Disruption of GMNC-MCIDAS multiciliogenesis program is critical in choroid plexus carcinoma development

Multiciliated cells (MCCs) in the brain reside in the ependyma and the choroid plexus (CP) epithelia. The CP secretes cerebrospinal fluid that circulates within the ventricular system, driven by ependymal cilia movement. Tumors of the CP are rare primary brain neoplasms mostly found in children. CP tumors exist in three forms: CP papilloma (CPP), atypical CPP, and CP carcinoma (CPC). Though CPP and atypical CPP are generally benign and can be resolved by surgery, CPC is a particularly aggressive and little understood cancer with a poor survival rate and a tendency for recurrence and metastasis. In contrast to MCCs in the CP epithelia, CPCs in humans are characterized by solitary cilia, frequent TP53 mutations, and disturbances to multiciliogenesis program directed by the GMNC-MCIDAS transcriptional network. GMNC and MCIDAS are early transcriptional regulators of MCC fate differentiation in diverse tissues. Consistently, components of the GMNC-MCIDAS transcriptional program are expressed during CP development and required for multiciliation in the CP, while CPC driven by deletion of Trp53 and Rb1 in mice exhibits multiciliation defects consequent to deficiencies in the GMNC-MCIDAS program. Previous studies revealed that abnormal NOTCH pathway activation leads to CPP. Here we show that combined defects in NOTCH and Sonic Hedgehog signaling in mice generates tumors that are similar to CPC in humans. NOTCH-driven CP tumors are monociliated, and disruption of the NOTCH complex restores multiciliation and decreases tumor growth. NOTCH suppresses multiciliation in tumor cells by inhibiting the expression of GMNC and MCIDAS, while Gmnc-Mcidas overexpression rescues multiciliation defects and suppresses tumor cell proliferation. Taken together, these findings indicate that reactivation of the GMNC-MCIDAS multiciliogenesis program is critical for inhibiting tumorigenesis in the CP, and it may have therapeutic implications for the treatment of CPC.


INTRODUCTION
The choroid plexus (CP) in each brain ventricle consists of stromal vasculatures ensheathed by epithelia [1][2][3]. The CP is responsible for the synthesis and secretion of cerebrospinal fluid in the central nervous system. Recent studies revealed multiciliated cells (MCCs) in the CP of the mouse [1][2][3]. Unlike ependymal cells that form multiple motile cilia to drive cerebrospinal fluid flow within the central nervous system after birth, MCCs in the CP epithelia arise during embryogenesis, display increased motility of their multiple cilia until birth, and experience a gradual regression in the motility during postnatal life [3][4][5]. Tumors of the CP comprise~20% of brain tumors diagnosed in children under 1 year of age [6,7]. Research aimed at understanding the origin and molecular characteristics of CP carcinoma (CPC) is essential for developing new therapies to improve clinical outcomes [8][9][10][11][12].
GMNC and MCIDAS both transcriptionally regulate multiciliogenesis in ependymal cells through the E2F4/5-DP1 transcription factors [15,16,29]. In contrast, Geminin antagonizes the transcriptional functions of GMNC and MCIDAS. And Geminin and GMNC play antagonistic roles in the maintenance of the stem and ependymal cell populations in the adult neurogenic niche, respectively [30,31]. CP epithelial cells are derived from neuroepithelial progenitors that express orthodenticle homeobox 2 (OTX2) and Growth differentiation factor 7 (GDF7). As these progenitors exit the cell cycle to undergo multiciliogenesis and differentiation, TAp73 is activated in MCCs, while aquaporin 1 (AQP1), transthyretin (TTR), and cytokeratins are upregulated in epithelial cells [32]. Our previous work showed that, in contrast to all other MCCs, TAp73 was dispensable for multiciliogenesis in the CP, suggesting that its differentiation program may be distinct from other MCC types [26]. Therefore, further analysis of the molecular mechanisms governing multiciliogenesis in the CP, as well as the functional significance of these cilia, will be important to understand their role in the pathology of both ciliopathies and CP tumor development.
Examination of human CP tumors revealed abnormal NOTCH activity in a subset of tumors [33], and we demonstrated that sustained NOTCH1 expression in mice led to CP papilloma (CPP) that arose from monociliated progenitors in hindbrain roof plate [34,35]. These progenitors proliferated in response to Sonic Hedgehog (SHH), but subsequently became quiescent after birth [34][35][36]. Here, we show that human CPC, and to a lesser extent CPP, display consistent defects in the GMNC-MCIDAS transcriptional program and amplifications of NOTCH pathway components. Using two distinct murine models, we found that CPCs in mice exhibit multiciliation defects and a deficient GMNC program. In addition, persistent NOTCH and SHH signals are sufficient to drive aggressive tumors in mice that resemble human CPC. These tumors display singular primary cilia resulting from the repression of the GMNC-MCIDAS multiciliation program by NOTCH. Biochemical or pharmacological disruption of the NOTCH complex restored multiciliation and suppressed tumor cell proliferation. Our findings indicate that the GMNC-MCIDAS transcriptional network is essential for MCC differentiation in the CP, and its activation can induce multiciliation and decrease CP tumor cell proliferation. These findings underscore the critical role of a compromised GMNC-MCIDAS multiciliogenesis program in CPC development and suggest that this could be exploited therapeutically to impair proliferation and promote tumor differentiation.

