Genetic variants of DOCK2, EPHB1 and VAV2 in the natural killer cell-related pathway are associated with non-small cell lung cancer survival

Abstract

Although natural killer (NK) cells are a known major player in anti-tumor immunity, the effect of genetic variation in NK-associated genes on survival in patients with non-small cell lung cancer (NSCLC) remains unknown. Here, in 1,185 with NSCLC cases of a discovery dataset, we evaluated associations of 28,219 single nucleotide polymorphisms (SNPs) in 276 NK-associated genes with their survival. These patients were from the reported genome-wide association study (GWAS) from the Prostate, Lung, Colorectal and Ovarian (PLCO) Cancer Screening Trial. We further validated the findings in an additional 984 cases from the Harvard Lung Cancer Susceptibility (HLCS) Study. We identified three SNPs (i.e., DOCK2 rs261083 G>C, VAV2 rs2519996 C>T and EPHB1 rs36215 A>G) to be independently associated with overall survival (OS) in NSCLC cases with adjusted hazards ratios (HRs) of 1.16 (95% confidence interval [CI] = 1.07-1.26, P = 3.34×10^-4), 1.28 (1.12-1.47, P = 4.57×10^-4) and 0.75 (0.67-0.83, P = 1.50×10^-7), respectively. Additional joint assessment of the unfavorable genotypes of the three SNPs showed significant associations with OS and disease-specific survival of NSCLC cases in the PLCO dataset (P _trend<0.0001 and <0.0001, respectively). Moreover, the survival-associated DOCK2 rs261083 C allele had a significant correlation with reduced DOCK2 transcript levels in lung adenocarcinoma (LUAD), while the rs36215 G allele was significantly correlated with reduced EPHB1 transcript levels in lymphoblastoid cell lines in the 1000 Genomes Project. These results revealed that DOCK2 and EPHB1 genetic variants may be prognostic biomarkers of NSCLC survival, likely via transcription regulation of respective genes.

Introduction

As the deadliest malignancy worldwide, lung cancer has contributed to a large portion of cancer-related deaths globally and caused about 228,820 new cases and 135,720 deaths in the United States in 2020 [1,2]. To date, lung cancer remains a heavy burden on public health of the general population. Accounting for about 85% of all lung cancer cases, the heterogeneous non-small cell lung cancer (NSCLC) comprises two broad subtypes: lung squamous cell carcinomas (LUSC) and lung adenocarcinomas (LUAD) [3]. Despite advances made in the treatment approaches, the prognosis of lung cancer patients remains poor, with <20% of individuals surviving up to 5 years after the diagnosis [4]. Moreover, clinical differences in the patients’ response to the same treatment have been observed [5], suggesting that host factors, including genetic factors such as single nucleotide polymorphisms (SNPs), could also play an essential role in outcomes of the patients; therefore, it is paramount to determine important genetic variants as biomarkers for evaluating therapeutic response and outcomes of NSCLC cases.

Because SNPs can affect their gene expression and thus functions, they may have a significant impact on cancer prognosis [6,7]. However, genome-wide association studies (GWASs) have detected only few novel and functional SNPs that predict survival of NSCLC patients, because GWASs on survival are exploratory and have focused on SNPs or genes with a stringent P-values of 5×10^-8 as a result of multiple testing comparison; meanwhile, the vast majority of the reported SNPs lack of functional annotations [8]. As a promising hypothesis-driven approach in the post-GWAS era, a biological pathway-based method has been used to reanalyze previously reported GWAS datasets and to evaluate the joint impact of SNPs on many genes of the same molecular pathway, which drastically improves the study power and facilitate subsequent functional analysis [9].

Cancer immunotherapy is a promising treatment strategy against cancer, through activating or boosting the immune system by mobilizing anti-cancer immunity, which potentially inhibits some types of cancer. For example, immune checkpoint inhibitors, e.g., programmed death 1 (PD-1) and ligand programmed death-ligand 1 (PD-L1), are applied as the first-line therapy for some advanced or metastatic NSCLC not expressing targetable genetic mutations [10]. In addition, natural killer (NK) cells, as the first line of defense, bridge and orchestrate immune responses, playing a vital anti-cancer role in the host response [11]. Since not all patients benefit from such immunotherapies, a major challenge is to identify individuals with inherited differences in the highly complex interaction involving the tumor and immune cells, including NK cells, for personalized treatment. Hence, we hypothesize that gene variants in the NK cell-related pathway involved in the anti-cancer immune response are associated with survival of NSCLC patients. By using two available published GWAS datasets of NSCLC cases, we tested our hypothesis.

