Cooperative integration between HEDGEHOG-GLI signalling and other oncogenic pathways: implications for cancer therapy

The HEDGEHOG-GLI (HH-GLI) signalling is a key pathway critical in embryonic development, stem cell biology and tissue homeostasis. In recent years, aberrant activation of HH-GLI signalling has been linked to several types of cancer, including those of the skin, brain, lungs, prostate, gastrointestinal tract and blood. HH-GLI signalling is initiated by binding of HH ligands to the transmembrane receptor PATCHED and is mediated by transcriptional effectors that belong to the GLI family, whose activity is finely tuned by a number of molecular interactions and post-translation modifications. Several reports suggest that the activity of the GLI proteins is regulated by several proliferative and oncogenic inputs, in addition or independent of upstream HH signalling. The identification of this complex crosstalk and the understanding of how the major oncogenic signalling pathways interact in cancer is a crucial step towards the establishment of efficient targeted combinatorial treatments. Here we review recent findings on the cooperative integration of HH-GLI signalling with the major oncogenic inputs and we discuss how these cues modulate the activity of the GLI proteins in cancer. We then summarise the latest advances on SMO and GLI inhibitors and alternative approaches to attenuate HH signalling through rational combinatorial therapies.


Introduction
The HEDGEHOG-GLI (HH-GLI) signalling plays a critical role in embryonic development, stem cell biology and tissue homeostasis, cellular metabolism, synapse formation and nociception ( Refs 1,2,3,4,5). Aberrant activation of the HH signalling has been linked to different aspects of cancer development, from initiation to metastasis 1 Core Research Laboratory, Istituto Toscano Tumori, Florence, Italy 2 Department of Oncology, Azienda Ospedaliero-Universitaria Careggi, Florence, Italy *Corresponding author: Barbara Stecca, Laboratory of Tumor Cell Biology, Core Research Laboratory, Istituto Toscano Tumori (CRL-ITT), Viale Pieraccini 6, 50139 Florence, Italy. E-mail: barbara.stecca@ ittumori.it (Ref.6).CanonicalHHpathwayactivationisinitiated by the binding of HH ligands to the transmembrane receptor PATCHED (PTCH), which relieves its inhibition on the transmembrane protein SMOOTHENED (SMO). Consequently, active SMO triggers an intracellular signalling cascade leading to the formation of activator forms of the GLI zinc finger transcription factors GLI2 and GLI3, which directly induce GLI1. Both GLI1 and GLI2 act as main mediators of HH signalling in cancer by controlling the expression of target genes.
Recent evidence suggests that GLI proteins can be directly and indirectly modulated by proliferative and oncogenic inputs, in addition or independent of upstream HH signalling. These mechanisms of aberrant, non-canonical HH-GLI pathway activation, apparently without known driver mutations in components of the pathway, have been associated with several types of human cancer (Ref. 7). In this review, we focus on the cooperative interaction between HH-GLI and other oncogenic signalling pathways. We first address the functions and post-translational modifications of the three GLI transcription factors, and the mechanisms that regulate their activity in cancer. We then review latest advances on SMO and GLI inhibitors and discuss approaches to attenuate HH signalling through rational combinatorial therapies.

