A Circle RNA Regulatory Axis Promotes Lung Squamous Metastasis via CDR1-Mediated Regulation of Golgi Trafficking

Lung cancer is the leading cause of cancer-related deaths worldwide and is an intrinsically highly metastatic disease. Even patients who undergo surgery at early stages have a high risk of recurrence (1). Despite lung squamous carcinoma (LUSC) being among the most lethal cancers, the mechanistic underpinnings of LUSC metastasis are poorly understood.

Although discovery of molecular aberrations in lung adenocarcinomas (LUAD) has led to development of effective targeted therapies (2), corresponding “drivers” in LUSCs have not been materialized. Extensive molecular profiling through The Cancer Genome Atlas (TCGA) has revealed LUSC tumors have nonrecurrent somatic mutations and are largely driven by copy-number alterations and distinct transcriptional programs (3, 4). Transcriptional profiling can stratify LUSC tumors into biologically distinct subtypes with different survival outcomes, indicating that regulation of gene expression in LUSCs has clinically relevant implications (5, 6).

Metastasis accounts for approximately 90% of cancer-related deaths (7), yet there is a surprising paucity of scientific evidence addressing the mechanisms that drive LUSC metastasis. The ability of cancer to spread is reliant on both cancer cell intrinsic and extrinsic factors. Gene expression programs within cancer cells facilitate egress from the tumor through blood or lymphatic vessels, extravasation, and colonization of distant organs (8). miRNAs regulate an increasing number of metastasis-relevant pathways through inhibition of target genes (9). However, the function of each miRNA is highly dependent on cancer type and even subtype (10, 11).

Recently, it was discovered that circular RNAs (circRNA) can regulate miRNA function. CircRNAs are covalently closed RNAs that are formed through an alternative splicing mechanism, which links the end of a 3′ exon with an upstream 5′ exon (termed a backsplice junction) and represent a new class of long noncoding RNAs (12). Some circRNAs act as miRNA sponges and contain multiple miRNA seed sequences (13, 14), but circRNAs may also act as protein scaffolds (15) or sequester RNA-binding proteins (RBP; ref. 16). Although expression of circRNAs is dysregulated in cancer, the repertoire of circRNA functions beyond miRNA sponging is still unclear. CircRNAs are structurally stable and released into the circulation within extracellular vesicles making them appealing biomarkers (17). Thus, characterizing the mechanism of action for circRNAs will facilitate biomarker development and identify possible therapeutic targets. Currently, therapeutic targeting of circRNAs is unexplored.

In this study, we evaluated whether miRNAs can regulate LUSC progression and discovered that miR-671-5p suppresses metastasis through the circRNA, CDR1as. By integrating miRNA and mRNA sequencing data from TCGA and profiling of metastatic LUSC models generated from in vivo passaging, we identified miR-671-5p as a key regulator of LUSC metastasis. Screening for miRNA targets revealed that miR-671-5p functions through CDR1as. While the proposed function of CDR1as is through sponging of miR-7 (13, 14), we instead describe a mechanism based on its antisense transcript, cerebellar degeneration related protein 1 (CDR1). CDR1 was highly linked to metastatic gene expression signatures and was necessary and sufficient to promote LUSC metastasis. CDR1 is a cryptic protein with an unusual sequence (composed predominantly of hexapeptide repeats) and no known function (18, 19). We found that CDR1 directly interacts with vesicular coat protein complexes and that inhibiting vesicular trafficking blocked CDR1-dependent migration. Using a targeted lipid nanoparticle to deliver miR-671-5p to target the CDR1as/CDR1 axis, we inhibited metastatic spread in vivo. These findings identify a previously unknown regulatory axis in LUSC progression and expand the functional repertoire of circRNAs in metastasis.

SK-MES-1 and H520 cells were obtained from the ATCC. HEK293T cells were kindly provided by Antonio Amelio (University of North Carolina at Chapel Hill, Chapel Hill, NC) and HCC2814 cells were obtained from University of Texas Southwestern Medical Center (Dallas, TX). SK-MES-1 cells were grown in minimum essential medium supplemented with 10% FBS, 1% penicillin–streptomycin, 1% nonessential amino acids, and 1 mmol/L sodium pyruvate. H520 and HCC2814 cells were grown in RPMI with 10% FBS and 1% penicillin–streptomycin. HEK293T cells were grown in DMEM with 10% FBS and 1% penicillin–streptomycin. UNCN3T cells and culture methods for them have been described in detail previously (20). All cell lines were tested to confirm the absence of Mycoplasma and grown at 37°C in 5% CO₂/95% air. Experiments were performed with cells at 60%–80% confluence. Cells were used fewer than 10 passages from thawing for in vivo experiments and fewer than 20 passages from thawing for in vitro experiments.

Cells were harvested from metastatic lesions by mechanical dissection in serum-free media containing 0.125% collagenase III and 0.1% hyaluronidase under sterile conditions. Minced tissues were filtered through a sterile 40-μm filter. After centrifugation, cells were incubated with 0.25% trypsin for 20 minutes at 37°C, agitating every 5–7 minutes. Cells were then grown in complete medium. Cells were expanded and FACS was performed using FITC-conjugated anti-human HLA-A,B,C antibody (32294x, Pharmingen) to remove any contaminating mouse cells.

