Evaluation of pre-PCR processing approaches for enumeration of Salmonella enterica in naturally contaminated animal feed




Three pre-PCR processing strategies for the detection and/or quantification of Salmonella in naturally contaminated soya bean meal were evaluated.

Methods and Results

Methods included: (i) flotation-qPCR [enumeration of intact Salmonella cells prior to quantitative PCR (qPCR)], (ii) MPN-PCR (modified most probable number method combined with qPCR) and (iii) qualitative culture enrichment PCR. The limit of quantification was 1·8 × 102 CFU g−1 (flotation-qPCR) and 0·02 MPN g−1 (MPN-PCR). Fifteen naturally contaminated Salmonella positive soya bean meal samples from one lot were analysed in parallel with the three methods, using 2·5, 50 and 25 g of feed, respectively, resulting in detection of Salmonella in 6, 15 and 9 bags. Enumeration resulted in 1·8 × 102–7·8 × 103 CFU g−1 (flotation-qPCR) and 0·024 to >5·2 MPN g−1 (MPN-PCR).


Except for differences in methodology, results obtained with the three techniques could be due to the presence of nonculturable Salmonella and/or a heterogeneous distribution of Salmonella in the material.

Significance and Impact of the Study

The evaluated methods provide different possibilities to assess the prevalence of Salmonella in feed, together with the numbers of culturable, as well as nonculturable cells, and can be applied to generate data to allow more accurate quantitative microbial risk assessment for Salmonella in the feed chain.


Animal feed can serve as a reservoir of Salmonella and contribute to the spread of the bacteria to the food production chain (Davies et al. 2004; Maciorowski et al. 2006b; Rostagno and Callaway 2011). Moreover, feed ingredients have been identified as an important source for introducing Salmonella in the feed production chain (Wierup and Häggblom 2010). To perform adequate quantitative microbial risk assessment (QMRA) for the feed chain, there is a need for data on the numbers of Salmonella present in different steps in the feed chain, including feed ingredients (Binter et al. 2011). However, there are few studies where Salmonella are enumerated in naturally contaminated feed, but numbers as low as <1 MPN per 100 g have been reported (Smeltzer et al. 1980; D'Aoust and Sewell 1986). In a more recent study, it was estimated that the level of contamination could be as low as a few CFU per kg of feed (Sauli et al. 2005). Furthermore, Salmonella are considered to be unevenly spread, that is, to have an aggregative distribution in the feed material (Jarvis 2008). Typically, only a fraction of samples from a batch or lot are tested positive for Salmonella (Andersson et al. 2010; Binter et al. 2011). To monitor Salmonella contamination in the feed production chain, it is essential to have accurate methods for detection, enumeration and isolation of Salmonella from feed materials (Löfström et al. 2004; Maciorowski et al. 2005, 2006a; Koyuncu et al. 2010; Löfström and Hoorfar 2012). There are several techniques available all with their strengths and weaknesses, applicable for different scenarios and sample types. In this study, three different pre-PCR processing strategies were evaluated on naturally contaminated feed samples.

One conventional technique that is suitable for enumeration of small numbers of Salmonella bacteria is the most probable number method (MPN) (Halvorson and Ziegler 1933; Blodgett 2010). In this method, triplicates or fivefold replicates are prepared from serial dilutions. All samples are tested by the horizontal standard culture-based method (International Organisation for Standardization 2002). The ratio of positive/negative in relation to the concentration results in a MPN g−1 value. The MPN method assumes that bacteria are distributed randomly within the sample and are separated (not clustered together). The growth medium and conditions of incubation have been chosen so that one viable cell is multiplied and can be detected. The main disadvantage with the MPN approach is that it is laborious and time consuming. Attempts have been made lately to overcome these obstacles by introducing a miniaturized MPN method (Pavic et al. 2010). However, to further decrease the work load and time, an MPN-PCR approach can be useful (Krämer et al. 2011). In MPN-PCR, the first part is based on culture enrichment [e.g. ISO 6579:2002 for Salmonella (International Organisation for Standardization 2002)], followed by real-time PCR. For Salmonella, this could mean that the samples are grown in buffered peptone water (BPW) and thereafter analysed by PCR on aliquots withdrawn from the enrichment broth. This approach was included in this study because it saves time and can also contribute to a lower rate of false-negative results as not so many steps including selective pressure are applied. Selective enrichment has been shown to pose a problem in some cases, for example nonmotile Salmonella cannot be found on media that rely on motility for selection, as the modified semisolid Rappaport-Vassiliades (MSRV) agar plates (Soria et al. 2011; Hello et al. 2012). Moreover, the use of selective media has been shown to contribute to the selection of more competitive strains when a mixed population of Salmonella are analysed (Singer et al. 2009; Gorski 2012), as might be the case for Salmonella positive feed lots.

