The Circular RNA circPRKCI Promotes Tumor Growth in Lung Adenocarcinoma Free

Copy number variation (CNV) is a form of genomic structural variation leading to gains and losses of DNA segments. Somatic CNVs are extremely common due to the genomic instability and play a significant role in tumorigenesis in many cancers, including colorectal, gastric, and lung cancers (1–4). During cancer development, proliferation-related genes are often amplified, in comparison with frequent loss of apoptosis effector genes. A large number of cancer-driving CNV loci that encode proteins have been successfully identified using high-throughput genome sequencing technologies (5, 6). Taking epithelial cancers as an example, integrated cancer genomic analysis and transgenic animal model have confirmed some well-known amplicons induced proto-oncogenic proteins like MYC (7) and PIK3CA (8), as well as deletions induced tumor suppressor–like RB1 (9) and PTEN (10).

Oncogenic proteins are not the only entities involved in tumorigenesis. Noncoding RNAs are also critical in cancer development, such as long noncoding RNA (lncRNA) PVT1 in the 8q24 “genomic desert” region, lncRNA FAL in the 1q21 amplification region (11), recurrently deleted lncRNA-PRAL on chromosome 17q13.1 (12), as well as the oncogenic miR-569 on the 3q26.2 amplicon (13). These findings indicated that noncoding RNAs, as the majority of transcriptome, may represent a large number of unexplored targets of genomic aberrations. Therefore, further exploration of the hidden noncoding transcripts within recurrent CNV loci in cancers is warranted.

Circular RNAs (circRNA), a naturally occurring family of noncoding RNAs, are involved in multiple biological processes (14, 15), including cancers (16, 17). One of the earliest characterized circRNAs was the sex-determining region of ChrY (Sry) in mice (18), and the well-known “miRNA sponging” function has been demonstrated for an antisense transcript to cerebellar degeneration-related protein 1 (CDR1as/ciRS-7; ref. 19), which contains 70 binding sites for miR-7 and suppresses miR-7 activity. Several cancer-derived circRNAs include circHIPK3 in multiple cancers (16), the hepatocellular carcinoma suppressor circMTO1 (20), the colon cancer progression promotor circCCDC66 (21), and gastric cancer highly expressed circPVT1 in the 8q24 amplification locus (22), which is highly expressed in gastric cancer. However, CNV-associated circRNAs in lung cancer have been reported rarely.

Lung adenocarcinoma is currently the most common histologic type of lung cancer (23). In the current study, we identified 107 differently expressed circRNAs in lung adenocarcinoma by using microarray, followed by an integrated analysis of published lung adenocarcinoma–specific CNV data derived from The Cancer Genome Atlas (TCGA; ref. 24). We further characterized a circular RNA, termed as circPRKCI, produced from the PRKCI gene at 3q26.2 amplicon. Subsequent biotin-coupled miRNA pulldown experiment revealed circPRKCI could function as a sponge for both miR-545 and miR-589, thus increasing expression levels of E2F7 and promoting tumorigenesis of lung adenocarcinoma.

All primary lung adenocarcinoma tissues and adjacent nontumor tissues were collected from patients who had undergone surgery at the Department of Thoracic Surgery, Nanjing Medical University Affiliated Cancer Hospital (Nanjing, China). All tumors and paired nontumor tissues were confirmed by experienced pathologists. Written informed consent was obtained from all patients. Collection of human tissue samples wad conducted in accordance with the International Ethical Guidelines for Biomedical Research Involving Human Subjects. This study was approved by the Ethics Committee of the Nanjing Medical University Affiliated Cancer Hospital and was performed in accordance with the provisions of the Ethics Committee of Nanjing Medical University. This study was approved by the Nanjing Medical University.

All cell lines [A549, NCI-H1975, NCI-H1703, NCI-H226, NCI-H46, PC9, NCI-H1299, SPC-A1, HCC827, and human bronchial epithelial cell (HBE)] were purchased from Shanghai Institutes for Biological Science, (Shanghai, China). NCI-H1975, A549, NCI-H1703, NCI-H226, NCI-H46, PC9, HCC827, and NCI-H1299 cells were cultured in RPMI1640 medium (KeyGene); SPC-A1 and HBE cells were cultured in DMEM medium (KeyGene), supplemented with 10 % FBS with 100 U/mL penicillin and 100 mg/mL streptomycin included. All cell lines were grown in humidified air at 37 °C with 5 % CO₂. Cell cultures were occasionally tested for mycoplasma (last tested 2016). Authentication of cells was verified by short tandem repeat DNA profiling within 6 months of use for the current study. The cells used in experiments were within 10 passages from thawing.

