| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
ANIMAL GENETICS |
Department of Animal Science, University of Padova, Viale dell’Università 16, 35020 Legnaro (PD), Italy
 |
Abstract |
|---|
 Top Abstract INTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONLITERATURE CITED |
|---|
Key Words: dry-cured ham • heritability • preweaning mortality • piglet • survival analysis
|
INTRODUCTION |
|---|
|
TopAbstractINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONLITERATURE CITED |
|---|
). Enhancing piglet survival is relevant both for ethical and economic implications and for the acceptability of the production system by the consumer. Additive genetic variation of piglet preweaning survival is large enough to be exploited in specific breeding programs, but selection for improvement of direct survival is difficult due to the binary nature of the trait and large environmental variance (Knol et al., 2002)
The threshold model is the method of choice for the analysis of binary traits (Gianola, 1982) and can account for the categorical nature of survival, but it suffers from a severe loss of information because piglets dying at d 1 or in wk 2 after birth are treated alike (Casellas et al., 2004). The analysis of failure time makes use of all available information and does not restrict observations to an arbitrarily defined point (Ducrocq, 1997). The availability of survival analysis techniques offers the opportunity of new approaches to the investigation of preweaning piglet survival (Ducrocq et al., 1988)
The relationships of piglet survival with production traits have been investigated in lines selected for increased efficiency of lean meat production. In Italy, the breeding goal for boar lines originating slaughter pigs for dry-cured ham production differs greatly from the one pursued to enhance efficiency of pork production, and the relationship between the breeding goal and piglet survival is currently unknown
The aims of this study were to investigate sources of variation of piglet preweaning survival in a crossbred slaughter pig population, the relationship of crossbred piglet survival with a total merit index (TMI) used for selection of breeding candidates in a Large White boar line, and to estimate variance components and genetic parameters through survival analysis techniques.
|
MATERIALS AND METHODS |
|---|
|
TopAbstractINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONLITERATURE CITED |
|---|
Animals and Data
Data used in this study were collected in the sib testing program of the C21 Large White boar line (Gorzagri) from 2000 to 2006. Boars from the C21 line are used on commercial farms as sires of crossbred pigs, which are fattened and slaughtered at heavy BW (165 kg) for production of dry-cured hams. In the sib testing program of the C21 line, crossbred paternal half sib families are produced by mating C21 nucleus boars to a group of crossbred sows, which are submitted to minimum intensity replacement policies. Crossbred sows originated from a cross involving boars of a synthetic line, derived from Large White and Pietrain breeds, and sows of a Large White line selected for maternal ability and prolificacy. Crossbred paternal half sib families provide the genetic evaluation program of C21 purebred breeding candidates with crossbred half sib phenotypes for quality traits of raw and dry-cured hams. Besides growth and residual feed efficiency, the breeding goal of the C21 line includes traits related to the quality of dry-cured ham. Selection is addressed to an intermediate optimum for marbling and for the amount of subcutaneous fat evaluated on the raw ham, to enhance the quality of fat covering, to reduce excessive ham roundness, and to reduce curing weight losses at a fixed level of dry-cured ham quality.
Data on survival of piglets at birth and up to weaning were routinely collected in the sib testing program since 2000 and included birth litter description (sow identification and parity, sire, date of farrowing, and size of the litter at birth), and individual piglet information (identification, sex, stillborn or alive at birth, weaning date or date of death if the piglet died during the suckling period, and date of transfer and foster dam identification for cross-fostered piglets). Cross-fostering occurred for 46% of piglets and was of similar proportion for male and female piglets. For piglets that died before weaning, survival time was computed as the difference between the date of death and date of birth, whereas for piglets still alive at weaning, survival time was computed as the difference between the date of weaning and the date of birth, but all of these records were considered censored records.
After application of editing procedures, which aimed to discard records with incomplete or inconsistent information (120 piglets) and with unknown sire (105 piglets), a total of 13,924 individual survival records of piglets (1,347 litters) sired by 189 C21 boars mated to 328 crossbred sows were available for the study.
