Ethical approval of this study was obtained from the Ethics Committee of the Faculty of Tropical Medicine, Mahidol University (approval number MUTM 2015-002-01, MUTM 2018-039-01, and MUTM 2021-055-01) and of 9 hospitals in Northeast Thailand, including Udon Thani Hospital (approval number 0032.102/318), Khon Kaen Hospital (approval number KE58068), Srinakarin Hospital (approval number HE581259), Nakhon Phanom Hospital (approval number IEC-NKP1-No. 15/2558), Mukdahan Hospital (approval number, MEC 010/59), Roi Et Hospital (approval number 166/2559), Surin Hospital (approval number 9/2559), Sisaket Hospital (approval number SSKH REC No. 034/2560), and Buriram Hospital (approval number BR 0032, 102.3/57).

A prospective observational study was conducted in 9 hospitals in Northeast Thailand from July 2015 to December 2018 (Chantratita et al., 2023). Written consents were obtained from the 1,361 adult male and female patients (age ≥ 15 years) with any specimens taken from any sites positive for B. pseudomallei by bacterial culture. Melioidosis patients who were pregnant, receiving palliative care, or incarceration were not enrolled in the study. B. pseudomallei were collected on the day of enrolment (day 0), and participants were followed up on days 5, 12, and 28 and every 2 months via phone call to collect data on the use of antibiotic therapy, clinical outcomes, and the occurrence of recurrent infections. Demographics and clinical data (age, sex, current illness, symptoms, vital signs, underlying morbidity, diagnosis, laboratory results, antimicrobial therapy, and clinical outcomes) of each participant were collected from medical records and hospital microbiology databases. Clinical and follow-up data were recorded in case report forms (CRFs).

Acute melioidosis was defined as the recent manifestation of melioidosis symptoms with a clinical sample culture growing B. pseudomallei, without any evidence of having (i) melioidosis before enrolment and (ii) recurrent infection during follow-up (Currie et al., 2000). Recurrent melioidosis was defined when a patient had (i) signs of infection determined by the attending physician and (ii) remained culture positive for B. pseudomallei at subsequent episodes after completion of antibiotic therapy. A patient with recurrent infection with the same B. pseudomallei genotype that was collected at the first episode was defined as relapse, whereas a patient who had recrudescence of a different genotype was defined as reinfection or was infected with multiple strains (Chantratita et al., 2023). A patient with no evidence of clinical response to oral antibiotic therapy and who subsequently remained culture positive for B. pseudomallei that shared the same genotype clone as the first isolate while undergoing antibiotic therapy was defined as having a persistent infection (Vargas et al., 2021).

Two B. pseudomallei isolates were longitudinally collected from any positive specimens of each melioidosis patient at two time points: on the first day of enrolment (first isolate) and on the day of the recurrent or persistent episode (second isolate). The identification of B. pseudomallei was performed in the microbiology laboratory at each hospital using culture, Gram staining, immunofluorescence assay (IFA), latex agglutination, and standard biochemical tests (Duval et al., 2014; Dulsuk et al., 2016). The bacterial identification was confirmed at the Faculty of Tropical Medicine, Mahidol University, using Matrix-Laser Absorption Ionization Mass Spectrometry (MALDI-TOF MS) as previously described (Suttisunhakul et al., 2017). Bacteria were sub-cultured on Ashdown agar plates, incubated at 37°C overnight, and stored at −80°C in trypticase soy broth containing 20% glycerol.

Genotypes of B. pseudomallei paired isolates collected at the first and second episodes of melioidosis were characterized by PFGE as previously described (Chantratita et al., 2008).

To investigate the phenotypic changes of B. pseudomallei, we compared the ability to induce cell-to-cell fusion between the first and second isolates from each relapse or persistent patients using a modified MNGC assay (Sangsri et al., 2020). Briefly, A549 cells were seeded at 1.5 × 10⁴ cells per well into a 96-well plate and incubated at 37°C in 5% CO₂ overnight. The cells were washed with PBS and infected with B. pseudomallei at MOI of 30:1 and incubated at 37°C for 2 h. The infected cells were washed with PBS and re-incubated with fresh RPMI medium (Gibco BRL, Grand Island, NY, USA) supplemented with 10% heat-inactivated fetal bovine serum (FBS) (HyClone, USA) and 250 μg/ml of kanamycin (Gibco BRL). At 10 h post-infection, the cells were washed and fixed with 4% paraformaldehyde for 30 min and stained with Giemsa stain (Merck KGaA, Darmstadt, Germany). The cells were then examined under a light microscope. The number of MNGC cells was analyzed using the ImageJ software version 1.52n.¹ An MNGC was defined as a cell having three or more nuclei (Whiteley et al., 2017). For each field, the total number of nuclei in MNGCs and the total number of MNGCs were counted. The MNGC efficiency was calculated using the formula: (number of nuclei in MNGCs/total number of nuclei) × 100. The MNGC size was calculated as follows: the total number of nuclei in MNGCs/total number of MNGCs. The MNGC formation assay was performed in three independent experiments, each was performed in triplicate.

