Gene expression diversity is crucial for crop breeding, yet population genomics has focused primarily on sequence polymorphisms. A single reference genome for RNA-seq cannot handle introgression bias. Here, we conduct RNA-seq for 328 wheat lines, including landraces and elite cultivars from China and the United States, to investigate the expression variation underlying agronomic traits. Leveraging pan-genome resources, we identify 23,296 more transcripts than using the Chinese Spring reference. We construct a pan-gene regulatory atlas through eQTL analysis, revealing the tight regulation of introgressed genes. We identify 299 high-confidence candidate genes for 34 agronomic traits and resistance to 8 Blumeria graminis f. sp. tritici isolates, more than one-fifth of which were absent from the Chinese Spring. Utilizing the Kenong 9204 mutant library, 73.7% of the candidates show significant phenotypic effects. Our work mitigates the reference bias and highlights the impact of breeding-driven directional expression changes on wheat adaptation and improvement.