Prostaglandin E1 Inhibits GLI2 Amplification–Associated Activation of the Hedgehog Pathway and Drug Refractory Tumor Growth

The evolutionarily conserved Hedgehog (HH) signaling pathway plays critical roles in embryonic patterning and adult tissue homeostasis (1, 2). Hyperactive HH signaling has been linked to a range of malignant tumors through tumor initiation, maintenance of tumor stem/progenitor cells, and support of tumor–stroma interaction (3, 4). Therefore, the HH signaling has emerged as a therapeutic target of interest for cancer therapy, and intensive efforts have been made to develop targeted pathway antagonists.

Mammalian HH signal transduction is controlled by the Patched1 (PTCH1)-mediated suppression of Smoothened (SMO), a seven-pass transmembrane protein that traffics continuously through the primary cilium (PC; refs. 5, 6). Inactive SMO failed to regulate the activity state of GLI2, the primary transcription activator of HH pathway, which thus was sequentially phosphorylated by protein kinase A (PKA), glycogen synthase kinase-3β (GSK-3β), and casein kinase 1 (CK1), and trafficked to the proteasome for degradation. On HH ligand [Sonic hedgehog (SHH), Desert hedgehog (DHH), or Indian hedgehog (IHH)] binding to the shared receptor PTCH1, the inhibitory effect on SMO is relieved, enabling SMO ciliary accumulation and activation (5, 6). Consequently, GLI2 translocates in activated full-length form from the cilium to the nucleus (7), where it induces orchestrated expression of target genes, including GLI1 and PTCH1.

Constitutive HH signaling contributes to tumorigenesis mainly through two types of mechanisms. First, ligand-independent hyperactive pathway activity within the tumor cell drives tumorigenesis in basal cell carcinoma (BCC), the most common cancer in Caucasian population (8), medulloblastoma, the most common childhood brain cancer (9), and rhabdomyosarcoma (10). Almost all cases of BCC are initiated by ligand-independent HH activity, most commonly through PTCH1 loss-of-function or SMO gain-of-function mutations (11, 12). Similarly, hyperactive HH signaling has emerged as the driver in approximately 30% of medulloblastoma through ligand-independent mechanisms including inactivating mutations in PTCH1 and SUFU, and genomic amplification of GLI2 (13–15). Second, HH pathway activation in surrounding stromal cells has been found to support the growth of tumor cells in a paracrine manner, whereby stromal cells receive HH ligand from tumor cells and secrete stimulatory factors in response for tumor progression (16). Such mechanism was documented in a broad range of malignancies, most notably those in blood, pancreas, lung, stomach, colon, and prostate (3). Clinical implications of the paracrine mechanism of action are yet to be clarified as most clinical trials using HH pathway antagonists to treat these cancers did not meet a positive conclusion (4). However, glasdegib was recently approved by the FDA for acute myeloid leukemia, thus highlighting potential expanded use of HH targeted cancer therapy beyond BCC and medulloblastoma (17).

Cyclopamine, a natural compound found in wild corn lily (Veratrum californicum), was identified as the first HH pathway inhibitor directly targeting SMO (18). Since then, many more SMO inhibitors have been developed, and several of them, including vismodegib, sonidegib, glasdegib, LY2940680, and BMS-833923, have delivered promising results in preclinical and clinical studies in HH-dependent cancers (3). Both vismodegib and sonidegib have been approved by the FDA for treatment of advanced BCC (19, 20). However, acquired resistance to vismodegib and sonidegib limits their long-term efficacy. Drug resistance can be acquired by genetic aberrations of multiple pathway components including SMO mutations, SUFU mutations, and GLI2 amplifications (21–24). Notably, intratumor heterogeneity of those drug refractory mechanisms was identified, further complicating the situation that next generation cancer therapy needs to tackle (21). In addition, current anti-SMO therapies failed to target primary tumors harboring mutations downstream of SMO level (25).

The emergence of multiple drug resistance mechanisms associated with current SMO antagonists and lack of therapies targeting HH pathway downstream of SMO level has prompted our investigations into alternative approaches. From a perspective of pathway epistasis, we reasoned that targeting hyperactive GLI2, the central transcription activator of the pathway, would potentially deliver more effective therapeutic interventions that may pan-inhibit various drug refractory mechanisms. Herein, we reported the discovery of a number of prostaglandins in a high-content screening for small molecules inhibiting GLI2 ciliary accumulation. We demonstrated that prostaglandin E1 (PGE1), an approved drug as a representative among the class, delivered cross-inhibitory activities against drug refractory SMO mutants and overexpressing GLI2. Mechanistic investigations revealed that PGE1 acts through E-prostanoid receptor 4 (EP4), which localizes to the PC, to elevate cyclic adenosine monophosphate (cAMP)-PKA signaling, thus leading to phosphorylation and subsequent ubiquitination primed degradation of GLI2. Furthermore, PGE1 effectively inhibits drug refractory tumor growth of GLI2-overexpressing medulloblastoma xenografts. In summary, our study identified prostaglandins as potential source of drug repurposing opportunities, which are capable of overcoming multiple drug resistance mechanisms associated with current generation SMO inhibitors and targeting medulloblastoma with broader molecular spectrum. In addition, the findings provide novel mechanistic insights furthering understanding of HH pathway modulation.