RESULTS
CPCs in humans exhibit reduced multiciliation and a deficient GMNC-MCIDAS program Most CP tumors in humans, especially CPCs, consist of monociliated tumor cells and frequently display large-scale genomic alterations [34,[37][38][39]. Analysis of published data revealed recurrent chromosomal changes that affect loci encompassing multiciliogenesis regulators, including GMNC on chromosome 3, that is lost in all hypodiploid CPCs, MCIDAS, CCNO, microRNA 449 (MIR449), and CDC20B, that are all located within the same locus of chromosome 5, and MYB on chromosome 6, that is lost in many CPCs (Fig. 1A). Conversely, N-acetyl galactosamine-type Oglycosylation enzyme GALNT11, a positive regulator of NOTCH signaling on chromosome 7, is gained in >80% CP tumors (Fig. 1A) [37][38][39][40]. In agreement, among 11 cases of human CPCs examined, most displayed significantly reduced or complete loss of GMNC expression, and GMNC expression was heterogeneous and only detected in a subpopulation of tumor cells (Fig. 1B, C). Decreased FOXJ1 expression was also observed in the majority of samples (Fig. 1B, C). A similar trend of reduced GMNC and FOXJ1 expression was observed in 31 human CPPs (Fig. 1B, C). Consistent with the requirement for GMNC and FOXJ1 in MCC differentiation, analysis of cilia marker ADP-ribosylation factorlike 13b (ARL13B) in six CPCs found that all were monociliated, while 11 of 17 CPPs analyzed were monociliated (Fig. 1B, D). Accordingly, RT-qPCR analysis revealed low levels of GMNC and MCIDAS expression in most human CP tumors compared to normal tissues, while FOXJ1 expression in CPCs was significantly lower than CPPs (Fig. 1E). In contrast, TAp73 expression in human CPC and CPP varied from significantly reduced to normal levels in similar proportions ( Supplementary Fig. S1A). Moreover, analysis of a published dataset revealed differential expression of genes involved in ciliogenesis in human CP tumors, contributing to significant enrichment of the pathway (Supplementary Fig. S1B, C) [37]. Thus, CPCs in humans are characterized by multiciliation defects and deficiencies in the GMNC-MCIDAS program, as well as recurrent amplification of NOTCH regulators.
Gmnc-Mcidas signaling is essential for generating multiciliated epithelia in the CP These results suggested that suppressing the MCC fate program controlled by GMNC and MCIDAS was a key step in the genesis of CP tumors. While GMNC was implicated in the formation of MCCs in the CP [15], a detailed examination of its role in the CP has not been carried out. The Gmnc conditional allele (Gmnc flox/flox ) has two loxP sites located in introns 3 and 5 that allow Cre-mediated deletion of exons 4/5 to generate the null allele (Gmnc −/− ) [15]. Using electron microscopy and immunostaining, we compared wild type CP to animals with a conditional deletion of Gmnc in the roof plate/CP by the Lmx1a-Cre transgene [41]. We found that both ependymal cells and the CP epithelium of Gmnc −/− animals were comprised solely of monociliated cells, compared to wild type controls that exhibited multiple basal bodies and multiciliation ( Fig. 2A, B; Supplementary Figs. S2A, B, S3A). RT-qPCR using primers from exons 4/5 revealed significantly reduced Gmnc levels in the CP from Lmx1a-Cre;Gmnc flox/− (Lcre;Gmnc flox/− ) mice at postnatal (P) day 7 (P7), consistent with efficient Gmnc disruption (Fig. 2C). Analysis of Ki-67 expression showed that both Gmnc −/− and wild type CP epithelial cells became postmitotic and the expression of epithelial markers cytokeratins, TTR, and OTX2 was comparable between Gmnc −/− and wild type CP, though Aqp1 expression was significantly increased in Gmnc −/− animals ( Fig. 2C; Supplementary  Fig. S3B-F).
Gmnc mRNA was detected in wild type CP epithelial cells adjoining the roof plate at embryonic (E) day 13.5 (E13.5), and Gmnc transcripts persisted in the epithelial cells of Gmnc −/− CP (Fig. 2D). RT-qPCR with primers from exons 2/3 showed increased Gmnc levels, and further sequencing revealed a mutant transcript with exon 3 spliced to exon 6 (designated as Gemc1 Δ4-5 ) ( Fig. 2C; Supplementary Fig. S4A). The altered splicing causes a frame shift and stop codon after a few amino acids, generating a truncated Gmnc transcript lacking crucial functional domains. The expression of GMNC targets Foxj1 and TAp73 in the CP and ependyma was markedly reduced in Gmnc −/− mice, whereas Gmnc overexpression stimulated TAp73  Table 1). Together, these data establish that Gmnc is essential for MCC differentiation and the expression of TAp73 and Foxj1 in the CP epithelium.
