Sixty-eight identified KPC-producing K. pneumoniae strains were isolated from different patients during the period June – December 2014 from the department of clinical microbiology in two tertiary hospitals in Beijing, China. Escherichia coli ATCC25922 was used as a reference strain. All strains were identified using the Vitek^® 2 Compact System (bioMérieux, Marcyl’Étoile, France).

Tigecycline and colistin standards were purchased from Sigma-Aldrich (St Louis, MO, USA). Amikacin standard was obtained from the National Institute for the Control of Pharmaceutical and Biological Products (NICPBP, Beijing, China). All the agents were prepared as fresh stock solutions in sterile distilled water on the day of use.

This study was performed in a standard Microbiological Lab in our hospital. We handle the clinical isolates according to the WHO Laboratory Biosafety Manual (WHO, 2004).

The MICs of tigecycline, colistin, and amikacin were determined by the agar dilution method according to CLSI (2014) guidelines. In brief, Mueller Hinton agar (Difco, Franklin Lakes, NJ, USA) plates containing a series of twofold concentration increments of each agent were prepared according to specifications. Then, approximate 10⁴–10⁵ CFU bacterial cells were inoculated by an autoclaved replicator and incubated at 35°C in ambient air for 20 h. The MIC was defined as the lowest drug concentration that inhibited the visible growth of colonies. All MIC determinations were conducted at least in triplicate on separate days. The MIC breakpoints of susceptibility for the three antibiotics were interpreted in accordance with the European Committee on Antimicrobial Susceptibility Testing standards (EUCAST): ≤1 mg/L for tigecycline, ≤2 mg/L for colistin, and ≤8 mg/L for amikacin (EUCAST, 2015).

In Brief, each strain underwent Enterobacterial Repetitive Intergenic Consensus-PCR (ERIC-PCR) typing as described previously (Smith et al., 2007; Ying et al., 2015).The primer pair ERIC1 and ERIC2 (Supplementary Table 1) were used to amplify intervening fragments of ERIC in the genomic DNA. Each reaction contained 2.5 μl of 10x PCR buffer, 2 μl dNTP mixture, 0.125 μl Taq enzyme, 0.5 μl each primer (10 μmol/L) and 1 μl DNA template, added to a final volume of 25 μl with ddH₂O. Thermal cycling was performed in an ABI 2700 thermal cycler (Applied Biosystems, Carlsbad, CA, USA): initial incubation at 94°C for 7 min; 30 cycles of 30 s at 94°C, 1 min at 52°C, and 8 min at 65°C; and a final incubation at 65°C for 10 min. The amplified products were resolved in 1.2% agarose gel electrophoresis stained with ethidium bromide. The data were analyzed with the software package BioNumerics 5.1 (Applied Maths, Austin, TX, USA) as previously described (Smith et al., 2007).

The MPCs of drugs alone and in combinations were determined by methods described previously (Zhanel et al., 2006; Ni et al., 2013). In brief, bacterial cells were grown overnight in fresh Mueller-Hinton broth (MHB) with vigorous shaking at 35°C in ambient air. Then the overnight cultures were diluted by 10-fold with MHB, and incubated at 35°C for 6 h. The growth was centrifuged (4000 × g for 5 min) to yield a high-density culture containing cells of ∼3 × 10¹⁰ CFU/ml. Fifty microliters of samples (approximate 1.5 × 10⁹ cells) were inoculated onto Mueller-Hinton agar plates (diameter, 90 mm) containing different concentrations of tigecycline, colistin and amikacin alone or in combination. For a given drug concentration, seven plates were used to ensure that the total number of cells tested was at least 10¹⁰. For counting the number of resistant colonies arising at a given concentration, another eight plates was inoculated with different number of bacterial cells (10³, 10⁵, 10⁷, and 10⁸). At the same time, the cell density of the culture was determined retrospectively by applying serial dilutions to drug-free agar plates. All plates were incubated at 35°C in ambient air for 3 days. The MPC was defined as the concentration that blocked growth when at least 10¹⁰ cells were applied to agar plates. The number of colonies arising in each plate of a given concentration was counted (if it was countable, typically,<500 CFU per plate) and the frequency of resistant mutants was defined as the total number of colonies in all plates with countable colonies divided by the total number of cells plated on these specific plates.

As mentioned above, the MSW of a single drug is a concentration range which approximately extends from the MIC to MPC, denoted by double-headed arrows in Figure 1A. The width of MSW can be expressed as a ratio (MPC/MIC), which is also termed as selection index (SI). As displayed in Figure 1B, no resistant mutants can recover above the MPC curve, and the MIC lines are approximate the lower boundary of the MSW in combination therapy (Zhao and Drlica, 2002). Therefore, the MSW of drug combinations (MSW_combination) can be measured as the area that extends from the MPC curve to the MIC lines of each drug (Zhao and Drlica, 2002).