For in vitro experiments, miR-671-5p or miRNA control mimics were purchased from Life Technologies and siRNAs were purchased from Sigma. Transfections of siRNAs and miR-671 mimics were performed using RNAiMAX (Invitrogen) at a final concentration of 20 nmol/L. For transfection of DNA, Lipofectamine 2000 (Invitrogen) was used.

Female athymic nude mice were purchased from the UNC Animal Studies Core. All animals used were between 6–10 weeks of age at the time of injection. For all animal experiments, cells were trypsinized, washed, and resuspended in Hank's Balanced Salt Solution (HBSS; Gibco) prior to injection. H520 or SK-MES-1 cells or related subclones were injected by an intrapulmonary technique (1 × 10⁶ cells in 50 μL 1:1 mixture of HBSS and BD Matrigel; BD Biosciences). For the intrapulmonary injections, mice were anesthetized with ketamine (80 mg/kg) + xylazine (8 mg/kg) + acepromazine (1 mg/kg) and placed in the right lateral recumbency. Following sterile skin preparation, an incision parallel to the rib cage between ribs 10 and 11 was made to visualize the lung through the intact thoracic pleura. A 1 mL tuberculin syringe with a 30-g needle was used to inject the cell suspension directly into the lung parenchyma at the left lateral dorsal axillary line. After injection, the skin incision was closed using surgery clips and the mice were turned on the left lateral recumbency and observed until fully recovered. In all experiments, 6–15 mice per group were used and mice were randomized before injection of cancer cells. For survival analyses, mice were monitored until they became moribund, and then sacrificed. For cross-sectional analysis, once mice in any group became moribund, they were all sacrificed, necropsied, and tumors were harvested. At the time of sacrifice, tumor weights, number, and location of lymphatic and distant metastases were recorded. Tissues used for IHC analysis were fixed in 10% neutral buffered formalin, and embedded in paraffin. Tissues were stained with either hematoxylin and eosin or Masson trichrome stain for histologic evaluation. Tissues for RNA and protein extraction were snap-frozen and stored at −80°C. RNA from tissues was isolated with TRIzol (Invitrogen) according to the manufacturer's instructions. Before qRT-PCR analysis, TRIzol-isolated RNA was treated with DNase using DNA-free DNA Removal Kit (Invitrogen) according to the manufacturer's instructions. Luciferase-labeled tumor progression was monitored weekly using an IVIS Lumina optical imaging system and Nano-Glo Luciferase Assay Substrate (Promega). Nano-Glo substrate was diluted in PBS and 250 μg/kg was administered 10 minutes before imaging.

Animals were cared for according to the guidelines set forth by the American Association for Accreditation of Laboratory Animal Care and the U.S. Public Health Service policy on Human Care and Use of Laboratory Animals. All mouse studies were approved and supervised by the University of North Carolina at Chapel Hill (Chapel Hill, NC) Institutional Animal Care and Use Committee. Tissue microarray (TMA) samples for lung squamous cell carcinoma and lung adenocarcinoma cancers were obtained and prepared following approval by the Institutional Review Board of UNC Chapel Hill.

The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE19 partner repository with the dataset identifier PXD012286.

More detailed methods are available in the Supplementary Materials and Methods. Sequences of short hairpin RNAs (shRNA) and qRT-PCR primers are listed in the Supplementary Data (Supplementary Table S1).

To develop metastatic models of LUSC, we chose SK-MES-1 and H520 human cell lines because they were recently found to be from aggressive mRNA subtypes associated with poor LUSC survival (5, 6, 21) and used an in vivo selection approach. We considered the highly prognostic value of lymph node (LN) metastases in lung cancer (1), including occult LN micrometastases (22, 23) and obtained subclones derived from LN metastases. While orthotopically injected parental SK-MES-1 cells metastasized to LNs in only 24% of mice, subclone SK-MES-LN1 (LN1) cells produced 100% LN metastases and very high rates of metastases to clinically relevant sites (e.g., adrenal glands and chest wall; Supplementary Fig. S1A and S1B). While mice injected with SK-MES-1 cells lived for >120 days, LN1-injected mice had substantially reduced survival (median, 41 days; P < 0.0001; Supplementary Fig. S1C). Similarly, H520 and H520-LN3 (LN3), which was generated from three in vivo passages, had LN metastasis rates of 20% and 100%, respectively (Supplementary Fig. S1D and S1E). In vivo passaging of H520 significantly reduced survival of orthotopically injected mice from >60 days to 39.5 days mean survival (P < 0.0001; Supplementary Fig. S1F). Consistent with the human disease and the importance of LN metastasis (22, 23), in both models, spontaneous development of LN metastasis significantly correlated with distant metastases (Spearman two-sided t test, P < 0.0001; Supplementary Fig. S1G). Indeed, recent evidence supports that metastatic cells within LNs can efficiently seed distant sites (24–26). There were no observed differences in proliferation between parental cell lines and respective subclones (Supplementary Fig. S1H). While distant metastases were most commonly found on the chest wall and in the adrenal glands, we also observed distant metastases in the optic nerve, base of tongue, and pancreas (Supplementary Fig. S2A–S2E). Histologic evaluation of LN1 orthotopic tumors revealed undifferentiated features, while LN3 showed central necrosis that is classically associated with LUSC. In both models, orthotopic tumors had considerable intratumoral collagen and LN metastases showed extracapsular extension (Supplementary Fig. S3). Metastasis is associated with epithelial-to-mesenchymal transition (EMT; ref. 27), to assess the role of EMT in the behavior of metastatic subclones, we profiled markers of EMT by qPCR. Relative to parental cell lines, both LN1 and LN3 had increased expression of SNAI2 and ZEB2, key transcriptional regulators of EMT (Supplementary Fig. S4). Taken together, the LN1 and LN3 models are highly metastatic and display typical features of human LUSC.

