How lipids affect the energetics of co-translational alpha helical membrane protein folding

Membrane proteins reside in a chemically complex environment. They must interact with the hydrophobic bilayer interior and the aqueous environment outside the bilayer, as well as the interface between the two. The final folded protein structure is crucial for correct protein activity, and for membrane proteins the hydrophobic bilayer is involved in the formation of this correct structure. This makes folding a complex process, with misfolding known to cause trafficking problems and also lead to disease [1]. Although the primary sequence of the protein dictates the final structure of a protein, as described by Anfinsen in 1973 [2], this has been shown to be an oversimplification. Folding is affected by many factors as well as the environment and folding can require assistance by chaperone proteins to prevent protein aggregation [3,4].

Folding is key to cellular life, and it is important to understand the molecular basis and driving forces behind this fundamental process. Membrane proteins are of high physiological importance and are significant drug targets. Their complex folding leads to problems expressing membrane proteins in vitro, making them more difficult to study. Folding studies have investigated the dependence of structure formation on the lipid environment [5–10] using full-length protein that has been expressed in cells and purified. Such work has been necessary to probe the mechanistic details including thermodynamics and how the lipid bilayer influences transmembrane (TM) helix insertion and protein folding. In vivo however, most alpha helical membrane proteins fold during chain elongation from the N- to C- terminus as they are translated by the ribosome, and they also insert into the membrane co-translationally [11–13]. This means that early helices can form tertiary interactions with each other in the bilayer while later helices are still being translated by the ribosome. Co-translational folding of alpha-helical proteins in vivo is assisted by membrane targeting and insertion apparatus. In E. coli, growing nascent chains are delivered to the SecYEG translocon by the signal recognition particle (SRP), and the ribosome continues to translate the elongating nascent chain while the translocon assists insertion of the TM helices across the bilayer [13].

Once the extending polypeptide emerges from the ribosome, all stages of membrane protein folding are affected by the lipid composition of the bilayer — association with the bilayer [9,14,15], insertion into the bilayer [15–25] and folding in the bilayer [5–7,9]. The translocon seems to be affected by the lipid composition, with its stability and function in E. coli shown to depend on the presence of cardiolipin [26]. There are thousands of different biological lipids with many different properties that can alter the properties of the bilayer (Figure 1), and in turn affect folding. Lipids can have saturated or unsaturated chains and symmetric or asymmetric chains. They can have different types of headgroups, which can be charged or neutral. Lipid headgroups also have different hydrogen-bonding capabilities [27], for example the ethanolamine headgroup in phosphatidylethanolamine (PE) lipids is able to form both intra- and inter-headgroup hydrogen bonds [28]. Bilayers can have a mixture of bilayer-forming and non-bilayer forming lipids, and these alter the lateral pressure profile across the membrane (Figure 1B). Cell membranes are composed of a mix of all these different types of lipids, altering the fluidity and thickness of the bilayer. The inner membrane of gram-negative bacteria such as E. coli contains mostly PE with phosphatidylglycerol (PG) and cardiolipin, while the cytoplasmic membrane of gram-positive bacteria contains mainly PG lipids and branched acyl chains [29]. Prokaryotes are able to alter their bilayer composition in response to things such as temperature, pH and pressure [30,31]. Different organelles in eukaryotes have different lipid compositions leading to membrane proteins experiencing different lipid environments during different stages of trafficking [10]. For example, the plasma membrane contains mostly phosphatidylcholine (PC) lipids and cholesterol, whereas the endoplasmic reticulum contains very little cholesterol but contains phosphatidylinositol [10]. Although we know the lipid compositions of various organisms and organelles, it is not well understood precisely what intermolecular forces and interactions underpin the bilayer influence on protein folding and stability.

