The vast majority of A. baumannii essential genes show significant chemical-gene interactions

To identify phenotypes for essential gene knockdown strains in the context of chemical treatment (i.e., chemical-gene interactions) in A. baumannii, we conducted a large-scale chemical genomics screen using our established inducible CRISPRi library in A. baumannii strain ATCC19606 (19606 throughout) [20]. This library consists of pooled CRISPRi strains with guides targeting 406 putatively essential genes derived from Tn-seq data in other strains [3,4], along with 1000 non-targeting control sgRNAs. For each targeted gene, we cloned four sgRNA spacers that exactly match the target sequence (perfect-match) and ten mismatch spacers with single-base variations from the target sequence that allow for titration of knockdown [22]. We performed competition fitness assays by inducing CRISPRi knockdown and adding chemicals at sublethal concentrations. These chemical concentrations ensured sufficient cell viability to preserve library diversity and representation, allowing us to measure phenotypic responses to treatment (Fig 1A). We screened the library against a diverse collection of chemical compounds, including clinically relevant antibiotics, heavy metals, and compounds with unknown mechanisms of action (Fig 1B). Knockdown strain abundance was analyzed by amplification and sequencing of the sgRNA spacer regions. We set quality control metrics—replicate correlation and sample complexity cutoffs—to remove samples with excessive noise or selective pressure (e.g., lethal chemical concentrations) that would skew calculations (S1A and S1B Fig; see Material and Methods for details). We verified that chemical solvent impacts were negligible as library composition for DMSO-only and mock treatment samples were highly correlated (r > 0.95) (S1C Fig).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Chemical genomics screen in A. baumannii essential gene library.

(A) Schematic depicting experimental setup for chemical genomics. (B) Proportional area chart of reported cellular targets of screen chemicals. (C) Histogram showing significant chemical-gene scores (|median log₂ fold change| ≥ 1 and Stouffer’s p < 0.05, calculated from perfect-match guides), across chemical conditions in our screen. Darker bars to the right of the dotted line represent at least one significant chemical interaction in this screen. (D) Stacked bar chart showing the sum of significant positive (pink) and negative (blue) CG scores across screen chemicals for each gene in the library.

https://doi.org/10.1371/journal.pgen.1011642.g001

To determine chemical-gene phenotypes from the pooled screens, we calculated chemical-gene (CG) scores for each gene as the median log₂ fold change (medL₂FC) of perfect guides with chemical treatment compared to induction alone. Because our A. baumannii CRISPRi system achieves only ~20-fold knockdown after 14 generations [20], perfect match guides cause only a partial knockdown. Positive CG scores indicate gene knockdowns with improved growth under chemical conditions relative to the untreated knockdown, whereas negative CG scores indicate chemical sensitivity or reduced growth. 93% (378/406) of the genes we investigated exhibited at least one significant CG score (medL₂FC ≥ |1|, p <0.05) upon knockdown, with a median of 14 significant chemical interactions per gene (Fig 1C). Most of these scores were skewed toward reduced growth, with ~73% (3895/5345) of significant CG scores being negative (Fig 1D). This is consistent with previous studies suggesting essential genes are more frequently connected by negative interactions, where two perturbations (e.g., a chemical stressor and a knockdown) cause further loss of fitness than expected based the effects of the individual perturbations [23]. Overall, this dataset provides thousands of unique chemical-gene phenotypes and serves as a detailed resource for A. baumannii essential gene responses to chemical stress.

Lipooligosaccharide transport inhibition enhances drug susceptibility through increased membrane permeability

To identify A. baumannii pathways crucial for chemical resistance, we ranked genes based on significant negative CG scores (medL₂FC ≤-1, p <0.05) across conditions and performed functional enrichment using the STRING database [24]. We found that knockdowns of the lipooligosaccharide transport (Lpt) system were significantly enriched (GO:0015920, enrichment score = 8.95, FDR = 1.04e-05), with negative CG scores for 70% of conditions in our screen (Fig 2B and 2C). Lpt is a highly conserved pathway required for lipopolysaccharide (LPS)/lipooligosaccharide (LOS) export to the outer leaflet of the outer membrane in Gram-negative bacteria. A. baumannii uses Lpt to transport LOS, which is comprised of lipid A and core oligosaccharide moieties found in LPS but lacks O-polysaccharide repeats (Fig 2A) [25]. Intriguingly, while LPS/LOS is essential in most Gram-negative bacteria, many strains of A. baumannii do not require LOS for survival, including the strain used for our library (19606) [26]. Indeed, knockdowns of lpx genes responsible for LOS synthesis showed fewer phenotypes than knockdowns of lpt needed for LOS transport (S2A Fig), suggesting that LOS transport unexpectedly plays a more crucial role in A. baumannii’s chemical sensitivity than LOS synthesis itself.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. lpt knockdowns are sensitized to antibiotics.

