The centrosomal deubiquitylase USP21 regulates Gli1 transcriptional activity and stability

Hedgehog (Hh) signaling plays an important role during development and has also been implicated in diverse malignancies, including basal cell carcinomas, medullo- and glioblastoma, as well as pancreatic, colon and breast carcinomas (Amakye et al., 2013; Takebe et al., 2015). Inhibitors of this pathway are under assessment in the clinic as potential anti-cancer therapeutics (Gonnissen et al., 2015; Takebe et al., 2015). The general architecture of the pathway was initially discovered in Drosophila (Nusslein-Volhard and Wieschaus, 1980), but much of the cascade is highly conserved from flies to vertebrates (Briscoe and Therond, 2013; Ingham et al., 2011). In contrast to many other signaling pathways, Hh signaling is constitutively repressed by the interplay of two multi-spanning transmembrane proteins, Patched (PTC in Drosophila, and PTCH1 and PTCH2 in mammals) and Smoothened (SMO). In the absence of a signal, Patched proteins inhibit SMO, and the key transcription factors, Ci (Cubitus interruptus) in Drosophila, and Gli2 and Gli3 in mammals, are converted to transcriptional repressors. Gli2 and Gli3 are phosphorylated by a series of kinases [protein kinase A (PKA), casein kinase 1 (CK1) and glycogen synthase kinase 3 β (GSK3β)], and in particular, Gli3 undergoes processing (limited proteolysis) in a phosphorylation-, ubiquitylation- and proteasome-dependent manner to a transcriptionally repressive form. Activation of the pathway is triggered by binding of Hh ligands to PTCH1 or PTCH2, which triggers their endocytosis and leads to de-repression of SMO, favouring activation of Ci/Gli proteins. These translocate into the nucleus and activate expression of Hh-responsive genes (Hui and Angers, 2011). In mammalian cells, activation of the cascade is localised at the primary cilium, a specialised microtubule-based surface organelle. Release of SMO inhibition leads to SMO, Gli2 and Gli3 translocation to the cilium, and subsequent translocation of the unprocessed transcriptionally active Gli proteins to the nucleus. One of the transcriptional targets is Gli1, which exclusively functions as a transcriptional activator in the pathway. It is the balance of active and repressive Gli proteins that determines signal strength and outcome.

Post-translational modifications play key roles in this cascade (Chen and Jiang, 2013; Gulino et al., 2012). Reversible ubiquitylation can either promote protein degradation, or alter subcellular localisation or activity of a protein. Multiple E3 ligases are thought to play crucial roles in Hh signaling – Cullin1 with the substrate adapter Slimb (in Drosophila) and βTRCP (in mammals), and Cullin3 with the substrate adapter Roadkill/HIB (in Drosophila) and SPOP (in mammals) and Itch have all been shown to ubiquitylate and destabilise Gli proteins (Jiang, 2006). Approximately 90 deubiquitylases (DUBs) are encoded in the human genome (Clague et al., 2013; Komander et al., 2009). This family of enzymes has been implicated in the regulation of many canonical signal transduction cascades – e.g. NFκB, Wnt, TGFβ and the Ras–MAPK pathway (Clague et al., 2012a; Hayes et al., 2012; Keusekotten et al., 2013). USP8 has been associated with this cascade in Drosophila, where it promotes recycling of SMO (Clague et al., 2012b; Li et al., 2012; Xia et al., 2012). However, the role of DUBs in the mammalian Hh-signaling pathway has only recently been explored (Zhou et al., 2015).

We have previously characterised USP21 as the only centrosome- and microtubule-associated DUB in a comprehensive localisation screen (Urbe et al., 2012). USP21 also associates with basal bodies in ciliated cells, and its depletion inhibits the formation of primary cilia, which play a crucial role in the initiation of Hh signaling in vertebrates. Here, we identify a new binding partner of USP21, the BTB-domain-containing cullin E3-ligase adapter protein KCTD6 (KCASH3), which has previously been implicated in Hh signaling (De Smaele et al., 2011). Importantly, we go on to show that either depletion or overexpression of USP21 suppresses Gli1-dependent transcription in human cells. Our results reveal that USP21 stabilises Gli1 at the centrosome and promotes its phosphorylation by PKA, thus contributing to the intricate compartmentalisation of the Hh-signaling pathway.

We have used a yeast two-hybrid (Y2H) approach to search for directly interacting proteins of USP21 that could act as substrates or substrate adaptors in the context of microtubule- and centrosome-dependent processes. Out of a panel of USP21 fragments, only those encompassing amino acids 1–174 and lacking residues 1–47 passed the suitability test by failing to auto-activate reporter genes (Materials and Methods; data not shown). We selected USP21[Δ1–47], which retains all microtubule-binding and centrosome-localisation determinants (Urbe et al., 2012), as a bait to screen a universal normalised human cDNA Y2H library (Fig. 1A). Seven out of ten isolated diploid colonies contained plasmids with in-frame annotated coding sequences, two of which encode ZNF350 (zinc-finger protein 350). The other proteins identified were potassium channel tetramerization domain containing 6 (KCTD6), WD-repeat-containing protein 47 (WDR47), spectrin-repeat-containing nuclear envelope 1 (SYNE1), membrane protein palmitoylated 1 55 kDa (MPP1) and Ankyrin-repeat-domain-containing protein 32 (ANKRD32) (Table 1, Fig. 1A).