During folding there are free energy contributions from solvation in the cytoplasm vs. in the membrane, association with the headgroup region of the lipids, inserting/partitioning of a TM helix across the bilayer, and folding once within the bilayer (Figure 2A,B). These free energies all contribute to the overall free energy landscape of membrane insertion and folding, and negative free energy contributions must balance against positive free energy contributions in order for folding to be successful. The free energy landscape includes enthalpic and entropic contributions, and measurements of thermodynamics and kinetics are used to define what the different contributions are at each stage of folding. During co-translational folding, the free energy landscape is likely to change with each amino acid translated and is challenging to measure both in vivo and in vitro. Thermodynamic and kinetic work therefore began on simpler systems such as peptides, and in vitro folding studies typically used fully translated proteins which had been unfolded in denaturant and refolded, with the change in structure and the kinetics between the two conformations measured (Figure 2B). A key advance for alpha-helical membrane protein folding in 1990 was the ‘two-stage model' [32,33], in which TM helices were deemed to be independently stable entities and insert into the bilayer in the first stage. In the second stage, these helices interact with each other in the bilayer to form the final folded structure. The study of the insertion of individual peptides is similar to the first stage, while some aspects of in vitro denaturant studies address stage two. However, this two-stage model is oversimplistic as not all helices are independently stable [34–37], and at the time there was limited detailed knowledge of co-translational insertion in vivo (as illustrated in Figure 3). Previous reviews have discussed in depth the energetics and mechanisms of membrane protein folding [34,38], and also the effect of lipids on folding [10]. This review will summarise the effects of lipids on the free energy landscape of alpha-helical membrane protein insertion and folding, using peptides and denaturant studies of full-length proteins. We will then discuss more recent work on co-translational folding, and the possible effects of the lipids on the co-translational folding free energy landscape.

Although the translocon is present in vivo to aid insertion and folding, when no translocon is present in vitro TM peptides and nascent alpha-helical membrane proteins can partition into the bilayer spontaneously [16–18,22,24,39–42] driven predominantly by favourable interactions between hydrophobic amino acid side chains and the hydrocarbon rich bilayer core. This has been taken advantage of in the use of peptides to investigate the thermodynamics governing TM integration into the bilayer (Figure 2A). Work on peptides has shown that the free energy associated with partitioning peptide bonds into the membrane interface and the bilayer core is unfavourable, and this is markedly reduced by hydrogen-bonding of the peptide backbone into an alpha-helical secondary structure [43,44]. Insertion of an unstructured peptide into the bilayer core is so energetically disfavoured that TM helices are only ever observed to insert across the membrane (or indeed exist within the membrane) when alpha-helical [34]. This also means that in vitro folding studies which measure partially unfolded states of full-length proteins are still biologically relevant, as TM helices are not expected to be fully denatured when associated with a bilayer. Even when in an alpha-helical structure, there is still a substantial free energy cost associated with the insertion of peptide bonds into the hydrophobic bilayer interior. In order to be stably inserted across a bilayer, TM domains must be rich in amino acids with hydrophobic side chains [45–47], which have favourable free energies of insertion that outweigh the energetic cost of peptide bond dehydration. Polar residues are rarely present at sites in TM helices facing the hydrocarbon bilayer core, and those which are located in such positions tend to be essential for protein function.

From work using peptides, hydrophobicity scales for TM insertion have been developed [47–49]. These scales have led to the development of prediction tools for the free energy of insertion of TM sequences [37,50]. Peptide work has also enabled the development of a biological hydrophobicity scale [51], and the position dependent contribution of each amino acid on insertion efficiency through the translocon has also been quantified [37]. It is therefore now possible to find whether there are any TM helices in a protein that are less likely to insert across a bilayer based on sequence alone.

