We are pleased to present two modeling projects that explore important aspects of biological processes and are of great relevance to us. The first project involves promoter prediction for the genes VbrR and Vbrk, so that we can provide the wetlab with the sequences they need to advance their efforts. At the same time, our second project focuses on receptor design for the membrane protein BlaR1, with the goal of improving our understanding of the binding interactions between the binding pocket and our chosen ligands. These efforts have the potential to provide valuable insights into receptor functionality and the world of biosensors, contributing to the fields of synthetic biology and biotechnology.
Our biosensor project utilizes the VbrK/VbrR two-component system (TCS) of Vibrio parahaemolyticus. In 2015, Li et al. identified the VbrK/VbrR response regulator pair to control the expression of β-lactamase, inducing antibiotic resistance in V. parahaemolyticus. In this system, VbrR regulates the expression of BlaA (vpa0477), which encodes for a class A chromosomal carbenicillin-hydrolyzing (CARB) β-lactamase (Chiou et al., 2015). The BlaA gene is located on the reverse strand on chromosome 2 (position 476708 ← 477559) of the V. parahaemolyticus genome (RIMD 2210633). However, the exact promoter region of BlaA and the binding site of the response regulator VbrR are unknown.
Figure 1: Visual Overview of the VbrR binding site and blaA's promoter prediction. VbrR interacts with the σ-factor and increases its binding affinity to the promoter, following a recruitment of an RNA-polymerase that initiates the transcription at the transcription start site (TSS).
Bacterial promoters are an essential part of the gene located upstream of the transcription start site (TSS). In those regions, transcription factors, such as σ70, bind and recruit an RNA-polymerase, initiating gene transcription. The structure of the binding sites of σ70 transcription factors in the promoter region are well-known and can usually be determined by a -35 and -10 region with respect to the TSS (cf. Fig. 1) (Nordheim et al. 2018). Response regulators of the OmpR/Phob superfamily, to which VbrR belongs (Cho et al. 2021), can interact with the σ70 factor and initiate and enhance its DNA binding affinity (Canals et al. 2021, Lou et al. 2023), thereby controlling the expression of the BlaA gene.
Our project aims to establish the VbrK/VbrR system in E. coli, to detect the presence of β-lactam antibiotics by replacing BlaA with a β-galactosidase as a reporter gene. Therefore, our designed plasmids would benefit from an accurate estimation of the promoter and VbrR's binding site. Since promoters are coherent in bacteria, we decided to search for the promoter and VbrR's binding site in the region between the gene start of BlaA and 300 bp upstream (477560 ← 477860). To simplify the visualization, we inverted the sequence of the promoter and BlaA and selected the first base pair (bp) of the promoter as the overall first bp.
Embarking on the journey of promoter prediction, we delve into the fascinating realm of uncovering regulatory elements that govern gene expression. A pragmatic strategy to detect promoters is to employ one of the many available promoter prediction tools. While early approaches to these tools primarily focused on identifying specific motifs within sequences, contemporary machine-learning techniques have expanded to integrate data from diverse biological sources (Cassiano and Silva-Rocha, 2020). We opted for SAPPHIRE (Coppens and Lavigne, 2020), a neural network-driven classifier tailored for σ70 promoter prediction in Pseudomonas and Salmonella because V. parahaemolyticus is closely related to both. Organisms of these genera belong to the class of gammaproteobacteria (Gupta et al. 2016, Gomila et al. 2015, Liu et al. 2022). It is important to note that machine learning tools like SAPPHIRE base their predictions on a finite dataset - in this instance known promoters from Pseudomonas and Salmonella.
Our project aims to establish the VbrK/VbrR system in E. coli, to detect the presence of β-lactam antibiotics by replacing BlaA with a β-galactosidase as a reporter gene. Therefore, our designed plasmids would benefit from an accurate estimation of the promoter and VbrR's binding site. Since promoters are coherent in bacteria, we decided to search for the promoter and VbrR's binding site in the region between the gene start of BlaA and 300 bp upstream (477560 ← 477860). To simplify the visualization, we inverted the sequence of the promoter and BlaA and selected the first base pair (bp) of the promoter as the overall first bp.
Figure 2: SAPPHIRE promoter prediction results A, B: Predicted promoter regions annotated towards p-value and approximated transcription start site (TSS). The black solid lines represent the TSS of blaA located at 300 bp. The black circles indicate the selected sequences from individual experiments. C: blaA and its potential promoter region. Yellow bars indicate the predicted promoter sequences by both experiments (Seq1, Seq2) resulting from the intersection of selected Pseudomonas- and Salmonella-related promotor sequences.
Ultimately, by harnessing the power of computational tools, we are able to illuminate the intricate landscape of promoter regions, enhancing our understanding of the genetic regulation of our reporter genes.
With our newfound insights into potential promoter sites in which σ-factors bind, we are well-equipped to ascertain the binding site for VbrR. The VbrK/VbrR TCS is also involved in repressing type 3 secretion systems (T3SS1). After activation of VbrK through S-nitrosylation of a cysteine residue, VbrR down-regulates the expression of exsC, a positive regulator of T3SS1. VbrR represses the expression of exsC by directly binding to the exsC promoter (Fig 3). In this case, VbrR competes with a σ-factor belonging to the σ70 family (Gu et al., 2020). Gu and colleagues determined the binding site of VbrR located in the exsC promoter.
