Structure of UBE2K–Ub/E3/polyUb reveals mechanisms of K48-linked Ub chain extension

Ubiquitin (Ub) chain types govern distinct biological processes. K48-linked polyUb chains target substrates for proteasomal degradation, but the mechanism of Ub chain synthesis remains elusive due to the transient nature of Ub handover. Here, we present the structure of a chemically trapped complex of the E2 UBE2K covalently linked to donor Ub and acceptor K48-linked di-Ub, primed for K48-linked Ub chain synthesis by a RING E3. The structure reveals the basis for acceptor Ub recognition by UBE2K active site residues and the C-terminal Ub-associated (UBA) domain, to impart K48-linked Ub specificity and catalysis. Furthermore, the structure unveils multiple Ub-binding surfaces on the UBA domain that allow distinct binding modes for K48- and K63-linked Ub chains. This multivalent Ub-binding feature serves to recruit UBE2K to ubiquitinated substrates to overcome weak acceptor Ub affinity and thereby promote chain elongation. These findings elucidate the mechanism of processive K48-linked polyUb chain formation by UBE2K.

This methodology is eminently transferable to other E2/E3 systems to allow such complexes to be trapped for structural studies. In this 2.4-Å resolution crystal structure, RNF38 primes UBE2K-Ub D , while the distal Ub of K48-linked di-Ub is accommodated between the UBA and UBC domains of UBE2K. This catalytic arrangement reveals a unique structural snapshot of K48-linked polyUb chain synthesis. We also elucidate a previously unknown role of the UBA domain in binding of diverse polyUb, thereby illustrating how UBE2K efficiently catalyzes polyubiquitination of substrates.

Results
RING E3s stimulate UBE2K. UBE2K is the only E2 that incorporates a UBA domain, a feature conserved across yeast (Ubc1) and human homologs ( Supplementary Fig. 1). The N-terminal region of UBE2K retains conserved residues found in E2s adapted to function with RING E3s, as highlighted by the UBE2D family (Extended Data Fig. 1a). To date, RING E3 binding and stimulation of UBE2K has been sparsely reported through limited functional assays, leaving the detailed mechanism as yet undetermined. This led us to explore which RING E3s function with UBE2K. Initially a panel of RING and U-box E3 ligases were reacted with UBE2K and wild-type Ub (Ub WT ) in a continuous turnover assay ( Fig. 1a and Extended Data Fig. 1b). Unlike the UBE2D family E2s, UBE2K primarily forms unanchored polyUb in the absence of substrate 5,8,16 . We therefore assessed the reaction products for the presence of K48-linked polyUb to examine whether the expected product was formed. Interestingly, not all RING E3s stimulated UBE2K and several reported to function with the UBE2D family produced K48-linked di-Ub with UBE2K. Comparing polyUb production by UBE2K in the absence or presence of RNF38 and BIRC2 demonstrated that formation of tri-Ub was also enhanced (Extended Data Fig. 1b). Taken together these assays suggest that several RING E3s function with UBE2K to stimulate transfer of Ub D to acceptor Ub. regardless of the methodology used to form the complex, Ub preferentially occupies the canonical Ub-binding site on the UBA domain 17   and mutagenesis analysis provide the only hint of how Ub is accommodated in UBE2K's acceptor site 17 . Thus it remains an open question as to how UBE2K initiates and extends K48-linked Ub chains. We reasoned that the questions pertaining to stimulation by RING E3 and accommodation of Ub as the acceptor could be combined and addressed within a single complex. This led us to load a Ub D via an isopeptide bond to the active site C92K of UBE2K C92K . UBE2K D124 juxtaposes the active site C92 and the corresponding Asp in UBE2I and UBE2N was shown to coordinate acceptor lysine and facilitate its deprotonation by lysine pK suppression 24,25 . We performed di-Ub formation at different pH values and found no shift in the pK a between UBE2K WT and UBE2K D124 variants, but UBE2K D124A and UBE2K D124R hindered di-Ub formation, and UBE2K D124N exhibited reduced activity at high pH ( Fig. 1b and Extended Data Fig. 2a-c). We posited that D124 has a minimal role in lysine pK suppression, but likely contributes important polar interactions with the acceptor lysine. Therefore D124 was selected as the acceptor cross-linking site and we chemically cross-linked D124C of Ub-C92K UBE2K D124C to K48C Ub via maleimide chemistry. However, this complex did not yield protein crystals with our RING E3 panel and therefore we cross-linked K48C Ub− 48 Ub to Ub-C92K UBE2K D124C ( Fig. 1c and Extended Data Fig. 2d,e) to increase the likelihood for crystallization. K97R was included to prevent nonproductive UBE2K-Ub formation. Following the convention of an established nomenclature system 26 , we designate this complex Ub-C92K UBE2K D124C−K48C Ub− 48 Ub.
