An ABC Transporter Drives Medulloblastoma Pathogenesis by Regulating Sonic Hedgehog Signaling

Medulloblastoma (MB) is the most common malignant pediatric brain tumor for which standard of care therapy includes tumor resection followed by radiation and chemotherapy (1). However, in children, the use of craniospinal irradiation has been deemed unacceptable due to the developmental side-effects and chemotherapeutic regimens have been prioritized. In the sonic hedgehog (SHH)-MB, patients with TP53 mutations and MYCN amplification exhibit poor clinical outcome, likely due to de novo resistance, representing the highest risk form of SHH-MB (2). Although initially promising, therapy targeted at the key SHH activator, smoothened (SMO), was shown ineffective as SHH-MB acquired therapy-induced SMO mutations, producing drug resistance and ultimately relapse (3, 4). An additional liability that limited the broad implementation of the SMO inhibitor, vismodegib was that children exposed to prolonged treatment developed irreversible growth plate fusions (5). As SHH cancers frequently harbor tumorgenic SHH pathway mutations downstream of SMO (e.g., SUFU; ref. 6), new targets amenable to manipulation are needed. To identify candidate new regulators that restrain or block the SHH pathway, we screened publicly available data sets to identify new candidate modulators of SHH-MB. The studies described herein identify a previously unknown regulator of the SHH pathway, an ATP-Binding Casette efflux transporter, ABCC4.

WT and Abcc4^−/− NIH3T3 cells were cultured in DMEM containing 10% bovine calf serum, 0.05 mg/mL penicillin and streptomycin, 2 mmol/L glutamine, 1 mmol/L sodium pyruvate, and 0.1 mmol/L nonessential amino acids (NEAA). Light2 cells were cultured in the same media and 0.4 mg/mL G418 and 0.15 mg/mL zeocin. Ptch1^−/−and SufuKD mouse embryonic fibroblast (MEF) cells were cultured in DMEM containing 10% FBS, 0.05 mg/mL pen/strep, and 2 mmol/L glutamine. DAOY cells were cultured in MEM/EBSS, 10% heat-activated FBS, sodium pyruvate, NEAA, 2 mmol/L glutamine, 1× pen/strep; UW228.3 cells were cultured in DMEM/F12,10% heat-activated FBS, 2 mmol/L glutamine, 1× pen/strep. NIH3T3 were obtained from ATCC, DAOY, and UW228.3 cells were from Dr. Clinton Stewart, SufuKD cell line from Dr. Young-Goo Han, Ptch1^−/− MEFs, Light2, and HEK293-SHH producing cell line are generous gift of Dr. Philip Beachy. Cells were used as described without further authentication from authors. Cells were Mycoplasma-free, tested using LookOut Mycoplasma PCR Detection Kit (Sigma), and used within 20 passages of thawing.

Mouse models of SHH-MB were Ptch1^+/−, Cdkn2c^−/− and Ptch1^+/−, Trp53^−/− were described previously (7). Human patient-derived orthotopic xenografts (PDOX) of human SHH MB, TB-13-5634 (MYCN amplified, TP53 mutation) and TB-14-1796 (MYCN amplified, TP53 mutation, PTCH1 deletion) were described previously (8). Human SHH-MBs tumor sections SC-18-3143 and SC-19-1060 were primary tumors from St Jude children's Research Hospital.

Tumor cells were orthotopically implanted into the cortex of female CD-1 nu/nu mice (6–12 weeks old). Animals were subsequently monitored daily for symptoms including doming of the head, ataxia, and reduced activity. Moribund mice were humanely sacrificed, and tumors were isolated. Animal experiments were performed in accordance to and approved by St. Jude Children's Research Hospital Animal Care and Use Committee.

Gene expression and heatmaps were extracted from Oncomine (www.oncomine.org). Survival data of patients with MB and curves were extracted from R2 genomics analysis and visualization platform (http://r2.amc.nl).

The network-based integrative NetBID (9) algorithm was applied to identify “hidden” drivers in SHH subgroup using gene expression profiles from patients with MB. A MB-specific interactome (MBi) was reverse-engineered using published microarray data (10–13) from platform HG-U133_Plus_2 (total 383 samples) against 1,621 transcription factors and 5,791 signaling proteins. The batch effects from different data sets was removed by using the “removeBatchEffect” function in limma (14). The data-driven MBi resulted in 1,567,381 edges and 26,126 nodes. NetBID was then applied (signed = TRUE) on comparing hub gene inferred activity between SHH versus all the other subgroups from the cohort GSE85217 (15). Finally, we subtracted the network of our hub genes of interest GLI1/2/3, ABCC4, and PRKACA from MBi and visualized their interconnections by Cytoscape. The NetBID package can be found online at: https://github.com/jyyulab/NetBID.

