Pharmacogenomic Characterization and Isobologram Analysis of the Combination of Ascorbic Acid and Curcumin—Two Main Metabolites of Curcuma longa—in Cancer Cells

Curcuma longa has long been used in China and India as anti-inflammatory agent to treat a wide variety of conditions and also as a spice for varied curry preparations. The chemoprofile of the Curcuma species exhibits the presence of varied phytochemicals with curcumin being present in all three species but AA only being shown in C. longa. This study explored the effect of a curcumin/AA combination on human cancer cell lines. The curcumin/AA combination was assessed by isobologram analysis using the Loewe additivity drug interaction model. The drug combination showed additive cytotoxicity toward CCRF-CEM and CEM/ADR5000 leukemia cell lines and HCT116p53+/+ and HCT116p53−/− colon cancer cell line, while the glioblastoma cell lines U87MG and U87MG.ΔEGFR showed additive to supra-additive cytotoxicity. Gene expression profiles predicting sensitivity and resistance of tumor cells to induction by curcumin and AA were determined by microarray-based mRNA expressions, COMPARE, and hierarchical cluster analyses. Numerous genes involved in transcription (TFAM, TCERG1, RGS13, C11orf31), apoptosis-regulation (CRADD, CDK7, CDK19, CD81, TOM1) signal transduction (NR1D2, HMGN1, ABCA1, DE4ND4B, TRIM27) DNA repair (TOPBP1, RPA2), mRNA metabolism (RBBP4, HNRNPR, SRSF4, NR2F2, PDK1, TGM2), and transporter genes (ABCA1) correlated with cellular responsiveness to curcumin and ascorbic acid. In conclusion, this study shows the effect of the curcumin/AA combination and identifies several candidate genes that may regulate the response of varied cancer cells to curcumin and AA.


INTRODUCTION
Curcuma longa L. belongs to the family Zingiberaceae, which is a perennial herb that measures up to 1 m height with a short stem. It is distributed throughout tropical and subtropical regions in the world, being widely cultivated in Asian countries, mainly in India and China (Kapoor, 1990). As a component of folklore medicine the use of C. longa has been documented both in Indian and Chinese cultures. The rhizomes are used as Ezhu according to the Chinese Pharmacopoeia (2010 edition; Zhao et al., 2010) and are a household remedy in Nepal (Eigner and Scholz, 1999). The long list of usages include antiseptic, analgesic, antiinflammatory, antimalarial, and insect repellant (Chaudhri, 1950;Li et al., 1998;Niederau and Göpfert, 1999;Tawatsin et al., 2001;Duke, 2002). Traditional Indian medicine claims to use its powder against biliary disorders, anorexia, coryza, cough, diabetic wounds, hepatic disorders, rheumatism, and sinusitis (Ammon et al., 1992). In ancient Hindu medicine, C. longa is extensively used for the treatment of sprains and swelling caused by injury (Ammon and Wahl, 1991). The antioxidant activity of C. longa is well-known (Ammon and Wahl, 1991;Anand et al., 2007).
The activity of many medicinal plants results from the interaction action of several constituents, which may cooperatively act in an additive or synergistic manner. It has been repeatedly observed that extracts of medicinal plants reveal better activities than their isolated single compounds at comparable equivalent concentrations of the active components. This phenomenon is attributed to the absence of interacting substances present in crude extracts. For example, extracts were obtained from the fresh herb of Artemisa annua L. either by soaking the herb in water followed by wringing out the juice by hand or by pounding the fresh herb to a pulp followed by squeezing out the juice. The extracts were then analyzed for artemisinin concentration and tested against malaria parasites.
It was found that the antiplasmodial IC 50 -values were 6-18-fold lower than was expected in terms of their artemisinin content suggesting that the activity of the extracts could not be entirely accounted by their artemisinin content (Wright et al., 2010). Another example are the Cinchona alkaloids. There are almost 30 alkaloids described in the bark of Cinchona officinalis L. The most well-known of these are quinine, quinidine, and cinchonine and cinchonidine. However, quinine is not the most potent of the alkaloids: quinidine, dihydroquinidine, and cinchonine all have consistently lower 50% inhibitory concentrations (IC 50 ) in vitro. The combination of quinine with quinidine and cinchonine is 2-10 times more effective in vitro against quinine-resistant strains, and the mixture of alkaloids reveals more consistent effects than any of the alkaloids singly used (Druilhe et al., 1988;Karle and Bhattacharjee, 1999). In pharmacokinetic synergy, substances with little or no bioactivity may assist the main active principle to reach the disease target by several mechanisms, e.g., improving bioavailability, or decreasing metabolism and excretion. Comparable effects are not yet known for C. longa and needs to be elucidated. In the present investigation, we addressed this question. We have chosen ascorbic acid (vitamin C, ascorbate, C 6 H 12 O 6 ) as phytochemical constituent of C. longa to investigate the additive or synergistic effects of its combination with curcumin.
Ascorbic acid is a ketolactone and a water-soluble antioxidant. There are two chemical forms of ascorbic acid: the reduced form (ascorbic acid; AA) and the oxidized form (dehydroascorbic acid; DHA; Mamede et al., 2011). AA is well-known for its potent antioxidant properties, as it is able to scavenge free radicals and reactive oxygen species (ROS). Thus, it has been associated with decreased oxidative stress in vivo Carr A. C. and Frei, 1999). High doses of AA can reduce inflammatory biomarkers such as C-reactive protein (CRP), tumor necrosis factor (TNF-α), interferon-γ (IFN-γ), and the interleukins IL-1, IL-2, IL-6, IL-8 (Gilliam and St Clair, 2011;Mikirova et al., 2012). AA is a cofactor in vivo for enzymes involved in the biosynthesis of collagen, carnitine, neurotransmitters, and neuropeptide hormones as well as enzymes involved in regulation of epigenetic or transcription factors (Rebouche, 1991;Du et al., 2012). Furthermore, it is a cofactor in the synthesis of the neurotransmitters norepinephrine, dopamine, and serotonin and neuropeptide hormones such as oxytocin (Harrison and May, 2009;May et al., 2012). Extensive studies have been carried out on AA in the treatment of cancer and vast literature exists on AA and cancer. In 1949, AA was first proposed to be used for cancer therapy (Klenner, 1949;Mc, 1952). The first comprehensive review on AA and cancer was published in 1979 and an updated review 25 years later (Cameron et al., 1979;González et al., 2005). AA may act as a prodrug causing the formation of AA radical and hydrogen peroxide in the extracellular space (Chen et al., 2007). In a clinical trial, AA exerted antitumor activity in patients with advanced cancer as a stand-alone therapy as well as in combination with other anticancer agents (Hoffer et al., 2008). The conjugation of AA with extracts of medical herbs stimulated apoptosis and disrupted the cell cycle in different cancer cell lines. Furthermore, AA was pro-oxidant generating hydrogen peroxide-dependent cytotoxicity toward various cancer cells without adversely affecting normal cells. AA together with sodium nitrite induced genotoxicity due to oxidative DNA damage. High concentrations of AA killed tumor cells in vitro with high efficiency and inhibited angiogenesis in mice bearing sarcoma (Chen et al., 2008;Kuroiwa et al., 2008;Verrax and Calderon, 2009;Yeom et al., 2009;Rozanova Torshina et al., 2010).
In this paper, we have chosen AA because it is widely distributed in many plants and it is also synthetically available. This allows us to investigate the cytotoxicity of the combination of curcumin and AA outside of the plant at exactly defined conditions to clarify their pharmacological effects in the combination. Our strategy was to investigate the combination by isobologram analysis and to investigate the genes which are related to the cytotoxicity induced by curcumin and AA in a broad spectrum of cancers. We postulate that the identification of genes that are specifically regulated by AA and curcumin could improve the understanding of the efficacy of C. longa in cancer treatment. We have systematically studied the microarray-based mRNA expression of genes, which influence the cellular response to curcumin and AA in the tumor cell line panel of the National Cancer Institute (NCI), USA to find possible mechanistic explanations for the interaction of these two compounds.