Two of the six new USP21-interacting proteins have been shown to localise to microtubule-based structures. WDR47 colocalises with cytoplasmic microtubules through its interaction with microtubule-associated protein (MAP)8 (Wang et al., 2012a), whereas MPP1 has been localised to the basal body and the ciliary axoneme in murine photoreceptors (Gosens et al., 2007). We were particularly intrigued by KCTD6, a BTB-domain-containing protein that has previously been shown to modulate Hh signaling in association with the related protein KCTD11 (De Smaele et al., 2011). We validated this interaction by co-immunoprecipitating epitope-tagged proteins from HEK293T cells (Fig. 1B). KCTD6 interacts preferentially with catalytically inactive USP21-C221S (USP21CS), which in turn isolates higher-molecular-mass species of mono- and di-ubiquitylated KCTD6 (Fig. 1B,C). USP21CS itself presents as a doublet, a phenomenon commonly observed for catalytically inactive mutants of the USP family – e.g. USP4 and USP15 – which tend to accumulate as a mono- and polyubiquitylated species (Fig. 1D) (Hayes et al., 2012).

KCTD6 binds to CUL3 through its BTB domain, suggesting that it acts as a classic substrate adapter of a BTB–CUL3–Rbx1 ubiquitin E3 ligase (De Smaele et al., 2011; Lange et al., 2012; Smaldone et al., 2015). We asked whether USP21 interacts with a fully assembled KCTD6–CUL3 ligase complex. Interactions between DUBs and E3 ligases have been established in the literature, but few examples have been analysed in detail (Hayes et al., 2012; Keusekotten et al., 2013; Komander et al., 2009; Schulein-Volk et al., 2014; Sowa et al., 2009; Villeneuve et al., 2013). Both active and inactive USP21 co-immunoprecipitate with CUL3 and vice versa (Fig. 2B; Fig. S1B). Intriguingly, immunoprecipitation of inactive USP21CS enriches a higher-molecular-mass species of CUL3 that could in principle correspond to a neddylated form, which is required for cullin–RING-ligase activation (Bennett et al., 2010). The Nedd8-activating enzyme (NAE) inhibitor MLN4924 did not affect association of FLAG–KCTD6 with endogenous CUL3, nor the amount of USP21 that bound to KCTD6 (Fig. S1A). Importantly, the additional CUL3 band was insensitive to MLN4924, suggesting that USP21CS promotes the stabilisation of ubiquitylated forms of both KCTD6 and CUL3 (Fig. S1B).

The interaction between USP21 and CUL3 is most likely indirect and mediated through KCTD6 given that a functional BTB domain is both required and sufficient (Fig. 2A,C,D). Two closely related KCTD-family members have also been implicated in Hh signaling (Canettieri et al., 2010). The KCTD6-interacting protein KCTD11 (also called KCASH1 or REN), but not KCTD21 (also called KCASH2), bound weakly to wild-type USP21 and strongly to catalytically inactive USP21 (Fig. S1C). However, in contrast to KCTD6, KCTD11 did not appear to be modified in GFP–USP21CS co-immunoprecipitates.

The best-characterised BTB-domain CUL3 adaptors contain defined substrate interaction domains, such as Kelch, MATH, PHR or zinc fingers (Stogios et al., 2005). No such domains have been described for KCTD proteins, yet the region comprising residues 114–237, which is located directly after its BTB domain, is sufficient for interaction with USP21 (Fig. 3A). The combined results of Y2H and pull-down analyses narrow down the USP21 binding site to the 114–187 region of KCTD6 (Fig. 3E). We hypothesised that this region contains a cryptic protein-interacting domain.

