FGF signalling is required for differentiation-induced cytoskeletal reorganisation and formation of actin-based processes by podocytes

Fibroblast growth factors (FGFs) constitute a large family of signalling molecules that participate in regulation of basic cellular processes such as proliferation, survival, migration and differentiation (Szebenyi and Fallon, 1999). FGFs have also been implicated in modulating morphogenetic processes involving cellular rearrangements and tissue remodelling (Hogan, 1999; Martin, 1998). For example, analysis of mouse embryos deficient in FGF receptor 1 (FGFR1) has shown that FGF signalling controls the movement of epiblast cells through the primitive streak and their conversion into mesenchymal cells during gastrulation (Ciruna et al., 1997; Deng et al., 1994; Yamaguchi et al., 1994). FGF signalling also appears to control epithelial to mesenchymal transitions (EMT) of other cell types (Migdal et al., 1995; Valles et al., 1990). For example, rat bladder carcinoma cells, which grow as epithelial cell clusters in vitro, are converted into migratory fibroblast-like cells after FGF stimulation (Valles et al., 1990). FGF signalling induces expression of the transcription factor slug (Buxton et al., 1997; Nieto et al., 1994; Savagner et al., 1997) and the intermediate filament protein vimentin (Schmid et al., 1979); both are markers of a transition to mesenchymal cell state.

Mature podocytes (or glomerular visceral epithelial cells) are highly differentiated cells within the kidney glomerulus that function in primary renal filtration (Saxén, 1987). During nephrogenesis podocytes originate as immature epithelial cells from the metanephric mesenchyme after induction by the ureteric bud (Saxén, 1987; Sorokin and Ekblom, 1992). During their subsequent differentiation, expression of epithelial markers ceases (Garrod and Fleming, 1990; Tassin et al., 1994), and mesenchymal markers such as vimentin and actin-associated synaptopodin are induced (Mundel et al., 1997a; Nagata et al., 1993). These changes coincide with extensive cytoskeletal reorganisation and extension of actin-based, branched cellular processes, the so-called foot processes (Saxén, 1987). Thus, podocytes appear to undergo EMT-like changes during their terminal differentiation. In support of this proposal, glomeruli of mice deficient for the homeobox transcription factor Pod-1 are nonfunctional due to a failure in terminal differentiation of podocytes (Quaggin et al., 1999), and morphological analysis showed that mutant podocytes were arrested in an epithelial state. Several FGFs and their receptors are expressed by nephrons (including podocytes [(Cancilla et al., 1999; Cauchi et al., 1996; Dono and Zeller, 1994; Mason et al., 1994; Ohuchi et al., 2000; Peters et al., 1992); and this study]) and are required for kidney morphogenesis (Celli et al., 1998; Ohuchi et al., 2000; Qiao et al., 1999; see also Discussion). Of relevance to the present study, FGF2 proteins are upregulated during podocyte differentiation and remain expressed in functional podocytes of the adult kidney (Dono and Zeller, 1994). However, the potential role of FGF2 signalling during podocyte differentiation and function remained unclear as no gross kidney defects were reported in mice carrying a Fgf2 loss-of-function mutation (Dono et al., 1998; Ortega et al., 1998).

To identify and study possible FGF signalling functions during podocyte differentiation and process formation, we used an established in vitro culture and cell differentiation system for mouse podocytes (Mundel et al., 1997b) based on conditional immortalisation (Jat et al., 1991). Conditionally immortalised mouse podocyte (MPC) cells cultured under permissive conditions grow as undifferentiated epithelial cells, whereas they develop mesenchymal characteristics similar to mature podocytes in vivo upon induction of post-mitotic differentiation. In particular, differentiating MPC cells reorganise their cytoskeleton and form branched actin-based processes, similar to foot processes (Majumdar and Drummond, 1999; Mundel et al., 1997b; Nagata and Watanabe, 1997). Here we have used the MPC cell culture system to study the differentiation potential of Fgf2 mutant podocytes in culture. Our studies indicate that FGF signalling is required for the EMT-like changes during differentiation, which results in actin cytoskeletal reorganisation and process formation.