Chemicals
All chemicals were of analytical grade. Curcumin, AA and DMSO were purchased from Sigma-Aldrich (Sigma-Aldrich Corp., St. Louis, MO, USA).

COMPARE and Hierarchical Cluster Analyses of Microarray Data
The cancer cell lines of the Developmental Therapeutics Program of NCI consisted of a series of non-small cell lung cancer, colon cancer, renal cancer, ovarian cancer cells, leukemia, melanoma, prostate carcinoma, breast cancer, and tumor cells of the central nervous system. Their origin and processing have been previously reported (Alley et al., 1988). The cytotoxicity induced by AA, curcumin, and standard anticancer drugs of the NCI cell line panel was measured by the sulforhodamine B assay. The 50% inhibition concentrations calculated from doseresponse curves and converted to logarithmic values [log 10 IC 50 (M)] have been deposited at the NCI database (http://dtp. cancer.gov/databases_tools/default.htm). The mRNA microarray hybridization of the NCI cell lines has been reported and deposited at the NCI website (http://dtp.cancer.gov/databases_ tools/default.htm). COMPARE analyses were performed to produce rank ordered lists of genes expressed in the NCI cell lines. The methodology has been previously described in detail as a tool to identify candidate genes for drug resistance and sensitivity. To derive COMPARE rankings, a scale index of correlation coefficients (R-values) was created from log 10 IC 50 (M) values of test compounds and microarray-based mRNA expression values. Greater mRNA expression correlated with enhanced drug resistance in the standard COMPARE approach, whereas greater mRNA expression in cell lines indicated drug sensitivity in reverse COMPARE analyses. Pearson's correlation test was used to calculate significance values and rank correlation coefficients as a relative measure for the linear dependency of two variables.
For hierarchical cluster analysis, objects were classified by calculation of distances according to the closeness of between individual distances by means of hierarchical cluster analysis. All objects were assembled into cluster trees (dendrograms). Merging of objects with similar features leads to cluster formation, where the length of the branch indicates the degree of relation. Distances of subordinate cluster branches to superior cluster branches serve as criteria for the closeness of clusters. Thus, objects with tightly related features were clustered closely together, while separation of objects in the dendrogram increased with progressive dissimilarity. Hierarchical clustering and heatmap analysis were performed using Euclidean distance and ward method implemented in "dist, " "hclust, " and "heatmap" functions in R programming (Eisen et al., 1998;Gu et al., 2016). The results were further confirmed using the CIM miner software by use of the one matrix clustered image map (CIM) https://discover.nci. nih.gov/cimminer/oneMatrix.do.

