ABC transporters that mediate uptake of small molecules and other compounds from the extracellular environment (importers) are generally categorized into three different types based on their overall transmembrane domain (TMD) architecture (Thomas et al., 2020). Type I importers adopt a fold similar to the maltose transporter system MalFGK₂E of E. coli (Oldham and Chen, 2011), whereas Type II importers resemble the vitamin B12 uptake system BtuC₂D₂F from E. coli (Hvorup et al., 2007). Type III importers are also known as the ECF transporters, and have a vastly different TMD fold and overall mechanism than other classical ABC transporter systems, and are not considered here in our current analysis of B. subtilis ABC transporters. Type I and II importers both utilize a periplasmic solute binding protein (SBP) to bind and transfer substrates to the TMD of the ABC transporter for further import (Figure 1).

Some ABC transporters have an extra regulatory domain that extends off of the NBD, and often plays a significant role in modulating the overall conformation and function of the complex. Examples of such regulatory domains include those in MalK of the E. coli maltose transporter system which bind the enzyme IIA^glc and the transcriptional activator MalT (Boos and Shuman, 1998; Chen et al., 2003), and those in Wzt of the WzmWzt O-antigen transporter of Aquifex aeolicus which recognize O-antigen modifications (Bi and Zimmer, 2020). In the analysis below we divide predicted B. subtilis ABC importer structures based on classification into Type I or II, and whether or not they possess regulatory domains on the NBDs.

The B. subtilis genome contains thirteen type I ABC importers that lack a regulatory domain on the NBD (Figure 1 and Table 1). For all such transporters, AlphaFold predicted the periplasmic SBP interacting with the top of the MSDs such that the small molecule binding cleft of the SBP is oriented to facilitate transfer of transported substrates to the MSDs. Such a configuration is highly reminiscent of the structure of E. coli maltose transporter in a pre-translocation intermediate state (Oldham and Chen, 2011). Although the type I importers adopt overall similar folds, slight structural diversity still exists within the configuration of individual complexes. For instance, comparing the prediction for YtlCDA and GlnQMPH (top row, Figure 1) reveals a domain swap configuration for one transmembrane helix of YtlCDA and no such domain swapping in the TMDs of GlnQMPH. Moreover, the prediction for the ArtRQP complex (bottom row, Figure 1) demonstrates that this transporter is unique among type I importers because the SBP is fused to one of the MSD transmembrane helices. Overall, the AlphaFold predictions of Type I transporters reveal a similar architecture with slight deviations between complexes that are involved in uptake of substances including amino acids, di- and oligopeptides, sulfonate, and phosphate (Table 1). The predictions presented here suggest that the overall structure of type I importers is conserved across similar complexes in the B. subtilis genome, and the ability of each individual ABC transporter to recognize and transport specific solutes is likely dependent on the specific amino acid composition of substrate binding sites in the SBP and TMDs, rather than gross differences in transporter structure.

Among type 1 importers there are an additional eleven ABC transporter complexes in which the NBDs contain a regulatory domain (Figure 2 and Table 2). Interestingly, it has been proposed that some NBDs can act as “multitask” ATPases to energize multiple different ABC transporters. One of the best examples of such an NBD is MsmX, which has previously been shown to power at least 6 different ABC transporter systems in B. subtilis (Leisico et al., 2020). AlphaFold was able to reliably predict (based on mean pLDDT score) complexes for these 6 transporter complexes each containing MsmX as an NBD (Figure 2). The regulatory domain extending off of MsmX and FrlP adopts a transporter-associated OB (TOBE) domain fold similar to that of Malk in the E. coli maltose transporter complex, which acts as a hub to recruit regulatory proteins (Boos and Shuman, 1998; Chen et al., 2003). In contrast to the TOBE domain observed in MsmX, the regulatory domain of MetN in the MetNPQ complex adopts a NIL domain fold (PFAM: PF09383) that is commonly found in the NBD of many ABC transporters. On the other hand, the regulatory domains found in OpuAA, OpuBA, and OpuCA are cystathionine beta-synthase (CBS) domains (PFAM: PF00571), which have been shown to act as small molecule binding molecular switches (Baykov et al., 2011), and likely play a regulatory role in the conformational switching and overall function of these ABC transporters. Aside from the additional regulatory domains extending off of the NBD, all twenty-four type I importers in B. subtilis were predicted by AlphaFold to adopt a highly similar overall architecture reminiscent of the maltose transporter.

