Limits...
Coexpression analysis of large cancer datasets provides insight into the cellular phenotypes of the tumour microenvironment.

Doig TN, Hume DA, Theocharidis T, Goodlad JR, Gregory CD, Freeman TC - BMC Genomics (2013)

Bottom Line: As a result genes expressed in specific cell types, or associated with certain biological processes are detected at widely variable levels across samples in transcriptomic analyses.Using a novel 3D network-based approach we have analysed six large human cancer microarray datasets derived from more than 1,000 individuals.Based upon this analysis, and without needing to isolate the individual cells, we have defined a broad spectrum of cell-type and pathway-specific gene signatures present in cancer expression data which were also found to be largely conserved in a number of independent datasets.

View Article: PubMed Central - HTML - PubMed

Affiliation: Centre for Inflammation Research, University of Edinburgh, The Queen’s Medical Research Institute, Edinburgh, UK.

ABSTRACT

Background: Biopsies taken from individual tumours exhibit extensive differences in their cellular composition due to the inherent heterogeneity of cancers and vagaries of sample collection. As a result genes expressed in specific cell types, or associated with certain biological processes are detected at widely variable levels across samples in transcriptomic analyses. This heterogeneity also means that the level of expression of genes expressed specifically in a given cell type or process, will vary in line with the number of those cells within samples or activity of the pathway, and will therefore be correlated in their expression.

Results: Using a novel 3D network-based approach we have analysed six large human cancer microarray datasets derived from more than 1,000 individuals. Based upon this analysis, and without needing to isolate the individual cells, we have defined a broad spectrum of cell-type and pathway-specific gene signatures present in cancer expression data which were also found to be largely conserved in a number of independent datasets.

Conclusions: The conserved signature of the tumour-associated macrophage is shown to be largely-independent of tumour cell type. All stromal cell signatures have some degree of correlation with each other, since they must all be inversely correlated with the tumour component. However, viewed in the context of established tumours, the interactions between stromal components appear to be multifactorial given the level of one component e.g. vasculature, does not correlate tightly with another, such as the macrophage.

Show MeSH

Related in: MedlinePlus

Conservation of transcriptional signatures in graph derived from skin cancer dataset (Pearson correlation threshold r = 0.80). In order to provide a clear view of transcripts/clusters within skin cancer graph it has been simplified. The network shown here has been constructed with a central framework of edges derived from the relationships between clusters and nodes representing the transcripts in each cluster joined to a central node representing the cluster with the graph laid out in 2D. Only clusters comprising more than 8 probesets were included. a) Colours represent different clusters in the skin cancer data, and b) overlay of clusters from the merged cancer (r = 0.6) graph displayed using larger nodes. Many of the housekeeping clusters (1–5, 11) can be seen to be conserved, as is a proportion of the cell cycle (6), macrophage (7), T-cell (8), ECM (9), interferon response (12), plasma cell (14), MHC class 1 (19), histones (20) and Affymetrix control (23) clusters. However it can also be seen that many of the skin cancer clusters are not represented in the merged cancer profile set, these transcriptional signatures being unique to skin cancers.
© Copyright Policy - open-access
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC3721986&req=5

Figure 6: Conservation of transcriptional signatures in graph derived from skin cancer dataset (Pearson correlation threshold r = 0.80). In order to provide a clear view of transcripts/clusters within skin cancer graph it has been simplified. The network shown here has been constructed with a central framework of edges derived from the relationships between clusters and nodes representing the transcripts in each cluster joined to a central node representing the cluster with the graph laid out in 2D. Only clusters comprising more than 8 probesets were included. a) Colours represent different clusters in the skin cancer data, and b) overlay of clusters from the merged cancer (r = 0.6) graph displayed using larger nodes. Many of the housekeeping clusters (1–5, 11) can be seen to be conserved, as is a proportion of the cell cycle (6), macrophage (7), T-cell (8), ECM (9), interferon response (12), plasma cell (14), MHC class 1 (19), histones (20) and Affymetrix control (23) clusters. However it can also be seen that many of the skin cancer clusters are not represented in the merged cancer profile set, these transcriptional signatures being unique to skin cancers.

Mentions: In order to confirm that the ‘core’ transcriptional signatures generated from the meta-analysis of six datasets are conserved in other cancer datasets, we mapped the signatures onto a number of completely independent tumour datasets derived from skin/melanoma [38], gastric cancer [47] and Hodgkin lymphoma [48]. In each case clusters derived from the meta-analysis of the six tumours identified corresponding clusters in these independent datasets. Shown here are the results of their comparison to a dataset consisting of primary skin cancers including basal cell carcinomas (BCC), squamous carcinomas (SCC) and melanomas, plus a number of metastatic melanomas [38]. Like the other independent datasets, this contained unique transcriptional signatures corresponding to the different tumour types represented in this dataset (Figure 6). However the core signatures were clearly also present. For example, cluster 16 (designated ‘macrophage’) in the skin cancer dataset was highly significantly enriched for genes found in the macrophage cluster in the ‘merged’ dataset (cluster 7 in Figure 5) (adjusted p-value = 1.3E-120) implying that these genes represent a true ‘functional unit’, in this case a cell signature. Similarly cell cycle, stromal and house-keeping clusters were also conserved in the skin cancer data (Table 1) and all other cancer datasets so far examined have all generated networks where the conserved signatures identified here have been found to be present.


