Chlorogenic acid effectively treats cancers through induction of cancer cell differentiation

Rationale: Inducing cancer differentiation is a promising approach to treat cancer. Here, we identified chlorogenic acid (CA), a potential differentiation inducer, for cancer therapy, and elucidated the molecular mechanisms underlying its differentiation-inducing effects on cancer cells. Methods: Cancer cell differentiation was investigated by measuring malignant behavior, including growth rate, invasion/migration, morphological change, maturation, and ATP production. Gene expression was analyzed by microarray analysis, qRT-PCR, and protein measurement, and molecular biology techniques were employed for mechanistic studies. LC/MS analysis was the method of choice for chemical detection. Finally, the anticancer effect of CA was evaluated both in vitro and in vivo. Results: Cancer cells treated with CA showed reduced proliferation rate, migration/invasion ability, and mitochondrial ATP production. Treating cancer cells with CA resulted in elevated SUMO1 expression through acting on its 3'UTR and stabilizing the mRNA. The increased SUMO1 caused c-Myc sumoylation, miR-17 family downregulation, and p21 upregulation leading to G0/G1 arrest and maturation phenotype. CA altered the expression of differentiation-related genes in cancer cells but not in normal cells. It inhibited hepatoma and lung cancer growth in tumor-bearing mice and prevented new tumor development in naïve mice. In glioma cells, CA increased expression of specific differentiation biomarkers Tuj1 and GFAP inducing differentiation and reducing sphere formation. The therapeutic efficacy of CA in glioma cells was comparable to that of temozolomide. CA was detectable both in the blood and brain when administered intraperitoneally in animals. Most importantly, CA was safe even at very high doses. Conclusion: CA might be a safe and effective differentiation-inducer for cancer therapy. “Educating” cancer cells to differentiate, rather than killing them, could be a novel therapeutic strategy for cancer.


Introduction
Cell differentiation is a programmed maturation of the immature stage which alters cell structure and function. The switch toward differentiation requires coordinated regulation of the expression of a group of genes at the post-transcriptional level involving epigenetic and/or miRNA-mediated mechanisms [1,2]. As the main characteristic features of cancer cells are sustained proliferation and vigorous migration, induction of cancer cell differentiation and correction of malignant behavior might constitute one of the Ivyspring International Publisher ideal approaches for cancer treatment. Differentiated cancer cells have low proliferation rate, reduced mobility, restricted ATP production and normalized cell morphology as compared to the undifferentiated ones [3][4][5]. However, studies of differentiation inducers, such as those of dimethyl sulfoxide (DMSO), arabinosylcytosine and actinomycin D, are stymied partially by their side effects [6]. Successful drugs in this field include arsenic trioxide (As2O3, ATO) and all-trans retinoic acid (ATRA). ATO induced cell differentiation in acute promyelocytic leukemia (APL) by, at least in part, attacking the PML moiety of the PML-RARα fusion products [7], whereas ATRA-stimulated APL cell differentiation through interacting with the nuclear receptor RAR [8,9]. Both drugs showed remarkable results in treating APL patients when used alone. In combination, the two could cure APL in many patients shedding new light on anticancer therapy [10][11][12]. Clinically, ATO and ATRA are effective mainly for specific types of leukemia and their side effects are related to heart, liver, and kidney [13]. Discovery of safe differentiating agents for solid cancers is highly desirable for the successful implementation of this approach in cancer treatment.
Although plants replace their foliage each year, placing an immensely heavy load on plant cell biosynthesis, it has been shown to be non-tumorigenic and plants very rarely develop cancer [14]. The rigid physical constraint by plant cell wall might be a contributing factor [14]; however, plants might possess innate mechanisms to control the meristematic tissue, such that they maintain the standard cycle of division-to-differentiation without erroneously progressing to malignancy. A challenging task in developmental biology is to learn the detailed regulatory mechanism of meristem maintenance and the transition to a differentiated cell. The process could include recruitment of cells within the meristem, cell division into primordium, post-primordial division, and differentiation into the mature plant organs, such as the leaf, flower, etc. [15]. Induction of differentiation in plants is associated with the plant's regulatory peptides and metabolites, which bring inter-meristematic communication and cell to cell interaction [16,17]. For example, the phytohormone auxin influx causes the root protophloem differentiation response [18,19]; plants also synthesize plenty of ascorbate as a cofactor with Fe 2+ /α-KG-dependent dioxygenases in the chloroplast of the meristem to control cell differentiation through epigenetic mechanisms [20]. Besides, reactive oxygen species (ROS) signaling cascades might be involved in plant cell differentiation [21]. Considering that plants have no immune system [22] with which cancer could be eliminated, an effective and well-established differentiation mechanism must be in place to correct malignant growth.
Several botanic compounds have been found to actively induce differentiation in mammalian cells, such as retinoic acid, matrine, shikonin, and cotylenin A, among others [8,[23][24][25]. Chlorogenic acid (CA) is a botanic compound (mw 354.31, Figure 1A) isolated from plants such as Eucommia almoides oliver or Lonicera confuse [26]. Chemically, it belongs to the phenylpropanoids with polyhydroxyphenol groups and has been reported to be non-toxic in animals and safe in humans [27]. The compound seems to be active against oxidation and inflammation [28,29], and has some regulatory activity on the immune system [30]. Recently, CA has been identified to be a potential drug for cancer and was approved by the China Food and Drug Administration (CFDA) first for phase I (NCT02728349, Apr. 2016) and then phase II (NCT03758014, Nov. 2018) clinical trials in glioma patients. Although the anticancer molecular mechanism of CA is far from clear, its unique effects against cancer provoked our interest to include CA in our investigation of potential botanic differentiation inducers. We found that CA treatment altered tumor cell behavior and programmed cancer cell differentiation both in vitro and in vivo. Human solid tumor cell lines from hepatoma, lung cancer, glioma, and colon cancer exhibited a similar response to CA. Here, we present data suggesting that CA is a novel and safe differentiation inducer for the treatment of solid tumors.

