Quantitative proteomic analysis of Parkin substrates in Drosophila neurons

Drosophila park gene was amplified from a Drosophila cDNA library (DGC realease 1.0, Berkeley Drosophila Genome Project) and FLAG-tag cloned at its 5’-end using the FLAG-parkin-Fw (GCCCTCGAGATGGATTACAAGGATGATGACGATAAGATGAGTTTTATTTTTAAATTTATTGCCAC) and parkin-Rv (GCCTCTAGATTAGCCGAACCAGTGGGCTCC) primers. This construct was then inserted into a pUASattb vector between the XhoI and XbaI sites. Ligase-dead FLAG-Parkin (Parkin^LD) was generated by mutating the C449 to S using the QuikChange Site-Directed Mutagenesis Kit (Stratagene) according to manufacturer’s instructions. The primers used for mutagenesis were C449S-Fw (GGAGCGAGATGGCGGTAGCATGCACATGGTCTGCACACG) and C449S-Rv (CGTGTGCAGACCATGTGCATGCTACCGCCATCTCGCTCC). Untagged human Parkin and Parkin (C431S) were amplified from pcDNA3.1(+)-HA-Parkin and pcDNA3.1(+)-HA-Parkin(C431S) respectively with primers forward (GCCGAAGCTTAACCATGATAGTGTTTGTCAGG) and reverse (AGTCTAGACTACACGTCGAACCAGTGGTCCTGGG). PCR products were inserted into pcDNA3.1(+) between HindIII and XbaI.

The following antibodies were used against Drosophila proteins: goat anti-biotin-horseradish peroxidase (HRP) conjugated antibody (Cell Signalling); chicken polyclonal anti-BirA antibody (Sigma); rabbit polyclonal anti-Parkin antibody [44]; mouse monoclonal anti-Syx1A antibody (DSHB); rabbit polyclonal anti-RdhB [45]; rabbit polyclonal anti-ArgK [46]; rabbit polyclonal anti-Vps4 [47]; rabbit polyclonal anti-Fax antibody (a gift from Eric Liebl); rabbit polyclonal anti-Ubiquitin antibody (Sigma). The following antibodies were used against Human proteins: goat polyclonal anti-VPS35 antibody (Abcam); mouse monoclonal anti-Cleaved Parp-85 fragment (Cell Signaling); mouse monoclonal anti-Parkin (Santa Cruz); rabbit polyclonal anti-PINK1 (Novus Biologicals); rabbit polyclonal anti-Miro1 (Sigma); rabbit polyclonal anti-Tim44 (Sigma); rabbit polyclonal anti-Tom20 (Sigma); mouse monoclonal (Abcam) and rabbit polyclonal (Sigma) anti-Actin. For monitoring the GFP pull-downs the following antibodies were used: monoclonal mouse anti-GFP antibody (Roche) and monoclonal mouse anti-Flag M2-HRP conjugated antibody (Sigma). Anti-mouse, rabbit and chicken HRP labelled secondary antibodies (Jackson ImmunoResearch Laboratories) and anti-guinea pig (Invitrogen) were used; and anti-mouse, rabbit and sheep IR 680 and IR800-coupled antibodies (LI-COR Biosciences).

UAS-BirA and UAS-( ^Bio Ub) ₆ -BirA [39] and their recombination with GMR-GAL4 flies for the study of ubiquitin proteomics has been previously described [43]. FLAG-tagged Parkin wild-type (Parkin^WT) and Parkin^LD flies were generated at Bestgene using the pUASattb constructs described above. Both UAS-Parkin ^WT and UAS-Parkin ^LD lines were independently crossed with GMR-GAL4,UAS-( ^Bio Ub) ₆ -BirA to finally generate: GMR-GAL4,UAS-( ^Bio Ub) ₆ -BirA/CyO;UAS-Parkin ^WT and GMR-GAL4,UAS-( ^Bio Ub) ₆ -BirA/CyO;UAS-Parkin ^LD . GMR-GAL4/CyO;UAS-BirA/TM6 and GMR-GAL4,UAS-( ^Bio Ub) ₆ -BirA/CyO flies were additionally used as controls. UAS-GFP, elav-GAL4, GMR-GAL4, Tub-GAL4, Da-GAL4, Ple-GAL4 flies were obtained from Bloomington Drosophila Stock Center. UAS-GFP ^CL1 flies were obtained from [48] and park ²⁵ /TM6b GFP-w ⁺ and UAS-park were obtained from [16]. Flies were grown in 12 h light-dark cycles at 25 °C and were fed with wheat flour and yeast food (1% agar, 5.5% dextrose, 3.5% wheat flour, 5% yeast, 0.25% Nipagen, 0.4% Propionic acid and 0.02% Benzalkonium Chloride in distilled H₂O).

