The Urokinase System of Plasminogen Activation and Prognosis in 2780 Breast Cancer Patients¹

Cancer cell invasion and metastasis result from a coordinated interaction between proteolytic enzymes degrading the ECM³and the adhesive proteins playing a role in cell attachment and migration. Data from preclinical and clinical studies point toward a central role for the uPA system in these processes (reviewed in Refs.1, 2, 3, 4). The serine protease uPA, which binds to a specific cell surface receptor uPAR (5, 6), facilitates the conversion of plasminogen into the serine protease plasmin. This wide-spectrum protease is able to degrade most components of the ECM directly or indirectly through activation of metalloproteinases, which subsequently degrade collagens and other matrix proteins (reviewed in Ref. 7). The activity of uPA can be inhibited by the serpin inhibitors PAI-1 and PAI-2 (8). In addition, most components of the uPA system of plasminogen activation have been linked to cell adhesion and migration through both proteolytic and nonproteolytic mechanisms (reviewed in Refs. 1, 2, 3). Cell migration requires the interaction of cell-bound adhesion receptors,such as integrins and uPAR, with their ECM-associated ligands such as vitronectin (9, 10, 11, 12). Binding of uPA, or fragments of uPA containing only the receptor binding domain, enhance binding of uPAR to vitronectin. PAI-1 can inhibit integrin and uPAR binding to vitronectin, thus directing a stepwise cell migration by allowing tumor cells to be attached or alternatively being detached from the ECM(10, 11, 13, 14).

Duffy et al. (15) were the first to link increased levels of uPA activity in breast tumor extracts with a high rate of relapse in patients with breast cancer. This important finding of an association between uPA and a poor prognosis has been confirmed by various research groups measuring uPA antigen levels(16) in breast tumors, as well as in a variety of other cancer types (reviewed in Refs. 4, 17). Interestingly,immunocytologically detected uPA-positive tumor cells in bone marrow from primary breast cancer patients were predictive of a poor prognosis(18). Moreover, as reported by Jänicke et al. (19), surprisingly at first, increased levels of the inhibitor PAI-1 were associated with a poor prognosis in primary breast cancer (4, 17), and like uPA, also in recurrent breast cancer treated with tamoxifen (20). These findings can now partly be explained by the recently ascribed role of PAI-1 in tumor cell adhesion and migration (1, 2, 3). As would be expected, high tumor levels of uPAR were associated with a poor prognosis (21, 22), and high levels of PAI-2 were associated with a favorable prognosis in patients with breast cancer(23, 24).

Because simultaneous measurement of the different components of the uPA system of plasminogen activation may provide more powerful prognostic information, we determined the levels of uPA, uPAR, PAI-1, and PAI-2 in breast tumors of 2780 patients and have correlated their levels with RFS and OS.

Analysis of RFS and OS was performed in 2780 patients with primary,operable, invasive breast cancer. The selection of samples was based on the availability of 3486 stored cytosol extracts (in liquid nitrogen),which remained after routine ER and PgR analysis. Exclusion criteria were: patient tissue that was taken from a biopsy only (such as inoperable T4 tumors or tissue that was not obtained from the primary breast tumor); previous diagnosis of carcinoma, with the exception of basal skin carcinoma and cervical cancer stage I; metastatic disease at diagnosis (M1 patients) or evidence of disease within 1 month of primary surgery; neoadjuvant therapy; and insufficient follow-up documentation. In the case of mastectomy after an initial lumpectomy because of residual disease, mastectomy is considered the primary treatment. Median age of the patients at surgery (modified mastectomy,1488 patients; breast conserving treatment, 1292 patients) was 57 years(range, 22–94 years). Radiotherapy was given to 75% of the patients:on the breast/thoracic wall in 1762 patients and/or on the axilla in 747 patients; and/or parasternal and/or supraclavicular lymph nodes in 873 patients. None of the 1405 node-negative patients received systemic adjuvant therapy. Of the 1375 node-positive patients, adjuvant chemotherapy (mainly cyclophosphamide/methotrexate/5-fluorouracil) was given to 446 patients (mainly premenopausal patients), whereas 208 patients received adjuvant hormonal therapy (mainly postmenopausal patients), either alone (185 patients) or in combination with chemotherapy (23 patients). All patients were examined routinely every 3–6 months during the first 5 years of follow-up and once a year thereafter. Of the 2780 patients included, 1297 patients (47%) showed evidence of disease during follow-up and count as failures in the analysis for RFS. One-hundred and seventy-two patients (6%) died without evidence of disease and were censored at last follow-up in the analysis of RFS. Nine-hundred and twelve patients (33%) died after a previous relapse. A total of 1084 (172 + 912) patients (39%)were failures in the analysis for OS. The median follow-up period of patients alive (n = 1,696) was 88 months(range, 1–207 months). Further characteristics of patients and tumors are listed in Table 1.

