Salmonella is one of the main causes of bacterial gastroenteritis worldwide. To date, more than 2,600 Salmonella serotypes have been reported globally (Graziani et al., 2017). Moreover, approximately one-tenth of individuals worldwide are infected by non-typhoidal Salmonella enterica isolates annually resulting in 33 million deaths [Salmonella (non-typhoidal), 2018]. However, among Salmonella serotypes, only a limited number causes human infection (Uzzau et al., 2000), such as Salmonella enterica serotype Enteritidis (S. Enteritidis) and Salmonella enterica serotype Typhimurium (S. Typhimurium); identification of these infective serotypes is important for public health. Salmonella is widespread in domestic and wild animals, particularly poultry and poultry products which have been identified as important sources of human salmonellosis (Sanchez et al., 2002; Antunes et al., 2016; McWhorter and Chousalkar, 2019). For example, in 2018, approximately 207 million eggs from Rose Acre Farms (Seymour, Indiana, United States) were contaminated with Salmonella enterica serotype Braenderup, infecting 45 people, and were recalled across the United States (Centers for Disease Control and Prevention, 2018a). Similarly, eggs from Gravel Ridge Farms (Cullman, Alabama, United States) contaminated with S. Enteritidis, infected 44 people and had to be recalled (Centers for Disease Control and Prevention, 2018b), and a raw frozen chicken product contaminated with multidrug-resistant Salmonella enterica serotype Infantis (S. Infantis) infected 129 people (Centers for Disease Control and Prevention, 2019). Currently, the most commonly used treatment method for Salmonella infection is antibiotic therapy. However, the unnecessary long-term use of antibiotics has led to the development of Salmonella resistance, which poses a major challenge to the treatment of salmonellosis. In particular, there are many serotypes of Salmonella, how to quickly type the strain and trace the source of salmonellosis outbreaks remains to be resolved.

Different molecular typing techniques, such as pulse-field gel electrophoresis (PFGE), multi-locus sequence typing (MLST), and multi-locus variable number tandem repeat (VNTR) analysis, have been developed to track the origins of bacterial disease. PFGE typing has a high resolution for analyzing the homology of pathogenic bacteria from different sources, tracing the origin of foodborne disease outbreaks, determining the diversity of pathogenic bacteria in food, and evaluating the relationships between pathogenic bacteria in food and the related environment (Pang et al., 2007; Best et al., 2009). However, there are no uniform standards for naming, making it difficult to share data and to distinguish among the long-term evolution process from small changes between different strains of the same clone. Although MLST technology compensates for the deficiency of PFGE, it has some limitations for the application of typing in some highly homologous pathogens, due to the conservation of housekeeping genes (Turki et al., 2014; Campioni et al., 2015).

In recent years, innovative and powerful typing methods based on clustered regularly interspaced short palindromic repeat (CRISPR) loci have been developed (Barrangou and Dudley, 2016) and applied to the typing of a variety of bacteria, such as Salmonella (Fabre et al., 2012) and Escherichia coli (Long et al., 2020). Through the study of S. Heidelberg typing, it is found that CRISPR typing can be useful for Salmonella foodborne outbreaks subtyping as well as serve as a complimentary tool to determine source attribution in foodborne outbreaks (Vincent et al., 2018; Yousfi et al., 2020). Researchers combined the multi-virulence gene locus sequence typing (MVLST) scheme with CRISPR typing to develop a CRISPR-MVLST method, which was used to genotype 171 Salmonella strains from nine serotypes. The results of this method were better than CRISPR and MLST and could be used as an important genotyping method in Salmonella outbreaks (Liu et al., 2011). Upon comparing the effects of PFGE and CRISPR-MVLST on typing 84 strains of Salmonella enterica serotype Newport, researchers found that the Simpson index D values of the two methods were both higher than 0.95, suggesting complementary with each other in practice (Shariat et al., 2013b). A new typing method, CRISPR locus spacer pair typing (CLSPT), analyzes only the two newly incorporated spacers adjoining the leader sequence in the two CRISPR loci. Use of this method to analyze 82 Salmonella strains of 21 serotypes showed that although this method was weaker than conventional CRISPR typing (CCT) and PFGE typing, it was better than MLST and did not require analysis of the full CRISPR sequence map; it was also low cost and more practical, requiring only simple operation procedures (Li et al., 2014). Sequencing of 156 Salmonella strains showed that the evolutionary pattern reflected by the CRISPR locus was different from that of phylogeny, and horizontal gene transfer and changes in the shared environment could affect the phylogenetic distribution of Salmonella (Timme et al., 2013). Among these above methods, it is either cumbersome or ignoring the differences in spacer sequences. Thus in order to make up for this shortcoming, a novel method should be developed.