Flies of indicated ages and genotypes were anesthetised with CO₂ on a pad, and 20 flies (10 male and 10 female) were randomly selected. After an hour of recovery, flies were transferred to a climbing vial and ability to climb was scored as followed. Flies were gently tapped to the bottom and the number of flies that reached the 10 cm mark at 30 s was counted three times, with 30 s interval.

One hundred newborn flies of the indicated genotypes were maintained in wheat flour and yeast food in 12 h light-dark cycles at 25 °C. The vials were changed every 2–3 days and the number of flies alive was counted.

Six heads (three males and three females) of adult flies were cut and homogenised in 60 μL of 4x Laemmli buffer with DTT. Samples were centrifuged 1 min at maximum spin and the supernatant was recovered. Cells were harvested using “Hot Lysis buffer” (2% SDS, 50 mM NaF and 1 mM EDTA at 110 °C; Fig. 8). Protein concentrations were determined by BCA protein assay (Pierce).

hTERT-RPE1-Parkin cells [32, 49] were cultured in Dulbecco’s modified Eagle’s medium (DMEM)/F12 with 10% FBS, 1% non-essential amino acids and 1% penicillin/streptomycin. Reverse transfection was performed using Lipofectamine RNAi-MAX (Invitrogen) and carried out for 72 h according to manufacturer’s instruction. Further 165 h double-hit knockdown was executed by transfecting again previously siRNA treated cells at 72 h. Cells were transfected with the following siRNA at a final concentration of 20nM: Non-Targeting siRNA oligo #1 (NT1) (ONTARGETplus: NT1; 5’-UGGUUUACAUGTCGACUAA-3’) and VPS35 (SMARTpool ONTARGETplus VPS35 siRNA oligo#5: 5’-GAACAUAUUGCUACCAGUA-3’; oligo#6: 5’-GAAAGAGCAUGAGUUGUUA-3’; oligo#7: 5’-GUUGUAAACUGUAGGGAUG-3’; oligo#8: 5’-GAACAAAUUUGGUGCGCCU-3’) from Dharmacon.

Human neuroblastoma SH-SY5Y cells were cultured in DMEM supplemented with 10% FBS (Thermo Scientific) and 1:100 penicillin/streptomycin at 37 °C. 300,000 cells were seeded in a 6 well-plate and incubated overnight under serum starvation. The following day, medium was removed, replaced by fresh DMEM and cells were transfected with 1 μg of FLAG-Ub, 1 μg of YFP-VPS35 together with 1 μg of pcDNA3.1 (control) or 1 μg of hParkin or 1 μg of hParkin (C431S) using Lipofectamine 3000 (Invitrogen) for 72 h according to manufacturer’s instruction.

Transfected cells were lysed in 500 μl of lysis buffer [50 mM Tris–HCl pH 7.5, 150 mM NaCl, 1 mM EDTA, 0.5% Triton, 1 × Protease Inhibitor cocktail (Roche Applied Science), 50 mM N-ethylmaleimide (NEM, from Sigma)] and collected for centrifugation at 14,000 × g for 10 min. The supernatant was mixed with 25 μL of GFP-Nanotrap beads (Chromotek GmbH) that had been pre-washed with dilution buffer (10 mM Tris–HCl pH 7.5, 150 mM NaCl, 0.5 mM EDTA, 1 × Protease Inhibitor cocktail, 50 mM NEM). The mixture was incubated at RT for 150 min with gentle rolling and centrifuged for 2,700 × g for 2 min. The supernatant was removed and the beads washed once with dilution buffer, three times with washing buffer (8 M Urea, 1% SDS in PBS) and once with 1% SDS in 1× PBS. The bound proteins were eluted with sample loading buffer (250 mM Tris–HCl pH 7.5, 40% glycerol, 4% SDS, 0.2% BPB) by heating at 95 °C for 10 min. Eluted samples were run into 4–15% Tris–Glycine gels.

For fly samples the amount of sample loaded largely was equivalent to one head per lane but was optimized for each antibody. Generally, 4–15% gradient TGX gels (Bio-Rad) were used and proteins were transferred to PVDF membranes using the iBlot system (Invitrogen). Following primary and secondary antibody incubation, membranes were developed using ECL kit (Biorad Clarity). Alternatively, western blots were scanned using the LI-COR Odyssey system (LI-COR Biosciences). Silver staining was performed with a SilverQuest kit (Invitrogen) following the manufacturer’s instructions. Dual-colour westerns were prepared by assigning independent colour channels to two independent westerns developed in the same membrane.

