J. Anim. Breed. Genet. ISSN 0931-2668
ORIGINAL ARTICLE
Phenotypic and genetic variation in longevity of Polish Landrace sows ska & T. Blicharski M. Sobczyn Institute of Genetics and Animal Breeding PAS, Magdalenka, Poland
Keywords Genetic parameters; longevity; removal risk; sow; survival analysis. Correspondence ska, Institute of Genetics and M. Sobczyn Animal Breeding PAS, Jastrzez biec, ul. Postez pu 36A, 05-552 Magdalenka, Poland. Tel: +48 227367128; Fax: +48 227561417; E-mail: [email protected] Received: 24 July 2014; accepted: 9 December 2014
Summary The influence of some production traits on the longevity of Polish Landrace sows was evaluated using survival analysis. Estimates of genetic parameters were obtained from the sire and animal components in linear and survival methodologies. Comparison between survival and linear models was based on heritabilities and ranking of estimated breeding values of sires. The same data set, 13 031 sows, was used for both methodologies, even in the presence of censored observations. The effects of herd*year and year*season of the first farrowing had the largest influence on the risk of culling of sows. Sows born in spring season (March–May) had a 24% (p < 0.001) lower hazard for removal than those born in winter (December–February). The age at first farrowing had a small but significant effect on culling: the hazard regression coefficient for this trait was 0.002 per day. Sows that had more piglets born alive and fewer stillborn in the first litter had a decreased risk of being culled. Within a contemporary group, slower growing gilts had decreased removal risk. The relative risk ratios show a marginal decreased rate of culling for sows with backfat thickness between 9.5 and 11 mm compared to the leaner sows. Loin depth had no effect on sow longevity. Heritability estimates ranged from 0.09 to 0.38 depending on the model and type of analysis. In survival analysis, all heritabilities for longevity were higher when analysed with sire models (0.21 and 0.38) compared to animal models (0.09 and 0.16). The use of animal or sire models in the linear analysis gave similar heritability estimates (0.12 and 0.10). Correlations between breeding values for sires were moderate and high, with absolute values from 0.51 to 0.99, depending on the model fitted and methodology. A stronger correlations within methodologies (0.83–0.99) than within models with different methodologies (0.51–0.63) were obtained.
Introduction During the last two decades, the increasing demand for leaner pork has resulted in breeding goals that focus more on reduction in backfat and improvement in growth rate to provide the required carcass quality. During the same time period the methods in which sows are managed and housed have moved towards a more intensive system, individual sow housing system
doi:10.1111/jbg.12135
has been changed to one where group housing is utilized, and herds producing piglets become larger. Introduced in 1998 in Poland, estimation of breeding values using BLUP technology and intensive selection have caused a remarkable genetic progress in reducing backfat for carcass quality and improving growth rate for production efficiency (Blicharski & Eckert 2011). The genetic evaluation of reproductive traits (number of piglets born alive and weaned) has been
© 2015 Blackwell Verlag GmbH
• J. Anim. Breed. Genet. 132 (2015) 318–327
ska & T. Blicharski M. Sobczyn
applied in Polish maternal pig lines for selection decisions since 2006. Less emphasis has been given for traits related to fertility like farrowing interval or interval between weaning and first service. Intensive selection on production and reproduction traits without considering functional traits can lead to decreased longevity (Engblom et al. 2008). Today, the majority of sows in the Polish commercial herds are removed in their early parities: sows are generally replaced at the third or fourth parity, and most cullings are unplanned, mainly due to reproductive disorders or locomotor problems. An increased herd life is associated with reduced replacement costs, higher productivity and increased possibility for selection on other traits (Hoge & Bates 2011). Despite the economic significance of longevity, there is currently no genetic evaluation for longevity of sows in Poland. Sow longevity is a complex trait, and it can be defined in different ways, most commonly as the ability of a sow to reach nth parity (stayability), total days in the breeding herd (length of productive life) and total piglets born in the sow’s lifetime (lifetime production), or a combination of the last two (pigs produced per day of herd life). There is a relatively high genetic correlation among the different sow longevity traits (Engblom et al. 2009; Sev on-Aimonen & Uimari 2012; Sobczy nska et al. 2013) suggesting that same genes impact the traits. Length of productive life combines several traits related to fertility, health and production like number of piglets born alive and weaned, milk production and leg soundness. Consensus has not been reached in the scientific literature regarding the relationships between longevity and growth and body composition traits (L opez-Serrano et al. 2000; Yazdi et al. 2000; Serenius & Stalder 2004; Knauer et al. 2010; Nikkil€ a et al. 2013). Genetic correlations vary across the trait definitions and among the populations studied. The genetic correlations between longevity and average daily gain and backfat obtained by Sobczy nska et al. (2013) in Polish Landrace breed were unfavourable but generally very low with high standard errors. However, the genetic correlations of longevity traits with lean meat percentage and phenotypic selection index reported in that study indicate that an antagonism exists between growth performance and sow longevity. Analysis of longevity traits using linear models may not be appropriate because of the skewed distribution and the bias caused by excluding the records of living sows or of assuming their records to be complete. Moreover, there are some factors that change throughout lifetime and affect an animal’s longevity. The management practices, natural environment and © 2015 Blackwell Verlag GmbH
• J. Anim. Breed. Genet. 132 (2015) 318–327
Study of longevity in Landrace sows
production level like number of piglets born in consecutive parities vary throughout a sow’s life. Survival analysis allows use of non-linear models and also tends to give higher estimates of heritability than does the linear model suggesting increased reliability of sires EBV and increased selection accuracy (Sewalem et al. 2005). The implementation of survival analysis in genetic evaluation (Ducrocq & S€ olkner, 1998) allows for processing censored as well as uncensored data, time-dependent covariates and random covariates. Therefore, the objective of this study was to estimate, using survival analysis methodology, the genetic variation for longevity, investigate relationships between longevity and production traits, compare estimates of genetic parameters from the sire and animal models and compare the sire rankings of longevity obtained from linear and survival analysis using the same data set.