We next examined critical components of the MCC transcriptional cascade, including Mcidas, Myb, and Ccno. All 3 genes were transiently upregulated in a subpopulation of epithelial cells next to the roof plate during development and their expression was lost in Gmnc −/− mice (  Table 1). Taken together, these results indicate that MCIDAS plays a critical role in multiciliogenesis downstream of GMNC in the CP.
The loss of Gmnc expression at early stages of tumorigenesis suggested that this may promote the reduced multiciliation observed in Rb1/Trp53-deficient CPC. To address this, we interbred Lcre;p53 cko ;Rb cko mice with Gmnc flox/− animals. The resulting Lmx1a-Cre;p53 cko ;Rb cko ;Gmnc flox/− (Lcre;p53 cko ;Rb cko ;Gmnc cko ) mice succumbed to CPCs that expressed OTX2 and showed proliferation levels similar to that of Lcre;p53 cko ;Rb cko mice ( Fig. 3A; Supplementary Fig. S7B). The expression of Ttr, Aqp1, Gdf7, and Foxj1 was significantly reduced, whereas TAp73 expression was more variable in CPC in these mice ( Fig. 3D; Supplementary  Table 1). A non-functional Gemc1 Δ4-5 mutant transcript was detectable in the CP of the Lcre;p53 cko ;Rb cko ; Gmnc cko mice, but its presence was dramatically reduced in tumor cells in these animals, indicating similar impairment in the upstream regulation of Gmnc (Fig. 3B, D).
Like CPP in Lcre;NICD1 mice, the abnormal CP growth in Lcre; Ptch cko ;NICD1 animals exhibited elevated expression of the SHH pathway targets Gli1 and Mycn, reduced Shh expression, as well as increased levels of NOTCH targets Hes1 and Hes5 (Supplementary Figs. S8B, S9A-C). Similar to Rb1/Trp53-deficient CPC, the abnormal CP growth in Lcre;Ptch cko ;NICD1 animals expressed OTX2, and showed reduced expression of AQP1, TTR, and cytokeratins, although some CP epithelial cells were mixed due to incomplete Cre-mediated activation of NICD1 (  Table 1). Thus, combined activation of the NOTCH and SHH pathways leads to increased proliferation and pathological cell overgrowth in the upper roof plate region, accompanied by a loss of differentiated epithelial cells in the CP. Moreover, these animals develop malignant CP tumors that closely match the characteristics of CPC in humans.

NOTCH activation leads to reduced multiciliation in CP tumors
In contrast to MCCs in the CP, NOTCH-driven CP tumors consisted of monociliated cells with decreased Foxj1 expression ( Fig. 5A) [34], suggesting that NOTCH might mediate reduced multiciliation in CP tumors. To address this, tumor cells from Lcre;NICD1 mice were treated with a recombinant amino-terminal fragment of SHH (ShhN) and a small molecule Inhibitor of Mastermind Recruitment 1 (IMR-1, or IMR-1A) to block the recruitment of Mastermind-like protein 1 (MAML1) to the NOTCH transcriptional complex, or infected with viruses expressing GFP fused to dominant negative MAML1 (dnMAML1) that disrupts the complex [48,49]. Remarkably, staining with the cilia markers ARL13B and γ-tubulin revealed GFP + multiciliated tumor cells within 72 h of treatment ( After a 7-day in vivo IMR-1 treatment from day E10.5, multiciliated tumor cells were detected in Lcre;NICD1 and Lcre; Ptch cko ;NICD1 animals at day E17.5 and day P7 (Fig. 6A, B). This was accompanied by a significant decrease in tumor cell proliferation, and a reduction of total tumor cell numbers by several folds at day P7 (Fig. 6C, D). Moreover, the expression of SOX2 in the ventricular zone, PAX6, and Atoh1 in progenitors derived from rhombic lips was comparable between wild type animals treated with IMR-1 or vehicle ( Supplementary Fig. S10C, D). Together, these results demonstrate that aberrant NOTCH signaling impairs MCC differentiation in the CP that can be rescued by NOTCH inhibition, leading to reduced tumor growth.

