Systemic genome screening identifies the outcome associated focal loss of long noncoding RNA PRAL in hepatocellular carcinoma

Systemic analyses using large‐scale genomic profiles have successfully identified cancer‐driving somatic copy number variations (SCNVs) loci. However, functions of vast focal SCNVs in “protein‐coding gene desert” regions are largely unknown. The integrative analysis of long noncoding RNA (lncRNA) expression profiles with SCNVs in hepatocellular carcinoma (HCC) led us to identify the recurrent deletion of lncRNA‐PRAL (p53 regulation‐associated lncRNA) on chromosome 17p13.1, whose genomic alterations were significantly associated with reduced survival of HCC patients. We found that lncRNA‐PRAL could inhibit HCC growth and induce apoptosis in vivo and in vitro through p53. Subsequent investigations indicated that the three stem‐loop motifs at the 5′ end of lncRNA‐PRAL facilitated the combination of HSP90 and p53 and thus competitively inhibited MDM2‐dependent p53 ubiquitination, resulting in enhanced p53 stability. Additionally, in vivo lncRNA‐PRAL delivery efficiently reduced intrinsic tumors, indicating its potential therapeutic application. Conclusions: lncRNA‐PRAL, one of the key cancer‐driving SCNVs, is a crucial stimulus for HCC growth and may serve as a potential target for antitumor therapy. (Hepatology 2016;63:850‐863)

S omatic copy number variations (SCNVs) are extremely common in cancer, (1,2) and, in some cases, focal SCNVs have led to the identification of cancer-causing genes and have suggested specific therapeutic approaches. (3)(4)(5) Genomic analyses of cancer samples by cytogenetic studies, (6) array-based profiling, (7,8) and, more recently, by targeted exome capture (9) have identified recurrent SCNVs that are associated with cancer. However, an important challenge is to identify the oncogene and tumor suppressor gene targets of driver SCNVs (which often encompass unknown genes) and to elucidate the functional roles of SCNVs. In the past two decades, great progress has been achieved in the identification of candidate hepatocellular carcinoma (HCC)-related protein-coding and microRNA genes in abnormal chromosome Abbreviations: ANOVA, analysis of variance; CCK-8, cell-counting kit-8; ChIRP, chromatin isolation by RNA purification; CNVs, copy number variations; dChIRP, domain-specific chromatin isolation by RNA purification; EMSA, electrophoretic mobility shift assay; GAPDH, glyceraldehyde 3phosphate dehydrogenase; HBV, hepatitis B virus; HBx, HBV X protein; HCC, hepatocellular carcinoma; HCV, hepatitis C virus; HSC90, heat shock protein 90; IHC, immunohistochemistry; kb, kilobases; lncRNA, long noncoding RNA; lncRNA-PRAL, p53 regulation-associated lncRNA; mRNA, messenger RNA; nt, nucleotide; Ops, oligonucleotide pools; qPCR, quantitative real-time PCR; RNA-FISH, RNA fluorescence in situ hybridization; SCNVs, somatic copy number variations; siRNA, small interfering RNAs; XAF1, XIAP-associated factor 1. regions. (10,11) Although genome-wide copy number analyses have identified a large number of HCC phenotype-associated SCNVs in intergenic regions of the human genome, it is difficult to explain the pathogenesis of these SCNVs because they cannot be directly linked to changes in protein content or function. (12) Long noncoding RNA (lncRNA) is a class of noncoding functional RNA that has recently attracted great research interest. (13,14) Indeed, roles of lncRNAs as tumor suppressors and oncogenic drivers have appeared in prevalent types of cancers, such as prostate cancer (15) and HCC. (16)(17)(18) Furthermore, recent studies show that some lncRNAs that locate at genomic fragile sites or abnormal regions are associated with cancer phenotype, (19,20) suggesting that these breakpointassociated lncRNAs might be important in the development and progression of human cancer as cancercausing genes. Therefore, linking cancer-associated SCNVs to lncRNAs will provide independent support for functional implications and lead to a greater understanding of cancer pathogenesis.
To address these challenges, we reanalyzed the copy number profiles of 286 HCC tissues and matched nontumor liver tissues (21) and identified regions in the HCC genome that have undergone recurrent high-level focal amplifications or deletions. By integrating copy number profiles with lncRNA expression signatures derived from our previous study, (16) we identified differentially expressed lncRNAs within the aberrant genome regions. Of these, p53 regulation-associated lncRNA (lncRNA-PRAL) was significantly underexpressed with recurrent genomic deletions in HCC. And lncRNA-PRAL genomic alterations were correlated with poor prognosis of HCC patients. Using small interfering RNA (siRNA)-mediated knockdown, adenovirusmediated overexpression and RNA pull-down, lncRNA-PRAL was found to be involved in p53mediated HCC growth and apoptosis.

