Four foods were tested in this study including cilantro, raw chicken thighs, liquid whole eggs, and peanut butter. Cilantro samples, provided by the Department of Agriculture and Rural Development in Lansing, Michigan or purchased from a local grocery store, were stored at 4°C prior to testing. Raw chicken thighs and liquid whole eggs, purchased from a local supermarket, were stored at 4°C prior to testing. Peanut butter samples, purchased from a local grocery store, were stored at room temperature.

Nine Salmonella enterica strains representing four serovars, Newport, Tennessee, Thompson, and Enteritidis, were used for food inoculation (Table 1). Two of the strains used to inoculate cilantro, S. Newport SALC14 and S. Tennessee SALC 76, were cultured from cilantro in our laboratory and the third strain, S. Thompson SALC 818, was chosen because S. Thompson was implicated in a cilantro outbreak in 1999 (Campbell et al., 2001). The three S. Tennessee strains used to inoculated peanut butter were isolated from peanut butter during a 2007 peanut butter outbreak (Wilson et al., 2016) and the S. Enteritidis strains used for the chicken and egg inoculations were isolated from the respective foods (Table 1). Culture stocks, stored at –80°C, were streaked onto Trypticase Soy Agar plates (Difco, Sparks, MD, USA) and incubated overnight at 35 ± 2°C. Bacterial cell suspensions, equal to 0.48–0.52 McFarland turbidity units (corresponding to approximately 1 × 10⁸ CFU/ml), were prepared in 0.85% sterile saline and serially diluted to approximately 28 CFU/mL for food inoculations.

Prior to inoculation, liquid whole eggs were homogenized in sterile beakers, and the raw chicken thighs were cubed into approximately 1-inch squares using sterile knives. All foods were aseptically portioned (25 g) into sterile Whirlpak bags (Nasco; Fort Atkinson, WI, USA), inoculated with approximately 28 CFU Salmonella per 25 g of food, and aged to simulate natural contamination at 4°C for 48–72 h (cilantro, liquid whole eggs, and chicken thighs) or at room temperature for 14 days (peanut butter). Un-inoculated controls samples were also prepared and aged in parallel for each food.

Aged cilantro samples were enriched in Lactose broth, UP or TSB prepared according to the FDA BAM. Peanut butter and additional cilantro samples were enriched in Modified Buffer Peptone Water [mBPW; Buffered Peptone Water (BPW, Difco^TM, Sparks, MD, USA), with 3.5 g of disodium phosphate and 1.5 g of monopotassium phosphate per liter]. Aged liquid whole eggs were enriched in TSB with ferrous sulfate prepared according to the FDA BAM. The aged chicken thighs were enriched in BPW, the non-selective pre-enrichment broth recommended in the MLG method for the isolation and identification of Salmonella from poultry.

Two modifications that reduce the selective strength of TT broth were compared to the standard TT broth formulation used in the FDA BAM. Standard BAM TT broth consists of a base broth to which Iodine-Potassium Iodide (I₂-KI) and brilliant green dye solutions are added on the day of use. Both TT modifications used in this study consist of the TT broth base without the addition of brilliant green dye. TT modification A (TT_A) consists of the TT broth base with 2.0% of the I₂–KI solution as per the BAM, and TT modification B (TT_B) contains only 1.0% of the I₂–KI solution. The FDA BAM, USDA MLG, and ISO methods all use a second selective enrichment broth called RV for Salmonella detection. The RV used in this study was prepared in accordance with the FDA BAM method.

