SCFFbxw5 targets MCAK in G2/M to facilitate ciliogenesis in the following cell cycle

The microtubule depolymerase Kif2C/MCAK plays important roles in various cellular processes and is frequently overexpressed in different cancer types. Despite the importance of its correct abundance, remarkably little is known about how MCAK levels are regulated in cells. Using comprehensive screening on protein microarrays, we identified 161 candidate substrates of the multi-subunit ubiquitin E3 ligase SCFFbxw5, including MCAK. In vitro reconstitution assays demonstrate that MCAK and its closely related orthologs Kif2A and Kif2B become efficiently polyubiquitylated by neddylated SCFFbxw5 and Cdc34, without requiring preceding modifications. In cells, SCFFbxw5 targets MCAK for proteasomal degradation specifically during G2/M. While this seems largely dispensable for mitotic progression, loss of Fbxw5 leads to increased MCAK levels at basal bodies, which impair formation of primary cilia in the following G1. We have thus identified a novel regulatory event of ciliogenesis that occurs already within the G2/M phase of the preceding cell cycle.


Introduction
In the past decade, primary cilia have gained massive attention due to the discovery of their involvement in a plethora of human diseases, ranging from organ-related disorders such as polycystic kidney disease to more pleiotropic syndromes like Bardet-Biedl 1-3 . By forming antenna-like membrane protrusions that enrich diverse cellular receptors, primary cilia are important signalling hubs for tissue development and homeostasis 4 . Ciliogenesis takes place in almost all human cells and requires drastic remodelling of centrosomes that includes their migration to the cortex and a microtubuledependent generation of a plasma membrane protrusion 5,6 . Upon re-entry into the cell cycle, primary cilia must be resorbed in order to release centrosomes for spindle formation and thus reappear only after mitotic exit 7 .
Recently, the ubiquitin proteasome system (UPS) has been identified as an important regulator of ciliogenesis [8][9][10] . Ubiquitylation defines the attachment of ubiquitin to substrate proteins via an enzymatic cascade, in which the final transfer to target lysine residues is catalysed by a so-called E3 ligase in a highly specific manner [11][12][13] . Since ubiquitin itself contains acceptor lysine residues for other ubiquitin moieties, chains of different linkages can be formed, some of which lead to proteasomal degradation 14 . Examples for ubiquitin-dependent control of ciliogenesis comprise the E3 ligase Mindbomb1 that antagonises ciliogenesis by targeting Talpid3, and the Cullin-RING ligases (CRL) Cul3-KCTD17 and SCF Fbxw7 that promote ciliogenesis via degradation of trichoplein and Nde1, respectively [15][16][17] .
CRLs constitute the biggest family of ubiquitin E3 ligases being responsible for up to 20% of proteolytic ubiquitylation events in cells 18,19 . The best-studied class of CRLs are SCF (Skp1-Cul1-F-box protein) E3 ligases that use Cul1 as a central scaffold protein. Cul1 recruits a small RING (really interesting new gene) containing protein called Rbx1 via its C-terminus and a substrate receptor module composed of Skp1 and one out of 69 interchangeable F-box proteins via its N-terminus 20,21 . Catalytic activity of SCF E3 ligases requires the modification of Cul1 with the ubiquitin-like modifier Nedd8. This induces conformational changes that facilitate ubiquitin transfer from the Rbx1-bound E2-Ubiquitin thioester to the substrate recruited by the F-box protein 22,23 .
In order to discover further substrates of Fbxw5, we now employed comprehensive substrate screening on protein microarrays and identified Kif2C/MCAK (mitotic centromere-associated kinesin) as an important novel target of SCF Fbxw5 . MCAK is the best studied member of the kinesin-13 family, which are non-motile kinesins that use their central motor domain to depolymerise microtubules (MTs) in an ATP-dependent manner [30][31][32] . MCAK is mostly known for its role in mitosis, where it localises to spindle poles, spindle MTs and centromeres exerting crucial roles in spindle formation and chromosome segregation [33][34][35] . Its activity is heavily regulated through phosphorylation by different mitotic kinases, such as AuroraA, AuroraB, Cdk1 and Plk1 [36][37][38][39][40] . Recently, MCAK has been implicated also in non-mitotic events, such as DNA damage repair and ciliogenesis 41,42 .
Here, we show that SCF Fbxw5 regulates MCAK protein levels by specifically catalysing K48 ubiquitin chains on MCAK in a highly efficient manner.
Although this process occurs during G 2 /M, loss of Fbxw5 does not provoke severe defects in mitosis but instead impairs ciliogenesis later in the following G 0 phase. Our work thus demonstrates an intriguing regulatory process required for ciliogenesis that takes place not during the event of ciliogenesis itself but rather within the preceding cell cycle.

