Mesothelial cells stimulate the anchorage-independent growth of human ovarian tumour cells.

Results are presented which show that mesothelial cells (MC) from ovarian cancer patients can both stimulate and inhibit the clonogenic growth of ovarian tumour cells (TC) in a dose-dependent fashion. TC lines from both non-ovarian and ovarian tumours were variable in their response to MC. Colony formation was rarely induced when the TC population was non-clonogenic and a bladder cell line showed inhibition of colony formation in the presence of MC. Primary tumour cultures from ovarian cancer patients also showed a variable response to MC. Fibroblasts from malignant, benign and non-neoplastic sources were significantly less effective in stimulating the clonogenic growth of responsive cell lines. Conditioned medium was a poor substitute for the presence of intact viable cells, and distance between feeder cell and TC was an important factor in determining the magnitude of response. A significant relationship between the feeder effect of MC and their proliferation in soft agar was observed when epidermal growth factor was used in the medium. The relevance of the findings in the context of the pattern of spread of ovarian cancer is discussed.

Summary Results are presented which show that mesothelial cells (MC) from ovarian cancer patients can both stimulate and inhibit the clonogenic growth of ovarian tumour cells (TC) in a dose-dependent fashion. TC lines from both non-ovarian and ovarian tumours were variable in their response to MC. Colony formation was rarely induced when the TC population was non-clonogenic and a bladder cell line showed inhibition of colony formation in the presence of MC. Primary tumour cultures from ovarian cancer patients also showed a variable response to MC. Fibroblasts from malignant, benign and non-neoplastic sources were significantly less effective in stimulating the clonogenic growth of responsive cell lines. Conditioned medium was a poor substitute for the presence of intact viable cells, and distance between feeder cell and TC was an important factor in determining the magnitude of response. A significant relationship between the feeder effect of MC and their proliferation in soft agar was observed when epidermal growth factor was used in the medium. The relevence of the findings in the context of the pattern of spread of ovarian cancer is discussed.
Dissemination of ovarian cancer beyond the confines of the coelomic cavity is a comparatively rare event for undefined reasons. The infrequent occurrence of haematogeneous metastases in patients with intractable ascites who have received peritoneovenous shunts (Tarin et al., 1984) argues favourably for the concept that locally produced factors within the coelomic cavity can stimulate the proliferation of ovarian tumour cells. Macrophages have been shown to promote the clonogenic growth of such cells (Buick et al., 1980;Welander et al., 1982;Hamburger et al., 1986) and it has been suggested that their presence in the peritoneal cavity produces favourable conditions for tumour growth. The ubiquitous distribution of the macrophage argues against this as the sole explanation for the restricted spread of ovarian cancer and other reasons must be sought. The mesothelial cell (MC) is unique to the peritoneal and pleural cavities and it was therefore postulated that this cell type might enhance the growth of ovarian carcinoma. A study was initiated to examine the above hypothesis and results are presented to support the concept partially. MC were not routinely maintained in the presence of epidermal growth factor (EGF) and hydrocortisone (HC) for the majority of this work, but this has become routine procedure following reports that longer term cultures of MC could be obtained by using these additives (Connell & Rheinwald, 1983). It has also been shown that EGF can enhance the feeder effect of fibroblasts (Rheinwald & Green, 1977;Taylor-Papadimitriou et al., 1977;Stanley & Parkinson, 1979;Peehl & Ham, 1980;Hamburger et al., 1981) and experiments were therefore also done to evaluate the effect of this altered methodology on the feeder effect of MC in soft agar.

Tumour cells
Four ovarian tumour cell lines (OAW 42 (Wilson, 1984), OAW 28, 32M, A7 (Abu Sinna et al., 1979)) and five primary ovarian tumour cultures (two of which were subsequently developed as continuous cell lines) were used as target cells. Eight other cell lines (PA 1, teratoma; BeWo, choriocarcinoma; FL, transformed amnion; HT29, colorectal carcinoma; ZR75, breast carcinoma; T24, bladder carcinoma; TRI26, cheek carcinoma; CCM and Go-GIJKT, astrocytomas) were also tested. All ovarian tumour cell cultures were maintained as monolayer cultures in Dulbecco's modification of Eagle's medium supplemented with 10 or 20% fetal calf serum, 20 IU insulin per litre, 1 mM glutamine, 1 mM sodium pyruvate, 20 pg ml -1 streptomycin, 20 IU ml-1 penicillin and 3.7 g litre-1 sodium bicarbonate (GM), with the exception of 41 M which was grown in a 50: 50 mix of GM and Ham's F12. The presence of epithelial tumour cells in the primary cultures from patients with ovarian carcinoma was confirmed using a combination of morphological criteria, karyology and immunohistochemical staining with HMFG2, a monoclonal raised against human milk fat globulin which recognises a determinant on normal epithelial cells (Taylor-Papadimitriou et al., 1981) and which has been shown to stain more than 94% of human epithelial carcinomas (Ward et al., 1987).
