Antibiotic Resistance Determinant-Focused Acinetobacter baumannii Vaccine Designed Using Reverse Vaccinology

As one of the most influential and troublesome human pathogens, Acinetobacter baumannii (A. baumannii) has emerged with many multidrug-resistant strains. After collecting 33 complete A. baumannii genomes and 84 representative antibiotic resistance determinants, we used the Vaxign reverse vaccinology approach to predict classical type vaccine candidates against A. baumannii infections and new type vaccine candidates against antibiotic resistance. Our genome analysis identified 35 outer membrane or extracellular adhesins that are conserved among all 33 genomes, have no human protein homology, and have less than 2 transmembrane helices. These 35 antigens include 11 TonB dependent receptors, 8 porins, 7 efflux pump proteins, and 2 fimbrial proteins (FilF and CAM87009.1). CAM86003.1 was predicted to be an adhesin outer membrane protein absent from 3 antibiotic-sensitive strains and conserved in 21 antibiotic-resistant strains. Feasible anti-resistance vaccine candidates also include one extracellular protein (QnrA), 3 RND type outer membrane efflux pump proteins, and 3 CTX-M type β-lactamases. Among 39 β-lactamases, A. baumannii CTX-M-2, -5, and -43 enzymes are predicted as adhesins and better vaccine candidates than other β-lactamases to induce preventive immunity and enhance antibiotic treatments. This report represents the first reverse vaccinology study to systematically predict vaccine antigen candidates against antibiotic resistance for a microbial pathogen.


Introduction
Acinetobacter baumannii is a Gram-negative opportunistic bacterial pathogen that is responsible for a diverse range of infections including ventilator-associated pneumonia, skin and wound infections, urinary tract infections, meningitis and bacteremia [1]. The majority of infections caused by A. baumannii are hospital-acquired, most commonly in the intensive care setting of severely ill patients. In addition, severe community-acquired pneumonia caused by A. baumannii has also been reported [2]. A. baumannii has become one of the most important and troublesome human pathogens with its increased number of infections and emergence of more threatening multidrug-resistant and pan-drug resistant strains [3]. Antibiotic resistance has greatly affected the effectiveness of antibiotic treatments. The development of alternative approaches is necessary.
Vaccination strategies are emerging as a viable option to prevent and/or treat multi-or pandrug-resistant infections. Although there is currently no licensed vaccine against A. baumannii infections, various vaccine candidates, including inactivated whole cell [4], outer membrane complexes (OMCs) [5], outer membrane vesicles (OMVs) [6], OmpA [7], Ata [8], Bap [9], K1 capsular polysaccharide [10] and PNAG [11], have recently been proven effective at some levels to protect against challenges in animals (mostly mice) from homologue strains and clonally distinct clinical isolates through active or passive immunization strategies. However, the clinical applications of reported vaccine candidates are limited by the potential regulatory and safety, and these antigens do not directly target antibiotic resistance [12]. β-lactamases are the most prevalent mechanism of antibiotic resistance in A. baumannii [13]. Ciofu et al. showed over 10 years ago that the animals immunized with β-lactamase proteins could induce strong neutralizing antibody responses and significant lower bacterial load and pathology [14,15]. This suggests an important solution against antibiotic resistance. However, no further progress has been reported along this line of research. Therefore, identification of more vaccine antigens that stimulate best immune responses against infections as well as antibiotic resistance is still a major task and challenge.
As an emerging and revolutionary vaccine development approach, reverse vaccinology (RV) starts with the prediction of vaccine protein targets by bioinformatics analysis of genome protein-coding sequences [16]. With the initial bioinformatics analysis, RV facilitates rapid vaccine design with less reliance on conventional animal testing and clinical trials. RV has been applied to the development of vaccines against a variety of pathogens such as serogroup B Neisseria meningitidis (MenB) [17], Bacillus anthracis [18], Streptococcus pneumoniae [19], Mycobacterium tuberculosis [20], and Cryptosporidium hominis [21]. More recently, several laboratories around the world began to identify A. baumannii vaccines candidates by RV approach [22][23][24][25]. Using 10 complete and 31 draft A. baumannii genomes, Moriel1 et al. applied RV and identified 42 A. baumannii antigens as potential vaccine targets [25]. By applying RV to analyze 14 A. baumannii genome sequences, Chiang et al. identified 13 novel proteins as potential vaccine candidates, and 3 out of these 13 antigens (OmpK, FKBP-type 22KD peptidyl-prolyl cis-trans isomerase (FKIB), Ompp1) were experimentally tested and proven to be highly immunogenic and conferred partial protection (60%) in a mouse pneumonia animal model [23]. Hassan et al. recently estimated the pan-genome of 30 complete A. baumannii genome sequences and identified 13 highly antigenic proteins as conserved immunogenic targets [24]. These three RV studies focused on the identification of conserved outer membrane and secreted proteins. The study conducted by Hassan et al. [24] also analyzed other aspects including transmembrane helices, 3D protein structural analysis, epitope mapping, protein function analysis. Out of 57 vaccine candidates predicted by Vaxign RV analysis using A. baumannii ATCC19606 genome sequences, Singh et al. specifically studied an outer membrane protein FilF and confirmed its value as a protective antigen [22]. However, no report is available on the usage of RV for vaccine design against A. baumannii antibiotic resistance.
Vaxign is the first web-based vaccine design program that predicts vaccine targets based on the RV strategy [26][27][28]. The Vaxign computational pipeline includes the prediction of many features, including subcellular localization, topology (transmembrane helices and β barrel structure), adhesin probability, sequence similarity to other pathogen sequences, similarity to host (e.g., human) genome sequences, and MHC class I and II epitope predictions [27]. In addition to the A. baumannii genome analysis by Singh et al. [22], the Vaxign RV approach has successfully predicted vaccine targets for many pathogens such as uropathogenic Escherichia coli [26], Brucella spp. [29], Rickettsia prowazekii [30], Streptococcus agalactiae [31], Corynebacterium pseudotuberculosis [32], and Campylobacter [33]. However, how Vaxign can be used for predicting antibiotic resistance determinants as vaccine candidates has not been demonstrated.
In this study, we used Vaxign and other bioinformatics methods to analyze 33 complete genome sequences and 84 representative antibiotic resistance determinants to predict A. baumannii protective vaccine antigens with a focused effort to identify antibiotic resistance determinants feasible for anti-resistance vaccine development.
In addition, our study collected 84 representative resistance genes from the ARDB-Antibiotic Resistance Genes Database (Available online: https://ardb.cbcb.umd.edu/), and NCBI Gene and Protein databases. Figure S1 provides the protein accession numbers and Vaxign analysis results for all these resistance genes. Extracted from various A. baumannii strains (e.g., A. baumannii AYE, ACICU, AB0057), these genes encode for proteins responsible for the resistance to most classes of antibiotics used in clinical practice, including β-lactams, aminoglycosides, quinolones, chloramphenicols, tetracyclines, sulfanilamide, tremethoprimand and polymycin. These resistance determinants utilize various antimicrobial resistance mechanisms, including enzymatic mechanisms, changes in OMPs and multidrug efflux pumps (Table 2). The above 33 complete genome sequences and 84 resistance determinants sequences were used to perform Vaxign RV analysis to identify A. baumannii vaccine candidate antigens. For the resistance determinants analysis, all 84 antibiotic resistance proteins were combined for a one single Vaxign pipeline analysis. Our overall RV prediction pipelines and results are shown in Figure 1 and are detailed in the following sections.

