Predictive Value of Plasma Hepatocyte Growth Factor/Scatter Factor Levels in Patients with Clinically Localized Prostate Cancer

Prostate cancer is the most common cancer and the second largest cause (9%) of cancer-related deaths among men in the Unites States. In 2007, it is estimated that there will be 218,890 new cases and 27,050 prostate cancer–related deaths (1).

The advent of prostate-specific antigen (PSA) screening has led to a stage migration with a decrease in the number of patients presenting with advanced cancer. Despite this, ∼5% to 25% of patients with presumed localized disease have metastases to lymph nodes (2, 3) and up to 30% of patients experience disease recurrence after presumed curative treatment. Several prediction tools such as Kattan-nomograms, Partin tables, algorithms, and artificial neural networks have been developed to predict the presence of lymphatic metastases but none of them are sufficiently accurate and need supplementation with novel biomarkers to improve predictive accuracy (4). Accurate prediction of lymphatic metastases is extremely important because therapeutic decisions such as the choice of treatment (surgery versus radiation, neoadjuvant therapy) and decision to perform and the extent of pelvic lymphadenectomy are affected by this information. Identification of patients who are likely to fail definitive local therapy would be helpful for selecting patients best suited for local adjuvant radiation therapy or clinical trials of early systemic intervention. Therefore, there is a clear need for novel biomarkers that are specifically associated with biologically aggressive prostate cancer for improved staging and prognostication.

Hepatocyte growth factor/scatter factor (HGF/SF) is a multifunctional cytokine that modulates processes such as growth, adhesion, motility, matrix degradation, invasion, and angiogenesis (5). HGF/SF also stimulates mitosis, inhibits apoptosis, and induces morphogenesis in ductal epithelial cells (5). c-Met, a transmembrane receptor tyrosine kinase, is the only known receptor for HGF/SF. The same roles of HGF/SF in normal cellular processes also make it an important promoter of tumor growth, invasion, and metastasis. HGF/SF has been, for example, shown to augment the adherence of tumor cells to the endothelium and enhance transmigration through extracellular matrix (6). Several studies have shown a relationship between serum and tissue concentration of HGF/SF with progression of several cancers such as gastric, breast, lung, and hypopharyngeal (7–11). In prostate cancer, however, the prognostic value of HGF/SF is not well studied. Studies have shown that HGF/SF levels are elevated in metastatic disease (12). However, whether HGF/SF can independently predict poor pathologic features or clinical outcomes is not known.

In this study, we investigated the ability of preoperative plasma levels HGF/SF to predict adverse pathologic features and disease recurrence in patients treated with radical prostatectomy and bilateral lymphadenectomy for clinically localized disease.

Patient population. All studies were undertaken with the approval and oversight of the institutional review board. The study comprised 421 consecutive patients treated with radical prostatectomy and bilateral lymphadenectomy for clinically localized prostatic adenocarcinoma between July 1994 and November 1997. No patients were treated preoperatively with either hormonal or radiation therapy and none had secondary cancers. Serum total PSA (T-PSA) was measured by the Hybritech Tandem-R assay (Hybritech, Inc.). The median age was 61.1 y (range 39.4-75.5 y). The median T-PSA was 5.6 ng/mL (interquartile range, 3.7-8.6; range, 0.1-99.3). Staff pathologists, who were blinded to clinical outcomes, examined all prostatectomy specimens. Evaluation of the radical prostatectomy specimen was done as previously described in accordance with the guidelines of the College of American Pathologists (13, 14).

We also assessed plasma HGF/SF levels in 64 consecutive healthy patients without cancer. These subjects were composed of men who participated in a prostate cancer screening program. They had no prior history of cancer or chronic disease, a normal digital rectal examination, and a T-PSA <2.0 ng/mL (15). The median age of the controls was 61.3 y (range 42.4-71.7 y).

Plasma HGF/SF measurements. Plasma samples were collected after a preoperative overnight fast on the morning of the day of surgery, at least 4 wk after transrectal guided needle biopsy of the prostate. Blood was collected into Vacutainer CPT (Becton Dickinson Vacutainer Systems) and centrifuged at room temperature for 20 min at 1,500 × g. The top layer corresponding to plasma was decanted using sterile transfer pipettes. The plasma was immediately frozen and stored at −80°C in polypropylene cryopreservation vials (Nalgene, Nalge Nunc). For quantitative measurements of HGF/SF levels, we used a commercially available quantitative immunoassay (R&D Systems). Every sample was run in duplicate, and the mean was calculated for data analysis. Differences between the two measurements were minimal (intra-assay precision coefficients of variation, 7.8 ± 9.6%).

