Single-particle EM reveals extensive conformational variability of the Ltn1 E3 ligase

Abstract

Ltn1 is a 180-kDa E3 ubiquitin ligase that associates with ribosomes and marks certain aberrant, translationally arrested nascent polypeptide chains for proteasomal degradation. In addition to its evolutionarily conserved large size, Ltn1 is characterized by the presence of a conserved N terminus, HEAT/ARM repeats predicted to comprise the majority of the protein, and a C-terminal catalytic RING domain, although the protein’s exact structure is unknown. We used numerous single-particle EM strategies to characterize Ltn1’s structure based on negative stain and vitreous ice data. Two-dimensional classifications and subsequent 3D reconstructions of electron density maps show that Ltn1 has an elongated form and presents a continuum of conformational states about two flexible hinge regions, whereas its overall architecture is reminiscent of multisubunit cullin–RING ubiquitin ligase complexes. We propose a model of Ltn1 function based on its conformational variability and flexibility that describes how these features may play a role in cotranslational protein quality control.

During protein production, a termination (or “stop”) codon on a template mRNA specifies the end of a polypeptide chain and recruits peptide release and ribosome recycling factors to terminate translation (1). Errors in gene expression generating mRNA lacking stop codons (“nonstop mRNA”) as well as defects in translation resulting in stop codon readthrough can result in the sequestration of ribosomes and the production of aberrant proteins (“nonstop proteins”) (2, 3). The quality of such newly synthesized proteins is carefully monitored through a process known as translational surveillance (2–5). During the past few years, pathways specifically dedicated to eliminate nonstop proteins have been discovered. This is controlled by a system known as tmRNA/ssrA in bacteria (4–6), and a critical component of this process that is conserved from yeast to humans is the Ltn1 E3 ligase (7). Ltn1 functions by associating with ribosomes and mediating the ubiquitylation and subsequent degradation of translationally arrested nonstop proteins (7). Experiments in yeast showed that the absence of Ltn1 leads to the toxic accumulation of such proteins, which may explain the embryonic lethality and neurodegenerative phenotypes of Ltn1/Listerin-mutant mice (7, 8). The elucidation of Ltn1’s role in yeast has identified how nonstop protein levels are controlled in eukaryotic cells, a process mechanistically distinct from the bacterial system. A better understanding of Ltn1 is thus expected to help clarify this recently discovered cellular process for protein quality control, while potentially shedding light on the role of this E3 in neurodegeneration.

E3s mediate ubiquitin transfer to target substrates, and most can be classified into one of two major families, defined by the presence of a RING (Really Interesting New Gene) or a HECT domain, with RING domain-dependent E3s accounting for the majority of known ubiquitin ligases (9). The members of this protein family are bisubstrate enzymes that bind to an E2-ubiquitin conjugate via the E3 RING domain, and to target substrates through auxiliary domains termed the substrate receptor. Thus, RING E3s act like scaffold proteins in bringing the two substrates into proximity and additionally stimulating ubiquitin release from the E2. Although, in some E3s, the RING domain and substrate binding site are present in separate subunits of a multiprotein complex, as, for example, within the cullin-RING ligase (CRL) superfamily (10), in other E3s, both functions reside in a single polypeptide (11). Ltn1 is believed to bind to its protein quality control substrates indirectly by using translationally arrested ribosomes as the equivalent of substrate receptors in cullin-dependent E3s (7). However, the exact biochemical mechanisms involved in the recognition of ribosomes, as well as Ltn1’s structure, remain unknown. Given Ltn1’s unusually large, evolutionarily conserved size across orthologues and its functional significance within the context of translational surveillance and neurodegeneration, we set out to determine its structure by using single-particle EM.

Here we present a 2D and 3D characterization of a continuum of conformational states of Ltn1, which provides direct structural evidence for large-scale conformational variability within an E3. The methods presented represent a large-scale dynamic characterization of a single protein with the use of single-particle EM and demonstrate a powerful approach for the assessment of conformational variability. These strategies may be applicable to the structural analyses of other small (<500 kDa) and highly dynamic macromolecules and macromolecular complexes.

