Distinct Cancer-Specific Survival in Metastatic Prostate Cancer Patients Classified by a Panel of Single Nucleotide Polymorphisms of Cancer-Associated Genes

Abstract

Individual genetic variations may have a significant influence on the survival of metastatic prostate cancer (PCa) patients. We aimed to identify target genes and their variations involved in the survival of PCa patients using a single nucleotide polymorphism (SNP) panel. A total of 185 PCa patients with bone metastasis at the initial diagnosis were analyzed. Germline DNA in each patient was genotyped using a cancer SNP panel that contained 1,421 SNPs in 408 cancer-related genes. SNPs associated with survival were screened by a log-rank test. Fourteen SNPs in 6 genes, XRCC4, PMS1, GATA3, IL13, CASP8, and IGF1, were identified to have a statistically significant association with cancer-specific survival. The cancer-specific survival times of patients grouped according to the number of risk genotypes of 6 SNPs selected from the 14 SNPs differed significantly (0-1 v. 2-3 v. 4-6 risk genotypes; P = 7.20 × 10⁻⁸). The high-risk group was independently associated with survival in a multivariate analysis that included conventional clinicopathological variables (P = 0.0060). We identified 14 candidate SNPs in 6 cancer-related genes, which were associated with poor survival in patients with metastatic PCa. A panel of SNPs may help predict the survival of those patients.

Introduction

Over the past few decades, a great deal of effort has been devoted to identifying the genes involved in prostate cancer (PCa), and recent genome-wide association studies using the array technique have identified several dozen promising candidate genes associated with PCa risk.^1-5 Most of the past studies attempted to identify genes involved in PCa susceptibility, while more recent studies have begun to consider the clinical significance of single nucleotide polymorphisms (SNPs) as predictive or prognostic markers.⁶ Accumulating data suggested that interindividual genetic variations represented by SNPs might partially affect the prognosis of PCa as well as the biological characteristics of the cancer itself by modulating the sensitivity to endocrine therapy or chemotherapy and the microenvironment around cancer cells.⁶ It is of clinical importance to identify such SNPs for the future optimization of individualized treatment plans.

We previously demonstrated that the SNPs of insulin-like growth factor 1 (IGF1) and cytochrome P450 19 (CYP19), which had earlier been implicated in PCa susceptibility,^7,8 might be markers for poor prognosis in patients with metastatic PCa.⁹ However, in most of the past studies, only a small number of genes were evaluated for their association with survival as a supportive analysis for a larger case-control study, and there have been few comprehensive SNP investigations based on this concept. Recently, several studies searched for SNPs related to the survival of lung and bladder cancer patients and reported some candidate SNPs or genes.^10-12 Meanwhile, in PCa, although a few studies tried to explore survival-associated SNPs in a large population-based cohort,^13,14 there was no study that exploratively investigated SNPs involved in the survival of patients with a specific disease status.

In this study, we conducted an exploratory study using a cancer SNP panel to identify SNPs associated with the survival of PCa patients with bone metastasis at initial presentation.

Results

Clinicopathological background of patients

The patient demographics and clinical characteristics are shown in Table 1. Eighteen patients received at least one course of docetaxel or docetaxel-containing chemotherapy during the follow-up period or before death. The median overall survival and cancer-specific survival times in the whole patient group were 4.62 and 4.77 years, respectively.

Table 1.

Patient Demographics and Clinical Characteristics

Variable	Mean ± standard deviation	Median	Range
Age, y	69.4 ± 8.8	70	45-89
Follow-up, mo	43.8 ± 33.4	37	1-210
PSA, ng/mL	1,088 ± 1,980	313	0.2-12,490
Hemoglobin, g/dL	13.3 ± 2.0	13.5	6.2-17.4
ALP, IU/L	654 ± 902	306	7-5,870
LDH, IU/L	299 ± 184	232	97-956
Variable		n	%
Gleason score
<6		15	8.1
7		26	14.1
8		39	21.1
9		81	43.8
10		13	7.0
Unknown		11	5.9
Site of metastases
Bone alone		95	51.3
Bone + lymph nodes		89	48.1
Bone + visceral organs		13	7.0
Initial treatments
Surgical castration alone		22	11.9
LH-RH analog alone		47	25.4
Combined androgen blockade		116	62.7

Note: ALP = alkaline phosphatase; LDH = lactate dehydrogenase; LH-RH = luteinizing hormone–releasing hormone; PSA = prostate-specific antigen.