(A) Graphical depiction of Lpt machinery and LOS trafficking to the outer membrane in A. baumannii. (B) Histogram of significant negative chemical interactions (medL₂FC ≤ -1 and Stouffer’s p < 0.05) across conditions. Bold line indicates location of lpt genes. (C) Heatmap displaying CG scores for lpt knockdown cluster. Y-axis clustering was conducted using the Ward method and Canberra distance across all library knockdowns. (D) Tigecycline and azithromycin MIC test strip assays for the lptA knockdown, lpxC knockdown, and control strain containing a nontargeting guide. Plates supplemented with 1mM IPTG; approximate MICs reported below images (μg/mL). (E) Ethidium bromide permeability assay for lptA knockdown and nontargeting control in 19606 or LOS^- backgrounds after induction; increased fluorescence over time indicates membrane permeability. Ribbons represent standard deviation (n=4).

https://doi.org/10.1371/journal.pgen.1011642.g002

We validated our screen results for individual knockdowns outside of the pooled context by performing quantitative minimum inhibitory concentration (MIC) assays using antibiotic strips. This revealed that an lptA knockdown was sensitized to several clinically relevant antibiotics (Fig 2D). In contrast, an lpxC knockdown was not sensitized (Figs 2D and S2B) although loss of LOS has been shown to increase antibiotic sensitivity [27]. Both the lpxC and lptA knockdowns remained sensitized to colistin (S2B Fig)—a cationic lipopeptide that disrupts membranes by targeting the lipid A moiety of LOS [28] —suggesting that these knockdowns do not completely eliminate LOS production or transport and knockdown strains still maintain some level of LOS in the outer membrane that can be targeted by colistin [29]. We therefore introduced CRISPRi either targeting lptA or containing a non-targeting control sgRNA into a mutant strain background that does not produce LOS (i.e., LOS^-), 19606 lpxC(S106R) [27]. In contrast to the 19606 lptA and lpxC knockdowns, LOS^- strains were fully resistant to colistin (S2B Fig). Additionally, the LOS^- strains showed considerable growth defects not exhibited by the 19606 knockdown strains (S2D Fig). Yet, the lptA knockdown strain was as similarly sensitized to levofloxacin as the LOS^- strains (S2B Fig). This suggested levofloxacin, an antibiotic targeting gyrase/DNA synthesis in the cell cytoplasm, could penetrate the cell envelope more effectively in both the lptA and the LOS^- strains.

These findings along with the sensitivity of lpt knockdowns to chemically diverse inhibitors pointed to a permeability defect. Therefore, we performed an ethidium bromide (EtBr) fluorescence assay to test for the knockdown’s effect on membrane permeability; in this assay, cells are briefly exposed to EtBr, which fluoresces after passing through the outer and inner membranes and binding to DNA. More permeable membranes will have higher fluorescence readings, and as expected, the LOS^- strains showed increased permeability compared to the 19606 nontargeting control. Surprisingly though, the 19606 lptA knockdown was substantially more permeable than the LOS^- and LOS^- lptA knockdown strains (Figs 2E and S2E). This indicates that Lpt deficiencies cause a further increase in membrane permeability when LOS is being synthesized, possibly due to LOS accumulating in and destabilizing the inner membrane. Taken together, these findings further underscore the critical role of the LOS transport system in A. baumannii in maintaining membrane integrity and mediating resistance to chemicals.

Taxon-specific determinants of chemical susceptibility and resistance

Essential genes that are highly divergent in, or exclusive to, A. baumannii may contribute to its unique drug resistance and could help guide the development of targeted therapeutics. To identify these genes, we determined the phylogenetic profiles of the genes represented in our library across a broad collection species covering the known diversity of Gammaproteobacteria [30]. We identified 15 candidate genes that were highly responsive to chemicals when knocked down but were rare in Gammaproteobacteria outside of the order to which A. baumannii belongs (Figs 3A and S4). These genes exhibited significant CG scores (medL₂FC ≥ |1|, p <0.05) for at least 10% of chemicals tested and likely play substantial roles in responding to chemical stressors.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Chemical interactions and conservation of essential genes.