Gmnc suppression by NOTCH mediates defective multiciliation in CP tumors
To understand the mechanisms of MCC regulation in tumor cells, we integrated RT-qPCR, RNAseq, and spatio-temporal gene expression data. Results from these assays consistently showed that both Foxj1 and Mcidas were expressed in tumor cells at lower levels than observed in wild type CP epithelium ( Fig. 7A; Supplementary Fig. S11A-D) [34]. As this suggested that upstream regulators of the MCC program were impaired, we examined Gmnc expression. Although Gmnc exhibited ubiquitous expression in the CP epithelium, we consistently observed decreased levels of Gmnc and its downstream target TAp73 in CP tumors (Fig. 7A, B; Supplementary Fig. S11A, C, D). This was accompanied by a transient increase in the expression of Gmnn, a gene that is normally associated with proliferation and was shown to antagonize GMNC transcriptional functions (Fig. 7B) [15,30]. These results demonstrate that the GMNC-MCIDAS program is profoundly repressed in NOTCH-driven CP tumors, and this can be modulated using NOTCH pathway inhibitors. asterisks) and wild type (arrow) mice. Boxed regions are magnified on the right. Notice that Gmnc −/− CP epithelial cell exhibits single basal body and solitary cilia compared to wild type epithelial cell with multiple basal bodies. CP choroid plexus, C cilia, BB basal body. Images are representative of at least three independent experiments. B The expression of acetylated α-tubulin (ac-α-tub, magenta) and γ-tubulin (green) is shown in the CP epithelial cells in newborn Gmnc −/− and wild type animals. Boxed regions are shown in higher magnification on the right. DAPI staining (cyan) labels nuclei. Scale bar, 20 µm. BB basal body. Images represent three independent experiments. C RT-qPCR analysis of the expression of Gmnc (primers/probe from exons 4/5 or 2/3), Aqp1 TAp73, and Foxj1 in the CP from Lcre;Gmnc flox/− and wild type mice at day P7 (n = 11 animals per genotype, mean ± s.e.m., paired t-test, ****P < 0.0001). Data represent three independent experiments. D Representative images of the expression of Gmnc and Foxj1 (upper panel), Mcidas and Myb (lower panel) are shown at day E13.5 in roof plate (upper roof plate marked by dotted lines) and CP in the hindbrain and the lateral ventricle in Gmnc −/− (arrowheads) and wild type (arrows) animals. Scale bars, 50 µm. Images represent at least three independent experiments. E The expression of ARL13B (yellow) and AQP1 (green) is shown in the CP epithelial cells in the hindbrain and lateral ventricles at day P7 in Mcidas −/− (arrowheads ) and wild type (arrows) animals. DAPI staining (cyan) labels nuclei. Scale bars, 10 µm. Results were obtained from three independent experiments.
To understand the role of the GMNC-MCIDAS program in defective multiciliation of CP tumors, myc-tagged GMNC or MCIDAS was expressed in tumor cells from Lcre;NICD1 mice using viral vectors. Enforced expression of Gmnc or Mcidas led to the formation of multiple cilia and reduced proliferation in infected tumor cells within 72 h (Fig. 7C, D; Supplementary Fig. S12A, B), phenocopying NOTCH inhibition with IMR-1 that significantly increased Gmnc levels in tumor cells (Fig. 7E). We subsequently eliminated Gmnc by crossing Gmnc flox/− and Lcre;NICD1 animals.  Fig. S12E). Therefore, these results indicate that monociliation in tumor cells is maintained through NOTCH suppression of GMNC-MCIDAS signaling and suggest that GMNC loss prevents the rescue of multiciliation defects by NOTCH inhibition.
Similar to Rb1/Trp53-deficient CPC, despite the suppression of the GMNC-MCIDAS program by NOTCH, combined loss of Gmnc and Patched1 failed to induce CPC in Lcre;Ptch cko ;Gmnc flox/− mice (Supplementary Fig. S13; Supplementary Table 1), suggesting that loss of GMNC-driven multiciliation in the CP is insufficient to replace NOTCH or Rb1/Trp53 deletion in CPC. Together, these data indicate that GMNC-MCIDAS program deficiencies critically mediate cilia defects in CPC to modulate tumor growth.

DISCUSSION
CPC clinical outcomes remain dismal, leaving patients vulnerable to devastating consequences [2,3]. The gross genomic alterations in CP tumors have made the identification of driving events and actionable targets difficult [43,44]. The GMNC-MCIDAS program promotes multiciliogenesis in different tissues, is required for MCC generation in mice, and mutations in both GMNC and MCIDAS have been identified in human ciliopathies [13,14]. The observation that there is consistent disruption of multiciliogenesis program and prevalence of solitary cilia in CPC indicates that CPC has characteristics of a ciliopathy and that therapeutic strategies aimed at restoring multiciliogenesis may suppress CP tumors.
Our findings revealed the interaction of the multiciliogenesis program, NOTCH, and SHH pathways during CP differentiation and tumorigenesis. NOTCH suppressed multiciliation of roof plate progenitors, thereby preserving cilia-based signaling activated by SHH from postmitotic MCCs in CP epithelium [33]. Conversely, SHH signaling enhanced Hes1 and Hes5 expression in the roof plate in Lcre;Ptch cko mice and NOTCH-driven CP tumors. The expanded upper roof plate in Lcre;Ptch cko ;NICD1 mice is consistent with the developmental origin and cilia defect of CPC being driven by NOTCH and SHH signaling. These animals represent an ideal therapeutic model for congenital or infantile CPC, a rare condition associated with high morbidity and mortality [50,51]. Indeed, NOTCH inhibition by IMR-1 rescued the cilia deficit by inducing multiciliated tumor cells, whereas SHH pathway inhibitors suppressed tumor cell proliferation (Fig. 8) [34]. Thus, further study of the interactions between the SHH and NOTCH pathways in CP tumors is warranted to determine the therapeutic potential of activators of multiciliation and cilia-dependent signaling [52][53][54][55][56].
