Basal cell carcinomas: attack of the hedgehog

BCCs and basal-cell nevus syndrome

The vast majority of BCCs occur sporadically, but there is one rare heritable disorder in which patients have a marked susceptibility to developing BCCs. This is basal-cell nevus syndrome (BCNS, also known as Gorlin syndrome or nevoid basal-cell carcinoma syndrome; see BOX 1).

Using family-based linkage studies of kindreds with BCNS, the locus carrying the causative mutant gene was mapped to human chromosome 9q22 (REF.19) and then to the patched 1 (PTCH1) gene^20–22. This finding was especially instructive for several reasons. First, there was previously only minimal insight into the molecular underpinnings of this cancer — p53 mutation in a sizable fraction of sporadic BCCs was essentially the only previously known molecular abnormality²³, and p53 mutations are common in non-cancerous skin of BCC patients as well²⁴. This lack of knowledge was due, at least in part, to the difficulty of growing human BCCs in tissue culture or as xenografts and to the lack of a satisfactory animal model. Mice treated with UV or ionizing radiation, or with chemical carcinogens, develop papillomas and carcinomas of the squamous but not basal cell lineage. rats treated with ionizing radiation do develop BCCs but they also develop other skin cancers in greater abundance. Second, the function of PTCH1 was broadly clear — its sequence identified it as the homologue of an already well-studied inhibitor of the HH signalling pathway that was known to be crucial for development in Drosophila melanogaster, and was found subsequently to occupy a similarly crucial role in mammalian development. Hence, it was straightforward to predict and demonstrate that its biallelic inactivation produced constitutive upregulation of HH signalling. Thus, PTCH1 functions as a classic tumour suppressor gene. Third, not only was HH signalling found to be upregulated in all studied sporadic BCCs, but also this upregulation is frequently accompanied by, and at least in part based on, mutations in PTCH1. Currently, it is thought that upregulation of HH signalling is the pivotal abnormality in all BCCs, and indeed there is some evidence that little more than HH upregulation is required for BCC carcinogenesis^25,26. Approximately 90% of sporadic BCCs have identifiable mutations in at least one allele of PTCH1 (often loss of the portion of chromosome 9q harbouring PTCH1), and an additional 10% have activating mutations in the downstream smoothened (SMO) protein, which presumably render SMO resistant to inhibition by PTCH1 (REFS 27–30).

Hedgehog signalling

Although a complete description of the intricacies of HH signalling is outside the focus of this review (see REFS 31–33 for reviews), some background information is necessary (FIG. 3). The HH signalling pathway is named after the family of extracellular HH ligands, of which there are three in mammals: sonic hedgehog (SHH), Indian hedgehog (IHH) and desert hedgehog (DHH). PTCH1 is the receptor to which the HH ligands bind, and such binding relieves the inhibition of the pathway induced by unbound PTCH1, specifically through SMO in a non-stoichiometric manner. Once relieved of inhibition, SMO sends signals through a series of interacting proteins, including suppressor of fused (SUFU), culminating in activation of the downstream Gli family of transcription factors, GlI1, GlI2 and GlI3, the founding member of which was identified as a gene amplified in glioblastoma³⁴. These transcription factor proteins exist in various forms, and GlI2 and GlI3 can be activators or suppressors of transcription; GlI1 seems to have only activator functions. The stability of these molecules is controlled by phosphorylation and ubiquitylation–proteolytic destruction, and processing from inactive to active suppressor or activator forms is accomplished, at least in part, by proteolysis^35–38.

The interactions of the components of the HH signalling machinery can occur at the cilium^39–42. SMO seems to be excluded from this structure when inactive, but resides within the cilium when signalling is activated⁴³. Target genes whose expression is upregulated directly by HH signalling in BCCs include PTCH1, providing a negative feedback that dampens of the pathway, GLI1, providing a positive feedback for the pathway, and HHIP, which encodes a HH binding protein^44,45. The expression of mrNAs encoding these proteins is routinely increased in BCCs.

