Foodborne illness due to Shiga toxin-producing E. coli (STEC) continues to be an important public health concern in Canada and around the world (EFSA, 2013; Gould et al., 2013; Thomas et al., 2015). A sub-group of STEC, the enterohemorrhagic E. coli (EHEC), causes infections that can result in serious medical conditions including bloody diarrhea, hemolytic-uremic syndrome (HUS), kidney failure and microangiopathic hemolytic anemia, and can occasionally be fatal (Karmali et al., 2010; Thomas et al., 2015). Consumption of foods contaminated with EHEC is an important cause of illnesses associated with this pathogen. High-risk foods, such as ground beef and produce, are thought to become contaminated through exposure to animal fecal matter, particularly from ruminant animals in which STEC bacteria are prevalent (Gill and Gill, 2010; Mathusa et al., 2010; Duffy et al., 2014). Surveillance and recall of EHEC-contaminated foods reduces risk for the consumer, and improved methods will enable more extensive testing and further reduce human-health risk attributed to this organism (EFSA, 2013; Catford et al., 2014; Duffy et al., 2014; Seys et al., 2015).

The most common EHEC is E. coli O157, which has been the focus of public health organizations. However, non-O157 EHEC foodborne illnesses have been increasingly identified (Johnson et al., 2006; Gould et al., 2013; Catford et al., 2014; Luna-Gierke et al., 2014). There are no biochemical features by which EHEC strains can be differentiated from commensal E. coli or other STEC that are not a public health concern. Nonetheless, it is universally recognized that foodborne EHEC can generally be defined on the basis of certain gene markers, including the Shiga toxin genes, stx1 or stx2, the intimin-coding gene, eae, along with markers for specific serogroups of concern (e.g., O26, O45, O103, O104, O111, O121, O145, and O157) (Blais et al., 2012; EFSA, 2013; Catford et al., 2014). Note that while most EHEC strains have both eae and stx genes, priority serogroups vary among countries.

The method for detection and isolation of EHEC used in food-testing laboratories at the Canadian Food Inspection Agency (Gill et al., 2012; Huszczynski et al., 2013; Blais et al., 2014a,b) as well as the ISO/CEN TS13136:2012 Technical specification (ISO, 2012) and the US MLG5B.05 (USDA-FSIS, 2014) methods, commonly used internationally, involve enrichment of samples in a selective broth and screening for the presumptive presence of EHEC using PCR methods targeting stx1 and/or stx2, and eae. One of the challenges of screening enrichment broths for EHEC is to distinguish samples with target EHEC carrying both stx and eae from samples containing mixed cultures in which these markers are present in different cells. Using current approaches, as much as 50% of samples identified as presumptive positives may be false positives, particularly in samples with high levels of non-target E. coli (Livezey et al., 2015; Delannoy et al., 2016). This high rate of false positives can negate the benefit of a screening procedure intended to identify presumptive EHEC (i.e., E. coli cells carrying both stx and eae) due to the need for unnecessary downstream processing of samples for the recovery and characterization of the target bacteria.

The aim of this work was to develop a screening method capable of distinguishing enrichments positive for EHEC from false-positive samples containing mixed cultures of eae-negative STEC and eae-positive E. coli using droplet digital PCR (ddPCR) technology (Figure 1). ddPCR is a quantitative PCR technique in which a standard PCR reaction mixture is distributed into thousands to millions of droplets prior to PCR amplification (Hindson et al., 2011; Pinheiro et al., 2012). The Bio-Rad QX200^TM implementation of this technology involves conversion of 20 μL reaction volumes into approximately 20 thousand one-nanoliter droplets followed by PCR amplification within each droplet. The concentration of the input sample is adjusted to achieve a distribution of less than one template molecule per droplet with a Poisson distribution of the template molecules among droplets (Pinheiro et al., 2012). TaqMan probe-based PCR assays are commonly used in this method, and amplification is determined by detection of fluorescence in each droplet by a droplet reader. The number of positive droplets indicates the number of template molecules in the sample and linkage between different targets (e.g., eae, stx) can be determined based on frequency of co-amplification of two targets within a droplet. In samples containing EHEC, co-amplification of eae/stx targets would be expected to be high; whereas in false-positive samples (i.e., mixtures of eae-negative STEC and eae-positive E. coli) the two targets would generally be amplified in separate droplets, with only a small number of droplets in which both targets were amplified due to presence of two different bacterial cells.

Here we describe the development of a Multiplexed, Single Intact Cell ddPCR (MuSIC ddPCR) assay targeting detection of stx and eae genes and ensuring presence of the two targets within a single bacterial cell. In this assay, suspensions of bacterial cells in enrichment broth are dispersed into droplets, followed by release of gDNA by heat lysis, and ddPCR analysis. The method was evaluated to determine performance for specific identification of EHEC in enrichment-broth samples containing excess amounts of non-target bacteria.