Materials and methods Data
Data included farrowing and growth records collected from the Polish Landrace (L) sows with at least one farrowing over the period from August 1994 until March 2011. In order to obtain reasonable contemporary groups, data were limited to herds of over 300 sows in the entire study period. The average number of sows recorded per herd was 543 (range: 303–1286, SD = 271). Sows on all farms were housed during gestation and farrowed indoors. Sows were usually mated naturally, and at farrowing, they were kept in individual crates. In general, the piglets were weaned at approximately 4 weeks of age. The longevity and reproductive traits were first farrowing age (AFF), number of piglets born alive in the litter (NBA), number of dead piglets in the first litter (ND1), length of productive life (LPL) defined as number of days between first farrowing and removal or the end of data collection. Length of production life was considered as functional longevity, reflecting the ability of sows to delay involuntary culling and the true longevity defined as the ability to delay any culling. The performance traits were daily gain (ADG), backfat thickness (BF) and Longissimus muscle depth (LMdepth) registered as a part of the on-farm testing programme by the breeding consultant. The growth rate was defined as the average weight gain per day from birth until the test day. The growth rate was adjusted to 180 day of age according to National Research Institute of Animal Production (PIB) guidelines prior to the analyses. 319
ska & T. Blicharski M. Sobczyn
Study of longevity in Landrace sows
Backfat thickness was defined as the mean of the depth of fat measured by ultrasound equipment (PIGLOG 105; SFKâ, Soborg, Denmark) at two points P2 and P4 (3 and 8 cm, respectively, away from the back mid-line behind the last rib). Longissimus muscle depth was measured behind the last rib approximately 8 cm off the mid-line by the same ultrasound equipment as BF. Backfat thickness and LMdepth were adjusted to a constant weight of 110 kg using the formula determined by PIB. The animals with extreme values for the age at first farrowing (≤250 days and ≥550 days) or with missing parity were excluded to remove questionable records. Only sows with for which identity of their sire and dam was known were included. The final data contained 13 031 sows from 24 herds. Records were treated as censored if sows were still alive at the end of the studied period: on 31 March 2011. Descriptive statistic for some production traits and reproductive performance, as well as survival analysis statistics, was presented in Table 1. Statistical analyses
The length of productive life was evaluated with survival and linear analyses, using a sire and animal
Table 1 Descriptive and survival statistics for Landrace sows included in the analysis Descriptive statistics Traits LPL NBA ND1 AFF ADG BF LMdepth Survival statistics N Records Elementary records Wright censored records Days Min censoring time Max censoring time Average censoring time Min failure time Max failure time Average failure time
Mean SD 596.8 417.3 11.5 1.8 0.44 0.89 343.6 37.9 627.7 73.1 10.8 1.9 51.3 5.5
Range 7–2784 0–22 0–14 252–547 407–940 5.5–25 34–81
13 031 67 237 2822 (21.6%) 28 2087 617.2 7 2784 591.2
LPL, Length of productive life; NBA, number of live-born piglets in the litter; ND1, number of dead piglets in the first litter; AFF, age at the first farrowing; ADG, adjusted daily gain; BF, adjusted backfat thickness; LMdepth, adjusted Longissimus muscle depth.
320
model. The same data set was used for both analyses, even in the presence of censored observations. Survival analyses were performed using the Survival Kit V6.0 (Ducrocq et al. 2010). Removal hazard during LPL was analysed with the following Weibull model: hðtÞ ¼ h0 ðtÞ exp ½xðtÞ0 b
ð1Þ
where h(t) is the hazard function of a sow’s LPL at time t, t is time measured in days from first farrowing until culling, death or censoring, h0 (t) is the baseline hazard function kq(kt)q1, assumed to follow a Weibull distribution with location (k) and shape (q) parameters, b is a vector of fixed and random effects and x(t) is the corresponding incidence matrix. Vector b contained covariates as follows: fixed time-dependent farm–year effects, changed at 1 January each year, fixed time-independent birth season effect with four classes: December–February (i), March–May (ii), June– August (iii) and September–November (iv), fixed time-independent year–season (listed above) of first farrowing with 64 classes, fixed time-independent regressions of age at first farrowing and number of dead piglets in the first litter, fixed time-dependent total number of piglets born in the litter, changed at each farrowing date, fixed time-independent effects of ADG, BF and LM depth classified into five categories taking cut-off points at percentiles 20, 40, 60 and 80 (Table 2) and random genetic component defining an animal or sire effect, depending on the model. The solution for fixed effects was expressed as hazard ratio (HR). Hazard ratio was defined as the ratio between the estimated hazard for being removed under the influence of certain factors and the estimated hazard for a reference class. The first class (the lowest traits values) for ADG, BF and LMdepth was considered as reference class. The reference class for season of birth was December–February. In the case of continuous variables, the estimates of b can be expressed in relative risk by exp(b). Two separate longevity analyses were carried out: the sire or animal variance for longevity was Table 2 Cut-off points for the factor variables included in the analysis Percentiles
ADG (g/day) BF (mm) LMdepth (mm)
20
40
60
80
566 9.4 46
603 10.3 50
638 11 53
687 12.2 56
© 2015 Blackwell Verlag GmbH
• J. Anim. Breed. Genet. 132 (2015) 318–327
ska & T. Blicharski M. Sobczyn