NetBID was used in gene set enrichment analysis: MBi network enabled prediction of regulons as gene sets and summarized differential gene expression profile between SHH versus other subgroups from cohort GSE85217. Differential activity P values were used as an enrichment score of each regulon. Gene set enrichment analysis (GSEA) results were verified by two other packages: JavaGSEA and fGSEA, with 1,000 permutations and default setting parameters.

Cells were transfected as described in Supplementary Materials and Methods. Twenty-four hours posttransfection, media was replaced with media containing 0.5% serum media with agonists (200 nmol/L SAG, SHH-conditioned media, or 10 μmol/L 20-OHC) for 16 hours unless indicated otherwise. Cylohexamide was used at 100 μg/mL in 0.5% serum media.

Cells were lysed in modified RIPA buffer (50 mmol/L Tris-Cl pH 7.4, 300 mmol/L NaCl, 2% NP-40, 0.25% deoxycholate) supplemented with 10 mmol/L N-etyl-malemaide (NEM), 1 mmol/L phenylmethylsulfonyl fluoride (PMSF), 1× EDTA-free protease inhibitor cocktail, 1× phosphatase inhibitor, and 10 μmol/L MG132 on ice for 30 minutes. Lysates subjected to SDS-PAGE and transferred to PVDF membrane. Immunoblot analysis was performed with the antibodies indicated in the figures and Supplementary Materials and Methods.

RNA was isolated (RNAeasy Kit) cDNA prepared by iScript reverse transcription reagent. Actin was used as the normalization control. Relative gene expression was calculated using the 2^−ΔΔC_t method.

Nuclear and cytosolic fractions were isolated using NE-PER (Thermo Fisher Scientific) nuclear and cytosolic extraction kit.

Light2 cells (16) were treated siRNA added by reverse transfection. After 24 hours, medium was replaced with 0.5% serum DMEM and pathway stimulation was initiated using indicated agonists for 48 hours. Gli1 luciferase activity was measured using the dual-luciferase assay system.

Cells were seeded on glass coverslips. The next day, cells were transfected with indicated siRNAs or plasmids. The next day, medium was replaced with 0.5% serum DMEM and agonists were added. At indicated time points, medium was removed, cells PBS rinsed, and fixed in 4% methanol-free formaldehyde. Cells were permeabilized and blocked in 0.2% triton x-100 with 10% normal goat serum, and incubated with primary antibodies. Subsequently, cells were incubated in fluorochrome-conjugated secondary antibodies and mounted on ProLong gold diamond with DAPI.

Cells were seeded onto a two-well Lab-Tek glass dish and transfected with cAMP or PKA sensors as described in Supplementary Materials and Method. Medium was replaced with phenol-red free, 0.5% serum media for 24 hours. Ten μmol/L forskolin and 100 μmol/L IBMX were added after baseline collection. Dual emission ratio imaging was performed using a Marianas spinning disk confocal imaging system on a Zeiss Axio Observer inverted microscope platform using a Zeiss Plan-Apochromat 63× (1.4 NA) oil objective. Images were taken every 60 s with an exposure time of 100 to 200 ms. Data were normalized by setting the emission ratio prior to small molecule addition to 1. % FRET change = (R_max – R_{newsteadystate})/R_max × 100, where R is the emission ratio.

For imaging using PKA sensors, 10 μmol/L H89 was added. AUC of cells expressing plasma membrane PKA sensor was measured by integrating the net decrease in FRET over time from 30 to 60 min using one as the baseline.

Brains, tumors, and spines were formalin-fixed, paraffin-embedded, and sectioned. Five-μmol/L thick sections were stained with hematoxylin and eosin and indicated antibodies (see Supplementary Materials & Methods).

For quantitative intensity measurements, images were acquired and compared with identical gain, laser power, and exposure time. Mean ciliary intensities were measured on Imaris.

Statistical parameters are reported within the figure legends. All statistical analyses, which include Student t test, one- or two-way ANOVA, Mann–Whitney, and Mantel–Cox test were calculated using GraphPad Prism 8.

