The transcriptomic response of Acinetobacter baumannii to colistin and doripenem alone and in combination in an in vitro pharmacokinetics/pharmacodynamics model

Abstract

Objectives

Colistin remains a last-line treatment for MDR Acinetobacter baumannii and combined use of colistin and carbapenems has shown synergistic effects against MDR strains. In order to understand the bacterial responses to these antibiotics, we analysed the transcriptome of A. baumannii following exposure to each.

Methods

RNA sequencing was employed to determine changes in the transcriptome following treatment with colistin and doripenem, both alone and in combination, using an in vitro pharmacokinetics (PK)/pharmacodynamics model to mimic the PK of both antibiotics in patients.

Results

After treatment with colistin (continuous infusion at 2 mg/L), >400 differentially regulated genes were identified, including many associated with outer membrane biogenesis, fatty acid metabolism and phospholipid trafficking. No genes were differentially expressed following treatment with doripenem (C_max 25 mg/L, t_1/2 1.5 h) for 15 min, but 45 genes were identified as differentially expressed after 1 h of growth under this condition. Treatment of A. baumannii with both colistin and doripenem together for 1 h resulted in >450 genes being identified as differentially expressed. More than 70% of these gene expression changes were also observed following colistin treatment alone.

Conclusions

These data suggest that colistin causes gross damage to the outer membrane, facilitates lipid exchange between the inner and outer membrane and alters the normal asymmetric outer membrane composition. The transcriptional response to colistin was highly similar to that observed for an LPS-deficient strain, indicating that many of the observed changes are responses to outer membrane instability resulting from LPS loss.

Introduction

Acinetobacter baumannii, an opportunistic Gram-negative pathogen, has caused major infection outbreaks in hospitals worldwide.¹ The hallmark feature of A. baumannii is its ability to acquire multiple antibiotic resistance determinants, making it one of the most significant threats to the therapeutic success of current antibiotics.² As there are almost no new drugs under development against Gram-negative bacteria, MDR strains of A. baumannii have the potential to cause unmanageable infections. Indeed, pan-drug-resistant strains have already been identified^3–5 and A. baumannii has been recognized as one of the six top-priority dangerous organisms by the Infectious Diseases Society of America.²

Colistin has been reintroduced as a last-line therapeutic option for the treatment of infections caused by MDR A. baumannii. Colistin was abandoned as a standard therapeutic option in the 1970s and currently <2% of strains show colistin resistance.⁶ However, colistin-resistant strains are increasingly being identified and clinical treatment failure due to development of colistin resistance has been observed.^7–10 Colistin is bactericidal against many Gram-negative bacterial species, but the mechanism of killing has not been fully elucidated. The interaction of the positively charged colistin molecule with the negatively charged LPS on the surface of Gram-negative bacteria is clearly critical for its bactericidal action, as the known mechanisms of colistin resistance all involve changes to the LPS structure.^11–13 However, the initial interaction with LPS is unlikely to be the direct cause of cell death and it has been speculated that colistin is able to induce damage to the outer and inner membranes in such a way that cell death occurs without causing leakage of cellular contents.^14,15 Production of hydroxyl radicals appears to play a role in bacterial killing by colistin.¹⁶ In a colistin-resistant clinical isolate, the expression of a number of enzymes involved in oxidative stress response significantly decreased.¹⁷ Furthermore, an A. baumannii superoxide dismutase (SOD) mutant displayed increased (2-fold) susceptibility to colistin.¹⁸ The interaction of colistin with intracellular protein targets, such as the respiratory chain enzyme type II NADH-quinone oxidoreductase, may also play a role in bacterial killing.¹⁹

Recent pharmacokinetic/pharmacodynamic (PK/PD) studies have demonstrated that currently recommended dosage regimens of colistin are suboptimal.²⁰ Colistin monotherapy can lead to rapid emergence of resistance in Gram-negative pathogens including A. baumannii.²¹ Furthermore, the high risk of nephrotoxicity precludes substantial increases in colistin dose.^20,22 In order to improve the clinical effectiveness of colistin and minimize potential emergence of resistance, combination therapy with colistin has been recommended and examined. Of note, colistin and the carbapenem antibiotic doripenem have shown synergistic killing of a range of Gram-negative pathogens including A. baumannii.^23–27

This study was undertaken to investigate the transcriptomic response of A. baumannii to colistin, doripenem and a colistin/doripenem combination using an in vitro PK/PD model to mimic their PK in patients.²³ These analyses show that colistin treatment alters the expression of a very large number of genes, with many of these involved in the synthesis and transport of membrane components. Combined colistin and doripenem treatment resulted in gene expression changes that were similar to those observed following colistin treatment alone, indicating that the colistin responses predominate in the bacterial response to the combined antibiotics.

Materials and methods

Bacterial culture

The A. baumannii strain ATCC 19606 was routinely cultured on Mueller–Hinton (MH) agar plates (Oxoid, Australia) and incubated aerobically for 24 h at 37°C. A. baumannii ATCC 19606 is susceptible to both colistin and doripenem, with MICs of both antibiotics of 1 mg/L. Broth cultures were grown in cation-adjusted MH broth (Oxoid; 20–25 mg/L Ca²⁺ and 10–12.5 mg/L Mg²⁺) and incubated aerobically for 16–18 h at 37°C with constant shaking at 120 rpm. The colistin-resistant, LPS-deficient lpxA mutant strain 19606R (MIC >128 mg/L) has been described previously.¹² The 19606R strain was grown on MH agar supplemented with 10 mg/L colistin, followed by growth in MH broth without colistin immediately prior to transcriptomic analyses (see details below). The stability of the lpxA mutation was confirmed by direct patching onto MH agar with or without colistin (10 mg/L); the mutation was 100% stable in the absence of colistin over the duration of the transcriptomic experiments (data not shown).

