Ubiquitin binding domains — from structures to functions

Monoubiquitin —- one fold, many dynamic conformations

The canonical ubiquitin fold is formed by a five-stranded β-sheet, a short 3₁₀ helix and a 3.5-turn α helix (FIG. 3a). The carboxyl-terminal tail of ubiquitin is exposed, which allows its covalent linkage to target proteins. Ubiquitin is a small protein of 76 amino acids with a surface area of only 4800 A². Most UBDs interact with a solvent-exposed hydrophobic patch, which includes Leu8, Ile44, and Val70 and is located in ubiquitin's β-sheet (FIG. 3a). Although most UBDs tend to target the same surface of ubiquitin, the amino acids that surround the hydrophobic patch are chemically diverse. As a consequence, the greater than two dozen UBDs characterized so far (TABLE 1) include a myriad of structural folds and unique binding modes.

X-ray and nuclear magnetic resonance (NMR) structures of ubiquitin bound to different UBDs indicate that it adopts slightly different conformations depending on its binding partner. Lange et al. used nuclear magnetic resonance residual dipolar couplings (RDCs) to analyse the full dynamic behaviour of ubiquitin on the picosecond to microsecond time scale in solution⁹. They demonstrated that ubiquitin exhibits the structural heterogeneity of its UBD-bound states, as each corresponded to a member of the ensemble of structures sampled by free ubiquitin over time (FIG. 3b). This structural heterogeneity contributes strongly to the adaptive interface of ubiquitin during binding events to different UBDs.

Ubiquitin chains — diverse cellular signals

An attractive model that describes how the outcome of ubiquitylation is determined in cells proposed that the destiny of ubiquitylated substrates relies on the ubiquitin linkage type and chain length¹⁰. This early model applies best to ubiquitin chains that are homogeneous, namely those assembled through a single linkage type. Structural studies have revealed significant differences between ubiquitin chains of Lys48 (FIG. 3c) and Lys63 (FIG. 3d)/linear linkages (FIG. 3e). Lys63- and linearly-linked diubiquitins adopt similar extended conformations with no contact between the two ubiquitin moieties¹¹. By contrast, the ubiquitin moieties of Lys48-linked tetraubiquitin pack against each other^{12, 13} with exchange in solution between the closed, packed structure and an extended, opened one¹⁴.

Importantly, structures of ubiquitin chains in complex with UBDs have revealed that Lys48-, Lys63- and linearly-linked ubiquitin molecules adopt a wide range of inter-moiety arrangements that depend on their binding partners (Supplementary information S1 (figure)). For example, the open conformation of Lys63-linked diubiquitin is stretched in the complex with the endosome-associated ubiquitin isopeptidase AMSH (associated molecule with the SH3 domain of STAM) to allow efficient cleavage of the isopeptide bond¹⁵. In contrast, Lys63-linked diubiquitin in complex with an antibody adopts a more compact conformation in which the two ubiquitin moieties are packed together with substantial surface contacts between them¹⁶. These structural studies have led to the appreciation of ubiquitin chains as flexible modules that can be shaped by their cellular context.

UBDs that bind monoubiquitin

A large majority of UBDs uses α-helical structures to bind a hydrophobic patch within the β-sheet of ubiquitin. The ubiquitin-interacting motif (UIM), inverted UIM (MIU or IUIM) and ubiquitin-binding zinc finger (UBZ) bind this region of ubiquitin with a single helix^22-27 oriented either parallel or anti-parallel to the central β-strand (FIG. 4b and Supplementary information S2 (figure)). Other ubiquitin-binding elements, including (UBA) (ubiquitin-associated) (FIG. 4a) and CUE (coupling of ubiquitin conjugation to endoplasmic reticulum degradation) domains^28-31, bind ubiquitin through two helices that are discontinuous. It is currently unclear why so many variations of helical structures have evolved to interact with ubiquitin and regulate its downstream signalling. Several examples of their involvement in specific ubiquitin signalling to regulate proteasomal functions, DNA damage response or receptor endocytosis are discussed below.