Materials and methods

Study populations

The discovery genotype dataset was from the GWAS of the Prostate, Lung, Colorectal and Ovarian (PLCO) Cancer Screening Trial with 1,185 eligible Caucasian cases. The PLCO trial was a large randomized cancer screening study performed in ten American medical centers from 1993 to 2011, which was funded by the National Cancer Institute (NCI). A total of 155,000 participants (77,500 men and 77,500 women of 55-74 years old) were included in the screening trial. All the participants were randomized to the intervention arm (with screening) and control (standard care) arms [12]. Genomic DNA extraction from whole blood specimens was performed, and the genotyping utilized Illumina Human Hap240Sv1.0 and Human Hap550v3.0 (dbGaP accession: phs000093.v2.p2 and phs000336.v1.p1) [13,14]. The PLCO trial was approved by the institutional review boards of all institutions involved, and informed consent was obtained from all participants.

The validation genotype dataset was from another GWAS performed in the Harvard Lung Cancer Susceptibility (HLCS) Study, which included 984 Caucasian NSCLC cases with available whole blood samples for DNA extraction. Genotyping was carried out with Illumina Human hap610-Quad arrays, and MaCH3.0 was utilized for analysis by referring to the sequencing data for Caucasians in the 1000 Genomes Project [15].

Utilization of both GWAS datasets had the approval from the Internal Review Board of Duke University School of Medicine (Project #Pro00054575) and the National Center for Biological Information (NCBI) for access to the de-identified database of genotypes and phenotypes (dbGaP; Project #6404). The features of the PLCO trial (n = 1185) and the HLCS study (n = 984) are shown in Table S1.

Gene selection and SNP imputation

Genes of the NK cell-associated pathway were identified in the Molecular Signatures Database (http://software.broadinstitute.org/gsea/msigdb/index.jsp) using “natural” AND “killer” AND “cell” as keywords. After excluding 369 duplicated and eight unavailable genes in NCBI, 267 genes were considered candidate genes for further analysis (Table S2). Imputation was carried out for the selected candidate genes with Minimac4 and the reference panel of the 1000 Genomes Project data (phase 3). Then, all the SNPs in the candidate genes and alongside the respective ±2 kb flanking regions were extracted and depicted in Figure 1. Finally, a total of 28,219 SNPs (2,959 genotyped and 25,260 imputed) from the PLCO trial were selected for further analyses (r-square ≥0.3, individual call rate ≥95%, minor allelic frequency ≥5%, and Hardy-Weinberg equilibrium ≥10^-5).

Figure 1.

Study flowchart. The overall procedures of the current study. Abbreviations: NK, Natural killer; SNPs, single-nucleotide polymorphisms; MAF: minor allelic frequency; HWE: Hardy-Weinberg Equilibrium; PLCO, The Prostate, Lung, Colorectal and Ovarian Cancer Screening Trial; HLCS, the Harvard Lung Cancer Susceptibility Study; NSCLC, non-small cell lung cancer.

Statistical analyses

The follow-up duration for the participants in both GWAS datasets spanned from NSCLC diagnosis to final follow-up or death. OS of NSCLC cases constituted the primary endpoint. In single-locus analysis, multivariable Cox proportional hazards regression was utilized for evaluating associations of all SNPs in the 267 NK cell-related pathway genes with OS in the additive genetic model after adjusting for age, sex, smoking status, tumor stage, histology, radiotherapy, chemotherapy, surgery, and the major four principal components (Table S3) of the PLCO genotyping dataset using the GenABEL package of R [16]. Because most of the investigational SNPs were in high linkage disequilibrium (LD) as a result of imputation, we then utilized Bayesian false discovery probability (BFDP) with a cut-off of 0.80 for multiple testing correction for reducing false-positives according to previous reports [17,18]. A prior probability of 0.10 was used for detecting a hazards ratio (HR) of 3.0 for associations with variant genotypes/minor alleles of investigational SNPs at P<0.05. Next, the chosen SNPs in the PLCO discovery dataset were subjected to subsequent validation with the HLCS dataset. To identify independent SNPs, a multivariate stepwise Cox regression model was used, adjusting for patient features and 41 SNPs predicting survival that were previously reported in the PLCO trial dataset. Finally, inverse variance weighted meta-analysis was performed for combining data from the discovery and validation sets using PLINK 1.90 with Cochran’s Q statistics and I ². As both datasets showed no heterogeneity (P _het>0.10 and I ²<50%), the fixed-effects model was used for data assessment. The identified SNPs were also visualized by Manhattan plots and regional association plots.

