The mechanisms of integral membrane protein biogenesis

Abstract

Roughly one-fourth of all genes code for integral membrane proteins that are inserted into the plasma membrane of prokaryotes or the endoplasmic reticulum of eukaryotes. Multiple pathways are used for the targeting and insertion of membrane proteins based on their topological and biophysical characteristics. Proteins that span the membrane multiple times face the additional challenges of intramembrane folding, and in many cases, assembly with other subunits. Recent biochemical and structural analyses have provided considerable clarity into the molecular basis of membrane protein targeting and insertion, with tantalising new insights into the poorly-understood biogenesis of multi-pass membrane proteins.

INTRODUCTION

A fundamental feature of all life is a plasma membrane that separates the intracellular and extracellular environments. Transport of metabolites and communication of information across otherwise impermeable membranes is mediated by integral membrane proteins. These membrane-spanning proteins represent ~25% of protein-coding genes in all organisms and they serve numerous crucial functions including ion and nutrient transport, signalling, pathogenesis, defence, and adhesion¹. Biogenesis of membrane-spanning proteins is therefore one of the most ancient biological processes, and the core machineries that mediate these processes are exceptionally broadly conserved.

In bacteria and archaea, which typically lack intracellular membrane compartments, membrane proteins are inserted directly into the plasma membrane where they function². Eukaryotes insert the vast majority of their membrane proteins into the endoplasmic reticulum (ER) membrane, where they fold and assemble before trafficking to their final destination³. Because the eukaryotic ER is evolutionarily derived from the prokaryotic plasma membrane⁴, the respective insertion machineries are related and share key mechanistic principles. The machineries for insertion into mitochondrial, chloroplast and peroxisomal membranes are reviewed elsewhere^5–9.

The membrane-spanning segments of a protein are typically alpha-helical, allowing the hydrophilic amide backbone of the protein to be shielded by hydrogen bonding from the hydrophobic membrane interior¹⁰. A second strategy to accomplish the same goal is by assembly of a beta-barrel. This class of membrane proteins, found in the outer membranes of bacteria, mitochondria, and plastids, is inserted by a unique machinery that is reviewed elsewhere^11,12. Here, we discuss alpha-helical membrane protein biogenesis at the eukaryotic ER and evolutionarily related prokaryotic plasma membrane.

Membrane protein biogenesis can be divided into four processes (Figure 1a). The first is targeting, the process by which a nascent protein is delivered to the membrane where it will be inserted. The second is membrane insertion of transmembrane domains (TMDs) in the appropriate topology, retaining some parts of the protein in the cytosol while translocating other parts across the membrane. In addition to these universal steps applicable to even the simplest membrane proteins, most require additional folding steps in the membrane, and many are assembled with obligate interaction partners. Although these steps of membrane protein biogenesis are typically studied separately, they are intimately linked and often occur simultaneously as the protein is being synthesised.

Fig. 1. |. Overview of integral membrane protein biogenesis.

a | The four major steps involved in membrane protein biogenesis. In this and subsequent figures, tapered extensions that flank TMDs indicate soluble regions of indeterminate length. When the flanking segment is necessarily short, such as the translocated segment of a tail-anchored membrane protein, a non-tapered line is used. An ellipsis is used to indicate further polypeptide that contains additional TMDs. b | The major classes of integral membrane proteins are indicated, with prominent examples from human cells listed below.

A full understanding of membrane protein biogenesis requires an appreciation of membrane protein diversity. The human genome contains ~5,000 integral membrane proteins whose ~20,000 TMDs vary widely in sequence, biophysical properties, location(s) within the protein, and topology (Figure 1b). A major consequence of this diversity is that the machinery tasked with membrane protein biogenesis must be highly accommodating. This is accomplished by different pathways for different classes of substrates (also termed clients) based on topology, biophysical property, or another distinctive feature. Thus, although the multiple often similar pathways in each following section can seem bewildering, their co-existence can be rationalised by the diversity of membrane protein substrates.

In this Review, we discuss the four basic steps of membrane protein biogenesis, highlighting the different mechanisms tailored for different types of substrates. Much of our focus will be on the mechanistic basis of membrane protein insertion and folding. These areas have recently seen the discovery of new machinery, mechanistic insights from structural and bioinformatics analysis, and the emergence of new concepts. The earlier step of targeting, a comparatively mature area, is summarised in less detail with references to other focused reviews. The final step of assembly, a poorly studied problem, is necessarily more speculative. In each section, we attempt to synthesise knowledge into a few major concepts and highlight the most pressing directions for future work.

TARGETING OF MEMBRANE PROTEINS

Membrane protein targeting to the eukaryotic ER or bacterial plasma membrane is generally mediated by the most N-terminal hydrophobic domain within the substrate. The targeting sequence can either be a cleavable signal peptide, typically at the N-terminus, or the first TMD anywhere within the polypeptide (Figure 1b). The hydrophobic domain of signal sequences is ~7–9 amino acids¹³, whereas TMDs are ~15–25 amino acids; both are widely variable in sequence and hydrophobicity¹⁴. This large diversity in targeting sequence location and biophysical properties means that a single targeting pathway or mechanism cannot accommodate all membrane proteins.

Three established targeting pathways are used depending on the position and hydrophobicity of a substrate’s targeting sequence (Figure 2a). A targeting sequence located at least ~65 amino acids from the substrate’s C-terminus is required for co-translational targeting by the signal recognition particle (SRP)^15–17. The basis of this distance constraint is SRP’s site of targeting sequence recognition at the mouth of the ribosome exit tunnel^18–20, which is positioned ~35 amino acids from the peptidyl-transferase centre deep inside the ribosome’s large subunit (Figure 2b). After recognition, SRP-mediated targeting to the ER takes ~5–7 seconds²¹, during which another ~30 amino acids can be synthesised. Co-translational targeting only occurs if translation does not terminate during this period, explaining why SRP requires targeting signals to be relatively far from the C-terminus.

Fig. 2. Membrane protein targeting to the endoplasmic reticulum.

| a | Targeting-sequence position and hydrophobicity influences the route taken to the ER membrane. b | Diagram showing the position of SRP on a translating ribosome. c | Histogram of all 235 predicted human TA proteins plotted by the hydrophobicity of their TMDs calculated using the Kyte-Doolittle (K-D) scale¹⁷⁹. The model EMC- and GET-dependent proteins SQS and VAMP2, respectively, are within the bins indicated. SEC61B can use either pathway, a property that seems to be shared by most TA proteins. d | Model illustrating the sequence of events that determine which targeting pathway is used by which type of substrate. Dedicated cytosolic targeting factors are yellow, general TMD-binding chaperones are green, and the initial interaction partner at the ER membrane in dark grey.

Membrane proteins whose sole targeting sequence is a TMD closer than ~65 amino acids from the C-terminus are termed tail-anchored (TA) proteins²², and they are targeted post-translationally using either general or specialised cytosolic chaperones. The TMDs of TA proteins vary widely in hydrophobicity (Figure 2c), a key feature that determines their mechanism of targeting²³. Those of high hydrophobicity, such as the protein VAMP2, are targeted by the ‘guided entry of TA protein’ (GET) pathway. Those of low hydrophobicity, such as SQS, instead use cytosolic chaperones and the ER membrane complex (EMC) for their biogenesis. Most TA proteins (such as SEC61B) can probably use either route for membrane insertion. We discuss the targeting pathways that utilise SRP, GET, and EMC in turn. Other poorly-studied potential targeting routes are discussed in Box 1.

Box 1: Other potential targeting pathways.