Since the discovery of the CRISPR locus in the 1980s, the functions and applications of CRISPR in prokaryotes have attracted much attention. For example, its spacer sequence has been used as an auxiliary tool for a variety of bacterial typing studies, epidemiological investigations, and evolutionary analyses. The CRISPR array has a high degree of polymorphism, which includes the unique biological and geographical characteristics of bacteria. However, few studies have reported its relationship with bacterial drug resistance, and the results of these studies in different bacteria have been inconsistent. For example, in Klebsiella pneumonia, Neisseria meningitidis, and Vibrio cholera, partly antibiotic resistance genes (ARGs) were positively associated with CRISPR-Cas system (Shehreen et al., 2019). But it is interesting that on the time scale of evolution, CRISPR-Cas has no significant effect on horizontal gene transfers (Gophna et al., 2015). It also has proved that the CRISPR-Cas system in E. coli does not affect the spread of plasmid and antibiotic (Touchon et al., 2012). And Shigella strains could regulate the activity of the CRISPR-Cas system by the insertion sequence (IS) elements that were identified in cas genes and then acquired exogenous resistance genes (Chen et al., 2019). In Salmonella, genome analysis has suggested that generally there are two CRISPR loci and a single type of CRISPR-Cas system, type I (mainly I–E) (Jansen et al., 2002; Grissa et al., 2007). Therefore, in this study, we aimed to analyze the occurrence and diversity of CRISPR arrays using whole-genome sequencing (WGS) data from 75 Salmonella isolates. Additionally, we established a new typing method based on CRISPR for Salmonella, and we further explored the relationships between the CRISPR-Cas system and drug resistance in Salmonella.

Recent reports have shown that the distribution of CRISPR spacer sequences of Salmonella is related to the serotype and subtype (Fabre et al., 2012). A CLSPT method based on the newly incorporated spacer for Salmonella enterica has been constructed; this is a simplified CRISPR-based typing method with higher consistency compared with traditional serotyping (Li et al., 2014). CRISPR combined with multi-site virulence sequence typing (CRISPR-MVLST) has been established in Salmonella and has proven to be a very effective typing technique in some serotypes (Shariat et al., 2013b, 2013a). Two CRISPR spacer sequences of Salmonella were combined with two virulence genes (fimH and sseL) and MLST to type Salmonella, and the results showed that the new genotyping method could improve the resolution compared with PFGE (Liu et al., 2011). However, because the above methods ignored differences in other spacer sequences, they showed weaker typing effects. In order to make up for this shortcoming, we developed a new method to distinguish Salmonella isolates in this work. Compared with the other subtyping methods (MLST, CCT, and CLSPT), the discrimination of CLTSST (86.46%) was weaker than that of CCT (87.75%), but stronger than those of MLST (85.66%) and CLSPT (85.52%). In these CRISPR-based typing methods, the CCT needed to analyze all spacer sequences in the strain. And the CLTSST only needed to know part of the spacer sequences, which not only had good typing effect, but also greatly reduced the cost. Therefore, the CLTSST could be used as the best method for typing different serotypes. The CLSPT was a simplified version of CCT. It only needed to analyze the newly inserted spacer sequences but ignored the first three spacer sequences that best reflected the original information of the strain and the CLTSST just made up for this deficiency. However, the CLTSST still had some shortcomings. For example, when there was no CRISPR in bacteria, this method could not be used. At the same time, this method still ignored part of the spacer sequence, so this method still needed to be further improved.

