Gene variants in the pheromone vomeronasal receptors and QTLs around behavioral and fat metabolism genes associated with altered feed efficiency in cattle

Abstract

Improving feed efficiency in cattle is increasingly important for both environmental and economic reasons. Although feed efficiency traits are under considerable genetic control, with an average moderate heritability estimate of 0.33, genetic evaluations are limited by the difficulties in measuring feed intake and the lack of records from most commercial herds. Most genetic evaluations rely on small numbers of records from research farms, resulting in under-represented genetic variation and pronounced sampling errors in heritability estimates. To enhance the discovery of genetic mechanisms underlying feed efficiency and to address measurement limitations and the under-representation of genetic variation, we used joint phenotypic and genotypic measurements from two distinct herds for GWAS and in-depth genomic analysis. By applying this approach, our exploratory analysis discovered fourteen significant markers with effects on residual feed intake (RFI) ranging from -1.41 to 1.44 kg/day. Quantitative trait loci (QTLs) enrichment analysis specifically pointed to traits that contributed to RFI, including dry matter intake (DMI), body weight (BW), and protein yield. Gene enrichment analysis, which was largely biased by a local cluster of vomeronasal receptor genes within a single ~ 500 kb region on BTA18, suggested three sets of genes of interest: a vomeronasal pheromone receptor cluster (VN1R1 and four additional response to pheromone genes on BTA18), genes linked to social and behavioral responses (EPC2 on BTA2; SYN3 on BTA5), and fat metabolism-related genes (KIF5C on BTA2; SV2B on BTA21). Of these candidate genes, likely functional amino acid (AA) variations were observed in the VN1R1 putative protein (314 AA) after screening a sample of 27 Israeli Holstein genomes. These functional variations included two truncation mutations that could encode 89 and 239 AA polypeptides. Consistent with these findings, whole-genome sequence data analysis of RFI-characterized Irish bulls identified a significant association between the 89 AA truncation and high RFI, further validating our results and indicating that although such variation was common, the presence of an intact VN1R1 receptor was associated with a beneficial effect on feed efficiency. Moreover, the 89 AA truncation was observed in diverse cattle breeds, including American, Israeli, Irish, and New Zealand Holstein. These findings are compatible with feed efficiency, a complex trait governed by neural (behavioral) and metabolic components. Further characterization of these factors would allow genetic selection to reduce feed costs and environmental footprints.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-37314-3.

Introduction

The dairy industry significantly contributes to the United States of America (US) and Israel (IL) agriculture, but feeding costs can account for 50–70% of dairy cattle production expenses¹. Inefficient feeding results in losses of nutrients to the environment due to poor conversion of nutrients into usable metabolites². Studies in beef and dairy cattle have consistently reported correlations between feed efficiency (FE) and FE-related traits, including methane emissions, with high FE animals producing significantly less methane³. This association underscores the dual economic and environmental benefits of improving feed efficiency through genetic selection, as it simultaneously enhances farm profitability by increasing feed nutrient conversion and mitigates losses of nutrients to the environment from livestock production systems⁴. Despite the benefits, selecting FE is challenging due to the need for extensive phenotypic records, which are costly and complex to measure^5–7. Consequently, FE is primarily studied in research herds and select farms, with residual feed intake (RFI) serving as an FE estimate. The RFI is defined as the difference between an animal’s actual feed intake and its expected feed intake after accounting for main energy sinks such as body weight (BW) and BW variation and energy-corrected milk (ECM). Thus, cows with negative RFI values consume less feed than expected for their production and body maintenance and are considered more feed-efficient, whereas cows with positive RFI values consume more feed than expected and are regarded as less efficient. Compared to simple FE measurements, RFI allows for improvement of FE that is largely independent of ECM production⁸. Recent meta-analysis of FE in Holstein cattle reported a wide range of heritabilities, ranging from 0.01 to 0.6, depending on the FE measurement type and experiment design⁷, with a moderate average heritability of 0.33^9,10.

Researchers have examined the genetic factors affecting FE in Holstein and Jersey cows across DNA, RNA expression, enzyme proteins, metabolism, and gut microbiome levels^2,10. Genome-wide association studies (GWAS) have identified genomic regions linked to FE traits, though these studies often use small research populations with limited genetic variation and statistical power¹¹. A recent meta-analysis reported 13 quantitative trait loci (QTLs) and their associated genes, including those related to olfaction and fat metabolism. The study also reported that candidate gene enrichment indicated the oxidative phosphorylation (OXPHOS) pathway⁷. The authors concluded that integrating data from multiple studies and conducting meta-analysis can overcome limitations in FE studies, such as insufficient population size⁷.

