Invasin allows efficient entry into mammalian cells by Yersinia pseudotuberculosis. It has been shown that the C‐terminal 192 amino acids of invasin are essential for binding of β1 integrin receptors and subsequent uptake. By analyzing the internalization of latex beads coated with invasin derivatives, an additional domain of invasin was shown to be required for efficient bacterial internalization. A monomeric derivative encompassing the C‐terminal 197 amino acids was inefficient at promoting entry of latex beads, whereas dimerization of this derivative by antibody significantly increased uptake. By using the DNA‐binding domain of λ repressor as a reporter for invasin self‐interaction, we have demonstrated that a region of the invasin protein located N‐terminal to the cell adhesion domain of invasin is able to self‐associate. Chemical cross‐linking studies of purified and surface‐exposed invasin proteins, and the dominant‐interfering effect of a non‐functional invasin derivative are consistent with the presence of a self‐association domain that is located within the region of invasin that enhances bacterial uptake. We conclude that interaction of homomultimeric invasin with multiple integrins establishes tight adherence and receptor clustering, thus providing a signal for internalization.
Adhesion and subsequent internalization within mammalian cells is a common strategy used by many pathogenic bacteria to establish a successful infectious niche (Finlay and Cossart, 1997). The Gram‐negative bacterium Yersinia pseudotuberculosis, which causes systemic disease after initial gastrointestinal colonalization, translocates across the ileum into deeper host tissue. In animal infection models, the organism is internalized by M‐cells which are intercalated into the epithelium overlaying ileal lymphoid follicles called Peyer's patches (Gruetzkau et al., 1990; Marra and Isberg, 1997). Efficient translocation into the Peyer's patch requires the bacterial protein invasin (Pepe et al., 1995; Marra and Isberg, 1997). Cellular internalization of Y.pseudotuberculosis can be reproduced in vitro using a number of normally non‐phagocytic cultured cell lines and has been used to identify critical components involved in target cell adherence and penetration (Devenish and Schiemann, 1981; Isberg, 1991). The most efficient factor that promotes internalization in culture is invasin. This 986 amino acid (108 kDa) bacterial outer membrane protein is encoded by the chromosomal inv gene which, when expressed in normally non‐adherent laboratory Escherichia coli strains, confers the ability to penetrate mammalian cells (Isberg and Falkow, 1985; Isberg et al., 1987). Homologous invasin proteins have been found in other pathogenic Yersinia species (Miller and Falkow, 1988; Simonet et al., 1996) and are highly similar to a family of proteins called intimins, involved in attachment of a variety of related Gram‐negative enteric pathogens to host cells (Jerse et al., 1990; Schauer and Falkow, 1993; Frankel et al., 1994).
Invasin promotes entry into eukaryotic cells by binding to at least five different members of the β1 integrin receptor superfamily (Isberg and Leong, 1990; Krukonis and Isberg, 1997). Integrins are αβ heterodimeric proteins found on the surface of most mammalian cells and are involved in a wide variety of adhesive functions, such as cell–cell interaction, cell migration, differentiation and adhesion. Members of this receptor family are able to bind extracellular matrix proteins (ECMs) as well as cytoskeletal components, thus providing a sophisticated communication system between the extracellular environment and the intracellular cytoskeleton (Hynes, 1992; Clark and Brugge, 1995).
Analysis of truncated derivatives has shown that a region corresponding to the extreme C‐terminal 192 amino acids of invasin is sufficient to promote integrin‐dependent cell adhesion and is necessary for both bacterial attachment and internalization by target cells (Figure 1A; Leong et al., 1990, 1995; Rankin et al., 1992). Invasin derivatives containing C‐terminal fragments shorter than 192 amino acids are unable to promote binding and uptake when immobilized by antibody on the surface of Staphylococcus aureus (Rankin et al., 1992; Saltman et al., 1996). One striking characteristic of the invasin C‐terminus is the presence of an intramolecular disulfide bond between residues 907 and 982 that appears to be required for integrin recognition (Leong and Isberg, 1993; Leong et al., 1993). Analysis of entry‐deficient mutants indicates that the single most critical residue for bacterial penetration is an aspartate at position 911 (Leong et al., 1995; Saltman et al., 1996). Substitutions at this residue render bacteria unable to enter into mammalian cells (Leong et al., 1995). D911 seems to have a cell adhesion function similar to that of the aspartate residue in the RGD‐containing motif of natural integrin ligands (Ruoslahti and Pierschbacher, 1987). As is true of fibronectin, a second region of the protein ∼100 amino acids N‐terminal to this critical site also has residues that contribute to receptor binding (Obara et al., 1988; Saltman et al., 1996).
Although the interaction of invasin with its receptors has been characterized in detail, less is known regarding the mechanism by which invasin triggers integrin‐mediated bacterial uptake. Efficient entry requires tight binding of the bacterium to its host cell, which results in the extension of tight‐fitting pseudopods around individual bacteria, each of which is internalized into a membrane‐bound phagosome (Isberg, 1989; Tran Van Nhieu et al., 1996). This zipper‐like process morphologically resembles conventional phagocytosis of complement‐coated particles by professional activated phagocytes (Swanson and Baer, 1995). Invasin‐mediated entry is somewhat different from natural ligand binding, as mere attachment of a substrate to β1 integrins is usually not sufficient to trigger the signal transduction pathway necessary for uptake. For example, bacteria coated with fibronectin adhere to host cells, yet they are internalized inefficiently, possibly because fibronectin binds with a lower affinity than does invasin to the integrin receptor (Van de Water et al., 1983; Hook et al., 1989; Tran Van Nhieu and Isberg, 1993a,b; Yang and Isberg, 1993). Thus, it has been assumed that the sole unique feature of invasin that ensures its ability to promote bacterial uptake is its high affinity binding of receptor.
In this study, we investigated invasin‐mediated uptake of a variety of invasin derivatives and report the identification and analysis of a region of the protein that enhances bacterial uptake. We present several lines of evidence that a domain within this region can promote self‐association. Inability to self‐associate is strongly correlated with an inability to promote efficient uptake of particles that specifically adhere to integrin receptors.