B. subtilis contains twelve type II importers that resemble the classical transporter BtuC₂D₂ with 20 TM helices and a SBP docking on top of the extracellular region of the MSDs (Figure 3 and Table 3; Hvorup et al., 2007; Locher et al., 2002). AlphaFold predicted a similar overall configuration for each of the type II importer complexes with NBDs separated in a nucleotide free conformation, and the SBP interacting with the top extracellular region of the MSDs. Among the twelve type II importers, the YvrABC complex is unique in that it is the only complex with regulatory domains extending off the NBDs into the cytosol (Figure 3). Like the E. coli transporter BtuC₂D₂, YvrABC is proposed to mediate uptake of vitamin B12 (cobalamin) in B. subtilis (Table 3). The regulatory domain of the YvrA subunit adopts what appears to be an FTH/DUF38 domain fold (PFAM: PF01827), which is presumed to act as a potential protein-protein interaction module. The presence of these regulatory domains in YvrABC suggests additional layers of regulating vitamin B12 import that may not be present in similar systems such as BtuC₂D₂ in E. coli.

Among all of the predicted B. subtilis ABC transporter complexes (both importers and exporters), the RbsACB and NupNOPQ type II importer complexes are unusual in that the two NBDs are fused together into a single polypeptide (Figure 3, red asterisk). Fusion of NBDs into a single polypeptide is relatively rare among prokaryotic ABC transporters, and it is currently unclear what if any role such a fusion may play in regulating transporter dynamics and function. It is interesting to note that sequence analysis of the RbsA and NupO NBDs suggests that one of the ATP binding sites in each of these proteins is degenerate (not competent for ATP binding/hydrolysis) as both proteins contain at least one ABC signature motif sequence that deviates significantly from the canonical LSGGQ sequence, in both proteins the second Walker A motif does not contain a lysine residue critical for coordinating bound ATP, and in the second Walker B motif of NupO the critical aspartic acid and glutamate residues are not present (Figure 3B). While degenerate ATP binding sites are not unusual in ABC transporters (Stockner et al., 2020), it is interesting that the only two NBDs that are fused into a single polypeptide in B. subtilis contain such a degenerate ATPase sequence. This observation raises intriguing questions about the evolutionary significance and potential functional implications of these degenerate ATP binding sites in the context of a fused NBD, which likely arose through gene duplication and eventual fusion. Further investigation is warranted to provide insight into how this unique structural feature may influence transporter activity and regulation.

