circPPFIA1s were downregulated in liver metastatic colon cancer cells
To search for metastasis-associated circRNAs, a previously established cell line model was used: primary colon cancer KM12C cells and its liver metastatic derivative KM12L4 cells (Supplementary Figure S1A) (21). The high metastatic potential of KM12L4 cells was verified by comparing their invasive and migratory abilities with those of parental KM12C cells. Transwell invasion and migration assays revealed that KM12L4 cells showed a higher number of invaded and migrated cells compared to KM12C cells, indicating that KM12L4 cells have a higher metastatic potential than KM12C cells (Figure 1A). To examine the degree of liver metastasis in vivo, KM12C and KM12L4 cells were injected into the spleen and the degree of liver metastasis was determined by visual counting and MRI. As expected, more liver metastases were found in mice injected with KM12L4 cells than in those injected with KM12C cells (Figure 1B; Supplementary Figure S1B).
A circRNA microarray was conducted to identify differentially expressed circRNAs between KM12C and KM12L4 cells. Twenty-nine circRNAs were differentially expressed more than two-fold. Nine circRNAs showed decreased expression in KM12L4 cells compared to KM12C cells. In contrast, the expression of 20 circRNAs was increased (Supplementary Figure S2A, S2B). Among them, hsa_circRNA_100873 (hsa_circ_0003429) showed the most significant decrease in expression. hsa_circRNA_100873 is an exonic circRNA generated from five exons (exons 17–21) of PTPRF interacting protein alpha 1 (PPFIA1) (Figure 1E). Interestingly, another PPFIA1-originated circRNA, hsa_circRNA_100872 (hsa_circ_0000337), was found in the list of differentially expressed circRNA (Figure 1C, 1D). The spliced length of hsa_circRNA_100872 generated from three exons (exons 17–19, 419 bp) is shorter than that of hsa_circRNA_100873 (702 bp). Hence, we named them circPPFIA1-L (long) and circPPFIA1-S (short), respectively (Figure 1E; Supplementary Figure S2B). According to the circBase database (www.circbase.org), 37 circRNAs are possibly generated from PPFIA1 (Supplementary Figure S3). However, there are very few circRNAs generated from PPFIA1 whose action mechanisms and roles have been identified.
To validate the microarray data, the expression levels of circPPFIA1-L and -S were determined by RT-qPCR and semi-qPCR (Supplementary Figure S4A, S4C). RT-qPCR analyses using specific primers recognizing their divergent region showed a considerable decrease in both circPPFIA1-L and -S in KM12L4 cells (Figure 1F; Supplementary Figure S4B). However, linear PPFIA1 mRNA levels were almost same. Similarly, semi-qPCR analyses revealed that KM12L4 expressed less of both circRNAs compared to KM12C cells without a significant change in linear mRNA levels (Figure 1G). Similar to the results in the model cell lines, the expression of circPPFIA1 in the tissues of colon cancer patients showed a decrease compared to normal tissues, although significant results were not obtained due to the small sample size (Supplementary Figure S5).
circPPFIA1-L and -S are highly stable circularized RNAs
The stability of circPPFIA1s was assessed by semi-qPCR or RT-qPCR after RNase R and actinomycin D treatment. Whereas RNase R degraded linear PPFIA1 mRNA, circPPFIA1-L and -S were highly resistant to RNase R (Figure 2A). Additionally, linear PPFIA1 mRNA was almost degraded at 24 h post-treatment with actinomycin D. However, neither circPPFIA1s was degraded (Figure 2B; Supplementary Figure S6A). These results indicate that circPPFIA1-L and -S are highly stable, which is a typical property of circRNAs.
To verify the junction sequence of circPPFIA1-L and -S, genomic DNA (gDNA) and cDNA were used for the PCR analysis with convergent and divergent primers. Whereas PCR products of the convergent primers were observed for gDNA and cDNA templates, the divergent primers generated PCR products only from cDNA (Figure 2C for circPPFIA1-L, 2E for circPPFIA1-S). In cDNA and gDNA, GAPDH was amplified only by the convergent primer (Supplementary Figure S6B). The back-spliced junction was amplified and verified by Sanger sequencing (Supplementary Figure S4D). We observed head-to-tail splicing between exons 17 and 21 in circPPFIA1-L and exons 17 and 19 in circPPFIA1-S, indicating that they have a circularized structure (Figure 2D, 2F; Supplementary Figure S6C).