β-sheets present in different structural domains can also engage in recognition of monoubiquitin. For example, the β-sheet of E2 ubiquitin-conjugating enzyme UBCH5c interacts with ubiquitin's hydrophobic Ile44-containing surface (FIG. 4c)³². UBCH5c functions with the E3 ligase BRCA1 (Breast Cancer Gene 1), a tumour suppressor protein with breast cancer-associated mutations. This non-covalent interaction between UBCH5c and ubiquitin activates the self-assembly of ubiquitin chains on UBCH5c and in turn the processive ubiquitylation by BRCA1³², which is crucial for coordinating DNA damage regulatory complexes (see below). Another example is the GLUE (GRAM-like ubiquitin-binding in EAP45) domain present in the EAP45 (ELL-associated protein of 45 kDa) subunit of the endocytic sorting complex ESCRT (endosomal sorting complexes required for transport). This domain folds into a split PH domain that also uses residues of β-strands S5 and S6 to bind the hydrophobic patch of ubiquitin; although in this case, additional residues from a loop and α-helix are used too^{33, 34} (FIG. 4d). The Pru (PH receptor for ubiquitin) domain present in the proteasomal receptor Rpn13 (Regulatory Particle, Non-ATPase-like 13) binds to the β-strand surface of ubiquitin with a PH domain; however, it does so in an entirely different manner³⁵. Three loops within its ubiquitin-binding domain form the binding interface to contact a larger surface compared to EAP45 and to form hydrogen bonds with His68 in ubiquitin (FIG. 4e). These properties contribute to its higher affinity for monoubiquitin³⁶. Hence, even when the same structural domain is used to bind the Ile44-containing surface of ubiquitin, recognition can be through different UBD surfaces and structural elements. This finding highlights the difficulty of identifying UBDs and of predicting ubiquitin-binding surfaces without experimental data.

The common use of the Ile44-containing surface by UBDs causes their binding to monoubiquitin to be mutually exclusive, suggesting that in most cases, a ubiquitin moiety is only acted upon by one UBD. It is possible that this mutual exclusivity is an important component of effective ubiquitin signalling, as it can prevent each ubiquitin from stimulating multiple processes, which could lead to discordance within the cellular context. An exception to this rule is demonstrated by variations in the ZnF domains; these recognize monoubiquitin by binding to three different regions on its surface (FIG. 4b). The NPL4 (nuclear protein localization 4) ZnF (NZF) domain³⁷, which serves as an ubiquitin-binding adapter protein in the endoplasmic reticulum (ER)-associated degradation (ERAD) pathway, and the ubiquitin-binding Zn fingers (UBZ domains) present in translesion synthesis polymerases (TLS)³⁸ bind to the β-sheet Ile44-containing hydrophobic surface of ubiquitin. By contrast, the RABEX-5 (Rab5 guanine nucleotide exchange factor) A20-type ZnF domain recognizes a polar surface of ubiquitin centred on Asp58²⁴, which empowers RABEX5 to act with ligase capacity toward trans substrates. A different binding surface is contacted by the deubiquitinase (DUB; also known as deubiquitylating or deubiquitinating enzyme) isopeptidase T (IsoT), which has a ZnF domain that binds to the C-terminal residues of ubiquitin³⁹. The unique ubiquitin binding mode of the IsoT ZnF domain ensures that it only disassembles unanchored chains⁴⁰ and suggests that it could act in conjunction with other UBDs that bind to ubiquitin's Ile44-containing surface. The impressive diversity of DUB structure and ubiquitin recognition strategies was recently described in an excellent review article⁴¹.

There is no evidence that binding affinity for ubiquitin can be predicted based on the UBD structural fold. By contrast, a larger range of affinities is found within the UBA domain family than between UBA domains and UBZ or CUE domains. There is also no clustering of particular UBD folds into specific effector proteins or to selective cellular functions. Rather, the same class of UBD, for example UBA domains, is found in a variety of proteins with numerous additional domains implicated in a wide spectrum of cellular functions that include endocytosis, proteasome signalling and apoptosis.

