In vitro folding and oligomerization of a membrane protein. Transition of bacterial porin from random coil to native conformation.

Porin, a channel-forming protein spanning bacterial outer membranes, was denatured in 6 M guanidinium hydrochloride or, alternatively, in sodium dodecyl sulfate at 95 degrees C. Circular dichroism spectra revealed that this protein, which in its native state consist of beta-pleated sheets as the sole detectable secondary structure, is transformed into random coil configuration in the chaotropic agent, or into alpha-helical structure in the detergent. From either state, the mature protein refolds in presence of amphiphilic molecules, attaining full structural and functional competence. As structural criteria, the native trimeric state was assayed by analytical ultracentrifugation, gel electrophoresis in sodium dodecyl sulfate, protease resistance, and circular dichroism spectroscopy. Channel formation in planar lipid bilayers reveals that the refolded protein is also functionally competent. It is concluded that the information required for the complete folding of porin is contained within the primary sequence of the mature polypeptide. The study of rapid refolding clearly reveals that this process occurs in the time range of seconds and that preexisting bilayers are not a prerequisite.

Porin, a channel-forming protein spanning bacterial outer membranes, was denatured in 6 M guanidinium hydrochloride or, alternatively, in sodium dodecyl sulfate at 95 "C. Circular dichroism spectra revealed that this protein, which in its native state consist of j% pleated sheets as the sole detectable secondary structure, is transformed into random coil configuration in the chaotropic agent, or into a-helical structure in the detergent.
From either state, the mature protein refolds in presence of amphiphilic molecules, attaining full structural and functional competence.
As structural criteria, the native trimeric state was assayed by analytical ultracentrifugation, gel electrophoresis in sodium dodecyl sulfate, protease resistance, and circular dichroism spectroscopy. Channel formation in planar lipid bilayers reveals that the refolded protein is also functionally competent.
It is concluded that the information required for the complete folding of porin is contained within the primary sequence of the mature polypeptide.
The study of rapid refolding clearly reveals that this process occurs in the time range of seconds and that preexisting bilayers are not a prerequisite.
Secretion of proteins across and their insertion into membranes have been investigated intensively over the past 10 years or so (for a review see Wickner, 1989;Saier et al., 1989;Verner and Schatz, 1988). Several mechanistic models have been proposed, some of which implicate additional proteinaceous components. In vivo studies as well as in vitro reconstituted systems have indicated that translocation requires partially unfolded precursors, with attached signal sequences (Park et al., 1988). Chaperone proteins (Hemmingsen et al., 1988) such as heat shock proteins (Pehlman, 1989), SecB (Collier et al., 19881, GroEL (Bochkareva et al., 1988), or "trigger" factor (Crooke et uZ., 1988) have been suggested to maintain precursors in a translocation and folding-competent state. These results are clearly distinct from early studies with soluble proteins (Anfinsen et al., 1961), which showed that the information contained in the protein sequence was suffi- cient for proper folding. In the work presented here, we have addressed the question whether an integral membrane protein, such as porin spanning Escherichiu coli outer membrane (Rosenbusch, 1974;Sass et al., 1989), can be folded from a random coil configuration into its native conformation, with all of its required functional and structural properties.

MATERIALS AND METHODS
Purification and Characterization-Porin was obtained from E. coli BE enveloDes.
bv extraction with octvl-POE.' as described Dreviouslv (Garavito-a&l gosenbusch, 1986). LIpopolysacharrides weie purified from the same bacterial strain (Biihler, 1986). For proteolytic digestion, trypsin (Sigma) was added to a final concentration of 10 mg/ ml. Incubations were allowed to proceed for 30 min at 37 "C and were stopped by the addition of soybean trypsin inhibitor (Merck) in 3fold molar excess. Digests were analyzed by SDS-gel electrophoresis. The oligomeric state was assessed by sedimentation equilibrium in a model E analytical ultracentrifuge (Beckman) using an An-J rotor. The secondary structure was deduced from circular dichroism spectroscopy (Cary 61 spectrophotopolarimeter) and ellepticity determined as described by Adler et al. (1973). Activity of the protein was assayed by electrical conductance measurements as described previously (Schindler and Rosenbusch, 1978 (Garavito and Rosenbusch, 1986) and used for structural tests (see above). For functional tests, vesicles were used and incorporated into bilayers as described (Dargent et al., 1987). ical of P-pleated sheet structures (Greenfield and Fassman, 1969) and consistent with previous observations (Rosenbusch, 1974;Markovic-Housley and Garavito, 1986). Treatment with a chaotropic agent such as guanidinium hydrochloride led to a random coil configuration as indicated by the absence of secondary structure. After boiling in SDS, two minima at 220 and 209 nm indicating a-helical structure formation were observed. Their ellipticity was weak ( [0]220 = -5220 and [f&9 = -7170 degrees+cm'.dmol-') as previously observed upon treatment of soluble proteins with SDS (Reynolds and Tanford, 1970b).

