Interpretation of transcript profiling.

To the Editor: We found the recent report by Liu et al1 using transcript profiling to examine genes regulated by insulin growth factor-1 interesting because of the biological and technological issues it raises. We applaud the efforts of the investigators to begin to address critically important and often neglected areas of microarray technology such as specificity and reproducibility. However, we take issue with some of the terminology, methodology, and underlying conceptual assumptions. The authors compared the results obtained from independent RNA samples extracted from control cardiomyocytes and assumed that any differences were attributable to “false-positive results” (page 1232). In fact, although the cardiomyocytes were “grown and harvested under the same protocol” (page 1232), it is quite possible—indeed likely—that true differences in transcript levels exist between such samples as a result of biological variation and genetic heterogeneity in the outbred animals used for this study. To rigorously address the false-positive (or negative) rate of any assay requires comparison of results to an independent, previously validated experimental standard. For transcript profiling, a variety of techniques including Northern blotting, RNase protection assays, and real-time PCR could serve this purpose but were not used in this study. Although labor-intensive, experimental verification of microarray results is required not only for validation of specific results but also to justify more global deductions about the specificity or sensitivity of the technology overall. Moreover, by using these postarray validation techniques, the investigator can reduce the filtering stringency applied to microarray data. In this way, changes in the expression of potentially important transcripts with low copy number may be identified and verified, thereby reducing the falsenegative rate. Thus, although we are pleased to see investigators beginning to address these important methodological issues and agree with the need for repetition in microarray experiments, we believe the results of the present study must be interpreted cautiously in the absence of independent experimental validation.