Conditionally immortalised MPC cell clones were isolated as previously described (Mundel et al., 1997b) from glomeruli of wild-type and Fgf2 mutant (Dono et al., 1998) adult mice of mixed genetic background harbouring a single copy of the H-2K^b−tsA58 large T-antigen (Jat et al., 1991). Initial characterisation revealed that 27 of the 30 isolated Fgf2 mutant MPC clones displayed morphological alterations as the ones shown in Fig. 2B,D. Three of 30 mutant MPC clones were morphologically indistinguishable from wild-type MPC cells (Fig. 2A) and not further analysed. Six (of eight) randomly chosen wild-type and six (of 27) Fgf2 mutant MPC clones were used to generate cell pools by mixing of equal numbers of cells from clones with similar proliferation characteristics under permissive conditions. MPC cell pools were used for all studies in order to exclude alterations of results due to clonal variations. Such MPC cell pools were repeatedly prepared from frozen stocks of the original clones (passages 3-6) and used for up to five passages only.

MPC cell pools were cultured in RPMI medium (containing 10% FCS and 100 U/ml penicillin-streptomycin, Sigma) under either permissive (33°C plus 20 units/ml γ-interferon) or nonpermissive (37°C without γ-interferon) conditions as described in detail previously (Mundel et al., 1997b). For cultures under nonpermissive conditions (to induce post-mitotic differentiation), MPC cells were seeded at a density of 0.5-1×10⁶ cells per 10 cm dish (NUNC) and split 1:5 after 3-4 days. For subsequent analysis by immunofluorescence, MPC cells were plated at a density of 0.1-0.5×10⁶ cells per cm² on collagen type-I-coated glass coverslips at day 3-4. Differentiated MPC cells were analysed in general after 7 days of culture under nonpermissive conditions.

Mitotic MPC cells were detected using the 5-bromo-2′-deoxy-uridine (BrdU) labelling and detection kit II (Boehringer Mannheim) as described (Mundel et al., 1997b). Proliferating cells in chicken embryonic mesonephric kidneys were labeled in organ culture for 48 hours with 100 μM BrdU. Mitotic and FGF2 expressing cells were detected on histological sections as described previously (Dono et al., 1998).

MPC cell morphology was assessed by microscopic observation of living cells. Images were captured using a COHU CCD camera mounted on a Nikon eclipse TE 200 microscope in combination with the Scion Image software program. For indirect immunofluorescence, cells cultured on glass coverslips were fixed with either methanol/acetone (1:1) for 15 minutes at −20°C (ZO-1) or 4% PFA for 10 minutes at ambient temperature (all other markers). Following fixation, cells were washed with PBS, permeabilised with 0.3% Triton X-100 in PBS for 5 minutes (PFA fixed cells only) and blocked in 3% BSA, 3% FBS and 0.2% gelatine in PBS for 1 hour at room temperature (RT). Subsequently, cells were incubated for 1 hour at RT with either antibodies against FGF2 (Dono and Zeller, 1994), synaptopodin (Mundel et al., 1997a), paxillin (Transduction Laboratories), vimentin (C-20; Santa Cruz) or ZO-1 (Zymed) in blocking solution. After washing with PBS containing 0.2% gelatine and 1% Triton X-100, cells were incubated for 30 minutes at RT with the appropriate secondary antibodies. FITC-conjugated phalloidin (0.5 μg/ml, Sigma) was used to visualise actin filaments and DAPI (Boehringer) for nuclei. Images of immunofluorescent cells were captured with the OpenLab 1.7.6 software program (Improvision) using a Photonic-Science cooled CCD camera attached to a Zeiss Axioscope microscope. All illustrations were assembled and processed digitally using Adobe Photoshop 5.0.

MPC cells were harvested after mild trypsin digestion and washed with ice-cold PBS containing 1 mM PMSF. Cell pellets were resuspended in ice-cold extraction buffer (20 mM Tris, pH 7.5, 0.5 M NaCl, 1% Triton X-100, 1 μg/ml aprotonin, 1 μg/ml leupeptin, 0.5 μg/ml pepstatin and 0.5 mM PMSF; all Sigma) at a concentration of approximately 10⁷ cells/100 μl. Suspensions were first homogenised, then sonicated on ice (20 Watts for 1 minute) and cleared by centrifugation (10,000 g for 30 minutes at 4°C). Separation of suspension into nuclear, cytoplasmic and membrane fractions was performed as described by (Dono and Zeller, 1994) with the following modifications: the crude cytoplasm was separated into membrane and cytoplasmic fractions by ultra-centrifugation (100,000 g for 30 minutes). The membrane pellets were dissolved in RIPA buffer (50 mM Tris-HCl, pH 7.4, 30 mM NaCl, 5 mM EDTA, 1% NP-40, 1% deoxychloate and 0.1% SDS). The protein contents of all total extracts and fractions were determined (Biorad kit) and extracts normalised. For FGF2 immunoblot analysis, 1 mg of total protein was heparin enriched. Antibodies against α-tubulin (Sigma) and c-Jun/AP1 (Santa Cruz) were used as controls to assess the quality of the nuclear-cytoplasmic fractionation. Immune complexes were visualised by chemoluminescence according to manufacturer’s instructions (Amersham). Three independent experiments yielded results very similar to the ones shown in Fig. 1E,F.