miRNAs can have remarkable roles in regulating a repertoire of metastatic programs in many cancers (28). To screen miRNAs with potential roles in LUSC metastases, we analyzed miRNA expression data from 348 patients with LUSC in TCGA. Using median cutoffs for each miRNA, 30 miRNAs were significantly associated with overall survival when individually considered (P < 0.05; Supplementary Table S2); therefore, this analysis was exclusively used as a screening tool to prioritize miRNA genes for experimental validation. Among these 30 genes, high expression was associated with poor survival (HR > 1) in only five miRNAs, while for the remaining 25 miRNAs, low expression was associated with poor survival (HR < 1; Fig. 1A and B). Altogether, these data suggest that suppression of miRNAs may generally lead to poorer outcomes in LUSC.

Next, we compared miRNA expression profiles between parental SK-MES-1 and the LN1 subclone using NanoString data. Similar to our TCGA findings (Fig. 1A and B), a majority of miRNAs were downregulated in the metastatic clones (Fig. 1C; Supplementary Table S3). Integrating miRNAs associated with survival and decreased in metastatic subclones, we selected 12 candidate miRNAs for validation (more details in Supplementary Materials and Methods). All miRNA candidates were decreased on the basis of qPCR in both metastatic LN1 and LN3 subclones relative to their respective parental cell lines (Fig. 1D). Of these candidates, miR-671-5p was the most consistently reduced (<20% of parental expression) in both LN1 and LN3. In each model, stable overexpression of miR-671 at physiologic levels (4- to 7-fold increases over control miRNA) significantly increased survival and decreased metastatic burden (Fig. 1E–J), implicating miR-671 as a novel metastasis suppressor in LUSC. As seen when comparing metastatic and parental cell lines, no appreciable effect on proliferation was observed in miR-671–overexpressing cells (Supplementary Fig. S5). Survival analysis based on the expression of the miR-671 gene and miR-671-5p and -3p isoforms in TCGA revealed that miR-671-5p [HR, 0.56; 95% confidence interval (CI), 0.39–0.80; P = 0.002], but not -3p (HR, 0.92; 95% CI, 0.64–1.33; P = 0.7), was strongly associated with LUSC survival (Fig. 1K–M). The expression of miR-671-5p and -3p was significantly correlated (Spearman two-sided t test, P < 0.0001; R = 0.19), but the expression of miR-671-5p was far more dynamic and significantly higher than -3p (Supplementary Fig. S6A and S6B).

To systematically identify biologically relevant miR-671-5p target genes in LUSC, we leveraged a previously described approach to integrate a linear regressions model of TCGA data and in silico predictions (11, 29). The miRNA target prediction algorithm, TargetScan (30), was used to predict miR-671-5p regulated genes (Supplementary Fig. S7A). Then, to enrich for genes relevant for LUSC biology, candidate genes had to meet four conditions based on TCGA analysis (n = 348): (i) inverse correlation with miR-671-5p (R < −0.25; FDR < 0.00001), (ii) survival relevance (P < 0.05; HR > 1.5), (iii) influence of miR-671-5p on target mRNA expression (>25%), and (iv) joint survival of target mRNA and miR-671 with more stringent criteria than for genes alone (P < 0.01; HR > 2). On the basis of these four criteria, 13 mRNAs were considered clinically relevant candidate miR-671-5p targets in LUSC (Supplementary Table S4). To validate these putative target genes, qPCR was performed on LN1 and LN3 cells stably expressing miR-671 or miR-671-5p inhibitors. Two previously validated targets (FOXM1; ref. 31 and CDR1as; ref. 32) were included as positive controls. Of all the genes tested, two genes (MSR1 and CD93) were lowly expressed in LN1 and LN3 cell lines and, therefore, excluded. Surprisingly, CDR1as and IL16 were the only target genes to show the predicted directionality in all four validation experiments (Supplementary Fig. S7B–S7E). Effects of miR-671 manipulation on CDR1as were notably stronger than on IL16 in most experiments. Ranking miR-671-5p target genes based on miRNA binding site strength revealed that CDR1as, a circular antisense transcript of the CDR1 gene (32), had a Context++ score of −3.7, several fold lower than the next lowest score of −0.53 for ANTXR2 (Supplementary Fig. S7F). Moreover, the predicted miR-671-5p binding site in CDR1as was complementary at 21 of 23 bases, resembling an siRNA (Fig. 2A and B). We confirmed that miR-671-5p potently decreased CDR1as expression in both of our metastatic subclones (Fig. 2C). These data show that miR-671-5p strongly targets the noncoding RNA, CDR1as. CDR1as was not detected in our initial screen because our target prediction was limited to linear, protein-coding mRNAs.

