Molecular Mechanisms behind Safranal’s Toxicity to HepG2 Cells from Dual Omics

The spice saffron (Crocus sativus) has anticancer activity in several human tissues, but the molecular mechanisms underlying its potential therapeutic effects are poorly understood. We investigated the impact of safranal, a small molecule secondary metabolite from saffron, on the HCC cell line HepG2 using untargeted metabolomics (HPLC–MS) and transcriptomics (RNAseq). Increases in glutathione disulfide and other biomarkers for oxidative damage contrasted with lower levels of the antioxidants biliverdin IX (139-fold decrease, p = 5.3 × 105), the ubiquinol precursor 3-4-dihydroxy-5-all-trans-decaprenylbenzoate (3-fold decrease, p = 1.9 × 10−5), and resolvin E1 (−3282-fold decrease, p = 45), which indicates sensitization to reactive oxygen species. We observed a significant increase in intracellular hypoxanthine (538-fold increase, p = 7.7 × 10−6) that may be primarily responsible for oxidative damage in HCC after safranal treatment. The accumulation of free fatty acids and other biomarkers, such as S-methyl-5′-thioadenosine, are consistent with safranal-induced mitochondrial de-uncoupling and explains the sharp increase in hypoxanthine we observed. Overall, the dual omics datasets describe routes to widespread protein destabilization and DNA damage from safranal-induced oxidative stress in HCC cells.


Introduction
Hepatocellular carcinoma (HCC) is the third-most leading cause of cancer death worldwide [1]. Risk factors for HCC include chronic infection with hepatitis B virus (HBV), hepatitis C virus (HCV), alcoholic liver disease, and nonalcoholic steatohepatitis (NASH). Aflatoxin-contaminated food, diabetes, obesity, certain hereditary conditions such as hemochromatosis, and various metabolic disorders increase the risk of acquiring HCC [2]. Treatment of HCC differs among patients according to their background and the tumor progression. Current therapies include transarterial embolization (TAE) (embolic particles without chemotherapy) or TACE (chemotherapeutic drugs and embolic particles), surgical resection, and liver transplantation [3]. Further research is essential to develop targeted novel therapies that address this lethal cancer and improve patient care. Natural products have been used to produce effective anticancer drugs and adjuvant chemotherapy drugs. The induction of autophagic apoptosis by naturally sourced antitumor chemicals has been recently investigated [4]. Here we focused on the compound safranal. Safranal is produced by Cuminum cyminum (cumin seed), Aspalathus linearis (rooibos), Centaurea sibthorpii, Calycopteris floribunda (ukshi), Citrus limon (lemon), Erodium cicutarium (common stork's-bill or pinweed), Sambucus nigra (elderberry), and the cyanobacterium Microcystis. The highest natural quantities of safranal can be found in saffron, the dried dark-red stigma of Crocus sativus (see Figure 1). Safranal is the most abundant molecule in saffron essential oil (~70%) [5]. Compounds in the spice saffron, including safranal, have antitumor effects in several tissues [6][7][8][9]. For example, safranal has anti-cancer activity in human colorectal cancer cells [8], human lung carcinomas [10], and neuroblastomas [11]. However, the metabolomic response of hepatocellular carcinoma cells (HCC) to endogenous saffron compounds, including safranal, has not yet been studied. Recent advances in metabolomic and transcriptomic technologies have accelerated dual omics analyses; in conjunction with advances in bioinformatics software, such as Mummichog (http://mummichog.org/, accessed on 20 March 2021), dysregulated metabolic reactions in model systems can be predicted with greater confidence. A dual omics approach to study safranal's effect on a liver cancer model would shed insight into its apoptosis induction mechanism.
We previously reported safranal-induced apoptosis in HCC cells using a HepG2 cell line [12]. The HepG2 line was chosen due to its widespread use in liver cancer research as an excellent model for HCC progression. Our extensive analyses into the mechanism behind safranal's anti-cancer mechanism were carried out using transcriptomics, cell cycle, and immunoblot analyses. We focused on critical regulators of the cell cycle, DNA damage repair, apoptosis, and effects on intrinsic and extrinsic initiator caspases and their implication in endoplasmic reticulum (ER) stress. Our work showed that safranal activates apoptosis in HCC, although further work was needed to detail the complex response to safranal and its anti-proliferative effects. We previously showed that safranal caused extensive DNA damage and elicited an unfolding response, although upstream mechanisms underlying safranal's apoptosis induction were unclear. Dual omics approaches, such as that presented in this study, offer an orthogonal route to validate potential system-wide metabolic dysregulation occurring upon safranal treatment. Yellow bubbles indicate increased, and blue indicates decreased compounds in HCC pellet extracts after safranal treatment.

