1-s2.0-s0168160506000195-main.pdf

biology 108 (2006) 246–254www.elsevier.com/locate/ijfoodmicro

International Journal of Food Micro

Use of a serological approach for prediction of Salmonella statusin an integrated pig production system

Nicolas Korsak a,⁎, Jean-Noël Degeye a, Grégory Etienne a, Jean-Marie Beduin b, Bernard China a,Yasmine Ghafir a, Georges Daube a

a University of Liege, Faculty of Veterinary Medicine, Food Science Department — Microbiology Section, Sart-Tilman, Bât B43, 4000 Liege, Belgiumb University of Liege, Faculty of Veterinary Medicine, Animal Production Department — Biostatistic, Economy and Animal Selection, Sart-Tilman,

Bât B43, 4000 Liege, Belgium

Received 13 July 2005; received in revised form 19 September 2005; accepted 30 September 2005

Abstract

Relevance of a Salmonella serological detection technique was studied from complete results obtained from 9 pigs fattening units. Feces andovershoes were sampled at different periods after starting fattening (2, 3 and 4 months) while caecal contents were taken on the slaughter line. Thebacteriological technique used was based on a Diasalm enrichment and a commercial test was used for serology on an average of ten animals perbatch. The aim of this work was to establish a correlation between serological results obtained at slaughter (10 samples/batch) and bacteriologicalresults. In this context, two types of logistic regression models were tested by considering alternatively serology and Salmonella detection in caecalcontents as the dependent variables. Firstly, beside the fact that all logistic regression models show weak correlations, the first finding was thatpositive results in overshoes taken at 2 and 3 months are slightly correlated with serological status of herds (odds-ratios of 4.96 and 2.55). Secondly,when batches were characterized as positive on the basis of serological results, the probability of Salmonella recovery in caecal contents was higherthan when the batches were considered as negative (odds-ratios comprised between 4.36 and 5.81).

A major conclusion is that serology can be used to follow the improvement of an integrated pig production system, but is not the unique solutionfor assessing risk of Salmonella shedding from specific herds.© 2006 Elsevier B.V. All rights reserved.

Keywords: Salmonella; Swine; Pigs; Serology; Logistic regression models; Caecal contents

1. Introduction

Salmonella is still amongst the leading causes of foodborneoutbreaks in industrialized countries. This was clearly demon-strated worldwide (Daube and Van Loock, 1997; Mead et al.,1999; Haeghebaert et al., 2002; Wegener et al., 2003). Besideimportance of this micro-organism for public health, anotheraspect is the cost generated by human salmonellosis. Aworkingdocument of the European Commission estimated that costslinked to foodborne salmonellosis ranged between 560 millionsand 2.8 billions € in Europe, where Salmonella was estimatedto be responsible of nearly 166,000 cases in 1999 (Anonymous,

⁎ Corresponding author. Tel.: +32 4 366 40 40; fax: +32 4 366 40 44.E-mail address: [email protected] (N. Korsak).

0168-1605/$ - see front matter © 2006 Elsevier B.V. All rights reserved.doi:10.1016/j.ijfoodmicro.2005.09.013

2001). Except for few countries, like Sweden, that maintain andcontrol Salmonella at a very low level in the pre-harvest stage,most of others initiate new control strategies in order toeradicate this zoonotic agent. These actions will be mandatoryin the future and especially in Europe, where several laws wererecently promulgated. In this context, directive 92/117/EEC wasrepealed and replaced by directive 2003/99/EC on the mon-itoring of zoonoses and zoonotic agents (Anonymous, 2003a).Furthermore, at the same time, a specific regulation for thecontrol of Salmonella and other specified food-borne zoonoticagents was voted (Anonymous, 2003b). This latest Europeanregulation requires that each member state set up a target fordifferent animal species and mandates that testing starts in amaximal delay after setting up targets. For example, in fatteningpig herds, for control of all Salmonella serotypes significant for

247N. Korsak et al. / International Journal of Food Microbiology 108 (2006) 246–254

public health, the target has to be established before December2007 and testing, 18 months thereafter (June 2009).

For this purpose, national authorities and researchers needfast and reliable methods for the detection of Salmonella at eachstep of the production chain, and especially during the pre-harvest stage (breeding and fattening). During the last decade,serological tests have been newly developed and marketed. Forexample, serology is used on a national scale in Denmark, wherethis program was initiated after an outbreak in 1993, caused bypork meat contaminated at the farm level (Nielsen et al., 1995;Mousing et al., 1997). This program has demonstrated its ef-ficiency by decreasing significantly the human cases of pork-related salmonellosis (expressed per 100,000 inhabitants): from22 in 1994 to 3 only in 2001 (Wegener et al., 2003).

However, despite great benefits of using serology like lowcost and high throughput, Nielsen et al. (1995) showed thatproblems may not be entirely solved by this technique. Amongthese, the variability in response of contaminated pigs and thelack of persistence of serological response can be reported.

The aim of the present work was to try to correlate serologicalresults obtained by sampling of muscles taken on the slaughterchain and bacteriological results obtained in farms and fromanalyses of caecal contents taken on the slaughter line.

2. Materials and methods

2.1. Animals, samples and bacteriological protocol

Nine herds were followed from a period ranging from April2001 to July 2002. The production system, sampling schemeand the analytical detection protocol for Salmonella werepreviously described (Korsak et al., 2003). In summary, theanalytical protocol lies on the use of Diasalm (Lab 537, Lab M,International Diagnostic Group PLC, Lancashire, UK), semi-solid medium close related to modified semi-solid Rappaport–Vassiliadis (MSRV), that allows detecting low levels ofSalmonella among important levels of competitive micro-org-anisms (De Smedt et al., 1986). Indeed, the formula of Diasalmhas proven to be superior for Salmonella recovery, especially inpoultry meat, than the classical Rappaport–Vassiliadis broth(Landman et al., 1996; De Zutter and Daube, 1998).