Structure of the K48-Ub 2 -UBE2K-Ub/RNF38 complex. Several RING E3 domains that were active in our RING panel survey were combined with this UBE2K complex for cocrystallization, but crystals were only obtained with RNF38. The Ub-C92K UBE2K D124C− K48C Ub− 48 Ub/RNF38 crystals resulted in a 2.4-Å resolution structure detailed in Fig. 1d and Supplementary Table 1. In the structure, Ub D is primed for transfer by RNF38, while the active site region of UBE2K encloses the distal Ub of K48-Ub 2 (hereafter referred to as Ub A ), which presents as an 'open' conformation (Extended Data Fig. 2f,g). The UBE2K D124C-BMOE-K48C Ub linkage juxtaposes the Ub-C92K UBE2K active site (Fig. 1e). To better visualize the actual active site, we modeled the wild-type residues and refined the model including energy minimization. This placed the ε-amino group of K48 of Ub A within ~6.4 Å of the UBE2K active site C92, suggesting that the structure approximates the transition-state complex of K48-linked Ub chain synthesis (Fig. 1f). K48 of Ub A is also in close proximity (~3.6 Å) to D124 of UBE2K, suggesting a role for this residue in coordinating acceptor lysine. The proximal Ub of K48-Ub 2 approaches the E64-T65 region of UBE2K, but does not make any contact; hence we focus the discussion on Ub A . Donor Ub is primed by RNF38 through a conserved network. In our complex Ub D is primed in a catalytically competent conformation by RNF38 with root mean square deviations (r.m.s.d.) across Cα atoms between 0.75 and 1.17 Å compared to UBE2D2-Ub/ RNF38 and other UBE2D-Ub/RING E3 complexes 25,27,28 ( Fig. 2a and Extended Data Fig. 3a,b). Furthermore, the position of the critical 'linchpin-Arg' R454 in RNF38 is superimposable between our structure and the UBE2D2−Ub/RNF38 complex 28 , suggesting this mechanism is conserved across E2s. To validate the importance of E2/Ub D /RING contacts, we made a series of mutations to UBE2K, Ub D and RNF38 and assessed their performance in a di-Ub formation assay. We designed UBE2K F62A,A102R and RNF38 M417A to disrupt the RNF38/UBE2K interaction and found both mutants impaired di-Ub formation in a RING-dependent manner (Fig. 2b,c and Extended Data Fig. 3c). The importance of RNF38's linchpin arginine was validated using the R454A mutant, which impaired activity (Fig. 2c). Ub D mutants I36A and Q40R, which were designed to disrupt the Ub D /RNF38 interface, were both defective compared to Ub WT in our assays (Fig. 2d). UBE2K α2-helix residues are involved in forming an interface with Ub D , and mutation of A115 to Arg within the α2-helix prevented di-Ub formation (Fig. 2e). D94 of UBE2K contributes to priming of the C-terminal tail of Ub D and D94A substitution eliminated di-Ub formation (Fig. 2e). UBE2K D94A and UBE2K A115R also exhibited a defect in di-Ub formation in the absence of E3 (Extended Data Fig. 3d), highlighting the importance of the primed UBE2K-Ub D conformation in both E3-dependent and E3-independent reactions.
UBC and UBA domains cooperate to stabilize acceptor Ub. Ub A is positioned between the UBC and UBA domains of UBE2K ( Fig.  3a and Extended Data Fig. 4a). The active site cross-over loop (residues 85-90) and the N-terminal region of α3-helix (residues 124-129) from the UBC domain form a network of contacts with the short 3 10 -helix region (residues 54-60) of Ub A . R54 of Ub A forms hydrogen bonds with carbonyl oxygens of P125 and Q126 of UBE2K. Within Ub A the guanidinium group of R54 is stabilized by π stacking with Y59, a feature not observed in other structures of Ub ( Fig. 3b and Extended Data Fig. 4b). The side chain and carbonyl oxygen of D58 of Ub A form hydrogen bonds with the side chain of S85 and amide of A128 of UBE2K, respectively (Fig. 3b). N60 of Ub A packs against A128 and N132 of UBE2K. A small hydrophobic contact is seen between V87 and T88 of UBE2K and T55 of Ub A (Fig. 3b). Additionally Q62 of Ub A contacts K186 in the UBA domain, while K63 of Ub A abuts L196 region in the UBA (Fig. 3c). Near the active site, E51 of Ub A is approaching hydrogen bond distance with K97 of UBE2K (Fig. 3b). Mutation of these key interface residues in UBE2K (S85L, T88D, D124R, K186A or L196A) or Ub A (E51A, R54G, D58A, Y59L, N60A, Q62A or K63A) reduced or abolished di-Ub formation (Fig. 3d,e). These results are consistent with prior mutagenesis studies, where docking methods were used to predict UBE2K/Ub A interactions 17 . The overall position of Ub A in the docked model is similar to the one observed in our structure. However, our structure reveals detailed molecular interactions and a role of the UBA domain in stabilizing Ub A that was not reported previously. Mutation of Ub A residues remote from the acceptor site (Ub L8A,I44A,V70A ) did not inhibit di-Ub formation (Fig. 3d). This observation suggests that Ub's canonical I44 patch plays a minimal role at the acceptor site of UBE2K.
UBE2K has a weak affinity for acceptor Ub. In multiple structures and studies, the UBA domain of UBE2K has been shown to bind Ub via the M172/L198 surface 17,23,29 . The nature of this interaction has been unclear since it is distant from the active site and therefore unlikely to contribute directly to Ub A recognition. This UBA surface has a strong affinity for Ub, making it difficult to assess Ub binding in the acceptor site. In an effort to measure Ub A -binding affinity, we generated UBE2K M172D,L198R to disrupt this canonical UBA/Ub interaction. Surface plasmon resonance (SPR) and NMR analysis demonstrated that UBE2K M172D,L198R had no detectable binding of Ub, whereas UBE2K T88D , a mutant defective in Ub A recognition (Fig. 3b,e), exhibited near-identical affinity for Ub as UBE2K WT (Extended Data Fig. 4c-h and Supplementary Fig.  3). Furthermore, the 1 H-15 N HSQC spectra of 15 N-Ub titrated with UBE2K WT and UBE2K T88D are nearly identical, while there were no chemical shift perturbations (CSPs) in 15 N-Ub upon titrating UBE2K M172D,L198R . These data support that Ub A binding is weak and the M172/L198 surface accounts for detectable Ub binding by the UBA domain. Kinetic analysis of di-Ub formation catalyzed by UBE2K WT and UBE2K M172D,L198R showed similar kinetics, suggesting that the M172/L198 Ub-binding site on the UBA domain has a limited role in Ub A recognition and catalysis ( Fig. 3f and Supplementary Fig. 4). The Michaelis-Menten curves were not saturated at concentrations of 1.2 mM Ub A , supporting a weak Ub A -binding affinity.
