Interfering with Resistance to Smoothened Antagonists by Inhibition of the PI3K Pathway in Medulloblastoma

Emergence of resistance to Smo inhibition

NVP-LDE225 is a potent and selective oral Smo antagonist from a novel structural class (Supplementary Fig. 1)(5). This molecule displaces the binding of the synthetic Smo Agonist 1.5 (6) to human and mouse Smo with an IC₅₀ of 11 and 12 nM, respectively, and in low nanomolar concentrations inhibits Hh-signaling in human and mouse cells (Supplementary Table 1) (5). In medulloblastoma tumors derived from Ptch^+/−p53^−/− mice (7) and implanted into nude mice, expression of the Hh pathway target gene Gli1 was completely suppressed by the oral administration of 20 mg/kg/day qd of NVP-LDE225 (Fig. 1A). Consistent with the dose-response for suppression of Gli1 mRNA (data not shown), treatment of tumor-bearing mice with NVP-LDE225 induced partial growth inhibition at 10 mg/kg/day and near complete regressions (84 to 92%) beginning at doses of 20 mg/kg/day (Fig. 1B). However, on day 13 of continuous dosing of NVP-LDE225 tumor re-growth was observed in all treatment groups indicating the development of resistance. Resistant tumors had a more heterogenous histological appearance compared to sensitive tumors (Supplementary Fig. 2). The development of resistance was also seen in mice treated with HhAntag (Fig. 1B)., a Smo antagonist from a structurally distinct class (8).

Although complete suppression of Gli1 mRNA in response to initial NVP-LDE225 treatment was seen, re-expression of Gli1 mRNA was observed in resistant tumors (Fig. 1A). A similar pattern of re-expression in resistant tumors was observed for several Hh pathway target genes such as Ptch1, Ptch2 and CyclinD1 (2) (Supplementary Fig. 3A). These data show that acquired resistance to Smo inhibition is associated with reactivation of Hh signaling.

To determine whether resistance and Gli1 reactivation was unique to the Ptch^+/−p53^−/− model, similar experiments were carried out using allografts derived from Ptch^+/−Hic1^+/− mice (9). Hypermethylated in Cancer-1 (Hic1) is a frequent target of epigenetic gene silencing in medulloblastoma (9). In the context of a Ptch^+/− background, heterozygous deletion of Hic1 leads to the development of Hh-pathway dependant medulloblastoma. Thus, nude mice were implanted subcutaneously with medulloblastomas derived from Ptch^+/−Hic1^+/− mice and treated with NVP-LDE225 orally at similar doses and schedules. Pronounced inhibition of Gli1, Ptch1, Ptch2 and CyclinD1 mRNA expression (Supplementary Fig. 4A and 5) and tumor regression (Supplementary Fig. 4B) were observed during initial treatment, followed by re-growth of some tumors (Supplementary Table 2) that had restored a variable degree of Hh pathway target gene expression1 expression (Supplementary Fig. 4A and 5). Similar observations were made in the Ptch^+/− allograft model (data not shown). The rapid emergence of resistance to NVP-LDE225 with restoration of Hh pathway target gene expression expression in both models is consistent with a marked addiction to Hh signaling in Ptch^+/− medulloblastoma. To confirm that NVP-LDE225 resistance resulted from a cell autonomous rather than an extrinsic mechanism such as decreased drug exposure, the ex-vivo proliferation (10) of tumor cells freshly isolated from sensitive (pre-treatment) and resistant tumors was measured in response to NVP-LDE225. Whereas growth of sensitive medulloblastoma cells was robustly inhibited (IC₅₀ 6 nM), growth of resistant tumor cells was largely unaffected by exposure to NVP-LDE225 (IC₅₀ >20 μM) (Fig. 1C) or to other Smo antagonists such as HhAntag and cyclopamine (Supplementary Table 3). Together these data suggested that cell autonomous mechanisms caused resistance to Smo inhibition in these tumors.