Total RNA was extracted from cultured cells or tissues using the RNeasy kit (Qiagen) according to manufacturer’s instructions. 10 μg of RNA were reverse transcribed with oligo(dT) and Superscript II (Gibco-BRL). The resulting cDNA samples were normalised by PCR using GAPDH primers as standard. Initially, 1/50 of the cDNA was used for PCR amplification using gene-specific primers and amplification conditions. The numbers of cycles for each gene was adjusted in test experiments to be within the linear range of detection. All semi-quantitative RT-PCR analysis was performed using RNA from at least three independently prepared batches and yielded results identical to the ones shown in Figs 3-5.

FGF2 proteins are upregulated in podocytes concurrent with their mitotic arrest in vivo (compare Fig. 1A with Fig. 1B) and expression persists into adulthood (Dono and Zeller, 1994). To analyse FGF2 protein accumulation in podocytes further, we have used cultured wild-type MPC cells (Mundel et al., 1997b). The subcellular distribution of FGF2 was determined by comparing proliferating with growth-arrested, differentiated MPC cells (the latter cultured for 7 days under nonpermissive conditions, see Materials and Methods). In proliferating MPC cells (Fig. 1C), FGF2 immunolabelling is weak and diffuse (similar to immature kidney glomeruli) (Dono and Zeller, 1994). By contrast, clear FGF2 labelling is detected predominantly in the nucleus of post-mitotic, differentiated MPC cells (Fig. 1D). This upregulation was confirmed by western blot analysis of total cell extracts (Fig. 1E). In addition, the subcellular distribution of the three FGF2 protein isoforms was determined by cell fractionation (Fig. 1F). Low levels of all FGF2 isoforms are detected in nuclear and membrane fractions of proliferating MPC cells (Fig. 1F, left panel). In post-mitotic MPC cells, high levels of the 21.5 kD and 22 kD FGF2 protein isoforms are detected predominantly in the nuclear fraction (Fig. 1F, right panel, lane Nuc) (Dono et al., 1998; Dono and Zeller, 1994). By contrast, the 18.5 kD FGF2 isoform is present in all fractions, but enriched in membranes and vesicles (Fig. 1F, right panel, lane Mem). The accumulation in the membrane fraction is indicative of possible FGF2 secretion and paracrine functions.

To address the functional significance of FGF signalling during podocyte maturation, kidneys of mice carrying an Fgf2 loss-of-function mutation (Dono et al., 1998) were analysed. In a large population of adult mice housed under non-SPF conditions only a very small fraction of Fgf2 mutant mice (10 of 266) displayed symptoms of renal failure with glomerosclerosis and podocyte damage. Such defects were not observed in a similarly large group of wild-type mice (G.D., unpublished). The rare and stochastic nature of this phenotype in adult mice suggests that Fgf2 deficiency is at best a contributing factor to renal failure, which is normally compensated for in vivo. In an attempt to circumvent this potential functional compensation, we isolated and cultured wild-type and Fgf2 mutant podocytes for comparative molecular analysis. Conditionally immortalised MPC cells were isolated from apparently normal kidneys of Fgf2 mutant adult mice carrying the inducible and temperature-sensitive large T-antigen transgene (Jat et al., 1991). For all phenotypic analysis shown in Figs 2-5, normalised pools of wild-type and Fgf2 mutant MPC cell clones were used to exclude clone-specific variation (see Materials and Methods). Wild-type and Fgf2 mutant MPC cells have similar proliferation and mitotic arrest kinetics (data not shown) as expected from conditional immortalisation (Jat et al., 1991). However, analysis of the growth characteristics under permissive conditions showed that Fgf2 mutant MPC cells tend to grow more as aggregates compared to their wild-type counterparts (compare Fig. 2A with Fig. 2B). Alterations in cell morphology became even more apparent upon shifting MPC cells to nonpermissive culture conditions. Post-mitotic, Fgf2 mutant cells remain tightly associated (Fig. 2D) and fail to form cellular processes (compare inset in Fig. 2C with inset in Fig. 2D) in contrast to the ‘arborised’ cell morphology of differentiated wild-type cells (Fig. 2C).

