Association between the genetic similarity of the open reading frame 5 sequence of Porcine reproductive and respiratory syndrome virus and the similarity in clinical signs of Porcine reproductive and respiratory syndrome in Ontario swine herds

Abstract

A study of Ontario swine farms positive for Porcine reproductive and respiratory syndrome virus (PRRSV) tested the association between genetic similarity of the virus and similarity of clinical signs reported by the herd owner. Herds were included if a positive result of polymerase chain reaction for PRRSV at the Animal Health Laboratory at the University of Guelph, Guelph, Ontario, was found between September 2004 and August 2007. Nucleotide-sequence similarity and clinical similarity, as determined from a telephone survey, were calculated for all pairs of herds. The Mantel test indicated that clinical similarity and sequence similarity were weakly correlated for most clinical signs. The generalized additive model indicated that virus homology with 2 vaccine viruses affected the association between sequence similarity and clinical similarity. When the data for herds with vaccine-like virus were removed from the dataset there was a significant association between virus similarity and similarity of the reported presence of abortion, stillbirth, preweaning mortality, and sow/boar mortality. Ownership similarity was also found to be associated with virus similarity and with similarity of the reported presence of sows being off-feed, nursery respiratory disease, nursery mortality, finisher respiratory disease, and finisher mortality. These results indicate that clinical signs of PRRS are associated with PRRSV genotype and that herd ownership is associated with both of these.

Introduction

Porcine reproductive and respiratory syndrome (PRRS) is a clinically variable disease that is generally characterized by its herd-level presentation rather than its effect on an individual pig. Reproductive failures in the sow population, especially abortions after 100 d of gestation, stillbirths, piglet mummification, and neonatal deaths, are a result of transplacental transmission of the causative agent, Porcine reproductive and respiratory syndrome virus (PRRSV). Disease in the weaned population consists of dyspnea and other signs of respiratory distress and an increased mortality rate. The virus can also cause disease in adult animals. Viremia in sows and boars results in pyrexia, inappetence, and death (1). In the finisher phase of production PRRSV causes respiratory disease, especially in the presence of other respiratory pathogens, such as Mycoplasma hyopneumoniae and Porcine circovirus type 2 (2).

The RNA genome of PRRSV consists of 9 open reading frames (ORFs) (3–5): ORFs 1a and 1b encode nonstructural proteins (6,7), ORFs 2a, 2b, 3, 4, and 5 encode envelope glycoproteins (5,8,9), and ORFs 6 and 7 encode the matrix protein and nucleocapsid protein, respectively (8,10). Perhaps because the protein it encodes is exposed to selective pressures by being located on the outside of the virus, ORF5 is a highly variable region of the genome (11). It has been used extensively for diagnostic identification of virus subtypes by restriction fragment length polymorphism (RFLP) (12–14), and sequencing of ORF5 has been used to measure phylogenetic distances in groups of PRRSV isolates (15–19).

Isolates of PRRSV that are genetically different have been shown in clinical trials to be associated with different clinical disease in young pigs (20) and in sows (21). There are few published reports of epidemiologic investigations that have explored the association between PRRSV genotype and clinical disease. Clinical disease has been shown to be strain-dependent in epidemiologic investigations in Illinois (15) and Ontario (22). This lack of field data on clinical disease with concomitant laboratory samples is likely due to the difficulty of collecting clinical data associated with PRRSV isolates from the field. The objectives of this study were to investigate the association between genetic similarity in the PRRSV ORF5 from diagnostic samples and similarity in the reported clinical signs of disease in the herds from which the samples originated and to investigate the impact of the presence of vaccine-like isolates in the data on this association.

