Early Career Research Award

The emerging complexity of protein ubiquitination

David Komander


Protein ubiquitination and protein phosphorylation are two fundamental regulatory post-translational modifications controlling intracellular signalling events. However, the ubiquitin system is vastly more complex compared with phosphorylation. This is due to the ability of ubiquitin to form polymers, i.e. ubiquitin chains, of at least eight different linkages. The linkage type of the ubiquitin chain determines whether a modified protein is degraded by the proteasome or serves to attract proteins to initiate signalling cascades or be internalized. The present review focuses on the emerging complexity of the ubiquitin system. I review what is known about individual chain types, and highlight recent advances that explain how the ubiquitin system achieves its intrinsic specificity. There is much to be learnt from the better-studied phosphorylation system, and many key regulatory mechanisms underlying control by protein phosphorylation may be similarly employed within the ubiquitin system. For example, ubiquitination may have important allosteric roles in protein regulation that are currently not appreciated.

  • deubiquitinase
  • linear ubiquitin
  • lysine linkage
  • NEDDylation
  • phosphorylation
  • SUMOylation
  • ubiquitination


There is hardly any protein in the cell that does not encounter a post-translational modification within its lifespan. The amino acids of proteins are subject to acetylation and methylation altering their charge, to lipidation anchoring them in membranes, or to glycosylation protecting them from hostile environments. Within the signalling cascades that the cell employs to respond to changes in its environment, two post-translational modifications have emerged to be the key players: modification by phosphate groups (phosphorylation) and by small proteins of the ubiquitin family.

Phosphorylation is the reversible addition of a phosphate group to a serine, threonine or tyrosine residue of a substrate protein. This modification was discovered in the mid-1950s, and was soon recognized to be a principal novel way to regulate enzyme function. In 1992, Edmond Fischer and Edwin Krebs received the Nobel Prize in Physiology and Medicine for this discovery [1]. Recent data indicate that the majority of the mammalian cellular protein content may be phosphorylated [2]. Phosphorylation itself changes the electrostatic properties of proteins, enabling conformational changes that can allosterically regulate the modified protein. Alternatively, phosphorylation can induce protein interaction, as it is sensed by specialized protein domains [3]. Phosphorylation is performed by 518 protein kinases in human cells [4], whereas ∼120 protein phosphatases remove phosphate groups. Recent decades have seen an explosion of data as more kinase cascades and networks have been studied genetically, biochemically and structurally.

Protein ubiquitination was discovered in the early 1980s as a post-translational modification, in which lysine residues are modified with the small protein ubiquitin, a 76 amino acid polypeptide of ∼8500 Da. Initially, this modification was found to regulate the half-life of proteins, since a ubiquitinated protein would undergo rapid ATP-dependent degradation mediated by the proteasome. In 2004, Aaron Ciechanover, Avram Hershko and Irwin Rose were awarded the Nobel Prize in Chemistry for their discovery of the ubiquitin system [5]. Ubiquitin is attached via its C-terminal glycine residue to the ε-amino group of a substrate lysine residue. This reaction is performed by a sophisticated three-step enzymatic cascade [6,7]. In an ATP-dependent first step, an E1 ubiquitin-activating enzyme ‘charges’ an E2 ubiquitin-conjugating enzyme with ubiquitin, i.e. the ubiquitin C-terminus is attached to the E2 catalytic cysteine residue via a thioester linkage. Three different types of E3 ligase, namely RING, U-box and HECT (homologous with E6-associated protein C-terminus) domains, act as adaptors that bind both substrate and charged E2. E3 ligases facilitate isopeptide bond formation between ubiquitin and substrate by hardly understood mechanisms. The human genome encodes two E1 enzymes, 37 E2 enzymes and >600 E3 ligases [811] (Figure 1). Ubiquitinated proteins are recognized by at least 20 specialized ubiquitin-binding domains [12]. Protein ubiquitination is reversible, and ∼85 deubiquitinases, specialized proteases that act on ubiquitin, are encoded in our genes [13,14].

Figure 1 Comparison of ubiquitination and phosphorylation systems

Ubiquitination and phosphorylation are reversible processes. The enzymes facilitating and removing the modifications are indicated, and numbers in parentheses indicate the number of human genes encoding the respective proteins. As for protein kinases and phosphatases, E2 conjugating enzymes and deubiquitinases also comprise several pseudogenes that lack catalytic residues and are hence inactive. The roles of inactive deubiquitinases are currently unknown, whereas the inactive E2 enzymes Uev1a and MMS2 have important roles in the assembly of Lys63-linked ubiquitin chains. K, lysine residue; P, phosphorylation; PBD, phosphate-binding domain; S, substrate; S/T/Y, serine/threonine/tyrosine residue; Ub, ubiquitin; UBD, ubiquitin-binding domain.

Forms of protein ubiquitination

The layouts of phosphorylation and ubiquitination have striking similarities, despite the different sizes of the modification. Enzymes able to attach the modification are in ∼5-fold excess of enzymes removing it, and specialized domains recognize proteins when they are modified (Figure 1). Ubiquitination, however, has an additional layer of complexity, greatly increasing its versatility: ubiquitin contains seven lysine residues itself, which can serve as acceptors of further ubiquitin molecules, leading to ubiquitin polymers.

Some proteins are ubiquitinated with only a single ubiquitin, on a single lysine residue (mono-ubiquitination) or on multiple lysine residues (multi-mono-ubiquitination) (Figure 2A). It was found that (multi-)mono-ubiquitination of cell-surface receptors triggers their internalization and subsequent degradation in lysosomes, or recycling to the cell surface [15]. The main mediator of these trafficking mechanisms is the ESCRT (endosomal sorting complex required for transport) machinery. This multimeric protein complex controls the sorting of endosomal cargo proteins, such as ubiquitinated cell-surface receptors, into internal vesicles of multivesicular bodies (reviewed in [16]). Mono-ubiquitination is also involved in the DNA-damage response [17,18], where mono-ubiquitination of histones or of the DNA sliding clamp PCNA (proliferating-cell nuclear antigen) has distinct functionality from the polyubiquitin modification on the same molecules (see below). Furthermore, mono-ubiquitination is being recognized as a crucial ‘priming’ event for the subsequent assembly of polyubiquitin chains, and often distinct E2 enzymes regulate the mono- and poly-ubiquitination of a substrate [1922].

Figure 2 Forms of ubiquitination

(A) The ubiquitin modification has three general layouts: mono-ubiquitination, multi-mono-ubiquitination and polyubiquitination. (B) Forms of homotypic polyubiquitination, where each ubiquitin chain contains a single linkage type. Individual linkages may lead to distinct ubiquitin chain structure, and Lys48- and Lys63-linked/linear chains have different conformations. We do not know structures of the remaining chain types. Multiple homotypic ubiquitin chains on the same substrate are possible. (C) Forms of heterotypic polyubiquitination. In mixed linkages, a ubiquitin chain has alternating linkage types. In branched or forked polyubiquitin chains, a single ubiquitin is extended at two or more lysine residues.

It was recognized early on, that protein degradation via the proteasome is initiated not by mono-ubiquitin, but by a polymeric chain of ubiquitin moieties linked through the Lys48 side chain [23]. Much has been learned since about how Lys48-ubiquitinated substrates are delivered to the proteasome, how the ubiquitin chains are recognized by the 26S proteasome through different ubiquitin binding subunits and how they are edited during the degradation process by deubiquitinases and additional E3 ligases (for a recent review, see [24]). E2 enzymes that exclusively assemble Lys48linkages were described and analysed [25,26], and Lys48-linked chains were thought to be the principal ubiquitin chain type.