Lys linkage-specific UBDs

Linkage-specific ubiquitin recognition contributes to the diverse set of functional outcomes associated with ubiquitylation. The pioneering work of Cecile Pickart and colleagues revealed diversity in ubiquitin chain recognition preferences for 30 different UBA domains. Although some UBA domains showed little discretion between ubiquitin chains of different linkage, others preferred Lys48-linked ubiquitin chains⁴². One of the Rad23 (RADiation sensitivity abnormal 23) human homologues, HR23a, which as discussed below is a ubiquitin receptor associated with substrate targeting to the 26S proteasome, has a C-terminal UBA domain that binds with 3.6-fold higher affinity to Lys48-linked chains compared to Lys63-linked ones⁴². This UBA domain, like many other UBDs, binds to monoubiquitin with significantly lower affinity, namely 70-fold lower than its binding to Lys48-linked ubiquitin chains⁴². The structure of the HR23a C-terminal UBA domain–Lys48-linked diubiquitin complex revealed that the UBA sandwiches between the two ubiquitin moieties to form unique contacts (including some interactions with the diubiquitin linker region) and a significantly larger binding surface than it could form with a single monoubiquitin, which explains its preference for the Lys48-linkage⁴³ (FIG. 5a).

Several UBDs can selectively bind to Lys63-linked ubiquitin chains. For example, NZF domains of TAB2 (TAK1 binding protein 2) or TRABID (TRAF binding domain) preferentially bind to Lys63- over Lys48-linked ubiquitin chains¹¹. The specificity determinants that promote NZF domain interaction with ubiquitin chains of certain linkage type are currently not known.

By contrast with other UBDs, DUBs must act at regions linking ubiquitin moieties into a chain in order to access the cleavage site between ubiquitin moieties. This family of proteins in particular exhibits a high degree of specificity for certain linkage types. For example, the DUBs IsoT and USP2 readily cleave Lys48- and Lys63-linked chains, but have significantly less activity towards linear chains¹¹. By contrast the DUB CYLD (cylindromatosis tumour suppressor), which is a negative regulator of NF-κB signalling⁴⁴, cleaves linear and Lys63-linked chains, but not Lys48-linked ones¹¹. Other examples of DUBs with explicit specificity exist for ovarian tumour (OTU) domains; the A20 OTU domain hydrolyses only Lys48-linked chains whereas the TRABID OTU preferentially binds and cleaves Lys63-linked chains¹¹.

Linker regions within a tandem repeat of UBDs can also define linkage specificity. For example, RAP80 (Receptor Associated Protein 80) targets BRCA1 to DNA damage-induced foci^45-47 through its two UIMs, which bind Lys63-linked ubiquitin chains, but not Lys48-linked ones⁴⁸. The sequence between the two RAP80 UIMs promotes an appropriate protein conformation such that the UIMs are positioned for efficient avid binding across a single Lys63 linkage, thus defining selectivity (FIG. 5b)⁴⁸. By contrast, ataxin-3, a DUB linked to the development of spinocerebellar ataxia type 3, contains two UIMs with a two-residue linker sequence that determines Lys48-specific binding^{48, 49}. The specificity of tandem UIMs is interchangeable between Lys63- and Lys48-linkages by swapping the linker sequences of RAP80 and ataxin-3⁴⁶.

Thus, tandem UIMs can be spatially arranged in the context of full size proteins such that simultaneous, high-affinity interactions are favored with one ubiquitin chain linkage, but unfavorable or impossible with other linkages. Hence, the coordinated action of multiple UBDs can be used to sort substrates towards specific functional pathways according to their ubiquitin chain linkage type. These findings have provided, for example, a moleculer basis to the observation that Lys63-linked chains has a role in DNA repair rather than proteasome targeting.

UBDs specific for linear ubiquitin chains

Linear ubiquitin chains in which ubiquitin monomers are conjugated through Gly to Met linkages (FIG. 3e) have been implicated in the activation of the NF-κB signalling pathways^{7, 8}. Several proteins that regulate this pathway, including NEMO (NF-κB essential modulator), ABINs (A20 binding inhibitor of NF-κB) and optineurin, contain the UBAN (ubiquitin binding in ABIN and NEMO) domain, which specifically binds to linear ubiquitin chains^{8, 11, 50, 51}. Binding of NEMO UBAN domain to monoubiquitin is undetectable in vitro, whereas binding to Lys63-linked diubiquitin is 100-fold weaker than to linear chains^{8, 50}. Specific mutations of NEMO that block interactions with linear ubiquitin chains impair the activation of the IKK (IκB kinase) complex and NF-κB in response to TNFα stimulation⁸.