Tumor tissues were stored in liquid nitrogen and pulverized in the frozen state with a microdismembrator as recommended by the European Organization for Research and Treatment of Cancer for processing of breast tumor tissue for cytosolic ER and PgR determinations (25). The resulting tissue powder was suspended in European Organization for Research and Treatment of Cancer receptor buffer (10 mmK₂HPO₄, containing 1.5 mm dipotassium chloride EDTA, 3 mmNaN₃, 10 mm monothioglycerol, and 10% v/v glycerol, pH 7.4). The suspension was centrifuged for 30 min at 100,000 × g at 4°C to obtain the supernatant fraction (cytosol). ER and PgR levels were determined by ligand binding assay or enzyme immunoassay, as described before(26).

uPA, uPAR, PAI-1, and PAI-2 levels were determined in breast tumor cytosols with ELISAs. For uPA and PAI-1, ELISA (27, 28)reagents have been used that are commercially available in assay kits(American Diagnostica, Greenwich, CT). The details of the assay procedures, including those of the specificity and performance of the uPAR and PAI-2 ELISAs, have been described elsewhere (22, 24, 29). To enable the assessment of the between-assay variations(% CV), in each assay-run an aliquot of a pooled breast cancer cytosol sample was analyzed. The between-assay CV was 12.7, 21.2, 14.3, and 8.4% for the uPA, uPAR, PAI-1, and PAI-2 assays, respectively. The within-assay CVs of samples measured in duplicate were all <5%. During the period of the assays (>3 years), there was no significant change in the levels of any of the four factors in the breast cancer cytosol pool. These results suggest that the factors were stable during long-term storage of the cytosols.

The strength of the associations between uPA, uPAR, PAI-1, and PAI-2 was tested with Spearman rank correlation(r_s). The associations of these factors with other variables were tested with the nonparametric Wilcoxon Rank-Sum test or Kruskal-Wallis test, followed by a Wilcoxon-type test for trend across ordered groups where appropriate. RFS and OS probabilities were calculated by the actuarial method of Kaplan and Meier (30). In our search for the best categorization of uPA, uPAR, and PAI-1, we have used IRA using RFS as end point (28, 31). After it had been established that in a test for trend using log-transformed variables increasing levels were significantly associated with RFS, an IRA was performed after correction for age/menopausal status, tumor size and grade, lymph node status, adjuvant therapy, and ER and PgR status. These factors defined the basic multivariate model for all patients that we incorporated in our analyses. Increasing PAI-2 levels were significantly associated with a favorable prognosis, however, only when uPA was also added to the basic multivariate model. For node-positive patients, nodal status was included as N>3 versus N1–3, and for node-negative patients, nodal status and adjuvant therapy were not applicable. The univariate and multivariate analyses, including tests for interactions,were performed using the Cox proportional hazards model(32). The assumption of proportional hazards was verified graphically. The associated likelihood ratio test was used to test for differences between models with variables included and excluded. In the multivariate analyses, the missing values for ER(n = 76), PgR (n = 120), and tumor grade (n = 677) were treated as separate groups to allow inclusion of all patients in all models. The results from Cox analyses, including the basic model and the components of the uPA system that significantly added to the model,were used to classify patients into risk groups using as cut points the 10, 25, 50, 75, and 90 percentiles of the calculated linear score from the Cox analyses. The resulting risk groups are visualized by Kaplan-Meier curves. All computations were done with the STATA statistical package, release 6.0 (STATA Corp., College Station, TX). All Ps are two-sided and relate to all available data during the total period of follow-up.

Levels of uPA, uPAR, PAI-1, and PAI-2 expressed as ng/mg of protein were measured in cytosols of primary breast tumors using specific ELISAs. The levels determined in 2780 patients (Table 1) are comparable with those obtained previously in fewer patients, with reported median levels in ng/mg of protein of 0.7 for uPA (27), 0.87 for uPAR (22), 15.2 for PAI-1 (28), and 2.26 for PAI-2 (24). For all four variables, the distribution was approximately log normal.

