BreastMark: Transcriptomic Datasets and Breast Cancer Outcome
Introduction: Breast cancer is a complex heterogeneous disease for which a substantial resource of transcriptomic data is available. Gene expression data have facilitated the division of breast cancer into, at least, five molecular subtypes, namely luminal A, luminal B, HER2, normal-like and basal. Once identified, breast cancer subtypes can inform clinical decisions surrounding patient treatment and prognosis. Indeed, it is important to identify patients at risk of developing aggressive disease so as to tailor the level of clinical intervention.
Methods: We have developed a user-friendly, web-based system to allow the evaluation of genes/microRNAs (miRNAs) that are significantly associated with survival in breast cancer and its molecular subtypes. The algorithm combines gene expression data from multiple microarray experiments which frequently also contain miRNA expression information, and detailed clinical data to correlate outcome with gene/miRNA expression levels. This algorithm integrates gene expression and survival data from 26 datasets on 12 different microarray platforms corresponding to approximately 17,000 genes in up to 4,738 samples. In addition, the prognostic potential of 341 miRNAs can be analysed.
Results: We demonstrated the robustness of our approach in comparison to two commercially available prognostic tests, oncotype DX and MammaPrint. Our algorithm complements these prognostic tests and is consistent with their findings. In addition, BreastMark can act as a powerful reductionist approach to these more complex gene signatures, eliminating superfluous genes, potentially reducing the cost and complexity of these multi-index assays. Known miRNA prognostic markers, mir-205 and mir-93, were used to confirm the prognostic value of this tool in a miRNA setting. We also applied the algorithm to examine expression of 58 receptor tyrosine kinases in the basal-like subtype, identifying six receptor tyrosine kinases associated with poor disease-free survival and/or overall survival (EPHA5, FGFR1, FGFR3, VEGFR1, PDGFRβ, and TIE1). A web application for using this algorithm is currently available.
Conclusions:BreastMark is a powerful tool for examining putative gene/miRNA prognostic markers in breast cancer. The value of this tool will be in the preliminary assessment of putative biomarkers in breast cancer. It will be of particular use to research groups with limited bioinformatics facilities.
Breast cancer is a complex heterogeneous disease which has traditionally been subclassified depending, amongst other factors, on the expression of different receptor proteins, such as estrogen receptor (ER), progesterone receptor (PR), and human epidermal growth factor receptor 2 (HER2). These 'biomarkers' allow us to tailor the level of clinical intervention. While ER-positive the second positive should be deleted tumours receive hormone therapies and HER2-positive cancers receive targeted therapies such as trastuzumab and lapatinib, 'triple negative' cancers lacking these markers currently have no targeted therapies and cause a disproportionate number of breast cancer deaths. In addition to the traditional classifications using these biomarkers, in recent years, whole genome DNA microarrays have been utilised to further classify this disease, initially into five molecular subtypes based on gene expression profiles, namely luminal A and luminal B (ER-positive tumours), HER2 (HER2-positive tumours), basal and normal-like tumours and subsequently into at least ten further molecular subtypes using both copy number and gene expression data.
It is crucially important to identify which breast cancer patients are at risk of developing a more aggressive phenotype so as to tailor the level of clinical intervention. Prognostic biomarkers, such as ER and HER2, can be used to assess the inherent likelihood of a patient exhibiting a particular outcome. However, within the subtypes defined by these classical markers, there is a wide spectrum of survival requiring the identification of additional novel prognostic markers. Also, the triple negative subtype has no such prognostic biomarkers currently in clinical use.
There is a great deal of transcriptomics data currently available to facilitate the identification of novel molecular biomarkers associated with breast cancer and its subtypes. Huge studies such as the 2,000 breast tumour profiles by Curtis et al. greatly aid in our understanding of breast cancer and facilitate the identification of novel intrinsic subtypes. The diverse nature of these datasets and the variability of the different microarray platforms themselves can affect the statistical power of such studies. Moreover, it is necessary to test the prognostic ability of markers in diverse datasets to avoid dataset-specific affects.
It is clear that the selection of markers could benefit greatly from the integration of datasets from multiple studies to increase confidence in the selected markers. To this end, we have developed an easy-to-use interface for our algorithm which allows identification of subsets of genes that are associated with disease progression in breast cancer or its subtypes, that is, a set of putative prognostic markers. This algorithm integrates gene expression data from DNA microarray studies and corresponding clinical data (hormone status, survival time, tumour grade, patient age and so on). In particular, it allows investigation of prognostic markers in the context of disease-free survival (DFS), distant disease-free survival (DDFS) and overall survival (OS).
