Metabolomics Reveals that Cysteine Metabolism Plays a Role in Celastrol-Induced Mitochondrial Apoptosis in HL-60 and NB-4 Cells

Recently, celastrol has shown great potential for inducing apoptosis in acute myeloid leukemia cells, especially acute promyelocytic leukaemia cells. However, the mechanism is poorly understood. Metabolomics provides an overall understanding of metabolic mechanisms to illustrate celastrol's mechanism of action. We treated both nude mice bearing HL-60 cell xenografts in vivo and HL-60 cells as well as NB-4 cells in vitro with celastrol. Ultra-performance liquid chromatography coupled with mass spectrometry was used for metabolomics analysis of HL-60 cells in vivo and for targeted L-cysteine analysis in HL-60 and NB-4 cells in vitro. Flow cytometric analysis was performed to assess mitochondrial membrane potential, reactive oxygen species and apoptosis. Western blotting was conducted to detect the p53, Bax, cleaved caspase 9 and cleaved caspase 3 proteins. Celastrol inhibited tumour growth, induced apoptosis, and upregulated pro-apoptotic proteins in the xenograft tumour mouse model. Metabolomics showed that cysteine metabolism was the key metabolic alteration after celastrol treatment in HL-60 cells in vivo. Celastrol decreased L-cysteine in HL-60 cells. Acetylcysteine supplementation reversed reactive oxygen species accumulation and apoptosis induced by celastrol and reversed the dramatic decrease in the mitochondrial membrane potential and upregulation of pro-apoptotic proteins in HL-60 cells. In NB-4 cells, celastrol decreased L-cysteine, and acetylcysteine reversed celastrol-induced reactive oxygen species accumulation and apoptosis. We are the first to identify the involvement of a cysteine metabolism/reactive oxygen species/p53/Bax/caspase 9/caspase 3 pathway in celastrol-triggered mitochondrial apoptosis in HL-60 and NB-4 cells, providing a novel underlying mechanism through which celastrol could be used to treat acute myeloid leukaemia, especially acute promyelocytic leukaemia.

and early haemorrhagic death if not promptly diagnosed 3,4 . With the introduction of all-trans retinoic acid (ATRA) and arsenic trioxide (ATO), APL treatment has achieved good outcomes. However, ATRA and ATO can lead to the development of differentiation syndrome and multiorgan toxicity, and a few patients do not respond favourably to the drugs 5,6 . Thus, the, development of less toxic and more effective agents to treat APL is urgent 7 Traditional Chinese medicine, as an important source of therapeutically effective drugs, has attracted increasing attention in cancer therapy 8 . Recently, celastrol an active component of Tripterygium wilfordii Hook f. (Supplementary Material: Fig. S1), is receiving increasing attention. Celastrol has shown great potential for the treatment of inflammation 9 , Gaucher disease 8 , obesity 10 and multiple cancers, such as prostate cancer 11 , breast cancer 12 , and leukaemia 13 .
The investigation of the effect of celastrol on AML and APL is an interesting topic, and has been widely conducted. Celastrol could significantly prolong the survival of mice in HoxA9/Meis1-induced AML model and APL model 13,14 . In fresh cells from patients of various types of AML, celastrol showed effect for the treatment of leukemia 13,15 . In addition, celastrol could eradicate leukemia stem cell which is the key cause of relapse 16,17 . Importantly, the previous study has demonstrated that celastrol showed stronger anti-tumour effect than ATRA in leukemia cells 13 . Celastrol is also found as a promising and unique agent for managing the sid e effects of ATRA application on APL 18 . Interestingly, the anti-tumour effects of celastrol have been consistently attributed to its ability to induce apoptosis in AML and APL NB-4 cells 15,[19][20][21] , but the mechanism is poorly understood.
HL-60 cells is a widely used model system for studying the molecular events of AML, which lack the t(15;17) translocation characteristic of most cases of APL 13,22,23 . However, HL-60 can respond to ATRA 22 , which is widely used as a cell line in the APL studies [24][25][26][27] . In our previous study 28 , consistent to previous reports 15,[19][20][21] , we also found celastrol caused apoptosis in HL-60 cells, indicating the key role of apoptosis in the effect of celastol in the treatment of acute leukemia.
Metabolomics, the systematic measurement and biological interpretation of metabolites within a biological sample, is used to study small molecules and is an integral technology for understanding the function of biological systems. Surveying these small molecules provides an overall understanding of biological mechanisms, thereby creating a more complete picture of the phenotype (the observable characteristics of a living system). In our previous study, we used metabolomics to study the underlying mechanism in HL-60 cells in vitro, and identified uridine was the most notable changed metabolite after celastrol treatment 28 . However, the metabolome changes in vivo are unknown. As we know, pathogenesis and therapeutic target of leukemia may be not limited in one pathway. Different and complementary conclusions may be reached by using omics analyses of in vitro and in vivo samples. The hypothesis of this study was that key metabolism changes extracted from metabolome of animal model could reveal mechanism underlying celastrol-induced apoptosis in AML, especially APL. Therefore, in the present study, we treated xenograft HL-60 cell-bearing nude mice with celastrol and used metabolomics to identify the key metabolic changes in tumour tissues in vivo. To validate the key findings with the focus on APL, we also applied NB-4 cell according to previous report 26 . NB-4 is a APL cell line, which has the t(15;17) translocation and also has response to ATRA 26. The role of key metabolic pathway in apoptosis was verified by disrupting the key metabolic changes in both cell lines, revealing that cysteine metabolism plays a key role in celastrol-induced ROS-dependent mitochondrial apoptosis in HL-60 and NB-4 cells.

