The hedgehog’s tale: developing strategies for targeting cancer

BCC

In the case of BCC, it has been estimated that essentially all tumors show evidence of constitutive HH pathway activity with 90% exhibiting loss of PTCH1 and 10% activating mutations in SMO²⁹. Although the cell of origin of most cancers remains elusive, good candidates have been identified for BCC. Previously, it was suspected that BCC arose from bulge stem cells in the hair follicle³⁰. However, targeted expression of an activated Smo gene in mice suggested that long-term resident progenitor cells of the interfollicular epidermis and upper infundibulum gave rise to tumors resembling BCC³¹. In contrast, in irradiated Ptch1^+/− mice, BCC were found to arise from keratin-15 expressing stem cells of the follicular bulge³². One explanation for the discrepancies in these results is that PTCH1 and SMO have additional distinct functions beyond the HH pathway that also influence tumorigenesis. For example, PTCH1 can sequester cyclinB in the cytoplasm³³ and this was observed in BCC from Ptch1^+/− mice, but not in those expressing activated SMO³². Furthermore, it appears that loss of p53 promotes BCC formation, in part, by increasing SMO expression³². The effect of p53 loss on SMO expression might also explain why homozygous loss of Trp53 dramatically accelerates medulloblastoma formation in Ptch1^+/− mice, whereas loss of Arf or heterozygous loss of Trp53, neither of which affect the transcriptional function of p53, do not increase tumorigenesis in Ptch1^+/− mice ³⁴. These findings show that activation of the HH pathway through loss of PTCH1 or by expression of a mutated form of SMO are mechanistically distinct. As pointed out by Wang et al³², although the HH pathway can be activated in cells that do not normally express SMO by expression of activated SMO or over expression of GLI2, loss of PTCH1 in cells that do not express SMO does not achieve the same effect.

The targeted expression of an activated form of GLI2 in stem cell populations in the skin generated a range of tumor phenotypes, depending on the cell of origin and level of GLI2 expression ³⁵. This study also showed that it is important to examine the histology of mouse skin tumors carefully as they may represent only a subset of the range of BCC types found in humans. Therefore, we should be cautious when comparing and interpreting data obtained from models based on different approaches.

Medulloblastoma

The situation in MB is less uniform, with up to 30% of tumors showing a gene expression signature indicative of HH pathway activation, but only half of these associated with loss of PTCH1, loss of SUFU or gain-of-function SMO mutations^36–38. Thus, a subset of MB exhibits HH pathway activation without evidence of mutations in PTCH1, SMO or SUFU.

Historically, medulloblastoma research was confined to a handful of cell lines and xenografts, although the limitations of such models, including their low predictive value, were well recognized³⁹. However, over the past several years, there has been explosive growth in the availability genetically engineered mouse (GEM) models for brain tumors. In MB, the majority of these models develop as a consequence of HH pathway activation⁴⁰. Mice heterozygous for Ptch1 provide a good model of Gorlin’s syndrome, exhibiting several of the phenotypes, including an increased incidence of MB (up to 25%, depending on the genetic background of the mice)^40–43. Tumor incidence increases to 100%, and the age of onset is decreased, in the absence of p53³⁴. Significant acceleration is also observed following exposure to ionizing radiation⁴⁴. Alterations in genes that function in DNA double-strand break repair and cell cycle control also collaborate with p53 loss to cause medulloblastoma in mice^45–47. Mutations in DNA repair genes seem to function by increasing the likelihood of chromosomal alterations in Ptch1 in mouse medulloblastoma⁴⁷, but mutations in DNA repair genes have not been described in human medulloblastoma. Additional mouse models of medulloblastoma have also been created by expressing transgenes with activating point mutations in SMO, that were first identified in BCC^{48, 49}, under the control of the Neurod2 promoter^{50, 51}. Tumors arising in one other mouse model, with heterozygous or homozygous loss of chemokine receptor type 6 (Cxcr6), exhibit an activated HH pathway and although Ptch1 is not mutated its expression is suppressed⁵². In addition, targeted expression of GLI1, GLI2 or HH ligands can induce medulloblastoma and tumor formation in the skin^53–55. Targeting expression of activated β-catenin to neural precursor cells in mice also results in the development of medulloblastoma, but in this case the WNT pathway is activated in tumors rather than the HH pathway⁵⁶. At present, the available genetically modified mouse models do not recapitulate the full genetic diversity of human medulloblastoma, but they do reflect many of the properties of human medulloblastoma⁴⁰. These models have made significant contributions to our understanding of the etiology of medulloblastoma and they have provided invaluable tools for proof-of-concept analysis of drugs that target the HH pathway.

