HDAC6 Inhibitors Sensitize Claudin-high Breast Cancer Cells to Cysteine Depletion via Activation of PKCγ Signaling


 IntroductionTriple-negative breast cancer (TNBC) is a highly malignant breast cancer type with poor prognosis and lacks effective therapy. TNBC is not responsive to targeted therapy for hormone receptors and often exhibit resistance to current chemotherapeutic agents. Targeting tumor metabolism is an emergent strategy to treat cancer. Therefore, identification of tumor metabolic deregulations and development of effective targeted therapies are urgently needed.MethodsWe performed the epigenetic compound library screening in claudin-high breast tumor cells and identified therapeutical sensitizers to overcome the drug resistance of targeted cysteine-dependence therapy. Gene expression profiling were generated to analyze signaling pathways induced by the combined tubacin and cysteine deprivation treatment. Specific inhibitors, shRNA, and CRISPR/Cas9 gene editing approaches were used to target cellular proteins HDAC6 and PKCγ and examine their roles in cell death. Cell viability, RT-qPCR, and Western blotting assays were performed in cysteine-independent tumor cells to examine the anticancer effects of combined tubacin and cysteine deprivation treatment. ResultsWe found that TNBC has differential death responses to cysteine deprivation and the cysteine-dependence of TNBC corelates with the expression levels of claudin genes in addition to the classical EMT markers. To overcome drug resistance in claudin-high/cysteine-independent breast tumor cells, HDAC6 inhibitors were identified by the epigenetic compound library screening as potent sensitizers that synergize with cysteine deprivation to eradicate cysteine-independent tumor cells. Unexpectedly, HDAC6 knockout did not recapitulate the HDAC6 inhibitors-mediated synthetic lethality, indicating that HDAC6 is not the actual target of HDAC6 inhibitors in this context. Transcriptomic profiling revealed that HDAC6 inhibitors synergizes with cysteine depletion to trigger a profound gene transcriptional program. Notably, a zinc-related gene response was observed to accompany with a prominent increase of labile zinc in cells during cell death. We further showed that activation of PKCγ signaling is required to interfere cellular zinc homeostasis and drive HDAC6 inhibitors-mediated cell death.ConclusionOur study demonstrated that HDAC6 inhibitors function as potent sensitizers to overcome the resistance of cysteine deprivation in claudin-high breast tumor cells. Identification of such sensitizers would make the targeted cysteine-dependence therapy applicable in various subtypes of breast cancer.


Introduction
Triple-negative breast cancer (TNBC) is a highly malignant breast cancer type with poor prognosis and lacks effective therapy. TNBC is not responsive to targeted therapy for hormone receptors and often exhibit resistance to current chemotherapeutic agents. Targeting tumor metabolism is an emergent strategy to treat cancer. Therefore, identi cation of tumor metabolic deregulations and development of effective targeted therapies are urgently needed.

Methods
We performed the epigenetic compound library screening in claudin-high breast tumor cells and identi ed therapeutical sensitizers to overcome the drug resistance of targeted cysteine-dependence therapy. Gene expression pro ling were generated to analyze signaling pathways induced by the combined tubacin and cysteine deprivation treatment. Speci c inhibitors, shRNA, and CRISPR/Cas9 gene editing approaches were used to target cellular proteins HDAC6 and PKCγ and examine their roles in cell death. Cell viability, RT-qPCR, and Western blotting assays were performed in cysteine-independent tumor cells to examine the anticancer effects of combined tubacin and cysteine deprivation treatment.

Results
We found that TNBC has differential death responses to cysteine deprivation and the cysteinedependence of TNBC corelates with the expression levels of claudin genes in addition to the classical EMT markers. To overcome drug resistance in claudin-high/cysteine-independent breast tumor cells, HDAC6 inhibitors were identi ed by the epigenetic compound library screening as potent sensitizers that synergize with cysteine deprivation to eradicate cysteine-independent tumor cells. Unexpectedly, HDAC6 knockout did not recapitulate the HDAC6 inhibitors-mediated synthetic lethality, indicating that HDAC6 is not the actual target of HDAC6 inhibitors in this context. Transcriptomic pro ling revealed that HDAC6 inhibitors synergizes with cysteine depletion to trigger a profound gene transcriptional program. Notably, a zinc-related gene response was observed to accompany with a prominent increase of labile zinc in cells during cell death. We further showed that activation of PKCγ signaling is required to interfere cellular zinc homeostasis and drive HDAC6 inhibitors-mediated cell death.