Predicted A. baumannii Vaccine Targets Based on RV Analysis of 33 Genomes
A. baumannii strain AYE is a representative multidrug-resistant strain that is epidemic in France [36]. Containing a resistance genomic island with 45 resistance genes clustered, this strain has been frequently used in whole genome sequence analyses (including RV analyses) [23,25,36,37]. The proteome of strain AYE includes 3712 proteins. In our Vaxign analysis, strain AYE was also used as the reference seed strain. As shown in Figure 1 and detailed below, Vaxign generated different results given different parameter settings.

Predicted A. baumannii Vaccine Targets Conserved in All 33 Genomes
Among all 3712 proteins of strain AYE, 2182 are conserved among all 33 strains. A BLASTp sequence comparison found that 639 of these 2182 proteins have homology with human proteins. Pathogen antigens having holomogy with host proteins are not preferred vaccine candidates since they likely induce autoimmunity or immune tolerance [38]. After removing these proteins and proteins with >1 transmembrane helices, we obtained 126 adhesin proteins. Proteins with a transmembrane domain less than or equal to one are selected since multiple transmembrane domains makes the purification of recombinant proteins difficult [38]. Adhesins are important for bacterial invasion and infection and are preferred vaccine candidates [38][39][40][41]. Outer membrane proteins (OMPs) and extracellular proteins are more likely to be protective vaccine candidates able to stimulate strong protective immunity [38,42]. Out of 126 adhesin proteins, 35 were predicted to be outer membrane or extracellular proteins ( Figure 1). Table 3 provides more detailed information about these 35 proteins. Specifically, these proteins include 27 OMPs and 8 extracellular proteins. The adhesin probabilities of these 35 proteins are between 0.513 to 0.744, which are all above 0.51, the cutoff of assigning a protein as an adhesin [43]. These 35 proteins have the lengths between 143 to 855 amino acids (Table 3).
To refine the selection, the antigenicity scores calculated by the VaxiJen v2.0 server (Available online: http://www.ddg-pharmfac.net/vaxijen/VaxiJen/VaxiJen) were applied. Our Vaxijen analysis found that all the 35 potential vaccine candidates have antigenicity scores from 0.42 to 0.82. An antigenicity score of over 0.40 is an indication of protein antigenicity [44]. Therefore, all 35 candidates were predicted to be antigenic.
For Cluster of Orthologous Groups (COG) protein functional categorization, 33 out of the 35 proteins fall into 7 different functional groups. Among these groups, "cell wall/membrane/envelope biogenesis" and "inorganic ion transport and metabolism" have 11 proteins each. Four COG groups have one protein each, including "secondary metabolites biosynthesis, transport, and catabolism", "posttranslational modification, protein turnover, chaperones", "cell motility, intracellular trafficking, secretion, and vesicular transport", and "cell wall/membrane/envelope biogenesis, inorganic ion transport and metabolism", and "function unknown" group has 7 proteins (Table 3).