(A) Dot plot depicting library chemical-gene scores and ortholog presence ratio (fraction of isolates possessing at least one ortholog out of the total number of analyzed isolates) across (A) representative non-Moraxellales Gammaproteobacteria or (B) non-A. baumannii Acinetobacter species. Dots in blue represent genes with <38% presence across representative groups and significant chemical-gene interactions in >10% of screen conditions. (C) Heatmap depicting CG scores for knockdowns of hits for genes rare outside A. baumannii.

https://doi.org/10.1371/journal.pgen.1011642.g003

Among others, this revealed that orthologs of A. baumannii lptC, which is involved in lipooligosaccharide transport, are considerably rare among Gammaproteobacteria (Figs 3A and S5). This suggests A. baumannii lptC is more divergent, in contrast to other lpt genes. Additional underexplored cell envelope genes—GO593_05760 (advA), GO593_07665 (blhA), and GO593_11915 (smc2), GO593_03530 (gtrOC1), GO593_06865 (wzx),—suggest distinct and under-characterized components of the A. baumannii cell envelope may play key roles in chemical response.

We then increased our resolution to pinpoint genes specific to A. baumannii compared to other Acinetobacter species. Three genes stood out as more responsive to chemicals upon knockdown: gtrOC1, wzx, and GO593_15125 (Fig 3B). Both gtrOC1 and wzx are involved in cell surface processes; wzx encodes a predicted polysaccharide flippase within a capsule biosynthesis locus [31], while gtrOC1 is part of a locus involved in LOS synthesis [32]. Orthologs to these genes are only sporadically found in Gammaproteobacteria outside A. baumannii (Figs 3A,3B,S3, and S4). For wzx, one explanation for this patchy occurrence of orthologs is the diversity of polysaccharides that the encoded protein transports [33,34]. Similarly, gtrOC1 has been implicated in a novel mechanism for core oligosaccharide assembly in A. baumannii [32]. Hierarchical clustering of gene knockdowns based on CG score correlations using Ward’s method confirmed association of gtrOC1 with LOS synthesis genes but revealed an unexpected clustering of wzx with thiD, a thiamin synthesis gene, for unknown reasons (Fig 3C).

GO593_15125 shows a similar ortholog distribution (Figs 3A, 3B, and S4) but encodes for an Arc-family repressor. In our previous work, we showed knockdown of a different Arc-family repressor caused significant fitness loss, likely due to its role in repressing toxic prophage genes [20]. GO593_15125 could play a similar role in regulating mobile genetic elements (MGEs) as it is linked to a predicted transposase and an antiphage defense island [35,36]). Additionally, the knockdown is most sensitive to levofloxacin, a DNA-damaging antibiotic (Fig 3C), and MGEs often modulate expression in response to DNA damage [37,38] and occasionally trigger DNA damage responses themselves [39]. Furthermore, this gene clusters with nusG, encoding a transcription elongation/termination factor (Fig 3C). NusG in E. coli induces the DNA damage response and rac prophage excision when depleted [40] and the essentiality of nusG is tied to suppression of toxic genes in rac [41], suggesting that GO593_15125 and nusG are both involved in prophage gene repression in A. baumannii.

These findings identify key chemically responsive genes in A. baumannii, which are less common in other Gammaproteobacteria. Their potential unique roles in bacterial survival provide a strong foundation for further investigation into their contribution to drug resistance.

An essential gene network in A. baumannii identifies novel gene connections

We next sought to identify further functional connections for these taxon-specific, chemically responsive genes. Using the CG scores, we calculated all pairwise correlations between genes then defined a significance threshold by generating an empirical null distribution through random permutation of the underlying data [7,16]. We used only correlations passing the significance threshold to construct a high-confidence essential gene network. Most connections in this network were between genes in different operons (Fig 4A), suggesting true functional relationships dominated over artifacts of CRISPRi polarity. As expected, genes in well-conserved pathways like translation and cell division showed numerous connections (Fig 4A). However, oxidative phosphorylation complexes were disconnected (Figs 4A and S6A), suggesting distinct functionalities outside of the electron transport chain (e.g., succinate dehydrogenase also functions in the TCA cycle).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. (A)

Essential gene network in A. baumannii. Nodes (genes) are connected by edges (phenotypic correlations, r>0.76). Solid lines represent genes across different operons; dotted lines represent genes within the same operon. Subnetworks depict (B) DNA polymerase and replication-related genes, (C) signal-recognition, and (D) cell division and cell wall synthesis. (E) Quantification of cell length of knockdowns from microscopy of cell wall gene knockdowns (N=200-400). (F) Representative microscopy images. Ruler is 5 microns.

https://doi.org/10.1371/journal.pgen.1011642.g004

Connections in this network between poorly understood, taxon-specific genes and well-characterized, conserved genes point to shared functions. Lpt genes formed a subnetwork with gtrOC1, supporting the gene’s role in LOS synthesis (S6B Fig). In comparison, wzx was located in a subnetwork with genes involved in thiamin synthesis, ferredoxins, iron-sulfur cluster biogenesis, isoprenoid synthesis, cytochrome bo₃ oxidase, and another monovalent cation/H+ antiporter, hinting at broader connections between aerobic metabolism and isoprenoid lipid-linked glycans operated on by Wzx [42] (S6A Fig).