Cell Viability Assay
Cell viability was evaluated by resazurin assay. This test is based on reduction of the indicator dye, resazurin, to the highly fluorescent resorufin by viable cells. Nonviable cells rapidly lose the metabolic capacity to reduce resazurin and thus produce no fluorescent signal. Glioblastoma and colon cancer cells were harvested with 0.25% trypsin/EDTA (Invitrogen, Germany) and diluted to a final concentration 5 × 10 4 cells/mL. One hundred microliters of the cell suspension were sowed into the wells of a 96-well-culture plate 1 day before treatment. However, for the leukemia cell lines 2 × 10 4 cells were sowed in a 96well-culture plate in a total volume of 100 µL for each well and then the cells were immediately treated. Marginal wells were filled with 200 µL of pure medium, in order to minimize effects of evaporation. Besides, wells filled with medium served as the negative control to determine background fluorescence that may be present. Then, cells were treated with different concentrations of curcumin, vitamin C alone, or combined. After 72 h, 20 µL resazurin (Sigma-Aldrich, Germany) 0.01% w/v in ddH 2 O was added to each well and the plates were incubated at 37 • C for 4 h. Fluorescence was measured on an Infinite M2000 Proplate reader (Tecan, Germany) using an excitation wavelength of 544 nm and an emission wavelength of 590 nm. Each assay was done at least two times, with six replicates each. The cytotoxic effect of the treatment was determined as percentage of viability and compared to untreated cells. The calculated cell viability (y-axis) was plotted against the log drug concentration (x-axis) using Microsoft Excel. The obtained curve was used to determine the IC 50 -value, which represented the concentration of the test compound required to inhibit 50% of cell proliferation.

Statistical Analysis
The Loewe additivity model was used to calculate synergetic drug interactions between curcumin and vitamin C in inhibiting cell growth. In this model, the combination index (CI) was defined as CI = (d1/D1)/(d2/D2), where D1 and D2 were the doses of drug 1 and drug 2 that produced an response Y (e.g., 50% inhibition of CCRF-CEM growth) when used alone, d1 and d2 were the doses of drug 1 and drug 2 in combination, which can generate the same response Y. If the CI is equal, less than or more than 1, the combination dose (d1, d2) is termed as additive, synergistic, or antagonistic, respectively. The drug interaction was illustrated geometrically as isobologram.
Pearson's correlation test was used to calculate significance values and rank correlation coefficients as a relative measure for the linear dependency of two variables. This test was implemented into the WinSTAT Program (Kalmia Co.). Pearson's correlation test determined the correlation of rank positions of values. Ordinal or metric scaling of data is suited for the test and transformed into rank positions. There is no condition regarding normal distribution of the data set for the performance of this test. We used Pearson's correlation test to correlate microarraybased mRNA expression of candidate genes with the IC 50 -values for curcumin and ascorbic acid.
The χ 2 -test was applied to bivariate frequency distributions of pairs of nominal scaled variables. It was used to calculate significance values (p-values) and rank correlation coefficients (R-values) as a relative measure for the linear dependency of two variables. This test was implemented into the WinSTAT program (Kalmia Co.). The χ 2 -test determines the difference between each observed and theoretical frequency for each possible outcome, squaring them, dividing each by the theoretical frequency, and taking the sum of the results. Performing the χ 2 -test necessitated to define cell lines as being sensitive or resistant to curcumin and ascorbic acid. This has been done by taking the median IC 50 -value (log 10 = −5.1 M) for curcumin and IC 50 -value (log 10 = −2.7 M) for ascorbic acid as a cut-off threshold.

Chemoprofiling of Different Curcuma Species
As a first step, we established chemoprofiles of three Curcuma species (C. longa, C. zedoaria, and C. xanthorrhiza) based on the chemical compositions of these species deposited at Dr. Duke's Phytochemical and Ethnobotanical Databases (http:// www.arsgrin.gov/cgi-bin/duke/farmacy2.pl). We subjected the chemical composition of these plants to hierarchical cluster analysis (Figure 1). A total of 114 phytochemicals have been included in the analysis, which are listed in detail in Supplementary Table 1. Three compounds were commonly found in all three Curcuma species (curcumin, D-camphor, and desmethoxycurcumin). Ten compounds were found in two of the three species, whereas all other compounds were found in only one Curcuma species. This specific distribution of phytochemicals enables specific clustering and separation of the Curcuma species.

