A Transposon Screen Identifies Loss of Primary Cilia as a Mechanism of Resistance to SMO Inhibitors

Aberrant Hedgehog (HH) signaling is implicated in many cancers (1) and is particularly critical in medulloblastoma, the most common malignant brain tumor in children, and in basal cell carcinoma (BCC), the most common human cancer overall (2–4). Hyperactivity of the HH pathway is frequently caused by inactivating mutations in Patched (Ptch), which encodes the receptor for HH ligands (5–7). In the absence of Ptch function, Smoothened (SMO), a G-protein-coupled receptor–like molecule, traffics into the primary cilia, a distinctive, microtubule-based signaling organelle. SMO functions within the primary cilia to inhibit the negative regulators Suppressor of fused (SUFU) and protein kinase A (PKA), triggering a signaling cascade that culminates in the nuclear import of GLI transcription factors, which activates an HH transcriptional program that drives proliferation and tumor growth.

The importance of the HH pathway in human cancers has stimulated great interest in developing targeted therapy that antagonizes HH signaling (1, 8). These efforts have resulted in the approval of SMO inhibitors as anticancer agents. In 2012 and 2015 respectively, vismodegib and sonidegib (NVP-LDE225) became the first FDA-approved SMO inhibitors for anticancer treatment in advanced BCC. Several clinical trials in medulloblastoma are in progress. Despite the initial success of SMO inhibitors in mouse models and subsequently in patients, long-term efficacy is limited by the emergence of drug resistance (9–13). Thus far, preclinical and clinical studies have uncovered a number of point mutations in SMO that confer resistance. Although most mutations located in the drug-binding pocket of SMO directly impair drug binding, an additional group of mutations distally located to drug-binding sites may have allosteric effects for drug binding or promote a constitutively activated state of SMO that is less sensitive to drug inhibition (9, 10). Alterations that activate HH signaling downstream of SMO, such as loss of SUFU or amplification of GLI2, as well as mutations that activate intersecting oncogenic pathways such as RAS/MAPK, PI3K, or aPKC, have also been reported (14–16). However, the scope of resistance mechanisms to SMO inhibitors and the process by which HH pathway–dependent tumors evolve and recur following therapy are not yet known.

To identify resistance mechanisms, we used a genome-wide transposon mutagenesis screen in HH pathway–dependent medulloblastoma cells and identified recurrent mutations in the ciliogenesis gene Oral-facial-digital syndrome 1 (Ofd1), in cells resistant to Smo inhibitors. We demonstrate that resistant tumors that lack primary cilia enter a “persister state” of slow, GLI2-dependent growth and show that this mechanism of therapeutic resistance occurs in patients.

To identify genetic changes capable of circumventing SMO inhibition, we launched a genome-wide transposon mutagenesis screen in SMB21 cells, a murine cell line derived from a Ptch^−/− medulloblastoma (16). SMB21 retains key characteristics of SHH-subgroup medulloblastoma, is dependent on HH signaling for proliferation, and is exquisitely sensitive to SMO inhibition (16). The engineered piggyBac transposon system utilized in this screen relies on a transposon vector that can either enhance or disrupt gene expression in the vicinity of the insertion site (Fig. 1A). To deploy the transposon-mediated screen, SMB21 parental cells were transfected with both piggyBac transposon and transposase vectors for mutagenesis, and were then selected for clones of cells able to grow in the presence of 1 μmol/L sonidegib (Fig. 1B). From 30 million cells, 29 resistant clones were isolated and 27 clones were successfully propagated. All resistant clones showed robust resistance to sonidegib in subsequent multidose survival assays (Supplementary Fig. S1A). A barcoded splinkerette-PCR approach coupled with parallel sequencing was utilized to identify transposon insertion loci in each clone (Supplementary Fig. S1B). A total of 182 transposon insertions were identified (Supplementary Fig. S1C and Supplementary Table S1). The two genes with the greatest number of recurrent transposon insertion sites were Sufu (13 clones) and Ofd1 (3 clones; Fig. 1C and D). The robust resistance effects in Sufu- and Ofd1-mutant cells were evident by the dramatic shift of the growth-inhibitory concentration of sonidegib (Fig. 1E) and were further validated using additional SMO inhibitors, vismodegib and cyclopamine (Supplementary Fig. S1D).

