| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
ANIMAL GENETICS |
* Statistical Animal Genetics Group, Institute of Animal Science, Swiss Federal Institute of Technology, ETH Zentrum, Zurich CH 8092, Switzerland; and
and
SUISAG, Allmend, CH-6204 Sempach, Switzerland
 |
Abstract |
|---|
 Top Abstract IntroductionMaterials and MethodsResultsDiscussionImplicationsLiterature Cited |
|---|
Key Words: Genetic Correlation • Growth • Meat Quality • Osteochondrosis • Pigs • Threshold Models
|
Introduction |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionImplicationsLiterature Cited |
|---|
; Jørgensen and Andersen, 2000). Additionally, OC lesions may cause or exacerbate leg weakness, another health problem in pigs (Häni et al., 1984; Lundeheim, 1987; Nakano et al., 1987). Both leg weakness and OC are of high economic, health, and welfare concern to countries with intensive pig production. Jørgensen and Andersen (2000) analyzed OC observed by x-raying live animals and reported heritabilities and genetic correlations among OC lesions, as well as between OC lesions and growth and feed efficiency. However, no study has considered genetic relationships between (pathologically scored) OC traits and meat quality and quantity traits.
Further, OC lesions are categorical traits (scores), which implies that threshold or nonlinear mixed models would be more appropriate than linear mixed models (e.g., Gianola and Foulley, 1983; Kadarmideen et al., 2000a,b; 2001). Threshold models have not yet been used to estimate genetic parameters for OC lesions in pigs.
Based on the foregoing arguments, the main objectives of this study were to estimate genetic and phenotypic parameters of OC lesions of station-tested pigs and to estimate their genetic and phenotypic correlations with meat quality and quantity, growth, and feed conversion traits. Linear mixed models (LMM) and generalized linear mixed model (GLMM) or threshold models were used to estimate parameters.
|
Materials and Methods |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionImplicationsLiterature Cited |
|---|
. Sibs of the selection candidates were station-tested to provide selected candidates for nucleus scheme in the station. Weight measurements on animals were taken at the beginning (approximately 30 kg) and at the end (approximately 103 kg) of the test. Animals were fed ad libitum, and feed consumed was known for test litters but not for individual pigs. All pigs were slaughtered using the facilities at the test station. Personnel of the test station took carcass measurements. Meat quality traits were measured in the longissimus dorsi muscle at the 10th dorsal vertebra.
Production, Meat Quality and Quantity, Growth, and Feed Conversion Traits
Several production and conformation traits are recorded on-station as a part of performance testing. Only the following production, meat quality and quantity, growth, and feed conversion traits were considered in this study: 1) Daily weight gain (DWG), which encompassed weight gain (g) on test from 30 to 103 kg of live weight. Weight at the end of test was computed from weight of the carcass assuming a ratio of carcass weight to live weight of 0.8. 2) Feed conversion ratio (FCR), which was defined as the ratio of weight of feed (kg) consumed to live weight gain (kg). The estimated individual feed intake was used to obtain an estimate of the individual feed conversion ratio. 3) Percentage of premium cuts (PPC); which was defined as the weight of ham, back, and shoulder without fat layer in proportion to carcass. The left side of each carcass was dissected and the cuts were weighed. 4) Percentage of i.m. fat (IMF); which was measured by infrared reflection with a Infra Alyzer 450 apparatus (Bran+Luebbe, Norderstedt, Germany). 5) The pH value at 45 min postmortem (pH1), which was measured with a pH-Meter 11-1 (Wintion AG, Gerzensee, Switzerland). 6) The pH value at 24 to 30 h postmortem (pH30), which was measured with a pH-Meter 11-1 (Wintion). 7) Reflectance measured at 24 to 30 h postmortem (H30), which was measured with a Unigalvo apparatus (Diffusion system Ltd., London, U.K.).
Osteochondral Disease
Recording of OC lesions in SUISAG began in 2002. Trained personnel at SUISAG conducted pathological/anatomical examinations of front and hind leg bones of slaughtered pigs and recorded OC lesions with a score of 1 to 6, with 1 = "normal" and 2 to 6 = "mildly to severely affected." Osteochondral lesions were observed in the following areas: head of humerus (HK), condylus medialis humeri (CMH), condylus lateralis humeri (CLH), radius and ulna proximal (RUP), distal epiphyseal cartilage of ulna (DEU), head of femur (FK), and condylus medialis femoris (CMF). Figure 1 shows the parts of bones examined manually for the presence or absence of different lesions and a minimum and maximum score given to each lesion. Traits DEU and CMF were scored from 1 to 6 because the lesions at these points were more variable than the others lesions (HK, CMH, CLH, RUP, and FK), which were scored from 1 to 4. More details and description of OC lesions in Swiss pig breeds are given by Schwörer et al. (1991).
