Potentially Functional Variants of HBEGF and ITPR3 in GnRH Signaling Pathway Genes Predict Survival of Non-Small Cell Lung Cancer Patients

Abstract

The gonadotropin-releasing hormone (GnRH) signaling pathway controls reproductive functions and cancer growth and progression. However, few studies investigated roles of genetic variants of GnRH pathway genes in survival of patients with non-small cell lung cancer (NSCLC). Therefore, we first evaluated associations between 22,528 single-nucleotide polymorphisms (SNPs) in 101 GnRH pathway genes and survival of 1,185 NSCLC patients using a dataset from Prostate, Lung, Colorectal and Ovarian (PLCO) Cancer Screening Trial. We found 572 SNPs to be significantly associated with overall survival (OS) of NSCLC (P ≤ 0.05, Bayesian false discovery probability ≤ 0.80). We then validated these SNPs in another dataset from 984 NSCLC patients in Harvard Lung Cancer Susceptibility (HLCS) Study. Finally, two independent SNPs (HBEGF rs4150236G>A and ITPR3 rs116454384C>T) remained significantly associated with NSCLC OS with a combined hazards ratio of 0.84 (95% confidence interval=0.76–0.92, P=0.0003) and 0.85 (0.78–0.94, 0.0012), respectively, and their genetic score (the number of protective genotypes) was associated with better OS and disease-specific survival (DSS) (P_trend=0.0002 and 0.0001, respectively). Further expression quantitative trail loci analysis showed significant correlations between ITPR3 rs116454384T genotypes, and higher mRNA expression levels in both whole blood and normal lung, and high ITPR3 mRNA expression levels in tumors was associated with better survival of NSCLC patients. Because ITPR3 mutations were rare in tumors, ITPR3 rs116454384C>T likely had an effect on cancer progression by regulating gene expression. Therefore, genetic variants of HBEGF rs4150236G>A and ITPR3 rs116454384C>T may be predictors for NSCLC survival, but HBEGF rs4150236G>A functional relevance remains to be determined.

Introduction

Lung cancer is the leading cause of cancer-related mortality in both men and women, with more than one million deaths each year worldwide¹. In the USA in 2020, it is estimated that there were approximately 228,820 new cases diagnosed with and 135,720 deaths from lung cancer². Non-small cell lung cancer (NSCLC) is the primary histological type, accounting for approximately 85% of all lung cancer patients. Although there have been advances in the treatment of NSCLC patients, including surgery, chemo-radiotherapy, molecular targeted therapy and immunotherapy, the 5-year overall survival (OS) rate of NSCLC remains only 18.1% in the United States³. Therefore, the discovery of suitable biomarkers would help improve early diagnosis and predict clinical outcomes as well as personalizing therapy of patients with NSCLC. It is well known that clinical characteristics, including age, sex, smoking status, histology, stage, and treatment options, are the main factors to influence lung cancer survival⁴. Moreover, increasing evidence suggests that genetic variations in critical genes play an important role in the tumorigenesis and progression of NSCLC^4–6. Indeed, recently, by using the pathway-based analytic approaches, several novel and biologically functional variants of the cancer-related pathway genes have been identified to be associated with lung cancer survival^4,7,8.

Accumulating evidence has demonstrated that the gonadotropin-releasing hormone (GnRH) signaling pathway not only is a key regulator of the reproductive system⁹, but also plays an important role in the control of tumorigenesis and progression in human cancers, including both reproductive and non-reproductive cancers^9–11. It is well known that GnRH triggers the synthesis and release of luteinizing hormone (LH) and follicular stimulating hormone (FSH) by pituitary gonadotropes and plays a central role in the regulation of gonadial development¹². GnRH specifically binds to its receptor (GnRH-R), which is an incipient key node in the GnRH signaling pathway and activates various intracellular mechanisms, having effects on cellular function by the GTP-binding protein-coupled receptor (GPCR) signaling¹³. It was reported that GnRH and GnRH-R were expressed in several types of cancer tissues, including NSCLC, indicating that the expression of GnRH may be associated with tumor progression¹⁴. Although the effect of GnRH on tumor progress is controversial, some studies revealed that GnRH has strong anti-proliferation and anti-metastasis properties in human cancers and is considered as a promising candidate for novel molecular-targeted strategies for the treatment of cancers¹⁰.

In addition, GnRH also activates and regulates multiple signaling pathways, such as the pathways of JNK/AP-1, calcineurin/NFAT, cAMP/PKA/CREB and mitogen-activated protein kinase (MAPK), all of which may play important roles in tumorigenesis and tumor progression in humans^10,15–17. For instance, GnRH is coupled to Gαq/11 G proteins to activate phospholipase C, which catalyzes the hydrolysis of phosphatidylinositol 4,5-bisphosphate to inositol 1,4,5-trisphosphate (IP3) and diacylglycerol (DAG)¹⁸; IP3 can stimulate the release of intracellular calcium and activate conventional protein kinase C (PKC) isoforms, and the inositol 1,4,5-triphosphate receptor type 3 (IP3R3 or ITPR3), a modulator for diverse cellular functions, is responsive to the binding of IP3, whereas the generation of DAG can lead to the activation of novel PKC isoforms^19,20. Furthermore, activation of PKC could result in transactivation of the EGF and MAPK signaling pathway^9,18. It is also demonstrated that GnRH causes the activation of phosphor tyrosine phosphatase (PTP), which leads to the de-phosphorylation of activated EGF-R and inhibition of EGF-R signal transduction as well as inhibition of the heparin binding-epidermal growth factor (HBEGF) signaling in ERα-negative breast cancer cells¹⁰. Taken together, these data suggest that aberrant activation of the GnRH pathway has a significant impact on tumorigenesis or tumor progression.