Survival Analysis
The individual piglet survival time was analyzed using survival analysis methodology. Preliminarily, the survivor function for the general population was estimated by the Kaplan-Meier method (Kaplan and Meier, 1958). The linear regression of ln{–ln[S(t)]} on ln(t), where S(t) is the Kaplan-Meier estimated survivor function, was considered to check the suitability of the assumption of a Weibull baseline hazard (Ducrocq et al., 1988). Because the relationship between ln{–ln[S(t)]} and ln(t) was not linear, the assumption of a Weibull distribution function for the baseline hazard was not suitable for these data and alternative models were considered. All of these models were from the group of proportional hazard frailty models (Cox, 1972) and of the general form:
where h(t)is the hazard of death at time t (age of piglets), h0(t) is the baseline hazard function, β1 is an unknown vector of fixed regression coefficients for a set of nongenetic time-independent effects, β2 is an unknown vector of fixed regression coefficients for a set of nongenetic time-dependent effects, x1 is a vector of indicator variables for nongenetic time-independent effects, t x2(t) is a vector of indicator variables for nongenetic time-dependent effects, u is an unknown vector of regression coefficients for random effects due to sires, z is a vector of indicator variables for sire effects, q is an unknown vector of regression coefficients for time-dependent random effects due to the nursed litter, and x2(t) is a vector of indicator variables for time-dependent nursed litter effects.
Three different models were considered. For the first model (model COX), the distribution function for the baseline h0(t) was left completely unspecified (Cox, 1972) and a semiparametric proportional hazard model was used. The second model (model WTD) was a parametric model where a Weibull distribution was assumed as a baseline distribution function after the inclusion of a time-dependent covariate [i.e., a Weibull time-dependent function was used as baseline function for this model (Tarrés et al., 2005; Casellas et al., 2006)]
An additional model (model GDM) was considered because it is well adapted for analyses of timing of events occurring in short periods of time and with high incidence of ties in timing of occurrence, which is a common situation in analysis of piglet preweaning survival (Casellas et al., 2004). This model, which is based on Prentice and Gloeckler (1978) and is a grouped data version of the proportional hazard model, does not make any assumption about the baseline distribution function, but it can be viewed as a fully parametric model that includes a time-dependent covariate that changes its value at each day of the observed time space (Ducrocq, 1999).
Model Selection for Fixed Effects
Before survival analysis, TMI of the sire of the piglet and the size of the nursed litter, which were considered in all models as potential effects influencing the hazard of death of piglets, were categorized (Table 1). Because the form of the relationship between these variables and the hazard was unknown, this ensured that no assumption had to be made about the form of that relationship. Before categorization, TMI of sires were standardized to a mean of zero and standard deviation of one.
|
TMI 1 SD; class 3: TMI >1 SD; the number of sires for piglets in class 1, 2, and 3 was 25, 132, and 32, respectively). The parity of the nurse sow (parity 1, 2, 3, 4, 5, 6, and 7 or more) and the class for the size of the nursed litter (class 1: 5 piglets, class 2: from 6 to 8, class 3: from 9 to 11, class 4: from 12 to 14, class 5: 
15 piglets) were effects included in β2 as time-dependent covariates that could change value in the time space as a consequence of cross-fostering, carried out to homogenize size of litters, and of piglet mortality. Besides these effects, model WTD also included a time-dependent covariate changing value at d 6 and 12 for all piglets. Cut points at 6 and 12 d of age of piglets were identified with a spline regression of the log of the Kaplan-Meier survival function on time following the same approach used by Tarrés et al. (2005). To do that, all possible combinations of 1, 2, ..., k cut points in the time space (preweaning period) were explored, avoiding combinations of adjacent days A preliminary analysis was carried out to identify fixed effects that were statistically significant at P < 0.05. Before ultimate rejection of effects which were not significant, each of them was tested with the group of fixed effects initially significant to determine whether any became significant. To test the proportional hazards assumption, time-dependent factors [i.e., interaction terms between the time-independent effects and function of time (changing at 6 and 12 d)], were defined. The inclusion of these interaction terms did not significantly increase (P > 0.05) the likelihood for any of the models analyzed (results not shown in tables), and thus, the proportionality hypothesis was not rejected.