Cytotoxicity of AA in the NCI panel of Cell Lines
We hypothesized that the cytotoxic effect of C. longa against cancer cells is not solely caused by its main compound, curcumin, but that other compounds may also contribute to this activity of the plant. To prove this hypothesis, we mined the NCI database for compounds found in C. longa and six compounds were identified (Figure 2A), i.e., AA, limonene, guaiacol, p-cymene, azulene, and curcumin. Although AA was not the most toxic compound among the six tested, we decided to continue our investigations with AA, because of its far distribution not only in C. longa but also in many other plants as well and it's enormous relevance for human health in general. Further, investigations were then carried out using AA in the NCI panel of cell lines. Leukemia and melanoma cell lines were most sensitive, while brain and lung cancer cell lines were the most resistant ones ( Figure 2B). Established anticancer drugs frequently show high sensitivity toward leukemia, but resistance toward melanoma. Hence, it is interesting that AA was active against melanoma cell lines.

Cytotoxicity of Curcumin and AA toward Drug Resistant Cancer Cell Lines
Drug-resistant cell lines with different resistance mechanisms (P-glycoprotein, EGFR, mutant p53) toward curcumin and AA were determined. All the cell lines were treated with varying concentrations of curcumin and AA for 72 h, their growth was inhibited in a dose-dependent manner, albeit at different efficacy. The IC 50 -values were calculated from the dose response curves and summarized in Table 1. Curcumin inhibited cell growth at lower concentrations than AA. The range of IC 50values was 5.22-58.3 µM for curcumin, while that of AA was 389.1-561.2 µM. The CCRF-CEM and CEM/ADR5000 leukemia cell lines were inhibited at concentrations of 5.22 and 6.33 µM, respectively. The U87MG. EGFR-transfectant glioblastoma cells exhibited sensitivity toward curcumin with an IC 50 -value of 46.1 µM, which was slightly lower than the  IC 50 -value of wild-type U87MG cells 49.6 µM. Interestingly, the HCT166p53 −/− colon cancer cell line was preferentially inhibited by AA with an IC 50 -value of 506.9 µM compared to HCT116p53 +/+ wild-type cells 561.2 µM. The degrees of resistance were calculated by dividing the IC 50 of the resistant cell line by the IC 50 of the sensitive cell line. Compared to the high degrees of resistance of these drug-resistant cell lines to standard drugs such as doxorubicin (Hall et al., 2009), curcumin and AA inhibited these cell lines with similar efficacies. The degrees of resistance were in a range of 0.9-1.41 (Table 1).

Cytotoxic Effects of Combination Treatments of Curcumin and AA
Next, we addressed the question, whether the combination of curcumin and AA exhibits additive or synergistic growth inhibition of cancer cells. We applied a universal reference model for evaluating the effects of drug interaction, i.e., the Loewe additivity model (isobologram analysis). The cancer cell lines were treated with varying concentrations of AA at indicated concentrations of curcumin for 72 h. In CCRF-CEM and CEM/ADR5000 cells, the IC 50 -values of curcumin in combination with AA were reduced by less than half of the IC 50 of curcumin alone. In the glioblastoma and colon cancer cell lines, the IC 50 -value of two of the curcumin concentrations (20% IC 50 curcumin and 40% IC 50 curcumin) decreased with increasing AA concentrations less than the IC 50 of curcumin alone. However, the two other concentrations of curcumin reduced the IC 50 in combination with AA by less than half of the IC 50 -value of curcumin alone. Dosenormalized IC 50 isobolograms for all cell lines were generated by plotting the combination treatment IC 50 -values of curcumin against AA. Additive effects were observed in CCRF-CEM and CEM/ADR5000 (Figure 3) as well as in HCT116p53 +/+ and HCT116p53 −/− cell lines (Figure 4), whereas supra/additive effects were visible in U87MG and U87MG. EGFR cells (Figure 5).

COMPARE and Hierarchical Cluster Analyses of mRNA Expressions
The transcriptome-wide mRNA expression of the NCI cell lines based on the Novartis microarray platform was investigated by COMPARE analyses and correlated to the log 10 IC 50 (M) values for curcumin and AA. This bioinformatical approach was performed to identify novel putative factors associated with cellular response to curcumin and AA. The top 20 genes with direct and top 20 genes with inverse correlation coefficient are shown in Tables 2, 3. These genes were subjected to hierarchical cluster analysis to analyze, whether the expression profiles of these genes may predict sensitivity or resistance of the cells to curcumin and AA. The mRNA expression of the identified genes were subjected to hierarchical cluster analysis and cluster image mapping (Figures 6, 7). The resulting dendogram with the cell lines analyzed on the left can be divided into five major clusters for curcumin and four clusters for ascorbic acid. Using the chi-square test, we analyzed whether the distribution of cell lines being sensitive or resistant to curcumin and AA was statistically significant. As shown in Table 4, the distribution of sensitive or resistant cell lines on the dendogram was significantly different indicating that cellular response to curcumin or AA was predictable by the mRNA expression of these genes. Therefore, it is interesting to know the function of these genes. The specific functions of the proteins encoded by the genes were diverse and included signal transduction, transcription factors, proteasome deregulation, apoptosis regulating genes, proliferation-related genes, proas well as anti-oxidative genes (Tables 2, 3). AA induced genes that belonged to the functional groups of transcription factors (TFAM, TCERG1, RGS13, and C11orf31), apoptosisregulating genes (CRADD, CDK7, CDK19, CD81, TOM1) and signal transduction genes (NR1D2, HMGN1, ABCA1, DE4ND4B, TRIM27). Curcumin induction resulted in DNA repair genes (TOPBP1, RPA2), mRNA metabolism genes (RBBP4, HNRNPR, SRSF4, NR2F2, PDK1, and TGM2), signal transduction genes (WWC1, DTX2, EGFR, CHRNB4, VPS41, CRIM1), proliferationrelated genes (RHOD), apoptosis-regulating genes (DFFB), and transporter genes (ABCA1).