It has been shown that Salmonella can survive for a long time in dry samples such as feed and, although sublethally damaged, still be infective after resuscitation (Wesche et al. 2009). However, detecting damaged cells might be difficult using culture-based methods, as these rely solely on the ability to grow the bacteria on liquid and/or solid culture media (Löfström et al. 2004; Koyuncu et al. 2010; Cocolin et al. 2011). There is also a risk of underestimating the numbers of damaged Salmonella cells using culture-based techniques. Moreover, in dry samples, such as feed, Salmonella can enter into a viable but nonculturable (VBNC) state, making it impossible to detect using standard culture-based methods (Löfström et al. 2004; Maciorowski et al. 2006a). Noticeably, to achieve a proper enumeration, but still be able to obtain an isolate for further subtyping, it is essential to supplement culture-based methods with culture independent procedures for the enumeration and isolation of Salmonella in samples such as feed.

One example of a culture independent method is flotation, a sample preparation technique based on buoyant density centrifugation, where particles and/or cells are separated based on their densities (Pertoft et al. 1977; Pertoft 2000). As applied in this study, the resulting sample after flotation consists of Salmonella cells free of most PCR-inhibitory substances originating from the feed sample, background flora, as well as DNA from dead Salmonella cells. Subsequently, cells can be analysed using qPCR and/or other methods for isolation, enumeration and characterization. Major drawbacks with this method is that it is laborious, time consuming and not suitable for high-throughput analysis of large sample volumes. Nevertheless, flotation has previously been used in the field of food and veterinary microbiology to quantify Salmonella from food and faecal material homogenized in liquid (Fukushima et al. 2007; Wolffs et al. 2007) and directly from carcass swab samples (Löfström et al. 2011), to quantify Staphylococcus aureus in milk (Aprodu et al. 2011) and Yersinia enterocolitica in pork (Wolffs et al. 2004), as well as to isolate Lawsonia intracellularis from faeces (Jacobson et al. 2009). Moreover, this technique has been shown to allow separation between living and dead Campylobacter cells (Wolffs et al. 2005a) and avoid detection of DNA from dead cells (Wolffs et al. 2005b).

PCR techniques have proven to be a valuable tool for detection and quantification of foodborne pathogens. However, there are some limitations to this technique and among the most important is the risk of PCR inhibition caused by complex biological samples [reviewed by Hedman et al. (2013)]. Inhibitory compounds originating from the sample and/or the sample preparation procedure might reduce the PCR amplification efficiency and influence PCR kinetics giving rise to bias in PCR-based quantification (Bar et al. 2012). This has also been demonstrated for feed samples (Löfström et al. 2004; Koyuncu et al. 2010). It is therefore essential to take the sample type that will be analysed into account, as different sample types with various chemical and structural composition could inhibit PCR with different mechanisms (Opel et al. 2010). This could for example result in a standard curve that is inaccurate when it is applied on a different matrix than it was originally validated for, as emphasized by others (Cocolin et al. 2011).

The objective of this study was to evaluate three pre-PCR processing strategies for detection and quantification of Salmonella in naturally contaminated animal feed samples, using soya bean meal as a model. The methods applied were as follows: (i) a flotation-qPCR protocol, previously developed to isolate and quantify Salmonella from carcass swab samples (Löfström et al. 2011); (ii) a MPN-PCR protocol modified to enable more accurate enumeration of low numbers of Salmonella;, and (iii) a culture enrichment PCR method for detection of Salmonella (Löfström and Hoorfar 2012).

Material and methods

Salmonella strain and feed samples

Salmonella ser. Cubana 548-03 and Salmonella negative reference soya bean meal were obtained from the National Veterinary Institute (SVA), Uppsala, Sweden. Naturally contaminated soya bean meal originating from one lot (30 000 tons in total, one sample taken per 2000 tons) was obtained from Svenska Foder, Lidköping, Sweden. Samples taken from the lot had been tested positive for Salmonella by standard sampling and analysis methods (Nordic Committee on Food Analysis 1999). Samples were stored cooled (2–5°C) before analysis to avoid microbial growth.

