Salmonella enterica and Escherichia coli in Wheat Flour: Detection and Serotyping by a Quasimetagenomic Approach Assisted by Magnetic Capture, Multiple-Displacement Amplification, and Real-Time Sequencing

Enterohemorrhagic Escherichia coli (EHEC) and Salmonella are of serious concern in low-moisture foods, including wheat flour and its related products, causing illnesses, outbreaks, and recalls. The development of advanced detection methods based on molecular principles of analysis is essential to incorporate into interventions intended to reduce the risk from these pathogens. In this work, a quasimetagenomic method based on real-time sequencing analysis and assisted by magnetic capture and DNA amplification was developed. This protocol is capable of detecting multiple Salmonella and/or E. coli organisms in the sample within less than a day, and it can also generate sufficient whole-genome sequences of the target organisms suitable for subsequent bioinformatics analysis. Multiplex detection and identification were accomplished in less than 20 h and additional whole-genome analyses of different nature were attained within 36 h, in contrast to the several days required in previous sequencing pipelines.

ABSTRACT

Food safety is a new area for novel applications of metagenomics analysis, which not only can detect and subtype foodborne pathogens in a single workflow but may also produce additional information with in-depth analysis capabilities. In this study, we applied a quasimetagenomic approach by combining short-term enrichment, immunomagnetic separation (IMS), multiple-displacement amplification (MDA), and nanopore sequencing real-time analysis for simultaneous detection of Salmonella and Escherichia coli in wheat flour. Tryptic soy broth was selected for the 12-h enrichment of samples at 42°C. Enrichments were subjected to IMS using beads capable of capturing both Salmonella and E. coli. MDA was performed on harvested beads, and amplified DNA fragments were subjected to DNA library preparation for sequencing. Sequencing was performed on a portable device with real-time basecalling adaptability, and resulting sequences were subjected to two parallel pipelines for further analysis. After 1 h of sequencing, the quasimetagenomic approach could detect all targets inoculated at approximately 1 CFU/g flour to the species level. Discriminatory power was determined by simultaneous detection of dual inoculums of Salmonella and E. coli, absence of detection in control samples, and consistency in microbial flora composition of the same flour samples over several rounds of experiments. The total turnaround time for detection was approximately 20 h. Longer sequencing for up to 15 h enabled serotyping for many of the samples with more than 99% genome coverage, which could be subjected to other appropriate genetic analysis pipelines in less than a total of 36 h.

IMPORTANCE Enterohemorrhagic Escherichia coli (EHEC) and Salmonella are of serious concern in low-moisture foods, including wheat flour and its related products, causing illnesses, outbreaks, and recalls. The development of advanced detection methods based on molecular principles of analysis is essential to incorporate into interventions intended to reduce the risk from these pathogens. In this work, a quasimetagenomic method based on real-time sequencing analysis and assisted by magnetic capture and DNA amplification was developed. This protocol is capable of detecting multiple Salmonella and/or E. coli organisms in the sample within less than a day, and it can also generate sufficient whole-genome sequences of the target organisms suitable for subsequent bioinformatics analysis. Multiplex detection and identification were accomplished in less than 20 h and additional whole-genome analyses of different nature were attained within 36 h, in contrast to the several days required in previous sequencing pipelines.

INTRODUCTION

The frequency and level of foodborne pathogen contamination of wheat flour have been recently determined to be greater than it was traditionally believed, and this food commodity is now considered capable of posing risk to consumers’ health (1–4). In the last 5 years, several flour-related recalls and outbreaks have been reported in different countries (5–9). Wheat flour is not normally consumed raw, but several infections have been due to cross-contamination before cooking or baking. In addition to cross-contamination, raw flour is an important component of uncooked dough, which is sometimes used as a component of cold treats, and some consumers ingest cookie dough and batter mix before baking (10). Hence, industry and health care authorities understand the necessity of developing new strategies for mitigation of microbiological risk in flour production, as well as improving its production safety measures in terms of monitoring and surveillance methods.

Salmonella and enterohemorrhagic Escherichia coli (EHEC) are the two bacterial foodborne pathogens linked to emerging gastroenteritis outbreaks caused by wheat flour (7, 11). Salmonella spp. are facultative anaerobic bacteria that include more than 2,600 serotypes and are linked to multiple types of foods (12). EHEC is a group of pathogenic E. coli capable of causing enterohemorrhagic colitis (13). E. coli serotypes O157, O26, O45, O103, O111, O121, and O145 are the most virulent Shiga toxin-producing bacteria that may cause complications such as hemolytic uremic syndrome. A recent extended surveillance study that tested more than 5,000 wheat kernel samples reported significant contamination levels of wheat with pathogens, especially Salmonella and E. coli, stressing the potential for wheat flour as a foodborne illness risk (14).

The detection of foodborne pathogens in foods is achieved by numerous methods (15). Conventional methods such as culture media are still considered the gold standard in many cases. However, they usually need several days to complete and have certain limitations in terms of specificity. Antibody-based methods are rapid and simple, but alone they may lack the sensitivity and specificity that processors require to monitor their products (16). Immunomagnetic separation (IMS) methods use immunomagnetic beads as the capture reagents to isolate pathogens from food samples. IMS beads have antibodies attached to their surface and are complementary methods used to improve sensitivity and/or specificity of detection protocols (17). In food safety detection, one of the first successful applications of IMS was to dramatically increase the recovery of E. coli O157:H7 from food and cattle fecal samples, revealing a markedly greater prevalence in bovine populations (18).

