To comprehensively dissect both conserved and species-specific functions of the EMC, we developed approaches to produce EMC for structure determination and broad mutational analysis (Figure 1A–D). We developed systems to produce robust quantities of pure intact yEMC and hEMC to determine structures for the two organisms in which different facets of EMC function have been described in detail (Jonikas et al., 2009; Christianson et al., 2012; Guna et al., 2018; Shurtleff et al., 2018). Parallel efforts converged on an approach involving FLAG affinity-tagging of the EMC5 C-terminus, which was performed for endogenous yEMC and recombinant hEMC in human embryonic kidney (HEK) cells (Figure 1—figure supplements 1–2, Supplementary file 1).

In parallel, to enable testing of hypotheses based on structures, we created a suite of human (K562) knockout cell lines deleted for individual hEMC subunits - hEMC1 (lumen), hEMC2 (cytoplasm), hEMC3, and hEMC5 (transmembrane) - and a series of reporters of EMC-dependent transmembrane protein biogenesis (Figure 1—figure supplements 3–4). Reintroduction of the wild-type hEMC subunits in the respective knockout cells fully rescued the knockout phenotype (Figure 1—figure supplements 5–6). This allowed for introduction of structure-based mutations in hEMC subunits into the respective knockout cells to determine features supporting biogenesis of fluorescently tagged versions of three different types of EMC clients: the transmembrane domain of a C-terminal tail-anchored transmembrane protein (squalene synthase, SQS^378-410) (Guna et al., 2018), a polytopic transmembrane protein that depends on the EMC N-terminal insertase activity (Beta 1 adrenergic receptor, B1AR) (Chitwood et al., 2018), and a polytopic transmembrane protein (Sigma intracellular receptor 2, TMEM97) whose biogenesis requires the Sec61 translocon but does not require a terminal helix insertase (Figure 1—figure supplements 3–6). Three individual EMC clients were fused to mCherry fluorescent protein and GFP separated by a P2A ribosomal skipping sequence. Translation of the described mRNA generates two products due to peptide bond skipping at the P2A sequence. For each molecule of the client-mCherry fusion, there is one GFP molecule. Reduction in mCherry levels relative to GFP reflects post-translational degradation of the client fused to mCherry. Each of the client reporters were introduced into five separate cell lines: wild-type K562 cells, hEMC1 knockout K562 cells, hEMC2 knockout K562 cells, hEMC3 knockout K562 cells, and hEMC5 knockout K562 cells. Monitoring the effect of an hEMC mutation on fluorescent reporter levels provided a quantitative measure of its impact on EMC-dependent biogenesis of each class of client protein. A number of mutations of varying severity, varying conservation between yeast and human (Figure 1—figure supplements 7–8), were designed and tested spanning the hEMC structure. Subsequently, these 49 mutations were mapped onto the structure grouped by reporter phenotype (Video 1, Supplement File 2). To allow for direct comparison of our structure-guided mutant phenotypes with those published recently by others (Pleiner et al., 2020; Bai et al., 2020; O'Donnell et al., 2020), we summarized all mutant data (Supplementary file 3). A subset of the mutant cell lines was validated by genotyping (Figure 1—figure supplement 9). Western blots against the endogenous hEMC subunits allowed us to control for mutational effects on the production and stability of the hEMC complex itself. We concurrently blotted for three clients, SQS, TMEM97, and BCAP31, to assay changes in endogenous protein levels for each of the mutations (Figure 1—figure supplements 5–6, Supplementary file 4). This strategy thus distinguishes effects resulting from a global disruption of the EMC complex from those caused by specific disruption of EMC function. These functional assays of the hEMC show a broad dependence of all these clients on the EMC, consistent with previous work (Shurtleff et al., 2018; Guna et al., 2018; Chitwood et al., 2018; Volkmar et al., 2019; Tian et al., 2019). In order to understand the mechanism of action, we will now go in more detail through several of the mutants with the strongest functional phenotypes in differing regions of the three-dimensional structure.

We determined structures of yEMC and hEMC — all showing overall compositional similarity, with regional conformational differences between the yeast and human complexes (Figure 2A–D). We obtained reconstructions of yEMC bound to an antigen binding fragment (Fab) and hEMC reconstituted both in detergent micelles and lipid nanodiscs, with the latter strategy yielding the most isotropic and highest resolution data. For yEMC+FabDH4 and hEMC, the global map resolutions reached 3.2 Å and 3.4 Å, respectively (Table 1, Figure 2—figure supplements 1–4). The cryo-EM maps allowed for de novo model building of both human and yeast complexes (Figure 2—figure supplements 5 and 6). As described in the following sections, our multiple EMC structures enable a broad survey of its conserved architecture, with variations between the structures pointing to conformational and compositional differences (Figure 2—figure supplements 7–8). We note that our maps and models are consistent with recent cryo-EM data from yeast EMC (Bai et al., 2020), human EMC (O'Donnell et al., 2020; Pleiner et al., 2020), and a crystal structure of human EMC2-EMC9 (O'Donnell et al., 2020; Figure 2—figure supplement 9).

The EMC is comprised of cytoplasmic, transmembrane, and lumenal domains arranged similarly for yeast and human, despite significant evolutionary separation (Figure 2A–B). For both species, subunits encompassing EMC2 to EMC7 form an interconnected core complex, while there is additional density capping both the cytoplasmic and lumenal domains of hEMC, occupied by an hEMC8/9 and an hEMC1 N-terminal expansion, respectively (Figure 2C–D). hEMC8 and hEMC9 are paralogs of each other, which have not been identified in yeast (Wideman, 2015). We modeled and depict only hEMC8 for clarity, but due to the 44% sequence identity with hEMC9 and both being present in the recombinant system we refer to this as hEMC8/9. The large hEMC1 insertion in hEMC constitutes the majority of a membrane distal beta-propeller domain protruding into the lumen, a feature missing from S. cerevisiae. Compared to other ER-resident proteins implicated in membrane protein biogenesis (Suloway et al., 2009; Pfeffer et al., 2017; Ramírez et al., 2019; McDowell et al., 2020; McGilvray et al., 2020), the arrangement of domains of the EMC is unusual with the transmembrane domain connecting prominent cytoplasmic and lumenal domains (Figure 2E). On a global level, the structure suggests complexities beyond those of some other ER machineries fulfilling select functions in transmembrane protein biogenesis.

