Outer membrane protein folding from an energy landscape perspective

More than 50 years of work on the folding pathways of water soluble proteins has yielded a plethora of detailed insights into the conformations visited by polypeptides along complex routes to their native conformations. These include the structural and energetic properties of transition states [1–4], partially folded intermediates [5–7], and lowly-populated ‘invisible’ states [8]. The role of molecular chaperones in assisting folding and preventing aggregation has also been studied extensively [9, 10], and powerful biophysical and structural methods are beginning to reveal how water soluble proteins fold in the crowded cellular environment [11], including during their synthesis on actively translating ribosomes [12–14]. By contrast, understanding the folding energy landscapes of membrane proteins (MPs) has lagged behind those of water soluble proteins, despite recent important progress. Here, we focus on outer membrane proteins (OMPs) from Gram-negative bacteria (for discussions of α-helical MPs see [15–22], and elsewhere in this issue). OMP assembly in vivo is complicated by the requirement to fold in an asymmetric lipid bilayer, as well as the need to cross the inner membrane and periplasmic space following synthesis in the cytosol. We highlight how these requirements impose constraints on the evolution of OMP sequences and how this influences the thermodynamics and kinetics of OMP folding. Finally, we discuss how cellular proteins may sculpt the folding energy landscape of OMPs to increase the rate and/or efficiency of folding and assembly in vivo.

For proteins to fold stably into the hydrophobic environment of a biological membrane their structure must fulfil a number of energetic requirements including: (1) the hydrogen bonding potential of their polar backbone carbonyl and NH groups must be mostly satisfied [23] to offset the energetic cost of peptide bond burial (~ 1.2 kcal/mol [24]); and (2) the amino acid side chain groups in contact with the acyl chains of the lipid bilayer must be predominantly hydrophobic. The secondary structures of α-helical and β-barrel membrane proteins allow them to meet these requirements in different ways [25, 26]. The residues within each helix in α-helical MPs make backbone hydrogen bonds, allowing the separate insertion of helices into the membrane bilayer which can subsequently associate laterally to form their final native structure [27]. By contrast, β-barrel OMPs form cylindrical structures by making hydrogen bonds between residues in different β-strands, potentially far from each other in sequence [28].

Despite their sometimes complex topology, OMPs are able to fold spontaneously in vitro from their denatured states in urea or guanidinium chloride into detergent micelles or lipid bilayers [29–33] in the absence of cellular factors. This observation is consistent with Anfinsen’s findings for the water soluble protein ribonuclease A that all the information for folding is contained within the amino acid sequence [34]. OMPs are structurally and functionally diverse [35, 36], with those of known structure containing 8–26 membrane-embedded β-strands in their native state [35, 37] (Fig. 1). Larger β-barrel proteins can be formed by the assembly of monomeric subunits, such as CsgG (9 subunits, 36 β-strands) [38, 39] and GspD (15 subunits, at least 60 β-strands) [40, 41]. The OM of Gram-negative organisms is an asymmetric bilayer consisting of inner and outer leaflets formed from phospholipid and lipopolysaccharide (LPS), respectively, densely packed with proteins (protein:phospholipid:LPS ratio of 5:1:1 (w/w) [42, 43]). The requirement for OMPs to fold into this crowded membrane leads to significant pressures on the evolution of OMP sequences. These constraints can be placed broadly into two categories: (i) sequence requirements for OMPs to fold to a stable and functional native state in the OM (Fig. 2); and (ii) sequence requirements for assisted OMP assembly in vivo (see below). A major constraint is the need for solvation of hydrophobic amino acid side chains in the membrane; hence, the lipid-facing residues in OMPs must be overwhelmingly hydrophobic (Fig. 2(i)) [44], providing the drive for spontaneous membrane insertion [45]. This is reminiscent of the manner by which the hydrophobic effect provides an entropic drive towards the folding of water soluble proteins [46, 47]. Mutations which decrease the hydrophobicity of lipid-facing residues can reduce the kinetics of OMP folding in vivo, leading to premature degradation by the periplasmic protease DegP [48]. Interestingly, the eight-stranded OmpA is able to tolerate substitution of approximately two-thirds of its lipid-facing residues to other hydrophobic residues without losing its ability to fold or to be assembled by in vivo cellular machinery [49].

