Expression of Thrombospondin-1 in Resected Colorectal Liver Metastases Predicts Poor Prognosis

Materials and Methods

Study population. All patients who underwent a liver resection for colorectal liver metastases at the University Hospitals of Leicester NHS Trust and the Royal Liverpool Infirmary from 1993 to 1999 were identified. All data derived from Liverpool were prospective, and all data from Leicester were retrospective from 1993 to 1999 and prospective from 1999 onwards.

The patients' medical notes were reviewed and a database compiled detailing patient demographics and a series of clinicopathologic factors (Table 1).

In Leicester, paraffin-embedded tissue blocks pertaining to each resected liver metastasis were obtained from the Department of Pathology. Tumor edge sections (abutting normal hepatocyte tissue) were selected with a Professor in Histopathology (J.L.J.). Paraffin-embedded tissue sections were similarly obtained from Liverpool. The liver metastatic tissue present in the Liverpool database had been prospectively selected with biopsies being taken from tumor edge. The Liverpool tissue bank also comprised an anonymized patient database detailing the same information as the Leicester database.

Determination of thrombospondin-1 expression. TSP-1 detection was detected using a standard Dextran Polymer Conjugate Two-step Visualization system Envision (DAKO, Ely, United Kingdom). All tissue sections were stained in a retrospective manner. Tissue sections were dewaxed, deparaffinized in xylene, rehydrated through a series of graded alcohols, and washed in water. Antigen retrieval was undertaken by immersing the slides for 10 minutes in a 300-mL solution of 0.1% trypsin (Becton Dickinson and Co., Meylan, France) and 0.1% calcium chloride (Fisher Scientific, Loughborough, United Kingdom), contained in a Hallendahl jar in a 37°C water bath at pH 7.8. The slides were rinsed in running water, bathed in TBS buffer for 5 minutes, and covered with 0.03% hydrogen peroxide for a further 5 minutes to reduce nonspecific staining. After rinsing in TBS, the slides were incubated with a 1:50 dilution of a mouse monoclonal primary antibody for TSP-1 (Ab4, Labvision, Newmarket, United Kingdom) at room temperature in a humid chamber for 30 minutes. After further washing in buffer for 5 minutes, the secondary peroxidase-labeled polymer conjugated to goat anti-mouse immunoglobulins was applied for 30 minutes. The sections were washed in buffer and TSP-1 antigen visualized using 3,3-diaminobezidinetetrahydrochloride chromogen and Mayer's hematoxylin counterstain. Slides were dehydrated through graded alcohols and coverslips applied.

Interpretation of slides. Slides were scanned by light microscopy at low power (×40 and ×100) and high power (×200). TSP-1 was localized in the tissue stroma and around blood vessels (perivascular). A semiquantitative grading scale was devised by a professor of histopathology. The pattern of expression was too complex for a percentage grading scale to be used; therefore, a scoring system (0-5) was devised to categorize the extent of immunoreactivity observed (Table 2).

Data analysis and statistics. Statistical analysis was done using SPSS version 11. (SPSS, Inc., Chicago, IL). The median score was calculated for the series and used to categorize tumors into those with greater than the median TSP-1 expression or less than the median expression. The χ² test was used to compare TSP-1 expression with the clinicopathologic variables. The Kaplan-Meier and log-rank survival curves were plotted to determine the relationship between perivascular and stromal TSP-1 and patient survival. Multivariate analysis was done using the Cox proportional hazard regression analysis.

Results

The total number of patients undergoing liver resection for colorectal liver metastases at our institutions during the study period was 197 (Leicester, n = 117; Liverpool, n = 80). Patients with a 30-day operative mortality (n = 7, 3.5%) and no or inadequate pathology specimens (n = 8) were excluded. The number of patients eligible for inclusion was thus 182. Of these, 135 patients remained alive at 2 years (75%) and 58 patients (32%) remained alive at 5 years after resection. Eighty-four patients received chemotherapy with a combination of 5-fluorouracil/folinic acid and leucovorin.

The clinicopathologic variables are listed in Table 2.

Forty-five tumor specimens (25%) expressed perivascular TSP-1 and 59 samples (33%) expressed stromal TSP-1 (Fig. 1A and B). In 124 tumors, no stromal or perivasvular TSP-1 was observed (Fig. 1C). In 36 tumors, both patterns of expression were observed. In 18 tumors, stromal TSP-1 was detected but no concomitant perivascular TSP-1 expression. In four tumors, perivascular immunoreactivity was detected with no concomitant stromal expression. Therefore, in 160 tumors, either both or neither patterns of expression were detected. These patterns of expression were analyzed separately given the distinct immunoreactivity observed that is in accordance with reports in the literature regarding the localization of TSP-1 in the extracellular matrix and surrounding smooth muscle vessel walls (7–10). No tumor cell expression of TSP-1 was observed. The median TSP-1 score for the series was zero expression; therefore, tumors were categorized into present versus absent TSP-1 expression.

