Gli2 modulates cell cycle re-entry through autophagy-mediated regulation of the length of primary cilia

The primary cilium, consisting of a microtubule-based axoneme, is an antenna-like organelle projecting from a specialized centrosomal structure, the basal body. Although the primary cilium has long been thought to be a vestigial cellular structure since its discovery in 1898 by Zimmerman, its crucial roles in mediating various signaling pathways have emerged in the last decade (Wheatley, 2005), including a role in mediating sonic hedgehog (SHH) signaling. When SHH binds to its receptor Patched-1 (Ptch1), Smoothened (Smo) is released from inhibition and allowed to enter the primary cilium, leading to the activation of Gli transcription factors (Haycraft, 2005). This process is required for many biological processes, such as cell cycle progression, embryonic development, metabolism, tissue homeostasis and tumorigenesis (Tran et al., 2014; Varjosalo and Taipale, 2008). Therefore, disruption of ciliogenesis leads to multiple dysfunction diseases, such as Joubert syndrome, Bardet–Biedl syndrome and Meckel syndrome (Craige et al., 2010; Goetz and Anderson, 2010; Lee and Gleeson, 2011; Waters and Beales, 2011; Williams et al., 2011).

Ciliogenesis is a complex process, beginning with docking of distal appendage proteins to the mother centriole, followed by fusion of ciliary vesicles, formation of transition zone and the extension of microtubule-based axoneme (Graser et al., 2007; Lu et al., 2015; Ott et al., 2018; Szymanska and Johnson, 2012; Tsang et al., 2008; Williams et al., 2011). In fibroblasts, these processes can be initiated by serum starvation, which forces the cells to enter a quiescent (G0) stage. Interestingly, serum starvation not only induces ciliogenesis, but also activates an acute autophagic response (Shang et al., 2011). It has been shown in recent studies that starvation-induced autophagy is required for ciliogenesis; in this process autophagy-related proteins (ATGs) are relocated toward the primary cilium and Ofd1, a key modulator for ciliogenesis and ciliary length, is removed from the centriolar satellites (Pampliega et al., 2013; Tang et al., 2013). In detail, if Hedgehog signaling is enhanced by applying a Smo agonist or overexpressing Gli1, autophagy is increased in a primary cilium-dependent manner (Pampliega et al., 2013). However, another report has shown that Hedgehog signaling impairs the formation of autophagosomes, which relies on the transcription factor Gli2 (Jimenez-Sanchez et al., 2012). Hence, the relationships between Hedgehog signaling, autophagy and ciliogenesis remain elusive.

As Gli2 mediates Hedgehog signaling-induced autophagy, which is important for the induction of ciliogenesis, we postulated that Gli2 may participate in the regulation of ciliogenesis. Thereafter, we utilized CRISPR/Cas9 technology to establish a Gli2^−/− cell line for examining the effects on ciliogenesis. Surprisingly, we found that Gli2^−/− cells possessed longer primary cilia after serum starvation. Notably, Gli2 depletion enhanced autophagy activity and reduced the Ofd1 protein level. Interestingly, this lengthening of the primary cilium delayed cell cycle re-entry after serum add-back and can be reversed by knocking down Kif3a, a kinesin-II motor protein required for ciliogenesis. Our results demonstrate a novel role for Gli2 that links ciliogenesis, autophagy and the cell cycle.

To generate a cell model to study the relationship between Gli2 and primary cilia, we established Gli2^−/− cell lines by utilizing CRISPR/Cas9 technology (Doudna and Charpentier, 2014). We constructed three plasmids expressing small guide RNAs (sgRNA) targeting exon 2 of the Gli2 gene (Gli2sgRNA#1, Gli2sgRNA#2 and Gli2sgRNA#3; Table S1). When these plasmids were each co-transfected with Cas9 cDNA, they reduced Gli2 protein level in NIH3T3 cells, with Gli2sgRNA#1 showing the highest efficiency (Fig. S1A). Thus, we used Gli2sgRNA#1 to create Gli2-knockout cells and identified four candidate clones (#2-8, #2-10, #3-11, and #4-27) that showed alterations in DNA sequences in exon2 (Fig. S1B). Among them, clone #4-27 represented the lowest query cover (10%, Table S2) and lowest Gli activity, as determined with a luciferase reporter assay (Fig. S1C). By utilizing TA cloning to separate each allele in clone #4-27, we found that one Gli2 allele has a deletion of 11 bp (six colonies) while the other has a 5 bp deletion (four colonies) at the sgRNA-targeted sites of exon 2 (Fig. 1A), leading to frameshift mutations. Western blot showed that Gli2 expression in these cells was undetectable (Fig. 1B). To confirm the specificity of Gli2 knockout in clone #4-27, we transfected Gli2 cDNA and found that the Gli2 protein expression and the Gli-luciferase activity can be rescued (Fig. 1B,C). Thus, we established a Gli2-knockout NIH3T3 cell line (termed NIH3T3^Gli2−/−) for our study.