We next evaluated the consequence of transposon insertions in Sufu and Ofd1. Although most of the transposons landed in introns and inserted in single alleles (Fig. 1C and D), they effectively disrupted gene transcription and eliminated SUFU or OFD1 protein levels in mutant cells (Fig. 1F and G; Supplementary Fig. S1E and S1F). As Ofd1 is on the X chromosome and the SMB21 parental line was derived from a tumor of a Ptch male, disrupting the single allele of Ofd1 is sufficient to eliminate all gene products (Fig. 1G). Intriguingly, complete loss of the SUFU full-length protein was observed in nearly all Sufu-mutant clones. Genomic copy-number analysis of the Sufu locus revealed loss of the second copy of the Sufu gene in 10 of 13 resistant clones with Sufu mutations (Supplementary Fig. S2). This finding indicates that a second genetic alteration event occurred in the Sufu locus in a subset of mutant cells, thus causing a complete loss of SUFU protein in these cells. Both Sufu- and Ofd1-mutant cells displayed persistent activation of HH signaling in the presence of SMO inhibitors, as demonstrated by expression of GLI1, a typical readout of HH signaling (Fig. 1H). These results suggest that mutations in Sufu and Ofd1 confer resistance to SMO inhibition through reactivation of HH signaling output.

Identification of recurrent insertions in Sufu provides a good validation of the screen, as SUFU is a well-known negative regulator of the HH pathway that functions by binding and sequestering GLI transcription factors, and mutations in Sufu have been identified in clinical tumors resistant to SMO inhibitors (9–11, 17). To establish a causal relationship between mutations in Ofd1 and the resistant phenotype, we first conducted an Flp/FRT-based rescue experiment (Fig. 2A). As the transposon is engineered with two FRT sites flanking the DNA cargo, lentivirus expressing Flp was used to remove the DNA cargo. In clones where the transposon is inserted in introns, such as R25, this approach successfully re-expresses OFD1 and restores sensitivity to sonidegib, as assessed by survival assays and by GLI1 expression (Fig. 2B–D). However, this strategy cannot revert disruptions in genes if transposons are inserted in exons, such as in clones R14 and R16, because the remaining transposon terminal repeat sequence outside of the FRT sites still causes truncated mRNA (Supplementary Fig. S3A and S3B). In a second rescue experiment, we asked whether expression of a full-length OFD1 rescues the resistant phenotype in Ofd1 mutants. Indeed, expression of a full-length OFD1 restored sensitivity to sonidegib in HH signaling and in cell viability assays (Fig. 2E and F). To independently demonstrate that loss of OFD1 mediates resistance to SMO inhibition, we specifically targeted the Ofd1 locus in parental cells using lentiviral-CRISPR/Cas9. Depletion of OFD1 confers resistance to SMO inhibition, demonstrating a causal relationship between loss of OFD1 and drug resistance (Fig. 2G and H).

Ofd1 is an X-linked ciliopathy gene essential for primary cilium formation (18), and mutations in Ofd1 cause oral-facial-digital syndrome type 1 (19). Immunostaining and electron microscopy demonstrated that primary cilia are absent in Ofd1-mutant medulloblastoma cells (Fig. 3A–C). To determine whether disruption of primary cilia per se confers resistance to SMO inhibition, we tested multiple genes essential for ciliogenesis including Ift88, encoding an intraflagellar transport protein, and Kif3a, encoding a subunit of the Kinesin-2 motor complex. Either shRNA-mediated depletion of IFT88 or CRISPR/Cas9-mediated depletion of KIF3A in SMB21 parental cells confers resistance to SMO inhibitors (Fig. 3D and E; Supplementary Fig. S3C and S3D). Similar to Ofd1-mutant cells, cells depleted of IFT88 or KIF3A exhibit persistent low-level activation of HH signaling (Fig. 3D; Supplementary Fig. S3C). Depletion of OFD1 or KIF3A also confers resistance in two additional medulloblastoma lines, SMB55 and SMB56 (Supplementary Fig. S3E and S3F). Given that hundreds of genes are critical for primary cilia formation (20), mutations in ciliary genes are likely to represent a broad new class of therapeutic resistance. To determine whether loss of primary cilia causes drug resistance in a nonspecific manner, we tested targeted and cytotoxic chemotherapy agents, including BKM120 (PI3K inhibitor), BEZ235 (a dual inhibitor for PI3K and mTOR), cisplatin, and vincristine. These agents exhibit similar potency profiles in both parental and cilia mutant cells (Supplementary Fig. S3G), suggesting that loss of primary cilia does not cause a general resistance profile but specifically results in resistance to SMO inhibitors.

