Parkin, A Top Level Manager in the Cell’s Sanitation Department

C-Terminal Mutations Affect Parkin Stability

Parkin is aggregation-prone and is dependent on chaperones for proper folding [29]. A series of engineered deletion mutations involving the 16 C-terminal amino acids showed that truncations exceeding 3 amino acids produced insoluble Parkin [29]. An ARJP mutation in which tryptophan 453 is replaced with a termination codon (W453X), lacks the last 13 amino acids and is insoluble; however, the chaperone, hsp70, which prevents aggregation of wild type Parkin also promotes folding of the W453X protein [29]. Using proteinase K digestion, it was found that the conformation assumed by W453X is different from that of wild type Parkin [29].

The insolubility of C-terminal deletions is specific to Parkin and is not shared by the very similar HHARI [30]. Nor do the C-terminal amino acids from HHARI rescue a hybrid Parkin-HHARI construct as the chimeric protein was rapidly degraded by the proteasome [30]. In Parkin, the amino acid F463 was found to be highly conserved between species, and when mutated to Ala, was found to form insoluble aggregates [30]. The role of the C-terminus in Parkin stability will be addressed in further detail in “Part VIII. Predicted RING structures”.

N-Terminal Mutations

The mutation R42P is located in the ubiquitin-like domain and is often mentioned in the literature. It appears that R42P protein is highly unstable, being rapidly degraded by the proteasome. Only under conditions of proteasome inhibition, can R42P be detected and then it is found in the insoluble fraction. This suggests that R42P protein assumes a non-native conformation which is rapidly degraded [30]. Interestingly, an N-terminally truncated variant of Parkin found in human brain and cultured cells that resulted from translation initiation at an internal methionine residue at position 80 [24] was soluble whereas full length Parkin carrying the R42P mutation was not [27].

Mutation of Parkin Cysteine Residues

Since 28 of the 35 cysteines in Parkin are involved in zinc binding, it was of interest to determine what would happen if any of those cysteines were mutated. Mutational analyses revealed that the great majority of the Parkin Cys residues are essential to maintaining Parkin solubility [21, 27]. Cysteine mutations that did not appear to affect solubility were C59A, C95A, C268A, C323A, C431A and C451A [21].

Finally, there are two interesting experimental outcomes regarding mutations. In one experiment, a double mutation (R42P and W453X) was expressed in the same protein. The R42P alone resulted in proteasomal degradation of Parkin, i.e., instability, whereas the W453X mutation alone resulted in insoluble aggregates. On expression in SH-SY5Y cells, the double mutant Parkin was only detected with proteasome inhibition, and then only in insoluble aggregates. Therefore, proteasomal degradation was dominant over aggregation, i.e., no aggregates formed with an intact ubiquitin proteasome degradation system [30]. The second interesting experimental outcome revealed that some innocuously appearing mutations, i.e., ones showing no signs of decreased Parkin solubility, sensitized Parkin to cell-induced stress [31]. That stress might include oxidative stress, proteolytic (unfolded protein) stress, s-nitrosylation, and/or neurotoxins. Wang et al. (2005) [31] then summarized the types of mutations according to their effects on Parkin, the first type by directly reducing or destroying catalytic function, the second type by changing solubility and location, and the third type by increasing sensitivity to environmental stresses.

Parkin and the Ubiquitin Proteasome System

In the UPS a small protein tag, ubiquitin, is covalently attached to the target protein through a series of steps, the last catalyzed by an E3 ubiquitin ligase. The first step is the ATP-dependent activation of ubiquitin by an E1 ubiquitin-activating enzyme. The activated ubiquitin is then transferred to a ubiquitin-conjugating enzyme, E2. The E3 ubiquitin ligase binds both the E2-ubiquitin and a specific protein target to which the ubiquitin is then covalently attached. Each step in the process becomes more restricted in its substrate, with the E3 ligase being the most specific (reviewed in Finley and Chau (1991) [32]; more recently in Deshaies and Joazeiro (2009) [35]). Parkin has been shown to have E3 ubiquitin ligase activity [36-39]. Consistent with a role in the ubiquitin proteasome system, Parkin has been shown to bind two subunits in the 26S proteasome, Rpn10 of the 19S regulatory subcomplex and the α4 subunit of the 20S proteolytic core [40, 41].

