Accurate and sensitive detection of Salmonella in foods by engineered bacteriophages

Salmonella is a major causative agent of foodborne illness and rapid identification of this pathogen is essential to prevent disease. Currently most assays require high bacterial burdens or prolonged enrichment to achieve acceptable performance. A reduction in testing time without loss of sensitivity is critical to allow food processors to safely decrease product holding time. To meet this need, a method was developed to detect Salmonella using luciferase reporter bacteriophages. Bacteriophages were engineered to express NanoLuc, a novel optimized luciferase originating from the deep-sea shrimp Oplophorus gracilirostris. NanoLuc-expressing bacteriophages had a limit of detection of 10–100 CFU per mL in culture without enrichment. Luciferase reporters demonstrated a broad host range covering all Salmonella species with one reporter detecting 99.3% of 269 inclusivity strains. Cross-reactivity was limited and only observed with other members of the Enterobacteriaceae family. In food matrix studies, a cocktail of engineered bacteriophages accurately detected 1 CFU in either 25 g of ground turkey with a 7 h enrichment or 100 g of powdered infant formula with a 16 h enrichment. Use of the NanoLuc reporter assay described herein resulted in a considerable reduction in enrichment time without a loss of sensitivity.


Results
Characterization of Salmonella phages SEA1 and TSP1. Preliminary studies led to the selection of two lytic Salmonella bacteriophages, SEA1 and TSP1, for assay development. Using one-step growth curves, the replication cycle time was determined to be 35-40 min and 60-70 min for SEA1 and TSP1, respectively. The burst size of SEA1 was found to be approximately 30 pfu per cell, while TSP1 produced a larger burst size of approximately 100 pfu per cell. Thus, the replication cycle time and burst size for SEA1 and TSP1 are similar to those reported for other Salmonella phages 28 .
To facilitate plasmid design for homologous recombination, DNA from SEA1 and TSP1 was extracted and sequenced. While genome curation and annotation are beyond the scope of this study, preliminary analysis revealed a genome size of approximately 162 kbp for SEA1 and 157 kbp for TSP1. BLAST analysis of SEA1 revealed considerable homology to Salmonella phage vB_SenM-S16 (NC_020416). This phage was previously described as a 160 kbp Myovirus possessing a remarkably broad host range strictly within the Salmonella genus 29 . BLAST analysis of TSP1 revealed considerable homology to Salmonella phage SFP10 (NC_016073). This phage was previously described as a 158 kbp Myovirus specific for E. coli O157:H7 and Salmonella isolates 30 . The morphology of SEA1 and TSP1 was visualized by transmission electron microscopy ( Fig. 1a,b). Both SEA1 and TSP1 have contractile non-flexible tails. Based upon these micrographs and supported by sequence homology, SEA1 and TSP1 are also predicted members of the Myoviridae family 31 . Construction of NanoLuc-expressing recombinant bacteriophages. Generation of NanoLucexpressing recombinant SEA1 and TSP1 was performed using homologous recombination (Fig. 2a,b). Homologous flanks were designed to direct insertion downstream of the predicted major capsid protein, a strategy previously used to generate luciferase reporter phage for Listeria 32 . This insertion site was not expected to disrupt any predicted genes. Recombination donor plasmids were generated containing these regions of homology flanking a codon-optimized NanoLuc gene under a T4 late promoter. Salmonella transformants containing these donor plasmids were infected with SEA1 and TSP1. Recombinant NanoLuc-expressing bacteriophages (SEA1.NL and TSP1.NL) were isolated from this reaction, passaged to purity, and confirmed by DNA sequencing.
