Sage Journals: Discover world-class research

Abstract

BACKGROUND:

Early recurrence is the main obstacle for long-term survival of hepatocellular carcinoma (HCC) patients after curative resection.

OBJECTIVE:

We aimed to develop a long non-coding RNA (lncRNA) based signature to predict early recurrence.

METHODS:

Using bioinformatics analysis and quantitative reverse transcription PCR (RT-qPCR), we screened for lncRNA candidates that were abnormally expressed in HCC. The expression levels of candidate lncRNAs were analyzed in HCC tissues from 160 patients who underwent curative resection, and a risk model for the prediction of recurrence within 1 year (early recurrence) of HCC patients was constructed with linear support vector machine (SVM).

RESULTS:

An lncRNA-based classifier (Clnc), which contained nine differentially expressed lncRNAs including AF339810, AK026286, BC020899, HEIH, HULC, MALAT1, PVT1, uc003fpg, and ZFAS1 was constructed. In the test set, this classifier reliably predicted early recurrence (AUC, 0.675; sensitivity, 72.0%; specificity, 63.1%) with an odds ratio of 4.390 (95% CI, 2.120–9.090). Clnc showed higher accuracy than traditional clinical features, including tumor size, portal vein tumor thrombus (PVTT) in predicting early recurrence (AUC, 0.675 vs 0.523 vs 0.541), and had much higher sensitivity than Barcelona Clinical Liver Cancer (BCLC; 72.0% vs 50.0%), albeit their AUCs were comparable (0.675 vs 0.678). Moreover, combining Clnc with BCLC significantly increased the AUC, compared with Clnc or BCLC alone in predicting early recurrence (all $P<$ 0.05). Finally, logistic and Cox regression analyses suggested that Clnc was an independent prognostic factor and associated with the early recurrence and recurrence-free survival of HCC patients after resection, respectively (all $P=$ 0.001).

CONCLUSIONS:

Our lncRNA-based classifier Clnc can predict early recurrence of patients undergoing surgical resection of HCC.

Keywords

Hepatocellular carcinoma prognosis lncRNA-based classifier recurrence

1. Introduction

Hepatocellular carcinoma (HCC), is one of the most common types of cancer worldwide and is the second leading cause of cancer deaths in men [1]. Clinically, resection is the most widely used potentially curative treatment for HCC patients [2]. However, a high incidence of recurrence leads to the poor survival of HCC patients [3, 4]. It has been reported that the recurrence rate peaked at 1 year after HCC resection, and this recurrence pattern was mainly due to invasion or metastasis from the primary tumors [5]. Therefore, it is urgent to establish a classifier that can effectively stratify the HCC patients according to the risk of early recurrence and consequently enable optimal postoperative adjuvant therapies.

Although traditional staging systems, such as Barcelona Clinical Liver Cancer (BCLC) system, were useful for treatment guidance, they are inadequate for predicting survival who undergo hepatic resection [6]. Considerable efforts have been devoted to identifying molecular markers for predicting the recurrence of HCC patients, including immunohistochemical markers, protein-coding gene markers, microRNAs, and epigenetic biomarkers [7, 8, 9, 10]. However, due to the inconsistency of analytical platforms and complex etiologies of HCC, these molecular markers have not been widely introduced into clinical practice as expected initially, and most of them still need to be validated.

LncRNAs are RNAs longer than 200 nucleotides (nt) and lacking protein-coding abilities [11]. Abnormal expression of lncRNAs has been implicated in tumor development and progression, and accumulating studies have shown that the aberrantly expressed lncRNAs could potentially mark the spectrum of disease progression, and some of them have exerted diagnostic or prognostic values [12, 13, 14, 15]. However, whether incorporating multiple lncRNAs could improve the prediction of early recurrence for HCC remains unknown.

In this study, a screen for differentially expressed lncRNAs between HCCs and the adjacent normal liver tissues resulted in the identification of 15 upregulated and 5 downregulated lncRNAs. We further investigated the expression profiles of the 20 dysregulated lncRNAs in a training cohort of 160 HCC patients, and identified a nine-lncRNA classifier (Clnc) for early recurrence by using a linear support vector machine (SVM) model. Then the prognostic value of Clnc was validated in a test set of 161 patients. We also assessed the ability of the classifier to separate patients with respect to survival.

