E3 ubiquitin-protein ligase TRIM21-mediated lysine capture by UBE2E1 reveals substrate-targeting mode of a ubiquitin-conjugating E2

The E3 ubiquitin-protein ligase TRIM21, of the RING-containing tripartite motif (TRIM) protein family, is a major autoantigen in autoimmune diseases and a modulator of innate immune signaling. Together with ubiquitin-conjugating enzyme E2 E1 (UBE2E1), TRIM21 acts both as an E3 ligase and as a substrate in autoubiquitination. We here report a 2.82-Å crystal structure of the human TRIM21 RING domain in complex with the human E2-conjugating UBE2E1 enzyme, in which a ubiquitin-targeted TRIM21 substrate lysine was captured in the UBE2E1 active site. The structure revealed that the direction of lysine entry is similar to that described for human proliferating cell nuclear antigen (PCNA), a small ubiquitin-like modifier (SUMO)-targeted substrate, and thus differs from the canonical SUMO-targeted substrate entry. In agreement, we found that critical UBE2E1 residues involved in the capture of the TRIM21 substrate lysine are conserved in ubiquitin-conjugating E2s, whereas residues critical for SUMOylation are not conserved. We noted that coordination of the acceptor lysine leads to remodeling of amino acid side-chain interactions between the UBE2E1 active site and the E2–E3 direct interface, including the so-called “linchpin” residue conserved in RING E3s and required for ubiquitination. The findings of our work support the notion that substrate lysine activation of an E2–E3-connecting allosteric path may trigger catalytic activity and contribute to the understanding of specific lysine targeting by ubiquitin-conjugating E2s.

The E3 ubiquitin-protein ligase TRIM21, of the RINGcontaining tripartite motif (TRIM) protein family, is a major autoantigen in autoimmune diseases and a modulator of innate immune signaling. Together with ubiquitin-conjugating enzyme E2 E1 (UBE2E1), TRIM21 acts both as an E3 ligase and as a substrate in autoubiquitination. We here report a 2.82-Å crystal structure of the human TRIM21 RING domain in complex with the human E2-conjugating UBE2E1 enzyme, in which a ubiquitin-targeted TRIM21 substrate lysine was captured in the UBE2E1 active site. The structure revealed that the direction of lysine entry is similar to that described for human proliferating cell nuclear antigen (PCNA), a small ubiquitin-like modifier (SUMO)-targeted substrate, and thus differs from the canonical SUMO-targeted substrate entry. In agreement, we found that critical UBE2E1 residues involved in the capture of the TRIM21 substrate lysine are conserved in ubiquitin-conjugating E2s, whereas residues critical for SUMOylation are not conserved. We noted that coordination of the acceptor lysine leads to remodeling of amino acid side-chain interactions between the UBE2E1 active site and the E2-E3 direct interface, including the so-called "linchpin" residue conserved in RING E3s and required for ubiquitination. The findings of our work support the notion that substrate lysine activation of an E2-E3-connecting allosteric path may trigger catalytic activity and contribute to the understanding of specific lysine targeting by ubiquitinconjugating E2s.
Tripartite motif (TRIM) 5 proteins constitute the largest subfamily of RING-type E3 ubiquitin ligases, with around 100 members in humans, and are associated with pathological conditions (1,2). RING-type E3s catalyze the direct transfer of ubiquitin (Ub), or a ubiquitin-like (Ubl) entity such as SUMO or NEDD8, from a thioester-linked E2-conjugating enzyme to specific substrates in the ubiquitination pathway (3,4). The multimodular TRIMs comprise an N-terminal RING domain, one or two B-box domains, a coiled-coil region, and a C-terminal substrate-binding domain (5) and predominantly support ubiquitination (6). TRIM21 functionality appears to rely upon its ability to specifically catalyze the formation of multiple Ub chain types, with several distinct E2s, in both nuclear and cytosolic cell compartments and onto a variety of different substrates. TRIM21 (also denoted Ro52 or SSA) was first identified as a major autoantigen in systemic lupus erythematosus and Sjögren's syndrome (7), and RING-domain specific patient autoantibodies impair TRIM21-mediated autoubiquitination by blocking the E2-E3 interaction (8). TRIM protein autoubiquitination in general has been shown to inhibit viral DNA synthesis, direct interferon regulatory factor signaling (9), and steer cellular differentiation (10). We and others have shown that both the cytosolic UBE2D1 (UbcH5a) and the nuclear UBE2E1 (UbcH6) (11,12) collaborate with TRIM21 in mediating polyubiquitination (13,14). Nuclear translocation of TRIM21 has been observed as a result of inflammatory signaling (12,15), and a splice variant, TRIM21␤, lacking part of the coiled-coil domain also demonstrated a predominantly nuclear localization (12). TRIM21 negatively regulates innate immune signaling by promoting Lys 48linked substrate ubiquitination of nuclear interferon regulatory factors (16 -19). TRIM21 also polyubiquitinates cytoplasmic targets such as the DDX41 DEAD-box protein (20) and mediates monoubiquitination of cytoplasmic substrates, including IKK␤ (21) and GMP synthase (22). Finally, TRIM21 autoubiquitination by consecutive Ube2W and Ube2N/Ube2V2 activity produces Lys 63 -linked Ub chains, both free and anchored to the TRIM21 N terminus, with a suggested role in virus neutralization (23).
Direct interaction between an E2 and a corresponding E3 is essential for RING-mediated ubiquitination, where the "linchpin" arginine residue in the RING domain (24) and the conserved "SPA" motif in E2 loop 7 (25) have been shown to be critical for enzymatic activity (3,4). Motifs flanking the E3 RING domain have been shown to stabilize the donor ubiquitin in a "closed state" most favorable for ubiquitin transfer (24, 26 -31). Such motifs without interacting with the substrate are able to turn on and off ubiquitination activity entirely in response to other signaling factors such as phosphorylation and/or multimodular domain interactions.
Knowledge of substrate-targeting modes and E3-catalyzed substrate transfer mechanisms in Ub-substrate conjugation is scarce because no structures of captured substrates have been determined for a Ub-conjugating E2-E3 complex. However, structural studies of larger multidomain complexes, including SUMO-and NEDD8-targeting E2s UBC9 and UBCH12 trapped in action with their substrates, have revealed how key residues around the E2 active site support SUMO/NEDD8 conjugation at specific substrate residues (32)(33)(34)(35)(36). Catalytically inactive modules can assist in positioning specific acceptor lysines from substrates into the E2 active site, thus placing substrate specificity partly outside of the direct E2-substrate interaction (34,37) and for human PCNA have been shown to support an alternate substrate entry path for SUMOylation (33). Whether Ub-conjugating E2s use the same specificity mechanisms is unclear, in particular because several key residues required for conjugation activity in SUMO-and NEDD8-ylating E2s are not conserved in Ub-conjugating E2s. Indeed, lack of detailed structures hampers the advancement of knowledge required to specifically target pathological conditions related to ubiquitination (1,2).
In this work, we have investigated the UBE2E1-TRIM21 interaction, where TRIM21 acts both as an E3 catalyst and as a substrate in autoubiquitination. Our resulting TRIM21-bound UBE2E1 crystal structure together with that of free UBE2E1 present structural snapshots that reveal an acceptor Ub-lysine recognition mode that is similar to the lysine entry path for human PCNA (33). Finally, we show how the presence of an acceptor lysine at the E2 active site confers substrate-induced conformational changes that extend to the E2-E3 direct interface, and we propose a model for how this could activate linchpin-mediated ubiquitination.

