Immunity proteins to pore‐forming colicins: structure‐function relationships

Colicin A and B immunity proteins (Cai and Cbi, respectively) are homologous integral membrane proteins that interact within the core of the lipid bilayer with hydrophobic transmembrane helices of the corresponding colicin channel. By using various approaches (exchange of hydrophilic loops between Cai and Cbi, construction of Cbi/Cai hybrids, production of Cai as two fragments), we studied the structure‐function relationships of Cai and Cbi. The results revealed unexpectedly high structural constraints for the function of these proteins. The periplasmic loops of Cai and Cbi did not carry the determinants for colicin recognition although most of these loops were required for Cai function; the cytoplasmic loop of Cai was found to be Involved in topology and function of Cai. The immunity function did not seem to be confined to a particular region of the immunity proteins.


Introduction
Pore-forming coiicins form voltage-dependent ion-channels in the cytoplasmic membrane of sensitive bacteria (Konisky. 1982). After these channels have formed, the cells are depleted of their intracellular K* (Bourdinaud et al., 1990), active transport is inhibited (Cramer et ai, 1983), and inorganic phosphate leaks out of the cells. The leakage causes intracellular ATP to be hydrolysed into ADP and AiVIP (Guihard etai. 1993).
Like many toxins, coiicins are made up of functional domains, each of which is associated with particular functions in the overall activity (Baty et ai, 1988). The Nterminal and the central domains of pore-forming coiicins are involved in their transport through the Escherichia coli envelope; the Oterminal domain carries the ionophoric activity (Lazdunski etai, 1988;Pattus etai, 1990).
The crystal structure of the pore-forming domain of colicin A (ColA) in aqueous solution was obtained at 2.4 A resolution (Parker etai, 1989;. This domain consists of a hydrophobic hairpin (helices 8 and 9) surrounded by eight amphipathic ^-helices. In vitro experiments performed on model membranes at low pH indicate that if the membrane potential is missing, the pore-forming domain of ColA remains embedded at the surface of the membrane (Geli etai, 1992;Lakey etai, 1993;Duche et ai, 1994). The hydrophobic hairpin of the homologous colicin El (ColEI) seems to assume a transmembrane orientation (Merril etai, 1990;Rath etai, 1991;Zhang and Cramer, 1992;Shin et ai, 1993). The structure of the membrane-spanning ion-channel has not been determined. It probably includes the hydrophobic hairpin and an amphiphilic hairpin that was identified in C0IEI as the voltage-sensitive region (Menill and Cramer, 1990;Abrams etai, 1991;Song e/a/., 1991;Zhang and Cramer, 1992).
To prevent the colicin they release in the extracellular medium from killing them, colicin-producing strains also synthesize immunity proteins (Geli et ai, 1986). According to their respective homologies, immunity proteins were classified into two groups (Schramm etai, 1988;Geli etai, 1989;. The first group comprises immunity proteins to coiicins A, B, and N, and the second comprises immunity proteins to coiicins El, la, and lb. Colicin A and B immunity proteins show 38% identity and 39% conservative substitutions; their sequences can be aligned with the introduction of very few gaps. Immunity proteins are integral inner membrane proteins (Goldman etai, 1985;fviankovich etai, 1986;Schramm et ai, 1988;Geli etai, 1988;Pugsley, 1988). They interact with the pore-forming domain of the corresponding colicin (fylankovich etai, 1984;Bishop etal., 1985;Benedetti et ai, 1991;Geli and Lazdunski, 1992a). The topologies of colicin A and El immunity proteins (Cai and Cei, respectively) are known: Cai has four transmembrane regions and its N-and C-termini face the cytoplasm (Geli et ai, 1989); Cei crosses the membrane biiayer three times .
The hydrophobic hairpin of the ColA pore-forming domain is the main determinant recognized by Cai; it was proposed that immunity protein function required helix-helix recognition within the lipid biiayer, involving the ColA hydrophobic hairpin (Geli and Lazdunski, 1992a, b). Another amphiphilic region of C0IEI, defined as a transmembrane helix of the channel in the open state, appears to be involved in the interaction between ColEI and Cei (Zhang and Cramer, 1993). This is consistent with membrane helix-helix interactions being the basis for the immunity function.
Recently, we have found that a fusion protein comprising the pore-forming domain of ColA attached to a yeast mitochondrial sorting sequence was cytotoxic for the producing cell and that the lethality was inhibited within the membrane by Cai (Espesset e(a/., 1994, accompanying paper).
In this study, we addressed the role of the small hydrophiiic loops (20 to 30 residues in length) of Cai in the protective function. These loops may interact directly with the hydrophilic loops of the membrane-spanning pore-forming domain of ColA or may influence the way in which the transmembrane helices of Cai functionally assemble. We also showed that Cai produced as two independent fragments did not assemble into a functional structure and that it was not possible to combine hydrophobic segments of two homologous immunity proteins to form a functional entity.

