Bacterial strains and plasmids used in the study are listed in Table 1. Carbapenem-susceptible (ISP50) and resistant (IRP41) P. aeruginosa isolates characterized in this study were obtained from sputum samples of the same patient with acute exacerbation of chronic bronchitis before and after treatment with biapenem for 4 days (dosage at 0.3 g × 2/day) at the Nankai University Affiliated Hospital, Tianjin, China. The 16S rRNA encoding gene was amplified (primers listed in Supplementary Table S1) and sequenced to identify the species of these two isolates (Spilker et al., 2004). Random amplified polymorphic DNA (RAPD) typing was carried out using primer 272 as described previously (Mahenthiralingam et al., 1996). Antibiotic susceptibility test was conducted using the disk diffusion method according to the manufacturer’s recommendations, and minimum inhibitory concentration (MIC) of antibiotics was determined by the two-fold serial dilution method. Susceptibility was interpreted according to the Clinical and Laboratory Standards Institute guidelines (CLSI 2011–2018).

For overexpression of oprD, a 1,802 bp oprD-containing fragment with its putative Shine-Dalgarno (SD) sequence was PCR amplified using ISP50 and IRP41 genomic DNA as templates (the used primers are displayed in Supplementary Table S1). The PCR products were digested with BamHI and HindIII, and then ligated into a shuttle vector pUCP24, resulting in pUCP24-oprD_ISP50 and pUCP24-oprD_IRP41, respectively. Constructs of pUCP24-ampC, pMMB-ampR_ISP50, pMMB-ampR_IRP41, pUCP24-ldcA_ISP50, pUCP24-ldcA_IRP41 and pUCP24-pbpC were all generated using similar procedures.

For deletion of the ldcA gene, a 836 bp fragment immediately upstream of the ldcA start codon and a 913 bp fragment downstream of the ldcA stop codon were PCR amplified, digested with EcoRI-BamHI and BamHI-HindIII, respectively. The two fragments were then ligated into pEX18Tc that was digested with EcoRI and HindIII, resulting in pEX18-ldcA. A pEX18-ampR was constructed by similar manipulations. Gene knockouts in P. aeruginosa were carried out by conjugal transfer of these plasmids followed by selection for single crossovers and then double crossovers as previously described (Schweizer, 1992).

Multilocus sequence typing (MLST) was performed following a previous description (Curran et al., 2004) with minor modifications to confirm the allelic profiles of the two isolates. Briefly, genomic DNA was isolated from overnight bacterial culture with a DNA purification kit (Tiangen Biotech, Beijing, China) for use as PCR template. The internal fragments of seven housekeeping genes (acsA, aroE, guaA, mutL, nuoD, ppsA and trpE) were amplified by PCR and sequenced using primers described previously (Supplementary Table S1) (Curran et al., 2004). Gene sequences were then submitted to the P. aeruginosa MLST database¹ for assignment of allelic numbers. A sequence type (ST) was assigned to each isolate by combining the seven allelic numbers.

The PAE-MHT assay was performed as described previously (Pasteran et al., 2011) with minor modifications. Briefly, an inoculum of Klebsiella pneumoniae ATCC 700603 was adjusted to a 0.5 McFarland turbidity standard (Mcfarland, 1907) followed by 10-fold dilution with sterile saline, and then inoculated onto the surfaces of Mueller-Hinton agar (Oxoid) plates (diameter, 10 cm) by swabbing. The plates were allowed to stand at room temperature. After 10 min, a disk of 10 μg imipenem/meropenem (Oxoid, United Kingdom) was placed in the center of each plate. Subsequently, one colony of each P. aeruginosa strain, grown overnight on LB agar plate, was inoculated onto the plate in a straight line from the edge of the disk to the periphery of the plate using a 1-μl loop (BOOPU Biological Technology, Changzhou, China). Presence of growth of the K. pneumoniae ATCC 700603 toward imipenem/meropenem disk was interpreted as carbapenemase positive.

Bacterial overnight culture was inoculated into fresh LB medium (1:50 dilution) and grown to an optical density of 1.0 (wavelength of 600 nm). Total RNA was extracted using an RNAprep Pure Cell/Bacteria Kit (Tiangen Biotech, Beijing, China) and reverse transcribed to cDNA with random primers and PrimeScript Reverse Transcriptase (Takara). The cDNA was mixed with indicated primers (Supplementary Table S1) and SYBR premix Ex Taq II (Takara), and then quantitatively amplified in a CFX Connect Real-Time system (Bio-Rad, United States). A 30S ribosomal protein encoding gene rpsL was used as an internal control.

