Targeted amplicon sequencing for enhanced detection of spiked Shiga toxin-producing Escherichia coli in ready-to-eat romaine lettuce: a proof-of-concept study

ABSTRACT

The early and accurate detection of low-level pathogenic and indicator organisms in fresh produce is critical for preventing widespread foodborne outbreaks. Contamination of leafy greens with foodborne pathogens, such as Shiga toxin-producing Escherichia coli (STEC), is a significant public health issue, making rapid and sensitive detection methods critical for mitigating outbreaks. Although next-generation sequencing (NGS) is a powerful tool for pathogen identification, challenges remain in detecting low contamination levels in food products. Here, we demonstrate the use of a custom targeted amplicon sequencing (TAS) primer panel targeting species with food safety concerns, including known human foodborne pathogens, opportunistic pathogens, and indicator organisms related to food spoilage. Using a quasi-metagenomics approach, this proof-of-concept study demonstrates that, compared to whole-metagenomic sequencing (WMS), TAS is a rapid and sensitive NGS-based method for detecting low levels of pathogens. Ready-to-eat romaine lettuce was spiked with STEC and incubated in enrichment medium. DNA was isolated at 0.5, 5, and 6 h, and libraries were prepared for both WMS and TAS. The results indicated that TAS was more sensitive than WMS not only at detecting the pathogen at the species level but also at identifying key virulence markers stx1 and stx2. Overall, our targeted sequencing approach provides a rapid and sensitive molecular method to detect and identify foodborne pathogenic bacteria, demonstrating its potential for application in food safety.

IMPORTANCE

Detecting low-level pathogenic and indicator organisms is critical to prevent foodborne outbreaks. Conventional methods lack speed and sensitivity. While next-generation sequencing methods, such as whole-metagenomic sequencing (WMS), offer a broad microbial landscape view, detecting pathogens at low concentrations within complex food matrices remains challenging. To address this, a targeted amplicon sequencing (TAS) panel was designed to identify species of food safety concern and key indicator organisms. This study demonstrates that TAS is more sensitive than WMS. The application of this TAS assay provides an important bridge between qPCR and WMS by detecting and characterizing pathogens that might be present in low numbers and otherwise missed in an enrichment. TAS allows multiplexing and overcomes the critical limitation of sensitivity in complex samples, providing a robust tool for food safety surveillance. Our findings demonstrate the potential use of targeted next-generation sequencing (NGS)-based methods to mitigate the risk of foodborne illnesses.

INTRODUCTION

High-throughput or massive parallel sequencing technologies have opened opportunities to advance genomic research (1–3). Among the various applications of next-generation sequencing (NGS), advancements in the finer characterization of bacterial genomes and accurate taxonomic identification of microorganisms in complex environments, particularly in food and food-related processes, are useful in outbreak tracing (4, 5). NGS methods can be used in two independent ways. The first and most common shotgun sequencing method is whole-genome sequencing (WGS) of a single cultured isolate. Data from WGS using core genome multilocus typing and whole-genome MLST schemes are used for high-resolution pathogen typing, epidemiology, susceptibility prediction, and virulence factor determination (6, 7). The second method is whole-metagenome sequencing (WMS), which involves sequencing all microorganisms in a sample without the need to selectively purify a single isolate. The high discriminatory power of NGS methods using shotgun sequencing has been shown to be valuable and increasingly popular for microbial typing and pathogen detection in food and for source attribution related to outbreak investigations or regulatory applications (8, 9).

Although NGS methods provide resolution and sensitivity for high-throughput surveillance, there are still significant challenges in detecting low levels of pathogens or subpopulations of unculturable pathogens present in samples at low levels with WGS or WMS sequencing approaches (10–12). These limitations on specificity and sensitivity are due to the depth of sequencing being compromised, depending on the number and complexity of the sample being sequenced. One way to overcome this problem is to use targeted amplicon sequencing (TAS), in which specific target genes are chosen for polymerase chain reaction (PCR) amplification, and then the amplicons are subjected to NGS. This allows for the sequencing of selected target genes specific for the pathogens of interest with increased sensitivity, better specificity, and ease of downstream analysis. Importantly, while broad NGS methods do not cover the qPCR diagnostic niche in food safety testing, TAS provides an important bridge technology, encompassing the low-level diagnostic capacity of qPCR with the high-resolution data of genome sequencing. TAS is now routinely used in the detection of biothreat agents, as well as in cancer diagnostics (13–15). Metagenomics sequences all DNA in a sample, enabling broad microbial detection, but complicating the identification of low-abundance organisms in food and environmental samples. In contrast, TAS strategically targets and amplifies predefined loci of interest, thereby avoiding the need for deep sequencing and making it a cost-effective method. Another advantage is that, unlike WMS, TAS is independent of sequencing depth and platform, offering flexibility across technologies, such as Illumina and PacBio. Furthermore, targeted sequencing approaches offer potential advantages in throughput and cost-effectiveness over WMS, as their targeted nature is amenable to higher sample multiplexing (16).