HUMAN TISSUES
All samples were collected with the informed consent of the patients and the experiments were approved by the ethics committee of Second Military Medical University (Shanghai, China). The human tissues are detailed in the Supporting Information.

COPY NUMBER DATA ANALYSIS
The raw copy number data were downloaded from Gene Expression Omnibus (GSE38323). The details of copy number data analysis are described in the Supplementary Information.

TUMOR GROWTH ASSAYS IN VIVO
The animal studies were approved by the institutional animal care and use committee of the Second Military Medical University. Additional details are described in the Supporting Information.

IMMUNOPRECIPITATION
For whole-cell extracts, cells were lysed in buffer containing 50 mM of Tris-HCl (pH 7.5), 150 mM of NaCl, 1% Triton X-100, and cleared by centrifugation. Immunoprecipitation with anti-HSP90 and p53 antibodies in whole-cell extracts was performed as the Pierce Co-Immunoprecipitation Kit (Thermo, Rockford, IL) protocol described.

STATISTICAL ANALYSIS
All the statistical analyses were performed using SPSS software (version 17.0; SPSS, Inc., Chicago, IL). For comparisons, one-way analyses of variance (ANOVAs), Fisher's exact test, chi-squared test, Wilcoxon's rank-sum test, Wilcoxon's signed-rank test, and two-tailed Student t test were performed, as appropriate. Cumulative survival probability was evaluated using Kaplan-Meier's method, and differences were assessed using the log-rank test. The SPSS software was also used to assess uni-or multivariate Cox's proportional hazards regression analyses with the hazard ratios and P values indicated. No statistical method was used to predetermine sample size, and investigators were not blinded to allocation during experiments.

ABERRANT lncRNAS AT CHROMOSOMAL BREAKPOINTS IN HCC
To find evidence of driver SCNVs in HCC genomes, we performed a data mining process using published data (GSE38323) and evaluated the frequency of SCNVs in HCC by subtracting those from paired nontumor liver tissues, in an effort to eliminate germline CNVs. A genome-wide view of segmented copy numbers revealed that most chromosomal arms undergo either copy number gain or deletion in a large proportion of the samples (Fig. 1A). Significantly, the genomic instability seems to be more remarkable in hepatitis B virus (HBV)-related HCC than in hepatitis C virus (HCV)-or alcoholrelated HCC (Fig. 1B). Notably, the intergenic regions also harbor more SCNVs in HBV-related HCC than in HCV-or alcohol-related HCC (Fig. 1B). Among the overlapped SCNVs in HBV-, HCV-, and alcoholrelated HCC, there are 43.2% that locate in the human genome intergenic regions (Fig. 1C). Considering an unanticipated and tremendous amount of noncoding sequences of human genome need to be transcribed, we further performed an association analysis of lncRNA expression microarray data (16) (GSE27462) with the SCNVs data according to their precise chromosome locations. This analysis identified 73 differently expressed lncRNA genes in the minimal deletion or amplification regions of HCC (Supporting Table 8). After a preliminary screen by strand-specific qPCR and confirmation by genomic real-time PCR (representative results are shown in Supporting Fig. 1), we found 11 lncRNAs with genomic copy number gains or losses in HCC; among these, six lncRNAs were frequently gained and five were frequently lost (Table 1).