Five lung adenocarcinoma tissues and paired nontumor tissues were used for microarray analysis. The microarray experiment was performed by Kangcheng Bio-tech Inc. The microarray data were submitted to the Gene Expression Omnibus, and the data can be accessed by the accession number GSE101586.

Tissue microarray (TMA) was constructed as described previously (25). Eighty-nine pairs of lung cancer tissues and adjacent nontumor tissues were used to construct the TMA. RNA chromogenic in situ hybridization (CISH) was performed to detect circPRKCI expression in TMA using digoxigenin-labeled probe (5′-GTATGCGAATTTGTTTTTCCAAAATAACATATCCCAATCA-3′). Briefly, after dewaxing and rehydration, the samples were digested with proteinase K, fixed in 4% paraformaldehyde, and hybridized with the digoxin-labeled probe overnight at 55°C. The samples were then incubated overnight at 4°C with an anti-digoxin mAb (Roche Applied Science). The sections were stained with nitro blue tetrazolium/5-bromo-4-chloro-3-indolylphosphate (NBT/BCIP) in the dark, mounted, and observed.

RNA extraction and qRT-PCR were performed as described previously (26). Genomic DNA (gDNA) was extracted from tissues or cultured cells according to the PureLink Genomic DNA Mini Kit protocol (Thermo Fisher Scientific, K182001). GAPDH, ACTB, and snRNA U6 were used as internal controls. All primer sequences are listed in Supplementary Table S1.

The cDNA and gDNA PCR products were investigated using 4% agarose gel electrophoresis with TBE running buffer. DNA was separated by electrophoresis at 110 V for 30 minutes. The DNA marker used was DL600 (KeyGen, Nanjing). The bands were examined by UV irradiation.

The subcellular localization of circPRKCI was detected using the PARIS Kit according to the manufacturer's protocol (Ambion, Life Technologies).

The siRNAs were provided by Life Technologies. The miRNA mimics and primers were provided by RiboBio. The full-length cDNA of human circPRKCI was synthesized by Invitrogen and cloned into the expression vector pCDNA3.1 (Clontech Laboratories, Inc.). The final construct was verified by sequencing. Plasmid vectors for transfection were prepared using DNA Midiprep Kits (E.Z.N.A Endo-Free Plasmid Mini Kit ΙΙ) and transfected into lung adenocarcinoma cells using Lipofectamine 3000 (Invitrogen). The siRNAs and miRNA mimic were transfected into lung adenocarcinoma cells using RNAiMAX (Invitrogen) according to the manufacturer's instructions. All siRNA sequences used are listed in Supplementary Table S1.

Cell proliferation was examined using a CCK-8 Kit (Roche Applied Science), EdU assay (RiboBio), and Real time xCELLigence analysis system (RTCA) following the research protocol afforded by the manufacturer (Roche Applied Science and ACEA Biosciences; ref. 27). Colony formation assays were performed to monitor lung adenocarcinoma cell cloning capability. Lung adenocarcinoma cells were transfected with si-circPRKCI or negative control (NC) and analyzed on a flow cytometer (FACScan; BD Biosciences) equipped with CellQuest software (BD Biosciences). Gefitinib (SML1657, Sigma) was dissolved in DMSO and used for in vitro studies at concentrations not exceeding 0.1% DMSO.

The EZMagna RIP Kit (Millipore) was used following the manufacturer's protocol. Lung adenocarcinoma cells were lysed in complete RNA immunoprecipitation (RIP) lysis buffer, and the cell extract was incubated with magnetic beads conjugated with anti-Argonaute 2 (AGO2) or control anti-IgG antibody (Millipore) for 6 hours at 4°C. The beads were washed and incubated with Proteinase K to remove proteins. Finally, purified RNA was subjected to qRT-PCR analysis.

The biotin-coupled miRNA pull-down assay was performed as described previously by Zheng and colleagues (16). Briefly, the 3′ end biotinylated miR-RNA mimic or control biotin-RNA (RiboBio) was transfected into SPC-A1 cells at a final concentration of 20 nmol/L for 1 day. The biotin-coupled RNA complex was pulled down by incubating the cell lysate with streptavidin-coated magnetic beads (Ambion, Life Technologies). The abundance of circPRKCI and E2F7 in bound fractions was evaluated by qRT-PCR analysis.

