A prognostic 11-DNA methylation signature for lung squamous cell carcinoma.

BACKGROUND: Lung squamous cell carcinoma (LUSC), as the second frequent subtype of lung cancer, causes lots of mortalities primarily due to a lack of precise prognostic markers and timely treatment intervention. Previous studies have constructed several risk prognostic models based on DNA methylation sites in multiple tumors, whereas, DNA methylation signature of LUSC remains to be built, and its predictive value need to be evaluated.
METHODS: The genome-wide DNA methylation data of LUSC samples was obtained from The Cancer Genome Atlas dataset. Univariate Cox analysis and the least absolute shrinkage and selection operator (LASSO) were implemented to identify DNA methylation sites related to overall survival of LUSC patients. Thus, we performed multivariate Cox regression to establish a DNA methylation signature. The Kaplan-Meier (K-M) survival curves and time-dependent receiver operating characteristic (ROC) curves were plotted to estimate the prognostic power of the signature. Comparison with other known prognostic biomarkers, our DNA methylation signature showed higher predictive specificity and sensitivity. In addition, multivariate Cox regression screened out independent prognostic factors and constructed a nomogram.
RESULTS: Several statistical methods were performed to construct an 11-DNA methylation signature. LUSC patients were divided into low- and high-risk group based on risk score, and high-risk group had a shorter survival time. According to the results of K-M and ROC analyses, the 11-DNA methylation signature showed significant sensitivity and specificity in predicting the LUSC patients' overall survival. Finally, we integrated some independent prognostic factors (risk score, metastasis stage, and tobacco smoking history) to construct a nomogram, which has excellent prognostic power and may provide guidance for the therapeutic strategies.
CONCLUSIONS: We constructed the first risk prognosis model based on DNA methylation site in LUSC, which showed better predictive ability. In addition, a nomogram integrating the DNA methylation signature, metastasis stage, and tobacco smoking history was developed. 2020 Journal of Thoracic Disease. All rights reserved.

Introduction

Lung cancer is the most frequent malignancy and the leading cause of cancer death worldwide (1) Lung squamous cell carcinoma (LUSC) is the second frequent subtype of lung cancer, accounts for approximately 30% (2,3). Although the progress has been made in early diagnosis and therapy, the 5-year survival of LUSC patients remains dissatisfactory. There is still a lack of effective biomarkers for identifying patients with high risk of recurrence and poor prognosis. Hence, there is an urgent need to find effective biomarkers to improve the ability of clinical prognosis prediction and make individualized therapy decisions.

Emerging studies have indicated that epigenetics plays a vital role in the occurrence, development, therapy response, and outcome of human tumors (4,5). The occurrence and development of cancer have been accompanied by abnormal DNA methylation, which has great potential as a biomarker of prognosis (6). For instance, p16 methylation induced paclitaxel resistance in non-small cell lung cancer (NSCLC), so it can predict paclitaxel chemosensitivity (7). KLF2 region 4 hypermethylation led to the downregulation of KLF2 and promoted the proliferation and metastasis in NSCLC cells (8). Downregulation of miR-1247 by DNA methylation promoted invasion and migration of NSCLC by targeting STMNI (9). Moreover, predictive models based on DNA methylation sites have been constructed in some tumors, such as ovarian serous cystadenocarcinoma, cutaneous melanoma, gastric adenocarcinoma and choroid plexus tumor (10-13). However, the prediction model based on DNA methylation sites remains to be constructed in LUSC.

In our present study, we aimed to construct a novel prognostic DNA methylation signature related to patient’s overall survival. The DNA methylation and follow-up data were downloaded from The Cancer Genome Atlas (TCGA) dataset. We performed several statistical methods, including univariate Cox regression, the least absolute shrinkage and selection operator (LASSO), and multivariate Cox regression methods to reduce dimensionality. As a result, an independent prognostic model based on 11-DNA methylation sites was successfully constructed. Besides, to improve the clinical practicability of the risk prognosis model, the metastasis stage and tobacco smoking history were integrated into the 11-DNA methylation signature and the nomogram was constructed.