For RNAseq, total RNA was extracted as described above, quantified and qualified by Agilent 2100 Bioanalyzer (Agilent Technologies, Palo Alto, CA, United States), NanoDrop (Thermo Fisher Scientific Inc.) and 1% agarose gel. Then, the rRNA was depleted using Ribo-Zero rRNA Removal Kit (Bacteria, Illumina). The mRNA was then fragmented and reverse-transcribed. The double-strand cDNA was purified, ends repaired and ligated with adaptors. After 11 cycles of PCR amplification, the PCR products were cleaned, validated with an Agilent 2100 Bioanalyzer (Agilent Technologies, Palo Alto, CA, United States), and quantified by Qubit 2.0 Fluorometer (Invitrogen, Carlsbad, CA, United States). The resulting libraries were subjected to sequencing using an Illumina HiSeq 2500 platform with a 2 × 150 paired-end configuration.

Sequence reads were mapped onto PAO1 reference genome (NC_002516.2) via software Bowtie2 (v2.1.0). Gene expression levels were analyzed with HTSeq (v0.6.1p1). Differentially expressed genes were identified by the DESeq Bioconductor package, with the fold-change larger than 2 and P-value no more than 0.05 as cutoff values.

Bacterial genomic DNA was extracted with DNA purification kit (Tiangen Biotech, Beijing, China). Fragments smaller than 500 bp were obtained from 200 ng genomic DNA by sonication (Covaris S220), followed by end treatment and adaptors ligation. Adaptor-ligated DNA fragments of about 470 bp were recovered using beads and then PCR amplified for six cycles, and the PCR products were cleaned up using beads, validated using a Qsep100 (Bioptic, Taiwan, China), and quantified by Qubit3.0 Fluorometer (Invitrogen, Carlsbad, CA, United States). Sequencing was carried out using a 2 × 150 paired-end (PE) configuration on an Illumina Hiseq instrument according to manufacturer’s instructions (Illumina, San Diego, CA, United States). The data were aligned to the PAO1 reference genome (NC_002516.2) via software BW2 (version 0.7.12). Single nucleotide variation (SNV) or InDel mutation were detected using the software Samtools (version 1.1) and the Unified Genotyper module from GATK.

The full-length oprD gene with its 105 bp upstream and 209 bp downstream region was amplified and sequenced using primers listed in Supplementary Table S1. The oprD gene sequence from each isolate was aligned with the oprD sequence of the reference strain PAO1². Analysis of ampR and ldcA were conducted with similar procedures.

Subcultured samples from equivalent number of bacterial cells of an optical density of 1.0 (wavelength of 600 nm) were separated on a 12% SDS-PAGE. The proteins were then transferred onto a PVDF (polyvinylidene difluoride) membrane and probed with a mouse monoclonal antibody against His-tag (Cell Signaling Technology, United States) and the RNA polymerase alpha subunit RpoA (Biolegend). Signals were generated using the ECL-Plus kit (Millipore) and detected by a Biorad imager (ChemiDocXRS).

Biofilm production was measured as described previously (Li et al., 2013). Overnight culture of each bacterium was diluted 50-fold in LB broth and incubated in each well of a 96-well plate at 37°C. After 24 h, each well was washed with phosphate buffered saline (PBS, 137 mM NaCl, 10 mM Na₂HPO₄, 2.7 mM KCl, pH 7.4) for three times, stained with 0.1% crystal violet, and then washed three times with PBS. Then, 0.2 ml of decolorizing solution [methanol (4): acetic acid (1): water (5)] was added into each well and incubated 10 min at room temperature. Each sample was then subjected to measurement for OD₅₉₀ using Varioskan Flash (Thermo Scientific, Netherlands).

Graphpad software was used to perform the statistical analyses. The real-time qPCR results were analyzed by Student’s t-test (two-tailed).