One possible explanation for this apparent discrepancy between intact kidneys and cultured MPC cells could be the loss of functional compensation by other FGFs in culture. Therefore, the levels of other FGFs and FGF receptors that are expressed in embryonic and/or adult kidneys (Cancilla et al., 1999; Ford et al., 1997; Mason et al., 1994; Ohuchi et al., 2000; Peters et al., 1992) were determined semi-quantitatively by RT-PCR (Fig. 3). The relative transcript levels in wild-type (+/+) and Fgf2 mutant (−/−) MPC cells were normalised and compared to the ones of dissected wild-type and mutant kidney cortex tissue (enriched in glomeruli). This analysis confirmed loss of Fgf2 mRNA in mutant MPC cells and kidneys (Fig. 3). Analysis of Fgf2 mutant kidney cortex tissue revealed no differences to wild-type tissue for any of the Fgf ligands and receptors analysed (Fig. 3, right panels). By contrast, significant changes in transcript levels were detected by comparing Fgf2 mutant (−/−) to wild-type (+/+) MPC cells (Fig. 3, left panels). In particular, Fgf1 transcript levels are upregulated in mutant MPC cells in comparison to wild-type cells (Fig. 3, upper panels). By contrast, the levels of both Fgf7 and Fgf10 transcripts are downregulated to low or undetectable levels in FGF2 mutant MPC cells (Fig. 3, upper panels), whereas Fgf8 expression remains normal (data not shown).

Fibroblast growth factor signals are transduced by high-affinity tyrosine kinase FGF receptors (FGFR) (Martin, 1998) and there is evidence that the IIIc isoforms of FGFR1 and FGFR2 have highest affinity for FGF2 (Ornitz et al., 1996). Although expression of the Fgfr1 isoform IIIc is not affected in mutant MPC cells, expression of the Fgfr2 isoform IIIc is downregulated compared with wild-type MPC cells (Fig. 3, lower panels). By contrast, the Fgfr1 and Fgfr2 IIIb isoforms are both upregulated in mutant MPC cells (Fig. 3, lower panels). In summary, the results in Fig. 3 show that expression of at least three Fgf ligands, which are normally upregulated upon MPC cell differentiation, are low or undetectable in mutant MPC cells. FGF signal reception is also altered in mutant cells, owing to changes in receptor isoforms. In agreement with this general alteration of FGF signalling, transient re-expression of FGF2 in FGF2 mutant MPC cells by DNA microinjection did not rescue the mutant phenotype (data not shown). It seems thus appropriate to refer to the phenotypes observed in mutant MPC cells (Figs 2; Fig. 4; Fig. 5) as defects caused by disrupting FGF rather than only FGF2 signalling.

The overall architecture of the actin cytoskeleton is similar in proliferating wild-type and mutant MPC cells (data not shown); however, striking differences become apparent upon induction of post-mitotic differentiation (Fig. 4A,B). In differentiated wild-type MPC cells, an ordered array of actin stress fibers extends throughout the cytoplasm and terminates in focal contacts located predominantly at the cell periphery (yellow arrowheads, Fig. 4A). By contrast, the actin filaments of post-mitotic mutant MPC cells are very disorganised (Fig. 4B). Large bundles of actin filaments are arranged in a belt-like structure around the perinuclear region and stress fibers fail to extend to the cell periphery (indicated by white arrowheads, Fig. 4B; the cytoplasm extends beyond the panel’s right margin). Accordingly, most focal adhesion contacts form in proximity of the perinuclear region instead of the cell periphery (compare Fig. 4A with Fig. 4B). In contrast to the actin cytoskeleton, the microtubular architecture of mutant MPC is not significantly altered (data not shown).