NMR approach to validating UBE2K acceptor Ub site. To validate the observed UBE2K/Ub A interactions we chemically cross-linked K48 of 15 N-Ub K48C directly to the active site C92 of UBE2K to overcome the weak binding affinity of the acceptor site ( Fig. 4a and Extended Data Fig. 5a,b). Cross-linking to UBE2K WT still allows for canonical binding of Ub to the UBA domain, potentially leading to ambiguous signals in NMR. Therefore, M172D and L198R were incorporated such that CSPs could be unambiguously attributed to Ub A binding. We generated UBE2K M172D,L198R,C92-K48C Ub( 15 N) and UBE2K ΔUBA,C92-K48C Ub( 15 N) and acquired 15 N-1 H HSQC spectra (Fig. 4b,c). As expected CSPs and/or signal attenuation of the signals originating from residues around K48 and residues that contact the UBC in our structure were observed for both conjugates (Fig. 4d,e). To support the suitability of our cross-linking strategy via UBE2K D124C and Ub K48C in our structure, we also generated UBE2K M172D,L198R,D124C-K48C Ub( 15 N) and found a similar 15 N-1 H HSQC spectrum and pattern of CSPs and signal attenuations (Extended Data Fig. 5c-f). Comparison of the spectra showed that the signal for K63 of Ub A was attenuated in the presence of UBA, but not with UBE2K ΔUBA , corroborating our trapped mimic structure (Extended Data Fig. 5g). We attribute the reduced attenuation of the signals observed for 15 N-Ub A in UBE2K ΔUBA,C92-K48C Ub( 15 N) to enhanced conformational freedom of Ub A in the absence of the UBA domain (Fig. 4d,e). Consistent with this notion, UBE2K K186A has reduced catalytic efficiency for di-Ub formation ( Fig. 3f and Supplementary  Fig. 4) supporting UBA's role in positioning Ub A for catalysis. Together these results demonstrate that the UBC active site region of UBE2K functions to orient Ub A to impart K48-Ub chain specificity, while the UBA domain serves to stabilize Ub A to enhance catalysis.
UBE2K is recruited to Ub-primed substrates through UBA. Analysis of copies of our structure from adjacent asymmetric units (ASUs) showed a potential interaction between K48-Ub 2 from one ASU binding in trans to the UBE2K UBA domain of another ASU, distal from the Ub A -binding site (Fig. 5a). Both Ub units of this neighboring K48-Ub 2 are in close contact with the same UBA, so we investigated the role of both binding sites. The canonical Ub-binding surface centered on M172 and L198 of UBE2K contacts the I44 patch of the proximal Ub from the neighboring ASU.   This interface represents a conserved binding mode between UBE2K and Ub, which can be observed in other structures with a varying degree of rotation between the two proteins (Extended Data Fig. 6). The neighboring distal Ub binds to UBE2K's UBA α5-helix and the preceding loop (residues E167 and R176) via its I36 patch, an interaction that has not previously been reported. This resulting 'open' conformation of di-Ub is unique among structures of K48-linked polyUb ( Fig. 5a and Extended Data Fig. 2f,g).
The role of UBA M172,L198 /Ub interaction in UBE2K-catalyzed polyUb formation remains unclear. We showed that UBE2K M172D,L198R had no defect in di-Ub synthesis when free unanchored Ub was used as the substrate (Fig. 3f). We wondered whether this Ub-binding surface of the UBA might serve to recruit UBE2K to ubiquitinated substrates to facilitate chain elongation. We generated a ubiquitinated substrate, SMAC-Ub 1-6 , using a combination of UBE2D2 and BIRC2 that allowed isolation of SMAC-Ub 1-2 , SMAC-Ub 3-5 and SMAC-Ub 5+ (Extended Data Fig. 7a-c). With this we could assess how the Ub-binding property of UBE2K's UBA domain mediated recruitment to our model substrate, SMAC-Ub n . At a concentration of 10 μM, monomeric Ub A , UBE2K WT and UBE2K M172D,L198R exhibited minimal di-Ub formation ( Fig. 3f and Supplementary Fig. 4). In contrast, at a 10-μM SMAC-Ub n concentration, UBE2K WT showed enhanced ubiquitination of SMAC modified with a minimum of three Ub molecules (Fig. 5b,c). It seems plausible that an increased number of Ub moieties on SMAC might provide more available Ub A for elongation, but UBE2K M172D,L198R , which was as active as UBE2K WT in di-Ub synthesis (Fig. 3f), showed no enhancement in ubiquitination of SMAC-Ub 3-5 and only a marginal increase in ubiquitination with SMAC-Ub 5+ ( Fig. 5c and Extended Data Fig. 7d). These findings suggest that the UBA M172,L198 -Ub interaction accounts for the bulk of enhanced ubiquitination in the context of Ub-primed substrates. Hereafter we selected SMAC-Ub 3-5 for further analysis. Mass spectrometry analysis revealed that several monoubiquitination sites, as well as a variety of Ub linkage types (K11, K27, K33, K48, K63) were present, consistent with reported properties of UBE2D2 (refs. 3,30,31 ) (Extended Data Fig. 7e,f). UBE2K WT catalyzed the transfer of Ub to 10 μM of SMAC-Ub 3-5 , whereas UBE2K M172D,L198R , UBE2K ΔUBA and UBE2D2 were defective (Extended Data Fig. 7g). Given that BIRC2 stimulates UBE2K (Extended Data Fig. 1b) and binds SMAC, we performed UBE2K-mediated SMAC-Ub 3-5 ubiquitination in the absence and presence of near-full-length BIRC2 147-C . Addition of BIRC2 147-C resulted in more Ub transfer to SMAC-Ub 3-5 by UBE2K WT and partially restored the ability of UBE2K M172D,L198R to transfer Ub to SMAC-Ub 3-5 (Fig. 5d). We introduced UBE2K E167A,C170R,R176A to address the secondary Ub-binding site in the UBA. UBE2K E167A,C170R,R176A was as competent in di-Ub synthesis as UBE2K WT (Extended Data Fig. 7h and Supplementary Fig. 5), but was defective in ubiquitination   of SMAC-Ub 3-5 compared to UBE2K WT (Fig. 5e,f). Comparing these UBE2K variants revealed Ub transfer to SMAC-Ub 3-5 by UBE2K M172D,L198R was reduced to ~25% of UBE2K WT , while UBE2K E167A,C170R,R176A was reduced to ~66% (Fig. 5f). SMAC-Ub 3-5 generated via UBE2D2 contained heterotypic Ub chain types.