Survival analysis

We narrowed the number of SNPs down to 176 at the first screening step using the minimum genotype frequency (MGF), and the top 14 SNPs were finally selected at the second screening step based on the false discovery rate (FDR) criterion (Table 2). Hardy-Weinberg equilibrium was assumed for all the SNPs. The selected SNPs were located in 6 genes: X-ray repair complementing defective repair in Chinese hamster cells 4 (XRCC4), postmeiotic segregation increased 1 (PMS1), GATA binding protein 3 (GATA3), interleukin 13 (IL13), caspase 8 (CASP8), and IGF1 (Table 2). The SNPs showed strong linkage disequilibrium with each other in 4 of 6 genes with multiple candidate SNPs (D′ = 1.000 in XRCC4 and IL13, 0.985-1.000 in PMS1, and 0.9862-1.000 in GATA3).

Table 2.

SNPs Selected for Association with Cancer-Specific Survival of Prostate Cancer with Bone Metastasis

Gene (location, function), SNP	Risk genotypes	MGFa	Log-rank χ2	P
XRCC4 (5p14, DNA strand-break repair)
rs2891980	GG (v. AG, AA)	45	9.49	0.0021
rs1805377	AA (v. AG, GG)	46	9.10	0.0026
PMS1 (2q31, DNA mismatch repair)
rs256550	GG, AG (v. AA)	46	8.39	0.0038
rs256552	GG, AG (v. AA)	46	7.82	0.0052
rs256564	AA, AG (v. GG)	46	7.27	0.0070
rs256563	GG, AG (v. AA)	46	6.84	0.0089
rs256567	AA, AG (v. GG)	44	6.04	0.0140
GATA3 (10p14, transcription factor)
rs570730	GG, AG (v. AA)	50	6.86	0.0088
rs10752126	GG, CG (v. CC)	49	6.85	0.0089
rs569421	GG, AG (v. AA)	49	5.21	0.0225
IL13 (5q31, cytokine)
rs1295686	AA, AG (v. GG)	50	6.22	0.0126
rs20541	AA, AG (v. GG)	48	6.06	0.0138
CASP8 (2q33, apoptosis)
rs2293554	AA (v. AC, CC)	48	5.46	0.0195
IGF1 (12q23, growth factor)
rs2162679	GG, AG (v. AA)	45	5.45	0.0196

Note: MGF = minimum genotype frequency; SNP = single nucleotide polymorphism.

^aMGF was defined as the lower frequency of 2 dichotomized genotype groups in each SNP.

In the leave-one-out cross-validation (LOOCV) analysis for predicting cancer-specific survival, a total of 165 patients were classified into high- and low-risk groups due to missing values of at least one SNP locus in 20 patients. The median cancer-specific survival time of the high-risk group was significantly shorter than that in low-risk patients (4.29 v. 7.09 years, respectively; P = 0.0050) (Fig. 1).

Figure 1.

Cancer-specific survival of patients according to risk categorization using candidate SNPs identified by array analysis. A prognostic scoring index using the 14 SNPs selecz the screening was developed by incorporating the difference in their effect sizes to classify high-risk and low-risk groups. Its predictive accuracy was assessed by the LOOCV analysis for the whole model building process. The median cancer-specific survival time of the high-risk group was significantly shorter than that in low-risk patients (P = 0.0050).

