Transposon insertion sequencing of Pseudomonas aeruginosa identifies multiple intersecting pathways essential for extreme colistin resistance Authors: Madeleine B. Vessely, Ryan P. Kich, Samuel W. M. Gatesy, Hanna K. Bertucci, Aliki Valdes, Clara Luczak, Shradha Rao, Artur Muszyński, Parastoo Azadi, Casey N. Kellogg, Brandon L. Jutras, John Mekalanos, Alan R. Hauser, Egon A. Ozer, Kelly E. R. Bachta Published: 2026 ABSTRACT Colistin is used to treat antibiotic resistant gram-negative infections, including those caused by Pseudomonas aeruginosa (Pa). Using a diverse collection of clinical isolates, we identified BWH047, a colistin-resistant isolate with an extremely high minimum inhibitory concentration (MIC, 1280 µg/mL). To characterize the genes conditionally essential for colistin resistance in BWH047, we employed transposon insertion sequencing and identified 20 gene candidates. In-frame deletion validated 75% of the candidates and identified genes in several new pathways that contribute to colistin resistance in Pa, including algU and wapH. We also identified several candidate genes from previously reported colistin resistance pathways (e.g., arn, pmrAB). We further investigated the impact of a colistin resistance-associated inner membrane DedA-family undecaprenyl phosphate flippase, which we named DpcA (DedA of Pseudomonas necessary for colistin resistance A). Deletion of dpcA in BWH047 restored sensitivity to colistin (MIC = 0.5 µg/mL) and resulted in several unique changes to the structure of lipopolysaccharide (LPS), including production of decreased amounts of the colistin resistance-conferring 4-amino-4-deoxy-L-arabinose (L-Ara4N) modification on lipid A. This work represents a robust analysis of colistin resistance in Pa and identifies intersecting pathways that contribute to extreme phenotypic resistance. Introduction Pseudomonas aeruginosa (Pa) is a gram-negative (GN) bacterium that causes a variety of healthcare-associated infections and is often extensively antibiotic resistant. Pa quickly acquires resistance to even the newest antimicrobial combinations, forcing clinicians to rely on antibiotics with higher toxicity. Colistin, one such drug, is a cationic antimicrobial peptide (CAMP) with activity against many GN pathogens, including Pa [1]. Although it was once sidelined due to toxic side effects, colistin has been re-introduced into clinical use, particularly in combination with other antimicrobials [2–4]. As such, there has been a rapid increase in colistin-resistant infections following its reintroduction [1]. Colistin (also known as polymyxin E), a cationic amphipathic cyclic lipopeptide antibiotic, is believed to work by disrupting lipopolysaccharide (LPS) in the GN outer membrane [5]. Interaction of colistin’s positively charged residues with the negatively charged phosphates on lipid A of LPS is thought to result in displacement of stabilizing divalent cations (e.g., Mg2+, Ca2+), leading to increased cell permeability and bacterial cell death [5,6]. To date, the major recognized mechanism of acquired colistin resistance involves modification of lipid A with positively charged moieties that decrease the net negative charge of LPS and consequently reduce colistin’s affinity for the bacterial outer membrane [7,8]. Lipid A modifications are mediated by sensor-kinase two component systems (TCS) that sense membrane disruption, cation displacement, or the presence of CAMPs and trigger transcription of genes that result in lipid A modification [9]. Although phosphoethanolamine modification of lipid A drastically increases colistin resistance in other bacterial species (e.g., Escherichia coli, Acinetobacter baumannii, and Klebsiella pneumoniae), this modification has little impact on colistin resistance in Pa [10,11]. In Pa, the only lipid A modification known to confer high-level colistin resistance is the addition of 4-amino-4-deoxy-L-arabinose (L-Ara4N) [9,12]. L-Ara4N synthesis, and its subsequent addition to lipid A, is mediated by the arn operon, which is activated by at least five TCS in Pa [13–20]. Using a patient-derived, multidrug-resistant isolate of Pa with extremely high phenotypic colistin resistance, we identified several new genes conditionally essential for high-level colistin resistance. In addition, we successfully validated well-described resistance mechanisms, lending credence to our approach. We specifically examined the role of a DedA protein, newly named DpcA (DedA of Pseudomonas necessary for colistin resistance A), in colistin resistance using mass spectrometry to characterize LPS and lipid A. We renamed this gene dpcA according to the precedent set in other GN bacterial genera including Burkholderia (dbcA) and Klebsiella (dkcA) [21,22]. We demonstrate that deletion of dpcA reduces the amount of aminoarabinose modifications found on lipid A, alters the glycosylation and acylation patterns of LPS and lipid A, and leads to colistin sensitivity. These results suggest that lipid A modification remains the critical step for colistin resistance in Pa. Our work facilitates a more complete understanding of high level colistin resistance by i) identifying several new resistance pathways, ii) examining the role of dpcA in colistin resistance, and iii) providing molecular insights into how we may prevent resistance to critical last resort antibiotics such as colistin and, by proxy, to naturally occurring cationic antimicrobial peptides. Results Characterization of the PABWH isolate collection From December 2015 through June 2016 at Brigham and Women’s Hospital in Boston, MA, Pa strains were isolated in the clinical microbiology laboratory (n = 100). For this study, 85 isolates (PABWH collection) were chosen representing the first isolate collected from an individual patient (Fig 1A). Whole genome sequencing was performed and revealed considerable strain diversity, including 54 distinct sequence types (STs), of which 4 were novel (ST-6655, 6656, 6658, 6659) (Fig 1A, S4 Table). Broth microdilution (BMD) was performed in triplicate to establish robust minimum inhibitory concentrations (MICs) for aztreonam, cefepime, ceftazidime, ciprofloxacin, meropenem, piperacillin-tazobactam, and colistin (Figs 1B, S1, S4 Tables). Within the PABWH collection, 25.9% of strains were non-susceptible (intermediate or resistant) to aztreonam, 18.8% to cefepime, 15.3% to ceftazidime, 51.8% to ciprofloxacin, 25.9% to meropenem, and 21.2% to piperacillin-tazobactam (S1 Fig, S4 Table). In total, four of the isolates (4.7%; PABWH066, PABWH074, PABWH088, PABWH103) met criteria for difficult-to-treat (DTT) P. aeruginosa based on the current definition [27]. Phylogenetic tree of PABWH isolates and their resistance profiles. (A) A maximum likelihood core genome phylogenetic tree was generated with the 85 strains in the PABWH isolate collection. BWH047 is denoted with an asterisk (*). Sequence type is indicated by the colored circles at the tree tips according to the legend. Isolate names are colored by Type 3 Secretion genotype (exoS, blue; exoU, brown; neither, gray; exoSU, pink). Source of isolation is displayed in the inner ring (blood, dark red; sputum, gray; urine, light yellow; wound/skin, purple). Susceptibilities to the indicated antibiotics are displayed in the outer rings (sensitive, green; intermediate, yellow; resistant, red). Susceptibility status was determined according to the CLSI guidelines (M100, Ed. 35) for each antibiotic [23]. (B) Colistin MIC (µg/mL), as determined by broth microdilution (BMD), is presented on the x-axis and number of isolates with each MIC is presented on the y-axis. Isolates with intermediate (yellow) and resistant (red) MICs are highlighted. BMD was performed at both standard concentrations and 10x standard concentrations to determine the actual MIC of isolate BWH047.Isolate BWH047 was a notable outlier within the PABWH collection due to its high colistin MIC (≥128 µg/mL). To determine the exact colistin MIC, we measured MICs at concentrations 10-fold (2.5-1,280 μg/mL) above those typically used for colistin. We found that the actual MIC of BWH047 was 1,280 μg/mL, among the highest ever reported for Pa [24–26]. The only other isolate within the PABWH collection that was colistin-resistant had an MIC of 4 μg/mL—the lowest MIC in the ‘resistant’ category according to Clinical and Laboratory Standards Institute (CLSI) guidelines, M100, Ed.35—and 320 times lower than BWH047’s MIC [23] (Fig 1B). Of note, in 2019, the CLSI updated the clinical breakpoints for colistin in Pa so that there is no longer a susceptible category—a colistin MIC of < 4 μg/mL is categorized as intermediate, while a resistant isolate has an MIC ≥ 4 μg/mL. As such, MICs presented in this paper are interpreted according to the most recent CLSI guidelines [23]. Long-read sequencing analysis of BWH047 revealed that it did not possess any plasmids and the sequence did not contain any of the reported mobile colistin resistance genes mcr 1-9. Therefore, we decided to investigate genetic mechanisms driving BWH047’s extreme colistin resistance. Identification of genes conditionally essential for extreme colistin resistance in BWH047 To address the question of why BWH047 exhibits such a high colistin MIC, we utilized transposon insertion sequencing (TnSeq). TnSeq is a technique that can be used to identify genes conditionally essential for bacterial survival in the presence of a stressor [28]. We created a BWH047 transposon (tn) library and subsequently exposed it to colistin at ¼ (320 µg/mL) and ½ (640 µg/mL) MICs (S2A Fig). Out of 6,424 total genes in the BWH047 genome, we were able to assess the impact of colistin exposure on transposon insertion recovery in 4,157 genes in the 320 µg/mL condition and 4,371 in the 640 µg/mL condition (S5, S6 Tables). For the remainder, we were unable to determine log2FC because there were too few reads per gene for analysis. Of note, BWH047 also has a 90,698 bp region within the bacterial chromosome at position 2,405,801–2,496,458 bp containing a ~45kbp duplication. The impact of transposon insertions in this region could not be assessed due to diploidy (S2B Fig). Insertion site reads were mapped to the BWH047 genome, and library density and quality were assessed in control (MHB II only) conditions using TIS tools [29]. After filtering, the BWH047:tn library consisted of 16,913 unique transposon insertion sites as defined by ≥ 3 reads in all control pools (n = 5) (S3 Fig). In total, 27 genes passed our threshold (log2FC ≥ -2.0, FDR < 0.05) in 640 μg/mL colistin, and 28 genes met the same criteria in 320 μg/mL colistin (Fig 2, Table S7). Of these, 20 candidate genes were shared in both (Fig 2, Table S7). These genes were deemed candidate conditionally essential genes for BWH047 colistin resistance. BWH047 genes conditionally essential for colistin resistance. Volcano plots showing TnSeq results following exposure of the BWH047 transposon library to (A) 320 and (B) 640 μg/mL colistin for 18 hours. The log2 fold change (log2FC) is presented on the x-axis and the inverse log10 of the false discovery rate (FDR) is presented on the y-axis. Red circles indicate conditionally essential genes that passed filtering conditions (log2FC ≤ -2.0, -log10FDR ≥ 1.3). All 20 overlapping conditionally essential genes are labeled in each upper right hand inset. (C) Graphical summary of TnSeq results by target gene validation status. To validate the TnSeq candidates, we attempted to make in-frame deletions of all of genes on the shared candidate list (Table 1, Fig 2A, 2B). We successfully created in-frame deletions in 17 out of 20 genes. For those we were unable to delete (galU, topA, and pgi), we found supporting data to suggest that they may be conditionally essential in certain media types and growth conditions, which may explain why bacteria with transposon insertions in these genes survived during MHBII growth, but we were unable to generate deletions [30,31]. Additionally, pgi and galU were previously identified as important for colistin resistance in a resistant Pa strain with a colistin MIC of 8 μg/mL, although neither gene target was independently validated [18]. 