Therefore, it is likely that genetic variation, including SNPs, in some key genes in the GnRH signaling pathway may be involved in the disorder or over-activation of the entire GnRH signaling pathway, modulating tumor growth and progression, but such genetic effects and their biological functions remain largely unknown. Therefore, in the present study, we investigated associations between potentially functional genetic variants in the GnRH signaling pathway genes and survival of NSCLC patients in a two-stage analysis of genotyping datasets extracted from two previously published genome-wide association studies (GWASs).

Materials and methods

Study populations

In the discovery stage, we obtained a genotyping dataset from the Prostate, Lung, Colorectal and Ovarian (PLCO) Cancer Screening Trial, which enrolled nearly 155,000 participants aged 55–74 from ten centers across the United States between 1993 and 2001²¹. Among all the participants, 1,185 Caucasian NSCLC patients with complete personal information including age, sex, smoking status, histology, clinical stage, treatment options, follow-up information and genotyping data were available for survival analysis. We used OS as the primary endpoint and also examined disease-specific survival (DSS). The follow-up time was defined from the diagnosis of NSCLC to the last follow-up or the time of death. In the PLCO trial, we extracted genomic DNA samples from the blood and genotyped with Illumina HumanHap240Sv1.0, HumanHap300v1.1 and HumanHap550v3.0 (dbGaP accession: phs000093.v2.p2 and phs000336.v1.p1)^22,23. To expand the genotyping data, we performed imputation with IMPUTE2 according to the CEU data from the 1000 Genomes Project (phase 1 release V3).

In the validation stage, we used another genotyping dataset that includes 984 histologically confirmed Caucasian NSCLC patients from the GWAS dataset of the Harvard Lung Cancer Susceptibility (HLCS) study²⁴, in which genomic DNA samples extracted from the patients’ blood were genotyped with Illumina Humanhap610-Quad arrays, and the genotyping data were also imputed by using MaCH1.0 based on the 1000 Genomes Project. Details of the patients from the HLCS study have also been described elsewhere²⁴.

The use of these two GWAS datasets was approved by both the Internal Review Board of Duke University School of Medicine (#Pro00054575) and the dbGAP database administration (#6404). The comparison of the characteristics between the PLCO trial (n=1,185) and the HLCS study (n=984) is presented in Supplementary Table 1.

Gene and SNP selection

We searched the GnRH pathway genes by using the Molecular Signatures Database with the keyword “GnRH” (http://software.broadinstitute.org/gsea/msigdb/index.jsp) and included 101 genes as the candidates for further analysis (Supplementary Table 2). SNPs within these genes and their ± 500-kb flanking regions were selected by the following quality control criteria: (1) a genotyping rate ≥ 95%, (2) a minor allelic frequency (MAF) ≥ 0.05, (3) Hardy-Weinberg equilibrium (HWE) P value ≥ 1×10⁻⁵, and (4) an imputation info score ³ 0.8. As a result, 2,378 genotyped SNPs were selected from the PLCO GWAS dataset and additional 20,150 SNPs were imputed (Supplementary Figure 1).

Statistical analysis

In the discovery stage, we performed multivariate Cox proportional hazards regression analysis in an additive genetic model with adjustments for age, sex, smoking status, histology, tumor stage, chemotherapy, radiotherapy, and surgery as well as the first four principal components identified from the GWAS dataset. We estimated associations between SNPs in the GnRH pathway genes and NSCLC survival by calculating hazards ratio (HR) and its 95% confidence interval (CI) with the GenABEL package of R software²⁵. For multiple testing correction, the false discovery rate (FDR) with a cut-off value of 0.200 was first used to assess the probability of false positives²⁶. Since the majority of SNPs were imputed with a high level of linkage disequilibrium (LD), we also used Bayesian false discovery probability (BFDP) with a cut-off value of 0.80 for multiple testing correction to reduce the probability of false positive findings as recommended²⁷. In the LD analysis, we selected representative SNPs in the identified important genes in high LD (r²>0.8) and functional SNPs according to functional annotation based on RegulomeDB (http://www.regulomedb.org/) and HaploReg v4.1 (https://pubs.broadinstitute.org/mammals/haploreg/haploreg.php). We assigned a prior probability of 0.10 to detect an HR of 3.0 for an association with variant genotypes or minor alleles of the SNPs with P ≤ 0.05. In the validation stage, we used Cox regression analysis with adjustment for age, sex, smoking status, histology, tumor stage, chemotherapy, radiotherapy, surgery and the first three principal components to validate the findings from the discovery stage.