Identification of an ‘uptake enhancer’ within invasin
To investigate invasin‐dependent entry in the absence of other bacterial factors, uptake of invasin‐coated latex beads was analyzed. Serial dilutions of two purified derivatives carrying the C‐terminal 497 and 197 amino acids of invasin (Inv497 and Inv197, Figure 1A; Materials and methods) were coated onto 1.1 μm inert latex beads, assayed for the number of molecules bound per bead and used to challenge HEp‐2 cells (Materials and methods). After a 1 h challenge of beads coated with a large excess of protein (>100 000 molecules per bead), internalized and extracellular beads were identified and quantitated by immunofluorescence microscopy. Control beads coated with bovine serum albumin (BSA) or maltose‐binding protein (MBP) rarely were found to be associated with HEp‐2 cells under the conditions used for the assay (<0.1%). More than 90% of Inv197‐ and Inv497‐coated beads observed were associated with HEp‐2cells, and a large number of bound beads were found to be internalized (Figure 1B–E). Control beads coated by fibronectin or polylysine were bound efficiently (80 and 60% of input, respectively) but, in marked contrast, only a few of these beads were found to be internalized (<1%, data not shown).
The uptake efficiency of the beads coated with Inv497 protein was significantly higher than that of Inv197 (Figure 1B–E) and was comparable to that seen with Y.pseudotuberculosis (24%; Marra and Isberg, 1997). Both invasin derivatives stimulated binding and uptake of beads in a concentration‐dependent manner; however, at all coating concentrations, Inv497‐coated beads were somewhat more adherent (Figure 2A) and internalized far more efficiently than beads coated with the Inv197 derivative (Figure 2B). The difference in uptake efficiency of cell‐bound Inv497‐ and Inv197‐coated beads was especially pronounced at low coating concentrations. For instance, ∼2000 Inv497 molecules were sufficient to mediate uptake of 20% of bound beads, whereas 10 times more Inv197 molecules were necessary to obtain the same efficiency of uptake. Moreover, particles coated with <2000 molecules of Inv197 were unable to promote cellular penetration. The effects observed were highly concentration dependent, as increasing the number of Inv197 molecules from 4×103/bead to 1×105/bead had little effect on the number of associated beads, but increased the internalization efficiency of bound beads 10‐fold (Figure 2). These results indicated the presence of a region that enhances uptake in the Inv497 protein and is absent from the Inv197 derivative.
The Inv497 derivative behaves in solution as if it has a large radius
Based on the concentration effects, we predicted that the low efficiency uptake of adherent beads coated by Inv197 may be due to an inability of the protein to form higher order complexes of invasin subunits in the bacterial outer membrane. A series of experiments was performed to determine if the invasin protein is able to interact with itself. To analyze the relative Stokes radius of Inv497 and Inv197, the proteins were size‐fractionated on a Superose 6 column (Materials and methods). No Inv497 protein was detected in the elution volume containing globular proteins of predicted mol. wt <60 kDa, where monomeric Inv497 with an apparent molecular mass of 52.6 kDa would be expected to fractionate (Figure 3A). Peak elution of Inv497 occurred in a region of the column predicted to contain spherical proteins of ∼200 kDa (Figure 3A). This indicates that the protein is either multimeric, has an elongated structure or both. In contrast, purified Inv197 protein fractionated with a peak elution consistent with a 20 kDa species, indicating that this derivative is monomeric (Figure 3A). Sedimentation equilibrium data, however, gave a contrasting result. For the most dilute initial concentration of protein analyzed, Inv497 sedimented as would be expected for a monomeric derivative with an apparent mol. wt of 51.4 kDa (Table I). The behavior of this protein, however, was highly non‐ideal. Sedimentation analysis of protein analyzed at a relatively low initial protein concentration (1.5 mg/ml) gave results that strongly deviated from the expected mol. wt of 52 600 (Table I). In addition, the plot of ln (concentration) versus r2 was non‐linear, indicating non‐ideal behavior of the protein (data not shown). These results may be due to charge effects, an elongated structure of the protein or self‐association during sedimentation.
In vitro and in vivo cross‐linking of invasin identifies invasin oligomers
As two physical techniques gave inconsistent results, the possibility of physical interaction between invasin monomers dependent on a region of the protein found in Inv497 but not in Inv197 was investigated further. Chemical cross‐linking experiments of Inv497 using formaldehyde, BS3 and Sulfo‐EGS generated two dominant species with apparent molecular masses of 120 kDa (35% of total Inv497 antigen) and 240 kDa (8% of total Inv497 antigen) for all cross‐linking agents used (Figure 3B, data for BS3 not shown). The difference in molecular mass between the two identified species and the Inv497 monomer (52.6 kDa) was consistent with the formation of both homodimeric and homotetrameric cross‐linking products. Formaldehyde resulted in cross‐links that were able to be disrupted by boiling and liberated the unaltered individual cross‐link component (Figure 3B, lane 5). It should be noted that even in the absence of cross‐linking agents, unboiled samples containing SDS gave a small amount of the predicted dimeric form of the Inv497 molecule detectable on Western blots (Figure 3B, lanes 2 and 6). The same cross‐linking treatment had no effect on the Inv197 protein and did not lead to the formation of higher molecular weight products under any conditions (data not shown).
To determine if the interaction between invasin monomers observed with the purified Inv497 occurs with intact invasin in the bacterial outer membrane, in vivo cross‐linking experiments were performed. An E.coli inv+ strain was subjected to cross‐linking by formaldehyde, BS3 or Sulfo‐EGS at 0°C; extracts were prepared, and the results were analyzed by gel electrophoresis and immunoblotting. In all extracts, the 108 kDa invasin polypeptide and several species lacking various amounts of the N‐terminus were detected, as noted previously (Figure 4B, lane 2; Isberg et al., 1987; Isberg and Leong, 1988). In extracts from intact bacteria subjected to cross‐linking by formaldehyde, two novel bands appeared that were of apparent molecular weights greater than the wild‐type invasin monomer (Figure 4B, lanes 3 and 4). Parallel experiments with cells lacking the inv gene showed no reaction with the anti‐invasin monoclonal antibody (Figure 4B, lane 1). The largest cross‐linking product had an apparent molecular mass of ∼400 kDa or higher, consistent with that of an invasin tetramer. Its weak intensity may be explained by difficulty in the electrophoretic transfer of higher molecular weight complexes. The smaller of the two cross‐linked bands had a mol. wt of ∼140 kDa, and its appearance was accompanied by the disappearence of the truncated 70 kDa invasin product upon addition of increasing amounts of cross‐linking reagents (Figure 4C, lanes 3 and 4). This is consistent with the 140 kDa cross‐reactive species being the dimeric form of the 70 kDa species. Identical results were obtained with two membrane‐impermeant high molecular weight cross‐linkers, Sulfo‐EGS (Figure 4C) and BS3 (Figure 4D), which were used to exclude the possibility of cross‐linking of any potential intracellular precursors of invasin.