10.1371/journal.ppat.1014168.t001Table 1Colistin minimum inhibitory concentration in deletion mutants and complements of conditionally essential genes identified by TnSeq.BWH047 StrainPAO1 OrthologPutative function of gene productTnSeq HitColistin MIC (μg/mL)Validated TnSeq Hitwildtype (WT)––1280∆mexAPA0425multidrug efflux RND transporter periplasmic adaptor subunit MexA1280 ∆mexB PA0426multidrug efflux RND transporter permease subunit MexBY640* ∆oprM PA0427multidrug efflux RND transporter outer membrane channel subunit OprMY8+ ∆dpcA PA4029DedA family proteinY0.5+ ∆dpcA + dpcA a –Chromosomally complemented ∆dpcA strain640 ∆algU PA0762RNA polymerase sigma factor AlgU (also known as RpoE)Y0.5+ ∆shaC PA1056monovalent cation/H+ antiporter subunit DY1280 ∆shaF PA1059Na + /H+ antiporter subunit GY640 ∆mexY PA2018multidrug efflux RND transporter permease subunit MexYY8+ ∆galU b PA2023UTP-glucose-1-phosphate uridylyltransferase GalUY-- ∆topA b PA3011type I DNA topoisomeraseY-- ∆arnB PA3552UDP-4-amino-4-deoxy-L-arabinose aminotransferase0.5 ∆arnC PA3553undecaprenyl-phosphate 4-deoxy-4-formamido-L-arabinose transferaseY0.5+ ∆arnC + arnC a –Chromosomally complemented ∆arnC strain640 ∆arnA PA3554bifunctional UDP-4-amino-4-deoxy-L-arabinose formyltransferase/UDP-glucuronic acid oxidase ArnAY0.5+ ∆arnT PA3556lipid IV(A) 4-amino-4-deoxy-L-arabinosyltransferase0.5 ∆arnEF d PA3557, PA35584-amino-4-deoxy-L-arabinose-phosphoundecapronol flippase subunits ArnE & ArnF0.5 ∆ugd PA3559UDP-glucose 6-dehydrogenaseY0.5+ ∆ugd + ugd a –Chromosomally complemented ∆ugd strain640 ∆pgi b PA4732glucose-6-phosphate isomeraseY-- ∆carB PA4756carbamoyl-phosphate synthase large subunitY160+ ∆speD2 PA4773adenosylmethionine decarboxylaseY1280* ∆speE2 PA4774spermidine synthaseY1280* ∆speH PA4775hypothetical proteinY640* ∆pmrA PA4776two-component system response regulator PmrAY0.5+ ∆pmrB PA4777two-component system sensor histidine kinase PmrBY0.5+ ∆dnpA PA5002PIG-L family deacetylaseY1280* ∆mig14 c PA5003antimicrobial resistance protein Mig-14-- ∆wapH PA5004glycosyltransferase1 ∆wapO PA5005carbamoyltransferase160a ∆dpcA was complemented at the neutral Tn7 site under a constitutive promoter. ∆arnC and ∆ugd were complemented in their native genomic positions under their native promoters.b Though identified as conditionally essential, deletion mutant was unable to be createdc We were unable to delete this gene. It was not an original TnSeq candidate but was located in an operon of interest.d MIC for ∆arnEF were assayed only in the double deletion mutant. Neither gene was identified as conditionally essential by TnSeq analysis.+, hits determined as conditionally essential by the TnSeq analysis that, when deleted, resulted in ≥8-fold decrease in colistin MIC from the parent strain*, hits determined as conditionally essential by the TnSeq analysis that, while not essential for colistin resistance themselves, neighbored genes that were, by deletion and MIC analysis, essential for colistin resistance (polar effects) Colistin minimum inhibitory concentration in deletion mutants and complements of conditionally essential genes identified by TnSeq. a ∆dpcA was complemented at the neutral Tn7 site under a constitutive promoter. ∆arnC and ∆ugd were complemented in their native genomic positions under their native promoters. b Though identified as conditionally essential, deletion mutant was unable to be created c We were unable to delete this gene. It was not an original TnSeq candidate but was located in an operon of interest. d MIC for ∆arnEF were assayed only in the double deletion mutant. Neither gene was identified as conditionally essential by TnSeq analysis. +, hits determined as conditionally essential by the TnSeq analysis that, when deleted, resulted in ≥8-fold decrease in colistin MIC from the parent strain *, hits determined as conditionally essential by the TnSeq analysis that, while not essential for colistin resistance themselves, neighbored genes that were, by deletion and MIC analysis, essential for colistin resistance (polar effects) Once candidate essential gene targets were deleted, all deletion strains were assayed by BMD to determine colistin MIC (Table 1). Ten deletion mutants had colistin MICs that were reduced by at least eight-fold compared to BWH047 (parental isolate), including seven which had MICs of 0.5 μg/mL, well below the CLSI threshold for Pa colistin resistance and consistent with previously susceptible isolates [23] (Fig 2C, Table 1). Hits were considered to be validated if clean deletion of the gene resulted in a≥8-fold decrease in colistin MIC from the parent strain, suggesting a role in extreme colistin resistance. Validated hits included: several members of the arn operon (arnA, arnC, ugd), the two-component system pmrAB, the carbamoyl synthesis gene carB, the resistance nodulation division (RND) efflux proteins mexY and oprM, the sigma factor algU (otherwise known as rpoE), and a DedA-family undecaprenyl phosphate (UndP) carrier protein hereafter designated dpcA for DedA of Pseudomonas aeruginosa required for colistin resistance A. While many of these genes and pathways have been associated with colistin resistance in prior studies in Pa and in other bacteria, this work represents a comprehensive analysis of the conditionally essential genes in a highly colistin resistant clinical isolate. Below, we discuss each of these gene groups in more detail. Deletion of arnA, arnC, or ugd resulted in a 2,560-fold decrease in colistin MIC from 1280 µg/mL to 0.5 µg/mL (Table 1). The arnBCADTEF-ugd operon is responsible for the biosynthesis and addition of 4-amino-4-deoxy-L-arabinose (L-Ara4N) to the phosphate moieties on lipid A and is a well-characterized mechanism of colistin resistance in many bacteria, including Pa [18,21,32]. Therefore, the identification of multiple genes located within this operon served as an internal positive control that validated our TnSeq experiment. Although we only identified three members of this operon in our experiment, it is well established that all genes in the arn operon are necessary for the synthesis and addition of L-4AraN to lipid A [33,34]. Thus, we investigated the conditional essentiality of additional genes in this operon (arnB, arnT, arnE/F) and found that all deletions resulted in a uniform colistin MIC of 0.5 μg/mL (Table 1). Finally, we successfully restored BWH047’s extreme colistin resistance phenotype through complementation of ΔarnC and Δugd with their native alleles, whereby ∆arnC + arnC and ∆ugd + ugd had colistin MICs of 640 μg/mL, which is within one 2-fold dilution of the parental isolate and within the acceptable range of error of the BMD assay (Table 1). Five two-component systems (TCS) that have been shown to induce arn expression in Pa—PmrAB, PhoPQ, CprRS, ColRS, and ParRS [12,17–20,35,36]. In BWH047, only one of these TCS was identified as conditionally essential for colistin resistance—PmrAB. As with the arn operon, deletion of either gene resulted in a 2,560-fold decrease in colistin MIC from 1280 µg/mL to 0.5 µg/mL (Table 1). Additionally, TnSeq identified three candidate targets near pmrAB including speD2, speE2, and speH; however, deletion of any of these three candidates resulted in no change in colistin MIC (Table 1). Intriguingly, these three candidate genes are located directly upstream of pmrAB in a proposed operon (S2C Fig), providing a likely explanation for their identification in the TnSeq experiment despite our inability to validate them. Whether speD2, speE2, speH, pmrA, and pmrB are all co-transcribed or are expressed as two separate operonic units is unclear [36,37]. TnSeq also identified conditionally essential targets involved in the modification of LPS core oligosaccharides (core-OS). Deletion of carB, a carbamoyl phosphate synthase that works with putative carbamoyl transferase WapO to carbamoylate sugar residues in the core-OS of Pa [38,39], resulted in a colistin MIC of 160 μg/mL, an 8-fold decrease from BWH047. Deletion of wapO, though not identified as a candidate in the original TnSeq analysis, resulted a similar MIC (160 µg/mL). One of our non-validating targets, dnpA (MIC 1280 µg/mL), was also located within the same operon as wapO (wapO-wapH-mig14-dnpA). Thus, we elected to investigate the conditional essentiality of the remaining genes in this operon, including wapH and mig14 (S2C Fig). To this end, we successfully generated a clean deletion of wapH. We were unable to delete the hypothetical gene mig14, which has been previously reported for Pa [40]. The colistin MIC of ΔwapH was 1 μg/mL (Table 1), which supports the hypothesis that transposon insertions in dnpA may have influenced expression of the remaining genes within the same operon. Polar transposon effects may have led to dnpA’s identification as conditionally essential when the true conditionally essential genes (wapH and wapO) were located elsewhere in the same operon. Both mexY and oprM were identified as conditionally essential for colistin resistance, and deletion of either resulted in a colistin MIC of 8 µg/mL. These genes code for members of the MexAB-OprM and MexXY-OprM resistance-nodulation-division (RND) transporters, which are well-characterized tripartite efflux systems in Pa. Both extrude a wide variety of antimicrobials and promote antimicrobial resistance; however, neither has been shown to export colistin or polymyxin B [41]. MexY spans the inner membrane and OprM is the outer membrane channel protein that is shared by several efflux systems including both MexAB-OprM and MexXY-OprM [42]. Interestingly, we also identified mexB as a candidate; however, deletion of mexB resulted in a MIC similar to the parental isolate (640 µg/mL). This suggests that the MexXY-OprM efflux pump, but not the MexAB-OprM efflux pump, contributes to colistin resistance in BWH047. Like speD2, speE2, and speH, we hypothesize that transposon insertions in mexB led to polar effects impacting the expression of oprM, which resides downstream of mexAB in a transcriptional unit (S2C Fig). Therefore, genetic proximity likely led to erroneous identification of mexB as a candidate. One of the more intriguing candidates identified was the sigma factor algU, which, when deleted, resulted in a colistin MIC of 0.5 µg/mL. AlgU, also known as rpoE in other GN bacteria, encodes the extracellular sigma factor (ECF) σ22 in Pa, which is most often studied in the context of mucoidy in chronic infections [43,44]. Next, we identified a DedA-family undecaprenyl phosphate flippase, named dpcA, as conditionally essential for colistin resistance in BWH047. Deletion of dpcA also resulted in a colistin MIC of 0.5 µg/mL, which could be complemented by expression of dpcA at a neutral site in the chromosome (Table 1). Previous work in Pa demonstrated that DpcA is critical for colistin resistance in the context of arn overexpression, that deletion of dpcA decreases the frequency of spontaneous colistin resistance mutations, and that DpcA binds undecaprenyl phosphate (UndP) [45]. However, the role of DpcA in aminoarabinosylation of lipid A was not confirmed. Finally, we were unable to validate a role in colistin resistance for five experimentally identified candidate genes: shaC, shaF, galU, topA, and pgi, either because