The E2F7-binding sites of miRNA were predicted by TargetScan (http://www.targetscan.org/vert_71/). The different fragment sequences were synthesized and then inserted into the pGL3-basic vector (Promega). All vectors were verified by sequencing, and luciferase activity was assessed using the Dual Luciferase Assay Kit (Promega) according to the manufacturer's instructions.

Female BALB/c nude mice (4 weeks old) were maintained under specific pathogen-free conditions and manipulated according to protocols approved by the Nanjing Medical Experimental Animal Care Commission. NC and si-circPRKCI transfected SPC-A1 cells were harvested. For the tumor formation assay, 1–2 × 10⁶ cells were subcutaneously injected into a single flank of each mouse. Tumor growth was examined every week, and tumor volume was calculated using the following equation: V = 0.5 × D × d² (V, volume; D, longitudinal diameter; d, transverse diameter).

Animal experiments were conducted in accordance with the Institute for Laboratory Animal Research Guide for the Care and Use of Laboratory Animals and followed protocols approved by the Animal Committee of Nanjing Origin Biosciences. BALB/c male nude mice, ages 4 to 6 weeks and weighing 20 to 25 g, were purchased from the Beijing Vital River Laboratory Animal Technology Co., Ltd. All animals were fed an autoclaved laboratory rodent diet.

Primary lung adenocarcinoma samples were cut into approximately 1-mm³ fragments in 0.1 mL 50% Matrigel Basement Membrane Matrix (BD Biosciences) and directly implanted into the subcutaneous space (n = 5 for each tumor sample). Patient-derived tumor xenografts (PDTX) were harvested and divided into three portions for the generation of the second in vivo passage xenograft tumors, protein and DNA/RNA extraction, and histopathologic examination. Mice with palpable tumors (∼200 mm³) were randomly divided into two experimental groups with intratumoral injection twice weekly for two weeks: 1.5 mg/kg control siRNA and si-circPRKCI. At the end of the experiment, tumors were weighed and processed for knockdown activity by qRT-PCR and further histologic analysis.

Western blot analyses were performed according to standard protocols as described previously (26). Anti-ACTB, anti-E2F7, and anti-Hedgehog acyltransferase (HHAT) were purchased from Abcam. Anti-CDKN1A, anti-CCND1, anti-KI67, and anti-SOX2 were purchased from Cell Signaling Technology.

Microarray data have been submitted to GEO database and can be accessed with the ID: GSE101586.

Differences between groups were assessed by a paired two-tailed t test. One-way ANOVA or the nonparametric Kruskal–Wallis test was applied to assess the relationship between circPRKCI expression and other characteristics. The strength of the association between continuous variables was tested with the Spearman correlation. Multivariate Cox regression was used to identify factors associated with survival of lung adenocarcinoma. All statistical analyses were performed using SPSS 20 software (Abbott Laboratories).

We performed microarrays to characterize the expression profiles of circRNAs and mRNAs in paired lung adenocarcinoma tissues and adjacent nontumor tissues from 5 patients with lung adenocarcinoma. A total of 107 circRNAs (P < 0.05 and fold change > 1.5) and 1,691 mRNAs (P < 0.05 and fold change > 2.0) were differentially expressed between the lung adenocarcinoma tumor tissues and paired adjacent normal tissues. Among the 107 differentially expressed circRNAs (Supplementary Table S2), 28 were upregulated and 79 were downregulated in lung adenocarcinoma tissues compared with nontumor tissues (Fig. 1A). These circRNAs and their host genes are located at diverse genomic regions. Notably, Circos plot showed that the expression levels of circRNAs did not correlate with the mRNA levels of their host genes (Fig. 1B).

To further screen candidate circRNAs, we extracted the previously published lung adenocarcinoma–associated somatic CNV data from TCGA (24) and found the 50 most frequently recurrent variation regions (both P and residual P < 0.05), which included 20 amplified (689 genes) and 30 deleted (2160 genes) regions (Supplementary Table S3). By intersecting the amplified regions with host genes of upregulated circRNAs within these loci, and the deleted ones with downregulated circRNAs, respectively, we identified two amplified host genes (PRKCI and P4HB) and three deleted host genes (RABL2B, FLI1, and TM4SF19; Fig. 1C). Because PRKCI has been previously demonstrated as an important stemness maintainer of lung cancer (28, 29) in 3q26.2 amplicon, which is one of the most frequent genomic aberrations in lung cancer (30), we therefore hypothesized that PRKCI gene generated circRNA (Hsa_circ_0067934, termed as circPRKCI below) might also play an important role in lung adenocarcinoma and eventually chose circPRKCI for further investigation.

circPRKCI is back-spliced of two exons (exons 15 and 16) of PRKCI gene (chr3: 170013698–170015181), located at 3q26.2 amplicon (Fig. 1D). We firstly verified its existence in many circRNA databases. According to the circBase database, circPRKCI is detected in many types of cancer cell lines, including K562 and A549 (http://www.circbase.org/cgi-bin/singlerecord.cgi?id=hsa_circ_0067934). The circNet database also supported the existence of circPRKCI (http://circnet.mbc.nctu.edu.tw/ by searching circPRKCI).

