High levels of contamination and antimicrobial-resistant non-typhoidal Salmonella serovars on pig and poultry farms in the Mekong Delta of Vietnam

SUMMARY We investigated the prevalence, diversity, and antimicrobial resistance (AMR) profiles of non-typhoidal Salmonella (NTS) and associated risk factors on 341 pig, chicken, and duck farms in Dong Thap province (Mekong Delta, Vietnam). Sampling was stratified by species, district (four categories), and farm size (three categories). Pooled faeces, collected using boot swabs, were tested using ISO 6575: 2002 (Annex D). Isolates were serogrouped; group B isolates were tested by polymerase chain reaction to detect S. Typhimurium and (monophasic) serovar 4,[5],12:i:- variants. The farm-level adjusted NTS prevalence was 64·7%, 94·3% and 91·3% for chicken, duck and pig farms, respectively. Factors independently associated with NTS were duck farms [odds ratio (OR) 21·2], farm with >50 pigs (OR 11·9), pig farm with 5–50 pigs (OR 4·88) (vs. chickens), and frequent rodent sightings (OR 2·3). Both S. Typhimurium and monophasic S. Typhimurium were more common in duck farms. Isolates had a high prevalence of resistance (77·6%) against tetracycline, moderate resistance (20–30%) against chloramphenicol, sulfamethoxazole-trimethoprim, ampicillin and nalidixic acid, and low resistance (<5%) against ciprofloxacin and third-generation cephalosporins. Multidrug resistance (resistance against ⩾3 classes of antimicrobial) was independently associated with monophasic S. Typhimurium and other group B isolates (excluding S. Typhimurium) and pig farms. The unusually high prevalence of NTS on Mekong Delta farms poses formidable challenges for control.


INTRODUCTION
Non-typhoidal Salmonella (NTS) is the generic name given to serovars belonging to Salmonella enterica subspecies I other than those that typically cause invasive disease in non-immunocompromised humans (i.e. S. Typhi and S. Paratyphi A). The majority of NTS infections do not result in clinical disease in animals. However, they are considered to be zoonotic and to have pathogenic potential in humans, typically causing self-limiting gastroenteritis [1]. In Vietnam NTS are responsible for up to 7% of cases of diarrhoea in children [2]. In addition, invasive infections due to NTS have become increasingly reported in Vietnam while infections due to S. Typhi are declining [3].
Due to economic and diagnostic limitations, full serotyping of NTS involved in disease in humans is not routinely performed in Vietnam, with most data on NTS serovar distribution originating from ad hoc research studies. For example, a study of 1419 paediatric cases of diarrhoea in 2009-2010 identified group B NTS in 58% of 77 identified NTS cases [4]. A previous study of 54 cases from five provincial hospitals in 2004 identified S. Typhimurium [phage type (PT) 90] as the most common serovar. The same study also identified S. Typhimurium PT90 as the second most common serovar in pigs after S. Anatum [5]. Monophasic S. Typhimurium (mST) variants (i.e. serovar 4, [5],12:i:-) lacking the first or the second phase H antigen due to a genetic deletion have been emerging since the 1990s worldwide, and are often associated with an increase in multidrug resistance (MDR) [6] but have not yet been reported in Vietnam.
A number of studies have investigated the presence of NTS in meat/carcasses in Vietnam [7][8][9][10][11][12]. However, to our knowledge, the farm-level prevalence of NTS and associated risk factors have not yet been systematically investigated in Vietnam; both of which are prerequisites for the development of control measures on farms.
We performed a survey on NTS in 341 pig and poultry farms in the Mekong Delta region of southern Vietnam using environmental faecal sampling methods. The aims of the study were: (1) to determine the prevalence and main serogroups of Salmonella, S. Typhimurium and mST in the main farmed species; (2) to describe antimicrobial resistance (AMR) patterns in NTS isolates; and (3) to investigate farm-level risk factors for NTS, S. Typhimurium, mST and MDR NTS.