The exterior interface of the cytoplasmic domain is formed by EMC2, EMC3, EMC4, and parts of hEMC8/9 (in human), while parts of EMC5, EMC2, and EMC8/9 are shielded from the cytoplasm (Figure 3A–B). The helical fold of EMC2 constitutes the central organizer of this platform, established by five or six TPR motifs in human versus yeast, respectively (Figure 3C). TPR domains are commonly found mediating protein-protein interactions and are present in numerous well-characterized chaperone-protein and other interaction networks (Blatch and Lässle, 1999; Scheufler et al., 2000; Schlegel et al., 2007; Assimon et al., 2015; Krysztofinska et al., 2017; Graham et al., 2019). Yeast EMC2 features a more curved helical arrangement with N- and C-terminal domains in closer proximity to each other than seen in hEMC2. Notably, the canonical peptide-binding TPR groove is occupied by the partially helical C-terminus of EMC5, which forms a large interaction surface with EMC2. To test the functional roles of this interaction, we mutated three residues within the hEMC2 TPR motif (hEMC2^{K125E + R126D + K127E}) or a single hEMC5 residue buried in the TPR-binding groove (hEMC5^F90A). The mutations on both sides of the interface decreased hEMC integrity by western blot, with a modest decrease of hEMC subunits for hEMC5^F90A and a strong reduction in the levels of several hEMC subunits for hEMC2^{K125E + R126D + K127E} (Figure 3C, Figure 3—figure supplements 1–2, Figure 1—figure supplements 5–6). This suggests that this interface might be critical for EMC complex assembly rather than EMC function.

The multi-protein cytoplasmic cap has distinct elements between hEMC and yEMC. Capping the cytoplasmic domain in hEMC is hEMC8/9; the functional roles of this cap-like structure are not yet clear. An hEMC8-9 heterodimer is not observed and our cryo-EM permits tracing with both the hEMC8 or hEMC9 amino acid sequence (Figure 3—figure supplement 3). Mass spectrometric analysis of our hEMC preparations reveals slightly higher abundance of hEMC8 to hEMC9 (Figure 1—figure supplement 2E–F, Supplementary file 1), so we modeled the cytoplasmic cap structure with the hEMC8 sequence. A groove on hEMC8/9 cradles an N-terminal peptide of hEMC4, which proceeds into the EMC4 segment that traverses over hEMC2 and the three-helix bundle of hEMC3 (Figure 3D). Although yEMC lacks EMC8/9, yEMC4 follows a similar binding trajectory along cytoplasmic yEMC2 and yEMC3 surfaces. A stretch of 20 hEMC4 amino acids (residues 23–42) after the hEMC8/9 binding site is only poorly resolved in our cryo-EM maps and predicted to be disordered (40% glycine content). This loop contains primarily polar amino acids, and traverses the top of the hEMC2 TPR domain. To see whether this dynamic hEMC2-hEMC4 interface played a role in client stabilization, we mutated two charged patches on hEMC2 to alanines (hEMC2^{E146A+E149A+Q150A}, hEMC2^{E168A+D170A+K173A}), lying in close vicinity to hEMC4^23-42 (Figure 3E–F). These mutants lead to a modest accumulation of the tail-anchored client (SQS^378-410) but did not affect polytopic client abundance or decrease of hEMC subunits (Figure 3E–F; Figure 1—figure supplement 5). Several mutants across the cytoplasmic domain showed similar phenotypes, supporting a key role in tail anchor protein biogenesis (Figure 3—figure supplements 1–2).

The transmembrane core of EMC is predicted to include contributions from each subunit except for EMC2 and, in humans, hEMC8/9 (Figure 2C–D). The EMC presents two distinct and structurally conserved cavities on opposite sides of the transmembrane core that differ in size, shape, subunit compositions and apparent function (Figure 4A–B). One cavity, which we refer to as the lipid-filled cavity, appears contiguous with the ER lipid environment (Figure 4A). The second cavity, which we refer to as the gated cavity, appears to open toward the cytoplasm in our structures and is more occluded by a transmembrane helix gate from the lipid environment (Figure 4B). Notable structural hallmarks present in both species include a superimposable core of nine transmembrane helices, a set of flexible gate helices, and an amphipathic EMC1 brace helix (Figure 4C).

Evaluating potential client paths from the cytoplasm into the transmembrane domain revealed a cavernous opening at the membrane-cytoplasmic interface of the gated cavity, wide enough to allow passage of a client helix, and tapering toward the lumen (Figure 4D). Consistent with its potential role as a cytoplasmic conduit into the EMC, the EMC3 portion of the cytoplasmic domain, which delineates this opening, sits approximately 45 Å from the lumenal side of the gated cavity. This dimension exceeds the thickness of the ER membrane (Mitra et al., 2004; Heberle et al., 2020; Cornell et al., 2020; Figure 4D). This cavity is lined primarily by EMC3, EMC4, and EMC6 (Figure 5A). Simulating the dimension of the first transmembrane helix of a known terminal insertase-client (B1AR - Chitwood et al., 2018) suggests that there is sufficient space for a client helix even in the client-free state of the EMC (Figure 5B). The gated cavity is hydrophilic on the cytoplasmic side and becomes increasingly hydrophobic toward the lumenal side (Figure 5C).

The entrance into the gated cavity interior (Figure 5A) is formed primarily by the EMC3 cytoplasmic domain. To test its function, charge swap mutations were introduced along the rim of this opening (hEMC3^{E63K + D213K + E223K}, hEMC3^{R59E + R62E + K216E}) (Figure 5D). These mutants resulted in loss of the tail-anchored client (SQS^378-410) and partial loss of the N-terminal insertase-dependent polytopic client (B1AR), reflecting a failure to support insertase activity. These mutants had no appreciable effect on the abundance of the polytopic transmembrane client (TMEM97) reporter (Figure 5E, Figure 5—figure supplements 1–2). A similar phenotype was observed with alanine substitutions for a pair of lysines at the periphery of this cytoplasmic rim (hEMC3^{K42A + K43A}) (Figure 5—figure supplement 2).

Having identified a functionally important entry route for terminal helix insertase clients, we next considered potential surfaces inside the cavity that might accommodate a client helix. A polar patch close to the membrane interior of this cavity was conspicuous, even though the specific amino acid residues are not strictly conserved (Figure 1—figure supplement 8). Mutating a pair of adjacent asparagine residues to equivalently sized but negatively charged aspartates (hEMC3^N114D+N117D) resulted in a dramatic decrease in SQS^378-410 reporter levels and no significant decrease in the other two client reporter levels (Figure 5F). Western blot analysis for this mutant showed wild-type rescue levels of hEMC subunits and a decrease in endogenous SQS levels (Figure 1—figure supplement 6). Meanwhile, mutating a neighboring positively charged residue to an alanine (hEMC3^R180A), a residue that is conserved in some of the YidC-superfamily insertase proteins (Anghel et al., 2017), resulted in partial loss of only the tail-anchored insertase client (SQS^378-410) (Figure 5—figure supplements 1 and 3).

Lastly, we surveyed residues closer to the hydrophobic lumenal side of the gated cavity. Lipid density was resolved at positions along the cavity in hEMC and yEMC cryo-EM maps (Figure 4B) and the properties of this hydrophobic seal to the lumen are conserved (Figure 5—figure supplement 4A–B). The importance of this hydrophobic seal is suggested by the strong effect of a structurally mild mutation of a conserved methionine to a leucine (hEMC3^M151L), which caused significant decrease in both SQS^378-410 and B1AR abundance (Figure 5G). Mutation of a neighboring aromatic residue (hEMC3^F148L), contacting both a lipid and a hEMC4 C-terminal transmembrane helix, caused a decrease in all three client types without altering the levels of hEMC subunits (Figure 5—figure supplements 1–2, Figure 1—figure supplement 6). Together these results indicate that proper EMC insertase function depends on the exact composition of the cavity and not simply on its hydrophobic nature.

