|Home | About | Journals | Submit | Contact Us | Français|
Individuals with germline mutations in the tumour suppressor gene CYLD are at high risk of developing disfiguring cutaneous appendageal tumours, the defining tumour being the highly organised cylindroma. Here, we analysed CYLD mutant tumour genomes by array comparative genomic hybridisation (aCGH) and gene expression microarray analysis. CYLD mutant tumours were characterised by an absence of copy number aberrations apart from loss-of-heterozygosity at chromosome 16q, the genomic location of the CYLD gene. Gene expression profiling of CYLD mutant tumours revealed dysregulated tropomyosin kinase (TRK) signalling with overexpression of TRKB and TRKC in tumours when compared to perilesional skin. Immunohistochemical analysis of a tumour microarray demonstrated strong membranous TRKB and TRKC staining in cylindromas, as well as elevated levels of ERK phosphorylation and BCL2 expression. Membranous TRKC overexpression was also observed in 70% of sporadic basal cell carcinomas. RNA interference mediated silencing of TRKB and TRKC, as well as treatment with the small molecule TRK inhibitor lestaurtinib, reduced colony formation and proliferation in three-dimensional primary cell cultures established from CYLD mutant tumours. These results suggest that TRK inhibition could be used as a strategy to treat tumours with loss of functional CYLD.
Germline truncating mutations in the tumour suppressor gene CYLD have been associated with three disfiguring hair follicle tumour syndromes; (i) familial cylindromatosis, (ii) Brooke-Spiegler syndrome and (iii) multiple familial trichoepitheliomas (Bowen et al., 2005). CYLD mutant patients develop three distinct types of cutaneous tumours, namely cylindromas, spiradenomas and trichoepitheliomas. The presence of painful spiradenomas and truncal and genital tumours can have a significant impact on quality of life, as can the surgery used to remove lesions as this may culminate in entire scalp removal. As such, the development of non-surgical approaches to control tumour burden are required (Rajan et al., 2009). Novel therapeutic approaches with inhibitors of the nuclear factor kappa beta (NFκB) pathway such as aspirin were proposed when CYLD was found to encode an ubiquitin hydrolase that negatively regulated this pathway (Brummelkamp et al., 2003; Kovalenko et al., 2003; Trompouki et al., 2003). However, initial clinical studies with topical salicylic acid demonstrated a significant response in only a small proportion of treated tumours (Oosterkamp et al., 2006). Therefore, a more detailed understanding of the molecular dysregulation caused by loss of functional CYLD is warranted and this may inform the development of novel therapeutic approaches.
CYLD encodes a ubiquitin hydrolase that cleaves lysine 63 (K63) linked ubiquitin chains (Brummelkamp et al., 2003; Kovalenko et al., 2003; Trompouki et al., 2003), the specificity of which depends on the presence of a B-box domain within the CYLD protein (Komander et al., 2008). In patients with germline CYLD mutations, the majority (94%) of mutations result in premature termination codons and predict translated truncated proteins that have reduced catalytic activity (Saggar et al., 2008) which in turn is thought to perturb regulation of the NFκB pathway. CYLD substrates including TRAF2, TRAF6 and NEMO which regulate canonical NFκB signalling are regulated by K63 ubiquitin tagging, and lack of functional CYLD results in constitutively active NFκB signalling (Brummelkamp et al., 2003; Kovalenko et al., 2003; Trompouki et al., 2003). CYLD interacts with and negatively regulates TAK1, reducing TAK1 mediated stimulation of IKK and hence activation of NFκB (Reiley et al., 2007). TAK1 also phosphorylates and activates MKK6 and MKK7, leading to the activation of p38 and JNK kinase pathways, which may also contribute to disease pathogenesis (Liu et al., 2006; Reiley et al., 2004). CYLD negatively regulates BCL3, preventing nuclear entry where it forms a dimer with p50/p52, resulting in transcription of genes involved in proliferation including cyclin D1 (Massoumi et al., 2006). CYLD was also shown to negatively regulate Wnt signalling, recognised to play a role in tumorigenesis, by deubiquitinating Dishevelled (Tauriello et al., 2010). Additional roles for CYLD have been suggested in the regulation of immunity, microtubule assembly, calcium ion channel regulation, facilitating entry into mitosis. (Gao et al., 2008; Reiley et al., 2006; Stegmeier et al., 2007; Stokes et al., 2006). How perturbation of these pathways by loss of functional CYLD contributes to tumour formation is not yet clear.
Apart from cylindromas, other cancers that lack functioning CYLD or have reduced expression of CYLD include hepatocellular and colonic carcinoma (Hellerbrand et al., 2007), multiple myeloma (Annunziata et al., 2007; Demchenko et al., 2010; Jenner et al., 2007; Keats et al., 2007), lung cancer (Zhong et al., 2007), prostate cancer (Kikuno et al., 2008) and malignant melanoma (Massoumi et al., 2009). In multiple myeloma, loss of heterozygosity (LOH) at 16q, the genomic location of the CYLD gene, is an indicator of poor prognosis and is associated with reduced overall survival (Jenner et al., 2007). Both deletions involving the locus of CYLD resulting in reduced copy number, as well as biallelic events, comprising a 16q deletion with mutations within the coding exons of CYLD in the remaining allele have been shown (Demchenko et al., 2010). Overexpression of NFκB target genes has been demonstrated in myeloma consistent with the proposed role of CYLD (Jenner et al., 2007).
