The first two authors contributed equally to this work.
Early Detection and Diagnosis
Evaluation of molecular forms of prostate-specific antigen and human kallikrein 2 in predicting biochemical failure after radical prostatectomy
Article first published online: 9 SEP 2008
Copyright © 2008 Wiley-Liss, Inc.
International Journal of Cancer
Volume 124, Issue 3, pages 659–663, 1 February 2009
How to Cite
Wenske, S., Korets, R., Cronin, A. M., Vickers, A. J., Fleisher, M., Scher, H. I., Pettersson, K., Guillonneau, B., Scardino, P. T., Eastham, J. A. and Lilja, H. (2009), Evaluation of molecular forms of prostate-specific antigen and human kallikrein 2 in predicting biochemical failure after radical prostatectomy. Int. J. Cancer, 124: 659–663. doi: 10.1002/ijc.23983
- Issue published online: 18 NOV 2008
- Article first published online: 9 SEP 2008
- Accepted manuscript online: 9 SEP 2008 12:00AM EST
- Manuscript Accepted: 14 AUG 2008
- Manuscript Received: 13 MAY 2008
- National Cancer Institute. Grant Numbers: P50-CA92629, SPORE Pilot Project 7
- European Union 6th Framework. Grant Number: LSHC-CT-2004-503011
- Swedish Cancer Society. Grant Number: 3555
Vol. 127, Issue 11, E2, Article first published online: 28 SEP 2010
- free PSA;
- human kallikrein 2;
- prostate cancer;
- biochemical recurrence;
- radical prostatectomy
Most pretreatment risk-assessment models to predict biochemical recurrence (BCR) after radical prostatectomy (RP) for prostate cancer rely on total prostate-specific antigen (PSA), clinical stage, and biopsy Gleason grade. We investigated whether free PSA (fPSA) and human glandular kallikrein-2 (hK2) would enhance the predictive accuracy of this standard model. Preoperative serum samples and complete clinical data were available for 1,356 patients who underwent RP for localized prostate cancer from 1993 to 2005. A case-control design was used, and conditional logistic regression models were used to evaluate the association between preoperative predictors and BCR after RP. We constructed multivariable models with fPSA and hK2 as additional preoperative predictors to the base model. Predictive accuracy was assessed with the area under the ROC curve (AUC). There were 146 BCR cases; the median follow up for patients without BCR was 3.2 years. Overall, 436 controls were matched to 146 BCR cases. The AUC of the base model was 0.786 in the entire cohort; adding fPSA and hK2 to this model enhanced the AUC to 0.798 (p = 0.053), an effect largely driven by fPSA. In the subgroup of men with total PSA ≤10 ng/ml (48% of cases), adding fPSA and hK2 enhanced the AUC of the base model to a similar degree (from 0.720 to 0.726, p = 0.2). fPSA is routinely measured during prostate cancer detection. We suggest that the role of fPSA in aiding preoperative prediction should be investigated in further cohorts. © 2008 Wiley-Liss, Inc.
Other than its established clinical application for the early detection of prostate cancer, prostate-specific antigen (PSA) represents one of the key variables in currently available prognostic models for clinically localized prostate cancer.1, 2 One of the applications of such models allows for assessment of risk of disease recurrence after local therapy. However, levels of total PSA (tPSA) in blood do not solely refer to the presence of cancer, but are also driven by benign prostatic hyperplasia or inflammatory processes. Furthermore, Stamey et al.3 have shown that in men with moderately elevated levels of tPSA (<10 ng/ml), PSA does not aid in predicting biochemical cure rates. This limits the application of tPSA as a predictor of stage and disease progression in populations in which PSA is regularly used for screening. Hence, there's a compelling need for additional biochemical markers to improve detection of clinically significant, biologically aggressive prostate cancer.
One such marker is human kallikrein-related peptidase 2 (hK2), a product of the human KLK2 gene and a serine protease that shares 80% sequence homology with PSA (hK3). The enzymes share the property of being expressed chiefly in the prostate under androgen regulation.4 Recent studies point to substantial diagnostic capacity of molecular isoforms of PSA and hK2 for prostate cancer detection and disease staging.5 Furthermore, hK2 has been shown to provide independent prognostic information over other established preoperative parameters. Previous data demonstrated that increasing levels of hK2 and a low ratio of free PSA (fPSA) to tPSA are linked to advanced prostate pathology, which is a well-established risk factor for disease progression after surgery.6, 7 Subsequently, it was hypothesized that hK2 may be a useful biomarker especially for the detection of biologically aggressive prostate cancers.