CDR1as (also known as ciRS-7), is a direct target of miR-671-5p as confirmed by luciferase assay (32) and Argonaut cross-linking and immunoprecipitation in mice and humans (33). CDR1as contains >70 repeats of the miR-7 seed sequence and is known to act as a miR-7 sponge (13, 14, 32). Unlike miR-671-5p, which acts through the endonuclease Ago2 (32), miR-7 binds more weakly through a canonical 7–8 nucleotide seed sequence (Supplementary Fig. S8A). Therefore, CDR1as can sequester miR-7 and associated RNA-induced silencing complexes leading to disinhibition of miR-7 targets (13, 14). To determine whether miR-671-5p silencing of CDR1as can increase free miR-7 and lead to decreased miR-7 targets, we measured the levels of nine validated miR-7 targets with a focus on those relevant to cancer biology (Supplementary Fig. S8B and S8C). Interestingly, we found no significant changes in miR-7 targets in LN1 or LN3 cells overexpressing miR-671. Of the nine genes tested, none showed more than a 30% decrease in cells overexpressing miR-671. These findings suggest miR-671-5p's role in LUSC metastasis may occur through CDR1as, although in an miR-7–independent manner. CDR1as has also been shown to increase the expression of the CDR1 mRNA, with which it shares a gene locus (32). CDR1 was discovered as a target of autoantibodies in paraneoplastic cerebellar degeneration (18, 19). Although CDR1 is highly expressed in the brain and upregulated in some cancer types, its biologic function is unknown. To confirm that miR-671-5p also reduced CDR1 expression in LUSC cells, we measured CDR1 by strand-specific reverse transcription followed by qPCR in cells transfected with miR-671-5p mimics. miR-671-5p strongly decreased CDR1 expression in both LN1 and LN3 subclones (Fig. 2D).

We next assessed the biological relevance of the CDR1as/CDR1 axis in lung cancer and metastases. Expression of CDR1as was measured using ISH on a clinically annotated TMA containing duplicate biopsies of LUSC, LUAD, and patient matched normal lung samples (Fig. 2E–G). Notably, we found that CDR1as was 10-fold higher in LUSC tumors than matched normal lungs (P = 0.0002) and 2-fold higher compared with LUAD tumors (P < 0.0001; Fig. 2E). High CDR1as expression was associated with worse survival in non–small cell lung cancer (NSCLC; HR, 1.39; 95% CI, 1.01–1.94; P = 0.048; Fig. 2G), suggesting a role in driving metastases. IHC for CDR1 also showed a significant increase in both LUAD and LUSC tumors compared with normal lung (Fig. 2H and I). CDR1 staining was present in the nucleus, cytoplasm, and perinuclear space and its localization varied between patients and cells within an individual tumor (Supplementary Fig. S9). High CDR1 expression was associated with worse survival in NSCLC (HR, 1.43; 95% CI, 1.02–2.01; P = 0.0429). Similarly, in TCGA, high expression of CDR1 in patients with LUSC was associated with significantly worse survival (HR, 1.57; 95% CI, 1.10–2.24; P = 0.019; Fig. 2J). To address the importance of the miR-671-5p/CDR1as/CDR1 axis in LUSC, we performed a dual survival analysis of miR-671-5p and CDR1 and found that low miR-671-5p/high CDR1 expression in tumors was associated with worse survival than high miR-671-5p/low CDR1 expression (HR, 2.094; 95% CI, 1.311–3.344; P = 0.002; Fig. 2K). Using TCGA to cluster LUSC tumors based on high CDR1/low miR-671-5p and low CDR1/high miR-671-5p expression, we identified 2,123 differentially expressed genes (Supplementary Fig. S10A; Supplementary Table S5). Gene set enrichment analysis (GSEA) in the high CDR1/low miR-671-5p cohort revealed dramatic enrichment for signatures of EMT and signaling of TGFβ (a potent inducer of EMT; Fig. 2L and M). These GSEA findings were powerfully matched by a gene ontology (GO) analysis of the genes upregulated in this cohort (Supplementary Fig. S10A, top), which showed that the most statistically significant GO term (P = 1.10E-49) was “cell adhesion” (Supplementary Fig. S10B; Supplementary Table S6); indeed EMT and cell–cell adhesion are intimately related (27). GSEA also identified a strong enrichment for genes upregulated when there is loss of the cell adhesion molecule, E-cadherin (CDH1; Supplementary Fig. S10C). Consistent with these findings, CDR1 expression across >1,000 cancer cell lines strongly correlated (Spearman two-sided t test, P < 0.0001) with several EMT genes (Supplementary Fig. S10D; Supplementary Table S7). Taken together, these findings imply that CDR1 expression is closely tied to poor survival and the expression of metastatic gene signatures in LUSC, including EMT.