Predicted A. baumannii Vaccine Targets Conserved in All 33 Genomes
Among all 3712 proteins of strain AYE, 2182 are conserved among all 33 strains. A BLASTp sequence comparison found that 639 of these 2182 proteins have homology with human proteins. Pathogen antigens having holomogy with host proteins are not preferred vaccine candidates since they likely induce autoimmunity or immune tolerance [38]. After removing these proteins and proteins with >1 transmembrane helices, we obtained 126 adhesin proteins. Proteins with a transmembrane domain less than or equal to one are selected since multiple transmembrane domains makes the purification of recombinant proteins difficult [38]. Adhesins are important for bacterial invasion and infection and are preferred vaccine candidates [38][39][40][41]. Outer membrane proteins (OMPs) and extracellular proteins are more likely to be protective vaccine candidates able to stimulate strong protective immunity [38,42]. Out of 126 adhesin proteins, 35 were predicted to be outer membrane or extracellular proteins ( Figure 1). Table 3 provides more detailed information about these 35 proteins. Specifically, these proteins include 27 OMPs and 8 extracellular proteins. The adhesin probabilities of these 35 proteins are between 0.513 to 0.744, which are all above 0.51, the cutoff of assigning a protein as an adhesin [43]. These 35 proteins have the lengths between 143 to 855 amino acids (Table 3).
To refine the selection, the antigenicity scores calculated by the VaxiJen v2.0 server (Available online: http://www.ddg-pharmfac.net/vaxijen/VaxiJen/VaxiJen) were applied. Our Vaxijen analysis found that all the 35 potential vaccine candidates have antigenicity scores from 0.42 to 0.82. An antigenicity score of over 0.40 is an indication of protein antigenicity [44]. Therefore, all 35 candidates were predicted to be antigenic.
For Cluster of Orthologous Groups (COG) protein functional categorization, 33 out of the 35 proteins fall into 7 different functional groups. Among these groups, "cell wall/membrane/envelope biogenesis" and "inorganic ion transport and metabolism" have 11 proteins each. Four COG groups have one protein each, including "secondary metabolites biosynthesis, transport, and catabolism", "posttranslational modification, protein turnover, chaperones", "cell motility, intracellular trafficking, secretion, and vesicular transport", and "cell wall/membrane/envelope biogenesis, inorganic ion transport and metabolism", and "function unknown" group has 7 proteins (Table 3).  Domains are distinct functional and/or structural units of proteins. A conserved domain footprint may reveal aspects of a protein's molecular or cellular function [45]. Of the total 35 potential vaccine candidates, 11 proteins were identified as TonB dependent receptors (PF00593), and most of them are putative ferric siderophore receptor proteins. An example TonB dependent receptor is CAM86022.1 (baumannii acinetobactin utilization, BauA), which has been identified to play an important role in iron uptake [46]. Another large group of proteins is defined as porin or efflux related proteins. Eight porin proteins, such as OmpA family lipoprotein (CAM87753.1) and OprD superfamily porins proteins (CAM88440.1 and CAM86576.1), were identified. Seven efflux related proteins, such as ATP binding cassette (ABC) superfamily transporter (CAM85361.1) and resistance nodulation cell division (RND) family transporter (CAM85249.1 and CAM85825.1), were predicted by Vaxign. In addition, fimbrial or pilus related protein (CAM87009.1 and CAM87933.1), copper resistance protein NlpE (CAM87612.1) and some other proteins with known or unknown functions were also identified. The detailed information is summarized in Table 3.

