Rationale for use of tetrameric branched Ub chains

The minimal branched Ub chain unit is commonly considered to be made up of three Ub moieties, with two distal Ub moieties linked to a single proximal Ub. This branching may create or disrupt interfaces for protein interactions compared to the unbranched chain. Nevertheless, we envisaged the use of branched tetrameric Ub (Ub4), wherein a single Ub branches off the center of a homotypic trimeric Ub (Ub3) chain ‘trunk’ (Fig. 1a). Notably, such branched Ub4 chains potentially encode additional information when compared to the minimal branched Ub3. This is because branched Ub4 not only possesses additional unique interfaces but can also be differentiated by the order of linkages (for example, K48-Ub branching off a K63-Ub trunk or vice versa). Consequently, for this study, we chose to use tetrameric branched Ub because crucial information may be overlooked with shorter branched Ub3.

Nomenclature to describe complex Ub chains

We incorporate various modifications such as substitutions, isotope labels and affinity tags into precise positions of branched and unbranched Ub4 chains in this study. To accurately describe the architecture of these complex Ub chains, we adapted the nomenclature introduced by Nakasone et al.27 to describe tetrameric branched and mixed chains (detailed examples are described in Extended Data Fig. 1). Because investigating heterotypic Ub chains is a rapidly expanding field, we believe that the timely adoption of one standardized nomenclature will avoid future confusion.

Assembly and structure of branched K48–K63 chains

We used two complementary enzymatic approaches of Ub chain assembly strategies that enable the assembly of well-defined, complex Ub chains. Previous approaches to generate branched Ub3 used a Ub moiety lacking the C-terminal glycine residues (UbΔC) (Extended Data Fig. 1a)9,25,26. As such chains lack the native C terminus on the proximal Ub, we adapted the ‘Ub-capping’ strategy to permit the assembly of longer and more complex chains (Extended Data Fig. 1b). Here, a blocking group, a ‘cap’, is installed at the C terminus of Ub and subsequently cleaved off by a DUB28. We used capped M1-linked Ub2 wherein the proximal Ub has a truncated C terminus and lysine-to-arginine substitutions such that only lysine residues from the distal Ub of this capped Ub2 are available for ligation to another Ub. This cap is removed using the M1-specific DUB OTULIN revealing a native C terminus on the now proximal Ub that is available for further ligation steps. Using this approach, we successfully assembled milligram quantities of pure K48–K63-branched Ub4 chains and confirmed their linkage composition using linkage-specific DUBs (Fig. 1b).

To gain insights into the structure of K48–K63-branched chains, we determined the crystal structure of the branched trimer (UbK48R, K63R)2–48,63Ub1–72 (Table 1). In this K48–K63-branched Ub3 structure, the K48-linked Ub adopts a closed conformation with interactions between the two I44 patches of the distal and proximal moieties (Fig. 1c), while the K63-linked Ub adopts an open, extended conformation. These closed and open Ub configurations have been observed previously for K48-linked and K63-linked Ub2, respectively (Fig. 1d,e)29,30, and for branched and mixed K48–K63-linked Ub3 in nuclear magnetic resonance spectroscopic analyses27.

Table 1 Crystallographic data collection and refinement statisticsIdentifying linkage-specific binders of branched Ub chains

To discover cellular proteins that bind to specific Ub chain architectures, we generated branched K48–K63-linked Ub4 and unbranched K48-linked or K63-linked Ub4 chains, covalently immobilized on agarose beads at the C terminus of the proximal Ub (Fig. 2a and Extended Data Fig. 2a). Crucially, immobilization by a defined anchor ensures that the branched interfaces are available for protein interaction. We then identified binding proteins using data-independent acquisition (DIA) MS/MS (Fig. 2a and Extended Data Fig. 2b). Analyzing the normalized binding Z scores of the 7,999 unique protein isoforms identified across the chain pulldown samples, we found 130 proteins with binding profiles that differed significantly from at least one other chain type (Fig. 2b and Supplementary Table 1).

Fig. 2: Identification of proteins binding to homotypic and heterotypic Ub4 chain architectures with K48 and K63 linkages.

a, Schematic workflow of pulldown from U2OS cell lysates using functionalized Ub4 chains and subsequent DIA MS/MS analysis. b, Heat map showing binding Z scores of 130 proteins with statistically significant differences in binding profiles identified in quadruplicate Ub chain pulldowns. Schematics of the chains used in pulldown are depicted on top. Spatial Euclidean distance computations were applied to rank proteins (left tree) and replicates (bottom tree). The six main distance clusters of proteins representing binding preferences are color-coded from red to blue.

