Localisation of granulocyte macrophage colony-stimulating factor in human long-term bone marrow cultures Biological and immunocytochemical characterisation

The role of growth factors in promoting the survival, proliferation and development of multipotent haemopoietic stem cells and their more developmentally restricted progeny has been unequivocally established using short-term in vitro clonogenic culture systems (Fauser and Messner, 1979; Johnson, 1984; Toogood et al., 1980). In other cul-ture systems, however, haemopoiesis occurs in the absence of added growth factors, provided that the haemopoietic cells are cultured in association with a bone marrow stroma (Dexter et al., 1984). The stroma of these ‘long-term bone marrow cultures’ consists of a complex network of fibrob-lasts, macrophages, adipocytes, endothelial cells, adventi-tial reticular cells and extracellular matrix molecules, which are cells that are representative of the bone marrow stroma in vivo. The adherent stromal cells exert their effect by direct cell contact with the haemopoietic cells, and if cell contact is prevented the haemopoietic cells die (Dexter et al., 1978; Dexter and Allen, 1984). This suggests that appre-ciable levels of diffusible growth factors are not normally produced by the stromal cells in these cultures. What then is the role of growth factors in stromal cell-mediated haemopoiesis?

With the exception of macrophage colony stimulating factor (M-CSF), it has been difficult to detect biologically active haemopoietic cell growth factors in the supernatant media from long-term bone marrow cultures (Lanotte et al., 1982; Heard et al., 1982; Shadduck et al., 1983). However, radioimmunoassay techniques and molecular probes for mRNA of haemopoietic colony stimulating factors (CSF) such as macrophage CSF, granulocyte-macrophage CSF and granulocyte CSF has shown that stromal cells are capable of producing at least some of these growth factors. Furthermore, stimulation of long-term bone marrow cultures with phytohaemagglutinin (PHA), lipopolysaccharide (LPS) or irradiation, results in the accumulation of appreciable levels of growth factors into the supernatant (Gualtieri et al., 1984; Li et al., 1987; Gimble et al., 1989). It is likely, therefore, that in normal ‘steady-state’ haemopoiesis these growth factors are indeed being produced but in relatively low abundance, and are able to stimulate their haemopoietic target cells only at a local level, i.e. in proximity to the producer cells or in regions where the growth factors are being sequestered. With respect to the latter suggestion, it has been shown that molecules of the extracellular matrix (particularly heparan sulphate) can sequester growth factors such as GM-CSF and interleukin-3 (IL-3), and ‘present’ them in a biologically active form to the appropriate haemopoietic target cells (Roberts et al., 1988). Additionally, extracts of bone marrow stromal cells are capable of stimulating the proliferation and development of granulocyte/macrophage progenitor cells (Gordon et al., 1987). Moreover, some of the haemopoeitic cell growth factors (e.g. the ligand for the tyrosine kinase receptor encoded by the c-kit proto-oncogene, known as stem cell factor, SCF) can be produced as transmembrane proteins (Huang et al., 1990; Anderson et al., 1990; Flanagan and Leder, 1990), and in this form they exert their effect only in the context of intimate cell to cell interactions. In other words, the requirement for intimate stromal cell/haemopoeitic cell contact in long-term bone marrow cultures probably reflects, at least in part, localised production of haemopoeitic cell growth factors.

To our knowledge, however, there has been no systematic study of the cells involved in the production/sequestration of haemopoietic growth factors or their longevity in long-term cultures. Here, we have used monoclonal anti-bodies to GM-CSF and IL-3 to investigate these points.