A non selective pre-enrichment was incubated during 24 h at37 °C after adding 225 ml of Buffered Peptone Water (BPW,CM509, Oxoid, Basingstoke, United Kingdom) to 25 g samplefecal matter (except for caecal contents where 5 g was used anddiluted in 45 ml of BPW) or after pouring 300 ml BPWonto theovershoes in the stomacher bag. Pre-enrichment was followedby the transfer of 100 μl (corresponding to three drops) on thecenter of a Diasalm Petri dish, which was then incubated at42 °C for 24 h. Isolation was achieved from the typical mi-gration zone on Diasalm onto xylose–lysine–tergitol 4 agarplate (63654, Biorad, Marnes La Coquette, France). This latestwas then incubated at 37 °C for 22 h. After this period,purification of characteristic colonies was operated by transfer-ring one colony in the bottom and on the slant of Triple SugarIron slants (tv5074d, Oxoid, Wesel, Germany) incubated at37 °C for 24 h. If the appearance on this latest could lead to

suspicion of the presence of Salmonella, confirmation was thenachieved by means of biochemical methods (e.g., API 20E,bioMérieux, Marcy-l'Etoile, France).

This study followed the fattening period and the slaughteringprocess of 59 batches of fattening pigs produced from 9 herds. Onaverage, 180 to 200 finishers composed each batch. For bac-teriological analyses on live animals, only fecal matter andovershoes sampled during the fattening period (after 2, 3 and 4months of fattening) were taken into account in order to establishthe matching between status in farm and serology at slaughter andbacteriological response in caecal contents at slaughter.

For each batch, the caecal contents were taken asepticallyfrom an average of 5 animals (min.: 4; max.: 8) on the slaughterchain just after the evisceration step. They were individuallytreated for analytical purpose.

All samples (on-farm and at slaughter) were put in an in-sulated box kept cold with ice, in order to avoid temperatureabuse and transported to the lab, were they were analyticallyprocessed in a delay not exceeding 12 h.

2.2. Serology

For serological examination, the same sampling procedurethan that used in the “Danish Surveillance Program” wasoperated (Mousing et al., 1997; Christensen et al., 1999). Cubesof approximately 3 cm of edge of diaphragmatic muscles weresampled on the slaughter line at the post-mortem inspection step.Antibodies quantification was performed on an average of10 samples (minimum: 5; maximum: 20 sampled animals fromthe batch) of meat juice collected, after freezing (−20 °C) andthawing, in a “meat juice collector” (Salmostore, LaborDiagnostik, Leipzig, Germany). This method was validated byNielsen et al. (1998), who showed that serological analysismuscle fluid is a useful alternative to serum analysis (relativespecificity of 0.91–1 and relative sensitivity of 0.80–0.89). Fivesamples out of the 10 samples for serology corresponded exactlyto the same animals used for the five caecal contents samplesintended for bacteriological analyses.

Serology was executed by using a commercial kit (Salmo-type® Pig LPS ELISA test, Labor Diagnostik, Leipzig, Ger-many), composed of six purified lipopolysaccharide fractionsthat enable detection of 90% of common serotypes. This test isspecific for serovars belonging to serogroups B, C1 and D1(Nielsen et al., 1995).

The manufacturer of this test provides five controls, thatcontains a pre-defined relative concentration of antibodies (0.1%for negative control, 18.5%, 33.9%, 55% and 79.7% of relativeantibody concentrations), allowing the setting up of a regressionline that explained the relationship between optical densitiesfrom the spectrophotometer and estimated level of antibodies inthe meat juice samples. Before processing, 30-fold dilutions ofeach pork meat juice was performed.

2.3. Statistical analyses

According to a previous study performed by analyzing com-plete results from 91 batches, for interpretation of serological

248 N. Korsak et al. / International Journal of Food Microbiology 108 (2006) 246–254

results and to enable evaluations of agreement parameters, thesame following cut-off values were used (Korsak et al., 2004).The serological cut-off values studied were OD10%, OD20%,OD30% and OD40% (optical density corresponding to anantibody concentration of 10%, 20%, 30% and 40%, respec-tively). For the batch cut-off, only the limit of 50% was takeninto account. Indeed, the aforementioned preliminary study hasshowed that a better concordance between serological status ofbatch at slaughter and bacteriological results for the fatteningperiod was obtained at this cut-off value of 50%.

The relationships between the categorization of batchesaccording to serology and Salmonella detection in caecal con-tents were studied. To do so, the 59 batches were classified in afourfold table in function of different situations encountered:number of batches where 50% of sampled pigs had an OD valuelower or higher than OD corresponding to 4 different antibodyconcentrations (10%, 20%, 30% and 40%) and number ofbatches where Salmonella was detected in caecal contents atabattoir (considered positive when at least 1 isolation has oc-curred from the caecal contents sampled on the slaughterline).The determination of agreement between tests was measuredthanks to the Kappa parameter, which was obtained with theequation:

j ¼ ð po− pcÞ1−pc

where po is the overall proportion of agreement obtainedbetween the tests to be evaluated while pc corresponds to theoverall proportion of agreement expected by chance (Fleiss, 1973).

When the ratio between the kappa value and its standarddeviation exceed the threshold of 1.96, the kappa value wasconsidered significantly different from zero (Pb0.05). The twoother statistics used were Chi-square and McNemar, which wereconsidered significant at level Pb0.05.