Results

Majority of Ltn1 Is Predicted to Fold as α-Helical Repeats of HEAT/ARM Type.

Ltn1 consists of an evolutionarily conserved N terminus (amino acids ∼18–322), a middle region (amino acids ∼323–1265), and an evolutionarily conserved C terminus (amino acids ∼1266–1562) that includes the catalytic RING domain (amino acids 1509–1562; Fig. 1 A and B). The results of protein motif analyses with the use of a publicly available bioinformatics platform predict that most of the protein, in particular the N terminus and middle region, is composed of a long stretch of HEAT (Huntingtin, Elongation factor 3, PR65/A subunit of protein phosphatase 2A, and TOR) or ARM (Armadillo) repeats [amino acids ∼44–1390; amino acids ∼40–1580 in the human orthologue (NP_056380.1); Fig. 1B] (12). Crystal structures of several HEAT repeat proteins have been solved, revealing that each ∼37- to 50-aa repeat consists of two antiparallel α-helices (three in the case of ARM repeats) joined by a short loop, and these may sequentially pack to form superhelices (13). To strengthen Ltn1’s domain prediction, we interrogated the protein’s secondary structure (14). The results of this analysis indeed suggest that Ltn1 is largely α-helical, composed of helices whose lengths are consistent with those found in more extensively characterized HEAT/ARM repeat proteins (13) (Fig. 1C and Fig. S1). Thus, although the middle region of the protein is not strongly conserved at the primary amino acid sequence level, it is likely that Ltn1 orthologues exhibit a similar overall structure.

Fig. 1.

Evolutionary conservation, domain, and secondary structure analysis of Ltn1. (A) Evolutionary conservation of Ltn1. Conservation for each orthologue is marked by solid lines and calculated with respect to the primary structure of S. cerevisiae Ltn1 using default parameters on the BLAST website (http://blast.ncbi.nlm.nih.gov/Blast.cgi). Residue numbers are at top. The lines indicate the length of conserved regions (by residue number). (B) Domain analysis of Ltn1 showing that predicted HEAT/ARM repeats span the majority of the protein, and that the RING domain responsible for E3 activity is located at Ltn1’s C terminus. (C) Secondary structure analysis showing predicted α-helices (light) and β-strands (dark).

Ltn1 Is an Elongated Protein That Exhibits Large-Scale Movement About Two Hinge Regions.

Purified and functionally active Ltn1 proteins (Fig. S2) are distinguishable as elongated and highly flexible particles (Fig. S3), indicating the presence of several different conformers in the data. To identify and reconstruct snapshots of conformational states of Ltn1, we collected a data set consisting of tilt-pair micrographs, taking advantage of the random-conical tilt (RCT) strategy for its ability to reconstruct electron density maps from independent 2D classes (15, 16). Such classes can be obtained by using standard alignment and classification methods in single-particle image processing (reviewed in ref. 17). To cope with the apparent high degree of heterogeneity within Ltn1, an iterative strategy was used to identify homogeneous classes and subsequently reconstruct electron density maps (Figs. S4 and S5). By using this workflow, we identified 22 conformational snapshots of Ltn1, which differed in the positions of the terminal regions (Fig. 2). Here, two volumes were defined as “distinct” when, after aligning their central backbones, a rotation of greater than 10° was observed in either of the two termini at an identical threshold level.

Fig. 2.

Ltn1 shows large-scale conformational variability about two hinge regions. (A) Aligned single particles, (B) class averages, (C) class correspondence through a heat map describing the percentage of reconstructed particles belonging to each conformation, and (D) reconstructed volumes of Ltn1, arranged according to movement about two hinge regions. (A, B, and D) Inward flexing of the left tail (terminus-1) varies vertically; rotations of the right tail (terminus-2) vary horizontally. All volumes in D are aligned to a 45° axis along their central region (displayed as a solid black line) and numbered with roman numerals (Table S1). Movement and corresponding angles of the termini from 0° are indicated by red arrows. Merged volumes for each column and row are boxed and labeled below and at right, respectively. Double asterisks indicate that the calculated rotation angles for terminus-1 slightly deviate from the ordering scheme and may represent intermediate states. An overlay of all 3D electron density maps and a schematic representation of Ltn1 mobility, displaying the full range of movements within the two termini, are displayed in the boxed region. (Scale bars: 200 Å.)

