The protein kinase PINK1 was recently shown to phosphorylate ubiquitin (Ub) on Ser65, and phosphoUb activates the E3 ligase Parkin allosterically. Here, we show that PINK1 can phosphorylate every Ub in Ub chains. Moreover, Ser65 phosphorylation alters Ub structure, generating two conformations in solution. A crystal structure of the major conformation resembles Ub but has altered surface properties. NMR reveals a second phosphoUb conformation in which β5‐strand slippage retracts the C‐terminal tail by two residues into the Ub core. We further show that phosphoUb has no effect on E1‐mediated E2 charging but can affect discharging of E2 enzymes to form polyUb chains. Notably, UBE2R1‐ (CDC34), UBE2N/UBE2V1‐ (UBC13/UEV1A), TRAF6‐ and HOIP‐mediated chain assembly is inhibited by phosphoUb. While Lys63‐linked poly‐phosphoUb is recognized by the TAB2 NZF Ub binding domain (UBD), 10 out of 12 deubiquitinases (DUBs), including USP8, USP15 and USP30, are impaired in hydrolyzing phosphoUb chains. Hence, Ub phosphorylation has repercussions for ubiquitination and deubiquitination cascades beyond Parkin activation and may provide an independent layer of regulation in the Ub system.
Ubiquitin phosphorylation by PINK1 kinase is involved in Parkin activation. Structural and biochemical work now shows that it also affects ubiquitin conformation, chain assembly and deconjugation, thus potentially affecting ubiquitin signaling more broadly.
PINK1 phosphorylates Ser65 both in free ubiquitin and in polyubiquitin chains.
Parkin is activated by phospho‐ubiquitin only in the presence of unphosphorylated ubiquitin.
Ser65 phospho‐ubiquitin interconverts between a major, canonical ubiquitin conformation and a previously unknown minor conformation.
Due to beta‐strand slippage, the ubiquitin C‐terminal tail is retracted in the minor conformation.
Ser65 phospho‐ubiquitin is charged onto E2 enzymes, but discharging with and without E3 can be inhibited.
Ser65 phospho‐ubiquitin‐incorporating polyubiquitin chains are less well hydrolyzed by all classes of DUB enzymes.
The covalent modification of Lys residues with the 76 amino acid protein ubiquitin (Ub) constitutes one of the most important cellular signals, most commonly targeting the substrate protein for proteasomal degradation (Hershko & Ciechanover, 1998). The last decade has revealed that ubiquitination is much more versatile and affects virtually all cellular processes, including transcription, translation, protein kinase, cytokine and DNA damage signaling, intracellular trafficking and most forms of protein degradation, such as ERAD, autophagy and mitophagy (Chen & Sun, 2009; Komander & Rape, 2012; Shaid et al, 2013; Randow & Youle, 2014). With such wide‐ranging roles, it is not surprising that imbalances in the Ub system lead to disease, and many proteins involved in assembly, binding or disassembly of Ub conjugates are mutated in human disorders. A prime example is Parkinson's disease, a neurodegenerative disorder characterized by loss of dopaminergic neurons, which affects 1–2% of the human population especially at an older age. Mutations in the Ub E3 ligase Parkin predispose individuals to autosomal recessive juvenile Parkinsonism (AR‐JP), a form of the disease where neurological symptoms show at an early age. Heterozygous mutations of Parkin have also been implicated in the more common, late‐onset form of the disease (Sun et al, 2006; Wang et al, 2008; Corti et al, 2011). Parkin belongs to the family of RBR E3 ligases (Wenzel & Klevit, 2012), and recent structural data revealed that Parkin is autoinhibited and requires activation (Riley et al, 2013; Trempe et al, 2013; Wauer & Komander, 2013). Parkin is activated by the protein kinase PINK1, which is stabilized on depolarized mitochondria upon mitochondrial damage (Youle & Narendra, 2011). Once activated, Parkin ubiquitinates numerous mitochondrial and cytosolic proteins (Chan et al, 2011; Sarraf et al, 2013) including mitofusins and Miro, eventually triggering mitophagy (Youle & Narendra, 2011).
Many substrates of PINK1 and mechanisms for PINK1‐mediated Parkin activation have been postulated (Wang et al, 2011; Okatsu et al, 2012; Chen & Dorn, 2013). Most recently, based on mass spectrometry and following from earlier findings that PINK1 phosphorylates the Parkin Ub‐like (Ubl) domain (Kondapalli et al, 2012), three groups showed that Ub itself is a substrate for PINK1, which phosphorylates Ub Ser65 exclusively, in vitro and in cells (Kane et al, 2014; Kazlauskaite et al, 2014; Koyano et al, 2014). Moreover, purified phosphoUb directly activated Parkin in the absence of PINK1, suggesting a new model of Parkin activation (Sauvé & Gehring, 2014; Zheng & Hunter, 2014) which is still being refined (Ordureau et al, 2014).
Aside from this gain‐of‐function role of phosphoUb in Parkin activation, it is unclear whether phosphorylation has structural consequences for Ub and whether phosphorylation affects other Ub‐mediated processes. These fundamental questions have not been addressed to date.
We here show that PINK1 is a Ub and polyUb kinase and reveal wide‐ranging consequences of Ub Ser65 phosphorylation. These include significant structural changes within phosphoUb wherein the protein interconverts between two conformations in solution, one of which has a retracted C‐terminal tail; this UbretraCT conformation has not been observed to date. While phosphoUb is still activated by E1 and charged onto E2 enzymes, discharging and polyUb chain assembly is impaired in several E2 enzymes and E2/E3 complexes. Moreover, 10 out of 12 tested deubiquitinases (DUBs) hydrolyze phosphoUb chains with significantly lower activity.