To determine whether inhibiting the CDR1as/CDR1 axis is necessary for the suppressive effect of miR-671-5p on LUSC metastasis, LN1 cells were stably transduced with shRNAs against CDR1 or CDR1as. Inhibition of CDR1/CDR1as with several independent shRNAs (Supplementary Fig. S11A–S11C) significantly increased survival of orthotopically injected mice from 53 days (control shRNA) to between 78 and 136 days (CDR1 and CDR1as shRNAs, respectively; Fig. 3A). In addition, inhibition of CDR1 or CDR1as markedly reduced LN metastases (Fig. 3B), in all four groups, and tumor burden as measured by luciferase expression (Fig. 3C and D). CDR1as shR2 showed the strongest inhibition of LUSC progression and metastasis, reducing LN metastases by >75% (P = 0.004) and tumor burden by 30-fold (P = 0.0002) relative to control (Fig. 3B and C). To test whether CDR1 is sufficient to promote LUSC progression and metastasis, SK-MES-1 and LN1 cells were stably transduced with the CDR1 open reading frame. CDR1 overexpression in SK-MES-1 cells was sufficient to increase distant metastases to the level of the aggressive LN1 model (P < 0.05), but had no significant effect on primary tumor size (Fig. 3E–G). CDH1 was decreased in tumors overexpressing CDR1, supporting a role for CDR1 in EMT (Supplementary Fig. S12). Taken together, the CDR1as/CDR1 axis is necessary and CDR1 alone is sufficient to drive LUSC metastasis. Importantly, overexpression of CDR1 was able to partially rescue the metastatic phenotype of CDR1as knockdown, showing that the biological function of CDR1as is at least, in part, dependent on its regulation of CDR1 (Fig. 3H and I).

To target the CDR1as/CDR1 axis therapeutically, we formulated lipid protamine hyaluronic acid nanoparticles (LPH-NP) containing miR-671-5p mimics (Fig. 4A). A characteristic core structure was seen by transmission electron microscopy (Fig. 4B) and LPH-NPs had a hydrodynamic diameter approximately 100 nm in size, as measured by dynamic light scattering, with a charge of +40 mV. LPH-NPs were decorated with the Sigma1R ligand amino-ethyl-anisamide (AE-AA) to facilitate uptake into tumor cells (34). Both LN3 and LN1 expressed Sigma1R on the cell surface as determined by biotinylation and isolation of cell surface proteins (Supplementary Fig. S13A). Also, SIGMAR1 expression in LUSC tumors was just more than half of that of EGFR (1,820 ± 1,200 and 3,630 ± 7,070, respectively), which is often amplified in LUSC (4), and more than 10 times higher than hormone receptors, ESR1 (90 ± 140), PGR (30 ± 40), and AR (20 ± 40, Supplementary Fig. S13B). Incorporation of AE-AA increased uptake of LPH-NPs carrying fluorescently tagged dsRNA (Fig. 4C). Given the higher expression of Sigma1R on the surface of LN3 cells (Supplementary Fig. S13), this model was chosen for further therapeutic experiments. LPH-NPs carrying fluorescently labeled dsRNA were used to measure the biodistribution of dsRNA in tumor-bearing mice. Forty-eight hours after a single intravenous injection, dsRNA accumulated in LN3 lung primary and metastatic tumors (Fig. 4D and E). Time–kinetic experiments revealed that miR-671-5p LPH-NPs decreased CDR1as expression by 60% relative to control LPH-NPs at 24 hours, and CDR1-positive cells by 70% at 48 hours (Supplementary Fig. S14A–S14C). Treatment with miR-671-5p nanoparticles for 3× per week for 3 weeks significantly decreased the number of LN metastases in LN3 mice (Fig. 4F and G) without evident toxicity or weight loss (Supplementary Fig. S15). In tumors collected 24 hours after the final nanoparticle injection, the number of CDR1-positive cells was significantly decreased in LN metastases of mice treated with miR-671-5p nanoparticles compared with controls (Fig. 4H and I). Interestingly, LN metastases had significantly higher numbers of CDR1-positive cells than primary lung tumors (Fig. 4I), consistent with CDR1 having a role in LUSC metastases.

