Abstract

   Circular RNAs (circRNAs) are a newly discovered type of biological
   molecule that belongs to the noncoding RNA family. Abundant evidence
   has shown that circRNAs are involved in the progression of various
   cancers. However, the particular functions of circRNAs in colorectal
   cancer (CRC) remain elusive. In this study, we investigated the
   differentially expressed circRNAs in three pairs of cancer tissue and
   adjacent normal tissue of CRC. We revealed that circGLIS2 expression
   was higher in CRC tissue and cell lines. Gain-and-loss function assays
   showed that circGLIS2 was involved in the regulation of cell migration.
   Moreover, overexpressing circGLIS2 in CRC cells activated the NF-κB
   pathway and induced pro-inflammatory chemokine production, which evoked
   tumor-associated inflammation through recruiting leukocytes. In turn,
   when the cancer cells were exposed to the supernatant of circGLIS2
   overexpressed cancer cells, they were endowed with the ability of
   migration and chemokines production. Furthermore, the rescue assay
   confirmed that circGLIS2 activated NF-κB signaling and promoted cell
   migration by sponging miR-671. Overall, our study reveals that
   circGLIS2, acting as a potential oncogene, maintains the abnormal
   activation state of the NF-κB signaling pathway via the miR-671 sponge
   mechanism in CRC cells. This study provides a scientific basis for
   targeting circGLIS2 in colorectal cancer interventions.

   Subject terms: Colorectal cancer, Oncogenes

Introduction

   Colorectal cancer (CRC) is the third most common malignancy and the
   fourth most common cause of death in cancers around the world^[42]1.
   With the development of early diagnosis and treatment, the survival
   rate of CRC has increased over the past few years in the United
   States^[43]2. Nevertheless, tumor metastasis is still a major factor
   hindering the therapeutic efficacy of CRC treatments. Therefore,
   studying the underlying molecular mechanism of CRC metastasis is highly
   important to improve CRC treatment.

   Circular RNAs (circRNAs), characterized by their back-spliced ring
   structure, are widely expressed in eukaryotes^[44]3,[45]4. Growing
   evidence has indicated that circRNAs participate in the metastasis of
   many kinds of cancers^[46]5–[47]7. Structurally, circRNAs are more
   stable than other kinds of RNAs, because of their tolerance of
   exonucleases. Functionally, circRNAs act as miRNAs sponges to regulate
   the expression of target genes via the miRNAs response
   element^[48]8,[49]9. This notion has been supported by a series of
   previous studies: circPVT1 functions as an oncogene to promote
   metastasis via miR-145 sponging in CRC^[50]10; circPDE8A promotes the
   invasive growth of pancreatic ductal adenocarcinoma cells by
   antagonizing miR-338 to enhance the MET pathway^[51]11; and fibroblast
   growth factor receptor 1 (FGFR1) derived circular RNA circFGFR1
   promotes non-small cell lung cancer progression and resistance to
   immune checkpoint inhibitor therapy via miR-381-3p/CXCR4 axis^[52]12.

   In this study, to explore the differentially expressed circRNAs in CRC
   development, we analyzed levels of circRNAs in paired CRC tissue and
   adjacent normal tissue by microarray. We identified that circGLIS2 is
   upregulated in CRC tissue and cell lines. We provide evidence that
   circGLIS2 promotes CRC cells migration by activating the NF-κB
   signaling pathway. We further showed that the supernatant collected
   from circGLIS2 overexpressed cells can activate the NF-κB signal
   pathway and promote cell motility. Furthermore, we observed that more
   leukocytes were recruited to the supernatant of circGLIS2 overexpressed
   cells compare to the vector control cells. Mechanically, circGLIS2
   sponge miR-671 to modulate NF-κB signaling pathway and consequently
   enhance the motility of CRC cells. These results indicated that
   circGLIS2/miR-671/NF-κB pathway axis regulated metastasis function via
   an autocrine-paracrine manner in CRC, which may be a promising target
   for CRC intervention.

Results

CircGLIS2 is upregulated in CRC

   CircRNAs plays an important role in many diseases, including cancer. To
   identify the circRNAs that is closely related to CRC, three pairs of
   human CRC clinical samples and adjacent normal tissue (Table [53]S1)
   were analyzed using a circRNA microarray assay. As shown in Fig.
   [54]1a, the volcano plot described 305 circRNAs that were
   differentially expressed in CRC (135 upregulated and 170
   downregulated). And the heatmap showed the top 20 differentially
   expressed circRNAs in CRC (Fig. [55]1b). CircGLIS2 (has_circRNA_101692,
   probe ID: ASCRP002039) was upregulated in the CRC samples and has not
   been studied in cancer yet. The level of circGLIS2 was quantified in
   various CRC cell lines. As shown in Fig. [56]1c, the expression of
   circGLIS2 was increased in CRC cell lines compared to the NCM460 colon
   normal cell line. To verify the research value of circGLIS2 in CRC, a
   CRC-related circRNAs dataset ([57]GSE126094) was attached. As shown in
   Fig. [58]1d, circGLIS2 was significantly upregulated in CRC samples
   compared with adjacent normal control.

Fig. 1. CircGLIS2 is overexpressed in colorectal cancer.

