Novel genetic variants in HDAC2 and PPARGC1A genes of the CREB-binding protein pathway predict survival of non-small cell lung cancer

Abstract

The CREB-binding protein (CBP) pathway plays an important role in transcription and activity of acetyltransferase that acetylates lysine residues of histones and non-histone proteins. In the present study, we hypothesized that genetic variants in the CBP pathway genes played a role in survival (OS) of non-small cell lung cancer (NSCLC). We tested this hypothesis using the genotyping data from the genome-wide association study (GWAS) dataset from the Prostate, Lung, Colorectal and Ovarian (PLCO) Cancer Screening Trial. In the single locus analysis, we evaluated associations between 13 176 (1107 genotyped and 12 069 imputed) single-nucleotide polymorphisms (SNPs) in 72 genes and survival of 1185 NSCLC patients. The identified 106 significant SNPs in the discovery were further validated in additional genotyping data from another GWAS dataset of 984 NSCLC patients in the Harvard Lung Cancer Susceptibility Study. The combined results of two datasets showed that two independent, potentially functional SNPs (i.e., HDAC2 rs13213007G>A and PPARGC1A rs60571065T>A) were significantly associated with NSCLC overall survival, with a combined hazards ratio (HR) of 1.26 [95% confidence interval (CI) = 1.09–1.45 and P=0.002] and 1.23 (1.04–1.47 and P=0.017), respectively. Furthermore, we performed an expression quantitative trait loci analysis and found that the survival-associated HDAC2 rs13213007A allele (GA+AA), but not PPARGC1A rs60571065A allele (TA+AA), was significantly associated with increased mRNA expression levels of HDAC2 in 373 lymphoblastoid cell lines. These results indicate that HDAC2 rs13213007A allele is a potential predictor of NSCLC survival, likely by altering HDAC2 expression.

1. INTRODUCTION

Lung cancer has been the leading cause of cancer-related mortality since 1985, with over a million deaths each year worldwide ^[1]. In 2018, it is estimated that there were approximately 234,030 new cases and 154,050 lung cancer deaths in the United States (https://seer.cancer.gov/statfacts/html/lungb.html). Non-small cell lung cancer (NSCLC) is the most common histological type of lung cancer, accounting for approximately 80–85% of all lung cancer diagnosis ^[2]. Over the past few decades, advances in surgical procedures, chemo-radiotherapy, targeted therapy, and more recently immunotherapy have contributed to modest improvements in the survival of lung cancer ^[3]; yet, the five-year survival rate is still at an underwhelming 15–20%. Previous studies have found that molecular and genetic factors may play an important role in lung cancer progression and outcomes ^[4]; thus, it is important to search for additional molecular markers that may predict the survival of NSCLC patients who would benefits from individualized therapies.

Single nucleotide polymorphisms (SNPs) can affect gene expression and functions and thus may be associated with susceptibility to cancer development and progression ^[5]. To date, several genome-wide association studies (GWASs) of lung cancer have discovered SNPs that are associated with cancer risk, but few have found markers predictive of outcomes of NSCLC patients. The vast majority of SNPs under GWAS investigation did not reach the stringent genome-wide level of significance, and most of those identified as significant SNPs are located within non-coding regions of the genome, making such discoveries fundamentally difficult to comprehend and harder to reveal the underlying mechanisms. As a result, it is recommended that research strategies in the post-GWAS era should include “discovery, expansion, and replication”, “biological studies”, and “epidemiologic studies” (https://epi.grants.cancer.gov/gameon/#funded) ^[6]. By using a hypothesis-driven approach of the pathway gene analysis pooling together GWAS datasets, one may have a better chance with an improved statistical power to identify novel loci with minor yet detectable effects, to examine functional consequences of such novel loci, and thus, to unravel possible mechanisms underlying the observed associations.

The CREB-binding protein (CBP) pathway regulates the post-translational modification with the activity of acetyltransferase that acetylates lysine residues of histones and non-histone proteins, such as p53 ^[7]. The CBP pathway is also involved in basic cellular functions, including cell growth, differentiation, DNA repair, and apoptosis ^{[8, 9]}. Studies have revealed oncogenic roles of the CBP pathway genes; for example, it has been shown that CBP is highly expressed in lung cancer cells and tumor tissues, upregulating hTERT expression and promoting tumor growth in human lung cancer cells ^[10]. Furthermore, CBP overexpression by members of the activator protein-1 (AP-1) family and downregulation of the retinoid acid receptor β might promote lung tumor proliferation ^[11]. Intriguingly, CBP exerts its actions mainly by acetylation of histones and other regulatory proteins ^[12], and could be fused to MOZ and several other gene products in acute myeloid leukemia by chromosomal translocations ^[13]. In addition, mice monoallelic for the CBP gene may induce multilineage defects in hematopoietic differentiation with an elevated chance of hematologic malignancies ^[14].

To date, the role of SNPs in the CBP pathway genes and their functionality related to tumor growth and progression are still unknown, and there are few reports about the SNPs in CBP pathway genes and its related gene-set in NSCLC prognosis. In the present study, by using two publicly available GWAS datasets, we performed a CBP pathway gene-set analysis to evaluate the association between genetic variants in this gene-set and the survival of NSCLC patients.