|
Animals in this data set were progenies of 385 sires and 1,175 dams. The data set comprised information from eight farms, 25 yr-month of testing, 59 stable periods, 57 slaughter day x slaughter house combinations, four breeds (SLW, SLR, SLWSL, and Duroc) and two sexes (female and castrated male). There were 1,808, 464, 340, and 98 pigs in this data set that belonged to SLW, SLR, SLWSL, and Duroc breeds, respectively. Among the 1,291 pigs scored for OC, 838, 269, 158, and 26 pigs belonged to SLW, SLR, SLWSL, and Duroc breeds, respectively. There were 1,328 litter effects, which are relatively large in size with respect to 2,710 animals in the data set. This was due to the data structure; only station-tested pigs were scored for OC lesions and not their entire contemporary population.
In addition to the above-described data set, a sub-data set was prepared for the purpose of single-trait GLMM or threshold model analysis of categorical OC lesions. Only 1,291 animals scored for OC lesions were considered. Figure 2 shows the distribution of number of animals by scores (1 to 5) for all OC traits (no animal had a score of 6). A reasonable number of observations in different categories was found only for DEU. The other OC lesions had more than 95% of observations with a score of 1 (= normal). Therefore, scores for all OC lesions, except DEU, were recoded such that animals with a score of 1 received a score of 0 (= healthy), and animals with a score of 2 and above received a score of 1 (= diseased). This grouping of categories resulted in a binary data set (0 or 1) with healthy and diseased pigs. With this recoding, incidences of binary OC lesions (except DEU) were as follows: 1.4% for HK, 10.0% for CMH, 2.0% for CLH, 4.0% for RUP, 0.01% for FK, and 30.0% for CMF.
|
.
|
 | [1] |
where y1, y2, and y3 were a vector of records for OC lesions; meat quality or quantity, and DWG or FCR traits, respectively; X1, X2, and X3 were design matrices relating fixed effects in b1, b2, and b3 to y1, y2, and y3, respectively. The fixed effects in the model involving OC lesions (b1), meat quality or quantity (b2), and DWG or FCR (b3) were:
 |
where µ = the overall mean; Farm = farm where animals originated; SP = stable period, which is a time period nested within stable in testing station; SID = slaughter day in slaughter house; YM = year and month of performance test; Breed = consisting of Swiss Large White, Swiss Landrace, Swiss Large White sire line, and Duroc breed; Sex = castrated male and female; WET = weight at end of test as covariate; Age = age (in days) at slaughter as a covariate.
Design matrices Z1, Z2, and Z3 related records to random additive animal genetic effects in vectors a1, a2, and a3. Design matrices, W1, W2, and W3 related records to vectors of random (common) litter effects 
1, 2, and 3. The vectors of random residuals were e1, e2, and e3.
Model assumptions were:
 |
with variances and covariances:
 |
and
 |
where * denotes the direct product between two matrices; A is the numerator relationship matrix among the animals; G, L, and R denote variance–covariance matrices among the three traits for additive genetic, litter and residual effects, respectively; and I represents the identity matrix. These variance-covariance assumptions mean that there are nonzero covariances among elements of vector a, , and e for different traits. Covariances between a, , and e are assumed to be null. Because OC records were missing in some animals, the residual elements corresponding to rows and columns of animals that have missing observations in residual variance-covariance matrix R were set to zero. Each trait fitted in Model [1] was assumed to follow multivariate normal distribution including OC lesions. Note that the GLMM method described below takes the actual distribution of OC lesions into account. Henderson (1984) described multiple-trait animal models and handling of missing observations for one or more traits in multitrait model equations as well as distributional assumptions.
Generalized Linear Mixed Model
Generalized linear mixed models (e.g., McCullagh and Nelder, 1989) can be used to analyze data with different distributions (e.g., Normal, Binomial, and Poisson). Threshold models or GLMM have been developed to estimate or predict effects on postulated continuous underlying "liability" variable, on which linearity is imposed (e.g., Gianola and Foulley, 1983; Kadarmideen et al., 2000a,b; 2001). Here, univariate GLMM is applied to analyze each OC lesion, separately. The link between observed categorical OC lesions and underlying liability to OC lesions is a "threshold point." In GLMM terminology, this is defined by a function called the link function (e.g., McCullagh and Nelder, 1989), which relates E(y|ß, u, ) to the observed mean.