For examining whether the arising colonies of each plate at a given concentration contain stable resistant mutants, two colonies of each plate were passaged five times on drug-free agars, and then the MICs were determined by the agar dilution methods. Real-time quantitative reverse transcription PCR (qRT-PCR) was used to detect the expression levels of the efflux pump gene acrB and the global transcriptional regulator ramA in tigecycline-resistant mutants, the expression level of pmrK that belongs to the pmrHFIJKLM operon in colistin-resistant mutants and the efflux pump gene acrD in amikacin-resistant mutants. The qRT-PCR was performed as previously described (He et al., 2015). Briefly, overnight bacterial cultures were diluted with MHB broth, and then were grown at 35°C with shaking of 200 rpm. Cell pellets were harvested at OD₆₀₀ = 0.6. The Purelink RNA Mini Kit (Ambion, Carlsbad, CA, USA) was used for extracting ribonucleic acid (RNA). The yield and quality of RNA were determined by a Nanodrop 2000C (Thermo, USA). Then the total RNA was reverse transcribed into cDNA using the PrimeScript RT Reagent kit (Tiangen, Beijing, China). Real-time qRT-PCR was performed using a LightCycler 480 II (Roche, Germany) with 40 cycles of 15 s at 95°C, 20 s at 52°C, and 30 s at 72°C, and SYBR Premix Ex Taq (TaKaRa, Dalian, China) was used to quantify the expression of the target gene. Expression of each gene was normalized to that of a housekeeping gene (rrsE). The parent strain was used as the reference strain. Besides, the mgrB gene of the colistin-resistant mutants and the aminoglycoside-modifying enzymes genes including aac (3)-I, aac (3)-II, aac (3)-III, aac (6)-Ib, aac (6)-II, ant (2″)-I, ant (3″)-I, and 16S rRNA methylase genes (armA, rmtA, rmtB, rmtC, rmtD, npmA) of amikacin-resistant mutants was amplified by PCR, and then the mgrB was sequenced by the Applied Biosystems 3730 sequence analyser (Applied Biosystems Inc., USA). In brief, the whole-cell DNA was extracted using the TIANcombi DNA Lyse & Det PCR Kit (Tiangen, Beijing, China) according to the manufacturer’s instructions. Each reaction contained 2.5 μl of 10x PCR buffer, 2 μl dNTP mixture, 0.3 μl Taq enzyme (5 U/μl; Tiangen, Beijing, China), 1 μl each primer (10 μmol/L) and 1 μl DNA template and 17.2 μl ddH₂O. The PCR method consisted of initial incubation at 94°C for 2 min; 35 cycles of 20 s at 94°C, 20 s at 55°C, and 30 s at 72°C; and a final incubation at 72°C for 10 min. The primers used in this study were shown in Supplementary Table 1.

MSW_combination of different drug pairs were compared and their differences were determined by One-way ANOVA and independent-sample t-test. We speculated the MSW_combination might be associated with the MSW of each single drug. Therefore, the relationship between the MSWs of single drugs and the MSW_combination was evaluated by Pearson’s correlation analysis. All analyses were performed with the software of SPSS 20.0. A P-value < 0.05 was considered statistically significant.

Among the 68 tested isolates, 85.3% (58 isolates) was susceptible to tigecycline, 97.1% (66 isolates) was susceptible to colistin, and 51.5% (35 isolates) was susceptible to amikacin. Thirty isolates were susceptible to all the three drugs, and they were mainly grouped into four clonal types (A, B, C, D) identified by ERIC-PCR (Figure 2). The MPC distributions of the 30 isolates are listed in Table 1. The MPC₉₀ for tigecycline, colistin, and amikacin were 16, >128, and 128 mg/L, respectively. One representative isolate of each clone type (Kp3 for A, Kp18 for B, Kp8 for C, and Kp23 for D) was randomly selected for evaluating the effect of antimicrobial concentrations on the selection of resistant mutants. As shown in Figure 3, a distinct plateau region was observed in the recovery curves of colistin. When the colistin concentration increased from 2- to 16-fold of the MIC, the frequency of resistant mutants did not decrease significantly.

The MPCs of three pairwise drug combinations were shown in Figure 4. The average MSW_combination of tigecycline-amikacin, tigecycline-colistin, and amikacin-colistin combinations for 4 strains were 11.99, 200.13, and 372.38, respectively. The MSW_combination of tigecycline-amikacin combination was much smaller than that of the other two combinations (P = 0.002). A correlation was found between the MSW_combination and the product of MSW of each single drug (MSW_A × MSW_B; r² = 0.68, P = 0.0153), and a strong correlation was found between log2 MSW_combination and log2 (MSW_A × MSW_B; r² = 0.96, P < 0.0001; Figure 5).

Combinations of 1MIC multiple amikacin and 1.5–2 MIC multiple tigecycline could restrict the recovery of resistant mutants on agar plates, while resistant mutants still recovered on plates containing 1 MIC multiple amikacin and 32 MIC multiple colistin in strain Kp3, Kp8, and Kp23 (Figure 4). As displayed in Table 2, 2 MIC multiple colistin in combination with 2 MIC multiple tigecycline or amikacin merely resulted in approximately 10-fold decrease in mutational frequency relative to their individual mutational frequencies. By contrast, combinations of 1MIC multiple tigecycline and 1MIC multiple amikacin could result in 1000- to 10000-fold reduction in mutational frequency for all strains.

The possible resistance mechanisms of recovered resistant mutants from the Kp3 isolate were further characterized. As shown in Figure 6, the average expression levels of ramA and acrB genes for 20 tigecycline-resistant mutants (Kp3-T1 to Kp3-T20) increased to 70.4 and 6.8 times compared with the parent isolate. The average expression level of pmrK genes for colistin-resistant mutants (Kp3-D1 to Kp3-D20) increased to 9.5 times compared with the parent isolate. But the expression levels of acrD genes were down-regulated in most amikacin-resistant mutants. Nine colistin-resistant mutants were detected to carry mutations in mgrB (GenBank accession numbers KX650149 for Kp3-D3; KX650150 for Kp3-D4; KX650151 for Kp3-D5 and Kp3-D8; KX650152 for Kp3-D6, Kp3-D13, Kp3-D15, and Kp3-D17; KX650153 for Kp3-D18), whereas PCR results of three mutants were negative. Because all the mutants were derived from the same parent isolate, we deemed that these three colistin-resistant mutants with negative PCR results had mutations in mgrB as well. Of the amikacin-resistant mutants, none were detected to harbor the screened aminoglycoside-modifying enzymes genes or 16S rRNA methylase genes.