Analysis and de novo design of membrane-interactive peptides

Abstract

Membrane–peptide interactions play critical roles in many cellular and organismic functions, including protection from infection, remodeling of membranes, signaling, and ion transport. Peptides interact with membranes in a variety of ways: some associate with membrane surfaces in either intrinsically disordered conformations or well-defined secondary structures. Peptides with sufficient hydrophobicity can also insert vertically as transmembrane monomers, and many associate further into membrane-spanning helical bundles. Indeed, some peptides progress through each of these stages in the process of forming oligomeric bundles. In each case, the structure of the peptide and the membrane represent a delicate balance between peptide–membrane and peptide–peptide interactions. We will review this literature from the perspective of several biologically important systems, including antimicrobial peptides and their mimics, α-synuclein, receptor tyrosine kinases, and ion channels. We also discuss the use of de novo design to construct models to test our understanding of the underlying principles and to provide useful leads for pharmaceutical intervention of diseases.

Peptide–membrane interactions play important roles in all organisms and viruses, performing a diverse array of functions. To name just a few, antimicrobial peptides (AMPs) and certain classes of antibiotics disrupt the membranes of their targets, apolipoproteins stabilize lipoprotein particles and bar-code them for internalization and processing, fusion peptides promote vesicle budding and fusion, and ion channel-forming peptides are found in both antibiotics and viruses. Indeed, membrane-interactive peptides comprise such a fundamental facet of the functional fabric of living systems that it is difficult to overstate the importance of improving our understanding of how their structures and dynamics lead to their remarkable properties. Here, we categorize distinct conformational themes and functional mechanisms of peptide–membrane interactions, drawing from peptides with important biological functions. We also discuss the use of de novo design to construct peptides that test hypotheses concerning the structures and functions of peptides. We will also discuss one example of the translation of an AMP mimetic to the clinic. The literature of peptide–membrane interactions is so immense that any review will necessarily be incomplete. To provide a modicum of focus, we will primarily limit our scope to ribosomally expressed peptides and proteins with native peptide backbones, rather than those synthesized via synthetases or extensive post-translational modification. We apologize in advance for drawing disproportionately from examples of work from our lab; they have been taken to illustrate principles rather than to claim any priority over many other contributions to this vast literature.

We first must define what a “peptide” is in this context. We will adopt the definition that membrane peptides are relatively short sequences (mostly < 50 residues) that form autonomous units capable of interacting with membranes. In the case of larger proteins, we will consider a membrane-interactive domain as a “peptide” if it preserves the membrane-interacting function of the larger protein when studied as an isolated peptide. For example, fusion peptides of viral fusion proteins would be considered as a class of membrane-interacting peptides. Also, the protein α-synuclein, which has a large number of tandemly repeated membrane-binding units, will be included in our discussion.

Intrinsically disordered membrane peptides.

Membrane-interacting peptides can be subdivided based on their conformations and localization in the membrane (Figure 1). One prominent class of membrane peptides are “natively disordered” in solution and on membranes. They interact with the membrane surfaces through highly dynamic electrostatic and hydrophobic interactions. Frequently, the peptides are rich in cationic residues, Lys and Arg, which are attracted to anionic groups in the headgroups of phospholipids, as well as aromatic residues that dip deeper to the acyl glycerol region of phospholipids.^1–3 Such conformational states have been identified as discrete kinetic intermediates in the process of insertion of helical proteins into membranes.⁴ In other cases, this dynamic disordered conformation represents the native mode of interaction observed at equilibrium. For example, a highly positively charged 25-residue peptide domain tethers the MARCKS protein to negatively charged lipids,⁵ and similar motifs are found in GTPases such as RAS, which helps localize the protein to phases rich in phosphatidylinositol lipids. It is likely that some non-helical AMPs similarly interact with membranes.⁶ Intrinsically disordered proteins can also form networks that interact with lipid bilayers through liquid–liquid phase separation to remodel membranes.^{7, 8}

Fig. 1: Illustrations of the different ways peptides can associate with membranes.

Peptides can interact with membranes in their (A) intrinsically disordered conformations or through formation of secondary structure. (B) Amphiphilic helices, which have hydrophobic and hydrophilic faces, bind to membrane surfaces such that the helix lies parallel to the interface. Peptide-membrane interactions can influence the insertion of transmembrane peptides (C) and their assembly into helical bundles (D).

The amphiphilic helix.

The amphiphilic helix is a ubiquitous motif that reprises frequently in the structures of membrane-interactive peptides. In this motif, hydrophobic and polar residues segregate onto opposing faces of the helix, imparting amphiphilic character ideally suited for binding to apolar/water interfaces such as those found at membrane surfaces. Such helices are also often termed “amphipathic helices”. However, following Tanford,⁹ we prefer the more euphonious and friendly term amphiphilic, etymologically stemming from “loving both” as opposed to “suffering both” when traced to the Greek roots.

As the apolipoproteins were first sequenced their potential to form amphiphilic helices became clear from physical models or when their sequences were drawn on helical wheels (a popular way to appreciate the orientation of sidechains when viewed axially down a helix).^{10, 11} The distribution of residues was striking, with apolar residues lining one face of the helix, flanked by Asp, Glu, Lys and Arg, arranged so that the positive charges flank the hydrophobic sector, and the acidic residues are furthest from the apolar region. This configuration was expected to foster a close orientation of the zwitterionic and acidic head groups in ion pairs, and the interaction of the apolar residues with the acyl-glycerols and acyl chains of phospholipids. Segrest and coworkers demonstrated the validity of this hypothesis through the design of peptides representing single repeats, as well as consensus sequences.^12–16 Depending on the apolipoprotein, such peptides bound to membrane surfaces, and also were able to stabilize disk-like assemblies (now designated nanodiscs) by binding to the exposed apolar regions of the phospholipids.^{17, 18}

Kaiser and Kézdy took this strategy one step further by designing peptides that idealize the structural and physicochemical properties proposed to be responsible for activity, rather than focusing on any natural sequence.^19–23 The resulting peptides were found to bind to and stabilize small unilamellar vesicles of a size and shape similar to that of high-density lipoproteins, and the concept was soon expanded to other peptides and proteins.^{20, 24, 25}

Cytotoxic peptides and antimicrobial peptides (AMPs)

As a graduate student working in the labs of Kaiser and Kézdy, one of us (WFD) expanded this concept to test the structural basis for the cytotoxic function of the bee venom peptide, melittin (although spelled with a single l, melittin is named for the honey bee, Apis mellifera, often causing confusion about its spelling in the literature). At the time it was unknown whether melittin elicited its toxic effects by a purely physical or receptor-based biochemical mechanism. Examining its structure on a helical wheel revealed an amphiphilic structure, but one that was distinct from apolipoproteins; in melittin 2/3 of the arc was composed of hydrophobic residues and the remaining 1/3 primarily small neutral polar residues. A highly basic hexapeptide resides C-terminal to the helical region, and melittin is devoid of any acid residues. A peptide designed to idealize melittin’s amphiphilic structure was several-fold more potent than melittin in its ability to lyse red cells and disrupt vesicles.^{26, 27} Also, as was the case for melittin, the peptide was in a random coil configuration in dilute solution, but became helical when allowed to self-associate or bind to apolar-water interfaces. Subsequently, sequences of a family of 14-residue peptides from wasp venom, the mastoparans,²⁸ were discovered to have an even simpler helical motif;^{28, 29} again about 2/3 of the residues were hydrophobic, but the positively-charged residues located to the polar side of the helix, rather in a distinct segment C-terminal to the helix as in melittin. Thus, a minimal structural feature for disruption of mammalian cells appeared to be a short, positively charged amphiphilic helix.