Given the utility of current risk models to predict disease progression, rather than replacing these models, the aim of this study was to determine whether the predictive accuracy of standard preoperative risk models for biochemical recurrence (BCR) can be enhanced with the addition of molecular PSA isoforms (fPSA) and hK2.
Material and methods
Patients and serum samples
Between January 1993 and December 2005, 4,737 patients with clinically localized prostate cancer (T1, T2 or T3 clinical tumor stage) underwent radical prostatectomy (RP) with staging lymphadenectomy at a single institution (Memorial Sloan-Kettering Cancer Center, New York, NY). Under an institutional review board-approved protocol investigating biochemical markers in prostate cancer, blood samples were obtained for 2,835 of these patients. Blood samples were drawn before RP 4 or more weeks after any prostatic manipulation (DRE, TRUS-guided biopsy), and immediately processed and frozen at −80°C until analysis. Patients with any neoadjuvant therapy (n = 376), prior surgical or medical treatment for benign prostatic hyperplasia (n = 45) or lack of clinical data (n = 1,058) were omitted from this study, leaving 1,356 patients with corresponding blood samples eligible for analysis. There were 146 recurrences and the median follow up for patients without recurrence was 3.2 years. The protocol was approved by the institutional review board and human tissue utilization committee, and informed consent had been obtained from all of the participating patients.
Clinical and pathologic evaluation
Definition of BCR
BCR was defined as postoperative levels of tPSA ≥ 0.40 ng/ml. The selection of this cut point was based on a previous evaluation, which demonstrated that a significant proportion of patients with a tPSA < 0.40 ng/ml did not experience further PSA rises.11 None of the patients had received any adjuvant therapy before evidence of cancer recurrence.
Measurements of biomarkers
Total and fPSA
To measure tPSA and fPSA, we used the commercial version of a previously reported dual-label assay (DELFIA Prostatus Dual Assay, PerkinElmer, Turku, Finland) that measures tPSA and fPSA on an equimolar basis.12 Detection limits are 0.04 ng/ml for fPSA (coefficient of variation [CV] 3.7% at 0.44 ng/ml and 17.9% at 0.10 ng/ml), and 0.05 ng/ml for tPSA (CV 5.0% at 2.32 ng/ml and 13.9% at 0.34 ng/ml). The percentage of fPSA (%fPSA) was calculated as %fPSA = fPSA/tPSA × 100.13–15
Total hK2 was measured according to a previously reported 3-step in-house assay design. It uses a combination of PSA-specific blocking monoclonal antibodies (6H10) together with capture- and Europium-labeled detection F(ab′)2 recognizing both PSA and hK2. Addition of PSA-specific blocking monoclonal antibodies enables specificity for hK2 as it eliminates cross-reaction of PSA to <0.001%. Selective hK2 detection was performed by using F(ab′)2 11B6 as a capture antibody, which is entirely specific for hK2, and Eu-labeled as a tracer. The functional detection limit is 0.003 ng/ml, defined as the concentration at which total assay imprecision ranged from 12.9% for low (0.0033 ng/ml) to 7.5% for high (0.48 ng/ml) hK2 levels.
fPSA and hK2 were not measured in a clinical setting, requiring frozen serum samples to obtain these measurements. To preserve frozen samples that need not be measured, we used a nested case-control study design matching on age and year of surgery. This design has been shown to be appropriate for determining associations and obtaining unbiased estimates of predictive accuracy.16 Therefore, the results currently presented would not importantly differ from that obtained from a Cox regression model. Cases were defined as patients with evidence of BCR at last follow up. Three controls were randomly selected for each case from the patient population at risk at the time of BCR. The patient population at risk for a given case included all patients at risk (alive and without BCR) at the time at which the case recurred. By this definition, another case is a potential control if he had a longer time to BCR. In other words, a patient who served as a control at a certain time point for a certain case could become a case once fulfilling the criteria for BCR himself. For the subgroup analysis of patients with moderately elevated tPSA levels, we re-matched within the patients with tPSA ≤ 10 ng/ml using the same methodology described above. The sttocc command in Stata was used to convert these survival-time data to case-control data.
