Multitrait Restricted Maximum Likelihood Estimates of Genetic and Phenotypic Parameters of Lifetime Performance Traits for Canadian Holsteins

Multitrait Restricted Maximum Likelihood Estimates of Genetic and Phenotypic Parameters of Lifetime Performance Traits for Canadian Holsteins L. K. JAIRATH, J. F. HAYES, and R. 1. CUE Department of Animal Science Macdonald Campus of McGill University 21,111 Lakeshore Road Ste. Anne de Bellewe, W,Canada H9X 3V9 ABSTRACT

and genetic progress for these traits has been well documented (32,35). However, high milk yields may be associated with physical or physiological changes that tend to limit further increases in productive or economic herd life (8, 10, 15); for instance, high yielding cows are more susceptible to mastitis, milk fever, and ketosis (27, 30). Dairy producers should be concerned more with maximizing the profitability of the cow than with maximizing milk yield (19, even if that approach implies suboptimal genetic progress for milk yield and component traits (22). For a cow to be profitable, the costs involved in raising cows must first be amortized over several lactations; only after such costs are recovered does milk revenue over direct costs (costs of feed, veterinary care, and insemination) become net profit. Long productive life not only reduces rearing costs per year of productive life but also allows exploitation of maximum milk capacity of the cow and increased voluntary culling. Selection goals for dairy cattle should reflect profit over lifetime of cows (2, 19, 31). but increased milk yield should be considered to be only part of the goal of maximizing profit. Most studies reporting on estimates of parameters of lifetime performance traits have included lifetime yield and longevity as traits of lifetime performance (5, 7, 11, 13, 18, 23, 26, 36, 37), although lifetime profit functions have also been considered (8,9, 15, 20). However, estimates of genetic parameters of lifetime performance traits from these studies have been inconsistent. To maximize genetic INTRODUCTION progress for lifetime profit, unbiased and preMilk yield and its components are the most cise estimates of genetic and phenotypic important traits of selection for dairy cattle, parameters of components of lifetime profit are needed. Knowledge of relationships among lifetime performance traits is important for prediction of expected correlated response to Received April 30. 1993. selection. Acceptad August 16, 1993. Data on 82,835 Holstein cows, daughters of 703 sires and with first calving from September 1979 to December 1984 from 2384 herds enrolled in the Quebec Dairy Herd Analysis Service, were used to estimate genetic and phenotypic parameters of partial and total lifetime perfmance traits with REML. The model included herd-year-season of fmt calving, age at first calving, and proven sires as fixed effects and young sires and residuals as random effects. Individual lactation records were precorrected for year-month of calving before lifetime totals were calculated. Only cows with at least 5 yr of opportunity for production were analyzed. The ranges of heritability estimates were .I1 to .I3 for lifetime production and profit, .07 to .09 for measures of longevity, and .28 to .32 for yield per day of productive life. Correlations among total lifetime yield and profit traits and among measures of longevity were 2.93. Genetic and phenotypic correlations, respectively, of early partial (two parities total) with total lifetime yield and profit and longevity traits ranged from .81 to .94 and .66 to .78. Selection on early performance seems to be desirable and, given the high positive genetic correlations, should increase both lifetime yield and longevity. (Key words: lifetime, Holsteins,, restricted maximum likelihood)

1994 J Dairy Sci 77:30>312

303

304

JAIRATH ET AL.

Most of the estimates of parameters of lifetime performance traits reported in the literature were obtained using ANOVA type estimators, based on the three methods of Henderson (17, 25). which assume that the data are randomly sampled and therefore tend to give biased estimates in the presence of selection. Also, in the majority of studies, estimates of correlations were obtained with univariate analysis procedures and, hence, were not free of the effects of selection on correlated traits (3, 16). Recent advances in computer technology have made feasible the use of multivariate REML methods that have optimal statistical properties and that, under certain conditions, can consider selection when variance components are estimated for animal breeding applications (17). No estimates of parameters of lifetime performance traits are available for Canadian Holsteins. The present study was undertaken to estimate genetic and phenotypic parameters of traits associated with part and total lifetime yield and profit, longevity, and yield and profit per day of productive life using multivariate REML methodology. MATERIALS AND METHODS

Data

A total of 929,054 lactation records, initiated and collected from September 1979 to December 1989 on 383,097 cows, were obtained from the Quebec Dairy Herd Analysis Service. Test day records for individual cows contained information on dollar value of milk, cost of feed, milk yield, and milk fat and protein percentages. This information was used to calculate cumulative milk, fat, and protein yields for individual lactations, value of cumulative milk, and cumulative feed cost for individual cows. Milk price and fat differential (without payment for protein) were reported for individual herds and reflect the average price received for milk after transportation and administrative charges were deducted. The milk pricing system did not change during the time of the data and was approximately W/hl of milk with a 60 to 646 fat differential for each .l%deviation from 3.6% fat. Forages were fed as part of the base ration at up to four different feeding levels in the herd. The amount conJournal of Dairy Science Vol. 77, No. 1. 1994

sumed by each cow was estimated based on the weight of each forage provided to each group adjusted according to the ratio of the cow’s BW to the group’s average BW. A more detailed description on feed intake and its measurement was given by Moore (19). Data Editing