At about this time, Boman discovered the peptide cecropin³⁰ in the course of studying innate immunity in insects.³¹ Cecropin was the first ribosomally synthesized AMP that was not toxic to eukaryotic cells. Cecropin had two basic amphiphilic helices; both were less hydrophobic than those found in melittin or the mastoparans. The labs of Merrifield^{32, 33} and DeGrado³⁴ noticed the amphiphilic character and showed the first helix alone to have antimicrobial activity. DeGrado also designed the first minimal synthetic model for AMPs with the sequence (LKKLLKL)₂.^{29, 34} Soon thereafter, Michael Zasloff made his seminal discovery of the first AMP, magainin.³⁵ Propelled by the importance of innate immunity to human health and Michael’s charismatic leadership, the field rapidly grew. Soon AMPs were found in many organisms including humans. Many were found to form amphiphilic helices, but others were small disulfide-rich mini-proteins (e.g., defensins), Pro/Arg-rich peptides, Trp-rich peptides, anionic AMPs, and β-hairpins.^36–44 Because many AMPs work by a biophysical mechanism, rather than by targeting a specific protein or ribosomal target, they have less tendency to develop resistance. The field of both designed and natural antimicrobial peptides has exploded, and there are now books and entire conferences devoted to the subject.^{45, 46} AMPs have diverse targets, but most target membranes as at least a part of their mechanisms of action.^{39, 47–49}

Much biophysical work has been devoted to understanding the mechanism by which AMPs disrupt bacterial membranes, and what differentiates AMPs from more toxic peptides such as melittin.^36–40 A reading of the literature highlights a diversity of mechanisms, ranging from channel formation to more generalized membrane disruption. While many papers in this volume discuss very detailed measurements of AMP–membrane interactions, here we present a less detailed understanding of the principles of AMP–membrane interactions, which guided the design of both peptide and the non-peptidic mimics of AMPs described in the subsequent section.

AMPs disrupt membranes at relatively high peptide:lipid ratios, when the surface concentration exceeds a certain threshold,⁵⁰ at which the peptides essentially carpet the membrane surface, leading to both generalized membrane disruption⁵¹ and heterogeneous pore formation.^{6, 50, 52–63} The asymmetric accumulation of such a large amount of a peptide on only one leaflet of the bilayer results in a large imbalance in chemical potential. For example, at a peptide/lipid ratio of 1:100 (corresponding to 1:50 if the AMPs localize to only the outer leaflet), there is a strong thermodynamic driving force to equalize the concentration on both sides of the bilayer. The gradient is roughly equivalent to the osmotic gradient created by introducing a solute at 1 M on just one side of a water-permeable membrane (in which case the mole fraction of the solute to water would be approximately 1:50 for a small molecule or ion). In addition to this strong thermodynamic driving force, the physical properties of the peptide contribute to lowering the kinetic barrier for membrane translocation. Insertion leads to expansion, disordering and thinning of the bilayer.^{61, 64} Moreover, asymmetric insertion leads to surface distortions such as buckling. In each case the bilayer is stressed, helping to reduce the energy of activation for AMP translocation and equilibration across the bilayer. Simulations suggest that the translocation step can also occur with a transient increase in ion permeability.^{65, 66}

AMPs also decrease the stability of bilayers and enhance the stability of nonlamellar phases including lipidic cubic phases and cylindrical micelles, with head groups projecting into a water channel and the apolar groups radiating outward; a similar structure might be formed as a defect within a bilayer, creating a pore that is lined by both headgroups and surface-absorbed peptides.^{52, 53, 55–60, 67–70} The resulting toroidal pore has a shape similar to the inside of a donut with a topology designated as negative Gaussian curvature (a shape in which there is the curvature is negative in one direction and positive in the orthogonal dimension). Gerard Wong has shown that antimicrobial peptides of many classes stabilize negative Gaussian curvature.⁷¹ His group has shown that Lys, Arg, and aromatic amino acids bias towards phases with negative Gaussian curvature and used this principle together with machine learning to design highly effective de novo AMPs.⁷²

Fully peptide-lined pores can also be formed by peptides such as alamethicin, which are sufficiently hydrophobic and long enough to span the bilayer (at least about 18–20 residues in α-helical peptides). For example, melittin binds initially as a random coil to acidic membranes, next it forms a laterally inserted helix, and ultimately forming a membrane-spanning helix at sufficiently high peptide:lipid ratios.^{4, 73, 74} The vertically inserted helix can next associate in a variety of stoichiometries to form heterogenous transient pores, which then disrupt ionic gradients required for cell viability.

In summary, AMPs have a common feature of being able to bind to the surface of the bilayers, by using a combination of electrostatics and hydrophobic interactions. They insert aromatic or hydrophobic groups sufficiently deep the bilayer to lead to a surface pressure imbalance, and modulation of the properties of the bilayer. The destabilization of the bilayer also decreases the activation energy for subsequent transitions of peptides into and/or across the bilayer. The accumulation of peptides on one side of the bilayer also leads to an imbalance in chemical potential providing a thermodynamic driving force for AMP translocation, which can occur with concomitant diffusion of ions and disruption of the membrane potential. Full-fledged pore formation can occur through the formation of channels that are either lined solely by peptides or a combination of peptides and lipid head groups. Different AMPs can access different mechanisms.

The selectivity of AMPs for bacterial membranes contrasts with toxins like melittin, which disrupt membranes of both mammalian and bacterial cells – a phenomenon that has been widely attributed to differences in the lipid compositions of different cell types. Bacterial cells are rich in acidic lipids compared to mammalian cells, and bacteria also lack cholesterol. In many cases,³⁶ subtle effects relating to phospholipid composition are involved; primary phospholipids in mammalian membranes include zwitterionic phosphatidylcholine (PC) and negatively charged phosphatidylserine (PS) headgroups, while bacterial membranes are rich in zwitterionic phosphatidylethanolamine (PE) and negatively charged phosphatidylglycerol (PG) lipids including cardiolipin. The differences in the physical properties of these various lipids can be as important as the overall charge of the membrane.^75–78

A number of features of a peptide define its selectivity. We have already seen that cytotoxic peptides tend to have greater hydrophobicity, leading to more non-discriminate binding than selective AMPs. For helical AMPs, the insertion of the peptides is thermodynamically linked with helix formation, so peptides that strongly favor the helical state are intrinsically predisposed to binding more tightly than closely related peptides with similar charge and hydrophobicity. Thus, for each scaffold there exists a delicate balance between charge, hydrophobicity and rigidity that must be fine-tuned to optimize both potency and selectivity. Finally, the response of a bilayer to bound AMPs likely varies with respect to differences in the lipid composition, so changes to not only binding affinity, but also biological activity likely contribute to selectivity.⁷⁶

Foldamers and polymers that mimic AMPs

Given the above mechanistic understanding of the structural features and mechanisms of action of AMPs, it became apparent that it should be possible to mimic their activities using molecules other than peptides composed of α-amino acids. Because the activities of many AMPs depends primarily on their overall physicochemical properties—rather than the fine details of their precise amino acid sequences—several groups designed “coarse-grained” molecules idealizing the amphiphilicity of natural AMPs.^{29, 37, 79, 80} Modifications on natural peptides have included the introduction of d-amino acids,^{81, 82} long acyl chains,^{83, 84} and cyclization.^{81, 85} The availability of β-peptides provided another avenue to probe the requirements for activity, as they can adopt distinct secondary structures such as “12-helices” and “14-helices”.^86–90 Our group^{91, 92} as well as those of Gellman⁹³ and Seebach⁹⁴ independently showed that β-peptides capable of forming amphiphilic 14- or 12-helices had potent antimicrobial activity. Oligomers with lengths of 10–15 residues that achieved an appropriate hydrophilic/lipophilic balance were selective for killing bacteria vs. mammalian cells. Oligomers that were too short were inactive, and these short sequences also failed to adopt the desired conformation. Moreover, foldamers that were too long or hydrophobic were toxic towards mammalian cells.⁹⁵ These studies were extended to a variety of different helical types.^{76–78, 96, 97} In a similar manner, Patch and Barron designed amphiphilic, helical, antibacterial N-substituted glycine oligomers (peptoids),⁹⁸ and Guichard and coworkers have synthesized antimicrobial foldamers based on a urea backbone.⁹⁹ Thus, by the early 2000s it had been well established that medium-sized molecules that adopt amphiphilic secondary structures were capable of acting as highly potent and selective antimicrobial agents.^100–102

Going one step further, the groups of DeGrado, Tew, and Klein collaborated on the design of antimicrobial oligomers that were more akin to small molecules than earlier foldamers.^103–108 Using a rigid arylamide as a framework they systematically introduced cationic groups (cat. 1 and cat. 2 in Figure 2) and hydrophobic groups to maximize activity against Staphylococcus aureus, while minimizing toxicity towards mammalian cells in vitro and in mouse models.^{107 109}

Fig. 2: Generic structure of a rigid platform used to design antimicrobial peptide mimics.

By systematically varying the cationic group (cat. 1 and 2) and the hydrophobic group (Hb), it was possible to maximize potency and efficacy:safety ratio. Brilacidin is undergoing clinical trials for bacterial infections and SARS-CoV-2.