To eliminate unknown contributions from such Ub chains, we treated polyubiquitinated SMAC with vOTU 32 to obtain monoubiquitinated SMAC-Ub 1-3 and then mixed with UBE2R1 to generate SMAC-K48Ub n and isolated SMAC-K48Ub 3-6 for analysis (Extended Data Fig. 8a-c). Similar to SMAC-Ub 3-5 , UBE2K M172D,L198R and UBE2K E167A,C170R,R176A were defective in ubiquitinating SMAC-K48Ub 3-6 compared to UBE2K WT (Extended Data Fig. 8d,e). These results demonstrated that both the canonical and newly identified secondary Ub-binding sites on the UBA domain Ub contribute to the recruitment of UBE2K to ubiquitinated substrates to facilitate Ub chain elongation.

The UBA domain selects Ub-Ub linkage types on the acceptor.
Closer inspection of the M172/L198 Ub-binding site reveals that a plausible isopeptide linkage can be modeled between K63 of the acceptor Ub and the C terminus of a symmetry-related Ub from the M172/L198 binding site (Fig. 6a). This suggests a potential mode for binding K63-linked polyUb acceptor, which is consistent with the recently proposed model for yeast homolog Ubc1 (ref. 33 ). We examined the preference of UBE2K WT and UBE2K M172D,L198R for different Ub acceptors. In a single time point reaction, both UBE2K variants appeared equally competent in Ub transfer when the acceptor was mono-Ub, K48-Ub 2 or K48-Ub 4 (Fig. 6b). Kinetic analysis of K48-linked tri-Ub formation showed similar catalytic efficiency by both UBE2K variants ( Fig. 6c and Supplementary Fig. 6a,b), suggesting that Ub/UBA binding is not a major factor for K48-linked polyUb acceptors. Similar to the situation with mono-Ub as the acceptor (Fig. 3f), the Michaelis-Menten curves could not be saturated, consistent with weak acceptor Ub-binding affinity. In contrast, both UBE2K variants showed enhanced Ub transfer with K63-Ub 2 acceptor compared to K48-Ub 2 acceptor (Fig. 6d). Kinetic analysis of tri-Ub formation using K63-Ub 2 acceptor showed that UBE2K WT exhibited a 20-fold higher catalytic efficiency for K63-Ub 2 (k cat /K m = 25,070 M −1 min −1 ; Fig. 6e and Supplementary Fig. 6c-e) compared to K48-Ub 2 (estimated k cat /K m of 1,230 M −1 min −1 ; Fig.  6c). This primarily results from an improved K m , since k cat is lower for K63-Ub 2 compared to K48-Ub 2 at higher acceptor Ub concentration. UBE2K M172D,L198R displayed reduced k cat and K m for K63-Ub 2 compared to UBE2K WT (Fig. 6e). The proximity of the extreme C-terminal residues E195, S199 and N200 of UBE2K to the potential K63 linkage led us to generate UBE2K E195R,S199A,N200A to examine its effect (Fig. 6a). UBE2K E195R,S199A,N200A exhibited similar activity as UBE2K WT with mono-Ub or K48-Ub 2 as acceptor, but was impaired with K63-Ub 2 (Fig. 6f,g and Extended Data Fig. 9). Thus our structure provides a probable mechanism describing how K63-linked polyUb acceptor is recognized by UBE2K. Together this suggests that the canonical Ub-binding surface of UBA and the UBE2K active site could cooperate to bind two consecutive Ub moieties in K63-linked polyUb to enhance catalysis. This mechanism seems to be unique for K63-linked polyUb, as the conformations of other di-Ub linkages preclude this binding mode.

Discussion
Our chemical biology approach allowed us to capture a transient state of UBE2K in complex with RING E3, Ub D and acceptor K48-Ub 2 . Our structure and functional assays reveal multiple roles for the UBA domain, provide an explanation for how RING E3s stimulate UBE2K and demonstrate how acceptor Ub is recognized with its K48 poised for catalysis. We showed that the UBC active site region defines the selectivity for K48-linked Ub chain synthesis, while the UBA domain contributes to processive polyUb formation by orienting Ub A at the active site and plays a role in recruitment to Ub-primed substrates. This is consistent with UBE2K being an elongating E2, suited to accommodate Ub rather than unmodified proteins, and reinforces the notion that diversity in regions outside the UBC domain of E2s are employed to control the Ub linkages they form.