   Fig. 1
   [59]Open in a new tab

   a Volcano plot describing the profile of circRNAs expression in CRC
   tissue and adjacent normal tissue. Significantly changed circRNAs are
   marked with red dots (P < 0.05, fold change > 2.0). b The heatmap for
   top 20 differentially expressed circRNAs in three cases of clinical
   samples, with rows referring to circRNAs and columns referring to the
   samples. c qRT-PCR comparing the expression of circGLIS2 in various
   colorectal cancer cell lines and NCM460 representing normal colon cell
   lines. d Scatter plots illustrating the expression fold change for
   circGLIS2 in CRC tissue compared with adjacent normal tissue from the
   [60]GSE126094 database (n = 10). Data are represented as mean ± SD.
   **P < 0.01.

CircGLIS2 is generated from the GLIS2 gene by back-splicing

   CircGLIS2, upregulated in CRC, is a novel circRNA generated from GLIS2.
   It consists of exons 2 and 3 of the GLIS2 gene. To verify the specific
   back-splicing sites, we designed divergent primers for specific
   back-splicing sites. In addition, convergent primers, which crossed
   exon 2 and 3 of GLIS2, were designed to work as normal control (Fig.
   [61]2a). We performed a set of PCR experiments to validate that
   circGLIS2 harbors a closed-loop structure that is not produced by
   trans-splicing or genomic rearrangement in CRC cells. First, the
   genomic DNA (gDNA) and transcriptomic complement DNA (cDNA) were
   obtained. We applied PCR for cDNA and gDNA using the divergent and
   convergent primers, respectively. The results showed that the
   convergent primers amplified GLIS2 from two kinds of DNA template.
   However, the divergent primers amplified the sequence that only existed
   in the cDNA template (Fig. [62]2b). Furthermore, RNase R treatment was
   used to validate the circular form of circGLIS2 because RNase R is an
   exonuclease that is incapable of cutting circRNAs. As shown in Fig.
   [63]2c, RNase R digestion cleared out most of the linear RNAs such as
   GLIS2 mRNA but circGLIS2 was more resistant. To verify the stability of
   circGLIS2, we used actinomycin D (ACT-d) to inhibit new RNA synthesis.
   As shown in Fig. [64]2d, circGLIS2 was more stable than the linear
   GLIS2 mRNA intracellularly. These data suggest that circGLIS2 is a
   circRNA that is generated from back-splicing of the GLIS2 gene
   transcript.

Fig. 2. CircGLIS2 is generated from exon 2 and 3 of the GLIS2 gene by
back-splicing.

   [65]Fig. 2
   [66]Open in a new tab

   a Schematic shows circGLIS2 is generated from exon 2 and exon 3 of
   GLIS2 by back-splicing. The divergent (red) and convergent (blue)
   primers were designed to identify the back-splicing structure of
   circGLIS2. b PCR analysis was performed to amplify the cDNA and gDNA of
   CRC cells, using the convergent (left) and divergent (right) primer,
   respectively. c Agarose gel electrophoresis (left) and qRT-PCR (right)
   were performed to detect circGLIS2 and linear GLIS2 mRNA after
   treatment with or without RNase R for 2 h. d circGLIS2 and GLIS2 mRNA
   levels were detected by qRT-PCR in DLD1 cells treated with actinomycin
   D (ACT-d) for 24 h. Data are represented as mean ± SD. **P < 0.01.

CircGLIS2 modulates CRC cell migration

   To investigate the role of circGLIS2 in CRC, we cloned the circGLIS2
   sequence into a circRNAs expression vector that consisted of two
   cyclization elements on the back-splicing site (Fig. [67]3a). The
   lentiviruses were generated using a second-generation lentivirus
   packaging system in which 293T cells were co-transfected with psPAX2,
   pMD2.G and the expression vector. DLD1 and HCT-8 were infected with the
   lentivirus and then screened with puromycin for a stably transfected
   cell line. First, we designed specific qPCR primers targeting the
   back-splicing site of circGLIS2 to detect circGLIS2 expression. The
   specificity of the primers was validated using a melting curve analysis
   of the qPCR-amplified product, which all showed a single peak (Fig.
   [68]S1). Next, qPCR assays demonstrated that circGLIS2 was stably
   overexpressed in the DLD1 and HCT-8 CRC cell lines (Fig. [69]3b). Then,
   the circGLIS2 overexpressing cells were subjected to a transwell assay
   to test their cell migration properties. As shown in Fig. [70]3c, d,
   overexpression of cricGLIS2 promoted cell migration of both the DLD1
   and HCT-8 CRC cell lines.

Fig. 3. CircGLIS2 promotes CRC cell migration.

   [71]Fig. 3
   [72]Open in a new tab

   a Schematic of circGLIS2 overexpression plasmid construction. b qRT-PCR
   validated circGLIS2 upregulation in DLD1 and HCT-8 cells transfected
   with circGLIS2 (cGLIS2). c Transwell assay showing the migration
   ability of DLD1 cells that overexpressed circGLIS2 (cGLIS2). d
   Transwell assay showing the migration ability of HCT-8 cells that
   overexpressed circGLIS2 (cGLIS2). e Schematic representation of the
   siRNA that targets the back-splicing junction of circGLIS2. f qRT-PCR
   assay was performed to validate the knockdown efficiency of the
   indicated siRNA in DLD1 and HCT-8. g Transwell assay showing the
   migration ability of DLD1 that were transfected with circGLIS2
   targeting siRNA. h Transwell assay showing the migration ability of
   HCT-8 that were transfected with circGLIS2 targeting siRNA. Scale bars:
   250 μm. Bars represent the number of migrated cells. Data are
   represented as mean ± SD. **P < 0.01.