For over two decades, molecular methods such as PCR have offered unique advantages over other detection protocols. These DNA/RNA-based methods are fast, accurate, and dependable and have constantly improved by advancements such as different real-time PCR chemistries, multiplex reactions, and extended automation (19). However, they can have high running costs, need sophisticated laboratory equipment, and are limited in the numbers of targets detected.

The continuous evolution of molecular methods has led to whole-genome sequencing (WGS) and metagenomics analysis-based detection and identification of pathogens in food matrices with abilities far more than just detection, including serotyping, characterization, phylogenetic placement, outbreak source tracking, and generating data on other organisms present in addition to the target cells (20, 21). As discussed above, these methods also need high-tech machines with relatively elevated costs of acquisition and maintenance and require extensively trained personnel. However, new technologies which are more affordable and easier to use continue to be developed.

Most standard detection and/or subtyping methods based on whole-genome sequencing (WGS) have traditionally required long enrichment of samples, isolation of the individual target organism, sophisticated DNA extraction and purification, sequencing library preparation, and, finally, performing a sequencing run for several days before the resulting sequences could be subjected to basecalling and any further analysis (22). This would lead to a total of up to 10 days or even longer for the entire process of pathogen detection/characterization from a contaminated food.

More recently, metagenomics sequencing has been demonstrated as an effective new approach for isolation-independent identification of foodborne pathogens such as Salmonella, E. coli, and Listeria from food samples (23–26). In this approach, pathogen detection and further subtyping are combined into a single workflow. However, analysis can be done and results obtained only after the sequencing run has been completed. Also, food composition and growth competition may negatively affect the detection.

One of the most recent advancements in sequencing technology is commercialization of a portable nanopore sequencing device which also has the ability of analyzing sequence data in real-time. As this technology has continuously evolved, it has developed a high degree of accuracy and can even be applied to field samples, leading to results in a few hours with limited laboratory equipment at hand (27, 28). In our previous study, we successfully developed a nanopore sequencing-based approach with real-time analysis capability for detection and subtyping of Salmonella in food for the first time (20).

In this study, we aimed to develop a robust method for extended simultaneous detection of Salmonella and E. coli in wheat flour by combining short-term enrichment, IMS, and multiple-displacement amplification (MDA) with the nanopore sequencing technology. This approach differs from conventional metagenomics sequencing by a combination of selective concentration and the nanopore shotgun sequencing of enriched IMS-MDA products. The approach was evaluated for multiple artificially inoculated flour samples and target organisms using a nanopore sequencing device followed by standard bioinformatics analysis and, for comparison, the bioinformatics analysis pipelines provided with the nanopore device.

RESULTS

Short-term enrichment.

The optimal medium for short-term enrichment of Salmonella and E. coli was selected by comparing lauryl sulfate broth (LSB), A1 broth (A1), E. coli broth (EC), tryptic soy broth (TSB), and buffered peptone water (BPW). Due to the low efficiency of A1 and EC for E. coli enrichment, they were not further tested for Salmonella. As seen in Table 1, TSB was the best medium for our purpose and hence was used for all subsequent experiments. It was always significantly better than the other enrichment media regardless of the bacteria or strain differences (P ≤ 0.05). Interestingly, in short-term enrichment at 42°C with 250 rpm shaking, E. coli seemed to grow faster than Salmonella based on the growth rates of strains used in this study.

TABLE 1.

Comparison of different broth media for short-term enrichment of Escherichia coli and Salmonella at 42°C^a

Enrichment time (h)	Mediab	Bacterial strain final count (log CFU/ml)c
		E. coli O157:H7	E. coli O26:H11	E. coli O121	Salmonella Typhimurium	Salmonella Tennessee
4
	A1	1.85a	3.45b	2.62ab	NA	NA
	EC	ND	2.38a	ND	NA	NA
	LSB	2.36a	3.41b	2.17a	3.80ab	4.24b
	BPW	3.82b	4.05c	3.99c	3.68a	4.11a
	TSB	5.24c	5.28d	5.34d	3.92b	4.50c
8
	A1	3.3A	5.43B	3.00A	NA	NA
	EC	ND	4.53A	ND	NA	NA
	LSB	4.16B	6.14BC	3.08A	7.06A	8.23A
	BPW	7.57C	7.44D	7.32B	7.33B	8.12A
	TSB	8.26D	8.35E	8.27C	7.57C	8.60B

^aInitial counts were approximately 1 log CFU/ml in all cases.

^bA1, A1 broth; EC, E. coli broth; LSB, lauryl sulfate broth; TSB, tryptic soy broth; BPW, buffered peptone water.

^cND, not detected; NA, not available. Values are the mean log-transformed numbers of cells per ml from at least 3 measurements. Different lowercase letters within the 4-h enrichment and different uppercase letters within the 8-h enrichment represent significantly different values (P ≤ 0.05).

IMS bead evaluation.

Nine out of fifteen E. coli strains were recovered with the IMS beads from pure cultures at more than 50% capture efficiency (CE), and five additional strains were captured at at least 30% CE (Fig. 1). Only one E. coli strain, O145 (TW08087), had a CE of 10%. All ten Salmonella strains were recovered at a minimum of 40% CE. Only one Salmonella strain was captured at more than 90% CE, compared to five E. coli strains. When the IMS capture test was applied to other bacterial genera that included ten Gram-negative and Gram-positive strains, only a Pseudomonas strain and a Lactococcus lactis strain were recovered at 10 and 12% CE, respectively. Either the remaining strains were not captured at all or the CEs were extremely low.

FIG 1.