To assess the localization of circRNA, a cellular fractionation assay was conducted. The level of α-tubulin and lamin B was determined to verify appropriate fractionation. Both circPPFIA1-L and -S were abundantly expressed in the cytosol (Figure 2G), which suggests that circPPFIA1s could function as molecular sponges of miRNA or RBP.
circPPFIA1s negatively regulated metastatic potential and liver metastasis of CRC
To investigate whether the knockdown of circPPFIA1-L and -S regulated the metastatic potential of KM12C cells, we designed siRNAs targeting the divergent junctions of circPPFIA1-L and -S (Supplementary Figure S7A for circPPFIA1-L and S7C for circPPFIA1-S). All designed siRNAs showed an efficient decrease in corresponding circRNAs, but barely influenced the expression of linear PPFIA1 (Figure 3A, 3C; Supplementary Figure S7B, S7D). An increase in invasive and migratory abilities was observed in circPPFIA1-L- and circPPFIA1-S-silenced KM12C cells (Figure 3B, 3D). The increased metastatic potential was observed with each siRNA. However, the knockdown of PPFIA1 mRNA did not influence the metastatic potential of KM12C cells (Supplementary Figure S8A, S8B). To exclude the possibility that the increased number of invaded and migrated cells observed after circPPFIA1s knockdown is attributed to increased cell growth, we examined the proliferation rate of circPPFIA1s-silenced KM12C cells. Neither circPPFIA1-L nor circPFIA1-S affected cell growth (Supplementary Figure S8C), demonstrating that circPPFIA1s inhibit the metastatic potential of KM12C cells without affecting cell growth. Notably, metastatic properties, including invasion and migration, were further increased when both circPPFIA1-L and -S were simultaneously silenced (Figure 3E, 3F; Supplementary Figure S7E).
To investigate whether knockdown of circPPFIA1-L and -S increased the liver metastasis of CRC in vivo, a splenic injection mouse model was used. An approximate four-fold increase in liver metastasis was observed in the mice injected with circPPFIA1-L-silenced KM12C cells (Figure 3G; Supplementary Figure S9A, S9B). The intrasplenic injection of circPPFIA1-S-silenced KM12C cells showed a more than four-fold increase in liver metastasis (Figure 3H; Supplementary Figure S9C, S9D). Based on these results, we confirmed that the knockdown of both circPPFIA1-L and -S potentiates metastatic potential and enhances the liver metastasis of CRC.
To examine whether the circPPFIA1s suppressed the metastatic potential of KM12L4 cells, we constructed overexpression vectors expressing each circRNA. Both vectors induced a significant increase in circPPFIA1-L and -S without any change in the linear PPFIA1 mRNA (Figure 4A, 4C). Increased expression of the circPPFIA1s resulted in the inhibition of the invasive and migratory properties of KM12L4 cells (Figure 4B, 4D). Overexpression of circPPFIA1-L dose-dependently suppressed the invasive ability (Supplementary Figure S10A), and similar results were obtained in all overexpressing cells (Supplementary Figure S10B–10D). We also confirmed that the reduction in metastatic abilities did not result from the inhibition of cell growth (Supplementary Figure S10E). The intrasplenic injection experiments revealed that KM12L4 cells with high levels of circPPFIA1s showed a decrease in liver metastasis (Figure 4E). The incidence of liver metastasis and the number of nodules were decreased in mice injected with circPPFIA1-overexpressing KM12L4 cells (Supplementary Figure S11).
To confirm that the inhibitory effects of circPPFIA1s on metastatic properties can be applied to other colon cancer cells, DLD1 and RKO colon cancer cells were used. Similar to the results in KM12C cells, the knockdown of circPPFIA1-L or -S caused an increase in metastatic abilities (Supplementary Figure S12A, S12B). Conversely, the increased expression of either circPPFIA1-L or -S diminished the number of invaded and migrated cells (Supplementary Figure S12 C, S12D). Thus, we demonstrate that circPPFIA1-L and -S negatively regulates liver metastasis in CRC.
miR-155-5p was identified as a sponging target of circPPFIA1s
Increasing evidence suggests that circRNAs act as miRNA sponges, thereby interrupting the inhibitory functions of miRNA. The cellular fractionation assays revealed that circPPFIA1-L and -S were abundantly located in the cytosol (Figure 2G), suggesting that they might function as competing endogenous RNAs (ceRNAs). Four bioinformatic prediction algorithms (ArrayStar, https://www.arraystar.com; circInteractome, https://circinteractome.nia.nih.gov; Starbase, http://starbase.sysu.edu.cn; and RNA22, https://cm.jefferson.edu/rna22) were used to search for circPPFIA1-interacting miRNAs. The only common prediction in all algorithms was miR-155-5p (Figure 5A; Supplementary Figure S13).
