In vitro folding studies of outer membrane beta-barrels have been invaluable in revealing the lipid effects on folding rates and efficiencies as well as folding free energies. Here, the biophysical results are summarized, and these kinetic and thermodynamic findings are considered in terms of the requirements for folding in the context of the cellular environment. Because the periplasm lacks an external energy source the only driving forces for sorting and folding available within this compartment are binding or folding free energies and their associated rates. These values define functions for periplasmic chaperones and suggest a biophysical mechanism for the BAM complex.
1. Introduction: the challenges of unfolded outer membrane proteins
On top of the requirement to find the native conformation out of all the many other conformational possibilities of a polypeptide chain, an unfolded outer membrane protein (uOMP) must achieve this folding feat by interacting with the complex anisotropic environment of a phospholipid bilayer. The uOMP must do so in the correct transbilayer orientation with its loops presented to the extracellular environment and its turns exposed in the periplasm . Further, it must insert itself into the correct biological membrane, which means that a uOMP nascent chain must traverse through the plasma membrane without folding into it either during or after its bilayer transmittal. And the uOMP must somehow make its way approximately 180 Å across an aqueous periplasmic compartment in which it has marginal—if any—solubility . Only then, when it reaches its outer membrane destination, should the uOMP fold. Because there is no ATP in the periplasm , a uOMP appears to do all this directional sorting and folding in the absence of an external energy source once secreted through the translocon.
Precise control of the OMP sorting and folding processes is paramount, because OMPs play many important roles in bacterial cells, and some of their functions are essential [4,5]. Equally important for control, however, is the possibility that OMPs could compromise the fitness of the cell if improperly inserted into the wrong membrane. Because many OMPs function as porins , the incorporation of these proteins into the energized plasma membrane would essentially poke holes in this bilayer. Even small, atomic-sized transbilayer passageways that are large enough only for hydrogen atoms would be a problem. Leakage of protons would dissipate the proton-motive force and compromise the ability of bacteria to synthesize ATP. This would undoubtedly kill the cell. Indeed, β-barrel toxins, such as alpha-haemolysin, as well as many α-helical antimicrobial peptides use this ‘hole poking’ mechanism to kill foreign cells [6–8].
2. Sorting across the periplasm and folding into outer membranes can be thought of as a ‘kinetic push, thermodynamic pull’ mechanism
In the absence of an external energy source to promote OMP sorting away from the plasma membrane and towards the outer membrane, the only chemical potentials available as driving forces are the OMP folding stabilities and/or the uOMP–chaperone binding free energies . The thermodynamic pull in this sorting hypothesis originates from the observations of unusually high OMP folding stabilities, ΔGFold. OmpLA, PagP and OmpW all have free energies of folding into large unilamellar vesicles (LUVs) in the range −18 to −32 kcal mol−1 [9,10]. These are very large values, and an interesting thought experiment can be constructed to consider their magnitudes: if an Avogadro's number of OMPs possessing the stability of OmpLA (−32 kcal mol−1) could be envisioned to exist within a Gram-negative bacterium, at equilibrium fewer than 100 of these molecules would be unfolded. Thus, the folding free energy change dictates that the population lies far, far in favour of the folded conformation. On this basis, it can be argued that a ΔGFold of this magnitude can easily serve as an ‘energy sink’ for the sorting process, because it is the most energetically favourable and last state for OMPs in bacterial cells.
For this hypothesis to be true, it must be the case that all uOMP/chaperone binding interactions are thermodynamically less favourable than the final OMP ΔGFold. Binding experiments conducted by several independent laboratories using a wide variety of uOMP sequences and chaperone expression constructs support this idea. As summarized in figure 1, the binding free energies of uOMPs to Skp, SurA, DegP or even themselves (e.g. uOMP aggregation) are all in the range −8 to −12 kcal mol−1 [2,11–13] (figure 1). The differences between these values and the OMP folding free energies are at least 10 kcal mol−1, a value that is a thermodynamic potential easily large enough to drive uOMPs into their native conformations.
