CRISPR 2 PCR and high resolution melting profiling for identification and characterization of clinically-relevant Salmonella enterica subsp. enterica

Background Nontyphoidal Salmonella spp. constitute a major bacterial cause of food poisoning. Each Salmonella serotype causes distinct virulence to humans. Method A small cohort study was conducted to characterize several aspects of Salmonella isolates obtained from stool of diarrheal patients (n = 26) admitted to Phayao Ram Hospital, Phayao province, Thailand. A simple CRISPR 2 molecular analysis was developed to rapidly type Salmonella isolates employing both uniplex and high resolution melting (HRM) curve analysis. Results CRISPR 2 monoplex PCR generated a single Salmonella serotype-specific amplicon, showing S. 4,[5],12:i:- with highest frequency (42%), S. Enteritidis (15%) and S. Stanley (11%); S. Typhimurium was not detected. CRISPR 2 HRM-PCR allowed further classification of S. 4,[5],12:i:- isolates based on their specific CRISPR 2 signature sequences. The highest prevalence of Salmonella infection was during the summer season (April to August). Additional studies were conducted using standard multiplex HRM-PCR typing, which confirmed CRISPR 2 PCR results and, using a machine-learning algorithm, clustered the majority of Salmonella serotypes into six clades; repetitive element-based (ERIC) PCR, which clustered the serotypes into three clades only; antibiogram profiling, which revealed the majority resistant to ampicillin (69%); and test for extended spectrum β-lactamase production (two isolates) and PCR-based detection of bla alleles. Conclusion CRISPR 2 PCR provided a simple assay for detection and identification of clinically-relevant Salmonella serotypes. In conjunction with antibiogram profiling and rapid assay for β-lactamase producers, this approach should facilitate detection and appropriate treatment of Salmonellosis in a local hospital setting. In addition, CRISPR 2 HRM-PCR profiling enabled clustering of S. 4,[5],12:i:-isolates according to CRISPR 2 locus signature sequences, indicative of their different evolutionary trajectories, thereby providing a powerful tool for future epidemiological studies of virulent Salmonella serotypes.


INTRODUCTION
Salmonella enterica can cause human gastroenteritis owing to inadequate hygienic standards of living and/or consuming poorly prepared fresh food. S. enterica has been subtyped to more than 2,463 serovars according to their antigenic properties (Kauffman, 1972); however, only a limited number of serovars cause human infection (Uzzau et al., 2000) and, thus, identification of these serovars is of public health importance. Distribution of Salmonella serovars is markedly different among continents (Galanis et al., 2006), but within a single country such as Thailand, Salmonella serovars slightly varies in different geographical regions, the most common clinical serovars being S. 4,[5],12:i:-, S. Anatum, S. Derby, S. Enteritidis, and S. Weltevreden (Bangtrakulnonth et al., 2004;Padungtod & Kaneene, 2006).
Emergence of multi-drug resistance (MDR) in the major prevalent non-typhoidal Salmonella serovars found in human gastroenteritis is related to an acquisition of antibiotic-resistant genes especially in S. Typhimurium (Velge, Cloeckart & Barrow, 2005). In Thailand, a significant increase in resistance to amikacin, kanamycin and second line antibiotic ceftriaxone is suggested to be related to overuse of antibiotics to livestock feed and in medical treatment (Boonmar et al., 1998). Also, an increase in resistance to quinolones such as nalidixic acid was reported (Sirsichote et al., 2010). Failure of treatment from third-generation cephalosporins, the extended spectrum β-lactamases (ESBLs), has created a major threat to treatment of MDR microbial pathogens and has stimulated studies on their epidemiology (Rupp & Fey, 2003). ESBL genes, such as blaOXA, blaSHV, blaTEM and the more recent blaCTX-M, generate different levels of virulence and transmission capability (Ewers et al., 2012). ESBL β-lactemase CTX-M is of particular concern due to its global dissemination through both clonal and horizontal gene transfer (Cantón, González-Alba & Galán, 2012).
The standard method for Salmonella identification is based on a culture method (ISO, 2002). The assay uses different selective media, followed by serotyping based on various combinations of O and H antigenic determinants (Kauffman, 1972). These methods require skilled personnel to perform the standardized protocols, which are laborious and timeconsuming. In order to reduce the turnover time for typing Salmonella spp., multiplex PCR of specific gene determinants was introduced, which depends on sequence polymorphisms of rfb locus and flagellar alleles (Kim et al., 2006;Masek et al., 2014). Other molecular modifications such as high resolution melting temperature (HRM)-PCR assay has been coupled to multiplex PCR for detection of polymorphisms of 16S rDNA (O'Regan et al., 2008), fljB, gyrB and ycf Q (Zeinzinger et al., 2012). HRM-PCR was applied to discriminate between two most prevalent clinical serotypes, namely, S. Enteritidis and S. Typhimurium (Bratčikov & Mauricas, 2009).
Clustered regularly interspaced short palindromic repeats (CRISPR) were discovered as a new family of repeated DNA sequences in many prokaryotes (Jansen et al., 2002). Their genetic signatures are characterized by repeated patterns of DNA, known as direct repeats (DRs), 24-47 bp in length, and DNA variable sequences (spacers) of 21-72 bp (Shariat & Dudley, 2014). Adjacent to the CRISPR locus are a ''leader sequence'' and cas (CRISPR-associated sequence) gene (Horvath & Barrangou, 2010). Salmonella spp. harbor two different CRISPR loci, namely, CRISPR 1 and CRISPR 2 that correlate with Salmonella serotype and multilocus sequence type (Fabre et al., 2012). Characterization of CRISPR alleles provide information of spacer content for performing Salmonella typing and subtyping (Fabre et al., 2012;Liu et al., 2011). CRISPR typing was recently applied to identify virulent Salmonella infection of chicken raised in a single farm as well as in farms from different areas (Fei et al., 2017;Li et al., 2018). More recently, CRISPR loci polymorphisms were utilized in a single-step assay to identify multiple Salmonella spp. contamination of a poultry sample (Thompson et al., 2018).
In order to reduce operational cost and turnover time of the traditional culture assay for detecting and typing Salmonella spp., a simple CRISPR 2-based conventional and monoplex HRM-PCR was developed in conjunction with HRM triplex and a novel machine-learning algorithm tool to analyze and identify Salmonella serotypes in correlation with repetitive element-based (ERIC)-PCR genotyping and a rapid ESBL test,and,in addition,to distinguish S. 4,[5],12:i:-isolates from different epidemiological settings.

