Early prediction of prostate cancer biochemical recurrence and identification of disease persistence using PSA isoforms and human kallikrein-2

1 Introduction

Prostate cancer (PC) affects approximately 12 percent of men during their lifetime [1] and is associated with a 3 percent risk of PC death [2]. Prostate-specific antigen (PSA), also known as human kallikrein-3 (hK3) is a biomarker secreted by the epithelial cells of the prostate gland that belongs to the family of serine proteases. Initially used in monitoring disease progression of PC patients, its use in screening, diagnosis, preoperative risk nomograms and follow up after primary therapy has become the current standard. After radical prostatectomy (RP), patients are surveilled by PSA measurements at three, six and twelve months, then every six months until three years, and then annually [3]. The level of PSA should be undetectable at 4–6 weeks postoperatively [4, 5]. A persistently elevated value after RP signals residual disease or micrometastases, while a fastly or slowly rising PSA during surveillance suggests metastases or local recurrence, respectively. The current definition of biochemical recurrence (BCR) after RP includes 2 consecutive PSA rises >  0.2 ng/ml and occurs in approximately 20–40%of patients [6, 7].

While the focus has been on developing new diagnostic markers for PC in the preoperative setting, that could either replace or enhance the performance of PSA, no alternatives have been approved in the postoperative period. Early prediction of BCR would be of paramount importance in planning adjuvant or salvage forms of therapy. The importance of identifying early patients at risk for BCR resides in the fact that the earlier radiation therapy (RT) is delivered postoperatively (either in adjuvant or salvage setting), the better seem to be the outcomes [8].

Although ultrasensitive PSA (uPSA) assays are able to detect minimal postoperative PSA levels of 0.001 ng/ml [9], the advantage of their use in clinical practice remains unclear.

New markers which have been extensively studied in the diagnostic setting of PC are recently being investigated as predictors of BCR in the preoperative setting. Among the studied markers are [–2]proPSA - a free-PSA (fPSA) isoform -and the derived formula Prostate Health index (PHI) [–2]proPSA/fPSA ^* √PSA. PHI has proved to be a valuable and approved diagnostic tool at both initial and repeat biopsy, enabling the prediction of significant PC [10, 11] and outperforming PSA and %fPSA in diagnostic accuracy in men with a PSA between 2–10 ng/ml [12].

Studies on the utility of PHI in the preoperative context confirmed its value as a predictor of BCR [13, 14], tumor volume [15] and pathologic outcomes [16, 17].

Another approved tool in PC diagnosis is the four kallikrein (4K) score test that includes the four parameters free PSA (fPSA), intact PSA, total PSA and human kallikrein-2 (hK2) and has shown excellent results in selecting patients with significant disease for prostate biopsy [18]. hK2 in particular belongs to the protein kallikrein family (such as PSA) and previous studies refer to its use in early detection of PC as an adjunct to PSA and fPSA [19], its value in BCR-prediction [20, 21] and the potential value of hK2/fPSA and hK2/%fPSA in detecting minimal PC disease [22].

To our knowledge, besides one study comparing the trends of [–2]proPSA and uPSA in the first year after RP in a high-risk population [23], and a second study [24] comparing the kinetics of [–2]proPSA and PSA in the first 3 years post-RP, no recent studies exist on the potential use of markers such as fPSA, %fPSA, hK2 and PHI to predict BCR and identify disease persistence (DP) in the postoperative setting.

The aim of our study was to investigate the use of these markers in both preoperative and early postoperative settings and examine their predictive power.

2 Materials and methods

3 Results

3.3 Univariate analysis of BCR and DP

Figures 1–3 depict the ROC curves obtained for BCR at the different time points using the different markers.

OPEN IN VIEWER

Fig. 1

ROC curves for all the markers at preoperative period and their relation to BCR. Red PSA (AUC 0.64), Blue [–2]proPSA (AUC 0.70), Green PHI (AUC 0.74), Brown hK2 (AUC 0.60), Violet fPSA (AUC 0.59), Black fPSA/PSA (AUC 0.60).

OPEN IN VIEWER

Fig. 2

ROC curves for all the markers at 1 month period and their relation to BCR. Red PSA (AUC 0.69), Blue [–2]proPSA (AUC 0.50), Green PHI (AUC 0.52), Brown hK2 (AUC 0.48), Violet fPSA (AUC 0.48), Black fPSA/PSA (AUC 0.59).

OPEN IN VIEWER

Fig. 3

ROC curves for all the markers at 3 months period and their relation to BCR. Red PSA (AUC 0.72), Blue [–2]proPSA (AUC 0.53), Green PHI (AUC 0.61), Brown hK2 (AUC 0.62), Violet fPSA (AUC 0.54), Black fPSA/PSA (AUC 0.62).

In the preoperative setting, the predictors of BCR were preoperative PSA (AUC 0.64; p-value 0.029), [–2]proPSA (AUC 0.70; p-value 0.002) and PHI (AUC 0.73; p-value 0.0003). Interestingly, the BCR predictive ability of PSA at diagnosis was surpassed by [–2]proPSA and, more importantly, PHI. In the postoperative period, PSA was the only marker that correlated with BCR, both at the time point 1 month (AUC 0.69; p-value 0.0047) and 3 months (AUC 0.72; p-value 0.0004). All other markers were devoid of ability to predict BCR in the postoperative period.

PSA (AUC 0.68; p-value 0.026), fPSA (AUC 0.66; p-value 0.048), PHI (AUC 0.75; p-value 0.002) and [–2]proPSA (AUC 0.73; p-value 0.006) were found to be significant predictors of DP in the preoperative period. Regarding early post-RP identification of DP, PSA, fPSA, [–2]proPSA and PHI were all statistically significant markers (p-value≤0.03).

