| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
ANIMAL GENETICS |
* IPG, Institute for Pig Genetics, PO Box 43, 6640 AA, Beuningen, the Netherlands;
and
Animal Breeding and Genetics Group, Wageningen University, PO Box 338, 6700 AH, Wageningen, the Netherlands
 |
Abstract |
|---|
 Top Abstract INTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONIMPLICATIONSLITERATURE CITED |
|---|
Key Words: meat quality • pig • quantitative trait loci • SSC4 • SSC11 • variance component
|
INTRODUCTION |
|---|
|
TopAbstractINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONIMPLICATIONSLITERATURE CITED |
|---|
; Hu et al., 2005; Wang et al., 2005). The analysis methods usually applied are based on regression (Knott and Haley, 1992; Knott et al., 1996) and are aimed at detecting linkage between markers and QTL. Unless the genotyped population is very large, the linkage analysis approach generally results in large confidence intervals for the QTL because of a relatively small number of meioses. Implementation of QTL in marker-assisted breeding programs often requires a greater map resolution. Genotyping of additional markers, eventually on additional families, might be used as a first step to increase the map resolution of QTL.
Recently, linkage disequilibrium (LD) mapping methods have been proposed for fine mapping, which aim at capitalizing on historical recombination events (Riquet et al., 1999; Meuwissen and Goddard, 2000). The classical linkage and LD analysis are complementary, and methods have been proposed that simultaneously model linkage and LD information (Meuwissen et al., 2002; Farnir et al., 2002). The method of Meuwissen et al. (2002) allows for simultaneous estimation of variance components for systematic, polygenic, and QTL variance. Compared with a classical regression analysis, the method can deal with additional relationships within and between families, increasing the power to detect QTL. Furthermore, utilizing LD information may result in a more accurate mapping of QTL. The actual contribution of LD depends on the patterns of linkage disequilibrium under study and the marker density. It has been shown that LD can extend over map distances of more than 10 cM in pigs (Nsengimana et al., 2004).
The aim of the current study was to confirm and fine map QTL on SSC4 and SSC11 by applying a combined linkage and LD analysis method on an increased data set.
|
MATERIALS AND METHODS |
|---|
|
TopAbstractINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONIMPLICATIONSLITERATURE CITED |
|---|
Genetic Material and Phenotype Measurements
The population used in this experiment was created by mating 17 sires of a synthetic Piétrain/Large White halothane free boar line (TOPIGS, the Netherlands) to 239 commercial crossbred sows with unknown pedigree. All piglets were born over a 2-mo period. Phenotypic measurements were taken for 30 traits on approximately 100 offspring per sire, resulting in a total of 1,855 animals recorded. A detailed description of the population and traits was presented by van Wijk et al. (2005, 2006).
To obtain a better phenotypic description of traits of interest, 9 additional variables were calculated based on the primary phenotypic measurements. These variables will be termed as derived variables. Calculation of these derived variables was as follows: average Japanese color score (AvgJCS) was calculated as an average of the Japanese color score measured at the rib and cut (JCScut) surface of the LM, assuming that both measurements could be considered as repeated measurements of the same trait. Japanese color scores are subjective observations, whereas Minolta measures are objective color measurements.
The Minolta measurements were used to calculate, using multiple regression, an estimated color (EstCol) variable explaining the AvgJCS following the formula: EstCol = [5.48 – (0.067 x Minolta L* LM) – (0.056 x Minolta b* ham) + (0.080 x Minolta b* LM) + (0.049 x Minolta a* ham)], where Minolta L* LM and Minolta b* LM were taken on the fresh cut surface of a 2.5-cm chop removed from the sirloin end of the boneless center cut loin using a Minolta CR 300 colorimeter set at C illuminant (Minolta camera, Osaka, Japan). Minolta b* ham and Minolta a* ham were taken from the fresh cut surface of the inside ham muscle using the same system. A mean value for the subjective and objective color measurements was calculated as the total color score (TotCol), following the formula: TotCol = (EstCol + AvgJCS)/2.
The chroma (C) and hue (H) values were calculated for the ham (hamC and hamH) and LM (loinC and loinH) measurements as: C = 
(Minolta a*2 + Minolta b*2), and H = tan–1(Minolta b*/Minolta a*). Chroma is a measure of color intensity, which increases when Minolta a* or Minolta b*, or both, increases. Hue indicates the degree of color change from red (low values of hue) to yellow (high values of hue; Setser, 1984). Finally, ham and LM gain (HamG and LoinG) were calculated as HamG = (boneless ham weight/cold carcass weight) x ADG, and LoinG = (domestic LM weight/cold carcass weight) x ADG.