NIH/3T3, HEK293T, Cos7, DAOY, and MCF7 cells were obtained from the ATCC. NIH/3T3 cells were cultured in DMEM supplemented with 10% (v/v) calf serum. HEK293T and Cos7 cells were cultured in DMEM supplemented with 10% (v/v) FBS, and DAOY cells were cultured in MEM supplemented with 10% (v/v) FBS. MCF7 cells were cultured in MEM supplemented with 10% (v/v) FBS and 0.01 mg/mL human recombinant insulin. 3T3/ARL13B::tagRFPT and 3T3/ARL13B::tagRFPT/EGFP::GLI2 stable cell lines were generated via lentiviral infection of NIH/3T3 cells. The 3T3/GLI-luc cell line was created for GLI-luciferase reporter assays via two rounds of infections using lentiviral particles harboring a GLI-responsive firefly luciferase reporter and a constitutive Renilla luciferase expression construct, respectively. Subclones overexpressing GLI2 or SMO-WT or SMO-D473H or SMO-W535L in 3T3/GLI-luc cells were generated. DAOY cell lines that overexpress SMO-D473H or GLI2 were also generated via lentiviral delivery. Med-113FH and Med-314FH tumor cells were obtained from the brain tumor resource lab (http://www.btrl.org). The passaging of Med-113FH and Med-314FH tumor cells was through serial transplantation into the cerebellum of immune-compromised mice. All cell lines were confirmed as Mycoplasma negative using the Mycoplasma PCR Detection Kit (HD01-0105, HD BIOSCIENCES). All primary cell lines were directly obtained from a public depository or from a commercial supplier and not additionally authenticated. Cellular experiments were performed within 10 passages after thawing.

Chemical libraries used in our screening include the Prestwick Chemical Library (Prestwick Chemical), the Spectrum Collection (Microsource Discovery Systems), the Library of Pharmacologically Active Compounds (LOPAC, Sigma), FDA-approved Drug Library (Topscience), and customized in-house compound libraries. Cyclopamine, sulprostone, and butaprost were purchased from Sigma. SAG was purchased from Millipore. Vismodegib and forskolin (FSK) were purchased from Selleck. All prostaglandins except sulprostone and butaprost were purchased from Cayman Chemical. SHH-N conditioned medium was collected as previously described (26), and its control medium was collected from wild-type Cos7 cells.

3T3/ARL13B::tagRFPT/EGFP::GLI2 cells were plated into 384-well imaging plate precoated with 0.1% gelatin (Sigma) at 1 × 10⁴ cells/well in 50 μL of media. After cells reached confluence (1–2 days), test compounds were added in 0.5% calf serum medium for 24 hours in the presence of SHH-N, and then cells were fixed with 4% paraformaldehyde and stained with Hoechst 33342 (H3570, Invitrogen) for imaging. Cells were imaged using Operetta High Content Screening System (PerkinElmer) with a 40× high NA objective. The Harmony 4.1 software (PerkinElmer) was used for high-content screening data management and image quantification. The identical microscopic setting and input parameters were performed throughout the imaging assay.

As for the wild-type NIH/3T3 cells for examining HH signaling activity, cells were plated at 2 × 10⁴ cells/well into 96-well assay plates and transfected the next day using Fugene HD (E2311, Promega) with a p8 × GLI binding sites (GLIBS)–firefly luciferase plasmid (27), a constitutive Renilla luciferase plasmid, and other DNA plasmids as indicated. After cells reached confluency in about 1 day, culture medium was switched to 0.5% calf serum medium and incubated for 36 hours with other reagents as indicated. For stable reporter cell lines, 3T3/GLI-luc or its derivatives were cultured in 96-well assay plates. Upon confluence, cells were treated with reagents in 0.5% calf serum medium as indicated for 36 hours. Then the firefly and Renilla luciferase activities were read sequentially by Luminescence Counter (PerkinElmer). The Renilla luciferase signal was used to normalize the firefly luciferase signal.

HEK293T cells were seeded in 96-well assay plates and cotransfected with a pGL4-CRE firefly luciferase construct and a constitutive Renilla luciferase construct. Twenty-four hours after transfection, cells were treated with compounds as indicated and incubated for another 36 hours before being processed for reading luciferase signals using Luminescence Counter (PerkinElmer). The firefly luciferase signal was normalized by Renilla luciferase signal.

The pCEFL-3 × Flag-Gli2 (Gli2-WT) plasmid was generated by replacing 3 × HA in plasmid pCEFL-3 × HA-Gli2 (37671, Addgene) with 3 × Flag. Site mutations were introduced into the Gli2-WT vector to generate pCEFL-3 × Flag-Gli2 ΔPKA (Gli2 ΔPKA) plasmid and pCEFL-3 × Flag-Gli2 ΔGSK-3β (Gli2 ΔGSK-3β) plasmid, whose mutated phosphorylation sites have been described as previously (28). Transfection was performed in NIH/3T3 cells using Fugene HD (E2311, Promega) according to the manufacturer's instructions. Transfected cells were treated with compounds in 0.5% calf serum for 36 hours before collecting the samples for examination.

Total RNA was isolated from cultured cells or snap-frozen tumors using the TRIzol Reagent (15596018, Thermo Fisher Scientific). Each RNA sample was treated with DNase I (AM1907, Thermo Fisher Scientific) at 37°C for 30 minutes to remove any contaminating genomic DNA and then used as a template for cDNA synthesis with the GoScript cDNA Synthesis Kit (Promega) according to the manufacturer's instructions. The High-Fidelity KOD-Plus-Neo (KOD-401, TOYOBO) was used for PCR amplification to determine gene expression. Primers for analyzing EP genes expression in mouse NIH/3T3 and human DAOY cells were listed in Supplementary Table S1.