Next, we examined the effects of Gli2 knockout on primary cilia by immunostaining the cells for the ciliary marker ADP-ribosylation factor-like protein 13B (Arl13b). First, we found that a comparable percentage of NIH3T3^WT cells and NIH3T3^Gli2−/− cells formed primary cilia after serum starvation for 24 h (Fig. S2A). Interestingly, when we measured the length of each primary cilium, the primary cilia of NIH3T3^Gli2−/− cells were ∼40% longer than those of wild-type NIH3T3 cells (NIH3T3^WT) (NIH3T3^WT=2.5±0.5 μm; NIH3T3^Gli2−/−=3.5±0.7 μm; mean±s.d.; Fig. 2A). This effect was reversed by expression of GFP–Gli2 in the NIH3T3^Gli2−/− cells (NIH3T3^Gli2−/−+Gli2-GFP=2.0±0.6 μm; Fig. 2A), demonstrating that the change of ciliary length was specific to Gli2 manipulations.

To further confirm this result, we used acetylated α-tubulin as a second ciliary marker and showed longer primary cilia in NIH3T3^Gli2−/− cells after serum starvation (NIH3T3^WT=1.9±0.4 μm; NIH3T3^Gli2−/−=2.5±0.4 μm; Fig. 2B). In addition, we separately applied two short hairpin RNAs (shRNAs) targeting the Gli2 gene and reduced the Gli2 protein level (Fig. S2B). Cells labeled with Arl13b–GFP again showed longer primary cilia when Gli2 was knocked down (sh-Gli2), whereas the primary cilia were moderately shorter when Gli2 was overexpressed (Gli2-OX) (control=3.7±0.2 μm; Gli2-OX=2.6±0.1 μm; sh-Gli2#1=5.1±0.3 μm; sh-Gli2#2=4.3±0.3 μm; Fig. S2C).

To trace the time course of ciliogenesis in both NIH3T3^WT and NIH3T3^Gli2−/− cells, we measured the ciliary length from the beginning (0 h after serum starvation, SS 0 h) to an extended period (48 h after serum starvation, SS 48 h). Although the primary cilia of these two cell lines before serum starvation were both short, NIH3T3^Gli2−/− cells exhibited longer primary cilia even after an initial 4 h of serum starvation. This effect persisted throughout the 48 h of serum starvation (Fig. 2C). Interestingly, both NIH3T3^WT and NIH3T3^Gli2−/− cells showed a similar percentage of ciliated cells throughout the entire course of ciliogenesis even though there was a slight increase in ciliated NIH3T3^Gli2−/− cells after 4 h of serum starvation (Fig. S2A). Taken together, these results indicate that manipulation of Gli2 level alters the length of primary cilia after serum deprivation.

Previously, serum starvation has been shown to not only lead to cell cycle exit, but also triggers autophagy-mediated Ofd1 removal, which is required for ciliary growth (Tang et al., 2013). Hence, we tested whether NIH3T3^Gli2−/− cells might have longer primary cilia after serum starvation because of an upregulation of autophagy. We first followed autophagy activities over time after serum starvation by measuring the protein level of the autophagosome marker, the lipdated form of LC3 proteins (also known as MAP1LC3), denoted LC3-II, and the degradation of the autophagy substrate p62 (also known as sequestosome 1, SQSTM1). We found that the protein level of LC3-II was significantly higher in NIH3T3^Gli2−/− cells than in NIH3T3^WT cells. Consistent with this, NIH3T3^Gli2−/− cells also exhibited faster p62 degradation (Fig. 3A), suggesting that the autophagy activity is raised when Gli2 is depleted. When treated with the lysosome inhibitor chloroquine (CQ), NIH3T3^Gli2−/− cells still exhibited a higher LC3-II level compared to wild-type cells 4 h after serum starvation (Fig. S3A), suggesting a higher level of autophagy.

