SEARCH

SEARCH BY CITATION

Keywords:

  • prostate cancer;
  • nomograms;
  • progression-free survival;
  • genetics;
  • single nucleotide polymorphisms

Abstract

  1. Top of page
  2. Abstract
  3. Introduction
  4. Patients and Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. Conflict of Interest
  9. References
  10. Supporting Information

What's known on the subject? and What does the study add?

  • Currently available nomograms to predict preoperative risk of early biochemical recurrence (EBCR) after radical prostatectomy are solely based on classic clinicopathological variables. Despite providing useful predictions, these models are not perfect. Indeed, most researchers agree that nomograms can be improved by incorporating novel biomarkers. In the last few years, several single nucleotide polymorphisms (SNPs) have been associated with prostate cancer, but little is known about their impact on disease recurrence.
  • We have identified four SNPs associated with EBCR. The addition of SNPs to classic nomograms resulted in a significant improvement in terms of discrimination and calibration. The new nomogram, which combines clinicopathological and genetic variables, will help to improve prediction of prostate cancer recurrence.

Objectives

  • To evaluate genetic susceptibility to early biochemical recurrence (EBCR) after radical prostatectomy (RP), as a prognostic factor for early systemic dissemination.
  • To build a preoperative nomogram to predict EBCR combining genetic and clinicopathological factors.

Patients and Methods

  • We evaluated 670 patients from six University Hospitals who underwent RP for clinically localized prostate cancer (PCa), and were followed-up for at least 5 years or until biochemical recurrence.
  • EBCR was defined as a level prostate-specific antigen >0.4 ng/mL within 1 year of RP; preoperative variables studied were: age, prostate-specific antigen, clinical stage, biopsy Gleason score, and the genotype of 83 PCa-related single nucleotide polymorphisms (SNPs).
  • Univariate allele association tests and multivariate logistic regression were used to generate predictive models for EBCR, with clinicopathological factors and adding SNPs.
  • We internally validated the models by bootstrapping and compared their accuracy using the area under the curve (AUC), net reclassification improvement, integrated discrimination improvement, calibration plots and Vickers' decision curves.

Results

  • Four common SNPs at KLK3, KLK2, SULT1A1 and BGLAP genes were independently associated with EBCR.
  • A significant increase in AUC was observed when SNPs were added to the model: AUC (95% confidence interval) 0.728 (0.674–0.784) vs 0.763 (0.708–0.817).
  • Net reclassification improvement showed a significant increase in probability for events of 60.7% and a decrease for non-events of 63.5%.
  • Integrated discrimination improvement and decision curves confirmed the superiority of the new model.

Conclusions

  • Four SNPs associated with EBCR significantly improved the accuracy of clinicopathological factors.
  • We present a nomogram for preoperative prediction of EBCR after RP.

Abbreviations
PCa

prostate cancer

RP

radical prostatectomy

BCR

biochemical recurrence

EBCR

early biochemical recurrence

SNPs

single nucleotide polymorphisms

AUC

area under the curve

KLK2

kallikrein-related peptidase 2

KLK3

kallikrein-related peptidase 3

SULT1A1

Sulfotransferase 1A1

BGLAP

bone gamma-carboxyglutamic acid-containing protein

UCSF-CAPRA

University of California, San Francisco Cancer of the Prostate Risk

Introduction

  1. Top of page
  2. Abstract
  3. Introduction
  4. Patients and Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. Conflict of Interest
  9. References
  10. Supporting Information

Up to 80% of patients with prostate cancer (PCa) are now diagnosed with clinically localized disease [1]. Radical prostatectomy (RP) is the most common curative treatment option for these patients. However, after RP, approximately 35% of patients will experience rising PSA levels, referred to as biochemical recurrence (BCR) [2]. Rising PSA levels will be the result of either a local recurrence as a consequence of non-radical local surgery, or systemic recurrence because of tumour dissemination before surgery.