To investigate the region of invasin responsible for cross‐linking, the identical experiments were performed on bacteria expressing the internal deletions InvΔKpn and InvΔHpa, lacking residues 608–794 and 804–850, respectively (Figure 4A). Both derivatives are surface exposed at almost the same levels as wild‐type invasin (see Table II). In the case of InvΔHpa (Figure 4D, lanes 4–6), bands of a higher molecular mass corresponding to a tetrameric form of the HpaI deletion product and a dimer of the shorter truncated product were detected. Interestingly, multimer formation of the InvΔHpa protein seemed to be somewhat more efficient than that of the wild‐type, with almost complete disappearance of the full‐length 104 kDa product (Figure 4D, lanes 5 and 6). In contrast, no cross‐linked product was detected with the InvΔKpn deletion product (Figure 4D, lanes 7–9), consistent with a loss of the capacity for protein–protein interaction. These results indicate that a region upstream from the C‐terminal cell‐binding region is required for cross‐linking, which is consistent with the cross‐linking data on the purified Inv 497 and Inv197 derivatives.
A binding‐ and uptake‐deficient mutation in Asp911 of invasin confers a dominant‐negative phenotype
Several residues of invasin have been identified that are necessary for efficient cellular penetration (Leong and Isberg, 1993; Leong et al., 1995; Saltman et al., 1996). The most critical residue for cell attachment and entry is Asp911 of the C‐terminal cell attachment domain (Figure 1A). Changes in this residue lower bacterial penetration to background levels and almost entirely abrogate cell attachment, such that only the most conservative mutation D911E retains detectable cell attachment activity (Leong et al., 1990, 1995). To assess further the ability of invasin to form multimers in vivo, we investigated the effects of the internal deletions, InvΔHpa and InvΔK, and three Asp911 point mutations on bacterial entry when co‐expressed with wild‐type invasin.
Escherichia coli strains harboring high copy number plasmids with the inv wild‐type or mutated derivatives in the presence or absence of a chromosomal inv gene were tested for cell entry into cultured HEp‐2 cells (Table II). Mutated invasin derivatives tested in the absence of the chromosomal inv gene showed significantly decreased cell binding and bacterial internalization efficiencies relative to the chromosomal inv+ wild‐type control (Table II). Notably, the internal invΔKpn deletion product which was unable to mediate invasin interaction in the cross‐linking experiments (Figure 4D) exhibited only 4% of wild‐type uptake efficiency, although it showed significant cell adhesion. Consistent with previous observations (Leong et al., 1995), strains expressing the mutated derivatives InvD911A, E or T lost activity completely and promoted neither cell binding nor entry.
When transformed into the strain XAcSu6(λgt11inv8–15) harboring a chromosomal inv gene, the plasmid‐encoded InvD911A or InvD911T derivatives showed significant dominant‐interfering effects on internalization (Table II). The InvD911T mutant strain was 50% less efficient in bacterial uptake, while InvD911A reduced internalization efficiency to ∼5% of wild‐type. Most importantly, neither point mutation interfered with cell attachment. In contrast, mutations that prevent invasin interaction (invΔKpn) or have a less severe effect on cell uptake (D911E) did not interfere with bacterial internalization by wild‐type invasin (Table II). Surprisingly, the invΔHpa deletion product did not confer a dominant‐interfering phenotype. This might be explained by the efficient InvΔHpa self‐association, observed in the cross‐linking experiments (Figure 4D), which may reduce the overall amount of non‐functional heteromultimeric complexes. Alternatively, the deletion product may result in a protein structure that is unable to associate with wild‐type invasin.
The interfering effect of the D911A mutation was also found in Y.pseudotuberculosis, although the uptake defect was less pronounced than that seen with E.coli (data not shown). In Yersinia, a lower plasmid copy number may explain the less drastic effect of the D911A mutation.
A region of invasin promotes protein–protein interaction
To map the region in invasin responsible for protein–protein interaction, the one‐hybrid λ repressor system was used (Hu, 1990). This allows testing of potential multimerization domains fused to the monomeric DNA‐binding domain (cIN) of λ repressor. Only multimeric λ repressor binds strongly to the λPR operator and efficiently represses transcription of a λPR‐lacZ reporter gene. Derivatives were constructed in which the DNA‐binding domain of cI was fused to the C‐terminal 478 and 202 amino acids of invasin (cIN–Inv478 and cIN–Inv202). Repression of transcription by the cIN–Inv fusion proteins was compared with: (i) the properties of intact λ repressor (cI); (ii) a cI‐leucine zipper chimera (cIN–GCN4) that behaves similarly to intact λ repressor (Hu, 1990); (iii) the DNA‐binding domain of cI alone (cIN); and (iv) the cI‐leucine zipper chimera interrupted by truncated lacZ sequences (cIN–LacZ′). All chimeric proteins were readily detected on SDS–polyacrylamide gels of cell extracts under inducing conditions, and their steady‐state levels were essentially identical under non‐inducing conditions, based on Western blot analysis (data not shown). Wild‐type λ repressor and the cIN–GCN4 chimera yielded 86 and 88% repression of λPR, respectively, whereas the monomeric cI N‐terminus showed only 41% repression, as expected (Figure 5, Materials and methods). Cells expressing the cIN–Inv478 protein exhibited 84% repression of λPR, essentially identical to the cIN–GCN4 chimera positive control (Figure 5, lane 6). In contrast, the cIN–Inv202 protein was unable to restore cI repressor activity (48% repression; Figure 5, lane 7), indicating that this chimera behaves like a monomeric protein. Strains expressing chimeric proteins that result in a functional dimeric or multimeric repressor, such as cI and cI–GCN4, confer immunity against the lytic phage λKH54(ΔcI). In contrast, clones expressing monomeric cI proteins are sensitive to the phage (Hu, 1990). Consistent with previous results, bacterial strains expressing cI–Inv478 were immune to the bacteriophage, implying multimer formation, while cIN–Inv202 failed to allow immunity to the phage (Figure 5).
