PIPKIγ targets to the centrosome and restrains centriole duplication

The centrosome, comprising one pair of centrioles and the surrounding pericentriolar material (PCM), is the major microtubule organization center and plays a crucial role in mitotic spindle assembly, primary ciliogenesis and cell morphogenesis (Debec et al., 2010; Nigg and Raff, 2009). In cycling cells, the centrosome duplicates once in S phase so that each daughter cell can inherit one centrosome after mitosis. The possession of an abnormal number of centrosomes results in mis-assembled spindles, and these frequently give rise to chromosome mis-segregation, genomic instability and aneuploid daughter cells (Ganem et al., 2009), linking centrosomal anomalies to many human diseases, including a variety of different aneuploid tumors (Ghadimi et al., 2000; Lingle et al., 2002; Nigg, 2006; Pihan et al., 1998; Yamamoto et al., 2004) and microcephaly syndromes (Megraw et al., 2011; Nigg and Raff, 2009). The procentriole is assembled by a small group of highly conserved proteins and is strictly limited to one per pre-existing centriole, with assembly occuring once per cell cycle. The availability and activity of these proteins at the centrosome have to be carefully controlled to maintain the fidelity of centriole and centrosome duplication (Bettencourt-Dias et al., 2005; Eckerdt et al., 2011; Kleylein-Sohn et al., 2007; Nigg, 2007; Nigg and Stearns, 2011).

It has been demonstrated that the polo-like kinase PLK4 is a key regulator of centriole formation and that excessive PLK4 activity leads to centriole amplification (Bettencourt-Dias et al., 2005; Eckerdt et al., 2011; Kleylein-Sohn et al., 2007; Nigg, 2007; Nigg and Stearns, 2011). PLK4 is recruited to the centrosome by CEP152 and controls the onset of centriole assembly together with this protein, which also provides a platform for centromere protein J (CENPJ or CPAP) to bind to the centrosome (Cizmecioglu et al., 2010; Dzhindzhev et al., 2010). PLK4 can carry out both autophosphorylation and the phosphorylation of other substrates, thereby priming either itself or a substrate (such as FBXW5) for ubiquitinylation and destruction. In this way, PLK4 can control its own availability (Brownlee and Rogers, 2012; Cunha-Ferreira et al., 2009; Guderian et al., 2010; Holland et al., 2010; Rogers et al., 2009) or that of other components of the centriole duplication machinery (such as SAS6) (Puklowski et al., 2011), to restrict centriole number. By contrast, protein phosphatase 2 (PP2A) was reported to promote centriole formation by facilitating SAS6 accumulation at the nascent centriole (Brownlee et al., 2011; Kitagawa et al., 2011; Song et al., 2011). These results support the existence of an intrinsic important balance between kinase and phosphatase activities during centriole duplication and highlight the importance of PLK4 to the centriole assembly machinery. In spite of the importance of controlling PLK4 activity accurately, the underlying mechanism is far from fully understood.

Phosphatidylinositol 4-phosphate 5-kinase type-1 gamma (PIP5K1C, hereafter referred to as PIPKIγ) is a lipid kinase that phosphorylates phosphatidylinositol 4-phosphate to generate phosphatidylinositol (4,5)-bisphosphate [PI(4,5)P₂]. To date, six alternatively spliced isoforms of PIPKIγ have been reported (Giudici et al., 2004; Giudici et al., 2006; Ishihara et al., 1998; Schill and Anderson, 2009; Xia et al., 2011). PIPKIγ isoforms are targeted to distinct subcellular sites. By regulating the dynamics of regional PI(4,5)P₂ pools, they are implicated in distinct cellular processes such as vesicular trafficking (Bairstow et al., 2006; Ling et al., 2007; Sun et al., 2013; Thapa et al., 2012; Thieman et al., 2009; Xiong et al., 2012), calcium signaling (Wang et al., 2004) or cell adhesion and migration (Di Paolo et al., 2002; El Sayegh et al., 2007; Ling et al., 2007a; Ling et al., 2002; Sun et al., 2007; Thapa et al., 2012; Thieman et al., 2009; Wang et al., 2004; Xiong et al., 2012). Moreover, PIPKIγ can regulate some proteins, such as E-cadherin (Ling et al., 2007), by directly interacting with them. Here, we report that PIPKIγ isoform 3 (PIPKIγ_i3) is targeted to the centrosome and negatively regulates centriole duplication. This lipid kinase associated with the proximal ends of parental centrioles in a cell-cycle-dependent manner, just like many other centrosomal proteins. Similar to PLK4 and CPAP, the residence of PIPKIγ at the centrosome depended upon CEP152. We show that loss of PIPKIγ results in centriole amplification without blocking the cells in S or G2/M phase, indicating that PIPKIγ is a negative regulator of centriole duplication. Indeed, further investigation revealed that PIPKIγ directly interacts with PLK4 and inhibits its activity. These results suggest the existence of an additional level of regulation of PLK4 activity and centriole biogenesis, acting in combination with phosphorylation-dependent protein ubiquitinylation and degradation.