The MuSIC ddPCR assay represents a novel approach for the detection of two genetic targets within a single, intact, bacterial cell. To our knowledge, this is a new application of ddPCR technology. In this study, MuSIC ddPCR has been applied to the specific detection of EHEC in enrichment broths, and was shown to be able to distinguish samples contaminated with EHEC from samples containing mixed cultures of eae-negative STEC and eae-positive E. coli, even in samples with low proportions of target EHEC (e.g., 0.01% of the total bacterial population). This method is intended for samples in which target EHEC are present at high concentrations (e.g., 10–500 cfu/μL), and would be appropriate for screening enrichment broths derived from foods.

One of the early challenges in the development of this method was the lot-to-lot variability in the performance of the degenerate primers used in the original implementation of the method. While this degeneracy may not impact standard qPCR methods, for the ddPCR assay, the reduced concentrations of specific primers likely led to the exhaustion of these reagents in individual droplets. Specific primers performed more consistently, even in samples in which there were up to 3 mismatches between primers and primer binding sites. Note that this assay is not appropriate for the detection of the stx2f subtype of the stx2 gene. Due to the high variability of the stx2f subtype relative to the other stx2 subtypes, none of the methods currently deployed for the detection of EHEC are capable of detecting this subtype (Wasilenko et al., 2012; Blais et al., 2014b). Given the evidence that EHEC with this toxin variant are associated with clinical infections (Jenkins et al., 2003; van Duynhoven et al., 2008; Friesema et al., 2015), improvement of current methods through the addition of specific stx2f primers and probes may be of value.

The use of intact bacterial cells was required to detect linkage between the eae/stx targets indicative of the presence of EHEC. These targets were not linked when gDNA extracts were used as template in the assay, probably due to fragmentation of the chromosomal DNA during its preparation (Figure 3). The stx and eae genes are widely dispersed on the E. coli chromosome (typically over 1 Mb); therefore, it is highly probable that chromosomal breakage between these targets would occur during the extraction procedure (Brisco et al., 2010; Malentacchi et al., 2015), and that the fragments would segregate into different droplets. In the case of the intact cell assay, the contents of the entire cell would be present within a droplet, and linked targets would remain in the droplet, even if the gDNA became degraded after the cell lysis. This is consistent with the higher linkage values observed for intact cells compared to gDNA (Figure 3). Optimization of droplet ddPCR often requires development of conditions in which there is a clear distinction between positive and negative partitions (Huggett et al., 2013). While this partitioning could be easily achieved with the gDNA extracts, more variability in the fluorescence amplitude intensity was observed when cell cultures were used.

In the conception of this method, it was predicted that amplification of two targets would be observed for every droplet containing an intact EHEC cell. In practice though, dual amplification was lower than predicted, ranging from 23 to 79% of the droplets expected to be positive for both targets, with the lowest linkages observed for undiluted food enrichment broth samples. This was not due to reagent limitations, or the use of intact cells, as dual target amplification was observed in 87–99.9% of control samples of E. coli transformed with a plasmid containing both targets. Restriction digestion of gDNA templates is commonly recommended for ddPCR methods to relieve tertiary structure and improve target accessibility (Hindson et al., 2011) and so this approach was evaluated. While conditions for the restriction digestion were not optimal due to the need to apply heat to lyse cells at the beginning of the reaction, thereby partially inactivating the restriction enzymes, significant increases in the fluorescent intensities of the stx-positive droplets were observed (Figure 4B) indicating an improvement in the amplification efficiency. This improvement may be due to reduction in tertiary structure near the stx genes, improving accessibility target sites. In this study, this did not translate into an increased proportion of droplets in which dual-amplification of the targets was observed; nonetheless, the use of thermostable restriction enzymes should be explored to determine if further improvement to the partitioning of the positive and negative droplets could be achieved. Another possible explanation for the low proportion of droplets with dual-target amplification is that degraded DNA released from dead cells provided single-target templates for this reaction. Removal of free gDNA could potentially be achieved by adding a DNase step before the cell lysis. Although variability in percentage linkage was high in the experiments conducted in this study, linkage in non-target mixed culture samples was consistently less than 2%, and EHEC-positive samples could be easily identified using a conservative cut-off value of >10% linkage.

