Determination of TGFβ1 protein level in human primary breast cancers and its relationship with survival

Transforming growth factor-beta (TGFβ)1 is thought to be implicated in breast cancer progression. However, data about the influence of TGFβ1 on breast cancer development are conflicting. To clarify the clinical relevance of TGFβ1, TGFβ1 protein level has been measured by enzyme-immoassay in 193 breast tumour samples. We found that 94.3% of patients expressed TGFβ1 with a range of 0–684 pg mg−1 protein. In the overall population, an increase of tumoral TGFβ1 was observed in premenopausal patients when compared to postmenopausal subgroup (P=0.0006). When patients were subdivided according to nodal status, TGFβ1 was correlated to type-1 plasminogen activator inhibitor in the node-negative subgroup (P=0.040). Multivariate analysis revealed that, after lymph node status (P=0.0002) and urokinase-type plasminogen activator (P=0.004), TGFβ1 was an independent prognostic marker for DFS (P=0.005) in the overall population. In the node-negative population, TGFβ1 was the prominent prognostic factor (P=0.010). In the same population, Kaplan–Meier curves demonstrated that high TGFβ1 level was correlated with a shorter disease-free survival (P=0.020). These data suggest that the measurement of tumoral TGFβ1 protein level, especially for node-negative patients, might help to identify a high-risk population early in tumour progression.

Transforming growth factor-beta (TGFb) belongs to a superfamily of secreted polypeptides, which regulate cell proliferation, differentiation, motility and apoptosis in a variety of different cell types (Roberts and Sporn, 1990). Three TGFb isoforms 1 -3 are ubiquitously expressed and have been detected in humans and other mammals. TGFb1 have been associated with both normal mammary gland development and breast carcinogenesis (Wakefield et al, 2000). Thus, in vivo, TGFb1 appears to regulate normal ductal and alveolar development in the mammary gland (Jhappan et al, 1993). Moreover, TGFb1 probably mediates the massive cell death and restructuring that takes place during postlactational involution of the mammary gland (Strange et al, 1992). Besides these physiological functions, there is considerable evidence that TGFb1 is implicated in several aspects of breast cancer onset and progression (Wakefield et al, 2000). A commonly held view is that TGFb1 prohibits tumour cell proliferation because TGFb1 is a potent growth inhibitor for nearly all epithelia in vitro (Roberts and Sporn, 1990). Moreover, TGFb1 can induce apoptosis, a process associated with tumour suppression, promote replicative senescence and exhibit negative regulation of angiogenesis (Alexander and Moses, 1995;Schwarte-Waldhoff et al, 2000;Perlman et al, 2001). Consistent with a tumoursuppressor role, transgenic mouse models have bring evidence that TGFb1 is able to protect against mammary tumour development in vivo, because either the suppression of TGFb1 or the inactivation of the TGFb signalling pathway results in loss of tumour prohibition and promotes carcinogenesis (Pierce et al, 1995;Bottinger et al, 1997). On the other hand, virally transformed tumorigenic mammary epithelial cell lines as well as most of the cell lines derived from invasive human breast carcinomas are resistant to the antiproliferative effects of TGFb1 in vitro and do not respond to treatment with TGFb1 in vivo. In a number of cases, this is attributable to inhibiting mutations in either TGFb type I or II receptors (Chen et al, 1998;Gobbi et al, 2000) or deregulation of the downstream signalling cascade (Xie et al, 2002).
In addition, there is increasing evidence that after cells lose their sensitivity to TGFb1-mediated growth inhibition, autocrine TGFb signalling may promote tumorigenesis. Consistent with a prooncogenic role for TGFb1 in late-stage cancer, elevated levels of TGFb1 are often observed in advanced carcinomas, and have been correlated with increased tumour invasiveness and disease progression in a variety of tumours such as malignant melanoma (Moretti et al, 1999) and colonic (Tsushima et al, 1996), ovarian (Bristow et al, 1999), and prostatic (Shariat et al, 2004) cancers. This suggests that secreting higher levels of TGFb1 may provide an advantage to tumour cells. Pro-oncogenic effects include direct effects of TGFb1 on tumoral cells, such as the stimulation of invasion and motility (Oft et al, 1996). Tumour TGFb1 may also indirectly promote cancer progression by promoting tumour vascularisation (Oh et al, 2000) and inhibiting mechanisms of immune surveillance (Letterio and Roberts, 1998).
In breast cancer, a number of studies have been engaged to evaluate the potential prognostic value of TGFb1. In most of these studies, tumoral TGFb1 expression has been examined either by immuno-histochemical assay or by Northern blot analysis. Although several groups demonstrated that increased TGFb1 was associated with more aggressive tumour behavior and poorer survival (Gorsch et al, 1992;Dalal et al, 1993;Walker et al, 1994), other authors reported the absence of correlation between disease progression and TGFb1 immunostaining or mRNA level (Dublin et al, 1993;Ghellal et al, 2000). Furthermore, a few number of studies demonstrated that TGFb1 was related to favorable outcome for patients with breast cancers (Murray et al, 1993;Kesari et al, 1999).
To date, quantitative determination of TGFb1 protein level has been performed exclusively in plasma of breast cancer patients. While some studies failed to reveal any change in plasma TGFb value in patients with breast cancer (Wakefield et al, 1995;Lebrecht et al, 2004), other reports demonstrate that patients with more advanced tumours have higher serum levels of TGFb1 (Kong et al, 1995;Ivanovic et al, 2003), suggesting that serum TGFb1 may reflect the severity of invasive breast cancer. However, these late studies have been performed on a small cohort of patients and potential prognostic value of TGFb1 has not been clearly determined. To clarify the relevance of TGFb1 as a prognostic marker in human breast cancer, we quantified TGFb1 protein level in 193 breast tissue specimens. The association between TGFb1 and the usual histological and biological parameters previously validated was examined. The prognostic relevance of TGFb1 for disease-free (DFS) and overall survival (OS) was studied for all patients by Cox multivariable analysis including the traditional prognostic markers.