A 1/10 dilution of cDNA was used as a template with a SYBR Green–based PCR Master Mix (11201ES08, YEASEN) on a Roche^@480 Real-Time PCR System. The relative expression levels of mRNAs were calculated using 2^–ΔΔC_t. Glyceraldehyde-3-phosphate dehydrogenase (Gapdh) was used as internal reference. The primer sequences used to quantify HH target genes are listed in Supplementary Table S1.

Cells were lysed in RIPA buffer (C1053, APPLYGEN) supplemented with 1 tablet of protease and phosphatase inhibitors (A32961, Thermo Fisher Scientific) per 10 mL RIPA buffer. Equal amounts of cell lysates were separated by SDS-PAGE and then transferred to polyvinylidene fluoride (PVDF) membranes. Subsequently, the membranes were blocked in 5% nonfat milk or 5% BSA in TBST buffer at room temperature for 2 hours and then incubated overnight at 4°C in blocking solution with the primary antibodies. After incubation overnight, membranes were washed 3 times per 10 minutes in TBST buffer, then incubated in 1:5,000 diluted horseradish peroxidase–labeled secondary antibodies (ZSBIO) at room temperature for 1 hour, washed for 3 times every 15 minutes with TBST buffer, and then detected with chemiluminescent reagents (32106, Thermo Fisher Scientific). As for detecting phosphorylation state of CREB, the blot on PVDF membrane was first developed with anti–p-CREB primary antibody (9198, Cell Signaling Technology), then the same membrane was stripped and reprobed for total CREB protein with anti-CREB primary antibody (9197, Cell Signaling Technology) and then with anti–β-actin (ab8226, Abcam) as an internal control. The stripping procedure was as following. The membrane was incubated with Western blot stripping buffer (BE6224, EASYBIO) for 30 minutes at room temperature in shaking motion and then washed 3 times per 10 minutes in TBST buffer. Other primary antibodies used in our study include mouse anti-GLI1 (2643, Cell Signaling Technology), Goat anti-GLI2 (AF3635, R&D), Sheep anti-GLI2 (AF3526, R&D), and mouse anti-EP4 (sc-55596, Santa Cruz Biotechnology).

For gene knockdown, the following shRNA plasmids: shEP4-1 (targeting sequence: 5′-GTACTGTTTCTGGACCCTTAT-3′) and shEP4-2 (targeting sequence: 5′-CAGTAAAGCAATAGAGAAGAT-3′), scrambled shRNA (targeting sequence: 5′-AACGTGATTTATGTCACCAGA-3′) were constructed and used in transfection studies. The transfection of shRNA plasmids was performed using Lipofectamine 2000 (11668019, Invitrogen) according to the manufacturer's protocol.

EP4 genomic mutations were generated by CRISPR-Cas9–mediated genomic editing. Single-guide RNA targeting murine EP4 genome was designed through the sgRNA design tool (https://portals.broadinstitute.org/gpp/public/analysis-tools/sgrna-design). Genome editing efficiency was examined through the Tide tool (https://tide.deskgen.com), and the selected sequence is: GGCGGCGTAGGCCGTTACGT. The annealed guide RNA oligos were inserted into a lentiviral plasmid digested by the BsmBI restriction enzyme. Genotyping was determined by sequencing PCR products amplified by the following primers: EP4 forward: TCTGTGCCATGAGCATCGAG; EP4 reverse: TCAGGACTTAGAAGGAAAAC. The PCR products with double peaks were then inserted into a pEASY-Blunt Zero Cloning Vector (CB501-01, TransGen Biotech) and sent for Sanger sequencing to verify the sequence of each allele.

The 3T3/ARL13B::tagRFPT cells were plated onto coverslips in 24-well plates (200,000 cells/well) and maintained in DMEM supplemented with 10% (v/v) calf serum until reaching confluency, at which time the culture medium was switched to 0.5% calf serum medium for 24 hours. After that, cells were fixed in 4% paraformaldehyde for 30 minutes, permeabilized in 0.3% Triton X-100 for 15 minutes, blocked in 2% BSA plus 0.3% Triton X-100 for 1 hour, and then incubated overnight at 4°C with mouse monoclonal anti-EP4 antibody (sc-55596, Santa Cruz Biotechnology, 1:200 dilution). After incubation overnight, cells were washed 3 times every 10 minutes, followed by the incubation with a secondary antibody, Alexa Fluor 488 goat against mouse IgG (H+L) (A-11001, Thermo Fisher S cientific, 1:500 dilution), at room temperature for 1 hour, then washed additional 3 times every 10 minutes. Cells were finally stained with Hoechst 33342 (H3570, Invitrogen) and washed, after which, the coverslips were mounted and samples were imaged. Images were collected using a Zeiss LSM 780 confocal microscope with a 63× oil objective and were processed with ZEN software (Zeiss).

For examining GLI2 ubiquitination, HEK293T cells were transiently transfected with DNA plasmids as indicated, and treated with or without PGE1 for 36 hours, all samples were treated with 10 μmol/L MG-132, a proteasome inhibitor, for 6 hours before collection. Cells were collected and lysed in lysis buffer plus protease and phosphatase inhibitors (A32961, Thermo Fisher Scientific) for 50 minutes on a rotor at 4°C. After 12,000 × g centrifugation for 15 minutes, the lysates were immunoprecipitated with 2 μg specific antibody overnight at 4°C, and 30 μL Protein A/G PLUS-Agarose (SC-2003, Santa Cruz Biotechnology) was washed and then added for an additional 3 hours. Thereafter, the precipitants were washed 5 times with lysis buffer, and the samples were boiled with loading buffer for 5 minutes and analyzed by immunoblotting to examine GLI2 ubiquitination. Antibodies against FLAG (F1804, Clone M2, Sigma) and HA (ab9110, Abcam) were used to examine GLI2 ubiquitination.