Denuturution
The folding state of porin was examined further using its typical migration patterns in SDS-polyacrylamide gel electrophoresis (Fig. 2). In 1% SDS, at temperatures below 75 "C, the protein migrated as undissociated trimers (Fig. 2A, lane a) with a mobility which for SDS-denatured proteins corresponds to a mass of about 95 kDa. If samples were heated to 95 "C for 1 min, the oligomers dissociated and migrated in the position typical for a denatured polypeptide of 37 kDa. Complete removal of endogenous lipids (phospholipids and lipopolysaccharides) was verified by the total absence of radio-activity in preparation purified from cells grown in presence of [32P]phosphate (Buhler, 1986). If the protein was reconstituted with added lipopolysaccharides, a ladder consisting of several bands exhibiting slightly lower mobility was observed (lane b). This is indicative of tightly associated glycolipids (Garavito et al., 1983;Buhler, 1986). If native trimers were treated with trypsin (lane c), no decrease in intensity of the band corresponding to the protein was detected, confirming that native porin is completely protease-resistant (Schindler and Rosenbusch, 1984). When protein denatured in guanidinium hydrochloride was applied to SDS-gel electrophoresis after removal of the chaotropic agent by extensive dialysis (Fig. 2B), the protein migrated as monomer regardless whether it was heat-treated or not. Treatment of this material with trypsin (lane c) completely degraded the polypeptide to small fragments that could not be detected in SDS gels. Lipopolysaccharide binding to the protein was no longer observed (lane b).
Refolding of Porin-Detergent was removed from the solution containing protein and phospholipids by extensive dialysis (4-72 h), yielding a turbid solution of vesicles. Gel electrophoresis of this material, resolubilized in SDS sample buffer, produced the patterns shown in Fig. 2C. Samples not heated above 75 "C yielded two bands (Fig. 2C, lane a). Quantification of the two bands indicated that, on average, 40% of the material was in the trimeric state. Recoveries up to 80% have been observed. More significant, this value never fell below 20%. The assignments of the native trimeric conformation was validated by the following criteria. First, the mobility of the band was the same as that observed with trimeric porin (Fig. 2A, lane a). When lipopolysaccharides were added, the characteristic bands with mobility slightly lower than that of pure protein were clearly visible (compare lanes b in Fig. 2, A and C). Moreover, it can be seen that the lower band (denatured protein) was fully sensitive to trypsin, whereas the upper one was not (lane c). Protease resistance of the refolded porin was also tested with other several proteases, all yielding the same result (not shown).
This protease resistance provided a simple method for the purification of trimers. Refolded porin was treated sequentially with trypsin and trypsin inhibitor and then chromatographed on the same anion exchange column as used for the purification in the presence of the detergent octyl-POE. Thus, removal of fragments, protease and inhibitor was achieved simultaneously with delipidation. Circular dichroism spectroscopy of this purified material (Fig. 1) yielded a spectrum that was typical of ,&pleated sheet structure and similar to that of undenatured porin. Analysis of the same preparation by sedimentation equilibrium analysis yielded a mass of 109 + 4 kDa, characteristic of trimers and indistinguishable from measurements of nondenatured porin (Rosenbusch, 1974). Finally, the purified, renatured trimers formed channels in reconstituted planar bilayers, with activation occurring in single steps and a conductance of 0.8 nS in 1 M KCl, as observed with porin extracted from E. coli membranes (Schindler and Rosenbusch, 1978).
In the standard procedure used for refolding, the precipitated porin was resolubilized in SDS after guanidinium hydrochloride had been removed by extensive dialysis. No resolubilization or refolding was observed in octyl-POE alone. We therefore tested the SDS properties that are essential for the recovery of native state. If nonionic or zwitterionic detergents were used, reproducible renaturation was observed but the yield was consistently low (lo-20% After mixing of denatured protein in 2% SDS with phospholipid-detergent mixed micelles, samples were withdrawn and treated sequentially with trypsin and trypsin inhibitor as described. To the sample at time 0, trypsin was added first, followed after a few seconds by the phospholipid-detergent solution. Positions of standard molecular mass marker proteins are shown on the left. recoveries comparable to those in SDS. Both sulfate and sulfonate detergents were less effective if their alkyl groups were octyl rather than dodecyl chains. When the short chain phospholipid diheptanoyl phosphatidylcholine (Eisele and Rosenbusch, 1989) was used, as the single amphiphile to substitute for both octyl-POE and soy bean lecithin, no refolding was observed.
Rapid refolding experiments (Fig. 3), involving dilution without dialysis, showed that renaturation occurred as soon as the protein was exposed to the phospholipids present in mixed micelles with detergent. Analysis of aliquots removed 30 s after mixing revealed the presence of protease resistant trimers, whereas digestion of denatured material occurred within seconds, as shown by the sample at time 0 (Fig. 3). Variations in the octyl-POE concentration in the refolding mixtures showed that high detergent concentrations (>5%) slowed down the kinetics and final yields of refolding to less than 10% after 4 h.