One of the candidate genes restricted to Moraxellales, GO593_00140, was linked with multiple DNA replication genes and an unannotated gene, GO593_11615 (Fig 4B). Knockdowns of GO593_00140 and GO593_11615 showed significant, negative CG scores in levofloxacin, which targets replication (S7A and S7B Fig). Despite poor conservation (Figs 2A and S4), structural prediction of GO593_00140 using Phyre2 showed a domain similar to that of E. coli HolD, the psi subunit of the bacterial DNA polymerase III holoenzyme (48% of the sequence modelled with 96.8% confidence despite only 12% sequence ID) (S7C Fig). A previous hidden Markov model (HMM)-based homolog search identified similar putative psi subunits in other Acinetobacter species [43], and our work provides phenotypic evidence to further support that this gene is involved in DNA replication in A. baumannii. GO593_11615 likely encodes a phage-associated transcriptional regulator, as the gene is situated within an intact prophage predicted by PHASTEST [44]. This protein has domain architecture similar to LexA, a conserved repressor of DNA repair pathways that senses DNA damage through an interaction with RecA-DNA filaments that triggers self-cleavage and subsequent de-repression of genes involved in the SOS response [45] (S7D Fig). Depletion of LexA is known to induce excision of some prophages and subsequent host lysis [37]. Interestingly, A. baumannii lacks lexA [46], suggesting this prophage encodes its own LexA-like protein as a host-independent strategy to detect DNA damage.

Other candidate genes, GO593_10640 and GO593_18720, grouped with the conserved signal recognition particle (SRP) genes ftsY and ffh as well as ybgC, a Tol-Pal-associated thioesterase linked to membrane integrity [47] (Fig 4C). GO593_10640 shows structural similarities to ion transporters and significant chemical responses consistent with ftsY and ffh knockdowns (S8A and S8B Fig). GO593_18720, however, showed CG scores that were smaller in magnitude, suggesting a less critical or defined role (S8A Fig). These genes may be directly or indirectly related to membrane protein localization and membrane integrity, though SRP mechanisms remain understudied in A. baumannii.

Our network analysis revealed a cell wall and division subnetwork (Fig 4D), containing three genes we identified as rare in other Gammaproteobacteria: GO593_05760 (advA), GO593_07665 (blhA), and GO593_11915 (smc2, or SMC_prok_B). These genes correlate with numerous well-characterized, highly conserved division genes, corroborating initial data from other studies suggesting roles for advA, blhA, and smc2 in cell division [3,48]. This subnetwork also featured another gene of unknown function, GO593_07290, which encodes for a SPOR domain-containing protein, annotated here as spor. While SPOR domains are widely conserved and are known to interact with cell wall peptidoglycan (PG) [49], functions of SPOR domain-containing proteins vary considerably. Indeed, one of the most extensively characterized SPOR domain proteins, FtsN from E. coli, does not require the SPOR domain for its essential function [50]. The short periplasmic region required for FtsN function, ^EFtsN [51], does not appear to be present at the amino acid level in GO593_07290, suggesting a functionally distinct role in cell division (S7E Fig).

To further support these genes’ roles in cell division, we generated individual CRISPRi knockdowns for blhA, smc2, and spor and examined cell shape with microscopy. Knockdown of advA has previously been shown to cause filamentation, a hallmark of impaired division [12]. Similarly, blhA, smc2, and spor knockdowns produced elongated or chained cells (Fig 4E and 4F), corroborating prior findings from a blhA transposon mutant [48]. We next attempted to determine the localization pattern of putative division genes using trans expression of fluorescent protein fusions in A. baumannii strain ATCC17978, due to complications caused by phase switching in 19606 Similar to what had been previously shown with AdvA [3], a fusion of superfolder GFP to the N-terminus of BlhA (sfGFP::BlhA) showed localization to the septum, or the site of division, suggesting that BlhA is a bona-fide member of the divisome (S9A Fig). In contrast, the SPOR domain protein localized to both the cell periphery and septum (S9B Fig), consistent with PG binding, but suggests a more generalized cell wall synthesis function. Interestingly, expression of the Smc2::sfGFP hybrid resulted in cell elongation and division defects, with the fluorescently tagged Smc2 protein forming elongated aggregates in the cells (S9C Fig). The impaired division phenotypes were verified with overexpression of Smc2 alone, suggesting precise protein levels are needed for proper cell division (S9D Fig). This trans expression additionally allowed us to disentangle the role of smc2 from polar knockdown effects on downstream gene zapA, a known division gene. As an important caveat, we note that our fusion constructs may not be expressed at the same level as their native counterparts and, therefore, may show aberrant localization. Nonetheless, we contend that these localizations are at least possible. Overall, our network reveals functional connections and underlying phenotypes linking crucial pathways in A. baumannii, such as cell division, and provides a basis for further targeted experiments for mechanistic analysis.