CYLD deficient mouse models with complete loss of CYLD expression do not develop spontaneous tumours, but show an increased susceptibility to cancer. This includes an increased susceptibility to chemical carcinogen induced cutaneous tumours (Massoumi, 2006) as well as colorectal tumours following dextran sulphate induced colitis (Zhang et al., 2006). Mice expressing truncated CYLD mutations that mimic human mutations in contrast, are not viable beyond a few hours after birth (Trompouki et al., 2009). This difference in the two phenotypes is intriguing and suggests that further work is required to understand genotype-phenotype correlations.
CYLD has been shown to interact with TRKA and plays a role in receptor internalisation and signal transduction (Geetha et al., 2005). TRK receptors were first recognised to have a tumourigenic role when an oncogenic TRK fusion protein was found in a proportion of colonic carcinomas (Martin-Zanca et al., 1986) and subsequently thyroid carcinomas (Butti et al., 1995). Dysregulation of TRK signalling has also been demonstrated in a number of epithelial tumours (Lagadec et al., 2009; Ricci et al., 2001; Weeraratna et al., 2000).
Here, we have performed a detailed molecular analysis of tumours from patients with CYLD mutations. This revealed up-regulation of TRK signalling in CYLD defective tumours. We went on to develop a three dimensional cell culture from primary tumour cells to demonstrate the potential of therapeutically targeting TRK for the treatment of these tumours.
Fresh frozen tumours from patients with germline CYLD mutations were microdissected to isolate tumour cells and perilesional control cells. Genomic analysis of cylindroma and spiradenoma tissue was performed on 12 tumours and 7 perilesional skin samples using a 32K bacterial artificial chromosome (BAC) tiling path array (Natrajan et al., 2009). This BAC array platform has been shown to be as robust as, and to have comparable resolution with high density oligonucleotide arrays (Coe et al., 2007; Gunnarsson et al., 2008; Tan et al., 2007). Comparative genomic hybridisation was performed using DNA extracted from peripheral blood leukocytes from the same patient as the reference. No obvious regional genomic amplification or deletion was noted in 12 tumour samples (Figure 1a). Moreover, no difference was seen between cylindromas (n= 7) and spiradenomas (n=5).
Given the apparent absence of genomic copy number variation observed using BAC arrays, we went on to perform high resolution SNP typing using a 550K SNP array platform (Illumina) on four cylindroma tumours and one control sample. This revealed copy number silent LOH for the entire arm of chromosome 16q in all four tumours but no other significant changes (Figure 1b).
LOH status at the site of the mutation was assessed in a further 27 tumours used in this study, which was possible as the genotype of each patient was known. LOH was assessed using genomic DNA from microdissected tumours by using restriction enzymes that specifically were unable to cut only the mutated allele in each tumour. Polymerase chain reaction (PCR) amplification was performed using primers that flanked exon 18, and the PCR product was subject to restriction enzyme digestion using enzymes chosen for pedigree specific mutations: HpyCH4V (c.2460delc) and Sml I (c.2469+1G>A). This revealed LOH at the site of the mutation in 75% of the 27 tumours, suggesting the mechanism resulting in LOH to be similar in the majority of tumours.
Gene expression profiling was performed on total RNA extracted from 42 microdissected samples, consisting of 19 cylindromas, 9 spiradenomas, 4 trichoepitheliomas and 10 perilesional controls. A bead array platform (Illumina) that is capable of assaying expression of 24,526 transcripts for each sample was used (April et al., 2009). Average signal values from replicate beads were summarised, log transformed, and normalised using quantile normalisation across all samples (Dunning et al., 2008; Workman et al., 2002). Unsupervised clustering analysis of the individual transcriptomes (24526 probes) by Euclidean distance revealed close clustering of the majority of cylindroma and spiradenoma tissue, and distinct clustering of the perilesional skin and trichoepithelioma tissue (Figure 2). This was corroborated with clustering by an independent technique that used multiple bootstrapping to determine the robustness of the clusters. Clustering of transcriptomes using individual probes (24526) was employed and dissimilarity was measured using the correlation method using a statistical package (pvclust) (Suzuki and Shimodaira, 2006). Clusters were maintained, however due to the small numbers of tumours used, statistical significance was seen between tumours and controls (p<0.01), but not between tumour subsets (Supplementary Figure 3). The unsupervised clustering analysis was further corroborated by a supervised clustering analysis, allowing class comparison between cylindroma and spiradenoma tumours. 19 cylindromas and 9 spiradenomas were subject to significance of microarrays analysis (SAM), which showed no significant difference between the tumour types when standard criteria were applied (Supplementary Figure 4). (Differentially expressed genes between the two groups had to pass a significance threshold of p<0.01 and 100 permutations of the data were analysed)
Quantitation of differential gene expression was determined using the bead array manufacturers’ custom algorithm (Illumina custom method- Illumina Inc, 2009). The statistical significance of differential gene expression between tumours and samples was calculated as a differential score. This calculation incorporated the difference in signal values between the two groups compared, corrected for appropriate negative control bead values, and was corrected for multiple hypothesis testing using a Bonferroni post test. We used stringent thresholds for filtering genes, with only transcripts that were differentially expressed with a p value of < 0.01 included for subsequent analysis (n=4492). Given the genomic and transcriptomic similarity seen in cylindromas and spiradenomas, these were pooled and analysed against pooled controls to provide enhanced statistical robustness.