Here, we used an MLST-like approach with core gene sequences to design a custom primer panel to amplify ten core gene regions (Table S1) from each of the targeted organisms of food safety concern in a single multiplex reaction for amplicon sequencing. The objectives of this study were twofold: (i) to demonstrate the utility of TAS for identifying microbial content at the species level and intraspecies diversity down to the strain level, and (ii) to test, as a proof of principle, the detection of Shiga toxin-producing Escherichia coli (STEC) spiked with different levels of colony-forming units (CFU) on ready-to-eat (RTE) romaine lettuce. Specifically, this study represents method development, where we compare the results of TAS with WMS using the same samples to determine the sensitivity of pathogen detection and demonstrate that TAS is more sensitive than WMS.

MATERIALS AND METHODS

Target primer panel design

Twenty bacterial pathogens affecting human health and nine other bacterial species associated with animals were selected for inclusion in the primer panel (Table S1). All the sequences were obtained from the NCBI database. Sixty-eight genomes of different genotypes from the following foodborne viruses were included: human adenovirus, Aichivirus A, human astrovirus, enterovirus, hepatitis A virus, hepatitis E virus, norovirus, rotavirus, and sapovirus. The virus genotypes were selected based on their established role in foodborne illnesses and their relevance to food safety. To account for temporal and geographic variations in circulating strains, multiple genotypes of key viruses were included, such as norovirus (GI, GII, GIV), hepatitis A and E, rotavirus (A, B, C), and human adenovirus (A–D). To represent the fungal kingdom, 35 species were included in this study. The selection was performed to ensure broad relevance to food safety and quality control. The targets included mycotoxin-producing molds, fungi that are indicative of food spoilage, and opportunistic pathogens that pose health risks. We also chose to include yeasts that are commonly used in food fermentation. Finally, Cyclospora cayetanensis, a foodborne coccidian parasite, was added to the set. This primer panel was designed to target bacterial, viral, fungal, and parasite pathogens that collectively account for the majority of foodborne disease burden, enabling comprehensive screening of the most clinically and epidemiologically relevant organisms encountered in food safety-related investigations. For each pathogen, core gene regions were chosen either from an existing MLST panel for that species or homologous genes from an MLST panel of a related species. Ten genes per pathogen (except for E. coli, we chose 12) were selected to account for any variations in the target that could lead to mispriming or no priming and increase detection accuracy. In addition to the core genes for E. coli, Salmonella enterica, and Listeria monocytogenes, virulence genes (for example, E. coli stx1 and stx2) were targeted for each of these three species. The desired amplicon size was set at approximately 600 base pairs (bp). The Primer3 software v2.7.0 (17) was used to design primers for all species, setting the optimal primer length to 20 and optimal Tm to 60°C. A consensus sequence from several strains of the species was used as input to maximize the universality of the primers for that species. The pathogen and genes that were used as targets to design the primers are shown in Table S1. The primer sequences (Table S1) were provided by Swift Biosciences (now Integrated DNA Technologies, Coralville, IA, USA) to design a custom kit using the multiplex pool Accel-Amplicon Panel approach. The primer panel CP-FD6054-48 (641 pathogen-specific primers) was used in this study. The initial evaluation was performed in silico using sequences from reference strains of foodborne pathogens. Three to eight genetically diverse representatives, with the exception of E. coli (n = 309), L. monocytogenes (n = 19), and Salmonella enterica subsp. enterica (n = 30), were used to screen the primers. Only primers that demonstrated 100% sequence identity to all reference genomes within the intended target species were selected for use. Primer specificity testing was limited to target species only, as amplicons from non-target organisms can be distinguished by sequence analysis of PCR products.

DNA standards

The Microbial Community DNA Standard (MCDS) containing 10 microbial species (Zymo Research Corporation, Irvine, CA, USA; Product D6305) was used as a positive control to evaluate the performance of the primer panel. This standard served as a reference material that can be used to compare new methods against established techniques and was chosen to evaluate potential biases in NGS methods. Of the 10 strains included in the MCDS, eight were bacterial (three gram-negative and five gram-positive) and two were yeast. The eight bacterial species were Bacillus subtilis, Enterococcus faecalis, Escherichia coli, Limosilactobacillus fermentum, Listeria monocytogenes, Pseudomonas aeruginosa, Salmonella enterica, and Staphylococcus aureus, while the two yeast species were Saccharomyces cerevisiae and Cryptococcus neoformans. WMS and TAS libraries were prepared as described below under library preparation.