Sensitive bioinformatics tools for detecting distant homologies (HHpred; Soding, 2005) and the Metaserver used to recognise protein folds (Bujnicki et al., 2001) failed to identify recognisable relatives of the KCTD6 C-terminal domain. Nevertheless, the region is rich in predicted secondary structure, suggestive of a well-folded domain. In favourable cases, ab initio modelling can accurately predict the overall fold of a queried protein through stitching together fragments of unrelated proteins (Rohl et al., 2004; Xu and Zhang, 2012). Using this approach, the most promising results were obtained for a region comprising residues 109–213, eliminating 24 residues from the C-terminus that some methods indicate might be intrinsically disordered. The favoured model from the QUARK ab initio modelling server bears a striking resemblance to the C-terminal Smr domain of human NEDD4-binding protein 2 (NEDD4BP2), with a Z-score of 7.1 for the match, well above the threshold of 2 generally taken as indicating significance [PDB code 2VKC (Diercks et al., 2008); Fig. 3B; Fig. S1D]. According to this model, the C-terminal section (residues 188–213) of this Smr-like (SmrL) domain, containing a helix and a short β-hairpin, is separate from the remainder of the structure, which thus might be able to fold independently. Indeed, separate ab initio modelling efforts using a shorter stretch of the peptide, comprising residues 109–187, produced results that included structures similar to those of the modelled SmrL domain (not shown). This might account for the observed association of USP21 with a C-terminally truncated KCTD6 (residues 114–187) in our Y2H screen.

Conversely the Δ1–184 fragment, encompassing the catalytic domain of USP21, is necessary and sufficient to bind to KCTD6 as well as to CUL3 (Fig. 3C,D). Taken together, our data indicate that CUL3, KCTD6 and USP21 can form a multimeric complex in which the catalytic domain of USP21 interacts with the C-terminal SmrL domain of KCTD6, whereas the BTB domain of the latter mediates interaction with CUL3 (Fig. 3E).

Based on previously described interactions between DUBs and E3 ligases, we asked whether USP21 might either stabilise or modulate the activity of the KCTD6–CUL3 E3-ligase complex. Neither small interfering (si)RNA-mediated depletion nor overexpression of USP21 altered overall CUL3 protein levels or the relative proportions of active (neddylated) and inactive (non-neddylated) forms (Fig. S2A,B). Unfortunately, none of the commercially available antibodies against KCTD6 that we tested recognised the endogenous protein. FLAG–KCTD6 protein levels were not significantly affected by overexpression of USP21 (Fig. 1B). We conclude that despite the enrichment of ubiquitylated KCTD6 and CUL3 in USP21CS immunoprecipitates, the stability of both proteins is not regulated by USP21. The steady-state protein levels of the sole known substrate of the previously described substrate of the previously described heterodimeric CUL3 E3-ligase substrate adapter complex KCTD6–KCTD11, HDAC1, were also unaffected by depletion or overexpression of USP21 (Fig. S2C,D).

KCTD6 has been proposed to act as a negative regulator of the Hh-signaling pathway (De Smaele et al., 2011). We have previously shown that USP21 depletion inhibits the formation of the primary cilium, which also plays a key role in Hh pathway activation in non-transformed cells. In order to assess a potential role of USP21 in Hh signaling, we first measured Gli1 mRNA levels as a transcriptional readout in NIH3T3 cells that had been stimulated with a Smoothened agonist (SAG) (Chen et al., 2002). Depletion of USP21 inhibited SAG-dependent accumulation of endogenous Gli1 mRNA levels (Fig. 4A).

The main transcriptional activator in the pathway is Gli1 itself, which operates through a positive auto-regulatory feedback loop (Hui and Angers, 2011). KCTD6, in concert with KCTD11, has previously been shown to target HDAC1 for ubiquitin-dependent degradation, thereby indirectly modulating the acetylation status and thus transcriptional activity of Gli1. With that in mind, we wanted to assess the ability of USP21 to affect Gli1 transcriptional activity in the absence of extrinsic pathway activation using a Gli1-dependent dual luciferase reporter gene assay. We opted for a heterologous co-expression system of hemagglutinin-tagged Gli1 (HA–Gli1) and GFP–USP21 in HEK293T cells, which permit high levels of transfection efficiency. Importantly, this approach introduces constitutively expressed Gli1 together with the Gli1-dependent reporter construct into the cells. This allows us to pinpoint the stage at which USP21 could regulate Hh signaling, downstream of SMO activation – i.e. downstream of the already established role in primary cilium formation. In this assay, siRNA-mediated depletion of USP21 decreased Gli1 transcriptional activity by ∼40% (Fig. 4B). Likewise, overexpression of GFP–USP21 inhibited Gli1 transcriptional activity to a similar degree, and this inhibition depended on its catalytic activity (Fig. 4C). Altogether, these data indicate that USP21 regulates the Hh-signaling pathway, and either directly or indirectly modulates Gli1 transcriptional activity.

We asked whether USP21 regulates the stability of Gli1 and turned to CRISPR-CAS9 technology to knockout USP21 in HEK293T cells, which, like many cancer cells, show autonomous expression of Gli1 (Fig. S3A) (Dahmane et al., 1997; Kinzler et al., 1987; Mazza et al., 2013; Nolan-Stevaux et al., 2009; Tariki et al., 2014). Transient transfection with plasmids encoding both CAS9–GFP and five distinct guide (g)RNAs, targeting the second exon encompassing the start codon of all annotated USP21 isoforms, resulted in a clear decrease of Gli1 protein levels by 25–50% (Fig. 5A). Note that in this configuration, residual USP21 could derive from non-transfected cells. Although these results suggest that USP21 stabilises Gli1, the observed decrease in endogenous Gli1 levels could, in principle, reflect transcriptional suppression (see also Fig. 4A).