Editing of data consisted of the following: 1) restricting age at first calving to 18 to 40 mo; 2) requiring all cows to have consecutive lactations, starting with first lactation; 3) restricting calving interval to 270 to 600 d; 4) removing records for cows with duplicate termination codes; 5) discarding records for cows with duplicate records; 6) requiring all cows to be in the same herd throughout their lives; 7) removing records for cows with code “sold for export”, “sold for dairy“, or “rented to”; and 8) allowing cows that remain in the herd at least 5 yr of opportunity for yield (i.e., records for a cow still in the herd at the time of obtaining the data were retained in the data only if 5 yr had elapsed since the cow’s first calving). The latter criterion meant that only records for cows calving for the first time prior to 1985 were included in the analysis. Furthermore, all remaining records were for cows with observations on milk, fat and protein yields, value of milk, and cost of feed for all lactations. Lactations starting after August 1988 were not included because they would involve a large number of incomplete lactations, and completed records after this date would tend to be shorter lactations. The edited data contained lifetime records on 90,560 cows. Tralta

The traits examined in this study included traits associated with lifetime yield (milk, fat, and protein yields) and profit (value of lifetime milk yield and lifetime fat differential over lifetime feed cost); yield and profit per day of productive life; and longevity (lifetime DIM, productive life, and number of lactations). TOtal for parities 1, 2, 3, and 4 of milk, fat, and protein yields and profit were included as partial lifetime traits. Total and partial lifetime yield and profit traits for an individual cow were obtained by accumulation of individual lactation yield and

GENETIC PARAMETERS OF LIFETIME TRAITS

profit records precorrected for the fixed effect of respective year-month of calving (Jairath et al., 1992, unpublished data). Productive life was defined as the total number of days from date of first calving to date of disposal or the last dry date if the cow remained in the herd. Per day of yield and profit traits were defined as total lifetime yield and profit divided by total days of productive life.

305

TABLE 1. Description of data for analysis of lifetime performance traits.

Rccords. no. S i , no.

Young sirts (random) Old sires (ked) Herds, no. Herd-year-seasons, no. Age. no. of levels Residual degrees of freedom

82,835 703 490

213 2,384 21,454 20 60,659

Statistical Analysis

Genetic and phenotypic parameters of lifetime performance traits were estimated with a 33-trait analysis (including first lactation traits of yield and profit, yield and profit per day, and percentages of fat and protein) using multivariate REML for a mixed model with an equal design matrix (17); herd-year-season and age at first calving were considered to be fixed effects, and sire and residuals were considered to be random effects. Proven sires were fixed effects to avoid selection bias (25, 35). Inclusion of records on daughters of proven sires increases the degrees of freedom for estimation of within-sire variance component and allows more accurate estimation of fixed effect of herd-year-season. There were 213 sires treated as fixed effects and 490 sires treated as random effects. All known relationships among sires were used to construct the relationship matrix. Inclusion of relationships should take into account most of the selection attributable to the four paths of improvement for dairy cattle (34). A restriction that a sire should have at least 10 daughters represented in at least 5 herds was imposed to ensure connectedness in the data and to reduce the bias caused by preferential treatment of daughters of a particular sire in a particular herd or group of herds. After these restrictions were imposed, lifetime records remained for 82,835 cows. The multivariate mixed model used for the analysis was

where Yiju is a vector of observations on the traits for cow ijkl, HYSi is a vector of fixed effects on the traits for herd-year-season i of first calving (21,454 levels, seasons were defined as March to August and September to

February), agq is a vector of fixed effects on the traits for age at first calving j in months (20 levels), Sk is a vector of effects on the traits for sire k (703 levels), and qju is a vector of random residuals associated with the traits for cow ijkl. Random sires and residuals were assumed to be distributed with zero means and (co)variances G*A and R*I, respectively, where G is the matrix of genetic (c0)variances among traits, A is the relationship matrix among sires, R is the residual (co)variance matrix among traits, I is the identity matrix, and the asterisk refers to the direct product of two matrices (28). Starting values for the analysis were taken from the average estimates of sire and residual variances published in the literature; the covariances were set to zero as their starting values. Iterations were continued until none of the parameters changed by >.05%. Heritability estimates and genetic and phenotypic correlations were computed from the converged estimates of sire and residual (co)variances. RESULTS AND DISCUSSION Means, Standard Deviations, and F l x d Effect8

Description of data and phenotypic means, standard deviations, and minimums and maximums of lifetime performance traits are

presented in Tables 1 and 2, respectively. Means for lifetime performance traits were comparable with those in other reports (5, 8, 11, 18, 37) but were generally lower for this population, possibly because all cows had not yet been culled and because of difference in editing criteria. The average number of lactations was 2.56. Average productive ,life was 833 d but ranged from 8 to 3569 d. CoeffiJwrnal of Dairy Science Vol. 77, No. 1. 1994