While the core transmembrane helices of the gated cavity are superimposable in all four of our EMC structures, the adjacent gate helices appear in different relative orientations. The structural variability likely reflects dynamics of the gate (Figure 4C). Comparing detergent and nanodisc maps for both species identified two major gate conformations (Video 2, Figure 5—figure supplement 5A). One of the conformations, referred to as the closed-gate conformation, results in a more occluded membrane cavity. The other conformation, referred to as the open-gate conformation, would provide space for client accommodation.

The C-terminal transmembrane helix of EMC4 and ensuing lumenal segment are well resolved in all four structures; however, other regions of EMC4, including the segment connecting the cytoplasmic domain to the transmembrane gate helices, were poorly resolved, perhaps owing to mobility. The yEMC detergent map, yEMC nanodisc map, and hEMC detergent map all show the unassigned helices in the closed conformation, preventing client residence in the gated cavity. By contrast, the hEMC nanodisc map reveals an open-gate conformation with the unassigned helices shifted away from the transmembrane core to provide space for a client (Figure 5B). Consistent with our observations, the closed transmembrane gate conformation can also be seen in recently published cryo-EM maps of hEMC (O'Donnell et al., 2020) and yEMC (Bai et al., 2020), which studied LMNG and digitonin-solubilized complexes, respectively (Figure 5—figure supplement 5B). We note that the conformational heterogeneity and concomitant lower resolution of the gate likely accounts for the challenges in making unambiguous subunit assignments (Figure 5—figure supplement 5C–E), reflected by the three different interpretations reported in recent structures (Pleiner et al., 2020; O'Donnell et al., 2020; Bai et al., 2020).

Considering the apparent flexibility of the gate, we sought to mutate the hEMC4 interfaces resolved in the cytoplasm versus the membrane. As described above, mutating residues that together form a composite-binding surface for the cytoplasmic domain of hEMC4 (hEMC2^{E146A+E149A+Q150A}, hEMC2^{E168A + D170A + K173A}, Figure 3E–F), we observed a modest accumulation of the tail-anchored insertase client (SQS^378-410). Likewise, mutating residues in the center of the gated cavity, close to one of the unassigned helices in the closed-gate conformation (hEMC3^{V118A + I122A}) (Video 2, Figure 5—figure supplement 1) led to an increase of SQS^378-410. This SQS^378-410 accumulation effect stands in contrast to mutating a residue that contacts the lumenal anchor of hEMC4 (hEMC3^F148L), which caused a reduction of SQS^378-410 levels (Figure 5—figure supplement 1).

In addition to the gated cavity, the EMC harbors another membrane-accessible cavity. The surface of the lipid-filled cavity includes contributions from EMC1, EMC3, EMC5, and EMC6 (Figure 6A). In our structures, the EMC2 N-terminus occludes cytoplasmic accessibility to this cavity (Figure 4D, Figure 6A–B). However, this cavity may be accessible from the membrane or the ER lumen. The respective distance from the cytoplasmic EMC2 N-terminus to the lumenal side of the lipid-filled cavity is approximately 35 Å across, which is close to the average ER membrane thickness (Mitra et al., 2004).

The lipid-filled cavity features a uniformly hydrophobic surface (Figure 6C) and superimposes across our ensemble of EMC structures. As noted, we resolved several lipids in our cryo-EM maps lining the cavity wall and modeled four POPC (1-palmitoyl-2-oleoyl-sn-glycero-3 phosphatidylcholine) molecules in the hEMC nanodisc map (Figure 6C). The residues in close proximity to these lipids are moderately conserved (Figure 5—figure supplement 4C–D). To characterize the functional role of the lipid-filled cavity, we mutated cavity-lining and lipid-proximal residues (Figure 6D, Figure 6—figure supplements 1–2). Most of these mutations resulted in an increased abundance of the tail-anchored reporter (SQS^378-410) and wild-type rescue levels for the other two reporters (B1AR, TMEM97). However, one lipid-proximal mutant showed decreased levels of all three client reporter types with varying severity (hEMC3^R13E) without altering overall EMC levels (Figure 6E). Western blotting for the endogenous SQS and TMEM97 revealed a decrease in endogenous SQS and TMEM97 levels for this mutant (Figure 1—figure supplement 6). An analogous mutation in Drosophila EMC3 was recently was reported to cause reduced levels of Rh1 in this mutant background (Xiong et al., 2020). The amphipathic EMC1 brace helix, which packs against the transmembrane helices of EMC5, is a structural hallmark of the lipid-filled cavity (Figure 6D). Here, mutating interfacial residues from hEMC5 (hEMC5^{H19L+S23A+Q26L}) caused a marked decrease in the N-lumenal polytopic reporter (B1AR) and no effect on either the tail-anchored client (SQS^378-410) or the polytopic client reporter (TMEM97) (Figure 6F). Unexpectedly, mutating interfacial residues from hEMC1 (hEMC1^F473Y+R487K) showed a diametrically opposed phenotype in which B1AR was unaffected, increased SQS^378-410 levels, and TMEM97 levels markedly decreased (Figure 6G). Another mutation in this brace (hEMC1^{M483A+R487H+Q491N}) resulted in a decrease in TMEM97 and no significant effect on the other two client reporters. An adjacent hEMC5^D44K mutations in the interfacial brace had yet different resulting client flow cytometry profiles, with an increase in SQS^378-410 and no effect on either of the polytopic client reporters (Figure 6—figure supplement 2B–C). The pleiotropic client phenotypes across the panel of interfacial brace mutants suggest that this feature is critical for multiple EMC functions.

Composed primarily of EMC1, EMC7, and EMC10, the extensive EMC lumenal domain (Figure 7A) is important for polytopic client biogenesis and interactions with lumenal chaperones (Luo et al., 2002; Shurtleff et al., 2018; Hiramatsu et al., 2019; Coelho et al., 2019). EMC7 and EMC10 are scaffolded on two beta-propellers of EMC1, one distal and the other proximal to the membrane. The lumenal cap differs between hEMC and yEMC, with a four-bladed distal beta-propeller in yeast and eight-bladed distal propeller the human complex (Figure 7B). All three lumenal EMC subunits have structural folds known to participate in protein-protein interactions (Reinisch and De Camilli, 2016). Mutations in this lumenal domain have been linked to loss of the EMC complex (Bircham et al., 2011), a trafficking delay for membrane protein Pma1 (Luo et al., 2002), and male infertility (Zhou et al., 2018).

