The animal feeding experiment was conducted at the BioResources Unit, Trinity College Dublin (TCD) Ireland according to European Union (EU) animal research welfare protocol, with approval for experimentation granted by the Department of Health and Children in Ireland (License number B100/3041). Fourteen, 4-week-old male, ob/ob (C57BL/6J) mice were purchased from Harlan, UK. The mice were acclimatised for 7 days during which time they received a purified control diet, before being assigned to one of two treatment groups for a 28-day period. During the intervention period, the animals were exposed to 12 hrs light/12 hrs dark cycles, maintained at a constant temperature of 22°C.
Dietary composition and preparation of the animal feeds
Diets were produced by Special Diets Service, Essex, UK and were received as 1 kg vacuum packed, heat sealed plastic bags. Low-CLA and high-CLA beef (0.53 and 2.65 w/w% of c9,t11 CLA, respectively) were provided by Teagasc (Grange Research Centre, Dunsany, Co. Meath). Test diet blends were prepared by mixing the beef component at a 36% inclusion rate to equal portions of wheat feed and maize (corn) feed. Final feeds were prepared by mixing 100 ml warm water with 100 g test diet blend. Dietary food intake was measured daily and the animals received freshly prepared food each day.
Blood sample and tissue collection and handling protocol
The mice were sacrificed at day 28 of the dietary intervention period. Food was removed from the cages at 6:00 pm and the animals were sacrificed the following morning between 8:00 am - 10:00 am, in the fasted state. The animals were euthanized using Carbon Dioxide (CO2) and cardiac puncture was performed to draw blood samples. Blood was transferred to a cooled sodium citrate blood vacutainer tube (BD Vacutainer, Dublin, Ireland) and centrifuged at 1500 rpm for 15 mins at 4°C, plasma was harvested, aliquoted and stored (-70°C). Tissue samples for gene expression analysis were harvested, immediately immersed in 0.5 ml RNALater (Ambion, AMS Techonology) and stored (-70°C). RNA was later extracted using a Qiagen RNeasy extraction kit, and outsourced to ServiceXS http://www.servicexs.com for hybridization to Affymetrix arrays, custom designed by the European Nutrigenomics Organization containing 15313 probesets. This platform is designated 'nugomm1a520177', and we used the 'entrezg' version 12.1.0 annotation from the MBNI custom cdf database http://brainarray.mbni.med.umich.edu/Brainarray/Database/CustomCDF/genomic_curated_CDF.asp, which reflects the latest remapping of Affymetrix probes based on current data in the NCBI database . The complete array data are available at the GEO database under accession GSE23337.
Determination & statistical analysis of plasma markers of metabolic syndrome
Plasma glucose concentrations were analysed using an endpoint enzymatic glucose oxidase, peroxidase, chromogen sequence, colorimetric assay (Biomérieux, France). A multiplex ELISA assay kit manufactured by Linco Research (Missouri, USA) was used to simultaneously quantify insulin, TNFα, MCP-1, resistin, and PAI-1 concentrations from mouse serum samples, while IL-6 and adiponectin were measured using ELISA kits from BioSources International (California, USA) and R&D Systems (Minnesota, USA), respectively. Plasma triglycerides (TAG) and cholesterol levels were measured using enzymatic-based assays from Randox Laboratories (Co. Antrim, UK), while plasma non-esterified fatty acids (NEFA) were quantified using a Randox NEFA kit. Insulin resistance (as defined by the homeostasis model assessment insulin resistance index; HOMAIR) was calculated as [fasting glucose (mg/dl) multiplied by fasting insulin (μU/ml)] divided by 22.5 . Significance of plasma marker level variation between groups was determined using ANOVA in conjunction with Tukey's honest significant differences test, which corrects for experiment-wise error rate.
Processing of microarray data, and single gene statistical analysis
Raw microarray data were first assessed for quality using a set of standard QC tests, including array intensity distribution, positive and negative border element distribution, GAPDH and ß-actin 3':5' ratios, centre of intensity and array-array correlation check. All QC tests were implemented in the R programming language , using the affyQCReport library . After quality assessment, all intensity values were background corrected and normalized (within each tissue group) using the GCRMA-slow method (which uses a slower and more exact optimization algorithm) . Probesets were filtered to remove genes with low or null expression, using a filter wherein probesets showing an intensity score less than 3 on more than 50% of the arrays were removed. Filtered adipose, liver and skeletal muscle datasets comprised 8575, 7781 and 8093 probesets, respectively. Single gene analysis was carried out using the LIMMA library , wherein linear models were fitted to each probeset on the array, to determine statistical significance of the effect of the high-CLA beef diet. Empirical Bayes statistics were generated using the eBayes() function, and resultant p-values were adjusted for multiple testing, using the Benjamini & Hochberg method .
Gene set enrichment analysis
A script in R was written to carry out gene set enrichment analysis on each tissue dataset, adapting the statistical code provided in the GSEAlm library in R (Additional file 1) . In addition to the typical single-direction enrichment, an additional test was included where absolute values of t-statistics were used, to detect bi-directional enrichment. T-statistics were extracted from linear models, which were fitted to each gene in a given gene set (i.e., KEGG pathway) - using 'diet group' as the predictor variable and 'expression level' as the response. These t-statistics (absolute values of t-statistics for bi-directional enrichment analysis) were then summed, and normalized for the number of genes in the gene set. Diet group labels were then randomized, as in a typical permutation test, and gene-set t-statistics were re-calculated using these randomized groupings. This permutation step was repeated 1000 times, and p-values were calculated by determining the proportion of permutation t-statistics that were closer to zero than the 'true' t-statistic. For instance, a p-value of 0.05 would be recorded if the original t-statistic were greater than more than 95% of the permutation t-statistics. These p-values were corrected using the Benjamini & Hochberg method . R scripts were written to produce summary plots of the results, and also to import KEGG pathway data, integrate microarray results, and export the annotated pathway to Cytoscape http://www.cytoscape.org for visualisation.
Regularized canonical correlation analysis
To determine canonical correlations between metabolic and transcriptomic data, gene expression and metabolic marker values were centered to 0 and scaled to have variance 1 (i.e, z-score normalized) within each diet group, to reveal the null correlations between gene expression and metabolic markers, irrespective of dietary treatment. To make CCA results more easily comparable to GSEA results, we used the subset of genes in our expression dataset with annotation to a KEGG pathway. The 'mixOmics' library of functions in R was used to carry out the analysis . Specifically, the rcc function was used to define the canonical correlations and the canonical variates, estim.regul for estimation of regularization parameters and the network function to produce the initial network of interactions. An additional script was written to output the R network to Cytoscape for visualization. Taking the group of genes with a correlation score of at least .65 (using 'threshold' argument of the network function in the mixOmics library; for further information on this association measure see ) with at least one plasma marker, Fisher's exact test was performed to define pathways that were significantly overrepresented among MetS-associated genes .