Overproduction of Cai confers immunity to colicin B
We have previously observed that Cai conferred partial immunity to colicin B (ColB); in contrast, Cbi is highly specific for ColB (Geli and Lazdunski, 1992a). In order to determine the regions of Cai that are involved in ColA recognition.
we performed NHgOH mutagenesis on plasmid plAI (carrying the ca/gene) and selected transfonnants insensitive to ColB (see the Experimental procedures). Three such clones were obtained whose insensitivities to ColB were con-elated with the mutated plAI piasmid. The clones were also immune to high concentrations of ColA and sensitive to ColEI. Sequencing of the plasmid DNA of each immune clone showed that all mutations were located near to or in the Pribnow box of the cai promoter ( Fig. 1 A) (Lloubes ef ai. 1986); they were predicted to increase the promoter strength (Mulligan etai, 1984). Because no antibody directed against Cai was available, it was not possible to confirm that Cai had been overproduced from the mutated variants of plAI. To determine whether cells overproducing Cai became immune to ColB, we used a Cai derivative (EpCai, encoded by plasmid pEpCai) whose production was inducibie. In EpCai, the first 12 residues of Cai are replaced with a 30-amino-acid epitope recognized by the monoclonal antibody (mAb) 1C11 (Geli etai. 1993). We then analysed the kinetics with which C600 pEpCai acquired immunity to ColA and ColB after the gene fusion had been induced ( Fig. IB): after 40 min, 100% of the cells were immune to ColA and 20% were immune to ColB; after 120min, 90% of the cells were immune to ColB (Fig. IB). By immunoblot analysis with the 1C11 mAb, we calculated that the amount of Ep-Cal required to protect the cells against ColB was 20-fold higher than that required to protect against ColA (not shown). The results indicated that cells overproducing Cai became immune to ColB and confirmed that Cai and Cbi were functionally related. B. Kinetics of acquisition of immunity to ColA and ColB after induction of EpCai production. C600 (pEpCai) was treated with mitomycin C (300ngml"'') to induce the synthesis of EpCai and assayed for colicin sensitivity at the indicated times. The percentage of survival was expressed as the ratio of the absorbance (at 600nm) of the culture incubated with ColA or ColB (1 mgml"'') to that of the control culture incubated with the dilution buffer. I, induced cells, NI. non-induced cells; ColA, percentage survival to ColA; ColB, percentage survival to ColB.

