Identification of conserved genomic signatures specific to Bifidobacterium species colonising the human gut

Abstract

Bifidobacterium species are known for their ability to inhabit various habitats and are often regarded as the first colonisers of the human gut. In the present work, we have used comparative genomics to identify conserved genomic signatures specific to Bifidobacterium species associated with the human gut. Our approach discovered five genomic signatures with varying lengths and confidence. Among the predicted five signatures, a 1790 bp multi-drug resistance (MDR) signature was found to be remarkably specific to only those species that can colonise the human gut. The signature codes for a membrane transport protein belonging to the major facilitator superfamily (MFS) generally involved in MDR. Phylogenetic analyses of the MDR signature suggest a lineage-specific evolution of the MDR signature in bifidobacteria colonising the human gut. Functional annotation led to the discovery of two conserved domains in the protein; a catalytic MFS domain involved in the efflux of drugs and toxins, and a regulatory cystathionine-β-synthase (CBS) domain that can interact with adenosyl-carriers. Molecular docking simulation performed with the modelled tertiary structure of the MDR signature revealed the putative functional role of the covalently linked domains. The MFS domain displayed a high affinity towards various protein synthesis inhibitor antibiotics and human bile acids, whereas the C-terminally linked CBS domain exhibited favourable binding with molecular structures of ATP and AMP. Therefore, we believe that the predicted signature represents a niche-specific survival trait involved in bile and antibiotic resistance, imparting an adaptive advantage to the Bifidobacterium species colonising the human gut.

Supplementary Information

The online version contains supplementary material available at 10.1007/s13205-023-03492-4.

Introduction

At the dawn of the last century, Henry Tissier of Pasteur Institute isolated a novel bacteria from the stools of healthy breastfed infants that, when ingested, can reduce the risk of acute diarrhoea in newborns (Tissier 1899; Lee and O’Sullivan 2010). Tissier described this novel microbe as gram-positive, curved, and bifurcated rod-shaped cells, later identified as Bifidobacterium bifidum. Ever since their discovery, the genus Bifidobacterium has been of immense importance. Members of this genus are often regarded as the first colonisers of the human gut (Turroni et al. 2022) and are believed to confer health-promoting benefits upon their host (Hidalgo-Cantabrana et al. 2017). The majority of these probiotic functions are exerted by modulating the immune, digestive and metabolic systems of the host (O’Callaghan and van Sinderen 2016) and through synthesising active compounds such as short-chain fatty acids (SCFAs), organic acids, and vitamins (Sarkar and Mandal 2016). Apart from the human gut, bifidobacteria have been identified in various ecological niches ranging from sewage to functional food products. However, they show a remarkable association with the gastrointestinal tracts (GIT) of social animals such as mammals, insects and birds, where offspring are dependent on parental care (Turroni et al. 2011; Hidalgo-Cantabrana et al. 2017). These associations are considered a result of their ability to colonise the host by vertical transmission from mother to offspring, a distinctive ecological feature of this genus that sets it apart from other gut commensals (Duranti et al. 2017).

According to the List of Prokaryotic names with Standing in Nomenclature (LPSN), there are 98 (sub)species with a validly published and correct name under this genus. At least nine out of these 98 (sub)species, B. bifidum, B. breve, B. longum, B. animalis, B. catenulatum, B. pseudocatenulatum, B. adolescentis, B. angulatum, and B. dentium have been found abundantly among the human intestinal microbiota (Arboleya et al. 2016). These species are thought to have coevolved with their hosts by gaining niche-specific adaptations that can enhance their survival in the harsh environment of human GIT. Some of these core survival traits include tolerance to acids and antibiotics, forming biofilm, transport of nutrients, and metabolism of a wide range of carbohydrates and amino acids (Sharma et al. 2018). A recent study reveals that many of these species have been associated with hominids over the past 15 million years, undergoing cospeciation within the respective host lineages across thousands of host generations (Moeller et al. 2016). During this course of time, each of these lineages would have developed unique genetic traits specific to the respective host environment they inhabit.

Studies that have attempted a comparative genomic analysis of the bifidobacteria have mainly focused on the core and pan-genome characteristics of this genus (Bottacini et al. 2010; Lugli et al. 2014; Sun et al. 2015). Most of these studies were intended to discover conserved functional traits that are assumed to be involved in the survival and probiotic characteristics. However, few studies have examined the host-specific adaptations common to bifidobacteria inhabiting the human gut. Some of the reported adaptive traits include glycolytic enzyme machinery for carbohydrate metabolism (Milani et al. 2016; Abdelhamid and El-Dougdoug 2021), sortase-dependent pili and extracellular sialidase for adhesion to the mucosal surface (Bottacini et al. 2014; Nishiyama et al. 2017), bile and antimicrobial resistance (Sarkar and Mandal 2016), and acid tolerance (Sánchez et al. 2007). Nevertheless, these studies do not confirm whether the identified host-specific conserved traits are unique to human gut colonising bifidobacterial species. Most of the reported conserved genetic traits are ubiquitously found in the genomes of other gut commensals, including species that are remotely related to bifidobacteria. As a result, there is a gap in the previously reported analyses towards understanding the host-specific genetic traits uniquely preserved in the bifidobacteria lineages associated with human GIT. The present study aims to bridge this gap by performing a comprehensive comparative analysis of different Bifidobacterium species regularly reported by various human gut microbiome studies. Our comparative genomic analysis was focused on finding conserved genomic signatures uniquely specific to the group of bifidobacteria under study. Further analyses were mostly focused on understanding the origin and functional role of the predicted signatures.