Ubiquitin is attached by formation of an isopeptide bond between the ubiquitin C-terminal Gly residue and the ε-amino group of a Lys residue in the targeted protein. There are at least three ways for a protein to be ubiquitinated: i) a chain of ubiquitin molecules formed by linkage to Lys 48 of the previously attached ubiquitin, ii) a chain of ubiquitin molecules linked through ubiquitin Lys 63, and iii) multimonoubiquitination, defined as multiple ubiquitin molecules per protein but attachment is one ubiquitin per lysine residue. Ubiquitin chains formed by Lys 48 linkage targets proteins for proteasomal degradation and a chain of at least four subunits is required for recognition and degradation [42]. Parkin can auto-ubiquitinate and in cell-free in vitro assays has been shown to multimonoubiquitinate itself [27, 43] as well as proteins that are in close proximity [43]. However, as discussed in the Hampe et al. report [27], the cell-free in vitro ubiquitination assays did not include other proteins that might be present in the cell. It was hypothesized that Parkin alone produces multimonoubiquitinated substrates whereas Parkin associated with other proteins conjugates polyubiquitin chains [27]. One of these other proteins may be an E4, (identified and described by Koegl et al. (1999) [44]. E4 ubiquitin ligases can be described as ubiquitin chain elongation factors. Without E4, substrate ubiquitination consists of only a few ubiquitin molecules, insufficient to accomplish proteasomal degradation [44]. Proteins such as Parkin, with ubiquitin-like domains at their N-terminus, are analogous to substrates of E4, (in particular of UFD2, a yeast member of the E4 family), and are likely to be degraded by the proteasome system [44].

Parkin has been shown to be auto-ubiquitinated when over-expressed in cultured 293, 293T and COS1 cells [36]; and co-transfection with the K48R ubiquitin mutation that prevents formation of K48- linked chains, inhibited Parkin ubiquitination [36]. This suggested that the ubiquitins were attached as Lys 48-linked chains. An unidentified mammalian E4 may facilitate K48-linked chains in cultured cells, but with no E4 in cell-free assays, Parkin is multimonoubiquitinated. Another study utilizing SH-SY5Y cells found that ubiquitinated Parkin could be identified only when cells were treated with the proteasome inhibitor MG132, suggesting that in neuronal cells with uninhibited proteasomes, ubiquitinated Parkin was rapidly degraded via the UPS [37].

As mentioned above, should the ubiquitin proteasome system become compromised in some way, proteins to be removed may be sequestered in the aggresome for eventual removal by macroautophagy. An aggresome is defined as a microtubule-dependent inclusion body that forms around the microtubule organizing center and is near to, or encompasses the centrosomes [33]. Proteins to be degraded by the aggresome pathway are actively transported along microtubules to the centrosome. At the same time, components of the UPS, e.g. ubiquitin, ubiquitin conjugating enzymes, proteasomal constituents, certain chaperones, and mitochondria are also recruited. These components are then organized into an aggresome (reviewed in Olanow (2004) [45]). Abnormal or damaged proteins are held within the aggresome by a cage of intermediate filaments, e.g. vimentin [33]. Parkin is actively involved in both the UPS and the aggresome system.

Parkin Substrates Ubiquitinated for Degradation

Of the many responsibilities assigned to the cellular Sanitation Department, Parkin participates either directly or by “committee” (as a member of a multi-protein complex) in removing i) short-lived proteins, ii) altered proteins, iii) excessive levels of proteins and iv) unfolded proteins. A short-lived Parkin substrate, SIM2, is a transcription factor located in the nucleus, and, like many short-lived proteins, contains a PEST degradation signal rich in the amino acids proline (P), glutamic acid (E), serine (S) and threonine (T). The human SIM2 protein is ubiquitinated by both Parkin and HHARI [46]. SIM2 is located in the nucleus whereas Parkin is found mainly in the cytoplasm [46]; however, the authors note that a very small amount of Parkin can be found in the nucleus and this nuclear Parkin is suspected of ubiquitinating SIM2 [46].

αSp22, a glycosylated form of α-synuclein, is an altered protein that is ubiquitinated by Parkin [47]. αSp22 accumulates in ARJP brain, but is undetectable in extracts from normal brain [47]. αSp22 appears to be unique to human brain as it could not be detected in lysates from HEK293 or SH-SY5Y cells transiently transfected with Parkin and α-synuclein, even when those cells were oxidatively stressed [48].

