Introduction

Gastric cancer is one of the leading causes of cancer mortality and morbidity, especially in East Asian countries [1,2,3]. Poor prognosis of unresectable gastric cancer along with limited available options for targeted therapeutic agents necessitates better understanding of the pathogenesis [4,5,6].

Adenosine deaminase acting on RNA 1 (ADAR1) is known to bind to double-stranded RNA (dsRNA) and induce A-to-I RNA editing, which is a process that modifies nucleotide sequences of RNA transcripts from adenosine-to-inosine in the absence of mutations in genomic DNA [7,8,9]. Because inosine is recognized as guanosine by the translational or splicing machinery, A-to-I editing of RNA can result in incorporation of different amino acids, alternative splicing, a gain or loss of binding sites for microRNA (miRNA), and subsequent alteration in protein and/or messenger RNA (mRNA) expression [10,11,12,13]. In addition, ADAR1 modulates miRNA processing through interactions with DICER or by affecting miRNA processing in an editing-independent manner, which in turn triggers complex gene expression changes [14, 15].

ADAR1 has various biological functions, including suppression of interferon signaling [16], prevention of autoimmune diseases by endogenous dsRNAs [17], maintenance of hematopoietic stem cells [18], and facilitation of responses to various stimuli, including heat shock, ultraviolet rays, and a viral infection [19]. Furthermore, an ADAR1 knockout is embryonically lethal in mice, indicating the indispensable roles of this enzyme in developmental processes [20].

Recently, multiple studies showed an oncogenic potential of ADAR1 and RNA editing in various cancer types [21, 22]. Nonetheless, these studies were mainly focused on the editing of coding regions, which are not the main editing targets of ADAR1, leaving ambiguity about the influence of this enzyme on the editing of 3′ untranslated regions (UTRs), which are main editing targets by being enriched in Alu repeats [23]. Furthermore, the influence of ADAR1 on the editing or biogenesis of miRNA is still unclear. We therefore sought to determine the editing-dependent and editing-independent functions of ADAR1 in gastric cancer, focusing on 3′-UTR editing and subsequent changes in expression of mRNAs as well as miRNAs.

Materials and Methods

Cell Cultures

AGS and MKN-45 human gastric cancer-derived cell lines and HEK-293T human embryonic kidney cells were purchased from the Korean Cell Line Bank (Seoul, Korea). AGS and MKN-45 were grown in the RPMI 1640 medium with 10% of fetal bovine serum. HEK-293T cells were grown in DMEM containing 10% of fetal bovine serum. All the cells were maintained at 37 °C and 5% CO2.

Cloning and Site-Directed Mutagenesis

Coding regions of the ADAR1 isoforms p110 and p150 were polymerase chain reaction (PCR)-amplified from an ADAR1 complementary DNA (cDNA) template or genomic DNA of AGS cells, and each PCR product was inserted into the pFLAG-CMV2 vector. The 3′-UTRs of phosphatase and actin regulator 4 (PHACTR4) and nuclear transcription factor Y subunit α (NFYA) were PCR-amplified from genomic DNA of AGS cells, and the amplicons were ligated into the pMIR-REPORT™ Luciferase vector (Applied Biosystems, CA, USA). Mutagenesis of the edited site (A-to-G) of PHACTR4 3′-UTR and of the seed-binding sequence of an NFYA 3′-UTR reporter was performed using QuikChange™ Site-Directed Mutagenesis Kit (Stratagene, CA, USA). The primer sequences for cloning and mutagenesis are shown in Supplementary Table 1.

Cell Transfection

To stably knock down ADAR1 in AGS and MKN-45 cells, HEK-293T cells were plated at a density of 1.5 × 105 cells/well in six-well plates and were incubated for 24 h. HEK-293T cells were cotransfected with the inhibitory plasmid (shADAR1, 1500 ng), the packaging plasmid psPAX2 (700 ng), and the envelope plasmid pMD2.G (700 ng) with Lipofectamine 2000 (Invitrogen, CA, USA). Viral supernatants were then harvested at 24–48 h after transfection and were used to transfect AGS and MKN-45 cells. The efficiency of 293T cell transfection was estimated by cotransfection with green fluorescent protein. To knock down the ADAR1 p150 isoform, short hairpin RNA (shRNA) targeting p150 isoform-specific sequences was employed (shADAR1-p150). To overexpress ADAR1 in AGS cells, 1500 ng of an ADAR1 expression vector encoding the p110 or p150 isoform was transfected into the AGS cell line, and the cells were harvested 48 h after the transfection.