Recent breakthroughs have been made in measuring the thermodynamic stabilities of OMPs, with data now available for the eight-stranded proteins OmpA [50–54], PagP [55–57] and OmpW [56] and the 12-stranded OmpLA [58] (Fig. 1). These experiments have shown that OMPs are highly stable, with ΔG°_UN values ranging from −3.4 to −32.5 kcal/mol [16, 50, 56], consistent with their common resistance to denaturation by SDS [59–61]. Available evidence suggests that ΔG°_UN is correlated with the water-to-bilayer transition free energy of the residues that are in contact with the bilayer [56], calculated using the Moon-Fleming hydrophobicity scale [62]. The high stability of OMPs in the OM, compared with their lower free energy of binding to periplasmic chaperones (−7 to −13 kcal/mol [56, 63–65]), may serve as a sorting mechanism for OMPs in the periplasm [56], with the free energy of folding providing the driving force for assembly into the OM (note that the periplasm lacks ATP) [66, 67]. So, by contrast with water soluble proteins, which when folded are often only marginally stable [68], the balance between stability and function is less important for OMPs because of their high thermodynamic stability and high energetic barriers to unfolding [69, 70]. Such features should aid the evolution of new functionality since destabilising mutations that enhance a new function can be readily tolerated [71]. An interesting example of the trade-off between activity and stability has been suggested for the acyltransferase PagP: PagP from Salmonella typhimurium is twofold less stable than that from Escherichia coli in dodecylphosphocholine (DPC) micelles (−6.5 kcal/mol and −10.5 kcal/mol, respectively), but has 15–20-fold greater catalytic activity [72]. Interesting examples also exist where the stability gained by completing the hydrogen bonding between β-strands in the β-barrel has been sacrificed for functional requirements. This is the case for the β-barrels of PagP [73], FadL [74] and OmpW [75], whose structures suggest they may undergo opening movements to allow lateral entrance of substrates. Structures of BamA, the major subunit of the BAM complex (see below), revealed incomplete hydrogen bonding between the N- and C-terminal β-strands (β1 and β16) [76–79]. This results in frustration in the native protein, enhanced dynamics in the frustrated residues and increased ruggedness in the folding energy landscape close to the native state [80], the significance of which is not yet fully understood.

OMPs must also tailor their sequences to the local chemical characteristics of the OM. This results in different frequencies of residues at different membrane depths [81]. In particular, the positioning of aromatic residues in native OMP structures shows a strong preference for the membrane–water interface (‘aromatic girdle’) (Fig. 2(ix)) [81, 82], where they contribute to OMP stability [32, 57, 83]. Charged residues are also favoured at the chemically complex interfacial regions [81], with OMPs obeying a positive-outside rule (i.e. positively charged residues are located predominantly on the extracellular surface of the outer membrane; Fig. 2(vi)) [84], in contrast to the ‘positive-inside’ rule observed for α-helical MPs (i.e. basic residues are enriched on the cytoplasmic side of the membrane) [85]. This patterning of charges may help ensure correct orientation of OMPs in the OM. Burial of different residues in the bilayer makes different contributions to OMP stability, and recent evidence suggests that the free energy of partitioning of side chains into membranes is also dependent on the β-barrel scaffold [62, 86]. Computational analysis of the transfer free energies of residues at different membrane depths suggests that OMP orientation within the bilayer is also influenced by the effect of residue position on stability in the asymmetric OM [45]. Consistent with this finding, specific LPS binding sites have been identified for E. coli FhuA (Fig. 2(v)) [87], OprH from Pseudomonas aeruginosa [88] and E. coli OmpF, with the latter being shown to be important for biogenesis [89]. Lipid–protein interactions are also important for hydrophobic matching (Fig. 2(ii)). Indeed, the average hydrophobic thickness of OMPs, measured by the distance between aromatic girdles in OMP structures (23.7 ± 1.3 Å) [90], closely mirrors that of the OM in simulations [91]. The requirement for specific interactions between neighbouring β-strands also constrains OMP sequences. Glycine–aromatic inter-strand pairings are often found between neighbouring β-strands in water soluble proteins [92], and are also common in native OMP structures [84]. These may be important for OMP stability [84, 92] and folding [93], as well as having possible functional roles, as shown for autotransporter assembly [94]. Similarly to water soluble proteins, specific residues may be conserved since they form stabilising quaternary interactions (Fig. 2(iv)), or are required for enzymatic activity (Fig. 2(vii)) or barrel dynamics (Fig. 2(iii)) [95].

Since the 1980s, experiments, simulations and theory have led to the view that funnel-shaped energy landscapes best represent the mechanisms of protein folding (Fig. 3), with the depth and width of wells in the landscape corresponding to the energy and conformational entropy, respectively [96–98]. The tension between the covalent connection between amino acid residues and the drive to minimise the contact free energy between each atom leads to frustration in the landscape [99]. Frustrated regions in proteins lead to increased ruggedness in the folding energy landscape, and hence kinetically trapped intermediate states become populated. Over 25 years of work on the folding of OMPs from denaturants into lipid membranes or detergent micelles in vitro has shown that OMPs fold on complex, rugged energy landscapes [29, 30, 100–102]. Early work on OmpA folding in vitro suggested a sequential pathway involving rapid formation of a collapsed state, followed by membrane absorption and possible partial membrane insertion of β-hairpins [103, 104]. OMP folding may, however, be more complex than a simple sequential route. For example, parallel folding pathways have also been proposed for OmpA [53, 105], FomA [69, 106] and PagP [107], dependent on the folding conditions, such as the pH, the nature of the lipid/detergent and the lipid:protein ratio. Recent, global analysis of the folding kinetics of tOmpA (the transmembrane domain of OmpA) into bilayers of different thicknesses, monitored by circular dichroism (CD) and SDS-PAGE, suggested a sequential model, with detours to off-pathway, misfolded states [108] (named a ‘predetermined pathway with optional errors’ [109]). This study revealed intermediates with a β-sheet content higher than that of the native state, which the authors propose may be due to transient β-strand formation by the extracellular loops on passage across the membrane [108]. Consistent with this view, recent in vivo experiments have shown that the assembly of BamA requires an extracellular loop (L6) to be buried within the newly forming β-barrel domain [110]. A model in which hydrophilic loops are tucked into folding OMP barrels to assist them across the hydrophobic membrane could also explain how lipids are prevented from entering the barrel lumen [110].

