Evaluation of enzyme-linked immunosorbent assays performed on milk and serum samples for detection of neosporosis and leukosis in lactating dairy cows

Abstract

Serum and milk samples from 1229 cows on 22 Ontario dairy farms were individually tested for antibodies specific for bovine leukosis virus (BLV) and Neospora caninum by enzyme-linked immunosorbent assay (ELISA). Antibodies against BLV were present in 361 serum samples (29.4%) and 369 milk samples (30.0%). Comparing the 2 tests, agreement was almost perfect (k = 0.86; 95% CI = 0.83 to 0.90) and the proportions of samples positive were not significantly different (P = 0.56). Both tests identified the same 3 herds free of bovine leukosis virus. Antibodies against N. caninum were detected in 138 serum samples (11.2%), and 111 milk samples (9.0%). Agreement between the 2 tests was moderate (k = 0.52; 95% CI = 0.43 to 0.59). Four herds were free of neosporosis by the serum test, while 10 herds were negative by the milk test. The ELISA on milk samples facilitates sample collection to classify herds free of BLV; the milk N. caninum ELISA was less reliable in predicting herd-level infection.

Introduction

Infectious agents found in apparently healthy animals, such as bovine leukosis virus (BLV) and Neospora caninum, are receiving increased attention due to regulations of the World Trade Organization (WTO) on animal health and animal movement between countries (1). Farm-level costs of infection may be due to altered milk production (2–4), risk of reproductive failure (4,5), and increased risk of culling (6); however, studies have failed to find a consistent association (2,3). In recent reports representing multiple Canadian provinces, serologic evidence of BLV exposure was found in 30.3% of cows with > 70% of herds having at least 1 positive cow, while N. caninum was reported in 11.4% of cows and > 60% of herds were classified as exposed based on seroconversion of 2 or more animals (7–11).

Most commercially available diagnostic procedures for BLV focus on identification of specific anti-BLV antibodies in serum from animals more than 6 mo old (12,13). The most frequently used serological tests are agar gel immunodiffusion (AGID) (14) and enzyme-linked immunosorbent assay (ELISA) (15,16). The estimated sensitivity (> 98%) and specificity (95% to 100%) of these tests indicate that they are reliable and accurate (15,16). Serologic tests have been the foundation for eradication strategies and certification of BLV-free herds, designed to protect access to international markets. Herd-level BLV status has been reported using an antibody detection ELISA conducted on bulk tank milk samples (17,18). Herd-level classification based on bulk milk reliably identifies BLV negative herds but the estimated specificity requires further investigation to confirm BLV positive herds. Determination of BLV infection status has been made possible through isolation of integrated proviral DNA using polymerase chain reaction (PCR) (19,20).

Neosporosis is associated with endemic herd infections in many countries (21–23). The gold standard for N. caninum infection is confirmation of characteristic lesions in the aborted fetus combined with PCR or immunoperoxidase staining of fetal tissue. In many situations fetal material is not available, thus a presumptive diagnosis is based on detection of antibodies to N. caninum in serum or milk. Several studies have reported the sensitivity and specificity of indirect fluorescent antibody tests (IFAT), a N. caninum agglutination test, and both competitive and indirect ELISA performed on both serum (24) and milk (7,25,26). These studies used sera from animals with pre-test information regarding abortion status or established a reference standard for assessing the characteristics of the assay under investigation (24,25).

Given the potential restrictions on animal movement, the potential for lost production and impaired reproduction, the utility of immunological assays that lend themselves to quickly and efficiently diagnosing a herd’s infection status is evident. While serological tests are available for the diagnosis of both diseases, serum collection and processing can be costly and time consuming. Milk-based ELISAs offer an attractive alternative to serology because the samples are relatively convenient and simple to collect, thus reducing costs.

The purpose of this study was to evaluate test agreement between 2 new commercially available BLV and N. caninum milk ELISAs and serological tests currently used in the industry as reference tests [BLV (16); N. caninum(25)] and in adherence to current test evaluation guidelines (27). Following initial evaluation of the skim milk BLV ELISA, the utility of using whole milk for the BLV ELISA was investigated to facilitate testing.