Structure-informed clustering of chemical perturbations reveals insights into inhibitor function

We next aimed to develop an approach for characterizing inhibitor function in A. baumannii that leveraged our chemical genomics dataset. To broadly define inhibitor impacts on cell physiology, we analyzed chemical-gene interactions at the pathway level rather than the individual gene level. This approach evaluated the L₂FCs of all guides, including mismatches, associated with a STRING functional pathway (“pathway effect”) [52]. Pathway level analysis provided increased statistical robustness (many more guides were measured at the pathway level than the gene level) at the cost of single gene resolution; this allowed us to detect many more significant pathways. We then clustered chemicals by pathway-level interactions, showing the relationship between chemicals solely based on the cellular pathways they impacted in our screen (Fig 5A). Inhibitors from the same antibiotic classes, such as quinolones (ciprofloxacin and levofloxacin) or aminoglycosides (amikacin, apramycin, and tobramycin), clustered as expected, validating our method.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Chemical grouping by pathway effects and structure.

(A) Dendrogram depicting hierarchical clustering (Ward method) of screen chemicals by similarities of effects across essential gene pathways. (B) Dendrogram displaying hierarchical clustering (Ward method) of chemical molecular fingerprints (ECFP6) by Tanimoto coefficient. Asterisks represent conditions with multiple chemical structures, in which a single structure was used (see S4 Table). (C) Sina plots depicting differential guide counts per million (CPMs) for protein translocation in carvacrol and thymol, or (D) for heme-copper terminal oxidase in tetracycline-class antibiotics, compared to corresponding no chemical controls. Chemical-gene interactions are described above each graph—up (positive), down (negative), or no change—with FDR values. Guides are weighted in this calculation by predicted efficacy, with perfect guides weighted at 100. Non-significant comparisons are shown with grey backgrounds.

https://doi.org/10.1371/journal.pgen.1011642.g005

To more systematically compare our results to inhibitors outside of established antibiotic classes, we performed molecular fingerprinting—a chemoinformatic approach that numerically represents unique chemical structure and properties of molecules as binary “fingerprints” (extended-connectivity fingerprints) [53,54]—and measured similarity between chemicals using the Tanimoto coefficient [55,56]. Clustering based on distances derived from Tanimoto coefficients verified that antibiotics with similar core structures (beta-lactams, fluoroquinolones, aminoglycosides, etc.) group together (Fig 5B). Comparing the pathway impact clustering with the structure clustering of these chemicals allowed us to investigate (a) consistent chemical classes with a not-yet-established mechanism of action, (b) structurally similar chemicals that have distinct effects on A. baumannii, and (c) structurally dissimilar chemicals that similarly affect A. baumannii physiology.

Carvacrol and thymol are chemicals that cluster tightly by both pathway effect and structure (Figs 5A,5B, and S10). These compounds exhibit antibacterial activity but lack a defined mechanism of action [57], though they have been proposed as antimicrobial agents or additives [58]. Our pathway analysis revealed that genes encoding NADH dehydrogenase (nuo genes) and protein transport complexes (ftsY, ffh, and sec genes) knockdowns were significantly sensitized while ubiquinone (ubi genes) knockdowns were more resistant (Figs 5C,S11A,S11B,S11C, and S11D). In particular, ftsY, encoding the SRP receptor protein, was a major negative outlier in both carvacrol and thymol (S11C and S11D Fig). These pathways influence membrane homeostasis and proton motive force (PMF), supporting previous studies stating that membrane disruption and depolarization underlie carvacrol and thymol antibacterial activity [59]. However, carvacrol and thymol cluster more distantly in their pathway effects from other known membrane disruptors, like chlorhexidine and EDTA [60,61] (Fig 5A). This suggests that the mechanisms of action for these phenolic monoterpenes are unique, potentially due to more specific effects on SRP-mediated protein transport.