The capture efficiency (CE) of immunomagnetic (IMS) beads for individual pure culture suspensions of Escherichia coli, Salmonella, and nontarget bacterial strains. CE was determined by dividing the viable count recovered after plating IMS beads by the initial count mixed with the beads.

Rapid extended detection of E. coli and Salmonella in flour using nanopore sequencing real-time analysis.

The total turnaround time for preparation of samples for sequencing was approximately 20 h, consisting of a 12-h short enrichment, 4.5 h for IMS-MDA, and 3 h for library preparation for nanopore sequencing. Performing sequencing on flour samples artificially inoculated with a single strain of E. coli or Salmonella for only 1 h generated approximately 15,000 reads. The average size of the reads was approximately 3 kb, with the majority of reads at 1 to 4.5 kb. However, there was also a range of reads from 4 to 10 kb that consisted on average of 10% of the total number of the reads.

After only 1 h of sequencing, the developed approach was used for the phylogenetic identification of wheat flour samples artificially inoculated with species of E. coli and Salmonella using the EPI2ME software on the Metrichor pipeline. Figure 2 depicts the taxonomic classification tree of the reads of two separate flour samples that had been inoculated separately with E. coli O157:H7 (ATCC 43895) and Salmonella Typhimurium (ATCC 14028). Even though the reads originating from naturally present Bacillus were predominant, the system was able to correctly discriminate both target bacteria. More interestingly, the highest hit numbers in the taxonomical listing of EPI2ME results were E. coli O157:H7 and Salmonella enterica serotype Typhimurium, respectively.

FIG 2.

Taxonomic classification tree identification of Escherichia coli (A) and Salmonella (B), individually inoculated to wheat flour at a minimum abundance cut-off ratio of 3% generated by EPI2ME after 1 h of sequencing in a nanopore sequencer. Flour samples were inoculated with 0.1 log CFU/g E. coli O157:H7 or Salmonella Typhimurium strains, enriched in tryptic soy broth at 42°C, and subjected to immunomagnetic separation and multiple displacement amplification before sequencing.

Sequencing results for flour samples inoculated with two more strains each of E. coli and Salmonella were also correct both for phylogenetic identification at the species level and in taxonomical listing. Namely, E. coli O145 (TW08087) and O26 (I2016003853) were both classified as E. coli in the phylogenetic tree generated and received highest read hits as E. coli O145:H28 and E. coli O26:H11, respectively. Also, the two Salmonella strains used (Newport and Tennessee) were identified correctly.

The sequencing run was continued for 15 h and data were obtained and analyzed at 1, 3, 6, 10, and 15 h run-time points. The total yield after a 15-h run for each barcode was on average 1 GB that consisted of 160,000 to 300,000 reads per barcode. The pore occupancy score on the flow cell at the beginning of the sequencing run was always >78% and most of the time was over 90%, gradually decreasing as time passed. The average quality score defined by the EPI2ME software was 9.5 to 10.5 in different runs performed. When the protocol was applied to uninoculated flour samples, no E. coli or Salmonella was detected. Figure 3 shows a taxonomic classification tree of a random uninoculated flour sample at a minimum abundance cut-off ratio as low as 0.1%. For control samples, sequencing was stopped after 7.5 h.

FIG 3.

Taxonomic classification tree of a control uninoculated wheat flour sample at a minimum abundance cut-off rate of 0.1% generated by EPI2ME after 7.5 h of sequencing in a nanopore sequencer.

This approach was evaluated for simultaneous detection of E. coli and Salmonella coinoculated into flour at 0.1 log CFU/g. Figure 4 shows codetection and identification of E. coli and Salmonella on a phylogenetic tree for E. coli O157:H7 (ATCC 43895) and Salmonella Typhimurium (ATCC 14028) generated by EPI2ME software on the Metrichore pipeline. The highest hit numbers in the taxonomical listing of EPI2ME results were also E. coli O157:H7 and S. enterica serotype Typhimurium, respectively, representing the exact match with strains used for inoculation.

FIG 4.

Phylogenetic tree of wheat flour samples coinoculated with Escherichia coli and Salmonella with 0.1 CFU/g of each organism at a minimum abundance cut-off rate of 0.5% generated by EPI2ME after 1 h of sequencing in a nanopore sequencer. Flour samples were inoculated with 0.1 log CFU/g E. coli O157:H7 and Salmonella Typhimurium strains, enriched in tryptic soy broth at 42°C, and subjected to immunomagnetic separation and multiple displacement amplification before sequencing.

Bioinformatics analysis.

For each inoculated flour sample at 1 log CFU/g E. coli and Salmonella inoculated either individually or in dual mode, a phylogenetic tree based on core genome single-nucleotide polymorphisms (SNPs) was constructed at each analysis time point (1, 3, 6, 10, and 15 h). Interestingly, data obtained and analyzed after only 1 h of sequencing run were enough to generate a tight clustering of the sample (target) and a sample of the spiked strain (reference). Figure 5 shows two representative phylogenetic trees of the samples analyzed at 1 h of sequencing run.

FIG 5.

Phylogenetic clustering of inoculated flour samples inoculated at 0.1 log CFU/g, subjected to nanopore sequencing on a MinIon device in a quasimetagenomic approach. (A) Phylogenetic tree of Salmonella Typhimurium (ATCC 14028) and its tight clustering with the reference strain obtained from a 1-h sequencing run of Salmonella individual inoculation, subjected to analysis. (B) Phylogenetic tree of E. coli O157:H7 (ATCC 43895) and its tight clustering with the reference strain obtained from a 1-h sequencing run of E. coli and Salmonella dual inoculation, subjected to analysis. Numbers correspond to the closest GenBank strains for reference. Additional data are in Table S1 in the supplemental material.