The toxicity of a third class of Parkin substrates, proteins that exceed certain concentration levels, was demonstrated by exogenous overexpression of these Parkin substrates in cell culture. Pael-R, p38, and Synaptotagmin XI are in this class of Parkin substrates. Pael-R is a G protein-coupled receptor thought to play a role in dopaminergic signaling. Over-expression of Pael-R produced insoluble, ubiquitinated protein and increased cell death [49]. Normal expression levels of Pael-R do not lead to insoluble accumulations in brain; however, insoluble Pael-R accumulates in ARJP brain [49]. Recently, a rat model of PD was developed by capitalizing on the fact that dopaminergic nerurons are especially sensitive to Pael-R accumulation [50].

The protein p38 is a structural component of the aminoacyl-tRNA synthetase complex involved in protein biosynthesis [51]. Over-expression of p38 in the dopaminergic cell line SH-SY5Y increased cell death by approximately 6-fold whereas over-expression of both p38 and Parkin reduced the amount of cell death when compared to p38 alone. This suggests that the toxic affect of increased p38 expression is partly alleviated by Parkin [51]. We assigned Synaptotagmin XI (SytXI), involved in the maintenance of synaptic function, to this category since a scan for SytXI mutations that might potentially produce disease in either inherited or sporadic PD was negative [52]. This suggested that protein levels rather than protein abnormalities were the key trigger for ubiquitination and degradation. However, this does not preclude SytXI degradation as a result of misfolding. In sporadic PD, SytXI is found in the core of Lewy bodies, suggesting that SytXI may be misfolded and targeted for degradation by the aggresome rather than by the proteasome pathway (discussed below). SytXI was identified as a Parkin substrate by yeast two hybrid followed by co-immunoprecipitation assays [53]. It was determined from the appearance of the bands on immunoblots that SytXI is polyubiquitinated and pulse-chase experiments using transiently transfected HEK293 cells showed that Parkin was responsible for the rapid turnover of SytXI.

Improperly folded proteins consititute a large number of proteins tagged for degradation. Upwards of 30% of newly synthesized proteins are improperly folded and the remaining 70% may at some later time be damaged or denatured during metabolic processes or by a change in the microenvironment of the cell (reviewed in Tanaka et al. (2004) [54]). Secretory proteins are folded within the ER; however, when the folding process fails, the unfolded proteins are removed to the cytosol for proteasomal degradation. Unfolded proteins produce “unfolded protein stress” which initiates a cellular “unfolded protein response”. One of the first steps in the unfolded protein response is an up-regulation of genes/proteins which aid in refolding or in protein degradation. One of these up-regulated genes, BiP, has become a marker for the unfolded protein response and was identified in the insoluble fractions of ARJP brain [49]. This suggests that the unfolded protein degradation pathway is activated in the absence of functional Parkin. An overwhelmed or impaired proteasomal system can also produce unfolded protein stress and the resulting unfolded protein response.

Parkin Activity in Response to Unfolded Protein Stress is Cell-Specific

Imai et al. (2000) [36] reported that Parkin specifically blocked cell death from unfolded protein stress and that Parkin activity was cell specific, i.e., the SH-SY5Y dopaminergic cell line had higher levels of protein ubiquitination than the HEK293 human embryonic kidney cell line. Parkin expression was up-regulated and cell death resulting from unfolded protein stress was reduced [36]. Although West et al. (2003) [55] were unable to confirm these results, a study utilizing primary cultures of rat hippocampal neurons and astrocytes showed that Parkin expression increased in the astrocyte, but not in the hippocampal cultures in response to unfolded protein stress [56]. Further, treatment of astrocytes with unfolded protein stress inducers resulted in formation of structures similar to aggresomes in appearance and location [56].

The “Lewy Body is a Failed Aggresome” Hypothesis

Lewy bodies are cytoplasmic inclusions that are found in sporadic PD (described in Olanow et al. (2004)) [45]. They are very similar to aggresomes, which form in response to unfolded protein stress or to an inefficient or inhibited proteasomal system (reviewed in Kopito (2000) [34]). Clearance of aggresomes may involve the macroautophagy system whereby the aggresome is engulfed by a double membrane to form an autophagasome which fuses with an endosome (or lysosome) to complete proteolytic digestion and clear the aggresome from the cell [34]. If aggresomal protein is not degraded, the structure may become an insoluble cellular inclusion. Due to the structural similarities between aggresomes and Lewy bodies, Olanow et al. 2004 hypothesized that Lewy bodies are failed aggresomes, i.e. they are unable to complete proteolysis successfully.

