Papers in Press. Published April 17, 2019 as doi:10.1373/clinchem.2018.301150 The latest version is at http://clinchem.aaccjnls.org/cgi/doi/10.1373/clinchem.2018.301150

Clinical Chemistry 65:7 Diagnostics 000–000 (2019)

Noncoding Serve as Diagnosis and Prognosis for Hepatocellular Carcinoma Chang Tan,1,2† Jingyi Cao,1,2† Lu Chen,3,4,5† Xiaochen Xi,1 Siqi Wang,1 Yumin Zhu,1 Liuqing Yang,1 Longteng Ma,6 Dong Wang,1 Jianhua Yin,6* Ti Zhang,3,4,5* and Zhi John Lu1*

BACKGROUND: Reliable noninvasive biomarkers for hep- fragment alone promoted cancer cell proliferation and atocellular carcinoma (HCC)7 diagnosis and prognosis clonogenic growth. are urgently needed. We explored the potential of not only (miRNAs) but other types of noncod- CONCLUSIONS: Our results show that various ncRNA ing RNAs (ncRNAs) as HCC biomarkers. species, not only miRNAs, identified in the small RNA sequencing of plasma are also able to serve as noninvasive METHODS: Peripheral blood samples were collected from biomarkers. Particularly, we identified a domain of sr- 77 individuals; among them, 57 plasma cell-free RNA pRNA RN7SL1 with reliable clinical performance for transcriptomes and 20 exosomal RNA transcriptomes HCC diagnosis and prognosis. were profiled. Significantly upregulated ncRNAs and © 2019 American Association for Clinical Chemistry published potential HCC biomarkers were validated with reverse (RT)-qPCR in an independent validation cohort (60–150 samples). We particularly in- vestigated the diagnosis and prognosis performance and Hepatocellular carcinoma (HCC) is the most common biological function for 1 ncRNA , RN7SL1,8 type of primary liver cancer (1). Viruses, like hepatitis B and its S fragment. virus, are crucial in the transformation into HCC (2). The overall 5-year survival rate for all stages of HCC is RESULTS: We identified certain circulating ncRNAs es- only 15%; however, if diagnosed early, the survival rate caping from RNase degradation, possibly through bind- can be up to 70% (1). Thus, noninvasive biomarkers for ing with RNA-binding proteins: 899 ncRNAs were early diagnosis of HCC are urgently needed. highly upregulated in HCC patients. Among them, 337 Although many plasma microRNAs (miRNAs) are genes were fragmented long noncoding RNAs, 252 genes differentially produced in HCC patients vs healthy indi- were small nucleolar RNAs, and 134 genes were piwi- viduals (3, 4), most are unreliable HCC biomarkers, ow- interacting RNAs. Forty-eight candidates were selected ing to lack of independent validation and their failure to and validated with RT-qPCR, of which, 16 ncRNAs distinguish HCC from hepatitis B virus. Recently, other were verified to be significantly upregulated in HCC, types of noncoding RNAs (ncRNAs) have gained atten- including RN7SL1, SNHG1, ZFAS1, and LINC01359. tion as potential biomarker candidates. For instance, the Particularly, the abundance of RN7SL1 S fragment dis- expression of long noncoding RNA (lncRNA) H19 is criminated HCC samples from negative controls (area increased in plasma compared to plasmas of under the curve, 0.87; 95% CI, 0.817–0.920). HCC healthy controls (5), and plasma concentrations of 2 ln- patients with higher concentrations of RN7SL1 S frag- cRNAs, SNHG1 and RMRP, are higher in ment had lower survival rates. Furthermore, RN7SL1 S patients vs cancer-free controls (6).

1 MOE Key Laboratory of Bioinformatics, Center for Synthetic and Systems Biology, School 7 Nonstandard abbreviations: HCC, hepatocellular carcinoma; miRNAs, microRNAs; of Life Sciences, Tsinghua University, Beijing, China; 2 Tsinghua-Peking Joint Center for ncRNAs, noncoding RNAs; lncRNAs, long noncoding RNAs; cfRNAs, cell-free RNAs; EVs, Life Sciences, School of Life Sciences, Tsinghua University, Beijing, China; 3 Tianjin Med- extracellular vesicles; tRNAs, transfer RNAs; HD, healthy donors; CHB, chronic hepatitis ical University Cancer Institute and Hospital, Department of Hepatobiliary Cancer, Na- B; exoRNA, exosomal RNA; srpRNAs, signal recognition particle RNAs; SRP, signal recog- tional Clinical Research for Cancer, Tianjin, China; 4 Key Laboratory of Cancer Prevention nition particle; piRNAs, piwi-interacting RNAs; snoRNAs, small nucleolar RNAs; RBPs, and Therapy, Tianjin, China; 5 Tianjin’s Clinical Research for Cancer, Tianjin, China; 6 De- RNA-binding proteins; mRNAs, messenger RNAs; RT, reverse transcription; seq, se- partment of Epidemiology, Second Military Medical University, Shanghai, China. quencing; NC, negative control; AUC, area under the curve. * Address correspondence to: J.Y. at 800 Xiangyin Rd., Shanghai 200433, People’s Re- 8 HumanGenes:RN7SL1,RNA,7SL,cytoplasmic1;SNHG1,smallnucleolarRNAhostgene public of China. E-mail [email protected]. T.Z. at Huan Hu Xi Road, Ti Yuan Bei, He 1; ZFAS1, ZNFX1 antisense RNA 1; LINC01359, long intergenic non-protein coding RNA Xi District, Tianjin 300060, People’s Republic Of China. E-mail [email protected]. 1359; H19, H19, imprinted maternally expressed transcript; RMRP, RNA component of Z.J.L. at School of Life Sciences, Tsinghua University, Beijing, China 100084. Fax +86- mitochondrial RNA processing endoribonuclease; XIST, X inactive specific transcript; 10-62789217; e-mail [email protected]. NEAT1, nuclear paraspeckle assembly transcript 1; MIR122, microRNA 122; HEIH, hepa- † C. Tan, J. Cao, and L. Chen contributed equally. tocellular carcinoma upregulated EZH2-associated long non-coding RNA; LINC01225, Received December 21, 2018; accepted March 12, 2019. LINC01225 ; CYTOR, cytoskeleton regulator RNA; MIR21, microRNA 21; Previously published online at DOI: 10.1373/clinchem.2018.301150 MIR192, microRNA 192; MIR801, microRNA 801; MIR1228, microRNA 1228. © 2019 American Association for Clinical Chemistry