While easier to work with than full-length membrane proteins, measuring the free energy of association and insertion of peptides experimentally is not trivial, as their hydrophobicity makes them prone to aggregation when not associated with a bilayer. Peptides are, however, particularly amenable to all-atom molecular dynamics (MD) simulations, which have proven to be a powerful tool to study the dynamic processes associated with peptide adsorption, folding, and bilayer insertion [52,53]. Recent advances in hardware and software have pushed the achievable timescales of MD simulations to the multi-microsecond range, but the processes involved in peptide insertion can take hours to observe experimentally [53]. In the specific case of membrane active peptides, it has been rigorously validated that the free energy of bilayer insertion is temperature independent over a wide range (as discussed in detail in [52]). As such, it is possible to quantitatively estimate the free energy of bilayer partitioning from high-temperature molecular dynamics (HT-MD) simulations, where transitions are accelerated. Ensemble average properties (e.g. free energies, average peptide orientation) derived from HT-MD have been shown to be in excellent agreement with those measured experimentally [52,53]. HT-MD has advanced our understanding of the molecular details of peptide insertion, as well as the properties which drive insertion such as peptide length and hydrophobicity [52]. Recent combined experimental-computational [41,42] studies have been used to show that the binding and insertion of certain small peptides is strongly dependent on the composition of the lipid bilayer. It was found that the peptides studied interacted differently with 1-palmitoyl-2-oleoyl-glycero-3-phosphocholine (POPC) and 1-palmitoyl-2-oleoyl-sn-glycero-3-phospho-(1′-rac-glycerol) (POPG) membranes and behaved differently in gel and fluid membranes. In some cases the peptides studied bound to the headgroups of either POPC or POPG membranes, but only folded into a helix in POPG (Table 1).

From these experiments on membrane active peptides, it can be inferred that the lipid bilayer can influence one of the earliest stages in folding — the interfacial partitioning of a TM into the headgroup region of the bilayer. An optimum lipid composition will therefore ensure that the initial interaction between the TM and the bilayer is favourable. This has implications for folding in vivo, as some proteins may interact less favourably with the native membrane environment, and therefore be less likely to insert across the membrane without aid from the translocon, and will perhaps be more likely to misfold.

The thermodynamics and kinetics of the folding of some full-length alpha helical membrane proteins has been measured in vitro. This information is usually gained by measuring the change in protein structure during unfolding by a denaturant, then measuring the refolding when the denaturant is removed (Figure 2A). These studies are performed on full-length protein which has been overexpressed in cells and purified, and there are many examples of such studies on folding in detergents [20,54–58] and bicelles [59,60]. The folding landscape will however be altered by the lipid composition that the protein is folding into (Figure 4). Lipid composition is also known to affect protein topology, the most dramatic and well-known example being the E. coli lactose transporter LacY [61–63]. It is known that in some cases the insertion into bilayers from a denatured state and the subsequent folding rate can be inhibited by a high lateral chain pressure (arising from using lipids such as those with PE headgroups), as it is harder to insert a fully translated protein across the membrane than it would be in lipids with a low lateral chain pressure (such as 1,2-dimyristoyl-sn-glycero-3-phosphocholine (DMPC) or 1,2-dioleoyl-sn-glycero-3-phosphocholine (DOPC)) [8,64]. Conversely, this high lateral chain pressure can lead to increased protein stability once in the bilayer [9]. In the case of LacY, 1,2-dioleoyl-sn-glycero-3-phosphoethanolamine (DOPE) and 1,2-dioleoyl-sn-glycero-3-phospho-(1′-rac-glycerol) (DOPG) are required for a high efficiency of reconstitution into liposomes from detergent and of refolding into liposomes from a urea-denatured state, but high DOPE becomes inhibitory [6] (Table 1).

Progress is starting to be made on measuring the free energy of folding in a lipid environment to specifically assess the effect of the lipid bilayer on thermodynamics. The bacterial leucine transporter (LeuT) is an neurotransmitter transport orthologue that is responsible for the transport of leucine in the hyperthermophilic bacteria Aquifex aeolicus [65]. To date, LeuT is the only alpha-helical protein to have its thermodynamic stability measured in a bilayer [5]. For these experiments, urea was used as a chemical denaturant to reversibly unfold LeuT in liposomes. By changing the lipid composition of the proteoliposomes, the effects of bilayer properties such as charge and lateral chain pressure on LeuT stability could be investigated. The stability of LeuT could be modulated by the properties of its surrounding bilayer, where addition of DOPE or DOPG to DOPC bilayers was found to increase its thermodynamic stability in liposomes [5]. This corresponds to either an increase in lateral chain pressure (DOPE) or an increase in charged headgroups (DOPG). Whilst these in vitro measurements are still far removed from what happens in the cell, they do allow key thermodynamic parameters in lipids to be attained. Kinetic information on LeuT folding and unfolding would further our understanding, as thermodynamic information can come from kinetic measurements of transition states and folding intermediates [7,9,14,60], and can help get a picture of the folding energy landscape and how it changes in response to different lipid environment.