Methods

Collection of DNA methylation and clinical data from TCGA and differential DNA methylation sites selection

Genome-wide DNA methylation data (level 3) and corresponding follow-up data of LUSC were downloaded from the TCGA dataset (http://cancergenome.nih.gov/). The DNA methylation data were detected by Infinium HumanMethylation450 BeadChip. Differential DNA methylation sites were identified between LUSC tissues and paracancerous tissues using the limma package (version 3.34.7; https://bioconductor.org/packages/release/bioc/html/limma.html). The selection criteria were fold change >2 or <0.5 and false discovery rate (FDR) <0.01. After removing tissues without survival records and follow-up time, filtered tissues were analyzed in the following study.

Development of DNA methylation signature in survival prediction

LUSC patients were classified into training set and testing set by random grouping method. All initial analyses were performed in the training set to construct a signature based on the DNA methylation site and validated the signature in the testing set. The DNA methylation sites associated with the overall survival of LUSC patients were screened using univariate Cox proportional hazard analysis with P<0.05 as statistical significance. LASSO analysis is a high-dimensional indicator regression method, which obtains a more refined model by compressing some regression coefficients. LASSO analysis was used to screen the critical DNA methylation sites from the significant DNA methylation sites in univariate Cox regression analysis using R with glmnet package (Version 3.0-2, https://CRAN.R-project.org/package=glmnet). Thus, we performed multivariate Cox regression, stepwise regression, to reduce dimensionality and establish a risk score formula weighted by the corresponding coefficients. The univariate and multivariate cox regression analysis used survival package (Version 2.41-1, http://bioconductor.org/packages/survivalr/) in R language. The risk score of each patient in training set was calculated according to the above formula. According to the median value, patients were classified into low- and high-risk groups. Survival difference between the low- and high-risk group was assessed by the Kaplan-Meier (K-M) survival analysis using R with survival package (Version 2.41-1, http://bioconductor.org/packages/survivalr/). To evaluate the predictive performance at 5 years of the DNA methylation signature, the time-dependent receiver operating characteristic (ROC) curve was performed using R with survivalROC package (Version 1.0.3, https://CRAN.R-project.org/package=survivalROC). Subsequently, we perform K-M survival analysis and ROC analysis to evaluate predictive accuracy of this signature in the testing set based on the same cutoff value. The area under the curve (AUC) is used as the evaluation criterion of the signature.

Construction and evaluation of nomogram

We performed multivariate Cox regression analysis to examine whether constructed DNA methylation signature is independent of other clinical data, consisting of age, gender, tumor stage, TNM stage, and tobacco smoking history. According to the results of multivariate Cox regression analysis, we constructed a nomogram for individualized prediction of overall survival to predict 1-, 3-, and 5-year overall survival using R with rms package (Version 5.1-4, https://CRAN.R-project.org/package=rms). Then, we performed ROC curve to appraise the predictive performance of this nomogram and only the DNA methylation signature (5-year survival).

Results

Data gathering and differential methylation analysis

A total of 412 samples with 485,577 DNA methylation sites were acquired from the TCGA dataset, including 370 LUSC tissues and 42 paracancerous tissues. Among them, 363 LUSC samples with clinical follow-up information were further randomly divided into two groups, 183 patients as a training set and 180 patients as a testing set. The clinical data of age, gender, race, tumor stage, TNM stage, and tobacco smoking history was summarized (). Compared with the paracancerous tissues, 15,343 differential DNA methylation sites were selected in LUSC tissues using fold change >2 or <0.5 and FDR <0.01 as the criteria.

