Normalization of a NAFLD gene expression signature

Abstract

We read with great interest the article by Moylan et al.1 in which the authors demonstrate that a 64-gene profile reproducibly differentiated severe nonalcoholic fatty liver disease (NAFLD) from mild NAFLD, and that a 20-gene subset within this profile correlated with NAFLD severity, independent of other known factors impinging on NAFLD progression. The need for accurate risk stratification for NAFLD is urgent because the disease is epidemic, imposing a public health burden, and cirrhosis and/or hepatocellular carcinoma (HCC) may develop on an NAFLD background. The authors unraveled, using microarrays and quantitative reverse-transcription polymerase chain reaction (qRT-PCR) in the livers of patients, a novel gene expression signature that independently correlated with NAFLD severity.1 Although these findings might help develop novel diagnostic tests and treatments that target NAFLD patients at greatest risk, there are some important aspects that need to be stressed. The choice of housekeeping (control) genes, whose expression is not susceptible to nutrients, is crucial to normalize results in gene expression studies involving liver tissue. Moylan et al. used ribosomal protein L35 (RPL35) as the only control gene for all analyses. Studies on the translatome (total proteins formed by mRNA translation) show tremendous fluctuations in polysome-associated RPL35 levels in the livers of obese versus chow-fed mice, and also between fasted and fed states.2 Gene expression of RPL35 is increased in HCC cell lines and tissues.3 Moylan et al. relied most likely on the fact that RPL35 levels did not vary significantly in the clinical samples analyzed. However, if the expression of a control gene is potentially heavily influenced by nutritional and proliferative status in the liver, that could pose a problem when defining a gene expression signature with clinical applications. There is now a strong consensus in the community on the validity of the “GeNorm” method, employing the geometric mean of multiple carefully selected housekeeping genes to achieve robust normalization of gene expression data.4, 5 It would be interesting to reassess the data obtained by Moylan et al. using three or more housekeeping genes for normalization. Without robust normalization strategies it will be hard to assess the prognostic power of gene expression signatures. Manlio Vinciguerra, Ph.D. University College London (UCL) Institute for Liver and Digestive Health UCL Medical School Royal Free Campus London, UK