According to the Flybase Drosophila genome database (Gramates et al., 2017), the dUsp36 gene encodes multiple putative transcripts but identification of full-length cDNAs supports the existence of only three of them (Figure 1A). The proteins expressed from these transcripts are identical except for their specific N-terminal domain (Figure 2A). The common part contains the Ubiquitin Specific Protease (USP) catalytic domain followed by a disordered domain (Gramates et al., 2017). When transfected into Drosophila S2 cells, V5-tagged isoforms display different subcellular localizations (Figures 1B–D): while the dUSP36-B isoform accumulates in the cytoplasm and at the nuclear membrane as shown by colocalization with Lamin (Figure 1B), the dUSP36-C and -D isoforms are localized in the nucleus (Figures 1C,D). However, under these overexpression conditions and in contrast to human USP36 (Endo et al., 2009; Sun et al., 2015a), their localization is not restricted to the nucleolus, highlighted by the nucleolar marker Fibrillarin, but expands to the whole nucleoplasm. To gain insight into the mechanisms controlling the subcellular localization of the dUSP36 isoforms, truncated constructs were produced (Figure 2A) and transfected into S2 cells: the 110–1038 construct which corresponds to the common part of the isoforms is localized in the nucleus (Figure 2C), as is the 477-1038 C-terminal construct (Figure 2E). On the opposite, the 110–670 construct, which contains the USP catalytic domain, is not only present in the nucleus but also in the cytoplasm (Figure 2D). These data place the sequence(s) responsible for the nuclear localization of the dUSP36-C and -D isoforms in the 670–1038 C-terminal domain, which is consistent with the identification of two putative Nuclear Localization Sequences (NLS, represented by black bars in Figure 2A) by NLS prediction programs (PSORT, NLS Mapper and SeqNLS). As these sequences are also present in the dUSP36-B isoform that is not localized in the nucleus, we hypothesized that the N-terminal forty amino-acid long domain specific of this isoform (Figure 2B) acts as a Nuclear Export Sequence (NES). Nuclear export of proteins occurs either through the classical nuclear export pathway mediated by the evolutionarily conserved CRM1 protein or through non-classical export pathways mediated by other importin β members. CRM1-dependent NESs are leucine-rich and typically contain large hydrophobic conserved residues separated by a variable number of amino acids (Fasken et al., 2000; Henderson and Eleftheriou, 2000; Xu et al., 2012). The primary sequence of the dUSP36-B specific domain is enriched in leucine residues and contains a CRM1-dependent NES consensus sequence (Figure 2B). This domain was fused to the transcription factor Relish (REL). Compared to the REL protein which is localized in the cell nucleus (Figure 2F), the fusion protein is excluded from the nucleus and accumulates in the cytoplasm (Figure 2G), demonstrating that the N-terminal domain specific of the dUSP36-B isoform has a NES activity.

These data show that the dUsp36 gene encodes two nuclear isoforms (dUSP36-C and -D) and one cytoplasmic isoform (dUSP36-B) which is exported from the nucleus due to the presence of a NES in its specific N-terminal domain.

To specifically inactivate each one of the dUSP36 isoforms, we performed three CRISPR-Cas9 mutagenesis targeting each one of the isoform-specific exons downstream of their respective ATGs (Yu et al., 2013; Port et al., 2014) (Supplementary Figure S1) and retained one nucleotide deletions inducing frameshift mutations for each one of the three isoforms (Figures 1A,E). In all subsequent experiments, each isoform-specific mutation has been analyzed in trans over the null dUsp36^Δ43 allele. This allele was previously generated by P element excision (Thevenon et al., 2009) and corresponds to a deletion of most of the dUsp36 gene, thus affecting all three isoforms (Figure 1A).

Western blot analysis using a specific anti-dUSP36 antibody (Buszczak et al., 2009) revealed that the dUsp36 gene produces the three isoforms at their expected size (Figure 1F). In dUsp36-B mutants, the dUSP36-B isoform is missing whereas the other two isoforms are normally expressed while in dUsp36-C and -D mutants, the dUSP36-C and -D isoforms are specifically absent, respectively (Figure 1F). These results show that each frameshift mutation efficiently and specifically inactivates the expected dUSP36 isoform.