Metabolite Extraction
Cells were treated with 500 uM safranal for 24 h. After 24 h, cells were centrifuged and both cells and conditioned media were frozen in liquid nitrogen. Cell pellets and conditioned media were thawed in 5 mL methanol (Sigma-Aldrich, Darmstadt, Germany) and sonicated for 30 min with a 296 W sonicator bath (Wiseclean; Witeg, Schönwalde-Glien, Germany) followed by vortexing. Cell extracts were filtered through 0.2 uM filters (Merck Millipore, Darmstadt, Germany) and stored in color-protected amber glass HPLC vials (Agilent, Santa Clara, CA, USA) at 4 • C until column loading.

LC/MS-QToF
All metabolites were dissolved in methanol to achieve separations of biomolecules on a reverse-phase C18 HPLC column. All samples were run in biological triplicates, with results from technical triplicates averaged for XCMS inputs. Eluted molecules went to an Agilent LC-MS QToF 6538 with accurate mass profiling with lock mass compounds (error < 2 ppm, Agilent, Santa Clara, CA, USA). Compounds were eluted from a reverse-phase C18 column using a semi-linear isopropanol gradient as in ref. [13], which was adapted from a shotgun lipidomics protocol [14] modified to capture a wide range of intracellular metabolites inclusive of high-molecular-weight, hydrophobic compounds. Briefly, individual components in the extracts were separated in a semi-linear hydrophobicity gradient using acetonitrile, ammonium formate, and isopropanol. The separation protocol was designed to send a wide range of intact biomolecules to a quadrupole time-of-flight mass spectrometer (qTOF-MS, Agilent, Santa Clara, CA, USA). This elution scheme was designed to elute small hydrophilic molecules first and gradually induce mobility in more massive, hydrophobic molecules such as phospholipids then triacylglycerols over 14-min runs. We achieved the separation of hydrophilic and hydrophobic compounds at the initial and final HPLC column elution endpoints using this extraction procedure and solvent strategy ( Figure 1).

Computational, Quantitative, and Statistical Analyses
Major peak clusters correspond with hydrophilic, mesophilic, and hydrophobic compounds in cell extracts. Molecular features detected by the MS were extracted for downstream untargeted metabolomics and multi-omics strategies.
Overall, 2484 compounds in the conditioned media and 5550 in the pellets of safranaltreated HCC cells had significantly different accumulation after 24 h (FDR-corrected p-value < 0.05; Figure 1 shows different metabolites at q < 0.001). HPLC column retention time correction was performed in XCMS with the adjustRtime method to align chromatograms [15][16][17][18][19]. The 'peakGroups' method was used for retention time correction from alignments of peak groups present in most/all samples. The 'adjustRtime-peakGroups' method facilitates accurate comparisons between HPLC runs.

Pathway Analysis
DEGs were analyzed through Ingenuity Pathway Analysis (IPA) (QIAGEN Inc., https://www.qiagenbioinformatics.com/products/ingenuitypathway-analysis, accessed 18 April 2021 [21]). The cutoff value for an FDR was set to 0.01. Core analysis was performed with default settings, and canonical pathways and associated toxicity functions were reviewed. Forty-five genes that were associated with HCC were re-analyzed (core analysis) for further toxicity function association. Transcriptomic data were used to retrieve ECs from PFAMs [22], and CAS numbers were obtained from metabolite assignments in METLIN (https://metlin.scripps.edu, accessed 4 May 2021). These ECs were uploaded to iPATH3 (https://pathways.embl.de/, accessed 29 May 2021) and used to generate the overlap pathways for dual omic comparisons.