Transcriptomic experiments

Experiments to examine the whole-genome transcriptional response of ATCC 19606 to colistin and doripenem alone and in combination were conducted using a one-compartment in vitro PK/PD model to mimic the PK of both antibiotics in patients.²³ Overnight cultures of A. baumannii were subcultured 1 : 100 into 10 mL of cation-adjusted MH broth and grown to an OD at 600 nm (OD₆₀₀) of ∼0.3. To achieve a starting inoculum of ∼10⁸ cfu/mL, 0.8 mL of the early logarithmic phase culture (above) was injected into the central reservoir of the PK/PD model and incubated until OD₆₀₀ reached ∼0.3 with the flow of medium temporarily halted (t = 0). Colistin and/or doripenem were then delivered into the central reservoir (and the medium container for colistin) to achieve constant concentration (0.2 or 2.0 mg/L for colistin) or declining concentration (C_max 25 mg/L with an elimination t_1/2 of 1.5 h for doripenem). Both profiles mirror the clinical PK of these antibiotics.²³ Control samples were grown without the addition of either antibiotic. Samples were collected at 0, 15 and 60 min for RNA isolation, measurement of viable cells and determination of antibiotic concentrations as described previously.^23,28 The 19606R strain was also treated with 2.0 mg/L colistin for 1 h as outlined above for the treatment of ATCC 19606.

Library preparation and RNA sequencing (RNA-Seq)

The construction of the cDNA library from purified RNA samples was performed using the TruSeq RNA Preparation Kit (Illumina). Ribosomal RNA (rRNA) was removed from the samples using the Ribozero kit (Epicentre). Approximately 300 ng of rRNA-depleted RNA was used as input, following the manufacturer's protocol (low throughput). Sequencing was conducted on either an Illumina MiSeq or an Illumina GA-IIx at the Micromon High-Throughput Sequencing Facility (Monash University). All GA-IIx samples were multiplexed in a single lane and sequenced as 100 bp paired-end (PE) reads. The MiSeq samples were sequenced as 50 bp single-end (SE) reads with all the first replicate set of samples multiplexed together and all the second replicate set of samples multiplexed together. As the MiSeq produces fewer data per lane than the GA-IIx, either four or three lanes were used (Replicate 1 and Replicate 2, respectively) for the multiplexed samples analysed on the MiSeq (Table S1, available as Supplementary data at JAC Online).

Bioinformatic analyses for transcriptomics

RNA sequence reads were independently aligned with the reference genome sequence of A. baumannii strain ATCC 19606 (NCBI accession NZ_GG704581.1) using SHRiMP software (available from the Computational Biology Laboratory, University of Toronto; http://compbio.cs.toronto.edu/shrimp/). For the 100 bp PE reads from the Illumina GA-IIx, only the left-end read was used for transcript mapping to be more consistent with the SE read data. The RNA sequence data from the biological replicates were analysed using the voom and limma methods.²⁹ Differentially expressed genes were defined as those with a change in expression of >2-fold (log₂ >1.0) with a false discovery rate (FDR) of <0.01. The differentially expressed genes identified in the draft ATCC 19606 genome were mapped to their closest orthologue (closest BLAST match) in the A. baumannii ACICU genome.³⁰ This allowed mapping of the data to annotated pathways and gene ontology (GO) terms via BioCyc.³¹ For identification of pathways and GO groups over-represented in each of the differentially expressed gene lists, enrichment analysis was conducted within BioCyc using Fisher's exact test and P < 0.01 to determine significance.

Data generated from two samples (one ATCC 19606 untreated sample and one ATCC 19606 sample treated with doripenem for 1 h) were discarded due to a technical problem that resulted in biased amplification of transcripts (data not shown). As a result, transcriptomic analysis was performed using three replicates for the ATCC 19606 untreated 1 h control (2 × MiSeq samples and 1 × GA-IIx sample) and from one sample for ATCC 19606 grown in the presence of doripenem for 1 h. The use of data from a single doripenem-treated sample was possible as the limma method allows for modelling of gene variation using the data from all other replicate samples (in all other conditions) with high confidence.²⁹ Indeed, the biological variability across replicates for all of the other conditions tested showed similar levels of variation, which supports the use of this approach. Furthermore, to empirically test the validity of this approach, we subsampled the colistin plus doripenem dataset and determined the number of differentially expressed genes when each of the single replicates was used or when both replicates were used together. Loss of the Replicate 1 data resulted in 54% fewer genes identified as differentially expressed while removal of the Replicate 2 data resulted in 33% fewer genes being identified as differentially expressed. Importantly, either 100% (Replicate 1) or 91% (Replicate 2) of the genes identified when only single-replicate data were used were also identified when both sets of data were examined. Thus, we are confident that the differentially expressed genes identified in the doripenem single-replicate data are robust, although it is possible that we have underestimated the true number of differentially expressed genes in this condition. The gene expression data in this study have been deposited in the NCBI Gene Expression Omnibus (GEO) database and are accessible through GEO series accession number GSE62794.

