenger and iron chelating agent could alleviated NB-induced cytotoxicity but had no significant effect on STA9090 (Fig. 1B–S1A). Moreover, we found that NB and the ferroptosis inducer Erastin-treated HepG2 cells displayed mitochondrial shrinkage and reduction of mitochondrial cristae, which are characteristic of ferroptosis, while some large mitochondria were observed after STA9090 treatment (Fig. 1C). Although the C11-BODIPY 581/591 staining and the level of MDA illustrated that the lipid peroxidation following NB and STA9090 treatment was increased (Fig. 1D–S1B), the ferroptosis inhibitor can rescue lipid peroxidation caused by NB rather than STA9090 (Fig. 1E–S1C-E). Meanwhile, we detected the expression of key ferroptosis proteins in HCC cells upon the inhibitor of Hsp90 treatment. The data demonstrated that the NB decreased the protein levels of SLC7A11 [37], FTH1 [38,39], and GPX4 in HCC cells (Fig. 1F). However, ACSL4 [40] and FSP1 [41] were not significantly changed after NB treatment (Fig. S1F). These results suggest that NB-induced ferroptosis in HCC cells may be primarily linked to redox imbalance and iron metabolism-related pathways, rather than ACSL4-mediated fatty acid metabolism pathways or FSP1-mediated ferroptosis pathways, which protect cells from ferroptosis via NADPH-dependent radical scavenging. Due to the significant decrease in GPX4 levels observed after NB treatment, we conducted in vivo validation and found that GPX4 decreased in xenograft tumor tissues of NB group instead of STA9090 group (Fig. G–H). GPX4 is crucial in suppressing lipid peroxidation, particularly phospholipid peroxidation. Lipidomics data based on xenograft tumor tissues show that the expression of phospholipids containing polyunsaturated fatty acids (PUFA), especially PC and PE, was significantly increased after NB treatment (Fig. 1I-J). Elevated PUFA levels make the double bonds in phospholipid membranes more prone to oxidation, leading to the formation of lipid peroxides. These data indicated that ferroptosis occurs in HCC cells induced by Hsp90 C-terminal domain inhibitor NB.

3.2
High level of Hsp90α and GPX4 indicates poor prognosis in patients with HCC
As an important catalyst and regulator of redox reactions in cells, GPX4 plays a role in maintaining cell homeostasis [42]. Although the cytoplasmic isoforms have high amino acid sequence identity, stress-inducible Hsp90α is dimerized frequently compared to constitutively expressed Hsp90β, which is more important for enhancing protein folding and stability [43,44], so we focus on the relationship between Hsp90α and GPX4 in HCC tissues. The Cancer Genome Atlas (TCGA) data showed that GPX4 and Hsp90α protein levels were elevated in the HCC tissues. With the increase of cancer stage, the decline in the degree of tumor differentiation, and the increase of lymph node metastases, the GPX4 and Hsp90α expression increased (Fig. 2A and B). Furthermore, the patients in the high-level group of GPX4 and Hsp90α showed significantly shorter survival (Fig. 2C). Then, the HCC samples and normal tissues obtained from Nanfang Hospital were used for examining the protein level of GPX4, which is consistent with the TCGA databases, and GPX4 is positively correlated with Hsp90α (Fig. 2D and E). Additionally, the levels of GPX4 and Hsp90α were elevated in three distinct HCC cell lines in comparison to L02, a normal liver cell line (Fig. 2F). These results suggested that the high expression of GPX4 and Hsp90α with poor prognosis, and both may jointly participate in the complex pathophysiological mechanisms of HCC.