MOLECULAR CHARACTERS OF lncRNA-PRAL IN HCC
To investigate whether abnormalities of DNA copy number might result in deregulation of lncRNA expression, we analyzed the concordance between the lncRNA gene copy number and transcript level in cohort 1 HCC tissues (Supporting Table 1). The lncRNA-PRAL (NCBI no.: AK128092) was found to be the most consistently and markedly reduced on both RNA transcript (P 5 0.000) and genome copy number levels (P 5 0.000; Fig. 2A), had the highest correlation coefficient (R 5 0.621; P 5 0.000) in HCC (Fig. 2B), and was most correlated with outcome of HCC patients (Table 1). Then, we detected the lncRNA-PRAL genomic and transcriptional level in 41 normal liver tissues (distal normal liver tissue of liver hemangioma) and 22 hepatitis liver tissues (Supporting Table 2), compared with HCC tissues from cohort 1, to find that hepatitis tissues had a lower lncRNA-PRAL expression with genomic deletion, whereas from hepatitis liver to HCC, lncRNA-PRAL exhibited a focal loss and significant underexpression (Fig. 2C,D).
Our results indicated that lncRNA-PRAL was transcribed with a poly A 1 tail (Supporting Fig. 2A). We next performed a rapid amplification of complementary DNA end analysis to identify the 5 0 and 3 0 ends of the lncRNA-PRAL transcript (Supporting Fig. 2B); the full-length lncRNA-PRAL sequence is presented in Supporting Fig. 3. The full-length transcription of lncRNA-PRAL was then validated by a northern blotting analysis (Supporting Fig. 2C). We applied RNA fluorescence in situ hybridization (RNA-FISH) to visualize the cellular localization and relative abundance of lncRNA-PRAL in HCCLM3 and SMMC-7721 cells. RNA-FISH demonstrated that lncRNA-PRAL localizes to both cytoplasmic and nuclear regions (Fig.  2E). Furthermore, RNA from nuclear and cytoplasmic fractions was analyzed by qPCR, revealing that lncRNA-PRAL was diffusely distributed both in the nucleus and cytoplasm (Fig. 2F). We observed a substantially lower level of lncRNA-PRAL expression in hepatoma cells (HCCLM3, Huh7, Hep3B, and SMMC-7721) than immortalized hepatocytes (L02 and QSG-7701; Supporting Fig. 2D). Then, we performed the siRNA-mediated knockdown and adenovirus-mediated overexpression of lncRNA-PRAL to explore the pathophysiological significance in hepatic cancerous cell lines HCCLM3 and SMMC-7721 (Supporting Fig. 2E,F).

