Study Population
The cooperator herd consisted of approximately 3400 lactating and dry cows located on a commercial dairy in Texas. Cows were milked 3x/d in a double-50 parallel parlor, with a rolling herd average >11,000 kg of milk. Because of expansion of the herd, just over one-half of the cows had been purchased from outside sources. Approximately 69% of the cows were in their first lactation, 16% were in their second lactation, and 15% were in their third or greater lactation, with an annual culling rate of 35 to 40%. Herd production and health events were recorded and tracked using a commercial on-farm computer software program (DairyComp 305; Valley Agricultural Software, Tulare, CA). All cows were housed in areas with access to shade; however, soakers and fans were not used. Although the farm had implemented a program to minimize MAP infection by managing colostrum based on ELISA test results from dams, the owner indicated that cows were not culled based on test results.
Sampling Scheme
Eligible cows were identified through the farm’s computerized records, excluding only those from the sampling frame that were both not pregnant and late in their lactation, which increased the likelihood of culling early in the study. Cows to be randomly sampled were stratified on the basis of DIM and lactation number; then, selection was weighted toward detecting those at greatest risk for identification of JD with 50% of those selected in their third or greater lactation, 35% in their second lactation, and 15% in their first lactation. Based on the historically observed differences between the proportion testing positive using the ELISA during summer months (approximately 8%) vs. winter months (approximately 13%), with 95% confidence and 80% power, 463 cows were required (one-sided test). Because we expected some attrition over a full year, resulting from normal herd culling procedures, we sampled more than were needed. Therefore, cows were randomly allocated into 3 separate monthly sampling groups of approximately 220 head each, which were resampled in staggered periods separated by 3-mo intervals, resulting in 4 sample collection periods per group. In the end, a total of 539 cows actually had repeated samples collected during the study period. Numbers of cows sampled initially (by cohort) and numbers lost to culling or death are presented in Table 1
.
Sampling Procedure
At the start of the study (month 1, October 2002), approximately 10 mL of blood were collected from each cow in the first group of cows. The sera from these blood samples were evaluated for antibodies to JD using a commercial ELISA test. For this first sampling period only, the 20% of cows with the highest S/P ratio, as reported by the laboratory, were then resampled shortly after the initial blood collection. At least 25 g of feces were collected from each cow and submitted to the laboratory for subsequent fecal culture. Fecal samples were collected directly from the rectum in sterile palpation sleeves using deep-well water from the milk house tap as the only lubricant. The same cows designated as the top 20% by S/P ratio [calculated using the formula provided by the manufacturer: (mean optical density (OD) of sample – OD of negative control)/(OD of positive control – OD of negative control); IDEXX Laboratories Inc., 2002] at the first testing period were re-sampled for fecal culture at each of the subsequent 3 seasonal sampling periods. This process was repeated for each group during the first 3 mo, then the cycle was repeated for the remaining 3 seasons, beginning with the first group of cows. The producer was blinded to the ELISA and fecal culture sample results of cattle during the study period.
Laboratory Procedure
Serum antibody.
After collection, all blood samples were taken to the Texas Veterinary Medical Diagnostic Laboratory (TVMDL) and centrifuged. Serum was decanted into serum tubes. A commercial solid-phase ELISA kit (HerdChek; IDEXX Laboratories Inc., West-brook, ME) was used to analyze serum from the blood samples collected. All experimental samples were analyzed in duplicate wells per laboratory protocol following the manufacturer’s instructions. After initial evaluation of the first 3 groups to determine the 20% of cows with highest S/P ratios, all samples (including those collected during the first season) were subsequently maintained frozen at –70°C upon arrival at TVMDL. At the completion of the collection period of the study, all serum samples were thawed and evaluated in duplicate per TVMDL protocol. To limit plate-to-plate and inter-operator variation, each set of samples collected per cow (up to 4) was analyzed on the same plate using a Biomek FX robotic device (Beckman Coulter, Fullerton, CA). Samples were allocated randomly on the plate to minimize potential bias caused by within-plate variation.
Fecal culture.
Upon arrival at TVMDL, all fecal samples were frozen and maintained at –70°C. The fecal culture procedure generally followed that of Stabel (1997). Throughout the study period, each sample was thawed and 2 to 3 g of feces were placed in a 50-mL centrifuge tube and filled to 35 mL volume with distilled water. Samples were mixed on a rotating mixer for 30 min. Samples were left standing for 30 min, and the supernatant fraction was decanted into a new 50-mL tube and centrifuged at 1700 x g for 20 min. Supernatant was discarded, and the pellet was resuspended in 30 mL of 0.9% cetylpyridinum chloride/1.9% brain heart infusion. Samples were then incubated overnight at 37°C as a decontamination step and then centrifuged at 1700 x g for 20 min. Supernatant was discarded, and the pellet was resuspended in 1 mL of sterile water with 50 µg/mL amphotericin B, 100 µg/mL vancomycin, and 100 µg/mL nalidixic acid. Samples were again incubated overnight at 37°C and then inoculated onto Herrold’s egg yolk medium (0.2 mL per tube) with 4 tubes containing Mycobactin J and one tube without. Tubes were placed in a slanted position with caps loosened and incubated at 37°C. Samples were checked for contamination, caps were tightened, and tubes placed in an upright position after 1 wk. Samples were then incubated for up to 15 wk and checked weekly for appearance of MAP. If no growth was visible after 15 wk, samples were determined to be negative. If colonies typical for MAP were observed at 15 wk (i.e., those growing on Mycobactin J only), they were stained using cold acid-fast stain. Acid-fast colonies were confirmed positive for MAP by PCR, a test used to detect IS900, a genomic insertion sequence specific for MAP using a protocol modified from Khare et al. (2004).