TSP-1 expression did not correlate with any of the clinicopathologic variables (Table 3). On univariate analysis, patients with tumors expressing perivascular TSP-1 had a significantly worse prognosis than those without perivascular TSP-1 expression (P = 0.01). Similarly, those patients expressing TSP-1 in the stoma also had a worse prognosis (P = 0.03; Fig. 2A and B). On multivariate analysis (Table 4) of all the clinicopathologic data listed in Table 2, and TSP-1 expression, bilobar disease and the expression of perivascular TSP-1 were independent negatively prognostic factors for patient survival (P = 0.009 and P = 0.01, respectively).

Discussion

Thrombospondin is a large glycoprotein, which is stored in the α granules of thrombin-stimulated platelets secreted in response to platelet-derived growth factor and incorporated into the extracellular matrix (Fig. 3). It is also a major secretory product of several vessel wall cells including smooth muscle cells, endothelial cells, and fibroblasts and is incorporated in the extracellular matrix (7–10). These may explain the observed perivascular immunoreactivity in this series.

Five subclasses of thrombospondin exist, TSP-1 to TSP-5, of which TSP-1 and TSP-2 are the most closely related structurally. TSP-1 consists of a heparin domain; a procollagen-like domain; type 1, 2, and 3 repeats; and a cell-binding domain (11).

The procollagen domain and the type 1 repeats, shared by TSP-1 and TSP-2, are believed to confer antiangiogenic properties and impair migration of endothelial cells in vitro (3). It is the multidomain structure of these molecules that enable it to interact with a variety of molecules.

TSP-1 modulates platelet aggregation, wound healing, protease activity, and cellular functions such as adhesion, motility, and growth (1). However, the precise function of thrombospondin in angiogenesis and tumor growth is not straightforward as controversy surrounds the actual role of thrombospondin in neovascularization. Evidence suggests that TSP-1 has both proangiogenic and antiangiogenic properties, as described below, thus displaying a regulatory role in tumor neovascularization.

Cellular adhesion of endothelial cells to the extracellular matrix is an important step in tumorigenesis. TSP-1 is believed to contribute to cellular adhesion by interacting with cell surface molecules. Examples of such surface molecules are the integrins; for example, α 3 β 1 integrin mediates the adhesion of TSP-1 to breast carcinoma (12) and anchorage-independent small cell lung cancer cell lines (13); the CD36 (14) and CD47 receptors in certain breast neoplasms (3); a novel receptor found in invasive malignant cells and endothelial cells in carcinoma of the lung (15); and heparins (16). TSP-1-mediated cellular adhesion can prevent cellular proliferation, and it has been shown in small cell lung cancer cell lines that it then promotes neuroendocrine rather than carcinogenic differentiation (13).

Cellular motility is promoted by chemotaxis in response to TSP-1. Receptors such as CD47 promote cellular motility by signaling pathways that modulate integrin-mediated cellular migration (17). It is also possible that different fragments of the thrombospondin molecule can be responsible for different angiogenesis-modulating properties that may often be opposing: a 25-kDa heparin-binding fragment induced a notably stronger angiogenic response in the rabbit cornea assay for neovascularization than the TSP-1 molecule as a whole (2), whereas a 140-kDa fragment, however, completely inhibited the angiogenic response (2). TSP-1 has also been shown to promote migration at high concentrations but actually inhibit migration at low concentrations if the whole thrombospondin molecule is considered thus emphasizing that different domains within the TSP-1 molecule may account for different tumor-modulating properties (3), further supported by studies in breast cancer cell lines in which the deletion of a pentapeptide sequence containing a cysteine residue near the COOH terminus of TSP-1 in proximity to the calcium-binding domain, resulted in increased tumor growth and metastatic potential (3).

TSP-1 may potentiate tumor invasion by up-regulating protease components of the plaminogen/plasmin system, such as thrombin, plasmin, and urokinase plasminogen activator (9); modulating the action of matrix metalloproteinases upon the extracellular matrix in a concentration-dependent manner; activating other growth factors such as transforming growth factor-β1 (9); and acting synergistically with other growth factors such as epidermal growth factor that stimulates mitogenesis in smooth muscle cells in vitro via an autocrine pathway (10). These processes are thought to ultimately result in the digestion of the extracellular matrix, promoting tumor invasion (9).