To characterize further the region of the invasin protein that is responsible for functional cI repressor formation, several deletion derivatives of cIN–Inv478 were constructed (Figure 5, lanes 8–10). The cIN–InvΔHpa and cIN–Inv509–710 proteins, carrying deletions in the invasin C‐terminus that destroyed the cell adhesion domain, showed full repressor activity (90 and 94% repression, respectively) and conferred resistance to bacteriophage λKH54, whereas the internal KpnI deletion derivative, cIN–InvΔKpn, was unable to repress λPR transcription fully, and bacteria expressing this derivative were sensitive to λKH54 (Figure 5, lanes 8–10). Thus, the region responsible for protein–protein interaction was localized in the internal conserved module II displayed in Figure 1A, consistent with the results from the in vivo cross‐linking experiments (Figures 3B and 4). Twelve different PCR‐derived cIN–Inv fusions encoding different portions of the invasin module II were analyzed to determine the smallest region capable of stimulating repression by cI. All cIN–Inv constructs containing amino acids 596–694 of the invasin protein fully repressed λPR‐lacZ expression, showing >79% repression, and were resistant to phage KH54 (Figure 5, lanes 11–13). Shorter derivatives failed to show repression (Figure 5, lanes 14 and 15), so this domain may be the minimal region necessary for self‐association.
Multivalency stimulates invasin‐mediated internalization
Several lines of evidence in this report indicate that a multimeric form of the invasin protein of Y.pseudotuberculosis may be the active player in the bacterial uptake process. Therefore, artificially produced bivalent invasin containing the C‐terminal cell adhesion region was tested for its ability to promote bacterial uptake relative to a monomeric derivative. The MBP–Inv197 hybrid protein was size‐fractionated, and the peak fraction corresponding to the 60 kDa monomer was isolated from aggregated protein found in this preparation (Saltman et al., 1996). This fraction was then incubated with latex beads coated with either purified monomeric Fab fragment directed against MBP, or purified anti‐MBP, the latter of which should generate bivalent invasin. The coating of the beads was adjusted to allow an identical amount of hybrid protein to be bound for each derivative, and beads adjusted in this fashion were used to challenge HEp‐2 cells to assay for internalization, using the fluorescence challenge strategy (see Materials and methods). Beads coated with the monomeric MBP–Inv197 protein bound equally well when coated in either a monovalent or bivalent fashion, yet the entry efficiencies were significantly affected by the valency of coating (Figure 6). Beads coated with monomeric Fab fragment and MBP–Inv197 were not internalized efficiently and uptake was only slightly higher than that observed with MBP, bound to either anti‐MBP antibody or Fab fragment (Figure 6). In contrast, up to 20% of the adherent beads coated with monomeric MBP–Inv197 that had been made artificially bivalent by antibody entered HEp‐2 cells (Figure 6, MBP–Ab/MBP–Inv197). Therefore, increasing the valency significantly enhanced uptake of this derivative without altering adhesion.
As seen above (Figures 1B–E and 2), beads coated with the Inv497 derivative bound somewhat better than Inv197‐coated beads. Surprisingly, beads coated with MBP–Inv497 and divalent antibody had a higher adhesion efficiency than those coated with Fab fragment‐linked Inv497. In contrast to the Inv197‐coated beads, dimerization of the Inv497 derivative had only a minor stimulatory effect on uptake efficiency of bound beads. Both Fab and antibody‐coupled Inv497 proteins resulted in 50–60% entry of all cell‐adherent beads (Figure 6), suggesting that the valency of Inv497 alone was sufficient to promote uptake of bound beads.
MBP–Inv497 internalization and integrin redistribution
It has been well established that ligand‐ and antibody‐induced clustering of β1 integrins can mediate transmembrane signal transduction, inducing a variety of cellular responses (Kornberg et al., 1991; Miyamoto et al., 1995a). Therefore, binding of invasin to β1 integrins was studied by analyzing the distribution of MBP–Inv derivatives on cultured cells after short incubation times. Size‐fractionated MBP–Inv197 and MBP–Inv497 were added directly to adherent HEp‐2 cells, incubated at 20°C for various times and fixed samples were probed by immunofluorescence to visualize protein distribution on the cell surface (Figure 7). In HEp‐2 cells exposed to MBP–Inv197, the distribution of the protein was somewhat concentrated on the cell margins, with few patches evident (Figure 7C and G). When cells were incubated with the size‐fractionated MBP–Inv497 derivative, however, the protein was observed as focal patches on the cell surface. Occasional redistribution of invasin–integrin complexes was apparent 7 min after addition of MBP–Inv497 to cells (Figure 7D), with significant accumulation of distinct punctate clusters found at 60 min after addition of the protein (Figure 7H). No punctate forms were observed for MBP–Inv197 at this time point (Figure 7G). Patches were also observed on control cells treated with anti‐β1 integrin antibody, although the aggregates were larger and less frequent than that seen for MBP–Inv497 (Figure 7B and F). No such forms were seen on cells exposed to purified MBP (Figure 7A and E).
The patches of invasin appear to be a reflection of the fact that the MBP–Inv497 derivative is internalized, whereas the the smaller derivative is not. This is based on analysis of the internalization of invasin derivatives and their ability to induce redistribution of integrin receptors. HEp‐2 cells incubated with the invasin derivatives MBP–Inv497 or MBP–Inv197 were probed differentially with anti‐MBP, either before or after permeabilization, to determine if the cell‐associated invasin was internalized (Figure 8). Confocal microscopy revealed that 7 min after binding, both of the derivatives were localized primarily extracellularly, with little intracellular staining by the antibody (Figure 8A and B; external invasin derivatives colored yellow or red). In contrast, by 60 min post‐incubation, the MBP–Inv497 derivative was largely internalized (Figure 8D; internal invasin visualized as green), whereas the the MBP–Inv197 was localized mostly at the edges of cells on the external surface (Figure 8C). Cells challenged with invasin derivatives in this fashion, fixed and probed with anti‐β1 integrin antibody, gave results that were consistent with the above analysis. HEp‐2 cells incubated with invasin derivatives for short periods of time or extensively with MBP–197 resulted in integrin receptor that was localized largely around the edges of cells, when viewed ∼0.6 μm above the adhesion plane (Figure 8E–G). In contrast, 60 min of incubation of HEp‐2 cells with the MBP–Inv497 derivative caused significant redistribution of the integrin receptor (Figure 8H). These results indicate that after binding to HEp‐2 cells, clustering and internalization of invasin, as well as redistribution of integrin receptor, were all enhanced for Inv497 relative to Inv197.
