Genomic investigation unveils colistin resistance mechanism in carbapenem-resistant Acinetobacter baumannii clinical isolates

ABSTRACT Colistin resistance in Acinetobacter baumannii is mediated by multiple mechanisms. Recently, mutations within pmrABC two-component system and overexpression of eptA gene due to upstream insertion of ISAba1 have been shown to play a major role. Thus, the aim of our study is to characterize colistin resistance mechanisms among the clinical isolates of A. baumannii in India. A total of 207 clinical isolates of A. baumannii collected from 2016 to 2019 were included in this study. Mutations within lipid A biosynthesis and pmrABC genes were characterized by whole-genome shotgun sequencing. Twenty-eight complete genomes were further characterized by hybrid assembly approach to study insertional inactivation of lpx genes and the association of ISAba1-eptA. Several single point mutations (SNPs), like M12I in pmrA, A138T and A444V in pmrB, and E117K in lpxD, were identified. We are the first to report two novel SNPs (T7I and V383I) in the pmrC gene. Among the five colistin-resistant A. baumannii isolates where complete genome was available, the analysis showed that three of the five isolates had ISAba1 insertion upstream of eptA. No mcr genes were identified among the isolates. We mapped the SNPs on the respective protein structures to understand the effect on the protein activity. We found that majority of the SNPs had little effect on the putative protein function; however, some SNPs might destabilize the local structure. Our study highlights the diversity of colistin resistance mechanisms occurring in A. baumannii, and ISAba1-driven eptA overexpression is responsible for colistin resistance among the Indian isolates. IMPORTANCE Acinetobacter baumannii is a Gram-negative, emerging and opportunistic bacterial pathogen that is often associated with a wide range of nosocomial infections. The treatment of these infections is hindered by increase in the occurrence of A. baumannii strains that are resistant to most of the existing antibiotics. The current drug of choice to treat the infection caused by A. baumannii is colistin, but unfortunately, the bacteria started to show resistance to the last-resort antibiotic. The loss of lipopolysaccharides and mutations in lipid A biosynthesis genes are the main reasons for the colistin resistance. The present study characterized 207 A. baumannii clinical isolates and constructed complete genomes of 28 isolates to recognize the mechanisms of colistin resistance. We showed the mutations in the colistin-resistant variants within genes essential for lipid A biosynthesis and that cause these isolates to lose the ability to produce lipopolysaccharides.


Bacterial isolates
A total of 1,214 consecutive non-duplicated clinical isolates of A. baumannii were collected during 2016 to 2019 as a part of routine diagnosis at Christian Medical College, Vellore, India.Among 1,214, 314 isolates were from blood and 900 isolates from endotracheal aspirate (ETA).All the isolates were identified at the species level as Acinetobacter baumannii calcoaceticus complex (Acb complex) using conventional biochemical methods (13).Matrix-assisted laser desorption ionization-time-of-flight mass spectrometry was used to confirm at the species level as A. baumannii.Further confirmation of Acb complex as A. baumannii was performed targeting chromosomally encoded bla OXA-51 -like gene by PCR (14).Only carbapenem-resistant A. baumannii isolates were included in this study.

Minimum inhibitory concentration by broth micro dilution
Minimum inhibitory concentration (MIC) value for colistin was determined for all the isolates using broth micro dilution (BMD) and interpreted accordingly (16).Escherichia coli ATCC 25922 and Pseudomonas aeruginosa ATCC 27853 were used as quality control (QC) strains.An E. coli mcr-1-positive isolate was used as an internal control.Two in-house Klebsiella pneumoniae QC strains BA38416 and BA25425 with colistin MIC values of 0.5 and 16 µg/mL, respectively, were also included in every batch of testing.

Whole-genome sequencing, assembly, and annotation
Among the 1,214 strains, a subset of 207 A. baumannii isolates (blood = 103, ETA = 103, and pus = 1) were selected based on the source for further characterization by whole-genome sequencing (WGS).The genomic DNA was extracted using QIAamp DNA Mini Kit (QIAGEN, Germany) according to the manufacturer's instructions, and WGS was performed.In brief, short-read sequencing was performed for all 207 isolates using IonTorrent Personal Genome Machine (Life Technologies, USA) with 400-bp read chemistry or by Illumina MiSeq as per the manufacturer's instructions.Furthermore, a subset of 28 isolates (blood, n = 20; ETA, n = 7; and pus, n = 1) were selected based on the International Clones (ICs )and fewer novel sequence types (STs) for long-read sequencing.The long-read sequencing was performed using the SQK-LSK108 Kit R9 version (Oxford Nanopore Technologies, UK) using the 1D sequencing method according to the manufacturer's protocol.To obtain complete genome, a hybrid assembly was performed on 28 genomes as described previously (17).

