Antibiotic susceptibility signatures identify potential antimicrobial targets in the Acinetobacter baumannii cell envelope

Abstract

A unique, protective cell envelope contributes to the broad drug resistance of the nosocomial pathogen Acinetobacter baumannii. Here we use transposon insertion sequencing to identify A. baumannii mutants displaying altered susceptibility to a panel of diverse antibiotics. By examining mutants with antibiotic susceptibility profiles that parallel mutations in characterized genes, we infer the function of multiple uncharacterized envelope proteins, some of which have roles in cell division or cell elongation. Remarkably, mutations affecting a predicted cell wall hydrolase lead to alterations in lipooligosaccharide synthesis. In addition, the analysis of altered susceptibility signatures and antibiotic-induced morphology patterns allows us to predict drug synergies; for example, certain beta-lactams appear to work cooperatively due to their preferential targeting of specific cell wall assembly machineries. Our results indicate that the pathogen may be effectively inhibited by the combined targeting of multiple pathways critical for envelope growth.

Fig. 1. Genome-wide profiling of gene–antibiotic interactions in A. baumannii by Tn-seq.

Diagram outlines the multiple parallel Tn-seq fitness profiling experiments as described in “Methods”. Sub-MIC drug concentrations used to achieve 20–30% growth rate inhibition are listed in Supplementary Table 1. For each antibiotic, genes contributing to intrinsic defense against each single drug were identified by using significance criteria (Methods). Fitness profiles across all conditions (phenotypic signatures) were then analyzed to identify novel gene relationships and discover multicondition discriminating genes. Drug abbreviations (Abbr.) and targets are listed. *CIP Tn-seq data used in these studies were described previously.

Fig. 2. Genes with interconnected functions show correlated phenotypic signatures.

a–f Genes within a shared functional pathway show relationships in their Tn-seq fitness signatures. a, c Heat maps (using blue-white-red scale at bottom) show normalized Tn-seq fitness in z-scored units for mutants in each gene (rows) grown in distinct conditions (columns). Characterized/annotated genes were first placed into shared functional pathways (Methods). We next identified additional genes (arrowheads), many of which were uncharacterized/unannotated, that correlate with each pathway by using hierarchical clustering with the entire set of A. baumannii phenotypic signatures. Parentheses denote domains identified from NCBI conserved domain database (CDD). In c, dashed black lines separate genes with opposing regulatory effects. b, d Heat maps (using yellow-white-blue scale at bottom) show Pearson correlation coefficient (r) matrices measuring relatedness of the corresponding Tn-seq fitness signatures in (a). Positive and negative r indicate correlation and anticorrelation, respectively. e, f Phenotypic signatures correlating with pAB3-encoded ACX60_RS18565. Tn-seq fitness (e) and correlations (f) are shown as above. N/A, no gene name or predicted protein domain. Source data are provided as a Source Data file.

Fig. 3. AdvA is a critical cell division protein connected by susceptibility phenotypes to BlhA.

a A. baumannii transposon mutants mapping throughout blhA and in a limited region of advA show decreased fitness with fluoroquinolones and β-lactams but not RIF. Bars show Tn-seq fitness values of individual Tn10 mutants at each locus across all tested banks in the indicated condition. b Validation of Tn-seq drug susceptibility phenotypes using defined mutants independently cultured in microplates. EGA746 (∆blhA, top, blue symbols) or EGA745 (∆advA/pMS88::advA, bottom, green symbols) were tested in parallel with the same WT control (black symbols). Data are presented as geometric mean (indicated by symbols) ±s.d. (indicated by area-filled dotted bands) from n = 3 independent cultures. Where not visible, s.d. is within the confines of the symbol. c Schematic of AdvA protein. Tn indicates approximate location of transposons in advA analyzed by Tn-seq, TM indicates predicted transmembrane helix. d, e AdvA is essential for colony formation. EGA745 or control were grown on solid medium at 37 °C (d). AFA11 (∆advA harboring T5lacP-advA-gfp) was grown on solid medium ± 0.5 mM IPTG (e). Images are representative of 18 (d) or 3 (e) parallel cultures. f AdvA is essential for growth in broth. AFA11 pregrown with 1 mM IPTG was diluted into LB ± 1 mM IPTG, followed by dilution into the same medium after 4 h. Growth was monitored by A₆₀₀ (1-cm cuvettes). Data points show geometric mean ± s.d. (n = 3 independent cultures). g AdvA level determines antibiotic susceptibility. AFA11 pregrown with 250 µM IPTG was washed, resuspended in LB with 5 or 125 µM IPTG, and cultured with WT control in microplates with or without the indicated sub-MIC antibiotic (Supplementary Table 1). Data are shown as geometric mean ± s.d. as in (b). h, i advA deficiency results in cell filamentation. The indicated strains grown with or without inducer were imaged via phase-contrast. Images are representative of three parallel cultures. Scale bar, 10 µm. j AdvA-GFP localizes to mid-cell at sites of cell division. WT A. baumannii harboring T5lacP-advA-gfp was cultured to mid-log phase with 50 µM IPTG and imaged by phase-contrast and fluorescence microscopy. Scale bar, 5 µm. Images are representative of two parallel cultures. Source data are provided as a Source Data file.