Then, the combined unfavorable genotypes of the identified SNPs were used to estimate their collective effects on survival probability. The Kaplan-Meier technique was utilized for survival function estimate. The Cochran’s Q-test was also carried out for comparing effect differences between subgroups in the stratified analysis. Subsequently, we carried out expression quantitative trait loci (eQTL) analysis for evaluating correlations of SNPs with the corresponding mRNA transcript levels utilizing a linear regression model in the R v3.6.2. The transcriptional expression data originated from three sources, including 373 European patients of the 1,000 Genomes Project [19], whole blood specimens from 369 subjects and non-diseased lung samples from 383 cases in the genotype-tissue expression (GTEx) project [20], and lung cancer tissues specimens from the The Cancer Genome Atlas (TCGA) database (dbGaP Study Accession phs000178.v10.p8) [21]. Next, bioinformatics functional prediction for the validated SNPs was performed with SNPinfo [22]. (https://snpinfo.niehs.nih.gov), RegulomeDB [23] (http://www.regulomedb.org), and HaploReg [24] (http://archive.broadinstitute.org/mammals/haploreg/haploreg.php) were utilized for predicting the functions of these validated SNPs. Differences in mRNA transcript levels between paired tumor specimens and adjacent noncancerous tissue samples from the TCGA database were evaluated by paired Student t-test. Moreover, associations of transcript levels with survival propability were assessed by the Kaplan-Meier method (http://kmplot.com/analysis/index.php?p=service&cancer=lung). SAS 9.4 (SAS Institute, USA) was utilized for data analysis, unless specified otherwise.

Results

Associations of SNPs in NK cell-asscociated pathway genes with NSCLC OS in the PLCO and HLCS datasets

The overall flowchart is presented in Figure 1. Baseline features for respective 1,185 and 984 NSCLC cases from the PLCO and HLCS study were reported previously [25]. The discovery PLCO dataset encompassed 28,219 SNPs (including 2,959 genotyped; 25,260 imputed SNPs) in the 267 NK cell-related pathway genes available for the initial analysis, of which 1,010 SNPs were statistically significant associated with OS in NSCLC (P<0.05) following multiple testing correction by BFDP≤0.8. After subsequent validation in the HLCS validation dataset, 16 SNPs still showed statistical significance.

Associations between independent SNPs and with NSCLC OS in the PLCO dataset

To determine whether the above-mentioned 16 SNPs were independently associated with NSCLC survival, adjustment was made for other covariates and previously reported SNPs in the same dataset, using the same Cox regression model. However, the individual genotyping and clinical data were only available for patients from the PLCO trial but not for those from the HLCS study. Therefore, we first performed stepwise multivariate Cox regression analysis for evaluating the impact of the 16 validated SNPs on OS only in the PLCO dataset. In stepwise Cox regression analysis, nine SNPs were identified with significant and independent associations with NSCLC OS. Subsequently, we further adjusted for other 41 additional SNPs that were previously reported to be associated with OS in the same PLCO dataset. As a result, three SNPs (DOCK2 rs261083 G>C, VAV2 rs2519996 C>T and EPHB1 rs36215 A>G) were still statistically significant in associations with NSCLC OS (P = 0.004, P = 0.019 and P<0.0001, respectively, in the additive model) (Table 1). Subsequent meta-analysis of the above-mentioned three independent SNPs yielded consistent results (P = 3.34×10^-4, 4.57×10^-4 and 5.00×10^-4, respectively) without heterogeneity across these two datasets (Table 2).

Table 1.

Three independent SNPs in a multivariate Cox proportional hazards regression analysis with adjustment for other covariates and 41 previously published SNPs for NSCLC in the PLCO Trial