A panel of 101 Salmonella isolates, derived from the Defense Science Office (DSO) of the Defense Advanced Research Projects Agency (DARPA), Systems and Assays for Food Examination (SAFE) Program, were cultured in selective RV, TT, TT_A, and TT_B broths to compare efficacy for Salmonella recovery. The SAFE collection is used at the FDA for validating new methods for Salmonella detection in food. The collection is divided into three groups representing a Salmonella subspecies set (55 strains), an outbreak cluster set (26 strains), and a food set (20 strains, Supplementary Table S1). Parallel sets of RV, TT, TT_A, and TT_B selective broths were inoculated with 10³ or 10⁵ CFU of each Salmonella isolate and incubated at 42 ± 2°C (RV) or 43 ± 0.2°C (TT, TT_A, and TT_B) for 24 ± 2-h. The following day 10 μl of each culture was plated onto Xylose-Lysine-Desoxycholate Agar (XLD, Becton, Dickinson and Company, Sparks, MD, USA) agar, and after a 24 h incubation at 35 ± 2°C, the plates were recorded as positive (black colonies) or negative (no growth) for Salmonella.

All food samples were processed following the FDA BAM workflow for the detection of Salmonella with a 24-h non-selective pre-enrichment and the standard and modified TT broths, as follows. On day one after aging, food samples were aseptically combined with 250mL of sterile non-selective pre-enrichment broth, massaged for 2 min, and incubated overnight at 35 ± 2°C (Figure 1). On day two, aliquots of the 24-h non-selective pre-enrichments were aseptically transferred to RV (100 μL), TT (1.0 mL), TT_A (1.0 mL), and TT_B (1.0 mL) selective enrichment broths, and incubated overnight at 42 ± 0.2°C (RV) or 43 ± 0.2°C (TT) (Figure 1). On day three, 10 μL of each selective enrichment was plated on Xylose Lysine Tergitol^TM 4 (XLT4, Becton, Dickinson and Company, Sparks, MD, USA) agar and incubated overnight at 35 ± 2°C. The remaining volumes of RV, TT, TT_A, and TT_B selective enrichments from cilantro samples were centrifuged at 7,100 rcf for 30 min to pellet the bacteria for DNA extraction, to represent 24RV, 24TT, 24TT_A, and 24TT_B time points used for 16S rRNA gene sequencing (Figure 1). Bacterial pellets were stored at –20°C. On day four, suspect black colonies from XLT4 plates were confirmed as Salmonella on the VITEK 2 system (BioMérieux, France).

In addition to the samples processed according the FDA BAM timeline, replicate samples of each food were incubated for a reduced time in their respective pre-enrichment broths (Figure 1). Specifically, on day one, the non-selective pre-enrichments were incubated at 35 ± 2°C for 5 or 6-h, and aliquots from each time period were transferred to RV (100 μL), TT (1.0 mL), TT_A (1.0 mL), and TT_B (1.0 mL) selective broths and incubated overnight at 42 ± 0.2°C (RV) or 43 ± 0.2°C (TT). On day two, each selective broth was streaked onto XLT4 agar and incubated overnight. Remaining volumes of RV, TT, TT_A, and TT_B selective broths from each time point were centrifuged at 7,100 rcf for 30 min to pellet the bacteria for DNA extraction to represent 5 and 6-h RV, TT, TT_A, TT_B samples (hereafter referred to as sRV, sTT, sTT_A, and sTT_B) for 16S rRNA gene sequencing (Figure 1). All bacterial pellets were stored at –20°C. On day three, one day earlier than the FDA BAM method, suspect black colonies from XLT4 agar plates were confirmed as Salmonella on the VITEK 2 system.

Cilantro non-selective pre-enrichments incubated for 5, 6, and 24-h were enumerated for Salmonella using the Most Probable Number (MPN) method described in the FDA BAM. Non-selective pre-enrichment samples were serially diluted (10-fold) in 0.85% NaCl. Selective TT_B was inoculated, in triplicate, with 1.0 mL of each dilution, incubated at 43 ± 0.2°C for 48-h, and then streaked onto XLT4. The XLT4 agar plates were examined for the presence of typical black Salmonella colonies, which were confirmed in the VITEK 2 system. The FDA BAM MPN calculator was used to compute the results (www.fda.gov/Food/FoodScienceResearch/LaboratoryMethods/ucm109656.htm).