Limit of detection of engineered bacteriophage reporters. Successful detection of pathogens must be capable of finding small numbers of contaminating cells. Although SEA1.NL and TSP1.NL were expected to produce considerable amounts of luciferase following infection, the exact number of required host cells to produce a detectable signal over background was unknown. To determine the limit of detection of these engineered phages, infections were performed with inoculums ranging from 1 to 10,000 CFU. After a 2 h infection, the amount of luciferase produced in each condition was determined with a luminometer following the addition of substrate (Table 1). Background from medium and reporter alone (negative control) was minimal, averaging www.nature.com/scientificreports/ 117 and 51 RLU for SEA1.NL and TSP1.NL, respectively. The standard deviation (SD) of background was also low, allowing signal from luciferase production to be easily recognized. Signal above background was detected, on average, from a single cell for both SEA1.NL and TSP1.NL. Average RLU from either reporter phage infection increased proportionately with the number of Salmonella cells, reaching values over 100 times background with only 100 CFU. Signal from TSP1.NL was consistently higher than SEA1.NL at equivalent cell counts, possibly reflective of the higher burst size of TSP1. RLU variability, as measured by coefficient of variation (CV), was expected at low cell counts, where the probability of no cells being present in a replicate is increased. Individual replicates that received no cells by chance will yield only a background signal, which will contrast starkly with replicates containing live cells. This effect would be most pronounced at one, two, and five CFU, where the highest variation is observed. These results confirm the functionality of SEA1.NL and TSP1.NL, revealing a clear correlation between average RLU and presence of Salmonella. Remarkably, a detectable signal above background could be demonstrated with a single log phase CFU after only a 2 h infection. This single cell signal was greater than twice background, a standard cutoff used by others to determine positive detection from luciferase reporter phages 25 .

Figure 2.
Generation of NanoLuc expressing bacteriophages by homologous recombination. (a) SEA1 recombination donor assembled in pUC57. ORF1 of SEA1 indicates the predicted major capsid protein while ORF2 has homology to head vertex proteins; (b) the TSP1 recombination donor assembled in pUC57. ORF1 of TSP1 indicates the predicted major capsid protein while ORF2 is a hypothetical protein of unknown function. Graphics were generated using SnapGene (GSL Biotech LLC, Chicago, IL, USA).  Table 2). While no false negatives were observed with these six strains, the lowest signal observed with SEA1.NL was a mere 2958 RLU compared to 6,195,000 RLU with TSP1. NL. This particular serovar has been the source of several food-related outbreaks, including contaminated infant milk products [33][34][35] . Overall, the median signal from positively detected Salmonella strains was 103,257,000 RLU for SEA1.NL and 206,576,768 RLU for TSP1.NL. These values are substantially above the 750 RLU threshold used and demonstrate the robust signal generation of these reporters.
Specificity of SEA1NanoLuc and TSP1NanoLuc detection. Methods facilitating detection of Salmonella contamination must possess sufficient specificity to limit cross-reactivity with the natural microbiome present in many food matrices. To determine the specificity of these NanoLuc reporter phages, an exclusivity panel of non-Salmonella strains was collected. Representatives of 14 species of Gram-positive and 26 species of Gram-negative bacteria were infected for 2 h with SEA1.NL or TSP1.NL and assessed for luciferase production. Unsurprisingly, no false positives were detected with Gram-positive bacteria, likely due to considerable differences in surface structures when compared to the Salmonella host (  Table 3). Of note, five of the eight false positives encountered with SEA1.NL were below 103,000 RLU, which is 1000-fold lower than the median RLU signal observed with Salmonella true positives. Only two false positive strains, one Escherichia coli and one Citrobacter sedlakii generated signal indistinguishable from most of the Salmonella strains. These results highlight the limited potential for false positives from related Gram-negative bacteria.
Detection of Salmonella contamination in food matrices. Based upon these results, SEA1.NL and TSP1.NL were combined into a single phage cocktail. TSP1.NL was expected to supplement the signal intensity of many Salmonella strains without compromising the specificity achieved with SEA1.NL alone. Further, an enrichment period was added to the workflow to promote recovery, growth, and detection of single cell Salmonella contamination. This method of phage detection is referred to as the PhageDx method for Salmonella.