Methods

The detailed workflow of the analysis is given in Supplementary File 1.

Retrieval of genome sequences and detection of genomic signatures

Genome sequences used in this comparative study were divided into two distinct groups, i.e. an inclusion group containing the genome sequences of organisms of interest and an exclusion group consisting of genomes of a background group of organisms. For the inclusion group, the Unified catalogue of Human Gastrointestinal Genomes (UHGG) (Almeida et al. 2021) was queried to identify 20 different Bifidobacterium species reported by various human gut microbiome studies. In total, 363 bifidobacterial genomes were included in the inclusion group, out of which 351 completed or draft quality genomes were obtained from NCBI, and the remaining 12 were retrieved from UHGG. The exclusion group consists of genomes from 16 non-Bifidobacterium species resulting in 98 completed or draft genomes of members belonging to the Bifidobacteriaceae family (Supplementary Table 1). A complete list of genomes of organisms used and their assembly details are available in Supplementary Table 2. Signatures were predicted using Neptune with default parameters by comparing the genomes of the inclusion group against the exclusion group. Neptune is a kmer-based signature prediction tool which employs probabilistic models to discover genomic signatures (Marinier et al. 2017).

Identification of orthologous sequences and phylogenetic analysis

Orthologous sequences were identified by comparing the predicted signature sequence against protein sequences from 101 representative genomes of the Bifidobacteriaceae family using BLASTP (Altschul et al. 1990). One-to-one identical partners of predicted signatures in each genome were determined by choosing a single best hit alignment with more than 60% query coverage and 30% identity against the target proteins. Multiple sequence alignment of the coding sequences of these proteins was computed using the MUSCLE algorithm (Edgar 2004) implemented in the Seaview package (Gouy et al. 2021). The alignment was further trimmed using the trimAI package (Capella-Gutiérrez et al. 2009) to reduce the gap-enriched sites in the final alignment. Additionally, poorly aligned sequences were manually inspected and removed before the analyses. Based on the final alignment, a maximum-likelihood phylogenetic tree was generated using the IQ-TREE package (Minh et al. 2020) to infer the evolutionary relationship of these homologous sequences. Additional tree decorations and labelling were performed using the Evolview v3 program (Subramanian et al. 2019).

Codon-based positive selection

Different models implemented within the Codeml program of the PAML package (Yang 2007) were used to determine the effect of episodic evolution in the nucleotide sequences of the predicted signature. EasyCodeML (Gao et al. 2019), a wrapper tool for the codeml package, was used to run both Branch models and Branch Site Models, which can identify signs of positive selection by computing the differences in ω value (ratio of non-synonymous to synonymous substitution rates) across branches, and branches and sites of a phylogenetic tree.

In the Branch models, the log-likelihood ratios of the one ratio model, where all the branches have a fixed ω value, is compared against the two ratio model, which fixes different ω values for the foreground branch having the signature sequences and the background branches consisting of all the other sequences. The Branch Site Models estimate heterogeneous values across different branches and sites. The log-likelihood ratio of branch-site model A which estimates ω > 1 along specified branches, was compared against Model A null which allows only neutral evolution and purifying selection where ω = 1 or ω < 1, respectively.

Domain architecture and functional prediction

NCBI Conserved Domain Database (CDD) (Lu et al. 2020) and InterPro server (Blum et al. 2021) were used to identify the various domains present in the protein sequence of the signature and to predict its function. Putative substrate binding sites of the domains were inferred from NCBI CDD. The Transporter Substrate Specificity Prediction (TrSSP) server was used to determine the substrate level specificity of the signature protein (Mishra et al. 2014). Support vector machine (SVM)-based computational models utilising amino acid index and position-specific scoring matrix (PSSM) profiles were used to predict the substrate level of signature for seven different transporter classes (amino acid, anion, cation, electron, protein/mRNA, sugar, and other transporters). The predicted signature was also inspected using a model that discriminates transporters from non-transporters.

Modelling of tertiary structure and molecular docking.

The amino acid sequence of the signature from B. bifidum (type species of Bifidobacterium genus) was used to predict the tertiary structure of the protein using AlphaFold2 (Jumper et al. 2021) algorithm implemented in the ColabFold platform (Mirdita et al. 2022). The modelled tertiary structure was then used to perform targeted docking of selected substrates against the different regions of the signature protein using Autodock Vina (Trott and Olson 2010). The docking operations were iterated for 50 independent runs for the same ligand. Among the different ligand binding models predicted, the most stable model was selected based on examining the binding affinity and various molecular interactions present in the complex. Additionally, the docked structures were analysed using PLIP (Protein–Ligand Interaction Profiler) to examine the various non-covalent interactions such as hydrogen bonds, hydrophobic interactions, salt bridges and other molecular forces present in the protein–ligand complexes (Salentin et al. 2015).