Brilacidin has undergone two phase II clinical trials for S. aureus infections and was shown to have an efficacy on par with daptomycin.¹⁰⁹ There is currently great interest in AMPs as potential antiviral agents against enveloped viruses,^110–113 and brilacidin has similarly been shown to be active against SARS-CoV-2.¹¹⁴ A phase II clinical trial for COVID is currently in progress (http://www.ipharminc.com/brilacidin-1).

The antibacterial mechanisms of action of brilacidin and closely related compounds have been studied by probing the transcriptional response and leakage of bacterial content in E. coli, as a Gram negative organism,¹¹⁵ and S. aureus as a Gram positive.¹¹⁶ These compounds cause significant changes in the permeability of the outer membrane of E. coli, similar to that observed in polymyxin B and nisin.^{117, 118} They cause comparatively little permeabilization of the inner membrane, although they lead to a loss of membrane integrity, reaching critical levels corresponding with the time required to bring about bacterial cell death. Transcriptional profiling of E. coli treated with sub-inhibitory concentrations of the arylamides also showed induction of genes related to membrane and oxidative stresses. The induction of membrane-stress response regulons coupled with morphological changes at the membrane observed by electron microscopy indicate that the activity of the arylamides at the membrane is the primary mechanism of action.¹¹⁵

In S. aureus brilacidin also causes membrane depolarization, to an extent comparable to that caused by the lipopeptidic drug, daptomycin.¹¹⁶ However, there was little leakage of cellular contents, ruling out mechanisms that involve large pores. Transcriptional profiling showed that the global response to brilacidin treatment is well-correlated with the corresponding response to daptomycin and a derivative of the cationic antimicrobial peptide, LL37, and induced regulons that respond to perturbation of cell wall and membrane function. These stress responses were mainly orchestrated by three two-component systems: GraSR, VraSR and NsaSR, which have been implicated in virulence and drug resistance against other clinically available antibiotics.

Coarse-grain MD allowed simulations with a relevant number of arylamides per bilayer⁶⁵ at coarse-grained oligomer/phospholipid (CGO/PL) from 1:256 to 1:14, spanning surface concentrations that experimentally give rise to from no lysis to very rapid vesicle lysis. At low CGO/PL ratios the antimicrobial inserts into the bilayer with its long axis parallel to the membrane surface and its apolar groups penetrating into the membrane. By contrast, at the high CGO/phospholipid ratios required for lysis of bilayers in experimental systems, the drugs initially insert with an ensemble of different angles. The very large imbalance in the concentration of the oligomers between the proximal versus the distal leaflets of the bilayer leads to a metastable state with pronounced buckling of the proximal leaflet of the bilayer. At longer times, the molecules diffuse to the opposite side of the bilayer. The CGOs frequently diffuse in pairs across the bilayer, often with accompanying ions and water molecules. Following the translocation step the oligomers are oriented predominantly parallel to the bilayer. However, they are less well ordered than in the simulations at low antimicrobial:phospholipid ratios. In the final configuration, the polarity gradient of the membrane is altered, and there is an increased permeability to water in the simulations. These simulations provide pictorial detail to prior models described above.

We hope the above section is helpful in defining the principles required for design of AMP mimetics. There are currently a number of AMPs and AMP mimetics undergoing clinical trials for various indications.^{119, 120} In addition to understanding the principles for design, which has been the primary focus here, it is also essential to understanding the biophysical bases for their action. The reader is pointed to other papers in this volume for a wealth of information on AMP-membrane interactions.

Dynamics, structure, and mechanism of membrane-bound α-synuclein

One of the most prominent examples of membrane-binding peptides, α-synuclein (Figure 3A), is also one the most complex.¹²¹ Although synuclein is sufficiently long to be considered as a protein, it has a number of repeating peptide units, and different portions of its sequence interact differently with the membrane that show the diversity of ways peptides can interact with membranes.

Fig. 3: Sequence and structures of α-synuclein.

(A) Helical wheel depicting the amphiphilic nature of α-synuclein’s 11-residue consensus sequence, which mediates membrane binding. (B) Alignment of the 11-residue consensus sequences of α-synuclein (α-syn) and apolipoprotein A1 (apoA1).¹²⁹ (C) Structural model of α-synuclein showing C-terminal dynamics in an extended, membrane-bound helix, which is supported by data from FRET,¹³⁰ EPR,^{131, 132} NMR,^{133, 134} and DMS.¹³⁵ (D) Cartoon model of α-synuclein tethering nearby vesicles.¹³⁶ (E) Hypothetical model of α-synuclein interacting with membranes of negative Gaussian curvature, such as those in fusion pores.¹³⁷

Originally discovered by screening cDNA expression libraries for proteins localized to synapses using antibodies raised against synaptic preparations from electric rays, α-synuclein was shown by electron microscopy to localize to the surface of synaptic vesicles, suggesting a direct peptide–membrane interaction.¹²² This hypothesis was further supported by similarity between the sequences of α-synuclein and apolipoproteins (Figure 3B),¹²³ which use amphiphilic helices featuring unique 11-residue repeating segments, to sequester and transport lipids.¹²⁴ α-Synuclein localizes specifically to synaptic vesicles, rather than the plasma membrane, and disperses from nerve terminals after stimulation,¹²⁵ suggesting that it might participate in synaptic vesicle trafficking. Indeed, early experiments on knockout animals revealed changes in stimulated dopamine release,¹²⁶ suggesting that α-synuclein plays a key role in neurotransmission. These early studies, as well as genetic¹²⁷ and pathological¹²⁸ links to Parkinson’s disease, motivated detailed study of the interactions of α-synuclein with lipid membranes, which have revealed a series of remarkable biophysical features that contribute directly to α-synuclein’s physiological and pathological activities.

Plotting the distribution of residues on a helical wheel revealed a facial separation of hydrophilic and hydrophobic residues, with cationic seams in between to potentially support membrane binding (Figure 3A).¹²³ Early experiments on purified α-synuclein confirmed this hypothesis, showing by circular dichroism that α-synuclein, which lacks defined structure in isolation,¹³⁸ adopts helical secondary structure upon association with lipid membranes.¹³⁹ Although these behaviors are shared with other membrane-binding peptides, α-synuclein’s length and hydrophobicity make it unique, and these features likely contribute to both its physiological and pathological activities.

Compared to other amphiphilic helices, α-synuclein is only modestly hydrophobic.¹⁴⁰ Its hydrophobic face consists primarily of smaller residues like Gly, Ala, and Val, which are interrupted by hydrophilic Thr residues. In contrast, the hydrophobic face of APLS (amphipathic lipid packing sensor) motifs, for example, often feature uninterrupted stretches of larger, greasier residues like Leu, Ile, Met, and Phe.¹⁴¹ α-Synuclein’s modest hydrophobicity tempers its affinity for lipid membranes. Unsurprisingly, for example, substituting the Thr residues with bulkier hydrophobic residues increases affinity, which in turn increases toxicity in cellular and animal models,¹⁴² suggesting that the affinity of α-synuclein for lipid membranes is tightly optimized. Moreover, the prevalence of Gly residues in this helix further attenuates membrane-binding affinity by increasing the entropic cost of helix formation, though the alignment of these Gly residues along one face of the helix suggests additional or more nuanced roles, perhaps mediating interactions with specific proteins or lipids. This tuning of α-synuclein’s membrane affinity not only increases the rate of exchange between the membrane surface and solution, which could be involved in its reported roles in synaptic vesicle trafficking;¹⁴³ it also enables α-synuclein to discriminate membranes with different physical and chemical properties.

For example, α-synuclein has a higher affinity for membranes of increasingly negative charge,¹³⁹ likely owing to the relative importance of electrostatic interactions, rather than strictly hydrophobic association, to drive membrane binding.¹⁴⁴ α-Synuclein’s modest membrane affinity has also been suggested to contribute to its ability to discriminate membranes with different packing or curvature.^{139, 145} The modest affinity likely provides insufficient energy to disrupt the packing of planar bilayers but could be sufficient to bind to the gaps between lipid molecules in highly curved membranes, such as those of synaptic vesicles. Moreover, theoretical models¹⁴⁶ and course-grained MD simulations¹⁴⁷ demonstrate that the partitioning of α-synuclein to a specific depth within the membrane, as tuned by its physicochemical properties, is sufficient to drive curvature formation.