Predicted A. baumannii Vaccine Targets Absent from 3 Antibiotic-Sensitive Strains
Among the 33 A. baumannii strains, three strains (i.e., AB307-0294, ATCC17978, and D1279779) are drug sensitive (Table 1). We hypothesized that those drug resistance determinants commonly found in MDR strains most likely do not exist in these drug sensitive strains. When we excluded any proteins in these three drug sensitive strains, 366 proteins were found in our Vaxign analysis ( Figure 1). These 366 proteins include many commonly known drug resistance determinants such as β-lactamases, aminoglycoside modifying enzymes, dihydrofolate reductase, chloramphenicol resistance protein, tetracycline resistance proteins and an efflux transporter. After applying no human similarity and transmembrane helix criteria, Vaxign predicted 15 adhesins. Fourteen of these 15 adhesins include many hypothetic proteins and one drug resistance determinant chloramphenicol acetyltransferase (CAM88367.1). Note that CAM88367.1 was predicted to have an unknown subcellular location (probability = 0.2). Interestingly, only one of these 15 adhesins exists in outer membrane or extracellular locations. Specifically, this protein is CAM86003.1, a 390 amino acid protein with an adhesin probability of 0.655 and an outer membrane location (probability = 0.952). This protein has been annotated as "conserved hypothetic protein; putative signal peptide" (Available online: https://www.ncbi.nlm.nih. gov/protein/CAM86003.1).
In addition to the absence from three drug sensitive strains, we added a criterion that a protein needed to be conserved in at least 10 other strains. With these two restrictions, Vaxign found 59 proteins from strain AYE (Figure 1). After using the criteria of no human similarity and transmembrane helix ≤1, we found 2 adhesins. These two adhesins include CAM86003.1 (as described above) and CAM86739.1. CAM86739.1 is another hypothetical protein with only 96 amino acids and an unknown function. Further analysis found that CAM86003.1 is conserved in 21 antibiotic-resistant strains, and CAM86739.1 is conserved in 19 antibiotic-resistant strains. These results suggest that these two proteins are possible protective antigens worth experimental evaluations.

Predicted A. baumannii Vaccine Targets Based on RV Analysis of 84 Antibiotic Resistance Determinants
Subcellular localization is a major selection criterion in RV analysis. To target against antibiotic resistance, we hypothesize that human antibody response plays a major role. Therefore, outer membrane and extracellular antibiotic resistance determinants, esp. those with high adhesin probabilities, are most favorable vaccine candidates against antibiotic resistance. Owing to their capability of hydrolyzing most found β-lactam antibiotics, β-lactamases form the most prevalent mechanism to β-lactam resistance. Typically produced in bacterial periplasms, β-lactamases can be packed inside outer membrane vesicles (OMVs) which can be released to the extracellular environment. The released β-Lactamase-containing OMVs then induce the production of anti-β-lactamse IgG [47,48]. In addition, Pseudomonas aeruginosa β-lactamase proteins have been found to induce strong neutralizing antibody responses and lower bacterial load in animal models [14,15]. Therefore, for the prediction of A. baumannii vaccine targets against antibiotic resistance, we also included those resistance determinants with periplasmic localization.
Our further multiple sequence alignment identified many sequence gaps exist among 11 representative β-lactamases (Figure 2a). The protein sequence identity among these 11 β-lactamases ranges from 11.64% to 66.26%, indicating significant sequence variations among β-lactamases. Among Ambler class Our phylogenetic tree analysis further shows the evolutionary relationships and distances among these 11 proteins (Figure 2b). Four classes of β-lactamases were separately clustered. Specifically, 5 class A β-lactamases were clustered together, 4 class B β-lactamases were grouped in its own cluster. The one class C and one class D β-lactamases were separated from the class A or B clusters.