Escherichia coli samples

Three STEC strains carrying different Shiga toxin allelic variants were selected to test the performance of the primer panel (Table 1). The first was the O157:H7 strain TW14359 or MI06-63 (GenBank CP001368), a clinical isolate associated with the 2006 bagged spinach (18). The other two strains were non-O157 STEC strains (O26:H11 and O121:H19). The selection of these strains was based on their serotypes, each of which has previously been associated with fresh produce outbreaks and has different Shiga toxin gene repertoires (Table 1).

TABLE 1.

STEC strains used in this study^a

Strain name	Serotype	stx	Other name	Source	Accession no.
EC1611	O157:H7	2a, 2c	MI06-63; TW14359	USA (MI), 2006 spinach outbreak	CP001368
EC1763	O26:H11	1a	TW14253	USA (MI), 1998 bovine feces	NJTV01
EC1992	O121:H19	2a	TW14945	USA (MI), 2002 human feces	NJXV01

^a Summary of STEC strains, including strain name, serotype, Shiga toxin (stx) subtype(s), alternate names, source, and GenBank accession numbers.

Metagenomic sample processing following spiking by STEC

Bagged chopped RTE romaine lettuce was purchased from a local grocery store and stored overnight at 4°C. The three STEC strains were grown overnight in buffered peptone water with pyruvate (mBPWp). Overnight cultures were diluted in a sterile saline solution. The CFU counts were verified by serial dilution plating on LB agar after incubation at 37°C for 24 h. Actual inoculum levels were confirmed to be within ±0.5 log CFU of target levels. For each strain, three filter whirlpak bags were used as technical replicates. RTE romaine lettuce (100 g) was added to 225 mL of mBPWp medium, and 100 µL of the diluted STEC cells at different CFU levels (1,000, 100, or 10 CFU) were then added to the romaine lettuce suspended in liquid medium, as described by Leonard et al. (19). Unspiked romaine lettuce was used as a control. Briefly, the bags were incubated with shaking at 185 rpm at 37°C for 30 min, and at this time point (0.5 h), 25 mL aliquots were withdrawn for processing as described below. The bags were then further incubated at 37°C for 4.5 h without shaking, and another 25 mL aliquot was withdrawn, and acriflavine (10 mg/L), cefsulodin (10 mg/L), and vancomycin (8 mg/L) were added. The bags were then incubated at 42°C for another hour, and another 25 mL aliquot was withdrawn. Bacteria from the 25 mL sample aliquots withdrawn at 0.5, 5, and 6 h were harvested by centrifugation at 5,000 rpm for 10 min, the supernatant was decanted, and the pellets were stored at −20°C until genomic DNA extraction.

DNA extraction, library preparation, and next-generation sequencing

Genomic DNA was isolated from 25 mL metagenomic samples using the DNeasy PowerSoil Pro kit (Qiagen, Carlsbad, CA, USA). DNA was quantified using a Qubit dsDNA BR assay kit and Qubit 3.0 fluorometer (Thermo Fisher Scientific, Waltham, MA). The WMS libraries were prepared using the NexteraXT DNA Sample Preparation Kit (Illumina, San Diego, CA, USA). One nanogram of genomic DNA was normalized to 0.2 ng/μL in 10 mM Tris-HCl (pH 8.5) and subjected to simultaneous fragmentation and adapter tagging using Amplicon Tagment Mix. The tagmentation reaction was performed at 55°C for 5 min in a thermal cycler, immediately followed by the addition of a neutralization buffer to stop the reaction. Next, libraries were amplified using the Nextera PCR Master Mix with dual-indexed primers (Nextera XT Index Kit v2) under the following cycling conditions: initial extension at 72°C for 3 min, denaturation at 95°C for 30 s, followed by 12 cycles of 95°C for 10 s, 55°C for 30 s, and 72°C for 30 s, with a final extension at 72°C for 5 min. PCR products were purified using AMPure XP magnetic beads (Beckman Coulter, Brea, CA, USA) at a 1.8:1 bead-to-sample ratio with two washes using freshly prepared 80% ethanol. Libraries were eluted in 30 μL of resuspension buffer. Library quality and size distribution were assessed using an Agilent 2200 TapeStation (Agilent Technologies, Santa Clara, CA, USA) with a High Sensitivity DNA Kit. Library concentrations were determined using a Qubit dsDNA HS Assay Kit. TAS libraries were prepared using a custom kit (CP-FD6054-48; Integrated DNA Technologies, Coralville, IA, USA) in accordance with the manufacturer’s protocol. Multiplex PCR amplification was performed using a two-stage thermocycling protocol optimized for multiple target specificities. The thermocycler was programmed with an initial denaturation step at 98°C for 30 s, followed by four touchdown cycles consisting of denaturation at 98°C for 10 s, combined annealing and extension at 65°C for 5 min, and a final extension at 67°C for 1 min. Standard amplification was then performed for 21 cycles of denaturation at 98°C for 10 s, annealing at 64°C for 1 min, and extension at 65°C for 1 min. The reaction was complete with an indefinite hold at 4°C. The touchdown strategy allows for higher stringency in the initial cycles to enhance primer specificity and minimize non-specific amplification, which is critical for successful multiplex reactions, while the subsequent temperature reduction and extended annealing times ensure efficient amplification of all target sequences. Size selection and clean-up were performed using AmpureXP beads at a ratio of 1:1.2, followed by two washes using freshly prepared 80% ethanol. Indexing sequencing adapters and the enzyme master mix were then added to each library (on the beads) and incubated at 37°C for 20 min. A second clean-up step was performed using AmpureXP beads at a ratio of 1:0.85 and eluted with 20 μL of post-PCR TE buffer. Quantification of the adapter-ligated libraries was performed by qPCR using the KAPA Library Quantification Kit (Roche Sequencing Solutions, Pleasanton, CA, USA). The library concentrations were measured using a Qubit 3.0 fluorometer using the Qubit dsDNA HS Assay kit (Thermo Fisher Scientific, Waltham, MA). To check the amplicon library size distribution, 2 µL of the library was run on a 2200 TapeStation using the High Sensitivity DNA kit. For both WMS and TAS, libraries were pooled at equimolar ratios and sequenced on the Illumina MiSeq using a 500-cycle MiSeq Reagent Kit v2 to generate 2 × 250 bp paired-end reads.