Immunity is iost when the periplasmic L3 loop is exchanged between Cai and Cbi
An Arg-136/Asp substitution in the second periplasmic loop (L3) of Cai abolishes the immunity function (Geli et ai., 1989) and the main ColA determinant for immunity recognition is located in the hydrophobic hairpin of ColA (Geli and Lazdunski, 1992a). To investigate whether L3 was involved in the specificity of colicin recognition. L3 of Cai was substituted for that of Cbi (Cai and Cbi sequences are aligned in Fig. 5 -see later). The resulting hybrid protein did not protect cells against low concentrations of either ColA or ColB ( Fig. 2 and Table 1). TaWe 1. Exchange of hydrophilic loops between Cai and Cbi and correlation with the immunity activity.
-indicates no clearing corresponding to active immunity protein, and ± indicates a turbid plaque of growth inhibition. b. Brackets indicate the replaced regions of Cai (see Fig. 2). The immunity gene and its variants were carried by a pBR derivative (plA2) in C600.
The major part of the first periplasmic loop {L1*, see Fig. 2) of Cai was aiso exchanged by the corresponding region of Cbi, but the hybrid immunity protein protected against CoiA and CoiB exactiy like the authentic Cai (Tabie 1). However, when four residues (lie-Glu-Gly-Arg) were inserted after Arg-46 of Cai (in LI), the immunity function was completely lost (not shown). Moreover, a Cai/ Cbi hybrid comprising LI * and L3 from Cbi did not confer immunity against either CoiA or CoiB ( Fig. 2 and Table 1). Therefore, although LI * was found to be sensitive to residue substitution only, L3 appeared to be totaily required in order for the immunity protein to be functionai.
The cytoplasmic L2 loop is required for immunity function L2 loops of Cai and Cbi are rich in positive charges and it has been shown that the way in which membrane proteins are topologically organized requires such positive charges (von Heijne, 1986). An Arg-92/Asp mutation in L2 of Cai was reported to result in an intermediate immunity level (Geii etai, 1989). We therefore investigated whether L2 was only involved in the topology of Cai or also had a role in the immunity function.
L2 of Cai was replaced with that of Cbi (Fig. 2). The hybrid protein protected against 100-fold less ColA than did the native Cai (Table 1). To study the topology of this hybrid protein, a tagged inducibie fusion protein (EpVBG7) was constructed: it consisted of EpCai fused to alkaline phosphatase (PhoA) in L3 (see the Experimental procedures). L2 of EpVBG7 was then replaced with that of Cbi. The PhoA activities of cells producing each protein, EpVBG7 and the modified EpVBG7, were compared: the activities were found to be identical for both proteins (not shown) and reflected the fact that L2 was able to promote the export of the PhoA moiety of EpVBG7. This identity indicated that the gross organization of the protein in the membrane was probably not affected when L2 was replaced with that of Cbi. Nevertheless, the hybhd immunity protein was unable to take up a functional structure, despite its overall topology being similar to that of the native protein.
A Cai/Cbi hybrid containing L2 and L3 of Cbi was also constructed: it had no protective activity, which was not surprising in view of the effect of the L3 substitution described above.

Separate transmembrane fragments of Cai do not iead to immunity activity
Several polytopic proteins produced as fragments are functional because the transmembrane elements reassociate within the membrane (Popot, 1993). As L2 seems to be essential in order for Cai to assemble functionally or for Cai to recognize ColA, Cai produced as two independent fragments should not be active.
A DNA sequence (DE1) containing a stop codon, a Shine-Dalgarno sequence, and a start codon was inserted in plasmid pEpCai between codons of cai corresponding to Asn-96 and Asn-97 (in L2); as a result, Cai should be expressed in two fragments: Ep-F(13-96) and F(97-178) (Fig. 3A). Cells transformed by the resulting plasmid (pEpCaiDEI) were completely sensitive to ColA after production of both fragments had been induced (Fig. 3B); this confirmed that L2 was required forthe immunity protein to be functional. To check that the fragments were expressed and membrane associated, we transferred the DE1 sequence into pEpVBG7: the modified plasmid (pEpVBG7DE1) encoded Ep-F(13-96) and a hybrid protein in which the PhoA protein was fused to the F(97-133) fragment (F(97-133)-PhoA) (Fig. 4A). The two fragments were produced; Ep-F(13-96) was found to be membrane associated but we were not able to probe its orientation by protease treatment because LI and L3 of Cai are fully resistant to extemally added protease in EDTA-treated cells (V. Geli, unpublished). PhoA activity was detected when F(97-133)-PhoA was produced (Fig. 4B); thus, at least the first transmembrane fragment of F(97-178) was correctly assembled into the membrane.