The formula for calculation of McNemar values, whose in-terest is to test the identity between false-positive and false-negative results, is the following (Wiberg and Noberg, 1996):

v2 ¼ ðja−bj2−1Þaþ b

where a is the number of false-negative results (samples positivefor bacteriological results from caecal contents but negative forserological examination) and b is the number of false-positiveresults (opposite situation).

For the choice of the better classification system in functionof the serological cut-off, the Chi-square test value need to besignificant, while McNemar not, in order to detect equally false-positive and false-negative results. To evaluate the serology asan alternative method, relative sensitivities (Serel) and relativespecificities (Sprel) were calculated considering the bacteriolo-gical technique (Salmonella detection in caecal content) asreference. Moreover, Youden's index (Serel+Sprel−100), thatpermits to view the better compromise between the two para-meters, was also assessed.

On a herd level, to evaluate the relationship (expressed withodds-ratio — OR) between the on-farm bacteriological results

and serology at slaughter or Salmonella detection from caecalcontents, logistic regression models were performed for differentstages of the fattening period: 2, 3 and 4 months. In the differentlogistic regressionmodels, serologywas considered positive if atleast 50% of sampled animals had an OD value higher thanOD20%. The software package used was R, a language spe-cifically developed for data analysis and graphics (Ihaka andGentleman, 1996). The dependent variable was alternatively theserology and the bacteriological result of caecal contents, whileother variables were considered as independent. Only the sta-tistically significant logistic regression models were retained inorder to calculate the OR. For this purpose, maximum likelihoodratio test, Wald Statistics (not shown) and pseudo R2 ofNagelkerke were used (Hosmer and Lemeshow, 1989; Bouyeret al., 1995).

For comparison of the distributions of the relative antibodyconcentrations between batches positive and negative in caecalcontents, the non-parametric Mann–Whitney test was used inorder to compare the two medians, given that the relative anti-body concentrations were non-normally distributed.

Finally, the herd seroprevalence (i.e. percentage of samplesabove the serological cut-off of OD20%) was plotted against thepercentage of positive samples in caecal contents.

3. Results

Table 1 shows the observed relationship between serologicalcategorization of batches and Salmonella detection in 5 g caecalcontents. Chi-square and McNemar values were statisticallysignificant for the first classification system (batches classifiedas in function of sampled pigs lower or higher than OD10%).Conversely, McNemar values were not significant for the threeother categorizations, meaning that there were no statisticaldifferences between false-positive and false-negative results(numbers of batches where at least 50% of pigs had OD valueshigher than the selected serological cut-off values but negative incaecal contents versus numbers of batches where at least 50% ofpigs had OD values lower than the selected serological cut-offvalues but positive in caecal contents). For all the serologicalclasses evaluated, the percentages of non-agreement values werehigh, ranging from 27% to 42%. For classification systems withMcNemar values not significant, a significant chi-square valueof 8.5 (Pb0.001) was noticed with classification relative toOD20%, meaning that, at this serological cut-off value, therelationship between the positive serological character of a batchand the probability of Salmonella detection in caecal contents isoptimized, although not perfect.

The Kappa values, ranging from 0.19 to 0.37, were sig-nificant for the two first serological classifications of batches(OD10% and 20%). The relative sensitivities were comprisedbetween 40% and 80%. The value of 66.7% was obtained whenthe batch classification was realized in relation to a serologicalcut-off of OD20%.

Fig. 1 reveals the association between seroprevalences(expressed in classes of percentages of samples above the cut-off of OD20%) and the recovery of Salmonella from caecalcontents. Despite the fact that there is no clear statistical

Table 1Relationship between categorization of herds and Salmonella detection in caecal contents

Salmonelladetection in caecalcontents (5 g)

% of non-agreement— Kappa values

Serel (CI)+Sprel (CI)−100=Youden's index2

Chi-squarevalue

McNemarvalue

Negative Positive1

Number of batches where b50% ofpigs had an ODN10%

22 3 42.4 — 0.21⁎ 80.0(59.8–100)+50.0(35.2–64.8)−100=30.0 4.12⁎ 12.96⁎⁎⁎

Number of batches where at least50% of pigs had an ODN10%

22 12


33 5 27.1 — 0.37⁎⁎ 66.7(42.8–90.5)+75.0(62.2–87.8)−100=41.7 8.47⁎⁎⁎ 1.56


11 10


35 9 30.5 — 0.19 40.0(15.2–64.8)+79.5(67.6–91.5)−100=19.5 2.25 0.06


9 6


37 9 27.1 — 0.25 40.0(15.2–64.8)+84.1(73.3–94.9)−100=24.1 3.78 0.06


7 6

OD: Optical density. C.I.: 95% confidence interval. Serel: relative sensitivity in %. Sprel: relative specificity in %.⁎pb0.05 ⁎⁎pb0.01 ⁎⁎⁎pb0.001 (not significant if nothing mentioned).1For bacteriology, a batch is considered positive if at least 1 caecal content has led to Salmonella detection, for serology, a batch is positive if at least 50% of sampled pigs hasan OD value higher than the serological cut-offs. 2The relative sensitivities and specificities were calculated with the bacteriological technique considered as reference.


relationship between the two variables, this figure indicates thattwo groups of classes can be outlined: below 40% ofseroprevalence, the mean percentage of Salmonella recoveryfrom caecal contents is 3.5% (7 positive out of a total of 200),while, above this limit, this value is 18.5% (20 positive samplesout of a total of 109) (data not shown).