Recently, Wong et al. (2008) [57], presented evidence suggesting that aggresome clearance by macroautophagy is determined by the components recruited to the aggresome along with the aggregated protein. Some proteins, such as the Parkin substrate p38 fail to recruit key macroautophagy components to the aggresome and are therefore autophagy resistant [57].

Parkin Contributes to Lewy Body Formation

Synphilin-1 is an α-synuclein interacting protein found in Lewy bodies [48] and confirmation of Synphilin-1 as a binding partner and substrate of Parkin led to the finding that E3 ligase activity of Parkin contributes to Lewy body formation. Lewy bodies are by definition, ubiquitinated cytoplasmic inclusions. Co-expression of Parkin, synphilin-1 and α-synuclein resulted in ubiquitinated Lewy-body-like inclusions [48]. However, without expression of Parkin, these inclusions were not ubiquitinated and therefore by definition were not Lewy bodies. That Parkin contributes to Lewy body formation was further supported by the finding that Parkin localized to Lewy bodies [58] and that 90% of nigral Lewy bodies are immunopositive for Parkin [59]. Parkin mutations that result in loss of Parkin ubiquitination activity or that result in Parkin misfolding and subsequent degradation would obviously prevent Lewy body formation, thereby providing an explanation for the absence of Lewy bodies in ARJP.

Proteasome Inhibition and Aggresome Formation

It is important to keep in mind that most of the Parkin E3 ligase activity assays were performed in cultured cells and therefore much of the ubiquitinated substrate would be rapidly cleared by the UPS before samples could be collected. Detection could occur only if the amount of substrate completely overwhelmed the cellular UPS or if the UPS were inhibited. At this point it is unclear whether formation of aggresomes occured in response to an experimentally inhibited UPS or to a “normal” clearing process for that particular substrate. Two factors may help to clarify the issue: i) is the substrate found in Lewy bodies in idiopathic PD?, and ii) is the protein substrate membrane-associated?. Transmembrane or membrane-associated proteins are more likely to be misfolded and removed via the aggresome degradation system [33]. Parkin substrates synphilin-1, p38, and synaptotagmin XI are found in Lewy bodies whereas CDCrel-1 and Pael-R are examples of membrane-associated proteins.

Membrane-Associated Parkin Substrates

CDCrel-1 is a member of the human septin protein family and is recruited to the centrosome along with Parkin [60]. Although ubiquitin modified CDCrel-1 was detected in the absence of transfected parkin, levels increased with expressed Parkin and CDCrel-1 was found to be degraded in pulse-chase assays [38]. Originally thought to be associated with synaptic vesicles, more recent evidence suggests that only a small fraction of CDCrel-1 is found in synaptic vesicles [61]. Although the role of CDCrel-1 in Parkinson’s disease has not yet been clarified, a number of possibilities that lead to inhibition of dopamine release have been proposed [38]. Observations of CDCrel-1 null mice revealed no significant disruption of brain function, suggesting that other septins likely compensated for loss of CDCrel-1 [62]. Another septin, SEPT5_v2a is a close homolog of CDCrel-1 [63] and accumulated in 3 of the 4 ARJP brains examined [63].

Pael-R, a toxic protein when over-expressed (discussed above), was also reported to accumulate in the endoplasmic reticulum during unfolded protein stress. The up-regulation of BiP (marker gene for unfolded protein response), confirmed that the cells were responding to unfolded protein stress. Six hours after proteasome inhibition, Pael-R was found in the endoplasmic reticulum; after 16 hours, Pael-R was located within aggresome-like inclusions at juxtanuclear sites [49].

Although Parkin may be found in aggresomes/Lewy bodies, it may be sequestered for different purposes. Proteasomal inhibition resulted in aggresomal sequestration of wild type Parkin presumably to aide in the degradation of unfolded proteins [64], However, several Parkin mutations, e.g. R256C, R275W, C289G, and C418R which reduce Parkin solubility are relocated from cytoplasm to aggresomes, presumably to degrade the insoluble Parkin [21, 28, 65, 66].

