Sixteen serotypes of Salmonella enterica subsp. enterica and other reference strains phylogenetically close to Salmonella spp., such as Escherichia coli, Klebsiella spp., and Shigella spp. (Table 1) were cultured in solid tryptone soya agar (TSA) or XLT4 (xylose, lysine, and tergitol 4) and liquid selective medium Rappaport Vassiliadis (RV). All strains were grown at 37°C for 12–18 h in liquid medium RV before DNA extraction.

A protocol equally efficient for both Gram-positive and Gram-negative bacteria was used with some modifications (Atashpaz et al., 2010). Briefly, 3 mL cell culture in the exponential phase was centrifuged at 5,000 rpm for 5 min at 4°C. Then, 800 μL lysis buffer [CTAB 2% (w/v), Tris−HCl 100 mM, NaCl 1.4 M, EDTA 20 mM, and LiCl 0.2% (w/v), pH 8.0] was added to the bacterial pellet. The samples were incubated at 65°C for 30 min for Gram-negative bacteria and 2 h for Gram-positive bacteria. Then, these were centrifuged at 10,000 rpm for 5 min at 4°C. The supernatant was transferred to a new tube where 1 volume of chloroform-isoamyl alcohol (24:1) was added to precipitate the proteins. The samples were gently mixed and centrifuged at 12,000 rpm for 8 min at 4°C. Two further extractions with chloroform-isoamyl alcohol (24:1) were performed on the aqueous phases for each sample. From the pooled supernatants, the DNA was precipitated by the addition of 1 volume of cold isopropanol and 100 μL of 5 M sodium acetate after overnight incubation at −20°C. Subsequently, the samples were centrifuged at 12,000 rpm for 15 min at 4°C and washed with 1 mL of 70% ethanol by inverting the tubes. Finally, once the ethanol had evaporated, the DNA was resuspended in 100 μL 1X TE buffer (Tris–HCl 10 mM, EDTA 1 mM pH 8.0) and stored at −20°C. The quality of the extracted genomic DNA was evaluated using UV spectrophotometry by measuring the absorbance ratios at 260 nm/280 nm and 260 nm/230 nm; DNA integrity was assessed using electrophoresis on 1% agarose gel.

The primers and probes (Biosearch Technologies, Petaluma, CA, United States, purified by HPLC) for each target are shown in Table 2.

Quantitative PCR was used as a preliminary test for evaluating the reaction efficiency of primers and probes for each target in simplex mode (CFX 96 Deep well- BioRad, cat. 1855196). Six dilutions of S. enteritidis reference strain ATCC 13076 DNA in 1X TE buffer were prepared (100–0.01 ng/μL). The 20 μL reaction contained 1X iTaq^TM Universal Probes Supermix (BioRad cat. 1725131), 400 nM primers, 300 nM probes, and nuclease-free water. 1X TE was used as a non-template control (NTC). All samples were run in triplicate. The amplification cycle consisted of initial denaturation at 95°C for 10 min, followed by 40 cycles of denaturation at 95°C for 15 s and annealing-elongation at 60°C for 30 s, with a heating ramp of 2°C/s. Efficiency was evaluated using Eq. 1,

where E is the efficiency and m the slope of a linear regression between the amplification cycle and the log concentration of each sample. The acceptability criterion for PCR efficiency was 90–100%.

In order to establish the best conditions to perform the validation assay for every target sequence, the annealing temperature through a gradient from 55–63°C was evaluated. In addition, the primer concentration was also evaluated from 300 to 900 nM, and the ramp rate between 1 and 2°C/s. The aim was to establish general conditions to amplify the target sequences under the same experimental conditions. After optimization, the best amplification temperature was 60°C with a ramp rate of 2°C/s and 600 nM concentration of primers.

After optimization, according to MIQE guidelines (Huggett et al., 2013; Supplementary Table S8), ddPCR assay was performed in simplex (spaQ), and duplex (hilA-siiA and invA-ttr) form: (i) according to the number of total reactions, a stock master mix with 1X ddPCR^TM Supermix for Probes (BioRad, CA, United States cat. 1863024), 600 nM primers, 300 nM probes, and nuclease-free water was prepared; (ii) for every triplicate 60 μL of the stock master mix was weighed, (iii) and finally 6 μL of DNA template (genomic DNA of S. enteritidis, measured gravimetrically) were added to complete 66 μL. Then to avoid pipetting bias, 21 μL PCR reaction mixture and 70 μL generator oil (BioRad cat. 183005) were loaded into an 8-well DG8 cartridge (BioRad cat. 1864008), for every replicate, in order to have the same dilution factor. Droplets were generated in a QX200 droplet generator, transferred to a 96-well plate (BioRad cat. 12001925), and sealed with a PX1 PCR plate sealer (BioRad cat. 1814000). Reactions were amplified in a CFX96 deep-well thermocycler (BioRad cat. 185–5196), and the cycling conditions were: 95°C for 10 min, 40 cycles at 95°C for 30 s, and 60°C for 1 min, with an overall ramp rate of 2°C/s. All the reactions were read in the QX200 droplet reader system (two detection channels FAM/EvaGreen and VIC/HEX) using the Quantasoft^TM software V1.7 from BioRad.