Next, we performed a meta-analysis to combine the results of both discovery and validation datasets by using PLINK 1.07, for which Cochran’s Q-test and the heterogeneity statistic (I²) were performed to assess the inter-study heterogeneity. If no heterogeneity was found between the two studies (Q-test P-value > 0.10 and I² < 50.0%), a fixed-effects model was implemented; otherwise, a random-effects model was applied. Pairwise LD was also estimated by using the data from 373 European individuals in the 1000 Genomes Project. To further identify novel and independent SNPs among the validated SNPs, we constructed a multivariate stepwise Cox model that included the first four principal components of the PLCO dataset and 23 SNPs previously published, in addition to the adjustment for available demographic and clinical variables. We also used the combined genotypes to evaluate the cumulative effects of the identified significant SNPs and the Kaplan-Meier curves to depict survival probability associated with their genotypes.

In the stratified analysis, we performed the heterogeneity test of associations among subgroups of each clinical characteristic by using the Chi-square-based Q-test, with P < 0.05 considered statistically significant for differences among the subgroups of each clinical characteristic. All statistical analyses were performed with SAS software (version 9.4; SAS Institute, Cary, NC, USA), if not specified otherwise. To illustrate the prediction accuracy of the model integrating clinical and genetic variables on NSCLC survival²⁸, we performed the receiver operating characteristic (ROC) curve and time-dependent area under the curve (AUC) with ROC-time package of R software (version 3.5.0). We also generated the LD map and haplotype blocks by Haploview software19, the Manhattan plot with the −log10 (P_adj) for all SNPs that passed QC, and the regional association plots by using Locus Zoom (http://http://locuszoom.sph.umich.edu).

Finally, we performed expression quantitative trait loci (eQTL) analysis²⁹ to evaluate correlations between genotypes of SNPs and mRNA expression levels of their corresponding genes by using RNA-sequencing data from lymphoblastoid cells derived from the same 373 individuals of European descent in the 1000 Genomes Project, and 369 whole blood samples and 383 normal lung tissue included in the Genotype-Tissue Expression (GTEx Analysis V7, dbGaP Accession phs000424.v7.p2) project^30,31. We examined the differences in mRNA expression levels in 111 pairs of lung cancer tissues and adjacent normal tissues from The Cancer Genome Atlas (TCGA) dataset by using a paired Student’s t-test. We also assessed the differences in mRNA expression levels in a larger, but not paired, dataset from TCGA (http://ualcan.path.uab.edu), and performed Kaplan-Meier survival analysis to assess the associations between mRNA expression levels of the important genes and survival probability (http://kmplot.com) of lung cancer patients. To assess the mutation rates of those identified important genes in lung tumor tissues, we also used the publicly available database of the cBioPortal for cancer Genomics (http://www.cbioportal.org). All statistical analyses were performed with the SAS software (version 9.4; SAS Institute, Cary, NC, USA) unless specified otherwise.

Results

Associations of SNPs in the GnRH signaling pathway genes with survival of NSCLC

Basic characteristics of the discovery dataset with 1,185 NSCLC patients from the PLCO trial and the validation dataset of 984 NSCLC patients from the HLCS study have been described elsewhere⁴. As shown in the working flowchart (Figure 1), we first used a single-locus multivariate Cox regression analysis in the discovery dataset to evaluate associations between 22,528 SNPs of GnRH signaling pathway genes and NSCLC OS with adjustment for age, sex, smoking status, histology, tumor stage, chemotherapy, radiotherapy, surgery and the first four principal components (Supplementary Table 3). After multiple testing correction by FDR first and then by BFDP, we identified 572 SNPs to be significantly associated with NSCLC OS (P < 0.05, BFDP ≤ 0.8). All these significant SNPs were further validated by the HLCS GWAS dataset, and finally, seven SNPs remained significantly associated with NSCLC OS. Further combined analysis of the two datasets for these seven SNPs showed their associations with a better NSCLC OS without heterogeneity. The details of associations between these seven SNPs and NSCLC OS are described in Table 1.

Figure 1. The flowchart of the present study.

Abbreviations: SNP, single- nucleotide polymorphism; PLCO, Prostate, Lung, Colorectal and Ovarian cancer screening trial; NSCLC, non- small cell lung cancer; HBEGF, heparin binding-epidermal growth factor; ITPR3, inositol 1,4,5-triphosphate receptor type 3.

Table 1.

Associations of seven validated significant SNPs with overall survival in both discovery and validation datasets from two previously published NSCLC GWAS datasets

