Limits...
Exon-array profiling unlocks clinically and biologically relevant gene signatures from formalin-fixed paraffin-embedded tumour samples.

Hall JS, Leong HS, Armenoult LS, Newton GE, Valentine HR, Irlam JJ, Möller-Levet C, Sikand KA, Pepper SD, Miller CJ, West CM - Br. J. Cancer (2011)

Bottom Line: Differential gene expression was confirmed using Quantigene, a multiplex bead-based alternative to qRT-PCR.Quantigene analysis of the top 26 differentially expressed genes correctly partitioned cervix samples as SCC or AC.FFPE samples can be profiled using Exon arrays to derive gene expression signatures that are sufficiently robust to be applied to independent data sets, identify novel biology and design assays for independent platform validation.

View Article: PubMed Central - PubMed

Affiliation: Translational Radiobiology Group, School of Cancer and Enabling Sciences, Manchester Academic Health Science Centre, The University of Manchester, Wilmslow Road, Manchester M20 4BX, UK.

ABSTRACT

Background: Degradation and chemical modification of RNA in formalin-fixed paraffin-embedded (FFPE) samples hamper their use in expression profiling studies. This study aimed to show that useful information can be obtained by Exon-array profiling archival FFPE tumour samples.

Methods: Nineteen cervical squamous cell carcinoma (SCC) and 9 adenocarcinoma (AC) FFPE samples (10-16-year-old) were profiled using Affymetrix Exon arrays. The gene signature derived was tested on a fresh-frozen non-small cell lung cancer (NSCLC) series. Exploration of biological networks involved gene set enrichment analysis (GSEA). Differential gene expression was confirmed using Quantigene, a multiplex bead-based alternative to qRT-PCR.

Results: In all, 1062 genes were higher in SCC vs AC, and 155 genes higher in AC. The 1217-gene signature correctly separated 58 NSCLC into SCC and AC. A gene network centered on hepatic nuclear factor and GATA6 was identified in AC, suggesting a role in glandular cell differentiation of the cervix. Quantigene analysis of the top 26 differentially expressed genes correctly partitioned cervix samples as SCC or AC.

Conclusion: FFPE samples can be profiled using Exon arrays to derive gene expression signatures that are sufficiently robust to be applied to independent data sets, identify novel biology and design assays for independent platform validation.

Show MeSH

Related in: MedlinePlus

A refined gene signature accurately partitions the NSCLC samples into histological subtypes. (A) Proportion of times the 2673 differentially expressed probesets appear as significant in 100 jackknifed data sets (with 10% and 30% of the samples removed) against rank in the original data set. (B) Hierarchical clustering of the 296 stable probesets on the 58 NSCLC samples. Genes and samples were clustered based on Pearson's correlation. The scaled expression of each probeset, denoted as the row Z-score, is plotted in red–blue colour scale with red indicating high expression and blue indicating low expression. The coloured bar above the heatmap indicates the histological classification: orange=SCC; green=AC; blue=misclassified samples. (C) Principal component analysis of the 296 stable probesets when applied to the NSCLC data set. The numbers represent the patient IDs. Different colours are used to represent the different histological subtypes: orange=SCC; green=AC; blue=misclassified samples. (D) Percentage variance explained by the first 10 principal components.
© Copyright Policy
Related In: Results  -  Collection


getmorefigures.php?uid=PMC3065290&req=5

fig6: A refined gene signature accurately partitions the NSCLC samples into histological subtypes. (A) Proportion of times the 2673 differentially expressed probesets appear as significant in 100 jackknifed data sets (with 10% and 30% of the samples removed) against rank in the original data set. (B) Hierarchical clustering of the 296 stable probesets on the 58 NSCLC samples. Genes and samples were clustered based on Pearson's correlation. The scaled expression of each probeset, denoted as the row Z-score, is plotted in red–blue colour scale with red indicating high expression and blue indicating low expression. The coloured bar above the heatmap indicates the histological classification: orange=SCC; green=AC; blue=misclassified samples. (C) Principal component analysis of the 296 stable probesets when applied to the NSCLC data set. The numbers represent the patient IDs. Different colours are used to represent the different histological subtypes: orange=SCC; green=AC; blue=misclassified samples. (D) Percentage variance explained by the first 10 principal components.

Mentions: Because of transcript fragmentation, there is scepticism over the reproducibility and clinical applicability of results generated from expression profiling FFPE samples. To address this, we assessed the stability of the list of 2673 differentially expressed probesets using jackknife analysis. We removed 10 or 30% of the samples from the original data set, and generated 100 jackknifed data sets for each subsample. The jackknifed data sets were analysed with LIMMA as before to determine how the perturbation affects the composition of the resulting gene lists. Removal of 10% of the samples modified differentially expressed probesets only moderately; with 1151 (43% of original DE probesets) probesets remained significant 95% of the time (Figure 6A). Omitting 30% of samples led to the number of significant probesets declining markedly to 296 (11% of original DE probesets). However, the resulting 296 probesets that were called significant 95% of the time were still sufficiently robust to discriminate between SCC and AC in the independent NSCLC data set (Figures 6B–D).