Patient population
This study involved 193 patients diagnosed and treated in Assistance Publique of Marseille (France) between early 1987 and late 1992. These patients were previously included in a retrospective multicenter study (Romain et al, 2000). Patients were selected according to the following criteria: (1) primary unilateral breast tumour; (2) previously untreated, no evidence of metastatic disease or any other malignancy at the time of diagnosis; (3) T1T2, N0N1 status according to UICC criteria; (4) o75 years old; and (5) surgery as the first treatment.
The patients were 29 -74 years old at diagnosis, with a median age of 57 years. In total, 37.3% of patients were premenopausal. A total of 83 patients presented a tumour size p2 cm; 48.7% of patients were node-negative (NÀ), 29.6% presented one to three axillary invaded nodes (N þ ) and 21.7% had more than three invaded nodes (N þ þ ). Among the 193 tumours graded according to Scarff, Bloom and Richardson classification (SBR), 25.5% were classified grade I, 52.8% were grade II, and 21.7% were grade III. Ductal carcinomas were diagnosed in 75% of patients, and invasive lobular carcinomas in 25% of patients.
The primary treatment was tumorectomy or quadrantectomy (92%) or modified radical mastectomy (8%) with axillary dissection, followed by radiotherapy in 98% of cases. Among the 121 postmenopausal patients, 21 received no adjuvant treatment whereas 33 received hormone therapy, 33 were treated with chemotherapy, and 34 received both treatments. Among the 72 premenopausal patients, 36 received no adjuvant therapy and 36 were treated as follow: 10 with chemotherapy, 22 with hormone therapy and four patients with both treatments. The median follow-up was 94 months (range, 1 -140). At the cutoff date of this study, 16 local recurrences, 42 metastasis and 28 deaths had been recorded. Tumour samples and clinical informations were obtained under Institutional Clinical Board approval.

Preparation of tumour tissue extracts and ER and PR assays
Tumour tissues were stored in liquid nitrogen and routinely assayed for estrogen (ER) and progesterone receptors (PR) levels, according the recommendations of the European Organization for Research and Treatment of Cancer (EORTC), as previously described (Foekens et al, 1989). Tumour tissues were pulverised in the frozen state with a microdismembrator (Braun, Melsungen, Germany) as recommended by the EORTC. The resulting powder was suspended in buffer containing 10 mM Tris-HCl pH 7.4, 1.5 mM EDTA, 10 mM Na 2 MoO 4 , 0.5 mM DTT and 10% glycerol. The suspension was centrifuged for 60 min at 105 000 g at 41C. The high-speed supernatants (cytosols) were collected and stored in liquid nitrogen. For all samples, cytosolic protein concentration was determined using BCA assay (Pierce Chemical, Rockford, IL, USA). ER and PR levels were determined by enzyme immunoassay as described previously (Foekens et al, 1989). To assess the between-assay variations, in each series of tests an aliquot of a pooled breast cancer cytosol sample was analyzed.
The remaining cytosols were frozen and stored in liquid nitrogen until used for the determination of thymidine kinase (TK) enzyme activity, urokinase-type plasminogen activator (uPA), type-1 plasminogen activator inhibitor (PAi-1), and TGFb1.
TK, uPA and PAi-1 assays TK enzyme activity was measured using the Prolifigen TK Radioenzymatic Assay (Sangtec Medical, Bromma, Sweden), with the modifications recommended by the EORTC Receptors and Biomarkers Study Group (Foekens et al, 2001).
uPA levels were measured with the Immunobind s uPA ELISA kit and PAi-1 levels by the Immunobind s PAi1 ELISA kit (American Diagnostic, Greenwich, CT, USA), according to the instructions of the manufacturer. Inactive and active forms of uPA are all recognised by the uPA ELISA kit, as is receptor-bound uPA and uPA complexed with PAi-1 and PAi-2. PAi-1 ELISA detects latent and active forms of human PAi-1 and PAi-1 complexes. The assay is insensitive to PAi-2.