While the differentiation of MCCs requires NOTCH inhibition, it is unclear precisely how NOTCH impacts MCC fate during tumorigenesis [57][58][59][60][61]. Our data shows for the first time that NOTCH suppresses the expression of Gmnc and Mcidas to impair multiciliation during tumorigenesis. Gmnc is required for MCC differentiation following NOTCH inhibition, indicating that GMNC-MCIDAS signaling is required downstream of NOTCH regulation and represents a potent anti-tumor mechanism in CP tumors (Fig. 8).
Consistent with previous studies, we found that the GMNC-MCIDAS program was required for multiciliation in the CP epithelium. As progenitor cells exit the cell cycle to undergo multiciliogenesis, the expression of Gmnc, Mcidas, Foxj1, and TAp73 was upregulated, as has been observed in other multiciliated tissues [26]. Ectopic expression of either GMNC and MCIDAS stimulated Foxj1 and TAp73 expression, whereas Gmnc loss prevented the activation of Mcidas, TAp73, and Foxj1, in contrast to Mcidas-deficient MCCs that showed expression of both Foxj1 and TAp73 [14,25]. Although both TAp73 and Foxj1 are sensitive to Gmnc status, loss of TAp73 failed to affect Foxj1 expression in the CP, as it does in other MCCs, indicating that TAp73 is not integrated into the Gmnc-Foxj1 axis in the CP [26]. Consistent with this, TAp73 expression varied greatly in Rb1/Trp53deficient CPC in mice, as well as CP tumors in humans. These results highlight the need to further analyze the similarities and differences between different MCC types.
Overall, this study shows that the GMNC-MCIDAS program is required for MCC differentiation in the CP. The impairment of the program by oncogenic signals including Rb1/Trp53 defects or NOTCH activation prevents multiciliation and facilitates proliferation of CP tumor cells (Fig. 8). Therefore, activation of multiciliogenesis may serve as a potential therapeutic strategy in a subset of CP tumors. As the early events leading to the activation of the GMNC-MCIDAS program remain poorly characterized, a detailed understanding of its regulation and functions will be critical for developing strategies to target this pathway for the treatment of CPC.

MATERIALS AND METHODS Animals
Gt(ROSA)26Sor tm1.Notch1Dam /J (Rosa26-NICD1) mice, B6N.129-Ptch1 tm1Hahn / J (Ptch flox/flox ) mice, B6.129P2-Trp53 tm1Brn /J (Trp53 flox/flox ) mice, Rb1 tm2Brn /J (Rb flox/flox ) mice, and C57BL/6 mice (all from Jackson Laboratory, Bar Harbor, ME, USA), Tg(Lmx1a-cre)1Kjmi (Lmx1a-Cre) mice, Gmnc tm1Strc (Gemc1 −/+ ) mice, and Gmnc tm1.1Strc (Gemc1 flox/+ ) mice were maintained by breeding with C57BL/6 mice (Supplementary Table 1). Animals were housed in the Animal Research Facility at New York Institute of Technology College of Osteopathic Medicine in accordance with NIH guidelines. All animal experimental procedures were approved by Institutional Animal Care and Use Committee (IACUC) and performed in compliance with national regulatory standards. Mcidas mutant mice were housed at the Biological Resource Center of the Agency for Science, Technology and Research (A*STAR) of Singapore, and experiments performed with these animals followed guidelines stipulated by the Singapore National Advisory Committee on Laboratory Animal Research. All experimental procedures at the Institute for Research in Biomedicine were conducted following European and National Regulation for the Protection of Vertebrate Animals used for experimental and other scientific purposes (directive 86/609), internationally established 3R principles, and guidelines established by the United Kingdom Coordinating Committee on Cancer Research.
