Computational Dissection of Tumor Expression Profiles.

Abstract

Gene expression profiling has yielded important insights about tumor biology that may improve treatment decisions in patients. However, it is difficult to collect a specimen of pure tumor cells, and thus microarray measurements usually reflect the contribution of tumor cells as well as stromal and other normal cells. We applied unsupervised matrix factorization methods to gene expression data to derive several sets of co-expressed genes, or signatures, that together comprise a set of independent descriptors of breast tumors. Some of these signatures correspond to specific cell types (adipocytes, lymphocytes, fibroblasts), while others reflect well-known tumor-intrinsic expression programs (ER, ERBB2, proliferation). We confirmed the specificity of the signatures using expression data from purified normal cells and tumor cell lines, microdissected tumors, and bulk tumors with corresponding histological cellularity estimates. In several data sets, cell-type signatures were highly variable and anticorrelated with tumor-intrinsic signatures, confirming that variation in normal cell content is a potential source of measurement bias.Overall, these results provide an intuitive framework for the interpetation of tumor expression profiles and may lead to an improved understanding of the physiological mechanisms involved in tumor development. Citation Information: Cancer Res 2009;69(24 Suppl):Abstract nr 1166.

Zonation of Ribosomal DNA Transcription Defines a Stem Cell Hierarchy in Colorectal Cancer

“…Raw cel files were normalized using procedure described below (see section Microarray analyses). Technical information concerning samples processing and hybridization was retrieved from the original CEL files: date of scanning were collected in order to define scan batches in each dataset separately; technical metrics described by Eklund AC and Szallasi Z in (Eklund and Szallasi, 2008) were computed and recorded as additional features for each sample. Expression data were corrected by metrics PM.IQR, RMA.IQR and RNA.DEG (Eklund and Szallasi, 2008) and by scanning day.…”

“…Technical information concerning samples processing and hybridization was retrieved from the original CEL files: date of scanning were collected in order to define scan batches in each dataset separately; technical metrics described by Eklund AC and Szallasi Z in (Eklund and Szallasi, 2008) were computed and recorded as additional features for each sample. Expression data were corrected by metrics PM.IQR, RMA.IQR and RNA.DEG (Eklund and Szallasi, 2008) and by scanning day. For doing so, a linear model was fitted separately for each probeset that included these metrics as the only explanatory variables, and the coefficients of such models were used to correct the expression values a-priori.…”

“…Standard quality controls were performed in order to identify abnormal samples regarding: a) spatial artifacts in the hybridization process (scan images and pseudo-images from probe level models); b) intensity dependences of differences between chips (MvA plots); c) RNA quality (RNA digest plot); d) global intensity levels (boxplot of perfect match log-intensity distributions before and after normalization and RLE plots); e) anomalous intensity profile compared to the rest of the samples (NUSE plots, Principal Component Analyses); f) impact of quality metrics (Eklund and Szallasi, 2008) on expression measures. Samples from batch "cm.1509" in the tumor dataset (refers to the Ephb2 populations, Figure 1) were a priori corrected gene-wise by RMA.IQR metric (Eklund and Szallasi, 2008) using a linear model with no more covariates included in it. No samples were excluded from the study due to quality issues.…”

Influence of pH on the cytotoxic activity of inositol hexakisphosphate (IP6) in prostate cancer

“…As previously described, PC-3 cells were treated with 2 mM IP6 solutions or with the corresponding volumes of buffered H 2 O. We chose the 2-mM IP6 dose in line with other studies by our group (Diallo et al, 2008) and others (Gu et al, 2009; Roy et al, 2009). Following a 72-h stimulation, IP6 at pH 12 significantly reduced cell number by 60.4% ( P  = 0.001, compared to H 2 O at pH 12, ANOVA), whereas a treatment with IP6 at pH 5 reduced the proliferation rate by 46.9% ( P  = 0.0029, compared to H 2 O at pH 5, ANOVA), and by 46.6% for the IP6 at pH 7 ( P  = 0.015, compared to H 2 O at pH 7, ANOVA; Figures 1C,D).…”

Immune translational control by CPEB4 regulates intestinal inflammation resolution and colorectal cancer development

“…Technical information concerning sample processing and hybridization was retrieved from the original CEL files: scanning dates were collected in order to define scan batches in each dataset separately. The technical metrics described by Eklund AC and Szallasi Z (Eklund and Szallasi, 2008) were computed and recorded as additional features for each sample. Microarray expression values were summarized at the gene level (entrez) using the first principal component of the probesets mapping to the same gene.…”