Most membrane proteins in bacteria are targeted co-translationally by SRP³⁵ or post-translationally using chaperones^63,64. However, some membrane proteins seem to be co-translationally targeted directly to SecY by the bacteria-specific ATPase SecA^180,181. The membrane protein is then presumably inserted via SecY’s lateral gate, with one flanking region being translocated through the SecY channel. The features that determine whether a membrane protein is recognised by SRP or SecA probably involve a combination of the TMD and its flanking regions¹⁸². Because SecA has a binding site for the signal sequences of secreted proteins¹⁸³, it is possible that low-hydrophobicity TMDs that resemble signal sequences are recognised similarly. As with secreted proteins¹⁸⁴, SecA might use repeated cycles of ATP binding and hydrolysis to push the translocated portion of a membrane protein through the central channel of SecY. Precisely how SecA-bound ribosomes are transferred to SecY remains unclear. The considerable overlap in the ribosome binding site of SecA and SecY suggests that handover may occur via a concerted mechanism similar to that observed with SRP and SecY¹⁸². As in eukaryotes, it seems that multiple targeting pathways are needed to accommodate the topological and biophysical diversity of the bacterial membrane proteome.

In yeast, some secretory or lipid-anchored proteins that normally engage SRP in wild type cells can nonetheless be targeted and translocated in SRP-lacking cells¹⁸⁵. Using a whole-genome high-content localisation screen of one such protein led to the definition of an SRP-independent (SND) targeting pathway composed of three interacting factors: SND1 in the cytosol and two ER membrane proteins SND2 (TMEM208 in humans) and SND3¹⁸⁶. Based primarily on synthetic genetic interactions, the SND pathway was proposed to target both SRP- and GET-pathway substrates to the Sec complex (composed in yeast of the Sec61 complex, Sec62, Sec63, Sec71, and Sec72) for translocation and membrane insertion. The membrane protein substrates of the SND pathway were proposed to be those whose first TMD is far from the N-terminus, but not so distal as to be tail-anchored. Because such TMDs would emerge from a translating ribosome, they should be recognised by SRP. It is mechanistically unclear why they seem to need a separate pathway. Additional work is required to assign specific roles to each SND factor, determine the mechanism(s) of substrate engagement by this pathway, determine whether the pathway operates co- or post-translationally, and clarify the role of the human SND2 ortholog^186,187.

The substrate ranges covered by the SRP, GET, and EMC targeting pathways are challenging to define precisely. Due to substantial overlap in their respective substrate ranges, deletion of any pathway has a strong impact only on the subset of substrates that cannot be accommodated by any other pathway. For this reason, the set of substrates whose biogenesis is impaired in the absence of a targeting factor is often only a small subset of the set of substrates that normally engage that factor. Thus, many substrates that are “SRP-independent” or “GET-independent” may normally engage the SRP or GET pathways. A pathway’s substrate range is best described by direct physical interaction analysis in unperturbed cells when all pathways are available. Although this has been achieved for SRP^35,36, it remains an important goal for GET, EMC, and SND pathways. Recognising what cannot be handled by known pathways provides substrates that can be used to discover new pathways.

Co-translational targeting by the signal recognition particle

During co-translational targeting, hydrophobic targeting sequences are recognised by SRP at the mouth of the ribosome exit tunnel and delivered to the SRP receptor (SR) at the ER membrane (Figure 2d, top row). The cycle of substrate recognition, targeting to SR, and recycling of SRP back to the cytosol is regulated through the cycle of GTP binding and hydrolysis by two universally conserved GTPases, one in the SRP54 subunit of SRP and the other in the alpha subunit of SR. The structures and molecular details of the SRP system have been extensively reviewed in detail elsewhere^15–17. Here, we consider comparatively recent new insights into the basis of SRP substrate range and specificity.

The minimum requirement for SRP recognition is a ~7 amino acid mostly hydrophobic sequence. Structures of the substrate-binding domain of SRP54 without and with substrate in prokaryotes^24–26 and eukaryotes²⁰ illustrates a conserved hydrophobic groove that explains SRP’s binding preference. Cryo-EM analysis of native mammalian SRP just prior to and immediately after substrate engagement suggests that the substrate-binding groove is normally auto-inhibited by a C-terminal amphipathic helix in SRP54²⁰. When a signal peptide or TMD emerges from the ribosome exit tunnel, its high local concentration near SRP54 displaces the auto-inhibitory helix, which then serves as a lid to help enclose substrate within the groove. Auto-inhibition may minimise promiscuous binding to non-substrates on and off the ribosome, thereby enhancing SRP specificity for bona fide signal sequences and TMDs emerging from a translating ribosome.

Promiscuous recognition of less hydrophobic sequences is also antagonised by the nascent polypeptide associated complex (NAC)^27–29, a highly abundant cytosolic factor that binds near the ribosome exit tunnel of most or all ribosomes³⁰ (Figure 2d). In the absence of NAC in vitro, SRP can bind to ribosomes translating proteins lacking a signal peptide or TMD, leading to their inappropriate targeting to the ER^{27,28,31–34}. Molecular insight for how NAC and SRP coordinate at the ribosome exit tunnel²⁹ awaits structural analysis. The biological importance of NAC for ensuring targeting specificity is highlighted by extensive mis-targeting of numerous proteins to the wrong cellular location, organelle stress, and lethality in C. elegans depleted of NAC²⁸. Several quality control pathways monitor failures in targeting, and these are discussed in Box 2.

Box 2: Quality control of failed membrane proteins.

The biogenesis of membrane proteins can fail at any of various steps ranging from its translation, targeting, insertion, folding and assembly. Each step is monitored by a combination of both general and TMD-specific quality control pathways. Incomplete translation due to a stalled ribosome is recognised by the ribosome-associated quality control (RQC) pathway [reviewed elsewhere^188–191]. RQC serves to target the partially synthesised protein for ubiquitin-mediated degradation, recycle or degrade the ribosome, and in some cases, degrade the mRNA. This pathway operates in the cytosol and at organelle membranes, including the ER.

Failure at the targeting or insertion step can lead to a membrane protein that is mis-localised to the cytosol. Several quality control factors deal with mis-localised membrane proteins depending on features of their TMDs. The cytosol contains several TMD-binding factors that also interact with E3 ubiquitin ligases. This includes BAG6, which associates with the ubiquitin ligase RNF126^192,193, and the Ubiquilin family of proteins that interact with yet unidentified E3 ligases⁶⁰. Mammals have four Ubiquilin family members, perhaps with different substrate ranges than each other and from BAG6, which seems to prefer particularly hydrophobic TMDs that normally use the GET pathway. Yeast do not have an obvious BAG6 homolog, but contains the Ubiquilin family member Dsk2 that might have a similar role. In addition, mis-localised membrane proteins can be directly recognised by the ER-resident E3 ligases Doa10 in yeast^194,195 and perhaps MARCH6 and TRC8 in mammals^196,197.

In addition to cytosolic mis-localisation, membrane proteins can be mis-inserted into the wrong organelle or in the wrong orientation. For example, the TMDs of nuclear-encoded mitochondrial membrane proteins can be similar to ER-destined membrane proteins. Similarly, many N_cyt signal anchors and signal sequences are similar to EMC-dependent N_exo signal anchors, differing primarily in their TMD-flanking charged residues. Thus, it seems likely that EMC sometimes inadvertently inserts terminal TMDs intended for mitochondria or N_cyt orientation. Recently, the P5A-ATPase family member ATP13A1 (Spf1 in yeast) was found to dislocate moderately hydrophobic TMDs containing flanking positive charge(s) facing the ER lumen¹⁹⁸. The mitochondrial outer membrane contains an analogous (although mechanistically distinct) ATP-dependent TMD dislocase termed ATAD1 (Msp1 in yeast)^199–201. Thus, TMD mis-targeting and mis-orientation appear to be sufficiently frequent occurrences to warrant robust and highly conserved quality control pathways for their mitigation.

Finally, membrane proteins that are inserted but fail to fold or assemble properly are recognised by organelle quality control pathways. At the ER, these pathways are collectively termed ER-associated degradation (ERAD) [reviewed elsewhere^202,203]. One mechanism of membrane protein recognition in ERAD seems to involve TMDs that expose hydrophilic side chains to the membrane. When correctly folded or assembled, such hydrophilic residues are shielded from the membrane, perhaps explaining why their persistent exposure is a reliable indicator of failed biogenesis. This mechanism of recognition is conceptually similar to how exposure of hydrophobic patches in the aqueous environments of the cytosol or ER lumen is used as a cue for soluble protein misfolding²⁰⁴.