deletion had no impact on colistin MIC (shaC, MIC 1280 µg/mL, and shaF, MIC 640 µg/mL; Table 1) or we were unable to generate in-frame deletions (galU, topA, and pgi). After accounting for genomic architecture, the overall validation rate of our experiment increased from 10/20 to 15/20, or 75% (Fig 2C), which aligns with similar TnSeq experiments [46–49]. Lipopolysaccharide (LPS) analysis of BWH047 WT and ∆dpcA Given the enrichment of lipopolysaccharide (LPS) modification genes in our experimental results and the fact that the bacterial outer membrane and its structures serve as the binding target for colistin, we hypothesized that the LPS structure in BWH047 could be contributing to its uniquely elevated colistin MIC. Thus, we decided to characterize the LPS of BWH047 and its colistin-sensitive derivatives ΔdpcA and ∆wapH. LPS from BWH047, ΔdpcA, and ΔwapH was isolated utilizing the hot phenol-water method [50] and visualized by silver staining after deoxycholate-polyacrylamide gel electrophoresis (DOC-PAGE). The laddering of O-antigen was reduced in the ΔdpcA strain compared to the WT BWH047 strain, consistent with the proposed role of DpcA in UndP recycling and UndP’s involvement in transporting O-antigen subunits (S4A Fig) [51,52]. ∆wapH exhibited an even more extreme truncated version of LPS lacking any visible O-antigen laddering (S4B Fig). LPS of BWH047 and ΔdpcA was further hydrolyzed with hydrochloric acid in methanol to determine the glycosyl and fatty acid residues (S4C and S4D Fig). Comparative GC-MS analysis showed that both strains contained glycosyl residues that could be attributed specifically to: (i) the O-antigen of Pa (N-acetylfucosamine (FucNAc) and N-acetylquinovosamine (QuiNAc)), (ii) the core-OS (3-deoxy-D-manno-oct-2-ulosonic-acid (Kdo)), and (iii) either in the core-OS or the O-antigen, (N-acetylgalactosamine (GalNac), glucose (Glc), N-acetylglucosamine (GlcNAc), mannose (Man), and rhamnose (Rha)). Heptose could not be identified in the applied chemical derivatization method, likely due to complete phosphorylation of the residue as previously reported [53]. In addition, we observed differences in the relative molar percentage of glycosyl residues in BWH047 and ΔdpcA LPS (Table 2). In ΔdpcA, we found a significant decrease in the relative content of FucNAc (50% to 1.2%) and relative increases in Rha (2.3% to 9.6%), Glc (40.7% to 66.2%), Kdo (2.5% to 14.6%), and GalNAc (0.7% to 3.6%). The relative content of GlcNAc also increased in ΔdpcA (2.7% compared to 0.2%). These findings are all consistent with the decrease of O-antigen laddering observed in ΔdpcA (S4A Fig). Of note, L-Ara4N was the only glycosyl residue attributed to lipid A that decreased in relative mole percentage in ΔdpcA (1.9%) compared to the parental isolate (3.0%). 10.1371/journal.ppat.1014168.t002Table 2Relative molar percentage of the glycosyl composition of lipopolysaccharide from BWH047 WT and ΔdpcA as determined by gas chromatography-mass spectrometry.Glycosyl ResidueStrainBWH047ΔdpcARhamnoseRha2.39.64-amino-4-deoxy-L-arabinoseaL-Ara4N3.01.9N-acetyl fucosamineaFucNAc501.2MannoseMan0.40.1GlucoseGlc40.766.23-deoxy-D-manno-oct-2-ulosonic acidKdo2.514.6N-acetylquinovosamiosylaQuiNAc0.1tr.N-acetylgalactosamineGalNac0.73.6L-glycero-D-manno-heptoseHeptr.tr.N-acetylglucosamineGlcNAc0.22.7aDue to the unavailability of authentic standards of FucNAc, L-Ara4N, and QuiNAc, we used the relative response factor of GlcNAc to calculate the relative mole percentage. Trace (tr.) was defined as a relative mole percentage < 0.1%. Relative molar percentage of the glycosyl composition of lipopolysaccharide from BWH047 WT and ΔdpcA as determined by gas chromatography-mass spectrometry. aDue to the unavailability of authentic standards of FucNAc, L-Ara4N, and QuiNAc, we used the relative response factor of GlcNAc to calculate the relative mole percentage. Trace (tr.) was defined as a relative mole percentage < 0.1%. We also compared the fatty acid compositions of BWH047 and ΔdpcA LPS using GC-MS (Table 3, S4C, S4D Fig). The LPS of BWH047 contained 3-hydroxydecanoic acid (10:0(3-OH)), 2-hydroxydodecanoic acid (12:0(2-OH)), 3-hydroxydodecanoic acid (12:0(3-OH)), hexadecanoic acid (palmitic acid, 16:0), dodecenoic acid (lauric acid, 12:0), and cis-8-octadecanoic acid (18:1∆8). ΔdpcA showed increased amounts of 2- and 3-hydroxydodecanoic acid compared to BWH047 (21% vs. 15% and 52% vs. 43%, respectively), and a reduced amount of palmitic acid (6% vs. 16%). We did not detect cis-8-octadecanoic acid in ΔdpcA LPS. Overall, fatty acid composition analysis of BWH047 and ΔdpcA LPS revealed both a high degree of heterogeneity in lipid A acylation status and isolate-specific acylation patterns. 10.1371/journal.ppat.1014168.t003Table 3Comparative fatty acid analysis (FAME) of lipopolysaccharide isolated from BWH047 WT and ΔdpcA.FAME (TMS)aStrain BWH047 ΔdpcA 3-hydroxydecanoic acid10:0 (3-OH)17182-hydroxydodecanoic acid12:0 (2-OH)15213-hydroxydodecanoic acid12:0 (3-OH)4352dodecanoic (lauric) acid12:043hexadecanoic (palmitic) acid16:0166cis-8-octadecanoic acid18:1∆85n.d.aValues presented are calculated based on relative response from the GC-MS El detector.Abbreviations: not detected (n.d.) Comparative fatty acid analysis (FAME) of lipopolysaccharide isolated from BWH047 WT and ΔdpcA. aValues presented are calculated based on relative response from the GC-MS El detector. Abbreviations: not detected (n.d.) Lipid A analysis of BWH047 WT and ∆dpcA To better characterize the structure of lipid A and to interrogate levels of L-Ara4N modification and the differences in fatty acid distribution, we released lipid A from LPS with a mild sodium acetate treatment and analyzed using MALDI-TOF MS (Matrix-Assisted Laser Desorption/Ionization Time-of-Flight Mass Spectrometry) (Fig 3A, 3B). For the peaks for which we were able to assign chemical compositions (Table 4), we calculated relative percentages of total lipid A with the indicated number of L-Ara4N modifications and acyl chains. A higher percentage of lipid A structures were modified by ≥ 1 aminoarabinose moiety in BWH047 compared to ΔdpcA (38.9% vs. 13.4%, respectively) (Table 5). Additionally, BWH047 had more penta-acylated lipid A compared to ∆dpcA (89.3% vs. 76.5%) and ∆dpcA had higher levels of tetra- (12.3% vs. 4.3%) and hexa-acylated lipid A (11.2% vs. 6.4%). Penta-acylation was the most predominant pattern in both WT and ΔdpcA (Table 6). 10.1371/journal.ppat.1014168.t004Table 4Proposed composition of selected peaks from BWH047 and ΔdpcA lipid A identified by Matrix-Assisted Laser Desorption/Ionization Time-of-Flight (MALDI-TOF) mass spectrometry.LabelBWH047∆dpcA[M-H-]Proposed Chemical Compositiona Observed m/z b Calculated m/z A1167.81167.7P, GlcNAc2, 10:0(3-OH), 12:0, [12:0(3-OH)]2B’1179.81179.81179.8P, GlcNAc2, [12:0]2, [12:0(3-OH)]2B1195.81195.8P, GlcNAc2, 12:0(2-OH), [12:0(3-OH)]2, 12:0C1247.81247.7P2, GlcNAc2, 10:0(3-OH),12:0, [12:0(3-OH)]2D’1259.81259.7P2, GlcNAc2, [12:0]2,12:0(3-OH)2D1275.81275.7P2, GlcNAc2, 12:0(2-OH), [12:0(3-OH)]2, 12:0D’ Na1281.81281.8P2, GlcNAc2, [12:0]2,12:0(3-OH)2 + NaE’1350.01350.01349.9P, GlcNAc2, 10:0(3-OH), [12:0]2, [12:0(3-OH)]2E1366.01366.01365.9P, GlcNAc2, 10:0(3-OH), 12:0(2-OH), [12:0(3-OH)]2, 12:0F’1376.01375.9P, GlcNAc2, 18:1∆8, [10:0(3-OH)]2, 10:0, 12:0 (3-OH)G1378.01378.01378.8L-Ara4N, P2, GlcN2, 10:0(3-OH), 12:0, [12:0(3-OH)]2F1392.01391.9P, GlcNAc2, 18:1∆8, [10:0(3-OH)]2, 10:0(2-OH), 12:0 (3-OH)H1402.01401.91401.8P2, GlcNAc2, [10:0(3-OH)]2, [12:0]2, 12:0(3-OH)I1404.11404.11404.0P, GlcNAc2, 18:1∆8, [10:0(3-OH)]2,12:0, 12:0(3-OH)J’1418.11417.8P2, GlcNAc2, [10:0(3-OH)]2, (12:0), [12:0(3-OH)]2K1422.01422.0P, GlnNAc2, 16:0, 10:0(3-OH), 12:0(2-OH), [12:0(3-OH)]2I Na1426.01426.11426.0P, GlcNAc2, 18:1∆8, [10:0(3-OH)]2,12:0, 12:0(3-OH) + NaL’1429.91430.01429.9P2, GlcNAc2,10:0(3-OH), [12:0]2, [12:0(3-OH)]2J1433.01433.8P2, GlcNAc2, [10:0(3-OH)]2, [12:0(2-OH)]2, 12:0(3-OH)L1445.91445.91445.9P2, GlcNAc2,10:0(3-OH), 12:0(2-OH), [12:0(3-OH)]2, 12:0L’ Na1451.91451.91451.9P2, GlcNAc2,10:0(3-OH), [12:0]2, [12:0(3-OH)]2 + NaL Na1467.91467.91467.9P2, GlcNAc2,10:0(3-OH), 12:0(2-OH), [12:0(3-OH)]2, 12:0 + NaM’1481.01481.01481.0L-Ara4N, P, GlcNAc2, 10:0(3-OH), [12:0]2, [12:0(3-OH)]2M1497.01497.01497.0L-Ara4N, P, GlcNAc2, 10:0(3-OH), 12:0(2-OH), [12:0(3-OH)]2, 12:0N’1520.11520.0P, GlcNAc2, [10:0(3-OH)]2, [12:0]2, [12:0(3-OH)]2N1536.11536.0P, GlcNAc2, [10:0(3-OH)]2, 12:0(2-OH), [12:0(3-OH)]2, 12:0O1537.11537.0L-Ara4N, P, GlcNAc2, 16:0, 10:0(3-OH), 12:0, [12:0(3-OH)]2P’1561.01561.01560.9L-Ara4N, P2, GlcNAc2, 10:0(3-OH), [12:0]2, [12:0(3-OH)]2P1577.01576.9L-Ara4N, P2, GlcNAc2, 10:0(3-OH), 12:0(2-OH), [12:0(3-OH)]2, 12:0P’ Na1583.01583.01582.9L-Ara4N, P2, GlcNAc2, 10:0(3-OH), [12:0]2, [12:0(3-OH)]2 + NaQ’1588.21588.1P, GlcNAc2, 16:0, 10:0(3-OH), (12:0)2, [12:0(3-OH)]2P Na1599.01599.01598.9L-Ara4N, P2, GlcNAc2, 10:0(3-OH), 12:0(2-OH), [12:0(3-OH)]2, 12:0 + NaR’1600.01600.11600.0P2, GlcNAc2, [10:0(3-OH)]2, (12:0)2, [12:0(3-OH)]2Q1604.11604.21604.1P, GlcNAc2, 16:0, 10:0(3-OH), 12:0(2-OH), [12:0(3-OH)]2, 12:0R1616.11616.11616.0P2, GlcNAc2, [10:0(3-OH)]2, 12:0(2-OH), [12:0(3-OH)]2, 12:0S’1668.21668.21668.1P2, GlcNAc2, 16:0, 10:0(3-OH), [12:0]2, [12:0(3-OH)]2S1684.21684.1P2, GlcNAc2, 16:0, 10:0(3-OH), 12:0(2-OH), [12:0(3-OH)]2, 12:0T’1692.11691.21691.9L-Ara4N2, P2, GlcNAc2, 10:0(3-OH), [12:0]2, [12:0(3-OH)]2T1708.11708.0L-Ara4N2, P2, GlcNAc2, 10:0(3-OH), 12:0(2-OH), [12:0(3-OH)]2, 12:0T’ Na1714.11714.01714.0L-Ara4N2, P2, GlcNAc2, 10:0(3-OH), [12:0]2, [12:0(3-OH)]2 + NaU1719.21719.31719.2L-Ara4N, P1, GlcNAc2, 16:0, 10:0(3-OH), [12:0]2, [12:0(3-OH)]2T Na1730.01730.0L-Ara4N2, P2, GlcNAc2, 10:0(3-OH), 12:0(2-OH), [12:0(3-OH)]2, 12:0 + Naa Due to the large number of species present in each strain, we assigned chemical compositions for a selection of peaks. Chemical structures of species with proposed compositions are depicted in Figs 3C and S6. Raw signal integration values are listed in S8 Table.b The m/z values obtained in the negative reflector were compared and matched with the calculated monoisotopic ions for the most probable structures.Abbreviations: phosphate (P), N-acetyl-glucosomine (GlcNAc), 4-amino-4-deoxy-L-arabinose (L-Ara4N), sodium (Na), 3-hydroxydecanoic acid (10:0 (3-OH)), 2-hydroxydodecanoic acid (12:0 (2-OH)), 3-hydroxydodecanoic acid (12:0 (3:OH)), dodecanoic (lauric) acid (12:0), hexadecenoic (palmitic) acid (16:0), cis-8-octadecanoic acid (18:1∆8) Proposed composition of selected peaks from BWH047 and ΔdpcA lipid A identified by Matrix-Assisted Laser Desorption/Ionization Time-of-Flight (MALDI-TOF) mass spectrometry. a Due to the large number of species present in each strain, we assigned chemical compositions for a selection of peaks. Chemical structures of species with proposed compositions are depicted in Figs 3C and S6. Raw signal integration values are listed in S8 Table. b The m/z values obtained in the negative reflector were compared and matched with the calculated monoisotopic ions for the most probable structures. Abbreviations: phosphate (P), N-acetyl-glucosomine (GlcNAc), 4-amino-4-deoxy-L-arabinose (L-Ara4N), sodium (Na), 3-hydroxydecanoic acid (10:0 (3-OH)), 2-hydroxydodecanoic acid (12:0 (2-OH)), 3-hydroxydodecanoic acid (12:0 (3:OH)), dodecanoic (lauric) acid (12:0), hexadecenoic (palmitic) acid (16:0), cis-8-octadecanoic acid (18:1∆8) 10.1371/journal.ppat.1014168.t005Table 5Relative percentage of lipid A structures with aminoarabinose modification(s).Strain # of L-Ara4N residues BWH047 ΔdpcA 061.1a86.6129.012.629.90.8aRelative percentages were calculated using relative intensity of peaks with proposed compositions listed in Table 4. Relative percentage of lipid A structures with aminoarabinose modification(s). aRelative percentages were calculated using relative intensity of peaks with proposed compositions listed in Table 4. 