One concern that may be raised about such high OMP thermodynamic stabilities is how they are degraded upon ageing. With such a high kinetic barrier to unfolding, it would be expected that such a process would require the input of energy, which is not available in the periplasm. However, under normal cellular growth conditions, the accumulation of OMPs should not be a serious problem in a bacterial cell, because they are diluted upon bacterial growth and division as new OMPs are biosynthesized . In addition, Escherichia coli and other Gram-negative bacteria dispose of their outer membranes by blebbing off outer membrane vesicles whose composition includes outer membranes, outer membrane proteins and periplasmic components [19–21].
A second point arising from ΔGFold values of these magnitudes is that these energies also render these OMPs kinetically stable in their final folded states, because the activation free energy for unfolding must at least be equal to the folding free energy. Figure 2 shows an energy diagram illustrating this point. A kinetic stability of this magnitude predicts that the unfolding rates will be extremely slow. Stated another way, the thermodynamic stability encodes a kinetic ratchet preventing backflow of folded OMPs into unfolded conformations, thereby rendering the sorting process irreversible. Although a kinetic ratchet could be considered kinetic control over the process, its existence is entirely a consequence of the OMP thermodynamic stabilities.
The kinetic push aspect of sorting across the periplasm derives from the observation that uOMPs do not spontaneously fold into bilayers with biological lipid compositions with high efficiency in vitro in the absence of a catalyst . No configuration of natural or mimicked E. coli lipids support fast folding in vitro: neither of the forms of E. coli lipids available for purchase from Avanti support high efficiency folding; nor do the lipids purified from E. coli; nor do bilayers prepared with synthetic lipids mixed at biological mole ratios; and the addition of lipopolysaccharide has no dramatic effect on folding [14,15]. Upon discovery, this was surprising and counterintuitive. How can it be that uOMPs fail to fold into a lipid bilayer with a lipid composition essentially the same as that it encounters in vivo? Once inserted, it seems unlikely that OMPs would be unstable in such a bilayer, eliminating a thermodynamic explanation for this finding. Rather, a simple answer to this question is that spontaneous folding into such bilayers is slow. The questions then become: what is the biological rationale for such slow folding? And what is the physical basis for such slow folding?
Some answers to these questions can be rationalized by recognizing that uOMP folding is not a unimolecular reaction. In contrast to the folding of small soluble proteins,1 the membrane protein folding reaction is minimally second order, because it involves two substrates: (i) the nascent polypeptide chain and (ii) the membrane. Two independent groups using different OMP proteins have experimentally confirmed this assertion: under pseudo first-order conditions (where the lipid concentration was in excess), the dependence of the apparent folding rate was observed to be first order with respect to the client for OmpX . Similarly, the apparent folding kinetics accelerate in a manner consistent with second-order reaction when the OmpA concentration is held constant and the lipid concentration is increased . Because the reaction is second order, the concentrations and structural features of both of these reactants will affect the observed folding rates, the final structures and the thermodynamic stabilities of OMPs. The cell will undoubtedly use these abilities to modulate folding rates in a manner that enhances its fitness and survival.
Accordingly, Gessmann et al.  demonstrated that the extremely poor folding of uOMPs into native lipid bilayers in vitro can be explained by a physical control of the folding rate encoded in the lipid head groups. To demonstrate that head group chemistries could slow folding, Gessmann et al. conducted host–guest experiments in a host background of phosphatidylcholine (PC) lipids with varying mole fractions of guest phosphatidylethanolamine (PE) or phosphatidylglycerol lipids. These experiments showed that increasing mole fractions of PE dramatically slow folding for all OMPs assayed in a dose-dependent manner. By extrapolation, the inclusion of PE in biological mole fractions (greater than 75%) would inhibit OMP folding on any reasonable biological timescale. Importantly, these experiments excluded a conclusion in which the uOMP polypeptides (or their preparation) were somehow incompetent for folding because—under otherwise identical experimental conditions—the uOMPs folded quickly and efficiently into 100% PC membranes.
