SOX9 transcription factor is involved in chondrocyte differentiation and male sex determination. Heterozygous defects in the human SOX9 gene cause campomelic dysplasia. The mechanisms behind SOX9 function are not understood despite the description of different target genes. This study therefore sets out to identify SOX9‐associated proteins to unravel how SOX9 interacts with the cellular transcription machinery. We report the ability of SOX9 to interact with TRAP230, a component of the thyroid hormone receptor‐associated protein (TRAP) complex. Both in vitro and in vivo assays have confirmed that the detected interaction is specific and occurs endogenously in cells. Using co‐transfection experiments, we have also shown that the TRAP230 interacting domain can act in a dominant‐negative manner regarding SOX9 activity. Our results add SOX9 to the list of activators that communicate with the general transcription machinery through the TRAP complex and suggest a basis for the collaboration of SOX9 with different coactivators that could contact the same coactivator/integrator complex.
Received February 1, 2002; Revised and Accepted May 28, 2002
Campomelic dysplasia (CD) is a severe human dwarfism syndrome characterized by malformations of many cartilage‐derived structures (1,2). This includes bowing and angulation of the tibiae and femora, bilateral talipes equinovarus, a small thoracic cage, a reduced number of ribs, facial abnormalities, and tracheomalacia leading to respiratory distress and neonatal death. Furthermore, ∼70% of campomelic patients show male to female sex reversal. Positional cloning and subsequent mutational analysis provided molecular evidence that the SOX9 gene is involved in both disorders (3,4). Mutation screens have identified hererozygous mutations in SOX9 throughout the open reading frame, some of them truncating the C‐terminal part of the molecule (5,6). The nature of these mutations indicates that this autosomal dominant disease is the result of haploinsufficiency. SOX9 is a transcription factor which contains a high mobility group (HMG)‐type DNA‐binding and bending domain and a transcription activation domain (TA‐domain), which is located at the C‐terminus of the protein (6). Evidence has been provided that chondrocyte‐specific markers such as Col2a1, the gene for type II collagen, and aggrecan, a gene encoding a sulfate proteoglycan, as well as Sertoli‐cell‐specific anti‐Müllerian hormone (also known as Müllerian inhibiting substance) gene, are direct target genes for SOX9 (7–10). In accordance with the skeletal defects seen in CD, SOX9 expression during embryogenesis occurs predominantly at sites of chondrogenesis where SOX9 accomplishes its role in both the initiation and the maintenance of chondrocyte differentiation (11). SOX9 is also expressed in genital ridges, where it appears to be upregulated in males at precisely the time when sex determination occurs (12,13). Finally, SOX9 is also expressed in heart, kidney, nervous system and otic vesicles of mouse embryos, locations where SOX9 functions still need to be studied further (12). At the cellular level, the SOX9 protein localization appears to be tightly regulated during sex determination: in the female genital ridge it remains cytoplasmic, whereas in males the protein is translocated to the nucleus (13,14). Thus, if SOX9 has an essential role in determining different cell fates during embryogenesis, nothing is known about the molecular mechanisms leading to SOX9‐dependent transcriptional activation. While SOX9 has been shown to cooperate with other SOX factors during chondrogenesis (15), it interacts with the orphan nuclear receptor SF‐1 in the activation of the AMH gene during testis differentiation (9,10). In order to understand the molecular mechanism of SOX9‐dependent transcriptional activation, a yeast two‐hybrid human embryonic cDNA library was prepared and screened using the SOX9 TA‐domain as bait. TRAP230, a component of the multi‐subunit human thyroid hormone receptor‐associated protein (TRAP)/Mediator complex (16,17), was identified as a SOX9 TA‐domain interacting protein. This interaction was confirmed by in vitro glutathione S‐transferase (GST) pull‐down experiments, co‐immunoprecipitation and immunofluorescence analysis. Finally, a transactivation inhibition assay indicated that the transcriptional activation capacity of the SOX9 TA‐domain is mediated via its interaction with TRAP230. These results strongly support a physiological role for the mammalian mediator complex in conveying the SOX9 activation signal to the transcriptional apparatus.
