Evolution of the Bovine TLR Gene Family and Member Associations with Mycobacterium avium Subspecies paratuberculosis Infection

Fisher, Colleen A.; Bhattarai, Eric K.; Osterstock, Jason B.; Dowd, Scot E.; Seabury, Paul M.; Vikram, Meenu; Whitlock, Robert H.; Schukken, Ynte H.; Schnabel, Robert D.; Taylor, Jeremy F.; Womack, James E.; Seabury, Christopher M.

Conditional logistic regression models were constructed for each of the 35 variable loci to estimate the relative odds of being infected with MAP based on the defined diagnostic criteria adjusted for the effects of herd using the PHREG procedure of SAS (SAS v. 9.2, SAS, Cary, NC). Effects of genotype were estimated using 3 different covariate specifications. First, an additive mode of inheritance was examined whereby the odds of infection associated with each additional copy of the minor allele was modeled as a single continuous covariate. Second, a recessive mode of inheritance was modeled, where the odds of infection in cattle homozygous for the minor allele were estimated relative to cattle heterozygous and homozyzgous for the major allele. Finally, each genotype was modeled as an indicator variable and effect estimates were generated for cattle homozygous for the minor allele, and for heterozygous cattle, both relative to cattle homozygous for the major allele. This allowed evaluation of assumptions in the additive model with respect to the effect of the additional copies of the minor allele being linear in the log odds, and potential intermediate effects of the minor allele not captured in the other models. Potential confounding by age was examined by including birth year as a fixed covariate (where available), and was defined as a change in the relative odds of greater than 20% after addition of the birth year term. For models where evidence of confounding by age was detected, birth year was retained in the model to adjust genotype estimates for this effect. With the exception of TLR1, TLR6, and TLR10, all single marker P-values were corrected for multiple testing by applying the FDR correction (http:// sdmproject.com/utilities/?show=FDR) [45] to the raw P-values derived from each investigated gene (locus-specific correction). Given the close physical proximity of TLR1, TLR6, and TLR10 on BTA6, these genes were considered a single locus for correction of multiple tests. However, it should be noted that none of the variable markers within TLR1 met our inclusion criteria (MAFs.0.01), and therefore, locus-specific correction was only applied to raw P-values from TLR6 and TLR10.