Ligation of Ub is a transient event that necessitates chemical cross-linking to capture the structure of the transition-state complex. Previous structural studies have used three-way chemical cross-linkers to connect the active site of E2 to Ub D and a lysine of the substrate, or a reactive cysteine on the E3 (refs. 34,35 ). Recently, the UBE2I-SUMO mimetic was engineered using a residue adjacent to the active site cysteine such that the catalytic cysteine of UBE2I is available for cross-linking with a lysine site (substituted as cysteine) on PCNA 36 . A similar strategy was adopted to generate a yeast CDC34 A141K -Ub mimetic to allow chemical cross-linking of the active site cysteines of CDC34 and Uba1 (ref. 37 ). We initially adopted a similar approach but found that the UBE2K-Ub mimetic could not assume the active closed conformation. Thus we preserved the established E2-Ub amide linkage mimetic and used an active site Asp, known to engage with substrate lysine 24,25 , for cross-linking with Ub A . This method should be applicable for trapping other E2-Ub/E3/Ub A complexes for structural analysis.
There is paucity in understanding how Ub A is recognized by E2 and/or E3 to enable linkage-specific polyUb chain assembly. K63-linked and M1-linked Ub chains syntheses have been previously captured in the structures of UBE2N~Ub/UEV1 and HOIP/Ub complexes, respectively 38-41 . Both complexes showed that Ub-binding domains, such as UEV1 and a zinc finger region embedded within the catalytic domain of HOIP, are responsible for orienting Ub A to confer linkage specificity. In contrast, UBE2K primarily employs residues surrounding the active site within the UBC domain to orient Ub A , while the UBA domain provides extra support in stabilizing Ub A to enhance catalysis. This is consistent with prior observations that UBE2K ΔUBA is sufficient to synthesize K48-linked Ub chains, but has a decreased rate of polyUb formation 29,42,43 . The 3 10 -helix region of Ub A forms a complementary interface with the UBE2K active site. This Ub surface was previously shown to bind Rabex-5 A20 zinc finger 44 . While structural modeling of other E2s, such as UBE2R1 and UBE2D family E2s known to assemble K48-linked Ub chains, onto our structure reveals a similar complementary interface, future studies will be required to address whether other E2s utilize a similar Ub A -binding mode. Although our structure includes K48-Ub 2 as the acceptor, we did not observed any interaction involving the proximal Ub. Our kinetic analysis of UBE2K WT -catalyzed di-Ub and tri-Ub formation using mono-Ub and K48-Ub 2 as the acceptor, respectively, showed that tri-Ub formation was marginally faster (Figs. 3f and 6c; ~1.5-fold). Since UBE2K WT and UBE2K M172D,L198R exhibited similar kinetics in these reactions, this marginal increase in rate is unlikely to be due to this UBA/Ub interaction. The relevance of proximal Ub molecules in polyUb chain elongation is a poorly understood area and warrants future investigation.
In addition to the canonical Ub-binding surface, the UBA domain harbors two weak Ub-binding surfaces, which we refer to as secondary and acceptor Ub-binding sites. The role of the canonical UBA/Ub interaction in UBE2K-catalyzed polyUb chain elongation remains elusive. We showed that UBE2K has weak affinity for Ub A and this probably helps to ensure that UBE2K does not assemble free unanchored Ub chains at cellular Ub concentrations. UBE2K WT and UBE2K M172D,L198R exhibited similar kinetics and activity in K48-linked di-Ub, tri-Ub and penta-Ub formation, suggesting that the canonical UBA/Ub interaction plays little role in unanchored K48-polyUb chain elongation. However, when substrate was primed with three or more Ub molecules on different sites that contain either K48-linked or heterogeneous Ub chains, both the canonical and secondary Ub-binding sites on the UBA domain promote Ub chain elongation on substrate. Our data suggest a model whereby these Ub-binding surfaces of the UBA domain provide avidity for Ub-primed substrates. By localizing UBE2K to Ub-primed substrates, where Ub concentration is enriched, the weak acceptor Ub affinity could be overcome. The presence of an E3 would further reduce the Ub concentration barrier by bridging UBE2K and Ub-primed substrate. We envisage that the spatial arrangement of Ub on the Ub-primed substrate probably matters, where different chain types and lengths emanating from different ubiquitination sites could influence how Ub binds the canonical, secondary and acceptor Ub-binding sites on UBA. If the arrangement permits multisite binding, the UBE2K active site could engage Ub A to enhance catalysis. The presence of K63-linked Ub chain on substrate could greatly enhance catalysis by improving the K m , in agreement with a recent finding for the yeast homolog Ubc1 (ref. 33 ). Furthermore, our structural data showed that the secondary Ub-binding site could potentially support a K48-linked Ub chain. By recognizing a diverse range of Ub conjugates and orienting Ub A , the UBA domain is an essential feature that enables K48-linked   Methods DNA construct design. UBE2K and related variants were expressed as 6×His-SUMO fusions in pRSF-DUET1. All Ub mutants and Ub lacking C-terminal diglycines (Ub74) used in assays were expressed with an N-terminal 6×His-TEV site in pRSF-DUET1. Untagged Ub WT , Ub K48C , Ub C-terminal 6×His (Ub(6×His)) and Ub K48C (6×His) were expressed in pET3a and purified as described 26,45 . Human E1 (UBA1) in pET21d was a kind gift from Professor Wolberger (Addgene, plasmid 34965) and subcloned into pRSF-DUET1 with a C-terminal GGS-6×His tag. Full-length RNF138 and RING domain only MDM2 419-C were expressed in His-Zbasic-CL7-SUMO vectors for use in the CL7/Im7 purification system 46 . Phospho-Y371 c-CBL was obtained from an established protocol 47 . UBE4B was obtained as a His-SUMO fusion following ref. 48 . The GST-TEV-RNF2(1-114) and His-TEV-BMI1(1-109) constructs were coexpressed to obtain the heterodimer. Full-length SIAH1 and CHIP were expressed as a His-MBP fusion and His-TEV fusion, respectively. Other RING E3s used, BIRC2 147-C , BIRC2 556-C , BIRC4 434-C , RNF114, RNF8 345-C , RNF38 389-C and DTX3L 544-C and were expressed as GST-TEV fusions in pGEX4T-1. For removal of C-terminal 6×His on Ub, YUH1 was expressed according to ref. 26 . His-ULP1 was expressed in pET28b. SMAC 56-C,V81D was expressed as a SUMO fusion in pRSF-DUET1.