It has also been suggested, however, that TSP-1 may inhibit tumor growth and invasion by binding to growth factors and proteases thus preventing the digestion of the extracellular matrix (3). Thrombospondin has been shown to inhibit cellular proliferation in vitro, as shown in carcinoma cell lines (18). Historically, TSP-1 has been regarded as a potent inhibitor of angiogenesis. TSP-1 is found in abundance in association with mature quiescent vessels but is absent from actively growing vessels (19, 20). It has been found to suppress capillary formation in vitro (20, 21), possibly by inhibiting chemotaxis, proliferation (3, 22), and the formation of focal adhesions in endothelial cells (23). It may be transforming growth factor-β coupled to TSP-1 that is responsible for this inhibitory effect on endothelial cell proliferation (24), but not all studies support this theory (3). TSP-1 also inhibits β fibroblastic growth factor–driven angiogenesis in vivo in the rodent eye model (25).

Few clinically based studies have examined the role of thrombospondin in tumorigenesis. Of particular interest are those that studied thrombospondin in colorectal cancer. To date, all studies examining the expression of TSP-1 in colorectal cancer pertain to primary tumors. The overall findings attribute an antiangiogenic role to thrombospondin, with improved patient survival in tumors expressing this glycoprotein, although direct comparisons cannot be made because different variables were examined by each study.

To our knowledge, this series is the first to examine the expression of TSP-1 in colorectal liver metastases.

Maeda et al. studied TSP-1 levels by immunohistochemical methods in 150 patients with primary colon cancer, 125 having undergone curative surgery, and 30 noncurative surgery. Fifty-nine percent of tumors were positive for TSP-1 (any degree of either cytoplasmic or membranous tumor cell staining was considered positive). TSP-1-positive tumors had lower microvessel counts compared with TSP-1-negative tumors and did not correlate with the development of liver metastases (liver metastases were more frequent in TSP-1-negative primary tumors). TSP-1 was not studied in the liver metastases and with increased disease-free survival rates (84% 5-year survival compared with 55% in TSP-1-negative tumors). Disease recurrence was also less frequent (8%) in TSP-1-positive patients (81% of TSP-1-negative tumors developed recurrence). This study therefore supports the role of TSP-1 as an inhibitor of angiogenesis by decreasing tumor vascularity thus reducing the risk of recurrence and prolonging patient survival (4). In another study by Maeda et al. examining the relationship among TSP-1, vascular endothelial growth factor (VEGF), and microvessel counts using 100 resected primary colon cancers; an inverse relationship between TSP-1 and microvessel counts; and a direct relationship between VEGF and microvessel counts were independently shown. Tumors with high VEGF and low TSP-1 had higher microvessel counts and by multivariate analysis worse prognosis with a 73% recurrence rate. Similarly, in hepatocellular carcinoma, high levels of TSP-1 and low levels of VEGF equated with low microvessel density (26).

Maeda et al. failed to show a direct relationship between TSP-1 and VEGF expression, but in endometrial cancer, strong thrombospondin expression to weak VEGF expression has also been found (wherein extracellular intratumoral or peritumoral thrombospondin was graded as weak, moderate, or strong; ref. 27). Thrombospondin expression did not correlate with pathologic features of the primary colorectal tumor, as shown by other researchers (4, 5, 28, 29).

Maeda et al.'s work also showed that TSP-1 expression independently correlated to better 5-year survival (91% versus 44% in TSP-1-negative tumors; ref. 5). Studies in breast cancer have described reduced TSP-1 levels in association with more aggressive tumors (3). No such antiangiogenic role was found for TSP-1 in endometrial cancer (27).

On initial inspection, our results may seem contradictory to Maeda et al. (4, 26) because we found that the expression of TSP-1 in colorectal liver metastases was negatively prognostic. However, on closer analysis, the differences may be explained by the site of expression of TSP-1. Maeda et al. graded the expression of TSP-1 in primary colorectal carcinoma tumor cells (cytoplasmic and membranous expression), whereas in our series of colorectal liver metastases, TSP-1 was expressed around blood vessels and in the stroma (Fig. 1A and B, respectively) and not in tumor cells. We may postulate, therefore, that TSP-1 is expressed differently in liver metastases when compared with the primary tumor, possibly showing different modes of action such as facilitating tumor invasion rather than acting as an antiangiogenic growth factor.