We next co-expressed HA-tagged Gli1 together with GFP–USP21 and fragments thereof in HEK293T cells. We observed a clear increase in Gli1 protein levels in cells that co-expressed full-length USP21 (Fig. 5B; Fig. S3B). In order to assess a potential interaction between these two proteins, we immunoprecipitated HA–Gli1 and probed for associated GFP–USP21. Full-length GFP–USP21, but not the shortest N-terminal fragments (comprising residues 1–120 or 1–174), associated with Gli1 (Fig. 5C). Both the unstructured N-terminal region (residues 1–210), required for microtubule and centrosome association, and the C-terminal USP21 region lacking residues 1–184 and harbouring the catalytic domain, were capable of binding to Gli1 independently of each other. Likewise, FLAG-tagged Gli1 and Gli2 were able to co-precipitate Myc-tagged USP21 from HEK293T cells (Fig. S3C,D).

Having shown above that the catalytic domain of USP21 also interacts with the KCTD6–CUL3 E3-ligase complex, we wondered whether KCTD6 would be able to influence Gli1 stability in this setting. Co-expression of KCTD6 with USP21 counteracted its ability to stabilise Gli1 (Fig. 5D). This suggests that the direct interaction of KCTD6 with the USP21 catalytic domain disrupts its regulation of Gli1.

The increase of Gli1 protein levels upon USP21 overexpression appears to be at odds with the observed inhibition of Gli1 transcriptional activity in this setting (Fig. 4C). On close inspection, we observed a higher molecular weight form of Gli1 in USP21-overexpressing cells, indicative of a post-translational modification (Fig. 6A). The intensity of this band was substantially reduced in cells that co-expressed the catalytically inactive USP21CS, and the band was absent in cells that expressed truncated versions of USP21 (Fig. S3B). Many components of the Hh-signaling pathway are regulated by acetylation, phosphorylation, ubiquitylation and sumoylation (Chen and Jiang, 2013; Gulino et al., 2012). In particular, Gli transcriptional activity is regulated by several kinases, thus we assessed whether USP21 induces Gli1 phosphorylation. Incubation of immunoprecipitated HA–Gli1 with calf intestinal alkaline phosphatase (CIP) induced loss of the upper band that was only present in cells that overexpressed full-length USP21 (Fig. 6B).

Phosphorylation of Gli transcription factors can either activate or inhibit their transcriptional activity. In the absence of pathway activation, sequential phosphorylation of Gli3, and to a lesser extent of Gli2, by PKA, CKI and GSK3β induces ubiquitin-dependent limited proteolysis to generate a transcriptionally repressive form, GliR (Briscoe and Therond, 2013; Hui and Angers, 2011). Phosphorylation by PKA alone leads to inactivation of Gli2 and Gli3 (Niewiadomski et al., 2014; Tuson et al., 2011), and has been proposed to inhibit nuclear localisation of Gli1 (Sheng et al., 2006). Finally, phosphorylation of Gli1 by ribosomal protein S6 kinase B1 in response to mTOR activation triggers its release from the repressor SuFu, leading to its transcriptional activation (Wang et al., 2012b). Guided by our functional experiments that indicated an inhibitory role of USP21 expression on Gli1-dependent transcription (Fig. 4C), we focused on PKA-dependent phosphorylation as a potential effect of USP21 activity. Co-expression of PKI, a well-established PKA-inhibitor (Scott et al., 1985), abolished the USP21-induced accumulation of phosphorylated Gli1 (Fig. 6C). Strikingly, the decrease of Gli1 phosphorylation was concomitant with a decrease in its protein levels, suggesting that USP21-induced phosphorylation of Gli1 is coupled with its stabilisation (Fig. 6C). Similarly, a pool of Gli2 appeared to be stabilised in a PKI-sensitive manner through co-expression of USP21 (Figs S3C,D and S4A). Altogether, these data indicate that USP21 stabilises inactive Gli1 and promotes its phosphorylation by PKA.

How then could the deubiquitylase USP21 induce Gli1 phosphorylation by PKA? We have previously reported that USP21 is most highly enriched at the centrosomes and basal bodies of ciliated cells (Urbe et al., 2012). PKA has previously been shown to be enriched and active at these two locations (Nigg et al., 1985; Terrin et al., 2012; Tuson et al., 2011). U2OS cells allow ready discrimination of intracellular structures owing to their planar geometry and expanded cytoplasm. In these cells, endogenous PKA colocalised with GFP–USP21 at the centrosome (Fig. 6D). Neither wild-type or catalytically inactive USP21 affected PKA protein levels (Fig. S4B). Given that full stabilisation of phosphorylated Gli1 requires the USP21 N-terminal microtubule and centrosome-targeting sequence, we asked whether USP21 influences the localisation of Gli1. Expressed on its own, HA–Gli1 presents with a diffuse, cytoplasmic and nucleoplasmic localisation. However, co-expression of either wild-type or catalytically inactive GFP–USP21 strongly promoted its recruitment to the centrosome (Fig. 6E).