In vitro ubiquitylation screening identifies 161 candidate substrates of SCF Fbxw5
In order to identify novel substrates of the SCF Fbxw5 complex, we performed an in vitro ubiquitylation screen on commercial protein microarrays (ProtoArray®v5.0, ThermoFischer Scientific) containing more than 9000 human proteins expressed and purified as GST-fusion proteins from insect cells. Conditions that we have previously used for in vitro ubiquitylation of the SCF Fbxw5 substrate Eps8 served as a blueprint for the screen 24 . Two arrays were incubated with recombinant E1, E2s (UbcH5b and Cdc34 together), Ubiquitin, neddylated SCF Fbxw5 complex (generated by a split and co-express method 43 ) and an ATP regenerating system to ensure reproducibility ( Fig 1A).
As a control, two arrays were probed with the same mix lacking SCF complexes. To detect ubiquitylated proteins, arrays were stringently washed, incubated with FK2 antibodies specific for conjugated ubiquitin 44 , scanned, quantified and normalised ( Fig 1B). As shown in Fig EV1A duplicate reactions were highly reproducible. SCF Fbxw5 -specific ubiquitylation substrates were then selected based on two criteria: first, signal intensity above an arbitrary threshold (i.e. 500 AU) and second, a more than 5 fold increase in signal intensity over control arrays lacking E3 ligase (Fig 1C and Fig EV1B). We identified a total of 161 candidates (Supplementary Table 1) fulfilling the above-mentioned criteria. Among them were the previously identified substrates Sas6 and Sec23b, demonstrating the capability of the in vitro system to identify SCF Fbxw5 targets (Fig 1C, note that Ask1 and Eps8 were not present on the array). Gene ontology enrichment analysis of cellular components via the DAVID webtool 45,46 revealed that a high proportion of these substrates localise to the cytoplasm, which is in line with the described cytoplasmic localisation of Fbxw5 25 . Furthermore, components of the cytoskeleton and vesicle-based transport were enriched among the SCF Fbxw5 substrates ( Fig 1D). This fits well to the previously identified substrates Eps8, Sas8 and Sec23b and suggests that Fbxw5 could act as a master regulator of cytoskeletal-dependent transport processes. In order to further validate our screen, we picked candidates from these categories and tested their ubiquitylation efficiency by SCF Fbxw5 in solution with purified substrates. All selected candidates were efficiently ubiquitylated ( Fig 1E) and a high proportion of these substrates interacted with Fbxw5 in coimmunoprecipitation experiments (Fig EV1C and D), demonstrating the high reliability of our screen.