Somatic mutations in BCCs

In general, BCCs seem to have relatively stable genomes — the few published studies suggest that they have lower levels of genomic instability than do many extracutaneous cancers⁴⁶. As noted above, BCCs routinely carry mutations in PTCH1 and TP53 and, in 10% of instances, in SMO^47,48. Mutations in several other genes encoding components of HH signalling have been sought but few have been identified, let alone confirmed⁴⁷. The mutations identified in PTCH1 and TP53 are frequently of a type that is consistent with their having been produced by UV radiation. This is true for BCCs that arise sporadically, and even more so for those large numbers of BCCs that arise in patients with xeroderma pigmentosum (XP), suggesting that repair of UV-induced DNA damage normally does reduce BCC carcinogenesis^49–51. Furthermore, this suggests that one reason for the increased incidence of BCCs in older people might be the reported reduction of DNA repair with ageing⁵².

Predisposing constitutional genetic variants

Because UV irradiation is a significant risk factor for BCC development, genes that control the extent of UV-induced DNA damage and those whose protein products effect repair of that damage are prime candidates for risk-modifying genes.

Variation in constitutive pigment and ability to tan seem to be correlated inversely with BCC development. Hence, it is not surprising that studies have been undertaken of the melanocortin 1 receptor gene (MC1R), the major known genetic variant contributing to the degree of skin pigmentation. MC1R encodes the receptor of α-melanocyte-stimulating hormone (αMSH), a proteolytic product of the larger protein encoded by the pro-opiomelanocortin gene (POMC). MC1R is highly polymorphic among normal individuals and is a prime contributor in determining whether the skin produces brown–black pigment (eumelanin) or red–yellow pigment (pheomelanin). Non-functional MC1r variants result in the production of pheomelanin and, in particular, the phenotype of red hair⁵³. Because people with red hair and fair pigment have an increased risk of skin cancers, those with variant MC1R alleles are at increased risk both of melanomas and of BCCs, and the increased risk is dose-dependent, that is, higher in carriers of two variant alleles than in carriers of a single variant allele^54–57. However, in both BCCs and SCCs of the skin, the association of variant alleles and increased risk persists even when corrected for skin pigmentation, thus suggesting a role for MC1R in susceptibility to skin cancers that is mediated by mechanisms other than control of pigmentation. Similarly, MC1R variants seem to contribute to susceptibility to melanoma by mechanisms beyond their effects on skin pigmentation^58–60. Indeed, some have suggested that αMSH can directly modulate keratinocyte proliferation and differentiation⁶¹. Furthermore, UV irradiation of keratinocytes can enhance POMC and αMSH production, which suggests a possible paracrine role for this hormone^62–64. Variation in BCC incidence is also associated with variants in the genes encoding tyrosinase, the rate-limiting enzyme in melanin formation, and agouti signalling protein, which inhibits the interaction of αMSH and MC1r. As with MC1r variants, their effects on pigmentation do not seem to explain all of their effects on BCC susceptibility⁶⁵.

Patients with XP have constitutional inactivating mutations in both alleles of certain genes that encode DNA repair proteins, in particular those involved in nucleotide excision repair, a process that is of crucial importance for removing UV-induced photoproducts from keratinocyte DNA. In patients with XP, this molecular importance is attested to clinically by the huge increase in relative risk of developing BCCs and other skin cancers and the onset of these cancers on average 50 years earlier than in those without such genetic impairment⁶⁶.

This finding raises the question of whether more common variants in genes encoding proteins involved in DNA repair might contribute to the relative risk of developing BCCs in patients without XP. Numerous studies have investigated the repair capacities of cells from patients with BCC and control patients, as well as the effects of common coding variants (predominantly single nucleotide polymorphisms) on BCC relative risk. In general, studies in patients with various cancers indicate some reduction of DNA repair⁶⁷. One group has published several studies that favour a reduced DNA-repair capacity (using the host cell reactivation assay) in lymphocytes from patients with BCC as opposed to controls⁶⁸, although their findings were not confirmed in a small independent study⁶⁹.