Conclusion
Our study demonstrated that HDAC6 inhibitors function as potent sensitizers to overcome the resistance of cysteine deprivation in claudin-high breast tumor cells. Identi cation of such sensitizers would make the targeted cysteine-dependence therapy applicable in various subtypes of breast cancer.

Page 3/33
Triple-negative breast cancer (TNBC) accounts for 15% ~ 20% of overall breast cancer cases and exhibits earlier age of onset, high metastasis, and aggressiveness with poor clinical outcomes shown by higher relapse and lower survival rates than other types of breast cancer [1][2][3]. TNBC contributes to the major mortality of breast cancer patients [4][5][6]. However, there is few clinically effective and targeted therapy for TNBC [7][8][9]. Treatments of TNBC patients are still limited to surgery, chemotherapy, or radiation since the absence of cell receptors makes targeted hormonal therapies impossible. About 50% of TNBC respond to conventional therapies, but the e ciency of treatments is limited by tumor metastasis and drug resistance [10,11]. Therefore, there remains an urgent need to develop novel therapeutic approaches to improve cancer outcomes and reduce patient mortality.
With omics technologies, researchers have gained a deep understanding of molecular complexity of cancer [1,12,13]. Metabolic deregulation is an emergent hallmark in many cancers [14,15]. Cellular metabolic rewiring often occurs with oncogenic alterations and microenvironmental adaptations to meet the demands of tumor cell survival, proliferation, and invasion [16][17][18]. For example, tumor cells acquire aerobic glycolysis (the Warburg effect) and redirect glucose metabolites into the pentose phosphate shunt pathway to maintain cellular redox and supply extra needs of nucleotides. Hence, these tumor cells are addicted to glucose and sensitive to inhibition of glycolysis by glucose analogues [19,20]. Similarly, alteration of amino acid metabolism has been observed in many cancers associating with genetic aberrations. Oncogenic MYC activates glutaminolysis and establishes an alternative citric acid cycle to meet the demands of lipid biomass for fast cell proliferation, and MYC-induced tumors typically addict to glutamine [21,22]. Overall, targeting metabolic vulnerabilities has been suggested as a promising targeted strategy to treat cancers.
The cysteine, mostly derived from extracellular disul de cystine, is involved in cellular glutathione (GSH) synthesis that removes cytotoxic ROS and reactive nitrogen through the action of glutathione peroxidases (GPXs) [26]. Cystine deprivation or blocking the cystine/glutamate antiporter (the system xC-) by erastin or sulfasalazine limits the synthesis of cellular GSH, which leads to accumulation of lipid peroxidation, oxidative damage, and ultimately ferroptosis [23]. Ferroptosis is a new form of regulated and iron-dependent cell death, which is mechanistically distinct from necroptosis, autophagy, and other forms of non-apoptotic cell death and mostly induced by lethal lipid peroxidation [27,28]. Targeting cysteine dependence/addiction could be an effective targeted cancer therapy since limiting single amino acid is relatively feasible and applicable to be achieved in vivo [29,30]. Recently, a human-engineered enzyme "cysteinase" has been developed to deplete serum cystine/cysteine to suppress tumor growth in mice with a safe and effective e cacy [31,32]. Cysteine dependence has been observed as a striking feature in TNBC [33]. However, only a subset of TNBC cells shows great sensitivity to cysteine depletion and the system xC-inhibitors. Little is known what mechanisms control the high-demand of cysteine in TNBC. Also, biomarkers are needed to identify to precisely determine the cysteine dependence in TNBC.
Epigenetic alterations including histone modi cations and DNA methylation have been suggested to direct cancer development and progression [34]. Distinct DNA methylation patterns have been observed to associate with luminal and basal subtypes of breast cancers when analyzing the omics of 802 breast tumor cases [35]. A high degree of DNA methylation deposits in aggressive metastatic breast cancer cells and a relative low level of histone acetylation and methylation associates with low prognostic breast cancers [35][36][37][38]. These suggest that the poised epigenetic states contribute to con gure different subtypes of breast cancer. Such epigenetic states in TNBC could acquire differentiated cysteine requirements in cells and dictate cancer cells with distinct cellular responses to cysteine deprivation. In addition, epigenetic changes during therapy are potential drivers of therapeutic drug resistance in cancer [39]. Therefore, modulation of epigenetic regulator's activity could render cysteine dependence and overcome drug resistance in tumor cells.
In this study, we found that the expression levels of claudin genes correlate with cysteine-dependence in TNBC and claudin-high TNBC cells are resistant to cysteine deprivation. We identi ed HDAC6 inhibitors as potent sensitizers by drug screening to overcome the drug resistance of cysteine depletion in claudinhigh tumor cells. Moreover, HDAC6 inhibitors, independent on their canonical target HDAC6, promote synthetic lethality of cysteine depletion through activation of PKCγ signaling. Our study revealed a new function of HDAC6 inhibitors that could be used as therapeutic adjuvants for the targeted cysteinedependence therapy to treat various subtypes of breast cancer.