Materials and methods

Animals

The study population consisted of herds that had previously participated in a N. caninum prevalence study (22) (17 herds with a N. caninum herd prevalence > 10% and 5 herds with a N. caninum herd prevalence < 7%). Herds were selected based on participation in routine milk recording through the Canwest Dairy Herd Improvement (DHI) association, and a willingness to participate. In total, 1229 lactating Holstein cows in 22 dairy herds in southwestern Ontario were enrolled. Herd size ranged from 22 to 210 (median = 50) lactating animals. Herds encompassed a variety of management strategies related to housing, nutrition, reproduction, culling, and biosecurity.

Sample collection

Herd managers were contacted prior to sampling in an effort to collect blood samples in close temporal proximity to the next scheduled DHI milk test. At the scheduled visit, blood samples were collected via coccygeal venipuncture from all lactating cows. Serum was harvested within 24 h of collection and stored in duplicate at −70°C until analyzed. Milk samples were collected from all lactating dairy cows at the next DHI test day by a customer service representative as part of the normal test day routine and preserved with bronopol. Collected milk samples were manually identified at the DHI milk lab and forwarded to a commercial laboratory (AntelBio Systems, Lansing, Michigan, USA) immediately after milk component analysis. Preserved milk samples were frozen within 48 h of collection and stored until all samples were collected and milk ELISAs were run in a single batch. The time between serum sample collection and milk collection ranged from 0 to 127 d. A total of 18 herds and 1073 cows had serum and milk samples collected within 28 d of each other. Cow level information, including date of most recent calving, milk production, somatic cell count, milk fat and protein percent, and reproductive status was collected at the DHI milk test.

Diagnostic testing

All serum samples were submitted at the same time to 2 American Association of Veterinary Laboratory Diagnostician (AAVLD) accredited laboratories (BLV, Animal Health Laboratory, Guelph, Ontario; Neospora, California Animal Health and Food Safety Laboratory System, Davis, California, USA). Serum anti-BLV antibodies were tested using a commercial ELISA according to the manufacturer’s instructions (16). Results from BLV were reported as positive or negative using the recommended ELISA optical density sample-to-positive ratio (S:P) ≥ 0.6. Sensitivity and specificity of the BLV ELISA relative to the agar gel immunodiffusion test (gold standard) at this threshold are 99.8% and 100%, respectively (16). Furthermore, the sensitivity of the BLV ELISA is constant across varying population prevalences (16). Serum antibodies against N. caninum were detected using a previously reported kinetic ELISA (25). Serum results for N. caninum were considered antibody positive for samples with an S:P ratio ≥ 0.45. Sensitivity and specificity of the N. caninum ELISA at this threshold are 89% and 97% respectively, assuming a 40% seroprevalence in the target population. Milk samples were sent to a commercial laboratory (AntelBio Systems) where they were tested using the IDEXX HerdChek* ELISA developed for detection of antibodies against BLV and N. caninum. The BLV skim milk ELISA was modified to use a 1:10 dilution of milk. This dilution was required to reduce the effects of carryover in the DHI milk sampling process and was calculated based on previous in-house evaluations. The BLV skim milk ELISA results were interpreted using corrected optical density (background subtraction; sample OD minus negative control OD) rather than the use of S/P ratio based on better performance using in-house comparisons. The BLV skim milk ELISA results were classified as positive with corrected optical density (OD sample — OD negative control) > 0.1. At the end of this trial, the milk ELISA was modified to expedite testing. The modified milk ELISA was validated for testing of whole milk instead of skim milk using the same dilution and threshold classification strategy. The IDEXX N. caninum milk ELISA was modified to a 1:2 dilution (25). Results were classified as positive for N. caninum milk samples with a corrected OD (background subtraction; sample OD minus negative control OD) > 0.1 to maximize the proportion of correctly classified cows based on in-house calculations.

Data analysis

Separately and repeated for each pathogen, the McNemar test was used to compare paired population proportions of positive results for the serum and milk ELISAs. Agreement between results for serum and milk samples was determined by calculating the κ statistic. A 95% confidence interval was calculated for the κ statistic using an analytical method (27). Data recorded by the reference laboratory included raw S:P ratio for the N. caninum serum samples and dichotomous outcomes (positive or negative) for the BLV serum samples. A Pearson correlation coefficient was calculated to estimate the degree to which the milk and serum results agree for each pathogen.