To improve genomic selection for FE, a USA (US) collaboration created a 6,000-cow dataset, translating to the Holstein Association’s Total Performance and the US Net Merit breeding indexes¹⁰. However, improvements remain limited due to the small number of phenotypic records, constrained genetic variation, low-to-intermediate heritability in cattle, and epigenetic effects¹⁰. It should be noted that the 488 US cow sample used in this study was part of this large dataset, whereas previous results for the 192 Israel (IL) cow sample required reevaluation³⁰.

Therefore, the objective of this study was to leverage the increased genetic variation from a combined US and Israeli Holstein dataset to (1) identify novel genomic regions associated with RFI using a genome-wide association study, and (2) perform functional annotation and gene enrichment analyses to elucidate the biological pathways underlying feed efficiency.

Results

Population structure and GWAS analyses

GWAS for RFI failed to obtain meaningful and significant results within the cow populations of each of the two research stations, likely due to the sample size and/or low genetic variation among the cows in each farm (Fig. 1). For instance, the computed genomic Restricted Maximum Likelihood (REML) for the pseudo-heritability in the IL farm indicated that genetic variation explains less than 0.1% of the RFI phenotypic variation (genetic variance < 0.01; phenotypic variance = 4.82). Therefore, our first objective was to combine the IL and US farm data to increase genetic variation, population size, and the likelihood of achieving results of biological significance.

Fig. 1.

Multidimensional scaling (MDS) plot of the studied population structure. MDS analysis was performed on the computed relationship matrix of the genome-wide pairwise distances generated using the identical by state (IBS) method. The projection of the first two dimensions is shown.

The multidimensional scaling (MDS) plot based on the identical by state (IBS) matrix showed two distinct, non-overlapping clusters corresponding to the IL and US herds (Fig. 1), confirming the distinct genetic backgrounds of the two populations. The Manhattan plot for the GWAS results for RFI is given in Fig. 2. Carrying out genomic REML (gREML) analysis, the EMMAX Pseudo-heritability by combining the data from both herds was 0.45, suggesting 45% of the variance among the cows’ RFI scores could be explained by considering all SNPs. After accounting for genetic relationships and for multiple tests, using GWAS, we found 14 significant markers with adjusted p-values ≤ 0.05 (Table 1).

Fig. 2.

Genome-wide association study (GWAS) for residual feed intake (RFI). (a) Manhattan plot of the RFI markers from the combined US and IL Holstein population. The chromosomal positions of 42,639 informative SNP markers were annotated on the ARS-UCD1.2 genome assembly (x-axis), and their corresponding nominal -log10 P-values for association with RFI were calculated using the EMMAX software³⁴ and adjusted for multiple testing by the Benjamini–Hochberg procedure (y-axis). The red horizontal line indicates the genome-wide adjusted significance threshold of p = 0.05 after correction for multiple testing. (b) Quantile–Quantile plot for RFI GWAS of the observed nominal –log10(p) values (y-axis) versus the expected –log10(p) values (x-axis), with a genomic inflation factor λ of 0.966.

Table 1.

Single-nucleotide polymorphisms (SNPs) that were significantly associated with residual feed intake (RFI) in a genome-wide association study (GWAS). 1: SNPs are sorted in ascending order according to the beta value, which represents the effect size. 2: Marker coordinates (chromosomal position in base pairs) are according to the bovine ARS-UCD1.2 assembly. 3: Substitution effect in units of the cows’ RFI scores (Kg/day). 4: The EMMAX GWAS nominal p-value. 5: The adjusted p-value, corrected using the Benjamini–Hochberg method for multiple testing.

CHR	SNP1	BP2	BETA3	SE	p4	p.adj5
21	Hapmap40073-BTA-53493	156,017,35	-1.42	0.30	3.20E-06	1.97E-02
1	BTB-01776422	104,314,388	-1.21	0.19	7.39E-10	1.90E-05
19	ARS-BFGL-NGS-116522	1,842,109	-1.06	0.24	1.52E-05	4.69E-02
5	ARS-BFGL-NGS-4826	118,861,762	-1.03	0.22	4.18E-06	2.12E-02
5	ARS-BFGL-NGS-112180	71,197,728	-0.93	0.20	2.52E-06	1.81E-02
9	Hapmap47111-BTA-83104	25,037,642	-0.91	0.21	1.49E-05	4.69E-02
15	ARS-BFGL-NGS-8859	3,952,769	-0.85	0.19	7.43E-06	2.92E-02
15	ARS-BFGL-NGS-109478	2,989,352	-0.82	0.17	1.94E-06	1.68E-02
6	BTB-00281962	110,562,628	-0.76	0.17	7.30E-06	2.92E-02
21	BTB-01644606	51,856,354	0.75	0.15	7.05E-07	8.21E-03
2	ARS-BFGL-NGS-83687	47,258,655	0.90	0.15	8.80E-10	1.90E-05
18	BTB-00730867	58,285,064	1.12	0.22	7.61E-07	8.21E-03
14	ARS-BFGL-NGS-100885	65,876,176	1.22	0.26	4.42E-06	2.12E-02
10	BTB-01660086	56,987,932	1.45	0.33	1.19E-05	4.28E-02