ABC transporters belonging to the type IV exporter class adopt a fold similar to the founding member Sav1886 from Staphylococcus aureus (Dawson and Locher, 2006), and display 12 long transmembrane helices that extend well into the cytosol where they engage the NBDs (Figure 4). Type IV exporters also invariably display a domain swapped configuration, with transmembrane helix 4 and 5 from one MSD interacting through the intervening coupling helix with the NBD from the opposite half of the transporter. Two out of the three B. subtilis ABC transporters that have had experimental structures determined (BmrA and BmrCD) belong to the type IV exporter class (Chaptal et al., 2022; Thaker et al., 2022; Tang et al., 2023). Experimental structures of BmrA have been determined in an inward facing nucleotide free conformation (Di Cesare et al., 2024), and ATP bound conformations with both NBDs dimerized around bound ATP molecules (Chaptal et al., 2022). The AlphaFold-2 Multimer prediction of BmrA more closely resembles the inward facing nucleotide free cryo-EM structure of the complex (Di Cesare et al., 2024), although a significant difference in the degree of opening between the NBDs is observed when comparing the predicted and experimentally determined structure (Figure 4B). This difference in conformation could potentially be attributed to the solubilization condition (detergent, nanodisc, amphipol, etc…) used in experimental structure determination procedures, which has been shown to heavily influence the conformation of ABC transporters (Hoffmann et al., 2024). In contrast, the AlphaFold-2 Multimer prediction of BmrCD is virtually superimposable with the cryo-EM structure of BmrCD in an inward facing conformation (Thaker et al., 2022; Figure 4C). AlphaFold-2 Multimer predicted the structure of all other type IV exporters to be in an inward facing nucleotide free conformation, except for the YwJAA complex that displays collapsed NBDs in a conformation that would be expected upon binding of ATP.

Among type IV importers the BmrCD and SunTT complexes stand out with distinct structural features. BmrCD contains an extracellular domain that adopts a Penicillin Binding Protein 4 C-terminal domain fold of currently unknown function (Thaker et al., 2022). On the other hand, the SunTT complex contains an N-terminal C39 peptidase domain and an overall configuration nearly identical to that observed for the “PCAT” transporter from C. thermocellum (Lin et al., 2015). The C39 peptidase domain of SunT cleaves the pro-peptide of the glycocin antibiotic sublancin to generate the mature antimicrobial peptide, which is then transported out of the cell by SunT (Håvarstein et al., 1995). Aside from BmrCD and SunTT all other AlphaFold predictions of B. subtilis type IV transporters are highly similar to classical ABC transporters such as Sav1886 or MsbA (Dawson and Locher, 2006; Mi et al., 2017). These transporters all display domain swapping of transmembrane helixes 4 and 5 across the dimer interface, and long transmembrane helices that extend well into the cytosol with coupling helices that engage the NBDs at an extended distance from the plane of the lipid membrane compared to other ABC importers and exporters.

Type V exporters (Table 5) are characterized by their distinct fold of the TMD and a lack of domain swapping in the transmembrane helices. Transporters that adopt the type V exporter fold are typified by the Wzm family of transporters in bacteria (Spellmon et al., 2022), and the ABCG (Taylor et al., 2017; Lee et al., 2016) and ABCA (Plummer-Medeiros et al., 2023) families in humans. B. subtilis contains 11 type V exporters that can be further divided into two classes, those with regulatory domains (5 transporters) or without regulatory domains (6 transporters) extending off of the NBD (Figure 5). Interestingly, AlphaFold predicted reasonable ABC transporter complex structures for all type V exporters with regulatory domains (Figure 5A), but generally struggled to provide accurate predictions for those without regulatory domains (Figure 5B). The regulatory domains on all type V exporters adopt a similar fold to that seen previously in the human ABCA1 (Plummer-Medeiros et al., 2023) and ABCA4 (Scortecci et al., 2021; Xie et al., 2021) structures with two alpha-helices and a 4-stranded anti-parallel beta-sheet (Figure 5A). Among type V exporters without regulatory domains the SkfEF, AlbCD, EcsAB, and YthPQ complexes are unique in that all of the TMD is encoded by a single polypeptide versus two individual polypeptides (Figure 5B). AlphaFold appears to struggle predicting an accurate structure for these complexes as evidenced by the large separation in the TMD region that is likely not representative of the true structure. Although these predictions can display a lower pLDDT score in the linker region connecting the TMD two helical bundles (Figure 5D), the mean pLDDT score and pTM score for most of these predictions are similar to other well predicted complexes. Thus, assessing the quality of a given prediction based solely on metrics such as pLDDT and pTM may fail to capture possible inaccuracies. Similarly, in the YhcGE and YybJKL complexes which contain two separate polypeptides for each individual TMD, AlphaFold appears to correctly model the fold and interaction of the TMD and NBD, but there is a large gap between the individual TMDs that seems unlikely to be accurate.