The ^BioUb pulldown was performed as previously described [39, 40, 43] using GMR-GAL4,UAS-( ^Bio Ub) ₆ -BirA/CyO; UAS-Parkin ^WT , GMR-GAL4,UAS-( ^Bio Ub) ₆ -BirA/CyO;UAS-Parkin ^LD , GMR-GAL4/CyO;UAS-BirA/TM6 and GMR-GAL4, UAS-( ^Bio Ub) ₆ -BirA/CyO flies. Briefly, 2–5 day old adult flies were fragmented by flash freezing in liquid nitrogen and shaking. A pair of sieves with a nominal cut-off of 710 and 425 μm was used to separate the heads from the rest of the fragments. About 0.5 g of heads were homogenized in 2.9 mL Lysis buffer (8 M Urea, 1% SDS and 50 mM N-ethylmaleimide in PBS, including a protease inhibitor mixture from Roche Applied Science). 250 μL of NeutrAvidin-agarose beads (Thermo Scientific) were used to purify ^BioUbiquitinated material and after stringent washes with a succession of buffers containing 8.4 M urea, 6 M Gdn-HCl, 1 M NaCl and 2%SDS, beads were finally eluted with 125 μL Elution Buffer (4X Laemmli Buffer and 100 mM of DTT). Recovered volume was ~150 μL with an average yield of 20–40%.

Eluted samples from three different experiments were briefly run on a SDS-PAGE, which was then cut in four slices, to separate the avidin band from the other proteins in the gel. Proteins were digested using automated in-gel digestion protocol [50]. The digested peptides were cleaned-up on Stage-tips [51]. The eluted peptides were resuspended in 0.1% formic acid and separated on a 15 cm reverse phase column (75 μm inner diameter, 3 μm reprosil beads, Dr. Maisch, GmbH, in house packed) using 5 to 50% acetonitrile gradient (VWR). Peptide ionization was performed on a Proxeon ionsource and sprayed directly into the mass spectrometer (Q-Exactive, Thermo Scientific). The MS was recorded with mass resolution of 60000, while the MS2 spectra were collected with a resolution of 35000 using a fixed fill time of 120 ms. For the analysis of the recorded mass spectra the MaxQuant software package (version 1.2.2.5) was used with 1% FDR for both peptides and proteins [52]. Searches were performed using the Andromeda search engine against the Uniprot Drosophila melanogaster database (downloaded 15.08.2015). As a fixed modification, cysteine carbamidomethylation was selected and as variable modifications, methionine oxidation, protein N-terminal acetylation and di-glycine addition on the ε-aminogroup of lysines. Two missed trypsin (full specificity) cleavages were allowed and the MaxQuant label free quantification for proteins was activated. The identified ubiquitination sites were checked visually using the Max-Quant View program and identification artefacts were removed as previously described [53].

Samples from three technical replicates were loaded on an 8% stacking gel and proteins compressed in one band. Proteins were digested in-gel using trypsin, and purified as described above. The peptides were separated on a 25 cm reverse phase column (75 μm inner diameter, 3 μm reprosil beads, Dr. Maisch, GmbH, in house packed) using 5 to 50% acetonitrile gradient (VWR). The peptides were ionized on a Nano3 ion source (ABSciex) and directly sprayed into a QTRAP 5500 triple-quadrupole mass spectrometer, run in MRM mode to detect branched ubiquitin peptides and their synthetic isotopically labelled counterparts [54, 55]. Each sample was injected as technical triplicates. Two transitions for each peptide, i.e., K6, K11, K48, K63, were acquired. After the measurement peaks were integrated and light/heavy ratios calculated using the Skyline software package [56] reflecting the abundance of the linkage type compared to the internal heavy standard.