An early biochemical recurrence (EBCR) after RP, especially within 1 year, has prognostic relevance, because it suggests that systemic disease was already present before surgery [3-5].

It is surprising that clinically or even pathologically localized PCa could trigger a systemic failure. So, how can we explain an early predisposition to systemic failure from supposedly localized stages? Can we expect a genetic predisposition to early systemic dissemination? Could we predict this predisposition preoperatively?

Knowing the genetic predisposition of a patient to early systemic dissemination, even though they may have a clinically localized PCa, would lead us to consider radiotherapy with hormonal adjuvant treatment instead of RP, or to suggest the inclusion of such patients in early adjuvant protocols or clinical trials, despite having pathologically localized PCa.

Recently, several germline genetic polymorphisms have been associated with the risk of developing PCa [6], its aggressiveness [7] and the risk of BCR [8]. We hypothesized that certain of those polymorphisms could also promote early systemic dissemination.

In this study, we analyse the association of common single nucleotide polymorphisms (SNPs) with the risk of EBCR within 1 year after RP, as a surrogate for systemic failure in clinically localized PCa.

In addition, we attempt to develop new preoperative nomograms to predict EBCR, combining standard clinicopathological parameters and SNPs. Finally, we compare the predictive accuracy of models with and without SNPs.

Patients and Methods

  1. Top of page
  2. Abstract
  3. Introduction
  4. Patients and Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. Conflict of Interest
  9. References
  10. Supporting Information

After exclusion of 33 patients because of missing data, a total of 670 patients were evaluated. All patients gave written informed consent. The study was approved by the Clinical Research Ethical Committee of University Hospital Vall d'Hebron (Barcelona), and it was in accordance with the Helsinki Declaration and the European Medicines Agency recommendations.

Study inclusion criteria were: (i) clinically localized PCa subjected to RP, (ii) without adjuvant treatment, (iii) followed until BCR or for at least 5 years after RP, and (iv) Caucasian origin. From January 2002 to May 2009, 703 patients with PCa were enrolled, from six institutions.

All patients were genotyped using a microarray with allele-specific probes for 83 SNPs, which have been selected for their association with PCa risk or aggressiveness according to published literature (see Supplementary material, Table S1). As the study was focused on germline variants, there is no concern about the time point of sample collection. Briefly, DNA from blood or saliva was used for amplifying target genes in six multiplex-PCRs. The PCR products were fluorescently labelled and hybridized (Ventana Medical Systems, Tucson, AZ, USA). The microarrays were scanned (Innopsys S.A., Carbonne, France) and genotypes were determined using MG1.0 software (Progenika Biopharma, Bilbao, Spain) [9, 10].

Age, preoperative PSA, clinical stage, biopsy Gleason score, and the SNPs, were analysed as candidate predictors. EBCR was defined as a PSA level >0.4 ng/mL [11] within 1 year of surgery. PSA was evaluated at 1.5–3 months after RP, and then, every 3–6 months depending on the previous value.

A preliminary variable selection was performed based on univariate association with EBCR for clinicopathological variables, and on allele association tests for SNPs, using chi-squared and Mann–Whitney tests. Subsequently, stepwise logistic regression was used to determine the optimal predictive model.

For multivariate prediction models, PSA was modelled as its natural logarithm, clinical stage and Gleason score were grouped into three categories, and a weighted risk score [12] was built using selected SNPs. For this purpose, we defined a new variable, inline image where one variable is considered per SNP, gK = 0, 1 or 2, depending on the number of risk alleles carried at the SNP k, and the weights were estimated using a logistic regression model. For the backward selection procedure, the threshold P-value was set at 0.1, and the stopping rule was based in Akaike's information criterion. Two predictive models were built, one based on clinicopathological variables, and the other adding the genetic score.