Analysis of the data shows an inherent flexibility of Ltn1, which produced a continuous range of structural conformations with two clear hinge regions. Fig. 2 displays the 22 RCT reconstructions of Ltn1 ordered according to the degree of flexibility about the hinges and arranged across two dimensions. Displayed are the aligned untilted particles (Fig. 2A), summed class averages (Fig. 2B), percentages of total particles within each class (Fig. 2C), and reconstructed volumes (Fig. 2D; Table S1 summarizes all final parameters). Among the identified conformers, lengths range from ∼17 to 21.5 nm, with widths of ∼10 to 13 nm (Table S1). For consistency, each displayed volume was low-pass–filtered to 40 Å and aligned to Ltn1’s central domain positioned at a 45° angle (Fig. 2D). This orientation allows us to focus on the movements in the terminal regions. The hinges identifiable in these 3D maps are at ∼2/5 and ∼3/5 of the distance from the left terminus, and their positions occur at approximately the ends of the solid black line (Fig. 2D, referred to as hinge-1/terminus-1 and hinge-2/terminus-2, respectively). Movement of terminus-1 of each reconstruction is evident across individual columns, whereas movement of terminus-2 is evident across individual rows, an aspect that can be more fully appreciated when each set is merged (Fig. 2D, Lower and Right). Additional conformers are likely present based on the observation that separate class averages may fit into the ordering scheme (Fig. S6). However, their reconstructions were poorly defined, most likely as a result of insufficient particle numbers, and we therefore cannot rule out the possibility that they may correspond to slightly altered views of an otherwise identical conformer.

Occasionally, a slight blurring is evident in the class averages in proximity to the termini, indicating some degree of heterogeneity within these regions and further demonstrating the likely continuum of flexibility of the protein. The two class averages in column 4 (Fig. 2D) both contain blurred density within terminus-2, and their terminal region undergoes a much larger transition of ∼140° between columns 3 and 5 compared with the more subtle variation of ∼30–35° between either the rightmost or leftmost three columns (Fig. 2D). We expect that the corresponding volumes represent a transition within the protein, wherein terminus-2 flips from an inward-folded (columns 1–3; Fig. 2D) to an outward-folded (columns 5–7; Fig. 2D) conformation. Therefore, the full spectrum of reconstructions can be roughly classified into one of these two categories, whereby terminus-2 accounts for the majority of the mobility seen within the protein.

Among most of the identifiable snapshots, multiple states of one terminus co-occur with a single conformation of the opposite terminus. This argues that the two movements are not directly coordinated, but rather occur independently with bidirectional freedom of motility, at least in the absence of a binding partner. However, given such large-scale mobility of opposing termini, together with minor structural variations within the aligned central regions in the merged 3D reconstructions, we cannot exclude the possibility that local fluctuations occur along the length of the protein. In addition, there appears to be some preference toward certain conformations vs. others, with a large proportion of particles belonging to the proposed transition for Ltn1’s terminus-2 represented by column 4 (Fig. 2D), as demonstrated by the heat map describing the percentage of total reconstructed particles represented by each conformer (Fig. 2C). Ltn1’s mobility can be better appreciated when the full spectrum of aligned conformers is merged into a single volume (Fig. 2D, Inset). Together, these data are indicative of fluctuation between different Ltn1 conformers, whereby the heterogeneity can be described by independent terminus mobility around two hinge regions.

Analysis of Truncated Ltn1 Allows for Orientation of N and C Termini.