α-Synuclein is also distinctive because of its length.¹⁴⁸ Preliminary sequence analysis suggested that the first 90–100 residues could form a continuous amphiphilic helix, far longer than most examples, the most notable exception being the perilipins that bind to lipid droplets.¹⁴⁹ Preliminary NMR experiments confirmed the membrane-binding region as the first 90–100 residues,¹⁵⁰ which include the 11-residue segments predicted to mediate lipid-binding; the remainder of the 140-residue protein remains disordered and unbound from the membrane. Subsequent studies disagreed about whether this region forms a long, single helix,¹³¹ or whether the membrane-binding domain is broken into two segments.¹⁵¹ The discrepancy highlights the effect that the membrane or membrane mimetic can have on peptide and protein structure. The first high-resolution NMR structure, solved in SDS micelles, suggested such a break.¹⁵² In contrast, studies by EPR,¹³² FRET,¹³⁰ and solution-¹³³ and solid-state NMR,¹³⁴ conducted with protein reconstituted in bilayers, instead favored a single, extended helix bound to the membrane. Consensus has emerged that this discrepancy likely arises from the curvature of the membrane mimetic; highly-curved species like SDS micelles induce breaking of the helix, while less curved membranes such as LUVs support helix formation across the entire membrane-binding domain.

The diversity of structures formed by α-synuclein has raised important questions about the functional and pathological roles of each individual species. To begin dissecting these roles, we adapted deep mutational scanning (DMS) to reveal the structure associated with a specific activity. In DMS, a pooled library of protein missense variants is screened for relative activity, which is inferred from changes in the frequency of each variant before and after selection.¹⁵³ By subjecting comprehensive libraries of missense variants to relatively weak selective pressure, such that most variants are retained in the population after selection, the relative effects of thousands of mutations can be determined and used to infer structural features of the protein that are responsible for activity.

We sought to determine the structural basis for a relatively simple activity of α-synuclein: its toxicity when ectopically expressed in yeast cells. By identifying mutations that interfere with formation of the toxic species and rescue the growth rate of yeast cells, we discovered that α-synuclein toxicity in yeast is driven by a dynamic, membrane-bound helix with increasing dynamics toward its C terminus.¹³⁵ For example, mutations that introduce polar groups onto the hydrophobic face reduce both membrane binding and toxicity, but this effect becomes less pronounced toward the dynamic C terminus. Similarly, helix-breaking proline residues reduce toxicity, again most potently at the N terminus. This result is in excellent agreement with models proposed to mediate vesicle clustering,¹³⁶ implicating this activity in toxicity, at least in yeast.

We then developed a simple structural model for this state (Figure 3C),¹³⁵ which we used to make thermodynamic predictions of the effect of each mutation on membrane binding, based on the known energy associated with positioning each amino acid to varying depths in the lipid bilayer.¹⁵⁴ These predictions were remarkably successful at predicting the effects in our cellular assay, providing strong support for the role of this species in toxicity. Our model also demonstrates that the entire helix contacts the membrane to similar depth, while the C terminus engages in increasingly frequent dynamic dissociation from the membrane surface. Again, these results from functional measurements are in remarkable agreement with inferences made from complementary spectroscopic methods, specifically NMR^{133, 134} and EPR.^{131, 132} The dynamics at the C-terminal end of the helix might result in part from the sequence of α-synuclein’s uniquely long amphiphilic helix. Although each of the repeated 11-residue binding segments encodes nearly three turns of an a-helix, the slight mismatch between the sequence and its encoded structure causes the hydrophobic face of the helix to gradually wrap around helix.¹⁵⁵ The energy associated with overwinding the helix to align the hydrophobic groups could prevent both ends from binding tightly.

The length of this helix creates interesting dynamical properties. In order to align the hydrophobic groups on one face of the helix, α-synuclein would have to overwind, and although the energy required to do so is small for individual residues or helical turns, that energy becomes significant for a helix of α-synuclein’s length. This energetic penalty to adopting the structure that would optimize the hydrophobic groups for membrane binding prevents the protein from fully associating with the membrane surface across its entire length.

Dynamics in the membrane-bound helix are hypothesized to have a variety of physiological roles. One putative role is the tethering of membrane surfaces to one another, release of the C-terminal region from the surface of one membrane allows it to associate with nearby surfaces,¹³⁶ which might mediate vesicle clustering (Figure 3D), one known roles of α-synuclein. Intriguingly, aberrant protein-rich inclusions in the brains of Parkinson’s patients called Lewy bodies are enriched in both α-synuclein and membrane-bound organelles and vesicles,¹⁵⁶ suggesting that this clustering activity might contribute to pathology, although this hypothesis is controversial.¹⁵⁷ Alternatively, dynamics in the membrane-bound state could support protein–protein interactions with other membrane-remodeling machinery, such as SNARE proteins.¹⁵⁸

Dynamics in the membrane-bound state can also promote membrane curvature;^{159, 160} as the surface concentration of α-synuclein increases, clashes between monomers involving the dynamic end of the helix and the disordered tail would be relieved upon increased curvature. Electrostatic repulsion between the acid tails of each monomer could further drive repulsion and increased curvature. These biophysical mechanisms might underlie α-synuclein’s ability to tubulate membranes.^{131, 161} Alternatively, the inability of the protein to associate with the membrane surface across its entire length might create an energetic driving force for the association with or formation of membranes with negative Gaussian curvature (Figure 3E), which has been observed in course-grained simulations,¹⁶² such as those that form during vesicle endocytosis or exocytosis. The energy release from binding of α-synuclein to such a surface could contribute to its observed ability accelerate fusion pore dilation.¹³⁷

Membrane-binding has a strong influence on the aggregation of α-synuclein,¹⁶³ which is implicated in Parkinson’s and related diseases. At low protein:lipid ratios, membrane binding suppresses α-synuclein aggregation by distracting individual monomers from self-association.¹⁶⁴ At higher concentrations, however, membrane-binding induces aggregation by concentrating monomers on the membrane surface. Moreover, the dynamic release of the C terminus of the amphiphilic helix from the membrane surface exposes the aggregation-prone NAC (non-amyloid component) region, which is then poised for self-association at high protein:lipid ratios.¹⁶⁵ Dysregulation of either the concentration of α-synuclein or its membrane affinity could therefore drive pathological aggregation.¹⁶⁶ In turn, α-synuclein aggregation can perturb its interaction with lipid membranes by either concealing¹⁶⁷ or revealing membrane-binding regions.¹⁶⁸

In summary, α-synuclein exemplifies many of the features of peptide-lipid interactions considered in this volume. In one single protein, we see surface helices that specifically recognize and bind to membranes with high curvature, using a combination of peptide-lipid interactions as well as generalized crowding effects. We also see multiple types of interactions exemplified in a single protein: α-synuclein adopts (1) a highly stable helix that is inserted into the bilayer surface near the N terminus; (2) a dynamic helix, which adopts a dynamic equilibrium between inserted helical conformations and unfolded structures; and (3) a fully unstructured C-terminal tail tethered to the membrane surface.

INSERTION, ASSEMBLY, AND DESIGN OF SINGLE-SPAN TM HELICES

Insertion of transmembrane helices

Membrane proteins are generally inserted into membranes via the translocon, and they complete the folding process in the membrane environment.^{169, 170} The helical conformation is stabilized in a membrane because the backbone amide-carbonyl hydrogen bonds are satisfied within the helix and sequestered away from the nonpolar environment of the bilayer.^{171, 172} In the two-stage model of membrane protein folding,^173–175 once inserted in the bilayer, the protein is then able to fold via the coalescence of the helices to form the native tertiary structure. The study of designed and natural transmembrane peptides has greatly informed our understanding of the forces that stabilize both the insertion and coalescence of helices in membrane proteins. Thus, TM peptides are excellent models for multi-span membrane protein folding. For the biophysicist, self-associating TM helices allow investigation of unconstrained inter-helical interactions with a clear unfolded state—a monomeric α-helix—where conformational specificity and thermodynamics can be simply evaluated by the oligomeric distribution. From a more biological perspective, half of all human membrane proteins have a single transmembrane helix.^{176, 177} Moreover, TM helices often do more than anchor proteins in membranes; they self-associate or interact with other membrane proteins to play vital roles from signaling to ion conduction, and often the aberrant assembly of TM helices is central to devastating diseases from cancer to Alzheimer’s disease.^{178, 179} Thus, the study of TM peptides is a fruitful endeavor for biophysicists and biologists alike.