MATERIALS AND METHODS
A directional human embryonic cDNA library was constructed in the EcoRI–XhoI site of the phagemid vector pAD‐GAL4, which includes the GAL4 activation domain, by using a Stratagene HybriZap two‐hybrid cloning kit. mRNAs were isolated from three complete 6–8‐week‐old human embryos. Human embryonic tissues were obtained from therapeutic termination (week of gestation) as part of a program approved by the French National Ethics Committee. A total of 106 independent clones were obtained. An insert analysis after PCR amplification of randomly picked clones from the library revealed an average size of insert of ∼1.2 kb (range: 125– 2300 bp). An aliquot of 50 µg of the resulting cDNA library was then transformed by 12 independent transformations into MATα Y187 yeast strain, plated on selective medium (leucine selection) and then pooled. A sample of 108 cells/500 µl were frozen in 15% glycerol rich medium. The resulting titer of the library was estimated to be ∼4.5 × 107 cells per vial after thawing. The human SOX9 bait plasmid was constructed by cloning a PCR‐amplified SOX9 TA‐domain (amino acids 408–509) in frame with a downstream GAL4 DNA binding domain in plasmid pGBT9, and then transformed into MATa CG1945 yeast cells. Yeast two‐hybrid screening was conducted according to the mating procedure described by Fromont‐Racine et al. (18) with minor modifications. For each screen, thawed cells from one vial of the yeast library were mixed with CG1945 cells transformed with the SOX9 TA‐domain bait plasmid. This mating was plated on complete medium for 5 h, and then transferred onto leucine– tryptophan–histidine‐defective (DO‐W‐L‐H) plates. After 3–5 days, positive clones were numbered and an overlay β‐galactosidase test conducted. Blue positive clones were then streaked on DO‐W‐L‐H plates. After PCR amplification using universal primers, sequencing of the positive clones was performed by the Sanger dideoxy chain termination method. The alignment program BLAST (GenBank) was used to determine protein similarity indices.
Protein interaction assays
Full‐length SOX9 sequence and TRAP230‐derived sequences [proline‐, glutamine‐ and leucine‐rich (PQL) and PQL– glutamine‐rich (OPA)] were fused to the GST sequence in pGEX‐4T‐1 vector and overexpressed and purified as described previously (9). For in vitro labeling, 35S‐labeled proteins were synthesized in vitro (TNT system, Promega) from a pcDNA3 (Invitrogen) vector. Glutathione–Sepharose‐immobilized proteins were then incubated with the corresponding labeled protein in 300 µl of TBST buffer (10 mM Tris–HCl pH 7.5, 100 mM KCl, 0.1% NP‐40, 0.5% bovine serum albumin) at 4°C for 30 min. After washing, bound proteins were eluted by addition of 5× Laemmli buffer, boiled, resolved by SDS–PAGE and visualized by autoradiography.
SOX9–TRAP complex interaction analysis
The thyroid hormone receptor (TR)–TRAP complex was isolated from a FLAG‐TR (f:TR) expressing cell line via an immobilized anti‐FLAG antibody (17). An aliquot of 5 µl of purified TR–TRAP complex (17) was incubated with 1 µg each of GST alone or GST‐SOX9 in BC buffer with 0.1% NP‐40, 1 mM β‐mercaptoethanol and either 100 or 150 mM KCl. After washing with the binding buffers, bound proteins were eluted with 0.1% Sarkosyl, resolved by 5–15% SDS–PAGE and visualized by silver staining (17).
The SOX9 TA‐domain (amino acids 408–509) was cloned into pBIND vector (Checkmate two‐hybrid System; Promega), thus creating a fusion construct encoding both the yeast GAL4 DNA‐binding domain and the SOX9 TA‐domain. A GAL4‐VP16 expression vector containing the VP16 TA‐domain (amino acids 411–490) fused to the GAL4 DNA‐binding domain was kindly provided by Dr M. Benkirane (CNRS, Montpellier, France). The partial TRAP230 cDNA M40, obtained from the yeast two‐hybrid screen, was cloned as an N‐terminal fusion with a sequence encoding the hemaglutinin (HA) tag (amino acids MAYPYDVPDYAEF) in a modified pcDNA3 vector. Transfection experiments were performed using Fugene‐6 reagent (Roche Diagnostics). COS‐7 cells were cotransfected with 10 ng of the pBIND SOX9‐TA expression vector, which includes the Renilla luciferase expression cassette, 0 (empty pcDNA HA vector), 10 or 20 ng of pcDNA HA‐M40 plasmid and 500 ng of GAL4 firefly luciferase reporter plasmid pG5luc (Promega) on 24‐well plates (4 × 105 cells per well). In each case, empty pcDNA HA was used to adjust plasmid quantity. Cells were harvested after a 48 h incubation. Firefly and Renilla luciferase activities were measured using the dual luciferase reporter assay system (Promega).