Although accurate detection of Salmonella cells was achieved in pure culture, food products are complex test matrices that may present additional challenges and complications. To determine if the PhageDx method could mediate detection in such environments, two distinct relevant food matrices were selected. As a representative of a ground meat product, raw ground turkey was chosen and has previously been associated with a nationwide outbreak in the United States 36 . For the second matrix, powdered infant formula was selected to model detection in dried food products, which itself has been linked to multiple outbreaks in infants 37 . Both matrices were pre-screened prior to use to evaluate the presence of pre-existing contamination. Salmonella was not detected endogenously from portions of either matrix used in this study. Although appearing free of Salmonella, each gram of homogenized ground turkey did yield approximately 40 CFU on non-selective media. The powdered infant formula used in this study was found to contain little endogenous flora (0 CFU per gram).
Portions (25 g) of raw ground turkey were either uninoculated, inoculated with a low-level of Salmonella (~ 1 CFU), or inoculated with a high-level of Salmonella (~ 10 CFU). Following a 7 h enrichment, samples were infected with a reporter phage cocktail for 2 h and checked for luciferase production. No false positives were detected among uninoculated test portions ( Table 4). As anticipated, partial positives for the "low" inoculum and all positives for the "high" inoculum were obtained with two Salmonella enterica subsp. enterica serovars. Importantly, the PhageDx assay agreed with a culture-based confirmation method for all 30 samples. Dynabead isolation and CHROMagar Salmonella were utilized for this comparison as both methods have demonstrated excellent performance in a variety of food matrices [38][39][40][41] . These results indicate that the PhageDx method is capable of accurately detecting low levels of Salmonella contamination in raw ground turkey.
Portions (100 g) of PIF were evaluated in a similar manner to raw ground turkey, except a 16 h enrichment was used. No false positives were detected among uninoculated test portions (Table 5). Both serovars of Salmonella enterica subsp. enterica produced the anticipated partial positives for the "low" inoculum and all positives for the "high" inoculum. As seen previously with ground turkey, the PhageDx method agreed with a culture-based confirmation method for all 30 PIF samples. These results indicate that the PhageDx method is also capable of accurately detecting low levels of Salmonella contamination in PIF.

Discussion
Rapid, accurate, and sensitive detection of foodborne pathogens is essential to maintain a safe and effective food supply. Despite achieving desired sensitivity and accuracy, many commercially available assays require extensive enrichment and operate under a timeframe of days, not hours. This study details the development and performance of the PhageDx rapid detection method for Salmonella, a prevalent and important contaminant of food products. To our knowledge, this method represents the first development and use of NanoLuc reporter bacteriophages in Salmonella.
Initial screening of 53 bacteriophages led to two promising candidates, SEA1 and TSP1, being chosen for assay development. These phages were predicted based on morphology to be members of the Myoviridae family (Fig. 1). NanoLuc-expressing recombinants of these phages (SEA1.NL and TSP1.NL) were engineered using www.nature.com/scientificreports/ homologous recombination (Fig. 2). These reporters were evaluated using a 2 h infection, no enrichment, and varying amounts of log phase Salmonella. A clear correlation between luminescent signal over background and Salmonella was observed with SEA1.NL and TSP1.NL (Table 1). Critically, a single CFU of Salmonella produced, on average, a detectable signal above background. This result underscores the advantages of tying reporter production to lytic phage, which rapidly adsorb, infect, replicate, and lyse host cells. The limit of detection of this method is also aided by the robust and specific luminescent signal produced by NanoLuc and its substrate, furimazine 23 . The reporter phages engineered in this study thus demonstrated remarkable sensitivity, a promising trait that can be leveraged to shorten assay time.