3.3
GPX4 binds to the C-terminal domain of Hsp90α
To further investigate the mechanism by which Hsp90α regulates ferroptosis, siRNA and plasmids targeting Hsp90α were transfected into the cells. It was found that silencing Hsp90α led to a suppression of GPX4 while the opposite results were attained when Hsp90α was overexpressed (Fig. 3A). The monomer of Hsp90 is composed of the NTD, middle domain (MD), and CTD. NTD has an ATP/ADP binding site, which is the specific binding target of STA9090. NB binds to the CTD and further inhibits the dimerization of Hsp90 and its function [45]. To study which domain of Hsp90 affects GPX4, we transfected the recombinant Hsp90α plasmids into HCC cells separately. The deletion of CTD resulted in a decrease in GPX4 protein levels (Fig. 3B and C). Molecular docking suggested that GPX4 binds to the CTD of Hsp90α (Fig. 3D–E, S2A). The immunofluorescence staining and Co-IP also confirmed that GPX4 was disassociated with Hsp90α in the treatment of NB rather than STA9090 (Fig. 3F–G, S2B). Furthermore, we found that CTD deletion inhibited GPX4-Hsp90α colocalization and led to a substantial reduction in GPX4 among the immunoprecipitates with anti-Hsp90α antibody, indicating that GPX4 was confirmed to bind to the CTD of Hsp90α (Fig. 3F–H).

3.4
Hsp90α C-terminal domain inhibition promotes GPX4 degradation via LC3/P62-mediated autophagy pathway
RT-qPCR suggested that NB elevated GPX4 mRNA levels in HCC cells and no significant difference was observed after CTD deletion (Fig. S3A). We next investigated whether inhibition of CTD has any effects on GPX4 protein stability. Cycloheximide (CHX), which can inhibit protein synthesis, was used to determine the protein turnover rate, and the GPX4 level was decreased in the NB and CTD deletion groups in a time-dependent manner, suggesting that the decrease in GPX4 content was dependent on protein degradation (Fig. 4A and B). Moreover, treatment with Chloroquine (CQ) to inhibit lysosomal degradation led to the stabilization of GPX4 protein levels, while MG132, a proteasome inhibitor, did not remarkably reverse the levels of GPX4 (Fig. 4C and D), indicating that the degradation of GPX4 is primarily mediated by lysosomal pathways rather than the ubiquitin-proteasome system. Additionally, the KEGG enrichment analysis illustrated that the Differentially Expressed Genes were closely associated with the lysosome pathway (Fig. S3B), and the LysoTracker staining showed that the fluorescence intensity was increased after NB treatment and CTD deletion, suggesting that the function of lysosome was hyperactive (Fig. 4E). The results of immunofluorescence indicated that the colocalization of Lamp1 and GPX4 in the cytoplasm increased after CTD inhibition (Fig. S3C). So, we hypothesized that the degradation of GPX4 might be through autophagy. Transmission electron microscopy revealed that after treatment with the Hsp90 C-terminal inhibitor NB, mitochondrial morphology changes associated with ferroptosis were observed, and an increased number of autophagosomes were detected (Fig. 4F). We illustrated the increased change of LC3II in response to CTD inhibition (Fig. S3D), as well as the LC3B and LAMP1 colocalization increasing (Fig. S3E). Besides, the combined use of CQ increased the LC3II protein level in NB-induced and CTD deletion transfected cells (Fig. 4G and H). Autophagic flux in HepG2 cells transfected with Tandem fluorescent-mRFP-GFP-MAP1LC3B-adenovirus showed increased autophagosomes and autolysosomes in CTD inhibition cells (Fig. 4I). Co-IP experiment confirmed the involvement of autophagy substrate P62 in the degradation of GPX4 (Fig. 4J and K). Moreover, results showed a decrease in lipid peroxidation when autophagy was blocked by CQ (Fig. S4A–B). Consistent with previous studies [46], RSL3 notably enhanced the expression of LC3II while blocking the levels of GPX4 and P62 (Fig. S4C). In summary, the CTD of Hsp90α is crucial for maintaining the stability of GPX4. Inhibition of CTD promotes the LC3/P62-mediated autophagy degradation of GPX4 and consequently facilitates ferroptosis.