2. Materials and methods

2.1 Patients and tissue samples

The patients enrolled in the study were diagnosed with primary HCC and underwent curative resection between 2001 and 2013 at the Sun Yat-sen University Cancer Center (Guangzhou, China). All the tissues were snapfrozen in liquid nitrogen or kept in RNAlater immediately after surgical resection and then stored at $-$ 80 ${}^{\circ}$ C until analysis. For classifier construction, a total of 321 HCCs were collected and were randomly divided into training (160 patients) or test set (161 patients). Most of the patients were HBV infected excepted for 9 patients who were HBV negative. Additional 20 paired HCCs and the corresponding non-cancerous liver tissues were also obtained to confirm the aberrant expression of lncRNAs candidates. The inclusion criteria for patient enrollment were as follows: (1) all the samples including HCC and adjacent non-tumorous liver tissues were histologically confirmed, (2) no local or systemic treatment had been conducted before the operation, and no other anticancer therapy was administered before relapse, (3) none of the patients had distant metastasis, (4) none of the patients had severe complications or other malignant diseases. Patients who developed HCC recurrence within 3 months after surgery were also excluded, as it is suggested that such recurrence is not real recurrence but preexisting metastases before HCC resection. After hepatectomy, all patients were followed-up at 2–4 months intervals in the outpatient clinic with regular surveillance for recurrence using the serum alpha-fetoprotein (AFP) level, abdominal ultrasonography, and chest radiography. When tumor recurrence or metastasis was suspected, further examinations, including CT and hepatic angiography, were performed to confirm the recurrence. Biopsies were taken when necessary. The median follow-up was 58 months (range 2–153 months). The clinicopathological characteristics of the patients in the training set and test sets are summarized in Supplementary Table 1. This study was approved by the Institute Research Ethics Committee at the Cancer Center under the Institutional Review Board approval number of GZR2019-086. Informed consent was obtained from all patients and the study conforms with The Code of Ethics of the World Medical Association (Declaration of Helsinki), printed in the British Medical Journal (18 July 1964).

2.2 Selection of candidate lncRNAs

Two strategies were used to select candidate lncRNAs. Firstly a bioinformatics analysis was conducted based on lncRNA expression profiles (GSE54238, https://www.ncbi.nlm.nih.gov/gds) of 10 normal livers (N), 13 early HCCs and 13 advanced HCCs as reported previously [16]. After removing the lncRNA genes with low abundance, and those lacking the histone modifications (H3K4me1, H3K4me3, and H3K27Ac) associated with active gene expression, as revealed by the ENCODE ChIP-seq data [17], the lncRNAs which differentially expressed in HCCs were subjected to subsequent analysis. Nineteen candidate lncRNAs (Supplementary Table 2) were finally obtained from microarray data after excluding a number of lncRNAs due to failure to amplify specific gene fragments. In addition, we included 11 other lncRNAs that were implicated in the development and progression of HCC by previously published studies(Supplementary Table 3). Altogether, a total of 30 genes were selected to analyze their expression levels in the paired HCC and adjacent non-tumor liver tissues. The validated lncRNAs that were abnormally expressed in HCCs were used for classifier construction.

2.3 RNA extraction, reverse transcription (RT) and quantitative PCR (RT-qPCR)

Total RNA from HCC and non-cancerous tissues were extracted using TRIzol ${}^{\@setsize{\scriptsize}{8pt}{\viipt}{\@viipt}\textregistered}$ Reagent (Invitrogen, Carlsbad, CA, USA) according to the manufacturer’s instructions. First-strand cDNA was synthesized using the M-MLV cDNA Kit (Promega, Madison, WI, USA). The expression levels of lncRNAs were detected by RT-qPCR using the SYBR Green qPCR master mix (Applied Biosystems, Foster City, CA, USA). All reactions were performed on a LightCycler ${}^{\@setsize{\scriptsize}{8pt}{\viipt}{\@viipt}\textregistered}$ 480 (Roche Diagnostics, Germany) and were run in triplicate in each experiment. Relative expression levels of lncRNAs were calculated using the 2 ${}^{-\Delta\Delta\text{Ct}}$ method, with U6 as an internal control. The sequences of all primers are listed in Supplementary Table 4.

2.4 Data process and supervised classifier construction

The gene expression levels were discretized before model construction. The point corresponding to the minimum value for (1-sensitivity) ${}^{2}$ $+$ (1-specificity) ${}^{2}$ in the receiver operating characteristic (ROC) curve [18] for the training set was identified as the optimal cut-off point (Supplementary Fig. S1). The expression value larger than the cut-off point was assigned to 1, otherwise was 0 for each lncRNA. The optimal cut-off value was applied directly to the test set without re-estimation.

The 160 patients in the training set were divided into two groups according to their recurrence state within 1-year post-operation. Then a linear SVM-based model with an exhaustive search method was used for the training cohort to construct a supervised classifier for predicting the postsurgical recurrence within 1 year. Each lncRNA combination was defined as one classifier. Leave-one-out (LOO) cross-validation was used to evaluate the performance of each classifier (Fig. 1). The optimal classifier was determined when the value of (1-sensitivity) ${}^{2}$ $+$ (1-specificity) ${}^{2}$ reached a minimum. The classifier obtained was then used to classify 161 patients in the test cohort to evaluate its robustness. R software version 3.3.1 (https://www.R-project.org/) and the e1071 package with version 1.6-8 (David Meyer, Austria) were used to do the SVM model analysis.

Figure 1.

Study design. A schematic overview summarizes the construction and validation of the classifier. A total of 30 candidate lncRNAs were selected to analyze their expression levels in 20 HCC and non-tumor tissues. Among them, 20 lncRNAs were abnormally expressed in HCCs and then applied to establish the prognostic model. A total of 321 HCC patients from Sun Yat-sen University Cancer Center (Guangzhou, China) were collected and distributed into the training cohort ( $N=$ 160) and the validation cohort ( $N=$ 161). HCC, hepatocellular carcinoma. SVM, support vector machine.