Transcriptome-wide footprinting of SRP-engaged ribosomes has verified conclusions derived from biochemical and structural analyses of SRP’s client preference. In bacteria, SRP typically engages the first TMD of membrane proteins³⁵, whereas in yeast, SRP engages both TMDs and cleavable signal sequences³⁶. The analysis of mRNA delocalisation from the ER upon rapid SRP depletion in yeast shows that the set of delocalised mRNAs generally matches those engaged by SRP in profiling studies³⁷. These results all arrive at the consistent conclusion that in eukaryotes, SRP engages and mediates the co-translational targeting of ribosomes that display a signal peptide or TMD at the exit tunnel. In prokaryotes, SRP’s substrate preference seems to be shifted toward higher hydrophobicity, making it more specific for TMDs than signal peptides^35,38.

Post-translational targeting by the GET pathway

TA proteins of moderate to high hydrophobicity are recognised and targeted by the GET pathway [reviewed in detail elsewhere^39–41] (Figure 2d, middle row). Here we use the nomenclature originally described for components of the yeast pathway, which has recently been applied to other organisms. Specificity of the GET pathway for hydrophobic TMDs is imparted by the homo-dimeric ATP-dependent chaperone GET3^42–44. The TMD-binding site spans the GET3 dimer and is a deep ~35 Å long hydrophobic groove that can house an α-helix of ~20 amino acids^45–47. GET3 is restricted to C-terminal TMDs because SRP’s ribosome association ensures that SRP has priority for non-C-terminal TMDs that emerge from the ribosome during translation. Thus, GET pathway substrates are defined by a combination of negative selection of non-TA proteins by SRP and positive selection of high hydrophobicity TMDs by GET3. Proteins that engage GET3 are targeted to the ER via a receptor composed of GET1 and GET2^48,49.

Loading of substrates onto GET3 occurs within a pre-targeting complex that additionally contains the TMD-binding chaperone SGTA (Sgt2 in yeast) and the GET4-GET5 complex (which in metazoans also contains the quality control factor BAG6)^43,44,50,51. SGTA engages a TMD shortly after nascent protein release from the ribosome^{43–45,51,52}. TMD capture might occur at the ribosome surface near the exit tunnel, where the yeast Get4-Get5 complex was recently found to bind⁵², consistent with the ribosome-binding capacity of mammalian GET4-GET5-BAG6 complex observed earlier⁴⁴. Dynamic binding of SGTA to GET5⁵³would allow it to sample the ribosome exit tunnel region to capture TMDs there. SGTA can also bind TMD substrates released into the cytosol^45,51 or acquire them from HSP70 family members via an HSP70-SGTA interaction⁵⁴. Although most HSP70 clients are soluble proteins engaged in folding, these are not transferred to SGTA because the latter has preference for longer hydrophobic segments that typify TMDs^43,55. Structural modelling of the SGTA family suggests its substrate binding site is more hydrophobic and larger than in HSP70, but smaller than in GET3⁵⁶.

SGTA-associated TMDs are then transferred to GET3 within the pre-targeting complex in a concerted handover reaction^43,50,51,57. TMDs that do not match GET3’s preference for high hydrophobicity are not transferred^43,55, providing a potential mechanism to prevent loading of mitochondrial membrane proteins whose TMDs are typically less hydrophobic. Thus, a series of chaperones with preference for increasingly longer and hydrophobic sequences (GET3>SGTA>HSP70) effectively “filters” potential substrates, loading onto GET3 only those TMDs of high hydrophobicity intended for the ER. Lower hydrophobicity TMDs remain on SGTA or engage some other cytosolic chaperone, as discussed next.

Chaperone-facilitated targeting

TA proteins with TMDs of low hydrophobicity probably get captured by SGTA similarly to TA proteins of the GET pathway^23,43. This is because SGTA has a broader substrate range and can bind TMDs that are shorter and less hydrophobic than GET3^51,55,56. However, these TMDs are not loaded effectively into GET3’s hydrophobic groove²³. Substrate binding by SGTA is highly dynamic⁵¹, and upon release, the TMD of such TA proteins seems to be recognised directly by EMC at the ER membrane^23,58,59. Until their recognition by EMC, these proteins are prevented from aggregation by dynamic binding and release from SGTA or other TMD-binding chaperones (Figure 2d, bottom row). Such chaperones are considered ‘targeting-permissive’ factors and include calmodulin^3,23, members of the UBQLN family⁶⁰, and heat shock proteins^61,62, the latter of which may serve a similar role in bacteria^63,64.

The lack of requirement for a dedicated targeting pathway may reflect the lower hydrophobicity of these proteins. This means they are less aggregation prone and can be kept soluble by abundant chaperones that dynamically bind and release them in the cytosol. Chaperone binding prevents inappropriate interactions while chaperone release allows opportunities for targeting via direct ER engagement. Dynamic substrate release from the chaperone is crucial for recognition by membrane factors because impeding release impairs targeting and membrane insertion²³. A conceptually similar mechanism is currently thought to facilitate mitochondrial and chloroplast targeting of nuclear encoded proteins^65–67.

TRANSMEMBRANE DOMAIN INSERTION

Once a polypeptide arrives at the membrane, its TMD(s) need to be inserted into the lipid bilayer. The partitioning of a hydrophobic segment of polypeptide into the hydrophobic membrane is an energetically favourable reaction^68–70. The two critical impediments to this process are off-pathway interactions prior to insertion and translocation of TMD-flanking hydrophilic polypeptide across the bilayer (Figure 3a, left). Off-pathway interactions such as aggregation are minimised during substrate delivery to the membrane by targeting factors and chaperones as discussed above. The second barrier, translocation of TMD-flanking hydrophilic domains concomitant with TMD insertion, is the key role of TMD insertion factors.

Fig. 3. Membrane protein insertion at the endoplasmic reticulum.

| a | Comparison of unassisted, insertase-mediated, and channel-mediated TMD insertion. b | Schematic of the open Sec61 complex containing an aqueous channel for protein translocation and a lateral gate for membrane insertion of hydrophobic segments. c-g | Key steps during insertion of a Type I membrane protein by Sec61 (c), Type II membrane protein by Sec61 (d), TA protein by the GET complex (e), TA protein by EMC (f) and Type III membrane protein by EMC (g).

The ER contains at least four insertion factors: the Sec61 complex, the GET1-GET2 complex, EMC, and the TMCO1 complex. The latter three factors contain subunits (GET1, EMC3, and TMCO1) that were recently appreciated to be evolutionarily and structurally related to each other and to members of the Oxa1 family of protein insertases^71,72. These collectively form the Oxa1 superfamily, an ancient class of proteins having arisen prior to the divergence of archaea and bacteria, and hence long before eukaryogenesis. Thus, its members are found in the bacterial and archaeal plasma membranes, the inner membranes of mitochondria and plastids (which are evolutionarily related to inner membrane of endosymbiont bacteria from which they evolved), and the eukaryotic ER.

As argued in detail later, all Oxa1 superfamily members may share a core mechanism of TMD insertion involving a hydrophilic vestibule to aid flanking domain translocation (Figure 3a, middle). This insertion-coupled translocation reaction is typically restricted to less than ~50 amino acids because Oxa1 superfamily proteins lack a membrane-spanning channel⁷³. By contrast, an aqueous channel housed in the Sec61 complex allows it to translocate hydrophilic polypeptide segments of unlimited length⁷⁴ (Figure 3a, right). This is why TMDs flanked by long hydrophilic regions use the Sec61 complex for their insertion. This basic concept, that short TMD-flanking domains are translocated via a hydrophilic vestibule within an Oxa1 superfamily member and long TMD-flanking domains are translocated through the Sec61 channel, is a useful to keep in mind as each insertion pathway and their respective substrates are described in turn below.