Quantitative reverse transcription PCR (qRT–PCR)

Primers for reverse transcription were designed using Primer-BLAST (www.ncbi.nlm.nih.gov/tools/primer-blast) (Table S2). Both RNA and DNA were quantified using a Qubit 2.0 Fluorometer (Life Technologies). Reverse transcription reactions were carried out on the same RNA preparations used for the RNA-Seq analyses. First-strand cDNA synthesis was carried out with 2–4 μg of RNA using 300 U of Superscript III Reverse Transcriptase (Invitrogen) and 300 ng of random hexamers in a 20 μL reaction at 42°C for 2.5 h. The cDNA samples were diluted 1 : 100 prior to qRT–PCR with the 20 μL PCR performed on a Mastercycler Ep Realplex (Eppendorf) using FastStart Universal SYBR Green Master Mix (Rox) (Roche) together with each gene-specific primer (1 μM) and 5 μL of diluted cDNA. The thermocycling parameters were 95°C for 2 min followed by 40 cycles of 95°C for 15 s and 60°C for 1 min. The concentration of each gene-specific template was determined by comparison with a standard curve constructed against known concentrations of ATCC 19606 genomic DNA. The gene expression levels of 11 housekeeping genes (gyrB, recA, rpoD, gpi, gltA, gdhB, cpn60, rpoB, pyrG, rplB and fusA) were analysed across all the RNA-Seq data and gyrB was chosen as the normalizer for all qRT–PCR reactions as it showed the least variability across conditions. Reactions without reverse transcriptase were used as negative controls for each reaction; no products were detected in these control reactions within 10 cycles of the reverse transcriptase-positive reactions. Melting curves were analysed for each reaction and these analyses indicated that all reactions amplified a single fragment. The relative standard curve method was used to determine gene expression levels. Transcript levels were determined from four different reverse transcription reactions and using three technical replicates per experiment. The details of the reporting of the qRT–PCR are compliant with the Minimum Information for Publication of Quantitative Real-Time PCR Experiments (MIQE) guidelines (http://miqe.gene-quantification.info/).³²

Results and discussion

Antibiotic treatments and transcriptional analyses using RNA-Seq

Experiments were performed using a one-compartment PK/PD model to mimic the PK of the various antibiotics in critically ill patients (Table 1). Cultures exposed to 0.2 mg/L colistin (sub-MIC concentration) all showed slight increases in viable cells (1.3–2.5-fold). Colistin treatment at 2 mg/L led to minor killing after 15 min (1.5-fold reduction) and significant loss of viability after 1 h (up to 10-fold). The largest loss of viability (∼20-fold) was observed in cultures treated with both colistin and doripenem for 1 h.

Table 1.

Culture conditions for transcriptomic samples

Strain (description)	Treatment	Time (min)	Sample	Log10 cfu (t = 0)	Log10 cfu (t = final)	Fold change (log10)
ATCC 19606 (WT, colistin susceptible)	untreated control	15	S1.1	7.70	8.03	0.3
			S1.2	7.95	8.07	0.1
	untreated control	60	S2.1	7.51	7.92	0.4
			S2.3	7.62	7.96	0.3
			S2.4	7.97	8.03	0.1
	2 mg/L colistin (CI)	15	S3.1	7.87	7.70	−0.2
			S3.2	7.78	7.56	−0.2
	2 mg/L colistin (CI)	60	S4.1	7.56	6.53	−1.0
			S4.2	7.56	6.53	−1.0
			S4.3	7.73	7.47	−0.3
			S4.4	7.87	7.79	−0.1
	25 mg/L doripenem (t1/2 1.5 h)	15	S5.1	7.33	7.61	0.3
			S5.2	7.72	7.88	0.2
	25 mg/L doripenem (t1/2 1.5 h)	60	S6.1	7.50	7.56	0.1
	2 mg/L colistin (CI) + 25 mg/L doripenem (t1/2 1.5 h)	60	S7.1	7.70	6.42	−1.3
			S7.2	7.69	6.42	−1.3
	0.2 mg/L colistin (CI)	60	S8.1	7.42	7.73	0.3
			S8.2	7.94	8.07	0.1
19606R (LPS deficient, colistin resistant)	untreated control	60	S9.1	6.87	6.57	−0.3
			S9.2	6.71	6.58	−0.1
	2 mg/L colistin (CI)	60	S10.1	6.90	6.70	−0.2
			S10.2	6.92	6.81	−0.1

CI, continuous infusion.

RNA-Seq analyses generated between 460 000 (doripenem treatment) and 3.5 million (ATCC 19606 untreated) reads that aligned uniquely with annotated coding sequences (Table S1). Initial analyses indicated that the gene expression changes in ATCC 19606 in response to 2 mg/L colistin were highly similar to those observed previously for the LPS-deficient A. baumannii strain 19606R grown in the absence of colistin.²⁸ However, these data could not be directly compared as the original analyses of the transcriptome of 19606R²⁸ were performed using RNA samples that had not been rRNA depleted and, as a result, significantly fewer reads aligned uniquely with annotated genes compared with the rRNA-depleted samples used here (average 220 000 reads mapped; Table S1). For this reason, we undertook transcriptomic analyses on rRNA-depleted RNA derived from the colistin-resistant strain 19606R grown with and without 2 mg/L colistin. The data from the rRNA-depleted samples used in this study generated a much larger set of differentially expressed genes (984 compared with 229)²⁸ (Table 2). Importantly, 75% of the previously identified genes²⁸ were also identified in this new analysis.