In summary, we show that PINK1‐mediated phosphorylation of Ub generates a functionally altered Ub that would be incompetent in various Ub‐regulated processes, consistent with a model that Ser65‐phosphorylated Ub is an independent signal with altered functions in cells.
PINK1 is a Ub kinase
Like others (Kane et al, 2014; Kazlauskaite et al, 2014; Koyano et al, 2014), we had also found that PINK1 is a Ub kinase while investigating the mechanism of Parkin activation (Fig 1A). In an attempt to reconstitute Parkin‐mediated ubiquitination of a reported substrate, Miro (Wang et al, 2011; Birsa et al, 2014), we performed a coupled kinase/ligase reaction with Tribolium castaneum (Tc) PINK1 (Kondapalli et al, 2012), human Parkin lacking the Ubl domain (Wauer & Komander, 2013), GST‐Miro (residues 1–580), E1, UBE2L3 (UBCH7) and wild‐type (wt) Ub. Tandem mass spectrometry on the reaction was searched for peptides containing phosphate incorporation. In these reactions, PINK1 phosphorylated Ub and targeted exclusively Ser65.
We confirmed that in vitro, Pediculus humanus corporis (Ph)PINK1 [related to TcPINK1 (Kondapalli et al, 2012)] mediated phosphorylation of Ub proceeds to completion, while a Ub S65A mutant is not modified, as assessed by Phos‐tag gels (Fig 1B). Reactions with TcPINK1 showed similar results and depended on active TcPINK1 (Supplementary Fig S1A).
To generate large amounts of Ser65‐phosphorylated Ub, 10 mg (1.15 μmol) of Ub was phosphorylated with enzymatic amounts of PhPINK1. Phosphorylation proceeds to completion as assessed by electrospray ionization mass spectrometry (ESI‐MS). Ub phosphorylation changes its isoelectric point from 6.6 to 5.7, enabling purification by anion exchange chromatography (see Materials and Methods). Purity of phosphoUb was confirmed by ESI‐MS (Fig 1C).
PINK1 is a polyUb kinase
While this demonstrated that PINK1 phosphorylates monoUb, it was not clear whether PINK1 could also phosphorylate polyUb chains. Phos‐tag gels of reactions using Lys63‐linked triUb as substrate resulted in multiple bands and diffuse signals, suggesting that more than one Ub molecule can be targeted by PhPINK1 in this chain type (Supplementary Fig S1B, C and E). For further analysis and quantification, we used AQUA‐based mass spectrometry (Kirkpatrick et al, 2006; Ordureau et al, 2014). Digestion of a ubiquitinated protein with trypsin leads to a signature mass addition of 114 Da on the modified Lys, corresponding to the C‐terminal GlyGly sequence of Ub. Spiking of tryptic digest mixtures with isotope‐labeled GlyGly‐modified standard peptides derived from each linkage site allows quantification of all chain types (Kirkpatrick et al, 2006). To account for the presence of phosphorylation at Ub Ser65, AQUA peptides incorporating pSer65 as well as (GG)Lys63/pSer65 were also included (see Supplementary Materials and Methods and Supplementary Table S1). With this, we followed phosphorylation of Lys63 hexaUb over time, revealing that all Ub molecules in the chain were phosphorylated by PhPINK1 within 1 h (Fig 1D). Lys63 chains adopt “open” conformation, while other chain types are more compact (Kulathu & Komander, 2012) and may not expose Ser65 for PINK1 phosphorylation. Still, we found that PhPINK1 phosphorylated all Ub moieties in tetraUb with Met1, Lys6, Lys11, Lys33, Lys48 and Lys63 linkages (Fig 1E and Supplementary Fig S1C–F), consistent with the dynamic nature of polyUb in solution (Ye et al, 2012). Hence, PINK1 is not only a Ub but also a polyUb kinase.
PhosphoUb activates and inhibits Parkin
We confirmed reported roles for phosphoUb in Parkin activation. While full‐length Parkin is inactive, addition of 20% purified phosphoUb led to robust activation in an in vitro auto‐ubiquitination reaction (Fig 1F). Assembled polyUb chains were detected with anti‐Ub antibody (FK2), which recognizes Ub and phosphoUb similarly (Supplementary Fig S1G).
Surprisingly, when phosphoUb was used as the only source of Ub in the reaction, Parkin was hardly active and did not form polyUb. Parkin showed very low activity with 0 or 100% phosphoUb, but polyUb chain generation was triggered by small amounts of phosphoUb in the reaction. Increasing the phosphoUb ratio diminished Parkin activity (Fig 1G). This unexpected behavior has also been reported by the Harper lab (Ordureau et al, 2014). It is known that the phosphoUb‐mediated activation of Parkin proceeds with a Ub mutant that cannot proceed through E1/E2/E3 cascades, pointing toward an allosteric mechanism (Koyano et al, 2014). The fact that phosphoUb is seemingly unable to be assembled into polyUb by Parkin in vitro, suggests that phosphorylation affects Ub structure and/or affects its ability to be passed through the ubiquitination cascade.