The Spearman rank correlation coefficients(r_s) between the various factors of the uPA system varied from r_s = 0.32 (between PAI-2 and PAI-1 or uPAR) to r_s = 0.59 (between uPA and PAI-1). Table 1 shows their relationships with patient and tumor characteristics. Levels of all four parameters were higher in ER- or PgR-negative tumors as compared with hormone receptor-positive tumors. The levels of uPA were not significantly related with age, menopausal status, or lymph node status. Well-differentiated and T_3/4 tumors had the lowest uPA levels, whereas moderately differentiated and T₂ tumors had the highest levels. In general, uPAR followed the same pattern of distribution over the various patient groups, and all of the relationships were statistically significant, i.e., the negative relationships of uPAR with age and menopausal status, and the positive associations between uPAR and tumor size and nodal status(Table 1). As was also observed for uPA, the highest uPAR and PAI-2 levels were found in moderately differentiated tumors. PAI-1 showed a positive relationship with tumor size, nodal status, tumor grade, age,and menopausal status. PAI-2 showed positive relationships with age and menopausal status, and negative relationships with nodal status and tumor size (Table 1).

To study which factors of the urokinase system add significantly to the prognostic information already provided by traditional prognostic factors, we first designed a basic multivariate model for all patients(Table 2). This model, which included age, menopausal status, tumor size and grade, the number of positive lymph nodes, adjuvant therapy, and ER/PgR status, was significantly associated with RFS(χ² = 517; df = 15; P < 0.0001) and OS(χ² = 519; df = 15; P < 0.0001). For both uPAR and PAI-2, the results of the IRA, after correction for the basic multivariate model,suggested that these variables could be considered as categorical variables with two cut points. The cut points chosen were 0.57 and 1.13 ng/mg of protein for uPAR and 0.62 and 4.05 ng/mg of protein for PAI-2. On the other hand, for uPA and PAI-1 the results were less clear and did not reveal a clear indication for cut points. However, to enable visualization of their relationships with RFS and for reasons of uniformity with the analyses of uPAR and PAI-2, we categorized uPA and PAI-1 at the levels of the two largest steps in the IRA. These levels were 0.19 and 1.21 ng/mg protein for uPA and 9.33 and 45.28 ng/mg protein for PAI-1. When each variable was added separately to the basic model for RFS, the addition of uPA resulted in aΔχ ² of 73.6 (df = 2), aΔχ ² of 19.9 (df = 2) for uPAR,a Δχ² of 81.4 (df = 2) for PAI-1, and a Δχ² of 4.6 (df = 2) for PAI-2. The RHRs and their 95% CI from the Cox multivariate analysis are listed in Table 2. High levels of uPA, uPAR, or PAI-1 were significantly associated with an early relapse, whereas PAI-2 did not significantly contribute to the basic multivariate model. Similar results were obtained in the analyses for OS (Table 2). The Kaplan-Meier curves for RFS as a function of the levels of uPA, uPAR,PAI-1, and PAI-2 are shown in Fig. 1. There were no statistically significant interactions between any of the prognostic variables in multivariate analysis for RFS or OS. Separate Cox multivariate analyses for RFS and OS were subsequently performed to establish whether combinations of uPA, uPAR, PAI-1, and PAI-2 would increase the prognostic strength of the classical prognostic factors already included in the basic model. When the four factors of the uPA system were added as log-transformed continuous variables to the basic model for RFS in all patients, high levels of uPA and PAI-1 were associated with a poor prognosis, whereas high levels of PAI-2 were associated with a favorable prognosis in the final model for RFS (for all, P < 0.001). The RHRs and 95% CIs for the components of the uPA system that are included as categorical variables in the final model for all patients and in subgroups of node-negative and node-positive patients are shown in Table 3. uPAR did not contribute to any of the models. Only in the analysis for OS in node-positive patients did PAI-2 not appear to be an independent prognostic variable. In all other analyses, uPA, PAI-1, and PAI-2 significantly added to the prognostic strength provided by the classical prognostic factors included in the basic model (Table 3). The increase in χ² caused by the addition of uPA,PAI-1, and PAI-2 to the basic model was 136 (df = 6) in the analysis for RFS and 120 in the analysis for OS. In node-negative patients with a χ² of 76 (df = 12) for the model including age and menopausal status, tumor size and grade, and ER/PgR status, the Δχ² as a result of the addition of uPA, PAI-1, and PAI-2 was 71 (df = 6)in analysis for RFS, and in the analysis for OS, theχ ² increased from 82 to 133 by the addition of the three factors. In node-positive patients, theχ ² increased from 221 (df = 14)for the model including age and menopausal status, tumor size and grade, the number of positive lymph nodes, adjuvant therapy, and ER/PgR status to 295 in the analysis for RFS, and from 207 to 297 in the analysis for OS.