Human MB has been stratified, based on molecular signature, into four primary subgroups, WNT (Wingless), SHH, Group 3, and Group 4. Interrogating a number of publicly available MB datasets, we found ABCC4 was highly expressed in the SHH subgroup (Fig. 1A; Supplementary Figs. S1A and S1B), although its DNA copy number was unchanged (Supplementary Fig. S1C). Gene expression analysis of human MB revealed that ABCC4, unlike other ABC transporters found in the brain, was uniquely coexpressed with SHH pathway genes including GLI1-3 (Fig. 1B). Using a large human MB gene expression dataset (15), we evaluated the link between ABCC4 expression and survival in the primary MB subgroups. High ABCC4 expression was significantly correlated with reduced overall survival of patients with SHH-MB (P = 0.0025; Fig. 1C). Other MB subgroups did not show this relationship (Supplementary Fig. S1D).

The strong upregulation of ABCC4 mRNA in SHH-MB led us to investigate if the GLI transcription factors targeted ABCC4. One of the largest MB datasets (GSE85217) revealed a positive correlation between GLI2 and ABCC4 mRNA levels (Fig. 1D). Notably, ABCC4 expression was the highest in the SHH subgroup, where GLI2 level was also the highest (orange points, Fig. 1D). In parallel with the expression correlation analyses, we used NetBID (9), a data-driven systems biology approach to reconstruct a MBi using microarray gene expression profiles of 383 patients with MB collected from published data sets (10–13) with the same platform ([HG-U133_Plus_2]). MBi identified ABCC4 as a modulator for SHH-MB, with the GLI and ABCC4 regulons being significantly enriched in SHH subgroup from GSE85217 data set (Fig. 1E). In the same data set, GLIs and ABCC4 MBi inferred activities were significantly higher in SHH-MB (Supplementary Fig. S1E) with a strong positive correlation for both inferred activities and mRNA level (Supplementary Figs. S1E and S1F).

Conserved GLI2 binding sites within the ABCC4 gene were identified and coincided with active enhancer marks (H3K27ac peaks) in patients with MB, providing further support for GLI2 contribution to ABCC4 regulation (Supplementary Fig. S1G). The contribution of GLI2 to Abcc4 expression was investigated in Ptch1^−/− MEFs to bypass the impact of SHH ligand-mediated activation. Ptch1^−/− MEFs were treated with GANT61, an SHH pathway inhibitor that blocks GLI1&2-mediated transcription by preventing GLI binding to DNA (17). GANT61 treatment dose-dependently suppressed GLI1 with ABCC4 showing a similar dose-dependent suppression, thus supporting the idea that ABCC4 is upregulated by constitutive SHH activation (Supplementary Fig. S1H).

Because SHH pathway activation is critical for granule neuron progenitors (GNP) proliferation and migration during cerebellar development (18), we interrogated the murine cerebellum during postnatal development (P0, P3, P7, and P13) for ABCC4 protein expression using IHC. At the peak SHH activity in the developing cerebellum, postnatal day 7 (P7), ABCC4 protein was barely detected, but it was expressed in meninges and in the choroid plexus as previously reported (Fig. 1F; ref. 19). ABCC4 expression in the cerebellum was generally very low at the start of SHH signaling (P0) all the way through the termination of signaling (P13; Fig. 1F). In contrast, in two different mouse models of SHH-MB (Ptch1^+/−;Cdkn2c^−/− and Ptch1^+/−;Trp53^−/−; ref. 7), ABCC4 was robustly expressed at the plasma membrane (Fig. 1G). Furthermore, strong ABCC4 immunoreactivity was observed in two PDOX of SHH-MB (8) and in primary tumor samples from SJCRH patients with SHH-MB (Fig. 1H and I). These results showing high expression of ABCC4 in SHH-MB are consistent with regulation by GLI transcription factors.

To determine if the amount of ABCC4 affects the canonical, ligand-activated SHH signaling, we overexpressed human ABCC4 in NIH3T3 cells, an SHH-responsive model. Enforced expression of ABCC4 alone is not sufficient to relieve PTCH1 repression of SMO to activate the pathway (Fig. 2A). In contrast, with an agonist that binds SMO (SAG), ABCC4 potentiates GLI1 expression in a dose-dependent manner (Fig. 2A).

As ABCC4 overexpression enhances SHH output, we tested if the converse was true: Does Abcc4 absence impair SHH transcription? Abcc4 knockdown by siRNA decreased Gli1 transcription as determined with a cell line expressing Gli1 reporter construct (Light2 cells; Fig. 2B). Furthermore, using various agonists, we investigated where Abcc4 affects the SHH pathway. Wild-type (WT), Abcc4 siRNA-transfected NIH3T3, or Abcc4-null cells were treated with SHH-conditioned media to activate the pathway at the level of PTCH1. SAG and 20-OHC act downstream of PTCH1 by directly binding to SMO. Irrespective of the agonist's mechanism of SHH pathway activation, ABCC4 knockdown or genetic ablation impaired signaling output (GLI1) at the protein (Fig. 2C and D) and mRNA levels (Supplementary Fig. S2A). Furthermore, level of other downstream SHH target genes, including Ptch1 and Ccnd1, were also suppressed by Abcc4-deficiency (Supplementary Fig. S2A). Together, these results suggest ABCC4 is required for optimal SHH pathway activation. Moreover, in a system where basal SHH activity is minimal such as NIH3T3 cells, ABCC4 modulates the pathway, but only after SMO is activated.

