Patients aged 12 years and older presenting with symptoms of salmonellosis, such as diarrhea, abdominal pain, or fever of unknown origin, were enrolled in the study. Participants were recruited from four primary health centers in the Mukuru kwa Njenga and Mukuru kwa Ruben informal settlements, as well as from Mama Lucy Kibaki Hospital, a Level-5 county referral facility in Nairobi, Kenya. The study was conducted between December 2020 and June 2023. Blood and stool samples were collected from participants, following their consent, for Salmonella culture and antimicrobial susceptibility testing (AST). Metadata, including the nearest landmark, were recorded and subsequently used to capture Global Positioning System (GPS) coordinates using the Epicollect5 data collection tool on a study mobile phone. To ensure optimal pathogen recovery, a sterile swab was used to aliquot a portion of each stool specimen into Cary-Blair transport medium. The specimens were then sealed in zip-lock bags and placed in a cooler box maintained at 2–8°C for transport to the laboratory at KEMRI.

Blood samples were collected for BACTEC Salmonella identification following standard procedures. The venipuncture site was disinfected using 70% alcohol swabs and allowed to stand for 30 s to minimize contamination. Using a sterile syringe, 8–10 mL of blood was drawn from the patients. The blood was immediately inoculated aseptically into BACTEC aerobic culture bottles and gently mixed to prevent clotting. Labeled samples were promptly transported to the laboratory at CMR, KEMRI, ensuring that room temperature was maintained.

The aliquoted stool sample on the cotton swab was inoculated into Selenite Feacal (SF) broth, an enrichment medium, and incubated aerobically at 37°C for 18–24 h. Following incubation, a loopful of the SF broth was streaked onto Xylose Lysine Deoxycholate (XLD) agar and incubated aerobically at 37°C for another 18–24 h. BACTEC vials containing blood samples were placed in the BACTEC blood culture system for incubation and detection of positive blood cultures. All positive blood cultures were subsequently streaked onto blood agar (BA) and XLD agar, followed by aerobic incubation at 37°C for 18–24 h. Non-lactose fermenting colonies, identified by their distinct brick-red appearance or brick-red colonies with black centers on XLD agar, were sub-cultured onto Tryptone Soya Agar (TSA), a non-selective medium. The colonies were characterized phenotypically using standard biochemical tests, including Triple Sugar Iron (TSI) agar, urea hydrolysis, and motility tests, and further confirmed using the Analytical Profile Index (API) system (Montalieu Vercieu, France).

Salmonella isolates were confirmed using the White-Kauffmann-Le Minor scheme (Grimont and Weil, 2007; Guibourdenche et al., 2010) with Salmonella polyvalent O antisera. Due to the unavailability of monovalent antisera in our laboratory for identifying less common Salmonella serovars, isolates that did not exhibit the characteristic antigenic profiles of S. Typhi (O:9, Vi:d:-), S. Typhimurium (4:i:1,2), or S. Enteritidis (9:g,m:-) were subjected to genomic DNA extraction, followed by whole genome sequencing for serovar identification and detection of antimicrobial resistance genes.

Antimicrobial susceptibility testing was performed using the disk diffusion method (Reller et al., 2009) for a range of antimicrobials commonly used to treat Salmonella infections, including ampicillin (10 μg), tetracycline (30 μg), co-trimoxazole (25 μg), chloramphenicol (30 μg), amoxicillin–clavulanate (20/10 μg), cefpodoxime (30 μg), ceftazidime (30 μg), ceftriaxone (30 μg), cefotaxime (30 μg), azithromycin (15 mg), ciprofloxacin (5 μg), nalidixic acid (10 μg), kanamycin (30 mg), and gentamicin (10 mg). The diameter of the zone of inhibition was measured after 18–24 h using a Vernier caliper, and results were interpreted in accordance with the Clinical and Laboratory Standards Institute (CLSI) guidelines for Salmonella (Performance Standards for Antimicrobial Susceptibility Testing (33rd ed.) (2023) (Clinical Laboratory Standards Institute., 2023).

