Putative vaccine candidates and drug targets identified by reverse vaccinology and subtractive genomics approaches to control Haemophilus ducreyi, the causative agent of chancroid

Abstract

Chancroid is a sexually transmitted infection (STI) caused by the Gram-negative bacterium Haemophilus ducreyi. The control of chancroid is difficult and the only current available treatment is antibiotic therapy; however, antibiotic resistance has been reported in endemic areas. Owing to recent outbreaks of STIs worldwide, it is important to keep searching for new treatment strategies and preventive measures. Here, we applied reverse vaccinology and subtractive genomic approaches for the in silico prediction of potential vaccine and drug targets against 28 strains of H. ducreyi. We identified 847 non-host homologous proteins, being 332 exposed/secreted/membrane and 515 cytoplasmic proteins. We also checked their essentiality, functionality and virulence. Altogether, we predicted 13 candidate vaccine targets and three drug targets, where two vaccines (A01_1275, ABC transporter substrate-binding protein; and A01_0690, Probable transmembrane protein) and three drug targets (A01_0698, Purine nucleoside phosphorylase; A01_0702, Transcription termination factor; and A01_0677, Fructose-bisphosphate aldolase class II) are harboured by pathogenicity islands. Finally, we applied a molecular docking approach to analyse each drug target and selected ZINC77257029, ZINC43552589 and ZINC67912117 as promising molecules with favourable interactions with the target active site residues. Altogether, the targets identified here may be used in future strategies to control chancroid worldwide.

1. Introduction

Chancroid is a sexually transmitted infection (STI) caused by the bacterium Haemophilus ducreyi. It is endemic in poor countries of Asia, Africa and Latin America [1], indicating that there is a close relationship between the social economic situation and the incidence of chancroid in a given population.

The World Health Organization estimated the global prevalence of the disease in around 7 million, in the 1990s. However, it is difficult to assess the current epidemiology of chancroid because of syndromic management of genital ulcer diseases and the lack of reporting and diagnostic tools [2–4].

Haemophilus ducreyi is a fastidious non-motile Gram-negative cocobacillus, negative catalase, D-glucose fermenter, has fine pili, and does not form endospores [5]. Moreover, it does not have a known animal or environmental reservoir and infects preferably the human mucosal epithelium, but it may also infect the keratinized stratified squamous epithelium causing ulcers, local pain and inguinal lymphadenopathy [6]. The bacterium remains in the extracellular environment along with polymorphonuclear leukocytes, macrophages, collagen and fibrin, where it is able to escape from phagocytes, which seems to be the main immune system evasion mechanism, an important factor in its pathogenesis. Furthermore, ulcerative lesions are associated with mononuclear cell infiltrates in dermis, consisting of a significant number of CD4⁺ lymphocytes, which may facilitate the co-infection mechanism by human immunodeficiency virus (HIV) [7].

Haemophilus ducreyi does not cause systemic infection, not even in HIV patients. However, extragenital lesions may occur, and they might result from auto-inoculation. Recent studies identified H. ducreyi causing non-sexually transmitted chronic cutaneous limb ulcerations in children and adults [8–10]. Despite the undefined relationship among genital and non-genital strains, genital strains are clustered into two classes (I and II), according to multiple phenotypes and genetic variations including the membrane protein DsrA, an important virulence factor. Based on proteomic analyses, H. ducreyi class I strains express the DsrA I form, whereas the class II lineages express the DsrA II form (e.g. [11]). Besides, during pathogenesis, H. ducreyi express other virulence factors, including lipooligosaccharides, pili, heat shock proteins, iron-regulated proteins or receptors, outer membrane proteins, toxins and other secreted products [12]. Most of the oligosaccharide structures from H. ducreyi have a lactosamine terminal, which binds to sialic acid. These lipooligosaccharides facilitate the adhesion of H. ducreyi to keratinocytes in vitro. Finally, H. ducreyi also express hemolysin that lyses erythrocytes, keratinocytes, fibroblasts, macrophages, and T and B cells [13].

Chancroid control is difficult because an H. ducreyi infection seems not to induce a protective immune response against subsequent infections, whereas a late hypersensitivity response against H. ducreyi may occur [14]. In fact, the lesions may persist for weeks or even months, and the skin may not recover completely without antimicrobial treatment [15]. This insufficient response might occur because a cell-mediated immunity may be only effective in eliminating intracellular bacteria while most part of H. ducreyi remains extracellular [16]. Even though it is not known which type of response is able to protect the organism against H. ducreyi infection, the fact that H. ducreyi is an extracellular bacterium may suggest the occurrence of a humoral immune response [14,17]. As far as we know, innate and acquired immune cells are recruited to the lesions, specially macrophages, dendritic cells, NK cells, polymorphonuclear leukocytes as well as memory and effector CD4⁺ and CD8⁺ T cells (revised by Janowicz et al. [18]). Despite this, it is still not entirely clear if Th1, Th2, Th17, Th9, Th22 and Treg immune response patterns are properly up- or down-modulated in patients with the different phenotypes of the disease. There is a need for additional in vitro and in vivo studies to definitively characterize the role of the cellular and humoral immune defences and to define the resistance or susceptibility of the patients affected with the disease.