Results
Structural and functional assembly of the UBE2E1 C -TRIM21 R complex TRIM21 ␣ (cytosolic) and ␤ (nuclear) isoforms are identical within the TRIM21 RING (residues 1-91; TRIM21 R ) fragment used in the complex crystal structure. Corresponding E2s UBE2D1 (cytosolic) and UBE2E1 (nuclear) are closely homologous. We affirmed experimentally that the respective cellular localization of TRIM21 ␣ and ␤ isoforms are indeed compatible with both UBE2D1 (cytosolic) and UBE2E1 (nuclear) E2 enzymes in HeLa cells, transfected with GFP/JRed-tagged constructs or stained with specific antibodies (Fig. S1, A and B).
The structure of a complex comprising the TRIM21 R domain with flanking helices and the catalytic core domain of UBE2E1 (residues 36 -193; UBE2E1 C ) was determined by crystallography to a resolution of 2.82 Å ( Fig. 1A and Table 1, PDB code 6FGA). The complex crystal structure includes four homodimeric TRIM21 R and seven UBE2E1 C entities in the unit cell, which together form four TRIM21 R :UBE2E1 C assemblies: two well-defined 2:2 complexes (chains IFDL and MBCO), one 2:2 complex with poor density in one of the E2 entities (chains NEAK), and one 2:1 complex (chains JHG) (Fig. 1A). TRIM21 R is predominantly a dimer both in crystal and in solution as estimated by small-angle X-ray scattering (SAXS) and analytical gel filtration (Figs. 1B and S3 and Tables S2 and S3). SAXS measurements of TRIM21 R reveal an overall similar shape in solution as in the corresponding crystal structure dimer (Fig.  1B). The quaternary arrangement of E2 and E3-RING dimer modules is highly similar to that in previously determined E2-RING-Ub assemblies (24, 26 -31) (Figs. 1C and S2B and  Table S1). Finally, to compare free and bound E2 states in this study, we obtained a crystal structure (1.4 Å) of similarly prepared free UBE2E1 C (Table 1, PDB code 5LBN), which is similar to matching residues within full-length UBE2E1 (PDB code 3BZH, r.m.s.d. 0.47 Å (38)) (Fig. S2C).
The interface connecting UBE2E1 C -H1, -L4, and -L7 with TRIM21 R -L1, -H1, and -L2 (where H represents helix and L represents loop) is well-defined in the TRIM21 R -UBE2E1 C crystal structure (Fig. 1, D and E, and Table S1). Significant amide chemical shift perturbations (CSPs) were observed by NMR in the direct interface (Fig. 1F), and a K d of 24 Ϯ 11 M was estimated for the TRIM21 R -UBE2E1 C interaction, based on CSPs in five titration points for nonbroadened residues (Fig. S4, A-C). Significant CSPs were also observed for residues in a contact network extending from the direct interface to the active-site region (Fig. 1F), in full agreement with previously proposed allosteric activation through the E2 core (39 -41).
To functionally probe the interface, we used mutational mapping assayed by autoubiquitination (8) and E2-Ub hydro-