To orient the N and C termini of Ltn1 in the structures we generated, we analyzed a truncated protein in which the first 476 aa have been deleted (Figs. S2 and S7). We will refer to the two proteins as Ltn1 (amino acids 1–1562; full-length) and ∆N-Ltn1 (amino acids 477–1562), respectively. To determine the types of structural features present in the ∆N-Ltn1 data, we performed the alignment and classification of ∆N-Ltn1 in an unsupervised fashion, followed by an RCT reconstruction of each class. Then, to identify the most probable region within the Ltn1 electron density to which ∆N-Ltn1 corresponds, we aligned the resulting classes to the full set of 22 class averages from Fig. 2. From the complete data set (Fig. S8), Fig. 3 A–D shows four representative examples of ∆N-Ltn1 2D averages and 3D reconstructions, aligned to and subsequently overlaid with their highest correlating Ltn1 counterparts. As is the case with Ltn1, ∆N-Ltn1 conformers present a continuum of movement about a hinge. For example, the seemingly unobstructed rotation of terminus-2 in Fig. 2 is also suggested by the ∆N-Ltn1 data, in which the density enclosed by the red boxed region in Fig. 3 flips upside down between columns (Fig. 3 A and B), indicative of a 180° rotation. Together, the reconstructions of different ∆N-Ltn1 conformers suggest that Ltn1’s N terminus corresponds to terminus-1, whereas its RING domain-containing C terminus corresponds to terminus-2.

Fig. 3.

Orientation of Ltn1’s N and C termini: unsupervised alignment and classification of raw ∆N-Ltn1 particles. Columns are labeled (A–D) and refer to distinct conformers of ∆N-Ltn1. Rows refer to 2D class averages of ∆N-Ltn1, highest correlating 2D class averages of Ltn1 aligned to the ∆N-Ltn1 conformer, RCT reconstruction of the ∆N-Ltn1 class, corresponding RCT reconstruction of the Ltn1 class, and an overlay of the two volumes, the latter now represented in mesh and colored gray. Red boxes around the electron density are indicative of a 180° rotation around the hinge. Approximate angles describing the relative extent of folding within the volume are also indicated. (Scale bars: 200 Å.)

Mobility of Ltn1’s Termini Is Evident in Vitrified Aqueous Preparations.

Given the unusually large number of conformational snapshots of Ltn1 identified by negative stain single-particle analysis, we set out to verify whether similar flexibility occurs in solution, which can be addressed by vitrification and cryo-EM (18). In contrast to negative stain, for which Ltn1 assumes a preferred orientation on the solid support film, in the vitreous ice, it adopts a variety of orientations (Fig. S9). By using a neutral starting model, the cryo-EM data set was refined by calculating one or four output reconstructions, which were subsequently subjected to variance analysis to show the primary regions of intra- and intermodel variability (Materials and Methods). As a result of the significantly lower signal-to-noise ratio of cryo-EM data and additional degrees of freedom resulting from a lack of preferred orientation on the carbon support, it was possible to recover density maps only for the four displayed (Fig. 4). In both cases, a high degree of heterogeneity is evident within the vicinity of the C-terminal hinge point, which is consistent with the larger magnitude of mobility that is observed in stain (compare Fig. 4 with Fig. 2). This data implies that the basic architecture of Ltn1 preserved in vitreous ice is similar to that observed by using negative stain, and that the predominant heterogeneity in the structure accounts for the mobility of its C terminus.

Fig. 4.

Mobility of Ltn1’s termini is present in frozen hydrated preparations: a neutral initial model (selected based on the highest cross-correlation coefficient between a cryo- and negative stain class average) was used to reconstruct one (Upper) or four (Lower) electron density maps. For the single-model refinement, a variance analysis was performed based on bootstrap resampling of the data (38). For the four-model refinement, a variance analysis was performed between the resulting electron density maps. Variance is displayed in red.

Structure of Ltn1 Is Reminiscent of Cullin Scaffold of Multisubunit CRL E3s.

A large subset of RING E3s belong to the CRL family of proteins, a group of multisubunit complexes consisting of a scaffolding cullin subunit, a small RING domain subunit, and interchangeable substrate binding subunits (10). The crystal structure of a CRL containing the CUL1 cullin scaffold (often termed SCF) showed that the RING and substrate binding domains are arranged at opposite ends of an elongated complex (Fig. 5A) (19). We note that the EM density of Ltn1, in particular the overall architecture of its backbone fold, is strikingly similar to the cullin scaffold of SCF and other CRL complexes (20) (Fig. 5 B and C). This was unexpected because, with the exception of the C-terminal RING domain, Ltn1 does not have any obvious homologues in the protein databank for any portion of its sequence and is predicted to be composed of HEAT/ARM repeat motifs.