Cell culture and reagents
All breast tumor cells and 293T cells were purchased from ATCC and maintained as per standard protocol in an incubator with 95% humidity and 5% CO 2 at 37 o C. Cells were cultured in DMEM with 10% heat-inactivated fetal bovine serum (FBS) and 1% penicillin-streptomycin. The cysteine de cient medium was prepared according to the previous report [40]. Erastin, selective HDAC6 inhibitors tubacin, tubastatin A, and CAY10603, Myriocin, and the metal chelator N, N, N′, N′-tetrakis (2-pyridylmethyl) ethylenediamine (TEPN) were obtained from Cayman Chemicals (Ann Arbor, Michigan, US); Z-VAD-FMK, necrostatin-1, and ferrostatin-1 were purchased from Calbiochem Research Biochemicals (Sigma). The molecular probe FluoZin-3 was purchase from Thermo Fisher Scienti c. All antibodies used in this study were listed in supplementary table 1.

Epigenetic compound library screening
Breast tumor cells HCC38 or T47D were seeded in two sets of epigenetic compound library plates under either cystine-rich or cystine depleted condition respectively. The cell viability was determined at 72 hours using the CellTiter-Glo assay kit (Promega).

Lentiviral cell infection
Viral particles were generated in 293T cells by transfecting lentiviral packaging plasmids using Lipofectamine 3000 (ThermoFisher Scienti c) and collected from cell media after 48 hours. The targeted cells were infected for 48 hours by indicated viruses and further stably selected by either puromycin or blasticidin (Cayman chemical). The pLentiCRISPR-v2-HDAC6 was purchased from GenScript with the targeting sgRNA sequence 5'-CAGTGCTACAGTCTCGCAC. All shRNAs targeting HDAC6 and PRKCG genes were purchased from Sigma. The pLKO1.0 plasmid was used as a control vector for infection.

Gene expression pro ling and geneset enrichment analyses (GSEA)
The gene expression pro le in HCC38 cells treated with either control, erastin, tubacin, or erastin and tubacin for 24 hours was analyzed by the GeneChip™ Human Gene 2.1 ST 24-Array Plate (ThermoFisher Scienti c). The data were deposited in GEO database (GSE154425). The probe intensities were normalized by RMA. The gene expression changes in HCC38 cells were derived by zero-transformation (Δlog 2 ) against those in the control condition. Probe sets that varied by 2-fold in at least 3 samples were selected for hierarchical clustering. The pathway enrichment was analyzed by Gene Set Enrichment Analysis (GSEA) using the G2 annotated-genesets with default criteria of 1000 permutations.
RNA extraction and real-time RT-PCR RNA was extracted from cells by RNeasy kit (Invitrogen). Total RNA (2 µg) was reverse-transcribed to cDNA and the quantitative PCR was performed using SYBR Green PCR mix (Applied Biosystems). The relative difference in mRNA expression was normalized with actin using the ΔΔCT method. All primers in this study were listed in supplementary table 2.