To further characterize circPRKCI, we designed two sets of primers: the divergent primers were used to amplify the circular transcripts and the convergent primers were used to detect the linear transcripts. The two sets of primers were then used to amplify the circular and linear transcripts of PRKCI in both cDNA and gDNA. PCR results indicated that the circular form was amplified using the divergent primers in cDNA but not gDNA. Convergent primers amplified in both cDNA and gDNA. GAPDH was used as a linear RNA control (Fig. 1E). RNase R is a 3′ to 5′ exoribonuclease that degrades linear RNA but does not act on circular RNA. As was expected, the linear transcripts of PRKCI were degraded by RNase R, whereas the circular transcripts of circPRKCI were resistant to RNase R treatment (Fig. 1F). Indeed, these data confirmed the existence of circPRKCI.

The expression of circPRKCI was detected in 48 pairs of primary lung adenocarcinoma and adjacent nontumor tissues using qRT-PCR. circPRKCI was highly upregulated in lung adenocarcinoma, with an average fold of 6.89 (P < 0.01; Fig. 2A). We next evaluated the association between circPRKCI and clinical and pathologic parameters (Supplementary Table S4). Patients with larger tumor size exhibited higher expression of circPRKCI (P = 0.001; Fig. 2B). Patients with tumor–node–metastasis (TNM) stage II–III exhibited higher circPRKCI expression than patients with TNM stage I (P = 0.009; Fig. 2C). Taken together, circPRKCI is upregulated in lung adenocarcinoma tissues and positively correlated with tumor size and TNM stage.

CircPRKCI expression was then detected in lung cancer tissues by CISH using TMA of 89 pairs of lung adenocarcinoma and adjacent nontumor tissues. There was also a positive correlation between circPRKCI expression and T stage and TNM stage (Fig. 2D; Supplementary Table S5). Kaplan–Meier survival curves showed that patients with higher levels of circPRKCI had a shorter overall survival [HR = 1.977; 95% confidence interval (CI), 1.153–3.579; P = 0.037; Fig. 2E]. Multivariate analyses indicated that high circPRKCI level is an independent poor prognosis factor for patients with patients with lung adenocarcinoma (HR = 2.664; 95% CI, 1.327–5.347; P = 0.006; Fig. 2F; Supplementary Table S6).

To determine whether PRKCI gene amplification increases circPRKCI expression, we first examined the PRKCI copy number and circPRKCI expression in lung adenocarcinoma cell lines. Compared with the internal control, CEP3, we observed amplification of the PRKCI gene in A549, SCPA1, and H1299 cells (Fig. 2G). Accordingly, the expression of circPRKCI in the three cell lines was significantly higher than that in other cell lines (Fig. 2H). Next, we examined PRKCI copy number variation and circPRKCI expression in 60 pairs of lung adenocarcinoma tumor tissues. The PRKCI copy number was amplified in tumor tissues compared with paired nontumor tissues and positively correlated with circPRKCI expression (R² = 0.4366, P < 0.01; Fig. 2I). These results suggested that high expression of circPRKCI is at least partially due to PRKCI gene amplification.

To investigate the biological function of circPRKCI, we designed two siRNAs: One is circPRKCI siRNA (si-circPRKCI), which specifically targets the back-splice junction site of circPRKCI, and the other is PRKCI siRNA (si-PRKCI), which targets the exon 18 of the linear transcript of PRKCI only (Fig. 3A). For ectopic overexpression of circPRKCI, exon 15 and exon 16 of PRKCI were cloned into expression vectors, together with upstream and downstream flanking introns to promote the formation of circPRKCI as in a previous study (Fig. 3A; ref. 31).