The use of cultures of intact bacterial cells and food enrichments in the MuSIC ddPCR assay instead of purified gDNA presented a number of challenges, likely due to presence of debris and particles in these samples. In this study, up to 6% of samples were lost due to blockages within capillaries which prevented droplet formation. In addition, droplet counts were not consistent. Under ideal conditions, droplet counts should approach 20,000, but when using cell cultures and enrichment broths, droplet counts varied from 10,000 to as high as 28,000. This is not surprising as the ddPCR reagents were developed for purified samples, and would not be expected to be optimal for these samples. Even with the use of purified gDNA, higher concentrations can increase viscosity of the sample, changing the average volume of the droplets (Hindson et al., 2011). In most cases, the problems with droplet generation were overcome with the use of 3–4 replicates per sample. The performance of the assay was also variable depending on the broth used for growth of the bacteria. Mean fluorescence amplitude was higher when cells were suspended in nutrient broth relative to BHI, mTSB and PBS (not shown). Partitioning of positive and negative droplets was also impacted in enrichments broths generated from leafy-green produce such as spinach or kale (Figure 7C), where dark green pigments were observed in the enrichment broths.

Despite apparent inefficiencies with the use of intact cells in the MuSIC ddPCR assay, the application of the percent linkage value, with positives defined as samples with greater than 10% linkage between stx and eae targets, provided a robust metric for distinguishing EHEC-positive samples from mixed cultures containing targets in different cells, even in samples containing high levels of non-target bacteria (e.g., 10,000-fold excess). In samples of mixed strains that would have generated false positives by traditional PCR-based methods, linkage between the two targets was extremely low, less than 0.5% in most samples. The use of undiluted food enrichments had some impact on the performance of the assay. For example, the higher degree of variability in percent linkage values for EHEC mixed at 1:1000 with food enrichment broths relative to the 1:100 mixtures (Figure 7A) may be due to the lower dilution used to achieve the optimal concentration of EHEC in the 1:1000 samples. Nonetheless, even in these samples, accurate identification of EHEC was achieved. Use of a positive control will be important for evaluating the impact of inhibition in different food matrices. This proof-of-concept study shows promising results for the use of MuSIC ddPCR with food enrichments broths containing food debris and high relative proportions of background bacteria. The performance of the method was robust in food-enrichment samples where concentration of target cells was between 50 and 2500 cfu/reaction. While this is a limited dynamic range for detection compared to other PCR-based approaches, in practice testing of up to three dilutions of the enrichment broths will enable accurate detection of EHEC-positive samples.

False-positive detection of EHEC is a significant challenge for food-microbiology laboratories (Krause et al., 2005; Shelton et al., 2006; Feng et al., 2010; Alonso et al., 2011; Delannoy et al., 2016). For example, in a study including 1739 beef enrichment broths, approximately half of the 180 enrichment broths positive for stx and eae markers were false-positives, and did not contain EHEC (Delannoy et al., 2016). The work required for the analysis of these false-positive samples is onerous, as many colonies would need to be screened to confirm either the presence or absence of EHEC in the sample. To achieve higher specificity, the Canadian (Blais et al., 2014b), ISO/CEN TS13136:2012 Technical specification (ISO, 2012) and the US MLG5B.05 (USDA-FSIS, 2014) methods also detect O-serogroup specific markers for priority serovars, that vary among jurisdictions. Unfortunately, serogroup markers can also be present in non-pathogenic strains, and not all of the clinically important EHEC fall within the priority serogroups (EFSA, 2013). New screening methods that use additional genes, associated with typical EHEC, have been effective in reducing false-positive detection of EHEC in food enrichment broths (Delannoy et al., 2013, 2016), but these rely on the detection of a number of different genes known to be associated with EHEC, none of which (individually) are present in all EHEC variants. Specific detection of samples containing typical EHEC, which are eae- and stx-positive, would enable identification of emerging pathotypes in foods.

The MuSIC ddPCR EHEC assay enables accurate detection of EHEC in enrichment broths based on detection of two diagnostic genetic markers (stx and eae) and determination that these occur in a single cell. While this method is somewhat more complex than standard PCR approaches, the benefit in reducing false-positive detection of EHEC outweigh the challenges associated with the implementation of the method. A more extensive validation of the assay with priority food types (e.g., leafy-green produce and raw meat) will be undertaken to enable deployment to food-testing laboratories. Integration of additional targets such as the stx2f toxin subtype and/or the aaiC and aggR markers associated with the seropathotype responsible for the 2011 European O104 outbreak (EFSA, 2013) may also be of value. The reduction in false-positive detection of EHEC associated with the application of the MuSIC ddPCR EHEC assay will enable high throughput screening for EHEC in food samples, ultimately reducing consumer exposure to EHEC-contaminated products.

Conceived and designed the experiments: TM, CC, BB, and AW. Performed the experiments: TM. Analyzed the data: TM and CC. Contributed reagents, materials, analysis tools: CC and BB. Wrote the paper: TM, CC, BB, and AW.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.