TMD insertion by the Sec61 complex

The Sec61 complex (termed the SecY complex in bacteria and archaea) is a universally conserved protein conducting channel used for both secretion and membrane insertion of proteins [reviewed extensively elsewhere⁷⁴]. The channel is housed in the largest subunit, SecY or Sec61α, with two small subunits located peripherally. The Sec61 complex is able to open axially across the membrane for polypeptide translocation and laterally into the membrane for TMD insertion (Figure 3b). Structures of SecY and Sec61 show it to be a pseudosymmetric membrane protein with its N- and C-terminal halves surrounding a central hourglass-shaped pore^75,76. The two halves come together like a clamshell, with a hinge in the back and a frontside lateral gate that can open toward the membrane. In the inactive state, the central pore is occluded by a short helix known as the plug.

Inactive and closed Sec61 can be opened for translocation when a signal peptide (or TMD) binds to and parts the lateral gate^77,78. Lateral gate opening leads to plug displacement, creating an open conduit across the membrane. The signal peptide binds to Sec61 with its N-terminus facing the cytosol and eventual cleavage site facing the lumen (Figure 3c). This position and orientation causes the polypeptide downstream of the signal sequence to be pulled into Sec61’s pore. From this point, further translational elongation results in translocation, and the hydrophobic signal peptide diffuses into the membrane where it is cleaved by the lumen-oriented active site of signal peptidase⁷⁹.

TMDs can be inserted into the membrane by Sec61 in two modes. In the first mode, a cleavable signal sequence has already threaded the polypeptide through the central pore within Sec61 when a TMD emerges from the ribosome (Figure 3c). The TMD therefore moves into an open Sec61 channel, where it can access the lipid bilayer via the lateral gate. It is thought that a dynamic lateral gate that is constantly sampling the open conformation allows the nascent TMD to partition into the membrane based on its hydrophobicity^10,80. Proteins that initiate insertion this way are called Type I membrane proteins (see Figure 1b) and represent roughly one-third of all membrane proteins made at the ER.

In the second mode, a TMD engages, opens, and passes through the Sec61 lateral gate using the mechanism described above for a signal sequence⁸¹ (Figure 3d, compare to Figure 3c). Hence, the N-terminal flanking domain is retained in the cytosol, the C-terminal flanking domain is threaded through the central channel, and the TMD is positioned at and subsequently passes through a parted lateral gate. Proteins that initiate their insertion by this mechanism are termed Type II membrane proteins (see Figure 1b), with their first TMD termed a Type II or N_cyt signal anchor (meaning the N-terminus faces the cytosol).

Proteins that initiate their membrane insertion using a TMD preceded by less than ~50 amino acids are called Type III membrane proteins (see Figure 1b), with their first TMD termed a Type III or N_exo signal anchor. Almost two-thirds of all membrane proteins begin with either a N_exo or N_cyt signal anchor. N_exo signal anchors have long been thought to insert via Sec61’s lateral gate similarly to N_cyt signal anchors⁸². However, Sec61 lateral gate inhibitors that block signal sequences and N_cyt signal anchors do not inhibit the insertion of N_exo signal anchors^83–85. More strikingly, immunodepletion of Sec61 had little effect on the insertion of all tested N_exo signal anchors, yet completely precluded signal sequence and N_cyt signal anchor function⁸⁶. These observations suggest that N_exo signal anchors are inserted by a qualitatively different mechanism than either signal sequences or N_cyt signal anchors. As discussed below, recent findings indicate that N_exo signal anchors can be inserted by EMC, providing one explanation for these otherwise unexpected findings⁸⁶.

Finally, several proteins are associated with eukaryotic Sec61 whose role in TMD insertion remains unclear. The translocon-associated protein (TRAP) complex and the translocating chain-associated membrane protein (TRAM) are found in many eukaryotes and facilitate co-translational Sec61 gating by weakly hydrophobic signal sequences in a mammalian in vitro system^87,88. A similar function is ascribed to the Sec62-Sec63 complex during post-translational signal sequence recognition^89–91. Of these factors, TRAM has also been observed to crosslink with low-hydrophobicity TMDs during or shortly after insertion^92,93, but a functional role has not been demonstrated. Some or all of these factors might perhaps assist insertion of certain TMDs similarly to signal sequences, but this has not been studied.

TA protein insertion by the GET complex

Membrane proteins engaged by GET3 are delivered to an ER-localised receptor composed of GET1 and GET2, both of which are three-TMD proteins containing cytosolic GET3-binding domains⁴⁸. Reconstitution experiments using yeast components have rigorously established that the Get1-Get2 complex is both necessary and sufficient for insertion of Get3-targeted TA proteins^94–96. The insertion reaction involves three main steps (Figure 3e): (i) initial engagement of the TA-Get3 targeting complex by the Get1-Get2 receptor; (ii) receptor-mediated release of TA protein from Get3; (iii) TMD insertion into the lipid bilayer.

Targeting complex engagement and TA protein displacement are mediated by the cytosolic domains of Get1 and Get2, both of which have partially overlapping binding sites on Get3^94,97. Because one TA protein binds to a Get3 homo-dimer⁴⁵, there are two potential binding sites on the TA-Get3 targeting complex. The binding of Get2’s cytosolic domain does not disrupt the TA-Get3 interaction, whereas the Get1 cytosolic domain can dislodge TA proteins from Get3^94,96–98. It is therefore clear that GET1 and GET2 collaborate to mediate targeting and targeting complex disassembly, although the precise order of interactions with GET3 is not resolved. In one model (shown in Figure 3e), GET2, whose GET3 binding domain resides on a long flexible tether, engages GET3 initially, brings the complex close to GET1, which then engages the other binding site on GET3 to release TA protein.

After TA protein release, the TMD is inserted into the lipid bilayer in a reaction that requires the TMD regions of the GET1-GET2 complex⁹⁵. A recent cryo-EM structure of the human GET1-GET2-GET3 complex observed a homo-dimer of the GET1-GET2 heterodimer⁹⁹. Notably, GET1 has the architecture and cytosol-facing hydrophilic vestibule predicted from its assignment as an Oxa1 superfamily member⁷¹. The stoichiometry of the GET1-GET2 complex during insertion is currently unclear. Because single Get1-Get2 heterodimers reconstituted into liposomes seem to be sufficient for insertion¹⁰⁰, the simplest model is that a GET3-TMD complex engages one GET1-GET2 heterodimer such that the substrate’s hydrophilic tail can access GET1’s hydrophilic vestibule (Figure 3e). From this intermediate, GET1-mediated dislodging of substrate from GET3 allows its TMD to enter the membrane, with the barrier to hydrophilic tail translocation eased by GET1’s hydrophilic vestibule. A single GET1-GET2 complex mediating insertion would be consistent with the likely stoichiometry of other Oxa1 superfamily members during TMD insertion^{58,59,101–106}.

Terminal TMD insertion by EMC

Several observations had long indicated that the GET pathway was not the only route for TA protein insertion at the ER. In both yeast and mammalian systems, the extent of GET pathway dependence for insertion varies widely among TA proteins¹⁰⁷. Mechanistically, TMD hydrophobicity of many natural TA proteins can be lower than what is efficiently accommodated by Get3^45,55. At least one explanation for these discrepancies came with the discovery that a widely conserved and large protein complex termed EMC¹⁰⁸ can insert low- and moderate-hydrophobicity TA proteins²³. The observation that many TA proteins have a TMD of intermediate hydrophobicity compatible with either pathway (Figure 2c) may explain why neither EMC nor GET subunits are essential in yeast but are lethal in combination¹⁰⁸.

EMC is an ER-resident 8- or 9-subunit complex (depending on species) with large cytosolic and lumenal domains connected by a 14-TMD membrane domain (reviewed elsewhere^109–111). Purified EMC reconstituted into synthetic liposomes is sufficient to mediate insertion of a TA protein upon release from a chaperone²³ (Figure 3f). The current model is that the cytosolic domain of EMC has one or more transient binding sites for moderately hydrophobic TMDs, from which the membrane can be accessed concomitant with tail translocation.