Random Effects and Heritability
Random effects included in all models were the effects of the sire of the piglet and of the nursed litter, which was treated as a time-dependent effect that, after the first day of life, could change value as a consequence of cross-fostering. Sire additive genetic effects, under polygenic inheritance, were assumed to follow a multivariate normal distribution: u ~ MVN (0, A
u2), where A is the additive genetic relationship matrix among sires and u2 is a variance component for sire effects. The effects of the nursed litter in q were assumed to be log-gamma distributed following a single parameter 
, from which the variance of the nursed litter effect q2 can be derived. Multivariate normal prior for the sire effect and log-gamma prior for the nursed litter effect were combined with the likelihood function of the data to obtain an expression proportional to the joint posterior density of all parameters (Ducrocq and Casella, 1996), and estimates of variance components were obtained by Laplacian approximation of the marginal posterior densities
Effective and equivalent heritabilities (Yazdi et al., 2002) of piglet preweaning survival were obtained as
and
where h2 and heq2 are effective and equivalent heritability, respectively, u2 and q2 are variance components for sire and nursed litter effects [calculated as q2 = tri-gamma(), where trigamma(.) is the trigamma function], respectively, and p is the average proportion of uncensored records. All analyses were carried out using the "Survival Kit" software, version 3.12 (Ducrocq and Sölkner, 1994).
|
RESULTS AND DISCUSSION |
|---|
|
TopAbstractINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONLITERATURE CITED |
|---|
The survival and hazard functions estimated by the Kaplan-Meier method are presented in Figures 1 and 2, respectively. Eighty-six percent of records were censored (animals still alive) at the end of the weaning period, which occurred at an average age of 28 d. Hence, piglet mortality from birth to weaning was 14%, and average failure time for uncensored records (death of piglets) was 6 d. The survival experience was not constant over the preweaning period. The estimated hazard function (Figure 2) indicates that the hazard of death for piglets progressively decreased from birth to d 14 and was much greater in the first week after birth than afterwards. As a consequence, nearly 70% of overall mortality (Figure 1) occurred during the first week of life. From d 14 onwards, the hazard remained unchanged. These results imply that each litter lost, on average, one piglet from birth to weaning, giving rise to important economic losses and ethical considerations. Svendsen and Bengtsson (1982) reported that a fraction ranging from 10 to 35% of newborn piglets may die within the first 3 wk of life. Moreover, over 50% of deaths occur in the first 3 d after birth (Dyck and Swierstra, 1987) with crushing accounting for 70 to 80% of deaths (English and Morrison, 1984). Most causes of death are due to interactions between the piglet and its environment (Le Dividich and Herpin, 1994). Also, low immune-competence at birth may play a role by increasing susceptibility to pathogens and leading to death in lactation (Xu et al., 2000).
|
|
Results of likelihood ratio tests (LRT), obtained with COX, for statistical significance of fixed effects are summarized in Table 2. All analyzed factors were significantly related to the risk of mortality when they were entered sequentially in the model or were excluded from the full model one at a time giving similar results and showing very little redundancy when explaining variation of the investigated trait. Parity of the nurse sow (P < 0.01) and TMI (P < 0.05) had less impact on preweaning survival than the one of the other effects. The year-month of birth had a marked influence on the hazard function (P < 0.001), as did cross-fostering (P < 0.001), sex (P < 0.001), and size of the nursed litter (P < 0.001). These results are in agreement with what was expected intuitively from percentages of uncensored records for levels of fixed effects reported in Table 1
.