Conditional logistic regression models were used to evaluate the association between predictors and BCR after RP. All biomarkers were entered as log-transformed continuous variables to model their nonlinear association with outcome. The base multivariable model included log tPSA, clinical stage (categorized as ≤ T2a and ≥ T2b), and biopsy Gleason grade (categorized as ≤6, 3+4, 4+3 and ≥8). We then constructed multivariable models with log hK2 or log fPSA as additional predictors. Predictive accuracy was defined in terms of the area under the ROC curve (AUC). All AUCs were calculated using 10-fold cross-validation. Confidence intervals for the difference in AUC from the base model were obtained using bootstrap methods, resampling within case-control groups, with 1000 replications. Subgroup analyses focused on men with moderately elevated PSA levels, who are more typical of contemporary patients with prostate cancer in the United States and in other countries, where tumors are generally detected by PSA screening.
The agreement between laboratory and clinical tPSA was assessed using the concordance correlation coefficient, which evaluates the degree to which pairs of observations are equivalent, and linear regression. All analyses were performed using Stata 9.2 (Stata, College Station, TX).
Overall, there were 146 cases with BCR and 436 controls (Supporting Table I). Of the 146 cases, 144 (99%) were matched to 3 controls. The remaining 2 cases (1%) were matched to only 2 controls because a total of 3 controls with matching age and year of surgery could not be found for them. Since age was a matching criterion, the age distribution between cases and controls was similar. Age was within 5 years of the matched case for all but 9 controls (2%). Year of surgery was matched exactly for all but 6 controls; these 6 controls varied from their case by 1 year of surgery. Clinical and pathological tumor characteristics were more unfavorable in cases than controls. For example, 63% of cases (n = 92) had a biopsy Gleason score >6 compared with only 23% of controls (n = 100); additionally, 57% of cases (n = 83) had advanced pathologic stage (>pT2) compared with only 22% of controls (n = 96).
Levels of tPSA, fPSA and hK2 are shown in Supporting Table II for the entire cohort and for patients with tPSA ≤ 10 ng/ml. Median levels of all biomarkers were higher in cases than controls. In the subgroup analyses for tPSA ≤ 10 ng/ml there were 70 cases and 209 controls. One case was matched to only 2 controls; the rest were matched to 3 controls. Overall, there were 34 patients who developed prostate cancer metastases, and 12 of them died of prostate cancer during the follow up. However, as the numbers were small we did not perform detailed analyses for these outcomes.
In univariate analyses, all variables were significantly associated with BCR in the entire cohort. When applied to the subgroup of patients with tPSA ≤ 10 ng/ml, only clinical stage and biopsy Gleason score were statistically significant predictors of BCR (Table I). Previously established predictors (stage, grade and tPSA) were combined in a prediction base model (Table II). The predictive accuracy of this model was high, with an AUC of 0.786 in the entire cohort. Adding fPSA and hK2 as predictors to the base model marginally enhanced the predictive accuracy to 0.798 in the entire cohort (difference in AUC 0.012; 95% confidence interval −0.002, 0.032). Of note, a statistically significant enhancement was observed when only fPSA was added to the base model (difference in AUC 0.013; 95% confidence interval 0.0004, 0.029). In the subgroup of men with tPSA ≤ 10 ng/ml, the addition of fPSA and hK2 enhanced the predictive accuracy of the base model to a similar degree (from 0.720 to 0.726; difference in AUC 0.006; 95% confidence interval −0.019, 0.074).