In previous studies, it has been shown that the C‐terminus of invasin is essential for binding and uptake, and can promote entry when coated on non‐adherent bacteria (Leong et al., 1990; Rankin et al., 1992). It has been assumed that bacterial uptake promoted by invasin is strictly a result of the ability of the protein to adhere to integrins. In this study, we report the identification and analysis of an additional property of invasin necessary for high efficiency uptake. Moreover, evidence is presented that a specific domain linked to this activity has the capacity to mediate intermolecular interaction. Using a bead challenge assay, we showed that particles coated with low concentrations of a purified fragment comprising the C‐terminal cell‐binding domain of invasin (Inv197) adhered to epithelial cells, but showed little internalization. In contrast, beads coated with a larger invasin derivative comprising the C‐terminal 497 amino acids of invasin (Inv497) were internalized far more efficiently even at low coating concentrations (Figure 2). Protein sequences present in Inv497, but absent in Inv197, appeared to enhance entry and contributed to the high efficiency of invasin‐mediated uptake.
The region of invasin necessary to promote enhanced uptake contains a specific domain that is capable of promoting intermolecular interaction and is located within a region of the protein required for chemical cross‐linking on the bacterial cell surface. We examined various invasin derivatives by chemical cross‐linking and found that the invasin wild‐type protein, as well as the deletion product InvΔHpa and the purified Inv497 derivative, formed species with apparent molecular weights consistent with multimerization. In contrast, no oligomeric complexes were found with the deletion product InvΔKpn and the Inv197 protein (Figures 3B and 4), indicating that the sequences responsible for protein–protein interaction are absent from these derivatives. The identification of dominant‐interfering mutations located in the C‐terminal integrin recognition site of the protein is also consistent with the property of multimerization.
Invasin sequences encompassing amino acids 596–694 of the invasin protein fused to the monomeric DNA‐binding domain of cI caused stimulation of repressor function (Figure 5). This region is located N‐terminal to the extracellular cell adhesion domain that recognizes integrin receptors (Figure 1A) and previously had not been assigned a specific function. Preliminary crystallographic analysis (Z.A.Hamburger and P.J.Bjorkman, unpublished data) of the Inv497 peptide indicate that the smallest region competent to confer high‐level repression demarks the endpoints of a distinct globular domain. This region stands out by having a large number of uncharged amino acids (76 out of 99 amino acids; Figure 9A).
The most striking result is the demonstration that bivalent invasin complexes are more powerful mediators of uptake than their monovalent counterparts (Figure 6). We showed that size‐fractionated monomeric invasin derivative Inv197 immobilized on latex beads by divalent antibodies promoted far more internalization than when it was immobilized by monomeric Fab fragments. Interestingly, the invasin protein of Y.enterocolitica, which permits 6‐ to 60‐fold less efficient uptake by cultured cells than the Y.pseudotuberculosis protein (Pepe and Miller, 1990), is missing the domain required for protein–protein interaction (Figure 9A). As expected from this study (Figure 2), high production levels of the Y.enterocolitica protein can compensate for the absence of this domain (data not shown). The intimins encoded by enteropathogenic E.coli strains (EHEC, EPEC) and Citrobacter freundii, which are homologous to invasin, also apparently lack this particular region (Figure 9A). Intimins are involved normally in the formation of special attachment and effacing structures (pedestals) on the host cell surface, which immobilize the bacteria extracellularly without causing large‐scale bacterial uptake (Jerse et al., 1990; Frankel et al., 1994). The behavior of the Y.enterocolitica invasin is similar to that of an invasin deletion derivative (InvΔKpn) that has most of the enhancer region removed. This derivative could not be cross‐linked (Figure 4) and promoted entry at only 4% the efficiency of the wild‐type protein (Table II). Thus, lack of multivalency strongly correlates with a lower uptake efficiency, emphasizing the importance of invasin–invasin interaction for the internalization process.
The simplest explanation for why uptake is enhanced by multimerization is that clustering results in increased affinity of invasin for its integrin receptors (Tran Van Nhieu and Isberg, 1993a). Previous studies have shown that two critical determinants that modulate invasin‐mediated uptake are integrin receptor density and the relative affinity at which the ligand binds receptor (Tran Van Nhieu and Isberg, 1993a; Leong, et al., 1995). Efficient uptake requires high affinity binding to the integrin receptor, while low affinity ligands promote bacterial adhesion to the mammalian cell without subsequent uptake (Tran Van Nhieu and Isberg, 1993a). Monomeric invasin may be of insufficient affinity to promote efficient uptake of Y.pseudotuberculosis, and multimerization would increase the binding affinity sufficiently to promote efficient uptake. A precedent for this phenomenon has been noted with the intracellular adhesion molecule‐1 (ICAM‐1), a potent β2 integrin ligand on lymphocytes. This protein forms homodimers at the host cell surface which enhance its activity in cell adhesion relative to its monomeric derivatives of ICAM‐1 (Miller et al., 1995).
An alternate model proposes that multimerization is required to send a necessary signal. Invasin multimers, containing multiple receptor‐binding domains, might interact simultaneously or cooperatively with several receptor molecules, inducing conformational changes and/or receptor clustering (Figure 9B). Results obtained in recent years provide evidence that multivalent ligand‐induced receptor oligomerization and clustering are required to trigger a large number of intracellular signaling events generated by integrin receptors (Kornberg et al., 1991; Miyamoto et al., 1995a,b). The elucidation of signal transduction pathways involved in integrin‐mediated signaling has revealed that the activity of tyrosine kinases and other signaling molecules as well as the participation of cytoskeletal components in these pathways is regulated by the nature of the bound ligand and the multimeric state of the integrin receptor (Miyamoto et al., 1995a,b). Similarly, oligomerization by multivalent invasin binding may generate the crucial signal to recruit appropriate signaling molecules and cytoskeletal elements for bacterial uptake (Figure 9B). By the clustering model, close proximity of receptors in a relatively small area of the cell surface would be essential for signal transmission. As receptor clustering is predicted to be sufficient to transmit an internalization signal, a high concentration of a monomer should promote entry as efficiently as a lower concentration of multimeric ligands (Figures 1 and 6).