Exon-array profiling unlocks clinically and biologically relevant gene signatures from formalin-fixed paraffin-embedded tumour samples.

Hall JS, Leong HS, Armenoult LS, Newton GE, Valentine HR, Irlam JJ, Möller-Levet C, Sikand KA, Pepper SD, Miller CJ, West CM - Br. J. Cancer (2011)

A refined gene signature accurately partitions the NSCLC samples into histological subtypes. (A) Proportion of times the 2673 differentially expressed probesets appear as significant in 100 jackknifed data sets (with 10% and 30% of the samples removed) against rank in the original data set. (B) Hierarchical clustering of the 296 stable probesets on the 58 NSCLC samples. Genes and samples were clustered based on Pearson's correlation. The scaled expression of each probeset, denoted as the row Z-score, is plotted in red–blue colour scale with red indicating high expression and blue indicating low expression. The coloured bar above the heatmap indicates the histological classification: orange=SCC; green=AC; blue=misclassified samples. (C) Principal component analysis of the 296 stable probesets when applied to the NSCLC data set. The numbers represent the patient IDs. Different colours are used to represent the different histological subtypes: orange=SCC; green=AC; blue=misclassified samples. (D) Percentage variance explained by the first 10 principal components.
© Copyright Policy
Related In: Results  -  Collection

Show All Figures
getmorefigures.php?uid=PMC3065290&req=5

fig6: A refined gene signature accurately partitions the NSCLC samples into histological subtypes. (A) Proportion of times the 2673 differentially expressed probesets appear as significant in 100 jackknifed data sets (with 10% and 30% of the samples removed) against rank in the original data set. (B) Hierarchical clustering of the 296 stable probesets on the 58 NSCLC samples. Genes and samples were clustered based on Pearson's correlation. The scaled expression of each probeset, denoted as the row Z-score, is plotted in red–blue colour scale with red indicating high expression and blue indicating low expression. The coloured bar above the heatmap indicates the histological classification: orange=SCC; green=AC; blue=misclassified samples. (C) Principal component analysis of the 296 stable probesets when applied to the NSCLC data set. The numbers represent the patient IDs. Different colours are used to represent the different histological subtypes: orange=SCC; green=AC; blue=misclassified samples. (D) Percentage variance explained by the first 10 principal components.
Mentions: Because of transcript fragmentation, there is scepticism over the reproducibility and clinical applicability of results generated from expression profiling FFPE samples. To address this, we assessed the stability of the list of 2673 differentially expressed probesets using jackknife analysis. We removed 10 or 30% of the samples from the original data set, and generated 100 jackknifed data sets for each subsample. The jackknifed data sets were analysed with LIMMA as before to determine how the perturbation affects the composition of the resulting gene lists. Removal of 10% of the samples modified differentially expressed probesets only moderately; with 1151 (43% of original DE probesets) probesets remained significant 95% of the time (Figure 6A). Omitting 30% of samples led to the number of significant probesets declining markedly to 296 (11% of original DE probesets). However, the resulting 296 probesets that were called significant 95% of the time were still sufficiently robust to discriminate between SCC and AC in the independent NSCLC data set (Figures 6B–D).

Bottom Line: Differential gene expression was confirmed using Quantigene, a multiplex bead-based alternative to qRT-PCR.Quantigene analysis of the top 26 differentially expressed genes correctly partitioned cervix samples as SCC or AC.FFPE samples can be profiled using Exon arrays to derive gene expression signatures that are sufficiently robust to be applied to independent data sets, identify novel biology and design assays for independent platform validation.

View Article: PubMed Central - PubMed

Affiliation: Translational Radiobiology Group, School of Cancer and Enabling Sciences, Manchester Academic Health Science Centre, The University of Manchester, Wilmslow Road, Manchester M20 4BX, UK.

ABSTRACT

Background: Degradation and chemical modification of RNA in formalin-fixed paraffin-embedded (FFPE) samples hamper their use in expression profiling studies. This study aimed to show that useful information can be obtained by Exon-array profiling archival FFPE tumour samples.

Methods: Nineteen cervical squamous cell carcinoma (SCC) and 9 adenocarcinoma (AC) FFPE samples (10-16-year-old) were profiled using Affymetrix Exon arrays. The gene signature derived was tested on a fresh-frozen non-small cell lung cancer (NSCLC) series. Exploration of biological networks involved gene set enrichment analysis (GSEA). Differential gene expression was confirmed using Quantigene, a multiplex bead-based alternative to qRT-PCR.

Results: In all, 1062 genes were higher in SCC vs AC, and 155 genes higher in AC. The 1217-gene signature correctly separated 58 NSCLC into SCC and AC. A gene network centered on hepatic nuclear factor and GATA6 was identified in AC, suggesting a role in glandular cell differentiation of the cervix. Quantigene analysis of the top 26 differentially expressed genes correctly partitioned cervix samples as SCC or AC.

Conclusion: FFPE samples can be profiled using Exon arrays to derive gene expression signatures that are sufficiently robust to be applied to independent data sets, identify novel biology and design assays for independent platform validation.

Show MeSH
Related in: MedlinePlus