This article is mentioned in:

Introduction

Hepatocellular carcinoma (HCC) is the sixth most common malignancy and the fourth leading cause of cancer-related death globally, and worldwide incidence increases by 3–4% per year (1,2). Hepatitis B virus (HBV) or hepatitis C virus (HCV) infection, alcohol consumption, diabetes, nonalcoholic fatty liver disease and smoking are known to be major risk factors for HCC (3,4). HCC is highly heterogeneous, and the pathogenesis is extremely complex. The progression of HCC involves multiple processes such as mutation and signaling pathway maladjustment, reflecting the interaction among multiple genes (5,6). Despite the development of various drugs and breakthroughs in diagnosis, the prognosis of HCC remains poor, with a 5-year survival rate of only 5% for patients with advanced HCC (7). Timely and effective assessment of prognosis is of great significance to guide the treatment. At present, there are no biomarkers that effectively predict the survival of patients with HCC, and thus, finding effective prognostic biomarkers for patients with HCC is crucial.

Long non-coding RNAs (lncRNAs), non-coding transcripts >200 nucleotides, serve important cellular functions such as in chromatin modification as well as transcriptional and post-transcriptional regulation (8,9). Increasing evidence demonstrates that aberrant expression of lncRNAs is associated with the occurrence and development of various human diseases, especially cancer (10–12). For example, the overexpression of lncRNA HOX Transcript Antisense Intergenic RNA (HOTAIR) was demonstrated to predict tumor recurrence after liver transplantation (13). There was also a significant association between HOTAIR expression and tumor progression in patients with HCC (14–16). Increased biallelic expression of H19 and IGF2 may participate in an epigenetic mechanism of HCC development and progression (17). The lncRNA GPC3-AS1 promotes HCC progression by epigenetic GPC3 activation (18). However, the role of lncRNAs in alcohol-related HCC remains unclear.

Alcohol is a dose-related risk factor known to be associated with more than 200 diseases, including HCC (19,20). Heavy drinkers are at 3- to 10-fold higher risk of hepatocellular carcinoma than non-drinkers (2). In addition, the overall survival rate is lower for patients with alcohol-related HCC than for those with non-alcohol-related HCC, suggesting that there may be a link between alcohol and prognosis (21). Thus, the present study aimed to examine whether lncRNAs are differentially expressed in the presence of alcohol consumption that may be used as prognostic markers in HCC, and whether these differences might influence the risk of HCC recurrence or death. Using data from The Cancer Genome Atlas (TCGA), the present study developed a risk-scoring system based on lncRNA levels that may be valuable for predicting the prognosis of patients with alcohol-related HCC.

Materials and methods

Patient selection and data collection

Profiles of lncRNA and mRNA expression in HCC patients were downloaded from the University of California Santa Cruz (UCSC) Xena server (https://xenabrowser.net/datapages/). Corresponding clinical information was obtained from TCGA (version 09–14–2017 for HCC). The patients in this dataset had histologically confirmed HCC. Patient data included a complete lncRNA expression profile, alcohol consumption status and survival data for determining OS and RFS. A total of 113 patients with HCC and alcohol consumption, and 224 without alcohol consumption were selected (Table I). This study complied with TCGA publication guidelines and policies (http://cancergenome.nih.gov/publications/publicationguidelines). No ethics approval was required for this study since data were obtained from TCGA.

Table I. Clinicopathological characteristics of 337 patients with alcohol- or non-alcohol-related hepatocellular carcinoma.

Table I.

Clinicopathological characteristics of 337 patients with alcohol- or non-alcohol-related hepatocellular carcinoma.

