Structures of the scanning and engaged states of the mammalian SRP-ribosome complex

1Roughly 20% of eukaryotic genes encode membrane proteins that must be inserted into the endoplasmic reticulum (ER) early in their biogenesis. The majority of these proteins are recognized co-translationally at the ribosome by the signal recognition particle (SRP) and targeted via the SRP receptor to the Sec61 translocon for insertion. SRP has several critical functions in coordinating this targeting reaction. First, SRP must selectively recognize its cargo via a hydrophobic targeting signal. Second, it must shield this hydrophobic domain to preclude inappropriate interactions. Third, cargo-loaded SRP must interact with its receptor and release the cargo to the translocon. All of these events must occur before substantial synthesis beyond the targeting signal precludes successful engagement of the translocon.

2In eukaryotes, SRP is composed of six protein subunits assembled on a ∼300 nucleotide RNA. The elongated assembly is operationally divided into an S domain and an Alu domain. The S domain is involved in recognizing the targeting signal and interacting with the SRP receptor. Its functionally essential SRP54 subunit contains three modules: an N domain that interacts with the ribosome near its exit tunnel, an M domain responsible for signal recognition, and a G domain that (together with the N domain) coordinates receptor targeting. The Alu domain interacts near the GTPase center at the interface between the two ribosomal subunits. This interaction is thought to compete with translation factors to slow nascent polypeptide elongation and provide additional time for successful targeting. Many prokaryotes lack an Alu domain and instead contain a simplified SRP composed solely of an SRP54 homolog and a short ∼110 nucleotide RNA.

3Despite extensive biochemical and structural studies, several basic aspects of SRP biology remain unresolved. First, the interactions between SRP and the ribosome are understood only in general terms, with regions of close proximity deduced from moderate resolution cryo-EM structures. The specific residues and chemical details of the binding surfaces are not known. Furthermore, the eukaryotic SRP-ribosome interaction has only been visualized for mammalian SRP bound to the plant ribosome. Although functional for translocation, this heterologous complex strongly arrests translation. By contrast, endogenous SRP in a homologous mammalian system has a remarkably subtle effect on translation elongation. Hence, the nature of the Alu domain-ribosome interaction and its relationship to translation factor binding in a native homologous system is unknown.

4Second, the molecular events leading to signal recognition by the SRP M-domain are also incompletely resolved. Structural analyses of the M domain show that it contains a large hydrophobic groove capable of housing a hydrophobic signal. While this has convincingly established the binding site, the configuration of the pre-engaged state of the M domain and how it transitions to the engaged state are less clear. An exposed hydrophobic groove, as seen in isolated M domain crystal structures (e.g., ), would seem energetically unfavorable in an aqueous environment and may exhibit promiscuous interactions. Indeed, other factors that bind hydrophobic clients contain hydrophobic grooves that are occluded in the unengaged state. Thus, it is not known how SRP's substrate binding groove is kept shielded until it encounters its hydrophobic clients specifically at the ribosomal exit tunnel.

5Third, the timing of SRP recruitment to the ribosome relative to its engagement of the targeting signal remains a source of considerable debate. The simplest and earliest idea was that SRP has high affinity and selectivity for a ribosome exposing a hydrophobic signal. However, such a model would not easily explain why far more abundant hydrophobic binding proteins do not outcompete SRP, why SRP cannot bind those same sequences post-translationally, and how SRP can rapidly find these transient and rare species to prevent aggregation. Thus, it was suggested that SRP may repeatedly scan all translating ribosomes until the appropriate cargo is encountered.

6Studies in the simplified bacterial system have extensively debated both the basic premise and mechanistic details of this scanning model. The extent to which these prokaryotic studies apply to eukaryotic SRP is unclear and has been minimally studied. Not only is its ribosome interaction more complex, but the Alu domain merits consideration of how SRP recruitment is integrated into the translation cycle. Furthermore, Escherichia coli SRP seems to scan any translating ribosome, while eukaryotic SRP may display a further preference for ribosomes containing a hydrophobic signal inside the exit tunnel. At present, the existence and properties of this scanning mode remain a point of uncertainty, but is crucial for understanding how hydrophobic signals are promptly found by SRP before their cytosolic exposure leads to off-pathway fates.