Transcripts of proteins that were previously noted to be overexpressed in cylindromas were also present and supported the enrichment of cylindroma tissue following microdissection. These included numerous laminins (Table 1) (Tunggal et al., 2002), collagens 4 and 7 (Bruckner-Tuderman et al., 1991; Timpl et al., 1984) and cytokeratins (Meybehm and Fischer, 1997; Tellechea et al., 1995) which were preferentially expressed in the tumours compared to perilesional epidermis.
Upregulation of NFκB target genes (http://people.bu.edu/gilmore/nfkb/target/index.html) was present as expected and served as an internal positive control. All tumours were compared to all controls, and the NFκB target genes that were differentially expressed with a p-value of less than 0.01 were tabulated in Table 1. To detect low level changes in gene expression between all tumours and all control tissue, gene set enrichment analysis (GSEA) was performed on data filtered at a threshold of p< 0.05 (Subramanian et al., 2005). This highlighted multiple gene sets that were involved in apoptosis and genes that were common to these sets were found to be members of or target genes of NFκB and JNK signalling pathways (Figure 3, Supplementary Figure 1).
Fold change (FC) analysis (Table 2) revealed overexpression of several members of the TRK signalling pathway compared to perilesional tissue, namely, TRKC (FC = 4.75x), Neurotrophin 3 (FC=3.00x) and Neurotrophin 4/5 (FC=3.15x) in the tumour group when compared to perilesional skin. This finding was corroborated by connectivity mapping analysis (Lamb, 2006) using Ingenuity Pathways Analysis (Ingenuity® Systems, www.ingenuity.com). This highlighted the concerted overexpression of multiple members of the Neurotrophin/TRK signalling pathway in tumour tissue compared to control tissue (p = 4.02 × 10−3). In addition to the transcripts related to the TRK signalling pathway detected by FC analysis above, TRKB, brain derived neurotrophic factor and PIK3R1, which encodes a regulatory subunit of PI3 kinase, a downstream mediator of TRK signalling, were elevated in expression (Figure 4). Moreover, TRKA, a previously described CYLD interacting protein (Geetha et al., 2005), was found to be downregulated, whilst the cognate ligand nerve growth factor (NGF) was upregulated. No obvious changes in copy number were found at the chromosomal location of the different TRK receptors and their respective cognate ligands suggesting that genomic amplification or deletion was not directly responsible for these changes in transcript level.
We investigated whether the overexpression of TRKB and TRKC observed at the mRNA level was reflected in elevated protein expression in CYLD tumours. We constructed a tumour tissue microarray (TMA) of CYLD related tumours - cylindromas, spiradenomas, trichoepitheliomas and perilesional skin and analysed this using immunohistochemistry. Expression of TRKB and TRKC was elevated in cylindroma, spiradenoma and to a lesser extent in trichoepithelioma tissue compared to perilesional skin (Figure 5a). Strong membranous staining was apparent in cylindroma cells, whilst diffuse cytoplasmic staining was noted in overlying keratinocytes. TRKC expression was mainly noted over the basophilic peripheral cells in cylindroma islands, whilst TRKB overexpression was non-discriminatory within tumour islands (Figure 5a). Quantification of membranous TRK expression was performed on each core of the TMA. A scoring scale, which assessed the intensity of membranous TRK staining and the proportion of tumour cells stained was used, (Figure 5 b-c), and the difference in expression between tumours and perilesional keratinocytes was found to be statistically significant p<0.05 (Tables (Tables33--4).4). Furthermore, cylindroma cells were strongly positive with antibodies against phosphorylated ERK (p44/42), a marker of active TRK signalling, mainly at the periphery of each tumour island. The antiapoptotic protein BCL2, a downstream target of TRK signalling, was also overexpressed in cylindroma tissue, showing strong perinuclear staining (Figure 5a)
To assess the functional significance of TRKB and TRKC in cylindromas, primary cell cultures were established from fresh tumour samples from the patients that contributed samples used for the molecular profiling and used for functional studies (Figure 6a). Cylindroma tissue was microdissected from normal perilesional tissue under a stereomicroscope, subject to enzymatic digestion to obtain a single cell suspension, and cultured under standard conditions (5% CO2, 21% O2). These primary cell cultures were used within four weeks of surgery to protect against culture induced genomic alterations. To validate these cultures, CYLD expression was characterised and these cells did not express full length CYLD, which was observed in perilesional fibroblast culture (Figure 6b). Cylindroma cells in culture expressed cytokeratins 6, 14, 17 and smooth muscle actin, markers that were shown to be present in cylindroma tumours in vivo, confirming the purity of the cultures (Figure 6c-6d). These primary cells were transduced with lentiviruses expressing short hairpin (sh) RNAs targeting TRKB and TRKC. The shRNA vectors carried a GFP reporter allowing monitoring of delivery and a puromycin gene allowing selection - cells were grown in media containing puromycin for 48 hours post transfection resulting in cultures that were approximately 100% GFP positive when analysed. TRKB and TRKC knockdown was assayed using immunoblotting (Figure 6e-6f) and after 10-14 days in culture, cells were fixed, stained and colonies were counted. Primary cells with knockdown of TRKB and TRKC demonstrated a modest reduction (20-40%; p<0.05) in colony formation (Figure 6g).