Overview of the HEDGEHOG-GLI signalling
The Hh signalling has been initially identified in Drosophila melanogaster, where it is required for determining proper embryonic patterning (Ref. 8). Smo protein is conserved and maintains its function in mammals, whereas there are two Ptch proteins in vertebrates. Hh ligand has diversified into Sonic (SHH), Indian (IHH) and Desert (DHH) Hedgehog, and the function of the downstream transcription factor Cubitus interruptus (Ci) has evolved into three GLI proteins: GLI1, GLI2 and GLI3. Here we focus on the function and regulation of the three GLI transcription factors and we present only a brief introduction of the key steps and components of vertebrate HH-GLI signalling upstream of GLI. The initiation of the HH signalling begins with the binding of one of the three HH ligands, each with distinct spatial and temporal expression patterns, to the 12-pass transmembrane protein receptor PTCH, which resides in the primary cilium, a non-motile structure that functions as a sensor and coordinator centre for the HH signalling ( Refs 9,10,11). Binding of HH ligands to PTCH relieves its inhibitory effect on the G-protein-coupled receptor-like SMO, which moves into the tip of the cilium and triggers a cascade of events that promote the formation of GLI activator forms (GLI-A). GLI2/3-A translocate into the nucleus and induce HH pathway target genes, including GLI1 ( Refs 12,13,14) (Fig. 1). In absence of HH ligands, PTCH inhibits pathway activation by preventing SMO to enter the cilium. This results in the phosphorylation and proteasome-mediated carboxyl cleavage of GLI3 and, to a lesser extent, of GLI2 to their repressor forms (GLI2/3-R; Refs 15,16). GLI1 is degraded by the proteasome and is transcriptionally repressed, with consequent silencing of the pathway. GLI1 acts exclusively as an activator, whereas GLI2 and GLI3 display both positive and negative transcriptional functions (Refs 15, 17, 18) (Fig. 1).
The HH target genes include GLI1, which further amplifies the initial HH signalling at transcriptional level and, therefore, is a reliable and robust read-out of an active pathway (Ref. 19). Other HH target genes are PTCH1 and HH interacting protein (HHIP1), which both mediate negative feedback by limiting the extent of HH signalling. The outcome of the HH signalling varies according to the receiving cell type, and includes a number of cell-specific targets mediating a variety of cellular responses: proliferation and differentiation (Cyclin D1 and D2, E2F1, N-Myc, FOXM1, PDGFRα, IGFBP3 and IGFBP6, Hes1, Neogenin), cell survival (BCL-2), self-renewal (Bmi1, Nanog, Sox2), angiogenesis (Vegf, Cyr61), cardiomyogenesis (MEF2C), epithelial-mesenchymal transition (Snail1, Sip1, Elk1 and Msx2) and invasiveness

The GLI transcription factors and their modifications
The three GLI transcription factors are members of the Kruppel family. They share five conserved C 2 -H 2 zinc-finger DNA-binding domains and a consensus histidine/cysteine linker sequence between the zinc fingers, and bind to the consensus motif GACCACCCA in the promoter of their target genes (Ref. 21). The sequence specificity of the GLI transcription factors, although, is not absolute, because they can recognise variant GLI-binding sites with relatively low affinity, still leading to strong transcriptional transactivation (Ref. 22).
GLI1 is a transcriptional target of GLI2 and GLI3 (Refs 17, 23) and a strong transcriptional activator. In both human and mouse cells, GLI1 protein is translated from alternative mRNAs that differ in their 5 ′ untranslated region and that are generated by exon skipping. The shorter mRNA shows the highest translation efficiency and it is the predominant transcript in proliferating cells and in basal cell carcinoma (BCC; Ref. 24). GLI1 mRNA also undergoes adenosine deamination acting on RNA (ADAR)-dependent A-to-I editing Key components of the mammalian HH signaling pathway Figure 1. Key components of the mammalian HH signalling pathway. In absence of HH ligands (a), PTCH inhibits SMO by preventing its entry into the primary cilium. GLI proteins are phosphorylated by PKA, GSK3β and CK1, which create binding sites for the E3 ubiquitin ligase β-TrCP. GLI3 and, to a lesser extent, GLI2 undergo partial proteasome degradation, leading to the formation of repressor forms (GLI3/2 R , red), that translocate into the nucleus where they inhibit the transcription of HH target genes. Full-length GLI may also be completely degraded by the proteasome. This process can be mediated by Spop  constitutive HH pathway activation. Defining the molecular mechanisms of ligand-independent activation of the signalling is crucial to determine whether a tumour might respond to the treatment with a HH inhibitor acting at the level of SMO or, in case the genetic alteration affects downstream components of the pathway, at the level of the GLI proteins.