Fbxw5 interacts with kinesin-13 family members
Due to its pivotal role in different cellular processes and its strong signal intensity in the screen, we focused on the candidate Kif2C/MCAK for further studies. Since ubiquitylation by SCF complexes requires substrate recruitment, we first tested if MCAK interacts with the substrate receptor Fbxw5. As shown by co-immunoprecipitation of tagged proteins, MCAK binds to full length and an F-box lacking mutant of Fbxw5, but not to Fbxw7, indicating an interaction that is independent of other SCF components ( Fig   2A). Accordingly, in vitro pull-down experiments using purified proteins mixed with competing E.coli lysates revealed stoichiometric precipitation of MCAK with the Fbxw5/Skp1 sub-complex with no other specific proteins present in the pull-down, confirming that MCAK binds Fbxw5 in a direct and efficient manner ( Fig 2B). In order to test if MCAK binds Fbxw5 also in intact cells, we carried out NanoBRET 47 experiments by overexpressing HaloTag(HT)-tagged Fbxw5 and MCAK fused to NanoLuc luciferase in HeLa cells. Here, Fbxw5 and MCAK generated significantly stronger signals than the negative controls ( Fig 2C), confirming that both proteins are also able to interact in living cells.
Human cells express two orthologs of MCAK -Kif2A and Kif2B -that have some overlapping and non-overlapping functions in cells 48,49 . Kif2A was present on the protoarray, but was below our threshold and Kif2B was not present on the array. Nevertheless, all three proteins share high sequence similarity and we therefore tested if these orthologs also interact with Fbxw5.
Indeed, Kif2B displayed a similar interaction with Fbxw5 in co-IPs and Kif2A precipitated weakly in these experiments (Fig 2D). However it did bind efficiently to Fbxw5 in in vitro pull-down assays (Fig 2E). Taken together, our results demonstrate that Fbxw5 can directly recruit all three kinesin-13 proteins.

SCF Fbxw5 ubiquitylates MCAK in a highly efficient and specific manner
Considering that MCAK was a strong hit in our protoarray screen, we were surprised by its rather modest ubiquitylation efficiency within our validation experiments ( Fig 1E). In order to investigate the ubiquitylation reaction in more detail, we generated neddylated SCF complexes in high amounts using the baculoviral expression system (Fig 3A and Fig EV2A and B). One difference between the screen and the control experiments was the nature of ubiquitin (FITC-labelled vs His 6 -tagged). We thus wondered if the His 6 -tag on ubiquitin had any impact on the ubiquitylation efficiency of MCAK. Indeed, while MCAK was only weakly ubiquitylated with His 6 -tagged ubiquitin, the reaction became much more accelerated using untagged ubiquitin ( Fig   EV2C). Of note, ubiquitylation efficiency of Eps8 was not impaired by the His 6tag ( Fig EV2D). In addition, the effect was specific for the His 6 -tag, as MCAK could be efficiently ubiquitylated with Flag-tagged ubiquitin (Fig EV2E). One explanation for this observation may be a higher positive net charge on the surface of MCAK that repels the His 6 -tag on ubiquitin, but it shows in general that care must be taken when using tags in ubiquitiylation experiments.
Using untagged ubiquitin for further characterisation, we could confirm that the slower migrating bands of MCAK in the reaction are due to modification by ubiquitin, as they were absent upon drop out of any essential component of the ubiquitin system ( Fig 3B). Furthermore, we found that the reaction was much more pronounced using Cdc34 as an E2 compared to UbcH5b and combination of both did not further improve the reaction (Fig 3B and C and Fig   EV2F). Unrelated proteins like Plk1 or RanGAP1 were not ubiquitylated ( Fig   3D) and the reaction was highly specific to SCF Fbxw5 as either replacing Cul1/Rbx1 with Cul4A/DDB1/Rbx1 ( Fig 3D and Fig EV2A, (Fig EV2J), demonstrating that the reaction does not require preceding post-translational modifications on the substrate. In line with the observed binding capability, E.coli-derived Kif2A and Kif2B were also efficiently ubiquitylated (Fig 3F), showing that SCF Fbxw5 is able to target all three orthologs.
Since ubiquitin can form chains of different linkage types with distinct physiological outcomes, we tested whether one of the two most prevalent ones (i.e. K48 and K63) are catalysed by SCF Fbxw5 on MCAK. Using UbcH5b as an E2, the ubiquitylation pattern of MCAK was not affected by any of the mutants including methylated ubiquitin that lacks functional acceptor aminogroups ( Fig 3G and Fig EV2K). This suggests that UbcH5b mainly catalyses mono-ubiquitylation on MCAK, as it has been shown also for the SCF βTrCP substrate IκBα 50 . Reactions using Cdc34 as E2 on the other hand showed massive loss of high molecular weight species of MCAK and Kif2A using either the K48R single or KK48,63RR double mutant, but not with K63R ubiquitin (Fig 3G and Fig EV2l). The pattern of K48R was the same as for methylated ubiquitin and displayed high similarity with the one obtained with UbcH5b, demonstrating that SCF Fbxw5 in concert with Cdc34 catalyses almost exclusively K48 chains on several lysine residues of MCAK and Kif2A.