Variables	Category	Frequency	HR (95% CI)a	P a	HR (95% CI)b	P b
Age	Continuous	1185	1.03 (1.02-1.05)	<0.0001	1.04 (1.03-1.06)	<0.0001
Sex	Male	698	1.00		1.00
	Female	487	0.76 (0.66-0.89)	0.005	0.70 (0.59-0.82)	<0.0001
Smoking status	Never	115	1.00		1.00
	Current	423	1.69 (1.26-2.27)	0.0004	2.10 (1.55-2.85)	<0.0001
	Former	647	1.67 (1.27-2.20)	0.0003	2.03 (1.52-2.70)	<0.0001
Histology	Adenocarcinoma	577	1.00		1.00
	Squamous cell	285	1.20 (0.99-1.44)	0.059	1.20 (0.99-1.46)	0.071
	others	323	1.32 (1.12-1.57)	0.001	1.41 (1.18-1.69)	0.0002
Tumor stage	I-IIIA	655	1.00		1.00
	IIIB-IV	528	2.84 (2.34-3.45)	<0.0001	3.40 (2.78-4.16)	<0.0001
Chemotherapy	No	639	1.00		1.00
	Yes	538	0.57 (0.48-0.68)	<0.0001	0.54 (0.45-0.65)	<0.0001
Radiotherapy	No	762	1.00		1.00
	Yes	415	0.93 (0.79-1.10)	0.415	0.98 (0.83-1.16)	0.821
Surgery	No	637	1.00		1.00
	Yes	540	0.20 (0.15-0.26)	<0.0001	0.18 (0.14-0.23)	<0.0001
DOCK2 rs261083 G>C	GG/GC/CC	694/416/75	1.14 (1.01-1.27)	0.029	1.20 (1.06-1.36)	0.004
VAV2 rs2519996 C>T	CC/CT/TT	1056/124/5	1.27 (1.03-1.55)	0.024	1.31 (1.05-1.63)	0.019
EPHB1 rs36215 A>G	AA/AG/GG	1030/150/5	0.79 (0.64-0.97)	0.024	0.63 (0.50-0.78)	<0.0001

Abbreviations: SNP: single-nucleotide polymorphisms; NSCLC, non-small cell lung cancer; PLCO, the Prostate, Lung, Colorectal and Ovarian cancer screening trial; HR: hazards ratio; CI: confidence interval.

^aAdjusted for age, sex, tumor stage, histology, smoking status, chemotherapy, radiotherapy, surgery, two other identified SNPs, PC1, PC2, PC3 and PC4.

^bOther 41 previously published SNPs from the same GWAS dataset were also included for further adjustment: rs779901, rs3806116, rs199731120, rs10794069, rs1732793, rs225390, rs3788142, rs73049469, rs35970494, rs225388, rs7553295, rs1279590, rs73534533, rs677844, rs4978754, rs1555195, rs11660748, rs73440898, rs13040574, rs469783, rs36071574, rs7242481, rs1049493, rs1801701, rs35859010, rs1833970, rs254315, rs425904, rs35385129, rs4487030, rs60571065, rs13213007, rs115613985, rs9673682, rs2011404, rs7867814, rs2547235, rs4733124, rs11225211, rs11787670 and rs67715745.

Table 2.

Associations of three significant SNPs with overall survival of patients with NSCLC in both discovery and validation datasets from two previously published GWASs

SNPs	Allelea	Gene	Chr	Position	PLCO (n = 1185)			HLCS (n = 984)			Combined-analysis
					EAF	HR (95% CI)b	P b	EAF	HR (95% CI)c	P c	P het d	I 2	HR (95% CI)e	P e
rs261083	G>C	DOCK2	5	169278751	0.24	1.17 (1.05-1.31)	0.006	0.21	1.16 (1.02-1.31)	0.023	0.879	0.0	1.16 (1.07-1.26)	3.34×10-4
rs2519996	C>T	VAV2	9	136769348	0.06	1.29 (1.05-1.58)	0.016	0.07	1.27 (1.06-1.54)	0.012	0.933	0.0	1.28 (1.12-1.47)	4.57×10-4
rs36215	A>G	EPHB1	3	134627850	0.07	0.76 (0.61-0.94)	0.010	0.05	0.75 (0.59-0.94)	0.015	0.879	0.0	0.75 (0.67-0.83)	5.00×10-4

Abbreviations: SNPs, single-nucleotide polymorphisms; Chr: chromosome; NSCLC, non-small cell lung cancer; GWAS, genome-wide association study; PLCO, the Prostate, Lung, Colorectal and Ovarian cancer screening trial; HLCS, Harvard Lung Cancer Susceptibility Study; EAF, effect allele frequency; HR, hazards ratio; CI, confidence interval; FDR, false discovery rate; BFDP, Bayesian false discovery probability; LD, linkage disequilibrium.

^aEffect/reference allele.

^bAdjusted for age, sex, stage, histology, smoking status, chemotherapy, radiotherapy, surgery, PC1, PC2, PC3 and PC4.

^cAdjusted for age, sex, stage, histology, smoking status, chemotherapy, radiotherapy, surgery, PC1, PC2 and PC3.

^d P _het: P value for heterogeneity by Cochrane’s Q test.

^eMeta-analysis in the fix-effects model.