	Patients (n=337)
Clinicopathological characteristics	n	%
Age, years
≤60	162	48.07
>60	175	51.93
BMI
<25	161	47.77
≥25	150	44.51
Not reported	26	7.72
Race
Non-Asian	186	55.19
Asian	141	41.84
Not reported	10	2.97
Sex
Female	107	31.75
Male	230	68.25
Hepatitisa
No	189	56.08
Yes	148	43.92
Cirrhosis
Non-cirrhosis	124	36.80
Cirrhosis	74	21.96
Not reported	139	41.25
Alcohol consumption
No	224	66.47
Yes	113	33.53
Histologic grade
G1-2	209	62.02
G3-4	123	36.50
Not reported	5	1.418
New tumor event
No	167	49.55
Yes	154	45.70
Not reported	16	4.75
Pathologic stage
Stage I+II	231	68.55
Stage III+IV	82	24.33
Not reported	24	7.12
Cancer status
Tumor free	181	53.71
With tumor	142	42.14
Not reported	14	4.15
Family cancer history
No	188	55.79
Yes	108	32.05
Not reported	41	12.17
Residual tumor
R0	301	89.32
Non-R0	29	8.61
Not reported	7	2.08
Vascular invasion
Negative	188	55.79
Positive	95	28.19
Not reported	54	16.02

a Hepatitis B or C. BMI, Body mass index; AFP, alpha fetoprotein.

Identification of lncRNAs differentially expressed between alcohol-related or non-alcohol-related HCC

After eliminating lncRNAs showing zero expression in >50% of all patients, the edgeR package in R (https://www.r-project.org/) was used to identify lncRNAs differentially expressed between patients with or without alcohol consumption (22). Differential expression was defined as log2fold change (log2FC) >1 and false discovery rate (FDR) <0.05. Differentially expressed lncRNAs were then presented in cluster heat maps and volcano maps generated using the packages gplots and heatmap in R.

Construction of lncRNA-based risk scoring systems

Standardized expression of lncRNAs in multiple tissues of the same patient were averaged. Univariate Cox analysis was then performed to screen differentially expressed lncRNAs to determine their significant relationship with OS or RFS, with the threshold set at P=0.05. Selected lncRNAs were included in subsequent multivariate Cox regression by the backward stepwise method in order to identify the best model. The expression level of each lncRNA was multiplied by the corresponding regression coefficient β and linearly combined to generate a risk scoring system:

Risk score=(β1 × expression level of lncRNA1) + (β2 × expression level of lncRNA2) + (β3 × expression level of lncRNA3) + (βn × expression level of lncRNAn).

This formula was used to calculate a risk score for each patient. The prognosis prediction performance of this risk score was assessed using time-dependent receiver operating characteristic (ROC) curves within three years (23). Patients with HCC were divided into a high- or a low-risk group according to the cut-off value of the median risk score, as demonstrated in non-cluster heat maps. Kaplan-Meier survival curves were generated and compared between high- and low-risk groups. All these analyses were conducted using R/Bioconductor (version 3.4.4, http://www.r-project.org/).

Prognostic performance of the risk scoring systems

To validate the prognostic performance of the risk scoring systems, univariate and multivariate Cox regression analyses were performed to determine whether the risk score was an independent factor for survival. This regression was performed in SPSS 16.0 (SPSS, Inc.), and a significance threshold of P=0.05 (two-sided).

Co-expression and functional enrichment analysis of related mRNAs

Pearson correlation was performed to screen for relationships between lncRNAs in the risk scoring systems and mRNAs based on data of 337 patients with HCC. Relationships were considered significant if the mRNA expression co-varied with that of lncRNAs with a two-sided absolute value of the Pearson correlation coefficient >0.30 and a z-test P<0.01. To obtain a deeper understanding of these mRNAs, enrichment analyses were performed using the Genomes pathway in the Kyoto Encyclopedia of Genes and Genomes by the package clusterProfiler in R (24). P<0.05 was considered to indicate a statistically significant difference.

Results

lncRNAs are differentially expressed in alcohol-related HCC or non-alcohol-related HCC

A total of 102 differentially expressed lncRNAs were identified, 47 (46.08%) of which were upregulated and 55 (53.92%) downregulated (Figs. 1 and 2); the first 20 up- and downregulated lncRNAs, together with the corresponding values for log2FC, P and FDR are demonstrated in Table II.

Figure 1. Volcano map of the differentially expressed lncRNAs between alcohol-related and non-alcohol-related HCC. Red spots represent upregulated genes, and green spots represent downregulated genes. lncRNA, long non-coding RNA; HCC, hepatocellular carcinoma; FC, fold change; FDR, false discovery rate.

Figure 2. Heat map based on the differentially expressed lncRNA between alcohol-related and non-alcohol-related HCC. lncRNA, long non-coding RNA; HCC, hepatocellular carcinoma.