In view of the population negligence in adopting prophylaxis methods and the lack of an effective vaccine, the only current treatment is the use of azithromycin or ceftriaxone (for pregnant) [19]. However, antibiotic resistance was notified in endemic areas [20]. More than that, studies with non-genital ulcers suggest the existence of azithromycin-resistant non-genital strains; or even the existence of environmental reservoir, or azithromycin usual dose resistance [21–23].

In this context, one of the aims of studying H. ducreyi pathogenesis is to investigate virulence factors that may be potential candidates for a vaccine development [24]. Owing to bioinformatics growth, the creation of a vaccine to prevent H. ducreyi outbreaks is closer to becoming possible (or feasible) [25–27], as well for other STIs, for example, the study performed by Jaiswal et al. [28] for the identification of putative vaccine and drug targets against syphilis.

Because of bacterial evolution and gain of new antibiotic resistance genes as a result of horizontal gene transfer (HGT) mechanism, it became necessary for the implementation of alternative strategies to fight infections [29]. The availability of pathogen and host genomes makes bioinformatics approaches more attractive [30,31]. Comparative and subtractive genomic approaches associated with an analysis of metabolic pathways efficiently contribute to the identification of non-host homologous pathogen essential proteins [32,33]. These pathogen essential and non-host homologous proteins are considered to be putative targets. When identified, these putative targets are a start point for drug and vaccine development studies. In the case of a vaccine, this target also needs to be able to elicit a proper and accurate adaptive immune response [34].

Along with the implementation of subtractive genomic approaches, reverse vaccinology involves the use of in silico steps to identify surface exposed or secreted immunogenic effective proteins which are virulence factors and are likely to bind MHC class I and II proteins for antigen presentation within the host [35,36]. There are some known reported features of effective vaccine candidate proteins which include: sub-cellular localization (the presence of signal peptides and transmembrane domains, for instance) and antigenic epitopes [37,38].

Despite the medical interest, there are only a few studies involving genomic analysis of H. ducreyi and most of them do not apply the reverse vaccinology and molecular docking approaches for vaccine or drug target prediction, opening new possibilities in the comparative genomic area.

2. Material and methods

2.1. Identification of data

The genome sequences of 28 H. ducreyi strains (table 1) were retrieved from GenBank database¹ available at the National Center for Biotechnology Information (NCBI).

Table 1.

General information about the 28 H. ducreyi strains used in this work.

strain	size (MB)	GC (%)	gene number	protein number	GenBank number	classification
Hd_35000HP	1.69	38.20	1668	1509	AE017143	genital class I
Hd_VAN1	1.66	38.10	1629	1468	CP015424	cutaneous
Hd_VAN2	1.58	37.90	1534	1374	CP015425	cutaneous
Hd_VAN3	1.66	38.10	1624	1463	CP015426	cutaneous
Hd_VAN4	1.67	38.10	1637	1475	CP015427	cutaneous
Hd_VAN5	1.66	38.10	1630	1470	CP015428	cutaneous
Hd_GHA1	1.62	37.90	1551	1389	CP015429	cutaneous
Hd_GHA2	1.63	37.90	1560	1377	CP015430	cutaneous
Hd_GHA3	1.73	38.20	1706	1540	CP015431	cutaneous
Hd_GHA5	1.73	38.20	1712	1538	CP015432	cutaneous
Hd_GHA8	1.76	38.20	1740	1567	CP015433	cutaneous
Hd_GHA9	1.77	38.20	1748	1573	CP015434	cutaneous
Hd_AUSPNG1	1.72	38.01	1695	1541	CM004377	cutaneous
Hd_ATCC 33940	1.57	37.80	1543	1383	FOJC00000000	—
Hd_CLU1	1.63	38.30	1595	1432	CP011218	cutaneous
Hd_CLU2	1.65	38.20	1610	1452	CP011219	cutaneous
Hd_CLU3	1.65	38.20	1611	1450	CP011220	cutaneous
Hd_CLU4	1.63	38.30	1590	1435	CP011221	cutaneous
Hd_CLU5	1.59	38.40	1557	1382	CP011227	cutaneous
Hd_GU1	1.66	38.30	1633	1455	CP011222	genital class I
Hd_GU2	1.76	38.70	1743	1555	CP011223	genital class I
Hd_GU3	1.70	3870	1671	1491	CP011224	genital class I
Hd_GU4	1.66	38.50	1640	1453	CP011225	genital class I
Hd_GU5	1.62	38.30	1574	1422	CP011226	genital class I
Hd_GU6	1.62	38.10	1566	1407	CP011228	genital class II
Hd_GU7	1.55	38.00	1504	1340	CP011229	genital class II
Hd_GU8	1.59	38.10	1539	1385	CP011230	genital class II
Hd_GU9	1.57	38.10	1522	1363	CP011231	genital class II

First, we applied the Rapid Annotation using Subsystem Technology (RAST) software [39] for reannotation of all 28 genomes. This tool homogenizes the genome annotations aiming to avoid unexpected results and incorrect interpretation of genes, which is a relevant step before the genome analyses.