Recent studies on the cell of origin of medulloblastoma have identified two potentially different tumor subtypes: that driven by HH pathway mutations and that driven by mutations in the WNT pathway. The origin of HH-medulloblastoma was suggested by the close resemblance of gene expression patterns in medulloblastoma cells with those of granule neuron progenitor cells⁵⁷. Genetic studies showed that, while activation of the HH pathway in neural stem cells, as well as granule neuron progenitor cells, results in medulloblastoma formation, stem cells must first make the transition to committed granule neuron progenitors^{58, 59}. In this case, it appears that oncogenesis results from constitutive activation of a signaling pathway, normally active in the cell of origin, which would otherwise be down regulated during the course of development. The WNT molecular subtype of medulloblastoma exhibits a distinct pattern of gene expression and appears to be derived from progenitor cells in the dorsal brainstem in which the HH pathway is inactive⁵⁶. However, the situation might not be so clear-cut, as analysis of the medulloblastoma genome recently revealed that some tumors can harbor both HH and WNT pathway mutations simultaneously, indicating that there may be some overlap among the cells of origin⁶⁰.

Rhabdomyosarcoma

Soft tissue tumors very similar to human rhabdomyosarcoma (RMS) are also seen in Ptch1^+/− mice, but the incidence is strongly influenced by genetic background⁶¹. Initial descriptions of Gorlin syndrome identified benign fetal rhabdomyoma and rare instances of several other tumors, including RMS⁶². Although it is often stated that RMS is associated with activating mutations in the hedgehog pathway, the role of the HH pathway in sporadic human RMS is not straightforward. RMS is a heterogeneous family of tumors, associated with the skeletal muscle lineage, encompassing several distinct molecular and histological subtypes, none of which have been specifically linked to HH pathway mutations⁶³. DNA sequence analysis of the PTCH1 coding region from 14 cases of RMS did not reveal any mutations⁶⁴. Although another study reported mutations in PTCH1 and SUFU in RMS, most of the analysis was restricted to loss of heterozygosity (LOH) analysis of PTCH1 and SMO in rhabdomyoma and only one case of RMS⁶⁵. Recently, low level amplification of GLI1 was reported in embryonal RMS and it was associated with increased expression of HH pathway target genes⁶⁶. In addition, comparison of gene expression patterns in human embryonal RMS indicated that approximately 29% exhibited evidence of HH pathway activity but this was always in association with either p53 or RB pathway signatures⁶⁷. However, the presence of the pathway signature does not necessarily mean that the tumor cells are dependent on SMO activity as analysis of the effects of cyclopamine on RMS in Ptch1^+/− mice showed that, while loss of PTCH1 may contribute to tumor initiation, it is not required for tumor maintenance⁶⁸.

Other types of cancer with altered HH pathways

Several different molecular lesions in the HH pathway have been described in tumors and, in each case, they result in increased transcriptional activity of the GLI1 and GLI2 transcription factors. Indeed, the first indication that genes in the HH pathway were associated with human cancer was the observation that GLI1 was amplified in glioblastoma⁶⁹. Although examples of GLI1 amplification have been described in childhood sarcoma⁷⁰, this does not appear to be a common primary mechanism underlying glioblastoma formation. The location of several other potential oncogenes (including MDM2 and CDK4) within the same amplicon has complicated interpretation of these data⁷¹. Translocations involving GLI1 and ACTB (encoding β-actin) have defined a new class of pericytoma⁷².