Preparation of Human Stomach Samples

Human gastric cancer samples and matched normal tissues from 16 patients were obtained from the Bio Resource Center (BRC) of Asan Medical Center, Seoul, Korea. The use of human tissue samples was approved by the Institutional Review Board of Asan Medical Center (IRB no: 2016-0931).

RNA Extraction, cDNA Synthesis, and Real-Time PCR

Total RNA was extracted by means of the TRIzol Reagent (Invitrogen, CA, USA). cDNA was synthesized with the PrimeScript First Strand cDNA Synthesis Kit (Takara, Japan) and subjected to quantitative polymerase chain reaction (qRT-PCR) analysis. The primer sequences for PCR are shown in Supplementary Table 2. The human glyceraldehyde 3 phosphate dehydrogenase (GAPDH) served as an internal control, and relative quantification was conducted by the 2−ΔΔCT method [24]. Quantitative assessments of miRNAs were performed on TRIzol-isolated RNA samples after reverse transcription with PrimeScript Kits (Heimbiotek, Korea), and cDNAs were analyzed using HB miR Multi Assay Kit™ System I (Heimbiotek, Korea).

Detection of Editing in RNA-Sequencing Data

The RNA-sequencing (RNA-Seq) data were generated on the Illumina HiSeq™ 2500 platform with Illumina TruSeq RNA Library Preparation Kits. The cleaned reads were aligned to the hg19 reference genome in the STAR software (v.2.5). After removal of PCR duplicates, variants were called in SAMtools (v.1.5). A series of quality filters were then applied to minimize the false-positive rate. Next, we dichotomized editing candidates into Alu and non-Alu regions. Additional information and detailed explanations are provided in Supplementary Methods.

Detection of Editing in miRNA-Seq Data

The small RNA-Seq was carried out by Theragen Etex Bio Institute (Suwon, Korea) on the Illumina platform (NEXTflex Illumina Small RNA-sequencing Kit v.3). The cleaned reads were mapped to the hg19 reference genome and to known pre-miRNA sequences (miRBase, release 21) in the Bowtie software (v.2.3.3). All variants were analyzed using prebuilt Perl scripts, and sequencing errors that cause false-positive calls were excluded by binomial statistics analysis [25]. In addition, we filtered known single-nucleotide polymorphisms (dbSNP build 138). Additional information and detailed explanations are provided in Supplementary Methods.

Sanger Sequencing

Editing rates obtained from RNA-Seq data were validated by PCR amplification and Sanger sequencing of DNA and cDNA from the inferred sites. Peak heights of nucleotide-editing targets were measured by means of ab1 peak reporters (https://apps.lifetechnologies.com/ab1peakreporter), and editing rates were calculated by dividing the height of peak G by the sum of A and G peak heights in sequencing electropherograms. Primers for Sanger sequencing are presented in Supplementary Table 1.

Sodium Dodecyl Sulfate–Polyacrylamide Gel Electrophoresis (SDS-PAGE) and Immunoblot Analysis

The protein levels of ADAR1, ADAR2, phosphoribosylaminoimidazole carboxylase and phosphoribosylaminoimidazole succinocarboxamide synthase (PAICS), PHACTR4, peptidylprolyl isomerase A (PPIA), V-Set and immunoglobulin domain containing 1 (VSIG1), thioredoxin-interacting protein (TXNIP), and β-actin were analyzed by immunoblot assays. Lysates from cells or patients’ samples (100 μg/well) were separated by SDS-PAGE in a 10% gel and were transferred to nitrocellulose membranes (Amersham, UK). The membranes were then probed with primary antibodies. Each target protein’s levels were normalized to those of β-actin, and protein concentrations were determined using the Pierce™ BCA Assay kit (Thermo Fisher Scientific, CA, USA) with bovine serum albumin as a standard. All immunoblots were performed three or more times for each experiment. The primary antibodies for immunoblots are shown in Supplementary Table 3.