We then followed Ofd1 protein level during the whole course of serum starvation. Similar to the course of p62 protein degradation (Sahani et al., 2014), we found that the Ofd1 protein level increased during the first 4–12 h and then decreased at the later stage of serum starvation in wild-type cells (12–48 h, Fig. 3A). Interestingly, the protein level of Ofd1 in NIH3T3^Gli2−/− cells was reduced within the first 4 h after serum starvation (Fig. S3B), and then underwent faster degradation at the later stage (12–48 h; Fig. 3A). To examine whether this decreased Ofd1 level reflects a reduction of Ofd1 at the centriolar satellites, we measured Ofd1 intensity specifically around the centrioles by immunostaining 24 h after serum starvation. Consistent with the above results, the Ofd1 intensity surrounding the centrioles was lower in NIH3T3^Gli2−/− cells (Fig. S3C), indicating that more Ofd1 is removed from the centriolar satellites in Gli2-depleted cells after serum deprivation for 24 h. Importantly, the increase in autophagy and decrease in Ofd1 in NIH3T3^Gli2−/− cells were not observed in serum-rich medium (Fig. S4A), as is also reflected by no difference in the percentage of ciliated cells (Fig. S4B) and ciliary length (Fig. 2C, time zero), indicating that this process is specific to serum starvation-induced ciliogenesis.

The increase in LC3-II protein level may reflect higher autophagic flux or a block of autophagosome–lysosome fusion. To dissect these possibilities, we transfected plasmids encoding RFP–GFP–LC3 in NIH3T3^WT and NIH3T3^Gli2−/− cells to detect autophagosome–lysosome fusion. Upon the onset of autophagy, RFP–GFP–LC3, which appears yellow, is recruited to autophagosomes. As autophagosomes fuse with lysosomes, GFP fluorescence is quenched under the acidic environment, and thus autolysosomes appear red (Orhon et al., 2016). Interestingly, both the numbers of autophagosomes (yellow) and autolysosomes (red) were higher in NIH3T3^Gli2−/− cells than that in NIH3T3^WT cells (Fig. 3B,C). Note that the ratio between autolysosome and autophagosomes was not changed in NIH3T3^Gli2−/− cells (Fig. 3D), indicating that the elevation of LC3-II in NIH3T3^Gli2−/− cells was due to an escalation of autophagic flux and not a blockade between autophagosomes and lysosome fusion. Taken together, we found that the autophagy flux is highly activated in NIH3T3^Gli2−/− cells upon serum starvation while basal autophagy is not significantly disturbed.

To investigate whether autophagy is responsible for the increase in ciliary length in NIH3T3^Gli2−/− cells, we treated NIH3T3^WT and NIH3T3^Gli2−/− cells with 3-methyladenine (3-MA), an autophagy inhibitor, during the first 4 h of serum deprivation. Indeed, 3-MA significantly decreased the LC3-II protein levels in both NIH3T3^WT and NIH3T3^Gli2−/− cells (Fig. 4A). Consistent with this, the protein level of Ofd1 was recovered in NIH3T3^Gli2−/− cells treated with 3-MA (Fig. S5A), indicating that 3-MA successfully blocked the autophagy-dependent Ofd1 removal. Remarkably, we found that while Gli2 knockout increased the ciliary length, 3-MA treatment brought the length back to normal in NIH3T3^Gli2−/− cells (NIH3T3^WT+DMSO=1.87±0.04 μm; NIH3T3^Gli2−/−+DMSO=2.48±0.05 μm; NIH3T3^WT+3-MA=1.49±0.06 μm; NIH3T3^Gli2−/−+3-MA=1.95±0.06 μm; Fig. 4B), indicating that the elevated autophagic flux during ciliogenesis plays a fundamental role for ciliary elongation in NIH3T3^Gli2−/− cells. To further test this possibility, we used shRNA targeting Atg3 along with a GFP reporter (sh-Atg3-GFP), an ATG protein crucial for autophagosome formation, as an additional tool to block autophagy (Mizushima and Komatsu, 2011). Cells infected with lentiviruses encoding sh-Atg3-GFP had significantly less LC3-II protein compared to the control group (Fig. 4C). Similarly, the primary cilia in NIH3T3^Gli2−/− cells had a normal length after sh-Atg3-GFP transfection (NIH3T3^WT+sh-Ctrl-GFP=2.07±0.04 μm; NIH3T3^Gli2−/−+sh-Ctrl-GFP=2.77±0.04 μm; NIH3T3^WT+sh-Atg3-GFP=1.22±0.03 μm; NIH3T3^Gli2−/−+sh-Atg3-GFP=1.78±0.03 μm; Fig. 4D).

To investigate whether autophagy-dependent Ofd1 removal is required downstream of Gli2 depletion for the ciliary lengthening, we transfected NIH3T3^WT and NIH3T3^Gli2−/− cells with Myc-tagged Ofd1 (pIRES2-Ofd1-Myc) or its control empty vector (pIRES2). After serum starvation for 24 h, Ofd1 overexpression reduced the percentage of ciliated cells as well as the ciliary length in NIH3T3^WT cells (Fig. S5B–D), consistent with previous findings (Tang et al., 2013). In NIH3T3^Gli2−/− cells, forced expression of Ofd1 brought the ciliary length back to normal (Fig. S5B,D). Taken together, these results suggest that the activation of autophagy-dependent Ofd1 removal is crucial for the ciliary lengthening in Gli2-knockout cells.