Chemicals
Chlorogenic acid (CA, Figure 1A) with a purity of more than 99% was acquired by the Jiuzhang Biochemical Engineering Science and Technology Development Co., Ltd. (Chengdu, Sichuan, China). For the in vitro experiments, CA was dissolved in dimethyl sulfoxide (DMSO, Sigma Aldrich, St. Louis, USA) at a concentration of 100 mM as a stock solution, followed by dilution in medium (or saline) prior to the experiments. DMSO was used as a vehicle reference in the study (as an untreated control group).
For the regulation of expression, cells were seeded the day before transfection into 12-well plates with antibiotic-free growth medium at 1 × 10 6 cells/well and cultured overnight until they reached 60-70% confluence. Transfection of the negative control (50 nM), mimics (50 nM), inhibitors (100 nM) and/or siRNAs (50 nM) was carried out using riboFECT™ CP transfection reagent (Ribobio, Guangzhou, China) for 24 h according to the manufacturer's protocol. The siRNAs used in this study were synthesized by Ribobio (Guangzhou, China).

Cell staining
Cells were rinsed three times in PBS and fixed with cold methanol, washed thoroughly with PBS, incubated with DAPI staining solution (1 µg/mL, Thermo Fisher Scientific, Waltham, MA, USA) for 30 minutes (min) and rinsed three times with PBS, followed by viewing using a fluorescence microscope (Olympus, Tokyo, Japan).

Cell viability and growth assay
Cell growth was examined using the 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT, Thermo Fisher Scientific, Waltham, MA, USA) assay for which 10 4 cells were seeded into each well of a 96-well plate. For growth analysis, the cells were treated with blank or 25 or 50 µM of CA every 24 h for 6 days. At different time points, the cells were stained with 50 μL of MTT reagent at 37 °C for 4 h. Spectrometric absorbance at 550 nm was detected using a microplate reader (BioTek, Vermont, USA). The doubling-time (Td) of the tumor cells was calculated using the following formula: where Nt is the cell number after treatment for 4 days.

Migration and invasion assay
Cells were treated with CA (25 μΜ or 50 μΜ) for 24 h, trypsinized and resuspended in serum-free medium. Next, 100 μL of the cell suspension (1 × 10 5 cells) was seeded into the upper chamber of the Transwell (Corning Costar, MA, USA) with a porous polycarbonate membrane (pore diameter of 8 μm). The lower chamber contained 700 μL of medium supplemented with 10% FBS. After 24 h incubation, the cells on the upper surface of the membrane filters were mechanically removed with cotton swabs. Migrated cells on the lower side of the filter were fixed with cold methanol, stained with 0.5% crystal violet solution for 120 min, washed with PBS, counted, and captured under a microscope. For the cell invasion assay, the upper chamber of the Transwell was coated with 60 µL of Matrigel (1:20 Corning Costar, MA, USA), and cells that invaded the lower surface of the membrane filter were analyzed as described above.

Vector construct
Oligonucleotide fragments, including the target region or mutant target region of the miRNA seed sequence AAGUGC, were synthesized (Invitrogen, CA, USA), annealed and ligated into the pmiRGlO dual luciferase miRNA target expression vector (Promega, WI, USA), and labeled pmiRGlo p21 position 1 (P1), position 2 (P2), pmiRGlo p21 mutant position 1 (mP1), or mutant position 2 (mP2). The EcoRV internal restriction site was used for clone confirmation. The bands were separated on a 1% agarose gel, stained with SYBR Green Safe and visualized under the UV trans-illuminator (Bio-Rad, Hercules, CA, USA).

Dual luciferase reporter assay
HEK293T cells, commonly used for the dual luciferase reporter assay were transfected with miR-20a, -93, -106b mimics (50 nM), or vector constructs (200 ng) using FuGENE @ HD transfection reagent (Promega) for 24 h. The cells were then added to 100 μL of Passive Lysis Buffer (PLB) and gently mixed by shaking for 15 min. The lysate was immediately measured using the Dual-Luciferase @ Report Assay System (Promega). Data were presented as a ratio of Firefly to Renilla luciferase activity (Fluc/Rluc). All assays were performed three times in triplicate.

Quantitative real-time PCR (qRT-PCR)
Total RNA was extracted from cells or tissues using the SV Total RNA Isolation System (Promega Biotech, Madison, USA). Subsequently, 2 μg of RNA was used for reverse transcription and the quantitative PCR reaction according to the manufacturer's instructions (Invitrogen). Mature miR-20a, -93 and -106b levels were measured using the Bulge-Loop miRNA qRT-PCR kit (Ribobio, Guangzhou, China) with U6 snRNA serving as an internal control. The resulting cDNA products were analyzed with the quantitative real-time PCR 7500 fast system (Applied Biosystems, Foster, USA). Total RNA (100 ng) and one-step power SYBR reagent (Invitrogen) were used for the detection of mRNAs. The mRNA expression of the housekeeping gene GAPDH was used for normalization. The primer sequences used in this study are listed in Table S1. Data analysis was performed using the 2 -ΔΔCt method.

Western blotting
Total proteins were extracted from cells or tissues using the M-PER mammalian protein extraction reagent (Thermo Pierce, Rockford, USA) and measured using the BCA Protein Assay (Thermo Pierce). Protein samples (50 μg per assay) were separated on SDS-PAGE gels and transferred to PVDF membranes. The membranes were blocked with 5% skim milk and then incubated overnight at 4 °C in the presence of primary antibodies. The primary antibodies for p21, p53, KHSRP, c-Myc, EPCAM, and β-actin were purchased from Cell Signaling Technologies (CST, Danvers, MA, USA), p-c-Myc antibody was from Santa Cruz (CA, USA), and IκBα, Rb1, STAT3, histone-H3, SUMO1, and SUMO2/3 antibodies were from the Proteintech Company (Chicago, IL, USA). The membranes were washed 3 times in Tris-buffered saline with Tween-20 and incubated with horseradish peroxidase-conjugated secondary antibody (CST, MA, USA) for 1 h at room temperature (RT). After washing 3 times with 15 mL of TBST, Immobilon Western Chemiluminescent HRP Substrate (Millipore, Billerica, USA) was added, and the bands were imaged with the Chemidoc XRS + electrophoretic imaging system (Bio-Rad, Hercules, CA, USA). Density scanning of each protein band was performed using Image Lab software (Bio-Rad, Hercules, CA, USA).