Postoperative follow-up. Patients generally were scheduled to have a digital rectal examination and serum T-PSA evaluation postoperatively every 3 mo for the 1st year, semiannually from the 2nd through the 5th years, and annually thereafter. Biochemical progression was defined as a sustained elevation, on two or more occasions, of T-PSA >0.2 ng/mL and was assigned to the date of the first value >0.2 ng/mL. Patients who received adjuvant radiation therapy before biochemical progression because of clinical or pathologic characteristics were considered to have disease progression from the date of the first T-PSA value >0.2 ng/mL.

Statistical analysis. Age, T-PSA, and HGF/SF were not normally distributed. Thus, nonparametric tests were used for all data analysis. Biopsy and pathologic Gleason sums were grouped into three categories (2-6 versus 7 versus 8-10). The lowest category (Gleason sum 2-6) was used as the reference category. Spearman's correlation (Rho) was used to assess the association between plasma HGF/SF level and continuous variables (i.e., age and T-PSA). The Wilcoxon-Mann-Whitney test and the Kruskal Wallis test were used to evaluate the association between HGF/SF and categorical clinicopathologic variables.

Univariate and multivariate logistic regression analyses were done to assess the relationship between plasma HGF/SF levels and pathologic features. Logistic regression models were created using the “backward elimination method.” Hosmer and Lemeshow goodness-of-fit test was done to assess the fit of all the models and none of the models were found to have a poor fit. The assumption of linear association between the ln(odds) and predictors was assessed graphically and was satisfied for all predictors.

Univariate and multivariate survival analyses were done using the Cox proportional hazards regression model. Time to disease recurrence was calculated as the interval from the date of surgery to disease recurrence. Models were created using the “backward elimination method.” Two models were created. The first model was a “preoperative” model designed to predict disease recurrence based on preoperative clinical and pathologic information. The variables considered for entry into the model included plasma HGF/SF level, age, preoperative T-PSA, biopsy Gleason sum, and clinical stage. The second model was a “postoperative” model and was based on both preoperative and postoperative clinical and pathologic features, including plasma HGF/SF level, age, preoperative T-PSA, pathologic Gleason sum, surgical margin status, extraprostatic extension, seminal vesicle invasion, lymph node metastases, and lymphovascular invasion. The proportionality assumption of the Cox models was checked by introducing time-dependent covariates in the models. The proportional hazard assumption was not violated for any of the models. Interactions between plasma HGF/SF levels and other covariates were tested and were not found to be significant.

For the regression analyses, the values of HGF/SF were divided by 100. Thus, the odds ratios (OR) and the relative risk estimates should be interpreted as the increase in odds/risk for each 100 pg/mL increase in plasma HGF/SF level.

Predictive accuracies of the logistic regression models were quantified with the area under the receiver-operating characteristic curve. For the Cox regression models, the predictive accuracy was quantified using the concordance probability estimate (16). For both the area under the curve and the concordance probability estimate, a value of 1.0 indicates perfect predictions, whereas 0.5 is equivalent to a toss of a coin. Internal validation was done using 1,000 bootstrap re-samples to reduce overfit bias. The average bootstrap-corrected predictive accuracy of every model was generated and was expressed as a percentage. The percentile method was used to generate the 95% confidence intervals (95% CI) for the predictive accuracies and their differences. The effect of including HGF/SF in prediction models was assessed as per the recommendations of Kattan (17).

Statistical significance in this study was set as P ≤ 0.05. All reported P values are two sided. All statistical analyses were done with SAS version 9.1 (SAS Institute, Inc.).

The median plasma HGF/SF level in prostate cancer patients was 505.5 pg/mL (interquartile range, 342.4-697.6; range, 102.3-2015.3) compared with the control group in whom the median HGF/SF plasma level was 397 pg/mL (interquartile range, 295.6-467.4; range, 93.9-571.2; P < 0.0001).

Association with clinical and pathologic features. Plasma HGF/SF levels did not correlate with age (Rho = −0.047, P = 0.34). Plasma HGF/SF levels were positively associated with serum T-PSA (Rho = 0.16, P = 0.001). The association between HGF/SF and clinical and pathologic features are shown in Table 1. HGF/SF levels increased with increasing biopsy and pathologic Gleason sums, but these differences did not reach statistical significance (P = 0.076 and P = 0.066, respectively). Patients with seminal vesicle invasion, lymph node involvement, or lymphovascular invasion had elevated HGF/SF levels compared with patients who did not have these features (all P < 0.05).

