Machine learning-based gait classification and genome-wide association identify a QTL for gait type in Colombian paso horses.

Coordinated mammalian locomotion relies on spinal circuits where DMRT3 regulates strides and alternative gaits. However, DMRT3 does not explain differences between the Colombian paso horse breed's specialized gaits, Colombian trocha and Colombian trot. We used inertial sensors and machine learning to develop accurate phenotyping (n = 225 horses), before performing genome-wide association analysis (n = 85 horses, 670K array). We identified a 2.43 Mb quantitative trait locus (QTL) on ECA16, where haplotypes featuring lead variants (rs1147402472, p = 1.95 × 10-8; rs1136628503, p = 8.52 × 10-8) explained 48.6% of gait variance (p < 0.001). The QTL contained 11 genes linked to neurodevelopment and muscle regulation, including LHFPL4, SRGAP3, and ATP2B2. Interaction analyses suggested a functional link between the latter genes. rs1147402472-C was common across diverse gaited horse breeds, but the Colombian trot-specific haplotype (rs1147402472-C/rs1136628503-T) was absent elsewhere. These findings demonstrate that fine-scale genetic differentiation at ECA16 underlies neural adaptations distinguishing complex locomotor traits and highlight the power of AI-assisted phenotyping in genomics.