Human epithelial colorectal adenocarcinoma cell line Caco-2 was purchased from Kunming Cell Bank. Cells were cultured in MEM-medium (for Caco-2 cells), supplemented with 10% fetal bovine serum (FBS), 100 U/ml penicillin and 100 μg/ml streptomycin (all available from Invitrogen, Grand Island, NY, United States). All cultures were maintained in a humidified environment with 5% CO₂ at 37°C.

Male or female C57BL/6 mice (12 weeks of age) were obtained from Guangdong experimental animal research center (certification code: SCXK (YUE) 2013-0002). All experiments were performed in the specific pathogen free unit of the Guangzhou University of Chinese Medicine (Animal experimental registration code: SYXK (YUE) 2013-0001). The study protocol was approved by the Ethics Committee for Animal Experiments of Guangzhou University of Chinese Medicine and all efforts were made to minimize suffering. Azoxymethane (AOM) (sigma, Cat. #A5486-25MG) was reconstituted to a final concentration of 10 μg/μL when used. Dextran sulfate sodium (DSS), 36–50 kDa (MP Biomedicals, cat. #160110-100G) was prepared to 2.5% DSS by dissolving 2.5 g DSS per 100 ml dH₂O when needed and changed every day for DSS cycle in experiment.

MicroRNA array: RNA of sufficient quality from 2 group samples (n = 2, total 4 samples) was submitted to KangChen-Biotech (Shanghai, China). RNA groups above mentioned by applying the miRCURY locked nucleic acid (LNA) microarray platform (Exiqon, Denmark). All procedures were carried out according to manufacturer’s protocol. Gene Pix 4000B scanner and GenePix Pro 6.0 software (Axon Instruments, Union City, CA, United States) were used to acquire images. Background subtraction and normalization were performed. Two-fold or larger change was set as a threshold of significant difference.

Pre-miR-31, Pre-miR-223, and Pre-let-7f were purchased from Ambion (Austin, TX, United States). Untreated Caco2 cells, growing exponentially, were plated at 2 × 10⁷/well in 2.5 ml DMEM medium for 24 h on six-well plates. Once cells reached about 50% confluence, transfection was conducted. Lipo2000 from Invitrogen (Carlsbad, CA, United States) was used in all transfection processes according to the manufacturer’s instructions.

The whole plant of Ilex rotunda was purchased from Guangzhou in 2016. Dr. Gao Feng authenticated the plant, and a voucher specimen was deposited in Shenyang Pharmaceutical University (Reference No. 2016-0105). The detailed process of isolation and identification for pedunculoside was described in the Supplementary Data.

Whole-cell lysates were prepared using 2% SDS, sonicated and centrifuged (15000 g) at 4°C for 15 min. The supernatants were boiled for 5 min and size-fractionated by SDS/PAGE (10% acrylamide). After transferring proteins on to nitrocellulose filters, the blots were incubated with primary antibodies recognizing APC, AMER3, SLC9A9, GPADH, β-actin, and factor in the Hippo pathway (Cell Signaling Technology, Beverly, MA, United States); following incubation with secondary antibodies, immunocomplexes were developed by using chemiluminescence.

Quantitative real-time PCR was performed using SYBR Premix EXtagII (TaKaRa, Dalian, China) in the PRISM 7900HT system (Applied Biosystems, Carlsbad, CA, United States). For miRNA quantification: Bulge-loop miRNA qRT-PCR Primer Sets (one RT primer and a pair of qPCR primers for each set) specific for miR-31-5p, miR-223-3p, and let-7f-5p were designed by RiboBio (Guangzhou, China). U6 small nuclear RNA (snRNA) was used as endogenous control. Total RNA was isolated from cells using an RNeasy Mini Kit (Qiagen, Valencia, CA, United States). RT-PCR for mRNAs were performed using ABI TaqMan Gene Expression Assays (Applied Biosystems, Foster City, CA, United States). RNA was reverse transcribed by using the high-capacity cDNA archive kit (Applied Biosystems). RT-PCR and subsequent calculations were performed by the Step One Plus Real-time PCR system (Applied Biosystems), which detected the signal emitted from fluorogenic probes during PCR. All samples were analyzed three times. The qRT-PCR results were expressed relative to GAPDH expression levels at the threshold cycle (Ct), which were then converted to fold change (2^-ΔΔCt).

