Early detection of drug-resistant Streptococcus pneumoniae and Haemophilus influenzae by quantitative flow cytometry

Early detection of drug resistance contributes to combating drug-resistant bacteria and improving patient outcomes. Microbial testing in the laboratory is essential for treating infectious diseases because it can provide critical information related to identifying pathogenic bacteria and their resistance profiles. Despite these clinical requirements, conventional phenotypic testing is time-consuming. Additionally, recent rapid drug resistance tests are not compatible with fastidious bacteria such as Streptococcus and Haemophilus species. In this study, we validated the feasibility of direct bacteria counting using highly sensitive quantitative flow cytometry. Furthermore, by combining flow cytometry and a nucleic acid intercalator, we constructed a highly sensitive method for counting viable fastidious bacteria. These are inherently difficult to measure due to interfering substances from nutrients contained in the medium. Based on the conventional broth microdilution method, our method acquired a few microliter samples in a time series from the same microplate well to exclude the growth curve inconsistency between the samples. Fluorescent staining and flow cytometry measurements were completed within 10 min. Therefore, this approach enabled us to determine antimicrobial resistance for these bacteria within a few hours. Highly sensitive quantitative flow cytometry presents a novel avenue for conducting rapid antimicrobial susceptibility tests.

www.nature.com/scientificreports/ such as dead bacteria and aggregates derived from these components reportedly interfere with specific and highsensitivity FCM 31,32 . Therefore, in this study, we confirmed the feasibility of conducting specific and highly sensitive FCM analysis with bacterial strains that require a growth medium containing more nutrients than usual 33 . In conventional studies using FCM, a method for predicting antimicrobial susceptibility has been investigated by comparing the ratio between plot groups consisting of scattered light or fluorescence intensity involved in cell activity 34,35 . Intercalators that permeate cell membranes, such as SYTO9, are used to measure whole bacterial counts. The impermeable intercalator propidium iodide which is taken up only by damaged bacteria is used to distinguish dead bacteria 32,33,35 . With the advent of quantitative FCM, the feasibility of applying FCM to drug susceptibility testing has recently been demonstrated 36,37 . However, these studies focused on bacteria that can grow in general media, such as Escherichia coli and Pseudomonas aeruginosa. As far as we know, detailed examinations of fastidious bacterial species have not yet been conducted 38 . Moreover, since the minimum inhibitory concentration (MIC) is considered according to a logarithmic distribution, consistent growth curves may not be obtained even if the same bacterial solution is used 39,40 .
We, therefore, combined quantitative FCM and a fluorescent intercalator to measure the time-dependent change in viable cells, using a few microliter samples from only 0.1 mL medium of the conventional microdilution method. Comparing the performance between the two, the feasibility of direct bacteria counting as a rapid phenotypical AST was validated.

Results
Viable cell counting from mixed medium solution. First, to validate the direct counting of bacteria in the medium by FCM, we analyzed the correlation between the numbers of bacteria and colony-forming units (CFUs). The target bacteria were S. pneumoniae, grown in a medium containing lysed horse blood, and H. influenzae, grown in a medium containing yeast extract and nicotinamide adenine dinucleotide. For live cell gating, we referred to a plot of a mixture of live and heat shock killed bacteria in a filtered buffer (Fig. 1a,e). The killed bacteria stained by propidium iodide (PI) had a stronger fluorescence intensity of FL3. In the case of S. pneumoniae, the fluorescence intensity by SYTO9 did not change with PI ( Fig. 1b) unlike in H. influenzae (Fig. 1f). In the non-staining method, the plots of bacterial cells obtained using forward-scattered light and side-scattered light were single-gated. Therefore, the bacteria were counted as a whole number. www.nature.com/scientificreports/ To determine the detection sensitivity, serial dilutions with a culture medium specific to the target bacterial species were prepared for both colony and FCM measurements. FCM was performed using a scattered light plot for non-staining bacteria and a fluorescence plot for staining bacteria. In the SYTO9/PI staining method, the gate indicating viable bacterial cells was obtained from the FL1 and FL3 plots (Fig. 1c,g).