2. MATERIALS AND METHODS

2.1. Study populations

As shown in the study flowchart (Figure 1), the discovery phase included 1,185 NSCLC patients obtained from the Prostate, Lung, Colorectal and Ovarian (PLCO) Cancer Screening Trial, after application and access approval from the National Cancer Institute (NCI). The PLCO is an NCI funded multicenter randomized trial of screening for cancer from ten medical centers in the United States between 1993 and 2011 ^[15]. The screening trial enrolled 77,500 men and 77,500 women aged 55–74. All individuals were randomized to either the intervention arm with screening or the control arm with standard care ^[15]. The PLCO trial collected blood specimens from the first screening visit and gathered extensive information about each individual, including smoking history, family history of cancer and demographic information ^[16]. All participants were followed for at least 13 years after the enrollment. Genomic DNA extracted from the blood samples was genotyped in a genome-wide association study (GWAS) with Illumina HumanHap240Sv1.0, Human-Hap300v1.1 and HumanHap550v3.0 (dbGaP accession: phs000093.v2.p2 and phs000336.v1.p1) ^{[17, 18]}. In 1,187 Caucasian NSCLC patients from the PLCO trial, two with missing follow-up information were excluded. Therefore, the eligible subsets of the PLCO lung cancer dataset for survival analysis included 1185 NSCLC patients, whose clinicopathological variables and genotype data were available. Tumor staging was determined according to the fifth edition American Joint Committee on Cancer staging system. The institutional review boards of each participating institution approved the PLCO trial and the use of biological specimens for further research, and all subjects signed a written informed consent permitting the research represented here ^[15].

FIGURE 1.

Study workflow chart. BFDP, Bayesian false-discovery probability; eQTL, expression quantitative trait loci; Harvard Study, Harvard Lung Cancer Susceptibility Study; PLCO, Prostate, Lung, Colorectal and Ovarian cancer trial; ROC, receiver operating characteristic; SNP, single-nucleotide polymorphism

The validation phase used the GWAS dataset from the Harvard Lung Cancer Susceptibility (HLCS) Study with 984 histology-confirmed Caucasian NSCLC patients. The histological classification of the tumors were recorded by two staff pulmonary pathologists at the Massachusetts General Hospital. The time of blood collection was within 1–4 weeks of the diagnosis for each patient. DNA was extracted from blood samples by using the Auto Pure Large Sample Nucleic Acid Purification System (QIAGEN Company, Venlo, Limburg, Netherlands). Genotype data were obtained by using Illumina Humanhap610-Quad arrays, and imputation was performed by using MaCH based on the 1000 Genomes project ^[19]. Details of the participants in the Harvard study were described previously ^[20]. The comparison of the characteristics between the PLCO and the Harvard study is presented in the Supplemental Table 1.

2.2. Gene and SNP selection

The genes or gene-set involved in the CBP pathway were selected through the Molecular Signatures Database (http://software.broadinstitute.org/gsea/msigdb/index.jsp), by the keyword “CBP”. After removing duplicated genes and excluding genes in the X chromosome, 72 genes remained as candidate genes for further analysis (Supplemental Table 2). We first performed imputation for the 72 genes plus the 500-kb flanking buffer regions by using IMPUTE2 and the 1000 Genomes Project data (phase 3) ^[19]. After imputation, we extracted all the SNPs in these genes and within their ±2 kb flanking regions according to the following criteria: a minor allele frequency ≥ 0.05, a genotyping rate ≥ 95%, and a Hardy-Weinberg equilibrium P value ≥ 1×10⁻⁵. As a result, 1107 genotyped SNPs were chosen from the PLCO GWAS dataset with an additional 12,069 SNPs as a result of imputation.

2.3. Statistical analyses

The follow-up time in both PLCO and HLCS datasets was from the diagnosis of lung cancer to the last follow-up or time of death. OS was the primary endpoint of the present study, and disease-specific survival (DSS) of lung cancer was also examined. In the single-locus analysis, multivariate Cox proportional hazards regression analysis was used to evaluate associations between each of the SNPs and survival (in an additive genetic model) with adjustment for age, sex, smoking status, histology, tumor stage, chemotherapy, radiotherapy, surgery and the top four principal components of the PLCO dataset using the GenABEL package of R software ^[21]. We first used the false discovery rate (FDR) to correct multiple testing with a threshold of 0.2.