Table S1

clinicopathological characteristics of the LUSC patients from TCGA datasets

Characteristics	Groups	Entire dataset (n=363)			Training dataset (n=183)			Testing dataset (n=180)
		N	%		N	%		N	%
Age, years	<70	185	51.0		91	49.7		94	52.2
	≥70	173	47.7		88	48.1		85	47.2
	Unknown	5	1.4		4	2.2		1	0.6
Gender	Female	94	25.9		48	26.2		46	25.6
	Male	269	74.1		135	73.8		134	74.4
T	T1	90	24.8		41	22.4		49	27.2
	T2	201	55.4		102	55.7		99	55.0
	T3	59	16.3		34	18.6		25	13.9
	T4	13	3.6		6	3.3		7	3.9
N	N0	233	64.2		117	63.9		116	64.4
	N1	95	26.2		49	26.8		46	25.6
	N2	29	8.0		12	6.6		17	9.4
	Unknown	6	1.7		5	2.7		1	0.6
M	M0	261	71.9		133	72.7		128	71.1
	M1	35	9.6		18	9.8		17	9.4
	MX	67	18.5		32	17.5		35	19.4
Tumor stage	Stage 1	170	46.8		83	45.4		87	48.3
	Stage 2	131	36.1		69	37.7		62	34.4
	Stage 3	55	15.2		26	14.2		29	16.1
	Stage 4	4	1.1		3	1.6		1	0.6
	Unknown	3	0.8		2	1.1		1	0.6
Tobacco smoking history	Lifelong non-smoker	13	3.6		6	3.3		7	3.9
	Current smoker	114	31.4		63	34.4		51	28.3
	Current reformed smoker for >15 years	54	14.9		32	17.5		22	12.2
	Current reformed smoker for ≤15 years	167	46		75	41.0		92	51.1
	Current reformed smoker, duration not specified	5	1.4		2	1.1		3	1.7
	Unknown	10	2.8		5	2.7		5	2.8
Race	White	273	75.2		141	77.0		132	73.3
	Black or African American	23	6.4		9	4.9		14	7.8
	Asian	7	1.9		6	3.3		1	0.6
	Unknown	60	16.5		27	14.8		33	18.3

LUSC, lung squamous cell carcinoma; TCGA, The Cancer Genome Atlas.

DNA methylation signature establishment and validation

As showed in the workflow diagram (), we used the training set to construct DNA methylation signature and validated the predictive ability of signature in the testing set. First, we carried out the univariate Cox regression to filter DNA methylation sites associated with overall survival of LUSC patients in training set. Then, 392 DNA methylation sites were significantly associated with overall survival of patients (P<0.01). Next, these selected DNA methylation sites were put into LASSO analysis. Therefore, 44 DNA methylation sites were selected as critical sites that were of significance in univariate analysis (). Multivariate Cox regression was performed on these 44 DNA methylation sites, stepwise regression and screening, and a risk prognosis model including 11-DNA methylation sites was determined as the optimal risk prognosis formula to predict overall survival (). The genes corresponding with these 11-DNA methylation sites were RASSF6 (Ras association domain family member 6), LHX5 (LIM homeobox 5), ZNF773 (zinc finger protein 773), HES7 (hes family bHLH transcription factor 7), APOBEC3C (apolipoprotein B mRNA editing enzyme catalytic subunit 3C), INSM2 (INSM transcriptional repressor 2), RPS18 (ribosomal protein S18), SPC25 (SPC25 component of NDC80 kinetochore complex), TRIM71 (tripartite motif containing 71), and ISL2 (ISL LIM homeobox 2), except for cg20565374. The correlation between the methylation degree of the DNA methylated sites screened and their corresponding gene expression was also analyzed ().

Figure 1

The workflow of construction of LUSC survival-related 11-DNA methylation signature. LUSC, lung squamous cell carcinoma.

Figure 2

Identification of key prognostic DNA methylation sites. (A) LASSO coefficient profiles of the DNA methylation sites; (B) partial likelihood deviance was plotted corresponding log (Lambda). LASSO, least absolute shrinkage and selection operator.