Feeder cells Cultures obtained from malignant ascites often gave rise to mixed populations of proliferating cells comprised of mesothelial cells (MC) and tumour cells (TC). Cultures in which TC either failed to grow or were initially absent were used as a source of MC. Characterisation involved recognition of the following criteria, some of which have been described in the literature (Mouriquand et al., 1978;Whitehead & Hughes, 1975): (i) alteration from an epithelial to a fibroblastic type of growth pattern on first subculture; (ii) small eccentric nucleus; (iii) extensive cytoplasm; (iv) frilled and folded cell margins; (v) densely eosinophilic perinuclear zone; (vi) rapid removal of cells from plastic substrate following PBS wash and exposure to 0.25% trypsin, 0.004% versene; (vii) rapid adherence to plastic; (viii) cell suspensions easily converted to monodispersed cells without clumps; (ix) chromosome number of 46XX with no abnormalities. Further characterisation of MC cultures was carried out using immunohistochemical techniques. Putative MC were stained with a panel of monoclonal antibodies, using a standard indirect immunoperoxidase technique.
Antibodies which were used included HMFG2 (Unipath), anti-vimentin and anti-keratin (Amersham International plc). In this study, cultured MC did not stain with HMFG2 (Wilson et al., 1987), although there are varied reports in the literature which describe positive staining of some benign MC in wax-embedded tissue (Ghosh et al., 1987) and negative staining in wet-fixed smears (Epenetos et al., 1982). MC used in this study stained positively for both vimentin and keratin, as has been previously reported (Connell & Rheinwald, 1983) whereas epithelial cells were, until recently, believed to express only keratin. However, co-expression of both types of intermediate filaments has been reported in ovarian carcinomas (Veda et al., 1987), and vimentin staining has not therefore been taken as definitive proof of MC without other confirmatory evidence. MC cultures were initially maintained on GM supplemented with 10% fetal Br. J. Cancer (1989), 59, 876-882 calf serum. Cultures were passaged weekly using a 1:3 split ratio and cells from confluent cultures at passage 2-4 were used for most experiments. In the latter part of the study MC were maintained on GM supplemented with EGF (5 ng ml -) and hydrocortisone (0.4 Mg ml -1) and cells were then used up to passage 7. Fibroblasts from a variety of sources including malignant tumours, benign tumours, normal endometrium and a cell line CCD-14SK from the skin of a 34-year-old normal female (American Cell Collection) were also used as feeder cells.
Soft agar assay Bases of 1 ml GM in 0.5% agar (Agar Noble) were prepared in 35mm petri dishes (Nunc) and allowed to solidify at 4°C for 10 min. Feeder cells were added as a single cell suspension in a 1 ml layer of GM in 0.5% agar. After solidification of the middle layer, tumour cells were added as a single cell suspension in a 1 ml layer of GM in 0.3% agar. Control plates contained a middle layer consisting of medium and agar only. Cell numbers used varied according to the experiment and will be detailed in appropriate sections. Four dishes per test condition were routinely plated and incubation was carried out at 37°C in an humidified atmosphere of 95% air/5% CO2. Colonies (>50 cells) were counted after 10-14 days, using an inverted microscope set at x 100 magnification.
Mesothelial cell:tumour cell ratio A series of plates were prepared containing standard 1 ml base layers overlaid with 101, 102, 103, 104, 101 and 106 MC in 0.5% agar in GM. Tumour cells were overlaid in a separate top layer at 103, 104 and 105 cells per dish in 0.3% agar in GM over each inoculum of MC or over a middle layer of GM in 0.5% agar. Dishes were scored for colonies after 7-10 days incubation at 37°C.

Conditioned medium (CM)
Growth medium which had been in contact with a confluent monolayer of MC for 3-5 days was used instead of MC in a soft agar assay. The middle MC-containing layer was omitted and CM was used undiluted in the top layer containing the tumour cells. The base layer contained GM, and CM was therefore present at a final dilution of 50%.