Discussion
The major contributions of this study are two-fold. First, we applied RV methods to analyze 33 complete A. baumannii genomes and identify many A. baumannii vaccine candidates using traditional whole genome analysis methods. Second, our RV methods with these 33 genomes and an additional collection of 84 representative antibiotic resistance determinants identified many vaccine candidates against antibiotic resistance. The second contribution also represents the first study to use an RV strategy for systematic vaccine design against antibiotic resistance for any microbial pathogen.
A very significant finding from our study is the list of 11 TonB-dependent receptors among all 35 predicted vaccine candidates conserved among 33 genomes ( Table 1). The RV studies by Moriel1 et al. [25] and Chiang et al. [23] did not identify any such receptor, and Hassan et al. [24] identified only one such receptor. TonB-dependent receptors are bacterial OMPs that primarily bind and transport ferric chelates called siderophores and also support the transport of vitamin B12, nickel, and carbohydrates in a TonB-dependent manner [49]. Iron transport into the cytosol is mediated by these specific TonB-dependent membrane receptors that recognize the iron-siderophore complexes. Iron acquisition is generally required for bacterial growth during infection [50]. The expression of a siderophore-ferric complex receptor is also critical for the establishment and persistence of A. baumannii infections [51]. It has been found that antibodies directed against proteins associated with iron uptake exert a bacteriostatic or bactericidal effect against A. baumannii in vitro [52]. Recently, the recombinant BauA, a member of TonB-dependent receptors, was experimentally verified as a valid protective vaccine antigen as shown by mouse challenge and passive serum immunization experiments [46]. Interestingly, surface-exposed iron receptors have also been used in the development of vaccines against other human pathogens e.g., uropathogenic E. coli [53]. Our identification of 11 TonB-dependent receptors provides more options of using TonB-dependent receptors in A. baumannii vaccine development. It would also be important to determine how these different TonB dependent receptors interact with each other.
Our RV analysis identified a putative pilus assembly protein (FilF) (CAM87933.1) and a fimbrial protein CAM87009.1. Recombinant FilF protein was experimentally verified to elicit a strong protective response against A. baumannii [22]. Fimbriae are proteinaceous filaments expressed on the surfaces of many pathogenic bacteria. These filaments are involved in bacterial adhesion to the host cells and are considered important for virulence [32]. Fimbriae are recognized as potential vaccine antigens of several pathogenic bacteria e.g., E. coli [33] and Bordetella pertussis [34]. Given its high adhesin probability, our study predicted the important role of CAM87009.1 in fimbrial adhesion. The role of CAM87009.1 as a protective antigen deserves experimental evaluation.
Porins are proteins that are able to form channels allowing the transport of hydrophilic compounds up to a certain size exclusion limit across lipid bilayer membranes [54,55]. Porins can play a variety of roles depending on the bacterial species, including transport of small molecules, maintenance of cellular structural integrity, bacterial conjugation and bacteriophage binding. Porins may also play a significant role in antibiotic resistance [56]. The porin protein OmpA is the most abundant surface protein and plays a role in the permeability of small solutes such as β-lactams and saccharides. In addition, OmpA of A. baumannii serves as an important virulence factor in the bacterial interaction with epithelial cells, induction of apoptosis of host cells, and dissemination of bacteria into the bloodstream [57]. OmpA has been identified as a highly promising candidate for active and passive immunization based on humoral immunodominance during lethal A. baumannii infection in mice [7]. Using an immunoproteome-based approach, Fajardo et al. also demonstrated that porin-related proteins including OmpA and OprB-like are highly abundant on the bacterial surface and can mount an immune response [56]. In our study, eight porin-related proteins including OmpA family lipoproteins (CAM87753.1), OprD superfamily porin proteins (CAM88440.1 and CAM86576.1), OprB family porin proteins (CAM85599.1, CAM86576.1), and Porin_7 superfamily proteins (CAM85174.1 and CAM85154.1) were identified. This list not only includes known vaccine candidates OmpA and OprB but also other porins whose potential roles as protective antigens and virulence factors are still unknown and are worth experimental testing.
Considering that OmpA is very conserved in the outer membranes of Gram-negative bacteria, it might be possible to develop a universal vaccine using OmpA against various bacterial pathogens. Chen et al. [58] compared the homology of OmpA proteins in three pathogenic Yersinia and found that their amino acid sequences have >98.6% identity. Cross-immunogenicity was also observed using three pathogenic Yersinia OmpAs [58], further supporting the speculation that OmpA is likely to be a common protective antigen against pathogenic Yersinia [58]. We performed a multiple amino acid sequence alignment analysis with 10 OmpA amino acid sequences from ten different bacteria, including 8 bacteria from the family Enterobacteriaceae (i.e., Escherichia coli, Klebsiella pneumonia, Serratia marcescens, Yersinia bercovieri, Proteus mirabilis, Salmonella enterica subsp. enterica serovar Typhimurium, Shigella flexneri, and Cronobacter sakazakii), Brucella abortus, and A. baumannii. Our study indicates that although the OmpA proteins from these 8 Enterobacteriaceae bacteria share high identities (70.06% to 97.38%), the sequence identities between A. baumannii (or B. abortus) and these 8 bacteria are only 21%-29% (data not shown). Therefore, an OmpA vaccine against many bacteria in one species or one family can likely be developed. However, it may be difficult to develop a universal OmpA vaccine against all bacteria in different families. Meanwhile, a possible effect of an OmpA vaccine on microbiota may need to be investigated if such an OmpA vaccine against many bacteria is developed.
Efflux pumps are transport proteins involved in the extrusion of toxic substrates (including antibiotics of multiple classes) from within cells into the external environment, leading to the prevention of intracellular accumulation of toxic compounds [59]. The overexpression of these transporters is associated with bacterial multidrug resistance [55]. Five superfamilies of efflux systems are found in A. baumannii: ATP-binding cassette (ABC) transporters, resistance-nodulation-cell division (RND) family, small multidrug resistance (SMR), multidrug and toxic compound extrusion (MATE) families, and the major facilitator superfamily (MFS) [60]. Our study found one ABC transporter, i.e., MlaD protein (CAM85361.1), which is a toluene tolerance efflux transporter. MlaD is predicted to be secreted to the extracellular environment and has high adhesin probability and antigenicity, suggesting its potential role as a protective vaccine antigen. Overproduction of RND efflux pumps has been found to be associated with MDR, virulence, and biofilm formation in A. baumannii [61]. Our genome sequence analysis found two RND transporters CAM85249.1 (CzcC) and CAM85825.1. Based on our BLAST sequence analysis, the CzcC sequence is assigned with NCBI reference sequence ID WP_005115306.1, which is a conserved RND transporter among over 100 A. baumannii strains (Available online: https: //www.ncbi.nlm.nih.gov/protein/WP_005115306.1). These two proteins are outer membrane proteins with high adhesin probability and antigenicity scores, suggesting their potential usage in vaccine development against A. baumannii infections. Considering their role in MDR, these two proteins are possible candidates for developing vaccines against drug resistance (see more discussion below).
Recent studies have identified the potential role of quorum sensing in antibiotic resistance [62]. A quorum sensing system is a cell density-based intercellular communication system, which utilizes hormone-like compounds referred to as autoinducers to regulate bacterial gene expression [63]. Quorum sensing can regulate multidrug resistance by upregulation of biofilm-associated genes and efflux pump genes [64,65]. Using an integrated proteomics study, Piras et al. [66,67] identified many quorum sensing-related proteins (e.g., LuxS) differentially expressed in multi-drug resistant E. coli compared to a control E. coli group, suggesting the association between quorum sensing and drug resistance. In A. baumannii, the quorum sensing system has also been identified and proved to play a key role in the regulation of bacterial virulence and biofilm formation [68]. The inhibition of the quorum-sensing activity in A. baumannii was found able to attenuate biofilm formation and decrease bacterial virulence [69,70], indicating that quorum sensing could be a promising target for developing new strategies against A. baumannii infection. More studies will be required to investigate how quorum sensing might be directly associated with antibiotic resistance in A. baumannii and how a vaccine targeting the quorum sensing mechanism might be effective against antibiotic resistance.
In addition to the conventional RV prediction, our study has a novel focus on identifying resistance-conferring proteins feasible for vaccine development against drug-resistant pathogens. Targeting vaccines against resistance determinants has been proposed to be a possible effective way to counteract selection pressure for antimicrobial resistance [71,72]. However, there is no report on using RV strategy to predict resistance determinants against antibiotic resistance. Our strategy of RV study of resistance determinants relies on two approaches: (1) Using Vaxign to compare MDR strains vs. three antibiotic-sensitive strains ( Figure 1); (2) Analysis of our collection of 84 resistance-conferring proteins from all reported A. baumannii strains. We assume that antibody response is the most important against antibiotic resistance. Therefore, our study focused on those OMPs, extracellular proteins, and adhesins, against which antibodies are very effective.