In this study, the lengths of different CRISPR arrays varied greatly, with the number of spacers ranging from 4 to 34. In some strains, spacers appeared repeatedly in multiple ways. The repetitive distribution of the spacer might be related to the sequence recombination of the repeat, and the recombination process might be accompanied by an increase and deletion of the spacer, as well as the position change of the spacer. This R-S recombination update could improve the adaptability of the strain. The structural features and sequence functions of Salmonella CRISPR were also explored. The DR sequences were different, and part of the spacer sequences was derived from foreign genes. We speculated that under antibiotic selection pressure, the bacterial CRISPR sequence may mutate, allowing the bacteria to acquire exogenous resistant genes and survive. This suggested that the variation in CRISPR DR sequences and the diversity of spacer sequences might be related to the resistance of Salmonella. The relationships between the number of CRISPR spacers and antibiotic resistance differed according to the type of antibiotic and the species of the strain. For example, the number of CRISPR spacers was positively correlated with SXT resistance but not correlated with AMP and cefazolin resistance in uropathogenic E. coli (Dang et al., 2013). By analyzing the relationships between the CRISPR-Cas system of 263 strains of E. coli and drug resistance, researchers found that there were no significant differences in the distributions of cas between sensitive and resistant strains and that drug-resistant plasmids could spread among CRISPR-positive E. coli (Touchon et al., 2012). However, the distribution of Enterococcus faecalis CRISPR was negatively correlated with bacterial resistance (Palmer and Gilmore, 2010). In this study, there were no close correlations between the CRISPR-Cas system and antibiotics in Salmonella isolates. But the aac(6')-Iaa was close positive correlation with the CRISPR1 or CRISPR2 and was close negative correlation with CRISPR1, CRISPR1 and CRISPR2. The bla_TEM-1B was close positive correlation with the CRISPR1 and was close negative correlation with CRISPR1 and CRISPR2. To date, no unified statement of the relationships between bacterial CRISPR and bacterial resistance to antibiotics has been proposed. This may be because of variations related to factors, such as the bacterial species, antibiotic species, host, strain isolation time, and geographic location. Therefore, further studies are needed to assess the relationships between bacterial CRISPR and antibiotic resistance.

In summary, we identified 517 unique spacers and 31 unique direct repeats in the CRISPR loci of these isolates. Further analysis of the identified spacers and repeats demonstrated specific patterns that harbored certain associations with genotype. We also established the CLTSST method and applied this method to the typing of different serotypes of Salmonella to explore the structure and function of CRISPR. Simultaneously, compared with the other three typing methods, we found that the discrimination of CLTSST was weaker than that of CCT, but stronger than those of MLST and CLSPT. In addition, we analyzed the relationships between CRISPR loci and Salmonella resistance and found that there were no close correlations between CRISPR loci and antibiotics but had close correlations between CRISPR loci and ARGs in Salmonella isolates.

The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found at: https://www.ncbi.nlm.nih.gov/, SAMN19819839, SAMN19819840, SAMN19819841, SAMN19819842, SAMN19819843, SAMN19819844, SAMN19819845, SAMN19819846, SAMN19819847, SAMN19819848, SAMN19819849, SAMN19819850, SAMN19819851, SAMN19819852, SAMN19819853, SAMN19819854, SAMN19819855, SAMN19819856, SAMN19819857, SAMN19819858, SAMN19819859, SAMN19819860, SAMN19819861, SAMN19819862, SAMN19819863, SAMN19819864, SAMN19819865, SAMN19819866, SAMN19819867, SAMN19819868, SAMN19819869, SAMN19820037, SAMN19820038, SAMN19820039, SAMN19820040, SAMN19820041, SAMN19819984, SAMN19819985, SAMN19819986, SAMN19819987, SAMN19819988, SAMN19819989, SAMN19819990, SAMN19819991, SAMN19819992, SAMN19819993, SAMN19819994, SAMN19819995, SAMN19819996, SAMN19819997, SAMN19819998, SAMN19819999, SAMN19820000, SAMN19820001, SAMN19820002, SAMN19820003, SAMN19820004, SAMN19820005, SAMN19820006, SAMN19820007, SAMN19820008, SAMN19820009, SAMN19820010, SAMN19812804, SAMN19812805, SAMN19812806, SAMN19812807, SAMN19812808, SAMN19812809, SAMN19812810, SAMN19812811, SAMN19812510, SAMN19812511, SAMN19812512, SAMN19812513.

HW and CuL designed the study and supervised the work. YW, YG, ChL, and BM participated, coordinated, and analyzed the data. CuL wrote the manuscript. All authors approved the final manuscript.

This work was supported by the National Natural Science Foundation of China (31830098 and 31772769), the China Agriculture Research System National System for Layer Production Technology (CARS-40-K14), and the Fundamental Research Funds for the Central Universities (SCU2021D006).

The authors declare that they have no competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.