Test performance, sensitivity, specificity, and the predicted probability of disease given a positive test are dynamic values reflecting the distribution of covariate factors in the sampled population (27,28). Specifically, experimental conditions may overestimate test sensitivity and specificity by testing populations with artificially elevated disease prevalence for a positive population and specific pathogen-free populations for a negative population (27). Due to a large discrepancy between the cow-level N. caninum seroprevalence and the seroprevalence used to validate the reference test, the relative impact of herd-level prevalence on the N. caninum milk ELISA performance was evaluated. The herds were separated based on herd prevalence of serum titers and the κ statistic calculated for both high prevalence (above the cow-level mean prevalence) and low prevalence herds.

To investigate the impact of specific cow level variables on the sample OD animal variables including the number of days between collection of serum and milk samples, days postpartum at sample collection, parity, milk production on the day milk samples were collected, milk fat and protein, linear score, and reproductive status (fresh, bred, pregnant), were investigated while controlling for the potential impact of clustering at the herd level. Using a population averaged model and after investigation of residuals, a linear prediction was generated to account for cow-level significant covariates using the same threshold values as applied on the corrected OD, and the κ statistic was recalculated. A κ statistic estimate was generated for both the unadjusted milk results relative to serum results and for the cow level variable adjusted milk OD relative to the serum results. The resultant κ statistic and their respective 95% CI were compared to estimate the impact of adjusting the OD for cow-level covariates. For all analyses, a commercial statistical program (Intercooled Stata 9; Stata Corp, College Station, Texas. USA) was used; probability values < 0.05 were considered significant.

Results

Bovine leukosis virus

The serum BLV prevalence was 29.4% (95% CI: 26.8% to 31.9%); within-herd prevalence ranged from 0.0% to 75.0%. Using the skim milk BLV ELISA the corrected optical density readings ranged from −0.0182 to 1.23. The skim milk BLV ELISA prevalence was 30.0% (95% CI: 27.4% to 32.6%), within-herd skim milk BLV ELISA prevalence ranging from 0.0% to 77.7%. Three herds were classified as BLV negative by both the serum and skim milk ELISA. The level of agreement between the serum and skim milk ELISA represents almost perfect agreement (κ = 0.86; Table 1). The McNemar test statistic indicates there was no difference in the proportion of animals that were positive in both tests.

Table 1.

Cross-classification results of an ELISA for antibodies against bovine leukosis virus (BLV) in serum samples from 1229 Holstein cows from 22 Ontario dairy herds and an ELISA for antibodies against BLV in milk samples from the same cows

	Results for milk samples
	Positive	Negative	Total
Results for serum samples
Positive	330	31	361
Negative	39	829	868
Total	369	860	1229

κ = 0.86 (95% CI = 0.83 to 0.90); McNemar χ² = 0.91, P = 0.34

The introduction of a whole milk ELISA required further evaluation. The whole milk ELISA was evaluated in a single herd (n = 251). The herd prevalence for the serum and skim milk ELISA was 45.0% and for the whole milk ELISA was 45.8%. Agreement between the whole milk and skim milk ELISA was almost perfect (κ = 0.97; Figure 1). Similarly, in a single herd of 251 Holstein cows the whole milk ELISA had almost perfect agreement with the serum ELISA (κ = 0.90; Table 2).

Figure 1.

Scatterplot of optical density (OD) of the sample minus the OD of the negative control from a skim milk ELISA and whole milk ELISA for antibodies against bovine leukosis virus (BLV) in milk samples from 251 Holstein cows from 1 Ontario dairy herd using the skim milk and whole milk ELISA from the same cows, overlaid by horizontal and vertical lines indicating the recommended cut-off values for each ELISA.

Table 2.

Cross-classification results of an ELISA for antibodies against bovine leukosis virus (BLV) in serum samples from 251 Holstein cows from a single Ontario dairy herd and an ELISA for antibodies against BLV in whole milk samples from the same cows using the refined milk ELISA

	Serum ELISA
	Positive	Negative	Total
Whole Milk ELISA
Positive	108	7	115
Negative	5	131	136
Total	113	138	251

κ = 0.90 (95% CI = 0.85 to 0.96); McNemar χ² = 0.33, P = 0.56; Pearson correlation coefficient (r) = 0.90

The adjusted OD on skim milk (raw absorbance — negative sample absorbance) was significantly increased among cows in 2nd or ≥ 3rd lactation (P < 0.01), relative to cows in their 1st lactation. There were also associations between absorbance measurement and days in milk at test day, linear score, and percent fat in the milk. Adjusting for these cow-level covariates did not improve the ability to correctly identify cows with or without BLV.