HIGH EXPRESSION OF lncRNA-PRAL INHIBITS HCC CELL GROWTH IN VITRO AND IN VIVO
Cell-counting kit-8 (CCK-8) assays indicated that cell proliferation was increased in HCCLM3 and SMMC-7721 cells when lncRNA-PRAL was silent (Fig. 3A), whereas the HCCLM3 and SMMC-7721 cells in which lncRNA-PRAL was overexpressed showed a lower degree of proliferation than the negative control cells (Fig. 3A). According to Annexin V/ fluorescein isothiocyanate/propidium iodide dual staining, the proportion of apoptotic HCCLM3 and SMMC-7721 cells with lncRNA-PRAL overexpression was increased compared to control cells (Fig. 3B). Additionally, the HCCLM3 and SMMC-7721 cells in which lncRNA-PRAL was underexpressed showed a lower degree of apoptosis than the negative control cells (Fig. 3B). However, lncRNA-PRAL cannot promote cell apoptosis in p53-deficient (Hep3B) or p53-mutant (Huh7) HCC cells (Supporting Fig. 4), whereas both HCCLM3 and SMMC-7721 are p53 wild-type cells.
To investigate the proapoptosis effect of lncRNA-PRAL in vivo, we transplanted HCCLM3 and SMMC-7721 cells overexpressing lncRNA-PRAL into the bilateral armpits of nude mice. We found that the growth of tumors from the lncRNA-PRALoverexpressing cells was significantly reduced in comparison to the growth of control cells ( Fig. 3C and Supporting Fig. 5A). Apoptosis regulation of lncRNA-PRAL was further confirmation by in situ terminal deoxynucleotidyl transferase dUTP nick end labeling assay of tumor tissue sections (Supporting Fig.  6). To further assess the biological therapeutic value of lncRNA-PRAL in vivo, we evaluated the therapeutic potential of adenovirus vector-mediated gene delivery The lncRNA expression ratio of HCC tissues to matched noncancerous liver tissues were obtained using strand-specific quantitative real-time PCR (n 5 20). § Expression and genomic level of lncRNA in HCC tissues were from qPCR detection in cohort 4. Value is shown in manner of "overall survival correlation/recurrence-free survival correlation." Agarose gel electrophoresis of real-time PCR products from cytoplasmic and nuclear RNA purification procedures. Strand-specific real-time PCR was performed using primers for lncRNA-PRAL (upper), human U2 snRNA (small nuclear RNA U2), and S14 (ribosomal protein S14) primers with either the cytoplasmic or nuclear RNA fraction isolated from HCCLM3 cells. The U2 snRNA and S14 primers were used to show the effective separation of the cytoplasmic and nuclear RNA. *P < 0.05; **P < 0.01; ***P < 0.001. Abbreviation: DAPI, 4 0 ,6-diamidino-2-phenylindole. of lncRNA-PRAL (AV-PRAL) in human HCC cell lines bearing a nude mouse subcutaneous model. Interestingly; tumor growth was significantly inhibited with injection of AV-PRAL ( Fig. 3D and Supporting Fig.  5B). These data demonstrate that the adenovirus vector system provides an effective means to deliver lncRNA to HCC tissue, and in vivo administration of AV-PRAL may have considerable potential for HCC gene therapy.

lncRNA-PRAL BINDS TO HSP90 AND REDUCES THE UBIQUITINATION OF P53
Although most of the sequence of the lncRNA-PRAL overlaps in antisense with the 3 0 untranslated region of XAF1 protein-coding gene, which is known to regulate apoptosis, (24,25) we did not find any significant changes of XIAP-associated factor 1 (XAF1) messenger RNA (mRNA) or protein in lncRNA-PRAL over-or downexpressed SMMC-7721 and HCCLM3 cells (Supporting Fig. 7). These results indicate that lncRNA-PRAL-regulated cell apoptosis is independent of XAF1.
Because binding to specific protein is one important pattern for lncRNA to implement their molecular functions, we performed an RNA pull-down experiment to identify proteins that bind to lncRNA-PRAL (Fig. 4A). Mass spectrometry analysis of the differentially displayed band revealed that HSP90 was the main protein associated with lncRNA-PRAL (Supporting Table 3). We confirmed the association between lncRNA-PRAL and HSP90 by western blotting using the proteins isolated from the RNA pulldown assays (Supporting Fig. 8A). RNA immunoprecipitation was performed using a specific HSP90 antibody to ensure that lncRNA-PRAL could be specifically immune precipitated from cell lysates (Supporting Fig. 8B). The electrophoresis mobility shift assay (EMSA) results further confirmed that lncRNA-PRAL could directly bind to HSP90 (Fig. 4B) in vitro.