TRIM21-UBE2E1 complex offers new insights in ubiquitination
lysis assays of an oxyester-bonded UBE2E1-Ub complex (27,34). By sequence homology, Arg 55 in TRIM21 corresponds to the catalytic linchpin residue in E2-mediated Ub conjugation (24). In agreement, both autoubiquitination and E2-Ub hydrolysis assays were inhibited in TRIM21 R -R55A (Fig. 1G). CSPs indicate that TRIM21-R55A interacts with UBE2E1 similarly as wildtype (WT) (K d Ϸ 50 M; Fig. S4, D and E), supporting that both ubiquitination and E2-Ub hydrolysis depend on the presence of a catalytic element and not simply on complex forma-tion (24,34). In the direct contact interface, a TRIM21 R -L20A mutation significantly reduces autoubiquitination (Fig. 1G), and very small NMR CSPs were observed for UBE2E1 C with TRIM21-L20A (Fig. S4F), indicating disrupted binding. Similarly, in the SPA motif of UBE2E1 C loop 7 (uL7), a UBE2E1 C -A142D mutant entirely disrupts the complex formation as observed by NMR (Fig. S4G), in agreement with the SPA region being critical for TRIM21-catalyzed conjugation activity (25,42). Altogether, three complete TRIM21 R :UBE2E1 C 2:2 assemblies are present in the ASU comprising chains MBCO, IFDL, and NEAK, and a 1:2 complex is present comprising chains JHG. B, bead models representing the solution structure of free TRIM21 R derived from the SAXS data using DAMMIF and assuming P2 symmetry (cyan) or no symmetry (blue dots) overlaid with the TRIM21 R dimer crystal structure. C, cartoon representation of UBE2E1 C (gray)-TRIM21 R (green/cyan) complex crystal structure with Zn 2ϩ shown as spheres (black); this coloring is maintained in Figs. 2-4. D, structure of UBE2E1 C -TRIM21 R direct interface (square in C; showing contacts in C). E, overview of TRIM21 R -UBE2E1 C contacts (lines), including hydrogen bonds or salt bridges and van der Waals interactions (black, solid); proposed interactions are in gray, dotted lines (44,45). F, combined 1 H and 15 N CSPs of 15 N-labeled UBE2E1 C in the presence of 2.0 eq of unlabeled TRIM21 R . Average CSP value is represented as the solid line; the dashed line is with one standard deviation added. Bars are colored according to reduced accessible surface area as determined by VADAR (73) from white (0%) to black (100%). Inset, cartoon representation of UBE2E1 C . Residues with significant CSPs are shown as red spheres, smaller for CSPs above average, bigger for CSPs above 1 S.D. from average, and colored salmon if buried (Ͼ20%). The active-site region is indicated (yellow). G, in vitro autoubiquitination assays with UBE2E1 and UBE2D1 show extent of TRIM21 R WT activity and loss of activity in TRIM21 R mutants as annotated. IB, immunoblotting. UBE2D1 shows a similar but not complete loss of autoubiquitination with TRIM21 R -R55K and -L20A mutants compared with UBE2E1 (Fig. 1G). The NMR CSP imprint of TRIM21 R on UBE2D1 is highly similar but slightly less stringent compared with that of UBE2E1 (Fig. S1C). Jointly, these observations could indicate more promiscuous and thereby more permissive catalytic activation of UBE2D1 by TRIM21, in agreement with earlier findings (43). Previous studies have suggested UBE2E1-Glu 105 (Asp in UBE2D1) and -Lys 109 as hot-spot residues in UBE2E specificity (44,45). We found small but distinct NMR CSPs for UBE2E1 C -Lys 109 (Fig. 1F) and for the corresponding Lys 63 in UBE2D1 (Fig. S1C), supporting a possible role for this residue in TRIM21 recognition.

TRIM21 R activity relies on a closed TRIM21 R -UBE2E1 C -Ub conformation
The ternary TRIM21 R -UBE2E1 C -Ub complex crystals resulted in low diffraction (Ͼ6 Å) and showed extensive line broadening by NMR experiments, suggesting dynamic properties. However, we could straightforwardly model the UBE2E1 C -Ub-TRIM21 R complex, supported by close E2-E3 structural similarity to a wealth of ternary E2-Ub-E3 complexes (24, 26 -31) (Figs. 2A, S6, and S7A). In this model, TRIM21 R residues Glu 12 , Glu 13 , Arg 67 Ј, and Asn 71 Ј (Ј symbol represents residues of the other RING protomer) hold positions that could stabilize a closed Ub conformation and thereby affect activity, as first shown for c-Cbl (46, 47) ( Fig. 2A). Indeed, E2-Ub hydrolysis was severely compromised for TRIM21 R mutants E12A, E13A, E13K, double mutant E12K/E13K, R67ЈA, and N71ЈA (Figs. 2B and S5) even if the E2-E3 interaction was retained as shown by NMR (Fig. S4H). The same TRIM21 R mutants are also poorly active in autoubiquitination assays, where intrinsic UBE2E1 autoubiquitination instead becomes visible in reactions with no or poorly functioning E3 ( Fig. 2C) (48). Interestingly, the single TRIM21 R -E12K mutant is as active as WT TRIM21 R in autoubiquitination assays but still shows greatly reduced activity in E2-Ub hydrolysis (Figs. 2B and S5). Indeed, a similar effect was observed for E12R in the related TRIM25, further supporting functional similarities between these TRIMs (30). Taken together, our results identify residues in TRIM21 R helices flanking the core RING motif that significantly affect Ub transfer, presumably by stabilizing a "closed" Ub conformation in a ternary complex.

Crystal capture of a TRIM21 Lys 61 acceptor lysine in the UBE2E1 active site
In the TRIM21 R -UBE2E1 C complex, we observed that the UBE2E1 C active site of the NEA chain assembly contacts Lys 61 in the TRIM21 R F chain. This chain is adjacent to the NEA assembly in the asymmetric unit (ASU) and is here labeled TRIM21 R Љ (Figs. 1A and 3A). The Lys 61 Љ side chain is well-

TRIM21-UBE2E1 complex offers new insights in ubiquitination
accommodated in a pocket lined by UBE2E1 C residues Asp 133 and Asp 163 (Fig. 3B). Within the resolution of the structure, the lysine ⑀ acceptor group could easily form hydrogen bonds with Asp 133 and Asp 163 side-chain carboxylates and is within 5 Å of the active cysteine (Fig. 3B). Asp 133 further bolsters the interaction by a hydrogen bond stabilizing the Asn 62 Љ side-chain amide (Fig. 3B). In the complex, the side-chain orientations of Lys 61 Љ, Asp 133 , and Asp 163 are all supported by well-defined electron densities (Figs. 3B and S7B). In contrast, the structure of free UBE2E1 C shows very poor density for Asp 163 despite the higher-resolution data, indicating a disordered orientation of this residue in the unbound state (Figs. 3C and S7C); similar disorder is observed also in substrate-free states of full-length UBE2E1 (PDB code 3BZH), UBE2D1 (PDB code 2C4P), and ubiquitin-conjugated UBE2D1 (PDB code 4AP4). Together, this suggests that lysine-coordinating residues are ordered on substrate lysine coordination in the active site.
We probed the nature of Lys 61 Љ as a possible target residue for autoubiquitination in several ways. First, a K61A mutation severely disrupts TRIM21 R autoubiquitination with both UBE2E1 and UBE2D1 (Fig. 4A). A TRIM21 R -N62A mutation similarly disrupts ubiquitination, whereas TRIM21 R -N62R ubiquitination is close to WT, both by UBE2E1 and UBE2D1 Because TRIM21 serves both as E3 and substrate in autoubiquitination, we critically interrogated whether our TRIM21 R -K61A mutation might itself impair the E3 activity of TRIM21 by disrupting interactions or catalytic functions. First, NMR CSP analysis shows that the UBE2E1 C -TRIM21 R -K61A binding pattern is highly similar to that of WT TRIM21 R (Fig. 4B). Second, in ubiquitin discharge assays, both K61A and N62A mutants are as active as TRIM21 R , suggesting that these mutant E3s fully retain their ability to catalyze the release of Ub (Fig.  S5). Third, to assay the capacity of TRIM21 R -K61A in catalysis of Ub conjugation, we performed a reconstitution experiment with the non-E2-binding TRIM21 R mutant L20A as a pseudosubstrate (Fig. 4, C and D). If deficient autoubiquitination in TRIM21 R -K61A is only due to the lack of a target lysine and not to deficient catalysis, then TRIM21 R -K61A should still be able to catalyze ubiquitination of the L20A mutant at its retained Lys 61 . In agreement with this, we found restored