0,0%

15,0

3,3%1,8%

4,8%

0%

9%

18%

27%

36%

45%

0-10 (20--105) 11-20 (10--55) 21-30 (6--30) 31-40 (2--10) 41-50 (4--2

% of samples above OD20% (nb of

% o

f p

osi

tive

sam

ple

s in

cae

cal c

on

ten

ts

Fig. 1. Relation between percentage of samples above OD20% and percentage of pIC 95% (upper limit).

Based on results derived from Table 1 and as described in theMaterials and methods section, for the different logistic regres-sion models, serology was considered positive if at least 50%of sampled animals had an OD value higher than OD20%;otherwise, serology was considered negative. When serology isthe dependent variable, the results of the logistic regression

0,0%

25,0%

27,6%

13,3%13,3%%

0) 51-60 (3--15) 61-70 (2--10) 71-80 (2--15) 81-90 (4--20) 91-100 (6--29)

batches--nb caecal contents analyzed)

ositive caecal contents. The plain lines above the histogram bars represent the

Table 2Logistic regression models of serology in function of herds and of bacteriological results obtained during the fattening period

Coefficients Odds-ratios of the different logistic regression models (CI)

Model 1:logit serology=α+∑β herd+γfeces2months



Model 4:logit serology=α+∑β herd+γovershoes2months



α – – – – – –β2 0.16 (0.08–0.32) 0.14 (0.07–0.28) 0.13 (0.06–0.25) 0.23 (0.11–0.46) 0.20 (0.10–0.39) 0.14 (0.07–0.28)β3 0.11 (0.03–0.42) 0.08 (0.02–0.30) 0.09 (0.02–0.32) 0.16 (0.04–0.59) 0.12 (0.03–0.41) 0.10 (0.03–0.37)β4 0.05 (0.02–0.16) 0.04 (0.01–0.12) 0.04 (0.01–0.13) 0.08 (0.03–0.23) 0.06 (0.02–0.16) 0.05 (0.02–0.15)β5 – – – – – –β6 0.37 (0.17–0.79) 0.27 (0.13–0.57) – – 0.38 (0.18–0.79) –β7 – 0.42 (0.19–0.93) 0.44 (0.20–0.99) – – –β8 – – 0.57 (0.34–0.97) – 0.54 (0.32–0.91) 0.56 (0.33–0.94)β9 0.10 (0.05–0.20) 0.07 (0.04–0.13) 0.07 (0.04–0.13) 0.15 (0.08–0.28) 0.08 (0.04–0.15) 0.08 (0.04–0.15)γ – 0.38 (0.20–0.74) – 4.96 (2.36–10.41) 2.55 (1.50–4.35) –Nagelkerkepseudo R2

0.294 0.309 0.309 0.326 0.316 0.313

–Means that odds ratios were not calculated given that Wald values (not shown) were not significant at Pb0.05 C.I.: 95% confidence interval. α: intercept. β2 to β9:herd component. γ: on-farm bacterial detection component.The “serology” variable was set positive if at least 50% of sampled animals had an OD value higher than OD20% (variable coded as 1); otherwise, serology wasconsidered negative (variable coded as 0).


show the differences in odds-ratio between herds comparedwith the herd 1, chosen by default as reference by the softwareprogram (Table 2). For feces and overshoes collected atdifferent stages of the fattening period, the probabilities ofhaving positive serological results are smaller for herds 2, 3, 4and 9, than for the reference herd. In the remaining herds, thedifferences were not systematically observed for all types ofsamples and sampling periods. For feces sampled three monthsafter starting fattening, a protective effect seems to beobserved, as shown by the odds-ratio lower than 1 (OR:

Table 3Logistic regression models of bacteriology in caecal contents in function of herds a

Coefficients Odds-ratios of the different logistic regression models (CI)

Model 7:logit caeca=α+∑β herd+γfeces2months+δserology



α – – –β2 0.11 (0.01–0.90) – –β3 – – –β4 – – –β5 – – –β6 – – –β7 – – –β8 – – –β9 – – –γ – – –δ 5.04 (1.35–18.78) 4.99 (1.34–18.57) 4.85 (1.30–18.1NagelkerkepseudoR2

0.286 0.289 0.281

–Means that odds ratios were not calculated given that Wald values (not shown) werherd component. γ: on-farm bacterial detection component. δ: serology component.The “serology” variable was set positive if at least 50% of sampled animals had anconsidered negative (variable coded as 0).

0.38; 95% confidence interval —CI: 0.20–0.74). The situationwas opposite for overshoes at 2 (OR: 4.96; CI: 2.36–10.41) and3 months after beginning of the fattening period (OR: 2.55; CI:1.50–4.35). For all these models (models 1 to 6), Nagelkerkepseudo-R2 values ranged from 0.294 to 0.326, meaning that, onaverage, only 30% of the variation was explained by theproposed logistic regression models.