Specifically, individuals with the EPHB1 rs36215 G allele had a favorable OS and DSS (P _trend = 0.017 and 0.018, respectively, in the trend test), while patients with both DOCK2 rs261083 C and VAV2 rs2519996 T alleles showed an elevated risk of death (P _trend<0.007 and P _trend = 0.016 for OS, respectively, and P _trend = 0.015 and P _trend = 0.048 for DSS, respectively, in the trend test) (Table 3). We also visualized the gene locations in the genome for all the significant SNPs by the Manhattan (Figure S1) and regional association plots (Figure S2).

Table 3.

Associations between the number of unfavorable genotypes of three independent SNPs with OS and DSS of NSCLC in the PLCO Trial

Alleles	Frequencya	OSb			DSSb
		Death (%)	HR (95% CI)	P	Death (%)	HR (95% CI)	P
DOCK2 rs261083 G>C
GG	691	463 (67.00)	1.00		415 (60.06)	1.00
GC	412	277 (67.23)	1.28 (1.10-1.49)	0.001	254 (61.65)	1.32 (1.13-1.55)	0.0006
CC	72	49 (68.06)	1.17 (0.87-1.58)	0.298	40 (55.56)	1.07 (0.77-1.48)	0.699
Trend test				0.007			0.015
Dominant
GG	691	463 (67.00)	1.00		415 (60.06)	1.00
GC+CC	484	326 (67.36)	1.27 (1.09-1.46)	0.002	294 (60.74)	1.28 (1.10-1.49)	0.0016
VAV2 rs2519996 C>T
CC	1049	699 (66.63)	1.00		628 (59.87)	1.00
CT	121	86 (71.07)	1.32 (1.05-1.65)	0.016	78 (64.46)	1.30 (1.03-1.65)	0.031
TT	5	4 (80.00)	1.33 (0.49-3.59)	0.575	3 (60.00)	1.02 (0.33-3.21)	0.971
Trend test				0.016			0.048
Dominant
CC	1049	699 (66.63)	1.00		628 (59.87)	1.00
CT+TT	126	90 (71.43)	1.32 (1.06-1.65)	0.014	81 (64.29)	1.29 (1.02-1.63)	0.034
EPHB1 rs36215 A>G
AA	1020	700 (68.63)	1.00		629 (61.67)	1.00
AG	150	84 (56.00)	0.74 (0.59-0.93)	0.010	76 (50.67)	0.74 (0.58-0.94)	0.013
GG	5	5 (100.00)	0.97 (0.39-2.38)	0.943	4 (80.00)	0.87 (0.32-2.35)	0.776
Trend test				0.017			0.018
Dominant
AA	1020	700 (68.63)	1.00		629 (61.67)	1.00
AG+GG	155	89 (57.42)	0.75 (0.60-0.937)	0.011	80 (51.61)	0.74 (0.59-0.94)	0.013
Change reference genotypes
AG+GG	155	89 (57.42)	1.00		80 (51.61)	1.00
AA	1020	700 (68.63)	1.33 (1.07-1.67)	0.011	629 (61.67)	1.35 (1.06-1.70)	0.013
NUGc
0	85	48 (56.47)	1.00		44 (51.76)	1.00
1	591	395 (66.84)	1.39 (1.03-1.88)	0.032	350 (59.22)	1.36 (0.99-1.87)	0.055
2	458	317 (69.21)	1.78 (1.31-2.41)	0.0002	291 (63.54)	1.80 (1.31-2.48)	0.0003
3	41	29 (70.73)	2.10 (1.32-3.35)	0.0018	24 (58.54)	1.87 (1.13-3.08)	0.015
Trend test				<0.0001			<0.0001
Dichotomized NUG
0-1	676	443 (65.53)	1.00		394 (58.28)	1.00
2-3	499	346 (69.34)	1.35 (1.17-1.56)	<0.0001	315 (63.13)	1.38 (1.19-1.60)	<0.0001

Abbreviations: SNP, single nucleotide polymorphism; NSCLC, non-small cell lung cancer; PLCO, Prostate, Lung, Colorectal and Ovarian cancer screening trial; HR, hazards ratio; CI, confidence interval; OS, overall survival; DSS, disease-specific survival. NUG: number of unfavorable genotypes.

^a10 missing date were excluded.

^bAdjusted for age, sex, smoking status, histology, tumor stage, chemotherapy, surgery, radiotherapy and principal components.

^cUnfavorable genotypes were DOCK2 rs261083 GC+CC, VAV2 rs2519996 CT+TT and EPHB1 rs36215 AA.