Dual Omics Systems Biology Resolves the HCC Safranal Response
We investigated the impact of safranal on HCC cells (HEP-G2) in vitro from metabolomic and transcriptomic perspectives by combining evidence from both approaches to understand the complex mechanisms underlying the HCC apoptotic safranal response. HCC cells from safranal-treated and control culture experiments were partitioned for transcriptomics (RNAseq) and metabolomics (HPLC-MS) (n = 3 biological replicates for each group and in each experiment). Then, we compared control and safranal-treated HCC cells using extracted metabolites and RNA for downstream metabolomic and transcriptomic analyses.
Overall increases in identifiable biomolecules in media extracts indicated higher cell lysis levels after treatment. Hydrophilic molecules in the conditioned media and pellets after safranal treatment were significantly increased (retention time < 5 min, see Figure 1). A significant decrease in large, hydrophobic molecules, such as tri-and di-glycerides (TAGs and DAGs, respectively) in the media of the safranal-treated cells was observed (Figure 1, see also Table S2).
Metabolites were mapped to known human metabolic pathways using the XCMS Systems Biology module [17,19]. Integrated pathway analysis uses multi-scale database matching and retrieval tools, including KEGG and BioCyc, to predict metabolic pathway dysregulation from MS-detected metabolite clusters (see Figure 2). The results of this semi-automatic annotation algorithm were manually compared with transcript expression values to infer reaction directions.
We manually extracted Enzyme Commission (EC) codes from differentially expressed and accumulated transcripts and compounds and revealed reactions with overlapping evidence from protein family (PFAM)-derived ECs and Chemical Abstract Service (CAS) registry-derived ECs. The Interactive Pathways Explorer (v3) tool was used to examine overlaps from metabolomic and transcriptomic datasets. Dysregulated metabolites from our HPLC-qTOF-MS studies (p < 0.001) were matched against the CAS registry to find associated ECs in Metlin (https://metlin.scripps.edu/, accessed 4 May 2021). The ECs were derived from their PFAM correlates using HMMsearch on translated transcripts, then ECdomainMiner to extract EC numbers from PFAMs with gold-level designations [22]. A total of 653 ECs from DEGs and 68 from CAS accessions in dysregulated metabolites were identified. Twenty-three ECs overlapped in these two datasets (Figure 3). The overlapping ECs corresponded to 47 distinct reactions, including reactions in the urea cycle, fatty acid elongation in mitochondria, arachidonic acid metabolism, pyrimidine metabolism, tyrosine metabolism, and tryptophan metabolism (Figure 3). Metabolic reactions with dual omics evidence for safranal perturbation can be visualized in the iPath metabolic network map in Figure 3 and can also be found in Tables S1 and S2.  [17]. These pathways were based on metabolic pathways annotated in the BioCyc4 (biocyc.org) and Uniprot6 (uniprot.org) databases. Their respective metabolites, degree fold-change (x-axis), and mass-to-charge ratios (m/z, heatmap) are shown. Dysregulated metabolites are indicated inside or adjacent to bars for increased (top) or decreased (below) metabolites in safranal-treated cells. Multiple adducts for the same compound are shown as stacked bars.
Other metabolomic and transcriptomic evidence also implied that the mitochondrial uncoupling process (e.g., futile, thermogenic cycling) was defunct in safranal-treated HCC cells. The generation of ROS is a hallmark of unchecked productive mitochondrial electron flow, and mitochondrial decoupling is a canonical cytoprotective response to the resultant oxidative damage [32,33]. Our data suggest that safranal-treated HCC may not correctly uncouple oxidative phosphorylation and, ultimately, accumulate cytotoxic levels of adenosines. Safranal was recently shown to inhibit the highly conserved ATP synthase [34]; hence, the accumulation of high levels of ATP precursors we observed may be due to mitochondrial electron transport chain (mETC) blockage at the final, productive step.  also Table S2). (C) Steroid biosynthesis dysregulation after safranal treatment (see also Table S2). (D) Overlap of metabolomics and transcriptomics dysregulated metabolic reactions from CAS and EC numbers. Metabolite accessions were retrieved from the Chemical Abstract Service (CAS) using the METLIN (https://metlin.scripps.edu, accessed 4 May 2021) batch search tool, ECs were retrieved from transcripts using EC Domain Miner [22]. The CAS and EC accession numbers were compared in iPATH3 (https://pathways.embl.de/, accessed 29 May 2021) to detect dysregulated.

Safranal Induces ROS-Damage Biomarkers and Antioxidant Gene Expression
Several metabolite and transcript dysregulation events coordinately suggested a prooxidant environment after safranal treatment. The up-regulation of an antioxidant defense network was outlined in our transcriptomic data, and increases in biomarkers for ROS damage were increased in our metabolomics data. For example, oxidized glutathione was significantly increased after treatment (GSSG; fold-increase = 236.6, p = 1.6 × 10 −5 ).
The protein KEAP1 was significantly up-regulated (base mean = 1526.5, t = 24.9, p = 0). KEAP1 interacts with NRF2, a master regulator of the antioxidant response [37][38][39][40]. The KEAP1 up-regulation reveals an antioxidant response to safranal in HCC cells and suggests that safranal induces a reactive oxygen species (ROS)-rich intracellular environment. Transcripts for the spermidine/spermine acetyltransferase (SAT1; SPD/SPM acetyltransferase), a key enzyme in the polyamine breakdown pathway, significantly decreased (6.1-fold decrease, p = 1.65 × 10 −193 ). SAT1 acetylates spermidine and modulates polyamine levels by regulating their transport at the cell membrane [41]. The downregulation of SAT1 implies that polyamine processing at the cell membrane is not functioning correctly.