Table 2.

Number of differentially expressed genes identified for each treatment

Strain (description)	Treatment	Time (min)	Number of differentially expressed genesa
			increased expression	decreased expression
ATCC 19606 (WT, colistin susceptible)	2 mg/L colistin (CI)	15	281	124
		60	312	157
	0.2 mg/L colistin (CI)	60	1	0
	25 mg/L doripenem (t1/2 1.5 h)	15	0b	0b
		60	45	0
	2 mg/L colistin (CI) + 25 mg/L doripenem (t1/2 1.5 h)	60	330	159
19606R (LPS deficient, colistin resistant)	no treatment		446	538
	2 mg/L colistin (CI)	60	452	538

CI, continuous infusion.

^aDifferentially expressed genes were identified as those genes with >2-fold expression change at an FDR of <0.01 compared with the ATCC 19606 untreated control strain.

^bThese comparisons also gave no significant differentially expressed genes at an FDR of <0.05.

RNA-Seq was conducted on either an Illumina MiSeq or an Illumina GA-IIx. To ensure that the data generated using the two different platforms (GA-IIx 100 bp PE versus MiSeq 50 bp SE) were comparable, replicate samples of the ATCC 19606 untreated control and the ATCC 19606 treated with 2 mg/L colistin for 1 h were analysed using both platforms (Table S1). Multidimensional scaling was used to compare the data generated for the 50 most variable genes across the two technologies (Figure 1) and clearly showed that the replicates of each sample clustered closely together irrespective of sequencing technology. Therefore, we concluded that the data from samples sequenced using GA-IIx or MiSeq could be combined for all further analyses.

Figure 1.

Multidimensional scaling plot of the 50 most variable genes (highest standard deviation) for the three ATCC 19606 untreated samples (S2.1, S2.3 and S2.4) and the four ATCC 19606 samples treated with 2 mg/L colistin monotherapy for 1 h (S4.1, S4.2, S4.3 and S4.4). The first set of replicates for each condition (S2.1, S4.1 and S4.2) were sequenced as 100 bp PE sequences on the Illumina GA-IIx and the second set of replicates for each condition (S2.3, S2.4, S4.3 and S4.4) were sequenced on the MiSeq as 50 bp SE sequences. Samples are labelled as in Table S1.

Following alignment of the RNA-Seq reads with the ATCC 19606 draft genome, lists of differentially expressed genes were generated for the following conditions relative to the respective untreated control: 2 mg/L colistin for 15 min (Tables S3 and S4) and 1 h (Tables S5 and S6); 25 mg/L doripenem for 1 h (Table S7); and a combination of 2 mg/L colistin plus 25 mg/L doripenem for 1 h (Tables S8 and S9). Furthermore, lists of differentially expressed genes were also generated for the LPS-deficient strain 19606R relative to the WT parent ATCC 19606 untreated control (Tables S10 and S11). The numbers of up- and down-regulated genes identified for each condition are given in Table 2. The analyses revealed that there was no significant change in the gene expression profiles of the ATCC 19606 strain over the time course of the experiment (ATCC 19606 untreated 15 min versus untreated 1 h). There was also no alteration in gene expression in response to 25 mg/L doripenem for 15 min (ATCC 19606 untreated at 15 min versus doripenem treatment for 15 min). Additionally, the gene expression profile of the LPS-deficient 19606R strain did not alter when exposed to colistin (19606R untreated versus colistin treatment with 2 mg/L for 1 h).

qRT–PCR confirmation of differential gene expression of selected genes

We selected three genes, lolA, baeS and pilM, for confirmation of differential gene expression by qRT–PCR (Figure 2). As observed in the RNA-Seq analyses, lolA and baeS were expressed at significantly increased levels in all samples relative to the untreated control (Figure 2). Similarly, as observed by RNA-Seq, pilM was expressed at reduced levels in the experimental samples relative to the untreated control (Figure 2). For all genes under all tested conditions, except baeS in the 19606R mutant, the magnitude of the gene expression changes measured by qRT–PCR was indistinguishable from those measured by RNA-Seq (P > 0.05). The expression of baeS in 19606R was measured as slightly more highly differentially expressed by qRT–PCR than by RNA-Seq (Figure 2).

Figure 2.

Real-time qRT–PCR confirmation of differential gene expression of selected genes. qRT–PCR was used to determine the expression of each of the genes lolA, baeS and pilM in RNA preparations from the LPS-deficient mutant 19606R, ATCC 19606 after colistin (CST) monotherapy for 1 h and ATCC 19606 after doripenem (DOR) monotherapy for 1 h. All data were normalized against the housekeeping gene gyrB and fold change determined relative to the ATCC 19606 untreated 1 h control. For comparison, the average fold change measured by RNA-Seq is also shown. Data shown are mean ± standard error.