Colony stimulating activity of cell-free ascites and serum from the same patient The colony stimulating activity (CSA) of cell-free ascites for TC has been described by others (Uitendaal et al., 1983;Broxterman et al., 1987). The supposition that preferential growth of ovarian cancer in the coelomic cavities is due to locally produced growth factors implies that there should be a difference in the growth promoting properties of serum and ascites, and paired samples of serum and ascites from the same patient were therefore tested for CSA, using OAW 42 as target cells. Aliquots of serum and ascites were clarified by centrifugation and used in the top tumour cell layer, as described for CM (see above).
Increasing distance between feeder cells and tumour cells A series of plates was set up containing four separate layers of soft agar. TC (OAW 42) and MC or fibroblasts (CCD-14SK) were added at 104 cellhml-1 and 105 cells ml-1 respectively (1OMC: ITC) such that feeder cells and TC were present in the same layer, adjacent layers or separated by one or two layers of soft agar. Each layer was allowed to solidify for 5min at 4°C before addition of the next.
Effect of epidermal growth factor (EGF) on the feeder effect of mesothelial cells EGF (Gibco) and hydrocortisone (Sigma) were added to GM to give final concentrations of 5 ng ml-I and 0.4 ,ug ml1 respectively. This supplemented medium was used as a diluent in the MC-containing middle layer. Control plates contained the supplemented middle layer without MC. The final concentration of EGF in the three-layer assay system was therefore 1.7 ng ml -1.

Ratio dependence
Preliminary results clearly showed that MC promoted the clonogenic growth of OAW 42 cells in soft agar, and that the effect was strongly ratio dependent.
Ovarian tumour cell lines Results obtained with three other ovarian TC lines showed that not all TC populations were responsive to the feeder effect of MC (lOMC: lTC). OAW 28 (P15) had a plating efficiency of 0.02% in the absence of MC which increased to 3.2% in their presence. However, 32M (P2-P13) and A7 (P82) were non-clonogenic in soft agar and did not form colonies in the presence of MC.
Primary ovarian tumour cultures Comparison of early (P26) and late (P102) cultures of OAW 42 revealed differences in the responsiveness of TC relating to passage level (Table I). Cells from P26 were responsive but the magnitude of the effect was considerably less than that obtained with P102. Several primary cultures were therefore tested to ensure that the phenomenon was not an artefact relating to long-term culture. Results for seven primary cultures are shown in Table II. Four cultures which formed colonies in soft agar responded to MC with an increase in plating efficiency (llM, x 2.4; 1B, x 14; 25M, x 2.3; 31M, x 5.3) but three others which were nonclonogenic remained non-clonogenic in the presence of MC (1S, 59M, 41M). 59M and 41M were subsequently developed into continuous cell lines.
Conditioned medium (CM) Addition of CM to soft agar cultures of OAW 42 significantly increased its plating efficiency in three separate experiments but the magnitude of the increase was considerably less than that achieved when intact viable MC were included in the soft agar (Table III).
CSA of cell-free ascites and serum from the same patient Pairs of serum/ascitic fluid taken at the same time from seven patients were tested for CSA against OAW 42. Results outlined in Table IV show that differences did exist between serum and ascities. Only 1/7 of the serum samples tested showed significantly greater activity than the corresponding ascitic fluid sample (81D; P<0.005). For the other pairs there was a significant decrease in CSA between serum and ascites in 5/6 pairs. However, cells grown with the human serum still showed significantly greater clonogenicity than those grown with GM in 3/7 pairs, similar clonogenicity in 2/7 pairs and significantly reduced clonogenicity in 2/7 pairs. Increasing distance between MC and TC The effect of increasing distance between target cell and feeder cell was determined on three occasions using 58MC and on three occasions using CCD-14SK at a ratio of 10 feeders: lTC. Results are shown in Table V for two experiments in which 58MC and CCD-14SK were set up at the same time using the same passage of OAW 42. The maximum feeder effect with both types of feeder cell was obtained when feeder cell and target cell were present in the same layer, although CCD-14SK were significantly less effective than 58MC. There was a gradual reduction in feeder effect as distance increased such that when two layers of agar were interspersed between the cell populations the feeder effect of 58MC was 50-52% of the maximum effect obtained, contrasting with 39-41% with CCD-13SK. In the two other experiments (data not shown) there were similar results. When all results obtained were combined a reduction of 55+8% was obtained for 58MC and of 27+22% for CCD-14SK; these values were significantly different (P<0.05; Student's t test). The gross morphology of the TC colonies also varied according to the relative positions of MC and TC. When the two cell types were mixed TC colonies were predominantly solid, containing small, densely packed cells. As distance increased the number of cystic colonies increased, and when MC were absent the colonies were entirely cystic. A total of 100 colonies were scored as dense, loose refractile or cystic, and these results are shown as histograms in Figure 2. The microscopic appearance of the three colony types are shown in Figures 3 and 4.