Source data

These 130 significant hits could be sorted into six main clusters of Ub chain interactors: proteins that mainly bind unbranched K63-linked Ub chains (clusters 1 and 2), branched K48–K63-linked Ub chains (clusters 3 and 4) and unbranched K48-linked Ub chains (clusters 5 and 6) (Fig. 2b). Gene ontology enrichment analysis revealed a strong association with Ub-related biological processes (Fig. 3a).

Fig. 3: Specific binders of K48–K63-branched Ub chains.

a, Sankey diagram connecting the six distance clusters to annotation clusters of DAVID gene ontology analysis colored by annotation enrichment score. b, Table of molecular functions and known binding motifs of proteins specifically associated with K48–K63-branched Ub chains (clusters 3 and 4). c, Schematic of p97 subcomplexes with varying Ub chain-binding preferences. d, Silver-stained SDS–PAGE analysis of HALO pulldown with recombinant HALO-tagged RFC1 UBD [190–246] and branched/unbranched Ub4 containing K48 and K63 linkages.

Source data

Proteins in cluster 1 preferentially associate with long K63-linked Ub chains (63Ub4 and (Ub)2–48,63Ub–63Ub) but not with K48-linked Ub or the single K63-linked Ub branching off a K48-linked Ub trunk (48Ub4 or (Ub)2–48,63Ub–48Ub). In contrast, the proteins in cluster 2 show a propensity to interact with the shorter K63-linked Ub2 present in the branch of (Ub)2–48,63Ub–48Ub. Proteins in these two clusters are strongly linked to biological processes associated with K63-linked ubiquitination, including protein unfolding and refolding, autophagy and protein sorting and endosomal transport (Figs. 2b and 3a). These include annotated K63-binding proteins such as the BRCA1 complex components ABRAXAS-1, BRCC36 and UIMC1/RAP80 (refs. 31,32,33), as well as endosomal trafficking-related proteins EPS15, ANKRD13D, STAM, STAM2, TOM1, HGS, TOM1L2 and TOLLIP (refs. 34,35,36,37,38). Furthermore, we identified less-studied proteins with annotated UBDs, such as CUEDC1 and ASCC2 (CUE), N4BP1 (CoCUN)39,40, CCDC50 (MIU)41 and RBSN (UIM), to preferentially bind K63 chains.

Cluster 5 comprised 18 proteins that primarily bind to K48 linkages in chains, regardless of whether they are within homotypic or branched architectures. Similarly, cluster 6 contained 17 proteins that bind strongly to unbranched K48-linked chains and weakly to branched (Ub)2–48,63Ub–48Ub, suggesting that these proteins either prefer binding to longer K48-linked chains (>2 Ub) or disallow binding to the single K48-linked Ub branching off a K63-linked Ub trunk ((Ub)2–48,63Ub–63Ub). Proteins predominantly binding to unbranched K48-linked Ub chains include the proteasomal Ub-binding component PSDM4/Rpn10, the segregase p97 and its substrate adaptors UBXN1, UFD1, NSFL1C/p47 and NPLOC4 (Fig. 3b). Other identified K48 binders are the proteasome shuttling factors RAD23A and RAD23B and the DUBs MINDY1, OTUD5, USP25 and USP48 (refs. 20,42,43). We also identified several proteins without annotated UBDs such as MTMR14, ZFAND6 and TBC1D17 as potential binders to K48-linked and K63-linked chains. However, it remains unclear whether these proteins directly bind to the Ub chains or whether they copurified as part of a multiprotein complex containing a UBD.