Data analysis

Fast QC was used to check the quality of the raw reads obtained from each sequence run (20). Reads with quality scores <Q20 were trimmed, and Illumina adaptor sequences were removed using the default parameters in Trimmomatic (21). For both WMS and TAS, bacterial classification was performed using MetaPhlAn (22) and Kraken2 (23), respectively, using GalaxyTrakr (24). The number of Kraken-classified reads was normalized per 1,000,000 reads sequenced for each sample to account for differences in sequencing depths across samples. The total read count ranged from 454,452 to 8,967,228 for WMS and 36,630 to 3,953,726 for TAS. ReCentrifuge was used to visualize taxonomic abundances (25). The E. coli MLST genes included in the TAS primer panel are aspC, clpX, fadD, icdA, lysP, mdh, uidA, adk, fumC, gyrB, purA, and recA. The reads for each sequencing run were matched by BLAST (26) to a reference, and a consensus sequence was formed for each gene after alignment with ClustalW version 2.1 (slow/accurate option, default parameters) (27). Sequences for the 12 E. coli MLST genes were concatenated and phylogenetic analysis was performed using MEGA v. 10.0.5 (28). The tree was rooted in E. coli K-12 strain MG1655 (GenBank U00096.3). Evolutionary history was inferred using the neighbor-joining method (29), and evolutionary distances were computed using the p-distance method (30) with pairwise deletion of gaps/missing data, and other parameters were set to default.

RESULTS

Evaluation of the TAS primer panel with ZymoBIOMICS Microbial Community DNA Standard

The ZymoBIOMICS MCDS is composed of genomic DNA from eight bacterial species and two yeasts. The bacterial species were Bacillus subtilis, Enterococcus faecalis, E. coli, Limosilactobacillus fermentum, L. monocytogenes, Pseudomonas aeruginosa, S. enterica, and Staphylococcus aureus, and the two yeasts used were Saccharomyces cerevisiae and Cryptococcus neoformans. This standard has been routinely used in various studies as a control for optimizing PCR methods and sequencing workflows, validating taxonomic classifications, and ensuring data quality in metagenomic research (31–34). We used 1 ng of MCDS to prepare libraries using the WMS and TAS protocols, followed by sequencing. The WMS and TAS results are shown in Fig. 1. The total number of reads was 1,534,650 and 2,309,624, respectively.

Fig 1.

The stacked bar chart shows the relative abundance (%) of microbial taxa identified from the ZymoBIOMICS Microbial Community DNA Standards analyzed using two methods: TAS (top) and WMS (bottom). Each color corresponds to a specific species: Escherichia coli, Salmonella enterica, Listeria monocytogenes, Enterococcus faecalis, Staphylococcus aureus, Limosilactobacillus fermentum, Bacillus subtilis, Cryptococcus neoformans, Saccharomyces cerevisiae, and Pseudomonas aeruginosa. (⁺) Species included in the TAS panel.

The number of reads mapped to each species is shown in Table 2. The average read lengths for the WMS and TAS data were 215 and 240 bp, respectively.

TABLE 2.