Our genome-wide Y2H screen yielded six new interacting proteins of USP21. The fact that WDR47 and MPP1 have been shown previously to associate with microtubules or the basal body and axoneme, respectively, makes these likely prospects for future studies (Gosens et al., 2007; Wang et al., 2012a). Here, we have focused exclusively on KCTD6, a CUL3-binding BTB protein that has previously been implicated in the Hh-signaling pathway. This pathway is of particular interest to USP21 function owing to its prominent association with centrosomes, basal bodies, microtubules and primary cilia. We have mapped this interaction to the catalytic domain of USP21 and a new cryptic protein-interaction domain, here called Smr-like (SmrL), that is unique to KCTD6. The Smr domain is generally associated with nuclease function (Fukui et al., 2007). However, the KCTD6 SmrL does not bear the positively charged surface of the NEDD4BP2 Smr structure that is characteristic of DNA- and RNA-binding proteins, nor a suitably positioned catalytic Glu residue (Zhang et al., 2014). Browsing the SCOP structure classification database (http://scop.mrc-lmb.cam.ac.uk/scop; Andreeva et al., 2008), at the level of protein folds that contain Smr domains, reveals a second theme of protein–protein interactions that is exemplified by bacterial sulphurtransferase TusA (Shi et al., 2010) and by the predicted dimeric form of NEDD4BP2 (according to the PISA resource; http://www.ebi.ac.uk/pdbe/pisa; Krissinel and Henrick, 2007). This is clearly consistent with the protein–protein interaction role determined here for the KCTD6 C-terminal domain and is supported by existing precedents of the same fold serving both DNA- and protein-binding functions, e.g. zinc fingers (Matthews and Sunde, 2002). The newly identified SmrL domain is restricted to KCTD6, although KCTD11 and KCTD21 could harbour their own distinct cryptic domains adjacent to the BTB domain. KCTD11, but not KCTD21, has been shown to heterodimerise with KCTD6, providing a possible explanation for its interaction with USP21 in pull-down experiments. USP21, unlike many other members of the USP family, does not contain insertions in its catalytic domain that could encode additional protein interaction motifs or domains. Thus, KCTD6 is the first protein, besides ubiquitin [and ISG15 (Ye et al., 2011)], to be identified as interacting with the catalytic core of USP21, and is a rare example of a core binding protein considering the USP family as a whole.

What then is the role of this interaction? E3-ligase–DUB interactions can protect the ligases from autoubiquitylation [e.g. MDM2 stabilisation by USP7 (Cummins and Vogelstein, 2004), BRAP/IMP stabilisation by USP15 (Hayes et al., 2012)]. Alternatively, the interactions might limit the levels of the respective DUBs or exert fine control over a common substrate. We have not found evidence that USP21 and KCTD6 have reciprocal effects on the stability of the other, nor does USP21 affect global CUL3 stability or neddylation status. However, the fact that co-expression of catalytically inactive USP21 results in the accumulation of minor mono- and di-ubiquitylated species of KCTD6 and CUL3 indicates the potential for USP21 to modulate their ubiquitylation status. Alternatively, USP21 could be acting directly to deubiquitylate the substrates of the cullin RING ligase, in an analogous fashion to USP28, which associates with the cullin-1-adapter FBW7 to deubiquitylate a range of substrates, including MYC (Schulein-Volk et al., 2014). In contrast to FBW7, we have very little knowledge of the substrates targeted by KCTD6. The closely related BTB proteins KCTD11 and KCTD21 undergo hetero- and homodimerisation to target HDAC1 for CUL3-dependent ubiquitylation and degradation (Canettieri et al., 2010). Degradation of HDAC1 promotes the accumulation of transcriptionally inactive acetylated Gli1. De Smaele and colleagues have reported that KCTD6, despite lacking a binding site for HDAC1, could dimerise with its paralogue KCTD11 (but not KCTD21) and contribute to HDAC1 ubiquitylation and degradation (De Smaele et al., 2011). Although we cannot exclude that USP21 directly stabilises KCTD6 substrates, we have not observed any global effects on HDAC1 protein levels in response to siRNA-mediated depletion or overexpression of USP21.