The features influencing the insertion of TM helices into membranes has been extensively. Sufficient hydrophobicity is required to enter the translocon and stabilize the helix in an apolar environment,¹⁸⁰ dictating the placement of apolar residues at most positions in the helix.^{154, 170, 181–183} Additionally, aromatic and basic residues are well accommodated in the headgroup region of a bilayer, where they can contribute to stability.^184–187 The residues in the headgroup region also define the transverse shift of a helix in a membrane, as does the presence of polar residues within the hydrophobic region.¹⁸⁸ Charged residues in the hydrophobic region of a TM peptide also have a large effect on the ability of peptides to insert into bilayers. Early work from Caputo and London^{189, 190} focused on three model TM Leu-rich peptides, each containing Asp at different positions in their hydrophobic core. When the Asp residue was protonated at low pH, the peptides inserted vertically into vesicles composed of dioleoylphosphatidylcholine (DOPC). When the Asp residues was ionized at neutral or high pH, the topography was altered in a manner that would allow the charged Asp residues to reside near the bilayer surface. More recently, Engelman and coworkers have developed a “pHLIP” peptide, which is a 36-aa peptide derived from the bacteriorhodopsin C helix with an acidic residue near the center of the TM helix. This peptide exists in three states: soluble in aqueous solution, bound to the surface of a membrane, and inserted across the membrane as an α-helix; decreasing pH stabilizes the inserted state.^{191, 192} An “ATRAM” peptide from the Barrera lab has similar physical characteristics.^{193, 194} Both insert unidirectionally into bilayers, and are capable of bringing cargos into cells. Given the acidic environment of tumors they show promise for targeted delivery and diagnosis of tumors.^195–197

There is generally a minimum length of about 20 residues (30 Å) to stably span the bilayer in a vertical orientation, and there are significant effects of a hydrophobic mismatch between the membrane and the peptide.^198–201 Shorter TM helices will necessarily experience a hydrophobic mismatch with the bilayer, leading to destabilization of the membrane lamellar phase.²⁰² On the other hand, TM helices with hydrophobic lengths significantly longer than 20 residues will tend to insert at an angle relative to the membrane normal, as this orientation allows more efficient burial of the TM helix.^{203, 204} Finally, while the rotation of a helix about its own axis does not affect the depth of insertion of an amino acid residue, rotation of a tilted about its axis places a sidechain into different positions in the bilayer, which can differ significantly near the headgroup region.^205–207 Thus, the membrane environment has a very significant impact on “setting up” a helix for favorable helix-helix interactions.²⁰⁸

Structural aspects of the assembly of transmembrane helices

Peptide–membrane interactions also play an essential role in defining helix–helix interactions in membranes.^{209, 210} The hydrophobic effect provided by burial of apolar side chains in a protein’s interior represents the predominant driving force for helix–helix association and folding in water, yet it is negligible in lipid membranes. What then drives assembly and folding of membrane proteins? Lipid-specific effects such as “solvophobic” exclusion likely contribute.^{211, 212} Oligomerization also relies on matching of the hydrophobic thickness of the TMD with that of the surrounding lipid. If the lipid bilayer is too thin or too thick, oligomerization can be reduced in model membranes.^{213, 214} Moreover, there can be very large effects when comparing native cell membranes with model membranes. These effects can even overwhelm stabilizing contributions from favorable helix–helix interactions, lateral packing pressure and excluded volume.²¹⁵ Thus, while the following section focuses on stabilizing helix–helix interactions, it is important to keep in mind that the association of helices represents a balance between membrane–protein and protein–protein interactions, which is sensitive to the membrane environment.

Structural informatics, peptide design, and site-directed mutagenesis of natural proteins have greatly advanced our understanding of the mechanisms of association of TM helices. We identify three general categories of interactions that stabilize protein structures. In some packings such as GX₃G motifs, the backbones of two helices approach closely forming favorable weakly polar interactions and efficient van der Waals packing. In other cases, electrostatic interactions and hydrogen bonds involving polar or charged amino acid sidechains help drive association. Finally, tight packing of apolar sidechains can be a strong driving force for association. We will consider each separately.

GX₃G (and SmallX₃Small) and Small-X₆-Small motifs.

In the GX₃G motif, first identified in glycophorin, Gly residues spaced four residues apart mediate a very close association of the backbone of two helices stabilized in part by C–H hydrogen bonds between the alpha CH groups and carbonyl groups of adjacent helices.^216–218 This non-classical hydrogen bonded interaction is favorable, and it also allows very efficient van der Waals packing interactions between proximal groups in the interacting helices.^219–221 In the classical GX₃G motif, the two interacting helices are parallel, and they have a right-handed crossing angle of about +40°. Variations on the GX₃G motif are seen in which one or both of the Gly residues are instead small side chains, such as Ala. It is noteworthy that these residues lose no side-chain conformational entropy when the helices associate, as is the case with larger, more flexible side chains. By contrast, the release of lipid tails to the bulk will increase the entropy of the lipids. The GX₃G motif mediates critical TM interactions in many biologically important single-pass proteins that include receptor tyrosine kinases. An extensive literature attests to the importance of the subject, and the interested reader is directed to a number of classical reviews from the labs of Engelman, Fleming, MacKenzie, Langosch, Akin, Senes, Schneider, and others.^{210, 217, 222–225} A more recent comprehensive review by Westerfield and Barrera is particularly insightful in its thoughtful treatment of the role of GX₃G and related motifs in receptor association.²²⁶

There are several variations on the GX₃G motif,²²⁷ in both TM peptides and larger protein structures.²²⁸ The classical GX₃G motif is C2 symmetric, and the interactions are identical on each helix, as in glycophorin A.²²⁹ Additionally, small amino acids can occur repetitively every four residues to create extended “glycine zippers”.²³⁰ In other cases, an asymmetric interaction utilizing the GX₃G motif on just one of two helices is seen in heterodimers. One example occurs in the integrins. In these heterodimeric proteins the TM helices of their alpha and beta subunits interact tightly in the resting state, but dissociate when they are activated for binding to extracellular proteins. Thus, the interaction strength between the two helices is finely balanced to allow the conformational change to occur in response to intracellular and extracellular signals. The interface of the αIIbβ3 TM domain heterodimer^231–233 consists of a tightly packed structure in which small and large residues interdigitate by efficient van der Waals packing along the heterodimer interface. Thus, a sequence motif in the αIIb TM domain, G-X₃-G-X₃-L, packs in a reciprocal manner with the β3 TM domain sequence V-X₃-I-X₃-G such that bulky residues from one TM helix contact a hole formed by a small Gly residue on the neighboring helix.²²⁷ A similar interaction occurs between the TM helices of a second β3 integrin, αvβ3, but using a different face of its TM helix to bind the αv TM helix the one used to interact with the αIIb subunit.²³⁴

The GX₃G motifs discussed above mediate interactions between parallel helices. An analogous type of interaction occurs between antiparallel helices with small residues spaced every 7 residues apart,^{154, 230, 235} designated here a Small-X₆-Small motif. The interaction is mediated by a mix of favorable electrostatic interactions between the peptide backbones, efficient van der Waals packing, and other interactions that depend on the sequence context. In this motif, the antiparallel helices have a left-handed crossing angle of about −10° to −30°, similar to classical coiled coils. Coiled coils have a characteristic 7-residue repeat, whose positions are designated “a” through “g” within a single heptad. By convention, the residues at the “a” and “d” positions of the heptad pack in the core of a coiled coil. The stability of water-soluble coiled coils scales with the size and hydrophobicity of the side chains at the “a” position increasing over the series Gly < Ala <Val < Ile.²³⁶ Just the opposite trend was seen in membrane-soluble coiled-coil peptides, Gly > Ala >> Val > Ile.²³⁷ Furthermore, the peptides in which the “a” residues were either Gly or Ala had a strong preference to form antiparallel “Ala-coil” structures (Figure 4) when in a membrane,^{237, 238} which appeared to be associated with alignment of dipoles from the amides of adjacent helices. Similar results are seen with Ser at the “a” position. Interestingly, Ser can stabilize both antiparallel packing as seen in protein structures²³⁹ and model peptides (Figure 4),²⁴⁰ as well as a parallel helical dimerization motif found in the murine erythropoietin receptor.^{241, 242}

Fig. 4: Inner tetramer of inactive channel form of designed ion channel from pdb 6yb1.

(A) Helical wheel diagram depicting key interactions, namely the Ser and Ala zippers, that stabilize the tetrameric assembly. (B) The inner tetramer is stabilized by a hydrogen-bonding network formed by Ser (purple), Thr (blue), and waters (not shown). Views of the Ala-coil and the Ser-coil interfaces reveal favorable interactions that mediate the formation of the tetramer.

Polar interactions.