The primary cilium is a signal transduction center for HH signaling and is essential for SMO activation and GLI processing (21, 22). In the absence of primary cilia, SMO cannot transduce positive signal to GLI. Nonetheless, Ofd1-mutant cells express HH pathway signature genes as assessed by Gene Set Enrichment Analysis (GSEA) of RNA-sequencing (RNA-seq) data, maintaining an active, albeit limited, HH signaling output (Fig. 4A and B). To ascertain how active HH signaling is maintained in cells lacking cilia, we examined expression and posttranslational processing of GLI transcription factors. In both parental and Ofd1 mutant cells, Gli2 mRNA is highly expressed, whereas no Gli3 mRNA can be detected by either qRT-PCR or RNA-seq (Supplementary Fig. S4). Thus GLI2, not Gli3, is the primary effector mediating HH signaling transcription in medulloblastoma cells. We then evaluated how posttranslational processing of GLI2 is regulated in both parental and cilia mutant cells. In parental cells, GLI2 is predominantly in the full-length form (GLI2-F) with a small fraction in the truncated repressor form (GLI2-R), whereas inhibition of Smo drastically reduces GLI2-F and increases GLI2-R (Fig. 4C). These data indicate that shutdown of HH signaling by Smo inhibitors in parental cells involves conversion of GLI2-F to GLI2-R. In contrast, in cilia mutant cells, only GLI2-F is detected and levels are not affected in the presence of Smo inhibitors (Fig. 4C and D), suggesting that impaired Gli processing due to cilia loss could explain persistent activation of downstream HH signaling. Cellular fractionation experiments showed that GLI2-F is associated with nuclear chromatin in cilia mutant cells, indicative of transcriptional activity (Fig. 4E and F). Importantly, knockdown of GLI2 effectively reduced both transcriptional output of HH signaling and cell viability in both parental and resistant cells (Fig. 4G and H), indicating that proliferation and survival of these cells depends on GLI2-mediated HH signaling output. Together, these results suggest that loss of cilia abolishes SMO-dependent full activation of HH signaling and simultaneously eliminates GLI2-R formation, resulting in low but persistent GLI2 activity and HH pathway transcriptional output. This trade-off enables cilia mutant cells to escape drug inhibition and maintain a “persister” state by surviving on this persistent low output of HH signaling. In stark contrast to cilia mutants, Sufu-mutant cells have drastically low levels of both full-length and repressor forms of GLI2 (Fig. 4C), indicating that a key role of SUFU is to sequester and stabilize GLI protein (23–25). These data highlight a mechanistic difference between Sufu mutation– and Ofd1 mutation–mediated activation of HH signaling.

To examine applicability of these effects to normal development, we performed CRISPR/Cas9-mediated knockout of Ofd1 in cerebellar granule cell precursors (GCP) from wild-type and Ptch^+/− mice, and measured HH signaling activity based on Gli1 expression levels. In the absence of primary cilia (sgOfd1 conditions), Gli1 levels were significantly reduced in both wild-type and Ptch^+/− cells compared with SAG-stimulated activation (Supplementary Fig. S5C). However, these levels of Gli1 were significantly higher than those in unstimulated control sgGFP cells, and there were no significant differences between wild-type and Ptch^+/− GCPs, indicating that this low but persistent activation of HH signaling is independent of Ptch status. Together, these data suggest that loss of primary cilia results in a constitutive low-level output of HH signaling in both tumor and developmental contexts.