The final member of the type V exporters in B. subtilis is the TagGH complex that is involved in transport of teichoic acids (Lazarevic and Karamata, 1995). The AlphaFold predicted structure of TagGH reveals a similar overall fold as the TarGH teichoic acid transporter from Alicyclobacillus herbarius (Chen et al., 2020), and predicts the complex to be in a nucleotide bound conformation similar to that observed in the TarGH cryo-EM structure. However, TagGH is unique in that the TagG NBD subunit contains a transmembrane domain that positions the c-terminus of the protein in the extracellular region (Figure 5C). The C-terminus of TagG is predicted as a well folded domain that has some homology to SH3b domains known to bind polymers such as teichoic acid in the cell-wall (Shen et al., 2020). A long (∼30 residues) and likely flexible linker connects the transmembrane helix and C-terminal domain of TagG, and AlphaFold places the C-terminal domain in the cytoplasm near the NBDs (Figure 5C). This is an incorrect positioning as the C-terminal domain will certainly reside on the extracellular side of the plasma membrane, and likely extends up from the plane of the bilayer towards the thick peptidoglycan layer surrounding B. subtilis. The incorrect positioning of this domain is likely a consequence of AlphaFold not having knowledge of the membrane location, and the low pLDDT score in the linker connecting the TagG NBD and C-terminal extension (Figure 5D). Nevertheless, the AlphaFold prediction provides first insight into the fold and plausible function of the TagG C-terminal domain.

Type VII exporters are also known as “mechanotransducers” as these ABC transporters do not necessarily move a substance across the membrane, but rather act to activate extracellular enzymes (Hao et al., 2024), extract lipoproteins from the membrane (Sharma et al., 2021), or pump small molecules through conduits to the extracellular space (Crow et al., 2017). The type VII exporter fold is typified by the structure of the MacB multidrug transporter from E. coli (Crow et al., 2017), and all type VII exporters have been found to have FtsX-like folds within the TMD architecture. In B. subtilis there are eight type VII exporters that can be divided into two distinct classes. The first class contains the FtsEX, YclHI, YknYZ, YtrEF, and YvrON complexes, all of which adopt a classical MacB type fold. These transporters are overall two-fold symmetrical, and contain two MSDs that each adopt FtsX-like 4-TM helical bundles for a total of eight transmembrane helices, as well as a large extracellular domain of diverse size and structure (Figure 6A). Interestingly, the type VII transporters appear to have some of the lowest pTM scores across all transporters in B. subtilis. However, we believe these below average scores are likely a consequence of poor confidence (pLDDT) in several dynamic loop regions and the greasy interaction between FtsX like folds of the two MSDs, rather than any major overall inaccuracy in the structural prediction. Aside from FtsEX which is known to activate a cell-wall hydrolase during cell division, the function of other type VII exporters within this category remains relatively obscure or ill-defined (Table 6).

The second category of type VII exporters is the “Bce-type” transporters which are named after the founding member BceAB from B. subtilis (George et al., 2022). There are three Bce-type transporters in B. subtilis (Figure 6B) that all adopt an overall configuration that is similar to that observed in the cryo-EM structure of nucleotide-free BceAB (George et al., 2022). Bce-type transporters are distinct from other type VII transporters in that they contain ten transmembrane helices encoded in a single polypeptide, contain only a single extracellular domain, and lack overall symmetry in the complex. The largest difference between the three Bce-type transporters in B. subtilis lies in the fold of the extracellular domain, which is proposed to mediate recognition of highly diverse antimicrobial peptides by each of the complexes (George et al., 2024). Bce-type transporters are known to form a membrane protein complex with a histidine kinase, which together work in tandem to sense antimicrobial peptides and initiate signaling to mediate resistance (George and Orlando, 2023; George et al., 2024).

