| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
1 Cooperative Resources International, Shawano, WI 54166
2 Iowa State University, Ames 50011
3 National Animal Disease Center-USDA-ARS, Ames, IA 50010
4 University of Wisconsin, River Falls 54022
5 Michigan State University, East Lansing 48824
Corresponding author: Gamal Abdel-Azim; e-mail:gamal{at}crinet.com.
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSACKNOWLEDGEMENTSREFERENCES |
|---|
Data were analyzed using a threshold model that included herd, calving year, season of calving, and parity as cross-classified fixed factors; and sire and cow within sires as random effects. Days at risk and days in milk at the beginning of a record were included by fitting the days as continuous covariates in the model. A heritability value of 0.202 ± 0.083 was estimated for generalized immunity. Heritability values of 0.141 and 0.161 were estimated for uterine infection and mastitis, respectively. Heritability of single noninfectious disorders ranged from 0.087 to 0.349. The amount of additive genetic variance recovered in the underlying scale of noninfectious disorders tended to zero when combining multiple conditions. The study supports combining infectious diseases into categories of interest but we do not recommend the same approach for noninfectious disorders.
Key Words: disease resistance • generalized immunity • genetic evaluation
Abbreviation key: BDIM = days in milk at the beginning of a record, CCL = cystic corpora lutea, COD = cystic ovarian disease, DAR = days at risk, GI = generalized immunity, ID = infectious diseases, IR = infectious reproductive conditions, MF = milk fever, TM = threshold model, UH = udder health, UI = uterine infection
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSACKNOWLEDGEMENTSREFERENCES |
|---|
Numerous studies have investigated the genetic basis of disease resistance in US Holsteins, emphasizing the genetic potential of improving animal health (e.g., Lin et al., 1989; Lyons et al., 1991; Uribe et al., 1995). This potential has been demonstrated in Scandinavian breeds where breeders have selected for several health traits. Significant genetic change for clinical mastitis in Norwegian cattle has also been found (Heringstad et al., 2003).
In addition, advances on the biostatistical side introduced new techniques for the analysis of binary and ordered categorical traits into which many health measurements could be classified. Better genetic parameter estimation is expected when a threshold model (TM) that assumes an unobservable underlying and heritable scale is postulated (Gianola and Foulley, 1983; Harville and Mee, 1984). In particular, genetic variation estimated from a TM is usually higher than genetic variation estimated by fitting a linear model to the observed categories (Abdel-Azim and Berger, 1999; Hansen et al., 2002). Further, the TM allows estimating probabilities of observing individuals between any 2 boundary points, given certain model effects. This leads to estimating the probability of disease occurrence for daughters of sires included in the analysis. This facility, available in a TM analysis, allows for a ranking criterion of sires that is more interpretable than mixed model solutions.
Diseases can be classified into infectious and noninfectious. Infectious diseases are thought to be controlled immunologically, and the level of immunity varies across animals according to many factors, including genetics. In the current study, resistance specific to a single disease as well as generalized immunity for multiple related diseases are examined. Selected noninfectious disorders, those that are not caused by pathogens, are also studied. Risk factors for liability to studied diseases are identified and their magnitudes and directions are tested statistically. Emphasis is given to the genetic basis of a large category of infectious diseases.
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSACKNOWLEDGEMENTSREFERENCES |
|---|
. Herds were required to be on DHIA and herdsmen agreed to perform complete health data recoding. Data recording was in 2 forms, electronically or by describing health conditions and treatments on forms provided to herdsmen. Data were collected periodically by personnel from Genex Cooperative Inc., and were entered into a database system at Iowa State University. A 3-digit numeric coding system was used to define health conditions. Data included 24,705 lactations of 14,473 Holstein cows. Cows were sired by 1,303 US Holstein sires. Health information included diagnosis of condition, treatment cost and level of treatment, and sometimes severity of condition, such as in clinical mastitis.
|
All cows in a herd were started at the same time. This introduced an additional source of variation because of the variable lactation stages of cows at their herd start date. The difference in days between a cow’s starting date and its most recent calving date was taken as days in milk for a lactation at the start of the data collection period (BDIM), and was used in the model of analysis as a covariate (de Haas et al., 2002). If the preceding calving of a cow occurred more than 450 d before the herd start date or if calving occurred at or after the herd start date, BDIM was assigned a value of 0. Eighteen percent of lactations had started before their corresponding herd start dates and average BDIM in data was 31 d. The other 82% were for subsequent lactations of cows that existed at the herd start date or for cows that entered the herd (as replacements) later in the study.