7Here we demonstrate that endogenous SRP is efficiently recruited to mammalian ribosomes containing a transmembrane domain (TMD) inside the ribosomal exit tunnel. The cryo-EM reconstruction of this native complex provided the first structure of SRP bound to the ribosome in a ‘scanning’ mode. A parallel structure of SRP bound after the TMD has emerged from the ribosome revealed the conformational changes that accompany TMD engagement by SRP. The structures show how the hydrophobic cavity of the SRP M domain is occluded by an amphipathic C-terminal ‘placeholder’ helix until its displacement by a bona fide TMD. Competition assays between SRP and a translational GTPase provide insight into how scanning and engagement are integrated during polypeptide elongation to maximize targeting efficiency while limiting TMD exposure to the cytosol. These findings have functional implications for the mechanism of membrane protein recognition and targeting.

8To examine the timing and specificity of SRP recruitment, a variety of stalled ribosome-nascent chain complexes (RNCs) were prepared by in vitro translation of truncated mRNAs in rabbit reticulocyte lysate, affinity purified via the nascent chain, and analyzed biochemically. Constructs and truncation points were designed to position a 20 residue TMD either within the exit tunnel or just emerged from the ribosome (Figure 1A). These two states were chosen to represent moments when SRP would be either scanning in anticipation of, or successfully engaged with, the TMD. Specificity was controlled by using a mutant (hereafter termed 3R) in which three hydrophobic residues in the TMD were changed to arginine.

9RNCs with an intact TMD, either exposed or buried in the tunnel, efficiently recovered SRP as judged by immunoblotting for SRP54 (Figure 1B). Furthermore, the SRP68 and SRP72 subunits were recovered at sufficient levels to be clearly visible on the stained gel of the purified complexes (Figure 1B). Indeed, immunoblotting of the unbound fraction after affinity purification revealed that ∼75% and 90% of SRP in the lysate was depleted by RNCs with the TMD in the tunnel or exposed, respectively (Figure 1C), despite these representing only ∼5% of total ribosomes. Appreciable SRP depletion was not observed with either of the 3R RNCs, and post-translational TMD-binding proteins such as TRC40 were not depleted by any of the constructs. These results illustrate that RNCs containing a TMD inside the ribosomal tunnel recruit SRP nearly as efficiently as RNCs containing an exposed TMD.

10The observation that the 3R mutant largely abolishes SRP recruitment suggests that shorter hydrophobic regions, such as those remaining on either side of the mutation, are insufficient for robust recruitment. Serial truncations verified that maximal recruitment required the presence of 14 hydrophobic residues in the tunnel, with lower (but detectable) recruitment once 10 or 11 residues has been synthesized (Figure 1D). Analysis of shortened TMDs were consistent with this result, showing sharply diminished recruitment with deletion of 4–8 residues, and no detectable recruitment after deletion of 10 residues (Figure 1E).

11Systematically varying the position of the full TMD inside the tunnel showed efficient SRP recruitment at all positions (Figure 1F), while the shortened ∆8 mutant showed low recruitment at all positions (data not shown). These data collectively suggest that SRP binds translating ribosomes in two distinct states: a ‘scanning’ mode when the TMD is inside the exit tunnel, and an ‘engaged’ mode where SRP interacts directly with the hydrophobic nascent chain. These scanning and engaged SRP-RNCs were assembled in total cytosol from endogenous mammalian components, thereby representing native complexes. Importantly, both complexes could be purified in sufficient amounts for structural analysis by electron cryomicroscopy (cryo-EM).

12Ribosomal particles from the scanning and engaged SRP complexes were visualized by cryo-EM and subjected to iterative in silico classification (Figure 2—figure supplement 1) to select RNCs containing both P-site tRNA (as a surrogate for the nascent chain) and SRP. Refinement of the resulting RNC populations resulted in final reconstructions at an overall resolution of 3.9 Å and 3.8 Å for the 80S ribosome bound to SRP in its scanning and engaged modes, respectively (Figure 2A,B, Table 1, Figure 2—figure supplement 2). The inherent flexibility of regions of SRP was reflected by the large variation in local resolution within the particle, which spans from ∼3.5 Å to greater than 7.5 Å resolution (Figure 2—figure supplement 3).

13In both structures, the ribosome is unratcheted and contains a canonical P-site tRNA and density for the nascent chain in the exit tunnel (Figure 2—figure supplement 4). Density for the mRNA codon in the P site is also present, but is poorly defined, likely reflecting increased flexibility due to truncation of the mRNA. Additional density representing the E-site tRNA is also observed, though its anticodon is disordered suggesting it is not in a single orientation. We do not observe any conformational changes in the ribosome exit tunnel in the scanning structure relative to either the engaged structure or earlier mammalian ribosome structures. Thus, SRP recruitment by a TMD inside the exit tunnel cannot be explained purely based on structural changes observed at this resolution, a point we address further in the ‘Discussion’.