Fig. 5.

The cullin component of CRLs displays less intrinsic flexibility than Ltn1. (A) Schematic representation of generic multisubunit CRL E3s, with an arrow indicating ubiquitin (Ub) transfer between an E2-conjugating enzyme and target substrate. (B) Cullin–RBX1 components of SCF from Zheng et al. (19) displayed side-by-side with the (C) average cryo-EM map of Ltn1 from Fig. 4. The cullin scaffold-containing CUL1 is colored yellow, and RBX1 (amino acids 19–106), the RING-domain subunit, is colored green. Functionally relevant regions are indicated. (D) RCT reconstruction of one conformer of cullin–RBX1 identified by negative stain EM at ∼40-Å resolution. (E) RCT reconstruction of the second conformer of cullin-RBX1 identified by negative stain EM at ∼40-Å resolution. (F) Overlay of the two conformers identified by EM, showing apparent flexibility in the vicinity of the region between the second and third cullin repeat. The cullin–RING component of SCF from Zheng et al. (19) was subjected to molecular dynamics simulations by Liu and Nussinov (21). (G) The starting Protein Data Bank structure in Liu and Nussinov (21) at 0 ns is fit to the reconstruction from D. (H) The ending Protein Data Bank structure in Liu and Nussinov (21) with maximum rotation angles is fit to the reconstruction from E. (I) The cullin–RING component of SCF is shown as ribbons, overlaid on a 40-Å electron density map simulated from itself, to depict a more accurate comparison with the resolution of EM maps.

The overall similarity between Ltn1 and the cullin core of CRLs prompted the examination of CUL1 by EM (these analyses used a complex of human CUL1 and the RING domain subunit RBX1: 89.7-kDa and 12.3-kDa proteins, respectively). Although CUL1’s large size and helical composition are predicted to be comparable to Ltn1’s, the purified CUL1–RBX1 complex displays limited flexibility under the conditions of our analyses (Fig. 5 D and F). Only two conformers of the particles were apparent, with flexibility appearing to reside between the second and third cullin repeats; no obvious intermediates could be identified. Interestingly, these exact same conformations within the cullin scaffold were independently identified by molecular dynamics simulations (21) (Fig. 5 G and H). These observations imply that Ltn1’s extensive flexibility observed by EM cannot be simply attributed to an artifact resulting from its large size and extended shape.

Discussion

In this study, we presented a comprehensive 2D and 3D characterization of Ltn1’s dynamic structure by single-particle methods and were able to directly visualize extensive conformational variability inherent to the 180-kDa protein. We show that Ltn1 contains two flexible termini that undergo distinct movement about two hinge regions, with the end harboring the ubiquitylation-mediating C-terminal RING domain exhibiting the largest mobility (Figs. 2, 4, and 6). Motif analyses predicted that the majority of the protein is composed of HEAT/ARM repeats, whose overall dimensions would fit comfortably into our EM densities (Fig. S10). Ltn1 is similar to the cullin subunits of CRL E3s in overall size and shape. However, in contrast to the large mobility range of Ltn1, the isolated cullin subunits of CRL complexes examined under our conditions displayed only small conformational variability that has been independently predicted by using molecular dynamics simulations (21), but not directly observed. Based on the combined bioinformatic and EM analysis of Ltn1, we suggest that a HEAT/ARM-repeat superhelix forms the backbone scaffold of the protein.

Fig. 6.

Model of Ltn1 function in the context of translational surveillance. Ltn1 contains a flexible region within (A) its N terminus and (B) its C terminus, which are predicted to (A) confer selective binding to the 60S ribosomal subunit and (B) facilitate the polyubiquitylation of nascent chains that are bound for proteasomal degradation. A full description is provided in the Discussion. In the diagram, the 80S ribosome is colored light blue (60S subunit) and light yellow (40S subunit). Multiple snapshots of Ltn1 are superimposed and centrally displayed. Four ubiquitins, attached to a substrate lysine on the nascent chain, are colored red and orange. All structures are drawn to scale to show true size variations. For simplicity, the nascent chain is artificially elongated and is displayed without chaperones or other associated proteins.

