Stress responsive miR-31 is a major modulator of mouse intestinal stem cells during regeneration and tumorigenesis

Intestinal regeneration and tumorigenesis are believed to be driven by intestinal stem cells (ISCs). Elucidating mechanisms underlying ISC activation during regeneration and tumorigenesis can help uncover the underlying principles of intestinal homeostasis and disease including colorectal cancer. Here we show that miR-31 drives ISC proliferation, and protects ISCs against apoptosis, both during homeostasis and regeneration in response to ionizing radiation injury. Furthermore, miR-31 has oncogenic properties, promoting intestinal tumorigenesis. Mechanistically, miR-31 acts to balance input from Wnt, BMP, TGFβ signals to coordinate control of intestinal homeostasis, regeneration and tumorigenesis. We further find that miR-31 is regulated by the STAT3 signaling pathway in response to radiation injury. These findings identify miR-31 as a critical modulator of ISC biology, and a potential therapeutic target for a broad range of intestinal regenerative disorders and cancers.


Introduction
The intestinal epithelium is one of the most rapidly renewing tissues, undergoing complete turnover in approximately 3 days (Leblond and Walker, 1956). This rapid turnover protects against insults from bacterial toxins and metabolites, dietary antigens, mutagens, and exposure to DNA damaging agents including irradiation. Upon insult, the rapid intestinal regeneration is particularly important as impaired regeneration can result in epithelial barrier defects that can lead to rapid dehydration and translocation of intestinal microbiota into the bloodstream. The processes of normal tissue turnover and intestinal regeneration are driven by intestinal stem cells (ISCs) that reside at the bottom of crypt and generate the precursors for the specialized differentiated cells (Barker, 2014;Li and Clevers, 2010).
It has been extensively reported that ISC compartment includes two functionally and molecularly distinct stem cell populations (Barker, 2014;Li and Clevers, 2010;Gehart and Clevers, 2015): The active crypt base columnar (CBC) stem cells , (Barker et al., 2007) and a more dormant, reserve ISC population that reside above the crypt base and exhibit no Wnt pathway activity, also referred as +4 cells due to their position at the crypt (Montgomery et al., 2011;Sangiorgi and Capecchi, 2008;Tian et al., 2011;Takeda et al., 2011;Li et al., 2014;Yan et al., 2012). The CBCs often identified and isolated based on the expression of Lgr5, a Wnt target gene (Barker et al., 2007). During homeostasis, steady-state proliferation of CBCs is driven by extrinsic niche signals -high canonical Wnt activity promotes CBC self-renewal and proliferation (Barker et al., 2007;Miyoshi, 2017) while BMP signals antagonize it (Kosinski et al., 2007). In contrast to the active CBCs, the reserve ISCs represent a slow-cycling population of stem cells that are resistant to high doses of ionizing radiation and appear dispensable for homeostasis (Sangiorgi and Capecchi, 2008;Yousefi et al., 2016). These reserve ISCs are identified through CreERT knockin reporter alleles at the Bmi1 and Hopx loci, as well as by an Tert-CreERT transgene (Montgomery et al., 2011;Sangiorgi and Capecchi, 2008;Tian et al., 2011;Takeda et al., 2011;Li et al., 2014). Reserve ISCs do not have an active Wnt signaling pathway and are refractory to Wnt signals in their resting state (Takeda et al., 2011;Li et al., 2014;Li et al., 2016). Although the activity of the BMP pathway has never been directly examined specifically in reserve ISCs, indirect evidence suggests that it may help to promote their dormancy (Reynolds et al., 2014;He et al., 2004; eLife digest Cells lining the inner wall of the gut help to absorb nutrients and to protect the body against harmful microbes and substances. Being on the front line of defense means that these cells often sustain injuries. Specialized cells called intestinal stem cells keep the tissues healthy by replacing the damaged and dying cells.
The intestinal stem cells can produce copies of themselves and generate precursors of the gut cells. They also have specific mechanism to protect themselves from cell death. These processes are regulated by different signals that are generated by the cell themselves or the neighboring cells. If these processes get out of control, cells can easily be depleted or develop into cancer cells. Until now, it remained unclear how intestinal stem cells can differentiate between and respond to multiple and simultaneous signals.
It is known that short RNA molecules called microRNA play an important role in the signaling pathways of damaged cells and during cancer development. In the gut, different microRNAs, including microRNA-31,help to keep the gut lining intact. However, previous research has shown that bowel cancer cells also contain high amounts of microRNA-31.
To see whether microRNA-31 plays a role in controlling the signaling systems in intestinal stem cells, Tian, Ma, Lv et al. looked at genetically modified mice that either had too much microRNA-31 or none. Mice with too much microRNA-31 produced more intestinal stem cells and were able to better repair any cell damage. Mice without microRNA-31 gave rise to fewer intestinal stem cellsand had no damage repair, but were able to stop cancer cells in the gut from growing.
The results showed that microRNA-31 in intestinal stem cells helps the cells to divide and to protect themselves from cell death. It controlled and balanced the different types of cell signaling by either repressing or activating various signals. When Tian et al. damaged the stem cells using radiation, the cells increased their microRNA-31 levels as a defense mechanism. This helped the cells to survive and to activate repair mechanisms. Furthermore, Tian et al. discovered that microRNA-31 can enhance the growth of tumors.
These results indicate that microRNA-31 plays an important role both in repairing gut linings and furthering tumor development. A next step will be to see whether cancer cells use microRNA-31 to protect themselves from chemo-and radiation therapy. This could help scientists find new ways to render cancerous cells more susceptible to existing cancer therapies. Kishimoto et al., 2015). During epithelial regeneration upon stresses, reserve ISCs give rise to Wnt high Lgr5 + CBCs that generate the precursor cells of the specialized differentiated cells (Tian et al., 2011;Takeda et al., 2011;Li et al., 2014). In addition, it has been documented that Lgr5-CreERT-or Bmi1-CreERT-marked cells can act as the cells of origin of intestinal cancer in mice (Sangiorgi and Capecchi, 2008;Barker et al., 2009). However, it remains unclear how ISCs differentially sense and respond to multiple signals under both physiological and pathological conditions, and whether these signals contribute to intestinal tumorigenesis.
MicroRNAs represent a broad class of 18-22 nucleotide noncoding RNAs that negatively regulate the stability and translation of target mRNAs. Mounting evidence indicates that microRNAs play important roles in stress-activated pathways (Leung and Sharp, 2010;Mendell and Olson, 2012;Emde and Hornstein, 2014) and in control of somatic stem cell fate and tumorigenesis (Gangaraju and Lin, 2009;Sun and Lai, 2013;Yi and Fuchs, 2011). Hundreds of microRNAs have been identified in the intestinal epithelium (McKenna et al., 2010). Global ablation of microRNA activity through genetic deletion of the microRNA processing enzyme Dicer demonstrated that microRNAs are critical for homeostasis of intestinal epithelium (McKenna et al., 2010). Recently, numerous reports demonstrate that specific microRNAs play important roles in the complex intestinal immune system and in the epithelium during homeostasis including miR-155, miR-29, miR-122, miR-21, miR-146a and miR-143/145 (Runtsch et al., 2014). Particularly, miR-143/145 are essential for intestinal epithelial regeneration after injury, acting non cell-autonomously in sub-epithelial myofibroblasts (Chivukula et al., 2014), indicating potential importance of microRNA activity in intestinal regeneration.
In the ISC compartment, the function of miR-31 is of a particular interest, as it becomes overexpressed in colorectal cancer (Bandrés et al., 2006;Cottonham et al., 2010;Wang et al., 2009;Yang et al., 2013) and increases during the progression of inflammation-associated intestinal neoplasia (Olaru et al., 2011). In addition, it has been reported that miR-31 is enriched in mammary stem/progenitor cells, suggesting a potential role in somatic stem cells (Ibarra et al., 2007). Here we utilized gain-and loss-of-function mouse models to show that a damage-responsive microRNA, miR-31 drives proliferative expansion of both active and dormant ISCs, and acts as an oncogene promoting intestinal tumorigenesis in different models. Our findings implicated miR-31 as a potential high-value therapeutic target for a broad range of intestinal regenerative disorders and cancers.