Remarkably, we identified seven proteins (cluster 3) that strongly associate with the two branched chain architectures but not with the unbranched K48-linked or K63-linked chains (Fig. 3b). These include proteins implicated in DNA replication (RFC1), histone deubiquitination (USP15), reading histone methylation (MORC3), ERAD (USP13 and DNAJB2) and peptide antigen loading (TAPBP) (Fig. 3b)44,45,46,47,48,49. Notably, 8 of the 15 identified proteins in clusters 3 and 4 contain annotated UBDs and Ub-interacting motifs (UIMs), suggesting that they may bind directly to the branched chains. Intriguingly, three of the eight proteins in cluster 4 (ATXN3, ZFAND2B and RHBDD1) also possess p97-binding motifs (VIM and VBM) along with UIMs (Fig. 3c). Interestingly, p97 is ranked between ATXN3, ZFAND2B and RHBDD1 and the established p97 substrate adaptors NPLOC4 and UFD1, which were previously shown to bind K48-linked Ub chains to initiate unfolding of modified client proteins50,51. Appropriately, we also detected additional p97-binding proteins or substrate adaptors (UBXN1 and NSFL1C/p47) in cluster 5, which mainly contained unbranched K48-linked Ub chain binders52,53. These results indicate the coexistence of p97 complexes functionalized with different substrate adaptors that confer a range of Ub chain preferences from unbranched K48-Ub to K48–K63-branched Ub chains (Fig. 3c).

We then attempted to validate our MS data in vitro using recombinant proteins for the identified branched-chain-specific binders. However, most interactors did not express as soluble full-length proteins, several of which are likely part of large multiprotein complexes in vivo. We, therefore, tested whether the specificity toward branched chain binding was encoded within the predicted UBDs of the proteins. Only the minimal UBD of RFC1 (amino acids 190–246) showed high specificity of binding to K48–K63-branched Ub4 chains with no detectable binding to the unbranched Ub4 controls (Fig. 3d and Extended Data Fig. 3a,b). Notably, the minimal UBD of RFC1 did not bind to K48–K63-branched Ub3. In contrast, the predicted UBDs from the other binders either did not bind to the Ub chains tested or lacked specificity (Extended Data Fig. 3a), suggesting that additional regions or cofactors may be required for branched Ub binding. In summary, these pulldown results reveal the existence of branched-Ub-specific binding proteins and demonstrate that cellular proteins can differentiate between tetrameric and trimeric branched chains, suggesting that the unique interfaces present in tetramers are being specifically recognized (Fig. 1a).

Ub linkage target identification by mass tagging (ULTIMAT) DUB assay monitors cleavage of individual Ub links

As we identified multiple DUBs in the pulldown with branched and unbranched Ub chains, we next investigated whether some DUBs can preferentially cleave branched Ub chains. However, conventional DUB assays, which monitor polyUb chain cleavage, lose information on which specific linkage within a Ub chain is cleaved54. Therefore, to overcome this limitation, we developed a precise, quantitative DUB assay, ULTIMAT. The principle of the ULTIMAT DUB assay relies on the use of substrate Ub chains in which each Ub moiety is of a discrete mass that can be distinguished by matrix-assisted laser desorption/ionization time-of-flight (MALDI-TOF) MS (Fig. 4a). After incubation with a DUB, the released monoUb species are detected using MALDI-TOF MS, enabling the identification and quantification of the exact linkage cleaved. The monoUb species were analyzed by MALDI-TOF MS and quantified relative to an internal standard of 15N-labeled Ub (Ub15N = 8,670 Da)55 (Fig. 4b). As controls, we first analyzed the activity of the K63-specific DUB AMSH and K48-specific DUB MINDY1, demonstrating that they only cleave the K63-linked or K48-linked Ub moieties, respectively (Extended Data Fig. 4a).

Fig. 4: Debranching activity of 53 human DUBs profiled with the ULTIMAT DUB assay.

a, Principle and schematic workflow of the ULTIMAT DUB assay. b, Mass spectrum of the four released Ub moieties detected in the ULTIMAT DUB assay and added Ub15N internal standard with indicated masses. Asterisks indicate metastable ion peaks. AU, arbitrary units. c, Screen of 53 human DUBs in duplicate using ULTIMAT DUB assay with K48-linked and K63-linked chains. The heat map shows individual data points of duplicate measurements of released Ub moieties normalized to the internal Ub15N standard and relative to the cleaved control substrate. The schematic of substrates and the location of Ub moieties are depicted above the heat map. Control substrates are either homotypic Ub chains of specific linkage type and length (for example, 632 = K63-linked Ub2), Ub with C-terminal tryptophan (-Trp), Ub modified with isopeptide-linked lysine (-Lys) or Ub with ester-linked threonine (-Thr).