3.5
Reduction of VDAC1-GPX4 interaction at K274 enhances VDAC1 carbonylation
Due to the indispensable role of GPX4 in maintaining redox homeostasis, it has been proven that low expression of GPX4 could promote protein carbonylation, common oxidative modification that can lead to structural instability and irreversible damage to protein function [47]. As mitochondrion is rich in iron and generates ROS, it is identified as a crucial area for ferroptosis. We speculate whether GPX4 will affect the local oxidative stress state of VDAC1. Here, we illustrated that NB significantly inhibited the interaction between GPX4 and VDAC1. Interestingly, we detected an increase in DNP protein levels in immunoprecipitates, indicating that NB treatment can promote carbonylation of VDAC1 (Fig. 5A and B).
Similar result was observed in GPX4 stably knocking down cells (Fig. 5C–S5A). Given the critical role of GPX4 enzyme activity in mitigating ferroptosis, we established an enzyme-dead mutant plasmid of GPX4 with a point mutation in the U73 active site from Sec-to-Ala (U73A) and found that transfection with this plasmid resulted in a significant increase in VDAC1 carbonylation. Importantly, treatment with NAC (N-Acetylcysteine), which can inhibit carbonylation [[47], [48], [49], [50]], effectively mitigated the increased carbonylation observed following U73A plasmid transfection (Fig. S5B). These results collectively demonstrate that GPX4 is indispensable for suppressing VDAC1 carbonylation, and this effect is largely mediated by its ability to maintain redox balance. Since the aldehyde 4-hydroxynonenal (4-HNE) can covalently bind to proteins’ amino acids via Michael addition reactions, introducing carbonyl groups and leading to protein carbonylation, studies have reported that the binding of proteins to 4-HNE is an indirect indicator of protein carbonylation [47]. Here, we observed that both NB treatment and si GPX4 significantly enhanced the colocalization of VDAC1 with 4-HNE (Fig. S5C–D).
To investigate the effects of Hsp90α perturbation on the VDAC1-GPX4 interaction and VDAC1 carbonylation, Co-IP experiments were performed under Hsp90α silenced conditions. The results revealed that si Hsp90α significantly reduced the interaction between VDAC1 and GPX4, accompanied by an increase in VDAC1 carbonylation levels (Fig. S5E). Further study revealed that the CTD deletion of Hsp90α enhanced the VDAC1 carbonylation (Fig. S5F).
Studies have shown that some amino acid residues, such as lysine (K), arginine (R), threonine (T), and proline (P) residues, have carbonylation sensitivity [51]. The protein docking illustrated that GPX4 binds the VDAC1 via the K274 site. The docking prediction shows that the interface binds E190 formed hydrogen bond and salt bridge with the K274, and the table indicates a short distance between the bridging atoms (Fig. 5D) (Supplementary Table 1). Besides, VDAC1-K274 site is essential for regulating mitochondrial homeostasis and cell survival and can also lead to its oligomerization [28,52], so we generated VDAC1 mutant with K-to-R (Lys-to-Arg) substitutions in K274 site. Co-IP and colocalization assays revealed a decreased interaction between GPX4 and VDAC1 after transfected K274R plasmid, and an increased level of VDAC1 carbonylation has been proved (Fig. 5E–G). The colocalization results further demonstrated that transfection with the VDAC1-K274R plasmid enhanced the overlap coefficient between VDAC1 and 4-HNE, indirectly indicating that reduced GPX4-VDAC1 interaction promotes VDAC1 carbonylation (Fig. S5G).
Besides, we measured mitochondrial ROS levels and found that treatments with NB, si GPX4, or transfection of plasmids encoding Hsp90α CTD deletion, GPX4-U73A, or VDAC1-K274R all significantly increased oxidative stress within mitochondria. Conversely, overexpression of GPX4 markedly reduced mitochondrial ROS levels. As the mitochondrial gatekeeper, VDAC1 is highly susceptible to oxidative attack, leading to carbonylation damage under these conditions (Fig. S6A). These results indicate that CTD of Hsp90 inhibition affects the interaction between GPX4 and VDAC1 and further enhances the carbonylation of VDAC1, with K274 identified as a critical site for this oxidative modification. Overall, the impaired redox homeostasis caused by the downregulation of GPX4 increased VDAC1 carbonylation. Additionally, we observed that silencing VDAC1 resulted in a decrease in GPX4 protein levels, while overexpression of VDAC1 led to an increase in GPX4 protein levels (Fig. S6B), suggesting a potential reciprocal regulatory relationship between the two proteins.