Table 1

The 11 prognosis-associated DNA methylation sites to construct the risk score system

Markers	Ref. gene	Coefficients	HR	P value
cg00224911	RASSF6	11.891	146,007.8	0.069
cg00802728	LHX5	4.359	78.17074	0.008
cg03612039	ZNF773	−2.344	0.095916	0.012
cg07148818	HES7	7.826	2,503.751	0.046
cg07186138	APOBEC3C	6.770	871.4044	0.063
cg11082362	INSM2	7.250	1,407.633	0.055
cg12086028	RPS18	2.692	14.76412	0.018
cg13605690	SPC25	29.905	9.71E+12	<0.001
cg18249634	TRIM71	−1.033	0.356052	0.04
cg20565374	chr17:20687569-20687913	−1.460	0.232338	0.103
cg20643871	ISL2	−1.473	0.229217	0.007

HR, hazard ratio.

Figure S1

Correlation between methylation levels of each DNA methylation site and expression of corresponding gene was assessed by Pearson’s correlation test.

Based on the corresponding coefficients of the prognostic methylation β-values, a risk score formula was generated for predicting prognosis. Risk score =11.891× β-value of cg00224911 + 4.359 × β-value of cg00802728 − 2.344 × β-value of cg03612039 + 7.826 × β-value of cg07148818 + 6.770 × β-value of cg07186138 + 7.250 × β-value of cg11082362 + 2.692 × β-value of cg12086028 + 29.905 × β-value of cg13605690 − 1.033 × β-value of cg18249634 − 1.460 × β-value of cg20565374 − 1.473 × β-value of cg20643871. Cg00224911, cg00802728, cg07148818, cg07186138, cg11082362, cg12086028 and cg13605690 were negative related to overall survival in LUSC patients while cg03612039, cg18249634, cg20565374 and cg20643871 were positive factors. To evaluate the predicted performance of 11-DNA methylation signature, patients were classified into high-risk (N=91) and low-risk (N=92) groups using the median score as the threshold. First, the distribution of risk score, survival status, and β-value of methylation sites was analyzed in the training set (), and then confirmed in the testing set (). We analyzed the β-value of each methylation site in the signature of the high- and low-risk groups in the training set (). K-M survival curves confirmed that the risk score was significantly related to overall survival and AUC is 0.787 (). Subsequently, the 11-DNA methylation signature was evaluated in testing set. Using the same risk score formula and threshold value, patients in testing set were divided into two groups: high-risk group (N=103) and low-risk group (N=77). The high-risk group also had a shorter survival time, and AUC was 0.750 (). The results demonstrated that our 11-DNA methylation signature performed significant sensitivity and specificity in assessing LUSC patients’ overall survival.

Figure 3

Distribution of the risk score, survival status, and β value of methylation sites in training set (A) and testing set (B). K-M survival curves along with the log-rank test and ROC analysis to evaluate performance of this risk score formula in training set (C) and testing set (D). AUC, area under the curve; K-M, Kaplan-Meier; ROC, receiver operating characteristic.

Figure S2

Compare the β-values of each DNA methylation sites between the high-risk and low-risk groups of LUSC patients in the training set. “L” represents low-risk group. “H” represents high-risk group. The difference between the high-risk and low-risk groups was determined by the log-rank test. *, P<0.05; **, P<0.01; ***, P<0.001; ****, P<0.0001.

Detection of predicted power of 11-DNA methylation signature in different clinical characteristics