In further analyses, we established prognostic scores based on the regression coefficients of the variables included in the final Cox multivariate models for all patients and separately for node-negative and node-positive patients. Kaplan-Meier curves were constructed for patients with the 10% lowest risk (group a), between 10% and 25%(group b), between 25% and 50% (group c), between 50% and 75%(group d), between 75% and 90% (group e), and the 10% highest risk for failure (group f). The actuarial RFS curve for all patients as a function of the prognostic score shows a significant separation for the different groups. At 10 years of follow-up, the difference between patients in the 10% highest and lowest risk groups was 66% in analysis for RFS (Fig. 2,A) and 61% in analysis for OS (Fig. 2 B).

The actuarial RFS curves as a function of the prognostic score for node-negative and node-positive patients are shown in Fig. 3, A and B, respectively. For node-negative patients, at 10 years the difference between the 10% lowest risk group(86 ± 4% relapse free) and the 10% highest risk group(27 ± 6% relapse free) was 59%. For node-positive patients, at 10 years the difference in RFS between the two extreme groups was 52% (61 ± 6% and 9 ± 3%for the 10% lowest and highest risk groups, respectively).

Determination of the tumor antigen levels of components of the uPA system may help to predict the time to disease recurrence and the overall survival rate in patients with primary breast cancer. This is of particular importance for patients with node-negative disease, who as a group have a relatively favorable prognosis. The goal of the present investigation was to evaluate a possible combined prognostic value of the four major components of the uPA system of plasminogen activation, uPA, uPAR, PAI-1, and PAI-2, in patients with primary invasive breast cancer.

Immunohistochemistry and in situ hybridization on breast cancer tissues showed that uPA, uPAR, PAI-1, and/or PAI-2 are expressed and synthesized by both tumor cells and host cells, including(myo)fibroblasts, endothelial cells, and phagocytic cells (reviewed in Refs. 1 and 4). A differential expression of the four components by the various cell types and an interplay between these cells may be necessary for the function of the system in various cellular processes at different stages of tumor progression(1). It has been shown that, in general, elevated antigen levels of uPA, uPAR, and PAI-1 determined in tumor extracts are associated with poor prognosis in a variety of cancer types. In contrast, increased PAI-2 antigen levels are associated with a favorable prognosis (reviewed in Refs. 1, 4, and17). The association of a high tumor level of PAI-1 with a poor prognosis in patients with primary breast cancer (19)has been explained by its inhibition of uPA activity and thus preventing degradation of tumor stroma, allowing new ECM to be formed(33), but it can also be attributed to its recently ascribed role in tumor cell adhesion and migration (10, 11, 13, 14). Furthermore, PAI-1 has been shown to be involved in uPAR clearance from the cell surface by promoting internalization of the formed ternary uPA:uPAR:PAI-1 complex (3, 34). Besides its antiproteolytic function, PAI-1 is necessary for focalized and optimal invasiveness (35), associated with angiogenic activity(36), and is essential for tumor cell invasion and tumor vascularization in PAI-1-deficient mice (37). Together these observations strongly imply that PAI-1 plays a primary role in tumor progression. Because we showed previously that the uPAR level in the cytosol was a much stronger prognostic factor than the uPAR level in the detergent extract, which also contains plasma membrane components (22), uPAR was measured only in the cytosols in the present study. The higher prognostic value of uPAR in cytosols was explained by its presentation as a water-soluble degradation product that has lost its lipid anchor because of the action of either proteases or phospholipases. The soluble uPAR may have been formed as a result of plasmin generation, which on cell surfaces happens in close vicinity to uPAR (38). Such a soluble form of uPAR has been detected in ascites fluid from patients with ovarian cancer(39). It was also found to be elevated in the blood of cancer patients (40, 41) and patients with paroxysmal nocturnal hemoglobinuria (42), when compared with healthy controls. Recently, high plasma levels of soluble uPAR were found to be correlated with a poor prognosis in patients with colorectal cancer(43).