SNP	Allelea	Gene	PLCO (n=1185)					HLCS (n=984)			Combined-analysis
			FDRa	BFDPa	MAF	HR (95% CI)b	Pb	MAF	HR (95% CI)c	Pc	Phetd	I2	HR (95% CI)e	Pe
rs2237078f	G>C	HBEGF	0.555	0.735	0.22	0.85 (0.74–0.96)	0.010	0.21	0.86 (0.75–0.99)	0.033	0.904	0	0.85 (0.78–0.94)	0.0010
rs4150230f	G>A	HBEGF	0.547	0.747	0.22	0.84 (0.74–0.96)	0.009	0.21	0.84 (0.73–0.96)	0.013	1.000	0	0.84 (0.76–0.92)	0.0003
rs4150232f	G>A	HBEGF	0.547	0.747	0.22	0.84 (0.74–0.96)	0.009	0.21	0.83 (0.72–0.96)	0.012	0.904	0	0.84 (0.76–0.92)	0.0002
rs4150234f	G>A	HBEGF	0.547	0.747	0.22	0.84 (0.74–0.96)	0.008	0.21	0.84 (0.73–0.96)	0.011	1.000	0	0.84 (0.76–0.92)	0.0003
rs4150236f	G>A	HBEGF	0.546	0.747	0.22	0.84 (0.74–0.96)	0.008	0.21	0.84 (0.73–0.96)	0.012	1.000	0	0.84 (0.76–0.92)	0.0003
rs13385f	G>A	HBEGF	0.556	0.735	0.22	0.85 (0.75–0.96)	0.011	0.21	0.86 (0.75–0.99)	0.030	0.902	0	0.85 (0.78–0.94)	0.0007
rs116454384	C>T	ITPR3	0.556	0.747	0.17	0.84 (0.74–0.96)	0.012	0.19	0.87 (0.75–1.00)	0.049	0.723	0	0.85 (0.78–0.94)	0.0012

Abbreviations: SNP, single nucleotide polymorphism; NSCLC, non-small cell lung cancer; GWAS, genome-wide association study; PLCO, Prostate, Lung, Colorectal and Ovarian cancer screening trial; HLCS: Harvard Lung Cancer Susceptibility; MAF, minor allele frequency; HR, hazards ratio; CI, confidence interval; FDR: false discovery rate; BFDP: Bayesian false discovery probability

^aFDR and BFDP were available in the PLCO dataset because the HLCS study provided only the summary data

^bObtained from an additive genetic model with adjustment for age, sex, stage, histology, smoking status, chemotherapy, radiotherapy, surgery, PC1, PC2, PC3, and PC4;

^cObtained from an additive genetic model with adjustment for age, sex, stage, histology, smoking status, chemotherapy, radiotherapy, surgery, PC1, PC2, and PC3;

^dP_het: P value for heterogeneity by Cochrane’s Q test;

^eMeta-analysis in the fixed-effects model.

^fSNPs rs2237078, rs4150230, rs4150232, rs4150234, rs13385 are high LD with rs4150236.

In further LD analysis of these seven replicated SNPs, except for rs116454384 in ITPR3, other six SNPs in HBEGF were in high LD with each other (all r² > 0.8) (Supplementary Figure 3a) by in silico SNP functional prediction (SNPinfo, RegulomeDB and HaploReg 4.1). In particular, we observed that the SNPs of rs4150236 in HBEGF and rs116454384 in ITPR3 are located in the enhancer-like H3K4me1 and H3K27ac in lung tissue or A549 EtOH 0.02pct lung carcinoma cell line, which is predicted to have putative regulatory sites, such as the enhancer histone modification (Supplemental Table 4). Thus, we selected these two representative SNPs (i.e. rs4150236 in HBEGF and rs116454384 in ITPR3) for further analyses and summarized the results in the Manhattan plot (Supplementary Figure 2) and the regional association plot (Supplementary Figure 3b and 3c). To further identify independent SNPs associated with NSCLC survival, we performed analysis with the multivariate stepwise Cox regression model, including the first four principal components of the PLCO dataset. When the two validated SNPs (rs4150236 and rs116454384) were added to the model with adjustment for the 23 previously published significant SNPs in the same PLCO GWAS dataset, both SNPs remained significantly and independently associated with NSCLC OS (Table 2).

Table 2.

Two independent SNPs in multivariate Cox hazards regression analysis with adjustment for other covariates and previous published SNPs in the PLCO Trial GWAS dataset

Variables	Category	Frequency	HR (95% CI)a	Pa	HR (95% CI)b	Pb
Age	Continuous	1185	1.03 (1.02–1.05)	<0.0001	1.04 (1.02–1.05)	<0.0001
Sex	Male	698	1.00		1.00
	Female	487	0.81 (0.69–0.94)	0.005	0.79 (0.68–0.93)	0.004
Smoking status	Never	115	1.00		1.00
	Current	647	1.64 (1.25–2.16)	0.0004	1.90 (1.43–2.54)	<0.0001
	Former	423	1.71 (1.27–2.29)	0.0003	1.96 (1.45–2.66)	<0.0001
Histology	AD	577	1.00		1.00
	SC	285	1.20 (0.99–1.44)	0.061	1.25 (1.03–1.51)	0.026
	Others	323	1.29 (1.08–1.53)	0.004	1.33 (1.11–1.59)	0.002
Stage	I-IIIA	655	1.00		1.00
	IIIB-IV	528	2.82 (2.32–3.42)	<0.0001	3.00 (2.46–3.66)	<0.0001
Chemotherapy	No	639	1.00		1.00
	Yes	538	0.58 (0.49–0.69)	<0.0001	0.58 (0.48–0.70)	<0.0001
Radiotherapy	No	762	1.00		1.00
	Yes	415	0.92 (0.78–1.09)	0.335	0.94 (0.79–1.11)	0.448
Surgery	No	637	1.00		1.00
	Yes	540	0.21 (0.16–0.27)	<0.0001	0.19 (0.15–0.25)	<0.0001
HBEGF rs4150236	GG/GA/AA	719/409/56	0.84 (0.74–0.95)	0.007	0.85 (0.74–0.97)	0.014
ITPR3 rs116454384	CC/CT/TT	807/334/37	0.85 (0.75–0.98)	0.021	0.85 (0.74–0.97)	0.020

Abbreviations: HR: hazards ratio; CI: confidence interval; SNP: single-nucleotide polymorphisms.