The animal experiments were not randomized, and both male and female animals were used for experiments at different time points. For analysis of the Gmnc-Mcidas program in MCC differentiation in the CP, Fig. 3 Disruption of GMNC-MCIDAS program mediates multiciliation defects in Rb1/Trp53-deficient CPC. A Hematoxylin and eosin (H&E) staining and Ki-67 expression are shown in CPC (arrowheads, boundary between tumor and unaffected brain region is marked by dotted lines) and the CP (arrows) from Lcre;p53 cko ;Rb cko and Lcre;p53 cko ;Rb cko ;Gmnc cko animals. Scale bars, 50 µm. Images are representative of at least three independent experiments. B RNAscope analysis of Gmnc and Mcidas expression in tumor cells (arrowheads, boundary between tumor and unaffected brain region is marked by dotted line) and the CP (arrows) in Lcre;p53 cko ;Rb cko and Lcre;p53 cko ;Rb cko ;Gmnc cko animals. Scale bar, 50 µm. Results were obtained from three independent experiments. C Representative images of immunofluorescence of ARL13B (yellow) and OTX2 (green) are shown in multiciliated epithelial cells (arrows) or monociliated tumor cells (arrowheads) in wild type and Lcre;p53 cko ;Rb cko animals, respectively. Scale bar, 10 µm. Data represent three independent experiments. D RT-qPCR analysis of gene expression in wild type CP and CPC from Lcre;p53 cko ;Rb cko and Lcre;p53 cko ;Rb cko ;Gmnc cko animals (wild type CP: n = 10; CPC: n = 11 for Lcre;p53 cko ;Rb cko animals, n = 10 for Lcre;p53 cko ;Rb cko ;Gmnc cko animals; mean ± s.e.m., one-way ANOVA, *P < 0.05, ****P < 0.0001, NS not significant). Data represent three independent experiments. E Representative images of immunofluorescence of ARL13B (red) are shown in tumor cells from Lcre;p53 cko ;Rb cko ; Gmnc cko animals infected with viruses expressing GMNC-myc (green) or MCIDAS-myc (green). OTX2 (green) labels tumor cells. DAPI staining (blue) labels nuclei. Scale bar, 20 µm. Three independent experiments were conducted. F Kaplan-Meier curve depicting the survival of Lcre; p53 cko ;Rb cko , Lcre;p53 cko ;Rb cko ;Gmnc −/+ , and Lcre;p53 cko ;Rb cko ;Gmnc cko animals compared to wild type mice. Fig. 4 Aberrant NOTCH and SHH signaling drive CPC in mice. A Wild type, Lcre;NICD1, Lcre;Ptch cko , and Lcre;Ptch cko ;NICD1 animals are shown at day E14.5. Notice the cranium defects resulting from enlarged and folded roof plate in the midbrain-hindbrain region of Lcre;Ptch cko and Lcre;Ptch cko ;NICD1 animals (white arrowheads). H&E staining and Ki-67 expression are shown of roof plate (upper roof plate marked by red lines) and the CP (black arrows) in the hindbrain in wild type and Lcre;Ptch cko animals, and CPP and abnormal CP growth (black arrowheads) in Lcre;NICD1 and Lcre;Ptch cko ;NICD1 animals, respectively. Enlarged roof plate disrupts the cranium in Lcre;Ptch cko and Lcre;Ptch cko ;NICD1 animals (red arrows). The upper roof plate is shown in higher magnification in the right (Lcre;Ptch cko ;NICD1 animal) and lower (wild type, Lcre;NICD1, Lcre;Ptch cko , and Lcre;Ptch cko ;NICD1 animals) panels. Scale bars, 100 µm. Quantification of Ki-67 expression in the upper roof plate and CP in the hindbrain is shown (wild type mice: n = 11; Lcre;NICD1 mice: n = 4; Lcre;Ptch cko mice: n = 3; Lcre;Ptch cko ;NICD1 mice: n = 7 for upper roof plate, n = 8 for the CP; mean ± s.e.m., one-way ANOVA, ***P < 0.001; ****P < 0.0001). Data are representative of at least three independent experiments. B Representative results of immunohistochemical staining for OTX2, and AQP1 are shown in the upper roof plate (marked by dotted lines) and the CP (arrows) in the hindbrain at day E14.5 in wild type and Lcre;Ptch cko animals, and CPP and abnormal CP growth (black arrowheads) in Lcre;NICD1 and Lcre;Ptch cko ;NICD1 animals, respectively. Residual AQP1-expressing epithelial cells (red arrowhead) are mixed with tumor cells in Lcre;NICD1 animals. Scale bar, 50 µm. Images represent at least three independent experiments.  NOTCH activation leads to reduced multiciliation in CP tumors. A Immunofluorescent staining for ARL13B (yellow) is shown at day E14.5 in the upper roof plate progenitors (marked by dotted lines and orange arrowheads) and the CP epithelial cells (arrows) in the hindbrain in wild type and Lcre;Ptch cko animals, and CPP and abnormal CP growth (white arrowheads and dotted lines) in Lcre;NICD1 and Lcre;Ptch cko ; NICD1 animals, respectively. GFP (green) labels tumor cells. DAPI staining (cyan) labels nuclei. Scale bar, 10 µm. Results were obtained from at least three independent experiments. The expression of ARL13B (red) is shown in tumor cells infected with viruses expressing GFP-tagged dnMAML1 or GFP (B), or treated with vehicle, or IMR-1/IMR-1A (C). GFP (green) labels infected or treated cells. DAPI staining (blue) labels nuclei. Scale bars, 20 µm. The percentage of multiciliated tumor cells after treatment is shown (n = 3, mean ± s.e.m., two-tailed unpaired t-test, **P < 0.01, ***P < 0.001). Results were obtained from at least three independent experiments, respectively. D The expression of Ki-67 (red) is shown in GFP + tumor cells treated with vehicle or IMR-1. Quantitation of Ki-67 expression is shown (n = 6 per treatment; mean ± s.e.m., paired t-test, ***P < 0.001). Results were obtained from at least three independent experiments. E RNAscope analysis of Foxj1 expression (red) is shown in tumor cells treated with vehicle or IMR-1. DAPI staining (blue) labels nuclei. Scale bar, 20 µm. Quantification of Foxj1 transcript is shown on the right (n = 7 per treatment; mean ± s.e.m., two-tailed unpaired t-test, **P < 0.01). Data are representative of three independent experiments.