The inclusivity of NanoLuc reporter phages was evaluated with diluted stationary phase cultures, no enrichment, and a 2 h infection. Using a detection threshold of 750 RLU, SEA1.NL was able to positively detect 99.3% of 269 Salmonella strains ( Table 2). The extensive range of this reporter suggests that SEA1 utilizes a receptor  www.nature.com/scientificreports/ common to almost all Salmonella. Of note, the Salmonella bacteriophage S16, which has considerable homology to SEA1, has been found to bind OmpC, an outer member protein that is well-conserved among Salmonella 29,42 . Regardless of the receptor used, the median signal from detected strains was over 100 million RLU, well beyond the threshold used (Supplementary Table 2 Overall, the broad inclusivity of SEA1.NL is well-suited for detection across the Salmonella genus, while TSP1. NL may provide a supportive benefit with signal intensity in particular serovars, such as Agona. Only two strains of Salmonella could not be detected by either reporter, one representative each of Salmonella enterica subsp. enterica serovars Enteritidis and Kentucky. Importantly, our results do not indicate an inability to detect these serovars entirely as 27 other Enteritidis and two other Kentucky strains were positive (Supplementary Table 1). The exact mechanism behind these two false negatives is unknown, although it is plausible these two strains lack a common Salmonella receptor. Alternatively, a litany of phage resistance mechanisms have been described such as those mediating restriction-modification, abortive infection, and extracellular matrix production 44 . Future studies may refine or expand upon the phage cocktail, highlighting the flexible and modular nature of this approach.
The exclusivity of each reporter phage was evaluated in the same fashion as inclusivity. Of 90 non-Salmonella strains examined, false positives were detected with eight strains for SEA1.NL (Table 3). These eight strains Table 4. Detection of Salmonella enterica in inoculated portions of ground turkey (25 g). Salmonella enterica subsp. enterica serovars were diluted from stationary phase cultures and inoculated into pre-screened portions of ground turkey. Strain 27869 (ATCC) and OCT084 (USDA) were used as serovar Newport and Muenster, respectively. Equilibrated samples were enriched for 7 h, and infected with a cocktail of SEA1.NL and TSP1. NL for 2 h before being assessed for luciferase production. Inoculums consisted of either no CFU "None", approximately 1 CFU per 25 g "Low" (1.1 CFU for Newport, 1.2 CFU for Muenster), or approximately 10 CFU per 25 g "High" (11.8 CFU for Newport, 8.9 CFU for Muenster). A detection threshold of 750 RLU was used to determine positive samples. Samples were confirmed by a culture-based method involving Dynabead isolation and plating on CHROMagar Salmonella.  Table 3). Although the mechanism behind this variation in signal intensity is not known, it is feasible that non-Salmonella may lack certain co-receptors conducive to phage infection. TSP1.NL, on the other hand, cross-reacted with only one strain from the entire exclusivity panel, the same Citrobacter sedlakii observed with SEA1.NL. This supports the previous notion regarding the narrower host range of this phage. TSP1.NL has considerable homology to the Salmonella bacteriophage SFP10, which was previously reported to recognize both Salmonella and E. coli O157:H7 30 . Despite this homology, TSP1.NL did not produce a positive result for any O157:H7 strain tested in our study, which included some of the same strains previously tested with SFP10. These data support the notion that TSP1.NL does not share this property with SFP10.