Our results suggest that endogenous USP21 positively regulates Hh signaling through two independent mechanisms: firstly, we have previously shown that USP21 depletion interferes with the formation of primary cilia, the specialised organelles that host the initiation of the Hh-signaling cascade in untransformed mammalian cells (Goetz and Anderson, 2010; Urbe et al., 2012). Secondly, we now show that depletion of USP21 not only inhibits SAG-induced transcription in ciliated NIH3T3 cells but also directly interferes with Gli1-dependent activation of a Gli-promoter-driven luciferase reporter in Gli1-transfected HEK293T cells. Although HEK293T cells have previously been shown to be capable of generating primary cilia (Gerdes et al., 2007; Lancaster et al., 2011), under the conditions of our assay (48 h in 0.5% fetal bovine serum) only 30% of cells do so. Whereas we cannot exclude that some aspects of exogenous Gli1-activation rely on primary cilium formation, taken together, our data showing (1) an interaction between USP21 and Gli1, (2) the regulation of PKA-dependent phosphorylation of Gli1 by USP21, and (3) the colocalisation of all three proteins at the centrosome in U2OS cells make a strong case for a regulatory role of USP21 that could be independent of cilium formation.

This positions USP21 downstream of SMO and Patched proteins, and highlights its potential as a new drug target to counteract Hh signaling. This could be of particular relevance in tumours that express aberrant levels of Gli1 or that are otherwise activated downstream of Patched proteins and are thus resistant to the SMO inhibitor Vismodegib (Amakye et al., 2013). In this context, it is interesting to note that USP21 has been identified in a large-scale screen as one of only few genes that has a significantly increased copy number count in breast tumour samples (Kan et al., 2010). The fact that both depletion and overexpression of USP21 interfere with Gli1-dependent transcription suggests that appropriate USP21 protein levels are key to ensuring optimal Hh-signaling outputs.

Intriguingly, overexpression of USP21 represses Gli1-dependent transcription, despite the fact that the total amount of Gli1 is clearly increased. Associated with this, we observed the accumulation of a phosphorylated pool of Gli1 in USP21-overexpressing cells. Although we have not identified the phosphorylation sites in this study, we provide evidence for the kinase involved in this modification. PKA has recently been shown to phosphorylate multiple serine residues that are conserved in all Gli proteins and to suppress Gli activity in the absence of a Hh signal (Niewiadomski et al., 2014). Our results suggest that USP21 overexpression facilitates these phosphorylation events by recruiting Gli1 either directly or indirectly to the centrosome, which harbours active PKA (Terrin et al., 2012). In ciliated cells, USP21 is localised at the centrosome-derived basal body, where PKA is thought to act on Gli proteins travelling to and from the cilium (Tuson et al., 2011). USP21 might thus promote coincidence of substrate and kinase at these locations. Our data suggest that expression levels of USP21 need to be tightly regulated because both its depletion and overexpression can hamper Hh signaling. However, the role of USP21 exceeds that of an adapter, given that both localisation determinants and catalytic activity are required for maximal stabilisation of phosphorylated Gli1 (Fig. 5B; Fig. S3A). It is conceivable that USP21 regulates both the phosphorylation status and the stability of Gli1 (and possibly Gli2) at the centrosome (Fig. 6F). Our finding provides a unique illustration of the close interplay of a DUB and a kinase at this location, and supports the concept of centrosomes functioning as intracellular signaling hubs or scaffolds (Arquint et al., 2014).

A recent report has identified USP7 as a DUB that is able to stabilise Gli-proteins, by opposing several Ubiquitin–E3-ligase complexes [Cullin1 (Slimb/βTRCP), Cullin3 (HIB/SPOP) and the HECT E3-ligase ITCH] that are involved in Gli protein processing or degradation in the absence and presence of Hh, respectively (Zhou et al., 2015). Our results suggest a more complex scenario in which the restricted localisation of USP21 facilitates the accumulation of an inactive pool of Gli1 at the centrosome. Extending this model to the basal body provides a means of orchestrating Gli protein activity upon entry into or exit from the primary cilium.

The coding sequence for USP21[Δ1-47] was subcloned from pCR4-TOPO-USP21[Δ1-47] into the pGBKT7 vector (Clontech). The resultant pGBKT7-USP21[Δ1-47] plasmid was sequence verified and transformed into Y2HGold yeast cells. Single colonies that failed to autonomously activate the reporter genes in the absence of prey proteins on tryptophan-lacking X-α-Gal plates, tryptophan-lacking X-α-Gal Aureobasidin-A plates and tryptophan-, adenine- and histidine-lacking X-α-Gal Aureobasidin-A plates were selected for the mating procedure. Briefly, Y2HGold pGBKT7-USP21[Δ1-47] cells were incubated with the human normalised cDNA Y187 prey library (Clontech) in yeast extract-peptone-adenine-dextrose (2×YPAD) medium for 23 h at 30°C. More than 2.4×10⁶ diploid clones were screened and selected for 5 days on synthetic dropout –Trp −Leu X-α-Gal Aureobasidin-A plates. Twenty clones were picked and streaked onto synthetic dropout –Trp –Leu –His −Ade X-α-Gal Aureobasidin-A plates for 7 days. Interacting proteins were identified by yeast colony PCR and subsequently by sequencing of the extracted plasmid.