Artificial contamination of feed for evaluation of flotation method

A dilution series of cold-stressed cells of the Salmonella ser. Cubana 548–03 was prepared as previously described (Löfström et al. 2009) and used to inoculate feed samples, in the same manner as for carcass swab samples, as previously described (Löfström et al. 2011).

Soya bean meal samples (2·5 g) were aseptically transferred to 50-ml Falcon tubes and were artificially contaminated with Salmonella ser. Cubana 548–03 prepared as described above. One ml from a tube with appropriate concentration of Salmonella cells from a dilution series in saline solution was dropped on the soya bean meal sample, trying to spread the liquid as much as possible. The tubes were stored at 5°C for 1 h to allow the cells to attach to the soya bean meal.

Four independent experiments (A–D) with artificial contamination and subsequent flotation were conducted, employing a total of 14 soya bean meal samples. The following numbers of Salmonella were used (CFU per 2·5-g sample): A: 0 (negative control), 4·6 × 101, 4·6 × 102 (duplicate samples), 4·6 × 103 (duplicate samples), 4·6 × 104, 4·6 × 105; B: 3·1 × 104, 3·1 × 106; C: 3·9 × 103, 3·9 × 105; D: 2·4 × 104, 2·4 × 106.


The flotation procedure was performed in 50-ml Falcon tubes as previously described (Löfström et al. 2011). Three different sterile colloidal solutions of nontoxic silanized silica particles in 0·9% NaCl, all formulated by QRAB (Alunda, Sweden) and produced by FertiPro N.V., Beernem, Belgium, were used. The solutions were as follows: BactXtractor-L (BX-L; density, 1·055 g ml−1), BactXtractor-M (BX-M; density, 1·102 g ml−1) and BactXtractor-H (BX-H; density, 1·297 g ml−1). The final densities of BX-L and BX-M were reached after dilution of BX-H with sterile 0·9% NaCl and measured using a DMA46 density metre (Instrument AB Lambda, Stockholm, Sweden). The flotation media were stored at 4°C. Briefly, discontinuous gradients were centrifuged for 45 min in a swing-out rotor at 3056 g at 5°C. Due to their buoyant densities, intact Salmonella cells from the added samples floated up and accumulated at the upper interphase between the medium and the top layers. After centrifugation, a 1·0 ml sample was withdrawn from the surface of the top layer using a pipette (top phase sample) and a 4·5-ml sample was taken from the upper interphase between the top layer and the medium layer (upper interphase sample), by pipetting through the top layer.

In addition, aliquots (100 μl) from the top phase sample were spread on xylose lysine deoxycholate (XLD) agar plates (Oxoid, Merck, Darmstadt, Germany). Four millilitres from the upper interphase samples was transferred to tubes containing 8 ml of sterile 0·9% NaCl solution, was mixed by inverting the tubes several times and centrifuged at 3056 g for 10 min at 5°C to pellet the cells. The supernatants were withdrawn, and the pellets stored at −20°C until qPCR analysis. The pellets from the upper interphase sample were resuspended in 50 μl of 0·9% NaCl solution, and 9 μl was used for qPCR analysis. The remaining 500 μl from the upper interphase samples were serially diluted in sterile 0·9% NaCl solution and 100 μl of appropriate dilutions plated on XLD agar plates in duplicates to determine the concentration of Salmonella in the upper interphase sample. The plates were incubated at 41·5 ± 1°C and the number of colonies counted after 24 ± 3 and 48 ± 3 h.

Preparation of qPCR standard curve

A standard curve previously used for enumeration of Salmonella in carcass swab samples (Löfström et al. 2011) was used for comparison to the standard curve for feed obtained in this study. To generate a standard curve that was specific for feed samples, a dilution series of cells of Salmonella ser. Cubana strain 548-03 was prepared as described above. To generate the feed sample background, 16 2·5 g samples of Salmonella negative soya bean meal were analysed using flotation and 4·5 ml samples were taken from the upper interphase. Four ml samples of the different tenfold dilutions, corresponding to 2·6 × 101–2·6 × 107 CFU per 4 ml and 3·6 × 101–3·6 × 107 CFU per 4 ml, for two independent replicates dilution series of Salmonella ser. Cubana, respectively, were added to a tube containing 4 ml of 0·9% NaCl solution and 4 ml of upper interphase sample. Samples were mixed by inverting the tube several times and centrifuged at 3056 g for 10 min. The supernatant was removed, and the pellet resuspended in 50 μl of 0·9% NaCl solution. Nine microlitres of these samples were analysed using real-time PCR (see below) in duplicate resulting in 16 data points. The log10-transformed Salmonella CFU per 4 ml sample values were plotted against Ct values, and a standard curve was obtained using linear regression (MxPro software v. 4.10, Stratagene, La Jolla, CA, USA). A project (analysis of different PCR runs in one data analysis) was constructed in the MxPro software, including the standard curve and all PCR runs with samples from experiments A–D.