To verify that miR-155-5p interacts with circPPFIA1, an ASO pulldown experiment was performed. First, we designed ASOs targeting the divergent sequences of circPPFIA1-L or -S. All designed ASOs for circPPFIA1-L and -S worked efficiently and miR-155-5p bound to circPPFIA1s (Supplementary Figure S14A, S14B). Repeated ASO pulldown experiments were conducted by the mixture of corresponding ASOs, and the level of miR-155-5p in pulldown materials was determined by RT-qPCR.
To confirm the interaction between cirPPFIA1s and miR-155-5p, an Argonaute 2 immunoprecipitation (Ago2 IP) experiment was performed. The introduction of pre-miR-155-5p into KM12C cells increased the enrichment of circPPFIA1-L and -S in Ago2 IP material compared to the control IgG (Figure 5C). Interestingly, the RT-qPCR results indicated that circPPFIA1-S was more enriched than circPPFIA1-L, assumingly due to the higher cytosolic levels of circPPFIA1-S. In addition, the direct interaction between circRNAs and miR-155-5p was examined by a luciferase assay. Two miR-155-5p MREs were found in exon 18, which is present in both circRNAs (details in Supplementary Figure S15A), and therefore, we constructed two luciferase vectors containing the wild-type or mutated sequence of miR-155-5p MRE (Supplementary Figure S15B). Luciferase activity was inhibited by overexpression of miR-155-5p in both vectors containing wild-type MRE. However, the luciferase expression was not affected in the case of mutated vectors (Figure 5D).
Although we confirmed that circPPFIA1s and miR-155-5p were bound, the knockdown of circPPFIA1-L, -S, or both did not affect the level of miR-155-5p (Figure 5E). Overexpression of miR-155-5p by introducing pre-miRNA into KM12C cells did not influence the expression of circPPFIA1s (Figure 5F). Similarly, the downregulation of miR-155-5p by anti-miRNA in KM12L4 cells did not result in the reduction of circPPFIA1s (Figure 5F). These results indicate that the circPPFIA1s and miR-155-5p did not affect each other’s expression. The effect of changes in ceRNA on the level of sponging miRNAs has not been fully elucidated. Due to their structural characteristics, circRNAs are not thought to be affected by their sponging miRNAs.
Next, we examined the regulation of metastatic potential by miR-155-5p. The overexpression of miR-155-5p in KM12C cells caused an increase in the number of invaded and migrated cells compared to the control miRNA (Figure 5G). In contrast, the inhibition of miR-155-5p suppressed the metastatic potential of KM12L4 cells (Figure 5H). The regulatory effect of miR-155-5p was confirmed in DLD1 and RKO cells. As observed in KM12C and KM12L4 cells, the metastatic potential was increased depending on the expression level of miR-155-5p (Supplementary Figure S16A for DLD1 and S16A for RKO). To verify the role of miR-155-5p in circPPFIA1s-mediated regulation of metastatic potential, a rescue experiment was conducted using a mixture of siRNAs targeting circPPFIA1-L and -S. As expected, the metastatic potential of KM12C cells was potentiated by the knockdown of circPPFIA1s. However, the inhibition of miR-155-5p reversed the increase in the invasive and migratory abilities of KM12C cells, indicating that an increase in liberated miR-155-5p is responsible for the function of circPPFIA1s (Figure 5I). Rescue experiments using each circPPFIA1-L and -S siRNA also showed similar results (Supplementary Figure S17).