TRIM21-UBE2E1 complex offers new insights in ubiquitination
ubiquitination by an equimolar mixture of K61A and L20A mutants, in particular by UBE2D1 but also by UBE2E1 (Fig.  4C). Taken together, these experiments show that TRIM21 R -K61A interacts with UBE2E1 similarly as WT and is catalytically active both in Ub discharge and conjugation, which together with the deficient autoubiquitination for K61A implies that Lys 61 is indeed targeted in autoubiquitination by both UBE2E1 and UBE2D1.
Based on our structure, we then probed the roles of the Lys 61 Љ-coordinating residues Asp 133 and Asp 163 in catalysis and substrate recognition. A Ub-conjugated UBE2E1 C -D163A mutant is hydrolyzed similarly as WT in the presence of , the UBE2E1 C -TRIM21 R complex (chains MBC) (E), and the TRIM21 R Љ-UBE2E1 C -TRIM21 R (chains NEAF; similar to B) (F). G, in vitro autoubiquitination assay shows that the acidic residue Asp 133 and Asp 163 mutants of UBE2E1 C are essential for substrate ubiquitination; remaining ubiquitination pertains to known E3-independent UBE2E1 C autoubiquitination (48). H, E2-Ub oxyester hydrolysis assay of UBE2E1c shows that Asp 163 does not affect Ub release, whereas D133S results in lost activity. TRIM21 (Fig. 3H), whereas the same mutation entirely abrogates TRIM21-mediated polyautoubiquitination (Fig. 3G), leaving only the known slow intramolecular UBE2E1-Lys 136 ubiquitination (48) at a position close to the catalytic Cys 131 (Fig. 3B). Thus, Asp 163 appears to be primarily involved in substrate recognition. These results are in full agreement with corresponding D117A mutations in UBE2D1 (27, 49 -51) and with the observation that serine phosphorylation in the corresponding position activates Ube2A for ubiquitination (for a review, see Ref. 57).

TRIM21-UBE2E1 complex offers new insights in ubiquitination
To assay the role of Asp 133 in substrate recognition, we had to consider that this conserved aspartic acid anchors to the Ub C-terminal tail in the closed state in a range of Ub-conjugating E2s while employing the same rotamer as Asp 133 in the free state ( Fig. 3C) (26,27,30,31,35,51). In SUMO-conjugating UBC9, a serine corresponding to Asp 133 in UBE2E1 anchors identically to the SUMO C-terminal backbone (35,36), suggesting that a D133S mutation in UBE2E1 could reveal a role in substrate recognition without distorting Ub anchoring. Indeed, as for D163A, we found that TRIM21-mediated substrate ubiquitination is interrupted by a D133S mutation, whereas UBE2E1 internal autoubiquitination to a lysine proximal to the active site can proceed (48) (Fig. 3G). However, in contrast to D163A, we found that that the D133S mutation also aborts TRIM21-mediated E2-Ub hydrolysis (Fig. 3H), suggesting an additional role for this residue in TRIM21-mediated catalysis.

Substrate-induced active-site remodeling extends to the RING linchpin
To investigate whether the substrate-induced reorientation of Asp 133 could induce further structural changes, we compared our three structures of free, E3-bound, and E3 ϩ substrate-bound UBE2E1 (Fig. 3, D-F). In the absence of E3 and substrate, the orientation of Asp 133 is stabilized by an intramolecular side-chain hydrogen bond to Asn 138 , for which two side-chain rotamers were identified in the crystal structure (Fig.  3D). In the E2-E3 complex, the Asn 138 side chain of UBE2E1 is constrained into a unique rotamer, supported by electrostatic interactions with TRIM21 R -Arg 55 , but with the hydrogen bond to Asp 133 maintained (Fig. 3E). Finally, in the substrate complex, the Asp 133 side-chain carbonyl shows favorable interactions with Lys 61 Љ and Asn 62 Љ (Fig. 3, B and F) hydrogenbonding network, a conservative TRIM21 R -R55K mutation would interrupt Asn 138 interactions. Indeed, an R55K mutation disables both UBE2E1-mediated ubiquitination (Fig. 1G) and UBE2E1-Ub hydrolysis (Fig. S5) even though CSPs suggest a maintained E2-E3 interaction (Fig. S4I). Taken together, an Arg 55 linchpin-connected, hydrogen-bonding network may be critical for TRIM21-mediated catalysis of ubiquitination.