Immunohistochemistry (IHC) and Immunofluorescence (IF) Assays (Confocal Microscopy)

A positive IHC result was defined by a cutoff expression of 2+ and above. For the negative controls intended to evaluate nonspecific staining, the primary antibodies were replaced with rabbit polyclonal IgG (Cell Marque, CA, USA) or mouse monoclonal IgG (Santa Cruz Biotechnology, Dallas, TX, USA). Breast cancer tissue served as a positive control for ADAR1 and PAICS. For PHACTR4, lung cancer tissue was used (data not shown). Alternatively, IF assays for ADAR1, PAICS, and PHACTR4 were performed to determine changes in the localization of the proteins. Detailed procedures of IHC and IF assays are described in the Supplementary Methods section.

Luciferase Assay

AGS cells were seeded in 24-well plates, 24 h prior to transfection. On the following day, 100 ng of the pMIR-REPORT™ Luciferase vector (Applied Biosystems, CA, USA), a wild-type UTR–containing pMIR vector, or the pMIR vector containing a mutated seed sequence were cotransfected with 10 ng of the TK-luc plasmid by means of Lipofectamine 2000 (Invitrogen). A miRNA-148a-3p mimic, miRNA-196a-3p mimic, or scrambled RNA duplexes were also cotransfected when necessary (Genolution, Korea). The activity of each luciferase reporter was measured via the Dual Luciferase Reporter Assay system (Promega, WI, USA). Relative luciferase expression was determined as the ratio of firefly to Renilla luciferase activities.

Web-Based Bioinformatic Assay Tools

Relations between miRNA expression and target mRNA expression were determined by interactive visual analysis with an miRNA prediction tool that integrates and visualizes miRNA–mRNA expression profile data (miRTarVis) [26]. In addition, miRNA targets were predicted via microRNA.org [27], and alterations in binding sites after editing were predicted in mirEdiTar [28].

Statistics

Data are presented as means ± standard deviations of two or three independent experiments. Comparisons between two groups were performed by Student’s t test or Chi-square test, as appropriate. Differences were considered significant when P values were < 0.05. The SPSS statistical package for Windows (v.23; SPSS) served for all statistical analyses.

Data Availability

All raw and processed data are freely available from the GEO repository (https://www.ncbi.nlm.nih.gov/geo/) under accession number GSE106874.

Results

Stable ADAR1 Knockdown Induced Changes in RNA Editing Levels in AGS and MKN-45 Gastric Cancer Cell Lines

To investigate ADAR1-dependent editing targets and expression changes of mRNAs upon the knockdown, both p110 and p150 isoforms of ADAR1 were stably knocked down in AGS and MKN-45 cells by lentiviral infection, which targets the 3′-UTR of the ADAR1 mRNA. Both mRNA levels (Fig. 1A) and protein levels (Fig. 1B) of ADAR1 decreased significantly upon the knockdown in AGS and MKN-45 cells. In contrast, the ADAR2 mRNA level was unaffected, and the protein was not detectable. IF images revealed that ADAR1 was the most abundant in the nucleus, especially in the nucleolar region, and the amount decreased after the knockdown (Fig. 1C).

Fig. 1
figure 1

Identification and gross analysis of RNA editing sites in two gastric cancer cell lines. qRT-PCR (A), immunoblot (B), and IF (C) assays of ADAR1 and ADAR2 expression in AGS and MKN-45 cell lines infected with the shADAR1 or control (shGFP) virus. In (C), the green signal on the left indicates ADAR1 protein, and the middle one is Hoechst staining. A merged image is shown on the right. Genomic distribution of edited sites in the Alu region in AGS (D) or MKN-45 cells (E). The number of base changes that occurred in Alu repeats and the change caused by the ADAR1 knockdown, as measured in AGS (F) or MKN-45 cells (G)