A binary response variable was created for a disease or a group of diseases of interest. If a cow contracted the disease at least once over a lactation, a response of 1 was attributed to the record. For a cow that calved during the monitoring period of a herd and that was never recorded for the disease or group of diseases of interest, a healthy record with a response of 0 was created. Table 2 lists 8 conditions of interest along with characteristics of their corresponding data sets. Information about recurrence of a condition was part of the data. Recurrence was not considered within a lactation but was considered across lactations of a cow, i.e., a second record for the cow was created if it contracted the same disease at least once in a subsequent lactation.
|
differed in size because a minimum of 2 cases per herd was applied, i.e., to include a herd in the analysis, 2 disease cases must appear in the herd. Therefore, herds with no disease conditions were excluded from the analysis, which avoided the extreme category problem (Harville and Mee, 1984). No similar restrictions were imposed on sires.
Diseases of Interest
Infectious diseases.
First, all available infectious conditions were used to identify animals that were superior in their generalized immunity (GI). Generalized immunity (including innate immunity) may be thought of as a combination of immune responses to a variety of immune system challenges to identify animals that are better able to respond to a variety of challenges. Selection for GI may genetically improve the overall health and performance of animals continuously exposed to unknown, new or old, infectious challenges. It may also be helpful in cases where selection for a single mode of resistance (i.e., single disease) adversely affects resistance to other diseases (Bishop, 2003).
At the immunological level, the distinction between selecting for resistance to a specific disease and generalized immunity may not be clear. For instance, allelic variants that positively affect immunity to many pathogens may be part of the innate immune system, hence their frequencies increase when selecting for a specific or generalized immunity. Mallard et al. (1998) (in pigs), and Pinard et al. (1992) and Sarker et al. (1999) (in chickens), demonstrated the feasibility of selecting for GI for several generations. In the current study, genetic parameters for resistance to several diseases (immunologically controlled) are estimated for the purpose of genetic improvement of GI. This is achieved using data on diseases that are immunologically controlled. In addition to pigs and chicken, Kelm (1998) and Kelm et al. (2001) quantified general immune responsiveness of dairy sires. Up to 14 immune response traits were measured and genetic parameters of the traits were estimated.
Infectious conditions considered included mastitis and udder infections (3801 conditions); reproductive, respiratory, navel, and kidney infections (2139 conditions); and skin and foot infections (257 conditions). These conditions are caused by pathogens and, hence, resistance is thought to be immunologically controlled. Lactational incidents, not multiple disease conditions over a lactation, constituted the response variable used in the analysis.
Other infectious diseases.
Liability to single infectious diseases was also studied. Mastitis [referred to as udder health (UH)] as a major infectious disease in the dairy cow was modeled. Mastitis conditions were reported according to infected quarters. In the current study, a cow that expressed symptoms of udder infection, regardless of the infected quarter, was assigned a response of 1.
In addition to mastitis, 3 reproductive infectious conditions were combined into one category and modeled as a single disease (IR). The category included uterine infection (661 conditions), retained placenta (322 conditions), and abortions caused by infection (325 conditions). Infection may not be the only cause of retained placenta but infection is a frequent consequence of retained placenta. The modeling of only uterine infection as a single reproductive condition was also studied and was compared with the IR category.
Noninfectious diseases.