3.6
VDAC1 carbonylation promotes its oligomerization
In the crosstalk between mitochondria and cytosol, VDAC1 is an essential channel in the OMM that is involved in transferring small molecular substances up to 5 kDa [26,53]. We hypothesized that the carbonylation of VDAC1 might activate its oligomerization. Protein carbonylation could arises from direct ROS attack on amino acid residues, leading to the introduction of carbonyl groups. In this study, NAC was used to inhibit this modification by scavenging of hydroxyl radicals (•OH). Western blotting revealed that treatment with NB notably enhanced the oligomerization of VDAC1, which could be reversed by NAC and the application of oligomer inhibitor VBIT4 (Fig. 6A, S7C). Alternatively, protein carbonylation can also be promoted through Michael addition reactions between lipid peroxidation products (such as 4-HNE) and nucleophilic amino acid residues in proteins, accordingly, rescue experiments using Fer-1 (Lipid peroxidation/Ferroptosis inhibitor) in H2O2-induced oxidative stress models showed that Fer-1 effectively suppressed H2O2-mediated VDAC1-carbonylation, demonstrating its anti-carbonylation activity (Fig. S7A–B). Building on this, we evaluated Fer-1's effect on NB-induced oligomerization and found that Fer-1 potently inhibited NB-triggered VDAC1 oligomerization (Fig. 6B). We also detected that the GPX4 knockdown cell line could lead to VDAC1 oligomerization, which was blocked in the presence of NAC or Fer-1 (Fig. 6C and D). The result confirmed that the enzyme-dead GPX4 promotes the carbonylation and oligomerization of VDAC1 (Figs. S5B and S7D). Further experiments on the transfection of K274R plasmid combined with NAC or Fer-1 indicated that VDAC1 carbonylation at the K274 site is critical for the formation of VDAC1 oligomerization (Fig. 6E and F). Results showed that NAC or Fer-1 can rescue the decrease in GPX4 levels after NB treatment, or K274R plasmid transfection (Fig. S7E–F). We also found that GPX4 overexpression significantly suppressed the VDAC1 oligomerization in NB treatment (Fig. S7G–H). This indicates that the reduction in GPX4-VDAC1 interaction and the subsequent increase in VDAC1 oligomerization appear to be closely related to redox-dependent mechanisms, specifically through GPX4-mediated VDAC1 carbonylation. Besides, results illustrated that VBIT4 could effectively inhibit lipid peroxidation and GPX4 overexpression decreased lipid peroxidation levels of NB treatment (Fig. S8A–B).
Biological process analyses were performed on GPX4 expression through CAMOIP online tool based on TCGA data. The results showed that the low expression of GPX4 promoted protein oligomerization and inhibited protein import into mitochondria (Fig. S8C). To further elucidate whether VDAC1 oligomers mediate the transport of ferroptosis-related proteins, whole-cell proteomic and mitochondrial proteomic analyses were performed to find the 111 proteins that decreased in the mitochondria while increasing in the cytosol with NB treatment. We intersected the driver genes of ferroptosis from FerrDb with the 111 proteins and finally obtained 8 proteins (Fig. 6G). After eliminating the proteins mainly located on the membrane, we selected HMOX1, which has a high correlation coefficient with ferroptosis in the database. Mitochondrial isolation tests illustrated that NB treatment increased HMOX1 in the cytosol and decreased it in the mitochondria, while VBIT4 inhibited this effect (Fig. 6H). Similar results were observed when the expression of GPX4 was knocked down (Fig. 6I). The colocalization of HMOX1-mitochondrion consistently supported these findings (Fig. 6J). VBIT4 can also effectively enhance the co-localization of HMOX1 and mitochondria after transfection with the K274R plasmid (Fig. S8D). In addition, the increase in intracellular Fe2+ concentration can be inhibited by DFO and VBIT4 (Fig. 6K–S8F). So, all the data suggest that iron overload contributes to ferroptosis. Thus, a new mechanism that carbonylation of VDAC1 activates its oligomerization and further increases HMOX1 from passing through the mitochondrion could promote ferroptosis.