DISCUSSION
Plants are perhaps earth's most accomplished chemists. They produce thousands of specialized secondary metabolites, and during the evolution of life plants developed multi-targeted chemicals to fulfill diverse tasks. C. longa contains some 30

U10868
Aldehyde dehydrogenase 3 family, member B1 Oxidizes medium and long chain saturated and unsaturated aldehydes. Protective role against the cytotoxicity induced by lipid peroxidation.

M62896
Annexin A2 pseudogene 1 Involved in cell proliferation and membrane physiology and is related to cancer progression.

BAIAP2
AB015020 BAI1-associated protein 2 Adapter protein that links membrane-bound small G-proteins to cytoplasmic effector proteins.

BAIAP2
AB015019 BAI1-associated protein 2 Adapter protein that links membrane-bound small G-proteins to cytoplasmic effector proteins. Participates in actin bundling, if associated with EPS8. Promoting filopodial protrusions.

BTFA1
AF038362 B-TFIID transcription factor-associated, 170kDa Regulates transcription in association with TATA binding protein (TBP). Removes TBP from the TATA box.

M23254
Calpain 2 (m/II) large subunit Calcium-regulated non-lysosomal thiol-protease, which catalyzes proteolysis of substrates involved in cytoskeletal remodeling and signal transduction.

U48861
Cholinergic receptor, nicotinic, beta 4 Leads to opening of an ion-conducting channel across the plasma membrane. Involved in vesicular transport in the osteoclast. Role in sialyl-lex-mediated transduction of apoptotic signals.

(Continued)
Frontiers in Pharmacology | www.frontiersin.org different phytochemicals. In the present study, we analyzed the cytotoxicity of a combination of two of these phytochemicals of C. longa, curcumin and AA. If applied alone, curcumin and AA were cytotoxic toward cell lines of different tumor types with curcumin exhibiting stronger cytotoxicity than AA. Our results are in line with other reports on the inhibitory activity of curcumin (Bimonte et al., 2016;Guzzarlamudi et al., 2016;Kasi et al., 2016;Ye et al., 2016;Yu et al., 2016;Zeng et al., 2016) and AA Fukui et al., 2015;Jacobs et al., 2015;Sunil Kumar et al., 2015;Venturelli et al., 2015). The anticancer effects of curcumin in vitro and in vivo are primarily due to the activation of apoptotic pathways in cancer cells as well as the inhibition of mechanisms related to the tumor microenvironments such as inflammation, angiogenesis, invasion, and metastasis. In particular, curcumin targets numerous therapeutically important cancer signaling pathways such as p53, Ras, PI3K, AKT, Wnt, β-catenin, mTOR, and so on. AA also reveals anticancer activity in vitro and in vivo, however at higher concentration than curcumin. A number of suggestions have been put forth on the potential mechanisms, by which AA causes death of cancer cells. The most common is Binds to the inner side of the nucleosomal DNA thus altering the interaction between the DNA and the histone octamer which maintains transcribable genes in a unique chromatin conformation.
HNRNPR AF000364 Heterogeneous nuclear ribonucleoprotein R Component of ribonucleosomes, which are complexes of >20 other heterogeneous nuclear ribonucleoproteins (hnRNP). hnRNP play an important role in processing of precursor mRNA in the nucleus.

KLHL35
AA471042 Kelch-like family member 35 A protein coding gene.

K02566
Kininogen 1 Role in blood coagulation. Inhibits the thrombin-and plasmin-induced aggregation of thrombocytes.

LRP10
AL080164 Low density lipoprotein receptor-related protein 10 Receptor involved in the internalization of lipophilic molecules and/or signal transduction. May be involved in the uptake of lipoprotein APOE in liver.

MAP2K2
L11285 Mitogen-activated protein kinase kinase 2 Catalyzes the concomitant phosphorylation of threonine and tyrosine residue in a Thr-Glu-Tyr sequence located in MAP kinases. Activates ERK1 and ERK2 MAP kinases.

MEGF8
AB011541 Multiple EGF-like-domains 8 Encodes a single pass membrane protein which participates in developmental regulation and cellular communication NUP160 D83781 Nucleoporin 160 kDa Involved in poly (A)+ RNA transport.
Catalyzes methyl esterification of L-isoaspartyl and D-aspartyl residues in peptides and proteins repair and/or degradation of damaged proteins.