MiR-31 expression pattern in intestine under physiologIcal and pathological conditions
Elevated miR-31 expression has been previously observed in colorectal cancers (Bandrés et al., 2006;Cottonham et al., 2010;Wang et al., 2009;Yang et al., 2013), however its expression in normal intestinal epithelium, particularly in ISCs, remains unclear. To begin addressing a potential role for miR-31 in the intestinal epithelium and ISCs, first we examined its expression pattern in intestine. MiR-31 expression levels are the highest in the Lgr5-GFP high crypt base columnar stem cells, intermediate in Lgr5-GFP low transit-amplifying cell population and the lowest in Lgr5-GFP neg populations ( Figure 1A). Higher level of miR-31 was also found in Hopx + reserve ISCs than that in bulk epithelial cells ( Figure 1A), based on isolation with Hopx-CreERT;mTmG alleles from mice 15 hr after tamoxifen injection. Consistently, in situ hybridization revealed that miR-31 expression levels are generally higher in the crypts than villi. MiR-31 is predominantly expressed in the epithelial cells of intestinal crypt, including stem cells and transit amplifying cells ( Figure 1B). Next, we examined miR-31 expression in response to intestinal injury. Mice were exposed to 12 Gy g-IR and then miR-31 expression was examined at various timepoints during the recovery phase. MiR-31 levels transiently and markedly drop by 24 hours (coincident with full proliferative arrest/DNA damage response), and then sharply upregulated 48 hours post-g-IR (during initiation of regenerative proliferation from the radioresistant ISCs), and then return to baseline levels within one week (after full recovery) ( Figure 1C). In situ hybridization reveals miR-31 expressing cells to be located in the regenerative foci known to exhibit high Lgr5 expression and Wnt pathway activity ( Figure 1D). Together, these data suggest a role for this microRNA in ISC-driven regeneration.
Crypt height (μm)  MiR-31 promotes intestinal epithelial cell turnover along the Cryptvillus axis To determine the function of miR-31 in the mouse intestine, we generated both gain-and loss-offunction mouse models. MiR-31 gain-of-function was achieved with a targeted, inducible Rosa26-rtTA;TRE-miR-31 mouse model (TRE-miR31) and doxycycline (Dox)-mediated induction of miR-31 in the intestinal epithelium was validated by qRT-PCR ( Figure 1-figure supplement 1A,B). For the loss-of-function, we generated constitutive miR-31 null mice using RNA-guided CRISPR/Cas9 nucleases ( Figure 1-figure supplement 1C). The 402 bp DNA fragment containing miR-31 was deleted in the knockout (KO) allele ( Figure 1-figure supplement 1D), which was validated by sequencing and qRT-PCR ( Figure 1-figure supplement 1E). We also generated a Villin-Cre-mediated intestinespecific conditional miR-31 null mice (cKO) using traditional homology-directed gene targeting ( MiR-31 induction in response to Dox administration in TRE-miR31 mice resulted in a significant reduction in body weight after 2 weeks ( Figure 1E) and intestinal lengths were moderately, but (Top) and TRE-miR31 (miR-31 overexpressing) intestinal section used as a positive control (Bottom). (C) qRT-PCR for miR-31 in the intestinal epithelium after exposure to 12 Gy g-IR at indicated time points. n = 3 biological replicates. (D) In situ hybridization for miR-31 in intestines without g -IR treatment (non-IR), and intestines 4 days after 12 Gy g-IR. Arrows, miR-31 positive regenerative foci. Dashes boxes indicate the high magnification images in right panels. Scale bar: 50 mm. (E) Quantification of body weight from M2rtTA and TRE-miR31 mice at the age of 8 weeks before and after Dox treatment for 2 weeks. Quantification of intestine length from M2rtTA and TRE-miR31 mice following 2 week Dox induction. n = 6 biological replicates. ***p<0.001. (F) Representative histologic images showing extension of crypt height in jejunum from TRE-miR31 mice, and quantification of crypt height from M2rtTA and TRE-miR31 intestine. Both M2rtTA and TRE-miR31 mice were treated with Dox for 2 weeks. n = 3 biological replicates. Scale bar: 50 mm. ***p<0.001. (G) Immunohistochemistry for Ki67 and quantification of Ki67 + cells per crypt in M2rtTA andTRE-miR31 jejunum, showing an expanded proliferative zone in TRE-miR31 mice following 2 weeks of Dox induction. n = 3 biological replicates. Scale bar: 50 mm. ***p<0.001. (H) Immunohistochemistry for cleaved-Caspase 3 (Casp3) and quantification of Casp3 + cells in the top of intestinal villi from M2rtTA andTRE-miR31 mice following 2 weeks of Dox induction. n = 3 biological replicates. 60 villi were quantified in each mouse. Scale bar: 100 mm. ***p<0.001. (I) Representative histologic images and quantification of crypt height in intestines from miR-31 +/À and miR-31 À/À mice at 2 months of age. Brackets mark crypts. Scale bar: 100 mm. n = 3 biological replicates. ***p<0.001. DOI: https://doi.org/10.7554/eLife.29538.003 The following source data and figure supplements are available for figure 1: Source data 1. Source data for Figure 1C significantly shorter than controls ( Figure 1E). Dox treatment of TRE-miR31 mice for 2 weeks resulted in expansion of intestinal crypts ( Figure 1F). Unexpectedly villus lengths were mildly shortened, and thus the total length of the crypt-villus was not significantly altered in TRE-miR31 mice ( Figure 1-figure supplement 2A). The expanded crypts were also found in the TRE-miR31 duodenum and ileum ( Figure 1-figure supplement 2B). The length of intestinal crypts in the control M2rtTA mice was not significantly altered at different time points in response to Dox treatment D). In contrast, crypts were significantly expanded in TRE-miR31 mice after 10 days of Dox treatment, this crypt expansion remained stable for up to 1 year with continuous Dox induction (Figure 1-figure supplement 2C-E). Given that crypt elongation reached maximal levels within 2 weeks of Dox induction, we conducted most of the subsequent assays at this time point. More mitotic cells were found in the TRE-miR31 crypts ( Figure 1G and Figure 1-figure supplement 3A,B), while more apoptotic cells were detected at the top of TRE-miR31 villi ( Figure 1H and Figure 1-figure supplement 3A,B). The number of Lgr5 + ISCs increased in TRE-miR31 mice after 10 day Dox treatment, while no significant difference was found between them after 7 days of Dox induction (Figure 1-figure supplement 3C,D). In addition, there were fewer differentiated cells including enteroendocrine, goblet and Paneth cells in TRE-miR31 intestine than the controls (Figure 1-figure supplement 4A,B), indicating an impaired cell differentiation. These results suggest that miR-31 induction accelerates the conveyer-belt movement of proliferative cells exiting the cell cycle and progressing into the villi to ultimately be shed into the lumen, which could comprise the differentiation of specialized intestinal cell types.
Next, we examined the consequence of miR-31 loss in both miR-31 germline knockout (KO) and Villin-Cre-driven intestinal epithelial conditional KO (cKO) mice. We followed these mice up to six months. Both miR-31 KO and cKO mice were viable and fertile with no apparent gross phenotypes observed. No differences in the body weight and intestinal length were found between control and miR-31 KO mice ( Figure 1-figure supplement 5A), and the transmission of miR-31 knockout alleles generally followed Mendelian ratios ( Figure 1-figure supplement 5B). Despite this, loss of miR-31 led to a significant reduction in crypt height with fewer proliferative cells ( Figure 1I and . These results suggest that miR-31 loss functions within intestinal epithelium. We further analyzed DNA synthesis and migration of epithelial cells along the crypt-villus axis after a single pulse of BrdU. Upward movement of BrdU + cells from crypts to villi was enhanced in TRE-miR31 mice, and this movement was impaired in miR-31 À/À mice (Figure 1-figure supplement 8). Taken together, these data indicate that miR-31 functions within the intestinal epithelium to maintain a proper balance between stem cell proliferation, differentiation, and epithelial cell death for optimal intestinal homeostasis.