Identification of structurally similar chemicals with differing effects on A. baumannii pathways could reveal how relatively small chemical modifications alter their physiological impact. To explore this, we scored how well chemical pairs clustered in pathway effect or in structure and calculated the differences (S12 Fig). We found that the pathways affected by the tetracycline class antibiotics minocycline, tetracycline, and tigecycline diverged despite their strong structural similarity (S13–S15 Figs). Tetracycline class antibiotics target the ribosome, but pathway effects and CG scores were similar for minocycline, tetracycline, and tigecycline treated cells (S14 Fig). The most distinct pathway effects across the tetracycline class occurred in genes encoding the cytochrome bo₃ oxidase, as cyo knockdown strains were severely depleted in tigecycline, but not significantly affected in minocycline and tetracycline (Fig 5D). We observed more subtle, but statistically significant differences in LOS transport and ATP synthesis pathways (S13 Fig), with Lpt encoding genes depleted from cells treated with minocycline and tetracycline but not tigecycline and ATP synthase gene knockdowns enriched in tetracycline and tigecycline but not minocycline. Although the precise mechanisms underlying these interactions are unknown, small changes in structure may affect permeability; for instance, RND-type efflux pumps that are sensitive to the cellular energy state are able to efflux tigecycline and tetracycline but not minocycline [62]. Tigecycline has a unique structural modification (a glycylcycline side chain) that may impact permeability [63] or could contribute to its clustering with other chemicals that affect oxidative phosphorylation and oxidative stress (e.g., phenazine and myricetin [64–67]). Additional work is required to validate and investigate the biological underpinnings of our tetracycline class results.

An important potential caveat our structure-informed clustering approach is that non-bioequivalent doses across chemical inhibitors could drive differential clustering and obscure the relationship between biological targets and pathway effects. One such likely example of this phenomenon in our data set is the carbapenem class antibiotic meropenem, which fails to cluster with other carbapenems we screened (doripenem and imipenem; S15 Fig). CG score (S15A Fig) and pathway analyses (S15B Fig) show a muted effect for meropenem versus other carbapenems, even on the known target pathway. Despite our efforts to normalize growth inhibition across chemicals, the sigmoid nature of dose-response curves can make it challenging to achieve precise bioequivalence in the context of partial inhibition. In the case of meropenem, it is likely that we simply failed to add a sufficient concentration to achieve bioequivalent inhibition. However, we note that our screening quality control steps, such as monitoring the diversity of non-targeting control guides as internal controls (see methods), likely guard against scenarios in which an excessive inhibition has been applied.

Bacterial strains and growth

All strains, plasmids, and oligos used in this study are listed in S1–S3 Tables. Strains were grown in EZ Rich Defined Media (Teknova) supplemented 40 mM sodium succinate as the carbon source (AbRDM) or AbRDM + 1.5% agar at 37°C unless otherwise noted. Selective antibiotics were used when necessary: for E. coli, 100 µg/mL ampicillin (amp) or carbenicillin (carb), 15 µg/mL gentamicin (gent), or 30 µg/mL kanamycin (kan); and for A. baumannii, 10 µg/mL polymyxin B (pmb), 150 µg/mL gentamicin (gent), 60 µg/mL kanamycin (kan). Diaminopimelic acid (DAP) was added at 300 µM to support growth of E. coli dap- donor strains. IPTG (isopropyl b-D-1-thiogalactopyranoside) (0 to 1 mM) was added where indicated in the figures or figure legends. Strains were preserved in 15% glycerol at -80°C.

General molecular biology techniques and plasmid construction

Plasmids, oligos, and other construction details are listed in S2 and S3 Tables. Oligonucleotides were synthesized by Integrated DNA Technologies (Coralville, IA). Plasmid DNA was purified using the GeneJet plasmid miniprep kit (K0503; Thermo Scientific); genomic DNA was purified using the GeneJet genomic DNA purification kit (K0721; Thermo Scientific). DNA fragments were amplified with Q5 DNA polymerase (M0491; New England Biolabs (NEB)) or OneTaq DNA Polymerase (NEB). All enzymes for restriction digests were from NEB. DNA fragments were spin-purified using DNA Clean & Concentrator kit (D4004; Zymo Research) after digestion or amplification. Ligations performed using T4 DNA ligase (M0202; NEB). Plasmids were assembled from restriction enzyme-linearized or PCR-amplified vector and PCR products or synthetic DNA fragments using the NEBuilder Hifi DNA assembly kit (E2621; NEB). Plasmids were transformed into electrocompetent E. coli cells using a Bio-Rad Gene Pulser Xcell on the EC1 setting. Sanger DNA sequencing was performed by Functional Biosciences (Madison, WI); full plasmid sequencing was performed by Plasmidsaurus.

