Metastatic cutaneous squamous cell carcinoma shows frequent deletion in the protein tyrosine phosphatase receptor Type D gene

Cutaneous squamous cell carcinoma (cSCC) is the second most common form of nonmelanoma skin cancer (NMSC), and its incidence is increasing rapidly. Metastatic cSCC accounts for the majority of deaths associated with NMSC, but the genetic basis for cSCC progression remains poorly understood. A previous study identified small deletions (typically <1 Mb) in the protein tyrosine phosphatase receptor Type D (PTPRD) gene that segregated with more aggressive cSCC. To investigate the apparent association between deletion within PTPRD and cSCC metastasis, a series of 74 formalin‐fixed paraffin‐embedded tumors from 31 patients was analyzed using a custom Illumina 384 SNP microarray. Deletions were found in 37% of patients with metastatic cSCC and were strongly associated with metastatic tumors when compared to those that had not metastasized (p = 0.007). Subsequent mutation analysis revealed a higher mutation rate for PTPRD than has been reported in any other cancer type, with 37% of tumors harboring a somatic mutation. Conversely, bisulfite sequencing showed that methylation was not a mechanism of PTPRD disruption in cSCC. This is the first report to observe an association between deletion within PTPRD and metastatic disease and highlights the potential use of these deletions as a diagnostic biomarker for tumor progression. Combined with the high mutation rate observed in our study, PTPRD is one of the most commonly altered genes in cSCC and warrants further investigation to determine its significance for metastasis in other tumor types.

Nonmelanoma skin cancer (NMSC) has a predicted prevalence equal to that of all other cancers combined and its incidence is increasing. 1 There are more than 81,500 new cases of NMSC diagnosed annually in the United Kingdom alone, placing a heavy burden on both patients and healthcare resources. 2 Despite accounting for only 20% of total NMSC cases, cutaneous squamous cell carcinoma (cSCC) is responsible for the majority of NMSC deaths, largely as a result of metastatic disease. A subset of immunosuppressed individuals, such as organ transplant recipients on long-term immunosuppressive drugs, develop particularly aggressive tumors with an increased risk of metastasis. 3 Risk factors for metastasis include poor differentiation of the tumor cells, large tumor size, tumor depth >5 mm, immunosuppression and localization on the ear and lip. [4][5][6][7] Metastatic cSCC is currently treated by surgical intervention and/or chemotherapy or radiotherapy and is associated with a poor outcome. 8 Despite the well-established role of ultraviolet radiation (UVR) in the etiology of cSCC, the molecular events underlying its development remain largely undefined. Inactivation of the tumor suppressor genes TP53 and p16 INK4A is a common and early event in cSCC pathogenesis and is characteristic of all histological grades of tumor. 9,10 Recently, genotyping of 60 cSCC identified high rates of loss of heterozygosity (LOH) of 3p and 9p in 65% and 75% of cases, respectively. Furthermore, reduced karyotypic complexity was observed in welldifferentiated (WD) cSCC, suggesting that low-grade tumors are a distinct subset of cSCC. 11 Histologically, WD tumors show less nuclear atypia and a higher degree of keratinization than moderately (MD) and poorly differentiated (PD) tumors. However, there are no established molecular pathways that separate these subtypes of cSCC.
The genotyping analysis cited above also identified small deletions (typically <1 Mb) in the protein tyrosine phosphatase receptor Type D (PTPRD) gene in 40% (4/10) of PD tumors, 16% (3/19) of MD tumors and only 6.5% (2/31) of WD tumors, suggesting an association with the more aggressive tumors. Furthermore, all three primary cSCCs that had metastasized had deletions within this locus. 12 PTPRD is one of 21 tyrosine phosphatase receptors and has emerged as a putative tumor suppressor in a number of different cancers. Homozygous deletion of PTPRD has been reported at a frequency of 8-20% in lung cancer, melanoma, neuroblastoma, glioblastoma multiforme, laryngeal SCC and lymphoma. [13][14][15][16][17][18][19] Mutations of the coding exons of PTPRD and methylation at the promoter region have also been reported in lung cancer, melanoma, glioblastoma, breast and colorectal cancers, indicating that multiple mechanisms of disruption affect this gene in cancer. 17,[20][21][22] The overexpression of PTPRD in vitro causes transient growth arrest and an increase in apoptotic cells, with the converse demonstrated for the knockdown of PTPRD. 17,21,23 Furthermore, PTPRD has been shown to dephosphorylate the tyrosine 705 residue of pStat3 in vitro, which is the first putative target of its catalytic phosphatase activity. 21 Collectively, these studies suggest that PTPRD has growth-suppressive activities in human cancer cells.
On the basis of the previous findings, we hypothesized that deletions within PTPRD may segregate with metastatic cSCC and that alternative mechanisms of disruption may also affect the gene, as demonstrated in other cancer types. To investigate this, we have extended the analysis of PTPRD in primary and metastatic cSCCs and found a positive association between the frequency of deletion and metastasis. We have also identified novel mutations in the coding exons of PTPRD and confirmed by qRT-PCR that PTPRD mRNA levels are lowest in high-grade cSCC. The association of this gene with metastasis may provide a potential prognostic biomarker, particularly for high-risk patients who present with multiple and/or large tumors.