MiR-31 is critical for intestinal epithelial regeneration following irradiation
The dynamic changes of miR-31 expression in response to irradiation prompted us to investigate its function during intestinal epithelial injury repair. Intestinal histology of cKO and control Vil-Cre mice was comparable two hours after 12 Gy g-IR ( Figure 3A). However, by 4 days post-g-IR, there were significantly fewer regenerative foci and fewer proliferative cells per regenerative focus in cKO mice ( Figure 3A). Consistently, intestinal regeneration in response to g-IR was significantly impaired in miR-31 À/À mice (Figure 3-figure supplement 1A,B). Conversely, in the intestine of TRE-miR31 mice pre-treated for 2 weeks with Dox, there were more regenerative foci with higher numbers of proliferative cells than in the control mice ( Figure 3-figure supplement 1A,B). These data suggest that miR-31 is important for intestinal epithelial regeneration in response to irradiation. To understand the phenotype resulting from miR-31 modulation, we assayed for apoptotic cells in cKO mice at early stages after irradiation. Loss of miR-31 increased apoptosis in the crypts 2 and 4 hours post-irradiation prior to any overt histological changes ( Figure 3B). Quantification of apoptotic cell position analysis reveals that apoptotic events occur with the highest frequently in CBC cells, but are still found in transit-amplifying and +4 zones of cKO crypts, compared to control mice ( Figure 3B). Further, flow cytometry for live cell and apoptotic markers within the Lgr5-GFP + population confirmed higher frequency of late apoptotic Lgr5 + cells (AnnexinV + /7AAD + ) and lower frequency of early apoptotic Lgr5 + cells (AnnexinV + /7AAD À ) and live Lgr5 + cells (AnnexinV -/7AAD -) in cKO mice, relative to controls (Figure 3-figure supplement 1C). These data suggest that loss of miR-31 increases apoptosis of Lgr5 + cells in response to irradiation. Next, we examined its effect on cell proliferation. Cell cycle analysis indicates that more Lgr5-GFP + cells resided in G0 relative to G1/ S/G2/M in cKO mice 2 hours after g-IR (Figure 3-figure supplement 1D). In agreement, expression levels of Lgr5 were dramatically up-regulated in TRE-miR31 mice and prominently down-regulated in Figure 2 continued n = 4 biological replicates. ***p<0.001. (D) Assessment of 1.5-hour-pulse EdU incorporation in Lgr5 + CBC cells in M2rtTA, and TRE-miR31 mice following 2 weeks of Dox treatment, and in miR-31 +/À and miR-31 À/À intestine. ***p<0.001. (E) Crypts purified from M2rtTA and TRE-miR31 mice grown in organoid cultures with Dox. Representative gross images of budding organoids, and quantification of budding and apoptotic organoids at day 7. Scale bar: 500 mm. n = 5 technical replicates. (F) X-gal staining showing lineage tracing events from Lgr5 + ISCs. Lgr5-eGFP-CreERT;R26-LSL-LacZ;TRE-miR31 mice and its control counterpart were pretreated with Dox for 2 weeks, injected with a single dose tamoxifen, and analyzed 2 and 4 days after injection. Scale bar: 100 mm. n = 3 biological replicates. (G) Quantification of the length of LacZ + cells and LacZ + units in Panel F. ***p<0.001. (H) qRT-PCR analysis for Hopx in intestines from M2rtTA, TRE-miR31, Vil-Cre and cKO mice. n = 3 biological replicates. **p<0.01; ***p<0.001. (I) Lineage tracing events from Hopx + ISCs. Hopx-CreERT;mTmG;TRE-miR31 mice and their control counterparts were pretreated with Dox for 2 weeks, injected with a single dose of tamoxifen, and analyzed 15 hr after injection. Hopx-CreERT;R26-LSL-LacZ;TRE-miR31 and their control counterparts were analyzed 4 days after inject with the same treatment. Scale bar: 100 mm. n = 3 biological replicates. (J) Quantification of Cleaved Caspase3 + cells at indicated positions in the intestinal crypts of Vil-Cre and miR-31 cKO mice in Figure  Reserve ISCs, marked either by Bmi1-CreER or Hopx-CreER reporters, have been reported to resist high dose of radiation, being able to replenish the depleted CBC compartment and regenerate the epithelium after irradiation (Sangiorgi and Capecchi, 2008;Tian et al., 2011;Takeda et al., 2011;Yan et al., 2012), (Yousefi et al., 2016). Thus, we examined the response of Hopx-CreERmarked reserve ISCs to 12 Gy g-IR upon miR-31 induction and deletion. Lineage-tracing assay revealed that miR-31 induction promoted epithelial regeneration from the Hopx + reserve stem cells ( Figure 3F and Figure 3-figure supplement 1E). Conversely, the number and the size of regenerative foci originating from Hopx-CreER;Rosa26-LoxP-Stop-LoxP-LacZ-marked cells were markedly reduced in miR-31 À/À mice ( Figure 3G). In line with this, the frequency of LacZ + /Ki67 + cells was significantly lower in miR-31 À/À mutants compared to controls ( Figure 3H). Taken together, miR-31 deficiency-mediated the reduction in proliferation and increase in apoptosis within both CBC and reserve ISC compartments can account for the impaired regeneration of miR-31 null intestine.