Comparison of microbial species abundances from ZymoBIOMICS Microbial Community DNA Standards using WMS and TAS methods^a

Species	GC content	WMS	TAS
Escherichia coli*	47	18514	106779
Salmonella enterica*	52	22540	97442
Listeria monocytogenes*	38	57354	5552
Enterococcus faecalis*	38	57390	10783
Staphylococcus aureus*	33	18255	1156
Limosilactobacillus fermentum	52	43682	1961
Bacillus subtilis	44	1241	1
Cryptococcus neoformans*	48	10267	61505
Saccharomyces cerevisiae*	38	8706	45
Pseudomonas aeruginosa	66	3111	1653

^a Abundance of microbial species detected using two sequencing methods: WMS and TAS. Species names are listed in the first column, followed by the GC content (%) of each species and the corresponding normalized reads for WMS and TAS. Species marked with asterisks (*) are included in the TAS primer panel.

The TAS primer panel targeted seven of the ten strains in the MCDS, and the primer panel was able to amplify and allow for the accurate identification of the seven strains. Among the foodborne pathogens, the gram-negative strains E. coli (47% GC) and S. enterica (52% GC) had high TAS reads compared to the gram-positive strains L. monocytogenes (38% GC), E. faecalis (38% GC), and S. aureus (33% GC), which have higher WMS read counts (Fig. 1). Of the two yeasts, S. cerevisiae (38% GC) had more WMS reads, and C. neoformans (48% GC) had more TAS reads, probably because of the higher GC content, which probably allowed for better amplification. The WMS data accurately identified the three strains that were not included in the primer panel, as expected. The three bacteria, B. subtilis (44% GC) and L. fermentum (52% GC), are gram-positive, while P. aeruginosa (66% GC), a gram-negative bacterium, all had higher WMS reads than TAS, as expected. Even though these strains were not included in the primer panel design, they amplified genes most likely due to mispriming or the conservation of primers targeting a related species and non-specific amplification.

Detection of E. coli using whole-metagenomic sequencing

RTE romaine lettuce was spiked separately with three STEC serotypes (O157:H7, O26:H11, and O121:H19) at 10, 100, and 1,000 CFU/100 g lettuce. Unspiked romaine lettuce from the same lot was used as a control to determine whether the microbial community in the romaine lettuce contained E. coli. The bags were processed at 0.5, 5, and 6 h post-inoculation, except that the 6 h sample was not collected for the unspiked samples. The number and percentage of reads that mapped to E. coli and Shiga toxin genes were assessed for all three strains and unspiked samples (Table 3).

TABLE 3.

Comparative analysis of WMS and TAS methods for the detection of STEC strains (O157:H7, O26:H11, and O121:H19) under varying CFU and time points^a

CFU	Time (h)	Total read count		Average read length (bp)		No. of E. coli reads (Kraken)		stx1/stx2
		WMS	TAS	WMS	TAS	WMS	TAS	WMS	TAS
Unspiked
0	0.5	618752	ND	251	ND	77	ND	−	ND
0	5	2022178	ND	228	ND	−	ND	−	ND
O157:H7
1,000	0.5	2397124	36630	220.1	242	692	3740	−	+
100	0.5	2054432	42670	222.8	235	96	3164	−	+
10	0.5	1872720	112402	238.3	211	177	196	−	+
1,000	5	619870	185408	250.6	241	179112	842143	+	+
100	5	667068	736492	250.6	221	205331	46158	+	+
10	5	563628	861678	250.7	220	69658	1274	−	+
1,000	6	2816922	2171138	234	251	317214	344640	+	+
100	6	1939090	1775322	232.5	251	260364	200954	+	+
10	6	1774236	1278908	235.6	250	42594	241744	+	+
O26:H11
1,000	0.5	1461300	2030560	154	80	124	79	−	+
100	0.5	1739294	3546366	91	82	30	44	−	+
10	0.5	1263392	3953726	191	84	13	126	−	+
1,000	5	457972	1849876	237	251	188485	109077	+	+
100	5	454452	1666900	237	250	138166	120753	+	+
10	5	624290	997256	227	246	7364	50481	−	+
1,000	6	899548	1951988	243	251	163409	128340	+	+
100	6	766086	1034306	227	250	152518	134995	+	+
10	6	485546	1300076	236	250	23922	89705	+	+
O121:H19
1,000	0.5	804206	1146408	210	251	31	517	−	+
100	0.5	705818	1380518	217	251	23	11	−	+
10	0.5	1565704	1273856	209	251	77	3628	−	+
1,000	5	2078380	855630	227	251	11066	34351	+	+
100	5	8967228	1338472	181	251	826	20983	+	+
10	5	3025906	1996282	234	251	100	1547	−	+
1,000	6	2930572	733832	238	246	23320	101328	+	+
100	6	3963632	1260550	224	251	703	23770	+	+
10	6	1183614	1005614	246	251	234	175	−	+

^a The table summarizes the total read counts, average read length (bp), number of E. coli reads per million, and detection of stx1/stx2 genes across different E. coli strains analyzed using WMS sequencing and TAS. Data were collected at multiple time points (0.5, 5, and 6 h) from 25 mL or (for O26:H11 and O121:H19) 225 mL of the cultures obtained from romaine lettuce spiked with 1,000, 100, and 10 STEC CFU. Unspiked samples were used as controls (ND, not performed). The presence (+) or absence (−) of stx1 and stx2 genes is noted for each strain; for O157:H7 and O121:H19, results shown are for the stx2 gene, and for O26:H11, results shown are for the stx1 gene.