Specificity of effect A number of non-ovarian TC lines were tested for responsiveness to the feeder effect of MC. The response was variable as was found with ovarian carcinoma (Table VI) with inhibition, stimulation and no effect observed in different cell lines. FL (transformed amnion), PAl (teratoma), HT29 (colorectal carcinoma) and Go-GIJKT (astrocytoma) all showed enhanced clonogenicity in the presence of MC; ZR75 (breast carcinoma), BeWo (choriocarcinoma), TR126 (cheek carcinoma) and CCM (astrocytoma) were all unresponsive; and T24 (bladder carcinoma) showed reduced clonogenicity in the presence of MC. Feeder cells from a variety of sources were also tested for their ability to promote the clonogenic growth of TC. These included MC from cancer patients (n = 13), fibroblasts from benign tumours (n =2) and malignant tumours (n= 1), as well as normal fibroblasts from endometria obtained at hysterectomy (n = 2), fibroblasts from a follicular cyst of pregnancy (n= 1) and a fibroblast cell line (CCD-14SK) from the skin of a 34-year-old female. Results are shown in Table   Table    Taking into account the enhanced clonogeneicity of the tumour cells in the presence of EGF, the magnitude of the reduction in feeder effect ranged from 1.04 to 12.14. The MC were also variably responsive to EGF and there was a significant relationship between the magnitude of the reduction in feeder effect and the response of MC to EGF (r=0.906, P.0.01), as evidenced by the increase in clonogenicity of MC in the presence of EGF. There was also a significant relationship between the control colony number of MC (i.e. in the absence of EGF) and the magnitude of the stimulatory effect of MC (r=0.712, P.0.05). It should be noted here that when these experiments were done MC were routinely maintained on GM supplemented with EGF and hydrocortisone. Under these growth conditions some colony formation occurred in soft agar even when the GM contained neither growth factor. The morphology of the colonies was distinctly different from TC colonies and since they were also present in a different layer from the TC, there was no possibility of confusion during colony counting. The results obtained when 32M was used as target differed in that these cells were unresponsive to MC but formed small clusters (6-12 cells) in the presence of EGF. Interestingly, in spite of their unresponsiveness to MC, their response to EGF was still reduced in the presence of MC.

Discussion
The ability of fibroblasts to promote the growth of TC and normal epithelium is well documented (Stanley & Parkinson, 1979;Taylor-Papadimitriou et al., 1977;Peehl & Ham, 1980;Hamburger et al., 1981;Brattain et al., 1982;Citron et al., 1986;Gallie et al., 1982;Kirk et al., 1981Kirk et al., , 1983Koopman & Cotton, 1984;Laboisse et al., 1981;Rheinwald & Beckett, 1981;Rosenstraus et al., 1984) but this is the first report on the feeder effect of MC for human ovarian cancer. It was originally hypothesised that locally produced factors could contribute to the restricted pattern of spread of ovarian cancer and the finding that MC can stimulate the clonogenic growth of ovarian TC supports this idea. Ascitic fluids from cancer patients also possess colony stimulating activity for some TC (Uitendaal et al., 1983;Broxterman et al., 1987) and this may be due to conditioning of the fluid by MC and/or TC. Although human serum/plasma also promoted clonogenic growth of TC above control levels (GM) in 3/7 samples tested, it was interesting to observe that the colony stimulating activity of serum was appreciably lower than ascitic fluid from the same patient in 5/7 pairs tested. This finding, together with the relative inactivity of conditioned medium compared with the continuous presence of viable cells, suggests that locally produced factors can stimulate clonogenic growth of ovarian TC and that these factors may not reach the systemic circulation in concentrations which are high enough to promote clonogenic growth. This is also suggested by the finding that increasing distance between MC and TC reduces the feeder effect. Although fibroblasts also stimulated the clonogenic growth of TC, this effect was significantly less than that obtained with MC. The observation that colony morphology changed depending on the distance between feeder cell and tumour cell was interesting. The cystic colonies represent a differentiated function of the epithelial tumour cells, arising due to fluid transport. It is possible that the dense colonies reflect a lack of differentiation since it has been shown that fibroblast feeder layers can inhibit the differentiation of retinoic acid-treated embryonal carcinoma cells (Rosenstraus et al., 1984). Even though the fibroblast is ubiquitously distributed in the body, a metastasising TC is unlikely to encounter fibroblasts in the same aPlating efficiency expressed in parentheses as a percentage of 1, i.e. maximum stimulation.