Typically, primary cilia form at the G0/G1 phase of the cell cycle and undergo disassembly when cells re-enter the cell cycle (Kim and Tsiokas, 2011). In previous reports, a longer primary cilium was shown to be associated with the delay in cell cycle re-entry (Kim et al., 2011). Therefore, we tested whether the elongated primary cilium delayed the process of ciliary resorption and cell cycle re-entry in Gli2-knockout cells by immunostaining and flow cytometry.

NIH3T3^WT and NIH3T3^Gli2−/− cells were synchronized at the G0 phase after serum starvation for 24 h. By immunostaining for markers labeling the primary cilium (Arl13b) and centrioles (γ-tubulin), we showed that, in NIH3T3^WT cells, the primary cilia were absorbed over time after serum re-stimulation, exhibiting a two-phase decrease both in ciliary length (Fig. 5A) and in the percentage of ciliated cells (Fig. 5B) as previously described (Pugacheva et al., 2007). However, Gli2 depletion slowed down this process, resulting in longer primary cilia and higher percentage of ciliated cells during ciliary resorption phase compared to wild-type cells (Fig. 5A,B).

Finally, by using the flow cytometry to decipher the phases of cell cycle, we found that NIH3T3^WT cells progressed to S-G2-M phase over time after serum was added back (Fig. 6B,C). However, the percentage of S-G2-M phase in NIH3T3^Gli2−/− cells did not increase as much as NIH3T3^WT cells after serum re-stimulation even for 24 h (NIH3T3^WT=53.9±1.3%; NIH3T3^Gli2−/−=39.2±1.0%; mean±s.e.m.; Fig. 6B,C), showing that there was a significant delay in cell cycle re-entry. As NIH3T3^WT and NIH3T3^Gli2−/− cells possess similar distribution of cells in each phase of the cell cycle under serum-rich conditions (Fig. S6A), we postulated that the delay in cell cycle re-entry in NIH3T3^Gli2−/− cells could be caused by the delay in ciliary resorption resulting from an lengthened primary cilium (Fig. 5A). To examine this hypothesis, we disrupted primary cilia by knocking down Kif3a, a kinesis II motor protein required for ciliogenesis. By immunostaining, we showed that applying the sh-Kif3a-GFP construct was able to eliminate primary cilia in both NIH3T3^WT and NIH3T3^Gli2−/− cells despite serum starvation for 24 h (Fig. 6A), consistent with the results in our previous report (Chen et al., 2018). Interestingly, ablating the primary cilium by means of sh-Kif3a-GFP slightly enhanced the cell cycle re-entry in NIH3T3^WT cells, consistent with previous findings seen upon Ift88 knockdown (Kim et al., 2011). To our surprise, sh-Kif3a-GFP restores the delay of cell cycle re-entry in NIH3T3^Gli2−/− cells after serum re-stimulation for 16 and 24 h (Fig. 6B,C). Taken together, these results demonstrate that the elongated primary cilium is responsible for the delay in cell cycle re-entry in NIH3T3^Gli2−/− cells.

In this study, we established the first Gli2-knockout fibroblast cell line using CRISPR/Cas9 technology (Fig. 1), which has become popular for cilia studies in different organisms (Breslow et al., 2018; Katoh et al., 2017; Li et al., 2015; Pusapati et al., 2018). We showed that both depletion and knockout of Gli2 expression increased primary cilia length after serum starvation (Fig. 2; Fig. S2C). Mechanistically, we found that the autophagy flux was increased in association with a decrease in Ofd1 in NIH3T3^Gli2−/− cells after serum starvation (Fig. 3; Fig. S3). Interestingly, pharmacological and genetic inhibition of autophagy, through 3-MA and sh-Atg3-GFP, respectively, rescued the Ofd1 protein level (Fig. S5) and alleviated the effect of growing longer primary cilia in NIH3T3^Gli2−/− cells (Fig. 4). In addition, NIH3T3^Gli2−/− cells showed a slower ciliary resorption (Fig. 5) and cell cycle re-entry after serum add-back; inhibition of ciliogenesis by means of Kif3a knockdown reversed this effect (Fig. 6). These results indicate that Gli2 regulates autophagy-mediated Ofd1 removal to control primary cilia formation and cell cycle progression in fibroblasts (Fig. 7).