Noninfectious diseases are hard to aggregate and analyze together because of the different mechanisms controlling each disease. In the current study, displaced abomasum and milk fever were analyzed as major noninfectious conditions. In addition, a category of ovarian cysts was created by combining 3 major cystic disorders of the ovary: follicular cysts (296 conditions), luteinized follicular cysts (609 conditions), and cystic corpora lutea (CCL) (404 conditions). The category is usually referred to in the literature as cystic ovarian disease (COD) and is frequently thought of as one disease, clinically defined as the presence of an anovulatory follicle-like structure greater than 2.5 cm for 10 d or more in the absence of a corpus luteum (Morrow, 1980; Garverick, 1997). To compare genetic parameters estimated for the COD category, with its 3 types of cysts, to estimates obtained by analyzing a single type of cysts, CCL was analyzed separately.
Measures of Disease Frequencies
Two main epidemiological measures were used to determine the frequency of occurrence of each disease, prevalence and incidence rate. Prevalence was expressed as the number of existing cases of a disease divided by the population size. Only conditions recorded in 1998 were used to compute the 1-yr prevalence of each disease (Rothman and Greenland, 1998). The use of conditions of all 4 yr to evaluate 4-yr prevalence would have introduced inaccuracies because not all herds were being recorded over the longer period. For the infectious disease (ID) category, there were 1067 infected lactations and 7431 total lactation records available in 1998 to compute period prevalence.
A second measure for disease frequency that incorporates data over the whole trial period is incidence rate (Rothman and Greenland, 1998; de Haas et al., 2002). Incidence rate is expressed as the total number of affected lactations divided by the sum of the total number of days on trial for all cows, i.e., sum of DAR. For the analysis of ID, there were 6,425,136 DAR. Incidence rate, as an epidemiological measure, is very flexible in allowing study subjects to enter the study at different times and it makes handling of culls or deaths easy, for these simply contribute fewer days to the denominator (Bhopal, 2002). It must be realized, however, that incidence rate is a crude measure that assumes that the disease occurs evenly over time and across subjects.
Statistical Analysis
Models.
Health data were analyzed with a threshold model (Harville and Mee, 1984). A linear mixed model was fit to an underlying continuous response variable that is both genetic and environmental in origin (Falconer, 1989). The model included herd, year of calving, season of calving, and parity as cross-classified fixed factors; and sire and cow within sire as nested random effects. Days at risk and BDIM were included by fitting the days as linear covariates in the model. Sire was treated as a random factor with a covariance structure proportional to the relationship matrix among sires. Relationships were traced through sires of sires and maternal grand sires of sires. Cow was also included as a random factor in the model to account for permanent cow effects. The linear model can then be expressed as
 |
where a, b, c, d, f, and g are mapping functions (Harville and Mee, 1984) identifying the herd, year, season, parity, sire, and cow that correspond to the ith lactation in data (i = 1 to N and N is the total number of records in a data set of Table 2); 
and 
are regression coefficients to be estimated; and DARi, BDIMi, and ei are days at risk, BDIM, and a residual component corresponding to the ith lactation, respectively. Six levels of season were considered that were functions of the month of calving so that seasonc(i) spans the 2 consecutive months, 2*c(i) – 1 and 2*c(i), where c(i) 
{1, 6}. Five levels of parity were considered, where the fifth level is cows in their fifth and greater lactations. Year effect was composed of 4 levels, 1996 to 1999, and levels of herd, sire, and cow factors are listed in Table 2. Herd, year, and season were fit as separate factors as opposed to a herd-year-season factor to avoid the extreme-category problem resulting from very few records in a herd-year-season subclass.
Higher order polynomials for DAR and BDIM were tested along with other factors in the model. Model selection was performed in S-PLUS by fitting generalized linear fixed models of the binomial family (Venables and Ripley, 1998).