calculated using a complete and a reduced model. The complete model included all variables listed in (1) and defined functional longevity. The reduced model contained herd–year, sire (or animal), season of birth, year–season of the first farrowing and age of the first farrowing terms and was referred to true longevity. To test the assumption for Weibull distribution, the log of minus the log of the Kaplan–Meier estimate of the survivor curve was plotted against log of time. The plot of this relationship produced approximately straight line (except for a period after the first farrowing when there was intensive culling), which indicates that the Weibull distribution fits data well (Figure 1). The proportion of explained variation by the model was calculated as R2 of Maddala R2M ¼ 1 ðLR =LU Þ2=n where LR and LU represent restricted and unrestricted maximum likelihoods and n denotes the total sample size. For proportional hazard model assuming a Weibull survival function, the heritability of longevity on the logarithmic scale was calculated as h2log ¼ 4r2s = ðp2 =6 þ r2s Þ for sire model and as h2log ¼ r2a = ðp2 =6 þ r2a Þ for animal model, where p2/6 is the variance of the standard extreme value distribution, r2s and r2a are sire and animal variance, respectively. The heritability on the original scale using the sire variance was calculated as h2ori ¼ 4r2s =½ðexp f1=qvgÞ2 ðp2 =6 þ r2s Þ for sire model and as h2ori ¼ r2a =½ðexp f1=qvgÞ2 ðp2 =6 þ r2a Þ for animal model (Yazdi et al. 2000), where q denotes shape parameter and v is –Euler’s constant. In the linear analysis, the following mixed model was fitted for LPL:
Study of longevity in Landrace sows
y ¼ Xb þ Zu þ e
ð2Þ
where y is a vector of observations of LPL, b is the vector of fixed effects, X is an incidence matrix for the fixed effects, the vector u contains random animal or sire effects, and Z is the associated incidence matrix, e is the vector of residual effects. For the linear model (2), the same fixed effects as for the complete Weibull model (1) were used. In the mixed linear analysis, the variance components for LPL, ignoring censoring, were estimated by the REML method with animal and sire models using the VCE 6.02 (Groeneveld et al. 2010). In addition, EBV for LPL for the sires was estimated under PEST package (Groeneveld et al. 2001) with the same statistical models. Models accounted for pedigree information up to the third generation of ancestors. The pedigree file for animal model included 16 814 animals. The number of sires and dams of the total population were 1368 and 15 451, respectively. The pedigree used in the sire model consists of 1225 animals and included pedigree information only for the sows’ sire (N = 1026). The number of daughters for each sire ranged from 1 to 230 (average 13). Correlations between the estimated breeding values from survival analysis and linear model were calculated to assess the agreement between the genetic predictions from the two methods. Estimated breeding values for sires with at least 10, 20, 30 and 40 uncensored daughter records obtained by survival and linear analyses were compared using Spearman’s rank correlation. The number of sires that met these criteria were 399, 197, 107 and 63, respectively. For survival analysis of LPL, a low value of EBV is desirable indicating a low removal hazard, whereas high values are desirable in the linear analysis indicating long herd life. Results and discussion
Figure 1 Log-cumulative hazard plot, S(t) – Kaplan–Meier survivor function, t – days from first farrowing.
© 2015 Blackwell Verlag GmbH
• J. Anim. Breed. Genet. 132 (2015) 318–327
The average LPL in Landrace sows was 597 days with large standard deviation (417). Exactly, the same LPL in Landrace sows obtained M esz aros et al. (2010) in Austria. Similar LPL was reported by Yazdi et al. (2000) for Landrace sows in Sweden, whereas the average LPL for Yorkshire sows in USA and cross-bred sows in Sweden was 485 days (Engblom et al. 2009; Hoge & Bates 2011). On average, uncensored females in this study had a herd life slightly shorter than censored ones (591 versus 617 days). Results obtained in different studies are difficult to compare due to different amount of censored records and length of the study period. 321
ska & T. Blicharski M. Sobczyn
Study of longevity in Landrace sows
Table 4 Hazard ratios (HR; Chi2 p-values in parentheses) for factors included in the animal model
Risk analysis for sow removal
The shape parameters of the Weibull distribution were very similar in the animal and sire analyses amounting 1.96 and 1.92, respectively. The significance of different covariates included in the full Weibull animal model is presented in Table 3. The hazard ratios for production traits as well as for the birth season and AFF are given in Table 4. Results of likelihood ratio test for significance of effects and risk ratios obtained from the sire model were very similar (not shown). Among the factors evaluated in the Weibull model, the herd*year effect had the largest influence on the risk of culling of sows. The proportion of explained variation by the model (R2 of Maddala) increased drastically (to 28%) when herd*year interaction was added. The second covariate that most influenced sow longevity was year*season of the first farrowing. These two factors explained nearly 97% of all the variation. Herd*year effect contains different management practices and housing systems among farms where improvements can probably reduce the unplanned removal of sows. High year*season effect of the first farrowing interaction may reflect fluctuation of pig prices at the market. These prices have shown a seasonal pattern and influence production and culling decisions. Earlier studies have found significant associations between the herd–year interaction and sow longevity (Yazdi et al. 2000; Serenius & Stalder 2004, 2007). Season of birth had a significant effect on the sows’ risk of being culled. Sows born in spring season (March–May) had a 24% (p < 0.001) lower hazard for removal than those born in the reference class (Table 4). Compared with sows born in winter season (December–February), sows born in autumn (SepTable 3 Likelihood ratio test when all covariates added to Weibull animal model sequentially Covariatea
d.f.
v2
p Value
R2M
Herd*year Year*season of first farrowing Season of birth AFF ADG BF LMdepth ND1 NBA Animal
296 63 3 1 4 4 4 1 1
4397.3 4003.9 98.9 43.3 7.8 10.1 3.2 2.7 227.2 RANDOM
0.000 0.000 0.000 0.000 0.100 0.038 0.525 0.100 0.000
0.286 0.475 0.479 0.480 0.481 0.482 0.482 0.483 0.491
a
See Table 1 for abbreviates.