Over the last decade, our understanding of the function that small non-coding RNAs known as microRNAs (miRNAs) play in an array of fundamental biological processes in both plants and animals has increased dramatically. These short endogenous non-coding RNAs act primarily by negatively regulating the expression of target mRNAs through translational inhibition and/or mRNA degradation. The complexity of post-transcriptional control of gene expression by miRNAs remains a significant challenge. Indeed, miRNAs have the potential to alter entire pathways due to their ability to target multiple genes simultaneously. The association of miRNAs with breast cancer has been well established. In fact, miRNAs have been identified as prognostic markers in breast cancer and associated with breast tumours defined by their HER2 or ER/PR status.
Approximately 50% of known human miRNAs are intronic (miRBase release 18, November 2011). Of these, 341 or roughly one third of human miRNA host genes are hybridized by probes on the U133plus2 Affymetrix gene chip. A number of studies have reported that many intronic miRNAs show significantly correlated expression profiles with their host genes. Estimates of the number of miRNAs whose expression profiles are significantly correlated with their host gene are as high as 70%. The expression of these miRNAs can, in some instances, be inferred from the expression of their host genes and can, therefore, be evaluated as putative prognostic markers in breast cancer and its subtypes using gene expression data.
We evaluated our approach using two commercially available gene expression-based prognostic tests in breast cancer, namely oncotype DX and MammaPrint. We also applied the algorithm to examine the expression of 58 receptor tyrosine kinases (RTKs) in the basal-like subtype of breast cancer. Using the 21 genes from oncotype DX and the 70-gene MammaPrint signature, we demonstrated the robustness of our approach and confirmed the prognostic value of these signatures. In the case of oncotype DX, we showed that the predictive strength of this test is centred on the five proliferation genes within the 21 gene set. We also identified six RTKs associated with poor prognosis in the basal breast cancer subtype. The feasibility of using miRNA host gene expression as a surrogate for miRNA levels was tested using known miRNA prognostic markers, mir-93 and mir-205. Although these markers were only identified in small patient cohorts, BreastMark was able to confirm the robustness of these prognostic markers across a far larger and diverse patient dataset. A web application for using this algorithm is currently available.
Abstract and Introduction
Abstract
Introduction: Breast cancer is a complex heterogeneous disease for which a substantial resource of transcriptomic data is available. Gene expression data have facilitated the division of breast cancer into, at least, five molecular subtypes, namely luminal A, luminal B, HER2, normal-like and basal. Once identified, breast cancer subtypes can inform clinical decisions surrounding patient treatment and prognosis. Indeed, it is important to identify patients at risk of developing aggressive disease so as to tailor the level of clinical intervention.
Methods: We have developed a user-friendly, web-based system to allow the evaluation of genes/microRNAs (miRNAs) that are significantly associated with survival in breast cancer and its molecular subtypes. The algorithm combines gene expression data from multiple microarray experiments which frequently also contain miRNA expression information, and detailed clinical data to correlate outcome with gene/miRNA expression levels. This algorithm integrates gene expression and survival data from 26 datasets on 12 different microarray platforms corresponding to approximately 17,000 genes in up to 4,738 samples. In addition, the prognostic potential of 341 miRNAs can be analysed.
Results: We demonstrated the robustness of our approach in comparison to two commercially available prognostic tests, oncotype DX and MammaPrint. Our algorithm complements these prognostic tests and is consistent with their findings. In addition, BreastMark can act as a powerful reductionist approach to these more complex gene signatures, eliminating superfluous genes, potentially reducing the cost and complexity of these multi-index assays. Known miRNA prognostic markers, mir-205 and mir-93, were used to confirm the prognostic value of this tool in a miRNA setting. We also applied the algorithm to examine expression of 58 receptor tyrosine kinases in the basal-like subtype, identifying six receptor tyrosine kinases associated with poor disease-free survival and/or overall survival (EPHA5, FGFR1, FGFR3, VEGFR1, PDGFRβ, and TIE1). A web application for using this algorithm is currently available.
Conclusions:BreastMark is a powerful tool for examining putative gene/miRNA prognostic markers in breast cancer. The value of this tool will be in the preliminary assessment of putative biomarkers in breast cancer. It will be of particular use to research groups with limited bioinformatics facilities.