Several regions of the lumenal domain form stabilizing interactions with the membrane cavities. The gate helices of the gated cavity are anchored via the embedding of EMC4’s C-terminus within the membrane-proximal EMC1 propeller. The lipid-filled cavity is connected to the ER lumenal domain via the amphipathic EMC1 brace helix, which is tethered to the membrane-proximal EMC1 beta-propeller. The connections between the lumenal domain and the transmembrane cavities could allow for conformational coupling during client handling. Indeed, superimposing the two conformations presented above, the open- and closed-gate states, revealed not only differences in the transmembrane domain but also a rotation of the lumenal domain relative to the membrane cavities (Video 3). The lumenal positioning is consistent for all three of our closed-gate conformation reconstructions (hEMC detergent, yEMC nanodisc, yEMC detergent). By contrast, the one map with an open gated cavity displayed a lumenal rotation and concomitant shifts in position of the hEMC1 brace helix (Figure 7—figure supplement 1). Indeed, our set of interfacial hEMC1 brace mutants described above (Figure 6F–G, Figure 6—figure supplement 2B–C), showed differing client phenotypes when mutated from either the hEMC1 or the hEMC5 side. This suggests a complex conformational interplay between lumenal and transmembrane domains during the engagement of diverse client types.

We investigated several known disease mutations in both conserved and human-specific regions of hEMC1 (Figure 7C–D, Figure 7—figure supplements 2–3; Harel et al., 2016; Abu-Safieh et al., 2013; Amberger et al., 2019). One of these disease-associated residues sits near the anchor point for the lumenal hEMC4 transmembrane gate helix (hEMC1^R881C), while the majority are found farther from the membrane (hEMC1^G868R, hEMC1^A144T, hEMC1^T82M) (Figure 7C–D, Figure 7—figure supplement 2B). Incorporating each of these disease mutations into our EMC functional assay resulted in lower levels of the N-cytoplasmic polytopic client (TMEM97) and an increase in the level of the tail-anchored client (SQS^378-410), discussed in more detail below.

Two different hEMC1 mutants associated with cerebellar atrophy, visual impairment, and psychomotor retardation (hEMC1^T82M, hEMC1^G868R) map to the hinge region between the hEMC1 beta propellers where hEMC7 binds (Figure 7D). Both the mutants at this protein-protein interface resulted in depletion of the N-cytoplasmic polytopic client (TMEM97). EMC7 and EMC10 form beta-sandwich domains on either side of the membrane-proximal beta-propeller of EMC1 and contact each other across the EMC1 surface. Consistent with our structures, coupling of these subunits is supported by the prior finding that in the absence of EMC7, EMC10 is also lost from the complex while the other EMC components appear unaffected (Shurtleff et al., 2018). EMC7 and EMC10 have been proposed to be auxiliary components with weaker phenotypes compared to core EMC subunits (Jonikas et al., 2009; Shurtleff et al., 2018; Dickinson et al., 2016). Upon deleting yEMC7, multi-pass transmembrane clients are retained in the ER but tail-anchored clients, including SQS-homolog Erg9, decrease in abundance (Shurtleff et al., 2018).

Several features of our data suggest dynamic association of hEMC7. Density for the hEMC7 beta-sandwich at the hinge between the two hEMC1 beta propellers was relatively weak in the consensus hEMC nanodisc map (Figure 2—figure supplement 4). Additional rounds of 3D classification revealed two distinct classes, one with clear density for hEMC7, and one with weak density in this region. Mass spectrometric analysis of purified hEMC, however, revealed that the abundance of hEMC7 was similar to that of the other hEMC components (Figure 1—figure supplement 2; Supplementary file 1). Both reconstructions, with and without density for the hEMC7 lumenal domain, displayed well-resolved density for hEMC10. Together, we conclude that hEMC7 is associated with hEMC1 in two different conformational states of hEMC7 with potentially distinct functions.

The OMIM database (Amberger et al., 2019) lists a mutation of unknown significance linked to retinitis pigmentosa (hEMC1^A144T) residing in the EMC1 distal propeller (Figure 7—figure supplement 2). Additionally, we also generated mutations in two surface exposed patches of the membrane-distal EMC1 beta-propeller projecting into the lumen (hEMC1^D31K, hEMC1^R69D, hEMC1^G71S, hEMC1^{H93D + E138D + N282K}, Figure 7E, Figure 7—figure supplements 2–3). Overall, these mutations displayed the same client effect: a decrease in the N-cytoplasmic polytopic client reporter (TMEM97), no change in the N-lumenal polytopic client reporter (B1AR), and accumulation of the tail-anchored client reporter (SQS^378-410). Upon identifying antibodies against yEMC, we observed that the top two antibodies bind to a similar extended loop in the distal yEMC1 beta-propeller, perhaps suggesting that this site is accessible for co-factor binding in the ER. Intriguingly, this region of the lumenal domain corresponds to the region where hEMC1 has an expanded distal beta-propeller. Taken together, the data provide evidence that the lumenal domain is functionally coupled to the broader EMC role in transmembrane client stabilization. Moreover, these data support that the EMC is acting as a holdase chaperone to shield polytopic clients from degradation while they are folding to their functional form.

EMC3’s fold at the interface between the cytoplasm and membrane forms the core of the gated cavity and is reminiscent of proteins from the YidC family of insertases (Borowska et al., 2015; Dalbey and Kuhn, 2015; Anghel et al., 2017). Indeed, mutations in either the cytoplasmic or transmembrane domains of EMC3 establish that these features are critical for terminal helix insertase activity. In light of our observation of multiple gate conformations, we speculate that these conformations modulate insertion and release into the ER membrane.

Notably, mutating the surface of the cytoplasmic cap, which extends beyond the EMC3 cytoplasmic helices toward EMC8/9, resulted in an unexpected increase in C-tail anchor client (SQS^378-410) abundance. Of the three clients analyzed, SQS was the only one to show enhanced levels. It is unclear if this enhancement is SQS-specific or representative more broadly of all post-translationally targeted EMC tail-anchored clients. Future studies will be required to address if this is due to regulated insertion of SQS by the EMC, parallel pathways for inserting SQS into the membrane (i.e. mediated by TRC40/GET), and/or slower cytoplasmic clearance of chaperone-bound SQS.

Post-translational insertase clients have previously been shown to be targeted to the ER by cytoplasmic chaperones (Guna et al., 2018). Structural analysis and coupled mutagenesis, from our and recent studies (O'Donnell et al., 2020; Pleiner et al., 2020; Bai et al., 2020), suggest that clients then engage the cytoplasmic domain of the EMC, the transmembrane gate opens, the terminal helix is inserted into the EMC-gated cavity, and then another conformational change would allow for release into the lipid bilayer (Figure 8A–B). Further studies are needed to establish the precise C-terminal client range, as most tail anchor clients have been shown to be inserted by the GET (in yeast) or WRB (in human) complexes (Denic et al., 2013; Mateja and Keenan, 2018).