The WMS data for the unspiked samples at 0.5 h showed 77 reads, which identified Escherichia at the genus level, but there were no reads for E. coli species or virulence genes. The 5 h unspiked sample data had no Escherichia reads. The metagenomic community analysis for the unspiked samples showed that the most prevalent genus at 0.5 h was Pseudomonas, and at 5 h was Priestia (Fig. S1I and J). For the spiked samples, WMS generated an average of 1,749,068 reads, with a minimum of 454,452 reads and a maximum of 8,967,228. Taxonomic classification using Kraken2 identified E. coli with an average of 89 reads that mapped to the species level at 10 CFU from the 0.5 h samples for all three strains. Our findings are similar to those of prior research on the detection of STEC from produce spiked with low CFUs (19). The average number of total reads from the WMS data for the O157:H7 strain was 1,633,899, for the O26:H11 strain was 905,764, and for the O121:H19 strain was 2,802,784. At 0.5 h samples, stx1 and stx2 were not detected in the three strains. For the O157:H7 strain at 5 h, stx2 was detected in the 100 and 1,000 CFU-spiked samples, but at 6 h, stx2 was detected in all three spike levels. For the O26:H11 strain at 5 and 6 h, stx1 was detected in all spiked samples except 10 CFU at 5 h. For the O121:H19 strain, stx2 was not detected at 10 CFU at both the 5 and 6 h time points, but could be detected at 100 and 1,000 CFU.

Detection of E. coli using target amplicon sequencing

The same DNA harvested at the time points mentioned above was used to set up a multiplex PCR as part of the custom targeted amplicon sequencing library preparation protocol. As expected, TAS analysis of the unspiked control samples did not detect any of the targeted pathogens, confirming the absence of background contamination from lettuce. For the spiked samples, TAS generated an average of 1,352,699 reads, with a minimum of 36,630 and a maximum of 3,953,726. Taxonomic classification using Kraken2 identified E. coli at the species level for all spiked samples and time points, with an average of 1,317 reads that mapped at the species level for 10 CFU from the 0.5 h samples for all three strains (Table 3). The average number of reads for O157:H7 was 800,072, the O26:H11 was 2,036,784, and for the O121:H19 strain was 1,221,240. For the O157:H7 and O121:H19 strains, stx2 and O26:H11 strain stx1 were detected in all spiked samples and time points tested. The taxonomic composition for both WMS and TAS data and abundance distribution of the bacterial communities in the spiked and unspiked samples from the 0.5 h time point and lowest CFU concentration used for the three strains and the non-spike control, as well as the 5 h time point for the non-spike control, were visualized using Krona charts (Fig. S1). Genus-level classifications using the WMS and TAS data are included in Table S2A and B, respectively.

Phylogenetic analyses post whole metagenomic and target amplicon sequencing

Phylogenetic analysis was performed to confirm that the spiked E. coli strains in the romaine lettuce samples were accurately identified and clustered correctly with their respective reference strains (Fig. 2 and 3).

Fig 2.

The tree was constructed using the maximum likelihood method in MEGA v. 10.0.5, with evolutionary distances computed using the p-distance method. Distances are expressed as the number of base substitutions per site (scale bar = 0.01). The clustering of strains within each serotype reflects their evolutionary relationships and responses to the experimental conditions. The branch lengths represent genetic divergence among the strains. Reference strains (EC1611 ref O157:H7, EC1992 ref O121:H19, EC1763 ref O26:H11) were included for evolutionary comparison. Strain names (EC1611, EC1992, EC1763) are presented along with their corresponding serotypes (O157:H7, O121:H19, O26:H11) and experimental conditions, including time points (0.5, 5, 6 h) and CFU counts (10, 100, 1,000 CFU).

Fig 3.

The tree was constructed using the neighbor-joining method in MEGA v. 10.0.5, and evolutionary distances were calculated using the p-distance method. Distances are measured as base substitutions per site (scale bar = 0.01). The clustering of strains within each serotype highlights their evolutionary relationships and responses to the experimental conditions. Branch lengths indicate the genetic divergence between the strains. Reference strains (EC1611 ref O157:H7, EC1992 ref O121:H19, and EC1763 ref O26:H11) were included for evolutionary comparison. Strain names (EC1611, EC1992, and EC1763) are shown with their respective serotypes (O157:H7, O121:H19, O26:H11), time points (0.5, 5, 6 h), and CFU counts (10, 100, 1,000 CFU).