Neospora caninum

The serum ELISA prevalence of N. caninum antibodies was 11.2% (95% CI: 9.5% to 13.0%); within herd prevalence ranged from 0% to 35%. The serum ELISA classified a total of 4 herds as free of N. caninum antibodies. The range of corrected ODs for this milk ELISA was from −0.087 to 0.71. The milk ELISA prevalence was 5.9% (95% CI: 4.5% to 7.2%); within herd prevalence ranged from 0% to 20.4%. Using the milk ELISA, the 4 negative herds from the reference test were correctly identified; however, an additional 6 herds were incorrectly identified as N. caninum negative. The level of agreement for the milk and serum samples was moderate (κ= 0.52; 95% CI = 0.43 to 0.59; Figure 2) and the McNemar test indicated that the proportions of positive results for milk and serum samples were significantly different (McNemar χ² = 46.3, P < 0.001). Exclusion of herds with a sampling lag > 28 d did not influence the estimate of agreement (κ = 0.52).

Figure 2.

Scatterplot of optical density (OD) of the sample minus the OD of the negative control from a milk ELISA and S:P ratio from a serum ELISA for antibodies against Neospora caninum from 1229 Holstein cows from 22 Ontario dairy herds, overlaid by horizontal and vertical lines indicating the recommended cut-off values for each ELISA.

To investigate the impact of herd prevalence of antibodies to N. caninum, test agreement was investigated separately among herds with serum ELISA cow-level prevalence > 11.2% (herds = 10; cows = 519), and herds with a low (< 11%) serum ELISA prevalence (herds = 12; cows = 710). Among high prevalence herds the estimate of test agreement improved slightly (κ = 0.55, 95% CI = 0.45 to 0.64; McNemar χ² = 32.1, P < 0.001). Test agreement decreased among low prevalence herds (κ = 0.37, 95% CI = 0.15 to 0.49; McNemar χ² = 14.2, P < 0.001).

The adjusted OD (raw absorbance — negative sample absorbance) was significantly increased among cows in ≥ 3rd lactation (P < 0.01), relative to cows in their 1st lactation and it tended to increase as the days in milk at test day increased (P = 0.07). The difference in absorbance decreased as days between milk and serum sample increased (P < 0.01). Exclusion of 4 herds (215 cows) with a lag greater than 28 d between serum and milk collection did not alter the estimate of test agreement. All herds tested after 28 d were classified as N. caninum negative. Adjusting the absorbance to account for cow-level covariates did not improve the level of agreement between the serum and milk ELISAs.

Discussion

Serologic diagnosis of BLV exposure using the IDEXX HerdChek* Anti BLV ELISA is an industry standard diagnostic test (16). Milk production, fat and protein production, and longevity studies have had variable associations with BLV serostatus (2–6). Despite these questionable production impacts, BLV is listed by the WTO as a potential barrier to animal movement and as such is of economic importance. The skim milk ELISA had almost perfect agreement with the existing serum ELISA. The McNemar test statistic indicated that the seropositive cows were also positive using the skim milk ELISA. Relative to the serum ELISA there were 31 false negatives and 39 false positive cows, which may represent a sample carryover effect introduced at the time of milk collection. Based on previous in-house research, in which the impact of milk carryover in the milk meter was investigated, a 1:10 dilution of milk was used to minimize the potential impact of this carryover.

At the herd-level there were no herds incorrectly classified as BLV negative by the skim milk ELISA relative to the serum ELISA. This level of agreement is rarely reported, indicating that the BLV skim milk ELISA has significant potential to facilitate milking herd classification. At the end of the study the milk preparation requirements were refined to expedite testing. Cross-classification of cows using both the skim milk ELISA and the whole milk ELISA test from a single large herd indicates that the refinement in the procedure did not alter classification of cows, and it did not impact the level of agreement between the serum and whole milk ELISA.

The observed lactation associated increase in prevalence of BLV was anticipated, as transmission of BLV occurs horizontally and throughout life (13,29). The significance of increased antibody prevalence in late lactation and associated with increased milk production and increased fat percentage is difficult to interpret in light of inconsistent production impacts based on BLV serostatus identified in large multi-herd studies (2–4,6). Herd selection did not require prior knowledge of BLV status and was representative of Ontario dairy herds, such that no selection bias is expected.