14The overall architecture of SRP in the engaged binding mode (Figure 2B) is very similar to earlier reconstructions of this state visualized using heterologous plant–mammalian complexes. As expected, the Alu domain is localized to the GTPase center between the 40S and 60S subunits (Figure 2C) and is connected by a flexible RNA linker to the S domain positioned at the exit tunnel (Figure 2D). At low-resolution, the observed density accounts for the majority of the SRP molecule, including most of the RNA, the Alu domain proteins (SRP9 and SRP14), and most of the S domain proteins (SRP54, SRP19, and RNA binding regions of SRP68). However, only the Alu domain and SRP54 were sufficiently well ordered to allow direct modelling into the density maps, and thus form the focus of our interpretations in the sections below. The others areas, whose functional relevance remain poorly understood, were rigid-body fit with minor adjustments from available crystal structures, and were not interpreted further here.

15The SRP density in the matched scanning mode structure was found to be remarkably similar in overall conformation, position, and relative occupancy of all domains (Figure 2A). In particular, the Alu domain, whose binding has been implicated in slowing translational elongation after SRP engages with a signal sequence, is positioned in the GTPase center. Similarly, the entire S domain is grossly indistinguishable, indicating that the TMD-interacting M domain of SRP54 is positioned at the exit tunnel even prior to engagement. This suggests that in mammals, SRP has a single stable binding site on the ribosome, and carries out its biological functions during both scanning and engagement from this position.

16The considerably higher resolution of these structures relative to earlier reconstructions permitted interpretation of several SRP-RNC interactions in atomic terms. SRP is anchored on the ribosome primarily by its SRP54 subunit near the exit tunnel (Figure 3A) and the Alu domain at the ribosomal GTPase center. Two loops in the SRP54 N-domain previously implicated in ribosome binding can now be sufficiently resolved for unambiguous placement of many protein side chains and analysis of the stabilizing chemical interactions (Figure 3B). For example, Thr21 and Asp19 of SRP54 are within hydrogen bonding distance of Lys46 in uL29 and the phosphate oxygen of A80 in 5.8S rRNA (Figure 3C). Similarly, the backbone of residue 67 is within hydrogen bonding distance of the carbonyl oxygen of residue 116 of uL23 (Figure 3D). These specific contacts, along with other electrostatic interactions involving the conserved basic residues in this region, provide the primary stabilization of SRP54 and serve to orient its signal-binding M domain at the ribosomal exit tunnel.

17Within the ribosomal GTPase center, the RNA portion of the well resolved Alu domain approaches the 60S subunit, interacting with the ribosomal stalk protein uL10, while the proteins SRP14 and SRP9 approach the 40S subunit, interacting with the backbone of the 18S rRNA (Figure 4A). The structure reveals that Asn77 and Lys95 of SRP14 are positioned within hydrogen bonding distance of the phosphate oxygens of U488 and A464 of the 18S rRNA (Figure 4B–D). Of note, alignment of our SRP structure bound to the mammalian ribosome with an earlier structure of mammalian SRP bound to the plant ribosome suggests changes in relative position of the S and Alu domains. Such changes, together with sequence divergence of uL10 and uL11, may explain why mammalian SRP arrests translation in a plant system while only subtly slowing translation in mammals.

18The main functional transition between a scanning and engaged SRP is substrate binding by the M domain. Our structures afforded the first opportunity to visualize the conformation of the scanning M domain and deduce changes that accompany TMD engagement. We began by modelling the engaged M domain (Figure 5A), guided by earlier crystal structures involving this region. The hydrophobic groove of this domain is occupied by density consistent with an alpha helix of ∼12 residues, accounting for roughly half of the nascent chain's TMD. This assignment is consistent with earlier structures and crosslinking analyses.

19After fitting the density corresponding to earlier crystal structures and assigning the substrate, two additional regions of helical density were observed abutting the TMD (Figure 5A). Several arguments suggest that these represent the C-terminus of SRP54. First, the C-terminal ∼70 residues, truncated from all earlier SRP54 crystal structures, are the only unaccounted amino acids in this region. The only other nearby polypeptide would be the hydrophilic nascent chain, which is unlikely to be helical or rigid. Second, this 70 residue stretch of mammalian SRP54 contains two conserved predicted helices of appropriate length (Figure. 5—figure supplement 1A). Third, these helices are amphipathic, making them ideal for surrounding the hydrophobic TMD (Figure 5—figure supplement 1B). We therefore posit that these two additional helices, provisionally termed αC1 and αC2 of SRP54, serve as an intramolecular ‘lid’ that surrounds the hydrophobic substrate to shield it from aqueous solvent (Figure 5B). Although we cannot unambiguously assign these helical regions to a particular sequence, we have provisionally named them based on the relationship between their apparent length in the structure and predicted size.