The remainder of this review will discuss the free energy contributions of co-translational folding and how the lipids affect TM helix insertion. Investigating the insertion of TMs and the folding of all TMs separately is useful both conceptually and experimentally but has limitations. Firstly, because folding is co-translational, folding of the initial TM helices begins while later helices are still being translated by the ribosome and inserted into the bilayer (see Figure 3). TM helices are often not independently stable and unable to insert into the bilayer by themselves [34–37]. Some helices are unable to insert into the bilayer alone, but can once other helices have been translated, or can move position or flip within the bilayer once inserted [17,62,63,66–73]. Folding studies therefore need to progress to co-translational studies in order to address how folding in vivo is governed. While thermodynamic data is lacking on co-translational folding, we can use previous studies on peptides and full-length proteins to make predictions of how lipids will affect the energetics of co-translational insertion.

Early work has measured apparent free energies of TM insertion co-translationally in vitro by glycosylation assays (Figure 5A) [35,36], and in vivo using force pulling assays with stalling sequences to ascertain the force acting on TM helices during translation [12,74,75] (Figure 5B). In glycosylation assays, two glycosylation sites are engineered into the protein flanking the TM of interest. Whether the construct is singly or doubly glycosylated depends on whether the TM traversed the membrane during synthesis, and the amount of singly and doubly glycosylated protein is compared with determine the apparent free energy of membrane insertion (called ΔG_app) [35,36]. In force pulling assays, an arrest peptide is engineered after the TM of interest to induce a stall in translation. The efficiency of stalling correlates with the efficiency of TM insertion into the membrane and is used as a proxy for the pulling force acting on the nascent chain during translation [74]. Both of these assays are powerful techniques for measuring TM insertion in native membranes, and in both the insertion efficiency of a helix can be compared either when in its native protein or on its own. The glycosylation assays were used to show that TM helices can move position once inserted, and that TM helices that cannot insert alone can do so in the full-length protein [35,36]. Force pulling assays have elucidated the contacts that form between TM helices, and between the nascent chain with the translocon during translation [12,74,75]. Both assays are however ensemble endpoint assays, so there is no information gained on the parallel folding pathways taken, or on folding kinetics. Force pulling assays have yet to be applied in vitro with defined synthetic membranes to investigate the specific effects of different lipids on TM insertion and folding.

Based on the peptide work and studies on full-length proteins discussed earlier, lipids are known to affect the initial TM-bilayer association. The interfacial partitioning of a TM into the headgroup region of lipids with a high lateral chain pressure, which corresponds to a low lateral headgroup pressure, is favourable as it relieves the stored curvature stress of the bilayer [8] (Figure 4). Cell-free expression systems have been used to directly study how specific changes in bilayer properties modulate co-translational TM insertion and folding (Figure 6). In comparison with in vivo expression, cell-free synthesis has the advantage that the composition of the bilayer structures present can be more easily controlled, and as such the influence of electrostatic and mechanical properties of the bilayer on folding can be investigated. These co-translational studies have found, with a few exceptions [17,21,76], that insertion is more favourable in lipids with charged headgroups (DOPG), or in high lateral chain pressure lipids (such as DOPE) [16–18,22,24,25,40]. This likely occurs because a charged headgroup or low lateral headgroup pressure allows favourable interfacial partitioning with the bilayer, making it more likely that a TM will insert across the bilayer and fold. Based on work on full-length proteins, it may be the case that once in the bilayer high lateral chain pressure increases folding rate and stability during co-translational folding, but this has not yet been confirmed experimentally. Interestingly, the native E. coli inner membrane is composed of 70% PE lipids and ∼30% PG lipids [10], corresponding to the favourable conditions found by co-translational studies. However, PE and PG lipids are not favoured by the E. coli protein DsbB during co-translational insertion and folding [17], implying other factors at work.