Sample collection
For the Illumina SNP microarray, 54 formalin-fixed paraffinembedded (FFPE) primary, recurrent and metastatic cSCC from 19 patients were obtained from the Department of Pathology, Barts and The Royal London NHS Trust. These included all cases of metastatic cSCC from the last 15 years for which the blocks were available for analysis. Five of these patients had an additional PD primary cSCC that was not associated with metastatic disease and was included for analysis. Fifteen recurrent or PD primary cSCC were obtained from a further 12 patients from the same time period. All had at least 3 years follow-up and no presentation of metastatic disease (see Supporting Information Table 1 for clinicopathological details). For the melanoma microarray, 36 primary, recurrent and metastatic FFPE tumors were obtained from 18 patients with metastatic disease and 40 primary and recurrent melanoma from 38 patients without metastatic disease. Melanoma FFPE samples were chosen retrospectively from the last 15 years. A range of Breslow thickness from <1 mm (good prognosis) to >4 mm (poor prognosis) were included. Venous blood samples were obtained from 22 of the patients with corresponding cSCC and 20 of the patients with melanoma.
Fresh-frozen tumor biopsies and short-term cultures for mutation and methylation analyses (n ¼ 41) are detailed in Supporting Information Table 2. All tumors had been previously analyzed for PTPRD genotype status. 11 Punch biopsies were obtained at the time of surgical resection and immediately snap frozen. Biopsies obtained for short-term culture were prepared directly from patient material as previously described. 24 The study was undertaken in accordance with ethical approval from the East London and City Health Authority Local Research Ethics Committee.

DNA extraction
To enrich for !70% tumor cell populations, DNA was microdissected from 10 lm FFPE sections using a reference H&E slide. DNA was extracted using the Qiagen DNA Mini Kit (Qiagen, UK) with an extended proteinase K digestion between 48 and 72 hr. After extraction, DNA was concentrated using YM-30 Microcon columns according to the manufacturer's instructions (Millipore, UK). DNA was extracted from 1 ml of venous blood using the Nucleon BACC kit (Amersham Biosciences, UK).