Our work identified Gli2 as a negative regulator for ciliary length, which is controlled through a balance between ciliary assembly and disassembly. Recent studies have shown that ciliary length can be regulated by proteins that mediate ciliary disassembly, including Nek2–Kif24, Aurora kinase A–HDAC6, and Nde1 (Kim et al., 2015, 2011; Maskey et al., 2015; Pugacheva et al., 2007). In addition, several proteins, such as dynein heavy chain-2 (DHC2, also known as DYNC2H1), dynein light chain Tctex-1 (also known as DYNLT1), and the centriolar protein calcineurin-2, are responsible for cilia length regulation (Palmer et al., 2011; Stevenson et al., 2018). On the other hand, several negative regulators for ciliogenesis, such as the APC-Cdc20 and Ofd1, have been reported to modulate the length of primary cilia (Tang et al., 2013; Wang et al., 2014). Upon serum starvation, Ofd1 is removed from centriolar satellites by autophagy via an integration with LC3, which allows the initiation of ciliary growth. Here, we found that Gli2 knockout enhanced the activation of autophagy required for Ofd1 removal during ciliogenesis (Fig. 3; Fig. S3). Importantly, we showed that inhibition of autophagy restored the primary cilia length and Ofd1 protein level in Gli2-knockout cells (Fig. 4; Fig. S5), suggesting that Gli2 regulates ciliary elongation, at least in part, through the autophagy pathway. Interestingly, although the inhibition of autophagy significantly reduced the ciliary length in NIH3T3^Gli2−/− cells, the average ciliary length was still slightly longer than that in control cells treated with autophagy inhibitors (Fig. 4), suggesting that additional mechanisms may participate downstream of Gli2 for regulating ciliary length. Our findings specifically link Gli2-dependent ciliary length control to the autophagy-mediated Ofd1 removal.

In this paper, we found that Gli2 depletion in NIH3T3 fibroblasts enhanced autophagic flux (Fig. 3; Fig. S3), possibly due to the reduction of Gli transcriptional activities as shown in the Gli reporter assay (Fig. 1C). This finding is consistent with previous findings with manipulations of SHH and Gli activities in several other systems. In cancer cells, inhibition of the SHH signaling pathway hinders their proliferation, in part through upregulation of autophagy (Tang et al., 2015). In pancreatic ductal adenocarcinoma cells, activation of SHH signaling inhibits autophagy, whereas pharmacological inhibition of Gli1 and Gli2, induces autophagy (Xu et al., 2014). Interestingly, not only serum starvation but also Gli2 mediates the SHH-dependent regulation of autophagy seen upon rapamycin treatment or trehalose induction (Jimenez-Sanchez et al., 2012). This regulation was likely meditated by the autophagy modulator, PERK (also known as EIF2AK3), an ER-resident protein responsible for the phosphorylation of eukaryotic initiation factor 2 α (EIF2α), leading to the inhibition of protein translation in response to ER stress (Jimenez-Sanchez et al., 2012). Here, we confirmed that the expression of Perk was highly elevated in NIH3T3^Gli2−/− cells upon serum starvation for 24 h by quantitative real-time RT-PCR (RT-qPCR) (Fig. S7A). Accumulating data implies that PERK may negatively regulate the mammalian target of rapamycin (mTOR) activity, leading to the activation of autophagy (Bobrovnikova-Marjon et al., 2012; Qin et al., 2010). Consistent with this, we found that the protein level of phosphorylated S6 ribosomal protein (p-S6) was decreased in NIH3T3^Gli2−/− cells compared to the level in NIH3T3^Gli2−/− cells, whereas the expression of S6 was the same in both cells, suggesting that the mTOR activity is be reduced in NIH3T3^Gli2−/− cells (Fig. S7B). Therefore, our findings of Gli2 regulation of autophagy is well in line with previous studies.

Here, we also showed that Gli2 depletion leads to late cell cycle re-entry while the ablation of primary cilia restores the cell cycle re-entry in NIH3T3^Gli2−/− cells (Fig. 6). Gli2 is known to regulate the cell cycle through the canonical Hh signaling pathway by elevating cyclin D, which is required for cell cycle progression at G0/G1 (Katoh and Katoh, 2009). For decades, it has been well characterized that cyclin D is crucial for G1/S progression by binding to CDK4 and CDK6, two cyclin-dependent kinases which phosphorylate retinoblastoma (Rb) protein, and thus release the inhibition of E2F transcription factors, which are necessary for driving genes required for G1/S transition (Bertoli et al., 2013).