Binding of SHH to PTCH1 triggers translocation and enrichment of SMO and GLI3 in primary cilia, which promotes formation of activated GLI3 (20). Blockade of SMO translocation to primary cilia has been shown to disrupt GLI activation (21). In addition, translocation and enrichment of SMO and GLI3 at primary cilia has been reported to occur within minutes of ligand addition (22). We assessed if ABCC4 affected this step. The rate of GLI3 and SMO translocation in Abcc4-null cells was almost identical to WT (Fig. 2E), suggesting the ciliary translocation and enrichment step, a prerequisite for canonical ligand-dependent pathway activation, does not depend on ABCC4.

SUFU associates with GLI3 to restrain its nuclear entry, thereby blocking its ability to activate transcription (23). To determine if ABCC4 genetically interacts with SUFU, we performed knockdown of Sufu in the Light2 cells (10). The resulting high luciferase activity was consistent with the expected Gli1-mediated transcriptional activation that was promoted by the absence of SUFU (Fig. 2F). When Abcc4 knockdown was superimposed on Sufu knockdown, the Gli1 reporter activity was further reduced (Fig. 2F) as was the endogenous Gli1 mRNA, indicating Abcc4 knockdown is not simply affecting the Gli reporter construct (Fig. 2G). In total, these results indicate that ABCC4 modulates the SHH pathway post-ciliary translocation of SMO and GLI3, thereby suggesting it might modulate GLI3-FL conversion to repressor (GLI3-R). Moreover, in the case of aberrant signaling (i.e., absence of SUFU, a gene that when mutated increases the predisposition to MB; ref. 6), ABCC4 appears to act downstream of SUFU (Fig. 2H).

Upon pathway activation, GLI3-A, a labile species, enters the nucleus, acting as a transcriptional activator (22, 23). Whether ABCC4 affects either the kinetics and/or the distribution of nuclear and cytoplasmic GLI3 was determined by subcellular fractionation with the nuclear fraction confirmed by the nuclear marker, Lamin A/C. As expected, the shorter, repressor form of GLI3, GLI3-R was mostly found in the nucleus (Fig. 3A). In the absence of Abcc4, at 16 hours post SAG treatment, GLI3-FL was almost undetectable in the nucleus (requiring a long exposure for detection) unlike WT cells (Fig. 3A). Nuclear GLI3-FL appeared to accumulate time-dependently in WT cells after SAG treatment, whereas in Abcc4-deficient cells, it only appeared transiently at 2 hours post-agonist addition (Fig. 3A and B). GLI3-R distribution was comparable between WT and siRNA-treated cells throughout the time course (Fig. 3B), which indicates that GLI3-R nuclear translocation is not defective in Abcc4-deficient cells. Importantly, Gli3 mRNA level is similar to WT cells at 16 hours, suggestive of a posttranslational effect on GLI3-FL (Supplementary Fig. S2A). Of note, compared with WT, ABCC4 only alters the cytosolic and nuclear distribution of only GLI3-FL with little impact on the distribution of GLI3-R.

Previous study showed that nuclear GLI3-FL is labile with a short half-life (23). Under basal conditions, ABCC4 absence did not alter the half-lives of both GLI3-FL and GLI3-R species compared with WT cells (Fig. 3C). However, upon SHH-activation, ABCC4 loss dramatically reduced the stability of GLI3-FL (Fig. 3D). Re-expression of human ABCC4 to levels similar to WT in Abcc4^−/− cells restored GLI3-FL stability to levels comparable to that of WT (Fig. 3D). SUFU, a known regulator of GLI3-FL stability (23, 24), also had a shorter half-life in the absence of ABCC4, which was also rescued by re-expression of ABCC4. These experiments indicate that ABCC4 modulates the stability of GLI3-FL.