Serologic detection of N. caninum exposure has been reported with variable success (30,31). In the present study we compared 2 N. caninum ELISA’s based on sonicated tachyzoite suspensions developed to detect serum antibody concentrations. Sensitivity and specificity of the serum-ELISA at an S:P ratio of 0.45, assuming a 40% seroprevalence in the target population, has been reported to be 89% and 97%, respectively (25). The seroprevalence in the present population was 11.2% — considerably lower. Given the lack of a true gold standard for cow level surveillance in the absence of abortion information, and the large discrepancy in the estimated population prevalence of N. caninum, analysis was limited to assessment of agreement between the 2 tests. The present study indicates that the milk ELISA to test for exposure to N. caninum in the dairy herds evaluated performed moderately well. At the herd level, the milk ELISA misclassified 6 herds as N. caninum negative. Previous evaluation of the IDEXX HerdChek^® Anti-Neospora ELISA using serum and milk samples reported an optimal sample to positive cutpoint for milk of 0.261 with a corresponding sensitivity and specificity of 90% and a Kappa of 0.8 (26). Previous comparisons have been conducted in populations with a much higher pre-test probability of infection, and a current history of Neospora-associated abortion. Improved sensitivity of serological diagnosis of N. caninum has been reported in the endemic state rather than the epidemic state and herd characteristics have been identified as a potential source of bias (26,32). In the current investigation, 17 herds had a diagnosed Neospora-associated abortion prior to 1999. Of these 17 high prevalence herds, 7 had moved into a low prevalence status prior to the start of this investigation.

The time between milk collection at DHI test day and serum collection ranged from 0 (the same day) to 127 d (median = 18 d) and there was a significant reduction of risk of seropositivity with increased time between milk and serum collection. The level of antibody in milk varies with month after calving and with lactation number (26,31). A single milk ELISA validation study reported increased positive values in a milk ELISA relative to a serum ELISA in late lactation (26), while a separate Dutch study reported the level of antibody was highest among fresh cows and cows at the end of lactation relative to cows in the middle of lactation (31). Despite this fluctuation in the magnitude of the response there is little evidence to support sero-status classification changes throughout lactation (31). Although enrolled herds had prior knowledge of their N. caninum serostatus (22), results of the current study were not reported to individual producers until both serum and milk analyses were complete. Thus, there was no opportunity for culling of animals based on serostatus prior to collection of milk samples. As the time between serum and milk collection increased there was an opportunity for some cows to be dried off or sold prior to milk collection. This would not have impacted the results as only cows with both serum and milk results were included in the final analysis.

From a herd management perspective, the lactation associated increase in the prevalence of N. caninum antibody was not anticipated. Given the anticipated increased risk of abortion among N. caninum-positive cows these cows are expected to have a shorter herd life. From a biological perspective there is an increased risk of N. caninum associated abortion among cows 4 years old and younger (33). Separation of true positive and false positives in this data set is not possible. Given the herds included in the study, there is potential for introduction of an age-related bias as these herds previously knew the serostatus for their animals which permitted selective culling of heifers from seropositive dams. Alternatively, the increased prevalence of N. caninum seroconversion among 2nd and ≥ 3rd lactation cows could suggest that among herds that have not experienced a N. caninum attributed abortion storm, there is no increased risk of culling among N. caninum positive animals (22).

The results of the McNemar test statistic indicate that the serum test positive cows were statistically different from the milk ELISA test positive cows. Given the significant McNemar test statistic, caution is warranted in interpretation of test agreement. The discrepancy in the populations that tested positive and negative will limit the utility of this test in determining the N. caninum status of individual animals in low prevalence herds. This difference does not preclude the use of the test as a herd-level screening test. Misclassification at the herd level was the result of 2 or more cows classified as negative with the milk ELISA relative to the serum ELISA. Thus application of this milk ELISA in the absence of clinical history of Neospora-associated abortion may contribute to the underestimation of the prevalence of N. caninum. The level of agreement increased among high prevalence herds and herds with a clinical history of Neospora-attributable abortions.

In conclusion, the use of a BLV whole milk-ELISA for classification of infection status shows promise. However the N. caninum milk-ELISA evaluated in this investigation indicates a limited application especially among low-prevalence herds.