Ser65 phosphorylation changes Ub structure
We first assessed biophysical and structural properties of phosphoUb. Ub is dynamic in solution yet generates excellent high‐resolution NMR spectra with typically ~72 resonances (Lange et al, 2008; Vögeli et al, 2009). To understand whether and how Ser65 phosphorylation affects Ub structure, we performed a phosphorylation reaction in situ, by adding PhPINK1/MgATP to 15N‐labeled Ub in a time‐dependent NMR experiment, which led to complete and specific Ser65 phosphorylation as confirmed by ESI‐MS (see Supplementary Materials and Methods). To our astonishment, the 2D‐NMR spectrum of phosphoUb showed 72 resonances that are similar to wt Ub and in addition 58 new resonances (Fig 2A). This suggested that phosphorylation of Ub at Ser65 generates two non‐identical Ub conformations. These results were corroborated using purified 13C, 15N‐labeled phosphoUb and it was possible to assign all resonances (Supplementary Fig S2). PhosphoUb in solution exists as a major species (70% as determined by relative peak intensities) that resembles wt Ub with resonances perturbed in the vicinity of pSer65 (Fig 2B). In addition, we observed a minor species (30%) with significant chemical shift perturbations (CSPs) as compared to wt Ub (Fig 2C) or to the major phosphoUb species (Fig 2D). ZZ‐exchange measurements revealed that the two species are in a slow equilibrium with a conversion rate of ~2 per second (Supplementary Figs S3 and S4).
The second, minor species of phosphoUb has to our knowledge not been observed before. Secondary structure calculations for the major and minor species using TALOS+ (Fig 2E, Supplementary Fig S5) revealed a similar secondary structure to wt Ub, indicating no global conformational changes or significant unfolding of the Ub molecule (Fig 2E) (see below).
To test whether the conformational change was restricted to monoUb, we characterized phosphorylated, 13C, 15N‐labeled Lys63‐linked polyUb, which due to its open conformation displays BEST‐TROSY spectra that are unperturbed by interface formation and are almost identical to Ub (72 peaks; Thr9, Glu24 and Arg74 resonances are exchange‐broadened, while Lys63 and Glu64 resonances are split due to proximal and distal differences for these residues, and one extra peak for the isopeptide bond between the distal Gly76 and the side chain of the proximal Lys63). We found that phosphorylation also generated a minor conformation in uniformly labeled Lys63‐linked di‐, tri‐ and tetraUb (Supplementary Fig S6 and data not shown) with 128 peaks in phosphorylated Lys63‐linked diUb. This indicates that phosphorylation‐induced conformational changes can take place in the context of polyUb. Future work will need to assess whether the conformational changes exist in distal, proximal or both Ub moieties, whether it occurs in all chain types, and whether this leads to changes to polyUb structure.
Phosphorylation affects Ub stability
Ub is highly stable and structurally unaffected by low pH, SDS or high temperatures [see e.g. (Ibarra‐Molero et al, 1999)]. However, a novel conformation of Ub and the observed conformational transition could result in a less stable molecule. Indeed, we found that melting temperatures were decreased by 10°C in differential scanning calorimetry (DSC) experiments, from 93°C for Ub to 83°C for phosphoUb (Supplementary Fig S7). This is consistent with an alternative, less stable conformation being present, but also confirms that neither of these conformations is unfolded at the temperatures of NMR measurements. However, it does suggest that thermal denaturation of phosphoUb occurs earlier, and standard protocols of Ub purification should be used with care (Koyano et al, 2014).
Crystal structure of phosphoUb
The two conformations of Ub were analyzed in more detail. We obtained crystals of phosphoUb and determined a structure to 1.9 Å resolution (Table 1). The new Ub crystal form (space group P1) contains eight phosphoUb molecules in the asymmetric unit (Fig 3A). Six of the molecules interact similarly via their Ile44 patches, resembling structures of Lys48‐linked diUb [pdb‐id 1aar (Cook et al, 1992)] (Fig 3B). Unambiguous electron density is observed for phospho‐Ser65 in all molecules (Fig 3C). The phosphate group forms a hydrogen bond with the backbone amide of Gln62, but no other contacts (Fig 3D). In wt Ub, the same hydrogen bond is formed via the hydroxyl group of Ser65 (Fig 3D). The small CSP for Gln62 in the major NMR species compared to wt Ub (Fig 2B) is consistent with the conservation of the hydrogen bond of this backbone amide. This and the fact that there are no significant conformational differences between the phosphoUb structures in the crystal and a Ub reference structure [pdb‐id 1ubq, RMSDs 0.45–0.69 Å (Vijay‐Kumar et al, 1987)] indicated that the major phosphoUb NMR species had crystallized. While there are no conformational changes, phosphorylation of Ser65 leads to significant changes in the electrostatic surface potential of Ub (Fig 3E), which may alter interactions with other proteins (see below and Discussion).
Characterization of the minor phosphoUb species
As crystallization of phosphoUb did not provide information on the more intriguing minor species of phosphoUb, this was further investigated using a variety of solution NMR techniques. While the significant CSPs of the minor species compared to wt Ub and major phosphoUb species suggest substantial changes, the secondary structure prediction showed only small changes (Fig 2E). It was unlikely that N‐ or C‐terminal regions of Ub were unfolded, as this would lead to loss of β‐sheet character in TALOS+ calculations (Supplementary Fig S5). Additionally, hetNOE measurements were collected to observe changes in local stability of the secondary structure (Supplementary Fig S8). Differences in motion on the pico‐ to nano‐second timescale are usually seen at the termini of proteins, resulting in reduced or negative hetNOE values due to increased flexibility and motion. Resonances of the first β‐strand (aa 2–6) have lower hetNOE values in the minor species as compared to the major species, indicating increased flexibility. Interestingly, Arg74 is significantly stabilized in the minor species. In wt Ub, this residue is part of the Ub C‐terminal tail (aa 74–76) and does not interact with the Ub core but is flexible in solution (Lange et al, 2008). The observed stabilization of Arg74 in the minor species is consistent with our secondary structure prediction that calculates the β5‐strand to be extended by one residue (Fig 2E).