Polar interactions, including salt bridges and hydrogen bonds provide a significant driving force for assembly of TM helices.^243–248 Numerous oncogenic mutations of receptor-mediated tyrosine kinases and other receptors involve the introduction of polar residues in the TM helices, which leads to enhanced association or alteration of the geometry of the TM helix-helix interaction (for reviews see 224, 241, 249, 250). Mutants with polar substitutions in their TM helices also figure largely in membrane protein misfolding diseases.²⁵¹

Studies with model TM peptides have shown that Asn, Gln, Glu, Asp (and to a lesser extent His) enhance TM helix association.^{2, 244, 245, 247, 248, 252} Presumably, the Glu and Asp residues are protonated when inserted deep into a bilayer.²⁵³ Thus, each of these residues that induce strong association have a common feature of having both hydrogen bond donors and acceptors, capable of mediating inter-helical interactions. Ser and Thr are exceptions, however. These residues generally form an intramolecular hydrogen bond to a carbonyl three or four residues above in the helix, so there is a reduced driving force for forming hydrogen bonds with adjacent TM helices.^{171, 254, 255} The position of side-chain hydrogen bonds in a TM helix and their surrounding sequence have significant effects on stability.^256–259 As compared to an interaction between apolar residues, Asn–Asn interactions range from highly favorable in the apolar region of the bilayer to unfavorable when buried in the interior of a water-soluble structure.²⁶⁰ Because Asn is capable of inducing or enhancing TM association, Asn residues have been scanned through the TM helices of several receptors, including the thrombopoietin receptor,²⁴⁹ the erythropoietin receptor,²⁴² and the integrin αIIbβ3,^{261, 262} leading to both activation and inhibition of signaling.

Intramembrane salt bridges can also be a potent driving force for the association of TM helices. For example, signaling complexes of immune cells employ multi-component protein receptors composed of distinct binding and signaling subunits^263–265 that assemble via helix–helix interactions of their transmembrane domains. The binding subunits associate with extracellular ligands, while the signaling subunit typically contains immunoreceptor tyrosine-based activation motifs (ITAMs), which couple to downstream phosphorylation and signaling pathways. The interaction between a pair of acidic Asp residues on the signaling dimer and a basic residue (Lys or Arg) on the ligand-binding coreceptor are required for complex assembly within the otherwise nonpolar membrane.^{266, 267} This 2-to-1 acidic to basic polar association motif is ubiquitous in multi-component receptor families including Fc receptors²⁶⁸ and T-cell receptor-CD3 complexes.²⁶⁹ For example, in the T-cell receptor complex, the positively charged Arg and Lys residues on the TCRαβ TMs interact with negatively charged Asp and Glu residues on the CD3δε and CD3 δε heterodimers, and the ζζ-homodimer.²⁶³ These interactions are necessary for proper assembly of the complex in the ER and subsequent transport to the cell surface.²⁶³ Furthermore, the 2-to-1 acidic to basic association is observed in a family of more than 20 receptors that contain the signaling subunit DNAX-activation protein 12 (DAP12), a disulfide-linked homodimer with a minimal extracellular region.^{265, 267, 270} Recent, MD simulations and density functional theory calculations of the transmembrane helices with differing protonation states of the Asp-Asp-Lys triad identified a structurally stable interaction in which a singly protonated Asp-Asp pair forms a hydrogen-bonded carboxyl-carboxylate clamp that clasps onto a charged Lys sidechain.²⁷¹ This polar motif was also frequently observed in a search of the Protein Data Bank-wide search, indicating that it is a widely used motif.

van der Waals packing of apolar sidechains.

In water the burial of apolar sidechains in a protein interior gives rise to a large hydrophobic driving force that is not operative in a membrane. Thus, membrane proteins utilize many of the interactions and motifs described above. Nevertheless, the great majority of contacts in membrane proteins involve interactions between apolar sidechains.²¹¹ Structural informatics suggests that these hydrophobic residues pack more efficiently in membrane proteins,^{228, 257, 272–274} and the resulting van der Waals (vdW) interactions provide a primary driving force for folding. On the other hand, apolar side-chains in the folded state might provide little net stabilization, because lipids interact with these side-chains similarly in the unfolded state. In this view, vdW packing is secondary to stronger than the forces and structural restraints, including conventional and CH-mediated hydrogen bonds,^{218, 219, 244, 248, 256} topology (i.e., loops, water-soluble domains, α-helical insertion),²⁷⁵ and weakly polar interactions.^{210, 276} Furthermore, experimental studies using site-directed mutagenesis suggest that apolar packing contributes similarly to stability in water-soluble and membrane proteins.²⁵⁷ Finally, attempts to design proteins using membrane-solubilized versions of peptides that assembly favorably in water, such as the leucine zippers resulted in very rather weak or structurally uncharacterized association in the absence of polar interactions.^{244, 277}

Recent work has, however, shown that van der Waals interactions can play a dominant role in the assembly of a natural membrane homopentameric peptide, phospholamban. Mravic and coworkers²⁷⁸ deciphered a steric code involving tight and regular packing that stabilizes natural membrane proteins, and used this insight to design self-associating pentameric peptides de novo. The structures of the resulting pentamers were solved to high resolution. Indeed, the assemblies were so stable that they remained partially associated after boiling in strong denaturant solutions. These packing motifs are frequently observed in numerous unrelated membrane protein families. These results reveal a substantial and widespread role for apolar packing in the folding, stabilization, and evolution of membrane proteins. Interestingly, a comparison of membrane-soluble and water-soluble sequences indicated that on average, the steric code was more stringent and the helices packed more tightly in the TM motifs versus water-soluble structures. Thus, achieving tight and specific packing would appear to be more essential in membrane proteins, which cannot rely on the hydrophobic effect.

A second recent example of the importance of van der Waals packing has been discovered through a random screening approach. DiMaio and coworkers have used clever genetic screens to discover “traptamers”²⁷⁹ that interact in a sequence-specific manner with receptor tyrosine kinases, including the erythropoietin receptor and platelet-derived growth factor Beta (PDGFβR).²⁸⁰ Their work began with small single-span TM protein, E5, a peptide from a tumor virus that targets and activates PDGFβR in a key step of oncogenesis.²⁸¹ E5 is a disulfide-bonded dimer and the dimeric conformation is required for activation PDGFβR. A single transmembrane Gln residue is required for activity in the context of the native sequence, but genetic selections^{282, 283} identified entirely hydrophobic sequences that are similarly able to activate PDGFβR in a highly specific manner.^284–287 Going one step further, DiMaio’s group has designed traptamers that composed of only Leu and Ile that target TM helices,²⁸⁸ again in sequence and target-specific manners. Finally, they have extended this approach to activator of the erythropoietin receptor²⁸⁹ and other TM proteins.²⁸⁷ It seems reasonable to assume van der Waals forces provide the a major driving force for assembly in these cases.

Combining a diversity of forces and motifs.

In the above sections, we have focused on examples that illustrate individual forces and motifs, but in typical membrane peptides and proteins, a number of stabilizing interactions are combined to achieve stability and function.^{211, 290–292} We have already seen the intimate interconnection between non-conventional CH···O=C hydrogen bonds and van der Waals packing in GX₃G motifs.²²¹ Also, π-stacking²⁹³ and hydrogen-bonded interactions can augment this motif. For example, strong hydrogen bonds between a Ser and His sidechain augments the affinity of a GX₃G motif in the transmembrane homodimer of BNIP.^{291, 292} Similarly, His37 forms multiple polar interactions in the homo-tetrameric four-helix bundle, M2 from influenza A virus.²⁹⁴ While these interactions help stabilize the tetramer,²⁹⁵ the tetrameric structure is retained in the H37A mutant, as is ion channel activity although the ion selectivity is altered by this substitution.²⁹⁶

Multiple types of zippers of the SmallX₆Small family can combine in protein assemblies larger than a dimer, when they are placed at two, non-overlapping positions of a helix. A recent crystal structure of a designed channel peptide in a non-conducting state illustrates the concept. Its structure displays two types of helix–helix interfaces, one featuring an Ala-coil and the other a Ser-coil. The designed sequence has a sequence repeat with small Ala or Ser residues of the heptad repeat (Figure 6). As mentioned above, small residues spaced 7 residues apart are the signatures of Ala and Ser zippers,^{228, 239, 297, 298} in which two helices wrap around one another in an antiparallel coiled coil. In this structure, the packing of small residues provides a strong driving force for assembly in a phospholipid bilayer, while the Ser residues make polar interactions. Both Ala coils and Ser zippers are observed widely in membrane proteins and have been used in membrane protein design.^{237, 240, 299}

Fig. 6: Designed ion channels.