Both the N-terminal (B1AR) and C-terminal insertase (SQS) clients depend on the EMC-gated cavity. Indeed, both the SQS tail-anchored helix and the first transmembrane helix of B1AR are moderately hydrophobic, with polar residues near the cytoplasmic end of the transmembrane helix, and both showed a strong dependence on the gated cavity. Nevertheless, our panel of mutants revealed some notable differences in the handling of these two client types. B1AR showed more dependence than SQS on the lipid-filled cavity in contrast to mutants elsewhere in the complex. Consistent with this, a number of mutations, primarily in the gated cavity, show residues of importance to both SQS and B1AR. However, there are also a number of mutations that appear to only affect SQS. One possible reason could be due to differences in the mechanism of initial engagement: SQS is targeted to the ER by cytoplasmic chaperones, while B1AR is targeted by SRP. Another key difference is that B1AR is polytopic and needs to overcome the additional challenge of tertiary transmembrane packing to reach its folded state. This work provides support for a model where the EMC inserts both types of terminal transmembrane helices into the gated cavity with differences in initial targeting and perhaps release into the lipid environment (Figure 8C–D). Future work will address the interplay between B1AR synthesis and its co-translational engagement with the translocon to ascertain whether there is a direct handoff between the translocon and the EMC or the EMC acts post-translationally to insert the N-terminal helix of B1AR.

Unlike the two terminal insertase clients we investigated, TMEM97 biogenesis was negatively impacted by mutation of the lumenal EMC1. The depletion of TMEM97 observed in these mutant backgrounds is consistent with the lumenal domain contributing to a holdase chaperone function, passively shielding its client while it is being synthesized and/or folded (Zhang et al., 2017). Interestingly, the diametrically opposed phenotype of mutants in the EMC lumenal domain on SQS raises the possibility that occupancy by one type of client can support an EMC conformation that is unfavorable for receiving the other. Alternative conformations could establish competition between client types for EMC occupancy. One explanation for this observation is that there is a conformational change between the insertase-active versus the holdase-active states. Interestingly, we identified at least two EMC conformations in our collection of structures, and EMC may adopt different conformations in various client and cofactor-engaged states.

In yeast, the polytopic clients co-purifying with the EMC are also glycosylated. One possible model is that the putative carbohydrate-binding domains in EMC7 or EMC10 directly contribute to engagement with client proteins. We speculate post-translational modifications on clients and the EMC could modulate function including client binding, chaperone binding, or regulating signaling in response to cellular cues.

Multi-pass transmembrane proteins require membrane factors to assist after insertion into the membrane to pack transmembrane helices in the correct order and topology. We propose that the EMC may act as a chaperone holdase to facilitate one of the following: helix and lipid packing, shielding from degradation while synthesis is in progress, or assisting in the assembly of multi-protein transmembrane complex formation. This is consistent with observations that in the absence of the EMC numerous integral membrane proteins are degraded (Shurtleff et al., 2018; Volkmar et al., 2019; Tian et al., 2019). Direct interactions with multi-pass transmembrane proteins have been shown previously (Shurtleff et al., 2018; Coelho et al., 2019). Furthermore, EMC dependence of internal transmembrane domain segments has also been established (Ngo et al., 2019; Hiramatsu et al., 2019). In the absence of yEMC7, a primarily lumenal subunit, a polytopic membrane protein was retained for longer in the ER, suggesting the possibility that yEMC7 may be involved in client release from the EMC. We propose a model where the EMC engages polytopic clients either during or directly after translation and remains bound until the client is released either to the membrane environment directly or handed off to client-specific and general ER chaperones (Figure 8E–F). It remains to be seen whether these polytopic clients directly engage with the lipid-filled cavity or the gated cavity or the lumen domain, what the determinants for engaging with a client or release into the membrane are, and how the EMC fits into the broader ER lumenal chaperone network.

Why does the cell use a multifunctional EMC molecular machinery rather than specialized machinery for each of the functions encompassed by the EMC? Considering that the cell already has general machinery (Sec61 translocon) and tail-anchor insertase machinery (GET/TRC complex), we speculate that the EMC coordinates biogenesis of diverse membrane proteins. Several observations suggest broader roles of the EMC as an integrator of information sensing the protein and lipid environment and coordinating its multiple activities, including the regulating the biogenesis of membrane proteins. For example, the initial identification of the EMC included numerous genetic interactions with both protein and lipid synthesis factors in yeast (Jonikas et al., 2009) and these disparate interdependencies have been subsequently observed in numerous species including human EMC (Lahiri et al., 2014; Tang et al., 2017; Guna et al., 2018; Volkmar et al., 2019; Volkmar and Christianson, 2020). Also, several client proteins are enzymes or cofactors involved in multiple stages of lipid synthesis or trafficking, and this may provide a unifying explanation for the range of genetic interactions and co-essentiality observations reported to date (Guna et al., 2018; Shurtleff et al., 2018; Volkmar et al., 2019; Tian et al., 2019; Wainberg et al., 2019; Corradi et al., 2019; Volkmar and Christianson, 2020). Perhaps by facilitating the insertion of sterol synthesis protein SQS, the EMC allows for modulation of local membrane thickness and lipid composition to accommodate differences within the broad range of membrane proteins being synthesized. In this regard, one structural feature of particular interest is the EMC1 amphipathic brace, which resides adjacent to the lipid-filled cavity. This conserved feature sits within the interfacial membrane boundary, raising the possibility that it can modulate the lipid or protein composition of this cavity. Notably, several other membrane proteins involved in ER homeostasis, including Opi1 and Ire1, also contain amphipathic helices that have been proposed to sense the properties of the lipid bilayer (Volmer et al., 2013; Jacquemyn et al., 2017; Halbleib et al., 2017; Hofbauer et al., 2018; Cho et al., 2019). Future work will explore how the EMC overall, and the EMC1 brace helix in particular, govern client release into the membrane, interface with the local structure of the lipid bilayer, and play roles in specific client-lipid interactions.

In addition to the three client classes we investigate here, it is clear that EMC has a broader range of clients including multi-protein assemblies (Richard et al., 2013; Talbot et al., 2019), lipid-modulating proteins (Volkmar et al., 2019), lipid-binding proteins (Salas-Estrada et al., 2018; Sejdiu and Tieleman, 2020), and those with helices that do not span the bilayer (Lin et al., 2019; Ngo et al., 2019). The compartmentalization and interdependence that we observe for effects of mutations on client handling provide a foundation for understanding this multifunctionality. We propose that the complexity of the EMC machine, combining insertase and holdase chaperone functions within one molecular machine, has arisen to mitigate the error prone biogenesis of a diverse range of membrane spanning proteins in the dynamic environment of the ER.

K562 dCas9 KRAB cells were grown in RPMI 1640 (GIBCO) with 25 mM HEPES, 2 mM l-glutamine, 2 g/L NaHCO₃ and supplemented with 10% (v/v) fetal bovine serum (FBS), 100 units/mL penicillin, 100 μg/mL streptomycin, 2 mM l-glutamine. HEK293T cells were grown in Dulbecco’s modified eagle medium (DMEM, GIBCO) with 25 mM d-glucose, 3.7 g/L NaHCO₃, 4 mM l-glutamine and supplemented with 10% (v/v) FBS, 100 units/mL penicillin, 100 μg/mL streptomycin. All cell lines were grown at 37°C. All cell lines were periodically tested for Mycoplasma contamination using the MycoAlert Plus Mycoplasma detection kit (Lonza).