Gene enrichment analyses

Next, we performed enrichment analyses for quantitative trait loci (QTLs) and gene pathways to further assess the possible roles of the identified significant markers in RFI. Previously, we found that about 90% of the marker pairs with r² > 0.9 (denoting the markers are in a tight linkage) are within ~ 200 kb distance from each other¹². Thus, for the QTL enrichment analysis, we defined genomic windows of ± 250 kb around each of the 14 lead SNPs, and we used the GALLO R package¹³ to intersect these regions with the cattle QTLdb. QTLs were counted by trait category within these windows, and a hypergeometric test was applied to evaluate whether particular QTL types (for example, milk protein yield) were over-represented compared to their genome-wide frequency.

Interestingly, QTL enrichment analysis found significant RFI markers enriched with protein and fat percentage and yield, BW, average daily gain, and dry matter intake (DMI), the components that compose RFI (Fig. 3). Additional enriched QTLs were associated with fertility, herd life, and health.

Fig. 3.

Bubble plot presenting the enrichment of previously reported QTLs with the identified RFI significant markers. The color denotes the enrichment-adjusted p-value (adj.pval) score, and the radius of the bubbles is proportional to the number of QTLs (N_QTLs). The y-axis represents the quantitative traits, and the x-axis denotes the affiliation of the QTL with the major trait class.

Gene enrichment analysis of the genes spanning the significant markers for RFI revealed significant enrichment of the pheromone vomeronasal receptor gene family (VN1Rs and V1Rs), all clustered in one locus spanning the BTB-00730867 SNP at BTA18 (Table 2). Previous evidence supports the role of pheromones and odorants in regulating mammalian appetite¹⁴, and a study of food-deprived male rats observed variations in neuronal activity of the vomeronasal pathway¹⁵. However, all of the identified genes come from a single 500 kb surrounding the SNP BTB-00730867 on BTA 18, and therefore cannot be considered as independent events on which the enrichment statistic score should have been based, yet these receptors were the only genes adjacent to this SNP marker, and thus point to the pheromone receptor pathway and warrant further examination.

Table 2.

Gene enrichment analysis of top-scored Gene Ontology (GO) terms obtained by the GeneAnalytics algorithm¹⁹. 1: GeneAnalytics score × 1000 is dependent on the number of matched genes and their scores, whereas each gene in each entity has a matching score based on the combination of annotations of all matched genes in the entity, and the entity type. The match quality was assessed by the GeneAnalytics algorithm, and considering multiple tests, the corrected p-value is categorized as high (p ≤ 0.05), medium (0.05 < p ≤ 1), and low (p > 1) quality matches. 2: Ratio between the number of matched genes and the number of genes listed in the entity.

Score 1	Entity type	Entity	Ratio 2	Matched genes
41.6 (high)	GO term	Response to pheromone	5/8	VN1R1, VN1R2, VN1R3, VN1R4, VN1R5
7.5 (high)	Disease	Autism spectrum disorder	10/5301	SYN3, SV2B, RPL10A, POP1, PPP2R1A, EPC2, LRFN5, KIF5C, CASP1, CC2D2A
10.9 (medium)	Pathway	tRNA processing	3/107	TRMT11, POP1, RTCB

Significant enrichment was also found for Autism Spectrum Disorder (ASD, Table 2). Interestingly, Synapsin III (SYN3 on BTA5) and Enhancer Of Polycomb Homolog 2 (EPC2 on BTA2), which are associated with the ASD pathway, were reported to affect BW and skeletal muscle development, respectively, in Sheep, Cattle, and Yak^16–18. Thus, our results associated specific pheromone receptor variants and behavioral/cognitive genes with animal production efficiency, which is controlled by a complex interplay of factors, with appetite and growth being central drivers.

Analysis by genomic position

Besides these analyses, we investigate the possible role of the genes in the highest proximity to the SNP markers in FE. We found that the ARS-BFGL-NGS-83687 marker on BTA2 is located within the gene, Kinesin heavy chain isoform 5C (KIF5C). In the Angus breed, KIF5C was associated with the Marbling trait, which describes the presence of intramuscular fat²⁰. The Hapmap40073-BTA-53493 marker on BTA21 is located at 200 Kb upstream of Synaptic vesicle glycoprotein 2B (SV2B). In mice, the regulation of Sv2b expression was correlated with altered insulin and BW²¹.