Combined and stratified analyses of the three SNPs independently associated with NSCLC OS in the PLCO dataset

To evaluate the collective impact of the above-mentioned three SNPs on NSCLC survival, the respective unfavorable genotypes (i.e., DOCK2 rs261083 GC+CC, VAV2 rs2519996 CT+TT, and EPHB1 rs36215 AA) were included to build a genetic score to categorize the totality of NSCLC patients into four groups based on the number of unfavorable genotypes (NUG). As illustrated in Table 3, multivariable Cox analysis showed that an elevated genetic score independently predicted an elevated risk of death (trend test: P<0.0001 for both OS and DSS). Next, the totality of patients were dichotomized into the low-risk (0-1 NUG) and high-risk groups (2-3 NUGs). In comparison with low-risk cases, high-risk cases showed a significantly elevated risk of death (OS: HR = 1.35, 95% CI = 1.17-1.56 and P<0.0001 and DSS: HR = 1.38, 95% CI = 1.19-1.60 and P<0.0001). We also generated KM survival curves to depict the associations of unfavorable genotypes with survival in NSCLC (Figure 2A-D).

Figure 2.

Kaplan-Meier (KM) survival curves for NSCLC patients based on the combined unfavorable genotypes of the three replicated SNPs in the PLCO trial. A. Based on 0, 1, 2, and 3 unfavorable genotypes in OS. B. Based on 0, 1, 2, and 3 unfavorable genotypes in DSS. C. Dichotomized groups of the unfavorable genotypes divided into 0-1 and 2-3 in OS from the PLCO trial. D. Dichotomized groups of the unfavorable genotypes divided into 0-1 and 2-3 in DSS from the PLCO trial. Abbreviations: OS, overall survival; DSS, disease-specific survival; PLCO, The Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial.

Then, stratification analysis was carried out to assess whether the impact of unfavorable genotypes assessed in combination on NSCLC survival were altered by sex, age, smoking status, tumor stage, histology, radiotherapy, chemotherapy and surgery in the PLCO dataset. Obvious differences in survival or interactions were not observed among the strata of these covariates in OS or DSS in NSCLC (P>0.05, Table S4).

Expression quantitative trait loci (eQTL) analysis

To investigate possible functions of these three SNPs, eQTL was performed to explore correlations of SNP genotypes with the respective mRNA transcript levels. In the TCGA dataset, the DOCK2 rs261083 C allele was significantly correlated with a reduced mRNA transcript levels in LUAD tissue specimens in both additive (P _additive = 0.025, Figure 3A) and dominant (P _dominant = 0.038, Figure 3B) models, but not observed in the 1000 Genomes Project database and LUSC from the TCGA dataset (Figure S3A-F). The EPHB1 rs36215 G allele was significantly correlated with decreased mRNA transcript levels of the gene in the dominant model (P = 0.045, Figure 3C) in the 1000 Gemomes Project database but not in another two models (Figure S3G, S3H) and no correlation between the EPHB1 rs36215 G allele and mRNA amounts was observed in genetic models in LUAD (Figure S3I) and LUSC (Figure S3J-L) from the TCGA dataset. And there was no significant correlation between the VAV2 rs2519996 T allele and mRNA transcript levels in all three genetic models in the 1000 Genomes Project (Figure S3M-O) and the TCGA dataset (Figure S3P-S). Additionally, eQTL analysis of the GTEx Project dataset was carried out, and these genotypes had no significant correlations with the respective mRNA transcript levels in either normal lung tissue specimens (n = 515) or whole blood samples (n = 515) (Figure S4).

Figure 3.

Associations of significant SNP genotypes with their respective mRNA amounts. A. rs261083 additive model in LUAD from the TCGA dataset. B. rs261083 dominant model in LUAD from the TCGA dataset. C. rs36215 dominant model from the 1,000 Genomes Project dataset. Abbreviations: NSCLC, non-small cell lung cancer; PLCO, Prostate, Lung, Colorectal and Ovarian cancer screening trial.