Results

Genomic signature prediction

Genomic signature in this work refers to a genomic locus or a stretch of nucleotide sequence particularly present in the genomes of a target (inclusion) group but sufficiently absent in the non-target (exclusion) group (Marinier et al. 2017). Since the members of the Bifidobacteriaceae family are believed to have evolved from a common ancestor, they share a pool of ancestral sequences common to themselves and other organisms. In addition, particular sequences might also be present in these species that are uniquely evolved in specific groups characterised by the environment they inhabit. Our approach was to discover these distinctively evolved genomic signatures and eliminate any shared sequences common to other organisms, including their adjacent taxonomic neighbours. Neptune predicted five such unique genomic signatures with varying lengths and confidence (Supplementary Table 3).

We investigated these sequences by mapping them against the reference proteins of different Bifidobacterium species to infer sequence annotations. Among the predicted signatures, a 1790 bp sequence annotated as MDR family MFS transporter protein (Signature 1), hereafter referred to as the MDR signature, had the highest score with nearly 78% in silico sensitivity and specificity concerning the inclusion and exclusion groups used for comparison. In addition, two hypothetical proteins (Signature 2 and 4), a multidrug resistance protein B (Signature 3) and a CBS domain-containing protein (Signature 5), were also predicted. Mapping the signatures against the complete genomes of species from the inclusion group revealed co-localisation of Signatures 1, 2, 3, and 5 in a single genomic locus of bifidobacteria having all five signatures (Fig. 1). Signature 4 was usually located in distant regions upstream from the rest of the signatures. The mapped genomic location of signatures in the majority of the genomes followed a similar trend. However, in some cases, the orientation of signatures showed variation between genomes (Supplementary File 2). Due to the lack of identical proteins with known biochemical functions, the sequences annotated as hypothetical proteins (Signature 2 and 4) could not be further analysed. Additionally, we could observe a certain level of overlapping and significant similarity of Signature 3 and Signature 5 with different regions of the MDR signature (Signature 1); therefore, they were also excluded from the downstream analyses.

Fig. 1.

Image of the mapped genomic location and orientation of the predicted five conserved genomic signatures in the complete genome of Bifidobacterium adolescentis (NZ_CP028341.1). This genome is used as a representative of the species from the inclusion group. Sig 1: MDR family MFS transporter, Sig 2: Hypothetical protein, Sig 3: Multidrug resistance protein B, Sig 4: Hypothetical protein, Sig 5: CBS domain-containing protein

The MDR signature (signature with the highest score) was initially compared with the representative genomes of different bifidobacteria species to examine its sensitivity. We detected genomic sequences highly identical to the MDR signature in eight species of Bifidobacterium typically found to be associated with the human gastrointestinal tract. The representative organisms having the best match for MDR signature in their genome are listed in Table 1; the complete result of the top 100 hits are available in Supplementary Table 4. In addition, our signature was found to cover all the seven “Human-Residential-Bifidobacterium” (HRB) (sub)species (B. adolescentis, B. breve, B. bifidum, B. catenulatum, B. longum subsp. infantis, B. longum subsp. longum and B. pseudocatenulatum), reported as the natural inhabitants of the human gut by Wong et al. (2020). Previous studies have isolated and identified B. angulatum from the human intestinal tract (Arboleya et al. 2016; O’Callaghan and van Sinderen 2016) and B. ruminantium from rumens of cattle (Biavati and Mattarelli 1991), but the human gut residential status of these species remains unclear. Comparing the MDR signature with genomes of different Bifidobacterium species, we found that the sequence exhibits high sensitivity towards members of bifidobacteria colonising the human gut.

Table 1.

Top-scoring BLASTn hits of the MDR signature in different Bifidobacterium representative genomes

Scientific name	Percentage identity	Query coverage	E value
Bifidobacterium longum subsp. infantis 157F	98.99	100%	0
Bifidobacterium ruminantium	98.88	100%	0
Bifidobacterium catenulatum DSM 16,992 = JCM 1194 = LMG 11,043	98.77	100%	0
Bifidobacterium bifidum NCIMB 41,171	98.6	100%	0
Bifidobacterium breve JCM 7017	98.49	100%	0
Bifidobacterium pseudocatenulatum	98.49	100%	0
Bifidobacterium angulatum DSM 20,098 = JCM 7096	98.38	100%	0
Bifidobacterium adolescentis	98.21	100%	0

On the other hand, we assessed the specificity of the signature by querying the sequence against the NCBI nt database, excluding the Bifidobacterium genus, using BLASTN. The result had multiple unspecific hits from a metagenomic assembly declared as ‘uncultured bacterium’ and a precise match (98.2% percentage identity and 100% query coverage) with one of the coding sequences from the partial genome of a Siphoviridae sp. isolate (Supplementary Table 5). Siphoviridae represents a family of tailed double-stranded DNA bacteriophages belonging to the order Caudovirales. Most of the accessible metagenomic reads from human gut viral genomic sequences belong to these phages, making them one of the dominant constituents of the human gut virome or phageome (Manrique et al. 2017; Fitzgerald et al. 2021). There are also reports of observing viral particles exhibiting morphotypes consistent with Siphoviridae phages in the culture supernatants of bifidobacterial strains (Mavrich et al. 2018). Therefore, the origin of the signature sequence found in the phage genome must be verified as it can be from a bifidobacterial host or any other member of the human gut microbial population. PHASTER and Islandviewer (Arndt et al. 2016; Bertelli et al. 2017) were used to scan the genome of B. bifidum to explore the possibility of the predicted signature being a prophage insertion element or a horizontally transferred gene originating from a different organism. PHASTER predicted three phage regions, and Islandviewer predicted 17 genomic islands but none of these regions had the predicted signature (Supplementary Table 6). Alternatively, the signature sequence from the phage genome was compared against the NCBI nt database to resolve its possible origin. The phage-derived sequence was highly homologous to the MDR signature from B. adolescentis (100% query coverage and 99.52% percentage identity), suggesting a bifidogenic origin of the sequence in the phage genome, which might have integrated during the induction phase of an early bacteriophage infection cycle involving B. adolescentis as host.