Using previously characterized rabbit polyclonal antibodies that recognize all of the PIPKIγ isoforms, we observed that PIPKIγ localizes to multiple subcellular sites including focal adhesions, adherens junctions and recycling endosomes (Ling et al., 2007; Ling et al., 2002), as well as the centrosome. To investigate the unexpected centrosomal localization of PIPKIγ, we generated new rabbit polyclonal and monoclonal antibodies to ensure that this localization pattern was not an artifact from the previous antibody. Both the purified polyclonal antibodies (supplementary material Fig. S1B) and a monoclonal antibody against PIPKIγ (Fig. 1A, left panel) stained the centrosome, which is highlighted by the centrosomal marker centrin 2 in HeLa cells. By contrast, purified rabbit polyclonal antibodies specifically recognizing PIPKIα or PIPKIβ (supplementary material Fig. S1A) did not show staining around the centrosome (Fig. 1A). The centrosomal binding of anti-PIPKIγ antibody was completely abolished after treatment with a PIPKIγ siRNA that targeted all PIPKIγ isoforms (Fig. 1B), or in the presence of excess PIPKIγ but not PIPKIα protein (supplementary material Fig. S1B), supporting the authenticity of this PIPKIγ localization. In addition, we isolated centrosomes from HeLa cells (Fig. 1C) or mouse kidney (supplementary material Fig. S1C) using sucrose gradient centrifugation and analyzed the fractions by western blot. Similar to other centrosomal proteins that were identified using this method (Kaplan et al., 2004), PIPKIγ co-fractionated with the centrosome components human SAS6 (hereafter referred to HsSAS-6), CPAP and γ-tubulin – a result that further supported the hypothesis that PIPKIγ associates with centrosomes as suggested by immunofluorescence microscopy. Moreover, treatment with nocodazole, an inhibitor of microtubule polymerization, had no effect on PIPKIγ signal at the centrosome, although it efficiently disrupted microtubule integrity (supplementary material Fig. S1D), indicating that the association between PIPKIγ and the centrosome is independent of the microtubule cytoskeleton. These results demonstrated that PIPKIγ was a bona fide component of the centrosome. In addition to its presence in HeLa cells, we also observed PIPKIγ at the centrosome in human MDA-MB-231 and U-2 OS cells and in murine IMCD3 and NIH3T3 cells (supplementary material Fig. S1E and data not shown), suggesting a pervasive, potentially functional role of PIPKIγ at the centrosome.

To determine which PIPKIγ isoform was targeted to centrosomes, we fused an HA tag to the N-terminus of all six known PIPKIγ splice variants. As shown in Fig. 1D and supplementary material Fig. S1F, only PIPKIγ_i3 displayed a centrosome-associated signal. Other isoforms showed vesicle-like cytoplasmic targeting, or localized to focal adhesions (PIPKIγ_i2), as reported previously (Di Paolo et al., 2002; Ling et al., 2002). PIPKIγ_i3 contains a unique 26-amino-acid insertion (amino acids 641-666) at the C-terminus, between the sequence representing the end of PIPKIγ_i1 and the beginning of the PIPKIγ_i2 C-terminal extension (Fig. 1E). However, the C-terminal region of PIPKIγ_i3 (amino acids 445-End) is predominantly cytoplasmic (Fig. 1D, right panel). The N-terminal fragment of PIPKIγ_i3 (amino acids 1-445), which is common to all PIPKIγ variants, localized to the centrosome (Fig. 1D, middle panel). These results suggest that the centrosome-targeting motif is likely to be located within the N-terminal 445 amino acids. Because PIPKIγ_i3 is the only isoform that showed centrosome targeting, it is possible that the unique C-terminal insertion of PIPKIγ_i3 might provoke a conformational adjustment that acts to expose the centrosome-targeting motif embedded in the first 445 amino acids.

To examine the fine structural details of the association of PIPKIγ with the centrosome, we co-labeled HeLa cells with antibodies against PIPKIγ and various centrosome components. The localizations of these proteins around the centrosome were analyzed using three dimensional structured illumination microscopy (3D-SIM). Co-staining for ninein (a subdistal appendage marker on the mother centriole) (Nakagawa et al., 2001) and centrin 2 (which marks the distal end of the centriole cylinder) (Paoletti et al., 1996) showed that PIPKIγ was positioned below these two proteins (Fig. 2A). Further examination indicated that PIPKIγ localized between ninein and C-Nap1, a linker protein between centrioles (Mayor et al., 2000) (Fig. 2B), as well as between C-Nap1 and outer dense fiber protein 2 (ODF2), which marks the distal appendage of the mature mother centriole (Piel et al., 2000) (Fig. 2C). In these images, PIPKIγ seemed to be distributed along the side of the centriole wall in the pericentriolar matrix and appeared likely to form a ring around the proximal end of centrioles with a diameter smaller than that of ninein.

To address whether the centrosomal localization of PIPKIγ is cell-cycle dependent, as is the case for many other centrosome proteins, we subjected HeLa cells to immunofluorescence microscopy to visualize PIPKIγ together with centrin 2 and DAPI. Based on the staining of centrin 2 and DAPI, cells were categorized as being in G1 phase (two centrin 2 foci and intact nuclear membrane), S phase (four clustered centrin 2 foci and intact nuclear membrane), G2 phase (two separated pairs of centrin 2 foci and intact nuclear membrane) and various stages of M phase (broken nuclear membrane and condensed chromosome DNA in varying arrangements). PIPKIγ staining around the centrosome was observed in cells at G1, S and G2 phase (supplementary material Fig. S1G). At prophase, the PIPKIγ signal diminished and it remained absent from centrosomes until late telophase, when PIPKIγ was recruited back to both centrosomes that would soon be inherited by each daughter cell (supplementary material Fig. S1G). Consistent with its centrosome targeting, the protein level of PIPKIγ decreased significantly as cells progressed through mitosis and then recovered at mitotic exit in synchronized HeLa cells (Fig. 2D). This followed the same trend as CPAP (Tang et al., 2009) (Fig. 2D), a CEP152-interacting protein that functions early in centriole duplication; however, this trend is different from that of HsSAS-6 (Fig. 2D), which is degraded during late M phase and is absent in G1 phase (Strnad et al., 2007), suggesting that PIPKIγ might have a specific function at the centrosome.