Compared with the negative control siRNA, the expression of circPRKCI was only downregulated by si-circPRKCI but was not affected by si-PRKCI (Fig. 3B). Also, the expression vector markedly increased the expression of circPRKCI compared with the empty vector (Fig. 3C). By using Cell Counting Kit-8 (CCK-8), colony formation, and 5-ethynyl2′deoxyuridine (EdU) proliferation assays, we determined that knockdown of circPRKCI greatly impaired the proliferation ability of SPC-A1 cells, whereas ectopic expression of circPRKCI promoted cell proliferation (Fig. 3D–F). We further evaluated whether circPRKCI affects apoptosis or cell-cycle progression of SPC-A1 cells. After treatment with si-circPRKCI, lung adenocarcinoma cells were arrested at G₁ phase (Fig. 3G), but circPRKCI knockdown did not affect apoptosis. Terminal dexynucleotidyl transferase–mediated dUTP nick end labeling (TUNEL) experiment also confirmed that knockdown of circPRKCI did not affect apoptosis in lung adenocarcinoma cells. The Matrigel assay showed that si-circPRKCI treatment markedly impaired the invasion capacity of SPC-A1 cells (Fig. 3H). All of the above experiments were also repeated in A549 cells, and similar results were achieved (Supplementary Fig. S1A–S1L). These in vitro experiments suggested that circPRKCI promotes the proliferation and migration of lung adenocarcinoma cells.

To explore the molecular mechanisms of circPRKCI-promoting lung adenocarcinoma tumorigenesis, we first determined the subcellular localization of circPRKCI in lung adenocarcinoma cell lines using the nuclear mass separation assay (Fig. 4A) and FISH analysis (Fig. 4B). We found more than 90% of circPRKCI was present in the cytoplasm.

Given that many circRNAs can function as miRNA sponges in the cytoplasm (32), we determined whether circPRKCI may also bind to miRNAs as a sponge and regulate targets via the competitive endogenous RNA (ceRNA) mechanism. We therefore analyzed the sequence of circPRKCI using the miRanda algorithm and identified five miRNA-binding sites with relatively high scores (miR-545, miR-589, miR-600, miR-144, and miR-670; Supplementary Fig. S2A). It is well known that miRNAs usually silence gene expression by combining with the AGO2 protein and form the RNA-induced silencing complex (RISC). In the context of ceRNA mechanism, it might be a prevalent phenomenon that AGO2 could bind with both circRNAs and miRNAs based on previous studies (16, 33, 34). We therefore conducted an RIP assay to pull down RNA transcripts that bind to AGO2 in SPC-A1 and A549 cells. Indeed, endogenous circPRKCI was efficiently pulled down by anti-Ago2 (Fig. 4C). To further detect whether circPRKCI could sponge miRNAs, we performed a miRNA pull-down assay using biotin-coupled miRNA mimics (miR-545, miR-589, miR-600, miR-144, and miR-670). Interestingly, circPRKCI was only efficiently enriched by miR-589 and miR-545, but not by the other three miRNAs (Fig. 4D). In addition, RIP assay revealed that miR-545 and miR-589 were efficiently pulled down by the anti-AGO2 antibody but not by the nonspecific anti-IgG antibody (Fig. 4E). Furthermore, silencing of circPRKCI did not affect the expression of miR-545 or miR-589, and transfection of miR-545 and miR-589 mimics did not affect the expression of circPRKCI (Fig. 4F and G), which indicated circPRKCI functions as an miRNA sponge without affecting the expression of sponged miRNAs.

Because there are few reports about miR-545 and miR-589, we next investigated the biological functions of miR-545 and miR-589. We found that they significantly inhibit the proliferation of lung adenocarcinoma cells and induce a G₁ phase arrest (Supplementary Fig. S2B–S2E). These results suggest that circPRKCI functions according to the ceRNA mechanism and serves as a sponge for miR-545 and miR-589, both of which suppressed the tumorigenesis of lung adenocarcinoma cells.

To identify the potential target genes of miR-545 and miR-589, we filtered 958 (total context++ score < −0.17) and 781 (total context++ score < −0.18) genes, respectively, by using the TargetScan prediction program. According to the ceRNA theory, circPRKCI expression is positively correlated with its target genes. Next, we filtered genes that positively correlated with circPRKCI expression in our microarray data. By plotting a Venn diagram using the 3 gene sets, we identified 7 candidate target genes (E2F7, CENPA, DNAJC3, FRK, LRRC17, MON2, and RMI2; Fig. 4H). To further verify the downstream targets of circPRKCI, mRNA levels of 7 candidate target genes were detected after silencing circPRKCI, and we found only E2F7 was downregulated (Fig. 4I). In addition, literature review suggests E2F7 is closely involved in tumorigenesis; we therefore reasoned that E2F7 may be a downstream target of circPRKCI.