The average microbial abundances for E. coli and Salmonella using the developed approach were approximately 46 and 15%, respectively. Figure 6 shows the microbial abundance at the genus level resulting from analysis of two different single-target spiked flour samples at 1 log CFU/g inoculum level. The changes in these percentages were not affected by the sequencing run time. However, when dual inoculations of targets were performed, E. coli abundance was not affected while Salmonella abundance was reduced to ∼2%.

FIG 6.

Microbial abundance at the genus level for two different flour samples inoculated at 0.1 CFU/g with E. coli and Salmonella, respectively, resulting from nanopore sequencing using a Minion device followed by a conventional bioinformatics analysis pipeline. Additional data are in Table S2 in the supplemental material.

Table 2 shows in detail the results of this IMS-MDA-assisted quasimetagenomics approach by the conventional bioinformatics pipeline after different Minion run times for samples containing individual inoculations. In contrast to the microbial abundance of the targets, which was not affected by the sequencing run time, genome coverage and hence sequencing depth and serotyping results were improved as the sequencing run time progressed. Data obtained and analyzed following a 15-h run time resulted in genome coverage of approximately 100% for both E. coli and Salmonella, regardless of single or dual inoculation at 1 log CFU/g (Tables 2 and 3). Regarding serotype prediction, any run time more than 3 h was enough to predict the exact serotype for Salmonella using the SeqSero software. In the case of E. coli, 10- and 15-h run times were necessary to predict O157 using the SerotypeFinder software in single and dual inoculated samples, respectively, using the O157:H7 strain in this study (Tables 2 and 3).

TABLE 2.

Metagenomics analysis at different sequencing run times using conventional bioinformatics pipeline^h

Sequencing time (h)	Raw-read serotypinga	% Genome coverageb	Sequencing depthc	Output (Mb)d	% Readse	% Abundancef	N50/total length (bp)g
Salmonella
1	N/A	64.82	1.17	37.66	12.66	14.49	N/A
3	(4:i:1,2)	91.23	2.92	93.65	12.68	14.54	N/A
6	(4:i:1,2)	96.43	4.27	136.83	12.75	14.60	N/A
10	(4:i:1,2)	99.55	8.15	260.79	12.74	14.60	N/A
15	(4:i:1,2)	100.00	18.82	602.21	12.75	14.60	4976/51793
E. coli
1	N/A	87.65	2.51	32.28	36.17	46.42	3255/3255
3	N/A	98.59	6.24	80.09	36.24	46.29	3233/3233
6	N/A	99.40	9.08	116.82	36.11	46.17	3591/3591
10	O157	99.70	17.23	221.97	36.10	46.16	6617/6617
15	O157	99.78	40.04	515.47	36.16	46.10	6076/78050

^aSerotype of Salmonella was predicted using SeqSero2, and serotype of E. coli was predicted using SerotypeFinder.

^bPercentage of reference genome that was mapped by sequencing reads.

^cThe ratio between total size of target sequences and the size of the reference genome.

^dTotal size (million bases) of raw read output.

^ePercentage of reads for each species (Salmonella/E. coli) among all sequencing reads.

^fPercentage of Salmonella/E. coli reads among all bacterial reads.

^gN₅₀ was calculated from de novo assemblies of sequencing reads classified as Salmonella or E. coli. Total length was the length of de novo assemblies.

^hFlour samples were inoculated with 0.1 log CFU/g E. coli O157:H7 or Salmonella Typhimurium strains, enriched in tryptic soy broth at 42°C, and subjected to immunomagnetic separation and multiple displacement amplification before sequencing. N/A, not available.

TABLE 3.

Metagenomics analysis at different sequencing run times using conventional bioinformatics pipeline^h

Sequencing time (h)	Raw-read serotypinga	% Genome coverageb	Sequencing depthc	Output (Mb)d	% Readse	% Abundancef	N50/total length (bp)g
Salmonella
1	N/A	67.93%	1.38	332.52	1.76	1.89	N/A
3	(4:i:1,2)	95.60%	4.07	987.70	1.75	1.88	N/A
6	(4:i:1,2)	97.42%	4.87	1180.6	1.75	1.88	N/A
10	(4:i:1,2)	99.53%	8.08	1967.6	1.75	1.87	N/A
15	(4:i:1,2)	99.89%	12.14	2948	1.76	1.87	N/A
E. coli
1	N/A	83.12%	2.13	39.19	23.27	33.61	3127/3127
3	N/A	98.63%	6.22	118.75	22.41	32.99	3074/3074
6	N/A	99.17%	7.42	141.83	22.37	32.92	3130/3130
10	N/A	99.64%	12.63	235.93	22.87	33.20	5152/10131
15	O157	99.70%	19.09	347.89	23.68	33.78	3789/12073

^aSerotype of Salmonella was predicted using SeqSero2, and serotype of E. coli was predicted using SerotypeFinder.

^bPercentage of reference genome that was mapped by sequencing reads.

^cThe ratio between total size of target sequences and the size of the reference genome.

^dTotal size (million bases) of raw read output.

^ePercentage of reads for each species (Salmonella/E. coli) among all sequencing reads.

^fPercentage of Salmonella/E. coli reads among all bacterial reads.

^gN₅₀ was calculated from de novo assemblies of sequencing reads classified as Salmonella or E. coli. Total length was the length of de novo assemblies.