TSP-1 may potentially be up-regulated by other cytokines. Kawakami et al., in a study of 53 primary colon cancers, showed higher expression of TSP-1 and TSP-2 mRNA (TSP-2 also being described as an antiangiogenic agent) in tumors expressing interleukin-10 mRNA. Vessel counts were also lower in tumors with increased interleukin-10 expression, but no relationship was found when compared with other angiogenesis markers (30).

Because up-regulation of TSP-1 may be a cytokine-dependent process, TSP-1 may equally need to interact with other cytokines to account for the antiangiogenic effects seen in some tumors.

In study of gene expression of TSP-1 and its receptor CD36 in 65 colon cancers, Tsuchida et al. showed that TSP-1 gene expression did not correlate to tumor vascularity, whereas CD36 gene expression correlated not only to lower vessel counts but to improved patient survival. Activation of CD36 is believed to activate signaling pathways that inhibit angiogenesis through ill-understood mechanisms possibly interacting in conjunction with protein kinases (31). The authors therefore concluded that the described inhibition of angiogenesis by TSP-1 may not be due to TSP-1 per se but to the presence of its receptor CD36, which may explain the discrepancies between studies regarding the role of TSP-1 in angiogenesis (32).

To analyze the roles of both TSP-1 and TSP-2 in the risk of the development of liver metastases, Tokunaga et al., in a study of 61 patients with colorectal cancer (28), found TSP-1 gene expression in 6.5% of primary tumors, TSP-2 gene expression in 28% of tumors, and gene expression of both markers in 34% of tumors. Both thrombospondin genes were more significantly expressed in the primary tumor than in the surrounding tissue. Interestingly, TSP-1 expression did not correlate to the development of liver metastases, comparable with Maeda et al.'s work (4), whereas, TSP-2 expression was associated with a lower incidence of metastatic disease: 15.8% compared with 43.5% incidence in tumors not expressing TSP-2. In addition, 71.1% of patients with hepatic disease did not express TSP-2 compared with 37.5% who did express this marker. The relationship between TSP-2 and VEGF was examined, which statistically showed better prognosis in patients with TSP-2 expression and no VEGF189 expression. Conversely, VEGF189 expression was associated with higher venous invasion. TSP-2 has been shown in endothelial migration assays to inhibit endothelial cell migration but in higher concentrations than TSP-1 (33).

Plasma thrombospondin levels have also been examined. Yamashita et al., in a study of 115 patients with colorectal cancer, measured the concentration of plasma thrombospondin levels by ELISA and correlated the results to Dukes' Stage and venous invasion. Plasma levels were statistically increased in Dukes' B to D compared with controls and in Dukes' A to C in tumors with no venous invasion and in Dukes' A, C, and D for tumors with venous invasion. Plasma levels were linearly higher in accordance to degree of tumor venous invasion. Venous invasion has been argued as resulting in increased thrombospondin levels, most likely from platelet activations or from the endothelium; however, the authors argue that another source must exist to account for the increase levels in patients with no venous invasion. They further argue that high plasma TSP-1 levels may promote metastatic disease as high TSP-1 has stimulated cell invasion in experimental models and indirectly, because patients with venous invasion have been shown to be at greater risk of developing hepatic metastases (6). Therefore, TSP-1 has been proposed as a marker of possible hepatic dissemination and may support the perivascular and stromal localization of TSP-1 in our series of liver metastases which contrasts the expression of TSP-1 in the primary tumor.

It can therefore be surmised that although the main role of thrombospondin seems antiangiogenic, the exact mechanisms by which it exerts its effects are open to speculation. Some studies have not shown a correlation between thrombospondin and markers of angiogenesis and others have further postulated that it may be the action either of separate thrombospondin domains or indeed of its receptor(s) rather that thrombospondin itself that accounts for the modulation of tumor vascularity. It has also been suggested that the actions of thrombospondin may result from coupling to other growth cytokines such as transforming growth factor-β. The presence of TSP-1 seems to confer a survival advantage, but more data are required.

In this series, the expression of TSP-1 in colorectal liver metastases correlated with poor patient prognosis, contrary to the survival advantage documented in primary colorectal cancers (see above). It may be postulated that in primary tumors, TSP-1 is acting in an antiangiogenic capacity, particularly in less advanced tumors. However, in metastatic deposits, TSP-1 may be acting in a proangiogenic manner or by promoting cell invasion as has been shown in vitro (9).

In conclusion, TSP-1 is negatively prognostic in colorectal liver metastases. The mechanism of this remains unclear and deserves further evaluation.