Lentivirus was generated by transfecting HEK39T cells with standard fourth-generation packaging vectors using TransIT-LT1 Transfection Reagent (Mirus Bio). Media was changed 10 hr post-transfection. Viral supernatant was harvested 60 hr after transfection, filtered through 0.45 μm PVDF filters and frozen prior to transduction.

A single and dual knockout guide system was developed in the pX458 backbone (Addgene plasmid # 48138) with guides targeting hEMC1, hEMC2, hEMC3, or hEMC5 (Key Resources table). Targeting guides were selected using the Broad’s guide selection tool (https://portals.broadinstitute.org/gpp/public/analysis-tools/sgrna-design). For the single hEMC5 knockout system, an hEMC5 targeting guide was cloned into pX458 by digesting with BbsI and ligating to annealed oligos for the hEMC5 sgRNA. For the dual knockout system, a four-step cloning process generated the final knockout plasmid: (1) Each of the two guides targeting the same locus were individually cloned into pX458. (2) Then pX458_sgRNA1 was digested with XbaI. (3) SgRNA2 cassette from pX458_sgRNA2 was PCR amplified with oligos containing overhangs spanning the XbaI cloning site and purified. (4) Finally, the final dual guide vector was generated by Gibson cloning (NEBuilder).

To generate the hEMC knockout cell lines, K562 dCas9 KRAB cells were nucleofected with the respective hEMC knockout plasmids using Lonza SF Cell Line 96-well Nucleofector Kit (V4SC-2096). Two days post-nucleofection, GFP-positive cells were single cell sorted into 96-well plates using BD FACS AriaII. After colonies from single cells grew out, genomic DNA was isolated using QuickExtract (Lucigen), the sgRNA-targeted sites were PCR amplified and then NGS-sequenced via Genewiz’s EZ-Amplicon service. Sequencing data was analyzed and aligned to the respective reference alleles in the human genome. Clones whose alleles harbored only indel mutations for hEMC1, hEMC2, hEMC3, and hEMC5 (full knockouts) respectively were further validated on the protein level.

Dual client reporters for TMEM97, ADRB1 (protein name: B1AR), and FDFT1 (protein name: SQS) were introduced lentivirally into each of the EMC1, EMC2, EMC3, and EMC5 knockout cell lines. TMEM97 and ADRB1 full-length sequences were used with a C-terminal tag -mCherry-P2A-GFP. The sequence for FDFT1 transmembrane domain (SQS^378-410) was tagged N-terminally with GFP-P2A-mCherry- and an opsin tag on the C-terminus as used in a prior study (Guna et al., 2018). Three days post-transduction, GFP/mCherry-positive cells were sorted on BDAriaII. Sequences for these constructs are available in the Supplementary file 5.

The EMC mutant genes were synthesized and cloned by Twist into pKDP119-SFFV-[insert site]-IRES-Puro-P2A-BFP. For hEMC subunit mutation details refer to the Key Resources Table, for sequences refer to Supplementary file 5. Mutant hEMC cell lines were generated by lentiviral introduction of the respective hEMC mutant subunit into the respective knockout cell lines (hEMC1, hEMC2, hEMC3, or hEMC5) containing the dual fluorescent reporters for each EMC client (pKDP110_ADRB1_mCherry_P2A_GFP, pKDP111_TMEM97_mCherry_P2A_GFP, or GFP_P2A_mCherry_FDFT1_TMD_opsintag). The expression of each fluorescent reporter was read out 6 days after puromycin selection in each of the hEMC mutant cell lines.

For each hEMC mutant cell line, 20,000 live cells were recorded on Attune NxT flow cytometer. FlowCal flow analysis package was used for analysis in Python. First, live cells were gated based on FSC/SSC. Then GFP (BL1-A) and mChery (YL2-A) were plotted for each mutant and control cell line. mCherry:GFP intensity ratios were calculated for individual cells in each cell line. Fluorescence ratios for each substrate in an hEMC mutant cell line were normalized to the mCherry:GFP ratio of the same substrate in the hEMC wild-type rescue cell line. Distributions of fluorescence ratios were plotted as histograms in Python using seaborn.

We performed bootstrap estimates of the mean of normalized mCherry/GFP ratio from the FACS data. For bootstrapping, we performed 1000 iterations with 50 cells/iteration to fit normal distributions. We performed two separate one-sided T-tests at a p-value cutoff of 0.01 between each mutant and the respective subunit WT to test for significant decreases or increases in ratios based on bootstrapped estimates of the mean. These statistics are contained in the files ‘filtered_final_pvalues.01cutoff.lo.csv’ and ‘filtered_final_pvalues.01cutoff.hi.csv’ respectively. Statistics were generated for the EMC-independent membrane protein controls (‘stats.membrane.controls.lo’, ‘stats.membrane.controls.hi’) and for the mCherry-p2a-GFP control (‘.mcherry.p2a.gfp.control.lo’, ‘stats.mcherry.p2a.gfp.control.hi’). Values can be found in Supplementary file 2.

Cell pellets were lysed using lysis buffer (20 mM Tris pH 7.5, 150 mM NaCl, 5 MgCl₂, 1% Triton x-100, 1 mM DTT, 24 U/ml) Turbo DNase (Ambion). Clarified lysate was quantified and samples were boiled with 4x LDS sample (Thermo Fisher, NP0007) buffer for 5 min at 95°C. Samples were separated on 4–12% or 12% Bolt Bis-Tris Plus Gels (Invitrogen, NP0322PK2). Proteins were transferred onto nitrocellulose membranes using Bio-Rad Trans-Blot Turbo transfer system. Membranes were blocked in Odyssey Blocking Buffer (LI-COR, 927–50000) for an hour at room temperature. Blocked membranes were incubated with primary antibody diluted in TBST and incubated overnight at 4°C on a shaker. Primary antibodies were detected by incubating membranes with 1:10,000 dilution of IRDye-conjugated (LI-COR) secondary anti-mouse and anti-rabbit antibodies for 1 hr at room temperature. Blots were visualized using LI-COR imaging system. The primary antibodies used in this study are listed in the Key Resources table.

Strain BY4741 and BY4742 were used as the wild-type parental strains for the creation of the yEMC overexpression strain. Yeast homologous recombination (Rothstein, 1991) was used to generate yeast strains. For the overexpression strain, the endogenous promotor for each yEMC subunit (yEMC1, yEMC2, yEMC3, yEMC4, yEMC5, yEMC6, yEMC7, yEMC10) were replaced with a TEF2 promoter. In addition, EMC5 was tagged at the C-terminus with linker-TEV-linker-3xFlag. Auxotrophic markers and drug selection markers in both BY4741 and BY4742 were employed to add this promoter modification to all these eight subunits and the two strains were crossed to create the resulting BY4743 strain used for immunoprecipitation. Endogenous EMC yeast strain was made using W303a wild-type parental background (leu2-3,−112; his3-11,−15; trp1-1; ura3-1; ade2-1; can1-100; MATa). Homologous recombination was used to integrate a linker-TEV-linker-3xFlag at the C-terminus of yEMC coding sequence. Genomic PCR was conducted to verify integration.