In a univariate Cox proportional hazard analysis of the association between cancer-specific survival and the clinicopathological variables, a higher alkaline phosphatase (ALP) level (hazard ratio [HR] = 2.92; 95% confidence interval [CI] = 1.79-4.77; P = 2.05 × 10⁻⁵), higher lactate dehydrogenase (LDH) level (HR = 2.70; 95% CI = 1.47-4.95; P = 1.34 × 10⁻³), and Gleason score of 9 or greater (HR = 2.44; 95% CI = 1.52-3.90; P = 2.01 × 10⁻⁴) were significantly associated with a shorter cancer-specific survival time (Table 3). The administration of docetaxel or docetaxel-containing chemotherapy did not significantly influence cancer-specific survival (P = 0.369). In the LOOCV analysis, the effect of the predicted risk classification based on genetic variables was statistically significant after adjusting for the clinicopathological variables (P = 0.0060 for the Wald statistic in the multivariate Cox regression model evaluated by the permutation method). These results suggest that the genetic classification was independent of the clinical prognostic variables.

Table 3.

Cox Proportional Hazard Regression Analysis of Factors Associated with Cancer-Specific Survival

Variable	Univariate analysis		Multivariate analysis (model 1)		Multivariate analysis (model 2)
	HR (95% CI)	P	HR (95% CI)	P	HR (95% CI)	P
Age (≥70 v. <70 y)	1.34 (0.85-2.11)	0.204
PSA (≥315 v. <315 ng/mL)	1.42 (0.90-2.24)	0.137
Hemoglobin (≤13.5 v. >13.5 g/dL)	1.14 (0.71-1.83)	0.589
ALP (≥350 v. <350 IU/L)	2.92 (1.79-4.77)	2.05 × 10−5	2.64 (1.51-4.62)	6.93 × 10−4	2.22 (1.32-3.73)	2.63 × 10−3
LDH (≥500 v. <500 IU/L)	2.70 (1.47-4.95)	1.34 × 10−3	1.55 (0.78-3.10)	0.215	1.22 (0.62-2.40)	0.570
Gleason score (≥9 v. <9)	2.44 (1.52-3.90)	2.01 × 10−4	2.10 (1.20-3.67)	8.95 × 10−3	2.16 (1.31-3.56)	2.69 × 10−3
Risk groupa (high v. low)	2.13 (1.28-3.52)	3.43 × 10−3	2.07 (1.18-3.62)	1.13 × 10−2
No. of risk genotypesb (4-6 v. 0-3)	3.21 (2.04-5.05)	4.70 × 10−7			3.06 (1.80-5.19)	3.58 × 10−5

Note: ALP = alkaline phosphatase; CI = confidence interval; HR = hazard ratio; LDH = lactate dehydrogenase; PSA = prostate-specific antigen.

^aGroup was classified by a leave-one-out cross-validation method.

^bThe number of risk genotypes was calculated using 6 representative single nucleotide polymorphisms selected from each candidate gene.

Next, we performed a survival analysis according to the number of risk genotypes within the 6 candidate genes. Since all the SNPs in a gene were in strong linkage disequilibrium to each other, the SNP with the smallest P value based on a log-rank test was picked to represent each gene (i.e., rs2891980 for XRCC4, rs256550 for PMS1, rs570730 for GATA3, rs1295686 for IL13, rs2293554 for CASP8, and rs2162679 for IGF1). We gave the same weight to each SNP because the HRs of risk genotypes in the representative SNPs were within the range of 1.71 to 2.00. The developed score based on the refined set of 6 SNPs was highly correlated with the original score based on the 14 SNPs (Spearman correlation: 0.88).