The UBAN domain of NEMO and linear diubiquitin form a heterotetrameric complex with two linear diubiquitin molecules on either side of the NEMO-UBAN coiled-coil dimer (FIG. 5c)⁸. Specificity for linear ubiquitin chains is provided by a continuous surface along the coiled-coil that interacts with the canonical Ile44-containing surface, the C-terminal tail of the distal ubiquitin and a novel interaction surface of the proximal ubiquitin⁸. Two loops, one from the distal ubiquitin molecule (from Pro37 to Gln40) and the other from the proximal ubiquitin (from Glu92 to Glu94), interact with each other and with the C-terminal tail of the distal ubiquitin (from Arg72 to Gly76). These residues form a structural core, which holds the two moieties in a semi-fixed orientation, and provides the basis for specificity towards linear ubiquitin chains.

The specificity of UBDs for ubiquitin linkages can provide the in vivo significance for the activation of the NF-κB pathway. The NZF domain of TAB2 binds preferentially to Lys63 chains and is important for recruitment and activation of the serine threonine kinase TAK1 (TGF-beta activated kinase 1), which is required for activation of the NF-κB and MAPK (mitogen-activated protein kinase) pathways upon TNF stimulation⁵². On the other hand, linear ubiquitin chains affect the IKK complex through direct binding to the UBAN domain in neighbouring NEMO, which in turn can activate the NF-κB pathway^{7, 8}.

It is likely that linear specific UBDs might be involved in regulation of additional cellular processes. For example, the UBAN containing protein ABIN-1 binds preferentially to linear chains and was suggested to act as a negative regulator of the NF-κB pathway⁵⁰. However, ABIN1 -/- mouse embryonic fibroblasts do not have any defect in TNFα-induced NF-κB activation. ABIN1-deficient mice die during the embryonic development due to massive apoptosis in the liver. It was suggested that ABIN1 functions as a ubiquitin-dependent anti-apoptotic sensor in mice⁵³.

ESCRT-ing ubiquitylated cargoes

Endocytic sorting of ubiquitylated surface receptors, such as epidermal growth factor receptor (EGFR), for lysosomal degradation via multivesicular bodies (MVBs) is performed by a series of ubiquitin receptors incorporated into larger complexes named ESCRT^{17, 54-56}. The machinery includes four complexes (ESCRT-0, ESCRT-I, ESCRT-II, and ESCRT-III), which are required for sequential sorting of cargoes. Structural details of isolated components and their complexes for each of the ESCRT assemblies have been revealed in recent years^{20, 57}.

The cores of these complexes are structurally conserved, mostly with α-helical domains. The regions responsible for ubiquitin recognition, UIM in ESCRT-0, UEV (ubiquitin-conjugating enzyme E2 variant) in ESCRT-I, NZF and PH in ESCRT-II, are at the peripheries of the ESCRT complexes to enable efficient recruitment of ubiquitylated cargoes. As discussed below, there are substantial variations in the mode of ubiquitin recognition by these domains, which vary from yeast to mammalian systems. Recognition of multiple ubiquitin moieties attached to a trafficking cargo must be robust in order to control sorting of often bulky cargoes, but cannot be too strong for efficient and seamless trafficking^{20, 54}. The appropriate level of ubiquitin interaction with the ESCRT complexes is accomplished by the multivalency of the ubiquitin-recognizing subunits and by local concentration of ubiquitin decorated cargoes.

A remarkable example of multivalent interactions is the use of double-sided ubiquitin binding modes in which α-helical (single, double and triple) structures form two ubiquitin binding surfaces. HRS (hepatocyte growth factor-regulated tyrosine kinase substrate) is a part of the ESCRT-0 complex and its UIM, which forms a short α-helix, binds two ubiquitin molecules with equal affinity. The crystal structure of the UIM in complex with two ubiquitin molecules shows that both ubiquitin moieties use the canonical Ile44-containing patch, but interact with the UIM on opposing sides with pseudo two-fold symmetry (FIG. 5d)²³. This binding mode is accomplished by a shift of two residues along the UIM sequence, which creates a very similar ubiquitin-binding surface on both sides of the helix.