Groundbreaking work in 1999/2000 by the late Cecile Pickart as well as Zhijian ‘James’ Chen has put another ubiquitin linkage on the map of ubiquitin research [27,28]. A protein complex was identified, consisting of an E2 enzyme, UBE2N/Ubc13, and one of two ‘pseudo-E2s’ (E2 fold proteins lacking catalytic residues) MMS2 or Uev1a, which together assembled Lys63-linked ubiquitin chains specifically. It was found that this chain type served non-degradative roles, and was involved in the DNA-damage response, as well as in signalling processes leading to the activation of the transcription factor NF-κB (nuclear factor κB) (reviewed recently in [29]).

Homotypic ubiquitin chains linked through the remaining five lysine residues (i.e. a chain only comprising Lys6, Lys11, Lys27, Lys29 or Lys33 linkages) have now also been found in cells (Figure 2B), and the emerging roles for the various ubiquitin modifications are discussed in more detail below.

In addition, it is also possible to assemble ubiquitin chains by head-to-tail assembly of ubiquitin moieties through the α-amino group at the N-terminus (Figures 2 and 3). Such head-to-tail, or linear, ubiquitin polymers are our genetic source of free cellular mono-ubiquitin; four ubiquitin genes in humans encode a total of 14 copies of ubiquitin, which need to be processed by dedicated deubiquitinases into mono-ubiquitin [13]. Linear polyubiquitin chains are also assembled in vivo from mono-ubiquitin, by a dedicated E3 ligase complex, LUBAC (linear ubiquitin chain assembly complex) [30]. Intriguingly, LUBAC functions with several E2 enzymes, one of which, UBE2K/E2-25K, had been shown to be specific for Lys48 chain assembly [26]. It hence appears that LUBAC is an E3 ligase that not only determines the linkage type, but also may override E2-encoded linkage preference. Exciting recent work has revealed unexpected functions of this chain type in signalling processes [3134]. A key difference between linear and lysine-linked chains is the chemistry of the linkage (isopeptide compared with peptide bond), which is important for recognition by deubiquitinases and ubiquitin-binding domains (discussed further below) [34].

Figure 3 Ubiquitin and its lysine residues

The structure of ubiquitin (green, PDB code 1UBQ [153]) reveals that all seven lysine residues (red, with blue nitrogen atoms) reside on different surfaces of the molecule. Met1 (with a green sulfur atom) is the linkage point in linear chains, and is spatially close to Lys63. The C-terminal Gly75-Gly76 motif involved in isopeptide bond formation is indicated (red oxygen atoms, blue nitrogen atoms). Lysine residues are labelled, and red numbers in parentheses refer to the relative abundance of the particular linkage in S. cerevisiae [36]. The proteomic study relied on His-tagged ubiquitin, and hence linear chains could not be determined (ND) [36]. The (tentative) roles of the particular linkages are indicated.

Finally, there is accumulating evidence for ubiquitin chains with mixed linkages, i.e. alternating linkage types in a single ubiquitin chain, and branched or forked ubiquitin chains, where more than one lysine residue of a single ubiquitin within a chain is extended (Figure 2C). Ubiquitin ligases such as CHIP [C-terminus of the Hsc (heat-shock cognate) 70-interacting protein] and Murf1 are highly promiscuous in vitro and, together with non-specific E2 enzymes such as UBE2D/UbcH5, assemble all possible linkages, including branched ubiquitin chains [35]. However, the in vivo abundance or relevance of these heterogeneous chains is not clear (see further discussion below).

Emerging complexity

When MS-based proteomics arrived at ubiquitination, it was found to be an immensely valuable tool to analyse ubiquitinated proteins, identify ubiquitination sites and to distinguish ubiquitin chain linkages. At once, all ubiquitin linkages in the cell could be analysed, and the results of this analysis were surprising [36,37]. In a recent quantitative study, the entire ubiquitin content of yeast was analysed, and relative abundances of ubiquitin–ubiquitin linkages were determined [36]. It was found that Lys48 linkages were the most abundant ubiquitin chain type, but accounted for only 29% of all ubiquitin linkages in the cell. Lys63 linkages accounted for another 17% of all linkages. Interestingly, Lys11 linkages were found to be almost as abundant as Lys48 linkages (28%), and Lys6, Lys27, Lys29 and Lys33 linkages followed with relative abundances of 11, 9, 3 and 3% respectively (Figure 3) [36].

MS has also shown that ubiquitin polymers can have more complex topologies. Ubiquitination of CyclinB1 comprises multiple mono- and short poly-ubiquitin chains of different linkages on the termini of the protein, which is sufficient to induce CyclinB1 degradation [37]. Also, doubly modified ubiquitin peptides indicative of branched ubiquitin chains have been detected. Ubiquitination of close-by ubiquitin lysine residues (Lys6/Lys11, Lys27/Lys29 or Lys29/Lys33) is a product of in vitro ubiquitination reactions [35], and Lys27/Lys29 forks have been detected in yeast cells [38]. At present, it is not clear just how complex the ubiquitination system is.

UBL (ubiquitin-like) modifications

To increase complexity further, ubiquitin is only one of 17 UBL molecules in mammalian cells, and this zoo of modifiers contains other well-studied molecules such as SUMO (small ubiquitin-related modifier) 1/2/3, Nedd8 (neural-precursor-cell-expressed developmentally down-regulated 8) and ISG15 (interferon-stimulated gene 15). All UBLs share the canonical ubiquitin fold, yet some UBLs, such as ISG15 and FAT10, possess two ubiquitin folds in tandem. A number of UBLs have been shown to modify proteins directly, employing very similar enzymatic machineries (reviewed recently in [39,40]). Atg8 family members {GABARAP [GABAA (γ-aminobutyric acid A) receptor-associated protein] 1/2/3 and MAP1LC3 (microtubule-associated protein 1 light chain 3) -A/-B/-C in humans}, modify lipids [41], and are involved in autophagy processes [42]. The UBL protein URM1 (ubiquitin-related modifier 1) is involved in sulfur transfer ([43,44] and reviewed in [45]).

Some UBL modifications, including SUMO2/3 and Nedd8 also form polymers. SUMO2 and SUMO3, but not SUMO1, are readily assembled into chains by the SUMO E2 conjugating enzyme Ubc9. Of the eight lysine residues in SUMO, Lys11 in SUMO2 and SUMO3 contains a SUMOylation consensus sequence, and SUMO chains from mammalian cells are predominantly Lys11-linked [46,47]. UBL modification can compete with ubiquitination if the same lysine residue is attacked. SUMOylation of IκB (inhibitor of NF-κB) can occur at the same lysine residue that mediates signal-induced polyubiquitination and degradation [48]. Similarly, MDM2 (murine double minute 2) SUMOylation prevents its autoubiquitination and degradation [49]. In other signalling molecules, such as PCNA, or the IKK (IκB kinase) subunit IKKγ/NEMO (NF-κB essential modulator), SUMOylation and ubiquitination have distinct signalling roles, not linked to degradation [21,50,51].

Two striking reports have recently identified a SUMO-directed ubiquitin ligase, RNF (RING finger protein) 4, which will polyubiquitinate SUMO chains [52,53]. It is to be anticipated that more UBL/ubiquitin cross-talk varieties exist; the possibilities of the ubiquitin/UBL system seem endless.