Discussion
This study used a dual omics workflow to study the phenotypic and transcript expression in HCC to safranal. Pathways involved in the oxidative stress response were predominant features with dysregulation evidence from both omics datasets. Their coordination reveals the apoptosis-inducing effects of safranal through redox instability and shifts in energy homeostasis, ultimately leading to widespread protein destabilization and DNA damage.
The evidence for an activated antioxidant defense in liver cells by a canonical mechanism, coupled with significantly increased oxidized GSSG, hypoxanthine, and various other ROS-damage biomarkers, outlines a highly pro-oxidant, ROS-rich environment induced by safranal. Farahmand et al. suggested that safranal acts as a minor pro-oxidant that activates endogenous antioxidant defenses [46]. The mechanism underlying safranal's induction of ROS may be based on its inactivation of endogenous antioxidant systems. For example, exogenously applied safranal inhibits catalase in plants [47]. Aside from catalase, the biliverdin-bilirubin cycle is important to maintain redox homeostasis in mammals [48]. When reduced by BLVRB, reactive oxygen species can be formed. BLVRB is a promiscuous enzyme and catalyzes many other pyridine-nucleotide-dependent reductions, including various flavins, biliverdins, quinone, and iron [49]. The reduction of biliverdin by biliverdin reductase A (BLVRA) to bilirubin is a central mechanism for redox regulation in humans; depletion of BLVRA sensitizes human diploid fibroblasts to oxidative stress [50]. Thus, the cycling of biliverdin to bilirubin and back constitutes a redox-modulation circuit that also appears to be perturbed after safranal treatment.
The decrease in the ubiquinol precursor 3-4-dihydroxy-5-all-trans-decaprenylbenzoate (3-fold decrease, p = 1.9 × 10 −5 ) signifies the loss of an important electron carrier in the mETC. With ubiquinol-10 in the mETC, electrons from upstream mETC components are more likely to result in ROS from electron leak. MTA is a substrate for MTA phosphorylase (EC 2.4.2.28) and produces adenine and S-methyl-5-thio-alpha-D-ribose 1-phosphate [51]. A further downstream product, adenylthiomethylpentose, lowers mammalian body temperature [52]. The increase of MTA suggests that intracellular cryopreservation mechanisms, such as non-shivering thermogenesis mediated by mitochondrial uncoupling, may decrease after safranal treatment.
Nitrogen metabolism appeared to be heavily affected by safranal. Urea, arginase, and nitric oxide (NO) precursor metabolic reactions were significantly dysregulated from transcriptomics and metabolomics evidence. Their disruption of these pathways leads to purine metabolism defects [53,54], in essence, simulating the genetic defects described by Pang et al. The inability to remove excess nucleotides or convert them to downstream products may be responsible for the hypoxanthine accumulation and subsequent oxidative damage in the HCC HEP-G2 cells. The significant increases observed in purine derivatives may be due to interruption of the mETC, which eliminates thermogenesis and promotes the accumulation of excessive ATP or ADP and reactive oxygen species levels due to the resulting electron leaks.

Conclusions
Overall, protein destabilization that leads to ROS generation appears to be a hallmark of safranal's mode of action on HCC cells. Save et al. found that safranal caused the inhibition and dis-aggregation of a wide variety of α-synuclein in E. coli [23]. This disassociation resulted from safranal binding to the hydrophobic parts of proteins, thus interfering with protein folding. Ultimately, interference in intracellular protein folding has a cytotoxic effect, consistent with our previous conclusions [5] and the results in this study. One of the key features we observed that may stem from protein destabilization was the accumulation of hypoxanthine. In itself, hypoxanthine has been shown to induce apoptosis through an ROS-dependant mechanism by the induction of oxidative stress in endothelial cells [36]. Thus, protein destabilization leading to hypoxanthine accumulation appears to be the primary mechanism of cytoxicity from safranal treatment of HCC cells.
Supplementary Materials: The following supporting information can be downloaded at: https: //www.mdpi.com/article/10.3390/antiox11061125/s1, The following are available online: Table  S1: Differentially-expressed genes and their predicted EC accessions, Table S2: tabular summary of XCMS suite results for triplicates of each control and treated cells, including m/z, fold-change, and retention times. Data S1: Hosted at zenodo.org

Conflicts of Interest:
The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.