TRIM21-UBE2E1 complex offers new insights in ubiquitination Residues in the UBE2E1 active-site entry path are conserved in Ub-conjugating E2s
In the Ub-conjugating UBE2E1, the entry path of the targeted Lys 61 Љ is guided by Asp 133 and Asp 163 , which jointly line a negative crevice extending to the active-site cysteine (Fig. 5, A and  B). Interestingly, ubiquitin-conjugating, lysine-targeting E2s either hold a conserved Asp 133 (Asp/Glu), or a large, negatively charged L7 loop adjacent to the Asp 133 position (UBE2E1 numbering; Fig. 6A). Similarly, at the Asp 163 position, E2s active in lysine-anchored ubiquitination either hold a conserved Asp/ Asn or a phosphorylatable serine (52, 53) (Fig. 6A). Conservation of this negative crevice at the active site suggests that the substrate entry path presented here for UBE2E1/TRIM21 could be accessible also to other Ub-conjugating E2s.
The targeted TRIM21-Lys 61 in the current complex is wellpositioned with respect to the active-cysteine compared with the substrate-containing structures obtained for SUMO-and NEDD8-conjugating E2s (Table S5). A SUMO substrate entry path similar to that employed in UBE2E1 was shown for yeast UBC9 sumoylation of human PCNA where the substrate is presented to E2 in a multimodular complex that steers the substrate into the E2 active site (Fig. 5D) (33). In contrast, in the human SUMO-conjugating UBC9 complex with the substrate RanGAP1, the targeted Lys 524 enters at nearly right angles to TRIM21-Lys 61 Љ, similarly directed by UBC9-Asp 127 (equivalent to UBE2E1-Asp 163 ). Here, the substrate entry is critically bol-stered by UBC9-Tyr 87 (35,36) (Fig. 5C), which only occurs in UBE2A and UBE2B among Ub-conjugating E2s (Fig. 5A). A UBE2E1-like acceptor lysine entry into UBC9 would be structurally hindered by a UBC9-Lys 101 -Asp 127 ion pair gate (Fig.  5C). Reciprocally, a UBC9-like acceptor lysine entry into the UBE2E1 active site would be repelled by the equivalent of UBE2E1-Lys 136 where a positive charge is conserved in Ubconjugating E2s (Fig. 5C). NEDDylation relies on a complex but specific multimodular assembly that optimally positions the modules of E3, E2, and NEDD8 for catalysis, resulting in a lysine entry path similar to that in SUMO-conjugating UBC9 but does not hold the entire SUMO conserved pattern for substrate recognition of acceptor residue (Fig. 5D) (34).

Discussion
In this work, we present the crystal structure of a TRIM21-UBE2E1 complex comprising the TRIM21 RING domain. Although biochemical and mutational data for this complex consistently agree with observations for other Ub-conjugating E2s, our structure also presents the capture of a substrate lysine acceptor targeted for RING-mediated ubiquitination. The functional consistency between our UBE2E1-TRIM21 complex and other E2-RING complexes makes it plausible that also other Ub-conjugating E2s could conjugate their substrates in a similar manner.

TRIM21-UBE2E1 complex offers new insights in ubiquitination
A common denominator for both Ub-and SUMOylation is the critical functional role for the residue corresponding to UBE2E1-Asp 163 , which is conserved in both Ub-and SUMO/ NEDDylating E2s (53) (Fig. 5A). This aspartic acid appears to coordinate the substrate lysine and prepare it for conjugation (35, 36) (Fig. 3). However, as shown here, the different substrate entry paths adopted by UBE2E1-TRIM21 and UBC9 -human PCNA on one hand and UBC9 -RanGAP1 and NEDD8 -RBL1-CUL1 on the other jointly support the presence of varied substrate entry paths to E2 active sites. This suggests that the proposed general "gateway" role of the aspartic acid, UBE2E1-Asp 163 (Fig. 5A), would primarily be to select and coordinate the acceptor lysine rather than to steer the substrate entry path. Taken together, this supports that, in addition to multimodular On-path and Ub-interacting residues are shown as filled circles, labeled and colored according to charge; Cys 131 is yellow and unlabeled throughout. Side-chain disorder is indicated by multiple, transparent circles, and hydrogen bonds are indicated as double arrows. Final transfer of Ub from E2 to substrate is indicated by a transparent brown arrow (E).