Flow cytometry analysis
Cells were harvested 24 h after treatment with DMSO (solvent control) or CA (25 μM or 50 μM) and rinsed twice with PBS. The cells were fixed with cold ethanol, washed with PBS twice, and centrifuged at 300 × g for 5 min. Next, 500 μL of propidium iodide (50 μg/mL) was added to each sample tube, followed by incubation for 30 min at 37 °C in the dark. The cell cycle distribution was analyzed using a C6 Flow Cytometer (BD FACSCalibur, Ann Arbor, MI). For CD44 expression, 1 μL of PE-conjugated CD44 antibody (1:100, Miltenyi Biotec, Cologne, Germany) was added to the cells, followed by a 30 min incubation, washing, and re-suspension in 500 μL of PBS for analysis.

Sumoylation assay
Nuclear proteins were extracted from sample cells or tissues using NE-PER nuclear and cytoplasmic protein extraction reagent (Thermo Pierce, USA) and measured with the BCA Protein Assay (Thermo Pierce). The sumoylation assay was performed using the EpiQuikTM Protein Sumoylation Assay Ultra kit (Epigentek, #P-8003, Farmingdale, USA). The standard procedure was as follows: primary antibody (IP-grade) or control was added to coat the microplate, which was evaporated at 37 °C for 120 min and the unoccupied sites were blocked with blocking buffer for 45 min at RT. SUMO assay buffer (SAB) was added to each well. Next, sample nuclear proteins (10 μg) were added to each sample well, the diluted detection antibody against SUMO proteins was added to each well for 60 min at RT and then color developer was added and incubated for 5 min. After the addition of stop solution, the absorbance was measured using a microplate reader (BioTek, USA).

Transcriptome and miRNA microarray expression profiling
Total RNA was extracted from CA-treated Huh7 cells, and the samples were used for whole human genome oligo microarray and miRNA microarray profiling. For the transcriptome microarray, RNA sample labeling and microarray hybridization were performed according to the Agilent One-Color Microarray-Based Gene Expression Analysis protocol (Agilent Technology). Briefly, total RNA was linearly amplified, labeled with Cy3-UTP, and then hybridized onto the microarray chips (4 × 44 K, Agilent, Santa Clara, CA, USA). After washing and staining the slides, the arrays were scanned using the Agilent Microarray Scanner (part number G2505C). The scanned images were analyzed using Agilent Feature Extraction Software (version 11.0.1.1).
The miRNA expression was measured by the Agilent microarray according to the standard Agilent microarray scanning process. The miRNA labeling was performed using the miRCURY™ Hy3™/Hy5™ Power labeling kit (Exiqon, Vedbaek, Denmark). The Hy3™-labeled samples were hybridized onto the miRCURYTM LNA Array (Exiqon, v.18.0), and the slides were scanned using the Axon GenePix 4000B microarray scanner (Axon Instruments, Foster City, CA). The scanned images were analyzed using GenePix Pro 6.0 software (Axon).

ATP and OCR measurement
Cellular ATP content was determined using the EnzyLight TM ATP assay kit (EATP100, Hayward, CA, USA) and operated according to the manufacturer's instructions. Briefly, 10 4 cells (per well) were seeded into the white opaque 96-well plate (Corning Costar, MA, USA) followed by transfection with negative control or SUMO1 siRNA in the presence or absence of CA (50 μM). Following 24 h incubation, the luminescence was measured after adding the reconstituted reagents of the kit, and the ATP concentration was calculated based on a calibration plot made with standard ATP solution (ranging from 3 μM to 30 μM).
Cellular oxygen consumption rate (OCR) test was performed using an XF24 Seahorse Analyzer (Seahorse Bioscience, North Billerica, MA, USA). Briefly, 2×10 4 cells were seeded into each well of the XF24 cell culture microplate with freshly prepared growth medium for 16 h and then treated with CA (0, 25, 50 μM) for another 24 h. The OCR value was detected with the XF Cell Mito Stress Test Kit (Seahorse Bioscience, North Billerica, MA, USA) according to the operation manual instructions. The mitochondrial inhibitors used were 1 μM oligomycin A, 1 μM carbonyl cyanide p-trifluoro-methoxyphenyl hydrazine (FCCP), and 0.5 μM Rotenone plus 0.5 μM antimiycin A (Ron/AA). All OCR values were normalized to the protein content (μg/μL) of each well, which were detected with the BCA protein assay kit.

Xenograft tumor model
Male NOD/SCID mice, weighing 18-22 g, were purchased from Vital River Laboratories (Beijing, China) and housed with free access to food and water. All experimental animal procedures were approved by the Animal Experimental Ethics Committee of the Institute of Materia Medica, Chinese Academy of Medical Sciences and Peking Union Medical College. To establish the mouse xenograft model, Huh7 or H446 cells (2 × 10 7 cells/mouse) in Matrigelcontaining culture medium were subcutaneously inoculated into the mice. The tumor-bearing mice were treated with CA when the xenograft tumor mass reached 100 mm 3 . The animals (7 per group) received an intraperitoneal (ip) injection of CA (directly dissolved in saline; 25, or 50, or 200 mg/kg/d) for 30 days, and the same volume of saline was used as a control. To terminate the experiment, the body weights of the mice were determined. The tumor volume (TV) was calculated with electronic calipers using the following formula: TV = (a × b 2 )/2 (where "a" is the largest superficial diameter and "b" is the smallest superficial diameter). The tumor inhibition rate (TIR, %) was calculated using the following formula: TIR = (the mean tumor weight (volume) of the control group − the mean tumor weight (volume) of the treated group) / the mean tumor weight (volume) of the control group × 100%. After sacrificing the mice, the tumor weights were determined, and the tumor tissue samples were collected for qRT-PCR and Western blotting. The tumor re-implantation test was carried out in which 25 mg of tumor tissues from each group was pooled, enzyme digested, homogenized, and diluted at 1:4 and 1:8 ratios in culture medium followed by reimplantation of the tumor cell suspension into naïve NOD/SCID mice (2 × 10 5 or 1 × 10 5 cells per mouse for the 1:4 or 1:8 dilutions, respectively). Tumor growth in the mice was monitored for 4 weeks.