Since the majority of SNPs were imputed and there was a high level of linkage disequilibrium (LD), we used the Bayesian false discovery probability (BFDP) with a cutoff value of 0.8 for multiple test corrections as recommended ^[22]. We assigned a prior probability of 0.10 to detect an HR of 2.0 for an association with variant genotypes or minor alleles of the SNPs with P < 0.05. Then, we chose SNPs for the HLCS GWAS dataset validation, which satisfied the following conditions: SNPs passed the threshold of BFDP ≤ 0.8, and tagging SNPs based on their LD. To identify independent SNPs, we included the validated SNPs in a multivariate stepwise Cox model with adjustment for demographic characteristics, previously published SNPs, clinical variables and the top four principal components of the genotyping data in the PLCO dataset. Combined-analysis of discovery and validation datasets was also performed to provide summary results. The fixed-effects model was applied If the Cochran’s Q-test P value > 0.100 and the heterogeneity statistic (I²) < 50%; otherwise, the random-effects model was employed. Kaplan-Meier curves were used to estimate the survival associated with the genotypes. The combination of risk or unfavorable genotypes was used to estimate the cumulative effects of the identified SNPs.

Functions of the validated SNPs were further predicted by HaploReg²³ , SNPinfo²⁴ and RegulomeDB²⁵. The criteria for functional SNPs were: 1) the SNPs were associated with survival in both the PLCO trial and HLCS study and 2) genotypes of the SNPs were associated with mRNA expression of their genes as shown in one of these databases. Expression quantitative trait loci (eQTL) analysis was further performed to assess correlations between SNPs and mRNA expression levels by using linear regression analysis with the R (version 3.5.0) software. mRNA expression data of genes were obtained from lymphoblastoid cell lines derived from the 373 European descendants included in the 1000 Genomes Project¹⁹, and from the whole blood and normal lung tissues in the genotype-tissue expression (GTEx) project²⁶. Using the data from the Cancer Genome Atlas (TCGA) database (dbGaP Study Accession: phs000178.v9.p8), we examined the differences in mRNA expression levels between paired tumor tissues and adjacent normal tissues by the paired t test²⁷. Next, we detected the association of mRNA expression and OS through Kaplan-Meier (KM) analysis (n=1,926) (http://kmplot.com/analysis/index.php?p=service&cancer=lung). Finally, receiver operating characteristic (ROC) curves were constructed and time-dependent ROC analysis were performed to examine the prediction accuracy of models integrating both clinical and genetic variables on NSCLC survival with the “timeROC” package in R (version 3.5.0)²⁸. Unless specified, all statistical analyses were performed using the SAS software (version 9.4; SAS Institute, Cary, NC, USA).

3. RESULTS

3.1. Associations between SNPs in the CBP pathway gene-set and NSCLC survival in both PLCO and HLCS datasets

The study flowchart is shown in Figure 1, and basic characteristics of 1,185 NSCLC patients have been described previously²⁹ (also see Supplemental Table 3). In the PLCO discovery dataset with an additive genetic model, the multivariate Cox model with adjustment for age, sex, smoking status, histology, tumor stage, chemotherapy, radiotherapy, surgery and the first four principal components (Supplemental Table 4) identified 405 SNPs that were significantly associated with NSCLC survival after multiple test correction by BFDP ≤ 0.8 but not by False Discovery Rate ≤ 0.2 – the results are summarized in a Manhattan plot (Figure 2A). Potentially functional SNPs were further validated by the HLCS dataset. As a result, two SNPs in two different genes (i.e., rs13213007 in HDAC2 and rs60571065 in PPARGC1A) were validated (Table 1), both of which were imputed. Further combined analysis of these SNPs of the two datasets showed a worse survival associated with the rs13213007 A and rs60571065 A alleles (P_adjusted = 0.0003 and 0.002, respectively), and no heterogeneity between the two studies were observed (Table 1).

FIGURE 2.

Functional and survival-associated SNPs. A. Manhattan plot of 13,176 SNPs of the CBP Pathway genes in the PLCO trial. The statistical values across the autosomes for associations between 13,176 SNPs and overall survival of patients with NSCLC are plotted as −log10 P values. The blue horizontal line indicates P = 0.05 and the red line indicates BFDP = 0.80. Abbreviations: NSCLC, non-small cell lung cancer; PLCO, Prostate, Lung, Colorectal and Ovarian cancer screening trial; BFDP, Bayesian false-discovery probability. B. Kaplan-Meier (KM) survival curves for NSCLC patients of two validated SNPs and combined unfavorable genotypes in the PLCO trial. (a) by 0, 1 and 2 unfavorable genotypes (log-rank test for trend: P) in OS, (b) by 0 and 1–2 unfavorable genotypes (log-rank test and multivariate analysis: P) in OS from the PLCO trial, (c) by 0, 1 and 2 unfavorable genotypes (log-rank test for trend: P) in DSS, and (d) by 0 and 1–2 unfavorable genotypes (log-rank test and multivariate analysis: P) in DSS from the PLCO trial. Abbreviations: NSCLC, non-small cell lung cancer; PLCO, Prostate, Lung, Colorectal and Ovarian cancer screening trial; OS, overall survival; DSS, disease-specific survival. C. eQTL analysis of HDAC2 rs13213007 and PPARGC1A rs60571065 genotypes and corresponding gene mRNA expression. All the data were from the 1000 Genome Project dataset. (a) rs13213007 additive model (P=0.043); (b) rs13213007 dominant model (P=0.080); and (c) rs13213007 recessive model (P=0.122) but no significant results for rs60571065 (P=0.556, 0.743 and 0.339, figures not shown). Abbreviations: eQTL, expression quantitative trait loci.