TRIM21-UBE2E1 complex offers new insights in ubiquitination
steering (33), E2 entry paths depend on small sequence variations, which may also guide substrate specificity.
Extending current views on substrate recognition, the acceptor lysine in the current structure is also coordinated by Asp 133 , which is uniquely conserved in Ub-conjugating E2s (Fig. 5A). At first glance, this could seem unexpected because, in the absence of substrate, residues equivalent to Asp 133 in other Ub-conjugating E2s were shown to anchor to the C terminus of Ub in its closed state (26,27,30,31,35,51). In the SUMO-conjugating UBC9, a serine in the position corresponding to Asp 133 in UBE2E1 fulfils the same role in anchoring the highly similar SUMO C terminus but does not coordinate the substrate lysine (35,36). If the sole and primary function of Asp 133 in UBE2E1 is to stabilize a reactive, closed-state Ub, then a conservative D133S mutation in UBE2E1 should also support ubiquitination, which is opposed to our findings (Fig. 3G).
Our results suggest that, in ubiquitination, the conserved Asp 133 might have dual roles in supporting the closed state of E2-Ub and in recognizing acceptor substrate lysines. Indeed, a structural overlay of our structure with the TRIM25-UBE2D1-Ub ternary complex (Fig. S7D) suggests that both UBE2E1-Asp 133 and TRIM21-Arg 55 would be prompted to release their Ub-stabilizing interactions in response to substrate binding. The release of Ub from its closed, E2-anchored state onto the substrate would then be triggered by the activesite coordination of the substrate lysine acceptor, by a chain of events affecting residues in the Lys 61 /Asn 62 (substrate) -Asp 133 (E2)-Asn 138 (E2) -Arg 55 (E3) contact chain (Fig. 6, B-E). By connecting the E2 active site with its corresponding E3, such a chain of activation would then imply that E3-catalyzed ubiquitination is jointly mediated by its substrate. The current structure, obtained in the absence of Ub, might then represent a model for a transition step where Ub is no longer anchored onto E2 in a closed conformation but is being released upon substrate conjugation (Fig. 6E).
The present structure together with mutational data suggests that TRIM21-UBE2E1 may hold substrate selectivity toward a K(N/R) pattern. Indeed, specific TRIM21-mediated monoubiquitination targets the 182 KK pattern of the substrate GMP synthase (22). In DDX41, one of two Ub-Lys 48 -conjugated lysines holds an 8 RKR motif (20). In IKK␤, a monoubiquitinated (21), TRIM21-mediated site at Lys 163 and all three Ub-Lys 48conjugated sites hold motifs where Lys is flanked by an amidecontaining side chain ( 162 HK, 418 KR, 555 KQ, and 703 KK), and monoubiquitination by UBE2E1 at histone H2A occurs at the 118 PKKT motif (54). TRIM21 itself contains several additional K(N/R/K) motifs outside the TRIM21 R domain that could be targeted by autoubiquitination. UBE2T in the Fanconi anemia pathway, which holds the conserved Asp 133/163 pattern, spontaneously ubiquitinates FANCL at 522 RKQ (55) (Fig. 6A). Finally, E2s specifically targeting hydroxyls, cysteines, lipids, or N termini do not hold the Asp 133/163 conservation but instead show high variability in these active site-proximal positions (Fig. 6A), which may further support the importance of E2 active-site interplay with the substrate anchor site to fine-tune specificity in ubiquitination.
In a larger context, autoubiquitination of TRIM proteins has been observed as a mechanism for antiviral defense and corre-lates with inhibition of retroviral transcription (56,57). In studies of TRIM5 assembly on capsids, a TRIM5-TRIM21 RING chimera spontaneously assembled into hexagonal two-dimensional arrays of TRIM dimers of antiparallel coiled coils, which resulted in the presentation of three RING domains at each hexagonal corner (58). It has been suggested that two of the RING domains could then dimerize and catalyze E2-mediated ubiquitination of the third RING (59). Our current structure supports this hypothesis by providing a detailed molecular mechanism for how such autoubiquitination occurs and a new structural scaffold for investigating how this could be facilitated in a trimeric arrangement. Further high-resolution structural analysis of TRIM substrate complexes with functionally complementary E2/E3/Ubl partners will be essential to map their structural and functional versatility and will advance the analysis of functional properties in multimodular ubiquitinating complexes.

Recombinant protein expression and purification
TRIM21 constructs were expressed in Escherichia coli BL21(DE3) Rosetta-2 cells, induced with 0.2 mM isopropyl D-1thiogalactopyranoside and 20 M ZnCl 2 . After 16 -18 h at 18°C, the cells were lysed by sonication in 50 mM Tris-HCl, pH 8.0, 300 mM NaCl, 10% glycerol (v/v), 10 mM ␤-mercaptoethanol, 20 M ZnCl 2 , and 5 units/ml DNase I (Roche Applied Science). The supernatant was purified on Ni 2ϩ -NTA-agarose resin (Qiagen) and eluted with 100 -150 mM imidazole buffer. The His 6 tag was cleaved off with tobacco etch virus recombinant protease (62) or thrombin as required. Cleaved protein was passed over Ni 2ϩ -NTA-agarose resin, and the flowthrough was collected, concentrated, and subjected to Superdex 75 gel filtration (GE Healthcare) in 20 mM Tris-HCl, pH 8.0, 100 mM NaCl, 10% glycerol (v/v), 100 M ZnCl 2 , and 10 mM ␤-mercaptoethanol. Buffer optimization was performed using static light-scattering StarGazer-384 (Harbinger), aiming for consistent high stability without signs of aggregation. Compared with our previous work (8), the stability of the TRIM21 RING was much improved by removal of the His tag, which in turn allowed for an increased ZnCl 2 content without precipitation.

TRIM21-UBE2E1 complex offers new insights in ubiquitination
propyl D-1-thiogalactopyranoside for 20 h at 20°C. Harvested cells were lysed by sonication in 50 mM Tris-HCl, pH 8.0, 300 mM NaCl, 10% glycerol (v/v), 10 mM ␤-mercaptoethanol, and 5 units/ml DNase I. The supernatant was applied to a 5-ml His-Trap column (GE Healthcare) and eluted with imidazole gradient. The His 6 tag was cleaved by thrombin (25°C, 4 h) followed by gel filtration (Superdex 200, GE Healthcare). Isotope-labeled proteins for NMR were expressed in M9 minimal medium supplemented with [ 13 C]glucose and/or 15 NH 4 Cl (Cambridge Isotopes). UBE2E1 C mutants were prepared similarly as UBE2E1 WT. Preparation of recombinant Ube1 was carried out as described earlier (61).

E2-Ub oxyester hydrolysis assays
To generate the E2-Ub conjugate, UBE2E1 C -S68R/C131S, denoted as UBE2E1 SC (100 M), His-tagged ubiquitin (120 M), and His-tagged human Ube1 (5 M) were incubated for 16 -18 h at 30°C in a reaction buffer containing 20 mM Tris-HCl, 200 mM NaCl, 5 mM ATP, 5 mM MgCl 2 , and 10 mM ␤-mercaptoethanol. The E2-o-Ub (o represents oxyester) conjugate was first purified by Ni 2ϩ -immobilized metal-affinity chromatography to separate the E2-o-Ub conjugate from unconjugated E2 followed by His tag cleavage and size-exclusion chromatography on a Superdex75 column. Purified E2-o-Ub (15 M) was mixed with TRIM21 R or TRIM21 R mutants (10 M) and incubated for 180 min at 27°C with samples taken at several time points (5,15,30,60,90, and 180 min). Reactions were stopped by addition of SDS Laemmli buffer and analyzed by SDS-PAGE stained with Coomassie Blue R-250. The E2-o-Ub conjugate quantification on stained gels was performed using ImageQuant (GE Healthcare). Reactions were performed in triplicates, and rates are given as mean Ϯ 1 S.D.

