Genetic control of temperament traits across species: association of autism spectrum disorder risk genes with cattle temperament

Background Temperament traits are of high importance across species. In humans, temperament or personality traits correlate with psychological traits and psychiatric disorders. In cattle, they impact animal welfare, product quality and human safety, and are therefore of direct commercial importance. We hypothesized that genetic factors that contribute to variation in temperament among individuals within a species will be shared between humans and cattle. Using imputed whole-genome sequence data from 9223 beef cattle from three cohorts, a series of genome-wide association studies was undertaken on cattle flight time, a temperament phenotype measured as the time taken for an animal to cover a short-fixed distance after release from an enclosure. We also investigated the association of cattle temperament with polymorphisms in bovine orthologs of risk genes for neuroticism, schizophrenia, autism spectrum disorders (ASD), and developmental delay disorders in humans. Results Variants with the strongest associations were located in the bovine orthologous region that is involved in several behavioural and cognitive disorders in humans. These variants were also partially validated in independent cattle cohorts. Genes in these regions (BARHL2, NDN, SNRPN, MAGEL2, ABCA12, KIFAP3, TOPAZ1, FZD3, UBE3A, and GABRA5) were enriched for the GO term neuron migration and were differentially expressed in brain and pituitary tissues in humans. Moreover, variants within 100 kb of ASD susceptibility genes were associated with cattle temperament and explained 6.5% of the total additive genetic variance in the largest cattle cohort. The ASD genes with the most significant associations were GABRB3 and CUL3. Using the same 100 kb window, a weak association was found with polymorphisms in schizophrenia risk genes and no association with polymorphisms in neuroticism and developmental delay disorders risk genes. Conclusions Our analysis showed that genes identified in a meta-analysis of cattle temperament contribute to neuron development functions and are differentially expressed in human brain tissues. Furthermore, some ASD susceptibility genes are associated with cattle temperament. These findings provide evidence that genetic control of temperament might be shared between humans and cattle and highlight the potential for future analyses to leverage results between species.


Irish national breeding program
Docility is scored subjective in both commercial and seedstock beef herds by trained classifiers contracted to the Irish Cattle Breeding Federation (DOYLE et al. 2018). Scoring is carried out on a linear scale from 1 to 10 where 1 represents aggressive and 10 represents docile. Data on docility were available on 3,356 purebred Angus (AA), 31,049 Charolais (CH), 3,004 Hereford (HE), 35,159 Limousin (LM), and 8,632 Simmental (SI) beef cattle scored between the years 2000 and 2016. All animals were scored between 6 and 16 months of age with a recorded sire and dam. Contemporary group was defined as herd-by-scoring date generated separately per breed. Each contemporary group had to have at least five records.

Generation of adjusted phenotypes
Prior to inclusion in the analysis, all beef cattle phenotypes were adjusted within breed in ASREML (GILMOUR et al. 2009) using the model: where y is the linear type trait, HSD is the fixed effect of herd by scoring date (11,130 levels), Sex is the sex of the animal (male or female), AM is the fixed effect of the age in months of the animal (11 classes from 6 to 16 months), DP is the fixed effect of the parity of the dam (1, 2, 3, 4 and ≥5), animal is the random additive effect of the animal, and is the random residual effect. The adjusted phenotype was the raw phenotype less the fixed effect solutions of HSD, sex, AM, and DP.

Genotype data
Animals were genotyped using either the Bovine Illumina SNP50, the Illumina High Density (HD), the Illumina 3k panel, the Illumina Low Density genotyping panel or a bespoke genotype panel (IDB) developed in Ireland (MULLEN et al. 2013). Each animal had a call rate ≥ 90% and only autosomal SNPs, SNPs with a known chromosome position on UMD 3.1, and SNPs with a call rate ≥ 90% within panel were retained for imputation.
All genotyped animals were first imputed to HD using FImpute2 (SARGOLZAEI et al. 2014) and then to whole genome sequence (WGS) using Eagle (version 2.3.2) (LOH et al. Of the edited dataset, imputed whole genome sequence data existed for 1,444 AA, 6,433 CH, 1,129 HE, 8,745 LM, and 1,698 SI. Quality control edits were imposed on the imputed sequence genotypes within each of the 6 breeds separately; all SNPs with a minor allele frequency (MAF) ≤ 0.002 or that deviated Hardy-Weinberg equilibrium (p < 1x10 -6 ) were removed. Further refinement of the dataset of the 23,943 animals with genotype and phenotype information in the present study was undertaken by removing regions of low imputation accuracy, perhaps due to local mis-assemblies or mis-orientated contigs. These regions were identified using an additional dataset of 147,309 verified parent-progeny relationships as described by PURFIELD et al. (2019). Following all SNP edits, 15,531,563, 16,110,951, 15,119,634, 15,336,651, 16,969,734 autosomal SNPs remained for analysis in the AA, CH, HE, LM, and SI populations, respectively.

Association analyses
The association analyses were performed within each breed separately using a linear mixed model in GCTA (YANG et al. 2011). Autosomal SNPs from the original HD density panel (i.e., 734,159 SNPs) were used to construct the genomic relationship matrix (GRM). The model used for the within-breed analysis was = + + + where y is a vector of preadjusted phenotypes, µ is the overall mean, x is the vector of imputed genotypes, b is the additive fixed effect of the candidate SNP to be tested for association, g ~ N(0, < = ) is the vector of additive genetic effects, where G is the genomic relationship matrix calculated from the HD SNP genotypes, and < = is the additive genetic variance, and ~ N(0,I = ) is the vector of random residual effects, and = is the residual variance.

Enrichment of ASD genes in bovine brain tissue
We used RNA-seq data from several cattle tissues to investigate tissue specificity of expression of ASD genes in cattle. A bovine gene expression atlas (CHAMBERLAIN et al. 2015) contains differential expression (DE) for 18 cattle tissues including brain, white blood, mammary and several others from one lactating Holstein dairy cow (3 technical replicates for each tissue type). For each tissue, we used the number of DE genes in bovine orthologous ASD genes (already mapped for enrichment in cattle temperament) and the overall number of DE genes to calculate log odds-ratio of up and down-regulation. We found that ASD genes were significantly more likely to be expressed in cerebellum and caudal lobe, exhibiting a positive log fold change in expression, when compared to all protein-coding bovine genes ( Figure S2).
We also combined both brain tissues to run a c2 test of independence and found that ASD genes are not a random sample of all bovine protein-coding genes as they displayed higher upregulation levels (P=7.2 x 10 -5 in Table S10).