On multivariate logistic regression analyses, the preoperative plasma HGF/SF level was an independent predictor of lymph node metastasis (OR,1.82, 95% CI, 1.33-2.49), seminal vesicle invasion (OR, 1.18; 95% CI, 1.06-1.3), and lymphovascular invasion (OR, 1.15; 95% CI, 1.004-1.33; Table 2). HGF/SF was not independently associated with extraprostatic extension.

The change in predictive accuracies after addition of HGF/SF to the base preoperative model containing preoperative clinical factors (T-PSA, biopsy Gleason sum and clinical stage) increased its accuracy for prediction of lymph node metastasis by 6.7% and brought the overall predictive accuracy to 98.4%. Addition of HGF/SF did not significantly increase the predictive accuracy for prediction of other pathologic features (Table 3).

Effect of inclusion of HGF/SF on predictive accuracy of the model to predict lymph node involvement was assessed as recommended by Kattan (17). This was done by creating four new models, each with one variable omitted (Table 4). Removal of HGF/SF led to the largest decrease in predictive accuracy (6.7%) compared with other variables (3.3% for Gleason sum, 0.7% for T-PSA, and 0.4% for clinical stage).

Association with disease recurrence. The median follow-up duration was 37 months (interquartile range, 21.9-65.1; range, 0.9-103.8 months) for patients who did not experience disease recurrence. Overall, 75 patients (17.8%) experienced disease recurrence within a median duration of 13.5 months after surgery (interquartile range, 6.4-32.6; range, 2.2-74.4 months).

Preoperative model. On univariate Cox proportional hazard regression analysis, preoperative plasma HGF/SF level predicted disease recurrence (relative risk, 1.13; 95% CI, 1.05-1.22; P = 0.001). On multivariate Cox regression analysis that controlled for the effects of preoperative T-PSA and biopsy Gleason sum, plasma HGF/SF level was an independent predictor of disease recurrence (Table 5). Interactions of HGF/SF with biopsy Gleason sum, T-PSA, and clinical stage were not statistically significant. The predictive accuracy (concordance probability estimate) of the base model without HGF/SF was 62.8% (95% CI, 58.3-67.2%). The addition of HGF/SF increased the predictive accuracy of the model to 63.9% (95% CI, 59.4-68.3%), that is, an increase of 1.06% (95% CI, 0%-3.8%).

Postoperative model. On multivariate Cox proportional hazard regression analysis, preoperative plasma HGF/SF level was not an independent predictor of disease recurrence after controlling for the effects of postoperative pathologic features. Pathologic Gleason sum (P < 0.0001), surgical margin status (P < 0.0001), and lymph node involvement (P = 0.01) were the only independent predictors of disease recurrence.

We found that preoperative plasma levels of HGF/SF was a strong predictor of the presence of lymph node metastasis in patients with clinically localized prostate cancer. Addition of HGF/SF to standard preoperative prediction variables improved their accuracy for prediction of lymph node metastasis by a clinically significant margin (i.e., 6.7%). In addition, HGF/SF also predicted the presence of seminal vesicle involvement and disease recurrence after surgery.

Elevated blood and tissue levels of HGF/SF have been associated with adverse pathologic features and poor outcomes in malignancies of the stomach, breast, lung, and hypopharynx (7–11). For example, in breast cancer patients, c-Met can be detected in the axillary drainage after surgery and correlates with lymph node involvement, positive margins, and lymphatic invasion (18). In addition, elevated serum levels of HGF/SF were associated with lymph node involvement and distant metastasis (11). In melanoma and hypopharyngeal cancer, expression of c-Met and HGF/SF was associated with lymph node metastases and decreased survival (8, 19).