Materials and methods

Diagnostic submissions from Ontario swine herds that tested positive for PRRSV by reverse-transcription polymerase chain reaction (RT-PCR) at the Animal Health Laboratory (AHL) of the University of Guelph, Guelph, Ontario, during the period September 1, 2004, to August 31, 2007, were eligible for the study. The RT-PCR was done as described by Cai and colleagues (13) to amplify 433 base pairs of ORF7 of PRRSV for cases before December 4, 2006. Samples from cases submitted to the AHL after December 4 were tested by a PRRSV multiplex real-time RT-PCR (Tetracore, Rockville, Maryland, USA). The AHL database was searched on 1 occasion for all PRRSV-positive submissions, referred to as “cases” by AHL, from the period September 1, 2004, to January 14, 2006. These retrospective cases were eligible only if samples from the original case had been stored by the AHL. Twice per week, starting January 15, 2006, the AHL database was searched for PRRSV-positive prospective cases.

For the purposes of this study the following definitions are applied to be in agreement with previous literature (23,24). The term ownership refers to any premises housing pigs under single corporate or private ownership. The term premises, refers to a single contiguous land parcel with 1 or more buildings housing pigs; an ownership may have 1 or more premises. The term herd refers to a group of pigs housed at the same premises at the same point in time. The unit of interest for the study was the herd. A premises and, by extension, a herd were eligible for inclusion in the study only once. The veterinarian listed on the case report was contacted to determine which premises had been sampled and to obtain contact information for the owner. The owners were contacted and asked if they would like to participate in the study. Permission was obtained for sequencing PRRSV-positive samples in the AHL database. Participating owners or managers were surveyed by telephone. The case, premises, and herd information was then entered into a study database in Microsoft Access (Microsoft Corporation, Redmond, Washington, USA).

The telephone survey was completed in approximately 30 min. Survey questions were about herd demographics and perceptions of clinical signs of PRRS in the herd. The latter question was phrased as follows: “At the time of the sample submission to AHL, what clinical problems did you have in the barn? Please indicate with ‘Yes’, ‘No’, ‘Don’t know’, or ‘Not applicable’ for each clinical problem. If you do not know whether a clinical problem occurred, please respond ‘Don’t know’. If the problem is not applicable to the barn (for example, if you only have nursery pigs, then the clinical problem ‘abortion’ does not apply), please respond ‘not applicable’.” The interviewer then listed each of the clinical signs, allowing the interviewee to respond to each: abortion, sows being off feed, stillborn piglets, weak-born piglets, sow and boar mortality, preweaning mortality, nursery mortality, nursery respiratory disease, finisher mortality, and finisher respiratory disease. The survey responses were recorded on paper and entered into the study database.

Reverse-transcription PCR was used as previously described (12) to amplify ORF5 of PRRSV from all samples included in the study. Only 1 isolate per herd was included in the study because of financial limitations. Also, because the isolates were from diagnostic submissions, some herds had only 1 isolate available, and therefore not all herds could have been tested multiple times. With the inclusion of only 1 isolate per herd, the presence of multiple strains at a single site at 1 point in time could have been missed (14). Previous work had shown that genetic variation of PRRSV ORF5 within the same herd is typically less than 5% when 2 samples are collected at close to the same time (25).

Amplification was done on the original stored sample submitted to the laboratory. The product was sequenced at the Guelph Molecular Supercentre, University of Guelph. All database sequences were aligned by means of MegAlign software, version 7 (DNASTAR Lasergene, Madison, Wisconsin, USA) with use of the Clustal W method followed by manual correction at any gaps created by the algorithm. A matrix of paired genetic similarities was constructed for all sequences in the database. Genetic similarity was calculated as the percentage of nucleotides shared between pairs of sequences within ORF5. Sequences were also compared with those of the 2 vaccine types being used in Ontario: Ingelvac PRRS MLV and Ingelvac PRRS ATP (Boehringer-Ingelheim, St. Joseph, Missouri, USA). The sequences of the 2 modified-live-virus strains were obtained from the online US National Center for Biotechnology and Information database under accession numbers AAD27656 and AAV71017, respectively. A database sequence was recorded as “wild type” if the similarity with either vaccine sequence was less than 98% and as “vaccine-like” if the similarity was 98% or greater with either vaccine sequence. The 98% cut-off was based on a previously published and commonly used rule for interpretation of PRRSV ORF5 sequence comparisons (14,16).