Similarly, a recent small study found reduced in vivo clearance of UV-induced photoproducts from DNA extracted from the skin of patients with BCC compared with DNA extracted from a control group without skin cancer⁷⁰. However, the overlap between individuals in the two groups was large, so variations in DNA repair, at least as captured by the assay used, cannot account for a large proportion of any differential susceptibility to BCC in the Finnish population studied.

Associations between DNA repair gene variants and BCC relative risk that reach (or at least closely approach) statistical significance have been reported for several DNA repair genes. Perhaps the most studied of these is the gene polymorphism that underlies the T241M substitution in XRCC3. Some studies^71–73 but not all^74,75 have found that this substitution is associated with a reduced relative risk for BCCs. Counterintuitively, this substitution is also associated with an increased risk for breast cancer⁷¹ and it does not seem to change DNA repair capacity, at least by the assay used. The usual explanation for such findings is that the identified polymorphism is not causative, but is simply located near (is in linkage disequilibrium with) the causative mutation, or it could be that the reported association results are spurious.

Studies of polymorphisms in other DNA repair genes have also suggested associations with the relative risk of developing BCCs, but other studies have failed to find the same association. Clearly, the relationships between identified associations and changes in DNA repair are confusing (BOX 2). Such uncertainties argue for the potential of unbiased, genome-wide association studies to identify polymorphisms that confer a relative susceptibility to BCC formation, and to uncover greater understanding of the mechanisms underlying BCC development.

A common variant in TP53 occurs at codon 72, and the two alleles encode either arginine or proline (Pro). Taken together, at least some of the published studies imply that the Pro allele enhances susceptibility to BCC, but perhaps only in people who are relatively more resistant to developing these tumours — those with darker skin pigmentation and a lack of variant MC1R alleles^76,77. There are similar published results for an association between this variant and susceptibility to cutaneous melanomas⁷⁸. Again, as with other studied polymorphisms, conflicting data arguing for no association have also been published⁷⁹. A further complication of the analysis is an association of the Pro allele with resistance to childhood sunburn. Thus, TP53 alleles may control not only susceptibility to skin cancer but also to sunburn, and the effects of one allele may be opposite on these two phenotypes and in counterintuitive directions. A biologically active polymorphism in MDM2, which influences p53 protein stability, was found not to be associated with BCC relative risk in one study that assessed more than 1,000 subjects⁸⁰.

One common polymorphism at exon 23 of PTCH1 encodes either Pro or leucine at codon 1315. In patients with more severe BCCs (multiple tumours, early onset), two studies have found an association with the genotype encoding Pro/Pro^55,81. However, in one study that was not restricted to patients with greater BCC severity, the individual single nucleotide polymorphism was not associated with BCC relative risk, although one haplotype that included this polymorphism was so associated. Possible associations with the many other genes encoding members of the HH pathway and/or genes encoding controls of this pathway remain to be studied. Finally, Balmain and colleagues reported the surprising finding that a Ptch1 polymorphism in the mouse is an important controlling factor in the susceptibility to mutant HRAS-induced skin tumours of the squamous lineage. They suggest that such polymorphisms may control the relative susceptibility to formation of SCCs versus BCCs⁸².

Molecular analyses are therefore consistent with the idea of genetic control of susceptibility versus resistance to BCC carcinogenesis. In particular, such analyses are consistent with the idea that resistance is associated with better protection against mutagenesis and with better repair of whatever DNA damage does occur. However, we have yet to identify convincingly the genetic underpinnings beneath the wide range of clinical outcomes in some people of northern European ancestry with pale complexions who sunbathe: some get many BCCs, some get many SCCs, some get fewer tumours, some get only precancerous carcinomas in situ, some get only wrinkles and some sustain no clinically apparent skin damage at all.

Pathway interactions and expression changes

Several groups have assessed genome-wide expression in keratinocytes in which HH signalling has been activated experimentally⁸³ and directly in BCCs. A question with the latter is the choice of control cells against which the BCC expression results are compared. Published studies have compared expression patterns of whole tumour with those of whole normal skin^84–86 or of cells at the periphery of BCC nests with those of cells of the basal layer of the interfollicular epidermis⁸⁷. Overall, perhaps as expected, many differences in expression have been found, but so far the field has not progressed sufficiently to enable firm conclusions to be drawn about which findings are reproducible across different platforms and investigators, and which changes are crucial to the aberrant behaviour of the BCC cancer cells.