The 50 g feed sample was placed in a stomacher bag, and 500 ml BPW (37°C) was added. The bag was then placed in a stomacher for 2 min at highest speed. Solid and liquid material was transferred in different amounts to containers A–H to generate various dilutions of the original feed slurry. The amounts of solid and liquid material transferred to each container, respectively, were as follows: containers A and B 60 ml solid and 140 ml liquid; C and D 15 ml and 35 ml; E and F 3 ml and 7 ml; G and H 0·5 ml and 1 ml. Fresh BPW was added to a total volume of 30 ml to containers E–H. This dilution scheme resulted in the following amounts of feed in the respective container: A and B 19·1 g feed; C and D 4·8 g; E and F 1·0 g; G and H 0·16 g (Fig. 1). Containers A–H were incubated at 37°C for 16 h. Thereafter, 5 ml of the supernatant from each container was transferred to 15-ml tubes, centrifuged at 3000 g for 5 min, and the pellets stored at −20°C until DNA extraction and PCR analysis as described below. Based on the PCR results, samples were reported as positive or negative for Salmonella and the MPN values calculated using the MPN calculator of Mike Curiale (http://www.i2workout.com/mcuriale/mpn/index.html).

Figure 1.

Overview of the experimental setup comparing results obtained using flotation-qPCR, MPN-PCR and qualitative culture enrichment PCR for enumeration and/or detection of Salmonella in naturally contaminated soya bean meal samples.

DNA extraction

DNA extraction was performed on the pellets using a KingFisher (Thermo Labsystems, Helsinki, Finland), as previously described (Josefsen et al. 2007), using a DNA isolation kit for blood, stool, cells and tissue (Magnesil KF, Genomic system, Promega, Madison, WI, USA) as specified by the manufacturer and 75 μl of magnetic particles. The obtained DNA was eluted in 100 μl distilled water at the last step of the DNA extraction.

Real-time PCR

A real-time PCR method (Malorny et al. 2004), targeting a region within the ttrRSBCA locus, for the specific detection of Salmonella was employed as previously described (Josefsen et al. 2007; Löfström et al. 2009) using a sample volume of 9 μl and an initial heating step of 10 min at 95°C to lyse the cells.

PCR data were analysed as previously described (Josefsen et al. 2007; Löfström et al. 2009) using the MxPro software v. 4.10. However, a fixed threshold of 1000 was employed to enable comparison of Ct values between different PCR runs. Unknown samples and standards in the different PCR runs were analysed together using the “Project” option in the MxPro software.

The naturally contaminated samples were also analysed using a qualitative real-time PCR method, as previously described (Löfström and Hoorfar 2012). Briefly, a 25 g sample of feed was mixed with 225 ml BPW and incubated at 37 ± 1°C for 18 ± 2 h. A 1 ml sample was withdrawn, centrifuged and the pellet used for DNA extraction using a magnetic beads-based method and real-time PCR as described above.

Analysis of naturally contaminated samples

Naturally contaminated soya bean meal samples were obtained in 15 bags as 250 g samples each representing a 2000 tons sublot from a 30 000 tons lot. From each original sample, different subsamples were analysed with the following methods: (i) flotation followed by qPCR and plating on XLD as for the artificially contaminated samples (duplicates of 2·5 g), (ii) MPN-PCR (single sample of 50 g) and (iii) culture enrichment PCR (single sample of 25 g).

Statistical analysis

The relationship between the Salmonella cells recovered by flotation and number of cells used for artificial contamination of samples was evaluated by plotting the log10-transformed Salmonella CFU per sample obtained from the PCR and selective plating on XLD, respectively, against the log10-transformed CFU per sample used for artificial contamination, as previously described (Löfström et al. 2011). Based on the visual impression of the plots, a simple linear regression model was used to describe how the log10-transformed Salmonella CFU per sample obtained from the PCR and selective plating on XLD changed across the log10-transformed CFU per sample. The linear regression was carried out using Microsoft Excel 2010 Data Analysis Toolpack using the function ‘regression’.