These isoform-specific mutations were then used to study the subcellular localization of the dUSP36 isoforms in vivo. In fat body cells of wild-type feeding third-instar larvae, the dUSP36 protein, visualized with a specific anti-dUSP36 antibody recognizing the three isoforms (Buszczak et al., 2009), is mainly nucleolar as evidenced by colocalization with the nucleolar marker Fibrillarin (Figure 3A). In dUsp36-B and -C mutants, no clear modification of dUSP36 expression is observed (Figures 3C,D) whereas in dUsp36-D mutants, the nucleolar expression of dUSP36 disappears and a faint residual expression is observed throughout the cell which may correspond to dUSP36-B or -C expression (Figure 3E).

Later in development, in wandering third-instar larvae, dUSP36 expression decreases in the nucleolus and increases both in the cytoplasm and at the nuclear membrane, as shown by colocalization with Lamin (Figure 3B). In dUsp36-B mutants, the cytoplasmic and perinuclear expression of dUSP36 is lost whereas the protein is still present in the nucleolus (Figure 3F). In contrast, in dUsp36-D mutants, only the nucleolar expression is affected whereas the cytoplasmic and perinuclear accumulation of dUSP36 is still observed (Figure 3H). Finally, as observed in younger larval fat body cells, the mutation of the dUSP36-C isoform does not significantly affect dUSP36 expression (Figure 3G).

Altogether, these data show that the dUSP36-B isoform is present in the cytoplasm and at the nuclear membrane whereas the dUSP36-D isoform is nucleolar, as observed for the human USP36 protein (Endo et al., 2009; Sun et al., 2015a). The dUSP36-C isoform is either not expressed in the larval fat body cells or at very low level that could contribute to the residual nuclear staining observed in dUsp36-D mutant cells (Figures 3E,H).

Analysis of the developmental effects of the isoform-specific mutations revealed that inactivation of the dUSP36-B and -C isoforms does not affect larval size (Figures 4C,D,K) whereas the specific mutation of the dUSP36-D isoform results in smaller larvae (Figures 4E,K). This growth phenotype is correlated with a strong reduction of the size of the larval fat body cells (Figures 4J,L). However, this growth defect is milder than the one observed in dUsp36 null mutants (Figures 4B,G,K,L).

While dUsp36 null mutants die during larval stages (Taillebourg et al., 2012), isoform-specific mutants are all viable (Figure 5). dUsp36-B and -C mutants do not display any growth defects or developmental delay (Figures 5B,C,M–O). This is not the case of the dUsp36-D mutants, which are smaller than control flies (Figures 5D,M) and have smaller wings (Figures 5H,N). In addition, they have shorter and thinner scutellar (arrows) and dorsocentral (arrowheads) bristles (Figure 5L) and display delayed development (Figure 5O).

Altogether, these results show that the dUSP36-D isoform is the main isoform involved in cell and organismal growth. The weaker phenotype of dUsp36-D mutants compared to null dUsp36 mutants suggests that the residual expression of dUSP36-B and/or -C isoforms may marginally contribute to cell growth, at least in the absence of dUSP36-D.

We next investigated the molecular interactions between the dUSP36-D isoform, dMYC and the E3 ligase AGO for numerous reasons. First, dUsp36-D and dUsp36 null mutants are phenotypically very similar to hypomorphic and null dMyc mutants respectively (Johnston et al., 1999; Pierce et al., 2004; Gallant, 2013; Grifoni and Bellosta, 2015). Moreover, human USP36 has been shown to regulate c-MYC stability in the nucleolus by antagonizing the activity of the E3 ligase Fbw7γ (Sun et al., 2015a). Lastly, the Drosophila Fbw7γ ortholog AGO is known to regulate dMYC stability (Moberg et al., 2004). Co-immunoprecipitation experiments with dUSP36-D show that it interacts with dMYC (Figure 6, panel IP V5 IB dMYC, lanes 6 and 8) and AGO (Figure 6, panel IP V5 IB HA, lanes 7 and 8). As shown previously (Moberg et al., 2004), we also observe an interaction between dMYC and AGO (Figure 6, panel IP dMYC IB HA, lanes 4 and 8). These data demonstrate that dUSP36-D, dMYC and AGO interact with each other. However, they do not tell whether these interactions take place as three different heterodimers or if these proteins are part of the same complex. Interestingly, AGO overexpression strengthens the interaction between dMYC and dUSP36-D (Figure 6, panel IP dMYC IB V5, compare lane 6 to lane 8). Moreover, immunoprecipitation of the endogenous dMYC protein (Figure 6, panel IP dMYC IB dMYC, lanes 1, 3, 5, and 7) allows the co-precipitation of dUSP36-D only when AGO is overexpressed (Figure 6, panel IP dMYC IB V5, compare lane 7 to lane 5). Taken together these data strongly argue that dUSP36-D, dMYC and AGO are part of the same macromolecular complex.