SCF Fbxw5 affects centrosomal MCAK levels in G 0
Since K48 chains provide an efficient signal for proteasomal degradation, we examined MCAK levels in RPE-1 cells upon Fbxw5 knockdown. Compared to non-targeting siRNA, MCAK amounts were slightly increased in asynchronously growing cells. Similar effects could be observed after G 2 arrest using the Cdk1 inhibitor RO-3306 51 and upon mitotic arrest using nocodazole, in which MCAK levels were in general higher 52 (Fig 4A and B).
However, the most pronounced difference appeared in quiescent cells that had been serum-starved for 24 hours. Here, MCAK levels went almost below detection in control samples, but were 4-fold higher upon Fbxw5 knockdown.
Due to this strong effect, we focused on serum-starved cells and used immunofluorescence to test if a specific sub-population of MCAK becomes increased after Fbxw5 knockdown. We detected one prominent signal that colocalised with centrosomes (shown by ODF2 staining) and proved to be specific as it disappeared after MCAK knockdown ( Fig 4C). In line with results from western blotting, this signal was on average significantly stronger after The observation that endogenous and mNG-tagged MCAK slowly disappear upon entry into a quiescent G 0 state in an Fbxw5-independent manner indicates the existence of another regulatory pathway. In line with this, proteasomal inhibition via MG-132 during the last 6 hours of serum starvation significantly increased total MCAK levels in control but also in Fbxw5 knockdown cells (Fig 5B). One likely candidate for an alternative regulatory pathway is the Anaphase-Promoting Complex/Cyclosome (APC/C) -another multi-subunit ubiquitin E3 ligase that has been shown to target MCAK via Cdc20 in HeLa cells 53 and via Cdh1 in vitro 54,55 . In order to test if the APC/C targets MCAK after mitotic exit, we again performed mitotic shake-off experiments with release into serum-free medium. This time, we added the APC/C inhibitors Apcin and proTAME after mitotic completion (~2 hours after shake-off and nocodazole washout). While Apcin prevents substrate recognition by masking the D-Box binding site of Cdc20, proTAME inhibits incorporation of both substrate receptors (Cdc20 and Cdh1) into the APC/C complex. Combination of both compounds has been shown to completely block APC/C activity 56,57 . As expected, total and centrosomal MCAK signals disappeared almost completely 20 hours after mitosis in DMSO-treated samples (Fig 5C and D). However, this effect was partially reverted in Apcintreated cells and almost completely reverted upon proTAME addition, suggesting that the APC/C is indeed responsible for MCAK removal upon entry into quiescence. Taken together, our findings show that after mitotic exit MCAK is predominantly regulated by the APC/C.

SCF Fbxw5 regulates MCAK in G 2 /M
The experiments described above did not reveal any stabilisatory effect on MCAK upon Fbxw5 depletion after mitotic exit. However, one striking difference that was revealed in the time course experiments (Fig 5A and Fig   EV3B) were the elevated levels of MCAK in Fbxw5 knockdown cells at time point zero. Accordingly, depletion of Fbxw5 led to elevated mNG-MCAK levels already in prometaphase ( Fig EV3C). As mNG-MCAK expression was induced only for 24 hours in this experiment, these results imply that Fbxw5 targets MCAK before mitotic exit. Since we were not able to detect ubiquitylated species of MCAK using different enrichment procedures such as TUBES 58 or FK2 antibodies (data not shown), we employed cycloheximide (CHX) chase experiments in different pre-mitotic cell cycle arrests to further narrow down the exact timing of the Fbxw5-dependent regulation. Whereas MCAK remained relatively stable for 6 hours in asynchronously growing and in S phase arrested cells, it became unstable in cells arrested in G 2 with the Cdk1 inhibitor RO-3306 in an Fbxw5-dependent manner ( Fig 6A). This is in line with our previous work on Eps8, which was also targeted during G 2 24 , indicating that SCF Fbxw5 is particularly active during this cell cycle stage.
To gain further evidence that Fbxw5 targets MCAK specifically during G 2 , we investigated MCAK amounts upon proteasomal inhibition. In contrast to asynchronously growing cells, MG-132 treatment under G 2 arrest led to a moderate but reproducible increase in MCAK amounts ( Fig 6B). Since MG-132 treatment did not further increase MCAK levels upon Fbxw5 knockdown, these results confirm that MCAK becomes proteasomally degraded in an Fbxw5-dependent manner during G 2 . Similar results were obtained for nocodazole-arrested cells. However, due to the incompatibility of CHX with mitotic arrested cells 59 , we were not able to further confirm this effect.
Nevertheless, our results together demonstrate that the activity of SCF Fbxw5 towards MCAK starts during G 2 , probably persist during mitosis and becomes undetectable after mitotic exit.