10.1371/journal.ppat.1014168.t006Table 6Relative percentage of lipid A structures with tetra-, penta-, or hexa-acyl chains.Strain # of acyl chains BWH047 ΔdpcA Tetra4.3a12.3Penta89.376.5Hexa6.411.2aRelative percentages were calculated using relative intensity of peaks with proposed compositions listed in Table 4. Relative percentage of lipid A structures with tetra-, penta-, or hexa-acyl chains. aRelative percentages were calculated using relative intensity of peaks with proposed compositions listed in Table 4. MALDI-TOF analysis of lipid A from BWH047 and ΔdpcA. Comparative Matrix-Assisted Laser Desorption/Ionization Time-of-Flight (MALDI-TOF) analysis of lipid A from BWH047 (A) and ΔdpcA (B) released by mild hydrolysis with sodium acetate at pH 4.5. The spectra were acquired in negative ionization mode ([M-H]-). The x-axis represents the mass to charge ratio (m/z) and the y-axis represents the intensity of each peak in arbitrary units (a.u.) x 105. (C) Structural predictions of selected peaks (identified by letter in panels A and B) differentially present in either BWH047 or ΔdpcA and their corresponding predicted m/z values. The detailed assignments of all peaks are listed in Table 4 and S6 Fig. For several prominent peaks in the BWH047 spectrum (m/z 1375.9 (Peak F’), 1391.9 (Peak F), and 1404.4 (Peak I), Fig 3A), we had difficulty solving structures. We elected to perform collision-induced tandem mass spectrometry (CID MS/MS) analysis of all three selected ions, which pointed to pentaacylated lipid A species substituted with one acyloxy-acyl cis-8-octadecanoic acid (18:1∆8) on C2 of GlcNAc II (F’, F, I, I Na) (S5 Fig). The lipid A species containing 18:1∆8 were only observed in WT. Peptidoglycan analysis of BWH047 and ∆dpcA Based on the role of DpcA as an UndP flippase and UndP’s essential role in peptidoglycan (PG) precursor transport [45,51], we postulated that ΔdpcA may have an altered PG composition compared to WT. However, liquid chromatography-mass spectrometry (LC-MS) PG composition analysis of BWH047, ΔdpcA, and ΔdpcA-comp strains showed identical PG profiles (S7 Fig; S12 Table). We were initially surprised by these results, but our observations are supported by recent data demonstrating that PG profiles are similar between PAO1 and PAO1ΔdpcA [54]. Klycheva et al. observed that loss of dpcA (which they refer to as dedA4) resulted in accumulation of the cytoplasmic PG precursor UDP-MurNAC pentapeptide but did not impact overall PG structure. Comparative expression analysis of pmrA, arnB, and dpcA Previous work has demonstrated that expression of pmrAB and the arn operon is induced upon colistin exposure. We sought to determine whether the expression of dpcA was also induced by colistin. To do so, we measured the expression of dpcA, arnB, and pmrA by digital PCR (dPCR) in BWH047 (colistin resistant) and PA14 (colistin susceptible, MIC 0.5 μg/mL) grown in the presence and absence of ½ MIC colistin (640 μg/mL and 0.25 μg/mL, respectively; S8 Fig). PA14 served as a comparator strain. Transcript levels were normalized to both the reference gene recA (S8A Fig) and to the amount of input RNA (S8B Fig). In PA14, colistin exposure significantly increased pmrA and arnB expression, while dpcA expression exhibited a modest, though statistically significant, decrease (S8 Fig; S13–S15 Tables). By contrast, in BWH047, pmrA and dpcA expression were unchanged upon colistin exposure. Expression of arnB showed a small, though statistically significant, increase upon colistin exposure when normalized to recA, which was not observed when normalized to input RNA (S8 Fig, S13–S15 Tables). Discussion By combining TnSeq with in vitro phenotypic testing, we identified multiple intersecting pathways that are essential for extreme colistin resistance in the multidrug-resistant Pa clinical isolate BWH047 (Fig 4). We validated both previously known colistin resistance genes (e.g., arn operon, pmrAB) and identified several understudied genes that contribute to colistin resistance (e.g., dpcA, algU, wapH). We chose to investigate one of these targets, dpcA, by characterizing its impact on Pa LPS structure. The lipid A of ΔdpcA possessed less aminoarabinosylation and altered glycosyl composition and acylation status compared to BWH047 WT. Functional model of all Pa BWH047 genes identified as conditionally essential for colistin resistance. (A) Colistin causes membrane instability due to displacement of stabilizing divalent cations. This disruption is sensed by PmrB. PmrB phosphorylates PmrA, which triggers the transcription of downstream genes including those in the arn operon. Following transcription of the arn operon, aminoarabinose (L-Ara4N) is synthesized and exported from the cytoplasm on the carrier molecule undecaprenyl phosphate (UndP) with the assistance of arnEF so it can be added on to lipid A. UndP is then recycled back to the cytoplasm by DpcA so it can facilitate further rounds of aminoarabinose modification. Created in BioRender. Bachta, K. (2026) https://BioRender.com/8zg018k (B) Detailed diagram of Pa LPS with the following modifications: aminoarabinose addition catalyzed by the Arn system, carbamoylization catalyzed by WapO following synthesis by CarB, and the addition of glucose (Glc) by WapH. GlcN, glucosamine; Kdo, 3-deoxy-d-manno-oct-2-ulosonic acid; GalN, galactosamine; L-Ala, 2-l-alanylation; L-Rha, rhamnose. Created in BioRender. Bachta, K. (2026) https://BioRender.com/cmm24hs (C) MexXY-OprM is conditionally essential for colistin resistance in Pa; however, the mechanism by which is facilitates resistance is unknown. Created in BioRender. Bachta, K. (2026) https://BioRender.com/bcq2rm0 (D) MucA senses membrane disruption and is then degraded to release AlgU to facilitate the transcription of membrane stress response genes; however, the mechanism by which AlgU promotes colistin resistance remains unknown. Created in BioRender. Bachta, K. (2026) https://BioRender.com/yor3ctpctp. In Pa, colistin resistance is overwhelmingly attributed to modification of the negative phosphate head groups on lipid A of LPS with the glycosyl residue L-Ara4N [9,12]. The products of the arn operon (arnBCADTEF-ugd) synthesize, transport, and attach L-Ara4N to lipid A [33]. In BWH047, we identified three members of this operon as conditionally essential candidates which, when deleted, validated by BMD. This suggests that lipid A modification with L-Ara4N remains the mainstay of colistin resistance even in extremely resistant Pa isolates (Fig 4A). It has been shown that transcription of the arn operon can be activated by five separate TCS that sense and respond to colistin or colistin-mediated changes in cation concentration [13–20]. Despite these reports, we identified conditionally essential candidate genes in only one of these systems, pmrAB. Intriguingly, our transcriptional analysis indicated that in BWH047, expression of pmrAB and the arn operon was not induced by colistin (S8 Fig, S13–S15 Tables). This contrasted with PA14, a colistin sensitive strain, in which expression of both pmrA and arnB increased upon colistin challenge as anticipated. By comparing transcriptional levels of pmrA and arnB between PA14 and BWH047, we observed considerably higher levels of expression in BWH047 (S8 Fig; S13–S15 Tables). The mechanism responsible for the elevated expression of pmrA and the arn operon in BWH047 is an area of future exploration. Intriguingly, we also identified three conditionally essential gene candidates (speD2, speE2 and speH) upstream of pmrAB (S2C Fig); however, deletion of any of these genes had little impact on the colistin MIC. This is consistent with the literature which shows that deletion of speD2, speE2, or speH does not affect colistin MIC despite resulting in increased bacterial death at early time points post colistin exposure [55,56]. In PAO1, speD2 and speE2 are annotated as paralogs of known spermidine synthesis genes speD and speE, whereas speH has no paralog [56]. Bolard et al. identified the speD2-E2-H locus as being upregulated in the colistin resistant PAO1 derivative strain AB16.2; however, deletion of the speD2-E2-H locus did not decrease the colistin MIC [55]. In PAO1, speD2-E2-H contributed to polyamine production (primarily norspermidine, with a secondary role in spermidine synthesis) in PAO1, which increased resistance to aminoglycosides but did not increase resistance to colistin [55]. Whether speD2-E2-speH-pmrA-pmrB are co-transcribed or expressed as two or three separate operonic units is unclear [36,37,57]. In BWH047, operonic prediction tools FgenesB (Softberry) and Operon-Mapper [58] predicted that all genes are located within one transcriptional unit. Therefore, the location of the speD2-E2-H genes directly upstream of pmrAB, their identification as conditionally essential by TnSeq, and their lack of phenotype in our BMD assays suggests that polar effects of transposon insertion and not the direct protein products of speD2-E2-H are responsible for the identification of speD2-E2-H as gene candidates. We identified several candidates involved in LPS biosynthesis. This is unsurprising given that LPS, more specifically lipid A, is the target of colistin [59]. First, we identified dnpA, a hypothetical gene located at the end of a LPS biosynthesis operon (wapO-wapH-mig14-dnpA), as conditionally essential (S2C Fig). Surprisingly, ∆dnpA had a MIC unchanged from that of BWH047 (Table 1). Like the speD2-E2-H-pmrA-pmrB locus, we hypothesized that transposon insertion into dnpA might impact expression of other members of the operon more directly involved in colistin resistance. We were unable to delete mig14 but successfully generated clean deletions of wapH and wapO, resulting in MICs of 1 µg/mL and 160 µg/mL, respectively. WapH is a hypothetical protein annotated as a putative glycosyltransferase [57]. In a 2011 study, Jochumsen et al. identified wapH as a putative colistin response gene but concluded that transposon insertion in wapH likely affected expression of the downstream mig14 gene in a polar fashion [40]. In contrast, our results implicate wapH itself in colistin resistance, despite being unable to assess the role of mig14 in colistin resistance in BWH047 because we were unable to delete it. In 2004, Matewish demonstrated that deletion of wapH in PAO1 resulted in truncated LPS missing portions of the outer core-OS [60]. LPS isolates from ∆wapH demonstrated a similar phenomenon (S4B Fig), which may explain the increased colistin sensitivity of this strain (Fig 4). The final member of this operon (S2C Fig), wapO, produces a hypothetical protein annotated as a probable carbamoyl transferase. Unlike the arn operon, pmrAB and wapH, deletion of wapO in BWH047 resulted in an intermediate colistin MIC of 160 µg/mL. In PAO1, mutations in wapO were identified as having minor contributions to extreme colistin resistance, but only in the context of prior mutations that increased the colistin MIC significantly from baseline [61]. More recently, Erdmann et al. defined adaptive mutations in wapO that emerged while serially passaging Pa in increasing concentrations of colistin [38,39]. In PAO1, transposon insertions in wapO resulted in a one-fold decrease in colistin MIC (0.5 µg/mL to 0.25 μg/mL) [38,39]. We hypothesize that the function of wapO may become increasingly important as colistin MIC increases, and that the low colistin MIC of PAO1 masks the ability to detect these more subtle phenotypes. Erdmann et al. also identified adaptive mutations in carB, a putative carbamoyl phosphate synthase and partner