The primary cell cultures described above expressed only low levels of TRK, an observation made in previous studies (Lagadec et al., 2009). To establish a more biologically relevant cylindroma culture model we seeded primary cells on three-dimensional (3D) polystyrene tissue culture scaffolds to allow enhanced cell–cell contact (Figure 7a). This caused increased expression of TRKB and TRKC when compared to matched cells grown in standard two dimensional (2D) culture (Figures 7b and 7c). To determine if TRK signalling was active, we examined phosphorylated ERK and BCL2 after stimulating the cultured cells with cognate ligands of the TRK receptors (NGF, BDNF and NT3). Stimulation indeed resulted in increased phosphorylated ERK above basal levels (Figure 7d). Antiapoptotic factor BCL2 has been shown to be upregulated in response to NT3 in oligodendrocyte progenitor cells (Saini et al., 2004). This was recapitulated in our model, where BCL2 expression was increased following stimulation with TRK receptor ligands (Figure 7e). To determine if these cultures were dependent on TRK signalling for survival, they were treated with small molecule TRK kinase inhibitors. As a control they were also exposed to salicylic acid which has previously been shown to effect survival in CYLD deficient cells (Brummelkamp et al., 2003). Cells were only modestly sensitive to salicylic acid (surviving fraction (SF50) 1.5mM), but were much more so to the pan-TRK inhibitors, lestaurtinib (CEP-701) (SF50, 4.57μM), AG879 (SF50, 4.21 μM) and K252a (SF50, 366 nM) (Figure 8). It seems possible that loss of CYLD catalytic activity accounts for the TRK dependency. To address this, we introduced a cDNA construct expressing CYLD into primary tumour cells, with the aim of assessing whether CYLD expression could rescue the phenotypes observed. However, as is often the case when genes are artificially expressed from cDNA constructs, the reintroduction of CYLD cDNA resulted in reduced cell viability, precluding further mechanistic studies (Supplementary Figure 5).
Cutaneous appendageal tumours seen in CYLD mutation carriers share similarities in cytokeratin profiles with sporadic basal cell carcinoma (BCC), that in humans are thought to be also derived from the hair follicle (Donovan, 2009). Dysregulation of several mutual oncogenic pathways are seen in cylindromas and BCCs. Wnt signalling is thought to play an oncogenic role in both, with expression of nuclear β-catenin demonstrated in both cylindromas (Tauriello et al., 2010) and human basal cell carcinoma (Kriegl et al., 2009). Also, the Sonic hedgehog pathway, recognised to depend on such canonical Wnt signalling for cutaneous tumour formation (Hoseong Yang et al., 2008), is an important final common pathway in both tumours. In a murine model that overexpresses GLI1 (Nilsson et al., 2000), tumours that had histological appearances that resembled both human basal cell carcinoma and cylindroma were seen. Diffuse light expression of TRK receptors has previously been noted in a small series of basal cell carcinoma using a pan TRK antibody (Chen-Tsai et al., 2004). Therefore, we went on to investigate if the membranous TRKB and TRKC overexpression we observed in CYLD mutant tumours was a feature that extended to sporadic skin cancers. We examined expression of TRK receptors in a human skin cancer TMA that included samples of squamous cell carcinoma (SCC; n=19), melanoma (MM; n=5) and basal cell carcinoma (BCC; n=23) using TRKA, TRKB and TRKC specific antibodies. Interestingly, membranous TRKC was overexpressed in 70% of BCC tumours (Figure 9), 5% of SCC but not melanoma. In some samples, TRKC staining was strongest at the invasive front where the tumour cells apposed stroma (Supplementary Figure 2). No membranous TRKA staining was seen, however some perinuclear staining, a feature associated with activated TRKA signalling (Wu et al., 2007) and was seen in a 39% of cases of BCC and 16% of cases of SCC. TRKB staining was seen in the cytoplasm in 13% of cases of BCC only, where it was seen in a few cells, but not at the cell membrane.
CYLD mutation carriers face repeated, disfiguring surgery to control tumour burden, and in a study of two large pedigrees, up to one in four affected patients underwent total scalp removal (Rajan et al., 2009). Painful eccrine spiradenomas, the development of conductive deafness caused by occlusive tumours in the ear canal and the risk of malignant transformation (Gerretsen et al., 1993) justify the development of non-surgical alternatives for this patient group. Here, we performed an unbiased approach using genomic and transcriptomic analysis of CYLD defective tumours. We demonstrated that cylindroma and spiradenoma tumours were genomically similar and stable, with LOH at 16q found in the majority of tumours in this study. The genomic stability seen across 16 tumours could suggest homogeneity in the mechanisms of pathogenesis and potentially in the response to therapeutic agents. Moreover, it seems possible that genomically stable tumours may be less liable to the acquisition of resistance mechanisms that arise through genomic diversity. We detected upregulation of transcripts of TRKB and TRKC and their cognate ligands NT3/NT4 and BDNF in cylindroma and spiradenoma samples. This was of particular interest as CYLD, a deubiquitinase with K63 linked ubiquitin specificity, has previously been shown to deubiquitinate TRKA and inhibit its internalisation from the cell membrane and subsequent downstream signalling via the phosphatidylinositol-3-kinase (PI3K) pathway (Geetha et al., 2005). Furthermore, recent studies have shown that BDNF is a NFκB target gene that may afford a neuroprotective role in astrocytes, supporting the finding that this TRK ligand is overexpressed by the cylindroma cells resulting in an autocrine loop (Saha et al., 2006). Validation of protein expression using TRK specific antibodies demonstrated strong membranous expression of TRKB and TRKC in cylindromas and spiradenomas, when compared to perilesional epidermis and hair follicles. Finally, downstream targets of TRK signalling such as phosphorylated ERK and BCL2 were overexpressed in the tumours.