Let xi denote row i of X, zi denote row i of Z, and wi denote row i of W, then the expectation of each observation yi can be written as a function of the fixed and random effects (e.g., Gianola and Foulley, 1983; Kadarmideen et al., 2000b; 2001):
 | [2] |
This GLMM model for OC lesions consisted of exactly the same effects as fitted for OC lesions in LMM Model [1]. Here fi is the link function, linking the value 
i to E(yI). Binary OC lesions data (all but DEU) were analyzed by GLMM with two kinds of link functions, fi: logit link and probit link.
The logit link, modeling the probability that an animal has a given OC lesion [P(y = 1)], is given by:
 | [3] |
The probit link, modeling the probability that an animal has a given OC lesion [P(y = 1)], is given by:
 | [4] |
where 
–1 is an inverse Normal cumulative density function. This "probit" model is essentially the "threshold model" used in animal breeding (e.g., Gionola and Foulley, 1983; Kadarmideen et al., 2000b; 2001).
The link function used for count (Poisson) data (here DEU) was a log link:
 | [5] |
Implementation
Variance-covariance parameters for all models were estimated using the software package ASREML (Gilmour et al., 2001). Heritabilities (h2), litter variances (L2) and all correlations (genetic-rg, litter-r, residual-re, and phenotypic-rp) were computed using estimated variance-covariance matrices. Variance-covariance components for the OC trait, FK, could not be estimated because this trait had extremely low incidence (0.01%). Generalized linear mixed models (logit or probit animal models) applied to most binary OC lesions (all but DEU) did not yield converged likelihood (or converged to illogical estimate), either due to very low incidences or due to animal models used or both. Estimated h2 for all OC lesions (by LMM) was mostly nonsignificant and therefore dropped in GLMM analysis. All final GLMM analyses were based on a sire model without relationship among sires and without litter effects.
Heritabilities using GLMM sire models were calculated as follows:
Probit model:
 |
Logit model:
 |
Poisson (Log) model:
|
|
where
and
were estimated sire and residual variances, respectively.
For binary data, estimates of h2 on the liability scales can be transformed to observed scale using the formula of Robertson and Lerner (1949) as follows:
 | [6] |
where
is the heritability on the observed (0/1) scale,
is the estimated heritability on the liability (logit or probit or Poisson) scale, p is a proportion of OC lesions in the data, and z is the height of the ordinate of normal distribution corresponding to a truncation point applied to p proportion of OC lesions.
|
Results |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionImplicationsLiterature Cited |
|---|
. Results showed that the SLWSL breed was more affected with OC in CMH and RUP, the Duroc breed was more affected in DEU and CMF, and the SLR breed was more affected in the CLH regions. There were no sex differences (Table 2), except for CMH and CMF lesions.
|
Heritability and Litter Variance for OC Lesions
The LMM estimates of h2 and the ratio of litter variance to L2, together with their standard errors are given in Table 3. For OC lesions, h2 estimates were low and ranged from 0.06 for HK to 0.16 for CLH with SE ranging from 0.05 to 0.09. Estimated genetic/phenotypic variance for FK was zero. Estimates of L2 were zero for HK, RUP, and CMF. For other OC traits, L2 ranged from 0.04 to 0.14, with SE ranging from 0.06 to 0.08.
|
. As mentioned earlier, for binary OC lesions (all but DEU), two GLMM models (logit and probit models) were fitted. For logit models, the highest h2 was for CMF (0.28), and the lowest h2 was for DEU (0.04), with SE ranging from 0.01 to 0.06. The L2 was estimable only for DEU and CMH (results are not shown). For probit models, the highest h2 was for CMH (0.42), and the lowest h2 was for CLH (0.18), with SE ranging from 0.04 to 0.32. The GLMM estimates of h2 on the logit and probit scales were transformed to the observed scale, using Equation [6]
. Heritabilities on the same observed scale were similar for GLMM and LMM methods. Generalized linear mixed model estimates of h2 on the original liability scale (i.e., logit or probit) were much higher (range 0.04 to 0.42) than LMM estimates of h2 (range 0.06 to 0.16), as would be expected from the theory.
|
. Only statistically significant genetic correlations (P < 0.05) and are of moderate magnitude (absolute value of at least 0.20) are discussed. There were 13 correlations that met these criteria (values shown as bold type in Table 5). Some unfavorable genetic correlation patterns were observed between HK and FCR (rg = 0.40; increased incidence of HK was associated with increased FCR, whereas genetic selection was for low FCR), CMH and DWG (rg = –0.44; increased incidence of CMH was associated with decreased DWG), CLH and PPC (rg = 0.21; genetic merit for high meat quantity increased incidence of CLH), CLH and pH1 (rg = –0.54; increased incidence of CMH was associated with very low pH1 values, genetic selection would be intended for optimal but not extreme pH values), RUP and PPC (rg = 0.32; genetic merit for high meat quantity increased incidence of RUP), and CMF and H30 (rg = 0.31; increased incidence of CMF lesions was associated with high reflectance or pale meat, whereas genetic selection is mostly for low incidence of pale color meat).