Re-activation of Hh signaling in resistant tumors

Next, we took an unbiased approach to the discovery of potential resistance mechanisms. Specifically, expression profiles of Ptch^+/−p53^−/− tumors were obtained using Affymetrix murine gene expression arrays, and were queried in a hypothesis-directed mode for signatures corresponding to a pattern of expression that was down-regulated upon short-term treatment with NVP-LDE225, but re-expressed in resistant tumors (Fig. 2A). Genes ordered by Spearman rank correlation with this pattern are shown in Fig. 2B and C. Gli1 emerged as the top ranking gene, closely followed by CyclinD1, Ptch1 and Hhip, three bona fida Hh pathway target genes (2). Using a Fisher’s Exact Test to probe gene-sets extracted from GeneGo Metacore database (11), the top ranked genes were shown to be enriched in the transcriptional targets of Gli2 (p=0.001, false discovery rate (FDR)=0.04) and Gli1 (p=0.003, FDR=0.08). Sensitive and resistant tumors from the Ptch^+/−Hic1^+/− model were subjected to the same analysis and also showed significant regulation of Gli2 (p=0.006) and Gli1 (p=0.04) target genes. In aggregate, these data confirm and expand on our initial observation that Hh signaling is reactivated in resistant tumors.

Smo mutations in resistant tumors

Resistance to inhibitors of BCR-ABL, c-KIT and EGFR1 kinases is most often ascribed to the development of mutations in the direct drug target (4). In addition, recently a mutation in Smo was described as mechanism of resistance to the Smo inhibitor GDC0449 (12). To determine whether point mutations in Smo might account for resistance to NVP-LDE225 and Gli reactivation in both medulloblastoma models, genomic DNA isolated from resistant tumors was subjected to PCR-based amplification and sequencing of all coding Smo exons. Surprisingly, missense point mutations in Smo were detected in only seven out of 135 resistant tumors, resulting in changes of leucine 225 to arginine (L225R), asparagine 223 to aspartic acid (N223D), serine 391 to asparagine (S391N), aspartic acid 388 to asparagine (D388N) and glycine 475 to serine (G457S) (Table 1 and Supplementary Fig. 6). All five mutations differ from the previously described aspartic acid 477 to glycine (D477G) mutation (12) but similarly abrogate or decrease Smo inhibition by NVP-LDE225 (Table 1 and Supplementary Fig. 7). Importantly these data suggest that nucleotide-based mutations in Smo are not a dominant driver of resistance to NVP-LDE225.

Gli2 amplifications in resistant tumors

To determine whether other genetic abnormalities including chromosomal alterations might account for resistance to Smo inhibition, genomic DNA isolated from three sensitive and three resistant Ptch^+/−p53^−/− tumors was subjected to genome-wide array-based comparative genomic analysis (aCGH). Two of the three resistant tumors showed a focal amplification of a region on chromosome 1 (1qE2-4) containing Gli2 (Fig. 2D). Next, copy number variations in Gli1, 2 and 3 was determined by quantitative PCR in a larger set of resistant tumor samples. This analysis revealed Gli2 amplification in 50% of Ptch^+/−p53^−/− (Fig 2E) and 20% of Ptch^+/− resistant tumors (data not shown), but not in sensitive tumors, nor in any of the Ptch^+/−Hic^+/− resistant tumors. No copy number changes for Gli1 and Gli3 were detected (data not shown). Gli2 amplifications appeared to be less frequent at higher doses but larger sample numbers will be required to draw a firm conclusion. In keeping with the notion that amplification would result in up-regulation of Gli2 mRNA levels, we found a strong correlation between expression and amplification where Gli2 mRNA expression was increased 2- to 20-fold in Gli2-amplified resistant tumors compared to sensitive tumors (Fig. 2F). A small number of resistant tumors demonstrated elevated Gli2 mRNA expression in the absence of clear amplification, suggesting that alternative mechanisms may also lead to Gli2 mRNA up-regulation. In the Ptch^+/−Hic^+/− model, Gli2 expression was not elevated compared to sensitive tumors (Suppl. Fig. 5B)