CEP152 is a scaffold protein localizing around the proximal end of centrioles and is required for the centrosomal localization of PLK4 and CPAP (Cizmecioglu et al., 2010; Dzhindzhev et al., 2010; Hatch et al., 2010). Because PIPKIγ also localizes around the proximal end of centrioles, we examined whether CEP152 was responsible for the targeting of PIPKIγ to the centrosome. The physical association between PIPKIγ and CEP152 was first determined using a co-immunoprecipitation assay. Indeed, HA-tagged PIPKIγ_i3 was pulled down with overexpressed GFP–CEP152 in HEK293T cells (Fig. 3A), suggesting that these two proteins have the potential to bind to each other. The CEP152 antibody recognized multiple bands (Fig. 3B), which could be all knocked down by CEP152-specific siRNA (Fig. 3D), suggesting that endogenous CEP152 exists in variant forms. Endogenous PIPKIγ co-immunoprecipitated with the endogenous CEP152 protein that migrated faster on SDS-PAGE gels (Fig. 3B). Although it is not known whether this PIPKIγ-associated CEP152 is an unmodified, truncated or alternatively spliced form of CEP152, our results support the existence of an in vivo physical connection between these two proteins. Because the recombinant full-length CEP152 expressed in Escherichia coli was highly degraded or insoluble no matter whether it was fused to a GST, His or maltose-binding protein (MBP) tag, we constructed and purified MBP-tagged CEP152 fragments and tested their interaction with His–PIPKIγ. Both the N-terminal 748 residues (CEP152^1-748) and the C-terminal 906 residues (CEP152^749-1654) pulled down PIPKIγ (supplementary material Fig. S2A). However, the N-terminal 217 residues of CEP152 (CEP152^1-217), to which PLK4 directly binds (Hatch et al., 2010), did not interact with PIPKIγ (supplementary material Fig. S2A). These data suggest that PIPKIγ might directly bind to the C-terminus of CEP152, but further investigation is needed to obtain a definitive conclusion because of the heavy degradation of our recombinant CEP152 polypeptides.

To understand the biophysical significance of the association between PIPKIγ and CEP152, we first tested whether they colocalized at the centrosome. It has been reported recently by multiple groups that centrosomal proteins, including CEP152, form a ring-like structure around the centriole (Fu and Glover, 2012; Lawo et al., 2012; Mennella et al., 2012; Sonnen et al., 2012). Consistently, our 3D-SIM images supported a ring-like colocalization between PIPKIγ and FLAG–CEP152 (Fig. 3C), suggesting that PIPKIγ localizes to the intermediate PCM around the proximal end of centrioles in a manner similar to CEP152 (Fu and Glover, 2012; Lawo et al., 2012; Mennella et al., 2012; Sonnen et al., 2012). More importantly, the colocalization between CEP152 and PIPKIγ reinforced the evidence for a physical interaction and suggested a functional association between these two proteins. Indeed, siRNA-mediated depletion of CEP152 (Fig. 3D) led to a loss of PIPKIγ signal from the centrosome (Fig. 3E,H), indicating that CEP152 is necessary for the recruitment of PIPKIγ to the centrosome. Moreover, depletion of CEP152 also eliminated the centrosome targeting of exogenous PIPKIγ_i3^1-445 (supplementary material Fig. S2B), further suggesting that CEP152 provides a necessary structural platform for the stable association of PIPKIγ with the centrosome.

In addition to CEP152 (Fig. 3F), CEP192 was recently shown to be important for the recruitment of centrosomal proteins (Sonnen et al., 2013). In CEP192-depleted cells, we found that the centrosomal signal of CEP152 was largely eliminated (Fig. 3G), as reported previously (Sonnen et al., 2013); however, the PIPKIγ signal around the centrosome was mostly retained (Fig. 3G,H). Combined with our previous observation that complete loss of CEP152 blocked the targeting of PIPKIγ to the centrosome (Fig. 3E,H), these results suggest that a very small amount of centrosomal CEP152 could be sufficient for the centrosomal recruitment of PIPKIγ.

The specific cell-cycle-dependent centrosomal localization of PIPKIγ led us to investigate the potential function of PIPKIγ at the centrosome. For this purpose, we knocked down PIPKIγ in HeLa cells using lentivirus-based PIPKIγ-specific short hairpin (sh)RNA (Fig. 4A). Strikingly, the loss of PIPKIγ resulted in an augmentation of centrin 2 foci (more than four per cell) in ∼20% of the cells (Fig. 4B,C). Importantly, centriole amplification caused by depletion of PIPKIγ was almost fully rescued by expression of RNAi-resistant PIPKIγ (Fig. 4D,E). A similar increase in the number of centrin 2 foci was also observed in HeLa cells treated with two distinct PIPKIγ siRNAs (siPIPKIγ-O1 and siPIPKIγ-O2) (supplementary material Fig. S3A,B) but not in cells depleted of PIPKIα or PIPKIβ (supplementary material Fig. S3F,G). This further confirmed that the phenotype was caused by PIPKIγ depletion. To determine whether the increased centrin 2 foci resulted from cell cycle arrest in S phase or G2/M phase, we examined the cell cycle distribution of PIPKIγ-depleted cells. The incorporation of BrdU (supplementary material Fig. S3C) and staining for proliferating cell nuclear antigen (PCNA) (supplementary material Fig. S3D) were used to determine the S phase population, and phospho-histone H3 (supplementary material Fig. S3E) was used to identify cells in G2/M phase. As shown in supplementary material Fig. S3C-E, no significant difference in these markers was observed between cells treated with siPIPKIγ and cells treated with control siRNA. Therefore, the excessive centrin 2 foci in PIPKIγ-depleted cells were not likely resulted from a sustained S, G2 or M phase. Furthermore, depletion of PIPKIγ, but not PIPKIα or PIPKIβ, caused an increase in centrosome number (more than two γ-tubulin loci per cell) in other cell types such as NIH3T3 (supplementary material Fig. S3H,I) and the renal collecting duct epithelial cell line IMCD3 (supplementary material Fig. S3J). Together, these results suggest that PIPKIγ likely inhibits centriole duplication.