|
) between OC lesions and other traits were all close to zero or nonsignificant (hence results not shown). Residual correlations, re, (regardless of statistical significance and magnitude) ranged from –0.14 for HK-FCR to 0.15 for IMF-CMH (results not shown in tables). Phenotypic correlations (rp) of OC lesions with meat quality, PPC and growth traits (regardless of statistical significance and magnitude) ranged from –0.07 for H30 and DEU to 0.14 for DWG and CMF, with SE in the range of 0.03 to 0.04. All rp mostly followed similar direction of relationship as genetic correlations (results of rp are not shown).
Correlations Among OC Lesion Traits
Genetic and phenotypic correlations (together with their SE) among OC lesions are given in Table 6. Among 15 estimates of rg, eight estimates were significant with |rg|
0.20 and are shown as bold type in Table 6. Genetic correlations that were significantly positive (range 0.57 to 0.69) were for the following combinations of OC lesions: HK–CMF, CMH-CLH, and CMH-RUP. This could be expected because these OC lesions were in the front leg bones (except CMF), and may have similar genetic factors influencing the occurrence of these lesions (Figure 1). Genetic correlations were negative (range = –0.21 to –0.40) between the following traits: HK-CMH, HK-RUP, CMH-DEU, CLH-DEU, and RUP-DEU. This indicates that these lesions are less likely to occur at the same time and/or genetic factors that influence the development of lesions may be different.
|
Genetic Parameters for Meat Quality and Quantity, Growth, and Feed Conversion Traits
Estimated h2 and the ratio of L2 to phenotypic variance, together with their SE are given in Table 3. The highest h2 was for IMF (0.66) and the lowest h2 was for pH1 (0.12), with SE ranging from 0.05 to 0.08. Estimates of L2 were the lowest (0.01) for H30 and highest (0.06) for IMF, with SE around 0.03. Estimated h2 and L2 for meat quantity (PPC) were 0.60 and 0.05, respectively. Estimated h2 and L2 for DWG were 0.28 and 0.11, respectively. Estimated h2 and L2 for FCR were 0.42 and 0.04, respectively.
Genetic correlations between meat quality, PPC, and growth traits are given in Table 7. Ten out of 21 genetic correlations were significant with |rg| 0.20 (shown as bold type in Table 7). Increased DWG was genetically correlated with decreased pH1 in carcass (rg = –0.31; very low pH1 being unfavorable), low FCR was associated with high PPC (rg = –0.66; favorable), and increased FCR was associated with increased IMF (rg = 0.34) and pH1 (rg = 0.32). This indicates that genetic selection for low FCR would have favorable genetic effects on IMF and pH1. Genetic correlation patterns between meat quantity and quality traits were: increased PPC was correlated with decreased IMF (–0.34), pH30 (–0.36), and H30 (–0.26). Increased IMF was highly correlated with low pH1 (rg = –0.39) and high reflectance (rg = 0.37). Traits pH 30 and H30 were negatively correlated (rg = –0.25), as expected (high pH increases darkness of the meat resulting in low reflectance).
|
|
Discussion |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionImplicationsLiterature Cited |
|---|
, Lundeheim (1987), and Jørgensen (1995). Multitrait analysis of osteochondrosis with other traits in this study was based on the hypothesis that osteochondrosis, which is believed to be a cause of leg weakness, would also have unfavorable (genetic) correlations with those traits. Results from this study indicated that this hypothesis could be true. The knowledge of inheritance of OC lesions and coinheritance or genetic correlations of these OC lesions with production and several meat quality traits would be essential if future breeding programs would include OC lesions (or leg weakness) in multitrait breeding goals (with other traits in the goal being, for example, production and meat quality traits).
Häni et al. (1984) studied the Swiss pig population for a relationship between OC and meat quality traits. The difference between this study and that of Häni et al. (1984) was that the latter used a sib-correlation method, which is not based on the currently used animal models of Henderson (1984). Further, Häni et al. (1984) study was based on data collected 20 yr ago (in 1984) and did not consider as many meat quality and quantity and growth traits as were considered here. The fundamental difference between our study and the study of Jørgensen and Andersen (2000) is that they did not estimate genetic correlations between OC lesions and meat quantity, meat quality traits (pH1, pH30, IMF, H30), and feed conversion ratio, which were estimated here. Additionally, their study used radiological (x-ray) data on OC lesions rather than actual pathological–anatomical examination of slaughtered animals.