MiR-31 activates the Wnt pathway and represses the BMP and TGFb pathways
Canonical Wnt pathway activity is a major driving force for self-renewal of CBCs and epithelial regeneration after injury , and, thus we examined the effect of miR-31 on Wnt activity. We utilized Axin2-LacZ Wnt reporter mice, which act as a broad readout for canonical Wnt activity, and normally showed its activity to be restricted to the base of crypts in control mice, as expected ( Figure 4A) (Davies et al., 2008). In contrast, Wnt pathway activity was strikingly absent from CBCs of miR-31 À/À crypts, appearing only faintly above the crypt base in the early TA zones ( Figure 4A). Conversely, Wnt activity was expanded in TRE-miR31 crypts ( Figure 4A,B). In agreement, the number of nuclear b-Catenin-positive cells was significantly reduced in miR-31 À/À intestinal crypts at 2 and 4 months of age ( Figure 4-figure supplement 1A). Conversely, they increase in TRE-miR31 crypts 14 days and 2 months after Dox induction (Figure 4-figure supplement 1B). Consistently, the expression levels of Ctnnb1 (encoding b-Catenin) and the Wnt targets, Ccnd1 (encoding Cyclin D1), Myc and Axin2 were significantly reduced in miR-31 À/À intestine both at the RNA and protein levels ( Figure 4C,D). In contrast, expression levels of the above genes were enhanced in TRE-miR31 intestinal epithelium following 2 weeks of Dox induction ( Figure 4E,F). The Quantification of Casp3 + cells at indicated positions in intestinal crypts of Vil-Cre and cKO mice 2 hrs post g-IR. Scale bar: 50 mm. n = 3 biological replicates, and 50 crypts were quantified in each single mouse. ***p<0.001. (C) qRT-PCR analysis for Lgr5 in intestines from M2rtTA, TRE-miR31, miR-31 +/À and miR-31 À/À mice 2 hrs, 2 and 4 days post 12 Gy irradiation. M2rtTA and TRE-miR31 mice were pretreated with Dox for two weeks. n = 3 biological replicates at each time points. *p<0.05; **p<0.01; ***p<0.001. (D) Schematic of Lgr5-eGFP-CreERT; R26-LSL-LacZ lineage tracing experiment after irradiation. X-gal staining showing lineage tracing events from Lgr5 + ISCs. Lgr5-eGFP-CreERT;R26-LSL-LacZ;TRE-miR31 mice and their control counterparts were pretreated with Dox for 2 weeks, injected with a single dose tamoxifen and then immediately exposed to 10 Gy g-IR, and analyzed 2 and 4 days after g-IR.  Figure 4G). To test whether Wnt activity is directly impacted by miR-31, we analyzed the effects of gain-and loss-of-function of miR-31 on expression of Wnt target genes in HCT116 human colorectal carcinoma cells. Ccnd1, Ctnnb1, Myc and Axin2 were markedly increased in miR-31 over-expressing cells, relative to controls ( Figure 4H). Conversely, these genes were downregulated upon miR-31 inhibition ( Figure 4H). Considering that HCT116 cells are heterozygous for a b-Catenin gain-of-function mutation at the Gsk3b target site S45 (Ctnnb1 +/S45mt ) (Ilyas et al., 1997), (Kaler et al., 2012), we examined b-Catenin protein levels. Consistently, b-Catenin was up-regulated in the presence of miR-31 mimics, and down-regulated upon miR-31 inhibition (Figure 4-figure supplement 1C). The Wnt reporter (Topflash/Fopflash) assay using HCT116 cells further confirmed that miR-31 induction enhanced Wnt activity, while inhibition of miR-31 repressed it ( Figure 4I). To test the functional relevance of miR-31 potentiation of canonical Wnt activity, we cultured organoids with varying combinations of miR-31 induction and R-spondin, the Lgr5 ligand. Wnt activation by R-spondin is critical for normal organoid growth and budding . Interestingly, we observed that miR-31 induction via TRE-miR31 was sufficient to maintain crypt organoid growth and budding in the absence of R-spondin ( Figure 4J,K) and that the Dox-treated TRE-miR31 organoids can be normally passaged at least five times, similar to the organoids cultured with R-spondin ( Figure 4L). Together, these findings demonstrate that miR-31 activates the canonical Wnt signaling in the crypts of small intestine.