Genotyping and Linkage Map Construction
Eight paternal half-sib (PHS) families had been used in the initial genome scan, with 6 and 4 markers on SSC4 and SSC11, respectively (van Wijk et al., 2006). Additional genotypes were generated in 2 steps. First, 9 additional PHS families were typed for the 10 markers on SSC4 and SSC11 that already had been used in the initial genome scan. Second, the whole population of 17 PHS families was typed for 18 additional markers on the 2 regions of interest. Nine of the markers are located on SSC4q and 9 on the p-arm and centromeric region of SSC11. The typed markers were selected based on their position in the 2 regions of interest, their informativeness, and scoring ability.
For SSC11, no informative markers were available for the ~25-cM interval between markers Sw1632 and S0071. Furthermore, to cover a larger genome region on SSC11, 7 markers were included that were located distal from marker S0071. The reasoning was that the QTL for LM weight with the greatest F-statistic were at marker S0071 in the initial genome scan. Genotypes were scored in duplicate and checked against pedigree information. The chrompic option of CriMap (Green et al., 1990) was used to check for double recombinants before final linkage map construction based on the Kosambi mapping function. Sex-average linkage maps with 15 and 13 markers on SSC4 and SSC11, respectively, were used in the QTL analysis.
Statistical Analysis
Linkage Analysis.
Linkage analysis (PHS) was performed using a classical regression interval analysis nested within half-sib families (Knott et al., 1996; de Koning et al., 1999) on phenotypic trait data precor-rected for systematic effects, as described by van Wijk et al. (2006). Segregating sire families were identified by individual family analysis. Significant thresholds were determined empirically for each trait by chromosome combination by performing 10,000 permutations (Churchill and Doerge, 1994).
Variance Component Analysis.
Alternatively, variance component-based linkage analysis (VC LA) and combined linkage and linkage disequilibrium analysis (LDLA) was performed using the method proposed by Meuwissen and Goddard (2000). The method models expected covariances between haplotype effects, which are proportional to linkage disequilibrium in the population, at a postulated QTL position. The method involves the following steps:
); using the LDLA package (Janss and Heuven, 2005). Calculation of identity by descent (IBD) probabilities for each putative QTL position given the marker scores results in a series of IBD probability matrices. Different from Meuwissen and Goddard (2001), pedigree information was not included when calculating IBD probabilities between base (parental) haplotypes. Identity by descent probabilities and likelihoods were evaluated at 4 points (putative QTL positions) within each marker bracket (i.e., evaluation points at 1, 25, 50, and 75% of the distance from the first marker to the next marker); and ) was used to calculate the maximum likelihood at each evaluation point using the appropriate IBD matrix. Phenotypes were analyzed using the following model:
 | [1] |
where Y is the vector of phenotypes, b is a vector of systematic effects, u is a vector of random additive polygenic effects of background loci, v is a vector of random additive effects due to the QTL, c is the vector of random litter effects, and e are the random residuals. The random effects were assumed normally distributed with mean zero and variances 
, 
, 
, and 
, respectively. The X, Z, W, and S are known incidence matrices for the effects of b, u, v, and c respectively.
In model [1], denoted as the full model, v represents the combined maternal and paternal haplotype effects as a single additive component (a single variance component). Maternal (vd) and paternal (vs) haplotype effects can be modeled separately (2 variance components) to allow for differences in effect (i.e., fitting parent-of-origin or breed-specific effects, or both):
 | [2] |
Considering the maternal (Wdvd) or paternal (Wsvs) component separately allows for a maternal- or paternal-only analysis.
Assuming unrelated base (parental) haplotypes (ignoring LD information by adjusting the IBD probabilities between base haplotypes to zero) results in a VC LA. Comparison of VC LA results with LDLA results gives insights into the contribution of LD. A paternal VC LA analysis resembles a paternal half-sib regression analysis.
Significance testing was based on a likelihood ratio test (LRT). The LRT was calculated as twice the difference between the ln(L) of the QTL model vs. the base model (model without a QTL effect). The 5% nominal significance level of the LRT was determined from a 
2 distribution of 1 df (model [1]) or 2 df (model [2]; df equal to the number of associated QTL effects).
|
RESULTS AND DISCUSSION |
|---|
|
TopAbstractINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONIMPLICATIONSLITERATURE CITED |
|---|
. The 15 markers on SSC4 span 91 cM, with an average marker interval of 6.1 cM. The 13 markers on SSC11 span 87 cM, with an average marker interval of 6.7 cM. The average information content along the chromosome was 0.69 for SSC4 and 0.68 for SSC11. The map of SSC11 contains a large gap of 31 cM. Available microsatellites in that region were tested but not informative on the sires. Generally, marker order was in good agreement with the USDA-MARC.2 genetic linkage map (Rohrer et al., 1996) except for SSC11 where marker Sw486 was positioned ahead of markers Sw1452 and Sw435. The 3 markers in the USDA map span an interval of only 2 cM, indicating tight linkage, and a different order is therefore possible. Map lengths were slightly greater in our data as compared with the USDA genetic linkage map, 21 and 10 cM for SSC4 and SSC11, respectively.