306

JAIRATH ET AL.

cients of variation for lifetime performance traits were higher than for other published investigations (14, 33, 37) with more restricted definitions of lifetime traits (e.g., 2 180 DIM, at least two calvings, and removal of terminal lactation records). Most lifetime performance traits tended to be skewed to the right. A preliminary analysis using REML and a subset of lifetime performance traits revealed the heritabilities and genetic and phenotypic correlations of log-transformed traits were unchanged from the untransformed traits. Based on these results, log transformations of lifetime performance traits were not utilized in this study. This result further supported the results of various authors (1, 12), indicating that REML estimators may be an appropriate choice even when normality does not hold. The fixed effect of age at fmt calving accounted for a very small (.l to .2%) but significant amount of variation (P S .01) for all lifetime performance traits.

Estimates of HerbbllMea for Ufetlme Performance T r a b

The estimates of heritabilities of partial and total lifetime production and profit traits ranged from .11 to .18 (Table 3). The estimates of partial lifetime traits were slightly higher than the estimates of total lifetime performance traits, as expected, because residual variation accumulates and variation increases as the length of herd life increases (Jairath et al.. 1992, unpublished data). Variance components among sire progeny groups increased, but variation within bull progeny groups increased even more, resulting in lower heritabilities for total lifetime performance traits. Heritability of lifetime milk (.13) was within the range of heritabilities (.09 to .18) reported in the literature (7, 11, 13). However, a somewhat higher estimate (.2S)was reported by Gill and Allaire (8). A limited number of herds in their study may have led to larger estimates of

TABLE 2. Phenotypic means. standard deviations, minimums and maximums, and cacfficients of variation for lifetime

performance traits. Trait'

-X

SD

MIL= kg FAT2, kg PRT2, kg PRFZ, s MLK3, kg FAT3, kg PRT3, kg PRF3, M L M . kg FAT4, kg PRT4, kg PRF4, s LTMILK. kg LTFAT, kg LTPRT, kg LTPRF, s LTDIM PLIFE NLACT MPDPL, kg FPDPL, kg PPDPL, kg PFPDPL, Yd

9396.28 335.18 298.82 2035.35 12269.8 437.16 389.74 2700.22 13,965.4 497.38 443.26 3109.92 15237.5 542.35 483.31 3430.98 717.41 833.63 2.56 19.88 .7 .62 1.12

4818.86 173.26 154.27 1190.26 7705.88 275.66 245.43 1867.07 9986.81 356.79 317.24 2408.42 12,494.4 445.43 395.82 3010.6 513.22 620.91 1.54 3.79 .14 .12 .ll

Maximum

cv

29.312.5 1037.38 947.47 9355.2 40.884 1543.85 1299.44 12,005.6 52,982.2 2065.16 1589.15 15,457.9 92,949.4 3228.93 2884.03 23,266.7 2885 3569 9 38.37 1.55 1.2 7

51.28 51.69 51.63 58.48 62.8 63.06 62.97 69.15 71.51 71.73 71.57 77.44 82 82.13 81.9 87.75 71.54 74.48 60.19 19.07 20.02 19.32 10.52

(96)

s

5.42 1.61 2.15 -1.73 5.42 1.61 2.15 -1.73 5.42 1.61 2.15 -1.73 5.42 1.61 2.15 -1.73 8 8 1 3.58 .11 .11 .81

'MIL? = TWo-paritY milk, MILK3 = thrcc-pkty d k . MILK4 = fm-parity milk, FAT2 = two-parity fat. FAT 3 = --panty fat, FAT4 = fOUr-parityfat, PR'I2 = t w 0 wprottin, PRT3 = thraGparity protein, PRT4 = fow-parity Proteh PRF2 = twO-p& profit. PRJ3 = thrcc-parity profit, PRF4 = fow-parity profit. LTMILK = Lifctimc d k , LTFAT = lifttim fat, LTPRT = life tin^ protein, LTPRF = lifetime p f i t , LTDIM = lifetime days in milk, PLIFE = days of productive lifc. NLACT = numba of ladatioas. MPDPL = milk per day of productive life, FPDPL = fat per day of productive life. PPDPL = protein per day of productive life, and PFPDK = profit per day of productive life.

