Developmental phosphoproteomics identifies the kinase CK2 as a driver of Hedgehog signaling and a therapeutic target in medulloblastoma

Inhibitors of the kinase CK2 may halt the growth of an aggressive form of medulloblastoma. A targeted, resilient treatment for medulloblastoma Medulloblastoma is an aggressive type of brain tumor that most often arises in children and lacks targeted therapeutic options. The subtypes driven by activity in the sonic hedgehog (SHH) pathway are particularly resistant to current drugs, such as those known as SMO inhibitors, which target this pathway. Purzner et al. used phosphoproteomics to track the development of mouse cells that give rise to medulloblastoma and identified the kinase CK2 as a likely target. CK2 inhibitors blocked the growth of SMO inhibitor–resistant, SHH-type human and mouse medulloblastoma cells and markedly extended the survival of tumor-bearing mice, in which the drug was well tolerated. One of the compounds also blocked the growth of tumors that had mutant CK2, suggesting that it is less susceptible to a common mode of drug resistance. A clinical trial is under way to test this inhibitor in pediatric patients. A major limitation of targeted cancer therapy is the rapid emergence of drug resistance, which often arises through mutations at or downstream of the drug target or through intrinsic resistance of subpopulations of tumor cells. Medulloblastoma (MB), the most common pediatric brain tumor, is no exception, and MBs that are driven by sonic hedgehog (SHH) signaling are particularly aggressive and drug-resistant. To find new drug targets and therapeutics for MB that may be less susceptible to common resistance mechanisms, we used a developmental phosphoproteomics approach in murine granule neuron precursors (GNPs), the developmental cell of origin of MB. The protein kinase CK2 emerged as a driver of hundreds of phosphorylation events during the proliferative, MB-like stage of GNP growth, including the phosphorylation of three of the eight proteins commonly amplified in MB. CK2 was critical to the stabilization and activity of the transcription factor GLI2, a late downstream effector in SHH signaling. CK2 inhibitors decreased the viability of primary SHH-type MB patient cells in culture and blocked the growth of murine MB tumors that were resistant to currently available Hh inhibitors, thereby extending the survival of tumor-bearing mice. Because of structural interactions, one CK2 inhibitor (CX-4945) inhibited both wild-type and mutant CK2, indicating that this drug may avoid at least one common mode of acquired resistance. These findings suggest that CK2 inhibitors may be effective for treating patients with MB and show how phosphoproteomics may be used to gain insight into developmental biology and pathology.


INTRODUCTION
A key challenge in developing cancer therapeutics is the identification of a target protein that is essential to the growth, survival, or metastasis of a tumor. One path to such proteins is to test developmental regulators that operate in normal cells from which the tumor is derived. A prime example is medulloblastoma (MB), the most common malignant pediatric brain tumor. Developmental, genetic, and transcriptional analyses have established a clear parallel between sonic hedgehog (SHH)-subtype MB and granule neuron precursors (GNPs) (1)(2)(3). During normal cerebellar development, GNPs proliferate extensively in response to hedgehog (Hh) signaling (2) before differentiating into granule neurons, the most abundant type of neuron in the brain (Fig. 1A) (4). In mice, this period begins at postnatal day 1 (P1), peaks at P7, and is largely complete by P14, at which point remaining GNPs have stopped dividing and have begun differentiating (5) into granule neurons (2,6). Sustained Hh target gene activity in GNPs, due to reduced function of pathway components that negatively regulate Hh signal transduction, such as Patched (PTCH1) (1), or to heightened function of activating Hh signal transducers or effectors, such as GLI2 (3), results in continued proliferation of GNPs beyond P14 and eventual rise of SHH-type MB.
To date, drugs for SHH-type MB inhibit Smoothened (SMO), a transmembrane protein that acts early in the Hh signal transduction pathway. Patients treated with SMO inhibitors initially have dramatic tumor regression but eventually develop resistance due to mutations in SMO or in genes encoding downstream components of the Hh pathway (7). Worse, at the time of diagnosis, 49% of infants and 59% of children have mutations downstream of SMO; hence, these tumors are resistant to SMO inhibitors from the start (8). Despite the great potential of Hh pathway inhibitors, children with MBs continue to receive multiple nontargeted therapies and, consequently, sustain long-term neurological and cognitive problems. There is a pressing need to identify novel drug targets that affect Hh signal transduction downstream of SMO, preferably at late steps in the pathway. Ideally, inhibition of this target using a specific drug would be refractory to single mutations of Hh pathway components.
To identify important candidate drug targets for MBs, and to explore regulation of Hh transduction, we performed a proteome-wide analysis of in vivo phosphorylation events occurring in murine GNPs during the initiation, peak, and completion of Hh-driven proliferation. Unlike genome-wide transcription assays, phosphoproteomics provided measurement of the modified state of proteins at each developmental stage and thereby inferred which kinases have stagespecific changes in activity. Kinases that are specifically active during the MB-like P7 stage of GNP development may also be required for MB growth. Kinases are critical regulators of Hh signaling (9,10) and are attractive drug targets, susceptible to inhibition by competitive inhibitors, small molecules that interfere with the binding of adenosine 5′-triphosphate (ATP) (11)(12)(13). If a kinase has multiple substrates that are important for Hh transduction and/or survival of MB cells, then the drug will have multiple restraining influences on tumor growth, and mutation of any single substrate is unlikely to cause drug resistance.