Interestingly, not only the Gli transcription factors, but also the primary cilia are tightly correlated with cell cycle progression. The assembly of the primary cilium usually occurs during G1/G0 phase; after that, the primary cilium rapidly becomes shorter as cells progress from G1 to S phase through the activation of Plk1 and Aurora A kinase activity, indicating that ciliary disassembly coincides with cell cycle re-entry (Fu et al., 2015; Izawa et al., 2015; Kim and Tsiokas, 2011; Pugacheva et al., 2007). The presence of primary cilia, especially a longer primary cilia functions as a brake for cell cycle re-entry (Kim et al., 2011; Li et al., 2011). For example, cells depleted of Nde1 possess longer cilia, which is responsible for the delay in cell cycle re-entry (Kim et al., 2011). In addition, overexpression a phospho-mimetic Tctex-1 (T94E), a negative ciliary regulator recruited to the transition zone during ciliary disassembly, leads to more cells re-entering to S-phase (Li et al., 2011). All these findings suggest that the primary cilium plays a critical role in cell cycle regulation.

Notably, as our experiments were conducted in medium without exogenous SHH, we did not see any significant difference in the phase of cell cycle distribution in NIH3T3^WT cells and NIH3T3^Gli2−/− cells under nutrient-rich conditions (Fig. S6A). However, when serum re-stimulation was performed after cells were synchronized in G0/G1 phase, cells with Gli2 depletion showed a delay in cell cycle re-entry, suggesting that Gli2 plays a specific role in delaying cell cycle re-entry but not cell cycle progression in general (Fig. 6B,C). Importantly, ablation of the primary cilium through Kif3a knockdown restored this process, implying that the primary cilium mediates the delay of cell cycle re-entry in NIH3T3^Gli2−/− cells (Fig. 6A,C). Therefore, in addition to the regulation on cyclin D upon SHH activation, we demonstrate that Gli2 plays a new role in regulating the cell cycle through the primary cilium.

Taken together, our findings suggest that Gli2 can regulate ciliary length through modulating autophagy-mediated Ofd1 degradation. This pathway may in turn facilitate cell cycle re-entry of resting fibroblasts (Fig. 7).

The shRNAs targeting Gli2, Kif3a, and Atg3 were obtained from National RNAi Core Facility in Taiwan. The targeting sequences for RNA interference (RNAi) are sh-Gli2#1: 5′-CACCAACCCTTCAGACTATTA-3′, shGli2#2: 5′-TGTGGAGGACTGCCTACATAT-3′, sh-Kif3a-GFP: 5′-ATATTGGGCCAGCAGATTATA-3′ (Chen et al., 2018), sh-Atg3-GFP: 5′-GTACATCACTTACGACAAATA-3′. pC2-Arl13b-GFP was a kind gift from Dr Olivier Ayrault at Institut Curie, France. pCEFLmGFP-Gli2 and pIRES2-eGFP-Ofd1-myc for Gli2 and Ofd1 overexpression were Addgene plasmid #37672 (deposited by Philip Beachy; Kim et al., 2009), Addgene plasmid #24560 (deposited by Jeremy Reiter; Singla et al., 2010), respectively. pUS2 and pUS2-Gli2 were generous gift from Dr Jenn-Yah Yu (Department of Life Sciences and Institute of Genome Sciences, Taipei City, Taiwan), who also provided the luciferase reporter constructs, 8×3′Gli-BS and 8×3′ mGli-BS (Lin et al., 2012), who received this plasmids from H. Sasaki (Sasaki et al., 1999). The promoter region for the luciferase reporter contains eight copies of wild-type Gli-binding motif (GAACACCCA); a mutant Gli-binding motif (GAAGTGGGA) served as a control.

NIH3T3 cells were obtained from ATCC (CRL-1658) and tested for contamination; thereafter, cells were maintained in Dulbecco's modified Eagle's medium (DMEM; Life Technologies), pH 7.3, supplemented with 10% calf serum, 1% 100 U/ml penicillin and 100 μg/ml streptomycin, and 1% L-glutamine at 37°C in humidified atmosphere with 5% CO₂ (Jheng et al., 2018). Cells were serum starved at the beginning of transfection in order to maintain cells in G0/G1 phase, which induces the growth of primary cilia. In experiments where autophagy was blocked, 3-methyladenine (10 mM, Sigma-Aldrich) was applied as previously described (Orhon et al., 2016). Plasmids were transfected into NIH3T3 cells via Lipofectamine 3000 reagent (Invitrogen) according to the manufacturer's recommended protocol. Primary cilia were observed by confocal microscopy (Zeiss LSM 700) after serum starvation and transfection.