^hFlour samples were coinoculated with 0.1 log CFU/g E. coli O157:H7 and Salmonella Typhimurium strains, enriched in tryptic soy broth at 42°C, and subjected to immunomagnetic separation and multiple displacement amplification before sequencing. N/A, not available.

DISCUSSION

Enrichment has been extensively used in multiple microbiological protocols as an amplification step to increase the chances of foodborne pathogen detection in food samples. Verhaegen et al. (29) reported that BPW worked better to enrich stressed Shiga toxin-producing E. coli (STEC) than other broth media they tested, including TSB. This is in contrast with our results of better performance in reaching larger numbers with TSB for flour. Similar to the results of this work, the same authors also reported that the use of selective media led to slower growth and lower performance. Use of nonselective media for this study was also based on the fact that it could improve early recovery (30).

Wang et al. (31) reported the suitability of TSB for STEC enrichment in combination with IMS and conventional plating. Stromberg and colleagues reported that TSB could successfully be used for STEC enrichment of fecal samples, while a modified version could further improve the enrichment capability (32). Comparison of TSB with other enrichment media in food samples has also been reported (33), emphasizing the suitability of TSB. In this work, we evaluated TSB for short enrichment of both E. coli and Salmonella. Although TSB was able to successfully enrich both target organisms significantly better than other broth media, E. coli growth was slightly faster both at 4- and 8-h enrichment times.

After the suitable medium for short-term enrichment was identified, a fixed 12-h enrichment time was used instead of testing different times. This was due to the results from our previous studies in which a 12-h enrichment was discovered to be the shortest time in which combination of enrichment, IMS, and MDA was capable of efficiently detecting target organisms at low inoculum levels while also supporting a good genome coverage where sequencing was to be performed (20, 34). Performance of IMS concentrated the target organism and boosted the enrichment. However, only by addition of MDA sequencing were we able to identify the target organism after a short run time and, more importantly, MDA addition resulted in improvement of the sequencing depth and quality. The methodology used in those previous publications was also applied in the present study. To confirm suitability of the enrichment-IMS-MDA approach to produce enough DNA concentration with suitable quality for nanopore sequencing, real-time PCR was used in the preliminary phase of the study prior to actual sequencing (data not shown).

Application of IMS as a complementary method to improve sensitivity and/or shorten the detection turnaround time has been widely used in food microbiology and safety. There are numerous publications reporting the combination of IMS with conventional and/or molecular methods (18). In this work, we combined IMS, MDA, short-term enrichment, and real-time analysis nanopore sequencing as a modified metagenomics approach in order to develop a rapid detection method suitable for flour and, accordingly, other food matrices upon evaluation. The other goal was to improve the chance of recovery for target strains with lower recoverability, using the developed multiaspect approach. However, it would be useful for future studies to assess the effect of multistrain contaminations on the efficiency of the approach. In this regard, new data becoming available, especially at molecular level such as WGS, may also be helpful for developing future strategies and methods (35).

By adding IMS, while the metagenome changed in favor of our target bacteria, enabling their detection at a low level such as 0.1 log CFU/g, the other major microbes, if present, could also be detected due to their natural presence in the food sample. In parallel, the developed approach could be used for subtyping of foodborne pathogens. Thus, our approach enables fast monitoring in which 1 h of sequencing allows initial analysis and, where potential pathogenic targets are detected, allows subsequent sequencing for more in-depth bioinformatics analysis.

The IMS beads used in this study were components of a commercial ATP kit designed for the immunomagnetic separation of E. coli and its detection by subsequent ATP bioluminescence assay in water samples. However, our preliminary tests indicated that they also captured Salmonella. This unique observation allowed us to develop this protocol to target both pathogens at the same time. The evaluation of IMS indicated there was variability in the capture efficiency among E. coli and Salmonella strains. This type of variation has previously been reported. In 2017, Kraft and colleagues reported that several factors such as serogroups, beads (different manufacturers), and food matrices could result in CE variations (36). Similar results were also reported by Calle et al., with more detail on IMS percentage of CE for STEC (37). The possibility of decreasing these changes by replacing conventional antibodies used in IMS beads with new natural aptamers provides a promising avenue for future studies in the field of biochemistry/immunology.

Selection of the library preparation kit PCR barcoding kit (SQK-PBK004) was based on its ability to work efficiently with a low starting DNA concentration, which is helpful in cases where food matrix composition would not allow a good yield after performing quasimetagenomics. In addition, the kit was able to separately barcode up to 12 samples to be mixed in a single run on the flow cell, resulting in reduced cost. The cost could be further reduced considering that with a maximum 15-h run time, the flow cell could be washed and used for 3 runs, covering a total number of 36 samples. This is due to the capability of the flow cell buffer and nanopore surface to run for 48 h either continuously or upon multiple washes. In this study, a maximum number of 4 samples were prepared and loaded into the flow cell for sequencing at one time.

A modification to the original kit protocol was that the optional fragmentation step was removed after our preliminary results of fragment analysis showed that most DNA sequences harvested after IMS-MDA were below 8 Kb, as previously discussed (data not shown). This also resulted in saving time and reducing cost per sample. Optimizing/replacing the MDA method in the future to be able to have a higher proportion of long sequences may increase speed as well as the precision of sequencing reads. Another key factor influencing sequencing output both in accuracy and yield, as well as flow cell health, is proper handling. Although preparing for this new technology and conducting the experiments are significantly easier than other available technologies for WGS, there is still need for precise handling with respect to multiple uses of a single flow cell, especially for library preparation, priming the flow cell, loading the library, and the wash steps.