Embryos were staged according to the recognized Carnegie stages. Sections were probed by an overnight incubation at room temperature with anti‐SOX9 (14) or anti‐TRAP230 (17) antibody. After washing in phosphate‐buffered saline, SOX9 primary antibodies were visualized with Fluorolink Cy2‐conjugated anti‐rat‐labeled goat antibodies (dilution, 1:200) and TRAP230 primary antibodies with Texas red‐conjugated streptavidin anti‐rabbit‐labeled donkey antibodies (dilution, 1:100). In each case, sections were incubated for 90 min under the same conditions described for the primary antibodies, and then mounted in FluorSave reagent (Calbiochem). Cell nuclei were visualized with Hoechst 33286. CV1 cells were transfected with the pcDNA‐HA‐M40 plasmid construct with or without pcDNA‐SOX9 using Fugene‐6 transfection reagent (Roche Diagnostics) according to the supplier’s instructions. After a 48 h incubation, cells were fixed and processed for immunofluorescence analysis with the appropriate antibody, anti‐SOX9 (14) or mouse monoclonal anti‐HA Tag (kindly provided by Dr C. Gauthier‐Rouvière, CNRS).
NT2‐D1 cell nuclei were prepared as described previously (19), sonicated and incubated with either SOX9 antibodies or the corresponding preimmune serum in the appropriate buffer. The immune complexes were adsorbed to protein A– Sepharose 4, washed and resolved by SDS–PAGE before immunoblotting with the anti‐TRAP230 antibody (17).
TRAP230 interacts with SOX9 in a yeast two‐hybrid assay
To identify SOX9 TA‐domain‐interacting proteins, a yeast two‐hybrid human embryonic cDNA library was synthesized and screened using the TA‐domain as bait. A total of 2157 His+ colonies were obtained from an estimated 106 transformants, and this number was reduced to 68 after a β‐GAL overlay test. A PCR probe was then derived from the pooled clones and used to group the clones in three subgroups according to their homology and estimated size upon agarose gel electrophoresis. After sequencing, two independent clones were shown to correspond to the C‐terminal part (amino acids 1650–2212) of the TRAP230 component (DDBJ/EMBL/GenBank accession no. AF117755) of the TRAP coactivator complex (17). This truncated TRAP230 cDNA was termed M40 and was shown to correspond precisely to the C‐terminal PQL and OPA domains (Fig. 1). Other TRAP230 domains include the N‐terminal leucine‐rich (L) and leucine‐ and serine‐rich (LS) domains. To further delineate which portions of the TRAP230/M40‐derived fragment were critical for this interaction, binding assays with M40 deletions were performed using the yeast two‐hybrid system. After subcloning M40‐derived sequences separately in the appropriate vector, each construct was tested against the SOX9 TA‐domain. The data in Figure 1 show that the complete PQL domain is both necessary and sufficient for the interaction with the SOX9 TA‐domain, since a deletion of the nine C‐terminal amino acids from this motif (amino acids 2078–2086, construct ΔM40.3) abolishes the interaction and since the C‐terminal OPA‐domain neither binds on its own nor affects the PQL interaction with the SOX9 TA‐domain (ΔM40.9, amino acids 2086–2212). Interestingly, in the yeast two‐hybrid assay, the M40 protein is also able to interact with the TA‐domains of two other SOX proteins, SOX8 (amino acids 220–301) and SOX10 (amino acids 365–466), of the subgroup E (20,21) (data not shown). This result suggests that the TA‐domains from SOX proteins of this subgroup could act via the same molecular process. The lack of homology between SOX8 transactivation (and TRAP230 interacting) domain with the TA‐domains of SOX9 and SOX10 could suggest that they all three adopt a similar 3D structure despite their difference in primary structure. As control, no such interaction was found when using three other full‐length SOX factors that belong to three others subgroups [SRY (group A), SOX3 (group B) and SOX11 (group C)] (data not shown).