Proliferating and postmitotic GNPs have distinct phosphoproteomes
Using a protocol summarized schematically in Fig. 1B, we identified 9311 unique phosphorylation events in GNPs purified from P1, P7, and P14 mouse cerebellum. The false discovery rate (FDR) was <1% (table S1). All samples were prepared in biological triplicate, and the results were strongly reproducible (R 2 = 0.76; Fig. 1C and fig. S1). A single MB sample generated from Ptch1 +/− mice (1,14) was included for comparison ( Fig. 1C and table S1). In keeping with previous literature (5), P1 and P7 Atoh1-positive cells were highly proliferative, whereas P14 GNPs were almost universally postmitotic ( fig. S2). Many changes in phosphorylation distinguished proliferating (P1 and P7) from early postmitotic (P14) GNPs (Fig. 1, C and D). Only 4.9% of the phosphorylated peptides changed in relative phosphorylation level more than twofold from P1 to P7 (3.6% increased, 1.3% decreased); in contrast, 21.4% of phosphorylated peptides changed more than twofold between P7 and P14 (18.2% decreased, 3.2% increased; Fig. 1D). Strikingly, the pattern of protein abundance and phosphorylation in P7 GNP samples was more similar to that of MB cells than to that of either P1 or P14 GNPs, demonstrating strong similarity between peak proliferating GNPs and MB even at the minute-to-minute kinetics of phosphorylation (Fig. 1C, asterisk). Peptides more highly phosphorylated in P1, P7, and MB relative to P14 included known pro-proliferative phosphorylation sites of the cell cycle regulator retinoblastoma 1 (RB1) ( fig. S2) and the proteins encoded by three of eight genes that commonly undergo copy number aberrations in human MBs (Gab1, Irs2, and Yap-1) (15).

CK2 substrates are more highly phosphorylated in GNPs isolated at P7
Motif analysis of phosphorylation events in GNPs that changed in frequency (>1.5 SD) over time suggested candidate kinases whose activity might be different during proliferation versus differentiation. Sixteen phosphorylation consensus motifs were prominent among 1522 protein sequences that had phosphorylation changes ( Fig. 2A). Four observed motifs [for cyclin-dependent kinases (CDKs), protein kinase A (PKA), and PKC] were consistent with past studies (4,10,16). Of the remaining motifs, 7 of 12 matched those targeted by the protein kinase CK2 (casein kinase 2)-the target motifs of which have acidic residues C-terminal to a phosphorylated serine or threonine

of 14
( Fig. 2, A to C, and table S2) (17). Notably, of 278 phosphorylated peptides with possible CK2 target motifs, 83.8% were more highly phosphorylated in P7 compared to P1 or P14. CK2 itself was not more highly expressed at P7 compared to P1 or P14 in terms of either protein or transcript abundance (18), which is likely why CK2 has not been previously identified in large-scale transcriptional assays of cerebellar development and MB (18,19). In keeping with this, many CK2 substrates did not change in phosphorylation abundance (table S1). Our findings therefore likely reflect regulated and localized change specifically in CK2 activity, which could have resulted from changes in kinase localization and/or trafficking of its subunits, co-regulator binding, phosphatase abundance, or accessibility of substrate motifs or docking sites (20). This highlights the importance of investigating kinase activity at its terminal readout: proteome-wide assessment of phosphorylation.

CK2 inhibition results in decreased GNP proliferation
Many peptides with CK2 consensus sites were more highly phosphorylated at the peak proliferative period (P7), suggesting that CK2 activity is important for GNP proliferation. We tested the role of CK2 in cerebellar development by injecting mice with the CK2 inhibitor 4,5,6,7-tetrabromobenzotriazole (TBB) (Fig. 3A) (21). Although several CK2 inhibitors exist, TBB was used because it is fairly selective (22), is cell-permeable (23), lacks short-term cytotoxicity in vivo (24), and is effective at preventing seizure activity in mice (24). Its potency suggests blood-brain barrier (BBB) permeability. Mice were treated twice a day, from P3 to P7, with TBB [30 mg/kg, intraperitoneally (ip)] (25) or vehicle control [dimethyl sulfoxide (DMSO)]. Four hours after the final injection, mice were sacrificed and their cerebella were fixed. TBB-treated mice had no change in cerebellar folia length, which is affected by non-GNP-derived neurons (Fig. 3, B and C) (26). In contrast, folia width, which depends on the number of proliferating GNPs and their progeny, granule neurons, was significantly reduced by TBB treatment (P < 0.01; Fig. 3, B and C). The number of Atoh1-positive (P < 0.01) and proliferating GNPs at P7 (P < 0.001) was significantly reduced in treated mice (Fig. 3, B and D). No significant change in the rate of apoptosis or the ratio of proliferation to differentiation was detected (Fig. 3, B and D). Similarly, GNPs cultured in the presence of SHH and CX-4945, a CK2 inhibitor that is highly specific and structurally distinct from TBB, had a dose-dependent decrease in proliferation and a correlating decrease in total Atoh1positive cells at 24 hours ( fig. S3), as assessed by single-cell immunofluorescence imaging. The decrease in folia width was likely to be due to reduced GNP proliferation in response to inhibition of CK2 during the critical postnatal period.