In general, using the first approach, we found 366 proteins that do not exist in any of the three antibiotic-sensitive strains (Figure 1). Out of these 366 proteins, our study found hypothetic outer membrane protein CAM86003.1 and hypothetic protein CAM86739.1, both having high adhesin probabilities. Using the second method, we predicted three extracellular and outer membrane proteins. In addition, we identified 21 periplasmic proteins, among which three have high adhesin probabilities.
RND efflux pumps are organized as a three-component system: a transporter (efflux) protein, located in the inner (cytoplasmic) membrane; a periplasmic accessory protein (also known as a membrane fusion protein (MFP); and an OMP channel, located in the outer membranes of Gram-negative bacteria. In Campylobacter jejuni, CmeC is an essential OMP of the CmeABC RND multidrug efflux pump. Zeng et al. discovered that the antibodies of the CmeC peptide inhibited the function of the CmeABC efflux pump and enhanced the susceptibility of C. jejuni to natural antimicrobial (bile salts) present in the intestine [73]. Purified recombinant CmeC also stimulated CmeC-specific serum IgG responses via oral vaccination in a chicken model of C. jejuni infection; however, the recombinant CmeC vaccination did not confer protection against C. jejuni infection [73]. It has also been demonstrated that inhibition of multidrug RND efflux pumps by efflux pump inhibitors (EPIs) is an effective approach to improve the antimicrobial drug susceptibilities of clinical A. baumannii isolates [74]. These findings suggest that outer membrane RND efflux pump proteins may be promising vaccine candidates to enhance clinical antibiotic activity and possibly prevent infections by MDR A. baumannii strains.
From our analyses of 33 genomes and 84 resistance determinants, three outer membrane RND transporters were identified as vaccine candidates against drug resistance. Our 33-genome analysis identified two RND type efflux pump proteins, CAM85249.1 (CzcC) and CAM85825.1. Since these two proteins were not reported in the literature or databases, they were not included in our original list of 84 resistance determinants. Our study with 84 resistance determinants found one outer membrane RND transporter (WP_000045119.1). To further evaluate their value as vaccine candidates, we used Vaxign to analyze C. jejuni CmeC (discussed above) [73] and compared different Vaxign analysis results. Interestingly, C. jejuni CmeC [73] was found to have adhesin probability (0.371) less than 0.51, suggesting a lack of the adhesin role [43]. In contrast, CzcC and CAM85825.1 had adhesin probabilities of 0.64 and 0.51, respectively, suggesting both are possible adhesin proteins. Given the importance of adhesins as likely protective antigens, CzcC and CAM85825.1 might be more likely than C. jejuni CmeC [73] to stimulate protection against virulent bacterial infection.
Our RV studies on the 33 genomes and 84 resistance determinants found two outer membrane porin (OprD)-like vaccine candidate proteins: CAM88440.1 in strain AYE and WP_000018327.1. OprD promotes the uptake of basic amino acids and small peptides containing these amino acids, and it can also serve as a specific channel for carbapenems by structural homology [75]. Carbapenems are antibiotics for treating infections caused by MDR bacteria. The loss of or reduced OprD confers resistance to carbapenems in P. aeruginosa [76,77]. However, Catel-Ferreira et al. demonstrated that the lack of OprD did not affect the susceptibility of A. baumannii to treatment with carbapenem antibiotics [78], suggesting that A. baumannii OprD is likely not involved in the carbapenem resistance mechanism. More investigation is needed to determine their exact role in antibiotic resistance and potential as vaccine candidates. Among 84 resistance determinants, QnrA (ADB64519.1) is the only predicted extracellular protein. The first discovered QnrA is coded by a 56-kb broad-host range conjugative plasmid, pMG252, that confers an unusual multidrug resistance phenotype, including resistance to quinolones, β-lactams, aminoglycosides, sulphonamides, trimethoprim, and chloramphenicol [79]. QnrA belongs to the pentapeptide repeat family and protects DNA gyrase and type IV topoisomerase from quinolone inhibition. The Qnr determinants have been identified in a series of enterobacterial species and nonenterobacterial Gram-negative species like P. aeruginosa and A. baumannii worldwide. Many epidemiological surveys show an association between Qnr-like determinants and the β-lactamases with the widest spectrum of activity, i.e., carbapenemases (Ambler class A or class β-lactamases) [80]. In addition to being extracellular, our analysis also found that QnrA has a high antigenicity score. Therefore, a vaccine targeting QnrA is likely to stimulate a strong antibody response against this protein and thus reduce or eliminate the bacterial resistance to many drugs.
β-Lactamases are likely the most critical antibiotic resistance enzymes produced by bacteria to provide multi-resistance to various β-lactam antibiotics. Previous reports show that β-lactamases can be used as vaccine candidates to stimulate protective immunity against antibiotic-resistant pathogens [14,15,81]. Ciofu et al. showed that animals immunized with the AmpC β-lactamase protein could induce strong neutralizing antibody responses and demonstrated a synergistic effect with ceftazidime treatment of resistant P. aeruginosa in a rat model of chronic lung infection [14]. This effect could be explained partially by the inactivation of the enzymatic β-lactamase activity by specific antibodies against β-lactamase. Ciofu [15] also found that a rat with chronic lung infection immunized with purified chromosomal β-lactamase showed significantly lower bacterial load and reduced lung pathology compared to non-immunized rats. Zervosen et al. [81] generated a subunit recombinant TEM-1 β-lactamase by insertion of a heat-stable enterotoxin sequence at position 197 of the TEM-1, and immunization of cattle with this hybrid β-lactamase protein resulted in high levels of the anti-TEM IgG in cattle sera that inhibited β-lactamase activity. These results support the feasibility of developing a preventive or therapeutic β-lactamase vaccine that would induce neutralizing antibodies to inhibit the activity of β-lactamase.
Our RV study shows that not all β-lactamases are equally feasible for β-lactamase vaccine development. Specifically, Figure 2 clearly shows the sequence differences and phylogenetic variation among different β-lactamases. In addition, among 39 β-lactamases, we found that all four CTX-M type β-lactamases have the highest adhesin probabilities. CTX-Ms have been detected in at least 26 bacterial species and are widespread not only in humans but also in animals and environments. CTX-M-2, -5, -15 and -43 were reported in A. baumannii. In this study, A. baumannii CTX-M-2, -5, and -43 enzymes showed preferred adhesion ability >0.51. How these periplasmic β-lactamases possibly become adhesins is unclear. It has been reported that β-lactamases can be released through outer membrane vehicles (OMVs) [47,48]. It is possible that these CTX-M type β-lactamases are released through OMVs and function as bacterial adhesins. Adhesins are usually virulence factors supporting bacterial invasion. Rao et al. also showed that the presence of PER-1 (a class A β-lactamase) is critical to cell adherence [82]. Interestingly, our Vaxign predicted PER-1 as a cytoplasmic protein with an adhesin probability =0.378. Given previous reports of β-lactamases being feasible vaccine candidates [14,15,81] and our Vaxign analysis of the differences among β-lactamases, we would recommend that A. baumannii CTX-M-2, -5, and -43 enzymes be better β-lactamase vaccine candidates, which could potentially stimulate strong immune responses that not only reduce antibiotic resistance but also prevent MDR bacterial invasion and infection.
It is possible to develop a combinatorial vaccine that includes a classical type vaccine candidate(s) against bacterial infections and an antibiotic resistance determinant vaccine candidate(s) against drug resistance. Our RV study provides many candidates for such multicomponent vaccine development. The usage of resistance-conferring proteins as antigens in multicomponent vaccines would exert consistent selection against resistance [83]. Furthermore, many antibiotic resistance determinants identified in our RV analysis are also OMPs, extracellular proteins, and adhesins, which are likely to stimulate strong immune responses to serve not only therapeutic but also preventive purposes. These vaccine antigen candidates can be used in different vaccine types such as subunit protein vaccines, epitope peptide vaccines, DNA vaccines, or recombinant vector vaccines.
Reverse vaccinology emphasizes bioinformatics analysis, which is the focus of the current research. It is clear that many valuable results have been identified from our bioinformatics study. It is noted that many published experimental studies [14,22,46,73,81] have experimentally verified many of our predictions. Considering that we did not use these published results ahead of our bioinformatics predictions, these experimental evidences have provided some proof-of-concept evaluations for our analysis. More wet-lab verification studies will definitely be needed in our or others' future research in order to eventually develop effective and safe vaccines against A. baumannii infections, especially those caused by antibiotic-resistant A. baumannii strains.