Source data

Having confirmed the robustness and reproducibility of this method, we proceeded to analyze a panel of 53 human DUBs for their activity toward homotypic and branched Ub substrates in comparison to a positive control substrate (Fig. 4c). As anticipated, no cleavage of K48-linked and K63-linked substrates was detected for the highly M1-specific DUB OTULIN or members of the UCH DUB family that prefer short and disordered peptides at the C terminus of Ub56,57. To our surprise, we did not observe UCHL5 to debranch K48–K63-branched Ub chains as previously reported26,58. This discrepancy is likely because of differences in assay conditions (Extended Data Fig. 4b). Because only about ~5% of K48–K63-branched Ub3 was cleaved by UCHL5, we conclude that K48–K63-branched Ub chains may not be the preferred substrate of UCHL5.

Members of the USP family, known to be less linkage-selective, displayed broad cleavage activity against all tested substrates, with a particular tendency to cleave from the distal end of the chain (Fig. 4c). Notably, we observed a moderate inhibitory effect of the branched chain architecture on CYLD activity, as previously reported4 (Fig. 4c). Importantly, we identified certain DUBs, such as MINDY family members and ATXN3, that showed a marked preference for cleaving branched Ub chains.

Unique Ub-binding site enables MINDY1’s debranching activity

In the ULTIMAT DUB assay, both MINDY1 and MINDY3 stood out for their high activity in cleaving K48 linkages off branched chains. The K48-specific DUB MINDY1 is an exo-DUB that favors long K48-linked chains as substrates and has five well-characterized Ub-binding sites on its catalytic domain59. It is, however, virtually inactive against shorter K48-linked Ub2; therefore, we could detect cleavage of only the distal Ub of 48Ub3 (Fig. 4c). Interestingly, the ULTIMAT DUB assay revealed that MINDY1 cleaved the distal K48-linked Ub off the branched chains more efficiently than the distal Ub of unbranched 48Ub3. To our surprise, we found that MINDY1 activity was also enhanced toward (Ub)2–48,63Ub–63Ub, a branched chain where a single K48-linked Ub branches off a K63-linked Ub3 trunk (Fig. 4c and Extended Data Fig. 4a).

We systematically analyzed the processing of K48–K63-branched chains by the MINDY DUB family (MINDY1–MINDY4). Comparing minimal catalytic domains to full-length MINDYs in an ULTIMAT DUB assay against branched and unbranched K48-linked and K63-linked substrates revealed that full-length MINDY1 cleaved 5.4-fold more branched chains than the distal Ub of unbranched 48Ub3 (Fig. 5a,b). This activity was only 2.8-fold higher for the catalytic domain, suggesting that the tandem MIUs have a role in effective branched-chain processing. In contrast, both full-length MINDY2 and the catalytic domain alone processed the distal K48-linked Ub of 48Ub3 and the two branched Ub4 chains with similar efficiency.

Fig. 5: Debranching activities of MINDY family and ATXN3: linkage specificities and identification of a K63-branch-binding site on MINDY1.

a, Schematic domain overview of active MINDY family members with highlighted catalytic cysteine residues. b, ULTIMAT DUB assay of catalytic domains and full-length constructs of MINDY family members against branched and unbranched K48-linked and K63-linked substrate chains. The heat map depicts individual data points of duplicate measurements of released Ub moieties normalized to the internal Ub15N standard and to the intensity of the distal Ub of 48Ub3. c, Crystal structure of the catalytic domain of MINDY1 in complex with 48Ub2 (PDB 6TUV)20 with MINDY1 residues colored by ScanNet binding probability score (blue, white and red) and Ub molecules in gray. Zoomed-in view of predicted K63-Ub-binding site, with residues shown as stick models. d, Silver-stained SDS–PAGE of DUB assays with catalytic domain of wild-type MINDY1 or point mutants in potential K63-Ub-binding site (L281A or V277R) screened against a panel of branched and unbranched K48-linked and K63-linked Ub chains. e, Schematic of six Ub-binding sites located in MINDY1’s catalytic domain with K48-linked Ub-binding sites in blue (S1, S1′–S4′) and K63-linked Ub site in red (S1′br). Site connectivity is indicated by dashed lines and the catalytic cysteine is indicated by a yellow star. f, Schematic domain overview of ATXN3 with highlighted catalytic cysteine residue. g, ULTIMAT DUB assay in duplicate with full-length and C-terminally truncated ATXN3 against branched and unbranched K48-linked and K63-linked substrate chains. The heat map shows individual data points of duplicate measurements normalized to the internal Ub15N standard as the absolute percentage of substrate linkage cleaved. h, Silver-stained SDS–PAGE analysis of DUB assays with full-length ATXN3 (top) and ATXN3 (1–260; bottom) against a panel of branched and unbranched K48-linked and K63-linked Ub chains. i, Silver-stained SDS–PAGE analysis of DUB assays with full-length ATXN3 against Ub4 chains of K63-linked 48Ub2 (63(48Ub2)2) and branched (Ub)2–48,63Ub–48Ub.