|
Regression Analysis
The classic paternal half sib (PHS) regression analysis on the increased data set identified 14 QTL at the nominal P = 0.05 threshold, from which 9 were on SSC4 and 5 on SSC11 (Tables 2 and 3). The nominal significance level was used to facilitate comparison of results between regression and VC analysis. Two of the QTL were significant at the P = 0.05 chromosome-wise level [i.e., JCScut on SSC4 and LM marbling (LMARB) on SSC11].
|
|
). The F-statistic profile was relatively flat with a wide interval of approximately 30 cM above the threshold. The extended population contributed to additional evidence for the presence of the QTL for JCScut on SSC4. Individual family analyses revealed that only 3 sires were heterozygous for the QTL affecting JCScut.
|
).
Although the power of this study was increased considerably (~25%) compared with the initial genome scan (Van Wijk et al., 2006), the suggestive QTL for LM weight at the centromeric region of SSC11 as identified in the initial genome scan (nominal P = 0.001) could not be confirmed. This QTL for LM weight was one of the most significant QTL identified in the initial genome scan but must be considered as a false positive based on this analysis.
Variance Component Analysis
The variance component QTL results are presented in Tables 2
and 3 for SSC4 and SSC11, respectively. For the VC method we used a likelihood ratio test without taking into account that multiple traits were analyzed and multiple tests were performed along the chromosome. In this study the nominal 5% significance level was applied. This threshold is used in similar studies (de Koning et al., 2003; Olsen et al., 2004; Uleberg et al., 2005; Gautier et al., 2005) and facilitates comparison of results. Estimates of the total variance explained by the QTL (hqtl2) are also presented in Tables 2 and 3. The hqtl2 were considerable and ranged from 1.8 to 13.2 percent. However, no clear relationship was observed between the significance and estimated QTL effects.
Variance Component Linkage Analysis:
The VC LA resulted in the identification of 10 effects (6 on SSC4 and 4 on SSC11) at the nominal level (P = 0.05).
On SSC4, the most significant effect (P < 0.01) was found for Minolta L* ham (HAML). Maternal- and paternal-only analyses revealed that the effect was segregating in the sire line (data not shown). This QTL, however, has not been identified with the paternal half-sib regression analysis. In this case, the 2 methods that were expected to give very similar results, PHS and VC LA paternal model, differed.
The QTL for JCScut on SSC4, which was most significant in the regression analysis, did not reach the significance threshold under the full model (Table 1 in italics, and Figure 1) but passed the threshold under the paternal only model (not shown). The F-statistic and LRT profiles were very similar, with the latter having an increased resolution (Figure 1).
On SSC11, the most significant effect was found for LMARB. This result corresponds with the regression analysis.
The VC LA analysis resulted in additional QTL compared with the regression linkage analysis due to combining paternal and maternal information. The increased number of significant effects under the full model nicely corresponded to results obtained with a maternal or paternal model (data not shown), which showed that the effects were generally either of maternal or paternal origin. Independent modeling of the paternal or maternal haplotype effects reveals the considerable contribution of the dam haplotypes to the variance of the measured traits. This is in agreement with the observed greater allelic diversity within the dams, which fits with the use of 3-way crossbred individuals as dams mated with pure line sires.
Linkage and Linkage Disequilibrium Analysis.
The LDLA analysis resulted in identification of 19 significant effects for 17 traits (Tables 2 and 3). The increased number of significant effects shows a considerable contribution of LD. For Chroma (HamC and LoinC) combining of information showed a positive effect on the power, resulting in increased LRT values compared with the individual underlying traits. However, LRT profiles of the derived variables were in general very similar to the LRT profiles of the underlying traits and are therefore not discussed. Accumulation of the different color measurements into compound variables did not lead to new disclosures.
The analysis for SSC4 revealed 10 effects (P < 0.05) for 8 traits (Table 2). Similarly, as in the VC LA analysis, the most significant effect was found for Minolta L* ham (HAML). The 2 related traits Minolta a* (LOINA) and chroma LM (LoinC) very clearly showed a 2-peak profile, suggesting 2 QTL affecting those traits. Likelihood ratio test profiles for JCScut and the 2 significant Minolta traits (HAML and LOINA) are presented in Figures 1 and 2, respectively.
|
). The LRT profiles for the different color traits (Figures 1 and 2) may suggest the presence of 2 QTL on the chromosome, with one of them located around the third marker with pleiotropic effect on the different color measurements. However, a 2 QTL model did not provide evidence for the presence of 2 QTL affecting the color traits (data not shown).
