Genome-wide profiling reveals alternative polyadenylation of mRNA in human non-small cell lung cancer

Twenty-six patients diagnosed with NSCLC (Table 1) were enrolled in this study. The protocol for using paired cancerous tissues and adjacent normal lung tissues was approved by Ethics Committee of The First People’s Hospital of Hangzhou. Fresh tissues were collected after surgery from each patient. All samples were snap frozen in liquid nitrogen within 30 min after surgical removal and then stored at − 80 °C. H&E staining was performed for validation of pathologic status of each specimen.

Human fetal lung fibroblast cell line HFL1 and lung cancer cell lines was purchased from ATCC (https://​www.​atcc.​org/​) and cultured in DMEM medium (Gibco, Waltham, MA, USA) supplemented with 10% fetal bovine serum (Gibco, Waltham, Massachusetts, USA), 100 U/mL penicillin and 100 μg/mL streptomycin at 37 °C humidified atmosphere with 5% CO₂. Log-phase growing cells were collected for subsequent gene expression detection and RNA interference experiment.

oligo siRNA (GGTGGATCGTTCTCTACGT) designed to target CSTF2 (Cleavage Stimulation Factor Subunit 2) was synthesized by RiboBio Co., Ltd. (Guangzhou, China). CSTF2 oligo siRNA was transfected into H460 cells at a concentration of 50 nM using Lipofectamine RNAiMAX Reagent (Invitrogen, Carlsbad, CA, USA). Cells were collected 48 h later after transfection for subsequent experiments, and quantitative real-time PCR was used to validate the efficiency of the RNA interference. Scramble siRNA Oligo was used as a control.

Total RNA were extracted from lung cancerous and paired adjacent normal tissues, or from cultured H460 cells using TRIzol reagent (Life Technologies, USA) according to the manufacturer’s protocol. The RNA samples were used to generate libraries for IVT-SAPAS analysis. The IVT-SAPAS was performed with the illumina HiSeq 2500 platform according to the manufacturer’s protocol [15]. Briefly, twelve libraries with distinct barcodes were pooled together and were sequenced with Hiseq 2500 with rapid run mode, and 20 dark cycles were taken and then 55 bp were further sequenced.

The raw reads from IVT-SAPAS analysis were mapped to the human genome (hg19) using Bowtie software [16], and internal priming was filtered by our in-house Perl scripts. Poly(A) sites of each sample were defined by clustering the unique mapped reads, and were then merged together across samples. The merged poly(A) sites were mapped to the 3′UTRs dataset constructed from UCSC known genes for further analysis. The expression levels of poly(A) sites were calculated by the clustered reads and scaled to the sample with the lowest number of raw reads.

To compare the 3′-UTR length across samples, we standardized the length by defining the longest 3′UTR as 1.0 and calculated the weighted mean of 3′-UTR length according to the reads and related position of multiple poly(A) sites for each gene. Genes with alternative poly(A) were identified by a test of linear trend with significant P-values paired to a false discovery rate with cutoff of 1% between the lung cancer tissue and paired para-carcinoma tissue. We denoted the 3′-UTR length of each gene identified in paired non-carcinoma tissue as 1 and that in cancerous tissue as 2, and used Pearson correlation to test the relationship between two variables. A positive R-value indicates that the genes in the cancerous tissue showing longer tandem 3′UTRs versus the paired non-cancerous tissue, and a negative R-value indicates that the genes in the cancerous tissue showing shorter tandem 3′UTRs versus the paired non-cancerous tissue.

Functional annotation of the DEGs and APA switching genes was performed using the DAVID (The Database for Annotation, Visualization and Integrated Discovery) Bioinformatics Resources (http://​david.​abcc.​ncifcrf.​gov/​), and genes with detected 3′UTRs length change were categorized for their significant enrichment according to Biological Process GO terms and pathways, with normalization of the background model for all transcripts found in both lung cancer and para-carcinoma tissues.