322
Covariatea
Class
No.b
HR
ADG
1c 2 3 4 5 1c 2 3 4 5 1c 2 3 4 5 1c 2 3 4
2311 2117 1961 1937 1883 2016 2023 2094 1909 2167 2609 2115 2086 1675 1724 2491 2489 2664 2569 10 209 10 209 10 209
1 1.08 (0.03) 1.05 (0.15) 1.03 (0.48) 1.07 (0.13) 1 0.95 (0.10) 0.97 (0.33) 1.03 (0.49) 1.03 (0.43) 1 1.04 (0.22) 0.99 (0.81) 1.03 (0.47) 0.98 (0.65) 1 0.76 (0.00) 0.99 (0.98) 1.08 (0.03) 0.91 (0.00) 1.04 (0.00) 1.002 (0.00)
BF
LMdepth
Season of birth
NBA ND1 AFF a
See Table 1 for abbreviates. Number of uncensored sows in each class. c Used as the reference class. b
tember–November) had 8% (p = 0.034) greater hazard for removal. It is well established (Black et al. 1999) that climatic conditions influence feed intake and metabolism in the pigs, and thus production and reproduction traits. Seasonal effects on fertility in gilts and sows have been showed by Tummaruk et al. (2000). The additional changes in R2 of Maddala were small when adding AFF, production and reproduction traits. Similar to many studies on Landrace and other pig breeds (e.g. Serenius et al. 2006, 2008; Serenius & Stalder 2007; Engblom et al. 2008; Hoge & Bates 2011), our analysis also revealed that age at first farrowing had a small but significant effect on culling. The hazard regression coefficient for AFF was 0.002 per day. Different results were obtained by Serenius & Stalder (2004) for two pig breeds: in Large White, negative genetic association between AFF and LPL (0.28) was found, whereas the sign of corresponding correlation was positive in Landrace (0.17). An increased AFF could be attributed to gilts that did not conceive on the first service. Delayed AFF could be considered an early indicator of reproductive fitness or lack thereof (Hoge & Bates 2011). On the other hand, the important factor in determining when to © 2015 Blackwell Verlag GmbH
• J. Anim. Breed. Genet. 132 (2015) 318–327
ska & T. Blicharski M. Sobczyn
mate or inseminate a gilt is probably prepubertal growth rate, which depends on environmental (herd management system) and genetic effects. The effect of NBA was highly significant (p < 0.0001). If a sow had one more piglet in her litter, her relative culling risk was reduced by 9%. Yazdi et al. (2000) also reported a decreased risk ratio with increasing litter size for both first and last litter. Similarly, Serenius & Stalder (2007) and Hoge & Bates (2011) reported that a small number of piglets born alive at first farrowing had a detrimental effect on LPL. The Landrace sows that had one more pig born alive in their first litter had a longer length of productive life of 4.6 days in the study by Sobczy nska et al. (2013). However, Serenius et al. (2006) concluded, comparing six genetic lines, that there was no clear association between litter size and LPL when culling due to poor reproductive performance was not practiced. Significant effect of NBA is not surprising as number of weaned piglets per parity is a main factor that determines profitability in piglet producing herds and sow productivity is a primary removal reason in voluntary culling. As ND1 increased by one piglet, the culling rate increased by 4%. This indicator of neonatal piglet mortality had a significant negative association with longevity. Hoge & Bates (2011) also concluded that Yorkshire sows that had fewer stillborn piglets in the first litter had a decreased risk of removal. The average daily gain had very low effect on the risk of culling as only sows with gains 567–603 g/day tended to have 8% more risk of being culled (p = 0.03) compared with sows from the reference group (ADG < 567 g/day). The effect of ADG was similar in other groups suggesting a negative phenotypic association between daily gain and LPL; however, it was not significant. In the studies of Serenius et al. (2006) and Knauer et al. (2010), the relationship between average daily gain and longevity was significant and negative in some genetic lines only. Hoge & Bates (2011) and Serenius & Stalder (2007) found that slower growing gilts tended to have a decreased risk of being culled. Similarly, L opez-Serrano et al. (2000) and Sobczy nska et al. (2013) reported unfavourable genetic correlations between growth rate and longevity in Landrace and Large White breeds, but in Landrace sows, this relationship was weak and not significant. Moderately unfavourable genetic correlations were found for growth rate and longevity by Nikkil€a et al. (2013) when REML and Gibbs sampling were used in analysis. Results obtained by Tarr es et al. (2006) suggest a negative phenotypic association between ADG after the growth test and LPL. Thus, it © 2015 Blackwell Verlag GmbH
• J. Anim. Breed. Genet. 132 (2015) 318–327
Study of longevity in Landrace sows
appears that ADG may affect sow longevity in some breeds or genetic lines while not in others; however, most studies imply that fast growing gilts have inferior longevity. Backfat thickness on the day of performance test had no effect on the risk of culling; however, the relative risk ratios show a marginal decreased rate of culling for sows with backfat thickness between 9.5 and 11 mm. The phenotypic or genetic association between backfat thickness and sow longevity is not clear. Two studies found unfavourable correlation, that is gilts with thicker backfat stayed and produced longer in the herd (L opez-Serrano et al. 2000; Hoge & Bates 2011), whereas some studies found no association (Yazdi et al. 2000; Serenius & Stalder 2004, 2007; Sobczy nska et al. 2013) and another study found that an intermediate optimum backfat (16–19 mm) at first farrowing was favourable for sow longevity (Tarr es et al. 2006). Results obtained by Serenius et al. (2006) indicate that gilt backfat level affects whether gilts successfully conceive and farrow their first litter, whereas the association between backfat and sow longevity is weaker for sows that complete their first parity. Loin depth had no effect on sow longevity in this study. This result is not consistent with the higher risk of culling in Duroc sows with loin depths
ORIGINAL ARTICLE
Phenotypic and genetic variation in longevity of Polish Landrace sows ska & T. Blicharski M. Sobczyn Institute of Genetics and Animal Breeding PAS, Magdalenka, Poland
Keywords Genetic parameters; longevity; removal risk; sow; survival analysis. Correspondence ska, Institute of Genetics and M. Sobczyn Animal Breeding PAS, Jastrzez biec, ul. Postez pu 36A, 05-552 Magdalenka, Poland. Tel: +48 227367128; Fax: +48 227561417; E-mail: [email protected] Received: 24 July 2014; accepted: 9 December 2014