|
. Estimated HR were similar for different models with the exception of HR estimates for the size of the nursed litter. For litters ranging from 12 to 14 piglets or with more than 14 piglets, the estimated HR were greater when obtained with WTD than when estimated by COX and GDM (1.78 vs. 1.40 for litters from 12 to 14 piglets; 2.98 vs. 1.60 for litters with more than 14 piglets)
|
; Becker, 1995; Knol et al., 2002). Becker (1995) reported that the increased mortality in males was due to more males being crushed and to chilling. Although the underlying mechanism responsible for this sexual dimorphism in preweaning mortality rates has not been elucidated, there are differences across sexes that may provide biological explanations to these observations. A greater basal concentration of cortisol observed in male piglets in comparison with female piglets (Ruis et al., 1997) may cause male piglets to be more susceptible to detrimental stress effects and to succumb to subsequent disease challenges
Cross-fostering exerted favorable effects on survival chances of piglets. Fostered piglets had 40% greater probability of survival than piglets raised by the biological mother. This is in agreement with results obtained by Knol et al. (2002) and Leenhouwers et al. (2001). Cross-fostering of piglets was performed to reduce variation in size of nursed litters and occurred for 46% of piglets. Cross-fostering is one of the most effective methods of increasing postnatal survival. This practice enhances the survival probability of small piglets, because they have to compete less to reach the last available teat (English et al., 1982) in litters of average size than in large or very large litters. Moreover, cross-fostered piglets might introduce a disease in their new litters and might be at less risk than their new littermates because of immunity due to colostrum suckled from the biological mother. This phenomenon decreases the average survival of piglets that are not cross-fostered (Knol, 2001). Because 70% of deaths occur during the first week of life, cross-fostering should be performed as soon as possible as reported in previous studies (Svendsen et al., 1986; Straw et al., 1998). A further biological explanation of these effects is related to the increase of piglet BW uniformity within a nursed litter caused by cross-fostering and to a possible association between within-litter variation of piglet weights and preweaning losses (English et al., 1982; Roehe and Kalm, 2000; Marchant et al., 2001; Milligan et al., 2002a,b). However, Milligan et al. (2001) reported that their data provided little support for the hypothesis that high birth weight variation resulted in decreased survival. Leenhouwers et al. (1999) did not find any relationship at the phenotypic level between the within-litter standard deviation of birth weight and the proportion of stillborn piglets
In this study, the group of sows originating cross-bred piglets was submitted to minimum replacement policies and culling occurred mostly for reduced fertility due to aging or occurrence of severe disease. As a consequence, the chance of repeated farrowing was not influenced by selection of sows based on reproductive performance. The survivability increased when piglets were nursed by sows of third, fourth, and fifth parity in comparison with first- and second-parity sows. However, the hazard of dying for piglets nursed by sows of sixth vs. seventh or greater parity did not differ (P > 0.05) from that of animals nursed by first- and second-parity sows. These results are consistent with those of several studies (Leenhouwers et al., 2001; Knol et al., 2002; Damgaard et al., 2003; Arango et al., 2005; Grandison et al., 2005). Conversely, Weary et al. (1998) reported a greater probability of crushing for greater parities. The authors justified this result because litters originated by sows of greater parity tended to exhibit less ADG from d 1 to 3 after birth, and because older sows were heavier and clumsier. However, their results did not allow clear conclusions about the causes of crushing because several potential predisposing factors (low early BW gains, high sow parity number, larger litter size, and low birth weight) were closely related. Another important aspect related to piglet preweaning mortality is colostrum production of sows. After birth, the piglet is fed initially with colostrum and after with milk, which exhibit high fat and low carbohydrate contents, implying that the intestine must be functional at birth and the piglet rapidly able to synthesize glucose by gluconeo-genesis to supply its