CK2 facilitates Hh signal transduction
Because GNP proliferation in neonatal mice is driven by Hh signaling, we investigated whether CK2 affects the response to SHH. Hh signal transduction is initiated when SHH ligand binds and inhibits the cell surface receptor PTCH1 (Fig. 4A) (1,27). SHH inhibition of PTCH1 unleashes the G protein (heterotrimeric GTP-binding protein)-coupled receptor SMO (28,29), which then inhibits a protein called suppressor of fused (SUFU), in turn an inhibitor of the transcription factor GLI2 (30). Inhibition of SUFU thus allows transcription of GLI2 target genes, such as GLI1. Note that CK2 is an acronym derived from the misnomer "casein kinase-2" and is unrelated in both structure and function to the known GLI2 regulator CK1.
Genetic and small-molecule inhibitor studies in NIH3T3 cells (spontaneously immortalized murine embryonic fibroblasts) confirmed a role for CK2 in Hh signal transduction. NIH 3T3 cells are routinely used to assay the Hh pathway because they contain all canonical components of the Hh transduction pathway and have a robust transcriptional response to SHH ligand in culture. In contrast, GNPs in culture exit the cell cycle and rapidly differentiate into non-SHH-responsive granule neurons. A change in Gli1 transcript expression in GNPs can result from perturbation of the Hh pathway or from non-Hh-dependent perturbations, such as cell cycle inhibitors or changes in ion homeostasis, so NIH3T3 cells are used instead.
CK2 is a heterotetrameric holoenzyme made up of four subunits: two catalytic  subunits (either two CSNK2A1 subunits or one CSNK2A1 subunit and one CSNK2A2 subunit) and two regulatory  subunits (CSNK2B). Genetic loss in mice of either CSNK2A1 or CSNK2B results in early embryonic death (31,32). CSNK2A2 can generally be replaced by CSNK2A1, and knockout of Csnk2a2 in mice results in infertility in males but otherwise has no apparent phenotypic consequence (31).
Small interfering RNA (siRNA)-mediated knockdown of the essential catalytic  subunit, CSNK2A1, in NIH3T3 cells strongly reduced the induction of Gli1 expression by SHH (Fig. 4B). As expected,   the effect of knockdown of only the nonessential  prime subunit, CSNK2A2, on Gli1 transcription was not significant. Knockdown of CSNK2B resulted in a significant but modest decrease in Gli1 induction, consistent with previous evidence (32) that even low CSNK2B protein abundance, as in Csnk2b knockdown cells ( fig. S4), is sufficient to make substantial amounts of CK2 holoenzyme. Each of the mRNA knockdowns was subunit-specific and did not reduce expression of the non-Hh target Smo (fig. S4).
Two structurally unrelated CK2 inhibitors, CX-4945 and TBB (21,33), each significantly reduced the expression of the SHH-induced GLI2 targets (Gli1, Gli2, and Ptch1) in NIH3T3 cells as early as 4 hours after drug administration (Fig. 4C). CK2 inhibition did not decrease the transcript abundance of the non-Hh target gene Smo (fig. S5). The rapid influence of CK2 inhibitors on Hh pathway target gene expression suggests that the slower (72-hour) effects of CSNK2A1 knockdown in NIH3T3 cells were due to direct action on Hh signal transduction rather than indirect effects from prolonged CK2 inhibition.

