The aim of this study was to perform a GWAS of bovine fatty acids based on summer milk samples and to compare them to previous results of a GWAS of fatty acids based on winter milk samples . For this GWAS we used different milk samples from largely the same set of cows with the same genotypes at a different stage of the same lactation. The main difference between the two seasons was the herd management, including feeding. In winter, all herds were kept indoors and fed silage, while in summer about half of the herds were grazing outside for at least part of the day and a few other herds were fed fresh grass. Diets of cows including fresh grass are known to alter milk fat composition e.g. [13, 14]. This was reflected in our results: summer milk contained more long chain fatty acids and less C16:0 compared to winter milk. The phenotypic correlations for the different fatty acids between summer and winter samples ranged between 0.36-0.67 (Table 2), indicating that the summer samples provide additional information compared to the winter samples. Genetic correlations between summer and winter samples of C4:0, C6:0, C12:0, C18:0, C10:1, C12:1, C14:1, C16:1, and C18:1 (ranging between 0.90-1, Table 2) showed that these fatty acids are genetically the same trait in summer and winter . The genetic correlations for C8:0, C10:0, C14:0, C16:0, and CLA were significantly different from one  but showed strong positive correlations (0.77-0.94, Table 2), suggesting that also for these fatty acids summer and winter samples have most genetic variation in common. It is important here to consider that strong positive genetic correlations are required to ensure that the traits have the same genetic background. However, for the summer milk sample to provide additional information to our previous GWAS phenotypic correlations should be weak.
This GWAS of fatty acids based on summer milk samples shows agreement with most associations detected in our previous GWAS of fatty acids based on winter milk samples . Three regions with major effects detected in the winter GWAS  were also found in the summer GWAS. On BTA 14 a dinucleotide polymorphism in DGAT1 is causing the major effect. The DGAT1 K232A polymorphism is known to be associated with fat content and composition, so our results are in line with other studies [1, 16, 17]. The rather large region associated with the short and medium chain saturated milk fatty acids on BTA 19 confirm previous linkage studies [18–20]. Several candidate genes related to fat synthesis are located in this region, e.g. ATP citrate lyase, sterol regulatory element-binding transcription factor 1, signal transducer and activator of transcription 5A, growth hormone, and fatty acid synthase. There might be more than one QTL in this region given the size of this region and the many candidate genes, but the actual polymorphism(s) causing the effect(s) has not yet been identified. On BTA 26 a polymorphism in SCD1 is causing the major effect. The gene SCD1 is known to be associated with medium-chain unsaturated fatty acids, so our results are in line with other studies [16, 21–24].
Our results from both GWAS studies also suggest that there are additional QTL on BTA 14 besides DGAT1 that were associated with fatty acids. These additional QTL on BTA 14 were located at 3.0-3.8 Mbp, 32.7 Mbp, 40.8-50.9 Mbp, and 73.0-76.5 Mbp, and confirm detected QTL for milk production traits in linkage analyses reviewed in . Candidate genes for these regions might be corticotropin releasing hormone at 30.5 Mbp, fatty acid binding protein 5 at 41.9 Mbp, and fatty acid binding protein 4 at 42.0 Mbp.
The three highly significant regions with major effects mentioned above were expected to be found in both GWAS studies. More interesting are the additional regions that were found in both GWAS studies such as the regions associated with more than two fatty acids: region 5c, 6, 13, 17b, and 27. Also worthwhile to mention are the ‘new’ regions that had a suggestive FDR between 5-20% and were not considered in the individual studies based on winter or on summer milk samples only, but became of interest because they were found in both studies. There are eight ‘new’ regions like this: region 1, 2a, 3, 5a, 10, 14b, 17c, and 24 (Table 4).
There are two different confirmation strategies regarding GWAS: replication and validation. Replication studies are meant to confirm that the actual association is a true association and should therefore be based on samples from the same population with minimal systematic differences [26, 27]. Validation studies are meant to see if the association can be generalized over different populations and should therefore be based on samples from a different population, where this population can be different concerning genetic background, phenotype definition, sampling strategy, and time point of investigation [26, 27]. A correctly performed replication is more likely to be successful in finding the same association again than validation, however, when an association is validated the associated SNP is probably closer to the actual polymorphism. In literature, replication and validation are often used interchangeably which complicates the interpretation of the results, especially when a study is called replication study but population, phenotype or study design are too different from the original study to be a replication study. Our GWAS has elements of a replication as well as of a validation study; it met criteria for a replication such as sufficient sample size, phenotypes were measured using the same method, same set of markers and a very similar population was used. It also met criteria for validation because the phenotypes were measured at different time points. The difference in season of measuring the phenotype provides additional information. Ideally an independent population of cows should have been sampled, but this was practically not feasible.
Replication of the study with largely the same set of animals and the same genotypes led to minor differences in LD and allele frequencies between studies, therefore it is more likely to confirm previously detected results . However, spurious associations due to population structure or genotyping errors are more likely to be detected twice using the same set of animals and genotypes.
Agreement between the two GWAS studies was based on a FDR threshold of 20% in each study. If the winter and summer GWAS would be independent, the FDR of a region found in both studies would be 4% (20%*20%) . A FDR of 4% gives enough reason to investigate such a region further. Lowering the threshold from 5% to 20% FDR resulted in the eight ‘new’ regions mentioned above, besides the regions that were already discovered in one of the individual studies and were in agreement with the other.
Even though we used largely the same set of animals, same genotypes, same phenotype measurement and a lower threshold for agreement between summer and winter GWAS not all regions were found in both summer and winter GWAS. This can be due to genotype by season interaction, due to lack of power (false negative QTL) or because these QTL were false positive QTL. It is not possible to determine which of these three reasons apply. However, the genetic correlations indicated that fatty acids are genetically similar traits in summer and winter, which suggests that genotype by season interaction may have only a small effect on the results. So lack of agreement between summer and winter GWAS is either due to lack of power or false positives.