MiR-31 contributes to tumor growth through Wnt activation and TGFb and BMP repression
Given that miR-31 promotes proliferation and inhibits apoptosis in the ISCs, it is plausible that miR-31 may function in intestinal tumorigenesis. Supporting this notion, miR-31 has been found to be upregulated in human colorectal cancers and in colitis (Bandrés et al., 2006;Cottonham et al., 2010;Wang et al., 2009;Yang et al., 2013). We tested the role of miR-31 in intestinal tumorigenesis and observed that miR-31 mimics promoted proliferation of HCT116, SW480 and LOVO colon cancer cells in vitro (Figure 7-figure supplement 1A). Conversely, inhibition of miR-31 with anti-miR-31 abrogated growth of these cells (Figure 7-figure supplement 1A). We further performed xenograft assays using miR-31 mimics-and inhibitor-treated HCT116 cells. Thirty days after grafting, tumor volume and weight were increased in miR-31 mimic-treated tumors, and markedly reduced in miR-31 knockdown tumors ( Figure 7A). The decrease in tumor size from miR-31 inhibition coincided with the reduction in Ki67 + and Cyclin D1 + proliferating cells ( Figure 7B and Figure 7-figure supplement 1B), and correlated with reduced Wnt activity and increased BMP and TGFb activities (Figure 7-figure supplement 1B). To verify these findings in more physiologically relevant settings, we examined tumor formation in the AOM-DSS (Azoxymethane-Dextran Sodium Sulfate) model of the inflammation-driven colorectal adenocarcinoma (De Robertis et al., 2011). In comparison with the controls, we observed a marked decrease in both tumor size and number in miR-31 À/À mice ( Figure 7C), along with a concomitant reduction in proliferating cells ( Figure 7D,E), and reduced Wnt pathway and increased BMP and TGFb activity ( Figure 7D,F). This tumor-promoting effect of miR-31 in mice became even more evident when miR-31 was deleted in Vil-Cre;Apc flox/+ mice. Intestinal adenomas form in this mouse model upon loss of heterozygosity at the Apc locus, which is relevant to human disease in that spontaneous loss of Apc is found in the vast majority of human colorectal cancer (Kinzler et al., 1991;Nagase et al., 1992). Loss of miR-31 in this animal model remarkably reduced tumor burden ( Figure 7G), which was associated with decreased Wnt activity, enhanced BMP and TGFb signaling, and decreased proliferating cells ( Figure 7H-J and Figure 7figure supplement 1C). Correspondingly, the miR-31 targets Axin1, Dkk1, Gsk3b, Smad4 and Bmpr1a were up-regulated in the miR-31 null tumors (Figure 7-figure supplement 1D). Together, these data demonstrate that miR-31 plays an oncogenic role in intestinal and colorectal tumorigenesis by mediating activation of Wnt and repression of BMP and TGFb signaling pathways.  Figure 5A. n = 3 biological replicates. (E) Western blotting for Axin1, Gsk3b, Dkk1, Bmpr1a and Smad4 in M2rtTA and TRE-miR31 intestine following 2 weeks of Dox induction. b-Tubulin was used as a loading control. n = 3 biological replicates. (F) Ratio of luciferase activity of miR-31 mimics versus scramble RNA in wild type and mutant 3'UTR constructs based on 3 independent experiments. *p<0.05; **p<0.01; ***p<0.001. (G) RNA crosslinking, immunoprecipitation, and qRT-PCR (CLIP-PCR) assay for Dkk1, Axin1, Gsk3b, Smad3, Smad4 and Bmpr1a upon Ago2 antibody immunoprecipitates in response to miR-31 mimics and scramble RNA (NC). IgG was used as a negative control. (H) qRT-PCR analysis for Axin1, Gsk3b, Dkk1, Smad3, Bmpr1a, Smad4 andTgfbr2 in miR-31 +/À and miR-31 À/À intestine 4 days post 12 Gy g-IR. n = 3 biological replicates. *p<0.05; **p<0.01; ***p<0.001. (I) Immunohistochemistry for Axin1, Gsk3b and Dkk1 in miR-31 +/À and miR-31 À/À intestinal crypts 4 days post 12 Gy g-IR. Scale bar: 25 mm. (J) Immunohistochemistry for p-Smad2/3, p-Smad1/5/8 and b-Catenin in miR-31 +/À and miR-31 À/À intestinal crypts 4 days post 12 Gy g-IR. Scale bar: 25 mm. Lastly, we asked how radiation injury induces miR-31 expression. We analyzed a 2 kb region upstream of the transcription start site of the miR-31 gene locus for the potential binding sites of transcription factors using the JASPAR database and identified one STAT3 and two NF-kB binding sites ( Figure 8A). Interestingly, the STAT3 and NF-kB signaling pathways were shown to be activated in response to g-IR, evidenced by p-STAT3 and p65 levels, respectively ( Figure 8B,C). The activation of the STAT3 pathway occurred mainly in the regenerative foci where miR-31 is highly induced, while NF-kB was more prominently activated in villi where little miR-31 is present and not in the regenerative foci ( Figure 8D). This suggested a link between STAT3 activity and miR-31 upon irradiation. To verify whether active STAT3 signaling could induce miR-31 expression, mICc12 intestinal epithelial cells were treated with IL-6, a known activator of the STAT3 signaling. Indeed, miR-31 expression was significantly induced upon IL-6 treatment ( Figure 8E), concomitant with the activation of the STAT3 pathway ( Figure 8F). In contrast, inhibition of STAT3 signaling with Stattic prominently dampened miR-31 induction response to IL-6 treatment ( Figure 8G), and reduced STAT3 signaling ( Figure 8H). This inhibitory effect on miR-31 expression was further validated using Stat3 siRNA ( Figure 8I,J). Importantly, miR-31 was induced by IL-6 in the organoid cultures, indicating that this is an epithelial cell-autonomous mechanism ( Figure 8K). Luciferase reporter assays reveal that IL-6 is able to induce its activity, while mutation of the p-STAT3 binding site blocked it ( Figure 8L). Furthermore, Chromatin Immunoprecipitation (ChIP) assays show that p-STAT3 is recruited to its binding site on the miR-31 promoter ( Figure 8M). Thus, our data strongly suggest that STAT3 activity potentiates miR-31 induction to promote crypt regeneration in response to radiation injury.