A crucial characteristic of a great prognostic signature should be independent or added to the clinical pathology prognostic factors currently in use. Clinicopathologic characteristics, including patients’ age, gender, tumor stage, TNM stage, and tobacco smoking history, have been considered as chief prognostic factors for patients with LUSC. In order to evaluate the independence and reliability of the 11-DNA methylation signature, patients were regrouped based on different clinical pathology features. Several factors were related to prognostic survival, consisting of age, gender, tumor stage, TNM stage, and tobacco smoking history. Age and gender were related to prognosis in NSCLC patients (14,15). All LUSC patients were classified into two groups according to their initial diagnosis age: <70 (N=185) and ≥70 (N=173), to analyze the prognostic predictive effect of this 11-DNA methylation signature in patients of different age groups. K-M curves suggested that overall survival time of high-risk group was worse in both age cohorts, with AUC values of 0.789 and 0.743, respectively (Figure S3A), indicating that the 11-DNA methylation signature was independent of age. Based on patients’ gender, patients were classified into 269 males and 94 females. The overall survival was significantly different between high- and low-risk groups, and AUC in male and female cohorts was 0.774 and 0.736, respectively (Figure S3B). The prognosis of patients in T1 and T2 was significantly better than patients in T3 and T4 (16). Compared with low-risk patients, the overall survival time of high-risk patients was significantly shortened, and the AUC in T1 and T2 (N=291) was 0.771. Nevertheless, in T3 and T4 (N=72), there was no significant difference in overall survival between the high- and low-risk groups (Figure S4A). Given that distant metastasis or lymph node metastasis can seriously affect the prognosis of patients, we regrouped patients according to whether the tumor has lymph node metastasis or distant metastasis. K-M and ROC analyses indicated that the prognosis of high-risk groups was significantly worse than low-risk groups (). The above results suggested that this 11-DNA methylation signature provides a superior reference for different distant metastasis or lymph node metastasis cohorts due to the effectiveness of risk stratification. Compared with early lung cancer, advanced lung cancer is more prone to recurrence and shorter survival time (17). As for tumor stage, we evaluated the predictive power of this 11-DNA methylation signature in stage 1 (N=170), stage 2 (N=131), stages 3 and 4 (N=59). In stages 1 and 2, the high-risk patients had obviously shorter overall survival, and AUC values in stages 1 and 2 cohorts were 0.774 and 0.762, respectively (Figure S4B). However, there was no significant difference in the overall survival of the high- and low-risk groups in stages 3 and 4, probably due to small numbers (Figure S4B). Tobacco serves as an important risk factor for NSCLC, approximately 80% of which is associated with smoking that closely related to DNA methylation (18-20). Based on the patient’s tobacco smoking history, patients were classified into three groups: current smoker (N=114), current reformed smoker for >15 years (N=54) and current reformed smoker for ≤15 years (N=167), and then to analyze the prognostic predictive efficiency of the 11-DNA methylation signature in patients of different tobacco smoking history. As shown, the difference in the overall survival between low- and high-risk groups was also significant, and AUC values of different smoking history groups were greater than 0.75 (). Results of K-M and ROC analyses according to various regrouping methods were also summarized in . The above results suggested that this 11-DNA methylation signature showed satisfactory availability when patients were regrouped according to different clinical pathology features, indicating that the 11-DNA methylation signature was an independent and applicative prognostic predictor of patients’ survival.

Figure 4

Kaplan-Meier and ROC analyses of patients with LUSC in different N cohorts: N0 (N=233) and N1 (N=124), respectively (A), different M cohorts: M0 (N=261) and M1 (N=35), respectively (B) and different smoking history cohorts: current smoker (N=114), current reformed smoker for ≤15 years (N=167), and current reformed smoker for >15 years (N=54) respectively (C). ROC, receiver operating characteristic; LUSC, lung squamous cell carcinoma.

Table S2

Kaplan-Meier and ROC analysis of various regrouping methods

Regrouping factors	Group	Sample size	Kaplan-Meier, P value	AUC
Age at diagnosis, years	<70	185	2.048e−04	0.789
	≥70	173	3.611e−08	0.743
Gender	Female	94	6.473e−03	0.736
	Male	269	2.568e−09	0.774
T	T1+2	291	4.749e−10	0.771
	T3+4	72	5.953e−02	0.713
N	N0	233	1.427e−06	0.788
	N1+2	124	2.832e−05	0.719
M	M0	261	7.314e−08	0.740
	M1	35	3.256e−03	0.901
Tumor stage	Stage 1	170	1.471e−06	0.774
	Stage 2	131	3.308e−05	0.762
	Stage 3 + stage 4	59	8.684e−02	0.726
Tobacco smoking history	Current smoker	114	2.838e−03	0.784
	Current reformed smoker for >15 years	54	4.32e−03	0.815
	Current reformed smoker for ≤15 years	167	1.325e−05	0.759

AUC, area under the curve; ROC, receiver operating characteristic.