TRK receptors have received considerable attention as mediators of neuronal growth, survival and maintenance (Huang and Reichardt, 2001), and have been shown to be expressed in hair follicles (Adly et al., 2005; Blasing et al., 2005). Their role in cancer was initially shown by uncommon oncogenic fusions involving C-terminal sequences of the NTRK1/NGF receptor gene with 5′ terminal sequences of various activating genes, such as TPM3, TPR and TFG in colon and papillary thyroid carcinoma (Butti et al., 1995; Martin-Zanca et al., 1986). Exploitation of the constitutively active prosurvival signalling that is usually restricted to neurons was thought to confer an advantage for the cancer cells. Recent advances in the understanding of complex signalling pathways involved in cancer (Luo et al., 2009) has suggested that some neural derived cancer cells, such as neuroblastoma, may become dependent on TRK signalling for survival, independent of mutations in TRK (Matsumoto et al., 1995; Scala et al., 1996). Subsequent demonstration of TRK expression associated with a survival advantage in epithelial tumours including breast (Lagadec et al., 2009), prostate (Weeraratna et al., 2000), lung (Ricci et al., 2001), extended the role of TRK signalling to non-neural derived tumours. At a genomic level, we established that neither TRKB and TRKC amplification, nor that of BDNF and NT3/4 was demonstrated to account for the overexpression seen in cylindroma tumours, suggesting that CYLD dysfunction perturbs normal TRK homeostasis.
The differential expression of TRKA, TRKB and TRKC in these tumours is intriguing. Whilst TRKA is regulated by CYLD, the interaction of CYLD with TRKB and TRKC remain to be clarified. The low perinuclear expression of TRKA, but conversely the high membranous expression of TRKB and TRKC suggests that not all TRKs are regulated by CYLD in the same manner. This has been shown to be true with other forms of ubiquitin modification including monoubiquitunation (Arévalo et al., 2006) or Lys48 polyubiquitination (Geetha et al., 2005). Arévalo et al. demonstrated that an E3 ubiquitin ligase, Nedd4-2, associated with the TRKA receptor and was phosphorylated upon NGF binding whilst Nedd4-2 did not bind or ubiquitinate related TRKB receptors, due to the lack of a consensus PPXY motif. These results indicate that TRK receptors are differentially regulated by ubiquitination to modulate the survival of neurons (Arévalo et al., 2006). Interestingly, Geetha et al. have demonstrated that the scaffold protein p62 which facilitates the interaction between CYLD and TRKA, is able to bind to specific lysine residues on TRKB (Lys 811) and TRKC (Lys 602 and 815) and mutation of these sites disrupted downstream signalling (Geetha et al., 2008). This suggests the possibility that CYLD could interact with TRKB and TRKC, but the precise mechanism of TRK specific homeostasis remains to be clarified.
We established TRK signalling was effective in cylindroma primary cell culture using recombinant human BDNF and NT3 for TRKB and TRKC respectively, and demonstrated increased expression of phosphorylated ERK and BCL2. To delineate the advantage conferred by TRKB and TRKC overexpression, we used a lentiviral mediated knockdown system. This demonstrated reduced colony formation in TRKB and TRKC deficient cells supporting the pathogenic role of these receptors.
We attempted to investigate if the catalytic activity of CYLD was implicated in the perturbed TRK signalling seen in these cultures. However, reintroduction of wild type-CYLD resulted in reduced cell viability preventing further investigation in this model (Supplementary Figure 5). The dependence of perturbed TRK signalling seen on NFκB and JNK signalling was investigated by growing primary cultures in NFκB and JNK inhibitors, however we were unable to conclusively show a clear effect on TRK levels (data not shown).
To investigate whether dysregulated TRK signalling in patients with truncating CYLD mutations represented a druggable pathway, we employed a 3D primary cell culture model. As seen in breast cancer (Lagadec et al., 2009), TRK expression in vitro was low when 2D cylindroma cell culture was compared to tumours. 3D culture was associated with an increase in expression of TRKB and TRKC, allowing us to use this as an in vitro model to assay cell viability in the presence of TRK inhibitors. The reduction in viability at micromolar concentrations of TRK inhibitors is encouraging, particularly as these tumours are amenable to transcutaneous or intralesional drug delivery. We attempted to determine whether the mechanism of action of TRK inhibition in our primary cell culture model was mediated by ERK or BCL2. This analysis proved not to be possible because of the high sensitivity of the primary cells to TRK inhibitors, resulting in only very few live cells after TRK inhibitor exposure. The result seen with lestaurtinib is particularly interesting as this agent was recently found in a screen of 2800 compounds to be a potent inhibitor of NFκB, suggesting it may target two key pathways in CYLD mutant tumours (Miller et al., 2010). Lestaurtinib is currently in phase 3 clinical trials (Shabbir and Stuart, 2010), and serum levels in patients on a 80 mg oral dose twice a day was 17-25 μm after taking the drug for 28 days (Marshall et al., 2005), suggesting that the sensitivity of CYLD deficient tumours suggested by our model may be achievable in patients. As novel agents with increased pan TRK specificity such as AZ-23 become available (Thresset al., 2009), patients with truncating CYLD mutations may be another step closer to non-surgical alternatives for this disfiguring disease.