Figure 2.

Performance evaluation of Clnc classifier for early recurrence prediction. A. ROC curves for the Clnc classifier and different clinical characteristics including tumor size, tumor number, differentiation, PVTT, and BCLC stage in predicting recurrence within 1 year after hepatectomy in the training (Left, $N=$ 160) and test (Right, $N=$ 161) sets. B. Kaplan-Meier curve for recurrence probability of the high-risk and low-risk groups stratified by Clnc in the training (Left) and test (Right) sets. In both data sets, log-rank $P$ -values $<$ 0.0001.

2.5 Statistical analysis

ROC curve analysis was carried out with MedCalc version 15.2.2, (MedCalc Software Ltd, Ostend, Belgium). The area under the curve (AUC) was used to measure prognostic or predictive accuracy. Survival analyses were performed using the GraphPad prism7.0 (GraphPad Software, San Diego, CA, USA). Survival curves are estimated for each group using the Kaplan-Meier method and the log rank test was used to compare the survival curves between groups. Regression analyses including Logistic regression and Cox Logistic regression were performed using the Statistical Package for the Social Sciences (SPSS) software (version 16, SPSS Inc. Chicago, IL). Logistic regression models were used to estimate the odds ratio (OR) and 95% confidence interval (CI), and to identify independent prognostic variables for early recurrence. The Cox regression models were applied to evaluate the prognostic variables for recurrence-free survival (RFS). Comparisons of quantitative data between two groups were analyzed by a paired $t$ -test. A Fisher’s exact test or likelihood ratio test was used to assess the association between Clnc and clinical pathological characteristics. DeLong method was used to test the statistical significance of the difference between the areas under two ROC curves. All the tests were performed using GraphPad prism7.0. A $P$ value of 0.05 or less is considered statistically significant.

Table 1
The predictive efficiency of Clnc and clinical features for early recurrence in the training and test sets

Variables	AUC (95%CI)	$P$ value	SEN (95%CI)	SPE (95%CI)
Training set ( $N=$ 160)
Clnc	0.753 (0.678–0.817)		0.726 (0.583–0.841)	0.780 (0.690–0.854)
Tumor size	0.614 (0.534–0.690)	0.0074	0.686 (0.541–0.809)	0.541 (0.443–0.637)
Tumor number	0.559 (0.479–0.638)	0.0003	0.275 (0.159–0.417)	0.844 (0.762–0.906)
PVTT	0.542 (0.461–0.621)	$<$ 0.0001	0.157 (0.070–0.286)	0.927 (0.860–0.968)
Differentiation	0.580 (0.499–0.657)	0.0058	0.490 (0.348–0.634)	0.670 (0.573–0.757)
BCLC	0.600 (0.519–0.676)	0.0067	0.392 (0.262–0.539)	0.807 (0718–0.874)
Test set ( $N=$ 161)
Clnc	0.675 (0.597–0.747)		0.720 (0.575–0.838)	0.631 (0.534–0.720)
Tumor size	0.523 (0.443–0.602)	0.0073	0.560 (0.413–0.700)	0.487 (0.390–0.583)
Tumor number	0.622 (0.543–0.698)	0.3228	0.380 (0.247–0.528)	0.865 (0.787–0.922)
PVTT	0.541 (0.461–0.620)	0.0012	0.100 (0.033–0.218)	0.982 (0.936–0.998)
Differentiation	0.597 (0.517–0.673)	0.1241	0.500 (0.355–0.645)	0.694 (0.599–0.778)
BCLC	0.678 (0.600–0.749)	0.9602	0.500 (0.355–0.645)	0.856 (0.776–0.915)

Abbreviations: AUC, area under the receiver operating characteristic curve; BCLC, Barcelona Clinic Liver Cancer stage; CI, confidence interval; N, number of patients; PVTT, portal vein tumor thrombus; SEN, sensitivity; SPE, specificity.

3. Results

3.1 Construction of the classifier

A group of 30 candidate lncRNAs was selected either from lncRNA microarray data or literature, and their expression levels in 20 paired HCC and the adjacent normal liver tissues were analyzed by RT-qPCR. The dysregulation of 20 lncRNAs was confirmed, with 15 upregulated and 5 downregulated lncRNAs (Supplementary Fig. S2). Further analysis showed that 3 lncRNAs were significantly associated with early recurrence ( $\leqslant$ 1 year) of HCC ( $P<$ 0.05, Supplementary Table 5) in the training set, suggesting that abnormally expressed lncRNAs may have predictive value for recurrence of HCC patients.