CK2 facilitates Hh signal transduction at late steps in the pathway
When NIH3T3 cells were stimulated with an SMO agonist "SAG" (100 nM) (34), the CK2 inhibitor TBB blocked the increase in Gli1 transcript that normally occurs in response to SAG, working as well as the SMO antagonist GDC-0449 (100 nM; also known as vismodegib) (Fig. 4D) (35). Therefore, CK2 regulates Hh signal transduction at or downstream of SMO. Treatment of Sufu −/− mouse embryonic fibroblasts (MEFs) (36) with TBB also strongly blocked transcription of Gli1 (Fig. 4E), indicating that CK2 is needed at late steps of the Hh signal transduction pathway, at or downstream of SUFU. Treatment with the SMO antagonist GDC-0449 (100 nM) did not block transcription of Gli1 in Sufu −/− cells, as expected, because SMO is upstream of SUFU in the pathway.
Our experiments demonstrating action of CK2 at a late step of the Hh pathway are consistent with previous work showing that the CK2 inhibitor TBB affects GLI2 protein stability, likely through regulation of GLI2 ubiquitination (37,38). In keeping with this, TBB reduced the amount of endogenous full-length GLI2 expressed in NIH3T3 cells (Fig. 4F), as previously shown (37). TBB did not affect the abundance of an N-terminal-deleted form of Gli2 (HA-Gli2N, hereafter simply Gli2N; Fig. 4F), stably transfected into NIH3T3 cells (39). Gli2N lacks sites required for ubiquitin-dependent degradation (40), remains constitutively nuclear, and normally causes constitutive induction of the GLI2 target gene Gli1. However, treatment with TBB still substantially lowered Gli1 mRNA expression in Gli2N-transfected cells (Fig. 4G). Together, the results suggest that CK2 regulates two steps in the Hh pathway: to stabilize GLI2 itself and to enhance the ability of GLI2 to activate its transcriptional targets. Our phosphoproteomic data point to several potential candidate substrates that might mediate effects of CK2 on GLI2-mediated transcription. For example, phosphorylated CK2 sites were found within enhancerlooping proteins [such as RING finger protein 1 (RING1), nipped-Blike protein (NIPBL), and mediator complex subunit 1 (MED1)] and epigenetic readers and writers [such as Polycomb repressive complex 1 (PRC1), PRC2, and histone deactylase (HDAC) components]. In two cases, the CK2 sites are well-established activating phosphorylation sites (in HDAC1 Ser 393/421 and HDAC2 Ser 294/422 ). CK2 may regulate accessibility of GLI2 target gene promoters or their interactions with enhancers.

CK2 inhibitors restrict the growth of mouse and human SHH MB cells in culture
SHH MBs occur because of sustained Hh target gene activity and require persistent action of the Hh signal transduction pathway for survival. Therefore, we reasoned that CK2 inhibitors may be effective of SHH MB recurrence (45), were the most sensitive cell population to treatment with CX-4945 (11.1% relative decrease in G 0 cells) but the least sensitive population to treatment with vismodegib (5.3% relative increase in G 0 cells; Fig. 5D). Control cell lines derived from diffuse intrinsic pontine glioma (DIPG), another aggressive pediatric brain tumor that is not Hh pathway-dependent, did not die in response to CX-4945 (Fig. 5E). were injected into the flanks of mice, and the resulting tumors were allowed to grow to 250 to 300 mm 3 (Fig. 6A). A parallel experiment was done using implants of Ptch +/− ;Tpr53 −/− MB cells, which lack the activating SMO mutation and are therefore susceptible to vismodegib. Preclinical work for the two FDA-approved Hh inhibitors (GDC-0449 and LDE225) relied on mouse flank allografts using the same cell lines as those used here (40,46,47). These mouse cells reportedly closely resemble human MB cells with respect to gene expression (41), response to treatment (40,(47)(48)(49), and even the spontaneous occurrence of treatment-induced, drug-resistant mutations (35,46). Treatment of mice harboring flank allografts with the CK2 inhibitor TBB [30 mg/kg, ip twice daily (BID)] significantly inhibited MB growth as early as 2 days after initiating treatment and, in vismodegib-resistant tu-mors, near-complete cessation of tumor growth (Fig. 6, B to D). CK2 inhibition prolonged survival in mice with cerebellar MB. Eighty thousand Ptch +/− ;Tpr53 −/− ;SmoD477G mouse MB cells per mouse were injected into cerebella of NSG mice (Fig. 6A). Seventy-two hours after tumor cells were injected, a collaborator who was blinded to treatment randomized the mice into two groups and began administering either the CK2 inhibitor CX-4945 (37.5 mg/kg) or vehicle control (DMSO). CX-4945 was used because it is the only CK2 inhibitor optimized for clinical use, is provided orally, and has been shown to be safe in humans with solid tumors (50). In agreement with previous trials using these cells (51), 100% of control mice died because of tumors by day 17; in contrast, 43% of mice treated with CX-4945 survived past 100 days, despite termination of treatment at day 30 ( Fig. 6E). CX-4945 was well tolerated, because body weights were similar between cohorts throughout the duration of treatment ( fig. S7).

CK2 expression correlates with 5-year survival in patients with SHH MB
Human clinical data were consistent with the idea of using CK2 inhibition to treat MB. Kaplan-Meier survival analysis has revealed that patients with SHH-MB who had low expression of mRNA encoding the main CK2 catalytic subunit  (CSNK2A1) in their tumors had a mean 82% 5-year survival, whereas patients with high expression of CSNK2A1 had a mean 61% 5-year survival (Fig. 6F). No correlation between CSNK2A1 expression and patient survival was observed in other MB subgroups (group 3, group 4, and WNT; fig. S8), which have been previously shown to not require Hh pathway activity for tumor growth (3).