Library growth experiment

The A. baumannii essential gene CRISPRi library (sJMP2949) was diluted 1000-fold from frozen stock (OD600=15) in 50 mL AbRDM (starting OD600= 0.015) and incubated in 250 mL flasks at 37°C with shaking until mid-log (OD600=0.2; ~3.5 hrs)(t0). This culture was then diluted to OD600 =0.02 in 4 mL media with 1mM IPTG and antibiotics (antibiotic stocks diluted 1:100 into media and used immediately; see S4 Table) in 14 mL snapcap culture tubes (Falcon 352051) and incubated with shaking at 37°C. At 18 hours, saturated cultures were serially diluted back to OD600=0.02 into fresh tubes containing the same media and incubated for another 18 hours at 37°C with shaking, for a total of ~14 doublings. Cells were pelleted from 1 mL culture in duplicate before (t0) and after growth in antibiotics.

Sample preparation and sequencing

Genomic DNA was extracted from cell pellets and resuspended in a final volume of 200 μL. The sgRNA-encoding region was amplified using Q5 DNA polymerase (NEB) in a 100 μL reaction with ~100 ng gDNA and primers oJMP697 and oJMP698 according to manufacturer’s protocol, using a BioRad C1000 thermocycler: 30s at 98°C, followed by 16 cycles of 15s at 98°C, 15s at 65°C, 15s at 72°C, then 72°C for 5 min. PCR products were spin-purified and eluted in a final volume of 20 μL.

Samples were sequenced by the UW-Madison Biotech Center Next Gen DNA Sequencing Core facility. PCR products were amplified with nested primers containing i5 and i7 indexes and Illumina TruSeq adapters, cleaned, quantified, pooled, and run on a Novaseq 6000 (150 bp paired-end reads).

Sequencing read quality control and processing

Guides were counted using a custom Python script designed to minimize noise and accurately count guides with overlapping targets. This script samples sequencing reads to identify barcode diversity, orientations, and offsets, and determines flanking sequences for correct barcode identification. This is particularly useful when guides differ by only one or two bases (e.g., mismatch guides) and correctly identifies unintended guides present in the library due to oligo synthesis errors. Only counts for library barcodes were used in downstream analyses here.

Coefficients of determination (R²) between samples were calculated from barcode counts per million (CPM), and conditions with replicates where R² < 0.5 were removed to ensure replicate agreement. Samples were also excluded if population diversity (Nb) was less than 10,000 for nontargeting control guides.

Log₂ fold changes and confidence values were computed using edgeR (S5 Table). Gene-level values were calculated by taking the median guide-level log₂ fold change for perfect match guides; confidence was determined by computing Stouffer’s p-value using guide-level FDR values (poolr R package) (S6 Table). Functional enrichment analyses were conducted using STRING-db version 11.5 with a custom A. baumannii ATCC 19606 proteome (Organism ID: STRG0060QIE).

Ortholog analyses

Genome assemblies of Gammaproteobacteria were collected from the NCBI Reference Sequence database (RefSeq release 213). The dataset comprises 2792 isolates distributed among 474 genera and 18 taxonomic orders.

Essential genes represented in the library were pulled from the genome assembly for A. baumannii ATCC 19606 (GCA_009759685.1). Orthology assignment was computed with the phylogenetic profiling tool fDOG v.0.1.26 [30], compiling a maximum of 25 core-orthologs for each gene using a taxonomic distance minimum of genus and maximum of class. Visualizations were performed using PhyloProfile [83].

The ortholog presence ratio for each gene was measured within a taxonomic group by calculating the fraction of isolates possessing at least one ortholog out of the total number of analyzed isolates. We evaluated these ratios at the genus level with non-ACB (Acinetobacter calcoaceticus-baumannii complex) isolates, at the order level with non-Acinetobacter Moraxellales, and at the class level with non-Moraxellales Gammaproteobacteria. Orthologs and ortholog presence ratios are listed in S7 Table with the full genome dataset and assemblies in S8 Table.

Creation of knockdown strains

As previously described [20,84], sgRNA-encoding sequences were cloned between the BsaI restriction sites of MCi plasmid pJMP2776. Two oligonucleotides encoding an sgRNA were designed to overlap such that when annealed, their ends were complementary to the BsaI-cut ends on the vector.