TABLE 1.

Combined analysis of the two unfavorable SNPs in PLCO and HLCS datasets

SNP	Allelea	Gene	PLCO (n=1,185)				HLCS (n=984)			Combined analysis
			EAF	HR (95% CI)b	Pb	BFDP	EAF	HR (95% CI)c	Pc	HR (95% CI)d	Pd	Phete	I 2
rs13213007	GG>A	HDAC2	0.26	1.18 (1.05–1.32)	0.004	0.58	0.29	1.14 (1.02–1.29)	0.022	1.16 (1.07–1.26)	0.0003	0.758	0
rs60571065	TT>A	PPARGC1A	0.11	1.22 (1.04–1.43)	0.015	0.77	0.10	1.21 (1.00–1.46)	0.050	1.22 (1.08–1.37)	0.002	0.948	0

Abbreviations: PLCO, Prostate, Lung, Colorectal and Ovarian Cancer Screening Trial; HLCS, Harvard Lung Cancer Susceptibility Study

^aMajor > minor 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;

^din fixed-effect model;

^eP value for heterogeneity by Cochrane’s Q test.

3.2. Identification of independent SNPs associated with survival of NSCLC in the PLCO dataset

Because the HLCS study provided only the summary data, we used only the PLCO dataset to identify independent SNPs after full adjustment for other co-variables. To identify potential functional SNPs associated with NSCLC survival with three online bioinformatics tools (i.e., RegulomeDB²⁵, SNPinfo²⁴ and HaploReg²³), we found that both of the validated SNPs were located in the intron regions with some functions. In the RegulomeDB²⁵, rs60571065 had a score of 5, indicating a potential transcriptional factor binding or DNase peak, while rs13213007 had no data (Supplemental Table 5); both of the validated SNPs had no obvious function based on SNPinfo²⁴ or HaploReg²³.

The two validated SNPs (i.e., HDAC2 rs13213007 and PPARGC1A rs60571065) were also included into a multivariate stepwise Cox model with adjustments for demographic and clinical variables as well as previously published SNPs and the first four principal components available in the PLCO dataset. As a result, both of the two SNPs remained independently associated with NSCLC survival (Table 2, also see Supplemental Table 6); a regional association plot of each SNP is shown in Supplemental Figure 1. In the PLCO dataset, patients with rs13213007 A or rs60571065 A alleles had an increased risk of death (P_trend = 0.004 or 0.015, Table 3). Compared with the reference genotype in a dominant genetic model, HDAC2 rs13213007 GA+AA and PPARGC1A rs60571065 TA+AA were associated with a significantly increased risk of death (HR=1.26, 95% CI=1.09–1.45 and P=0.002 for rs13213007 GA+AA; and HR=1.23, 95% CI=1.04–1.47 and P=0.017 for rs60571065 TA+AA).

TABLE 2.

Predictors of OS obtained from stepwise Cox regression analysis in the PLCO dataset

Variablesa	Category	Frequencyb	HR (95% CI)	P
Age	Continuous	1,174	1.04 (1.02–1.05)	<0.001
Sex	Male	694	1.00
	Female	480	0.82 (0.70–0.95)	0.010
Smoking status	Never	114	1.00
	Current	416	1.78(1.32–2.39)	<0.001
	Former	644	1.78(1.35–2.35)	<0.001
Histology	Adenocarcinoma	575	1.00
	Squamous cell	283	1.29 (1.07–1.56)	0.008
	others	316	1.34 (1.13–1.59)	<0.001
Tumor stage	I-IIIA	654	1.00
	IIIB-IV	520	2.97 (2.45–3.61)	<0.001
Chemotherapy	No	638	1.00
	Yes	536	0.56 (0.46–0.66)	<0.001
Radiotherapy	No	760	1.00
	Yes	414	0.96 (0.86–1.13)	0.617
Surgery	No	634	1.00
	Yes	540	0.19 (0.15–0.25)	<0.001
HDAC2 rs13213007 G>Ac	GG/GA/AA	648/450/76	1.16 (1.03–1.30)	0.012
PPARGC1A rs60571065 T>Ac	TT/TA/AA	925/231/18	1.21 (1.04–1.42)	0.017

Abbreviations: CI = Confidence interval; HR = Hazards ratio; PLCO = Prostate, Lung, Colorectal and Ovarian cancer trial; OS = Overall survival; SNP, Single Nucleotide Polymorphism

^aStepwise analysis included age, sex, smoking status, tumor stage, tumor histology, chemotherapy, radiotherapy, surgery, top four principal components and eight published SNPs and two new validated SNPs (HDAC2 rs13213007 and PPARGC1A rs60571065 in an additive model);

^bTen observations missing of clinical variables (two of tumor stage and eight of chemotherapy/ radiotherapy/ surgery); one missing for TLE1 rs199731120 C>CA (published SNP); 1,174 patients remained for the stepwise analysis;

^cTwo SNPs has been identified in the present study.