There are other examples of such double-sided UIMs including human co-chaperone HSJ1 [because of human nomenclature](heat-shock protein DnaJ homolog) and kinase MEKK1 (MAP kinase kinase kinase 1), endocytic regulators EPS15 and EPS15R (epidermal growth factor receptor substrate 15 and EPS15 related, respectively). Interestingly Vps27, a yeast homologue of HRS, has two single-sided UIMs (Supplementary information S2a (figure)) whereas mouse EPS15 also has tandem UIMs, but only one of them is double-sided (Supplementary information S2b (figure)). It is conceivable that the double-sided UIM and/or tandem UIM arrangements evolved from copies of single-sided UIMs to provide additional levels of control.

An example of double-sided ubiquitin binding is also found in the ESCRT pathway. GGA (Golgi-localized, gamma-ear-containing) and TOM1 (Target of Myb1) function as ESCRT-0 complex associated proteins^{58, 59}. The GAT (GGA and TOM) domains of GGA3 and TOM1 fold into a three-helix bundle to form two ubiquitin binding surfaces, a high affinity one formed by its first and second helices (Supplementary information S2c (figure))^{58, 60} and a low affinity binding surface formed by its second and third helices^{61, 62}. Like in the case of UIMs, the dual ubiquitin recognition of these proteins might enhance their affinity towards multiply ubiquitylated surface receptors.

Multisite interactions in DNA damage pathways

A prominent example among the various DNA damage pathways regulated by ubiquitin signals⁶³ is translesion DNA synthesis (TLS), which responds to damage-induced stalling at replication forks and provides a good example of the involvement of multiple UBDs in regulating transcriptional response to DNA damage⁶⁴. In the TLS pathway, DNA damage promotes monoubiquitylation on Lys164 of proliferating cell nuclear antigen (PCNA), a processivity factor that forms a sliding clamp around DNA. PCNA weakly interacts with numerous DNA polymerases, but monoubiquitylation of PCNA favours its preferential interaction with Y-family TLS polymerases, which possess UBDs (ubiquitin-binding motif (UBM) and UBZ). These error-prone polymerases are able to bypass the lesion, which would otherwise block DNA replication. Functional integrity of UBDs is essential for proper localization of TLS polymerases to replication foci and cellular survival following UV irradiation^{38, 65}. UBZ and UBM domains bind monoubiquitin with comparable affinities, but they interact via distinct surfaces: UBZ requires the Ile44-containing hydrophobic patch whereas UBM binds independently of Ile44^{22, 38}.

The presence of multiple functional UBMs in mammalian polymerases REV1 (reversionless 1) and pol iota led to the hypothesis that these UBMs interact with multiple ubiquitylated components of replication foci. In accordance, mutation of either of REV1's UBM domains partially reduces its localization to stalled replication foci induced by UV light⁶⁶. The effect is more severe in pol iota, as mutation of its UBM domains completely blocks its accumulation at replication foci during S phase, and this result is independent of DNA damage³⁸. The prominent role of the UBM domains does not seem to apply to all organisms, since UBM1 of Rev1 is not functional in yeast cells⁶⁶. Swapping experiments have revealed that specific UBDs in host proteins are linked to distinct functional roles, which cannot be assumed by other UBDs. For instance, an exchange of the UBZ domain of Werner helicase interacting protein 1 (WRNIP1), which is involved in regulation of genome stability, for pol iota's UBM1, both of which can bind ubiquitin, abolishes foci formation of WRNIP1⁶⁷, suggesting that different UBDs have distinct functional roles in the context of full size proteins.