Mutation analysis of lpxACD and pmrCAB genes
The lpxA, lpxC, lpxD, pmrA, pmrB, and pmrC sequences were extracted from whole genomes of A. baumannii isolates, and an in silico mutation analysis was conducted in all the genomes using BLAST analysis (https://blast.ncbi.nlm.nih.gov) and compared with the reference strain of A. baumannii ATCC 17978 (GenBank accession number CP000521).Other A. baumannii reference strains, such as AYE (GenBank accession number NC010410), ACICU (GenBank accession number NC010611), and ATCC 19606 (GenBank accession number CP046654), were included in the analysis to identify the genetic polymorphisms.Detection of mcr genes was performed using either ResFinder or in silico BLAST analysis against the reference sequences of mcr genes reported so far (21).In addition to the mutation analysis, all the 28 complete genome sequences were characterized manually for other colistin resistance mechanisms including inactivation of lpxA and lpxC genes due to insertion of ISAba11 element and overexpression of pmrC homolog, eptA, due to upstream insertion of ISAba1 element.

In silico analysis of protein structures by modeling
The protein sequences of LpxA, LpxC, LpxD, PmrA, PmrB, and PmrC were subjected to protein BLAST search against Protein Data Bank (PDB) (22) to verify the availability of their three-dimensional (3D) structures.The protein BLAST returned 100% identity with LpxA and PmrA sequences but not with the other proteins.Therefore, PDB has the structural information of LpxA (PDB ID: 4E6U) and PmrA (PDB ID: 7MUS); the 3D structures were downloaded in PDB format.The 3D structures of LpxC (AF-A3M9 × 5-F1) and LpxD (AF-A3M650-F1) were obtained from the AlphaFold protein structure database and retrieved.No structural information regarding PmrB and PmrC is currently available.To perform in silico protein modeling, iterative threading assembly refinement (I-TASSER) was used to predict the 3D structures of PmrB and PmrC (23).I-TASSER generated five different models, and the confidence of each model is quantitatively measured by C-score.which typically ranges from −5 to 2. C-score is a confidence score that was calculated based on the significance of threading template alignments and the convergence parameters of the structure assembly simulations.A higher C-score value signifies a model with a high confidence and vice versa.The mutant-type (MT) protein structures were obtained by replacing the specific point mutations in their respective wild-type (WT) protein structures using the Swiss PDB viewer package (24).PROSA (25) and PROCHECK (26) were used to check the quality of the predicted structures for further analysis.The modeled structures were validated from the percentage of residues in the Ramachandran favored region and the Z-score obtained from PROSA.The protein functional domains were determined from the InterPro server (http://www.ebi.ac.uk/ interpro/), which analyzes the function of proteins by organizing them into families, domain prediction, and critical sites.InterPro uses predictive models or signatures provided by different databases to classify the proteins (27).

Thermodynamic stability assessment
The effect of point mutations on the stability of the proteins was evaluated using the DUET online tool.The tool combines two complimentary approaches including site-directed mutator (SDM) and cutoff scanning matrix (mCSM).The results obtained by these approaches were combined by support vector machines.The tool was trained on experimental thermodynamic data sets derived from the PROTHERM database.DUET showed the stability of proteins through changes in the unfolding Gibbs free energy (ΔΔG kcal/mol) between the WT and the mutant proteins.Additionally, the mutations are classified as destabilizing (ΔΔG < 0) or stabilizing (ΔΔG > 0) (28).

AST status of the clinical isolates and the determination of MIC for colistin by BMD
When we analyzed the AST profile, we found that 1,212 isolates (blood, n = 313 and ETA, n = 899) were carbapenem resistant.Of the remaining two isolates, one from ETA was pan-susceptible (AB01, aAccession no.CP040080), while the other isolate (AB025, accession no.CP050432) was found to be resistant to cephalosporins, fluoroquinolones, aminoglycosides, tetracyclines, tigecycline, and trimethoprim-sulfamethoxazole and was identified as multi-drug resistant.Thirty-four isolates from blood (10.8%) and 74 isolates from ETA (8.2%) were colistin resistant.The colistin MIC range, MIC 50 , and MIC 90 of the blood and ETA isolates are tabulated (Table 1).

