Phenotypic plasticity allows a single genotype to maintain its fitness across different environments. This facilitates colonisation of new niches but can be further refined or even lost as lineages diversify. The toxic Heliconiini butterflies have biochemical plasticity: they either sequester their cyanogenic glucosides (CG) from their larval hostplant or biosynthesize them when compounds for sequestration are not available. Here, we trace the evolution of CG biosynthesis in Heliconiini butterflies, a fundamental component of this biochemical plasticity. We first reconstructed the evolutionary history of biochemical plasticity in Heliconiini using chemical data from over 700 individuals, demonstrating that plasticity was ancestral in the tribe but subsequently lost in a few clades, such as the Sapho clade specialized in CG sequestration. In lepidopterans, CG biosynthesis has previously been characterized in the moth Zygaena filipendulae, as the genes CYP405A2, CYP332A3, and UGT33A1. Thus, we CRISPR-edited CYP405 in Heliconius erato, and confirmed that CYP405-knockout caterpillars do not biosynthesize CGs. We identified the CYP405As and CYP332As in other lepidopterans and found that both genes were independently co-opted into CG biosynthesis in the Heliconiinae butterflies and Zygaena moths. While most lepidopterans have a CYP332A, CYP405A is mostly restricted to butterflies and has been duplicated in all Heliconius species. Although several CYP405A copies were found in the Sapho clade, most of them lack structurally important P450 domains, which explains the loss of biochemical plasticity via specialization in CG sequestration. This represents one of the few examples of plasticity-first evolution in which the genetic mechanisms associated with its refinement are known