Assessment of flotation setup and influence of feed matrix on qPCR

To adjust the previously designed flotation setup (Löfström et al. 2011) to feed samples, an optimization was performed, where it was found that the optimal sample weight was 2·5 g per 50-ml tube (data not shown). After flotation, the components of the feed sample were distributed between the three density layers for the analysed samples. In most cases, fat and some small feed residues were located in the top layer (density of < 1·055 g ml−1), while the remaining parts of the feed sample was located around the lower interphase. Salmonella cells were concentrated at the upper interphase. Thus, the previously designed setup enabled the separation of Salmonella from the feed samples.

To investigate whether the feed matrix had an influence on the PCR amplification efficiency, a standard curve was constructed. Sixteen 2·5 g samples of Salmonella negative soya bean meal were subjected to flotation and samples from the upper interphase were collected. These were mixed with Salmonella in different concentrations and analysed with qPCR to construct a PCR standard curve to enable quantification of Salmonella in naturally contaminated samples. The standard curve (y = −3·3x + 44·8, where y is the Ct value and x no. of Salmonella in log CFU per sample) had a linear range of 102–10CFU per sample, and an R2 value of 0·99.

Evaluation of flotation method on artificially contaminated feed samples

Four biological replicates (experiments A–D, in total 14 feed samples) were included using artificial contamination levels from 101 to 106 CFU per 2·5 g sample (Fig. 2). After flotation, samples were withdrawn from the surface of the top layer and from the upper interphase, and the number of Salmonella cells was estimated. Results show that a very small fraction of the cells (≤0·1%) of the artificially inoculated Salmonella cells were found in the top phase when enumeration was carried out by plating on XLD agar. For the upper interphase samples, Salmonella was enumerated using both plating on selective XLD and the optimized real-time PCR procedure. There was no significant difference in the slopes of the two regression lines for the culture-based and qPCR methods, indicating that the difference in the quantities obtained by the two methods over the tested concentration range was constant (Fig. 2; Table 1). Visual inspection of the curves indicates homogeneity of the variation along the number of artificially inoculated samples (i.e. homogeneous variation of observed values around the regression line). None of the samples were inhibitory to PCR.

Table 1. Results of linear regression analysis of artificially contaminated carcass swab and feed samples analysed by flotation combined with qPCR and plating on XLD agar plates
AnalysisSample type R 2 Slope (95% CI)Intercept (95% CI)Reference
PCRSwab0·981·1 (1·0–1·2)−0·5 (−0·9 to −0·1)Löfström et al. (2011)
Feed0·931·0 (0·8–1·2)0·3 (−0·6 to 1·2)This study
CultureSwab0·981·0 (1·0–1·1)−1·1 (−1·4 to −0·8)Löfström et al. (2011)
Feed0·960·9 (0·7–1·0)−0·3 (−0·8 to 0·3)This study
Figure 2.

Scatter plot of the quantification of Salmonella in artificially contaminated soya bean meal samples after flotation using plating on selective XLD agar plates (○) and quantitative PCR (●). Linear regression equations and R2 values are given next to the linear regression lines for the two methods. The 1 : 1 line is shown dotted.

The developed method allowed accurate quantification from 4·6 × 102 to at least 3·1 × 106 CFU Salmonella per feed sample of 2·5 g, corresponding to 1·8 × 102 to at least 1·2 × 106 CFU g−1 feed using qPCR (without preceding DNA extraction) or selective plating on XLD agar (Fig. 2). For the uninoculated control sample and the sample inoculated with 1·8 × 10CFU g−1 feed, Salmonella was not detected after flotation using either PCR or plating on selective agar plates.

Quantification of Salmonella in naturally contaminated feed samples using flotation-qPCR

To validate the performance of the flotation protocol for naturally contaminated samples, 15 bags of soya bean meal, taken from one batch, were investigated by duplicate analyses of feed from the same bag resulting in 30 data points (Table 2). The soya bean meal had previously been found positive for Salmonella using culture-based reference methods in the quality control at the feed producing company. By the flotation-qPCR method, Salmonella was found in 7 of the 30 subsamples analysed (6/15 bags). Four of the seven positive samples had levels below the limit of quantification (LOQ) 1·8 × 102 CFU g−1 feed and could therefore only be semiquantified to <1·8 × 102 CFU g−1 feed, whereas the remaining three were quantified to 1·8 × 102, 1·7 × 103 and 7·8 × 103 CFU g−1 feed, respectively.

