Soil samples were collected in January 2021 from 100 points within each of five sites, each 35 m², around the Ashanti region of Ghana as previously described and shown in Figure 1 (Oduro et al., 2022). Sampling environment and sampling site details are presented in Figure 2F and Supplementary Table 2, respectively. Soils were enriched for B. pseudomallei using a combination of methods previously described. The enriched samples were preserved by combining 1.2 ml of the resulting cultures with 0.3 ml of 80% sterile glycerol and frozen at −80°C for shipment and storage. Frozen cultures were thawed and used directly to prepare crude DNA extracts for screening by PCR. PCR-positive samples were colony purified on Ashdown’s agar and pure cultures of suspected B. pseudomallei were utilized for DNA isolation and phenotypic characterization.

Briefly, crude DNA extracts were prepared by lysing 100 μl of frozen stocks at 95°C for ten minutes. Cell debris was pelleted by centrifugation and 2 μl of the resulting supernatant was used as template in the PCR reaction as previously described (Novak et al., 2006). Each extract was analyzed in duplicate, and each reaction plate contained positive controls including B. pseudomallei strain Bp82 crude extract and purified Bp82 genomic DNA, as well as negative controls including B. thailandensis strain E264 crude extract and water. To identify extracts containing B. pseudomallei genomic DNA, cutoff values were generated by plotting the results of true positive samples versus true negative samples using Graphpad Prism (version 9.0, Boston, Massachusetts USA).¹ Frozen stocks from sites determined to be positive by PCR were used to inoculate Ashdown’s agar plates which were incubated at 35°C for 96 h. Colonies presenting a morphology consistent with B. pseudomallei (i.e., flat, wrinkled, purple colonies) were selected for genetic characterization.

Suspected B. pseudomallei colonies were inoculated into 5 ml tryptic soy broth (TSB) and incubated at 37°C overnight while shaking. The following day, 1 ml was withdrawn, the cells were pelleted, and genomic DNA was isolated using the MasterPure Complete DNA and RNA Purification Kit (LGC Biosearch Technologies; Middleton, WI) according to the manufacturer’s instructions. Purified genomic DNA was quantified using Qubit dsDNA BR assays (ThermoFisher Scientific; Waltham, MA, USA) and a Qubit 4 fluorometer. DNA integrity was evaluated by electrophoresis using Genomic DNA ScreenTape, genomic reagents and a TapeStation 4150 (Agilent Technologies; Santa Clara, CA, USA).

Each sampling site of 35 m² was divided into a 10 × 10 grid of 3.5 m² (Oduro et al., 2022). To select colonies that represented each site, we divided each site into four quadrants of 25 sampling points. Due to the uneven distribution of PCR-positive sites in sampling site C (Figure 2), the upper left quadrant was reduced to 20 points and the lower left quadrant was expanded to 30 points (Figure 2). The highest quality DNA sample from each quadrant was selected for sequencing. If a given quadrant had only one sample, then that sample was chosen regardless of DNA concentration. Specific locations are indicated with a star in the grids of Figure 2.

Short-reads shotgun libraries were prepared using NEBNext Ultra II FS DNA Library Prep Kit for Illumina (New England Biolabs; Ipswich, MA, USA) following the manufacturer’s instructions. Briefly, 26 μl of DNA at 2–4 ng/μl was first fragmented enzymatically via 20 min incubation at 37°C. Next, hairpin sequencing adaptors, containing 5′-dT overhangs and a U ribonucleotide in the hairpin loop, were added to the fragmented DNA by ligation, and subsequently cleaved at the U sites. The libraries were then amplified and indexed by PCR using NEBNext Unique Dual Indexes. Prior to sequencing, the libraries were evaluated for quality using Agilent D1000 kit (Agilent Technologies; Santa Clara, CA, USA). The libraries that passed quality control were then quantified using Qubit dsDNA BR assay (ThermoFisher Scientific; Waltham, MA), pooled, and sequenced using a NovaSeq 6000 S4 Reagent Kit, v1.5 300 cycles, and a NovaSeq 6000 sequencer (Illumina; San Diego, CA, USA). Long-reads libraries were prepared using Ligation kits with native barcodes (Oxford Nanopore Technologies, Oxford, UK) following the manufacturer’s instructions. Briefly, 8 μl of DNA at ∼50 ng/μl was first polished for ligation using NEBNext FFPE DNA Repair Mix. The end-prepped DNA was then indexed by incubation with ONT Native Barcodes and Blunt/TA Ligase Master Mix (NEB). The indexed libraries were then pooled and ligated to ONT AMXII sequencing adaptors using NEBNext Quick T4 DNA ligase. Free adapters and short library fragments were eliminated by cleaning with AMPure beads (Beckman Coulter; Brea, CA) and ONT Long Fragment Buffer. Finally, the libraries were sequenced using a MinION_Flow Cell (R9.4.1) and a MinION-MK1C sequencer (Oxford Nanopore Technologies, Oxford, UK).

