Influence of model selection and data structure on the estimation of genetic parameters in honeybee populations

Abstract

Estimating genetic parameters of quantitative traits is a prerequisite for animal breeding. In honeybees, the genetic variance separates into queen and worker effects. However, under data paucity, parameter estimations that account for this peculiarity often yield implausible results. Consequently, simplified models which attribute all genetic contributions to either the queen (queen model) or the workers (worker model) are often used to estimate variance components in honeybees. However, the causes for estimations with the complete model (colony model) to fail and the consequences of simplified models for variance estimates are little understood. We newly developed the necessary theory to compare parameter estimates that were achieved by the colony model with those of the queen and worker models. Furthermore, we performed computer simulations to quantify the influence of model choice, estimation algorithm, true genetic parameters, rates of controlled mating, apiary sizes, and phenotype data completeness on the success of genetic parameter estimations. We found that successful estimations with the colony model were only possible if at least some of the queens mated controlledly on mating stations. In that case, estimates were largely unbiased if more than 20% of the colonies had phenotype records. The simplified queen and worker models proved more stable and yielded plausible parameter estimates for almost all settings. Results obtained from these models were unbiased when mating was uncontrolled, but with controlled mating, the simplified models consistently overestimated heritabilities. This work elucidates the requirements for variance component estimation in honeybees and provides the theoretical groundwork for simplified honeybee models.

Sequence-based genome-wide association studies reveal the polygenic architecture ofVarroa destructorresistance in Western honey beesApis mellifera

“…In some countries, such as Germany, beekeepers are grouped in breeding organisations, that make extensive use of artificial insemination to track their mating making the construction of breeding schemes easier (Lorenz 2016). In this context, some studies (M. Du et al 2022a;M. Du et al 2022b;Hoppe et al 2022;Uzunov et al 2022;Bernstein, M. Du, Z. G. Du, et al 2023) described the statistical models and sampling strategies that can be successfully applied to implement genomic selection in honey bees.…”

“…In this context, some studies ( [65], [66], [67], [68], [69]) described the statistical models and sampling strategies that can be successfully applied to implement genomic selection in honey bees. In contrast in other countries, as in France, there is limited consensus between beekeepers on breeding practices, hence making selection schemes for beekeeping difficult to describe and manage.…”

Consequences of incorrect genetic parameter estimates for single‐trait and multi‐trait genetic evaluations in honeybees

“…How to cite this article: Du, M., Bernstein, R., Hoppe, A., & Bienefeld, K. (2022). Consequences of incorrect genetic parameter estimates for single-trait and multi-trait genetic evaluations in honeybees.…”

“…Thus, the genetic impact is split into a maternal queen effect and a direct worker group effect, much like the separation into maternal and direct effects in other agricultural species (Willham, 1963). The separation of genetic variance into two effects complicates the accurate estimation of honeybee‐specific trait parameters substantially (Andonov et al, 2019; Du et al, 2022; Guichard et al, 2020; Willam & Eßl, 1993). The imprecise paternal pedigree information due to the honeybee queen's mating behaviour with multiple drones from different colonies in mid‐air further impedes precise parameter estimations (Andonov et al, 2019; Du et al, 2022; Zakour et al, 2012).…”

“…The separation of genetic variance into two effects complicates the accurate estimation of honeybee‐specific trait parameters substantially (Andonov et al, 2019; Du et al, 2022; Guichard et al, 2020; Willam & Eßl, 1993). The imprecise paternal pedigree information due to the honeybee queen's mating behaviour with multiple drones from different colonies in mid‐air further impedes precise parameter estimations (Andonov et al, 2019; Du et al, 2022; Zakour et al, 2012). This results in a high probability of erroneous genetic parameters as input data for the BLUP procedure—even more so, when data paucity forces incipient breeding programmes to adopt genetic parameters that were estimated in other populations (Uzunov et al, 2022).…”