Fig. 4. Phenotypic signatures defined by hydrophobic compound sensitivity reveal connection of pbpG to LOS synthesis.

a Fitness profile clusterogram of genes for which knockout causes preferential but differing hypersensitivity to hydrophobic (RIF and AZITH) and amphipathic (COL and PB) antibiotics. Tn-seq data were subjected to PCA to identify discriminating genes whose fitness values differed as a function of hydrophobicity annotation [high, xlogp3 ≥ 4; low, xlogp3 < 4; amphipathic, polymyxin antimicrobial peptides; NA not applicable (no drug); Supplementary Table 1] (ANOVA, q = 0.0003). Heatmap shows normalized fitness in z-scored units. Dashed boxes indicate clusters of phenotypic signatures defined by differential defects with hydrophobic and amphipathic drugs. Arrowheads highlight cell-wall enzymes clustering with distinct surface polysaccharide synthesis pathways. b Bars show fitness values of individual transposon mutants at each locus across all tested banks as in Fig. 3. c, d Validation of Tn-seq selective drug hypersensitivities using deletion mutants ∆pbpG (c) or ∆pbp1B (d) vs. isogenic WT. Data points show geometric mean ± s.d. (n = 3 independent cultures, except for WT + MEC, n = 2 independent cultures) as in Fig. 3. e ∆pbpG strain shows detergent hypersensitivity phenotype matching that of lpsB LOS glycosyltransferase mutant by colony formation efficiency (CFE) assay. Data points show geometric mean ± s.d. (n = 4 independent cultures). f Reintroduction of pbpG reverses hypersusceptibility to RIF and SDS. Strains harbored vector (pYDE153) or vector containing pbpG (pYDE210). Bars show geometric mean CFE ± s.d. (n = 3 independent cultures). g, h pbpG knockout results in reduced LOS levels. g LOS (bottom) and total protein (top) were detected in cell lysates separated by SDS-PAGE. Image is representative of three experiments. h LOS levels in regions indicated in g were normalized to total protein content. Values are shown relative to total normalized LOS levels in WT. Bars show mean ± s.d. from n = 7 (WT, ∆pbpG) or n = 5 (∆lpsB and ∆lpxL) biologically independent samples. Total normalized LOS levels of each mutant were compared to WT by two-way ANOVA with Dunnett’s multiple comparisons test. *p < 0.0001; n.s. not significant (p = 0.1652). Source data are provided as a Source Data file.

Fig. 5. Morphology-specific susceptibility signatures uncover phenotypes and proteins linked to the Rod-system.

a Exposure of A. baumannii to sub-MIC β-lactams causes target-specific morphotypes. SLB (0.25 µg/ml) causes growth as extended rods. CEF and AZT cause a similar filamentous morphology. MEC (16 µg/ml) causes loss of rod shape. IPM and A22 cause a similar spheroid morphology. Images were acquired with phase-contrast. Scale bar, 5 µm. b Tn-seq fitness clusterogram showing subset of genes for which inactivation causes selective hypersensitivity to antibiotics generating filaments vs. spheres. Tn-seq data from the indicated conditions were subjected to PCA using morphology annotations (indicated above the heatmap), and discriminating genes having significantly different fitness with cell wall perturbations causing filamentation compared to other conditions were identified (two-sided t test, q = 0.025). Heatmap shows normalized fitness in z-scored units. Dashed box indicates cluster of canonical Rod-system genes with three uncharacterized genes (arrowheads). c Fitness values of individual transposon mutants at each locus across all banks as in Fig. 3. d, e Validation of Tn-seq hypersensitivities using deletion mutants and WT cultured independently. Data points show geometric mean ± s.d. (n = 3 independent cultures) as in Fig. 3. f Domain predictions in the indicated proteins, listed with NCBI locus tag and protein annotation. ElsL does not contain a predicted signal peptide. g SLB susceptibility on solid medium was analyzed via CFE assay. Data points show geometric mean ± s.d. (left, n = 3; right, n = 4 independent cultures). h CFE with SLB (0.25 µg/ml) vs. no drug was measured with strains harboring pEGE305 (vector), pYDE135, (pbp2) or pEGE308 (elsL). Bars show geometric mean ± s.d. (left, n = 4; right, n = 3 independent cultures). i ElsL, ElsS, and DacC determine Rod shape. Mutants (bottom) and WT control (top) grown without antibiotics to mid-log phase were imaged with phase contrast. Scale bar, 5 µm. Images are representative of 2 (elsL), 3 (elsS), or 4 (dacC) parallel cultures. Insets show 2×-magnified views of representative bacteria. j Two-hybrid interactions of ElsS and Rod-system proteins. Proteins fused to a T25 or T18 CyaA fragment were tested in E. coli for LacZ reporter activity on X-gal plates. Images are representative of 6 co-transformants patched in parallel. Source data are provided as a Source Data file.

Fig. 6. Synergistic inhibition of A. baumannii by pairing Rod-system-targeting and Divisome-targeting antibiotics.

a Diagrams showing the two modes of cell wall growth in rod-shaped bacteria which are governed by distinct biosynthesis systems (Rod-system vs. divisome). Listed next to each diagram is the subset of antibiotics that preferentially target the respective PG synthesis system. b Heat map shows average Log2FIC scores resulting from pairwise interactions among seven antibiotics via the diagonal sampling method from n = 2 independent determinations. Blue indicates synergistic pairs, white indicates additive pairs, and red indicates antagonistic pairs. *Average Log2FIC was significantly different from 0 in one-sample t test (two-tailed p < 0.05; exact p values shown in Supplementary Table 3). c Validation of drug–drug interactions via checkerboard assay. Heat map shows bacterial growth in microplate wells containing no drug (lower left wells) or increasing amounts of each drug alone or in pairwise combinations. Drug concentrations increase linearly from left to right along x-axis, and bottom to top along y-axis. “Div” refers to Divisome, “Rod” refers to Rod-system. Source data are provided as a Source Data file.