An intracellular bacterial survival assay was performed to compare the ability of B. pseudomallei pairs to grow within the host cells. The assay was performed in two independent experiments as previously described (Saiprom et al., 2020), but with some modifications. Briefly, A549 cells were seeded at 1.5 × 10⁴ cells per well into a 96-well plate, washed with PBS, and infected with B. pseudomallei at MOI of 30:1 as described in the MNGC assay. The infected cells were washed with PBS and re-incubated with fresh RPMI medium supplemented with FBS (HyClone) and 250 μg/ml of kanamycin (Gibco BRL). The cells were then washed and lysed with 0.1% v/v Triton X-100 (Sigma) and performed colony count at 2, 4, 6, 8, and 10 h post-infection. Each experiment was performed in triplicate.

The intracellular growth of B. pseudomallei and cell-to-cell fusion were validated by immunostaining and visualization using confocal microscopy (Sangsri et al., 2020). Briefly, A549 cells were seeded at 1 × 10⁶ cells per well on a sterile glass coverslip in a 6-well plate and incubated at 37°C in 5% CO₂ overnight. The cells were washed with PBS and infected with B. pseudomallei at MOI of 30:1 and incubated for 2 h. The infected cells were then washed with PBS and re-incubated with fresh RPMI medium supplemented with 10% FBS and 250 μg/ml of kanamycin. At 8 h post-infection, the cells were washed and fixed with 4% paraformaldehyde for 30 min and permeabilized with 0.5% Triton X-100 for 30 min. The permeabilized cells were incubated with 2.5 μg/ml of 4B11 monoclonal antibody specific to B. pseudomallei capsular polysaccharide at 37°C for 1 h. The cells were further incubated with goat anti-mouse IgG conjugated with Alexa Fluor 488, phalloidin conjugated with Alexa Fluor 647, and Hoechst 33258, at dilution of 1:1,000, to stain B. pseudomallei, cytoplasm, and nucleus, respectively. The stained cells were then washed three times with PBS. The coverslips were mounted on glass slides using 8 μl of ProLong Gold antifade reagent (Invitrogen). Staining cells and bacteria were imaged with a laser scanning confocal microscope (LSM 700; Carl Zeiss, Germany) using a 20 × and 100 × objective lens with oil immersion and Zen software (2010 edition, Carl Zeiss). The excitation and emission wavelengths for Alexa Fluor 488, Hoechst 33258, and Alexa Fluor 647 were 496/519, 352/461, and 594/633, respectively. The number of intracellular survival bacteria was calculated from 20 images (by 20 × objective lens) of two independent experiments using the ImageJ software.

Growth curve analysis was performed to determine the ability of B. pseudomallei to grow in an enrichment medium as previously described (Saiprom et al., 2020). A single colony of B. pseudomallei was inoculated in 3 ml of Luria-Bertani (LB) broth and incubated at 37°C with shaking at 200 rpm for 18 h. Culture bacteria were centrifuged at 12,000 rpm for 5 min, washed with PBS, and resuspended with PBS. The bacterial suspension adjusted to the optical density (OD) at 600 nm to obtain a bacterial concentration of approximately 1 × 10⁸ CFU/ml. The bacterial suspension was added to 10 ml of LB broth to make a final concentration of 1 × 10⁵ CFU/ml. Cultures were incubated at 37°C with shaking at 200 rpm. The viable count was performed at 0, 2, 4, 6, 8, 10, 12, and 24 h post-infection. The growth curve was determined from two independent experiments.