Finally, using the entire set of patients, we developed a multivariate model for prognostic prediction using both the genetic score and the clinical prognostic factors to establish a method of independent, external validation for future studies. Each patient was assigned to 1 of 3 groups: a favorable-risk group with 0 to 1 risk genotypes, an intermediate-risk group with 2 to 3 risk genotypes, and a poor-risk group with 4 to 6 risk genotypes based on the genetic score using 6 SNPs. The cancer-specific survival times were significantly different among the 3 groups, with median values of 13.3, 7.0, and 3.8 years for the favorable-, intermediate-, and poor-risk groups, respectively (log-rank, P = 7.20 × 10⁻⁸) (Fig. 2). In the multivariate Cox proportional hazard analysis with this model, the risk classification (poor v. favorable or intermediate risk) (HR = 3.06; 95% CI = 1.80-5.19; P = 3.58 × 10⁻⁵) was again indicated as an independent variable, predicting cancer-specific survival along with the ALP level (HR = 2.22; 95% CI = 1.32-3.73; P = 2.63 × 10⁻³) and Gleason score (HR = 2.16; 95% CI = 1.31-3.56; P = 2.69 × 10⁻³) (Table 3). For subgroup analyses (Fig. 3), patients were divided into subgroups on the basis of ALP levels (<350 IU/L or ≥350 IU/L) and Gleason scores (<9 or ≥9). Then, cancer-specific survival times were compared in each subgroup stratified by the number of risk genotypes (0-3 v. 4-6). In the subgroup with ALP <350 IU/L, the cancer-specific survival time for patients with 4 to 6 risk genotypes was significantly worse than that for patients with 0 to 3 risk genotypes (P = 1.69 × 10⁻⁷). No significant difference in survival was observed between patients with 0 to 3 risk genotypes and those with 4 to 6 risk genotypes in the subgroup with ALP ≥350 IU/L (P = 0.681). With regard to Gleason scores, there were significant differences in cancer-specific survival between patients with 0 to 3 risk genotypes and those with 4 to 6 risk genotypes in both Gleason score subgroups (<9: P = 0.0004; ≥9: P = 0.0004).

Figure 2.

Cancer-specific survival of patients according to the number of risk genotypes of 6 representative SNPs selected from 6 candidate genes. Each patient was assigned to 1 of 3 groups according to the number of risk genotypes in 6 representative SNPs selected from each candidate gene: 0-1, 2-3, and 4-6. The cancer-specific survival times differed significantly among the 3 risk groups (P = 7.20 × 10⁻⁸).

Figure 3.

Cancer-specific survival based on the number of risk genotypes for 6 representative SNPs in the following groups: (A) ALP <350 IU/L, (B) ALP ≥350 IU/L, (C) Gleason score <9, and (D) Gleason score ≥9. There were significant differences in cancer-specific survival between patients stratified by the number of risk genotypes (0-3 v. 4-6) in the subgroups shown: (A) P = 1.8 × 10⁻⁷, (C) P = 0.0004, and (D) P = 0.0004.

The same exploratory approach was performed for selecting SNPs associated with overall survival. However, none of the SNPs met the screening criteria of MGF and FDR employed in this study.

Discussion

A comprehensive SNP analysis was performed to identify SNPs related to the survival of metastatic PCa patients using a cancer-related SNP panel on which SNPs in various types of genes related to cell cycle regulation, apoptosis, DNA repair, proliferation, migration, steroid metabolism, among others, were integrated. Based on the results of this analysis, we selected 14 candidate SNPs in 6 genes and showed that they had the potential to increase the accuracy of survival prediction. Among the 6 genes involved, 2 were related to DNA repair (XRCC4 and PMS1), and the others were involved in the regulation of a transcription factor (GATA3), immune system (IL13), apoptosis (CASP8), and cell growth (IGF1). Most of the SNPs were mapped on either the intron or 3′-untranslated region, and one of the SNPs (rs1805377 in XRCC4) altered a splicing acceptor site. Meanwhile, only one exonic SNP (rs20541 in IL13) comprised a nonsynonymous amino acid substitution, which changed a glutamine to an arginine at position 144 (144 Glu>Arg).