To characterize the cellular effects of CDR1 overexpression, we assessed proliferation and migration in our LUSC cell lines. While modest effects on proliferation and colony formation were observed (Supplementary Fig. S16A–S16C), overexpression of CDR1 in parental SK-MES-1 cells was sufficient to significantly drive chemotactic migration to the level of the LN1 subclone (Fig. 5A). Similarly, SK-MES-1-CDR1 cells moved with greater velocity (P < 0.01) and traveled a greater distance (P < 0.05) compared with SK-MES-1-GFP cells in a random migration assay on type 1 collagen (Fig. 5B). These experimental outcomes are aligned with the GO analysis of the genes upregulated in high CDR1/low miR-671-5p, which shows that one of the most significant biological process in these patients is “positive regulation of cell migration” (Supplementary Fig. S10B), intended by an increased frequency, rate, or extent of migratory capabilities. To understand the mechanism by which CDR1 promotes metastasis, we identified protein interacting partners of CDR1 using epitope-mediated immunoprecipitation followed by mass spectrometry (IP-MS). A total of 34 proteins were associated with CDR1 in both 293T and LN1 cells stably expressing CDR1-Flag (Fig. 5C and D), and 19 of 34 proteins were localized to the Golgi apparatus and/or endoplasmic reticulum (ER; Fig. 5C). It is noteworthy that EMT-driven Golgi compaction has recently been linked to promotion of LUAD metastasis (35). Several CDR1-interacting partners were validated by immunoprecipitation–Western blotting (Fig. 5E). CDR1 interacted with coatomer protein I (COPI) complex subunits, COPA and COPE, which mediate transport between the ER and Golgi (36), as well as several members of the adaptor protein 1 (AP1) complex (AP1G1, AP1G2, and AP1S1). The AP1 complex serves as an adaptor for clathrin, traffics between the trans-Golgi network and endosomes, and is important for basolateral polarized sorting (36).

To explore the interaction between CDR1 and vesicular proteins, we performed immunocytochemistry for CDR1. CDR1 staining was observed in the nucleoplasm, ER, and Golgi structures (Fig. 6A), consistent with IP-MS results (Fig. 5C). Coimmunostaining for COPA and AP1G1 showed that CDR1 partially colocalized with COPA and AP1G1 in perinuclear structures consistent with Golgi (Fig. 6A; Supplementary Fig. S17A; Supplementary Table S8). CDR1 also partially colocalized with COPA and AP1G1 in HCC-2814 cells, which express intermediate endogenous levels of CDR1 (Supplementary Fig. S17B; Supplementary Table S8). Proximity ligation assays confirmed that CDR1 interacts with COPA and AP1G1 in LN1 and LN3 cells at endogenous levels (Fig. 6B). Consistent with CDR1 regulation of Golgi-mediated processes, air–liquid interface cultures of the human bronchial epithelial cell line, UNCN3T, overexpressing CDR1 displayed a striking increase in mucin production (Supplementary Fig. S18). Furthermore, use of brefeldin A to inhibit Arf1-GTPase, which is required for AP1 and COPI membrane anchorage (37), abolished CDR1-dependent migration and haptotaxis (Fig. 6C and D). Similarly, in epistasis experiments, silencing of COPA, but not AP1G1, blocked CDR1-dependent migration (Fig. 6E), suggesting that COPA is functionally downstream of CDR1-mediated migration.

Considering that Golgi complex reorientation is essential during cell migration (38), we tested whether CDR1 expression altered Golgi orientation during a scratch assay. Consistent with a promigratory phenotype, SK-MES-1-CDR1 cells invaded the scratch in significantly greater numbers than SK-MES-1-EV cells (Fig. 7A and B). The orientation of the Golgi in cells at the scratch edge was scored as either on the side of the nucleus facing toward the scratch (toward) or on the side of the nucleus facing away from the scratch (away). On the basis of this analysis, a significantly higher proportion of CDR1 cells had Golgi oriented toward the scratch than empty vector (EV) cells (Fig. 7C), indicating that CDR1 increases Golgi orientation toward the direction of migration.

ER to Golgi trafficking is required for the secretion of prometastatic factors. To determine whether CDR1 overexpression has a functional effect on ER to Golgi trafficking, we utilized the retention with selective hooks assay (39) to quantitatively measure the rate of ER to Golgi trafficking. SK-MES-1-EV or -CDR1 cells were transfected with streptavidin binding protein (SBP) fused to a secreted soluble GFP (ssSBP-GFP) and streptavidin fused to the invariant chain of the MHC (Ii) (Str-Ii), which acts as an ER hook. ssSBP-GFP was retained in the ER until the addition of biotin, which released ssSBP-GFP to traffic into the Golgi. Fluorescence images were captured every minute starting 5 minutes after biotin addition for 45 minutes. GFP accumulation in the Golgi, labeled with mApple-Sialyltransferase (mApple-SiT), was quantified (Fig. 7D and E; Supplementary Videos S1 and S2). Both SK-MES-1-EV and -CDR1 showed an accumulation of GFP in the Golgi over the imaging period (Fig. 7F). We observed a small population of fast trafficking cells (>3-fold from baseline at 50 minutes after biotin addition; Fig. 7G). There was a significantly higher proportion of “fast” trafficking cells in CDR1 (4/11 cells) compared with EV cells (0/11 cells; P < 0.05; Fig. 7H). These data demonstrate that CDR1 can promote ER to Golgi trafficking. Together, these findings represent the first functional characterization of CDR1 and identify it as a key driver of LUSC metastatic biology, which is dependent on Golgi trafficking.

Therapeutic strategies to inhibit metastatic spread in aggressive cancers, such as LUSC, are urgently needed. To identify key regulators of LUSC metastasis, we screened miRNAs associated with patient survival and metastatic phenotypes and identified miR-671-5p as an antimetastatic miRNA. Expression of miR-671-5p was strongly associated with patient survival and genetic overexpression or therapeutic delivery of miR-671-5p was capable of inhibiting metastasis. The circRNA, CDR1as, was strongly inhibited by miR-671-5p, as was its antisense transcript, CDR1. This CDR1as/CDR1 axis was associated with worse survival in patients with LUSC and inhibition of CDR1as or CDR1 decreased tumor burden and metastases. Overexpression of CDR1 was sufficient to promote metastatic spread and rescue loss of CDR1as.