DNA was extracted from Salmonella isolates using the GenElute™ Bacterial Genomic DNA Kit (Sigma-Aldrich, Missouri, USA) for whole genome sequencing (WGS), following the manufacturer's instructions. Library preparation for WGS was performed by SeqCoast Genomics (Portsmouth, New Hampshire, USA) using the Illumina DNA Prep Tagmentation Kit with unique dual indexes. Sequencing was conducted on the Illumina NextSeq 2000 platform with a 300-cycle flow cell kit, generating 2 × 150-bp paired-end reads, as previously described (Grant et al., 2023). A PhiX control (1%−2%) was included in the sequencing run to optimize base calling. Demultiplexing, read trimming, and run analytics were carried out using DRAGEN v3.10.12, an onboard analysis software of the NextSeq 2000.

Raw reads were quality-trimmed using Trimmomatic (version 0.39) (Bolger et al., 2014), followed by error correction using SPAdes (version 3.13.1) (Bankevich et al., 2012). Assembly into contigs was performed using SPAdes, wrapped in Unicycler (version 0.4.4) (Wick et al., 2017), with read mapping conducted using Bowtie2 (Langmead and Salzberg, 2012) and alignment processing utilizing SAMtools (Li et al., 2009). Assembly polishing was carried out with Pilon (version 1.24) (Walker et al., 2014), also wrapped in Unicycler. Gene prediction and functional annotation were performed using BAKTA (version 1.5.1) (Schwengers et al., 2021). The annotation pipeline was as follows: prediction of protein-coding genes using Prodigal, tRNA identification using tRNAscan-SE (Chan and Lowe, 2019), tRNA and tmRNA identification using Aragorn (Laslett and Canback, 2004), prediction of rRNA sequences using Infernal and the Rfam database (Nawrocki and Eddy, 2013), CRISPR prediction using PILER-CR (Edgar, 2007), antimicrobial resistance gene identification using AMRFinderPlus (Feldgarden et al., 2021), prediction of signal peptides using DeepSig (Savojardo et al., 2018), prediction of transposases using ISFinder (Siguier et al., 2006), and computation of codon usage biases for each amino acid using the codonUsage.py script (Garber, 2024) available at https://github.com/Arkadiy-Garber/BagOfTricks.

Enterobase Salmonella database, a web-based platform for genome analysis accessible at https://enterobase.warwick.ac.uk/species/index/senterica, was used for the identification of NTS serovars (Zhou et al., 2020). The platform utilizes WGS data to assign Salmonella strains to specific serovars through its integration with tools like SISTR (Salmonella in Silico Typing Resource) (Yoshida et al., 2016) and Multi-Locus Sequence Typing (Achtman et al., 2012). Enterobase combines the functionalities of both SISTR1 and SeqSero2 for in silico serotyping of Salmonella genomes (Zhou et al., 2020). WGS data in FASTA format were uploaded to the database for analysis. Results obtained for each strain included the serovar name, sequence type and predicted antigenic profile. Enterobase provided predictions for the expression of the O antigen, as well as the flagellar antigens H1 and H2, based on the analysis of the O antigen gene cluster and the fliC and fljB genes.

The Enterobase SNP (Single Nucleotide Polymorphism) phylogenetic tree web-based tool (Alikhan et al., 2018) was used to construct a phylogenetic tree for the Salmonella isolates based on whole genome sequences. This analysis aimed to investigate the genetic relationships and evolutionary patterns of the isolated Salmonella strains, alongside several additional strains from different regions with available whole genome sequences from the NCBI public database. Salmonella bongori (GenBank Accession: SAMN43895974) was included as the outgroup. The resulting phylogenetic tree was downloaded in Nerwick format and uploaded to Microreact (https://microreact.org/) for further visualization. Associated metadata was exported in CSV format to Microreact, and a Newick-format phylogenetic tree was generated, reflecting branch lengths that correspond to SNP/genetic distances.