Like Ltn1, E3s of the CRL family have also been shown to contain flexible regions (22–26). For example, it is known that the linkage between The CUL1–RBX1 subunits can be highly flexible, regulated by and dependent on posttranslational modification of the complex with the ubiquitin-like protein Nedd8 (22). Furthermore, additional flexibility for certain CRL complexes has been observed at the level of the substrate receptor (26) and at the level of cullin linkage to the substrate adapter subunits (23). The latter implied a ∼100-Å zone for the ubiquitylation of a diverse set of substrates involved in DNA lesion repair. These examples suggest that flexibility plays an important role for E3 function, and that its nature and extent will vary depending on the biological function of the protein.

Based on our understanding of Ltn1 in mediating polyubiquitylation of ribosome-stalled nonstop proteins, we can hypothesize a general role of flexibility for this biological function (Fig. 6). Currently, the regions involved in the Ltn1–ribosome association, and whether this association occurs directly or indirectly, are not understood. Drawing on the parallels with CRLs, as well as on the observations that Ltn1 associates with the 60S ribosomal subunit (7, 27) and that its N terminus is conserved in evolution, we suggest that this portion of the protein may contain a site that confers specificity in ribosome attachment (Fig. 6A). It is conceivable that the ribosomal binding site for Ltn1’s N terminus may lie far from the nascent peptide exit tunnel, explaining the accompanying N-terminal flexing within this region, as well as Ltn1’s evolutionarily conserved large size. The prevalence of HEAT/ARM repeats within Ltn1’s polypeptide chain may serve as a long spacer that confers structural malleability and/or extension around the ribosome in a manner that may be analogous to the signal recognition particle (28). Regardless of the manner of association, the key outcome of the Ltn1–ribosome interaction is expected to be the positioning of the RING domain-containing C terminus in proximity to the nascent chain as it exits the ribosomal tunnel. At that point, the range of motion in Ltn1’s C-terminal end would facilitate the attachment of ubiquitin to the appropriate substrate lysine by the RING domain-bound E2-ubiquitin complex (Fig. 6B). The enormous heterogeneity of quality control substrates and the dynamic nature of nascent polypeptides emerging from ribosomes implies that target lysines are likely not to be positioned in predetermined locations that are optimized through evolution for accepting ubiquitin, as may be the case for nonquality control (i.e., regulatory) substrates; it is thus just as likely that a ubiquitin-accepting lysine would be present near the ribosome tunnel exit as it is some distance away (Fig. S11 A and B), and the continuous flexibility within Ltn1 may exist to cope with this matter. In addition, flexibility in Ltn1’s C terminus is expected to facilitate the subsequent elongation of a polyubiquitin chain (Fig. 6B). Such a role for flexibility in optimizing multiple orientations of E2–Ub complexes relative to “moving targets” during polyubiquitin chain elongation is a function that is probably common among many E3s (Fig. S11C).

To date, direct evidence for conformational variation within E3s has been limited to crystallographic (22–24, 26, 29), NMR (30), and small angle X-ray scattering (22, 30) experiments. When pursuing structural studies of flexible macromolecules, crystallography provides a wealth of information at atomic precision and can account for several key structural states related to the function of the protein, but is hampered by the constraints of the crystal lattice and may not be well suited to characterize a large number of conformers. NMR is challenging for structures larger than ∼30 kDa, and small angle X-ray scattering provides an indirect readout in the form of a 1D scattering profile that must usually be accompanied by previous structural data together with predictive methods to interpret dynamic behavior (e.g., molecular dynamics simulations). In contrast, single-particle EM is capable of providing a broad assessment of conformational heterogeneity (31, 32). Its integration with other structural and biochemical methodologies is therefore expected to allow for a more detailed understanding of the mechanisms of protein dynamics. The approaches developed here for classifying conformational snapshots of Ltn1 by using iterative alignment and classification image processing strategies open up the possibility for the performance of analyses on other highly flexible macromolecules to enable direct visualization of individual proteins and protein complexes. They may serve as valuable tools to complement structural studies of E3 function, given the emerging theme of conformational variation observed in these proteins.