of wapO, using experimental evolution to increase colistin resistance [38,39]. We, too, identified carB as conditionally essential for colistin resistance. Whereas Erdmann et al. found that deletion of carB did not impact PAO1’s colistin MIC, we determined that, like ∆wapO, the colistin MIC of ΔcarB was 160 μg/mL. Erdmann et al. postulated that WapO and CarB were part of a pathway responsible for carbamoylation of sugars in the LPS core-OS (Fig 4) [38]. Because mutations in both genes emerged during the development of colistin resistance, but complete gene deletion did not result in a change in MIC, they postulated that a decrease in carbamoylation correlated with increased colistin resistance [38,39]. However, by utilizing BWH047, we observed the opposite—the activity of WapO and CarB, and thus the carbamoylation pathway, were required for full colistin resistance. TnSeq also identified candidates in two resistance-nodulation division (RND) efflux systems: mexY, mexB, and oprM. These genes code for members of the MexAB-OprM and MexXY-OprM complexes, which are well-characterized Pa tripartite efflux systems that extrude a wide variety of antimicrobials and contribute to antimicrobial resistance [41]. MexB and MexY span the inner membrane, MexA and MexX are periplasmic membrane fusion components, and OprM is the outer membrane channel protein that is shared by both efflux systems [42]. Interestingly, we found that deletion of mexY or oprM, but not mexB, significantly reduced the colistin MIC of BWH047 (Table 1). Further investigation into this phenomenon showed that mexA, mexB, and oprM are co-transcribed as a single transcriptional unit [62]. We hypothesize that like the speD2-E2-H system, mexB was identified as a candidate essential gene because of polar effects on expression of oprM. Thus, MexXY-OprM, but not MexAB-OprM, contributes to colistin resistance in BWH047. The literature is divided on the importance of MexAB-OprM and MexXY-OprM in Pa colistin resistance. Initial work to determine the substrates of these efflux pumps showed that polymyxin B, a cationic antimicrobial peptide closely related to colistin, was not a substrate for either efflux system [41]. Puja et al. showed that the MexXY-OprM system is intricately connected with arn- and pmrAB-mediated colistin resistance in Pa, but that this phenomenon had no impact on L-Ara4N modification of lipid A [55,56]. It is possible that MexXY-OprM is effluxing some as-yet unidentified compound that modifies lipid A or LPS and prevents colistin’s attraction to the bacterial cell. Ultimately, how this particular efflux pump regulates colistin resistance or cross talks with the more dominant arn- and pmrAB-mediated mechanisms of colistin resistance remains unclear. Additionally, we identified AlgU as important for colistin resistance in BWH047. Otherwise known as RpoE, AlgU is an extra-cytoplasmic sigma factor that regulates alginate production and mucoidy in Pseudomonas [43,44,63–65]. AlgU is sequestered at the inner membrane by transmembrane anti-sigma factor MucA, which is degraded through a proteolytic cascade upon sensing membrane stress [66]. AlgU also downregulates motility and virulence gene expression and upregulates stress response pathways [43,44,63–65]. In Salmonella enterica and Escherichia coli, the algU ortholog rpoE was recently demonstrated to be necessary for colistin resistance [67]. Interestingly, Zeng et al. found that in E. coli and S. enterica, deletion of rpoE did not impact lipid A modification with L-Ara4N [67]. Instead, they attributed the colistin sensitivity of the ΔrpoE strain to an alteration in phospholipid composition, which resulted in decreased membrane asymmetry and increased cell permeability [67]. We have yet to determine the mechanism by which algU is essential for colistin resistance in Pa, but this remains an active area of inquiry. Our most interesting candidate gene from TnSeq, now named dpcA, is orthologous to PAO1 PA4029. Deletion of dpcA led to a complete loss of colistin resistance. Until recently, the function of DpcA (DedA) orthologs remained unknown. Within the past few years, research in Klebsiella pneumoniae, Burkholderia thailandensis, Vibrio cholerae, and Staphylococcus aureus revealed that DedA is a flippase that recycles undecaprenyl phosphate (UndP) transporting it from the outer surface of the inner bacterial membrane and back into the cytoplasm [21,22,68–70]. In addition to transporting LPS O-antigen subunits, peptidoglycan precursors, and glycosyl residues, UndP also ferries the lipid A-modifying glycosyl residue L-Ara4N across the cytoplasmic membrane in GN bacteria [33]. Based on these studies, a model emerged wherein DedA flippases play an important role in colistin resistance by facilitating the recycling of UndP carrier molecules back into the cytoplasm once they deliver their L-Ara4N cargo to the periplasmic space. Uncharged UndP is then reloaded with L-Ara4N, facilitating further rounds of lipid A modification that led to increasing colistin resistance (Fig 4) [69,71]. In support of this model, deletion of dpcA orthologs in either K. pneumoniae or B. thailandensis significantly decreased the amount of L-Ara4N modification of lipid A [21,22]. Until very recently, DpcA was unstudied in Pa. Two papers revealed its importance for UndP recycling and implicated it as having a role in colistin resistance [45,54]. A lack of DpcA led to decreased UndP recycling and diminished UndP pools on the cytoplasmic face of the bacterial inner membrane. Thus, fewer UndP molecules are available to transport the critical building blocks of LPS [45,54]. However, this work was completed in the colistin-sensitive strain PAO1. We further characterized dpcA’s function in BWH047 by investigating LPS structure. Overall, deletion of dpcA had pleotropic effects on LPS including a higher proportion of core-OS sugars (Rha, Glc, Kdo, and GlcNAc) than BWH047 (Table 2), an increase in the content of palmitic fatty acid (16:0) and loss of the unusual cis-8-octadecanoic fatty acid (18:1∆8) (Table 3), an increase in tetra- and hexa-acylated lipid A (Table 6), and a decrease in the number of L-Ara4N moieties on lipid A (Table 5). With respect to the LPS core-OS, UndP ferries O-antigen subunits from the cytoplasm to the periplasm [51,52,72]. Deletion of DpcA likely results in decreased transport of O-antigen glycosyl residues to the outer membrane leading to shorter LPS O-antigen, as observed in S4A Fig. GC-MS and MALDI-TOF MS/MS data revealed that the lipid A of BWH047, unlike ∆dpcA, contained higher levels of palmitic acid (16:0) and an unusual cis-8-octadecanoic (18:1∆8) fatty acid which was totally absent in ∆dpcA (Table 3, Figs 3, S4–S6). While 18:1 fatty acids have been reported as a component of lipid A in Pseudomonas syringae and in Francisella tularensis [73,74], 18:1∆8 has not previously been reported as component of lipid A in Pa. The functional significance of the 18:1∆8 acyl chain in lipid A has not been investigated. Future investigation will be necessary to determine biosynthetic pathways responsible for its inclusion in BWH047 lipid A, if the presence of 18:1∆8 leads to increased cationic antimicrobial peptide resistance and membrane plasticity, and if and how this modification impacts lipid A recognition by the host immune system. In accordance with our hypothesis, the lipid A of ∆dpcA contained a lower proportion of structures that were modified by either one or two L-Ara4N moieties than in BWH047 (Tables 2, 4, 5 and Fig S6). In summary, deletion of dpcA in BWH047 had pleiotropic and wide-ranging effects on the structure of LPS and lipid A which extend beyond its impact on L-Ara4N modification. Notably, unlike pmrA and arnB, dpcA expression was not induced by colistin exposure in either BWH047 or PA14 (S8 Fig, S13–S16). DpcA expression remained consistent, regardless of the expression of pmrA and arnB. We believe the DpcA-mediated UndP recycling pathway is essential for colistin resistance and acts downstream of the PmrAB and Arn systems (Fig 4A). Therefore, extreme colistin resistance in BWH047 likely depends on both elevated pmrA/arnB expression and on DpcA activity, which facilitate continuous UndP recycling and ongoing aminoarabinosylation of lipid A. Overall, TnSeq identified multiple pathways conditionally essential for colistin resistance in the extremely colistin resistant clinical Pa isolate BWH047 and uncovered a complex web of interconnected networks that were necessary for this extreme phenotype (Fig 4). This work represents one of the most thoroughly validated TnSeq experiments to date. We identified 20 conditionally essential gene candidates and successfully deleted 17 target genes and 6 ineighboring genes of interest. This resulted in a validation rate of 75%, wherein 15 of our 20 target genes were essential for or located in operons with genes essential for extreme colistin resistance. Taken together, TnSeq proved highly successful in identifying multiple intersecting pathways that lead to extreme colistin resistance in Pa and highlighted additional mechanisms that deepen our understanding of colistin resistance. Materials and methods Ethics statement The collection and use of Pa clinical isolates and associated data represented in this study was reviewed and approved by the Partners Human Research Committee under Protocol #2015P001829/BWH on September 15, 2015. The IRB waived the requirement for informed consent, as the research involved only existing bacterial isolates and retrospective clinical data. Bacterial strains and growth conditions Pa clinical isolates were collected from Brigham and Women’s Hospital (BWH) clinical microbiology laboratory in Boston, MA between 2015 and 2016. Isolates were collected from a variety of patient sites (Fig 1, S4) and were selected and screened for antibiotic resistance to commonly used antibiotics including aztreonam, anti-pseudomonal cephalosporins (cefepime and ceftazidime), piperacillin-tazobactam, ciprofloxacin, colistin, and meropenem. BWH047 was originally described in 2019 [75]. The remainder of the PABWH isolates are described in the present study (Fig 1, S4 Table). Escherichia coli strain TOP-10 (Invitrogen) was used for cloning, and E. coli strains S17.1 λpir and SM10 λpir were used to introduce plasmids into Pa. Unless otherwise specified, all bacteria were grown at 37°C in LB or cation-adjusted Mueller Hinton Broth (MHBII). When necessary, antibiotics were used at the following concentrations: for Pa, irgasan 5 μg/mL, hygromycin 500 μg/mL and for E. coli, hygromycin 100 μg/mL. Additional details on strains and plasmids used in this study can be found in S2–S4 Tables. Whole genome sequencing of PABWH isolates and isogenic mutants Briefly, genomic DNA from each PABWH isolate was harvested from a single colony and sequenced on the HiSeq 4000 Platform (Illumina, Inc., San Diego, CA) generating 2 x 150 bp paired end reads. Adapter sequences were removed, and reads were quality trimmed using Trimmomatic v.0.36 [76]. De novo assembly was performed using SPAdes v3.9.1 [77]. Quality control was performed by aligning trimmed reads to assembly contigs using bwa v0.7.15 [78]. All contigs shorter than 200 bp or with an average fold coverage of <5x per base were removed. To complete the draft genome sequence of BWH047 [75], Oxford Nanopore MinION sequencing was performed on a R9.4.1 flow cell after genomic DNA library preparation using the Oxford Nanopore SQK-LSK109 kit. A consensus de novo assembly was generated using Trycycler v.0.5.0 [79] from long read assemblies generated using Raven v.1.5.3 [80], Flye v.2.9.5 [81], and Miniasm v.0.3 [82]. The assembly was polished first with the Oxford Nanopore