Non-ubiquitinated/background proteins were excluded by a 4-fold change threshold of the average LFQ intensity between ^BioUb, ^BioUb + Parkin^WT or ^BioUb + Parkin^LD against the BirA sample. LFQ intensity values of ubiquitinated proteins were imported to Perseus software (http://​www.​perseus-framework.​org/​) and the imputation tool was used to replace missing values. Subsequently, values were grouped in categories and two T-tests were performed: ^BioUb + Parkin^WT vs ^BioUb and ^BioUb + Parkin^WT vs ^BioUb + Parkin^LD. Proteins with a p-value smaller than 0.05 and a fold change bigger or smaller than 2 in any of these two tests were selected for further filtering. Proteins with less than two unique peptides in all the three ^BioUb + Parkin^WT (enriched) and ^BioUb + Parkin^LD (reduced) were excluded. To further exclude false positives, all unique peptide profiles of these proteins were analysed. Peptides that appeared only in one condition were excluded and the average peptide intensity was calculated for each peptide in each condition. On the one hand, average of all peptide intensities between ^BioUb + Parkin^WT vs ^BioUb or ^BioUb + Parkin^LD were calculated and only the proteins whose total fold change was bigger (enriched) or smaller (reduced) than two were further selected. On the other hand, the average of all fold changes of peptide intensities between ^BioUb + Parkin^WT vs ^BioUb or ^BioUb + Parkin^LD minus the SEM (Standard Error of the Mean) was required to be bigger or smaller than two. Only the proteins that successfully passed all the tests were selected as: Parkin substrates (35) and less ubiquitinated in a Parkin - dependent manner (2).

A two sample T-Test for Parkin proteomics was performed using Perseus software applying default parameters: t-test; S0; Side = both; threshold p-value = 0,05. Imputation tool was used as default: width = 0,3; down shift = 1,8; mode = separately for each column. Other statistical analysis was performed with Graphpad Prism6. For the GO Term identification g:Cocoa tool of g:Profiler web server was used using archive “Ensembl 79, Ensembl Genomes 26 (r1395)”. p-values are indicated as *p ≤ 0.05, **p ≤ 0.01 and ***p ≤ 0.001. Functional connectivity analysis was performed using STRING (http://​string-db.​org/​) web tool.

For GFP pull-downs protein bands were analysed and quantified by densitometry using ImageJ software. Statistical significance in Western blotting quantification was evaluated using an analysis of variance (ANOVA) complemented by Tukey’s honest significance difference test (Tukey’s HSD) performed in GraphPad PRISM software. Statistical significance levels are annotated as ns, not significant; **, p < 0.01 (mean ± SEM, n = 4).

In order to study Parkin E3 ubiquitin ligase activity in Drosophila neurons, we over-expressed FLAG-Parkin^WT (Fig. 1a) and a catalytically inactive or ligase-dead mutant, FLAG-Parkin^LD, using the neuronal specific elav-GAL4 driver (Fig. 1b). As expected, heterozygous FLAG-Parkin flies expressed lower Parkin levels compared to homozygous Parkin flies (Fig. 1c, Additional file 1: Figure S1A). Neuronal over-expression of Parkin^WT during Drosophila development compromises climbing ability of the fly in an age and dose-dependent manner, mirroring the phenotype of Parkin null flies (Fig. 1d, Additional file 1: Figure S1B). Expression of a ligase-dead form of Parkin generates a much milder climbing impairment. The N-terminal FLAG tag did not contribute to this behaviour as both untagged Parkin^WT and 8HA-Parkin^WT flies display the same phenotype (Additional file 1: Figure S1C). High-level expression of Parkin^WT also led to a decreased life span (Fig. 1e) and, in distinction to Parkin^LD flies, Parkin^WT flies displayed a Parkinsonian phenotype, suffering from tremors and spontaneous spasms (Additional file 12: Video S1).

We recently introduced a pipeline for the identification of ubiquitinated proteins in a fly model that constitutively expresses biotinylated ubiquitin (^BioUb). Here, we have applied this approach to identify physiologically relevant Parkin substrates in adult Drosophila neurons (Fig. 2a). ^BioUbiquitin together with either Parkin^WT (WT) or Parkin^LD (LD) was over-expressed in Drosophila photoreceptors, a specialised type of neuron, using GMR-GAL4 driver (Fig. 2b). Flies over-expressing just BirA (C) and ^BioUbiquitin (Ub) were used as background control and steady state ubiquitination control respectively. Endogenous deubiquitinating enzymes (DUBs) digest the ^BioUbiquitin construct releasing BirA (Fig. 2c) and ubiquitin. Each ubiquitin contains a short N-terminal sequence, which is then recognised and biotinylated by BirA, resulting in a biotin-tagged ubiquitin (^BioUb) that is efficiently handled by the cascade of ubiquitin conjugating enzymes (Fig. 2d) [39‐43]. ^BioUb conjugates were isolated on Neutravidin beads (Fig. 2e, f) and subsequently identified by Mass Spectrometry (MS).