Next, we expressed human ABCC4 in Abcc4^−/− cells to determine if ABCC4 reexpression affects nuclear GLI3-FL and GLI1. ABCC4 expression restored the nuclear levels of both GLI1 and GLI3-FL (Fig. 3E). Because ABCC4 has the ability to export cAMP (25), we evaluated whether elevation of intracellular cAMP generally reduces nuclear GLI3-FL and GLI1 levels. WT cells were treated with forskolin, which by adenylyl cyclase activation elevates intracellular cAMP concentration. Like absence of ABCC4, forskolin treatment blocked GLI1 expression, however in contrast to ABCC4 loss, forskolin treatment increased GLI3-R without affecting nuclear GLI3-FL level (Fig. 3E, last lane). Thus, ABCC4 deficiency promotes nuclear GLI3-FL loss, whereas an overall rise in intracellular cAMP elevation by forskolin, did not, revealing a distinct ABCC4-mediated regulatory mechanism.

We further confirmed ABCC4 affects GLI3-FL level, but not GLI3-R in human SHH tumor cell lines, DAOY and UW228.3 (26, 27). These human cell lines are extensively used to study MB biology. ABCC4 expression was higher in UW228.3 cells compared with DAOY cells (Fig. 3F). ABCC4 knockdown using pooled siRNA reduced GLI3-FL level without changing GLI3-R level in both cells lines (Fig. 3G). Moreover, exposure to ABCC4 inhibitors, MK571 and Ceefourin, also reduced GLI3-FL level (Supplementary Fig. S2B). These findings are consistent with our observation in NIH3T3 cells where ABCC4 regulates GLI3-FL level. Importantly, upon reduction of ABCC4 level, the stem cell marker Nestin was also reduced in both human tumor cell lines (Fig. 3G). Together, these data showed agreement between the mouse and human cellular models that show ABCC4 modulating GLI3 level and having an overall role in SHH pathway.

Because ABCC4 has the ability to export cAMP (28–30), we next determined if ABCC4 modulates SHH signaling by modulating cAMP level. NetBID predicted that the PKA regulon was significantly lower in SHH-MB, which is the opposite of the pattern for the GLI transcription factors and ABCC4 expression (Fig. 4A; Supplementary Fig. S1E). ABCC4 is capable of exporting cAMP in NIH3T3 cells as acute treatment with forskolin and IBMX, reagents that dramatically elevate cAMP level, revealed cAMP export from Abcc4-deficient cells was about two-fold less than NIH3T3 cells (Fig. 4B). Addition of exogenous human ABCC4 to Abcc4^−/− cells restored cAMP export (Fig. 4B). We did not observe a significant change in bulk intracellular cAMP level when SHH pathway was active (Supplementary Fig. S3A) and the amount of PKA was comparable between WT and Abcc4^−/− cells (Supplementary Fig. S3B).

cAMP signaling elicits specific responses to various cellular stimuli. Such specific responses are achieved through compartmentalization of cAMP signaling (31). We hypothesized that ABCC4 might affect the SHH pathway by specifically regulating plasma membrane cAMP. To interrogate cAMP changes at distinct subcellular compartments, we utilized genetically encoded cAMP biosensors (ICUE3 sensors) targeted to the plasma membrane, cytosol, and nucleus (32, 33). Cells expressing these biosensors were treated with forskolin to activate adenylyl cyclase and IBMX to inhibit phosphodiesterases (PDE; Fig. 4C). These treatments allowed us to concentrate on cAMP changes predominantly mediated by ABCC4 efflux, and not degradation by PDEs. Upon treatment, the cyan to yellow emission ratio was increased, indicative of elevated membrane cAMP levels (Fig. 4C and D). Addition of forskolin and IBMX rapidly increased cAMP level in both WT and Abcc4^−/− cells in all subcellular compartments monitored (Fig. 4D; Supplementary Fig. S3C). No significant difference was observed in either the amplitude of maximum response or the time to reach maximum response between the genotypes (Supplementary Fig. S3C). As expected, the cAMP response at the plasma membrane was faster than the cytoplasm and nucleus (Supplementary Figs. S3C and S3D).

Following the maximum cAMP response, cells reestablished a new, lower steady-state cAMP level (Fig. 4C). Abcc4 deficiency significantly impaired the creation of a lower steady-state cAMP at the plasma membrane (which we refer to as “FRET change”; Fig. 4D). This contrasts with the cytosol and nucleus, where the cells' established a new steady-state level that appeared unaffected by Abcc4 loss (Fig. 4D; Supplementary Fig. S3C). Together, these results highlight that ABCC4 is capable of determining the duration of sustained cAMP concentration at the plasma membrane.