Studying polyubiquitination

The last decade has seen hundreds of papers published on function, structure and physiological roles of Lys48- and Lys63-linked chains in ubiquitination; however, only few reports deal with the remaining chain types, some of which have not been studied at all. A likely reason for this is the fact that currently only Lys48- and Lys63-linked chains are readily available, whereas synthesis of the remaining chain types has not been achieved. Below I review approaches for the studies of polyubiquitination currently available.

Mass spectroscopy

As mentioned above, MS is an invaluable technique to study polyubiquitination. The C-terminal sequence of ubiquitin conveniently reads Leu73-Arg74-Gly75-Gly76. Trypsin, which cleaves after lysine and arginine residues, leaves the Gly-Gly tail attached to the modified lysine residue. Furthermore, it will no longer cleave after the modified lysine residue itself. Hence, a tryptic peptide digest of a ubiquitinated protein has two signatures. First, it will have lost one cleavage site, as the modified lysine residue is no longer recognized by trypsin; secondly the same lysine residue has an additional mass corresponding to the Gly-Gly motif (114 Da), which can be readily resolved by MS [38,54]. This technique has revealed the chain types assembled by the HECT E3 ligase KIAA10 (Lys29 and Lys48 linkages) [55], by the Rsp5 HECT ligase (Lys63 linkages) [56], by the BRCA1 (breast-cancer early onset 1)/BARD1 (BRCA1-associated RING domain 1) E3 ligase complex (Lys6 linkages) [57] and by Pellino RING ligases (Lys63 linkages) [58].

Challenging for MS is the analysis of branched or forked ubiquitin chains, in which one ubiquitin in a polymer is extended at two lysine residues. Tryptic digests will only reveal doubly modified peptide for a Lys6/Lys11 branch, a Lys27/Lys29 branch or a Lys29/Lys33 branch (due to other lysine and arginine residues in the ubiquitin sequence); however, such peptides have been found in vitro and in vivo [35,38]. Also mixed chains, in which alternating linkages exist in the same polymer (Figure 2), cannot be identified unequivocally by current MS techniques [59,60].

Generation of free ubiquitin chains in vitro

Free, i.e. unattached, ubiquitin chains of defined linkages are important tools to understand ubiquitin chain structure, as well as chain-type-specificity of deubiquitinases and ubiquitin-binding domains [34]. Protocols for generation of Lys48- and Lys63-linked ubiquitin chains have been published [6264]. Lys63-linked chains can be generated by the E2 enzyme UBE2N/Ubc13, which, together with Uev1a or MMS2, will generate unattached ubiquitin chains that can be purified to homogeneity [62,63]. The E2 enzymes UBE2R1/Cdc34 or UBE2K/E2-25K assemble free Lys48-linked chains in vitro [64,65]. Linear chains can be generated by molecular biology techniques, by cloning one ubiquitin sequence in front of another [66]. Lys29-linked chains have been made by HECT ligase-mediated assembly, using the HECT domain of KIAA10 together with the promiscuous E2 enzyme UBE2D1/UbcH5 and a K48R ubiquitin mutant [67]. However, Lys29-linked or other chain types have not yet been made in large enough quantities to study them substantially. It is a major challenge in the field to identify the enzymes making the remaining ubiquitin chain types.

Ubiquitin mutants

Mutants of ubiquitin are widely used to study the linkage type of a ubiquitination reaction in vitro and in vivo. Two complementary sets of lysine mutants are available. In the first set, each ubiquitin mutant has one of seven lysine residues mutated to arginine (i.e. single point mutants). The second, complementary set, has ubiquitin mutated at all lysine residues except one (i.e. six lysine mutations per mutant) [68]. Together with a lysine-less mutant (all seven lysine residues mutated to arginine) and methylated ubiquitin (in which also the N-terminal α-amino group in blocked), the entire set comprises 16 ubiquitin mutants. Using this set, it was shown that IKK activation depends on Lys63 of ubiquitin [27], and that LUBAC assembles linear chains [30].

Ubiquitin chains assembled from ubiquitin mutants that comprise a single lysine residue only, are available commercially for the study of, e.g., deubiquitinase specificity. In vitro studies using such mutant ubiquitin chains, however, is prone to artefacts, as the ubiquitin surface has been altered significantly by mutating the six lysine residues. This might affect recognition by deubiquitinases and ubiquitin-binding domains, and ubiquitin chain quarternary structure may be affected.

Moreover, the use of transiently expressed ubiquitin mutants in cells has several caveats. Ubiquitin is very abundant in mammalian cells (estimated concentration 20 μM). Ubiquitin-knockout, -knockdown or -knockin approaches are difficult due to the four ubiquitin genes that need to be targeted, the intrinsic high stability of ubiquitin protein, and lethality induced by loss of ubiquitin. Hence, wild-type ubiquitin is always present, and will be incorporated together with mutant ubiquitin into chains. Such level of genetic interference is currently only possible in Saccharomyces cerevisiae, where this strategy has recently led to an amazing number of insights regarding abundance (see above), and functions (see below) of ubiquitin linkages [36].

A second important point is that not all ubiquitin mutants behave like wild-type ubiquitin. This is illustrated by the fact that, when Lys6 of ubiquitin is mutated and expressed in cells, high-molecular-mass polyubiquitinated species accumulate without being degraded, showing that incorporation of Lys6 mutant ubiquitin into chains somehow blocks proteasome function [69].

Chain-specific antibodies and affinity reagents

An important advance is the development of linkage-specific antibodies against Lys63, Lys48 and linear linkages, which will be useful to better understand the in vivo involvements of these chain types [31,70,71]. Antibodies against the remaining chain types are clearly required, yet again due to the high abundance of ubiquitin, more sophisticated phage display technologies will have to be employed to achieve this aim.

Finally, ubiquitin-binding domains and deubiquitinases exist which have remarkable intrinsic specificity in vitro [34,7274]. Although such in vitro analyses suffer greatly from unavailability of the remaining chain types, recent structural studies have further verified and rationalized the observed specificities [32,62,64,75,76] (see below). Hence, specific ubiquitin-binding domains and deubiquitinases can be used as linkage-specific analytical reagents, and may become valuable tools to study the ubiquitin system.

Ubiquitin chain structure

It is remarkable that many enzymes specifically assemble and disassemble only particular linkages. Furthermore, specific ubiquitin-binding domains have been described that can distinguish between chain types. This strongly suggests that different linkages are regulated and recognized independently.

However, this poses the question: how can proteins distinguish between polymers of equivalent units, i.e. ubiquitin? Linking two ubiquitin molecules via an isopeptide bond results in ubiquitin dimers with the same mass and charge, no matter which lysine residue was used. The answer to this question lies in the structure of the ubiquitin chain itself.

Ubiquitin has a globular structure (Figure 3), which is stable at high temperatures and in extremely acidic conditions. Important features of this structure are the C-terminal Leu-Arg-Gly-Gly motif, which extends from the core and is flexible, and a prominent hydrophobic surface patch centred on Ile44, which is commonly used for interaction with ubiquitin-binding domains and deubiquitinases [12,13]. Importantly, all lysine residues of ubiquitin reside on different surfaces of the molecule, and point in different directions (with the exception of the N-terminal amino group of Met1, which is spatially close to Lys63) (Figure 3).