Differentially expressed genes in response to colistin treatments

Colistin treatment (2 mg/L) of ATCC 19606 resulted in 405 differentially expressed genes after 15 min (281 up-regulated and 124 down-regulated) and 469 differentially expressed genes after 1 h (Table 2 and Tables S3–S6). Of the genes identified at 15 min, 75% were also identified as differentially expressed at 1 h (Figure 3). Furthermore, the magnitude of expression changes was very similar between the two timepoints (Figure 3); most genes rapidly responded to colistin (within 15 min) and remained at similar expression levels after 1 h. Only a single gene, HMPREF0010_02579, was identified as differentially expressed in response to the sub-MIC concentration (0.2 mg/L) of colistin for 1 h (∼0.2× MIC), indicating that significant transcriptional changes only occur in the presence of colistin concentrations that cause cell damage. This is consistent with the known concentration-dependent killing induced by colistin.³³ The gene HMPREF0010_02579 was also differentially expressed following all colistin treatments, doripenem treatment for 1 h, combined colistin/doripenem treatment (1 h) and in the LPS-deficient strain (19606R untreated versus ATCC 19606 untreated). Together, these data suggest HMPREF0010_02579 plays a critical role in the response to antibiotic treatment and membrane perturbation. The encoded protein shares identity with hypothetical proteins within the Acinetobacter genus but not to any functionally characterized proteins. Thus, full characterization of this protein is clearly warranted.

Figure 3.

Comparison of gene expression changes in response to colistin monotherapy at 15 min and 1 h. (a) Venn diagram showing overlap of gene expression changes in response to colistin treatment at 15 min and 1 h. (b) Gene expression values for all genes significantly differentially expressed at 15 min and at 1 h relative to the 15 min untreated control.

Pathway and GO enrichment analysis was used to identify groups of genes highly over-represented in the differentially expressed gene lists. In the analyses of ATCC 19606 treated with colistin (2 mg/L, 15 min and/or 1 h), these included genes for protein transport (GO:0015031; 13 genes with P = 1 × 10⁻⁴ at 15 min and 12 genes with P = 0.002 at 1 h), membrane assembly (GO:0071709; 4 genes with P = 4 × 10⁻⁴ at 1 h) and the fatty acid degradation β-oxidation pathway (13 genes with P = 4 × 10⁻⁵ at 15 min and 10 genes with P = 0.004 at 1 h) (Tables S3 and S5). Gene groups over-represented in the down-regulated gene lists included those involved in fatty acid and lipid biosynthesis (six genes with P = 1 × 10⁻⁴ at both 15 min and 1 h) and biotin synthesis (three genes with P = 2 × 10⁻⁴ at both 15 min and 1 h) (Tables S4 and S6). Importantly, a large number of genes (65% at 1 h) that responded to colistin monotherapy (2 mg/L) were also differentially expressed in the 19606R LPS-deficient strain grown in the absence of colistin (as previously identified²⁸ and confirmed here), indicating that the response of ATCC 19606 to colistin is highly similar to the response to the total loss of LPS from the outer membrane (Figure S1 and Tables S12 and S13). These included the genes encoding proteins involved in lipoprotein transport (lolA, lolB, lolD and lolE), poly-N-acetyl glucosamine (PNAG) biosynthesis (pgaA, pgaB, pgaC and pgaD), retrograde phospholipid transport (mlaC and mlaD), regulation of envelope stress responses (baeS and baeR) and efflux (adeI, adeJ, adeK, macA, macB and tolC). Together, these data strongly suggest that colistin treatment induces disruption/loss of LPS from the bacterial surface and many of the gene expression changes that occur in response to colistin treatment are aimed at stabilizing the damaged outer membrane.

Differentially expressed genes in response to doripenem treatment

No differentially expressed genes were identified in ATCC 19606 after 15 min of doripenem treatment. However, after 1 h of treatment, 45 genes were identified as differentially regulated; all showed increased expression relative to the untreated control (Table 2 and Table S7), including those involved in biofilm formation (GO:0042710; 2 genes with P = 0.001) and carbohydrate degradation (3 genes with P = 0.003). Interestingly, >93% of these genes (42/45) were also identified as differentially expressed in the 19606R LPS-deficient strain and in ATCC 19606 following 1 h treatment with either 2 mg/L colistin alone or colistin and doripenem combined (Table 3 and Figure S2). These data suggest that a core set of genes respond to antibiotic or general stresses, including those involved in drug efflux (macA, macB, tolC and mexB), PNAG biosynthesis and transport (pgaA, pgaB and pgaC), retrograde phospholipid transport (HMPREF0010_02608 and mlaC) and lipoprotein transport (lolA and lolB).

Table 3.

Up-regulated genes common to ATCC 19606 treated with colistin, doripenem and the combination at 1 h