Many other tumors lacking HH pathway mutations have been reported to be sensitive to SMO inhibitors in vitro and in vivo¹⁴. This has led to estimates that up to 25% of human tumors may depend on HH pathway activity for growth⁷³ and, as a consequence, a broad range of tumors was included in the early clinical trials of SMO inhibitors. Initially, these findings were interpreted to suggest that there was an autocrine feedback loop in some tumors expressing HH ligands^74–77 However, subsequently, it was shown in xenograft models that while some human tumors do make HH ligands, the target genes upregulated by these ligands are expressed in stromal cells of mouse origin⁸. In these cases, treatment with SMO inhibitors slowed tumor growth, but did not eliminate tumors under standard xenograft conditions⁸. In a mouse model of pancreatic cancer, inhibition of HH signaling enhanced the delivery and response to chemotherapy by depleting tumor-associated stromal tissue⁷⁸. Thus, in addition to targeting tumor cells directly, SMO inhibitors may also show therapeutic benefit through effects on stromal cells.

Some leukemias were also reported to depend on SMO signaling for growth^{79, 80}, but this was disputed in subsequent studies on B-cell chronic lymphocytic leukemia indicating that although GLI1 had an autonomous role in promoting cell survival, this was not dependent on HH ligands or SMO⁸¹. The large body of work on the role of SMO signaling in cancer models supports a broader use of SMO inhibitors in tumors other than BCC and a subset of MB. The difficulty at present is to identify the tumors that would benefit from treatment, as the presence of HH ligands or a HH pathway signature alone, does not guarantee a response.

Preclinical development

Cyclopamine, and other naturally occurring inhibitors of SMO, are not suitable as therapeutic agents because of poor solubility, low potency, rapid clearance, non-specific toxicity and chemical instability⁸⁵. This encouraged a search for novel inhibitors with preferential characteristics for drug development. HH-mediated GLI transcription activity provided a highly appropriate biomarker for identification of small molecule inhibitors as GLI1 and GLI2 contribute directly to oncogenic activity^{55, 86–88}. Cell-based screening approaches, using GLI-dependent transcriptional reporters, proved remarkably adept at identifying small molecule inhibitors of SMO^{89, 90}. Thus, it appears that SMO is a highly druggable target and a range of compounds, many with distinct structures from cyclopamine, have been uncovered that are capable of binding to the same site to block signaling (reviewed in⁹¹). Medicinal chemistry approaches optimized the properties of several lead compounds resulting in a 100-fold increase in potency compared to cyclopamine as measured in gene expression assays^{8, 92, 93}.

Pre-clinical proof-of-concept studies of SMO inhibitors, to establish efficacy in cancer models, have been both challenging and controversial. The advent of GEM models of cancer offered hope that new approaches would be more predictive than the venerable xenograft models. Indeed, a study of acute myeloid leukemia (AML) allograft transplant models, expressing different translocation fusion genes, shows close concordance with responses seen in the respective pediatric populations⁹⁴. Nevertheless, the issue remains hotly debated with strong proponents arguing in favor of xenograft models, others supporting GEM models and some adamant that mouse models of any kind cannot predict clinical outcome.

Most studies of HH pathway activity in cancer employed either human tumor cell lines or xenograft transplantation models treated with cyclopamine to demonstrate dependency on HH pathway activity. Reduced rates of cell proliferation or tumor growth were interpreted to mean that SMO activity was critical for growth of the respective tumors^{74, 75, 95}. However, different groups have reported contradictory results, even when using the same cell lines or mouse strains carrying identical transplantable tumors^{8, 96–100}. Taken together, a number of variables in the experimental approaches and application of the model systems used can account for these discrepancies.

There are no standard approaches to determining whether a particular cell line or tumor depends on HH pathway activity. Therefore, different marker genes have been used to assess pathway activity, and a range of quantitative and semi-quantitative approaches employed to measure gene expression levels, making it problematic to compare results from different studies. Cyclopamine is a toxic compound⁸⁵ and it can cause growth inhibition, independently of HH pathway activity, when used at high concentrations depending on the cell type^{8, 101}. This means that it is critical to use cyclopamine at a concentration that inhibits the HH pathway but does not result in non-specific growth inhibition. This is very hard to do because there is no agreed up standard method to measure HH pathway activity. As a consequence, while some studies report concentrations of cyclopamine of 3 μM and above result in non-specific growth inhibition^{8, 101}, others claim that cyclopamine can function as a specific HH pathway inhibitor at levels up to 30 μM¹⁰⁰. Although several studies employed complementary genetic approaches, such as short interfering RNAs to inhibit GLI1 or GLI2, as independent confirmation of the effects of cyclopamine, this approach does not demonstrate dependence on SMO. For example, GLI1 and GLI2 have been shown to function independently of HH pathway activity and SMO in B-cell chronic lymphocytic leukemia⁸¹. Thus, the fact that GLI1 and GLI2 are required for tumor cell viability does not necessarily mean that SMO is active in these cells and therefore does not confirm the effects of cyclopamine. In vivo, it is difficult to reach systemic levels of cyclopamine that completely inhibit HH pathway activity because of associated toxicities^{85, 101}. Thus, it is likely that the role of HH pathway activity in human cancer has been overestimated as a consequence of the prevalent use of cyclopamine at toxic concentrations.