long reads using medaka v1.11.3 and subsequently with the Illumina short reads using polypolish v0.5.0 [83] and the polca module of MaSuRCA v4.0.9 [84]. Annotation was performed using the NCBI Prokaryotic Genome Annotation Pipeline 6.10 [85]. Whole genome phylogenetic analysis Illumina reads from each isolate were aligned to the complete genome sequence of strain BWH047 using bwa v0.7.15 [78]. Single nucleotide variants (SNVs) were identified using bcftools v1.9 [86] using the bcftools_filter software (https://github.com/egonozer/bcftools_filter) to remove variants with SNV quality scores less than 200, read consensus less than 75%, read depths less than 5, read numbers in each direction less than 1, or locations within repetitive regions as defined by blast alignment of the reference genome sequence against itself. A maximum likelihood phylogenetic tree was generated from the core genome alignment with IQ-TREE v2.2.0 using the ModelFinder function to estimate the best-fit nucleotide substitution model by means of Bayesian information criterion (BIC) [87,88]. Tree visualization and annotation was performed in R v4.4.2 using ggtree v3.14.0 [89]. Generation of gene deletions and complemented strains in Pa BWH047 Upstream and downstream fragments surrounding the gene of interest (GOI) were amplified from BWH047 genomic DNA using primers GOI-5–1-HindIII with GOI-5–2, and GOI-3–1 with GOI-3–2-HindIII (S1 Table). The resulting upstream and downstream fragments, which each contained a 24-bp overlapping linker sequence (TTCAGCATGCTTGCGGCTCGAGTT in primers GOI-5–2 and GOI-3–1), were used as the templates for overlap extension PCR and were subsequently PCR amplified with primers GOI-5–1-HindIII and GOI-3–2-HindIII to create a single linear fragment. This was combined with the integration proficient vector pEX18Hyg [90] previously digested with HindIII (New England Biolabs) via Gibson assembly (New England Biolabs Gibson Assembly Cloning Kit) to generate allelic exchange vectors for all genes of interest (pEX18HygΔGOI, S2 Table). These vectors were transformed into E. coli TOP-10 (Invitrogen) and plated on LB agar plates supplemented with 100 µg/mL hygromycin. Plasmids were miniprepped (GeneJET Plasmid Miniprep Kit, Thermo Scientific) and verified by Sanger Sequencing (ACGT, Inc.). Vectors were transformed into E. coli SM10 λpir and E. coli S17.1 λpir. Following conjugation and allelic exchange with BWH047, gene deletions were verified using Sanger sequencing (ACGT, Inc.) Gene deletions were complemented in one of two ways. First, arnC and ugd were complemented by re-introducing the native allele into the deletion strain at its original position in the genome. Briefly, GOI-5–1-H and GOI-2-H were used to amplify the entire native allele. As above, Gibson assembly (New England Biolabs Gibson Assembly Cloning Kit) was used to ligate this complementation fragment into HindIII-digested pEX18Hyg generating pEX18Hyg-comp-GOI. Then, these two vectors were transformed into E. coli TOP-10 (Invitrogen), miniprepped (GeneJET Plasmid Miniprep Kit, Thermo Scientific), verified by Sanger Sequencing (ACGT, Inc.) and transformed into E. coli SM10 λpir and S17.1 λpir. Following conjugation and allelic exchange with BWH047ΔGOI, complementation was confirmed both by sanger sequencing (ACGT, Inc) and phenotypically through determination of colistin MIC (Table 1). Alternatively, dpcA was complemented by introducing the dpcA sequence at the Tn7 site of BWH047∆dpcA using pCOMP-hyg-npt2, a version of the pUC18T mini-Tn7T-Gm vector [91] modified with a constitutive npt2 [92] promoter and hygromycin resistance. To construct pCOMP-hyg-npt2, the vector pUC18T-mini-Tn7T-Gm-eYFP [91] was first digested with EcoRI and NsiI (NEB) and Gibson assembly was used to ligate this vector with a gene block containing a FRT-flanked hygromycin resistance cassette to generate pUC18T-mini-Tn7T_hygR_eYFP. Subsequently, pUC18TminiTN7T_hygR_eYFP was digested with BsrG1 and Kpn1 to remove the eYFP cassette. Then, Gibson assembly was used to ligate a gene block (IDT) containing both a constitutive synthetic lac promoter (PA1/04/03) [93] and a multiple cloning site was into the cut pUC18TminiTN7T_hygR to generate pCOMP-hyg. To provide alternative promoter expression, pCOMP-hyg was again digested with BstBI and KpnI and Gibson assembled with a gene block from IDT containing the npt2 promoter to generate pCOMP-hyg-npt2 (gift of J. Allen). This final plasmid was verified using whole plasmid sequencing through Plasmidsaurus (San Francisco, CA). Finally, to generate pCOMP-hyg-npt2-dpcA, the sequence of the dpcA was amplified from BWH047 using primers pCOMP-hyg-npt2-dpcA_5 and pCOMP-hyg-npt2-dpcA_3 (S1 Table) and Gibson assembly was used to ligate dpcA with BstBI-digested pCOMP-hyg-npt2. This final dpcA complementation vector was transformed into E. coli TOP-10 cells (Invitrogen), sequence verified via Sanger sequencing (ACGT, Inc.) and transformed into E. coli SM10 λpir. E. coli SM10 λpir containing pCOMP-hyg-npt2-dpcA, E. coli SM10 λpir containing pTNS2, E. coli HB101 containing pRK2013, and ΔdpcA were mated for 2 hours at 37˚C [91]. Resulting colonies resistant to both hygromycin and irgasan were screened using PCR for insertion of the dpcA allele at the Tn7 site. Complementation was confirmed by both Sanger sequencing (ACGT, Inc.) and evaluation of colistin MIC (Table 1). Determination of minimum inhibitory concentrations (MIC) The MICs for the PABWH collection and all BWH047 isolate derivatives were determined in triplicate in compliance with the broth microdilution (BMD) protocol as outlined by Clinical and Laboratory Standards Institute (CLSI) M100-Ed35 [23]. The following antibiotics were prepared from commercially available sources and were used to assess MICs: aztreonam, cefepime, ceftazidime, ciprofloxacin, meropenem, piperacillin-tazobactam. For most antibiotics, MICs were performed in Mueller Hinton Broth (MHB) except for colistin MICs, which were performed in cation-adjusted MHB (MHBII). Pa ATCC control strain 27853 was used for quality control for broth microdilution experiments according to CLSI guidelines. Colistin MICs for all PABWH isolates were initially determined using the standard colistin concentrations (0.25-128 μg/mL). Once it was established that BWH047 had a colistin MIC ≥ 128 µg/mL, colistin BMDs were performed using 10x the traditional concentrations (2.5-1280 μg/mL) to determine an exact MIC value. Construction of transposon mutant library in P. aeruginosa isolate BWH047 The pSAMHyg suicide plasmid for Himar1 mariner transposon (tn) mutagenesis [94] was transformed into E. coli SM10 λpir. Equal volumes of E. coli SM10 λpir containing pSAMHyg were mixed 1:1 with BWH047 and 125 x 10µL mate spots were spotted onto LB for 1 hour and 15 minutes. Mating conditions were optimized to ensure a minimum of 400,000 colonies of BWH047:tn were obtained. All mate spots were scraped and resuspended into a total volume of 8 mL of LB. 50 μL of this suspension was then plated onto large agar plates (n ~ 150) supplemented with irgasan (5 μg/mL) and hygromycin (500 μg/mL) and incubated at 37°C overnight to select for colonies containing a transposon insertion event. The following day, all irgasan (5 μg/mL)/hygromycin (500 μg/mL) resistant colonies were scraped and resuspended in a total volume of 15mL of LB supplemented with 25% glycerol. The final BWH047:tn library was aliquoted and stored at −80°C for further analysis. Transposon Insertion (TnSeq) experimental design An aliquot of the BWH047 transposon library was thawed and revived in MHBII at 37°C for approximately 2 hours. Following this incubation, the revived library was added to MHBII containing colistin at these final concentrations: 640 μg/mL, 320 μg/mL, and 0 μg/mL. Five experimental replicates were used per antibiotic concentration (S2A Fig). Cultures were incubated at 37°C with shaking for 18 hours. The colistin concentrations and library outgrowth time were optimized (1/4 MIC, 320 µg/mL and 1/2 MIC, 640 µg/mL) to provide strong selective pressure to facilitate drop out of strains with disrupted conditionally essential genes and to mirror the time point at which MICs are typically assessed (16–20 hours post inoculation) [23]. Following incubation, bacteria were harvested via centrifugation at 13,200 x g for 2 minutes and genomic DNA was extracted DNA was prepared for sequencing according to the methods outlined in Kazi et al 2020 [95]. Briefly, genomic DNA was sheared to approximately 250 bp (Covaris Focused Ultrasonicator). Transposon-DNA junctions were amplified, and barcodes were added using previously designed primers [95] (S1 Table). Library pools were quantified using Kapa library quantification kit (Roche) and Quant-iT dsDNA Assay Kit, high sensitivity (HS) (Invitrogen). Pools were sequenced on an Illumina MiSeq using the Miseq v2 reagent kit (50-cycle) generating 50 bp single-end reads containing the 6 bp barcode sequence, 22 bp Himar transposon sequence, the TA dinucleotide insertion site, and the 22 bp genome sequence flanking the TA site. Illumina reads were processed with TIS_Tools (github: https://github.com/egonozer/TIS_tools). Briefly, the transposon sequence was identified in each read, allowing for up to a 1 base mismatch. Next, reads were trimmed to a final length of 24 bp containing only the TA dinucleotide insertion site and the 22 bp of BWH047 genome sequence that flanked the TA dinucleotide. Then, the number of sequences for each TA insertion site was counted and grouped within each library and reads were mapped to the BWH047 complete genome sequence using Bowtie v1.3.1 [96]. Reads were excluded from analysis if they aligned to ≥ 1 genomic site or contained > 1 base mismatch. Finally, the alignment position and strand (+ or -) were assigned before outputting a file associated with each experimental replicate. The filtering done to determine high confidence TA insertion sites, the percentage of reads retained for analysis after each filter was applied (separated by exposure condition—320 μg/mL, 640 μg/mL, and MHB II only), and the percentage of TA sites included in the final analysis are depicted in S3 Fig, and the underlying data is in S9 and S10 Tables. TnSeq data analysis and identification of conditionally essential genes To identify genes conditionally essential for colistin resistance in BWH047, a modified version of the ESSENTIALS pipeline [97] was implemented (github: https://github.com/egonozer/essentials_local). Briefly, reads aligning to flanking sequences in the reference BWH047 sequence were identified, and sites which had a combined total of < 3 reads in any of the control libraries (MHBII only, 0 µg/mL) were filtered out. Additionally, any TA sites located in the last ~ 5% of a protein-coding sequence were excluded from analysis. Next, a Loess normalization was executed to reduce the bias of higher read counts near the chromosomal origin of replication given its enhanced TA site accessibility during chromosomal replication. Finally, a Fisher’s exact test followed by Benjamini-Hochberg correction was employed to determine which genes were conditionally essential for survival in the presence of colistin. We defined a gene as conditionally essential if it met the following criteria: log2 fold change <−2.0 and false discovery rate (FDR) < 0.05 when comparing transposon insertion count per gene when grown in the presence of either 640 μg/mL or 320 μg/mL colistin treatment (treatment) compared to transposon insertion count when grown in MHBII media alone (control, 0 µg/mL). To analyze the validity of our pipeline, we also utilized the modified ESSENTIALS pipeline to identify conditionally essential genes in