Fabs were identified as described in these studies (Kim et al., 2011; Wu et al., 2012). Overexpressed yEMC solubilized in DDM as described above was biotinylated and streptavidin magnetic beads were used to capture yEMC, which was then subjected to a Fab phage library. Unbound Fabs were washed away and then binding Fabs were eluted and analyzed by ELISA. Two Fabs were identified binding yEMC, Fab DH4 and DE4.

Plasmid with either Fab DH4 or DE4 were transformed into BL21 Gold Star cells and plated onto agarose plates with 2x YT + 2% glucose + Ampicillin. Cultures were inoculated from resulting colonies for overnight growth at 30°C into 2xYT + 2% glucose + Amp. In the morning dilute overnight culture to OD₆₀₀ of 0.05 in 1L, in a 2.8 L flask of 2xYT + 0.1% glucose + Amp. Grow the culture at 180 rpm at 37°C shaker until OD₆₀₀ of 0.6, then, switch to shaking at 19°C for 1 hr. Next, induce with 0.4 mM IPTG. Shake at 180 rpm at 19°C for 18–20 hr. Spin 1L cultures down at 3500 rpm in large Beckman Centrifuge at 4°C for 20 min in (8.1 rotor). Discard media and gently resuspend cell pellet in ice-cold 20 ml in Buffer 1 (0.2 M Tris pH 8.0, 0.5 mM EDTA, 0.5 M Sucrose) on ice. Transfer the resuspended cells from step 2 into two smaller JLA 25.5 centrifuge tubes. Add 20 mL of ice cold ddH₂O with 2x protease inhibitor cocktail (Roche Complete Ultra, Millipore Sigma 5056489001) from step 3 to the resuspended pellets. Incubate at on ice for 1 hr occasionally swirling samples gently. Spin periplasmic fractions at 13,000 x g for 15 min, 4°C, rotor 25.50. Wash 500 µL Ni resin (Qiagen, Ni-NTA, 30210) per periplasmic fraction four times in Buffer 2 (50 mM Tris pH 8.0, 250 mM NaCl). Add MgCl₂ and imidazole to a final concentration of 10 mM to each periplasmic fraction. Add beads to periplasmic fractions and nutate at 4°C for 2 hr. Spin down beads at 2000 x g, 10 min, 4°C. Transfer beads either to a 50-mL gravity column. Wash the beads with 20 column volumes of Buffer 3 (50 mM Tris pH 8.0, 500 mM NaCl, 20 mM Imidazole). Elute protein with three column volumes of Buffer 4 (50 mM Tris pH 8.0, 500 mM NaCl, 300 mM Imidazole). Analyze eluate by SDS-PAGE 4–12% Invitrogen (Invitrogen, NP0321PK2). Fabs as two bands run around 30 kDa in reducing conditions, or 50 kDa in non-reducing conditions. Dialyze eluate O/N in Dialysis cassette 10 kD molecular weight cutoff at 4°C against 150 mM KOAc, 20 mM HEPES pH 6.8.

The OE-Emc5-3xflag yeast strain were grown in YEPD media in a 40 L fermenter, harvested and flash frozen in liquid nitrogen. Cell pellets were thawed and diluted in lysis buffer (50 mM HEPES pH 6.8, 150 mM KOAc, 2 mM MgOAc, 1 mM CaCl2, 0.2M Sorbital, 2x Protease Inhibitor). Bead beating (10 times → 1 min on, 2 min off) was used to lyse cells. For 25 g of cells, 0.1 mm cold beads were added and lysis buffer up to the top of the 50 mL canister. After lysis, beads were filtered and solution centrifuged at 10,000 xg for 10 min. Supernatants were ultracentrifuged at 42,000 RPM (Ti 45 rotor) for 2 hr. Supernatant was discarded. Membrane pellet was combined with the lipid layer, and resuspended in lysis buffer and then a precooled dounce homogenizer was used to dounce 20 times. Membranes were aliquoted and flash frozen in liquid nitrogen. On ice, 150 mL of solubilization buffer (50 mM HEPES pH 6.8, 150 mM KOAc, 2 mM MgOAc, 1 mM CaCl2, 15% glycerol, 1% b-DDM, 2x Protease Inhibitor) was added incrementally to 7.5 g of thawing membranes, nutated at 4°C for 1 hr in JA 25.5 rotor tubes, and centrifuged at 20,000 rpm for 45 min. Meanwhile 2.5 mL of αFLAG agarose beads (Millipore A2220) were rinsed in 50 mL of low-salt buffer (50 mM HEPES pH 6.8, 150 mM KOAc). Supernatant was added to αFLAG beads and nutated at 4°C for 2 hr. Resulting solution was applied over a glass column. After flowing through unbound solution, αFLAG beads were washed with 100 mL low-salt buffer, 100 mL high-salt buffer (50 mM HEPES pH 6.8, 300 mM KOAc, 0.05% b-DDM), and 100 mL low-salt buffer. αFLAG beads were resuspended in 10 mL of low-salt buffer and 300 µL of TEV (1.15 mg/mL) was added and nutated overnight at 4°C. Removed supernatant from beads by low-speed spin and applied over 500 µL of NiNTA beads equilibrated with low-salt buffer to remove excess TEV. Flow through glass column and collect supernatant. Using a 100 kD concentrator (Millipore, UFC910008) solution was concentrated to 2 mg/mL. Concentrated EMC protein was applied to the Akta Explorer Superose 6 Increase column (Cytiva, 29091596) for size exclusion chromatography in the size exclusion buffer (20 mM HEPES pH 6.8, 150 mM KOAc, 0.05% b-DDM). Fractions were evaluated by SDS-PAGE Coomassie stain and negative stain electron microscopy then EMC peak fractions were pooled and incubated with 2x molar excess of Fab, either Fab DH4 or Fab DE4, for 30 min on ice. Solution was applied to Akta Explorer Superose 6 Increase for size exclusion of Fab bound EMC. Resulting EMC-Fab fractions were evaluated by SDS-PAGE Coomassie stain and EMC-Fab peak fractions were pooled.

Yeast was grown in rich media (YPAD) in a 65L fermenter until OD 2.6. Cell pellets were harvested and flash frozen in liquid nitrogen. Pellets were ground using three cycles in a French press. As above, the resulting solution was ultracentrifuged to separate membranes, dounced to homogenize, and flash frozen in liquid nitrogen. Thawed membranes were solubilized in 1% b-DDM (Anatrace, D310) nutating at 4°C for 1 hr then centrifuged to separate solubilized membranes from the pellet. Supernatant was applied to equilibrated αFLAG beads, nutated at 4°C for 1 hr, and applied over a disposable plastic column at 4°C. αFLAG beads were washed with low-salt buffer and high salt buffer. Then washed with low-salt buffer with b-DDM+CHS (Anatrace, CH210) (10:1) in place of b-DDM. αFLAG beads were then transferred to a 15-mL Eppendorf tube for TEC cleavage and nanodisc reconstitution.