Total RNA extracted from cells using TRIzol Reagent (Life Technologies, USA) was used for cDNA synthesis with using SuperScript^® III Reverse Transcriptase (Invitrogen, Carlsbad, CA, USA) and an oligo (dT) primer. 1.0 μg of total RNA was used for reverse transcription (RT) according to the manufacturer’s recommendations. qRT-PCR was performed in 10 μL of quantitative reaction mixtures containing 5 μL of SYBR Premix ExTaq (Takara, Japan), 0.4 μM each of forward and reverse primers, 1 μL cDNA, and nuclease-free water. The program used for qRT-PCR consisted of a denaturing cycle of 2 min at 95 °C, 45 cycles of PCR (95 °C for 10 s, 60 °C for 30 s) and a melting cycle of 60 °C for 15 s. The primer sequences is listed as below: CSTF2 qF 5′-TCGTTCTCTACGTTCTGTGTTCGTG-3′; CSTF2 qR 5′-ACCCTTTGGCTTTCCTGTCTCTCT-3′; β-actin 5′-qFGAGAAAATCTGGCACCACACCTT-3′; β-actin qR 5′-GCACAGCCTGGATAGCAACGTA-3′; The mRNA level of CSTF2 gene in H460 cells was normalized to β-actin mRNA level. Results of real-time PCR were analyzed using the 2^−ΔCT method to compare the transcriptional levels of CSTF2 gene in lung cancer cells to that in normal control.

Cells were collected and lysed in RIPA buffer containing protease and phosphatase inhibitors. 20 μg of cell lysates were used for western blot analyses using antibodies against CSTF2 and GAPDH (Cell Signaling Technology, USA). Simply, proteins were separated by SDS-PAGE and transferred to the polyvinylidene difluoride membrane. The membrane was then incubated with primary antibody at 4 °C overnight, followed by incubation of secondary antibody for one hour. Finally, the target proteins detected by chemiluminescence kit. The western blot signal was detected by ChemiDoc XRS + (BioRad, USA).

From a total of 41,773,101 average reads per each paired sample that were obtained from IVT-SAPAS analysis, we found an average read of 17,260,816 that meet the criteria for data analysis after data filtering, genome alignment and orientation filtering (Additional file 1: Table S1). Of them, 5,303,272 reads correspond to APA sites, with near 69.7% reads being mapped to the transcript of the UCSC ends database and Tian’s database, and 1.5% and 1.3% of the reads being mapped to the noncoding gene or 1 kb downstream of the UCSC canonical genes defined using the GENCODE gene loci, respectively (Fig. 1a). In addition, there were 5.5%, 15.6% and 1.7% of reads were mapped in the introns, 3′UTRs or coding sequences (CDSs), respectively.

We conducted a gene expression survey for lung cancer tissue versus paired non-carcinoma tissue by calculating the number of transcript reads. Based on this analysis, we identified a total of 14,030 genes with distinct difference of expression. Of them, 1439 (10.26%) genes showed up-regulated expression and 1364 (9.72%) genes showed down-regulated expression in lung cancerous tissue (Fig. 1b). KEGG (Kyoto Encyclopedia of Genes and Genomes) enrichment analysis further revealed that the up-regulated genes in cancer tissues were enriched in signaling pathways primarily associated with cell cycle and biosynthesis of amino acids (Fig. 1c), and the down-regulated genes were enriched in pathways of neuroactive ligand-receptor interaction and cytokine–cytokine receptor interaction (Fig. 1d).

Previous studies have demonstrated that highly proliferative cells or cancer cells often expressed substantial amounts of mRNA isoforms with shortened 3′UTRs [6, 17]. Studies also revealed that shortened 3′UTRs correlated with poor prognosis of cancer, including lung cancer [13, 18]. We thus performed a comparison analysis for the tandem 3′UTRs lengths of these identified genes. For this, we set up a standard of the 3′UTR length by designating the longest 3′UTR as 1.0 and calculated the 3′UTR length of each gene using weighted mean of 3′UTR length with multiple APA sites to reduce the variance of 3′UTR length. We detected shortened 3′UTR in collected cancer tissues from 96.2% (25/26) patients, as shown in Fig. 2a. We also performed a test of linear trend to determine the tandem 3′UTR length switching. Based on the statistical significance (FDR = 0.01, |r| ≥ 0.1), we identified total of 7639 genes with potential APA switching in 26 paired clinical samples. Of interest, we noticed that lung cancer cells tend to have shortened 3′UTRs of these identified genes (1855 genes showing shorten 3′UTR and 58 genes showing elongated 3′UTR). Additional file 1: Table S2 shows the gene numbers with identified APA switching for each patient.

To determine the significance of these APA site-switching events, we performed functional annotation using the web-accessible DAVID program. The analysis indicated that genes with shorten 3′UTR are enriched in signaling pathways such as mTOR signaling, ubiquitin mediated proteolysis and RNA degradation (Fig. 2b). With gene ontology (GO) enrichment analysis, we also identified the top 30 enriched genes with shorten 3′UTR determined in these clinical samples (Fig. 2c).