Three independent experiments were performed for each condition (Additional file 2: Figure S2A), generating 1522 protein identifications by mass spectrometry (Additional file 9: Table S1). After background subtraction of the BirA control, 1132 proteins were classed as ubiquitinated by virtue of their enrichment (Fig. 3a). Most of the ubiquitinated proteins were identified in each of the three different conditions (Fig. 3b), suggesting that the ectopic expression of Parkin does not perturb the overall global ubiquitin economy. In fact, the number of proteins detected to be ubiquitinated in each pulldown was very similar across all samples (Additional file 3: Figure S3A) and Label Free Quantification (LFQ) values of independent pulldowns were highly correlated (Additional file 4: Figure S4). For proteins not directly quantitated in one experimental condition (LFQ = 0) we used the Perseus imputation tool to assign an arbitrary value within the lower detection range (Additional file 3: Figure S3B).

Fifty-six proteins were discovered to be both more than 2-fold enriched and reach a significance of p < 0.05 (Perseus T-test, see Materials and Methods). In contrast, 30 proteins were reduced (using the same criteria) in purified ubiquitinated material from Parkin^WT over-expressing neurons compared to controls (Fig. 4, Additional file 6: Figure S6A). We next applied a further set of criteria to select the most robust Parkin responders (35 enriched, two reduced). These all provide at least two unique peptides in all three biological replicates and display a fold change bigger than two as average of all unique peptides (see Materials and Methods). Seven of the 35 enriched proteins are annotated as mitochondrial proteins, 12 as cytosolic and 13 as part of endomembrane system according to UNIPROT database (http://​www.​uniprot.​org/​) (Table 1, Additional file 10: Table S2). These include three sub-units of the v-ATPase required for endosome and lysosomal acidification, two proteins associated with the endosomal sorting complexes required for transport (ESCRT) machinery (ALiX, Vps4) and the PD-associated retromer component, Vps35. The membrane receptor NinaE (Opsin/Rhodopsin homolog) and the cytosolic calcium binding protein, Cbp53E are robustly, less ubiquitinated in a Parkin E3 ubiquitin ligase activity dependent manner.

Trypsin treatment of ubiquitinated proteins leaves a di-Gly signature residue at the ubiquitination site. Although the majority of peptides we use for quantitation are not themselves ubiquitinated, we have captured a set of 111 ubiquitinated peptides (Additional file 11: Table S3). Such di-Gly peptides were detected for five of the Parkin substrates identified as described above (Cisd2, EIF2B-beta, Pdh, Porin and SesB) and for the two proteins that were less ubiquitinated in a Parkin context (NinaE and Cbp53E) (Fig. 5). Di-Gly peptides corresponding to the following ubiquitination sites, Cisd2 K83, EIF2B-beta K30, Pdh K40, Porin K52 and SesB K120 were enriched in Parkin^WT over-expressing flies. However, Pdh K85 was found in a Parkin-independent manner, as similar intensities were detected in all samples. Di-Gly peptides for Cbp53E K262 and NinaE K372 were reduced in Parkin^WT flies. Moreover, 3 di-Gly peptides within Parkin itself were also detected that showed differing enrichment patterns. K35 was more ubiquitinated in Parkin^WT compared to both Parkin^LD and control. K40 was similarly ubiquitinated in Parkin^WT and Parkin^LD, but not in control and K128 was much more ubiquitinated in Parkin^LD compared to Parkin^WT and control.

Ubiquitin chains can be assembled through isopeptide bonds with any of the seven internal lysine residues. Each linkage type generates a unique peptide signature upon trypsin digestion that can be discriminated by mass spectrometry and quantitated by comparison with isotopically labelled standard [55]. We adopted this strategy to identify the linkage types of the ubiquitin chains that are generated by Parkin over-expression in vivo, by analysis of isolated ^BioUb material. As judged by western blotting neither the overall pattern of ubiquitination nor the total amount of ubiquitinated material were altered by Parkin^WT over-expression in Drosophila neurons (Fig. 6a). The same conclusion can be obtained based on the Venn diagrams that compare the ubiquitomes identified upon Parkin over-expression with the control samples (Fig. 3b). However, we find a marked increase in the representation of K6 linked ubiquitin chains in Parkin^WT compared to control ^BioUb and Parkin^LD samples (Fig. 6b, Additional file 6: Figure S6B).

We sought to validate ubiquitination of selected Parkin substrates by direct western blotting of proteins for which antibodies were available. For ArgK, RdhB and Vps4 we observe a fraction that is uniformly upshifted to a degree consistent with monoubiquitination in a manner that is stringently dependent on Parkin activity (Fig. 7a). In contrast, Fax, a protein that was not specifically enriched following Parkin expression in the mass spectrometry data set, is indiscriminately ubiquitinated in all samples (Fig. 7b). Parkin is enriched largely in an unmodified form in the eluates of flies over-expressing Parkin (Fig. 7c). This likely represents Parkin that is captured by virtue of a thioester-bonded ubiquitin at its active site cysteine, which is then released by the reducing conditions of the subsequent processing steps. The residual minor ubiquitinated fraction is linked via isopeptide bonds, suggesting that a small fraction of Parkin is itself ubiquitinated.