Ligand-dependent autocrine/juxtacrine activation of the pathway (Type II) has been identified in the last fewyearsindifferenttypesofcancers,includinglung, pancreas, gastrointestinal tract, prostate and colon cancers, glioma and melanoma (Refs 62,67,68,69,70,71,72,73,74,75,76,77,78). In this case, tumours show increased HH ligand expression, in absence of genetic aberrations of HH pathway components, and respond to HH stimulation in cell-autonomous manner. This concept is supported by a number of experimental data showing that: (i) tumour cells, but not the surrounding stroma, express HH ligands and downstream HH signalling components (e.g. PTCH1, GLI1) (e.g. Refs 67, 74, 78); (ii) tumour cell growth could be inhibited by RNAi-mediated knockdown of SMO or GLI1 and by treatment with cyclopamine (a SMO antagonist) in vitro and in xenograft models in vivo; (iii) metastatic growth could be prevented in vivo upon RNAi-mediated knockdown of SMO or GLI1 (Ref. 67). These effects appear to be specific, because GLI1 epistatically rescues the inhibition of metastatic colonies obtained with SMO silencing (Ref. 67).
Ligand-dependent paracrine activation of HH pathway (Type IIIa) is a mode of action that resembles the physiological HH signalling occurring during embryo development. In this case, HH ligands secreted by cancer cells activate HH signalling in the surrounding stroma rather than in the tumour itself. The mechanisms by which the HH signalling pathway and the tumour stroma interact during paracrine signalling are not completely understood. However, activation of HH signalling in the tumour-associated stroma might lead to the production of growth factors (e.g. VEGF, IGF) and stimulation of other signalling pathways (e.g. Wnt, Interleukin-6) that in turn create a favourable microenvironment sustaining the growth and progression of the tumour (Ref. 79). Evidence supporting this mechanism has accumulated from studies in human tumour xenograft models of pancreatic and colorectal cancers that express high levels of HH ligands, in which increased expression of HH targets is detected specifically in tumour-infiltrating mouse stromal cells (Ref. 79). Interestingly, growth of mutant Kras-driven tumours is reduced in mice lacking Gli1 in the pancreatic microenvironment compared to wild-type mice (Ref. 80).
Similarly, the reverse paracrine HH pathway activation (Type IIIb) has been described in an experimental model of glioma (Ref. 81) and in haematological malignancies such as B-cell lymphoma and mantle cell lymphoma (MCL; Refs 82,83). According to this modality, HH ligands are secreted by the tumour microenvironment (bone marrow stromal cells or endothelial cells) and activate the pathway on tumour cells, thus affecting its growth.

HH-GLI signalling in cancer stem cells (CSCs)
Multiple lines of evidence indicate that HH-GLI pathway plays a role in the maintenance and regulation of CSCs in several types of cancer. Selfrenewal, survival and tumourigenicity of CD133 + glioblastoma CSCs require SMO and GLI1 activity, as shown by their inhibition with cyclopamine and RNA interference ( Refs 75,84 The critical tumourigenic role of HH pathway is further highlighted by its activity in CSCs, through the subverted regulation of stemness genes, such as NANOG and SOX2, which are overexpressed in certain cancer types. More specifically, the HH pathway has been shown to directly regulate NANOG transcription through GLI1 and GLI2 in neural stem cells (Ref. 94). In line with these findings, NANOG has been shown to act as a mediator of the HH-GLI signalling in regulating in vivo growth of glioblastoma CSCs (Ref. 95). Similarly, HH-GLI signalling regulates the expression of SOX2 in neural stem cells and medulloblastoma (Refs 96,97). Recently, we showed that both GLI1 and GLI2 bind to SOX2 promoter in melanoma cells and that SOX2 function is required for HH-induced self-renewal of melanoma CSCs (Ref. 98). Altogether, these findings suggest that aberrant HH signalling induces a number of stemness factors, that might play a critical role in the acquisition of a more undifferentiated and aggressive state through a process similar to reprogramming.