Journal of Dairy Science Vol. 77,No. 1, 1994

PARAMETERS GENETIC PAR fuvImERs OF LIFETIME TRAITS

of sampling. sampling. Schneeberger heritability because of (26) and Casanova (26) reported a similar estimate !'i $1111:83 (24) k (.24) Brauenvieh. but for Swiss Brauenvieh. The heritability ~I~~~~~~~~~~~~~~~N F ~ 11II ...l r-:r-:r-:r-:~~~~~ .~~~~~-: of Metime I, was somewhat lower than cl lifetime fat, fat., .I .11, ~::.c E the range of .12 to .18) of most estimates estimates ( (.12 .18) in the E;~~I~~~~~~~~~~~~ ~N~ d$ literature literature (11, (11, 13). Literature estimates estimates of of 1 3 ) . Literature F ~ t'-:r--:~~~~~~~~~c:~~~c: ~...l .25 (s) (8) and .35 .35 ( (15), P-1 heritability, .25 15),were relatively cl higher for lifetime lifetime profit t han the estimate than estimate of l~ !:Ie. .12 .12 obtained in the present study, study, possibly !( lifetime definition of lifetime ~I~~~~~~~~~~~~~~~~ .s~ because of a different definition ~"i profit. Lifetime profit, profit., as used by Gill and S profit. !( ~ Allaire (8) (8) and and Lin Lin and and Allaire Allaire (15), (15), was was de2~.s All& defined from birth to the the date in & II date of disposal disposal or last if a cow remained in the herd. date if .as processing Lifetime profit profit functions functions reported reported in the literalitera84 Lifetime in the complete definition ture were more complete in their definition and ~ii c.. ~~~~~~~~~~~-:~~~~ .s~ of input and output variaconsidered a number of II .ar;. ~'"" bles (e.g., insemination, veterinary health and I' S ~1;fi' care, care, labor and salvage value) in addition to E ~ '9 j feed consumption and milk yield, yield, which were Ir, i~] considered L study. The estimate considered in in the the present present study. The estimate ~ d b a ~ II !5 of for of .12, heritability lifetime protein .12, was I' 4 2~I~~~~~~~~~~~~~~~~ 1~: also No estimates literature low. NO estimates were in the literature Oc.."a ~ lifetime protein traits. traits. !..:~ for lifetime ::.c estimates of longevity longevity traits traits of II~S Heritability estimates ~I~~~~~~~~~~~~~~~~ «~e.' lifetime DIM, and DIM, productive life, total number lifetime ~V1 2~15l .av of lactations lactations were low and ranged from .07 .07to :!i, d o~. .09 (Table (fable 4). Literature Literature estimates estimates of heritabil's o;o;o;~~~~:q~~~~~~~~ for longevity longevity measured variously ranged ~~I' ~ ity 0 (5, 21) to .39 from (5, 21) .39 (38); (38);values near .15 .15 were 1 most frequent F frequent (10, (10,12, 12, 18, 18, 22, 22, 34). 34). Smith Smith and E .sP .~ - 4 Quaas Quaas (29), (29).reported heritabilities heritabilities of herd life life :$II ;.§"; of .13 .13 and .06 from from statistical statistical methods that ~ included censored records. records. Estimates Estimates of heritaF : :~ ~~ .... c.. bility for length of productive life were low, ~I~~~o;~~~~~~~~~~o;~ 2 low, '5~ .085 for both definitions (4), I) true stayability .OS5 definitions (4), 1) stayability ~ 6 2) 2) functional functional or "milk-corrected" stayabilstayabil....... ti and ity. l~ e. Heritability estimates estimates of yield per day of ' ' t;:.~ productive life traits in the present study c...j f ranged from from .28 to .32 (fable 4) 4) and were .28 to .32 (Table and were ~~~~~~~~~~~~~~~~ IIII '§.:=II ranged higher than than the corresponding corresponding estimates estimates from ~ ~t: EZ record recod of performance data for Canadian Hol~01c steins (13) (13) but lower than the estimates estimates of .49 -49 steins 8;~~~~~~~~~~~~~~~ E .40obtained using paternal paternal half-sib analyanaly:!i 'r'..: and .40 sis by Evans et al. (5) (5) and Gill and Allaire (8), ';~J! sis ~ c respectively. However, using *& II respectively. However, using daughter-dam daughter-dam 11 « d~ ~~~~~~~~~~~~~~~~ $ E a comparisons, Gill and Allaire (8) also comparisons, also reported o~:= No estimates estimate (.28). (.a). estimates were ~ g: a IIII a much lower estimate ~ in the literature literature for heritability of protein per II 61; ~~~~~~~~~~~~o;~~~ ~ ..... day of productive productive life. Heritability of profit per QJ productive life life was very low (.03). (.03). day of productive N "" ~ ~~~~ Because this this trait trait is is standardized, standardized, higher higher Because '@ ~~~ ~~~ ~~~:"t~«~i= ~ 'e.·s g i heritabilities were expected, expected, similar similar to other

o~

·~1 ~ ~

"0

i

~ .~

~ ~

307 307

.[

* i~

k

?.*

~ i,~~~~~~~~~~~~q~~~

4

tl~I~~~~~~~~~~~~~~~~ 'i~

'iI

d8 II~I~~~~~~~~~~~~~~~~ 5 &a

'S

!

i

o

'il -

~

.~

'j 3

'' '

!I~ B i

1'~I~~~o;~~~~~~~~~~~~ ~~·i.g ~

!

g

I8

~I~o;~~:q~~~~~~~~~~~

] i·

·tl~

4~ $7

11~

f! $5 ~'i~

·B

zj 1 J

t

I...., ~

4 5,_ ~

'C

,..;

~ ~

Q

-~l~ ~ ~~~~if~~~ if~ ItJ ~~...l ·i#

Journal Joumpl of Dairy Science Scitncc Vol. Vol. 77, No.1. No. 1, 1994 1994

308

JAIRATH ET AL.