Discrimination accuracy of the two models was compared using the area under the curve (AUC) [13], along with the net reclassification improvement and the integrated discrimination improvement [14]. Calibration was assessed graphically, and clinical utility was studied using Vickers' decision curves [15]. All analyses were performed using R programming language v.2.11.1 with the RMS, HMISC and PROC libraries added and HELIXTREE software.

Results

  1. Top of page
  2. Abstract
  3. Introduction
  4. Patients and Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. Conflict of Interest
  9. References
  10. Supporting Information

Among 670 patients, 13.3% had a PSA level >0.4 ng/mL within 1 year of RP (Table 1). Our cohort included patients with clinically localized PCa (T1–T2) and half of them were T1c. More than 66% had a preoperative PSA level <10 ng/mL and 76% had a Gleason score <7. Understaging or positive surgical margins were found in c.30% (Table 1).

Table 1. Descriptive statistics of the clinicopathological variables for 670 patients undergoing radical prostatectomy included in the study
VariableTotal, N = 670
  1. PSA, prostate-specific antigen; IQR, interquartile range; BCR, biochemical recurrence.

Preoperative PSA 
Mean, ng/mL10.09
Median (IQR), ng/mL7.94 (5.88)
<4, %5.2
4–6.9, %34.8
7–9.9, %26.9
10–19.9, %25.8
≥20, %7.3
Age at diagnosis 
Mean, years63.84
Median (IQR), years64 (8)
<54, %6.1
55–59, %17.5
60–64, %27.3
65–69, %30.4
≥70, %18.7
Biopsy Gleason sum, % 
2–676.1
719.7
8–104.2
Clinical stage, % 
T1c49.6
T2a–T2b35.4
T2c15.1
Gleason sum at surgery, % 
2–656.1
732.8
8–1011.0
Pathological stage, % 
T2a–T2b21.5
T2c50.4
T3–T428.1
Seminal vesicle involvement, % 
Negative94.0
Positive6.0
Lymph node involvement, % 
Negative98.4
Positive1.6
Surgical margins, % 
Negative69.6
Positive30.4
BCR within 1 year, % 
No86.7
Yes13.3

Gleason score, clinical stage, preoperative PSA, and four SNPs located at KLK3 (kallikrein-related peptidase 3 gene; –5429T/G, rs2569733), KLK2 (kallikrein-related peptidase 2 gene; Arg250Trp, rs198977), SULT1A1 (sulphotransferase 1A1 gene; Arg213His, rs9282861) and BGLAP (bone γ-carboxyglutamic acid-containing protein gene; –198T/C, rs1800247) genes showed independent association with EBCR (Tables 2 and 3).

Table 2. Univariate associations between baseline preoperative clinicopathological variables and early biochemical recurrence within 1 year of radical prostatectomy
VariableOverallNo BCRBCRP value
670 (100)581 (86.7)89 (13.3)
  1. Results of chi-squared and Mann–Whitney tests. PSA, prostate-specific antigen; IQR, interquartile range; BCR, biochemical recurrence.

Biopsy Gleason sum, n (%)   <0.001
2–6510 (76.1)461 (79.4)49 (55.0) 
7132 (19.7)103 (17.7)29 (32.6) 
8–1028 (4.2)17 (2.9)11 (12.4) 
Clinical stage, n (%)   0.003
T1c332 (49.6)301 (51.8)31 (34.8) 
T2a–T2b237 (35.4)201 (34.6)36 (40.4) 
T2c101 (15.0)79 (13.6)22 (24.8) 
PSA, ng/mL, Median (IQR)7.9 (5.9)7.5 (5.3)11.4 (9.7)<0.001
Table 3. Frequency distributions, odds ratios and univariate association P values for the presence of biochemical recurrence by genotype, for single nucleotide polymorphisms on KLK3 (rs2569733), KLK2 (rs198977), SULT1A1 (rs9282861), and BGLAP (rs1800247) genes
 Genotype frequencies, n (%)Frequency of risk genotype-carriers*, n (%), and ORs
TTTGGGTT*TG or GGOR95% CIP value
  1. OR, odds ratio; 95% CI, 95% confidence interval; T, thymine; G, guanine; C, cytosine; A, adenine; BCR, biochemical recurrence; KLK2, kallikrein-related peptidase 2; KLK3, kallikrein-related peptidase 3; SULT1A1, Sulfotransferase 1A1; BGLAP, bone gamma-carboxyglutamic acid-containing protein.