Table II. Differentially expressed lncRNAs in patients with alcohol-related or non-alcohol-related hepatocellular carcinoma.

Table II.

Differentially expressed lncRNAs in patients with alcohol-related or non-alcohol-related hepatocellular carcinoma.

Top 20 upregulated lncRNAs				Top 20 downregulated lncRNAs
lncRNA	logFC	P-value	FDR	lncRNA	logFC	P-value	FDR
AC090921.1	5.26	1.35×10−50	7.79×10−47	ERVH48-1	−3.77	1.25×10−21	1.45×10−18
AL451074.6	3.15	2.17×10−35	6.29×10−32	AP000439.3	−4.73	3.49×10−17	2.89×10−14
AC007938.3	1.68	3.78×10−22	7.29×10−19	AC139749.1	−3.34	5.15×10−17	3.72×10−14
AC105446.1	2.58	6.40×10−22	9.26×10−19	LINC00473	−4.00	1.51×10−16	9.73×10−14
AC012640.1	1.94	2.45×10−18	2.36×10−15	LINC01480	−2.26	4.15×10−12	2.00×10−9
AC025627.1	1.26	6.31×10−13	3.65×10−10	AP000757.1	−1.61	2.31×10−10	7.86×10−8
AL445483.1	1.73	1.36×10−12	7.14×10−10	UCA1	−2.54	7.54×10−9	1.90×10−6
MYOSLID	1.59	1.35×10−11	6.03×10−9	AC064807.2	−1.85	9.35×10−9	2.16×10−6
EMX2OS	1.63	5.63×10−11	2.33×10−8	AL359853.1	−1.98	1.29×10−8	2.82×10−6
AC156455.1	1.18	1.15×10−10	4.17×10−8	PLCE1-AS1	−2.03	2.86×10−8	5.71×10−6
AP003354.2	1.27	6.76×10−10	2.17×10−7	AC002398.2	−2.03	9.19×10−8	1.77×10−5
AC007220.1	1.14	1.70×10−9	5.18×10−7	AC079305.1	−1.12	2.12×10−7	3.71×10−5
U62317.1	1.63	1.93×10−9	5.58×10−7	AC007099.1	−2.10	4.35×10−7	6.80×10−5
LINC02043	1.22	2.25×10−9	6.19×10−7	LINC01229	−1.32	7.07×10−7	1.02×10−4
AC068473.3	1.41	3.83×10−9	1.01×10−6	AC098869.2	−1.26	7.52×10−7	1.06×10−4
AC069431.1	1.42	8.70×10−9	2×10−6	LINC00624	−1.28	1.16×10−6	1.51×10−4
LINC01615	1.40	1.32×10−8	2.82×10−6	AC005674.1	−1.06	1.83×10−6	2.11×10−4
AC079779.2	1.50	2.35×10−8	4.86×10−6	AL161772.1	−1.24	2.26×10−6	2.52×10−4
AC062004.1	1.40	1.11×10−7	2.07×10−5	BX640514.2	−1.79	2.93×10−6	3.08×10−4
AL391056.1	1.574	1.21×10−7	2.18×10−5	AC023154.1	−1.44	3.44×10−6	3.46×10−4

[i] lncRNAs, long noncoding RNAs; logFC, log2fold change; FDR, false discovery rate.

Risk-scoring system based on lncRNA expression and OS

Univariate Cox analysis identified six lncRNAs that were significantly associated with OS: AC012640.1, AC013451.2, AC062004.1, LINC02334, AC090921.1 and LINC01605. The first four were independent prognostic indicators of OS based on multivariate Cox regression (Table III). The resulting risk scoring system was: Risk score=(0.186 × AC012640.1) + (0.363 × AC013451.2) + (−0.243 × AC062004.1) + (−0.275 × LINC02334).

Table III. Four lncRNAs were correlated with overall survival in the best statistical model.

Table III.

Four lncRNAs were correlated with overall survival in the best statistical model.

lncRNA	β	HR	z	P-value
AC012640.1	0.186	1.205	1.71	0.088
AC062004.1	0.243	0.784	−1.98	0.047a
LINC02334	−0.275	0.759	−2.23	0.026a
AC013451.2	0.363	1.437	2.42	0.016a

a P<0.05. lncRNAs, long noncoding RNAs; HR, Hazard ratio.