Activation of HH-GLI signalling in human cancers
The initial link between HH signalling and cancer came from the finding that loss of function mutations in PTCH1 gene are associated with a rare and hereditary form of BCC, basal cell nevus syndrome (BCNS) (also known as Gorlin syndrome) (Refs 59, 60, 99). BCNS is an autosomal dominant disorder with two distinct sets of phenotypes; increased risk of developing cancers such as BCC, medulloblastoma, rhabdomyosarcoma and meningioma, as well as developmental defects, including bifid ribs and ectopic calcifications (Ref. 58), that reflect the involvement of HH pathway in many developmental processes. Consistent with the risk for specific cancers in Gorlin syndrome, sporadic BCCs and at least a subset of medulloblastomas (MBs), are the tumour types that show the strongest association with aberrant HH pathway activation, both in humans and in experimental mouse models. Activation of HH pathway in BCC and MB occurs through direct genetic alterations of HH pathway genes. Sporadic BCC and MB, a malignant brain tumour in children, harbour high frequency of inactivating mutations in PTCH1 ( This observation is in agreement with the clinical course of BCC in BCNS patients, where BCCs occur preferentially on sun-exposed areas of the body (Ref. 110). BCNS patients are predisposed to BCC, MB and rhabdomyosarcoma, but they are not at increased risk to develop other cancer types, such as glioma, breast or prostate cancers. Genetic mouse models and identification of genetic mutations in BCC and MB have suggested that aberrant activation of HH signalling is required and sufficient for the development of these cancers. In other types of cancer activation of HH signalling might require additional alterations/ mutations in other signalling pathways to contribute to tumour development.
Glioma is the most frequent tumour of the central nervous system and can be classified into four grades, with glioblastoma multiforme (GBM) being the most aggressive. GLI1 was originally identified as a gene amplified in malignant GBM (Ref. 65), although its amplification is detected in a small fraction of gliomas (Refs 111, 112). The landscape of driver genomic alterations in glioblastoma has been recently revealed, suggesting that ligand-independent activation of the HH pathway is not frequent (Ref. 113).
Nevertheless, several reports support a role for HH signalling in gliomas. For instance, expression of components of HH signalling is observed in gliomas of different grades, with SHH expression mostly confined to the surrounding endothelial cells and astrocytes. Activation of the pathway sustains growth, survival and stemness of glioma cells and progenitors ( Refs 75,77,84). Consistently, inhibition of HH signalling by cyclopamine treatment or by overexpression of miR-326, which targets SMO, decreases glioma growth, stemness and tumourigenicity ( Refs 75,84,114).
There are strong indications that the HH pathway is involved also in human breast cancer (BC), the leading cause of cancer death among women. High expression of components of HH pathway, including GLI1, is associated with a higher risk of recurrence after surgery and poorer prognosis ( However, recent data obtained in murine models of PDAC propose a controversial role for HH signalling in PDAC. In fact, activation of HH signalling has been shown to induce stromal hyperplasia and reduce epithelial growth, thus restraining tumour. Conversely, HH pathway inhibition accelerates tumour progression because, although reducing desmoplasia, it promotes proliferation and vascularisation of the tumoural epithelium, which exhibits a more undifferentiated phenotype (Refs 140, 141).
HH signalling is involved in prostate cancer (PC). Aberrant activation of HH signalling in PC might result from loss of SUFU or by ligand-dependent activation of the pathway due to high expression of SHH (Ref. 62). However, it is not clear whether HH activation occurs in a paracrine and/or autocrine/juxtacrine manner. Evidence suggests that the PC cells secrete HH ligands that activate the pathway in the surrounding stromal cells, which in turn produce factors promoting cancer cells proliferation (Refs 142, 143, 144). Conversely, other reports indicate the presence of a cellautonomous activation of HH signalling in PC cells, whose proliferation is greatly decreased by cyclopamine treatment. The expression of HH ligands and of target genes in the tumour epithelium is higher than in the normal adjacent tissue and correlates with Gleason score, metastasis and poor prognosis (Refs 62, 73,74,145). The HH effector GLI2 is highly expressed in PC where it enhances proliferation, cell survival and tumourigenicity (Refs 146, 147). Multiple evidence suggests an interplay between HH and androgen signalling.