Structural analysis of the available chain types has shown that Lys48- and Lys63-linked chains have entirely different structures. A number of studies have analysed isolated Lys48 dimers and tetramers by crystallography and NMR methods [7784]. Lys48-linked chains adopt a compact fold under physiological conditions. The hydrophobic Ile44 patches of two linked ubiquitin moieties interact, and the linking residues are closely packed against the ubiquitin units. A tetramer of Lys48-linked ubiquitin forms a pseudo-tetragonal structure, where contact surfaces exist between each molecule (Figure 4A). However, Lys48-linked chains are not rigid, but are in dynamic equilibrium and can also adopt more open conformations [83,84]. Furthermore, UBA (ubiquitin-associated) domains can insert between the interacting Ile44 patches (see below and Figure 5A). Also in the process of deubiquitination, a Lys48-linked chain has to be partly unfolded, as deubiquitinases closely contact the linkage region during hydrolysis.

Figure 4 Structures of ubiquitin chains

(A) Structure of Lys48-linked tetra-ubiquitin (PDB code 2O6V [82]). Proximal (white) and distal (black) molecules are labelled. Proximal/distal describes the position relative to the substrate, see cartoon on the left. In Lys48-linked chains, all ubiquitin molecules interact with each other, and the Ile44 patches are not exposed. (B) Lys63-linked ubiquitin chains display an open conformation, both in the crystal structure (PDB code 2JF5 [34]) and in solution [81,85]. The Ile44 patches (shown as blue surface on the molecules) are exposed, and can adopt different relative positions due to the flexibility in the ubiquitin chain. (C) Linear ubiquitin chains display an equivalent conformation to Lys63-linked chains in the crystal structure, as both molecules crystallize under similar conditions in the same crystallographic setting (PDB code 2W9N [34]). The chemical differences in the linker may account for differences, especially during hydrolysis by deubiquitinases [34].

Figure 5 Structures of Lys48, Lys63 and linear chains bound to proteins

(A) Structure of Lys48-linked di-ubiquitin (left, PDB code 1AAR [77]), and complex with the Lys48-specific UBA2 domain of hHR23A (middle, PDB code 1ZO6, [88], co-ordinates kindly provided by D. Fushman) and with the UBA domain of Mud1 (right, [64], co-ordinates kindly provided by J.-F. Trempe). The MUD1 complex is a docking model derived from chemical shift perturbations of the ubiquitin surfaces as intermolecular NOEs (nuclear Overhauser effects) could not be obtained. Both UBA domains insert into the Lys48 dimer to interact with both Ile44 patches. The structures were superimposed on the proximal (white) ubiquitin molecule. (B) The extended conformation of the Lys63 di-ubiquitin (left, PDB code 2JF5 [34]) is exploited by the deubiquitinase AMSH-LP which maximally extends the linker region (second from left, PDB code 2ZNV [76]). The tandem UIM domains of RAP80 also interact with an extended conformation of the Lys63-linked chains, and contact both Ile44 patches, highlighting the rotational flexibility of the Lys63-linked chain (second from right, PDB code 3A1Q [75], co-ordinates kindly provided by S. Fukai). In contrast, interaction of Lys63 di-ubiquitin with an antibody Fab fragment kinks the di-ubiquitin molecule, to interact with the linkage region (right, PDB code 3DVG [71]). (C) The extended conformation of linear chains (left, PDB code 2W9N [34]) is kinked upon interaction with the NEMO UBAN domain (middle and right, PDB code 2ZVN [32]). The crystallized NEMO construct containing the UBAN domain forms a symmetrical elongated α-helical dimer (middle and right, PDB code 2ZVN [32]). The crystal structure (right) revealed a 2:2 complex (two di-ubiquitin molecules bind two NEMO molecules). For clarity, the middle image only shows one linear di-ubiquitin to indicate the kinked conformation, compared with the unbound di-ubiquitin (left, molecules superimposed on the proximal ubiquitin).

Lys63-linked chains have an entirely different three-dimensional structure compared with Lys48-linked chains [34,81,85] (Figure 4B). Lys63-linked chains are extended and adopt an open conformation, where the individual ubiquitin molecules are not interacting with each other. The open conformation of Lys63-linked chains results in freely accessible Ile44 patches. Ubiquitin molecules in a Lys63-linked chain can rotate without restraints, allowing the Ile44 patches to adopt to ubiquitin-binding domains {as shown recently in the RAP80 (receptor-associated protein 80) complex structure [75]; see below and Figure 5B}. Owing to the unrestrained flexibility of the linker, Lys63 chains can also adopt more compact conformations (as seen in the antibody complex; see below and Figure 5B).

The linkage residue Met1 in linear ubiquitin chains is spatially close to Lys63, and hence linear ubiquitin chains adopt the same open conformation as Lys63-linked chains in protein crystals [34] (Figure 4C). The major difference between these chain types is the chemistry of the linkage residues, as a peptide bond in linear chains, with a methionine side chain branching off, is conformationally more restrained compared with a more flexible lysine-linked chain [34].

At present, we do not know the structures of the remaining ubiquitin chains, and this is an important challenge for future research. The remarkable differences between Lys48- and Lys63-linked chains are reflected in their degradative compared with non-degradative roles (see below), and it is easy to comprehend how proteins and deubiquitinases will interact with one chain type, but not another, as both chain types present different surfaces and environments for interaction. As discussed below, the remaining chain types have different roles in cells. Hence, it is to be anticipated that the remaining five chain types also adopt distinct conformations, in order to serve their apparent non-redundant cellular functions.

It is striking that some ubiquitin-binding domains can even distinguish between the conformationally identical linear and Lys63-linked chains, as we have shown recently [34]. This implies that not only overall conformation, but also more subtle differences in chain topology and linker sequence, are being utilized by the sophisticated ubiquitin-binding domains now emerging.

Structural insights into specific ubiquitin chain recognition

Ubiquitin chain recognition is mediated by the deubiquitinases that hydrolyse these chains, but also by at least 20 types of ubiquitin-binding domain. These domains are generally small (<100 amino acids) and, in the simplest case, just consist of a single ∼20 residue helix (ubiquitin-interacting motifs, UIM and MIU). Despite the fact that ubiquitin exists predominantly in polymers within cells, most deubiquitinases and ubiquitin-binding domains have only been characterized structurally in complex with mono-ubiquitin. We have only recently come to appreciate the first complex structures of di-ubiquitin bound to proteins.

Recognition of Lys48-linked chains

UBA domains are ∼40 amino acids long and form a three-helix bundle. These versatile domains are commonly found in ubiquitin-interacting proteins, as they show a wide variety of specificities. In fact, Lys48-specific, Lys63/linear-specific and non-specific members have been identified [86,87]. NMR analysis of two Lys48-specific UBA domains bound to di-ubiquitin has been reported [64,88]. These UBA domains can insert between the apparently tight Lys48 di-ubiquitin interface, and hence interact with the Ile44 patch of both ubiquitin molecules [64,88] (Figure 5A). In contrast, recent work has shown that the two UIM motifs of the proteasomal S5a subunit bind to a more open conformation of Lys48 polymers [89].