TABLE 3.

Associations of the two validated SNPs in the CBP pathway with OS of NSCLC in the PLCO dataset

	OS Univariate analysis				OS Multivariable analysisa
	Frequency				Frequency
Genotype	(n=1,185)	Death (%)	HR (95% CI)	P	(n=1,175)	Death (%)	HR (95% CI)	P
HDAC2 rs13213007 G>A
GG	656	431 (65.7)	1.00		648	424 (65.4)	1.00
GA	453	311 (68.7)	1.19 (1.03–1.38)	0.018	451	309 (68.5)	1.26(1.09–1.47)	0.002
AA	76	56 (73.7)	1.28(0.97–1.70)	0.079	76	56 (73.7)	1.24 (0.93–1.64)	0.145
Trend				0.008				0.004
GA+AA	529	367 (69.4)	1.21(1.05–1.39)	0.008	527	365 (69.3)	1.26 (1.09–1.45)	0.002
PPARGC1A rs60571065 T>A
TT	942	629 (66.8)	1.00		934	621 (66.5)	1.00
TA	225	159(70.7)	1.11 (0.93–1.32)	0.235	223	158 (70.9)	1.22 (1.03–1.46)	0.025
AA	18	10 (55.6)	0.78 (0.42–1.46)	0.441	18	10 (55.6)	1.42 (0.76–2.67)	0.276
Trend				0.555				0.015
TA+AA	243	169 (69.6)	1.08 (0.91–1.29)	0.351	241	168(69.7)	1.23 (1.04–1.47)	0.017
NUGb
0	522	340	1.00		516	334	1.00
1	554	380 (68.6)	1.17 (1.01–1.35)	0.039	550	377 (68.6)	1.28 (1.10–1.48)	0.002
2	109	78 (71.6)	1.32 (1.03–1.69)	0.027	109	78 (71.6)	1.50 (1.17–1.93)	0.001
Trend				0.008				<0.001
0	522	340 (65.1)	1.00		516	334 (64.7)	1.00
1–2	663	458 (69.1)	1.19 (1.03–1.37)	0.015	659	455 (69.0)	1.31 (1.14–1.52)	0.0002
Phenotype missingc	N/A				10

Abbreviations: SNPs = Single nucleotide polymorphisms; OS = Overall survival; NSCLC = Non-small cell lung cancer; PLCO = Prostate, Lung, Colorectal and Ovarian cancer trial; NPG = Number of protective genotypes; HR = Hazards ratio; 95%CI = 95%Confidence interval; N/A = Not applicable.

^aMultivariate Cox hazards regression analyses were adjusted for age, sex, smoking, stage, histology, chemotherapy, radiotherapy, surgery, and top four principal components;

^bNUG=number of unfavorable genotypes; NUG were HDAC2 rs11606640 GA+AA and PPARGC1A rs62053220 TA+AA;

^cTwo observations missing of tumor stage and eight observations missing of chemotherapy/radiotherapy/surgery in PLCO dataset.

3.3. Combined and stratified analysis of the two independent and functional SNPs in the PLCO dataset

To provide a better estimation of the hazards for survival, we combined the risk genotypes (i.e., rs13213007 GA+AA and rs60571065 TA+AA) into a genetic score, which was used to divide all NSCLC patients into three groups. As shown in Table 3, in the multivariate analysis, an increased genetic score was associated with worse survival (trend test: P<0.001). After dichotomizing the genetic score, we re-grouped all the patients into a low-risk group (0 risk score) and a high-risk group (1–2 risk scores); patients in the high-risk group had significantly poorer survival compared with the low-risk group (HR=1.31, 95% CI=1.14–1.52 and P=0.0002). Kaplan-Meier survival curves were generated to depict the associations between risk genotypes and NSCLC OS (Figure 2Ba and 2Bb) and DSS (Figure 2Bc and 2Bd).

To assess the ability of risk genotypes to predict outcomes, we compared the area under the ROC curve (AUC) from the model with clinical variables to that from the model including both clinical variables and risk genotypes. The addition of risk genotypes to the prediction model of five-year OS increased the AUC from 87.0% to 87.27% (P=0.375, Supplemental Figure 2a); similarly, the addition of risk genotypes to the prediction model of five-year DSS increased the AUC from 86.7% to 87.01% (P=0.391, Supplemental Figure 2b). Finally, the time-dependent AUC curve was generated to quantify the ability of risk genotypes to predict NSCLC survival through the entire follow-up period (Supplemental Figure 2c and 2d).

We further analyzed the effect of combined risk genotypes on NSCLC OS, and whether it was modified by age, sex, smoking status, histology, tumor stage, chemotherapy, radiotherapy and surgery (Table 4). The results showed no significant interactions (P>0.05) except with smoking status (P=0.002). Hence, as shown in Supplemental Tables 7 and 8, we further assessed the associations of HDAC2 rs13213007 and PPARGC1A rs60571065 with OS of NSCLC in three subgroups of never, current and former smokers and found that cigarette smoking and HDAC2 rs13213007 had an interactive effect on NSCLC OS, where the highest risk was observed amongst current smokers. Additionally, never smokers were at a higher risk than former smokers.