We then asked whether dMYC quantity and ubiquitination levels are regulated by dUSP36-D and AGO. To this end, the dMYC protein was expressed alone (CTL) or in combination with the wild-type dUSP36-D protein (dUSP36-D^WT), a mutated version of dUSP36-D devoid of catalytic activity (dUSP36-D^mut) or the E3 ligase AGO (Figure 7A). As already described (Moberg et al., 2004), when AGO is expressed dMYC quantity is decreased (Figures 7A,A’) and its ubiquitination is increased (Figures 7A,A”). On the opposite, when the wild-type dUSP36-D protein is expressed, dMYC quantity is strongly increased (Figures 7A,A’) which is correlated with a sharp drop of its ubiquitination levels (normalized according to the quantity of immunoprecipitated dMYC) (Figures 7A,A”). Expression of the mutated dUSP36-D protein has no significant effect on dMYC quantity nor on its ubiquitination levels (Figures 7A–A”) indicating that dUSP36-D acts on dMYC through its catalytic activity.

Loss of function experiments were also performed using specific dsRNAs targeting the dUSP36-D isoform in cells overexpressing dMYC (Figure 7B). Silencing of dUsp36-D significantly decreases the quantity of overexpressed dMYC (Figures 7B,B’) and increases its ubiquitination level (Figures 7B,B”). Taken together, these results show that dUSP36-D deubiquitinates and stabilizes dMYC.

The interaction between dUSP36-D and AGO (Figure 6) prompted us to investigate whether AGO is also a substrate of dUSP36-D. To test this hypothesis, a tagged version of AGO was expressed in S2 cells alone (CTL) or in combination with either the wild-type dUSP36-D protein or its mutated version (Figure 7C). The wild-type dUSP36-D protein, but not the mutated form, significantly increases the quantity of AGO protein (Figures 7C,C’) and decreases its ubiquitination (Figures 7C,C”), suggesting that AGO is indeed a substrate of dUSP36-D catalytic activity. Loss of function experiments strengthen this conclusion since silencing of dUsp36-D using an isoform-specific dsRNA diminishes the quantity of AGO protein (Figures 7D,D’) and increases its ubiquitination (Figures 7D,D”). These data thus show that dUSP36-D deubiquitinates and stabilizes AGO.

S2 cells overexpressing a V5-tagged version of dUSP36-D and treated with the proteasome inhibitor MG132 show an accumulation of dUSP36-D ubiquitinated species (Supplementary Figure S2) suggesting that dUSP36-D is subjected to ubiquitination prior to proteasomal degradation. As the two proteins interact (Figure 6), we investigated the role of the E3 ligase AGO in dUSP36-D ubiquitination. We observed indeed that dUSP36-D levels are decreased by AGO overexpression (Figures 8A,A’) whereas its ubiquitination is concomitantly increased (Figures 8A,A”) indicating that dUSP36-D is ubiquitinated by AGO, which triggers its degradation.

During the course of our experiments, we noticed that the catalytic inactive form of dUSP36-D is systematically present in fewer quantities than the wild-type protein in cell lysates (Figures 7A,C). Quantification of the amount and ubiquitination level of wild-type and mutant dUSP36 proteins confirmed that the catalytically inactive form of dUSP36-D is both less abundant (Figures 8B,B’) and more ubiquitinated than the wild-type protein (Figures 8B,B”). These results suggest that the wild-type dUSP36-D protein is able to regulate its own ubiquitination level.