2.2. Identification of intra-species conserved non-host homologous proteins

After applying RAST, we used global alignment tools to align all exported CoDing Sequences (CDS). For the orthology definition process, we used the orthoMCL software [40], using all-versus-all blast analyses with an E-value cutoff of 1 × 10⁻¹⁰ and the Markov Cluster Algorithm, commonly called MCL algorithm. We considered core genome the CDS shared by all 28 strains. To avoid autoimmunity, the drug and/or vaccine targets must be non-homologous to human proteins, so for the identification of non-host homologous target, we also used orthoMCL (E-value 1 × 10⁻¹⁰) to compare the core genome with the human genome.

2.3. Identification of pathogenicity islands

Genome plasticity is the genome dynamic property that involves DNA gain, loss or rearrangement. In this step, the genomic plasticity analyses focused on the identification of genomic islands, that are genome regions potentially obtained by HGT. The main features used to predict genomic islands are: deviations in genome signature (GC content and codon usage, for instance); the presence of transposases and high concentrations of virulence, resistance, metabolic and symbiotic factors; the presence of insertion sequences or flanking tRNA genes, size ranging from 6 to 200 KB; and/or the absence in non-pathogenic related organisms. For that, we used the Genomic Island Prediction Software (GIPSy) tool [41].

2.4. Reverse vaccinology approach for the prediction of putative Haemophilus ducreyi vaccine targets

For the prediction of vaccine targets, we used the subtractive genome approach adapted from Jaiswal et al. [28]. Firstly, we used the core genome, that consists of the pathogen essential genes, and then we performed BLASTp analyses and predicted the non-host homologous targets. From this non-host homologous conserved proteome, we predicted the subcellular localization of all proteins, using the SutfG+ software, which classifies the proteins in cytoplasmic, secreted, putative surface exposed (PSE) and membrane proteins according to the presence or absence of signal peptides, retention signals and transmembrane helices. After identifying the secreted proteins, we submitted this data to Vaxign tool [42] to analyse the proteins according to their adhesion probability (greater than 0.51) and MHC I and MHC II binding properties. We looked for cleavage sites and transmembrane helices using SignalP [43] and Transmembrane Helix prediction server, based on hidden Markov model (TMHMM) [44], respectively, and predicted some functional domains using InterProScan [45]. Finally, we analysed the putative targets regarding their presence in pathogenicity islands (PAIs).

2.5. High-throughput structural modelling

In this subsection, we analysed the cytoplasmic (SurfG+) non-host homologous conserved proteins present in the core genome and in the predicted pathogenicity islands (GIPSy). We submitted the multi-fasta files containing amino acid sequences of the cytoplasmic proteins to the MHOLline tool [46], that uses the HMMTOP, BLAST, BATS, MODELLER and PROCHECK software for the prediction of protein three-dimensional modelling.

2.6. Assessment of essential genes

Essential genes control major cellular functions of microbes, which makes essentiality check an important parameter for identification of potential targets [47]. To identify conserved essential targets of H. ducreyi, we submitted the set of core conserved proteins to the Database of Essential Genes (DEG) [48] for identity analyses. DEG contains all the essential genes currently available. Basic parameters used for BLAST against DEG were set as default including bit score of 100 and E-value cut-off of 1 × 10⁻⁴.

2.7. Ligand libraries, virtual screening and molecular docking analysis

Molecular docking was performed using Autodock Vina software [49] with a ligand library of ZINC Natural Products (NP) (11 203 compounds) [50]. The three-dimensional structures of all target proteins were examined and converted to the required PDBQT format using Auto Dock tool (ADT) and MGL Tools (v. 1.5.4) [51]. The ligand molecules were prepared by converting .MOL2 to .PDB file format using OpenBabel (v. 2.3.1) [52] program. After converting file in .PDB format, the Gasteiger atomic partial charges were assigned and converted all the ligand compounds to the PDBQT format by using prepare_ligand4.py script on terminal. The box size was calculated following the protocol outlined by the authors of Vina [49] for each target. Blind docking was performed for the identification of most effective binding site of these proteins. For virtual screening of all ligand compounds shell script vina_screen_local.sh was used. The top-ranked ligand molecules were identified by using a Python script vina_screen_get_top.py. The three-dimensional poses of docked molecules were analysed in Chimera [53] and for two-dimensional structure representation PoseView was used [54]. Molecular function (MF) and biological process (BP) for each protein were determined from UniProt [39]. KEGG was used for the biochemical pathway analysis [55], SurfG+ software was used for subcellular localization [56] and virulence was determined using GIPSy [41].

3. Results and discussion

The key steps for target identification, the methodologies used and the total number of proteins described in each step are summarized on the workflow in figure 1.

Figure 1.

Designed workflow with the methodologies used and the total number of proteins identified in each step. CDS, coding DNA sequence; PSE, putative surface exposed; PAIs, pathogenicity islands.

3.1. Identification of intra-species conserved non-host homologous proteins and pathogenicity islands

We compared 28 genomes of H. ducreyi strains (table 1), using H. ducreyi 35000HP as the reference strain to perform the orthoMCL [40] analysis. Coding DNA sequences shared by all strains are part of the core genome, which correspond to 1257 CDSs. Among these CDSs, considering the human genome as the host genome, we found that 847 are non-host homologous proteins.