   We also designed two siRNAs that specifically targeted the
   back-splicing site for a loss-of-function study (Fig. [73]3e). First,
   we verified the known efficiency of the siRNA via RT-qPCR. As shown in
   Fig. [74]3f, both si-1 and si-2 could reduce the expression of
   circGLIS2 in DLD1 and HCT-8 cell lines. Then, as we expected, the
   downregulation of circGLIS2 inhibited cell migration in the DLD1 and
   HCT-8 cell lines (Fig. [75]3g, h). These data indicate that circGLIS2
   is involved in the regulation of CRC cell migration.

NF-κB signaling pathway is involved in circGLIS2 regulated cell migration

   To explore the basic mechanism underlying the effect of circGLIS2 on
   cell migration, we screened the signaling pathways that may participate
   in circGLIS2 functions using the Arraystar Human LncRNA Microarray V4.0
   and KEGG analysis of the cricGLIS2 overexpressed DLD1 cell line. The
   data of the microarray profiling have been deposited in Mendeley Data
   (DOI: 10.17632/fms7h9xwby.1). In contrast to the control cells, 545
   genes were upregulated and 166 genes were downregulated in the
   circGLIS2 overexpressed cells (Fold Change cut-off: 2.0). KEGG pathway
   enrichment analysis indicated that the differently expressed genes
   induced by circGLIS2 overexpression were largely involved in the TNFα
   and the NF-κB signaling pathway (Fig. [76]4a). Accordingly, we
   conducted the luciferase vector pGL4.32 containing five copies of an
   NF-κB response element (NF-κB-RE), which drives expression of the
   luciferase reporter gene luc2P (Photinus pyralis) to detect the
   activity of NF-κB pathway in cells with circGLIS2 overexpression. As
   shown in Fig. [77]4b, NF-κB signal was activated after the transfection
   of overexpressed circGLIS2 in DLD1 cells. Given that p65 plays a
   crucial role in the NF-κB signaling pathway, we also investigated
   whether circGLIS2 could induce p65 activity. Western blots showed that
   the level of phosphorylated p65 (p-p65) had increased following the
   overexpression of circGLIS2 (Fig. [78]4c) and that circGLIS2
   overexpression increased the nuclear localization of p65 (Fig. [79]4d).
   To further confirm the relationship between circGLIS2 and NF-κB
   signaling pathway, p65 targeted siRNA was applied to knockdown the
   level of p65 (Fig. [80]4f). Accordingly, the pro-migration effect was
   eliminated under the p65 knockdown circumstance (Fig. [81]4e). These
   data suggest that circGLIS2 is a positive regulator of the NF-κB
   signaling pathway.

Fig. 4. CircGLIS2 activates the NF-κB signaling pathway in colorectal cancer
cells.

   [82]Fig. 4
   [83]Open in a new tab

   a The top 10 significantly enriched KEGG pathways of the differentially
   expressed genes in DLD1/cGLIS2 cells compared to DLD1/VC cells. The
   y-axis shows the name of the pathway, and the x-axis shows the
   enrichment factor. b Luciferase reporter assays were performed to
   detect NF-κB signaling pathway activity in DLD1/cGLIS2 cells compared
   with DLD1/VC cells. c Whole cell lysis (WCL) western blot assay was
   used to detect the levels of p65 and p-p65 after circGLIS2
   overexpression in DLD1 cells. d Cytoplasm and nuclei protein separation
   checked with a Western blot assay in order to detect the p65 level in
   nuclei. LAMB1 and GAPDH were used as nuclear and cytoplasm markers,
   respectively. e Transwell assay illustrated the migration ability of
   DLD1/VC (cVC) and DLD1/cGLIS2 (cGLIS2) cells after knockdown of p65 via
   siRNA transfection. Scale bars: 250 μm. Bars represent the number of
   migrated cells. Data are represented as mean ± SD. *P < 0.05,
   **P < 0.01. f Western blot assays were performed to validate the
   knockdown efficiency of the siRNA targeted p65.

CircGLIS2 enhances NF-κB signaling pathway and cell motility potential of CRC
cells by an autocrine mechanism

   The above data suggest that circGLIS2 was a potential regulator of
   NF-κB signal pathway, which plays an important role in the tumor
   associated microenvironment (TME). We hypothesize that circGLIS2
   overexpressing cancer cells manipulate cancer development via TME
   construction. To test this hypothesis, we first transfected the vector
   pGL4.32 into 293T cells to detect the effect of cell supernatant on
   NF-κB signal pathway. Supernatants of circGLIS2 overexpression cells
   and the vector control cells were collected as condition medium. We
   found that the activity of luciferase in NF-κB was significantly
   increased by the circGLIS2 overexpression condition medium (Fig.
   [84]5a). Additionally, we also confirmed that the p65 phosphorylation
   level was elevated in CRC cells when the cells were treated with
   circGLIS2 overexpression condition medium (Fig. [85]5b). To further
   investigate the interaction between circGLIS2 associated
   microenvironment and cancer cell motility. We designed an in vitro
   model for measuring cell migration after the condition medium treatment
   (Fig. [86]5c). CRC cells were pretreated with the condition medium for
   24 h and then measure their cell motility by the transwell assay. We
   found that circGLIS2 overexpression condition medium endowed CRC cells
   with greater motility (Fig. [87]5d). Collectively, our data suggested
   that the supernatant of circGLIS2 overexpressing cells was sufficient
   to enable the NF-κB signal pathway activation and cell motility
   promotion.