Safety assessment of CA in vivo
KM mice were purchased from the Vital River Laboratories (Beijing, China) and randomly divided into 4 groups. Animals (n = 10 per group, ♀ × 5, ♂ × 5) were treated intraperitoneally with a single dose of 1000, or 500, or 250 mg/kg CA, and saline was used as a control. Mouse survival and body weight changes were examined daily for all groups for 7 days. Next, 200 µL blood samples were collected from each mouse via retro-orbital bleeding, and levels of ALT, AST, BUN, and CRE in serum were detected.

CA concentration in animal blood and brain
Wistar rats (male, 250 ± 20 g) were acquired from Vital River Laboratories (Beijing, China). CA in saline was administered to the rats (ip, 75 mg/kg) and three rats per time point were exsanguinated at 1, 2, 4, 7, 10, 20, 30, 45, 60, and 120 min post-injection. Blood was collected into a heparinized test tube and centrifuged at 4500 rpm for 10 min. Plasma was stored at −80 °C and thawed at room temperature prior to use. For analysis, 20 μL of CA (2.5-5000 ng/mL an equivalent gradient concentrations), 20 μL of puerarin (25 ng/mL, internal standard, Baoji Herbest Bio-Tech, Shanxi, China) and 20 μL of 2 M HCl was added to 100 μL of plasma followed by 700 μL of acetonitrile. The mixture was vortex-mixed for 5 min, centrifuged at 4500 rpm for 10 min, and the supernatant acetonitrile extraction solution was evaporated to dryness with a gentle stream of nitrogen gas at 30 °C. The residue was reconstituted in 50 μL of the initial mobile phase by vortexing for 3 min and centrifuged at 13000 rpm for 10 min. Finally, 5 μL of supernatant was injected into the UPLC-MS/MS system (Agilent Corporation, MA, USA). Values were presented as the mean ± SD of 3 rats per time point. To determine the pharmacokinetic parameters of chlorogenic acid, the concentration-time data were analyzed using the DAS (Drug and Statistics for Windows) software package (version 2.0, Chinese Pharmacological Association, China).
To measure CA in rat brain, CA in saline was ip administered to the Wistar rats, with 3 rats for each group. At 1, 2, 4, 7, 10, 20, 30, 45, 60, 120, 180, 300, and 420 min post-CA injection, the rats were administered with pentobarbital sodium (ip, 60 mg/kg) to induce deep anesthesia, followed by surgery and injection of saline into the ventricle to remove blood from the blood vessels. Rat brain was then homogenized, and the CA concentration in the brain was determined using a similar chemical detection method in the UPLC-MS/MS system as mentioned above.

Treating glioma with CA in animals
For the C6 glioma in rats, CA (purity > 99%) was from the Jiuzhang Biochemical Engineering Science and Technology Development Co., Ltd. (Chengdu, Sichuan, China). Temozolomide (TMZ, purity over 98%) was purchased from Sigma-Aldrich (St. Louis MO, USA). The rat glioma C6 cells were obtained from the Cell Center, Peking Union Medical College, and cultured in DMEM with 10% FBS, penicillin (100 units/mL) and streptomycin (100 μg/mL). Male Wistar rats (260-280 g) were from Vital River Laboratories (Beijing, China).
Before the surgical procedures, male Wistar rats were anesthetized by an intraperitoneal injection of pentobarbital sodium (60 mg/kg). The rat head was shaved and disinfected with 70% ethyl alcohol, and then the animal was fixed in a stereotactic apparatus. A midline incision was generated on the dorsal aspect of the rat head to expose the bregma. A burr hole was then generated in the cranial bone 0.2 mm anterior to the bregma and 3 mm lateral to the sagittal suture. A 25 μL Hamilton microsyringe was connected to the manipulating arm of the stereotactic device. The tip of the microsyringe was inserted 7 mm beneath the dura and then withdrawn 1 mm. Next, 10 μL of C6 glioma cells (4 × 10 5 ) were injected stereotactically into the striatum. The injection was conducted over 10 min, after which the syringe was maintained in the brain for another 10 min and then slowly retracted. The wounds were cleaned, and the animals were then returned to their cages and received the standard rat diet and water ad libitum.
After 11 days of inoculation, 18 rats developed tumors based on magnetic resonance imaging (MRI) scanning using the Bruker PharmScan system (Bruker Biospin, Karlsruhe, Germany). The tumor-bearing rats were then randomly assigned to three treatment groups (n = 6), including the model group, TMZ group and CA group. Another six rats were designated as the sham-operated group (Sham group), which received an injection of 10 μL of cell culture medium without glioma cells using the same surgical procedure. The rats in the TMZ group were orally treated with a solution of TMZ (20 mg/kg/d) for five consecutive days, and those in the CA group were intraperitoneally administered CA (75 mg/kg, bid) for seven consecutive days. The body weights of the rat were measured every day. The tumors were scanned by MRI on days 11 and 18 after tumor cell injection. The experiment was then terminated after the second MRI scan.

Statistical analysis
Heterogeneity of variance among groups was first examined by F-tests, followed by difference analyses performed with the unpaired 2-tail Student's t-test using GraphPad Prism 5.0 software (San Diego, CA, USA), p < 0.05 was considered significant.

CA changed tumor cell behavior by inducing differentiation in vitro.
The effect of CA was investigated in a variety of solid tumor lines, such as human hepatoma (Huh7) and lung cancer (H446) cells, for its anticancer effects. Compared with the untreated groups, CA slowed down the proliferation rate of tumor cells. In the Huh7 cells, the doubling time (Td) on day 4 of treatment was extended from 1.15 to 1.28 days at 25 µM CA concentration and up to 1.60 days at 50 µM. For the H446 cells, Td was extended from 1.04 to 1.10 (CA 25 µM) and 1.12 days (CA 50 µM) ( Figure 1B). The cell mobility assay showed that the number of invading ( Figure 1C) and migrating cells ( Figure 1D) was significantly lower in the Huh7 and H446 cells treated with CA (25 or 50 µM for 24 h) as compared with the untreated controls. CA appeared to reduce the tumor cell growth rate and inhibited cell mobility through a non-cytotoxic mechanism. Figure 1E shows the morphological changes in the Huh7 and H446 cells after exposure to CA. With respect to the untreated cells, the cell nuclei underwent karyopyknosis and exhibited a kidney shape after treatment with 25 or 50 μM of CA for 48 h. This phenotypic change suggested a transition toward cell maturation. Cancer cell differentiation was also supported by the cell cycle arrest at the G0/G1 phase after treating the Huh7 or H446 cells with CA (25 or 50 μM for 24 h). Compared with the solvent controls, the percent of cells in the G 0 /G 1 phase increased, and S-phase cells decreased in the two cell lines indicating a slowed-down biosynthesis ( Figure 1F) and cell maturation [32]. The G 2 /M phase remained stable. Human glioma cells showed a similar response to CA (see the glioma section below).