Bacterial genomic DNA was extracted from B. pseudomallei isolates using QIAamp DNA Mini Kit (Qiagen, Germany). Genomic DNA was subjected to the 150-base-read library preparation. WGS was performed in the Illumina HiSeq2000 system with 100-cycle paired-end runs at Wellcome Sanger Institute, Cambridge, UK. To enhance the sensitivity and specificity of within-host mutation analysis, the genomic DNA of a paired isolate from a patient P17 (whose second isolate showed a dramatic change in MNGC formation compared with the first isolate) was analyzed using a long-read sequencing on a MinION sequencer (Oxford Nanopore Technologies, Oxford, UK). FastQC version 0.11.3² was used to check the quality of all genomes. The accession number for all genomes is listed in Supplementary data 1.

The de novo assembly of short-read data was performed for all B. pseudomallei genomes using Velvet version 1.2.10 (Zerbino and Birney, 2008). For each genome, this resolved into an average of 132 contigs (range 74–210 contigs), thereby providing a high resolution for detecting genomic alternations between the primary and relapse and persistent strains. The complete hybrid assembly was performed on long- and short-read sequence data of a paired isolate from P17 using Unicycler version 0.8.4 (Wick et al., 2017). The complete hybrid assembly was used as an internal reference for enhancing the sensitivity and specificity of within-host mutation analysis of isolates from P17. Our hybrid assembly produces two contigs, corresponding to two chromosomes of B. pseudomallei. All assemble genomes were annotated by employing Prokka version 1.13.4 (Seemann, 2014) using B. pseudomallei K96243 genome as a reference. This led to an average of 5,925 genes being annotated per genome (range 5,809–6,089), falling within the same range of previous literature (Holden et al., 2004; Nandi et al., 2010; Chewapreecha et al., 2019).

We studied within-host alterations of all paired isolates collected from relapse and persistent patients by comparing genomes of B. pseudomallei pairs isolated from the same patients. Variants including SNPs and indels were called from the genomic comparison using Snippy version 4.6.0.³ To avoid false variants, we filtered out reads with < 10 read depth and found a frequency of < 90%. All variants identified by Snippy were curated manually by visualizing with the Tablet software.⁴ The functions of mutated genes were annotated using Clusters of Orthologous Group (COG) terms of B. pseudomallei K96243 available on the Burkholderia Genome Database (Winsor et al., 2008).

Data were analyzed using the R version 4.0.4 and GraphPad Prism 10 software (GraphPad Inc.). Firth logistic regression (Puhr et al., 2017) was used to analyze risk factors for relapse and persistent infections using R-package logistf⁵. The Mann–Whitney U test was performed to analyze the differences in the median of MNGC efficiency, MNGC size, and the number of intracellular bacteria between the first and second isolates from patients with relapsed or persistent infections.

We enrolled 1,361 melioidosis patients from 9 hospitals in Northeast Thailand, of whom 1,046 (76.8%) survived and were discharged from the hospital, allowing us to follow up to investigate the prevalence of relapse, persistence, and reinfection. Of all 1,046 patients, 927 patients had acute melioidosis, 89 patients were suspected to have a history of melioidosis at least 8 days before enrolment, while 23 and 7 patients were suspected to have recurrent and persistent infections, respectively. Of the 23 suspected recurrent patients, we excluded 7 patients, since the subsequent B. pseudomallei isolates were not collected from the recurrent episode. Therefore, we analyzed the remaining 16 suspected recurrent patients and 7 suspected persistent patients using PFGE, MLST, and WGS (Figure 1). Genotyping analyses of all B. pseudomallei paired isolates from these patients revealed 13 (1.2%) patients with relapse, 7 (0.7%) with persistent infection, and 3 (0.3%) with reinfection or polyclonal infection (Figure 1; Supplementary data 1). The genomic evidence supporting the relapse and persistent patients is given in Supplementary data 2. The median duration of relapse and persistent was 245 days (range 164–518, IQR = 181–278) and 74 days (range 38–174, IQR = 68–103), respectively.

Relapse and persistent infections represent treatment failure in melioidosis. Understanding the risk factors associated with relapse and persistent infections is fundamental to the control of infection. Since the number of relapses and persistent infections was low, we performed Firth logistic regression to identify risk factors for the combined number of patients with relapse and persistent infections (N = 20). Firth logistic regression is an approach for the analysis of rare event data. Results in Table 1 demonstrated that the regimen and duration of oral antibiotic therapies were the factors associated with relapse and persistent infections (P = 0.005 and P = 0.031, respectively). For a regimen of oral antibiotic therapy, patients who were treated with amoxicillin, ciprofloxacin, clindamycin, and/or rifampicin were more presented in relapse and persistent than patients with acute infection [7/20 (35.0%) vs. 106/927 (11.4%)]. Similarly, patients who were treated with trimethoprim/sulfamethoxazole were more presented in acute infection than patients with relapse and persistent infections [693/927 (74.8%) vs. 10/20 (50.0%)]. For the duration of antibiotic therapy, patients who received any of these oral antibiotics (trimethoprim/sulfamethoxazole, doxycycline, augmentin, amoxicillin, ciprofloxacin, clindamycin, and rifampicin) for 3–6 months were not likely to be at risk of relapse and persistent than those with acute [8/20 (40.0%) vs. 567/927 (61.2%)]. However, underlying morbidity, clinical manifestation, regimen, and duration of initial IV antibiotic therapy were not associated with the risk of relapse and persistent infections.