Functional annotations of the predicted signature

Functional characterization and domain architecture prediction of the signature were made using the NCBI subfamily protein architecture labelling engine (SPARCLE) implemented in the CDD database by querying protein sequences of signature from B. bifidum (WP_017143680.1). The following conserved domains were predicted for the signature protein; an MFS_LmrB_MDR_like domain (CDD:341,046) (35–471) belonging to the Bacillus subtilis lincomycin resistance protein (LmrB) and similar multidrug resistance (MDR) transporters of the Major Facilitator Superfamily, and two tandem repeats of the cystathionine-β-synthase (CBS pair) domains (CDD:417,695) (533–666). InterPro's domain prediction was also consistent with the CDD results, where two functional domains were predicted for the signature protein; a major facilitator superfamily domain (IPR020846) (27–476) and the CBS domain (IPR000644) (530–680) (Supplementary File 3).

Since the predicted domain architecture of the identified genomic signature points to the putative transporter activity of the protein, we evaluated its transporter role and substrate specificity using the TrSSP server. The TrSSP server calculated substrate specificity scores for each of the seven transporter classes and one for the non-transporter using the signature sequence's evolutionary (PSSM) and amino acid composition profiles. The signature was identified as a transporter having specificity for substrates other than the following classes: amino acid, anion, cation, electron, protein/mRNA, and sugar.

In a previous study by Gueimonde et al. (2009), two putative homologous MDR genes BL0920 and Bbr0838 were characterised from the genomes of B. longum subsp. longum and B. breve respectively. Both the proteins were found to harbour a permease domain belonging to MFS transporters and a catalytic CBS domain at the C-terminal end. Functional characterisation of these proteins revealed their bile inducible expression and putative role in bile tolerance (Gueimonde et al. 2009; Ruiz et al. 2012). The predicted MDR signature protein is remarkably homologous to both BL0920 and Bbr0838 proteins with more than 96% identity and 100% query coverage. Therefore, it is already established that the predicted MDR signature sequences can be associated with the efflux of bile acids and confer bile resistance to the microbes harbouring it. The MDR family of MFS transporters are primarily involved in the secondary transport of a broad spectrum of drugs and toxins from the microbial cells (Kumar et al. 2020; Drew et al. 2021). Since the signature protein's MFS domain is identical to the lincomycin resistance protein and has already been characterised to be involved in bile efflux, this region of the MDR protein might be involved in the efflux of toxic compounds such as antibiotics and bile acids. Although the role of CBS domain attached to the C-terminal end of the MFS transporter region has not yet been investigated, it has been previously shown to regulate the activity of linked protein by interacting with adenosyl carrying ligands such as AMP, ATP and S-AdoMet (Scott et al. 2004).

Phylogenetic analysis and positive selection

Orthologous sequences of MDR signature were identified in the genomes of 91 members of the Bifidobacteriaceae family. All the identified orthologues are found to be transporters of the major facilitator superfamily (MFS) involved in the efflux of multiple drugs. The maximum likelihood phylogenetic tree constructed using the multiple sequence alignment of these orthologue nucleotide sequences revealed a distinct sub-cluster (marked in red) formed by signatures from 6 HRB species and B. angulatum, making this cluster highly specific to the human gut bifidobacterial species (Fig. 2). Within the signature group, the sequence from B. pseudocatenulatum forms a discrete branch from the other signatures. MDR sequences of the other six species have a common ancestry, but they have later diverged into two separate lineages.

Fig. 2.

Maximum likelihood tree of 91 orthologous sequences identified from the representative genomes of various organisms under the Bifidobacteriaceae family. Leaf backgrounds are coloured based on the isolation source. The inner strip represents the various domains present in the sequence. The outer strip denotes the BLAST percentage identity of each sequence with the MDR signature. Internal nodes involved in the divergence and domain fusion events are labelled as A, B and C

The sequences that form the deeper clade originating from the common node (node A, Fig. 2) shared by MDR signatures and the other 15 transporters are derived from various Bifidobacterium species inhabiting the gastrointestinal tracts of different primate groups. Consequently, the ancestral node A can be considered an ancient MFS transporter belonging to the members of bifidobacteria inhabiting the gut of different primate species. An initial divergence due to changes in the MFS domain at the ancestral node A is evident due to the formation of two distinct clusters (internal nodes B and C) of the MFS proteins. These changes in the MFS domain have ultimately resulted in the low sequence homology observed between the MDR signature (clade C) and the proteins of clade B (outer strip Fig. 2). However, within the clade C, the multidrug efflux protein of B. samirii has a certain level of homology with the MFS domain of the MDR signature protein. It is important to note that within the clade C, the multidrug efflux protein of B. samirii mainly differs from the MDR signature in terms of the absence of the CBS domain in the former (Supplementary File 4; inner strip Fig. 2).