Autoubiquitination activity reaction
Autoubiquitination assays were performed in 20-l reactions containing 0.50 M TRIM21 R or variants thereof, 100 ng of E1, 500 ng of UBE2E1, and 2.5 g of ubiquitin in a buffer containing 50 mM Tris-HCl, 2.5 mM MgCl 2 , 0.5 mM DTT, and 2 mM ATP. Each reaction mixture was incubated for 2 h at room temperature and terminated by addition of 5 l of 5ϫ SDS-PAGE sample buffer containing 100 mM Tris-Cl, 10% (w/v) SDS, 0.5% (w/v) bromphenol blue, and 500 mM DTT followed by boiling. The total reaction mixture was loaded onto a 4 -20% gradient gel, separated by SDS-PAGE, and transferred to a polyvinylidene difluoride membrane for immunoblotting against ubiquitin.

Subcellular localization of UBE2E1, UBE2D1, TRIM21, and TRIM21␤
HeLa cells were chosen for the localization experiments based on their morphology with a large thin cytoplasm when cultured on microscopic slides. Cells were cultured on Nunc Lab-Tek II chamber slides (Thermo Scientific). For transfection, 500 ng of plasmid (pJRed-UBE2D1, pEGFP-TRIM21, or pEGFP-TRIM21␤) was used together with X-tremeGENE 9 reagent (Sigma-Aldrich, Merck). After 48 h, cells were washed with PBS before fixation with 4% paraformaldehyde for 10 min at 4°C. For immunostaining, cells were fixed with 4% paraformaldehyde for 10 min at 4°C and then permeabilized with 0.2% Triton X-100 followed by a blocking step with 5% fetal bovine serum in PBS for 30 min. 1 g/ml rabbit anti-human UBE2E1 (ab36980, Abcam) was used as primary antibody and incubated for 60 min. Bound antibodies were detected by Alexa Fluor 594 -conjugated donkey anti-rabbit antibodies in a 1:400 dilution (Molecular Probes).
Nuclei were counterstained with 4Ј,6-diamidino-2-phenylindole (Molecular Probes) in PBS for 2 min, and slides were mounted in Prolong Gold antifade mounting medium (Invitrogen) under a coverslip. Rinsing in PBS was performed two to four times between each step, and all steps but the fixation were performed at room temperature. A laser-scanning confocal microscope was used to assess and document the cells (63ϫ magnification).

Analytical gel filtration
Tricorn Superdex 75 10/300 was used to perform the analytical gel filtration in an Ä KTA purifier using a standard lowmolecular-weight calibration kit (GE Healthcare) to calculate void volume by blue dextran and molecular weight calibration curve from the standard proteins therein. TRIM21 and UBE2E1 constructs were analyzed in a concentration range of 50 -400 M with an injected volume of 100 l and a flow rate of 0.5 ml/min.

TRIM21-UBE2E1 complex offers new insights in ubiquitination
the Sparky program (T. D. Goddard and D. G. Kneller, SPARKY 3, University of California, San Francisco). The backbone assignment was manually performed, assisted by the COM-PASS software (64).
Titrations experiments were carried out in a buffer containing 50 mM Tris-HCl, pH 8.0, 100 mM NaCl, 10 mM ␤-mercaptoethanol, 500 M ZnCl 2 , and 90% H 2 0, 10% 2 H 2 0 (v/v). CSP data were collected at 30°C in 1 H, 15  To identify significant CSPs, a cutoff of two standard deviations from the trimmed mean was calculated in an iterative procedure as described (65).
K d values were calculated by a nonlinear least-squares analysis using the following equation, where [P] T and [L] T are the total protein (NMR labeled) and ligand (unlabeled) concentrations at each aliquot, ⌬␦Ј is the change in peak position with each aliquot, and ⌬␦Ј max is the change in shifts between apo and fully bound states of the protein, P. K d values were only calculated for residues that show significant chemical shift perturbations upon TRIM21 binding and have signal intensities above the noise level. The dissociation constant of UBE2E1 C -TRIM21 R binding is an average over values obtained from fitting titrations on a perresidue basis for residues in UBE2E1 C -H1, -L4, and -L7. K d values obtained for residues in UBE2E1 C -H2 were averaged separately, as CSPs observed for this region likely originate from allosteric effects.

Crystallization
Purified E2, UBE2E1 C , and TRIM21 R were mixed in 1:1.2 and 1:2 ratios, incubated overnight, and then concentrated to 35 mg ml Ϫ1 . Initial crystal hits were optimized in both sitting-drop and hanging-drop vapor diffusion at 4°C with a reservoir solution containing 100 mM Bicine, pH 9.0, and 5% (w/v) PEG 6000. Final crystals were obtained in the above-described reservoir conditions with 12.5% (v/v) glycerol and 5% (v/v) ethylene glycol and flash frozen in liquid nitrogen. Crystals belong to C2 space group with cell dimensions of a ϭ 103.811 Å, b ϭ 95.834 Å, c ϭ 235.043 Å, ␣ ϭ ␥ ϭ 90.0°, and ␤ ϭ 93.15°with a solvent content of 54%. For the free E2 structure, UBE2E1 C was concentrated to 18 mg ml Ϫ1 and crystals were optimized in sitting-drop vapor diffusion at 4°C. The initial crystals were obtained in 0.1 M sodium citrate, pH 6, and 8% (w/v) PEG 8000 at 4°C. Final single crystals were obtained from the hanging-drop method in the same reservoir condition with added 10% (v/v) glycerol and flash frozen in liquid nitrogen.