Clinical similarity was calculated for each clinical sign as a binary outcome for each pair of herds as follows: 1 if the response as to whether a particular clinical sign was present in the herd was “Yes” for both herds of a pair or “No” for both herds of a pair, and 0 if the response was opposite for the 2 herds. Ownership similarity was also calculated for each pair of herds, as follows: 1 if both herds of a pair were under the same ownership and 0 if the members of a pair were under different ownership.

Association between genetic similarity and clinical similarity was first tested with the Mantel test (26), which measures the correlation between 2 variables in matrix form by means of a Pearson correlation coefficient and a Spearman rank correlation coefficient; these coefficients are assigned a degree of significance by simulation. Correlation of the reciprocal transform of the genetic distance against the clinical similarity was also tested by the Mantel test, as indicated by Mantel (26). For all tests, only comparisons between herds that had the phase of production to which the clinical sign pertained were included. Therefore, the dataset for a clinical sign pertaining to sows would be different from the dataset for a clinical sign in the finisher or nursery phase of production. The association between sequence similarity and ownership similarity was also tested by the Mantel test and by linear regression.

The associations between clinical similarity, sequence similarity, and ownership similarity were further tested with the generalized additive model (GAM) because of unexpected results of the Mantel test, which showed a significant but very small correlation between sequence similarity and clinical similarity. The GAM was used to allow for nonlinearity that could not be adequately modeled by logistic regression. Univariable logistic GAM models were built to test the association between clinical similarity and percent sequence similarity with thin-plate regression splines and a maximum allowable knot value of 20. The association was first tested for all the herds in the dataset and then tested for only the herds in which a wild-type virus was identified. Univariable models were built to test the association between ownership similarity and clinical similarity for each clinical sign. For those clinical signs which a significant association was found, ownership similarity was included as a parameter in subsequent multivariable models. The multivariable GAM was a logistic model with the binary outcome of clinical similarity and the independent variables being a smooth function of genetic similarity and a parametric variable for ownership similarity.

The GAM models for the association between clinical similarity and percent sequence similarity were interpreted in the following way. The probability that a clinical sign was similar in 2 herds was considered to be significantly greater than expected by chance if the lower 95% confidence interval (CI) of the GAM model, calculated by a normal approximation, crossed above a line equal to the logit of the proportion of clinical similarity in the dataset. Models were further investigated only when the P-value for the nonparametric smoothing term was less than 0.05 and the model prediction illustrated a region where the probability of clinical similarity was significantly different from what was expected by chance. Pearson residuals and Cook’s distance were calculated for each of the final models. The maximum 1% of Pearson residuals, the minimum 1% of Pearson residuals, and the maximum 1% of observed Cook’s distance values were sequentially dropped from the dataset and the models refitted to determine if each model was greatly influenced by these observations. Refitted models on datasets with observations removed were fitted in the same manner as the original model except that the maximum knot value was set to the nearest integer greater than the total degrees of freedom from the original model.

The CIs, calculated by a normal approximation, were considered to be potentially biased estimates because of the data structure (27). The data structure could have an inherent dependent structure and could be interpreted as having an inflated sample size because of the use of similarities between pairs of herds rather than a measure at the herd level. Using similarity between pairs of herds results in the number of observations in an analysis being equal to the original number of herds² (a comparison matrix) minus the original number of herds (subtracting the diagonal) divided by 2 (using the upper triangle of the symmetric comparison matrix). The values in the comparison matrix may also be clustered within herds. This would result in the variation of similarity within a column in the matrix being less than the variation between columns. Because of these potential violations of the model assumptions, 95% simulation envelopes were calculated for each regression line to improve confidence in the model interpretation. The simulation envelopes were calculated by repeating the regression after randomly shuffling the dependent variable 5000 times. After each permutation the regression line was recorded. The 95% simulation envelope consisted of those points lying with the central 95% of points along the regression line assessed at each 0.1% genetic similarity. If the observed regression line was outside of the 95% simulation envelope it was considered to be significantly different from what would be expected by chance.