Droplets were classified as PCR-positive or PCR-negative according to a threshold fluorescence value set manually using the Quantasoft^TM software V1.7. Technical reasons for excluding data from the analysis were: (i) the total number of droplets <12,000 (BioRad, 2018) and (ii) wells with amplitude different from those of other wells with the same target. This variance indicated poor droplet generation, possibly due to poor handling or mixing of samples, and is likely to generate erroneous concentration values. Reference material ERM-AD623 (White et al., 2015) was used as the amplification control for assessing the performance of the thermocycling parameters.

Data were exported to a spreadsheet to calculate the concentration of the sample in cp/μL in the ddPCR reaction, with a droplet volume of 0.819 ± 0.017 nL (k = 2; Eqs 2, 3), as long as the confidence limits were associated with Poisson distribution.

where:

C_sample: Sample concentration (cpuL)

_λ : Copies/partition

N: Number of negative partitions

P: Number of total partitions

v: Droplet volume

D: Total dilution of the sample from the master mix.

The droplet volume was calculated as the mean of values reported previously (Corbisier et al., 2015; Dagata et al., 2016) and the uncertainty was estimated based on a rectangular distribution, to encompass any influence of droplet volume variability and to avoid underestimation of measurement uncertainty (Košir et al., 2017; Emslie et al., 2019).

The performance characteristics evaluated were specificity, working interval, equivalence between simplex and duplex assay, precision as repeatability and intermediate precision, the LOD, the LOQ, and measurement uncertainty.

From the optimized extraction process, 1 mL of 4430 ng/μL of DNA was obtained, 1:4 dilution was prepared as a working solution (WS; see Supplementary Figure S1). Preliminary tests using qPCR were performed to determine whether all gene sequences amplified correctly and whether PCR inhibitors were present. The qPCR amplification efficiencies calculated for the five targets were between 90 and 99% (Table 3), indicating that inhibitors were not present in the starting genomic DNA. The dilutions and log concentration showed a good correlation, indicating the suitability of qPCR for the studied targets.

The DNA of sixteen serotypes of Salmonella enterica subsp. enterica (Table 1) was amplified using the optimized ddPCR method. On the other hand, the DNA of four enterobacteria groups closely related and not related to the species under study were also evaluated. Each group contained 1 ng/μL DNA of each species, and they were classified as (i) Gram-positive bacteria group: Staphylococcus aureus subsp. aureus, Bacillus cereus, and Enterococcus faecalis; (ii) Shigella group: Shigella sonnei, Shigella boydii, and (iii) SHEC Escherichia Group: E. coli O104: H4, E. coli O145: NM, E. coli O157: H7. As a positive control for this assay, a DNA mixture of 1 ng/μL of four serotypes of Salmonella spp., (S. typhimurium, S. typhi, S. derby, and S. paratyphi A.) was used.

The specificity of the selected sequences was evaluated initially using an in silico analysis. Then, sixteen serotypes of Salmonella spp. (Table 4) were evaluated using a ddPCR assay. The results indicated that a positive amplification response was generated for all Salmonella serotypes and that negative amplification results were generated for all Enterobacteria strains evaluated.

Using S. enteritidis ATCC 13076 DNA, five WS from 10 to 0.001 ng/μL (as nominal concentration), were prepared. Serial gravimetric dilution from WS to ddPCR master mix reaction were measured in triplicate on three different days (see Supplementary Tables S1–S5). The working interval was determined using least squares regression analysis of the linear relationship between each dilution and their corresponding concentration (cp/μL) using the ddPCR method (Table 5). Dilution value is used to analyze the working interval as the nominal concentration is just an estimate of the real concentration. An example is presented in Figure 1. Grubbs’ test was performed to determine outliers on one hilA level replicate. The results for the NTC were negative.

The acceptance criteria for linearity were as follows: a slope statistically different from zero and correlation coefficient >0.99 (Deprez et al., 2012). The method showed excellent linearity between 1 and 8,000 cp/μL in reaction (except for siiA, the interval of which was from 8 to 8,000 cp/μL). Therefore, the working interval for the whole method was established as 8–8,000 cp/μL.

As part of the validation process, the ability of the method to run in a duplex form (invA-ttr and hilA- siiA) was also evaluated through regression analysis. Both methods were compared over the working interval, using data of four concentration levels (8–8,000 cp/μL) in triplicate and on three different days (Figure 2; see Supplementary Table S6). A slope statistically equal to 1 and correlation coefficient >0.99 were set as acceptance criteria.