Table 2. Results from quantitative and qualitative analysis of Salmonella in naturally contaminated soya bean meal samples using flotation combined with qPCR or selective plating on XLD plates, a modified most probable number combined with PCR (MPN-PCR) and a qualitative culture enrichment PCR method*
Sample no.No. of Salmonella recovered using method (CFU g−1 or MPN g−1)Results from culture enrichment PCR
Flotation + qPCRFlotation + plating on XLDMPN-PCR (95% CI)
  1. *ND: Not Detected; N/A: Not Applicable; † Below the limit of quantification.

  2. ‡+: Detection; −: No detection of Salmonella in 25 g feed.

16·9 × 101†/1·8 × 102ND/ND0·35 (0·092–1·3)+
22·0 × 101†/NDND/ND0·65 (0·17–2·6)+
3ND/NDND/NDN/A (> 5·2)+
4ND/9·3 × 101ND/ND0·14 (0·039–0·51)+
5ND/NDND/ND0·65 (0·17–2·6)
6ND/NDND/ND0·33 (0·086–1·3)+
7ND/1·7 × 103ND/ND0·074 (0·019–0·29)
8ND//NDND/ND1·8 (0·43–7·4)
9ND/NDND/ND0·024 (0·0035–0·17)+
10ND/NDND/ND0·084 (0·022–0·32)
11ND/NDND/ND0·33 (0·086–1·3)+
127·8 × 103/NDND/ND0·053 (0·012–0·23)+
13ND/NDND/ND0·024 (0·0035–0·17)
14ND/1·7 × 101ND/ND0·042 (0·0088–0·2)+
15ND/NDND/ND0·14 (0·04–0·51)

Quantification of Salmonella in naturally contaminated feed samples using MPN-PCR

Subsamples (50 g) from the same 15 bags as for flotation-qPCR were analysed in parallel with MPN-PCR (Table 2). Results show that Salmonella was found in all of the analysed samples, mostly with low levels (median 0·14 MPN g−1 and range 0·024–1·8 MPN g−1 for the quantifiable samples). One sample fell out of the quantification range due to too high concentration (all samples from the tested dilutions were positive) and could thus only be semiquantified to >5·2 MPN g−1 feed.

Detection of Salmonella in naturally contaminated feed samples using qualitative PCR

Subsamples (25 g) from the 15 bags of naturally contaminated soya bean meal samples were also analysed in parallel by a qualitative culture enrichment PCR method (Table 2). Results show that Salmonella was found in 9 of 15 bags. No PCR inhibition was noted for any of the samples.


In this study, three different pre-PCR processing approaches to investigate the presence and numbers of Salmonella in naturally contaminated feed samples were evaluated. The three methods were different with regard to their sampling volumes, time consumption, range of quantification, types of cells detected and the principle of analysis (Table 3). As the numbers of Salmonella in feed has been reported to be low, together with an uneven distribution of cells that are sublethally damaged, methods need to be adjusted to enable quantification and/or detection. Two of the tested methods (MPN-PCR and culture enrichment PCR) are both based on the same principles; pre-enrichment of feed in nonselective BPW followed by DNA extraction of an aliquot and real-time PCR analysis. However, in the culture enrichment PCR, it is only possible to determine whether the sample is positive or negative for the tested sample portion (25 g). If we assume that 1 CFU can be detected in the 25 g sample, this gives us a semiquantification of ≥1 CFU per 25 g (≥0·04 CFU g−1). MPN-PCR is based on repetitive analysis of different dilutions of the same sample, and by adjusting the dilution scheme, it was possible to obtain a method with a quantification range of 0·02–5·2 CFU g−1. Because both culture enrichment PCR and MPN-PCR rely on a culture-based enrichment step to multiply the Salmonella cells, they fail to detect cells that are not able to grow and multiply except at concentrations higher than 4 × 104 CFU g−1 (Löfström et al. 2004). These high levels have rarely been reported for Salmonella in feed. Thus, a culture-based approach will always underestimate the total number of living Salmonella in harsh environments where cells are likely to be damaged and nonculturable. For example, it has been shown that the culturable number of Salmonella cells in soya bean meal decline by 1–2 log CFU g−1 during storage (Koyuncu et al. 2013). This could be explained by cells entering into the VBNC state. Hence, there is a need to use alternative strategies to enable enumeration of the nonculturable fraction. Flotation has been proposed as an alternative pre-PCR processing strategy to use for this purpose.