The favourable interaction of an elongating nascent chain with lipid headgroups observed in vitro fits neatly with a recent model of translocon-assisted insertion in vivo. This model states that the interfacial partitioning of the TM with the lipid headgroups is the important first step for translocon-guided insertion into the membrane [34]. For a range of peptides, transfer free energies for the spontaneous transition from interfacially bound to transmembrane state have been shown to be quantitatively similar to those measured for translocon-assisted peptide insertion [77]. This suggests that for both spontaneous and translocon-assisted insertion, the peptide transitions from an interfacially bound to transmembrane state — although presumably via different pathways [78]. It is therefore reasonable to hypothesise that the properties of the bilayer interface may affect both translocon-assisted and spontaneous insertion through similar mechanisms, and thus measurements on spontaneous insertion in vitro may give clues to the effects of lipids on membrane protein insertion in vivo. Importantly, this would suggest that the free energies measured for spontaneous insertion of short peptides, or indeed of helix-by-helix co-translational insertion of membrane proteins, would be similar to those expected for translocon-assisted insertion and therefore means that computational and different experimental techniques can complement each other, and each fill in different pieces of information.

We know from previous studies on peptides and full-length proteins that lipids are very important in insertion and folding. Once a nascent chain has emerged from the ribosome exit tunnel, the lipids within the surrounding native membrane are an integral part of membrane protein folding, and therefore also directly affect the folding energy landscape. Lipids affect the initial TM-lipid interaction, where headgroup charge and low lateral headgroup pressure can promote TM association with the bilayer. Higher lateral chain pressure lipids are also known to inhibit TM insertion, but conversely also increase protein stability. We now need to understand precisely how the findings from in vitro studies translate to co-translational folding, in both spontaneous folding scenarios and in in vivo translocon-assisted folding. Some E. coli proteins favour the lipids found in E. coli inner membranes, but not all, and it is not understood why this is the case. To fully understand this, we need to build a better, higher temporal resolution understanding of co-translational folding of membrane proteins. Then we can attempt to measure the energetics of the entire system, and how the lipids affect this.

• Membrane proteins are important drug targets and their misfolding is implicated in many diseases. There are many different types of lipids in cellular membranes, and how these different lipids influence membrane protein folding is not well understood.

• Membrane proteins interact with the interfacial region of the bilayer during translation prior to insertion across the bilayer. Lipid lateral chain pressure and headgroup charge are known to affect protein insertion, protein stability and folding rates.

• Future directions should include folding studies which assess the thermodynamics and kinetics of co-translational folding in different lipids. An advance in computational methods would greatly aid these studies.

The authors declare that there are no competing interests associated with the manuscript.

N.J.H, R.A.B and P.J.B wrote the article with contributions from S.G.B, G.A.P.

We acknowledge funding from the Wellcome Trust Investigator Award 214259/Z/18/Z to PJB. We thank Dr Heather Findlay for a critical reading of this manuscript.

DMPC

1,2-dimyristoyl-sn-glycero-3-phosphocholine (14:0 PC)

DOPC

1,2-dioleoyl-sn-glycero-3-phosphocholine (18:1 (Δ9-Cis) PC)

DOPE

1,2-dioleoyl-sn-glycero-3-phosphoethanolamine (18:1 (Δ9-Cis) PE)

DOPG

1,2-dioleoyl-sn-glycero-3-phospho-(1′-rac-glycerol) (18:1 (Δ9-Cis) PG)

DPoPC

1,2-dipalmitoleoyl-sn-glycero-3-phosphocholine (16:1 (Δ9-Cis) PC)

DPoPE

1,2-dipalmitoleoyl-sn-glycero-3-phosphoethanolamine (16:1 (Δ9-Cis) PE)

HT-MD

high temperature molecular dynamics

LysoOPC

1-oleoyl-2-hydroxy-sn-glycero-3-phosphocholine (18:1 Lyso PC)

LysoPPC

1-palmitoyl-2-hydroxy-sn-glycero-3-phosphocholine (16:0 Lyso PC)

MD

molecular dynamics

PE

phosphatidylethanolamine

POPC

1-palmitoyl-2-oleoyl-glycero-3-phosphocholine (16:0-18:1 PC)

POPG

1-palmitoyl-2-oleoyl-sn-glycero-3-phospho-(1′-rac-glycerol) (16:0-18:1 PC)

SDS

sodium dodecyl Sulfate

TM

transmembrane