The technique has been described in detail elsewhere (Coutinho et al., 1990). Briefly, red blood cell (RBC)-depleted bone marrow aspirates at concentrations of 1.5×10⁶ cells/ml were incubated in 25 cm² tissue culture flasks (NUNC) in a 10 ml total volume of 10% FCS, 10% horse serum, 5×10⁻⁷ mol/l hydrocortisone sodium succinate and single-strength Iscove’s medium (350 mosmol/kg) supplemented with penicillin and streptomycin. The cultures were gassed with 5% CO₂ in air and incubated at 33°C. After 7 days and subsequently at weekly intervals, the cultures were fed by removing half of the supernatant and replacing this with an equal volume of fresh growth medium. Over a two-to three-week period, an adherent stromal cell layer is established that contains islands of haemopoietic activity (Toogood et al., 1980). Subse-quently, these cultures can be maintained for periods greater than 12 weeks before haemopoiesis begins to decline (Coutinho et al., 1990).

The GM-CSF monoclonal antibodies used in this study were pro-duced by one of us (D.F.) in the Simon Flavell Leukemia Research Laboratories, Southampton. The method of production was, briefly, as follows: spleen cells from Balb/c mice immunised with non-glycosylated rhGM-CSF (Glaxo) were fused with NS-1 myeloma cells in the presence of polyethylene glycol 4000. Hybridoma supernatants were screened for antibody activity to GM-CSF by enzyme-linked immunosorbence assay (ELISA). The two IgM antibodies selected for this study were DF 2740 and SF GM-CSF2. Control serum and control Ig subtype-specific mono-clonal antibodies were obtained from Dr D. Flavell and Dr P. Stern (Paterson Institute Cancer Research, Manchester). Commercial anti-GM-CSF antibody was purchased from Genzyme (Koch-Light). Monoclonal antibodies to rhIL-3 were supplied by the Genetics Institute. Monoclonal antibodies to α-tubulin and actin were a gift from Dr T. Sherwin (University of Manchester).

rhGM-CSF (Glaxo) was run on a 12.5% sodium dodecyl sulphate/polyacrylamide gel (SDS/PAGE) along with relative mol-ecular mass markers and transferred by blotting on to nitrocellu-lose paper (NC). Strips of NC paper were probed with each monoclonal antibody (ascites diluted 1:100 in antibody buffer) and the purified control antibody (DFT1 or DFB2). Following incubation with biotinylated sheep anti-mouse antibody (diluted 1:300 in PBS) the NC strips were developed using the ABC method (Dakopatt) and diaminobenzidine tetrahydrochloride (DAB) sub-strate (Sigma).

The long-term bone marrow cultures were gently washed with phosphate buffered saline (PBS) at room temperature and adherent layers were fixed in situ for 10 to 15 minutes in freshly prepared 3% paraformaldehyde in PBS. After fixation, cells were washed with PBS and the growing surface of the flask was removed and cut into 2 cm squares under PBS to prevent air drying. Some of the preparations were permeabilised in 50% ethanol for 10 minutes at room temperature or methanol at −20°C for 5 minutes. All preparations were carefully washed in PBS and non-specific binding sites were blocked by incubating with 5% bovine serum albumin (BSA) in PBS for 30 minutes. It is important to protect cells from desiccation during the incubation peri-ods and cell preparations were maintained in a humidified atmosphere. The preparations were again washed to remove unbound BSA, then incubated with hybridoma supernatants or primary antibodies for 30-40 minutes at 4°C. Binding was visualised using a secondary anti-mouse antibody fluorescein-conjugated and diluted 1:15 in PBS/1% BSA. Non-permeabilised samples were treated in a similar manner after first omitting the ethanol or methanol step. All preparations were mounted cell side uppermost, on a microscope slide in non-fade mountant (Uvinert, BDH) with a coverslip over the cell layers and observed using a fluorescence microscope.

The organisation of F-actin was monitored using rhodamine-labelled phallacidin (Molecular Probes Inc.). Preparations were processed as for indirect immunofluorescence, but before adding the primary antibody they were incubated in rhodamine-phal-lacidin in PBS for 20 minutes at room temperature. All preparations were washed in PBS, then stained with monoclonal anti-GM-CSF and visualised with a secondary anti-mouse fluorescein-conjugated antibody. Similarly, permeabilised preparations were first incubated with either anti-tubulin or anti-vimentin and visualised using an anti-mouse fluorescein-conjugated secondary antibody.