glucose-dependent tissues and to oxidize fats. In this context, ingestion of colostrum, which provides both energy and maternal antibodies protecting the piglets until their immune system matures, is of utmost importance for survival. There are several sow-related factors, including health, premature farrowing, changes in reproductive hormones and metabolism, parity, nutrition, and genetics, that might be involved in colostrum production (Le Dividich et al., 2005). There is no clear evidence of parity effects on colostrum production by the sow (Le Dividich et al., 2005). Inoue et al. (1980) and Klobasa et al. (1986) reported that first parity sows have lesser colostrum IgG concentrations than multiparous sows
Consistent with several studies (Kerr and Cameron, 1995; Knol et al., 2002; Grandison et al., 2005), the probability of survival decreased for piglets joining small (less than 6 piglets), large (from 12 to 14 piglets), or very large (more than 14 piglets) litters in comparison with litters of intermediate size (from 6 to 11 piglets). Because of occurrence of cross-fostering, a few litters were classified as litters of small size and were likely to be those of sows with physiological inabilities or difficulties to have a normal gestation. For large or very large litters, increased risk of preweaning mortality might have been caused by excessive crowding and reduced milk availability for piglets of limited ability to compete for suckling. Large litters and wide ranges of birth weights are claimed to cause decreases in piglet survival because of competitive exclusion of light lit-termates from access to productive teats (English and Morrison, 1984)
The year-month of birth had a marked influence on piglet survival. Piglet preweaning mortality changed across years and across months of the same year as a consequence of changes in the hazard due to several sources of variation such as climate, epidemiologic and management effects. The magnitude of the estimated HR for year-month of birth effects (data not presented) changed erratically across year-month classes and did not exhibit a consistent trend over time. Roehe and Kalm (2000) analyzed preweaning mortality in piglets and found that year-season was the most important fixed effect for preweaning mortality
For model WTD, the instantaneous mortality rate was more pronounced during the first week of life, diminished from d 6 to 12 (HR = 0.21), and was very low from d 12 up to weaning (HR = 0.06). Heterogeneous mortality rates in different periods caused the Weibull distribution to fail in the validation of the baseline distribution, and a time-dependent distribution was needed. In our analysis, the parametric survival function has been replaced with piecewise survival functions whose slopes change at given points, as suggested by Yazdi et al. (2002).
Relationship Between Sire Total Merit Index and Piglet Survival
The TMI of boars exhibited significant relationships with piglet preweaning survival (Table 2). Piglets originated by top TMI boars exhibited a 17% (20% for WTD) greater instantaneous risk of mortality than did piglets sired by intermediate or low TMI boars. The Kaplan-Meier estimate of survival functions stratified according to TMI class are presented in Figure 3. Some studies investigated the genetic relationship between piglet survival and performance traits such as back-fat and fat and protein deposition in pig lines selected for efficiency of pork production (Herpin et al., 1993; McKay, 1993; Knol, 2001). The genetic correlation between piglet survival and backfat has been reported to be moderate (Knol et al., 2002). Knol (2001) reported that selection to reduce backfat is expected to increase birth weight and decrease piglet survival. Unfortunately, the relationship of piglet survival with production traits has been investigated only in lines used for pork production. In Italy, the breeding goal for boar lines originating slaughter pigs for dry-cured ham production differs greatly from the one pursued to enhance efficiency of pork production. Class of TMI exhibited an unfavorable relationship with survival of piglets. Biological explanations for these results could be related to selection of the C21 boar line, which did not include any crossbred survival trait during the studied period. Another interpretation is that, in the past, one of the major goals for this line was to reduce excessive marbling of the raw ham. This trait is related to fat deposition and a reduced ability to metabolize triglycerides; it might play an important role in the thermoregulatory ability of the neonate and survival.