The human SLC9A9, APC, and AMER3 3′ untranslated region (UTR) fragment containing putative binding sites for miR-223-3p or let-7f-5p were amplified by PCR from human genomic DNA. The mutant 3′-UTRs were obtained by overlap extension PCR. The fragments were cloned into a pmirGLO reporter vector (Promega, Madison, WI, United States), downstream of the luciferase gene, to generate the recombinant vectors pmirGLO-WT and pmirGLO-MUT. For the luciferase reporter assay, Caco2 cells were co-transfected with miRNA (pre-miR or scrambled-miR negative control) and reporter vectors (pmirGLO-WT reporter vectors or pmirGLO- MUT reporter vectors), using lipo2000. Luciferase activities were measured with a Dual-Luciferase assay kit (Promega, Madison, WI, United States) according to manufacturer’s instructions at 24 h post-transfection. Experiments were repeated three times in triplicate.

Weigh mice (n = 15 for each group) and inject AOM (12.5 mg/kg) i.p. using a 1 ml sterile syringe. After 1 week 2.5% DSS was prepared to drink for 5 days, followed by 2 weeks of regular water and normal feed. Then start a second cycle of 2.5% DSS for 5 days followed by 2 weeks of recovery. At the same time, peduncloside (20 mg/kg, p.o. per day) treatment was started at the fourth week and continued for 5 weeks. Mice were killed 2 weeks after the end of the last round of DSS by cervical dislocation, the whole colon was removed and split in two along the central axis for measurement and histopathological staining. And the colon tissues were stored at -80°C for RNA and protein expression analysis.

We used paraffin-embedded sections of colonic samples from normal, CAC model and peduncloside treatment mice for the immunohistochemistry analysis. Phospho-YAP (Ser127) (CST #13008), YAP/TAZ (CST#8418), APC (ARG52491) antibodies were used according to manufacturer’s instructions and incubated overnight at 4°C. DAB kit and Hematoxylin nuclear counterstain were used for staining as the instructions.

Results are expressed as mean ± S.D. in experiment. Statistically significant values were compared by using one-way ANOVA analysis of Prism 6, and P < 0.05 indicated statistical significance.

C57BL/6 mice were used to construct the CAC model induced by AOM/DSS method. The microscopic results (H/E) showed that severe intestinal dysplasia, even mucosa neoplasia, could be observed for the mice in the model group at 8 weeks, while mice in the control group bear healthy intestine without any atypical hyperplasia (Figure 1A), suggesting that mice in the 8-week model group are at the precancerous stage. For we were focused on the exploration of the miRNA-based mechanism for CAC prevention, which means we were inclined to determine the stage right before the tumor initiation, and therefore this 8-week CAC model of severe atypical hyperplasia perfectly fell into the scope of our interests. Moreover, the miRNA array was performed to examine the dysregulated miRNAs in mice at 8 weeks. Comparing to those in the control group, mice in the model group showed much higher levels of miR-31-5p and miR-223-3p, but a lower level of let-7f-5p (Figure 1B). Subsequently the expression of these miRNAs were further confirmed by RT-qPCR method, and the result of which proved the up-regulated miR-31-5p and miR-223-3p expression, along with the attenuated let-7f-5p level in the mice of the CAC model group (Figure 1C–E). Thus, miR-31-5p, miR-223-3p, and let-7f-5p were considered to be associated with the oncogenesis for this 8-week CAC model.