Next, the Mobile-CRISPRi system was transferred to the attTn7 site on the chromosome of A. baumannii by quad-parental conjugation of three donor strains - one with a mobilizable plasmid (pTn7C1) encoding Tn7 transposase, another with a conjugal helper plasmid (pEVS104), and a third with a mobilizable plasmid containing a Tn7 transposon encoding the CRISPRi system - and the recipient strain A. baumannii ATCC 19606 or ATCC 17978. Briefly, 100 µL of culture of donor and recipient strains were added to 600 µL LB, pelleted at ~8,000 × g, washed twice with LB prior to depositing cells on a nitrocellulose filter (Millipore HAWP02500) on an LB plate, and incubated at 37°C, ~4 h. Cells were removed from the filter by vortexing in 500 μL PBS and selected on LB-gent plates at 37°C.

MIC test strip assays

500 μL of culture (OD=0.1, or 0.5 for LOS- strains) was spread on Mueller-Hinton agar (MHA) plates and allowed to dry for 30 minutes. A sterile MIC test strip was added to the plate and incubated overnight at 37°C. The MIC was determined as the value where the zone of inhibition meets the strip.

Spot assays

Strains were grown on LB agar overnight. Cell scrapes were resuspended from plates into 500 μL PBS, and normalized to an OD₆₀₀ of ~1.0. 10 μL of ten-fold dilutions of these resuspensions were then spotted on AbRDM and AbRDM +1mM IPTG agar plates.

Membrane permeability assay

ATCC 19606 and LOS- strains containing sgRNAs targeting lptA or a non-targeting control (sJMP4539 and 4324, respectively) were grown on LB agar or LB agar containing 1mM IPTG overnight. Cell scrapes were resuspended from plates into 500 μL PBS and normalized to an OD600 of 0.3. Ethidium bromide (EtBr; Bio-Rad #1610433) was added to cell suspensions at a final concentration of 10 ug/mL, and OD600 and fluorescence at 545nm excitation/600 emission were read in a plate reader (Tecan Infinite 200 Pro M Plex). Readings occurred over 100 cycles (~75 min).

Chemical comparison analyses

After edgeR was used to import, organize, filter, and normalize counts, the limma package with voom method was used to perform gene set testing using functional groups predicted by STRING-db (custom proteome; Organism ID: STRG0060QIE). Chemical effects on pathways were scored for direction and significance, then transformed into a correlation matrix. Chemicals were clustered using hierarchical clustering with the Ward method, and a dendrogram visualized the clustering based on pathway impacts.

SMILES strings for each chemical were parsed using the rcdk package in R to generate molecular fingerprints. For combination or mixed-structure antibiotics, a representative structure was selected (e.g., trimethoprim for trimethoprim-sulfamethoxazole or polymyxin E1 for colistin). Tanimoto coefficients were calculated to measure chemical similarity and generate a similarity matrix, which was then clustered using the Ward method to produce a chemical structure dendrogram.

To assess the relationships between chemical pairs, each dendrogram was iteratively cut at all k-values (1-45), and pairs were scored if they clustered together. The final association scores are the sums across all k-values. Differences in association scores between pathway-effect and structure were plotted to highlight disparities. Additionally, the sums of association scores for pathway-effect and structure were plotted to highlight similarities.

Protein structural prediction

Structures of GO593_10640 and GO593_00140 (with chi subunit ortholog) were predicted from amino acid sequence using Alphafold2 [77,85] and AlphaFold-Multimer [86]. Predicted structures were visualized using PyMOL (Schrödinger).

Cell division protein localization and microscopy

Expression vector plasmids were transferred by conjugation into A. baumannii ATCC 17978; briefly, 100 µL of culture of donor and recipient strains were mixed and deposited on a nitrocellulose filter (Millipore HAWP02500) on an LB plate, and incubated at 37°C, ~2 h. Cells were removed from the filter by vortexing in 500 μL PBS and selected on LB-kan plates at 37°C.

Cultures were grown from a starting OD600~0.01 to mid-log at 37°C for ~3 hours in AbRDM + 1mM IPTG (or 500 uM IPTG for Smc2 and Smc2-sfGFP overexpression strains, due to toxicity). 1 mL of culture was centrifuged (8000xg for 2 min) and resuspended in equal volume PBS. Cells were fixed with paraformaldehyde (final concentration 5%) and quenched with equal volume 2.5M glycine. 10 μL of each sample were spotted on a glass slide for microscopy.

Bacteria were imaged with a Nikon Ti-E inverted microscope with an Orca Fusion BT digital CMOS camera (Hamamatsu) using NIS-Elements. Fluorescence images were collected using Prior Lumen 200 metal halide light source and a FITC-specific filter set. Cell length analysis was done with MicrobeJ [87].