Crystallography structure determination
Diffraction data for UBE2E1 C -TRIM21 R crystals were collected at BL14.1 beamline at BESSY Synchrotron (Berlin, Germany) and screened for TRIM21 R presence by testing for diffraction at the Zn 2ϩ -absorption peak wavelength. The protein complex structure was solved by the three-wavelength multiple anomalous dispersion method using the anomalous signal from the two Zn 2ϩ atoms in TRIM21 R . The location of Zn 2ϩ atoms and initial density modification were performed using SHELX (66) and its graphical user interface HKL2MAP (67) with a SHELXE-estimated mean figure of merit of 0.642 and pseudofree correlation coefficient of 69%. For the structure refinement, we used the inflection point data set merged with "Fridel pairs ϭ true" (2.82 Å) instead of "Fridel pairs ϭ false" (3.1 Å) during the multiple anomalous dispersion phasing method. Our first model was built using the CCP4 (68) software Buccaneer (69) and completed by manual model building in Coot (70). The molecules in the asymmetric unit were initially refined with local noncrystallographic symmetry (NCS) restraints in BUSTER and later with Phenix_Rosetta (71) that does not use NCS but improved the local geometry as judged by MolProbity. For final refinement, we uploaded the unmerged XDS_ASCII.HKL inflection point data set with the STARANISO web server (http://staraniso.global phasing.org/cgi-bin/staraniso.cgi) 7 (80) that performs an elliptical resolution cutoff for anisotropically diffracting crystals. Despite having a few diffraction spots to 2.57-Å resolution in the best-diffracting direction, we decided to remove the data in the 2.82-2.57-Å interval because spherical completeness was only 15% in that interval. After removing that interval, the spherical/elliptical completeness was 50.7/66.5% in the highest-resolution shell (2.91-2.82 Å), and overall the spherical/elliptical completeness was 91.2/93.6% in the 47.9 -2.82-Å interval ( Table 1). The final model was generated using local NCS restraints and jelly-body refinement in REFMAC5 (72) with 96.6/3.2/0.2% of the amino acids in the preferred/allowed/disallowed regions of the Ramachandran plot. UBE2E1 C crystals were produced from the same material as in the UBE2E1 C -TRIM21 R crystals, and data were recorded at the same beam time. Diffraction data were collected at BL14.1 beamline at BESSY Synchrotron. The structure was determined by molecular replacement in MOLREP using PDB code 3BZH (38) as a search model followed by manual model building in Coot and refinement in REFMAC5. All data collection and refinement statistics are summarized in Table 1.

Model building and structural presentations
The model of UBE2E1-Ub-TRIM21 R was generated by superimposing UBE2E1 C (module F) in the UBE2E1 C -TRIM21 R structure (PDB code 6FGA) onto the E2 module of the TRIM25-UBE2D1ϳUb ternary complex (PDB code 5FER; 7 Please note that the JBC is not responsible for the long-term archiving and maintenance of this site or any other third party-hosted site.

TRIM21-UBE2E1 complex offers new insights in ubiquitination
C␣ r.m.s.d., 0.54 Å). The resulting UBE2E1-Ub conjugate in which Ub is in a closed conformation shows essentially no clashes with UBE2E1 C or TRIM21 homodimers in the 6FGA structure, supporting its relevance in a ternary Ub-E2-E3substrate complex. All figures were generated using PyMOL Molecular Graphics System, Version 1.2r3pre (Schrödinger LLC).

Structural interface analysis
The web server VADAR (Volume Area Dihedral Angle Reporter) (73) was used for structure evaluation, including hydrogen-bonding partners and accessible surface area for both TRIM21 R (this study; PDB code 6FGA) and TRIM25 (PDB code 5FER) dimer analyses. Side chains were considered buried if their level of exposure was less than 20%.

SAXS sample preparation, data acquisition, analysis, and modeling
SAXS data were acquired for TRIM21 R using the ESRF BM29 SAXS beamline (74,75) with a robotic sample changer (76) and a Pilatus 1M detector (Dectris). SAXS data were also acquired using the Anton Paar SAXSess at Linköping University for TRIM21 R . SAXS samples were prepared by extensive dialysis against their buffer: 50 mM Tris-HCl, pH 8.0, 150 mM NaCl, 10 mM ␤-mercaptoethanol, and 500 M ZnCl 2 . For SAXSess measurements only, 10% glycerol was added to the buffer solution. Exact solvent blanks for all measurements were obtained from the last dialysis step. Table S3 provides the SAXS data acquisition parameters, sample parameters, and software used for data reduction to I(q) versus q, where I(q) ϭ 4sin/, 2 is the angle between the incident and scattered X-rays, and is their wavelength), analysis, and interpretation. All data were placed on an absolute scale using the scattering from pure H 2 O (SAXSess) or incident beam flux (BM29). I(q) versus q for the protein was obtained by subtraction of the solvent scattering from that of protein ϩ solvent. For the BM29 data, solvent measurements taken immediately before and after the protein ϩ solvent measurement were averaged to optimize solvent subtraction. As there was no discernible concentration dependence to I(q) for TRIM21 R , SAXSess data from the highest concentration samples were averaged to improve signal to noise. Molecular weights for the proteins were estimated using the method of Orthaber and Glatter (77). Values for contrast and partial specific volumes were determined using the MULCh program (78) with the known chemical compositions of samples and solvent. SAXS data analysis and modeling were performed using the tools of the ATSAS program package (79). The online interface and software used are listed in Table S3. Default software parameters were employed unless otherwise specified. The ESRF and SAXSess data overlay well (Fig. S3D), and a Kratky (Fig. S3E) plot of the ESRF data shows the expected bell shape for a globular, mostly folded protein with a rising profile at high q-values, indicating some degree of flexibility in the structure. The Guinier results for the ESRF and SAXSess data were the same within error (Table S4), but the SAXSess data were measured to lower minimum q-values and hence are more reliable for d max determination in P(r) calculations compared with the ESRF data. Indeed, the latter consistently showed a similar shape to that obtained with the SAXSess data with a weak tail to longer r values of indeterminate length that increased the apparent R g values. Furthermore, the molecular weight values were slightly more consistent with full RING dimerisation for the higher concentration SAXSess data. Therefore, the DAMMIF models in Fig. 1B were obtained using the SAXSess data-derived P(r) (Fig. S3F) and represent the averaged and filtered models from 20 independent DAMMIF calculations. Normalized spatial discrepancy values were 0.735 (assuming P1, i.e. no specific symmetry) or 0.496 (assuming P2 symmetry), indicating similar structural solutions for all calculations.