To assess the potential mechanisms of 3′UTR shortening in human lung cancer, we examined the RNA expression of genes (Additional file 1: Table S3) that their encoding proteins may correlate to 3′UTR length editing. Of total 93 candidate genes, we found that 28 genes displayed significant changes of up-regulation and 5 genes of down-regulation. Our results also revealed that CSTF2 was one of the top genes with most significant up-regulation in cancerous tissues versus paired non-cancerous tissues (Fig. 3a and Additional file 1: Tables S3, S4). Of interest, we detected higher protein expression of CSTF2 in six NSCLC cancer cell lines versus normal lung HFL1 cell line. Of these established cancer cell lines, H460 cells showed highest expression level of CSTF2 (Fig. 3b). Our qRT-PCR analysis further confirmed the higher mRNA expression of CSTF2 in H460 cells when comparing to HFL1 cells (Fig. 3c). We further assess the expression pattern of CSTF2 in lung cancer samples in The Cancer Genome Atlas database (TCGA), and the results from 100 paired TCGA clinical samples showed near twofolds change of up-regulated CSTF2 expression in cancer cells versus normal lung cells, which is in consistence with the CSTF2 expression pattern of our clinical samples (Fig. 3d). We thus hypothesize that the CSTF2 gene may play a key role in regulation of 3′UTR length in lung cancer cells.

To test this, we used RNA interference to knock down the CSTF2 expression in H460 cells, and determined the APA site-switching events with IVT-SAPAS analysis. We observed near 80% knocking-down effect of siRNA on CSTF2 expression (Fig. 4a), and a total of 10,235 genes with significant change after siRNA transduction was detected in H460 cells. Of them, 5694 genes (55.63%) are with APA site-switching events, 269 genes (2.62%) show expression difference, and 78 genes (0.76%) show distinct change of 3′UTR length. Of note, the global comparison shows consistence of IVT-SAPAS reads for genes with APA site switching between data sets for the clinical samples and for H460 cells, however, significant differences are observed for genes with differential expression or with different 3′UTR in length, whereas more reads were detected with clinical sample set (Additional file 1: Table 5S). In addition, we noticed that the IVT-SAPAS reads from the clinical samples and H460 cells showed similar distribution pattern of poly(A) signals (6-nucleotides), which is consistent with the previously reported [15] and demonstrates the validity of the IVT-SAPAS analysis used in our study capturing poly(A) site (Fig. 4b).

In IVT-SAPAS assay for H460 set, we detected elongation of 3′UTR length for 60 genes in CSTF2-siRNA transducted cells (Fig. 4c). Of interest, we found that, of the 1855 genes with shorter 3′UTR length identified in human lung cancer tissues (Fig. 4d), 38 genes displayed elongation of 3′UTR length when CSTF2 expression was knocked down in H460 cells (Fig. 4e and Table 2). The functional annotation of these genes further revealed that eleven of these 38 overlapping genes can be annotated into signaling pathways of insulin signaling, lipid metabolism and MAPK signaling. For example, FGF2 (fibroblast growth factor 2, uc003iev.1), which shows 0.3 fold-change of down-regulation in cancer tissue (P-value = 1.85e−4), has been demonstrated to play vital roles in tumor angiogenesis and lymphangiogenesis of lung cancer [19, 20]; MAP4K4 (mitogen-activated protein kinase kinase kinase kinase 4, uc002tbc.2), which shows 1.2 fold-change of upregulation in cancer tissue (P-value = 0.037), contributes to cell proliferation, anchorage-independent growth and cancer cell migration of lung cancer [21]. Of interest, the analysis for the distributions of poly(A) reads showed switching of distal stop codon in clinical samples may indicate cancer cells prefer to use the shorten length of transcripts for both FGF2 and MAP4K4 when comparing to normal cells, and the knocking down of CSTF2 expression in H460 cells could lead to changes of transcript lengths of these genes according to the detected numbers and sites of poly(A) reads (Additional file 2: Fig. S1). These findings thus not only demonstrate the vital role of CSTF2 in regulation of 3′UTR length for cancer cell-associated genes, but also indicate a potential of CSTF2 or its protein family may also serve as oncogene(s) driving carcinogenesis of NSCLC.