Fig. 5. CircGLIS2 overexpressed cancer cell-derived supernatant induces NF-κB
signaling and motility activation.

   [88]Fig. 5
   [89]Open in a new tab

   a Luciferase report assay of pGL4.32 transfected 293 T cells co-culture
   with the supernatant of DLD1 cells (cVC or cGLIS2). b Western blot
   assay showing the p-p65 and p65 levels in DLD1 cells co-culture with
   the supernatant of circGLIS2 overexpression (cG) or vector control
   (cVC) CRC cells. c Schematic protocol for the supernatant associated
   transwell assay. d Transwell assay showing the migration ability of
   DLD1 that were co-culture with the supernatant of circGLIS2
   overexpression or vector control CRC cells. Scale bars: 250 μm. Bars
   represent the number of migrated cells. Data are represented as mean ±
   SD. **P < 0.01.

CircGLIS2-overexpressed CRC cells amplify the chemokines milieu to recruit
neutrophils via autocrine and paracrine manners

   To identify what factors contribute to the tumor microenvironment, we
   further analyzed the specific gene set of differentially expressed
   genes caused by overexpression of circGLIS2 in DLD1 cells (Fig.
   [90]6a). It is worth noting that some chemokines (CXCL1, CXCL8, and
   CCL20) were elevated in DLD1-circGLIS2 cells compared to the vector
   control cells, which can be confirmed through knocking-down circGLIS2
   (Fig. [91]6b). We next investigated whether the condition medium from
   circGLIS2 overexpression cells could also induce the expression of
   CXCL1 and CXCL8 in CRC cells. As shown in Fig. [92]6c, when CRC cells
   were exposed in the condition medium of circGLIS2-overexpressing cells,
   the expression of CXCL1 and CXCL8 were significantly increased. This
   data indicated that circGLIS2-overexpressing cells amplified the
   chemokines milieu within the TME via autocrine manner. CXCL1 and CXCL8
   are known chemokines for neutrophils. Thus, we speculated that
   circGLIS2 overexpression cells could recruit leukocyte via the
   construction of tumor associated microenvironment. We designed an in
   vitro chemotaxis assay to evaluate the intercellular role of the
   circGLIS2 associated microenvironment (Fig. [93]6d). We collect the
   leukocytes from healthy donors via the ACK lysis process. The
   leukocytes were added to the upper chamber, while the supernatant of
   cricGLIS2 overexpression cell or vector control cell was add to the
   lower chamber. Surprisingly, the conditioned medium of circGLIS2
   overexpressed cells recruit more cells to the lower chamber (Fig.
   [94]6e). Also, we identified the properties of leukocytes in the upper
   and lower chamber via flow cytometry and we found that the supernatant
   of circGLIS2 overexpressed cell attracted more CD15^+ neutrophils in
   the lower chamber (Fig. [95]S2). Moreover, silencing circGLIS2
   suppressed the chemotaxis of leukocytes toward cell supernatant (Fig.
   [96]6f). These results suggest that cricGLIS2 cells form a unique tumor
   microenvironment enriched with neutrophils via autocrine and paracrine
   manners.

Fig. 6. CircGLIS2 overexpressed cancer cells construct chemokines milieu to
recruit neutrophils from peripheral blood.

   [97]Fig. 6
   [98]Open in a new tab

   a A heatmap showing chemokines expression in circGLIS2 overexpressed
   and vector control DLD1 cells (n = 2). Upregulated chemokines were
   marked in red and downregulated chemokines were in green. b qRT-PCR
   analysis of CXCL1, CXCL8 and CCL20 expression in circGLIS2
   overexpressed (up) or silenced cells (down). c qRT-PCR analysis of
   CXCL1 and CXCL8 expression in DLD1 cells co-cultured with the
   supernatant of circGLIS2 overexpression or vector control CRC cells. d
   Schematic protocol for the in vitro chemotaxis assay of
   circGLIS2-associated supernatant. e Transwell assay analysis of
   chemotactic activity of leukocytes toward the supernatant of DLD1 cells
   (cVC or cGLIS2). f Transwell assay was used evaluated the chemotactic
   activity of leukocytes toward the supernatant of DLD1/cVC and
   DLD1/cGLIS2 cells transfected with si-NC or si-cGLIS2. Data are
   represented as mean ± SD. **P < 0.01.

CircGLIS2 binds to miR-671 to activate the NF-κB signaling pathway

   To further explore the molecular mechanism by which circGLIS2 activates
   the NF-κB signaling pathway, we first determined the subcellular
   location of circGLIS2 in CRC cells. qPCR analysis of the subcellular
   fractionated RNA showed that circGLIS2 predominantly resided in the
   cytoplasm of CRC cells (Fig. [99]7a). FISH experiments also showed the
   same results (Fig. [100]7b). There are many possible mechanisms of
   cytoplasmic circRNAs that have been reported, including sponging miRNAs
   and interacting with other RNA binding proteins. We analyzed the
   protein binding function of circGLIS2 using the CircInteractome
   database^[101]13. The results indicated that circGLIS2 probably
   interacts with AGO2, which is a very common miRNA interaction protein.
   Along this line, a RAP assay showed that an anti-AGO2 antibody did
   enrich circGLIS2 (Fig. [102]7c). Therefore, we speculated that
   circGLIS2 might function by interacting with miRNAs.

Fig. 7. CircGLIS2 binds to miR-671 to modulate the NF-κB signaling pathway.