Celastrol inhibited tumour growth and increased apoptotic gene expression in HL-60 cells in vivo.
In the nude mouse xenograft tumour model, 2 mg/kg celastrol decreased tumour volume and weight ( Fig. 1a-c). The TUNEL assay results showed that celastrol induced apoptosis in tumours (Fig. 1d). In addition, we found increased protein levels of p53, Bax, cleaved caspase 9 and cleaved caspase 3 in the celastrol treatment group in vivo (Figs. 1e, S2-S5).
Metabolomics profiles showed that cysteine metabolism was the key metabolic alteration after celastrol treatment in HL-60 cells in vivo. A total of 129 metabolites were found in tumour tissues (Supplementary Material: Fig. S6). After false discovery rate (FDR) correction, 121 metabolites were found to be significantly changed (q<0.05) (Supplementary Material: Table S1). A total of 43 metabolites significantly increased, while 78 significantly decreased. Principal component analysis (PCA) is an unbiased and unsupervised model that can reveal global celastrol-induced metabolic changes in tumour tissue. We established a 3D PCA model and found good separation between the treatment and control groups (Fig. 2a). The above results indicate that along with the significant growth-inhibitory and pro-apoptotic effect of celastrol on tumour tissue, the metabolome was also dramatically disrupted in tumour tissue, a quite important finding in the search for the underlying mechanism. Among the metabolites, L-cysteine and adenine showed the most marked change. These metabolites were significantly decreased-indeed, totally undetectable-after celastrol treatment in tumour tissues, indicating the importance of L-cysteine metabolism and adenine metabolism in celastrol-induced apoptosis in tumour tissues. To further identify key metabolic changes, the changed metabolites were imported into the "enrichment analysis" module (www.metaboanalyst.ca/), and protein biosynthesis and glutathione metabolism were identified as the key changed pathways (Fig. 2b, Supplementary Material: Table S2). Importantly, glutathione metabolism was directly linked to L-cysteine metabolism; in addition, acetylcysteine and homocysteine, the upstream metabolites of L-cysteine metabolism, were also significantly decreased after celastrol treatment (Fig. 2c,d). These results indicated the importance of cysteine metabolism in HL-60 cells in vivo after celastrol treatment. Raw metabolomic data can be found in the Supplementary Material: Table S3.
Celastrol decreased L-cysteine levels and induced significant accumulation of ROS in HL-60 cells, which could be reversed by acetylcysteine. To verify the change in cysteine metabolism in vivo, we analysed L-cysteine in HL-60 cells in vitro and found that it was decreased after celastrol treatment in a dose-related manner, as assessed by Spearman correlation analysis (rs = −0.4522, p = 0.0265) (Fig. 3a). As www.nature.com/scientificreports www.nature.com/scientificreports/ cysteine and its related glutathione metabolism were enriched in vivo and the accumulation of intracellular ROS is one of the most important upstream stimuli of p53 activation in apoptosis 29 , the above metabolomics findings prompted us to focus on the intracellular ROS level after celastrol treatment, as ROS might be the intermediate linking the observed deficiency in oxidized glutathione and its upstream metabolites with the decreased anti-oxidative capacity and increased apoptosis in HL-60 cells after celastrol treatment. As shown in Fig. 3b, ROS was detected in control HL-60 cells, and the ROS positive control reagent Rosup led to a dramatic increase in the ROS level in the treated cells, indicating the efficiency of the ROS detection method. The intracellular ROS level was significantly increased in a dose-dependent manner after celastrol treatment (Fig. 3b). Acetylcysteine is an www.nature.com/scientificreports www.nature.com/scientificreports/ upstream metabolite of cysteine metabolism and was a significantly decreased metabolite after celastrol treatment (Fig. 2). Acetylcysteine exerts an anti-oxidant effect related to its role as a metabolic precursor of glutathione 30 . Based on the metabolomics findings, we used acetylcysteine to disrupt cysteine metabolism in order to verify the www.nature.com/scientificreports www.nature.com/scientificreports/ connection among cysteine metabolism, ROS and apoptosis. The ROS level in HL-60 cells treated with both celastrol and acetylcysteine was drastically decreased to the level in control cells, indicating that cysteine metabolism plays a key role in the induction of ROS by celastrol (Fig. 3c). Fig. 3d, celastrol treatment induced significant apoptosis of HL-60 cells, and acetylcysteine treatment exerted no effect on apoptosis. The apoptosis rate of cells treated with both celastrol and acetylcysteine was drastically decreased to the level of control cells, indicating that cysteine metabolism plays a key role in celastrol-induced apoptosis of HL-60 cells.