Acquired antimicrobial resistance genes and known resistance-associated point mutations in the Salmonella-assembled nucleotide sequences were identified using NCBI's Antimicrobial Resistance Gene Finder Plus (AMRFinderPlus, v3.12.8) (Feldgarden et al., 2021), as outlined in the tool's documentation (https://github.com/ncbi/amr/wiki).

To identify plasmids carrying acquired antimicrobial resistance (AMR) genes in the genomic DNA of isolated Salmonella strains, the PLASMe plasmid detection tool (v1.1) was used (https://github.com/HubertTang/PLASMe). The tool employs an alignment component to detect closely related plasmids, while order-specific Transformer models predict more diverged plasmids (Tang et al., 2023). Assembled Salmonella genomes were used for this analysis.

Salmonella biofilms were grown on non-treated polystyrene 96-well plates (Corning, Kennebunkport, ME) as previously described (González et al., 2018). The biofilm formation was evaluated under four conditions: in the absence of both cholesterol and bile, in the presence of cholesterol but without bile, in the presence of bile but without cholesterol, and in the presence of both cholesterol and bile. Cholesterol was incorporated by precoating the wells with cholesterol solution (prepared by mixing 5 mg/mL cholesterol and 1:1 isopropanol:ethanol) which was then air-dried overnight (Crawford et al., 2008). A single colony of non-typhoidal Salmonella strain, grown on XLD agar, was inoculated into 3 mL of TSB and incubated overnight at 37°C with gentle rolling. The OD₆₀₀ of the overnight culture was measured using a NanoDrop spectrophotometer (TSB served as the blank) and normalized to OD₆₀₀ = 0.8 in 1:20 TSB. The normalized culture was then further diluted 1:10 in 1:20 TSB or 1:20 TSB containing 2.5% human bile, and 100 μL of the resulting suspension was added to each well of the microtiter plate (three wells per isolate). The plates were incubated at 25°C on a Fisherbrand™ nutating mixer (Thermo Fisher Scientific, Hampton, NH) set to 24 rpm for 24 h. After incubation, the plates were emptied and washed twice in a bucket of reverse osmosis (RO) water before heat-fixing at 60°C for 1 h. Biofilms were then stained with crystal violet solution (33.33% crystal violet, 60% 1xPBS, 33.33% methanol and 3.33% isopropanol), and bound crystal violet was eluted using 33% acetic acid. The OD₅₇₀ of the eluted solution was measured to quantify biofilm formation. Each strain was analyzed in triplicate, and the entire process was repeated at least three times for each of the four conditions.

GraphPad Prism 9.5 was used to analyze the biofilm formation data. One-way analysis of variance (ANOVA) was performed to assess the significance of differences in biofilm formation across the four different in vitro conditions. To assess differences between specific pairs of conditions, Student's t-test was applied. The following comparisons were made: absence of both cholesterol and bile vs. presence of cholesterol, and presence of bile vs. presence of both bile and cholesterol. A p-value of less than 0.05 (p < 0.05) was considered statistically significant.

Between December 2020 and June 2024, a total of 2,634 patients aged ≥12 years old were recruited for the study to identify Salmonella infections and detect asymptomatic carriage of S. Typhi (Muturi et al., 2024a). Among these, 25 (0.9% positivity rate) had NTS species other than S. Typhimurium and S. Enteritidis isolated from their stool or blood samples. A concurrent study by our group, conducted in the same setting and period, reported a 1.5% isolation rate of S. Enteritidis and S. Typhimurium among children under 5 years of age (Kering et al., 2024). Of the 25 patients, 8 were female, and 17 were male. The age distribution of the patients was as follows: 5 were under 20 years of age, 9 were aged 20–29 years, another 9 were in their 30 s, and 2 were over 40 years old. The patients exhibited a range of gastrointestinal and systemic symptoms, including abdominal pain, diarrhea, vomiting, headache, fever, abdominal distention, and cough (Supplementary Table 1). Among the isolates, 24 were recovered from stool samples, while only one was isolated from a blood sample. During the same period, S. Typhi infections were confirmed in 1.5% of patients aged ≥12 years old presenting with typical symptoms of typhoid fever (Muturi et al., 2024a). As illustrated in Figure 1, areas identified as hotspots for NTS infections coincided with hotspots for typhoid fever.