One explanation for the significant CSPs in the minor form and overall intact Ub structure could be dramatic changes in the hydrogen bonding pattern, which would affect the chemical environment of the entire Ub core. This was investigated using long‐range HNCO experiments (Cordier et al, 2008; Nisius & Grzesiek, 2012). In a standard HNCO spectrum, the HN of a residue is correlated with the C′ resonance of the preceding residue via direct one‐bond 1JNC′ coupling. By adjusting the HNCO pulse sequence for prolonged N‐C′ magnetization transfer (total transfer time increased to 133 ms), weak h3JNC′ couplings across hydrogen bonds can be recorded and assigned with reference to a standard HNCO spectrum (Fig 4A). This analysis yielded a hydrogen bonding network for the β‐sheet region of both phosphoUb species (Fig 4B). The major form of phosphoUb has a similar pattern to that of wt Ub published previously (Nisius & Grzesiek, 2012) although some correlations could not be resolved (Fig 4B, compare left and middle). The minor species of phosphoUb maintains similar contacts between β‐strands β1 and β2 and partially between strands β3 and β4, suggesting that these core interactions are unaffected by Ub phosphorylation. Strikingly, the hydrogen bonding pattern of the β‐strand β5 to both β3 and β1 is entirely different (Fig 4B, compare middle and right). None of the contacts of wt Ub or phosphoUb major species are maintained, but a distinct set of contacts are formed between Ile44‐Val70, Arg42‐Arg72, Leu69‐Phe4 and Leu71‐Lys6. These contacts are beyond hydrogen bonding distance in any Ub molecule structurally resolved to date, suggesting that the β5‐strand has shifted with respect to the Ub core to form the observed hydrogen bonds.
These results are corroborated by through space NOE analysis (Supplementary Fig S9), which confirmed that the major conformation had contacts that were consistent with the crystal structure and that the minor conformation had a shifted β5 strand as suggested by the long‐range HNCO experiment.
Ser65 phosphorylation induces β5‐strand slippage in phosphoUb
Closer inspection of Ub structure and sequence reveals that such a β‐strand slippage is entirely feasible. The β5‐strand has four Leu residues (Leu67, Leu69, Leu71 and Leu73), and Leu67 and Leu69 are integral to the Ub core (Fig 4B and C). The intermittent residues His68, Val70 and Arg72 are solvent‐exposed (Fig 4C). Modeling of a β‐strand slippage of two residues, that is, replacing Leu67 with Leu69, Leu69 with Leu71, etc., reveals a model for the minor phospho‐Ub species in which the C‐terminal tail of Ub has retracted into the Ub core by two amino acids (Fig 4D). The generated model is consistent with all data: (i) The four measured long‐range HNCO contacts are now satisfied (Fig 4A, B and D), (ii) the through space NOE contacts are reconciled (Supplementary Fig S9), (iii) retraction of the C‐terminal tail would lead to core contacts and stabilization of Leu73 and Arg74 as observed in the hetNOE experiment (Supplementary Fig S8), (iv) the C‐terminal tail would extend the β5‐sheet, as was predicted in the secondary structure analysis (Fig 2E, Supplementary Fig S5) and (v) all residues in β5, the preceding loop (aa 62–66), in β1 (aa 2–7) and in β3 (Ile44 and Phe45) have changed their environment significantly, explaining the large chemical shift perturbations (Fig 2A–D). The largest CSP in the minor species occurs for Gln62 (Fig 2C), indicating a drastically altered environment around the Gln62 amide, possibly due to loss of the conserved hydrogen bond of the backbone amide (see above).
Retracting the Ub C‐terminus would extend the phospho‐Ser65 containing loop (aa 62–66), which in the hetNOE data appears to be rigid, suggesting that pSer65 forms new interactions on the Ub surface. Likewise, retraction of the important Ub C‐terminus would affect interactions with most enzymes in the Ub system. Clearly, this model for a new Ub conformation requires further structural validation, which could be obtained by identifying binding partners or mutations that stabilize this species (see Discussion).
Together, the data show that phosphorylation has dramatic effects on Ub structure. It changes the surface properties of the major species, and, more intriguingly, generates a previously unobserved Ub conformation (which we will refer to as the UbretraCT conformation). Ser65 phosphorylation leads to a β‐strand slippage by two residues, which is enabled by the sequence of the β5‐strand. These changes in Ub structure may have many repercussions on the Ub system, which relies on highly conserved and regulated interactions of Ub with the many enzymes facilitating its attachment and removal.
Effects of phosphoUb on the Ub system
Ub is almost invariant in evolution, is identical in higher eukaryotes and differs only by one non‐conservative and two conservative mutations between Saccharomyces cerevisiae and humans. PINK1 from all species tested to date phosphorylates the invariant Ub Ser65, suggesting that the gain‐of‐function of phosphoUb in Parkin activation and mitophagy is evolutionarily conserved. However, Ub plays many roles, and it is currently unclear whether phosphoUb has other, potentially detrimental effects on the Ub system. The structural data described above (Figs 2, 3, 4) show that phosphoUb is structurally distinct and could potentially adopt new functions. For example, we posit that the minor species could be recognized by as‐yet unknown phosphoUb binding proteins. It is also clear that in the major phosphoUb species, the main Ub interactions sites, the Ile44 and the Ile36 patches, are structurally intact and extensively utilized in crystal lattice interfaces (Fig 3). Nonetheless, all surfaces of Ub are highly conserved, and most have been implicated in protein interactions during Ub chain assembly, binding and/or disassembly.
Consequences of phosphoUb on E2 charging and discharging
The E1‐mediated charging of E2 enzymes is a sophisticated multi‐step reaction in which Ub is engaged in a variety of ways (Schulman & Harper, 2009). We screened 13 of the 19 human E2 enzymes that are charged by the Ub E1 UBA1, to find that all tested enzymes were charged by Ub and phosphoUb similarly (Fig 5).