Single mutation in CK2 causes resistance to TBB but not CX-4945
To test the potential for rapid emergence of resistance to CK2 inhibitors, four Ptch +/− mouse MB cell lines (MB21, MB53, MB55, and MB56) were serially passaged for 1 month in the presence of TBB In wild-type CK2, the binding of TBB requires interaction with both the hinge region and Lys 68 (53). Our substrate-specific docking experiments revealed that, when TBB was inside the enlarged cavity of the D175H mutant protein, TBB was too small to efficiently interact with both the hinge region and Lys 68 (Fig. 7B). In contrast, CX4945 spanned the entire catalytic site and maintained the same crystallographic binding pattern as in the wild-type protein (Fig. 7C), anchoring both the hinge region at 2.94 Å and to Lys 68 at 2.80 Å through hydrophilic interactions. These data indicate that the efficacy of CX-4945 in blocking growth of the MB cell lines-with wild-type and mutant CK2 alikemay be through direct binding-mediated inhibition of CK2.

DISCUSSION
By screening the phosphoproteome during developmental periods when GNPs switch from proliferation to differentiation, we identified CK2 as having increased activity on multiple substrates, specifically in the proliferative, MB-like, P7 stage. The work presented here has already directly led to a phase 1/2 study investigating the use of the CK2 inhibitor CX-4945 in patients with SHH MB. The rapid translation from a phosphoproteomic screen to a clinical trial reflects favorable aspects of the approach to cancer drug target identi-fication. Traditional approaches to discovering novel drug targets often rely on identification of mutations that appear within tumors. However, tumor cells carrying any mutation that is advantageous to cell growth and survival will come to predominate, even if that advantage is small. A great challenge is to determine the biological function of each mutated protein in the context of any one tumor and then distinguishing proteins with activity only slightly advantageous to tumor growth from those that are essential to survival of the tumor. Drugs that target the former have only incremental benefit to patient survival, whereas the latter would be potentially clinically transformative. In contrast, a better understanding of proteins that regulate proliferation versus differentiation in the developmental lineage from which the tumor arose, as provided by our developmental phosphoproteomic approach, can identify important tumor drivers and their likely biological function in proliferation. The critical role of CK2 in GNP development may be the reason why CK2 inhibition resulted in robust tumor stasis and long-term regression of SHH MB but had less robust, often transient, benefit in many other cancer types (54,55). Another reason for the rapid path to clinic was the identification of a drug target that is readily conducive to small-molecule inhibition. Kinases, as would be identified through our phosphoproteomic approach, are often readily inhibited by small molecules that compete with ATP binding in the activation cleft. CK2 had many commercially available small-molecule inhibitors and a human-ready compound. This allowed us to bypass lengthy and costly drug development work as would be required for many other protein classes that have undesirable features of a drug target (presence of metal ions, lack of hydrogen bond donors and acceptors, need for adaptive changes in conformation, lipophilicity, and featureless binding sites) (56).
The many potential roles of CK2 in MB, and its influence on two late steps in SHH signal transduction, make it unlikely that a mutation in any one substrate will result in drug resistance, which is another common limitation of many targeted therapies. As we demonstrated, mutations within CK2 itself can still cause resistance. Yet, unlike offtarget resistance resulting from mutation of any one of many proteins, predictable resistance due to a mutation in CK2 itself can be tackled through combinations of CK2 inhibitors and/or rational drug design. The quantitative developmental phosphoproteomics approach that we used enabled identification of a previously unknown regulatory mechanism of the Hh signaling pathway and a potential new treatment for this devastating pediatric cancer. Similar approaches may work for other cancers where the developmental cell of origin can be found.