Results

A total of 37 veterinarians serving 106 owners with more than 1 premises in the study and 148 owners with a single premises enrolled in the study were contacted. The proportion of herds reporting each of the clinical signs, the total number of herds of each herd type, and the proportion of herd pairs in which the reported clinical signs were similar are presented in Table I. The associations by Spearman rank correlation coefficient between percent sequence similarity and clinical similarity for each clinical sign with use of the untransformed data are shown in Table II. The results of the correlation tests using the Pearson correlation coefficient and the reciprocal transformed data did not yield different information about the relationship between percent sequence similarity and clinical similarity, so the specific values are not reported.

Table I.

Proportion of all herds and only those in which wild-type Porcine reproductive and respiratory syndrome virus (PRRSV) was identified that reported each of the clinical signs and proportion of pairs of herds that had similar reported clinical signs

	Proportion of herds reporting clinical signsa (%)		Proportion of pairs with reported similarityb (%)
Clinical sign	With wild-type virus	All herds	With wild-type virus	All herds
Abortion	97/176 (55.1)	99/195 (50.8)	7737/15 400 (50.2)	9411/18 915 (49.8)
Sows off feed	97/176 (55.1)	103/195 (52.8)	7737/15 400 (50.2)	9439/18 915 (49.9)
Stillborn piglets	89/176 (50.5)	95/195 (48.7)	7657/15 400 (49.7)	9415/18 915 (49.8)
Weak-born piglets	94/176 (53.4)	101/195 (51.8)	7692/15 400 (49.9)	9421/18 915 (49.8)
Sow/boar mortality	58/176 (33.0)	61/195 (31.3)	8556/15 400 (55.6)	10 741/18 915 (56.8)
Preweaning mortality	102/176 (58.0)	107/195 (54.9)	7852/15 400 (51.0)	9499/18 915 (50.2)
Nursery respiratory disease	117/171 (68.4)	133/197 (67.5)	8217/14 535 (56.5)	10 794/19 306 (55.9)
Nursery mortality	125/171 (73.1)	141/197 (71.6)	8785/14 535 (60.4)	11 410/19 306 (59.1)
Finisher respiratory disease	66/147 (44.9)	89/178 (50.0)	5385/10 731 (50.2)	7832/15 753 (49.7)
Finisher mortality	66/147 (44.9)	87/178 (48.9)	5385/10 731 (50.2)	7836/15 753 (49.7)

^aThe number of herds in the denominator changes for each clinical sign depending on how many herds were in the applicable stage of production.

^bThe total number of pairs in the denominator is related to the total number of herds making up the pairs: number of pairs = (number of herds² – number of herds)/2.

Table II.

Results of Mantel test for correlation between clinical similarity and genetic sequence similarity with use of the Spearman rank correlation coefficient for untransformed data

Clinical sign	Spearman rank correlation coefficient	P-value
Abortion	0.09	< 0.001
Sows off feed	0.02	0.02
Stillborn piglets	0.03	0.006
Weak-born piglets	0.03	0.003
Sow/boar mortality	0.0005	0.48
Preweaning mortality	0.04	0.001
Nursery respiratory disease	−0.0001	0.50
Nursery mortality	−0.01	0.71
Finisher respiratory disease	−0.009	0.97
Finisher mortality	−0.02	0.99

The results of the univariable GAM models for an association between clinical similarity and percent sequence similarity were different when all virus types were included in the analysis compared with when only wild-type viruses were included (Figure 1). This was also true for the models in Figure 2, but for the sake of brevity the principle is illustrated with just the 2 models in Figure 1.

Figure 1.

Generalized additive models (GAMs) for the association between percent sequence similarity and similarity of the clinical signs abortion and stillbirth when all virus types are included (top 2 graphs) and when only wild-type viruses are included (remaining 4 graphs). The 95% confidence intervals (CIs; shaded areas) for the top 4 graphs were generated by normal approximation. The dashed line indicates the logit of the proportion of pairs of herds reporting clinical similarity; this line is considered to be the null hypothesis of the model, where the odds of clinical similarity are no greater than chance alone. The lowest 2 graphs are shown with a 95% simulation envelope (shaded areas) based on 5000 random model permutations; this region is considered to be consistent with the null hypothesis.