KLK3, -5429 T/G (rs2569733)        
No BCR331 (57.0)224 (38.6)26 (4.5)331 (57.0)250 (43.0)1.931.18–3.16<0.01
BCR64 (71.9)23 (25.8)2 (2.2)64 (71.9)25 (28.1)   
Total395 (59.0)247 (36.9)28 (4.2)395 (59.0)275 (41.0)   
KLK2, Arg250Trp (rs198977)CCCTTTCCCT* or TT*OR95% CIP value
No BCR259 (44.6)253 (43.5)69 (11.9)259 (44.6)322 (55.4)1.691.06–2.690.025
BCR28 (31.5)45 (50.6)16 (18.0)28 (31.5)61 (68.5)   
Total287 (42.8)298 (44.5)85 (12.7)287 (42.8)383 (57.2)   
SULT1A1, Arg213His (rs9282861)GGGAAAGG*GA or AAOR95% CIP value
No BCR277 (47.7)254 (43.7)50 (8.6)277 (47.7)304 (52.3)1.781.12–2.810.014
BCR55 (61.8)27 (30.3)7 (7.9)55 (61.8)34 (38.2)   
Total332 (49.6)281 (41.9)57 (8.5)332 (49.6)338 (50.4)   
BGLAP, -198 T/C (rs1800247)TTTCCCTTTC* or CC*OR95% CIP value
No BCR354 (60.9)190 (32.7)37 (6.4)354 (60.9)227 (39.1)2.001.27–3.14<0.01
BCR39 (43.8)45 (50.6)5 (5.6)39 (43.8)50 (56.2)   
Total393 (58.7)235 (35.1)42 (6.3)393 (58.7)277 (41.3)   

We generated two predictive models: a baseline model using clinicopathological variables (Fig. 1) and another one using clinicopathological variables and the genetic score, constructed from the four SNPs independently associated with EBCR (Fig. 2). Both models showed good discrimination. The AUC of the baseline model was 0.728 (95% CI 0.674–0.784), and the AUC of the model with the genetic score was 0.763 (95% CI 0.708–0.817) (Table 4). The latter showed a significant increase in discrimination ability compared with the baseline model (AUC difference, 0.034, P = 0.025) [13] (Fig. 3).

figure

Figure 1. Predictive clinicopathological nomogram of early biochemical recurrence (EBCR) for clinically localized prostate cancer, within 1 year of radical prostatectomy. psa, prostate-specific antigen.

Download figure to PowerPoint

figure

Figure 2. Predictive clinicopathological-genetic nomogram of early biochemical recurrence (EBCR) for clinically localized prostate cancer, within 1 year of radical prostatectomy. PSA, prostate-specific antigen; SNP, single nucleotide polymorphism; the SNP represents the weighted risk score built from the SNP information.

Download figure to PowerPoint

figure

Figure 3. Receiver operator curves of predictive models (red line: clinicopathological model; blue line: clinicopathological-genetic model).

Download figure to PowerPoint

Table 4. Multivariate models to predict the probability of early biochemical recurrence within 1 year of radical prostatectomy
VariablesOR95% CIP value
  1. OR, odds ratio; 95% CI, 95% confidence interval; T, thymine; G, guanine; C, cytosine; A, adenine.