Collection of A. baumannii Genome Sequences and Antibiotic Resistance Determinants
The information for the 33 completed and annotated A. baumannii genomes was retrieved from the NCBI database (Available online: http://www.ncbi.nlm.nih.gov/genome/). In addition, 84 A. baumannii antibiotic resistance proteins, which cause the resistance of A. baumannii strains to different classes of antimicrobial agents, were also retrieved from the ARDB-Antibiotic Resistance Genes Database (Available online: https://ardb.cbcb.umd.edu/) and NCBI Gene and Protein databases.

Vaxign Calculation of Sequence-Derived Features
Thirty-three completed and annotated A. baumannii genomes with their NCBI BioProject numbers were used for Vaxign dynamic analysis (Figure 1). Given the BioProject numbers, Vaxign automatically retrieved the protein sequences of the chromosome and any possible plasmid(s) of the A. baumannii genome from the BioProject database. For each protein, Vaxign calculated many features including subcellular localization, transmembrane domain prediction, adhesin probability, conservation among different A. baumannii strains and homology with human proteins. The programs used in the Vaxign calculation include PSORTb2.0 for subcellular localization prediction [84], TMHMM for transmembrane helix topology analysis [85], and SPAAN for adhesin probability calculation [43]. Vaxign optimizes these programs in a seamless integrative pipeline. After each genome was separately processed, Vaxign then implemented an OrthoMCL program to identify conserved sequences among genomes [86]. The default Blast E-value threshold of 10 −5 was set for OrthoMCL processing [27]. The homology of the antigen candidates to the host (in this study, only human was considered) was tested using Blastp with an E-value threshold of 10 −6 .