Available experimental evidence suggests a concerted mechanism for OMP folding in vitro, in which the final folding step and membrane insertion occur concurrently. Fluorescence quenching experiments showed that all four β-hairpins of the OmpA barrel cross the bilayer simultaneously [111], and similar kinetics were observed for the formation of OmpA secondary and tertiary structure (observed by CD and cold SDS-PAGE, respectively) on folding into diC_12:0PC (DLPC) liposomes [112]. Consistent with this, at least partial structure is formed in all eight β-strands of PagP in its folding transition state [55]. Further, hydrogen-deuterium exchange (HDX) experiments, which monitored OmpX folding into detergent micelles, found that the rate of hydrogen bond formation was the same between all β-strands and synchronised with tertiary structure formation [113].

OMPs can also self-associate in their unfolded aqueous states [114], and can populate folded and unfolded dimers and trimers during folding experiments [115], as well as species with higher apparent molecular weights [108], adding further complexity to experiments tracking OMP folding mechanisms. The requirement for OMP sequences to contain alternating polar and non-polar residues that have a high propensity to form ordered aggregates [116] favours the formation of off-pathway, misfolded states. Indeed, similarly to the α-helical MP LacY [117], OmpA has been shown to form amyloid-like fibres in vitro in the absence of chaperones [118]. As fibrillar protein species can be associated with cellular toxicity [119, 120], this highlights the importance of chaperones in preventing aggregate formation in OMP folding (Fig. 3).

In vitro studies have highlighted the importance of the membrane environment on the kinetics of folding and insertion of OMPs [101, 121]. Bilayers that are more fluid, thinner, contain more unsaturated chains, and have increased curvature stress enhance OMP folding rates and yields [30, 33, 112, 121–124]. Bilayer properties also affect OMP thermodynamic stability; one study found that increasing curvature stress, by substituting C_16:0C_18:1PC (POPC) lipids for C_16:0C_18:1PE (POPE) in liposomes formed from POPC containing a 7.5% mole fraction of C_16:0C_18:1PG (POPG), increased the stability of OmpA. Conversely, substitution of POPC for shorter chain PC lipids (such as diC_10:0PC) decreased stability [50]. In vitro studies have also begun to explore the influence of chaperones [105, 125, 126] and BAM protein components [127–132] on OMP folding in vitro. One key result from these studies is that when E. coli polar native lipids are used to create liposomes, there is a requirement for cellular folding factors to assist OMP folding [33, 127, 128], rationalising the conservation of these folding factors across bacterial species [133–135].

Great progress has been made in our understanding of how OMPs fold in vitro into bilayers formed from different lipids. Similarly, impressive breakthroughs have been made in understanding the factors required for OMP assembly in vivo. How the pathways of OMP folding in vitro compare with those in vivo, however, remains unclear, partly due to a lack of studies that use direct biophysical measurements of OMP folding into physiologically relevant membranes to complement powerful, but indirect, in vivo methods. In vivo, chaperones and the BAM insertase are required to sculpt the OMP folding energy landscape to ensure rapid and efficient folding into the crowded OM. Several models for the mechanism of BAM catalysis of OMP folding have been proposed. All involve membrane destabilisation close to the BamA barrel, with some implicating a more direct role of BamA itself in the formation and stabilisation of OMP folding intermediates. Recent progress in structural analyses of BAM in different conformational states, combined with the development of biochemical and biophysical methods able to track the progress of OMP folding [221], promise to cast new light on the mechanisms of OMP folding, and how OMPs are recognised by their chaperones and released for folding by BAM. Such knowledge will aid the task of bringing our understanding of the folding mechanisms of OMPs to the level of detail achieved for water soluble proteins. It will also allow realistic models of the energy landscapes of OMP folding to be drawn, including how the landscape is modulated by the asymmetric OM and the BAM complex. Combined with impressive achievements in genetic and cellular studies of OMP assembly, there is no doubt that exciting discoveries will be reported in the years ahead, fuelled by the incentive that such knowledge may lead to the generation of much-needed new anti-bacterial agents that target OMP biogenesis.