|
Parameters of the approximated marginal posterior distributions of sire and nursed litter variance components and estimates of effective and equivalent heritability for COX, GDM, and WTD models are reported in Table 4. Estimated variance components and heritabilities obtained using different survival models were similar with the only exception of the WTD estimate of the nursed litter variance, which was slightly greater than the corresponding estimates obtained with COX and GDM. The estimated nursed litter variance component was much greater than the sire variance estimate, confirming the importance of the common environment generated by the nurse sow as a key factor affecting the survival of piglets before weaning (Casellas et al., 2004; Wolf et al., 2007). This effect was included as a time-dependent effect to account for variation in litter membership due to cross-fostering. Piglets of a nursed litter share common environmental effects due to occurrence of infectious diseases such as diarrhea and incidentals such as diseased udders, and are affected by the maternal ability of the nurse sow. The nurse sow exerts effects on piglet survival that are strictly environmental for the piglets, but for the sow, are affected by both genetic and environmental components. In this study cross-fostering occurred for 46% of the piglets. This raised a question in simultaneously modeling maternal and permanent environmental effects. For a piglet that was not moved to a different litter, accounting for both the biological mother and the nurse sow effects was not feasible, because these effects were confounded. In the present study, the choice was to model the permanent environmental effect determined by the nursed litter
|
. When a parametric (WTD and GDM) or semiparametric (COX) model was used in the construction of the likelihood function, it was observed that the resulting marginal posterior densities of 
were very similar, with a slightly larger variance in the case of COX. In spite of a very high censoring rate, the standard deviations of the posterior densities were small, possibly due to the size of the data set and the good pedigree structure, which allowed precise estimation of sire variance (Ducrocq et al., 2000). The estimated sire variance obtained in this study using proportional hazard models cannot be compared directly with estimates obtained in other studies where alternative methodologies were used (e.g., linear or threshold models) because of scale differences. Van Arendonk et al. (1996) and Knol et al. (2002) used linear models ignoring the categorical nature of the trait. Few studies have addressed estimation of variance components for piglet mortality using threshold models (Roehe and Kalm, 2000; Grandison et al., 2002; Arango et al., 2005), and only one modeled the trait at the individual piglet level (Grandison et al., 2002)
|
. There is large variation across estimates obtained in different studies. Lamberson and Johnson (1984) reported an estimate of heritability for preweaning survival of 0.03, whereas Ferguson et al. (1985) reported values of 0.14 and 0.18 for Yorkshire and Duroc, respectively
In our analysis, using the formula of Yazdi et al. (2002), the estimated heritability, in the unrealistic situation of no censoring, was 0.14. After correction for the large censoring rate (86%), the equivalent heritability was very low (0.03) and comparable with estimates reported by other authors (Lamberson and Johnson, 1984; Knol et al., 2002; Damgaard et al., 2003; Casellas et al., 2004; Wolf et al., 2007)
Piglet preweaning survival additive genetic variance is large enough to allow economically viable selection as suggested by Knol et al. (2002). Furthermore, the implementation of a routine genetic evaluation based on survival models is feasible, even for large populations. Inclusion of the results of such an evaluation in breeding programs seems possible and is probably advisable, for economic as well as for ethical reasons. However, the relationships between survival and breeding goal traits should be considered to optimize selection strategies.
Sire Rankings Under Different Models
The relationships between sire rankings obtained using different survival models are depicted in Figure 5. The rank correlations between COX and GDM, COX and WTD, and GDM and WTD were 0.99, 0.98, and 0.97, respectively. Changes in sire rankings due to use of different models were limited and occurred preferentially at intermediate rank positions. Because COX is a semiparametric model, it is less sensitive to an incorrect model choice. Nevertheless, when analyzing short periods of time, the dates of failure are rather broadly grouped, and the time scale has to be considered as discrete. This is a common situation in the analysis of piglet preweaning survival, and the assumption of continuity in the baseline hazard distribution and absence of ties between ordered failure time (Cox, 1972) associated with the use of proportional hazard models may be violated. Because the Weibull model is a priori not sensitive to ties, and WTD is more flexible than a pure Weibull model (Yazdi et al., 2002; Tarrés et al., 2005), WTD seems to be a better option than COX for the analysis of piglet preweaning survival. Results of this study indicate that, as suggested by Casellas (2007), the fitting of parametric survival models can be easily improved with the simple addition of a time-dependent effect. Although the high flexibility of COX is advantageous, semi-parametric approaches imply greater demands in computational requirements and time needs (Ducrocq et al., 2000). Hence, WTD seems to be a more advantageous model for genetic evaluation of piglet preweaning survival because the vector of first derivatives of the log-likelihood function is much easier to compute and the Hessian matrix is usually very sparse (Ducrocq et al., 2000).
|
|
Footnotes |
|---|
2 Corresponding author: alessio.cecchinato{at}unipd.it
Received for publication December 21, 2007. Accepted for publication May 5, 2008.
|
LITERATURE CITED |
|---|
|
TopAbstractINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONLITERATURE CITED |
|---|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