We next investigated how ABCC4 absence impacts the direct downstream target of cAMP, protein kinase A (PKA). PKA activity at the plasma membrane and cytosol was monitored using PKA activity FRET biosensors (AKAR4 sensors; ref. 33). The AKAR4 sensor harbors an FHA1 phosphoamino acid binding domain and a PA substrate peptide. When PKA activity is high, the substrate peptide is phosphorylated, the FHA domain then binds to the phosphorylated substrate thus inducing the FRET event (33, 34). A high ratio of YFP over CFP emission signifies strong PKA activity. We hypothesized that in Abcc4^−/− cells, the majority of the sensors are phosphorylated in the resting state. As such, inhibition of PKA by H89 will decrease the FRET ratio as a consequence of sensor dephosphorylation by endogenous phosphatases. When WT and Abcc4^−/− cells expressing membrane PKA sensors were treated with H89, there was a much greater decrease in the FRET ratio for cells lacking ABCC4 (Fig. 4E). This indicated that the amount of phosphorylated sensor was much greater at the resting state in Abcc4^−/− cells compared with WT, suggesting a greater PKA activity. This difference between the genotypes was not observed for the cytosolic PKA sensors and indicates that ABCC4 specifically influences plasma membrane PKA activity (Fig. 4E). In total, these results show that the plasma membrane cAMP and PKA activity are regulated by ABCC4.

As GPR161 is implicated in cAMP/PKA-dependent basal repression of SHH signaling (35, 36), we investigated if GPR161 might be an intermediary in the altered response to the SHH agonist in Abcc4^−/− cells. Notably, Abcc4^−/− cells had more GPR161 positive cilia (Fig. 4F). Exit of GPR161 from primary cilia is required for strong SHH pathway activation. After SHH pathway activation, more GPR161 was retained in primary cilia in Abcc4^−/− cells compared with WT cells, which may translate into SHH pathway repression (Fig. 4F, WT 1 hour vs. Abcc4^−/− 1 hour). Importantly, restoration of ABCC4 to levels similar to WT in Abcc4^−/− cells reduced the proportion of GPR161-positive cilia during signaling (Fig. 4F). Given the role of ABCC4 in modulating cAMP, we hypothesized that cAMP itself might influence GPR161 localization during signaling. To test this hypothesis, cells were treated with forskolin and IBMX, which increases cAMP level, in the presence of SHH (Fig. 4F, green bars). We observed that when cAMP is high, GPR161 is retained in the primary cilia even under pathway activating conditions (Fig. 4F, compare WT 1 hour SHH and WT 1 hour SHH + forskolin + ibmx). These results further demonstrate that ABCC4, by modulating membrane cAMP, modulates cAMP-responsive components of the SHH pathway, such as the level of GPR161 in the primary cilia.

Next, we utilized a phospho-specific antibody that recognizes phosphorylation of PKA substrates at the consensus PKA motif, RRXS*/T* as an orthogonal approach to measure PKA activity. The magnitude of PKA substrate phosphorylation in cells lacking Abcc4 under basal conditions was greater than WT cells (Fig. 4G). Extending this, 1 hour after signaling initiation, the number of phosphorylated substrates appears more extensive in Abcc4^−/− cells (Fig. 4G), which may in part be related to retention of GPR161 at primary cilia. These results further support ABCC4 role in modulating plasma membrane cAMP level and PKA activity during initiation of SHH signaling.

Increased PKA activity can alter the overall phosphorylation status of PKA targets including GLI2/3 and SUFU, subsequently altering GLI signaling output (37, 38). Consistent with this, we observed, by quantitative phosphoproteomics, differential phosphorylation of SUFU, GLI2, and GLI3 in Abcc4^−/− cells compared with WT (Supplementary Fig. S3E).

Aberrant, disease-inducing, ligand-independent SHH signaling might also be modulated by ABCC4. Loss of Ptch1 relieves the repression of SMO, allowing it to ultimately activate the GLI transcription factors. In contrast, Sufu mutations found in MB can produce a truncated, inactive protein, causing the constitutive presence of GLI proteins in the nucleus (6). In Ptch1^−/− MEFs and Sufu knockdown NIH3T3 cells (SufuKD), ABCC4 was upregulated at both the protein and mRNA level compared with WT NIH3T3 cells (Fig. 5A). Importantly, suppression of Abcc4 in both Ptch1^−/− and SufuKD cells resulted in reduction in the activity of the SHH pathway as measured by GLI1 at the protein and mRNA levels (Fig. 5B–D).

Acquired resistance to SMO inhibitors reactivates SHH pathway, secondary to mutations within SMO such as W535L (SMOA1 mutant; refs. 4, 39). The oncogenic SMO mutant resulted in a ligand-independent expression of GLI and increased ABCC4 expression (Fig. 5E). SMOA1 expression increased Gli1 mRNA level even in the absence of ligand, indicative of aberrant signaling, which was strongly reduced in Abcc4-null cells (Fig. 5F). These results show that aberrant SHH pathway can be markedly reduced by suppression of ABCC4.