The NovaSeq data were processed using MetaDetector as previously described (Adhikari et al., 2024). Briefly, via this pipeline, reads were trimmed and filtered using BBDuk (v38.96) (Bushnell, 2014) and the resulting data were assembled using SPAdes (v.3.15.3) (Bankevich et al., 2012). The reads were mapped back to the contigs using BBMap (v38.96) (Bushnell, 2014), and the reads and contigs were classified using DIAMOND BLAST against NCBI’s non-redundant protein database (nr accessed February 10, 2023) (Buchfink et al., 2015). The ONT reads were assembled using dragonflye (v1.0.14) (Kolmogorov et al., 2019). In addition to these two methods, trimmed and filtered NovaSeq reads were further subsampled to 20 million (M) reads using seqtk (v1.3-r106) and combined with ONT reads for a hybrid assembly using UniCycler (v0.5.0) (Wick et al., 2017).

The most cohesive UniCycler assemblies were selected from each site (except for SiteE where Drangonflye assembly was used) and manually extracted using Bandage (v0.9.0) (Wick et al., 2015). The extracted genomes were validated using Qiagen CLC Genomics Workbench (v23.0.2)² by mapping the subsampled 20M NovaSeq reads and Oxford Nanopore Technologies (ONT) reads against the draft genome and using the Analyze Contigs function to identify problematic regions and Basic Variant Detection to identify regions with differences in the reads and contigs (CLC Microbial Genomics Module 23.0). The problematic regions and differences were then manually resolved. This process was repeated until there were no problematic regions or differences identified, resulting in a high-quality draft genome.

Genome annotation was performed using EDGE Bioinformatics (v2.4.0 BDRD 2023FEB09) (Li et al., 2017) with Prokka (v1.14.0) (Seemann, 2014). Virulence factors were identified using EDGE with ShortBRED (v0.9.4M) (Kaminski et al., 2015), antibiotic resistance genes were identified using EDGE with Resistance Gene Identifier (RGI v5.0.1; database v3.0.7) (Jia et al., 2017), and insertion sequences were identified using BLAST against the ISFinder database (accessed 2023-05-04) (Siguier et al., 2006). The MLST database for B. pseudomallei was downloaded via CLC on March 28, 2023, and May 04, 2023. CLC Type with MLST Scheme function was used with the SPAdes contigs, and the high-quality draft genomes, to identify sequence types for each of the samples. CLC Genomics Workbench was used to pull the core genome multilocus sequence typing (cgMLST) database and perform the cgMLST to determine a possible cgST. The cgMLST database has 4,071 alleles and requires a minimum fraction of 0.50 to have confidence in typing.

A total of 1,777 complete assemblies were downloaded from GenBank and nine assemblies associated with Africa were downloaded from PubMLST on May 04, 2023. Snippy (v4.6.0) (Seemann, 2015) was used to generate an alignment of the core single nucleotide polymorphism (SNP) genomes using B. pseudomallei Mahidol-1106a (Assembly Accession GCA_000756125.1) from Thailand as the reference. Using the core SNP alignment, IQ-Tree (v1.6.10) (Nguyen et al., 2015) was used to generate a Maximum Likelihood tree with automatic model testing using TVM+F+ASC+G4 and 1,000 bootstrap and BEAST (v1.10.4) was used to generate a Bayesian tree using HKY model and 1M MCMC chains. Mauve (v2.4.0) (Darling et al., 2004) was used to perform a whole genome alignment of the three high-quality draft genomes and reference B. pseudomallei Mahidol-1106a.