The performance of SEA1.NL and TSP1.NL in food matrices was evaluated as part of the PhageDx Salmonella assay. Critically, this method involves minimal processing, no sample cleanup, and can be adapted as needed depending on the properties of the matrix. When compared to a culture-based method, the PhageDx assay correctly identified 100% of artificially contaminated raw ground turkey samples with a 7 h enrichment ( Table 4). Detection of a single Salmonella CFU in 25 g of product, a required sensitivity benchmark, was achieved. Importantly, no false positives were observed, suggesting that the natural flora of this matrix (40 CFU per g) did not Table 5. Detection of Salmonella enterica in inoculated portions of powdered infant formula (100 g). Salmonella enterica subsp. enterica serovars were diluted from stationary phase cultures, dried down, and inoculated into pre-screened portions of PIF. Strain SL476 (FDA) and 52317.1 (USDA) were used as serovar Heidelberg and Reading, respectively. Equilibrated samples were enriched for 16 h, diluted ten-fold, and infected with a cocktail of SEA1.NL and TSP1.NL for 2 h before being assessed for luciferase production. Inoculum consisted of either no CFU "None", 1 CFU per 100 g "Low", or 5 CFU per 100 g "High". A detection threshold of 750 RLU was used to determine positive samples. Samples were confirmed by a culture-based method involving plating on CHROMagar Salmonella. www.nature.com/scientificreports/ contain problematic Enterobacteriaceae strains or burdens. Similar performance was achieved in PIF, where the PhageDx assay also correctly identified 100% of artificially contaminated PIF samples with a 16 h enrichment ( Table 5). The lack of false positives in this matrix was unsurprising, given the lack of natural flora observed in this study (0 CFU per g). The ability of phage reporters to function in various food matrices has been previously observed and is further supported by this study 48,49 . In summary, the PhageDx method agreed 100% with the longer culture-based method in both food matrices and was capable of accurate and sensitive detection of Salmonella contamination. Previous attempts to utilize bacteriophage for Salmonella detection in food matrices have been met with mixed success. Utilizing a combination of bacteriophage and immunomagnetic separation, previous studies have successfully detected one to three CFU per 25 g of food 50 . This method required 20 h to complete but, as a result of the limited host range of the phage (SJ2), could only reliably detect Salmonella enterica subsp. enterica serovar Enteritidis. Bacteriophage have also been combined with real-time PCR to achieve rapid detection of low burdens in spiked chicken samples 51 . This approach was able to detect eight Salmonella CFU per 25 g portion within 10 h. Once again, however, the limited host range of the phage restricted reliable detection to one serovar, Enteritidis. Similar to the approach used in this study, several recombinant luciferase reporter phages have previously been assessed in food matrices. A luciferase-expressing recombinant of the temperate phage SPC32H demonstrated the ability to detect as few as 22 CFU per g (550 CFU per 25 g) in only 2 h from food matrices 49 . Although promising with sensitivity and speed, this phage was specific for the serovar Typhimurium, preventing its use in Salmonella species detection. P22, another temperate bacteriophage of Salmonella, was also assessed as a luciferase recombinant 20 . This reporter phage demonstrated excellent performance in feed and environmental samples with a 16 h assay time. As with all other described approaches, detection of Salmonella by P22 recombinants was limited by the narrow host range of this phage and was serovar dependent. Thus, while previous bacteriophage-based methods have achieved excellent sensitivity and time to results, the PhageDx method uniquely affords broad detection of Salmonella independent of serovar in food matrices.
Ultimately, new methods support the continued goal of preventing contaminated products from reaching consumers. Shorter enrichment times are highly desirable, allowing issues to be detected early with limited product holding time. The PhageDx Salmonella assay leverages the sensitivity of two engineered NanoLuc-expressing bacteriophages to achieve rapid detection of single cell Salmonella contamination. This study demonstrates the noteworthy capabilities of bacteriophage reporter assays to facilitate accurate pathogen detection in a variety of matrices.

Methods
Bacterial strains. All bacterial strains used in this study were obtained from the American Type Culture Wild-type phage isolation. The Salmonella bacteriophage SEA1 was obtained from Dr. Francisco Diez-Gonzalez's laboratory at the University of Minnesota. SEA1 is a broad-spectrum Salmonella phage of the Myoviridae family previously isolated from waste effluents 52 . Salmonella phage TSP1 was isolated from sewage samples obtained from the Metropolitan Waste Water Treatment Plant in St. Paul, Minnesota. Samples were clarified by centrifuging in a swinging bucket rotor at 4700 × g for 10 min and filtering the supernatant through a 0.45 µm filter (Nalgene, Rochester, NY). A mixture containing 2 mL of this filtrate, 1 mL of 3 × TSB, and 150 µL of an overnight culture of Salmonella enterica subsp. enterica ser. Typhimurium (ATCC 19585) was incubated at 37 °C with shaking at 225 rpm for 18 h. After this incubation, the sample was once again centrifuged at 4700 × g for 10 min and the supernatant filtered through a 0.45 µm filter. The presence of phage was initially evaluated by spot testing on ATCC 19585 and confirmed by plating for single plaques using the classical overlay method 53 . Individual plaques were picked, resuspended in TSB, and subsequentially plated again for single plaques. Single plaque selection was repeated five times to obtain pure, single phage cultures.