The QTL for JCScut is in agreement with previous findings. Ovilo et al. (2002) and Nii et al. (2005) reported meat color QTL in the same region. de Koning et al. (2001) and Malek et al. (2001) published meat color QTL on SSC4 positioned more to the centromere or distal on SSC4, although confidence intervals are usually large in a regression analysis with marker intervals that are commonly used in a genome scan.
The QTL for boneless ham weight (BHAM) is in line with the many QTL affecting different muscle mass and back fat measurements reported on SSC4 (see PigQTLDB; Hu et al., 2005).
The LDLA results for SSC11 are presented in Table 3. A total of 9 effects were identified at the nominal P < 0.05 level. Likelihood ratio test profiles for the meatiness traits LM depth (LMd), meat percentage (PLEAN), and ADG are presented in Figure 3. Profiles for the marbling traits are presented in Figure 4.
|
|
The LRT profiles for the meatiness traits LM depth (LMd), meat percentage (PLEAN), and ADG (Figure 3) are very similar and suggest a single QTL affecting the different traits.
Most effects on SSC11 were not identified in the preceding genome scan (Van Wijk et al., 2006) because the 40 cM region beyond marker S0071 was not covered.
Comparison of Analysis Methods
In the literature, few studies have reported on a comparison of data analyzed with the different methods (de Koning et al., 2003; Nagamine et al., 2004; Grapes et al., 2004; Kolbehdari et al., 2005). The regression and variance component method have similar power in a simple pedigree structure. In complex pedigrees, the variance component method is thought to achieve greater power to detect QTL. Furthermore, the variance component method uses potential information from segregation on the maternal side (Nagamine et al., 2004). Other advantages of the VC method are simultaneous estimation of (non)genetic effects, less sensitivity for small family sizes or less informative markers, or both (Kolbehdari et al., 2005). Also, the VC method provides estimates of the effect of each haplotype, which links up with breeding value estimation.
Significant QTL obtained with the regression analysis were not necessarily significant in the VC LA analysis or vice versa, although, results of both methods were comparable. Ten traits were (near) significant in both analyses, whereas 4 traits were significant in the regression analysis and not in the VC LA analysis (Tables 2 and 3). Another 4 traits were significant in the VC LA analysis and not in the regression analysis, all segregating on the maternal side, which explains why these were not found in the regression analysis.
The LDLA analysis resulted in the identification of an increased number of significant effects compared with the VC LA or regression analysis. Seventeen traits showed an increase of LRT value for LDLA compared with VC LA. Despite the greater number of significant effects found with the LDLA analysis, LD did not always positively influenced the LRT value. Six effects were reduced by inclusion of LD. A reason might be that the LDLA method assumes that all individuals are descendants from a common base population. This is not necessarily true for a cross between different breeds. Uleberg et al. (2005) presented an adjusted version of the LDLA method accounting for the fact that parents from different breeds do not (necessarily) descend from a common base population. Uleberg et al. (2005) ignored the IBD probabilities between base individuals of the 2 parental breeds. In this way base individuals of the different breeds are considered completely unrelated despite very similar haplotypes. However, it is questionable whether haplotypes in different breeds can be considered completely unrelated in reality. In order to use crossbred data more correctly, further developments of the LDLA method may be required. In the current method, alike in state probabilities estimated based on the genotypes are used in calculation of IBD probabilities. In crossbred data alike in state probabilities estimated based on sire or dam genotypes could be used to calculate IBD probabilities between pairs of haplotypes of paternal or maternal origin.
The expectation is that LD in particular will contribute in case of greater marker densities. The effective marker density will depend on the LD in the population. Nsengimana et al. (2004) investigated the extent of LD in 2 genomic regions on SSC4 and SSC7 and concluded that genome-wide association studies are feasible in commercial pig populations at marker densities of 5 to 10 cM. In our study, no common haplotypes, shared by segregating sires, could be identified associated with the effects for JCScut of LMARB. Based on this study, it can be concluded that greater marker densities are required for the identification of common haplotypes across families associated with segregating effects.
|
IMPLICATIONS |
|---|
|
TopAbstractINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONIMPLICATIONSLITERATURE CITED |
|---|
|
Footnotes |
|---|
2 Corresponding author: Rik.van.Wijk{at}ipg.nl
Received for publication February 3, 2006. Accepted for publication August 23, 2006.
|
LITERATURE CITED |
|---|
|
TopAbstractINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONIMPLICATIONSLITERATURE CITED |
|---|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