Based on the hypothesis that incorporating multiple lncRNAs could improve prediction performance, we incorporated all the 20 lncRNAs into a linear SVM model to build lncRNA-based classifiers for early recurrence. After an exhaustive search of all possible combinations and LOO cross-validation, a combination of nine lncRNAs, named as Clnc, was finally selected (Supplementary Fig. S3), which includes AF339810, AK026286, BC020899, HEIH, HULC, MALAT1, PVT1, uc003fpg, and ZFAS1(Supplementary Table 5). The discriminative value for Clnc as measured by the AUC was 0.753 (sensitivity, 0.726; specificity, 0.78) for predicting early recurrence after HCC resection, which was significantly higher than that of other traditional clinical features including tumor size, tumor number, portal vein tumor thrombus (PVTT), differentiation and BCLC stage (Fig. 2A, Left panel and Table 1). For the 160 patients, 61 were discriminated as high-risk for early recurrence and 99 as low-risk by Clnc. The recurrence probability in the high-risk group was remarkably higher than that in the low-risk group (60.7% (37/61) vs 14.1% (14/99) and the cumulative recurrence probability in the high-risk group was also significantly higher than that in the low-risk group (Fig. 2B, Left panel). Consistently, survival among high-risk patients was poorer than that of the low-risk patients (Supplementary Fig. S4; Left). To facilitate the usage of this classifier, we constructed a web-based application that allows the users to determine the risk for early recurrence of each patient according to the relative expression values of lncRNAs that constitute Clnc in HCC tissues (https://lncrnaspredicthcc.shinyapps.io/Shiny_APP/). All the scripts for calculation have already been embedded in this application, and a template for data input was shown in Supplementary Table 6.

Table 2
Logistic regression analysis to evaluate factors that were associated with early recurrence for patients in the training and test sets

Variables	Univariate analysis		Multivariate analysis
	OR (95% CI)	$P^{*}$	OR (95% CI)	$P^{*}$
Training set
Age ( $\leqslant$ 55 vs $>$ 55 years)	0.897 (0.445–1.810)	0.762
Gender (female vs male)	0.604 (0.248–1.472)	0.267
AFP ( $\leqslant$ 25 vs $>$ 25 ng/ml)	1.425 (0.704–2.885)	0.325
Tumor size ( $\leqslant$ 5 vs $>$ 5 cm)	2.581 (1.280–5.205)	0.008
Tumor number (1 vs $>$ 1)	2.048 (0.917–4.574)	0.08
HBV infection (no vs yes)	NA
PVTT (no vs yes)	2.349 (0.828–6.665)	0.109
Differentiation (I/II vs III/IV)	1.950 (0.989–3.844)	0.054
Cirrhosis (no vs yes)	1.180 (0.500–2.785)	0.705
Child-Pugh (A vs B)	2.234 (0.536–9.316)	0.27
BCLC (0/A vs B/C)	2.704 (1.294–5.647)	0.008	2.738 (1.166–6.428)	0.021
Clnc (low vs high risk)	9.360 (4.361–20.092)	$<$ 0.0001	9.408 (4.298–20.592)	$<$ 0.0001
Test set
Age ( $\leqslant$ 55 vs $>$ 55 years)	0.583 (0.273–1.243)	0.162
Gender (female vs male)	1.035 (0.397–2.698)	0.945
AFP ( $\leqslant$ 25 vs $>$ 25 ng/ml)	2.326 (1.098–4.925)	0.027	2.93 (1224–7.017)	0.016
Tumor size ( $\leqslant$ 5 vs $>$ 5 cm)	1.206 (0.616–2.359)	0.585
Tumor number (1 vs $>$ 1)	3.923 (1.782–8.633)	0.001
HBV infection (no vs yes)	0.440 (0.060–3.219)	0.419
PVTT (no vs yes)	6.056 (1.133–32.368)	0.035
Differentiation (I/II vs III/IV)	2.265 (1.141–4.496)	0.019
Cirrhosis (no vs yes)	1.235 (0.453–3.371)	0.68
Child-Pugh (A vs B)	2.972 (0.763–11.582)	0.116
BCLC (0/A vs B/C)	5.937 (2.758–12.781)	$<$ 0.0001	6.175 (2.663–14.314)	$<$ 0.0001
Clnc (low vs high risk)	4.390 (2.120–9.090)	$<$ 0.0001	4.326 (1.946–9.617)	0.0003

Abbreviations: AFP, Alpha-fetoprotein; BCLC, Barcelona Clinic Liver Cancer stage; CI, confidence interval; HBV, hepatitis B virus; N, number of patients; NA, not available; OR, odds ratio; PVTT, portal vein tumor thrombus. ${}^{*}$ Likelihood ratio test; NOTE: Variables with $P<$ 0.05 in univariate analysis were included as covariates in multivariate analysis.

Figure 3.

Comparison of predictive power for early recurrence by the Clnc, BCLC stage, and their combinations. ROC curves in detecting HCC with early recurrence from training cohort (A) and test cohort (B) by the Clnc classifier, BCLC stage, and their combinations. The combination of Clnc and BCLC show the highest sensitivity in both training and test cohorts.