CA regulated expression of the differentiation-related genes.
To understand the potential molecular mechanism(s) underlying the phenotypic change by CA, we investigated the differential mRNA expression. Huh7 and non-cancer liver MIHA cells were treated with 50 μM of CA for 24 h, followed by transcriptome microarray profiling. As shown in Figure 2A, CA increased expression of key genes in Huh7 cells associated with differentiation, for instance, KHSRP, p53, and p21. Simultaneously, CA treatment decreased the expression of genes associated with poor differentiation, such as c-Myc and CD44, among others. These differentiation-related genes were identified based on the gene function description in the Gene Bank database (https://www.ncbi.nlm.nih.gov/gene/) followed by literature review for confirmation. To validate the results, expression of the corresponding mRNAs and proteins of KHSRP, p53, p21, c-Myc, and CD44 was examined in Huh7 and H446 cells using qRT-PCR, Western blotting, or FCM analysis. The mRNA results were consistent with those from the transcriptome microarray ( Figure 2B). Expression of p21, p53, and KHSRP proteins was significantly increased by CA, while phosphorylated c-Myc and total c-Myc protein expression dramatically decreased after CA treatment (25 or 50 μM) in both Huh7 and H446 cells ( Figure 2C; right side: density scanning results). EPCAM is also a gene closely related to poor differentiation [33], and its expression was found to be decreased by CA in both cancer cell lines ( Figures  2B-C), although its expression reduction in microarray was not substantial. CD44-positive cells declined by CA (50 μM) in Huh7 (ΔMFV = -36.8%) and H446 cells (ΔMFV = -33.6%) in the FCM analysis ( Figure 2D). It appeared that CA caused a differentiation shift in cancer by regulating genes associated with cell differentiation.
We also examined seven other human solid tumor cell lines: U87MG (glioma), M059J (glioma), HCT-116 (colon carcinoma), NCI-H358 (non-small cell lung cancer), A549-5FU (lung cancer cells resistant to 5-FU), SK-LU-1 (lung adenocarcinoma), and Bel-7402 (hepatoma). Similar to the Huh7 and H446 cells, CA dose-dependently decreased p-c-Myc expression and increased p21 expression in all seven cell lines confirming the differentiation-inducing effect of CA in solid tumors ( Figure 2E, upper; density scanning results, Figure S1A). Furthermore, four non-cancer human cell lines were tested for comparison, 2 liver cell lines (CCC-HEL-1 and MIHA) and two lung cell lines (WI-38 and MRC-5). As shown in Figure 2E (lower; for density scanning, see Figure S1B), in contrast to cancer cells, CA treatment showed no effect on p21 expression in these lines but decreased the phosphorylation of c-Myc to p-c-Myc in all four cell lines. In the microarray analysis with MIHA, CA regulated the expression of a set of differentiation-related genes (Figure 2A) different from those seen in the CA-treated Huh7 cells. The reason for the different response to CA between cancer and non-cancer cells remains unclear. Considering that the four non-cancer cell lines were isolated, immortalized in vitro, and cultivated in laboratories for years, their gene expression profiles might have shifted from their original normal patterns. Thus, primary human hepatocytes HH were examined. As shown in Figure 2E (lower) and Figure  S1B, in contrast to the Huh7 cells following exposure to CA, HH cells showed almost no change in the expression in p-c-Myc, total c-Myc and p21. Thus, the cell differentiation response to CA appears to occur in the order of cancer cells > non-cancer immortalized cells > normal primary cells.
Since meristematic tissues are considered stem cells of plants, we induced human pluripotent stem (iPS) cells to examine the CA differentiation-inducing effect. We found that the mRNA expression of well-differentiated genes (like p53, p21) was significantly increased whereas that of poorly differentiated genes (like c-Myc, EPCAM, and CD44) declined in response to CA treatment (25 or 50 μM, 24 h) ( Figure S2A). The results were confirmed by Western blotting ( Figure S2B).

CA elevated p21 expression by down-regulating the expression of the miR-17 family.
As gene expression is largely controlled by the miRNA machinery, we next investigated whether CA regulated differentiation-related genes through miRNAs. The miRNA microarray profiling showed that the expression of the miR-17 family member's miR-20a, -93, and -106b was down-regulated in a dose-dependent manner after exposure of the cells to CA (25 or 50 μM for 24 h) ( Figure 3A). The results were then confirmed in both Huh7 and H446 cells by the qRT-PCR assay ( Figure 3B). The expression of the 3 miR-17 members was also decreased by CA (25 or 50 μM for 24 h) in iPS cells ( Figure S2C). Subsequently, we analyzed three miRNA-related databases, the MiRNA.org (www.microrna.org), TargetScan Human (www.targetscan.org) and PicTar (pictar.mdcberlin.de), to identify the potential target genes of miR-20a, -93, and -106b. We found that the p21 3' UTR (untranslated region) sequence had two sites (with a sequence of 5'-GCACUUU-3'), position 468-474 and position 1148-1154, which matched the seed sequence (5'-AAAGUGC-3') of the miR-17 family ( Figure 3C), suggesting that p21 is a possible target gene of the miR-17 family members. We next investigated whether p21 expression was influenced after diminution of the three miR-17 family members. As shown in Figure 3D, transfection of Huh7 or H446 cells with the miR-20a, -93, or -106b specific inhibitor (single-chain nucleotide complementary to the three miRNAs mature sequence) increased p21 mRNA expression. Consistent with this observation, elevated p21 protein was detected in the Western blotting assay ( Figure 3E). The dual luciferase reporter assay demonstrated that overexpression of miR-20a, miR-93, or miR-106b mimics (50 nM) reduced the luciferase activity of the vectors (200 ng) with the p21 3' UTR binding sites. The ratio of Firefly to Renilla luciferase activity (Fluc/Rluc) ranged from 32.6%-48.3% of that of the empty vector control. In contrast, overexpression of the miR-20a, miR-93, or miR-106b mimics (50 nM) in cells with vectors carrying mutated p21 3' UTR binding sites (200 ng) showed minor effects on luciferase activity, and the Fluc/Rluc value ranged from 70.6%-89.6% of the control ( Figure 3F, scheme). The results indicated that the two binding sites in the p21 3' UTR sequence were the responsive elements for the three miR-17 family members.
Also, transfection of the CA-treated Huh7 or H446 cells with mimics of miR-20a, miR-93, miR-106b or p21-siRNAs (si-p21) eliminated the p21 up-regulatory effect of CA and increased c-Myc and CD44 expression ( Figure S3). As p21 has a positive role in cell differentiation, in contrast to that of c-Myc and CD44, it appeared that through down-regulation of miR-20a, -93, and -106b expression, CA stabilized the 3' UTR of p21 mRNA, increased the intracellular level of p21 protein, and triggered the differentiation cascade in tumor cells.