To test this possibility, we examined whether the expression of PIPKIγ could reverse centrosome amplification caused by hydroxyurea (HU) treatment. As shown in Fig. 4F, U2 OS cells were treated with HU for 24 h before being transfected with or without wild-type or kinase-dead HA–PIPKIγ. This experimental design limited the impact of overexpressed PIPKIγ to S phase and excluded the potential influence on other phases of the cell cycle (Hemerly et al., 2009). As shown in Fig. 4G, HU-induced centrosome amplification in U-2 OS cells was significantly suppressed by the overexpression of HA–PIPKIγ, suggesting that PIPKIγ is a negative regulator of centrosome duplication. Furthermore, the expression of both wild-type and kinase-dead PIPKIγ could rescue centriole over-duplication caused by PIPKIγ depletion or HU treatment, suggesting that the production of PI(4,5)P2 might not play an important role in centriole biogenesis.

To further characterize these excessive centrin 2 foci in PIPKIγ-depleted cells, we checked the localization of several centriolar proteins, including centriolar coiled protein of 110 kDa (CP110, which localizes at the distal ends of centrioles) (Kleylein-Sohn et al., 2007), centrin 2 (localizes at the distal ends of centrioles), centrosomal protein of 135 KDa (CEP135, which localizes at the proximal end of centrioles) (Kleylein-Sohn et al., 2007) and CPAP (localizes at the proximal ends of centrioles) (Kleylein-Sohn et al., 2007). At 48 h after HeLa cells were infected with lentivirus carrying control or PIPKIγ shRNA, we observed excessive CP110 and centrin 2 foci (more than four per cell) that associated with each other in PIPKIγ-depleted cells. However, the centrosome number (represented by γ-tubulin foci) in these cells was normal (two per cell). The numbers of CEP135 and CPAP foci were also normal (two per cell) in PIPKIγ-depleted cells (Fig. 5A,C). These results suggest that the additional centrin 2 and CP110 foci in PIPKIγ-depleted cells might correspond to over-duplicated procentrioles that were still engaged with the parental centrioles. Indeed, 72 h after lentivirus infection when the disengagement of nascent centrioles occurred, excessive CEP135, CPAP and γ-tubulin foci (more than two per cell) were observed in PIPKIγ-depleted cells (Fig. 5B,C). Moreover, transmission electron microscopy analysis of serial-sectioned PIPKIγ-depleted HeLa cells showed two procentrioles (P1 and P2, Fig. 5D, left and middle panels) attached to the proximal wall of a mother centriole (identifiable by the distal appendages) (Fig. 5D, right panel), further supporting the hypothesis that the lack of PIPKIγ leads to procentriole amplification.

It has been reported that overexpression of PLK4 (Habedanck et al., 2005) or HsSAS-6 (Leidel et al., 2005; Strnad et al., 2007) leads to biogenesis of multiple procentrioles around one pre-existing centriole. PIPKIγ depletion also yielded more than one procentriole per parental centriole, suggesting a connection with PLK4 or HsSAS-6. To test this, we designed specific siRNAs to knock down PLK4 (supplementary material Fig. S3K) or HsSAS-6 (supplementary material Fig. S3L). When PLK4 or HsSAS-6 was co-depleted along with PIPKIγ, the centriole over-duplication phenotype caused by PIPKIγ exhaustion was completely rescued (Fig. 5E), suggesting that PIPKIγ functions against PLK4 and/or HsSAS-6 at the early stage of centriole biogenesis. In agreement with this, depletion of PIPKIγ, without affecting the level of PLK4 (supplementary material Fig. S3M), synergistically enhanced the centriole amplification that resulted from PLK4 overexpression (supplementary material Fig. S3N). In addition, the centriole amplification induced in U-2 OS cells by Myc–PLK4 overexpression was significantly suppressed by expression of either wild-type or kinase-dead HA–PIPKIγ (Fig. 5F), further suggesting that PIPKIγ might counteract PLK4 function.

Like PIPKIγ, PLK4 (the key regulator of centriole duplication) localizes in the PCM around the proximal end of centrioles in a CEP152-dependent manner (Cizmecioglu et al., 2010; Dzhindzhev et al., 2010; Hatch et al., 2010). To investigate the functional correlation between PIPKIγ and PLK4, we first tested whether they physically associate with one another. Because the endogenous level of PLK4 is undetectable by immunoblotting, we examined whether endogenous PIPKIγ associates with overexpressed GFP-tagged PLK4. GFP–PLK4 co-immunoprecipitated with PIPKIγ, indicating that they form a complex in vivo (Fig. 6A). To understand the spatial correlation between PIPKIγ and PLK4, we developed a mouse monoclonal PLK4 antibody. This antibody specifically recognized purified MBP–PLK4 but not MBP or MBP–CEP152^1-748 (supplementary material Fig. S4A) and it stains PLK4 at the centrosome. This staining was abolished by two independent PLK4-specific siRNAs (supplementary material Fig. S4B). These results indicate that this new PLK4 antibody can be used to visualize endogenous PLK4 by immunofluorescence microscopy. Images obtained by 3D-SIM showed that our PLK4 antibody nicely stained the centrosome with a pattern similar to that seen for endogenous CEP152, the binding partner of PLK4 (supplementary material Fig. S4C), further endorsing the specificity of this antibody. Colocalization of endogenous PIPKIγ with endogenous PLK4 at the centrosome was observed (Fig. 6B), suggesting that the working spaces of these two proteins overlap. Both PIPKIγ and PLK4 displayed a ring-like structure with a similar shape and diameter to that of CEP152 (Fig. 3C and Fig. 6B) (Sir et al., 2011). This is in line with the notion that PLK4 and CEP152 interact and colocalize (Cizmecioglu et al., 2010; Dzhindzhev et al., 2010; Hatch et al., 2010) and position PIPKIγ at the inner to intermediate region of the PCM (Sonnen et al., 2012).