Survey design and data collection
The survey was conducted between February and May 2012 in Dong Thap (Vietnam), a Mekong Delta province characterized by high density of small-scale pig and chicken farms, as well as large numbers of duck flocks typically reared in synchronicity with the rice production cycle. The study covered 4/12 districts, three species (chicken, ducks, pigs) and three farm size categories. Farms were randomly selected from the district farm population census as previously described [13]. When a farmer refused participation, his farm was replaced by the next available one. Questionnaires (available upon request) were used to obtain basic farm demographics as well as other parameters related to the farmer and the farm management. Demographic statistical data pertaining to the commune (density of humans, pigs and poultry) were obtained from the Vietnamese Government Statistical Office (GSO).

Sampling methods
We performed environmental sampling using swabs to collect naturally pooled faeces [14][15][16]. A fixed number of boot swabs were collected from small (n = 3), medium (n = 4) and large (n = 5) farms. This sampling scheme was formulated as a compromise between sensitivity, simplicity and the requirement for standardization. Since one pair of boot swabs is roughly equivalent to sampling 60 individual faecal droppings [17], we assumed that our sampling method was equivalent to sampling a constant proportion of animals from each farm. Briefly, areas dedicated to housing the farms' target species were divided into equally sized sections. From each section a pair of boot swabs was taken by walking over 50 steps with a shuffling motion focused on areas where fresh droppings were visible. Where walking on housing areas was not possible (i.e. sows in crates/farrowing pens, chickens in cages or stilt houses), hand-held gauze swabs were used to sample faeces from 10 sampling points (∼25 g per swab). Recent studies have shown that three pairs of boot swabs used per flock are equivalent to sampling methods used in the European Union NTS surveys [18].
Isolates were assigned to a serovar by multilocus sequence typing (MLST) on DNA extracted using the Wizard Genomic DNA extraction kit (Promega, USA) from one full loop of pure NTS colonies of overnight culture in nutrient agar media. PCR amplification was performed on the seven MLST loci aroC, dnaN, hemD, hisD, purE, sucA and thrA, as described previously [22]. PCR products were purified using QIAquick column-based PCR purification system (Qiagen, USA) and sequenced in forward and reverse directions using the Big Dye Cycle Sequencing kit (Applied Biosystems, USA) on an ABI 3770 automatic sequencer according to the manufacturer's instructions. After generating a sequence type, strains were inferred to serotypes according to the data available on the MLST database (http://mlst. warwick.ac.uk/mlst/dbs/Senterica/). Due to economic limitations, we performed selective MLST-based serovar identification on one isolate representing each phenotypic pattern, defined by the combination of (a) serovar/serogroup (S. Typhimurium, mST, other group B, group C, group D, and 'other group'); and (b) sensitivity results for 10 antimicrobials. The tested isolate was chosen at random.