TGFb1 measurement
TGFb1 levels in breast tumour cytosols were measured by ELISA. This assay used monoclonal antibody (R&D Systems, UK) as capture antibody and biotinylated polyclonal antibody (R&D Systems, UK) as detection antibody. The assay specifically measures active TGFb1 forms. To measure total TGFb1 present in tumour samples, biologically latent TGFb1 was activated by acid-treatment. For this purpose, cytosols were diluted with four volumes of DPBS buffer (2.7 mM KCl, 137 mM NaCl, 1.5 mM KH 2 PO 4 , 3.2 mM Na 2 HPO 4 , 1 mM CaCl 2 , 0.5 mM MgCl 2 , pH 7.4). Samples were then incubated for 15 min at room temperature in the presence of 0.02 vol of 1 N HCl, then neutralised with equal volume of 1 N NaOH. ELISA analysis was performed in 96-well plates following the instructions of the manufacturer (R&D Systems, UK). Recombinant human TGFb1 (R&D Systems, UK) was used as standard at 0 -1000 pg ml À1 . A preliminary evaluation was performed to assess the buffer compatibility and the parallelism of sample dilutions. The inter-and intra-assay (n ¼ 10) CVs of a pool of tumour extracts with mean value of 134.1 pg TGFb1 per mg protein were 7.5 and 3.9% respectively.

Statistical analysis
The strength of the associations of TGFb1 with other variables was tested with Spearman rank correlation. The associations of TGFb1 (used as continuous variables) with other variables (used as grouping variables) were examined using Mann -Whitney U test (two categories), or in the case of more than two ordered categories by Kruskal -Wallis test. Survival curves were generated using the method of Kaplan and Meier and the log-rank test for trend was used to examine survival data. For the univariate survival analysis, DFS time (the interval between date of surgery and primary failure defined as a locoregional and/or distant recurrence) and OS time (the interval between date of surgery and death by any cause) were used as follow-up parameters. P-values p0.05 were considered as significant.
Cox multivariate regression analysis was used to evaluate the prognostic value of TGFb1 in the overall or N þ or NÀ populations. Multivariate analysis was performed with variables eliminated in a step-down fashion. Variables with a Pp0.05 were retained in the final multivariate models. Hazard ratios (HR) derived from the estimated regression coefficients, are presented with their 95% confidence intervals (CI).
In regard to the variations of ER levels observed in premenopausal vs postmenopausal patients (39), all tumours were considered to be estrogen receptor-negative (ERÀ) if ER values o15 fmol mg À1 protein; for the premenopausal population, tumours with ER 15 -205 fmol mg À1 protein (75th percentile) were classified ER þ , whereas tumours of postmenopausal patients were considered as ER þ when ER level was 15 -377 fmol mg À1 protein (75th percentile). In both pre-and post-menopausal populations, ER þ þ represents tumours with ER values exceeding the 75th percentile. In all cases, the tumours were considered to be PRpositive if values exceeded 20 fmol mg À1 protein. For all others biological parameters, cutpoints corresponded to the 25th and 75th percentiles of the distribution (see Table 1).

Clinicopathological characteristics
The clinicopathological characteristics of the patients are presented in Table 1. Patients were characterised according to their age, hormonal (menopausal) and steroid receptor status, tumour grade according the SBR grading system, histology and size of the tumour, and the axillary nodal status.

Biological characteristics of the breast cancer samples analyzed
The distribution of biological factors in breast cancer samples are listed in Table 2. A wide inter-patient variability in the levels of all the parameters measured could be observed. ER, PR and TK levels were previously determined in our laboratory and integrated elsewhere in other published study (Romain et al, 1995(Romain et al, , 2000. uPA and PAi-1 levels ranged from 0.01 to 1.39 ng mg À1 protein (median, 0.20) and from 0.27 to 54 ng mg À1 protein (median, 6.0), respectively. TGFb1 was detectable in 94.3% of samples and its concentration ranged from 0 to 684 pg mg À1 protein, with a median at 86.7 pg mg À1 protein.

