New breeding value evaluation of fertility traits in Finnish mink

  • View

  • Download

Embed Size (px)



    New breeding value evaluation of fertility traits in Finnish mink


    MTT Agrifood Research Finland, Biotechnology and Food Research, Biometrical Genetics, FI-31600 Jokioinen, Finland

    AbstractLitter size (LS) has been included in the Finnish mink breeding goal for several generations. Still, the phenotypic trend inthe average number of pups per mated female has slightly decreased while animal size (AS) has increased. The aim of thisstudy was to estimate genetic parameters for pregnancy rate (PREG) and felicity (FEL), and their genetic correlations to LSand AS. The estimated heritabilities were low for PREG (0.032) and FEL (0.026). The genetic correlations between LS andPREG (0.34), and LS and FEL (0.53) were clearly positive. Thus, on average females having genetically larger LS havehigher PREG and FEL. The genetic correlation between AS and PREG was low (0.13), and correlation between AS andFEL was moderate (0.27) indicating that larger animals are more likely barren or lose their kits during pregnancy or rightafter birth.

    Keywords: Fertility, litter size, mink, pregnancy.


    The main goals in Finnish mink breeding have been

    improved fur quality, and increase in body size and

    litter size (LS). Consequently, average pelt size has

    increased considerably. However, at the same time

    the average number of kits per mated female has

    slightly decreased in Finland as well as in other

    countries (Hansen & Berg, 2007, 2008; Hansen,

    2009). Increase in body size may have lead to smaller

    LS (Hansen & Berg, 2007, 2008; Hansen, 2009;

    Koivula et al., 2009b, 2010). This seems to be a

    problem also in other species because when selecting

    for body size negative genetic trend has often been

    observed in traits measuring reproduction and

    survival despite their importance to profitability

    (Peura et al., 2007; Goddard, 2009; Koivula et al.,

    2009a). One reason is a strong negative genetic

    correlation between large animal size (AS) and LS.

    For example in minks this correlation has varied

    from 0.18 to 0.28, (Lagerkvist et al., 1994;Rozempolska-Rucinsca, 2004; Peura et al., 2007;

    Koivula et al., 2009b, 2010), and in blue foxes

    negative correlation has been even higher (0.36 to0.43) (Peura et al., 2007; Koivula et al., 2009a).

    Fertility can be measured in many ways. In

    Finland, farmers record mink LS at two weeks after

    whelping, but also barren females, aborting females

    or females losing their kits are recorded in a routine

    recording scheme. However, breeding values are

    based on LS only, and barren females or females

    aborting or losing their kits are not included in the

    breeding value evaluation. In blue foxes heritability

    of pregnancy rate (PREG) was 0.028 and felicity

    (FEL) 0.049, and the genetic correlations between

    LS and PREG and LS and FEL were clearly positive

    (Koivula et al., 2009a). Thus, it was possible to

    include these traits in breeding programmes of

    Finnish blue foxes.

    The aim of this study was to estimate genetic

    parameters for PREG (representing the proportion

    of females whelping, i.e. the non-barren females)

    and FEL (including both aborting females and

    females losing all kits after birth), and their genetic

    correlations to LS and AS. In addition, we examined

    genetic trends in the traits studied.

    Correspondence: M. Koivula, MTT Agrifood Research Finland, Biotechnology and Food Research, Biometrical Genetics, FI-31600 Jokioinen, Finland.

    Tel: 358-40-1960986. Fax: 358-3-4188-3244. E-mail:

    Acta Agriculturae Scand Section A, 2011; 61: 16

    (Received 14 September 2010; revised 1 November 2010; accepted 2 November 2010)

    ISSN 0906-4702 print/ISSN 1651-1972 online # 2011 Taylor & FrancisDOI: 10.1080/09064702.2010.538715