Confirming the association of vomeronasal receptors with RFI

To further assess the association of the five candidate genes with RFI, we screened complete genomic variation data of 27 Israeli bulls (as described in the method section), filtering for likely functional variants (e.g., missense, stop-gain/loss, frameshift). This analysis identified two deletion variants in the VN1R1 gene’s Coding Sequence (CDS) at positions 137 and 662 (Table 3). Four additional missense variants in KIF5C, SV2B, were predicted to be tolerated by the SIFT algorithm²². We then turned to search data deposited in GenBank that combined whole-genome and RFI data for cattle, preferably Holstein. Only 14 Short Read Archive (SRA) Holstein genomes with RFI data were detectable using the terms “Holstein” and “RFI” in a text search of this database; all of them belonged to the BioProject of Irish bulls (PRJNA889458). This project included 28 bulls divergent for residual feed intake, half Charolais and half Holstein-Frisian breed, categorized as low or high RFI, thus seven bulls in each category, combining breed and RFI status.

Table 3.

Five VN1R1 functional variations. 1: Sequence positions are given for the chromosomal base-pair position (BTA18, build ARS-UCD2.0), for variation between the reference (REF) and the alternative (ALT) alleles within the mRNA coding sequence (CDS), and for the amino-acid (AA) substitution within its corresponding putative translation (Protein). 2: FS—indicates frameshift affecting several AAs.

Position1			Substitution		SIFT22
BTA18	CDS	Protein	REF/ALT	AA
64,905,158	137	46	C/DEL	I/FS 2	Affect protein function with a score of 0.00
64,905,296	275	92	C/T	T/M	Tolerated with a score of 0.36
64,905,387	366	122	G/A	M/I	Tolerated with a score of 0.81
64,905,683	662	221	G/DEL	R/FS	Affect protein function with a score of 0.00
64,905,736	715	239	C/A	L/M	Affect protein function with a score of 0.00

We further expanded the characterization of the VN1R1 variants by using SRA genomic data (PRJEB59761, PRJNA889458, and PRJNA277147), including assembling VN1R1 reads from these genomes and other cattle genome builds (GCA_002263795.4, GCA_021347905.1). This resulted in the detection of additional variants (Table S2) that encoded non-synonymous and deletion mutations in five positions (Table 3). Since no or low polymorphism was detected in CDS positions 366, 662, and 715 across the Irish bulls, we examined the association between variable CDS positions (137 and 275) and the RFI score (Table 3). A significant association with RFI (P_χ2 = 0.0053) was detected only for CDS position 137 (deletion allele), which could cause a frameshift in the translated protein, with a SIFT score of 0, likely leading to protein loss of function. The aberrant shifted allele was overrepresented in high-RFI individuals; thus, the intact gene is the FE-beneficial allele (Table 4). TBLASTN searches against genomic SRA libraries of Israeli (PRJEB59761), Irish (PRJNA889458), and USA (PRJNA277147) projects detected a high frequency (0.258–0.575) of frame-shifted alleles (Table S1). Annotated according to the interrogated AA (Table 5), the 46-AA position was frequent in all three populations, whereas the two other forms (221-AA and 46-AA combined with 221-AA) were identified in the Israeli population only (Table S1).

Table 4.

Contingency table examining RFI association of VN1R1 non-synonymous mutations in Irish bulls.1: FS—indicates frameshift affecting several amino acids. 2: χ2 test P-value.

Phenotype	CDS Position
	275 T	275 M	137I	137FS1
High RFI	7	21	13	15
Low RFI	10	18	23	5
P-value2	0.3833		0.0053

Table 5.

Fifteen amino acid probes interrogating five functional variations of the bovine VN1R1 protein. 1:Variable amino acids are denoted with an underlined bold font.

Position 46	Position 92	Position 122	Position 221	Position 239
> 137I1 NFTLLTGHNLRPIDPI  > 137T1 NFTLLTGHNLRPIDPT	> 275T1 DDTGCKLVFYFHRVAT  > 275M1 DDTGCKLVFYFHRVAM	> 366M1 ALKLKPSIWRWMELQM  > 366I1 ALKLKPSIWRWMELQI	> 662R1 VLFLGRHKRRIQRICSHR  > 662Q1 VLFLGRHKRRIQRICSHQ	> 715L1 PRPSCEGRATRTVLVL  > 715M1 PRPSCEGRATRTVLVM
> 137I21 ILMQLVIANATVLFSK  > 137T2 TSCNWSSPMPQFFSLK	> 275T2 TGVSFSTTCLFNGFQA  > 275M2 MGVSFSTTCLFNGFQA	> 366M2 MRALRFIAFCCFLCWI  > 366I2 IRALRFIAFCCFLCWI	> 662R2 QRICSHRVSPRPSCEGRA  > 662Q2 QRICSHQSPPDLPVRAEP	> 715L2 LVSSFVTFYTVYIILT  > 715M2 MVSSFVTFYTVYIILT
> 137I3 NLRPIDPILMQLVIAN  > 137T3 NLRPIDPTSCNWSSPM	> 275T3 LVFYFHRVATGVSFST  > 275M3 LVFYFHRVAMGVSFST	> 366M3 SIWRWMELQMRALRFI  > 366I3 SIWRWMELQIRALRFI	> 662R3 VSPRPSCEGRATRTVLVL  > 662Q3 SPPDLPVRAEPHALSWSW	> 715L3 GRATRTVLVLVSSFVT  > 715M3 GRATRTVLVMVSSFVT