20Relative to the engaged state, the scanning structure showed markedly underrepresented density in the region of αC2, but clear helical density within the SRP hydrophobic groove (Figure 6A, compare top and bottom panels). This suggests that in the absence of a TMD substrate, αC2 shifts to occupy the hydrophobic groove. This helix may be somewhat dynamic since a small amount of density is still observed in the alternate position of αC2 (Figure 6A, bottom panel). Although this is our favored hypothesis, we cannot formally exclude the possibility that the density in the hydrophobic groove represents the nascent chain. However, this seems unlikely because a hydrophilic sequence would be disfavored in the hydrophobic groove, this region of polypeptide is unlikely to be helical, earlier crosslinking data suggest SRP does not interact with hydrophilic segments of nascent chain, and disappearance of the αC2 density would remain unexplained. Thus, we propose that the amphipathic αC2 normally resides in the hydrophobic groove, but is displaced by TMD engagement to its ‘lid’ position between the M domain and exit tunnel (Figure 6B).

21Our structures indicate that Alu domain occupancy at the GTPase center is mutually exclusive with translational GTPase interactions that occur with each cycle of nascent chain elongation. Yet, 35 cycles of translational elongation separate our two structures, and SRP does not arrest translation even after successful engagement. To investigate this apparent incompatibility, we sought to examine the relationship between translational GTPase binding and the SRP-ribosome interaction in the scanning vs engaged modes. Such an experiment is complicated because neither dominant negative eEF1 nor eEF2 would bind our staged SRP-RNC complexes: eEF1 ternary complex binding requires an appropriate codon in the ribosomal A site, while eEF2, when in its GTP-bound state, preferentially acts on pre-translocation RNCs in the hybrid state. Furthermore, adding mutant versions of these factors to our reactions would inhibit translation. To bypass these technical obstacles, we utilized the GTPase Hbs1 as a surrogate for eEF1 and eEF2. The G-domain of all three factors share a high degree of sequence and structural homology, and the binding of Hbs1 to the ribosomal GTPase center is well characterized. Importantly, a GTPase deficient dominant negative mutant (Hbs1-DN) added to our translation reactions will stably bind to the translation factor binding site only when ribosomes reach the end of the truncated message, allowing us to query the consequence of GTPase center occupancy for the SRP-RNC interaction without inhibiting translation.

22Translation extracts were supplemented with increasing amounts of Hbs1-DN, and used for the synthesis and purification of scanning and engaged RNC complexes via the nascent chain. Hbs1-DN binding, observed in both cases, reduced SRP recovery by up to ∼70% with the scanning RNCs, but had no effect on engaged SRP-RNC complexes (Figure 7A). Because a substantial excess of Hbs1-DN relative to ribosomes was used, its failure to displace engaged SRP suggested the existence of a ternary complex between the ribosome, Hbs1-DN, and SRP. The relatively modest displacement of scanning SRP further hinted that this too may permit ternary complex formation.

23To test this idea, scanning and engaged HA-tagged RNCs were assembled in the presence of FLAG-tagged Hbs1-DN, affinity purified via the FLAG tag, and tested for SRP recovery (Figure 7B). Consistent with the competition experiment, equal amounts of ribosomes were recovered from both reactions in an Hbs1-dependent manner, but substantially less SRP was found with scanning vs engaged RNCs. Thus, when SRP is in its scanning mode, binding of Hbs1-DN can partially (but not completely) displace SRP. However, once the TMD is exposed, both Hbs1-DN and engaged SRP are able to stably associate with the ribosome.

24Such an observation can be explained by the available structural information. Due to the flexible RNA linker that connects the two halves of SRP, Alu domain dissociation need not lead to immediate dissociation of SRP. Rather, the SRP S domain interactions could retain SRP on the ribosome, with dissociation governed by the off rate of this complex. For E. coli SRP, which lacks an Alu domain, this off rate is ∼0.1–0.3 s−1, probably explaining why scanning-mode SRP is lost during the 2–3 hr needed for RNC affinity purification. By contrast, the additional interaction between the TMD and the M domain reduces this off rate effectively to zero relative to the timeframe of this experiment, explaining why Hbs1-DN cannot displace engaged SRP. Other translational GTPases (eEF1, eEF2, and the release factor complexes) would likely share the same qualitative relationship with the different states of ribosome-bound SRP, explaining why SRP, even after RNC engagement, is compatible with active elongation.