Next we tested the direct interaction between PIPKIγ and PLK4 with in vitro pull-down assays using purified MBP–PLK4 and His–PIPKIγ. PIPKIγ clearly shows direct binding to PLK4 (Fig. 6C). Using a series of truncated recombinant PLK4 fragments (Fig. 6F), we found that the cryptic polo-box (CPB) domain of PLK4 is necessary and sufficient for PIPKIγ interaction (Fig. 6D). Using truncated PIPKIγ proteins, we narrowed down the PLK4-binding region within PIPKIγ to the kinase domain (Fig. 6E,F). The N- or C-terminus of PIPKIγ alone could not bind to PLK4, and the PIPKIγ N-terminus is not necessary for PLK4 binding (Fig. 6E,F). These results strongly support the notion of a physical association between PIPKIγ and PLK4 and suggest a functional consequence of this interaction.

Because both CEP152 and PIPKIγ bind to the CPB domain of PLK4, we tested whether PIPKIγ and CEP152 compete for the interaction with PLK4. Because the full length CEP152 and the protein fragments were highly degraded when purified from E. coli, we produced the ³⁵S-labeled CEP152 N-terminal fragment (³⁵S–CEP152^1-748) by in vitro translation. Consistent with the literature (Cizmecioglu et al., 2010; Dzhindzhev et al., 2010; Hatch et al., 2010), MBP-tagged full-length PLK4 pulled down ³⁵S–CEP152^1-748 well. Although PIPKIγ could still be pulled down by MBP–PLK4, the addition of increasing amounts of PIPKIγ did not affect the binding of CEP152 to PLK4 (Fig. 7A), suggesting that PIPKIγ and CEP152 do not bind to the same region of PLK4.

Well-controlled activity of PLK4 is essential for centriole duplication and fidelity (Bettencourt-Dias et al., 2005; Eckerdt et al., 2011; Kleylein-Sohn et al., 2007; Nigg, 2007; Nigg and Stearns, 2011). PLK4 can phosphorylate both itself (Brownlee and Rogers, 2012; Cunha-Ferreira et al., 2009; Guderian et al., 2010; Holland et al., 2010; Rogers et al., 2009) and its potential in vivo substrate CEP152 (Cizmecioglu et al., 2010; Dzhindzhev et al., 2010; Hatch et al., 2010). Therefore, we overexpressed FLAG-tagged CEP152^1-748 in HeLa cells and purified this protein by immunoprecipitation. The precipitates obtained by normal mouse IgG or by anti-FLAG antibody were incubated with purified MBP–PLK4 to perform a kinase assay. Results from these experiments demonstrated that PLK4 could phosphorylate both itself and CEP152 (Fig. 7B) but not PIPKIγ (Fig. 7C), indicating that PIPKIγ would not compete with PLK4 substrates. We next tested whether the interaction with PIPKIγ influences PLK4 activity (Fig. 7D). Because PIPKIγ and PLK4 are both kinases and they might inhibit one another by competing for ATP, we compared the PLK4 activity with or without varying amounts of the wild-type (WT) or kinase-dead (KD) PIPKIγ. PIPKIγ-KD does not bind to ATP (Kunz et al., 2000) but binds to PLK4 in vitro at a comparable level to its wild-type counterpart (unpublished data). Both PIPKIγ-WT and PIPKIγ-KD inhibited PLK4 activity in a dose-dependent manner (Fig. 7D-F). When equal amounts of PIPKIγ and PLK4 were added, PIPKIγ-WT or PIPKIγ-KD resulted in a ∼20% inhibition of PLK4 autophosphorylation and CEP152 phosphorylation (Fig. 7D-F). When the amount of PIPKIγ-WT or PIPKIγ-KD was increased to threefold that of PLK4, >40% of PLK4 autophosphorylation and ∼50% of CEP152^1-748 phosphorylation were inhibited (Fig. 7D-F). PIPKIγ-WT and PIPKIγ-KD consistently repressed PLK4 at a comparable level, suggesting that this inhibition was not caused by competition for ATP. These results suggest that PIPKIγ regulates PLK4 activity by binding to PLK4. To explore how PIPKIγ inhibits PLK4 activity, we determined whether the binding of PIPKIγ affects PLK4 dimerization, which is important for the self-activation of PLK4 (Guderian et al., 2010). As shown in Fig. 7G, overexpressed Myc–PLK4 were pulled down by using purified MBP–PLK4. This interaction was not interrupted by the addition of increasing amounts of recombinant PIPKIγ (Fig. 7G), indicating a more complicated mechanism that will require future exploration. Nevertheless, in the context that PLK4 activity is crucial for PLK4 to initiate procentriole biogenesis and achieve self-regulation, our results suggest that PIPKIγ could restrain centriole duplication by binding to and inhibiting PLK4 kinase activity.