Table 3

The predicted power of Clnc in the subgroups of patients in the combined training and test cohort

Variables	No. of patients	Rec after 1 year ( $N=$ 220)	Rec within 1 year ( $N=$ 101)	Odds ratio (95% CI)	$P^{*}$
Combined cohort	321			6.217 (3.684–10.490)	$<$ 0.0001
Low-risk		155	28
High-risk		65	73
AFP
$\leqslant$ 25 ng/ml	118			8.739 (3.287–23.233)	$<$ 0.0001
Low-risk		67	7
High-risk		23	21
$>$ 25 ng/ml	203			5.188 (2.774–9.703)	$<$ 0.0001
Low-risk		88	21
High-risk		42	52
Tumor size
$\leqslant$ 5 cm	151			6.901 (3.073–15.501)	$<$ 0.0001
Low-risk		86	12
High-risk		27	26
$>$ 5 cm	170			5.334 (2.671–10.653)	$<$ 0.0001
Low-risk		69	16
High-risk		38	47
Tumor number
1	256			5.927 (3.207–10.955)	$<$ 0.0001
Low-risk		131	19
High-risk		57	49
$>$ 1	65			8.000 (2.642–24.221)	$<$ 0.0001
Low-risk		24	9
High-risk		8	24
Differentiation
I/II	201			7.771 (3.764–16.044)	$<$ 0.0001
Low-risk		109	13
High-risk		41	38
III/IV	120			4.472 (2.049–9.762)	$<$ 0.0001
Low-risk		46	15
High-risk		24	35
Cirrhosis
No	53			3.200 (0.921–11.115)	0.106
Low-risk		28	7
High-risk		10	8
Yes	268			7.147 (3.982–12.828)	$<$ 0.0001
Low-risk		127	21
High-risk		55	65
BCLC
0/A	239			5.972 (3.083–11.568)	$<$ 0.0001
Low-risk		129	16
High-risk		54	40
B/C	82			6.500 (2.473–17.081)	$<$ 0.0001
Low-risk		26	12
High-risk		11	33

Abbreviations: AFP, Alpha-fetoprotein; BCLC, Barcelona Clinic Liver Cancer stage; CI, confidence interval; OR, odds ratio; Rec, recurrence. ${}^{*}$ Calculated by Fisher exact test.

3.2 Validation of the Clnc classifier

The predictive efficiency of the Clnc was then evaluated in the test set. Clnc correctly predicted the early recurrence for 36 of 50 cases within the test set, and the odds ratio was 4.39 (95% CI 2.12–9.09), $P<$ 0.0001, Table 2). The AUC for Clnc to predict early recurrence was still significantly higher than that for tumor size and PVTT (Fig. 2A, Right panel and Table 1). The probability of early recurrence for the high-risk group was remarkably higher than that for the low-risk group (46.8% (36/77) vs 16.7% (14/84)). Moreover, the sensitivity of Clnc was significantly higher than that for BCLC (0.72 vs 0.50), albeit comparable AUCs were noted (Table 1). In contrast, the individual lnRNA that constitute the Clnc showed limited discriminative power as revealed by the AUC (Supplementary Fig. S5). Survival analysis showed that the Clnc defined high-risk patients consistently showed much higher 5-year cumulative recurrence rate (72.9% vs 55.6%; average RFS: 21.5 vs 37.2 months; HR 2.093, 95% CI 1.479–3.372, $P<$ 0.0001, Fig. 2B, Right panel) and poorer overall survival (Supplementary Fig. S4), which suggested that Clnc could effectively separate high-risk patients from low-risk ones.

BCLC staging system has been widely used as a prognostic factor, we then evaluated whether combining the Clnc and BCLC stage could increase the prediction efficiency. As shown, the combination of Clnc and BCLC (Clnc $+$ BCLC) for predicting early recurrence after resection performed better than Clnc or BCLC stage alone (AUC 0.795; 95% CI 0.724–0.855; Fig. 3A) in the training set. The similar results were obtained in the test set (AUC 0.751; 95% CI 0.677–0.816; Clnc $+$ BCLC vs Clnc: $P=$ 0.035; Clnc $+$ BCLC vs BCLC: $P=$ 0.002; Fig. 3B). The above results showed that the integrated Clnc and BCLC stage could significantly increase the predictive power of Clnc.

3.3 Clnc is an independent prognostic factor for early recurrence and RFS in HCC patients

To explore the independence of Clnc from other clinical or pathological factors in predicting early recurrence, multivariable logistic regression analysis was performed using an entering variable selection method. Factors that were significantly associated with early recurrence in the univariate logistic regression analysis were included in the regression equation to adjust the coefficient. Clnc remained a powerful and independent factor to predict early recurrence in the training set (OR 9.408, 95% CI 4.298–20.592, $P<$ 0.0001), which was consistent with that in the test set (OR 4.326, 95% CI 1.946–9.617, $P=$ 0.0003; Table 2). To detect whether Clnc had a similar prognostic value for RFS, multivariable Cox regression analysis was conducted. The results showed that Clnc was an independent prognostic factor for RFS in both training (HR 2.443, 95% CI 1.601–3.728, $P<$ 0.0001; Supplementary Table 7) and test cohorts (HR 1.928, 95% CI 1.279–2.906, $P=$ 0.002; Supplementary Table 7).