The aim of this study was to identify candidate SNPs and cancer-related genes involved in the survival of PCa patients with bone metastasis. Although the percentage of patients who are diagnosed with PCa with distant metastasis has decreased to as low as 1% in countries where prostate-specific antigen (PSA) screening is widely prevalent,¹⁵ we expect the present results to have clinical significance not only for such patients but also potentially for patients with locally advanced or recurrent PCa. One of the goals of this study was to actively intervene in the therapy to prolong survival by the individual genomic background based on progression-related SNPs in addition to conventional markers. However, it is to be noted that the SNPs in the 6 genes selected in this study are not candidates for predictive markers but prognostic markers at present. To optimize the treatment using this panel of SNPs, there will be a need for further investigations in which the responses to particular treatments are compared between patients with different genotypes along with sufficient validation studies using a large-scale sample set. The other goal was to identify genes involved in PCa progression. We believe that revealing the function of these genes will help to delineate the molecular mechanism underlying the progression of PCa as well as to the development of therapeutic agents targeting those genes.

We consider that a cohort of metastatic PCa patients is one of the most suitable clinical models for assessing the individual genetic background of cancer progression. To assess the influence of inherited genetic factors on cancer progression, several cohorts of patients with different clinical backgrounds can be chosen for the analysis. In patients with localized PCa, there are many treatment options, and it can be difficult to distinguish the results due to genetic factors and those due to treatment-specific factors. Meanwhile, although there was a more than 10-year gap in the entry period in this study, the patients with metastatic PCa at diagnosis were treated according to a fairly uniform therapeutic protocol consisting of androgen suppression therapy followed by estrogens, steroids, and chemotherapeutic agents. Although patients with castration-resistant PCa are often treated with docetaxel, it is suggested that this agent had only a limited effect on survival in this study design.¹⁶

Using the cohort as a clinical model for assessing cancer progression, various biological mechanisms are speculated to explain why each candidate SNP affects survival.⁶ First, the SNP may affect the cancer biology, including cell cycle regulation, tumor transformation, apoptosis, cell adhesion, or migration. The SNPs directly relevant to cancer biology may influence patient survival as well as a susceptibility or clinical phenotype of cancers. Second, the SNPs may affect the response to endocrine therapies or chemotherapies and adverse effects. Recent genetic and pharmacological studies have revealed that many SNPs in genes involved in drug and hormone metabolism and disposition exerted an influence on circulating or intracellular levels of drugs or hormones.^17,18 Furthermore, the response to cytotoxic agents is also considered sensitive to genes other than drug-related genes, such as DNA repair genes. The clinical response modulated by functional SNPs of those genes may affect survival. Thirdly, the SNPs may have an effect on the intracorporeal environment and microenvironment around cancer cells, thus affecting the invasive and metastatic potential.¹⁹ Furthermore, intracrine androgen levels that may be affected by variants of steroid hormone-related genes in the prostate might play a role in the growth of castration-resistant cancer cells.²⁰

We acknowledge that this study has several limitations. First, the possibility of patient selection bias cannot be eliminated due to its retrospective multi-institutional study design. The majority of the patients enrolled in this study are incidence cases, but the possible inclusion of prevalence cases could have affected the results. Such a bias could have led to a slightly better survival time than previously reported by eliminating patients with extremely short survival times. Second, if such patients were excluded from the analysis, less frequent but more important SNPs related to the extremely aggressive phenotype could have slipped out of the screening for candidate SNPs. To explore such rare SNPs, a higher statistical power with a larger number of carefully selected samples is needed. In this study with a relatively small number of patients, a higher cutoff of MGF was required to achieve sufficient statistical power. Thus, quite common SNPs that have a modest influence on survival were considered to be detected in this study. Thirdly, the validation of the results was still insufficient due to the lack of the number of patients. Additionally, patients enrolled in this study were entirely from Japan. The expected prevalence of SNPs will differ relative to other populations. External validation with a larger series of patients or other ethnicities will be required to translate these results into clinical applications. A multi-institutional prospective trial for validating the candidate SNPs as prognostic markers for PCa patients with bone metastasis (UMIN trial ID: UMIN000009785) is now ongoing by us and other members in Japan.