^aStepwise analysis included age, sex, smoking status, tumor stage, histology, chemotherapy, radiotherapy, surgery, PC1, PC2, PC3, PC4 and SNPs.

^b23 published SNPs were used for post-stepwise adjustment: rs779901, rs3806116, rs199731120, rs10794069, rs1732793, rs225390, rs3788142, rs73049469, rs35970494, rs225388, rs7553295, rs1279590, rs73534533, rs677844, rs4978754, rs1555195, rs11660748, rs73440898, rs13040574, rs469783, rs36071574, rs7242481, rs1049493.

As shown in Table 3 for the 1,185 NSCLC patients in the PLCO dataset with complete adjustment, patients with the HBEGF rs4150236 A allele or ITPR3 rs116454384 T allele had a reduced risk of death (rs4150236: P_trend = 0.008 for OS and P_trend = 0.004 for DSS; rs116454384: P_trend = 0.012 for OS and P_trend = 0.011 for DSS). More specifically, compared with their wild genotypes in a dominant genetic model, patients with HBEGF rs4150236 GA/AA or ITPR3 rs116454384 CT/TT genotypes had a significantly reduced risk of death (HBEGF rs4150236: HR = 0.80, 95% CI = 0.69–0.92 and P = 0.002 for OS; 0.79, 0.67–0.92 and 0.003 for DSS; ITPR3 rs116454384: 0.84, 0.72–0.98 and 0.025 for OS; 0.82, 0.70–0.97 and 0.018 for DSS) (Table 3 and Figure 2).

Table 3.

Associations between two significantly independent SNPs and survival of NSCLC patients in the PLCO Trial

Genotype	No. of patients	OS			DSS
		Death (%)	Multivariate analysisa		Death (%)	Multivariate analysisa
			HR (95% CI)	P		HR (95% CI)	P
HBEGF
rs4150236 G>Ab	1175
GG	713	500 (70.1)	1.00		453 (63.5)	1.00
GA	406	255 (62.8)	0.79 (0.67–0.92)	0.002	229 (56.4)	0.79 (0.67–0.93)	0.004
AA	56	34 (60.7)	0.88 (0.62–1.26)	0.492	27 (48.2)	0.78 (0.53–1.16)	0.217
Ptrend test				0.008			0.004
GA/AA	462	289 (62.6)	0.80 (0.69–0.92)	0.002	256 (55.4)	0.79 (0.67–0.92)	0.003
ITPR3
rs116454384 C>Tb	1175
CC	804	550 (68.4)	1.00		499 (62.1)	1.00
CT	333	220 (66.1)	0.86 (0.74–1.01)	0.072	192 (57.7)	0.84 (0.71–1.00)	0.048
TT	38	19 (50.0)	0.64 (0.40–1.01)	0.055	18 (47.4)	0.66 (0.41–1.05)	0.082
Ptrend test				0.012			0.011
CT/TT	371	239 (64.4)	0.84 (0.72–0.98)	0.025	210 (56.6)	0.82 (0.70–0.97)	0.018
Number of protective genotypes (NPGs)b,c
0	489	354 (72.4)	1.00		324 (66.3)	1.00
1	539	342 (63.5)	0.76 (0.65–0.88)	0.0003	304 (56.4)	0.75 (0.64–0.88)	0.0005
2	147	93 (63.3)	0.71 (0.56–0.89)	0.004	81 (55.1)	0.68 (0.53–0.87)	0.003
Ptrend test				0.0002			0.0001
0	489	354 (72.4)	1.00		324 (66.3)	1.00
1–2	686	435 (63.4)	0.75 (0.65–0.86)	<0.0001	385 (56.1)	0.74 (0.63–0.86)	<0.0001

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

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

^b10 missing date were excluded.

^cProtective genotypes were HBEGF rs4150236 GA/AA and ITPR3 rs116454384 CT/TT.

Figure 2.

Prediction of survival with genotypes of HBEGF rs4150236, ITPR3 rs116454384 and combined protective genotypes.

(a) Kaplan–Meier survival curves with additive model for the overall survival of HBEGF rs4150236 genotypes and (b) dominate model; (c) Kaplan–Meier survival curves with additive model for the disease- specific survival of HBEGF rs4150236 genotypes and (d) dominate model; (e) Kaplan–Meier survival curves with additive model for the overall survival of ITPR3 rs116454384 genotypes and (f) dominate model; (g) Kaplan–Meier survival curves with additive model for the disease- specific survival of ITPR3 rs116454384 genotypes and (h) dominate model; (i) Kaplan–Meier survival curves for the overall survival of the combined protective genotypes and (j) dichotomized groups of the NPG in the PLCO dataset; (k) Kaplan–Meier survival curves for the disease- specific survival of the combined protective genotypes and (l) dichotomized groups of the NPG in the PLCO dataset.