Recent structures of yeast and mammalian EMC reveal that, as was predicted⁷¹, EMC3 is an Oxa1 superfamily member with a cytosol-facing hydrophilic vestibule. Based on this similarity, one model for TA protein insertion is that this vestibule, near which the membrane might be distorted⁵⁹, facilitates translocation of the C-terminal hydrophilic tail (Figure 3f). Consistent with this idea, mutations in or near this vestibule can impair insertion^59,105,106. EMC was noted to contain a second potential route into the membrane on the opposite side where a shallow cytosolic cradle leads into a membrane-embedded hydrophobic groove^{58,59,105,106}. Having two different routes into the membrane might broaden EMC’s substrate range by accommodating TMDs with different biophysical features, an idea worth exploring in future studies.

In addition to TA proteins, EMC can also mediate co-translational insertion of N_exo signal anchors displayed on translating ribosomes⁸⁶ (Figure 3g). Although N_exo signal anchors are of the opposite topology to TA proteins, they are nonetheless similar in containing a short translocated domain less than ~50 amino acids. In addition, most EMC-dependent N_exo signal anchors also have partial hydrophilic character, similar to EMC-dependent TA proteins. These observations suggest that one general class of substrates for EMC-mediated insertion are TMDs close to either terminus of a protein.

A major unresolved issue with N_exo signal anchor insertion is precisely when EMC acts¹¹⁰. The most likely possibility is after SRP-mediated targeting to SRP receptor but before ribosome docking on Sec61 (Figure 3g). Because SRP and Sec61 occupy overlapping sites on the ribosome, the handover reaction must necessarily involve an intermediate when the signal anchor is released from its binding site on SRP, but cannot engage Sec61 until SRP dissociates from the ribosome. Importantly, the large size of SRP and its receptor¹¹² necessarily means the ribosome is sufficiently far from the membrane to allow EMC’s cytosolic domain to approach the ribosome exit tunnel to access the substrate.

This putative intermediate would necessarily be short-lived because of Sec61’s high affinity for its binding site at the exit tunnel¹¹³. After ribosome binding to Sec61, EMC’s cytosolic domain can no longer get within ~100 Å of the ribosome exit tunnel, which is now very close to the membrane. It is attractive to postulate that the limited time available for EMC precludes sub-optimal hydrophobic sequences (e.g., those intended for N_cyt insertion by Sec61) from promiscuous insertion. The mechanism of co-translational triage of TMDs for insertion by EMC versus Sec61 remains to be resolved.

Principles of TMD insertion by the Oxa1 superfamily

Oxa1 resides in the mitochondrial inner membrane and mediates insertion of many inner membrane proteins including components of the respiratory chain complexes. The earliest studies on Oxa1 recognised its similarity to Alb3 in the inner chloroplast membrane and YidC in the bacterial plasma membrane, both of which were later shown to function in membrane protein insertion^114–118. More recently, GET1, EMC3, TMCO1, and archaeal proteins of unknown function were found to be evolutionarily related to the Oxa1 family⁷¹. Recent structures demonstrate that the conserved core of YidC^119,120 is found in an archaeal protein Ylp1¹²¹, yeast and human EMC3^{58,59,105,106}, TMCO1¹⁰⁴ and GET1⁹⁹. These observations, beyond their interest for protein evolution, allow experimental results from the study of one family member to be interpreted in light of mechanistic insights from other members. In what follows, we attempt to synthesise the biochemical and structural analysis of various insertase family members into a set of shared core principles.

The conserved element of all Oxa1 superfamily members is a three-TMD core arranged to form a cytosol-facing hydrophilic vestibule embedded in the membrane (shown in Figure 4a for YidC, GET1 and EMC3) [recently reviewed¹²²]. Most but not all family members contain one or more basic amino acids in the vestibule. The first and second TMDs of the core are linked by a cytosolic coiled-coil that is highly variable in length and conformation among family members. In most cases, Oxa1 members form a complex with evolutionarily conserved partners whose functions are unclear. Although TMCO1 and Ylp1 initially seemed to contain only the core subunit, co-evolutionary analyses suggest they may interact with C20orf24 and Mj0606, respectively, to form complexes structurally and evolutionarily related to the GET1-GET2 and EMC3-EMC6 complexes⁷². Beyond these shared features, Oxa1 superfamily members are elaborated with additional TMDs (e.g., YidC and Oxa1), additional domains in the lumen or cytosol, or they function in the context of larger protein assemblies (e.g., EMC, TMCO1). Elements common to all (or nearly all) Oxa1 superfamily members are likely to participate in the conserved function of TMD insertion, whereas the divergent elements may provide regulatory, member-specific, or organism-specific functions.

Fig. 4. Structure and function of the Oxa1 superfamily insertases.

| a | Some examples of Oxa1 superfamily members (blue) found in the prokaryotic plasma membrane (YidC) and eukaryotic ER (GET1 and EMC3). Closeup views of the conserved 3-TMD core of each family member are shown below. The cartoon at right illustrates a cross-section at the plane of the vestibule where the membrane might be distorted or thinned. YidC is a single polypeptide, whereas GET1 forms an obligate complex with GET2 (orange), both of which participate in recruiting the targeting factor GET3 (gray). Only one of two GET1-GET2 complexes observed in the structure is shown; the other is behind the displayed complex facing the opposite direction. EMC3 is part of a 9-protein complex that includes EMC6 (orange), which is structurally and evolutionarily related to GET2. Other EMC subunits are shown in gray. Note that the membrane domains of EMC4, EMC7, and EMC10 probably reside in front of EMC3 in this view but were not resolved in this structure. PDB IDs: YidC, 3WO6; human GET complex, 6SO5; human EMC, 6WW7. b | Model for substrates favoured and dis-favoured for insertion by Oxa1 superfamily insertases such as YidC, the GET1-GET2 complex, and EMC. TA proteins with positive charges in the translocated tail are inserted into mitochondria. Signal anchors with a positively charged or lengthy N-terminal flanking domain are inserted by the Sec61 complex.

The hydrophilic cavity can be considered a partial channel across the membrane where a segment of the substrate’s flanking hydrophilic domain can move part of the way toward the lumen. This would reduce the energetic barrier to complete translocation if the lipid bilayer is locally distorted to shorten the distance between the head groups of each leaflet [reviewed elsewhere⁷³]. Such a mechanism would allow short polypeptide segments, but not long or folded domains, to cross the membrane concomitant with TMD insertion. The basic amino acid(s) inside the vestibule would impose the additional constraint of disfavouring vestibule entry of substrates with high positive charge. Thus, substrates for Oxa1 superfamily members are TMDs flanked by relatively short unstructured segments of polypeptide with relatively few basic amino acids; TMDs whose context does not meet these criteria would be rejected and instead inserted by other machinery (Figure 4b).

Consistent with this model, terminal TMDs that are established to be directly inserted by YidC, EMC, and the GET complex have short unfolded translocated regions, and increasing the length of this tail impedes insertion in the GET pathway⁴² and probably others¹²³. The need for a hydrophilic cavity with accompanying distorted membrane would explain why the cytosolic domains of the Get1-Get2 complex at the membrane, while sufficient for Get3 targeting and substrate release, does not mediate effective TMD insertion. Oxa1 superfamily members also have the capacity to translocate hydrophilic loops between sequential TMDs^117,124. Such loops are presumably translocated through the hydrophilic vestibule concomitant with insertion of the two adjacent TMDs, either in rapid succession or together as a hairpin. Reconstitution studies of YidC-mediated two-TMD insertion support this mechanism of hairpin insertion^125,126, whereas artificial lengthening of the translocated domain between a TMD pair precludes YidC-mediated insertion¹²⁴.

The discrimination against flanking basic residues might explain why N_cyt signal anchors close to the N-terminus are not inappropriately inserted in the N_exo orientation by EMC (or YidC in bacteria) even if EMC or YidC initially gets this opportunity. The same mechanism could be used by EMC and the GET1-GET2 complex to reject TA proteins intended for mitochondria, which typically have positive charges in their translocated tail. Thus, Oxa1 superfamily members may contribute to the long-observed “positive-inside” rule at the bacterial plasma membrane and eukaryotic ER, in which the cytosolic flanking domains of TMDs are enriched in positively charged amino acids¹²⁷. Such rejection is unlikely to be perfect, but any inappropriately inserted TMDs might be rectified by quality control mechanisms (see Box 2).