Vaxign Results Analysis
The final calculated results of the Vaxign pipeline execution were visualized using the Vaxign web program. The Vaxign filtering program was also used to identify proteins that met pre-defined criteria. The Vaxign results were also downloaded to an Excel file and separately analyzed. The Vaxign RV analysis results of the 33 A. baumannii genomes are available openly on the Vaxign web page (Available online: http://www.violinet.org/vaxign) for free exploration.

Antigenicity Prediction
Antigenicity scores of potential vaccine candidates were calculated by the VaxiJen v2.0 server. This software uses the z-descriptor composed of multiple physicochemical properties of proteins to predict their antigenicity from FASTA-submitted amino acid sequences using partial least squares discriminant analysis (DA-PLS). The antigens having values more than 0.4 were considered potentially antigenic as described by Doytchinova and Flower [44].

BLAST Analysis and COG Functional Annotation of Predicted Vaccine Candidates
Blastp (Available online: https://blast.ncbi.nlm.nih.gov/Blast.cgi) was used to find the similarity of the potential vaccine candidates to proteins of known function.

PFam Conserved Domain Analysis
The PFam database (Available online: http://pfam.xfam.org/) was used to search the conserved domains present in candidate proteins [88].

Conclusions
Using 33 complete A. baumannii genomes and 84 representative antibiotic resistance determinants, we applied the Vaxign reverse vaccinology pipeline and other bioinformatics methods and identified many vaccine antigen candidates for the rational development of vaccines against A. baumannii infections and antibiotic multi-drug resistance. Our study predicted 35 protective antigens that are conserved in all 33 genomes, have no homology with human proteins, have less than 2 transmembrane helices, have high adhesin probabilities, and exist in outer membrane or extracellular locations. These 35 antigens include 11 TonB dependent receptors, 8 porins, 7 efflux pump proteins, and two fimbrial proteins. By comparing the genome sequences of 3 antibiotic-sensitive strains and 30 antibiotic-resistant strains, we identified many vaccine candidates including hypothetic outer membrane adhesin CAM86003.1 and hypothetic adhesin CAM86739.1. By comparatively analyzing 84 resistance determinants and 33-genome sequences, we further identified many feasible anti-resistance vaccine candidates such as three RND type outer membrane efflux pump proteins, one extracellular protein (QnrA), and three CTX-M type β-lactamases. Out of 39 β-lactamases, our study showed that A. baumannii CTX-M-2, -5, and -43 enzymes are likely better β-lactamase vaccine candidates than other β-lactamases. To our knowledge, our study is the first to apply reverse vaccinology for systematically predicting vaccine candidates against antibiotic resistance of a microbial pathogen.