An ESCC EC109 cell line was obtained from the Cell Bank of the Chinese Academy of Sciences (Shanghai, China). The Japanese ESCC KYSE30 cell line was generously provided by Dr Guan Xinyuan from Sun Yat-sen University Cancer Center (SYSUCC). These two cell lines were maintained in RPMI-1640. The HEK293T cell line (used for lentivirus packaging) was obtained from SYSUCC and maintained in DMEM. We added 10% fetal bovine serum and 1 mM penicillin/streptomycin to the media.

For siRNA transfection, Lipofectamine RNAiMAX reagent (Invitrogen) was used according to kit directions. siRNAs were purchased from RiboBio (Guangzhou, China). For gene transient transfection, coding sequences were cloned into pcDNA3.1, and the plasmids were transfected with Lipofectamine 2000 (Invitrogen). For gene stable transfection, pLVX-IRES-Puro lentiviral expression vector (Clontech) was used according to the manufacturer's instructions.

Cell growth was measured by MTT assay as previously reported (20), using 2×10³ cells seeded into each well of a 96-well plate. A clone formation assay was carried out as previously reported (15), using 1×10³ cells seeded into each well of a 24-well plate.

Apoptosis was measured using an Annexin V-FITC and PI staining kit (Bestbio, Shanghai, China) according to the manufacturer's protocol. Briefly, cells were collected, washed and resuspended in 400 µl binding buffer containing 5 µl Annexin V-FITC and 10 µl PI. After incubation, samples were assessed with flow cytometry. We summed percents of early apoptotic cells (Annexin V-positive, PI-negative) and late apoptotic or necrotic cells (Annexin V/PI-double-positive) to obtain percent apoptotic cells.

Animal experiments were approved by the Institutional Animal Care and Use Committee at SYSUCC. Five-week old female BALB/C-nu/nu nude mice were purchased from Experimental Animal Center of Guangdong Province and used for the tumorigenicity assay. Briefly, 1×10⁶ cells resuspended in RPMI-1640 were injected into each mouse scapula. After 4 weeks, mice were sacrificed and xenograft tumors were removed, photographed, and weighed.

High-quality total RNA was extracted with TRIzol (Invitrogen), and polyA⁺ mRNA was selected with oligo(dT) magnetic beads and chemically fragmented. RNA fragments were reverse transcribed to cDNA and the ~200 bp cDNA fraction was selected to construct a sequencing library. RNA-Seq was carried out on a HiSeq 2000 platform according to Illumina's protocol. After filtration, clean reads were aligned to the human reference genome (hg19) with Tophat2 (21). Single nucleotide variants were assessed with GATK2 (22). SnpSift was used for variant functional annotation (23).

Site-specific RNA editing was measured with RT-PCR and direct sequencing (24). Editing was calculated with ImageJ software (http://rsb.info.nih.gov/ij/). Primers for PCR include FLNA-forward (CTTTGTTCCCGCTGAGATGG), FLNA-reverse (TTCCAGACCTGCTCCGTAAG), FLNB-forward (GAAGAACTCACACTGCGTCC), FLNB-reverse (CCTGCTCGGGTGGTGTTAAT), COPA-forward (GCGAGGCATTGACTTCCATA), COPA-reverse (GGTCTCTGCGGAAAGTCTGA), IGFBP7-forward (CTGCCCCTCTCCTCTTCCT), IGFBP7-reverse (GGGATTCCGATGACCTCACA).

Western blotting was carried out according to published standard methods. Antibodies used included ADAR2 (Sigma, 1:1,000), IGFBP7 (Abcam, 1:1,000), Matriptase (GeneTex, 1:1000), pAkt (Ser473) (CST, 1:1,000), total Akt (CST, 1:1,000), pBAD (SAB, 1:1,000), BAD (CST, 1:1,000), GAPDH (CST, 1:1,000), His-tag (CST, 1:1000), c-Myc-tag (MBL, 1:1,000).