Discussion
The intestinal epithelium is one of the most rapidly renewing tissues (Leblond and Walker, 1956). Those Lgr5 + CBC stem cells residing at the base of crypts maintain the proliferative capacity necessary to meet this demands of high-turnover tissue, which is driven by activation of the canonical Wnt pathway, as well as repression of BMP signaling (Li and Clevers, 2010), (Li et al., 2014), (Kosinski et al., 2007). Wnt pathway activity and BMP inhibition are believed to be the niche for cycling CBCs. However, it is largely unknown how those Lgr5 + CBCs integrate the signals of Wnt antagonists and activators of BMP and TGFb. Here we show that the miR-31 activates Wnt signaling by directly repressing a cohort of Wnt antagonists Dkk1, Axin1 and Gsk3b, and represses BMP/  Figure 7-figure supplement 1B. ***p<0.001. (C) Representative photograph of distal colon resected from WT and miR-31 À/À mice at the end of AOM-DSS protocol. Frequency and tumor size of inflammation-driven colorectal adenomas in mice treated with the AOM-DSS protocol, with or without miR-31 deletion. n = 6 mice per group, *p<0.05; **p<0.01. (D) H&E, and immunohistochemistry for Ki67, b-Catenin and p-Smad1/5/8 in adenomas of WT and miR-31 À/À mice resulting from AOM-DSS treatment. Scale bar: 100 mm. (E) Quantification of Ki67 + cells in Panel D. ***p<0.001. (F) Western blotting for p-Smad2/3, p21, Axin1, b-Catenin, Cyclin D1 in adenomas of WT and miR-31 À/À mice resulting from AOM-DSS treatment. b-Tubulin was used as a loading control. (G) Representative photograph of intestine resected from Vil-Cre;Apc fl/+ and Vil-Cre;Apc fl/+ ;miR-31 À/À mice at 6 months of age. Arrows point to tumors. Quantification of tumor number and tumor volume in intestines from these mice. n = 6 biological replicates. ***p<0.001. (H) Representative histology of intestine resected from Vil-Cre;Apc fl/+ and Vil-Cre;Apc fl/+ ;miR-31 À/À mice at 6 months of age. Arrows point to tumors. Scale bar: 2.5 mm. (I) Immunohistochemistry for b-Catenin and quantification of nuclear b-Catenin positive cells in Vil-Cre;Apc fl/+ and Vil-Cre; Apc fl/+ ;miR-31 À/À tumors. (Black, Vil-Cre;Apc fl/+ ; Blue, Vil-Cre;Apc fl/+ ;miR-31 À/À ). n = 6 biological replicates. Scale bar: 50 mm. ***p<0.001. (J) Immunohistochemistry for p-Smad2/3, p-Smad1/5/8 and Ki67 in Vil-Cre;Apc fl/+ and Vil-Cre;Apc fl/+ ;miR-31 À/À tumors. Scale bar: 50 mm. TGFb signaling by directly inhibiting activators of the pathways, Smad3, Smad4 and Bmpr1a, pointing to an important role of miR-31 acting as a rheostat to integrating niche signals sensed by cycling CBCs. In agreement with this point, our in vivo analysis demonstrated that miR-31 induction increases the number of Lgr5 + CBCs whereas miR-31 deletion reduces CBC frequency. Niche Wnt signals likely originate from sub-epithelial telocytes whose presence is required for CBC activity, and possibly to a lesser extent from Paneth cells, who secrete Wnt ligands but are dispensable for CBC activity (Durand et al., 2012;Aoki et al., 2016;Sato et al., 2011;Kim et al., 2012;San Roman et al., 2014;Kabiri et al., 2014). BMP antagonists noggin and gremlin are similarly secreted by submucosal tissues below the crypts (Kosinski et al., 2007), repressing the BMP signaling in CBCs. Thus, sub-epithelial mesenchyme constitutes an extrinsic niche for cycling ISCs. In contrast to secretory signals from an extrinsic niche, miR-31 appears to be an intrinsic coordinator of these extrinsic niche signals, supporting canonical Wnt and represses BMP/TGFb signals within CBCs. Thus, we identify miR-31 as a cell-autonomous post-transcriptional regulator of the ISC niche, maintaining proliferative capacity of cycling CBC cells. In addition, we also noticed that miR-31 loss resulted in an increased apoptosis in CBC cells, suggesting the importance of miR-31 in maintaining cell survival. The molecular mechanism by which miR-31 protects against apoptosis warrants future study. The response to high dose of g-IR can be separated into two distinct stages. First, within 24 hours, the majority of CBCs die via apoptosis and subsequent mitotic death, caused by residual misrepaired and unrepaired of DNA double-strand breaks (Hua et al., 2012). Next, between 24 hours and 4 days after g-IR, rare surviving CBCs and quiescent reserve ISCs enter the cell cycle and form regenerative foci that produce mitotically active Lgr5 + cells that repair lost epithelium (Yousefi et al., 2016;Hua et al., 2012). We assume that reserve ISCs also undergo the same process, although lack of direct evidence. In line with this, miR-31 is dramatically reduced within the first 24 hours post g-IR, most likely due to loss of CBCs. Loss of miR-31 led to an marked increase in apoptosis in both CBCs and +4 cells 2 hours post-g-IR. Based on our data, we conclude that during the first stage miR-31 acts as an anti-apoptotic factor, protecting CBCs and reserve ISCs against apoptosis. During the second stage, the surviving stem cells start proliferating to repopulate the depleted intestinal epithelium. The surviving stem cells are relatively damage-resistant (Tian et al., 2011;Takeda et al., 2011;Li et al., 2014;Yousefi et al., 2016;Ritsma et al., 2014), a property attributed to their quiescence, a state likely maintained by BMP/TGFb signaling and inactivation of Wnt signaling (Li et al., 2014;Yousefi et al., 2016;He et al., 2004). We show that miR-31 is prominently induced at the regenerative foci 36 hr post-g-IR and that miR-31 activates Wnt, and represses BMP/ TGFb activities. This points to the potential importance of miR-31 in activating the surviving ISCs. Given BMP/TGFb inhibiting ability of miR-31, we speculate that the homeostatic insensitivity of reserve ISCs to Wnt ligands (Yan et al., 2012) results from their having active BMP and TGFb pathways, that must be suppressed for cells to become competent to respond to Wnt ligands. Our findings suggest that miR-31 functions as an activator of dormant reserve ISCs. We also want to mention that the expression patterns of Bmi1 and Hopx are not specific to +4 position, as both of these transcripts are found non-specifically throughout the crypt base (Li et al., 2014;Muñoz et al., 2012;Itzkovitz et al., 2011). This means that miR-31-activated stem cells represent a complex population including +4 cells, surviving Lgr5 + cells, and those TA cells dedifferentiated in response to irradiation. Taken together, our findings suggest that miR-31 functions as the anti-apoptotic factor in ISCs during the early post-g-IR stage, and, potentially, serves as the cell-intrinsic activator of surviving blotting for STAT3 and p-STAT3 in mICc12 cells treated with STAT3 siRNA. (K) qRT-PCR analysis for miR-31 in cultured organoids treated with IL-6. n = 4 technical replicates. ***p<0.001. (L) Luciferase activity in lysates of mICc12 cells transfected with luciferase reporter plasmids of pGL3-basic empty vector (basic), wild type miR-31 promoter or mutant promoter with mutation of p-STAT3 binding sites. ***p<0.001. (M) Chromatin immunoprecipitation (ChIP) assay carried out on mICc12 cells using antibodies against p-STAT3 and Histone 3. The antibody against Histone 3 was used as a positive control. The enrichment of p-STAT3 binding to miR-31 promoter was quantified using qPCR. ***p<0.001. DOI: https://doi.org/10.7554/eLife.29538.039 The following source data is available for figure 8: Source data 1. Source data for Figure 8. DOI: https://doi.org/10.7554/eLife.29538.040 ISCs regenerative foci promoting regeneration. Future studies will be needed to comprehensively test this idea.
Many reports have showed that miR-31 is overexpressed in CRC tissues (Bandrés et al., 2006;Cottonham et al., 2010;Wang et al., 2009) and increases in progressively during progression from normal to inflammatory bowl disease (IBD) to IBD-related neoplasia (Olaru et al., 2011). We demonstrate that miR-31 promotes tumor development using several models, including cancer cells xenografting, AOM-and DSS-induced inflammation-driven tumors, and Apc-loss driven tumors, characterized by activated Wnt, and repressed BMP/TGFb signalings. Indeed, several reports showed that miR-31 is overexpressed in colorectal cancer (CRC) tissues (Bandrés et al., 2006;Cottonham et al., 2010;Wang et al., 2009). Wnt signaling is aberrantly up-regulated in CRCs, which due primarily to mutations in the Wnt antagonist APC (Novellasdemunt et al., 2015). Our current study suggests that miR-31 up-regulation might also contribute to Wnt activation in CRCs. In addition, decreased BMP and TGFb signaling is also often found in CRCs (Bellam and Pasche, 2010;Hardwick et al., 2008), and can be the consequence of miR-31 upregulation. As such, our data suggests that miR-31 acts as the oncogenic microRNA in CRCs. Moreover, tight association between miR-31 induction and STAT3 pathway activation in intestinal tissues is worth noting. Our molecular data suggest direct activation of miR-31 expression by STAT3 signaling pathway. Indeed, many reports showed that constitutive activation of STAT3 is frequently detected in primary human colorectal carcinoma (Kusaba et al., 2005;Corvinus et al., 2005) and contributes to invasion, survival, and growth of colorectal cancer cells (Tsareva et al., 2007;Lin et al., 2005). Therefore, our current study suggests a signaling pathway involving STAT3, miR-31 and WNT/BMP/TGFb that promotes colorectal tumorigenesis.
In summary, we propose a model in which miR-31 functions as a cell-intrinsic master modulator of the intestinal stem cell niche signaling during normal homeostasis, regeneration and tumorigenesis ( Figure 9). During homeostasis, miR-31 functions to integrate niche signals, supporting canonical Wnt activity and represses BMP/TGFb signaling pathways within cycling CBC stem cells. MiR-31 is stress inducible and plays an important role in epithelial regeneration. In response to high dose of g-IR, miR-31 is markedly induced via STAT3 signaling pathway, and appears capable of regulating the activation state of a population of dormant, radiation resistant reserve ISCs during regeneration. Further, we demonstrate that miR-31 acts as an oncomiR in promoting tumor growth.

