Rapid susceptibility profiling of carbapenem-resistant Klebsiella pneumoniae

The expanding global distribution of multi-resistant Klebsiella pneumoniae demands faster antimicrobial susceptibility testing (AST) to guide antibiotic treatment. Current ASTs rely on time-consuming differentiation of resistance and susceptibility after initial isolation of bacteria from a clinical specimen. Here we describe a flow cytometry workflow to determine carbapenem susceptibility from bacterial cell characteristics in an international K. pneumoniae isolate collection (n = 48), with a range of carbapenemases. Our flow cytometry-assisted susceptibility test (FAST) method combines rapid qualitative susceptible/non-susceptible classification and quantitative MIC measurement in a single process completed shortly after receipt of a primary isolate (54 and 158 minutes respectively). The qualitative FAST results and FAST-derived MIC (MICFAST) correspond closely with broth microdilution MIC (MICBMD, Matthew’s correlation coefficient 0.887), align with the international AST standard (ISO 200776-1; 2006) and could be used for rapid determination of antimicrobial susceptibility in a wider range of Gram negative and Gram positive bacteria.


Preparation of bacteria.
Fluids for bacterial preparations and acoustic flow cytometer operation were filtered at 0.1 µm prior to use to minimise particulate contamination. A 1 mL aliquot of bacterial suspension was centrifuged, washed and resuspended in 1 ml filtered Hank's Buffered Salt Solution (HBSS) and diluted in series to 1:1000. SYTO ® 9 stain working stock solution (1 µl) was added to the final dilution at a final concentration of 5 µM and incubated for 5 minutes before determination of bacterial count by flow cytometer (Attune, ThermoFisher, Eugene, OR, USA), which was used to prepare a standardised inoculum density for susceptibility testing.
An aliquot of bacterial suspension was added to 50 mL centrifuge tubes (Corning, New York) containing 9 mL of pre-warmed (37 °C) filtered MHB to produce a final density of 5 × 10 5 −1 × 10 6 bacteria per mL. This suspension was incubated at 37 °C with shaking at 100 RPM for 30 minutes to obtain an actively dividing culture. An aliquot of antibiotic working solution from the previously described dilution series (appropriate to the concentration tested) was then added to each tube before a further 30-minute incubation with shaking, during which time a microbroth dilution (MBD) plate was prepared for overnight incubation at 37 °C 20 . One millilitre of bacterial suspension from each antibiotic concentration was harvested by centrifugation at 7800 × g for 5 minutes, washed, resuspended and diluted 1:10 in filtered HBSS in a light-impermeable microcentrifuge tube, then stained with SYTO ® 9 at a final concentration of 5 µM, and incubated for 5 min before AFC sampling. Hoechst 33342 dye (NucBlue, Thermo Fisher Scientific, Eugene, OR, USA) and a SYTO ® 9/propidium iodide (PI) combination (Thermo Fisher Scientific) were used in a series of replicate experiments (data not shown).
Acoustic Flow Cytometer (AFC) operation and data analysis. The acoustic flow cytometer was calibrated at the beginning of each acquisition session in accordance with the manufacturer's instructions (ThermoFisher Scientific). Flow cytometer settings were: Forward Scatter (FSC) voltage 3100, FSC threshold 4 × 1000 AND, blue laser 1 (BL1 -530/30 nm) voltage 1900, BL1 threshold 1 × 1000 AND, high sensitivity, flow rate 25 µL/minute, and an acquisition volume of 125 µL. Acquisition halted after collection of 20,000 events across all gates, or after 3 minutes, with each sample acquired in technical triplicate.
Collected data were exported in the FCS 3.0 file format and analysed in Flow v10.0 (FlowJo LLC, Ashland, OR, USA) by a single user, blinded to the MIC results by broth microdilution (MIC BMD ).
Digital fluorescence microscopy. A 1 mL aliquot of growth from each antimicrobial concentration was harvested while conducting AFC measurements of antimicrobial-exposed bacteria, centrifuged at 7800 × g for 5 minutes, and resuspended in 10 µl of HBSS. A 0.1 µL aliquot of SYTO ® 9 was added to each tube (final concentration 50 µM) and incubated for 5 minutes. A 2 µl aliquot was placed on a poly-L-lysine slide, sealed beneath a coverslip, and observed at 60x magnification by digital fluorescence microscopy. Samples were observed on the EVOS-FL digital fluorescence microscopy platform (Thermo Fisher, Eugene OR), with a representative field of view captured for each sample across the antimicrobial dilution series. High-resolution images were acquired using a Nikon Ts2R Eclipse inverted digital fluorescence microscope (Nikon, Tokyo, Japan). Qualitative susceptibility testing. We sought to determine whether a qualitative susceptibility test could be developed using the FAST platform. Subsequent to quantitative MIC FAST determination, the flow cytometer data were re-analysed in silico to produce a limited sub-set for qualitative susceptibility determination. The six antibiotic dilutions (0 mg/L, 0.25 mg/L, 1 mg/L, 2 mg/L, 4 mg/L and 16 mg/L) most relevant to qualitative susceptibility assessment were used. In silico analysis was restricted to the first technical triplicate of each recorded sample. Gating strategies remained consistent, with the addition of a gate restricting analysis to only those events recorded in the first 60 seconds of acquisition. Isolates were defined as meropenem susceptible (S) or non-susceptible (NS) using EUCAST clinical breakpoints for Enterobacteriaceae (S ≤2 mg/L, NS >2 mg/L).