Establishment of the nomogram

According to the results from univariate analysis, histologic grade, tumor stage, lymph node stage, metastasis stage, and tobacco smoking history were significantly related to overall survival of patients with LUSC (). Through multivariate analysis of the above factors, metastasis stage and tobacco smoking history and the risk score, independent and stable prognostic factor (), were used to construct a nomogram (). Compared with the 11-DNA methylation signature, the nomogram shows higher accuracy of 5-year survival prediction (AUC =0.811, ).

Table 2

The univariable and multivariable Cox regression analysis of the 11-DNA methylation signature in LUSC patients

Variables	Univariate analysis				Multivariate analysis
	HR	95% CI of HR	P value		HR	95% CI of HR	P value
Age, years
<50	1 (reference)	–	–		–	–	–
50–59	1.398	0.382–5.116	0.613		–	–	–
60–69	2.072	0.639–6.722	0.225		–	–	–
70–79	2.124	0.662–6.813	0.205		–	–	–
≥80	2.143	0.474–9.689	0.322		–	–	–
Gender
Female	1 (reference)	–	–		–	–	–
Male	1.036	0.680–1.579	0.87		–	–	–
T
T1	1 (reference)	–	–		1 (reference)	–	–
T2	1.164	0.734–1.844	0.5186		1.0809	0.6552–1.7831	0.760717
T3	1.386	0.754–2.546	0.2928		0.6722	0.2622–1.7229	0.408144
T4	2.918	1.357–6.273	0.0061		0.9810	0.2627–3.6634	0.977234
N
N0	1 (reference)	–	–		1 (reference)	–	–
N1	0.998	0.663–1.504	0.9927		0.6139	0.3031–1.2437	0.175590
N2	1.851	1.040–3.293	0.0362		0.7302	0.2082–2.5612	0.623388
M
M0	1 (reference)	–	–		1 (reference)	–	–
M1	3.520	2.151–5.760	5.47e−07		2.9969	1.6852–5.3295	0.000186
Tumor stage
Stage 1	1 (reference)	–	–		1 (reference)	–	–
Stage 2	1.135	0.742–1.736	0.55989		1.3480	0.6438–2.8227	0.428389
Stage 3	2.018	1.273–3.199	0.00284		2.0929	0.5505–7.9574	0.278436
Stage 4	4.015	1.242–12.976	0.02021		2.5101	0.6633–9.4986	0.175282
Tobacco smoking history
Smoking	1 (reference)	–	–		1 (reference)	–	–
Less 15	0.5029	0.2776–0.9112	0.02340		0.5464	0.2894–1.0319	0.062437
More 15	0.5477	0.3711–0.8082	0.00243		0.6039	0.4050–0.9006	0.013383
Risk score
Low	1 (reference)	–	–		1 (reference)	–	–
High	2.869	1.943–4.238	1.18e−07		2.738	1.812−4.138	1.74e−06

LUSC, lung squamous cell carcinoma; HR, hazard ratio; CI, confidence interval.

Figure 5

Development of nomogram for lung squamous cell carcinoma. (A) The nomogram for predicting probabilities of patients with 1-, 3- and 5-year overall survival; (B) ROC curve based on the 11-DNA methylation signature and nomogram for overall survival probability. ROC, receiver operating characteristic; AUC, area under the curve.