We then evaluated the prognostic value of Clnc in subgroups of different clinical stages. As shown, Clnc could identify patients with a high risk of early recurrence from low-risk patients in the subgroups divided according to AFP, tumor size, tumor number, differentiation, cirrhosis, and BCLC stage in the combined group including all the training and test cohorts (Table 3). These results suggested that Clnc had the prognostic potential for a subgroup of HCC.

4. Discussion

Early postsurgical recurrence remains a major cause of cancer-related mortality. Identifying patients at high risk of early relapse after resection could guide postoperative adjuvant therapies that confer the best survival and spare the low-risk patients from over chemotherapy. In this study, we developed a nine-lncRNA classifier, Clnc, and showed that Clnc could effectively discriminate patients at high risk of early recurrence from low-risk patients. Moreover, the Clnc classifier is an independent prognostic factor for the RFS of HCC patients.

Currently, BCLC staging has been widely accepted as a powerful prognostic predictor and acted as the main reference to direct the treatment strategies in HCC. However, as a result of the heterogeneity of cancer at the molecular and genetic levels [19, 20], even patients with the same stage can have very diverse survival outcomes [21]. Recent studies have suggested a few lncRNAs with prognostic values in HCC, yet most of them only tested an individual lncRNA in a small size of patients, and the accuracy and robustness of these lncRNA markers need to be further validated. Here, we showed that Clnc was a significant predictive classifier for early recurrence and RFS in the validation cohort in multivariate analysis (Table 2 and Supplementary Table 7). Moreover, Clnc can identify high-risk patients in different subgroups divided according to AFP, tumor size, tumor number, differentiation, and even those who are in the early BCLC stage (Table 3), which suggests that Clnc may be used to refine the current prognostic factors. Furthermore, combining Clnc and BCLC together could significantly increase the accuracy of prognostic prediction.

Early recurrence represents a true metastasis and is associated with the dissemination of primary HCC tumor cells. Of the nine lncRNAs that make up the Clnc, five of them, including ZFAS1, HULC, HEIH, PVT1, and MALAT1, are well-known lncRNAs that were dysregulated in HCCs and have been implicated in the development and progression of HCCs. For instance, ZFAS1 functions as an oncogene in HCC progression by abrogating the tumor-suppressor ability of miR-150, and high levels of ZFAS1 in HCCs are correlated with poor recurrence-free survival [22]. Loss of function studies showed that knockdown of HULC, HEIH, or PVT1 suppresses migration and invasion of HCC cells [23, 24, 25], characteristics that contribute to the early dissemination of tumor cell, suggesting that overexpression of these lncRNAs may play a role in tumor metastasis. Consistently, retrospective cohort studies have revealed that patients with high expression levels of HULC, HEIH or PVT1 exhibited poor recurrence-free survival [23, 26, 27]. MALAT1 is upregulated in HCC and acts as a proto-oncogene by Enhancing mTOR-Mediated Translation of TCF7L2 [28]. Moreover, MALAT1 overexpression inhibits, while MALAT1 deficiency promotes migration and invasion of HCC cells, and higher MALAT1 expression in HCC patients was associated with worse RFS [29]. Taken together, our findings suggested that integrating multiple lncRNAs that were dysregulated in HCCs might develop an excellent prognostic classifier for HCCs.

Notably, the lncRNA genes we selected in this study concluded only those differentially expressed between HCC and the non-tumor liver tissues, considering that the aberrantly expressed lncRNAs in cancer could mark the spectrum of disease progression and these lncRNAs may have the potential to serve as biomarkers for early relapse. Given the heterogeneity of tumors, lncRNA genes showing similar expression levels between cancer and normal tissues may possibly predict the recurrence of HCC patients with high accuracy. Thus, a more sophisticated screening strategy to discover the prognosis related lncRNAs, ideally those differentially expressed in subgroups of HCC patients with different prognosis, may guarantee a more robust classifier. Besides, this study concludes only patients from one center, and the AUC for the classifier is moderate. Future efforts will focus on optimizing the algorithm to boost the AUC and multi-center external validation to assure the general applicability of our model.

In conclusion, we developed a lncRNA-based classifier that could identify the patients at risk for postsurgical recurrence of HCC, which might provide patients with a chance of optimal postoperative adjuvant therapies.

Authors’ contributions

Conception: Shuai He and Jin-E Yang.

Interpretation or analysis of data: Shuai He, Jin-Feng Li, Hao Tian, Ye Sang, Xiao-Jing Yang, and Gui-Xin Guo.

Preparation of the manuscript: Shuai He and Jin-E Yang.

Supervision: Jin-E Yang.

Supplementary data

The supplementary files are available to download from http://dx.doi.org/10.3233/CBM-210193.

sj-docx-1-cbm-10.3233_CBM-210193.docx - Supplemental material

Supplemental material, sj-docx-1-cbm-10.3233_CBM-210193.docx

Footnotes

Acknowledgments

This work was supported by the National Natural Science Foundation of China (81572400 and 81872259).