TABLE 4.

Stratified multivariate analyses for association between unfavorable genotypes and OS in NSCLC patients in the PLCO trial

Characteristics	No. of 0 risk genotypes		No. of 1–2 risk genotypes		Multivariate analysisa		Pinterb
	All	Death (%)	All	Death (%)	HR (95% CI)	P
Age (years)
≤ 71	277	168 (60.7)	357	251 (64.7)	1.31 (1.07–1.61)	0.009
> 71	239	166 (69.5)	302	224 (74.2)	1.35 (1.10–1.65)	0.005	0.697
Sex
Male	273	190 (69.6)	422	315 (74.6)	1.33 (1.11–1.60)	0.002
Female	243	144 (59.3)	237	140 (59.1)	1.32 (1.03–1.69)	0.029	0.906
Smoking status
Never	57	28 (49.1)	57	34 (59.7)	1.44 (0.81–2.57)	0.212
Current	182	110 (60.4)	235	156 (66.4)	1.72 (1.33–2.23)	<0.001
Former	277	196 (70.8)	367	265 (72.2)	1.08 (0.90–1.31)	0.409	0.002*
Histology
Adenocarcinoma	259	149 (57.5)	316	197 (62.3)	1.36 (1.09–1.70)	0.007
Squamous	129	87(67.4)	155	104 (67.1)	1.32 (0.98–1.79)	0.069
Others	128	98 (76.6)	188	154 (81.9)	1.43 (1.10–1.87)	0.009	0.780
Tumor stage
I-IIB	287	133 (46.3)	367	181 (49.3)	1.23 (0.98–1.55)	0.077
IIIA-IV	229	201 (87.8)	292	274 (93.8)	1.36 (1.12–1.64)	0.002	0.524
Chemotherapy
No	272	147 (54.0)	366	219 (59.8)	1.33 (1.07–1.65)	0.009
Yes	244	187 (76.6)	293	236 (80.6)	1.22 (1.00–1.50)	0.049	0.847
Radiotherapy
No	323	179 (55.4)	438	270 (61.6)	1.35 (1.11–1.64)	0.002
Yes	193	155 (80.3)	221	185 (83.7)	1.25 (1.00–1.56)	0.055	0.397
Surgery
No	283	242 (85.5)	352	323 (91.8)	1.34 (1.13–1.59)	0.0008
Yes	233	92 (39.5)	307	132 (43.0)	1.33 (1.01–1.76)	0.041	0.771
Phenotype missingc	10

Abbreviations: CI = Confidence interval; HR = Hazards ratio; NSCLC = Non-small cell lung cancer; OS = Overall survival; PLCO = Prostate, Lung, Colorectal and Ovarian cancer trial.

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

^bP_inter: P-value for interaction analysis between characteristic and number of protective genotypes;

^cTwo observations missing of tumor stage and eight observations missing of chemotherapy/radiotherapy/surgery in PLCO dataset.

3.4. in silico functional validation

Experimental data from the ENCODE project²⁵ (Supplemental Figure 3a and 3b), revealed the two SNPs (i.e., HDAC2 rs13213007 and PPARGC1A rs60571065) to be located in a DNase I hypersensitive site, where the DNase hypersensitivity and histone modification H3K4 acetylation indicated some strong signals for active enhancer and promoter functions. The evidence from the DNase cluster and transcription factor CHIP-seq data²⁵ predicted that rs60571065 is located in the Barhl2/ Msx-3/ Sox1 motif as shown by the position weight matrix (Supplemental Figure 3c, 3d and 3e), and that the minor allele may affect the binding activity to have an impact on the transcription factors.

In addition, we performed eQTL analysis to correlate genotypes of the SNPs and mRNA expression levels using data of the 373 European descendants available from the 1000 Genomes Project¹⁹. Only the HDAC2 rs13213007 A allele (GA+AA) showed a significant correlation with increased mRNA expression levels of the gene (P=0.043, Figure 2Ca), while this was not the case for the PPARGC1A rs60571065 A allele; however, in the whole blood and lung tissue data of the GTEx project, both SNPs were not associated with expression levels of HDAC2 and PPARGC1A (Supplemental Table 9). Taken together, these findings suggest that HDAC2 rs13213007A allele, but not PPARGC1A rs60571065A allele, may influence HDAC2 gene expression at the transcription level.

Finally, to find the expression of these genes in NSCLC, we compared the mRNA expression levels of these genes in 111 paired NSCLC tumor and adjacent normal tissue samples obtained from the TCGA database. Expression levels of HDAC2 were higher in the tumor tissues (P<0.001), compared with the adjacent normal tissues (Supplemental Figure 4a, 4b and 4c), and the higher expression levels were associated with a poorer NSCLC OS (Supplemental Figure 5a). On the other hand, the expression levels of PPARGC1A were lower in the tumor tissues (P<0.001), compared with the adjacent normal tissues (Supplemental Figure 4d, 4e and 4f), and the higher expression levels were associated with a better NSCLC OS³⁰ (Supplemental Figure 5b).