   [103]Fig. 7
   [104]Open in a new tab

   a Subcellular distribution of circGLIS2 was detected by qRT-PCR, GAPDH
   was used as the cytoplasm-located (Cyto) control and U6 was used as the
   nucleus-located (Neu) control. b FISH illustrated that circGLIS2 was
   predominantly located in the cytoplasm. Oligo DNA probes target
   circGLIS2 was labeled with Cy5 (red) and the nuclei was stained with
   DAPI (blue). Scale bars: 25 μm. c RIP-IP assays were performed to
   co-immunoprecipitation the Ago2 complexes from DLD1 cells. qRT-PCR was
   used to detect the circGLIS2 level and FOX was used as the positive
   control. d A schematic drawing showing the putative binding sites of
   miR-671 on circGLIS2. e The luciferase activity of wild-type circGLIS2
   or mutant circGLIS2 after transfection with miR-671 mimic or miR-NC. f
   NF-κB signaling pathway activity was assessed in cells transfected with
   miR-671 or miR-NC. g Western blot assay showing the p-p65 and p65
   levels in DLD1/cVC and DLD1/cGLIS2 cells transfected with different
   amounts of miR-671 or miR-NC. Data are represented as mean ± SD.
   **P < 0.01.

   To validate this hypothesis, the CircInteractome bioinformatics
   databases were used to predict miRNAs that may bind to
   circGLIS2^[105]13. Based on the number of pairing sequence and
   TargetScan miRNA perdition score, three miRNAs were selected for
   transwell assay validation. Among the candidate miRNAs, miR-671
   significantly inhibited cell migration in DLD1 cells (Fig. [106]S4). To
   detect miR-671 expression in CRC tissue, we referred to TCGA dataset
   and revealed that miR-671 was downregulated in colon and rectum
   adenocarcinoma (Fig. [107]S3). Also, miR-671 has been reported to
   regulate cancer progression and immunomodulation^[108]14. Thus, we
   hypothesized that circGLIS2 might regulate the NF-κB signaling pathway
   by competitively binding to miR-671 (Fig. [109]7d). We used a
   dual-luciferase reporter system to validate the interaction between
   circGLIS2 and miR-671. First, we generated wild-type and mutated
   vectors of circGLIS2 validated by Sanger sequencing. Then, the
   wild-type or mutated circGLIS2 vectors were co-transfected with miR-671
   into the DLD1 cell line. The dual-luciferase reporter assays indicated
   that the relative luciferase activity had decreased after
   co-transfection with circGLIS2-WT and miR-671 (Fig. [110]7e). We
   evaluated the activity of the NF-κB pathway in cells after transfection
   with miR-671 and found that miR-671 suppressed NF-κB signaling in DLD1
   cells (Fig. [111]7f). To investigate whether circGLIS2 exerts its
   function via miR-671, rescue experiments were performed with circGLIS2
   ectopic expressing cells with the presence of miR-671 dosage
   transfection. MiR-671 reversed the ability of circGLIS2 to up-regulate
   p-p65 in CRC cells (Fig. [112]7g). Also, miR-671 reversed the ability
   of circGLIS2 to promote the migration of DLD1 and HCT-8 cells (Fig.
   [113]8a, b). However, we did not observe a significant reversal effect
   in the cell chemotaxis experiments (Fig. [114]8c). These findings
   demonstrated that circGLIS2 could bind to miR-671 to increase the
   activity of the NF-κB signaling pathway and facilitate its own
   migration.

Fig. 8. MiR-671 attenuates the function of circGLIS2 in cell migration but
not affecting chemotaxis potential.

   Fig. 8
   [115]Open in a new tab

   a Cell migration activity was assessed for DLD1/VC and DLD1/cGLIS2
   cells that were transfected with different amounts of miR-671 or
   miR-NC. b Cell migration activity was assessed for HCT-8/VC and
   HCT-8/cGLIS2 cells that were transfected with different amounts of
   miR-671 or miR-NC. Scale bars: 250 μm. c Transwell assay was used to
   evaluate the chemotactic activity of leukocytes toward the supernatant
   of DLD1/cVC and DLD1/cGLIS2 cells transfected with miR-NC or miR-671.
   Data are represented as mean ± SD. *P < 0.05, **P < 0.01.

Discussion

   circRNAs have long been recognized as a byproduct of gene expression in
   mammalian cells^[116]15. Recently, with the development of RNA-seq and
   bioinformatics technologies, many circRNAs have been found to have
   specific roles in various diseases, including cancer^[117]16. It has
   been reported that, circMAN2B2, circPRKCI, circPIP5K1A, and circPVT1
   can promote tumor growth and invasion in lung cancer^[118]17–[119]20.
   In addition, circBANP can increase the metastatic phenotype in
   CRC^[120]21, and circ-102004 can promote cancer cell proliferation in
   prostate cancer^[121]22. Furthermore, circ-0013958 might be used as a
   potential biomarker in liquid biopsies for lung adenocarcinoma^[122]23.
   CircPAN3 is enriched in refractory acute myeloid leukemia and might
   serve as a chemo-sensitive predictor for the patients^[123]24. Thus, it
   is essential to analyze the differentially expressed circRNAs in CRC
   and elucidate its basic mechanism in cancer progression.