Isolation and identification of Salmonella isolates in stool of diarrheal patients
Stool samples were collected from patients (n = 26) with diarrhea during May 2016 to July 2017 at Phayao Ram Hospital, Phayao province, Thailand. Diarrhea is defined as patients showing related symptoms of fecal incontinence diagnosed by a physician and participants signed the consent forms after brief explanation of the research project. Samples were incubated in buffered peptone water (Oxoid, Hampshire, UK) at 37 • C overnight, plated on SS and XLD agar (Oxoid, Hampshire, UK), and suspected Salmonella colonies indicated by using a triple sugar iron (TSI) slant (Oxoid, Hampshire, UK) assay and lysine iron agar (LIA) (Biomedia, Nontanuri, Thailand). Conventional biochemical tests (glucose fermentation, lactose oxidation, gas and H 2 S production, urease assay, methyl red staining, and indole, motility and Voges-Proskauer tests) were performed according to ISO 6579:2002.
The study protocols were approved by the Ethical Committee of Phayao University (no. 57 02 04 0020).

Determination of β-lactam antibiotic resistance profile and ESBL phenotype
Susceptibility to β-lactam antibiotics was performed using a disk diffusion method according to Clinical and Laboratory Standards Institute (CLSI) guidelines (CLSI, 2016) using ampicillin (AMP, l0 µg), cefotaxime (CTX, 30 µg), ciprofloxacin (CIP, 5 µg), ertapenem (ETP, 10 µg), and gentamycin (GEN, 10 µg) discs (Oxoid, Hampshire, UK), with Escherichia coli ATCC 25922 as control. ESBL phenotype was evaluated using a doubledish method (CLSI, 2016) employing CTX alone and in combination with clavulanic acid (10 µg) (Oxoid, Hampshire, UK), with in-house known ESBL-producing and -negative Escherichia coli strains as controls. A rapid ESBL was slightly modified from the ESBL NDP (Nordmann, Dortet & Poirel, 2012), which is based on observation of color change due to acid production of ESBL-producing bacteria. Salmonella isolate was inoculated in one ml of NB broth (Oxoid, Hampshire, UK) and incubated at 37 • C overnight, each experiment conducted in duplicate. Then culture was centrifuged at 13,000 g for 2 min, pellet washed twice with 700 µL of 1X 10 mM Tris HCl pH 8.0 containing 1 mM EDTA (TE) buffer and mixed with 100 µL of lysis buffer [(B-PERII Bacterial Protein Extraction Reagent)] (Pierce/Thermo Scientific, Villebon-sur-Yvette, France), followed by 100 µL of revelation solution (0.05% (w/v) phenol red) with and without 6 g/ml cefotaxime and incubation at 37 • C for 10 min; yellow color in the former indicating positive result.