We showed that the interrelationships between uPA, uPAR, PAI-1, and PAI-2 were all positive and statistically significant, with Spearman correlations ranging from 0.32 to 0.59. Similar positive relationships between two or more of these factors have been reported before(21, 22, 24, 28, 44, 45, 46). In the present study, all four variables were negatively related with ER and PgR. The levels of uPA were in general not significantly related with poor prognostic characteristics. However, the levels of uPAR and PAI-1 were in most analyses weakly, although significantly, associated with poor prognostic features, whereas PAI-2 levels were associated with favorable prognostic characteristics. Because most relationships were weak yet statistically significant, probably as a result of the large numbers included, several of the observed associations may hardly be of biological relevance. With respect to prognosis, we show that high tumor levels of uPA, uPAR, and PAI-1 were associated with poor, and of PAI-2 with a favorable, RFS and OS for patients with primary breast cancer. Moreover, when added separately to the basic multivariate model including traditional prognostic factors, uPA, uPAR, and PAI-1 all gave additional prognostic information. PAI-1 appeared to be the second strongest prognostic factor after nodal status, superior to the established factors such as tumor size (Table 2). In contrast, PAI-2 levels were not significantly associated with prognosis in multivariate analysis when corrected for the contribution of traditional prognostic factors. PAI-2 was significantly associated with a favorable prognosis when added to the multivariate models together with uPA and PAI-1. This may seem odd, however, we have shown before that PAI-2 was only of prognostic value in tumors with high levels of uPA and not in those with low uPA levels (24). This phenomenon of PAI-2 being an independent favorable prognostic factor in the presence of uPA could be attributable to the positive association between the levels of PAI-2 and the other three components of the uPA system that are related to a poor prognosis. When combining the various factors of the urokinase system in multivariate analyses for RFS and OS, uPAR did not further contribute to the models in which uPA and/or PAI-1 were included.

It has been argued recently by Powles (47) that “the time has come to individualize adjuvant chemotherapy, basing it more on the biological characteristics of individual tumors rather than on the widespread treatment of large groups of patients.” In this line of thought, we aimed to establish a prognostic score based on the contribution of various traditional and tumor cell biological prognostic factors. Using a score based on the traditional prognostic variables and the independent biological variables uPA, PAI-1, and PAI-2, we were able to obtain survival curves that showed a wide separation between patients in the various risk groups of patients, as well as in nodal subgroups of patients. It should, however, be emphasized that the largest power of the models in all patients are derived from the classical prognostic factors, particularly nodal status and tumor size. In analysis for RFS in all patients, theχ ² of the basic model containing the classical prognostic factors was 517 (df = 15). The independent biological factors uPA, PAI-1, and PAI-2 only added moderately to this model (increase in χ² of 136, df = 6). Importantly, for node-negative patients, this increase inχ ² caused by the addition of the three factors was 71 (df = 6). This is relatively high compared with the χ² of 76 (df = 12) already provided by the classical prognostic factors.

In the present study, the levels of the components of the urokinase system of plasminogen activation were determined with specific ELISAs performed on tumor extracts, which does not discriminate between the cell type that expresses the specific factor. Nevertheless, the measured levels correlated with prognosis in several cancer types in many published studies. An advantage of using ELISAs is that the assays can easily be subjected to external quality control programs(48). With respect to immunohistochemical assessment of the components of the uPA system, discrepant results on the localization of the different factors have been published by various groups. It has been argued by Andreasen et al.(1) that to ascertain the specificity of immunohistochemical stainings and to obtain conclusive results, special care should be taken regarding a number of control experiments and that, although immunohistochemical studies may reveal where the various components are present, the localization of the active components remains elusive (1).

Several of the components of the uPA system are potential targets for antiangiogenic, anti-invasive, and/or antimetastatic therapy, and various different approaches to interfere with the expression or reactivity of uPA or uPAR at the gene or protein level have been proven successful. Such therapeutic approaches include the application of antisense oligonucleotides, antibodies, enzyme inhibitors, and recombinant and synthetic uPA and uPAR analogues (reviewed in Refs.1, 3, and 4). Because the uPA system plays an important role in tumor cell adhesion and migration as well, treatments aimed at interfering with the nonproteolytic properties of this multifactorial system may also prove beneficial.

We are indebted to Joan Bolt-de Vries and Erica Binnendijk-Noordegraaf for technical assistance and to the Nijbakker-Morra Foundation, Amsterdam, the Netherlands, for the donation of the ELISA washer. We gratefully express our thanks to the surgeons, pathologists, and internists of the St. Clara Hospital Rotterdam, Ikazia Hospital Rotterdam, and St. Fransiscus Gasthuis Rotterdam for the supply of tumor tissues and/or for assisting us in the collection of the clinical follow-up data.