Long-term androgen deprivation in PC leads to a strong up-regulation of HH signalling, which is also observed in androgen-independent (AI) PC cells (Refs 145, 148, 149). Overexpression of GLI1 and GLI2 enhances androgen-specific gene expression, indicating that HH signalling supports androgen signalling even in absence of androgen and in AI prostate cancer cells (Ref. 150).
HH pathway plays a role also in the most lethal form of skin cancer, malignant melanoma. A recent global genomic screening of 100 melanomas revealed few missense mutations in the core genes of the HH pathway (PTCH1, SMO, SUFU, GLI1, GLI2 and GLI3) (Ref. 151), although their potential oncogenic function remains to be determined. Several studies report an active role for HH signalling in melanoma. Human melanomas express components of HH pathway (Ref. 78) and about half of melanoma cell lines express high levels of SMO, GLI2 and PTCH1 and low levels of the negative regulators PKA and DYRK2 compared to melanocytes (Ref. 152). Interestingly, high HH pathway activity is associated with decreased post-recurrence survival in metastatic melanoma patients (Ref. 152). Moreover, we previously showed that growth and metastasis of human melanomas xenografts in nude mice can be blocked by local or systemic treatment with cyclopamine. Cyclopamine treatment drastically reduces tumour growth also in melanomas induced by oncogenic NRAS in a Tyrosinase-NRAS Q61K ; 78). Two recent studies confirmed and extended these findings; the SMO antagonist sonidegib has shown to reduce proliferation of human melanoma cell lines and to decrease human melanoma xenograft growth in nude mice (Refs 152, 153). Interestingly, one of the two studies showed a stronger inhibition of proliferation in BRAF mutant cell lines than in BRAF wild-type cells and a modest but significant effect combining BRAF and Hedgehog inhibitors (Ref. 152), suggesting that a combined therapy targeting both mutant BRAF and HH pathway could be beneficial in patients with mutated BRAF and activated HH signalling. Activation of HH pathway might also play a role in melanoma progression, by contributing to the acquisition of an invasive behaviour. Melanoma cells with high GLI2 expression are characterised by an invasive and metastatic phenotype, associated with loss of E-cadherin and secretion of metalloproteases, and metastasise to bone more quickly than cells with low GLI2 expression The activation of HH signalling in these diseases likely results from the integration of deregulated oncogenic inputs that contribute to the direct activation of the GLI proteins. Different haematological malignancies also show different modalities of HH signalling activation, which has been proposed to be paracrine mainly in CLL and plasma cell myeloma, both paracrine and autocrine in DLBCL and autocrine in ALL, AML and ALK+ALCL.

Modulation of HH-GLI signalling by oncogenic pathways
The activity of HH-GLI signalling observed in human cancer is the result of its functional interaction with other pathways and of the direct or indirect regulation of the final effectors of the HH signalling by oncogenes and tumour suppressors (Fig. 2) (Fig. 2b). Likewise, transforming growth factor β (TGF-β) stimulation leads to a SMAD3-dependent induction of GLI2, which in turn increases GLI1 expression (Ref. 208). TGF-β also induces Kindlin-2, which increases GLI1 protein levels by inhibiting GSK3β. GLI1, in turn, represses Kindlin-2 creating a regulatory loop (Ref. 209) (Fig. 2b). Activation of HH pathway in some tumours results from the increase of HH ligands. For instance, ERα pathway in gastric cancer (Ref. 210) or NF-kB in pancreatic cancer cells (Refs 211, 212) directly increase SHH expression, leading to enhanced proliferation and resistance to apoptosis. Direct induction of SHH is also mediated by p63β, p63γ and TAp73β, which bind to SHH promoter (Ref. 124) (Fig. 2b).