Association of the 11-DNA methylation signature with tumor recurrence and distant metastasis

We next studied the utility of the risk score in assessing tumor recurrence and distant metastasis of LUSC. Clinical and demographic features, including age, gender, race, tumor stage, TNM stage and tobacco smoking history were included in the analysis. The risk score of patients with metastasis (N=32) was significantly higher than those without metastasis (N=253) (). Similarly, the risk scores of patients with tumor recurrence (N=89) were significantly higher than those with no tumor recurrence (N=206) (). Collectively, these results indicate that risk scores can be used to predict tumor recurrence, metastasis, and surveillance.

Figure 6

Association of the 11-DNA methylation signature with tumor recurrence and distant metastasis. (A) The risk score in LUSC patients with non-metastasis and metastasis; (B) the risk score in LUSC patients with non-recurrence and recurrence. **, P<0.01; ***, P<0.001. LUSC, lung squamous cell carcinoma.

Comparison of the 11-DNA methylation signature with other known prognostic biomarkers

Previous studies have focused on building predictive signatures using protein-coding genes or miRNAs or lncRNAs. For instance, cathepsin B (CTSB) is a predictor of poor prognosis and promotes tumor metastasis and might have the potential to be a therapeutic target for LUSC (21). Zhang et al. constructed a prognostic signature using 17 mRNAs and a miRNA in LUSC (22). Based on lncRNA expression, Wang et al. identified eight lncRNAs as a prognostic signature (23) and Tang et al. constructed a predictive 5-lncRNA model (24). PD-L1 can severe as a poor prognostic signature in LUSC patients (25). CD271 promoted cell proliferation and was related to the poor prognosis of LUSC (26). RBMS3, as a tumor suppressor gene, inhibited the occurrence and development of LUSC (27). The expression of RRM1 and ERCC1 was related to the better prognosis of patients with LUSC (28). Li et al. identified methylation-driven genes and used four methylation driving genes GCSAM, GPR75, NHLRC1 and TRIM58 as prognostic indicators of LUSC. They used the average methylation level of the methylation-driven gene to build a prognostic model, instead of methylation sites (29). To evaluate whether our DNA methylation signature has a robust and reliable performance advantage, we compared the sensitivity and specificity of our DNA methylation signature with other known prognostic signatures in the same 363 patients with LUSC (). According to the results of the ROC analysis, the predicted performance at 5 years of our 11-DNA methylation signature was better than other known prognostic biomarkers, including mRNAs, miRNAs, and lncRNAs. All the above results showed that the 11-DNA methylation signature had better stability and reliability and was currently the best predictor of overall survival in predicting LUSC patients.

Figure 7

ROC curves were used to assess the sensitivity and specificity of the 11-DNA methylation signature and other known biomarkers in predicting the overall survival of LUSC patients. AUC, area under the curve; LUSC, lung squamous cell carcinoma.

Discussion

Despite advances in the prevention, diagnosis, and therapy of LUSC over the past few decades, the 5-year survival rate remains low, less than 15 percent (17). Therefore, the prognostic prediction of LUSC patients is critical to the selection and improvement of appropriate treatment options. To distinguish between high- and low-risk patients for more effective management, previous studies developing a series of molecular biomarkers related to the prognosis of LUSC patients have focused on protein-coding genes or miRNAs or lncRNAs while ignoring the impact of methylation on patient’s survival. With the deepening of epigenetic research, increasing evidence has shown that DNA methylation is critical to gene regulation and is early events of some tumors. DNA methylation is one of the earliest detectable neoplastic changes that give it a unique advantage as cancer diagnosis and prognosis biomarkers (30-32). In addition, a prognostic signature formed by combining multiple DNA methylation sites has higher sensitivity and specificity than a single DNA methylation site (33). Our study emphasized the potential role for a combination of epigenetic biomarkers in improving prognosis prediction and providing tailored therapeutic decisions, as well as providing alternative biomarkers and therapeutic targets for LUSC patients.