Detection sensitivity was defined as the point at which a significant difference was obtained. This was determined using Dunnett's multiple comparison test with reference to the negative control. The concentration of S. pneumoniae (22 events/test; P = 0.0003) was detected with 4-fold more sensitivity using fluorescent staining than with the non-staining method (86 events/test; P < 0.0001). Since the FCM measurement used a diluted sample, the actual measured concentration was multiplied by the dilution rate. Therefore, using the fluorescence method, the sensitivity of counting the viable S. pneumoniae cells from the culture medium was found to be approximately 3 × 10 4 CFU/mL. However, a substantial difference was not observed between the non-staining method for detecting all bacteria and the fluorescence method targeting only viable bacteria. In H. influenzae, the detectable point was 3099 events/test (P = 0.0004) for the non-staining method. The detectable point was identified with less sensitivity than that of S. pneumoniae due to the interference of contaminants in the medium at the gate of the scattered light plot. The fluorescent staining method maintained a sensitivity of 12 events/test (P = 0.0022), which was 2 × 10 4 CFU/mL in terms of concentration in the culture medium. This sensitivity was equal to or higher than that observed for S. pneumonia. This suggests that the fluorescent staining method is 64-fold more sensitive than the non-staining method. The measured values of H. influenzae from the non-stained samples were affected by contaminants in the medium from yeast extract and nicotinamide adenine dinucleotide components (Supplemental Fig. 1). However, accurate whole bacterial cell counting was possible up to the maximum detectable points of the equipment (approximately 5 × 10 5 events/test) using the non-staining method. Since the initial bacterial concentration via the general AST method is 5 × 10 5 CFU/mL, these sensitivities of the fluorescent staining method were sufficient to reveal the changes in bacterial counts upon exposure to antimicrobial agents.
Nonlinear logarithmic regression analysis showed a high positive correlation between the CFUs measurement and FCM measurement, regardless of medium, species, and staining ( Fig. 1d,h). Fluorescent staining and FCM measurement were completed within 10 min. Consistent antimicrobial susceptibility testing. Next, we sought to confirm the consistency of FCM and MIC with broth microdilution (BMD). We observed bacterial growth using FCM by cultivating bacteria in a medium containing an antimicrobial agent at certain concentrations. This allowed us to determine antibacterial resistance, as defined by the CLSI 19 . As shown in Table 1, S. pneumoniae ATCC 6303 was confirmed to be penicillin G-sensitive, while S. pneumoniae ATCC 49619 was penicillin G-resistant but cefotaxime-sensitive (PRSP: Penicillin G Resistant S. pneumoniae), and S. pneumoniae ATCC 700677 was resistant to both penicillin G and cefotaxime (MRSP: Multi-drug resistant S. pneumoniae). H. influenzae ATCC 49766 was susceptible to ampicillin and did not produce beta-lactamase. H. influenzae ATCC 33533 was resistant to ampicillin due to its beta-lactamase production but was susceptible to ampicillin-sulbactam. H. influenzae ATCC 49247 did not produce beta-lactamase but was resistant to both ampicillin and ampicillin-sulbactam.
Before FCM measurement, the bacterial suspensions were prepared and inoculated at approximately 5 × 10 4 CFU/well on a microplate. This was done according to the standard BMD method. Subsequently, 5 µ L samples collected at different time points were stained in wells and measured by FCM, as described above. Table 1. Antimicrobial resistance of the bacterial strains used in this study. Antimicrobial resistance was determined from the minimum inhibitory concentration (MIC) based on the triple measurement obtained via broth microdilution (BMD). Penicillin G was the parenteral breakpoint, and ampicillin was the nonmeningococcal breakpoint. The beta-lactamase was confirmed to be TEM-1 or ROB-1 by polymerase chain reaction (PCR) of Haemophilus influenzae. The detection time for Streptococcus pneumoniae was defined as the time when 60% or more bacterial growth was observed compared to the initial bacterial count. The detection time of H. influenzae was determined after at least 2 h of incubation, at which time the observed bacterial growth was compared to the bacterial count at 1 h of incubation. www.nature.com/scientificreports/ Initially, 500-1000 cells were measured for S. pneumoniae (equal to 6.5 × 10 5 to 1.3 × 10 6 CFU/mL) and 100-300 cells/events (1.3 × 10 5 to 3.9 × 10 5 CFU/mL) for H. influenzae. Both PRSP and MRSP were grown in the wells containing penicillin G at breakpoint concentrations (Fig. 2). Growth of the bacterial count was confirmed 1 h after inoculation, and resistance was confirmed. In cefotaxime, the results of FCM were also consistent with those of BMD. MRSP proliferation was only observed in 1 μg/mL cefotaxime; moreover, resistance was confirmed within 90 min. MRSP showed resistance to both penicillin G and cefotaxime. These results indicate that FCM could determine a non-susceptible bacterium in 90 min ( Table 1).