These findings are consistent with miRNA profiling studies that found miR-671-5p among miRNAs associated with increased survival in stage I NSCLC (40). In agreement with previous studies, we found that miR-671-5p targeted the circRNA, CDR1as (32, 33). However, CDR1as was not detected by a bioinformatic screen for miR-671-5p targets. This highlights a weakness of standard miRNA target prediction approaches, which can overlook miRNA–noncoding RNA interactions.

We found that CDR1 promotes metastasis through Golgi orientation and trafficking. Intrinsic functions of Golgi trafficking are critical for metastasis (35) and cell migration (38). We report that CDR1 overexpression can increase cancer cell motility, chemotaxis, and haptotaxis, and epistasis experiments revealed that this gain of function is dependent on Golgi trafficking. Golgi trafficking can also promote metastasis through extrinsic mechanisms by secretion of proinvasive and proangiogenic factors (41). These secreted factors require processing and trafficking in the ER–Golgi. Using nontransformed human bronchial cells, we found that CDR1 expression increased mucin secretion. CDR1 overexpression in LUSC cells facilitated Golgi orientation toward the direction of migration and increased the rate of ER to Golgi trafficking, suggesting CDR1 may also have effects on the secretome and extracellular state. In addition to the Golgi, CDR1 also localized to the nucleus as detected by immunostaining and interacted with several nuclear proteins in our IP-MS analysis. Further investigation of the role of CDR1 in the nucleus is warranted.

Understanding of the cross-talk between circRNA and miRNAs is rapidly expanding. Although circRNAs were discovered decades ago, they were initially dismissed as splicing errors (42). The renaissance of circRNAs began with the discovery that circRNAs are abundant and functional in mammals (13, 14). These early reports indicated that circRNAs could function as miRNA sponges, and suggested that this mechanism could influence cancer biology. In particular, CDR1as acts as a sponge for the tumor suppressor, miR-7 (13, 14). CDR1as is highly expressed in the brain and full body knockout models of CDR1as resulted in a brain-specific decrease in miR-7, increase in miR-7 targets, and mild neuropsychologic deficits, suggesting sponging of miR-7 by CDR1as stabilizes rather than inhibiting miR-7 (33). In neurons, excess miR-7 can also enhance the silencing of CDR1as by miR-671-5p (43). Importantly, no adverse effects of CDR1as deletion were found outside the central nervous system, making CDR1as a highly appealing therapeutic target for cancer.

There are >70 miR-7 seed matches decorating CDR1as, which is an unusually high amount. For most circRNAs there are no more miRNA binding sites than expected by chance (44). Other known functions for circRNAs include acting as a decoy for RBPs and serving as protein scaffolds. However, little is known about mechanisms for circRNAs in cancer beyond miRNA sponging (17). CDR1as is also an unusual circRNA in that it is a natural antisense transcript (32). Natural antisense transcripts can increase or decrease expression of their complimentary mRNA. In some cases, this is through complimentary base pairing (45). Given that the entire CDR1 mRNA is complementary to CDR1as, CDR1as likely stabilizes CDR1 directly through base pairing. Consistent with our findings in LUSC, the function of CDR1as in melanoma is also miR-7 independent (46). Interestingly, in melanoma, CDR1 was not found to be expressed and CDR1as instead functions through an RBP-mediated mechanism to inhibit metastasis (46). Thus, CDR1as functions in a context-dependent manner to enact different, even divergent, cancer phenotypes.

The repeated miR-7 seed sequence on CDR1as is highly conserved. Interestingly, this seed sequence corresponds to the hexapeptide repeat that is present in CDR1. Therefore, it is unclear whether protein sequence, miRNA seed, or both drive evolutionary conservation. We found miR-671-5p's role in suppressing metastasis occurred through a CDR1as/CDR1 axis and that CDR1 alone was sufficient to promote metastasis. CDR1 is highly expressed in the brain and acts as an onconeural antigen in paraneoplastic disorders (18, 19). The predicted structure of CDR1 is an α-helix with a 30-amino acid segment of β-sheet conformation at its C-terminus. Repeated sequences and α-helices are characteristic of a wide range of proteins, including structural proteins, RNA polα, and fungal surface proteins (18, 19). Therefore, it is difficult to discern the function of CDR1 by sequence alone. On the basis of direct protein–protein interaction and immunostaining, we determined that CDR1 localized to multiple cellular compartments and was associated with COPI and AP1 vesicles. Furthermore, epistasis experiments revealed that inhibition of the COPI subunit, COPA, blocked CDR1-dependent migration. Given the importance of secretion and retrograde transport in neurotransmission and the high expression of CDR1 in the brain, these findings have important implications in neurobiology.