Fbxw5-dependent regulation of MCAK in G 2 /M is required for ciliogenesis
Since excess activity of MCAK has been shown to increase mitotic duration 35,60 , we speculated that its regulation by Fbxw5 in G 2 /M may be an essential process for mitosis. However, we did not observe a striking increase in the mitotic index of RPE-1 cells ( Fig EV4A) and mild effects such as a delay in prometaphase may well be due to other Fbxw5 substrates like Sas6 or Eps8, which have been shown to impact on mitotic progression, too 24,25 .
A second period during which elevated MCAK levels could be detrimental is the G 0 phase. In fact, overexpression of kinesin-13 proteins (including MCAK) has recently been demonstrated to impair ciliogenesis in RPE-1 cells upon serum starvation 42 . Consistent with this, we observed a dramatic decrease in ciliated cells upon MCAK overexpression in serum-starved RPE-1 cells using our doxycyclin-inducible system. Interestingly, this defect was dependent on the timing of MCAK overexpression relative to serum withdrawal. Whereas inducing MCAK overexpression before serum starvation provoked a massive loss of ciliated cells, it had only little impact when initiated in cells that had been already serum starved for 24 hours (Fig 7A). Regardless of the timing, cilia that were still present upon MCAK overexpression showed a significant reduction in their length that correlated with MCAK intensities at basal bodies ( Fig EV4B and C). Taken together, these results suggest that excess centrosomal MCAK strongly inhibits the de novo generation of cilia, but only decreases the lengths of axonemes on pre-existing ones without leading to a complete resorption in most cases.
Since loss of Fbxw5 increased MCAK levels at centrosomes during serum starvation ( Fig 4C), we wondered if Fbxw5 knockdown induces similar defects in ciliogenesis of wild type RPE-1 cells. Indeed, knockdown of Fbxw5 with three different siRNAs led to a significant reduction in ciliated cells (Fig 7B).
Similarly to MCAK overexpression, remaining cilia under these conditions displayed on average much shorter axonemes ( Fig EV5D). Strikingly, simultaneous knockdown of MCAK almost completely rescued both Fbxw5dependent ciliogenesis phenotypes, indicating that these effects are due to elevated levels of MCAK and not of any other substrate. In conclusion, our data demonstrate that loss of Fbxw5 impairs ciliogenesis via elevated MCAK levels at the beginning of G 0 .

Discussion
Using comprehensive substrate screening on protein microarrays, we identified MCAK as a bona fide substrate of SCF Fbxw5 , assigned its regulation pathway to the G 2 /M phase of the cell cycle and demonstrated that this process is required for ciliogenesis upon entry into a quiescent state. In addition, our in vitro ubiquitylation screening approach provides a useful and reliable source for potential Fbxw5 substrates. The number of 161 candidate substrates may seem high at first glance, but one has to keep in mind that proteins are probed here in a cell-free system. Fbxw5 may regulate some of its substrates only in certain cell types or during specific developmental stages. In contrast to cell-based screens, the protein microarray method is unique in yielding a comprehensive list of potential substrates without being limited to a particular cellular context. a common mechanism that allows timely formation of primary cilia in the following G 1 /G 0 phase. It will be interesting to see, how such preceding regulatory events impact on cilia-dependent developmental programs within multicellular organisms and whether they play a critical role in ciliopathies.  Beloshistov for critical reading of the manuscript and the whole Melchior lab for helpful discussions, reagents and advise.