Generation of high titer stocks of wild-type and recombinant bacteriophages.
High titer wildtype and recombinant phage stocks were made using broth lysates. To this end, 100 mL of logarithmic (log) phase Salmonella cells at an OD 600 of 0.2 were infected at a multiplicity of infection (MOI) of 0.05. SEA1 and SEA1.NL used strain ATCC 14028 while TSP1 and TSP1.NL used strain ATCC 19585. After allowing 5 min for adsorption, infected cells were diluted into 400 mL of prewarmed TSB and incubated at 37 °C with shaking at 250 rpm until lysis was apparent. The phage lysate was clarified by centrifuging at 14,900 × g using a type 19 rotor in an Optima XE-90 ultracentrifuge (Beckman Coulter, Brea, CA) for 10 min at 4 °C. Phages were concentrated by centrifuging again at 14,900 × g for 2 h at 4 °C. The phage pellet was resuspended in TMS buffer (50 mM Tris-HCl pH 7.8, 10 mM MgCl 2 , and 300 mM NaCl) then treated with RNase and DNase I. Phages were further purified on a sucrose density gradient (10-30%) in TMS. The phage band was removed and sedimented at 107,200 × g using a SW41 Ti rotor in an Optima XE-90 ultracentrifuge for 30 min at 4 °C. Finally, the phage pellet was resuspended in SM buffer ( www.nature.com/scientificreports/ by heating 5 × 10 9 plaque forming units (pfu) at 90 °C for 5 min. DNA was purified from protein by three phenol/ chloroform extractions. After removal of phenol/chloroform, 0.1 volume of 3 M sodium acetate and two volumes of ethanol were added to aqueous phase. DNA was precipitated at − 80 °C, pelleted, then washed two times with 70% ethanol. The DNA pellet was dried and then resuspended in deionized water and used for sequencing. The burst size and replication cycle time of phages were determined using a traditional one-step growth curve on their respective host strains 54 .
Transmission electron microscopy of SEA1 and TSP1 was performed using 400 mesh grids coated with a thin carbon film. Glow discharged grids were floated on cesium chloride density gradient purified phage samples, then stained with 2% uranyl acetate. Images were captured on a Tecnai G2 Spirit BioTWIN at 30 kV.

Construction of homologous recombinant plasmids.
Plasmids were designed to generate Nano-Luc-expressing recombinant bacteriophages through homologous recombination (HR). Constructs containing a codon-optimized NanoLuc (Promega Corp., Madison, WI) gene under a T4 late promoter and flanked by regions of homology to the respective phage genome were designed. Codon optimization for Salmonella was performed using a codon optimization tool (Integrated DNA Technologies, Coralville, IA). Homologous flanks were designed to direct insertion downstream of the predicated major capsid protein. The SEA1 cassette consisted of 500 bp upstream of the desired insertion site, followed by the σ70 promoter − 10 consensus sequence, a Shine-Dalgarno ribosomal entry site consensus sequence, a NanoLuc codon optimized for Salmonella, then 500 bp of downstream phage SEA1 sequence. The TSP1 cassette was similarly designed with the following exception. A 300 bp downstream homologous sequence was used, followed by a stop codon and transcriptional terminator. Differences in design were due to initial difficulties in the construction of the TSP1 HR plasmid which were overcome by the addition of a stop codon and transcriptional terminator. Constructed cassettes targeting TSP1 and SEA1 were assembled and inserted into the multiple cloning site of pUC57 by GeneWiz (South Plainfield, NJ). These recombinant constructs were expected to facilitate insertion of NanoLuc into a phage late gene region without disrupting any predicted genes.