the MHBII-only media control conditions as previously described [29]. We then compared the list of identified essential genes in MHBII to the two gold-standard sets of essential genes identified by Van Maele et al. as essential for growth in all Pa TnSeq experiments (GOLD_84) and essential for growth of PA14 in LB (GOLD_115) [98]. BWH047 homologs of the PA14 loci from GOLD_84 and GOLD_115 were determined using BLASTp (BLAST+ v2.16.0), and all met the following identity criteria: percent identity >98%, query coverage ≥ 99% [99]. In our BWH047 transposon library grown in MHBII, 84/84 of the GOLD_84 genes and 113/115 of the GOLD_115 genes were identified as essential, lending credence to our analysis pipeline. The essential genes for growth in MHBII are listed in S11 Table along with the corresponding PA14 loci of the GOLD_84 and GOLD_115 genes. Extraction and purification of Pa LPS Twenty liters of BWH047 and ΔdpcA were grown in LB to exponential phase (OD ~0.8), at which point cultures were fixed in 1% phenol for 1 hour and then pelleted at 8,000 x g for 20 minutes at 4°C. LPS was extracted from cell pellets by adding an equal volume of preheated 90% liquefied phenol and gently stirring for 20 minutes at 68°C, according to the Westphal extraction procedure [50]. After LPS extraction, the mixture was cooled in an ice bath and centrifuged at 10,000 x g for 20 minutes at 4°C. The water layer was collected, and the remaining phenol phase was extracted an additional two times with water, following the same protocol as above. The combined extracts of all three water layers and the residual phenol layer were extensively dialyzed against deionized water (12–14,000 MWCO), freeze-dried, and washed with 9:1 (v/v) ethanol in water at 4°C. Nucleic acids and proteins were removed by treating overnight with Benzonase (37°C), followed by proteinase K (37 °C) digestion (50 mM MgCl2•6H2O and 20 mM NaOAc•3H2O) and dialysis (12–14,000 MWCO) at 4°C against several exchanges of dH2O. Dialyzed phenol and water phases were then freeze-dried, and the LPS preparations were suspended in deionized water. These LPS fractions were further purified by ultracentrifugation at 100,000 × g at 4°C for 16 h. Enzymatic purification steps and ultracentrifugation were repeated twice to prepare ultrapure LPS. In addition, purified LPS was deprived of phospholipids and free fatty acids using 90% EtOH washes (completed three times) followed by short sonication at 0˚C and collection of LPS by centrifugation at 3,500 x g. A trace of coextracted phospholipids was removed from the final LPS preparation using three 90% EtOH washes, facilitated by short sonication at 0˚C and centrifugation at 3,500 x g. The ultrapure LPS was used for all chemical studies. DOC-PAGE analysis of LPS Purified Salmonella typhimurium, P. aeruginosa BWH047, ∆dpcA, and ∆wapH LPS were resolved on a 18% polyacrylamide gel (PAGE) containing deoxycholic acid (DOC) detergent [100]. The gel was developed using a silver stain reagent kit (Bio-Rad) after oxidation with sodium periodate. Chemical composition analysis of LPS A simultaneous analysis of glycosyl residues and fatty acids constituting BWH047 and ∆dpcA LPS was achieved through sample derivatization to O-trimethylsilyl (TMS) methyl ethers for sugar residues and fatty acid methyl esters (FAME) and TMS-FAME derivatives for hydroxylated fatty acids. Briefly, samples were methanolyzed with 1M HCl-methanol at 80°C for 18 h, re-N-acetylated at 100°C for 1 h, and O-trimethylsilylated with Tri-Sil reagent at 80°C for 30 min. Each sample was supplemented with myoinositol, which was used as an internal standard [101,102]. The analysis of TMS and FAME derivatives was performed by gas chromatography-mass spectrometry (GC-MS) on an Equity-1 (Supelco) fused silica capillary column (30 m length × 0.25 mm ID × 0.25 μm film thickness) interfaced with a Hewlett-Packard HP5890 gas chromatograph equipped with a mass selective detector 5970 MSD. The oven temperature was initially set at 80°C for 2 min, then increased to 140°C at a rate of 20°C/min with a 2 min hold and then raised to 200°C at a rate of 2°C/min, followed by an increase to 250°C at a rate of 30°C/min with a 5 min hold. The data were processed using Agilent ChemStation. Isolation of lipid A from intact LPS To prevent the loss of acid-labile L-Ara4N and phosphate residues, ultrapure LPS from Pa BWH047 and ∆dpcA was dissolved in 10 mM sodium acetate pH 4.5 for 1h at 100˚C with gentle agitation. Lipid A was recovered from the sodium acetate solution by threefold extraction with chloroform and, finally, the extraction of the combined chloroform phases with water. The remaining chloroform phase was dried, and purified lipid A was used for MALDI-TOF MS analysis. Colistin exposure for gene expression analysis Four colonies each of BWH047 and PA14 were inoculated into MHBII and grown overnight at 37°C. From each overnight, subcultures were grown to mid-log phase and cultures were normalized to OD600 of 0.4. OD-standardized cultures were then combined in equivolume amounts with MHBII or with MHBII containing colistin such that the final concentration was ½ MIC (640 μg/mL for BWH047 and 0.25 μg/mL for PA14). Cultures were grown for 1 hour before RNA was stabilized using RNAProtect Bacterial Reagent (QIAGEN). RNA was purified and extracted utilizing the RNEasy Mini Kit (QIAGEN) according to the manufacturer’s instructions with the inclusion of the on-column DNAse treatment (QIAGEN). RNA was aliquoted to prevent freeze-thaw and stored immediately at -80°C. RNA purity was assessed using the NanoDrop One Microvolume UV-Vis Spectrophotometer. RNA concentration was evaluated utilizing the Quant-iT RNA Reagent and Assay Kit (ThermoFisher). cDNA conversion was performed with the QuantiTect Reverse Transcription Kit (QIAGEN) according to the manufacturer’s instructions, including the gDNA Wipeout step, with 1920 ng input RNA (normalized across samples); standard cycling conditions were used. cDNA was stored at -20°C and aliquoted to reduce freeze-thaw cycles. RNA extraction, cDNA conversion, and digital PCR Digital PCR (dPCR) was performed on the QIAcuity One Digital PCR System using 8.5k 96 well nanoplates (QIAGEN) according to the manufacturer’s instructions. Each reaction contained 1.2 μL cDNA, forward and reverse primers (listed in S1 Table) at a final concentration of 0.4 μM, 4 μL 3X EvaGreen PCR Master Mix (QIAGEN), and nuclease-free water in a total reaction volume of 12 μL. Primers were designed using PrimerQuest (IDT), and specificity was verified by in silico BLAST analysis and PCR using genomic DNA of BWH047 and PA14 prior to dPCR. dPCR cycling conditions consisted of an initial denaturation at 95°C for 2 min followed by 40 cycles of: 95°C for 15 sec, 58.2°C for 15 sec, 72°C for 15 sec; all followed by a final cooling step at 40°C for 5 min. The annealing temperature of 58.2°C was selected as the optimal annealing temperature based on gradient PCR using dPCR primers (S1 Table). Fluorescence was imaged using the QIAcuity Software Suite (version 3.2.0.0) with exposure of 300ms and gain of 6. Positive and negative partition thresholds were set automatically by the software followed by manual adjustment to maximize cluster separation. A no-template control (NTC) was included for each primer set on every plate, and a no reverse transcriptase control was included for pooled samples from each condition to confirm the absence of genomic DNA contamination. All wells were included in the analysis regardless of partition count (S13 Table). Absolute copies/µL of target genes were normalized to either absolute copies/µL of reference gene recA (S14 Table) or to amount of input RNA (S15 Table). Gene expression analysis recA was selected as the reference gene for dPCR normalization based on its use as a housekeeping gene in prior P. aeruginosa studies [103]. Statistical analysis was performed in GraphPad Prism (version 11.0.2). For each gene, colistin-exposed and untreated paired biological replicates (n = 4) were compared using a ratio paired t-test. This approach was applied independently to recA, arnB, pmrA, and dpcA expression, normalized either to recA (target/recA concentration ratio) or to input RNA mass (copies per ng of input RNA). To account for multiple comparisons across the four genes tested within each normalization scheme, p-values were adjusted using the Benjamini-Hochberg method to control the false discovery rate (FDR), and adjusted q-values below 0.05 were considered statistically significant. This study adheres to the dMIQE guidelines for performing and reporting digital PCR experiments [104]; the completed dMIQE checklist is provided in S16 Table. MALDI-TOF MS and MS/MS analysis of lipid A Approximately 2 µL of purified Pa BWH047 and ∆dpcA lipid A dissolved in chloroform:methanol (3:1, v/v) was mixed with 0.5 M 2,4,6-trihydroxyacetophenone matrix (Sigma) in methanol at a 1:1 ratio (v/v), and 0.5-1 µL was immediately applied to a Matrix-Assisted Laser Desorption/Ionization (MALDI) plate using a microcapillary. Spectra were acquired on a MALDI-TOF (time of flight) mass spectrometer RapifleX TissueTyper system (Bruker Daltonics, Billerica, MA, USA) in negative reflector ionization mode [M-H]-. From select observed precursor ions, MS/MS tandem mass spectrometry was performed on the same instrument in TOF/TOF geometry using a single-beam laser and nitrogen collision-induced dissociation (CID) in the negative ionization mode, with an isolation window of ±10MhZ and 4000 fragment shots. All data were processed using flexAnalysis (v. 4.2) MS software (Bruker Daltonik, GmbH). Comparative analysis of LPS and lipid A composition Monosaccharide composition of LPS isolated from each strain (Table 2) was determined by calculating the relative molar percentage of each monosaccharide from response factor–corrected values based on the response factor of original monosaccharide standards GC–MS (S4 Fig). Due to the unavailability of authentic standards for FucNAc, L-Ara4N, and QuiNAc, we used the relative response factor of GlcNAc. Fatty acid composition of LPS isolated from each strain (Table 3) was calculated using the relative response from the GC-MS EI detector, where the sum of all integrated fatty acid peak areas equals 100%. The peak intensities (S8 Table) for all peaks with assigned proposed structures of interest (Table 4, S6 Fig) were divided by the sum of all peak intensities with proposed structures to determine a relative percentage of various glycosyl residues, acylation patterns, and L-Ara4N in each sample with the indicated characteristic. Peptidoglycan purification and analysis Peptidoglycan (PG) was isolated and purified from 350 mL of mid-log phase cultures using previously published procedures [105–107]. BWH047, ΔdpcA, and ΔdpcA-comp were grown overnight, diluted 1:100 into 350 mL of fresh LB, and cultured to mid-log exponential growth phase. Cells were then pelleted, washed twice with PBS, and frozen at -80°C until PG purification. To isolate PG sacculi, the washed cell pellets were thawed on ice, resuspended in cold PBS, and added dropwise to boiling 10% SDS (5% SDS final concentration). After 1 hour, the insoluble material was collected by ultracentrifugation (242,500 x g, 60 min, 30ºC) and the resulting pellets were washed five times with warm (40ºC) autoclaved dH2O. Intact peptidoglycan sacculi were treated with chymotrypsin (0.3 mg/mL) overnight with shaking (540 rpm) at 37ºC to remove bound proteins. Excess chymotrypsin was denatured, and cleaved material solubilized, by adding 0.5% SDS (final, v/v) and incubating each preparation for 30 minutes at 80ºC. Samples were washed three times using warm (40ºC) autoclaved dH2O