Let y, [Y1, ..., YN]' denote the underlying continuous response variable, w, [W1, ...., WN]' denote the observable dichotomous response variable, then y could be expressed in matrix form as
 | ([1]) |
where X, [X1 X2], and Z, [Z1 Z2], are known matrices; b, [b1 ' ]', is a vector of unknown fixed effects; a, [u v]', is a vector of unknown random effects of sires (u) and cows (v); and e is a vector of residuals. Further, u ~ MVN(0, 
Du) v ~ MVN(0, Dv) and e ~ MVN(0, I), where Du = 
A and Dv = 
I, with A representing the relationship matrix among sires, = 
/ and = /. Finally, y ~ MVN(Xb, V) with V = I + ZDZ ', where var(a) = D which is a block diagonal matrix with two blocks, Du and Dv. The relationship between Yi and Wi is taken to be
 | ([2]) |
In the analysis of our data, a standardized version of the TM equations was solved and estimates for fixed effects were obtained.
Sire solutions were obtained and represented sire transmitting abilities. Estimate of the parameters and were obtained and h2 was estimated as 4/
p2 and c2, proportion of the cow variance to the total variance, was estimated as /p2, where p2 = 1 + + . Estimates of and along with their asymptotic variance covariance matrix were obtained by Matvec (Wang et al., 2003). Average information-REML, which is implemented in Matvec, with a probit link function was used in the analysis except for ID and UH where a Logit link function was used. With low incidence in data as encountered with single conditions, a probit link function was slower but more robust than a logit link function that did not always converge.
Standard errors of h2 were estimated as 
, where Var(h2) was approximated by, h4{Var()/ 2 + Var(p2)/p4 – 2[Var() + Cov(, )]/( + 2 + )}. Similarly, the standard error of c2 was approximated as the square root of c4{Var()/2 + Var(p2)/p4 – 2[Var() + Cov(,)]/ ( + 2 + )}. Variance of ratios can be found in Mood et al. (1974).
Hypothesis testing for important contrasts was performed on the underlying scale, model [1]. Seasonal and parity differences were tested for the health conditions studied to determine their magnitudes and directions as risk factors. Linear contrasts were tested in Matvec (Wang et al., 2003).
|
RESULTS AND DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSACKNOWLEDGEMENTSREFERENCES |
|---|
.
|
, the h2 estimate of ID was 0.202 ± 0.083. An estimate of the parameter c2 was also obtained, 0.083 ± 0.0527. The moderate heritability value for ID indicates the potential of selection to genetically improve generalized immunity. Compared with heritabilities of GI estimated from immune response traits, the current value is not as high. Pinard et al. (1992) obtained a heritability estimate of 0.31, and Sarker et al. (1999) obtained heritability estimates of 0.61 and 0.52 in selected lines for serum IgM and IgG levels, respectively.
Days at risk showed a positive relationship with occurrence of the diseases. Table 4 lists regression coefficients for both DAR and BDIM for all disease traits. An estimate of 0.0021 for was found and was significantly different from 0 (0.0021 ± 0.0003, P <0.01). Figure 1A shows the effect of adjusting for DAR. Two fixed effect models with and without DAR were fit to the same data. When the DAR covariate was removed, disease occurrence was overestimated. Overestimation was more severe in lactations with fewer than 150 DAR. A slight underestimation was also noticed very late in lactation or with lactations that continued for over 350 d (Figure 1A).
|
|
, which shows that removing both year and DAR from the model further introduced more bias compared with Figure 1A, where year is included.
The regression coefficient of BDIM () was also significantly different from 0 (–0.0055 ± 0.0007, P < 0.01) but negative, which indicated more disease occurrences early in lactation, or equivalently reduced immunity with less BDIM.
Season solutions showed a positive difference (0.461 ± 0.082, P < 0.01) between season 1 and season 4. Solution for season 1 was 0.31 and for season 4 was –0.15. This difference indicates less disease occurrence or equivalently better GI for cows that calved in summer than in winter.
Disease occurrence across different parities did not seem to differ greatly. Lactation number was, however, a significant factor in the model and the solutions indicated that older cows (those in their fifth or greater lactation) had the poorest GI compared with cows in their first to fourth lactation. Average solution for lactations 1 to 4 was –0.028 vs. 0.12 as the solution for lactation 5. The null hypothesis Ho: 
for the difference between parity(5) and the average of all other parities was rejected (–0.146 ± 0.058, P <0.05) confirming the reduced GI of older cows in data.