U2OS, HEK293T and NIH3T3 cells were cultured in Dulbecco's modified Eagle medium (DMEM) with Glutamax (Gibco, Carlsbad, CA). Media were supplemented with 10% FBS, 1% nonessential-amino-acid solution and 1% penicillin-streptomycin. All cell lines were routinely tested for mycoplasma contaminations. Transfection of plasmids into HEK293T and U2OS cells was performed with Genejuice (Novagen). For the dual luciferase reporter assays (Promega), plasmids were transfected with Lipofectamine 2000 (Invitrogen) into HEK293T cells (DMEM, 0.5% FBS). Cells were fixed or lysed 24 h after transfection. For siRNA experiments, cells were treated with non-targeting (NT3) or target-specific siRNA oligonucleotides (Dharmacon On-Target Plus, Thermo Fisher Scientific; Table S1) at 40 nM final concentration, once or twice over a period of 96 h (HEK293T) or 120 h (U2OS). NIH3T3 cells were treated with a 12 nM final concentration, twice over a period of 120 h. All siRNA experiments were performed using RNAiMAX (Invitrogen). For Hh-pathway activation, the NIH3T3 cells were first serum-starved in DMEM with Glutamax for 24 h before addition of 100 nM SAG or DMSO for 4 h.

Cells were lysed, and mRNA was extracted using the RNAeasy mini kit (Qiagen). cDNA synthesis was performed using 1 µg RNA with RevertAid H-minus M-MuLV reverse transcriptase (Fermentas) using an oligo-dT primer (Promega). Quantitative real-time reverse-transcription PCR was performed in triplicate using iTaq Universal SYBR Green Supermix and a CFX Connect Real-Time PCR detection machine (Bio-Rad). Primer sequences are listed in Table S2.

HEK293T cells were transfected with a total of 2.4 µg of DNA per 1×10⁶ cells. At 24 h after transfection, the cells were washed in cold PBS and lysed in RIPA lysis buffer (10 mM Tris-HCl pH 7.5, 150 mM NaCl, 1% NP40, 0.1% SDS, 1% sodium deoxycholate) or in NP40-lysis buffer (0.5% NP40, 25 mM Tris-HCl pH7.5, 100 mM NaCl, 50 mM NaF) supplemented with 10 mM N-ethylmaleimide, mammalian protease inhibitor (Sigma) and phosphatase inhibitor cocktail (Roche). Lysates were incubated with primary antibodies and protein-G– or protein-A–agarose beads (P4691 and P2670, Sigma) or with anti-FLAG M2 Affinity Gel (Sigma, A2220) for 1 h at 4°C. Beads were washed with RIPA or YP-IP buffer (0.1% NP40, 25 mM Tris-HCl pH 7.5 and 150 mM NaCl) 3–5 times and once with 10 mM Tris-HCl pH 7.5. Immunoprecipitated samples and 10 µg of the input were evaluated by western blotting.

Cultured cells were lysed with RIPA or NP40 lysis buffers (see above). Proteins were resolved using SDS-PAGE (Invitrogen NuPage gel 4–12%), transferred to nitrocellulose membrane and probed with primary antibodies over night. For anti-ubiquitin antibodies, membranes were boiled for 30 min in distilled water and then blocked with 0.5% fish skin gelatin in TBST. Otherwise, blocking and antibody solutions contained 5% milk. Visualisation and quantification of western blots were performed using an Odyssey infrared scanner (LI-COR Biosciences, Lincoln, NE).

Antibodies were as follows: anti-USP21 (Sigma, HPA028397; 1:1000), anti-Gli1 (clone L42B10, Biolabs, 2643S; 1:1000), anti-CUL3 (Bethyl, A301-109A; 1:1000), anti-ubiquitin (clone P4G7, Covance, MMS-258R; 1:1000), anti-HA (clone 16B12, Covance, MMS-101P; 1:1000), Polyclonal anti-FLAG (Sigma, F7425; 1:1000), Monoclonal anti-FLAG (clone M2, Sigma, F3165; 1:1000), anti-GFP and anti-RFP (sheep and rabbit, respectively, gift of Ian Prior, University of Liverpool, Liverpool, UK; 1:1000), anti-β-actin (Abcam, ab6276; 1:5000), anti-pericentrin (Abcam, ab4448; 1:1000), anti-PKAC (BD BioSciences, 610980; 1:1000) antibodies. Anti-HDAC1 antibody was obtained from Santa-Cruz (sc-6298; 1:1000), and anti-MYC antibody from Millipore (clone 4A6, 05-724; 1:1000). MLN4924 was obtained from Millennium Pharmaceuticals.

