Identification of a two-marker-haplotype on Bos taurus autosome 18 associated with somatic cell score in German Holstein cattle

Background The somatic cell score (SCS) is implemented in routine sire evaluations in many countries as an indicator trait for udder health. Somatic cell score is highly correlated with clinical mastitis, and in the German Holstein population quantitative trait loci (QTL) for SCS have been repeatedly mapped on Bos taurus autosome 18 (BTA18). In the present study, we report a refined analysis of previously detected QTL regions on BTA18 with the aim of identifying marker and marker haplotypes in linkage disequilibrium with SCS. A combined linkage and linkage disequilibrium approach was implemented, and association analyses of marker genotypes and maternally inherited two-marker-haplotypes were conducted to identify marker and haplotypes in linkage disequilibrium with a locus affecting SCS in the German Holstein population. Results We detected a genome-wide significant QTL within marker interval 9 (HAMP_c.366+109G>A - BMS833) in the middle to telomeric region on BTA18 and a second putative QTL in marker interval 12-13 (BB710 - PVRL2_c.392G>A). Association analyses with genotypes of markers flanking the most likely QTL positions revealed the microsatellite marker BMS833 (interval 9) to be associated with a locus affecting SCS within the families investigated. A further analysis of maternally inherited two-marker haplotypes and effects of maternally inherited two-marker-interval gametes indicated haplotype 249-G in marker interval 12-13 (BB710 - PVRL2_c.392G>A) to be associated with SCS in the German Holstein population. Conclusion Our results confirmed previous QTL mapping results for SCS and support the hypothesis that more than one locus presumably affects udder health in the middle to telomeric region of BTA18. However, a subsequent investigation of the reported QTL regions is necessary to verify the two-QTL hypothesis and confirm the association of two-marker-haplotype 249-G in marker interval 12-13 (BB710 - PVRL2_c.392G>A) with SCS. For this purpose, higher marker density and multiple-trait and multiple-QTL models are required to narrow down the position of the causal mutation or mutations affecting SCS in German Holstein cattle.


Background
Udder health, somatic cell score and subclinical and clinical mastitis remain major challenges for the economy of milk production in respect to milk production efficiency and animal health and welfare. Several studies have attempted to identify chromosomal regions, genes and polymorphisms that influence udder health in order to improve breeding strategies. SCS has been used as an indicator of udder health and is implemented in routine sire evaluations in many countries [1]. SCS has a low to medium heritability (h 2 = 0.15; [1]) and a strong correlation to mastitis in the German Holstein population (r g = 0.84; [2]). However, selection on low SCS as well as on decreased mastitis incidence is hampered by three aspects: first the low heritability of SCS and liability to mastitis, second the difficulties in recording mastitis related data and third by potential population-wide antagonisms between milk production traits (milk, fat and protein yield) and udder health [1,2]. Recently, marker assisted selection (MAS) has been determined as a promising tool to improve current selection strategies based on phenotypic data [3]. MAS implements genetic marker information of confirmed QTL regions to identify individuals with favourable genetic background concerning the trait of interest. Thus, the confirmation and fine mapping of known QTL regions and the estimation of QTL effects will advance the use of MAS.
For clinical mastitis (CM) or SCS, QTL have been detected on nearly all autosomes [4] and several studies repeatedly detected QTL for SCS or CM in the telomeric region of BTA18 [5][6][7][8][9][10][11][12]. In addition, Kühn et al. [13] have shown in a proof-of-principle approach that information of 5 markers located in the telomeric region of BTA18 indeed enabled successful MAS, which identified halfsib heifers prior to first calving that exhibited significant differences in SCS after parturition.
The intention of this study was to further analyse the middle to telomeric region of BTA18 with the aim of identifying markers and marker haplotypes in linkage disequilibrium (LD) with SCS in German Holstein cattle to improve MAS for udder health. Therefore, we increased the marker density in the telomeric region on BTA18 and selected four functional candidate genes within the QTL regions reported by [5,7,11]. Polymorphisms detected within these candidate genes were used as additional markers for fine mapping previously identified QTL regions and to analyse effects of candidate gene polymorphisms on SCS in the German Holstein. In the present study, we detected a genome-wide significant QTL in the middle to telomeric region on BTA18. Furthermore, we analysed effects of maternally inherited marker haplotypes and identified a two-marker-haplotype associated with SCS in German Holstein cattle.