TABLE 4. Heritabilities(diagonal), genetic comlations (below diagonal), and phenotypic correlations (above diagonal) of longevity and per day of productive life traits. Traits'

LTDIM

PLIFE

NLAm

MPDPL

FPDPL

PPDPL

PFPDPL

LTDIM PLIFE NLACT IUPDPL FPDPL PPDPL PFPDPL

.09 1.a0 .98 .85 .58 .74

1 .oo .OS .98 .85 .58 .74

.97 .97

.62 .62

.58 .58 .57

.64 .64

.44

.46

.so

.07 .82 .54 .70

.46

.a .32 .a .85 .56

.80 .30 .72 .51

.62 .93 .86 .28 '55

.49 .48 .30 .29 .32 .03

ILTDIM = Lifethe days h millr. PLIFE = days of productive life, NLACT = number of lactations, MPDPL = milk per day of productive life, FPDPL = fat per day of productive life, PPDPL = protein per day of productive life, and PFPDPL = profit per day of produaivc life. SE 5 .03, 5 .12, and S .15 for heritability and phenotypic and genetic correlations, respectively.

traits consisting of up to four lactations. Correlations among adjacent partial lifetime and between adjacent partial lifetime and total lifetime yield and profit traits were >.9 but decreased as the distance between them increased, this relationship was more evident for phenotypic correlations than for genetic correlations. Genetic and phenotypic correlations among total lifetime performance traits (Table 3) and among the three longevity traits (Table 4) were all close to 1, which agreed with other reports (11, 13, 20) and indicates that many of the same factors are involved in controlling these traits. Genetic correlations of partial lifetime traits and measures of longevity were higher than corresponding phenotypic correlations, especially for early performance traits consisting of Estimates of Genetic and Phenotypic totals for parities 2 and 3 (Table 5). Genetic Correlations Among Lifetime and phenotypic correlations of longevity traits Performance T r a b with partial and total lifetime traits increased Genetic and phenotypic correlations of par- from early partial lifetime to total lifetime. The tial lifetime with total lifetime yield and profit importance of increasing the number of lactatraits are presented in Table 3. Genetic correla- tions to increase total yield and profit was tions of partial with total lifetime traits were indicated by the large phenotypic correlations very high, consistent, and >.9 in most cases. (.97 and .95, respectively). High genetic correHigh genetic correlations can arise from lations of total lifetime yield and profit traits pleiotropy and also because early life yield is a and measures of longevity indicate that cows part of total lifetime yield (i.e., a part to whole with long herd life were also high for genetic relationship). The corresponding phenotypic merit of total lifetime performance traits. This correlations were relatively lower (.74 to .77) correlation is mostly due to number of lactafor the partial lifetime trait consisting of pari- tions and, in part, is a result of management ties 1 and 2, higher (.86 to .89) for the partial and culling for yield. Genetic correlations of partial and total lifelifetime trait consisting of parities 1, 2, and 3, and highest (.93 to .95) for the partial lifetime time performance traits with yield and profit

per day traits reported herein. Only one estimate (.31 f .lo) for profit per day of productive life was in the literature (8). These results show that heritability of yield per day of productive life seems to be higher than heritability of all other lifetime performance traits discussed herein. Possible explanations include the following. First, to the extent that yield is included in culling, larger differences among means of progeny groups tend to increase heritability estimates for a lifetime trait if lower yielding cows were culled and if mature survivors continued to yield with no increase in averaging calving interval. Second, lifetime traits are affected by herd life, which has low heritability, whereas per day of productive life traits are standardized for herd life.