ATCC 19606 locus tag	ACICU locus tag	ATCC 19606 predicted function	ACICU predicted function	Colistin		Doripenem		Combined colistin + doripenem
				expression ratio (log2)	FDR	expression ratio (log2)	FDR	expression ratio (log2)	FDR
00179	02365	PgaA, biofilm synthesis protein	hypothetical protein	3.6	7.6E-11	2.3	2.6E-04	3.3	4.5E-09
00180	02363	PgaB, xylanase/chitin deacetylase	xylanase/chitin deacetylase	3.6	7.3E-10	1.9	5.0E-03	3.4	2.3E-08
00181	02362	PgaC, glycosyltransferase	glycosyltransferase	3.5	3.3E-10	1.9	3.2E-03	3.2	9.4E-09
00185	02358	hypothetical protein	hypothetical protein	6.7	7.3E-08	4.9	1.3E-03	7.1	8.1E-07
00186	02357	hypothetical protein	hypothetical protein	5.4	4.3E-08	3.5	2.5E-03	5.7	5.2E-07
00247	02289	conserved hypothetical protein	hypothetical protein	2.7	4.5E-04	3.7	8.9E-03	2.8	1.3E-03
00284	02257	transaldolase	transaldolase B	1.3	1.7E-06	1.1	8.9E-03	1.3	2.5E-05
00333	02113	UDP-glucose 4-epimerase	UDP-glucose 4-epimerase	1.4	1.3E-05	1.6	3.9E-03	1.2	6.0E-04
00495	01935	non-haem chloroperoxidase	α/β superfamily hydrolase/acyltransferase	1.9	1.4E-05	2.1	8.9E-03	1.8	2.0E-04
00500	01925	DJ-1/PfpI family protein	putative intracellular protease/amidase	1.6	2.0E-05	2.1	3.1E-03	1.4	6.9E-04
00595		excinuclease ABC, A subunit	absent	1.0	9.0E-04	2.0	3.2E-03	1.4	3.1E-04
00992	01426	catalase hydroxyperoxidase II	catalase	1.8	1.6E-05	2.0	8.9E-03	1.3	2.1E-03
00995		hypothetical protein?	absent	2.7	8.5E-05	3.4	5.0E-03	2.7	3.1E-04
01282	00738	conserved hypothetical protein	hypothetical protein	2.2	7.7E-06	2.4	2.5E-03	2.0	1.6E-04
01333	00789	outer membrane lipoprotein LolB	outer membrane lipoprotein	4.0	1.3E-10	2.2	1.3E-03	3.9	4.5E-09
01378	00835	type VI secretion system OmpA/MotB	outer membrane protein, related peptidoglycan-associated (lipo)protein	2.9	5.3E-09	1.7	1.3E-03	2.8	2.3E-07
01511	00937	conserved hypothetical protein	hypothetical protein	3.0	4.4E-06	3.1	4.4E-03	3.3	8.8E-06
01565	00989	heat shock protein	heat shock protein	2.7	2.9E-10	1.0	7.4E-03	2.5	2.3E-08
01712	00546	MacA, membrane-fusion protein	membrane-fusion protein	5.3	2.1E-09	3.0	1.3E-03	4.7	1.7E-07
01713	00545	MacB, macrolide transporter	peptide ABC transporter permease	5.4	4.3E-10	2.6	2.5E-03	4.7	4.4E-08
01714	00544	TolC, RND efflux transporter	outer membrane protein	5.0	7.6E-11	2.2	2.6E-03	4.4	6.8E-09
01945	00305	hypothetical protein	hypothetical protein	7.4	9.2E-09	5.4	3.9E-04	8.0	3.4E-08
02071	02904	multidrug resistance protein mexB	cation/multidrug efflux pump	3.1	8.1E-11	1.3	9.7E-03	2.8	6.0E-09
02249	02661	periplasmic/secreted protein	periplasmic/secreted protein	4.2	3.3E-10	2.3	1.3E-03	4.2	9.4E-09
02462	03447	hypothetical protein	hypothetical protein	4.9	1.8E-07	4.0	1.3E-03	5.3	1.1E-06
02553	03355	haemerythrin HHE cation binding domain-containing protein	hypothetical protein	1.8	2.8E-06	2.0	2.5E-03	1.8	2.8E-05
02568	03340	secreted protein	hypothetical protein	3.5	6.2E-09	1.9	8.9E-03	3.9	3.4E-08
02579	03329	hypothetical protein	hypothetical protein	5.5	2.7E-13	3.2	1.4E-07	5.3	2.3E-11
02607	03300	MlaC, toluene tolerance protein Ttg2D	ABC-type transport system	3.6	1.9E-11	1.3	1.3E-03	3.3	4.3E-09
02608	03299	MlaB?, conserved hypothetical protein	putative toluene-tolerance protein (Ttg2E)	3.5	1.9E-11	1.2	3.0E-03	3.2	4.3E-09
02675	03230	conserved hypothetical protein	hypothetical protein	3.8	1.9E-11	1.8	6.0E-04	3.4	4.3E-09
02727	03145	glycosyltransferase	glycosyltransferase	4.4	4.3E-09	2.5	9.9E-03	4.3	2.6E-08
02733	03139	conserved hypothetical protein	hypothetical protein	7.9	3.4E-08	5.5	8.7E-04	8.1	4.0E-07
02739	03132	conserved hypothetical protein	hypothetical protein	7.3	8.7E-09	4.4	1.4E-03	7.1	1.6E-07
02740	03131	BaeS, two-component signal transduction system sensor kinase component	sensor kinase component of a two-component signal transduction system	3.1	1.9E-11	1.2	2.9E-03	2.5	4.5E-09
02888	02981	outer membrane lipoprotein carrier protein LolA	outer membrane lipoprotein-sorting protein	4.9	1.3E-10	2.6	5.4E-04	4.7	7.6E-09
03145	01159	peptidase M48 family protein	Zn-dependent protease with chaperone function	2.8	3.5E-10	1.2	5.7E-03	2.8	1.2E-08
03296	00065	conserved hypothetical protein	hypothetical protein	4.5	7.7E-08	3.1	8.9E-03	4.7	2.5E-07
03355	00054	conserved hypothetical protein	hypothetical protein	5.2	6.5E-09	2.6	9.0E-03	5.4	1.2E-07
03356	00053	conserved hypothetical protein	hypothetical protein	8.3	1.0E-07	6.0	1.3E-03	8.2	1.6E-06
03654		hypothetical protein	absent	6.3	2.0E-08	4.2	1.3E-03	6.6	1.7E-07
03655	00035	conserved hypothetical protein	hypothetical protein	4.4	5.3E-09	2.7	2.9E-03	4.9	2.3E-08

