Factors within skin cancer that contribute to metastasis

Skin cancer is the most frequent cancer worldwide and accounts for 1 in every 3 cancers diagnosed. Skin cancer comprises of melanoma, arising from the melanocytes of the skin and keratinocyte carcinomas which arise in the keratinocytes and include cutaneous squamous cell carcinoma (cSCC) and basal cell carcinoma. cSCC predominantly affects the older generation and is one of the most common types of cancer capable of metastasising, with 5 year survival rates reported as <30%. Although less common, melanoma can affect all ages and has one of the highest rates of metastasis of any cancer, with 5 year survival rates as low as 23%, depending on whether or not distant metastasis has occurred. There are currently very few prognostic markers capable of predicting metastasis in these skin cancers. Presently in the UK, melanoma is graded according to the American Joint Committee on Cancer Guidelines (AJCC) whereas cSCCs are categorised as high or low risk according to the British Association of Dermatologists’ guidelines (BAD). Other staging systems have been proposed but most of them also rely on histological features such as differentiation, diameter, depth, site and/or Clark’s level, amongst others. This study aimed to identify factors within cSCC and melanoma which contribute to metastasis using a mass spectrometry based proteomics approach. A method to extract protein from formalin fixed paraffin embedded (FFPE) samples was developed and optimised. Proteins were extracted from 24 FFPE surgically excised primary cSCC (P-NM) and melanoma (Pmel-NM) tumours which had not metastasised at 5 years post-operatively and from 24 FFPE surgically excised primary metastatic cSCC (P-M) and melanoma (PmelM) tumours which had metastasised. A total of 144 and 31 significantly differentially expressed proteins between metastatic and non-metastatic samples were identified in the cSCC and melanoma groups respectively. KEGG, gene ontology, weighted gene co-expression network analysis (WGCNA) and ingenuity pathway analysis (IPA) highlighted several key pathways likely to be involved in development of metastasis in cSCC and melanoma. Multiple reaction monitoring (MRM) of two proteins, ANXA5 and DDOST, verified the original differences in levels of these proteins in cSCC and also validated these findings in an independent sample cohort. Additionally, MRM analysis and machine learning revealed that the combination of ANXA5 and DDOST levels could correctly predict metastasis better than any guideline in current clinical use, with an AUC of 0.929, sensitivity and specificity of 88.24% and 94.12% respectively. However, MRM was technically challenging in the melanoma group and was not able to verify the original melanoma mass spectroscopy results. Machine learning and modelling of histological characteristics from cSCC samples was subsequently undertaken to see whether it was possible to improve on current prediction of prognosis with these readily available parameters. Surprisingly, this produced a prediction model with an Area Under the Curve (AUC) of 0.997 and a sensitivity and specificity of 94.1% and 100% respectively. Despite this model not requiring any additional work over and above that which is already currently reported histologically when cSCCs are routinely excised in the UK, it was better than the aforementioned ANXA5 and DDOST model and moreover, than any guideline in clinical use at the present time. Moreover, this model has the potential to be integrated into a clinical setting with relative ease and speed. This study has identified a number of factors, including key pathways that likely contribute to metastasis in cSCC and melanoma. In addition, the combination of proteomics, machine learning and mathematical modelling has identified key prognostic indicators in cSCC and has demonstrated that this approach may have potential to do likewise in many other cancer types