To test this hypothesis, the V5-tagged catalytic mutant form of dUSP36-D was expressed alone (CTL) or in combination with Flag-tagged wild-type or mutant dUSP36-D proteins (Figure 8C). We observed that the V5-tagged dUSP36-D protein immunoprecipitates the Flag-tagged dUSP36-D proteins (Figure 8C, panel IP V5 IB Flag) indicating that dUSP36-D can dimerize or that multiple copies of dUSP36-D are part of the same complex. Moreover, expression of dUSP36-D^WT-Flag increases the quantity of dUSP36-D^mut-V5 (Figures 8C,C’) whereas it decreases its ubiquitination level (Figures 8C,C”) arguing that dUSP36-D is able to deubiquitinate itself.

Taken together, our results show that dUSP36-D is ubiquitinated by the E3 ligase AGO and is able to promote its own deubiquitination (auto-deubiquitination).

Our previous data indicate that dUSP36-D, AGO and dMYC are part of the same complex in which AGO ubiquitinates and destabilizes dMYC and dUSP36-D, and dUSP36-D deubiquitinates and stabilizes dMYC, AGO and itself. dUSP36-D and AGO are thus expected to have antagonistic functions on dMYC-induced cell growth. The in vivo relevance of these results was investigated by looking at genetic interactions between dUsp36, dMyc, and Ago.

To this end, various transgenes were expressed in a few cells of the larval fat body using the Flpout system (Pignoni and Zipursky, 1997) to generate a chimeric tissue and compare the size of the transgene expressing cells to that of wild-type cells in the same context (Figure 9). The transgene expressing cells were easily detected because of the co-expression of an H2B-RFP construct (Langevin et al., 2005). When a control dsRNA transgene targeting the Luciferase gene (Luc-IR) is expressed, the ratio between the sizes of the transgene-expressing (red) and the wild-type cells is equal to one (Figure 9I), indicating that cell growth is not affected. This Luc-IR transgene was then used to balance the number of transgenes in subsequent experiments. Expressing dsRNA transgenes targeting either all dUSP36 isoforms (Figure 9A) or only the dUSP36-D isoform (Figure 9B) significantly reduces cell size (Figure 9I). As observed in previous experiments (Figure 4), inactivation of all dUSP36 isoforms displays the strongest effect (Figure 9I). In contrast, as previously shown (Parisi et al., 2013), cells overexpressing the dMYC protein show a strong increase in size (Figures 9C,I). Inactivating dUsp36 in this context significantly suppresses the effect of dMyc overexpression (Figures 9D,I). When dMyc is silenced using a partially efficient dsRNA transgene, cell size is moderately but significantly reduced (Figures 9E,I). Co-inactivating the dUSP36-D isoform enhances this effect (Figures 9F,I). Finally, inactivating ago significantly increases cell size (Figures 9G,I) and co-inactivation of the dUSP36-D isoform counteracts this effect (Figures 9H,I).

A second set of genetic interactions was analyzed at the tissue level (Figure 10) using the MS1096 wing-specific GAL4 driver (Brand and Perrimon, 1993; Capdevila and Guerrero, 1994). We observed that silencing of dUsp36 expression reduces wing size (Figures 10B,K). Two different dsRNA transgenes targeting dMyc were also tested for their ability to affect wing size: the first transgene (attP40 dMyc-IR, Figure 10C) has no visible effect (Figure 10K) whereas the second (attP2 dMyc-IR, Figure 10E) has a moderate but significant effect on wing area (Figure 10K). Although producing no phenotype when expressed alone, the attP40 dMyc-IR transgene drastically enhances the wing phenotype induced by the dUsp36 inactivating transgene (Figures 10D,K), indicating a synergistic relationship between dMyc and dUsp36. A similar phenotypic enhancement is observed with the attP2 dMyc-IR transgene (Figures 10F,K). Lastly, two dsRNA transgenes targeting ago were expressed in the wing imaginal disk and result in significantly increased wing size (Figures 10G,I,K). Co-silencing dUsp36 and ago rescues this wing phenotype (Figures 10H,J,K).