We chose to investigate the expression of TRK in other human skin cancer models. The over expression of membranous TRKC in a high proportion of basal cell carcinoma is interesting. In sporadic BCC, LOH at 9p, the locus for PTCH1, has been found in the majority (93% in this series) of basal cell carcinoma (Teh et al., 2005), whilst 16q, the locus for CYLD, appears rarely affected. TRKC signalling in non melanoma skin cancer has not been characterised, but there are data from other models to support that TRK signalling could confer a proliferative advantage to cells where constitutively active SHH signalling following loss of functional PTCH could occur. Synergy between NT3, the cognate ligand for TRKC, and SHH has already been demonstrated in murine motorneurons where both are important for growth, and use of NT3 antibodies in this system reduces growth of these neurons (Dutton et al., 1999). TRKC also directly interacts with bone morphogenetic signalling receptors (BMPRI) in murine epithelial cells, and inhibits the tumour suppressor effects of BMP signalling (Jin et al., 2007) in a mouse colon cancer model. BMP and SHH signalling are tightly balanced, and BMP2, the ligand of BMPRI has been shown to oppose SHH mediated proliferation in murine cerebellar granular precursor cells via regulation of N-myc (Alvarez-Rodríguez et al., 2007). Furthermore, BMP signalling is protective in cutaneous carcinoma, and a SMAD 4 conditional knockout expressed in murine keratinocytes has indeed shown to result in spontaneous squamous and basal cell carcinomas (Qiao et al., 2006). It is hence conceivable that TRKC mediated inhibition of BMP signalling, on a mutated PTCH background in basal cell carcinoma may play a pathogenic role. The localisation of TRKC at the invasive margin of basal cell carcinoma tumour islands suggests that this role may be in invasion (Supplementary Figure 2). The mechanism that results in TRKC overexpression in basal cell carcinoma however remains to be explored. However, TRK receptors may represent a novel therapeutic target both in tumours with loss of CYLD function but also for basal cell carcinoma, the commonest cancer in humans.
Tumour tissue was obtained from patients undergoing surgery indicated to control tumour burden or for symptomatic relief, under regional ethical committee approval (REC REF:06/1001/59). Patients provided blood samples from which genomic DNA was extracted. Patients had been genotyped from all three pedigrees, and were confirmed to have the same mutation when entered into the study. These mutations were c.2460delc, c.2469+1G>A and c.2290del5 (Accession No. NM_015247).
All samples for microarray studies were snap frozen immediately after surgery, with detailed location maps of the tumours made at the time. Tumour blocks were mounted in optimal cutting temperature compound and sectioned on a cryostat at 12 μm onto RNAse free Superfrost coated slides (VWR). All subsequent steps were performed in RNAse free solutions. Tissue sections were fixed in 100% ethanol at −20°C for 2 minutes and then washed in 70% DEPC treated ethanol and 50% DEPC treated ethanol for 2 further minutes. Sections were stained with crystal violet solution (Ambion) for 40 seconds and then dehydrated in increasing concentrations of ethanol. Slides were then stored on dry ice until needle microdissection was performed. Microdissection was performed at 4°C using a Leica dissecting microscope. The dissected samples were placed directly into Trizol ® (Invitrogen UK). Needle homogenisation was performed and the sample was frozen at −80°C until RNA and DNA extraction was performed.
Trizol® (Invitrogen UK) was used to extract RNA and DNA from the samples. The standard supplied protocol was used, with glycogen added to aid precipitation of RNA into the aqueous phase. RNA was quantified using a Nanodrop spectrophotometer (Nanodrop). RNA integrity analysis was performed using a chip based electrophoresis kit to confirm preservation of the 18S and 28S ribosomal bands (RNA 6000 Nano and Pico Chip kits; Agilent 2100 Bioanalyzer).
Genomic DNA from peripheral blood was extracted using the QIAGEN whole blood kit according to manufacturer’s protocols.
The 32K BAC re-array collection (CHORI) tiling path aCGH platform was constructed at the Breakthrough Breast Cancer Research Centre, as described earlier (Marchio et al., 2008). This type of BAC array platform has been shown to be as robust as and to have comparable resolution with high- density oligonucleotide arrays. DNA labelling, array hybridisations and image acquisition were performed as described earlier (Turner et al., 2010). aCGH data were pre-processed and analysed using an in-house R script (BACE.R) in R version 2.9.0 as previously described(Natrajan et al., 2009). After filtering polymorphic BACs, a final data set of 31544 clones with unambiguous mapping information according to the August 2009 build (hg19) of the human genome (http://www.ensembl.org) was smoothed using the circular binary segmentation (cbs) algorithm (Natrajan et al., 2009). A categorical analysis was applied to the BACs after classifying them as representing amplification (>0.45), gain (>0.12 and <0.44), loss (< −0.12), or no-change according to their cbs-smoothed Log2 ratio values. ArrayCGH data were deposited at ROCK (http://brcabase.icr.ac.uk/index.jsp), at URL: http://rock.icr.ac.uk/collaborations/Rajan/Cylindroma.