25This logic suggests that a population of ribosomes bound to both a translation factor and SRP might be present in our EM samples. The use of a truncated mRNA for sample preparation precludes eEF1 binding, but eEF2-ribosome-SRP complexes were feasible. Such a population was not identified in our scanning dataset, likely due to the lower stability of this ternary complex when SRP is bound in its scanning mode (Figure 7B). By contrast, the engaged dataset contained a small subset of particles in which density was observed for both eEF2 and the S domain of SRP bound in its canonical location at the exit tunnel (Figure 7C). At a lower threshold, density for the flexible RNA linker and the Alu domain could also be observed, with the Alu domain at the so-called right foot of the 40S subunit (Figure 7, Figure 7—figure supplement 1). The relatively poor density for this half of SRP suggests it is probably dynamic when the Alu domain cannot bind its canonical site at the GTPase center. Whether the weakly observed alternate Alu domain position on the 40S represents a physiologically relevant binding site remains to be determined. Nevertheless, this structure supports our conclusion that the Alu domain can be displaced from the GTPase center by translation factors without dissociating SRP from the ribosome.

26The compatibility of SRP and translation factor binding to the same ribosome suggested that scanning SRP need not dissociate between rounds of polypeptide elongation. To investigate this idea, we designed an experiment to determine if scanning SRP can be observed on actively translating ribosomes. A stop codon was positioned 35 residues from the beginning of the TMD, ensuring that translation would terminate before emergence of the TMD from the exit tunnel (Figure 7D, diagram). During translation, a sub-population of ribosomes containing nascent peptides long enough to expose the N-terminal epitope tag would also contain the TMD inside the tunnel. As SRP cannot interact with its clients post-translationally, its recovery with the nascent chain would necessarily be via the ribosomal scanning mode.

27Affinity purification of RNCs after 3 min of translation recovered substantially more SRP with RNCs containing a TMD than the 3R mutant (Figure 7D, right panel). Parallel reactions in which translation initiation was inhibited at 3 min and the ribosomes were allowed to complete translation produced full-length protein that recovered neither ribosomes nor SRP. These observations suggest that SRP is recruited in a TMD- and nascent chain-dependent manner to actively translating ribosomes despite never exposing the TMD outside the exit tunnel. Thus, SRP might be able to continuously scan translating ribosomes as a TMD elongates through the exit tunnel, thereby permitting engagement with essentially no opportunity for TMD exposure to the cytosol.

28In this study, we have characterized an anticipatory scanning mode of SRP-ribosome interaction, validated its existence during ongoing translation, and provided the first structure of this complex. Together with our engaged SRP complex, these structures represent markedly improved views of many functional regions of mammalian SRP and provide the first entirely homologous SRP-RNC structures assembled with native endogenous factors. Additional experiments analysing the relationship between SRP and a translational GTPase permit us to consolidate our findings into a dynamic model of nascent chain scanning and TMD capture by SRP during the earliest stages of membrane protein biosynthesis (Figure 8).

29Once as few as 12–14 residues of a TMD enter the ribosomal exit tunnel, an RNC becomes competent to recruit SRP in its scanning mode. SRP binding appears to favor an unratcheted ribosome and requires an unoccupied GTPase center. An A site tRNA, while not observed in our structure due to the use of a truncated mRNA, is sterically compatible with SRP binding. Thus, Alu domain recruitment could potentially occur immediately after dissociation of eEF1 or eEF2, from the ribosome. The S domain could presumably bind transiently anytime, and may increase local SRP concentration to facilitate the Alu domain interaction. The proportion of each ∼200 ms translation cycle during which RNCs are compatible for Alu domain binding remains to be determined. Regardless, a twenty residue TMD would have approximately 21 translation cycles to recruit SRP while it is still inside the tunnel.

30Once recruited, SRP would be ribosome-associated via both the S and Alu domains. Arrival of the next translation factor would displace the Alu domain, while the S domain can remain fixed due to the flexible RNA linker between these two domains (Figure 8A). Whether SRP completely dissociates from the ribosome would depend on the off-rate of the S domain relative to the next Alu domain binding opportunity within the translation cycle. Estimates from the E. coli system suggest that the S domain dissociation would be far slower than an elongation cycle, and thus the Alu domain could re-bind and detach repeatedly while SRP nevertheless maintains ribosome association during ongoing translation. This would explain how we are able to isolate scanning mode SRP-RNC complexes from an active translation reaction (Figure 7D) and recover at least some GTPase-SRP-RNC ternary complex in the scanning state (Figure 7B).