Although primary cilia are absent in both R25 and R14 due to mutations in Ofd1, these clones exhibit different levels of HH signaling output and growth rates. R25 cells have lower levels of GLI1 expression and grow much more slowly than R14 cells (Fig. 5A and B). Parental cells depleted of OFD1 or KIF3A exhibited a phenotype similar to R25, with low activation of HH signaling and slow growth (Supplementary Fig. S5A and S5B), indicating that loss of primary cilia alone is sufficient to confer resistance by enabling tumor cells to attain the “persister” state. As persister cells have the potential for further genetic alteration and tumor evolution, we postulated that R14 may harbor a relevant alteration in addition to cilia loss. Because we observed frequent genomic loss of the second Sufu allele in Sufu-mutant clones, we examined the genomic copy number of Sufu in Ofd1-mutant clones. Quantitative PCR revealed only one copy of the Sufu gene in R14, whereas two copies remained in R25 (Fig. 5C). Moreover, heterozygosity for Sufu in R14 is associated with reduced protein levels of SUFU (Fig. 5D). Based on these observations, we hypothesize that in the absence of primary cilia, sequestration of GLI2 by SUFU becomes a bottleneck that limits GLI2 activation and nuclear translocation. Therefore, losing one copy of the Sufu gene in this sensitized condition drastically enhances GLI2 activation and causes a “persister” state of slow-growing resistant cells to evolve into rapidly growing resistant tumors. To test this hypothesis, we systematically adjusted SUFU levels in Ofd1 mutant cells. Notably, a 4-fold increase in SUFU suppressed cell proliferation rate and dampened HH signaling in Ofd1 (R14) cells, whereas proliferation rate and HH pathway activity of parental cells were not perturbed by a similar increase in SUFU (Fig. 5E and F). Furthermore, redistribution of GLI2-F from the nucleus into the cytoplasm was observed upon increase of SUFU expression only in Ofd1 (R14) mutant cells (Fig. 5G and H), resulting in reduction of GLI2-F in the nucleus and decreased GLI1 expression. Together, these data indicate that in the absence of cilia, SUFU acts in a quantitative, dose-dependent manner in limiting GLI2-mediated HH signaling output in tumor cells. To examine whether this synergistic interaction between loss of cilia and Sufu heterozygosity occurs in a broad range of contexts, we turned to a genetically defined system, Sufu^+/+ and Sufu^+/− mouse embryonic fibroblast (MEF) cells. As observed for SMB cells, heterozygosity of Sufu significantly reduced SUFU protein and mRNA levels (Supplementary Fig. S6A and S6B). Importantly, depletion of OFD1 induces greater GLI1 expression in Sufu^+/− than in Sufu^+/+ cells (Supplementary Fig. S6C). Taken together, these data indicate that Sufu heterozygosity, which is not sufficient to trigger activation of HH signaling on its own, can synergize with loss of primary cilia to augment pathway activation (Fig. 5I). This surprising synergy provides a new route for robust activation of downstream HH signaling in the absence of primary cilia and may play a role in a range of developmental and disease contexts.

To determine if cilia mutant cells generate resistant tumors in vivo, Ofd1-mutant cells were orthotopically transplanted into mice. Both parental and Ofd1-mutant cells initiate tumors in the brain. Notably, although sonidegib abrogates growth of parental cells, Ofd1-mutant cells exhibit complete resistance to sonidegib treatment (Fig. 6A). We next examined whether loss of primary cilia occurs in preclinical in vivo models in the context of acquired resistance. SMB21 parental cells were transplanted into nude mice and treated with sonidegib. After the initial robust response to sonidegib treatment, resistant tumors developed spontaneously in all animals (Fig. 6B). Immunostaining with cilia markers revealed that resistant tumors had a greater percentage of unciliated cells than tumors that were not treated with sonidegib (Fig. 6C and D). As a control, we examined the prevalence of cilia in SMB cells expressing Smo^D477G, a SMO mutant resistant to sonidegib. Xenografts of these resistant cells have the same level of cilia formation before and after treatment with sonidegib (Fig. 6C and D). Together, these data indicate that loss of primary cilia confers resistance to SMO inhibitors in vivo.

To determine whether loss of primary cilia occurs in cancer patients treated with SMO inhibitors, we analyzed datasets of BCC, a common cancer that is the FDA-approved indication for SMO inhibitors (10). Analysis of sequencing data from 11 resistant and 48 untreated patients (“resistant” and “untreated”) revealed many mutations in ciliary genes (Supplementary Table S2). Importantly, the incidence of mutations in ciliary genes is significantly greater in resistant samples compared with untreated tumors (Fig. 6E and F). Furthermore, in 1 patient, we identified both a SUFU mutation and an OFD1 mutation in the same resistant sample. In addition, resistant samples from 2 other patients showed a loss of one copy of chromosome 10q including the SUFU locus, as well as mutations in ciliary genes. Notably, loss of one copy of chromosome 10q is one of the most common copy-number aberrations in the SHH subgroup of medulloblastoma (37/133, 28%; ref. 17). Mutations in any one of the ciliary genes in this large set of patients may lead to rapidly growing resistant tumors. Together, these results provide clinical evidence that loss of primary cilia, alone or in combination with additional alterations, constitutes a route for resistance to SMO inhibition.