Summary The influence of some production traits on the longevity of Polish Landrace sows was evaluated using survival analysis. Estimates of genetic parameters were obtained from the sire and animal components in linear and survival methodologies. Comparison between survival and linear models was based on heritabilities and ranking of estimated breeding values of sires. The same data set, 13 031 sows, was used for both methodologies, even in the presence of censored observations. The effects of herd*year and year*season of the first farrowing had the largest influence on the risk of culling of sows. Sows born in spring season (March–May) had a 24% (p < 0.001) lower hazard for removal than those born in winter (December–February). The age at first farrowing had a small but significant effect on culling: the hazard regression coefficient for this trait was 0.002 per day. Sows that had more piglets born alive and fewer stillborn in the first litter had a decreased risk of being culled. Within a contemporary group, slower growing gilts had decreased removal risk. The relative risk ratios show a marginal decreased rate of culling for sows with backfat thickness between 9.5 and 11 mm compared to the leaner sows. Loin depth had no effect on sow longevity. Heritability estimates ranged from 0.09 to 0.38 depending on the model and type of analysis. In survival analysis, all heritabilities for longevity were higher when analysed with sire models (0.21 and 0.38) compared to animal models (0.09 and 0.16). The use of animal or sire models in the linear analysis gave similar heritability estimates (0.12 and 0.10). Correlations between breeding values for sires were moderate and high, with absolute values from 0.51 to 0.99, depending on the model fitted and methodology. A stronger correlations within methodologies (0.83–0.99) than within models with different methodologies (0.51–0.63) were obtained.
Introduction During the last two decades, the increasing demand for leaner pork has resulted in breeding goals that focus more on reduction in backfat and improvement in growth rate to provide the required carcass quality. During the same time period the methods in which sows are managed and housed have moved towards a more intensive system, individual sow housing system
doi:10.1111/jbg.12135
has been changed to one where group housing is utilized, and herds producing piglets become larger. Introduced in 1998 in Poland, estimation of breeding values using BLUP technology and intensive selection have caused a remarkable genetic progress in reducing backfat for carcass quality and improving growth rate for production efficiency (Blicharski & Eckert 2011). The genetic evaluation of reproductive traits (number of piglets born alive and weaned) has been
© 2015 Blackwell Verlag GmbH
• J. Anim. Breed. Genet. 132 (2015) 318–327
ska & T. Blicharski M. Sobczyn
applied in Polish maternal pig lines for selection decisions since 2006. Less emphasis has been given for traits related to fertility like farrowing interval or interval between weaning and first service. Intensive selection on production and reproduction traits without considering functional traits can lead to decreased longevity (Engblom et al. 2008). Today, the majority of sows in the Polish commercial herds are removed in their early parities: sows are generally replaced at the third or fourth parity, and most cullings are unplanned, mainly due to reproductive disorders or locomotor problems. An increased herd life is associated with reduced replacement costs, higher productivity and increased possibility for selection on other traits (Hoge & Bates 2011). Despite the economic significance of longevity, there is currently no genetic evaluation for longevity of sows in Poland. Sow longevity is a complex trait, and it can be defined in different ways, most commonly as the ability of a sow to reach nth parity (stayability), total days in the breeding herd (length of productive life) and total piglets born in the sow’s lifetime (lifetime production), or a combination of the last two (pigs produced per day of herd life). There is a relatively high genetic correlation among the different sow longevity traits (Engblom et al. 2009; Sev on-Aimonen & Uimari 2012; Sobczy nska et al. 2013) suggesting that same genes impact the traits. Length of productive life combines several traits related to fertility, health and production like number of piglets born alive and weaned, milk production and leg soundness. Consensus has not been reached in the scientific literature regarding the relationships between longevity and growth and body composition traits (L opez-Serrano et al. 2000; Yazdi et al. 2000; Serenius & Stalder 2004; Knauer et al. 2010; Nikkil€ a et al. 2013). Genetic correlations vary across the trait definitions and among the populations studied. The genetic correlations between longevity and average daily gain and backfat obtained by Sobczy nska et al. (2013) in Polish Landrace breed were unfavourable but generally very low with high standard errors. However, the genetic correlations of longevity traits with lean meat percentage and phenotypic selection index reported in that study indicate that an antagonism exists between growth performance and sow longevity. Analysis of longevity traits using linear models may not be appropriate because of the skewed distribution and the bias caused by excluding the records of living sows or of assuming their records to be complete. Moreover, there are some factors that change throughout lifetime and affect an animal’s longevity. The management practices, natural environment and © 2015 Blackwell Verlag GmbH
• J. Anim. Breed. Genet. 132 (2015) 318–327
Study of longevity in Landrace sows
production level like number of piglets born in consecutive parities vary throughout a sow’s life. Survival analysis allows use of non-linear models and also tends to give higher estimates of heritability than does the linear model suggesting increased reliability of sires EBV and increased selection accuracy (Sewalem et al. 2005). The implementation of survival analysis in genetic evaluation (Ducrocq & S€ olkner, 1998) allows for processing censored as well as uncensored data, time-dependent covariates and random covariates. Therefore, the objective of this study was to estimate, using survival analysis methodology, the genetic variation for longevity, investigate relationships between longevity and production traits, compare estimates of genetic parameters from the sire and animal models and compare the sire rankings of longevity obtained from linear and survival analysis using the same data set.