Acetylcysteine reversed celastrol-induced apoptosis of HL-60 cells. As shown in
Acetylcysteine reversed the dramatic decrease in the mitochondrial membrane potential and upregulation of apoptosis-related proteins. ROS are mainly produced by mitochondria 31 . To further identify the intermediate linking cysteine metabolism, ROS and apoptosis, we measured the mitochondrial membrane potential. The mitochondrial membrane potential was significantly decreased after celastrol treatment, and this decrease was dramatically reversed by the addition of acetylcysteine, indicating that cysteine metabolism-initiated, ROS-triggered apoptosis might occur through the mitochondrial pathway after celastrol treatment (Fig. 3e). We next assessed the effect of celastrol on the protein levels of p53, Bax, cleaved caspase 9 and cleaved caspase 3. As shown in Figs. 3f, S7-S10, the protein levels of p53, Bax, cleaved caspase 9 and cleaved caspase 3 were significantly increased after celastrol treatment, and this increase was dramatically inhibited in the presence of acetylcysteine, indicating that cysteine metabolism plays a role in celastrol-induced mitochondrial apoptosis in HL-60 cells. whether celastrol operated through the same metabolic pathway to induce apoptosis in different model systems, we used human APL NB-4 cells. Similar to its effects on HL-60 cells, celastrol induced ROS accumulation and apoptosis in NB-4 cells; effects were reversed by acetylcysteine (Fig. 4a,b). In addition, L-cysteine, the key metabolite of cysteine metabolism, was significantly decreased after celastrol treatment, validating the changes in cysteine metabolism in different cell models (Fig. 4c).