Source data

MINDY3 demonstrated comparable activity against the distal Ub of unbranched 48Ub3 and branched (Ub)2–48,63Ub–48Ub but, strikingly, it was 4.4 times more active at cleaving the K48-linked distal Ub off the K63-Ub trunk in (Ub)2–48,63Ub–63Ub (Fig. 5b). These data suggest a specific role of MINDY3 in removing K48-Ub chain linkages branching off K63-Ub chains. In contrast, MINDY4 efficiently cleaved distal K48 linkages in both unbranched 48Ub3 and branched [Ub]2–48,63Ub–48Ub but displayed reduced processing of (Ub)2–48,63Ub–63Ub (0.5-fold) (Fig. 5b). In summary, we found that each MINDY family member has a unique cleavage profile for branched K48–K63-linked Ub chains with MINDY1 and MINDY3 demonstrating a preference for branched substrates.

MINDY1 and MINDY2 have five defined Ub-binding pockets for K48-linked Ub on the catalytic domains20. However, these previously identified Ub-binding sites (S1, S1′–S4′) would not be able to accommodate a K63-linked Ub of a branched K48–K63-linked Ub chain, as the K63 residue of the proximal Ub in the S1′ pocket is situated opposite to these known K48-binding sites (Fig. 5c). To understand how branched chains are bound, we analyzed the protein-binding probability of MINDY1 surface residues using ScanNet60, which predicted a high-confidence binding patch adjacent to the S1′ pocket of MINDY1 near the K63 residue of the proximal Ub of 48Ub2 bound to MINDY1 (Fig. 5c). We hypothesized that substituting the residues in this potential K63-linked Ub-binding site in MINDY1 should affect the cleavage of branched K48–K63-linked Ub chains but not unbranched K48-linked chains. Indeed, MINDY1 V277R or L281A substitutions abolished the cleavage of (UbK48R, K63R)2–48,63Ub and K48–K63-branched Ub4 while processing of unbranched K48-linked Ub3 was unaffected (Fig. 5d), providing evidence that the catalytic domain of MINDY1 has a sixth Ub-binding site that recognizes K63-linked branched Ub (S1′br site) that is distinct from the other five previously identified K48-linked Ub-binding sites (Fig. 5d). Importantly, MINDY1 was unable to cleave mixed, unbranched Ub4 containing both K48 and K63 linkages, confirming that the enhanced cleavage activity is specific to K48 linkages present within K48–K63-branched chains and does not result from a combination of K48 and K63 linkages per se (Extended Data Fig. 5a). In addition, MINDY1 was unable to cleave other branched Ub3 chains containing K11–K48 or K29–K48 linkages, which agrees with the distant positions of the other lysine residues of the proximal Ub in the S1′ binding site relative to the S1′br site (Extended Data Fig. 5b,c).

ATXN3 is a K63-specific debranching enzyme

The ULTIMAT DUB assay screen revealed the p97-associated DUB ATXN3, previously considered to cleave long K63-linked chains22, to have tenfold higher cleavage activity toward the distal K63-linked Ub in the two branched Ub4 substrates compared to the control Ub-Thr substrate21. However, unbranched 63Ub3 and the proximal K63-linked Ub were not cleaved (Fig. 4c). ATXN3, a member of the Josephin family of DUBs, has an N-terminal catalytic domain followed by a helical extension, tandem UIM (UIM1–UIM2) and a third C-terminal UIM (UIM3) (Fig. 5f). We generated truncated versions of ATXN3 to dissect the potential roles of p97 and the various UBDs in ATXN3 toward debranching activity. An ULTIMAT DUB assay comparing truncated ATXN3 versions revealed that the catalytic domain and the tandem UIM (ATXN31–260) are the minimal domains required for efficient cleavage of the branched chain architectures (Fig. 5g), while hydrolysis of the control substrate Ub-Thr was unaffected. Next, we conducted a gel-based time-course experiment comparing the activity of full-length ATXN3 and ATXN31–260 (Fig. 5h and Extended Data Fig. 5d). While unbranched 48Ub4 was a poor substrate and ATXN3 did not cleave 63Ub4 or branched (Ub)2–48,63Ub, we observed that both ATXN3 constructs remarkably cleaved about 50% of the tetrameric branched chains within 5 min (Fig. 5h).