Tissue microarray (TMA) blocks with 116 ESCC samples, and three pairs of frozen ESCC primary tumor tissues and adjacent normal esophageal tissues were obtained from the Biobank of SYSUCC. Tissue samples were surgically removed between 2011 and 2014. Each participant signed an informed consent form and the study was approved by the Human Ethics Committee.

Immunohistochemical staining for pAkt (antibody was purchased from CST) and pBAD (antibody was purchased from SAB) was conducted as previously reported (25). Two investigators scored immunoreactivity independently, and staining was scored as follows: i) percent positively staining cells: zero (0–10%), 1 (11–25%), 2 (26–50%), 3 (51–75%), and 4 (76–100%); and ii) staining intensity: zero (no signal), 1 (weak), 2 (moderate), and 3 (strong). Staining scores were multiplied by percent scores for a final score (range 0–12).

Unless otherwise specified, data are means ± SD of three experiments. SPSS (version 16.0) software was used for data analysis. P<0.05 was considered statistically significant.

Previously, we reported that ADAR2 expression was downregulated in ESCC compared with adjacent normal tissues, and patients with less tumor ADAR2 expression had worse prognoses (3). Furthermore, we found that several known editing targets of ADAR2 were edited in normal esophageal tissue, but they were almost not edited in the seven ESCC cell lines (data not shown), which suggested that ADAR2 lost its function in ESCC. To investigate the ADAR2 role during ESCC progression, we used ADAR2 lentivirus to overexpress ADAR2 in ESCC KYSE30 and EC109 cell lines. Overexpressing lines were KYSE30-AR2 and EC109-AR2, and the control lentiviral-infected cells were KYSE30-PLVX and EC109-PLVX. Upregulation of ADAR2 was confirmed with western blotting (Fig. 1A). In vitro studies confirmed that cells transfected with ADAR2 had less growth (Fig. 1A) and colony formation (Fig. 1B) compared to controls. Annexin V and PI staining revealed that overexpression of ADAR2 induced apoptosis (Fig. 1C). In contrast, knockdown of ADAR2 in KYSE30-AR2 and EC109-AR2 reversed cell growth and apoptosis effects to some extent (Fig. 2). Moreover, ADAR2 overexpression decreased tumor growth in xenograft tumor models (Fig. 1D and E).

To explore target genes edited by ADAR2 in ESCC, we used RNA-seq assay in KYSE30-PLVX and KYSE30-AR2 cells. We noted 19 A-to-G single nucleotide variation sites, G fractions of which were ≥10% higher in KYSE30-AR2 than in KYSE30-PLVX cells. These were selected as potential A-to-I editing sites catalyzed by ADAR2. We validated these sites by assaying corresponding sequences of DNA and mRNA in KYSE30-AR2 and KYSE30-PLVX cells with direct sequencing. Finally, transcripts of four genes were identified as substrates of ADAR2 in ESCCs, including IGFBP7, FLNB, COPA, and FLNA (Fig. 3A). IGFBP7 was the only one reported to be associated with apoptosis (26–29), so we focused on IGFBP7. The editing site in IGFBP7 identified with RNA-seq was located at position 284 of the coding sequence, changing codon 95 from AAG (lysine) to AIG (arginine), henceforth termed K95R. K95R was validated in EC109-AR2 successfully (data not shown). We constructed and transfected three plasmids expressing wt ADAR1 and ADAR2, and RNA editing function loss type ADAR2 (ADAR2-mut) caused by deaminase domain mutation (30), respectively (Fig. 3B). Direct sequencing showed that only wt ADAR2 could edit IGFBP7 K95R, and neither ADAR1 nor ADAR2-mut could (Fig. 3C). We then measured IGFBP7 editing in 32 pairs of ESCC tumors and adjacent normal tissues. The data show that, similar to ADAR2 protein, IGFBP7 K95R editing decreased in tumors (Fig. 3D). Moreover, IGFBP7 K95R editing was moderately correlated with ADAR2 expression in 116 ESCC tumors (r=0.52, P<0.001) (Fig. 3E). Thus, these results suggested that IGFBP7 K95R is an ADAR2-specific editing target in ESCCs.