To identify global changes in cell state that are associated with loss of cilia and might be indicative of resistance to SMO inhibitors, we compared expression profiles of cilia mutant cells with parental SMB cells in our model system. A total of 236 gene signatures were evaluated by GSEA, including 186 gene sets derived from the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway database and 50 hallmark gene sets from the Molecular Signatures database (26). Genes involved in myogenesis and cytokine–cytokine receptor interaction are among the most upregulated signatures, whereas genes involved in oxidative phosphorylation and ribosome function are among those most downregulated (Supplementary Fig. S7A). These data suggest that metabolic state and protein translation are altered by cilia loss, and these changes can potentially be explored as an indicator for clinical resistance. To evaluate the clinical relevance of this expression profile, we constructed a gene signature based on the top 50 differentially upregulated genes and the top 50 downregulated genes, defined as the “Ciliomor” (Ciliopathy tumor) signature (Supplementary Fig. S7B). We first validated this signature in medulloblastoma cells independently created by CRISPR/Cas9-mediated mutations in Ofd1 and Kif3a. Using single-sample GSEA (ssGSEA), a Ciliomor signature score was calculated for each sample. Both Kif3a and Ofd1 knockout cells are indeed associated with higher scores of Ciliomor signature compared with control cells (Supplementary Fig. S7C). Using this signature, we applied the same ssGSEA analysis to a cohort of patients with BCC treated with SMO inhibitors to generate an enrichment score in each patient sample (9). This analysis showed resistant samples preferentially associated with higher scores of Ciliomor signature (Fig. 6G and H), supporting the notion that cilia loss might be associated with clinical resistance to SMO inhibitors. Due to lack of resistant medulloblastoma samples, we have also applied ssGSEA analysis in a cohort of treatment-naïve medulloblastoma samples (17). We found that 3 of 73 cases exhibited high Ciliomor signature scores (Supplementary Fig. S7D). This preliminary observation raises a possibility of preexisting resistance to SMO inhibitors among these samples. Examination of pre/posttreatment patient cohorts will be necessary to confirm the implication in medulloblastoma.

To identify therapeutic strategies that overcome this new type of resistance, we explored epigenetic and transcriptional targeting strategies. Parental, Sufu-mutant, and Ofd1-mutant cells all responded to arsenic trioxide (ATO; Fig. 6I), a reported Gli inhibitor (27), in accordance with GLI2 knockdown results shown above. Preclinical studies indicated that a BET bromodomain inhibitor, JQ1, which affects the epigenetic landscape (28), is effective in treating the SHH subgroup of medulloblastoma (29). We found that treatment with JQ1 significantly reduced cell proliferation and survival in parental, Sufu-mutant, and Ofd1-mutant cells (Fig. 6J). Together, these data suggest that intervention with epigenetic and transcriptional inhibitors has the potential to overcome diverse mechanisms of resistance to SMO inhibitors.

Here, we developed a piggyBac transposon system as a new screening platform for identification of drug resistance mechanisms. This approach provides a fast, efficient, and systematic approach parallel and complementary to library-based functional screens such as CRISPR, RNAi, and ORFeome expression library. Compared with library-based approaches, this approach has several unique advantages. First, interrogation of the genome is not limited by the design and content of libraries, and thus has the potential to uncover functionally relevant regions previously overlooked. Second, DNA transposons can be easily engineered to incorporate new features such as selection markers or regulatory elements to facilitate both gain-of-function and loss-of-function screens. Third, transposon-mediated mutations may easily be reversed to validate causal relationship, which is key for successful functional screens. However, one major limitation of insertional mutagenesis methods is their low probability of mutating both alleles of a gene in the same cell. Homozygous mutations require two independent transposon insertion events that mutate both alleles, or else depend on a collaborating genomic event to inactivate the second allele. The exception is for genes on the X chromosome. Ofd1 is on the X chromosome, and parental SMB21 cells are derived from a male mouse. Thus, one transposon insertion is sufficient to cause complete loss of Ofd1 function.

Recurrent mutations in Sufu (13) and Ofd1 (3) suggest that our screen has achieved a good degree of coverage of the haploid genome. Even though our screen did not identify any resistant clone in which both alleles are mutated by transposons, the screen was able to detect the transposon insertions that cooperate with other genetic events. In the majority of 13 Sufu-mutant clones, the transposon-mutated allele in combination with genomic loss of the second Sufu allele resulted in complete loss of SUFU. We speculate that the frequent loss of the second Sufu allele is partially due to an unstable genome in parental SMB21 cells and is a favorable event selected by drug inhibition. Notably, loss of Sufu genomic region is one of the most frequent copy-number variation events in human medulloblastoma, hence constituting a physiologically relevant event. Intriguingly, we found more Sufu mutants (13) than Ofd1 mutants (3). This ratio suggests that the Sufu locus may be a hotspot for more frequent transposon landing, or Sufu mutations may enable more robust resistance.