3.7
Hsp90 inhibitor NB suppresses HCC cells proliferation via VDAC1 oligomerization-dependent ferroptosis
To investigate the biological role and regulatory mechanisms of ferroptosis in NB-treated HCC cells, our study focused on the key ferroptosis regulator GPX4 to observed the cell proliferation. The results demonstrated that NB treatment significantly suppressed the colony-forming capacity, DNA replication activity, and 3D spheroid formation ability of HCC cells. These inhibitory effects were partially reversed by ferroptosis inhibitors Fer-1 and DFO (Fig. 7A–F, S9A-D). Furthermore, overexpression of GPX4 effectively rescued NB-induced proliferation inhibition and spheroid formation capacity decline (Fig. 7G–I), indicating that GPX4 downregulation-induced ferroptosis is an important mechanism underlying NB's anti-proliferative effects in HCC cells. Base on previous studies demonstrating that NB induces ferroptosis by downregulating GPX4 to activate VDAC1 oligomeric pore opening and iron overload, we combined the VDAC1 oligomerization-specific inhibitor VBIT4 with NB treatment or GPX4 knockdown (sh GPX4). Treatment with VBIT4 significantly reversed NB and sh GPX4-induced reductions in colony formation and restored the EdU-positive cell ratio and 3D spheroid formation capacity (Fig. 7J–O). Additionally, CCK-8 assays, intrahepatic tumor model, and subcutaneous xenograft models consistently demonstrated that GPX4 downregulation suppressed tumor proliferation, whereas its upregulation promoted tumor growth (Fig. S9G–J). These results indicate that the induction of ferroptosis is a critical factor in NB-mediated inhibition of cell proliferation, and GPX4 downregulation and further VDAC1 oligomerization serve as a key step in suppressing HCC cell proliferation by NB.

3.8
VDAC1 oligomerization inhibitor VBIT4 reverses NB-induced tumor growth inhibition via blocking ferroptosis
To validate the above findings in vivo, we subcutaneously xenografted HepG2 cells into nude mice. Experimental results demonstrated that NB, a C-terminal inhibitor of Hsp90, significantly suppressed tumor proliferation, with markedly reduced tumor volume compared to the control group. Notably, the VDAC1 oligomerization inhibitor VBIT4 effectively reversed NB's inhibitory effects on tumor growth (Fig. 8A–C). In these in vivo experiments, no significant changes in body weight were observed across treatment groups, indicating that both compounds influenced tumor progression without adversely affecting normal growth or overall health (Fig. 8D).
IHC analysis demonstrated elevated levels of the lipid peroxidation marker 4-HNE and reduced GPX4 expression in NB-treated tumors (Fig. 8E). Furthermore, ferroptosis-related proteins, including the ferroptosis suppressors GPX4, SLC7A11, and FTH1, were markedly downregulated in the NB group, and these changes were counteracted by VBIT4 co-treatment (Fig. 8F). Frozen tissue sections were subjected to immunofluorescence staining. Consistent with cellular observations, NB treatment suppressed the colocalization of HMOX1 with Mitotracker, whereas VBIT4 co-treatment significantly enhanced their overlap coefficient. These data suggest that NB inhibited tumor growth by promoting VDAC1 oligomerization to driving ferroptosis in HCC (Fig. 8G).