   Here, by combining circRNAs array analysis with GEO database
   verification, we found that circGLIS2 was upregulated in CRC. CircGLIS2
   is located on the plus strand of chromosome 16p13.3 and is derived from
   the known protein-coding gene GLIS2. As the host gene of circGLIS2,
   GLIS2 plays the role of a transcription factor and may act as an
   oncogene or tumor suppressor depending on the cell circumstances.
   Previous reports showed that GLIS2 is usually involved in the
   development of cancer as fusion genes. For example, GLIS2 fused with
   CBFA2T3 directly contributes to leukemogenesis and presents an
   aggressive phenotype^[124]25. ETO-GLIS2 fusion gene oligomerization
   promotes megakaryocytic differentiation^[125]26. These facts suggested
   that GLIS2 has various gene structures and functions in the genome.
   However, the circRNAs derived from GLIS2 as well as their function in
   cancer progression are still poorly understood. Here, the gain-and-loss
   function assays showed that consistent expression or silencing of the
   expression of circGLIS2 in CRC cell lines significantly promoted or
   inhibited cell migration, respectively. These data suggest that
   circGLIS2 is involved in the progression of CRC via modulating the
   migration phenotype of the cancer cells. Similar to other noncoding
   RNAs, circRNAs are involved in cellular gene regulation through
   post-transcriptional modification^[126]27,[127]28. Emerging research
   has indicated that circRNAs function as “miRNAs sponges” to
   competitively bind to miRNAs^[128]4. Subsequently, our further data
   showed that circGLIS2 settled in the cytoplasm and was incorporated
   into the AGO-RISC complex in CRC cells. Our study focused on the
   interaction between circGLIS2 and miR-671 in CRC. We found that miR-671
   inhibited the NF-κB signaling pathway in CRC cells. Also, miR-671
   attenuated NF-κB signaling activity that was elicited by circGLIS2
   overexpression. In the rescue experiments, we observed a gradual
   inhibition of cell motility with miR-671 transfection in DLD1 and
   HCT-8. Moreover, we revealed that miR-671 was significantly
   downregulated in CRC tissue compared to the normal tissue. It has been
   known that constitutive activation of NF-κB signaling pathway plays a
   role in the progression of cancers^[129]29. Previous reports showed
   that abnormally activated NF-κB signaling was found in 40% of CRC
   patients. TRAF6 activates NF-κB via the nuclear shuttling effect of p65
   in CRC^[130]30. RNA helicase p68 enhanced NF-κB target genes expression
   by directly promoting p65 transcription^[131]31. Our research
   introduced a new mechanism that helps understand the NF-κB pathway in
   CRC. We found that circGLIS2 activated NF-κB pathway through
   competitively binding to miR-671. However, we also noted that circGLIS2
   could interact with other miRNAs including miR-512, miR-874 and
   miR-940, since these miRNAs also have targeting sequences and high
   scores in bioinformatics prediction. We also cannot rule out whether
   circGLIS2 would function by binding to proteins or translating
   peptides. The involvement of miR-671 clarified the role of circGLIS2 in
   NF-κB and cell migration, but miR-671 does not affect the leukocyte
   recruitment. This question awaits further investigation.

   Our results showed that more neutrophils were recruited when the
   leukocytes were exposed to the supernatant of circGLIS2 overexpressed
   cells. Recent studies have highlighted an important role of neutrophils
   in manipulating cancer metastasis^[132]32. In primary cancers,
   infiltrating neutrophils promote cancer cell dissociation by secreting
   TGF-β and MMP9^[133]33,[134]34. In the circulating cancer cells (CTC),
   neutrophils physically bind with circulating cancer cells to promote
   intravasation of CTC^[135]35. Notably, an increase in the ratio of
   neutrophil-to-lymphocyte has been observed in many kinds of cancers and
   has become a poor survival indicator in cancer patients^[136]36. This
   indicated that circGLIS2 might induce a pro-metastasis microenvironment
   to promote metastasis of cancer cells.

   In summary, all of the data indicate that the upregulated circGLIS2
   contributes to pro-metastasis microenvironment via activation of NF-κB
   signaling pathway and recruitment of leukocytes in CRC. Mechanically,
   circGLIS2 acts as a “miRNAs sponge” to inhibit miR-671 function and
   subsequently manipulate the NF-κB signaling axis. These findings
   highlight a new mechanistic connection between NF-κB and circGLIS2 in
   the construction of a pro-metastasis microenvironment of CRC. These
   results may provide new insights and lead to new therapeutic strategies
   for CRC prevention and treatment.

Materials and methods

Tissue samples and ethics statement

   Three pairs of CRC tissues and matched adjacent noncancerous colorectal
   tissues were collected from surgical resections at the Sixth Affiliated
   Hospital, Sun Yat-sen University (Guangdong, China). None of the
   tissues were exposed to chemo- or radio-therapies prior to surgery. The
   tissues were frozen and stored in liquid nitrogen until use. Whole
   blood samples were preserved at 4 °C after collection from healthy
   donor. The leukocyte were purified by ACK red blood cells depletion kit
   (Leagene, Beijing, China), according to the manufacturer’s manual. The
   protocol and the use of clinical materials for research purposes were
   approved by the Ethics Committee of the Sixth Affiliated Hospital of
   Sun Yat-sen University (ethical approval number 2018ZSLYEC-008).