Full size image

RNA-Seq Reveals a Characteristic Editing Pattern upon the ADAR1 Knockdown

We first focused on analysis of Alu regions, because these conserved repetitive sequences are known to be frequent targets of RNA editing [23]. In Alu region, editing most commonly occurred as the A-to-I type (98.2%) in the 3′-UTR (38.9%) or intron (34.4%) in the AGS cell line (Fig. 1D, F). Similarly, in Alu regions of the MKN-45 cell line, the A-to-I type constituted the bulk of editing (99.9%), which mostly involved either the 3′-UTR (26.4%) or an intron (42.4%, Fig. 1E, G). Upon the ADAR1 knockdown, a decrease in the number of editing candidates was observed throughout all genomic regions in which RNA editing occurred (Fig. 1D, E). Likewise, the number of candidates for A-to-I editing decreased by 76.2 and 66.7% in AGS and MKN-45 cells after the ADAR1 knockdown, respectively (Fig. 1F, G). Editing candidates in non-Alu regions of mRNA in the MKN-45 cell line were similar to those of Alu region, with predominance of 3′-UTR and introns (58.6%) and A-to-I editing (63.8%, Supplementary Fig. 1B, D). In contrast, non-Alu regions of the AGS cell line RNA showed a lower proportion of 3′-UTR and introns (49.6%) and of A-to-I editing (35.2%, Supplementary Fig. 1A, C). We next listed edited positions located in the 3′-UTR in each cancer cell line from the highest editing level (Supplementary Tables 4 and 5) and from the highest number of sites of editing within their 3′-UTR (Supplementary Tables 6 and 7) because the protein and/or mRNA expression of these candidates is more likely to be affected by RNA editing. In addition, genes highly or frequently edited in the 3′-UTR of their mRNA in both gastric cancer cell lines were also analyzed and were regarded as priority targets for confirmation of editing (Supplementary Tables 8 and 9).

Editing-Dependent Regulatory Effect of ADAR1 on the 3′-UTR of PHACTR4 mRNA

Because 3′-UTR is a well-known editing target region of ADAR1, we determined whether editing of 3′-UTRs of a specific gene would alter miRNA binding and subsequent protein and mRNA expression. Among edited targets that were identified in RNA-Seq analyses, PHACTR4 mRNA was the target that could explain the phenomenon. RNA-Seq data revealed that editing affected 54% of PHACTR4 transcripts in the 3′-UTR (chr1;28826190), and this percentage decreased to 15% after the ADAR1 knockdown (Fig. 2A). Indeed, Sanger sequencing of the site chr1;28826190 showed that 59.8% of control cDNA copies were affected by editing, and this percentage decreased to 26.1% after the ADAR1 knockdown, in the absence of mutations in genomic DNA (Fig. 2B). Subsequent qRT-PCR analyses of PHACTR4 expression showed a significant increase in mRNA expression upon the ADAR1 knockdown (Fig. 2C). Furthermore, a 1.6-fold increase in the PHACTR4 protein level was detected by immunoblotting (Fig. 2D, E, Supplementary Fig. 2A for repeated results), although no significant change was found when the p110 or p150 isoform of ADAR1 was overexpressed (Supplementary Fig. 2B, see “Discussion”). PHACTR4 localization did not change after the ADAR1 knockdown (Supplementary Fig. 2C). After that, to determine whether the increased PHACTR4 level was due to the p110 or p150 isoform of ADAR1, we transfected shRNA specifically targeting the ADAR1 p150 isoform (shADAR1-p150) but detected no alteration in the PHACTR4 level (Supplementary Fig. 2D). We next questioned whether this expression change is mediated by RNA editing, observed in Fig. 2A, B. The edited position in PHACTR4 mRNA 3′-UTR corresponding to 28826190 of chr1 was predicted to create a novel binding site for miRNA-196a-3p (Fig. 2F). We therefore performed a UTR luciferase reporter assay to confirm that the alteration of the seed-binding sequence in PHACTR4 mRNA 3′-UTR can affect PHACTR4 expression. Indeed, luciferase reporter assays of altered seed-binding UTR sequences of PHACTR4 mRNA revealed a significantly reduced signal from the A-to-G mutant owing to miRNA-196a-3p (Fig. 2G, H for the level of miRNA-196a-3p).

