Forty-eight farms in Ontario were conveniently selected and visited once or twice between two periods of sample collection from August to December 2010, and from January 2011 and May 2011. For the first period of sample collections, 28 farms were visited. One farm was visited twice in the first period, and visited again in the second period. In the second period, a total of 11 farms from the first period were visited again, and 20 farms were visited for the first time. Some farms that were visited from the first period were not re-visited because they did not have suckling pigs or they were no longer in business.
In both periods of sample collection, for farrowing operations, pooled-fecal samples were collected from gestating sows, weanling pigs, and grower-finisher pigs. In addition, a pooled sample was taken from the manure pit. In the second period of sample collection, fecal samples were also collected from lactating sows and their litters. A lactating sow fecal sample was collected by randomly selecting a sow, and manually obtaining a rectal sample with a gloved hand. A pooled-fecal sample of the lactating sow’s litter was also obtained by selecting three suckling pigs in the litter and collecting the feces from the piglet’s rectum. If suckling pigs did not defecate sufficient material during the sampling process, fresh piglet feces were obtained from the farrowing area of the litter. Three fecal samples were collected from lactating sows, and three pooled fecal samples were collected from their respective litters from each farrowing operation.
A pooled-fecal sample of gestating sows was obtained by selecting 6 sows from different areas of the barn, and obtaining a rectal sample. A pooled-fecal sample of weanling pigs or grower-finisher pigs was obtained from randomly selecting 6 pens, observing the pigs defecating, and then immediately collecting the fresh feces from the floor of the pens. On farrow-to-feeder farms sampling was similar to farrow-to-finish farms except two pooled samples were collected from weaning pigs. On grower-finisher farms, 3 pooled-fecal samples were taken from the pigs present, as well as a pooled manure pit sample. All farms on the study used a liquid manure system and a sample from the holding tank or manure pit was obtained by filling a plastic bottle from the manure pit. Liquid manure from 3 locations in the pit and at 3 different depths was combined to create the pooled sample.
A survey was administered to collect information on each farm for different management factors including the following: type of operation (farrowing-to-finish, farrowing-to-feeder, or grower-finisher operation), total number of barns, herd size (number of sows, gilts, weanling pigs, and grower-finisher pigs), sow flow, weanling flow, grower-finisher flow (all-in/all-out or continuous), number of farrowing rooms and crates, the presence of other agricultural species (other species), drug usage and vaccination, history of diarrhea outbreak in suckling pigs or post-weaning pigs in the past year, and treatment of the diarrhea problem. A total of 46 farms were surveyed; two farms were not surveyed because they were no longer in business.
For bacterial culture, samples were weighed and 4–5 serial 10-fold dilutions were performed in phosphate buffered saline, pH 7.2. Fecal dilutions were plated onto selective SFP (Shahadi Ferguson Perfringens) medium (Difco, Detroit, MI, USA). The SFP media contains 5% egg yolk emulsion (Oxoid, Nepean, ON), 12 μg/ml kanamycin sulphate, and 30 IU/ml polymyxin B sulphate (SFP Selective Supplement, Oxoid). The plates were transferred to an anaerobic jar, with anaerobic atmosphere provided by GasPak (BD, Sparks, MD, USA) and incubated at 37°C overnight. Colonies with characteristic C. perfringens colonial and microscopic morphologies, as well as lecithinase activity on the egg yolk agar, were counted and counts converted to colony-forming units (CFU) per gram.
Genotyping of isolates
Clostridium perfringens colony lysates were examined by multiplex polymerase chain reaction (PCR) for toxin genes using the method described by Albini and colleagues . Five isolates per sample were genotyped for cpa, consensus cpb2 (cpb2) and atypical cpb2 (atyp-cbp2). The isolates recovered during the first period of the study were also genotyped for the major toxin genes beta toxin (cpb), enterotoxin (cpe), epsilon toxin (etx), and iota toxin (itx), NetB toxin (netB) and large clostridial cytotoxin (tpeL). Since these were all found to be absent they were not examined in the second period. The primer sets for the major toxin genes and cpb2 used were designed using the reported nucleotide sequences . The primer sets for netB, tpeL, and atyp-cpb2 were designed for this study using standard approaches and appropriate bacterial control strains. The Roche LightCycler® 480 SW 1.5 software package was used to analyze the data. Relevant positive control strains were included in all batched PCR tests.
The data were entered in a spreadsheet (Microsoft Excel 2007; Microsoft Corp., Redmond, WA) and imported to Stata 10 Intercooled for Windows XP (StataCorp LP, College Station, TX) for statistical analysis.
Three outcomes were measured, which included C. perfringens type A (CpA) count in fecal samples, the presence of cpb2-carrying C. perfringens type A (cpb2-CpA) isolates in fecal samples, and the presence of atyp-cpb2-carrying C. perfringens type A (atyp-cpb2-CpA) isolates in fecal samples. The independent variables included in multivariable analysis were stage of production (main effect), type of operation, total number of barns, herd size (small or <1000, medium or ≥ 1000–3000, and large or >3000), and the presence of other species.
A mixed linear regression modeling method with farm as a random effect was used to analyze the association between CpA count in fecal samples and independent variables. Intra-class correlation coefficient was calculated as
. In addition, two separate logistic regression models were fitted to investigate the association between the presence of cpb2-CpA and atyp-cpb2-CpA in fecal samples and independent variables. Intra-cluster coefficient was calculated as
by assuming that sample level variance on the logit scale was π2/3, π = 3.1416.
Univariable analysis of independent variables and their association with CpA count, and presence of cpb2-CpA and atyp-cpb2-CpA isolates in fecal samples was performed using a single linear or logistic regression. Variables with P <0.20 were selected for inclusion in multivariable analyses which was modeled at the sample level with farm as random variable.
A manual stepwise procedure was used to build the multivariable models. Pair-wise correlation coefficients were calculated between independent variables, and coefficients with an absolute value greater than 0.8 were considered colinear. If two independent variables were significantly correlated, the more informative variable was used in the final model. A variable was identified as a confounder if it changed the coefficient of the main effects by 20% or more when the potential confounder variable was removed. If a variable was determined to be a confounder, it was included in the final model regardless of its statistical significance. Interaction was evaluated between all independent variables and the main effects model. Each interaction term was assessed for statistical significance with the main effects, and the interaction terms with P <0.05 were included in the final model. Interaction terms that were not significant in the final model were removed if removal of the interaction term did not result in a significant change in the likelihood ratio test (logistic regression) and partial F-test (linear regression).
For linear multivariable analysis, a log10 transformation was used on the C. perfringens count (CFU/g) in order to meet the normal distribution and homoscedasticity assumptions for the outcome of interest. The Cook-Weisberg test was performed for the linear multivariable model with farm as a fixed effect, and the assumption of homoscedasticity was met if P >0.05. The Shapiro-Wilk normality test was used to test the normal distribution of log10 CFU/g with farm as a fixed effect, and the assumption of normality was met if P >0.05. The standardized residuals were graphically assessed for outliers, and the leverage values were graphically assessed.
The logistic regression multivariable model with farm as a fixed effect for the presence of cpb2 in C. perfringens type A isolates was assessed for goodness-of-fit, which was indicated by P >0.05 in the Hosmer-Lemeshow test. Pearson residuals were graphically assessed for outliers, and the leverage values were graphically assessed.
The Spearman’s rank correlation test was used to analyze the association between the toxin genes in lactating sows and their litters.