ATXN3 was previously reported to prefer cleaving long K63-linked Ub chains and K63 linkages in mixed, unbranched Ub chains containing K48 and K63 linkages22. It is worth noting that the ‘mixed’ chain used in the previous study was assembled by ligating two wild-type K48-linked Ub2 using the K63-specific E2 enzymes UBE2N and UBE2V1. Such an assembly would result in a mixture of branched and mixed Ub4 chains, as one 48Ub2 molecule could be ligated to the proximal or distal Ub moiety of the other 48Ub2 (that is, creating branched (Ub)(Ub–48Ub)–48,63Ub or mixed Ub–48Ub–63Ub–48Ub) (Fig. 5i). To directly compare ATXN3 activity against mixed and branched chains, we compared the ability of ATXN3 to cleave the mixed chain,63(48Ub2)2 and branched (Ub)2–48,63Ub–48Ub. While only a small fraction of the mixed 63(48Ub2)2 was cleaved to 48Ub2 after 2 h, the majority of branched (Ub)2–48,63Ub–48Ub was debranched within 30 min (Fig. 5i), demonstrating that branched rather than mixed K48–K63-linked Ub chains are the preferred substrates of ATXN3.

Engineering a branched K48–K63-Ub-specific nanobody

To enable the facile detection of branched chains, we set out to develop nanobodies61. Using a synthetic yeast surface display nanobody library62, we devised a screening strategy to obtain nanobodies capable of selectively binding to K48–K63-branched Ub chains (Fig. 6a). In four rounds of negative and positive selection, we removed undesired binders to unbranched K48-linked or K63-linked Ub chains and enriched for binders to K48–K63-branched Ub3 ((UbK48R, K63R)2–48,63Ub1–72-AVI*biotin), respectively. A promising candidate nanobody, NbSL3, had submicromolar affinity (KD = 740 ± 140 nM) for (Ub)2–48,63Ub and exhibited good solubility in bacterial and mammalian cell expression (Fig. 6b,c and Extended Data Fig. 6a,b).

Fig. 6: Engineering of the K48–K63-branched Ub-specific, high-affinity nanobody NbSL3.3Q.

a, Schematic workflow of nanobody selection and maturation using yeast surface display screening using biotinylated (B), Avi-tagged Ub chains immobilized on magnetic streptavidin beads (Strep). b, Sequence alignment, CDRs and secondary structure elements of NbSL3 and its variants. The four substitutions of the maturation from NbSL3 to NbSL3.3Q are indicated by red triangles. c, ITC analysis of first-generation nanobody NbSL3 and matured third-generation nanobody NbSL3.3Q binding to branched K48–K63-linked Ub3. DP, differential pressure. d, Silver-stained SDS–PAGE analysis of in vitro pulldown with NbSL3.3Q-immobilized agarose beads against a panel of branched and unbranched Ub3 chains. e, Silver-stained SDS–PAGE of DUB assay with full-length ATXN3 and (UbK48R, K63R)2–48,63Ub–48Ub1–72 incubated at 30 °C for 2 h following the addition of K48–K63-branched Ub-specific nanobody NbSL3.3Q. f, Cocrystal structure of NbSL3.3Q (yellow) in complex with (UbK48R, K63R)2–48,63Ub1–72 (blue, red and gray) in cartoon representation with semitransparent surface, rotated by 120°. Zoomed-in views of nanobody interactions in proximity to K48 (right) and K63 (left) linkages shown as stick models. Interatomic distances are indicated by black dashed lines with distance measurements in Å.

Source data

To improve the affinity and specificity of NbSL3, we performed affinity maturation using site-directed saturation mutagenesis to randomize individual amino acid positions in the complementarity-determining regions (CDRs) of the candidate nanobody, resulting in a diverse NbSL3-based yeast library with ~2 × 108 unique nanobody sequences. After four rounds of negative and positive selection, we identified nanobodies (NbSL3.1–NbSL3.4) exhibiting affinities in the low-nanomolar range (~1–100 nM) for K48–K63-branched Ub chains (Extended Data Fig. 6c). Next, we combined the substitutions of the top two nanobodies (NbSL3.3Q) (Fig. 6b). Strikingly, NbSL3.3Q demonstrated picomolar