References

Bray

Ferlay

Soerjomataram

Siegel

R.L.

Torre

L.A.

and Jemal

, Global cancer statistics 2018: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries, CA Cancer J Clin 68 (2018), 394–424.

Villanueva

, Hepatocellular carcinoma, N Engl J Med 380 (2019), 1450–1462.

Tabrizian

Jibara

Shrager

Schwartz

and Roayaie

, Recurrence of hepatocellular cancer after resection: Patterns, treatments, and prognosis, Ann Surg 261 (2015), 947–955.

European Association for the Study of the Liver, EASL Clinical Practice Guidelines: Management of hepatocellular carcinoma, J Hepatol 69 (2018), 182–236.

Imamura

Matsuyama

Tanaka

Ohkubo

Hasegawa

Miyagawa

Sugawara

Minagawa

Takayama

Kawasaki

and Makuuchi

, Risk factors contributing to early and late phase intrahepatic recurrence of hepatocellular carcinoma after hepatectomy, J Hepatol 38 (2003), 200–207.

Furukawa

Shiba

Horiuchi

Shirai

Haruki

Fujiwara

Sakamoto

Gocho

and Yanaga

, Survival benefit of hepatic resection for hepatocellular carcinoma beyond the barcelona clinic liver cancer classification, J Hepatobiliary Pancreat Sci 24 (2017), 199–205.

Krishnan

M.S.

Rajan Kd

Park

Arjunan

Garcia Marques

F.J.

Bermudez

Girvan

O.A.

Hoang

N.S.

Yin

Nguyen

M.H.

Kothary

Pitteri

Felsher

D.W.

and Dhanasekaran

, Genomic analysis of vascular invasion in HCC reveals molecular drivers and predictive biomarkers, Hepatology 73 (2021), 2342–2360.

Nault

J.C.

and Villanueva

, Biomarkers for hepatobiliary cancers, Hepatology 73(Suppl 1) (2021), 115–127.

Chan

A.W.H.

Zhong

Berhane

Toyoda

Cucchetti

Shi

Tada

Chong

C.C.N.

Xiang

B.D.

L.Q.

Lai

P.B.S.

Mazzaferro

Garcia-Finana

Kudo

Kumada

Roayaie

and Johnson

P.J.

, Development of pre and post-operative models to predict early recurrence of hepatocellular carcinoma after surgical resection, J Hepatol 69 (2018), 1284–1293.

10.

Iizuka

Oka

Yamada-Okabe

Nishida

Maeda

Mori

Takao

Tamesa

Tangoku

Tabuchi

Hamada

Nakayama

Ishitsuka

Miyamoto

Hirabayashi

Uchimura

and Hamamoto

, Oligonucleotide microarray for prediction of early intrahepatic recurrence of hepatocellular carcinoma after curative resection, Lancet 361 (2003), 923–929.

11.

Statello

Guo

C.J.

Chen

L.L.

and Huarte

, Gene regulation by long non-coding RNAs and its biological functions, Nat Rev Mol Cell Biol 22 (2021), 96–118.

12.

Jiang

Luo

Wang

Wen

Liao

Yang

and Wang

, A novel three-lncRNA signature predicts the overall survival of HNSCC patients, Ann Surg Oncol 28 (2021), 3396–3406.

13.

Bhan

Soleimani

and Mandal

S.S.

, Long noncoding RNA and cancer: A new paradigm, Cancer Res 77 (2017), 3965–3981.

14.

Jiagen

Zhaoli

Liqing

Chengcheng

Max Yifan

Yibo

Suya

Fang

Susheng

and Xiaoli

, LncRNA profile study reveals a three-lncRNA signature associated with the survival of patients with oesophageal squamous cell carcinoma, Gut 63 (2014), 1700–1710.

15.

Cai

Yang

Chen

Huang

and Miao

, Signature panel of 11 methylated mRNAs and 3 methylated lncRNAs for prediction of recurrence-free survival in prostate cancer patients, Pharmgenomics Pers Med 14 (2021), 797–811.

16.

Yuan

S.X.

Wang

Yang

Tao

Q.F.

Zhang

Wang

L.L.

Yang

Liu

Wang

Z.G.

Q.G.

Fan

Liu

Sun

S.H.

and Zhou

W.P.

, Long noncoding RNA DANCR increases stemness features of hepatocellular carcinoma by derepression of CTNNB1, Hepatology 63 (2016), 499–511.

17.

Ernst

Kheradpour

Mikkelsen

T.S.

Shoresh

Ward

L.D.

Epstein

C.B.

Zhang

Wang

Issner

Coyne

Durham

Kellis

and Bernstein

B.E.

, Mapping and analysis of chromatin state dynamics in nine human cell types, Nature 473 (2011), 43–49.

18.

Hanley

J.A.

and McNeil

B.J.

, The meaning and use of the area under a receiver operating characteristic (ROC) curve, Radiology 143 (1982), 29–36.

19.

Wang

Khatib

S.A.