The functional significance of Gli2 amplification was investigated in freshly isolated cells from a Gli2-amplified and Gli2 overexpressing, NVP-LDE225-resistant tumor transfected with Gli2 siRNA. As shown in Fig. 2G, treatment with Gli2 siRNAs resulted in a 50% to 70% knock-down of Gli2 mRNA. This was correlated with inhibition of proliferation, and also led to the suppression of Gli1 mRNA, a well defined transcriptional target of Gli2 (13). Moreover, preliminary data indicated that knock-down of Gli2 resulted in the partial reconstitution of sensitivity to NVP-LDE225. The IC₅₀ for growth inhibition for NVP-LDE225 shifted from >20 uM to 0.1 μM in the presence of Gli2 knock-down (data not shown). These data show that Gli2 is a critical effector of tumor cell growth downstream of mutations in the Ptch receptor, and can act as a mediator of resistance to Smo inhibitors.

Upregulation of Igf-R/PI3K signaling in resistant tumors

Gli2 amplification was frequent in resistant tumors, however it was not uniform. Furthermore, Gli2 amplifications were not detected in Ptch^+/−Hic1^+/− resistant tumors, suggesting that tumors may escape Smo inhibition through alternative routes. To search for such a mechanism, we probed the Ptch^+/−p53^−/− medulloblastoma expression dataset with an expression vector emphasizing the emergence of new or up-regulated pathways not regulated by short-term treatment with NVP-LDE225 (Fig. 3A). We created a rank-ordered list of genes differentially expressed (by t-test) between resistant tumors and all other tumors (control and treated sensitive), and reapplied the previously-described gene-set analysis in order to detect signaling pathways preferentially enriched during the emergence of resistance (Fig. 3B, C and Supplementary Table 4).

Strikingly, in this analysis three of the most highly ranked pathways (Akt, PIP3 and Igf-1R) were directly related to Igf-1R/PI3K signaling (Fig. 3B), strongly suggesting that a compensatory upregulation of PI3K/mTor signaling might contribute to development of resistance. Intriguingly, this mechanism of resistance has been observed previously with EGFR inhibitors (4).

The 29-gene PI3K gene-set in Fig. 3C derived from the Akt, PIP3 and Igf-1R pathway category was surveyed across an additional 3 vehicle (sensitive) and 16 resistant Ptch^+/−p53^−/− tumors (Supplementary Fig 8). The “PI3K signature score” of each tumor, defined as the average z-score of all genes in the 29-gene set, was upregulated (p<0.01) in 11/16 resistant tumors but not in the three sensitive tumors. Gli2 amplification and Smo mutation were detected in 6/16 and 1/16 resistant tumors, respectively. Upregulation of the PI3K signature was detected in tumors with and w/o Gli2 amplification.

To address the functional relevance of the PI3K upregulation, we asked whether a combination of NVP-LDE225 with the PI3K class I inhibitor NVP-BKM120 could block the development of resistance. Following continuous treatment with NVP-LDE225, tumors derived from the Ptch^+/−p53^−/− model regressed but then started to re-grow as seen previously (Fig. 3D and E). However, the combination of NVP-LDE225 and NVP-BKM120 delayed tumor re-growth. NVP-BKM120 as single agent, on the other hand, had no effect on tumor growth. Western blot analysis of tumors from different treatment group demonstrated increased PI3K/mTor pathway activation in resistant tumors as measured by phosphorylation of S6 and 4EBP1 and its inhibition in tumors treated with NVP-BKM120 or the combination (Fig. 3I).