Parkin-dependent Vps35 ubiquitination could not be validated in flies by western blotting, due to the lack of available working antibodies for Drosophila Vps35. We instead performed in human cells an in vivo ubiquitination assay we recently developed [42, 43, 57, 58]. The main advantage of this method is that, due to the very stringent washes performed, the entire ubiquitin signal detected corresponds exclusively to the protein of interest and cannot be derived from other proteins in the cell. We tested this assay on SH-SY5Y neuroblastoma cells, which allowed us to further test whether Parkin-dependent ubiquitination of VPS35 is conserved in human cells.

SH-SY5Y cells were transfected with YFP-VPS35, FLAG-Ub and either WT or LD untagged human Parkin. Cells were then subjected to a GFP pulldown and the eluted samples monitored by Western blotting in order to compare ubiquitination levels of VPS35 across the different samples (Fig. 8a, Additional file 7: Figure S7). Since the ubiquitination signal proceeds only from YFP-VPS35, quantification of the data (Fig. 8b) is provided as ratios of the ubiquitin signal (monitored through the anti-Flag western) over the VPS35 intensity (monitored through the anti-GFP western). A significant increase in the ubiquitinated fraction of VPS35 was observed when WT Parkin was co-expressed, relative to either control (only endogenous Parkin or LD Parkin co-expression).

Our results confirm a linkage between Parkin and the PD-associated protein VPS35, which has been previously revealed by genetic studies [59]. Both the proteomic data and the in vivo ubiquitination assay coupled to subsequent western blot validation suggest that VPS35 may be a substrate of Parkin, in line with previous studies that analysed Parkin substrates following acute mitochondrial depolarisation [38]. The mechanism that links VPS35 to PD is unclear, although recent evidence indicates that VPS35 regulates mitochondrial homeostasis by regulating fission-fusion dynamics [60, 61] and the trafficking of mitochondrial-derived vesicles (MDV) [62, 63]. We and others have recently used retinal pigment epithelial cells (RPE1) that stably over-express YFP-Parkin as a convenient system for analysis of Parkin-dependent mitophagy [32, 49]. We tested whether VPS35 is necessary for PINK1/Parkin - dependent mitophagy in these cells following acute mitochondrial depolarisation. However, siRNA-mediated knockdown of VPS35 had no effect upon mitophagy as judged by the time-dependent loss of several mitochondrial marker proteins (Fig. 8c).

Our finding that neuronal Parkin over-expression during neurodevelopment leads to multiple Parkinsonian-like defects conflicts with the simple notion that Parkin is neuroprotective [29]. However, recent cell culture and in vivo studies have shown that Parkin activity can also sensitize towards cell death in certain contexts [30‐33]. We observed that the severity of Parkin over-expression phenotype is associated with its E3 ubiquitin ligase activity in a dosage dependent manner, which is further augmented during aging. Nevertheless, the fact that flies over-expressing Parkin^LD still have a mild phenotype, suggests that this mutant shows a dominant-negative effect that is not linked to the ligase function. Further in vivo experiments are necessary to unravel Parkin function at different developmental stages in different tissues.

Identification of Parkin substrates is crucial in order to comprehend the still elusive aetiology of PD. While early studies stressed the notion that Parkin ubiquitinates and degrades several aggregation-prone toxic proteins, currently Parkin is mainly associated with mitophagy, through ubiquitination of several OMM proteins [35, 36, 64‐66]. Parkin is predominantly held in an inactive conformation at steady state [20] and PINK1 dependent phosphorylation of both ubiquitin and Parkin is necessary for its activation [19‐25]. Mitochondrial depolarisers have been proven to efficiently activate Parkin through PINK1 accumulation, but most cell lines show low levels of Parkin, precluding detectable levels of Parkin-dependent mitophagy and ubiquitination. Moreover, the ubiquitinated fraction of proteins is usually low due to their fast proteasomal degradation and/or the activity of deubiquitinating enzymes (DUBs) [67]. Therefore, most studies have been performed under conditions of Parkin over-expression using non-neuronal in vitro cell culture models, usually employing proteasome inhibitors and mitochondrial depolarisers. However, these acute conditions are not encountered in physiological settings. In the present study we have sought for in vivo Parkin substrates in Drosophila neurons, using a fly model that co-expresses constitutively biotinylated ubiquitin together with FLAG-tagged Parkin. While it has been shown that tagging human Parkin can lead to some constitutive activation [68], Drosophila Parkin has an extended N-terminal sequence relative to the human protein and addition of a further short sequence tag is not predicted to impact on the structure (Additional file 8: Figure S8). If such a scenario was extant, it would simply shift an equilibrium and potentially allow more sensitive detection of Parkin substrates without the recourse to acute drug treatments used in previous studies [27, 34, 36‐38]. Treatments with mitochondrial depolarising drugs will lead to PINK1 accumulation and activation of Parkin specifically at mitochondria. Our study provides a complementary resource, which may capture some Parkin substrates that are PINK1-independent. The notion of alternative activation routes for Parkin has been previously proposed [69] and it is supported by data suggesting a role for Parkin in xenophagy that might be PINK1-independent [70].