Figure 2.

Models for the association between percent sequence similarity and the similarity of 5 other clinical signs. These GAMs include only wild-type viruses and are shown with a 95% simulation envelope based on 5000 random model permutations; this region is considered to be consistent with the null hypothesis.

Figure 1 shows that for similarity of the clinical sign of abortion, for all virus types the prediction of the GAM model had a positive slope in the region above 85% percent sequence similarity. The prediction increased above the clinical similarity that would be expected by chance alone at a percent sequence similarity of 92.5%. However, the predicted odds of clinical similarity started to decrease as the sequence similarity approached 100%, and the lower limit of the 95% CI crossed below what would be expected by chance at a percent sequence similarity greater than 99%. Similar, counterintuitive results were found for the clinical sign of stillbirth: the estimate of clinical similarity was not different from what was expected by chance in the region close to 100% sequence similarity.

Figure 2 illustrates the predictions from the other significant models of association between clinical similarity and sequence similarity for wild-type viruses only with simulation envelopes. These include sows being off feed, weak-born piglets, sow/boar mortality, preweaning mortality, and respiratory signs in the nursery. The significance levels and estimated degrees of freedom for the models in Figures 1 and 2 are listed in Table III. The odds of similarity of preweaning mortality were found to be significantly greater than expected by chance when the sequence similarity was greater than 93% but less than 99% in the model with only wild-type viruses. The odds of similarity of sow/boar mortality were significantly greater than expected by chance when the sequence similarity was greater than 98% or less than 82%.

Table III.

Results of univariable generalized additive models for wild-type viruses only. For each clinical sign, two models are presented: The association between clinical similarity and ownership similarity; the association between clinical similarity and sequence similarity

	Ownership similarity		Smooth term of sequence similarity
Clinical sign	Odds ratioa	P-value	Estimated degrees of freedomb	P-valuec
Abortion	—	0.56	6.0	< 0.0001
Sows off feed	7.39	0.06d	1.0	0.001
Stillborn piglets	—	0.32	4.1	< 0.0001
Weak-born piglets	—	0.82	1.0	< 0.0001
Sow/boar mortality	—	0.39	6.8	< 0.0001
Preweaning mortality	—	0.86	8.0	< 0.0001
Nursery respiratory disease	5.16	0.008d	6.1	0.007
Nursery mortality	6.34	0.011d	1.0	0.07
Finisher respiratory disease	6.81	0.0003d	1.2	0.23
Finisher mortality	2.49	0.021d	1.0	0.50

^aThe multiplicative odds of a clinical sign being similar in 2 herds under the same ownership as compared with 2 herds under different ownership.

^bAn indication of the order of fitting that the model is estimating; reported for the association between percent sequence similarity and clinical similarity.

^cA measure of the probability of the slope of the model being equal to 0 across the range of the independent variable.

^dClinical signs for which the ownership similarity would be included in a final multivariable model.

Two PRRSV vaccines were reported to be used in the herds that participated in the study, Ingelvac PRRS ATP in 15 herds and Ingelvac PRRS MLV in 61 herds. After adjusting for ownership, use of the ATP vaccine was associated with a 3.4% higher similarity of the isolated virus with the known vaccine strain than in the herds that did not receive the vaccine (P < 0.001). No association was detected between use of the MLV vaccine and isolated virus type.