Statistical analyses
Data from all farms were combined into a unique dataset aiming to model the overall probability of a swab testing positive for NTS. Since the number of farms in each stratum (i = 1, . . ., 36) was known from census data, prevalence estimates and odds ratio (OR) estimates, regression models were adjusted by assigning a stratum-specific sampling weight (W i ) to each observation unit (swab): where N T is the total number of farms ( 35 248) and N i is the number of farms sampled in each stratum. Standard errors were corrected to take into account potential similarities between farms in each stratum [23]. The prevalence of a positive boot swab per farm was compared for each species. Risk factor analyses at swab level were performed using hierarchical logistic regression with 'farm' as the clustering variable for the following outcomes: (1) prevalence of NTS; (2) prevalence of S. Typhimurium-positive results (vs. any other result, including negative); and (3) prevalence of mST result (vs. any other result, including negative). Each NTS isolate was characterized by a profile defined by its unique combination of type (S. Typhimurium, mST, other group B, group C, group D, other group) and its susceptibility/resistance against the 10 antimicrobials tested. Shannon-Weaver entropy (H) was used to estimate the farm-level diversity of isolates from the same farm. H was defined as: where p i is the proportion of isolates with the ith resistance profile, and R is the total number of profiles. Higher values of H correspond to greater diversity in AMR profiles.
The prevalence of AMR against all antimicrobials tested (and their corresponding 95% confidence intervals; CI) was calculated for S. Typhimurium, mST and other group B serovars, groups C, and D, and other group serovars separately. Hierarchical logistic regression models with the variable 'farm' as a random effect were built using data on all NTS isolates to investigate the outcome of 'probability of a multidrug resistant (MDR) strain', defined as resistance against 53 classes of antimicrobials. Variables investigated in all models were: (1) type of farm, defined by the target species and size (nine categories); (2) age, gender and years of experience of the farmer; (3) use of antimicrobials on the farm over the last 3 months; (4) use of any Salmonella vaccines on target species on farm; (5) history of disease on farms (enteric, respiratory or other); (6) presence of chickens/ducks/ pigs as 'non-target species'; (7) presence of cattle/buffalo/dog/cat; (8) type of water (well, municipal); and (9) frequency of rodent and wild bird sightings. In all models candidate variables were those significant (<0.05) in the univariable models for any of the three outcomes. Variables were ranked by their degree of significance and were included in the model starting with the most significant ones using a step-wise forward approach [24]. In each of the final multivariable models, variables were retained if their P value was <0·05 for any of the outcome variables. All interactions between significant variables in the model were assessed. The suitability of each new variable included in the model was assessed using the Wald test [23]. All interactions between final significant variables were tested. All statistical analyses were performed using R packages 'survey' and 'epicalc' (http://www. r-project.org).

RESULTS
The total number of farms in the sampling frame as well as the actual number of farms sampled (sampling units) and their associated sampling weights are shown in Supplementary Table S1. A total of 341 farms were sampled (117 chicken, 118 duck, 104 pig farms) using a total of 1349 swabs, and 739 NTS isolates were recovered. Since small chicken farms were by far the predominant farm category (66·2% of all farms), swabs from these farms were assigned the largest sampling weight.

Risk factors for NTS, S. Typhimurium and mST
The nine farm-type categories were combined into four groups, (a) chicken farms (baseline); (b) duck farms; (c) small pig farms; and (d) medium and large pig farms, given the observed statistically significant differences in risk between (b), (c), and (d) and baseline. Swabs from duck farms were associated with a higher prevalence of NTS compared to chicken swabs (OR 21·2). Compared to chicken farms, pig farms had an increased risk, although the magnitude of risk was higher for large farms (OR 11·9, compared to OR 4·9 for smaller pig farms). Other factors associated with NTS were 'rodents seen at least weekly' (OR 2·3), and male farmer (OR 2·5). The interaction between 'high frequency of rodent sightings' and farm type indicated an additional increase in risk for small pig farms, large pig farms and chicken farms with high frequency of rodent sightings (OR 8·8, 21·3, and 1·8, respectively) (compared to chicken farms with low frequency of rodent sightings). By contrast, duck farms with a high frequency of rodent sightings had no increased risk (vs. duck farms with low frequency of rodent sightings) (OR 0·79, P = 0·26). Of the three species, ducks had the greatest prevalence of S. Typhimurium, although it was only significantly greater compared to pigs (but not chickens). The human density at commune level was a strong protective factor for S. Typhimurium (log number of humans/km 2 , OR 0·5).
Duck flocks had a greater probability of testing positive with mST compared to chickens or pigs (OR 6·8), as did farms located in areas of high density of pig farms (OR 1·6). By contrast, supplying animals with municipal water was a strong protective factor (OR 0·1) for all species (Table 2).
A total of 97/103 pig farmers reported that they had vaccinated their herds against Salmonella over the previous year, using both inactivated and attenuated preparations containing S. Cholerasuis and/or S. Typhimurium strains. There was no difference in prevalence between vaccinated and unvaccinated herds (data not shown). None of the chicken and duck farms reported using Salmonella vaccines.