We next investigated if suppression of Abcc4 alters SHH-MB pathogenesis. SHH-MB arising from Ptch1^−/−;Trp53^−/− tumors were lentivirally-transduced with an all-in-one CRISPR-Cas9 plasmid encoding Cas9 in cis with GFP, with or without Abcc4 gRNA (targeting exon 4 of mouse Abcc4). Multiple gRNAs were designed and tested but gRNA targeting exon 4 reduced ABCC4 to the greatest extent. These tumor cells were then implanted into the cortices of naïve recipients and disease progression was monitored. In two independent experiments, among a total of 10 mice that were implanted with Abcc4 gRNA-infected tumor cells, only 50% of the mice developed pathologically discernible MB, whereas in the control group, all of the mice readily developed MB (Fig. 5G). Although 50% of mice implanted with tumor cells expressing Abcc4 sgRNA developed tumor, their disease showed significantly delayed progression (P = 0.0012; Fig. 5H). When moribund mice were sacrificed, tumor cells were isolated and analyzed by immunoblotting to evaluate SHH signaling. Because the GFP is in cis with Cas9 and linked through a P2A sequence, GFP level was used to measure transduction efficiency. All GFP levels in tumors were similar, indicating the plasmid containing guide RNA targeting Abcc4 and Cas9-GFP was taken up by the tumors and that Cas9 was expressed. As we demonstrated in NIH3T3 cells with knockdown of Abcc4, SHH target genes including GLI1 and Cyclin D2 were decreased in tumors where ABCC4 level was reduced (Fig. 5I). Notably, among the Abcc4 gRNA cohort that developed tumor, ABCC4 levels were much lower than the control gRNA cohort (Fig. 5I). Furthermore, Nestin, a marker for stem cell, whose expression has been shown to contribute to tumorigenesis, was reduced in tumors that received Abcc4 gRNA (Fig. 5I). Together, these results show that ABCC4 is required for maximal tumor expansion and SHH pathway activation in vivo.

Finally, to explore the molecular connections between ABCC4 and GLIs, we extracted a subnetwork from the human MB interactome. We found the GLI transcription factors share mutual downstream targets, not only within GLI family members, but also with ABCC4. Interestingly, only GLI2 showed a direct regulatory relationship with ABCC4. In addition, PKA coregulated multiple target genes with GLI family members too (Fig. 6). In total, these results integrate ABCC4 into a regulatory network important in SHH-MB pathogenesis.

In this study, mining of clinical data coupled with a systems biology approach suggested that ABCC4 was a potential modulator SHH-MB pathogenesis. To test this hypothesis, we used cellular model systems and found that ABCC4, by regulating membrane cAMP levels and subsequently GPR161 localization and GLI3 amount, functions in both normal and aberrant SHH signaling. Extending this to a preclinical model that faithfully recapitulates human SHH-MB (7, 40, 41), we validated that ABCC4 is indeed a disease-modifying regulator of SHH-MB, a finding supporting the significant association between poor prognosis in SHH-MB and high ABCC4 expression. Ablation of Abcc4 was especially salutary when applied to a murine model of SHH-MB as lifespan was extended significantly and associated with a strongly reduced tumor burden. Although ABCC4 overexpression alone does not spontaneously activate the SHH pathway, when it is activated, ABCC4 function is crucial to elicit maximal activation. These findings highlight how upregulation of ABCC4 expression promotes a heretofore unknown “feedforward” mechanism in the SHH pathway, which suggests that reduction in ABCC4 function might be a general strategy to yield overall improvements in SHH-MB outcomes.

We propose the following model for ABCC4 modulation of SHH signaling (Fig. 7). The enrichment of cilia with GLI2/3 after SHH activation is unimpaired by ABCC4 deficiency (Fig. 7B). However, sustained plasma membrane cAMP with ABCC4 loss delays ciliary exit of GPR161 and is closely related to PKA hyperactivation (Fig. 7C). Increased PKA activity enhances phosphorylation of PKA targets including GLI3, which correlates with its reduced stability and is likely related to an alteration in its phosphorylation. Importantly, reexpression of ABCC4 rescues the stability of GLI3 and restores GPR161 exit from cilia. Thus, further elucidation of the crosstalk between the GLI transcription factors and ABCC4 may provide additional mechanistic insights into approaches to minimize the positive regulation of SHH signaling by ABCC4 (Fig. 6).