Mechanics of Aggresome Formation and Microtubules

Proteins destined for aggresomal degradation are actively transported along microtubules to the centrosome as are components of the UPS, e.g. ubiquitin, ubiquitin conjugating enzymes, proteasomal constituents, certain chaperones, and mitochondria. These components are then organized into an aggresome (reviewed in Olanow et al. (2004)) [45]. When the cellular proteosomal system is inhibited, Parkin is recruited to the centrosome; removal of proteosomal inhibition reverses the centrosomal binding and Parkin is dispersed [60, 67]. This is accomplished by histone deacetylase 6 (HDAC6) which appears to sense proteosomal efficiency [67]. Generally, ubiquitinated proteins that are transported to the centrosome bind to the C-terminal ZnF-IBP motif of HDAC6 (discussed in Jiang et al. (2008)) [67]. Parkin however, binds to two other regions in HDAC6, DD1 and DD2. The interaction between Parkin and HDAC6 is direct and not mediated by microtubules. Recruitment of Parkin to the centrosome was shown to be dependent on HDAC6/dynein motor protein binding, and dispersal dependent on HDAC6 switching to the kinesin 1 protein (KIF5) [67]. HDAC6 deacetylates tubulin, and although apparently not acting directly on Parkin, the tubulin deacetylase activity of HDAC6 is required for Parkin recruitment to and dispersal from the centrosome [67].

Considering the importance of microtubules for centrosomal recruitment of Parkin and substrate proteins, it comes as no surprise that Parkin, as a top level manager in the cell’s Sanitation Department takes some responsibility for microtubule quality control and stabilization. Microtubules are formed from α/β tubulin heterodimers, and monomer levels of α and β tubulin are closely regulated so that neither is in excess. In addition, correct folding of tubulin is a complex process that results in a high level of misfolded tubulin (discussed in Ren et al. (2003)) [68]. Parkin co-immunoprecipitated with α or β-tubulin monomers in HEK293 cells and in rat brain lysates, and with α/β tubulin heterodimers in rat brain lysates [68]. A third type of tubulin, ɣ-tubulin, co-localized with Parkin in Lewy bodies (reviewed in Olanow et al. (2004) [45], McNaught et al. (2002)) [69].

Parkin interacts with microtubules, binding very tightly even in the presence of 3.8 M NaCl or 0.5 M urea and acting as a MAP (microtubule-associated protein [70]). Microtubules were stabilized when bound to wild type Parkin in the presence of the microtubule de-polymerizing agent, colchicine [70]. Parkin alone protected against the microtubule de-polymerizing PD toxin, rotenone, as shown by loss of protection in β-lymphocyte and fibroblast cell cultures derived from PD patients carrying Parkin mutations [71]. Parkin/microtubule interactions that function to stabilize microtubules do not appear to interfere with E3 ligase activity [68].

Parkin-Protein Interactions

Generally, Parkin binding sites were determined by immunoprecipitation of the Parkin substrate or interacting protein using various fragments of Parkin. Several studies used different Parkin domains, parts of domains, or combinations of domains in their binding assays, thereby making comparisons between reported Parkin/protein interaction sites difficult. Keeping this in mind, certain generalizations regarding Parkin-protein binding can be made.

One important observation is that Parkin-protein interaction sites are spread over the entire Parkin molecule (Table 1). RING2 appears important for many proteins, either as the single binding site, as a part of a larger binding site, or as an alternative binding site. RING2 alone appears sufficient for binding Synphilin 1 [48], Pael-R [49], and UbcH13 [95] SIM2 [46] and α4 [41] require the adjacent IBR in addition to RING2, and the entire RBR is needed for DJ-1 [80] and hSel-10 [77]. Some proteins interact with RBR domains other than RING2. PCNA binds to a fragment containing both RING1 and the IBR (although the two domains were not tested separately) [78]; and both CHIP and hsp70 bind to RING1. Several proteins bind to domains outside of the RBR. RING0 is an alternative site for tubulin [70] and HDAC6 binding [67]; BAG5 appears to be the only protein that interacts solely with RING0 [73]. Rpn10 appears to be the only protein discovered thus far that interacts only with the ubiquitin-like domain [40]; Sept5_v2a interacts with either the ubiquitin-like domain or RING1-IBR of the RBR region [63].