Phenotypic characterization such as enzymatic activity and carbohydrate utilization were evaluated using the Analytical Profile Index system, specifically API 20NE (BioMerieux; Cambridge, MA, USA). Antibiotic sensitivity testing (AST) using the disk diffusion method and detection of B. pseudomallei capsular polysaccharide by the Active Melioidosis Detect (InBios International Inc, Seatle, WA, USA) lateral flow immunoassay were conducted as previously described (Oduro et al., 2022).

Soil cultures were generated from a soil sampling expedition in Ghana using the selective enrichment process described by Trinh and colleagues (Trinh et al., 2019). We first sought to screen these cultures to down-select from our 500 soil samples, using PCR-specific for B. pseudomallei. Real time PCR targeting orf2 of the B. pseudomallei Type III Secretion System is considered 100% specific for B. pseudomallei because this region is not present in any near neighbor species (Novak et al., 2006). Ct values of true positive results derived from crude extracts of B. pseudomallei strain Bp82, prepared as described in the “Materials and methods” section, were plotted against the negative values to generate a Receiver Operator Characteristic (ROC) curve to identify a cutoff value. As depicted in Supplementary Figure 1, the area under the curve (AUC) was 1.0 [95% Confidence Interval (CI) 1.0 to 1.0]. The ROC curve identified a PCR cutoff value for 100% sensitivity and 100% specificity (Table 1). By applying a cutoff value of >30.47 to DNA extracts from each of the 100 points across each site, we found the positive rates as follows: Site A, 45%; Site B, 56%; Site C, 43%; Site D, 70% and Site E, 60%. These results confirm our previous observation that B. pseudomallei is ubiquitous throughout the Ashanti region of Ghana, although the distribution is not even (Figure 2 and Supplementary Table 2) (Oduro et al., 2022).

We performed whole-genome sequencing using both short read and long read platforms on 21 isolates from the five sampling sites around the Ashanti Region of Ghana (Figures 1, 2) with de novo assembly of the resulting sequencing data (Supplementary Table 3). Each genome from the same site was found to be very similar (i.e., a minimum of 97.92% nucleotide identity percentage); therefore, one representative genome from each site was deemed to be sufficient rather than producing high-quality drafts of duplicate genomes. To that end, five genomes from four sampling sites including one each from SiteA_10J, SiteC_5E, SiteD_1A, SiteD_10J, and SiteE_6D were manually closed to high quality draft genome status (Supplementary Table 4). An isolate from Site B was not chosen due to a high degree of sequence similarity to isolates from Site A.

Isolates from three of the sampling sites represented previously identified MLST Sequence Types (ST). ST930 was found in site A and site E and is represented by isolates found in four countries including the soil of Nigeria, located to the East of Ghana (Savelkoel et al., 2023). ST1749 from sampling site C is represented by a single isolate from Mexico.³ The genome of isolates from sampling site D was found to have a novel sequence type, now designated ST2058 (Table 2).

Phylogenetic analyses showed that the genomes from Ghana grouped within a clade associated with isolates from the Americas. The isolates from site A (ST930) and the two isolates designated with the new strain type ST2058 grouped within a sub-clade that includes isolates from Ghana-neighboring Burkina Faso, whereas the isolate with strain type ST1749 from Site C grouped closely with isolates from Aruba (Figure 3). We also observed that all genomes obtained from these sites contained the Yersinia-like fimbrial (YLF) (also referred to as fimbria/pilus outer membrane usher protein) gene. The presence of YLF is in agreement with a 2014 study by Gee et al. (2017) that established that YLF gene is associated mainly with isolates from locations other than Australia.

We conducted phenotypic characterization, including enzymatic activity, carbon utilization, capsular polysaccharide (CPS) detection and AST analyses of representative isolates of each of the STs including Site C, point 5E (C5E) representing ST1749, Site E point 2C (E2C) representing ST930 found in sites A and E. We selected two isolates representing our new ST from sampling site D (D1A and D10J). These isolates were from the farthest points on the sampling grid (Figure 2) and were selected due to minor differences in their genome sequences that could have translated into phenotypic variations. Each isolate was oxidase positive, was CPS positive, and produced an API 20NE profile of 1156577, one of the most common B. pseudomallei profiles (Table 3) (Amornchai et al., 2007). Each isolate was also susceptible to common antibiotics used in the management of melioidosis (Table 4).