In this scoring system, increased expression of AC012640.1 and AC013451.2 predicted worse OS (β>0), whereas increased expression of AC062004.1 and LINC02334 predicted better OS (β<0).

Based on their risk scores, patients were classified as at low or high risk of poor OS using the median risk score as cutoff (Fig. 3A). Kaplan-Meier curves demonstrated that patients with high risk had significantly lower OS at 3 and 5 years compared with that of patients with low risk (Fig. 4A). The area under the ROC curve (AUC) for the risk scoring system was 0.721 (Fig. 5A).

Figure 3. Non-cluster risk heat map of long non-coding RNA-based risk scoring system for overall survival (A) or recurrence-free survival (B). The risk value gradually increases from left to right.

Figure 4. Kaplan-Meier survival curves of overall survival (A) or recurrence-free survival (B) according to the risk cutoff point. Horizontal axis represents the survival period and the vertical axis represents the frequency.

Figure 5. ROC curves analysis of the risk scoring system for overall survival (A) or recurrence-free survival (B). ROC, receiver operating characteristic; AUC, area under the curve.

The risk scoring system was used to predict OS of patients with different clinicodemographic characteristics. This is an important test of the scoring system because of the heterogeneity of HCC and the large number of factors that influence prognosis.

Univariate analysis identified risk score, family cancer history and vascular invasion as significantly associated with OS, but not age, body mass index (BMI), ethnicity, sex, hepatitis, cirrhosis, histological grade of cancer, new tumor event, pathology stage, cancer status or residual tumor. Multivariate Cox regression identified the following as independent predictors of poor OS: Risk score [hazard ratio (HR) 3.393, 95% confidence interval (CI) 1.597–7.210] and vascular invasion (HR 2.146, 95% CI 0.903–5.104, Table V).

Table V. Univariate and multivariate Cox regression analysis for overall survival.

Table V.

Univariate and multivariate Cox regression analysis for overall survival.

	Univariate Cox regression			Multivariate Cox regression
Variables	P-value	HR	95% CI	P-value	HR	95% CI
Risk score (high/low)	0.003a	4.949	1.735–14.117	0.001a	3.393	1.597–7.210
Age (>60/≤60)	0.202	1.816	0.726–4.543
BMI	0.060
<25	Reference
≥25		0.245	0.073–0.818
Not reported		0.779	0.096–6.340
Race	0.423
Non-Asian	Reference
Asian		0.352	0.053–2.322
Not reported		1.915	0.115–31.769
Sex (Male/Female)	0.143	2.590	0.725–9.251
Hepatitis B or C	0.542
No	Reference
Yes		1.489	0.415–5.346
Cirrhosis	0.216
Non-cirrhosis	Reference
Cirrhosis		3.183	0.524–19.324
Not reported		0.572	0.178–1.843
Histologic grade	0.586
G1-2	Reference
G3-4		1.326	0.481–3.649
New tumor event	0.406
No	Reference
Yes		1.126	0.216–5.883
Not reported		16.771	0.260–1080.40
Pathologic stage	0.093
Stage I+II	Reference
Stage III+IV		2.857	0.935–8.730
Not reported		3.173	0.740–13.617
Cancer status	0.998
Tumor free	Reference
With tumor		1.052	0.196–5.635
Not reported		0.000	0.000
Family cancer history	0.031			0.126
No	Reference	Reference
Yes		0.667	0.234–1.899		0.868	0.419–1.799
Not reported		0.076	0.011–0.528		0.206	0.045–0.953
Residual tumor	0.611
R0	Reference
Non-R0		0.469	0.105–2.092
Not reported		0.000	0.000
Vascular invasion		0.006a		0.005a
Negative	Reference	Reference
Positive		3.019	0.798–11.416		2.146	0.903–5.104
Not reported		8.383	2.264–31.038		3.577	1.652–7.748

a P<0.05. BMI, Body mass index; AFP, α fetoprotein; HR, Hazard ratio; CI, Confidence interval.