IGFBP7 was previously reported to be an apoptotic promoter in several other cancers (26–29). So, we investigated the role of IGFBP7 in ADAR2-mediated apoptosis in ESCCs. At 48 h after transfection of IGFBP7 siRNAs and the siRNA negative control (nc) in KYSE30-AR2 and EC109-AR2, cells were harvested and apoptosis was assayed. Data show that knockdown of IGFBP7 abolished ADAR2-induced apoptosis (Fig. 4A). Transfection of ADAR2-mut, the RNA editing function loss type of ADAR2, did not induce apoptosis as wt did (Fig. 4B). Thus, the pro-apoptotic ability of ADAR2 might depend on IGFBP7 editing. However, overexpression of wt IGFBP7 (IGFBP7-wt) could promote ESCC apoptosis as well as the edited type (IGFBP7-K95R) (Fig. 4C and D). We conjectured that RNA editing could fine tune IGFBP7 function by affecting its RNA or protein stability, however, this fine-tuning was trivial when cells expressed exogenous IGFBP7 abundantly.

According to previous reports, IGFBP7 is a secreted protein and IGFBP7 K95R editing is located within a matriptase protease recognition site (PRS) and inhibited matriptase mediated proteolysis in test tubes (31). In our study, we validated this phenomenon in culture medium of ESCC cells. Matriptase and IGFBP7 (IGFBP7-wt and IGFBP7-K95R) expression vectors were co-transfected into KYSE30 and EC109 as described in Fig. 4E. Then, 48 h after transfection, media were collected and assayed with western blotting. Data show that IGFBP7-wt could be partially cleaved at the 27 kDa C-terminus and the 8 kDa N-terminus (only the intact and 27 kDa C-terminus was measured with western blotting) by overexpression of matriptase, whereas little IGFBP7-K95R was cleaved (Fig. 4E). Moreover, western blotting showed that ADAR2 overexpressing xenograft tumor had less cleaved endogenous IGFBP7 (Fig. 4F). Thus, RNA editing of K95R can protect IGFBP7 against matriptase proteolysis in ESCC culture and xenografts. Measuring IGFBP7 expression in 3 pairs of surgically removed ESCC tumors and adjacent normal tissues revealed that more cleaved IGFBP7 was apparent in tumors (Fig. 4G), indicating a positive correlation between IGFBP7 editing and tissue stability.

Full-length IGFBP7, and not the truncated C-terminus, was reported to bind to IGF1R and inhibit downstream Akt signaling in mouse embryonic fibroblasts (MEFs) (17). We investigated the function of intact and cleaved IGFBP7 in ESCC cells with regard to apoptosis. The intact form, N- and C-terminus of IGFBP7 expression constructs (each contained a BM40 signal peptide sequence in the vector) were transfected into ESCC cells separately. Only the intact form induced apoptosis 48 h after transfection (Fig. 4H and I). Thus, protein integrity is essential for IGFBP7 pro-apoptotic ability.

IGFBP7 was reported to mediate apoptosis via regulating Akt signaling (17), so we validated this phenomenon in ESCC (Fig. 5A), and investigated whether ADAR2 could influence Akt signaling. As expected, ADAR2 overexpression in KYSE30 and EC109 cells inhibited phosphorylation of Akt and a downstream molecular BAD (BCL2 associated agonist of cell death) (Fig. 5B). However, when IGFBP7 was knocked down, this effect was abolished (Fig. 5B). Tissue arrays containing 116 ESCC tumors were immunohistochemically assayed for pAkt and pBAD, and data show that both were negatively correlated with ADAR2 (Fig. 5C and D) or IGFBP7 K95R editing (Fig. 5E and F) to a certain degree. Therefore, ADAR2 might protect BAD from phosphorylation and induce apoptosis by promoting IGFBP7-mediated Akt signaling inhibition.