Fig. 2
figure 2

PHACTR4 is an editing-dependent target of ADAR1. A Raw RNA-Seq data of PHACTR4 (chr1;28826190) in control and ADAR1 knockdown AGS cells. The A-to-I editing position is boxed, indicating that the 54.4% editing level (31 guanosines among a total of 57 reads) in the control sample (left) and 15.4% (six guanosines of 39 sequence reads) in an ADAR1 knockdown sample (right). Visualization was conducted via the SAMtools’ tview function, which depicts sequences aligned in the reverse complement direction by means of lowercase letters. B Sanger sequencing results on PHACTR4 (chr1;28826190) in genomic DNA, control, and ADAR1 knockdown cDNA. Arrows indicate the A-to-I edited sites, which constitute 59.8% in control cDNA (middle), but this percentage diminished to 26.1% in shADAR1 cDNA (right). qRT-PCR (C), immunoblot (D), and their quantification data (E) after the ADAR1 knockdown. F A novel binding site for miRNA-196a-3p was predicted to be created after editing of the UTR at position 28826190 of PHACTR4 cDNA. A graph showing a luciferase assay performed with unedited [marked as UTR(A)] or edited [UTR(G)] reporter plasmids cotransfected with a miRNA-196a-3p mimic or scrambled control (G) and qRT-PCR results for miRNA-196a-3p in the same samples (H); n.s. not significant

Full size image

Of note, we observed that three other targets examined did not show significant alteration in their protein levels despite the obvious presence of ADAR1-dependent RNA editing and/or a change in the mRNA level. Sanger sequencing of the most highly edited site of PAICS mRNA (corresponds to chr4;57327087) in AGS cells revealed that 93.7% of control cDNA copies were edited, and this percentage decreased to 61.6% upon the ADAR1 knockdown, in the absence of mutations in genomic DNA (Supplementary Fig. 3A). Nevertheless, no detectable changes in expression levels or localization of PAICS were detected by qRT-PCR, immunoblotting, or IF assays of ADAR1 knockdown cells and ADAR1-overexpressing cells (Supplementary Fig. 3B–E). Similarly, the editing of mRNA corresponding to position chr7 44841489 in the PPIA gene did not alter protein expression although mRNA levels significantly decreased (Supplementary Fig. 4A–C). Likewise, the editing of mRNA corresponding to position chrX 107321144 in the VSIG1 gene did not alter protein expression although mRNA levels significantly decreased (Supplementary Fig. 4D–F).

ADAR1 Immunohistochemistry Scores Are Elevated in Gastric Cancer Specimens and Inversely Correlate with PHACTR4 Levels

Next, we investigated the effect of RNA editing on the expression of genes PHACTR4 and PAICS in patients’ samples. For this purpose, IHC staining of ADAR1, PHACTR4, and PAICS was performed on 16 gastric cancer tissue samples and their paired normal gastric mucosa samples. The clinical and histological details of the patients are described in Supplementary Table 10. Notably, the percentage of cases positive for ADAR1 expression was significantly higher in tumor samples than in paired normal samples (87.5 vs. 56.3%, P = 0.049, Supplementary Table 11, Fig. 3A, B, N). In contrast, the percentage of cases positive for PHACTR4 expression was lower in gastric cancer tissues than in matched normal tissues (18.8 vs. 56.3%, P = 0.028, Supplementary Table 11, Fig. 3G, H, N), in agreement with the data in Fig. 2. By contrast, the percentages of cases positive for PAICS expression did not differ significantly between tumorous and paired normal tissues (25.0 vs. 25.0%, P = 1.000, Supplementary Table 11, Fig. 3D, E, N). In normal gastric tissues, PHACTR4 was positive in 63.6% of body-type glands (Fig. 3L, M), but was negative in all antral-type glands (Fig. 3J, K), suggesting specific roles of this protein in the secretion of body-type gastric glands.

Fig. 3
figure 3

IHC analysis of ADAR1 and PHACTR4 reveals their inverse correlation in gastric tumor specimens. A–I Representative IHC, corresponding hematoxylin and eosin (H&E) staining, and negative control images of ADAR1 (A–C), PAICS (D–F), and PHACTR4 (G–I) in gastric tumor specimens (magnification ×200). J–M IHC and corresponding H&E images of PHACTR4 in the antral gland (J and K) and corpus gland (L and M) in normal gastric tissue (×200). N The IHC score of ADAR1, PAICS, and PHACTR4 in tumors and corresponding normal tissues (n = 16)

Full size image

Sequencing of miRNAs in Gastric Cancer Cell Lines Reveals miRNAs Edited by ADAR1

We next examined another mode of ADAR1-mediated gene expression change: via the regulation of miRNA. We detected miRNA editing using small RNA-Seq of control and ADAR1 knockdown AGS and MKN-45 gastric cancer cells. Overall, editing occurred at 84 and 81 positions of miRNAs of AGS and MKN-45 cells, respectively, with 33 positions of miRNAs being edited in both cell lines (Fig. 4A). Four miRNAs (miRNA-627, let-7a-1, let-7a-3, and miRNA-1260a) in AGS cells and two miRNAs (miRNA-1260a and let-7e) in MKN-45 cells contained edited sequences in their seed regions (Supplementary Table 12).