Models of (A) the homotetrameric LS2, and (B) the homohexameric LS3 channels show pore-facing Ser residues with Leu at well-packed helix–helix interfaces. (C) Designed Zn²⁺/H⁺ antiporter Rocker (pdb 2MUZ) binds zinc through two 4-Glu, 4-His sites in the channel.

Analysis and design of functional membrane-spanning TM helices and helical bundles

Modulation of protein–protein interactions in the membrane.

The association of TM helices is often a critical feature in the assembly and function of single-span TM proteins. Peptides and small molecules that disrupt or modulate association provide very useful tools to probe the contribution of TM association to the function of membrane proteins—and they can translate to useful therapeutics. For example, Eltrombopag was the first small molecule drug approved that targets the TM domain of a single-pass membrane protein, and it is clinically approved for multiple indications.³⁰⁰ This drug targets the TM domain of the thrombopoietin receptor near His99.^{249, 301} There are several approaches to design peptides and small molecules that modulate receptor function, ranging from random screening/synthetic optimization to computational design of TM peptides. Many excellent reviews of the field are available, so we will only introduce the topic briefly.^{226, 302}

The simplest way to design peptide modulators of TM proteins is to synthesize or express a peptide spanning the sequence of the TM helix of interest. This was first demonstrated for glycophorin A in 1976 by Furthmayr and Marchesi,^{303, 304} who showed that a TM peptide was able to shift the equilibrium of the full-length protein from a dimer to a monomer by hetero-association. Since then, this approach has been widely used to study association of numerous proteins, as reviewed recently.^{226, 302} Going a step further, Barrera and coworkers have developed the introduced a Glu residue into the TM helix of EphA2, rendering it water-soluble, but still capable of inserting into membranes in a controllable manner.^305–307 This peptide activates EphA2 by interacting with the TM helix and its juxtamembrane domain. Attesting to its specificity, this peptide only inserts into membranes at neutral pH in the presence of the TM region of EphA2.

A second approach involves computational design of peptides that are designed to target the TM domains of proteins in a sequence- and structure-dependent manner. The GX₃G motif has been used as a basis to design “Computed Helical Anti-Membrane Protein” (CHAMP) peptides, which bind to the TM helices of single-span membrane proteins that have this motif in their sequence.³⁰⁸ A second designed CHAMP helix is docked to make good interactions with the peptide, then its sequence is optimized to recognize the specific sequence surrounding the GX₃G motif of the target. This method has been used to specifically recognize the αIIb, αv, and β1 subunits of integrins. They stabilize an activated form of the integrin by blocking the interaction of the alpha and beta TM helices^{278, 308–310} and have been helpful in determining the contribution of TM helix heterodimerization to overall activation of the integrin αIIb.³¹¹ Similar computational methods have now been expanded to a growing number of targets as reviewed recently.^{226, 302}

The third approach involves genetic screening to identify modulators of a given activity. A classical method to determine associations is through the construction of trans dominant negative mutations. In early work, this approach was elegantly employed by Pinto and Lamb to determine the tetrameric association state of the M2 proton channel from influenza A virus, through quantitative analysis of the effects of a dominant negative mutant on the proton channel activity of the protein. The elegant work of DiMaio and coworkers is also covered above in the section on van der Waals interactions.

Irrespective of the method used to design TM peptides, there are a number of technical issues that need to be carefully considered and appropriate controls conducted. If TM peptides are synthesized and added exogenously, one must tackle the problem of their poor solubility and tendency to aggregate in solution. While adding short poly-Lys sequences has been employed, it is important to conduct appropriate controls with mutants or scrambled peptides as the addition of a poly-Lys tag can be perturbing. An alternative is to use a short polyethylene glycol sequence of defined length³⁰⁸ in place of the Lys residues. Even after tagging, the peptides can have poor properties, which necessitates addition of organic co-solvents or micelles. Again, rigorous controls for these variable need to be conducted. An alternative approach is to express the peptide genetically for biosynthetic incorporation into membranes. Again, however, care needs to be taken to assure that the peptides are expressed appropriately in the desired target membrane, and to assure that they are not triggering an unfolded protein response in the ER or other off-pathway effects.

Analysis and design of self-assembling peptide channels.

Ion channels are essential for all electrical activity in the central nervous system, and for the translocation of ions, water, and small molecules across cellular membranes. Given our advanced understanding of the TM helices and helical bundles, we should now be able to design functional membrane proteins from first principles. Here, we focus primarily on designed peptide channels that transport protons and ions, highlighting natural model systems that have illuminated principles underlying channel activity and defining key considerations in peptide-membrane interactions that allow us to fine-tune ion conduction.

Our understanding of the detailed molecular mechanisms of ion conduction has progressed in a number of distinct stages.³¹² Before the sequences and structures of ion channel proteins were available, detailed studies of the structural basis of ion conduction focused on simple peptides, primarily gramicidin A (GA) and alamethicin. Next, as the sequences, but not yet atomic-level structures, of channels began to emerge, the de novo design of model peptides that mimic the pores of helical ion channels became an active area of research,^{313, 314} and studies of GA and alamethicin continued apace. As structures of natural ion channel proteins became increasingly available, they received increasing attention. Finally, we have reached a level of understanding of both membrane protein folding and ion channel function that it should now be possible to design channel proteins from first principles. The first forays into the field of de novo design have focused on testing and sharpening principles of folding and conduction, but applications³¹⁵ for sensing, water purification, and ion separation could follow.

Several milestones in the field of de novo design of ion channels have already been achieved. In early work, the design of peptides with well-defined rates of ion conduction and selectivity were reported, but high-resolution structures were not available.^{313, 314, 316} Very recently, the design of peptides and proteins that assemble into channels have been reported by Baker and coworkers, but channel recordings were limited to macroscopic conductance measurements and the per/protein or per/channel rate of conductance was not reported.³¹⁷ The design of channels with well-defined channel properties and high-resolution structures is imminent, and we will probably see papers on this subject emerge in the next year. However, for now, the only published report focuses on a peptide that uses a dynamic rocking mechanism to transport Zn(II).^{299, 318} Below, we briefly summarize work on the natural peptide channels, gramicidin A and alamethicin, as well as a natural four-helical proton channel. We then highlight progress in the design of peptides that assemble into ion-conducting channels.

Gramicidin A and Alamethicin.

Gramicidin A from Bracilus brevis, a bacteriostatic peptide comprised of alternating d- and l-amino acids, forms head-to-head dimers in lipid membranes to selectively transport monovalent cations, including H⁺, K⁺, and Na⁺.^319–323 Carbonyl groups of the peptide bonds line the pore, and although they are hydrogen-bonded to amide NH groups, they are able to tilt to stabilize transient cations. The conformational dynamics of the β-helices drive ion transport, and conduction properties are affected by its interactions with the surrounding lipid.^324–326 Inspired by β-helix structure of gramicidin, early peptide designs focused on synthetic polypeptides that formed similar helical architectures that open to allow for the permeation of ions.^{327, 328} Taking a different approach, Ghadiri and colleagues similarly designed cyclic tetra- and hexapeptides that assemble into cation-selective channels.³²⁹

Unlike gramicidin, alamethicin, an antibiotic peptide from Trichoderma viride, forms “barrel stave” pores lined by α-helices, which more closely resemble ion channel proteins.³⁰⁰ This 20-residue peptide consisting of interspersed α-aminoisobutyric acid residues is postulated to associate into n = 4–8 α-helical bundles with its polar face, which contains the Gln and Glu residues, pointed to the center of the pore.^{73, 300, 330–337} While a structure exists for the monomer, there is no structure available for the alamethicin channel.³³⁸ Nevertheless, MD simulations of various oligomeric states (n = 5–8) reveals stability in all of the oligomeric states which corroborates findings on the multiplicity of conductance levels observed for this channel.³³⁹ Also, a lower-resolution model of the channel has been determined by atomic force microscopy and is consistent with the barrel stave model.³⁴⁰

M2, an example from nature that illustrates the requirements for selective proton transport.

The proton channel from influenza A virus is a small membrane protein, and the founding member of the viroporin family or proteins.^341–346 M2’s proton channel is essential to the life cycle of the virus; it mediates the acidification of the interior of the virus,^347–349 necessary for release of viral RNA. Following the discovery of M2, a large number of viruses have been found to also express small membrane proteins, many of which have ion channel activity.^341–346 Most topically, the structure of the E-protein, a virporin from SARS-CoV-2 has been solved at moderate resolution by solids NMR,³⁵⁰ and is an area of very active investigation.

