Pan-cancer Analysis And Construction Of Cancer Prognostic Models Based On Pan-cancer Molecular Biomarkers

Cancer,known as the "the king of the diseases",is a serious threat to human health.It is one of the leading cause of death globally.It can originate in various tissues and organs,but has a common feature�abnormal cell growth.During thousands of years of struggle against cancer,human exploration of cancer has never stopped.The rapid development of science and technology helps deepening our understanding of cancer.With the help of gene sequencing technology,we have already known that tumor cells are transformed from normal cells through a series of gene mutations.Cells lose the ability to regulate their own growth,resulting in abnormal proliferation and differentiation.Then tumor cells are no longer controlled by the body and ultimately destroy the normal organs.In recent decades,with the deepening understanding of cancer,various methods of cancer treatment continue to emerge,such as chemotherapy,radiotherapy,targeted therapy and immunotherapy.But up to date,tumors are still far from cured.In 2012,the Cancer Genomics Atlas launched the Pan-Cancer analysis project.Pan-cancer analysis investigates the similarities and differences across various cancer types using genomic,transcriptomic,proteomic and other omics data,which will benefit clinical diagnosis and treatment.Tumor heterogeneity is one of the major challenges facing tumor research.The heterogeneity occurs between tumor cells with differences in genome,transcriptome and proteome,resulting in the distinct molecular mechanisms underlying the occurrence and development of tumors.Thus,patients with the same tumor type may have quite different prognoses(i.e.survival time,recurrence and metastasis)and treatment responses.It has brought great challenges to the diagnosis and treatment of tumors.Hence accurate diagnostic methods and treatment strategies for different molecular subtypes can significantly improve the diagnosis accuracy and treatment effect of tumor patients.On the other hand,studies have shown that there are also similarities across tumors besides differences.For example,tumors of different tissues or organs may contain common molecular features,in which case a drug targeting the common feature can be used for precise treatment.In summary,pan-cancer analysis aims to explore the differences and similarities across tumors,which will deepen our understanding of tumors.It will help uncovering molecular mechanisms in the occurence and development of tumors and promote clinical precision diagnosis and treatment.Our work contains three parts of studies: pan-cancer analysis across tumor types;pan-cancer analysis across tumor individuals;construction of prognostic models based on pan-cancer molecular biomarkers.The first part is pan-cancer analysis across tumor types.We collected 144 gene microarray datasets containing 98 tumors and their subtypes from the GEO database,which contained a total of 2,362 gene expression profiles.Gene expression signature is a set of characteristic genes that can reflect the variation of gene expression levels between tumor and normal tissues.We used the hierarchical majority-voting scheme to aggregate the gene expression profiles in each dataset to form the corresponding gene expression signatures.Then we used the gene enrichment analysis method to measure the similarities between different datasets.Based on the tumor similarity matrix,we used hierarchical clustering to explore the similarities and differences between tumor types.The results of the analysis showed that most of the clusters contained datasets of the same organ,such as brain cancer,renal cancer,lymphoma,colorectal cancer and ovarian cancer.We also found a pan-cancer cluster containing six tumors types,i.e.breast cancer,adrenocortical cancer,lung cancer,bladder cancer,colorectal cancer and pleural mesothelioma.The second part of the tumor is pan-cancer analysis across tumor individuals.We collected RNA-seq data of 13 tumor types from TCGA and selected 633 samples containing both tumor tissues and normal tissues.Based on the transcriptional response profiles derived from tumor and adjacent normal tissues,633 samples were divided into 10 clusters by consensus clustering.There were significant differences in the gene expression,enrichment pathways and immune levels among the 10 clusters.One of the clusters included bladder urothelial carcinoma,breast cancer,head and neck squamous cell carcinoma and uterine corpus endometrial carcinoma.Head and neck squamous cell carcinoma in the consensus clustering results was divided into two distinct subtypes.One subtype was enriched in cell cycle-related pathways while the other subtype was enriched in cell adhesion-related pathways.They might respond to different treatment strategies.In addition,we found 92 genes that are consistently upregulated in tumor tissues.They might be involved in the regulation of cell cycle and apoptosis.We selected FAM64 A and TROAP genes for experimental validation.RNA interference experiments showed that knockdown of either FAM64 A or TROAP in MDA-MB-231 cells could significantly inhibit the growth of tumor cells.The third part is construction of prognostic models based on pan-cancer molecular biomarkers.We selected 75 prognostic molecular biomarkers from multiple tumors through literature research.These biomarkers almost covered all the signaling pathways involved in the development of tumors,which can be called pan-cancer molecular biomarkers.Immunohistochemistry staining was performed to quantitatively detect the expression levels of 75 biomarkers in non-small cell lung cancer.Survival analysis showed that 41 and 22 biomarkers were associated with clinical prognosis in lung adenocarcinoma and lung squamous cell carcinoma,respectively.But the number of overlaps between them was quite small.Then we selected prognostic biomarker signatures of lung adenocarcinoma and lung squamous cell carcinoma by random forest method,and constructed prognostic models for them by SVM method.The results showed that the prediction performance of the two models was good.In summary,we carried out pan-cancer analysis from different perspectives in this work.We used different methods to investigate the differences and similarities between tumors.It will be of value in clinical diagnosis and treatment of cancer.There are two innovations in this work:1.Normal tissues were involved in pan-cancer analysis for the first time.We found novel perspectives based on the similarities and differences between tumors.Our study suggests that adjacent normal tissues,as the good controls in tumor research,can provide additional information and play a role in pan-cancer analysis;2.We used a variety of molecular biomarkers from different tumor types in non-small cell lung cancer and obtained two prognostic models with good predictive performance.It is suggested that the pan-cancer molecular biomarkers have the ability of predicting survival in different tumor types,which may be useful for the clinical prognosis of tumor.