The apparent rate of folding in vitro is so slow using E. coli lipids (approx. hours to days) as to be disregarded as biologically meaningful. But another important comparison here is the rate at which uOMPs bind to chaperones or self-associate. Using stopped flow experiments, Zhao and co-workers  showed that the half time for uOmpC binding to either SurA or Skp was of the order of milliseconds. The binding to DegP was slightly slower, on the scale of seconds . Danoff & Fleming  showed that uOMPA171 self-aggregation had an apparent half-life for formation of the order of minutes. Thus, all known interactions in which a uOMP can participate occur orders of magnitude faster than the spontaneous folding into membranes composed of biological lipids. Unfolded OMPs are therefore kinetically partitioned onto a productive folding pathway by essentially being kinetically pushed away from the outer surface of the plasma membrane.
This kinetic partitioning sets in motion a sorting process in the direction of the outer membrane that is not currently well understood. As cartooned in figure 1, both genetic and biochemical experiments indicate that there is a functional interaction between SurA and components of the BAM complex [23–27]. Confirming this finding in vitro, Hagan et al.  demonstrated that the inclusion of SurA increased the efficiency of OmpT folding by a reconstituted BAM complex. The biological importance of Skp is less well understood. Although a Δskp surA::kan double null strain was synthetic lethal in rich media , the skp gene has not shown any genetic interactions with components of the BAM complex. Still, Skp appears to maintain uOMPs in a folding-competent, unfolded conformation because a uOMP captured by a Skp protein and subsequently presented with membranes will fold into those bilayers in vitro [11,29]. Despite the fact that these details are not well understood, it is clear that folding into outer membranes in vivo involves the BAM complex.
We have hypothesized that the main function of the BAM complex should be to reduce the activation barrier to uOMP folding at the inner surface of the outer membrane. This idea originates in the observation that the phospholipid head group composition in this membrane leaflet is similar to that found in the outer leaflet of the inner membrane: it is enriched in PE lipids, which are known to slow spontaneous folding. Therefore, to ensure that uOMPs can fold efficiently into bacterial outer membranes, a catalyst would be required in vivo. Keeping in mind that uOMP folding is minimally a second order reaction, a folding catalyst could accelerate folding by stabilizing a transition state of either the uOMP or the membrane or both. Because it is an evolutionarily conserved and essential E. coli protein, the central subunit of the BAM complex, BamA, is well positioned in cells to execute these functions.
To demonstrate that BamA could accelerate folding of client OMPs, Gessmann et al.  conducted folding experiments of client uOMPs either in the absence or in the presence of pre-folded BamA in vesicles containing PE lipids. In all cases, pre-folded BamA accelerates the folding of the client OMP. Although the BamA-enhanced initial rate increased with increasing substrate concentration (figure 3), these experiments could not definitively establish that BamA behaved according to classic enzyme kinetics. In fact, if it does function like that, the initial rate versus [substrate] plot did not display the expected shape of a square hyperbola, which implies that the Km would be high. Such a finding seems to indicate that any interactions between BamA and a client uOMP are thermodynamically quite weak.