The remaining five exporters in B. subtilis (Table 7; YhcHI, YhaQP, LnrLMN, NatAB, and YjkBA) display AlphaFold predicted structures that appear to be a hybrid between different classes of transporters. The YhcHI, YhaQP, LnrLMN, and NatAB complexes display a similar TMD architecture as type V exporters with a short stature and lack of domain swapping in the TMD region, but also display relatively large extracellular domains (Figure 7A). The extracellular domains of these transporters are similar in size and location to the beta jelly-roll domains found in the type VI transporter complex LptB₂FG from Gram-negative (Li et al., 2019; Owens et al., 2019). For this reason, we have designated YhcHI, YhaQP, LnrLMN, and NatAB B. subtilis transporters as hybrids between type V and type VI transporters. YhaQP, LnrLMN, and NatAB all display domain swapped extracellular domains that adopt a similar fold of a five-stranded beta-sheet with flanking alpha-helices (Figure 7A). Searches for similar domains or protein folds for these extracellular domains revealed no known protein domains, suggesting a novel overall fold. In contrast to YhaQP, LnrLMN, and NatAB, the YhcHI complex displays a smaller extracellular domain composed of two alpha-helices that form a coiled-coil. Interestingly, among the transporters classified as type V/VI hybrids, the only complex that lacks regulatory domains on the NBD is NatAB (Figure 7A).

The final ABC transporter complex in B. subtilis is the YjkBA complex, which has an unusual topology unlike any of the other transporters found in this organism (Figure 7B). The YjkBA complex displays a total of fourteen transmembrane helices that are arranged in a domain swapped configuration similar to that seen in type IV transporters (Figure 4). However, unlike type IV transporters the transmembrane helices of the YjkA do not extend far below the plane of the lipid bilayer like type IV transporters, and as a result the YjkB NBDs are located closer to the lipid bilayer than they are in type IV transporters. Moreover, YjkA lacks an N-terminal “elbow helix” that is typical of type IV transporters (Thomas et al., 2020). The short stature (location of the NBDs close to the membrane) and compact nature of the TMD in the YjkBA complex is somewhat reminiscent of type V transporters. For these reasons we have designated YjkBA as a hybrid type IV/V exporter, which adopts an overall architecture unlike any ABC transporter of known structure.

During preparation of this manuscript the latest version of AlphaFold (AlphaFold-3) was released (Abramson et al., 2024). Although the source code for this new implementation of AlphaFold was not made available, the ability to perform predictions between proteins and a limited list of small molecules was provided in the form of AlphaFold Server.¹ Among the list of available ligands is the nucleotide ATP, which provides an opportunity to explore the ability of AlphaFold-3 in predicting the conformational changes that are thought to be initiated by ATP binding to ABC transporters. To explore this possibility, we performed AlphaFold-3 predictions either in the absence or presence of ATP molecules for the three B. subtilis ABC transporters that have had experimental structures determined (George et al., 2022; Tang et al., 2023; Di Cesare et al., 2024). In the absence of ATP, the AlphaFold-3 predictions for BmrCD, BmrA, and BceAB reveal an essentially identical prediction to that which was provided by AlphaFold-2 Multimer (Figure 8, left). When two copies of an ATP molecule were included in these predictions, AlphaFold-3 placed each ATP molecule in the correct location between the Walker A and signature motif of each NBD in the ABC transporter complexes, and in all three complexes inclusion of ATP induced a conformational shift towards a nucleotide bound conformation (Figure 8, left).