Risk-scoring system based on lncRNA expression and RFS

Univariate analysis identified 11 lncRNAs that were significantly correlated with RFS: ERVH48-1, LINC02043, LINC01605, AC062004.1, AL139385, AC007938.3, AC090921.1, AC025580.1, AC012640.1, C10orf91, and LINC01589. Multivariate analysis demonstrated the first five to be independent prognostic indicators of RFS (Table IV). The resulting risk scoring system was: Risk score=(0.3529 × ERVH48-1) + (0.3499 × LINC02043) + (0.1701 × LINC01605) + (−0.3531 × AC062004.1) + (−0.1924 × AL139385).

Table IV. Five lncRNAs were correlated with recurrence-free survival in the best statistical model.

Table IV.

Five lncRNAs were correlated with recurrence-free survival in the best statistical model.

lncRNA	β	HR	z	P-value
ERVH48-1	0.3529	1.4232	2.59	0.0096a
LINC02043	0.3499	1.4189	3.13	0.0017a
AC062004.1	−0.3531	0.7025	−3.12	0.0018a
LINC01605	0.1701	1.1855	2.19	0.0285a
AL139385	−0.1924	0.8249	−2.23	0.0259a

a P<0.05. lncRNAs, long noncoding RNAs; HR, Hazard ratio

In this scoring system, increased expression of ERVH48-1, LINC02043, and LINC01605 predicted worse RFS (β>0), whereas increased expression of AC062004.1 and AL139385 predicted better RFS (β<0).

Patients were classified as at low or high risk of poor RFS (Fig. 3B). Kaplan-Meier curves demonstrated that patients with high risk had significantly lower RFS at 3 and 5 years compared with that of patients with low risk (Fig. 4B). AUC for the risk scoring system was 0.777 (Fig. 5B).

Univariate analysis identified that risk score and vascular invasion were significantly correlated with RFS, but not age, BMI, ethnicity, sex, hepatitis, cirrhosis, histology grade, new tumor event, pathology stage, cancer status, family cancer history or residual tumor. Multivariate analysis identified the independent predictors to be risk score (HR 2.895, 95% CI 1.491–5.621) and vascular invasion (HR 2.398, 95% CI 1.104–5.210, Table VI).

Table VI. Univariate and multivariate Cox regression analysis for recurrence-free survival.

Table VI.

Univariate and multivariate Cox regression analysis for recurrence-free survival.

	Univariate Cox regression			Multivariate Cox regression
Variables	P-value	HR	95% CI	P-value	HR	95% CI
Risk score (high/low)	0.009a	4.883	1.485–16.051	0.002a	2.895	1.491–5.621
Age (>60/≤60)	0.064	2.343	0.952–5.767
BMI	0.255
<25	Reference
≥25		0.533	0.092–3.074
Not reported		0.087	0.004–1.772
Race	0.403
Non-Asian	Reference
Asian		4.064	0.514–32.105
Not reported		1.065	0.034–33.341
Sex (Male/Female)	0.680	1.523	0.207–11.220
Hepatitis B or C	0.210
No	Reference
Yes		3.102	0.528–18.223
Cirrhosis	0.340
Non-cirrhosis	Reference
Cirrhosis		2.228	0.297–16.737
Not reported		0.291	0.037–2.284
Histologic grade	0.489
G1-2	Reference
G3-4		0.705	0.262–1.899
New tumor event	0.860
No	Reference
Yes		316078	0.000–3.495×1066
Pathologic stage	0.464
Stage I+II	Reference
Stage III+IV		1.896	0.638–5.632
Not reported		0.997	0.127–7.803
Cancer status	0.286
Tumor free	Reference
With tumor		2.632	0.323–21.431
Not reported		11.747	0.399–345.797
Family cancer history	0.242
No	Reference
Yes		0.726	0.147–3.588
Not reported		0.129	0.010–1.703
Residual tumor	0.749
R0		Reference
Non-R0		0.930	0.090–9.614
Not reported		0.321	0.015–6.981
Vascular invasion	0.039a			0.001a
Negative	Reference	Reference
Positive		10.023	1.408–71.326		2.398	1.104–5.210
Not reported		3.990	0.369–43.169		4.732	2.235–10.019

a P<0.05. BMI, Body mass index; AFP, α fetoprotein; HR, Hazard ratio; CI, Confidence interval.