In many cases, alternate routes of administration were used to deliver cyclopamine. For example, in some instances cyclopamine was delivered by subcutaneous inoculation^{95, 96}. This often caused ulceration at the injection site as the vehicle contained alcohol. This route is not directly comparable to systemic delivery and it is not clear if it resulted in the same amount of drug delivery to the tumor cells. When the route of administration was changed to oral delivery, it was not possible to achieve the same degree of tumor inhibition¹⁰². In most cases the level of cyclopamine achieved in tumor tissues was not determined, however, subcutaneous inoculation may have resulted in greater bioavailability, increased levels of exposure and higher specific, as well as non-specific, toxic effects.

Of much greater concern are studies in which cyclopamine was injected directly into the tumor mass¹⁰³. In some studies the administration route was described as both proximal and directly into the tumor mass^{104, 105}. In these cases, the actual concentration of cyclopamine to which tumor cells are exposed is extremely high and would likely result in non-specific toxicity. Finally, although the standard xenograft approach recommends treating tumors only after they are fully established as transplants (approximately 200 – 400 mm³) some studies used much smaller tumors, even as little as 10 mm³, before they are fully established¹⁰⁵. Thus, a series of methodological differences in preclinical experimental design, combined with the high degree of non-specific toxicity associated with cyclopamine, may account for some of the conflicting findings reported in the literature. These reports may have significantly overstated the potential of SMO inhibitors for treating human cancer.

Compounding these problems is the fact that pharmaceutical companies generally decline to supply compounds currently in development for such studies, because of fears that negative results would compromise the approval process. Therefore, most investigators relied on the use of cyclopamine, despite reservations about its properties, to test the contribution of SMO to tumor growth in vitro and in vivo. The notable exception was a set of compounds identified by Curis Inc., and subsequently developed in collaboration with Genentech Inc., that were used to show efficacy of SMO inhibitors in BCC explant cultures^{29, 106} and in mouse models of medulloblastoma^{52, 101}. One particular compound, a benzimidazole termed HhAntag, provided the most compelling preclinical data on the efficacy of SMO inhibitors. Oral delivery of HhAntag eradicated large medulloblastomas arising spontaneously in the cerebellum of Ptch1^+/−;Trp53^−/−mice¹⁰¹. However, tumor cell lines derived from these medulloblastomas, as well as allograft tumors made using these cell lines, were completely resistant to the inhibitory effects of HhAntag because the HH pathway was dramatically downregulated as soon as these cells were propagated in culture. By contrast, allografts derived directly from the medulloblastomas that were never grown in culture exhibited dramatic sensitivity, with large tumor masses (200 mm³) regressing after only 4 days of treatment⁵². These model studies predicted that BCC and the HH pathway subtype of medulloblastoma would exhibit dramatic responses to SMO inhibitors in the clinic and led to the inclusion of patients with medulloblastoma in the initial clinical trials.