To verify whether E2F7 was the direct target of miR-545 and miR-589, we first performed the miRNA biotin pull-down assay. We found that both miR-545 and miR-589 could significantly enrich the 3′UTR of E2F7 mRNA (Fig. 4J). After transfection of either miR-545 or miR-589 mimics, both the mRNA and protein levels of E2F7 were significantly decreased (Fig. 4K and M). Furthermore, we cloned the wild-type and mutant (predicted miR-545 and miR-589–binding sites were mutated) 3′-UTR of E2F7 mRNA and performed a dual luciferase reporter assay. Compared with the control RNA group, both miR-545 and miR-589 mimics efficiently inhibited luciferase activity of wild-type group but not mutant one. However, after mutating the binding sites, the inhibitory effect was abolished (Fig. 4L). These results suggested that both miR-545 and miR-589 bind to the 3′-UTR of E2F7 and directly downregulate E2F7 expression.

E2F7 was previously identified as a negative regulator of CDKN1A transcription and subsequently upregulated downstream CCND1 so as to induce cell-cycle arrest (35, 36). We confirmed that both miR-545 and miR-589 mimics upregulated CDKN1A while downregulating CCND1 as compared with control RNA (Fig. 4M). In addition, by analyzing the RNA sequencing and miRNA microarray data derived from TCGA, we found expression levels of both miR-545 and miR-589 are negatively correlated with E2F7 (Fig. 4N). It has been reported that E2F7 promotes cancer cell proliferation (35). In support of this finding, silencing of E2F7 indeed inhibited lung adenocarcinoma cell proliferation and induced G₁ phase arrest (Supplementary Fig. S3A–S3D). Taken together, these data indicated that both miR-545 and miR-589 could directly downregulate E2F7 and thereby regulate its downstream signaling.

To validate whether circPRKCI promotes cell proliferation via the circPRKCI-miR-545/589-E2F7 axis, we first confirmed that silencing circPRKCI decreased the protein levels of E2F7, whereas overexpressing circPRKCI increased the protein levels of E2F7. The protein levels of downstream CDKN1A and CCND1 were also altered accordingly (Fig. 5A). Second, in the expression cohort of 48 patients with lung adenocarcinoma, circPRKCI expression was positively correlated with mRNA levels of E2F7 (R² = 0.49, P < 0.01; Fig. 5B). Then, we designed rescue experiments using miRNA-545 and miR-589 mimics (Fig. 5C). At protein level, miR-545 and miR-589 partially reversed the effects of circPRKCI on E2F7, CDKN1A, and CCND1 in SPC-A1 cell (Fig. 5D). More importantly, as revealed by colony formation, CCK-8, and EdU assays, both the miR-545 and miR-589 mimics could partially rescue the proliferation-promoting effect induced by circPRKCI and the combination of two miRNA mimics showed a stronger rescue effect (Fig. 5E; Supplementary Fig. S4A–S4C).

To further determine whether circPRKCI exerts the proliferation-promoting effect dependent on the miR-545/589–binding sites, we constructed circPRKCI vectors with mutated binding sites of miR-545 (circPRKCI Mut1), miR-589 (circPRKCI Mut2), and both binding sites (circPRKCI Mut1+2), respectively. Colony formation and EdU assay showed that mutation of the miR-545 or miR-589–binding site could partially reverse the cell proliferation induced by circPRKCI, and mutation of both miR-545 and miR-589–binding sites completely abolished the enforced cell proliferation induced by circPRKCI (Fig. 5F; Supplementary Fig. S5A–S5C). Taken together, we demonstrated that circPRKCI promotes lung adenocarcinoma cell proliferation via the circPRKCI–miR-545/589–E2F7 axis.

PRKCI and SOX2 are coamplified in 3q26. Previous work (28) has demonstrated PRKCI could regulate SOX2-mediated HHAT expression, a key component of Hedgehog pathway. PCR in lung adenocarcinoma tumor tissues showed circPRKCI expression was not correlated with PRKCI (R² = 0.0026, P = 0.72) or SOX2 (R² = 0.049, P = 0.170; Supplementary Fig. S6A and S6B). SOX2 and HHAT expression were not altered at the protein level after silence or overexpression of circPRKCI (Supplementary Fig. S6C and S6D). To see whether SOX2 has synergistic effect to circPRKCI or not, we silenced circPRKCI and SOX2 alone or both in lung adenocarcinoma cells and performed colony formation and Edu assay. The results showed that silencing circPRKCI or SOX2 alone inhibited cell proliferation, while silencing both somehow enforced the inhibitory effect (Supplementary Fig. S6E–S6H). These lines of evidence suggest circPRKCI might not function through PRKCI/SOX2 signaling pathway.