In addition, we performed the prediction of genomic islands. The most closely related organisms to H. ducreyi are pathogenic either to human or to animals, or do not have a sequenced complete genome, which makes them non-applicable to the genomic islands prediction task. To avoid false negative results, we used two closely related non-pathogenic organisms to human to consolidate both results: Haemophilus somnus 129Pt strain, a bovine preputial isolate, and Manheimia haemolytica USMARC_2286 strain. For this approach, we applied the GIPSy software [41]. As we could conclude after performing MEGA analysis, and some studies have shown, instead of being closely related to other species from Haemophilus gender (composed by pathogenic organisms), H. ducreyi is closer related to other species from the Pasteurallaceae family. Other studies performed phylogenomic analysis of H. sommus 129Pt and M. haemolytica USMARC_2286, which are comensal organisms in animals, and also are part of the family Pasteurallaceae, showing close relationship to H. ducreyi [57–60].

After applying GIPSy analysis and predicting three pathogenicity islands, we plotted the results in figure 2, using BRIG software [61]. Pathogenicity island prediction is very important for the knowledge about the virulence factors encoded and their mobility and structure, in understanding the bacterial evolution and the interactions between pathogen and eukaryotic host cells [28], thus these PAI host virulence factors that may be desirable vaccine candidates and elicit an immune response.

Figure 2.

Pathogenicity islands (PAIs) prediction [41] of 28 H. ducreyi (Hd) strains using Haemophilus somnus 129Pt (Hs) and Mannheimia haemolytica USMARC_2286 (Mh) as non-pathogenic organisms. All the H. ducreyi strains, H. somnus and M. haemolytica were aligned using H. ducreyi 35000HP strain as a reference. From the inner to outer circle: Hd_ATCC33940, Hd_AUSPNG1, Hd_VAN1, Hd_VAN2, Hd_VAN3, Hd_VAN4, Hd_VAN5, Hd_GHA1, Hd_GHA2, Hd_GHA3, Hd_GHA5, Hd_GHA8, Hd_GHA9, Hd_CLU1, Hd_CLU2, Hd_CLU3, Hd_CLU4, Hd_CLU5, Hd_GU1, Hd_GU2, Hd_GU3, Hd_GU4, Hd_GU5, Hd_GU6, Hd_GU7, Hd_GU8, Hd_GU9, Hs_129PT and Mh_USMARC_2286. In the last ring, we plotted the PAI result: PAI1, PAI2 and PAI3. GC, guanine-cytosine; PAI, pathogenicity island; Hd, Haemophilus ducreyi; Hs, Haemophilus somnus; Mh, Mannheimia haemolytica.

The absence of part of PAIs in some strains might be related to the plasticity/infection pattern virulence genes, and multiple phenotypes and genetic variations that classify the strain of H. ducreyi in three groups: non-genital, genital class I and genital class II. This can be inferred by the lack of part of PAI 2 in genital class II strains (Hd_GU6, Hd_GU7, Hd_GU8 and Hd_GU9). PAIs are very unstable regions that may be acquired or even deleted through the course of time. Also, their absence may be related, in some cases, to the presence of draft genomes in the dataset used in these analyses. PAIs are important as they represent a class of GEIs that carry virulence genes. Therefore, high concentrations of the two following subsets of genes would be expected inside PAIs: (i) shared genes that are shown by the two or more, but not all, strains and (ii) singletons (strain specific) [62].

3.2. Prediction of candidate vaccine target for Haemophilus ducreyi

During the process of reverse vaccinology, we considered all genomic sequences from H. ducreyi. Then, we analysed the genes that are conserved among the different genomes, essential to the pathogen and non-host homologous. Also, we predicted the subcellular localization, selecting those that are secreted proteins, surface-exposed proteins and membrane proteins, because they are most probably to be antigenic proteins and can be promptly recognized by the immune system [32], analysing their MHC I and MHC II binding properties and adhesion probability greater than 0.51 and absence of similarity to mammalian proteins [42]. Last, virulence factors are better targets and are often encoded in PAIs [32]; aiming to find the best targets we investigated their association to PAIs. Therefore, proteins encoded by shared PAIs are appropriate candidates; however, this step does not exclude the targets from the step before. The presence in PAIs indicates that the proteins may be important virulence factors and should be considered first [62].

The protein subcellular localization was predicted using SurfG+ software [56]. We identified 847 gene products and 332 as putative surface-exposed proteins, secreted proteins, or membrane proteins (table 2). Then, we submitted these 332 proteins for analysis in Vaxign [42], looking for proteins with adhesion probabilities greater than 0.51. From those, we identified 31 proteins. Next, we analysed those 31 proteins against DEG and found that 13 (table 3) out of the 31 targets are strictly essential to the pathogen, considering an E-value cut-off of 1 × 10⁻⁴ [48], and two of them are present on PAI1 and PAI2 (A01_0690 and A01_1275, respectively).

Table 2.

Subcellular localization of H. ducreyi conserved non-host homologous proteins. PSE, putative surface exposed.

localization	no. proteins
cytoplasmic protein	515
membrane protein	159
PSE	117
secreted protein	56

Table 3.