|Univariate||Multivariable, full model|
|Odds ratio||95% CI||p||Odds ratio||95% CI||p|
|Cases = 146, controls = 436|
|T2b/c/T3||2.27||1.53, 3.37||1.34||0.80, 2.26|
|Biopsy Gleason score|
|3+4||1.49||0.92, 2.07||4.46||2.36, 8.44|
|4+3||2.74||1.97, 3.51||13.2||5.56, 31.3|
|≥8||2.49||1.71, 3.26||11.61||4.89, 27.6|
|log total PSA||2.75||2.04, 3.73||<0.0005||4.49||2.63, 7.67||<0.0005|
|log free PSA||1.42||1.09, 1.85||0.009||0.43||0.26, 0.71||0.001|
|log hK2||1.74||1.37, 2.21||<0.0005||1.20||0.87, 1.66||0.3|
|tPSA ≤10 ng/ml, cases = 70, controls = 209|
|T2b/c/T3||2.17||1.26, 3.72||1.47||0.76, 2.85|
|Biopsy Gleason score|
|3+4||0.64||−0.20, 1.47||1.84||0.78, 4.35|
|4+3||2.69||1.48, 3.90||11.44||3.25, 40.3|
|≥8||2.39||1.33, 3.44||10.18||3.27, 31.7|
|log total PSA||1.45||0.86, 2.47||0.17||2.6||1.14, 5.92||0.02|
|log free PSA||0.85||0.56, 1.28||0.4||0.51||0.25, 1.06||0.07|
|log hK2||1.12||0.82, 1.52||0.5||1.09||0.74, 1.61||0.6|
|AUC||95% CI for difference from base||p value for difference from base|
|Cases = 146, controls = 436|
|+free PSA||0.799||0.0004, 0.029||0.017|
|+free PSA & hK2||0.798||−0.002, 0.032||0.053|
|tPSA ≤ 10 ng/ml, cases = 70, controls = 209|
|+free PSA||0.737||−0.010, 0.072||0.13|
|+free PSA & hK2||0.726||−0.019, 0.074||0.2|
We also applied these analyses to a different prediction model, which also includes tPSA, but is mainly based on postoperative pathological parameters: RP specimen Gleason score, extracapsular extension, seminal vesical invasion, lymph node invasion and surgical margin status (Supporting Table III). No enhancement in the predictive accuracy could be shown when fPSA or hK2, or both were added to this prediction model (Supporting Table IV).
Comparison of serum levels of tPSA between fresh and frozen samples
As described above, blood samples analyzed in this study were drawn before RP 4 or more weeks after any prostatic manipulation, and subsequently immediately processed, aliquotted and frozen at −80°C until analysis in this study. At the same time, tPSA levels in serum were determined in the clinical setting before patients underwent RP. We performed a subanalysis, in which we compared the concordance of levels of tPSA in both samples: (i) in the prospectively collected, aliquotted and frozen samples that were then thawed and retrospectively measured (termed laboratory samples), and (ii) in the clinical samples that were analyzed for tPSA immediately on the day of venipuncture. The reason for this subanalysis was to address the question and hypothesis of possible biodegradation of PSA in the frozen laboratory samples. In a total of 427 patients, the date of the last clinical PSA measurement before RP and the date of the aliquot of the frozen laboratory sample were the same (i.e., the fresh sample, analyzed in the clinical setting, and the frozen sample, analyzed in this study, originated from the same initial venipuncture). Linear regression analysis demonstrated a statistically significant slope of 1.22, with an intercept of −0.49, and r2 = 0.9540, when tPSA values of clinical samples were compared with the corresponding laboratory samples (Supporting Fig. 1A); the concordance correlation coefficient was 0.94. Although the intercept of −0.49 showed the possibility of bias, this was highly influenced by 11 patients with outlying tPSA values (>30 ng/ml); when these patients were excluded from the linear regression, the slope was 1.16 and the intercept was −0.055 (Supporting Fig. 1B).
This subgroup of 427 patients with matching laboratory and clinical samples translated to 137 cases (94%) and 389 matched controls for the case-control analysis. The predictive accuracy of the base model for this subgroup of patients, when calculated with the tPSA values that were obtained from the laboratory samples, showed an AUC of 0.782, vs. an AUC of 0.773 when calculated with tPSA levels from the clinical samples. Therefore, we can be confident that our conclusions were not affected by biodegradation of tPSA.
In this study, we examined the ability of PSA isoforms (fPSA) and hK2 to improve upon previously validated risk-prediction models. Our findings suggest that an improvement in the AUC can be achieved by adding both fPSA and hK2 to the standard nomogram that contains only clinical stage, biopsy Gleason score and tPSA.