Custom Illumina array
A 384 SNP Oligo Pool All (OPA) was designed using the Illumina Assay Design Tool, with $6 Kb spacing across the PTPRD locus. Each SNP was chosen according to the manufacturer's recommendations (see Supporting Information). DNA (250 ng) was prepared with the OPA according to the GoldenGate Assay and hybridized to the 96-sample Universal Array Matrix. 25 Gene Calls were extracted using the Bead Scanner, and data were analyzed using BeadStudio Genotyping Module. 26 Five SNPs failed to meet the standard GenCall cutoff score of 0.4 and were excluded from further analysis, along with one metastatic cSCC sample and four melanoma samples that had call rates lower than 50%. Call rates of !70 were obtained for the remaining samples.
Eleven cSCC (from a total of eight patients) and 17 melanoma (from 15 patients) were excluded from LOH analysis because of scattered B allele frequency (BAF) distribution.
Deletions were only considered genuine events if there was a reduction in the log R ratio that was concomitant with an altered pattern of the BAF, as described by Assie et al. 27 The coordinates for deletion were taken as the first and last SNP in a run of at least five consecutive SNPs that met the above criteria. Statistical significance for the comparison of PTPRD deletion between groups was determined using the 2Â2 Fisher's test function in the ''R" programming environment.  Patient codes are ''PM'' for primary-metastatic and ''P'' for poorly differentiated tumors that did not metastasize. Dark gray shading indicates tumors in primary-metastatic series that were not available for analysis. Sample IDs in brackets are those assigned in Ref. 11 . PUVA is psoralen ultraviolet A treatment and indicates a cSCC arising in a patient who had been treated with PUVA for a pre-existing skin condition. LOH indicates loss of heterozygosity across the entire locus. 1 Both these tumors occurred in close proximity to each other, and it was not possible to distinguish clinically between two separate primary tumors or a primary and recurrence.

Sequencing
DNA extracted from 12 microdissected fresh-frozen cSCC was subjected to whole genome amplification using Repli G (Qiagen) and used in PCR as a template for 35 primer sets designed to amplify all coding exons of PTPRD (for primer sets, see Ref. 12 ). DNA extracted from seven short-passage cultured cSCC was also amplified in PCR without undergoing prior whole genome amplification. PCR products were purified using ExoSap and used directly in sequencing with BigDye v3.1 and the Applied Biosystems 96 capillary Prism 3700 sequencer. Mutations were confirmed by sequencing a second PCR product from the original nonamplified DNA sample and determined to be somatic by sequencing matched blood DNA.

Pyrosequencing
Genomic DNA (200 ng) from 25 fresh-frozen cSCC and 15 cell lines was sodium bisulfite modified using the Zymo EZ DNA methylation kit (Zymo, USA) according to the manufacturer's instructions. Unmethylated human DNA extracted from normal peripheral blood and CpGenome Universal Methylated DNA (Chemicon International, USA) were included as negative and positive controls, respectively. Approximately 10 ng of modified DNA was amplified in a 20 ll PCR reaction as previously described. 28 Primer pairs were designed using the Pyrosequencing Assay Design Software (Biotage, Sweden), see Supporting Information for primer sequences. Primer sets were demonstrated to amplify only bisulfite-modified DNA using nonmodified genomic DNA as a template. Amplified products were analyzed for methylation using the PyroMark ID pyrosequencer with PyroGold Q96 reagents according to the manufacturer's instructions (Qiagen, UK). Data were analyzed using Pyro-MarkID software (Qiagen, UK) and assessed using the percentage of T and C incorporation at each CpG dinucleotide.

Results
To investigate the proposed association between deletions within PTPRD and metastatic cSCC, we developed a custom Illumina SNP microarray designed to cover the locus at high resolution. The mean age of patients with metastatic disease upon presentation of the primary tumor was 68.4 years (range: 54-91), and an approximately equal number of patients were immunosuppressed (n ¼ 10) and immunocompetent (n ¼ 9). By using this array, we were able to successfully analyze 73 of 74 FFPE cSCC tumors for deletion within PTPRD, indicating high compatibility for this assay with FFPE material. Four control FFPE tumors, for which matched fresh-frozen or cultured cell equivalents had been previously genotyped using the Affymetrix 250K SNP microarrays, 11 were included as controls for the detection of copy number change. One control sample displayed LOH across the entire locus, one had a heterozygous deletion and two had homozygous deletions. In all four cases, an identical pattern and location of copy number/deletion was detected in the FFPE samples, suggesting good reliability (Figs. 1a and  1b).
PTPRD is frequently deleted in primary-metastatic and aggressive cSCC In total, deletions within PTPRD were found in primary and/ or metastatic tumors from 37% (7/19) of patients with metastatic disease ( Table 1). Five of the deletions were homozygous, occurring on a background of LOH and two were heterozygous deletions with a normal copy number across the rest of the locus. The primary-metastatic series from a further seven patients showed LOH at the locus, with an overall LOH rate for this sample subset at 63% (12/19; Supporting Information Table 1). Of the PD cSCC included on the array, 22% (4/18) showed a deletion within PTPRD, of which two were homozygous on a background of LOH and two were heterozygous deletions with no LOH (Table 1). Three PD cSCC also showed LOH in the absence of deletion (Supporting Information Table 1). For statistical analysis, SNP microarray data of the PTPRD locus from the 60 cSCC analyzed by Purdie et al. 11 were combined with the series of tumors analyzed in our study to give a total of 96 tumors from 80 patients Table 2). This included one additional patient with metastatic disease. This analysis revealed a significant association between primary tumors that metastasized and deletion at the PTPRD locus (p ¼ 0.007). To exclude the possibility that patient effect may be skewing the data for the 13 patients from whom multiple primary tumors were analyzed, the test was repeated with these patients excluded. This still showed a significant association between deletion within PTPRD and metastasis (p ¼ 0.002). There was also a significant association between PD tumors and PTPRD deletion when compared to WD and MD tumors (p ¼ 0.01), which is consistent with the more aggressive nature of PD tumors.