The E1‐generated E2~Ub conjugate is relatively stable until discharged by a ligase. In NMR and SAXS ensembles of charged E2 enzymes, Ub is flexibly attached to the E2 and can take many conformations (Pruneda et al, 2011). Importantly, Ser65 is remote from the active site and does not interact with the enzyme (Fig 5B, Supplementary Fig S10A), suggesting that the E2~phosphoUb conjugate may behave similarly to an E2~Ub conjugate.
A subset of E2 enzymes ubiquitinated themselves on one or multiple sites in vitro, and this was in most cases independent of Ub phosphorylation (Fig 5A). We also noted considerable auto‐polyubiquitination of UBE2E1 and UBE2T in our assays. UBE2S and UBE2R1 (cdc34) assemble free Ub chains linked via Lys11 and Lys48, respectively. Interestingly, while UBE2S assembled Lys11‐linked chains regardless of whether Ub was phosphorylated or not, UBE2R1, UBE3E1 and UBE2T were significantly impaired in discharging and chain formation when phosphoUb was used in the reaction (Fig 5A and C, Supplementary Fig S10B–D). This was quantified by AQUA mass spectrometry indicating GlyGly‐modified Ub peptides in E2 reactions with phosphoUb were significantly reduced (Fig 5C, Supplementary Fig S10B–D). Hence, while UBE2R1, UBE2T and UBE2E1 were charged with phosphoUb (Fig 5A), they were unable to generate significant levels of phosphoUb chains.
The most dramatic effects of phosphoUb on E2 chain elongation were observed with UBE2N (UBC13), which in complex with the inactive E2‐fold proteins UBE2V1 (UEV1A) or UBE2V2 (MMS2) assembles free Lys63‐linked polyUb. UBE2N was charged identically with Ub or phosphoUb as observed for the remaining E2s (Fig 5A and D). Addition of UBE2V1 led to robust Lys63 chain assembly with Ub, but the enzyme complex was inactive with phosphoUb (Fig 5D). This finding can be explained structurally from the complex of S. cerevisiae Ubc13~Ub with Mms2 (Eddins et al, 2006). The acceptor Ub is bound by Mms2, which positions it such that Lys63 points toward the active site. Ser65 of the acceptor Ub interacts directly with Mms2, and although the side chain does not make any contacts, phosphorylation would clash with Glu18 of Mms2 (Fig 5E). All functionally important residues in Ubc13/Mms2 are invariant in human UBE2N/UBE2V1. A reaction with Ub K63R as a donor Ub and Ub lacking the C‐terminus (Ub ∆GG) as an acceptor Ub assembled Lys63‐linked diUb specifically (Fig 5F). Increasing amounts of phosphoUb ∆GG did not affect diUb generation (Fig 5F), suggesting that this species was unable to compete with Ub ∆GG for UBE2V1 binding. Hence, the defect in UBE2N/UBE2V1 chain assembly is due to blocked acceptor Ub binding, consistent with the structure.
In summary, E2 charging with phosphoUb is unaffected for the majority of E2 enzymes, showing that E1 does not differentiate between Ub and phosphoUb. Importantly, some E2 enzymes are impaired in polyUb chain formation with phosphoUb, and, for example, the UBE2N/UBE2V1‐mediated assembly of Lys63‐linked chains is switched off by phosphoUb.
Effects of phosphoUb on E3‐mediated ubiquitination
E2 enzymes are discharged with the help of E3 ligases, which either induce a high‐energy state of the E2~Ub complex (RING, U‐box E3 ligases) or accept Ub to form a second thioester intermediate (HECT, RBR E3 ligases) (Berndsen & Wolberger, 2014). Recent structural work has revealed how RING E3 ligases generate a spring‐loaded E2~Ub complex (Dou et al, 2012; Plechanovova et al, 2012; Pruneda et al, 2012), and Ser65 of Ub is not involved in E2 or E3 interactions in any of the studied trimeric complexes (Supplementary Fig S11A). Likewise, Ub Ser65 is not contacted in HECT and RBR Ub complex structures (Supplementary Fig S11B and C). However, as seen with E2 enzymes, a switch from Ub to phosphoUb could lead to differences in ligase activity due to effects on acceptor Ub binding or may change linkage composition during chain formation.
We tested GST‐tagged RING domains of cIAP1 (aa 363–614) and TRAF6 (aa 1–285 and 50–285) in autoubiquitination reactions with Ub‐ or phosphoUb‐charged UBE2D1 and UBE2D3 (Yang et al, 2005). UBE2D1 and UBE2D3 work with most E3 ligases, are non‐specific for substrate modification and assemble multiple linkage types in vitro (Kim et al, 2007; Dynek et al, 2010). Autoubiquitination of GST‐cIAP1 proceeded with Ub and phosphoUb similarly with either E2 enzyme, leading to a depletion of monoUb and to qualitatively similar chain laddering on GST‐cIAP1 (Fig 6A and B, Supplementary Fig S11D). AQUA mass spectrometry revealed that GST‐cIAP1/UBE2D1 assembled chains to a similar extent (Fig 6C) and with similar composition with Ub and phosphoUb, namely Lys63 (~50%), Lys48 (~25%), Lys11 (~18%), and Lys6 and Lys27 linkages in smaller amounts (Fig 6D). The high amounts of Lys63 linkages (> 50% in both samples) were interesting, as this contrasts the finding of UBE2N/UBE2V1 where this linkage type was no longer assembled (Fig 5D). Hence, phosphoUb has no impact on cIAP1‐mediated chain assembly.