Protein immunoblotting analysis
Proteins were extracted from the cells using the RIPA extraction and lysis buffer (Sigma) with the protease and phosphatase inhibitor cocktail (ThermoFisher Scienti c). Protein concentrations were determined by BCA protein assay. Equal amounts of protein were loaded for the immunoblot analyses. The signal was detected by the ECL plus Western blotting detection system (Amersham) and visualized by LAS-4000 lumino image analyzer.

Cell viability and cytotoxicity
Cell viability was measured by either trypan blue cell counting or the relative ATP level using the CellTiter-Glo assay kit (Promega) and evaluated by crystal violet staining. Cell cytotoxicity was measured using the CytoTox-Fluor™ cytotoxicity assay kit (Promega).

Microscopic Imaging
To analyze intracellular zinc levels, 7 × 10 3 cells were seeded in the 96-well plate under various treatments. In the end, the cells were stained by 2 µM FluoZin-3 molecular probe (ThermoFisher Scienti c) at 37 °C for 30 min. Dialyzed FBS medium was used to ensure that the medium was zinc-free.

Cellular zinc level measurement
Total cellular zinc including labile and bound zinc was determined using ICP-OES (CHORI Elemental Analysis Facility) [41]. Brie y, 3 × 10 6 HCC38 cells under different treatments were collected and fully dissolved in 0.25 mL OmniTrace 70% HNO 3 (EMD Chemicals) by microwave digestion. Samples were diluted to 5% HNO 3 with OmniTrace water and analyzed with the use of a Vista Pro ICP-OES (Varian Vista Pro). Zinc was measured at the 213-nm wavelength with a detection range between 0.005 and 5 ppm. All associated reagents and plasticware were certi ed as trace metal-free or tested for trace metal contamination. Zinc concentrations were normalized by total protein mass. All samples were analyzed in triplicate.

Statistical analyses
The signi cance of differences between groups was determined using a t test. Statistical analysis was performed using GraphPad Prism 8.0 Software. A p-value < 0.05 was considered statistically signi cant. Data were presented in gures as mean ± standard deviation (SD).

HDAC6 inhibitors sensitize the claudin-high TNBCs to cysteine deprivation
Cysteine dependence/addiction is a novel feature in mostly aggressive TNBC cells [42]. However, only a subset of TNBC exhibited cysteine-dependence, such as HBL100 and MDA-MB-231 cells, which underwent rapid necrotic cell death in response to erastin, an inhibitor of cysteine transport, while HCC70 and HCC38 cells exhibited no signi cant cell death (Fig. S1A). To examine whether gene expression programs correlate differential cysteine requirement in TNBC, the gene expression pro ling data of various TNBC cells (GSE69017) [43] were subjected to the cluster analysis. We found that two subgroups of TNBCs, with differentiated claudin gene expression, correlated with their cysteine-dependence, which were named claudin-high and claudin-low TNBC (Fig. S1B),. Gene Set Enrichment Analysis (GSEA) showed that the claudin-high TNBC has high epithelial gene expressions but low metastasis-related gene expressions, such as epithelial-mesenchymal transition (EMT) markers CDH1 and VIM ( Fig. S1C and S1D), which were con rmed by quantitative RT-PCR (Fig. S1F). The genes with potential DNA and histone methylation in their promoters are suppressed in claudin-low TNBC cells (Fig. S1E), suggesting a role of epigenetic regulation in cell identity. In addition, analysis of gene pro ling data from 812 TCGA invasive breast carcinoma tumors [44] indicated that the claudin genes are highly correlated with each other but have less or no correlation with the EMT markers CDH1 and VIM (Fig. S1G). These data suggest that the claudin genes should be supportive or better biomarkers to determine tumor cysteine-dependence in vivo.
Epigenetic regulation could determine cysteine-dependence in TNBC, since epigenetic alterations are often associated with cancer progression and therapeutic drug-resistance in breast tumors [36][37][38]. To overcome the drug resistance in claudin-high TNBCs, we examined whether modulation of epigenetic regulator's activity can sensitize cysteine-independent tumor cells to cystine depletion. To that end, we developed a synthetic-lethal library screen strategy, by which the cell survival rate of cysteine-independent and claudin-high cancer cells was examined under cotreatment of epigenetic inhibitor and cystine deprivation. 140 inhibitors of epigenetic regulators were included in the epigenetic compound library, and the screening was performed in claudin-high TNBC (HCC38) and luminal (T47D) cancer cells. The epigenetic compound (2 µM) was applied to these cells in either cystine-replete (+ Cys) or cystine-depleted (-Cys) condition and cell survival was measured by cellular ATP levels. By drug screening, three inhibitors of the histone deacetylase 6 (HDAC6), Tubacin, CAY10603, and Tubastatin A, were identi ed and dramatically induced synthetic-lethal death in both HCC38 and T47D cells under the cystine-depleted condition ( Fig. 1A and 1B). We con rmed that tubacin indeed induced signi cant synthetic cell death in