3.5. Mutation analysis

We accessed the mutation status of HDAC2 and PPARGC1A in lung tumor tissues by using the public database of the cBioPortal for Cancer Genomics. As shown in Supplemental Figure 6a and 6b, HDAC2 had a low somatic mutation rate in NSCLC (1.05%, 12/1144) in the 2016 TCGA study³¹, LUAD (1.64%, 3/183 and 1.24%, 7/566) in the Broad³² and TCGA PanCan studies²⁷, respectively; and LUSC (0.41%, 2/487) in the TCGA PanCan study. In contrast, PPARGC1A had a relative higher somatic mutation rate in NSCLC (3.06%, 35/1144) in the TCGA 2016 study³¹, LUAD (3.28%, 6/183; 3.89%, 22/566; and 3.89%, 22/566) in the Broad³² and TCGA PanCan studies²⁷, respectively; and LUSC (3.37%, 6/178; 2.26%, 11/487; and 0.98%, 5/511) in the TCGA pub, TCGA PanCan and TCGA²⁷ studies, respectively. Given the rarity of mutations in these two genes, our results suggest that the functional SNPs in HDAC2 may play a relatively important role in the dysregulation of mRNA expression in tumor tissues, whereas the mutations might also play a role in altered functions and expression of PPARGC1A in addition to the causal SNPs.

4. DISCUSSION

In the present study, we found two genetic variants (i.e., HDAC2 rs13213007 G>A and PPARGC1A rs60571065 T>A) in the CBP pathway gene-set to be significantly associated with survival of patients with NSCLC in both the PLCO trial and the HLCS GWAS datasets. The results suggest that patients with risk genotypes (i.e., HDAC2 rs13213007 GA+AA or PPARGC1A rs60571065 TA+AA) had a worse prognosis. Furthermore, HDAC2 rs13213007 G>A appears to influence HDAC2 mRNA expression, which makes this SNP-associated risk of death biologically plausible.

Histone deacetylases (HDACs) function to deacetylate the ε-amino group of lysine residues of both histone and non-histone substrates⁷. HDACs are grouped into class I–IV enzymes³³, i.e., Class I contains HDAC1, 2, 3 and 8, class II HDAC4, 5, 6, 7, 9 and 10, and class IV HDAC, whereas Class III enzymes contains the SIRT deacetylases³⁴. Studies suggest that HDAC2 is crucial for embryonic development, growth regulation, and affects cytokine signaling relevant for immune responses³⁵ and is often overexpressed in lung cancer and other tumors³⁶. Both published in vitro and in vivo experiments have indicated that the aberrant regulation of HDAC2 may confer an oncogenic potential for lung cancer cells and hepatocellular carcinoma cells by deregulating apoptosis and expression of cell cycle proteins³⁷, while overexpression of HDAC2 has been found to be associated with a shorter relapse-free survival after radical prostatectomy in prostate cancer³⁸. Furthermore, the upregulation of HDAC2 may enhance proliferation of gastric cancer cells by deregulating cell cycle proteins³⁹, and HDAC2 may also play a fundamental role in the Myc-mediated oncogenesis⁴⁰. These are consistent with our findings that HDAC2 expression was increased in lung cancer tissues, compared with normal tissues in the TCGA dataset. However, the detailed mechanisms of HDAC2 SNPs underlying the observed death-risk associations warrant further investigations.

Few studies have reported the role of HDAC2 SNPs in cancer outcomes. One study found that the rs11391 SNP in the 3’-UTR of HDAC2 was a potential genetic marker of poor prognosis for NSCLC⁴¹, while another study identified several functional SNPs in HDAC2 and evaluated their associations with clinical outcomes in hepatocellular carcinoma, but no significant associations were found⁴². In the present study, we showed that the HDAC2 rs13213007 A allele was associated with a poorer survival of NSCLC, likely due to its association with an increase in mRNA expression levels of the gene; moreover, HDAC2 mRNA expression levels were found to be higher in tumor tissues, and higher expression levels were also associated with a worse survival in NSCLC. According to the ENCODE project database²⁵, HDAC2 rs13213007 is located in a DNase I hypersensitive site with considerable levels of histone modification H3K4 acetylation, which may lead to an enhanced transcriptional activity. Therefore, it is likely that that HDAC2 might act as an oncogene in NSCLC.

Similarly, there are also few studies that have reported the role of PPARGC1A SNPs in cancer outcomes. Previous studies have reported that abnormalities in cell metabolism are associated with tumorigenesis where mitochondria are key regulators⁴³. PPARGC1A is a known regulator of mitochondrial biogenesis and is a multiple-function transcriptional coactivator that has been noted to be associated with various human diseases, including type II diabetes mellitus⁴⁴, coronary disease⁴⁵ and others. One study utilized ChIP-seq to obtain PPARGC1A binding sites across the genome in hepatoma cells HepG2 and the subsequent conserved-motif analysis showed that the majority of the PPARGC1A binding sites were located in multiple regulatory factor binding regions⁴⁶. To date, however, the studies on the expression and function as well as oncogenic roles of PPARGC1A in lung cancer are inconsistent. For example, one report found that PPARGC1A might be an oncogene⁴⁷, while the other demonstrated that it might be a suppressor gene⁴⁸.