A surprising finding of our study is identification of loss of primary cilia as a new mechanism of resistance to SMO inhibitors. The primary cilium is notable for supporting maximal activation of HH signaling and is essential for SMO activation and conversion of GLI2 into GLI2-A. Paradoxically, we find that loss of primary cilia causes resistance to SMO inhibitors. We show that cells lacking primary cilia express HH pathway signature genes and maintain a persistent low-level activation of HH signaling. Our results further indicate that primary cilia are required for conversion of GLI2 into GLI2-R and for the complete cessation of HH signaling in the presence of SMO inhibitors, suggesting that cells without cilia evade the effect of SMO inhibitors by eliminating GLI2-R. Consequently, cilia mutant cells grow slowly, but can survive as persister cells during drug treatment (Fig. 7). Interestingly, Ift80 and Rpgrip1l were additional ciliogenesis genes with a single hit in our screen (Supplementary Table S1). Further experimental verification would be needed to determine whether these mutations are driver or passenger events.

We also uncovered a new evolutionary path wherein cilia mutant tumors in the persister state can progress into a fast-growing state. Loss of one copy of Sufu synergizes with the loss of primary cilia to accelerate resistant tumor regrowth. Mechanistically, in the absence of cilia-dependent GLI2 processing, cytoplasmic sequestration of GLI2 by SUFU becomes a bottle-neck step that restrains GLI2 activation. Therefore, any reduction of SUFU protein in this sensitized condition synergizes with loss of primary cilia to enhance GLI2 activation and proliferation, resulting in the evolution of more aggressive tumor growth. It is conceivable that other alterations such as activation or amplification of Gli2 may be an alternative way to synergize with loss of primary cilia to promote resistance to SMO inhibitors.

In clinical sequencing data from BCC tumors, we identified many mutations in ciliary genes (Supplementary Table S2) and demonstrated that these mutations are enriched in drug-resistant samples. Moreover, we identified co-occurrence of SUFU and OFD1 mutations in one resistant sample. Together, these results provide clinical evidence that loss of primary cilia, alone or in combination with additional mutations, constitutes a route for bypassing SMO inhibition. Given that hundreds of genes are critical for primary cilia formation (20), mutations in ciliary genes are likely to represent a broad new class of therapeutic resistance that shares key pathologic characteristics.

Do mutations in ciliogenesis genes cause cancer in general? HH pathway–dependent tumors are predominantly initiated by loss of Ptch or activating mutations in Smo that cause robust activation of HH signaling. Our study demonstrated the significance of cilia loss in the context of drug resistance. Given that this is a general mechanism for low-level activation of HH signaling in both tumor resistance and developmental settings, it is conceivable that this mechanism could also play a role in tumorigenesis. Our findings predict that cilia loss in combination with additional alterations, such as Sufu reduction, may be a potent oncogenic driver. Indeed, removal of primary cilia has been reported to cooperate with constitutive activation of GLI2 to promote tumor initiation in mouse models (30, 31). Because GLI2 amplification and loss of SUFU genomic region are common events in human medulloblastoma and BCC, sequencing analyses of sufficient number of clinical tumor samples should be able to determine whether loss of primary cilia synergizes with other alterations and thereby contributes to oncogenesis.

Finally, this study reveals two different strategies to overcome diverse mechanisms of resistance to SMO inhibitors that could inform future therapies. Consistent with evidence that GLI2 is necessary for the growth of resistant tumors that lack primary cilia, ATO, which is reported to abrogate GLI2 function, can be used to treat these resistant tumors. Therapeutic intervention of epigenetic states by bromodomain inhibitors such as JQ1 provides an alternative approach to counter resistance that will enable enhanced treatment of patients with HH-dependent tumors.

Additional details are provided in the Supplementary Data.

All experimental procedures were done in accordance with the National Institutes of Health guidelines and were approved by the Dana-Farber Cancer Institutional Animal Care and Use Committee. The nu/nu mice were obtained from Charles River Laboratories.

SMB cell lines (SMB21, SMB55, and SMB56) were derived from spontaneous medulloblastoma tumors in Ptch^+/− mice. They were established in our lab in 2010 and were described and characterized in ref. 16. SMB cells exhibited LOH for Ptch and mutations in Trp53 as described. The propagated cells were authenticated by genotyping Ptch locus and testing sensitivity to sonidegib. SMB cells were cultured as neurospheres at 37°C in a humidified incubator with 5% CO₂ in DMEM/F12 media (2% B27, 1% Pen/Strep). To passage cultures, cells were dissociated with accutase and plated 1:3 in fresh media.