We identified 37 proteins whose ubiquitination pattern changes in a Parkin E3 ubiquitin ligase activity dependent manner. 35 proteins were enriched and 2 were reduced in purified ubiquitinated material of Parkin over-expressing head extracts (Fig. 9). Western blot validation of proteins for which Drosophila antibodies were available confirmed several of these proteins as authentic in vivo Parkin substrates. A number of mitochondrial proteins were identified, including proteins known to be ubiquitinated during mitophagy, such as Cisd2/CISD1,2 (MitoNEET), Porin/VDAC1,2,3 and Tom70 [27, 34, 36‐38, 71]. Nevertheless, in comparison to other recent Parkin proteomic studies, our Parkin substrate data set shows less enrichment of mitochondrial proteins, probably reflecting the lack of acute mitochondrial depolarization [27, 34, 37, 38]. Whilst some of the mitochondrial proteins are ultimately matrix associated, they are synthesised in the cytosol and imported through the TIM/TOM complex providing the opportunity to encounter cytosolic Parkin. Interestingly, the presence of ubiquitinated proteins within the yeast mitochondrial matrix has recently been reported [72]. In fact, a recent paper suggested that Parkin ubiquitinates and regulates the stability of the mitochondrial matrix enzyme 17-β hydroxysteroid dehydrogenase type 10 (HSD17B10) [73]. Further proteins previously detected also include Sply/SGPL1, VPS4, VPS35, ALiX/PDCD6IP and CG5525/CCT4 [37, 38]. To our knowledge, the remaining 27 enriched proteins are novel Parkin substrates, which are predominately cytosolic and endosomal proteins. Inversely, NinaE/Opsin, Rhodopsin and Cbp53E/Calbindin were reduced in purified ubiquitinated material after Parkin over-expression indicating that Parkin can indirectly regulate the ubiquitination status of these proteins. It is also plausible that Parkin-dependent ubiquitination followed by rapid degradation result in reduced protein levels leading to decreased ubiquitination status of these substrates. Importantly, decreased calbindin mRNA and protein levels are linked to neurodegeneration in PD [74‐77].

The fact that there is no singular pathway controlled by substrates described here suggests that under in vivo physiological conditions Parkin may have broader roles beyond the control of mitophagy. For example, we note prominent representation of endosomal-associated proteins, including three sub-units of the v-ATPase required for endosome and lysosomal acidification, two proteins associated with the endosomal sorting complexes required for transport (ESCRT) machinery (ALiX/PDCD6IP, VPS4) and the PD-associated gene and retromer component VPS35. Moreover, proteins involved in protein synthesis, metabolic process, cell death, calcium signaling and immune response are also present (Additional file 10: Table S2). Further study of Parkin substrates will reveal whether they act in an unexplored common pathway.

A single ubiquitin can be attached to a unique (monoubiquitination) or several (multiple-monoubiquitination) lysine residues of substrate proteins. Ubiquitin can be generated from these stubs to form chains of different topologies (M-1, K6, K11, K27, K29, K33, K48, K63). This provides a complex “ubiquitin code” that regulates a myriad of biological processes. Whilst K48 regulates protein turnover, through proteasomal degradation, and K63 is mainly involved in endosomal trafficking and DNA repair, the function of the “non-canonical” chains is less well defined [78, 79]. Parkin has been reported to be able to promote monoubiquitination as well as multiple chain types such as K6, K11, K27, K48 and K63 ubiquitin chains in diverse contexts through interaction with different E2s [64, 80‐83]. Moreover, Parkin has been shown to autoubiquitinate itself via K6 ubiquitin chains and the deubiquitinase USP8 has been suggested to remove these K6 chains to regulate Parkin stability and its recruitment to mitochondria [84]. Furthermore, USP15, USP30 and USP35 have been shown to deubiquitinate Parkin substrates and/or regulate mitophagy following acute mitochondrial depolarisation in cells [32, 38, 85, 86]. While USP15 has been reported to have a preference for K48 and K63 [85], USP30 has been reported to favour K6 chains disassembly [87].