Risk factors for MDR
MDR was associated with pig farms (vs. other species) (OR 4·5) and with mST (OR 18·5), other group B serovars (OR 13·6), groups other than B, C, or D (OR 6·3), and S. Typhimurium (OR 6·2) (vs. groups C and D). A high density of pig farms in the commune was also a risk. The calculated OR of 1·4 is equivalent to an OR of 6·8 for a farm located in a commune with a density of 86 pig farms/km 2 (75% quantile) vs. farms in communes with density of 16 pig farms/km 2 (25% quantile). The human density was not associated with increased resistance and was therefore excluded from the model (Table 4).

Serovar allocation to representative AMR/serogroup patterns by MLST
Of isolates other than S. Typhimurium and mST (n = 632) there were 63 unique patterns reflecting the combinations of NTS group (groups B, C, D, and other) and AMR test result. A total of 14 different serovars were identified by testing one randomly selected isolate within each pattern. A total of 391/632 isolates (61·9%) corresponded to seven patterns of isolates either  The remaining 56 patterns (i.e. with resistance pattern other than tetracycline-only), representing 241 isolates, and their allocated serovar identity are presented in Table 5. The most commonly identified serovars were S. Derby (n = 9 patterns), S. London (n = 7), S. Anatum (n = 7), S. Indiana (n = 6), and S. Rissen (n = 5).
In chicken farms, patterns assigned to S. London and S. Anatum were predominant; in duck farms S. Indiana and S. London; and in pig farms S. Derby and S. London were predominant.
In our survey duck farms had the highest farm-level NTS prevalence (94·3%). A possible explanation is the relatively less attention to biosecurity and the characteristic management system of duck flocks that involve grazing on rice fields [28]. A previous study also reported a higher animal-level prevalence in ducks (8·7%) compared to pigs and chickens (5·2-7·9%) in the Mekong Delta, although these differences were not significant [29]. Data from Denmark (2012), where duck flocks are routinely monitored, indicate a higher prevalence of NTS infection in duck flocks compared to chicken flocks (46·0% vs. 0·8%) [30]. Our data also indicated that, of the three species, ducks also had the highest per sample prevalence. This probably reflects higher levels of animal-level infection, since the sensitivity of pooled samples is a function of within bird-level prevalence [31].
A high frequency of rat sightings on farms was a significant risk factor. Studies have documented an association between rodents and increased risk of NTS on farms [32,33]. Rats are extremely prevalent in the Mekong Delta. A study on 600 rats trapped in several Mekong Delta provinces indicated high NTS prevalence (19·3%) (40% for rats trapped in Dong Thap). A total of 94% rats were colonized by either NTS serovars London, Weltevreden or Derby, all identified in our study [34]. Most farm buildings in the Mekong Delta are not rat-proof, and pest control measures are normally insufficient. In these circumstances rodents can easily access feed storages where they often defecate. This may plausibly contribute to infection of farmed animals through consumption of contaminated feed. The significant (protective) interaction between ducks and rodents suggests that, unlike in the case of pigs and chickens, rats probably play a lesser role in transmission of NTS to ducks, and other sources linked to their grazing practices may be more relevant. However the low levels of biosecurity in Mekong Delta farms suggest the potential role of many potential vectors and reservoirs. We report a higher risk in larger pig farms (n > 50), in agreement with previous studies [35,36], reflecting the challenge in controlling NTS in pig production, probably due to the continuous introduction of new NTS strains via replacement of breeding stock, as well as the intrinsic difficulties of adequately disinfecting occupied buildings.
Although most NTS infections in animals are asymptomatic, some host species-adapted serovars such as S. Cholerasuis and S. Typhisuis (pigs) and S. Pullorum and S. Gallinarum (chickens) often cause fatal disease [37]. In pigs, septicaemic episodes of S. Typhimurium have been reported [38]. In Mekong Delta pig farmers there is a strong perception that NTS infections cause significant production losses, and this explains the high proportion of farm owners reporting routine vaccination (94%).