Chang

C.W.

Heinrich

Dominguez

D.A.

Forgues

Candia

Hernandez

M.O.

Kelly

Zhao

Tran

Hernandez

J.M.

Davis

J.L.

Kleiner

D.E.

Wood

B.J.

Greten

T.F.

and Wang

X.W.

, Single-cell atlas of tumor cell evolution in response to therapy in hepatocellular carcinoma and intrahepatic cholangiocarcinoma, J Hepatol 10.1016/j.jhep.2021.06.028 (2021). doi: 10.1016/j.jhep.2021.1006.1028.

20.

Heinrich

Craig

A.J.

Heinrich

Greten

T.F.

and Wang

X.W.

, Understanding tumour cell heterogeneity and its implication for immunotherapy in liver cancer using single-cell analysis, J Hepatol 74 (2021), 700–715.

21.

Kinoshita

Onoda

Fushiya

Koike

Nishino

and Tajiri

, Staging systems for hepatocellular carcinoma: Current status and future perspectives, World J Hepatol 7 (2015), 406–424.

22.

Xie

Shen

Cheng

Shi

Deng

Chen

Shen

Peng

Zhan

and Zhu

, Amplification of long noncoding RNA ZFAS1 promotes metastasis in hepatocellular carcinoma, Cancer Res 75 (2015), 3181–3191.

23.

S.P.

H.X.

J.D.

Wang

Y.J.

Wang

C.Y.

Zhang

H.M.

Zhang

R.X.

Zhang

J.J.

Yao

and Shen

Z.Y.

, LncRNA HULC enhances epithelial-mesenchymal transition to promote tumorigenesis and metastasis of hepatocellular carcinoma via the miR-200a-3p/ZEB1 signaling pathway, Oncotarget 7 (2016), 42431–42446.

24.

Luo

Chen

Wang

Lai

and Ye

, Long non-coding RNA PVT1/miR-150/HIG2 axis regulates the proliferation, invasion and the balance of iron metabolism of hepatocellular carcinoma, Cell Physiol Biochem 49 (2018), 1403–1419.

25.

Cao

Y.M.

Liu

and Liu

, Silence of lncRNA HEIH suppressed liver cancer cell growth and metastasis through miR-199a-3p/mTOR axis, J Cell Biochem 120 (2019), 17757–17766.

26.

Ding

Yang

Xiao

Peng

Cheng

Xie

Zhou

and Zheng

, Long non-coding RNA PVT1 is associated with tumor progression and predicts recurrence in hepatocellular carcinoma patients, Oncol Lett 9 (2015), 955–963.

27.

Yang

Zhang

Huo

X.S.

Yuan

J.H.

Yuan

S.X.

Zhu

Zhou

W.P.

Yang

G.S.

Wang

Y.Z.

Shang

J.L.

Gao

C.F.

Zhang

F.R.

Wang

and Sun

S.H.

, Long noncoding RNA high expression in hepatocellular carcinoma facilitates tumor growth through enhancer of zeste homolog 2 in humans, Hepatology 54 (2011), 1679–1689.

28.

Malakar

Stein

Saragovi

Winkler

Stern-Ginossar

Berger

Pikarsky

and Karni

, Long noncoding RNA MALAT1 regulates cancer glucose metabolism by enhancing mTOR-mediated translation of TCF7L2, Cancer Res 79 (2019), 2480–2493.

29.

Xie

S.J.

Diao

L.T.

Cai

Zhang

L.T.

Xiang

Jia

C.C.

Qiu

D.B.

Liu

Sun

Y.J.

Lei

Hou

Y.R.

Tao

Y.X.

Xiao

Z.D.

and Zhang

, mascRNA and its parent lncRNA MALAT1 promote proliferation and metastasis of hepatocellular carcinoma cells by activating ERK/MAPK signaling pathway, Cell Death Discov 7 (2021), 110.

A long non-coding RNA-based signature predicts early recurrence in hepatocellular carcinoma

Abstract

BACKGROUND:

OBJECTIVE:

METHODS:

RESULTS:

CONCLUSIONS:

Keywords

1. Introduction

2. Materials and methods

2.1 Patients and tissue samples

2.2 Selection of candidate lncRNAs

2.3 RNA extraction, reverse transcription (RT) and quantitative PCR (RT-qPCR)

2.4 Data process and supervised classifier construction

Table 1 The predictive efficiency of Clnc and clinical features for early recurrence in the training and test sets

3.1 Construction of the classifier

Table 2 Logistic regression analysis to evaluate factors that were associated with early recurrence for patients in the training and test sets

3.3 Clnc is an independent prognostic factor for early recurrence and RFS in HCC patients

4. Discussion

Authors’ contributions

Supplementary data

sj-docx-1-cbm-10.3233_CBM-210193.docx - Supplemental material

Footnotes

Acknowledgments

References

Table 1
The predictive efficiency of Clnc and clinical features for early recurrence in the training and test sets

Table 2
Logistic regression analysis to evaluate factors that were associated with early recurrence for patients in the training and test sets