Evaluation of colistin resistance mechanisms by WGS
Sequencing analysis of both carbapenem/colistin-resistant A. baumannii (CR-ColRAB) (n = 115) and carbapenem-resistant colistin-susceptible A. baumannii (CR-ColSAB) (n = 92) revealed the presence of multiple amino acid substitutions in the lpx genes.We found single amino acid substitution in the lpxA gene (Y131H), four in the lpxC gene (C120R, P148S, F230Y, and N287D) and seven in the lpxD gene (Q4K, V63I, V93I, E117K, G166S, T287I, and S299P) when compared with the ATCC 17978 reference strain.Three amino acid substitutions including Q4K, V63I, and E117K were identified in lpxD, in both CR-ColSAB and CR-ColRAB isolates.Y131H, C120R, and N287D substitutions were identified in all the sequenced isolates and the three reference strains.It is important to mention that amino acid substitutions such as V93I, G166S, T287I, and S299P in the lpxD were not previously found.Similarly, we identified two substitutions in the lpxC gene (P148S and F230Y) that were never reported before.Sequencing of pmrAB genes identified a single amino acid substitution in pmrA (M12I) and seven amino acid substitutions in pmrB (A138T, L168S, A226T, V300E, G315A, P360Q, and A444V).Of note, only two mutations, T7I and V383I, were identified in the pmrC gene specific to colistin-resistant A. baumannii.The 28 complete genomes of A. baumannii were characterized for other colistin resistance mechanisms, such as the insertional inactivation of lpxA or lpxC by ISAba11 and the presence of ISAba1 upstream of eptA.Two CR-ColSAB blood isolates, AB016 (CP040259) and AB017 (CP050385), showed the presence of ISAba11, but no disruption of lpxA or lpxC was observed.At least one copy of the eptA gene was identified among the 10 and 6 genomes of CR-ColSAB from blood and ETA, respectively.In contrast, more than one copy of eptA gene was present in 12 genomes (seven CR-ColSAB and four CR-ColRAB isolates from blood and one CR-ColRAB isolate from ETA).
According to Fig. 1, ISAba1 is upstream of pmrC in AB02, upstream of eptA in AB03, but downstream of eptA in AB04.In addition, ISAba1 is in the reverse orientation compared to pmrC/eptA in AB02 and AB03 but is in the same orientation as eptA in AB04 (but downstream) (Fig. 1).
Two of the CR-ColSAB isolates from blood had ISAba1 insertion but not to the upstream of the pmrC or eptA genes (AB010 [CP040053] and AB011 [CP040056]), whereas three CR-ColSAB isolates from blood (AB08 [CP038500], AB015 [CP050403], and AB025 [CP050432]) have other family transposases (Fig. 2).Interestingly for one of the CR-ColRAB isolates, AB06, we detected a single amino acid substitution in the lpxA gene (Y131H), two substitutions in the lpxC gene (C120R and N287D), and two substitutions in the lpxD gene (P148S and G315D); however, we did not detect other known mutations or mechanisms that result in colistin resistance.The summary of the findings of CR-ColSAB and CR-ColRAB isolates are listed in Table 2.
Finally, it is important to mention that among the colistin-resistant isolates that we tested, we did not identify any of the reported plasmid-mediated colistin resistance determinant, such as mcr-1 or its homologs.
We found lineage-specific amino acid substitutions such as Q4K, V63I, and E117K within lpxD gene, among isolates belonging to IC8, IC7, and IC2 clades, respectively.We also identified that isolates belonging to the IC2 clade carry substitutions, such as M12I in the pmrA gene and the combination of A138T plus A444V within the pmrB gene.The effect of mutations was evaluated through in silico analysis.

Structural prediction of the protein and the impact of SNPs on stability
The 3D structural information for PmrB and PmrC was unavailable in the PDB database.Thus, we modeled the structures of PmrB and PrmC from their primary sequences using I-TASSER.The primary analysis generated five different structural models.The model with a C-score of −2.32 in the PmrB protein and the model with a C-score of 1.13 in the PmrC were selected as these models displayed higher confidence based on the C-scores.We further validated these structures individually.The modeled structures of PmrB and PmrC proteins showed >95% of residues in the favored and allowed regions, and their corresponding Z-scores were −5.56 and −5.96, respectively.Therefore, these structures were selected for further mapping analysis.The SNPs identified in this study were mapped onto the respective protein structures as shown in Fig. 3 (LpxACD) and Fig. 4 (PmrCAB).
The domains of LpxACD and PmrCAB, predicted by the InterPro server, are listed in Table 3.The possible structural impact of the mutations observed in this study was examined and is given in Table 4. Based on domain screening, in LpxA, even though the mutation was not present in the UDP N-acetylglucosamine O-acyltransferase domain, Y131H was found to be destabilizing.In LpxC, two SNPs (P148S and D159N) were mapped in the ribosomal protein domain that might destabilize LpxC, whereas the other SNPs (C120R in the N-terminal and N287D in the C-terminal) were found to be stabilizing the LpxC structure.Both SNPs (E117K and Q4K) in LpxD and an SNP (M12I) in PmrA were not mapped in their domain region and might stabilize the protein structures.Interestingly, we have noted that all SNPs (A138T, G315D, and A444V) in PmrB destabil izes the protein although A138T substitution was not present in the histidine kinase domain.In PmrC, nine SNPs were predicted.Among these, one SNP (F166L) was mapped in the phosphoethanolamine domain, while the other SNPs (R348K, A370S, and V383I) were mapped in the sulfatase.We found that only two SNPs (T7I and R348K) might be stabilizing the structure, whereas all other SNPs have a destabilizing effect.The exact relevance of these SNPs on the protein activity needs to be biochemically evaluated.