Climatic Data
Climatic conditions were assessed using data from a local weather station, including monthly mean, monthly mean minimum, and monthly mean maximum temperatures (mean minimum and mean maximum refer to daily measurements). An on-farm weather station was erected during the study; however, because data were not available for the entire study, the highly correlated data (R2 = 0.997) from the local weather station (distance = 15 km) was used instead.
Statistical Analyses
Variables assessed in the analyses included the following: seropositivity of cows to MAP antigen (binary classification via ELISA); S/P ratio results (interval variable from ELISA); fecal culture result (binary classification); age (based on lactation number); stage of lactation (DIM); and mean, mean daily minimum, and mean daily maximum monthly temperatures. Lactation was evaluated as a 4-level categorical variable to avoid statistical model instability, attributable to insufficient numbers in the highest lactations (i.e., 5 to 8). Categories 1 to 3 corresponded with lactations 1 to 3; category 4 included lactations 4 to 8. Days in milk also were evaluated as a categorical variable representing physiologically important stages of lactation. Category 1 included 0 to 30 DIM, category 2 included 31 to 60 DIM, category 3 included 61 to 150 DIM, category 4 included 151 to 305 DIM, and category 5 included >305 DIM.
Effects of various current and lagged temperature-related seasonal climatic factors on the risk of cows exhibiting seropositivity were tested in a generalized linear modeling framework (PROC GENMOD; SAS Inst., Inc., Cary, NC) using a binomial distribution and a logit link function. Using the repeated statement (by cow) and an auto-regressive [AR(1)] correlation structure, a generalized estimating equation (GEE) was utilized to adjust for the within-cow dependence of the outcome variable over repeated sampling (Diggle et al., 2002). Potential explanatory variables that were analyzed as categorical variables included sampling cohort; season of sampling; sampling month; fecal culture result; lactation number (categorized); and mean, mean daily minimum, and mean daily maximum temperatures for the month of, month before, and 2 mo before sampling. Effect of climatic factors on actual continuous S/P ratio was assessed using a mixed modeling framework (PROC MIXED; SAS Inst., Inc.) with random effects for cow (with an AR(1) correlation structure) and fixed effects for climate and cow-level factors such as lactation number. The S/P ratio was not log-transformed because negative S/P values were possible. This model utilized the same variables as the GLM framework.
Effect of climatic factors on S/P ratio (categorized into quintiles) was assessed in a generalized linear modeling framework using a multinomial distribution and a cumulative logit link function (McCullagh and Nelder, 1989; Hardin and Hilbe, 2003). Using the repeated statement (by cow) and an independent correlation structure, a GEE was utilized to adjust for the within-cow dependence of the outcome variable. An AR correlation structure would have been ideal, but the software procedure would not permit a correlation structure other than independent (Hardin and Hilbe, 2003) for the cumulative logit model. However, the GEE parameter estimates remain robust even if correlation structure is misstated (SAS, 1996). The GEE parameter estimates are based on robust estimates of the standard errors derived from the empirical covariance matrix.
A transitional model based on a first-order Markov-chain for binary data (Diggle et al., 2002) was used to model the dependence of change or no change in positive-negative status and was based on the same explanatory variables as described previously. A cow was considered to be positive if analysis of serum antibodies by ELISA resulted in an S/P ratio 
0.25 (IDEXX Laboratories, 2002). A logistic regression was performed conditioned on the previous response (i.e., positive or negative). This provided consideration for the probability that each subsequent positive or negative test result was conditional (i.e., not independent) on the previous result, while continuing to consider the effects of other explanatory variables.
Fecal culture results were described on the basis of both the proportion of cows exhibiting seropositivity and previous results of a positive or negative test (ELISA or fecal culture). The probability of a cow testing positive for MAP via fecal culture also was evaluated for the effects of mean daily maximum temperature in month of, month before, and 2 mo before sampling along with lactation number in a GLM framework (PROC GENMOD) using a binomial distribution and a logit link function. Using the repeated statement (by cow) and an auto-regressive (AR(1)) correlation structure, a GEE was utilized to adjust for the within-cow dependence of the outcome variable.
To assess potential biases affecting the long-term follow-up of the cohorts of cattle, the effect of each of seropositivity and S/P ratio (categorized into quintiles) on the risk of culling was assessed using logistic regression (SPSS 12.0; SPSS Inc., Chicago, IL). In addition, survival analysis using Cox regression (SPSS 12.0) was used to assess within-lactation time to culling based on seropositivity and quintiles of S/P ratio. Both of the culling analyses were multivariable, adjusted for lactation number and DIM at the time of selection.
Answers to all your Biology Questions