PLA2R1
U17034 Phospholipase A2 receptor 1 Receptor for secretory phospholipase A2. Activation of the mitogen-activated protein kinase (MAPK) cascade to induce cell proliferation, production of lipid mediators, and selective release of arachidonic acid in bone marrow-derived mast cells. Involved in responses in proinflammatory cytokine productions during endotoxic shock.

X74039
Plasminogen activator, urokinase receptor Role in localizing and promoting plasmin formation. Mediates the proteolysis-independent signal transduction activation effects of uPA.

U77718
Pinin, desmosome associated protein Transcriptional activator of the E-cadherin gene. Regulation of alternative pre-mRNA splicing. Regulates specific excision of introns in specific transcription subsets. Involved in the establishment and maintenance of epithelia cell-cell adhesion. Potential tumor suppressor for renal cell carcinoma.

D38552
Peptidylprolyl isomerase domain and WD repeat containing 1 Accelerates the folding of proteins. Involved in pre-mRNA splicing.

(Continued)
Frontiers in Pharmacology | www.frontiersin.org Stimulates the hydrolysis of glucosylceramide by beta-glucosylceramidase and galactosylceramide by beta-galactosylceramidase. Behaves as a myelinotrophic and neurotrophic factor.

RABAC1
AJ133534 Rab acceptor 1 General Rab protein regulator required for vesicle formation from the Golgi complex. May control vesicle docking and fusion.

RGS13
AF030107 Regulator of G-protein signaling 13 Inhibits signal transduction by increasing the GTPase activity of G protein alpha subunits.

SETD4
AB004848 SET domain containing 4 Role in protein encoding.

SMA4
X83300 Glucoridase Required for signal transduction and it also defines a conserved family of transforming growth factor beta pathway components.

TAF9
U21858 TATA box binding protein (TBP)-associated factor Gene regulation associated with apoptosis. Regulation of RNA polymerase II-mediated transcription.

TCERG1
AF017789 Transcription elongation regulator 1 Transcription factor that binds RNA polymerase II and inhibits the elongation of transcripts from target promoters in a TATA box-dependent manner.

M62810 transcription factor A Mitochondrial transcription regulation.
Maintenance of normal levels of mitochondrial DNA. Organizing and compacting mitochondrial DNA.

TOM1
AJ006973 Target of myb1 Involved in intracellular trafficking. Probable association with membranes.

D83004
Ubiquitin-conjugating enzyme E2N The UBE2V1-UBE2N and UBE2V2-UBE2N heterodimers catalyze the synthesis of non-canonical "Lys-63"-linked polyubiquitin chains. This type of polyubiquitination does not lead to protein degradation by the proteasome. Mediates transcriptional activation of target genes. Plays a role in the control of the cell cycle and differentiation. Plays a role in the error-free DNA repair pathway and contributes to the survival of cells after DNA damage. Induction and expression of NF-kappa-B and MAPK-responsive inflammatory genes.