Candidate vaccine target for H. ducreyi identified using Vaxign. TMHMM, transmembrane helix prediction server; PSE, putative surface exposed; SEC, secreted; MEM, membrane.

protein	protein ID	new locus tag	old locus tag	gene	localization	signal-P	TMHMM	InterProScan	product	adhesin probability	length (AA)	molecular weight (Da)
A01_0245	WP_010944348.1	HD_RS01300	HD0318	—	PSE	no	1	tetratricopeptide-like helicaldomain	putative fimbrial biogenesis and twitching motility protein PilF-like protein	0.550	181	20 238
A01_0449	WP_010944572.1	HD_RS02385	HD0588	lptC	PSE	no	1	LptC related	lipopolysaccharide export system protein LptC	0.673	193	21 798
A01_0540	WP_010944666.1	HD_RS02835	HD0697	—	PSE	no	1	thioredoxin-like fold/thioredoxin domain/alkylhydroperoxide reductase subunit C/thiolspecific antioxidant	thioredoxin-like protein	0.563	172	19 907
A01_0584	WP_010944716.1	HD_RS03070	HD0755	—	PSE	yes (between 20 and 21)	0	LysM domain/peptidase M23/duplicated hybrid motif	lipoprotein NlpD	0.541	372	40 279
A01_0690	WP_010944817.1	HD_RS03600	HD0881	—	PSE	no	2	Rossmann-like alpha/beta/alphas and wich fold/domain of unknown function DUF218	probable transmembrane protein	0.533	248	27 983
A01_1166	WP_010945328.1	HD_RS06065	HD1481	nlpc	PSE	yes (between 31 and 32)	0	endopeptidase, NLPC/P60 domain	probable lipoprotein nlpC precursor	0.519	71	19 295
A01_1607	WP_010945753.1	HD_RS08300	HD2025	—	SEC	yes (between 22 and 23)	0	TonB-dependent receptor, plug domain/TonB-dependent receptor, beta barrel	outer membrane receptor proteins, mostly Fe transport hgbA (haemoglobin and haemoglobin-haptoglobin-binding protein)	0.574	972	110 921
A01_1275	WP_010945431.1	HD_RS06605	HD1622	—	SEC	yes (between 22 and 23)	0	no	protein of unknown function DUF882	0.615	214	24 576
A01_0506	WP_010944628.1	HD_RS02660	HD0650	—	SEC	yes (between 22 and 23)	0	RuvA domain 2-like/competence protein ComEA, helix-hairpin-helix domain	DNA uptake protein and related DNA-binding proteins	0.551	119	12 905
A01_0636	WP_010944761.1	HD_RS03335	HD0812	—	SEC	yes (between 19 and 20)	0	solute-binding protein family3/N-terminal domain of MltF	arginine ABC transporter, periplasmic arginine-binding protein ArtI	0.615	239	26 261
A01_0935	WP_010945090.1	HD_RS04895	HD1191	bamA	SEC	yes (between 17 and 18)	0	POTRA domain/ POTRAdomain, BamA/TamA-like/bacterial surface antigen (D15)	outer membrane protein assembly factor YaeT precursor	0.598	793	88 661
A01_1304	WP_010945461.1	HD_RS06760	HD1655	—	SEC	no	0	TamA, POTRAdomain 1/POTRA domain, BamA/TamA-like/bacterial surface antigen (D15)	hypothetical protein	0.599	602	69 191
A01_0770	WP_010944920.1	HD_RS04080	HD0991	focA	MEM	no	6	aquaporin-like	formate efflux transporter (TC2.A.44 family)	0.542	274	29 917

Also, the relevant information for these 13 proteins was accessed, searching for cleavage sites and transmembrane helices using SiganlP [43] and TMHMM [44], respectively, and predicting some functional domains using InterProScan [45]. Proteins having molecular weight less than or equal to 110 kDa are considered to be more effective targets because they can easily be purified and subjected to vaccine development [63]. Molecular weights of targeted proteins were calculated using UniProt [39]. All predicted proteins are within this range, and thus may be effective vaccine targets.

The A01_1275 predicted protein is a secreted uncharacterized protein similar to an ABC transporter substrate-binding protein. The A01_0690 PSE predicted protein belongs to the Rossmann-like alpha/beta/alpha sandwich fold domain and to a domain of unknown function (DUF 218) and has a putative active site YdcF-like. The Rossmann-like threefold is a protein structural motif found in proteins that bind nucleotides, such as enzyme cofactors FAD, NAD+ and NADP+. The DUF 218 domain contains several highly conserved charged amino acids, suggesting this may be an enzymatic domain. This family includes SamA that is involved in vancomycin resistance. This protein may be involved in murein synthesis [64,65].

Most of the reported vaccine targets include surface exposed or secreted proteins which are more likely to be antigenic or virulent and thus are considered as suitable vaccine candidates [36]. Although we prioritize predicted proteins in PAIs, other PSE and secreted proteins are also potential vaccine targets, such as the lipoprotein NlpD protein (A01_0584) and the secreted arginine ABC transporter (A01_0636). A01_0584 belongs to the LysM domain that is a widespread protein module involved in binding peptidoglycan in bacteria and chitin in eukaryotes. The domain was originally identified in enzymes that degrade bacterial cell walls [66] and functions as a signal for specific plant-bacteria recognition in bacterial pathogenesis [67]. The predicted protein A01_0636 belongs to the solute-binding protein family 3/N-terminal domain of MltF, involved in active transport of solutes across the cytoplasmic membrane. Besides, it is important to mention the identification of other secreted protein (A01_0506), a DNA uptake protein and related DNA-binding protein, from the competence protein ComEA, helix-hairpin-helix domain, responsible for the ability of a cell to take up exogenous DNA from its environment, resulting in transformation. This process is regulated in response to cell–cell signalling and/or nutritional conditions and it is widespread among bacteria, which is probably an important mechanism for the horizontal transfer of genes [68].