4
Discussion
Considering that Hsp90 is involved in the initiation and progression of a variety of malignancies, the research of the inhibitors has become one of the current hotspots. Here, we reported the detailed mechanism of Hsp90ɑ CTD inhibition inducing ferroptosis in HCC by iron overload and lipid peroxidation.
GPX4 is necessary for reducing ROS levels and performing lipid repair functions [[54], [55], [56]]. Earlier research has indicated that Hsp90 N-terminal inhibitors can prevent ferroptosis and increase the protein levels of GPX4 [21,57]. In this study, we discovered that Hsp90 CTD inhibition decreases GPX4 protein levels and encourages ferroptosis, whereas N-terminal inhibitors STA9090 cannot achieve this effect (Fig. 1). N-terminal inhibitors of Hsp90 can trigger the activation of heat shock factors and cause heat shock response, which is possibly the primary reason why they did not reduce GPX4 levels.
Recently, ferroptosis was identified as a kind of autophagy-dependent cell death. The ferroptosis inducer Erastin and RSL3 are involved in promoting the degradation of ferroptosis-related proteins via NCOA4-related ferritinophagy and CMA-mediated autophagy [15,58,59]. Our current research demonstrated that the CTD of Hsp90α stabilizes GPX4 by directly binding to GPX4, preventing its degradation via the LC3/P62 autophagy pathway (Fig. 3, Fig. 4). This finding highlights the importance of Hsp90's structural domain specificity in regulating ferroptosis.
Given that GPX4 is known as an antioxidant enzyme in cellular and mitochondria is an important source of ROS, it plays an indispensable role in mitochondria-related ferroptosis. The research illustrated that DHODH and mitochondrial GPX4 constitute the ferroptosis defense system in mitochondria by reducing ubiquinone to ubiquinol [60]. Moreover, as the gatekeeper of mitochondrial, VDAC functions in regulating the metabolism of mitochondrial and cell fate. It is shown that the inhibition of Nedd4 blocked the degradation of VDAC2/3, which increased the ferroptosis induced by Erastin [61]. Lip-1 can protect the heart from I/R damage by suppressing the expression of VDAC1 and its oligomerization [62]. Inhibition of VDAC1 oligomerization is effective in reducing ferroptosis of hepatocytes [29]. Herein, we demonstrated that the suppression of GPX4 is involved in promoting the carbonylation of VDAC1 in a redox-dependent manner, leading to the formation of oligomers of VDAC1 and further enhanced ferroptosis.
Carbonylation is an irreversible type of protein modification caused by the introduction of carbonyl derivatives, which can alter the conformation, charge, or steric hindrance of proteins, thereby impacting their functional activity [63,64]. The initiation of carbonylation can be achieved by direct attack of amino acid residues in proteins by ROS (such as •OH) to oxidize their side chains and generate carbonyl groups, or by introducing carbonyl groups through highly reactive aldehydes such as 4-HNE [65]. As the mitochondrial protein, VDAC1 is highly vulnerable to ROS-mediated oxidative attack and prone to carbonylation. In Alzheimer's disease-related research, the lipid peroxidation environment promotes an increase in MDA, which facilitates VDAC1 carbonylation. Proteomic studies have identified VDAC1 carbonylation in the brain tissues of Alzheimer's disease patients [66,67]. Due to GPX4's unique ability to scavenge free radicals and suppress lipid peroxidation, studies have shown that GPX4 inactivation elevates lipid peroxide production, leading to STING carbonylation at the C88 residue [47]. Indicating that GPX4 may influence the carbonylation of specific proteins by modulating localized oxidative stress. Our study demonstrates that a reduction in VDAC1-GPX4 interaction leads to increased VDAC1 carbonylation and subsequent oligomerization, as well as mitochondrial damage (Fig. 5, Fig. 6). It has been reported that an aniline-based probe has identified carbonylation modification sites on VDAC2 during ferroptosis [68]. Here, we observed that mutation of the VDAC1-GPX4 binding site at VDAC1-K274 resulted in increased VDAC1 carbonylation and subsequent oligomerization. This demonstrates that dissociation of the VDAC1-GPX4 complex reduces localized antioxidant capacity at VDAC1, thereby promoting its carbonylation (Fig. 5, Fig. 6E-F). Additionally, transfection with enzyme-dead GPX4 plasmid also leads to VDAC1 carbonylation and oligomerization (Figs. S5B and S7D). Highlighting that the decreased interaction between VDAC1 and GPX4 results in diminished antioxidant protection and the role of GPX4's antioxidant properties on mitochondria in protecting VDAC1 from oxidative damage.
Interestingly, some studies have shown that overexpression of VDAC1 can inhibit GPX4 levels [69]. In contrast, our research indicates that silencing or overexpressing VDAC1 can either suppress or enhance GPX4 protein levels (Fig. S6B), suggesting a mutual regulation relationship between the two. Oxidative stress affects the level of VDAC1, changes the permeability and ion balance of mitochondrial membrane, and increases the level of ROS. As an intracellular antioxidant enzyme, GPX4 can catalyze the reduction of peroxides to alcohols using GSH as a co-substrate, thereby protecting cells from oxidative damage and maintaining the redox homeostasis within cells, and the decrease of GPX4 can promote the increase of oligomers by promoting the carbonylation of VDAC1. However, the specific mechanism by which VDAC1 regulates GPX4 and the feedback mechanism requires further exploration.
As a versatile protein, VDAC1 is capable of facilitating the transport of metabolites or large proteins across the mitochondria through the formation of oligomers. Proteomics and ferroptosis database found that HMOX1 may be the molecule that VDAC1 oligomerization promotes ferroptosis. The increase of free iron in cardiomyocytes induced by HMOX1 leads to the peroxidation of the mitochondrial membrane [24]. Excessive activation of HMOX1 can degrade heme to ferrous and promote iron overload [70]. Our results supported that HMOX1 in mitochondria can be translocated into the cytoplasm through the VDAC1 oligomerization pore, and induce iron overload, indicating that the transportation of HMOX1 is a potential source of iron overload that can result in ferroptosis during VDAC1 oligomerization is activated (Fig. 6H–K). In vivo experiments demonstrated that the C-terminal inhibitor of Hsp90, NB, exhibited significant tumor-suppressive effects, while the VDAC1 oligomerization inhibitor VBIT-4 effectively reversed both the antitumor activity and ferroptosis-inducing effects of NB (Fig. 8).
In summary, we identified that Hsp90α CTD inhibition disrupts its interaction with GPX4 and promotes LC3/P62-mediated autophagy degradation of GPX4. The decrease in GPX4 and the reduction of GPX4-VDAC1 interaction promotes VDAC1 carbonylation and subsequent oligomerization, ultimately promoting the release of HMOX1 and leading to cellular iron overload, which is a crucial factor in ferroptosis induced by VDAC1 oligomerization. This finding underscores the complex role of oxidative modifications and VDAC1 oligomerization in regulating ferroptosis pathways. Additionally, our results indicate a critical role for the VDAC1 oligomerization pathway in mediating the antitumor effects of Hsp90 inhibition, particularly through the induction of ferroptosis. Therefore, strategies to enhance VDAC1 oligomerization could potentially synergize with Hsp90 CTD inhibition to amplify ferroptosis in cancer cells, offering new avenues for developing combination therapies or overcoming resistance in oncology.