ZNF195
AF003540 Zinc finger protein 195 May be involved in transcriptional regulation. that AA is a precursor for hydrogen peroxide (H 2 O 2 ) generation, which is considered to be preferentially cytotoxic to cancer cells (Parrow et al., 2013). Exposure to AA concentrations of up to 5 mM for 1 h resulted in decreased survival of cancer cells and cell death was dependent on H 2 O 2 production mediated by extracellular AA oxidation (Chen et al., 2005). Furthermore, intercellular metals contribute to the production of H 2 O 2 , with AA losing an electron to form a radical molecule. The free electron is donated to a transition metal. This reduced metal is then available to react with molecular oxygen resulting in the generation of H 2 O 2 . In the presence of AA, H 2 O 2 reacts with another transition metal ion such as ferrous ion to generate a hydroxyl radical (Kehrer, 2000). The tumor suppressor p53 may also play a role for this activity. P53-positive cell lines were more sensitive to both AA and H 2 O 2 treatment than p53-deficient ones (Kim et al., 2012). Thus, having two compounds from one plant with diverse mechanisms of action, we were interested to evaluate, whether their combination would reveal synergistic, antagonistic, or additive interactions. Understanding drug-drug interactions always represents a critical issue in the drug development process, since clinically relevant changes in exposure of co-administered drugs can lead to reduced efficacy or, conversely, adverse drug reactions, depending on the therapeutic window of the drugs. The latter becomes especially important with anti-cancer medications, since they are typically administered at or close to the maximally tolerated dose (Waters, 2015). In this study, we applied isobologram analyses to evaluate the nature of interaction of curcumin and AA. Isobologram analysis is considered as gold standard to provide evidence for drug interactions. The combination of curcumin and AA exhibited additive effects in leukemia and colon cancer cell lines and supra-additive effects in the glioblastoma cell lines. Our hypothesis is that both natural products may work together by different molecular pathways to achieve their overall cytotoxicity. Unlike chemically synthesized drugs, natural products might be active at lower doses and over longer periods of incubation, which could further support the appearance of additive effects.
We further analyzed molecular determinants of sensitivity and resistance of cancer tumor cell lines toward curcumin and AA. We correlated the IC 50 -values expressed on induction by curcumin and AA of 60 tumor cell lines by COMPARE analysis of microarray-based transcriptome-wide mRNA expression levels of these cell lines (Scherf et al., 2000). We identified genes from diverse functional groups, which were associated with response of the tumor cells toward curcumin and AA. Under curcumin treatment, these groups of genes included DNA repair, mRNA metabolism, signal transduction, angiogenesis, proliferation, apoptosis etc. While for AA treatment, the COMPARE analysis provided genes that are involved in signal transduction, transcription factors, and apoptosis. Although the exact function of these genes for cellular responsiveness to curcumin or AA treatment is still unknown, we have some clues of explanation. On treatment with AA the following genes were downregulated: SETD4, TAF9, PNN, CDK7, TFAM, and PPWD1. SETD4 is a methyltransferase, which is involved in carcinogenesis. Its down-regulation suppressed cellular proliferation and delayed the G1/S cell cycle transition without affecting apoptosis. Furthermore, its knockdown decreased cyclin D1 (Faria et al., 2013). The TATA-binding protein associated factor 9 (TAF9) interacts with oncogenic GLI family members to form GLI-TAF9 binding, which is important for carcinogenesis activity and malignant growth (Yoon et al., 2015). PNN is a nuclear and cell adhesion-related protein participating in the regulation of gene  Cluster 3 12 20 Cluster 4 0 6 Chi-square P = 0.050 The median log 10 IC 50 -values (-5.1 M for curcumin and -2.7 M for ascorbic acid) were used as cut-off to separate tumor cell lines as being "sensitive" or "resistant." expression and thereby, positively promoting cell-cell adhesion, and negatively affecting cell migration and cell proliferation (Shi et al., 2001). Cyclin-dependent kinase 7 (CDK7), which promotes transcription during the cell cycle, is critical for the survival of cancer cells. The inhibition of CDK7 suppressed proliferation and induced apoptotic cell death (Wang et al., 2015). Mitochondrial transcription factor A (TFAM), a member of the high mobility group (HMG) box protein family, is required for mitochondrial DNA replication and transcription. HMG proteins are often overexpressed in cancer cells and are involved in apoptosis regulation (Krynetskaia et al., 2009;Vander Heiden et al., 2009). TFAM may play a significant role in tumorigenesis (Guo et al., 2011). PPWD1 has a well-characterized peptide domain (WD40 domain for PPWD1), and this domain has also a critical role in carcinogenesis. The WD40 domain mediates signal transduction and transcriptional regulation during cell cycle and apoptosis. PPWD1 may serve as target for drug development (Davis et al., 2008;Jeon et al., 2014). From the above discussion, AA mechanism revolves around the genes affecting cell proliferation and cell cycle events; it can be assumed that these genes contribute to sensitivity of the tumor cells to AA. On the other hand, genes potentially responsible for responsiveness on curcumin treatment were identified and curcumin downregulated, i.e., AK2, PDK1, NR2F2, DFFB, MTHFD2, and ALDH3B1. Adenylate kinases (AKs) represent enzymes that catalyze reversible high-energy phosphoryl transfer reactions between adenine nucleotides in the intermembrane space. During periods of metabolic stress, AK2 increases the amount of available adenosine monophosphate and therefore activates downstream ATP-sensing mechanisms-such as AMPactivated protein kinase (AMPK)-to regulate the cellular metabolism. Inhibition of AK2 expression significantly inhibited the proliferation of cancer cells (Dzeja and Terzic, 2009). PDK1 plays a key role in several cancer types. Alterations of PDK1 are critical for oncogenic PI3K signaling. PDK1 has an essential role in regulating cell migration, especially in the context of PTEN deficiency. Downregulation of PDK1 levels inhibits migration and metastasis. PDK1 inhibitors may be useful to prevent cancer progression and abnormal tissue dissemination (Raimondi and Falasca, 2011). The nuclear receptor subfamily 2, group F, member 2 (NR2F2) is a master regulator of angiogenesis and acts as oncogene in prostate and other human cancers. NR2F2 is robustly expressed in the stroma of healthy ovary with little or no expression in epithelia lining the ovarian surface, clefts, or crypts. The pattern of NR2F2 expression was severely disrupted in ovarian cancers, in which decreased levels of stromal expression and ectopic