Differential mRNA expression analysis in target tissues

In comparison with paired adjacent noncancerous tissues, DOCK2 expression in tumor tissues samples was markedly lower in LUAD, LUSC and combined LUAD+LUSC samples (all P<0.001; Figure 4A), and elevated DOCK2 transcript levels were also associated with reduced risk of death as depicted by a KM survival curve of lung cancer found in the online data (Figure S5A) (http://kmplot.com/analysis/index.php?p=service&cancer=lung). In contrast, EPHB1 transcript levels were higher in LUAD, LUSC and combined LUAD+LUSC samples (P = 0.001, <0.001 and <0.001, respectively) in comparison with paired adjacent normal lung tissue specimens (Figure 4B). Meanwhile, these elevated transcript levels had no association with elevated risk of death (Figure S5B). Likewise, VAV2 gene expression was remarkably elevated in LUAD and combined LUAD+LUSC samples (P = 0.0013 and 0.001, respectively) but not in LUSC (P = 0.1411; Figure 4C); meanwhile, elevated VAV2 mRNA transcript levels had no association with a high risk of death (Figure S5C).

Figure 4.

DOCK2, EPHB1 and VAV2 mRNA expression in lung cancer tissue and adjacent noncancerous lung tissues specimens in the TCGA dataset. A. Elevated DOCK2 expression was detected in noncancerous tissue samples versus both LUAD tissues and LUSC tissue specimens. B. Elevated EPHB1 expression was found in both LUAD and LUSC tumor tissue specimens in comparison with the noncancerous tissue samples. C. Elevated expression of VAV2 was found in the LUAD tissue samples in comparison with the noncancerous tissue samples. Abbreviations: NSCLC, non-small cell lung cancer; LUAD, lung adenocarcinoma; LUSC, lung squamous cell carcinoma.

Bioinformatics analyses for functional prediction

In silico prediction of the functions of the above-mentioned three significant SNPs was carried out with online bioinformatics tools (i.e., SNPinfo, RegulomeDB, and HaploReg). We found that a DOCK2 rs261083 G>A change might alter protein motifs, while VAV2 rs2519996 C>T and EPHB1 rs36215 A>G changes might potentially affect enhancer histone marks (Table S5). Additionally, according to experimental data in the Encyclopedia of DNA Elements (ENCODE) project, no potential functions for these independent SNPs were predicted (Figure S6).

Discussion

In the current study, we demonstrated that DOCK2 rs261083 G>C, VAV2 rs2519996 C>T, and EPHB1 rs36215 A>G in the NK cell-related gene-set significantly predicted OS and DSS in Caucasian NSCLC patients. Additionally, the DOCK2 rs261083 C allele affected DOCK2 mRNA transcript levels in LUAD specimens in the TCGA dataset, while the EPHB1 rs36215 G allele was significantly correlated with mRNA transcript levels in lymphoblastoid cells from 373 European individuals in the 1000 Genomes Project.

Considering gene expression data in the TCGA database, DOCK2 might represent a suppressor gene, because reduced DOCK2 mRNA transcript levels were correlated with an elevated risk of death in patients with both LUAD and LUSC. However, the rs261083 C allele appeared to affect DOCK2 mRNA expression levels in LUAD but not LUSC. This might result from the small number of tumor specimens available for analysis; alternatively, there might be distinct molecular mechanisms and genetic heterogeneity between these subtypes of NSCLC [26,27]. On the other hand, both EPHB1 and VAV2 seemed to have some oncogenic features, because higher mRNA transcript levels were associated with an elevated risks of death in NSCLC cases. Moreover, the rs36215 G allele predicted a reduced risk of death and low EPHB1 mRNA transcript levels in normal lymphoblastic cell lines, although the rs2519996 T allele did not predict VAV2 gene expression. Jointly, these findings suggest that the observed associations of genetic variants in the NK cell-associated genes with NSCLC survival are biologically plausible.

DOCK2, also called dedicator of cytokinesis2 and located on chromosome 5, is critical for lymphocyte migration and regulates T cell responsiveness [28]. Functionally, DOCK2 regulates not only the differentiation of NKs, plasmacytoid dendritic cells and T helper 2 cells [29] but also cell motility, polarity, adhesion, proliferation and apoptosis by activating Rac [30]. DOCK2 is also likely involved in the carcinogenesis. For instance, a report showed DOCK2 overexpression is associated with good acute myeloid leukemia (AML) prognosis [31]. Additionally, DOCK2 overexpression was reported to be correlated with a favorable prognosis in colorectal cancer with overtly enhanced CD8+ lymphocytes infiltration [32]. The above findings suggest that DOCK2 might be a potential tumor suppressor, because DOCK2 was suppressed in tumor tissues in comparison with benign tissues and that elevated DOCK2 gene expression predicted good prognosis in LUAD cases.