Multinucleated giant cell formation (MNGC) formation assay was performed on paired isolates from all 20 patients to determine the alteration of cell-to-cell fusion ability. We observed variations in MNGC formation ability among B. pseudomallei isolates, with overall median MNGC efficiency of 8.0% (range 1.2–88.3%, IQR = 4.6–9.3%) and median size of 9 nuclei/MNGC (range 4.8–177.2, IQR = 6.8–11.7) (Figure 2). Comparison between the paired isolates from all 13 relapse patients showed no significant difference between the first and second isolates on MNGC efficiency and size (Figures 2A, B). All but one pair of the first and second isolates from seven persistent patients showed similar MNGC efficiency and size. Interestingly, the second isolate DREX512 from P17 showed significantly higher MNGC efficiency and size than the first isolate DR60101A (efficiency 88.3 vs. 3.6%, P < 0.0001; size 177.2 vs. 5.5, P < 0.0001) (Figures 2A–C). DR60101A and DREX512 were isolated 98 days apart. The dramatic change in MNGC efficiency of the second isolate compared with the first isolate suggests that B. pseudomallei increased cell-to-cell fusion induction during infection in patient P17.

To determine whether the second isolate from P17 has also changed the ability to grow within the host cells, we performed intracellular survival assays of paired isolates from P17 and reference stain K96243 in A549 cells at 2, 4, 6, 8, and 10 h post-infection. Our experiment demonstrated that the second isolate DREX512 could survive and increased its intracellular growth over time after infection compared with the first isolate DR60101A and B. pseudomallei K96243, suggesting that increased cell-to-cell fusion efficiency was related to intracellular replication in DREX512 (Figure 3A). The median number of intracellular bacteria of the second isolate DREX512 was significantly higher than that of the first isolate DR60101A at 4 (10,125 ± 2,125 CFU/ml vs. 5,000 ± 2,350 CFU/ml, P = 0.002), 6 (23,875 ± 5,687 CFU/ml vs. 11,250 ± 2,300 CFU/ml, P = 0.002), and 8 h (238,750 ± 71,250 CFU/ml vs. 84,500 ± 28,375 CFU/ml, P = 0.002) post-infection, but not different at 10 h post-infection (925,000 ± 168,750 CFU/ml vs. 750,000 ± 512,500 CFU/ml, P = 0.192). However, we further validated the intracellular growth of bacteria by visualization using confocal microscopy at 8 h post-infection. Our confocal microscopy confirmed that DREX512 formed larger MNGC and multiplied better in the host cells than DR60101A (144 ± 74 vs. 40 ± 23 bacteria cells/image, P < 0.0001) and K96243 (median 45 ± 27 bacteria cells/image, P < 0.0001) (Figures 3B, C).

To examine whether the second isolate DREX512 also enhances the growth in the enrichment medium compared with the first isolate DR60101A and the reference strain K96243, we performed two experiments of growth curve analysis of these isolates in LB at 8 time points. Our assay demonstrated that the growth rate of DREX512 was comparable to DR60101A and K96243 at every time point (Supplementary Figure 1). This result indicated that DREX512 normally replicated in LB broth but increased its replication in the host cells.