Variables included in the clinicopathological model   
Preoperative PSA (log), ng/mL1.851.42–2.41<0.001
Biopsy Gleason sum  <0.001
7 vs <72.371.40–4.010.001
>7 vs <76.152.57–14.74<0.001
Clinical stage  0.018
T2a–T2b vs T11.91.11–3.230.019
T2c vs T12.381.26–4.510.008
Variables included in the clinicopathological-genetic model   
Preoperative PSA, ng/mL1.831.40–2.40<0.001
Biopsy Gleason sum  <0.001
7 vs <72.401.40–4.120.001
>7 vs <76.592.69–16.12<0.001
Clinical stage  0.017
T2a–T2b vs T11.881.09–3.250.023
T2c vs T12.341.21–4.530.011
SNP genotyping2.071.50–2.87<0.001
KLK3 genotype (TT vs TG or GG)0.290.11–0.770.013
KLK2 genotype (CT or TT vs CC)2.041.02–4.100.044
SULT1A1 genotype (GG vs GA or AA)0.470.21–1.070.070
BGLAP genotype (TC or CC vs TT)2.481.15–5.330.020

We performed an internal validation using 10 000 bootstrap samples with similar proportion of EBCR to the original database, using the procedure described by Harrell et al. [16]. Bias-corrected AUCs for the two models were 0.714 and 0.748, respectively.

The improvement of the model with the genetic score was analysed through the net reclassification improvement category-free [17] and the integrated discrimination improvement. The analysis showed a net reclassification improvement increase of 60.7% for events and a decrease of 63.5% for non-events (P < 0.001), and an integrated discrimination improvement superiority of the model with the genetic score (P < 0.001).

Both models showed good calibration (Figs 4 and 5). Although calibration is not a good metric for model comparison [18], better calibration was found in the high probability range for the model with the genetic score. Finally, decision-curve analysis showed a superior clinical benefit of the model with the genetic score, particularly for intermediate-risk patients for whom classic predictive models are least accurate (Fig. 6).

figure

Figure 4. Calibration plot of clinicopathological model of early biochemical recurrence within 1 year of radical prostatectomy.

Download figure to PowerPoint

figure

Figure 5. Calibration-plot of clinicopathological-genetic model of of early biochemical recurrence within 1 year of radical prostatectomy.

Download figure to PowerPoint

figure

Figure 6. Decision-curve analysis of clinicopathological (Model 1) and clinicopathological-genetic (Model 2) models. SNP, single nucleotide polymorphism.

Download figure to PowerPoint

Discussion

  1. Top of page
  2. Abstract
  3. Introduction
  4. Patients and Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. Conflict of Interest
  9. References
  10. Supporting Information

We report the identification of four common SNPs located in KLK2, KLK3, SULT1A1 and BGLAP genes, independently associated with the risk of EBCR. In parallel, we developed a model to predict EBCR within 1 year of RP, based on classic preoperative clinicopathological variables. The model showed high discrimination capacity and correct calibration, and confirmed the predictive ability of clinical variables included in previously published nomograms [19-21]. In addition, the incorporation of a genetic score, based on those four SNPs, resulted in a significant improvement of the model in terms of both discrimination and calibration. The gain in accuracy was more noticeable in intermediate-risk and high-risk patients. The decision-curve analysis showed a greater net benefit at the same threshold as the clinical-genetic model, in agreement with the results of net reclassification improvement and integrated discrimination improvement.

The detection of systemic recurrences after RP is not feasible with objective techniques such as bone scans or CT scans, until advanced stages and a long time after RP (i.e. 8 years) [22]. To maximize the probability of having patients with systemic recurrences, we defined patients with EBCR as patients with detectable PSA in their first control after RP, or with rising PSA levels during the first year after RP [3-5]. Therefore, we used EBCR as a surrogate for a high probability of systemic recurrence. Lymph node/seminal vesicle involvement or high prostatectomy Gleason score could also be considered as risk factors of systemic recurrence [23], but they are not known preoperatively.

