Comments on the use of a single or multiple probe-set approach for microarray-based analyses of routine molecular markers in breast cancer

Abstract

In a recent paper, Spyratos et al. [1] dealt with the relevant issue of the routine use of DNA-microarray-based analyses as alternative method to evaluate molecular markers in breast cancer. In particular, they tested the reliability of single Affymetrix probe-sets, in evaluating the performance of microarray data for prediction of estrogen (ER), progesterone receptor (PR), and HER2 status (as determined by immunohistochemistry). Since not all Affymetrix probe-sets are equivalent for a given gene, they separately analyzed each probe-set for each gene and compared microarray data to quantitative RT-PCR results of 45 genes including ESR1, PGR, and ERBB2 (respectively coding for ER, PR, and HER2 protein). As expected, they confirmed the unequal quality of quantification of the different probesets for ESR1 (9 probe-sets), PGR (2 probe-sets), and ERBB2 (3 probe-sets) by comparison to protein expression. As a similar result was observed for other 42 genes analyzed, they concluded that microarray results from Affymetrix device may greatly vary according to the probe-set chosen providing misleading interpretation. The clinical relevance of such a conclusion is not negligible taking into account the decisional role in the clinical setting played by some biomarkers. In a series of in silico studies [2–4], aimed to evaluate a selected panel of genes experimentally recognized as involved in specific biological pathways, we dealt with a similar problem. We approached it adopting a conservative analytical strategy aimed to avoid the loss of potentially useful information because of probe-sets dropping away and, at the same time, to evaluate the specific contribution of each probe-set to gene expression properly. As each Affymetrix probe-set is characterized by an individual specificity and sensitivity that differently contribute to gene expression value, we calculated gene expression mean value after weighting each probe-set for its own sensitivity and specificity scores (information freely available at the GeneAnnot web site [5]). Specifically, the sensitivity score is the fraction of probes in a probe-set that match a respective gene, i.e., the number of matching probes in the given probe-set to a certain gene, divided by the total number of probes in this probe-set (which is usually 11 in probe-sets of U133A-B and U Plus 2.0 array sets). The specificity score indicates to what extent probes of a probeset bind to genes. It sums up the number of matching probes while giving lower weight to probes that match additional genes, and eventually divided by the total number of probes that matched any gene. As an example of the information provided by GeneAnnot system, Table 1 shows nine genes we selected for our studies [2–4] and analyzed in the paper by Spyratos et al. [1]. All these genes are recognized by at least two probe-sets. Notably, most of the probe-sets have sensitivity and specificity score equal to 1, indicating that all probes in the probe-set match the gene (sensitivity) in an exclusive manner (specificity). Conversely, other probe-sets have high sensitivity but low specificity (i.e., the probes match at least two different genes) or high specificity but low sensitivity (i.e., not all probes in the probe-set match the gene). Figure 1 shows a representative example of the pattern of expression of three genes (ESR1, CD24, and CD44), respectively, recognized by 6, 8, and 9 probe-sets, in a series D. Coradini (&) F. Ambrogi P. Boracchi Department of Clinical and Community Health Sciences, Medical Statistics, Biometry and Bioinformatics, University of Milano, Via Vanzetti 5, 20133 Milan, Italy e-mail: danila.coradini@yahoo.it