So far, not much is known about the ideal follow up of patients after RP. Not only no other marker with the exception of PSA has been approved for disease surveillance and BCR identification, but also the schedule of PSA testing is a matter of debate, with some experts advocating a risk-adapted follow up versus the current ‘one-size-fits-all’ standard schedule of PSA measurements at 3,6,12 months after RP then semiannually until 3 years, and then annually [3].

After surgery, the level of PSA should drop to undetectable levels within 21–30 days, considering that the half-life of PSA is approximately 3.2 days [28]. Therefore, there is a window of opportunity for the early use of markers of disease recurrence and persistence.

An early and more accurate prediction of BCR and DP is of the utmost importance. It would have an impact on patient counseling, selecting patients for nuclear imaging techniques such as 68GaPSMA-11 Positron Emission Tomography and risk-adapting surveillance schedules. It could select a group of patients who would benefit of a stricter follow up protocol and would benefit from adjuvant or early-salvage therapy. Although adjuvant therapy is considered in patients with adverse pathological features (extracapsular extension, seminal vesicle invasion and positive surgical margins), an overtreatment risk of approximately 50%[29] should not be neglected. Early prediction of BCR would thus spare these patients from unnecessary morbidity.

Despite recent advances in PC diagnosis brought by PSA isoforms such as [–2]proPSA and PHI, their use in BCR prediction and in surveillance is virtually unexplored. Preoperative risk nomograms such as the MSKCC nomogram [25] which guides the surgeon in decision making concerning pelvic lymph node dissection and nerve-sparing procedures based on the probability of extraprostatic disease, lymph node and seminal vesicle invasion and risk of BCR still rely, among other parameters, on serum PSA.

The present work aimed to investigate the predictive ability of PSA isoforms to detect BCR and DP at three distinct time points: preoperatively and in the early post-operative period (one month and 3 months after RP).

With the present work we showed that PHI (AUC 0.73; p-value 0.0003) and [–2]proPSA (AUC 0.70; p-value 0.002) are the most significant predictors of BCR in the preoperative period, and have a higher predictive ability than serum PSA (AUC 0.64; p-value 0.029). Both markers were independent predictors of BCR at multivariate analysis (Table 3), and in the case of PHI a clear advantage was shown (AUC 0.86) over PSA (AUC 0.80) with p-values <  0.0001 for both markers.

This finding is in accordance with previous results by Lughezzani et al. [14] and Maxeiner et al. [13] who tested preoperative PHI values in populations of 313 and 437 patients respectively and confirmed the value of PHI as an independent biomarker of BCR prediction.

There are not many studies on the use of PSA isoforms in the surveillance period. We found that only PSA was able to correlate with BCR at the first (AUC 0.64; p-value 0.03) and third month (AUC 0.72; p-value 0.0004) after RP and all other markers tested such as fPSA, %fPSA, hK2, hK2/fPSA, hK2/%fPSA, [–2]proPSA and PHI were devoid of any predictive value (p-value>0.05). Similar findings were obtained by S. De Luca et al. [23] who tested [–2]proPSA in a group of high-risk patients at 3-month intervals in the first year after RP. Conversely, Casale et al. [24] concluded that [–2]proPSA could be of use in detecting recurrence earlier than standard PSA in a cohort of 134 patients after RP with a follow up of three years. The main limitations of this study were, according to the authors, a low rate of observed BCR and a high rate of [–2]proPSA false positive results.

As far as we are aware, the present work is the first recent study testing the remaining PSA isoforms in the postoperative period and their relation to the occurrence of BCR.

Disease persistence is defined in most studies as a detectable level of PSA >  0.1 ng/ml within 4–8 weeks after RP and occurs in 5–20%of patients [30]. This may be the result of residual benign or malignant prostate tissue or the presence of micrometastases. In our cohort of 128 patients, 15 patients (11.7%) developed DP. We tested the aforementioned markers and concluded that preoperatively both [–2]proPSA (AUC 0.73; p-value 0.0055) and PHI (AUC 0.75; p-value 0,0021) were superior to PSA (AUC 0.68; p-value 0.026) in predicting DP. In the surveillance period, PSA, fPSA, [–2]proPSA and PHI were the markers significantly correlating to the presence of DP. However, these results need to be taken with caution due to the low number of patients with DP in our cohort. For this reason, multivariate analysis was not performed in this group.

Besides recent nuclear imaging studies on the detection of DP [31] there are currently no studies on potential serum markers signaling DP, with the exception of studies using PSA and uPSA.

The significance of our work resides in the finding that both [–2]proPSA and PHI are stronger predictors of BCR and DP than the current serum PSA at the preoperative setting, so their use in preoperative nomograms and validation by larger scale studies would be a logical next step.

Conversely, the use of any other tested marker besides PSA brings no benefit in the period post-RP.

Although our study is innovative in testing PSA isoforms both at the preoperative and postoperative periods, analysed in the same laboratory, by the same operator, in a homogenous population, it is not devoid of limitations. The relatively small number of patients enrolled, the low rate of BCR and DP, and the use of stored serum samples albeit the study’s prospective design are the main drawbacks of our study.

We believe that cross-validation of models including [–2]proPSA and especially PHI in a preoperative nomogram, on a larger independent population, in a prospective and multicentre setting would be of a great value.

5 Conclusions

proPSA and PHI outperform serum PSA as predictors of BCR and DP preoperatively. PHI is the strongest independent predictor of BCR. At the early surveillance period only PSA correlated to BCR, and there is no additional benefit of using its isoforms in detecting DP. Further research on the use of [–2]proPSA and PHI in BCR prediction is needed to prove its clinical usefulness.