31Early recruitment followed by maintenance of the M domain at the exit tunnel would allow for continuous sampling of the nascent chain until the TMD emerges. The M domain is optimally pre-positioned for this function, with the hydrophobic groove located ∼20 Å from the exit site. Although the precise path of the nascent chain is not known, the limited space in this region means it necessarily passes the M domain (Figure 8B, left panel). We envisage a dynamic αC2 helix that transiently exposes the hydrophobic groove to provide opportunities for nascent chain sampling. If a segment of the nascent polypeptide is sufficiently hydrophobic to replace the αC2 helix, it would successfully engage SRP54 (Figure 8B, right panel). In this model, the biophysical properties of αC2 effectively set the hydrophobicity threshold for client engagement. After nascent chain binding to the M domain, αC2 would occupy a position on the opposite side of the hydrophobic groove, oriented with its hydrophobic face toward the substrate, forming part of a protective lid. The conformation of this engaged state may be flexible to accommodate different substrates, as suggested by other variant configurations (Figure 8—figure supplement 1). Regulated access to a hydrophobic binding groove and substrate shielding by an amphipathic helix appears to be a general principle that is shared by the Get3 targeting factor.

32In the engaged state, the SRP S domain is able to remain stably bound to the ribosome via its dual interactions with the exit tunnel and nascent chain, ensuring shielding of the hydrophobic substrate. Elongation factors could still access the GTPase center by displacing the Alu domain, which must swing away from the ribosome. Its local concentration would necessarily remain high due to S domain tethering, presumably improving its ability to compete with eEF1 and eEF2. This is the likely cause of the reported slowing of translation by SRP that provides a subtly increased time for successful delivery to the translocon.

33It is noteworthy that the minimum hydrophobic element for robust early recruitment of SRP is longer than the 7 to 9 hydrophobic residues within most cleavable N-terminal signal peptides. Early recruitment may not be critical in this case because signal peptides are more soluble and less aggregation-prone than TMDs, and sufficiently short to be fully shielded by the hydrophobic groove of the M domain. By contrast, TMDs responsible for targeting are markedly more hydrophobic and between 17 to 24 residues long. They are effectively insoluble in aqueous solvent and too long to be completely shielded by the M domain, making their efficient and early targeting more important. Moreover, early recruitment would provide SRP the opportunity for first refusal of an emerging TMD, which could allow SRP to out-compete other more abundant binding factors such as TRC40, SGTA, Hsp70, and calmodulin.

34Another important functional implication of early recruitment is increased time for targeting the RNC to the translocon. We cannot, at this level of resolution, see any differences in the configuration of the SRP54 NG domain between the scanning and engaged states. This suggests that the scanning state should be competent for interaction with the SRP receptor, potentially permitting targeting while the TMD is still inside the tunnel. Based on the ∼35 residue length of the exit tunnel, recruitment before the TMD is synthesized provides an additional ∼7 s for targeting relative to a model where targeting only occurs after TMD emergence.

35This extra time would be sufficient for delivery to the translocon before or as the TMD emerges, essentially eliminating its cytosolic exposure and allowing its synthesis to occur directly at Sec61. Indeed, recent ribosome profiling experiments in yeast suggest that many proteins arrive at the Sec61 translocon before the TMD is exposed. Such early delivery may be particularly important for proteins whose downstream domains may interfere with correct translocation such as downstream TMDs of multispanning membrane proteins or rapidly folding domains. Our findings suggest that early SRP recruitment is another mechanism for lengthening the targeting window, and may operate in conjunction with translational slowing and strategic positioning of slowly decoded rare codons.

36A major unresolved question is the mechanistic basis for how a TMD can influence SRP binding from inside the exit tunnel. One possibility involves conformational changes on the ribosome surface triggered by TMD interactions with the proteins or RNA lining the tunnel. However, we have been unable to detect any differences in the tunnel or surface regions for SRP binding by comparing our scanning RNCs with those containing hydrophilic sequences. Instead, we currently favor a kinetic model in which a hydrophobic sequence in the tunnel biases ribosomes into a conformation favorable for Alu domain binding. If the presence of a hydrophobic sequence in the ribosomal tunnel would preferentially extend those stages of the translation cycle compatible with SRP binding, this could provide a mechanism for tunnel-initiated SRP recruitment.