Selection of candidate genes
Based on the positional information derived from the previously mentioned QTL regions identified by [5,7,11] and preliminary results from microarray experiments, an intensive literature research of genes bearing potential function in innate immune defence, immune defence, mammary gland development or udder morphology was performed. To enable the investigation of positional candidate genes within the telomeric region on BTA18, a comparative map for BTA18 and Homo sapiens autosome 19 (HSA19) was constructed to take advantage of the more advanced gene annotation of the human genome (Additional file 1: marker table). Four candidate genes were selected: calmodulin 3 (phosphorylase kinase, delta) (CALM3), Hepcidin antimicrobial peptide (HAMP), cadherin 1, type 1, E-Cadherin (ephitelial) (CDH1), and poliovirus receptor-related 2 (herpesvirus entry mediator B) (PVRL2). Calmodulin 3 is a ubiquitously expressed Ca 2+ -binding protein that is involved in many Ca 2+ modulated signal pathways. It was selected due to its influence on smooth muscle contraction [14,15] and the potential effects on milk leaking and milk flow, which is a trait with a substantial genetic correlation to SCS (r g = 0.4; [1]). HAMP was selected as an immunological candidate gene because of its antifungal and antibacterial activity [16], the toll-like receptor-4 dependent induction by bacterial pathogens in myeloid cells [17] and to a lesser extent due to its function as a key regulator of iron metabolism [18]. For CDH1 and PVRL2 initial results from own microarray experiments indicated their differential expression in clinically unaffected heifers with different predisposition to udder infection. Additionally, both genes are involved in cell-cell junctions that have a strong impact on tissue development. E-Cadherin, a Ca 2+ -dependent cell-cell adhesion molecule, is involved in tissue and organ development as part of the cadherin-catenin-complex [19,20], and poliovirus receptor-related 2, a Ca 2+ -independent immunoglobulin-like cell adhesion molecule, is involved in the organisation of intercellular junctions as part of the nectin-afadin-complex [20][21][22]. Furthermore E-Cadherin can serve as a receptor for pathogens and is involved in their internalization [23,24].

Screening for polymorphisms
Polymorphisms within the candidate genes were detected by comparative sequencing of genomic DNA from heifers selected for QTL alleles associated with high or low SCS. The heifers originate from most likely QTL heterozygous sires selected from the German Holstein population based on marker information regarding a confirmed QTL for SCS on BTA18. The selection strategy for the heifers and their phenotypes are described by Kühn et al. [13] in detail. For sequencing, we selected three pairs of halfsibs, where one daughter inherited the SCS increasing paternal chromosomal region (q) and the other inherited the SCS decreasing paternal chromosomal region (Q). In addition, two animals were screened for variants which originated from a genetically divergent Charolais × Holstein F 2 cross background.
For each gene except CDH1 the entire gene (large introns excluded), 800 to 1500 bp of the promoter region and up to 500 bp downstream of the transcripts were investigated. For CDH1, only the genomic sequence spanning exon 13 to 15 was analysed. Primer information and the genomic position for each primer are given in additional file 2: primer table sequencing. Sequencing was performed by amplification of genomic DNA and subsequent sequence analysis with the DYEnamic ET Terminator Cycle Sequencing reaction and the Mega-BACE™1000 DNA Analysis System (GE Healthcare, Munich, Germany). For evaluation of polymorphisms BioEdit 7.0.5.2 [25] was used. Essentially, polymerase chain reaction (PCR) primers were used for sequencing. Additional sequencing primers were used only for longer PCR fragments or PCR fragments that were difficult to sequence.

Families
The pedigree material used for genotyping included a total of 1,054 animals originating from six paternal halfsib families. Some of the animals are a subset of the granddaughter designs previously described by [7,11]. Numbers of sons per grandsire ranged from 60 to 353, with an average family size of 175 sons. The German genetic evaluation center (VIT) in Verden, Germany provided additional pedigree information including non-genotyped ancestors of genotyped animals (7,627 animals).