Similarly, in clade B, the MFS protein of B. lemurum differs from the rest of the proteins of this clade in terms of the presence of the CBS domain in the former (inner strip Fig. 2). This indicates that after the initial divergence, the ancestral nodes B and C would have been subjected to a domain fusion event involving MFS and CBS domains leading to a second round of divergence. As a result, some of the lineages have a C-terminally linked CBS domain, such as the MFS transporter of B. lemurum in clade B and MDR signatures in clade C, and the remaining with the default organisation of having only a single MFS domain, such as in B. samirii in clade C and the other 13 proteins in clade B (Fig. 2).

The likelihood ratio test (LRT) obtained from the branch models indicates a significant difference (p < 0.01) in the ω value among branches of the signature group and the rest of the tree (Supplementary Table 7). The evolutionary constraints acting on the branches of signatures after their divergence appears to be different from the rest of the branches in the tree. The likelihood of two branch-site models was compared to understand the nature of selection happening on the sequences of the signature group. Branch site analysis revealed that the branches and particular sites of the signature group are undergoing strong positive selection with an ω value = 49.63 (Supplementary Table 8). Twenty codon positions were identified from the sequences of the signature group as potential sites under positive selection by the Bayesian Empirical Bayes (BEB) method (Supplementary Table 7 and 8). Six of these sites have more than 90% posterior probability, and 2 sites have more than 95% posterior probability of being positively selected. Sites with a higher probability for positive selection were located primarily on the α-helices and inter-helical loops adjacent to the transmembrane helices of the MDR signature protein (Supplementary File 5).

Tertiary structure modelling

The modelled protein structure has two topologically distinct regions that correspond with the two functional domains present in the protein, namely, the MFS and CBS domains (Fig. 3A). The overall 3D protein model consists of α-helix (71%), β-strand (5%) and coil (24%) regions. The MFS domain region of the protein is enriched with bundles of α-helical structures, whereas the CBS domain region of the protein consists of both α-helical and β-strand structures.

Fig. 3.

Overview of the modelled tertiary structure of MDR signature protein. A Predicted protein model with two topologically distinct domains, namely, MFS (blue) and CBS (red) domain, B Arrangement of the MDR protein between inner (blue) and outer (red) layers of a lipid membrane

The majority of the α-helical structures in the MFS domain of the proteins are, in fact, transmembrane helical segments. Therefore, to identify these segments and to understand their spatial arrangement with respect to a lipid bilayer membrane, we analysed the modelled MDR protein structure using the MEMEMBED (Nugent and Jones 2013) tool hosted in the PSIPRED webserver (McGuffin et al. 2000). The MEMEMBED tool positioned 14 transmembrane (TM) segments between the two bi-lipid membrane layers (Fig. 3B). The TM segments consist of the following residues; TM1 (28–48), TM2 (63–81), TM3 (93–112), TM4 (114–136), TM5 (152–170), TM6 (181–200), TM7 (217–235), TM8 (238–259), TM9 (281–299), TM10-11 (320–359), TM12 (377–395), TM13 (408–428) and TM14 (456–474). The canonical structure of the MFS fold involves 12 transmembrane helical segments arranged into two bundles of six helices, connected to each other by a long cytoplasmic loop extending from the TM6 to TM7 (Drew et al. 2021). However, some members of the MFS protein family are found to deviate from this characteristic structure by having 14 TM segments with additional two transmembrane helices inserted between TM6 and TM7. The modelled protein seems to follow a similar trend of having 14 transmembrane helices similar to that of MFS proteins from drug:H + antiporter 2 (DHA2) families (TC number 2.A.1.3) and peptide transporter PepTSt. The MFS domain is linked to the CBS domain by a long alpha helix structure extending from the inner leaflet of the bi-lipid membrane to the cytoplasm of the cell.

The structural organisation of the CBS domain is remarkably different from the MFS domain by having three copies of tandem pair of anti-parallel β-sheets flanked by short α-helical structures. The canonical conserved structure of the CBS domain is of a β-α-β-β-α secondary structure configuration (Baykov et al. 2011). However, our modelled protein structure appears to be either missing the first β-strand or this region is not clearly predicted, resulting in an α-β-β-α arrangement. Several such cases have been reported where the first β-strand is either not resolved properly as a β-sheet or cannot be predicted reliably (Ignoul and Eggermont 2005). The CBS domain pairs 1 and 3 (N–C) are arranged pseudo-dimerically through their β-sheets, forming a cleft at their centre that can serve as active ligand-binding regions.

The characteristic structure of the MFS fold was first described from the crystal structures of lactose symporter LacY (Abramson et al. 2003) and the glycerol-3-phosphate antiporter GlpT of Escherichia coli (Huang et al. 2003). To understand the structural homology shared by the MDR signature with other MFS proteins, we compared the modelled MDR structure with the crystallographically resolved reference structures of LacY and GlpT available in the PDB database. The alignment of these structures revealed a high structural resemblance between the MDR signature and the two reference MFS proteins. When aligned with LacY, the MDR signature gave an RMSD value of 15.86 A⁰ across all 391 pairs and with GlpT, it was 24.08 A° across all 412 pairs. However, when the amino acid sequences of these proteins were compared, we could not observe any significant similarities between the MDR signature and the other two proteins.