From each model which the smooth terms had associated P-values of less than 0.05 and a region where the point estimate crossed outside of what was expected by chance, 162 observations were removed. These observations represented the 1% most influential observations or the 1% greatest outliers. When these were removed, the overall shape of the models did not change for the clinical signs: abortion, sows being off feed, stillborn piglets, weak-born piglets, preweaning mortality, and nursery respiratory disease. The point where the 95% CI crossed outside of what was expected by chance also did not change for these models. However, the model for sow/boar mortality did change: the negative slope in the region of less than 82% sequence similarity indicated that an increase in sequence similarity resulted in a decrease in the probability that 2 herds had a similar clinical experience; this was also found when the 95% simulation envelope was produced for the model (Figure 2). When the top 1% of observations based on their value of Cook’s distance were removed from the model the negative slope in the region of less than 82% sequence similarity became shallower, and the peak in the model prediction was at 0.5 logit compared with 1.2 logit in the model with all of the data. This indicated that a small number of observations were responsible for these results. Without the most influential observations the model prediction was greater than would be expected by chance alone in the region of more than 98% sequence similarity.

The results of GAMs using normal approximation did not differ from the interpretation of models using simulation analyses. This may be because of a lack of autocorrelation in the sequence similarity matrix data considered likely to be present (27). Such autocorrelation might arise if the variation in observations within a row or column in the similarity matrix was lower than the variation between rows or columns. As an example, a single virus that was very different from the rest of the viruses would result in a sequence similarity matrix with low variability within a row and a column but increased between-row variability. The use of simulated null envelopes to address the potential model violations is an important final step to ensure that the results are valid.

Multivariable GAMs to test the association between clinical similarity and percent sequence similarity accounting for herd ownership were completed for each of the clinical signs: sows being off feed, nursery mortality, and nursery respiratory disease because both univariable models indicated an association (Table III). One multivariable model was found to have significant associations of both ownership similarity and nursery respiratory disease similarity with percent sequence similarity. This model indicated that herds with similar ownership had 5.7 times higher odds of reporting similar clinical signs with respect to respiratory disease in the nursery (P = 0.005) and that the association for sequence similarity had become essentially unimportant but had a significant smooth term (P = 0.01) (Figure 3).

Figure 3.

Example of GAM predictions and CIs for the relationship between sequence similarity and similarity of nursery respiratory disease when ownership similarity is included in the model.

The distributions of sequence similarities among wild-type and vaccine-like viruses are presented in Table IV. The association between percent sequence similarity and ownership similarity is shown in Table V. The univariable associations between ownership similarity and clinical similarity and between percent sequence similarity and clinical similarity are described in Table III.

Table IV.

Distribution of percent sequence similarity of open reading frame 5 of the PRRSV genome among wild-type viruses and among virus isolates similar to the 2 modified-live-virus strains used for vaccination in Ontario, contained in the Boehringer-Ingelheim Ingelvac PRRS ATP and Boehringer-Ingelheim Ingelvac PRRS MLV vaccines

		Similarity (%)
Virus strain	Number of isolates	Mean	Median	5th percentile	95th percentile
Wild type	176	87.2	86.6	82.9	94.8
Ingelvac ATP	3	97.9	97.7	97.4	98.5
Ingelvac MLV	22	97.9	97.8	97.1	98.8

Table V.

Correlation from the Mantel test and association from linear regression between ownership similarity and genetic similarity

Production phase	Pearson correlation coefficient	P-value for Mantel test	Intercept for linear regression	Slope	P-value for slope
Herds with sows	0.05	0.001	87.6	6.03	< 0.0001
Herds with nursery pigs	0.11	0.001	87.6	7.53	< 0.0001
Herds with finisher pigs	0.06	0.002	87.4	3.50	< 0.0001
All herds	0.12	0.001	87.7	4.59	< 0.0001

Discussion

Viruses that are genetically similar resulted in similar clinical disease for 6 of the clinical signs asked about in this study. In a previous investigation, some RFLP types of PRRSV were associated with the presence of specific clinical signs of disease in herds (22). This indication that genotype plays a role in clinical disease led us to hypothesize that sequence similarity and clinical similarity would be strongly correlated. However, the correlation coefficients from the Mantel test were small, consistently less than 0.1, indicating a weak-positive correlation between the similarity in PRRSV sequence and clinical similarity (Table II). These results led us to believe that a potential association between sequence similarity and clinical similarity might not be consistent throughout the entire range of sequence similarity in the dataset and prompted the deeper investigation of the data with the GAM.