These in vivo results strongly support the notion that dMyc and dUsp36 act in the same pathway controlling cell growth whereas dUsp36 and ago have antagonistic functions. They are entirely consistent with our previous biochemical experiments. Altogether, our data show that dUSP36-D stabilizes dMYC and promotes cell growth while AGO destabilizes dMYC and inhibits cell growth.

The cDNAs for the dUSP36-B (LD40339), -C (AT24152) and -D (AT31021) isoforms were obtained from the Drosophila Genomics Resource Center and cloned by PCR into the pAc5.1-V5His plasmid (Invitrogen) to express C-terminally V5-tagged proteins. The dUSP36-D catalytic dead mutant was produced by mutating C181 to S and H439 to N using the QuickChange XL Site-Directed Mutagenesis Kit (Stratagene). The plasmids expressing the tagged AGO (FMO08124) and dMYC (FMO12803) proteins have also been obtained from the Drosophila Genomics Resource Center.

Flies were reared at 25°C except otherwise stated on standard cornmeal–yeast medium.

The dUsp36-D-IR plasmid was generated by cloning a hairpin sequence corresponding to the specific 5′ exon of this isoform (sequence available on request) into the pWIZ transgenesis vector (Lee and Carthew, 2003) and transgenic Drosophila strains were established (BestGene Inc).

dUsp36^{Δ 43}, a null dUsp36 allele, and the dUsp36-IR transgenic line have already been described (Thevenon et al., 2009). The UAS-Luc-IR (BL#35788), UAS-dMyc (BL#9674), UAS-dMyc-IR (BL#43962) and UAS-ago-IR34 (BL#34802) strains (Ni et al., 2009, 2011) were obtained from the Bloomington Drosophila Stock Center. The UAS-H2B-RFP transgenic line (Langevin et al., 2005) was obtained from Dr. Y. Bellaiche.

For the Flpout method, a FRT-flanked cassette blocking expression of the GAL4 gene is excised upon heat-shock induced expression of the FLP recombinase (Pignoni and Zipursky, 1997). This mitotic recombination event leads to the expression of the GAL4 gene and is transmitted across mitosis, generating clones of cells in which GAL4 expression is activated. Spontaneous activation of the GAL4 transcription factor has been reported and allows for the induction of GAL4 expressing cells without heat shock (Hennig et al., 2006).

For CRISPR-Cas9 mutagenesis, the gRNA sequences designed to target each one of the dUSP36 isoforms were selected using the “CRISPR Optimal Target Finder” website¹. These sequences were cloned into the pCFD3 plasmid (Port et al., 2014). The pCDF3-B, -C or -D plasmids were injected into the nos-Cas9 CFD2 (Port et al., 2014) recipient strain (BestGene Inc.) and the resulting male founders were crossed with yw, TM3/TM6 females. Their progenies were then screened either by T7 endonuclease I assay (for B and C) or by phenotype (for D). The presence of the mutations was then confirmed by sequencing. For dUsp36-B, the progeny of 30 founder males was screened, six mutations were identified. For dUsp36-C, the progeny of 25 founder males was screened, 11 mutations were identified. For dUsp36-D, the progeny of 51 founder males was screened, 27 mutations were identified. The sequences of the recovered mutations are given in Supplementary Figure S2. The sequences of the oligonucleotides used for screening and sequencing are available on request.

Drosophila S2 cells were maintained at 25°C in Schneider’s Drosophila medium supplemented with 10% heat-inactivated serum (FCS, Invitrogen). DNA transfections were performed 48 h prior to cell lysis using Transfectin (Bio-Rad) according to the manufacturer’s instructions. Gene inactivation was achieved by incubating cells with double strand RNA (dsRNA) for 48 h. DNA templates for dsRNA synthesis were generated by PCR (MEGAscript RNAi kit, Ambion) using the primers designed from Heidelberg Fly Array RNAi libraries².