Cai function requires the integrity of its transmembrane heiices
We then tested whether a functional hybrid immunity protein could be constructed from the hydrophobic stretches of different immunity proteins. Cbi/Cai hybrids were generated by homologous recombination. Hybrid proteins resulting from homologous recombination are perfect fusions (see the Experimental procedures). After cells had been transformed with plasmids encoding recombinant immunity proteins, clones insensitive to ColA were selected: the fusion sites of all 70 selected clones were located upstream of Leu-30 of Cai (Fig. 5). All these hybrid proteins conferred levels of immunity that were higher even than those of the native Cai (Table 2), presumably because the fusion genes were under the control of the cbi promoter, which is stronger than the cai promoter. The DNA regions encoding hybrid immunity proteins of 20 clones sensitive to ColA were also sequenced: most of them resulted from frameshifting. However, three  ColA(mgmr 'f 10-^ 10"^ 10-" 10"+ 1 10-^ 10-^ 10 + + + + + + + + _ __ _ _ _ + + + + + + -+ + + -+ + + -+ + + + + + + + + + + + -----^ 10 " 10 + The colicin activity was analysed as described for Table 1. b. Brackets indicate the positions of the fusion sites in the hybrid immunity proteins (see Fig. 4).
clones carried plasmids encoding perfect hybrid immunity proteins; the fusion sites of these inactive Cbi/Cai hybrids were located in LI, H2 and L3, respectively (Fig. 5, Table 2). Only two transformants insensitive to ColB were obtained; the fusion sites were located in the extreme C-terminal part of Cbi (Fig. 5). These hybrid proteins displayed the same immunity activities as Cbi (Table 2).
These results indicated that only the first two-thirds of H1 of Cai were able to be replaced by the corresponding region of Cbi to give an active immunity protein, interestingly, immunity was conserved when the regions Cai(1-30) or LI* were exchanged, although immunity was lost when the region Cai(1-60) was exchanged (see above and Fig. 5, Table 1). This suggested that the region Cai(30-Leu-Cys-lle-Phe-Val-Val-Tyr-37) contained determinants for ColA recognition cr that this region was specifically required for interaction with another transmembrane hydrophobic helix of Cai.