Journal of Dairy Science Vol. 77, No. 1, 1994

GENETIC PARAMETERS OF LIFETIME TRAITS

309

TABLE 5. Phenotypic and genetic correlations of part and total lifetime yield and profit traits with longevity traits. ~

~~~

Phenotypic comlations

Genetic correlations

Trait'

LTDIM

PLIFE

NLACT

LTDIM

PLIFE

NLACT

MILK2 FAT2 PRT2 PRF2 MILK3 FAT3 PRT3 PRF3 MILK4 FAT4 PRT4 PRF4 LThULK LTFAT LTPRT LTPRF

.77 .77 .77 .74

.76 .76 .76 .73 .88 .87 .88

.70 .70 .70

.89 .83 .88 .82 .95 .9 1 .94 .90 .97 .95 .97 .94 .99 -97 .98 .98

39 .83 .88 .82 .95 .91 .94 .90 .97 .95 .97 .94 .98 .97 .98 .97

.85 .77 32 .77 .9 1 .87 .90 .86 .94 .92 .94 .91 .97 .95 .96 .95

.88

.88 .88

.85 .94 .94 .94 .91 .98 -98 .98

.66 .83 .83 .83 .79 .90

.84

.94 .93 .94 .91 .98 .91 .98 .95

.%

.90 .90 .86 .95 .95 .95 .92

'MILK2 = Two-parity milk, MILK3 = three-parity milk. MILK4 = four-parity milk, FAT2 = two-parity fat, FAT3 = --parity fat, FAT4 = f O U r - M t y fat, PRT2 = two-parity protein, PRT3 = thne-parity protein, PRT4 = four-parity protein, PRF2 = two-parity profit, PRF3 = three-parity profit, PW4 = four-parity profit, LTMILK = lifetime milk, LTFAT = lifetime fat, LTPRT = lifetime protein, LTPRF = lifetime profit, LTDM = lifetime days in milk, PLIFE = days of productive life, NLACT = number of lactations. SE 5.03 for genetic correlations only.

TABLE 6. Phenotypic and genetic correlations of partial and total lifetime production and profit with production and profit per day of productive life traits. Phenotypic correlations Trait'

MPDPL

FPDPL ~

MILK2 FAT2 PRT2 PRF2 MILK3 FAT3 PRT3 PRF3 MILK4 FAT4 PRT4 PRF4 LTMILK LTFAT LTPRT LTPRF

.70 .65 $67 .67 .73 .68 .7 1 .70 .73

.69 .71 .70 .71 .67 .69 .68

PPDPL ~

.62 .69 .63 .65 .65 .7 1

.66 .68 .65 .70 .67 .68 .63 .68

.64 .66

.7 1 .70 .72 .70 .74 .73 .75 .72 .74 .73 .75 .72 .7 1 .70 .72 .70

Genetic correlations PFPDPL

~

hPDPL

FPDPL

PPDPL

PFPDPL

.94 .80

.64

.so

.86 .71 .78 .63 .83

.78 .88 .81 .81 .80

.69

.88

.77 .61 .80 .67 .75 .58 .75 .63 .71

.83

.67 .69 .69 .68 .60 .62 .62 .62 .54 .57 .56 .57 .47 .48 .48 .49

~~

A7 .47 .47 .43 .48 .48 .48

.90 .87 .95 .84

.91 .90

.44 .46 .46 .46

.85 .91

.42 .42 .42 .42 .39

.92 .91 .83 .88 .90

.94

.81

.SO .87 .84

.78 .78 .84

32

'MILK2 = Two-pdty milk. MILK3 = thne-parity milk. MILK4 = four-parity milk. FAT2 = two-parity fat, FAT3 = three-paritY fat. FAT4 = fOlU-@ty PRT2 = tweparity prottin, PRT3 = thnc-@ty protein. PRT4 = fow-parity Pmteh = tWo-pdty profit, PRFJ = thneparity profit, P W 4 = four-parity profit, LTMILK = lifetime milk, LTFAT = lifetime fat, LTPRT = lifetime protein. LTFW = lifetime profit, MPDPL = milk per day of productive life, FPDPL = fat per day of productive life. PPDPL = protein per day of productive life, and PFPDPL = profit per day of productive life. SE S -07 (only for genetic cornlation).

Journal of Dairy Science Vol. 77, No. 1, 1994

310

JAIRATH ET AL.

per day of productive life traits (Table 6) were higher (.47 to .95) than the corresponding phenotypic correlations (.39 to .75), which is consistent with results in the literature (8, 13, 26). Gill and Allaire (8) reported higher correlations of total performance traits with profit per day than with milk yield per day of productive life, in contrast to the results of the present study. In general, the patterns of genetic and phenotypic correlations of yield and profit per day of productive life traits with each partial and total lifetime performance trait were similar; genetic correlations of profit per day traits, however, decreased slightly from partial to total lifetime performance (Table 6). Genetic correlations of longevity trait with yield and profit per day of productive life traits (Table 4) ranged from .46 to .85 and were highest for milk per day of productive life (.82 to .85), followed by protein per day of productive life (.70 to .74), fat per day of productive life (.54 to .58), and profit per day of productive life (.46 to 50). The corresponding phenotypic correlations ranged from .44 to .64. Phenotypic correlation of protein per day of productive life exceeded slightly that of milk per day of productive life. Correlations of individual per day traits with the three measures of longevity were almost similar. However, Gill and Allaire (8) r e p o d estimates of genetic correlations of length of productive life and number of lactations completed with milk per day (.40and .42) and profit per day (.64 and .65) that were slightly Iower than the estimates of corresponding phenotypic correlations (.48 and .48) and (.69 and .70). These correlations were in contrast to the correlations reported in the present study. Two explanations are plausible. First, the range of standard errors reported in the study of Gill and All& (8) was very large (.01 to 40);although estimates of genetic correlations are usually subject to rather large sampling errors (6, 24), the estimates obtained in the present study were more precise. Second, the data in Gill and Allaire (8) were limited to only eight herds and included only records for those cows that had at least two calvings and at least one completed lactation; terminal lactation records were excluded from the data. Also, lifetime profit function was defined differently. Among per day of productive life traits, milk per day of productive life was most Journal of Dairy Science Vol. 77. No. 1, 1994