Materials and methods Data
Data included farrowing and growth records collected from the Polish Landrace (L) sows with at least one farrowing over the period from August 1994 until March 2011. In order to obtain reasonable contemporary groups, data were limited to herds of over 300 sows in the entire study period. The average number of sows recorded per herd was 543 (range: 303–1286, SD = 271). Sows on all farms were housed during gestation and farrowed indoors. Sows were usually mated naturally, and at farrowing, they were kept in individual crates. In general, the piglets were weaned at approximately 4 weeks of age. The longevity and reproductive traits were first farrowing age (AFF), number of piglets born alive in the litter (NBA), number of dead piglets in the first litter (ND1), length of productive life (LPL) defined as number of days between first farrowing and removal or the end of data collection. Length of production life was considered as functional longevity, reflecting the ability of sows to delay involuntary culling and the true longevity defined as the ability to delay any culling. The performance traits were daily gain (ADG), backfat thickness (BF) and Longissimus muscle depth (LMdepth) registered as a part of the on-farm testing programme by the breeding consultant. The growth rate was defined as the average weight gain per day from birth until the test day. The growth rate was adjusted to 180 day of age according to National Research Institute of Animal Production (PIB) guidelines prior to the analyses. 319
ska & T. Blicharski M. Sobczyn
Study of longevity in Landrace sows
Backfat thickness was defined as the mean of the depth of fat measured by ultrasound equipment (PIGLOG 105; SFKâ, Soborg, Denmark) at two points P2 and P4 (3 and 8 cm, respectively, away from the back mid-line behind the last rib). Longissimus muscle depth was measured behind the last rib approximately 8 cm off the mid-line by the same ultrasound equipment as BF. Backfat thickness and LMdepth were adjusted to a constant weight of 110 kg using the formula determined by PIB. The animals with extreme values for the age at first farrowing (≤250 days and ≥550 days) or with missing parity were excluded to remove questionable records. Only sows with for which identity of their sire and dam was known were included. The final data contained 13 031 sows from 24 herds. Records were treated as censored if sows were still alive at the end of the studied period: on 31 March 2011. Descriptive statistic for some production traits and reproductive performance, as well as survival analysis statistics, was presented in Table 1. Statistical analyses
The length of productive life was evaluated with survival and linear analyses, using a sire and animal
Table 1 Descriptive and survival statistics for Landrace sows included in the analysis Descriptive statistics Traits LPL NBA ND1 AFF ADG BF LMdepth Survival statistics N Records Elementary records Wright censored records Days Min censoring time Max censoring time Average censoring time Min failure time Max failure time Average failure time
Mean SD 596.8 417.3 11.5 1.8 0.44 0.89 343.6 37.9 627.7 73.1 10.8 1.9 51.3 5.5
Range 7–2784 0–22 0–14 252–547 407–940 5.5–25 34–81
13 031 67 237 2822 (21.6%) 28 2087 617.2 7 2784 591.2
LPL, Length of productive life; NBA, number of live-born piglets in the litter; ND1, number of dead piglets in the first litter; AFF, age at the first farrowing; ADG, adjusted daily gain; BF, adjusted backfat thickness; LMdepth, adjusted Longissimus muscle depth.
320
model. The same data set was used for both analyses, even in the presence of censored observations. Survival analyses were performed using the Survival Kit V6.0 (Ducrocq et al. 2010). Removal hazard during LPL was analysed with the following Weibull model: hðtÞ ¼ h0 ðtÞ exp ½xðtÞ0 b
ð1Þ
where h(t) is the hazard function of a sow’s LPL at time t, t is time measured in days from first farrowing until culling, death or censoring, h0 (t) is the baseline hazard function kq(kt)q1, assumed to follow a Weibull distribution with location (k) and shape (q) parameters, b is a vector of fixed and random effects and x(t) is the corresponding incidence matrix. Vector b contained covariates as follows: fixed time-dependent farm–year effects, changed at 1 January each year, fixed time-independent birth season effect with four classes: December–February (i), March–May (ii), June– August (iii) and September–November (iv), fixed time-independent year–season (listed above) of first farrowing with 64 classes, fixed time-independent regressions of age at first farrowing and number of dead piglets in the first litter, fixed time-dependent total number of piglets born in the litter, changed at each farrowing date, fixed time-independent effects of ADG, BF and LM depth classified into five categories taking cut-off points at percentiles 20, 40, 60 and 80 (Table 2) and random genetic component defining an animal or sire effect, depending on the model. The solution for fixed effects was expressed as hazard ratio (HR). Hazard ratio was defined as the ratio between the estimated hazard for being removed under the influence of certain factors and the estimated hazard for a reference class. The first class (the lowest traits values) for ADG, BF and LMdepth was considered as reference class. The reference class for season of birth was December–February. In the case of continuous variables, the estimates of b can be expressed in relative risk by exp(b). Two separate longevity analyses were carried out: the sire or animal variance for longevity was Table 2 Cut-off points for the factor variables included in the analysis Percentiles
ADG (g/day) BF (mm) LMdepth (mm)
20
40
60
80
566 9.4 46
603 10.3 50
638 11 53
687 12.2 56
© 2015 Blackwell Verlag GmbH
• J. Anim. Breed. Genet. 132 (2015) 318–327
ska & T. Blicharski M. Sobczyn