To investigate the biological function of circPRKCI in vivo, we established a xenograft tumor model in nude mice. SPC-A1 cells transfected with si-circPRKCI or a control siRNA were subcutaneously injected into nude mice. The tumors derived from cells transfected with si-circPRKCI had a smaller size and lower weight compared with those derived from cells transfected with the control siRNA (Fig. 6A). IHC revealed that tumor tissues collected from the si-circPRKCI group had fewer E2F7 and CCND1-positive cells but more CDKN1A-positive cells when compared with control group (Fig. 6B). In addition, it is well known that PDTXs maintained better cell differentiation ability, morphology, and architecture of the original patient tumors compared with the cell line–derived xenografts; therefore, PDTX is considered as a translational model for cancer research (37). We then developed a PDTX model from a female patient with lung adenocarcinoma and evaluated the therapeutic potential of circPRKCI by intratumoral injection of cholesterol-conjugated si-circPRKCI and a control siRNA (4 times and twice a week). As a result, treatment of si-circPRKCI significantly inhibited growth of PDTX in vivo (Fig. 6C), suggesting that circPRKCI could serve as a promising therapeutic target of lung adenocarcinoma.

Because EGFR tyrosine kinase inhibitors (EGFR-TKI) have been widely used in patients with lung adenocarcinoma with EGFR mutations, to find whether circPRKCI could influence the therapeutic effects of EGFR-TKI, we performed proliferation assay in HCC827 cells (with EGFR E746-A750 deletion, which is sensitive to EGFR-TKI) by using RTCA system. As was expected, either EGFR-TKI (gefitinib) alone or silencing circPRKCI alone showed cytotoxic effect in HCC827 cells. Combining gefitinib with silencing circPRKCI showed remarkably stronger inhibitory effect compared with gefitinib or si-circPRKCI alone (Fig. 6D), which suggested that combined inhibition of EGFR and circPRKCI might have potential synergistic inhibition effect.

Recent high-throughput sequencing studies have demonstrated that the human genome is actively transcribed, and noncoding transcripts are abundant in the human transcriptome. circRNAs are a type of noncoding RNAs that have recently attracted great research interest. The majority of circRNAs are generated by exon circularization from precursor RNAs; thus, the expression of circRNAs would be affected by genetic alterations, like translocation and CNV (22). Guarnerio and colleagues have proved that chromosomal translocations give rise to fusion circRNAs and the fusion circRNA generated by MLL/AF9 translocation is oncogenic and promotes leukemia progression (38). However, the function of deregulated circRNAs caused by CNVs remains largely unknown.

In the current study, we filtered several deregulated circRNAs that are localized in recurrent CNV locus in lung adenocarcinoma by integrated analysis of microarray and TCGA data. Among them, we characterized that circPRKCI is a highly upregulated circRNAs in lung adenocarcinoma due to the amplification of 3q26.2 locus. During the period of investigating circPRKCI in lung adenocarcinoma, we interestingly found circPRKCI was also upregulated in esophageal squamous cell carcinoma (39). However, the molecular mechanism of circPRKCI, as well as the correlation between increased circPRKCI expression and 3q26 amplification, remains unclear at that moment. circPRKCI is derived from exons 15 and 16 of its host gene PRKCI, which is a known proto-oncogenic gene in lung cancer that could drive a stem-like phenotype by competing with SOX2 (28). It has been proven that amplification of the 3q26.2 locus leads to increased PRKCI expression (40), and our data also demonstrated that PRKCI gene amplification was positively correlated with circPRKCI expression. In addition to lung cancer, accumulating evidences have revealed the amplification of 3q26.2 is frequently identified in multiple cancers, such as esophageal cancer (41), ovarian cancer (42), thyroid cancer (43), and testicular germ cell tumor (44). At this region, many proto-oncogenic genes have been identified, including not only coding genes like PRKCI, SEC62 (45), and MDS1/EVI1 (46), but also some noncoding gens like miR-551b-3p (42) and miR-569 (13). These data indicated the noncoding transcripts at the 3q26.2 amplicon also play an important role in cancer development. On the other hand, gene amplification leads to overexpression of both linear and circular transcripts. Considering that many oncogenes generate circular transcripts, like circFOXO3 (14), it is plausible to hypothesize that circular transcripts of those oncogenes may also play a role in the pathogenesis of cancer independent of their linear transcripts.