Population-based data have shown that levels of PSA and hK2 are significantly elevated in blood up to 20 years before men were diagnosed with clinically significant prostate cancer.17 It is also believed that the tight compartmentalization of PSA and hK2 in the normal prostate gland is altered in prostatic disease. The disruption of the continuous layer of basal cells results in loss of the normal glandular architecture and allows substantial leakage of various proteins into circulation.18, 19 This development suggests that detection of increased extracellular kallikrein forms offer very sensitive means to detect changes in prostate tissue architecture that may signify prostate cancer progression.20 The covariance of hK2 and PSA concentrations in blood has been determined to be generally <60%, suggesting that hK2 might have utility as an independent biomarker for prostate cancer.21
Several studies have evaluated hK2 as a prognostic tumor marker. Darson et al.22 examined paraffin specimens from RP and found that hK2-specific immunostaining in lymphatic metastases and in high-grade tumors was more intense compared with well-differentiated tumors and benign tissues. Serum hK2 measurements were significantly elevated in locally advanced tumors compared with tumors that were organ confined.23, 24 Furthermore, on univariate analysis, hK2 alone was shown to be a significantly superior predictor of BCR compared with tPSA (concordance index of 0.739 for total hK2 vs. 0.566 for tPSA, p < 0.0005). Combining tPSA, fPSA and total hK2 in a multivariable model improved prediction compared with use of each biomarker individually.7 We have previously reported that pretreatment measurements of hK2 and fPSA were helpful in identifying patients at risk for biochemical failure after RP.5 Overall, the predictive accuracy of the base model (preoperative tPSA, clinical stage and biopsy Gleason grade) was not improved after the addition of fPSA or hK2 (0.813 vs. 0.818). However, for men with moderate tPSA elevation (tPSA ≤ 10 ng/ml), addition of fPSA and hK2 data increased the predictive accuracy (from the base model concordance index of 0.756–0.815, p = 0.005).5 In our present study, although addition of fPSA and hK2 improved predictive accuracy for the entire cohort as well, the improvement was less robust (0.786–0.798). Additionally, in the cohort mentioned above,5 hK2, but not fPSA, was significantly associated with BCR on multivariable analysis in the subgroup of patients with a PSA of <10 ng/ml. In our cohort, we found the opposite: fPSA, but not hK2, was significantly associated with BCR. We believe that our results support the hypothesis that the combination of fPSA and hK2 (in addition to tPSA, stage and grade) can enhance the predictive accuracy of BCR. We do not know, however, whether the main contribution to this enhancement is due to fPSA or hK2.
One of the critical aspects for patients and physicians is to better understand the risk a given prostate cancer poses on an individual basis. Currently available pretreatment staging tools can help in the decision-making process, but may lack specificity. We believe that the addition of new variables, such as serum peptide protein profiling, may improve our ability to better classify the biologic potential of individual prostatic tumors, and to distinguish between indolent and aggressive cancers. Improvement in the predictive accuracy may imply enhanced ability to counsel patients before treatment about their risk of BCR.
Several limitations may have influenced the validity of our findings. As suggested in multiple prior analyses on the value of fPSA and hK2, it is possible that our data may at least be partly influenced by suboptimal sample collection (we analyzed serum, not plasma), storage, and preanalytical processing, although the serum measurements were attempted to be made reliable and reproducible by following standard and precise processing protocols (separation of serum from blood cells immediately after venipuncture, subsequent immediate freezing at −80°C, analysis of samples immediately after thawing at 4°C), as well as limiting interference by replacing our monoclonal capture and tracer antibodies with recombinant Fab fragments.14 Hence, limited stability of fPSA and hK2 after venipuncture, reported also for long-term frozen serum samples as used in our analyses, instead of fresh plasma samples, may have influenced the accuracy of measurements of the analytes.25, 26 To address this question, we performed linear regression analysis for tPSA levels that were determined in the clinical setting, in comparison to the tPSA values in the frozen aliquots that were obtained through immunoassay measurements in this study. As mentioned, after excluding influential observations, this subanalysis demonstrated a statistically significant slope of 1.16, with an intercept of −0.055. In other words, a man with a PSA of 1 ng/ml in the clinical sample would have a PSA of 0.95 ng/ml in the frozen laboratory sample, and then each additional 1-ng/ml increase in PSA level in a clinical sample would correspond to a 1.16-ng/ml increase in PSA level in the frozen laboratory sample.