Although HH signalling activation is regulated by many phosphorylation events, only few phosphatases have been described to modulate the pathway. In Drosophila PP4 and PP2A act as negative and positive modulators of HH signalling, acting at the level of Smo and Ci, respectively (Ref. 213). Recently, the oncogenic wild-type p53-induced phosphatase 1 (WIP1) has been described to cooperate with SHH to enhance tumour formation in SHH-dependent medulloblastoma (Ref. 214). Our group showed that WIP1 phosphatase activity enhances GLI1 function in melanoma by increasing GLI1 nuclear localisation, protein stability and transcriptional activity, whereas its inhibition reduces self-renewal and tumourigenicity of melanoma cells with activated HH signalling (Ref. 215).
A negative reciprocal regulation is observed between GLI1 and the tumour suppressor p53. p53 inhibits the activity, nuclear localisation and protein levels of GLI1 in neural stem cells and glioblastoma cells (Ref. 216 (Fig. 2b). In BC, high levels of liver kinase B1 (LKB1) are associated with low levels of HH signalling activation (Ref. 219). Another suppressor of HH signalling is REN KCTD11 , which is often deleted in medulloblastoma and it has been shown to retain GLI1 in the cytoplasm, reducing its transcriptional activity (Ref. 63).
Regulation of HH signalling occurs also at epigenetic level. Menin, the gene mutated in multiple endocrine neoplasia type 1, recruits the protein arginine methyltransferase 5 (PRMT5) to growth arrest-specific 1 (Gas1) promoter. The consequent Gas1 repression prevents the binding of Shh to Ptch1, thus resulting in reduced HH pathway activity (Ref. 220). Different components of HH signalling are also targets of micro-RNAs (miR). miR-125b and miR-326, which target SMO, and miR-324-5p, which targets both SMO and GLI1, are downregulated in HH-driven MB and contribute to sustain tumour growth (Ref. 221). In glioblastoma the miR-302-367 cluster inhibits clonogenicity and stemness of glioblastoma stem cells, through downregulation of CXCR4/SDF1 and consequent reduction of SHH, GLI1 and NANOG levels (Ref. 222).

Inhibitors of HH-GLI signalling
Current HH pathway antagonists can be classified according to what level of the pathway they modulate: (i) HH/PTCH interaction; (ii) SMO translocation and activation; (iii) GLI nuclear translocation and transcriptional activation (Fig. 3).   The use of SMO inhibitors has been associated with the acquisition of resistance to SMO inhibitors, mostly described in medulloblastoma, as a consequence of (i) mutations in human SMO In studies investigating systemic treatments with SMO inhibitors, a common set of adverse effects has been observed, including muscle spasms, loss of taste (dysgeusia), hair loss (alopecia), fatigue, nausea, diarrhoea, decreased appetite, weight loss and hyponatraemia (summarised in Table 1). It is likely that hair loss, altered taste and diarrhoea are directly related to the inhibition of the intended molecular target (SMO), since HH signalling is known to be active in hair follicle, taste buds and gastrointestinal tract ( Refs 256,257,258). Therefore, these effects are unlikely to be avoided by modifying the molecular structure of the agents. Possible strategies to lessen these effects would be to perform interval dosing of single agent or lower doses in combination with other agents (see later). Although most of the side effects of SMO inhibitors are mild to moderate (grade 1/2, Table 1), in some cases their severity has caused 50% of dropouts (Ref. 244) and raised concerns about long-term treatment in patients with BCC, typically a nonlife-threatening cancer. One way to avoid or reduce such effects in BCC might be to use these inhibitors topically, limiting systemic exposure. A study employing topical treatment of LDE-225 for 4 weeks documented an effective reduction in tumour size or clinical clearing that correlated with effective inhibition of HH signalling (Ref. 259).