Osteochondral Disease: Incidences and Heritabilities
Very low incidence of mild to severe OC lesions (Figure 2) was similar to the incidence found in Swiss pigs approximately 20 yr ago by Schwörer et al. (1991) and in Danish pigs by Jørgensen (1995). Jørgensen (1995) found no significant differences between sexes for OC incidences, which is similar to our findings. Occurrence of OC lesions was more frequent in SLW than SLR breed, similar to the reports of Schwörer et al. (1991). Estimated heritabilities for OC lesions from LMM in this study were lower than those reported by Jørgensen and Andersen (2000). This is the first study that applied threshold models to the analysis of osteochondral lesions in pigs, and hence our results could not be compared with any study so far. Kadarmideen et al. (2000b) had shown that threshold animal models for binary traits were computationally more unstable (and statistically less appropriate) than threshold sire models and provided possible reasons. One of the reasons was the very low rate of incidence, which was the case in this study. Hence sire (threshold) models were used here.
Genetic Correlations Between OC Lesions and Meat Quality and Quantity and Growth Traits
Overall OC lesions showed unfavorable genetic correlations with meat quantity (PPC). Statistically significant genetic correlations (and above the absolute value of 0.20) show that some OC lesions also have unfavorable genetic effects on meat quality and production traits. Unfavorable (significant) genetic correlations between DWG and CMF (0.31; Table 5) were also found by Jørgensen (1995) and Jørgensen and Andersen (2000). No genetic correlations have been reported for feed conversion ratio and OC lesions; therefore, our results (high FCR related to high incidence of HK lesions; rg = 0.40) could not be compared. However, this is indeed an important result that should be considered in the breeding program.
Genetic and Phenotypic Correlations Among OC Lesions
Osteochondral lesions on the humerus were correlated with changes in the proximal radius: proximal edge of radius and synovial fossa of radius. However, there were no significant correlations between osteochondrosis in the elbow and stifle joints (see Figure 1 for names and positions). Some correlations among OC lesions were nonsignificant or had high SE, which was due to the fact that the prevalence of severe joint lesions (osteochondritis dissecans) was very low.
Genetic correlations between different OC lesions mostly agreed with those reported by Jørgensen and Andersen (2000). In general, low residual and phenotypic correlations among OC lesions suggest that, although these traits are genetically related, they might have different environmental or nongenetic causes. Similar findings were reported by Jørgensen and Andersen (2000).
General
Estimated genetic and phenotypic parameters for meat quality and quantity traits, growth rate, and FCR were similar to the findings of Hermesch et al. (2000a,b) and Knapp et al. (1997).
Genetic and phenotypic correlations between osteochondrosis and meat quality, growth, and feed conversion traits reported in this study could be useful in selection. Some of the genetic correlation estimates were found to be statistically nonsignificant or had poor accuracy in this study. Significant and accurate estimates for OC (especially for genetic correlations) could be obtained if more animals are performance tested and scored for OC lesions for many more years. This may, however, be financially prohibitive for the commercial pig breeding sector (currently approximately 3,000 animals are performance tested). Jørgensen and Andersen (2000) used radiological (x-ray) data of osteochondrosis lesions taken on live animals. This x-ray option opens up the possibility of performance testing a large number of animals for osteochondrosis lesions in less time and probably at low cost.
|
Implications |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionImplicationsLiterature Cited |
|---|
|
Footnotes |
|---|
2 Correspondence – phone: +41 (0)1-632-3266; fax: +41 (0)1-632-1260; e-mail: haja.kadarmideen{at}inw.agrl.ethz.ch.
Received for publication April 5, 2004. Accepted for publication July 14, 2004.
|
Literature Cited |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionImplicationsLiterature Cited |
|---|
This article has been cited by other articles:
|
|
 |
|
  H. N. Kadarmideen and L. L. G. Janss Population and systems genetics analyses of cortisol in pigs divergently selected for stress Physiol Genomics, March 14, 2007; 29(1): 57 - 65. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 H. N. Kadarmideen and L. L. G. Janss Evidence of a Major Gene From Bayesian Segregation Analyses of Liability to Osteochondral Diseases in Pigs Genetics, November 1, 2005; 171(3): 1195 - 1206. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 J. Arango, I. Misztal, S. Tsuruta, M. Culbertson, and W. Herring Study of codes of disposal at different parities of Large White sows using a linear censored model J Anim Sci, September 1, 2005; 83(9): 2052 - 2057. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