In the prostate, the stromal cells secrete HGF/SF, whereas the epithelial cells express its receptor c-Met (20). HGF/SF stimulates the proliferation of DU145 cancer cell lines while inhibiting the growth of normal prostate epithelial cell lines (21). Incrementally higher c-Met levels are seen in prostate cancer cell lines of more malignant potential and c-Met is preferentially expressed in androgen-independent prostate cancer cells (22, 23). Moreover, c-Met is preferentially expressed in lymph node and bone metastasis compared with benign prostatic hyperplasia and primary prostate cancer (24–26). However, despite the extensive basic science research on the role of HGF/SF in prostate cancer, there are only few clinical or translational studies. Serum levels of HGF/SF have been shown to be higher in men with metastatic prostate cancer compared with those with localized disease, who in turn have higher levels compared with healthy controls (27–29). Pisters et al. (24) showed that tissue expression of c-Met was associated with higher tumor grade as well as lymph node and distant metastasis. Humphrey et al. (12) found that serum levels of HGF/SF predicted survival in patients with metastatic prostate cancer on univariate, but not multivariate, analysis.

We found that preoperative plasma HGF/SF predicted metastasis to lymph nodes. Currently available imaging modalities such as computed tomography, magnetic resonance imaging, and positron emission tomography lack adequate sensitivity to detect node metastasis (30). Prediction tools such as the Partin tables and the Kattan nomogram that use T-PSA, Gleason sum, and clinical stage reported a predictive accuracy between 76% and 83% (3, 31, 32). The addition of HGF/SF to the preoperative Kattan nomogram increases its predictive accuracy by a prognostically significant margin (i.e., 6.7%).

After an external validation in large multi-institutional data sets, our findings could help in the management of patients with clinically localized prostate cancer. For example, an elevated preoperative HGF/SF level indicative of a high risk for lymph node metastasis could lead to the performance of standard or preferentially extended pelvic lymphadenectomy. Thus, the use of a biomarker such as HGF/SF may allow for better selection of high-risk patients with an increase in oncologic efficacy and decrease in morbidity by avoiding unnecessary procedures in lower-risk patients. Moreover, addition of preoperative HGF/SF addition into the classic nomograms may improve their accuracy for prediction of pathologic features and clinical outcomes. Another potential use of HGF/SF is to select patients best suited for clinical trials of neoadjuvant and adjuvant therapy. HGF/SF may also help in avoiding the morbidity of salvage therapies in patients with disease recurrence as they are likely to be ineffective in the setting of lymph nodal metastases (33). Finally, HGF/SF is a potential target for novel chemotherapeutic agents. NK4, for example, is a competitive HGF/SF antagonist and has been shown to reduce the growth, adhesion, migration, and invasion of prostate cancer cells (34, 35). Inhibition of invasion and migration of HGF/SF–mediated prostate cancer cells by a hammerhead ribozyme has also been described (36). Further studies may elucidate whether HGF/SF–targeted therapy would improve outcomes in patients with prostate cancer.

Our study is limited by its retrospective nature, small sample size, low prevalence of nodal metastases, the lack of a standard template for pelvic lymph node dissection, the use of disease recurrence as the outcome rather than metastasis or survival, and the relatively short follow-up. The standard lymph node sampling may have failed to detect all nodal metastases that could have been detected by a more extended lymphadenectomy (37).

The addition of preoperative plasma HGF/SF improved the predictive accuracy by a small margin of 1% for recurrence. Although the clinical significance of this is arguable, the degree of improvement was statistically significant. In addition, the hazard ratio for HGF in preoperative models is 1.07. HGF was treated as a continuous variable and thus a hazard ratio of 1.07 means that the risk of recurrence increases by 7% for every 100 pg/mL increase in HGF levels. Thus, for a patient whose HGF level is 500 pg/mL higher, their risk is 35% higher. Furthermore, an accuracy gain of 1% implies that 1 of 100 additional patients will be correctly classified. Although such an increase may seem trivial, the effect of a 1% gain is substantially more relevant on a larger scale, when thousands of patients are considered. A small accuracy gain is also important when the nomogram is used for risk stratification within a clinical protocol. A more accurate nomogram will contribute to better distribution between study arms. Finally, a seemingly trivial gain is always important to patients who deserve the most accurate prognosis.

Preoperative plasma levels of HGF/SF independently predict the presence of lymph node metastasis and improve the predictive accuracy of classic nomograms for prediction of lymph node metastasis by statistically and clinically significant margin. HGF/SF also predicts the probability of disease recurrence after surgery. After validation in large, multicenter, prospective studies, HGF/SF should be considered for addition into pretreatment predictive tools for pathologic stage and disease recurrence.

No potential conflicts of interest were disclosed.

We thank Drs. Michael Kattan, Nancy Obuchowski, and Mithat Gonen for their guidance during the statistical analysis of this article.