Molecular docking

The generated 3D model of the MDR signature was used for molecular docking simulations to understand the putative role of the two functional domains present in the protein. From the initial predictions, it was evident that the MDR signature is involved in transporting drug molecules or toxins out of the microbial cell. Therefore, in the first set of docking experiments, we examined the interaction of various protein synthesis inhibitor antibiotics and human bile acids with the MFS region of the protein. Molecular structures of eight such antibiotics and eight principal human bile acids were downloaded from PubChem and docked against the MFS region of the protein (Fig. 4 and Fig. 5). All the docked complexes were found to exhibit favourable binding affinity (ΔG < 0 kcal/mol), indicating their potential for interaction with the transmembrane helices of the protein (Tables 2 and 3). The amino acid residues 133 PRO, 298 ASN, 384 GLN, 392 ASN, 405 ALA, 408 ASN and 412 ASN are found to be frequently involved in various molecular interactions with the ligand molecule (Figs. 4 and 5). Similarly, TM4, TM9, TM12 and TM13 of the MFS domain seem to mediate the binding and translocation of various substrates.

Fig. 4.

Overview of the docking simulations performed with molecular structures of antibiotics against the MFS domain-containing region of the predicted protein model. A Amikacin, B Chloramphenicol, C Clindamycin, D Gentamicin, E Kanamycin, F Lincomycin, G Streptomycin, and H Tetracycline. Protein structures are represented as blue sticks and ligand structures as orange sticks. The hydrophobic interactions are represented by grey dashed lines and hydrogen bonds by blue solid lines

Fig. 5.

Overview of the docking simulations performed with molecular structures of human bile acids against the MFS domain-containing region of the predicted protein model. A Chenodeoxycholic acid, B Cholic acid, C Deoxycholic acid, D Glycochenodeoxycholic acid, E Glycocholic acid, F Lithocholic acid, G Taurochenodeoxycholic acid, and H Taurocholic acid. Protein structures are represented as blue sticks and ligand structures as orange sticks. The hydrophobic interactions are represented by grey dashed lines, hydrogen bonds by blue solid lines and salt bridges by yellow dashed lines

Table 2.

The estimated binding affinity of various antibiotics with the MFS domain of the signature protein model

Ligand	*ΔG (kcal/mol)	#Hydrophobic interactions (n)	#Hydrogen bonds (n)	Antibiotic sensitivity from literature (Charteris et al. 1998; Mättö et al. 2007)
Amikacin	– 7.45 ± 0.14	2	12	Resistant
Gentamicin	– 7.64 ± 0.11	1	10	Resistant
Kanamycin	– 7.02 ± 0.28	2	12	Resistant
Streptomycin	– 8.13 ± 0.09	1	12	Resistant
Tetracycline	– 8.76 ± 0.17	2	8	Moderately susceptible
Chloramphenicol	– 6.70 ± 0.01	4	4	Susceptible
Clindamycin	– 7.10 ± 0.14	4	5	Susceptible
Lincomycin	– 7.13 ± 0.06	4	5	Susceptible

The number of various molecular interactions was predicted using the protein–ligand interaction profiler (PLIP) server

^*Average ΔG ± standard deviation obtained from 50 docking iterations

^#Reported from the docked complex having the best number of possible interactions among the 50 docking iterations

Table 3.

The estimated binding affinity of various human bile acids with the MFS domain of the signature protein model

Ligand	*ΔG (kcal/mol)	#Hydrophobic interactions (n)	#Hydrogen bonds (n)	#Salt bridges (n)
Chenodeoxycholic acid	– 7.80 ± 0.11	6	4	0
Cholic acid	– 7.92 ± 0.10	4	5	0
Deoxycholic acid	– 8.00 ± 0.11	6	5	1
Glycochenodeoxycholic acid	– 8.57 ± 0.12	7	5	1
Glycocholic acid	– 8.35 ± 0.23	6	6	1
Lithocholic acid	– 7.98 ± 0.14	6	3	1
Taurochenodeoxycholic acid	– 8.51 ± 0.19	6	7	1
Taurocholic acid	– 8.39 ± 0.28	4	8	1

The number of various molecular interactions was predicted using the protein–ligand interaction profiler (PLIP) server

^*Average ΔG ± standard deviation obtained from 50 docking iterations

^#Reported from the docked complex having the best number of possible interactions among the 50 docking iterations