For examining GLI2 phosphorylation, HEK293T cells were transiently transfected with DNA plasmids as indicated. Thirty-six hours after treatment with or without PGE1, cells were collected and processed for immunoprecipitation assay following the above procedures. Then the samples were analyzed by immunoblotting to examine GLI2 phosphorylation. Antibodies against FLAG (F1804, Clone M2, Sigma) and phospho-(Ser/Thr) PKA substrate (9621, Cell Signaling Technology) were used to examine GLI2 phosphorylation.

Cells were plated at 3,000 cells per well in 96-well plates and treated with drugs as indicated for 72 hours. The CCK-8 reagent (B34302, Biomake) was directly added into the plate at 10 μL/well and incubated at 37°C for 4 hours. Cell viability was assessed using a PerkinElmer plate reader.

The patient-derived orthotopic xenograft experiments were approved by the Institutional Animal Care and Use Committee of Institute of Zoology, Chinese Academy of Sciences, and conducted according to institutional guidelines. The surgery of orthotopic xenograft experiments was carried out in 6- to 7-week-old NOD-Prkdc^scidIl2rg^tm1/Bcgen mice (BIOCYTOGEN) following anesthetized by i.p. injection of 400 mg/kg 2,2,2-Tribromoethanol (T1420, TCI). The head of mouse was properly placed in a stereotaxic apparatus in a sterile environment, and then a small incision was made starting between the ears and ending near the back of the skull using a sterile scalpel. Then, a 0.8-mm-diameter burr hole is placed in the calvarium using a hand-held microdrill (78001, RWD Life Science) with a 0.8-mm burr (78042, RWD Life Science) using the following coordinates: 2 mm posterior to the lambdoid suture, 2 mm lateral to the midline, and 2.5 mm ventral from the surface of the skull. Note that 3 × 10⁵ Med-113FH cells or Med-314FH cells in a volume of 3 μL were stereotaxically injected into the cerebellum using a 33G needle mounted at a Hamilton syringe (7634-01, Hamilton) at an infusion rate of 1 μL/min. After injection, the needle was kept in place for about 5 minutes to equilibrate the pressure within the cranial vault. The incision was closed with two to three interrupted sterile sutures (CR434, Jinhuan Medical). Mice were randomized into different groups (n = 5 in each group) according to the proper luminescence signal. For drug treatment in vivo, vismodegib was used at 30 mg/kg for oral administration once a day, and PGE1 was used at 15 mg/kg for i.p. administration once a day. Tumor growth was monitored weekly by bioluminescence acquisitions using the IVIS Spectrum Imaging System (PerkinElmer). Survival data were collected throughout these orthotopic xenograft experiments.

The flank xenograft experiments were approved by the Institutional Animal Care and Use Committee of Institute of Zoology, Chinese Academy of Sciences, and conducted in compliance with institutional guidelines. In brief, 5 × 10⁶ GLI2-overexpressed DAOY cells in a total volume of 100 μL of 1:1 mixture of PBS:Matrigel were injected subcutaneously at the flank of each 6- to 7-week-old NOD-Prkdc^scidIl2rg^tm1/Vst mice (VITALSTAR). When tumors were grown to a median size of 100 mm³, mice were randomized into three groups (n = 8 in each group) and treated with solvent only, vismodegib (30 mg/kg, daily oral administration), and PGE1 (15 mg/kg daily i.p. administration) for 51 days, respectively. Tumor volume was measured with calipers once every 3 days and calculated by the formula: length × width × width × 0.5. At the end of the treatment period, each tumor was harvested and divided for qRT-PCR, hematoxylin and eosin (H&E), and IHC analyses.

Tumor samples were freshly collected, formalin-fixed, and paraffin embedded. Before staining, sections were rehydrated by xylene and a series of descending concentrations of ethanol, and then rinsed by deionized H₂O. For H&E staining, sections were incubated with H&E following standard procedures. For IHC staining, sections were performed with sodium citrate buffer for heat-induced epitope retrieval, and then incubated with 1:600 diluted rabbit anti-Ki67 antibody (ab15580, Abcam) at 4°C overnight, washed 3 times every 10 minutes with PBS, and then detected with the anti-rabbit DAB kit (PV-9001, ZSBIO). Slides were rinsed by deionized H₂O, dehydrated by a series of ascending concentrations of ethanol and xylene, and then cover slipped. The resulting slides were digitally scanned at 40× magnification on an Aperio VERSA 8 system (Leica Biosystems).

Data were shown as mean ± SD throughout the paper. Two-way ANOVA was used for comparing tumor growth curves. Log-rank (Mantel–Cox) test was used for comparing survival curves. For other comparisons, two-tailed Student t tests were used to generate P values, and a P value less than 0.05 was considered statistically significant. Plots and statistical tests were performed by GraphPad Prism software. Asterisks (*) always represent degree of significance as follows. *, P < 0.05; **, P < 0.01; ***, P < 0.001; and NS, not significant.

