Leipart, V.; Amdam, G. V.; O'Brien, S.; Pigott, E.; Dodds, G.; Ihle, K.

Managed honey bees (Hymenoptera: Apidae: Apis mellifera L) have experienced unsustainably high rates of annual loss driven by several interacting factors, most notably pests, pathogens, pesticides, and poor nutrition. Breeding bee stocks that can cope with these challenges is a priority. Advanced molecular methods (marker-assisted selection, MAS) have enhanced the breeding efficiency of domesticated animals in recent years, but have not contributed strongly to honey bee stock improvements. This is largely because desirable traits of bees usually emerge from collective phenotypes of workers (sterile females) instead of from the breeding individuals (queens and male drones). For collective phenotypes, single genes typically have small, additive effects, so identifying impactful MAS targets is challenging. Here, we provide proof of concept for a new approach to honey bee breeding through MAS using the multifunctional protein Vitellogenin (Vg), a protein known to interact with and mitigate the primary drivers of colony loss. Our pipeline leverages cutting-edge, artificial intelligence (AI)-driven protein structure modeling algorithms to predict the effects of genetic variants of Vg on relevant molecular functions including lipid, zinc, and DNA binding. Following the AI-powered Vg variant selection step, we use a combination of standard apicultural techniques and DNA sequencing validation to breed honey bee queens homozygous for the desirable Vg allele. Our protocol can kick-start a new area of modernized bee breeding: an AI-enhanced MAS system that allows cost-effective and nimble development of stocks to meet urgent and long-term needs of stakeholders.