Fig. 4
figure 4

Small RNA-Seq in gastric cancer cell lines identifies miRNAs edited by ADAR1. A The Venn diagram showing the number of edited miRNAs in AGS and MKN-45 cell lines. B The number of miRNAs whose expression is changed after the ADAR1 knockdown in AGS (filled bars) and MKN-45 cells (open bars). Ten most highly up- or downregulated miRNAs in AGS (C) or in MKN-45 (D) cell lines upon the ADAR1 knockdown

Full size image

Integration of miRNA and mRNA Expression Changes by a Bioinformatic Analysis Identified a Few Hub miRNAs Affecting Expression of Multiple mRNAs

miRNA expression alterations due to the ADAR1 knockdown differed between the two cell lines. Upon the ADAR1 knockdown, in AGS and MKN-45 cells, expression of 56 and 53 miRNAs, respectively, increased by more than 50%, whereas expression of 56 and 24 miRNAs, respectively, decreased by more than 50% (Fig. 4B). In the AGS cell line upon the ADAR1 knockdown, miRNA-302a-3p was the most downregulated miRNA (88.0%) (Fig. 4C, Supplementary Table 13). In contrast, miRNA-148a-3p showed the most drastic downregulation (72.8%) in the MKN-45 cell line, after the ADAR1 knockdown (Fig. 4D, Supplementary Table 14). Furthermore, to understand the gene-regulatory network controlled by ADAR1, we combined the RNA-Seq and small RNA-Seq data showing differentially expressed miRNAs and mRNAs after the ADAR1 knockdown. To construct a regulatory network, we used open-source program miRTarVis to combine miRNA expression changes with those of target mRNAs that had inversely correlating expression levels. The results are presented in Fig. 5. Specifically, in MKN-45 cells, miRNA-148a-3p and miRNA-215-5p had the strongest effects on target mRNAs (Fig. 5A, B), whereas miRNA-302a-3p, miRNA-30c-5p, miRNA-615-3p, and miRNA-107 were predicted to influence target mRNAs in AGS cells (Fig. 5C, D).

Fig. 5
figure 5

Combinatory analysis of RNA-Seq and small RNA-Seq data revealed several hub miRNAs regulating mRNA expression by ADAR1, in an editing-independent manner. The node-link diagram (A) and a tree map image (B) of integrated data on miRNA and target mRNAs in MKN-45 cell lines and AGS cell lines (C and D). See “Methods” for details

Full size image

Editing-Independent Regulation of miRNA-148a-3p/NFYA by ADAR1

Among several ADAR1-dependent hub miRNAs identified in Fig. 5, we verified by qRT-PCR whether the miRNA-148a-3p level decreased after the ADAR1 knockdown in the MKN-45 cell line. Indeed, the level of miRNA-148a-3p was found to decrease by 40.1% after the knockdown (Fig. 6A). Among the target mRNAs predicted to be regulated by miRNA-148a-3p, we examined NFYA as a target gene to confirm. The results indicated a 38.2% increase in the NFYA mRNA level (Fig. 6A) and a similar increase at the protein level after the ADAR1 knockdown (Fig. 6B and Supplementary Fig. 5A). By means of the prediction program, a wild-type (WT) and mutant NFYA UTR reporter vector were generated for the seed-binding sequence of miRNA-148a-3p (Fig. 6C). In a luciferase assay, a significant decrease in the luciferase activity of the WT reporter was restored by mutagenesis of the seed-binding sequence (Fig. 6D). We also performed qRT-PCR to quantify miRNA-148a-3p and confirmed that the target miRNA level substantially increased after miRNA-148a-3p mimic treatment (Fig. 6E). Moreover, the miRNA-148a-3p level of AGS cells was also found to decrease by 23.9% in RNA-Seq data. We therefore expanded our investigation to determine changes in the AGS cell line, which also showed a significant decrease in the NFYA mRNA and protein levels in the AGS cell line (Fig. 6F, G). Additionally, in the AGS cell line, miRNA-302a-3p expression was expected to decrease by 88.0% after the ADAR1 knockdown. Indeed, the decrease in the miRNA level was confirmed by qRT-PCR (Supplementary Fig. 5B, C), which led to upregulation of TXNIP mRNA (Supplementary Fig. 5D), one of the expected targets of miRNA-302a-3p (Fig. 5C, D).