The results of sedimentation analysis of the soluble C‐terminus of invasin (Table I), the low efficiency of cross‐linking (Figure 4) and crystallographic analysis of the C‐terminal 497 amino acids (Z.A.Hamburger and P.J.Bjorkman, unpublished data) indicate that invasin may exist primarily as a monomer in the absence of membrane localization. By limiting diffusion of the protein to two dimensions, the outer membrane may stabilize transient multimer formation, particularly for individual processing products of invasin. In fact, preferential multimerization can be observed for a 70 kDa C‐terminal invasin fragment, which was cross‐linked quantitively (Figure 4), indicating that it may exist entirely in a multimeric form. Further studies on the oligomerization state of invasin in the presence and absence of receptor binding should reveal if multimerized species are bound more efficiently to the integrin receptor.
Enteropathogenic Yersinia have developed a special strategy to usurp normal cellular functions (Tran Van Nhieu et al., 1996). Similar strategies have been seen in a large number of pathogens (Finlay and Cossart, 1997). In the present work, we have identified a novel role for multivalent ligand interaction in uptake and proposed invasin‐induced receptor oligomerization as a crucial step in efficient bacterial uptake. Future work should provide more information about the specific nature of the association between invasin and its receptor and allow deeper insight into integrin‐mediated signal transduction pathways.
Materials and methods
Bacterial strains, cell culture and media
Bacterial strains used in this study are listed in Table III. Overnight cultures of E.coli were grown at 37°C, Yersinia strains were grown at 28°C in Luria–Bertani medium. HEp‐2 cells were cultured in RPMI 1640 media (Irvine Scientific) supplemented with 5% newborn calf serum (Life Technology Inc.) and 2 mM glutamine at 37°C in the presence of 5% CO2.
DNA manipulations, plasmids and oligonucleotides
Preparations of plasmid DNA and phages, restriction digestions, ligations and transformations were performed as previously described (Sambrook et al., 1989; Miller, 1992). PCR reactions were performed in a standard 100 μl mix for 20 cycles in a DNA thermal cycler PTC‐200 (MJ Research). PCR products were purified with the QIAquick Kit (Qiagen) before and after restriction digestion of the amplification products.
Plasmids used in this study are listed in Table III. Plasmids pRI284 and pRI285 were constructed by fusing a PCR‐derived fragment of the inv gene of Y.pseudotuberculosis encoding the C‐terminal 197 and 497 amino acids of invasin to the malE gene of pMal‐c1 (New England Biolabs) inserted in the XbaI and HindIII sites. Plasmids pPD208 (cIN–inv478) and pPD210 (cI–inv202) were constructed by fusing a PCR‐derived fragment containing codons 509–986 and 795–986 of the Y.pseudotuberculosis inv gene in‐frame with cIN at the SalI site on vector pJH391 (Hu, 1990). The following primers were used. Forward primer inv478, 5′‐TAACGTCGACCGTCATTGGTGATGGC‐3′; forward primer inv202, 5′‐CCCTGTCGACGGTACCTACGCTGACC‐3′; reverse primer inv478 and inv202, 5′‐CCAGGATCCTGGGCCGTAAGATCGG‐3′. The inv in‐frame deletions of plasmids pPD211, pPD212 and pPD213 were derived from plasmid pPD208 by EcoRV (cI–inv509–710), KpnI (cI–inv509–608,795–986) or HpaI (cI–inv509–803,851–986) digests and religations. The plasmids pPD223, pPD226–227 and pPD244 were constructed by inserting PCR‐derived SalI–BamHI fragments containing codons 542–694 (pPD223), 575–694 (pPD226), 575–657(pPD227) and 593–694 (pPD244) using the upstream primers 5′‐GGGGGGTCGACGGTGATAACCACCAATAATGGTGCG‐3′, 5′‐GGGGGGTCGACCGTGACGGTAGTCACAGCAGAAGTGG‐3′ and 5′‐GGGGGGTCGACAAGGGTACTATCGCGGCGGATAAATCC‐3′, and the respective downstream primers 5′‐CCGCGGGATCCCTTAATCTGCCGTGAAATTAACCG‐3′ and 5′‐CCGCGGGATCCCTTAGCCGTCATTGTGATCCGTGATAAC‐3′. Oligonucleotides were synthesized by the Howard Hughes Medical Institute Microchemical Facility, Harvard Medical School.
Escherichia coli strains JH372 harboring derivatives of pJH391 were grown overnight at 37°C. To lyse the cells, 20 μl of 0.1% SDS and 40 μl of chloroform were added to 100 μl of the culture and incubated for 5 min at room temperature. Assay of β‐galactosidase was performed as described by Miller (1992) and activity was calculated as follows: β‐galactosidase activity = [OD420 assay]×6.75/[OD600 cells]×min (time of reaction)×0.1 ml (reaction volume).
In vitro and in vivo cross‐linking
For in vitro cross‐linking, various concentrations of formaldehyde (37% w/w, Fisher) or Sulfo‐EGS (Pierce, USA) were added to 10 μg of purified Inv497 protein in 25 mM HEPES (pH 7.0), 1 mM EDTA. Samples were incubated for 30 min at room temperature with gentle shaking rotation, and reactions were stopped by adding ethanolamine (pH 7.4) in a final concentration of 50 mM. Prior to 10% SDS–PAGE, a portion of a sample containing no or 0.8% formaldehyde was heated for 15 min at 100°C, to reverse the cross‐linking. Samples were analyzed by 10% SDS–PAGE, followed by immunoblotting using anti‐invasin mAb3A2 (Leong et al., 1990). For in vivo cross‐linking, XL1blue harboring invasin derivatives was grown in L broth with 100 μg/ml ampicillin at 37°C to an OD600 = 0.7. The cells were harvested by centrifugation, washed once, either with NaPO4 buffer (pH 7.5) for formaldehyde or 10 mM HEPES (pH 7.5), 1 mM EDTA for BS3 (Pierce) and Sulfo‐EGS (Pierce) cross‐linking, and resuspended in the identical buffers to an OD600 = 1.0. The reactions were allowed to proceed at 0°C for 1 h with shaking and were terminated in the presence of ethanolamine, as above. Samples were washed twice with the reaction buffer containing ethanolamine. Pellets were resuspended in 40 μl of sample buffer and aliquots of 10 μl were analyzed by 10% SDS–PAGE followed by immunoblotting using invasin antibodies mAb3A2 and mAb2A9 (Leong et al., 1990).