37Evidence for preferential SRP binding to some but not other states of the ribosome comes from experiments using different translation elongation inhibitors. Cycloheximide, an elongation inhibitor that causes ribosomes to pause in a presumably unratcheted state, can rescue translocation defects caused by limiting SRP availability. By contrast, another elongation inhibitor anisomycin, which pauses ribosomes at the ratcheted state, does not rescue under similar conditions. This implies that situations preferentially favoring the cycloheximide state may permit SRP recruitment. Thus, it is plausible that a TMD inside the tunnel, or elongation of hydrophobic amino acids, would subtly influence the translation cycle to favor this state. Indeed, sequences inside the ribosomal tunnel are known to influence the translation cycle, while the biophysical characteristic of the aminoacyl-tRNA in the A site influences the time spent in different ribosome states. A model invoking the Alu domain would also explain why sequence-independent scanning in the E. coli system appears to operate differently than the TMD-dependence observed in eukaryotes.

38Finally, it is worth noting that tools are now readily available to analyze early recruitment of SRP in vivo. The combination of ribosome profiling together with selective retrieval of SRP complexes should permit the mapping of all SRP-containing RNCs. Such analyses can provide a global view of not only SRP substrates in vivo, but also the precise timing of its recruitment for different types of cargos. We anticipate that, based on our analyses in the mammalian system, early recruitment should be seen for proteins with the most hydrophobic targeting sequences. Such proteins are also the most aggregation-prone when targeting fails, perhaps providing a strong selection pressure for evolution of an early targeting mechanism.

39An SP64 vector-based construct encoding the transferrin receptor TMD (AIAVIVFFLIGFMIGYLGYA) inserted into Sec61β was modified to contain an N-terminal affinity tag (3xFLAG or 3xHA). Phusion mutagenesis was used to generate the following TMD mutants: 3R (AIAVIRRRLIGFMIGYLGYA), ∆4 (AVIVFFLIGFGYLGYA), ∆6 (AIVFFLIGGYLGYA), ∆8 (AVFFLIGYLGYA), and ∆10 (AFFLGYLGYA). A stop codon was introduced by Phusion mutagenesis at position 110 for the experiment in Figure 7D. The mammalian expression construct for Hbs1-DN has been described, and was purified as before. Antibodies against uL6 (anti-L9) and uS9 (anti-S16) were from Santa Cruz Biotechnology (Dallas, TX). Anti-SRP54 was from BD Biosciences (San Jose, CA). The antibody against TRC40 has been described. Anti-Flag and HA resin and 3X Flag and HA peptides were obtained from Sigma (St. Louis, MO).

40Preparation and purification of stalled RNCs was performed as previously described. Briefly, the template for in vitro transcription was prepared by PCR from the constructs above using a 5′ primer just preceding the SP6 promoter, and a 3′ primer at the desired site of truncation (or downstream of the stop codon for full length products). All 3′ primers for truncations encoded a final valine codon, whose peptidyl-tRNA is least labile to hydrolysis. PCR products were purified and used for in vitro transcription and translation in rabbit reticuloclyte lysate as described. Translations were for were for 20–25 min at 32°C. For Figure 7A, Hbs1-DN was included at 3.3 nM, 17 nM, 33 nM and 133 nM. For Figure 7B, Hbs1-DN was included at 133 nM. For Figure 7D, actively translating, unstalled complexes were produced by allowing 3 min of translation at 32°C followed by either a rapid fivefold dilution in chilled buffer 1 (50 mM HEPES pH 7.5, 200 mM KAc, 15 mM MgAc2, and 1 mM DTT), or addition of 70 μM aurin tricarboxylic acid (ATA) for the remainder of a 30 min translation before dilution.

41All samples were then affinity purified via either the nascent chain or the tag on Hbs1-DN using anti-Flag or HA resin. The Flag tag was used on the nascent chain in all experiments except those that included Hbs1-DN, which instead used HA-tagged nascent chains. For biochemical analysis, the affinity resin was either added directly to undiluted translation reactions, or after dilution in buffer 1. The ratio of affinity resin to translation reaction was typically 1:50. Binding was performed in batch at 4°C for ∼1–2 hr with gentle mixing, transferred to a micro-spin column, washed with ∼25 vol (relative to resin) of buffer 1, and eluted for 30 min at 22°C in the same buffer supplemented with 0.2 mg/ml of the appropriate peptide. Samples for structural analysis were washed as above in buffer 1 containing an additional 200 mM or 400 mM KAc for the scanning and engaged complexes, respectively. Samples were eluted in buffer 1 supplemented with 3 mM GDPCP, concentrated by centrifugation (50,000 rpm for 75 min in a TLA55 rotor), and resuspended in a volume of buffer 1 to achieve ∼125 nM ribosomes in the presence of 100 μM GDPCP.