The phylogenetic clustering distance between the metagenomics sample (target) and the WGS of the inoculated reference strain was accurate in single and/or dual inoculation (Fig. 5). A high degree of accurate match was observed for both scenarios with only 1 h of sequencing after the 12-h enrichment followed by IMS-MDA of the artificially inoculated flour samples. On the other hand, raw-read serotyping using SeqSero2 (21) was able to detect the exact serotype using data obtained after only 1 h of sequencing regardless of single or dual inoculation. In contrast, serotyping results for the reference E. coli O157:H7 strain used by the SerotypeFinder (38) could not reveal any details more than O157 level at 10 and 15 h sequencing. This difference probably relates to the different accuracy levels for these tools or their suitability for analyzing sequence ranges resulted from the MinIon device.

When Salmonella and E. coli were separately inoculated (0.1 log CFU/g), their calculated abundances in the final quasimetagenome using the conventional bioinformatics pipeline were approximately 15% and 46%, respectively (Table 2). When they were coinoculated, the abundances declined (Table 3), which could be expected due to the possible split of the available IMS beads to capture both Salmonella and E. coli. However, interestingly, the extent of the reduction was greater in Salmonella, which declined to about 2%, while E. coli declined to approximately 33%. Considering that the percentage of CE for IMS with single pure cultures was not significantly better for E. coli over Salmonella, the reason for this observation might be the higher affinity of IMS beads for E. coli over Salmonella where both are present. Alternatively, the fact that E. coli grew slightly faster during the early enrichment hours than Salmonella might result in higher E. coli numbers upon short-term enrichment of coinoculated samples. A good target for future studies would be using genus-specific IMS beads for each individual target organism to see if it can mitigate such biases for cocontaminated samples.

In conclusion, the developed approach has several advantages, including the ability to read long sequences, the possibility of real-time analysis, and the relatively high level of accuracy (20, 39). Furthermore, these techniques are constantly becoming less expensive as new devices become available, such as the most recent advancement termed flongle, a disposable replacement for the original MinIon flow cells, which can substantially reduce the sequencing cost for samples, including, but not limited to, bacteria and viruses (40). We believe that upon optimization and precise adaptation to food matrices and targets, this method can be a suitable platform for simultaneous identification and subtyping of foodborne pathogens with the ability to support further in-depth downstream analysis.

MATERIALS AND METHODS

Bacteria and growth conditions.

A total of 37 bacterial strains were used in this study. The seventeen E. coli strains used were O157:H7 (ATCC 43895, 121583, and LC-40), O121 (TW08980, I2016000899, and I2016012950), O111 (3077-99 and 3107-02), O26 (I2016003853), O26:H11 (3013-03 and 3235-99), O103:H2 (3002-98 and 3425-01), O145:NT (TW08087), O145:H28 (145-2), O45:H2 (TW14003), and ATCC 15597. All strains with the TW codes were obtained from the Michigan State University STEC Center (http://shigatox.net/stec/cgi-bin/index). Strains I2016000899, I2016003853, and I2016012950 were obtained from the Minnesota Department of Health.

Ten Salmonella enterica strains of different serotypes were used: Typhimurium (2009K-0300 and ATCC 14028), Agona (SLR141, F5567), Anatum (6802), Enteritidis, Braenderup (H9812), Newport (MH57137), Saintpaul (E2008001236), and Tennessee (07-0191). S. enterica serotype Typhimurium ATCC 14028 was purchased from the American Type Culture Collection (https://www.atcc.org/). All other Salmonella strains, E. coli strains without a designated source, and other bacterial genera used were provided by our culture collection at the Center for Food Safety, University of Georgia, Griffin Campus.

Ten strains belonging to other bacterial genera used in this study were Listeria monocytogenes (Jalisco, G1091), Cronobacter sakazakii (4921), Pseudomonas aeruginosa (NCTC 10332), Bacillus cereus (3812), Shigella sonnei (6129), Enterococcus faecium (NRRL-B2354), Staphylococcus aureus (NCTC8532), Streptococcus thermophilus (ATCC 19258), and Lactococcus lactis. These bacteria were selected because they may also commonly be present in food, including flour, and to confirm the exclusivity of the IMS step of the procedure against them.

All stock cultures were stored at –70°C in tryptic soy broth (TSB; Difco Laboratories, Sparks, MD, United States) supplemented with 20% (vol/vol) glycerol. All bacterial stock cultures were subjected to two consecutive transfers (24 h at 37°C) before direct use. In addition, working cultures of each strain were prepared using Luria-Bertani broth (LB; Difco Laboratories, Sparks, MD, United States), stored at 4°C, and refreshed on a monthly basis. These working stocks were used to prepare inoculation cultures.

Inoculation procedures.

Each strain was inoculated into 10 ml TSB (Difco) and incubated at 37°C for 24 h with shaking (200 rpm), resulting in stationary-phase cultures at an approximate concentration of 9 log CFU/ml. Appropriate 10-fold dilutions of these cultures were prepared using sterile 0.1% buffered peptone water (BPW; Neogen, Inc., East Lansing, MI, United States) and used for different subsequent experiments.

For flour inoculation, all-purpose wheat flour with an average protein content of 10% and approximate microflora background of 1 to 4 log CFU/g was used. In brief, 1 ml of a 2 log CFU/ml suspension of each individual strain in 0.1% BPW (Difco) was aseptically spot inoculated into 10-g flour portions in sterile stomacher bags (Nasco Whirl-Pak, Janesville, WI, United States) inside a biosafety cabinet and hand mixed for 5 min or until no clumps were observed. After addition of extra 90-g flour amounts to each inoculated 10-g portion, they were mixed for 3 min by hand and then two sets of stomaching for 3 min at 260 rpm were performed (Seward Stomacher, 400 Lab Systems, Norfolk, United Kingdom). A final 3-min hand mix was performed at the end and the resulting flour samples with approximate 1 log CFU/g contamination level were used for further experimentation.