Discussion
It has been previously shown that it was possible to replace the N-and C-terminal cytoplasmic ends of Cai by unrelated amino acid sequences without affecting its function (Geli etai, 1988). Various single point mutations in the polar loops of Cai, L2, and L3 decrease the immunity function (Geli et ai. 1989), whereas those of the ColEI immunity protein (Cei) tolerate a high degree of substitution Zhang and Cramer, 1993). However, the main determinant necessary for coiicins A, B, and El to be recognized by immunity proteins is located in the transmembrane regions of their pore-forming domains (Geli and Lazdunski, 1992a;Zhang and Cramer, 1993). Thus it was suggested that colicin transmembrane helices are recognized by the immunity protein within the lipid bilayer. In addition, it was proposed that the immunity proteins function independently of colicin translocation systems (Geli and Lazdunski, 1992a;Benedetti et ai. 1992;Zhang and Cramer, 1993). Lateral diffusion of the immunity proteins in the membrane would ensure rapid recognition of collcin molecules (Zhang and Cramer. 1993).
Moreover, we have recently shown that, when produced in E. coii. a chimeric protein comprising the ColA porefomiing domain fused to a mitochondrial intermembrane space sorting sequence formed a channel in the inner membrane of the cells; the channel formed independently of the Tol proteins and it was inhibited by Cai (Espesset et ai, 1994, accompanying paper). Thus, it appears that immunity proteins interact within the core of the lipid bilayer with the channel in the open state (Geli and Lazdunski, 1992a;Zhang and Cramer, 1993). However, this does not exclude a role for the polar regions of Cai.
The results reported in this study indicated that: (i) Cai overproducing cells became immune to ColB; (ii) L1* and L3 of Cai and Cbi did not carry, by themselves, the determinants for colicin recognition, although most of these loops were required for Cai function; (iii) when L2 of Cai was replaced with that of Cbi the immunity function was lost but this was not due to modified topology; (iv) Cai produced as two membrane fragments was not functional; and (v) fusion sites of functional Cbi/Cai hybrids were located upstream of Leu-30 of Cai. These results showed that although Cai and Cbi are homologous (38% identity and 39% conservative substitutions), only limited regions were able to be exchanged without affecting function.
Two roles can be proposed for the hydrophilic loops of Cai and Cbi: (i) they may interact on both sides of the cytoplasmic membrane with different parts of the colicin molecule to prevent the pore-forming helices from assembling; (ii) they may allow the transmembrane helices of Cai to assemble functionally. These two roles may or may not be mutually exclusive.
The first proposition implies that assembly of Cai into the membrane would not be perturbed if its hydrophilic loops were exchanged. When L2 of Cai was replaced with that of Cbi, the PhoA activity of the modified EpVBG7 protein was not affected; thus it seemed that the topology of Cai was not affected when its polar regions were replaced with those of Cbi. Furthermore, if the polar regions of Cai interacted with the ColA pore-forming domain in the open state, this domain would be expected to expose regions on the cytopiasmic side of the inner membrane to interact with 12 of Cai. However, studies on the gating of ColEI channels (Slatin e/a/., 1986;Raymond etai. 1986;Jakes etai, 1990;Abrams etai. 1991) and ColEI topographic studies (Song etai, 1991;Zhang and Cramer, 1992) indicate that the loops of ColEI connecting the four putative transmembrane helices on the cytoplasmic side of the inner membrane are very short and uncharged. As ColA is homologous to ColEI, it is unlikely that its cytoplasmic loops interact with L2 of Cai. Thus we favour the second proposition: the hydrophilic loops may allow the transmembrane helices of Cai to assemble functionally.
Conversely, the periplasmic loops of Cai may interact with ColA on the periplasmic side of the inner membrane. Indeed, Cai was inactivated when both L1 and L3 were replaced with those of Cbi. However, the activity of Cei is not affected even if its peripheral domains are extensively modified. This suggests that these regions are not critical for activity (Zhang and Cramer, 1993). In contrast, hydrophobic intramembrane helix-helix interactions are important for both ColA and ColEI (Geli and Lazdunski, 1992a;Zhang and Cramer, 1993).
The differences in permisslvity with regard to alterations of peripheral loops in Cai and Cei may reflect structurefunction relationships. The structures of these proteins require that: (i) the transmembrane helices assemble either into a three-a-helix bundle (Cei) or four-a-helix bundle (Cai), and (ii) these bundles should be able to interact with hydrophobic helices of coiicins. The structurai constraints imposed on helix orientation should be much less marked with a three-ot-helix bundle than with a four-a-helix bundle. Thus the bundle of Cai helices may be destabilized when its peripheral loops are modified, whereas the Cei ct-helix bundle may be more robust.
Cai produced as two membrane fragments was unable to assemble into a functional protein. Transmembrane fragments of several proteins behave as autonomous folding domains (Koster and Braun, 1990;Popot and Engelman, 1990;Bibi and Kaback, 1990;Lemmon et ai, 1992;Kahn and Engeiman, 1992;Maggio etai. 1993). These transmembrane regions may undergo tight intermolecular associations that may not be modified when the domains function. In contrast, because intramembrane hydrophobic interactions are involved in the function of immunity proteins (Geli and Lazdunski, 1992a;Zhang and Cramer, 1993), inter-helix affinities within immunity proteins may not be very high so that the helices can interact with the Immunity proteins to pore-fonning coiicins 1117 hydrophobic helical hairpins of the pore-forming coiicins. This is consistent with independent membrane fragments of Cai not being able to assemble into a functional complex and with L2 being required. Alternatively, proteins with only four transmembrane segments may have a reduced probability of assembling correctly when produced as two separate fragments.
The protective function of the Cbi/Cai hybrids indicates that the hydrophobic regions of Cai and Cbi cannot be combined although they are homologous. The results also suggest that the last seven amino acids of HI of Cai (residues 30 to 36) are specifically required for function. These data suggest that either the functional packing of Cai requires H1 to interact with another Cai transmembrane region or that H1 specifically recognizes ColA.
Immunity protein mutants able to protect against coNcin mutants which bypass the wild-type immunity protein may help better define how Cai and pore-fonning helices interact.