Recognition of Lys63-linked chains

Three crystal structures of Lys63-linked di-ubiquitin bound to the deubiquitinase AMSH-LP {AMSH [associated molecule with the SH3 (Src homology 3) domain of STAM (signal-transducing adaptor molecule)]-like protein} [76], to the tandem UIM domains of RAP80 [75], and to a Lys63-specific antibody fragment [71] have been published in the last year (Figure 5B). The Lys63-specific deubiquitinase AMSH-LP in complex with Lys63 di-ubiquitin showed how this enzyme exploits the open conformation of Lys63-linked chains. The ubiquitin dimer is extended maximally, and the linkage region forms extensive contacts with the protein (Figure 5B). To achieve specificity further, AMSH-LP also contacts the Lys63 neighbouring residues, Gln62 and Glu64; this sequence context is only present around Lys63, but not other lysine residues in ubiquitin [13,76].

The tandem UIM domain of RAP80 has also been crystallized with Lys63-linked di-ubiquitin (Figure 5B). RAP80 and several other proteins contain two UIM motifs, separated by a linker of evolutionarily conserved length. This places the UIMs at a particular distance, allowing Lys63-chain interaction with significantly higher affinity compared with linear or Lys48-linked chains [75,90] (Figure 5B). Consistently, increasing or shortening of the linker abrogated this selectivity [75,90]. Different linker length may confer selectivity for other ubiquitin chain types, and the two-residue linker between the tandem UIMs of ataxin-3 confers Lys48-specificity [90]. The crystal structure also explained the inability of RAP80 to interact with linear ubiquitin chains, as the Met1 group cannot be reached by the C-terminus of the second molecule without distortion of the binding modules [75].

Intriguingly, a structure of a Lys63-specific antibody Fab fragment bound to Lys63-linked di-ubiquitin has revealed a different non-elongated structure for a Lys63-linked ubiquitin chain [71]. The ubiquitin molecules are bent along the flexible linker so that the linker region is exposed to interact with the antibody directly. This highlights the intrinsic flexibility of this chain type, and future structures of Lys63-linked chain in complex with proteins will probably reveal further novel interaction modes.

Recognition of linear ubiquitin chains

We and others have recently shown that UBAN domains [ubiquitin-binding domains found in ABINs (A20-binding inhibitors of NF-κB) and NEMO] selectively bind linear chains [3234,91]. NEMO was originally identified to bind selectively to Lys63-linked ubiquitin chains [92,93]; however, recent reports have shown biophysically that the affinity of NEMO for linear chains is significantly higher [32,33,91], and that, in competition experiments, NEMO will select only linear chains from a mixture of linear, Lys63- and Lys48-linked chains [34]. A crystal structure of the NEMO UBAN in complex with linear di-ubiquitin has revealed the molecular basis for UBAN specificity (Figure 5C) [32]. The region of NEMO comprising the UBAN domain forms a parallel symmetric coiled-coil dimer, and the UBAN domain of each NEMO molecule interacts with one linear di-ubiquitin molecule (Figure 5C). The linear di-ubiquitin is kinked along the linker, and NEMO contacts the residues involved in the linkage directly, thus explaining the specificity for linear chains. Linear ubiquitin binding to NEMO also induces a conformational change in NEMO [32]. The entire NEMO protein is an elongated coiled-coil molecule, and a conformational change might allosterically activate the interacting IKK subunits [32,94]. UBAN domains are also found in ABIN proteins and optineurin, and ABIN1 and ABIN2 have also been shown to preferentially interact with linear chains [32,34].

Physiological roles for protein ubiquitination

It is now clear that the different ubiquitin chain types may adopt distinct conformations that enable interactions with linkage-specific ubiquitin-binding domains and deubiquitinases. Consistently, different ubiquitin linkages have been implicated in distinct physiological processes (Figure 3), although the roles of atypical chains (i.e. linkages other than Lys48 or Lys63) are still elusive, and some chain types have hardly been studied. Each linkage type is discussed below according to its abundance in S. cerevisiae (Figure 3).


Lys48-linked chains are the canonical form of polyubiquitin, and, for a long time, were believed to be the principal chain type. Lys48-linked ubiquitin tetramers are the minimal recognition motif for the proteasome [65], and a Lys48-ubiquitinated substrate will be degraded within minutes in cells. The proteasome contains a variety of ubiquitin receptors and deubiquitinases, as well as some ubiquitin ligases, which edit ubiquitin chains [24]. Ubiquitin itself is not degraded, but is recycled by the action of proteasome-associated deubiquitinases [13]. The complex proteasomal machinery is the focus of much research, and has been reviewed extensively (most recently in [24]).

In the last few decades, degradative ubiquitination of proteins was commonly associated with Lys48 linkages without any validation of the chain type. The abundance of atypical linkages, especially of Lys11 chains that also are degradative signals, suggests that many earlier reports might have missed interesting uncommon ubiquitination events.


Despite being as abundant as Lys48 in S. cerevisiae, there are only few reports dedicated to Lys11-linked chains. Early work, and more recent work in yeast and mammalian cells, unambiguously shows that Lys11 chains serve as potent proteasomal degradation signals [36,95,96]. Proteins from diverse cellular processes were identified as being modified and regulated by Lys11-linked chains, suggesting that Lys11 is employed in many different pathways [36]. For example, Lys11-linked chains have been implicated in ERAD (endoplasmic-reticulum-associated degradation). Yeast Ubc6, an important E2 in ERAD, was proposed to mediate assembly of most Lys11 linkages [36]. In human cells, Lys11-linked ubiquitin chains co-purified with a number of UBA/UBX family proteins [97]. UBX domains interact with the AAA (ATPase associated with various cellular activities) protein Cdc48/p97 [98], another important regulator of ERAD [99], and other ubiquitin-mediated processes [100,101].

Interesting work by Rape and colleagues has revealed a role for Lys11-linked chains in the mammalian cell cycle [96]. The key E3 ligase in this process, the APC/C (anaphase-promoting complex/cyclosome), assembles Lys11-linked chains on substrates that need to be degraded during the cell cycle, underlining the degradative potency of this chain type. The E2 enzyme UBE2C/UbcH10 functions with the APC/C to make short Lys11 chains specifically [96]. There is no yeast orthologue of UBE2C/UbcH10, and the yeast APC/C was shown to make Lys48 polymers using the E2 enzymes Ubc1 and Ubc4 [20]. Whether Lys11 chains are involved in cell-cycle regulation in yeast is currently not clear, but it appears that this essential cellular process employs different types of ubiquitin chains in yeast and higher eukaryotes.


A significant body of work on this chain type has now identified important non-degradative roles for Lys63 polymers in endocytosis, in the DNA-damage response and in cell signalling (reviewed recently in [29]).

Endocytic processes, especially the down-regulation of cell-surface receptors, rely on recognition of ubiquitinated receptors by the ESCRT machinery [16]. Initially, mono-ubiquitination was implicated in these processes (see above); however, it has now become clear that many cell-surface receptors are Lys63-ubiquitinated before internalization [102105]. Evidence for Lys63 involvement in the endocytic pathway also comes from the identification of an endocytosis-regulating deubiquitinase, AMSH, which has exquisite intrinsic Lys63 specificity [73,76].

Lys63-ubiquitination is also intimately involved in the DNA-damage response. Early data in yeast showed that mutating endogenous ubiquitin at Lys63 to arginine sensitized cells to DNA damage [28]. Elegant work in yeast has subsequently shown that ubiquitination of the DNA sliding clamp PCNA determines how cells replicate past damaged DNA ([21,106] and reviewed in [107]). Mono-ubiquitinated PCNA triggers translesion synthesis, which is an error-prone way to replicate DNA. However, if the ubiquitin chain on PCNA is extended with Lys63 linkages, an error-free template switching pathway will be initiated [107]. Also, histones are ubiquitinated. In human cells, histone H2A was shown to be modified with Lys63-linked chains after DNA damage, mediated by the E3 ligases RNF8 and RNF168 [108111]. This orchestrates the recruitment of DNA-repair proteins, including 53BP1 (p53-binding protein 1) and BRCA1 (reviewed in [112]). Histones and PCNA are unlikely to be the only targets of ubiquitination at sites of DNA damage, and this is an active area of research.