Using the standard online software (TargetScan and RNA22), we found that there are predicted binding sites of miR-223-3p within the 3′-UTR of APC (adenomatous polyposis coli) mRNA (Figure 2A), suggesting that miR-31-5p directly interacts with the APC mRNA so as to regulate its expression. To examine the effect of miR-223-3p on APC expression, miR-223-3p was overexpressed by transfection with the miR-223 precursor (pre-miR-223) in Caco2 cells. As shown in Figure 2B, transfection of pre-miR-223 increased the miR-223-3p level remarkably and decreased the APC protein level (Figure 2C,D). Furthermore, luciferase reporter assays were conducted and revealed that miR-31-5p inhibited WT, but not the Mut luciferase activity of APC in Caco2 cells (Figure 2E). Taken together, these results appear to indicate that APC is a direct target gene of miR-223-3p in Caco2 cells.

Through the target analysis via TargetScan, predicted binding sites of let-7f-5p within the 3′-UTR of the SLC9A9 and AMER3 mRNAs were found (Figure 3A), indicating let-7f-5p could suppress SLC9A9 and AMER3 expression post-transcriptionally by directly interacting with SLC9A9 and AMER3 mRNAs. To verify this predicted regulatory effect of let-7f-5p, transfection experiment of let-7f-5p precursor (pre-let-7f) was conducted to increase its level. Transfection of let-7f-5p precursor significantly increased the let-7f-5p level, and the overexpressed let-7f-5p potently attenuated SLC9A9 and AMER3 expression (Figure 3B,C). Moreover, luciferase reporter assays revealed that let-7f-5p inhibited WT, but not the Mut luciferase activity of SLC9AG and AMER3 in Caco2 cells (Figure 3D,E). On the whole, these results suggested that SLC9AG and AMER3 are direct target genes of let-7f-5p in Caco2 cells.

Emerging evidence indicates that miR-31 plays a dual role in tumorigenicity. However, whether miR-31-5p plays as an oncogene in CAC and its potential target molecules are still unclear. Aforementioned results of miR-31-5p, whose basal level in CAC group was almost a hundred times as high as in the control group (Figure 1B,C), indicated the potential importance of miR-31-5p in CAC. Furthermore, we found miR-31-5p could modulate Hippo pathway in Caco2 cells via repressing LATS2 although miR-31 was reported to target LATS2 in other cancers (Gao et al., 2017; Luan et al., 2017). First, miR-31-5p was overexpressed by transfection of its precursor (pre-miR-31) (Figure 4A). Western blot analysis showed that LATS2 was significantly decreased by miR-31-5p overexpression (Figure 4B). Second, to investigate if miR-31-5p could affect the up-stream genes of LATS2 so as to decrease LATS2 protein level, MST1 and MOB1 were examined through western blot analysis after pre-miR-31 transfection. The results revealed that miR-31-5p specifically targeted LATS2 without affecting LATST1, MST1 and MOB1 expression (Figure 4B). Subsequently, we found that the decrease of LATS2 by miR-31-5p definitely affected the down-stream genes of LATS2, such as p-YAP and YAP/TAZ (Figure 4B). Thus, these evidence indicated that in Caco2 cells, miR-31-5p modulates the Hippo pathway by specifically targeting LATS2.

MiR-31 was reported to be overexpressed in UC, and was deemed as one of the prognostic biomarkers for stage II and III colon cancer (Gao et al., 2017; Jacob et al., 2017; Luan et al., 2017). This prompted us to investigate if inflammation-induced factors would cause the overexpression of miR-31-5p so as to exacerbate the tumorigenesis of colon cancer. Thus, miR-31-5p and its down-stream gene c-MYC were evaluated in the LPS-treated Caco2 cells. The results showed that only long-term treatment of LPS (72 h) could lead to miR-31-5p overexpression (Figure 4C). And the down-stream gene c-Myc of miR-31-5p was also up-regulated by 72-h treatment of LPS in Caco2 cells as expected (Figure 4D). Interestingly, similar result was observed in the miRNA array for the 5-week CAC model, in which comparing to the 8-week CAC model, miR-31-5p was basically not up-regulated (Figure 4E,F). All these results suggested that long-term inflammation in intestinal cells may contribute to the aberrant expression of miR-31-5p and thereby its down-stream oncogenic factors.