Bacterial pellets from the selective RV, TT, TT_A, and TT_B enrichment samples were lysed for 20-min at room temperature with 100 ng/uL lysozyme (Sigma–Aldrich; Saint Louis, MO, USA), followed by DNA extraction on the QIACube using the QIAamp DNA mini protocol (QIAGEN, Germany).

A two-step PCR amplification strategy was used to generate barcoded 16S rRNA gene amplicons with four sets of PCR primers targeting the Variable 1 to Variable 3 (V1 to V3) regions of the gene. These primers have added base pairs upstream of the 16S rRNA gene aimed at adding nucleotide base diversity to the initial sequencing cycles on the Illumina MiSeq (Supplementary Table S2). Four master mixes were prepared, one for each primer set, consisting of 3.0 ng template DNA, 5.0 μl 5× Omni Klentaq Master Mix (DNA Polymerase Technology, Inc., St. Louis, MO, USA), 10.0 μl PCR Enhancer Cocktail (DNA Polymerase Technology, Inc., St. Louis, MO, USA), and 1.5 μl each forward and reverse primers (600nM final concentration). PCR cycling conditions consisted of a 2-minute denaturation at 94°C, followed by 25 cycles of 40 s at 94°C, 15 s at 56°C, 40 s at 68°C, and a final extension at 68°C for 5 min. The 16S rRNA gene PCR amplicons were visualized on 1% agarose gels and, purified using the QIAquick PCR purification kit (QIAGEN, Germantown, MD, USA) if multiple bands were present.

The second set of PCR reactions used the 16S rRNA gene amplicon PCR products as template with the Illumina Nextera XT indexing primers to generate unique PCR amplicons that were compatible with the Illumina MiSeq sequencing chemistry. Briefly, the 16S rRNA gene amplicons were quantified using the Qubit 2.0 Fluorometer (Life Technologies, Carlsbad, CA, USA), then amplified in 50 μl reactions with ∼150 ng amplicon products, 5× Omni Klentaq Master Mix, and 5 μl each of the Illumina i7 and i5 indexing primers from the Nextera XT Indexing Kit (Illumina Inc., San Diego, CA, USA). Indexing PCR cycling conditions included initial 3 min and 30 s holds at 72 and 98°C, respectively, followed by 5 cycles of 10 s at 98°C, 30 s at 63°C, and 3 min at 72°C, and a final hold at 10°C.

Libraries were size selected using Agencourt AMPure XP beads (Beckman Coulter, Brea, CA, USA) to an average library size of 600 bp, and then quantified using the Qubit 2.0 Fluorometer. Library quality was verified on the Agilent Technologies 2100 Bioanalyzer (Agilent Technologies, Inc., Santa Clara, CA, USA) using the DNA 1000 chip kit. Size selected libraries were normalized to 2 nM using the SequelPrep Normalization Plate Kit, pooled, and then measured using High Sensitivity Qubit Reagents (Life Technologies, Carlsbad, CA, USA). The library pools were denatured with 0.1N NaOH, serially diluted to 8pM with HT1, spiked with 10% 12.5pM PhiX174 bacteriophage DNA, and then sequenced using a 600-cycle MiSeq Reagent Kit v3 (Illumina Inc., San Diego, CA, USA) in four sequencing runs, two with 96 samples, one with 86 samples, and one with 87 samples.

Raw paired-end reads output by the MiSeq platform were merged into consensus fragments by FLASH (Magoc and Salzberg, 2011) and subsequently filtered for quality (max error rate 1%) and length (minimum 200bp) using Trimmomatic and QIIME (Caporaso et al., 2010; Kuczynski et al., 2011; Bolger et al., 2014). Spurious hits to the PhiX control genome were identified using BLASTn and removed. Passing sequences were trimmed of primers and evaluated for chimeras with UCLUST (de novo mode) (Edgar et al., 2011). Chloroplast contaminants were detected and filtered using the RDP classifier with a confidence threshold of 80% (Wang et al., 2007). Sequences were further screened for unknown contaminant using a sensitive BLASTn search against the GreenGenes database (McDonald et al., 2012). To normalize across samples, the 16S rRNA gene sequence profiles were subsampled to an even level of coverage prior to downstream comparisons. High-quality 16S rRNA gene sequences were assigned to a taxonomic lineage using Resphera Insight (Baltimore, MD, USA; www.respherabio.com). Mock community analyses compared Resphera Insight assignments to default UCLUST-Ref and RDP pipelines implemented within QIIME (Caporaso et al., 2010; Kuczynski et al., 2011).