Statistical analysis.
Statistical software (Prism v 6.1, GraphPad, San Diego, CA, USA) was used to analyse both SIR categorization and quantitative MIC results. SIR results were analysed using a χ 2 format. Clinical laboratory test performance measurements (sensitivity, specificity, positive predictive value, negative predictive value, Matthews Correlation Coefficient) were used to assess the ability of the FAST method to correctly determine carbapenem susceptibility. The correlation between MIC BMD and MIC FAST was analysed by calculating Spearman's coefficient for non-parametric data. The MIC data were plotted on a log-log biaxial plot, using the microdilution results as the determinant.
Discrepancy investigation. We examined in more detail all isolates demonstrating anomalous S/NS categories (MIC BMD vs. MIC FAST ), or MIC discrepancies outside the accepted tolerance of the microbroth dilution assay (+/− one two-fold dilution step). These isolates were subcultured once per day for three days to exclude the possibility of low prevalence contamination of cryo-preserved stocks by bacteria other than Klebsiella species. Any isolates displaying variable colony appearance on solid media had each observed colony morphotype sub-cultured separately, and their identity verified. We reconfirmed the molecular basis of carbapenem resistance using mechanism-specific PCR assays 5,8 . In cases where contaminants or complex resistance mechanisms were identified, isolates were subjected to a further round of FAST following determination of identity and molecular basis of resistance.  Subpopulation investigation. Populations observed on bi-variate flow cytometry plots that seemed to segregate into two populations across a dilution series were observed in many isolates during validation. To investigate this, we selected a demonstrative example (K16, an IMP-4 producing isolate) and subjected it to our FAST assay. We referred to this as "Day one". One mL of the 4 mg/L meropenem exposed culture was harvested, washed in fresh HBSS to remove the presence of meropenem, and inoculated into fresh TSB to provide input for a second round of FAST on "Day two". MIC FAST , population shapes, and progression to susceptibility-associated signature were compared between both experiments.

Results
A rapid flow-assisted susceptibility test for meropenem. Following our initial development process, we developed a new method ( Fig. 1) by which susceptibility to meropenem can be assayed in K. pneumoniae.
Using an acoustic flow cytometer to obtain optimal resolution of small particles, and a nucleic acid intercalating fluorophore to discriminate bacterial events from background debris, optimal results were achieved with SYTO ® 9. We used changes in size, shape, cytoplasmic volume and overall event numbers to predict susceptibility to meropenem in 1 hour, and MIC in 3 hours.
Defining meropenem susceptibility by AFC. Susceptibility to meropenem was defined by careful pairing of observed shifts in FSC and BL1 fluorescence (530/30 nm -ideal collection for SYTO ® 9) in bi-axial AFC plots, and observation of bacterial structures consistent with meropenem compromise by fluorescence microscopy. Exposure of actively dividing meropenem-susceptible isolates to inhibitory concentrations of the drug has been demonstrated to produce a range of cellular morphotypes; cells elongate, swell, balloon, and eventually proceed to complete cell lysis as they become compromised 21 . When microscopy and biaxial AFC plots were compared, an increased prevalence of aberrant cell morphotypes (consistent with meropenem compromise) was found to correlate with an increase in FSC, increased BL1 fluorescence, and formation of populations that contour independently on biaxial plots. In a meropenem-susceptible isolate, these changes were observed at the lowest concentration tested ( Fig. 2A), whereas in an isolate with a raised meropenem MIC, these changes were not observed at concentrations below the MIC BMD (Fig. 2B). When a non-susceptible isolate was exposed to concentrations approaching or exceeding its elevated MIC BMD , we observed forward scatter and BL1 changes associated with susceptibility. We refer here to this progressive change in morphotype approaching, and exceeding the MIC BMD as the susceptibility-associated signature.   (Fig. 3A). Using the technical triplicate of the unexposed bacteria with the median BL1-H geometric mean fluorescence intensity, the auto gate tool (FlowJo) defined a gate that bounded all contoured events on a bivariate contour plot of FSC-H vs BL1-H at the 10% threshold (Fig. 3B). This gate and the events it bounded were referred to as the Unexposed Cell Morphotype (UCM), and the gate was then applied consistently to all samples across the antimicrobial agent dilution series. The absolute count of event numbers in this gate was calculated to give a comparable measure of UCM for each sample, standardised by volume (events/µL). Changes in the prevalence of morphotypes when bacteria were exposed to meropenem at concentrations approaching or exceeding the MIC BMD were evident as a distinct susceptibility-associated signature. Iterative comparisons between UCM event rates/µL in antibiotic-exposed samples and the unexposed control samples were used to determine the flow-associated susceptibility test MIC (MIC FAST ).

