This Article Full Text (PDF) Purchase Article View Shopping Cart Alert me when this article is cited Alert me if a correction is posted Services Email this article to a colleague Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Save to my personal folders Download to citation manager Rights & Permissions Citing Articles Citing Articles via Google Scholar Google Scholar Articles by Koscielny, S. Articles by Hill, C. Search for Related Content PubMed PubMed Citation Articles by Koscielny, S. Articles by Hill, C. Related Articles Related Reply Social Bookmarking                            What's this?

Validation of Microarray Data by Quantitative Reverse-Transcriptase Polymerase Chain Reaction

Institut Gustave Roussy, Villejuif, France

To the Editor:

The article by Espinoza et al¹ is an example of poor methodology resulting in a misleading conclusion.

The purpose of the study was to reproduce with quantitative reverse-transcriptase polymerase chain reaction (qRT-PCR) the results obtained with the 70-gene expression profile in breast cancer described by van 't Veer.² The expressions of 60 of these 70 genes were measured by qRT-PCR in 96 new patients. For each patient, a distance/similarity parameter was calculated between his gene expression and the gene expression of patients with poor prognosis (the poor gene profile).

Different cutoff values were tested for this distance and the cutoff leading to the minimum P value of the log-rank test for disease-free survival curve was retained. As a reference to support this minimum P value method, the authors quote a paper that demonstrates its lack of validity and explains that the chance of declaring a quantitative factor as prognostically relevant when in fact it has no influence on survival is approximately 50% with an intended level of 5%.³ The problem with this method is well identified^4,5 and is one of the issues in the recent recommendations for the reporting of tumor marker prognostic studies.⁶

The decision rule (ie, transformation for a patient of 60 gene expression measurements in one prognostic classifier) defined with this method is the best possible for this training set of 96 patients, and its validity on another set of patients has not been investigated.

The authors justify not validating their findings by stating "because the profile has been previously validated in the original micro-array study, we think that the possibility of a chance association is low." This justification is not acceptable. The profile published by van ’t Veer is made of both a list of genes and a decision rule to classify new patients. Since the decision rule used by Espinoza et al is not the same as the one published by van ’t Veer, the results have to be validated. Whatever the list of genes, one might find a decision rule that classifies correctly patients in the training set by chance alone, and validation is crucial.

The conclusion "qRT-PCR reproduced the results obtained with microarrays for a prognostic gene signature" is misleading. Since the patients, the list of genes, and the decision rule are different, the study provides absolutely no support that qRT-PCR reproduces microarray data.

Moreover, we investigated the correlation between gene expressions evaluated by the qRT-PCR and microarrays in 42 colon tumors. Thirty-nine genes were randomly selected for analysis by qRT-PCR. Overall, the correlation between the 42 x 39 = 1,628 microarrays and quantitative RT-PCR was 0.68. The correlation was 0.5 for seven of the 39 genes. This means that microarray and qRT-PCR data are not deterministically related and therefore not interchangeable. Consequently, a decision rule established on microarray data has little chance to be valid with qRT-PCR, even if the patients and the list of genes were the same.

Authors’ Disclosures of Potential Conflicts of Interest

The authors indicated no potential conflicts of interest.

REFERENCES

1. Espinosa E, FresnoVara JA, Redondo A, et al: Breast Cancer prognosis determined by gene expression profiling: A quantitative reverse transcriptase polymerase chain reaction study. J Clin Oncol 23:7278-7285, 2005[Abstract/Free Full Text]

2. van ’t Veer LJ, Dai H, van de Vijver MJ, et al: Gene expression profiling predicts clinical outcome of breast cancer. Nature 415:530-536, 2002[CrossRef][Medline]

3. Schumacher M, Holländer N, Schwarzer G, et al: Prognostic factors studies, in Crowley J (ed): Handbook of Statistics in Clinical Oncology. New-York, NY, Marcel Dekker, 2001, pp321-378

4. Hilsenbeck SG, Clarck GM, McGuire WL: Why do so many prognostic factors fail to pan out? Breast Cancer Res Treat 22:197-206, 1992[CrossRef][Medline]

5. Farraggi D, Simon R: A simulation study of cross-validation for selecting an optimal cutpoint in univariate survival analysis. Stat Med 15:2203-2213, 1996[CrossRef][Medline]

6. McShane LM, Altman DG, Sauerbrei W, et al: Reporting recommendations for tumor marker prognostic studies (REMARK). Statistics Subcommittee of the NCI-EORTC Working Group on Cancer Diagnostics. J Natl Cancer Inst 97:1180-1184, 2005[Abstract/Free Full Text]

CiteULike    Complore    Connotea    Del.icio.us    Digg    Facebook    Reddit    Technorati    Twitter    What's this?

Related Reply

• In Reply:
  Enrique Espinosa, Juan Ángel Fresno Vara, Andrés Redondo, and José Javier Sánchez
  JCO 2005 23: 9440 [Full Text]

This Article Full Text (PDF) Purchase Article View Shopping Cart Alert me when this article is cited Alert me if a correction is posted Services Email this article to a colleague Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Save to my personal folders Download to citation manager Rights & Permissions Citing Articles Citing Articles via Google Scholar Google Scholar Articles by Koscielny, S. Articles by Hill, C. Search for Related Content PubMed PubMed Citation Articles by Koscielny, S. Articles by Hill, C. Related Articles Related Reply Social Bookmarking                            What's this?