THE THREE BASIC STEM-LOOP MOTIFS AT THE 5 0 END OF lncRNA-PRAL ARE REQUIRED FOR THE ASSOCIATION BETWEEN HSP90 AND P53
To monitor the ability of different truncated fragments of lncRNA-PRAL binding to HSP90, we predicted lncRNA-PRAL structure using the RNA Structure (version 5.3) and Vienna RNA Package (1.8.5). According to the predicted lncRNA-PRAL structure, we truncated lncRNA-PRAL into fragments S1, S2, S3, and S4 as shown in Supporting Fig. 10A. Our results revealed that the 1,180-nucleotide (nt) fragment (S1) at the 5 0 end of lncRNA-PRAL mediated binding to HSP90 (Fig. 5A). Interestingly, a 2,583-nt fragment at the 5 0 end (S2) including S1 could not bind to HSP90 (Fig. 5A). We next analyzed the RNA secondary structure and identified three basic stem-loop motifs (motifs A, B, and C) located close to one another in the S1 fragment ( Supporting Fig. 10B). These motifs are also located close to one another in   Fig. 10A). Proteins isolated from the RNA pull-down assays were identified by western blotting analyses using specific anti-HSP90 or p53 antibodies. (B) The S1 fragment of lncRNA-PRAL was further truncated around the basic stemloop motif to monitor the ability of shorter versions binding to HSP90 by western blotting. (C) Up: dChIRP oligonucleotide design strategy. Biotinylated antisense OPs are designed to tile motif A, B, C, and fragment S3 of lncRNA-PRAL. Light blue solid line, lncRNA-PRAL; dotted line, antisense oligonucleotide; green patch, possible binding protein. Down: Workflow of dChIRP. Whole HCCLM3 cells are crosslinked to preserve protein/nucleic acid interactions. Sonication is used to shear nucleic acids. Next, the mixture is subdivided into five equal samples. OP is added to each sample, then the biotinylated oligonucleotides, RNA targets, and cross-linked proteins are purified using magnetic streptavidin beads. (D) Fragments of lncRNA-PRAL are enriched >1,000-fold over the abundant GAPDH mRNA in dChIRP samples. LacZ ChIRP does not enrich for lncRNA-PRAL over GAPDH. Average of technical triplicates 6 standard deviation shown. (E) The protein fraction from each lncRNA-PRAL dChIRP sample was analyzed by immunoblotting against HSP90, p53, and GAPDH. Motif A and C efficiently recovered HSP90 and p53. LacZ ChIRP recovered no detectable protein. GAPDH was not detected in any sample. b-actin was used as the input control. n 5 2. (F) Western blotting analyses for total and nucleus p53 protein expression in HCCLM3 cells transfected with PRAL full-length vector, S1 fragment vector, S3 fragment (lncRNA-PRAL with S1 fragment deletion) vector, and null vector as negative control, respectively. Abbreviation: RT-PCR, real-time PCR. the lncRNA-PRAL full-length transcript (Supporting Fig. 10A). However, those motifs are separate from one another in the S2 fragment (Supporting Fig.  10C), which may impede their association with HSP90. We further truncated the S1 fragment into shorter fragments that retained the basic stem-loop motifs (Supporting Fig. 10D) to monitor the ability of shorter versions binding to HSP90, and our results indicated that these shorter fragments lacked the ability for binding to HSP90 (Fig. 5B). In order to dissect the HSP90-and p53-binding domains of lncRNA-PRAL within its native cellular context, we further performed modified domain-specific chromatin isolation by RNA purification (dChIRP), a technique described previously. (22,23) First, biotinylated antisense 20-mer oligonucleotides were designed according to the reported principle, (22) and the oligonucleotides were divided into domain-specific oligonucleotide pools, such that pools target motif A, B, C, and fragment S3 of lncRNA-PRAL (Fig. 5C). We used a similar set of probes that targeted the lacZ mRNA as a negative control. Next, whole HCCLM3 cells were cross-linked to preserve protein-RNA interactions. Sonication was used to solubilize the fraction and shear nucleic acids. Oligonucleotide pools (OPs) were added to each sample and allowed to hybridize. Then, the biotinylated OP, hybridized RNA, and associated proteins were purified on magnetic streptavidin beads (Fig. 5C). To confirm that dChIRP could recover the intended fragments of lncRNA-PRAL, we purified the RNA fraction from the dChIRP samples and analyzed RNA recovery by qPCR using primers for motif A, B, C, fragment S3, and GAPDH that should not be enriched by lncRNA-PRAL dChIRP. We confirmed that lncRNA-PRAL dChIRP specifically retrieved lncRNA-PRAL, whereas LacZ ChIRP did not enrich for lncRNA-PRAL (Fig. 5D). Finally, the recovered material from each dChIRP sample was analyzed by immunoblotting. We found that motif C and motif A OP recovered p53 and HSP90, whereas the motif B and fragment S3 OPs recovered no detectable p53 or HSP90 (Fig. 5E). As negative controls, the LacZ OP recovered no proteins, and GAPDH was not detected in any sample. In general, the three basic stem-loop motifs at the 5 0 end of lncRNA-PRAL are required for the association between HSP90 and p53. Then, we conducted the experiment to detect the function of the short versions of lncRNA-PRAL in p53 activation. Deletion of S1(S3), which mediates the HSP90 association, abolished the ability of lncRNA-PRAL in increasing the p53 protein levels both in total and nucleus protein, whereas S1 fragment could enhance the accumulation of p53 (Fig. 5F).

