Thirty clinical Elizabethkingia spp. strains collected in China and American Type Culture Collection (ATCC 13253) were used in this study (Sun et al., 2014, 2015). The identification and antimicrobial susceptibility of 31 strains were tested using the automated VITEK 2 Compact system (BioMérieux, France) with the gram-negative identification card (GN) and AST-GN16 card, respectively. All strains were grown aerobically in trypticase soy broth (TSB; Sigma-Aldrich, United States) at 37^°C overnight and then seeded on a blood agar plate (BioMérieux, France) for single colonies.

A standard assay (CLSI, 2019) was performed in a 96-well microplate (Bio-Kont, China). The tested bacterial colonies on each blood agar plate were aseptically transferred into Mueller-Hinton (MH) broth (BD Biosciences, United States), and a homogenous suspension with a density equivalent to 0.5 McFarland’s standard was prepared. The bacterial suspension was then diluted with MH broth at a ratio of 1:200. The bacterial solution (100 μL) was inoculated into wells with sequentially diluted antimicrobials and incubated at 37^°C at 180 rpm for 16 h. Aztreonam, cefepime, imipenem, ticarcillin/clavulanic acid, amikacin, tobramycin, minocycline, and levofloxacin were used in this study. The concentrations of the eight antimicrobials were based on the breakpoints of other non-Enterobacteriaceae listed in the guidelines M100 (CLSI, 2019). MIC values and the results of antimicrobial susceptibility were interpreted based on the guidelines M100 (CLSI, 2019).

D₂O (99% D atom, Sigma-Aldrich, United States) labeling was performed as previously reported (Berry et al., 2015; Song et al., 2017). In this study, the three breakpoint-concentrations of each antimicrobial were tested, instead of multi-proportion dilution method. The benefit was that the number of tests was reduced, which increased detection efficiency and reduced time. Three to five isolated colonies selected from each blood agar plate were subjected to single cell Raman-DIP. Bacterial suspension was prepared for AST. One hundred microliters of inoculated bacterial solution was added to each 96-well plate with standard concentrations of antimicrobial suggested in the guidelines M100 (CLSI, 2019) and incubated at 37°C at 180 rpm for 1 h. Sixty-six microliters D₂O was added to each well, and then incubated for two more hours. The cells were washed with sterile deionized water three times and resuspended in 20 μL of sterile deionized water. The cell suspension (2.5 μL) was transferred onto an aluminum-coated slide which was placed on the x-y-z motorized stage of a confocal Raman spectrometer (P300, Hooke Instruments, China) with 100 × magnifying dry objective (Olympus, Japan). The laser power used on the samples was approximately 5 mW. At least 20–30 single cell Raman spectra were recorded by a –70°C cooled charge-coupled device (PIXIS 100 B, Princeton instruments, United States) with an integration time of 8 s per spectrum.

The FMS-007 strain was used as an example to simulate bloodstream infection samples. The bacterial contents in the simulated samples ranged from 10⁶ to 10⁷. Each simulated sample (1 mL) was collected and diluted ten times (1 mL sample was added to 9 mL sterile normal saline). The diluted samples (10 mL) were injected into a serum separator tube with gel and clot activator (Becton, Dickinson and Company, United States), statically incubated at 37°C for 1 h, and then centrifuged to remove the majority of blood cells at 2,500 rpm for 2 min. Bacteria were present in the supernatant, washed three times with sterile water, and resuspended in 1 mL of sterile water. AST was performed as described previously. This protocol was reviewed and approved by the Medical Ethics Committee of Zhejiang Chinese Medical University (Approval No. 20211202-4).

The range of the Raman spectra was extracted from 400 to 3,400 cm^–1. The cosmic rays and baselines were corrected in Labspec6 (Horiba JY, Tokyo, Japan) with 10-degree linear baseline fitting algorithm. Then, the spectra were normalized using a total area below all the peaks within the spectral range. The Raman spectra for carbon-deuterium (C-D) peaks (2,040–2,300 cm^–1) and carbon-hydrogen (C-H) peaks (2,800–3,100 cm^–1) were obtained. The C-D ratio was calculated (C-D/C-D+C-H) and normalized (C-D ratio of treated group minus C-D ratio of no deuterium and no antimicrobials controls). The impact of treatment was determined by the relative metabolic rate (defined as the ratio of normalized C-D of the treatment to that of the no drug control; Yi et al., 2021). Graphpad Prism was used for statistical analysis. Statistical tests performed were indicated in figure legend. Numerical data were shown as mean ± SEM. P values of less than 0.001 were considered statistically significant.

The AST results of 31 strains were listed in Table 1 and they were consistent with previous reports (Burnard et al., 2020; Wang et al., 2020). Collectively, our previous studies of around 150 clinical isolates indicated that the resistance rates of aztreonam, ampicillin, cefazolin, ceftriaxone, imipenem, amoxicillin/clavulanic acid, tobramycin, and nitrofurantoin were >97% in the tested strains (Sun et al., 2014; Li et al., 2020). However, ciprofloxacin and levofloxacin all demonstrated susceptibility rates of >60%.