Among the different antibiotics, tetracycline and streptomycin seem to have a high affinity toward the MFS region of the protein as their binding complexes have relatively lower binding energy. Variation in the number of predicted intermolecular forces, such as hydrogen bond and hydrophobic interactions, could be observed concerning the antibiotic involved in the complex. In general, amikacin and kanamycin exhibited a high number of molecular interactions with the MDR transporter compared to the rest of the antibiotics. Furthermore, we compared our findings from docking simulations with previously reported antibiotic susceptibility profiles of different strains of bifidobacteria (Table 2). We could observe a substantial correlation between the antibiotic sensitivity of bifidobacteria and the number of intermolecular hydrogen bonds. Generally, the docking complexes involving antibiotics to which the bifidobacterial strains are reported to be resistant have a higher number of hydrogen bonds (n ≥ 10). On the other hand, the other antibiotics to which Bifidobacterium strains are susceptible showed relatively fewer hydrogen bonds (n ≤ 8) in their complexes. The complex formed by tetracycline, to which Bifidobacterium strains are moderately susceptible, had the lowest ΔG value ( – 8.76 ± 0.17 kcal/mol) of all the antibiotics compared. Based on these results, we believe that the MDR protein might be directly involved in the efflux of antibiotics such as amikacin, streptomycin, kanamycin, and gentamicin, where intermolecular hydrogen bonds play a vital role in mediating the substrate binding and subsequent translocation activity of the protein. The eight bile acid structures used for docking also showed high affinity towards the transmembrane domain of the protein (Fig. 5 and Table 3). All the docking complexes had ΔG values ≤  – 7.8 kcal/mol, indicating a relatively stable protein–ligand binding interaction. Docking complexes involving bile acids has a higher number of hydrophobic interactions compared to that of antibiotics. Except for chenodeoxycholic acid and cholic acid, all the other bile acid structures formed a single salt bridge with the MFS domain of the signature protein.

Additionally, we also performed docking simulations targeting the CBS domain of the protein (Fig. 6 and Table 4). Based on the evidence sourced from various literature, we examined the potential of the CBS domain of MDR signature in interacting with adenosyl carrying molecules by docking it with the structures of adenosine monophosphate (AMP) and adenosine triphosphate (ATP) molecules. Favourable binding energies were predicted for the complex formed by both these ligands. The binding of ATP to the CBS domain has relatively more hydrogen bonds than AMP (Table 4), but a single salt bridge formation mediated by 649 LYS is consistently found in both complexes (Fig. 6). Additionally, pi-stacking interactions are also predicted in both the complexes contributing to more stability of the adenosyl-MDR complex. Therefore, we predict that the AMP and ATP molecules can potentially interact with the CBS domain of MDR protein through the formation of intermolecular interactions such as hydrogen bonds, salt bridges and pi-stacking interactions.

Fig. 6.

Overview of the docking simulations performed with molecular structures of AMP and ATP against the CBS domain of the predicted protein model. Protein structures are represented as blue sticks and ligand structures as orange sticks. The hydrogen bonds are represented by blue solid lines, salt bridges by yellow dashed lines, and pi stacking interactions by green dashed lines

Table 4.

The estimated binding affinity of various adenosyl carriers with the CBS domain of the signature protein model

Ligand	*ΔG (kcal/mol)	#Hydrogen bonds (n)	#Salt bridges (n)	#Pi-stacks (n)
Adenosine monophosphate (AMP)	– 6.96 ± 0.10	7	1	2
Adenosine triphosphate (ATP)	– 7.08 ± 0.23	13	1	3

The number of various molecular interactions was predicted using the protein–ligand interaction profiler (PLIP) server

^*Average ΔG ± standard deviation obtained from 50 docking iterations

^#Reported from the docked complex having the best combination of all three non-covalent interactions among the 50 docking iterations

Discussion

Comparing the publically available genomes of bifidobacteria reported to be associated with the human gut, we discovered a potential genomic signature with adequate sensitivity and specificity for Bifidobacterium species inhabiting the human gut. To identify the different Bifidobacterium species associated with the human gut, we examined the unified catalogue of gastrointestinal genomes comprising reference genomes of 4644 prokaryotes derived from the human faecal samples. A major limitation of this approach is that many of the Bifidobacterium species included for comparison were not, in fact, natural inhabitants of the human gut but are otherwise detected in the human faecal samples, possibly as a consequence of diet or by natural contamination (Turroni et al. 2009; Wong et al. 2020). However, the predicted MDR signature was remarkably specific to only the species that can colonise the human gut. As a consequence, the MDR signature sequence was not present in any of the following transient (sub)species native to hosts other than humans, such as B. animalis lactis from dairy products (Meile et al. 1997), B. animalis animalis, B. thermophilum and B. pseudolongum from pigs (Lamendella et al. 2008), and B. gallinarum and B. pullorum from chicken (Biavati et al. 2000). It should also be noted that the MDR signature was absent in species such as B. dentium and B. scardovii that are found to be associated with the human microbiota inhabiting different body sites such as the oral cavity (Ventura et al. 2009) and blood (Toh et al. 2015) respectively. Overall, the MDR signature may represent a niche-specific survival trait essential for surviving and colonising the human gut.

Phylogenetic analyses, based on the MFS protein-coding genes from the members of the Bifidobacteriaceae family homologous to the MDR signature sequence, also suggest a lineage-specific origin of the MDR signature exclusively in species inhabiting the human gut. The estimated phylogeny of the protein indicates that the MDR signature has a more ancestral relationship with the MFS transporters of bifidobacteria inhabiting the gut of the primate group rather than the MFS transporters of bifidobacteria inhabiting various human body sites other than the gut. Therefore, the MDR signature in bifidobacteria may have evolved by first undergoing divergence from an ancestral MFS transporter belonging to the members of the primate gut microbiome (node A Fig. 2). This initial divergence might have been followed by a domain fusion event resulting in the incorporation of the CBS domain to the C-terminal end of the MFS protein (nodes B and C Fig. 2). Subsequently, the MDR protein would have acquired independent changes to become exceptionally different from its homologous counterparts, forming a distinct cluster in the phylogenetic tree. Some of the changes accumulated by the MDR signature after its divergence were adaptive in nature with signs of positive selection. The majority of the positively selected sites with significant posterior probability were located on the interhelical loops connecting the transmembrane helices of the MFS domain. The amino acid residues in the interhelical loops are reported to influence the dynamics and structural stability of the transmembrane segments (Tastan et al. 2009). The observed positively selected changes may therefore be affecting the organisation of protein in the lipid membrane and conformational changes of the protein while transporting substrates.