1, MC and TC in same layer; 2, MC and TC in adjacent layers; 3, MC and TC separated by one layer; 4, MC and TC separated by two layers; 5, TC only.     numbers as those of MC in the coelomic cavity and an excess of feeder cells is obviously necessary for stimulation. It is therefore suggested that the coelomic cavity can provide a uniquely favourable environment for TC growth, in that mesothelial cells are present in large numbers. The generally restricted spread of ovarian cancer beyond the confines of the coelomic cavity and the failure of TC which reach the .nna I .... r qL VV" r n [ systemic circulation via peritoneovenous shunts to proliferate (Tarin et al., 1984) may reflect this requirement. Mechanisms underlying the epithelial-mesenchymal interactions have not been elucidated and no clear picture emerges from a study of the literature. Variable factors have included source of feeder cell, type of target cell, method of culture, treatment of feeder cell and type of medium. The general consensus appears to be: (i) that fibroblasts from any source including normal adult, neonatal, tumour-derived, embryonic and non-human can stimulate epithelial cell growth; (ii) that a variety of normal and transformed cell types can respond to the feeder effect; (iii) that co-cultivation of feeder cells and epithelial cells gives considerably better growth than does the use of conditioned medium. Contradictions to these general conclusions may be found, however. Kirk et al. (1981Kirk et al. ( , 1983 concluded that the feeder effect was dependent on non-proliferation of feeder cells since monolayers of feeders were inhibitory whereas agar-suspended feeders stimulated the growth of target cells. These authors also found differences in the inhibitory behaviour of fibroblasts from different sources. Thus, neonatal human lung fibroblasts were inhibitory in monolayer, kidney tumour derived fibroblasts were stimulatory and fibroblasts from the skin of a Down's syndrome neonate were both inhibitory and stimulatory depending on the ratio of feeder cell to target cell (Kirk et al., 1983). Other authors have found no differences between fibroblasts from different sources (Taylor-Papadimitriou et al., 1981;Gallie et al., 1982), while in the present study only one (endometrium derived fibroblasts) of fourteen feeder cell populations failed to enhance tumour growth. Although Kirk et al. (1983) reported that mitomycin C treatment of monolayers abolished their inhibitory effect, other workers have found that untreated monolayers could enhance tumour growth (Brattain et al., 1982;Gallie, et al., 1982). Results from the present study suggest that variability between difference studies may be due to a combination of differences in feeder cell to target cell ratios and the distance between feeder cell and target cell. In the two studies where monolayers and target cells were separated by a layer of agar either no stimulation of target cell growth (Hamburger et al., 1981) or inhibition of target cell growth (Kirk et al., 1983) was found. This idea is supported by the finding that feeder effect decreased with increasing distance between feeder cell and target cell.
The finding that EGF reduced the feeder effect of MC contrasts with the results of other workers showing an enhancement of the feeder effect in the presence of EGF (Rheinwald & Green, 1977;Stanley & Parkinson, 1979;Taylor-Papadimitriou et al., 1977;Peehl & Ham, 1980;Hamburger et al., 1981). In all these studies the feeder cells were either irradiated or treated with mitomycin C. The dual finding of significant correlations between: (i) response of MC to EGF and reduction in feeder effect and (ii) colony formation by MC in the absence of EGF and magnitude of the feeder effect, provides strong evidence for an association between proliferation and feeder effect, as previously reported by Kirk et al. (1981Kirk et al. ( , 1983.
It would appear that the effector molecule may be inhibitory or stimulatory depending on the target cell and the relative numbers of MC and TC. 32M was unresponsive to MC but had a reduced response to EGF in their presence. While this may have been due to reduced availability of EGF because of binding by MC, it could also have been due to inhibition of 32M by MC, a phenomenon which could only be observed when the cells were actually stimulated to proliferate by EGF. Further support for the idea of a dual role for the effector molecule comes from the marked inhibition of clonogenicity of T24 in the presence of MC. Feeder effect was also inhibitory when TC were present in excess of MC.
Whether any of these interactions are of relevance in vivo cannot be directly concluded from these results but they do provide indirect evidence for the view that epithelialmesenchymal interactions within the coelomic cavity may be important in influencing the pattern of spread of ovarian cancer.