How centrioles are built and how the centriole assembly is limited to one daughter per mother in a single cell cycle are associated puzzles that have not been fully resolved. In addition to the key players identified in recent years (Brito et al., 2012; Nigg and Stearns, 2011), here we report PIPKIγ as a new component of the centriole duplication machinery that restricts centriole formation by potentially limiting PLK4 activity. As a negative regulator of centriole biogenesis, the association of PIPKIγ with the centrosome is likely to be sensitive and carefully regulated, which could explain why PIPKIγ was not described in the recent centrosome proteome studies (Andersen et al., 2003; Keck et al., 2011; Müller et al., 2011; Müller et al., 2010; Nogales-Cadenas et al., 2009; Ren et al., 2010; Zellner et al., 2011). Nevertheless, our results from multiple experiments (Figs 1, 2) provide solid evidence supporting the physical association of endogenous PIPKIγ with the centrosome. The overall levels and centrosome association of PIPKIγ are regulated through the cell cycle, as is the case for many other centriolar proteins, however, this regulation operates in a rather unique way. Compared with the positive regulators of centriole biogenesis, such as PLK4, SAS6 and SCL-interrupting locus protein (STIL), the levels of which are highest in M phase and drop sharply at mitotic exit, PIPKIγ is regulated in the opposite direction: it is downregulated during mitosis but increases rapidly immediately after mitosis. In addition, the association of PIPKIγ with a centriole is first seen at the G1–S boundary when the procentriole starts to emerge from the centriole (supplementary material Fig. S1G). This timing is similar to that of centrin 2, SAS-6 and CPAP, which are recruited to centrioles in G1/S phase to support procentriole assembly (Kleylein-Sohn et al., 2007; Paoletti et al., 1996; Strnad et al., 2007; Tang et al., 2009). Although it is not fully understood why PLK4, SAS-6 and STIL accumulate in M phase when the budding of the procentriole has completed, the cell-cycle-dependent expression and centrosome association of PIPKIγ serves its function well: it counteracts the centriole assembly machinery and inhibits the formation of excessive procentrioles.

Although only PIPKIγ_i3 (and not other PIPKIγ isoforms) is targeted to the centrosome when overexpressed, the unique insert sequence in PIPKIγ_i3, which distinguishes it from other PIPKIγ isoforms, does not contain a centrosome-targeting sequence (Fig. 1E). Instead, the centrosome-targeting signal seems to be located in the kinase domain that is commonly present in other PIPKIγ splicing isoforms and is highly conserved among all type I PIPK members; however, none of the other isoforms shows centrosomal localization. This suggests that the centrosome-targeting signal in PIPKIγ is normally hidden and needs to be modified in a certain way in order to be exposed. It is plausible that dimerization of PIPKIγ (Rao et al., 1998) covers the centrosome-targeting motif in the kinase domain. The unique C-terminal insertion of PIPKIγ_i3 might alter the protein conformation and expose the centrosome-targeting motif. Because only a fraction of PIPKIγ_i3-expressing cells showed PIPKIγ_i3 at the centrosome, it is also possible that other modification mechanisms, which could have been exhausted by overexpressed PIPKIγ_i3, are necessary to achieve a more efficient exposure of the centrosome-targeting motif in PIPKIγ. Similar to other centrosomal proteins (Fu and Glover, 2012; Mennella et al., 2012; Pelletier and Yamashita, 2012; Sonnen et al., 2012), both endogenous and overexpressed PIPKIγ exhibited a toroid-shaped distribution around the centriole in most cells, suggesting that the association of PIPKIγ with the centrosome follows the same common mechanism as other centrosomal proteins. This was reinforced by the observation that the stable association between PIPKIγ and the centrosome strictly relies on the existence of CEP152 but not CEP192. This stable association is likely to be achieved through a physical interaction between PIPKIγ and CEP152. Indeed, the PIPKIγ toroid overlapped well with the CEP152 toroid, indicating that PIPKIγ, like CEP152, is an intermediate PCM protein (Sonnen et al., 2012). However, PIPKIγ is absent from the centrosome in M phase when CEP152 is still present, suggesting that CEP152 might not be sufficient to recruit PIPKIγ to the centrosome. Further investigations are required in order to decipher the signal code that governs the cell-cycle-dependent association of PIPKIγ with the centrosome.

Traditionally, PIPKIγ regulates a variety of cellular processes by targeting to specific subcellular locales and providing PI(4,5)P₂ to the proximate downstream effector. This leads to the question of whether PI(4,5)P₂ has a role in centriole duplication. Using GFP-tagged phospholipase C delta pleckstrin homology (PLCδ PH) domain, we failed to see clear GFP signal at the centrosome owing to the strong membrane and cytosolic background (data not shown). Antibody against PI(4,5)P₂ also failed to highlight any specific subcellular locations, including the ones where PI(4,5)P₂ plays important roles, such as focal adhesions and clathrin-coated pits (data not shown). However, it has been reported that there are membrane vesicles around the centrosome (Foraker et al., 2012; Westlake et al., 2011) and that phosphoinositide-binding proteins, such as clathrin (Ford et al., 2001) and FYVE-domain-containing centrosomal protein (FYVE-CENT) (Sagona et al., 2010) localize at the centrosome and regulate centrosome integrity (Foraker et al., 2012) – suggesting that phosphoinositides, including PI(4,5)P₂, might occur and function at the centrosome. Nevertheless, in our hands both the wild-type and kinase-dead forms of PIPKIγ could similarly rescue the centriole or centrosome amplification caused by PIPKIγ depletion, HU treatment or PLK4 overexpression, suggesting that PI(4,5)P₂ is not essential in limiting centriole duplication. We reported previously that PIPKIγ can regulate some cellular process independently of PI(4,5)P₂ production. For example, PIPKIγ functions as a scaffold between E-cadherin and the clathrin-adaptor complex AP1B and thereby facilitates the transport of E-cadherin from the trans-Golgi network to the recycling endosome (Ling et al., 2007). In this process, the physical interaction of PIPKIγ with E-cadherin, but not the PI(4,5)P₂ generated by PIPKIγ, is crucial. We propose that the binding of PIPKIγ to PLK4 might impair the conformation of the protein and/or the access of ATP or other PLK4 substrates such that PLK4 activity is weakened. Although this model has not been tested in vivo owing to the highly restricted expression of endogenous PLK4 (Guderian et al., 2010; Holland et al., 2010; Sillibourne et al., 2010), it is in agreement with our observation that loss of PIPKIγ causes PLK4-dependent centriole amplification without affecting cell cycle progression.