Substrate access to the insertases may be regulated

Given the simple paradigm of a membrane-thinning hydrophilic cavity, why are Oxa1 superfamily members typically part of larger complexes or embellished with additional modules (Figure 4a)? One possibility might be to regulate substrate access to the hydrophilic cavity. This seems to be the case in the GET pathway, where Get2’s cytosolic domain is important for delivering substrates to Get1^94,96. In addition, it is attractive to posit that the hydrophilic cavity in Get1 is normally partially occluded, either by Get2 or a yet unseen ‘closed’ Get1 conformation, until engaged by substrate-bound Get3. Thus, the cavity would be contextually gated rather than constitutively residing in the ER in an energetically unfavourable ‘open’ conformation.

In the case of EMC, the hydrophilic cavity of EMC3 seems to be partially occluded by EMC4, EMC6, EMC7 and possibly EMC10^{58,59,105,106,111}. An attractive model is one where one or more of these subunits regulate access to EMC3’s insertase module. How this regulation could be achieved is unclear, but one possibility is that substrate binding to the cytosolic domains of EMC induces conformational changes that better expose the hydrophilic cavity of EMC3. This is consistent with the observation that mutations in the cytosolic domain of EMC relatively far from EMC3’s hydrophilic vestibule impede insertase function^105,106.

Whether YidC or Oxa1 undergo any conformational changes that regulate their hydrophilic cavities is unclear. Crosslinking studies with YidC suggest that the cytosolic coiled coil connected to two TMDs of the three-TMD core can potentially interact with SRP¹²⁸ and the ribosome^102,129. Perhaps SRP or ribosome binding to the coiled coil is coupled to an opening of a normally closed hydrophilic cavity. TMCO1, about which very little is known, also has a cytosolic coiled coil that interacts with the ribosome¹⁰⁴ and seems likely to have an interaction partner (C20orf24) related to GET2 and EMC6. Whether these features regulate its putative insertase activity is unknown. Conformational changes in the Oxa1 superfamily insertases during their functional cycle are poorly understood and warrant future biophysical analysis.

MEMBRANE PROTEIN FOLDING

TMDs of multi-pass membrane proteins are often packed together using non-hydrophobic amino acids (Figure 5a). Furthermore, TMDs that contribute to channels, ligand binding sites, or catalytic sites typically contain polar or charged amino acids. Calculations of insertion propensity¹³⁰ suggest that the TMDs of multi-pass membrane proteins are systematically lower hydrophobicity than the TMDs of single-pass membrane proteins (Figure 5b). Furthermore, exposed hydrophilic side chains within the membrane would be recognised for quality control by ER-associated degradation pathways (Box 2). Thus, successful folding of multi-pass membrane proteins requires the biogenesis machinery to not only insert partially hydrophilic TMDs, but also temporarily stabilise and shield them in the membrane until their successful assembly with other TMDs. The mechanisms of multi-pass membrane protein biogenesis are just beginning to be defined.

Fig. 5. Biogenesis of multi-pass membrane proteins.

| a | The linear and folded states of the β1-adrenergic receptor (ADRB1, PDB ID 2VT4) illustrates how most of the hydrophilic side chains (blue, red and yellow) in the TMD become buried upon folding. b | Violin plots showing the hydrophobicity of TMDs in single-pass and multi-pass proteins. Hydrophobicity was calculated as the predicted energy of membrane insertion, where a negative value favours insertion and positive value disfavors insertion¹³⁰. c | The PAT complex engages and protects nascent TMDs with hydrophilic residues until they are buried in the protein interior upon folding. Substrates that are not shielded are potential targets for quality control (QC). Note that PAT complex shielding could occur regardless of the route of TMD insertion. d | Cryo-EM structure of the ribosome-bound multi-pass translocon (PDB ID 6W6L). The view from the ER lumen also indicates the potential positions of C20orf24 (adjacent to its likely interaction partner TMCO1), and Asterix (adjacent to its interaction partner CCDC47). e | Model for protein biogenesis by the multi-pass translocon. In the intermediate depicted, the first TMD is being held by the PAT complex until TMD2 and TMD3 are inserted as a unit using the hydrophilic groove of the TMCO1 complex. The lipid cavity may be the site where multiple substrate TMDs can fold protected from aggregation and inappropriate interactions. f | Hypothetical assembly factors temporarily shield hydrophilic regions of TMDs in the unassembled state. This may serve to stabilise substrates in the membrane until their assembly with interaction partners.

Partially hydrophilic TMDs engage the PAT complex

Pioneering experiments investigating multi-pass membrane protein biogenesis used chemical crosslinking to find proteins adjacent to various insertion intermediates of the 7-TMD protein rhodopsin^131,132. In addition to the ribosome-associated Sec61 complex, prominent crosslinks were seen between TMD1 of rhodopsin and an unidentified ~10 kD protein provisionally termed PAT-10 (for protein associated with the translocon). This crosslinking partner was recently identified to be Asterix and shown to tightly interact with CCDC47 to form the PAT complex¹³³ (Figure 5c). Site-specific photo-crosslinking experiments indicate that Asterix directly engages substrate TMDs inside the membrane¹³³, whereas parallel structural studies described later show that CCDC47 binds to Sec61-bound ribosomes¹⁰⁴.

Mutagenesis experiments suggest that hydrophilic amino acids within a substrate’s TMD are essential for PAT complex engagement¹³³. This is noteworthy because essentially all multi-pass membrane proteins contain TMDs that have partial hydrophilic character that must be packed against other TMDs in the final structure (Figure 5a). The PAT complex seems to selectively engage and presumably protect semi-hydrophilic TMDs until their intramolecular interaction partners are synthesised. Consistent with this function, depletion of either PAT complex subunit impairs stable expression of several multi-pass membrane proteins without any obvious effect on single-pass membrane proteins¹³³.

Precisely what goes wrong during substrate biogenesis without the PAT complex is not known. The simplest explanation is that in the absence of a stabilising chaperone, a partially hydrophilic TMD might engage quality controls factors¹³⁴ (Box 2) or even slip out of the membrane^135–137, leading to promiscuous degradation. Such off-pathway fates would compete with on-pathway folding, explaining why the phenotypic effects on membrane proteins are partial. This is analogous to how many cytosolic proteins can fold to at least some degree without chaperones, whose primary role is to minimise off-pathway outcomes.

The PAT complex seems to be essential at the organismal level^138,139, causes ER stress (presumably due to excess membrane protein misfolding) when deleted in cells¹³⁸, and has been conserved broadly across eukaryotes. The remote yeast homologs of Asterix (YPR063C) and CCDC47 (YNR021W) have not been studied yet but could provide a useful system for large-scale genetic and phenotypic analyses. Future studies should investigate the mechanistic and structural basis of substrate recognition by Asterix and the precise consequences of its absence for insertion and folding. As discussed below, the PAT complex probably cooperates with other putative chaperones, insertases, and Sec61 to ensure membrane protein folding. Among these additional factors, members of the Oxa1 superfamily may have chaperone functions in addition to their insertase roles (see Box 3).

Box 3: Oxa1 superfamily members may also operate as chaperones.

The observation that newly inserted TMDs near the SecY complex can crosslink to YidC has long suggested that YidC might have a potential chaperone function²⁰⁵. This idea is attractive because YidC’s hydrophilic cavity used for polypeptide translocation serves as an ideal binding site for a partially hydrophilic TMD. In vivo experiments showing an effect of YidC depletion on insertion of some substrates^{63,116,125,206–209} and folding of others^210–212 has supported this dual-function model for YidC. Whereas the insertion function is strongly supported by reconstitution studies in vitro^213–219, the chaperone function is less well defined and difficult to disentangle from YidC’s insertase role.