Hydroxylamine mutagenesis of pi A1 and selection of CoiB-immune ciones
Hydroxylamine mutagenesis was performed as described (Silhavy etai. 1984). Briefly, 5ng of purified plasmid plAI DNA was incubated for 3h at 65 C in a total volume of 500 (il containing lOOmM phosphate buffer, pH6, 1 mM EDTA, and 0.5 M NH2OH. The reaction was stopped by passing the mix through a spun column of Sephadex G50 (Pharmacia). The DNA was precipitated and resuspended in H2O at a concentration of 200 ng ml -'. C600 competent cells (0.1 ml) were transformed by electroporation with 200 ng of mutagenized plasmid. The celts were incubated for 45 min in 500 MI of LB medium and then 2(ig of purified ColB was added. Cells were incubated for 10 more min and were plated on ampicillin Luria-Bertani plates. Plasmids were isolated from selected transformants and used to transform C600 cells. The killing activity of ColB was tested on the transformants to check that the immune phenotype was plasmid-linked. Plasmids of immune clones were sequenced.

Substitution of Cai hydrophilic loops LI, L2. and L3 for the homologous loops of Cbi
The C/al restriction site of plAI was destroyed to obtain plA2. Acc\, Cla\. AflW, and Eco47lll restriction sites were created by site-directed mutagenesis in plA2 at codons 35, 56, 89. and 144 of the cat gene, respectively. Pairs of oligonucleotides (5-A TAC ATT AAT GAG CCG TAT TCA CAA GTA CTG  TAT TAC CTG TAC AAC AAA GTC GCA TTC CTG CCA T Plasmid plA2 was cut either by >iccl + C/al, AflW-t-BglW, or Hpal + Eco471ll. 0g/tl and Hpal are natural sites of plA2. Pairs of oligonucleotides were hybridized and ligated into plA2 cut with the appropriate restriction enzymes. DNA encoding the hybrid immunity proteins was sequenced. The region encoding the complete L3 of Cai was replaced with that of Cbi by site-directed mutagenesis. Arg-124 and Phe-126 were substituted for Tyr and Leu, respectively, by using the following mutagenic oligonucleotide: 5'-C TGT AGT TAA CTC GAG ATT GTA GAA GCA GAA TAG-3 .

Production of Cai as two independent fragments
A pair of oligonucleotides (5-TTA AGA ATC AGA AAA TTA ATC AAC TTA GAG GTA AAT ATG AAC AGC GAC AGG AAC ACT GTA CTT A-3\ 5-GATC AAG TAG AGT GTT CCT GTC GCT GTT CAT ATT TAC CTC TTA GTT GAT TAA TTT TCT GAT CT-3') was introduced into plA2 cut with AflW and BglW. This synthetic DNA fragment contains a stop codon followed by a Shine-Dalgarno sequence and an ATG codcn (see underlined sequences above) after the codon corresponding to Asn-96 of Cai. This DNA modification was also transferred into pEpCai and pEpVBG7.

Construction of Cbi/Cai hybrid proteins
Plasmid plAB (5 ng) carrying cbi and calm tandem was linearized with EcoRV (this restriction site is located between the two genes), treated with exonuclease Hi and mung-bean nuclease as previously described (Geli and Lazdunsto, 1992a), and used to transform strain C600. Transformants were toothpicked onto plates on which 100 |jl of purified ColA or ColB solution (20ngmr"^) had been spread. ColA-or ColB-sensitive and -insensitive clones were identified. Their DNAs were analysed by restriction mapping. Homologous recombination between cbi and cai resulted in a fixed deietion (Tomassen et ai, 1985). The hybrid genes resulting from homologous recombination were selected and sequenced.