EPHB1 (Eph receptor B1), an Eph family member, modulates not only adult angiogenesis but also embryonic vascular system development [33]. A previous study revealed that high EPHB1 transcript levels in NSCLC tissue specimens in comparison with adjacent noncancerous normal lung tissue specimens; Furthermore, EPHB1 overexpression induces migration and invasion and is also associated with reduced survival of lung cancer patient [34]. These findings indicate that EPHB1 may have an oncogenic effect in line with the above finding that the EPHB1 transcript levels were elevated in lung cancer tissue specimens in comparison with paired noncancerous lung tissue specimens from the same patients. However, no study has reported associations of genetic variants of EPHB1 with NSCLC survival. As shown above, the EPHB1 rs36215 variant G allele predicted a reduced risk of death in NSCLC cases, compared with individuals with the wildtype allele, possibly because the EPHB1 rs36215 G allele decreased EPHB1 gene expression. Since OS was comparable between the higher and lower expression of EPHB1 groups of NSCLC cases, additional molecular mechanisms might control EPHB1 expression in cancer, which needs further investigation.

VAV2, (vav guanine nucleotide exchange factor 2), located on chromosome 9, is one of the essential regulators of immune function [35]. The newly described VAV2 has abnormal expression in several malignancies [36-38], and the VAV2-Rac1 pathway is involved in cancer development [39]. Reports assessing VAV2 in NSCLC are scarce. A report suggested VAV2 modulates vimentin-associated FAK activation and controls lung cancer cell adhesion in vitro [40]; Another study showed that miR-331-3p suppresses epithelial-to-mesenchymal transition (EMT), migration and metastatic potential by interacting with ErbB2 and VAV2 via the Rac1/PAK1/β-catenin signaling in NSCLC [41]. Furthermore, host Vav2 deficiency decreases microvascular density and tumor growth and/or survival, in Lewis lung carcinoma models [42]. These findings suggest that VAV2 might have an oncogenic function in the tumor microenvironment in the NSCLC biology, corroborating the above-mentioned finding of VAV2 overexpression in lung cancer specimens in comparison with paired noncancerous tissue samples form the same patients. However, no reports have assessed the roles of VAV2 genetic variants in NSCLC prognosis. As shown above, the VAV2 rs2519996 variant T allele was not found to affect VAV2 mRNA transcript levels, although it predicted an elevated risk of death in NSCLC, in comparison with the wildtype. Furthermore, VAV2 mRNA amounts were markedly increased in lung cancer tissue specimens compared with noncancerous counterparts, although VAV2 upregulation was not correlated with reduced survival. Thus, the molecular mechanisms of VAV2 in predicting survival of NSCLC patients need to be further investigated.

The limitations of the current study should be mentioned. First, only Caucasian patients were included in the two available GWAS datasets, which may have reduced the generalizability of the findings to other ethnic groups. Secondly, the exact molecular mechanisms underpinning the above-mentioned associations of these independent SNPs with NSCLC survival remain unknown and should be further explored. Thirdly, no detailed information about the therapies administered to the patients was available in the PLCO trial, which makes it impossible to be adjusted, such as immunotherapies, in the analysis. Finally, despite the relatively large sample size of the PLCO trial, the numbers of patients in various subgroups remain relatively small, likely reducing the statistical power in the stratified or subgroup analyses.

Overall, we demonstrated that three novel independent functional SNPs (DOCK2 rs261083 G>C, VAV2 rs2519996 C>T, and EPHB1 rs36215 A>G) were associated with survival in NSCLC cases in both the PLCO trial and the HLCS study. We also showed that DOCK2 rs261083 C likely has an effect on survival in NSCLC cases, possibly by altering the targeted gene expression. Our findings indicate that these three SNPs might represent new prognostic biomarkers of NSCLC survival, once further validated in additional lung cancer patient populations and upon further mechanistic assessment.

Disclosure of conflict of interest

None.

Abbreviations

SNP

single nucleotide polymorphism

NK

natural killer

NSCLC

Non-small cell lung cancer

LUAD

lung adenocarcinoma

LUSC

lung squamous cell carcinoma

GWAS

Genome-Wide Association Study

PLCO

the Prostate, Lung, Colorectal and Ovarian Cancer Screening Trial

HLCS

Harvard Lung Cancer Susceptibility

OS

overall survival

DSS

disease-special survival

eQTL

expression quantitative trait loci

TCGA

The Cancer Genome Atlas

HR

hazards ratio

CI

confidence interval

GTEx

genotype-tissue expression project

NUG

number of unfavorable genotypes

DOCK2

Dedicator of cytokinesis 2

EPHB1

Erythropoietin-producing hepatocellular carcinoma (Eph) Receptor B1

VAV2

vav guanine nucleotide exchange factor 2