calculated using a complete and a reduced model. The complete model included all variables listed in (1) and defined functional longevity. The reduced model contained herd–year, sire (or animal), season of birth, year–season of the first farrowing and age of the first farrowing terms and was referred to true longevity. To test the assumption for Weibull distribution, the log of minus the log of the Kaplan–Meier estimate of the survivor curve was plotted against log of time. The plot of this relationship produced approximately straight line (except for a period after the first farrowing when there was intensive culling), which indicates that the Weibull distribution fits data well (Figure 1). The proportion of explained variation by the model was calculated as R2 of Maddala R2M ¼ 1 ðLR =LU Þ2=n where LR and LU represent restricted and unrestricted maximum likelihoods and n denotes the total sample size. For proportional hazard model assuming a Weibull survival function, the heritability of longevity on the logarithmic scale was calculated as h2log ¼ 4r2s = ðp2 =6 þ r2s Þ for sire model and as h2log ¼ r2a = ðp2 =6 þ r2a Þ for animal model, where p2/6 is the variance of the standard extreme value distribution, r2s and r2a are sire and animal variance, respectively. The heritability on the original scale using the sire variance was calculated as h2ori ¼ 4r2s =½ðexp f1=qvgÞ2 ðp2 =6 þ r2s Þ for sire model and as h2ori ¼ r2a =½ðexp f1=qvgÞ2 ðp2 =6 þ r2a Þ for animal model (Yazdi et al. 2000), where q denotes shape parameter and v is –Euler’s constant. In the linear analysis, the following mixed model was fitted for LPL:
Study of longevity in Landrace sows
y ¼ Xb þ Zu þ e
ð2Þ
where y is a vector of observations of LPL, b is the vector of fixed effects, X is an incidence matrix for the fixed effects, the vector u contains random animal or sire effects, and Z is the associated incidence matrix, e is the vector of residual effects. For the linear model (2), the same fixed effects as for the complete Weibull model (1) were used. In the mixed linear analysis, the variance components for LPL, ignoring censoring, were estimated by the REML method with animal and sire models using the VCE 6.02 (Groeneveld et al. 2010). In addition, EBV for LPL for the sires was estimated under PEST package (Groeneveld et al. 2001) with the same statistical models. Models accounted for pedigree information up to the third generation of ancestors. The pedigree file for animal model included 16 814 animals. The number of sires and dams of the total population were 1368 and 15 451, respectively. The pedigree used in the sire model consists of 1225 animals and included pedigree information only for the sows’ sire (N = 1026). The number of daughters for each sire ranged from 1 to 230 (average 13). Correlations between the estimated breeding values from survival analysis and linear model were calculated to assess the agreement between the genetic predictions from the two methods. Estimated breeding values for sires with at least 10, 20, 30 and 40 uncensored daughter records obtained by survival and linear analyses were compared using Spearman’s rank correlation. The number of sires that met these criteria were 399, 197, 107 and 63, respectively. For survival analysis of LPL, a low value of EBV is desirable indicating a low removal hazard, whereas high values are desirable in the linear analysis indicating long herd life. Results and discussion
Figure 1 Log-cumulative hazard plot, S(t) – Kaplan–Meier survivor function, t – days from first farrowing.
© 2015 Blackwell Verlag GmbH
• J. Anim. Breed. Genet. 132 (2015) 318–327
The average LPL in Landrace sows was 597 days with large standard deviation (417). Exactly, the same LPL in Landrace sows obtained M esz aros et al. (2010) in Austria. Similar LPL was reported by Yazdi et al. (2000) for Landrace sows in Sweden, whereas the average LPL for Yorkshire sows in USA and cross-bred sows in Sweden was 485 days (Engblom et al. 2009; Hoge & Bates 2011). On average, uncensored females in this study had a herd life slightly shorter than censored ones (591 versus 617 days). Results obtained in different studies are difficult to compare due to different amount of censored records and length of the study period. 321
ska & T. Blicharski M. Sobczyn
Study of longevity in Landrace sows
Table 4 Hazard ratios (HR; Chi2 p-values in parentheses) for factors included in the animal model
Risk analysis for sow removal
The shape parameters of the Weibull distribution were very similar in the animal and sire analyses amounting 1.96 and 1.92, respectively. The significance of different covariates included in the full Weibull animal model is presented in Table 3. The hazard ratios for production traits as well as for the birth season and AFF are given in Table 4. Results of likelihood ratio test for significance of effects and risk ratios obtained from the sire model were very similar (not shown). Among the factors evaluated in the Weibull model, the herd*year effect had the largest influence on the risk of culling of sows. The proportion of explained variation by the model (R2 of Maddala) increased drastically (to 28%) when herd*year interaction was added. The second covariate that most influenced sow longevity was year*season of the first farrowing. These two factors explained nearly 97% of all the variation. Herd*year effect contains different management practices and housing systems among farms where improvements can probably reduce the unplanned removal of sows. High year*season effect of the first farrowing interaction may reflect fluctuation of pig prices at the market. These prices have shown a seasonal pattern and influence production and culling decisions. Earlier studies have found significant associations between the herd–year interaction and sow longevity (Yazdi et al. 2000; Serenius & Stalder 2004, 2007). Season of birth had a significant effect on the sows’ risk of being culled. Sows born in spring season (March–May) had a 24% (p < 0.001) lower hazard for removal than those born in the reference class (Table 4). Compared with sows born in winter season (December–February), sows born in autumn (SepTable 3 Likelihood ratio test when all covariates added to Weibull animal model sequentially Covariatea
d.f.
v2
p Value
R2M
Herd*year Year*season of first farrowing Season of birth AFF ADG BF LMdepth ND1 NBA Animal
296 63 3 1 4 4 4 1 1
4397.3 4003.9 98.9 43.3 7.8 10.1 3.2 2.7 227.2 RANDOM
0.000 0.000 0.000 0.000 0.100 0.038 0.525 0.100 0.000
0.286 0.475 0.479 0.480 0.481 0.482 0.482 0.483 0.491
a
See Table 1 for abbreviates.