Phenotypes
The phenotype information for SCS was provided by the VIT as daughter yield deviations (DYD) for the first lactation (Table 1). SCS is the log 2 transformed somatic cell count (log 2 (somatic cell count/100000) + 3). DYD for SCS were calculated based on a random regression test day model [26]. The reliabilities associated with the DYD were expressed as the number of effective daughter contributions (EDC) as described by [26]. DYD and EDC for genotyped animals were obtained from the official release of the April 2008 routine genetic evaluation.

Marker Set
The marker set included a total of 28 markers covering the telomeric region of BTA18 from CDH1 to DIK4013. Six of the 28 markers were already genotyped within previous QTL mapping studies [5,7,11] and include an erythrocyte antigen marker. The other 22 markers were newly selected and genotyped. Fifteen new microsatellite and seven new single nucleotide polymorphism (SNP) markers were chosen based on the information of the putative QTL positions reported by [5,7,11]. The fifteen new microsatellite markers were selected using the bovine linkage map of the United States Department of Agriculture's Meat Animal Research Center (MARC USDA) [27]. SNP markers were selected from all detected polymorphisms within candidate genes based on allele frequencies, position (coding-or non-coding region), effect of the SNP (synonymous or nonsynonymous) and whether they are in linkage disequilibrium to each other.  Table 2). The enzymes used for detection of RFLPs were identified with the NEBcutter V2.0 webtool [28,29] and PCR primers were designed with primer analysis software Oligo 4.1 (National Bioscience Inc., Plymouth, MN, USA). For the RFLP assays, a PCR specific for each SNP was used to amplify genomic DNA, and PCR products were subsequently incubated for 8 h with SNP specific restriction enzymes. For visualization of the RFLP a 2.5% agarose gel was used. Arithmetic mean and standard deviation as well as lowest and highest daughter yield deviation for SCS considering all animals are given. SCS is the log 2 transformed somatic cell count (log 2 (somatic cell count/100000) + 3).

Genotyping
Primer design for pyrosequencing was performed with Pyrosequencing™ Assay Design Software (Biotage AB, Uppsala, Sweden). A SNP specific PCR was used for amplification of genomic DNA and the products of both PCR were merged and analysed with the PSQ™HS 96A pyrosequencing system (Biotage AB, Uppsala, Sweden) in a multiplex run.

Linkage Map
The genetic linkage map was calculated based on a refined marker order using CRIMAP software [30]. The marker order on BTA18 used for the calculation was evaluated in two steps. First, information from published linkage maps [31,32], RH-maps [33], and human and bovine sequenceassemblies [34] were merged and compared to own linkage-mapping results. Second, in a region including the markers BMON117, DIK4672, BMS833, DIK4232 and BB710, no unequivocal marker order was obtained. Thus, the marker order in this region was verified by RH-mapping using the 12000 rad whole-genome radiation hybrid panel [35] and RH-MAP3.0 software [36]. For some marker groups, a recombination rate of zero was calculated. To avoid technical difficulties arising in the calculation of transmitting probabilities the marker spacing was set to small values greater than zero (Additional file 1:  marker table).

QTL Mapping
A combined linkage and linkage disequilibrium analysis (LALD) was performed using the software system TIGER [37]. TIGER is a UNIX script linking several individual Fortran programmes to perform combined linkage and linkage disequilibrium analysis. Six steps are implemented in the script. First, allele frequencies and transmit-ting probabilities for each putative QTL position are calculated using BIGMAP [38]. The putative QTL positions were considered as the midpoint of each marker interval, resulting in a total of 27 putative QTL positions. Second, the identical by descent (IBD) sub-matrices for each putative QTL position are computed based on the gene dropping procedure described by [39][40][41] and third, the IBD sub-matrices are tested for positive definiteness and inverted. The software program COBRA [42] computes a condensed gametic relationship matrix and its inverse at each putative QTL position for the calculation of gametic effects. Transmitting probabilities and IBD sub-matrices are used for the set up of the condensed gametic relationship matrix. Finally, an LALD analysis is performed analysing every putative QTL position with restricted maximum likelihood (REML) methods applied in ASReml [43]. A detailed description of the QTL mapping procedure and the model applied in ASReml is given by [12].
Analyses were conducted using a likelihood ratio test, where the REML of the full model was compared with the REML of the model missing the QTL effect. Chromosomewide and genome-wide significance thresholds were determined as restricted log likelihood ratio (RLRT) equivalents of logarithm of odds (LOD) units [44] where a LOD > 2 indicates chromosome-wide (RLRT = 9.2) and a LOD > 3 indicates genome-wide significance (RLRT = 13.8) [45]. Confidence intervals were estimated using the LOD drop-off method described by [44].