A total of 16 NTS serovars, rarely reported in this region, were isolated from samples collected from 25 patients. S. Newport was the most prevalent, from six participants (4 male and 2 Female), followed by S. Breda (3 patients; 2 female and 1 male), S. Eastbourne (2 cases: both male), and S. Orion (2 patients: both male) (Table 1). Each of the remaining 12 participants had a distinct Salmonella serovar isolated from their samples, representing 12 additional serovars (Figure 2). Notably, S. Newport isolates were classified into three sequence types: ST46, ST31, and ST166. These serovars displayed different antigenic profiles and were classified into 9 distinct serogroups (Table 2).

Resistance to the tested antimicrobials was observed in only 3 out of 25 Salmonella isolates. Two S. Newport strains, both classified as ST31, exhibited resistance to ampicillin and tetracycline. Additionally, one S. Newport strain (ST166) showed resistance to trimethoprim-sulfamethoxazole and tetracycline. Genomic analysis revealed that all three isolates carried acquired antimicrobial resistance (AMR) genes, but none exhibited known resistance-associated point mutations. The two S. Newport ST31 strains each carried the bla_TEM−1 and tet(A) resistance genes. The S. Newport ST166 strain carried multiple resistance genes, including tet(A), aph(6)-Id, dfrA14, aph(3′')-Ib, and sul2.

The two S. Newport ST31 isolates, KEMRI2023_Np50 and KEMRI2023_Np106, each carried two AMR genes on separate contigs. The beta-lactamase gene bla_TEM−1 was located on a contig with sequences closely resembling those of Shigella dysenteriae strain BU53M1′s plasmid (GenBank accession: NZ_CP024467.1). The tetracycline resistance gene tet(A) was found on a contig that shared high sequence similarity with the plasmid p65COLEC-2 from Escherichia coli strain 65COLEC (GenBank accession: NZ_CP070916.1). Meanwhile, S. Newport ST166 (strain KEMRI2023_Np96) contained all of its AMR genes (tet(A), aph(6)-Id, dfrA14, aph(3′')-Ib, sul2) on a single contig, which showed high sequence similarity to a multidrug-resistant (MDR) plasmid from Citrobacter freundii (plasmid CF_20.1, GenBank accession: NZ_MW115421.1) (Table 3).

The ability of the NTS isolates to form biofilms under various in vitro conditions showed significant statistical variation. Biofilm formation differed significantly across four tested conditions: (i) in the absence of both cholesterol and bile, (ii) in the presence of cholesterol but absence of bile, Figure 3A, (iii) in the presence of bile but without cholesterol and (iv) in the presence of both cholesterol and bile (p < 0.0001) (Figure 3B). Notably, the isolates formed significantly stronger biofilms in cholesterol-coated wells but failed to form biofilms when exposed to 2.5% human bile. Among the 25 strains tested, 12 exhibited strong biofilm formation (OD₅₇₀ > 1.0), 11 displayed moderate biofilm formation (OD₅₇₀ between 0.5 and 1.0), and only 2 produced weak biofilms (OD₅₇₀ < 0.5) on cholesterol-coated wells without bile. Serovars that formed strong biofilms included S. Breda (including strain KEMRI_2023Br19 isolated from blood sample), S. Hann, and S. Eastbourne, while S. Chicago and S. Kentucky were among those that formed weak biofilms.