The best studied chaperone substrate for YidC is the LacY sugar transporter. Folding of LacY, as monitored by conformation specific antibodies and intramolecular disulphide crosslinking, was observed to be impaired in E. coli acutely depleted of YidC^211,212. LacY was found to have inserted in its normal topology based on resistance to alkaline extraction and cysteine accessibility of inter-TMD loops. Because LacY can physically interact with YidC, the effect on folding seems to be direct. However, it can be challenging to exclude indirect effects of YidC depletion (which rapidly causes extensive compensatory responses) or an impairment in some aspect of LacY insertion, which YidC is capable of mediating²²⁰. Despite these caveats, the concept of YidC acting as a chaperone is attractive and might also apply to other Oxa1 superfamily members.

In this regard, EMC is noteworthy. Proteomic analyses of EMC-depleted cells show changes in the levels of various types of membrane proteins^221,222. Because many of these are neither tail-anchored nor begin with an N_exo signal anchor, it has been speculated that EMC performs functions beyond insertion of terminal TMDs^105,222. Although it is difficult to exclude indirect effects, one explanation is if EMC also mediates insertion of TMD pairs linked by a short lumenal loop. A second non-mutually exclusive possibility is that EMC, either by itself or via interaction partners, acts as a chaperone during membrane protein biogenesis.

Co-immunoprecipitation of EMC with various multi-pass membrane proteins, and in some cases their substrate-specific maturation factors, supports a post-translational role in early biogenesis²²². Genetic and proximity-biotinylation experiments have also suggested a co-translational function for EMC during biogenesis of certain multi-pass membrane proteins²²². Although this idea is attractive and analogous to YidC’s co-translational function near SecY^{128,205,223,224}, EMC’s large cytosolic domain limits its access to ribosome-associated Sec61. Hence, EMC can only engage a nascent membrane protein after parts of it diffuse ~100 Å away from the ribosome exit tunnel or the ribosome detaches from Sec61. This steric constraint might help explain why the best co-translational substrate candidates (e.g., Yor1, Fks1, and viral polyproteins) are large multi-pass proteins with one or more long inter-TMD loops.

A second line of evidence for an insertase-independent function for EMC comes from mutagenesis studies¹⁰⁵. EMC mutations at regions far from the EMC3-EMC6 insertase module can impair maturation of a membrane protein (TMEM97) without affecting known insertase substrates. Seeking a mechanistic explanation for such substrate-specific mutants may reveal currently unexplored aspects of EMC function. Toward this end, it will be important to analyse different co- and post-translational biogenesis intermediates for direct EMC interactions using approaches such as site-specific crosslinking and eventually, structural methods.

A specialised translocon for multi-pass membrane proteins

The initial study of the PAT complex did not investigate its interactions with or position relative to the Sec61 complex. Instead, unexpected insight into this issue has come from a parallel seemingly unrelated study investigating the interaction partners and structure of the Oxa1 superfamily member TMCO1¹⁰⁴. Purification and proteomic analysis of TMCO1-containing ribosomes revealed the Sec61 complex, CCDC47, and a three-protein complex composed of TMEM147, Nicalin, and NOMO1. The absence of CCDC47’s partner Asterix in this proteomic experiment is probably explained by Asterix’s paucity of suitable tryptic fragments¹³³. Thus, one can reasonably postulate that this TMCO1-containing translocon contains several key elements for membrane protein biogenesis: a protein conducting channel (Sec61 complex), an Oxa1 superfamily insertase complex (TMCO1 and C20orf24), and an intramembrane chaperone (the PAT complex).

Consistent with this function, sequencing of the mRNAs recovered with TMCO1-containing ribosomes revealed an exceptional enrichment of those coding for multi-pass membrane proteins¹⁰⁴. Analysis of EAAT1, a trimeric multi-pass membrane protein whose mRNA was enriched in TMCO1-containing ribosomes, showed markedly reduced levels in cells depleted of CCDC47, TMCO1, or components of the TMEM147 complex. A moderate-resolution cryo-EM reconstruction of TMCO1-containing ribosomes revealed the positions of CCDC47, TMCO1, and the TMEM147 complex relative to the Sec61 complex¹⁰⁴ (Figure 5d). These components (and potentially yet unidentified factors) constitute a translocon that appears to be specialised for multi-pass membrane protein biogenesis.

The three factors (TMCO1 complex, PAT complex, and TMEM147 complex) define a cavity, presumably containing lipids, on the hinge side of Sec61 opposite its lateral gate (Figure 5d). The exit tunnel of the ribosome is positioned where this cavity abuts Sec61, providing the nascent protein potential access to Sec61’s lateral gate and the multi-pass translocon components. Thus, TMDs emerging from a ribosome bound to this multi-pass translocon would have (at least) two insertion routes, Sec61 or TMCO1, depending on the substrate.

For example, TMDs followed by a lengthy lumenal domain might engage Sec61 so its channel can be utilised for soluble domain translocation (e.g., Figure 3d). By contrast, TMDs flanked by only short translocated segments could use TMCO1 by a mechanism typical for Oxa1 superfamily members. Notably, there seems to be sufficient space between the ribosome and membrane to accumulate two TMDs and a short loop, thereby allowing their concerted insertion via TMCO1 (Figure 5e). In this way, poorly hydrophobic internal TMDs of multi-pass membrane proteins, many of which cannot engage Sec61 effectively, could nonetheless be inserted by TMCO1.

The mammalian multi-pass translocon may be analogous to the prokaryotic ‘holotranslocon’ containing the SecY complex (which also contains SecE and SecG), YidC, the SecD-SecF complex, and YajC¹⁴⁰. Both translocons contain a SecY family channel, an Oxa1 superfamily insertase (e.g., YidC), and a lipid-filled cavity. Although many TMDs might be able to use either the channel or the insertase for insertion, the unique functionality of the SecY family is its capacity to translocate lengthy soluble domains across the membrane. Conversely, Oxa1 superfamily members might be better at other reactions, such as insertion of two-TMD pairs or low-hydrophobicity terminal TMDs, that are less suited for the SecY family. A translocon with both a protein translocation channel and an insertase may therefore provide the requisite flexibility to accommodate the diverse TMDs and loops of multi-pass membrane proteins.

A potential site for membrane protein folding

The lipid-filled cavity of the mammalian multi-pass translocon¹⁰⁴ is large enough to house multiple substrate TMDs with their exposed hydrophilic parts being temporarily chaperoned by the PAT complex or other translocon components (Figure 5e). A protected cavity would facilitate intramembrane folding of nascent multi-pass proteins without off-pathway interactions with bulk ER proteins or quality control factors. This mechanism is analogous to cytosolic protein folding inside a chaperonin chamber¹⁴¹. In the chaperonin example, the surface properties of the chamber’s interior promote folding of substrates. It will be interesting to explore whether the interior features of the multi-pass translocon cavity similarly facilitate productive TMD-TMD interaction to catalyse intramembrane folding. At present, the bacterial holotranslocon has only been visualised at low resolution so the placement of its constituents must be considered provisional¹⁴⁰. Nevertheless, it seems to also contain a lipid-filled cavity, potentially providing a protected site for membrane protein folding¹⁴².

Crosslinking analyses of multi-pass membrane proteins suggest that different TMDs dynamically sample multiple protein environments at different stages of synthesis^131,143. Furthermore, it has been observed that up to six TMDs of a multi-pass membrane protein can be accommodated in a urea-extractable state at the translocon prior to their eventual membrane integration¹⁴⁴. These findings can potentially be rationalised by a model where multiple substrate TMDs insert into and fold within the multi-pass translocon cavity. Directly testing this idea will require increasingly precise assays for different steps in multi-pass membrane protein insertion and folding, combined with the capacity to generate, analyse, and structurally characterise key intermediates in the process.