Association analysis
To investigate the association of candidate gene polymorphisms and markers flanking interval 9 and interval 12-13  with SCS, a mixed model including a random polygenic effect and the fixed effect of marker genotypes was applied in ASReml: where y is a vector of phenotypic observations (DYD) for sires, μ is the overall mean, MG i is the fixed effect of the marker genotype i, a j is the random polygenic effect of animal j and e ij is the random residual. The polygenic effect that accounts for the family structure of the population was estimated using an extended pedigree of non-genotyped ancestors of genotyped animals including a total of 7,627 animals. To account for multiple testing, a 5% experiment-wise significance threshold was obtained by Bonferroni correction of the nominal p-value assuming a 5% Type 1 error (p exp = 0.0057).

Analysis of maternally inherited two-marker haplotypes and two-marker-interval gametes
Due to the limited number of sires, paternally inherited chromosomes could have a strong impact on genotype effects estimated in their offspring. To exclude these effects, maternally inherited two-marker-intervals including flanking markers of most likely QTL position were investigated. The most probable linkage phases of genotyped sires were calculated to determine the maternally inherited haplotypes using BIGMAP. The TIGER software system was then applied to estimate maternally inherited gametic effects for the putative QTL positions in interval 9 and interval 12 based on IBD sub-matrices. Finally, the estimated gametic effects of maternally inherited twomarker-interval gametes were analysed and plotted using SAS software (SAS Institute, Cary, NC, USA).
To verify differences in estimated two-marker-interval gamete effects, we performed an association analysis including a fixed maternally inherited two-marker-haplotype effect independent of the IBD sub-matrices. For this purpose, the same model was used as for marker genotypes (1), except that the fixed genotype effect was replaced by the fixed maternally inherited two-markerhaplotype effect. Two-marker-haplotypes with the highest and lowest mean for IBD gametic effects were tested against the total of all other haplotypes within respective intervals. For interval 12-13 additionally the most frequent two-marker-haplotype was investigated.

Screening for polymorphisms
Sequence analyses were based on the provisional reference sequences obtained from NCBI [46]. For CALM3 [GenBank: NM_001046249; GeneID: 520277], the entire gene (10728 bp) was resequenced including ~1.3 kb upstream from the transcription start (assumedly pro-moter region). A total of eleven polymorphisms were detected (Additional file 4: polymorphisms): ten SNP and one 12 bp deletion within the assumed promoter region.
For HAMP [GenBank: NM_001114508; GeneID: 512301], in silico analyses revealed that this gene is not annotated in Btau4.0 but has been annotated in NCBI Build3.1. The sequence of HAMP is still located on BTA18 in Btau4.0 [GenBank: NC_007316.3: 476114 bp-477570 bp]. The whole gene (2777 bp) was resequenced, and three SNP were detected (Additional file 4: polymorphisms), one SNP within intron 1 and two SNP within the first 110 nucleotides downstream of HAMP transcript.
For CDH1, only the coding sequence was provided as provisional mRNA reference sequence [GenBank: NM_001002763; GeneID: 282637]. Due to discrepancies between the mRNA reference sequence and the genome assembly sequence, only the genomic reference sequence was used for sequence comparisons. The genomic sequence including exon 13 to 15 (1673 bp) was investigated and six SNP were detected (Additional file 4: polymorphisms) two of them were located in exon 13 and cause an amino acid substitution.

Markers and Map
The genetic marker map covered 50 cM of the telomeric region of BTA18 (Additional file 1: marker table). Marker intervals ranged from 0.05 cM to 7.9 cM with an average marker interval of 1.85 cM. The number of alleles for each marker ranged from two for SNP to 34 for the erythrocyte antigen marker. The marker order is in good agreement with previously published linkage-and RH-maps [31][32][33] and no discrepancies to the bovine sequence assembly Btau4.0 [49] were observed (Additional file 1: marker  table).