Comparison of growth media for short-term enrichment.

Three E. coli strains (O157:H7 ATCC 43895, O26:H11 3337-99, and O121 I201612950) and two Salmonella strains (Typhimurium ATCC 1402 and Tennessee 07–0191) were individually inoculated into 10-ml test tubes containing five different broth media at a final concentration of approximately 1 log CFU/ml to identify the best culture medium for short-term enrichment of both target organisms. Lauryl sulfate broth (LSB), A1 broth (A1), E. coli broth (EC) (Sigma-Aldrich, St. Louis, MO, United States), TSB, and BPW (Difco) were used.

Inoculated tubes were incubated at 42°C for 4 or 8 h with 250 rpm shaking. After incubation, appropriate dilutions of the short-term incubated media were prepared and spread plated on tryptic soy agar (TSA; Difco). After a 24-h incubation of plates at 37°C, the colonies were counted, and the counts were transformed to log CFU/ml. These log-transformed numbers were used to compare short-term enrichment efficacies of the different media.

Immunomagnetic separation and capture efficiency calculations.

A commercial immunomagnetic separation kit consisting of primary antibody (IMS217:EcAb) and secondary goat anti-rabbit IgG IMS beads (IMS217:RbIgG) was used in this study (Virusys Corp., Taneytown, MD). The primary function of this kit was for detection of E. coli as a component of an IMS-ATP kit for the estimation of E. coli by ATP measurement. The kit was evaluated for inclusivity for target organisms, exclusivity for nontarget organisms, and capture efficiency (CE) with modifications to the manufacturer’s guidelines, using a total of 15, 10, and 10 individual strains of E. coli, Salmonella, and other bacterial genera, respectively.

In brief, initially 1.5-ml microcentrifuge tubes were coated with 700 μl of coating solution from the kit. The coating solution was transferred to tubes which were rotated and inverted until all inner surfaces were covered by the coating solution. The solution was then transferred to a second set of tubes one at a time until all were coated. Individual 1 ml 0.1% BPW aliquots containing 4 log CFU of each strain previously grown overnight in TSB, centrifuged, and diluted were transferred to each tube, 5 μl of primary antibody was added, and tubes were incubated at room temperature (RT; 23 ± 2°C) for 20 min on a rotator at 18 rpm (HulaMixer sample mixer; Thermo Fisher Scientific, Waltham, MA, USA).

Secondary antibody attached beads were then added (100 μl) and another incubation was performed, this time for 45 min at RT with 18 rpm rotation (Thermo Fisher Scientific). Samples were then put on a magnetic concentrator (DynaMag-2 Magnet; Thermo Fisher Scientific) and gently tilted (90°) for 2 min. After leaving the magnet (facing down) on a desk for 1 min, the supernatant was discarded and beads were harvested in 50 μl of phosphate-buffered saline (PBS) (Bio-Rad Laboratories, Hercules, CA) with 0.05% Tween 20 (Thermo Fisher Scientific). After transferring the beads to new 1.5-ml microcentrifuge tubes with coated inner surfaces as previously described, the original tubes were washed with an extra 25 μl of the harvesting solution and this was added to the first recovered portion. The new tubes were put back on the magnet and tilted (180°) for 2 min. Subsequently, two rounds of washing using PBS with 0.05% Tween 20 were performed and the final IMS recovered beads were dissolved in 100 μl of 1× PBS. Appropriate 10-fold serial dilutions were prepared using 0.1% peptone water and plated on TSA (Difco) for microbiological analysis. Recovered colonies were counted and used to calculate capture efficiency (CE) of the IMS kit and confirm its inclusivity and exclusivity for target and nontarget microorganisms, respectively.

The CE, defined as the percentage fraction of the total bacteria retained on the surface of the IMS beads (41), was calculated with the equation based on the cells bound to the beads as follows: CE (%) = C_b/C₀ × 100, where C₀ is the total number of cells present in the sample (CFU/ml) and C_b is the number of cells bound to the IMS beads at the end of the separation process.

IMS-MDA in flour.

The IMS procedure was slightly modified to fit the requirements for analyzing flour-enriched samples and also to suit the following MDA step. Briefly, flour samples (25 g each) inoculated with approximately 1 log CFU/g bacterial strain of choice as described in the inoculation procedures were enriched using 225 ml TSB (Difco) at 42°C with shaking (250 rpm) for 12 h in 500-ml flasks. Upon enrichment, the enrichment was left in the biosafety cabinet for 5 min to settle and 50 ml of the upper liquid phase of enriched flour mixture was aseptically collected in two 50-ml conical tubes, each containing a 25 ml portion.

A centrifugation of 100 × g for 10 min (4°C) was performed to further reduce the solid particles present. Subsequently, supernatant was carefully transferred to new 50-ml conical tubes and subjected to another 10 min centrifuge (4°C) at 3,214 × g. After centrifugation, the supernatants were discarded and pellets from duplicate tubes were recovered in 1 ml PBS. This was used for the IMS procedure as previously discussed. The only two differences were that three rounds of final PBS washes were performed and the final bead-bacterium component was harvested and resuspended in 9 μl of sample buffer from the MDA kit.