EBCR is associated with metastases [22] and PCa-specific mortality [24, 25]. For this reason, accurate EBCR risk assessment is critical. With this aim, a postoperative nomogram to predict EBCR within 2 years of RP was reported [26]. However, the model was based on postoperative variables and is not intended for preoperative use. In contrast, we have developed a model based on preoperative variables and germline genetic variants to predict the risk of EBCR within 1 year of RP. Patients at high risk of EBCR might be eligible for radiotherapy with concomitant hormonal therapy or subjected to RP in early adjuvant systemic treatment clinical trials, despite conflicting results [27, 28]. Hence, our predictions could assist clinicians in disallowing RP alone or considering multimodal approaches or early adjuvant therapies.

Patients with pathological features associated with local recurrence (e.g. pT3a/pTxR1) could have been excluded to best evaluate the associations with systemic recurrence. However, pathological features of those patients were not known preoperatively, so their exclusion would have prevented use of the nomogram as a preoperative tool. Hence, all patients were included irrespective of their pathological features, as in most published preoperative nomograms [29]. Instead, our study focused on EBCR within the 1st year to minimize the chances of including local recurrences.

Our observed rate of EBCR (13.3%) within 1 year of surgery is close to the 13.1% and 8.9% within 2 years, reported by Walz et al. [26]. The BCR rate within 1 year in those cohorts is unknown. Despite using a more sensitive definition of BCR (0.1–0.2 vs 0.4 ng/mL), Walz et al. [26] reported lower BCR rate than that observed in our study. Several factors might explain this: (i) an artefactually high proportion of recurrence due to retrospective recruitment of patients; (ii) a lower rate of non-palpable T1c tumours in our cohort than in the others (49.6% vs 62.8% and 66.7%), which suggests a poorer prognosis of our patients; (iii) patients having EBCR with PSA level >0.1–0.2 ng/mL within 2 years are likely to reach 0.4 ng/mL in a short period of time, suggesting similarity between these series; (iv) differences in ethnic composition between cohorts may involve different inherited genetic or environmental risk factors influencing the discrepancy. The latter hypothesis emphasizes the interest of studying the genetic contribution in PCa prognosis.

To date, D'Amico risk classification [19], the University of California, San Francisco Cancer of the Prostate Risk (UCSF-CAPRA) score [20], and the Stephenson nomogram [21] can be cited among the main preoperative models for prediction of BCR [30]. Those nomograms incorporate the same clinical variables as our model, except for the last two, which include the number of positive cores (non-available in our cohort) and the UCSF-CAPRA, which includes the age (non-significant in our analysis). We observed 76% of cases with biopsy Gleason score <7. Traditionally, high-risk patients are more frequently given radiotherapy with hormone therapy rather than radical prostatectomy in our healthcare setting [5]. This may have resulted in a slight enrichment in not so aggressive disease in our cohort. However, despite being higher, our 76% is very close to the percentage of Gleason score <7 reported in other cohorts from widely validated preoperative nomograms (e.g. 72%, 68%, 70% and 74%, for D'Amico, Stephenson, Walz, and UCSF-CAPRA, respectively). Hence, we consider that this issue would not jeopardize the applicability of our nomogram. Another model incorporating immunofluorescent biomarkers has been reported [31]. None of those models predicts BCR within 1 year of RP, which makes their comparison with our model difficult. Nevertheless, the c-index reported for their external validations (D'Amico [19, 32] 65.5–70.4%; UCSF-CAPRA [20, 31] 68–81%; Stephenson [21, 31] 75.2–79%; Donovan [30] 73%), and that obtained in our clinicopathological model (original 72.8%; bootstrap-corrected 71.4%) confirms a highly similar discrimination ability. Interestingly, the addition of genetic variables to the clinicopathological model resulted in a significant improvement in discrimination ability (original 76.3%; bootstrap-corrected 74.8%).