Currently, the best-studied downstream mediators remain those on which investigators have chosen to focus as candidate targets. Among these are upregulated plateletderived growth factor receptor-α (PDGFRα)⁸⁸, the upregulated apoptosis inhibitors BCl2 (REFS 89,90) and CASP8 and FADD-like apoptosis regulator (CFlAR)⁹¹ and the downregulated apoptosis inducers CD95 (FAS)^92,93 and BMI1 (REFS 94,95). The ‘wiring’ downstream of HH signalling activation seems to differ in various tissues. Thus, inhibition of mitogen-activated protein kinase signalling inhibits the growth of HH pathway-stimulated BCC cells but not that of cerebellar granule precursor cells^88,96. But, in truth, we have only a limited understanding of which downstream expression changes are actually crucial for HH-induced BCC carcinogenesis. More data have been published during the past few years indicating that other signalling pathways may have profound effects on HH signalling in cancers (see below and BOX 3), although most of these studies did not address BCCs specifically.

Prevention

Although the use of sunscreens has not so far been associated with a reduction in BCCs, there is hope that their use earlier in childhood might reduce later BCC carcinogenesis. Clinical trials addressing this hypothesis will not be completed in the near future, and so counselling of sun avoidance must currently rely on ‘best guesses’. One alternative to sun-protection measures is that of artificial tanning by systemic administration of αMSH¹²⁵, and indeed tanned skin is less susceptible to UV-induced DNA damage¹²⁶. As compared with UV-induced tanning, αMSH-induced tanning would seem to have the advantages of being non-mutagenic, non-promoting and non-immunosuppressive, but its clinical practicality has yet to be demonstrated, and current development efforts involve parenteral drugs, a route of administration that is not likely to gain universal acceptance for UV protection.

Oligonucleotides that mimic the free ends of telomeres enhance DNA repair, stimulate tanning¹²⁷ and have anticancer efficacy in various cancer models, including UV-treated Ptch1^+/− mice¹²⁸. Another strategy to enhance DNA repair capacities is to apply extra copies of DNA repair enzymes. Surprisingly, such proteins applied in liposomes can penetrate skin and accumulate in the nucleus of UV-damaged cells in an apparently catalytically active form such that significant reduction in UV-induced skin changes, including experimental carcinogenesis, can be seen¹²⁹. Such topical applications over one year significantly reduced the development of new skin cancers in patients with XP¹³⁰.

Systemic retinoids have been reported to be effective against new BCC development in patients with BCNS^131,132 or XP¹³³, but have shown no protective effect in trials in patients who are at high risk of developing sporadic BCCs^134,135. Surprisingly, prolonged topical application of tazarotene can cure 25—50% of sporadic human BCCs^136–138. Topical tazarotene effectively prevents BCC carcinogenesis in the Ptch1^+/− mouse^139,140, and we (D. Bickers and E.H.E.) are now conducting a multicentre clinical trial of its efficacy in patients with BCNS. Finally, one clinical trial of dietary intervention in patients who were at high risk of developing new sporadic skin cancers found that a low-fat diet (20% of calories as fat as opposed to the usual 35—40%) was associated with a reduction in the number of new BCCs¹⁴¹.

Treatment —narrowing the hedgehog’s powers

Surgical excision of BCCs is currently by far the most commonly used treatment, and control approaches 100%. However, this high cure rate is accompanied by the inevitable discomfort and scarring of surgery, and the high frequency of BCCs makes them an expensive effort for those who foot the bill. As PDGFrα has been reported to mediate some of the downstream effects of HH signalling in BCCs, use of approved agents that inhibit this receptor kinase would be sensible⁸⁸. These include sorafenib and imatinib (Gleevec) but no study of their efficacy in BCC has been published. One systemic chemotherapy reported to have some efficacy for locally uncontrollable BCCs is the combination of paclitaxel and carboplatin¹⁴². With the evidence for HH activation in many types of visceral cancers, the rationale for development of HH inhibitors (HHIs) has become compelling. Consequently, at least half a dozen pharmaceutical companies have embarked on HHI development, and BCCs that are not controllable by local therapies are potential targets for trials of HHIs.