Coexpression analysis of large cancer datasets provides insight into the cellular phenotypes of the tumour microenvironment.

Doig TN, Hume DA, Theocharidis T, Goodlad JR, Gregory CD, Freeman TC - BMC Genomics (2013)

Conservation of transcriptional signatures in graph derived from skin cancer dataset (Pearson correlation threshold r = 0.80). In order to provide a clear view of transcripts/clusters within skin cancer graph it has been simplified. The network shown here has been constructed with a central framework of edges derived from the relationships between clusters and nodes representing the transcripts in each cluster joined to a central node representing the cluster with the graph laid out in 2D. Only clusters comprising more than 8 probesets were included. a) Colours represent different clusters in the skin cancer data, and b) overlay of clusters from the merged cancer (r = 0.6) graph displayed using larger nodes. Many of the housekeeping clusters (1–5, 11) can be seen to be conserved, as is a proportion of the cell cycle (6), macrophage (7), T-cell (8), ECM (9), interferon response (12), plasma cell (14), MHC class 1 (19), histones (20) and Affymetrix control (23) clusters. However it can also be seen that many of the skin cancer clusters are not represented in the merged cancer profile set, these transcriptional signatures being unique to skin cancers.
© Copyright Policy - open-access
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC3721986&req=5

Figure 6: Conservation of transcriptional signatures in graph derived from skin cancer dataset (Pearson correlation threshold r = 0.80). In order to provide a clear view of transcripts/clusters within skin cancer graph it has been simplified. The network shown here has been constructed with a central framework of edges derived from the relationships between clusters and nodes representing the transcripts in each cluster joined to a central node representing the cluster with the graph laid out in 2D. Only clusters comprising more than 8 probesets were included. a) Colours represent different clusters in the skin cancer data, and b) overlay of clusters from the merged cancer (r = 0.6) graph displayed using larger nodes. Many of the housekeeping clusters (1–5, 11) can be seen to be conserved, as is a proportion of the cell cycle (6), macrophage (7), T-cell (8), ECM (9), interferon response (12), plasma cell (14), MHC class 1 (19), histones (20) and Affymetrix control (23) clusters. However it can also be seen that many of the skin cancer clusters are not represented in the merged cancer profile set, these transcriptional signatures being unique to skin cancers.
Mentions: In order to confirm that the ‘core’ transcriptional signatures generated from the meta-analysis of six datasets are conserved in other cancer datasets, we mapped the signatures onto a number of completely independent tumour datasets derived from skin/melanoma [38], gastric cancer [47] and Hodgkin lymphoma [48]. In each case clusters derived from the meta-analysis of the six tumours identified corresponding clusters in these independent datasets. Shown here are the results of their comparison to a dataset consisting of primary skin cancers including basal cell carcinomas (BCC), squamous carcinomas (SCC) and melanomas, plus a number of metastatic melanomas [38]. Like the other independent datasets, this contained unique transcriptional signatures corresponding to the different tumour types represented in this dataset (Figure 6). However the core signatures were clearly also present. For example, cluster 16 (designated ‘macrophage’) in the skin cancer dataset was highly significantly enriched for genes found in the macrophage cluster in the ‘merged’ dataset (cluster 7 in Figure 5) (adjusted p-value = 1.3E-120) implying that these genes represent a true ‘functional unit’, in this case a cell signature. Similarly cell cycle, stromal and house-keeping clusters were also conserved in the skin cancer data (Table 1) and all other cancer datasets so far examined have all generated networks where the conserved signatures identified here have been found to be present.

Bottom Line: As a result genes expressed in specific cell types, or associated with certain biological processes are detected at widely variable levels across samples in transcriptomic analyses.Using a novel 3D network-based approach we have analysed six large human cancer microarray datasets derived from more than 1,000 individuals.Based upon this analysis, and without needing to isolate the individual cells, we have defined a broad spectrum of cell-type and pathway-specific gene signatures present in cancer expression data which were also found to be largely conserved in a number of independent datasets.

View Article: PubMed Central - HTML - PubMed

Affiliation: Centre for Inflammation Research, University of Edinburgh, The Queen’s Medical Research Institute, Edinburgh, UK.

ABSTRACT

Background: Biopsies taken from individual tumours exhibit extensive differences in their cellular composition due to the inherent heterogeneity of cancers and vagaries of sample collection. As a result genes expressed in specific cell types, or associated with certain biological processes are detected at widely variable levels across samples in transcriptomic analyses. This heterogeneity also means that the level of expression of genes expressed specifically in a given cell type or process, will vary in line with the number of those cells within samples or activity of the pathway, and will therefore be correlated in their expression.

Results: Using a novel 3D network-based approach we have analysed six large human cancer microarray datasets derived from more than 1,000 individuals. Based upon this analysis, and without needing to isolate the individual cells, we have defined a broad spectrum of cell-type and pathway-specific gene signatures present in cancer expression data which were also found to be largely conserved in a number of independent datasets.

Conclusions: The conserved signature of the tumour-associated macrophage is shown to be largely-independent of tumour cell type. All stromal cell signatures have some degree of correlation with each other, since they must all be inversely correlated with the tumour component. However, viewed in the context of established tumours, the interactions between stromal components appear to be multifactorial given the level of one component e.g. vasculature, does not correlate tightly with another, such as the macrophage.

Show MeSH
Related in: MedlinePlus