Animal experiments
All mouse experiment procedures and protocols were evaluated and authorized by the Regulations of Beijing Laboratory Animal Management and strictly followed the guidelines under the Institutional Animal Care and Use Committee of China Agricultural University (approval number: SKLAB-2011-04-03).

Mouse strains
To generate TRE-miR-31 transgenic mice, the mmu-miR-31 sequence was amplified using the following primers: Forward 5'-CTCGGATCCTGTGCATAACTGCCTTCA-3' (BamHI site was added), and Reverse 5'-CACAAGCTTGAAGTCAGGGCGAGACAGAC-3' (HindIII site was added), and was inserted into pTRE2 vector (Clontech) to generate a pTRE2-miR31 construct. TRE-miR31 transgenic mice were produced using standard protocols and crossed with Rosa26-rtTA mice which harboring the modified reverse tetracycline transactivator (M2rtTA) targeted to and under transcriptional control of the Rosa26 locus. Constitutive miR-31 À/À mice were generated using CRISPR/Cas9 approach at the Nanjing Animal Center, and 402 bp DNA fragment containing miR-31 was deleted to produce the null allele. Conditional miR-31 KO allele was generated at the Shanghai Model Animal Center, the first exon (14806-15522) of miR-31 was targeted with flanking LoxP sites resulting in the 2 LoxP locus. Villin-Cre (Vil-Cre) mice were purchased from the National Resource Center of Model Mice (stock number:T000142). mTmG, Lgr5-eGFP-CreERT, Apc floxed, and Rosa26-LSL-lacZ mice were obtained from Jackson Laboratories (stock number: 007576, 008875, 009045 and 009427). Hopx-CreERT mice were obtained from John Epstein laboratory. Axin2-LacZ mice were obtained from Yi Zeng laboratory.

Cell culture
HCT116, SW480 and LOVO human colorectal cancer cell lines are purchased from American Type Culture Collection (ATCC) and the mouse mICc12 intestinal epithelial cell line was obtained from the Institute of Interdisciplinary Research (Fudan University, Shanghai, China) who originally obtained them from Dr A Vandervalle (Institut National de la Santé et de la Recherche Mé dicale, Faculté X, Paris, France). They were confirmed to come from a mouse cell line by Beijing Microread Genetics Co., Ltd using STR profiling. No cell lines are on the list of commonly misidentified cell lines. We have tested for mycoplasma contamination using a Mycoplasma Detection Kit, and no mycoplasma contamination was detected in any of the cultures. These cell lines were cultured in DMEM/F12 medium. The sequence of miR-31 inhibitor is 5'-AGCUAUGCCAGCAUCUUGCCU-3'. The sequence of Scramble RNA is 5'-CAGUACUUUUGUGUAGUACAA-3'. The Sequence of miR-31 mimics: 5'-AGGCAAGAUGCUGGCAUAGCU-3' 3'-CUAUGCCAGCAUCUUGCCUUU-5' The sequence of negative control for miR-31 mimics: 5'-UUCUCCGAACGUGUCACGUUU-3' 3'-ACGUGACACGUUCGGAGAAUU-5'.