highly related, genetically, but profit per day of productive life was least related, genetically and phenotypically, to all partial and total lifetime performance traits, including traits of longevity (Tables 4 and 6). The estimates of genetic and phenotypic parameters of lifetime performance traits in the present study were mostly free from the effects of selection on milk yield. The REML analysis used for these estimates accounts for selection if the information on which the selection was practiced is included in the analysis. Inclusion of the information was facilitated by accommodation of phenotypic and genetic correlations for partial lifetime and lifetime performance traits in the multitrait analysis. CONCLUSIONS

For any genetic improvement program of livestock to be meaningful, the breeding objective must be decided before selection criteria are chosen. For dairy cattle, the breeding objective should reflect profit over lifetime of the cow, and total profit and average profit per day have both been used to reflect the profitability of the cow. Meaningful arguments to select one of these criteria were lacking until Van Arendonk (31), using replacement theory, showed that profit per day does not reflect profitability of cows within a breed correctly unless no variation occurs for herd life and that average profit per day can be used to compare profitability of cows that differ for length of productive life only in situations of identical replacement. The low estimates of heritability for total lifetime performance traits suggest that direct selection holds little promise for enhancing lifetime performance of cows because response to selection will be very slow. Another practical problem is the large number of daughters (100 to 200 daughters per sire) needed to attain reasonable reliability of genetic evaluations for traits with such low heritabilities. Many bulls do not have a large number of daughters and farmers are not likely to base selection of bulls on genetic evaluations with such low reliability. Selection on the basis of per day of productive life (the traits with high heritabilities) might favor daughter groups with very high yields and short herd life. Selection on lifetime performance is not practical because

GENETICS-P

delaying of selection until information is available on lifetime traits greatly increases the generation interval (with resulting decrease in genetic gain) and the costs of maintaining potential replacement stock. Alternatively, selection for lifetime performance trait could be achieved through selection for one or more of the associated traits. Selection on early performance traits seems more desirable for changing lifetime performance. Given the high genetic correlations of early lifetime traits with all lifetime performance traits including longevity, selection for early lifetime traits would indirectly improve lifetime performance and hence lifetime profitability. Although selection on first lactation traits improves all measures of lifetime performance traits, as information becomes available on lifetime performance of halfsisters and close ancestors, it should be incorporated in choosing young sires for enhancing lifetime performance. However, very little opportunity for selection among cows exists for lifetime performance traits because most of the selection practiced is based on the performance in the latest or some earlier lactation and not for any of the traits expressed later in life. In this sense, most of the selection pressure that can be applied among cows is automatic; those dams remaining longest in the herd leave the largest number of offspring. ACKNOWLEDGMENTS

We thank K. Meyer for the use of her computer programs and Quebec Dairy Herd Analysis Service for providing data. We also acknowledge the work of Susan Joyal, S. des Marchais, H. G. Monardes, and R. K. Moore in preparing the data file. We acknowledge financial assistance from Le Minist&rede 1’Enseignment Sup6rieur et de la Science and Consei1 des Recherches en Pkhe et en Agroalimentaire du Qu6be.c and the Canadian Association of Animal Breeders. REFERENCES

1 Banks, B. D.. I. L. Mao, and J. P. Walter. 1985. Robustness of the restricted maximum likelihood estimator derived under normality as applied to data with skewed distributions. J. Dairy Sci. 68:1785. 2 Bumside, E. B., A. E. Mcclintodr. and K. Hammond. 1984. Typc, production and longevity in dairy cattle: a

OF LIFETIME TRAITS

311

review. Anim. Brced. Abstr. 52171. 3Cue. R. I., H. G. Monardes, and J. F. Hayes. 1987. Comlations between production traits in first lactation Holstein cows. J. Dairy Sci. 702132. 4 Ducrocq, V., R. L.Quaas, E. J. Pollak, and G. Casella. 1988. Length of productive life of dairy cows. 2. Variance component estimation and sire evaluation. J. Dairy Sci. 71:3071. 5 Evans, D. L., C. Branton, and B. R. Farthing. 1964. Heritability estimates and interrelationships among production per day of productive life, longevity, breeding efficiency and type in a herd of Holstein cows.J. Dairy Sci. 47(Suppl. 1):699(Abstr.). 6Falconer, D. S. 1989. Introduction to Quantitative Genefics. Longman Scientilic and Technical,London,