DISCUSSION
In this study, we found 8%-11% prevalence of colistin-resistant A. baumannii.Similarly, reports from other countries also suggest increased colistin resistance rates.For instance, 16.7% from Bulgaria, 19.1% from Spain, and 27% from Greece have been reported (29)(30)(31)(32).Therefore, these increasing prevalence of colistin resistance rate highlights the importance of routine testing and understanding of colistin resistance mechanisms (3).
Previous studies reported genetic polymorphisms within the lpxACD and pmrCAB operons that led to altered or loss of LPS/lipooligosaccharide production (9,10).In this study, we identified several SNPs within the lpxACD operon.The amino acid substitutions identified in this study in the lpxD gene (Q4K, V63I, and E117K) are similar to the previous studies (33,34).Recent studies suggest that mutations in the pmrB gene, which encodes for histidine kinase, are one of the major contributors for colistin resistance in A. baumannii (31,33,35,36).
Moffatt and colleagues have reported that inactivation of the lpxA or lpxC genes caused by insertion of ISAba11 elements resulted in the loss of LPS production that confers resistance to colistin (10).In addition, inactivation of the lpxC gene due to insertion of ISAba11 has also been reported (37).Though the current study showed the presence of ISAba11 in two CR-ColSAb isolates, disruption of lpxA or lpxC was not observed.
A single substitution M12I was detected in the pmrA gene, which encodes for the response regulator of the pmrCAB TCS.This substitution is thought to be associated with colistin heteroresistance (38).However, we have not detected other known substitutions such as G54E within the pmrA gene, which is reported to be associated with conferring high colistin resistance in A. baumannii either alone or in combination with other genes (36,39).In a recent study, it has been described that several substitutions, including G54E identified within the receiver domain of the pmrA response regulator, are responsible for colistin resistance in A. baumannii.Furthermore, G54E substitution alone or in combina tion with mutations in other genes can confer significantly high colistin resistance up to >256 or 512 µg/mL in A. baumannii (40).Previous studies detected A138T substitution in pmrB in addition to other substitutions (31,33,36).Also, a recent study by Srisakul et al. (41) found S14P and A138T in pmrB and reported that it could be correlated with colistin resistance.However, the current study identified only A138T, and the exact effect of this particular substitution needs further investigation.
Gerson and colleagues have reported the presence of additional substitutions in both the lpxD and pmrB genes (36).Interestingly, we found that similar substitutions in the lpxD gene (such as E117K) co-occurred with substitution in the pmrB gene (A138T and A444V) while analyzing CR-ColRAB and CR-ColSAB isolates.A similar observation was also reported by Zafer et al. (42).These findings suggest that not all substitutions are associated with colistin resistance, and this highlights the importance of consider ing their genetic background in addition to SNPs.Thus, further investigation on the role of these substitutions in lipid A modification and the resulting colistin-susceptible phenotype is required.It is evident that colistin resistance mechanisms in A. baumannii are much more complicated than imagined.
The pmrC gene encodes for pEtN transferase, which is necessary for LPS biogenesis (38).Lesho et al. showed the presence of an alternate gene, eptA, with pEtN transferase activity among the clinical isolates of A. baumannii (43).The length of the pmrC and eptA genes is approximately 1,650 bp, and it was found that they are homologous proteins with 93% amino acid identity thereby suggesting similar enzymatic activity (7).Two earlier reports have suggested that in the absence of the pmrA gene-mediated expression of pmrC, transposition of an insertion element, such as ISAba1, might lead to overexpression of the highly similar pEtN transferase, EptA, which allows addition of pEtN to lipid A and results in colistin resistance among the clinical isolates of A. baumannii (3,7).We found the presence of eptA in both CR-ColRAB and CR-ColSAB isolates similar to previous reports (3,7,8,43).However, the presence of the ISAba1 element to the upstream region of the pmrC/eptA loci was identified only among the CR-ColRAB isolates, which might lead to overexpression of pEtN transferase activity and increased colistin resistance.This observation is in agreement with previous studies (3,8,43).Recently, it has been reported that the disruption of the eptA gene by insertion of the ISAba125 element or increased expression of the eptA gene due to the insertion of ISAba1 in the reverse orientation to the upstream of eptA is also associated with colistin resistance (3).Though the difference in the orientation or the position of the ISAba1 insertion with respect to the eptA locus was identified in this study, further studies are warranted to understand the impact on the expression on eptA.
In summary, our study provides characterization of multiple resistance mechanisms that could be responsible for the emergence of colistin resistance among A. baumannii clinical isolates.Previously reported amino acid substitutions as well as other substitu tions that are not described previously within the lpxD, pmrA, and pmrB genes were identified in this study isolates.In particular, the presence of E117K in the lpxD gene along with A138T/A444V in the pmrB gene suggests a novel synergistic activity for the occurrence of colistin resistance.In this study, two colistin-susceptible isolates harbored ISAba11, and studies to understand the role of ISAba11 toward colistin resistance are essential.Though the altered expression pattern of pmrC and addition of pEtN transferase to the lipid A are regulated by PmrAB TCS, this may not be the sole contributor of colistin resistance.The presence of the additional eptA gene encoding pEtN transferase and the insertion of the ISAba1 element to the upstream of the eptA/pmrC loci among the colistin-resistant isolates might be associated with colistin resistance.The exact resistance mechanism that contributes to colistin resistance was not elucidated for two colistin-resistant isolates (AB05 and AB06) in this study and requires further investigations including transcrip tomic analysis.
It must be noted that most of the amino acid substitutions in lpxACD and pmrCAB identified from our study were present both in colistin-resistant as well in susceptible isolates except the substitutions of T7I and V383I in the pmrC gene that are present only in the colistin-resistant isolate, AB06.Even though both substitutions were not present in the functional domain of the protein, V383I destabilizes the PmrC protein structure, whereas T7I stabilizes the protein.To the best of our knowledge, these two substitutions within the pmrC gene are novel and not reported elsewhere.Similarly, we observed certain amino acid substitutions in the lpxD (V63I and G166S) and the pmrC (V118F, I131V, I131N, V151A, Q232H, D298G, I342T, K531T, and K514N) genes that are present only in colistin-susceptible isolates.
With respect to the epidemiologic aspect of the CR-ColSAB and CR-ColRAB observed in this study, we found several lineage-specific mutations, such as Q4K that belongs to IC8, V63I in IC7, and E117K/M12I/A138T in IC2 clades.Earlier studies reported IC2 as the predominant lineage involved in outbreaks (8,13).We consider these as novel findings because of the associations of the SNPs with the clones, which are not reported earlier.
Overall, the present study highlights the diversity of colistin resistance mechanisms among the clinical isolates of A. baumannii.