Our study first identified differential methylation sites according to genome-wide DNA methylation analysis. We performed COX regression and ROC analysis to identify an 11-DNA methylation signature that was significantly related to overall survival of LUSC patients. To detect the predictive performance and independence of the 11-methylation signature, patients were regrouped based on different clinicopathological features (age, gender, tumor stage, TNM stage, and tobacco smoking history). We used K-M and ROC analysis to estimate the prognostic ability of the 11-DNA methylation signature in different subgroups. Based on the risk scores of the 11-DNA methylation signature, we performed risk stratification and survival prediction for LUSC patients. In addition, comparison of our 11-DNA methylation signature with other known prognostic biomarkers indicates that it has significantly higher sensitivity and specificity in the prognosis prediction of LUSC. Among these 11 methylation sites, 10 sites have corresponding reference genes. RASSF6 is a tumor suppressor with methylation of its promoter region leading to decreased expression, thereby promoting melanoma development and brain metastasis (34,35). ZNF773 has a higher level of DNA methylation in human papillomavirus-related oropharyngeal squamous cell carcinoma compared to normal samples (36). HES7 is a biomarker gene for early epithelial-mesenchymal transition in lung adenocarcinoma (37). SPC25 increases tumor stem cell characteristics in NSCLC and pancreatic cancer, and enhances cell proliferation and poor prognosis of breast cancer (38-40). TRIM71 promotes cell proliferation in NSCLC and hepatocellular carcinoma (41,42). Nevertheless, the relationship between five of these ten corresponding reference genes (LHX5, APOBEC3C, RPS18, ISL2, and INSM2) and tumor biology and the related molecular mechanisms have not been studied.

Gene expression is affected by epigenetic changes, and inactivation of tumor suppressor genes caused by DNA methylation is related to occurrence and development in multiple tumors, consisting of LUSC (43,44). Although DNA methylation can affect gene regulation, there are a few exceptions (45,46). In our signature, the expression of APOBEC3C, LHX5, SPC25, RPS18, and ZNF773 were negatively related to the methylation levels (P<0.05), but no association between the expression and the methylation level of other five genes (HES7, INSM2, ISL2, RASSF6, and TRIM71). Further, we will focus on verifying the biological functions of these 11-DNA methylation sites and their corresponding genes through more experiments, which may provide more targets and therapeutic decisions.

To improve a more sensitive and specific prognostic signature for LUSC, we constructed a prognostic nomogram that combines the 11-DNA methylation signature with distant metastasis of the patient’s tumor and smoking history and demonstrates more satisfied predictive performance. To apply the model to the clinic in the future, more clinical investigations are needed to assess the robustness of this 11-DNA methylation signature. It is undeniable that there may be some deviations in the process of constructing a model by selecting prognostic-related DNA methylation sites. The correlation analysis suggests that subsequent research should focus on the combination of mRNA and DNA methylation signature to construct better prognostic biomarkers.

Conclusions

In conclusion, we constructed the first risk prognosis model based on DNA methylation site in LUSC, which had better stability and reliability and was currently the best predictor of overall survival in predicting LUSC patients. In addition, in order to better apply the risk prognosis model to clinical decision-making, a nomogram integrating the DNA methylation signature, metastasis stage, and tobacco smoking history was developed.

Kaplan-Meier and ROC analyses of patients with LUSC in different age cohorts: <70 (N=185) and ≥70 (N=173), respectively (A) and in different gender cohorts: male (N=269), female (N=94), respectively (B). AUC, area under the curve; ROC, receiver operating characteristic; LUSC, lung squamous cell carcinoma.

Kaplan-Meier and ROC analyses of patients with LUSC in different T cohorts: T1, T2 (N=291) and T3, T4 (N=72), respectively (A) and in different stage cohorts: stage 1 (N=170), stage 2 (N=131), and stages 3, 4 (N=59), respectively (B). AUC, area under the curve; ROC, receiver operating characteristic; LUSC, lung squamous cell carcinoma.

The article’s supplementary files as