The auto-phosphorylation of PLK4 triggers both activation and degradation of PLK4 (Guderian et al., 2010; Holland et al., 2010; Sillibourne et al., 2010). It has been shown that the degradation of PLK4 requires multi-site phosphorylation, which leads to a model in which PLK4 is destroyed once its activity reaches a certain threshold level (Guderian et al., 2010; Holland et al., 2010; Sillibourne et al., 2010). Our results suggest that PIPKIγ binds to PLK4 and inhibits its activity, providing another level of regulation in addition to phosphorylation. This seems crucial to maintain the appropriate activity of the centriole duplication machinery because loss of PIPKIγ leads to the generation of multiple procentrioles and overexpression of PIPKIγ blocks centrosome overduplication. Our results suggest that CEP152, by engaging both the positive regulator PLK4 and its negative regulator PIPKIγ at the proximal ends of parental centrioles, provides a physical platform for the accurate control of centriole assembly and homeostasis. Future studies will be necessary to provide greater insight into the functional consequences of the physical interaction between PIPKIγ and PLK4. It is also important to investigate whether PLK4 is the only centrosomal protein regulated by PIPKIγ and whether PI(4,5)P₂ has a functional role in other centrosome-related cellular processes.

HeLa, U-2 OS, MDA-MB-231, NIH3T3 and HEK293T cells were maintained in DMEM containing 10% FBS. IMCD3 cells were cultured in DMEM:F12 containing 10% fetal bovine serum (FBS). HeLa and HEK293T cells were transfected using Lipofectamine 2000 (Invitrogen) or FuGENE 6 Transfection Reagent (Promega). U-2 OS cells were transfected using X-tremeGENE 9 (Roche) or Amaxa Cell Line Nucleofector Kit V (Lonza). siRNA duplexes were introduced into cells using Lipofectamine RNAiMAX (Invitrogen). HeLa cells were synchronized with a standard double-thymidine block [two 16 h thymidine (Sigma, 2 mM) treatments separated by a 9 h release]. U-2 OS cells were synchronized with 16 mM hydroxyurea (Sigma) for 24 h. The tetracycline-inducible cell-line expressing Myc-tagged PLK4 was kindly provided by Erich Nigg (Kleylein-Sohn et al., 2007). Myc–PLK4 expression was induced by the addition of 1 µg/ml doxycycline (Clontech).

Rabbit monoclonal antibody against PIPKIγ was generated at Epitomics by immunizing rabbits with recombinant His-tagged PIPKIγ. Before collecting the spleens, anti-sera were harvested and purified against an affinity column conjugated with MBP-tagged PIPKIγ C-terminus (amino acids 469-668) to obtain polyclonal PIPKIγ antibody. Rabbit polyclonal antibodies against PIPKIα and PIPKIβ were obtained by immunizing rabbits with recombinant GST–His-tagged PIPKIα or PIPKIβ. Anti-sera were purified with affinity column conjugated with the MBP-tagged C-terminus of PIPKIα (amino acids 451–562) or PIPKIβ (amino acids 397-540). Mouse monoclonal anti-PLK4 antibody was generated at Abmart, Inc. Rabbit polyclonal antibodies against the following proteins were used: phospho-histone H3 (Ser10) (Millipore), CPAP (Proteintech), β-actin, CEP135, GFP (Abcam), CEP152 (Bethyl Laboratories, A302-480A), CEP192 (a generous gift from Erich Nigg) (Schmidt et al., 2009) and CP110 (Proteintech). Mouse antibodies against the following proteins were used: centrin 2 (20H5, a generous gift from Jeffrey Salisbury), ODF2 (Abnova), C-Nap1 (BD Biosciences), His (HIS.H8, Millipore), GFP (Roche); α-tubulin (DM1A), γ-tubulin (GTU-88), FLAG (M2) and HA (HA-7) were from Sigma; Myc (9E10), HsSAS-6 (91.390.21) and cyclin B (GNS1) were from Santa Cruz. Anti-ninein (Y-16) antibody was from Santa Cruz and maltose binding protein antibody (7G4) was from Sigma. cDNAs encoding full-length or truncated human PIPKIγ (Ling et al., 2002), PLK4 and CEP152 were obtained by PCR and subcloned into pCMV–HA, pcDNA3–FLAG, pGEX–4T-1, pET-28, pET-42 or pMAL–C2X constructs. pcDNA3-Myc–PLK4 was obtained from Addgene. HA–PIPKIγ kinase dead (KD) and RNAi-resistant HA–PIPKIγ-WT or –KD, which contain mismatches to the HsPIPKIγ-O1 siRNA, were generated using QuikChange Site-Directed Mutagenesis Kit (Agilent Technologies). All constructs were verified by sequencing.

Lentiviruses carrying shRNA were constructed by cloning shRNA oligonucleotides (Invitrogen) into the pLKO.1 vector (AddGene). The shRNA sequence targeting human PIPKIγ was 5′-GTCGTGGTCATGAACAACA-3′. An interfering RNA sequence targeting luciferase (5′-GTACCTGTACTTCATGCAG-3′) was used as negative control. All constructs were confirmed by sequencing. HEK293T cells were co-transfected with pLKO.1–puro carrying shRNA, pCMV–VSVG and pCMV-R8.91 (gifts from Gaoxiang Ge, Shanghai Institute of Biochemistry and Cell Biology, Chinese Academy of Sciences, China) at a ratio of 2∶1∶1 using FuGENE 6 (Promega). The supernatant was collected 48-72 h after transfection. Subconfluent HeLa cells were infected in the presence of 8 µg/ml polybrene (Sigma).