For models trying to explain Salmonella recovery incaecal contents in function of the herds of origin, the bac-teriological detection during fattening and the serological

nd bacteriological results of feces and overshoes collected during fattening

Model 10:logit caeca=α+∑β herd+γovershoes2months+δserology



– – –0.10 (0.01–0.82) 0.10 (0.01–0.85) –– – –– – –– – –– – –– – –– – –0.23 (0.06–0.93) 0.23 (0.06–0.88) –– 0.31 (0.11–0.91) –

2) 5.41 (1.45–20.14) 5.81 (1.58–21.35) 4.36 (1.14–16.67)0.292 0.314 0.286

e not significant at Pb0.05 C.I.: 95% confidence interval. α: intercept. β2 to β9:

OD value higher than OD20% (variable coded as 1); otherwise, serology was

Table 4Summary of bacteriological and serological for the followed herds in farm and at slaughter

Herds (nbof batchesfollowed)

Batches withdetection ofSalmonellain caecalcontents

Salmonella detection in on farm sampling Mean antibodyconcentration (nb)(1)

Salmonellaisolation fromcaecal contents(%)(2)

2 mo 3 mo 4 mo

Feces Overshoes Feces Overshoes Feces Overshoes

1 (15) Batches+: 8 3/8 4/8 1/8 2/8 1/8 3/8 66.9 (76) 54.0 (141) 16/74 (21.6)Batches−: 7 2/7 2/7 0/7 2/7 0/7 1/7 38.9 (65)

2 (7) Batches+: 0 – – – – – – 13.6 (75) 0/40 (0)Batches−: 7 0/7 0/7 0/6 0/6 0/7 0/7

3 (2) Batches+: 0 – – – – – – 15 (20) 0/10 (0)Batches−: 7 0/2 0/2 0/2 0/2 0/2 0/2

4 (5) Batches+: 0 – – – – – – 9.2 (51) 0/25 (0)Batches−: 5 0/5 0/5 0/5 0/5 0/5 0/5

5 (1) Batches+: 0 – – – – – – 8.4 (10) 0/5 (0)Batches−: 1 0/1 0/1 0/1 0/1 0/1 0/1

6 (4) Batches+: 0 – – – – – – 25.3 (41) 0/20 (0)Batches−: 4 0/4 0/4 0/4 0/4 0/2 0/2

7 (3) Batches+: 0 – – – – – – 24.1 (30) 0/15 (0)Batches−: 3 0/3 0/3 0/3 0/3 0/3 0/3

8 (8) Batches+: 5 0/5 0/5 3/5 1/5 0/5 2/5 26.6 (79) 32.3 (109) 8/50 (16.0)Batches−: 3 0/3 1/3 0/3 2/3 0/3 0/3 47.2 (30)

9 (14) Batches+: 2 0/2 0/2 0/2 0/2 0/2 0/2 10.1 (21) 11 (143) 3/70 (4.3)Batches−: 12 0/10 0/10 0/12 2/12 0/12 0/12 11.2 (122)

Total (59) Batches+: 15 3/15 4/15 4/15 3/15 1/15 3/15 42(3) (176) 26.4 (620) 27/309 (8.7)Batches−: 44 2/42 3/42 0/43 6/43 0/42 1/42 20.1(4) (444)

(1)Data expressed in percent (in parentheses: number of total individual values analyzed by for serology).(2)Positive samples for Salmonella detection on total samples investigated (in parentheses: percentage positive sample).(3)Median: 23.1%.(4)Median: 9.3%.


status defined at slaughter, fewer coefficients were significantthan for the above mentioned models (Table 3). However, ineach case, positive serological results at slaughter seem to belinked with a higher likelihood of Salmonella isolation incaecal contents at slaughter (models 7 to 12). Odds-ratios weresystematically higher than 1, with values ranging from 4.36(model 12) to 5.81 (model 11), with the lower border of 95%confidence interval always greater than 1. In contrast, thevariable of Salmonella detection in overshoes at 3 months offattening harbored an odds-ratio significantly inferior to 1 (OR:0.31; CI: 0.11–0.91).

For caecal contents, a mean prevalence of 8.7% (27 positivesamples out of a total samples of 309) was obtained (Table 4).Great differences between the mean Salmonella relative anti-body concentrations and the prevalence in on-farm sampleswere noticed between herds. A relationship between high pre-valence rate of Salmonella in caecal contents and high meanantibodies concentration seems to be observed, especially inherd 1. For all batches positive in caecal contents (n=15), themean relative antibody concentration was 42%, while this valuewas 20.1% for negative batches (n=44) (Table 4). The medianscompared with the Mann–Whitney test were highly significantbetween the two groups (median for batches positive: 23.1%;median for batches negative: 9.3%; z-value=7.43, Pb0.001)(Table 4).

Surprisingly, in herd 8, the mean relative antibody concen-tration was lower for animals sampled in batches whereSalmonella was detected in caecal contents than for batchesthat were negative. For herds 6 and 7, mean antibodies con-

centrations were high (above 20%) but no Salmonella wasdetected in caecal contents.

All isolated serotypes from herds 1 and 8 belonged to theserovar Typhimurium (serogroup B), except in the case ofherd 1 for feces and overshoes at 2 months where Infantis(serogroup C1) was isolated (data not shown). In contrast, inherd 9, a more diverse variety of serotypes was identified:Livingstone (serogroup C1) in overshoes at 3 months offattening, Mbandaka (serogroup C1), Enteritidis (serogroupD1) and Eboko (serogroup C2–C3) from caecal contents(data not shown).

4. Discussion

In this experiment, serology, allowing Salmonella antibodyquantifications, was used. Reference to several countries thatused this method on a routine basis for several years can be usedand serological analyses allow to treat a lot of samples together(Nielsen et al., 1995; Blaha, 2003; Lo FoWong et al., 2003; Hurdet al., 2004). These samples may be collected during theslaughter process or in the farms during the surveillance programof animal diseases (e.g. during control of Aujeszky disease).Moreover, the method may be cheaper than classical bacterio-logical techniques, especially when analyzing many samples.