QTL Mapping
The restricted Log Likelihood Ratio profile of the combined LALD analysis is shown in Figure 1. Two peaks exceeding the genome-wide significance level can be observed. The maximum of the first peak is located in marker interval 9 at position 71.775 cM with HAMP_c.366+109G>A and BMS833 as flanking markers. The maximum of the second peak is located in marker interval 12 at position 77.6 cM with the flanking markers BB710 and PVRL2_c.-1268G>C. The LOD drop-off method was used to estimate confidence intervals for each of the peaks. An approximate 96% confidence interval included marker intervals 3 to 9 for the first peak and the marker intervals 2 to 15 for the second peak.

Association analysis
An association analysis between the candidate gene polymorphisms and SCS was performed to evaluate the influence of the selected candidate genes on the variation in SCS in German Holstein cattle. The markers BMS833 and BB710 were also included in the association analysis, as they are flanking markers of the maxima observed in the LALD test statistic. The association analyses (

Analysis of maternally inherited two-marker haplotypes and two-marker-interval gametes
To exclude any specific sire gamete effects that occur in a paternal halfsib design, maternally inherited gametic effects for two-marker-interval gametes were analysed and plotted for the putative QTL positions in interval 9 and interval 12-13. In interval 9 (HAMP_c.366+109G>A -BMS833), six maternally inherited two-marker-allele combinations occurred. Two-marker-interval gametes including BMS833 alleles 115 and 119 were excluded, because they both occurred only once (Table 4; Figure  2A). Gametes carrying the BMS833 allele 117 in interval 9 have a mean estimated maternally inherited gametic effect of 0.022 (± 0.003) and 0.028 (± 0.003), respectively, whereas gametes carrying allele 113 of marker BMS833 have mean effects of -0.0006 (± 0.0014) and -0.0059 (± 0.0022), respectively. Investigating the gametic effect of each flanking marker of interval 9 alone ( Figure 2B), revealed that the main differences in gametic effects in the two-marker-interval results from the discrimination by microsatellite marker alleles. The difference in the mean effects for two-marker-interval gametes carrying the two LALD Profile PVRL2_c.392G>A (interval 13) are located in close vicinity within the PVRL2 gene and showed a high linkage disequilibrium (r 2 = 0.68). In interval 12-13 (BB710 -PVRL2_c.392G>A), two-marker-interval gametes with twelve different two-marker-allele combinations were observed. Two-marker-interval gametes carrying the two BB710 alleles 243 and 245 were excluded, because they both only occurred once. In addition, the frequencies for allele combinations 253-G and 257-G were smaller than 1% and therefore gametes carrying these two allele combinations are not included in Figure 3 (Table 4). Analo-gous to the situation in interval 9, the SNP alleles of PVRL2_c.392G>A seem to have only a small influence on discriminating estimated two-marker-interval gamete effects ( Figure 3B), because the difference in the mean effects discriminated by the alleles is 0.0078. For microsatellite marker alleles ( Figure 3A), two-marker-interval gametes carrying the allele 249 have mean effects of 0.0378 (± 0.0018) and 0.0176 (± 0.0009) and gametes carrying allele 253 have mean effects of -0.0150 (± 0.0021) and -0.0253 (± 0.0012). The biggest differences in the mean estimated gametic effects was observed for the two two-marker-interval gametes 249-G (0.0378 (± 0.0018)) and 253-A (-0.0253 (± 0.0012)). Thus, the difference in the mean maternally inherited gametic effects equals 0.0631, which is equivalent to 0.16 phenotypic standard deviations.
To validate the differences observed between maternally inherited two-marker-interval gametes, we performed a direct association analysis for maternally inherited twomarker-haplotypes without considering IBD coefficients in interval 9 (HAMP_c.366+109G>A -BMS833) and interval 12-13 (BB710 -PVRL2_c.392G>A) ( Table 5). Only the two-marker-haplotypes with the highest and lowest mean for IBD gametic effects were tested against the total of all other haplotypes. For interval 12-13, we also tested 255-A, because it was the most frequent haplotype (frequency at 40%) in the data set. In interval 9, no significant association of the maternally inherited haplotypes G-113 and G-117 were observed, whereas haplotype 249-G (p = 0.027) in interval 12-13 seems to be associated with SCS in the German Holstein population. Summary of results from association analyses for marker genotypes based on a mixed model (1) including a fixed genotype effect and a random polygenic effect. The experiment-wise 5% significance level is p exp = 0.0057.