Surprisingly, with TRAF6, the same E2 enzymes, UBE2D1 and UBE2D3, assembled chains processively with Ub but not with phosphoUb (Fig 6E and F). This was observed for several GST‐TRAF6 constructs that include the RING domain as well as zinc finger modules (Fig 6E and F, Supplementary Fig S11E). AQUA mass spectrometry confirmed this result, showing that only a small fraction of phosphoUb is involved in chain linkages (Fig 6G and H). The low activity with phosphoUb was TRAF6 dependent (Supplementary Fig S11F). These results are unexpected as they suggest that TRAF6 is able to recognize the Ser65 phosphorylation status.
To understand whether phosphoUb affects other ligase classes, we tested NEDD4L, a Lys63‐specific enzyme (Kamadurai et al, 2009) with the E2 enzyme UBE2L3 that specifically transfers Ub onto the catalytic Cys of HECT and RBR E3 ligases (Wenzel et al, 2011). NEDD4L/UBE2L3 assembled Ub chains with Ub and phosphoUb identically (Fig 6I). Also, the bacterial effector HECT‐like enzyme NleL (Hospenthal et al, 2013) assembled chains similarly with Ub and phosphoUb (Fig 6J, Supplementary Fig S11G).
Finally, we tested another member of the RBR E3 ligase family, HOIP, which is part of the linear Ub chain assembly complex (LUBAC) (Kirisako et al, 2006). The HOIP RBR domain extended C‐terminally by a Ub binding domain assembles Met1‐linked chains processively in vitro (Smit et al, 2012; Stieglitz et al, 2012). Interestingly, this reaction proceeded with Ub, but was inhibited with phosphoUb (Fig 6K, Supplementary Fig S11H). The structural basis for this is unclear. A structure of a minimal HOIP construct sufficient for Met1‐chain assembly bound to two Ub molecules (Stieglitz et al, 2013) showed that Ser65 is solvent‐exposed in both donor and acceptor Ub (Supplementary Fig S11C). This suggests that the extended construct used in the assay forms additional Ub interactions involving Ser65 of donor or acceptor Ub that cannot be formed when Ser65 is phosphorylated. Alternatively, the phosphate affects either access to or the pKA of the acceptor α‐amine of Met1.
In summary, although testing only a small subset of E3 ligases, we uncovered that phosphoUb can lead to marked defects in polyUb chain generation.
Attributing phosphoUb ligation defects to the major conformation
Known structures of the ubiquitination machinery are incompatible with the observed minor phosphoUbretraCT conformation. An interesting finding with phosphomimetic Ub mutants confirmed that the surface charge changes of the major conformation in phosphoUb accounted for the defects in the ligation deficiency in TRAF6/UBE2D and UBE2N/UBE2V1.
Parkin activation in cells could be mimicked by phosphomimetic Ub mutants, where the phosphorylated Ser65 is replaced with Asp or Glu (Kane et al, 2014; Koyano et al, 2014). We tested whether phosphomimetic mutants displayed two conformations in NMR experiments, by analyzing spectra of 15N‐labeled Ub S65E and S65D (Supplementary Fig S12A). Both mutants did not show the additional peaks of the UbretraCT conformation, but revealed a spectrum with 69 resonances (Thr9, Glu24 and Ala46 were exchange‐broadened as compared to Ub) resembling the major conformation of phosphoUb (Supplementary Fig S12B). This is important as it suggests that phosphomimetic Ub mutants may not reconcile all effects of Ub Ser65 phosphorylation, and results from cell biology studies should be interpreted with care.
Still, this result was useful as it enabled us to understand whether the phosphoUbretraCT conformation or the major phospho‐Ub conformation accounted for the observed defects in chain assembly. Indeed, consistent with structural considerations (Fig 2E), Ub S65E is unable to perform ubiquitination with UBE2N/UBE2V1 (Supplementary Fig S13A). Similarly, for TRAF6/UBE2D, Ub S65E significantly inhibited chain assembly, although not to the same extent as phosphoUb (Supplementary Fig S13B). There could be multiple reasons for the difference, and it is possible that the conformational equilibrium of phosphoUb may exaggerate the observed loss of activity.
Lys63‐linked phosphoUb chains are recognized by TAB2
Ubiquitin binding domains (UBDs) decode Ub signals, and some preferentially interact with Ub polymers (Husnjak & Dikic, 2012). As mentioned above, the principle UBD recognition site on Ub, the Ile44 patch, is not grossly affected in the major conformation of phosphoUb, and we would expect small UBDs to interact similarly with Ub and phosphoUb. The case may be different for chain interacting UBDs. The TAB2 NZF domain “bends” a Lys63‐linked diUb to interact with its two Ile44 patches simultaneously (Kulathu et al, 2009; Sato et al, 2009), yet Ser65 of either molecule is solvent‐exposed (Fig 7A).
To test whether TAB2 binds poly‐phosphoUb, we performed pull‐down experiments using GST‐tagged TAB2 NZF domain (Kulathu et al, 2009), revealing strong interactions with Lys63‐linked but not Met1‐linked tetraUb regardless of phosphorylation status (Fig 7B). TAB2 has previously been used to sense Lys63‐linked chains on mitochondria after depolarization (van Wijk et al, 2012), and it can be assumed that the detected chains were partly phosphorylated.
