Journal
MOLECULAR CANCER
Volume 16, Issue -, Pages -Publisher
BIOMED CENTRAL LTD
DOI: 10.1186/s12943-017-0673-0
Keywords
Breast cancer; Cell lines; Hypoxia; Co-expression; Co-regulation; Genome-wide analysis; Clustering; Bioinformatics; UNCLES
Categories
Funding
- Brunel University London
- Cancer Research UK, EU framework 7
- Oxford NIHR Biomedical Research Centre
- Breast Cancer Research Foundation
- National Science Foundation of China [61520106006]
- National Science Foundation of Shanghai [16JC1401300]
- MRC [G0300648] Funding Source: UKRI
- Cancer Research UK [11359, 18974, 23969, 16466] Funding Source: researchfish
- Medical Research Council [G0300648] Funding Source: researchfish
- National Institute for Health Research [NF-SI-0611-10163] Funding Source: researchfish
Ask authors/readers for more resources
Background: Hypoxia is a characteristic of breast tumours indicating poor prognosis. Based on the assumption that those genes which are up-regulated under hypoxia in cell-lines are expected to be predictors of poor prognosis in clinical data, many signatures of poor prognosis were identified. However, it was observed that cell line data do not always concur with clinical data, and therefore conclusions from cell line analysis should be considered with caution. As many transcriptomic cell-line datasets from hypoxia related contexts are available, integrative approaches which investigate these datasets collectively, while not ignoring clinical data, are required. Results: We analyse sixteen heterogeneous breast cancer cell-line transcriptomic datasets in hypoxia-related conditions collectively by employing the unique capabilities of the method, UNCLES, which integrates clustering results from multiple datasets and can address questions that cannot be answered by existing methods. This has been demonstrated by comparison with the state-of-the-art iCluster method. From this collection of genome-wide datasets include 15,588 genes, UNCLES identified a relatively high number of genes (> 1000 overall) which are consistently co-regulated over all of the datasets, and some of which are still poorly understood and represent new potential HIF targets, such as RSBN1 and KIAA0195. Two main, anti-correlated, clusters were identified; the first is enriched with MYC targets participating in growth and proliferation, while the other is enriched with HIF targets directly participating in the hypoxia response. Surprisingly, in six clinical datasets, some sub-clusters of growth genes are found consistently positively correlated with hypoxia response genes, unlike the observation in cell lines. Moreover, the ability to predict bad prognosis by a combined signature of one sub-cluster of growth genes and one sub-cluster of hypoxia-induced genes appears to be comparable and perhaps greater than that of known hypoxia signatures. Conclusions: We present a clustering approach suitable to integrate data from diverse experimental set-ups. Its application to breast cancer cell line datasets reveals new hypoxia-regulated signatures of genes which behave differently when in vitro (cell-line) data is compared with in vivo (clinical) data, and are of a prognostic value comparable or exceeding the state-of-the-art hypoxia signatures.
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available