Discussion

A primary limitation of our study is the modest sample size (n = 680), which is relatively small for a GWAS of a polygenic trait like RFI. Consequently, the statistical power to detect small-effect QTLs was limited, and reported associations are at risk of being false positives. Therefore, our results should be interpreted with caution and viewed as an exploratory effort to generate hypotheses that require validation in larger, independent populations.

Despite the limited sample size, integrating phenotypic and genotypic data from two distinct herds increased the genetic variation available for analysis. For instance, the minor allele frequency (MAF) of the top significant SNP marker BTB-01660086 on BTA2 is 1.5% and 16.5% in the US and IL herds, respectively, and 5.2% in the pooled population. This pattern shows that the two herds contribute different allele frequency spectra at RFI-associated loci, so that some alleles are rare or nearly fixed in one herd but segregate at moderate frequency in the other. Moreover, in the IL herd, the higher MAF occurs in a population with relatively high within-herd kinship, which suggests that much of the segregation is confined to specific pedigrees. In contrast, the additional carriers in the US herd are not, or only remotely, related to the IL cows, as indicated by the population structure and relationship analyses (e.g., Fig. 1). Combining the two herds, therefore, increases the effective segregation both by raising the overall allele frequency and by adding genetically less related carriers. This helps explain why the single-herd GWAS had limited power, whereas the joint analysis of the combined population identified significant associations beyond the simple increase in sample size and allowed the detection of 14 significant RFI QTLs.

Two of the herein identified SNP markers, BTB-00730867 on BTA18 and BTB-01776422 on BTA1, are in proximity to RFI QTLs previously reported around UMD3.1 BTA18 57-58 Mb and UMD3.1 BTA1 104 Mb, respectively^7,23, and in a previous meta-analysis that integrated data from 47 studies and reported an additional 11 QTLs of those identified in our study⁷. Hence, our study provides valuable preliminary insights, although further studies should validate the results before concluding data reliability to outline the genetic mechanisms underlying RFI. The RFI generally measures the differences between actual and expected feeding based on the animal’s BW and production; thus, it reflects FE²⁴. Interestingly, the QTL enrichment analysis pointed to traits that are components of the RFI calculation, such as DMI and BW. While not a novel biological discovery, this result provides important supporting evidence that our analysis successfully captured biological signals related to feed efficiency, despite the study’s limitations.

One interesting outcome of this study is the implication of chemosensory receptors, particularly pheromone receptors, in feed efficiency. These vomeronasal receptors (VRs), which include V1Rs and V2Rs located in the vomeronasal organ, are believed to detect pheromones primarily. Although they are G protein-coupled receptors (GPCRs) involved in chemosensation, like olfactory receptors (ORs), they do not share significant sequence similarity with ORs, whose gene family has a common frequency of genes and pseudogenes in the bovine genome (> 1000 genes²⁵). Our gene enrichment analysis identified genomic regions harboring a cluster of pheromone receptors, which suggests that one or more of these genes are involved in FE regulation. Indeed, previous work suggested these genes regulate feed intake and appetite-related traits¹⁴. This aligns with growing evidence that polymorphisms in odorant and pheromone receptor genes can influence feeding behavior. For instance, in humans and rodents, it has been demonstrated that genetic variation in specific odorant receptor genes has been associated with different feeding behaviors, food choices, and regulating energy balance. Additional evidence has demonstrated the physiological role of OR in livestock’s appetite regulation, as previously summarized¹⁴. Olfactory receptors are not only essential for smelling environmental odors but also play physiological roles in regulating appetite. They are part of a broader chemosensory system (including vomeronasal pheromone receptors) that transmits chemical signals (like odors and pheromones) to the brain’s feeding centers¹⁴. Expression of BTA18 VRs as recorded by the Cattle-GTEx project²⁶ indicated that these receptors are expressed along this transmission pathway, with VN1R1 expression peaking in the cerebral cortex (Fig. 1S). VN1R1 and VN1R4 present a similar expression pattern that is distinguished from V1R416 and V1R418 (Fig. S1). VN1R1 and VN1R4 are expressed across multiple bovine tissues, mainly in the brain, digestive, and reproductive systems. Interestingly, both genes are expressed in the olfactory, nasal, and tonsillar pharynx, part of the organs where taste perception takes place²⁷. On the contrary, V1R416 and V1R418 are mainly expressed in reproductive and embryonic tissues (Fig. S1).