Jw

7 Gaalaas, R. F., and R. D. Plowman. 1963. Relationship M e e n longevity and production in HolsteinFriesian cattle. J. Dairy Sci. 46:U. EGill, G. S.. and F. R. Allaire. 1975. Relationship of first ladation performance to lifetime production and economic efficiency. J. Dairy Sci. 591319. 9Gilmore, J. A. 1977. The relationship of milk yield and other traits measured early in life to a dairy cattle profitability model including health and opportunity costs. Ph.D. Diss.. no^ Carolina State Univ., Raleigh. 10Gravext, H. 0. 1985. Genetic factors controlling feed efficiency in dairy cows. Livest. Prod. Sci. 13:87. 11 Hargrove, G.L.,J. J. Salazar, and J. E. Legates. 1969. Relationship among first lactation and lifetime measmments in a dairy population. J. Dairy Sci. 52:651. 12 Harville, D.A. 1977. Maximum likelihood approaches to variance component estimation and to related problems. J. Am. Stat. Assoc. 72320. 13 Hoque. M.,and J. Hodges. 1980. Genetic and phenotypic parameters of lifetim production traits in Holstein cows. J. Dairy Sci. 63:1900. 14 Lamon, C. J., A. B. Chapman, and L. E. Casida. 1951. Butterfat production per day of life as a criterion of stleuion in dairy cattle. J. Dairy Sci. 34:1163. 15Lin, C. Y.. and F. R. A u k . 1978. Efficiency of selection on milk yield to a fixed age. J. Dairy Sci. 61: 489. 1 6 h , C. Y..and A. J. Lee. 1985. Multitrait estimation of relationship of first lactation yield to body weight changes in Holstein heiim. J. Dairy Sci. 68:2954. 17Meyer, K. 1985. Maximum likelihood estimation of variance components for a multi-variate mixed model with equal design matrices. Biometrics 41:153. 18 Miller, P. D.,L.D.Van Vleck, and C. R. Henderson. 1967. Relationships among herd life, milk production. and calving interval. J. Dairy Sci. 50:1283. 19Moore. R. B. 1991. Parameter estimates between production and management traits in first lactation using milk recording data. Ph.D. Diss., Univ. Guelph. Guelph, ON, Can. 20Nomran, H. D.,B. G. Cassell, R. E. Pearson, and G. R. Wiggans. 1981. Relation of first lactation production and conformation to lifetime performance and profitability. J. Dairy Sci. 64:104. 21 parktr, J. B., N. D. Bayley. M. H. Fohrman, and R. D. Plowman. 1960. Fectors influencing dairy cattle longevity. J. Dairy Sci. 43:401.

Journal of Dairy Science Vol. 77, No. 1. 1994

312

JAIRATH ET AL.

22 Pearson, R. E., and R. H. Miller. 1981. Economic definition of total performance, b e g gcals, and breeding values for dauy d e . J. Dairy Sci. 44357. 23 Plowman, R. D., and R. F.Gaalaas. 1960. Heritability estimates of longevity in Holstein-Friesian cattle. J. Dairy Sci. 43:401. 24 Reeve, E.C.R. 1955. The variance of genetic comlation coefficient. Biometrics 11:357. 25Robertson. A. 1977. The effect of selection on the estimation of genetic m t e r s . Z. Tien. Zuechtungsbid. 94:131. 26 Schneeberger, M..and Casanova, L. 1990. Relationship between longevity and milk production of Swiss Braunvieh d e . Page 371 in Roc. 8th C o d Aust. Assoc. Anim. Bred. Genet. Massey Univ. h s . Palmerston North, NZ. 27 Schultz, L. H. 1968. Ketosis in dairy d e . J. Dairy Sci. 51:1133. 28 Searle, S. R. 1982. Matrix Algebra Useful for Statistics. John Wiley & Sons, New York, NY. 29Smitb, S. P., aad R. L. Quaas. 1984. Productive lifespan of bull progeny groups: failure time analysis. J. Dairy Sci. 672999. 30 Solkner, J. 1989. Genetic relationships between level of production in diffcrcnt ladations. rate of maturity and longevity in a dual purpose cattle. Livest. prod. Sci. 23:33.

Journal of Dairy Science Vol. 77, No. 1, 1994

31 Van Arendonk, J.A.M.1989. Use of profit equations to determine reMve economic value of dairy cattle herd life. J. Dairy Sci. 741101. 32Van Tassell, C. P., and L. D. Van Vleck. 1991. Estimates of genetic selection differentials and generation intervals for four paths of selection. J. Dairy Sci. 74:1078. 33Van Vkck, L. D. 1944. F i t lactation performance and herd life. J. Dairy Sci. 47100. 34 Van Vleck, L. D. 1977. Theoretical and actual genetic progrtss in dairy d e . Page 543 in Roc.Int. Conf. Quant. Genet. Iowa State Univ. Press, Ames. 35 Van Vleck, L. D. 1985. Including records of daughters of selected bulls in estimation of sire components of variance. J. Dairy Sci. 68:23%. 36Van Vleck, L. D. 1990. Observations on selection advances in dairy cattle. Page 433 in Proc. 2nd Int. C o d Quaat. Genet. S i a u r Assoc., Inc. Sunderland, MA. 37 White, J. M.,and J. R. Nicholas. 1965. Relationships bctween first lactation, later performance, and length of herd life in Holstein-Friesian cattle. J. Dairy Sci.

48:468. 38 Wilcox, C. J., K. 0. Pfau, and J. W. Bartlett. 1957. An investigation of the inheritance of female reproductive performanct and longevity, and their relationship within a Holstein-Friesian herd. J. Dairy Sci. 40:942.