Cells at a concentration of <10⁵/ml were inoculated in Iscove’s medium supplemented with 20% FCS, 1% BSA and 10% 5637 conditioned medium or appropriate concentrations of the recom-binant growth factors. The three purified recombinant growth fac-tors used were obtained from Amgen (rhG-CSF 5×10³ units/ml), Biogen (rhGM-CSF 5×10⁴ pg/ml) and the Genetics Institute (IL3 20 units/ml) and used at the concentrations indicated. Dilutions of each factor were prepared under sterile conditions in PBS containing 0.1% BSA. Anti-GM-CSF antibodies were used over the range of concentrations indicated in the text. Agar was added to a final concentration of 0.33% and the culture mixture was plated in volumes of 1 ml in 35 mm Petri dishes (Falcon), in duplicate. After an 11-day incubation period at 37°C in a fully humidified atmosphere of 5% CO₂ in air, colonies were scored using a dissecting microscope.

The monoclonal antibodies DF 2740 and SF GM-CSF2 recognise recombinant human GM-CSF as shown by west-ern blotting, in Fig. 1. Both antibodies reacted with a protein band at 14 kDa, though with varying intensities. To determine the specificity of these antibodies, studies based on their ability to inhibit colony formation in the presence of recombinant growth factors were performed. Table 1 shows the degree of inhibition of GM-CFC colony formation in the presence of antibody and growth factors. Both antibodies compete strongly for rhGM-CSF and native GM-CSF present in 5637 conditioned medium. There was little or no cross-reactivity with rh-IL3 or rhG-CSF. One micro-gram of the commercial antibody completely neutralises 50 units of GM-CSF and there is no detectable cross-reaction with human G-CSF or IL-3 as determined by neutralisation studies (Genzyme). The IL-3 antibody was also effective in inhibiting colony formation of cells stimulated with recom-binant human IL-3 (data not shown).

In LTBMC, the pattern of staining of antibodies DF 2740, SF GM-CSF2 and the commercial anti-GM-CSF antibody were strikingly similar in the non-permeabilised cultures. In fibroblasts (one of the most abundant cell type in the early stages of formation of stroma in long-term (LT) cul-tures) the stain was deposited in dense patches along the membrane boundary of these cells (Dexter and Allen, 1984). This can be seen more clearly when the fluorescent picture is compared with that in phase-contrast (Fig. 2A and B) and illustrates one of the main advantages of the immunofluorescence technique when used in conjunction with phase-contrast morphology: it allows simultaneous immunological study of the different cellular elements present in the adherent layers of the bone marrow stroma. The staining was observed between adjacent cells and on the cell surfaces, suggesting secretion or deposition of the growth factor, possibly on extracellular matrix, in certain areas. These labelled areas resembled fibrillar patches that increased in abundance with age of the culture, but had almost disappeared by the twelfth week (Fig. 2C-F). Most macrophages could be recognised in the preparations by the bright yellow/orange autofluorescence of their secondary lysosomes, as indicated in Fig. 2G. Anti-GM-CSF staining of macrophages was heterogeneous, some cells bound the antibody whereas other macrophages were clearly unlabelled (Fig. 2G and H). By about the fourth week of culture, the adherent stromal layer was almost confluent and the adipocytes began to increase in numbers. Staining of these adipocytes with anti-GM-CSF was rather diffuse, as shown in Fig. 2E. At this time, haemopoietic foci composed of early progenitor cells are also apparent under the adherent layer, forming characteristic ‘cobblestone’ regions. Fig. 3A shows one such focus site under fluorescence microscopy and the same area viewed by phase-contrast in Fig. 3B. It is evident that the progenitor cells within these foci are unstained, as only the fibrillar patchy staining asso-ciated with the stromal cells (seen in Fig. 2C-F) is observed. By the sixth week the stroma had completely covered the base of the flask and became increasingly more multilayered. It was difficult to distinguish labelling on any indi-vidual cells after this period.