Functional analysis of co-expressed lncRNA and mRNAs

KEGG pathway analysis revealed that co-expressed lncRNA and mRNAs that correlated with OS were involved mainly in chemical carcinogenesis, cytochrome P450-mediated drug metabolism and retinol metabolism (Fig. 6A). Co-expressed lncRNAs and mRNAs that correlated with RFS were involved mainly in cell cycle and carbon metabolism (Fig. 6B).

Figure 6. KEGG pathway analyses. Top 10 pathways of mRNAs that co-expressed with lncRNAs in the risk scoring system for (A) overall survival or (B) recurrence-free survival. KEGG, Kyoto Encyclopedia of Genes and Genomes; lncRNA, long non-coding RNA.

Discussion

HCC is a major health problem worldwide with poor overall prognosis (25,26). Most patients with HCC are diagnosed at advanced stages (III–IV) (27). Earlier diagnosis and more reliable prognosis, based on suitable biomarkers, are crucial for improving the management and therefore outcomes of patients with HCC. Accumulating evidence has suggested that the abnormal expression of lncRNAs is associated with the recurrence, metastasis and prognosis of HCC (28–30). Since the prognosis in HCC may differ depending on whether it is alcohol-related or not, the present study developed a risk scoring system based on lncRNA expression to evaluate the risk of poor OS or RFS in alcohol-related patients with HCC. The results of the present study may suggest good potential for lncRNAs to be prognostic biomarkers in alcohol-related HCC.

The results of the present study demonstrated that the risk-scoring system and vascular invasion were important independent predictors of prognosis in the sample of patients with HCC. The AUCs for OS and RFS risk scoring systems were high, suggesting good predictive power. Thus, an lncRNA-based risk scoring system may be used to estimate the risk scores of different alcohol-related patients with HCC, predict survival and determine treatment.

Previous studies have identified lncRNAs as prognostic biomarkers for HCC using the TCGA database (31,32). To the best of our knowledge, the present study is the first to analyze alcohol-related HCC. The present study identified eight lncRNAs as potential prognostic biomarkers for alcohol-related HCC. Among them, LINC01605 has been demonstrated to be upregulated in bladder cancer tissues and may be associated with poor prognosis (33), whereas ERVH48-1 has been identified as a prognostic biomarker for tongue squamous cell carcinoma (34). The remaining potential biomarkers from the present study (AC012640.1, AC013451.2, AC062004.1, LINC02334, LINC02043, and AL139385) do not appear to have been analyzed in detail. The eight lncRNAs in this model appear to be involved in chemical carcinogenesis, metabolism and the cell cycle. Investigating these lncRNA-mediated pathways may provide new insights into the development of alcohol-related HCC.

There are some limitations in this study. First, HCC treatment types were not included in the multivariate Cox regression due to lack of data. Second, Cox analyses may be less accurate because some clinical data were missing for some patients. Third, the sample was relatively small, and as a result the present study could not divide the samples into training and test dataset for determining and validating the model. Thus the findings of the present study should be verified and extended in larger studies.

Despite these limitations, the results of the present study suggested that an lncRNA-based risk scoring system may predict the risk of poor prognosis in patients with alcohol-related HCC. Eight lncRNAs are independent clinicopathological variables for alcohol-related HCC.

Acknowledgements

Not applicable.

Funding

This study was supported by the Guangxi Key Research and Development Program (grant. no. GuiKeAB16380215), Guangxi First-class Discipline Project for Basic Medicine Sciences (grant. no. GXFCDP-BMS-2018) and Guangxi Medical University Training Program for Distinguished Young Scholars.

Availability of data and material

The datasets used during the present study are included in this published article.

Authors' contributions

YUL. and JY designed and performed the research, analyzed and interpreted the data, and drafted the manuscript. JW collected and analyzed the data. YOL and JZ conceived the study, designed the methodology and reviewed the manuscript. All authors read and approved the final version of the manuscript.

Ethics approval and consent to participate

Not applicable.

Patient consent for publication

Not applicable.

Competing interests

The authors declare that they have no competing interests.

References