Udder Health.
A moderate heritability value was estimated for mastitis, 0.161 ± 0.078. The model did not include the cow factor because it hindered convergence. The model included an extra covariate for DAR2 as the highest order significant polynomial for days at risk. Higher order polynomials for BDIM were not significant and were not included. The h2 value estimated from data supports estimates reported in Uribe et al. (1995) and further assures the potential of genetic improvement of resistance to mastitis.
A positive regression coefficient, 0.0053 ± 0.00122, was estimated for DAR and was significantly different from 0 (P < 0.01). The positive value indicates that disease frequency increases with DAR. A negative coefficient was estimated for BDIM (–0.0065 ± 0.00089). The value is larger in magnitude than the corresponding value estimated for ID indicating more importance to the stage of lactation when modeling udder health. The negative coefficient implies more mastitis in early lactation stages.
Season solutions showed increased udder health problems in winter than in summer, which is in agreement with other studies (e.g., Lescourret et al., 1995). Season 4 was the best and season 1 was the worst with a difference of 0.484 ± 0.099 (P < 0.01). Lactation solutions showed a trend toward more udder health problems with older cows but with no significant difference between lactations 1 and 4. Solutions for lactations 1 to 
5 were –0.24, –0.13, 0.0, 0.03, and 0.24, respectively.
Infectious reproductive conditions.
A heritability of 0.057 ± 0.0285 and c2 of 0 were estimated for reproductive conditions immunologically controlled (infectious). This included uterine infection, abortion, and retained placenta. The analysis of only uterine infection (UI) as a single reproductive disorder resulted in h2 estimate of 0.141 ± 0.0734 and c2 estimate of 0. A positive regression coefficient on DAR was estimated for the UI data. The effect itself was important for modeling uterine infection ( = 0.0011 ± 0.00036, P < 0.01). Season effect was also important for modeling UI with more disease in cows that calved in winter than in summer.A contrast of season(4) – season(1) showed a significant difference (P < 0.01) of –0.382 ± 0.0846. Levels of parity did not differ significantly from each other indicating similar risks for contracting the disease across cows in different parities.
Disease Resistance of Immunity-Unrelated Disorders
In the following, results of selected single diseases and a class of related ovary disorders are presented. Incidence rate and prevalence of immunity-unrelated disorders are generally less than infectious diseases whose resistance is immunologically controlled (Table 3). This resulted in smaller data sets and consequently larger standard errors of estimated genetic parameters.
Displaced abomasum.
Cow variance converged to 0. Only 5 cows had the condition twice making repeatability difficult to estimate. The h2 estimate, 0.087 ± 0.041, was similar to that of Lyons et al. (1991). Regression coefficient on DAR, , was positive but not significantly different from 0. Regression coefficient on BDIM, , on the other hand was significantly lower than 0 (P < 0.05) indicating more occurrences in the early stages of lactation.
Parity was an important risk factor for displaced abomasum; a positive association of parity with the occurrence of this disease was found, and solutions were 0.0, 0.10, 0.15, 0.28, and 0.23 for parities 1 to 5. The difference, parity(5) – parity(1), was significantly different from 0 (0.234 ± 0.073, P < 0.05). Season solutions showed more occurrences of displaced abomasums in cows that calved in winter and spring than those calved in summer and fall.
Cystic ovarian disease.
When the 3 major categories of ovarian cysts, cystic corpora lutea, follicular cysts, and luteinized follicular cysts were analyzed as cystic ovarian disease (COD), the heritability estimated was near 0 (0.030 ± 0.0470). A higher estimate was obtained when the single condition, CCL, was modeled alone, resulting in a heritability estimate of 0.110 ± 0.062 [note that the presence of CCL may not be considered a pathological condition if the CCL produce progesterone (Morrow, 1980)]. Results show that combining the 3 cystic disorders of the ovary, despite being closely related, resulted in almost no additive genetic variance (u2 = 0.007 for cystic ovarian disease vs. 0.028 for CCL, Table 3).