Abbreviations: NPG, number of protective genotypes; PLCO, The Prostate, Lung, Colorectal and Ovarian Cancer Screening Trial.

Combined effects of the two independent SNPs in the PLCO dataset

To provide a better estimation of the hazards of survival, we combined the protective genotypes (i.e., rs4150236 GA/AA and rs116454384 CT/TT) into a genetic score as the number of protective genotypes (NPGs), which divided all NSCLC patients into three groups: zero, one, and two NPGs. As shown in Table 3, an increased NPG was associated with better survival after adjustment for other covariates (P_trend = 0.0002 and 0.0001 for OS and DSS, respectively). To dichotomize for better survival analysis, we re-grouped all the patients into a low-protective-genotypes group (0 NPGs) and a high-protective-genotypes group (1–2 NPGs). Compared with the 0 NPGs, 1–2 group NPGs were associated with significantly better survival (adjHR = 0.75, 95% CI = 0.65–0.86, P < 0.0001 for both OS and DSS), which were further depicted in Kaplan-Meier survival curves (Figure 2).

Stratified analysis for associations between NPGs and NSCLC survival

We further performed the stratified analysis to evaluate the possible modification effect of protective genotypes on survival of NSCLC by age, sex, smoking status, histology, tumor stage, chemotherapy, radiotherapy and surgery in the PLCO dataset. As a result, compared with patients with 0 NPGs, patients with 1–2 NPGs exhibited a significantly better survival for OS and DSS in subgroups of males (P = 0.0006 and 0.0005, respectively); age < 71 or ≥ 71 (0.012 or 0.020 for OS, and 0.023 or 0.008 for DSS); former or current smokers (0.010 or 0.034 for OS and 0.012 or 0.024 for DSS); adenocarcinoma or squamous cell carcinoma (0.013 or 0.004 for OS and 0.033 or 0.0004 for DSS); I-IIIA or IIIB-IV tumor stage (0.028 or 0.011 for OS and 0.008 or 0.023 for DSS); no or received chemotherapy (0.019 or 0.003 for OS and 0.014 or 0.004 for DSS); no or received radiotherapy (0.004 or 0.005 for OS and 0.007 or 0.005 for DSS); and without or with surgery (0.014 or 0.011 for OS and 0.016 or 0.006 for DSS). However, no significant interaction was found between genotypes/genetic score and other covariates on NSCLC OS and DSS (P_inter > 0.05 for all, Supplementary Table 5).

The ROC curves and time-dependent AUC

We further assessed predictive values of the two SNPs with time-dependent AUC and ROC curves at the 12^th, 24^th, and 60^th month (5-year) in the PLCO dataset (because we only had the summary genotyping data from the HLCS study). Compared with the model for all covariates including age, sex, smoking status, histology, tumor stage, chemotherapy, radiotherapy, surgery and first four principal components, the time-dependent AUC plot with addition of the independent SNPs did not improve prediction performance of the model at the 12^th (1^st year) and 60^th month (5^th year). That is, the AUCs for OS changed from 85.73% to 85.84% (P = 0.600) for the 1^st year and from 88.59% to 88.75% (P = 0.512) for the 5^th year (Supplementary Figure 4b and 4d, respectively); the AUCs for DSS changed from 86.07% to 86.26% (P = 0.425) for the 1^st year and from 88.54% to 88.76% (P = 0.451) for the 5^th years) (Supplementary Figure 4f and 4h, respectively). However, the AUC and ROC curves at the 24^th month (or the 2^nd year) suggested that the prediction performance of the model was improved significantly: the AUCs changed from 86.65% to 87.12% (P = 0.016) for OS and from 86.61% to 87.25% (P = 0.033) for DSS (Supplementary Figure 4c and 4g, respectively).

The eQTL analysis

To further explore potential functions of the two independent SNPs, we performed the eQTL analysis to identify the correlations between genotypes of the SNPs (HBEGF rs4150236 and ITPR3 rs116454384) and mRNA expression levels of their corresponding genes by using genomic data from 373 lymphoblastoid cell lines derived from individuals of European descent in the 1000 Genomes Project and 369 whole blood and 283 lung normal tissues in the GTEx project. We found that the rs4150236 variant AA genotype was borderline significantly correlated with an increased expression level of HBEGF mRNA (P = 0.052, Supplementary Figure 5a), compared with the GG/GA genotypes in the recessive model in the 1000 Genomes Project, while this was not the case in the additive and dominant models, nor in whole blood data (Figure 3a) and lung normal tissues (Figure 3b) of the GTEx project. For the ITPR3 rs116454384 variant T allele, there was also no significant difference in the 1000 Genomes Project (Supplementary Figure 6a). However, in the whole blood data and lung normal tissues of the GTEx project, the variant rs116454384 T allele was associated with higher expression levels of ITPR3 mRNA (in additive model, P = 0.9×10⁻⁵ for 369 whole blood individuals and P = 0.010 for 383 lung normal tissues) (Figure 3d and 3e, respectively).