CA-caused c-Myc SUMOylation resulted in miR-17 family decline and p21 elevation
c-Myc is known to be an important regulator for miRNAs [34] Figure 4A, right), showing the negative relationship between c-Myc and p21. The results were confirmed at the protein level by Western blotting in which CA showed an advantage in elevating p21 protein expression superior to that of 10058-F4 ( Figure 4B). Moreover, down-regulated expression of miR-20a, miR-93, and miR-106b ( Figure  4C) was also observed in both tumor cell lines treated with CA or with the c-Myc inhibitors. This finding suggested that the down-regulation of c-Myc by CA might up-regulate p21 expression through abrogating the expression of miR-17 family members.
We also detected an increase in the small ubiquitin-like modifier 1 (SUMO1) protein expression but not SUMO2 and SUMO3 proteins in the Huh7 and H446 cells cytoplasm following CA treatment ( Figure  4D, lower). It significantly promoted the SUMO1 in the cytoplasm and simultaneously decreased the levels of p-c-Myc ( Figure 4D, upper) in the cell nucleus. CA (50 µM) increased the stability of the SUMO1 mRNA transcripts in Huh7 and H446 cells with an increase in T1/2 by 3.75-or 1.9-fold, respectively ( Figure  4E). The CA-caused SUMOylation of c-Myc was dose-dependent and selective as it was negative for other proteins such as IκBα, Rb1 and STAT3 ( Figure 4F). It appeared that CA stabilized the SUMO1 mRNA, increased the intracellular SUMO1 protein, selectively promoted sumoylation of c-Myc instead of phosphorylation, and thus decreased the expression of the miR-17 family members [34], leading to an elevation of p21 protein in tumor cells.
We then assessed which regulatory sequences in the SUMO1 mRNA 3' UTR region mediated the effect of CA on SUMO1 mRNA stability. The SUMO1 mRNA 3' UTR sequence contains 4 AU-rich elements (ARE1-4) with a total length of 1073 bp.  Data were normalized to EV as 100%. EV stands for empty vector of pIS0. All data are presented as mean ± SEM of 3 independent experiments. * p < 0.05, ** p < 0.01 and *** p < 0.001 (treated vs untreated control).

CA induced differentiation in mouse xenograft tumors.
To investigate the differentiation-based therapeutic effect of CA in vivo, we used the Huh7 hepatoma and H446 lung cancer xenograft models in the NOD/SCID mice. CA treatment with ip administration of 25 mg/kg/d, 50 mg/kg/d, or 200 mg/kg/d slowed down the growth of the tumor mass, yielding 44.0%, 80.1% or 83.1% tumor volume inhibition of Huh7 tumors and 39.9%, 84.6%, or 86.3% tumor reduction of H446 tumors, respectively ( Figure  5A). Accordingly, tumor weight was declined by CA as well ( Figure 5B). Increasing the CA dosage up to 200 mg/kg/d (ip) did not cause further decline of the tumor mass but accelerated the entry of the xenograft tumor into the non-growth plateau ( Figure 5A). No animal toxicity was observed. As both the Huh7 and H446 tumors maintained non-growth after CA treatment (50 or 200 mg/kg/d) for 12 or 18 days ( Figure 5A), we assumed that, similar to the iPS cells, most of the remaining tumor cells, including tumor stem cells, might have been differentiated by CA. By analyzing differentiation-related genes, we found that c-Myc, EPCAM, and CD44 mRNA levels decreased in both Huh7 and H446 ( Figure 5C) xenograft tumors in the mice treated with CA (ip, 25 mg/kg/d), while the p21 mRNA level increased ( Figure 5C) compared with the saline-treated control. Western blotting showed that CA treatment reduced the level of p-c-Myc protein but elevated the p21 protein level in the tumor tissue ( Figure 5D). Furthermore, the expression of miR-20a, miR-93, and miR-106b was down-regulated both in Huh7 and H446 ( Figure 5E) xenografts in mice treated with CA (ip, 25 mg/kg/d). Sumoylation of c-Myc was also increased after the CA treatment ( Figure 5F). The results obtained for CA-induced solid tumor differentiation in vivo were consistent with those observed in vitro.
Next, we digested and homogenized 25 mg of the tumor tissue pooled from the saline control mice (n = 5) or from the CA-treated (50 mg/kg/d) mice (n = 5), followed by reimplantation of the tumor cell suspension (at 1:4 and 1:8 dilutions) into naïve SCID mice (n = 4 for each group). As displayed in Figure 5A (inserts), on day 28 after re-implantation, no tumors developed in the mice inoculated with Huh7 tumor cells collected from the CA-treated mice (both 1:8 and 1:4 dilutions). On the other hand, there was tumor development in mice implanted with saline-treated controls with all mice tumor-positive when implanted with cells at a 1:4 dilution and half of the mice growing tumors at a 1:8 dilution. For the H446 lung cancer xenografts, while all mice inoculated with the tumor suspension from saline-treated H446 tumor-bearing mice developed tumors (for both 1:4 and 1:8 dilutions), the mice implanted with a 1:8 dilution originating from xenografts of mice treated with CA did not develop tumors; but half of them developed tumors at a 1:4 dilution. These results suggested that CA might decrease tumorigenicity through differentiating cancer cells in the tumor mass.
The safety test showed that all mice survived for 7 days after single dosing of CA ranging from 250 to 1000 mg/kg (ip). Compared with the saline control group, CA ip injection caused no change in body weight ( Figure 5G) and no damage to the liver and kidney function, with serum AST, ALT, BUN and CRE levels within the normal range ( Figure 5H). These observations were consistent with excellent biocompatibility and safety of CA in humans. When the blood concentration of CA (ip, 75 mg/kg) was examined in the SD rats (3 rats per time point), the average Cmax of CA in the ip injection was 52.9 µg/mL (147 µM), which was higher than the drug concentration used in the in vitro experiments ( Figure  5I).