Introduction
Breast cancer is a complex heterogeneous disease which has traditionally been subclassified depending, amongst other factors, on the expression of different receptor proteins, such as estrogen receptor (ER), progesterone receptor (PR), and human epidermal growth factor receptor 2 (HER2). These 'biomarkers' allow us to tailor the level of clinical intervention. While ER-positive the second positive should be deleted tumours receive hormone therapies and HER2-positive cancers receive targeted therapies such as trastuzumab and lapatinib, 'triple negative' cancers lacking these markers currently have no targeted therapies and cause a disproportionate number of breast cancer deaths. In addition to the traditional classifications using these biomarkers, in recent years, whole genome DNA microarrays have been utilised to further classify this disease, initially into five molecular subtypes based on gene expression profiles, namely luminal A and luminal B (ER-positive tumours), HER2 (HER2-positive tumours), basal and normal-like tumours and subsequently into at least ten further molecular subtypes using both copy number and gene expression data.
It is crucially important to identify which breast cancer patients are at risk of developing a more aggressive phenotype so as to tailor the level of clinical intervention. Prognostic biomarkers, such as ER and HER2, can be used to assess the inherent likelihood of a patient exhibiting a particular outcome. However, within the subtypes defined by these classical markers, there is a wide spectrum of survival requiring the identification of additional novel prognostic markers. Also, the triple negative subtype has no such prognostic biomarkers currently in clinical use.
There is a great deal of transcriptomics data currently available to facilitate the identification of novel molecular biomarkers associated with breast cancer and its subtypes. Huge studies such as the 2,000 breast tumour profiles by Curtis et al. greatly aid in our understanding of breast cancer and facilitate the identification of novel intrinsic subtypes. The diverse nature of these datasets and the variability of the different microarray platforms themselves can affect the statistical power of such studies. Moreover, it is necessary to test the prognostic ability of markers in diverse datasets to avoid dataset-specific affects.
It is clear that the selection of markers could benefit greatly from the integration of datasets from multiple studies to increase confidence in the selected markers. To this end, we have developed an easy-to-use interface for our algorithm which allows identification of subsets of genes that are associated with disease progression in breast cancer or its subtypes, that is, a set of putative prognostic markers. This algorithm integrates gene expression data from DNA microarray studies and corresponding clinical data (hormone status, survival time, tumour grade, patient age and so on). In particular, it allows investigation of prognostic markers in the context of disease-free survival (DFS), distant disease-free survival (DDFS) and overall survival (OS).
Over the last decade, our understanding of the function that small non-coding RNAs known as microRNAs (miRNAs) play in an array of fundamental biological processes in both plants and animals has increased dramatically. These short endogenous non-coding RNAs act primarily by negatively regulating the expression of target mRNAs through translational inhibition and/or mRNA degradation. The complexity of post-transcriptional control of gene expression by miRNAs remains a significant challenge. Indeed, miRNAs have the potential to alter entire pathways due to their ability to target multiple genes simultaneously. The association of miRNAs with breast cancer has been well established. In fact, miRNAs have been identified as prognostic markers in breast cancer and associated with breast tumours defined by their HER2 or ER/PR status.
Approximately 50% of known human miRNAs are intronic (miRBase release 18, November 2011). Of these, 341 or roughly one third of human miRNA host genes are hybridized by probes on the U133plus2 Affymetrix gene chip. A number of studies have reported that many intronic miRNAs show significantly correlated expression profiles with their host genes. Estimates of the number of miRNAs whose expression profiles are significantly correlated with their host gene are as high as 70%. The expression of these miRNAs can, in some instances, be inferred from the expression of their host genes and can, therefore, be evaluated as putative prognostic markers in breast cancer and its subtypes using gene expression data.
We evaluated our approach using two commercially available gene expression-based prognostic tests in breast cancer, namely oncotype DX and MammaPrint. We also applied the algorithm to examine the expression of 58 receptor tyrosine kinases (RTKs) in the basal-like subtype of breast cancer. Using the 21 genes from oncotype DX and the 70-gene MammaPrint signature, we demonstrated the robustness of our approach and confirmed the prognostic value of these signatures. In the case of oncotype DX, we showed that the predictive strength of this test is centred on the five proliferation genes within the 21 gene set. We also identified six RTKs associated with poor prognosis in the basal breast cancer subtype. The feasibility of using miRNA host gene expression as a surrogate for miRNA levels was tested using known miRNA prognostic markers, mir-93 and mir-205. Although these markers were only identified in small patient cohorts, BreastMark was able to confirm the robustness of these prognostic markers across a far larger and diverse patient dataset. A web application for using this algorithm is currently available.
Source...