E3 Ubiquitin Ligase Activity

A number of studies have suggested that requirements for binding are less stringent than are requirements for ubiquitination or degradation of Parkin substrates. There are some general “rule of thumb” guidelines involving mutations that can be useful in examining binding sites. For instance, if a mutation is known to change the conformation of a molecule and also affects binding properties, then the binding site may be dependent on the conformation of the molecule. If the binding site requires non-consecutive regions of the molecule, then likely the site requires a structure that pulls those disparate regions together. And finally, if a binding site is a small specific primary sequence of amino acids, then the tertiary structure is less likely to affect binding unless the site is obscured by the tertiary structure. In such cases a mutation may actually increase binding if it denatures that small region of the molecule, perhaps by disruption of zinc ion binding that stabilizes the obscuring structure. Examples include: i) synphilin-1 which maps to RING2. As expected, familial-PD mutations in RING1 appeared to have no effect on binding; however, those in RING2 and the IBR actually increased binding [48]. At this point it is unclear why binding would increase. Although the ARJP mutations T240R and R256C (in RING1) as well as the truncating mutation W453X either had no effect or actually increased binding, ubiquitination of synphilin-1 was adversely affected by all three of these mutations [48]. Interestingly, a later investigation of the effect of mutations on Parkin E3 ligase activity identified the mutation T240R as belonging to a set of mutations (K161N, T240R, Q311X, T415N and P437L) that are considered “ligase dead” since they neither auto-ubiquitinate nor substrate ubiquitinate [26]. R256C belongs to a set (R256C, G328E, G430Q, C431F, and W453X) that auto-ubiquitinates but not substrate ubiquitinates [26]; ii) tubulin is bound, but not ubiquitinated, in the presence of Parkin mutations K161N in RING0, T240R in RING1, and C431A in RING2 [68]. Again, these belong to the set of “ligase dead” mutations [26]; iii) binding and ubiquitination of Parkin substrates (CDCrel-1, p38 and α-tubulin) or associated proteins (ɣ tubulin, hsp70 and the α4 subunit of the proteosome) were tested with ten Parkin mutations A82E, K161N, K211N, R256C, G328E, R42P, R275W, C289G, C418R, and C441R [27]. The binding interactions between these proteins and mutant Parkin variants did not appear to vary significantly from that of wild type Parkin. However, utilizing a cell-free in vitro ubiquitination assay that measured Parkin auto-ubiquitination, only C418R and C441R (which disrupt zinc ion binding in RING2) inhibited ubiquitination. Since the assay involved Parkin auto-ubiquitination we would expect mutations R256C and G328E (in the set of mutations that auto-ubiquitinate) to be unaffected; however, K161N was not “ligase dead” as might be expected.

It should be noted that there are many discrepancies in reports of substrate ubiquitination, some likely due to effects of other proteins in cell-based assays. And “in vitro” assays utilizing recombinant Parkin purified from reticulocytes may be affected by protein contaminants. This was discussed thoroughly in the Hampe et al. report [27].

General vs Specific Ubiquitination in the Presence of Parkin

Although Parkin expression increased ubiquitination of α- and β-tubulin in HEK293 cells, the authors also showed data suggesting that exogenous expression of Parkin may enhance ubiquitination levels for many different proteins [68]. Certainly, there is little doubt that Parkin recognizes specific proteins for ubiquitin modification; however, it is possible that Parkin alone or as a component of a multi-protein complex, perhaps with chaperone complex members, can recognize something more general such as exposed hydrophobic regions. Whether Parkin switches between a specific substrate to more general substrates, and if so, determining the mechanism for such a substrate switch, is an interesting question for possible exploration. Similarly, many questions remain regarding the circumstances under which a Parkin substrate is “chosen” for binding and/or ubiquitination,

Structure

The three RING domains (0, 1 and 2) of the Parkin protein were modeled using the I-TASSER program that generates structure predictions by first identifying templates (proteins with similar fold) for the query protein from the Protein Data Base (PDB) library. I-TASSER then assembles the fragments derived from the templates into full-length models using the parallel replica-exchange Monte Carlo simulations [100].

The models generated for each Parkin domain were analyzed to identify plausible locations of zinc ion binding sites based on sequence alignment and the position of zinc coordination sites in templates identified by threading algorithms. Zinc ions were added near the plausible binding sites and the structure was refined using the Gromacs molecular dynamics simulation package [101] for the purpose of refining the local structure of the backbone and side-chain atoms. The position of ligand residues in modeled structures of the three RING domains (0, 1, and 2) suggest that each domain can bind two zinc ions. We also modeled the Ubl and the IBR domain of Parkin. However, as our results are in agreement with the known structure of other ubiquitin conjugating proteins and with those presented earlier by Tomoo et al. (2008) [22] and Beasely et al. (2007) [17], we will not present our results here.

Model of RING1

The fold of RING1 closely matches the RING finger domain of human UbcM4-interacting protein (PDB: 1WIM). The predicted structure of the RING1 domain (Fig. 2A) and the threading alignment with the identified template, suggest that RING1 adopts the classical zinc ligand spacing [12, 102, 103] (Fig. 1B, 2B) and assumes a typical cross brace structure (Fig. 1C) with residues C238, C241, C260 and C263 as ligands in the first zinc co-ordination site and C253, H257, C289 and C293 present in the second site.