Differentially expressed genes in response to colistin and doripenem combination therapy

Growth of the ATCC 19606 strain with colistin/doripenem combination therapy for 1 h resulted in increased and decreased expression of 330 and 159 genes, respectively, compared with the untreated control (Table 2 and Tables S8 and S9). Comparison of the differentially expressed genes identified following combination therapy with those identified following monotherapy with either colistin or doripenem identified a core set of 42 genes up-regulated under all three treatments (Table 3 and Figure S2). Genes over-represented in this set of 42 genes included those involved in biofilm formation (GO:0042710; 2 genes with P = 0.001; ACICU_02362 and 02363), carbohydrate degradation (3 genes with P = 0.003; ACICU_02113, 02257 and 02363) and membrane components (GO:0016020; 8 genes with P = 0.007; ACICU_00544, 00545, 00546, 00789, 00835, 01159, 02904 and 03131). Of the 489 genes differentially expressed in response to the colistin and doripenem combination therapy, 341 (70%) of these were also identified as differentially expressed in response to colistin 2 mg/L monotherapy, suggesting that the gene expression changes resulting from the colistin and doripenem combination therapy were driven predominantly by the action of colistin (Figure S2). Interestingly, of the 84 genes up-regulated in response to the combined treatment, but not the monotherapy treatments, 10% encoded predicted transcriptional regulators (HMPREF0010_00156, 00515, 00577, 00628, 00680, 02475, 02927 and 03265); the genes that these proteins regulate are unknown.

Functional importance of differentially expressed genes

The genes macA, macB, tolC and mexB (HMPREF0010_01712, 01713, 01714 and 02071) were highly up-regulated in response to both colistin and doripenem treatment (1 h) and in the 19606R LPS-deficient strain, indicating that their increased expression is a crucial response to treatment with either antibiotic and also to membrane perturbation. The MacAB-TolC resistance–nodulation–division (RND) macrolide transporter system in Escherichia coli is involved in drug efflux,³⁴ but MacA has also been shown to bind LPS and may be involved in glycolipid transport under certain conditions.³⁵ Thus, increased expression of this system may act to both increase efflux of toxic compounds and restore glycolipid components of the damaged outer membrane. Pseudomonas aeruginosa MexB (49% identity with A. baumannii MexB) is involved in carbapenem efflux.³⁶ The AdeIJK efflux system was also up-regulated in response to colistin treatment and LPS loss and has a known involvement in the efflux of β-lactams, chloramphenicol, tetracyclines and ethidium bromide.^37,38 AdeI, AdeJ and AdeK share 48%, 59% and 43% identity with P. aeruginosa MexA, MexB and OprM, respectively.

The genes encoding the Lol lipoprotein transport system (lolA, lolB, lolD and lolE; HMPREF0010_02888, 01333, 02125 and 02124, respectively) were up-regulated in ATCC 19606 treated with colistin (2 mg/L mono- or combined therapy) and in the 19606R LPS-deficient strain, as shown here and reported previously,²⁸ but only lolA and lolB were up-regulated following doripenem treatment. In E. coli, the Lol system (LolABCDE) is essential for lipoprotein transport from the inner to the outer membrane.^39,40 In A. baumannii ATCC 19606, no clear homologue of LolC was identified but LolE may be bifunctional as it had significant identity with both the E. coli LolC protein and the LolD protein. The increased expression of this system in response to colistin and in LPS-deficient cells suggests a requirement for increased lipoprotein at the cell surface. This may help stabilize the outer membrane in the LPS-deficient (19606R) or LPS-depleted (colistin-treated ATCC 19606) cell. The increased expression of lolA and lolB during doripenem monotherapy suggests that lipoproteins may play an as yet unknown role in response to this antibiotic.

In Gram-negative bacteria, the inner and outer leaflets of the outer membrane are primarily composed of phospholipids and LPS, respectively.⁴¹ The Mla system is proposed to maintain outer membrane asymmetry by transporting excess phospholipids in the outer leaflet back to the inner membrane.⁴² In this study, the increased expression of the Mla system (mlaC, mlaD and HMPREF0010_02608) in response to colistin treatment suggests that membrane asymmetry has been disrupted, consistent with the prediction that colistin induces cell death via lipid exchange between the inner and outer membrane.¹⁴ HMPREF0010_02608 and mlaC were also up-regulated following doripenem treatment, but it is unclear what role this system might play in response to doripenem.

The pgaABCD genes encode proteins involved in the synthesis and transport of PNAG;⁴³ all were up-regulated in response to colistin (mono- and dual therapy) and in the 19606R LPS-deficient strain. PNAG is a secreted exopolysaccharide critical for biofilm formation in A. baumannii⁴³ and has been shown to be produced at increased levels in the LPS-deficient strain 19606R.²⁸ PNAG imparts increased polymyxin resistance to E. coli biofilms, probably due to its positive charge having a repulsive effect on cationic antibiotics.⁴⁴ The increased PNAG production by A. baumannii in response to colistin treatment may also act to reduce the charge-based interaction of colistin with the bacterial outer membrane.