Clinical trials

The results of the first phase I trial of the SMO inhibitor GDC-0449 reported that 19 out of 33 patients with BCC, and one patient with medulloblastoma, exhibited either a partial or complete response to this novel therapy⁶. An unprecedented 50% response rate was observed in patients with metastatic BCC^{3, 6}. A dramatic, albeit transient, responses was also reported in an adult with metastatic medulloblastoma⁴, and encouraging results in a Phase I pediatric medulloblastoma clinical trial were reported at the 2010 American Society of Clinical Oncology meeting¹⁰⁷, indicating that the preclinical data on BCC and medulloblastoma were indicative of a response. In the one case of metastatic adult medulloblastoma, the patient subsequently relapsed because of mutation of the drug-binding site in SMO². This unfortunate circumstance provided strong affirmation of SMO as a drug target and echoed early experiences with Imatinib (Gleevec), the prototypic molecular targeted therapy¹⁰⁸. Resistance to inhibitors also arises readily in animal models as a consequence of mutations in SMO, amplification of GLI2 or amplification of CCND1 (encoding cyclin D1)^{2, 109, 110}. It is likely that drug resistance will be an important aspect of clinical treatment with HH pathway inhibitors and compounds have already been identified that can overcome resistance resulting from SMO mutations¹¹⁰. However, this approach will not be successful in resistant tumors that acquire mutations downstream of SMO. Some hope has been offered by the recent finding that blocking phosphatidylinositol-3 kinase (PI3K) activity inhibits the growth of certain resistant tumors¹⁰⁹.

In contrast to the dramatic clinical results reported in early BCC and medulloblastoma patients, no major responses to SMO inhibitors have yet been reported in other cancers⁶. Although Phase I trials are designed to test drug safety, not efficacy, the positive effects on BCC and medulloblastoma were very obvious. In addition, two trials of GDC-0449 have been closed to patient accrual (NCT00739661 and NTC00636610, listed on http://clinicaltrials.gov). In the case of advanced ovarian cancer, GDC-0449 was being used in a maintenance setting as a single agent in a phase II clinical trial, but it did not sufficiently extend the median time to recurrence. In the case of colon carcinoma, a phase II combination therapy clinical trial of GDC-0449 plus bevacizumab (Avastin) failed because treatment did not extend the time from randomization to disease progression or death. A major caveat in the interpretation of the results of the initial clinical trials is that patients were not selected based on the presence of HH pathway activity in tumor tissue (Box 2). Previous experience with protein kinase inhibitors showed that therapeutic effects are often not detected without molecular stratification of patients. It is important to learn from these experiences by including tests that determine the suitability of patient populations for treatment with HH pathway inhibitors, as well as assays for monitoring the effect of treatment on the target at all stages of the drug development process¹¹¹.

Currently, seven compounds that bind and inhibit SMO are listed for use in a range of advanced cancers in more than 40 different clinical trials (Table 1). The most widely used compound is the first-in-class Vismodegib (GDC-0449), developed by Curis and Genentech^{6, 92, 112}.

In mammals, HH pathway activity depends on the primary cilia¹¹³. HH binding to PTCH1 allows SMO to translocate to the ciliary membrane where it relieves repression of GLI1 and GLI2 by SUFU. Although cyclopamine binding to SMO promotes translocation to the primary cilia through the intraflagellar transport pathway, it does not activate GLI1 or GLI2 ^{114, 115}. In contrast, the antagonists SANT-1 and SANT-2, which bind to the same site on SMO, do not promote cilial translocation of SMO^{114, 115}. These findings imply that SMO has multiple conformations that could be exploited for drug development. Future research on the mechanisms responsible for SMO function and the downstream events leading to GLI activation will be particularly important if gain-of-function mutations in SMO occur frequently in drug-resistant tumors. At present, it is not clear which of the many SMO inhibitors currently in drug development would be least affected by acquired mutations in SMO.

Side effects

SMO inhibitors have exhibited remarkably few and only relatively modest side effects in adult patients (Table 2)⁶. However, when given to young mice, for as little as 2–4 days, HhAntag caused dramatic and permanent defects in bone growth¹¹⁶. Short-term treatment of 10-day old mice resulted in malformation of the epiphysis and growth plate. The columnar organization of chondrocytes in the growth plate was disrupted, and the cartilage structure appeared dysplastic. These observations are consistent with the critical role the hedgehog pathway is known to have in bone development¹¹⁷. Deletion of Indian Hedgehog (Ihh) causes embryonic lethality in mice¹¹⁸, conditional ablation results in a phenotype similar to that seen in mice treated with HhAntag¹¹⁹, and hypomorphic mutations of IHH in humans cause acrocapitofemoral dysplasia ¹²⁰. The several roles that HH signaling has in development, including the postnatal formation of the cerebellum^{121, 122}, raise many concerns about potential side effects that may be seen in the youngest patients (Table 2).