Genomic DNA was extracted from microdissected tissue and assayed on an Illumina 550K SNP array according to manufacturers protocols. SNP array data were deposited at ROCK (http://brcabase.icr.ac.uk/index.jsp), at URL: http://rock.icr.ac.uk/collaborations/Rajan/Cylindroma.
Polymerase chain reaction (PCR) was performed using primers (5′-3)′ CYLD18F –GAGAGCTTAAGCAGATGGAA and CYLD 18R –TAAACAGAAAAGGCAAAAGC that flanked exon 18, and the PCR product was subject to restriction enzyme digestion using HpyCH4V (c.2460delc) and Sml I (c.2469+1G>A). Digested products were separated on an agarose gel and visualised using ethidium bromide on an ultraviolet transilluminator. LOH was determined by the absence of a band that corresponded with loss of the wild-type allele in both cases.
The Illumina WG-DASL platform was used for each sample that allowed for monitoring of gene expression of 24,526 transcripts. 50ng of total RNA from each sample was converted to cDNA as per the supplied protocol and then hybridised to Illumina Human 8 v3 chips. A confocal Illumina Infinuim bead reader then scanned the chips and the results were extracted using Illumina Beadstudio software. Microarray gene expression profiles were deposited at Arrayexpress (www.ebi.ac.uk) at accession number E-MTAB-352.
Raw gene expression values were normalised using quantile normalisation. Differential expression analysis was performed using Beadstudio software to detect significantly different expressed genes. Differences in signal values for each gene was scored with a differential score (DS). Bonferroni testing was performed to correct for multiple hypothesis testing, and the results were ranked in order of differential score with a differential score ranging from +374 to −374, with +/−65 corresponding to a p value cut off of <0.01. Signal values from genes that fulfilled this level of significance were analysed for fold change, and results were tabulated in Table 1. Pathway discovery was further aided by Ingenuity pathway analysis software. Finally, a parallel analysis was carried out using gene set expression analysis (GSEA). For this analysis, probe values (24526) were collapsed to gene symbols (18630), with the max probe value used for replicate probes. After filtering, 867 gene sets were used in the analysis. Gene sets were permutated 1000 times, before being ranked in order using a “difference of classes” metric, with a weighted enrichment statistic applied. 56 gene sets were significantly enriched at a nominal p-value of <0.01.
Total RNA was extracted from cylindroma primary cell cultures as described above, subject to DNAse digestion and then reverse transcribed using a Superscript III kit (Invitrogen) according to manufacturers instructions. cDNA was then used for quantitative PCR with the following primers (5′-3)′: (GAPDH F – ATGGGGAAGGTGAAGGTCG, GAPDH R –GGGGTCATTGATGGCAACAATA, NTRK2 F – TGTTCAGCACATCAAGCGACA, NTRK2 R – GCTCAGGACAGAGGTTATAGCAT, NTRK3 F TCCGTCAGGGACACAACTG, NTRK3 R –GCACACTCCATAGAACTTGACA) with a SYBR green containg Taq polymerase master mix (Sigma), in an ABI 7900HT thermal cycler. Gene expression was normalised to GAPDH and comparisons in expression were made using the 2−ΔΔCT formula as the PCR reactions had similar efficiencies of amplification.
Homogenised tumour samples, and lysates were sonicated in lysis buffer (65mM Tris, 50mM NaCl, 5mM EDTA, 1mM PMSF, 1mM Na orthovanadate, 0.1%SDS, 2% TritonX and 1% NP-40) with added phosphatase and protease inhibitors (Roche). 5-20 ug of total protein was loaded onto SDS-PAGE gels, and then transferred to PVDF membranes. Immunoblots were probed with the following antibodies: TRKB, TRKC, BCL2, p44/42, β-actin (Cell signalling Technology) and visualised with appropriate HRP conjugated secondary antibodies using ECL-Plus (GE healthcare). Immunoblots were exposed to Biomax XR film and processed routinely.
A custom made tissue microarray was made from single 2mm cores from representative tumour areas and was used to validate protein expression in tissue sections. The TMA contained 10 cylindroma samples, 4 spiradenoma samples and 2 trichoepithelioma samples. Cylindromas and spiradenomas were chosen from both scalp and torso, and trichoepitheliomas were from the face. Human skin cancer TMA slides (A216) were purchased from Cepheid UK.
Tissue sections from the TMA were dewaxed with xylene and rehydrated in graded ethanols, before undergoing microwave antigen retrieval in sodium citrate buffer (pH 6.0). Tissue sections were then blocked with peroxidase blocking agent (DAKO) and then incubated with the following primary antibodies: TRKA, TRKB, TRKC, p44/42 (phospho ERK) and BCL2. Tissue sections were then washed in phosphate buffered saline and probed with secondary HRP-conjugated antibodies, and staining was visualised with 3,3′-Diaminobenzidine (DAB). Haematoxylin was used as a nuclear counterstain. Perilesional epidermis was used as a negative control, and anagen hair follicles in perilesional tissue served as a positive control for TRK antibodies.