The genes comprising the BaeSR two-component signal transduction system were up-regulated in A. baumannii treated with colistin and in the LPS-deficient strain.²⁸ baeS was also up-regulated following doripenem treatment and baeR showed increased expression, but below the level required for significance (2.1-fold, FDR = 0.05). In E. coli, BaeSR controls the expression of numerous efflux systems including MdtABC, ArcD and TolC in response to cell envelope stress.^45–47 In A. baumannii, colistin and doripenem treatment also led to the increased expression of multiple efflux systems including MacAB-TolC, MexB and AdeIJK; it is likely that some or all of these systems are under the control of the BaeSR system.

Numerous pili biosynthesis genes were down-regulated in ATCC 19606 following colistin (2 mg/L) and colistin plus doripenem treatment and in the LPS-deficient strain (as reported previously²⁸ and confirmed here). These included pilM and pilN (HMPREF0010_02514 and 02515, respectively), the pap pilus assembly genes fimA, fimB/papD, papC and fimA (HMPREF0010_00597–00600, respectively) and the chaperone/usher system genes csuA and csuB (HMPREF0010_00110 and 00111, respectively). We propose that the outer membrane containing depleted levels of LPS (following colistin treatment) or no LPS (LPS-deficient strain 19606R) is destabilized, which leads to reduced expression of these membrane-spanning structures.

The majority of genes differentially expressed in response to colistin were also differentially expressed in the LPS-deficient strain 19606R (grown without colistin). However, some genes responded only to colistin treatment. These included genes involved in fatty acid β-oxidation/degradation, which were up-regulated (including fadA, fadB and fadE) (Table S14; P = 4 × 10⁻⁵), and genes involved in fatty acid synthesis, which were down-regulated (Table S15). We hypothesize that the change in expression of these pathways is required to respond to colistin damage at the bacterial surface and for the outer membrane remodelling that follows. In the already LPS-deficient strain 19606R, the outer membrane has presumably stabilized and therefore altered expression of these pathways is not required.

Colistin treatment increases expression of genes encoding SODs

The current evidence on the importance of free radicals in the mechanism of bacterial killing by antibiotics is contradictory.^48–50 Preliminary studies in A. baumannii indicate that colistin action may be in part mediated by the toxic activity of hydroxyl radicals¹⁶ or by inhibition of respiratory chain enzymes such as NADH-quinone oxidoreductase.¹⁹ A. baumannii ATCC 19606 contains two genes predicted to encode SOD enzymes: HMPREF0010_02336 (sodB encoding a predicted FeSOD) and HMPREF0010_02564 (encoding a predicted Cu-ZnSOD). Mutation of the A. baumannii ATCC 17978 sodB has been shown to result in a 2-fold increase in colistin susceptibility.¹⁸ Increased expression of the ATCC 19606 sodB was specific to colistin treatment, indicating that hydroxyl radicals may be involved in colistin antibacterial activity. The expression of HMPREF0010_02564 was increased in response to colistin exposure and also in the untreated 19606R LPS-deficient strain, indicating that this enzyme responds to generated free radicals or imbalance of intracellular redox potential as a consequence of both LPS loss and LPS deficiency.

The NADH-quinone oxidoreductase is a critical respiratory chain multisubunit complex⁵¹ that in A. baumannii ATCC 19606 is predicted to be encoded by 13 colocated genes (HMPREF0010_01253–01265). These genes were consistently up-regulated in the LPS-deficient 19606R strain (8/13 genes showed >2-fold up-regulated expression with FDR <0.003; 5/13 genes showed increased expression of between 1.6- and 2-fold with FDR ≤0.01). However, expression was unchanged following antibiotic treatment, indicating that this system is up-regulated only in A. baumannii that has lost all LPS production.

Conclusions

This is the first known study to examine the transcriptomic responses to colistin and/or doripenem treatments using an in vitro PK/PD model to simulate dosage regimens in patients. Our results are in agreement with the hypothesis that colistin induces significant membrane damage, shedding of LPS and lipid exchange between inner and outer membranes. Surprisingly, the transcriptional response to doripenem monotherapy involved mostly genes that were also differentially regulated in response to colistin, despite the antibiotics having very different mechanisms of action and predominantly involved genes that encode drug efflux and stress response proteins. Interestingly, doripenem appears to contribute negligibly to the changes in the global gene expression profiling caused by the colistin/doripenem combination. Our study highlights the potential of transcriptomics in understanding the synergy of antibiotic combination therapy.

Funding

The project was supported by Australian National Health and Medical Research Council (NHMRC; APP1046561) and award number R01 AI079330 from the National Institute of Allergy and Infectious Diseases. Z. F. L. is supported by the National Natural Science Foundation of China (81128016). J. L. is an Australian NHMRC Senior Research Fellow.

Transparency declaration

None to declare.

Disclaimer

The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institute of Allergy and Infectious Diseases or the National Institutes of Health.

Supplementary data

Tables S1–S15 and Figures S1 and S2 are available as Supplementary data at JAC Online (http://jac.oxfordjournals.org/).