We expanded our studies to the Ptch^+/−Hic^+/− model that lacked Gli2 amplifications but similarly to the Ptch^+/−p53^−/− demonstrated upregulation of the PI3K/mTor pathway in resistant tumors (Supplementary Table 5). Combination treatment of Ptch^+/−Hic^+/− tumors with NVP-LDE225 and the dual PI3K/mTor inhbitor NVP-BEZ235 resulted in a marked delay of tumor re-growth compared to treatment with single agent NVP-LDE225 (Fig. 4A and B). Moreover, while only 1 complete regression was observed after 61 days of treatment in the group treated with NVP-LDE225 alone, 5 complete regresssions were observed in the combination group (Supplementary table 6). Similar results were obtained following treatment of NVP-LDE225 in combination with the mTOR inhibitor RAD001 (Supplementary Fig. 9A and B). In keeping with these data PI3K/mTor pathway activation as measured by S6 and 4EBP1 phosphorylation was up-regulated in resistant tumors and was inhibited in tumors treated with NVP-BEZ235 (Fig. 4C) or RAD001 (Supplementary Fig. 9C).

In summary, our data in two different models of medulloblastoma using three different inhibitors of the PI3K/mTor pathway provide a very strong rationale that the combination of Smo antagonists with PI3K/mTor inhibitors may delay or prevent the development of resistance to Smo inhibitors in medulloblastoma.

Medulloblastoma allograft studies

Tumors derived from Ptch^+/−p53^−/− and Ptch^+/−Hic^+/− transgenic mice were serially passaged as fragments in nude mice. For efficacy studies tumor fragments were dissociated into single cells and 5×10⁶ cells were allografted into nude mice and treated with small molecule inhibitors as previously described (5,24). Tumor volumes were measured three times a week and calculated using the ellipsoid formula: (length × width²)/2. NVP-LDE225 was formulated as diphosphate salt in 0.5% Methylcellulose and 0.5% Tween 80 (Fisher), NVP-BKM120 in 0.5% Methylcellulose, NVP-BEZ235 in 1 volume of NMP (1-methyl-2-pyrrolidone, Sigma-Aldrich) and 9 volumes of PEG300 (Sigma-Aldrich), and RAD001 in water. Doses are expressed as free-base equivalents. Tumors were harvested for analysis 4 h after the last dose. All animal studies were carried out according to the Novartis Guide for the Care and Use of Laboratory Animals.

Ex-vivo medulloblastoma assay

Tumors were minced and a single cell suspension was prepared using the Papain dissociation system (Worthing Biochemical Co) as previously described (10). Cells were resuspended in serum-free Neurobasal medium (with B27 supplement) (Invitrogen) and plated in 96-well plates at a density of 3×10⁵ cells/well in 200 ul medium. Serial dilutions of inhibitors were prepared in DMSO and added at 1 μl per well. Cells were incubated for 48 hours and [³H]-thymidine was added for the last 8 hours to assess cell proliferation. Incorporated radioactivity was quantified as previously described (10). siRNA transfections with scrambled siRNAs (ON-TARGETplus Non-targeting Pool, Dharmacon) and mGli2 siRNAs (Gli2 SiGENOME, Dharmacon) were applied using the DharmaFECT1 transfection reagent according to the manufacturer’s instructions (Dharmacon).

Immunoblot analysis

Cell lysates were prepared and analyzed by immunoblot for phospho-S6 (S235/236), total S6, phospho-4EBP1 (T37/46) and total 4EBP1 (CST #2211, 2217, 2855 and 9452, respectively) as described (25)

Smo binding and cell-based assays

Agonist displacements assays and TM3-Gli-luciferase assays were performed as previously described (24). HEPM cells (ATCC #CRL-1486) were cultured in Minimum Essential Medium (Gibco) supplemented with 10% FCS. Cells were plated at 5×10⁴ cells/well in 96-well plates and stimulated with recombinant SHH (R&D systems 1845-SH) for 48 h in the presence of serial dilutions of NVP-LDE225. Gli1 mRNA levels were determined at the end of the assay as described below. Smo point mutations were introduced into pcDNA3.1 containing HA-tagged Smo using the QuickChange Lightning Site-directed mutagenesis kit (Strategene). Wild-type and mutant smo constructs were transiently expressed in C3H10T1/2 (ATCC #CCL-226) with a Gli-luciferase reporter and pRL-TK expressing renilla luciferase using GeneJuice transfection reagent as described (12). Serial dilutions of NVP-LDE225 (0.001 to 10 μM final concentration in assay) were added 24 hours after transfection. Firefly and renilla luciferase activity was detected after an additional 24 h with the Dual-Luciferase reporter assay system (Promega). Percent inhibition was calculated relative to DMSO control.