Despite evidences of a relationship between the serologicalresponse and the bacteriological results obtained in swine herds,there still remain major drawbacks for this type of analysis(Nielsen et al., 1995). Among these, one has to mention theimportant diversity of antibodies responses for animals


contaminated by Salmonella spp., lack in several cases ofpersistence of the serological response and the poor relationshipbetween serological status and persistence of Salmonella insome internal organs. Kranker et al. (2003) have alsodemonstrated that there were important variations in theserological reaction in contaminated piglets for several char-acteristics like the following ones: estimated shedding time, ageat maximum bacteriological prevalence and age at maximumseroprevalence. Another survey conducted in the USA hasshown permanent overestimations of apparent prevalences withserology compared with that obtained from bacteriologicalanalyses of feces, caecal contents and lymph nodes (Hurd et al.,2004).

However, Nielsen et al. (2001) have proven that a connectionexists between the proportion of herds in which Salmonella wasfound in samples of caecum contents and the Danish serologicalSalmonella index calculated on the basis of the past three monthsserological results of the herd. Indeed, the observed values couldbe fit with a regression line with a R2 value of 0.85.

In our study, for logistic regression models, a batch wasassessed as seropositive when at least 50% of results of sampledmeat juices were above OD20% (Korsak et al., 2004). Table 1and Fig. 1 of the present study confirm that this categorization ofherds seems more suitable than the others in order to ascertainthe probability of Salmonella recovery in the caeca at abattoir.This situation maximizes Kappa value and the combination ofrelative sensitivity and specificity. However, with this classifi-cation system, a lot of false-positive (33.3%, i.e. 100−Serel) andfalse-negative (25%, i.e. 100−Sprel) results was noticed.

The choice of OD20% for serological cut-off is partially inaccordance with the lowering adaptation of the serological cut-off in the Danish Salmonella Surveillance program fromOD40% to OD20% (Nielsen et al., 2001). Another survey haspinpointed that herds with high seroprevalence tended to have agreater likelihood of Salmonella isolation from pen samples(Stege et al., 2000). This latter observation was also noticed byLo Fo Wong et al. (2003), who showed a link between thewithin-herd seroprevalence (at cut-off OD10%) and the within-herd bacteriological prevalence, estimated by the probability ofdetecting Salmonella in 20 pooled pen fecal samples. Anotherinvestigation on 1,658 finishers provided by 167 herds showedthat an association exists between the herd seroprevalence(percentage of samples above the cut-off of OD20%) and theprobability of Salmonella isolation from 25 g caecal contents(Sorensen et al., 2004). Depending on the serological cut-offchosen, these authors assessed the Spearman's correlations at avalue comprised between 0.41 and 0.42.

In France, Proux et al. (2000) have also tested a completeELISA system that enables in theory to detect 100% of in-fected pigs. They observed a significant correlation betweenthis serological technique and Salmonella recovery fromlymph nodes.

The major findings of the present study were the following.Firstly, an association was noticed between Salmonella recoveryfrom overshoes sampled at 2 and 3 months of fattening and thepositive serological status at slaughter. Next, the relation be-tween bacteriological results in feces and serology was less

evident, except a slight “protective” effect concerning bacteri-ological results at 3 months of the fattening period. Finally, theprobability of Salmonella detection in 5 g caecal contents wasabout 5 times higher (OR comprised between 4.36 and 5.81)when the batch was characterized positive by serological ex-amination than if the batch was negative. In our study, we werenot able to assess completely the risk of caecal contents con-tamination at slaughter, given that the variances of the differentlogistic regression models were around 30%, meaning that onlya third of the variance was explained by the proposed models.

The fact that Salmonella was far more detected from over-shoes than from feces seems in accordance with an earlierpublished study. This was especially true for different pro-duction stages like lactating and young sows, weaned pigs andthe fattening stages (Korsak et al., 2003).

It is widely assumed that bacteriological classical techniquesapplied on matrixes containing fecal matter in order to detectSalmonella in contaminated farms have poor sensitivity (Bagerand Petersen, 1991; Nielsen et al., 1995). In this way, Hurd et al.(2004) showed that the relative sensitivity of a unique 1 g fecalsamples was only 13.3%. This very low value may explain partlythe observed differences of contamination levels between thefollowed fattening herds. Another explanation could be the lownumbers of sampled caecal contents for two herds (herds 6and 7) were mean antibodies concentrations were above 20%,while no Salmonella could be detected in caecal contents fromthese herds. This poor sensitivity is essentially due to inter-mittent shedding of low numbers of Salmonella excreted fromanimals categorized as carriers (swine contaminated bySalmonella, but harboring no symptoms of salmonellosisusually present in contaminated human). However, despitethis poor sensitivity, one has always to consider this method asreference. The classical bacteriological technique enablesfurther research on a epidemiological point of view by allowingfurther characterization of isolated strains (e.g. serotyping,phage typing, Pulse Field Gel Electrophoresis, antimicrobialresistance,…). A lot of experiments have still to be done in thefuture in order to improve these methods.

A special attention has to be dealt with the volume of fecessampled for bacteriological examination. In a survey conductedin USA, an improvement in sensitivity was observed whenincreasing sample weight of analyzed feces from 1 to 10 g andfrom 10 to 25 g (Funk et al., 2000). In this context, one has tomention that our experiment relies on detection of Salmonellafrom 5 g of caecal content. Greater percentages of positiveresults might have been obtained if a larger amount of caecalcontents has been sampled on the slaughter line. Anotherpossibility consisted of pooling 3 or 5 caecal contents for anal-ytical purpose.

