Metagenome-enabled Models Improve Genomic Predictive Ability and Identification of Herbivory-limiting Genes in Sweetpotato

Plant-insect interactions are often influenced by host- or insect-associated metagenomic community members. The relative abundance of insects and the microbes that modulate their interactions were obtained from sweetpotato () leaf-associated metagenomes using quantitative reduced representation sequencing and strain/species-level profiling with the Qmatey software. Positive correlations were found between whitefly () and its endosymbionts (, , and spp) and negative correlations with nitrogen-fixing bacteria that implicate nitric oxide in sweetpotato-whitefly interaction. Genome-wide associations using 252 975 dosage-based markers, and metagenomes as a covariate to reduce false positive rates, implicated ethylene and cell wall modification in sweetpotato-whitefly interaction. The predictive abilities (PA) for whitefly and abundance were high in both populations (68%-69% and 33.3%-35.8%, respectively) and 69.9% for . The metagBLUP (gBLUP) prediction model, which fits the background metagenome-based Cao dissimilarity matrix instead of the marker-based relationship matrix (G-matrix), revealed moderate PA (35.3%-49.1%) except for (3%-10.1%). A significant gain in PA after modeling the metagenome as a covariate (gGBLUP, ≤11%) confirms quantification accuracy and that the metagenome modulates phenotypic expression and might account for the missing heritability problem. Significant gains in PA were also revealed after fitting allele dosage (≤17.4%) and dominance effects (≤4.6%). Pseudo-diploidized genotype data underperformed for dominance models. Including segregation-distorted loci (SDL) increased PA by 6%-17.1%, suggesting that traits associated with fitness cost might benefit from the inclusion of SDL. Our findings confirm the holobiont theory of host-metagenome co-evolution and underscore its potential for breeding within the context of G × G × E interactions.