Protein expression and purification. All proteins were expressed in BL-21 Rosetta 2 (DE3) chemically competent Escherichia coli (Novagen). LB cultures were supplemented with 1 mM MgSO 4 and grown at 37 °C until optical density (OD 600 ) of 0.8, at which the temperature was reduced to 16 °C and cultures were induced with 0.3 mM IPTG for ~16 hours. Cultures of RING E3s were supplemented with 400 μM ZnSO 4 30 min before induction. Cells were collected at 3,500g and resuspended in IMAC buffer A (20 mM Tris-HCl, 500 mM NaCl, 1 mM TCEP, pH 7.4), lysed at 12,000 psi using a microfluidizer, cleared at 18,000g for 45 min and the supernatant was passed through a 0.45-μm syringe filter before loading on an affinity column. His-SUMO, His-MBP or His-TEV proteins were loaded on a 5-ml His-Trap column (GE Life Sciences), washed for 30 column volume (CV) in IMAC buffer A and eluted in IMAC buffer B (20 mM Tris-HCl, 500 mM NaCl, 300 mM imidazole, 1 mM TCEP, pH 7.4). GST-tagged proteins were loaded on a 5-ml GST trap column (GE Life Sciences), washed for 20 CV and eluted in GST elution buffer (20 mM Tris-HCl, 500 mM NaCl, 1 mM TCEP, 20 mM reduced glutathione, pH 8.0). Elutions were dialyzed against 20 mM Tris-HCl, 200 mM NaCl, 0.5 mM TCEP, pH 8.0 overnight at 4 °C. SUMO constructs were cleaved using 5 μM ULP1 and TEV cleavage carried out at 1:50 (mg/mg) ratio of TEV protease/protein. Protease was removed using a 15-ml His-Trap column before a final size-exclusion step on a HiLoad 26/600 Superdex 75 column (GE Life Sciences) in SEC buffer (20 mM Tris-HCl, 200 mM NaCl, 0.5 mM TCEP, pH 7.5). CL7 fusions were loaded on an Im7 column (TriAltus Bioscience), washed in 30 CV in CL7 buffer (20 mM Tris-HCl, 700 mM NaCl, 1 mM TCEP, pH 7.5) and cleaved on-column by injecting 1 mg of ULP1 per ml of Im7 resin allowing for a one-hour incubation. Proteins were directly loaded onto a HiLoad 26/600 Superdex 75 column from the Im7 column in SEC buffer and ULP1 was removed by addition of Ni-NTA. Human E1 was subjected to anion exchange and a final size-exclusion step on a HiLoad 26/600 Superdex 200 column (GE Life Sciences) in SEC buffer. 15 N-Ub K48C (6×His) was expressed with 15 N-ammonium chloride (Cambridge Isotope Laboratories) as the sole nitrogen source in autoinducing media according to ref. 45 .

UBE2K-Ub chemical cross-linking.
A 50 μM portion of UBE2K C92K,K97R,D124C,C170S was loaded with 1 mM His-TEV-Ub WT and 3 μM E1 in 50 mM Tris-HCl, 20 mM ATP, 20 mM MgCl 2 , 1 mM TCEP, pH 9.5 buffer at 30 °C for 20 hours. The His-TEV-Ub-C92K UBE2K D124C product was isolated in a Ni-NTA step, cleaved with TEV and the protease was removed using another Ni-NTA step. The Ub-C92K UBE2K D124C intermediate was further purified using a HiLoad 26/600 Superdex 75 column in 20 mM Tris-HCl, 200 mM NaCl, 5 mM TCEP, pH 7.5. In the next phase, 20 mM TCEP was added to K48C Ub− 48 Ub(6×His) before loading on a HiPrep 26/10 Desalting column (GE Life Sciences) in cross-linking buffer (25 mM sodium phosphate, 200 mM NaCl, 2 mM EDTA, pH 7.0). Fractions containing the fully reduced K48C Ub− 48 Ub(6×His) were collected and 15-fold molar excess of BMOE was immediately added. Ub-C92K UBE2K D124C was reduced in 20 mM TCEP and desalted, while BMOE was reacted K48C Ub− 48 Ub(6×His) desalted directly into reduced Ub-C92K UBE2K D124C and allowed to react at ambient temperature for 4 hours. This cross-linking reaction was diluted 10-fold in 25 mM Tris-HCl, 500 mM NaCl, 3 mM β-mercaptoethanol (BME), pH 7.5 buffer and loaded onto a 5-mL His-Trap column. After loading, the column was washed for 30 CV in IMAC buffer A, and eluted in 5 CV using IMAC buffer B. The elution was concentrated in a 10,000 MWCO centrifugal filter unit (Merck-Millipore) and loaded onto a HiLoad 26/600 Superdex 75 column in 20 mM Tris-HCl, 200 mM NaCl, pH 7.5. For NMR UBE2K C92−K48C Ub( 15 N) and UBE2K D124C−K48C Ub( 15 N) conjugates were created following the same procedure, but 25 mM sodium phosphate, 120 mM NaCl, pH 7.0 buffer was used in the final gel filtration step. Protein concentration was determined by Bio-Rad protein assay or by using molar extinction coefficient at 280 nm (ε 280 ) of 22,500 cm  Fig. 7a-c). To generate SMAC-K48Ub n , SMAC-Ub n was reacted with GST-vOTU 32 to remove residual polymeric Ub to generate multiple monoubiquitinated SMAC-Ub 1-3 and the DUB was removed by a glutathione-sepharose purification step. A 75 μM portion of primed monoubiquitinated SMAC-Ub 1-3 was reacted as above, except for the omission of BIRC2 and substitution of UBE2D2 for UBE2R1. SMAC-K48Ub n was obtained through the procedure above and defined lengths were isolated in the final size-exclusion step (Extended Data Fig. 8a-c).