The sampling of pigs in the great intestine may appear easierto perform in the slaughterhouse environment but is not ap-propriate, due to possible contamination of this part of intestinefrom water of the dehairing machine device (Korsak et al.,2003). It seems that Salmonella detection from caecal contentsreflects more the bacteriological status of swine just beforeslaughtering. Pigs may reactivate and excrete Salmonella frominternal organs such as tonsils and lymph nodes from a delay as


short as 2 h. This was clearly demonstrated by Berends et al.(1996) who refer to the stress undergone by pigs during transportand lairage in order to explain this situation. This indicates thatSalmonella isolation in this viscera may not entirely reflect thebacteriological situation of the farm of origin, but also may bedue to contamination during transport or lairage. Negative effectof this last stage, especially when bad sanitizing procedures areapplied, was shown by Hurd et al. (2001).

Our findings have not revealed the major interest of usingserology alone during the last month of the fattening stage, inorder to organize and plan a strategic slaughtering of pigs (i.e.slaughter of non contaminated pigs before those considered assuspect), given that the result of a serological test has to betreated with caution. Indeed, it was earlier shown that relativesensitivity and negative predictive value were close to 80% and90%, respectively, indicating the important numbers of falsepositive and false negative results obtained with this serologicaltechnique (Korsak et al., 2004). The maximal relative sensitivityassessed in the present study was only 80%. On an animal level,when comparing the recovery of Salmonella from one animal inits caecum and its relative antibody concentration (referenceOD20%), nearly 40% of results were not well classified: 37% ofpigs where Salmonella was recovered from caecal contents hadrelative antibody concentrations lower than 20% and, con-versely, 33.7% of pigs where no Salmonella was detected fromcaecal contents had relative antibody concentrations higher than20% (data not shown). This means that on an individual scale,the prediction between Salmonella in caecal content and thequantification of antibodies was very low and not reliable.

In caecal contents, a mean prevalence of 8.74% (27 positiveout 309) was obtained. This percentage is quite similar to that ofLetellier et al. (1999) who revealed a percentage of 5.2% insamples of 1 g of caecal contents.

The final conclusion of this work may be that serology usedalone at a whole cannot be the unique solution in order tocharacterize the risk of shedding Salmonella from suspectedherds, bacteriological analyses remaining the reference forimplementing a Salmonella-free pig production system. Furtherresearch is needed for setting up in Belgium a system for thecharacterization or classification of herds based upon serolog-ical results obtained from several samplings at the fattening unitor at slaughter.

Acknowledgements

The authors wish to thank the Ministère de la RégionWallonne (Direction Générale de l'Agriculture) and the FederalMinistry of Agriculture and Small Enterprises for financing theproject.

References

Anonymous, 2001. Proposal for a Directive of the European Parliament and ofthe Council on the monitoring of zoonoses and zoonotic agents, amendingCouncil Decision 90/424/EEC and repealing Council Directive 92/117/EEC/⁎ COM/2001/0452 final— COD 2001/0176. Official Journal C 304 E,30/10/2001 P. 0250–0259.

Anonymous, 2003a. Directive 2003/99/EC of the European Parliament and of theCouncil of 17 November 2003 on the monitoring of zoonoses and zoonoticagents, amending Council Decision 90/424/EEC and repealing CouncilDirective 92/117/EEC. Official Journal L 325, 12/12/2003 P. 0031–0040.

Anonymous, 2003b. Regulation (EC) No 2160/2003 of the European Parliamentand of the Council of 17 November 2003 on the control of salmonella andother specified food-borne zoonotic agents. Official Journal L 325, 12/12/2003 P. 0001–0015.

Bager, F., Petersen, J., 1991. Sensitivity and specificity of different methods forthe isolation of Salmonella from pigs. Acta Veterinaria Scandinavica 32,473–481.

Berends, B.R., Urlings, H.A.P., Snijders, J.M.A., Van Knapen, F., 1996.Identification and quantification of risk factors in animal management andtransport regarding Salmonella spp. in pigs. International Journal of FoodMicrobiology 30, 37–53.

Blaha, T., 2003. Implementing a Salmonella monitoring programme for pork inGermany. In: Leontides, L. (Ed.), Proceedings of 5th InternationalSymposium on The Epidemiology and Control of Foodborne Pathogens inPork, Heraklion (Greece). October 1–4.

Bouyer, J., Hémon, D., Cordier, S., Derriennic, F., Stücker, I., Stengel, B.,Clavel, J., 1995. Epidemiologie, Principes et Méthodes Quantitatives.Inserm, Paris, pp. 267–290.

Christensen, J., Baggesen, D.L., Soerensen, V., Svensmark, B., 1999.Salmonella level of Danish swine herds based on serological examinationof meat-juice samples and Salmonella occurrence measured by bacterio-logical follow-up. Preventive Veterinary Medicine 40, 277–292.

Daube, G., Van Loock, F., 1997. Surveillance des toxi-infections d'originealimentaire en Belgique. Archives of Public Health 55, 351–361.

De Smedt, J.M., Bolderdijk, R., Rappold, H., Lautenschlaeger, D., 1986. RapidSalmonella detection in foods by motility enrichment semisolid Rappaport–Vassiliadis medium. Journal of Food Protection 49, 510–514.

De Zutter, L., Daube, G., 1998. Improved isolation of Salmonella from meatusing the diagnostic semisolid Salmonella agar Diasalm. Proceedings ofcongress “Foodborne pathogens: detection and typing.” Internationalsymposium sponsored by ISF/ISO/AOAC. Den Hague, The Netherlands.April 20–21.

Fleiss, J.L., 1973. The measurement and control of misclassification error, In:John Wiley and sons (Ed.), Statistical Methods for Rates and Proportions,1st edition. Wiley-Interscience, New York, pp. 140–154.

Funk, J.A., Davies, P.R., Nichols, M.A., 2000. The effect of fecal sample weighton detection of Salmonella enterica in swine feces. Journal of VeterinaryDiagnostic Investigation 12, 412–418.