Most part of the structure of an outer membrane protein is embedded within the membrane and may contain sequence variability of the protective epitopes located at the external loops, enough to escape immunity [32]. However, according to Fusco et al. [27], adherence of pathogenic bacteria to eukaryotic cells or to host extracellular matrix proteins is thought to be the first step in most infections. Blocking this interaction may, therefore, be an effective approach in preventing infection. So, as a preventative strategy, we included the identification of proteins involved in adherence, such as secreted outer membrane proteins A01_1304, A01_0935 and A01_0245. The first two proteins are part of the BamA family involved in assembly and insertion of beta-barrel proteins into the outer membrane [69], and A01_1304 also belongs to the bacterial surface antigen D15 domain. By contrast, A01_0245 is required for pilus stability and for pilus functions such as adherence to human cells [70,71].

DsrA was not predicted as part of the core genome, which may be accounted both to its deletion on some strains or to the great number of draft genomes used here. However, as we want to identify targets that may be used against any strain, we need to be very stringent in this case.

Previous in vitro and in vivo studies have identified H. ducreyi transcripts expressed during human infection and some of these proteins were described in our results. Bauer et al. [72] used selective capture of transcribed sequences with RNA isolated from pustules of three volunteers infected with H. ducreyi, and RNA isolated from broth-grown bacteria. They identified several genes encoded by the bacteria that have been characterized as potential virulence determinants, including the predicted outer membrane protein A01_1304 [72]. Another transcriptomic study, performed by Gangaiah et al. [73], determined the H. ducreyi transcriptome in biopsy specimens of human lesions and compared it to the transcriptome of bacteria grown to mid-log, transition and stationary phases. Compared to the inoculum (mid-log phase), H. ducreyi harvested from pustules had several upregulated genes involved in nutrient transport, anaerobiosis and fermentation in vivo, such as the predicted formate efflux transporter focA (A01_0770), and virulence determinants, including hgbA (A01_1607), that encodes a protein required for haemoglobin uptake. However, the study only determined the transcriptome at one-time point from the entire lesion. The expression of other virulence factors may be differentially expressed over time or from bacteria that occupy distinct microniches within the lesions [73].

During the last century, several conventional approaches have been successfully used for vaccine development requiring cultivation of the pathogen and using biochemical, immunological and microbiological methods [74]. However, this method is time-consuming, identifies only abundant antigens that may or may not provide immunity, and frequently fails when the pathogen cannot be cultivated under laboratory conditions. Revolution in the genomic era allowed the field of computational vaccinology to deliver best results for the design of vaccines starting from the prediction of all antigens in silico [75]. The development of a global vaccine against Neisseria meningitidis strains was the first application of immune-informatics in the field of vaccinology [74]. Afterwards, many efficient vaccines were introduced based on reverse vaccinology, such as vaccines against Listeria monocytogenes [76], Malarian protozoans [77], Streptococcus pneumoniae, Porphyromonas gingivalis, Chlamydia pneumonia, Staphylococcus aureus [74,75]. Thus comparative genomics, subtractive genomics and reverse vaccinology have been successfully applied for vaccine research.

3.3. High throughput structural modelling

Aiming to identify not only vaccine targets but also novel drug targets against H. ducreyi, we performed modelome prediction analyses. For that, the 515 non-host homologous cytoplasmic proteins (table 2), predicted using SurfG+ software [56], were submitted to MHOLline tool [46], which uses the HMMTOP, BLAST, BATS, MODELLER and PROCHECK software for the prediction of protein three-dimensional modelling. We took for consideration only the first two classifications (very high and high) from G2 model groups for further analyses. We found 51 proteins with very high quality and 135 proteins with high quality. To confirm that these core genome products on G2 are essential to H. ducreyi, we applied the DEG analysis using default values. The results show that 49 out of 51 very high-quality proteins and 114 out of 135 high-quality proteins are essential to H. ducreyi. Aiming to get more restrictive results, an E-value of 1 × 10⁻⁵⁰ was also applied, predicting 40 very high- and 80 high-quality essential proteins. Altogether, we predicted 120 non-host homologous essential cytoplasmic proteins.

Out of 120 proteins, we selected only cytoplasmic proteins that were located on PAI regions. Then, three proteins were identified as essential and also present on PAIs (A01_0698, A01_0702 and A01_0677), which were further submitted to molecular docking analyses.

3.4. Virtual screening and molecular docking

The lower energy score indicates better protein–ligand binding compared to high energy values [78]. In this work, for each target protein (FbaA-Fructose-bisphosphate aldolase class II, DeoD-Purin nucleoside phosphorylase and Rho, Transcription termination factor), ZINC Natural Product (11 203 compounds) library was used for docking analysis. All the compounds from this library were used one by one for the identification of the best-ranked molecules that showed favourable binding with the target. The biological importance of each target is described in table 4. The name of the molecules, Vina binding affinity for the identified molecules, number of predicted hydrogen bonds with interacting residues are shown for each target in table 5. The predicted configuration of the best-docked molecules is shown for each protein target in figures 3–5.