At first glance, the findings illustrated in Figure 1 seem to contradict the suggestion that very similar viruses have the most similar clinical presentation. However, once the vaccine-like viruses were removed from the analyses the model for abortion appeared very similar except for the slope continuing to be positive up to 100%. The model for similarity of stillbirth became less convoluted, and the region close to 100% sequence similarity showed significantly increased odds of clinical similarity. It was hypothesized that this change was due to comparison between 2 vaccine-like viruses that were very similar but not clinically similar (Table IV) because the clinical signs of disease in these herds are not thought to be caused by the vaccine-like strain that was recovered from them. The disease may have been caused by another, undetected PRRSV strain or a different pathogen.

A previous observational study measured similarity in the clinical signs: abortion, sow mortality, preweaning mortality, respiratory disease in nursery pigs, respiratory disease in grower/finisher pigs, and longer time to market for grower/finisher pigs (15). The investigators found significant correlation between similarity of sow mortality reports and similarity of PRRSV ORF5 by use of the Mantel test. In contrast, the present study found a nonlinear association (Figure 2), and therefore the Mantel test could not detect significant correlation (Table II).

The univariable GAM used for testing the association between similarity in nursery respiratory disease and percent sequence similarity resulted in a significant smooth term (Table III). However, the predictions of the model illustrate that this association cannot be easily interpreted (Figure 2). The null hypothesis of the test is that there is no association between the variables anywhere in the range of the independent variable (28). Because of the very broad potential for rejecting the null hypothesis, we defer to the simulated CIs, which tell us that there is no association between PRRSV ORF5 similarity and similarity of reported respiratory disease in the nursery. This conclusion is reinforced by the results from the multivariable model of the association of nursery respiratory disease similarity and sequence similarity with ownership similarity. The smooth term indicates little or no change in the odds of similarity of nursery respiratory disease with increasing percent sequence similarity (Figure 3). Pairs of herds with the same ownership were more likely to be clinically similar than those with different ownership up to a sequence similarity of 98%. At greater sequence similarity ownership was no longer important. This indicates that ownership is more important in predicting similarity in reports of respiratory disease in the nursery unless the 2 herds have highly similar viruses, in which case similarity in ownership does not play a role.

The similarity in the reports of sows being off feed and of weak-born piglets was associated with the outcome of sequence similarity (Table III). The GAM indicated that these relationships were linear and had the anticipated positive slope; however, both models had a region of high genetic similarity that crossed into the null envelope (Figure 2). The data do not offer strong evidence for these clinical signs, and these findings are interpreted as only a tendency for the clinical sign to be associated with the genotype of the virus.

The results for the similarity of reports of preweaning mortality indicated a general positive trend across the range of sequence similarity. However, the GAM illustrated that the slope of the line became negative for isolates that were more than 96% similar. This is weak evidence that genotype is related to preweaning mortality in a herd.

Ownership in general was found to play an important role in sequence similarity, as reported in Table V. This was most noticeable in the herd pairs that included sows or nursery pigs and less marked in herds with finisher pigs. Herds with finisher pigs may have more unidirectional animal flow with respect to the rest of an ownership structure in order to not pose as great an infection risk to the system. This may be why two herds with finisher pigs from the same owner had viruses that were 3.5% more similar than herds from different ownerships, while in herds with sows or nursery pigs from the same owner the virus similarity was 6% to 7% higher than those from different owners. The multivariable model for nursery respiratory disease with ownership similarity and sequence similarity (Figure 3) also illustrated the relative importance of ownership in clinical disease. The model indicated that, although both variables were significant, the slope was essentially flat for the effect of sequence similarity. Ownership appeared to be more important than type of virus in the herd. Ownership similarity is a proxy for factors related to management that may, therefore, in turn be related to clinical disease in a herd.