The essential role of GLI2 ciliary accumulation in HH pathway activation presents a novel opportunity for high-content drug discovery targeting GLI2. We thus developed a high-content screening method for small molecules that inhibit GLI2 translocation to the cilia. To facilitate identification of the PC, we first introduced a construct of ADP-ribosylation factor-like GTPase 13B (ARL13B) tagged with red fluorescent tagRFPT into HH-responsive NIH/3T3 cells. ARL13B is a GTPase required for ciliogenesis (29). Subclones of this cell line were created by introducing EGFP::GLI2 expression lentivirus particles. Mouse GLI2 was used in the EGFP fusion construct. A clone with low EGFP::GLI2 expression, designated 3T3/ARL13B::tagRFPT/EGFP::GLI2 cell line, was selected for further study. In this cell line, EGFP::GLI2 mirrored previously reported endogenous GLI2 ciliary trafficking behaviors (30, 31). EGFP accumulation in the PC was observed upon treatment with SHH-N, the SHH N-terminal signal peptide, or SAG, a small-molecule SMO agonist. And its SHH-N–induced ciliary accumulation was attenuated by vismodegib or cyclopamine (Supplementary Fig. S1A and S1B). Moreover, GLI-dependent transcriptional activity responds to these treatments as expected for both 3T3/ARL13B::tagRFPT/EGFP::GLI2 cell line and its parental 3T3/ARL13B::tagRFPT cell line (Supplementary Fig. S1C). These data indicate that EGFP::GLI2 is a bone fide reporter of GLI2 ciliary localization well-suited to high-content screening. To probe the transcriptional output of the HH pathway activity, we next generated another NIH/3T3 reporter stable cell line designated as 3T3/GLI-luc, in which firefly luciferase is driven by GLI-dependent transcriptional activity through 8 tandem repeats of GLIBS (32), and a constitutive Renilla luciferase reporter serves as an internal control (Supplementary Fig. S1D–S1F). We overexpressed mouse GLI2 in this 3T3/GLI-luc cell line (named 3T3/GLI-luc/GLI2) via delivery of lentiviral particles containing a CMV-driven mouse GLI2 expression construct (Supplementary Fig. S1G). Exogenous GLI2, which can be readily detected by Western blot, elicited robust luciferase reporter activity in comparison with the parental cell (Supplementary Fig. S1H and S1I).

Using the 3T3/ARL13B::tagRFPT/EGFP::GLI2 cell line, we conducted a high-content screening for small-molecule inhibitors of SHH-N–induced ciliary accumulation of EGFP::GLI2 (Supplementary Fig. S2A). Compound libraries screened herein include FDA-approved drugs, candidates being studied in clinical trials and compounds with annotated biological functions. Through this screening, we identified several hits that include known HH pathway inhibitors, thus validating the assay (Supplementary Fig. S2B). Among the hits were 6 prostaglandins, including prostaglandin A1 (PGA1), prostaglandin D1 alcohol (PGD1 alcohol), 5-trans prostaglandin D2 (5-trans PGD2), PGE1, prostaglandin J2 (PGJ2), and 15-deoxy-Δ^12,14-prostaglandin J2 (15-deoxy-Δ^12,14-PGJ2; Fig. 1A–C).

In agreement of GLI2 amplification as a drug refractory mechanism for vismodegib, no inhibitory effect against GLI2-induced luciferase activity was observed for vismodegib at concentration up to 10 μmol/L, way higher than its saturated dose against wild-type activity (Fig. 1D; Supplementary Fig. S1F). In contrast, all 6 prostaglandins identified in the primary high-content screening and additional 16 prostaglandin analogs inhibit GLI2-induced pathway activity in the GLI-luciferase assay (Fig. 1D; Supplementary Fig. S3A–S3P). Among these prostaglandins, PGE1, also known as alprostadil, is an FDA-approved drug commonly used for the treatment of pulmonary hypertension, erectile dysfunction, and peripheral artery occlusive disease (33, 34). Therefore, we used PGE1 as a representative of these prostaglandins in further investigations. We confirmed that PGE1 effectively inhibits expression of GLI2 target genes, including Gli1 and Ptch1, whereas vismodegib and cyclopamine showed no inhibitory effects (Fig. 1E).

Given that PGE1 overcomes drug resistance introduced at the GLI level, we suspect, from an epistasis standpoint, it would also overcome other major drug refractory mechanisms introduced by SMO mutations and show activity with a broader spectrum. Therefore, we examined HH pathway activity mediated by two drug refractory SMO mutants, SMO-D473H and SMO-W535L (also known as SMO-M2; refs. 8, 23), which were identified in patients who experienced devastating cancer relapses during vismodegib treatments (21, 24). In agreement with this expectation, PGE1 effectively suppressed both vismodegib-resistant mutants in luciferase reporter assays of HH pathway activity (Fig. 2A and B), and while examining endogenous Gli1 and Ptch1 mRNA expression levels (Fig. 2C–E), and endogenous GLI1 protein level (Fig. 2F). All 6 prostaglandins identified from the primary high-content screening inhibited SMO mutants with equivalent potency to wild-type SMO (SMO-WT; Fig. 2B; Supplementary Fig. S4A–S4E). Consistently, no IC₅₀ shift was observed for PGE1 when escalating concentrations of SAG, a potent HH pathway agonist by binding and activating SMO protein (35), were applied, in contrast to results obtained for vismodegib (Fig. 2G and H). In addition to pathway activity introduced by SAG, that introduced at a further upstream level by SHH-N, can also be inhibited by PGE1, probed by various measurements (Supplementary Fig. S5A–S5C). Taking these data together, PGE1 showed pan-inhibition for multiple drug-resistant causes of SMO targeted cancer therapy, including GLI2 overexpression and SMO mutations, and PGE1 likely functions at a downstream level from SMO.