The first well-studied HHI is the plant alkaloid cyclopamine¹⁴³. Indeed, one intrepid group applied cyclopamine topically and reported regression of four sporadic BCCs¹⁴⁴. Cyclopamine is a competitive inhibitor of SMO signalling, binding directly to the protein^145–147, and inhibits the growth of malignant cells driven by HH activation¹⁴⁸. Infinity Pharmaceuticals in Cambridge, Massachusetts, USA is developing cyclopamine derivatives with better pharmacological and inhibitory properties as potential HHIs, and the company expects to launch a phase 1 trial of one of these in 2008.

The first HHI tested in phase 1 trials was a Curis–Genentech compound (Curis 61414), which produced neither clinical changes in the tumours nor reductions of mrNA encoding the HH target gene GLI1 when applied topically to sporadic human BCCs (see Curis website 2006 press release in Further information). However, GDC-0449, a second Curis–Genentech HHI molecule, seemed to have minimal toxicity and to cause clinically significant benefit in eight out of nine patients with metastatic or locally advanced BCCs when administered orally in a phase 1 trial¹⁴⁹. Genentech expects shortly to initiate phase 2 trials of this molecule in patients with advanced BCCs and in patients with advanced colorectal and ovarian cancers.

The practical barriers to the development and use of HHIs in localized BCCs include the high cure rate with surgery, despite the unattractive aspects of such procedures described earlier. If systemic delivery of HHI were to be used, the agent would have to be essentially 100% free of adverse extracutaneous effects. One potential on-target adverse effect might be on normal tissue stem cells such as those of the brain, the maintenance of which in mice is supported by HH^150,151. loss of these brain stem cells in mice has been reported to give cognitive defects¹⁵². Indeed, the minimally annoying dysgeusia (distortion or loss of the sensation of taste) and alopecia seen in the systemic HHI trial could be caused by on-target effects of inhibition of HH function normally required for maintenance of the olfactory bulb, tongue papillae and hair follicle^{150,153–155}. Such therapy-induced loss occurring in humans with pancreatic cancer might be tolerable; not so for patients with the usual BCCs. HHIs also can cause premature closure of epiphyses¹⁵⁶, a reminder that development is not complete at birth and a potential contraindication to HHI treatment of childhood medulloblastomas.

Why are BCCs so much more common than other human cancers?

The answer to this question traditionally has been that they arise in an organ that is subject to enormous mutagenic insults exogenously and endogenously as a ‘cost’ of the continued lifelong high turnover of the epidermis. Viewed in molecular terms, it might also be the case that, unlike in the development of many visceral cancers, disruption of merely one anticancer mechanism — restraint of HH signalling — seems to be enough to allow their growth. However, viewed in another context (prostate cancer in the elderly, for example), perhaps their frequency is not so high after all. Moreover, the apparently high incidence might be, in part, a product of the fact that these lesions are obvious to the naked eye. Indeed, as we have been able to examine internal tissues more closely, the incidence of ‘incidentalomas’ rises dramatically, with the accompanying quandary of what is the proper medical intervention. If the diagnosis of BCCs were to depend on organ dysfunction rather than appearance, their incidence might plummet to less than that of many classical visceral cancers. In this context, the skin may serve as a useful model for how medicine will learn to deal with the increasing numbers of ‘incidentalomas’ that our improving diagnostic acumen uncovers.

Why does the incidence of BCCs not rise proportionally to the amount of UV exposure as in SCCs?