This was essential for evaluating the specificity and precision of the TAS approach, particularly in the context of complex metagenomic samples containing diverse microbial populations. By examining the clustering patterns, we verified that the TAS method could reliably distinguish and classify spiked strains according to their genetic relatedness in the presence of potential interference from non-target microbial DNA. Even in samples where data are missing due to low coverage, they are properly clustered but with a long branch. At 0.5 h for the WMS data, there were insufficient reads to be included in the phylogenetic analysis. For the TAS data, only the O26:H11 at 0.5 h spike with 1,000 CFU did not have enough coverage to be included in the tree. The only exception where we did not see a tight cluster in the WMS tree was at O121:H19 in the 5 h and 10 CFU spiked samples (Fig. 2). Using both WMS and TAS data, the strains were clustered based on their serotypes, as shown in Fig. 2 and 3, respectively.

DISCUSSION

This study describes the use of TAS in the detection of foodborne pathogens and compares it with WMS. We specifically compared the performance of WMS and TAS in detecting STEC in RTE romaine lettuce at different CFU and time points. The specificity of our primer set was evaluated using a combination of both in silico bioinformatics analysis and experimental validation with representative target organisms in the ZymoBIOMICS MCDS. In silico evaluation using reference strain sequences was the first step in primer validation. While the analysis included genetically diverse representatives for most target organisms, the coverage was expanded for certain species of particular concern, E. coli, L. monocytogenes, and Salmonella enterica subsp. enterica because these are common foodborne pathogens. The stringent criterion requiring 100% sequence identity across all reference genomes within each target species helped to ensure theoretical specificity. The in silico analysis was restricted to available reference sequences in the databases, which may not capture the full genetic diversity present in environmental and clinical isolates that could be encountered in real-world applications. Although 100% sequence identity was achieved within the target species, the computational approach cannot account for potential secondary structure effects or primer-template interactions that might occur under actual multiplex PCR conditions. The ZymoBIOMICS MCDS was chosen as a control for this study, as it was specifically designed to include organisms spanning a broad GC content range to evaluate bias across different sequencing workflows. The most likely explanation for the higher TAS reads for species with higher GC content is compared to species with lower GC content due to better annealing and thereby amplification. As expected, the three species B. subtilis, S. cerevisiae, and P. aeruginosa showed fewer TAS reads as they were not included in the primer panel design. While the MCDS provides a standardized approach for specificity testing, it represents a defined subset of microbial diversity rather than a comprehensive coverage of all possible target organisms. However, the systematic composition of this standard, combined with our bioinformatics analysis across broader sequence databases, provided strong evidence for primer specificity within the tested parameters. Our primary objective was to compare the performance of WMS and TAS for pathogen identification rather than achieving absolute quantification against theoretical values (CFUs). We demonstrated that library preparation methods, which involve PCR amplification, can have significant effects on GC bias in metagenomic approaches. Although the sequencing depths achieved by both methods covered the microbial diversity in mock communities (Fig. 1 and Table 2), significant differences were observed in their sensitivity, specificity, and sequencing variability between the two sequencing methods. WMS provides broader taxonomic coverage as it offers a holistic, unbiased approach that captures the entire microbial community within a sample and accurately identifies species not targeted by TAS. This makes WMS particularly useful for quasimetagenomic characterization (35), enabling the detection of both STEC and other microbial taxa present in the sample. Such insights are invaluable for understanding the microbial ecology, potential cross-contamination, and community dynamics in food matrices. The TAS exhibited higher sensitivity for low-abundance and GC-rich species (Fig. 1 and Table 2), which is attributed to the targeted design. The biases observed in the TAS and WMS were most likely due to the GC content. TAS demonstrated a preference for GC-rich genomes, as shown in Table 2, illustrated by higher read counts for E. coli (47% GC) and S. enterica (52% GC). In contrast, WMS captured species with a low GC content, such as S. aureus (33% GC) and L. monocytogenes (38% GC). These findings align with those of previous studies highlighting the influence of GC content on sequencing efficiency, indicating the need for careful consideration when designing custom primer panels for pathogen detection (36). Future studies with more stringent PCR cycling conditions could help reduce non-specific amplification, even though the highly multiplexed nature of our panel allows for effective allele differentiation once sequencing data are aligned to reference sequences.