Sputum samples from a patient with acute exacerbation of chronic bronchitis were collected before and after treatment with biapenem for 4 days (dosage at 0.3 g × 2/day). Two mucoid isolates were selected, one before the biapenem treatment which was sensitive to imipenem (ISP50), and the other one after the biapenem treatment which showed resistance against the imipenem (IRP41). PCR amplification of the 16S rDNAs as well as their sequence analysis indicated that both of them were P. aeruginosa (Supplementary Figure S1A). The MLST analysis of both isolates revealed an allelic profile for acsA, aroE, guaA, mutL, nuoD, ppsA and trpE as 17, 5, 12, 3, 14, 4, 7, respectively, corresponding to the same sequence type, ST244. RAPD typing was further performed on the IRP41 and ISP50 isolates, which also suggested that they were of the same clone (Supplementary Figure S1B).

Imipenem susceptibilities of IRP41 and ISP50 were found different based on the preliminary analysis with VITEK automatic microbe analysis instrument (data not shown). Therefore, we further examined the imipenem susceptibilities by both MIC and disk diffusion methods, with the results shown in Table 2 and Supplementary Figure S2. Initial isolate (ISP50) was found to be susceptible to the imipenem (53.3 mm/diameter of inhibition zone), while later isolate (IRP41) was resistant to the imipenem (10.7 mm/diameter of inhibition zone), with a 32-fold increase in MIC of imipenem. Although both strains were susceptible to meropenem (another carbapenem antibiotic) according to the Clinical and Laboratory Standards Institute guidelines (CLSI 2011–2018), IRP41 displayed a smaller diameter of inhibition zone (24 mm) compared to that of ISP50 (44 mm) (Supplementary Figures S2A–D). In addition, MIC of meropenem/biapenem in IRP41 showed two-fold/eight-fold increase compared to ISP50 (as displayed in Table 2).

Since the production of horizontally acquired carbapenem-hydrolyzing enzymes has been defined as a contributory factor in clinical P. aeruginosa isolates with carbapenem resistance (Potron et al., 2015), we tested if the IRP41 produced carbapenemase using a PAE-MHT assay. As the results shown in Supplementary Figures S2E,F, compared to the PA-NK41, which produces a KPC2 carbapenemase, no carbapenemase activity was observed in the IRP41 or ISP50 strain, suggesting that the carbapenem resistant phenotype of isolate IRP41 is not due to the presence of any horizontally acquired carbapenemases.

To elucidate the mechanism of reduced susceptibility to carbapenem in isolate IRP41, we compared the global gene expression profiles between strains ISP50 and IRP41. Expression levels of 284 genes were altered comparing the two strains (Table 3 and Supplementary Table S2). Among them, ampC gene, encoding the β-lactamase, displayed a 285-fold higher mRNA level in IRP41 than that in ISP50 (Table 3). To confirm the up-regulation, we further examined the mRNA levels by real-time qPCR. As shown in Figure 1A, consistent with the RNAseq result, the mRNA level of ampC was 413-fold higher in the strain IRP41 than that in ISP50. To further determine if the increased ampC expression contributed to the reduced susceptibility to carbapenem in IRP41, the ampC gene was overexpressed in the ISP50 strain background and the MIC of carbapenems was examined. As shown in Supplementary Table S3 and Supplementary Figure S3A, overexpression of the ampC (1,239-fold increase compared to that in ISP50/pUCP24) reduced the susceptibility of ISP50 to imipenem/meropenem/biapenem, with 16/2/4-fold increase in MIC. These results suggested that increased expression of the ampC indeed contributed to the decreased susceptibility of IRP41 to carbapenems, at least partially.

Besides the ampC, a pbpC gene, encoding penicillin binding protein 3a, displayed an eight-fold up-regulation in IRP41 (Table 3). Real-time qPCR confirmed the increased expression of the pbpC in IRP41 (Figure 1B). However, overexpression of the pbpC (Supplementary Figure S3B) in ISP50 (ISP50/pUCP24-pbpC) had no effect on the bacterial MIC against imipenem/meropenem/biapenem (Supplementary Table S3), although it restored the MIC of carbenicillin in PA14pbpC::Tn to that of wild type PA14 strain (Supplementary Table S3). These results indicated that the elevated pbpC expression did not contribute to the carbapenem resistance in IRP41.