Figure 3. Correlations between genotypes of the significant SNPs and their corresponding mRNA expression levels.

For GTEx project, The HBEGF rs4150236 A allele was no associated with mRNA expression levels of HBEGF in (a) normal lung tissue and (b) whole blood; (c) higher expression levels of HBEGF were associated with a better survival in patients with lung cancer (http://kmplot.com/analysis/index.php?p=service&cancer=lung); (d) The ITPR3 rs116454384 T allele was associated with higher mRNA expression levels in normal lung tissue and (e) whole blood from the GTEx project; (f) higher expression levels of ITPR3 were associated with a better survival in patients with lung cancer (http://kmplot.com/analysis/index.php?p=service&cancer=lung).

Differential mRNA expression analysis and survival of NSCLC

To find molecular mechanisms of the HBEGF and ITPR3 genes in the progression and survival of NSCLC, we first assessed mRNA expression levels of the two genes in 111 pairs of lung cancer tissues including 60 lung adenocarcinoma (LUAD) and 51 lung squamous cell carcinoma (LUSC) and adjacent normal tissue samples in NSCLC obtained from the TCGA database. As shown in Supplementary Figure 5b, compared with adjacent normal tissues, tumor tissues had a lower mRNA expression level of HBEGF in 60 LUAD and 51 LUSC (all P = 0.001). Meanwhile, the mRNA expression levels of ITPR3 were not significantly different in the pairs of LUAD and LUSC NSCLC tissues (P = 0.862) (Supplementary Figure 6b).

We then compared mRNA expression levels of these two genes in 59 adjacent normal lung tissues, 515 LUAD tissue samples, and 503 LUSC obtained from The Cancer Genome Atlas (TCGA) database (ualcan.path.uab.edu/home). As shown in Supplementary Figure 5c, the mRNA expression levels of HBEGF were all significantly lower in LUSC and LUAD than that in normal lung tissues (P = 1.69×10⁻¹² for LUAD and P = 1.64×10⁻⁹ for LUSC), while the mRNA expression levels of ITPR3 were significantly higher in LUAD (P = 1.11×10⁻¹⁰) and non-significantly lower in LUSC than that in normal lung tissues (P = 0.302) (Supplementary Figure 6c).

Additionally, we also evaluated the correlation between mRNA expression levels of these two genes and OS of NSCLC patients from the TCGA dataset (www.kmplot.com) and found that the patients with high mRNA expression levels of ITPR3 had a better NSCLC OS (HR = 0.72; 95% CI = 0.64–0.82; Log-rank P = 6.3×10⁻⁷, Figure 3f). However, the impact of mRNA expression levels of HBEGF on OS was not statistically significant in the TCGA database (HR = 1.09; 95% CI = 0.96–1.24, Log-rank P = 0.16, Figure 3.c). Taken all the data together, we do not have sufficient evidence to judge whether these genes are oncogenes.

Mutation analysis

Because frequently mutated genes in tumor tissues would have a much greater impact on patients’ survival than SNPs in the same genes, we further investigated the mutation status of HBEGF and ITPR3 in lung tumor tissues by using the public database of the cBioPortal for Cancer Genomics. As shown in Supplemental Figure 5d, HBEGF had a much low somatic mutation rate in different NSCLC datasets (0.56% in TCGA pub, 0.21% in PanCan and 0.09% in TCGA 2016), while ITPR3 also displayed a low mutation rate in different NSCLC datasets (6.25% in MSKCC, 4.11% in TCGA 2016 and 1.33% in MSKCC 2018, Supplemental Figure 6d). These low mutation frequencies in both HBEGF and ITPR3 unlikely had a significant effect on the expression levels of these two genes in NSCLC tumors. Therefore, the roles of SNPs in HBEGF and ITPR3 in regulating gene expression and NSCLC survival need further research.

Discussion

In the present study, we investigated associations between 22,528 genetic variants of 101 genes in the GnRH signaling pathway and NSCLC survival using genotyping data 1,185 NSCLC patients from the PLCO trial and another 984 NSCLC patients from the HLCS study. We found that HBEGF rs4150236A and ITPR3 rs116454384T variant genetypes were significantly associated with a better NSCLC survival in US Caucasian populations. We also provided biological evidence that the protective ITPR3 rs116454384 variant T allele, but not HBEGF rs4150236 variant A allele, was associated with high mRNA expression levels of ITPR3 in both whole blood cells and normal lung tissues from individuals of European descent. Furthermore, patients with high mRNA expression levels of HBEGF and ITPR3 had a better NSCLC OS. These data imply that the HBEGF rs4150236 variant A or ITPR3 rs116454384 variant T alleles may play a role in survival of NSCLC patients, possibly by modulating the mRNA expression levels of their related genes, particularly for ITPR3 rs116454384 variant T allele. These provide further support for biological plausibility for the observed SNP-survival associations.