All siRNA oligonucleotides were obtained from Invitrogen. HsPIPKIγ-O1, 5′-GCGTGGTCAAGATGCACCTCAAGTT-3′; HsPIPKIγ-O2, 5′-CCTACAGGTTCATCAAGAAACTGGA-3′; mouse PIPKIγ, 5′-GCGAGAGAGAGGATGTGCAGTATGA-3′; PLK4-O1, 5′-CACTGGTTTGGAAGTTGCAATCAAA-3′; PLK4-O2, 5′-AGGAGGTGTGTGTGGAGCTTGTAAA-3′; HsSAS-6, 5′-AGAAAAGCACGTTAATCAGCTACAA-3′ (Strnad et al., 2007); CEP152-O1, 5′-CAGAACAACTGAAATGGCTCTGGAA-3′; CEP152-O2, 5′-CAGCGTTTGCTGGGTGCAACTCAA-3′; CEP192-O1, 5′-CCCAAAGGAAGACATTTTCATCTCT-3′; CEP192-O2, 5′-ATCAGACAGAGGAATCAATAATAAA-3′ (Gomez-Ferreria et al., 2007; Sonnen et al., 2013; Zhu et al., 2008); HsPIPKIα-O1, 5′-TTGAAAGGTGCCATCCAGTTAGGCA-3′; HsPIPKIα-O2, 5′-TGTTAAGAAGTTGGAGCACTCTTGG-3′ (Mellman et al., 2008); HsPIPKIβ, 5′-CAGCAAAGGGTTACCTTCCAGTTCA-3′. Stealth RNAi Negative Control siRNA (Invitrogen) was used as negative control. Similar results were obtained by using both siRNA oligonucleotides (O1, O2) for depletion of HsPIPKIγ, PLK4, CEP152, CEP192 or HsPIPKIα, respectively. Unless stated otherwise, results for oligonucleotide 1 (O1) are shown.

Immunoprecipitation was performed as previously described (Ling et al., 2002) using immunoprecipitation (IP) buffer [20 mM HEPES-KOH (pH 7.2) 10 mM KCl, 1.5 mM MgCl₂, 1 mM EDTA, 1 mM EGTA, 150 mM NaCl, 0.5% NP-40, Complete Protease Inhibitor Cocktail (Roche), PhosSTOP Phosphatase Inhibitor Cocktail (Roche) following the manufacturer's standard protocols]. For indirect immunofluorescence, cells were grown on glass coverslips, fixed with ice-cold methanol for 10 min, permeabilized and stained as described previously (Ling et al., 2002). Fluorescence images were acquired using a Nikon TE 2000-U with Metamorph (Molecular Devices). Z-Series were taken at 0.1-µm steps. Three-dimensional structured illumination microscopy (3D-SIM) was performed on a ELYRA Superresolution Microscopy system (Zeiss) equipped with an alpha ‘Plan-Apochromat’ ×100/1.46 digital interference contrast (DIC) oil immersion objective and Andor iXon 885 EMCCD camera, following standard protocol. Sections were acquired at 0.125 mm z-steps. Color channels were aligned using the alignment parameter from control measurements with 0.5 µm diameter multispectral fluorescent beads (Zeiss). Structured illumination reconstruction and image processing was performed with the ZEN software package (Zeiss). Final image processing was performed using Adobe Photoshop.

GST or MBP-fused proteins were incubated with His-tagged proteins and glutathione-sepharose beads (GE Healthcare) or amylose resin (New England Biolabs) in binding buffer [25 mM Tris (pH 7.6), 150 mM NaCl, 1 mM dithiothreitol (DTT), 0.5% Triton X-100, 10% glycerol] for 2 h at 4°C. The precipitates were washed five times with binding buffer before being analyzed by immunoblotting. The CEP152 N-terminus (amino acids 1-748) protein was generated using the TNT Quick Coupled Transcription/Translation System (Promega), in the presence of [³⁵S]methionine (Perkin-Elmer), used for pull-down assay and analyzed by autoradiography.

The cells were fixed for 1 h using Trump's fixative (Electron Microscopy Sciences). Embedding and serial sectioning of the cell samples was performed according to standard procedures at the Electron Microscopy Core Facility, Mayo Clinic. Specimens were observed in a JEOL 1400 transmission electron microscope (JEOL) operating at 80 kV.

PLK4 activity was determined following a previously published approach (Hatch et al., 2010; Holland et al., 2010). Briefly, a FLAG-tagged CEP152 fragment (CEP152^1-748) was overexpressed in HeLa cells and immunoprecipitated with anti-FLAG antibody. 0.2 µg of purified MBP–PLK4 was incubated with or without recombinant CEP152 N-terminus and/or an appropriate amount of wild-type (WT) or kinase-dead (KD) His–PIPKIγ in kinase buffer with 5 µCi [³²P]ATP (Perkin Elmer) and 33 µM ATP (New England Biolabs) for 30 min at room temperature. ImageJ (NIH, Bethesda, MD) was used to quantify autoradiographs.

Centrosomes were isolated from HeLa cells or mouse kidneys, based on a published approach (Meigs and Kaplan, 2008).

Data represent the mean and standard deviation (SD) from at least three independent experiments. Significance was calculated by Student's t-test using Excel software (Microsoft). A P-value of <0.05 was considered statistically significant.

We sincerely thank Bing Huang and Trace Christensen (Electron Microscopy Core Facility, Mayo Clinic) for technical assistance with electron microscope studies, Gaoxiang Ge (Shanghai Institute of Biochemistry and Cell Biology, Chinese Academy of Sciences) for kindly sharing the lentivirus system, and Jason L. Bakeberg and Cynthia Hommerding at Mayo Clinic for technical assistance with ultra-speed centrifugation and fluorescence microscopy. We thank Gina Razidlo, Barbara Schroeder and Mark McNiven for their stimulating discussion.