Accumulating evidence demonstrated that olfactory pathways via the endocannabinoid system likely link energy-sensing signals, olfactory processes, and food intake²⁸. Although there is sparse evidence for pheromone receptor effects on cattle feeding, our results provide a supportive indication for this concept. Since one of the top markers for RFI in our analysis is located within a cluster of the vomeronasal genes, a possible explanation is that differences in smell or pheromone signaling might alter feed intake or eating behavior. Such a mechanism could operate through neural pathways, where an animal with a specific variant might experience feed differently (e.g., enhanced palatability), thus affecting how much it eats and how efficiently it utilizes feed. In addition, pheromone receptors could play a role in mediating social or stress cues that can indirectly impact feeding. Interestingly, we observed the occurrence of truncated VN1R1 genes in diverse cattle breeds, including American, Irish, and Israeli Holstein, as well as in the Holstein genomic build based on a bull originating from New Zealand. A high frequency of truncated VN1R1 alleles was detected in all examined populations (Table S1). Thus, the indication that this allele is not beneficial in the Irish population suggests that selection for the intact gene has promising potential for improving FE, if similar results are replicated in additional studies and in several unrelated populations. The occurrence of recombination within the relatively short (945 bp) VN1R1 CDS is likely to be a rare event; thus, the presence of all four possible combinations between the frameshift mutations suggests these emerged in the distant past. This abundance of aberrant VN1R1 alleles stands in contrast with our observation that the intact VN1R1 receptor is associated with a beneficial effect on feed efficiency and thus suggests a possible hidden genetic conflict that supports the propagation of the truncated forms, similar to previously reported selection conflict for fertility and production-associated traits^29,30. For instance, a conflict could arise from selection by breeding that is directed to prefer high production rather than FE, which is also affected by feed intake and metabolic efficiency. Studies have shown that production traits do not necessarily improve automatically when selection targets RFI, and vice versa. In a multi-herd analysis of first-lactation Holsteins²⁹, favorable genetic correlations between DMI and fat and protein yields (0.43 and 0.50, respectively) were reported, whereas the genetic correlations between these production traits and RFI were negative (-0.07) and smaller than 0.10, respectively, indicating near independence at the genetic level. Taken together, our data are consistent with the involvement of chemosensory receptors with a plausible biological mechanism of feed efficiency. If validated, it can open new avenues, such as managing olfactory environments or selective breeding for certain olfactory receptor variants to modulate feed intake and efficiency in dairy cows. In the latter case, our results suggested that for breeding, prioritizing bulls and cows that carry intact VN1R1 receptors may be successful in freeing herds from the aberrant genes, as was achieved for Complex Vertebral Malformation (CVM) through routine DNA testing and culling known carriers³¹.

In addition to the vomeronasal receptors and their associated pathways, our analysis pointed to fat metabolism and weight-gain genes. Notably, pathway analysis of feed efficiency candidates in dairy cows consistently identifies lipid metabolism among the top biological processes, underpinning the importance of energy metabolism and growth regulation to feed efficiency. This concept was demonstrated in dairy cattle and feed-efficient beef cattle (with low RFI) that were shown to upregulate genes for fatty acid transportation and β-oxidation in the liver^7,32. Collectively, our results and the evidence from the literature suggest that, at least in part, feed efficiency is a trait composed of appetite regulation, behavioral response, and fat metabolism. It should be noted that in this study, our discussion of other behavioral and fat metabolism genes potentially affecting RFI was limited. In our sample, we did not detect likely functional protein-coding variants in these candidates, in contrast to the variants observed in the pheromone vomeronasal receptors. Nevertheless, these genes remain relevant, as they may be influenced by other forms of genetic variation. In particular, most causal variants for complex traits are assumed to be regulatory and located in non-coding regions, so functional effects in these genes may still arise through regulatory mechanisms rather than protein sequence changes^33,34.

Conclusions

This study provides preliminary evidence that genetic factors related to pheromone sensing and fat metabolism may play a role in feed efficiency in dairy cattle. Our findings are consistent with a model where feed efficiency is influenced by both appetite regulation and metabolic processes. The identified variation in pheromone receptors could contribute to differences in feeding behavior, while the implicated fat metabolism genes may influence how dietary energy is converted into productive output. Our results highlight the complexity of feed efficiency, suggesting it is governed by both neural (behavioral) and metabolic components. Further validation of these candidate genes in larger populations is warranted. In breeding programs where genomic selection indices already incorporate RFI or closely related feed efficiency traits, dense SNP panels will likely capture the variation at such loci through linkage disequilibrium. However, in many herds, RFI is not represented in the selection objective because recording intake phenotypes is limited. In such cases, once confirmed, selecting against unfavorable VN1R1 alleles and other validated variants could provide a complementary route to promote feed efficiency, alongside the ongoing genomic selection schemes.