1

Copyright (C) 2019 by The American Association for Clinical Chemistry Published research on cell-free RNA (cfRNA) in BLOOD SAMPLING AND CELL-FREE RNA PURIFICATION cancer suggests that RNA fragments exist in plasma or Peripheral whole blood samples (5–10 mL) were col- extracellular vesicles (EVs), such as exosomes and mi- lected from participant patients in K2EDTA tubes. crovesicles (7). These RNA fragments may be processed Tubes were inverted 8–10 times and centrifuged (820g, posttranscriptionally and may be developmentally regu- 10 min, 4 °C). Blood cells were removed by use of a lated and tumor-specific. For example, EVs from human centrifuge at high speed (14000g, 1 min, 4 °C). Super- primary fibroblasts contain both full-length human hY5 natant was collected, aliquoted, and stored at Ϫ80 °C, and shorter processed fragments, whereas cancer cell- within 2 h after collection. derived EVs only contain a processed hY5 fragment, A modified protocol was used to isolate cfRNA from which promotes apoptotic cell death (8). In another ex- plasma with ZYMO Quick-cfRNA Serum & Plasma kit ample, the 5Ј fragment of glycine transfer RNA (tRNA) (R1059). DNase I was added, followed by Proteinase K represses the expression of genes associated with the en- treatment. cfRNA was purified with Zymo-Spin IC Col- dogenous retroelement, MERVL, in both embryonic umn according to the manufacturer’s protocol. stem cells and embryos (9). Although biological functions for fragments of Y cfRNA LIBRARY CONSTRUCTION RNA and tRNA in plasma and EVs have been demon- We used NEBNext® Multiplex Small RNA Library Prep strated, little research has focused on other types of Set to construct the small RNA libraries, and Illumina cfRNA fragments, such as lncRNAs (10). XIST has a HiSeq ϫ10 sequencing platform to sequence libraries structural fragment that is necessary for X – and generate 150-bp paired-end reads. inactivated gene silencing (11). LncRNA NEAT1_2 also has structural fragments (12) that promote NEAT1_2 DATA PROCESSING AND DIFFERENTIAL EXPRESSION stabilization. ANALYSIS Little is known about the biological functions of Library quality was evaluated with FastQC (version extracellular RNA fragments and the utility of extracel- 0.11.7) on raw read1. Adapters were removed by cut- lular RNA fragments as diagnostic biomarkers. To un- adapt (version 1.16) with the parameter setting of: –a ravel the function of extracellular RNA fragments and AGATCGGAAGAGCACACGTCTGAACTCCAGT- find robust noninvasive biomarkers for HCC, we se- CAC. Trimmed reads were mapped to the human quenced extracellular RNA from plasma and exosomes rRNAs, and rRNAs were removed with Bowtie2 (version and comprehensively interrogated their transcriptome. 2.2.9) with the following settings: –norc, –sensitive-local, –no-unal. Unmapped reads from the rRNA alignment Materials and Methods step were sequentially aligned to human miRNAs, piwi- interacting RNAs (piRNAs), Y RNAs, signal recognition STUDY DESIGN In the identification phase of our studies, 45 plasma particle RNAs (srpRNAs), tRNAs, small nucleolar cfRNA libraries containing 31 HCC, 11 healthy donors (snoRNAs), small nuclear RNAs, vault RNAs, lncRNAs, (HDs), and 3 chronic hepatitis B (CHB) samples (15 messenger RNAs (mRNAs), and transcripts of uncertain CHB individuals pooled into 3 libraries) were sequenced. coding potential. The human piRNA genome sequence Four exosomal RNA (exoRNA) libraries containing 2 was downloaded from piwiRNABank, and the other hu- HCC (10 HCC individuals pooled into 2 libraries) and 2 man genome sequences were downloaded from GEN- HD (10 HD individuals pooled into 2 libraries) were also CODE (Release v27). sequenced. In the quantification phase, differentially ex- Published data sets were downloaded from NCBI pressed genes were identified. In the validation phase, Gene Expression Omnibus under accession numbers performance assessments of selected RNA biomarkers GSE100207 (exoRBase), GSE77509 (long RNA-seq), were evaluated (see Fig. 1 in the Data Supplement that and GSE76903 (small RNA-seq). accompanies the online version of this article at Because of the fragmentation and scarcity character- http://www.clinchem.org/content/vol65/issue7). istic of cfRNA reads, we divided transcripts into 30-bp Study participant characteristics are presented in Ta- bins with 15-bp overlap. Raw read counts of each bin ble 1 and File 1 in the online Data Supplement. No were counted with featureCounts (Version 1.5.2). Dif- significant differences were found in age and sex between ferential expression analysis was calculated at bin level to the identification and validation cohort. In the HCC avoid the dilution phenomenon caused by uneven reads group, fewer patients had lower serum ␣ feto-protein distribution per transcript. Wilcoxon rank sum test and concentrations in the identification cohort than in the edgeR (Version 3.20.9) were used to find differentially validation cohort (52% vs 75%, respectively). The HCC expressed bins, selected as HCC biomarker candidates ͉ ͉ Ͼ Barcelona Center Liver Cancer stages ratio was well bal- (cutoff: log2(fold change) 1 and false-detection rate anced between the identification and validation cohort. Ͻ0.05).