Because uOMP folding is minimally second order, it is also worth considering how BamA may function to influence the membrane conformation. A large literature of folding studies in vitro has demonstrated that forms of the bilayer expected to have defects accelerate uOMP folding. Under otherwise identical experimental conditions, unfolded OMPs fold faster into membranes with shorter acyl chains as compared to membranes composed of longer acyl chains [14,22]. The accelerated folding in thinner membranes was attributed to an increased prevalence of bilayer defects, consistent with data showing that thinner bilayers are more permeable to solutes  and with simulations that demonstrate that thinner bilayers have higher incidences of spontaneous pore formation . The introduction of a double bond in the phospholipid acyl chain leads to increased folding when compared with the saturated form of a phospholipid with the same chain length . In identical lipid compositions, the bilayer architecture strongly influences folding rates and efficiencies as demonstrated by the observations that small unilamellar vesicles support more efficient folding of uOMPs than LUVs [14,22,32–34]. And more recently, folding acceleration was observed to be catalysed by bilayer defects created within a single lipid composition and architecture . Danoff & Fleming conducted folding of OmpA171 in DMPC LUVs from just below (20°C) to just above (26°C) the phase transition temperature. Over this minute temperature range, folding was dramatically accelerated at the lipid Tm while extremely slow both below and above this temperature . The significance of this result lies in the fact that gel and fluid phases of a lipid bilayer coexist at the Tm with a heterogeneous boundary where they meet. Membranes are well known to be more permeable to solutes and to contain defects at the Tm [36–39]. Notably, the acceleration of uOMP folding in vivo may represent a previously unrecognized role for this membrane conformation.
A physically sensible mechanism for the BAM complex would be one that capitalizes upon the physical conditions that are known to accelerate the intrinsic folding reaction for OMPs. From the large body of work described above, the creation and/or stabilization of membrane defects would be one such mechanism. Remarkably, when the structure was solved, the N. gonorrhoeae (Ng) BamA showed the induction of significant membrane defect over the course of a microsecond long, molecular dynamics simulation. Along the side of NgBamA where β-strand 1 meets β-strand 16, Noinaj et al. observed a large, 16 Å decrease in the dimyristoyl-PE (DMPE) bilayer thickness . Shown in figure 4, the bilayer thinning near NgBamA in this region is due to a gel-like to liquid-like transition in the phospholipids, which is exactly the kind of membrane structure known from experiments to accelerate uOMP folding.
The data to date therefore point to an essential function of BamA as a stabilizer of an unusual structure in the bilayer. This functional role would also explain the relatively weak interaction between BamA and a client uOMP. Because it is itself a transmembrane protein, BamA is well positioned to modulate the structure of its surrounding membrane. Moreover, if BamA shows the formidable thermodynamic stability observed for OmpLA, PagP and OmpW, it is tempting to speculate that some of its folding free energy is used to induce bilayer defects. Wu et al.  have already observed that the native asymmetric bilayer may be slightly thicker than E. coli β-barrels, so it can be imagined that lipids within the vicinity of BamA would need to distort their structures to hydrophobically match the thickness of the BamA short side. Even if this does not involve a local gel-to-liquid membrane structure, uOMP intrinsic folding kinetics clearly show that any inhomogeneity in lipid packing represents a membrane structure that accelerates uOMP folding. BamA is well suited to create them, and it is important to note that BamA stabilization of excited lipid structures is not necessarily mutually exclusive to other proposed models for how BamA functions, such as lateral gate opening  and/or templating of a budding baby barrel .
In conclusion, measurements of the folding free energies of outer membrane β-barrels have yielded insight into the energetics of how the polypeptide sequence for a membrane protein folding encodes its native structure. Not only have these data informed on the intellectually stimulating question of how sequence encodes structure, the results of these experiments have led to a sorting hypothesis in vivo that incorporates all known thermodynamic and kinetic findings for uOMP chaperone and folding reactions.
All data were previously published.
The author declares no competing interest.
The author is supported by grants from the NIH (R01 GM079440) and the NSF (MCB1412108).
We thank Prof. James C. Gumbart for sharing trajectory files from reference .
One contribution of 12 to a theme issue ‘The bacterial cell envelope’.
↵1 N.B. Soluble protein folding is not technically a unimolecular reaction either, as it involves rearrangement of water molecules. However, this point is not usually considered because the water activity is not thought to change in an experimentally observable manner under most in vitro conditions.
- Accepted June 9, 2015.
- © 2015 The Author(s)
Published by the Royal Society. All rights reserved.