Materials and methods

Study population and phenotypes

Dry matter intake (DMI), milk yields (MY), and body weight (BW), all measured in kilograms, as well as fat, protein, and lactose components of milk, were collected from Holstein cows at the two research stations, including 488 cows in the USDA Animal Genomics and Improvement Laboratory (AGIL) in Beltsville, Maryland, USA³⁵, and 192 cows in the Agricultural Research Organization (ARO) dairy herd in Rishon-Letzion, Israel³⁶. The US 488 cows descended from 117 sires and 357 dams, whereas the IL 192 cows descended from 56 sires and 159 dams (Table S2). For the US sample, the RFI data were collected continuously from the early 2000s to the mid-2010s, using the methods previously reported^8,37. For the IL sample, the RFI data were collected continuously from 2012 to 2019³⁶.

Residual feed intake (RFI) calculation

Calculation of the RFI (kg/day) was by statistical determination of the deviation of the actual DMI of a cow from the expected intake, based on the average of the cohort (cows managed the same and fed the same diet at the same time), calculated for each cohort according to the equation:

where DMIi is the observed daily dry matter intake of cow i (kg/day), ECM is the daily energy-corrected milk yield of cow i (kg/day), and WOLi is the week of lactation of cow i. The coefficients were determined for each cohort to yield the highest R² and lowest intercept for the relationship between predicted and actual DMI³⁸. ECM yield [kg/d of standard milk containing 3.5% fat, 3.5% protein, and 5% lactose, with energy value of 0.714 Mcal/kg milk (NRC 2001)] was calculated as:

where MY is milk yield (kg/day) and mf, mp, and ml are milk’s fat, protein, and lactose concentration (%), respectively. The cows were multiparous, high-yielding (> 35 kg/day), and healthy at early mid-lactation, 60–180 days in milking. All cows were housed in the experimental dairy farms equipped with an individual feeding system. Computerized monitoring for each cow was performed for the number of meals per day (24 h), meal size and duration, diurnal eating over day and night, and daily feed intake. A comprehensive description of the full assay and feeding systems is detailed in^8,36,37,39. DMI was calculated manually daily by subtracting the orts from the served meal.

Genomic and bioinformatics analysis

Genotypes were determined using various Illumina bovine beadchips. Israeli animals were genotyped using the Illumina BovineSNP50 BeadChip, while the US animals were genotyped using a combination of Illumina BovineSNP50 and BovineHD BeadChips. Thus, a final dataset with genotypes for 50,392 SNPs shared by all cows was generated as described before⁴⁰. This was achieved by using Plink⁴¹ to merge the IL and US datasets with the –merge option, and SNPs with opposite strand annotation between platforms were identified by an in-house script and flipped before merging. Quality control on SNPs was performed as described in^42,43 using PLINK⁴¹. We removed SNPs with minor allele frequency (MAF) lower than 0.01 using –maf 0.01 and SNPs with a call rate lower than 70 percent (missing in more than 30 percent of cows) using –geno 0.3, to exclude very rare variants or those with little information for association testing. After these filtering steps, 42,639 informative SNPs remained and were used for GWAS carried out by the EMMAX software (version EMMAX-intel64-20120205)⁴⁴. The primary aim of the GWAS was to map SNP markers and genomic regions associated with RFI. At the same time, we needed to account for non-genetic differences between herds and for the genetic relationships among cows, so that the resulting association tests would not be confounded by farm management or population structure. To account for population structure and farm-of-origin effects, a genome-wide association study (GWAS) was performed using a single-SNP mixed linear model in the EMMAX software. The model was as follows:

where y is the vector of RFI phenotypes; β is a vector of fixed effects containing the additive SNP effect, and the farm of origin (US vs. IL); X is the corresponding design matrix; g is a vector of random polygenic effects distributed as , and e is a vector of residuals with . The matrix G is the genomic relationship matrix (GRM), which summarizes the realized genetic relationships between cows and was computed in EMMAX from the filtered SNP genotypes using the command emmax-kin-intel64 -v -s -d 10, where -v requests verbose output, -s specifies the standard symmetric kinship matrix format used by EMMAX, and -d 10 sets the numerical precision of the kinship coefficients, following Kang et al.⁴⁴. The farm of origin was included as a fixed class effect to correct for systematic management and environmental differences between the two herds, while the GRM term accounted for population structure and relatedness at the genomic level.