CRediT authorship contribution statement
Jieyou Li: Investigation, Validation, Visualization, Writing – original draft. Guibing Wu: Data curation, Investigation, Validation. Hairou Su: Investigation, Data curation. Manfeng Liang: Investigation, Data curation. Shengpei Cen: Investigation. Yandan Liao: Investigation. Xiangjun Zhou: Data curation, Investigation, Methodology. Guantai Xie: Investigation. Zihao Deng: Investigation. Wenchong Tan: Methodology. Yan Li: Investigation. Wang Xiao: Methodology, Resources. Lixia Liu: Methodology, Project administration. Jinxin Zhang: Investigation, Methodology. Zhenming Zheng: Software. Yaotang Deng: Software. Yaling Huang: Investigation. Xiongjie Shi: Investigation. Yilin Liu: Investigation. Guowei Zhang: Methodology, Resources. Xuemei Chen: Conceptualization, Funding acquisition, Writing – review & editing, Project administration, Supervision.

Ethics statement
The collection of clinical samples and the animal studies have been reviewed by the Ethics Committee of Southern Hospital of Southern Medical University and the Ethics Committee of Experimental Animals of Southern Medical University, respectively. The manuscript has been approved by all the authors.

Data availability statement
All data generated or analyzed during this study are included in this published article and the supplementary files.

Funding statement
This study was supported by 10.13039/501100001809National Natural Science Foundation of China (No. 82072105, No. 81673216), Guang Dong Basic and Applied Basic Research Foundation (No.2025A1515012191) to Xuemei Chen; 10.13039/501100001809National Natural Science Foundation of China (No. 82130054, 81971783) to Fei Zou.

Declaration of competing interest
All authors declare that no conflict of interest exits that could bias the outcome of our study.