Two Illumina MiSeq runs multiplexed with 96 samples produced a total of 9,631,743 and 6,949,843 quality filtered 16S rRNA gene sequences with average (average ± SD) read lengths of 412±19 and 378±27, respectively. Eighty-seven multiplexed samples in a third Illumina MiSeq run generated 8,875,924 quality-filtered sequences with an average read length of 409±24. A fourth MiSeq run of 86 samples yielded 7,631,971 quality filtered reads with aver age read lengths of 399±24. Raw 16S amplicon sequences reported in this paper have been deposited in the NCBI SRA repository (BioProject accession no. PRJNA343808).

To perform an external validation of the species level accuracy of Resphera Insight, we utilized 110 whole-genome shotgun datasets from the GenomeTrakr Project (NCBI Project ID PRJNA183844) designated as novel S. enterica isolates. Raw paired-end sequences were filtered for quality and length, followed by merging of overlapping sequences using FLASH (Magoc and Salzberg, 2011). Merged reads were then screened for 16S rRNA fragments using Bowtie2 (Langmead and Salzberg, 2012) against a broad database of 16S rRNA gene sequences, with additional BLAST-based filtering to confirm location specific query matches to a reference S. enterica 16S rRNA gene (Salmonella enterica ssp. enterica serovar Typhimurium strain LT2; NCBI accession NR_074910.1).

Passing sequences were submitted to Resphera Insight (Baltimore, MD, USA; www.respherabio.com) for high-resolution taxonomic identification. We also compared performance of Insight to the RDP classifier and UCLUST reference algorithms implemented within QIIME (Kuczynski et al., 2012; Navas-Molina et al., 2013). Our primary measure of performance was the Diagnostic True Positive rate (DTP), defined as the percentage of reads with an unambiguous assignment to S. enterica, and we evaluated differences in accuracy associated with changes in read length and gene position.

Z tests were performed to compare the recovery rates of Salmonella SAFE isolates inoculated into TT, TT_A, and TT_B. Z tests were also used to compare the efficacy of Lactose, mBPW, TSB, and UP non-selective enrichments at 5, 6, and 24 h for recovery of Salmonella from selective RV, TT_A, TT_B, and TT and to compare the efficacy of the selective enrichment broths. For the ladder comparisons the p values were adjusted for multiple comparisons, using the Benjamini and Hochberg method (R software v3.2) after continuity corrections were implemented (SAS 9.4).

Since a reduced non-selective pre-enrichment will likely have fewer metabolically active cells of Salmonella, as compared to a 24 h pre-enrichment, we first evaluated the sensitivity of standard and modified RV and TT selective enrichment broths at a low inoculum level to mimic these conditions. To do this we directly inoculated RV and TT broths with 10³ CFU/mL of each of the 101 SAFE isolates. We also directly inoculated TT and RV broths with a high inoculum of 10⁵ CFU/mL as control samples. Among the selective broths inoculated with 10³ CFU/mL Salmonella, the TT_B and TT_A selective broths had significantly higher recovery rates of 92 and 88% (P ≤ 0.001) compared to the FDA BAM TT formulation (17%) (Table 2; Supplementary Table S1). The Outbreak cluster set had the lowest recovery from the BAM TT broth of 4% (Table 2; Supplementary Table S1). Furthermore, the Outbreak cluster and Food sets of Salmonella SAFE strains were recovered at 100 and 90% in TT_B, respectively, both significantly higher than the FDA BAM TT formulations (Table 2; Supplementary Table S1). Interestingly, the recovery of Salmonella from BAM RV (58%) was also lower than TT_A (88%) and TT_B (92%) for the entire SAFE collection (Table 2; Supplementary Table S1).