Determination of Salmonella serotypes by HRM-PCR
DNA was extracted from Salmonella isolates as previously described (McNerney et al., 2017). In brief, one ml aliquot of an overnight culture was sedimented as described above, washed twice with 400 µL of TE buffer, resuspended in the same volume of TE buffer, incubated at 80 • C for 20 min, and cooled to ambient temperature. Then, a 50 µL aliquot of lysozyme solution (10 mg/mL) was added and the solution incubated at 37 • C for one hour with occasionally shaking, followed by addition of 75 µL of 10% SDS/proteinase K (10 mg/mL) solution, vigorous vertexing and incubation at 65 • C for 10 min. Following addition of 100 µL of 5 M NaCl and 100 µL of prewarmed (65 • C) 10% N-cetyl-N,N,N,-trimethyl ammonium bromide (CTAB)/ 5 M NaCl solution, the mixture was further incubated at 65 • C for 10 min, followed by addition of 750 µL of chloroform : isoamyl alcohol (24 :1) and centrifugation at 11,000 g at 4 • C for 5 min. DNA in the upper aqueous was precipitated with ethanol, resuspended in 50 µL of double-distilled water and stored at −20 • C until used.
Multiplex HRM-PCR was performed using a combination of primers to amplify fljB (170 bp), gyrB (171 bp) and ycf Q (241 bp) (Table 1). HRM-PCR mixture (10 µL) contained 1 µL of DNA, 0.1 pmol of gyrB, 0.075 pmol of fljB and 0.075 pmol of ycfQ primer pairs and 2 µL of HOT FIREPol EvaGreen: no ROX Mix (Solis Biodye, Tartu, Estonia). Thermocycling was performed in a BIO-RAD CFX96 TM Real-Time System (Bio-Rad, Hercules, CA, USA) as follows: 95 • C for 12 min, followed by 40 cycles of 95 • C for 10 s, 60 • C for 10 s and 72 • C for 20 s. Samples were then heated at 95 • C for 1 min, cooled to 40 • C for 1 min and then heated from 70 to 95 • C at 0.2 /s, with 25 fluorescence data acquisitions/ • C. HRM profiles were generated using a Precision Melt Analysis software V 1.2 (BIO-RAD, Hercules, CA, USA) with sensitivity setting at 0.30, temperature shift at threshold 5, pre-melt normalization range from 80.87 to 81.51 • C, and post-melt normalization range from 87.17 to 87.92 • C. Following normalizing and temperature shifting, difference plots were generated relative to HRM profile of S. Bareilly (as baseline).

Hierarchical clustering of HRM curves using dynamic time warping (DTW) algorithm
A dendrogram of normalized HRM curves was constructed using a DTW algorithm to determine distance measurements (Lu et al., 2017); the entire dendrogram construction was performed in Python programming language (Van Rossum, 1995). In short, a smooth spline approximation was determined from each normalized HRM curve using cubic splines of splrep function in scipy module (Jones, Oliphant & Peterson, 2001) followed by a rate curve calculated from negative first derivative of the resulting spline employing a splev function. The curve was then z-normalized using a zscore function to calculate DTW distances, in which lˆ2-norm is the distance function (only between 80 and 94 • C). Hierarchical clustering based on DTW distances was performed based on a neighbor-joining method to generate a dendrogram.

Analysis of ERIC PCR amplicon profile and phylogenetic tree
Amplicon patterns generated by ERIC-PCR were analyzed and employing a curve-based algorithm (Pearson correlation) together with an open-source GelJ software (Heras et al., 2015) to create a similarity scale, and analysis of constructed phylogenetic tree clusters was performed by an unweighted pair-group using arithmetic averages algorithm (UPGMA).

Characterization of CRISPR 2 amplicon sequences
CRISPR 2 DNA sequences in fasta format were uploaded to https://crisprcas.i2bc.parissaclay.fr/CrisprCasFinder/Index with default setting. Information of CRISPR 2 region (molecular signatures of CRISPR 2 locus), e.g., DR consensus sequence, DR length and number of spacers, were provided directly in the form of DNA sequences or numeral data.

Identification of bla alleles
Reaction mixture (10 µL) for multiplex PCR of bla alleles of interest contained 1 µL of DNA, bla primer sets (IDT) ( Table 1) and 2 µL of HOT FIREPol Blend Master Mix Plus 10 mM MgCl 2 (Solis Biodye) and thermocycling was conducted as described above but using the following conditions: 95 • C for 12 min; 30 cycles of 95 • C for 40 s, 60 • C for 40 s and 72 • C for 60 s; and a final step at 72 • C for 7 min. Amplicons were analyzed (using 1.5% agarose gel) and recorded as described above.