322
Covariatea
Class
No.b
HR
ADG
1c 2 3 4 5 1c 2 3 4 5 1c 2 3 4 5 1c 2 3 4
2311 2117 1961 1937 1883 2016 2023 2094 1909 2167 2609 2115 2086 1675 1724 2491 2489 2664 2569 10 209 10 209 10 209
1 1.08 (0.03) 1.05 (0.15) 1.03 (0.48) 1.07 (0.13) 1 0.95 (0.10) 0.97 (0.33) 1.03 (0.49) 1.03 (0.43) 1 1.04 (0.22) 0.99 (0.81) 1.03 (0.47) 0.98 (0.65) 1 0.76 (0.00) 0.99 (0.98) 1.08 (0.03) 0.91 (0.00) 1.04 (0.00) 1.002 (0.00)
BF
LMdepth
Season of birth
NBA ND1 AFF a
See Table 1 for abbreviates. Number of uncensored sows in each class. c Used as the reference class. b
tember–November) had 8% (p = 0.034) greater hazard for removal. It is well established (Black et al. 1999) that climatic conditions influence feed intake and metabolism in the pigs, and thus production and reproduction traits. Seasonal effects on fertility in gilts and sows have been showed by Tummaruk et al. (2000). The additional changes in R2 of Maddala were small when adding AFF, production and reproduction traits. Similar to many studies on Landrace and other pig breeds (e.g. Serenius et al. 2006, 2008; Serenius & Stalder 2007; Engblom et al. 2008; Hoge & Bates 2011), our analysis also revealed that age at first farrowing had a small but significant effect on culling. The hazard regression coefficient for AFF was 0.002 per day. Different results were obtained by Serenius & Stalder (2004) for two pig breeds: in Large White, negative genetic association between AFF and LPL (0.28) was found, whereas the sign of corresponding correlation was positive in Landrace (0.17). An increased AFF could be attributed to gilts that did not conceive on the first service. Delayed AFF could be considered an early indicator of reproductive fitness or lack thereof (Hoge & Bates 2011). On the other hand, the important factor in determining when to © 2015 Blackwell Verlag GmbH
• J. Anim. Breed. Genet. 132 (2015) 318–327
ska & T. Blicharski M. Sobczyn
mate or inseminate a gilt is probably prepubertal growth rate, which depends on environmental (herd management system) and genetic effects. The effect of NBA was highly significant (p < 0.0001). If a sow had one more piglet in her litter, her relative culling risk was reduced by 9%. Yazdi et al. (2000) also reported a decreased risk ratio with increasing litter size for both first and last litter. Similarly, Serenius & Stalder (2007) and Hoge & Bates (2011) reported that a small number of piglets born alive at first farrowing had a detrimental effect on LPL. The Landrace sows that had one more pig born alive in their first litter had a longer length of productive life of 4.6 days in the study by Sobczy nska et al. (2013). However, Serenius et al. (2006) concluded, comparing six genetic lines, that there was no clear association between litter size and LPL when culling due to poor reproductive performance was not practiced. Significant effect of NBA is not surprising as number of weaned piglets per parity is a main factor that determines profitability in piglet producing herds and sow productivity is a primary removal reason in voluntary culling. As ND1 increased by one piglet, the culling rate increased by 4%. This indicator of neonatal piglet mortality had a significant negative association with longevity. Hoge & Bates (2011) also concluded that Yorkshire sows that had fewer stillborn piglets in the first litter had a decreased risk of removal. The average daily gain had very low effect on the risk of culling as only sows with gains 567–603 g/day tended to have 8% more risk of being culled (p = 0.03) compared with sows from the reference group (ADG < 567 g/day). The effect of ADG was similar in other groups suggesting a negative phenotypic association between daily gain and LPL; however, it was not significant. In the studies of Serenius et al. (2006) and Knauer et al. (2010), the relationship between average daily gain and longevity was significant and negative in some genetic lines only. Hoge & Bates (2011) and Serenius & Stalder (2007) found that slower growing gilts tended to have a decreased risk of being culled. Similarly, L opez-Serrano et al. (2000) and Sobczy nska et al. (2013) reported unfavourable genetic correlations between growth rate and longevity in Landrace and Large White breeds, but in Landrace sows, this relationship was weak and not significant. Moderately unfavourable genetic correlations were found for growth rate and longevity by Nikkil€a et al. (2013) when REML and Gibbs sampling were used in analysis. Results obtained by Tarr es et al. (2006) suggest a negative phenotypic association between ADG after the growth test and LPL. Thus, it © 2015 Blackwell Verlag GmbH
• J. Anim. Breed. Genet. 132 (2015) 318–327
Study of longevity in Landrace sows
appears that ADG may affect sow longevity in some breeds or genetic lines while not in others; however, most studies imply that fast growing gilts have inferior longevity. Backfat thickness on the day of performance test had no effect on the risk of culling; however, the relative risk ratios show a marginal decreased rate of culling for sows with backfat thickness between 9.5 and 11 mm. The phenotypic or genetic association between backfat thickness and sow longevity is not clear. Two studies found unfavourable correlation, that is gilts with thicker backfat stayed and produced longer in the herd (L opez-Serrano et al. 2000; Hoge & Bates 2011), whereas some studies found no association (Yazdi et al. 2000; Serenius & Stalder 2004, 2007; Sobczy nska et al. 2013) and another study found that an intermediate optimum backfat (16–19 mm) at first farrowing was favourable for sow longevity (Tarr es et al. 2006). Results obtained by Serenius et al. (2006) indicate that gilt backfat level affects whether gilts successfully conceive and farrow their first litter, whereas the association between backfat and sow longevity is weaker for sows that complete their first parity. Loin depth had no effect on sow longevity in this study. This result is not consistent with the higher risk of culling in Duroc sows with loin depths