The general transport mechanism of the MFS proteins comprises an “alternating-access model”, where a single substrate-binding cavity of the protein is alternatively exposed to either side of the membrane by undergoing structural transitions (Quistgaard et al. 2016; Drew et al. 2021). The transport cycle of MFS proteins broadly involves three distinct conformational states: inward-facing, occluded, and outward-facing. The modelled MFS region of the MDR signature had high homology with the crystallographically resolved representative structures of MFS proteins LacY and GlpT described as having an inward-facing conformation. In the inward-facing conformation, the substrate-binding cavity of the protein is exposed to the cytoplasm, where it can interact with various cellular contents that need to be expelled from the cell. The MDR-MFS proteins are generally involved in the efflux of structurally unrelated drugs and toxins out of the cells providing an adaptive advantage to the microbe for surviving and colonising hostile environments such as the human gut. In the case of MDR signature protein, we relied on molecular docking simulation to assess the potential of modelled MFS domain in translocating toxic substrates such as antibiotics and human bile acid, commonly encountered by microbial cells residing in the human gut. The antibiotics used for docking studies are protein synthesis inhibitors which can accumulate inside the microbial cell and target ribosomal protein synthesis machinery. Similarly, human bile acids are known to penetrate cells and cause irreversible damage to the membrane and DNA of microbial cells. Docking simulation studies showed high compatibility and affinity of the substrate-binding cavity of the MFS domain for the molecular structures of these compounds. Therefore, we predict that the MDR signature is involved in the possible efflux of these compounds conferring antibiotic and bile resistance to the bifidobacteria colonising the human gut. The major intermolecular force mediating the interaction between antibiotics and the MDR transporter seems to be hydrogen bonds. Intriguingly, the hydrogen bonds are highly dependent on the pH of the surrounding environment. Since the pH of the human gut is naturally acidic, the antibiotics bound to the substrate-binding cavity of the protein can be easily released when the MFS domain of the protein is in an outward-facing conformation. Since both groups of these substances, antibiotics and bile acids, commonly occur in the human gut and are toxic to the microbial cells, the MDR signature may be an important survival trait for the species colonising the human gut.

Despite being a member of a large family of secondary active transporters, the MDR signature is remarkably distinct from the other MFS proteins of this class in terms of the unique nucleotide sequence encoding the protein as well as the covalently linked MFS and CBS domains they possess. MFS transporters with fused C-terminal CBS domains are common in many members of the phylum Actinobacteria (Willson et al. 2019). It has also been reported in B. breve and B. longum, where a homologous MFS efflux pump involved in bile resistance was found to harbour covalently linked MFS and CBS domain (Gueimonde et al. 2009; Ruiz et al. 2012). However, the role of the fused CBS domains in these proteins was not investigated in detail. In other proteins with the CBS domain, such as AMP-activated protein kinase (AMPK) and ATP-dependent magnesium transporter MgtE, it was generally involved in regulating the catalytic/transporter activity of the protein by sensing the levels of adenosyl-carrying ligands such as AMP and ATP respectively (Scott et al. 2004; Hirata et al. 2014). In light of this evidence, we conducted molecular docking simulations of modelled MDR protein with structures of ATP and AMP, where both molecules exhibited favourable binding with the CBS domain.

Based on the evidence collected from our study, we hypothesise that the predicted genomic signature is imparting an adaptive advantage to the Bifidobacterium species in surviving and colonising the human gut. Single proteins, such as the MDR signature, harbouring more than one domain with different roles can provide a functional advantage to the organism possessing it (Willson et al. 2019). In the case of MDR signature, the catalytic MFS domain is involved in the possible efflux of drugs and toxins such as antibiotics and bile. In contrast, the C-terminally linked CBS domain is believed to be involved in regulating the transporter activity by sensing the levels of adenosyl carriers such as AMP and ATP. Although further experimental validations are required for our findings, we believe that the observations made in the present study will advance the understanding of the host-specific adaptations exhibited by gut commensals such as Bifidobacterium. We anticipate that our findings will guide future investigations, especially in the area of probiotics and the functional food industry, in evaluating and designing the probiotic strains used for supplementation. It would be beneficial for such strains to have host-specific survival traits, such as the discovered MDR signature protein, that can potentially increase their chance of survival in the human gut. Additionally, we believe our findings would also stimulate further research in understanding the mechanism and strategies of multidrug resistance, one of the rising concerns of the twenty-first century.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

AOK: Methodology, Software, Validation, Formal analysis, Investigation, Data Curation, Writing—Original Draft, Writing—Review & Editing and Visualisation. TP: Conceptualization, Methodology, Validation, Investigation, Resources, Writing—Review & Editing, Data Curation, Visualisation, Supervision, Project administration and Funding acquisition.

Funding

This research received no external funding.

Data availability

The datasets generated during the current study are available from the corresponding author upon reasonable request.

Declarations

Conflict of interests

The authors declare that they have no conflict of interest.