Regardless of the existence of an antimicrobial, all H. influenzae strains grew for 1-1.5 h after inoculation and then stagnated for 0.5-1 h (Fig. 3). Subsequently, after 2 h or more, they either died or proliferated. Therefore, the resistance was determined when the bacteria have grown compared to the number of bacteria 1 h after inoculation. Ampicillin resistance was detectable in both strains at 2 h, and resistance to the combination of ampicillin-sulbactam was detectable at 2.5 h (Table 1).
Incidentally, to confirm the effect of volume reduction due to multiple sampling from the same well, the remaining sample was finally cultured for up to 22 h. The resistance predicted from FCM measurements were consistent with the results of BMD.

Discussion
In this study, we showed the feasibility of rapid and accurate counting of viable bacterial cells using highly sensitive and quantitative FCM. This method was coupled with a culture-based testing method and a nucleotide intercalator. Furthermore, we confirmed that this method can contribute to the rapid detection of antimicrobial resistance in bacteria. Surprisingly, clinically important resistance of S. pneumoniae could be detected within 1 h. This demonstrates that the culture time required to obtain results can be reduced by 95% or more compared with the conventional BMD method. Additionally, this method was able to detect the antimicrobial resistance of S. pneumoniae and H. influenzae earlier than the previously reported method of rapid measurement by FCM, which could only be applied to bacteria such as E. coli 36 .
Although there have been many reports of viable cell counting or monitoring using FCM since the 1980s, they were either clinically impractical due to the use of FCM equipment that lacked sensitivity or were nonquantitative 36,41 . These methods required complicated analytical logic and time-consuming sample preparation. Furthermore, there are no reports on viable cell counts of Streptococcus or Haemophilus species by FCM, especially the detection of antimicrobial resistance. By excluding the interference of contaminants from the culture medium, the method described here was able to count viable S. pneumoniae and H. influenzae cells with high  www.nature.com/scientificreports/ sensitivity. Although the regression line between CFUs measurement and FCM counting shows high correlation values, FCM counting of S. pneumoniae have a tendency to be higher than CFU measurements. This may be because FCM counting were lower due to the effects of self-lysis during the test or frictional heat of smearing on the agar medium 42 . H. influenzae showed negative sections in fluorescent staining, suggesting that they were not completely stained because of media contaminants or adsorption on tubes while preparing dilution samples for FCM measurements 43 . Previous studies have revealed that 36,41 this method enables the counting of both viable and dead cells of various bacterial species, such as E. coli, P. aeruginosa, Staphylococcus aureus, and Enterococcus faecalis in addition to species reported in this paper (data not shown). However, in the preliminary experiment as some plots of the dead cells prepared via heat shock disappeared, a precise measurement of the dead cells could not be conducted. The discrepancy in SYTO9 staining with PI between S. pneumoniae, a gram-positive coccus and H. influenzae, a gram-negative bacillus (Fig. 1b,f), would be related to factors such as membrane structures as well as membrane permeability 44 .
In this report, the resistance of S. pneumoniae and H. influenzae could be detected from 5 μL samples with a sensitivity of approximately 3 × 10 4 CFU/mL or more. Since the initial bacterial concentration at the time of inoculation in the conventional BMD method is 5 × 10 5 CFU/mL, our method requires a sample volume of 1 µ L or less per well using a 96-well microplate. As the influence of sampling is likely to be small, we can carry out early detection of antimicrobial resistance and confirm the results of culturing along with the conventional BMD method. In the future, detection accuracy can be further improved by using other fluorescent reagents, such as fluorescein isothiocyanate isomer, which can provide further information of proteins or enzymes in bacteria compared to intercalators only targeting for nucleic acids 45 .
As the performance of FCM equipment has recently improved, there have been studies measuring timedependent bacterial changes or the accurate quantitative measurement of viable cells. However, most of these studies took single measurements per vial. On the other hand, our method allowed continuous measurement from one well and analysis of the details of proliferation events. For S. pneumoniae, bacterial growth could be detected within 1 h. It may be possible to detect growth even earlier than 1 h by increasing the number of measurement time points. Interestingly, we found that H. influenzae grew for approximately an hour after inoculation, regardless of being resistant or susceptible. This may be because of the post-antibiotic effect. Once bacteria were exposed to an antimicrobial agent, growth was inhibited even after removing the antimicrobial agent 46,47 . These phenomena suggest that monitoring the bacterial count is important for the reliable prediction of antimicrobial resistance 36 .