epithelial expression were exhibited. Targeting NR2F2 expression in ovarian cancer cell lines enhanced apoptosis and increased proliferation (Hawkins et al., 2013). DFFB contributes to both chromosomal condensation and DNA degradation during apoptosis, decreased DFFB expression favors DNA damage, which in turn may contribute to both tumorigenesis and better response to DNA damaging chemotherapy (McDonald et al., 2005). MTHFD2 mRNA and protein expression is markedly elevated in many cancers and correlated with poor survival in breast cancer. MTHFD2 is integral to mitochondrial one-carbon metabolism, a metabolic system recently implicated in rapid cancer cell proliferation. Synthesis of one-carbon units carried by the tetrahydrofolate (THF) cofactor is important for proliferating cells, required for nucleotide synthesis and methylation reactions. MTHFD2 is a bifunctional enzyme, catalyzing the NAD + dependent CH2-THF dehydrogenase and CH + -THF cyclohydrolase reactions within the mitochondria. Within the mitochondrial folate pathway, MTHFD2 is of special interest, because MTHFD2 was one of the most consistently overexpressed mRNAs genome-wide across 19 different tumor types. The MTHFD2 protein is specifically expressed in transformed cells, but not the stroma surrounding the tumor tissues. MTHFD2 by RNAi impairs proliferation in a variety of cancer cell lines, independent of the tissue of origin, and decreases invasion and migration in breast cancer cell lines. MTHFD2 is broadly required for cancer cell proliferation and viability (Lehtinen et al., 2013;Nilsson et al., 2014). ALDH3B1 is a metabolically active enzyme with distinct specificity for various aldehyde substrates, particularly medium-, and long-chain aliphatic aldehydes. These substrates include many products that are formed during LPO, such as hexanal, 4-hydroxy-2-nonenal (4-HNE), octanal, and trans-2-nonenal. ALDH3B1 plays an important physiological role against cellular oxidative stress by detoxifying aldehydes derived from oxidative processes, such as ethanol metabolism and LPO (Marchitti et al., 2010). Our pharmacogenomics data shows curcumin suppressing cell proliferation by downregulation of anti-apoptotic genes and cell surface adhesion molecules. Curcumin is also seen to regulate cellular metabolism and inhibition of angiogenic cytokines. We can then suggest that the downregulated genes affect the sensitivity of the tumor cells to curcumin.
Previously, we reported the mRNA expression profile induced by curcumin in the NCI cell line panel, which was merged from four different microarray platforms (Novartis, Stanford, Chiron, and Genelogic;Sertel et al., 2012). In the present investigation, we focused only on the Novartis microarray platform for the comparison of curcumin and vitamin C and to reduce the degree of complexity. If we compared the top ranked genes in the previous analysis with those of the present investigation, we found some genes in common (MTHFD2, AK2, NFATC21P, BTAF1, RBBP4), although the majority of genes were different in both analyses. This result points to an observation that was frequently made by many investigators: different microarray platforms deliver different results. Nevertheless, several biological functional groups were found to be common in our previous and the present paper, e.g. cell cycle, DNA damage response, cell migration, inflammation, signaling pathways, and apoptosis-regulating genes. To our opinion, microarray data are reliable, if they are used for the generation of testable hypotheses. In this respect, both of our microarray analyses were useful. Microarray data represents the starting point for the elucidation of modes of action of cytotoxic compounds rather than completed end results.
Cluster analyses were applied in the present investigation under the assumption that responsiveness of cancer cells might be predicted by using gene expression patterns and that appropriate gene expression profiles might be sufficient to predict whether a cancer cell line is sensitive or resistant to a cytotoxic compound (Sertel et al., 2010). Curcumin revealed two clusters with predominantly sensitive and three with predominantly resistant cell lines. For AA cluster analysis revealed two clusters containing mainly resistant and two clusters containing mainly sensitive cell lines in a comparable fashion to curcumin. The prediction of sensitivity or resistance to cytotoxic drugs by mRNA expression profiles is interesting in the context of individualized or precision medicine, because it may open the possibility to determine prior to treatment, whether or not a tumor will respond to specific drugs. Our data demonstrate that this may not only be feasible to established anticancer drugs, but also to investigative natural products such as curcumin or AA.
The fact that the mRNA expression profiles induced by curcumin and AA related to different gene expression patterns may be related to different modes of actions of both compounds. Medicinal herbs generally contain mixtures of active compounds, which may interact in an additive or synergistic manner. Synergistic interactions may need common mechanisms e.g., a common specific pathway that they inhibit. From an evolutionary point of view, synergistic interactions need co-evolutionary selection pressures to evolve. Hence, it can be speculated that synergisms are less likely to occur than additive effects. Therefore, additive drug interactions can be more frequently found in medicinal herbs. Compounds with different modes of action can efficiently and sufficiently fulfill the requirements for plants to survive under specific evolutionary selection pressure. This may explain that we found additive rather than synergistic interactions in isobologram analyses between curcumin and AA in the panel of cell lines tested. This observation is in accordance with previous data with several cytotoxic compounds from Artemisia annua L., which also showed additive rather than synergistic interactions (Efferth et al., 2011).
In summary, we have identified some genes which were downregulated by AA and curcumin. The genes may be responsible for cellular responsiveness of the varied cancer cells to AA and curcumin treatments. The cellular activities tackled by the downregulated genes include inhibition of cell proliferation and cell cycle activities, reduction in cellular metabolism, downregulation of anti-apoptotic gene products and inhibition of angiogenic cytokines. The two natural products seem to induce cytotoxicity by different mechanisms and this may lead to achieve tumor eradication in vivo. The varied cellular functionalities represented by the genes downregulated may support additive effects observed in the isobologram analyses.

AUTHOR CONTRIBUTIONS
Title selection and research design: TE, EO. Laboratory experiments, results generation, data analysis, and interpretation: EO, OK. Manuscript writing and submission: EO. Manuscript writing: HG. Advising in research design, results interpretation, and manuscript writing: TE.