Cell culture and transfection

   The normal human colon epithelial cell line NCM460 was maintained in
   our lab and the CRC cell lines DLD1, HCT-8, HCT116, RKO, HT-29 and
   HCT-15 were obtained from the American Type Culture Collection (ATCC,
   Virginia, USA). The NCM460 cell line was cultured in DMEM (Gibco, New
   York, USA) supplemented with 10% fetal bovine serum (FBS, Gibco, New
   York, USA) and 100 U/mL penicillin-streptomycin (Gibco, New York, USA).
   The other CRC cell lines (DLD1, HCT-8, HCT116, RKO, HT-29, and HCT-15)
   and their corresponding construction cell lines were cultured in
   RPMI1640 (Gibco, New York, USA) supplemented with 10% FBS and 100 U/mL
   penicillin-streptomycin. All cells were cultured in a humidified
   atmosphere with 5% CO[2] at 37 °C. Cells were passaged every 2–3 days
   to maintain cell viability.

Microarray analysis

   For circRNA expression profiling, Arraystar Human circRNA Array v2
   analysis was performed by Kangcheng Biotech (Shanghai, China) according
   to Arraystar standard protocols. Briefly, total RNAs were extracted
   from the paired tissue samples of three CRC patients by TRIzol reagent
   (Invitrogen, Oregon, USA). Then the RNAs were digested with RNase R
   (Epicentre, Wisconsin, USA) to remove the linear RNA population. The
   remained RNAs were amplified and transcribed into fluorescent cRNA and
   then hybridized onto the Arraystar Human circRNA Array (8 × 15 K,
   Arraystar). After washing the slides, the arrays were scanned by the
   Agilent Scanner G2505C. Agilent Feature Extraction software was used to
   analyze acquired array images. Quantile normalization and subsequent
   data processing were performed using the R software package. Volcano
   Plot filtering was used to identify the statistical significance of
   differentially expressed circRNAs between two groups.

   For mRNA expression profiling, Arraystar Human LncRNA Microarray V4.0
   was used to analyze protein-coding transcripts by Kangcheng Biotech
   (Shanghai, China). Briefly, mRNA was purified from total RNA after
   removal of rRNA (mRNA-ONLY™ Eukaryotic mRNA Isolation Kit, Epicentre).
   Then, the RNA sample was transcribed into fluorescent cRNA (Arraystar
   Flash RNA Labeling Kit, Arraystar). The labeled cRNAs were purified by
   RNeasy Mini Kit (Qiagen) and then hybridized onto the LncRNA expression
   microarray slide. The hybridized arrays were washed, fixed and scanned
   by Agilent DNA Microarray Scanner G2505C. Agilent Feature Extraction
   software (version 11.0.1.1) was used to analyze acquired array images.
   Quantile normalization and subsequent data processing were performed
   using the GeneSpring GX v12.1 software package (Agilent Technologies).
   Differentially expressed mRNAs were identified through P-value/FDR and
   Fold Change filtering.

RNA extraction and quantitative real-time polymerase chain reaction (qRT-PCR)

   Total RNA was extracted using TRIzol reagent (Invitrogen, Oregon, USA),
   reverse transcription was performed using PrimeScript RT Master Mix
   (Takara Biomedical Technology, Beijing, China), and quantitative
   real-time PRC was performed using SYBR Green Premix Ex Taq II (Takara
   Biomedical Technology, Beijing, China) according to the manufacturer’s
   instructions. The results of the transcript levels were analyzed by the
   2-delta-delta Ct method. All quantitative real-time PCR primers were
   synthesized in Sango Biotech (Shanghai, China) and the primers sequence
   shown as follows: circGLIS2-forward, CAGCAGCTCGCTGTCCCCCGAGCG;
   circGLIS2-reverse, GTTGGAGGTGGCAGCAGGCAGTGG. GAPDH-forward,
   GCACCGTCAAGGCTGAGAAC; GAPDH-reverse, TGGTGAAGACGCCAGTGGA. U6-forward,
   CTCGCTTCGGCAGCACA; U6-reverse, AACGCTTCACGAATTTGCGT. And the divergent
   and convergent primer set for circGLIS2 PCR identification were shown
   as below: divergent primer pair, CTTCCTGCTGAACTCCAAGT and
   GCTTGGTGATACTCAGCTTC; convergent primer pair, TTCCTGCGTGAAGACCAGCT and
   GAACTTGGAGTTCAGCAGGA.

Cell migration assays

   The cell migration activity was assessed using 24-well transwell insert
   chamber plates (Corning, Massachusetts, USA). Briefly, Cells were
   harvested then suspended in serum-free 1640 medium. Then cells were
   transferred in to the upper compartment of the chamber at 0.2 million
   in 200 μL. A total of 600 μL of complete 1640 medium supplemented with
   10% FBS was added to the lower chamber. The cells were then incubated
   for 24 h. The cells that did not migrate were scraped off with a cotton
   swab and the cells were stained with a crystal violet solution
   (Beyotime, Jiangsu, China). The cells were then counted and analyzed.
   These experiments were repeated three times.

Western blot

   Cells were harvested and lysed with RIPA buffer (Beyotime, Jiangsu,
   China) followed by BCA analysis to determine the protein concentration.
   The samples were separated by SDS-PAGE and transferred onto
   nitrocellulose membranes. After incubation with the primary antibodies
   anti-p65 (1:1000; AF0246, Beyotime, Jiangsu, China), anti-phosphor p65
   (1:1000; AN371, Beyotime, Jiangsu, China), anti-GAPDH (1:10000;
   10494–1-AP, Proteintech Group, Illinois, USA), anti-LAMB1 (1:5000;
   23498–1-AP, Proteintech Group, Illinois, USA), the membranes were then
   incubated with a secondary antibody (1:5000; #7074, Cell Signaling
   Technology, Massachusetts, USA). After washing, signals were detected
   using a chemiluminescence system (Bio-Rad, California, USA) and
   analyzed using Image Lab Software (Bio-Rad, California, USA).