Proteasomal ubiquitin receptors

In Saccharomyces cerevisiae, there are five known ubiquitin receptors that can support protein degradation by the proteasome; two of these, Rpn10 (S5a in humans)⁷² and Rpn13³⁶, are proteasome subunits, whereas the other three, Rad23, Dsk2 (dominant suppressor of KAR1), and Ddi1 (DNA-damage inducible 1), bind to the proteasome's RP reversibly. Orthologues of these proteins exist in higher eukaryotes, some of which also have multiple paralogues. In yeast, none of these proteins are essential and even strains with all five of these receptors deleted are viable, suggesting that other components can mediate substrate docking to the proteasome in their absence³⁶; some candidates have been implicated through crosslinking experiments^{73, 74}. In mice, RPN10 and RAD23 are essential^{75, 76} and might therefore have more specialized roles in higher eukaryotes.

RPN10 contains UIMs, which bind monoubiquitin and ubiquitin chains⁷⁷. The two UIMs of S5a adopt helical configurations and bind monoubiquitin independently due to their separation by flexible linker regions (Supplementary information S2d (figure))²⁷. In binding to Lys48-linked diubiquitin however the two UIMs bind simultaneously to the two ubiquitin moieties for a markedly increased affinity⁷⁸. S5a binds Lys48- and Lys63-linked chains equivalently, but exhibits significantly weaker affinity for a mixture of Lys29- and Lys6-linked chains⁴². It also binds Lys11-linked ubiquitin chains, in accordance with their ability to effectively target substrates to the proteasome for degradation⁷⁹.

In contrast to S5a, RPN13 has only one ubiquitin interacting surface, which occupies the same structural domain used to dock this receptor into the proteasome^{35, 36}. As discussed above, this domain folds into a PH domain and three of its loops bind ubiquitin (FIG. 4e)³⁵. With the exception of the shorter Rpn13 protein from S. cerevisiae, Rpn13 binds the DUB Uch37 and thereby recruits it to the proteasome's RP^80-82. It seems to bridge substrate docking with ubiquitin chain deconjugation. Human RPN13 exhibits strong affinity for even monoubiquitin³⁶ and is therefore expected to bind to chains of all linkage type. It does exhibit preference however for the proximal over the distal ubiquitin of Lys48-linked diubiquitin due to favorable contacts with the linker region³⁵. This preference is expected to render the distal ubiquitin moiety available for other interactions, such as with Uch37. RPN13 binding to ubiquitin chains and UCH37 might simultaneously facilitate UCH37's distal end⁸³ deubiquitylating activity^80-82 by orienting the ubiquitin moieties in a configuration favorable to hydrolysis. Future experiments are needed to test this model and whether RPN13's ubiquitin-binding capacity contributes to its activation of UCH37.

Shuttling ubiquitin receptors

Proteins containing ubiquitin-like (UBL) and UBA domains are able to recruit ubiquitylated substrates to the proteasome for degradation^84-88. Their UBL domain can bind to a scaffolding protein in the base of proteasome's regulatory particle, Rpn1 (S2 in humans)⁸⁹, as well as to Rpn13³⁶ and S5a^90-92, while their UBA domains bind to ubiquitin^{93, 94}. Depending on their cellular protein levels UBL-UBA containing proteins can also inhibit the degradation of ubiquitylated substrates⁸⁸. Such inhibition has been hypothesised to occur because UBA domains sequester ubiquitin chains to in turn prevent deubiquitylation, which is prerequisite to substrate hydrolysis by the 20S core particle of the proteasome^{88, 95, 96}.

The different UBL-UBA family members exhibit different ubiquitin-binding preferences and modes. A human Rad23 orthologue, HR23a, effectively sequesters ubiquitin moieties of chains up through eight ubiquitin monomers⁹⁷ and, as discussed above, its C-terminal UBA domain sandwiches between the ubiquitin moieties of Lys48-linked diubiquitin, making close contact with the linker region connecting the two moieties (FIG.3a)⁴³. These properties explain HR23a's ability to inhibit substrate deubiquitylation and are markedly different from Rpn13's binding to and activation of the proteasomal DUB Uch37. They presumably make Rad23 proteins better shuttling factors for passing substrates to the proteasome, as substrates will more likely remain ubiquitylated during transport. There is evidence that certain proteasome substrates require specific ubiquitin receptors^{85, 88}. More studies are needed to define substrate selectivity requirements for specific receptors; however, specificity is expected to be particularly applicable in higher eukaryotes in which Rad23 proteins⁷⁶ and S5a⁷⁵ are essential for viability.