SNP-expression associations referred to in Fig. 3 and Supplementary Table 6 were carried out using linear models that included surrogate variables generated as described above, age, life stage (as defined in Fig. 1b), an interaction of age and life stage, sex, race, and the SNP under investigation. The SNP was included as a continuous variable, coded as 1,2,3; that is, an additive or ‘dosage’ model rather than a categorical or ‘co-dominant’ model was used. Association methods used to generate hits for the genetic distance used in Supplementary Fig. 2, part 1, were identical to those described above except for the omission of the race and sex terms in the linear model. To confirm that this analytical framework is capable of discovering associations between genetic and transcriptional metrics as analysed here, we conducted a positive control analysis (Supplementary Fig. 2, part 2; negative control also included).