To establish the Gli2-knockout cell line, we used CRISPR/Cas9 technology as previously described (Doudna and Charpentier, 2014). Briefly, we designed three different Gli2 sgRNAs (Gli2sgRNA#1, Gli2sgRNA#2, and Gli2sgRNA#3; Table S1), with the MIT CRISPR design tool (http://crispr.mit.edu/), that targeted the exon 2 region of the Gli2 gene. Next, we annealed two oligonucleotides in each sgRNA pair and extended them to make a 100 bp double-stranded DNA fragment by using Phusion polymerase (NEB). Subsequently, these double-stranded DNA fragments were inserted into the pCR-Blunt II-TOPO vector. NIH3T3 cells were transfected with pCP-Blunt II-TOPO-sgGli2 and Cas9 constructs by Lipofectamine 3000 (Invitrogen), followed by determination of the knockout efficiency through western blotting.

After transfection for 5 days with Gli2sgRNA#1 and Cas9 by Lipofectamone 3000 (Invitrogen), cells were selected and transferred to 96-well plates with 1 cell per well. Once these cells reached 60% confluency, they were sub-cultured into a single well of a 48-well plate. As these cells reached 80% confluence, we sub-cultured the cells into three 60 mm dishes for genomic DNA extraction and protein analysis.

The collected cells were suspended in 500 μl lysis buffer (Goal Bio), followed by addition of 10 μl (20 mg/ml) proteinase K (BIOTOOLS) at 55°C for 1 h. Next, we purified the gDNA by using protein precipitation buffer (Goal Bio) according to the manufacturer's protocol. The primers for sequencing the Gli2 exon 2 are: forward, 5′-TATGAGCCTCTGAGATGGAG-3′ and reverse, 5′-ACAAGAAAGCATCAGAGCAC-3′.

Cells were washed in PBS and lysed in 1× RIPA buffer (50 mM Tris-HCl pH 8.0, 150 mM NaCl, 1% Nonidet P-40, 0.5% sodium deoxycholate, 0.1% SDS) with protease inhibitor cocktail (Sigma Aldrich) and phosphatase inhibitors (Roche). Lysates were sonicated in icy water and cleared by centrifugation (21,000 g for 15 min), and protein concentration was determined by using the Pierce BCA protein assay kit (Thermo Scientific), followed by electrophoresis on a 0.1% SDS, 40% acrylamide/bis-acrylamide gradient gel (TOOLS), and transfer onto a PVDF membrane (Millipore). Primary antibodies used were: rabbit anti-Gli2, 1:500 (Abcam, cat. no. ab167389); rabbit anti-LC3, 1:500 (Proteintech, cat. no. 14600-1-AP); rabbit anti-Ofd1, 1:500 (Abcam, cat. no. ab222837); rabbit anti-P62, 1:1000 (Proteintech, cat. no. 18420-1-AP); rabbit anti-phospho-S6 ribosomal protein, 1:1000 (Cell Signaling Technology, cat. no. #2211); rabbit anti-S6 ribosomal protein, 1:1000 (Cell Signaling Technology, cat. no. #2217); mouse anti-GAPDH, 1:10,000 (Proteintech, cat. no. 60004-1-Ig); mouse anti-β-actin, 1:10,000 (Proteintech, cat. no. 60008-1-Ig), and mouse anti-α-tubulin, 1:5000 (Proteintech, cat. no. 66031-1-Ig). Next, horseradish peroxidase (HRP)-conjugated secondary antibodies were applied as below: anti-rabbit-IgG (1:10,000, Sigma Aldrich), and anti-mouse-IgG (1:10,000, GeneTex). Signals were generated with ECL-Plus reagent (Millipore), and detected under the Luminescence/Fluorescence Imaging System LAS-4000 (Fujifilm). Signal quantification was performed with Image-J software (NIH).

NIH3T3^WT and NIH3T3^Gli2−/− cells in 24-well plates were transfected with 0.025 μg US2-renilla Luciferase, and 0.3875 μg firefly luciferase with either normal or mutant Gli-binding sites as the promoters. Gli2 overexpression was achieved by transfecting 0.3875 μg of the pLKO-Gli2 construct whereas 0.3875 μg PLKO construct was used as control. The dual luciferase assay (Promega) was performed according to the manufacturer's protocol. The relative luciferase intensity is calculated as previously described (Lin et al., 2012): (GliBS firefly luciferase/Renilla luciferase in the same cells)/(mutant GliBS firefly/Renilla luciferase in the same cells).

