نوع مقاله : مقاله پژوهشی
عنوان مقاله English
نویسندگان English
Extended Abstract
Introduction: Feed efficiency and enteric methane emission are economically and environmentally critical traits in dairy cattle, yet both are difficult and expensive to record and have low to moderate heritability, which limits the accuracy of conventional genomic prediction based on the host genome alone (GBLUP). A growing body of evidence shows that the rumen microbiome is partly shaped by host genetics and contributes substantially to feed efficiency and methanogenesis, giving rise to the concept of microbiability — the proportion of phenotypic variance explained by microbiome composition. This raises a timely holo-omic question: can integrating host-genomic and rumen-microbiome information improve the prediction of these traits, and does the value of the microbiome depend on the trait's microbiability and on whether the goal is predicting the animal's own performance or its transmissible breeding value? The objective of this study was to address these questions under controlled conditions in which the genetic and microbial architectures are known.
Materials and Methods: The stochastic simulation was developed for a dairy cattle population of 800 animals. Rather than assuming prediction accuracy, a host genome of 2500 independent single-nucleotide polymorphisms (minor-allele frequencies drawn from U(0.05, 0.50); genotypes under Hardy–Weinberg equilibrium) was simulated in Python 3.12 (NumPy 2.4, SciPy 1.17). Host additive genetic values were drawn directly from N(0, Gσ²g), so no historical population, linkage disequilibrium or discrete QTL were involved. Standardized abundances of 150 rumen taxa were generated from a normal distribution; for each taxon, 30% of the variance was controlled by 500 randomly chosen host SNPs and 70% was environmental (microbiome heritability 0.30). A genomic relationship matrix (G) and an analogous microbiome relationship matrix (B) were constructed. Microbiability was defined as m² = σ²m/σ²P, the proportion of phenotypic variance explained by the microbiome effect m ~ N(0, Bσ²m); σ²m covers both the host-controlled and the environmental part of the microbiome and is not an additive genetic variance. Two traits with contrasting architectures were modeled: residual feed intake (RFI; heritability 0.25, microbiability 0.15) and methane emission (CH4; heritability 0.12, microbiability 0.30). Three models were compared by repeated five-fold cross-validation, with variance components estimated by REML: GBLUP (genome only), microbiome BLUP (MBLUP, microbiome only) and a combined model (GMBLUP, genome plus microbiome). Crucially, predictive accuracy was evaluated against two distinct targets: (i) the animal's own genetic-plus-microbiome merit (g + m), relevant to predicting its own performance; and (ii) the transmissible breeding value — the heritable component passed to offspring — relevant to genetic selection. The sensitivity of model performance to trait microbiability was assessed by varying it from 0.05 to 0.45 at a fixed heritability of 0.20, and a genome-by-microbiome interaction term gm ~ N(0, (G∘B)σ²gm), with σ²gm = 0, 0.05, 0.10 and 0.15, was introduced to test whether additive models capture non-additive merit. Models were compared by paired t-tests across replicates with Bonferroni correction.
Results and Discussion: For predicting the animal's own merit (g + m), integrating the microbiome markedly improved accuracy (mean over 30 replicates; MSE reported alongside accuracy in Table 2). For methane — a highly microbiome-driven trait — own-merit accuracy rose from 0.17 ± 0.05 with GBLUP (MSE 0.41) to 0.67 ± 0.03 with MBLUP (MSE 0.24) and 0.68 ± 0.03 with GMBLUP (MSE 0.23); for residual feed intake, the combined model (0.45, MSE 0.32) outperformed both genome-only (0.22, MSE 0.38) and microbiome-only (0.40, MSE 0.35) models (all P < 0.001). For the transmissible breeding value, GBLUP and GMBLUP performed comparably (RFI: 0.24 vs 0.25, P > 0.05; CH4: 0.18 vs 0.20, P = 0.02), while MBLUP alone was distinctly less accurate for this target (RFI 0.02, CH4 0.03), consistent with most of the microbiome signal not being heritably transmitted to offspring. The sensitivity analysis (Table 3) and the interaction test (Table 4). The sensitivity analysis confirmed that the gain from adding the microbiome to own-merit prediction grows sharply with microbiability (from +0.04 to +0.37), whereas the gain for the breeding value is positive but much smaller and grows only modestly (from 0.00 to +0.11). The interaction test confirmed that a purely additive model tracks the additive value well regardless of interaction strength, while its correlation with total (additive + interaction) merit declines as the genome-by-microbiome interaction grows (gap rising from 0 to about 0.09). Thus, the microbiome is highly informative for an animal's own performance, and adding it did not hurt breeding-value prediction. However its contribution there remained small relative to MBLUP's own-merit gains. These patterns are broadly consistent with reports that the holobiont and microbiome contribute substantially to methane and feed-efficiency phenotypes, while underscoring the distinction between predicting phenotype and predicting heritable, transmissible merit.
Conclusion: Integrating rumen-microbiome information with the host genome substantially improves prediction of an animal's own feed efficiency and, especially, methane emission, and this benefit scales directly with the trait's microbiability. For the transmissible breeding value, genome-only (GBLUP) and combined (GMBLUP) prediction performed comparably in this reimplementation, whereas microbiome-only prediction (MBLUP) was clearly less accurate for this target — consistent with most of the microbiome signal not being transmitted to offspring, even though adding it to the genome did not measurably reduce accuracy here. Microbiome data are therefore most valuable for predicting an animal's own performance — supporting management, feeding and culling decisions, and on-farm phenotyping of strongly microbiome-driven traits such as methane — while selection on the host genome remains the practical basis for permanent, cumulative genetic improvement. A holo-omic strategy should distinguish these two objectives: microbiome-informed prediction for managing low-methane, efficient cattle, and the (partly heritable) microbiome component together with the host genome for long-term genetic selection.
کلیدواژهها English