Discussion
In an approach to identify marker and marker haplotypes affecting SCS in the German Holstein population, we analysed association of marker genotypes and maternally inherited two-marker-intervals flanking putative QTL positions detected in an LALD analysis for SCS on BTA18, and showed that the two-marker-haplotype 249-G of interval 12-13 (BB710 -PVRL2_c.392G>A) is in LD with SCS in the German Holstein population.
Initially, we detected a genome-wide significant QTL for SCS on BTA18 (Figure 1). The maxima of the QTL test statistic were in interval 9 (HAMP_c.366+109G>A -BMS833) and in interval 12 (BB710 -PVRL2_c.-1268G>C). In the same region telomeric on BTA18, several studies reported QTL for SCS. Kühn et al. [7], Brink [5] and Xu et al. [11] all identified a QTL for SCS in the German Holstein population at the telomeric end of BTA18 near marker TGLA227. Schulman et al. [10] reported a QTL for SCS as well as for mastitis at the telomeric end of BTA18 in Finnish Ayrshire cattle, and Ashwell et al. [50] localized a QTL for SCS in the middle to telomeric region on BTA18 at marker BM2078 in US Holstein, whereas Lund et al. [8] reported a QTL for SCS in Finnish Ayrshire, Swedish Red and White, and Danish Red with the maximum of the test statistic in the middle of BTA18 between the markers ILSTS002 and BMS2639. Interestingly Lund et al. [8] as well as Xu et al. [11] indicated that there might be more than one QTL for SCS on BTA18, but in both studies no significant evidence could be provided for a second QTL. Due to the lower marker density in previous studies and different approaches to detect QTL, a direct comparison between our results and previously mentioned studies is impeded. Nevertheless, the confidence intervals of the maxima observed in our LALD analysis did not include marker TGLA227 at the telomeric end of BTA18, and the QTL position reported by Lund et al. [8] as well as the assumption of a second QTL by Xu et al. [11] further in the middle of BTA18, indicated that we identified a second QTL for SCS in German Holstein cattle and possibly discovered a third QTL on BTA18 in our studies. Further indications for a third QTL on BTA18 in German Holstein cattle arise from association and haplotype analyses but no formal proof for a third QTL is given in our studies. The weak association observed for these markers is also in accordance to our LALD analysis, because interval 13 and 14 are within the genome-wide significance threshold in LALD analysis. Additionally, the second maximum observed in our LALD analysis in interval 12 is in an immediately adjacent interval and the putative QTL position in interval 12 is approximately 1 MB upstream of PVRL2_c.392G>A. For CALM3_c.3+1678C>T, only interval 16 (DIK4234 -CALM3_c.3+1678C>T) is within the chromosome-wide significance level in our LALD test statistic confirming the weaker association with SCS observed. To further test the results of our LALD analysis and the association of BMS833 genotypes with SCS on a population wide level, we investigated maternally inherited two-marker-intervals for interval 9 (HAMP_c.366+109G>A -BMS833) as well as for interval 12-13 (BB710 -PVRL2_c.392G>A). First we analysed effects of maternally inherited two-marker-interval gametes estimated based on IBD coefficients and second we performed a direct association analysis for maternally inherited two-marker-haplotypes of interval 9 and interval 12-13 without considering IBD coefficients. Summary of results from an association analyses for maternally inherited two-marker-haplotypes based on a model (1) including a fixed twomarker-haplotype effect and a random polygenic effect. Individual two-marker-haplotypes were compared to all other maternally inherited haplotypes in the data set. Haplotype frequencies were determined by direct counting.
For interval 9 (HAMP_c.366+109G>A -BMS833), maternally inherited two-marker-interval gametes showed differences in estimated effects for SCS ( Figure 2). The variance of estimated gametic effects was higher for twomarker-interval gametes in interval 9 compared to twomarker-interval gametes in interval 12-13 and single marker analyses revealed that the microsatellite marker alleles are the main force in discriminating the effects of maternally inherited two-marker-interval gametes ( Figure  2, Figure 3). Association analysis for maternally inherited two-marker-haplotypes of interval 9 showed that none of the maternally inherited two-marker-haplotypes in interval 9 (HAMP_c.366+109G>A -BMS833) are in LD with SCS in the German Holstein population. However, association analysis of BMS833 genotypes showed an association with SCS in our half-sib design indicating that the association is due to linkage but not linkage disequilibrium of the BMS833 locus with the causal mutation affecting SCS.
For interval 12-13 (BB710 -PVRL2_c.392G>A), the biggest difference in gametic effects estimated for maternally inherited two-marker-interval gametes was observed between 249-G (0.0378 (± 0.0018)) and 253-A (-0.0253 (± 0.0012)), where a positive mean of estimates indicates an unfavourable effect on SCS (high number of cells) and a negative mean of estimates indicates a favourable effect on SCS (low number of cells). Association analyses for the maternally inherited two-marker-haplotypes of interval 12-13 showed that 249-G (p = 0.027) is associated with SCS at the nominal 5% significance level in the German Holstein population. The weak association of PVRL2_c.392G>A genotype with SCS within the families we investigated and the association of maternally inherited two-marker-haplotype 249-G with SCS indicates that PVRL2_c.392G>A is not the causal mutation affecting SCS in German Holstein cattle, but the causal mutation has to be located near or within marker interval 12-13 (BB710 -PVRL2_c.392G>A). Combining results obtained for markers BMS833 and PVRL2_c.392G>A it still remains unclear whether there are two mutations (two QTL), one located near marker BMS833 (interval 9) and one near or within interval 12-13, or only one mutation, presumably in interval 12-13, affecting SCS in the middle part of BTA18.
The candidate gene polymorphisms we investigated were not the causal mutations affecting SCS in German Holstein cattle. However, the results of association analyses for single marker genotypes and maternally inherited twomarker-haplotypes indicated that HAMP and PVRL2 were selected within the region harbouring at least one QTL for SCS. Particularly PVRL2 still remains interesting. On the one hand, the association of the PVRL2_c.392G>A genotype and that of the 249-G haplotype suggests that another polymorphism within PVRL2 is the causal mutation affecting SCS in the German Holstein population. On the other hand, PVRL2 was selected as a gene with possible impact on mammary gland development or udder morphology and several studies have detected QTL for udder conformation on BTA18 [51,52]. Therefore, it is also possible that PVRL2 does not directly affect SCS but affects udder conformation traits like udder depth or fore udder attachment that are correlated with SCS [1].