Model of RING0

Our predicted structure for the RING0 domain is different from the previously proposed model. For example, a model for RING0 structure was proposed by Hristova. et al. consisting of two zinc binding sites containing (C4 and C3H) separated by a linker region (Fig. 1A). Because the linker region separating the zinc binding motifs is very large (26 amino acid) and because HOIL-1, the only other RBR protein to contain a zinc binding motif upstream of RING1, has only one zinc binding motif, Hristova et al. (2009) [19] proposed a model in which RING0 binds two zinc atoms in an independent and linear fashion rather than in a cross brace structure. However, a clear picture of the binding mode of zinc ions in the RING0 domain remained elusive.

Structure modeling of RING0 using the automated fold recognition procedure for RING0 did not identify any significant templates and therefore the model quality was low because the templates had diverse folds. Accordingly, the template proteins in the PDB library were filtered based on the presence of bound zinc ions and a linker region of at least 20 residues between the zinc coordination sites. The search resulted in identification of three DNAJ/hsp40 cysteine rich domains (PDB: 1exk), with sequence conservation mainly at/or near ligand binding sites.

The overall topology of the predicted model for RING0 (Fig. 3A and B) is based on the DNAJ/hsp40 homology and differs from that proposed by Hristova et al. (2009) [19] in that the two zinc binding sites are interleaved, with a V-shaped extended β-hairpin topology in the linker region, similar to the cysteine rich domain of DNAJ. The long linker apparently does not impair binding of zinc ligands. The rest of the domain contains only a limited amount of regular secondary structure. The first binding site consists of Ligands one (Cys 150), two (Cys 154), seven (C212) and eight (H215), while the second is composed of Ligands 3-6 (C166, C169, C196, and C201). While the predicted model requires experimental validation, it provides the first plausible three dimensional model for zinc binding in RING0.

Model of RING2

A current model for RING2 structure is based on zinc ion ligand residues C418, C421, C431, H433, C436, C441, and C449 identified by Morett and Bork (1999) [104]. The model fits the classical ligand spacing [12, 102, 103, 105] (Fig. 2B), however spatial limitations prevent it from easily fitting into a cross-brace structure (Fig. 1C). In the Morett and Bork model, C418, C421, C431, H433 coordinate to the first zinc ion, but Capilli et al. (2004) showed that the RING2 domain of HHARI, a protein that shares both the RBR domain structure and the E3 ligase function of RING2 with Parkin, did not form a cross brace structure and binds only one zinc ion [18]. By analogy it was suggested that the RING2 domain of Parkin may also bind just one zinc atom (Fig. 4A). However, this contrasts with the zinc titration analysis that suggested that Parkin can bind a total of eight zinc ions, two each for RING0, RING1, IBR, and RING2 [19].

While analyzing the predicted models of RING2 and the alignment with identified template proteins (RING2 of HHARI and IBR domain of RING finger domain of protein 31), we found that another conformation was possible for Parkin RING2 that uses a different set of ligands to accommodate two zinc ions. This model suggests that C418, C421, C436 and C441 coordinate to the first zinc ion while C446, C449, C457 and H461 coordinate to the second zinc ion (Fig. 4B, 4C). The zinc coordination sites in the predicted model are similar to the IBR domain of RING finger protein 31, which also binds two zinc ions. With regard to spacing, the previous model of RING2 follows the Lovering and Freemont RING spacing, except for the presence of 4 rather than 2 amino acids between ligand 5 and 6 (Fig. 2B). Our predicted structure follows the Lovering and Freemont RING spacing except there are 4 instead of 2 amino acid residues between ligand 4 and 5, and 3 rather than 2 residues between ligand 7 and 8 (Fig. 4C). A major point of difference is that in our predicted structure, the RING2 domain boundary, defined as being 4 residues before the first Cys and 6 residues after the last ligand (a histidine rather than cysteine in our predicted structure) is followed by only 4 amino acid residues rather than the 6 residues, as defined by Freemont [103]. The predicted RING2 structure does not form the classical cross brace structure and the last amino acid involved in zinc binding is His461 (Fig. 4B). Therefore our predicted structure varies in spacing, in the RING domain boundary, and also in location of the histidine. Wong et al. (2007) [21] had earlier proposed a similar first zinc ion ligand binding site (C418, C421, C436 and C441) for RING2 based on homology modeling.