The 24-residue TM helix of the M2 proton channel from the influenza A virus is the minimal construct of the M2 polypeptide that still retains its proton channel activity.³⁵¹ Like M2 it assembles into a parallel homotetramer with a water-filled channel lumen composed of apolar residues that leads to the proton-shuttling His37 tetrad (Figure 5). Very high-resolution crystal structures of M2 (<1.1 Å),^{294, 352, 353} together with solution³⁵⁴ and solids NMR structures,^{355, 356} have revealed the interactions that stabilize the protein, and the path taken by protons passing through the channel. Protons enter through a narrow hydrophobic sphincter and then travel along “water wires” to protonate a critical His37 residue. High-resolution crystallographic structures^{352, 353} reveal that the backbone carbonyls of the pore-facing hydrophobic residues, along with the interlumenal waters, constitute the extensive hydrogen-bonding network that both lowers the energy barrier for conductance and stabilizes excess positive charge during proton conduction.³⁵⁷ Based on different solids NMR measurements there is currently an unresolved question of whether or not the His37 residues are hydrogen bonded to one another.^358–360 In every independent crystallographic or solution NMR structure of the channel that we have solved, the His37 residues are either fully hydrated in one conducting state or indirectly involved in edge-on aromatic interactions with strong water-mediated hydrogen bonds in a second conformational state,^{294, 352, 353, 361–365} which agrees well with the model of M. Hong.³⁵⁸ The conformation of the protein varies with the degree of protonation of His37, and in the proton-conduction cycle, this region of the channel becomes more and less open as protons come on and off.³⁶⁶ Voth’s group has used classical MD, reactive MD, and QM calculations to evaluate the conduction mechanism, which are able to quantitatively predict the conductance of the channel as well as the mechanism of binding to small molecule drugs.^{357, 367–369}

Fig. 5: Key structural features that define the proton channel activity for M2.

The TM domain of M2 forms a proton-conductive channel. Here, we see the drug, amantadine, binding to the channel in the hydrophobic gasket with its ammonium group pointed towards the water-filled pore. The ammonium group acts as a proton and locks the interlumenal waters near the gating His in place, essentially blocking proton conduction.

Just below the His37 residue, a ring of four Trp41 residues form tight aromatic–aromatic or aromatic–cation interactions with His37, which is further stabilized by a solvent-mediated hydrogen bond to Asp44.³⁷⁰ This gate is interrupted when the His37 residues reach a critical protonation state of approximately +3 and a proton can move into the interior. Thus, M2 has a transporter-like mechanism.^{351, 371–374}

The contribution of each residue to the free energy of tetramerization of the channel has been evaluated in a series of mutants in micelles and bilayers. No single residue is absolutely required for tetramerization, although some substitutions result in a loss of thermodynamic stability,^{295, 375, 376} particularly when Ala is substituted for His37. On the other hand, these mutants have substantially altered conductance and/or proton selectivity.²⁹⁶ It is noteworthy that all the hydrogen bonds within the TM region are water-mediated rather than direct, with the only direct interaction being a salt bridge at the hydrated exit of the channel. Even the polar ammonium group of channel-blocking drugs form indirect, water-mediate hydrogen bonds to the channel (Figure 5).^{362–364, 377} This finding underscores the importance of considering water in the stability and function of this channel.

Finally, no discussion of M2 would be complete without mentioning some early debates, which are now resolved. Early work questioned whether a second, cytoplasmic helix was needed for the structural integrity of the protein, but this was addressed thoroughly by deletion mutagenesis of M2 expressed in mammalian cells³⁵¹ and by solving crystallographic³⁶² and NMR structures³⁷⁸ with and without C-terminal extensions. Instead, the C-terminal cytoplasmic helix was found to be important for budding of the virus.^{71, 379} Also in early work, Chou et al. suggested that the channel-blocker amantadine (which was clearly observed in a crystal structures^{361, 364, 377}) might instead bind on the outside of the channel.³⁵⁴ This debate was also resolved when Chou was able to solve a structure with drug bound in the interior pore.³⁸⁰

De novo designed ion channels.

Designed peptide channels offer the opportunity to decipher the complexities of channel function in simple, minimalist systems. Beyond the considerations for peptide–peptide and peptide–membrane interactions in the design of membrane-soluble helical bundles, in designing functional ion channels, we need to account for water- and ion-accessibility of the pore. In the late 1980s, DeGrado and coworkers designed the first TM α-helical peptide channel using a repeating sequence of Leu and Ser: H₂N-(Leu-Ser-Leu-Leu-Leu-Ser-Leu)₃-CONH₂ and H₂N-(Leu-Ser-Ser-Leu-Leu-Ser-Leu)₃-CONH₂, or LS2 and LS3, respectively.^{313, 316} The Leu residues were chosen to pack well at helix–helix interfaces, while the Ser residues are important for both packing and the stabilization of a proton-conducting pore. The packing of small Ser residues near the core and larger Leu sidechains at the helical interfaces dictated a tetrameric arrangement for the proton channel, LS2. The Ser hydroxyls, together with water molecules, appeared to create a proton conduction pathway via a water-hopping mechanism. LS3 formed hexameric channels with a pore large enough to accommodate a solvated ion in the hexameric bundle. While crystallographic structures were not available at the time, a large body of subsequent data, including the incorporation of LS2 peptides onto a rigid tetrameric scaffold,³⁸¹ supported the hypothetical structures and conduction model.^{313, 314, 316, 382–384}

Recently, Woolfson and colleagues approached the design of α-helical channels by converting water-soluble coiled coils into membrane-spanning assemblies.³⁸⁵ Unlike previous designs which borrowed from the D4 domain of E. coli polysaccharide transporter Wza,³⁸⁶ these designs were purely de novo. Their work began with a careful consideration of the features required to pack helices into various sized helical bundles of various sizes.³⁸⁷ They showed that the packing of small residues at positions designated “e” and “g” of a coiled coil dictate the formation oligomeric bundles with sizes ranging from 5 to 8 helices, depending on the nature of the small residues and other features. In the initial work, the residues in the core at positions “a” and “d” were large Leu and Ile residues, respectively. They reasoned at least some of these Leu and Ile residues could be changed to combinations of Ser and Thr to form water-filled pores, which was validated by crystal structures of the water-soluble peptides. Next, they converted the peptides to membrane peptides by introducing apolar residues at the exterior positions.³⁸⁵ Single-channel recordings were consistent with the designs, suggesting that the designed structures had been realized. However, the peptides failed to crystallize in the active channel-forming states. Nevertheless, an unanticipated structure of a non-conducting antiparallel tetrameric state (Figure 4) illustrated important design principles of de novo membrane protein design.

The most ambitious functional membrane protein designed to date is a TM four-helix bundle, Rocker (Figure 6), that transports first-row transition metal ions Zn²⁺ in exchange for protons.^{299, 318} Rocker has two 4Glu, 4-His Zn²⁺ binding sites, which can alternately accommodate protons and metal ions. A computational design algorithm was used to stabilize two energetically degenerate asymmetric states of the protein while destabilizing a competing fully symmetrical state which might otherwise bind metal ions too tightly and impede motions required for ion transport. The computed TM bundle formed a dimer of dimers with two non-equivalent helix–helix interfaces (Figure 6). A “tight interface” had a small inter-helical distance (8.9 Å) stabilized by efficient packing of small Ala residues in an Ala-coil motif, which are seen in the crystal structure of the protein. The “loose interface” had a larger interhelical distance of 12.0 Å and was less well packed. Solids NMR was thus used to define the structure of the dimer of dimers. The resulting membrane-spanning four-helical bundle transported first-row transition metal ions Zn²⁺ and Co²⁺, but not Ca²⁺ across membranes. Vesicle flux experiments show that as Zn²⁺ ions diffuse down their concentration gradients, protons were antiported. These experiments illustrate the feasibility of designing membrane proteins with predefined structural and dynamic properties. These studies illustrate the how multiple forces, including structural motifs and diverse polar interactions can be combined to build functional assemblies.

OUTLOOK

This review has given only the smallest glimpse of the vibrant and growing field of peptide–membrane interactions. Peptides have proven to be outstanding systems for understanding larger proteins, particularly for deciphering the rules of membrane peptide/protein association, insertion, and folding. This field has also spawned antimicrobial peptides and peptide mimetics currently in the clinic, while also inspiring new areas of antimicrobial polymer science.^{103, 106, 388–395} Clearly, the future is very bright.