Conclusion
In summary, our results suggest that the chromosomal region including interval 9 (HAMP_c.366+109G>A -BMS833) and interval 12-13 (BB710 -PVRL2_c.392G>A), in the middle to telomeric region on BTA18 has a strong impact on SCS in the German Holstein population. The analyses of maternally inherited two-marker-interval gamete effects and the association of the maternally inherited two-marker-haplotype 249-G of interval 12-13 (BB710 -PVRL2_c.392G>A) with SCS indicates that micosatellite marker BB710 could be a suitable candidatemarker for MAS, but association of microsatellite marker BB710 with SCS has to be verified. To confirm the association of the two-marker-haplotype 249-G with SCS and approve the hypothesis of two QTL in this region a further investigation is necessary. Thus, a mapping of udder conformation traits including a multiple-trait and multiple-QTL model might be useful to verify the existence of two QTL, and whether they are both directly affecting SCS or one is affecting a correlated trait. Likewise, a higher marker density within this region has to be achieved and families segregating for different BB710 alleles have to be identified. Hence, it might be useful to cover the region including interval 9 and interval 12-13 with equally distributed SNP to narrow down the position of the casual mutation or mutations affecting SCS in the German Holstein population by a further fine mapping approach.

Authors' contributions
BB carried out the genotyping work, the polymorphism screening, the linkage map construction, participated in the statistical analyses and drafted the manuscript. CB facilitated the statistical analyses by the development of software packages, participated in the statistical analyses and helped drafting the manuscript. MM participated in the development of software packages and the statistical analyses. NR participated in design and coordination of the study. CK devised the design of the study, coordinated the study, and participated in the statistical analyses and in drafting the manuscript. All authors read and approved the final manuscript.