Of the three animal species investigated, serovars S. Typhimurium and mST were more common in ducks. This is in contrast with European countries where these serovars are predominantly isolated from pig farms and pork [25,39,40]; however, the size of the duck industry in the EU is very small and data on NTS prevalence in ducks is limited.
The negative association between mST and municipal water suggests that water sources other than municipal water (i.e. river, well or rain water) may represent potential risks, although no specific water source categories remained significant in the multivariable models.
With the exception of tetracyclines, against which most (77·8%) isolates were resistant, levels of resistance against ampicillin, chloramphenicol, and nalidixic acid Table 3. were generally lower (22·8-27·0%) compared to NTS isolates from meat samples in northern Vietnam (24·5-39·8%) [12,41]. We found a strong association between group B serovars and AMR, independent of the animal host species, with the highest prevalence of MDR corresponding to mST, followed by 'other' group B serovars. Compared to chicken and duck isolates, a higher proportion of NTS isolates from pigs were antimicrobial resistant. In addition AMR was partly explained by the density of pig farms, suggesting that environmental contamination with antimicrobials or other AMR organisms from pig farms in the vicinity could have an impact beyond the pig farms themselves. Nevertheless, our results are in agreement with routine monitoring data collected in the EU where resistance levels in isolates from pigs against tetracyclines, ampicillin and sulphonamides were higher than in broiler isolates (55-59% vs. 20-27%). By contrast, EU chicken NTS isolates had moderate to high levels of resistance against nalidixic acid and ciprofloxacin (23-24%), whereas pig isolates had low levels of resistance (2-3%) [42]. However, surveillance data on AMR from animal NTS in the EU indicate resistance based on EUCAST epidemiological cut-off values, not clinical breakpoints. Because of this, the observed differences should be viewed with caution.
The emergence of mST worldwide has been associated with clonal groups resistant to ampicillin, sulphonamides, streptomycin and tetracyclines [45,46].
Although we did not test for streptomycin resistance, our results confirm a high proportion (31·1%) of mST co-resistant against ampicillin, sulphonamides and tetracyclines, vs. 16·9% in other serovars (data not shown).
We used MLST to test isolates representing unique serogroup/AMR combinations, excluding fully susceptible or tetracycline resistant-only isolates. Unfortunately, we could not perform MLST testing on all 739 isolates due to economic limitations. Clearly testing was insufficient for the (more common) fully susceptible or tetracycline-only susceptible serovars. It is expected that more economic alternatives to conventional serotyping will become available in developing countries to enable the characterization of NTS on farms and the investigation of human outbreaks [39].
In summary, we demonstrated an exceptionally high prevalence and high diversity in NTS serovars across farms in the Mekong Delta. Of the three species investigated, ducks had the highest NTS prevalence although pigs were associated with the highest levels of MDR. Levels of AMR were considerably high for most antimicrobials investigated, except for amoxicillin/clavulanic acid, ciprofloxacin and thirdgeneration cephalosporins.
Our study confirmed the widespread presence of MDR mST in Mekong Delta farms, as reported in other countries. Farming practices in the Mekong Delta are typical of many developing countries in the region. There is a concern that fast increases in animal production are not matched by increases in biosecurity and testing programmes for foodborne pathogens. In Vietnam, virtually no NTS serotyping is conducted in human or animal diagnostic   laboratories. However, given the country's rapid economic development and the increasing potential for export markets this may soon change. The unusually high prevalence, the predominance of mixed-species farming without adequate biosecurity, and the abundance of vectors (rats) suggest that control of NTS on farms in the Mekong Delta of Vietnam will be particularly challenging.