HBEGF is located on chromosome 5q31.3, contains six exons and encodes HBEGF, a 19–23 kDa protein of 208 amino acids³². As a member of the EGF family, HBEGF was initially found in human macrophages and identified as a protective cytokine on different target cells in the intestine^33–35. Subsequently, the expression of HBEGF was detected in several human cancers. For example, one study found that HBEGF mRNA was expressed in seven human glioma cell lines with expression levels of two- to five-fold higher than that of normal brain tissues in eight of 11 glioblastoma patients³⁶, while another study reported that HBEGF enhanced the growth of human pancreatic cancer cells in an allelic dose-dependent manner³⁷. In a study of 108 Japanese patients, investigators found that HBEGF expression to be significantly increased in advanced ovarian cancer, compared with that in normal ovaries, and that the higher expression levels of HBEGF were significantly associated with poor clinical outcomes^38,39.

For lung cancer, one study reported that HBEGF was highly expressed in a subset of lung cancer patients, in which proliferation was dependent on the HBEGF signaling and that silencing of HBEGF with RNA interference suppressed cell growth, leading to G1/S cell cycle arrest in HBEGF-positive lung cancer cells, which supports an oncogenic role of HBEGF in lung cancer cells⁴⁰. The same study also found that the expression of HBEGF correlated with EGFR expression in primary lung tumors was associated with a poor survival of 287 Taiwanese lung cancer patients, suggesting the involvement of the EGFR/HBEGF signaling in lung tumor progression⁴⁰. Although it was not clear whether the HBEGF rs4150236G>A had an effect of its gene expression, we observed that mRNA expression levels of HBEGF were higher in normal lung tissues than in tumor tissues and that the higher HBEGF mRNA expression was associated with a better survival in over 1,000 NSCLC patients in the TCGA database. In particular, such observations are also consistent with our findings, in that NSCLC patients with the HBEGF rs4150236 variant A allele had a better survival. However, we did not have the evidence for an association between HBEGF rs4150236 variant A allele and mRNA expression levels of the gene, this SNP was associated with considerable levels of monomethylation of lysine 4 (H3K4Me1) and histone modification of H3K27 acetylation (H327Ac) enrichment, which may have some effects on gene activation and expression⁴¹. Taken together, the exact molecular mechanisms of HBEGF in lung cancer need to be further explored.

ITPR3 encodes a receptor for the inositol 1,4,5-trisphosphate, a second messenger that mediates the release of intracellular calcium, that plays a key role in exocrine secretion underlying the energy metabolism and cell growth, and ITPR3 also plays a critical role in regulating cellular proliferation, activation and apoptosis of cancer cells^20,42. However, the functions of ITPR3 in tumor progression, for lung cancer in particular, remain unclear⁴³. One study reported that FBXL2 knockdown caused accumulation of ITPR3, which is a major player in Ca²⁺⁻dependent apoptosis, causing an increase in Ca²⁺mediated apoptosis in lung cancer A549 cell lines and limiting tumor growth⁴⁴. In contrast, one recently published study observed that overexpression of ITPR3 could result in reduced apoptosis in breast and colon cancer cells^20,42, while knockdown or silencing of ITPR3 enhanced apoptosis in breast, colon and renal cancer cells^20,42,43.

In the present study, we found that a better survival of NSCLC was associated with the ITPR3 rs116454384T allele that was also associated with significantly higher mRNA expression levels of ITPR3 in whole blood and normal lung tissues, which in turn were associated with better survival in NSCLC. Furthermore, ITPR3 rs116454384 is located in the intron region of ITPR3 with considerable levels of H3K4Me1 and H3K27ac enrichment, a possible mechanism by which this SNP regulates gene expression⁴¹. Therefore, the roles of ITPR3 and related SNPs in tmor progression and the mechanisms involved in survival of NSCLC patients remain to be determined.

To our knowledge, the present study is the first that focuses on associations of potentially functional genetic variants in the GnRH pathway genes with survival of NSCLC patients by a two-stage analysis of two previously published GWAS datasets. However, there were some limitations. First, because both available GWAS datasets came from populations of European descendants, it is uncertain whether these results are generalizable to other ethnic populations. Second, detailed genotype information and clinical outcomes data of the HLCS study were not accessible for us to do additional combined modeling and stratified analysis. Third, because of no direct evidence of biological experiments, the biological mechanisms by which the identified SNPs may influence tumor progression and thus survival of NSCLC needs to be further explored. Finally, our results need to be validated in larger patient populations.

Abbreviations:

GnRH

Gonadotropin releasing hormone

NSCLC

non-small cell lung cancer

OS

overall survival

DSS

disease specific survival

GWAS

genome-wide association studies

SNP

single-nucleotide polymorphisms

PLCO

Prostate, Lung, Colorectal and Ovarian Cancer Screening Trial

HLCS

Harvard Lung Cancer Susceptibility

NCI

National Cancer Institute

MsigDB

Molecular Signatures Database

KEGG

Kyoto Encyclopedia of Genes and Genomes

HR

hazards ratio

CI

confidence interval

MAF

minor allelic frequency

HWE

Hardy-Weinberg equilibrium

FDR

false discovery rate

BFDP

Bayesian false discovery probability

LD

linkage disequilibrium

eQTL

expression quantitative trait loci

TCGA

the Cancer Genome Atlas

AUC

area under the receiver operating characteristic curve

ROC

receiver operating characteristic curve

HBEGF

heparin binding-epidermal growth factor

ITPR3

inositol 1,4,5-triphosphate receptor type 3