The GWAS was carried out using the command: emmax-intel64 -v -d 10 -t tped -p pheno -c COV -k GRM –o OUT. Where -t specifies the genotype file prefix, -p the phenotype file, -c the covariate file, -k the genomic relationship matrix, and -o the output prefix. The allele substitution effects and standard error, the nominal probabilities for the hypothesis of no effect, and the EMMAX genomic pseudo-heritability (gREML) were computed. This heritability is a measure of the proportion of phenotypic variance explained by the GRM and is similar to a heritability estimate from REML in Genomic Best Linear Unbiased Prediction (GBLUP) but is based on an empirical GRM and assumes SNPs with no major effects. To account for multiple testing, we use the Benjamini-Hochberg (BH) procedure⁴⁵. Finally, we computed the genomic inflation factor λ⁴⁶ to evaluate hidden population structure, which retained a value of 0.966, denoting no inflation. A Quantile–Quantile plot for the RFI p-values was created using the R package qqman⁴⁷. For the population structure analysis, the IBS similarity matrix was calculated using the Plink⁴¹ –genome flag, and a MDS analysis was carried out with –cluster and –mds_plot arguments.

Quantitative trait locus (QTL) annotation was performed as described before⁴⁸. The annotation file of the QTL database for cattle was obtained from the Animal QTLdb⁴⁹, and the bovine Gene Transfer Format (GTF) genome annotation file was obtained from Ensembl⁵⁰. Both files are based on the bovine reference genome assembly ARS-UCD1.2⁵¹, corresponding to this study’s marker coordinates. The main function find_genes_qtls_around_markers() from the GALLO package¹³ was applied to annotate the SNP markers by using input files containing the SNP positions (chromosome and base pair) and reference database files in GTF obtained for the genes from Ensembl and the General Feature Format (GFF) file for RFI QTLs from the Animal QTLdb. Gene enrichment analysis was performed using the GeneAnalytics and Enrichr server^19,52, which can identify gene enrichment for several terms and data sources, including diseases, pathways, Gene Ontology (GO) terms, and tissue expression. The window size of the genomic interval used for enrichment was set to 250Kb.

Genomic sequence analyses and searches

For the identification of additional variation in the candidate genes, we first combined a joint Variant Call Format (VCF) file of 27 Israeli Holstein bulls that we previously submitted to the SRA repository (ENA BioProject PRJEB59761), following our previously described WGS analysis pipeline^53,54. All identified variants in the joint VCF file underwent comprehensive annotation by the Ensembl VEP⁵⁵.

In addition, to assess the likely functional variants identified in the VN1R1 candidate gene, its exon sequences were downloaded from NCBI (https://www.ncbi.nlm.nih.gov), and SRA genomic reads were mapped to these templates using GAP5 software⁵⁶. For further characterizing functional variation observed in the encoded protein of the candidate gene, on a larger scale, we used NCBI BLAST⁵⁷, e.g., three probes interrogated each of the five VN1R1 functional variations, whereas the critical variation was positioned in the middle, the left, and the right terminals of the probes (Table 5). These probes were used as queries in batch TBLASTN searches against genomic SRA libraries.

To determine the allelic frequency of VN1R1 functional variations, we used read counts in SRA libraries of Israeli (PRJEB59761), Irish (PRJNA889458), and US (PRJNA277147) projects. Reads encoding the mutations were detected using probes for AA positions 46 and 221 (Table 5) in TBLASTN searches against the raw genomic data. The χ2 test was applied to examine the association between sequence variants and RFI in Irish cattle (PRJNA889458).

Abbreviations

BLAST

Basic local alignment search tool

EMMAX

Efficient mixed-model association expedited

FE

Feed efficiency

RFI

Residual feed Intake

MDS

Multidimensional scaling

SRA

Short read archive

IBS

Identical by state

CDS

Coding sequence

TBLASTN

Translated BLAST

VCF

Variant call format

Author contributions

ES, GEL, and MG designed the study. AS, LY, NB, YB, AS, MCZ, RLB, ES, GEL, and MG collected, analyzed, and interpreted the data. MG and ES wrote the manuscript. MG, ES, GEL, and RLB substantively revised the manuscript. All authors read and approved the final manuscript.

Funding

This work was supported by grants from the ISF 946/19 and the Israeli Dairy Board, given to MG.

Data availability

This study was based on previously reported data^35,36, BioProjects PRJEB59761, PRJNA889458, PRJNA277147, and ENA BioProject PRJEB59761.

Competing interests

The authors declare no competing interests.

Ethics approval and consent to participate

This study was based on previously reported data^35,36; therefore, no animals were used specifically for this study, and therefore exempt from animal care and use approval. The authors have not stated any conflicts of interest.

Consent for publication

Not applicable.