CA was effective for glioma cells in vitro and in animal models.
Besides liver and lung cancer, we also investigated the efficacy of CA in glioma cells. As is evident from Figure 2E, human glioma cells were sensitive to CA, showing reduced p-c-Myc and elevated p21 expression similar to the results observed in hepatoma, lung cancer as well as colon cancer. Further investigation of CA in gliomas was designed to identify specific biomarkers or indication of differentiation in brain tumors. Toward this end, the cellular expression of Tuj1 and glial fibrillary acidic protein (GFAP) in glioma cells was assessed as signature markers for neurons and astrocytes, respectively [35,36]. Immunofluorescent staining showed increased expression of Tuj1 and GFAP both in the U87MG and M059J glioma cells after CA treatment (25 or 50 μM, 24 h) indicating the shift toward neurological characteristics ( Figure 6A). Since cancer cells exhibit increased glucose utilization and energy metabolism [5,37], mitochondrial function was examined in glioma cells with or without CA treatment. As shown in Figure 6B, CA treatment decreased energy metabolism in mitochondria indicating reduced ATP production in the U87MG and M059J glioma cells by about 21%. Also, silencing SUMO1 expression by transfection of SUMO1-specific siRNA abolished the effect of CA on ATP reduction suggesting a SUMO1-dependent action of CA on mitochondria ( Figure 6B). Thus, it appears that SUMO1 is critical for the mechanism of action of CA. Oxygen consumption rate (OCR, both basal or maximal) in the two glioma cell lines was reduced after treating the cells with CA (25 μM or 50 μM, 24 h; Figure 6C), indicating inhibition of oxidative phosphorylation in the mitochondrial respiratory chain by CA. Accordingly, the cell mobility assay showed that the number of invading ( Figure 6D, upper) and migrating glioma cells ( Figure 6D, lower) was significantly reduced in the M059J or U87MJ cells treated with CA (25 or 50 µM for 24 h) as compared to the untreated ones possibly due to the decline in the ATP level. Finally, in the neurosphere culture, we observed inhibition of the sphere formation by CA both in U87MG and M059J cells ( Figure 6E) indicative of decreased malignancy in the CA-treated glioma cells [38,39]. These changes of differentiation markers and cell behavior in glioma strongly suggest maturation toward differentiated cancer cells.
The pharmacokinetic study in rats revealed that CA was able to cross the blood-brain barrier and enter the brain tissue 10-20 min after ip injection (75 mg/kg, Figure 6F). The maximum brain concentration of CA in the rats was 50.7 µg/kg (n = 3). CA administration (75 mg/kg, ip, bid) caused significant inhibition of tumor growth of the C6 glioma in rats (** p < 0.01, n = 6 for each group, Figure 6G, right) with an inhibition rate of 57% of the tumor volume by MRI. The CA anticancer efficacy was comparable to that of temozolomide (TMZ; # p < 0.05, CA group vs TMZ group), the first-line drug for glioma in the clinic ( Figure 6G, ref. 40). Continuous injection of CA showed no toxicity in the rats as evaluated by the rat body weight measurements ( Figure 6G, left). The sham group (n = 6) showed no tumors in the brain based on MRI detection (data not shown).