Tissue sections were assessed for staining in 5 randomly selected high power fields per core on the TMA. A modified version of an established score to assess membranous staining of Her-2 was developed (Kavanagh et al., 2009).To quantify TRK staining, tumour cells were scored as having circumferential strong membranous staining (3), strong non-circumferential staining (2), weak staining (1) or no staining (0). Proportion of such cells stained were counted per HPF on a 63x objective on a Leica DMR microscope and scored as follows: 75-100% of cells (3), 50-75% (2), and less than 50% (1). A total score of up to 6 was possible per field per core and scores were compared to perilesional keratinocytes for statistical significance using a t-test.
Cylindroma samples were micro dissected using a Zeiss microscope following surgery and tumour pieces were enzymatically digested in 2.5% Trypsin. A cell suspension was then counted for viable cells using Trypan blue and the cells were plated in collagen coated tissue culture plasticware and grown in keratinocyte serum free medium. Lentiviral transfection was performed with shRNA against TRKB, TRKC and compared to a non-silencing control (Openbiosystems). Cells were grown in puromycin selection media till all cells were green fluorescent protein positive and then used for experiments. For colony forming assays, cells were plated at a low density of 4000 cells per well of a 6 well plate. These cells were maintained for 10-14 days, before cells were fixed in 3.3% trichloroacetic acid, and stained with 0.57% sulphorhodamine in 1% acetic acid. Plates were then scanned and colonies with more than 10 cells were counted. For short term viability assays, cells were grown in a 96 well format for 96 hours, before cell viability was assayed using a luminescent assay that lysed cells and used liberated ATP as readout for viability (CellTiter Glo - Promega). This was carried out according to standard manufacturers protocols and luminescence was read on a luminometer (Luminoskan – Ascent). For 3D cultures, cells were seeded on a polystyrene collagen coated scaffold (Reinnervate, Durham, UK), grown for 28 days and then treated with drugs. Cells were grown in the following inhibitors were used, diluted in DMSO, with a final concentration of DMSO at 0.1%: K252a and Tyrphostin AG879, Salicylic acid (SIGMA), Lestaurtinib (Toronto Research Chemicals). After 14 days, CellTiter Glo (Promega) was added to the media according to standard manufacturers protocols and luminescence was read on a luminometer (Luminoskan – Ascent). All experiments were done in triplicate.
Cells were trypsinised, resuspended in PBS and fixed by adding ice cold ethanol whilst gently vortexing the tube, until a final concentration of 70% ethanol was reached. Cells were fixed for 1 hour at 4° C. Cells were then pelleted and resuspended in Cystain Solution (Partech, UK) which stained the nuclear DNA with DAPI. Cells were stained for 1 hour in the dark at 4° C, before analysis on a FACS Canto II flow cytometer (Becton Dickson, UK) using a UV laser. Cells were analysed for total DNA content and a minimum of 10000 events were used for cell cycle data. All cell cycle data was analysed using Modfit (Verity Software House, USA)
Expression levels of 24K genes from 32 microdissected tumours were pooled and compared against 10 pooled perilesional controls and subject to gene set expression analysis. Sets of genes recognised to modulate apoptosis were found to be enriched in the tumour sample. Leading edge analysis of these gene sets revealed NFκB target genes to be frequently present in these samples.
A skin cancer tissue microarray containing 23 BCCs, 19 SCCs and 5 malignant melanomas with two cores per case was immunostained with antibodies against TRKA, TRKB and TRKC. (a-j) TRKA and TRKC staining varied across basal cell carcinoma tumour samples, with membranous TRKC seen at the invasive edge adjacent to stroma, indicated with white arrows Figures S2 (a,c,f and h) are enlarged in (d,e,i and j). TRKA staining was cytoplasmic, but predominantly perinuclear at the invasive front, indicated with black arrows. Figures S2 (a-c, f-h) were taken at 5x original magnification and Figures S2 (d-e, i-j) were taken at 20x original magnification. Scale bars represent a distance of 100μm.
Clustering of transcriptomes using individual probes (24526) was employed and dissimilarity was measured using the correlation method using a statistical package (pvclust) (Suzuki and Shimodaira, 2006). Clusters seen in Figure 2 were maintained, however due to the small numbers of tumours used, statistical significance was seen between tumours and controls (p<0.01), but not between tumour subsets. pvclust provides two types of p-values: AU (Approximately Unbiased – numbers in red) p-value and BP (Bootstrap Probability – numbers in green) value. AU p-value, which is computed by multiscale bootstrap resampling, is a better approximation to the true unbiased p-value than the BP value computed by normal bootstrap resampling. Numbers in grey are edge values.
Significance of microarrays analysis, a form of supervised clustering analysis was performed between 19 cylindroma and 9 spiradenoma samples, and showed no significant difference between samples types.
Cells were plated in equal numbers and transfected with empty vector, wild type CYLD and red fluorescent protein, and selected with puromycin. At Day 5 post selection, cells were trypsinised and viable cells counted. (a) Expression of CYLD was assessed using immunoblotting. (b) Small numbers of viable cells were seen with CYLD reintroduction. (c) FACS analysis was performed to determine changes in cell cycle and highlighted a sub G0/G1 peak suggestive of apoptosis. The experiments were performed in three different primary cell cultures derived from separate tumours.
This work was funded in part by grants from the Newcastle Healthcare Charities Trust, the North East Skin Research Fund, Breakthrough Breast Cancer, Cancer Research UK and the Medical Research Council. Neil Rajan is a MRC Clinical Training Fellow.
Conflicts of interest
The authors declare no conflicts of interest