NIH3T3^WT and NIH3T3^Gli2−/− cells were plated on glass coverslips coated with 0.2 mg/ml poly-D-lysine (Sigma Aldrich) and ciliogenesis was induced by serum starvation for 24 h. The cells were then fixed for 15 min in 4% paraformaldehyde (PFA) at room temperate followed with cold methanol/acetone (1:1) for 5 min, washed in PBS and incubated in PBS with 0.2% Triton X-100 (PBST) for 15 min. The cells were blocked in 10% normal goat serum (NGS) and 5% bovine serum albumin (BSA) in PBS for 1 h at room temperature and were incubated overnight with primary antibody at 4°C. Primary antibodies were used at the following concentrations: mouse anti-Arl13b, 1:500 (NeuroMab, cat. no. 73-287); mouse anti-acetylated α-tubulin, 1:500 (Abcam, cat. no. ab24610); and rabbit anti-γ-tubulin, 1:1000 (Abcam, cat. no. ab11317); rabbit anti-Pcm1, 1:500 (Bethyl Laboratories, cat. no. A301-149A); rabbit anti-Ofd1, 1:500 (Abcam, cat. no. ab222837); rabbit anti-Myc tag (Proteintech, cat. no. 16286-1-AP). The cells were washed in PBS and incubated in secondary antibodies (conjugated to Alexa Fluor 488, 546, or 647, 1:500 for each antibody) for 2 h at room temperature. Finally, the coverslips were counterstained with 0.5 μg/ml DAPI (Invitrogen). VECTASHIELD^® Mounting Medium Media (Vector Laboratories) was added before sealing the slides.

Flow cytometry (BD FACSCanto) was applied to observe cell cycle distribution of NIH3T3^WT and NIH3T3^Gli2−/− cells. For analysis of the DNA content, cells were fixed with 0.5% formaldehyde and stored at −20°C for no longer than 24 h. After fixation, cells were stained with 20 μg/ml propidium iodide (PI, Sigma Aldrich) plus 200 μg/ml RNase A (Sigma Aldrich) in PBS and subjected to flow cytometry analysis. At least 10,000 cells were analyzed per sample.

Total RNA extraction from NIH3T3 cells was performed with Quick-RNA™ MiniPrep (ZYMO RESEARCH) according to manufacturer's protocol. Removal of genomic DNA was achieved through DNase I treatment, as provided in the kit, and cDNA was synthesized by using the iScript™ cDNA Synthesis Kit (BIO-RAD) following the provider's instructions. RT-qPCR was performed with a OmicsGreen qPCR Master Mix (Omics Bio). Primers used in RT-qPCR for Perk expression were: forward: 5′-CACAGGGACCTCAAGCCTTC-3′; reverse: 5′-GTCCCTACTTGTCCCGTGTG-3′.

Glass coverslips were imaged under an inverted laser scanning confocal microscope (LSM-700, Zeiss). The excitation wavelengths were 647 nm for infrared, 546 for red fluorescence, 488 nm for EGFP, and 405 nm for DAPI. The Z-stack section was set for 0.1 μm for observation of primary cilia. Analysis of images were performed with Zen software (Zeiss) or ImageJ software (NIH).

The measurement of the LC3 puncta (Fig. 3B) was performed as previously described (Dupont et al., 2014; Yoshii and Mizushima, 2017). In brief, cells were transfected with plasmids encoding RFP–GFP–LC3 under basal (serum-rich) as well as serum-free conditions. Images were analyzed by using the ITCN plugin in ImageJ software. Individual puncta were measured and defined as one dot according to the instructions from Academic Technology at Keene State College (https://www.youtube.com/watch?v=PqHFsmS1_JY). To measure the density of the puncta, the number of dots were normalized to the total area in each cell. Next, the density of LC3 puncta upon serum starvation was further normalized to that under the basal condition for each group.

For measurement of Ofd1 intensity around the centrioles (related to Fig. S4C), both NIH3T3^WT and NIH3T3^Gli2−/− cells stained with Ofd1 and γ-tubulin were imaged under the same criteria, followed by analysis using ImageJ. Centered on the γ-tubulin signal, we measured the signal intensity of Ofd1 within a circular area of 34.9 μm². Next, the signal intensity of Ofd1 in NIH3T3^WT cells was normalized to 1, and then the signal intensity of Ofd1 in NIH3T3^Gli2−/− cells was calculated relative to that in NIH3T3^WT cells.

Statistical significance was determined with Excel and SPSS software. Two-tailed tests were performed and P<0.05 was considered as statistical significant. Data are shown as mean±s.e.m. or s.d. as indicated in figure legends. Bar graphs were drawn with Excel (Microsoft) and Prism (GraphPad) software. When the sample number was less than five, bar graphs along with individual data point are shown.

We thank Dr Feng Zhang at MIT for providing a platform for optimized CRISPR designs. We also thank Pei-Jun Hsen, and Yung-Yu Lu for confocal imaging assistance. We especially thank Dr Olivier Ayrault (pC2-Arl13b-GFP), Jenn-Yah Yu (Gli-luciferase vectors) and the National RNAi core of Taiwan for providing plasmids.