2 Clinical Chemistry 65:7 (2019) ncRNAs for HCC Diagnosis and Prognosis Biomarkers

REVERSE TRANSCRIPTION qPCR VALIDATION incubation for 3 h, mixed medium was assessed with the Total RNA was extracted from plasma with TRIzol re- Nanodrop2000 (Thermo) at 490 nm. agent (13). Total RNA was reverse-transcribed for polyA tailing-based (200 ␮L) and random hexamer primer- COLONY FORMATION ASSAY based (5 ng) reverse transcription (RT)-qPCR reactions. Colony formation assays were performed to assess the We analyzed data by the comparative Ct method, with effects of candidate RNAs (16). Three overexpression glyceraldehyde 3-phosphate dehydrogenase as an exoge- cell lines and an empty-vector transfected cell line was nous control. mRNA was isolated from plasma with expanded and harvested with trypsin-EDTA. After incu- miRcute Serum/Plasma miRNA Isolation Kit (TIANGEN, bating at 37 °C for 2 weeks, cells were stained with 0.5% DP503). RT reactions and single-primer qPCR were con- crystal violet for 5 min and colonies were counted with ducted with miRcute Plus miRNA First-Strand cDNA ImageJ. Synthesis Kit and miRcute Plus miRNA qPCR Detec- tion Kit (TIANGEN, KR211 and FP411). MIR1228 DATA AVAILABILITY was used as a reference gene. Primer sequences for RT- The high-throughput sequencing data have been submit- qPCR are listed in File 2 in the online Data Supplement. ted to the GEO under accession number GSE123972. We obtained primers with the best performance that have the single and correct amplification products. Each Results Ct value is the average of 3 replicates, and the detection limit is lower than 0.1 pg. RNA SPECIES DETECTED IN PLASMA AND EXOSOME Total RNA concentration, raw-reads quality, depth of

SURVIVAL ANALYSIS sequencing, and abundance of reads mapped to each We downloaded 157 RNA-sequencing (seq) data from RNA type were determined in the 45 plasma cfRNA the Cancer Genome Atlas liver hepatocellular carcinoma libraries and 4 exoRNA libraries (see Fig. 2 in the online Data Supplement). One HCC cfRNA library and 1 HD patient database along with clinical data from the cfRNA library were excluded from analysis owing to poor National Cancer Institute Cancer Genomics Hub quality. Among these raw reads, 85%–90% mapped to (CGHub) (14). In the Kaplan–Meier survival analysis, the human GRCh38 reference sequence. Although the the survival data included vital status and days to death. library construction procedure did not include rRNAs Patients were classified into low-expression (98 patients) depletion step, the ratio of rRNAs to all mapped reads and high-expression (59 patients) groups according to was approximately 5%, suggesting that the blood RNA whether their RN7SL1 S fragment reads per kilobase per transcriptome was captured without rRNA contamina- million were lower or higher than the mean expression tion. This result is in accordance with previous studies value among all patients. Kaplan–Meier survival analysis (17, 18). (15) was applied in a 5-year survival analysis, and the As previously reported, the most abundant RNA log-rank test was used to ascertain the significance via the species detected by plasma and exosome RNA-seq were survival package. miRNA and Y RNA (Fig. 1A). To compare the distribu- tion of RNA species among plasma, exosome, and tissue, EXOSOME ISOLATION AND CHARACTERIZATION we downloaded 3 published HCC data sets (exosome Plasma from 5 individuals was mixed with protease long RNA-seq, tissue long RNA-seq, and tissue small inhibitor mixture (Cwbio, cw2200) and centrifuged RNA-seq) (8, 19). miRNA was the most abundant RNA (12000g, 20 min, 4 °C). After resuspending in cold PBS, type across the 3 small RNA-sequencing methods. ␮ sample was filtered (0.22 m). Cold PBS (20 mL) was Among the leftover RNA species, plasma and exosome added and then centrifuged (110000g, 4 °C, 4 h). The small RNA-seq detected more Y RNA, whereas tissue pellet containing the exosomes was resuspended in cold small RNA-seq identified more piRNA (Fig. 1A). Com- PBS. The modal value, concentration, and size distribu- pared to 2 sets of long RNA-seq data, 3 sets of small tion of the isolated exosomes were acquired by the Nano- RNA-seq detected very few reads originating from long sight NS300 (Mastersizer) and transmission electron RNA species, such as mRNA and lncRNAs (Fig. 1A). For microscopy. all RNA types, the length distribution peaks were at 21 nt and 32 nt, which were attributable to the most abundant MTS CELL PROLIFERATION ASSAY small ncRNAs (miRNA and Y RNA), in accordance with HepG2 cells were transfected with negative control (NC, Yuan et al. (20) (see Fig. 3 and 4A in the online Data empty vector) or RN7SL1 S fragment overexpression vec- Supplement). We also detected small RNA fragments tor and then cultured at an initial density of 1 ϫ 103 that originated from long ncRNAs, such as lncRNAs and cells/well. At the indicated times (0 h, 24 h, 48 h, 72 h, srpRNAs. Although small ncRNAs were abundant in 96 h), 20 ␮L of 5 g/L MTS (Promega) was added. After plasma and exosomes, the diversity of these small