The ceRNA hypothesis suggests that RNA transcripts, including mRNAs, lncRNAs, pseudogenes, and circRNAs, could communicate with each other via miRNAs. They usually bind to miRNAs and then regulate expression of RNA transcripts harboring the same miRNA-binding sites, constructing a complex posttranscriptional regulatory network (47). circRNAs were first observed decades ago; however, their functional roles in biological processes have not been well characterized until recently (48). Owing to its biological stability, many researchers have shown that circRNAs are perfect miRNA sponges. ciRS-7 is the first circRNA that is confirmed as a miRNA sponge, which contains more than 70 binding sites of miR-7, and is involved in the development of brain (19). They demonstrated “ciRS-7 serves as a binding platform for Ago2 and miR-7” and “the widespread Ago occupancy is caused by miR-7-directed association of AGO2 proteins to the prevalent and conserved miR-7 target sites in the ciRS-7 RNA.” Thereafter, many other studies have also validated that circRNA could bind with AGO2 and “sponge” miRNAs (16, 33, 34). In the current study, we also found Ago2 could pull both miR-545/589 and circPRKCI. Taking the literatures and our data together, it might be a broad phenomenon that both circRNAs and miRNAs could bind with AGO2, despite the underlying mechanisms need further investigation.

A growing number of studies have shown that several circRNAs that can sponge miRNAs have been reported in various types of cancer. For an instance, circHIPK3 is upregulated in multiple solid cancers and binds to miR-124 so as to inhibit cancer growth (16). Subsequent studies found that circHIPK3 could also bind to miR-558 (49) and miR-30a-3p (50). Other circRNAs with similar ceRNA functions have also been reported, such as circMTO1 binding to miR-9 (20), as well as circCCDC66 binding to miR-33b and miR-93 (21). It suggests that one circRNA could bind to more than one miRNA, and consistent with these findings, our data showed circPRKCI binds to both miR-545 and miR-589. Considering the relative low abundance of circRNAs and various miRNA-binding sites within mRNA and circRNA transcripts, sponging multiple miRNAs by a single circRNA could be a general mechanism in cells.

Finally, we also evaluated the clinical and translational relevance of circPRKCI. Results of RT-PCR and CISH both showed circPRKCI is highly expressed in lung adenocarcinoma tissues, and further analyses indicated high expression of circPRKCI correlates with higher tumor grade and poor prognosis of lung adenocarcinoma. These lines of evidence demonstrate that circPRKCI is a potential biomarker for lung adenocarcinoma. Considering the translational significance of circPRKCI, in the PDTX model, intratumor injection of siRNA inhibited lung cancer growth in vivo. Moreover, in EGFR-mutant HCC827 cells, silence of circPRKCI greatly increased the inhibitory effect of gefitinib. These data indicate that circPRKCI may be a potential therapeutic target for lung adenocarcinoma.

In summary, as shown in Fig. 7, we identified a CNV-associated circular RNA, circPRKCI, which is overexpressed in lung adenocarcinoma tissues, at least in part due to the amplification of 3q26.2 locus, promotes proliferation and tumorigenesis of lung adenocarcinoma. circPRKCI binds to both miR-545 and miR-589 and subsequently inhibits their suppressing capability on E2F7. Our study provides a novel potential target for patients with lung adenocarcinoma in the sight of circular RNAs.

No potential conflicts of interest were disclosed.

Conception and design: M. Qiu, W. Xia, R. Yin, Jun Wang, L. Xu

Development of methodology: M. Qiu, W. Xia, R. Chen, S. Wang, J. Hu

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): S. Wang, Y. Xu, Z. Ma, T. Fang, J. Hu

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M. Qiu, W. Xia, S. Wang, Y. Xu, W. Xu, E. Zhang, R. Yin, Jun Wang, L. Xu

Writing, review, and/or revision of the manuscript: M. Qiu, S. Wang, Z. Ma, R. Yin, Jun Wang, L. Xu

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): Jie Wang, J. Hu, G. Dong, L. Xu

Study supervision: R. Yin, Jun Wang, L. Xu

This work was supported by the National Natural Science Foundation of China (81372321, 81472200, 81572261, and 81702256), Innovation Capability Development Project of Jiangsu Province (BM2015004), Project of Jiangsu Provincial Medical Talent (ZDRCA2016033), and the Key Project of Cutting-edge Clinical Technology of Jiangsu Province (BE2016797). M. Qiu was supported in part by the Postdoctoral Fellowship of Peking-Tsinghua Center for Life Sciences. We greatly appreciate Dr. Mark K. Ferguson's (Department of Surgery and the Cancer Research Center, the University of Chicago Medicine & Biological Sciences) kind help for manuscript language editing.

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.