Table 4.

Drug target prioritization parameters and functional analysis of the protein targets.

protein gene and protein ID	new locus tag	old locus tag	official full name	molecular weight (Da)	functions	cellular component	pathways	virulence	essentiality
A01_0698, DeoD, WP_010944825.1	HD_RS03640	HD0889	Purine nucleoside phosphorylase, DeoD	25972.90	MF: Purine nucleoside phosphorylase activity BP: nucleoside metabolic process	cytoplasmic	metabolic pathway	yes	yes
A01_0702, Rho, WP_010944829.1	HD_RS03660	HD0895	Transcription termination factor, Rho	46924.16	MF: ATP binding, helicase activity, RNA binding, RNA-dependent ATPase activity BP: DNA-template transcription, regulation of transcription	cytoplasmic	RNA degradation	yes	yes
A01_0677, fbaA, WP_010944802.1	HD_RS03540	HD0864	Fructose-bisphosphate aldolase class II, fbaA	39248.52	MF: Fructose-bisphosphate aldolase activity, ZINC ion binding, BP: glycolytic process	cytoplasmic	carbon metabolism	yes	yes

Table 5.

Autodock Vina binding affinity of ZINC Natural Product and predicted hydrogen bonds for the selected best-ranked molecules against each protein as drug target.

drug targets	compounds		Autodock Vina binding affinity	H-bond	residues
A01_0677 Fructose-bisphosphate aldolaseclass II, fba	ZINC NP compounds	ZINC77257029ZINC67911804 ZINC67910940	−11, 1−10, 7−10, 7	642	HIS 226, GLY 179, LYS 230, ASN 224, THR 178, GLU 147, SER 146 VAL 225, GLY 63 PRO 231, GLY 223
A01_0698 Purin nucleoside phosphorylase, DeoD	ZINC NP compounds	ZINC43552589 ZINC67910930 ZINC67911626	−10, 2 −10, 3 −9, 9	2 7 3	ARG 217, ARG 87 GLY 20 SER 90, GLU 179
A01_0702 Rho, Transcription termination factor	ZINC NP compounds	ZINC67912117 ZINC67912113ZINC14615857	−10, 5 −9, 2−8, 2	4 64	TYR 254, ARG 276, LYS 231 THR 211, ALA209, VAL 207, ILE 208 ALA 209

Figure 3.

(a) Three-dimensional flat ribbon representation of docking analysis for the structure of A01_0677 (Fructose-bisphosphate aldolase class II) with compound CID-ZINC77257029. (b) Three-dimensional surface representation with 30% transparency, (c) three-dimensional surface representation of docking with the same compound and (d) two-dimensional representation of Fructose-bisphosphate aldolase class II with compound CID-ZINC77257029.

Figure 5.

(a) Three-dimensional flat ribbon representation of docking analysis for the structure of A01_0702 (Rho, Transcription termination factor) with compound CID-ZINC67912117. (b) Three-dimensional surface representation with 30% transparency, (c) three-dimensional surface representation of docking with the same compound and (d) two-dimensional representation of Rho, Transcription termination factor with compound CID-ZINC67912117.

Based on the structural comparison with a crystallography structure of the A01_0677 (fbaA), Fructose-bisphosphate aldolase template (PDB ID: 1dos, fructose 1,6-bisphosphate aldolase from Escherichia coli), the active site residues involved in hydrogen bond interactions with the ZINC metal ion are HIS110, GLU174, HIS226 and HIS264. One of these residues, HIS226, was predicted to make hydrogen bonds with the compound ZINC77257029 with a binding affinity of −11.1. A01_0677 (fbaA), Fructose-bisphosphate belongs to FBPases that are already reported as targets for the development of drugs for the treatment of non-insulin dependent diabetes and potential target for drug development against pathogenic bacteria, including H. ducreyi [26,79,80].

A01_0698 (DeoD, Purine nucleoside phosphorylase) is an important protein for the synthesis of nucleoside and deoxynucleoside catabolic, located on the linkage map of E. coli [81]. Escherichia coli Purin nucleoside phosphorylase along with its natural substrate creates an enzyme/prodrug combination for potential gene therapy. Purine nucleoside phosphorylase is responsible for the reversible catalysis of phosphorolysis of Purine nucleosides or 2′-deoxynucleosides to the free base and sugar phosphate. Escherichia coli Purin nucleoside phosphorylase also allows the activation of prodrugs that are not substrates for enzymes in humans. Systemic treatment with relatively non-toxic adenosine analogues, such as 9-(2-deoxy-β-d-ribofuranosyl)-6-methylpurine (MeP-dR), has resulted in the selective in vivo killing of ovarian, hepatoma, central nervous system and prostate tumours transfected with the E. coli gene Purin nucleoside phosphorylase [82]. Based on a comparison with a crystallographic structure of the A01_0698 (deoD), Purine nucleoside phosphorylase template (PDB ID: 1VHW, DeoD, Purine nucleoside phosphorylase), none of the active site residues were identified. The docking analysis was performed to identify the maximum number of hydrogen bond interactions and the number of residues interacting with the compounds from ZINC Natural Product library. The interaction of compound ZINC43552589 is shown in figure 4. The nucleotide biosyntheses are recognized as potential targets in antibacterial therapy. The protein DeoD, Purine nucleoside phosphorylase was reported as potential target for drug development against pathogenic bacteria Mycobacterium tuberculosis and Bacillus anthracis [83,84].