Conjugation of near-infrared fluorescence labeled Ub.
Reduced GGSC-Ub K48R was exchanged into cross-linking buffer using a Zeba Spin desalting column (Thermo Fisher Scientific). IRDye 800CW Maleimide (LI-COR Biosciences) was resuspended in DMSO and added at a threefold molar excess to GGSC-Ub K48R and allowed to incubate at room temperature for 90 min. Labeling was quenched with the addition of 3 mM BME and excess dye was removed and buffer-exchanged to 25 mM Tris-HCl, 150 mM NaCl, pH-8.0 using a using a Zeba Spin desalting column. The final stock 200 μM of infrared-labeled GGSC-Ub K48R (Ub * ) was mixed with unlabeled Ub K48R in a 1:80 molar ratio to obtain a working stock.
Kinetic analysis of di-Ub and tri-Ub formation. Ub K48R was used to charge all UBE2K variants to prevent the labeled donor Ub from acting as the acceptor. 32 P-Ub K48R was prepared as described previously 47 . The charged reactions were performed in 50 mM Tris-HCl, pH 7.6, 50 mM NaCl, 10 mM ATP, 10 mM MgCl 2 , 1 mg ml −1 BSA, 3 μM UBA1, 9 μM UBE2K variant and 13.5 μM 32 P-Ub K48R at 37 °C for 30 min followed by equilibration at 22 °C for 10 min. A 6-μl portion of charged reaction was added to 12 μl of chased reaction containing the indicated final concentrations of Ub74 (Fig. 3f, Extended Data Fig. 7h and Supplementary Figs. 4 and 5), K48-linked di-Ub (Fig. 6c and Supplementary Fig.  6a,b) or K63-linked di-Ub (Fig. 6e and Supplementary Fig. 6c,d) at 22 °C and stopped at the indicated time (30 s or 1 min) by adding 6 μl of the reaction to 2× LDS loading buffer. Under these reaction conditions ~10-15% of UBE2K~3 2 P-Ub was consumed at the highest acceptor Ub concentration, and less than 10% at other concentrations. Thus, the reaction satisfied the initial rate requirements for kinetic analysis. The reactions were resolved by SDS-PAGE, dried, exposed to a phosphorimager, scanned with a Typhoon PLA 7000 (GE Healthcare) and quantified using ImageQuant (GE Healthcare). Known amounts of 32 P-Ub K48R were resolved by SDS-PAGE, dried and analyzed together with the reaction products. Di-Ub or tri-Ub products and charged UBE2K~3 2 P-Ub were quantified. Initial rate is expressed as moles of di-Ub or tri-Ub formed per minute per moles of UBE2K~3 2 P-Ub or per minute. All reported kinetic parameters were determined from initial rates by fitting two independent datasets to the Michaelis-Menten equation using Prism (GraphPad). In cases where Michaelis-Menten curves were not saturated, k cat /K m was estimated from the slope of the linear portion of the curve.
Continuous ubiquitination assays. RING stimulation of UBE2K (Fig. 1a and Extended Data Fig. 1d) was carried out in assay buffer (25 mM Tris-HCl, 50 mM NaCl, pH 8.0) with 15 mM ATP, 15 mM MgCl 2 , 1 μM UBA1, 200 μM Ub WT and 5 μM UBE2K at 37 °C. The concentration of E3s was kept at 5 μM. UBE2K D124 variants (Fig. 1b) were assessed under identical conditions. Reactions were stopped at indicated time points with LDS loading buffer and resolved by SDS-PAGE. Gels were transferred to nitrocellulose membrane (Bio-Rad) using the Trans-Blot Turbo system (Bio-Rad) for 10 min with constant 1.3 amps. Total protein was visualized with Ponceau-S, imaged using Bio-Rad ChemiDoc, destained and the membrane was blocked in 5% (w/v) BSA in TBST (20 mM Tris-HCl, 150 mM NaCl, 0.1%(w/v) Tween-20) for 30 min at ambient temperature. Primary antibodies were incubated at 4 °C for 12 hours using a 1:5,000 dilution of mouse anti-Ub (PD41, Santa Cruz Biotechnology) and 1:3,000 dilution of rabbit anti-K48-Ub (clone Apu2, Merck) in 2.5% (w/v) BSA in TBST. The membrane was washed three times for 5 min in TBST and incubated with secondary antibodies: IRDye 800CW goat anti-mouse IgG (LI-COR Biosciences) in a 1:15,000 dilution and IRDye 680RD goat anti-rabbit IgG (LI-COR Biosciences) in a 1:15,000 dilution, for 1 hour at ambient temperature. Two washes with TBST and a final in TBS were carried out before imaging on a LI-COR Odyssey CLx.