Relationships between TGFb1 and clinicopathological and biological parameters
When the correlation between TGFb1 and each of the others parameters was examined in the overall population, no significant correlation could be observed between TGFb1 and the biological and clinicopathological variables, except the hormonal status. Thus, premenopausal patients were found to express higher TGFb1 levels than postmenopausal patients (114 vs 86 pg mg À1 protein, P ¼ 0.0006) (not shown). When patient population was subdivided according to pathological nodal status, TGFb1 remained correlated to the hormonal status in both node-negative (P ¼ 0.012) and node-positive (P ¼ 0.008) subgroups (Table 3). Moreover, a positive association between TGFb1 and PAi-1 (P ¼ 0.040) was observed in the node-negative population.

Prognostic relevance
The impact of TGFb1 on OS and DFS was determined in the overall population and node-negative/node-positive subsets. When 25th  and 75th percentiles of the distribution were used as cutoff values, TGFb1 appeared significant (P ¼ 0.020) for DFS in the overall population ( Figure 1A). The 10-year probability of DFS was 86.4% for patients with low TGFb1 levels (o42 pg mg À1 protein), 72% for the intermediate group (42 -148 pg TGFb1/mg protein) and 61% for patients with TGFb1X148 pg mg À1 protein. The patients were then dichotomised according to their nodal status. While TGFb1 was found to have no significant impact on DFS in the nodepositive subgroup (not shown), high TGFb1 levels were significantly associated with poor DFS in the node-negative population (P ¼ 0.02) ( Figure 1B). Thus, among the node-negative patients with low TGFb1 level, no relapse (DFS ¼ 100%) were observed, whereas 17 and 38% relapses were observed for patients with intermediate and high TGFb1 expression levels, respectively. Unlikely, the level of TGFb1 had no impact on OS, neither in the overall population nor in the node-positive/node-negative groups (not shown). A Cox multivariate analysis was performed to evaluate whether TGFb1 might significantly add to the contribution of the traditional prognostic factors. A significance level of 5% in the univariate analysis was chosen as the criterion for entering variables (SBR grade, nodal status, ER, uPA, PAi-1, TK and TGFb1) (not shown). The analysis was conducted in the overall population and node-negative/node-positive subgroups ( Table 4). The analysis performed for OS revealed nodal status (P ¼ 0.0003) and SBR grade (P ¼ 0.0008) as independent parameters in the overall population. The prominent predictor for OS was SBR grade (P ¼ 0.004) and ER (P ¼ 0.020) in the node-positive population, and uPA (P ¼ 0.010) in the node-negative subgroup. In addition of nodal status (P ¼ 0.0002) and uPA (P ¼ 0.004), TGFb1 was independently associated to poor DFS in the overall population (P ¼ 0.005). Whereas PAi-1 appeared as the prominent independent predictor for the node-positive patients (P ¼ 0.019), the parameter associated with DFS in the population without nodeinfiltration was TGFb1 (P ¼ 0.010).