We next explored potential mechanisms underlying PGE1 inhibition against GLI2. Previous reports suggest that PGE1 might function through activation of cAMP signaling (36, 37). To this end, we employed a luciferase reporter controlled by cAMP responsive element (CRE). PGE1 indeed activated CRE-luciferase reporter in a dose-dependent manner (Fig. 3A), similar to a known cAMP agonist FSK (Supplementary Fig. S6A; ref. 38). To test whether PGE1 inhibition of GLI2 works through a cAMP-PKA–dependent mechanism, we generated a GLI2 mutant where PKA phosphorylation sites were altered (GLI2 ΔPKA), along with another GLI2 variant for comparison, where phosphorylation sites of GSK-3β were mutated (GLI2 ΔGSK-3β; Fig. 3B; ref. 28). In the GLI-luciferase assay, we found that PGE1 and FSK failed to inhibit GLI2 ΔPKA–induced HH pathway activation, whereas wild-type GLI2 WT and GLI2 ΔGSK-3β remain unaffected (Fig. 3C), suggesting PGE1′s GLI2 regulation specifically dependent on PKA-mediated phosphorylation.

To further study whether PGE1 regulation of GLI2 is mediated by PKA, we took advantage of PKI, an inhibitor peptide of PKA (39). We also generated a PKI mutant (PKI mut) control, mutating essential arginines at 20 and 21 positions (40). As expected, administration of PKI, but not its mutant form, significantly decreased CRE-luciferase signals stimulated by either PGE1 or FSK (Supplementary Fig. S6B). In support of PGE1 functioning through PKA-mediated GLI2 regulation, PKI rescued the PGE1 inhibitory effect on GLI2-induced HH pathway activity (Fig. 3D) and GLI2 protein level (Fig. 3E). Furthermore, using a polyclonal antibody developed against phosphorylated substrates of PKA, we observed that PGE1 treatment results in GLI2 phosphorylation, and such phosphorylation was abrogated by PKI (Fig. 3F). Together with GLI2 mutagenesis analyses (Fig. 3B and C), these results demonstrate PGE1 inhibits GLI2 activity by modulating PKA-mediated phosphorylation of GLI2.

The decreased GLI2 protein level on introduction of PGE1 (Fig. 3E) prompted us to further examine the potential effect of PGE1 on GLI2 ubiquitination. Previous studies suggest that GLI2 protein stability may be regulated by a sequence of events from multisite phosphorylation, ubiquitination, and consequent degradation by the proteasome (41, 42). To this end, we transfected Flag-Gli2 WT construct into HEK293T cells with HA-ubiquitin construct in the presence or absence of PGE1. Strikingly, PGE1 treatment increased the ubiquitination of GLI2 in comparison with a vehicle control. Treatment with PKI, not its mutant, reversed this effect (Fig. 3G). These data support a working mechanism of PGE1 regulation of GLI2 mediated by a cascade of events, including activation of PKA, followed by GLI2 phosphorylation, ubiquitination, and subsequent degradation.

Having identified the cAMP-PKA-ubiquitination regulatory cascade of PGE1 on GLI2 activity, it is tempting to ask that through which receptor(s) does the drug trigger such effect. There are four E-prostanoid receptors for PGE1, designated EP1, EP2, EP3, and EP4 (43). To determine which EP subtype(s) transduces the PGE1 signal, we first examined their expression and found that only EP1 and EP4 were detected in the NIH/3T3 cells, HH-responsive cells used in primary screening and secondary studies above, thus excluding EP2 and EP3 as the functional receptor for PGE1 action in this scenario (Fig. 4A). Next, we took advantage of the selective EP agonists, including sulprostone (agonist of EP1 and EP3; ref. 44), butaprost (EP2 receptor–specific agonist; ref. 45), and rivenprost (EP4 receptor–specific agonist; ref. 46). In GLI-luciferase assays, qRT-PCR analyses of endogenous Gli1 and Ptch1, and Western blot examination of GLI1 protein level, rivenprost elicited a similar effect on HH signaling compared with PGE1, whereas neither sulprostone nor butaprost was active, indicating EP4 receptor–mediated PGE1′s effect on HH signaling (Fig. 4B–D).

We next further addressed the question of whether EP4 mediates the PGE1-induced HH pathway inhibition by knocking out EP4 using the CRISPR-Cas9 approach (Supplementary Fig. S7A and S7B). In strong contrast with wild-type cells, treatment of PGE1 in EP4 knockout monoclonal cell lines failed to inhibit HH pathway activation induced by SHH-N, probed by mRNA levels of Gli1 and Ptch1 and protein level of GLI1 (Fig. 4E and F). Consistently, PGE1 also failed to inhibit SAG-induced HH pathway activities in EP4 knockout cells (Supplementary Fig. S7C and S7D). These findings directed us to examine the subcellular localization of EP4 in the NIH/3T3 cells. Immunofluorescence imaging analysis showed that EP4 colocalized with the PC marker ARL13B, in agreement with its role in PGE1 regulation of HH pathway (Fig. 4G). Taken together, we concluded that PGE1 initiates the cAMP-PKA-ubiquitination regulatory cascade of GLI2 through acting on EP4 receptor, possibly on the PC (Fig. 4H).