SOX9‐TRAP230 interaction in vitro
In order to test this interaction with the full‐length proteins, and because the full‐length TRAP230 is toxic for yeast, GST pull‐down assays were performed. Assays were carried out using a bacterially expressed and purified GST‐SOX9 fusion protein and in vitro translated 35S‐labeled TRAP230. As shown in Figure 2A, TRAP230 binds to the GST‐SOX9 fusion protein, but not to GST alone. This experiment confirms the yeast two‐hybrid result and also shows that the full‐length SOX9 and TRAP230 proteins can interact. This interaction was confirmed by a reverse experiment using 35S‐labeled SOX9 and GST‐TRAP230 PQL+OPA or PQL domain fusion proteins (Fig. 2B). However, when 35S‐labeled SOX9 was deleted for the TA‐domain (ΔCT), the interaction with either GST‐TRAP230 PQL+OPA or GST‐TRAP230 PQL domains was abolished. This confirms that the SOX9 TA‐domain is required for the interaction with the PQL motif of TRAP230.
SOX9‐TRAP230 interaction in vivo
These results prompted us to assess a possible SOX9– TRAP230 interaction in vivo. To determine whether endogenous SOX9 and TRAP230 were complexed in cells, SOX9 was immunoprecipitated from nuclear lysate of a human cell line (NT2‐D1) that expresses the SOX9 protein (9). Immune complexes were subsequently resolved by SDS–PAGE followed by western blotting with antibodies against TRAP230 (17), as shown in Figure 3. This analysis revealed a unique band with the expected size for TRAP230 (∼230 kDa), thus strongly indicating an in vivo interaction between SOX9 and TRAP230 (Fig. 3, lane 1). A control analysis performed with preimmune serum showed no TRAP230 precipitation, confirming the specificity of the interaction (Fig. 3, lane 2). Another control analysis (Fig. 3, lane 3) shows the specificity of the TRAP230 antiserum for TRAP230 in the total NT2‐D1 extract. Another way to demonstrate an intracellular interaction of two nuclear proteins is to delete the nuclear localization signal in one of them, and to show by transfection and immunofluorescence analyses that the intact nuclear protein can drive the mutated protein into the nucleus. For this purpose we first verified that the M40 clone, when fused to an N‐terminal HA‐tag sequence and transfected into CV‐1 cells, effectively encoded a cytoplasmic protein (Fig. 4B); this contrasts with the full‐length TRAP230, which is nuclear even when overexpressed (data not shown). This observation suggests that nuclear localization signals from TRAP230 are missing in the HA‐M40 clone. When cells were co‐transfected with HA‐M40 and SOX9 expression plasmids in the same type of experiment, the HA‐M40 protein was localized to the nucleus (Fig. 4E) and showed a staining pattern very similar to that of SOX9 (Fig. 4D). This result demonstrates that the SOX9 protein can drive the portion of TRAP230 containing only the PQL and OPA domain into the nucleus, confirming that these two proteins are interacting in vivo. As control, a construct encoding SOX9 deleted for its TA‐domain was unable to drive the HA‐M40 protein into the nucleus (data not shown).
SOX9 interacts with the TRAP complex
While these experiments suggest an in vivo interaction between SOX9 and TRAP230, they do not prove that SOX9 can interact with TRAP230 when it is part of a pre‐formed TRAP complex. To show this, the GST‐SOX9 fusion protein was incubated at either 100 or 150 mM KCl with a TR–TRAP complex purified from HeLa cells (17), a cell line that does not express SOX9 (data not shown). After extensive washing with the binding buffer, bound proteins were eluted with Sarkosyl and analyzed by SDS–PAGE and silver staining (Fig. 5). The TR–TRAP complex bound specifically to the GST‐SOX9 fusion protein relative to GST alone at 100 mM KCl, the profile of bound polypeptides being essentially the same as that of the complex that was affinity purified through f:TR (17). This interaction is relatively weak, since the complex was not bound at 150 mM KCl, suggesting that SOX9 alone may not be sufficient to efficiently stabilize the TRAP complex on the regulatory region of a target gene. Additional transcription factors that interact either with SOX9, such as steroidogenic factor 1 in the control of the anti‐Müllerian hormone (AMH) gene (9), or with other TRAP proteins could reinforce the stabilization of TRAP complex on the promoter.
Also of note from this analysis is that TR remains bound to the TRAP complex when the latter is bound to SOX9 (Fig. 5, lane 7). This indicates that two activators (TR and SOX9) can bind simultaneously to the same TRAP complex, consistent with previous observations for TR and Gal4‐VP16 (17).
