Email updates

Keep up to date with the latest news and content from BMC Molecular Biology and BioMed Central.

Open Access Research article

Use of Maximum Likelihood-Mixed Models to select stable reference genes: a case of heat stress response in sheep

Magdalena Serrano1*, Natalia Moreno-Sánchez1, Carmen González1, Ane Marcos-Carcavilla1, Mario Van Poucke2, Jorge H Calvo3, Judit Salces1, Jaime Cubero4 and María J Carabaño1

Author Affiliations

1 Departamento de Mejora Genética Animal, INIA, Ctra. de la Coruña km 7.5, Madrid, Spain

2 Department of Nutrition, Genetics and Ethology, Faculty of Veterinary Medicine, Ghent University, Heidestraat 19, B-9820 Merelbeke, Belgium

3 Unidad de Tecnología en Producción Animal, CITA, Zaragoza, Spain

4 Departamento de Protección Vegetal, INIA. Ctra. de la Coruña km 7.5, Madrid, Spain

For all author emails, please log on.

BMC Molecular Biology 2011, 12:36  doi:10.1186/1471-2199-12-36

Published: 17 August 2011

Abstract

Background

Reference genes with stable expression are required to normalize expression differences of target genes in qPCR experiments. Several procedures and companion software have been proposed to find the most stable genes. Model based procedures are attractive because they provide a solid statistical framework. NormFinder, a widely used software, uses a model based method. The pairwise comparison procedure implemented in GeNorm is a simpler procedure but one of the most extensively used. In the present work a statistical approach based in Maximum Likelihood estimation under mixed models was tested and compared with NormFinder and geNorm softwares. Sixteen candidate genes were tested in whole blood samples from control and heat stressed sheep.

Results

A model including gene and treatment as fixed effects, sample (animal), gene by treatment, gene by sample and treatment by sample interactions as random effects with heteroskedastic residual variance in gene by treatment levels was selected using goodness of fit and predictive ability criteria among a variety of models. Mean Square Error obtained under the selected model was used as indicator of gene expression stability. Genes top and bottom ranked by the three approaches were similar; however, notable differences for the best pair of genes selected for each method and the remaining genes of the rankings were shown. Differences among the expression values of normalized targets for each statistical approach were also found.

Conclusions

Optimal statistical properties of Maximum Likelihood estimation joined to mixed model flexibility allow for more accurate estimation of expression stability of genes under many different situations. Accurate selection of reference genes has a direct impact over the normalized expression values of a given target gene. This may be critical when the aim of the study is to compare expression rate differences among samples under different environmental conditions, tissues, cell types or genotypes. To select reference genes not only statistical but also functional and biological criteria should be considered. Under the method here proposed SDHA/MDH1 have arisen as the best set of reference genes to be used in qPCR assays to study heat shock in ovine blood samples.