Transfection was achieved using 5 μg of mPBase plasmid, 5 μg of PB transposon plasmid, and 30 μL of Fugene6 (Roche) per million SMB21 cells. After 1 week of selection with puromycin (1 μg/mL), 30 million cells were plated in soft agar. After 4 to 8 weeks of 1 μmol/L sonidegib selection, 29 individual resistant clones were isolated. Twenty-seven clone cell lines were successfully established. Transposon insertion identification was described in Supplementary Methods.

Transposon cargo was flanked by two FRT sites. Its removal was achieved using lentiviral-delivered Flpo. Removal of transposon cargo rendered cells sensitive to puromycin. For transposon cargo removal, 0.5 × 10⁶ SMB cells were infected with 1 mL of viral supernatant of LV-tdTomato or LV-Flp. Forty-eight hours after infection, blasticidin (2 μg/mL) was added, and cells were selected for 3 weeks.

SMB cells were seeded in 96-well plates (3 × 10⁴ cells per well). Serial dilutions of the relevant compound in DMSO were used, yielding final drug concentrations ranging from 10 μmol/L to 0.01 nmol/L. In all cases, the final volume of DMSO did not exceed 1%. Cells were incubated for 72 hours following addition of sonidegib, vismodegib, or cyclopamine, or for 96 hours following the addition of ATO or JQ1. Cell viability was measured using CellTiter 96 Aqueous One Solution (Promega) and calculated as a percentage of control (DMSO-treated cells). A minimum of three replicates were performed for each cell line and drug combination. Survival curves were modeled using a nonlinear regression curve fit with a sigmoid dose–response, and displayed using GraphPad Prism 5. Note that 10 mmol/L stocks of compounds were made in DMSO and stored at −20°C. Sonidegib (LDE225), vismodegib (GDC-0449), and cyclopamine were purchased from Selleck Chemicals. ATO was purchased from Sigma Aldrich. JQ1 was purchased from Cayman Chemicals.

For immunostaining, cells were fixed in 4% paraformaldehyde, permeablized with PBS containing 0.2% Triton X-100, blocked with PBS containing 0.1% Triton X-100, 5% NGS, and then incubated with primary and secondary antibody solutions and DAPI. Primary antibodies used: Gamma-Tubulin (Sigma; 1:250) and Acetylated Tubulin (Invitrogen; 1:250). Secondary antibodies were conjugated to Alexa Fluor 488 or 568 (Invitrogen; 1:200) or Cy5 (Jackson ImmunoResearch Laboratories; 1:200).

For immunoblots, cell lysates were made in RIPA buffer (50 mmol/L Tris-HCl, pH 7.4, 150 mmol/L NaCl, 0.25% deoxycholic acid, 1% NP-40, 1 mmol/L DTT, 10 mmol/L NaF, 1 mmol/L NaVO₃, and 1 mmol/L PMSF) with protease inhibitor cocktail (Roche). Lysates were quantified (Bradford assay), normalized, reduced, denatured (95°C), and resolved by SDS gel electrophoresis on 4% to 12% Bis-Tris Glycine gels (Invitrogen). Proteins were transferred to PVDF membranes (Bio-Rad) and probed with primary antibodies recognizing GLI1 (Cell Signaling Technology; 1:1,000), SUFU (Cell Signaling Technology; 1:1,000), Actin (Sigma; 1:10,000), GLI2 (R&D; 1:3,000), KIF3A (Sigma; 1:1,000), IFT88 (Sigma; 1:1,000), OFD1 (a gift from Jeremy Reiter; 1:1,000; ref. 32), Vinculin (Sigma-Aldrich; 1:10,000), SFPQ (Abcam; 1:1,000), and Histone H2B (Cell Signaling Technology; 1:1,000). Secondary antibodies are either horseradish peroxidase–linked (Bio-Rad) or infrared dye–labeled (LI-COR). Proteins were visualized using LI-COR Odyssey Imager, film, or ImageQuant LAS 4000 imager.