Clinical Chemistry 65:7 (2019) 3 Fig. 1. Characteristics of cell-free RNA-seq. (A), Reads distribution of each RNA species among plasma, exosome, and tissue. Small and long RNA data were analyzed separately. (B), Left panel, number of genes identified by plasma small RNA-seq belonging to each noncoding RNA species. Right panel, reads distribution of noncoding RNA species among all HCC individuals.

ncRNAs was lower than that of lncRNAs (Fig. 1B). 40% to 70% among different HCC samples (see Fig. 4B These statistical results indicate that long and small in the online Data Supplement). Owing to variability in ncRNAs are potential HCC biomarkers. these genes, we used the Wilcoxon rank sum test to find differentially expressed genes. Finally, we identified 899 SELECT UPREGULATED LNCRNAS AS HCC BIOMARKER highly expressed ncRNAs in HCC patient plasma [with Ͼ CANDIDATES the cutoff: log2(fold change) 1 and false-detection rate In plasma cfRNA-seq, the 20 genes with the highest Ͻ0.05, see File 3 in the online Data Supplement]. Al- abundance across all HCC samples were Y RNA and though most reads were mapped to miRNA and Y RNA miRNAs. The proportion of the top 20 genes varied from transcripts, the majority of differentially expressed genes

4 Clinical Chemistry 65:7 (2019) ncRNAs for HCC Diagnosis and Prognosis Biomarkers

were lncRNAs and snoRNAs (Fig. 2A). Among the up- RN7SL1 S FRAGMENT CAN PROMOTE CANCER CELL regulated genes, 37.9%, and 28.4% were lncRNAs and PROLIFERATION AND CLONOGENIC GROWTH snoRNAs, respectively. In accordance with the reads distribution pattern in We selected 22 lncRNA, 11 snoRNA, 1 piRNA, and plasma cfRNA sequencing, RN7SL1 S fragment reads 1 srpRNA as candidates on the basis of differential ex- were enriched in HCC exosome compared to HD exo- pression analysis. We also selected 8 miRNAs and 5 ln- some (Fig. 3A). Indeed, dynamic feedback between cRNAs (see Table 2 in the online Data Supplement), breast tumor cells and stromal cells, via release of exo- which were reported as noninvasive HCC biomarkers somes containing RN7SL1, has been shown to activate previously (21–26). Among the previously published pattern recognition receptors of tumor cells, and pro- HCC biomarkers, expression levels of 4 out of 5 mote tumor growth and metastasis (27). To determine lncRNAs (HEIH, LINC01225, CYTOR, NEAT1) and 3 whether RN7SL1 is loaded as cargo in circulating EVs, out of 7 miRNAs (MIR21, MIR192, MIR801) were in we isolated exosomes from plasma. There were no obvi- accordance with published results (see Table 2 and ous differences in exosome size between HD and HCC Fig. 5 in the online Data Supplement), suggesting that plasma (Fig. 4A). The shape of exosomes was typical of lncRNAs are reliable HCC plasma biomarkers. EVs. The RN7SL1 S fragment was predominantly en- riched in exosomes in the HD or HCC group compared S FRAGMENT OF RN7SL1 AS A ROBUST HCC BIOMARKER to plasma supernatant, and the HCC group was signifi- Among the 35 candidates selected from our discovery cantly higher than the HD group. MIR122 was used as a cohort, 16 candidates were verified to be significantly NC because this miRNA is mainly present in exosome- upregulated in HCC plasma in our validation cohort free plasma supernatant (Fig. 4B). The above result indi- (see Table 3 in the online Data Supplement), such as cates that RN7SL1 might escape from RNA degradation RN7SL1, ZFAS1, SNHG1, and LINC01359. Among by the protection of exosome membrane and binding them, RN7SL1 was the most significantly upregulated proteins. It is also consistent with a previous study in gene in HCC samples (P Ͻ 0.0001) (Fig. 2B). Based breast cancer, which showed that exosomal RN7SL1 was involved in the tumor microenvironment (27). on our results, and the abnormal transcription of To examine the biological function of exosomal RN7SL1 in breast cancer (27), we chose RN7SL1 for RN7SL1, the full-length and S fragment of RN7SL1 were detailed validation and functional studies. MIR192 overexpressed in HepG2 cells, and the empty vector was used as a positive control. acted as a NC. Proliferation of HepG2 cells increased Two structural fragments, the S fragment and the remarkably on overexpression of full-length and S frag- Alu fragment, had different RNA-binding proteins in ment of RN7SL1, compared to the NC (Fig. 4C). Fur- RN7SL1. By comparing the reads distribution tracks of thermore, the overexpression of full-length and S frag- HCC samples with HD and CHB samples, we found ment of RN7SL1 significantly enhanced clonogenic that there was a significant increase of reads coverage in formation of HepG2 cells (Fig. 4D). Thus, the exosomal the RN7SL1 S fragment region in the HCC group (Fig. RN7SL1 S fragment promoted cancer cell proliferation 3A). We examined the expression level of full-length and differentiation. RN7SL1 and RN7SL1 fragments separately in the vali- dation cohort by 2 different methods: (a) the polyA RN7SL1 S FRAGMENT FOR HCC DIAGNOSIS AND PROGNOSIS tailing-based reverse-transcribed method, which is suit- The diagnosis performance of plasma cell-free RN7SL1 S able for examining short and fragmented cfRNAs, and fragment was compared to the full-length transcript in (b) the traditional random hexamer primer-based the validation cohort. The area under the curve (AUC) reverse-transcribed method, which is suitable for exam- of plasma RN7SL1 S fragment (AUC, 0.87; 95% CI, ining long cfRNAs (see Methods). All products were con- 0.817–0.920) was higher than that of the full-length firmed by Sanger sequencing after TA cloning (see Fig. 5 transcript (AUC, 0.75; 95% CI, 0.671–0.826) and in the online Data Supplement), and the length of each exosomal RN7SL1 S fragment (AUC, 0.75; 95% CI, product is shown in Fig. 3B. The RN7SL1 S fragment 0.526–0.975), indicating that plasma RN7SL1 S frag- detected by polyA tailing-based RT method demon- ment is a better HCC diagnostic biomarker (see Fig. 7 strated the highest significance and fold change (10.29, in the online Data Supplement). The fold change of P Ͻ 0.0001; Fig. 2B). The Alu fragment did not show the RN7SL1 S fragment in stage A vs HD was a little significant differences between HCC and CHB or HD. bit lower than that of stages B and C (see Fig. 8 in the The fold change of S fragment in HCC plasma vs HD online Data Supplement). This result was consistent plasma was higher than that of HCC tissue vs HD tissue with the ROC analysis result, showing that the AUC (see Fig. 6 in the online Data Supplement). These results of stage A (AUC, 0.78; 95% CI, 0.695–0.869) was suggest that the RN7SL1 S fragment discriminates HCC also a little bit lower than that of stages B (AUC, 0.93; plasma from CHB and HD plasma. 95% CI, 0.869–0.986) and C (AUC, 0.92; 95% CI,