Days at risk had a positive, unfavorable, relationship with incidences of COD ( = 0.0037 ± 0.00038, P <0.01). Days in milk, on the other hand, had a negative relationship with incidences of COD ( = –0.0019 ± 0.0008, P < 0.05) indicating more COD occurrences in early stages of lactation. Season of calving was identified as a significant risk factor. Cows that calved in winter were more likely to develop COD than those that calved in summer and fall. The COD data showed a difference of –0.259 ± 0.0597, (P < 0.01) for (Season3+Season4)/2 – (Season1+Season6)/2. However, there is no consensus in the literature about calving season as a risk factor for COD (Garverick, 1997). Finally, there was a positive, unfavorable association between high parity and the development of COD but there was no significant difference between the first 2 lactations and older lactations.
Milk fever.
The heritability estimate of milk fever as a single metabolic disorder was 0.349 ± 0.179. The value is much higher than the 0.09 estimated by Uribe et al. (1995) but smaller than the 0.40 estimated by Lyons et al. (1991) and similar to the 0.30 estimated by Lin et al. (1989). There was no evidence that season of calving was a significant risk factor but lactation number was a risk factor indicating higher incidences of milk fever in cows in their fourth or higher parities compared with cows in their earlier parities. The null hypothesis Ho:
was rejected (–0.871 ± 0.117, P < 0.01). Parity is largely accepted in literature as a major risk factor for milk fever (e.g., Houe et al., 2001; Ostergaard et al., 2003) acting in the same direction as found in the current study.
Sufficiency of Health Data to Estimate Cow Variance
Cow was accounted for in the model of analysis, [1], as a random effect to adjust for permanent cow effects associated with repeated health records on the same cow. Cow factor included nonadditive genetic, three-quarters of the additive genetic, and other permanent nongenetic effects. Table 5 shows that health data did not always contain sufficient information to obtain reasonable estimates of cow variance. The parameter was either imprecisely estimated as for the ID category, or an estimate was not possible or converged to 0 in other cases. It can also be seen from Table 5 that recurrence of another affected lactation on the same cow was rare (columns ">1" and ">2"). Rarity of repeated affected lactations as a possible reason for 0 estimates of the parameter was verified by excluding extreme cows with multiple affected lactations and repeating the analysis for the ID category, i.e., after excluding records of 287 cows with 2 or more affected lactations (Table 5). The latter analysis resulted in a 0 estimate of Further, repeating the analysis without cows with 3 or more affected lactations yielded a positive, but insignificant, estimate of , 0.0313 ± 0.1926.
|
as opposed to the interest in that is entirely additive genetic.
Correlations Among Predicted Sire Transmitting Abilities for Disease Traits
Random sire effects of model [1] can be viewed as sire transmitting abilities. Correlation coefficients among predicted transmitting abilities of sires with 10 or more daughters for the 8 disease traits were calculated and the coefficients are listed in Table 6.
|
contributed to the correlations. |
CONCLUSIONS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSACKNOWLEDGEMENTSREFERENCES |
|---|
The amount of additive genetic variance recovered in the underlying scale of noninfectious disorders tended to 0 when combining multiple conditions. Our study supports combining infectious diseases into categories of interest but does not support the same approach for noninfectious disorders.
Season of calving and parity are important environmental risk factors for disease liability. Statistical analyses showed that cold months increased the risk of contracting diseases and that resistance deteriorated in older cows. Our study also showed the importance of accounting for days at risk and days in milk at the beginning of a record when analyzing health records.
|
ACKNOWLEDGEMENTS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSACKNOWLEDGEMENTSREFERENCES |
|---|
Received for publication May 25, 2004. Accepted for publication October 27, 2004.
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSACKNOWLEDGEMENTSREFERENCES |
|---|
This article has been cited by other articles:
|
|
 |
|
  J. Arango, I. Misztal, S. Tsuruta, M. Culbertson, and W. Herring Study of codes of disposal at different parities of Large White sows using a linear censored model J Anim Sci, September 1, 2005; 83(9): 2052 - 2057. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