CDX1 is responsible for the function of circPPFIA1/miR-155-5p
By screening targets of circPPFIA1/miR-155-5p using prediction algorithms, six genes were identified (Supplementary Figure S18). CDX1, a tumor-suppressor, was selected by RT-qPCR validation and reference search for further studies (Figure 6A). Western blot and RT-qPCR assays revealed that CDX1 was highly expressed in KM12C cells compared to KM12L4 cells (Figure 6B). The effect of miR-155-5p on the expression of CDX1 was tested using pre- and anti-miR-155-5p in KM12C and KM12L4 cells, respectively. The overexpression of miR-155-5p decreased CDX1 protein and mRNA expression levels in KM12C cells. Conversely, CDX1 was upregulated by decreasing miR-155-5p (Figure 6C) in KM12L4 cells. Direct interaction between CDX1 mRNA and miR-155-5p was assessed by Ago2 RIP and luciferase experiments. The enrichment of CDX1 mRNA in Ago2 IP material was enhanced by the overexpression of miR-155-5p and was lowered by the inhibition of miR-155-5p compared to the control (Figure 6D). One MRE of miR-155-5p in the sequence of the 3'UTR of CDX1 mRNA was found (Supplementary Figure S19). To confirm the binding of miR-155-5p to CDX1 mRNA, luciferase vectors containing the wild-type or mutated sequence of the miR-155-5p binding site were manufactured. The overexpression of miR-155-5p significantly decreased luciferase activity; in contrast, the mutation of the binding sequence blocked the miR-155-5p-mediated inhibition of luciferase activity (Figure 6E).
We found that circPPFIA1s associated with miR-155-5p and mitigated its inhibitory function. Therefore, we tested whether circPPFIA1s regulated CDX1 expression. The knockdown of circPPFIA1-L or -S decreased the expression level of CDX1 protein and mRNA in KM12C cells (Figure 6F). Ago2 RIP and luciferase experiments were carried out to verify that miR-155-5p was required for the regulation of CDX1 by circPPFIA1s. The knockdown of circPPFIA1-L and -S increased the enrichment of CDX1 mRNA in Ago2 IP materials (Figure 6G) resulting from an increase in liberated miR-155-5p via a decrease in the level of circPPFIA1 as a ceRNA (Figure 6G). Increased levels of functional miR-155-5p due to the knockdown of circPPFIA1-L and -S also lowered luciferase expression in the wild-type but not in the mutant vector (Figure 6H). These results indicate that the inhibitory effect of miR-155-5p on CDX1 expression is enforced by the knockdown of circPPFIA1s. We also tested whether the overexpression of circPPFIA1s can upregulate CDX1 expression. The expression levels of CDX1 protein and mRNA were increased by the overexpression of circPPFIA1s (Figure 6I), and as expected, the enrichment of CDX1 mRNA in Ago2 IP was lowered in circPPFIA1-overexpressing cells (Figure 6J).
The above results indicated that the inhibition of miR-155-5p by anti-miR reversed the increased metastatic potential due to the knockdown of circPPFIA1s (Figure 5I). Accordingly, we assessed the expression level of CDX1 in the same samples. Decreased CDX1 expression due to the knockdown of circPPFIA1s was restored by introducing anti-miR-155-5p into KM12C cells (Figure 6K). To examine whether CDX1 is associated with metastatic potential, the invasive and migratory abilities of CDX1-silenced KM12C cells were assessed. We found that siRNA that targets CDX1 mRNA efficiently decreased the expression of CDX1 (Figure 6L). Transwell assays revealed that the knockdown of CDX1 enhanced invasive and migratory abilities (Figure 6M). These results indicate that liberated miR-155-5p by the knockdown of circPPFIA1s suppressed CDX1 by directly binding to its mRNA.
HuR is identified as a circPPFIA1s-interacting RBP
By predicting circPPFIA1s-associated RBPs using three algorithms (circInteractome, RBPDB, and StarBase), HuR was identified as a putative sponging RBP of circPPFIA1s (Figure 7A; Supplementary Figure S20A). Moreover, the association of HuR with circPPFIA1s was confirmed by a computational prediction (RBPmap, http://rbpmap.technion.ac.il). To verify the direct interaction between circPPFIA1s and HuR, ASO pulldown was followed by western blot analysis. We observed that HuR was bound to both circPPFIA1-L and -S (Figure 7B; Supplementary Figure S20B). In addition, the association of HuR with circPPFIA1s was examined by HuR RIP experiments. Semi-qPCR results showed that circPPFIA1s were more enriched in HuR IP than in IgG IP. These results indicate that HuR, as a sponging RBP of circPPFIA1s, is closely implicated in the anti-metastatic function of circPPFIA1s.
Next, the expression of HuR was compared in KM12C and KM12L4 cells. Interestingly, the western blot results revealed that the expression level of HuR in both cells was almost similar (Figure 7D). Moreover, the cellular localization of HuR did not differ between cells (Figure 7E). We investigated the effect of the circPPFIA1s on HuR expression and vice versa. When circPPFIA1-L and -S were silenced in KM12C cells, the expression level and cellular localization of HuR were unchanged (Figure 7F, 7G). Moreover, the knockdown of HuR by two independent siRNAs did not cause notably altered expression levels of circPPFIA1-L and -S in KM12L4 cells (Figure 7H). Based on these results, we assumed that circPPFIA1s may affect the regulatory functions of HuR, such as stabilization or translational activation of its target mRNA, without any change in the expression and localization of HuR.