Prediction of MIC BMD by FC.
Our initial range-finding series demonstrated close correspondence between the meropenem concentrations that caused appearance of the susceptibility-associated signature in each of 10 isolates and their corresponding MIC BMD values. We compared numbers of events bounded by the UCM gate per µL (UCMµ) in antibiotic-exposed samples with unexposed control samples, with a particular focus on cell numbers falling into and out of gated regions in those samples displaying the susceptibility-associated signature. We observed that the flow cytometer results accurately predicted meropenem MIC BMD when a cut-off point was established as the first concentration in an antimicrobial dilution series in which two or more of the technical replicates had less than 30% of events falling into the UCM gate when compared to the unexposed control (Table 1 and Fig. 4A). We refer to this concentration as the MIC FAST . Paired MIC BMD and MIC FAST results for the entire isolate collection are shown in Table 2. There was a strong positive correlation between MIC BMD and MIC FAST across the entire isolate collection (Spearman r = 0. 913, p < 0.0001, sensitivity 1.00, specificity 0.90, positive predictive value 0.86, negative predictive value 1.00 and Matthew's correlation coefficient of 0.878) (Fig. 4B). MIC FAST determination required 158 minutes from actively growing culture (35 minutes of incubation, 12 minutes for manual handling, 108 minutes for data acquisition, and 3 minutes of data interpretation from a pre-prepared workspace template). Qualitative meropenem susceptibility was assessed for the entire isolate collection from the previously described data subset. Three isolates were incorrectly determined; two isolates (KS1, OXA-181, and 500638, pAmpC) were incorrectly categorised as susceptible despite being non-susceptible (MIC FAST 2, MIC BMD 4) however, this two-fold inter-test variation is within the accepted tolerance of the broth microdilution assay. Isolate K16 (IMP-4, MIC BMD 16; MIC FAST 2) was the subject of extensive further investigation. Despite these isolates, the FAST susceptible/non-susceptible threshold/interpretive criterion was highly concordant with broth microdilution-derived susceptibility (χ 2 = 37.03, df = 1, p < 0.0001, sensitivity 1.00, specificity 0.90, positive predictive value 0.875, negative predictive value 1.00 Matthew's correlation coefficient 0.887). Based on the conditions selected for the data set assembly, the theoretical time-to-result for this qualitative test was 54 minutes from actively dividing culture (35 minutes of incubation, 11 minutes for manual handling, six minutes for data acquisition, and two minutes of data interpretation from prepared workspace template).

FAST can be applied to other carbapenems.
To examine the applicability of the FAST method to other carbapenems, carbapenem-susceptible (ATCC 700603) and -resistant (ATCC BAA 1705) control strains of K. pneumoniae were exposed to analytical grade meropenem, imipenem, ertapenem, and therapeutic grade meropenem. There was no difference between S/NS categorisation between MIC BMD and qualitative FAST S/NS across all tested conditions (Table 3).