In the TCGA database, we found that mRNA expression levels of PPARGC1A were higher in normal lung tissues than in tumor tissues (n=111), while its lower expression levels were associated with a worse survival in NSCLC. However, PPARGC1A was also reported to act as an oncogene, because its expression levels were upregulated in tumor tissues (n=3), facilitating lung cancer metastasis⁴⁹. It is possible that the expression levels of PPARGC1A in lung cancer tissues may be affected by other factors, such as an imbalanced activation by risk genotypes, which may cause abnormal expression of PPARGC1A, in addition to mutations. Based on the ENCODE Project data, PPARGC1A rs60571065 is located in a DNase I hypersensitive site, indicating its potential function to modify gene expression. Clearly, additional studies are warranted to investigate possible mechanisms underlying PPARGC1A rs60571065-associated death risk of NSCLC.

In the clinical setting for NSCLC patients with EGFR mutations, treatment with epidermal growth factor receptor-tyrosine kinase inhibitors (EGFR-TKIs) has resulted in an increased survival rate in comparison with the standard chemotherapy⁵⁰. Because EGFR-TKIs have been used as the first-line treatment in the application of concurrent or sequential chemo-radiotherapy regimens, different treatment options should be incorporated into the phenotype analysis⁵¹. Recently, the 3rd-generation EGFR-TKIs (EGFR-TKI3rd) have displayed remarkable efficacy against the T790M-related resistance in mutation-positive NSCLC⁵². Despite dramatic initial responses, a genetically driven resistance to the targeted therapy eventually emerged; thus, limiting their effectiveness in prolonged treatments⁵². Some reports suggest that it is due in part to differences in co-occurring genetic alterations, such as mutations of TP53 or PTEN⁵³, but few reports have addressed the roles of SNPs in other oncogenes. In the present study, we found that risk genotypes of HDAC2 rs13213007 promoted its mRNA expression levels, suggesting that the risk genotypes may have an impact on T790M-related resistance in NSCLC, which may provide a clue for individualized therapy⁵⁴.

While tobacco smoking is a major risk factor for lung cancer, smoking behavior could enhance the adverse genetic effect of the variants by increasing c-Jun expression in lung cancer⁵⁵. In the present study, we found that smoking might increase the risk of death associated with HDAC2 rs13213007 SNP in patients with NSCLC. It is likely that smoking could affect the function of epigenetics genes or the H3K4 acetylation activity associated with HDAC2, a similar phenomenon seen on other cancers, including cancers of the pancreas, stomach, bladder, head and neck, and the colorectum⁵⁶.

Although we observed associations between genetic variants in two CBP pathway genes and NSCLS survival along with some functional evidence, the exact molecular mechanisms are still unclear. Further functional experiments are required to corroborate the present findings. Additionally, since both the discovery and validation datasets used in the present study were from Caucasian populations, the results may not be generalizable to other ethnic populations. Although the sample size of PLCO was relatively large, the number of patients in each subgroup were still relatively small, which might have reduced the statistical power to detect a small effect in one particular subpopulation. Furthermore, since only a few clinical factors were available for additional analysis, other covariates, such as performance status and treatments, were not available for further adjustments. Additionally, it is worth noting that the detailed genotype and phenotype data from the HLCS study were not accessible for us to perform additional stratified analysis in a larger sample size.

In conclusion, two independent functional SNPs (i.e., HDAC2 rs13213007 G>A and PPARGC1A rs60571065 T>A) were found to be significantly associated with NSCLC survival in both the PLCO trial and the HLCS GWAS datasets. The combined analysis revealed that the two SNPs had a significant association with survival and that patients with more risk genotypes had a worse prognosis, possibly due to the progression acceleration and metastatic effects of overexpressed HDAC2. Knowing the characteristic overexpression of HDAC2 in cancer, the discovery of the novel genetic variant of HDAC2 rs13213007 may serve to showcase additional insights for potential therapeutic capabilities of HDAC inhibitors and in cancer treatment. The findings in the present study also provides a solid foundation for further functional studies to identify molecular mechanisms underlying the observed death-risk associations in the progression of NSCLC.

ABBREVIATIONS

NSCLC

Non-small cell lung cancer

SNP

single nucleotide polymorphism

CBP

CREB-binding protein

GWAS

Genome-Wide Association Study

PLCO

the Prostate, Lung, Colorectal and Ovarian cancer screening trial

HLCS

Harvard Lung Cancer Susceptibility study

OS

overall survival

DSS

disease-specific survival

LD

linkage disequilibrium

BFDP

Bayesian false discovery probability

eQTL

expression quantitative trait loci

TCGA

the Cancer Genome Atlas

ROC

receiver operating characteristic

HR

hazards ratio

CI

confidence interval

AUC

area under the curve