The best-understood role of Lys63-linked chains, however, is in cytokine signalling (reviewed in [113]). Stimulation of cells with cytokines such as TNFα (tumour necrosis factor α) eventually leads to the activation of two protein kinase complexes, the TAK1 [TGFβ (transforming growth factor β)-activated kinase 1] and IKK complexes, and eventually to the activation of the NF-κB transcription factor. TAK1 can activate IKK, and IKK phosphorylates IκB, leading to degradation of IκB (dependent on Lys48 linkages) and subsequent activation of NF-κB. It was shown that activation of IKK by TAK1 is dependent on Lys63-linked polymers, which are assembled upon receptor stimulation by TRAF (TNF receptor-associated factor) E3 ligases [27]. Many molecules within the receptor complexes, including TRAFs themselves, but also RIP1 (receptor interacting protein 1), IRAKs (interleukin-1 receptor-associated kinases) and components of the TAK complex, are targets for Lys63-polyubiquitination [27,58,92,114]. The TAK1 complex assembles at this scaffold of Lys63-polymers, mediated by the Lys63-specific ubiquitin-binding domains of its subunits TAB (TAK1-binding protein) 2 or TAB3 [74,115]. This oligomerization is thought to trigger cross-phosphorylation and activation of TAK1; however, it is also possible that ubiquitin plays an active allosteric role in TAK1 activation. Also the NEMO subunit of the IKK complex was shown to bind Lys63 chains [92,93]; however, recent data suggest involvement of linear ubiquitin chains at the level of NEMO (see below).

Like for endocytosis, identification of a Lys63- (and linear) specific deubiquitinase, the tumour suppressor CYLD (cylindromatosis), underlined the non-degradative role of polyubiquitination in cytokine signalling. Interestingly, Lys63-specific deubiquitinases have also been identified in the Wnt signalling pathway {TRABID (TRAF-binding domain) [116]} and in the IRF (interferon-response factor) pathway {DUBA (a deubiquitinase), [117]}, suggesting further important roles of Lys63-linked chains in cell signalling.


A heterodimeric RING E3 ligase complex consisting of the breast cancer-susceptibility protein BRCA1 and BARD1 was shown to mediate Lys6-linked polyubiquitination [57,118,119]. However, the physiological roles of this chain type are currently unclear. Individuals who carry mutations in the BRCA1 gene are predisposed to early-onset breast and ovarian cancer [120]. BRCA1/BARD1 are localized at sites of DNA damage, through binding of their adaptor protein RAP80 (see above) to Lys6- and Lys63-linked ubiquitin chains [121]. Hence, both chain types may be involved in DNA repair.


The three ubiquitin lysine residues closest in sequence space, Lys27, Lys29 and Lys33, prove to be a challenge for MS (owing to small tryptic digest fragments), which might be a reason that their roles are more elusive. No cellular role has been associated with Lys27-linked polymers, although this linkage corresponds to ∼10% of all ubiquitination events, at least in yeast [36]. A U-box-containing protein, mammalian Ufd2a, was found to catalyse ubiquitin linkages through Lys27 and Lys33 [122]; however, yeast Ufd2 assembles Lys48-linked chains [123]. The Ufd2 protein belongs to a class of E4 ubiquitin chain-elongating enzymes [124].


Three independent reports have associated Lys29-linked polyubiquitination with HECT E3 ligases. You and Pickart [67] described a HECT E3 ligase, KIAA10, which mediated primarily Lys29- (and to a lesser extent Lys48- and Lys6-linked) ubiquitination in vitro, together with the promiscuous E2 enzyme UBE2D1/UbcH5. The HECT E3 ligase Itch also assembles this chain type on Deltex, a regulator of Notch signalling. Lys29-ubiquitinated Deltex was degraded, by lysosomal rather than proteasomal degradation pathways [125,126]. Lys29-linked polyubiquitin chains were also implicated in the UFD (ubiquitin fusion degradation) pathway, which operates through another HECT E3 ligase, Ufd5 [127]. Attachment of a ubiquitin to the N-terminus of proteins (e.g. fusion with a linear non-cleavable linkage) leads to the extension of the attached ubiquitin with a Lys29- or Lys48-linked chain, resulting in efficient degradation of the fusion protein [127]. Many members of the HECT E3 ligase family are unstudied to date, and it will be interesting to study their specificities and their role in Lys29-linkage assembly.

Several members of the AMPK (AMP-activated protein kinase)-related family of protein kinases [including NUAK1 (NUAK family, SNF1-like kinase 1) and MARK4 (microtubule-affinity-regulating kinase 4)] are ubiquitinated with Lys29- and/or Lys33-linked polyubiquitin chains. A deubiquitinase, USP9x, was identified to regulate this event [128]. Ubiquitination of these kinases was shown to block their activation, as mutants that did not bind to the deubiquitinase were hyper-ubiquitinated and not phosphorylated and activated in cells [128]. Interestingly, MARK kinases contain a UBA domain that interacts directly with the kinase domain [129,130]. This UBA domain does not bind to canonical ubiquitin chains with notable affinity [131,132]. Whether Lys29/Lys33 ubiquitination blocks the activating phosphorylation directly, whether the UBA domain plays a role in this regulation and to what extent the ubiquitinated kinases are degraded are not clear at the moment.

Linear chains

Linear chains are the source of cellular mono-ubiquitin, which is translated as a polyubiquitin precursor and processed post-translationally by specialized deubiquitinases [13]. Linear chains can also be assembled on protein substrates. In 2006, Iwai and colleagues described the LUBAC E3 ligase complex, consisting of two RING domain proteins, HOIL-1 (haem-oxidized iron-regulatory protein 2 ubiquitin ligase-1) and HOIP (HOIL-1-interacting protein), which together exclusively assemble linear ubiquitin chains (see above) [30]. Mouse genetics have recently described a physiological role for linear polyubiquitin chains in the NF-κB pathway [31,133]. The IKK complex component NEMO, an important regulator and adaptor molecule, was found to be the target of linear ubiquitination, and this modification was shown to be required for NF-κB activation [31]. As mentioned above, NEMO itself binds linear chains specifically [3234,91], and NEMO binding to linear chains is important for downstream signalling and NF-κB activation ([32] and reviewed recently in [133]).

Mixed and branched ubiquitin linkages

In vitro ubiquitination is commonly promiscuous, and chains of alternating linkages may be formed readily [35]. It is currently difficult to estimate the abundance of mixed-linkage ubiquitin chains in vivo. The SUMO-directed ubiquitin ligase RNF4 might indicate a common principle, in that it might be sufficient to target ubiquitin ligases (such as Lys48-specific systems) through ubiquitin-binding domains to ubiquitin chains of a different linkage. This could be considered as a form of ubiquitin chain editing.

Branched ubiquitin chains, i.e. where one ubiquitin is extended at two or more lysine residues, are not processed efficiently by the proteasome [35], questioning the in vivo abundance and the roles of branched chains. Interestingly, addition of a ubiquitin-binding protein, the proteasomal subunit S5A, to an in vitro reaction commonly producing branched chains, decreases branched conjugates significantly [134]. This has wide implications, as it suggests that cells furnish E3 ligases with chaperones that prevent formation of branched chains [134].