In contrast to our earlier work using this same bioUb pulldown and Western blotting approach [39, 40, 42, 43], most of the specific proteins we have validated here by western blotting were found to be monoubiquitinated. This agrees with previous observations demonstrating that Parkin is able to generate monoubiquitinated conjugates [81, 82]. Our ubiquitin chain analysis showed that Parkin preferentially promotes K6 chain formation in contrast to some literature reports indicating that Parkin promotes a more diverse set (K6, K11, K48 and K63) after mitochondrial depolarisation [83, 88]. However, in line with our observations, recent ubiquitin chain analysis showed that K6 is preferentially formed in whole lysates from Parkin over-expressing HEK and SH-SY5Y cells after mitochondrial depolarisation [87].

Despite the fact that the exact molecular mechanism leading to PD is unclear, the set of genes linked to PD can be divided in two main pathways. Parkin, DJ-1, PINK1 and FBXO7 are linked to mitochondrial homeostasis [65, 66, 89, 90], while α-synuclein (SNCA), LRRK2, VPS35, EIF4G1 and ATP13A2 are involved in endosomal-lysosomal trafficking [7‐9]. LRRK2, RAB7L1 and VPS35 have been shown to act in a common pathway [91, 92] and VPS35 has recently been shown to genetically interact with EIF4G1 in a pathway upstream of α-synuclein [93]. Furthermore, VPS35 impairment leads to lysosomal dysfunction and α-synuclein pathology [94, 95].

The finding that Parkin ubiquitinates VPS35 is of considerable interest, as it could represent a connection between mitochondrial homeostasis and endo-lysosomal pathways, unifying the PD-associated genes in a common network [96]. In fact, recent studies in Drosophila reported that Vps35 genetically interacts with parkin [59]. Our identification of VPS35 as a Parkin substrate strengthens this association. Moreover, VPS35 has been found to be involved in mitochondrial homeostasis [60, 61], therefore it is tempting to consider that Parkin-dependent ubiquitination of VPS35 may regulate mitophagy. However, our results show that Parkin-dependent mitophagy is not affected by VPS35 knockdown, suggesting that VPS35 is not an essential regulator of Parkin-dependent mitophagy in a mammalian cell system subjected to acute depolarisation. Nor do we see any changes in VPS35 levels following Parkin activation (Fig. 8c) or with control RPE-1 cells which do not express Parkin (not shown). Alternatively, Parkin - dependent-ubiquitination of VPS35 could be involved in the regulation of mitochondrial-derived vesicles (MDV), which have been reported to traffic defective mitochondrial proteins to peroxisomes and lysosomes in a more specific manner prior to the whole organelle impairment [62, 63, 97]. A further possibility is that Parkin ubiquitinates and regulates endo-lysosomal pathway components through the ubiquitination of several proteins, including VPS35 and others described here. During the preparation of this manuscript Parkin has been reported to modulate endo-lysosomal trafficking through Rab7 ubiquitination, but also to influence the distribution of VPS35 and the levels of the retromer cargo protein ci-M6PR [98]. Other recent papers have also linked Parkin to endo-lysosomal trafficking [99‐102]. The involvement of endocytic trafficking in PD is gaining momentum with the finding that two main PD-associated genes, PINK1 and LRRK2, converge and lead to the phosphorylation of common Rab proteins, small GTPases involved in membrane trafficking [103‐105].

Interestingly, both VPS35 and Parkin mutations are also associated with Alzheimer’s Disease (AD) [106‐108]. VPS35 interacts with the amyloid precursor protein (APP) sorting receptor, SorL1, which is also an AD causing gene, to regulate APP processing and Aβ production [109‐113]. Parkin has recently been identified as an AD risk gene [108] and decreased Parkin levels have been found in AD patient cerebrospinal fluids [114]. Remarkably, Parkin over-expression has been shown to ameliorate AD symptoms by increasing the clearance of amyloid loads [115, 116], which may suggest that defects in a common endo-lysosomal pathway are shared between PD and AD. In this scenario, Parkin may have a role in endosomal trafficking control through ubiquitination of VPS35, with a potential impact on both PD and AD. Certainly, further work is required in order to explore this hypothesis.