With regard to study limitations, the survey was completed in such a way as to ask the producers whether they considered each clinical sign to be a “problem” in the herd. The results suggest that viruses that are more similar are associated with more similar clinical disease, although not for all clinical signs associated with PRRS. The clinical signs that were associated are also those that tend to be the most dominant and easy to identify in an outbreak. Abortion, stillbirth, and preweaning mortality may be subject to less interpretation about how a problem is defined as opposed to respiratory disease in any of the production phases. In a previous study of the association between clinical similarity of PRRS and similarity of the PRRSV ORF5 gene (15), the information about clinical presentation in the herds was collected from the herd veterinarian, which may have resulted in less variation in the interpretation of the presence or absence of a clinical sign in a herd compared with interviewing a producer because a veterinarian has the benefit of a larger frame of reference, having visited many herds with varying clinical conditions. Herd managers, as opposed to veterinarians, were surveyed in the present study because the managers had daily contact with the animals in the herd and a clear knowledge of both clinical disease and other production practices that were part of the survey. The herd managers were also chosen for their ability to focus on what they considered to be the most important feature of the particular outbreak of PRRS that the virus isolate represented. This may, however, have resulted in recall bias toward the most memorable events of a PRRSV outbreak, especially when substantial time had passed between the outbreak and the interview. Using production records to assess the clinical severity of a PRRSV outbreak would have reduced the variability in interpretation of the clinical picture among herds. Production records were not chosen for the present study in order to include all possible herd types. Because herds with detailed production records may be managed differently from those without, including only the former in the study would have biased the results.

The difference between inclusion and exclusion of vaccine-like viruses from the data indicates that vaccine virus in a population can greatly affect the interpretation of the epidemiologic features of a disease. Many of the herds in the present study group used an attenuated live PRRSV vaccine. Only a single PRRSV isolate was collected from each herd, so if a vaccine-type virus was identified, it was the only virus that could be attributed to the clinical disease with this dataset. However, it may be that another PRRSV type was present in the herd that resulted in the clinical outbreak that prompted the sample submission to the laboratory. The lack of association between clinical similarity and sequence similarity for herds with vaccine-type viruses is an indication that the vaccine-like viruses were not causing the reported disease in these herds. In the present study it was not feasible to sequence more than 1 virus from each herd.

The presence of other pathogens in the herds in this study was unknown to us and could have been responsible for some of the reported clinical signs. As is true of all observational studies, it is therefore important that the present results be interpreted as an association, and not a causative link, between ORF5 genetic similarity among herds and clinical similarity.

The use of Boehringer-Ingelheim Ingelvac PRRS ATP was associated with recovery of a similar virus. The use of vaccines in the study population could have biased the results toward the recovery of vaccine virus over the recovery of wild-type virus, especially if vaccine was used in response to the outbreak of PRRS. This could have resulted in our missing wild-type viruses in some herds from which vaccine virus was recovered.

In conclusion, this research indicates that PRRSV sequence similarity is related to clinical similarity at the herd level. This association was found for clinical signs in the sow barn and was detected in the range of sequence similarity from 92.5% to 100% among wild-type viruses. This finding agrees with the results of previous work indicating that clinical PRRS is variable by PRRSV genotype. Because a relationship between clinical similarity and sequence similarity was not found for vaccine-like viruses it is thought that these are not frequently causing clinical PRRS in Ontario. The present results are based on data collected from field cases of PRRS and herd-level clinical presentation of PRRS. This is in contrast to previous research that has focused on comparing the clinical disease in individual animals that were experimentally infected. Although the ability to characterize clinical disease at the herd level from a diagnostic sample is not as precise as would be possible in an experimental study, the present results offers insight into the way PRRSV behaves in the field that complements the previous experimental results. Future work focused on the clinical variability of PRRS by PRRSV genotype should focus on prospective investigations of PRRSV-negative herds with ongoing collection of production data as part of the research. When outbreaks of PRRS occur, one could be more confident that the clinical signs and production losses are due to the outbreak of a specific genotype of PRRSV. This would also allow for more specific assessment of the production losses from and the longevity of an outbreak.