Clinical Chemistry 65:7 (2019) 5 Fig. 2. RN7SL1 S fragment was significantly upregulated in HCC plasma. (A), RNA type distribution of upregulated genes determined by comparing HCC plasma with control. Percentages of each type are indicated on the right side. (B), Validation of selected lncRNA candidates by use of RT-qPCR. Four genes (ZFAS1, RN7SL1, SNHG1, LINC01359) were verified inacohortofHD=30,CHB=20,andHCC=100,and2differentRT-qPCRstrategieswereusedtoverifytheSdomainofRN7SL1withdifferent lengths of its fragments. SNHG1 and LINC01359 dropped out in certain samples, and the number of positive amplifications was indicated in the figure. P1 and P2 referred to 2 different regions of SNHG1.

6 Clinical Chemistry 65:7 (2019) ncRNAs for HCC Diagnosis and Prognosis Biomarkers

Fig. 3. Reads distribution of RN7SL1 for cfRNA-seq and exoRNA-seq. (A), Reads distribution of RN7SL1 for cfRNA-seq and exoRNA-seq. The secondary domains of RN7SL1 are shown at the top of the integrative genomics viewer tracks. (B), The validation of RN7SL1 RT-qPCR products by agarose gel electrophoresis and TA cloning with Sanger sequenc- ing. † indicated the length of the RT-qPCR product verified by Sanger sequencing.

0.850–0.998) (Fig. 4E). Furthermore, the expression These results indicate that plasma RN7SL1 S fragment level of the RN7SL1 S fragment clearly classified liver is a good diagnostic and prognostic biomarker, al- hepatocellular carcinoma patients into 2 subclasses though RN7SL1 S fragment alone may not be a robust with different survival rates, whereas RN7SL1 full- biomarker for HCC early diagnosis. length transcript did not (Fig. 4F). Results of different cutoffs for patient classification are shown in Fig. 9 in INTERACTIONS WITH RNA-BINDING PROTEINS MAY PROMOTE the online Data Supplement. These cutoffs generated RN7SL1 STABILITY consistent results, suggesting the robustness of this Certain RNAs might be able to avoid RNase degradation marker in the survival prediction. Additionally, no sig- by binding with specific proteins and as cargo loaded nificant correlations were observed between ␣-feto- in exosomes (28, 29). We found that high-abundance protein or alkaline phosphatase and RN7SL1 expres- exoRNAs tend to bind with more RNA-binding proteins sion (see Fig. 10 in the online Data Supplement). (RBPs) than low-abundance and background group (P

Clinical Chemistry 65:7 (2019) 7 Fig. 4. RN7SL1 S fragment was enriched in exosome and could promote cancer cell growth. (A),CharacterizationsofexosomesthatpurifiedfromHDandHCC.ThecurveindicatesthediameterdistributionofexosomesbyNanosightand the transmission electron micrograph shows the external morphology of exosomes. (B), RN7SL1 was enriched in exosomes and differentially expressed between HCC and HD. MIR122 was selected as a supernatant-enriched marker. (C), RN7SL1 and S fragment promoted HepG2 cell proliferation. Cell proliferation was assessed by the 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide assay at 24 h, 48 h, 72 h, and 96 h, which is shown as the absorbance at OD490. *P < 0.05 compared with the NC treatment group. (D), RN7SL1 and S fragment enhanced the capacity of HepG2 cells colony formation. ***P < 0.001 compared with the NC treatment group. (E), ROC analysis of plasma RN7SL1 S fragment performance as a biomarker for different HCC stages in the validation cohort. S fragment yielded an AUC of 0.782 with 68.5% sensitivity and 72.5% specificity for HCC stage A; S fragment yielded an AUC of 0.927 with 98.6% sensitivity and 74.4% specificity for HCC stage B; S fragment yielded an AUC of 0.924 with 97.1% sensitivity and 76.2% specificity for HCC stage C; S fragment yielded an AUC of 0.869 with 90.0% sensitivity and 70.0% specificity for HCC all stage. (F), Kaplan–Meier analysis of overall survival in the Cancer Genome Atlas liver hepatocellular carcinoma cohort. Participants were classified according to the expression of RN7SL1 S fragment. The P value for Kaplan– Meier analysis was determined with the log-rank test. NS, not significant; Exo, exosome; Sup, supernatant.