The improvement achieved by including the SNPs is modest, albeit consistently significant across all tests evaluated. The consistent improvement observed shows how genetic factors can enhance the accuracy of PCa prognostic models.

We have previously reported the usefulness of common SNPs for postoperative 5-year BCR predictions [10]. In the present study, we have identified four SNPs independently associated with the preoperative risk of EBCR within 1 year. These findings complement our previous results on postoperative long-term BCR predictions [10]. Two of those SNPs, located at the KLK2 and SULT1A1 genes, were also identified in the postoperative study whereas the other two, on KLK3 and BGLAP genes, were not. A potential reason for that is that certain germline SNPs may contribute to specific histological phenotypes, which once expressed, are reflected in the pathological variables preventing the causal SNPs from remaining in the models. Although that may be more obvious in the postoperative model, SNPs could also have an impact in preoperative variables. The SNPs on KLK2, KLK3 and BGLAP genes, have been previously associated with aggressive disease by different authors, which supports our results. In contrast, the SNP on SULT1A1 has been associated with PCa risk but, to our knowledge, not with PCa aggressiveness. Hence, validation of the latter in external cohorts would help to confirm our results.

The KLK3 gene encodes PSA, a prostate-specific and androgen-induced protease. Several SNPs throughout the kallikrein gene region on chromosome 19q13.33 have been consistently associated with PCa risk, aggressiveness and PCa-specific mortality [10, 33, 34]. One of those SNPs (–5429T/G, rs2569733), which belongs to a major linkage block in the upstream enhancer region of KLK3, has been significantly associated with increased PSA levels and PCa risk. Conversely, we found that carriers of the G allele had a decreased risk of BCR. Some authors have reported that this association might be due to a PSA bias [35]. Individuals with the PSA allele might be biopsied earlier because of increased PSA levels, and so, have lower Gleason score and less aggressive PCa. Therefore, the SNP may not be aetiologically implicated in PCa [34]. However, this hypothesis only partially explains the observed associations [7]. Indeed, in our study we could not find an association between SNP rs2569733 in KLK3 gene and histological grade or clinical stage. Moreover, we found that the SNP was independently associated with the risk of EBCR, and the SNP remained significant in the multivariate model that included PSA, Gleason score and stage. In agreement with our results, Gallagher et al. [7] reported that the association of SNPs in the KLK3 gene with PCa-specific mortality remained significant in a model that also included PSA and stage, which strengthens the hypothesis that this locus may play a biological role in PCa aggressiveness.

Another human kallikrein is hK2 protein (kallikrein-related peptidase 2), which is codified by the KLK2 gene. We have analysed a non-synonymous polymorphism at codon 250 of the KLK2 gene (Arg/Trp, rs198977). This functional SNP maps at 19q13.4 chromosome, close to one of the most well-established susceptibility loci for PCa6 [36]. We found that the T allele was also associated with increased risk of EBCR.

Sulphotransferase 1A1 activates dietary carcinogens and metabolizes protective agents [37]. The SNP rs9282861 on the SULT1A1 gene (Arg213His, SULT1A1*1/SULT1A1*2) leads to decreased enzyme activity and thermostability [38]. Decreased SULT1A1 levels and enzymatic activity have been associated with decreased PCa risk [39]. Consistently, we found that carriers of the SULT1A1*2 allele had a decreased risk of EBCR, suggesting a protection against PCa progression.

Overexpression of the osteocalcin (BGLAP) gene has been reported in metastatic bone tumours, including PCa [40]. One SNP in the promoter region of the BGLAP gene (–198T/C, HindIII, rs1800247) has been associated with PCa risk [38]. Our study showed that patients carrying this variant were at increased risk of EBCR.