Animal experiments: Statistics, blinding, and randomization
Given that CK2 promotes Hh signal transduction, we estimated sample size on the basis of previous work performed with Hh inhibitors.
Given the dramatic SHH MB tumor regression previously demonstrated with the Hh inhibitor vismodegib (57), we conservatively predicted that CK2 inhibitors would have an effect size of at least 2.2. Therefore, if  error = 0.05 and power = 0.85, then five animals are required per condition. To allow for potential exclusion of one to two animals per condition because of poor tumor engraftment, we used seven to eight animals per condition. Ultimately, we had 100% engraftment [and which met the inclusion criteria (tumor = 250 to 300 mm 3 )]; thus, all animals were included in the analyses of the tumor growth experiments. Mice were randomized into control or treatment groups daily. Daily randomization ensured that quicker growing tumors were equally represented in both groups. For cerebellar tumor studies, mice were randomized by a blinded collaborator 72 hours after implantation. A blinded collaborator treated and recorded weight, morbidity, and survival. Prism statistical analysis software was used (GraphPad Software) to calculate significance as determined by two-tailed t tests for each paired data point (Fig. 6, B and C) or log-rank Mantel-Cox test (Fig. 6E) for survival studies. All studies were performed in compliance with institutional regulations. A previous study investigating the effect of Hh inhibition in the developing cerebellum showed that mice with inactive Hh signaling have about 50% reduction in the EGL size, with an effect size of about 5.7 (control: about 80, SD = 7; treatment: 40, SD = 7) (6). Therefore, we conservatively predicted that CK2 inhibitors would be at least 25% as potent as Hh inhibitors, giving an effect size of 1.4. Therefore, with  error = 0.05 and power = 0.85, we expected that we would need and therefore used 10 animals per condition for the developmental experiments. Mice were randomized to treatment or control group at P3, with each litter being equally divided between the two groups. Images were randomized, and all measurements (length, width, Atoh-1, Tag1, and TUNEL staining) were performed by a blinded collaborator. Cerebellar folia length was compared between matched sections of control and treated mice by measuring the maximal length from the mid-base to the tip of the culmen (lobule IV/V). Culmen width was measured 600 m from the base of the folia between lobules V and VI. Prism statistical analysis software was used (GraphPad Software) to calculate significance as determined by two-tailed t tests. For the Hh epistasis assays specifically, all measurements were taken in biological triplicate, and Prism statistical analysis software was used (GraphPad Software) to calculate significance as determined by twotailed t tests.

Human Kaplan-Meier curves
Kaplan-Meier survival curves on the human MB data sets were generated using GraphPad Prism, and significance was measured by log-rank Mantel-Cox test.

Sample preparation for quantitative MS analysis
GNP samples were prepared as previously described (60), with the following modifications. All solutions are reported as final concentrations. Lysis buffer [8 M urea, 1% SDS, 50 mM tris (pH 8.5), protease and phosphatase inhibitors from Roche] was added to the cell pellets to achieve a cell lysate with a protein concentration between 2 and 8 mg/ml. A micro-BCA (bicinchoninic acid) assay (Pierce) was used to determine the final protein concentration in the cell lysate.
Proteins were reduced and alkylated as previously described. Proteins were precipitated using methanol/chloroform. In brief, four volumes of methanol were added to the cell lysate, followed by one volume of chloroform, and finally three volumes of water. The mixture was vortexed and centrifuged to separate the chloroform phase from the aqueous phase. The precipitated protein was washed with one volume of ice-cold methanol. The washed precipitated protein was allowed to air dry. Precipitated protein was resuspended in 4 M urea, 50 mM tris (pH 8.5). Proteins were first digested with LysC (1:50; enzyme/protein) for 12 hours at 25°C. The LysC digestion was diluted down to 1 M urea, 50 mM tris (pH 8.5) and then digested with trypsin (1:100; enzyme/protein) for another 8 hours at 25°C. Peptides were desalted using a C 18 solid-phase extraction cartridges, as previously described. Dried peptides were resuspended in 200 mM EPPS [4-(2-hydroxyethyl)piperazine-1-propanesulfonic acid] (pH 8.0). Peptide quantification was performed using the micro-BCA assay (Pierce). The same amount of peptide from each condition was labeled with tandem mass tag (TMT) reagent (1:4; peptide/TMT label) (Pierce). The 6-and 10-plex labeling reactions were performed for 2 hours at 25°C. Modification of tyrosine residue with TMT was reversed by the addition of 5% hydroxyl amine for 15 min at 25°C. The reaction was quenched with 0.5% trifluoroacetic acid, and samples were combined at a 1:1 ratio. Combined samples were desalted and offline-fractionated into 24 fractions, as previously described.

Liquid chromatography-MS3 spectrometry
Twelve fractions (every other one) of the 24-peptide fraction from the basic reverse-phase step were analyzed with a liquid chromatography (LC)-MS3 data collection strategy on an Orbitrap Fusion mass spectrometer (Thermo Fisher Scientific) equipped with a Proxeon EASY-nLC 1000 for online sample handling and peptide separations. About 5 g of peptide resuspended in 5% formic acid + 5% acetonitrile (ACN) was loaded onto a 100-m-inner diameter fused-silica microcapillary with a needle tip pulled to an internal diameter less than 5 m. The column was packed in-house to a length of 35 cm with a C 18 reverse-phase resin (GP118 resin 1.8 m, 120 Å; Sepax Technologies). The peptides were separated using a 120-min linear gradient from 3 to 25% buffer B (100% ACN + 0.125% formic acid) equilibrated with buffer A (3% ACN + 0.125% formic acid) at a flow rate of 600 nl/min across the column. The scan sequence for the Orbitrap Fusion began with an MS1 spectrum [Orbitrap analysis, resolution 120,000 (400 to 1400 m/z scan range); AGC target, 2 × 10 5 ; maximum injection time, 100 ms (dynamic exclusion of 75 s)].