DUBs hydrolyze phosphoUb chains with lower activity
Finally, we analyzed the ability of DUBs to hydrolyze phosphoUb chains. The human genome encodes ~80 active DUBs, > 50 of which contain Ub‐specific protease (USP) domains. The remaining enzymes belong to four structurally distinct families (Komander et al, 2009; Clague et al, 2013). USP enzymes such as USP2 (Fig 8A) engulf ~40% of the solvent accessible surface of a bound distal Ub (Komander et al, 2009). Importantly, Ser65 is involved in these interactions, and phosphorylation may affect DUB recognition (Fig 8A). Indeed, when we tested USP2 activity against ubiquitinated GST‐cIAP1 (Fig 6A) as a substrate in DUB assays, we found that USP2 was markedly less active against poly‐phosphoUb (Fig 8B) and only cleaved phosphoUb chains at later time points or at higher enzyme concentration (data no shown). Identical behavior was found for USP8, USP15 and USP30 (Fig 8C–E). Also, the Josephin‐family DUB Ataxin‐3 showed less activity against poly‐phosphoUb (Fig 8F). Interestingly, USP8, USP15, USP30 and Ataxin‐3 have all been implicated in Parkin regulation (Durcan et al, 2012, 2014; Bingol et al, 2014; Cornelissen et al, 2014). In contrast, known non‐specific enzymes USP21 and the small viral OTU domain DUB vOTU (Akutsu et al, 2011; Ye et al, 2011) were able to cleave polyUb and poly‐phosphoUb similarly (Fig 8G and H).
USP30 was significantly less active toward cIAP polyUb chains as compared to other USP DUBs. To test whether this was due to a different chain preference, we performed a specificity analysis against a panel of differently linked diUb, revealing that USP30 preferred Lys6‐linked chains (Fig 8I). This is interesting as Lys6‐linkages are enriched in Parkin substrates (Durcan et al, 2014; Ordureau et al, 2014). USP30 was still less active toward phosphorylated Lys6‐linked diUb (Fig 8J).
Finally, we tested the activity of linkage‐specific DUBs against defined phosphorylated substrates. The JAMM‐family DUB AMSH is Lys63‐specific (McCullough et al, 2004), but is less active against phosphorylated chains (Fig 8K). Similarly, Lys29/Lys33‐specific TRABID, Lys48‐specific OTUB1 and Met1‐specific OTULIN are impaired in cleaving phosphorylated chains (Fig 8L–N).
Structurally, some of these results are surprising. While USP Ub binding may explain the weak activity of this enzyme class against poly‐phosphoUb, the outlier USP21 shows that subtle differences in the binding site apparently enable a USP domain to target phosphorylated chains (Supplementary Fig S14A). Ser65 in distal or proximal Ub is solvent‐exposed in known structures of OTU DUBs, such as OTUB1 (Juang et al, 2012; Wiener et al, 2012) (Supplementary Fig S14B) and OTULIN (Keusekotten et al, 2013; Rivkin et al, 2013) (Supplementary Fig S14C). Similarly, Ser65 in the distal Ub is solvent‐exposed in the Ataxin‐3‐like structure (Weeks et al, 2011) (Supplementary Fig S14D). In AMSH‐LP, Ser65 of the proximal Ub is close to the enzyme (Sato et al, 2008) and may account for the observed reduced cleavage activity (Supplementary Fig S14E).
The fact that 10 out of 12 tested DUBs have notably weaker activity against chains that contain phosphoUb suggests that phosphorylated chains are more stable as compared to unphosphorylated chains. Importantly, chains can be phosphorylated by PINK1 (Fig 1), and poly‐phosphoUb exists in cells (Ordureau et al, 2014). This together with the inhibition of ligases could change the dynamics of modified substrates, their degradation rates or signaling capacity. Hence, Ub phosphorylation may have consequences on the dynamics of the entire Ub system.
PINK1‐mediated phosphorylation of Ub at Ser65 has dramatic consequences for Ub structure, and key processes in the Ub system, namely Ub attachment and removal.
It could be expected that phosphorylation of Ub would change its surface properties due to the addition of a negative charge. The obtained high‐resolution crystal structure and solution studies agree that the majority of phosphoUb is structurally similar to wt Ub. To our amazement, NMR studies showed a second, minor conformation of phosphoUb, which is in slow exchange with the major conformation. Strikingly, the minor conformation shows distinct hydrogen bonding patterns and long‐range NOEs for its C‐terminal β5‐strand, which can only be structurally satisfied when this strand is shifted by two residues. Our phosphoUbretraCT model explains numerous observations and is structurally feasible due to the existence of four Leu‐Xaa repeats in the β5‐strand that would allow a shift of two residues without significantly disturbing the Ub core. β‐strand slippage is a known phenomenon occurring in, for example, Arf family GTPases (Pasqualato et al, 2002) and Serpins (Zeraik et al, 2014). In these proteins, strand slippage is induced by changes of nucleotide (GDP/GTP) in the binding pocket. While conceptually similar, we are not aware of a phosphorylation‐induced β‐strand slippage as reported here, and to our knowledge, the UbretraCT species has not been observed previously.
Structurally, β5‐strand slippage would have significant implications: Firstly, shifts of His68 and Val70 would disrupt the important Ile44 hydrophobic patch which is involved in most Ub interactions described to date (Husnjak & Dikic, 2012; Komander & Rape, 2012). Secondly, the UbretraCT conformation can presumably not be recognized by enzymes of the assembly cascade (E1, E2, E3) as well as DUBs since these proteins rely on interactions with the extended Ub C‐terminus. Thirdly, retraction of the C‐terminus may change the dynamics and structure of polyUb chains, adding further complexity to the already vast structural landscape adopted by Ub polymers (Kulathu & Komander, 2012).
The relevance and potential roles of the UbretraCT conformation remain unclear. Its existence raises fundamental questions for Ub and Ubl conformation, dynamics and folding. Biologically, the UbretraCT conformation could be recognized as a distinct signal. One could speculate that a mitophagy adaptor or other phosphoUb binding proteins may preferentially recognize the UbretraCT conformation of phosphoUb, and it will be interesting to identify such proteins.