Having gained molecular insights of PGE1 regulation of GLI2 activity, we next explored its potential application in treating refractory tumors associated with current SMO targeted cancer therapies. Previous cancer studies in the field have utilized tumor allografts derived from genetically modified mice, which were also limited in just examining drug refractory SMO mutants (47, 48). To minimize interspecies variation, we first used DAOY cell line, a human SHH subtype medulloblastoma cell line with known PTCH1 mutations (49, 50). It has been widely used as a SHH subtype medulloblastoma model for HH signaling studies (49–53). EP1, EP2, and EP4 receptors were detected through RT-PCR in DAOY cells (Supplementary Fig. S8A). Among the selective EP agonists, only rivenprost elicited inhibitory activity similar with PGE1 against the HH pathway activity in DAOY (Supplementary Fig. S8B), consistent with our observations in mouse NIH/3T3 cells (Fig. 4B–D). We thus concluded that PGE1 also acts on EP4 in DAOY cells. To model drug resistance arising from both SMO and GLI2 abnormalities, we generated DAOY sublines in which human SMO-D473H and human GLI2 were stably overexpressed, respectively. Cell viability and expression of pathway target genes (GLI1 and PTCH1) in wild-type DAOY cells, SMO-D473H–overexpressed DAOY cells, and GLI2-overexpressed DAOY cells significantly decreased upon PGE1 treatments, whereas SMO-D473H–overexpressed DAOY cells and GLI2-overexpressed DAOY cells showed expected resistance to both vismodegib and cyclopamine (Fig. 5A–F). To determine whether PGE1 acts on HH signaling via PKA activation in DAOY cells, we overexpressed GLI2 ΔPKA and found that PGE1′s effect on cellular viability was absent (Supplementary Fig. S8C). In addition, PGE1′s inhibitory activity against HH pathway was no longer observed upon knockdown of EP4, reinforcing the role of EP4 for PGE1 (Supplementary Fig. S8D). These results demonstrated PGE1′s effect against tumor cell growth in vitro and the underlining HH pathway activity.

We next focused on GLI2-overexpressed DAOY cells in tumor xenograft studies in vivo, which also gained a growth advantage over wild-type DAOY cells (Supplementary Fig. S8E). Administrations with 15 mg/kg PGE1 on a daily basis led to significant tumor growth inhibition (Fig. 6A and B). On the contrary, saturated treatments with 30 mg/kg vismodegib on a daily basis conferred no inhibition of tumor growth (Fig. 6A and B). Of note, previous studies showed a daily treatment regime with 25 mg/kg vismodegib delivered complete blockade against the growth of Ptch1^+/− murine tumor allografts (54). In agreement with a PGE1 inhibitory action on GLI2, PGE1 attenuated GLI1 and PTCH1 expression in the tumors (Fig. 6C). H&E staining revealed that PGE1-treated tumors showed evidence of necrosis as seen by destruction of organized nests of basophilic tumor cells and pyknosis (black arrowhead), whereas vehicle or vismodegib-treated tumors showed nests of well-organized tumor cells (red arrowhead; Fig. 6D). Consistent with tumor growth inhibition delivered by PGE1, Ki67-positive proliferating cells markedly decreased in comparison with vehicle and vismodegib controls (Fig. 6E and F).

Two caveats can be found in the above DAOY-based studies: one is that the cell line might lose its accuracy in modeling the original tumor after prolonged passaging and in vitro culturing; the other is that xenografting to the flank did not accurately reflect real tumor environment of medulloblastoma—the cerebellum. Therefore, we further tested the effect of PGE1 using two patient-derived orthotopic xenograft models, Med-113FH and Med-314FH, harboring PTCH1 mutation and GLI2 amplification, respectively (55). Both of them express EP4 (Supplementary Fig. S9A). Med-113FH cells stably expressing a firefly luciferase reporter were transplanted into the cerebellum of immune-compromised mice, and tumors were measured with IVIS imaging system. We treated the orthotopic allografts of Med-113FH with vehicle control or 15 mg/kg PGE1, and observed a significant decrease in orthotopic tumor growth in PGE1-treated mice, as well as an increase in overall survival in PGE1-treated mice (Fig. 7A–C). As expected, PGE1 treatment resulted in significant reduction of the expression of endogenous GLI1 and PTCH1 mRNAs (Supplementary Fig. S9B). Meanwhile, we observed a decreased protein level of GLI2 and an elevated protein expression of p-CREB in Med-113FH orthotopic allografts treated with PGE1, supporting a consistent mechanism of action in Med-113FH tumors (Supplementary Fig. S9C). In a similar approach, orthotopic cerebellum tumors implanted with a firefly luciferase–labeled subline of Med-314FH were treated with vehicle control or 30 mg/kg vismodegib or 15 mg/kg PGE1. We observed marked reduction in the growth of Med-314FH orthotopic allografts in response to PGE1 but not vismodegib, together with an improved overall survival for Med-314FH orthotopic allografts treated with PGE1 but not vismodegib (Fig. 7D–F). Consistently, in Med-314FH orthotopic allografts, we observed downregulation of endogenous GLI1 and PTCH1 mRNA expression, decreased GLI2 protein expression, and increased p-CREB protein expression after PGE1 treatment (Supplementary Fig. S9D and S9E).

It has been reported that PGE1 can efficiently cross the blood–brain barrier (BBB) in multiple mammals (56–58). And administration of PGE1 in newborns resulted in neurological and electroencephalographic changes, indicating distribution in the brain (59). Therefore, BBB might not be an obstacle for repurposing PGE1 to treat medulloblastoma, a brain disease. In addition, it was reported that PGE1 is metabolized at a high speed in patients (60, 61). Therefore, we also examined its two main metabolites in human, 13,14-dihydro-PGE1 and 15-keto-13,14-dihydro-PGE1 (62). Both of them inhibit GLI2 activity, the former displays a comparable potency to PGE1, whereas the latter is less active (Supplementary Fig. S10A and S10B), thus suggesting that its fast metabolism unlikely limits its activity against GLI2-driven drug refractory human tumor growth.