Whole-genome sequencing (WGS) was performed on the genomic DNA extracted from 20 within-host strain pairs obtained from the patients with relapse (13 pairs) and persistent (7 pairs) infections. Using the Snippy and Tablet software, we identified one indel and 16 SNP mutations of 6 of 20 paired isolates collected from 5 of 13 patients with relapse (P1–P3, P10, and P12) and 1 of 7 patients with persistent infection (P17) (Table 2; Figures 4A, B). Therefore, the within-host strain pairs from P4–P9, P11, P13–P16, and P18–P20 shared identical genome. A relatively large indel mutation (189 bp) was detected in the TetR family regulatory protein (amrR, BPSL1805) of a paired isolate from a relapse patient (P12). This structural variant was suspected to be associated with decreased susceptibility to meropenem in our previous study (Fen et al., 2021) and is being validated in future studies. Of the 16 SNPs, 11 were non-synonymous, 2 were synonymous, and 3 were in the intergenic region (Figures 4A, B), indicating the occurrence of purifying pressure on synonymous than on non-synonymous variants. The non-synonymous mutations were distributed across 11 genes involved in multiple functions, including transcription (2 genes), general function prediction (3 genes), cell wall biogenesis (1 gene), inorganic ion transport and metabolism (2 genes), energy production and conversion (1 gene), and virulence (1 gene) (Table 2). To further characterized within-host evolution of B. pseudomallei, we performed spectrum mutations analysis of the 16 SNPs. We found that A:T > G:C nucleotide transition was the most common (37.5%), while A:T > T:A and A:T > C:G transversion was 6.2% (Figure 4C). Of the SNPs observed in this study, 50% was G:C > A:T transition and G:C > T:A transversion, the patterns of which putatively occur due to oxidative damage to the bacterial DNA (Ford et al., 2011; Dillon et al., 2015).

For all 20 patients, the mutation rate ranged from 0 to 18.6 SNPs per genome per year (Supplementary data 3). However, an average mutations rate for all within-host strain pairs was not estimated, since an increase in within-host SNPs was not related to the time between isolates collection (R² = 0.0709, P = 0.1) (Figure 4D). We observed that paired isolates from P17 had the highest number of SNPs (5 SNPs) with a short period of time between isolates collection (98 days). The patient was previously diagnosed and completely treated for TB. For the first episode, the patient was admitted with acute respiratory failure and finally diagnosed with pneumonia melioidosis. The patient received IV ceftazidime for 10 days and was discharged to home 10 days after admission. The patient was prescribed ciprofloxacin for 180 days as oral antibiotic therapy. At 60 days after discharge, the patient was reported not responding to therapy and was readmitted to the same study hospital with pneumonia melioidosis at 98 days after the first admission. Although the IV antibiotic therapy data were not collected during the second episode, the patient was reported to receive oral ciprofloxacin for 180 days which was provided during the first episode. The first and second B. pseudomallei isolates were collected from sputum at the first and second episodes, respectively. After the second episode of admission, the patient did not respond to therapy and was readmitted two times to the community hospital. Finally, the patient died 332 days after the first admission or 231 days after readmission at the persistent episode.

Whole-genome sequencing (WGS) analysis of the paired isolates from P17, which differed in intracellular replication and cell-to-cell fusion, identified 5 non-synonymous mutations in 5 different genes, including chromate transporter (chr, BPSL0285), hydrolase (BPSL3337), NAD (P) transhydrogenase subunit alpha (BPSL2887), ferric uptake regulator (fur, BPSL2943), and type VI secretion system (T6SS-5) protein TssB-5 (tssB-5, BPSS1496) (Table 2). Since the indel and structural mutations were not detected, we hypothesized that the 5 non-synonymous mutations may play a role in the phenotypic alternations in this strain. The functions of chromate transporter, hydrolase, NAD (P) transhydrogenase subunit alpha, and ferric uptake regulator proteins were not obviously related to within-host infectivity or pathogenesis of Burkholderia species (Table 2). The enhancement of intracellular replication and cell-to-cell fusion in a persistent isolate of P17 may not be related to the non-synonymous mutations of these 4 genes. Interestingly, we found a non-synonymous mutation in tssB-5. This gene encodes an essential component of T6SS-5, a virulent factor of B. pseudomallei, involved in intracellular replication and MNGC formation (Pilatz et al., 2006; Burtnick et al., 2011; Chen et al., 2011; French et al., 2011; Hopf et al., 2014; Schwarz et al., 2014; Toesca et al., 2014; Nguyen et al., 2018). We postulated that a non-synonymous mutation in the tssB-5 gene shifted B. pseudomallei to have stronger intracellular survival and MNGC phenotypes. Unfortunately, according to biosafety regulation, tssB-5 is a highly restricted pathogenic sequence of B. pseudomallei K96243 and Burkholderia mallei ATCC 23344. We could not synthesize the tssB-5 mutant fragment for mutagenesis to investigate the effect of this mutation on intracellular replication and cell-to-cell fusion of the second isolate DREX512 from P17.