Integration of the NanoLuc into the phage genome by homologous recombination. Electrocompetent Salmonella were generated as described previously 55,56 . Salmonella enterica subsp. enterica ser. Typhimurium strains ATCC 14028 and ATCC 19585 were selected as recombinant hosts for SEA1 and TSP1, respectively. Electrocompetent bacteria were combined with 100 ng of homologous recombination plasmid DNA and subjected to a 1.8 kV single pulse using a MicroPulser electroporation apparatus (Bio-Rad Laboratories, Hercules, CA). Transformants were isolated following overnight growth on Luria-Bertani (LB) agar containing 100 µg/mL carbenicillin.
Resistant colonies were selected, grown in LB containing 100 µg/mL ampicillin, and infected with SEA1 or TSP1 respectively at various MOIs (0.1-10). Samples were incubated at 37 °C with 220 rpm shaking for 3 h. Following infection, cultures were centrifuged for 2 min at 6800 × g. The supernatant was collected, filtered through a 0.45 µm filter, and washed with TMS on a 100 kDa pore protein concentrator PES column (Pierce Biotechnology, Rockford, IL). This was plated as previously described for single plaque isolation. To identify recombinants, candidate plaques were picked, mixed with a diluted overnight culture, and monitored for luciferase expression. Once a NanoLuc-producing isolate had been found for each bacteriophage, it was sequentially passaged at least four times from a single plaque to ensure the stability and purity of the recombinant. After isolation, high titer stocks of SEA1NanoLuc (SEA1.NL) and TSP1NanoLuc (TSP1.NL) were prepared as described previously. Homologous recombination was verified by genome sequencing, as described previously, and confirmed that the desired recombinants had been generated.
Limit of detection of phages SEA1NanoLuc and TSP1NanoLuc. Log phase Salmonella cells (OD 600 of 0.1-0.5) were diluted in TSB to obtain desired CFU/mL. Salmonella enterica subsp. enterica ser. Typhimurium (ATCC 19585) was used for TSP1.NL while Salmonella enterica subsp. enterica ser. Choleraesuis (ATCC 7001) was used for SEA1.NL. 100 µL was then transferred to a 96-well plate and infected with 10 µL of phage reagent (1.2 × 10 7 pfu/mL in TSB) for 2 h at 37 °C. Luciferase detection solution was prepared as a master mix for each experiment, consisting of 50 µL of NanoGlo buffer, 15 µL Renilla lysis buffer, and 1 µL of NanoGlo substrate (Promega Corp., Madison, WI) per sample well. Following infection, 65 µL of this luciferase detection solution was added to each well and the samples read in a GloMax Navigator luminometer (Promega Corp., Madison, WI) using a 3 min wait time and 1 s integration. Six to ten replicates of each dilution were measured and results averaged. Signal output was relative light units (RLU). Wells containing no Salmonella were used to determine background from media, phage, and detection reagents alone.
Inclusivity and exclusivity of phages SEA1NanoLuc and TSP1NanoLuc. Inclusivity and exclusivity assays were carried out to determine the coverage and specificity of recombinant phages. Overnight stationary phase cultures were diluted in TSB to an OD 600 of 0.2 (approximately 1.6 × 10 8 CFU/mL). Aliquots of 100 µL were transferred to 96-well plates and infected with 10 µL of phage reagent. After 2 h of incubation at 37 °C, 65 µL of luciferase detection solution was added to each well. Luminescence was measured as previously described. Positive results were evaluated using a cutoff of 750 RLU.

Inoculation of raw ground turkey and powdered infant formula (PIF).