MEMBRANE PROTEIN ASSEMBLY

Roughly half of membrane proteins are part of multi-protein complexes, many with other membrane proteins^145–147. The mechanism of assembly of two or more membrane proteins within the lipid bilayer is poorly understood. The simplest model is that the individual subunits diffuse until they encounter their partner(s). This mechanism, although plausible for certain simple complexes or in reconstituted systems, is unlikely to be the primary strategy used in a crowded cellular environment. The main reason is that unassembled orphan subunits are prone to aggregation and targets for quality control, as evident from the well-known phenomenon of subunit degradation when its interaction partner is eliminated^148–150. Avoiding these off-pathway fates between completion of synthesis and assembly probably requires chaperones to temporarily shield assembly intermediates from quality control.

A ‘placeholder’ mechanism for subunit stabilisation

Although assembly chaperones for membrane proteins are not well defined, hypotheses about their plausible characteristics and mechanism can be gleaned from known factors for soluble protein complex assembly. For example, hemoglobin assembly is facilitated by an assembly factor that specifically binds to and temporarily shields the surface of the alpha subunit intended for interaction with the beta subunit^151,152. During EMC assembly, cytosolic EMC2 temporarily interacts with WNK1 until the latter is displaced by cytosolic EMC8 binding to the same site¹⁵³. Similarly, subunits of the 19S proteasome base interact with factors that shield regions which eventually dock onto the 20S proteasome core¹⁵⁴. Thus, membrane protein assembly factors may similarly bind temporarily to inter-subunit interfaces (Figure 5f). A worthwhile endeavour is to search for such factors and analyse their role in assembly of membrane protein complexes.

Such ‘placeholder’ chaperones might be generic in many cases given that intramembrane interfaces between subunits often share similar features such as polar side chains. Thus, a factor capable of dynamically covering such surfaces may be sufficient to temporarily shield a newly made subunit for long enough to find its partner. Acquisition of such a chaperone at the site of synthesis (e.g., as part of the multi-pass translocon) would provide an initial opportunity for assembly before competing with quality control factors in the case of failure. Thus, factors implicated in membrane protein folding are reasonable candidates for also aiding assembly by remaining associated with their substrates. Indeed, both EMC and YidC have been implicated in biogenesis of multi-protein complexes^155–157, although how they mediate these functions remains speculative.

In addition to shielding proteins from premature degradation, assembly chaperones may also prevent poorly hydrophobic TMDs from membrane dislocation. The single TMDs of some T-cell receptor subunits¹³⁷ and certain TMDs of multi-pass protein subunits have been observed in the cytosol or lumen in their unassembled state^135,136,158. Because topologically incorrect subunits would not be able to assemble, it is attractive to posit the existence of factors that stabilise an otherwise unstable topology until assembly. This too could be performed potentially by multi-pass translocon subunits that remain associated after translocon disassembly at the end of substrate synthesis. As with cytosolic protein complexes, some of these stabilising chaperones might be substrate-specific. Indeed, T-cell receptor subunits engage a yet-unidentified T-cell specific ER factor during, but not after assembly¹⁵⁹.

Assembly factor discovery from neurobiology and virology

In addition to insertases and chaperones potentially doubling as assembly factors, other candidates can be mined from orthogonal fields. For example, ion channels, neurotransmitter receptors, and sensory receptors have been extensively analysed for factors involved in their productive expression. Proteomic, loss-of-function, and gain-of-function screens have found a variety of genes, many of which operate in the secretory and endocytic pathways^{155,160–167}. Conserved ER resident protein hits are excellent candidates for factors involved in the early biogenesis of receptors or channels. In support of this idea, EMC subunits, PAT complex subunits, and other components of the multi-pass translocon were hits in one or more such screens^{155,165–168}. Dissection of other hits from such studies may yield insights that have general applicability.

For different reasons, another field from which biogenesis factors consistently emerge is virology. Numerous genetic screens for host factors involved in an enveloped virus’s life cycle also find ER biogenesis factors such as the signal peptidase complex, oligosaccharyl transferase complex, and EMC^169–172. Viral membrane proteins are sometimes complicated and produced from polyproteins, perhaps requiring chaperones and assembly factors. Indeed, there is a long history of investigating protein biogenesis in the ER by the study of virus glycoproteins¹⁷³ such as influenza hemagglutinin^174,175. Mining the many functional screens for ER-resident host factors, then analysing them in focused biochemical assays, may yield one route to understanding assembly of multi-protein complexes.

CONCLUSIONS AND OUTLOOK

The steps of membrane protein biogenesis from targeting to assembly are understood in decreasing mechanistic depth. The initial concept of membrane protein biogenesis¹⁷⁶, where a single linear pathway and uniform machinery sequentially interprets hydrophobic elements as they emerge from the ribosome, has been diversified into multiple pathways at each step. Understanding how the nascent polypeptide is routed toward one machinery versus another based on its sequence features, has emerged as a key goal. The eventual aim is to explain in molecular terms the specific steps taken by each major class of membrane protein to achieve its final assembled state.

This aim seems to be nearing completion for most types of single-pass membrane proteins. Biochemical reconstitution, structures of key factors, in vivo analysis, and evolutionary considerations now lead to a mostly unified and consistent view. Nevertheless, small membrane proteins, many of which were overlooked in early genome annotations^177,178, are poorly studied and the basis of their biogenesis in the correct topology is not clear. Even for seemingly settled substrates such as N_exo signal anchors of Type III proteins, a role for EMC (and the uncertain relationship to Sec61) has only recently emerged. And the functions of SND components, whose deletion impacts the localisation of some proteins, is not understood.

The richest avenues for future work lie in the steps of multi-pass membrane protein biogenesis, including multi-subunit assembly. Here, our understanding is still at the stage of compiling a reasonably complete parts list. Given the highly pleiotropic effects of perturbing core pathways in cells, caution is warranted in assigning molecular functions based primarily on endpoint phenotypes of reporters. Complementary biochemical studies will be needed to demonstrate a direct effect at a specific step of substrate biogenesis. This time-tested two-pronged strategy that was so successful in dissecting simpler biogenesis pathways will undoubtedly be valuable for multi-pass membrane proteins. Thus, devising sensitive quantitative cellular reporters and setting up biochemical reconstitution systems where topology and folding can be measured precisely will be important.

GLOSSARY

Oxa1 superfamily

an evolutionarily-related group of membrane protein insertases that includes: Oxa1 in the inner mitochondrial membrane; YidC in the bacterial inner membrane; Ylp1 in the archaeal plasma membrane; Alb3 in the chloroplast inner membrane; GET1, EMC3, and TMCO1 in the eukaryotic endoplasmic reticulum

Intramembrane chaperone

factors that promote folding in the membrane by temporarily shielding partially hydrophilic TMDs of nascent polypeptides until their successful assembly with other TMDs

Assembly factor

factors that promote the assembly of two or more proteins, possibly by temporarily shielding their inter-subunit interfaces

Insertase

a transmembrane protein containing a hydrophilic groove that facilitates translocation of short polypeptide segments across the membrane concomitant with TMD insertion

Sec61 complex

a heterotrimeric protein complex that translocates hydrophilic polypeptide segments across the membrane through an aqueous channel and inserts hydrophobic domains into the membrane through a lateral gate. It is called the SecY complex in prokaryotes

Targeting sequence

the sequence element in a protein that directs its delivery to a specific membrane in the cell. For membrane proteins, the targeting sequence is typically a cleavable N-terminal signal peptide or the first transmembrane domain (TMD)

Single-pass membrane protein

proteins spanning the membrane once

Multi-pass membrane protein

proteins spanning the membrane more than once

Type I membrane protein

Signal-peptide containing membrane proteins oriented with their mature N-terminus facing the lumen (a topology that is also termed N_exo) following signal-peptide cleavage

Type II membrane protein

Membrane proteins oriented with their N-terminus facing the cytosol (a topology that is also termed N_cyt

Type III membrane protein

Membrane proteins oriented with their N-terminus facing the lumen (a topology that is also termed N_exo); these proteins typically possess a short (<50 amino acids) N-terminal flanking region

Tail-anchored membrane protein

Membrane proteins whose only transmembrane domain lies within ~65 amino acids of the C-terminus, and are oriented with the N-terminus facing the cytosol. These are sometimes called Type IV membrane proteins