HDAC6 inhibitors synergize with erastin to induce a mixed necrotic and apoptotic cell death
To characterize the synthetic-lethal death pathway induced by erastin and HDAC6 inhibitors, we examined various death and signaling markers and protective effects of different cell death inhibitors. The combined erastin and tubacin treatment caused activation of death signaling (phosphorylation of p38) and increase of DNA double-strand breaks (phosphorylation of H2AX) with a partial cleavage of PARP1 and caspase-3 ( Fig. 2A and 2B), while either erastin or tubacin treatment alone failed to activate such death markers and signaling. Acetylation of α-tubulin, the substrate of HDAC6, was strongly induced by tubacin. The synergistic cell death was only partially blocked by the pan-caspase inhibitor Q-Vad, but fully rescued by the ferroptosis and necroptosis inhibitors Ferrostatin-1 and Necrostatin-1 (Fig. 2C-D), and which abolished most cell death markers and signaling (Fig. 2E), indicating a largely necrotic cell death or ferroptosis. Similarly, the cell lethal effects induced by CAY10603 was blocked by all three cell death inhibitors (Fig. 2F). These data suggested that HDAC6 inhibitors synergize with erastin induced a mixed necrotic and apoptotic cell death program in cysteine-independent tumor cells.
HDAC6 is not required for tubacin to promote synergistic cell death.
HDAC6 has been suggested as a therapeutic target since high level of HDAC6 has been reported in smallsize and ER/PR positive breast tumors and to confer drug resistance [45,46]. Therefore, we determined whether inactivation of endogenous HDAC6's activity by shRNA is able to mimic the lethality-promoting effect of HDAC6 inhibitors in our context. Unexpectedly, silencing of HDAC6 expression by different shRNAs did not promote cell death in response to erastin, although the protein level of HDAC6 was knocked down and the acetylation level of tubulin protein was signi cantly increased ( Fig. S3A-C). We questioned whether shRNA is unable to inactivate HDAC6 enough to the same level of HDAC6 inhibitors. To that end, we employed CRISPR/Cas9 gene editing to knockout HDAC6 expression in claudin-high TNBC HCC38 and MDA-MB-436 cells and luminal T47D cancer cells. Independent HDAC6-null cell clones were isolated based on the complete deletion of HDAC6 protein expression and increased acetylation of tubulin ( Fig. 3A-C upper panel). Surprisingly, knockout of HDAC6 could not promote the synthetic-lethal response with erastin, which was shown by no signi cant ATP decrease and cell loss in different HDAC6null cell clones (Fig. 3A-C lower panel and 3E). Erastin with tubacin, but not erastin alone, induced similar cell death (Fig. 3A-C lower panel and 3E) and activated similar death markers (Fig. 3D) in both HDAC6null and vector cells. The sphingolipid biosynthesis was previously identi ed as a so-called off-target of tubacin [47]. However, myriocin, an inhibitor of sphingolipid synthesis, failed to promote cell death with erastin in cysteine-independent tumor cells ( Fig. S3D-E). We excluded the involvement of sphingolipid biosynthesis in this context. Taken together, the data strongly suggest that the death-promoting effect of HDAC6 inhibitors is independent on endogenous HDAC6.
Tubacin synergizes with erastin to activate a lethal gene transcriptional program Although the molecular target(s) of HDAC6 inhibitors in our system remains elusive, dissecting the HDAC6 inhibitors-mediated gene transcriptional program might uncover the underlying death-promoting mechanism and help to identify new molecular targets of HDAC6 inhibitors. To that end, we examined the gene expression pro ling in HCC38 cells under various conditions by Affymetrix microarray (Human Gene 2.1 ST Array). RNA was collected from cells in triplicate treated with either control, erastin, tubacin, or erastin plus tubacin and subjected to microarray analysis (GSE154425). By the supervised cluster analysis, we found that either erastin or tubucin alone induced mild or few changes of gene expression. However, the combination of tubacin and erastin dramatically triggered induction or repression of a large group of genes (Fig. 4A), such as apoptotic genes (Bim, BNIP3, and Puma) and genes involved in endoplasmic reticulum and oxidative stress (ATF3, CHOP, and HOMX1). GSEA revealed that the gene response induced by tubacin and erastin mimics the gene changes by photodynamic therapy (PDT) (Fig. 4B), which is a well-established cancer therapy utilizing a light-absorbing molecule or visible light irradiation to cause tumor ablation [48]. RT-qPCR and immunoblotting data con rmed the pro ling data that the apoptotic genes were strongly induced by the combination treatment, but not by either erastin or tubacin alone (Fig. 4C-D). Similarly, erastin alone did not activate apoptotic genes in the HDAC6-null cells (Fig. 4E-F). These data suggested that tubacin and erastin induces a lethal gene response but in an HDAC6-independent manner.