Materials and Methods

Experimental procedures are summarized below. All details can be found in SI Materials and Methods.

Cloning, Expression, and Purification of Yeast Ltn1 Recombinant Proteins.

Ltn1 and Ltn1_477-1562 (ΔN-Ltn1) constructs with N-terminal His₆-Smt3 tags were generated and used for protein expression. Cells were harvested, lysed by sonication, and clarified by centrifugation. Proteins were first purified by Ni-NTA resin chromatography, followed by size-exclusion chromatography. The His₆-Smt3 tag was then removed by Ulp1. After tag removal, proteins were further purified by ion-exchange chromatography, pooled, and concentrated.

Cloning, Expression, and Purification of Human CUL1-RBX1.

The CUL1–RBX1 complex was purified from a 100-mL culture of Hi5 insect cells (10⁶ cells/mL) infected with recombinant baculoviruses for 40 h to express full-length human hexahistidine-tagged CUL1 and untagged RBX1. Proteins were purified through Ni-NTA resin, then desalted.

EM Specimen Preparation.

For all negative stain samples, 3 μL of sample at a concentration of 0.01 mg/mL was applied to a C-flat grid with 2-−m diameter holes overlaid by thin (∼1.5 nm) continuous carbon, stained with 2% uranyl formate (wt/vol), as described previously (33). For cryosample preparation, 3 μL of Ltn1 at a concentration of 0.25 mg/mL was applied to a C-flat grid with a 1.3-μm hole diameter, allowed to adsorb for 30 s, then plunged into liquid ethane by using a manual plunger.

Negative Stain EM Data Collection and Image Processing.

For Ltn1, data were acquired using a Tecnai F20 Twin transmission electron microscope operating at 120 kV, using a dose of ∼15 electrons per square Ångstrom (e-/Ų) and a nominal underfocus ranging from 1.2 to 2.6 µm. Tilt pairs were collected at a nominal magnification of 50,000× at a pixel size of 0.218 nm at the specimen level. All images were recorded with a Gatan 4k × 4k pixel CCD camera by using the Leginon data collection software (34) and processed by the Appion software package (35). After removing bad particles, 31,297 untilted particles were put into an iterative procedure consisting of alignment, classification, and RCT reconstruction from every class average (Figs. S4 and S5). We retained all unique volumes for which, after aligning the central backbone to all comparable conformers, a rotation of greater than 10° was observed in either of the two termini at an identical threshold level. Class averages that were separated at the level of classification but produced density maps with less distinguishable terminus rotations of <10° were grouped into a single reconstruction, but only if the result produced an improvement in resolution. ∆N-Ltn1 data were acquired in an identical manner as Ltn1 and provided 13,980 particles for the iterative RCT procedure. CUL1–RBX1 data were collected at a nominal magnification of 80,000×, with 9,515 particles left for RCT reconstructions.

Cryo-EM Data Collection and Image Processing of Ltn1.

Data were acquired using a Tecnai F20 Twin transmission electron microscope operating at 120 kV, using a dose of ∼30 e-/Ų and a nominal underfocus ranging from 3 to 5 µm. A total of 380 images were automatically collected and recorded as described earlier, at a nominal magnification of 80,000× at a pixel size of 1.37 Å at the specimen level. A neutral starting model that was low-pass–filtered to 60 Å and did not contain evident terminus mobility was used for 3D reconstructions, all of which were performed with the RELION package (36). Variance analysis of the single model was performed from 20,000 bootstrap volumes in the SPARX package (37) and by using codimensional PCA (38). To show intermodel heterogeneity, variance analysis between the density maps that resulted from a four-model refinement in RELION was calculated. All maps were displayed in UCSF Chimera (38).

Accession Numbers.

Structures from the negative stain data set of Ltn1 were deposited in the Electron Microscopy Data Bank under accession codes EMD-2248–EMD-2269. The aligned negative stain data set of Ltn1 is made publicly available for interactive viewing by focused classification using customized masks at http://maskiton.scripps.edu.