We inoculated a second set of BAM RV and TT broths with the SAFE Salmonella collection at 10⁵ CFU/mL and were able to recover 97 and 75% of the isolates, respectively, indicating that for some of the SAFE panel strains, the BAM selective broths require more than10⁵ CFU/mL in the food non-selective pre-enrichment for Salmonella recovery (Table 2; Supplementary Table S1). The Subspecies set of Salmonella isolates had the lowest recovery levels in the 10⁵ CFU/mL BAM RV (95%) and TT (75%, Table 2). The Outbreak cluster and Food sets of SAFE strains were recovered in 100% of the BAM RV, and in 73 and 80% in BAM TT broths, respectively, at this inoculum level (Table 2).

In summary, we determined that TT_A and TT_B broths significantly improved the recovery of Salmonella (inoculated at 10³ CFU/mL) and yielded positive results in 88 and 92%, respectively, of the SAFE isolates tested, as compared to the BAM TT and RV recovery rates of 17 and 58%, respectively (Table 2).

Before testing our reduced non-selective pre-enrichment protocol, we first tested two non-selective pre-enrichment broths (Lactose and mBPW), with three TT selective broth formulations (TT, TT_A, TT_B) and RV, for the recovery of Salmonella from cilantro, to determine the efficacy of these different media formulations on food samples. For these experiments, cilantro samples were inoculated with Salmonella at ∼28 CFU/25 g, and aged at 4°C prior to the 24-h non-selective pre-enrichment. The 24-h non-selective pre-enrichments were inoculated into the four selective RV and TT broths, and then plated on XLT4 for Salmonella isolation and confirmation (Figure 1). The 24-h Lactose broth pre-enrichments (n = 13) had recovery rates of 77% (10/13), 69% (9/13), 92% (12/13), and 92% (12/13) from the RV, TT, TT_A, and TT_B selective enrichments, respectively (Table 3). In comparison, selective enrichment broths inoculated with 24-h mBPW non-selective pre-enrichments (n = 33) resulted in 100% recovery rates in all four of the selective enrichment broths, suggesting that mBPW improved resuscitation of Salmonella in cilantro (Table 3). Enumeration of three Lactose and two mBPW 24-h non-selective pre-enrichments using the MPN method revealed Salmonella levels of 0.00 in Lactose broth and 7.44 ± 1.15 log MPN g^-1 in mBPW, corroborating the lower recovery rates from Lactose broth in 24-h non-selective pre-enrichments (Table 4).

To optimize a shortened enrichment time for Salmonella, the four selective broths (RV, TT, TT_A, and TT_B) were tested in parallel with mBPW and Lactose non-selective pre-enrichments incubated for 5 and 6-h, instead of the 24-h time period recommended in the BAM (Figure 1). Salmonella was recovered one day earlier on XLT4 from 97% (32/33) of cilantro samples when TT_B broths were inoculated with 6-h mBPW pre-enrichments, which were found to harbor 2.00 ± 0.42 MPN g^-1 Salmonella (Tables 3 and 4). In contrast, Salmonella recovery was only 69% (9/15) when TT_B was inoculated with a 6-h Lactose pre-enrichment, which had a correspondingly lower level of Salmonella contamination, 0.37 ± 0.32 MPN g^-1 (Tables 3 and 4). In the other selective broths inoculated with 6-h mBPW and Lactose pre-enrichments, recovery, respectively, decreased to 85% (28/33) and 46% (6/13) in TT_A, 3% (1/33) and 15% (2/15) in TT, and 9% (3/33) and 38% (5/13) in RV (Table 3). Recovery of typical-Salmonella colonies on XLT4 was possible but not consistent using 5-h resuscitated mBPW and Lactose broth samples, ranging from 0 to 52% depending on the selective broth used (Table 3). It is notable that the XLT4 plates originating from TT_B broth inoculated with 5 and 6-h non-selective pre-enrichments had little to no background flora, whereas those from 24-h non-selective pre-enrichments were consistently mixed cultures.