Doxycycline induction and isolation of intestinal epithelium
For the induction, 2 mg/mL Dox (Doxycycline hyclate, Sigma) was added to the drinking water along with 1% w/v sucrose. Mice were induced at 8 weeks of age. To isolate intestinal epithelial cells, mouse intestine was dissected longitudinally and rinsed three times with ice-cold 1x DPBS, then cut into 2-4 mm long pieces, incubated in 1x DPBS containing 2 mM EDTA and 0.2 mM DTT for 30 min at 4˚C on a rotating platform. Suspended cells were then collected folowing gentle vortexing. To isolate intestinal crypts, rinsed small intestine was cut-opened and and villi were scraped using coverslip glass, the technique which left the crypts attached. Crypts were then detached after tissue incubation in 1x DPBS with 2 mM EDTA for 30 min at 4˚C with gentle vortexing. Isolated crypts were counted and pelleted as previously described (Sato et al., 2009).

Flow cytometry
Dissected intestine was incubated with 5 mM EDTA and 1.5 mM DTT in HBSS for 30 min at 4˚C. Single cell suspension was produced following Dispase (BD Biosciences) treatment and passing cells through 40 mm cell strainer. Flow cytometry analysis was performed using BD LSR Fortessa cell analyzer (BD Biosciences). PI-negative cells were selected, then gated for single cells based on the forward-scatter height vs. forward-scatter width (FSC-H vs. FSC-W) and side-scatter height vs. sidescatter width (SSC-H vs. SSC-W) profiles. The size of the nozzle for all sorting runs was 100 mm (20 psi). Lgr5-eGFP + cells were quantified by flow cytometry in TRE-miR31;Lgr5-eGFP-CreERT and M2rtTA;Lgr5-eGFP-CreERT mice after two weeks of Dox treatment. Lgr5-eGFP + cells in miR-31 +/À ; Lgr5-eGFP-CreERT and miR-31 À/À ;Lgr5-eGFP-CreERT mice were quantified using the same method.

Hybridization
For miR-31 in situ hybridizations, digoxigenin (DIG)-labeled probes (Exiqon) were used following the manufacturer's protocol. Both DIG-labeled miR-31 and scrambled probes (Exiqon) were hybridized at 61˚C. U6 probe was used as the positive control. In situ signals were detected by staining with Anti-DIG-AP antibody (Roche) and developed using BM purple substrate (Roche).

Dual luciferase activity assays
To generate reporter constructs for luciferase assays, 300-600 bp fragments in length containing predicted miR-31 target site in the 3'UTRs of Axin1, Dkk1, Bmpr1a, Gsk3b, Smad3 and Smad4 were cloned into the psiCHECK-2 vector (Promega) between the XhoI and NotI sites immediately downstream of the Renilla luciferase gene. To generate reporters with mutant 3'UTRs, nucleotides in the target site complementary to the sequence of the miR-31 seed region sequence were mutated using QuikChange Site-Directed Mutagenesis kit according to the manufacturer's protocol (Stratagene). 293T cells were seeded in 96-well plates one day before transfection. 10 ng of each reporter construct was co-transfected with miR-31 mimics or scramble RNA at a final concentration of 50 nM into 293T cells using Lipofectamine 2000 according to the manufacturer's protocol (Invitrogen). After 24 hr, firefly and renilla luciferase activities were measured with the Dual-Glo luciferase assay system according to the manufacturer's instructions (Promega) and then be calculated using this formula (WT-mimics/WT-mimics NC) /(MUT-mimics/MUT-mimic NC).

Irradiation injury
For irradiation, 2-month-old adult mice were subjected to 12 Gy g-IR and executed at appointed time.

Establishment of the AOM-DSS mouse model
Seven week-old control and miR-31 À/À mice were intraperitoneally injected with AOM (Sigma-Aldrich,) at 10 mg/kg body weight. One week after AOM injection, mice were treated with the socalled DSS cycle, comprised of two steps in which mice were fed with 2.5% (w/v) DSS (molecular weight 36,000-50,000, MP Biomedicals) for 7 days followed by 14 days of normal water feeding. Mice were subjected to a total of three DSS cycles. After treatment, mice were sacrificed and distal colon tissues were collected and tumor number and volume were evaluated.
Luciferase assay for miR-31 promoter activity The transcript of primary miR-31 is located at Chromosome 4, NC_000070.6 (88910557..88910662, complement) in the mouse genome. The upstream 2 kb region of transcript start site (TSS) was identified as the miR-31 promoter in this study, which is located at Chromosome 4, NC_000070.6 (88910663..88912663) and was cloned into the pGL3-Basic reporter constructs. The binding site of STAT3 is located at 88911572-88911582. The binding site 1 and 2 of p65 are located at 88912038-88912048 and 88912409-88912419, respectively.

Chromatin immunoprecipitation (ChIP) assay
ChIP assay was performed according to the manufacturer's protocol with minor modifications, using Simple-ChIP enzymatic chromatin immunoprecipitation kit (Cell Signaling Technology). The sonicated nuclear fractions were divided for input control and for overnight incubated at 4˚C with p-STAT3 or the positive control with H3, negative control with IgG. The recruited genomic DNA from the ChIP assays was quantified by qPCR with primers specific to p-Stat3 binding elements of the miR-31 promoter regions. Primers were as follows: p-STAT3-binding site forward: 5'-TCCAGG-CAAGAAAGTGAGGG À3'; p-STAT3-binding site reverse: 5'-TGAGTAACAGTGCAACAGAGC-3'.
RNA crosslinking, immunoprecipitation, and qRT-PCR (CLIP-PCR) assay CLIP-PCR assay performed as previously described with modification (Wang et al., 2015). Cells were treated with scramble RNA or miR-31 inhibitor, and then harvested after being irradiated at 400 mJ/cm 2 twice. They were then re-suspended in PXL buffer with RNAsin (Promega) and RQ1 DNAse (Promega), and spun at 15000 rpm for 30 min. Supernatant was collected. Protein A Dynabeads (Dynal, 100.02, Thermo Fisher) and goat anti-rabbit IgG (Jackson ImmunoResearch,) or Ago2 antibody were incubated for 4 hr at 4˚C with rotation. The supernatant was added to the beads for 2-4 hr at 4˚C. Beads were then washed twice and digested with Proteinase K (4 mg/ml) for 20 min at 37˚C. RNA was then extracted using Trizol Reagent (Invitrogen) and quantified by qRT-PCR.

Statistical analysis
All analyses were performed in triplicate or greater and the means obtained were used for independent t-tests. Asterisks denote statistical significance (*p<0.05; **p<0.01; ***p<0.001). All data are reported as mean ±SD. Means and standard deviations from at least three independent experiments are presented in all graphs.