SOX9 co‐localizes with TRAP230 in human embryonic chondrocytes
To show whether SOX9 and TRAP230 proteins are co‐localized in embryonic cells, their subcellular localizations were next analyzed by immunofluorescence experiments using confocal laser scanning microscopy. Staining with both SOX9 and TRAP230 antisera revealed a nuclear localization for the two proteins in chondrocytes from a 6‐week‐old human embryo (Fig. 6A and B) and a good co‐localization of the two proteins was observed after summation of the two staining patterns (Fig. 6C).
Dominant‐negative TRAP230 (M40 subfragment) blocks SOX9 TA‐domain activity
The interaction between TRAP230 and the SOX9 TA‐domain suggests that TRAP230 could mediate the transcriptional activation signal from enhancer‐bound SOX9 to RNA polymerase II on a target gene. One way to support this is to determine whether overexpression of TRAP230 leads to a positive effect on the activity of SOX9 in a cotransfection experiment. However, in our hands ectopic TRAP230 had no effect on SOX9 function, probably because TRAP230 functions as a stoichiometric component of a TRAP complex whose other subunits are tightly regulated. We therefore tested whether HA‐M40 could serve as a dominant‐negative mutant that blocks the activity of the SOX9 TA‐domain. To look only at the activity of this domain and to avoid involvement of any other transcription factor interacting with SOX9, a vector expressing a GAL4 DNA‐binding domain‐SOX9 TA‐domain fusion protein was cotransfected with a GAL4‐responsive luciferase reporter (pG5Luc) and increasing concentrations of the HA‐M40 expression vector. GAL4‐SOX9TA displayed a strong transactivation capacity on pG5Luc and coexpression of HA‐M40 showed an induced dose‐dependent inhibition of this activity (Fig. 7). Maximum inhibition of ∼67% was observed at 20 ng of HA‐M40 expression vector, compared to 1% inhibition with the empty expression vector. The failure to demonstrate complete inhibition at even higher levels of HA‐M40 (data not shown) probably reflects the presence of endogenous TRAP230 protein. At the same time, western blot analysis has shown that the inhibition was not due to a decrease of the GAL4‐SOX9 TA protein level (data not shown). As a control for the specificity of this inhibition, the same concentrations of HA‐M40 were tested on VP16 TA‐domain activity. Indeed, the VP16 TA‐domain is known to target the TRAP complex through the TRAP80 component (17). As expected, HA‐M40 did not inhibit VP16 TA‐domain activity (Fig. 7). To conclude, our results suggest that M40 effectively competes with endogenous TRAP230 in its ability to relay SOX9 signals to the general transcription machinery, and that transactivation by the SOX9 TA‐domain is mediated via its interaction with the PQL motif of the TRAP230 protein.
The TRAP coactivator complex was first biochemically identified through its intracellular association and affinity copurification with epitope‐tagged TR (22). This large multimeric complex of proteins that range in size from 20 to 240 kDa was shown to enhance in vitro transcriptional activation by TR in a chromatin‐free assay (22–24). The TRAP complex is in fact similar if not identical to other subsequently identified mammalian complexes such as DRIP, ARC, NAT, SMCC and Mediator (17). The TRAP complex, like these other complexes, has been found to bind to diverse transcriptional activation domains and to be required for their function. These data suggest that the TRAP complex could form a direct link between specific DNA‐binding transcription activators and RNA polymerase II, thus forming a novel pathway for communication from gene‐specific activators to the general transcriptional machinery. Different components of the TRAP complex were previously shown to be the targets of diverse activators (17). Thus, TRAP220 was shown to directly interact with TR, and later with several other nuclear receptors, in a ligand‐dependent fashion (24). In contrast, TRAP80 was identified as a target for p53 and VP16 (17) and SUR2 as a target for E1A and Ets proteins (25). We have now extended this list to the SRY‐related protein SOX9 (and probably to the related group E SOX factors) by demonstrating its ability to directly bind TRAP230, as well as the complete TRAP complex. This interaction has been confirmed for SOX9 both in vitro and in vivo. Interestingly, it may now be possible to classify the transcription factor activation domains according to the TRAP protein that they contact. For that purpose, a yeast two‐hybrid screen with the TRAP230 PQL domain could provide the opportunity to start such a classification.