Deletions of PTPRD in metastatic cSCC are clonal
Including the additional metastatic tumor from the series analyzed by Purdie et al., 11 six of eight (75%) patients with metastatic disease and deletion of PTPRD had an identical pattern of deletion throughout the individual tumor series. This included the primary tumor and subsequent recurrent and metastatic tumors (Fig. 1c). In the remaining two primary-metastatic series (PM10 and PM17), deletion of PTPRD was detected in the metastatic tumors only and not in the corresponding primary tumor that was analyzed.

PTPRD deletions were not detected in melanoma tissue samples
To investigate if deletions within PTPRD may also be associated with metastasis in other skin tumors, DNA was extracted from a series of 76 FFPE primary and metastatic melanoma samples and analyzed by the custom SNP microarray. This revealed no deletions within PTPRD in any of the melanoma samples, regardless of metastatic potential or Breslow thickness. There was also a lower rate of LOH than PTPRD deletions frequently target the 5 0 untranslated region PTPRD deletions have been reported in a wide range of different cancers. Figure 2 shows the location of deletions from the primary-metastatic cSCC series reported here as well as those cited in the literature from seven studies on lung cancer, neuroblastoma, glioblastoma multiforme, melanoma, cSCC and follicular lymphoma. 11,[13][14][15][16] In total, 15% (7/46) of deletions extend into the protein-coding region, with the remainder located exclusively in the 5 0 untranslated region (UTR). Furthermore, 60% (28/47) of the deletions include exon B7 of the 5 0 UTR, suggesting that this exon is a common target for deletion.