Our RNAseq analysis also revealed that armR, encoding MexR antirepressor (ArmR), was up-regulated five folds in IRP41. It had been demonstrated that MexR represses the expression of mexAB-oprM efflux pump (Daigle et al., 2007) whose over-expression confers meropenem resistance in P. aeruginosa (Okamoto et al., 2002). However, RNAseq results showed no significant difference in the transcriptional levels of the mexAB-oprM between the two strains, and this is confirmed by real-time qPCR (Figures 1C,D). Further sequence analysis showed that no mutation happened in the mexR, armR or mexR–mexA intergenic region of the IRP41 and ISP50 strains (data not shown).

To further elucidate the molecular mechanisms of the decreased susceptibility to carbapenem in IRP41, genome reference mapping was performed to identify the accumulated mutations in the genomes of IRP41 and ISP50 in comparison with PAO1 reference genome (see text footnote 2). Between strains ISP50 and IRP41, there are 84 frameshifts (deletion/insertion) and 377 nonsynonymous single nucleotide variations (SNV) (including 17 early stops) (Supplementary Table S4). Among them, a G831A substitution in the oprD gene resulted in a premature termination of the OprD in the IRP41 strain. To confirm the result, oprD genes were PCR amplified from the genomic DNA of strains ISP50 and IRP41. Sequence analysis of the amplicons revealed that oprD in ISP50 was the same as that of PAO1, while the oprD from IRP41 showed a G831A substitution, which confirmed the reference mapping results. To assess if the G831A substitution in oprD contributed to the reduced susceptibility of the IRP41 to carbapenems, the oprD gene from both ISP50 and IRP41 were expressed in the IRP41 strain background. As can be observed in Table 2, introduction of the oprD_ISP50 reduced the MIC of imipenem/meropenem/biapenem in IRP41, with 16/2/2-fold decrease, while the oprD_IRP41 did not. These results indicated that the G831A substitution in oprD indeed disrupted its function and lead to the reduced susceptibility to carbapenem in strain IRP41.

AmpR is a transcriptional regulator encoded immediately upstream of the ampC, regulating the ampC expression (Kong et al., 2005). The genome reference mapping result revealed that the ampR sequence of IRP41 is identical to that of PAO1, while in strain ISP50 an “A” was substituted by “G” at 821^th position of the ampR gene, resulting in an E274G substitution. These results were further confirmed by sequencing of the PCR amplicons. To test if the nonsynonymous mutation in ampR of ISP50 resulted in a lower level of ampC expression, both ampR genes from IRP41 and ISP50 were cloned and expressed in the background of ISP50. As shown in Supplementary Figure S3C and Supplementary Table S3, expression of the ampR_ISP50 conferred ISP50 strain a higher ampC mRNA level and increased MIC against ampicillin compared to that of ampR_IRP41, although it had no observable influence on the MIC of imipenem/meropenem/biapenem. The ampR_ISP50 and ampR_IRP41 were further introduced into PAO1ΔampR, and their effects on MICs to imipenem, meropenem, biapenem and ampicillin in PAO1ΔampR were examined. As shown in Supplementary Table S3, both ampR_ISP50 and ampR_IRP41 restored the MIC against ampicillin and biapenem in PAO1ΔampR to that in wild type PAO1 strain, while in contrast to ampR_IRP41, the ampR_ISP50 conferred PAO1ΔampR a repeatable two-fold increase in MIC against imipenem compared to that of wild type PAO1. Unexpectedly, no influence on MIC of meropenem was observed in PAO1ΔampR. These results suggested that A821G SNV of ampR in fact confers slightly higher ampC expression-activating capacity than the wild type AmpR, thus resulting in higher, rather than lower ampC in ISP50.