Ferroptosis is a new form of necrotic cell death, iron-dependent and characterized by the accumulation of lipid peroxidation products and reactive oxygen species (ROS) [49,50]. We did not nd any changes in genes involved in cellular iron metabolism. Interestingly, the genes involved in zinc transport and storage, such as ZnT1, ZnT2, and the metallothioneins MT1G/1H/1M, were highly induced by erastin and tubacin (Fig. 4A). RT-qPCR con rmed that these zinc-related genes were strongly induced by the combined treatment, but not by either erastin or tubacin alone (Fig. 5A-B). Similarly, Cay10603 along with erastin also strongly activated the zinc-related genes (Fig. 5C). Previous reports showed that the increase of zinc in cells causes ROS production and induces a mixed type of cell death, including apoptosis and necrosis [51,52]. Therefore, we examined whether the zinc is involved in tubacin mediated death-promoting effects. The FluoZin™-3, a Zn 2+ -selective indicator, was used to detect cellular labile zinc. Indeed, the level of labile zinc was highly increased in cells by erastin and tubacin, but not signi cantly by either erastin or tubacin alone (Fig. 5D and S4A). Increased labile zinc mostly accumulated in nucleus, as it co-localized with nuclear DNA stained by DAPI. Next, the total cellular level of zinc, including labile and bound zinc, was determined by ICP-OES. We found that the total cellular level of zinc was not signi cantly altered under any conditions (Fig. 5E). These suggested that the increased labile zinc was mostly released from cellular proteins or compartments but not imported from culture media. To test whether the increased labile zinc mediated the death-promoting effects, TPEN, a zinc ion chelator [53], was used to chelate the labile zinc in cells. However, TPEN could not protect cells from cell death induced by erastin with tubacin ( Fig. 5F and S4B). In summary, these results suggested that the zinc-related gene response and increase of labile zinc are likely the outcomes of cell death, but not the death-promoting mechanism of HDAC6 inhibitors.
Inhibition of PKC suppresses the death-promoting effects of tubucin in claudin-high TNBC Since zinc can function as a structural or modulatory component of many regulatory and signaling proteins [54,55], the increase of labile zinc in cells could indicate alterations of cellular protein function or signaling. Previous reports indicated that the zinc release from protein kinase C (PKC) is a common event during PKC activation by reactive oxygen species [56,57]. Therefore, we examined whether activation of PKC is required for the death-promoting effects of HDAC6 inhibitors. We found that Gö 6983, a PKC inhibitor with a broad inhibitory spectrum [58], fully rescued cells from cell death induced by erastin and tubacin, while Gö 6976, a selective inhibitor of PKCα/β, had no protective role ( Fig. 6A and S5A).
Immunoblotting and RT-qPCR analysis con rmed that Gö 6983 abolished the induction of cell death genes and phosphorylation of PKC substrates triggered by the combined erastin and tubacin (Fig. 6B-C  and S5B). Moreover, Gö 6983 signi cantly suppressed the zinc-related gene response and increase of labile zinc in cells (Fig. 6D-E and S5C). These results suggested that activation of PKC, but not PKCα/β, is required for tubacin to promote cell death in cysteine-independent cells.
PKCγ is required for the tubacin-mediated synthetic-lethality Next, we examined which member of PKC family kinases is required for the death-promoting effects of tubacin. Two additional PKC inhibitors with a slightly different inhibitory spectrum, bisindolylmaleimide I and sotrastaurin, were used. Similar to Gö 6983, bisindolylmaleimide I strongly protected cells from cell death induced by erastin and tubacin, but sotrastaurin failed to rescue cells from the stress (Fig. 7A-B). Based on the protective role and inhibitory spectrum of various PKC inhibitors, PKCγ is the candidate kinase since it is inhibited by Gö 6983 and bisindolylmaleimide I, but not by Gö 6976 and sotrastaurin.
Immunoblotting analysis showed that there was a transient phosphorylation of PKCγ in the early phase during the combined erastin and tubacin treatment, which was associated with the phosphorylation of PKC substrates, while the phosphorylation of PKC δ/θ occurred in the later phase (Fig. 7C). Furthermore, knockdown of PKCγ by shPKCγ signi cantly alleviated cell death induced by erastin and tubacin in cysteine-independent TNBCs (Fig. 7D and 7F). In consistent with its protective role, PKCγ knockdown suppressed induction of cell death genes and phosphorylation of H2AX (Fig. 7E). Taken together, activation of PKCγ is required for the tubacin-promoting synthetic-lethality in cysteine-independent TNBCs.