PTPRD mutations are common in cSCC
There was insufficient material from the FFPE cSCC tumor samples to sequence the entire 35 coding exons of the PTPRD gene. Therefore, sequencing was performed in 12 fresh-frozen microdissected cSCC tissue samples and six short-term passage cSCC cultures that we had previously analyzed for PTPRD genotype status. 11 Samples with a homozygous deletion of PTPRD were not included. The samples included three primary tumors from the metastatic series analyzed by the Illumina microarray, four PD, six MD and five WD tumors that had not metastasized. In addition, sequencing data from PM1 that we had previously analyzed were included. 12 In total, 37% (7/19) of tumors had a novel coding region mutation and all were found to be somatic upon sequencing matched blood DNA (Fig. 3, Supporting Information Fig. 1). This is the highest frequency of PTPRD mutation reported to date in any cancer type. 17,21,22 The majority of mutations (7/10) were missense changes spread throughout the PTPRD protein, including one in the active phosphatase subunit of the catalytic domain (Table 3). One further change was a stop mutation-W775X-predicted to truncate the protein in a fibronectin subunit upstream of the catalytic domain. The remaining two changes were either synonymous or in the 3 0 UTR. Of the seven tumors that showed a PTPRD mutation, four were from PD cSCC or primary tumors that had metastasized. The remainder were WD (2/7) or MD (1/7) cSCC that had not metastasized. Furthermore, the mutations in two of the seven tumors (SCC10 and SCC33) revealed biallelic disruption of PTPRD, combining LOH with mutation of the remaining allele. Two additional primary tumors that had metastasized also showed potential biallelic disruption with ''two-hits" within PTPRD; however, it was not determined if the two events were on separate alleles. PM1 had a coding sequence mutation combined with a heterozygous deletion in the 5 0 UTR that extended into the coding sequence and PM6 had two mutations.
Promoter methylation is not a common mechanism of PTPRD inactivation in cSCC PTPRD has two distinct 5 0 UTR isoforms that arise from different promoters. Pyrosequencing was used to interrogate part of each promoter region for methylation changes relative to normal skin. The longer (L) isoform has a canonical 698bp CpG island located at the transcription start site and has been previously shown to be methylated in breast, colorectal cancer and glioblastoma, whereas the shorter (S) isoform does not have a defined promoter region and has not been previously investigated. In the absence of a CpG island at the designated start site of the S isoform, a 1,223-bp CpG island for which EST evidence supports the generation of PTPRD transcripts was analyzed. Primers were designed for the L isoform promoter region to overlap with the region previously reported as methylated in other cancer types. A panel of 25 microdissected fresh-frozen tumors, eight normal skin samples and 15 cell lines were tested for methylation at both promoter regions. This included 8 of 12 fresh-frozen tumors and all cell lines that had been used for sequence analysis of PTPRD (Supporting Information Table 2). No methylation was identified in the normal skin samples or tumor tissue, indicating that this is not an important mechanism of disruption in this tumor type. Similarly, the cSCC cell lines did not show methylation. The only cell line to show positive methylation at both CpG islands was the HeLa cell line.

PTPRD is expressed at lower levels in high-grade tumors
The expression level of PTPRD was measured in 16 laser capture microdissected cSCC, which included seven WD tumors, seven MD or PD tumors and two lymph node metastases. Fold change differences were determined relative to the average level of PTPRD expression across six normal skin samples. Expression levels were twofold increased in WD cSCC when compared to normal skin, whereas MD and PD showed a decrease in the levels of mRNA (Fig. 4). Intriguingly, tumors with a deletion showed an increase in expression of PTPRD when compared to tumors of the same histological grade without a deletion.