  • Material and methods

    Mink data were obtained from the Finnish Fur

    Breeders Association. The data had information

    from 3.7 million animals. Data for the variance

    component estimation were sampled from the full

    data. Sampling was done by farm. The complete

    pedigree contained about 4.1 million animals from

    136 farms. The pedigree had many disconnected

    subpopulations, so it had to be pruned with Relax2

    (Stranden & Vuori, 2006) to have only informative

    animals. In the end, the sample had observations

    from 12 farms having 69,441 animals born in years

    19982006. The pedigree file contained 93,632animals.

    The analysed traits were the first parity LS,

    PREG, FEL and AS. LS was recorded as numbers

    of kits alive two weeks after whelping. PREG and

    FEL were binary (1/0) traits, value 0 representing

    the event when the female was barren or aborted/

    lost her kits. Females were scored as pregnant when

    she showed visual signs of pregnancy. If pregnancy

    was recorded, PREG1, and also if PREG recordwas missing but FEL and LS records exist. If

    abortion or kit loss was observed, females were

    recorded as FEL0, similarly if she was recordedas pregnant and LS was missing. Because all

    pregnant females and all females giving birth or

    losing kits are not observed, PREG and FEL are

    always approximations. Abortion and kit loss after

    birth was treated as a single FEL trait, because the

    number of observations for aborting females was so

    low that it would have been difficult to analyse it as a

    separate trait. AS was graded subjectively by the

    farmer. The grading scale ranged from 1 (smallest)

    to 5 (largest). The recommendation was that the

    average AS should be close to 3 within farm and


    Restricted maximum likelihood (REML) esti-

    mates of (co)variance components were calculated

    using DMU software (Madsen & Jensen, 2000). The

    multi-trait animal model was:


    where y is a vector of observations, b is the vector of

    fixed effects, c is the vector of random effect of the

    litter in which the female is born, and a is the vector

    of random genetic effects for animal and e is the

    random residual, and X, W and Z are known

    incidence matrices for the fixed and random effects.

    Random effects were assumed to be independent

    and normally distributed. In particular, c N(0;C0I); a N (0;G0A); e N (0;R0I)where C0 is common litter effect (co)variance

    matrix, G0 is direct additive genetic (co)variance

    matrix, A is numerator relationship matrix and R0residual (co)variance matrix.

    The fertility traits were exclusive by nature of their

    definition: when PREG had value 0, both FEL and

    LS information were missing; when PREG had value

    1, and FEL had value 0, then LS was missing. Thus,

    LS was observed only when both PREG and FEL

    had value 1. Consequently, the residual covariance

    between LS and PREG, LS and FEL, and PREG

    and FEL was assigned as non-existing (zero) because

    these trait combinations are not present in the data,

    and thus, cannot be estimated.

    Fixed effects for the traits were studied with the

    general linear model by excluding random effects

    other than the residual (SAS, 2004). Fixed effects

    for LS, PREG and FEL were farmyear, time ofbirth for animal (three classes: 99119, 120140 and141160 days from the beginning of the year,reflecting timing of birth and thus also age of animal)

    and number of matings (three classes: 1, 2 or 2mating/season). Fixed effects for AS were farmyear,time of birth for animal, sex of animal (three classes:

    male, female and unknown) and age of dam (three

    class: 1, 2 or 3-years-old).Heritability (h2) and proportion of common litter

    variance (c2) for the traits were calculated as h2s2a=(s2as2c s2e ); and c2s2c =(s2as2c s2e );where s2a; s2c and s2e are trait variances of additivegenetic, common litter environment and residual,

    respectively. Linear animal model was used to

    analyse PREG and FEL, although theoretically a

    threshold model would be more appropriate for

    analysis of binary data (Gianola, 1982). Heritability

    calculated on the observed binary scale varies with

    incidence because the amount of variance due to

    measurement error depends on the incidence. To

    overcome this, heritabilities were converted from the

    binary to the continous scale using Dempster and

    Lerner (1950) formula:


    where h2 is the heritability in the continuous scale,

    h201 is the corresponding heritability calculated on the

    binary scale, p is the incidence of affected individuals

    in the population, and z is the ordinate of the

    standard normal density function on the threshold

    corresponding to the incidence p.

    In addition to the genetic parameters, genetic

    trends for the studied traits were assessed by

    examining standardised estimated breeding values

    (EBV). EBVs were calculated with MiX99

    (Stranden & Lidauer, 1999; Vuori et al., 2006).

    The largest subpopulation was used for EBV calcu-

    lations. The data included observations from

    395,233 animals and the pedigree had 451,643

    animals from years 19882006 in 59 farms. The

    2 M. Koivula et al.

  • model used in the EBV calculation was the same

    as in the variance component analysis but the

    variances were the obtained REML estimates.

    Breeding values were standardised to year 2003

    with mean 100 and SD 10 in order to make

    comparison of years and EBVs of different traits