Discussion
In the xenograft tumour mouse model, celastrol inhibited tumour growth, induced apoptosis, and upregulated pro-apoptotic proteins (Fig. 1). Unbiased metabolomics profiling revealed that a decrease in the metabolites involved in cysteine metabolism was the key metabolic change after celastrol treatment (Fig. 2). Cysteine metabolism and its derived metabolites are related to cells' anti-oxidative stress capacity 32 , and the final product of cysteine metabolism is glutathione. Glutathione is an important biological anti-oxidant and can protect important cellular components from ROS-induced damage 33 . ROS are formed by the transfer of one electron from a redox donor to molecular oxygen (O 2 ) 31 . Emerging evidence suggests that ROS are critically involved in cancer cell functions 34 . Cancer cells, which have higher levels of basal ROS than normal cells, are likely to be more vulnerable to damage by further ROS insults 35 . Hypothesis-free screening indicated that the decrease in the anti-oxidative capacity and accumulation of intracellular ROS in HL-60 cells were key events after celastrol treatment. The increased ROS level was then verified in vitro (Fig. 3).
In this study, we found that the mitochondrial membrane potential significantly decreased after celastrol treatment (Fig. 3). Mitochondria, one of the most important cellular organelles, can produce energy, generate ROS, regulate cell signalling and participate in biosynthetic metabolism 36 . Mitochondria can impart considerable flexibility to tumour cell survival and growth under harsh environmental conditions, such as nutrient depletion, hypoxia, and cancer treatment exposure, and many classical hallmarks of cancer result in altered mitochondrial function. Therefore, mitochondria are key factors in tumourigenesis and treatment.
The tumour suppressor p53 plays a pivotal role in regulating the cell cycle, genomic integrity and apoptosis 37 . Among signalling components, p53 has one of the highest sensitivities to the cellular redox state 38 . p53 mutations often occur in most cancers. However, in the vast majority of AML cells, wild-type p53 is retained 39 . ROS may attack proteins and nucleic acids, and this event is followed by upregulation of wild-type p53. Upregulation of p53 was also verified in this study after celastrol-induced elevation of intracellular ROS levels (Fig. 3). Notably, as a sensor of cellular stress, p53 is a critical initiator of the intrinsic apoptotic pathway under robust or sustained stress 40 .
Apoptosis plays a critical role during various physiological (vertebrate development and postnatal tissue homeostasis) and pathological processes, including tumourigenesis. Resistance to apoptosis is a hallmark of cancer cells 41,42 . Therefore, targeting apoptosis for anticancer therapy is promising. The mitochondrial pathway of apoptosis is activated following cellular stress (DNA damage, nutrient deprivation, endoplasmic reticulum stress, and so on) 43 . The mitochondrial pathway of apoptosis is essential for metazoan development, tissue homeostasis, and cellular responses to therapeutics 43 . In human cancers, signalling via the mitochondrial (intrinsic) pathway of apoptosis is frequently impaired 44 . BCL-2 family proteins, especially Bax, are the central regulators of the intrinsic apoptotic pathway via the control of mitochondrial outer membrane permeabilization (MOMP) 45 , which is frequently the decisive event preceding cell death 46 . In the present study, we verified the pro-apoptotic effect of celastrol on HL-60 cells via the mitochondrial pathway (Fig. 3). We further found the clearance of ROS reversed the upregulation of p53 and the increased apoptosis induced by celastrol (Fig. 3). The elevated ROS is an activating signal for p53-mediated apoptosis in leukemia 35,47 . Therefore, the elevation of ROS induced p53-mediated mitochondrial apoptosis in leukemia cells after celastrol treatment.
Acetylcysteine is an upstream metabolite of cysteine metabolism and was significantly decreased after celastrol treatment (Fig. 2). Acetylcysteine can support cysteine metabolism to increase glutathione production in order to increase the anti-oxidative stress capacity 30 . In this study, when acetylcysteine was added, the mitochondrial membrane potential was dramatically recovered, and the increases in intracellular ROS, p53 expression and apoptosis were inhibited (Fig. 3), verify ing the involvement of a cysteine metabolism/ROS/p53/Bax/caspase 9/caspase 3 pathway in the apoptosis of HL-60 cells after celastrol treatment.
Studying the effect and underlying mechanism in NB-4 cell lines could enhance the significance of the current findings in APL 26 . In this study, the effect of celastrol on cysteine metabolism and the rescue effect of acetylcysteine on celastrol-induced ROS and apoptosis were also verified in NB-4 cells, indicating that celastrol operates through the same metabolic pathway to induce apoptosis in HL-60 and NB-4 cells (Fig. 4).

conclusions
In conclusion, by combining an in vivo unbiased metabolomics analysis with follow-up in vitro experiments, we found consistent results indicating that cysteine metabolism was the key metabolic change in HL-60 cells and NB-4 cells after celastrol treatment. This finding linked the depleted intracellular anti-oxidative capacity, increased ROS accumulation, and increased mitochondrial apoptosis in human HL-60 and NB-4 cells (Fig. 5), thus providing a novel underlying mechanism through which celastrol might be used to treat AML, especially APL.

Metabolomics analysis.
Sample preparation was performed according to our previous report 52 .
Homogenized tumour tissues (50 mg) (n = 9 per group) were mixed with 150 μl of water and 600 μl of methanol. Tissues were ultrasonicated for 5 min (power: 60%, pulses: 3/3), and the supernatant was obtained after centrifugation at 20,000 × g for 15 min. After drying, the residue was reconstituted for metabolomics analysis. The extracted samples were subjected to unbiased metabolomics analysis according to protocols in our previous studies 52,53 . Metabolomics analysis was performed on a UPLC Ultimate 3000 system (Dionex, Germering, Germany) coupled to a Q-Exactive Orbitrap mass spectrometer (Thermo Fisher Scientific, Bremen, Germany). The UPLC Ultimate 3000 system was used with a C18 column (Hypersil Gold, 1.9 μm, 100 mm × 2.1 mm; Thermo Fisher Scientific), which was maintained at 40 °C. Mobile phase A was 0.1% formic acid in ultra-pure water, and mobile phase B was acetonitrile acidified with 0.1% formic acid. Chromatographic separation was conducted using a multistep gradient with a flow rate of 0.4 ml/min over 15 min. The UPLC autosampler temperature was set at 4 °C, and the injection volume for each sample was 10 μl. All samples were analysed in a randomized fashion to avoid complications related to the injection order. The Orbitrap mass spectrometer was equipped with a heated electrospray source using full scan analysis (70 to 1050 amu) in both positive and negative ion models simultaneously. For both the positive and negative ion modes, the mass spectrometer parameters were set as follows: spray voltage of 3. Quality control (QC) sample was prepared by mixing each original sample according to a previous metabolomics study 54 , and was analysed in parallel with original samples.
Targeted L-cysteine analysis. HL-60 cells and NB-4 cells were plated at a density of 2 × 10 5 cells/well in six-well plates and exposed to celastrol for 24 h. Cells were harvested and washed three times with ice-cold PBS. Then, 1 ml of ice-cold 50% methanol was added, and all cells were lysed by ultrasonication for 1 min (power: www.nature.com/scientificreports www.nature.com/scientificreports/ 60%, pulses: 3/3). Following centrifugation (16,000 × g, 10 min, 4 °C), the supernatant was used for detection. L-cysteine was detected using the UPLC Ultimate 3000 system coupled to the Orbitrap mass spectrometer. The conditions of the chromatograph and mass spectrometer were the same as those described in the "Metabolomics analysis" section. The [M+H]+ ions of L-cysteine at m/z 122.02703 were monitored.