Cells (5 × 10⁶) in 100 μL were injected subcutaneously into the right flank of nu/nu mice (6–8 weeks old). Tumor volumes were measured twice a week and calculated using the formula V = 0.5 × a × b², where a and b are the shortest and longest perpendicular tumor diameters, respectively. When tumors reached 150 mm³, animals were randomly separated into treatment groups (5 mice per group). Sonidegib was administered at 80 mpk by oral gavage once daily. Sonidegib was formulated as diphosphate salt in 0.5% methylcellulose and 0.5% Tween 80 (Fisher). Mice were euthanized when tumors exceeded 2,000 mm³. Tumor progression data were initially reported (16). The Kaplan–Meier survival analysis and cilia analysis are reported in this study.

Cells (5 × 10⁵) in 2 μL were injected orthotopically into the cerebellum of nu/nu mice (6–8 weeks old). Injection coordinates are (relative to bregma): anterior, 6.5 mm; lateral, 1.0 mm; depth, 2 mm. After engraftments were confirmed, mice were randomly separated into treatment or control groups (8 mice per group). Animal were treated with vehicle control, sonidegib (80 mpk daily p.o.), or JQ1 (50 mpk daily i.p.). Tumor growth and response to drug treatments were monitored by measurement of whole-body bioluminescence once a week. Animals were euthanized once they entered moribund status (end point for survival analyses). Tumor growth and metastatic spread were quantified by measuring bioluminescence imaging intensity over the head or over the spinal cord, respectively.

Publicly available whole-exome sequencing of resistant and untreated BCC samples was analyzed based on reported single-nucleotide variants and insertions/deletions (original data doi: 10.1016/j.ccell.2015.02.001; ref. 10).

Mutations in ciliary genes were determined using the SYSCILIA gold-standard list of 303 genes for known ciliary components (20).

The Ciliomor signature was defined by the top 50 differentially upregulated genes together with the top 50 differentially downregulated genes in cilia mutant cells compared to parental sensitive cells. Expression heat map of Ciliomor signature genes is shown in Supplementary Fig. S7.

ssGSEA, an extension of GSEA, calculates an enrichment score for each sample against one gene set (33). The ssGSEA enrichment score represents the degree to which the genes in this gene set are coordinately upregulated or downregulated within a sample. ssGSEA was applied to generate an enrichment score of Ciliomor signature in each patient sample. Gene expression data of normal skin (n = 8), sensitive (n = 4), and resistant BCC (n = 9) samples were obtained from GEO (GSE58375; ref. 9). Gene expression data of treatment-naïve medulloblastoma samples were from GEO (GSE49243; ref. 17).

Statistical analyses were performed with the Student t test, one-way ANOVA with Dunnett or Bonferroni post hoc test as indicated. A P value of 0.05 or less was considered statistically significant. All data analyses were performed using Microsoft Excel or GraphPad Prism 5.

RNA-seq data are accessible through GEO series accession number GSE103538.

J.F. Kelleher is Head of Discovery Biology at Kymera Therapeutics and has ownership interest (including patents) in the same. No potential conflicts of interest were disclosed by the other authors.

Conception and design: X. Zhao, E. Pak, J.F. Kelleher, R.A. Segal

Development of methodology: X. Zhao, E. Pak, K.J. Ornell, T. Ponomaryov

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): X. Zhao, E. Pak, K.J. Ornell, M.F. Pazyra-Murphy, E.L. MacKenzie, E.J. Chadwick

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): X. Zhao, E. Pak, K.J. Ornell, M.F. Pazyra-Murphy, E.L. MacKenzie, R.A. Segal

Writing, review, and/or revision of the manuscript: X. Zhao, E. Pak, K.J. Ornell, T. Ponomaryov, J.F. Kelleher, R.A. Segal

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): X. Zhao, E. Pak, K.J. Ornell, M.F. Pazyra-Murphy

Study supervision: X. Zhao

We thank Stephan Teglund for Sufu^+/− and Sufu^+/+ MEF cells, Jeremy F. Reiter for OFD1 antibody, Anthony E. Oro for sharing unpublished information, Harvard Medical School EM facility for help with TEM images, Lurie Family Imaging Center at Dana-Farber Cancer Institute for preclinical studies, Molecular Biology Core Facility at Dana-Farber Cancer Institute for NGS service, Rodent Histopathology Core at Dana-Farber/Harvard Cancer Center (supported by NIH 5 P30 CA06516), and members of the Segal laboratory for scientific discussion and critical reading of the manuscript.

The work was funded by grants from the NIH: P01 CA142536, Alex's Lemonade Stand Foundation, and Emerald Foundation (to R.A. Segal), and F31CA183145 (to E. Pak).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.