MUN, the concentration of urea in milk, serves as a practical window into how efficiently a cow converts dietary nitrogen into milk protein. Dairy cattle typically convert only about 25 percent of ingested nitrogen into milk, with the rest excreted in urine and feces, contributing to ammonia volatilization, nitrate leaching, and nitrous oxide emissions. Because MUN is routinely measured in milk recording programs at no additional cost, it has become an attractive target for both management interventions and genetic improvement aimed at reducing the environmental footprint of dairy farming.

The research team, led by Daniel George Bratu and colleagues at the University of Life Sciences “Regele Mihai I” from Timișoara, genotyped 406 cows—324 Romanian Simmental, 57 Brown Swiss, and 25 Holstein—at 42,440 single nucleotide polymorphisms (SNPs) after quality control. The animals contributed 13,722 test-day MUN records, with a median of 25 records per cow and a maximum of 147. The cows came from two farms: the Research and Development Station for Bovine in Arad, which housed 347 animals, and a private farm in Timiș County with 59 cows.

Under a well-calibrated genomic mixed model with a genomic inflation factor of 1.034, no SNP reached genome-wide significance, and three suggestive signals failed to survive false discovery rate control, with a minimum q-value of 0.723. The number of suggestive leads matched almost exactly what would be expected by chance alone—42,440 SNPs multiplied by the suggestive threshold of 7.149 × 10⁻⁵ yields approximately 3.0 expected hits. The three suggestive SNPs mapped near DGKI on chromosome 4, PLPP3 on chromosome 3, and OXR1 on chromosome 14, none overlapping previously reported MUN quantitative trait loci.

The most striking result, however, was not what the scan found but what it un-found. Under a conditional-residual phenotype—the default output of standard mixed-model software—the SNP AX-124375815 on chromosome 18 near ZNF423 reached genome-wide significance at P = 4.42 × 10⁻⁷. But when the researchers switched to a marginal adjusted mean phenotype, holding the covariate model fixed, the signal collapsed to P = 0.060. A factorial experiment crossing phenotype definition with covariate specification revealed that the phenotype switch alone drove the collapse, acting through standard-error inflation rather than any change in the effect size, which remained stable at approximately 0.35 to 0.41 across all four analytical combinations.

The mechanism is subtle but consequential. In a two-step genome-wide association study, repeated per-cow measurements are typically condensed into a single value through a mixed model. The conditional residual implicitly subtracts the estimated, shrunken animal effect—the best linear unbiased prediction, or BLUP—in addition to fixed effects. Because shrinkage increases with the number of records, cows with many records are pulled strongly toward the population mean, while cows with few records retain most of their apparent deviation. The adjusted phenotype then encodes record count, a purely technical quantity, alongside biology. In this dataset, the between-cow standard deviation of the conditional phenotype fell from approximately 2.76 mg/dL among cows with two to three records to approximately 0.17 mg/dL among cows with eleven or more records—a compression of roughly 2.6-fold relative to the corrected definition.

The researchers demonstrated the artifact’s signature directly: the Spearman correlation between record count and the absolute deviation of the per-cow phenotype from the farm mean was −0.458 under the conditional definition (P = 2.0 × 10⁻²²) but only +0.025 under the marginal definition (P = 0.62). This diagnostic, computable within any farm to avoid confounding recording depth with herd effects, offers a portable safeguard for other small two-step livestock studies. The artifact is most severe when record counts vary widely, the trait is at least moderately repeatable, and the per-animal phenotype is taken as a mean of conditional residuals—conditions that coincide frequently in smallholder and local-breed datasets.

Beyond the phenotype-definition lesson, the study confronted an irreducible structural confound: breed and farm were inseparable, with Brown Swiss confined to one farm and Holstein to the other. Farm alone explained 84.7 percent of the variance in the corrected phenotype, and within Simmental—the only breed present at both farms—farm still accounted for 80.8 percent of the variance. The chromosome 18 minor allele was concentrated in the single-farm breeds, so no model could fully separate a between-farm allele-frequency difference from a breed or management effect. A within-Simmental analysis, the primary deconfounding check, reduced the signal to P = 0.795 with the effect direction reversing.

Narrow-sense heritability could not be separately identified in this mostly unrelated cohort, as the genomic relationship matrix showed a mean off-diagonal of −0.0025, leaving additive-genetic and permanent-environment variances confounded. The researchers therefore treated repeatability as an upper bound for heritability, reporting an intraclass correlation of 0.243 in the full recording cohort of 954 cows and 0.388 in the matched genotyped subset. A formal power analysis showed the scan had 80 percent power only for a locus explaining approximately 6.9 percent of phenotypic variance at the genome-wide threshold, while typical MUN loci explain only 2 to 3 percent.

The study also tested whether established MUN loci from larger populations—55 reference loci compiled from prior studies, including the casein cluster, PAEP, and DGAT1—showed any signal in this cohort. None did, whether tested per locus, by region, or as an aggregate genomic partition, consistent with the cohort being roughly 50 to 85 times smaller than the discovery studies. The researchers frame the environmental relevance of MUN around nutrition and management rather than genomic selection: each 1-percentage-point change in dietary crude protein shifts MUN by approximately 1.1 mg/dL, and a 1 mg/dL reduction in population-mean MUN corresponds to approximately 15.5 grams per day per cow of avoided urinary nitrogen at a body weight of 600 kilograms. The study delivers the first MUN baseline for Romanian dual-purpose cattle, a rigorous null scan, and a demonstrated caution regarding phenotype definition that extends well beyond this population.

Subject of Research: Genetic architecture and repeatability of milk urea nitrogen in Romanian dual-purpose cattle and the impact of phenotype definition on genome-wide association results

Article Title: Repeatability and genetic architecture of milk urea nitrogen in a multi-breed Romanian dual-purpose cattle population: a genome-wide association study and the impact of phenotype definition

Article References: Bratu, D. G., Blaga-Voicu, Ş., Zanfira, B., Mircu, C., Spătaru, I. I., Torda, I., Cocioba, V. M., Mizeranschi, A. E., Ilie, D. E., Cziszter, L. T., Vizitiu, D. A., Boldura, O. M., & Huţu, I. (2026). Repeatability and genetic architecture of milk urea nitrogen in a multi-breed Romanian dual-purpose cattle population: a genome-wide association study and the impact of phenotype definition. Frontiers in Animal Science, 7, Article 1937481. https://doi.org/10.3389/fanim.2026.1937481

Image Credits: AI Generated

DOI: 10.3389/fanim.2026.1937481

Keywords: milk urea nitrogen, genome-wide association study, dual-purpose cattle, phenotype definition, Romanian Simmental, genomic relationship matrix, repeatability, nitrogen efficiency, ZNF423, BLUP shrinkage, livestock genetics, population stratification