LC-MS3 data analysis
A suite of in-house software tools was used for .RAW file processing and controlling peptide and protein-level FDRs, assembling proteins from peptides, and protein quantification from peptides, as previously described. MS/MS spectra were searched against a UniProt human database (accessed February 2014), with both the forward and reverse sequences. Database search criteria are as follows: tryptic with two missed cleavages, a precursor mass tolerance of 50 parts per million (ppm), fragment ion mass tolerance of 1.0 Da, static alkylation of cysteine (57.02146 Da), static TMT labeling of lysine residues and N termini of peptides (229.162932 Da), and variable oxidation of methionine (15.99491 Da). TMT reporter ion intensities were measured using a 0.03-Da window (6-plex) or 0.003-Da window (10-plex) around the theoretical mass/charge ratio (m/z) for each reporter ion in the MS3 scan. Peptide spectral matches with poor-quality MS3 spectra were excluded from quantitation (<100 summed signal-to-noise ratio across 6 channels and <0.5 precursor isolation specificity for 6-plexes or <200 summed signal-to-noise ratio across 10 channels and <0.5 precursor isolation specificity for 10-plexes).

LC-MS3 sample comparison
Only peptides identified with full confidence were included in the analysis (binomial probability threshold of P < 10 −6 , occurrence threshold = 20). Each biological replicate was median-adjusted to 10 to account for differences in sample loading, resulting in the following changes: proteomic data set: :P14 ratios were determined, and phosphopeptide changes were normalized to protein changes. Values outside of 1.5 SD were considered "significant changers." Motif analysis of significantly changing phosphopeptides was performed using Motif-X, as previously described (61,62). Briefly, Motif-X is an iterative statistical approach to identifying protein phosphorylation motifs in large-scale phosphoproteomic data sets built on a greedy recursive search of the sequence space to identify highly correlated residue/ position pairs with the lowest P values. Here, a binomial probability threshold of P < 10 −6 and occurrence threshold of 20 were used.

GNP proliferation
To determine relative proliferation of Atoh1(+) GNPs at P1, P7, and P14, cerebella were dissected from Atoh1/nGFP transgenic mice (63) 1 hour after EdU injection [50 mg/kg using a stock (5 mg/ml) diluted in phosphate-buffered saline (PBS)]. Brains were dissected and fixed in 4% paraformaldehyde (PFA) overnight and then transferred into 20% sucrose for 24 hours. Fixed whole cerebella were mounted in optimal cutting temperature (OCT) embedding medium, sectioned at 20 m, and then air-dried for 20 min before storing at −20°C for up to 2 months. Sections were blocked in 0.2% Triton X-100 and 2% goat serum diluted in sterile PBS for 1 hour at room temperature. EdU staining was performed as per the manufacturer's instructions (Click-iT Plus EdU Alexa Fluor 647 Imaging Kit, catalog no. C10640, Life Technologies) and counterstained with Hoechst 33258 (final concentration, 1 l/ml in PBS) at room temperature for 10 min. Images were taken on a Leica TCS SP8 confocal microscope. Total EdU signal within Atoh1(+) regions was quantified using ImageJ (six cerebella per time point, four to seven folia per cerebellar section) after background subtraction via rolling ball radius (50 pixels). Prism statistical analysis software was used (GraphPad Software) to calculate significance as determined by two-tailed t tests.

Single-cell immunofluorescence imaging (GNP and MB)
Before cell culture, 96-well glass imaging plates (Cellvis P96-1.5H-N) were incubated with poly-d-lysine (A-003-E, Millipore) at 100 g/ml for 2 hours followed by laminin (CC095, Millipore) at 10 g/ml overnight at 37°C. GNPs or MB cells were plated at 1 × 10 5 cells per well and grown at 37°C. After treatment, the cells were fixed with 4% PFA for 10 min at room temperature. The cells were then blocked using 5% donkey serum, 1% BSA, and 0.2% Triton X-100 for 1 hour at room temperature. Primary antibodies (Abs) were incubated overnight at 4°C. The primary Abs used are as follows: anti-NeuN mouse monoclonal at 1:200 (MAB377, Millipore) and phospho-Rb (Ser 807/811 ) rabbit monoclonal at 1:1000 (clone D20B12, 8516, Cell Signaling Technology). Cells were costained with rhodamine phalloidin (R415, Molecular Probes). Donkey anti-immunoglobulin G (IgG) secondary Abs against mouse and rabbit conjugated to Alexa Fluor 488 and Alexa Fluor 647 at 1:500 (Jackson ImmunoResearch) were used. All cell imaging was performed using the ImageXpress Micro XLS Widefield High Content Screening System (Molecular Devices) using 20× (0.45 or 0.75 numerical aperture) Nikon objectives. The intensity of fluorescence in each cell was automatically calculated using custom MATLAB scripts: Nuclei were segmented using 4′,6-diamidino-2-phenylindole, as previously described (64); downstream analysis was performed in R; cell cycle classification was performed using Gaussian finite mixture modeling with the mclust package (65); and cell cycle clustering analysis in GNPs used the log 10 -transformed median intensity of each cell, measuring Math1-GFP, phosphorylated Rb, and NeuN, as well as the area of the cell. All studies were done with six individual cultures.

