Pathway analysis

To identify functional processes enriched among genes with the most extreme relative expression, functional gene annotations were compiled separately from the Gene Ontology (GO)⁴⁴ and Mouse Genome Informatics (MGI) Mammalian Phenotype (MP)⁴⁵ databases (4th July 2018). GO annotations derived from the following evidence codes were excluded: IEA (Inferred from Electronic Annotation); NAS (Non-traceable Author Statement); and RCA (inferred from Reviewed Computational Analysis).

To identify sets of genes with which to perform pathway analyses, all expressed genes were ranked by their expression during the developmental stage of interest then divided into ten equal bins. Each decile was tested for genetic association enrichment via a two-tailed competitive test in MAGMA⁴². Significantly associated deciles (following correction for ten tests) were then merged and tested for GO/MP term enrichment (GO and MP terms tested separately) relative to a background set of all expressed genes, using a Fisher’s exact test. Following Bonferroni correction for multiple testing of all GO or MP terms, significantly enriched terms were subjected to a refinement procedure: deciles were re-tested for enrichment of each term following the removal of genes from another term. This was done in a step-wise fashion whereby genes from the smallest term were removed first and any terms of weaker enrichment that were no longer significant on re-test (determined by uncorrected P-value in Fisher’s exact test) were removed from the analysis.

To investigate whether genes corresponding to refined GO/MP terms were more enriched for genetic association than other genes in significantly associated deciles, genes common to both refined terms and significant decile(s) were subjected to one-tailed gene-set association analyses conditioning on all genes within the significant decile(s).