Fig. 6
figure 6

Downregulation of miRNA-148a-3p by the ADAR1 knockdown increases the expression of NFYA. Along with ADAR1 mRNA (A, left) and miRNA-148a-3p levels (A, middle), the NFYA mRNA (A, right) and protein expression (B) were determined after the ADAR1 knockdown in the MKN-45 cell line. C An alignment image showing the miRNA-148a-3p seed sequence interacting with the 3′-UTR of NFYA mRNA. Luciferase assay results involving the wild-type NFYA UTR reporter (WT) or a mutant in the seed-binding site of miRNA-148a-3p (MUT), along with an miRNA-148a-3p mimic or scrambled control (D) and qRT-PCR results for miRNA-148a-3p in the same samples (E). ADAR1 mRNA (F, left) quantified along with NFYA mRNA (F, right) as well as their protein expression (G) in the AGS cell line after the ADAR1 knockdown

Full size image

Discussion

In this study, we demonstrated widespread RNA editing in 3′-UTRs of two gastric cancer cell lines. We also observed editing-dependent regulation of PHACTR4 mRNA and protein levels in cancer cell lines and in tumorous and normal samples obtained from patients, as indicated by consistent inverse correlations between ADAR1 and PHACTR4 expression levels (Fig. 7). In addition, we showed that decreased miRNA-148a-3p expression in ADAR1 knockdown cells led to increased mRNA and protein expression of target gene NFYA. These data revealed editing-independent effects of ADAR1 on the expression regulation between miRNAs and their target mRNAs (Fig. 7).

Fig. 7
figure 7

A summary of effects of the ADAR1 knockdown in gastric cancer cell lines. The editing-dependent pathway is demonstrated by a loss of the miRNA-196a-3p binding site in the 3′-UTR of PHACTR4 mRNA with a subsequent increase in its expression. By contrast, a decrease in the miRNA-148a-3p level and the subsequent increase in NFYA expression reflect the editing-independent pathway

Full size image

Altered expression of target genes after RNA editing has been reported in multiple studies, and the best-characterized consequence of RNA editing involves changes in protein amounts due to editing of a coding or UTR region of a target mRNA. In gastric cancer cells, oncogenic activities of unedited PODXL mRNA are abrogated by ADAR2-mediated editing of its codon 241 [29]. Similarly, RNA editing leads to a serine-to-glycine substitution at residue 367 of AZIN1 and reportedly promotes hepatocellular carcinogenesis [30]. Herein, PHACTR4 mRNA was the only edited target that showed editing-dependent protein expression changes in gastric cancer cells. PHACTR4 was previously identified as a suppressor of tumorigenesis and of normal cell proliferation. Moreover, PHACTR4 deletion in lung, breast, and colorectal cancer cells is associated with loss-of-function mutations at multiple tumor sites [31]. In addition, PHACTR4 is known to perform important regulatory functions in developmental processes, such as neural tube and optic fissure closure [32] and directional migration of enteric neural crest cells [33]. Hence, our findings of decreased PHACTR4 levels in cancer tissues and inverse correlations with ADAR1 indicate PHACTR4’s involvement as tumor suppressor in gastric cancer.

NFYA is known to bind to the CCAAT motif, which is in the promoter region of a number of genes including cyclin B1, von Willebrand Factor, and TP53 (p53) [34,35,36,37,38]. NFYA has been shown to either activate or suppress the transcription of its target genes, which control various processes from hematopoietic stem cell self-renewal to the response to DNA damage [34,35,36,37]. When it comes to carcinogenesis, however, NFYA has generally not been found to be mutated or overexpressed [39]. Rather, its binding motif—CCAAT box—is reportedly enriched in the promoters of genes overexpressed in a number of cancer types, resulting in its procarcinogenic role in breast, colon, thyroid, and prostate carcinomas [40,41,42,43,44]. Given that ADAR1 has generally been regarded as procarcinogenic, the upregulation of NFYA by the ADAR1 knockdown in gastric cancer cells seems intriguing. Because NFYA exerts its main function via binding to the promoters of target genes, a future study should be focused on the identification of the ADAR1-dependent NFYA target genes in gastric cancer to understand the functional impact of ADAR1 via NFYA.