Human samples
CP specimens were procured with informed consent from patients following the requirements by institutional review boards at Shanghai East Hospital, Sanford Burnham Prebys Medical Discovery Institute, and University Medical Center Hamburg-Eppendorf. All CP specimens from Boston Children's Hospital were obtained under an approved institutional review board protocol. All tissues were handled in accordance with guidelines and regulations for the research use of human brain tissue set forth by the NIH (http://osp.od.nih.gov/o_ce-clinical-research-andbioethics-policy). Diagnoses of human CP specimens from Boston Children's Hospital were reviewed by two neuropathologists (HGWL, S. Santagata) using standard WHO criteria [62].

Cell culture
Multiple sets of tissue specimens were collected from animals of appropriate genotype and maintained in culture. Gender information is not available for animals collected at day P7. Primary CP tumor cells were cultured as described previously [34]. Dissected specimens were dis- , D, S10A, B, S11D, S12A-E). HEK293, AD-293, and mIMCD3 cells were tested regularly for mycoplasma. Given the short time (<8-10 days) during which CP cells were maintained as primary cultures supplemented with antibiotics, we did not test for mycoplasma contamination.

Viruses
PmeI-linearized pShuttle-vectors carrying different cDNA fragments were introduced into the replication-deficient adenoviral vector pAdEasy-1 through homologous recombination in BJ5183 cells (Agilent Technologies). Successfully recombined adenoviral vector was verified by sequencing. The adenoviral plasmid was linearized by PacI digest and transfected into AD-293 cells (Agilent Technologies) to produce recombinant viral particles. All the procedures of production, purification, and use of adenoviruses were approved by Institutional Biosafety Committee.

RT-qPCR, in situ hybridization and RNAscope
Multiple sets of tissue specimens were collected from animals. Gender information is not available for animals collected at days P0 and P7. Total  Table 2) [34]. Transcript levels were determined as the number of transcripts of genes of interest relative to those of Actb (mouse) or GAPDH (human) and normalized to the mean value of control samples. The results for each set of specimens were obtained by averaging Fig. 6 NOTCH inhibition restores multiciliation in CP tumors. Representative images of immunofluorescent staining for ARL13B (A, yellow; B, red) are shown in tumor cells at day E17.5 (A) and tumor cells isolated at day P7 (B) from Lcre;NICD1 (A, B) and Lcre;Ptch cko ;NICD1 (A) animals treated with vehicle or IMR-1 from day E10.5 to day E16.5. Boxed region of ciliated cells is magnified in lower panel (A). DAPI staining (A, cyan; B, blue) labels nuclei. Scale bars, 5 µm (A), 10 µm (B). Results were obtained from at least three independent experiments. C Quantification total tumor cell numbers isolated at day P7 is shown in Lcre;NICD1 animals treated as described in A and B (n = 5 animals per treatment; mean ± s.e.m., two-tailed unpaired t-test, ***P < 0.001). D The expression of Ki-67 (red) in NICD1 + /GFP + tumor cells at day E17.5 is shown in Lcre; NICD1 animals treated with vehicle or IMR-1 from day E10.5 to day E16.5. DAPI staining (blue) labels nuclei. Scale bar, 20 µm. Quantification of Ki-67 expression in tumor cells is shown (right panel: n = 9 animals for vehicle, n = 8 animals for IMR-1; mean ± s.e.m., two-tailed unpaired ttest, **P < 0.01). Data are representative of three independent experiments. transcript levels of technical triplicates and used for subsequent analyses. Exclusion was applied when one of the triplicates was a significant outlier, and the assay was repeated in independent experiments to validate the exclusion. For analysis of GMNC, MCIDAS, and FOXJ1 expression, human samples used included 10 CPPs, 8 CPCs, and 1 for brain, trachea, lung, testis, and epididymis. For analysis of Gmnc-deficient CP samples, animals included: 11 Lmx1a-Cre;Gmnc flox/− and wild type animals, respectively. For CP tumor analysis, samples used included: wild type CP: n = 10; CPC: n = 11 from Lmx1a-Cre;p53 flox/flox ;Rb flox/flox mice, n = 10 for Lmx1a-Cre; p53 flox/flox ;Rb flox/flox ;Gmnc flox/− animals. For NOTCH-driven CP tumors, animals examined included: wild type, Lmx1a-Cre;NICD1 (n = 3 for each at genotype at each time point). For gene expression analysis of infected cells, 3 independent samples for infected and control cells were used, respectively.
In situ hybridization was performed as described at the In Situ Hybridization Core facility at Baylor College of Medicine [63].

