Reply

We thank Dr. Vinciguerra for his comments about our Hepatology article (1) in which we reported an independently-validated 64 gene profile that reproducibly differentiated severe nonalcoholic fatty liver disease (NAFLD) from mild NAFLD. We agree that there is a great need for accurate risk stratification in NAFLD and feel that our article brings us one step closer to unraveling some of the key pathways differentiating patients at risk for clinically important outcomes. Dr. Vinciguerra questioned the method used for normalization of our gene expression data. We agree that this is an important aspect of data analysis and are grateful for the opportunity to clarify the points he raised.

In our study, normalization of the gene expression microarray analyses of the discovery and validation cohorts was performed using GCRMA (GeneChip Robust Multi-array Averaging). (2) GCRMA is a widely used normalization method built upon Robust Multi-array Average (RMA) normalization. It involves background adjustment, quantile normalization, and median-polish summarization on Affymetrix microarray probe-level data. In a recent review article, McCall and Almudevar cited previous works that suggest GCRMA and RMA are among the best performers for normalization of microarray gene expression data. (3) It is important to emphasize that the gene profile identified in our discovery cohort was validated in a second, independent cohort of patients. QRT-PCR was applied as yet another method for validation of the microarray data. Ribosomal protein 35 (RPL35) was only used as an internal control in the QRT-PCR validation. That QRT-PCR validation was performed for 8 genes that were found to be significantly differentially expressed in the microarray analysis of both the initial and validation cohorts (both of which used GCMRA for normalization). Other findings from the microarray analysis were further confirmed by immunohistochemistry to assure that observed differences in mRNA expression led to differences at the level of protein expression.

As Dr. Vinciguerra correctly points out, there are multiple methods that have been used to select internal control genes for normalization of transcript quantification assays, such as QRT-PCR. Because commonly used individual reference genes may not be stably expressed in all tissues, geometric averaging of expression of several different reference genes (“GeNorm”) has been suggested as an approach to overcome this limitation.(4, 5) An alternative strategy is to use a gene with stable expression as demonstrated by microarray analysis to serve as the internal control for QRT-PCR validation.(6) Despite evidence that expression of RPL35 mRNA and/or its protein product may vary with feeding and obesity in mice and in HCC cell lines (7, 8), RPL35 was stably expressed across the extremes of human NAFLD severity in our microarray analysis, so we selected it as an internal control for our QRT-PCR validation study. Reassuringly, another group subsequently reported that normalization using this method was superior to the “GeNorm” method for QRT-PCR analysis of hepatic gene expression.(6) It will be important to determine if other groups can independently reproduce our findings.