8 Clinical Chemistry 65:7 (2019) ncRNAs for HCC Diagnosis and Prognosis Biomarkers

Fig. 5. Structural domains and RBP binding sites of RN7SL1. (A), A schematic diagram of the secondary structure indicating RN7SL1 and its binding protein. The red line indicates the RN7SL1 region that is highly expressed in HCC. (B), The crystal structure indicates the S domain of RN7SL1 and its binding protein. The red nucleic acid region indicates the S domain fragment of RN7SL1 that is highly expressed in HCC. RBD, RNA binding domain.

value Ͻ2.2eϪ16) (see Fig. 11 in the online Data HCC patients, binds SRP (Fig. 5, A and B). In addi- Supplement). tion, we found a large number of RBPs that bind with RN7SL1 is a component of the signal recognition the S fragment region of RN7SL1 by analyzing RBP particle (SRP) cytoplasmic ribonucleoprotein complex binding sites identified by CLIP-seq (see Fig. 12 in the (30) (Fig. 5A). By analyzing the crystal structure of online Data Supplement). By analyzing RNA produc- srpRNA and protein complexes, we found that the S tion levels of these RBPs between HCC patient tissue fragment of RN7SL1, which is highly expressed in and NCs, we found that 7 RBPs are significantly up-

Clinical Chemistry 65:7 (2019) 9 regulated in HCC samples (see Fig. 13 in the online misidentified (36). Whether these genes are truly dif- Data Supplement), especially for IGF2BP3, which was ferentially expressed needs further validation in indi- reported to serve as an m6A reader and promote the viduals by a low-throughput method like RT-qPCR. stability of their target RNAs (31). In addition to Previously, Chen et al. reported the increased pro- IGF2BP3, other RBPs are also related to RNA m6A duction of the BC200 lncRNA (related to RN7SL1)in modifications, such as METTL13, which was reported cancer cells (37). Abdelmohsen, et al. reported that to play a role in promoting glioma stem-like cells RN7SL1 was highly expressed in liver cancer tissues (38). growth and self-renewal through enhancing SOX2 sta- They found that RN7SL1 could interact with TP53 bility (32). These results indicate that the RN7SL1 S mRNA to reduce p53 . Additionally, an fragment may escape RNA degradation via m6A mod- exRNA database, ExoRBase, shows highly expressed ification. Differences in m6A modification levels be- RN7SL1 in exosomes of HCC patients (7). However, tween RN7SL1 S and the Alu fragment may explain these studies did not investigate the expression and func- why the RN7SL1 S fragment accumulates in HCC tion of extracellular RN7SL1 or examine the potential for patients but not healthy individuals. this ncRNA serving as a biomarker. Despite the potential of circulating ncRNAs as bio- Discussion markers, there are several critical problems that prevent the use of cell-free ncRNAs for clinical diagnosis and We identified a domain of srpRNA RN7SL1 that existed prognosis. So far, no standardized protocols are estab- stably in plasma and exosomes and was found to discrim- lished for sample collection, library construction, data inate HCC patients from healthy and CHB with good normalization, and analysis. Additionally, there is no sensitivity and specificity. Our study shows that various standard intrinsic RNA that could serve as a housekeep- noncoding RNA species, not only miRNAs, can be iden- ing gene, and this limitation results in poor comparability tified from the small RNA-seq library of plasma that have between different studies. The control (MIR1228)we the potential to serve as biomarkers for HCC diagnosis used (39) was reported to be stable in plasma and still and prognosis. needs to be validated in further studies. Moreover, the Compared with traditional imaging methods, non- patient follow-up time was usually very short, thus lim- invasive RNA biomarkers are more dynamically regu- iting the analysis of plasma samples for prognosis. To lated, tissue-specific, and abundant in extracellular envi- improve the robustness of extracellular ncRNAs as clini- ronments. Many papers have reported finding potential cal prognostic markers, collecting plasma data sets with HCC miRNA biomarkers. Because other kinds of adequate clinical information is needed, and the determi- ncRNAs have been discovered in plasma or EVs, recent nation of an accurate, significant, and practical baseline studies have focused on analyzing the aberrant expression needs further validation in a large cohort of clinical of other types of ncRNAs for the purpose of finding samples. reliable HCC diagnosis biomarkers. In our study, we established an efficient labora- tory protocol for cfRNA extraction and sequencing library construction. Although a previous study (33) Author Contributions: All authors confirmed they have contributed has shown that pooling samples can save cost without to the intellectual content of this paper and have met the following 4 much loss of precision, pooling could still result in requirements: (a) significant contributions to the conception and design, false positives in RNA-seq and miss the information of acquisition of data, or analysis and interpretation of data; (b) drafting individual patients. These limitations make the pool- or revising the article for intellectual content; (c) final approval of the ing strategy somewhat ill-suited for cancer study if no published article; and (d) agreement to be accountable for all aspects of further validation is conducted, owing to the hetero- the article thus ensuring that questions related to the accuracy or integ- rity of any part of the article are appropriately investigated and resolved. geneity of cancer. It has been reported that pooling strategies may help to reduce the biological variability J. Cao, statistical analysis; L. Chen, provision of study material or and thus increase the power of statistical tests (34, 35), patients; X. Xi, statistical analysis; L. Yang, statistical analysis; L. Ma, provision of study material or patients; D. Wang, administrative sup- based on the assumption that the signal derived from port; J. Yin, provision of study material or patients; T. Zhang, administra- pooled samples is equivalent to the average of expres- tive support, provision of study material or patients. sion signals from an individual-based design. This as- Authors’ Disclosures or Potential Conflicts of Interest: Upon man- sumption likely is not true for some deregulated genes, uscript submission, all authors completed the author disclosure form. Dis- particularly for tumor samples with high heterogene- closures and/or potential conflicts of interest: ity. For instance, Rajkumar et al. reported that pooling Employment or Leadership: None declared. strategies displayed good sensitivity (approximately Consultant or Advisory Role: None declared. 90%) and specificity (approximately 86%) in RNA- Stock Ownership: None declared. seq, whereas some differentially expressed genes were Honoraria: None declared.