MIC BMD vs MIC FAST discrepancy analysis.
Only five of 48 (10.4%) isolates showed discrepancies between MIC BMD and MIC FAST that resulted in a different meropenem S/NS assessment. The first, isolate 374, was found to contain a low-prevalence Staphylococcus aureus contaminant. Analysing the pure K. pneumoniae growth produced perfect concordance between MIC BMD and MIC FAST . Three of these isolates, two with an OXA-48-family enzyme and one with an IMP-4 (3000763, KS11 and K23 respectively) had MIC BMD and MIC FAST values within the two-fold dilution tolerance of the BMD method, but straddled the EUCAST breakpoint. This is an error of classification, not an inaccuracy of our method. The remaining isolate (K16, IMP-4) initially produced a MIC BMD of 16 mg/L and a MIC FAST of 2 mg/L. This was sub-cultured once again to check for purity, whereupon smooth and rough colony variants were observed. Retesting of each colony

Identification of subpopulation and persisting populations in isolates with discrepant MIC
results. On Day 1, isolate K16, classified as susceptible by MIC FAST , was found to contain a population of bacterial events consistent with unexposed cell morphotypes that persisted until 16 mg/L (Fig. 6 -Day 1). On Day 2 the bacterial population characteristics exhibited a different progression towards the susceptibility-associated signature across the same meropenem dilution range. Bacterial cells had a much higher forward scatter, without an associated BL1 increase on Day 2, and starting at 4 mg/L a subpopulation of cells again became evident ( Figure  Day 2). Subpopulations such as these have been observed across approximately one third of isolates assayed in our collection (n = 17).