Multiple factors have to be taken into consideration when trying to interpret these findings: First, biodegradation of the protein (PSA) itself could certainly, at least to a limited degree, contribute to a decrease in the measurable level of PSA in a frozen sample. Nevertheless, this process is very unlikely to be represented in such a linear manner as we were able to demonstrate in our analysis. Second, patient samples were collected from 1993 to 2005. During those years, important changes and developments with regard to immunoassay techniques that were used to detect PSA in serum or plasma had been made: First assays that were used were less sensitive (i.e., detection limits of PSA were higher). But also simply interassay variability may be a factor that could explain the significant, but constant differences between PSA levels that were determined between the clinical samples and PSA levels that we measured in the aliquotted, frozen laboratory samples, which were analyzed much later and provided the data that went into the main analysis in this study. However, although the differences in the PSA levels in serum were significant, we were able to show a linear relationship, which would preclude biodegradation in the frozen samples being the major reason for these differences. Our most important overall conclusion on this subanalysis is that PSA values obtained through measurements of the laboratory samples are in fact highly representative of the clinical values. This conclusion may seem a very important result of our analysis, as this finding confirms the biostability of tPSA samples that are stored and frozen for analyses at later time points.
In conclusion, our study replicated previous findings that serum hK2 and fPSA were associated with BCR after local therapy. However, there was only a small increase in predictive accuracy when these markers were added to a standard multivariable model. This enhancement seemed to be largely contributed by fPSA. Given that fPSA is often routinely measured to aid detection of prostate cancer, incorporation in risk prediction models would be straightforward. Further studies are necessary to elucidate the role of these markers in predicting disease progression after local therapy.
- 5Risk assessment for biochemical recurrence prior to radical prostatectomy: significant enhancement contributed by human glandular kallikrein 2 (hK2) and free prostate specific antigen (PSA) in men with moderate PSA-elevation in serum. Int J Cancer 2006; 118: 1234–40., , , , , , , , , , .
- 8AJCC Cancer Staging Manual, 5th edn., Philadelphia, PA: Lippincott Raven Publishers, 1997., , , , , , .
- 16The statistical evaluation of medical tests for classification and prediction. Oxford Statistical Science Series: Oxford University Press, 2003. 25..
Additional Supporting Information may be found in the online version of this article.
|IJC_23983_sm_suppinfofigure1a.tif||342K||Supporting Figure 1. – (a) Comparison of clinical total PSA values, and “laboratory” total PSA values measured in this study (entire group); r2 = 0.9540, slope = 1.22, intercept/constant = −0.49; y = −0.49 + 1.22x. The thick black line is the fitted linear regression line; for comparison, the perfect relationship (clinical tPSA = laboratory tPSA) is shown as the think black line. The concordance correlation coefficient was 0.94.|
|IJC_23983_sm_suppinfofigure1b.tif||334K||Supporting Figure 1. – (b) Comparison of clinical total PSA values, and “laboratory” total PSA values measured in this study; excluding 11 outliers with a total PSA >30 ng/ml; r2 = 0.8574, slope = 1.16, intercept/constant = −0.055; y = −0.055 + 1.16x. The thick black line is the fitted linear regression line; for comparison, the perfect relationship (clinical tPSA = laboratory tPSA) is shown as the think black line. The concordance correlation coefficient was 0.88.|
|IJC_23983_sm_suppinfotables1-4.doc||122K||Supporting Table 1. Clinical and pathological characteristics of cases and controls Supporting Table 2. Median tPSA, fPSA, and hK2 levels among all patients and among patients with tPSA ≤10 ng/ml Supporting Table 3. Multivariable results of conditional logistic regression for prediction of biochemical recurrence following radical prostatectomy among all patients and among patients with total PSA ≤10 ng/ml, controlling for pathologic tumor characteristics Supporting Table 4. Accuracy (defined in terms of the area under the ROC curve [AUC]) of a postoperative pathological prediction base model (including prostatectomy specimen Gleason score, ECE, SVI, LNI, SM, and total PSA), compared to this base model supplemented by free PSA and/or hK2 levels|
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.