ABCC4's role in SHH-MB disease progression is important in the following context: (i) high ABCC4 expression is prognostic of poor overall survival in human SHH-MB; (ii) ABCC4 expression is correlated to GLI2, which is amplified in 29% of human SHH-MB cases (15); (iii) loss of Abcc4 suppressed aberrant SHH signaling caused by Ptch1 and Sufu inactivation, genetic lesions that drive MB; (iv) a MB-specific interactome revealed ABCC4, an ATP-binding cassette transporter, as a key player in SHH-MB. To that end, we employed a mouse model of SHH-MB harboring either one Ptch1 allele with Trp53 loss (7), genetic alterations typically found in human SHH-MB (42). ABCC4 is highly expressed in these tumors (Fig. 1G and H). Targeted Abcc4 ablation decreased tumor incidence, extended survival of tumor-bearing mice, and reduced SHH signaling even in tumors without complete Abcc4 ablation. Collectively, these results show ABCC4 regulates SHH-MB progression.

Inhibition of ABCC4 might be a novel strategy to block SHH-MB disease progression and improve overall survival for several reasons. First, this approach seems advantageous because of where ABCC4 affects the SHH pathway. ABCC4 acts downstream from SMO activation, which is where the most aggressive cases of MB have driver mutations in either SUFU or GLI2 amplification (15). Further, as cAMP modulates phosphorylation of PI3K (43), cAMP regulation by ABCC4 might attenuate PI3K activity. In adult SHH-MB, proteins in the PI3K/Akt/mTOR pathway are recurrently mutated (12). Importantly, in adult patients, studies showed that inhibition of PI3K phosphorylation is effective in inhibiting MB growth (44, 45).

There are important additional reasons for the appeal of this approach, not the least of which has been described herein, but also because current frontline MB chemotherapy includes the camptothecins, topotecan and irinotecan, well-known ABCC4 substrates (19, 46). Thus, an effective ABCC4 inhibitor would be envisioned as both restraining MB tumor cell growth and decreasing tumor cell survival by affecting ABCC4 function, while concomitantly blocking camptothecin export, thereby further increasing the intratumoral concentration and tumorocidal activity. Further targetting ABCC4 might be beneficial in treating other cancers where high ABCC4 expression appears to negatively impact survival. Interrogation of public databases reveals that high ABCC4 expression predicts poor survival for multiple cancers (Supplementary Figs. S4A–S4F). One possible explanation for this might be that high ABCC4 affects response to chemotherapy, however this possibility seems unlikely given the range of cancer chemotherapeutics used to treat these different cancers. Alternatively, we favor a more fundamental biological role for ABCC4. This is in part based on studies in hematopoietic stem cell progenitors, where ABCC4 was shown to transport bioactive endogenous signaling molecules including prostaglandins, leukotrienes, and cAMP(47). Furthermore, in human leukemic cell lines modulation of ABCC4 levels altered monocytes differentiation and maturation, suggesting a role in cancer stem cell biology (25). Thus, it seems likely that ABCC4's ability to transport endogenous biologically active molecules might be key for its impact on multiple cancers.

No potential conflicts of interest were disclosed.

Conception and design: J. Wijaya, B.T. Vo, M.F. Roussel, J.D. Schuetz

Development of methodology: J. Wijaya, B.T. Vo, J. Peng, J. Yu

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.):J. Wijaya, Y. Wang, J. Peng, J. Zhang

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J. Wijaya, J. Liu, B. Xu, G. Wu, L.J. Janke, B.A. Orr, J. Yu, J.D. Schuetz

Writing, review, and/or revision of the manuscript: J. Wijaya, B.T. Vo, J. Liu, G. Wu, B.A. Orr, J. Yu, M.F. Roussel, J.D. Schuetz

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J. Wijaya

Study supervision: M.F. Roussel, J.D. Schuetz

This study was supported by P30 CA021765 Cancer Center Support grant (to J.D. Schuetz, M.F. Roussel), R01AG053987 (to J. Peng), PO1CA096832 (to M.F. Roussel), and ALSAC of St. Jude Children's Research Hospital. We thank Drs. Young-Goo Han and Stacy Ogden, Frederique Zindy, and Schuetz lab members for helpful discussions. Microscopy images were acquired at SJCRH Cell & Tissue Imaging Center. We thank Drs. Victoria Frohlich, Aaron Pitre, Jennifer Peters, and Sharon King for technical help with microscopy analyses and data acquisition. We thank the staff of Flow Cytometry, Cell Sorting Shared Resource and the Center for In vivo Imaging and Therapeutics.

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.