Precharged ubiquitination assays. Charging of UBE2K was carried out in assay buffer with 20 mM ATP, 20 mM MgCl 2 , 3 μM UBA1, 18 μM Ub * and 10 μM UBE2K for 30 min at 37 °C. UBE2K ΔUBA required 60 min for optimal charging. For unlabeled Ub D the same concentration was used. The charging reaction of UBE2K~Ub * was diluted 1:1 into the acceptor Ub conjugate to start each reaction. The final acceptor concentration of mono-Ub was 100 μM for Figs. 2c-e and 3d,e and Extended Data Figs. 3c,d and 9a. K48-Ub 2 as acceptor was used in a final concentration of 50 μM for Fig. 6b,d and Extended Data Fig. 9c, while K63-Ub 2 as acceptor was used at 40 μM for Fig. 6d,f. The final SMAC-Ub n concentration was 10 μM. The reactions were stopped with LDS loading buffer and resolved by SDS-PAGE. In-gel fluorescence was detected using an Odyssey CLx imaging system and analyzed with ImageStudio v.5 and the Fiji distribution of ImageJ 57 . Total protein was detected with InstantBlue (abcam) and Coomassie-stained gels were imaged using a LI-COR Odyssey CLx imaging system and a conventional scanner.
pH-dependent di-Ub formation assay. Buffers covering pH 6.5 to 10.5 were prepared in 50 mM Tris-HCl and150 mM NaCl at the desired pH. Charging of 30 μM UBE2K was carried out in 20 mM Tris-HCl, 100 mM NaCl, 20 mM ATP, 20 mM MgCl 2 , pH 8.0 buffer with 3 μM UBA1 and 25 μM Strep-Tag II-Ub K48R for 30 min at 37 °C. The charging reaction of UBE2K~Ub was diluted 10-fold into the indicated pH buffer with a final concentration of 100 μM Ub 74 as acceptor to begin the reaction. Samples were taken at 2 min, mixed with LDS loading buffer and resolved by SDS-PAGE. Protein gels were transferred to nitrocellulose membranes as described in the section on 'Continuous ubiquitination assays' and incubated with the mouse anti-Strep-Tag II (Merck) primary antibody in a 1:2,000 dilution overnight. The secondary IRDye 800CW goat anti-mouse IgG (LI-COR Biosciences) in a 1:15,000 dilution was used to detect Strep-Tag II-Ub K48R using a LI-COR Odyssey CLx. The intensity of the di-Ub band was calculated as a fraction of total Strep-Tag II-Ub K48R following a similar approach 41 and the pK a was determined from sigmoidal standard titration curve (equation (1), Prism) where η H is the Hill slope and minUb 2 and maxUb 2 correspond the minimum and maximum fraction of Ub 2 formation, respectively.
Solution NMR experiments. NMR data were acquired on a Bruker Avance III 600 MHz spectrometer with a cryogenic triple resonance inverse probe. UBE2K conjugates were exchanged into NMR buffer (20 mM sodium phosphate, 100 mM NaCl, 7.5% D 2 O, pH 7.0) at a final concentration of 50 μM. 1 H-15 N HSQC spectra were recorded at 298 K with 128 scans and 64 complex points using a sweep width of 36 parts per million (ppm) in the 15 N dimension. All spectra were processed with 256 points in the indirect dimension using Bruker TopSpin v.3.5 patch level 7 and analyzed using CARA and NMRFAM-SPARKY 58 . CSPs were calculated using the following: where, for a given residue at state A and B, δ H and δ N are the difference in proton and nitrogen chemical shifts, respectively.
SPR binding and analysis. SPR binding experiments were performed at 25 °C on a Biacore T200 instrument using a CM-5 chip (GE Healthcare) with coupled anti-GST antibody as described previously 48 . GST-tagged UBE2K variants were coupled on the chip and a serial dilution of Ub in running buffer containing 25 mM Tris-HCl, pH 7.6, 150 mM NaCl, 1 mM DTT and 0.005% (v/v) Tween-20 was used as analyte. Two technical replicates were performed and data were analyzed with BIAevalution (GE Healthcare) and fit to a one-site binding model using Prism (GraphPad).

Data availability
Coordinates and structure factors have been deposited with the protein data bank under accession code 7OJX. The data that support the findings of this study are available within the paper and its supplementary information file. Source data are provided with this paper. DNA constructs are available upon request from the corresponding author.
UBE2S UBE2W       , and 5x molar excess of (h) UBE2K M172D,L198R . No measurable binding is observed for UBE2K M172D,L198R in either SpR or NMR. T88 is located at the UBE2K/Ub A binding interface and UBE2K T88D is defective in Ub A catalysis (Fig. 3e), yet UBE2K WT and UBE2K T88D have near identical affinity for Ub. These data suggest that the canonical M172/L198 Ub binding surface contributes to the bulk of Ub binding affinity detected and that the UBE2K acceptor site has weak affinity for Ub A . (e,f) Our structure of Ub A (yellow) at the acceptor site of UBE2K(UBC cyan and UBA blue) with signal attenuation (red sticks) mapped from d and (Fig. 4d)