Acting at the level of HH/PTCH interaction
Interference with the interaction between HH ligands and PTCH has been shown to attenuate HH signalling in experimental models. The monoclonal antibody 5E1 blocks the binding of HH ligands to PTCH1 with low nanomolar potency (Ref. 260). This antibody has been widely used in experimental studies to demonstrate HH dependency in tumour models, but it has not advanced to clinical settings. Recently, a novel neutralising antibody acting on SHH and IHH with low picomolar affinity has been reported (Ref. 261). Moreover, two small molecules have been described; robotnikinin binds to and inhibits SHH protein (Ref. 262), whereas RU-SKI, an inhibitor of HH acyltransferase, hampers SHH palmitoylation and blocks HH signalling (Ref. 263).

Acting at the level of GLI
The development of molecules able to target directly the GLI, the final effectors of the HH signalling, would provide a good approach to block both canonical and non-canonical HH pathway activation and perhaps overcome anti-SMO drug resistance. Unfortunately, so far only few molecules acting on GLI proteins have been identified and their use is only limited to preclinical studies. A cell-based screening for inhibitors of GLI1-mediated transcription identified two structurally different compounds, GANT61 and GANT58. Both are capable of interfering with GLI1 and GLI2-mediated transcription and inhibit tumour cell growth in a GLI-dependent manner (Ref. 264). A screening of natural products identified physalins F and B as inhibitors of GLI-mediated transcriptional activity (Ref. 265). More recently, HPI-1/4 were described to act at or downstream of SUFU through various mechanisms, such as interfering with GLI processing or GLI activation. In particular, HPI-1 and HPI-4 have been shown to increase the proteolytic cleavage of Gli2 to its repressor form, whereas HPI-4 also decreases Gli1 stability (Ref. 266).
Arsenic trioxide (ATO), an already approved therapeutic for acute promyelocytic leukaemia, inhibits the GLI transcription factors ( Refs 267,268). Mechanistically, ATO directly binds to GLI1 protein and inhibits its transcriptional activity (Ref. 268 Pyrvinium, an FDA-approved anti-pinworm agent, has recently been shown to inhibit Gli activity and enhance Gli degradation in a CK1αdependent manner (Ref. 269). Consistent with its activity on the downstream mediators of the HH signalling, pyrvinium is able to inhibit the activity of a vismodegib-resistant SMO mutant (D473H) and Gli activity resulting from loss of Sufu, as well as to reduce in vivo growth of Ptch +/− MB allografts (Ref. 269).
Recently, the structural requirements of Gli1 for binding to DNA where clarified and a small molecule (Glabrescione B) that binds Gli1 zinc finger and interferes with its interaction with DNA was identified (Ref. 270). Glabrescione B is an isoflavone naturally present in the seeds of Derris glabrescens. Remarkably, as consequence of its strong inhibition of Gli1 activity, Glabrescione B inhibits growth of Hh-dependent BCC and MB tumour cells in vitro and in vivo as well as selfrenewal ability and clonogenicity of CSCs (Ref. 270).
Inhibition of BET bromodomain proteins has recently emerged as a novel strategy to target epigenetically the Hh pathway transcriptional output (Ref. 271). The BET bromodomain protein BRD4 is a critical regulator of GLI1 and GLI2 transcription through direct occupancy of their promoter. Interestingly, occupancy of GLI1 and GLI2 promoters by BRD4 and transcriptional activation at cancer-specific GLI promoterbinding sites are markedly inhibited by the BET inhibitor JQ1. In Ptch-deficient MB and BCC mouse models and patient-derived tumours with constitutive HH pathway activation, JQ1 decreases tumour cell proliferation and viability in vitro and in vivo, even in presence of genetic alterations conferring resistance to SMO inhibition (Ref. 271). These findings suggest that BET inhibition could be effective against tumour cells that evade SMO antagonists through mutation of SMO or amplification of GLI2 and MYCN, although the potential toxicities of BET inhibitors remain to be elucidated.