One possibility for this comes from the observation that sun exposure activates DNA repair mechanisms¹⁵⁷: perhaps mutations of PTCH1 are more susceptible to repair than mutations in the genes that underlie SCCs. However, no evidence for such differential repair is available. One alternative is suggested by the observation that vitamin D inhibits HH signalling through binding to SMO protein¹⁵⁸. Thus, more frequent UV exposure might maintain a level of keratinocyte vitamin D that is sufficient to inhibit the growth of BCCs but not of SCCs. A potential flaw in this is that one study suggested that chronic sun exposure fails to increase cutaneous vitamin D levels even while increasing internal levels¹⁵⁹. Consistent with this idea, however, is the finding that skin production of 7-dehydrocholesterol, the precursor molecule that UV radiation converts to vitamin D, wanes in the elderly, and this parallels the increasing incidence of BCCs with ageing¹⁶⁰. More generally, the increasingly studied and discussed anticancer effects of vitamin D stores might also have some anti-BCC carcinogenesis effect. Another possibility might be that the photoimmunosuppression that occurs with sun exposure affects SCCs more than BCCs; indeed, in immunosuppressed organ transplant patients, the incidence of SCCs increases considerably more than the incidence of BCCs. However, at least in Ptch1^+/− mice, anti-rejection drugs can allow more robust BCC carcinogenesis¹⁶¹, and so the immune system ordinarily may provide at least some protection against BCC as well as SCC carcinogenesis. Better understanding might come were we able to identify clinically inapparent precursor lesions.

Why do BCCs so rarely metastasize?

Not only do BCCs essentially never spread to distant regions, but also the limitation of their growth to that by local extension allows careful, microscopically controlled surgical excision to give a cure rate approaching 100%. One possibility might be that physicians remove BCCs before they grow to a size that allows the accumulation of the additional mutations needed to metastasize. Yet patients with BCNS may have hundreds of BCCs; commonly, BCCs in these patients are removed only when they impinge on sensitive structures such as the eye, and most are treated with a ‘wait and see’ approach. Another possibility would be that activated HH signalling is incompatible with metastasis. But prostate cancer metastases have higher HH signalling activity than do the primaries from which they arose¹⁶², suggesting that this might not be the answer. A third possibility might be that specialized abnormal stroma is required, and indeed there is some evidence that BCCs are especially dependent on stroma, at least for their experimental transplantation to other sites in humans¹⁶³. In addition, loss of specialized stroma might be one reason for the marked downregulation of HH signalling when experimental BCCs or medulloblastomas are transferred from the host to tissue culture. But in most patients with BCCs, there would seem to be no lack of UV-damaged stroma available as a ‘soil’ for BCCs to find if this were the main cause of their lack of metastasis. Finally, BCCs tend to have relative genomic stability, and perhaps it is this that provides the barrier to further DNA abnormalities that might confer metastatic potential. But why do BCCs not acquire genomic instability? Might such instability be incompatible with their successful local growth?

What can we learn about BCCs that may further inform our knowledge of visceral cancers?

If hedgehog activation at first induces keratinocyte senescence, as has been suggested in experimental models of HH signalling activation¹⁶⁴, why does senescence not occur with physiological HH signalling during development, hair follicle cycling and experimental models with conditional Gli overexpression?²⁵ A greater understanding of the HH pathway and the induction of senescence is needed to address this question. It is possible that the degree of deregulation of HH signalling found in human cancer is enough to overcome induction of senescence. Indeed, it will be interesting to determine how much HH signalling is required to drive cancer development in different tissues. Analysis of tissues from the recent phase 1 clinical trials of HHIs in patients with BCC will also provide important data on whether a complete absence of HH signalling is required for a clinical response or whether a reduction is all that is needed. later stage trials of HHIs in BCC will also be able to address whether relapse or resistance occur in patients with advanced BCC. Finding mechanisms of resistance, should they occur, might help the development of HHIs for other cancers such as medulloblastoma, in which HH signalling is deregulated. As BCCs seem to have relatively little genomic instability, it is reasonable to hope that mutations may not be as frequent a mechanism of resistance as is the case with more genetically unstable cancers. Finally, it will be important to establish the signalling pathways downstream of the Gli transcription factors that drive carcinogenesis. These might provide opportunities for development of drugs with perhaps a better therapeutic index (for example, sparing stem cells that, at least theoretically, might be the target of collateral damage with HH inhibition) than that of direct inhibition of HH signalling.