Two additional non-selective pre-enrichment broths, TSB and UP, were tested for their ability of to recover Salmonella one day earlier from selective enrichments. For these studies RV and TT_B were inoculated since the data support improved recovery of Salmonella. Similar to the mBPW results, TT_B resulted in higher Salmonella recovery compared to RV when plated on XLT4. Both UP and TSB broths resulted in recoveries of 100% (5/5) in TT_B and 40% (2/5) in RV using 6-h pre-enrichments that had 2.10 ± 0.52 log MPN g^-1 (UP) and 2.16 ± 0.20 log MPN g^-1 (TSB) Salmonella, respectively (Tables 3 and 4). Recovery of Salmonella was also 100% when either 5-h UP (n = 5) or TSB (n = 5) non-selective pre-enrichments containing 1.79 ± 0.17 and 1.36 ± 0.81 log MPN g^-1 Salmonella, respectively, were inoculated into TT_B (Tables 3 and 4). However, recovery decreased to 20% (1/5, UP) and 40% (2/5, TSB) when the same non-selective pre-enrichments were inoculated into RV broth (Table 3). Salmonella was recovered from RV and TT_B control samples inoculated with 24-h UP and TSB non-selective pre-enrichments 100% of the time, and similar to the mBPW 24-h non-selective pre-enrichments, the log MPN g^-1 Salmonella reached 8.74 ± 1.09 in UP samples and 8.10 ± 1.12 in TSB (Tables 3 and 4).

A final set of experiments with raw chicken thighs (n = 4), liquid whole eggs (n = 3), and peanut butter (n = 3) confirmed early detection of Salmonella in these foods. TT_B broths inoculated with 6-h non-selective pre-enrichments of chicken, eggs, or peanut butter were positive for Salmonella in all three foods. Additionally, using 5-h pre-enrichments, 100% of chicken and peanut butter, and 67% (2/3) of egg samples were Salmonella-positive. The recovery of Salmonella from RV broth inoculated with 6-h non-selective pre-enrichments was 100% in chicken thighs, 33% (1/3) in eggs, and 67% (2/3) in peanut butter. All 24-h chicken, egg, and peanut butter non-selective pre-enrichments resulted in positive RV selective enrichments.

In summary we significantly improved Salmonella recovery from cilantro by reducing the length of non-selective pre-enrichment in mBPW from 24 to 6-h, and lowering the selective strength of the BAM TT broth by removing brilliant green dye and reducing the concentration of I₂–KI. These changes to the FDA BAM method resulted in the recovery of Salmonella one day early. Additionally, although a larger sample population is required for confirmation, our preliminary results with raw chicken thighs, liquid whole eggs, and peanut butter support that Salmonella recovery can be improved in these commodities. Taken together, our data from 6-h non-selective pre-enrichments of cilantro (excluding lactose replicates), chicken thighs, liquid whole eggs, and peanut butter enabled the recovery of Salmonella 1 day earlier than the BAM method in 98% (52/53) of the samples using TT_B and 26% (14/53) using RV supporting an overall improvement. In contrast, the FDA BAM TT broth resulted in very low recoveries of Salmonella from 6-h cilantro non-selective mBPW (3%) and Lactose (15%) pre-enrichments (Table 3).

We also tested cilantro samples in TSB and UP non-selective broths incubated for 5 and 6-h and found that Salmonella recovery was 100% in TT_B broth which is higher than the recovery from RV. However, with the small sample sizes (five samples for each broth) tested we did not have the statistical power to show a significant improvement with TT_B broth (Table 3).