Although AlphaFold-3 reliably predicted the general ATP binding sites in each complex, the conformational shifts observed in each complex show some discrepancies to experimentally determined structures. The AlphaFold-3 prediction of BmrA bound to ATP is nearly identical to the experimentally determined cryo-EM structure (PDB: 7OW8) (Di Cesare et al., 2024), with only slight discrepancies in the relative position of the loop extending between transmembrane helix 1 and 2 (Figure 8, right). On the other hand, the AlphaFold-3 prediction of BmrCD in the presence of ATP reveals significant differences in the positioning of the NBDs and coupling helices relative to the experimentally determined cryo-EM structure (PDB:8FHK) (Tang et al., 2023). In the AlphaFold-3 predicted structure of BmrCD with ATP the NBDs are not fully dimerized to sandwich ATP between each NBD, and further conformational change (toward the experimentally determined structure) would be required for BmrCD to adopt a conformation that is competent for ATP hydrolysis (Figure 8, right). Finally, the AlphaFold-3 predicted structure of BceAB with ATP molecules is strikingly different than the experimentally determined cryo-EM structure (PDB:7TCH) (George et al., 2022). In the AlphaFold-3 prediction of BceAB the NBDs are not fully dimerized in a conformation that would support ATP hydrolysis, the transmembrane helices do not collapse into a more compacted state as seen in the cryo-EM structure, and the large extracellular domain is in a completely different conformation than that observed in the cryo-EM structure. Thus, while AlphaFold-3 was able to predict conformational changes elicited by ATP binding, in two out of the three cases explored here the predicted structures of ABC transporters bound to nucleotides (ATP) deviate in several ways from the same structures determined through experimental methods. The implications of these findings are explored in further detail in the following discussion.

The inventory and assembly of B. subtilis ABC transporters previously assembled by Quentin Y. et al. was initially used as a template to inspect the genomic neighborhood, predicted function, and AlphaFold predicted 3-dimensional structure of every plausible ABC transporter. For these initial searches and visual inspection we utilized Subtiwiki (Pedreira et al., 2021), which provided a convenient format to analyze both the genomic context and predicted structure of individual ABC transporter components. From this analysis we compiled a new inventory of ABC transporter complexes in B. subtilis complete with updated protein nomenclature, predicted subunit stoichiometries, and biological function. Individual protein amino acid sequences for every component of every ABC transporter complex were compiled into a database that was subsequently used to generate input files for AlphaFold predictions. For the SBP of ABC importers the signal sequence that directs protein translocation to the extracellular space was left intact, and can often be visualized in the resultant structural predictions as an unstructured segment extending off of the NBD. Inclusion of the SBP signal sequence did not impair prediction of any ABC importer structures.

AlphaFold predictions were run on a local GPU (NVIDIA RTX-3090) utilizing the LocalColabFold² version of ColabFold (Mirdita et al., 2022). An input.csv file containing a list of all ABC transporter complexes with correct amino acid sequences and stoichiometry was used as an input to predict all structures using the AlphaFold2-Multimer version 2 model (Evans et al., 2022). Predictions were run using the default LocalColabFold settings with 3 prediction recycles and no templates from the PDB or custom PDB templates provided. In order to save computational time, the AlphaFold predictions did not include addition of hydrogens or AMBER relaxation/energy minimization. Output PDB files, pLDDT confidence scores, and predicted aligned error (PAE) were visually inspected on an individual basis for each ABC transporter complex prediction. For each ABC transporter complex AlphaFold generated five atomic models ranked in order of predicted accuracy. For simplicity, only the top ranking model for each ABC transporter complex is displayed in this manuscript.

To identify common domain folds and protein families for regulatory and extracellular domains we employed several bioinformatic tools. The InterPro database (Apweiler et al., 2001) was used to search for protein domains by querying the amino acid sequences corresponding to individual regulatory or extracellular domains. Interpro results were used to identify individual protein domains and their common biological functions. Additionally, the FoldSeek server (van Kempen et al., 2024) was used to identify similar transporters that match the unique topologies found in the extracellular domains of the Type V/VI hybrids.

AlphaFold structure predictions were initially visualized and compared using UCSF Chimera (Pettersen et al., 2004). All figures were generated with UCSF ChimeraX software (Pettersen et al., 2021).