Using ubiquitin mutants, Ben-Saadon et al. [135] have found evidence for branched ubiquitin chains. The E3 ligase complex Ring1B/BMI1 assembled Lys6/Lys27/Lys48 mixed linkages on itself through auto-ubiquitination. All three lysine residues were required on ubiquitin for this reaction to be as processive as with wild-type ubiquitin. Interestingly, auto-ubiquitination was required for the efficient in vitro mono-ubiquitination of the physiological target of the ligase, histone H2A [135]. More work and new methods are required to understand the in vivo abundance and relevance of mixed- and branched-linkage ubiquitin chains.

Consequences of protein ubiquitination

Proteasomal degradation

As outlined above, different chain types have been implicated in distinct cellular processes and signalling pathways. However, the recent proteomic analysis that revealed the high abundance of atypical chain types indicated a common function for these ubiquitin chains: proteasomal degradation. All chain types, except for Lys63-linked polymers, accumulated in yeast cells when the proteasome was inhibited, although to varying degrees. Purified polyubiquitinated substrates encompassing all linkages were efficiently degraded by purified proteasomes in vitro, with Lys48-ubiquitinated substrates being most rapidly degraded [36]. It is debated whether Lys63-linked chains serve as proteasomal signals. Lys63-polyubiquitinated substrates, although still being degraded in vitro, did not accumulate in cells after proteasome inhibition [36]. In contrast, the yeast ubiquitin ligase Rsp5 attaches exclusively Lys63-linked chains to model substrates, which are subsequently readily degraded by proteasomes in vivo [56].

If indeed all ubiquitin linkages may target proteins for proteasomal degradation, the ubiquitin system appears to be very redundant. This poses the question of why sophisticated machineries exist to specifically make particular chain types.

An answer to this conundrum might be that different chain types have distinct regulators upstream of the proteasome. Ubiquitinated proteins are often transported to the proteasome by shuttling factors, i.e. proteins that bind the ubiquitin chains, and that interact with proteasome subunits. If those factors had different affinities for, e.g., Lys11- and Lys48-linked polymers, this may affect the kinetics of degradation and impose a hierarchical order of degradation signals. Similarly, the different ubiquitin receptors on the proteasome itself might distinguish between chain types, and degrade some chains faster than others. A second important layer of regulation may be imposed by deubiquitinases, some of which are intrinsically specific for particular linkages [13,34]. Specific deubiquitination may also change the kinetics of degradation: if some ubiquitin linkages are more resistant to deubiquitination, such signals may be ‘one-way tickets’ to the proteasome. In contrast, ubiquitin chains prone to hydrolysis may not be stable enough to facilitate proteasomal degradation. Such chains may, however, serve allosteric functions, such as attracting high-affinity binders that modify the substrate. It is easy to imagine that some aspects of cellular regulation, such as the cell cycle, should not be regulated by cohorts of deubiquitinases, but instead use a dedicated chain type that is hydrolysed by few tightly regulated enzymes.

It is possible that the atypical chain types have additional non-degradative functions, in analogy to Lys63-polyubiquitination, which is largely non-degradative but can act as proteasomal signal.

Protein interaction

Ubiquitin polymers serve as interaction sites for proteins, and at least 20 different types of ubiquitin-binding domain have been identified [12]. Ubiquitin recognition is the common principle for both degradative and non-degradative ubiquitin functions. The proteasome itself has at least two ubiquitin-binding subunits and, in addition, employs adaptor proteins that carry ubiquitinated proteins [24]. The ESCRT machinery contains numerous ubiquitin binding entities that recognize ubiquitinated cargo [16]. The TAK1 and IKK kinase complexes involved in NF-κB activation bind to ubiquitin chains, and ubiquitin binding induces their activation [113]. Ubiquitination at sites of DNA damage leads to assembly of DNA repair complexes [29]. Interestingly, proteins involved in these processes employ highly specific ubiquitin-binding domains, which recognize Lys48, Lys63 and linear ubiquitin chain types specifically [32,34,75].

Hence, ubiquitin-binding domains mediate specificity in the ubiquitin system. As outlined above, different chain types regulate different aspects of cellular biology, and therefore also the remaining chain types must be recognized specifically by proteins. Through the study of the remaining chain types, novel specific ubiquitin-binding domains are likely to be discovered.

Allosteric protein regulation?

Ubiquitination has the potential to not only induce interactions of proteins, but also allosterically regulate protein function. In principle, covalent attachment of one protein to another, such as modification with ubiquitin or ubiquitin-like molecules, has great potential to rearrange protein loops, induce order in formerly disordered regions or stabilize the active conformation of an enzyme. Furthermore, ubiquitinated proteins often contain ubiquitin-binding domains, and hence may be regulated by cis-interaction and corresponding regulatory conformational changes. Curiously, this role of ubiquitin or ubiquitin-like domains has not been studied in great detail.

Ubiquitination can inactivate proteins without inducing their degradation. The E2 enzyme UBE2R1/Cdc34 is inhibited after ubiquitination of N-terminal lysine residues. This ubiquitination event prevents charging of Cdc34 by the E1 enzyme [136]. The deubiquitinase USP25 is subject to SUMOylation, which inhibits its catalytic activity against ubiquitin chains, probably through steric inhibition imposed by the SUMO molecule [137]. Many deubiquitinases are ubiquitinated themselves and interact with E3 ligases [13]. Whether deubiquitinases may be inhibited by ubiquitination directly has not been studied in detail.

Interestingly, there are also two reports where ubiquitin and UBL modification has an activating effect. Todi et al. [138] have shown that ubiquitination within the N-terminal Josephin deubiquitinase domain of ataxin-3 activates the enzyme. Structurally, in order to become active, a large helical lever has to move to reveal the active site of the Josephin domain ([139–141] and reviewed in [13]). Hence, it is tempting to speculate that ubiquitination stabilizes this open conformation of ataxin-3.

The best understood example of allosteric activation by a ubiquitin-like modification is the activation of cullin SCF (Skp1/cullin/F-box) ligases by NEDDylation. Nedd8 is the most closely related ubiquitin-like molecule (42% identical). Like the other UBLs, it utilizes a specific E1 (UBA3–NAE1 heterodimer) and E2 (UBE2M/Ubc12) enzyme [142]. NEDDylation of cullins was known for some time to activate this large class of ubiquitin E3 ligases [143], but only through recent work by the Schulman laboratory have the structural consequences of Cullin NEDDylation been appreciated [144].

SCF ligases are large multiprotein complexes, which comprise dedicated substrate-binding subunits, e.g. F-box proteins, and also a RING E3 ligase subunit [145] (Figure 6). Early structural work showed that cullins separate the tightly bound RING subunit from a substrate through an extended helical domain [146], posing the steric problem to bridge a gap of ∼50 Å (1 Å=0.1 nm) between E2/E3 and substrate (Figure 6A). Recently, crystal structures of Cul-5 in its un-NEDDylated and NEDDylated forms revealed that, upon NEDDylation, a large conformational change occurs. The attached Nedd8 molecule interacts in cis with a remote Nedd8-binding patch on Cul-5 [144]. These movements release the tightly bound RING E3 ligase subunit, and this newly achieved flexibility allows the E3 to reach and efficiently ubiquitinate the substrate (Figure 6B). Cullin NEDDylation affects SCF-mediated ubiquitination in various ways [147]. It bridges the gap between E3 ligase and substrate, it increases the affinity of SCF for charged E2 enzyme and it stabilizes the transition state, increasing the kcat for the ubiquitination reaction [147].