FIG 1
FIG 1 Genetic arrangement of pmrC/eptA with upstream presence of ISAba1 among complete genomes of colistin-resistant A. baumannii (AB02, AB03, and AB04).The direction of the arrow represents the orientation; pmrC/eptA is shown as red arrows, ISAba1 as purple arrows, pmrA as blue arrows, and pmrB as green arrows.The genetic arrangement of isolates AB05 and AB06 has eptA and pmrC without ISAba1.

FIG 4
FIG 4 Mapping of mutations in PmrCAB.(A) PmrA, domain signal transduction response regulator in red and domain OmpR/PhoB-type DNA binding in yellow; (B) PmrB, histidine kinase domain in magenta; and (C) PmrC, domain phosphoethanolamine transferase in orange and domain sulfatase in green.Key regulatory amino acids are represented in stick model, pale green color, and mutated amino acids are represented in stick model, blue color.

TABLE 1
MIC range, MIC 50 , and MIC 90 of colistin for clinical isolates of A. baumannii

TABLE 2
Characterization of various colistin resistance mechanisms among complete genomes of A. baumannii (n = 28) (Continued) a Underlined text indicates novel SNPs identified in this study.

TABLE 3
Domains of LpxACD and PmrCAB predicted by the InterPro server

TABLE 4
Predicted effect of protein stability in the presence of amino acid mutations in the LpxACD and PmrCAB proteins