Phylogenic tree based on multiplex HRM-PCR of clinical Salmonella isolates and association with other characteristics
A DTW algorithm was applied to transform HRM-PCR profiles into numerical data to allow construction of a phylogenetic tree (Lu et al., 2017), resulting in grouping of the 26 clinical isolates into six clades (Fig. 3). S. 4,[5],12:i:-, S. Enteritidis and S. Stanley was clustered in clade 2, 4 and 6, respectively while HRM types of unidentified serotypes were mainly distributed in clade 3 and 5 except HRM_5 (S. Weltevreden) and isolate 412 (HRM_4,S. 4,[5],12:i:-) was in clade 1 and 3 respectively. It is worth noting that S. 4,[5],12:i:-isolates were clustered in clades 2 (n = 8) and 3 (n = 1), but three were unassigned. Ampicillin resistance was predominantly associated with blaTEM, and two ESBL-producing Salmonella isolates (based on both double-dish and the rapid ESBL assay) harbored blaCTX group 1. One ESBL-producing S. 4,[5],12:i:-, isolate 412 (assigned by ERIC-PCR, Fig. 2) demonstrated a multidrug-resistant phenotype, including CIP resistance.    Table 2 HRM temperature-shifted difference profiles (relative to S. 4,[5],12:i:-isolate 76) of CRISPR 2 uniplex amplicons of the 11 S. 4,[5],12:i:-isolates from this study and other previously identified samples produced three different profiles (clusters 1-3) (Fig. 5). Cluster 1 contained five known human and four from minced pork samples, cluster 2 four isolates from this study and three known human isolates and cluster 3 seven isolates from this study (including Salmonella isolate 345 and 392 that were not grouped in any of the six clades inferred from multiplex HRM PCR profiling (Fig. 3) and Salmonella isolate 412) and one known human isolate ( Table 2). The three ESBL-producing Salmonella clinical isolates in cluster 1 carried a novel variant DR consensus sequence (Fabre et al., 2012).

Seasonal prevalence of Salmonella serotypes collected from stools of patients, Phayao Ram Hospital
Over a period of one calendar year, infection at Phayao Ram Hospital of Salmonella serotypes identified using multiplex HRM-PCR profiling peaked during the summer period (April to August) (Fig. 6), with, as expected, S. 4,[5],12:i:-being predominant, but all serotypes were represented. Although minor peaks were discernable, they are not of significance owing to the limited number of Salmonella isolates examined.

DISCUSSION
In Thailand, Salmonella spp. is the leading cause of food poisoning with approximately 359-389/100000 of the population affected (National Disease Surveiilance, 2017). In Phayao province, the estimated number of cases of food poisoning increased from 143 to 288/100,000 between 2015 to 2016, but with no reported mortality (National Antimicrobial Resistance Surveillance Center, 2017). Salmonella spp. isolated from stool of patients, especially from infants, are consistent among studies irrespective of geographical locations (Pulsrikarn et al., 2013;Parry & Threlfall, 2008). There were 105 cases admitted to Phayao Hospital from Salmonella food poisoning from January to December 2014 (Phayao Hospital, 2015). Multiplex HRM-PCR assay of Salmonella serotypes, although not able to type all eight serotypes responsible for food poisoning, is adequate in identifying the majority of pertinent clinical Salmonella serotypes in particular S. 4,[5],12:i:-and S. Enteritidis that are frequently identified as the serotypes responsible for human and veterinary infections worldwide (Frasson et al., 2016). The method is less time consuming, with a turn-around time of approximately 8-12 h, compared to 3-5 days using standard conventional technique 2 HRM-PCR profiling in our hands indicated S. 4,[5],12:i:-isolates could be readily differentiated, possibly reflecting differences in evolutionary trajectories of the S. 4,[5],12:i:isolates from this region of the country.
The limited number of S. 4,[5],12:i:-isolates available in this particular area necessitates further studies of a larger cohort of S. 4,[5],12:i:-isolates not only from the study area but from various epidemiology backgrounds to enable correlation of important virulent features with epidemiological settings. Interestingly, application of CRISPR 1 and 2 PCR methods for Salmonella identification allowed detection of multiple Salmonella serovars in a single sample (Thompson et al., 2018). This application of CRISPR 2 PCR for identification of multiple Salmonella serotypes in a specimen will be further explored.

CONCLUSIONS
The study demonstrates the ability of applying multiplex HRM-PCR profiling and a novel protocol of CRISPR 2 uniplex PCR together with HRM profiling to rapidly and conveniently type clinically important Salmonella isolates. In conjunction with a rapid ESBL test, antibiogram profile and detection of β-lactamase genes should provide a powerful strategy for predicting antibiotic resistance phenotypes associated with particular Salmonella serotypes. Moreover, CRISPR 2 HRM-PCR profiling provides a novel tool for classification of S. 4,[5],12:i:-according to evolutionary trajectories stemming from various epidemiological settings.