Discussion
Taken together, these data show that PTPRD is disrupted in cutaneous SCC by multiple mechanisms and that deletions within PTPRD are associated with the clinically more aggressive tumors. This is the first study to specifically associate deletions of PTPRD with metastatic tumors and is consistent with previous reports showing reduced mRNA expression of PTPRD in higher grade neuroblastoma, breast and colorectal carcinomas. 20,30 Furthermore, a previous study of neuroblastoma  demonstrated a PTPRD deletion in a metastatic tumor of the bone marrow that was not present in the primary tumor from which it was derived, suggesting that the link with metastasis may also extend to additional cancer types. 16 In summary, deletions within PTPRD were identified in 37% (7/19) of patients with metastatic disease, and mutations were present in 37% (7/19) of tumors analyzed. Methylation of the promoter regions was not observed in cSCC tissue samples or cell lines. No deletions were identified in either primary or metastatic melanoma samples, suggesting that the association of PTPRD deletion and metastasis may be specific to cSCC, rather than skin cancers per se. This finding is in contrast to a study by Stark and Hayward 15 that found deletion of PTPRD in 9% of melanoma cell lines. The difference in the rate of deletion reported by the two studies could reflect the use of primary melanoma tissue in our study when compared to cell lines in the previous publication and could indicate either an acquired change in the cell lines or masking of the deletions in our study by stromal contamination of the tumor tissue. However, the latter is unlikely as all samples were crudely microdissected in an identical manner to the cSCC in which the deletions were readily identified. PTPRD belongs to a family of 21 transmembrane protein tyrosine phosphatase receptors (PTPRs) that are implicated in ligand-controlled protein dephosphorylation, cell-cell adhesion and cell-matrix adhesion. 31 Many different family members have been implicated as putative tumor suppressor genes, including PTPRG, PTPRK and PTPRJ, which are dis-rupted in nasopharyngeal carcinoma, lymphoma and pancreatic cancer, respectively. [32][33][34][35] Conversely, PTPRA has growthpromoting effects and is overexpressed in oral SCC, colorectal and gastric carcinomas, suggesting a diverse role for this phosphatase family in cancer. [36][37][38] The role of PTPRD has not been fully characterized, and determining its function in both normal skin and cSCC is essential to understand its part in cSCC progression. A recent publication has shown that PTPRD dephosphorylates the pStat3 protein in vitro, suggesting a potential mechanism through which PTPRD inactivation may contribute to tumorigenesis. 21 Upregulation of pStat3 has been identified in cSCC in comparison to normal skin (including a limited number of metastatic lesions), which is supportive of a role in cSCC tumorigenesis. 39,40 The effect of the deletion within PTPRD is currently unknown. As illustrated in this publication, only 15% of deletions described in the literature extend into the protein-coding region, with the remainder located in the 5 0 UTR of the gene. In addition, a study by Nair et al. 30 revealed aberrant splicing of exons within the 5 0 UTR of PTPRD in the absence of deletion. Exon B7 was one of the most frequently spliced exons, mimicking the effect of genomic deletions. This suggests that the 5 0 UTR is the target of multiple mechanisms of disruption in cancer and may play a key role in regulating the expression of PTPRD. Upstream noncoding regions are important regulators of protein expression affecting the localization, efficiency of translation and stability of mRNA. Long 5 0 UTRs (>200 bp) usually have a higher degree of secondary . qRT-PCR of PTPRD mRNA levels in cSCC compared to normal skin. PTPRD expression levels were averaged for each tumor, normalized to HPRT and fold change differences calculated with the DDCt method using a panel of six normal skin samples. The denotation '_del' indicates those samples with a known deletion of PTPRD. *Indicates samples with such low levels of PTPRD expression they were beyond the detection limits of the assay. structure and are frequently found in genes that encode tightly regulated proteins such as transcription factors, tumor suppressors and proto-oncogenes. 41 Deletion of exons within long 5 0 UTRs can decrease the complexity of mRNA and is typically associated with increased protein expression. The higher mRNA expression level seen in cSCC with a deletion suggests that the mRNA may be stabilized in these tumors, which would be predicted to increase protein expression. Unfortunately, the commercial antibodies that are currently available are not effective for visualization of PTPRD in vivo, so we were not able to continue these observations at the protein level.
Despite the unknown function of the 5 0 UTR deletions, the association with metastatic cSCC is of clinical relevance, as they could potentially be used as a genetic marker to identify risk of disease progression in high-risk patients. Currently, only clinicopathological criteria are available for assessing metastatic risk, and the only suggested biomarker for metastasis-the epidermal growth factor receptor-shows inconsistent protein expression that does not correlate with tumor grade or progression. 7,42,43 The metastatic rate for cSCC is between 0.1 and 9.9% and represents the main cause of death for patients suffering from NMSC. [4][5][6][7] For tumors with a depth >6 mm, the metastatic rate rises to 16% and high-risk patients such as those on immunosuppression are at an increased risk of developing aggressive cSCC with a corresponding higher rate of metastasis. 3,7,44 For these patients in particular, a biomarker to distinguish primary tumors with an increased risk of metastasis would be a highly desirable clinical tool, enabling more efficient targeted management such as sentinel lymph node surveillance. Given that the majority of the deletions found within individual metastatic series were clonal, a cytogenetic-based approach using probes within the most frequently deleted region would detect the majority of primary cSCC with deletion. Such cytogenetic-based methods are routinely used for diagnostic purposes in hematological and solid malignancies, where they represent an integral part of the clinical management of some cancers.
In conclusion, the data presented here show PTPRD to be an important gene that warrants further investigation to define its role in cSCC progression. The association between deletion and metastasis indicates its potential as a diagnostic biomarker, particularly for high-risk patients.