Discussion
TNBC is the most challenging subtype of breast cancer and overall its survival rate remains very low [59].
Developing novel targeted therapies and de ning associated biomarkers are urgent needed for precise TNBC treatment. Targeting cysteine-dependence can eradicate a subset of TNBCs with highly epithelialto-mesenchymal transition (EMT) but less responsive in many other TNBCs and luminal breast cancer [42]. The early reports indicated that the claudin-low tumors are an aggressive subtype of breast cancer with poor prognosis that enriches tumor-initiating cells with high expression of EMT markers [60,61]. Our gene expression analysis showed that the expression of claudin genes, such as CLDN 3, CLDN4, and CLDN7, are highly correlated with classical EMT markers CDH1 and VIM genes in vitro and can differentiate cysteine-dependence in TNBC. However, the expression of claudins are highly correlated with each other in invasive breast tumors in vivo, but not greatly correlated with the expression of CDH1 and VIM. This discrepancy suggests that the claudin genes should be used as additional biomarkers to guide the precise application of targeted cysteine-dependence therapy in TNBC patients.
Since many TNBC and luminal breast cancer are cysteine-independent, optimization of targeted cysteinedependence therapy will be required for its broad and effective application. Epigenetic alterations can change tumor identity and metabolism, contribute to breast tumor heterogeneity, and acquire drug resistance in cancer [36,38]. Many epigenetic activators or inhibitors have been used as therapeutic adjuvants to increase chemotherapy e cacy and overcome drug resistance [34,39]. In line with these, we identi ed that the HDAC6 inhibitors can render cysteine-independent tumor cells sensitive to cysteine deprivation. HDAC6 is a unique member of the HDAC family that can regulate cell proliferation, metastasis, and invasion in tumors, and can also drive tumor progression and confer drug resistance in some cancers [62][63][64][65]. These observations suggest that HDAC6 is likely a reasonable molecular target in our system. However, we showed that knockout of endogenous HDAC6 protein did not mimic HDAC6 inhibitors to synergize with erastin to induce cell death in claudin-high TNBC. This indicates that the synergistic effect of HDAC6 inhibitors is via a new cellular molecule, which is entirely independent of HDAC6. Previous report showed that the sphingosine biosynthesis is an off-target of tubacin [66], but we ruled out the involvement of the sphingosine pathway in our system. More studies are needed in order to identify new molecular target(s) of HDAC6 inhibitors since identi cation of such new targets of HDAC6 inhibitors would help us reveal new genes or pathways involved in tumor cysteine-addiction, and which can be used as direct targets in combination with cysteine deprivation for cancer treatment. Many studies had attempted to use HDAC6 inhibitors to treat cancer, but the therapeutic outcomes of these inhibitors should be carefully examined in order to avoid any off-targeting in patients.