Electron microscopy and image acquisition
Transmission electron microscopy was performed as described previously [34]. The investigator was blinded to group allocation. A whole-mount bright field was obtained using a Nikon SMZ1000 Stereomicroscope. Light and fluorescent microscopic images were obtained by a Nikon Eclipse 90i microscope system, a Nikon confocal microscope system A1 + (Nikon Instruments, Melville, NY, USA), and a ZEISS LSM 980 with Airyscan 2 confocal microscope (Carl Zeiss Microscopy, LLC, White Plains, NY, USA).

Statistical analysis and reproducibility
Multiple specimens were collected from independent samples or animals for each treatment or genotype. Pilot studies were conducted, and results from these studies were used to determine the choice of sample size for the experiment. A group size of n = 10 (5 experimental, 5 control) will provide 90% power to detect a 22% change in assay results. No randomization was used to determine how samples were allocated to experimental groups. Both male and female animals were used for experiments. Experiments were repeated with similar results to eliminate the effects of gender and age on experimental findings. Information on experiment replication is provided in legends for figures and supplemental figures. Statistical analyses were performed with GraphPadPrism 9.0 (GraphPad Software Inc., La Jolla, CA, USA). All pooled data were expressed as the mean ± standard error of the mean (SEM). Variation within each group of data was examined based on the differences between each data point and the mean of the group. The Kolmogorov-Smirnov test was used to test the normal distribution of the data. Differences between two groups were compared using paired t-test or unpaired two-tailed t-test. Differences between multiple groups were analyzed with ANOVA followed by Tukey's multiple comparisons test. Results were considered significant at *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001.

Accession numbers
Published data sets of human CP tumors (GSE14098, GSE60886) were downloaded from the GEO database. Hierarchical clustering was performed using Genesis (http://genome.tugraz.at/genesisclient/genesisclient_description. shtml). Pathway analysis using the GeneGoMetaCore Analytical Suite (http:// genego.com; GeneGo) was used to score and rank pathways enriched in data sets by the proportion of pathway-associated genes with significant expression values. RNA-seq data (BioProject ID, PRJNA282889) were analyzed.   7 Gmnc suppression by NOTCH mediates multiciliation defects in CP development and tumorigenesis. A Median FKPM (fragments per kilobase of exon per million reads mapped) values of genes in NOTCH-driven CP tumors and wild type CPs (n = 3 specimens per time point, mean ± s.e.m., two-tailed unpaired t-test, *P < 0.05; **P < 0.01). B RT-qPCR analysis of NOTCH-driven CPP and wild type CP (n = 3 animals per time point, mean ± s.e.m., two-tailed unpaired t-test, ***P < 0.001, ****P < 0.0001). Three independent experiments were conducted. C The expression of ARL13B (red) is shown in tumor cells infected with viruses expressing GMNC-myc, MCIDAS-myc, or control only. GMNC-myc (green), or MCIDAS-myc (green) labels infected cells. DAPI staining (blue) labels nuclei. Scale bar, 20 µm. Quantification of the percentage of MCCs in infected cells is shown on the right (n = 4 per treatment, mean ± s.e.m., one-way ANOVA, *P < 0.05). Results were obtained from three independent experiments. D The expression of Ki-67 (red) is shown in tumor cells from Lcre;NICD1 mice infected with viruses expressing GMNC-myc. GMNC-myc (green) labels infected cells. DAPI staining (blue) labels nuclei. Scale bar, 20 µm. Quantification of Ki-67 expression in tumor cells from Lcre;NICD1 mice infected with viruses expressing GMNC-myc or control vectors is shown in the lower panel (n = 5 per treatment, mean ± s.e.m., paired t-test, ****P < 0.0001). Data represent at least three independent experiments. E Representative images of Gmnc expression (green) by RNAscope are shown in tumor cells treated with vehicle or IMR-1. DAPI staining (blue) labels nuclei. Scale bar, 20 µm. Quantification of Gmnc transcript is shown (n = 7 per treatment; mean ± s.e.m., two-tailed unpaired t-test, ***P < 0.001). Three independent experiments were conducted. F The expression of ARL13B (red) is shown in Gmnc-deficient tumor cells treated with vehicle or IMR-1/IMR-1A. GFP (green) labels tumor cells. DAPI staining (blue) labels nuclei. Scale bar, 20 µm. Data represent five independent experiments. G RT-qPCR analysis of Foxj1 expression in tumor cells treated with vehicle or IMR-1 (tumors from Lcre;NICD1 mice: n = 6 per treatment; tumors from Lcre;NICD1;Gmnc flox/− mice: n = 4 per treatment, mean ± s.e.m., paired t-test, *P < 0.05, NS, not significant). Results were obtained from three independent experiments. H Quantification of Ki-67 expression is shown in Gmnc-deficient tumor cells treated with vehicle or IMR-1 (n = 6 per treatment, mean ± s.e.m., two-tailed unpaired t-test, NS not significant). Three independent experiments were conducted.