THE ASSOCIATION OF lncRNA-PRAL WITH HCC PROGNOSIS
To further investigate whether the lncRNA-PRAL genomic copy number deletion correlated with the survival of HCC patients, we investigated two independent cohorts of HCC patients: cohort 2 comprising 189 patients and cohort 3 comprising 102 patients (Supporting Table 1). Kaplan-Meier analysis revealed that low lncRNA-PRAL genomic DNA level in HCC tissues was significantly correlated with markedly reduced tumor-free survival and overall survival in HCC patients (Fig. 6A,B). Significantly, in the two independent HCC cohorts, the multivariate analysis confirmed that low lncRNA-PRAL genomic DNA level in HCC tissues was an independent predictor for the reduced tumor-free survival of HCC patients (Supporting Tables 4 and 5). Moreover, the integrated survival analysis with genomic level of lncRNA-PRAL, tumor number, and portal vein thrombus indicated a more efficient prognostic predict for HCC patients (Supporting Fig. 11). Taken together, the clinical data on our patients support the conclusion that genomic DNA deletion of lncRNA-PRAL locus is the potent genetic factor predisposing HCC patients with poor prognosis of the disease.
Tumor suppressor gene p53 has been described to be inactivated in HCC through various mechanisms. (30,31) In order to verify the relationship between lncRNA-PRAL deletion and the level of p53 in clinic samples, we determined the level of p53 in HCC tissues using immunohistochemistry (IHC) prepared from cohort 2 patient samples without p53 deletion, R249S mutation, or HBV X protein (HBx) expression (n 5 91). Among these patient samples, strong p53 staining, with a score of IHC intensity over 2 (11), was detected in 37 of 47 tumor specimens with higher lncRNA-PRAL genomic levels. In contrast, 27 of 44 tumor samples with lower lncRNA-PRAL genomic levels showed very low p53 levels with a score of IHC intensity less than 1 (1) (Fig. 6C). More important, lower lncRNA-PRAL genomic DNA levels in patients with lower p53 IHC score showed a much worse prognosis than higher lncRNA-PRAL genomic DNA levels in patients with higher level of p53 (Fig.  6D). These results demonstrated that genomic alterations of lncRNA-PRAL was a clinical risk factor and showing the quantitative evaluation of p53 staining intensity from IHC analysis. Plot of a box plot (25%-75%) with whiskers to minimal and maximal of all the score data was used. The statistical differences between the two groups were analyzed by one-way ANOVA. **P < 0.01. (D) Tumor-free survival (left) and overall survival (right) between HCCs with lower and higher lncRNA-PRAL genomic DNA in cohort 2 HCC without p53 deletion, R249S mutation, or HBx expression. The median value in each cohort was chosen as the cut-off point. LINE1 (long interspersed element-1) as the genomic level internal control.