To try and determine which cells were producing and possibly secreting GM-CSF, cells from the LTBMCs were permeabilised and the cytoplasm was stained with the anti-GM-CSF antibodies. Again, all antibodies showed similar staining patterns. Fluorescence microscopy of these cultures revealed an extensive fibrillar network of long fine processes extending throughout the cytoplasm of both fibroblasts and endothelial cells (Fig. 3C and D). This cyto-plasmic staining with the anti-GM-CSF antibodies, although reminiscent of patterns seen with cytoskeletal anti-bodies of proteins such as tubulin and vimentin, was quite distinct. As cultures aged and became more multilayered, the fibrillar ‘cytoskeletal’ pattern became less clear, although staining was still evident (Fig. 3E). Macrophages again showed a variety of labelling in the permeabilised cultures. Those cells at the surface of the stroma were often brightly labelled whilst labelling of the ‘spread’ macrophages within the stroma was variable (Fig. 3E). Surprisingly, few granulocytes or their precursors within the stroma were stained (compare Fig. 3E and F). This was not due to any difficulty experienced by the antibody in pene-trating the multilayer, as granulocytes in comparable cultures, treated in the same manner and labelled with an actin antibody, were vividly labelled (Fig. 3G and H). By the end of three months in culture, there was virtually no labelling of any adherent cells with anti-GM-CSF. This correlates well with the haemopoietic activity of long-term cultures, which begins to decline after eight to ten weeks in culture and is virtually absent at three months (Coutinho et al., 1990). Any staining with anti-GM-CSF observed at this time was very weak in areas distributed randomly through-out the adherent layer. It was not possible to identify whether the staining was associated with any individual cells, owing to the intricate meshwork of interconnected cells at this stage of the culture. Though vesicles of mature adipocytes were clearly seen in phase-contrast, the diffuse labelling previously observed was absent.

Attempts to label either the cell surface or cytoplasm of adherent cells with the monoclonal antibodies to human IL-3 proved negative.

In control experiments, irrelevant antibodies of the same isotype also gave negative results and autofluorescence of the macrophages only was observed. A second series of control experiments, where the primary antibody was omit-ted and cells incubated with secondary antibody alone, also resulted in unlabelled cells.

Having identified potential GM-CSF in the LTBMC by immunocytochemistry and observing the ‘cytoskeletal’ pat-tern of staining in the permeabilised cells, we tried to clarify whether the anti-GM-CSF antibodies co-localised with known cytoskeletal proteins. In double-labelling experiments, the distribution of GM-CSF antibody was not coin-cident with that of actin as determined using rhodamine-phallacidin (Fig. 4A and B). Actin could be seen as linear filaments criss-crossing throughout the body of the cell whereas anti-GM-CSF appeared as a dense mass of fila-ments extending from the nucleus to the cell membrane. Tubulin staining was not typical of that seen with non-haemopoietic fibroblasts (Fig. 4C) and was clearly distinguished from the anti-GM-CSF labelling. Similarly, the staining pattern seen with anti-GM-CSF was also distinct from that observed with anti-vimentin antibodies (Fig. 4D).