Haeghebaert, S., le Querrec, F., Gallay, A., Bouvet, P., Gomez, M., Vaillant, V.,2002. Les toxi-infections alimentaires collectives en France, en 1999 et2000. Bulletin Epidémiologique Hebdomadaire 23, 105–109.

Hosmer, D.W., Lemeshow, S., 1989. Applied Logistic Regression. Wiley, NewYork, p. 307.

Hurd, H.S., McKean, J.D., Wesley, I.V., Karriker, L.A., 2001. The effect oflairage on Salmonella isolation from market swine. Journal of FoodProtection 64, 939–944.

Hurd, H.S., McKean, J.D., Griffith, R.D., Rostagno, M.H., 2004. Estimation ofthe Salmonella enterica prevalence in finishing swine. Epidemiology andInfection 132, 127–135.

Ihaka, R., Gentleman, R., 1996. R: a language for data analysis and graphics.Journal of Computational and Graphical Statistics 5, 299–314.

Korsak, N., Jacob, B., Groven, B., Etienne, G., China, B., Ghafir, Y., Daube, G.,2003. Salmonella contamination of pigs and pork in an integrated pigproduction system. Journal of Food Protection 66, 1126–1133.

Korsak, N., Leroy, B., Etienne, G., Degeye, J.-N., Clinquart, A., Daube, G.,2004. Comparaison des analyses sérologiques et bactériologiques dedétection des salmonelles dans le suivi sanitaire d'exploitations porcines.In: INRA, ITP (Eds.), 36èmes Journées de la Recherche Porcine, 3–5 Février2004, Paris, France.

Kranker, S., Alban, L., Boes, J., Dahl, J., 2003. Longitudinal study of Salmo-nella enterica serotype Typhimurium infection in three Danish farrow-to-finish swine herds. Journal of Clinical Microbiology 41, 2282–2288.

Landman, W.J.M., Hartman, E.G., Doornenbal, P., 1996. Detection ofSalmonella spp in poultry samples: comparison between Diagnostic Semi-


Solid Agar (Diasalm) and Rappaport Vassiliadis Broth (RV). De Ware(-n)Chemicus., vol. 26, pp. 234–237.

Letellier, A., Messier, S., Quessy, S., 1999. Prevalence of Salmonella spp. andYersinia enterocolitica in finishing swine at Canadian abattoirs. Journal ofFood Protection 62, 22–25.

Lo Fo Wong, D., Dahl, J., Van der Wolf, P., Wingstrand, A., Leontides, L., VonAltrock, A., 2003. Recovery of Salmonella enterica from seropositivefinishing pig herds. Veterinary Microbiology 97, 201–214.

Mead, P.S., Slutsker, L., Dietz, V., McCaig, L.F., Bresee, J.S., Shapiro, C.,Griffin, P.M., Tauxe, R.V., 1999. Food-related illness and death in the UnitedStates. Emerging Infectious Diseases 5, 607–625.

Mousing, J., Jensen, P.T., Halgaard, C., Bager, F., Feld, N., Nielsen, B., Nielsen,J.P., Bech-Nielsen, 1997. Nation-wide Salmonella enterica surveillance andcontrol in Danish slaughter swine herds. Preventive Veterinary Medicine 29,247–261.

Nielsen, B., Baggesen, D., Bager, F., Haugegaard, J., Lind, P., 1995. Theserological response to Salmonella serovars typhimurium and infantis inexperimentally infected pigs. The time course followed with an indirect anti-LPS ELISA and bacteriological examinations. Veterinary Microbiology 47,205–218.

Nielsen, B., Ekeroth, L., Bager, F., Lind, P., 1998. Use of muscle fluid as a sourceof antibodies for serologic detection of Salmonella infection in slaughter pigherds. Journal of Veterinary Diagnostic Investigation 10, 158–163.

Nielsen, B., Alban, L., Stege, H., Sorensen, L.L., Mogelmosse, V., Bagger, J.,Dahl, J., Baggesen, D.L., 2001. A new Salmonella surveillance and controlprogramme in Danish pig herds and slaughterhouses. Berliner undMünchener Tierärztliche Wochenschrift 114, 323–326.

Proux, K., Houdayer, C., Humbert, F., Cariolet, R., Rose, V., Eveno, E., Madec,F., 2000. Development of a complete ELISA using Salmonella lipopoly-saccharides of various serogroups allowing to detect all infected pigs.Veterinary Research 31, 481–490.

Sorensen, L., Alban, L., Nielsen, B., Dahl, J., 2004. The correlation betweenSalmonella serology and isolation of Salmonella in Danish pigs at slaughter.Veterinary Microbiology 101, 131–141.

Stege, H., Christensen, J., Nielsen, J.P., Baggesen, D.L., Enoe, C., Willeberg, P.,2000. Prevalence of subclinical Salmonella enterica infection in Danishfinishing pig herds. Preventive Veterinary Medicine 44, 175–188.

Wegener, H.C., Hald, T., Lo Fo Wong, D., Madsen, M., Korsgaard, H., Bager,F., Gerner-Smidt, P., Molbak, K., 2003. Salmonella control programs inDenmark. Emerging Infectious Diseases 9, 774–780.

Wiberg, C., Noberg, P., 1996. Comparison between a cultural procedure usingRappaport–Vassiliadis broth and motility enrichments on modifiedsemisolid Rappaport–Vassiliadis medium for Salmonella detection fromfood and feed. International Journal of Food Microbiology 29, 353–360.

1-s2.0-s0168160506000195-main.pdf

Documents