Discussion
The present study showed, for the first time, that CA induced differentiation in solid tumors and inhibited cancer cell proliferation, mobility, invasion, ATP production in mitochondria, and tumor clone formation. These changes were accompanied by normalization in morphology as well as altered gene expression in cancer cells, favoring a shift toward differentiation. Mechanistic studies revealed that CA increased SUMO1 protein expression by stabilizing SUMO1 mRNA, thus causing SUMOylation of c-Myc and decreasing p-c-Myc, which in turn downregulated the expression of the three miR-17 family members. The decrease in miR-20a, miR-93 and miR-106b levels promoted p21 expression leading to cell differentiation. Activation of the "c-Myc SUMOylation/miR-17 reduction/p21 elevation axis" by CA resulted in cell differentiation and attenuated the cancer behavior of solid tumor cells.
Our study showed that the SUMOylation of c-Myc, down-regulated miR-17 family members and up-regulated p21 are important steps for CA-mediated differentiation of cancer cells. SUMOylation is a posttranslational modification that represents a process of protein conjugation with SUMO proteins 1-4. SUMO1, SUMO2, and SUMO3 are rich in vertebrate tissues, and SUMO4 is expressed only in the kidney, lymph nodes and spleen [41]. SUMOylation controls cell cycle progression and embryo development [42,43]. Functional proteins, such as MDM2, Rb, and STAT3, among others, can be modified by SUMOs [44][45][46]. Of these SUMO proteins, SUMO2-mediated SUMOylation plays an important role in ATO-induced differentiation therapy [47]. The c-Myc protein contains 10 lysine residues that can be conjugated with the SUMO protein to modify its oncogenic activity [48]. Herein, we showed that CA increased SUMO1 expression and boosted c-Myc SUMOylation. The SUMOylation at the lysine sites might protect c-Myc from phosphorylation at the S62 and/or T58 sites in its C-terminus [49,50]. The decrease in p-c-Myc in response to CA suppressed the transcriptional activity of the miR-17 family consistent with previous reports [51,52]. For mitochondria, increased expression of SUMO1 reduced ROS production probably through its action on NADPH oxidases [53]. Our results showed that the declined ATP production by CA was SUMO1-dependent, demonstrating, for the first time, the role of SUMO1 in the oxidative phosphorylation for ATP production. The relationship between CA-caused ATP reduction and SUMO1 function in mitochondria needs to be further investigated. Data are presented as mean ± SEM, ** p < 0.01, *** p < 0.001 (CA treated versus untreated control. Scale bars indicate 10 μm). (E) Glioma cell sphere formation of U87MG (upper) and M059J (lower) cells after exposure to CA stained with 0.4% Trypan blue, calculated and observed at 200 × magnifications under the microscope. Data are presented as mean ± SEM, ** p < 0.01, *** p < 0.001 (CA treated versus untreated control. Scale bars indicate 10 μm). (F) CA concentration in the brain was measured at consecutive time points post injection (ip) with 3 rats for each time point. (G) CA treatment (75 mg/kg, ip, bid, for 7 days) did not significantly reduce the body weight of the rats as compared to the sham group (left), and inhibited the C6 glioma tumor growth in rats with a higher inhibitory efficacy than that of temozolomide (right; 20 mg/kg/d, oral, for 5 days). Data are presented as mean ± SD, # p < 0.05, CA group versus TMZ group; * p < 0.05, ** p < 0.01, CA or TMZ group versus model group; n = 6 for each group).
In the present study, we discovered that CA acts as a suppressor for three members of the miR-17 family. It is well recognized that miRNAs can act as oncomiRs or antioncomiRs in cancers [54][55][56][57]. Numerous miRs have been reported to participate in cell cycle regulation, drug resistance, tumor differentiation, or cell "stemness" [58][59][60][61]. As miR-20a, miR-93, and miR-106b share the same critical seed sequence, they are assigned to the miR-17 family (http://www.mirbase.org/). It has been reported that the upregulation of miR-20a, miR-93 and miR-106b is associated with a poor prognosis in breast cancer, pediatric glioma and hepatocellular carcinoma and is linked to the promotion of tumor proliferation and invasion [62][63][64][65]. In the present study, we identified miR-17 family members as oncomiRs that function by inhibiting p21.
Recently, we have been interested in meristematic tissues in plants in search of effective, but non-toxic, tumor differentiation inducers. In this context, CA has been shown to stimulate a complex signaling network that causes phenotypic transformation toward cell differentiation. A decrease in gene expression in response to CA was observed in a group of genes associated with poor differentiation. Among these genes, CD44 and EPCAM are known to be highly expressed in progenitor cells as well as in undifferentiated stem cells [33,66]. Of the oncogenes affected by CA, c-Myc plays an important role in malignant progression by activating a variety of genes (such as those encoding CDKs) and promotes the expression of oncomiRs. For instance, there are four c-Myc E-box binding sites in the CDK4 promoter region, which elevate CDK4 expression and play a critical role in cell-cycle progression by mediating its oncogenic function [67]. Furthermore, c-Myc drives the metabolism of glucose and glutamine for the production of ATP needed for cell proliferation [68]. High expression of c-Myc promotes the expression of multi-oncomiRs, such as miR-17-5p, -92, -106b, -19a, -19b, and miR-20a [69]. Other genes, which likely promote differentiation and show increased expression levels by CA, include MAMLD1, KHSRP, LPAR4, KLHL22, p53, and p21 [70][71][72][73][74][75][76][77]. For example, p21 is well-recognized as an inhibitor of CDKs and acts at the G1/S phase checkpoint to determine the cell fate [78].
Human p21 is an important cellular component that responds to cellular stress and DNA damage [79]. It acts on multiple pathways, leading to cell growth suspension, differentiation, or cellular senescence [80]. For example, p21 prevents retinoblastoma protein (pRb) from undergoing phosphorylation [81] and increases the E2F-associated phosphor-protein (EAPP) level to maintain gene regulation by E2F transcription factors in a repressed status and to induce cell cycle arrest [82]. P21 also interacts with proliferating cell nuclear antigen (PCNA), which is linked to DNA metabolism and repair in response to radiation damage [83]. As p21 is widely accepted as a positive factor for differentiation or cellular senescence, we used it as one of the important differentiation indicators in the study. Using bioinformatics and the dual luciferase reporter assay, we discovered that the miR-17 family members were able to bind the 3' UTR sequence in p21 through complementary base pairing and down-regulated p21 expression. Thus, the suppression of miR-17 family members by CA resulted in increased expression of p21, which then acted as a tumor suppressor protein to induce cell cycle arrest at the G1/S checkpoint and tumor differentiation. However, the involvement of other pathways could not be excluded, as cell differentiation represents a complex process that requires the synchronized operation of several genes/proteins.
In animal experiments, we showed that CA treatment terminated tumor growth. The detection of the genes/proteins associated with differentiation in tumors from the CA-treated animals revealed a profile identical to that observed in vitro. Through tumor re-implantation, we showed that CA also differentiated cancer stem cells. This inference was based upon the observation that the residual tumor cells from the CA-treated animals largely lost the ability to grow new tumors in vivo. It is of note that the tumor inhibition effect of CA in glioma was not as good as that in hepatoma or lung cancer, probably because of the low CA concentration in the brain. The mouse experiments showed a CA concentration in brain (the original CA compound) lower than that used in cell culture. The possible explanations for the CA's effect could be at least 1) CA has 6 structure-similar metabolites in vivo which might also be active against cancer [84], and 2) CA is a substrate of P-glycoprotein (P-gp) [85,86]; thus, the in vivo metabolites of CA might increase CA intracellular concentration by blocking P-gp efflux competitively, forming a synergistic action against glioma.
The CA-induced differentiation response in cancer cells was not detectable in primary human cells, probably because they are already differentiated cells. In animal experiments, CA terminated tumor growth by inducing differentiation and most of the tumor tissue taken from the CA-treated mice was no longer tumorigenic. More significantly, ip administration of CA exhibited no toxicity even at high doses. More research is needed to elucidate the molecular details of the differentiation of cancer cells induced by CA.
Taken together, we found that CA functioned as a safe differentiation inducer for solid tumors. After observing the exceptional therapeutic effect and excellent safety by CA in the recurrent high-grade glioma patients in the Phase I study (NCT02728349), CA has been approved by China FDA for Phase II trial (NCT03758014). Our study strongly suggested that "educating" cancer cells, rather than killing them, is a good option for future anticancer therapy. We consider CA as a promising therapeutic agent for cancer.