Multi-Omics Analysis Reveals the Pan-Cancer Landscape of Bone Morphogenetic Proteins
Total Page:16
File Type:pdf, Size:1020Kb
DATABASE ANALYSIS e-ISSN 1643-3750 © Med Sci Monit, 2020; 26: e920943 DOI: 10.12659/MSM.920943 Received: 2019.10.24 Accepted: 2020.01.27 Multi-Omics Analysis Reveals the Pan-Cancer Available online: 2020.02.12 Published: 2020.04.05 Landscape of Bone Morphogenetic Proteins Authors’ Contribution: AB Wen-Li Luo Department of Orthopedics, Ningbo Hangzhou Bay Hospital, Ningbo, Zhejiang, Study Design A ABCDEFG Ming-Xing Luo P.R. China Data Collection B Statistical Analysis C ABCD Rong-Zhen He Data Interpretation D ACD Lv-Fang Ying Manuscript Preparation E A Jian Luo Literature Search F Funds Collection G Corresponding Author: Ming-Xing Luo, e-mail: [email protected], [email protected] Source of support: Departmental sources Background: Bone morphogenetic proteins (BMPs) are widely involved in cancer development. However, a wealth of con- flicting data raises the question of whether BMPs serve as oncogenes or as cancer suppressors. Material/Methods: By integrating multi-omics data across cancers, we comprehensively analyzed the genomic and pharmacoge- nomic landscape of BMP genes across cancers. Results: Surprisingly, our data indicate that BMPs are globally downregulated in cancers. Further genetics and epi- genetics analyses show that this abnormal expression is driven by copy number variations, especially hetero- zygous amplification. We next assessed the BMP-associated pathways and demonstrated that they suppress cell cycle and estrogen hormone pathways. Bone morphogenetic protein interacts with 58 compounds, and their dysfunction can induce drug sensitivity. Conclusions: Our results define the landscape of the BMP family at a systems level and open potential therapeutic oppor- tunities for cancer patients. MeSH Keywords: Bone Morphogenetic Proteins • Epigenomics • Gene Expression Profiling Full-text PDF: https://www.medscimonit.com/abstract/index/idArt/920943 1946 — 9 17 Indexed in: [Current Contents/Clinical Medicine] [SCI Expanded] [ISI Alerting System] This work is licensed under Creative Common Attribution- [ISI Journals Master List] [Index Medicus/MEDLINE] [EMBASE/Excerpta Medica] NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0) e920943-1 [Chemical Abstracts/CAS] Luo W.-L. et al.: DATABASE ANALYSIS The landscape of bone morphogenetic proteins in cancers © Med Sci Monit, 2020; 26: e920943 Background Survival analysis and cancer subtype analysis Bone morphogenetic proteins (BMPs) were first identified as We used sample barcodes to match gene expression and clin- proteins that are responsible for bone formation. They were ical survival data. We used the median RSEM value to divide first reported to induce the formation of cartilage in vivo in samples into high and low gene expression groups. Then, we 1988 [1] and were believed to be members of the transform- used the R package survival to assess to the survival time and ing growth factor b (TGF-b) family [1]. BMPs are highly con- survival status of the 2 groups. We constructed a Cox ratio-haz- served and have existed for over 700 million years [2], which ard model for each gene and plotted the Kaplan-Meier curve indicates the importance of BMPs in vertebrate physiology. using the log-rank test for each gene. We performed ANOVA They were later shown to regulate a broad spectrum of bio- to find subtype-specific genes. logical processes such as cancer development [3]. The BMP family consists of over 10 components [4], some of which are Single-nucleotide variation analysis currently in clinical use. Perturbations of BMPs gives rise to a wide range of clinical traits, including tumorigenesis. We obtained single-nucleotide variation data from The Cancer Genome Atlas, and we used maftools (https://bioconductor.org/ In the context of cancer, a few studies have shown that BMPs packages/release/bioc/html/maftools.html) to analyze the SNV have widely altered expression in tumor tissues [3]. For exam- data. SNV percentage refers to the number of mutated sam- ple, BMP2 is downregulated in prostate cancer samples and ples divided by the number of cancer samples. is correlated with cancer progression [5]. In contrast, BMP2 is significantly upregulated in head and neck squamous cell car- Copy number variation analysis cinoma [6]. BMP4 was reported by various research groups to have a dual role, acting as an oncogene or a tumor suppres- CNVs are divided into 2 subtypes – heterozygous CNV and ho- sor [7,8]. A wealth of conflicting studies indicated that the mozygous CNV. These 2 subtypes represent only 1 chromo- same BMP ligand can act differently depending on the cancer some or 2 chromosomes simultaneously with CNV. Based on type, and it appears that multiple members in the BMP family the percentage statistics of CNV subtypes, GISTIC (http://soft- have distinct functions [3,9]. These seemingly conflicting re- ware.broadinstitute.org/cancer/software/genepattern/modules/ sults raise the critical question of how they are involved and docs/GISTIC_2.0) was used to process CNV data, and the cal- how they influence cancer progression. There is a pressing culations used original CNV data and RNA-seq data. need to define the role of the BMP family across cancer types. Methylation analysis Here, by using high-throughput sequencing data from a large cohort of patients, we explored the genomic and pharmacoge- We retrieved methylation data from the NCI Genomic Data nomic interactions of BMP genes across cancers to assess how Commons (https://gdc.cancer.gov/). Only cancers paired with BMP family genes are upregulated or downregulated and to normal data were included for analysis. The mRNA expression determine the cancer hallmark signaling pathways they are and methylation data were merged according to their TCGA associated with. Our work helps define the landscape of the barcode. We used the methylation of the promoter region of BMP family at the systems level and provides potential ther- matched genes. We examined the relationship between gene apeutic opportunities for cancer patients. expression and methylation level according to the correla- tion coefficient of humans. Genes with FDR less than 0.05 were retained. Material and Methods Pathway activity quantification Gene expression analysis RPPA data from The Cancer Proteome Atlas (TCPA) were used We limited our analysis to cancer types with over 10 pairs of to calculate the scores of 10 cancer-related pathways. Reversed matched tumors and normal samples. We conducted paired dif- Phase Protein Array (RPPA) is a technology with a process sim- ferential gene expression analysis of tumors and their matched ilar to Western blot analysis. The RPPA data were centered normal samples. The fold change shows the mean gene ex- and the relative protein levels were obtained by normalizing pression of tumors by dividing the mean gene expression of the standard deviations for all samples of each component. normal samples. P values were adjusted by FDR. Genes with fold change (FC >2) and significance (FDR >0.05) were used for further analysis. Indexed in: [Current Contents/Clinical Medicine] [SCI Expanded] [ISI Alerting System] This work is licensed under Creative Common Attribution- [ISI Journals Master List] [Index Medicus/MEDLINE] [EMBASE/Excerpta Medica] NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0) e920943-2 [Chemical Abstracts/CAS] Luo W.-L. et al.: The landscape of bone morphogenetic proteins in cancers DATABASE ANALYSIS © Med Sci Monit, 2020; 26: e920943 A BMP8A log2 FC BMP1 ≥3 1.5 BMP8B 0 GDF11 –1.5 BMP4 ≤–3 BMP15 BMP10 FDR GDF2 0.05 10–5 BMP2 10–10 BMP7 <10–15 BMP3 GDF7 BMP6 BMP5 A CA AD AD LIHC KIRC KIRP KICH BL ST THC BRCA CO PRAD B C Heterozygous apmlicationHeterozygous deletion Hete CNV% Heterozygous apmlicationHeterozygous deletion Homo CNV% GDF7 5 GDF7 5 10 10 GDF2 20 GDF2 20 40 40 GDF11 60 GDF11 60 BMP8A 100 BMP8A 100 BMP7 BMP7 BMP6 BMP6 BMP5 BMP5 BMP4 BMP4 BMP3 BMP3 BMP2 BMP2 BMP15 BMP15 BMP10 BMP10 BMP1 BMP1 A A A A CA CA AD AD AD AD CA CA AD AD AD AD SCNA Type LIHC LIHC KIRC KIRC KIRP KIRP KICH KICH LIHC LIHC KIRC KIRC KIRP KIRP KICH KICH BL BL ST ST UCEC UCEC BL BL THC THC ST ST BRCA BRCA UCEC UCEC CO CO PRAD PRAD THC THC BRCA BRCA CO CO PRAD PRAD Deletion Amplication Figure 1. Bone morphogenetic protein family genes are commonly dysregulated in human cancers. (A) The CDK gene is downregulated in a wide range of cancers. Copy number variation of the CDK gene includes (B) heterozygous amplification and (C) homozygous amplification. Drug response analysis with a family error rate of less than 0.025 per tail. The Pearson correlation coefficient of the labeled drug target pair was com- To analyze the correlation between gene expression and drug pared to the same number of correlation coefficients produced sensitivity, we downloaded the region under the drug dose- by random sampling correlation. response curve (AUC) values and the gene expression profiles of all cancer cell lines. Fisher’s Z transform was used to nor- malize the transcription level to the Pearson correlation coeffi- cient of AUCs, with a Bonferroni-corrected 2-tailed distribution Indexed in: [Current Contents/Clinical Medicine] [SCI Expanded] [ISI Alerting System] This work is licensed under Creative Common Attribution- [ISI Journals Master List] [Index Medicus/MEDLINE] [EMBASE/Excerpta Medica] NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0) e920943-3 [Chemical Abstracts/CAS] Luo W.-L. et al.: DATABASE ANALYSIS The landscape of bone morphogenetic proteins in cancers © Med Sci Monit, 2020; 26: e920943 A B BMP7 BMP2 BMP8B GDF2 BMP3 BMP15 GDF7 BMP5 BMP2 BMP1 BMP4 GDF11 GDF11 Symbol Symbol BMP3 BMP8A BMP8A BMP1 BMP6 GDF7 GDF2 BMP6 BMP5 BMP8B BMP15 BMP4 BMP10 BMP7 A A A A CA CA AD AD AD LIHC LIHC KIRC KIRC KIRP KIRP KICH BL BL ST UCEC THC THC BRC BRC CO CO PRAD PRAD Cancer types Cancer types Methylation Di (T–N) –Log10 (FDR) 10 20 30 Spearman correlation coecient –Log10 (P.