Discussion
Antimicrobial susceptibility profiling of carbapenem-resistant K. pneumoniae by acoustic flow cytometer predicted both quantitative (MIC) and qualitative (susceptible/non-susceptible) carbapenem susceptibility. While flow cytometry has been used for antimicrobial susceptibility testing before [11][12][13][14][15][16][17][18] , our FAST assay is the first reported description of a validated method to generate a clinically-relevant quantitative end-point. Furthermore, our rapid phenotypic determination of antimicrobial susceptibility accurately predicts the qualitative result, and is therefore a significant step towards alignment of laboratory testing with clinical decision timelines. Broth microdilution is too labour-intensive for use in most clinical laboratories, which favour other methods of susceptibility determination. We present performance statistics for our qualitative susceptibility test but to demonstrate the power of single-cell level analysis rather than to expect immediate adoption of this assay in the clinical laboratory. To the prescribing physician, rapid qualitative susceptibility represents an ability to align the decision-making process of antibiotic prescribing to the best-practice ideals of effective anti-microbial stewardship 7 .
The FAST method is suitable for application as a rapid method to determine carbapenem resistance phenotype on the grounds of a strong correlation between MIC BMD and MIC FAST . MIC FAST follows a pre-determined heuristic to generate quantitative results, rather than relying on potentially user-biased subjective end-points. Our use of workspace templates allowed replication of results by non-specialists after minimal instruction by a skilled operator using a proprietary software package (FlowJo). Furthermore, any flow cytometry software capable of generating a contouring output should be suitable. The FAST assay is underpinned by the reproducible flow cytometry model of a complex series of physiological interactions we established. Forward Scatter (FSC) is often used a surrogate for particle size, but this oversimplifies the dynamic behaviour of non-spherical particles 17 . There is much more information in this single measurement than the size and orientation of a particle passing through the flow cytometer. For example, changes in granularity and autofluorescence profiles also alter the absolute numbers of photons reaching the FSC detector, and in similar manner, photons absorbed and emitted by fluorescence signals can alter FSC measurements 17,18 . Our choice of fluorescent dye (SYTO ® 9) ensured that measurements in the BL1 channel (530/30 nm) contained information on DNA content, cytoplasmic volume and autofluorescence. Observed staining intensity profiles from a rigorously controlled experimental method offer additional insight into physiological properties such as membrane permeability and dye molecule efflux 14 . Isolates with the osmoporin ompK36 third eyelet insertion mutation (ins AA 134-135 GD 5, 6 ) displayed a reduced BL1 intensity.
This porin mutation excludes positively charged compounds such as SYTO ® 9 22 and has been shown to correlate with high-level meropenem resistance 23,24 . The consistency of our observations across a collection of isolates from such diverse geographic origins and resistance mechanisms supports a conserved bacterial physiology.
The physiological response we detected by the FAST method after antimicrobial exposure resembles the range of carbapenem-induced morphotypes described previously 21 . Arrested cell division after inhibition of penicillin-binding proteins 25,26 leads to an overall increase in cellular DNA and increases the DNA-bound SYTO ® detectable in BL1. The overall decrease in cell numbers by fluorescence microscopy and flow cytometry, and the corresponding increase in flow cytometer event populations with low forward scatter and varied BL1, is likely to reflect mixed cell debris from bacterial cell lysis during antimicrobial exposure. The broth microdilution MIC method relies on a subjective end-point and requires extended incubation 27 , allowing persistence of resistant sub-populations after inhibition of the susceptible majority of bacteria [28][29][30] . The FAST method measures the resistance phenotype of all bacterial cells in each aliquot, and adds to the evidence that carbapenem-resistant Enterobacteriaceae are phenotypically heterogeneous [28][29][30] . We postulate that broth microdilution over-simplifies the test method and overestimates the dose required to demonstrate antimicrobial efficacy. Highly resistant  bacterial sub-populations have been implicated in failed meropenem monotherapy before 28,31 . These bacteria may respond to meropenem combination therapy provided sufficient breakthrough growth has not occurred 28,31 . Identification of these features of bacterial susceptibility in a shorter time could become the basis of more timely antimicrobial treatment guidance 28,29,31 . While our method eliminates the necessity of the secondary culture step required for either broth microdilution or other growth-dependent quantitative susceptibility determination 27 , further advances are needed to purify bacteria directly from patient samples so that laboratory results are available to the physician within a shorter time frame, particularly for patients with sepsis and other severe infections. Discrepancies of ≥2 two-fold dilutions were observed between MIC BMD and MIC FAST for pAmpC-and IMP-4-producing isolates. These types of resistance cause inducible meropenem resistance 30,32 . Induction of meropenem-resistant pAmpC-producing K. pneumoniae has been demonstrated after accumulation of transpeptidation by-products in the cytosol [32][33][34][35] . Selection of low-prevalence sub-populations with constitutive AmpC can also lead to rapid, time-dependent shifts in the overall resistance phenotype 33,34 . Induction of expression does not occur within 30 minutes of antimicrobial exposure and may therefore contribute to discrepancies between MIC BMD and MIC FAST . In the case of IMP-4-producing isolates, high-level induced meropenem resistance is thought to be caused by intrinsically-resistant sub-populations 30 . The presence of persistent bacterial populations at higher meropenem concentrations in the UCM gate indicates a sub-population of inducible IMP-4-mediated meropenem resistant cells. Identification of inducible resistance is a challenge with any antimicrobial susceptibility test, but determination of the result shortly after the start of antimicrobial exposure should reduce the complex effects of prolonged antimicrobial exposure and improve the accuracy of test endpoints. Figure 5. Differences in MIC FAST were observed between colony variants of K. pneumoniae isolate K16. When subculturing IMP-4 producing K. pneumoniae isolate K16, a rough and smooth colony variant was observed. The smooth colony variant produced an MIC FAST of 2 mg/L. The rough colony produced an MIC FAST of 64 mg/L and, at 2 mg/L, was observed by AFC to contain a population consistent with a non-susceptible phenotype. Both MIC FAST results were concordant with the MIC BMD results.
The precision of our method for determining quantitative and qualitative susceptibility to meropenem in Klebsiella species compares favourably with the current international standard, while returning results in 1 (qualitative) to 3 (quantitative) hours after receipt of primary culture -in most cases a full 24 hours earlier than current standard practice. Transition from subjectively interpreted end-points to objectively-generated, single bacterial cell analysis can improve the resolution of an antimicrobial susceptibility test, without sacrificing either precision or specificity. Figure 6. Resistant sub-populations were observed in K. pneumoniae isolate K16 across two days of selective passage and FAST: Day One -IMP-4 producing K. pneumoniae isolate K16 was found, at 2 mg/L meropenem, to contain a minority population of cells with a phenotype consistent with unexposed cells (remaining within the Unexposed Cell Morphotype gate -indicated by arrow). This subpopulation persisted, at a diminished frequency, at 16 mg/L meropenem while the majority of cells display a compromised phenotype (shifted outside the gate). Day 2 -The 2 mg/L culture of K16 from Day 1 was subcultured and subjected to FAST on the following day. The isolate displayed an increased MIC (4 mg/L), delayed progression to the emergence susceptibility-associated signature, with most events consistent with a non-susceptible phenotype at 2 mg/L. Most events at 16 mg/L were consistent with a susceptible phenotype, however a small subpopulation remained inside the Unexposed Cell Morphotype gate.