42Purified samples were applied to glow-discharged holey carbon grids (Quantifoil R2/2), which had been coated with a ∼70 Å thick layer of amorphous carbon. Using an FEI Vitrobot, 3 μl of sample was applied to the grid, followed by a 30 s incubation at 4°C, 3 s of blotting, and flash-cooling in liquid ethane. Data were collected on an FEI Titan Krios at 300 KV using FEI's automated single particle acquisition software and defocus values of 2–3.5 μm. Images were recorded using a back-thinned FEI Falcon II detector at a calibrated magnification of 104,478 (pixel size of 1.34 Å). Individual frames from the detector were recorded as previously described.

43Contrast transfer function parameters were estimated using CTFFIND3, and micrographs that had evidence of astigmatism or drift were discarded. All automated particle picking, 2D and 3D classifications, and refinements were performed using RELION as described below. Unsupervised 2D class averaging was used to discard any non-ribosome particles, resulting in a combined 326,981 and 639,184 particles for the engaged and scanning samples, respectively. Iterative rounds of 3D classification were then utilized to identify the population of ribosomes bound to SRP (Figure 2—figure supplement 1).

44For the scanning sample, 27% of selected 80S particles (171,143) were computationally classified as being in an unratcheted conformation and bound to a P-site tRNA. As weak density for SRP in this population could be observed, a further focused classification was performed utilizing a mask around the observed extra-ribosomal density to identify 16% of this unratcheted population (27,627 particles) that contained SRP. An additional focussed classification using a mask around the P-Site tRNA identified a final 27,415 particles that are unratcheted, bound to SRP, and contain improved density for the nascent chain. The somewhat weaker density observed for the nascent chain in the scanning complex (Figure 2—figure supplement 4) might be due to either increased flexibility of the nascent chain when it is not anchored at the N-terminus by an interaction with SRP, or decreased occupancy resulting from hydrolysis during sample freezing that was not resolved by computational sorting. Similarly, in the engaged sample, 58% of ribosomes (189,099 particles) were unratcheted and contained tRNA, while 28% of these particles (52,061) were bound by SRP. In order to isolate the population of ribosomes bound to both SRP and eEF2 in the engaged sample, a mask for SRP was used to further sub-classify the population of ribosomes bound to eEF2 (20%: 64,416) resulting in a final population of 8418 particles.

45Final 3D refinements of the resulting populations were performed without external masking, utilizing statistical movie processing, and particle polishing. This resulted in final reconstructions at overall resolution of 3.9 Å, 3.75 Å, and 5.0 Å for the scanning, engaged, and ternary complex structures, respectively, using the gold-standard FSC = 0.143 criteria.

46All models were initially built and refined in the density for the higher resolution engaged complex, and then rigid body fit into the density for the scanning sample. The resulting scanning model required only minor adjustments to the M domain to fit the respective density (e.g., Figure 6). The 60S subunit, 40S body, and 40S head were individually placed using models of the porcine ribosome, and the P-site tRNA was homology modelled using the bacterial structure. SRP was built using a combination of crystal structures of its individual domains, which were modified to fit the observed density as described below.

47Within the S domain, the N-domain of SRP54 was built primarily using the solution structure of the human domain (PDB ID: 1WGW) with minor modification to the loops that interact with the ribosome. The G-domain was modelled using the structure from (PDB ID: 3KL4). The M-domain itself was modelled using the crystal structure of the human M-domain (PDB ID: 1QB2), as well as models of the occupied M-domain from homologous species, with modifications to account for the interface with the ribosome and other SRP components. Due to lower resolution in this region, the S-domain RNA, SRP19, and a portion of SRP68 were simply rigid-body fit using the model from (PDB ID: 4P3E).

48The Alu domain was built based on models from, including the proteins SRP9 and SRP14 (PDB ID: 1E8O), and the Alu RNA (assembled from PDB IDs: 1E8O and 1E8S). All models were built in COOT. Refinement of the 40S subunit plus Alu domain, and 60S subunit plus SRP54, were carried out individually using using REFMAC v5.8 as previously described. Secondary structure restraints were generated in ProSMART, and nucleic acid base-pairing and stacking restraints were generated as before and were maintained throughout refinement to prevent over-fitting. Local resolution was calculated using ResMap and all figures were generated using Pymol and Chimera.