Acting on other proteins/pathways that modulate HH signalling
Other compounds might inhibit HH signalling by targeting proteins and/or pathways that modulate GLI transcription factors. For instance, forskolin inhibits HH signalling by activating PKA, which in turn is involved in the phosphorylation of GLI2/ GLI3, leading to their proteolytic processing into C-terminally truncated repressor forms (Ref.

Evidence for rational combinations Combination of SMO inhibitors and other agents in preclinical studies
Support for combinatorial strategies is derived from the increasing amount of experimental data showing evidence of non-canonical HH signalling activation in tumours (summarised in Table 2). Combined inhibition of HH and MEK or AKT has been shown to yield additive/synergistic effects in reducing melanoma and cholangiocarcinoma cell proliferation in vitro ( Refs 78,275). Combination of EGFR and SMO inhibitors has been described in several preclinical models. In pancreatic cancer cells, treatment with cyclopamine and EGFR inhibitor gefinitib decreased tumour growth rate and  Recently, cyclopamine has been shown to act synergistically with WIP1 inhibitor CCT007093 in reducing in vitro growth of patient-derived melanoma cells and BC cell lines (Ref. 215). These data suggest a possible novel therapeutic approach for tumours expressing high levels of WIP1 and with activated HH pathway, such as a subset of MB, gliomas and melanomas ( Refs 215,293,294,295). Targeting WIP1 in tumours with wild type p53 would lead not only to restoration of p53 tumour suppressor activity (Ref. 296), which in turn might inhibit GLI1 (Ref. 216), but also to a direct attenuation of GLI1 function (Ref. 215), resulting in a stronger inhibition of the HH pathway. This is particularly relevant to melanoma, as nearly 90% of human melanomas express functionally defective wild-type p53 and restoration of p53 function has recently been suggested as an alternative for melanoma therapy (Ref. 297). Moreover, this approach based on WIP1-p53-GLI1 axis might inhibit not only the growth of tumour bulk, but also that of putative CSCs (Ref. 215).

Clinical trials of SMO inhibitors in combination with other targets
Based on the crosstalk between HH signalling and other pathways, several combinations with SMO inhibitors are being evaluated in clinical trials (  (Table 3). Interestingly, the placebo group had a slightly better overall response than vismodegib-treated group (51% versus 46%), probably reflecting differences in safety and tolerability, as vismodegib-chemotherapy combination is less well tolerated compared with placebo-chemotherapy combination. In a pilot study, 25 patients with metastatic pancreatic adenocarcinoma were treated with a combination of vismodegib and gemcitabine. Vismodegib treatment for 3 weeks led to downregulation of GLI1 and PTCH1 in posttreatment biopsies in the majority of patients, without significant changes in the CSC compartment compared with baseline. However, vismodegib and gemcitabine were not better than gemcitabine alone in the treatment of metastatic pancreatic cancer (Ref. 299).
Vismodegib is also being tested in combination with the mTOR inhibitor sirolimus, and in combination with the gonadotropin-releasing hormone agonist leuprolide or goserelin in metastatic pancreatic cancer and locally advanced prostate cancer, respectively (Table 3). In addition, clinical studies combining vismodegib with the Notch pathway inhibitor RO4929097 in advanced BC and sarcoma are ongoing. Multiple combination studies with sonidegib and BMS-833923 are either recruiting or ongoing. For instance, phase 1 studies of sonidegib in combination with PI3K inhibitor buparlisib in several types of advanced solid tumours, or in combination with BCR-ABL inhibitor nilotinib in patients with chronic myeloid leukaemia are recruiting (see Table 3 for details). Results from these clinical trials will address the applicability of SMO inhibitors in combination with other targets in multiple cancer types.

Perspectives
Over the last decade, knowledge of the HH-GLI signalling has greatly increased, enabling a better understanding of the interaction of the major oncogenic pathways during tumourigenesis. Despite these advances, our understanding of this signalling pathway is far from complete and many important questions remain to be answered. For example, which are the mechanisms of gene