In spiked romaine lettuce samples, TAS outperformed WMS in the detection of low-abundance STEC serotypes (Table 3). TAS identified stx1 and stx2 genes across all time points and spiked samples tested, as shown in Table 3, highlighting its superior sensitivity, while WMS could not detect stx1 and stx2 genes at 0.5 h with all three CFU. At 5 h, while providing broader coverage of the microbial community, the detection of species in low abundance using WMS was inconsistent. At 5 h, WMS detected the stx2 gene in the 100 and 1,000 CFU spiked samples for both O157:H7 and O121:H19 strains, but not in the 10 CFU spiked samples. In contrast, stx1 in the O26:H11 strain was detected at 5 h at 10, 100, and 1,000 CFU using WMS. This indicates that pathogenic markers, such as stx1 and stx2, are less likely to be detected by WMS at earlier time points, like 0.5 or 5 h, when spiked with 10 CFU. This is likely because WMS provides a broad snapshot of the entire microbial community, and the abundance of the pathogens in the sample affects their detection, which can make detecting low-level pathogens challenging due to background noise from other organisms. In addition, if the pathogen grows poorly, its representation in the data remains low. TAS can amplify target regions, but enrichment may still be necessary to achieve adequate detection, depending on the level of pathogens present. The intended application of TAS should be understood as a complementary tool to enhance the sensitivity of metagenomic identification of organisms present at low levels, rather than as a standalone detection method for microbiological hazards. It is well known that matrices of complex food samples can affect DNA isolation and amplification, thereby introducing a bias. Similarly, as observed here, the low GC content of pathogens can affect their detection, and future studies will focus on optimizing the conditions for their detection. The differences in sequencing depth between WMS and TAS, shown in Table 3, highlight why TAS is a better strategy for detecting low-level pathogens that may be present in food and environmental samples. WMS relies on cultural target amplification to enhance the detection of low-abundance targets, whereas TAS employs molecular amplification using specific primers to amplify genomic regions of interest. However, the enhanced sensitivity of TAS in this preliminary study suggests its utility in food safety applications in outbreak investigations, foodborne illness surveillance, and routine food safety assessments (37).

Phylogenetic analysis confirmed the robustness of TAS in accurately clustering spiked E. coli strains with their respective reference strains across different time points and CFU tested, even in the presence of complex microbial backgrounds. TAS demonstrated specificity in producing accurate phylogenetic relationships (Fig. 3). WMS offers broader taxonomic profiling and showed accurate clustering for all spiked samples at the 5 and 6 h time points, except for the 100 CFU spiked sample for the O121:H19 strain at 5 h (Fig. 2). As there were not enough reads for STEC at the 0.5 h time points for all strains at all spiked CFUs, they were excluded from the WMS tree. These findings reinforce the utility of TAS in pathogen surveillance, particularly when they are present in low amounts in the presence of the matrix microbiome. The complementary strengths of WMS and TAS suggest that hybrid approaches could provide both broad taxonomic coverage and precise strain-level identification, thereby enhancing microbial surveillance in food safety. By combining the sensitivity of TAS for targeted gene detection with the broad coverage of WMS, a more comprehensive and efficient workflow for pathogen detection can be achieved.

Conclusion

The detection of pathogens, such as STEC, in fresh produce remains a public health and food safety concern. This study introduces the concept of using TAS for risk assessment and compares it with WMS for detecting spiked STEC in RTE romaine lettuce, highlighting their respective strengths, limitations, and applicability to food safety.

TAS demonstrated higher sensitivity and specificity for stx1 and stx2 detection, particularly at low CFU spiked levels. Our findings also highlight that TAS can be a powerful complement to qPCR, where qPCR can confirm presence, TAS provides deeper resolution through in-depth amplicon characterization and inherent subtyping abilities. Meanwhile, WMS, with its broader microbial resolution, complements TAS by offering valuable insights into microbial diversity and community dynamics. A combined approach that leverages the strengths of both methods is promising for enhancing pathogen detection and food safety monitoring, ultimately safeguarding public health. While these preliminary results demonstrate the potential of TAS for enhanced detection, validation studies following established food microbiological tests will be addressed in future studies.

DATA AVAILABILITY

The sequencing data presented in this study have been deposited in the NCBI Sequence Read Archive repository (accession number PRJNA1034160). Non-sequencing data relevant to the study are included in this article or provided as supplementary material.

SUPPLEMENTAL MATERIAL

The following material is available online at https://doi.org/10.1128/spectrum.01022-26.

Table S2A. spectrum.01022-26-s0001.xlsx.

Genus-level taxonomic classification using WMS data.

Figure S1. spectrum.01022-26-s0002.pdf.

Metagenomic community analysis.

Table S1. spectrum.01022-26-s0003.xlsx.

Primer panel specifications.

Table S2B. spectrum.01022-26-s0004.xlsx.

Genus-level taxonomic classification using TAS data.

ASM does not own the copyrights to Supplemental Material that may be linked to, or accessed through, an article. The authors have granted ASM a non-exclusive, world-wide license to publish the Supplemental Material files. Please contact the corresponding author directly for reuse.