In silico analysis of mutant versus wild-type binding of TBB and CX-4945
All the crystal structures of CK2 were retrieved from the Protein Data Bank (PDB). To perform docking and MD simulations, the crystal structure of human CK2 in complex with CX4945 was used (PDB code: 3PE1). However, to compare the binding motif of TBB and to evaluate its position in the structure 3PE1, the crystallographic complex TBB/CK2 was also considered (PDB code: 1J91). The crystal structures were processed to remove unwanted ligands and water molecules. Hydrogen atoms were added to the protein structure using standard geometries (66). To minimize contacts between hydrogens, the structures were subjected to AMBER99 force-field minimization until the root mean square of conjugate gradient was <0.1 kcal mol −1 Å −1 (1 Å = 0.1 nm), keeping the heavy atoms fixed at their crystallographic positions (66). A similar protocol was used also to minimize the structures subjected to in silico mutagenesis. To strictly calibrate the high-throughput docking protocol, a small database of known CK2 inhibitors was built and a set of docking runs was performed. After the calibration phase, TBB and CX-4945 were docked directly into the ATP-binding site of selected CK2 crystal structures, by using AutoDock software (66). MD simulations of the considered structures (parameterized with AMBER99) were performed with NAMD 2.8 (67) to verify their stability over time; in particular, a 100-ns NPT (isothermal-isobaric) ensemble (1 atm, 300 K) MD simulation was performed after an equilibration phase of 1 ns (positional restraints were applied on carbon atoms to equilibrate the solvent around the protein) (68,69).

Flank allograft studies
In vivo assays of CK2 inhibitor efficacy were performed in accordance with protocols approved by the Institutional Animal Care and Use Once tumors reached 250 to 300 mm 3 , mice were randomized to treatment versus control groups. Treatment groups received TBB (30 mg/kg, ip BID; reconstituted in DMSO at 50 g/l) or GDC-0449 (37.5 mg/kg, ip BID; reconstituted in DMSO at 62.5 g/l) (S1082, Selleckchem). Control mice were treated with an equivalent volume of DMSO. Corn oil was used in all groups to a total volume of 180 l per injection, and all doses were given with a 28-gauge insulin syringe. Tumors were measured, and mice were weighed daily. Once tumors reached 1.5 cm 2 or met institutional euthanasia criteria, mice were sacrificed and tumors were harvested. Growth curves were generated using GraphPad Prism, and significance was measured by two-way analysis of variance (ANOVA).

Cerebellar allografts
Cells from Ptch +/− ;Tpr53 −/− ;SmoD477G mice were suspended in sterile PBS to an adjusted final concentration of 80,000 cells per 2-l injection. NOD.Cg-Prkdc scid Il2rg tm1Wjl /SzJ (NSG) mice (6 to 7 weeks old; The Jackson Laboratory) were anesthetized and fixed to a stereotactic frame, and the head was sterilized. A small sagittal incision was made in the mouse's scalp to expose the area surrounding lambda. The skull was sterilized, and a 0.7-mm drill bit was used to create a midline hole located 2 mm down from lambda. A Hamilton syringe with affixed 26-gauge needle was lowered to a depth of ~3 mm and then withdrawn to 2.75 mm before injecting 2-l (80,000) cells at 0.5 l/min. The drill hole was closed with a thin slip of bone wax, and the wound was closed. Mice were randomized, and a blinded collaborator began treatment at 72 hours with either CX-4945 (37.5 mg/kg, orally BID; reconstituted in DMSO) or DMSO (equivalent volume to treatment group, dosed ip BID). Corn oil was used as vehicle for both CX-4945 and DMSO. Weight, morbidity, and survival were measured by a blinded collaborator.

Mouse strains
All animal experiments were performed in accordance with protocols approved by the Institutional Animal Care and Use Committee at Stanford University. Male and female wild-type CD1-Elite mice (strain 482, Charles River) were sacrificed at P1, P7, and P14 for initial phosphoproteomic study. Female and male Atoh1/nGFP transgenic mice (63) were used for cerebellar development studies. Female Nu/ Nu mice (6 to 7 weeks old; strain 088, Charles River) were used for flank studies, and female NOD.Cg-Prkdcscid Il2rgtm1Wjl/SzJ (NSG) mice (6 to 7 weeks old; strain 005557, The Jackson Laboratory) were used for cerebellar tumor studies.