DISCUSSION
Porin is fully denatured either in guanidinium hydrochloride or in SDS. In the chaotropic agent, the protein dissociates quantitatively into its constituent polypeptide chains which, according to circular dichroism spectra, exist in random coil configuration.
Due to unfavorable signal to noise ratios at wavelengths shorter than 205 nm, the presence of less than 5% a-helical configuration cannot be ruled out. Since no indication of even very limited quantities of a-helix is observed in the native protein, it seems highly unlikely that a small segment of the polypeptide could assume such a configuration or retain such secondary structure in guanidinium hydrochloride.
In SDS, less than 1% of incompletely denatured protein would be detected, due to the sensitivity of the gel system and the resistance of the undenatured protein to enzymatic hydrolysis. Moreover, as recoveries of refolded protein reach values as high as 80%, the probability of porin molecules escaping denaturation seems negligible. Refolding, either by rapid dilution or after extensive dialysis, was demonstrated to be complete by the full recovery of the native structure of porin and of its channel conductance. This renaturation requires an amphiphile, preferentially SDS, although others such as zwitterionic detergents yielded significant renaturation. Whether these amphiphiles are critical in the formation of intermediates cannot presently be determined.
Earlier experiments (Dargent et al., 1987) showed that amphiphiles are required for the insertion of native porin into preexisting membrane (planar bilayer), although concen-trations well below their critical micelle concentration were sufficient.
Membrane protein folding and insertion might therefore be viewed to occur in two steps: folding to a conformation close to the native state, and its subsequent insertion into the lipid bilayer, the latter being aided by amphiphiles. This proposal seems compatible with recent evidence obtained in uivo with lactose permease (Roepe and Kaback, 1989). Furthermore, earlier attempts to refold porin in the absence of amphiphiles (Rosenbusch, 1974;Markovic-Housley and Garavito, 1986) have shown that, although the p-sheet structure was recovered, no other characteristics of native porin were present.
How could the pathway of the folding and insertion of porin into the membrane be visualized? Bacteriorhodopsin has been shown to reassume its native structure and functional properties after complete unfolding (Huang et al., 1981). In that case, insertion was proposed to occur by the partitioning of overall hydrophobic helical hairpins, or corresponding helical domains (Popot et al., 1987), a concept of micro-assembly in accordance with the proposal advanced by Engelman and Steitz in 1981. Though this may be a valid hypothesis for bacteriorhodopsin, it seems unlikely to account for membrane insertion of porin, with its many polar residues dispersed over the entire sequence (Paul and Rosenbusch, 1985). Even if minor a-helical domains that would have escaped detection were present, they alone could hardly account for membrane insertion of the entire protein. As in P-pleated sheets, the hydrogen bonds occur between, rather than within peptide strands, at least four segments would be necessary for thermodynamically favorable insertion (Tanford and Reynolds, 1976), a reasoning which does not even include the large number of polar side chains in porin.
By analogy to the situation in situ, preexisting membranes have been tacitly assumed to be a prerequisite for membrane protein assembly. To address the question of whether such an assumption was justified, we have allowed refolding of porin with phospholipids in presence of detergent in excess of its critical micelle concentra$on.
The presence or absence of structures larger than 50 A, typical of mixed micelles (Zulauf and Rosenbusch, 1983), was ascertained by quasi-elastic light scattering studies. As no structures larger than that size were found in conditions under which refolding occurred with significant yields, the presence of membranes seems not to be required for the proper folding of porin.
Are such observations limited to porin only, or can they be extended to other proteins such as membrane receptors? It may be provocative to investigate the applicability of spontaneous folding to other membrane proteins at a time when overexpression of receptors appears potentially so fecund. Irrespective of such considerations, our results clearly show that all the information necessary for the proper folding of the protein to its native state is contained within the mature segment of the polypeptide and that the signal peptide, as well as other components, may affect the targeting of the proteins and the kinetics of their folding process.