In conclusion, we have found that miR-671-5p inhibits LUSC metastases by targeting the poorly understood CDR1as/CDR1 axis. Although previously only known to be associated with paraneoplastic cerebellar degeneration (18, 19), we found that CDR1 promotes metastases via increased cell migration through Golgi trafficking. To our knowledge, this is one of few reports to show a circRNA driving metastasis in vivo and the first for CDR1as. In addition, we describe the first therapeutic targeting of a circRNA. These findings reveal a complex regulatory network used to promote LUSC metastases and may have important implications in the understanding of the progression of other cancer types. Furthermore, these findings reveal a broader functional repertoire for circRNAs, which can be targeted through RNAi-based strategies.

E.B. Harrison reports grants from Lung Cancer Initiative of North Carolina and NCI during the conduct of the study. A.E.D. Van Swearingen reports grants from NIH (R01CA215075) during the conduct of the study. T.J. Goodwin reports grants from NIH during the conduct of the study. S. Cohen reports grants from NIH during the conduct of the study, and NIH and Alzheimer's Association outside the submitted work. C.V. Pecot reports other compensation from EnFuego Therapeutics (founder and equity in a biotechnology company investigating RNA interference approaches to treat cancer) outside the submitted work. No potential conflicts of interest were disclosed by the other authors.

E.B. Harrison: Formal analysis, funding acquisition, validation, investigation, visualization, writing-original draft, writing-review and editing. A. Porrello: Data curation, software, formal analysis, investigation, visualization, writing-review and editing. B.M. Bowman: Formal analysis, investigation, writing-review and editing. A.R. Belanger: Formal analysis, validation, investigation, writing-review and editing. G. Yacovone: Formal analysis, validation, investigation, writing-review and editing. S.H. Azam: Investigation, writing-review and editing. I.A. Windham: Formal analysis, visualization, writing-review and editing. S.K. Ghosh: Validation, writing-review and editing. M. Wang: Investigation, writing-review and editing. N. Mckenzie: Investigation, writing-review and editing. T.A. Waugh: Validation, investigation, writing-review and editing. A.E.D. Van Swearingen: Investigation, writing-review and editing. S.M. Cohen: Investigation, writing-review and editing. D.G. Allen: Investigation, writing-review and editing. T.J. Goodwin: Investigation, writing-review and editing. T. Mascenik: Investigation, writing-review and editing. J.E. Bear: Resources, writing-review and editing. S. Cohen: Methodology, writing-review and editing. S.H. Randell: Resources, investigation, writing-review and editing. P.P. Massion: Resources, writing-review and editing. M.B. Major: Investigation, writing-review and editing. L. Huang: Resources, investigation, writing-review and editing. C.V. Pecot: Conceptualization, resources, formal analysis, funding acquisition, investigation, visualization, writing-original draft, writing-review and editing.

The authors acknowledge members of the Pecot laboratory for helpful discussions and feedback. The authors would like to especially thank Drs. Yongjuan Xia and Nana Feinberg from the UNC Translational Pathology Laboratory for their help with ISH, IHC, and processing of the tissue microarrays and the UNC Animal Histopathology Core. Confocal microscopy was performed with equipment and assistance from the UNC Microscopy Services Laboratory, Department of Pathology and Laboratory Medicine. We also thank Dr. Antonio Amelio for providing the Fluorescent-Nanoluciferase plasmids and Drs. Gianpietro Dotti and Hongwei Du for GFP-luciferase labeling of cell lines. The authors acknowledge the following financial support: The UNC Translational Pathology Laboratory was supported, in part, by grants from the NCI (5P30CA016080-42), NIH (U54-CA156733), NIEHS (3P30 EOS010126-17), UCRF, and NCBT (2015-IDG-1007). C.V. Pecot was supported, in part, by the NIH grant no. R01CA215075, a Mentored Research Scholar Grants in Applied and Clinical Research (MRSG-14-222-01-RMC) from the American Cancer Society, the Jimmy V Foundation Scholar award, the UCRF Innovator Award, the Stuart Scott V Foundation/Lung Cancer Initiative Award for Clinical Research, the University Cancer Research Fund, the Lung Cancer Research Foundation, the Free to Breathe Metastasis Research Award, and the Susan G. Komen Career Catalyst Award. A.R. Belanger was supported by a grant from the National Heart, Lung and Blood Institute of the NIH under award number HL007106-38, an ASCO Young Investigator Award, and by the Lung Cancer Initiative of North Carolina. S.H. Azam and I.A. Windham were supported, in part, by a grant from the National Institute of General Medical Sciences under award 5T32 GM007092 and GM119999, respectively. E.B. Harrison was supported by a grant from the NCI of the NIH under award number T32CA196589 and by the Lung Cancer Initiative of North Carolina. S.M. Cohen and I.A. Windham were supported, in part, by an NIH R35 grant award number GM133460. S.H. Randell was supported, in part, by NIH 5P30DK065988. The UNC Microscopy Services Laboratory and Lineberger Comprehensive Cancer Center Animal Histopathology and Animal Studies Cores were all supported, in part, by an NCI Center Core Support grant (CA016086) to the UNC Lineberger Comprehensive Cancer Center. The UNC Flow Cytometry Core Facility was also supported, in part, by the North Carolina Biotech Center Institutional Support grant 2012-IDG-1006.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.