In the course of searching for a functional outcome of RNA editing, we found that the RNA-binding proteins such as HuR can affect protein localization [45]. The HuR protein is another well-known RNA-binding protein, which may interact with ADAR1 via binding to the same transcript [46]. It is therefore plausible that editing of 3′-UTR may alter the binding site for other RNA-binding proteins and subsequently alter localization of such a protein. However, we could not find an alteration in the protein localization of PAICS and PHACTR4 in the present study (Supplementary Figs. 2C and 3E). Moreover, no change in protein levels of PAICS and PPIA was observed despite active editing in their 3′-UTRs. These data can be interpreted in the following three ways: First, in contrast to the edited sites in coding regions, up to 39 edited sites were found in the 3′-UTR of a single gene, as reported previously [47]. These edited sites may create novel seed sites for respective miRNAs or vice versa, and the change at the protein level is likely to be the sum of the changes from these miRNA-binding sites. Second, RNA editing is reversible and is mediated by the ADAR1 enzyme, which is ubiquitously present and is indispensable for living. Unlike genomic mutations, which are irreversible, those due to ADAR1 activities may lead to increased expression or altered localization during development after exposure to deleterious stimuli. Indeed, recent research showed that nuclear export of the ADAR1 p110 isoform to the cytoplasm under stressful conditions, such as heat shock or UV irradiation, inhibits Staufen 1-mediated mRNA decay [19]. Likely, ADAR1 was also downregulated after hydrogen peroxide treatment through a STAT-2-dependent pathway, which in turn modulates PKR activity [48]. Therefore, specific roles of ADAR1 in the regulation of PAICS or PPIA or VSIG1 under stimulation-related and developmental conditions require further studies. Third, in the present study, we implemented an shRNA-mediated knockdown of ADAR1 because the two gastric cancer cell lines express a substantial amount of ADAR1 (Supplementary Fig. 6). Nonetheless, a large proportion of editing targets in a specific gene consists of so-called weakly edited sites. Hence, for these targets, a decrease in the editing rate after the ADAR1 knockdown may have a minor impact, whereas overexpression analysis of these targets with low editing levels will likely be more informative.

In this study, we applied an integrative bioinformatics tool to link a considerable number of up- or downregulated miRNAs after the ADAR1 knockdown to their predicted mRNA targets. Substantial numbers of mRNA alterations were observed, which may mostly be due to a secondary effect upon the ADAR1-mediated expression change of miRNAs. Although this way of analyzing targets may have limitations (e.g., wrong miRNA–mRNA target prediction, involvement of a transcription factor or epigenetic modifier targets), it seems to be a useful way to estimate the changes at a glance. We applied this bioinformatic tool to address the coordinated changes in expression of specific miRNAs and mRNAs after the ADAR1 knockdown; this analysis is limited to the assessment of RNA editing-independent data. Greater prediction accuracy of RNA editing analyses may be achieved by integrating predictions of seed sequence editing with miRNA abundance data.

Notably, we found that relatively small numbers of miRNAs were edited in their seed sequences (n = 5). This result is consistent with other studies [49, 50], likely reflecting inhibited miRNA biogenesis in edited primary miRNAs. Moreover, editing of miRNA-1260a occurred at rates of up to 100%, and these did not decrease after the ADAR1 knockdown, thereby warranting further studies of this target. A recent functional study on ADAR2 revealed its participation in gastric cancer, even though its expression is low [29].

There are some limitations of our study. First, a relatively small number of patients’ samples was examined. Second, the functional implication of ADAR1 in gastric cancer was not demonstrated in the current study, leaving unclear the role of ADAR1 in carcinogenesis. Nonetheless, the results of the present study provide important clues to the understanding and development of novel therapeutic strategies against gastric cancer.

In conclusion, we demonstrated that ADAR1 regulates post-transcriptional gene expression in gastric cancer through both RNA editing-dependent and -independent mechanisms. Further research combining the functional data on these ADAR isoforms will help researchers to fully understand the physiological meaning of RNA editing.