10 Clinical Chemistry 65:7 (2019) ncRNAs for HCC Diagnosis and Prognosis Biomarkers

Research Funding: Z.J. Lu, National Key Research and Development Expert Testimony: None declared. Plan of China (2016YFA0500803), National Natural Science Founda- Patents: None declared. tion of China (31771461, 31522030), Fok Ying-Tong Education Foundation, Beijing Advanced Innovation Center for Structural Biol- Role of Sponsor: The funding organizations played a direct role in the ogy, Bio-Computing Platform of Tsinghua University Branch of China design of study, review and interpretation of data, preparation of man- National Center for Protein Sciences (Beijing). uscript, and final approval of manuscript. References

1. Ghouri YA, Mian I, Rowe JH. Review of hepatocellular ing cancer through the power of torrential data. Data- 27. Nabet BY, Qiu Y, Shabason JE, Wu TJ, Yoon T, Kim BC, et carcinoma: epidemiology, etiology, and carcinogene- base (Oxford) 2014;2014. al. Exosome RNA unshielding couples stromal activa- sis. J Carcinog 2017;16:1. 15. Efron B. Logistic regression, survival analysis, and the tion to pattern recognition receptor signaling in cancer. 2. Ashtari S, Pourhoseingholi MA, Sharifian A, Zali MR. Kaplan–Meiercurve.JAmStatAssoc1988;83:414–25. Cell 2017;170:352–66. Hepatocellular carcinoma in Asia: prevention strategy 16. Friedenstein AJ, Gorskaja UF, Kulagina NN. Fibroblast 28. Mitchell PS, Parkin RK, Kroh EM, Fritz BR, Wyman SK, and planning. World J Hepatol 2015;7:1708–17. precursors in normal and irradiated mouse hematopoi- Pogosova-Agadjanyan EL, et al. Circulating microRNAs 3. Xu J, Wu C, Che X, Wang L, Yu D, Zhang T, et al. Circu- etic organs. Exp Hematol 1976;4:267–74. as stable blood-based markers for cancer detection. lating microRNAs, miR-21, miR-122, and miR-223, in 17. Godoy PM, Bhakta NR, Barczak AJ, Cakmak H, Fisher S, Proc Natl Acad Sci USA 2008;105:10513–8. patients with hepatocellular carcinoma or chronic hep- MacKenzie TC, et al. Large differences in small RNA 29. Viereck J, Thum T. Circulating noncoding RNAs as bio- atitis. Mol Carcinog 2011;50:136–42. composition between human biofluids. Cell Rep markers of cardiovascular disease and injury. Circ Res 4. Qu KZ, Zhang K, Li HR, Afdhal NH, Albitar M. Circulating 2018;25:1346–58. 2017;120:381–99. microRNAs as biomarkers for hepatocellular carcinoma. 18. Tan YW, Ge GH, Pan TL, Wen DF, Chen L, Yu XJ, et al. A 30. BateyRT,RamboRP,LucastL,RhaB,DoudnaJA.Crystal J Clin Gastroenterol 2011;45:355–60. serum microRNA panel as potential biomarkers for structureoftheribonucleoproteincoreofthesignalrec- 5. Zhang K, Luo Z, Zhang Y, Zhang L, Wu L, Liu L, et al. hepatocellular carcinoma related with hepatitis B virus. ognition particle. Science 2000;287:1232–9. Circulating lncRNA H19 in plasma as a novel biomarker PLoS One 2014;9. 31. Huang HL, Weng HY, Sun WJ, Qin X, Shi HL, Wu HZ, et for breast cancer. Cancer Biomark 2016;17:187–94. 19. Yang Y, Chen L, Gu J, Zhang HS, Yuan JP, Lian QY, et al. al. Recognition of RNA N6-methyladenosine by IGF2BP 6. Lin YL, Leng QX, Zhan M, Jiang F. A plasma long non- Recurrently deregulated lncRNAs in hepatocellular car- proteins enhances mRNA stability and translation. Nat coding RNA signature for early detection of lung cancer. cinoma. Nat Commun 2017;8. Cell Biol 2018;20:1098. Transl Oncol 2018;11:1225–31. 20. Yuan T, Huang X, Woodcock M, Du M, Dittmar R, Wang 32. Visvanathan A, Patil V, Arora A, Hegde AS, Arivazhagan 7. Li S, Li Y, Chen B, Zhao J, Yu S, Tang Y, et al. ExoRBase: a Y,etal.PlasmaextracellularRNAprofilesinhealthyand A, Santosh V, Somasundaram K. Essential role of databaseofcircRNA,lncRNAandmRNAinhumanblood cancer patients. Sci Rep 2016;6:19413. METTL3-mediated m6A modification in glioma stem- exosomes. Nucleic Acids Res 2018;46:D106–D12. 