In conclusion, we identified 14 candidate SNPs in 6 cancer-related genes associated with cancer-specific survival (CSS) in PCa patients with bone metastasis at the initial diagnosis. Using a panel of the SNPs, the prediction of the survival and optimization of the individualized treatment for patients with advanced PCa may be possible in the future.

Materials and Methods

Patients

From July 1980 to September 2008, 191 native Japanese patients with PCa with bone metastasis at initial presentation were enrolled in this study. They were diagnosed at Akita University Hospital and its related community hospitals, Kyoto University Hospital, Tohoku University Hospital, and Hirosaki University Hospital, and had undergone no previous treatments. Pathological diagnosis was made by prostate needle biopsy, and metastasis was identified by radiography, computed tomography, or bone scintigraphy. All the patients were initially treated by endocrine therapy. Combined androgen blockade was defined as luteinizing hormone–releasing hormone analog plus antiandrogens. After treatment failure with the primary therapy, optional therapies, including other antiandrogens, estrogens, steroids, chemotherapeutic agents, or palliative radiation, were added to or substituted for the preceding therapies.

Pathological grading of needle biopsy specimens was performed by local pathologists. Pretreatment hemoglobin, ALP, LDH, and PSA levels before the initial treatment of PCa were obtained from medical charts. An independent end-point reviewer at each institution determined the cause of death on the basis of standardized extractions from the patient files without providing genotype data of each patient.

Genotyping analysis

Approximately 2 ug of genomic DNA was prepared from a peripheral blood sample of each patient using a QIAamp Blood Kit (Qiagen, Hilden, Germany) or standard phenol-chloroform extraction. Genotyping was performed using a Cancer SNP Panel (Illumina Inc., San Diego, CA) that contained 1,421 SNPs in 408 cancer-related genes selected from the SNP500 Cancer Database.²¹ This panel included several genes related to the synthesis or metabolism of androgens and steroid hormones. Since 6 of 191 samples were eliminated from this study due to unsuccessful genotyping (sample call rate: 96.9%), a total of 185 patients were subjected to statistical analyses. The locus success rate and overall genotype call rate were 98.2% and 95.5%, respectively.

Statistical analysis

The end points of this study were death from any cause and PCa-specific death, and the survival time was calculated from the date of diagnosis to the day of death. The log-rank test was used to test for an association with survival in either the dominant, recessive, or additive model for a variant allele of each SNP, and the most suitable model with the smallest P value was adopted for each SNP. The MGF was defined as the lower frequency of 2 dichotomized genotype groups in each SNP. We used a cutoff of 0.40 for the MGF with the strategy to detect a common, high-frequency genetic abnormality.

We used a cutoff of 0.30 for the FDR to correct for the multicity problem in multiple testing. We performed a LOOCV analysis. Specifically, in each fold of LOOCV, we selected SNPs by multiple testing from scratch, developed a compound covariates prediction score for the training set, and classified the test sample into high- or low-risk groups depending on whether the test sample score was greater than the median of the prediction score in the training set.²² After completion of the LOOCV analysis, we compared the survival time between the 2 groups for the entire sample. To assess the statistical significance of a log-rank test statistic for no survival difference, we performed a permutation method with 2,000 repetitions of the whole process of the LOOCV analysis after random permutations of the survival times.²³

To compare the survival times according to clinicopathological factors, patients were dichotomized by the median value. The influences of SNPs and the clinicopathological factors on survival were assessed by a Cox proportional hazard regression analysis. After the LOOCV analysis, we also fit a multivariate Cox model with both the predicted risk classification based on genetic variables (i.e., the predicted high- and low-risk groups) and the clinicopathological factors as the covariates. Overall and cancer-specific survival times were estimated using the Kaplan-Meier method. SPSS software version 16.0 (SPSS Japan Inc., Tokyo, Japan) and SAS 9.2 (SAS Institute Inc., Cary, NC) were used for survival analyses, and 2-sided P values of less than 0.05 were considered to indicate a statistical significance.