Covalent coupling of protein and antibodies to latex beads
About 108 latex beads (1.1 μm diameter, Sigma) were placed in glass centrifuge tubes and washed sucessively in 1 ml of phosphate‐buffered saline (PBS) and 1 ml of 0.2 M Na2HCO3 (pH 8.5), 0.5 M NaCl (coupling buffer) and were resuspended in 100 μl of coupling buffer. Purified Inv197 and Inv497 proteins (isolated as described earlier) were added in a range of concentrations from 0 to 1 mg/ml. Proteins were allowed to adsorb to the beads for 30 min at 37°C. Subsequently, 20 μl of biotinylated BSA (1 mg/ml) was added and incubated for an additional 30 min at 37°C. After adding 500 μl of coupling buffer to the beads, the solution was sonicated for 20 s, and 500 μl of 20 mg/ml BSA in coupling buffer was added and incubated at 37°C for an additional hour to complete blocking. The beads were washed in PBS containing 10 mg/ml BSA and stored in 200 μl of PBS containing 2 mg/ml BSA at 4°C. The coupling efficiency, and protein concentrations of the starting solution added to beads and the supernatant before addition of BSA were determined using the BCA Kit (Pierce). For anti‐MBP Ab/Fab‐coated beads, the beads were prepared in a similar manner and coated at 200 μg/ml. To allow coupling of the invasin derivatives, the Ig‐coated beads were incubated with 50 μg/ml of size‐fractionated MBP–Inv197 and MBP–Inv497 protein in PBS containing 1% BSA at 4°C overnight, washed three times with PBS to remove unbound protein and stored in PBS with 0.2% BSA.
To determine the number of invasin molecules on beads, purified proteins were labeled with carboxytetramethyl rhodamine, succinimidyl ester (Molecular Probes, Inc.) as described (Rankin et al., 1992) or analyzed by surface enzyme‐linked immunosorbent assays (ELISAs), as follows: 1:2 dilutions of 108 beads were washed in PBS and incubated with primary anti‐invasin antibody mAb3A2 at room temperature in PBS containing 2% goat serum for 1 h. Subsequently, beads were washed three times with PBS and incubated with anti‐mouse IgG–alkaline phosphatase at room temperature for an additional hour. After three subsequent washes alkaline phosphatase assays were performed with 1 mg/ml σ104 (Sigma) in AP‐buffer (100 mM Tris–HCl pH 9.5, 5 mM MgCl2, 150 mM NaCl2) and quantitated at 405 nm with a microtiter spectrophotometer (BioRad). The amount of invasin on the bead was then quantitated and calculated from standard ELISAs of the same protein coated on plastic. To this end, serial dilutions of the protein were incubated on 96‐well plates for 16 h at 4°C, and the amount of bound protein was determined by subtraction of the amount of remaining protein in the supernatant from that of the in‐put protein.
Internalization of bacteria into mammalian cells, quantitation of cell binding of invasin‐expressing bacteria and quantitation of invasin proteins on the surface of bacteria
Penetration of bacteria into cultured mammalian cells was assayed by gentamicin protection assays, as described (Leong et al., 1990). Approximately 5×105 E.coli bacteria expressing the wild‐type inv gene or various inv mutations were centrifuged onto a subconfluent HEp‐2 cell monolayer and incubated for 90 min at 37°C. Extracellular bacteria were then killed by a 90 min treatment in the identical media containing 50 μg/ml gentamicin. The surviving intracellular bacteria were released with 0.1% Triton X‐100, and viable counts were titered on bacteriological media. For each individual clone, the relative level of bacterial uptake was determined by calculating the number of colony‐forming units that arose relative to the total number of bacteria in the assay.
Surface expression of invasin proteins was determined by mAb3A2 and mAb2A9 probing. Bacteria were incubated in PBS containing 2% goat serum and 1 μg/ml primary anti‐invasin antibodies mAb3A2 or mAb2A9 followed by reprobing with 5 μg/ml goat anti‐mouse IgG alkaline phosphatase (Zymed) for 1 h at room temperature. The cell density was determined at A600, and the bacteria were incubated in AP‐buffer with 1 mg/ml σ104 alkaline phosphate substrate to assay for bound antibody, as described earlier.
Cell binding of bacteria expressing invasin derivatives was determined as described (Leong et al., 1995).
Expression and purification of invasin proteins
Twenty liters of SR2 (pRI284) or SR2 (pRI285) were grown at 28°C in L broth to an A600 = 0.6. Isopropyl‐β‐d‐thiogalactopyranoside (IPTG) was added to final concentration of 1 mM to induce the expression of MBP–Inv fusions. The cells were grown for an additional 2 h before being harvested. Frozen cell pellets were resuspended in 50 ml of 10 mM Tris pH 8.0 plus protease inhibitor cocktail containing 5 mM phenylmethylsulfonyl fluoride (PMSF), 10 mM pepstatin (Sigma), 10 mM E64 protease inhibitor (Boehringer Mannheim, Germany), 20 μM leupeptin (US Biochemical) and 10 μM chymostatin (Sigma). The MBP–Inv197 protein was purified by affinity chromatography on cross‐linked amylose, as described previously (Leong et al., 1990; Rankin et al., 1994). MBP–Inv497 was purified by ion exchange chromatography, with slight modification of previous procedures (Leong et al., 1995). A cell extract containing MBP–Inv497 was loaded onto a HiLoad 26/10, Fast Flow Q Sepharose column (Pharmacia, Biotech, Inc.) equilibrated with 10 mM Tris pH 8.0 and eluted with a continous salt gradient (0–500 mM NaCl). The MBP–Inv497 protein that eluted at ∼100–120 mM NaCl was pooled, precipitated in 40% ammonium sulfate, resuspended in 10 mM Tris pH 8.0 plus protease inhibitor cocktail and dialyzed twice against 100 volumes of the same buffer. Subsequently, the protein was reloaded on the Q Sepharose column and pure peak fractions were pooled. To liberate Inv 497 and Inv197 proteins from the MBP moiety, both proteins were cleaved by factor Xa (New England Biolabs) adding 0.5 mg of protease per 100 mg of fusion protein in factor Xa cleavage buffer (15 mM Tris pH 8.0, 100 mM NaCl, 2 mM CaCl2), and incubated at 14°C for 12 h. The protease reaction was stopped by 10 μM 1,5‐DNS‐GGACK‐HCl inhibitor (Calbiochem) and the MBP protein was removed by loading the samples on a cross‐linked amylose column equilibrated with 15 mM Tris–HCl (pH 8.0). Pure Inv497 and Inv197 were collected from the column flowthrough, and protein concentrations were determined by the BCA protein assay (Pierce).