The results from this study show that GM-CSF or an immunologically related protein is produced by adherent cells of the long-term bone marrow stroma. The antibodies raised against recombinant human GM-CSF recognised the recombinant protein on western blots and neutralised its biological activity in colony assays. Additionally, the anti-bodies inhibited GM colonies in the presence of 5637 conditioned medium, indicating recognition of native GM-CSF. Although considerable progress has been made in the analysis of the stromal components in the long-term cul-ture system, the contribution of each stromal cell type to the support of haemopoiesis has not been determined. Most of the cells of the bone marrow stroma can be induced to produce haemopoietic growth factors but whether these cells produce the same factors in the steady-state is unclear. Our finding that GM-CSF is present in bone marrow stromal cells is in agreement with those of others (Gualtieri et al., 1987; Gordon et al., 1987; Eaves et al., 1991; Charbord et al., 1991). In addition, we have shown that the factor is predominantly produced by fibroblasts and endothelial cells, as detected by immunofluorescense. The high local extracellular concentrations of antibody associated with fibroblasts and endothelial cells appear to confirm growth factor binding in these areas. As the binding was in such close proximity to the surfaces of the cells (Fig. 2A) in the early stages where the cultures were sparse, it is logical to assume that it is synthesised by these cells. Additionally, the cytoplasmic labelling of both cell types raises the dis-tinct possibility that these cells both synthesise and secrete GM-CSF. It is not certain how the factor is retained at the cell surface but it is tempting to speculate that this may be via an association with the extracellular matrix (ECM). Previous work has already demonstrated that factors such as GM-CSF can bind to components of the ECM and retain biological activity (Roberts et al., 1988). Clearly, binding to the ECM would limit distribution very tightly to the cells expressing growth factor and would also provide a means of concentrating the factor. From the results obtained, GM-CSF was seen to be concentrated in discrete fibrillar patches, broadly distributed over the surface of the adherent layer (Fig. 2C-F). Whether these patches are associated with molecules of the ECM has not been determined. How-ever, if GM-CSF is bound, it would help to explain the observation that only extracts from cultured bone marrow stroma, but not supernatants, were capable of stimulating GM-CFC (Gordon et al., 1987).

Heterogeneity in macrophage labelling was apparent in the cultures, since not all macrophages bound anti-GM-CSF. Two populations of macrophages have been identi-fied in LTBMC; resident macrophages of the bone marrow and macrophages developing from monocytes (Mori et al., 1990). The labelled and unlabelled cells may reflect these two populations and the labelling maybe a result of sequestering GM-CSF produced by other cells. There was a sub-population of ‘spread’ macrophages that stained with the GM-CSF antibody and it is possible that this small population may be producing the growth factor.

Anti-GM-CSF staining appeared to correlate with the haemopoietic activity of the cultures. This activity is often measured as the yield of progenitor cells (GM-CFC) recovered in the culture supernatant, with production of these progenitor cells declining after the eighth to tenth week in culture. At the peak of haemopoietic activity (4-7 weeks) the cell preparations were abundantly covered with the characteristic patches associated with GM-CSF staining (Fig. 3E). As activity declined, staining was reduced in parallel and was virtually absent by week 12 (Fig. 3F). It may be that the decline in GM-CSF production is at least partially responsible for the fall in the number of GM-CFC progenitors. In agreement with this, Eaves et al. (1991) suggested that GM-CSF, either alone or together with other growth factors produced by stromal cells, may be a direct positive mediator of primitive progenitor activation in the LTBMC system.

In contrast, all attempts to label stromal cells with anti-IL-3 monoclonal antibodies were unsuccessful. Neverthe-less, this provided a good negative control and confirmed results of other workers, who were all unable to detect IL-3 in LTBMC by either biological assays or conventional (non-PCR) molecular techniques (Spooncer et al., 1986; Li and Johnson, 1985).

The specificity of the antibody had been previously demonstrated, in that the anti-IL3 antibody blocked the formation of colonies stimulated with human recombinant IL3 (data not shown). From the immunofluorescence results we conclude that the antibody is probably not reacting with vimentin, tubulin or actin, as the immunofluorescent pat-terns are quite distinct from that of GM-CSF. In addition, the fluctuation of GM-CSF binding over three to twelve weeks in culture would be inconsistent with an antibody recognising a stable cytoskeletal protein. These studies produced some interesting results. Although the stroma contains several cell types, only two of them, fibroblasts and endothelial cells, are predominantly involved in the pro-duction of GM-CSF in LTBMC. At least two populations of macrophages exist but only a small fraction of these may be involved in the secretion of GM-CSF. GM-CSF can be localised to certain areas where it may be bound to the ECM. Finally, staining with the antibodies appears to parallel haemopoietic activity. Further experiments involving double labelling and in situ hybridisation using RNA probes to growth factors should enable us to determine the nature of the factor-producing cells.

This work was supported by the Cancer Research Campaign. We thank Sandra Rutherford for her expert technical assistance.