Table 3. Comparison of the three evaluated methods for detection and quantification of Salmonella in feed samples
MethodAmount of sample (g)Total time of analysis (h)Detection level (CFU g−1 or MPN g−1)Range of quantification (CFU g−1 or MPN g−1)Type of cells detectedPrinciple of analysis
  1. N/A: Not Applicable.

  2. a

    Cells with density 1·05–1·10 g ml−1.

  3. b

    Cells that can grow in BPW at 37°C.

  4. c

    If sufficient amounts are present.

Flotation + qPCR2·541·8 × 1021·8 × 102 to at least 1·2 × 106

Culturable cellsa

Nonculturable cellsa

Separation of Salmonella cells from sample based on differences in densities.

Quantification of DNA in cells by PCR.

Comparison to standard curve.


Culturable cellsb

DNA from dead cellsc

Parallel analysis of several dilutions of the sample.

Growth in nonselective media.

Extraction of DNA.

Detection of DNA in cells by PCR.

Calculation of no. of cells based on MPN table.

Culture Enrichment PCR25200·04N/A

Culturable cellsb

DNA from dead cellsc

Growth in nonselective media.

Extraction of DNA.

Detection of DNA in cells by PCR.

The performance of the previously developed flotation protocol to isolate and enumerate Salmonella directly from the solid sample carcass swabs (Löfström et al. 2011) was in this study confirmed on a second sample type, soya bean meal. The buoyant density of most of the feed particles was found to be >1·102 g ml−1 which is similar to previously reported values for other types of biological samples (Lindqvist 1997; Fukushima et al. 2007; Wolffs et al. 2007; Löfström et al. 2011). Similar to when the assay was applied on carcass swabs, it was not possible to separate all background flora and sample parts from the Salmonella cells. As feed and carcass swabs are fundamentally different sample types, for example concerning the texture, moisture content and chemical composition, the generality of the flotation method has been demonstrated. Direct separation, detection and quantification of cells from a solid matrix rather than a preprocessed or homogenized sample lead to a better sensitivity compared with previously published protocols (Fukushima et al. 2007; Wolffs et al. 2007), mostly due to fewer steps in the analysis chain and a larger sample volume. The LOD for the feed flotation protocol is much higher compared with MPN-PCR, but comparable to other molecular techniques, as well as to direct plating on selective agar (Malorny et al. 2008). The LOQ is similar to the LOQ obtained for carcass swabs using the same protocol, whereas the LOD was higher (Löfström et al. 2011). However, because the LOD is estimated from few replicates at the lowest concentration (<100 CFU per sample), further studies are needed to confirm the LOD.

A PCR step was included as the final step in all three investigated methods, either to detect the presence of cells (MPN-PCR and culture enrichment PCR) or for quantification of the numbers of cells (flotation-qPCR). Using PCR presents several challenges, including the possibilities of PCR inhibition from substances present in biological samples (Hedman et al. 2013). In all three methods, this was taken into account by the inclusion of an internal amplification control (IAC) to control total inhibition of the PCR. An additional challenge is encountered when PCR is used for absolute quantification, applying a standard curve with known quantities. In this case, the standard curve can be skewed if one does not take into account the influence that different sample types have on, for example the linear range, detection level and amplification efficiency of the PCR step. It has previously been demonstrated that various feed materials contain substances that influence PCR and that this effect differs between different types of feed (Löfström et al. 2004; Koyuncu et al. 2010). In this study, it was shown that the standard curves for feed samples (using soya bean meal as a model) differed from the one that was previously being used for carcass swabs (Löfström et al. 2011). Carcass swabs analysed with the flotation protocol was shown not to have a negative influence on the PCR, whereas feed particles with the same buoyant density as for Salmonella, coextracted from the upper interphase after flotation, was shown to be PCR inhibitory. Due to this PCR-inhibitory effect, it was not possible to use the approach applied by Löfström et al. (2011) where a PCR standard curve based on cells that were added to pure flotation media and thereafter treated as the unknown samples extracted after flotation was used to calculate the no. of CFU in the sample from the qPCR Ct values. If there are PCR inhibitors that in a systematic manner affects the PCR amplification efficiency, as was noted for the feed samples, this might lead to a constant underestimation of the true cell counts. These results stress the importance of applying a separate standard curve for each matrix, or at least investigate the influences of the matrix on the results, similar to what has previously been suggested by others (Cocolin et al. 2011).