To construct an in vitro model for the screening of chemopreventive agents against CAC, Caco2 cells were transfected with precursors (for miR-31-5p and miR-223-3p) or the siRNA (for let-7f-5p) to simulate the aberrant expression pattern of miRNAs in the aforementioned CAC mice model. To be noted, western blot analysis showed that pedunculoside, a natural ursane type of triterpenoid, reversed the decreasing tendency of LATS2 caused by miR-31-5p overexpression (Figure 5A). Immunofluorescent staining substantiated that pedunculoside could reverse the miR-31-5p-induced increase of YAP/TAZ (Figure 5B), which is the down-stream gene of LATS2, and therefore abrogated the mRNA overexpression of oncogenic genes caused by miR-31-5p overexpression, such as CTGF, AREG and CYR61 that are down-stream genes of LATS2/YAP/TAZ in the Hippo pathway (Figure 5C). The similar outcome was also found in miR-223-3p overexpression, and the results of which showed that pedunculoside could restore APC protein level lowered by miR-223-3p overexpression (Figure 5D,E). Although pedunculoside did not reverse the effect of let-7f-5p on AMER3 and SLC9A9 (data not shown), this “miR-31-5p and miR-223-3p counteractive” feature of pedunculoside still makes it a good candidate for CAC prevention.

To further substantiate the screening results in the aforementioned Caco2 cell model, the in vivo experiment was performed using the same aforementioned CAC model (Figure 6A). Histological assessment (HE, 100×) showed that in the CAC group the normal structure of glands disappeared with enlarged nucleus, and in the pedunculoside treatment group glands were arranged in disorder with a reduced number of enlarged nucleus (Figure 6B). The event summarization for the dysplasia suggested the different degree of colonic dysplasia: 5 severe, 2 moderate and 1 mild colonic mucosa dysplasia were observed in CAC model group, while 1 severe, 2 moderate and 6 mild dysplasia appeared in pedunculoside treatment group (Figure 6C). As shown in Figure 6D, the CAC model group had an increased thickness of colonic muscle layer compared with control group (99.67 ± 3.24 μm vs. 250.7 ± 10.31 μm, P < 0.01). However, treatment with pedunculoside alleviated the thickened muscle layer caused by AOM/DSS (150.20 ± 7.16 μm, P < 0.01). Moreover, the abnormal morphology showed colonic mucosa hyperemia and even tumor formation (in situ carcinoma) in CAC model, whereas no in situ carcinoma were observed in the mice treated with pedunculoside for 5 week (from 4 to 8 week) as shown in Figure 6E. The colon length of CAC mice was significantly shorter than that of the control group (10.1 ± 1.37 cm vs. 8.02 ± 0.40 cm, P < 0.05) (Figure 6F). In contrast, pedunculoside treatment (20 mg/kg) could prevented the shortening of the colon length induced by AOM/DSS (8.64 ± 0.34 cm, P < 0.05) as shown in Figure 6F. RT-qPCR results showed that only LATS2 mRNA was abrogated in CAC model. And the decrease of LATS2 mRNA level was reversed by pedunculoside (Figure 7A). Immunohistochemical images for APC showed the strong positive in the cytoplasm, very weak positive in the CAC group, and a relatively stronger positive in the pedunculoside group (Figure 7B); On the contrary, immunohistochemical results for YAP showed that the YAP level was very low in normal colon segment, while a significantly enhanced expression of YAP was observed in nucleus for the CAC group.

On the contrary, p-YAP and APC protein expression were both decreased in abnormal dysplasia part compared with positive cytoplasm expression in normal mucosa (Figure 7B). Moreover, the miRNA levels for the pedunculoside group were also evaluated. To be interest, although miR-223-3p and miR-31-5p were up-regulated in CAC model group, pedunculoside specifically down-regulated miR-31-5p level, without affecting miR-223-3p expression (Figure 7C). Subsequently, the protein levels of APC and factors in the Hippo pathway were examined by western blot analysis (Figure 7D). The results showed that the down-stream genes of miR-31-5p and miR-223-3p, including APC and p-YAP, were all decreased in CAC group, which were also rescued by pedunculoside treatment.