The demonstrated nuclear colocalization of SOX9 and TRAP230 in human embryonic chondrocytes is consistent with a function for this interaction in the transcription events that trigger chondrocyte differentiation. For other SOX9‐expressing tissues, such as embryonic male gonads, colocalization of both proteins was more difficult to visualize, probably because of a weak SOX9 expression level and difficult access to the SOX9 protein by the antibody. Further experiments using other antibodies will be required to confirm this point.
To show the functional significance of the TRAP230–SOX9 interaction, and given the likelihood that TRAP230 may function optimally only in conjunction with other associated TRAP proteins (22), we tested the ability of the SOX9‐interacting C‐terminal domain of TRAP230 (encoded by the M40 clone) to act as a dominant negative. The demonstrated ability of this domain to inhibit SOX9‐mediated activation indicates that an interaction with TRAP230 is required for the biological function of the SOX9 TA‐domain. Nevertheless, our results do not exclude the possibility that other regions of SOX9 contact the TRAP complex. Indeed, the region adjacent to the SOX9 TA‐domain, also called the ‘PQA motif’ (4), seems to contribute to the function of the TA‐domain (26). Nevertheless, this region is poorly conserved in the SOX9 protein between different species, and because of its PQA‐rich sequence may only work as a flexible hinge for the TA‐domain.
Our findings confirm the emerging view that distinct activators not only interact with the complete TRAP complex, but that they may do so through different TRAP subunits and in a concerted fashion (16). The protein composition of the TRAP complex also is somewhat variable and may contribute, with different transcription factors, to cell specificity. Thus, TRAP230, like TRAP220 and TRAP240, shows a relatively looser association with the TRAP core complex and may even be missing in some TRAP complexes in the cell (16). One prediction from these observations is that specific activators may interact with, and stabilize interactions of, free or loosely bound TRAPs with the TRAP complex. Another prediction is that different DNA‐bound transcriptional activators may function synergistically on a common target gene through simultaneous interactions with a single TRAP complex, thus facilitating its promoter recruitment and functional interactions with components of the preinitiation complex. Cooperative interactions of SOX9 with SOX5 or SOX6 have been implicated in type II collagen gene activation during chondrogenesis (15), whereas activation of AMH expression during sex determination involves SOX9 interactions with the orphan receptor SF‐1 (9). These observations lead to the hypothesis that, in each case, the additional SOX9 coactivators may interact with the TRAP complex through distinct components. This hypothesis is now under investigation. The possibility that other components of the TRAP coactivator complex could, for instance, contribute directly to sex determination by interacting with SOX9 or with other members of the combinatorial transcriptional complex involving SOX9 has recently been reinforced. After a large‐scale screen for genes involved in mouse gonad development, TRAP100 expression was observed at the time of the sex determining process specifically in male genital ridges and a few days later in testis cords (27). This sex‐dimorphic expression could suggest that the male differentiation program requires specific TRAP proteins because of the involvement of a different set of transcription factors.
SOX9 mutational analysis in patients with CD has revealed mutations throughout the gene, including missense mutations in the HMG domain and C‐terminal truncations (6). In the latter case, these mutations often create termination mutations that would abolish interactions with TRAP230, which in turn could lead to failure of target gene activation. The relatively broad expression patterns of both SOX9 (28) and TRAP230 (17,29) suggest that both proteins interact with each other in physiological situations other than chondrogenesis, and may imply a more global role for this interaction, which needs to be studied further.
A recent paper has identified sop‐1 (30), the TRAP230 homolog in Caenorhabditis elegans, as a modulator of the bar‐1/β‐catenin‐stimulated pathway, a component of the Wnt signal transduction system. This action could reinforce the capacity of different SOX family members to interfere with the ability of TCF/LEF factors to bind β‐catenin and to modulate Wnt signaling pathways (31,32). Experiments are in progress to test this hypothesis.
In summary, our identification of TRAP230 as a transcriptional coactivator for SOX9 provides further support for a general role for the TRAP complex in the regulation of transcription by connecting specific transcription factors and the general transcription machinery.
We thank Ms Brigitte Moniot for skillful technical assistance, and Mr Patrick Atger for high quality photographic processing. This work was supported by the European Economic Community through the fifth framework Program, no. GLG2‐CT‐1999‐00741, ARC grant no. 5210 to P.B., a grant from the Deutsche Forschungsgemeinschaft to G.S. (Sche 194/11‐3) and a grant from the NIH to R.G.R.