When comparing results obtained for artificially contaminated samples, flotation combined with qPCR systematically gives higher estimates of the numbers of Salmonella in feed compared with flotation followed by culture on selective agar plates. The same trend was noted for recovery of Salmonella in carcass swabs using the same protocol (Löfström et al. 2011), as well as for Staphylococcus aureus in milk samples analysed by flotation (Aprodu et al. 2011). Löfström et al. hypothesized that this was most likely due to the inability of damaged cells to grow under the applied selective pressure on the agar plates. This is also somewhat confirmed by the fact that it was not possible to grow any Salmonella on selective plates after flotation for naturally contaminated samples that were positive by PCR (Table 2). Several studies have shown that it is hard to recover Salmonella from naturally and artificially contaminated feed samples from PCR positive samples by culture-based methods (Löfström et al. 2004; Koyuncu et al. 2010). Other explanations might be that several cells aggregate giving rise to only one CFU when grown on an agar plate (Rossmanith and Wagner 2011), variable genome content of Salmonella (Lebaron and Joux 1994) and/or detection of DNA originating from dead cells. The last alternative is less plausible as it has been shown that free DNA and dead Salmonella cells have different densities compared with intact cells (Wolffs et al. 2005b, 2007).

The levels of Salmonella in the analysed batch of feed were shown to be low (median 14 MPN per 100 g) using the MPN-PCR approach (Table 2). The numbers obtained are comparable to the scarce data that can be found in the literature, estimating levels typically <50 MPN per 100 g or most often <1 MPN per 100 g feed (Smeltzer et al. 1980; D'Aoust and Sewell 1986; Sauli et al. 2005; Binter et al. 2011). However, when using flotation combined with qPCR numbers as high as 7·8 × 103 CFU g−1 Salmonella was obtained. The methods use different principles for detection of the target bacteria and different sample sizes which could explain some of the differences (Table 3). The results from the analysis on naturally contaminated soya bean meal indicated a poor correlation between repeated measurements from the same original 250 g sample. This agrees with observations from other lots (unpublished results) and indicates that the heterogeneous distribution of Salmonella may largely be due to a ‘nugget effect’, that is, aggregates of bacteria being bound to feed particles rather than a large-scale spatial distribution or ‘hot-spots’. The lot analysed in this study is likely to have been reloaded and mixed several times during its way from the crushing plant, which may account for the apparent lack of spatial distribution (i.e. correlation between repeated measurements from the same sample). Nevertheless, the results illustrate that measuring the levels of Salmonella in a feed lot with a reasonable precision require the analysis of several hundred grams of material in addition to representative sampling. Thus, the design of the MPN-PCR reflects that a labour intensive quantitative method with high precision in the estimate of Salmonella levels in one sample may be less fit for purpose than a method that allow the parallel analysis of multiple samples with moderate precision. The number of samples analysed in this study is limited and therefore these data cannot be used to estimate the levels of Salmonella in the feed lot. However, results obtained demonstrate the applicability of the three methods to estimate prevalence and numbers of Salmonella in feed ingredients.

In conclusion, this study compares the performances of three methods for detection and/or quantification of Salmonella in feed. The methods meet different needs. Culture enrichment PCR can be used for routine screening of Salmonella as it is a relatively simple and automated procedure (Löfström and Hoorfar 2012) and can also to some extent be used to assess the quantity of Salmonella in a batch of feed by repetitive analysis of multiple samples using a systematic sampling plan (Wierup and Häggblom 2010; Binter et al. 2011). MPN-PCR can be used when there is a need to quantify low numbers of Salmonella in feed with a somewhat easier protocol that is adjusted to quantify low levels compared with standard MPN techniques. Flotation-qPCR can be a useful tool when the aim is to extract nonculturable cells that will be used for further analysis downstream, for example source tracking, but is not useful for feed with low numbers due to the small volume analysed and the relatively high detection level. However, combining these tools gives a good opportunity to assess not only the prevalence of Salmonella in feed, but also the numbers of culturable, as well as nonculturable cells. Hereby, data can be generated that will give more accurate QMRA for Salmonella in the feed chain.


Kirsten Michaëlis is acknowledged for excellent technical assistance. The feed samples were kindly provided by Svenska Foder, Lidköping, Sweden and the BactXtractor medium by FertiPro, Belgium. This study was financially supported by the European Union project BIOTRACER (FOOD-2006-CT-036272) and the Swedish Research Council for Environment, Agricultural Sciences and Spatial Planning. The funders had no role in study design, data collection and analysis, conclusions, decision to publish, or preparation of the manuscript.

Conflict of interest

The authors have no conflict of interest.