Gene expression analysis

RNA isolation from tumors, cDNA synthesis and real-time quantitative PCR for Gli1 were performed as described (24). PCR probes used: mGli1: Mm494646m1 mGli2 Mm01293117m1,, mPtch1: Mm00436026m1, mPtch2: Mm00436047m1, mCyclinD1: Mm00432359m1 (Applied Biosystems).

Generation of labelled cDNA and hybridization to 430_2 murine arrays (Affymetrix) were performed as described (26). Expression values were normalized using the Affymetrix MAS5.0 algorithm. Probe sets with MAS5.0 expression below 100 in 90% of samples were excluded from analysis. The remaining probe sets were compared with the pattern in Figure 2A using a Spearman rank correlation, and tested for differential expression as in Figure 3A using a homoscedastic t-test. Using the gene selection methods above, probesets with nominal p-values <0.01 were assessed for membership in gene sets (transcription factor sets and canonical pathways) extracted from the GeneGo Metacore database (11). Significance values were calculated using a Fisher’s Exact test and Benjamini-Hochberg FDR correction (27). Genes with multiple probesets were considered for set membership if any constituent probesets met the selection criteria. Mouse genes were converted to human homologs using the NCBI homologene database, August 2009 build (http://www.ncbi.nlm.nih.gov/homologene).

DNA copy number and sequence analysis

Genomic DNA was extracted from tumors using the DNeasyBlood and Tissue kit (Qiagen) and subjected to copy number analysis with the Agilent Mouse CGH 244K array containing 244,000 features with a median probe spacing of 7.8 Kb per manufacter’s instructions. The data was analyzed using the Agilent G4175AA CGH

Analytics 3.4 software for copy number alterations. Genomic copy number for Gli1, Gli2 or Gli3 was determined using custom designed quantitative PCR reagents synthesized by Applied Biosystems (Gli1: forward primer CATTGCCTTTTCTCCTTGTCATCTG, reverse primer GGCGGTCCAGGGAGACT, probe CACCTGTGTCTCGCCGTC; Gli2: forward primer CCCGTGGGTCTTCTCTCTGA, reverse primer GACAGGGCTGCCACTTAGG, probe CCTCCACAGGCCTCC; Gli3: forward primer CTCATCTTTTCCCTGCCTTCCA, reverse primer ACATGTAATGGAGGAATAGGAGATGGA, probe CCTCATGATGTCTGGCATC). qPCR was carried out in 384-well plate and run with the 7900HT Fast Real-Time PCR System (Applied Biosystems) using the default cycling method (50°C for 2 minutes, 95°C for 10 minutes followed by 40 cycles at 95°C for 15 seconds and 60°C for 1 minute). The reaction volume was 12 ul and the mixture included 900 nM of each primer, 250 nM of probe, 1X taqman universal master mix (containing PCR buffer, nucleotides and Taq DNA polymerase) and 10 ng genomic DNA template. Copy numbers were calculated with 6 replicates reactions for each DNA sample with a standard curve constructed with four replicates of 4-fold serial dilution of normal genomic DNA template, ranging from 40 ng to 0.04 ng. Copy number was calculated as 2(T_DNA/C_DNA) where T_DNA and C_DNA are the calculated amount of test gene DNA at the recorded Ct for tumor and calibrator respectively. Copy number data for Gli2 was normalized to Gli3 using the formula 2(T_dna/L_dnaT)/(C_dna/L_dnaC) where T_dna and C_dna represent the amount of Gli2 calculated amount DNA in the tumor and calibrator respectively and L_dnaT and L_dnaC are the corresponding amount of their Gli3 DNA at the recorded Ct values. Mutation analysis was performed by sequencing of PCR-amplified exon sequences of Smo at Agencourt.0