Measurement of intracellular ROS. Intracellular ROS generation was measured by a ROS assay with
DCFH-DA (Beyotime, Shanghai, China) according to a previous report 55 . Rosup was used as the positive control reagent according to the manufacturer's instructions. In brief, HL-60 cells were plated at a density of 2 × 10 5 cells/ well in six-well plates and exposed to celastrol (0.125, 0.25 and 0.5 μM) for 24 h. Then, cells were incubated with DCFH-DA (10 μM) for 30 min at 37 °C. DCFH-DA is hydrolysed by intracellular esterase to produce a nonfluorescent DCFH product, that can then be oxidized by ROS to produce a highly fluorescent DCF product. For the follow-up study, HL-60 cells and NB-4 cells were treated with 0.1% DMSO, celastrol (0.5 μM) and celastrol (0.5 μM)+acetylcysteine (5 mM) for 24 h. The level of ROS was measured by flow cytometry (BD Biosciences, NJ, USA). All experiments were repeated at least three times.
Determination of apoptosis by flow cytometric analysis. HL-60 and NB-4 cells were treated with 0.1% DMSO, acetylcysteine (5 mM), celastrol (0.5 μM) and celastrol (0.5 μM)+acetylcysteine (5 mM) for 24 h. The concentration of acetylcysteine was determined according to a previous report 56 . The percentage of apoptotic cells was assessed using a FITC Annexin V Apoptosis Detection Kit I (BD Biosciences, NJ, USA) as described in the manufacturer's instructions.
Measurement of the mitochondrial membrane potential. The mitochondrial membrane potential was measured with a JC-1 Mitochondrial Membrane Potential Assay Kit (Yeasen, Shanghai, China) according to the manufacturer's instructions, as described previously 57 . In brief, HL-60 cells were treated with 0.1% DMSO, acetylcysteine (5 mM), celastrol (0.5 μM) and celastrol (0.5 μM) + acetylcysteine (5 mM) for 24 h. Then, cells were incubated with JC-1 for 20 min in the dark at 37 °C. After being washed twice with staining buffer, cells were analysed by flow cytometry (BD Biosciences, NJ, USA). JC-1 forms aggregates and emits red fluorescence to indicate a normal mitochondrial membrane potential. This red fluorescence was collected as the PE signal, while the accumulation of green fluorescent monomers due to depolarization of the mitochondrial membrane potential was indicated by an increase in FITC fluorescence.
Statistical analyses. Statistical analysis of the data was performed using the Stata statistical package (Version 9.2, Stata Corp, LP); "R" (http://cran.r-project.org/), a freely available open source software package; and SIMCA-P software (Umetrics, Sweden). The metabolomics data were imported into SIMCA-P software for multivariate analysis. All data were UV-scaled and automatically log transformed by the software. PCA was applied to identify the metabolic changes between groups. The differences in metabolomics data between the treatment group and control group were identified with t-tests. To improve the statistical robustness, we used the FDR correction for multiple testing. We imported the changed metabolites into the "enrichment analysis" module to analyse the key metabolic changes (www.metaboanalyst.ca/). Statistically significant differences between two groups and among more than two groups were determined using a two-sided, unpaired t-test and one-way ANOVA with Dunnett's post hoc test, respectively. Values are expressed as the means ± standard errors. A p value of <0.05 was considered statistically significant.

Data availability
The raw metabolomic data are available in the electronic Supplementary Material.