21. Zhou J, Yu L, Gao X, Hu J, Wang JP, Dai Z, et al. like cells maintenance and radioresistance. 8. Chakrabortty SK, Prakash A, Nechooshtan G, Hearn S, Plasma microRNA panel to diagnose hepatitis B 2018;37:522–33. Gingeras TR. Extracellular vesicle-mediated transfer virus-related hepatocellular carcinoma. J Clin Oncol 33. Kendziorski C, Irizarry RA, Chen KS, Haag JD, Gould of processed and functional RNY5 RNA. RNA 2011;29:4781–8. MN. On the utility of pooling biological samples in mi- 2015;21:1966–79. 22. Yang F, Zhang L, Huo XS, Yuan JH, Xu D, Yuan SX, et al. croarray experiments. Proc Natl Acad Sci USA 9. Sharma U, Conine CC, Shea JM, Boskovic A, Derr AG, Long noncoding RNA high expression in hepatocellular 2005;102:4252–7. BingXY,etal.BiogenesisandfunctionoftRNAfragments carcinoma facilitates tumor growth through enhancer during sperm maturation and fertilization in mammals. of zeste homolog 2 in humans. Hepatology 34. Churchill GA. Fundamentals of experimental design for Science 2016;351:391–6. 2011;54:1679–89. cDNA microarrays. Nat Genet 2002;32:490–5. 10. Sugimoto Y, Vigilante A, Darbo E, Zirra A, Militti C, 23. ZhangC,YangX,QiQ,GaoYH,WeiQ,HanSW.LncRNA- 35. Simon RM, Dobbin K. Experimental design of DNA mi- D’Ambrogio A, et al. hiCLIP reveals the in vivo atlas of HEIH in serum and exosomes as a potential biomarker croarray experiments. Biotechniques 2003:16–21. mRNA secondary structures recognized by Staufen 1. in the HCV-related hepatocellular carcinoma. Cancer 36. Rajkumar AP, Qvist P, Lazarus R, Lescai F, Ju J, Nye- Nature 2015;519:491–4. Biomark 2018;21:651–9. gaard M, et al. Experimental validation of methods for 11. Wang CG, Wang LZ, Ding Y, Lu XY, Zhang GS, Yang JX, 24. Wang X, Zhang W, Tang J, Huang R, Li J, Xu D, et al. differential gene expression analysis and sample pool- et al. LncRNA structural characteristics in epigenetic LINC01225 promotes occurrence and metastasis of ing in RNA-seq. BMC Genomics 2015;16. regulation. Int J Mol Sci 2017;18. hepatocellular carcinoma in an epidermal growth fac- 37. Chen W, Bocker W, Brosius J, Tiedge H. Expression of 12. Yamazaki T, Souquere S, Chujo T, Kobelke S, Chong YS, tor receptor-dependent pathway. Cell Death Dis neural BC200 RNA in human tumours. J Pathol Fox AH, et al. Functional domains of NEAT1 architec- 2016;7. 1997;183:345–51. tural lncRNA induce paraspeckle assembly through 25. Li J, Wang XC, Tang JW, Jiang RQ, Zhang WJ, Ji J, Sun 38. Abdelmohsen K, Panda AC, Kang MJ, Guo R, Kim J, phase separation. Mol Cell 2018;70:1038–53.e7. BC. HULC and Linc00152 act as novel biomarkers in Grammatikakis I, et al. 7SL RNA represses p53 transla- 13. Chomczynski P, Sacchi N. The single-step method of predicting diagnosis of hepatocellular carcinoma. Cell tion by competing with HuR. Nucleic Acids Res RNA isolation by acid guanidinium thiocyanate- Physiol Biochem 2015;37:687–96. 2014;42:10099–111. phenol-chloroform extraction: twenty-something years 26. Wang ZM, Zou QT, Song M, Chen J. NEAT1 promotes 39. Hu J, Wang Z, Liao BY, Yu L, Gao X, Lu SH, et al. Human on. Nat Protoc 2006;1:581–5. cell proliferation and invasion in hepatocellular carci- miR-1228 as a stable endogenous control for the quan- 14. Wilks C, Cline MS, Weiler E, Diehkans M, Craft B, Martin noma by negative regulating miR-613 expression. tification of circulating microRNAs in cancer patients. C, et al. The Cancer Genomics Hub (CGHub): overcom- Biomed Pharmacother 2017;94:612–8. Int J Cancer 2014;135:1187–94.

Clinical Chemistry 65:7 (2019) 11 本文献由“学霸图书馆-文献云下载”收集自网络，仅供学习交流使用。

学霸图书馆（www.xuebalib.com）是一个“整合众多图书馆数据库资源，

提供一站式文献检索和下载服务”的24 小时在线不限IP 图书馆。 图书馆致力于便利、促进学习与科研，提供最强文献下载服务。

图书馆导航：

图书馆首页 文献云下载 图书馆入口 外文数据库大全 疑难文献辅助工具