Synaptopodin is a protein associated with actin filaments in differentiated podocytes and neurons (Mundel et al., 1997a). Both in vivo and in vitro, the expression of this differentiation marker is activated during actin-based foot process formation (Mundel et al., 1997a; Mundel et al., 1997b) (Fig. 4C). In accordance with the morphological alterations seen in mutant MPC cells (Fig. 2D) (Fig. 4B), synaptopodin expression is much lower in mutant MPC cells compared with wild-type cells after induction of differentiation (Fig. 4D,E). Interestingly, the podocyte lineage marker WT-1 (Wilm’s Tumour antigen) (Mundlos et al., 1993) is also markedly downregulated in proliferating and post-mitotic mutant MPC cells (Fig. 4E). Instead, podocalyxin and Pod-1, two additional podocyte lineage markers (Kerjaschki et al., 1984; Quaggin et al., 1999), remain expressed at levels similar to those seen in wild-type cells (Fig. 4E). Thus, although many properties of differentiated podocytes are disrupted, mutant MPC cells mostly maintain podocyte lineage characteristics.

The results shown in Figs 2, Fig. 4 reveal a block in postmitotic differentiation of mutant MPC cells. To identify the underlying molecular defects causing this phenotype, we analysed genes implicated in the EMT-like transition that podocytes undergo during onset of differentiation. The intermediate filament protein vimentin is a marker of mesenchymal cells (Schmid et al., 1979) and is expressed by mature podocytes (Fig. 5A) (Holthofer et al., 1984; Yaoita et al., 1999). In agreement, vimentin expression is upregulated in differentiating wild-type MPC cells (Fig. 5A,E), whereas it remains at low levels in mutant MPC cells (Fig. 5B,E). This defect is paralleled by a marked upregulation of the two epithelial markers cytokeratin and desmocollin type-2 (Dsc2) (Franke et al., 1979; Koch et al., 1992) in mutant MPC cells (Fig. 5E; and data not shown). Furthermore, the tight junction protein ZO-1 (Stevenson et al., 1986) relocalises from an apical position to the slit-diaphragm forming area of foot processes during podocyte differentiation (Schnabel et al., 1990). Differentiated wild-type MPC cells express predominantly one of the two ZO-1 isoforms (Fig. 5E), which localises to discrete areas of the cellular processes (Fig. 5C, arrow). Instead, mutant MPC cells express both ZO-1 isoforms (Fig. 5E), which are uniformly distributed at cell-cell interfaces (Fig. 5D). Again, this ZO-1 expression profile is reminiscent of epithelial cells (Balda and Anderson, 1993). The zinc finger transcription factor slug has been implicated in FGF-induced EMT switches (Savagner et al., 1997) and marks mesenchymal cell state (Nieto et al., 1994). As expected, slug expression is upregulated in differentiated wild-type MPC cells, whereas it remains undetectable in mutant MPC cells (Fig. 5E). Taken together, these results indicate that the molecular pathway(s) leading to induction of mesenchymal differentiation are blocked in MPC cells with disrupted FGF signalling.

The results presented here show that differentiation is blocked in post-mitotic mutant MPC cells isolated from Fgf2 mutant adult mice. As a consequence, these cells fail to re-organise their actin cytoskeleton and extend cellular processes. By contrast, symptoms of renal failure are very rare in Fgf2 mutant mice, suggesting functional compensation. Such striking differences in phenotypic manifestation between cells in intact organs and cells in culture have been previously observed. For example, telomerase- (Blasco et al., 1997) and vimentin- (Eckes et al., 1988) deficient cells isolated from apparently normal mice develop striking phenotypes in vitro. In the case of Fgf2 mutant podocytes, it is possible that other FGFs produced by podocytes, mesangial or endothelial cells compensate for the lack of FGF2 in vivo. Developing and adult kidneys express FGF7 and FGF10, and loss-of-function mutations in either of these FGFs causes kidney defects in mouse embryos (Ohuchi et al., 2000; Qiao et al., 1999). Indeed, the specific loss of FGF7 and FGF10 in mutant MPC cells together with their normal expression in Fgf2 mutant kidneys indicates that loss of these FGFs (in addition to FGF2) is responsible for the in vitro phenotypes observed. Thus, isolation and culture of podocytes from Fgf2 mutant kidneys resulted in removal of possible functional compensation by other FGFs. Upregulation of Fgf1 expression by Fgf2 mutant MPC cells may be indicative of compensatory cross-regulation, but is obviously not sufficient to compensate for the combined loss of FGF2, FGF7 and FGF10 in vitro. Therefore, it is appropriate to interpret the observed EMT-like cell differentiation defects as a consequence of disrupting/altering FGF signalling rather than loss of FGF2.