Figure 6 Allosteric regulation of cullins by NEDDylation

Cullin NEDDylation activates the ubiquitin E3 ligase function of SCF ligases. (A) In the structure of un-NEDDylated Cul-5, the RING subunit Rbx1 is tightly bound. Upon interaction of Rbx1 with Nedd8-charged Ubc12, auto-NEDDylation commences on a nearby lysine residue of Cul-5. (B) NEDDylation induces a large conformational change, as the Nedd8 binds in cis to a remote site of Cul-5. This unleashes Rbx1, which interacts now with higher affinity with the E2 enzyme Cdc34. Release of the RING subunit also allows it to reach the substrate, promoting processive ubiquitination. The models were generated from superposition of Cul5-Rbx1 in un-NEDDylated and NEDDylated states (PDB codes 3DPL, 3DQV [144]) on to the corresponding parts of the SCF core structure of Cul1–Rbx5–Skp1–FboxSkp2 (PDB code 1LDK [146]). Individual subunits are shown in different colours and labelled accordingly, and S indicates the position of the substrate.

The structure of NEDDylated Cul-5 is the first molecular description of a protein modified with ubiquitin-like molecules. To date, no structure of any ubiquitinated protein (i.e. with ubiquitin linked in an isopeptide bond) has been structurally resolved. Therefore we do not currently understand whether ubiquitination of substrates, especially with mono-ubiquitin or non-canonical chains, can have consequences other than creating interaction interfaces. The example of NEDDylated Cul-5, with its large-scale movements induced by the modification, demands more efforts to understand the role of ubiquitin in substrate protein remodelling.

Phosphorylation compared with ubiquitination

Different, yet similar

As detailed above, ubiquitination is much more complex than phosphorylation, mainly due to the ability of ubiquitin to form polymers. However, in a most simplistic view, both phosphorylation and ubiquitination have two principal outcomes.

On one hand, both modifications lead to the generation of new interaction sites, which can be recognized by dedicated protein domains. Owing to the much larger polymeric ubiquitin modification, it is no surprise that the number of structurally unrelated ubiquitin-binding domains exceeds the number of phosphate-binding domains.

On the other hand, phosphorylation has been shown to regulate protein function allosterically. This is most prominent for the activation of protein kinases themselves, many of which require phosphorylation in the activation segment to achieve catalytic activity. The aforementioned example of cullin auto-NEDDylation, leading to a conformational change activating the E3 ligase activity, strikingly resembles this theme. The roles of ubiquitin as an allosteric regulator of proteins are currently underappreciated.

Cross-talk between ubiquitination and phosphorylation

Both ubiquitination and phosphorylation are principal components of many signal transduction processes, and hence it is not surprising that much cross-talk exists between these systems (reviewed in [148]). In many cases, degradative ubiquitination is induced by prior phosphorylation. This is mediated by SCF ligases, which utilize phosphopeptide-binding domain proteins as substrate-binding adaptors. A prominent example is β-TRCP (β-transducin repeat-containing protein), which targets SCF to phosphorylated ‘degron’ motifs and induces substrate degradation by polyubiquitination with Lys48-linked chains [149].

Recent work shows that phosphorylation can directly activate E3 ligase activity. Pellino E3 ligases are activated after phosphorylation by IRAK kinases upon cytokine stimulation [58]. Interestingly, Pellinos also contain a cryptic FHA (forkhead-associated) phospho-binding domain in their N-terminus, which is required for interaction with active (i.e. phosphorylated) IRAK kinases [150]. Hence, it appears that activating phosphorylation of IRAK kinase serves both to attract and to activate Pellino proteins. Pellino will subsequently polyubiquitinate IRAK with Lys63-linked ubiquitin chains, leading to activation of TAK1 and IKK kinases and, eventually, NF-κB. Similarly, it has recently been shown that TRAF2 E3 ligase activity is activated by phosphorylation [151]. Clearly, further examples of cross-talk will be discovered with the ubiquitination system becoming better defined.


Although the complexity of ubiquitination seems challenging, exciting discoveries await once appropriate reagents, such as the atypical chain types themselves, are available. Deregulation of ubiquitin-mediated signalling is increasingly implicated in human diseases such as cancer and infection by pathogens. With a deeper understanding of the multiple roles of ubiquitin in cell signalling, the ubiquitin system will attract further interest from the pharmaceutical industry. Maybe the current focus on protein kinases as drug targets of the 21st Century [152] might shift to the enzymes in ubiquitination cascades soon.

Early Career Research Award Delivered at Appleton Tower, University of Edinburgh on 2 April 2009David Komander


I am grateful for a Beit Memorial Fellowship for Medical Research that funded parts of my postdoctoral work. The Medical Research Council funded my Ph.D. work with an MRC Predoctoral Fellowship, and is now funding research in my laboratory.


I thank the members of my group, especially Masato Akutsu, Anja Bremm, Yogesh Kulathu, Julien Lichesi, Satpal Virdee and Yu Ye, for their shared excitement in this field and for the many stimulating discussions. I am grateful to David Barford, Sonja Flott, Yogesh Kulathu, Anja Bremm and Satpal Virdee for constructive comments on the manuscript. I am also grateful to the many collaborators and colleagues who have provided help and reagents while we are starting to explore the ubiquitin system. I would like further to thank my former supervisors David Barford, Dario Alessi and Daan van Aalten for their continuous support, and in particular David Barford for sparking my interest in the ubiquitin system, and for allowing me to continue many of my postdoctoral projects as a group leader.


  • Early Career Research Award:

Abbreviations: ABIN, A20-binding inhibitor of nuclear factor κB; AMSH, associated molecule with the SH3 (Src homology 3) domain of STAM (signal-transducing adaptor molecule); AMSH-LP, AMSH-like protein; APC/C, anaphase-promoting complex/cyclosome; BRCA1, breast-cancer early onset 1; BARD1, BRCA1-associated RING domain 1; ERAD, endoplasmic-reticulum-associated degradation; ESCRT, endosomal sorting complex required for transport; HECT, homologous with E6-associated protein C-terminus; HOIL-1, haem-oxidized iron-regulatory protein 2 ubiquitin ligase-1; IκB, inhibitor of nuclear factor κB; IKK, IκB kinase; IRAK, interleukin-1 receptor-associated kinase; ISG15, interferon-stimulated gene 15; LUBAC, linear ubiquitin chain assembly complex; MARK4, microtubule-affinity-regulating kinase 4; Nedd8, neural-precursor-cell-expressed developmentally down-regulated 8; NEMO, nuclear factor κB essential modulator; NF-κB, nuclear factor κB; PCNA, proliferating-cell nuclear antigen; RAP80, receptor-associated protein 80; RNF, RING finger protein; SCF, Skp1/cullin/F-box; SUMO, small ubiquitin-related modifier; TAK1, TGFβ (transforming growth factor β)-activated kinase 1; TAB, TAK1-binding protein; TNF, tumour necrosis factor; TRAF, TNF receptor-associated factor; UBA, ubiquitin-associated; UBAN, ubiquitin-binding domains found in ABINs and NEMO; UBL, ubiquitin-like; UIM, ubiquitin-interacting motif


View Abstract