Patients at high risk of EBCR are more likely to develop metastatic disease. Considering the costs and decreased quality of life derived from metastatic disease, the most cost-effective strategy for the management of high-risk patients is the one that maximizes progression-free survival [41]. This highlights the need for improved risk classification methods. In a previous study, Zubek et al. [42] show the cost-effectiveness of a tissue-based protein assay for the prediction of BCR. Compared with such tests (i.e. tissue-based or serum-based patterns of protein or RNA expression), SNP-based tests have become cheaper and faster in recent years, and so more suitable for clinical routine testing [43]. Therefore, considering the reduced cost of an SNP assay, and the increased clinical benefit derived from its use, the cost-effectiveness of the new test presented is warranted.

Despite the significant improvement achieved by the inclusion of those four SNPs, there is still scope for progress. Most of the candidate SNPs analysed in our study had been originally associated with PCa risk rather than PCa progression. Hence, the potential phenotypic impact of those SNPs on the clinicopathological variables could have partially obscured their contribution to the predictive accuracy of said variables. In this respect, we hypothesize that the analysis of a panel of SNPs on genes specifically associated with the ability of PCa circulating cells to migrate, or to their tissue-specific tropism, such as nodes or bone, could further improve the prediction of EBCR. The finding of an osteocalcin gene polymorphism associated with EBCR in our study, and the known involvement of this gene in bone metastasis, could support this hypothesis.

Several limitations may apply to our study. First, we developed a multicentric retrospective study with its potential intrinsic limitations. For example, certain EBCR risk factors were not available in our cohort (e.g. number of positive cores) or were not equally recorded. Indeed, a systematic review and re-grade of all cases by a single uropathologist might have helped to increase the homogeneity of our data and the applicability of the results to contemporary cohorts. Nevertheless, a recent comparison of the predictive accuracy of 2001 Partin tables versus a new preoperative nomogram based on PSA level, disease stage and Gleason score, complying with the 2005 International Society of Urological Pathology consensus on Gleason grading, showed roughly similar performance in a series of 1188 patients with PCa [44], which suggests that this limitation may not substantially reduce the clinical utility of the new nomogram. Nonetheless, further validation of the nomogram in external, contemporary series would be desirable as it would strengthen our results. Second, the effect of unfavourable alleles in ethnically different populations should be confirmed. Third, it would be interesting to explore whether newly identified PCa susceptibility loci [10] could further improve the prediction of EBCR. Finally, despite 1-year BCR risk being a clinically relevant endpoint, the impact of those SNPs on metastasis and PCa-specific mortality may also be analysed in future studies.

In conclusion, we have developed a new nomogram for preoperative prediction of BCR at 1 year using clinicopathological and genetic risk factors. The addition of genetic polymorphisms significantly improves the predictive power of classic nomograms, and opens the way for adding new biomarkers in future updates.

Acknowledgements

  1. Top of page
  2. Abstract
  3. Introduction
  4. Patients and Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. Conflict of Interest
  9. References
  10. Supporting Information

Dr María J. Viso from the Department of Biochemistry, ‘Miguel Servet’ University Hospital, Zaragoza, Spain, assisted with the collection of blood samples.

Conflict of Interest

  1. Top of page
  2. Abstract
  3. Introduction
  4. Patients and Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. Conflict of Interest
  9. References
  10. Supporting Information

Ä€ngel Borque is a paid Consultant to the Sponsor; Jokin Del Amo, Diego Tejedor, Laureano Simon and Antonio Martínez are Employees of the Sponsor and Patent Inventors; Juan Morote is Patent Inventor; Marta Artieda is Employee of the Sponsor.

References

  1. Top of page
  2. Abstract
  3. Introduction
  4. Patients and Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. Conflict of Interest
  9. References
  10. Supporting Information

Supporting Information

  1. Top of page
  2. Abstract
  3. Introduction
  4. Patients and Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. Conflict of Interest
  9. References
  10. Supporting Information
FilenameFormatSizeDescription
bju11333-sup-0001-si.doc111K

Table S1 Description of the single nucleotide polymorphisms (SNPs) analysed by the DNA microarray.

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.