Despite the fact that terminally differentiated podocytes retain a basolateral asymmetry, the morphological and molecular changes that take place during onset of differentiation, both in vivo and in vitro, are similar to an EMT (Holthofer et al., 1984; Mundel et al., 1997b; Schnabel et al., 1990). In particular, differentiating podocytes loose epithelial characteristics such as desmosomal adhesion plaques and cytokeratins, whereas they acquire mesenchymal features such as vimentin expression (Schmid et al., 1979) (this study). The inability of mutant MPC cells to develop such mesenchymal characteristics indicates that FGF signalling is required to induce EMT-like changes. In support of this, epiblast cells require FGF signalling to undergo the EMT during gastrulation (Ciruna et al., 1997) and epithelial cells in culture acquire mesenchymal properties upon FGF overexpression (Migdal et al., 1995). A downstream target of FGF-mediated effects on EMT is the transcription factor slug (Buxton et al., 1997; Savagner et al., 1997), whose expression is disrupted in mutant MPC cells. In fact, overexpression of slug has been shown to trigger an EMT and results in downregulation of desmosomal cadherins (Savagner et al., 1997). In agreement, the disruption of FGF signalling and failure to activate slug in mutant MPC cells is accompanied by a very significant upregulation of the desmosomal cadherin Dsc2. These studies show that FGF signalling acts upstream of slug activation and Dsc2 repression during initiation of an EMT or establishment of mesenchymal characteristics.

During terminal differentiation, podocytes extend an ordered array of actin-based foot processes, which entwine glomerular capillaries and form an essential component of the filtration unit (Saxén, 1987). The inability of mutant MPC cells to initiate these EMT-like changes leads to a block in development of actin-base cellular processes. Interestingly, these defects are reminiscent of alterations occurring during onset of glomerular disease in vivo. Damaged podocytes retract their foot processes, detach from the glomerular basement membrane and frequently revert to a more epithelial morphology (Bariety et al., 1998; Whiteside et al., 1993). These changes in podocyte morphology are often accompanied by a significant expansion of their cytoplasm (Autio-Harmainen et al., 1981), a phenomenon seen often in mutant MPC cells. Together with the present study, these results indicate that FGF signalling might regulate differentiation or maintenance the intricate array of podocyte foot processes in interaction with neighbouring podocytes and the glomerular basement membrane (Saxén, 1987).

Finally, podocytes seem to share some similarity with neurons as both cell types extend complex actin-based processes during terminal differentiation (Reeves et al., 1978; Zigmond et al., 1999). Axonal path finding by neurons is controlled by (de)polymerisation of actin-based filopodia emerging from the growth cone located at the distal tip of the growing axon (Zigmond et al., 1999). Disrupting FGF signalling in cultured cerebellar neurons prevents axonal outgrowth (Saffell et al., 1997), whereas FGF stimulation of cultured retinal neurons promotes axonal outgrowth and target recognition (McFarlane et al., 1995). FGF signalling has also been shown to regulate actin filament reorganisation of endothelial cells during wound healing (Wang and Gotlieb, 1999), a process which is delayed in Fgf2 deficient mice (Ortega et al., 1998). Furthermore, it has recently been shown that embryonic cells extend cytonemes, which are long, actin-based cellular processes growing towards signalling centers (Ramirez-Weber and Kornberg, 1999). Interestingly, FGF signalling stimulates both outgrowth and orientation of cytonemes similar to FGF functions during axonal path finding (see above). In summary, FGF signalling seems to regulate aspects of growth and/or maintenance of actin-based processes during differentiation of various cell-types.

We thank W. Kriz, P. Mundel, J. Reiser, K. Amann and A. Zuniga for reagents, advice on kidney morphology and establishing MPC cell cultures. We are grateful to T. Bouwmeester, S. Cohen and M. Way for advice, support and critical input into this study. We are indebted to J. Boonstra, S. Chabanis, C. Niehrs, M. Way and A. Zuniga for critical comments on the manuscript. This study was supported by EMBL and in parts by a grant from the Thyssen foundation.