A portion of the MBP–Inv proteins aggregate into soluble higher molecular weight complexes. To isolate MBP–Inv197 and MBP–Inv497 hybrid proteins from larger complexes, protein preparations were subjected to gel filtration chromatography on a 10 mm×30 cm Superose 12 column (Pharmacia Biotech, Inc.). The column was equilibrated with 20 mM HEPES (pH 7.0) and calibrated using a variety of protein standards with known molecular sizes: 669 kDa thyroglobin, 440 kDa ferritin, 232 kDa catalase, 158 kDa aldolase, 67 kDa BSA and 43 kDa ovalbumin. Samples (200 μl) of the MBP–Inv fusion proteins (5 mg/ml) were applied to this column and collected in 200 μl fractions. Proteins eluted in the void volume were discarded and peak fractions corresponding to the MBP–Inv197 and the MBP–Inv497 species in the included volume were pooled and concentrated using a Centricon 30 filter (Amicon, Beverly, MA). Thereafter, samples were refractionated on an identical gel filtration column to ensure that no aggregation of the purified sample occurred on storage and concentration.
Analytical gel filtration of purified Inv197 and Inv497 cleaved from MBP was identical to above, using 1 mg/ml Inv197 and 4 mg/ml Inv497, respectively. Elution profiles were monitored by UV absorption, and 200 μl fractions were collected and probed by immunoblotting using anti‐invasin mAb 3A2.
Purification of anti‐MBP and anti‐MBP–Fab fragments
A 500 μl aliquot of anti‐MBP serum was diluted with 1 ml of 10 mM Tris–HCl (pH 7.5) and incubated overnight at 4°C in a roller with 5 ml of a 50% protein A–Sepharose slurry (Pharmacia Biotech, Inc.) equilibrated in the same buffer. The material was poured into a 1×5 cm column (BioRad) and washed with 20 ml of 10 mM Tris–HCl (pH 7.5) and 20 ml of 100 mM Tris–HCl (pH 7.5). Antibodies were eluted with 100 mM citric acid (pH 3.5). The pH of the eluted antibodies was readjusted to pH 7.5 with 1 M Tris–HCl (pH 8.0). To isolate Fab fragments, the purified anti‐MBP IgG was incubated in 100 mM sodium acetate (pH 5.5), 50 mM cysteine, 1 mM EDTA containing 10 μg/ml papain/mg of IgG at 37°C overnight. The Fab fragments subsequently were separated from uncleaved Fab and Fc as described (Harlow and Lane, 1988).
Approximately 5×104 HEp‐2 cells were seeded and grown on coverslips placed inside individual wells of 24‐well cell culture plates (Costar). Cell monolayers were washed three times with PBS and incubated in RPMI 1640 medium supplemented with 20 mM HEPES (pH 7.0) and 0.4% BSA (binding buffer), before the addition of ∼5×106 beads (corresponding to 100 beads per cell). Beads were centrifuged onto the cell monolayer (1000 r.p.m., 5 min) and incubated at 37°C in a humidified atmosphere of 5% CO2. At 1 h post‐infection, the cells were washed three times with PBS and fixed with 2% paraformaldehyde in PBS for 20 min at room temperature. For differential staining of adherent and internalized invasin‐coated latex beads, fixed cell monolayers were incubated sequentially in PBS containing 2% goat serum for 10 min, PBS containing 2% goat serum and 1 μg/ml anti‐invasin mAb3A2 for 1 h at room temperature, and PBS containing 2% goat serum and 5 μg/ml anti‐mouse IgG–fluorescein isothiocyanate (FITC) for 1 h at 37°C. To detect beads blocked with biotinylated BSA, 5 μg/ml FITC–streptavidin (Pierce) was used. For anti‐MBP‐ or anti‐MBP Fab fragment‐coated beads, 1 μg/ml rabbit anti‐MBP (New England Biolabs) and 5 μg/ml anti‐rabbit IgG–FITC (Zymed) was used. Three washes in PBS were performed at the beginning and end of each incubation step. Stained samples were mounted in PBS containing 0.1% p‐phenylenediamine, 80% glycerol (v/v). Quantification of adherent and internalized latex beads was performed using a Zeiss Axioskope (Jena, Germany) with a fluorescence filter set. The total number of adherent and internalized beads was determined under phase contrast for 200 individual cells, and the number of adherent and intracellular beads per cell was determined as the mean of 50 individual cells. Error bars represent the standard deviations of mean values from four groups of 50 cells counted in various regions of the stained specimens. Data presented correspond to one of three typical independent experiments.
For clustering experiments, HEp‐2 cells were seeded and grown on coverslips placed inside individual wells of 24‐well cell culture plates (Costar). Cell monolayers were washed three times with PBS and incubated in binding buffer (see above) for 10 min at room temperature, before the addition of 10 μg/ml MBP–Inv197, 10 μg/ml MBP–Inv497 or 5 μg/ml anti‐β1 invasin mAb VD1 (Tran van Nhieu and Isberg, 1993a). The cells were incubated for 7 or 60 min at room temperature and washed three times with PBS to remove unbound protein or antibody. Subsequently, the cells were fixed with 2% paraformaldehyde and incubated successively with anti‐MBP (NE Biolabs) and goat anti‐rabbit Ig‐FITC (Zymed) or goat anti‐mouse Ig‐FITC (Boehringer Mannheim) in PBS containing 2% goat serum. Three washes in PBS were performed at the beginning and end of each incubation step. Stained samples were mounted in PBS containing 0.1% p‐phenylenediamine, 80% glycerol (v/v), and clusters visualized by microscopy, as described above.
We would like to thank Dr James Hu for providing strains and phages for the one‐hybrid assay system, Drs John Leong and Eric Krukonis for providing strains, and Dr Carol Kumamoto for anti‐MBP antiserum. We thank Zsuzsa A.Hamburger and Dr Pamela J.Bjorkman for unpublished information, and Dr Pak Poon for running the sedimentation analysis. We also thank Drs Dorothy Fallows, Martin Fenner and Joseph Vogel for helpful discussion and critical reading of the manuscript. This work has been supported by the Howard Hughes Medical Institute and the NIH‐Grant RO1‐AI23538. P.D. is a recipient of a research fellowship of the Deutsche Forschungsgemeinschaft.
- Copyright © 1999 European Molecular Biology Organization