Cell lysis was performed in RIPA buffer (50 mM Tris-HCl, 150 mM NaCl, 1 mM EDTA, 1% IGEPAL, 0.5% sodium deoxycholate) supplemented with a protease inhibitor cocktail (Sigma) and with the pan-deubiquitinases inhibitor PR-619 (Sigma-Aldrich). For co-immunoprecipitation assays, lysates were precleared with Protein A or G-Sepharose beads (Sigma-Aldrich) for 4 h at 4°C with rotation. Immunoprecipitations were performed in RIPA buffer. Immune complexes were precipitated with protein A or G-Sepharose beads with the indicated antibody overnight at 4°C, the beads were washed 4 times with RIPA buffer and bound proteins were eluted using Laemmli Sample Buffer (Bio-Rad) supplemented with 10% of β-mercaptoethanol and boiled. Whole cell lysates were diluted with Laemmli Sample Buffer 4x (Bio-Rad) supplemented with 10% of β-mercaptoethanol, boiled at 95°C for 5 min and directly used for immunoblotting.

Protein lysates and eluates were separated on SDS-PAGE gels (TGX Stain Free from Bio-Rad). The total amount of protein was detected directly in the gels using the Image Lab Stain Free Gel protocol. Proteins were transferred to PVDF membranes (Bio-Rad). Membranes were saturated for 1h in TBS/0.1% Tween-20/5% BSA or skimmed milk, incubated with the appropriate primary antibody for 1 to 3 h at RT or overnight at 4°C in TBS/0.1% Tween-20/1% BSA or skimmed milk before three washes in TBS/0.1% Tween-20 (TBST) and then incubated an additional hour at RT with corresponding secondary antibodies coupled to horse radish peroxidase (HRP). After three rinses with TBST, the membranes were revealed with the Luminata Forte Western HRP substrate (Millipore) using the Chemidoc imaging system (Bio-Rad). Quantifications were performed using the Image Lab software (Bio-Rad). The extent of the smear of ubiquitinated proteins taken into account for quantification is indicated by the double arrows on Figures 6, 7.

The anti-dUSP36 antibody, a kind gift from Dr. M. Buszczak, has been already described (Buszczak et al., 2009) and was used at a 1/2500 dilution. The following antibodies were used following the manufacturer’s instructions: monoclonal anti-V5 (Invitrogen), monoclonal anti-Flag clone M2 (Sigma), rat anti-HA High Affinity (Sigma), rabbit anti-dMYC (Santa Cruz Biotechnology, sc28207), monoclonal anti-ubiquitinated conjugates FK2 (Enzo). The secondary antibodies HRP-coupled goat anti-mouse, -rabbit and -rat are from Sigma.

For immunocytochemistry, S2 cells were fixed for 10 min in 4% paraformaldehyde, rinsed twice in PBS, blocked for 1 h in PBS, 0.1% Triton X-100, 5% normal goat serum and incubated for 1 h with a rabbit anti-V5 antibody (1/500, Invitrogen) and either with monoclonal anti-Lamin Dm0 ADL84.12 (1/200, Developmental Studies Hybridoma Bank) or anti-Fibrillarin 38F3 (1/500, abcam) antibodies. Lateral lobes of third instar larval fat bodies were dissected in PBS, fixed for 30 min in 4% paraformaldehyde and rinsed twice in PBS. The samples were then blocked for 1 h in PBS, 0.1% Triton X-100, 5% normal goat serum and incubated overnight at 4°C with the anti-dUSP36 (1/200) and either with monoclonal anti-Lamin Dm0 ADL84.12 (1/200, Developmental Studies Hybridoma Bank) or anti-Fibrillarin 38F3 (1/200, abcam) antibodies. Secondary antibodies were coupled to Alexa594 or Alexa488 (1/500, Invitrogen).

After mounting in DAPI-containing Vectashield (Vector Laboratories, H-1200), the samples were imaged with a 40x or 63x magnification (oil immersion) using a Leica TCS SP2 confocal microscope and the LCS software. All the pictures shown are representative of the whole tissue and of the observations made from different animals. Cell areas were automatically measured using the cell image analysis software CellProfiler (Carpenter et al., 2006). The analysis pipelines are available on request.

Third instar larvae, wings, thoraces and adults were imaged using a Keyence VHX-5000 numerical microscope. Larval size and wing area were blindly measured manually using the Fiji/ImageJ software (National Institute of Health).

All statistical analyses were performed in Prism 6 (GraphPad).