According to Figure 2, the results obtained did not differ when invA and ttr were measured independently or in combination. Identical results were obtained for hilA and siiA target sequences (see Supplementary Figure S2). This allowed simultaneous evaluation of four different targets in two duplex reactions with the same confidence.

In addition, the amplitude heat plots of 1D and 2D duplex reactions showed clear populations of droplet distribution in positive and negative partitions, minimizing misclassification of droplets (rain) and absence of cross-reaction (Figure 3). As DNA distribution follows a random pattern, in the 2D plot, droplets are cluster in four groups: (i) double negatives, (ii) FAM positives and HEX negatives, (iii) FAM negatives and HEX positives, or (iv) double positives; not all positive partitions are double-positive in the duplex assay (Figure 3C) due to some possible fragmentation process that affect DNA integrity (Furuta-Hanawa et al., 2019); Table 6 shows the variation of clusters fraction with DNA concentration.

Five concentration levels covering 1–8,000 cp/μL in ddPCR reaction, in triplicate reactions on three different days (see Supplementary Tables S1–S5), relative repeatability standard deviation (Eq. 4), and relative intermediate standard deviation (Eq. 5), were calculated using one factor ANOVA were calculated using one-factor ANOVA (Deprez et al., 2016).

where

MS_{witin run} : Within run mean squares calculated using one-way ANOVA

MS_{between run} : Between run mean squares calculated using one-way ANOVA

n_replicates: Number of replicates per run

C_{sample, mean}: Averaged copy number concentration calculated over all runs.

Repeatability was below 5% for the three highest concentration levels and increased to 10% for the fourth one. However, for the lowest concentration (approximately 1 cp/μL in reaction), repeatability, was beyond 20% (Figure 4A), while intermediate precision was below 5% in almost the entire interval, except for ttr, and siiA (Figure 4B). When MS between <MS within, the S_{interm, rel} was considered negligible with respect to S_repeat,rel.

Limit of quantification represents the lowest copy number of the analyte that can be determined with acceptable performance. We have defined the performance criteria as relative repeatability standard deviation <20%, according to the precision study. The higher values represent a significant increase in the measurement uncertainty. Thus, the LOQ based on the response of all the genes in this study was established as 8 cp/μL (Figure 4A).

Limit of detection is defined as the lowest analyte concentration that can be distinguished from zero with a specified level of confidence. LOD was determined as the lowest concentration level, where at least three positive droplets were present in all three replicates (Morisset et al., 2013), or equivalent in at least nine positive partitions in the pooled replicates. Amplifications were performed in a simplex mode for spaQ and duplex mode for invA-ttr and hilA-siiA. Five gravimetric serial dilutions (1–0.1 cp/μL in the ddPCR reaction) were prepared and run in triplicate. Figure 5 presents the results in which each bar represents the total positive partitions obtained for three replicates compared to a total negative NTC. The estimated LOD was 0.5 cp/μL in the ddPCR reaction for three positive droplets (on average) in at least 12,000 total partitions.

Based on precision data and considering the mathematical model, the measurement uncertainty estimation was established: first, all possible sources of uncertainty were considered (Figure 6), and they were grouped according to their relation and quantified. Then, all contributions were combined to have standard uncertainty (Eq. 6; Joint Committee for Guides in Metrology [JCGM], 2008).

where

uModel2 corresponds to the uncertainty contribution provided by the mathematical model (Eq. 2) of the measuring method and includes the uncertainty provided by the partitioning of the sample (λ factor), gravimetric dilution, and droplet volume (Eq. 7),

and uPrecision2, which corresponds to the precision uncertainty provided by the total variation of each concentration level, was defined as the highest degree of dispersion obtained between replicates srepeat2 and between different days sinterm2 (Eq. 8),

n: Days of measurement.

According to the validation data for all target sequences, the relative measurement uncertainty associated with the mathematical model varied between 1,6 and 9% (Table 7). The maximum contribution corresponded to λ factor (50–99%) at all levels (see Supplementary Figure S3). After combining the mathematical model and measurement uncertainty contributions (Eq. 6), the combined uncertainty was calculated for every target in every concentration level (Figure 7). The relative standard uncertainty value ranged between 2 and 12%, while its lower value was obtained at the concentration of 800 cp/μL in the ddPCR reaction.

The correlation between the targets invA-ttr and hilA-siiA was evaluated to determine the contribution source to the uncertainty associated with covariance between target sequences in duplex assays. Considering that all target sequences are single copy, the correlation was evaluated, keeping the primers concentration constant for one target sequence (600 nM), while this was varied for the other one as 0, 150, 300, 450, 600, 750, and 900 nM. In all cases, the probe concentration was 300 nM. The slope and correlation coefficients were calculated and their significance analyzed using the student’s t-test with a 95% confidence.

As the slopes were statistically equal to zero and the correlation were not significant (Figure 8 and Supplementary Figure S4) for any of the target sequences, then, there is not a covariance contribution to the combined uncertainty (Joint Committee for Guides in Metrology [JCGM], 2008).