Figure 4.

(a) Three-dimensional flat ribbon representation of docking analysis for the structure of A01_0698 (Purin nucleoside phosphorylase) with compound CID-ZINC43552589. (b) Three-dimensional surface representation with 30% transparency, (c) three-dimensional surface with 30% transparency representation of docking with the same compound and (d) two-dimensional representation of Purine nucleoside phosphorylase with compound CID-ZINC43552589.

A01_0702 (Rho, Transcription termination factor) is a necessary protein for transcription termination in E. coli. This protein is essential for the viability of the cell and required for the factor-dependent transcription termination by RNA polymerase enzyme. It is a homo-hexameric protein that preferably binds to C-rich sites in the transcribed RNA. It uses the RNA-dependent ATPase activity and successively ATPase-dependent helicase activity to unwind RNA–DNA hybrids, eventually releasing the RNA from a transcribing elongation complex. Over the past few decades, studies have emphasized Rho as a molecule and have shown much of its mechanistic properties. The crystal structure solved in recent past could explain many of its physiological functions in terms of its structure. Though a lot is known about Rho, still various fundamental questions relating to Rho recognition sites, translocation of Rho along the nascent transcript, differential ATPase activity in response to different RNAs, interactions with elongation complex and finally unwinding and release of RNA remain unclear [85]. Bismuth-dithiol solutions have been shown to inhibit the Rho transcription termination factor selectively in E. coli [86]. In a comparison between the crystallographic structures of the Rho, Transcription termination factor template (PDB ID: 3ice, Rho, Transcription termination factor), none of the active site residues were identified. Our docking analysis predicted some of the residues (TYR 254, ARG 276, LYS 231, ALA337, THR211, ALA209, VAL207, ILE208, ALA209) showing the lowest binding affinity and making hydrogen bonds with the compounds of ZINC NP library. The binding mode of compound ZINC67912117 is shown in figure 5. According to Botella et al. [87] Rho, Transcription termination factor is a drug target against M. tuberculosis. They found that the inhibition of Rho may provide an alternative strategy to treat tuberculosis [87].

Fusacandin A (ZINC77257029) is a natural product, isolated from Fusarium sambucinum and it has an antifungal activity [88]. The drug molecule 2,3-dihydroamentoflavone (ZINC43552589) is isolated from the plant Cycas beddomei and it is a derivative of biflavonoids, amentoflavone and hinokiflavone. The biological activities of amentoflavone and its derivatives have inhibitory effects on lipid peroxidation and expression inhibition of Epstein–Barr virus genes and also possess antifungal, antibacterial, antiviral and anti-HIV activities [89]. And the drug edgeworside A (ZINC67912117) is isolated from the plant Edgeworthia chrysantha found in eastern Asia and its roots and flowers are used as therapeutics, which have antibacterial effects [90,91].

Among the drug-like molecules, ZINC77257029, ZINC43552589 and ZINC67912117 were predicted to bind the protein targets (Fructose-bisphosphate aldolase class II, fbaA, Fructose-bisphosphate aldolase class II, fbaA and Rho, Transcription termination factor) with good affinity. ZINC77257029, ZINC43552589 and ZINC67912117 were top-ranked molecules, however with much higher energy scores (less negative) than the top compounds from the ZINC Natural Products. Thus, the identification of molecules in our in silico study strengthens that identified molecules can be potentially used as new drugs for the treatment of chancroid disease.

4. Conclusion

Owing to the negligence in preventing worldwide STIs and the acquisition of antibiotic resistance, some infections have reemerged, such as syphilis, requiring new strategies for STI control. Likewise, there is a need for preventive measures against possible outbreaks of chancroid, which can be done by the creation of vaccines and drugs against H. ducreyi. Therefore, the current study successfully applied in silico reverse vaccinology and subtractive genomic approaches, using the genome repertory of 28 H. ducreyi strains and selecting the best putative vaccine and drug targets. We predicted two and three essential virulent non-host homologous vaccine and drug targets, respectively, applying a molecular docking approach to analyse each drug target. These results allow the development of therapeutic composition after further experimental validations.

Data accessibility

The genomes are available at GenBank as described in table 1.

Authors' contributions

A.S. carried out all the data analysis and participated in conceiving and designing the study and drafted the manuscript; A.K.J. participated in molecular docking analyses and helped draft the manuscript; S.T. participated in molecular docking analyses; L.C.O. participated in sequence alignments; D.B., V.A. and C.J.O. reviewed the manuscript and the data for confirmation of results and gave insights; S.C.S. conceived, designed and coordinated the study, and helped draft the manuscript. All the authors gave their final approval for publication.

Competing interests

We declare we have no competing interests.

Funding

This work was supported by grants from the Brazilian funding agencies Fundação de Amparo à Pesquisa de Minas Gerais (FAPEMIG; http://www.fapemig.br/) and Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES; http://www.capes.gov.br/).