Single cell Raman-DIP was used to measure the metabolic activity of bacteria at the single-cell level (Berry et al., 2015). Elizabethkingia spp. strains were incubated in MH medium containing 40% heavy water for 2 h, the deuterium in the medium was incorporated into the cellular biomass and produced a new peak at 2,040–2,300 cm^–1 (C-D band) on a background of 2,800–3,100 cm^–1 (C-H band; Figures 1A,D). Imipenem was used as an example to detect metabolism under antimicrobial treatment using single cell Raman-DIP. After exposure to imipenem, the intensity of the C-D band in strain FMS-007 remained at the same level as the drug-free control, in which no imipenem was added (Figure 1A), and the C-D ratio (Figure 1B) were calculated. The number of biological replicates in each group was shown in supplementary data (Supplementary Table 2). This indicated that bacteria were still metabolically active under imipenem up to 16 μg/mL, resulting in an MIC of >16 μg/mL. Since 16 μg/mL was the resistant breakpoint of Elizabethkingia spp. for imipenem, strain FMS-007 was regarded as resistant to imipenem. The C-D band of strain TZ-3 was not detected when imipenem was present at a concentration of 16 μg/mL (Figures 1C,D), indicating that the bacterial metabolic activity was inhibited and the MIC was determined to be 16 μg/mL. The results showed that single cell Raman-DIP could be used to detect microbial metabiotic activity and antimicrobial susceptibility of Elizabethkingia spp.

The antimicrobial susceptibility of five Elizabethkingia spp. strains to eight antimicrobials was tested using single cell Raman-DIP and classical BMD methods. The results of single cell Raman-DIP and BMD were summarized in Figure 2 and Table 2, respectively. The breakpoints of antimicrobial susceptibility according to the guidelines M100 (CLSI, 2019) were used in this study.

To compare the results from Raman-DIP and BMD, we used 3 cutoff values for relative metabolic rate, i.e., 0.6, 0.7, and 0.8. Antibiotic treated single cells with relative metabolic rate below the cutoff value were considered to be inhibited by the antibiotic. The cutoff value of 0.7 was shown to give the highest consistency with conventional BMD approach (Supplementary Table 1), and thus the corresponding concentration of relative metabolic rate < 0.7 was used to interpret the results of antimicrobial susceptibility. The results of seven antimicrobials were consistently detected by single cell Raman-DIP and classical BMD methods, including aztreonam, imipenem, ticarcillin/clavulanic acid, amikacin, tobramycin, minocycline, and levofloxacin. However, the AST results of the two methods were different for cefepime. Five strains were all resistant to cefepime tested by single cell Raman-DIP, but the results of testing by BMD were sensitive or intermediate.

In the AST results tested by single cell Raman-DIP, all five strains were resistant to five antimicrobials (a total of eight antimicrobials tested), including aztreonam, cefepime, ticarcillin/clavulanic acid, amikacin, and tobramycin (Figure 2). ATCC 13253, HS-2, NB-46, and TZ-3 were sensitive to minocycline, while FMS-007 was intermediate. Except for strain HS-2, which was intermediate, they were resistant to imipenem. FMS-007 and TZ-3 were sensitive to levofloxacin, but ATCC 13253, HS-2, and TZ-3 were resistant. This implied that minocycline and levofloxacin were probably most suitable drugs for Elizabethkingia spp. infection.

Based on the above results, single cell Raman-DIP could be used to detect metabiotic activity of Elizabethkingia spp. Because Elizabethkingia spp. were more sensitive to minocycline and levofloxacin, the two antimicrobials were tested against the other 26 Elizabethkingia spp. The cutoff value of 0.7 was used to interpreted the AST of other 26 Elizabethkingia spp. The results from single cell Raman-DIP and classic BMD methods were shown in Figure 3 and Table 3. The AST of minocycline tested by BMD (Supplementary Table 3) and single cell Raman-DIP was 100% consistent, while the consistency of levofloxacin was approximately 87.10%.

Bloodstream infection caused by Elizabethkingia spp. has been reported (Reed et al., 2020; Kuo et al., 2021). In this study, the blood samples were diluted ten times and injected into a serum separator tube with gel and clot activator to remove blood cells (Figure 4). It was a much more convenient and effective way to remove blood cells than using ammonium-chloride-potassium (ACK) buffer and repeated washing (Han et al., 2020; Yi et al., 2021). The relative metabolic rate directly measured from bacterial single-cells in the blood sample were shown in Figure 5. The number of biological replicates in each group of simulated blood sample was shown in supplementary data (Supplementary Tables 4, 5). The results indicated that the pathogens in the blood samples were susceptible to minocycline and levofloxacin since their relative metabolic rate were below 0.7 when the concentrations were at the breakpoints concentration for susceptible. These results were the same as the clinical isolates results (Figure 2 and Table 2). The metabolic activity of bacteria in the blood sample with bacterial concentration as low as 10⁶ CFU/mL (except for one case where single cells were not spotted in bacterial contents of 10⁶ treated with 16 μg/mL minocycline) can be determined by Raman-DIP without strain sub-culture. These results demonstrated that this method could be directly integrated into current clinical protocol for sepsis and halve the report time for AST (Figure 4).