Chemotaxis assay

   Leukocytes were isolated from whole blood using ACK red blood cells
   depletion kit (Leagene, Beijing, China), according to the
   manufacturer’s protocol. Purified leukocytes were counted and 2 × 10^5
   cells were added to the upper chamber. The supernatant of DLD1/VC or
   DLD1/circGLIS2 cells was added to the lower chamber to test the
   chemotaxis effect. After 2 h of incubation, the number of leukocytes
   translocated to the lower chamber were counted using The Bio-Rad TC20
   automated cell counter (Bio-Rad, California, USA). Subsequently,
   leukocytes obtained from the lower chamber were labeled with FITC
   conjugated CD3 (Biolegend, California, USA) and APC conjugated CD15
   (Biolegend, California, USA) antibodies to test the identity of the
   leukocytes recruited by the supernatant of CRC cells. The samples were
   analyzed on FACSCanto Flow Cytometry (BD Biosciences, California, USA)
   according to the user’s manual.

RISC Ago2 immunoprecipitation

   RISC immunoprecipitation assays were performed with an Ago2-RIP Kit
   (Millipore, Massachusetts, USA) as directed by the manufacturer.
   Briefly, DLD1-circGLIS2 cells were cross-linked through 16%
   formaldehyde treatment and lysed in co-IP lysis buffer containing
   proteinase inhibitors and RNase inhibitor. The lysate was incubated
   with anti-Ago2 antibody overnight. Agarose beads were added to the
   co-IP sample to precipitate the RISC-Ago2 complex. The beads were
   washed with co-IP lysis buffer six times. The RNA was purified using
   TRIzol reagent (Invitrogen, Oregon, USA) according to the
   manufacturer’s instructions.

RNase R treatment

   Total RNA was extracted using TRIzol reagent according to the
   manufacturer’s instructions. The RNA was incubated with 20 U RNase R at
   37 °C for 2 h in the buffer provided with the kit (Epicentre,
   Wisconsin, USA).

Luciferase reporter assay

   Cells were co-transfected with plasmids containing the targeted
   sequences of the wild or mutant fragment from the circGLIS2 and miRNAs
   mimics using Lipofectamine3000 (Invitrogen, Oregon, USA) according to
   the manufacturer’s instructions. Then, 24 h after transfection, firefly
   luciferase activities were measured by using a dual-luciferase reporter
   assay kit (Promega, Massachusetts, USA), with renilla activity used as
   the internal control. Finally, the ratios of firefly to renilla
   luminescence represent the binding activity of circGLIS2 and miRNAs.

   For the NF-κB signaling pathway report system, cells were transfected
   with the pGL4.32/NF-κB reporter plasmid (Promega, Massachusetts, USA)
   for 24 h, and firefly luciferase activity was then measured by using a
   luciferase reporter assay kit (Promega, Massachusetts, USA). The
   luminescence was normalized against the protein concentration.

Fluorescence in situ hybridization (FISH)

   Specific DNA probes to the circGLIS2 back-splicing site were applied
   for in situ hybridization. In brief, cells were seeded onto a coverslip
   following by fixation with 25% formamide (Sigma-Aldrich, Missouri,
   USA). The protein fraction was removed by 20 µg/mL proteinase K
   (Sigma-Aldrich, Missouri, USA) treatment for 10 min. The sample was
   then dehydrated in a gradient of ethanol solutions. Before the
   hybridization, the Cy5-labeled probes (Sangon Biotech, Shanghai, China)
   were diluted in hybridization buffer at 10 ng/mL. Following a 95 °C
   denaturing process, they were chilled on ice. The sample was covered
   with the denatured probes and incubated in a 60 °C incubator overnight.
   The coverslip was gently washed with 2× SSC buffer (Leagene, Beijing,
   China) 5 times. The nuclei were stained with Hoechst 33342 (Invitrogen,
   Oregon, USA). The fluorescent images were acquired on a Leica TCS SP8
   DLS confocal microscope system (Leica, Hesse, Germany). The Cy5signal
   excited in response to the 649 nm laser and had an emission wave at
   666 nm. The Hoechst 33342 signal was excited by a 346 nm laser and had
   an emission wave at 460 nm. FISH probes are as follows: FISH-circGLIS2
   Cy5-TGCCTCCAGTGCCCAGTGCCTCGGTTCCTGCGTGAAGACCAGCTGG; FISH-NC Cy5-
   TGGTGTCCCCTGGACGTGGCTACACGTAGCACTCGGACGTGGCTA.

Statistical analysis

   All statistical analyses were carried out using GraphPad Prism
   (GraphPad Software, California, USA). Quantitative values are expressed
   as means ± standard deviation (SD) of at least three independent
   repetitions. The difference between means was evaluated by unpaired
   two-tailed Student’s t test. A probability value of P < 0.05 was used
   as the criterion for statistical significance.

Supplementary information

   [137]Supplementary figures legends^ (14.4KB, docx)
   [138]Table S1^ (12.6KB, docx)
   [139]figure S1^ (2.4MB, jpg)
   [140]figure S2^ (707.1KB, jpg)
   [141]figure S3^ (497.5KB, jpg)
   [142]figure S4^ (976.6KB, jpg)

Acknowledgements