To test whether HuR can regulate metastatic potential, the invasive and migratory abilities of KM12L4 cells were examined by transwell assays. The knockdown of HuR dramatically decreased the number of invaded and migrated cells (Figure 7I). We also determined whether HuR is required for the increased metastatic potential of KM12C cells by lowering the expression of circPPFIA1s. Increased metastatic potential by the knockdown of circPPFIA1s was reversed through HuR silencing (Figure 7J). This indicated that HuR is required for the increase in metastatic ability due to the knockdown of circPPFIA1s. As expected, when the expression level of HuR was lowered, the metastatic potential was decreased, and when circPPFIA1s were silenced, KM12C cells showed high metastatic potential. However, the knockdown of both HuR and circPPFIA1s decreased the invaded and migrated cell number compared to those of circPPFIA1-silenced cells
RAB36 is involved in the control of metastatic potential by circPPFIA1s/HuR
By comparing and analyzing the HuR CLIP-sequencing results with the list of genes upregulated under the three described conditions, 48 out of 62 genes (approximately 77% of the total merged genes) were found to be putative HuR target genes (Supplementary Figure S21). Among these genes, RAB36 was selected as a HuR target gene using the reference investigation (Figure 8A). To verify that RAB36 is a HuR target, an HuR RIP experiment was conducted. The level of RAB36 mRNA was more enriched in HuR IP compared to IgG IP (Figure 8B). Western blot and RT-qPCR analyses indicated that the expression levels of RAB36 protein and mRNA were higher in KM12L4 cells (Figure 8C). Moreover, the knockdown of HuR by two independent siRNAs downregulated RAB36 protein and mRNA (Figure 8D).
As an oncogene, the main mechanism of HuR is the stabilization of target mRNA by directly binding to its 3'UTR, which results in the upregulation of the target gene. Therefore, we determined whether HuR increased the stability of RAB36 mRNA. The decreased expression of RAB36 by knockdown of HuR was confirmed using the mixture of HuR siRNAs (Figure 8E). The knockdown of HuR induced a more rapid decrease in RAB36 mRNA compared to that in the control (Figure 8E). The estimated half-lives of RAB36 mRNA in the control and HuR-silenced KM12L4 cells were 5.4 h and 3.1 h, respectively. Next, we investigated the functional role of circPPFIA1s in HuR-mediated RAB36 regulation. The knockdown of circPPFIA1-L and -S increased the expression level of RAB36 protein and mRNA (Figure 8F). The HuR RIP experiment indicated that the knockdown of circPPFIA1s enhanced the interaction between HuR and RAB36 mRNA, which allowed HuR to stabilize RAB36 mRNA (Figure 8G). Although the estimated half-lives of RAB36 mRNA in the control was approximately 3.1 h, it increased to 5.8 h and 5.6 due to the knockdown of circPPFIA1-L and -S, respectively (Figure 8H). Conversely, the overexpression of circPPFIA1-L and -S induced a decrease in RAB36 protein and mRNA (Figure 8I) and lowered the enrichment of RAB36 mRNA in HuR IP materials (Figure 8J).
We assessed the expression level of RAB36 in the same samples, because RAB36 was identified as a HuR target. Increased level of RAB36 by the knockdown of circPPFIA1s was lowered by HuR silencing (Figure 8K). This indicates that the liberation of HuR by decreasing the levels of circPPFIA1s is required for highly metastatic phenotypes. We also tested whether RAB36 is involved in the invasive and migratory abilities of KM12L4 cells. Introducing a siRNA that targets RAB36 mRNA efficiently decreased the expression of RAB36 protein and mRNA in KM12L4 cells (Figure 8L). The metastatic potential was also diminished by the knockdown of RAB36 (Figure 8M).
Our findings are summarized by a schematic illustration in Figure 9. Briefly, two circPPFIA1s, generated from the exons of the PPFIA1 gene, are downregulated in the liver metastasis of CRC. They are mainly present in the cytosol, which allows them to function as molecular sponges. As tumor suppressors, circPPFIA1-L and -S negatively control the metastatic potential of CRC via two pathways: as a sponge of miR-155-5p, upregulating CDX1 expression; and as a sponge of HuR, downregulating RAB36 expression.