ldcA (PA5198) encodes a cytosolic LD-carboxypeptidase which removes the terminal amino acid D-alanine from the tetra peptide of the peptidoglycan (Korza and Bochtler, 2005). Our genome reference mapping and PCR amplicons sequencing results revealed that a C was deleted at 445 bp of the ldcA gene in strain IRP41, leading to a frameshift mutation after R149 of the LdcA. To test if this frameshift mutation in ldcA contributed to the elevated expression of the ampC in IRP41, we expressed the ldcA_IRP41 or ldcA_ISP50 (same as ldcA of PAO1) in the IRP41 background. As shown in Figure 2A, the relative mRNA level of ampC in IRP41/pUCP24-ldcA_ISP50, but not in IRP41/pUCP24-ldcA_IRP41, was restored to that in ISP50/pUCP24, and the MIC to imipenem/meropenem/biapenem in IRP41/pUCP24-ldcA_ISP50 was decreased by two folds (Table 2). To confirm the expression level of ampC gene, a C-terminal His-tagged ampC driven by its native promoter was integrated into the genomes of ISP50 and IRP41. As shown in Figure 2B, the AmpC-His level was much higher in IRP41 than that in ISP50, and overexpression of ldcA_ISP50, but not ldcA_IRP41, reduced the amount of AmpC-His in the IRP41 to almost that of the ISP50. These results demonstrated that frameshift-mutation of the ldcA in IRP41 contributes to the increased expression of the ampC in IRP41.

To further confirm the effect of ldcA on ampC repression, the whole ldcA open reading frame was deleted in PA14 and PAO1 backgrounds, resulting in PA14ΔldcA and PAO1ΔldcA, respectively. Similar to previous report in E. coli (Templin et al., 1999), both of the ldcA deletion mutants showed a reduced growth rate and proneness to lysis during stationary growth phase (Supplementary Figure S4). Interestingly, the expression levels of ampC displayed 2.6-fold and 4.0-fold increase in the ΔldcA mutants compared to their parental strains PAO1 and PA14, respectively (Figure 3A). Accordingly, the MIC of ampicillin in PAO1ΔldcA and PA14ΔldcA displayed a two-fold and four-fold increase compared to their respective parental strains (Table 2). We further deleted the ldcA gene from the ISP50 strain. Consistent with the above observations, the expression level of ampC in the ISP50ΔldcA increased 1.8-fold compared to that in the ISP50 strain (Figure 3B). The difference in fold increase of ampC mRNA levels between ISP50 and IRP41 (413 folds) vs. between ISP50 and ISP50ΔldcA (1.8 folds) suggested that other factors also contributed to the increased expression of the ampC in IRP41.

In addition, we also examined the effect of ldcA mutation on MICs against piperacillin, ceftazidime, aztreonam, ciprofloxacin and tobramycin. Except for a two-fold decrease against ceftazidime and meropenem in PAO1ΔldcA, no MICs change was observed for piperacillin, aztreonam, ciprofloxacin and tobramycin in ldcA mutants (Table 4). However, we didn’t obtain the IRP41ΔldcA, although deletion of ldcA in IRP41 strain had been tried for several times. We assume that it may be due to presence of some unknown factors to prevent the recombination in IRP41.

Since AmpR is the regulator of ampC, and its 274^th amino acid is different between IRP41 and ISP50, we further compared the ampC expression levels and the MIC to ampicillin/imipenem/meropenem/biapenem between ISP50ΔldcA expressing AmpR_ISP50 and AmpR_IRP41. As shown in Supplementary Figure S3D, consistent with that in the ISP50 strain, expression of AmpR_ISP50 in ISP50ΔldcA (ISP50ΔldcA/pMMB-ampR_ISP50) resulted in a higher mRNA level of ampC compared to that of ISP50ΔldcA/pMMB-ampR_IRP41, as well as a two-fold increase in MIC of ampicillin (Supplementary Table S3). Similar to that in ISP50 strain, overexpression of both AmpR_IRP41 and AmpR_ISP50 showed no influence on MIC of imipenem/meropenem/biapenem in the ISP50ΔldcA strain. Again, these are consistent with the earlier observation that the AmpR point mutation in ISP50 confers a higher transcriptional activator activity on ampC.

Furthermore, as the expression level of pbpC is much higher in IRP41 than that in ISP50, we overexpressed pbpC in the ISP50ΔldcA background to see if the elevated pbpC expression had any effect on the ampC mRNA levels. However, the mRNA level of ampC was not affected by the overexpression of pbpC in ISP50ΔldcA (data not shown), consistent with such null effect observed in the ISP50 background (Supplementary Table S3 and Supplementary Figure S3E).

Our RNA-seq result also revealed an increased expression of pslD-pslL operon (Table 3) which is involved in biofilm formation. Accordingly, we examined and compared the biofilm formation between IRP41 and ISP50 strains. As the results shown in Figure 4, the IRP41 strain produced more biofilm than the ISP50 strain.