The intrinsic target of HDAC6 inhibitors in our context remains unknown, but we found that tubacin with erastin triggered many stress and apoptotic gene responses, one of which mimics the photodynamic therapy (PDT) gene response [48]. The tubacin-promoting cell death occurred as a mixed type of cell death, which includes apoptosis, necroptosis, and ferroptosis. Ferroptosis is an iron-dependent cell death driven by lipid peroxidation with necrotic features [28,49,50]. The transcriptional pro ling data did not show any iron-related gene response, but a prominent zinc-related gene response was activated by the combined erastin and tubacin, which was con rmed by a signi cant increase of labile zinc in cells. We found total zinc, including labile and bound forms, was not signi cantly changed in treated cells, which indicated that increase of labile zinc was likely released from cellular proteins or compartments but not imported from extracellular culture media.
Zinc is a trace but essential metal micronutrient and is integral to many enzymes and regulatory proteins, and functions as a signaling messenger in cells [67,68]. We found that cellular zinc homeostasis is deregulated by HDAC6 inhibitors and cysteine depletion, but the increase of labile zinc does not promote cell death. Therefore, the increase of labile zinc in cells is a possible indicator of intracellular signaling perturbations under the combined treatment. It has been reported that reactive oxygen species (ROS) can cause zinc releasing from oxidized metallothioneins and zinc-ngers of abnormal proteins [69]. More speci cally, it has been shown that zinc can modulate the activity of PKC family isozymes via zinc nger domains and activation of PKCα/β by ROS stress can release zinc directly from PKC into cytoplasm [56,57]. Different isozymes of PKC family play roles in multiple cellular processes including proliferation, survival, invasion, angiogenesis [70]. Indeed in our study, inhibition of PKC abolished the release of labile zinc and cell death. Speci cally, activation of PKCγ is required for the tubacin's death-promoting effects. As one of conventional PKC isozymes, PKCγ is mainly present in the brain and has very few studies in cancer [71]. Recent studies showed that activation of PKCγ increases the migratory capacity in colon cancer [72,73]. Our study identi ed a new role of PKCγ in breast cancer that the signaling of PKCγ is important to mediate the cysteine-dependence.

Conclusion
We found that cysteine-dependence in TNBC associates with the expression levels of claudin genes in addition to the classical EMT markers. HDAC6 inhibitors identi ed by the epigenetics compound library screening overcome drug resistance in claudin-high TNBC and promote the synthetic-lethality of cysteine deprivation. Importantly, HDAC6 inhibitors execute their death-promoting effects independent on their canonical target HDAC6 protein. In addition, HDAC6 inhibitors synergize with cysteine restriction to trigger a mixed type of cell death by induces a profound gene transcriptional program via the PKCγ signaling. Identi ed such HDAC6 inhibitors will support the broad translational application of targeted cysteinedependence therapy in various types of breast cancer. The authors declare that they have no known competing nancial interests or personal relationships that could in uence the work reported in this paper.

Funding
This study was partially supported by the National Institute of Health (Grant # 1R15CA246336-01) to X.T. and the Research excellence fund of Michigan Technological University (Grant # 2015-025) to X.T.

Availability of data and materials
The datasets generated and used in the current study are available in the Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/geo/) and The Cancer Genome Atlas (TCGA) database (https://www.cancer.gov/about-nci/organization/ccg/research/structural-genomics/tcga).