Declaration of interest
The authors declare no conflict of interests.

Material and Methods
Cell culture were cultured in DMEM supplemented with 2 mM glutamine and 10% FBS.

RPE-1 hTERT cells (ATCC
Hela cells were cultured in DMEM Glutamax TM supplemented with 10% FBS. All cell lines were authenticated at the DKFZ Genomics and Proteomics Core Facility (Heidelberg), regularly tested for mycoplasm contamination, kept constantly below confluency and were cultivated at 37°C with 5% CO 2 for no longer than 8 weeks (~20 passages). In order to induce ciliogenesis, cells were washed 2x with phosphate-buffered saline (PBS) and then incubated with DMEM/F12 medium lacking FBS.

Plasmid transfection
Hek293T cells were transiently transfected 1 day after seeding in 15

Generation of neddylated Cul1 and Cul4A complexes
The bacteria-derived split Cul1A-Cul1B/Rbx1 subcomplex was obtained from

In vitro ubiquitylation
In vitro ubiquitylation was performed as described previously 24   For western blotting, proteins were resolved on 7.5% or 7.5-15% gradient SDS-polyacrylamide gels. After electro-transfer onto PVDF membranes using wet-tank blotting systems (self-made), membranes were stained with coomassie (0.1% coomassie brilliant blue R250 in 50% ethanol and 10%

IP and western blot analysis
acidic acid) and cut in order to simultaneously detect two proteins from the same blot. After destaining (50% ethanol, 10% acidic acid) and washing with

Image analysis
Image analysis was performed using FIJI 74  Analysis of MCAK signals at ODF2 in Fig 5D and Fig EV3B was performed accordingly, except that here ROIs were generated by thresholding after subtracting two maximum intensity projection images subjected to two different Gaussian blurs (σ = 2 or 3 + threshold 25 ( Fig 5D) and σ = 3 or 4 + threshold 3 (Fig EV3B)) in order to enable use of same settings for images with highly diverse MCAK signal intensities.

Incucyte
Cells were seeded on 12-well plates 24 hours before imaging with an Incucyte® S3 Live-Cell Analysis System. Images were taken using the 20x air objective. H2B-mRuby signals were used to count all cells automatically using the same settings for all images via the Incycyte software. Mitotic cells were counted manually and the percentage was afterwards calculated.

Plasmid generation
All plasmids used in this study are shown in Appendix Table 1     Asterisk indicate an unspecific protein from E.coli Note: Kif2B purification from Sf21 cells was much less efficient and it was therefore not included in the experiment. Source data for C are presented in Source Data Table 1. Right: Same reaction except that Cul1~Nedd8/Rbx1 complexes were replaced by Cul4A~Nedd8/DDB1/Rbx1 complexes (see Fig EV2A and B).
Bottom blots: Same blots as above incubated with anti-Fbxw5 antibodies.
Autoubiquitylation of Fbxw5 indicates activity of the E3 ligase complex. E. in vitro ubiquitylation experiment as in B except that different F-box/Skp1 sub-complexes (Fbxw5/Skp1, Fbxl2/Skp1 or Fbxw7/Skp1) were used (see Fig   EV2h). F. in vitro ubiquitylation experiment as in B using E.coli-derived MCAK, Kif2A or Kif2B as substrates and Cdc34 as E2. G. in vitro ubiquitylation experiment as in B using either wild type, K48R, K63R, KK48,63RR or methylated ubiquitin (K methyl , see Fig EV2k). Top: Cdc34 used as E2. Bottom: UbcH5b used as E2.    Model: MCAK is regulated by two distinct E3 ligases at different time points.
While SCF Fbxw5 targets MCAK for degradation during G 2 /M, the APC/C takes over after mitotic exit. Via the SCF Fbxw5 pathway, cells ensure that MCAK levels are kept low upon entry into G 0 and thus permit ciliogenesis in the following cell cycle. Source data for A, B and C are presented in Source Data Table 1.