Although we confirmed the effects of volume reduction due to multiple sampling from the same microplate well using S. pneumoniae and H. influenzae, further studies need to be done with other strains or resistant strains. Moreover, since this method is based on BMD, it is necessary to pay attention to the MIC distribution, especially in the sub-MIC well. For example, the results of growth changes depend on the number of bacteria, compatibility with the medium, the number of plasmids in the bacteria, and other factors. It will, therefore, be necessary to combine molecular biological techniques in the future, such as the detection of resistance factors 48 .
In this study, we found that FCM presents a novel avenue for conducting rapid ASTs. This platform may be advanced with technological innovations such as FCM imaging. With these technologies, further improvements in infectious disease treatment and suppression of drug-resistant bacteria are expected.

Methods
Strains and target antimicrobials. Three strains each of S. pneumoniae and H. influenzae were used: one antimicrobial-susceptible strain and two resistant strains were prepared. Viable cell counting was performed using S. pneumoniae ATCC 49619 and H. influenzae ATCC 49766, which are used as control strains in CLSI testing 19 .
Confirmation of antimicrobial resistance. The antimicrobial resistance and MIC profiles of the strains were determined by the manual method, according to the BMD described in the CLSI criteria 30 . Cation-adjusted Mueller-Hinton broth (Becton Dickinson, MD, USA), which had lysed horse blood added to a final concentration of 5%, was used as a medium for S. pneumoniae. The Haemophilus test medium for H. influenzae was used as described in the CLSI criteria. This medium was prepared using cation-adjusted Mueller-Hinton broth, nicotinamide adenine dinucleotide, and yeast extract. The production of H. influenzae beta-lactamase was confirmed by PCR using previously reported primers synthesized by Eurofins Genomics (Tokyo, Japan) 53 .
Counting of colony forming units. For CFUs measurement, samples of 100 to 300 CFU/20 μL were selected from the dilution series used for FCM quantitative viable cell counting, described later, and directly applied to different agar media for each bacterial species. S. pneumoniae strains were cultured in 5% sheep blood agar medium (Eiken Chemical, Tochigi, Japan) for 18 h at 35 °C. H. influenzae was cultured in chocolate agar medium (Eiken Chemical, Tochigi, Japan) in 5% CO 2 at 35 °C. The number of cultured colonies was multiplied by the dilution ratio and used as a standard for determining the number of viable bacteria. Samples were subsequently incubated at room temperature in the dark for 5 min. FCM measurement was performed after fivefold dilution with PBS to suppress background fluorescence. All fluorescent staining reagents were purchased from Thermo Fisher Scientific (Waltham, MA, USA). FCM measurement was performed using RF-500 (Sysmex, Kobe, Japan) equipped with a blue semiconductor laser (488 nm wavelength). The built-in front scattering (FSC) and side scattering (SSC) detectors were used to detect scattered light. An FL1 filter (527+/−15 nm, SYTO9) and FL3 filter (695+/−25 nm, PI) was used for fluorescence detection. The measurement volume was set to 20 μL and the flow rate was 1.85 μL/s. Flowcytometric viable cell counting. For viable cell monitoring by FCM, a 5 μL sample was collected from a 100 μL medium containing the bacteria. The sample was taken from the same microplate well of BMD and mixed with PBS containing SYTO9 (5 μM) and PI (15 μM). Next, each sample was incubated at room temperature in the dark for 5 min. FCM measurement was performed after 5-fold dilution with PBS to suppress background fluorescence. The microplate with inoculated wells was immediately returned to the incubator to restart the culture. The FCM measurement was operated on RF-500 software (Sysmex, Kobe, Japan). Bacterial counting was performed using FCS Express 6 RUO Edition (De Novo Software, Los Angeles, CA, USA).

Statistical analysis.
The detection limits of the viable cell counting were analyzed by Dunnett's multiple comparison test. This test was used to compare the measured value at each time point and the negative control value using StatFlex Ver7.0 (Artec Inc., Osaka, Japan). In these analyses, a p value less than 0.05 was considered significant. Regression analysis of the viable cell counts was performed with the GraphPad Prism 6 software (GraphPad Software Inc., San Diego, CA, USA).