Among species within the family Enterobacteriaceae, there can be high levels of similarity in 16S rRNA gene sequences, and as a result, many bioinformatics tools maintain poor sensitivity to detect S. enterica at the species level. Therefore, we employed the Resphera Insight algorithm for high-resolution taxonomic assignment, which we validated for accuracy on 110 novel isolates of S. enterica from the GenomeTrakr Project. Overall, between 16S rRNA gene positions 27 and 534, the V1 to V3 region sequenced in this study, Resphera Insight achieves diagnostic true positive rates up to 99.8% for S. enterica with improved accuracy associated with increased read length. In contrast, RDP and UCLUST were unable to achieve a DTP above 0.1% in the V1 to V3 regions (Supplementary Figure S1). It is notable that RDP and UCLUST had higher DTP at the 3′ end of the 16S rRNA gene at positions 800 to 900, and the highest DTP percentages with these algorithms were observed with smaller 16S rRNA gene fragments (Supplementary Figure S1).

We employed 16S rRNA gene sequencing to characterize the microbial communities in selective RV, TT, and TT_B broth samples originating from 5, 6, and 24-h cilantro non-selective pre-enrichments (Figures 1 and 2). The samples sequenced included RV inoculated with 5, and 6-h non-selective pre-enrichments (sRV, n = 6), or 24-h non-selective pre-enrichments (24RV, n = 9); TT_B inoculated with 5, and 6-h non-selective pre-enrichments (sTT_B, n = 8) or 24-h non-selective pre-enrichments (24TT_B, n = 9); and TT inoculated with 6-h non-selective pre-enrichments (sTT, n = 2), or 24-h non-selective pre-enrichments (24TT, n = 2) (Figures 1 and 2).

Average S. enterica proportional abundances were 92% for sRV, 87% for 24RV, 92% for sTT_B, 40% for 24TT_B, 0.07% for sTT, and 58% for 24TT (Figure 2). Although all RV enrichments were similar in appearance when streaked on XLT4 plates, a multivariate linear regression analysis of Salmonella proportional abundances showed that sRV significantly increased Salmonella proportional abundances relative to 24RV (P = 1.58e-04) (Figure 2). Salmonella proportional abundances were also significantly higher in sTT_B compared to 24TT_B (P = 2.40e-05), supporting the observed reduction in background flora on XLT4 plates derived from the sTT_B enrichments (Figure 2, data not shown). Not surprisingly, the Salmonella abundances were significantly lower in sTT than in the 24TT selective enrichments (P = 2.0e-03) (Figure 2). The 24TT_B and 24TT selective enrichments had statistically similar proportional abundances of Salmonella while the abundances in 24RV were significantly higher than 24TT_B (P = 2.59e-05) (Figure 2). The proportional abundances of Salmonella in the sRV and sTT_B enrichments were statistically similar (P = 8.39e-01) as were those in 24RV and 24TT BAM enrichments (P = 1.51e-01). Finally, the proportional abundances of Salmonella in the sTT_B enrichments were significantly higher than sTT since TT broth is too harsh for a shortened non-selective pre-enrichment (Figure 2).

Taken together, these 16S rRNA gene sequencing data suggest that a 24-h non-selective cilantro pre-enrichment allows resident bacterial species to resuscitate, thrive, and compete with Salmonella during selective enrichment, whereas the shortened non-selective pre-enrichment is insufficient for resuscitation of resident bacteria and gives Salmonella a competitive edge. In either case, those bacteria that can metabolize the selective agents in RV and TT_B will grow and thrive with Salmonella during the selective enrichments (Figure 2). The 16S rRNA gene profiles of TT_B samples identified these competitive families as non-Salmonella Enterobacteriaceae, Planococcaceae, Peptostreptococcaceae, and Bacillales Incertae Sedis XII. Although bacteria in these families compete with Salmonella in TT_B, most are Gram-positive and were not observed on XLT4 plates. However, members of the non-Salmonella Enterobacteriaceae family (i.e., Citrobacter sp. and Enterobacter sp.) were commonly isolated from XLT4 after enrichment in RV and TT_B.