Bio-Beads SM-2 (Bio-Rad) were prepared ~400 µL biobeads, rinsing with EtOH, and then water four times. Yeast Extract Total (Avanti Polar Lipids, 190000 C-100mg) was prepared by transferring chloroform resuspended solution to a glass vial, drying the lipids into a film with nitrogen gas, drying in a vacuum desiccator overnight, and then solubilizing the lipids first in water and then in size exclusion buffer with DDM+CHS by bath sonication, aliquots stored at −20°C until use. A total of 200 µL of TEV protease (5 mg/mL) and 150 µL of 1 mg/mL Yeast Total Extract solubilized in b-DDM+CHS, at room temperature for 30 min. Then added MSP1D1, purified as described previously (Ritchie et al., 2009), to a ratio of 200:10:1 (Yeast total extract:MSP1D1:EMC), at 4°C for 10 min. Then activated Bio-Beads SM-2 (Bio-Rad), ~300 µL, were added and nutated overnight. On-bead reconstitution employed adapted from Laverty et al., 2019. In the morning, ~100 µL more Bio-Beads SM-2 (Bio-Rad) were added and 2x molar excess of FabDH4, nutated for another hour. Beads and solution applied to an EconoPac column (Bio-Rad). Flow through was collected and solution was applied to a 100 kD (Amicon) concentrator. Resulting concentrated EMC was applied to the Akta Explorer Superose 6 Increase column for size exclusion chromatography. Peak fractions were pooled for SDS-PAGE Coomassie stain, negative stain, and cryo-EM evaluation. Key reagents used are provided in the Key Resource Table.

Structural biology applications used in this project were compiled and configured by SBGrid (Morin et al., 2013). The yeast EMC structure was built de novo using Coot (version 0.8.7 and 0.9) and UCSF ChimeraX (Goddard et al., 2018). Visible secondary structure was built by hand for the entire structure using overlays of the yEMC detergent consensus map as well as the yEMC nanodisc unsharpened and sharpened map. Starting with the best resolved transmembrane helices, sequence was placed for each of the predicted transmembrane helices, using TMHMM (Krogh et al., 2001), in the yEMC proteins. Visual inspection for landmark residues (tryptophan, tyrosine, leucine, and proline) in the sequences that correlated with the position of well densities as well as fit correlation in UCSF Chimera was computed to assign identities for yEMC1, yEMC3, yEMC5, and yEMC6. Connectivity between the EMC1 assigned helix to the lumenal domain was used to start assigning sequence for the lumenal portion of EMC1. Secondary structure prediction was computed for all yEMC proteins using Phyre2 (Kelley et al., 2015) and Quick2D, a tool within the Max-Plank Institute for Developmental Biology Bioinformatics Toolkit that visualizes several different secondary structure predictors (Jones, 1999; Cuff and Barton, 2000; Ouali and King, 2000; Rost, 2001; Lupas et al., 1991; Jones et al., 1994; Ward et al., 2004; Peng et al., 2006; Obradovic et al., 2005). Secondary structure prediction was used to check and guide sequence assignment of beta strands and helices. Next several homology models were computed and overlain for yEMC2, with a predicted TPR structural domain, using Robetta (Raman et al., 2009; Song et al., 2013), I-TASSER (Zhang, 2008; Roy et al., 2010; Yang et al., 2015), Phyre2 (Kelley et al., 2015), and RaptorX (Källberg et al., 2012). These were used in addition to secondary structure prediction to guide sequence assignment, loop building, and helical packing. Fab DH4 starting structure was computed using Phyre2 1-to-1 threading against a crystal structure of a monoclonal Fab (PDB 1M71, Vyas et al., 2002). EMC3, EMC5, and EMC6 were built off of the transmembrane helices using sphere refinement, real space refinement, regularization, and visual monitoring of the Ramachandran plot in Coot. EMC7 and EMC10 both form beta sandwich folds on the exterior of the EMC1 lumenal domain, beta strand sequence was placed for both in both densities, position of aromatic residues and loop length differed between the two allowing assignment of each. After building EMC1-3, EMC5-7, and EMC10, there remained several transmembrane helices and a beta strand fitted into the lumen but not connected to EMC1, EMC7, or EMC10. The resolution of the lumenal domain is better than 3 Å in most parts allowing for sequence placement of the EMC4 C-terminus and C-terminal transmembrane helix. The connectivity of the transmembrane helix to the cytoplasmic domain was not resolved. However, there was an additional poorly resolved short helix and loop density in the cytoplasmic domain which was assigned to EMC4.

Two poorly resolved transmembrane helices remained, however, due to the fact they did not have clear connectivity to any built strand, poly alanine alpha helices were built in but not assigned to a yEMC protein (Figure 5—figure supplement 5). EMC4 had density in the cytoplasmic domain as well as the lumenal domain, suggesting that it has either one or three transmembrane passes. EMC7 and EMC10 were predicted to have transmembrane helices, however, the connection between the lumenal densities and those predicted transmembrane helices was not clear. Additional density that was not built into was visualized in UCSF ChimeraX (Goddard et al., 2018) and allowed for subsequent assignment of several glycosylated residues and one POPC molecule. Each subunit was built in a separate pdb file and subjected to iterative rounds of phenix.real_space_refine (Adams et al., 2011; Liebschner et al., 2019) into segmented maps preceded and followed by adjustment in Coot. Manual assignment of secondary structure restraints was used and improved during Phenix refinement. Once all of the well-resolved secondary structure was assigned to yEMC subunits, PDBs were combined and subjected to iterative rounds of phenix.real_space_refine (Adams et al., 2011; Afonine et al., 2018; Liebschner et al., 2019) in the unsharpened and then sharpened maps. Loops were built back where the connectivity was clear and then refined again in Phenix and Coot. PDBs were prepared for refinement steps using phenix.reduce to add hydrogens throughout refinement steps, ReadySet to generate cif restraints, and Phenix PDB preparation tool for creating mmCIF files for deposition. Representative regions of the model as well as the map-to-model FSC can be found in Figure 2—figure supplement 5. Model for the yEMC nanodisc sample was used to generate reference model restraints for phenix real space refinement of yEMC DDM model.

T-coffee PSI-Coffee extension (Notredame et al., 2000) was used to compute sequence alignments between yEMC, hEMC, and homologous proteins (Figure 1—figure supplements 6–7, Figure 3—figure supplement 3, Figure 5—figure supplement 3). Outputs of these alignments were visualized in Jalview (Waterhouse et al., 2009) for figure creation and colored by ClustalX convention.

All figures were assembled and edited in Adobe Illustrator. Figure 1 and Figure 1—figure supplement 4 were created using BioRender. All the visualization, structure figures, and structure videos were made using UCSF ChimeraX 1.0 (Goddard et al., 2018) and UCSF Chimera 1.14 (Pettersen et al., 2004). Flow cytometry plots were generated in Python and labeled in Adobe Illustrator.