DISCUSSION
The reduced response to TGFb in some tumour systems appears to involve multiple mechanisms, including loss of functional TGF-b receptor proteins (Grady et al, 1999;Fukai et al, 2003). In addition, mutations of downstream TGF-b signalling pathway genes have also been shown to result in a loss of responsiveness to TGF-b1 (Wang et al, 2000;Maliekal et al, 2003). In contrast to many other tumours, structural lesions of TGFb signal transducers appear to be rare in breast cancers (Chen et al, 1998;Xie et al, 2002;Jeruss et al, 2003). This suggests that, in a number of circumstances such as cell dedifferentiation, the normal function of TGFb1 in breast epithelial cells might be abrogated on behalf of oncogenic function.  Whereas TGFb1 seems to be confirmed as a marker of bad prognostic in a number of human tumours such as colorectal (Tsushima et al, 1996;Picon et al, 1998) and prostatic (Ivanovic et al, 1995;Shariat et al, 2004) cancers, the impact of TGFb1 on the progression of breast cancer remains uncertain. As for carcinomas in other organs, TGFb1 expression is often increased locally and systemically in advanced breast cancers, particularly at the leading invasive edge of the tumour and in metastasis (Dalal et al, 1993;Walker et al, 1994;Chakravarthy et al, 1999). Nevertheless, whereas the elevated expression of TGFb1 is described to associate with disease progression in a number of studies (Gorsch et al, 1992;Dalal et al, 1993;Walker et al, 1994), others studies failed to reveal diagnostic or predictive value of TGFb1 for breast cancer patients (Dublin et al, 1993;Murray et al, 1993;Kesari et al, 1999;Ghellal et al, 2000).
It is notable that, except when measured in plasma, most of the authors employed semiquantitative immunohistochemical staining to evaluate TGFb1 protein level in breast samples. In this study, we have measured for the first time the level of TGFb1 protein in breast tumour samples by ELISA, in order to examine potent correlations with clinical features. Using this assay, we found 94.3% of patients expressing TGFb1 with a range of 0 -684 pg mg À1 protein and a median value of 86.7 pg mg À1 of protein.
In agreement with a number of previous studies, we show in the overall population that TGFb1 was correlated only with menopausal status. Thus, a moderate but significant increase of tumoral TGFb1 level was observed in premenopausal patients when compared to postmenopausal subgroup (P ¼ 0.0006). These data are in apparent opposition with some published studies, indicating that estradiol decreased the production of TGFb1 by breast cancer epithelial cells in vitro (Knabbe et al, 1987;Philips and McFadden, 2004). Nevertheless, the diminution of TGFb1 in postmenopausal patients might reflect adaptability process of tumoral cells to the profound hormonal modifications, which occur during menopause.
The increased expression of uPA has been reported to be associated with poor prognostic for patients with breast cancer (Duffy et al, 1998). Paradoxically, its inhibitor PAi-1 has also been described to contribute to the malignant phenotype of tumour cells (Look et al, 2002;Schrohl et al, 2004). Thus, PAi-1 might promote the development of tumoral angiogenesis through the stabilisation and maturation of new vessels (Bajou et al, 2004). Interestingly, TGFb1 was also found to be positively correlated to PAi-1 in the node-negative subgroup (P ¼ 0.040). The activity of PAi-1 is tightly regulated on the transcriptional level, and TGFb1 is the major regulator of PAi-1 expression and in turn of local PAi-1 activity (Westerhausen et al, 1991). Thus, in the node-negative population, the upregulation of PAi-1 by TGFb1 might constitute an early event that promotes further progression of breast tumours. This is in agreement with our data indicating that TGFb1 is an indicator of bad prognostic for breast cancer patients. Thus, multivariate analysis revealed that, after lymph node status (P ¼ 0.0002) and uPA (P ¼ 0.004), TGFb1 was an independent prognostic marker for DFS (P ¼ 0.005) in the overall population. Furthermore, TGFb1 remained the prominent prognostic factor in the node-negative population (P ¼ 0.010). In this late population, Kaplan -Meier   Figure 1 The probability of DFS for overall (A) and node-negative (B) populations according to TGFb1 levels. 1, negative (TGFb1o42 pg mg À1 protein); 2, low (42 pg mg À1 protein pTGFb1 o148 pg mg À1 protein); 3, high (TGFb1X148 pg mg À1 protein). curves further demonstrated that high level of TGFb1 was correlated with a shorter disease-free survival (P ¼ 0.020). Conversely, TGFb was not a prognostic factor for OS in the nodenegative population. However, it has to be mentioned that at the cutoff date of the study, three deaths had been recorded in the node-negative subgroup. This is probably insufficient to distinguish a potential influence of TGFb on overall survival, in this population. Whereas clinical studies in breast cancers have led to conflicting results, our data suggest that TGFb1 has the potential to promote metastasis and recurrence for patients with breast carcinomas. It has to be noted that patients included in this study have not received modern chemotherapy, which could influence the outcomes. The fact that prognostic value of TGFb1 was observed in node-negative population strongly suggests that TGFb1 interferes at early stages of tumour progression, probably by making cell environment favorable for metastatic spread.
Although the lymph node status is one of the best prognostic factors in breast cancer, it is not sufficiently accurate to predict the clinical course of the disease. Indeed, 20 -30% of node-negative breast cancer patients will experience disease recurrence and metastatic dissemination. Whereas numerous predictive factors have been characterised thus far, early prognostic markers that interfere at the beginning of tumour progression are scarce. The prognostic significance of high TGFb1 level on DFS observed in node-negative breast cancer patients suggest that the determination of tumoral TGFb1 status might help to identify a high-risk population early in tumour progression, for which a more appropriate therapy should be established. In this context, it appears fundamental to confirm the prognostic value of TGFb in a large cohort of node-negative patients. Furthermore, as total TGFb (active plus latent forms) has been measured in our study, it would be helpful to determine the respective role for latent and active TGFb as prognostic markers in breast cancers.  OS and DFS,in overall population (n ¼ 193), and in node-positive (n ¼ 99) and node-negative patients (n ¼ 94)