The MDA procedure was performed using an Illustra GenomePhi V2 DNA amplification kit (GE Healthcare Life Sciences, Piscataway, NJ, USA). Briefly, the bead-bacterium complexes were resuspended in 9 μl of sample buffer and incubated at 95°C for 3 min. The tubes were then immediately cooled on ice and 9 μl of reaction buffer and 1 μl enzyme mix were added and mixed by gentle pipetting. This allowed the annealing of random hexamer primers. Reaction tubes were then incubated at 30°C for 2 h for amplification followed by 65°C treatment for 10 min to inactivate the ϕ29 DNA polymerase enzyme. These final MDA products of approximately 20 μl were immediately used for the following procedures or stored at –20°C until used.

Quasimetagenomic nanopore sequencing of flour samples.

Shotgun sequencing was performed using a portable MinIon device (Oxford Nanopore Technologies, Oxford, UK). The sequencing library was prepared from the quasimetagenomics samples using a PCR barcoding kit (number SQK-PBK004, Oxford Nanopore Technologies). Application of this kit was possible using as little as 100 ng of genomic DNA. Prior to library preparation, all DNA sample purities were measured using a Nanodrop ND-1000 UV/VIS spectrophotometer (Thermo Fisher Scientific). The DNA input mass was also measured for each sample at the beginning of the library preparation as well as during the procedure as needed using a Qubit 3 fluorometer (Thermo Fisher Scientific). After necessary measurements, library preparation was performed following the manufacturer’s guidelines.

Briefly, IMS-MDA-resulting DNA fragments were repaired using NEBNext end repair/dA-tailing module (E7546; New England Biolabs, Ipswich, MA, USA) and cleaned using Mag-Bind Total Pure NGS (Omega Bio-Tek, Norcross, GA, USA) and two 70% ethanol washes. The end-prepped DNA was mixed with barcode adaptors and blunt/TA ligase master mix (New England Biolabs) and, after a 10-min incubation, was cleaned again as discussed above. A 14-cycle PCR was then performed using unique barcode primers for each sample DNA and 2× LongAmp Taq master mix (New England Biolabs) to amplify and barcode each set of DNA library. The final DNA batch recovered was eluted in 10 μl of 10 mM Tris-HCl pH 8.0 with 50 mM NaCl. Rapid adaptor ligation was then performed using 1 μl of the provided adaptor and the resulting 11 μl of DNA library was either immediately subjected to the next steps or stored at –20°C for priming and loading the SpotOn on the flow cell attached to a MinIon device for sequencing.

The SpotOn flow cell (R9.4; Oxford Nanopore Technologies) was primed according to the manufacturer’s guidelines, and samples were mixed with sequencing buffer (SQB) and loading beads (LB) and loaded to the flow cell through the SpotOn sample port. In cases of multiple samples with different unique barcodes being loaded to the flow cell in the same run, they were adjusted to the same DNA mass and mixed together in the loading solution while retaining the final load volume in the library preparation guideline. Subsequently the sequencing run was started and continued using the MinKNOW software with fast basecalling enabled for a maximum of 15 h and basecalling was performed using Guppy software (v3.1.5). The basecalled reads were collected at 1, 3, 6, 10, and 15 h of sequencing and subjected to the nanopore analysis platform as well as conventional bioinformatics analysis.

Bioinformatics and data analysis.

For the nanopore technology-based analysis, basecalled reads of each individual time point from sequencing runs were uploaded to the Metrichor analysis platform (Oxford Nanopore Technologies) using EPI2ME (v2.59.1896509) desktop agent. Using this software, sequences were subjected to the “what’s in my pot” (WIMP) pipeline for detection, metagenomics analysis, and phylogenetic tree building.

In parallel, conventional bioinformatics analysis was also performed. Adaptor trimming and barcode demultiplexing were done using Porechop v0.2.4 (https://github.com/rrwick/Porechop). Quality of the sequencing reads was assessed using NanoPlot v1.18.1 (42). The trimmed and demultiplexed reads were classified using Kraken2 (43) and reads classified as E. coli and/or Salmonella were extracted for further analysis. Miniasm 0.3-r179 was used for de novo assembly of the extracted reads, and Racon v1.4.1 was used to correct raw contigs of the draft assembly. Finally, serotyping was performed using SeqSero2 (21) and SerotypeFinder (38) for Salmonella and E. coli, respectively.

For SNP detection and phylogenetic analysis, core genome SNPs were identified using CFSAN SNP Pipeline v0.8.0 using default parameters (44). Genomes of S. enterica Typhimurium strain SL1344 (ATCC 14028; NCBI assembly accession no. NC_016810.1) and E. coli O157:H7 Sakai (ATCC 43895; NCBI assembly accession no. NC_002695.2) were used as reference genomes for reads mapping for Salmonella and E. coli, respectively. Read mapping for nanopore sequencing reads was performed using BWA-MEM (45) with the ‘-x ont2d’ option. SNP matrices and alignments with concatenated SNPs were produced by custom Python scripts from each metagenomics or WGS sample processed by the CFSAN SNP pipeline. Finally, maximum likelihood (ML) phylogenetic trees were constructed using PhyML (46).

Statistical analysis.

Means of at least 3 measurements of microbial populations (log CFU/ml) from short-enrichment in different media were subjected to analysis of variance (ANOVA). Fisher’s least significant difference (LSD) test was applied for the comparisons. All analyses were performed using IBM SPSS Statistics version 25 (SPSS Institute, Chicago, IL). The significance of difference was defined at P values of ≤0.05.