using ultracentrifugation, as described above. Soluble muropeptides were produced by incubating intact chymotrypsin-treated sacculi with mutanolysin (62 U/mL final) and 5 μM NaHPO4/NaH2PO4, pH 5.5 overnight with shaking (540 rpm) at 37ºC. Afterwards, undigested material was re-treated with mutanolysin, as described above, and incubated for an additional 4–6 hours. Mutanolysin was heat-inactivated (10 min, 100ºC) and soluble muropeptides were harvested via centrifugation at 22,000 x g for 30 minutes. Isolated supernatant containing soluble muropeptide fragments was frozen (-80ºC) and dried (FreeZone Benchtop Freeze Drier, LABCONCO). The purified muropeptides were resuspended in 150 μL of saturated 0.5 M sodium borate buffer, pH 9.25. A 50 μL aliquot of 100 mg/mL NaBH4 was added dropwise to reduce each sample and, after 1 hr incubation, the reactions were quenched with approximately 10 μL of formic acid (final pH 3). The samples were then frozen and lyophilized as described above. Reduced muropeptides were resuspended in H2O:MeOH (200 μL, 9:1, v:v) containing 0.1% formic acid and sonicated in a water bath (Fisher Scientific) for 10 minutes. Soluble reduced muropeptides were collected by transferring 180 μL of supernatant into an LC-MS vial after centrifugation at 13,000 x g at 4°C for 15 min. Samples were then stored at -20°C until LC-MS analysis. LC-MS analysis of muropeptides Analyses were performed as published previously on a Shimadzu LCMS9030 QToF system coupled to a LC-40B X3 UPLC, a SIL-40C X3 autosampler (10°C), and a CTO-40C column oven (40°C) [106,107]. Borohydride-reduced mutanolysin-digested muropeptides were separated on a Waters BEH C18 column (2.1 x 50 mm, 1.7μm particle size), at a flow rate of 0.3 mL/min. Solvent A was water and Solvent B was MeOH, with both solvents containing 0.1% formic acid (v/v). The initial solvent condition was 99:1 (A:B) for 3 min. This was followed by stepwise linear gradients first to 8% Solvent B (12 min), then to 20% Solvent B (24 min), and finally 95% Solvent B (25 min), followed by a hold at 95% Solvent B until 29 min. The column was then re-equilibrated to 1% Solvent B (1 min of 1% Solvent B, followed by a 5 min re-equilibration time). The reduced muropeptides were subjected to electrospray ionization in positive ion mode using data-dependent acquisition. Interface voltage was 4.50 kV at 300°C with a desolvation temperature of 526°C and a DL temperature of 250°C. Gas flows for nebulizing, heating, and drying gases were 2, 10, and 10 L/min, respectively. The m/z range for the precursor MS scan was 100–2000. Muropeptide analysis peak areas and percentages are listed in S12 Table. Supporting information List of primers used in this study. (XLSX) List of plasmids used in this study. (XLSX) List of strains in this study. (XLSX) List of PABWH Collection Isolates and associated metadata. Contains source, sequence type, T3SS genotype, MIC, and SIR data. (XLSX) ESSENTIALS bioinformatic pipeline output from transposon insertion sequencing experiment exposing BWH047 to 320 µg/mL of colistin. (XLSX) ESSENTIALS bioinformatic pipeline output from transposon insertion sequencing experiment exposing BWH047 to 640 µg/mL of colistin. (XLSX) Gene targets from the transposon insertion experiment that passed the filtering conditions for both 320 µg/mL and 640 µg/mL colistin. (XLSX) MALDI-TOF MS raw signal integration values for BWH047 and ΔdpcA lipid A released using a mild procedure (pH 4.5). (XLSX) TIS_tools bioinformatic output for TnSeq read processing. (XLSX) TIS_tools bioinformatic output for TnSeq TA site analysis. (XLSX) ESSENTIALS bioinformatic pipeline output from transposon insertion sequencing experiment exposing BWH047 to MHBII media only, including comparison with GOLD_84 and GOLD_115 PA14 loci. 84/84 GOLD_84 homologs were identified as essential for growth in MHBII.113/15 GOLD_115 homologs were identified as essential for BWH047 growth in MHBII; bolded and italicized genes (from GOLD_115) were not identified as essential for growth in MHBII in BWH047. (XLSX) Muropeptide analysis of P. aeruginosa BWH047 and its derivatives. Includes raw peak areas and percentage of total peak areas for both replicates in all strains. (XLSX) Raw results from digital PCR run evaluating expression of arnB, dpcA, pmrA, and recA in BWH047 and PA14. (XLSX) Expression (cp/μL) of indicated genes in BWH047 and PA14 in the presence and absence of colistin, normalized to cp/μL recA. (XLSX) Expression (cp/μL) of indicated genes in BWH047 and PA14 in the presence and absence of colistin, normalized to input RNA. (XLSX) dMIQE2020 checklist for digital PCR experiment (adapted from the dMIQE Group & J Huggett, 2020 [104]. (XLSX) Antibiotic resistance profiles of the PABWH P. aeruginosa isolate collection. Minimum inhibitory concentrations as determined by BMD for aztreonam (A), cefepime (B), ceftazidime (C), ciprofloxacin (D), meropenem (E), and piperacillin-tazobactam (F) for all isolates within the PABWH collection (n = 85). The x-axis is the concentration of antibiotic in µg/mL and the y-axis is total number of isolates with each MIC. MICs are reported as susceptible (S, green), intermediate (I, yellow), or resistant (R, red), based on current CLSI guidelines (Ed. 100, M35). (H). Summary antibiogram for all 8 antibiotics assayed by broth microdilution for the PABWH isolate collection including colistin. The y-axis represents the percent of isolates that are sensitive (green), intermediate (yellow) or resistant (red) for each antibiotic list on the x-axis. (PPTX) TnSeq Experimental Design, Transposon Insertion Density Map, and Operonic Diagrams of TnSeq Hits. (A) Experimental design of TnSeq experiment challenging BWH047 transposon library with ¼ and ½ MIC colistin. Created in BioRender. Bachta, K. (2026) https://BioRender.com/os150y1. (B) Linearized heatmap showing distribution of transposon insertion sites across the BWH047 6.86Mbp genome with sites with greater frequency of insertions present as yellow and areas of lower density transposon insertions present as blue. The black region around 2.5Mbp represents a large ~90kbp region containing a duplication for which transposon depth was unable to be evaluated because the genes are diploid. Insertions per site is calculated in 1,000 bp ranges. (C) Operonic context for TnSeq hits with validated genes or validated neighboring genes. Minimum inhibitory concentrations of individual gene deletions indicated underneath. TnSeq gene hits from original screen indicated with *; – indicates gene deletion was not pursued. Created in BioRender. Bachta, K. (2026) https://BioRender.com/ggazop0. (PPTX) Quality Control for the creation of BWH047 transposon library and experimental execution. (A) Filtering schematic depicting methodology for determining high-confidence, analyzable TA sites present in the MHB II media control pools (n = 5). Created in BioRender. Bachta, K. (2026) https://BioRender.com/dyzeaz0.(B) Mean percentage read retention ± 1 SD from TnSeq by experimental condition. For each of the three experimental conditions (320 and 640 μg/mL colistin and MHBII), the mean percentage is calculated from 5 replicates. (C) Mean percentage TA site occupancy by experimental condition ± 1 SD. For each of the three experimental conditions (320 and 640 μg/mL colistin and MHBII), the mean percentage is calculated from 5 replicates, and the denominator is the total number of genomic TA sites (n = 107,934). Analyzable TA sites (n = 16,913) are defined as TA sites present in all 5 MHB II media control pools with ≥ 3 reads in all pools. All data presented in this figure represents TIS_tools preprocessing outputs and is shown in S9 and S10 Tables. (PPTX) Comparative analysis of LPS from selected strains of P. aeruginosa BWH047. The purified LPS was resolved on PAGE and visualized with silver after periodate oxidation. (A) LPS isolated from Salmonella typhimurium (control), P. aeruginosa BWH047, ∆dpcA, and (B) ∆wapH. Extracted LPS from both P. aeruginosa BWH047 WT (C) and ∆dpcA (D) was further hydrolyzed with 1M HCl-MeOH, re-N-acetylated, and the resulting monosaccharides and fatty acid residues were converted to TMS-methyl glycosides and TMS-fatty acid methyl esters (FAME-TMS, for hydroxyl fatty acids) and fatty acid methyl esters (FAME, for non-hydroxylated fatty acids) and analyzed by GC-MS. The summary of the glycosyl composition is listed in Table 2, and of FAME/FAME-TMS in Table 3. Rha, rhamnose; Ara4N, 4-amino-4-deoxy-L-arabinose; FucNAc, N-acetyl fucosamine; Man, mannose; Kdo, 3-deoxy-D-manno-oct-2-ulosonic-acid; QuiNAc, N-acetylquinovosamine; Glc, glucose; GlcNAc, N-acetylglucosamine; GalNAc, N-acetylgalacatosamine; Ino, inositol; 10:0, decanoic acid; 10:0(3-OH), 3-hydroxydecanoic acid; 10:0(2-OH), 2-hydroxydecanoic acid, 10:2(3-OH), 3-hydroxy-decadienoic acid; 16:0, hyexadecanoic (palmitic) acid; ∆8, cis-8cis-8-octadecanoic acid. Uncropped gel pictures for S4A and S4B are located in S1 Raw Images. (PPTX) Collision-Induced Dissociation (CID) MS/MS analysis of selected ions corresponding to lipid A substituted with 18:1, cis-8-octadecanoic acid. (A) Parent ion at m/z 1404 (Peak I) was fragmented in GC MS/MS. Detected fragment ions (peaks corresponding to fragments from proposed structures) are presented in bolded red alongside corresponding portions on the proposed structures (dashed lines, inset). (B) Parent ions at peaks F’ and F at m/z 1375.9 and 1391.9, respectively, were fragmented using GC MS/MS. Detected fragment ions (peaks corresponding to fragments from proposed structures) are presented in bolded red alongside corresponding portions on the proposed structures (dashed lines, inset). Peak F’ (m/z 1375.9) and Peak F (m/z 1391.9) are related by the addition of an –OH group, so only one probable structure is depicted (B/C, inset) representing both peaks with the aforementioned –OH group depicted in blue. (PPTX) Structural heterogeneity of P. aeruginosa BWH047 lipid A revealed by MALDI-TOF-MS. The listed, proposed lipid A structures (A-U) correspond to the major ions detected during MS analysis (Fig 3 and Table 4) and highlight the proposed structures of lipid A. Structures and m/z values marked with (‘) indicate a different structural version of lipid A deprived of hydroxyl group (-OH) on the acyloxy acyl chain. “Na” indicates the presence of a sodiated version of lipid A. (PPTX) DpcA does not alter peptidoglycan composition. (A) Peptidoglycan was purified from biological replicate cultures, digested with mutanloysin, and the muropeptide profile was analyzed by LC-MS. Strains included BWH047 (WT, black), ∆dpcA (blue), and complement (dpcA comp, purple). (B) Quantification of relative peak area from data collected in ‘A’. No statistical significance observed between strains using 2-Way ANOVA (p > 0.05) with multiple comparisons and Tukey post-hoc test. Raw peak areas and peak area percentages are listed in S12 Table. (PPTX) Expression of target genes in the presence and absence of colistin in BWH047 and PA14. (A) pmrA, arnB, and dpcA copies (cp)/µL normalized to recA cp/µL. (B) recA, pmrA, arnB, dpcA cp/µL normalized to input RNA concentration. Benjamini-Hochberg-adjusted p-values below 0.05 indicated. N = 4 biological replicates. The y-axes are plotted on a logarithmic (base 10) scale. The underlying data presented in this figure is shown in S14 and S15 Tablesand the raw expression data is in S13 Table. (PPTX) Uncropped silver-stained PAGE gels depicting LPS from selected strains of P. aeruginosa BWH047. (A) LPS isolated from Salmonella typhimurium (control), P. aeruginosa BWH047, ∆dpcA, and (B) ∆wapH. These represent the full gel pictures from cropped gels represented in S4A and S4B Fig. (PDF)