Cloning primer sequences used in this work are given in Table S3. pCXN2-FLAG-KCTD6, pCMV-3HA-Gli1 (G933D), pGLB3B-12Gli-Luc and pRL-Renilla-TK plasmids have been described previously (Canettieri et al., 2010; De Smaele et al., 2011). KCTD6 was amplified from a liver cDNA library, inserted into pCR4-TOPO (Invitrogen) and subcloned into pCMV-Tag2B-FLAG vector. The DNA encoding KCTD6[114–237] was amplified from pCR4-TOPO-KCTD6, inserted into pET151 vector and subcloned into pCMV-Tag2B-FLAG vector. pGFP-GW, pGFP-GW-USP21, pGFP-GW-USP21CS, pGFP-C1-USP21[1–174], pGFP-C1-USP21[1–210], pGFP-C1-USP21[Δ1–47], pGFP-C1-USP21[Δ1–121] and pGFP-USP21[Δ1–184] plasmids have been described previously (Urbe et al., 2012). pCDNA4-CUL3-[1-250]-FLAG and pCDNA4-CUL3-H2×5 H5×2 [L52A, E55A, Y58A, R59A, Y62A, Y125A and R128A] encoding Cul3[1-250] and Cul3m, respectively, were kindly provided by Michael Rape (University of California, Berkeley, CA). MYC, MYC–USP21 and MYC–USP21CS plasmids have been described previously (Urbe et al., 2012). pCXN2-FLAG-KCTD11, pCXN2-FLAG-KCTD21 and pCDNA-FLAG-Gli1 have been described previously (Canettieri et al., 2010; De Smaele et al., 2011). Murine Gli2 cDNA was kindly provided by Hiroshi Sasaki (Osaka University, Osaka, Japan) and sub-cloned by means of BamHI–NotI restriction digestion in a pCDNA3 vector engineered to contain an in-frame 5′ FLAG sequence.

CRISPR plasmids expressing Cas9–GFP from a CMV promoter and selected gRNAs from a U6 promoter were kindly provided by Horizon Discoveries (Cambridge, UK). The gRNAs used were designed to target the second exon of the USP21 gene, encompassing the start codon (Table S4). PKI–Cherry plasmid was kindly provided by Oliver Rocks (Max Delbrück Center for Molecular Medicine, Berlin). Smoothened agonist (SAG, CAS 364590-63-6) was purchased from Calbiochem.

Cells were lysed in NP40 lysis buffer supplemented with 10 mM N-ethylmaleimide, mammalian protease and phosphatase inhibitors. Lysates (500 µg) were incubated with protein-G beads and anti-HA antibody for 1 h at 4°C. Beads were washed three times with YP-IP buffer, and once with 1× CutSmart buffer (NEB). Beads were resuspended in 40 µl CutSmart buffer and 3 µl of CIP (NEB) and incubated for 1 h at 30°C, and the reaction was stopped by adding 5 mM of EDTA and subsequent incubation at 65°C for 30 min. Equivalent reaction volumes were loaded onto a gel and evaluated by western blotting.

Possible distant homology of the C-terminal domain of KCTD6 with known domains or structures was assessed using HHpred (https://toolkit.tuebingen.mpg.de/hhpred; Soding, 2005) and the Metaserver (https://genesilico.pl/meta2/; Bujnicki et al., 2001). Ab initio modelling was performed with ROSETTA (Rohl et al., 2004) or QUARK (Xu and Zhang, 2012). QUARK returns ten ranked fold predictions derived from clustering of ab initio models. Likewise, ten candidate structure predictions were derived from 1000 ab initio models using the clustering algorithm of the ROSETTA suite. Structural similarity of predictions to known folds was assessed with DALI (Holm and Sander, 1993), and the classifications of proteins of interest were derived using the SCOP database (Andreeva et al., 2008).

Cells on coverslips were rinsed once with PBS, fixed at −20°C with methanol, blocked in 10% goat serum, and incubated with primary and secondary antibodies in 5% goat serum (1 h each), all in PBS. Secondary antibody was conjugated to Alexa-Fluor-594 or Alexa-Fluor-355. Immunofluorescence images were taken using a 3i spinning disc microscope (63× oil objective) and a digital camera (Hamatsu, CMOS). The images were then processed using SlideBook software and Photoshop CS4 Version 11.0 (Adobe). Pearson's correlation coefficients were calculated using Coloc2 Plugin on Fiji.

P-values are indicated as *P<0.05, **P<0.01, ***P<0.001 and ****P<0.0001 and derived either by two-tailed paired t-test or, for multiple comparison analysis, by one-way ANOVA and Dunnett's post-hoc test using GraphPad Prism6.

The authors thank Oliver Rocks, Michael Rape and Horizon Discovery for generously providing plasmids and antibodies, and Rebecca Eccles and Oliver Rocks for providing comments to the manuscript.