To further explore whether additional drug repurposing opportunities exist, lastly we examined 4 FDA-approved prostaglandins (misoprostol, latanoprost, tafluprost, and travoprost) and 1 prostaglandin having undergone clinical trials (rivenprost; Supplementary Fig. S11A). Among them, misoprostol and rivenprost showed similar potency with PGE1 in antagonizing GLI2 activity, whereas the other three, including latanoprost, tafluprost, and travoprost, had no inhibitory effects (Supplementary Fig. S11A). This observation implied that additional opportunities other than PGE1 potentially exist for drug repurposing. Taken together, these results support that PGE1, as a representative of many prostaglandins identified in this study, would provide potential opportunities for further therapeutic translation targeting tumors refractory to current generation SMO antagonists.

Drug resistance is a major and devastating challenge associated with targeted cancer therapies. Those targeting HH pathway is no exception. Superior to the strategy targeting drug refractory SMO mutants, an alternative to target downstream GLI2 amplification, as demonstrated in the current study, shows effects of a broader molecular spectrum and pan-inhibition against multiple drug refractory mechanisms, including SMO mutagenesis. Using PGE1 as a representative of prostaglandins identified from our screening, a novel mechanism of GLI2 regulation triggered by its action on EP4 receptor was discovered (Fig. 4). More importantly, using human medulloblastoma xenograft models that displayed resistance to vismodegib, PGE1 demonstrated significant inhibition against tumor growth (Fig. 6A and B, Fig. 7D–F), thus highlighting potential opportunities for future clinical translation.

We demonstrated PGE1 acts through EP4 to inhibit the HH signaling. Intriguingly, we observed EP4 localization on the PC, the central cellular organelle for HH signaling transduction (Fig. 4G). A hypothetical model was proposed where PGE1 binds to EP4 receptor on the cilium, thereby activating cAMP-PKA signaling to promote GLI2 phosphorylation and subsequent degradation in an ubiquitin-proteasome–dependent manner (Fig. 4H). In support of the model, we found that GLI2 ΔPKA was resistant to PGE1 treatment and PKI rescued PGE1′s effect on GLI2 stability and transcriptional activity (Fig. 3C–E).

Current study used PGE1 as a representative for mechanistic investigations. Nonetheless, several prostaglandins showed similar pan-inhibition against pathway activation at multiple levels, including that associated with drug refractory mechanisms (Figs. 1D and 2B; and Supplementary Fig. S4A–S4E). It was reported that although functioning through distinct receptors, intracellular signaling transduction stimulated by these prostaglandins converges into cAMP elevation (63). In contrast, latanoprost, tafluprost, and travoprost, which are known to enhance cellular Ca²⁺ levels through FP receptor (63), but have no effect on cAMP level, are inactive in inhibition of GLI2 activity (Supplementary Fig. S11A). We observed that FP receptor is expressed in the NIH/3T3 cells used in this experiment (Supplementary Fig. S11B), thus ruling out the possibility that lack of inhibition against GLI2 activity is due to lack of FP receptor expression.

The concentrations of PGE1 required for full inhibition of HH pathway activity in our cell-based assays (5–10 μmol/L, Fig. 2B and H; and Supplementary Fig. S5A) are comparable with clinically relevant level at 10 μmol/L reported for human cardiac regeneration (64), implying that dosing unlikely being a limiting factor for further clinical translation. It is also worth pointing out that PGE1 is not the only FDA-approved drug among active prostaglandins identified in the current study (Supplementary Fig. S11A). Therefore, we provided a rich source of potential opportunities for medicinal chemistry optimization, if necessary before entering clinical trials, and drug repurposing, taking advantages of their well-characterized pharmacokinetics and safety profiles.

No potential conflicts of interest were disclosed.

Conception and design: F. Wu, Y. Wang

Development of methodology: F. Wu, C. Zhang, C. Zhao

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): F. Wu, C. Zhang, Z. Teng

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): F. Wu, H. Wu, T. Jiang, Y. Wang

Writing, review, and/or revision of the manuscript: F. Wu, T. Jiang, Y. Wang

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): C. Zhang, C. Zhao, H. Wu

Study supervision: T. Jiang, Y. Wang

We highly appreciate Dr. Andrew P. McMahon (University of South California, Los Angeles) for reading and discussion of this article. We are grateful to Dr. Yongbin Chen (Kunming Institute of Zoology, Chinese Academy of Sciences, Kunming, China) for sharing p8 × GLIBS-firefly luciferase plasmid. We thank the Brain Tumor Resource Lab at Fred Hutchinson Cancer Research Center for offering Med-113FH and Med-314FH human SHH medulloblastoma lines. We thank Dr. Changmei Liu (Institute of Zoology, Chinese Academy of Sciences, Beijing, China) for the gift of mouse ovary samples. We also thank Dr. Wei Li (Institute of Zoology, Chinese Academy of Sciences, Beijing, China) for assisting with protein ubiquitination techniques. We would like to thank colleagues in our lab for helpful discussions and our colleagues from the Zhongguacun Park Campus of the Institute of Zoology at the Chinese Academy of Sciences for sharing instruments, technical assistance, and helpful discussions. This study was supported by the National Natural Science Foundation of China (No. 91957121, 31571514 to Y. Wang), Beijing Municipal Natural Science Foundation (No. Z190013 to Y. Wang), Capital's Funds for Health Improvement and Research (No. CFH 2018-2-2042 to T. Jiang), the Hundred Talents Program of Chinese Academy of Sciences, and State Key Laboratory of Stem Cell and Reproductive Biology (to Y. Wang). A patent covering the novel findings of prostaglandins and analogs modulating HH pathway, GLI2 activity, and tumor growth has been filed.

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.