Biochemically, it was not clear whether phosphoUb can be assembled into chains and whether chains incorporating phosphoUb are recognized by UBDs and hydrolyzed by DUBs. Using representative members from all branches of the Ub assembly and disassembly cascades, we demonstrated that, in principle, E1 activation, E2 charging and E3‐mediated discharging can progress with phospho‐Ub, although we made several surprising observations. Most strikingly, the Lys63‐specific assembly complex UBE2N/UBE2V1 was inactivated by phosphoUb, which can be explained structurally. PhosphoUb also affected discharging in other E2 enzymes; the reasons for this are less clear.
We also investigated two RBR, two HECT and two RING E3 ligases, uncovering surprising differences for the latter. With identical E2 enzymes, the well‐studied RINGs of cIAP1 and TRAF6 processed Ub and phosphoUb differently, and our results suggest that TRAF6 recognizes Ub and is unable to bind phosphoUb. The two tested HECT E3 ligases worked similarly with Ub and phosphoUb, while the RBR enzyme HOIP and Parkin are less active with phosphoUb.
Ub phosphorylation severely affects DUB activity, and 5 out of 6 USP enzymes, 4 out of 5 OTU enzymes and the tested JAMM and Josephin DUBs were significantly less active against phosphoUb. We were surprised that all DUBs implicated in Parkin biology and mitophagy were less active with phosphoUb; we wonder whether a phosphoUb‐specific DUB exists. Since phosphoUb chains are more resistant to DUB activity, they likely constitute a more stable Ub signal.
Our work also revealed some insights for the PINK1/Parkin system. We show that, at least in vitro, Ub tetramers comprising six of the eight linkage types can be phosphorylated at every Ub in the chain. Harper and colleagues showed that this likely holds true in vivo, where PINK1 phosphorylates polyUb of various chain linkages on mitochondria (Ordureau et al, 2014). This is interesting from a kinase perspective since, apparently, PINK1 is able to access each Ub molecule within chains to phosphorylate Ser65. Moreover, Ser65 is in a folded and structurally stable environment in Ub, not exposing any stretch of primary sequence. This indicates that PINK1 likely recognizes a three‐dimensional motif present in Ub and the Parkin Ubl domain, rather than, as more common for kinases, a peptide motif.
According to proteomic analysis, PINK1 phosphorylates 2.5–3% of total Ub and 10–20% of mitochondrial Ub upon mitochondrial depolarization (Koyano et al, 2014; Ordureau et al, 2014). Different models have been proposed for phosphoUb‐mediated Parkin activation, and the sequence of events is still debated (Zheng & Hunter, 2013; Kane et al, 2014; Kazlauskaite et al, 2014; Koyano et al, 2014; Ordureau et al, 2014). We here show that for phosphoUb to activate Parkin, unphosphorylated Ub must be present. This was also shown by Harper and colleagues, although in their case, the effects of pure phosphoUb appear less dramatic (50% reduction in chain formation) (Ordureau et al, 2014). The mechanistic basis for this is not clear and will need to be revealed in structural studies.
In summary, we show that phosphoUb, beyond the described gain‐of‐function in Parkin activation, has primarily detrimental effects on the Ub system, partly due to significant structural perturbations (Fig 9). Many issues remain, in particular regarding specific recognition of phosphoUb by (novel?) UBDs, specific reversal of phosphoUb signals by DUBs, and regarding functional consequences from inactivating E2/E3 ligase systems. Structurally, it is of prime importance to understand how phosphoUb is recognized by Parkin, how PINK1 can phosphorylate Ub specifically, and whether the UbretraCT conformation has additional binding partners.
It is also clear that the identification of Ub phosphorylation by PINK1 is the tip of the iceberg with regard to Ub modifications. Several Ser/Thr/Tyr residues of Ub, including Thr7, Thr12, Ser57 and Tyr59, are known to be phosphorylated by as‐yet unknown kinases (http://www.phosphosite.org), and all may have dedicated functions and pathways. This vastly increases the complexity in the Ub system and requires careful systematic future studies.
Materials and Methods
Please find complete Materials and Methods in the Supplementary Information online.
Crystallographic data for Ser65‐phoshoUb have been deposited with the protein data bank, accession number 4wzp.
DK, KNS, TW, CG, MAM and MG performed biochemical experiments, and TW and DK solved and refined the phosphoUb crystal structure. KNS established mass spectrometry including AQUA and PRM analysis in the laboratory, performed all mass spectrometry and analyzed the data. JNP generated reagents and provided expertise and protocols. CMJ performed stability measurements. JW and SMVF performed NMR analysis. KNS, TW and CG discovered ubiquitin phosphorylation by PINK1 in vitro. DK wrote the manuscript with input from all authors.
Conflict of interest
DK is part of the DUB Alliance that includes Cancer Research Technology and FORMA Therapeutics, and is a consultant for FORMA Therapeutics.
We would like to thank Wade Harper and Alban Ordureau for pSer65 AQUA peptides and for sharing unpublished information, Thomas Mund for the gift of NEDD4L protein, Marc Pittman and Sylvie Urbe for the gift of DUBs, Arthur Pardi for sharing Mathematica scripts, Paul Elliott for help with data collection, and beam‐line staff at Diamond Light Source, beam line I‐03. We also thank members of the DK lab for reagents and advice and Trevor Rutherford and Paul Elliott for critical comments on the manuscript. This work was supported by the Medical Research Council [U105192732], the European Research Council , the Lister Institute for Preventive Medicine and the EMBO Young Investigator Program. MAM is supported by the Boehringer Ingelheim Foundation and JNP is an EMBO Long‐Term Fellow.
FundingMedical Research Council U105192732
This is an open access article under the terms of the Creative Commons Attribution 4.0 License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited.
- © 2014 MRC Laboratory of Molecular Biology. Published under the terms of the CC BY 4.0 license