Raw ground turkey (85% lean/15% fat Jennie-O, Wilmar, MN) was pre-screened using the PhageDx and culture-based confirmation method described below. Samples of ground turkey were also homogenized and plated on a non-selective agar (tryptone soy agar) to evaluate pre-existing contamination levels. Once the absence of endogenous Salmonella www.nature.com/scientificreports/ had been confirmed, the matrices were inoculated with the indicated Salmonella enterica serovars. Strain 27869 (ATCC) and OCT084 (USDA) were used, serovars Newport and Muenster, respectively. A liquid inoculum culture was prepared by transferring a single Salmonella colony from a TSB plate into TSB broth and incubating the culture for 18-24 h at 37 °C. Following incubation, the culture was diluted to the target level in buffered peptone water (BPW) (Oxoid, Hampshire, United Kingdom). Inoculums were plated to determine CFU level. Target CFUs levels were an average of 2-20 CFU/mL for low level and 20-100 CFU/mL for high level inoculums. Aliquots of 100 µL of designated inoculum were used to inoculate turkey samples. Based on averaged replicate plating, CFU inoculum per 25 g sample were 1.1 and 1.2 for low level and 11.8 and 8.9 CFU for high level of serovar Newport and Muenster, respectively. Prior to analysis samples were held for 48-72 h post-inoculation at 2-8 °C to allow for equilibration. Low level inoculated samples were expected to yield fractional positive results (25-75% positive), and a high level expected to yield all positive results. Negative control samples were uninoculated. All samples were assessed in a blinded manner as testers were unaware of the inoculum given to each test portion. For PIF (Up and Up milk-based infant formula with iron, Target, Minneapolis, MN), Salmonella enterica was also grown in TSB for 18-24 h at 37 °C. Strain SL476 (FDA) and 52317.1 (USDA) were used, serovars Heidelberg and Reading, respectively. The culture was diluted in BPW, reconstituted in PIF and placed into a speed vacuum for 4-8 h until the sample was completely dried. Contaminated PIF was allowed to equilibrate for 2-4 weeks at room temperature (20-25 °C). After equilibration, an aliquot of dried inoculum was resuspended in 1 mL of BPW and plated to determine CFU level. Using this determined CFU, dried inoculum was then diluted into additional PIF matrix to achieve a low level (1 CFU/100 g) or high level (5 CFU/100 g). Low level inoculated samples were expected to yield fractional positive results (25-75% positive), and a high level expected to yield all positive results. Negative control samples were uninoculated. Samples of PIF were screened prior to inoculation to evaluate pre-existing contamination as described above for ground turkey. As with ground turkey, PIF was also assessed in a blinded manner.
PhageDx Salmonella detection assay for raw ground turkey and PIF. Raw ground turkey (25 g) was placed in a filter bag (Nasco WhirlPak, Fort Atkinson, WI), homogenized, and enriched with pre-warmed (41 °C) BPW in a 1:3 ratio (25 g ground turkey: 75 mL BPW) for 7 h at 41 °C. Powdered infant formula (100 g) was placed in a sample bag with pre-warmed (37 °C) BPW in a 1:3 ratio (100 g PIF: 300 mL BPW) and enriched for 16 h at 37 °C. After enrichment, a 150 µL direct sample for raw ground turkey, or a 150 µL 1:10 diluted sample for PIF, was transferred to a 96-well plate. Volumes of 10 µL of phage reagent were added and samples were incubated at 37 °C for 2 h. Then, 65 µL of luciferase detection solution was added. Luminescence was measured as previously described. Ground turkey samples and powdered infant formula samples with signals ≥ 750 RLUs were considered positive.

Data availability
Unannotated raw genome assemblies of SEA1 and TSP1 are available upon request. All other data generated or analyzed during this study are included in this published article (and its Supplementary information files). Researchers receiving resources generated in this study may be asked to sign a Materials Transfer Agreement that covers potential commercial use.