Our RING2 structural model can explain several previously puzzling observations about Parkin. One is the number of zinc ions bound by Parkin,. Our model predicts two for RING2 for a total of eight in full length Parkin as demonstrated by Hristova et al. (2009) [19]. Another is the observed instability of Parkin C-terminal deletions and chimeric Parkin-HHARI proteins [29, 30]. In our predicted RING2 structure the deletion resulting from the W453X mutation would delete the final two ligands, C457 and H461 of the second zinc binding domain, explaining why this mutant spontaneously adopted an unfolded conformation [29]. Further, any C-terminal deletion greater than three amino acids resulted in the loss of Parkin solubility [29]. While H461 is the last ligand and fifth residue from the C-terminal, the importance of W462 in Parkin has remained largely unknown. The role of the highly conserved tryptophan residues in RING structures, e.g. Parkin W453 and W462, have been thought to be important for making interactions with other proteins, however the new insight into the metal-ion-recognition property of the indole ring of tryptophan suggests that the aromatic rings of these residues can form non-covalent interactions with zinc ions to provide rigidity to the RING structure and/or cause conformational changes leading to binding of zinc ions [106]. Therefore, as shown by Winkelhofer et al. (2003) [29] the last three residues, F463, D464, and V465, are the only dispensable C-terminal amino acids in Parkin RING2.

In contrast to Parkin, HHARI tolerated C-terminal deletions [30]. To address this difference, chimeras of the two proteins were prepared in which C-terminal fragments of HHARI from residues 376, 379 or 395 onwards, were substituted for Parkin residues 449 or 453 onwards. The chimeras were assayed for stability in situ [30]. In all of the chimeric proteins, the final two putative zinc ligands (C457 and H461) in our predictd RING2 model were substituted by amino acids G, Y, H, W, K or D. In one case, a His residue replaced the C457 ligand, however this construct was still missing the zinc binding ligand at position 461. All of the chimeras proved to be unstable, supporting the hypothesis that Parkin RING2 contains two zinc binding sites, and, as predicted by the model, both are important for Parkin stability.

Our RING2 model also explains the unexpected result regarding the behavior of the C431A mutation in an ambitious study in which the Parkin cysteines were individually mutated to determine which ones affected Parkin solubility [21]. The C431A mutation remained soluble, which was puzzling as all other cysteine mutants resulted in insoluble Parkin. Although C431 is considered as a ligand in the first zinc binding site in the Morett and Bork RING2 structure, C431 is not a ligand in either of the two zinc binding sites of our predicted RING2 structure (Compare Fig. 4A with 4B). This provides further support for the predicted RING2 structure.

Another previously puzzling observation came from Hampe et al. (2006) [27] in which mutation of either C418 or C441 in Parkin resulted in loss of autoubiquitination activity. Using the Morett and Bork (1999) ligand selection, C418 was a part of the first zinc binding site, whereas C441 was in the second zinc binding site. If Parkin RING2 was similar to HHARI RING2, we would expect the second zinc site to be unimportant for catalytic activity and hence the mutation of C441 would retain activity. However, that is not the result obtained by Hampe et al. (2006). Using our predicted RING2 structure, ligand C441 is a member of the first zinc binding site and therefore, like C418, should inhibit Parkin ubiquitination activity when mutated. This is consistent with the Hampe et al. (2006) [27] results.

From observations of zinc binding sites in RING domains discussed in Kentsis et al. [107], it appeared that zinc binding sites may show negative cooperativity in zinc binding, i.e. binding of zinc in one binding site reduces the affinity for binding of zinc to a second site. Considering that the new conformational structure proposed for the RING2 domain would explain several earlier observations, we hypothesize that Parkin assumes this conformation, at least part of the time. Whether or not the two zinc binding sites remain occupied continually, or whether Parkin can function (or remain soluble) with just one zinc ion bound remains to be determined. Indeed, domains with just one bound zinc ion have been shown to self-assemble into stable multi-RING structures that enhance the function of individual RING domains [107]. Based on these and related observations [108-110], an attractive possibility is that, in response to changing physiological conditions, a transition between bound and unbound zinc ions at one of the two zinc binding sites of Parkin RING2 promotes a change in local structure that acts as a mechanical switch, providing a mechanism whereby Parkin-protein association may be altered resulting in a change in function. Although an interesting hypothesis, a significant amount of work remains to be done with Parkin RING2 in exploring that hypothesis.