Figure 3: Identified binding site of VbrR by Gu et al. in the exsC promoter. The figure is adapted from Gu et al. (Gu et al., 2020). The black arrow indicates the TSS of exsC.
Under the assumption that the VbrR binding site upstream of the BlaA promoter would be similar to this 49 bp binding site, we performed local alignments against our 300 bp long sequence containing the putative binding site.
Local alignments were used to determine whether two sequences (or subsets of each) are a close match to another. To achieve this all possible ways to match the two sequences were given a score according to the scoring function S(ai, bj), whereby ai, bj denote an observed nucleotide at position i, j of the two sequences a and b, respectively. Since we were not interested in the best hit, i.e. the alignment with the highest score, only, we clustered the first 1000 sequences, sorted by their descending scores, whenever the region covered by two alignments was similar. If they exceeded a limit of 20 bp on one side of the prior sequence, the sequences were discarded from the cluster.
Results are shown in Fig. 3. In total, 5 bins were created using our procedure (Fig. 4A), whereby the first 4 bins hold about 90% of all the alignments (Fig. 4B) and were therefore selected as potential VbrR binding sites (Fig. 4C).
Figure 4: Local alignment results using the 49 bp sequence identified by Gu et al. (Gu et al. 2020). A: Best 1000 local alignments clustered into 5 clusters. The top score refers to the best alignment included in the cluster. B: Number of sequences belonging to each cluster. C: Position of cluster 1 to 4 in the potential promoter region.
A subsequent study by Hong and colleagues (Hong et al. 2022) resolved the crystal structure of the VbrR-DNA complex. Similar to other response regulators of the OmpR/PhoB superfamily, VbrR exhibits an N-terminal receiver domain (RD) and a C-terminal DNA-binding domain (DBD), and exists mainly as a dimer in solution. Hong et al. identified two 7 bp DNA half-sites S1 (TTCTAAT) and S2 (TTCATCG) within the VbrR binding site that is bound by the two DBD units of the dimer. Notably, those two sites are solely separated by 2bp.
On top of aligning the 49 bp binding site, we decided to map the two motifs against our sequence to identify possible DNA half-sites that allow binding of VbrR. To that end, we used alignments but only evaluated the gap size between the two half-sites and the number of mismatches. A gap size of up to 6 bp and a maximum of 5 mismatches were allowed. No binning was applied at this stage. Fig. 5A lists all hits based on their gap size and number of mismatches. Even though no perfect alignment was found, we decided to favor fewer mismatches and accept larger gap sizes. Thereby, the locations of sequences 4 and 6 were selected as potential binding sites (Fig. 5B) as they also overlapped with the determined binding site in Fig 4.
Figure 5: Alignment results of the two half-site sequences S1 and S2 identified by Hong et al. (Hong et al. 2022). A: Number of mismatches of the 6 alignments between the aligned half-sites and the promoter region. The gap size indicates the gap size between the two half-sites. B: Position of the 6 half-site alignments in the promoter region. Pinkish bars indicate the selected binding sites.
Our promising discoveries pave the way for a deeper understanding of VbrR's binding behavior.
Finally, we focused on integrating the gathered data. The exact interaction between VbrR and the σ70 factor remains unknown. However, by integrating several studies connected to VbrR's superfamily, and predicting potential promoter sequences using SAPPHIRE, we are now able to determine VbrR's binding sites and their potential function. All predicted promoter regions and VbrR's binding sites are shown in Fig. 6.
By referring to the knowledge from the OmpR/PhoB superfamily and the direct interaction with the binding sites, we are able to classify our suggested binding sites of VbrR into two categories: enhancing (green), located upstream of the -35 region, and σ70-activating (orange), directly located in the -35 and -10 regions. We predict that the promoter region is located within sequence positions 477560 ← 477668 (108 bp), and the enhancing region is positioned at 477707 ← 477761 (54 bp).
Figure 6: Combination of the results of all three approaches. (SAPPHIRE, local alignments, half-sites). Results show a promoter region (orange) and a potential enhancer region (green).
With those predictions we are now able to improve our plasmids in terms of size and functionality. However, experimental work is required to validate the predicted binding sites, promoter regions, and their function. We plan to perform Electrophoretic Mobility Shift Assays (EMSA) that would allow us to determine whether VbrR binds to specific DNA fragments. To validate the enhancing potential of VbrR binding sites, we will conduct in-gel fluorescence assays employing GFP, aiming to discern any observable increase in protein expression upon their inclusion.
Existing, more expensive, antibiotic detection systems for fresh and wastewater have been designed to very precisely determine the concentration of a specific molecule. For example, complex and expensive electrochemical and optical biosensors were built to determine the presence of Penicillin G or Ampicillin in nanogram range. (Zeng et al. 2022)
In their natural state, our two-component systems cannot distinguish between Penicillin G, Ampicillin, or other types of β-lactam antibiotics found in wastewater probes. They can only determine the presence or absence of any β-lactam antibiotic. To tackle this, we wanted to establish a toolbox of receptors, each with a high specificity for a specific β-lactam antibiotic. Our goal is to create two variants of the BlaR1 receptor, each variant exhibiting increased specificity for a particular β-lactam antibiotic (as shown in Fig. 1).
Figure 1: Illustration showing the naive BlaR1 receptor (yellow receptor) that binds to all types of β-lactam antibiotics. BlaR1 will be mutated and optimized for high binding affinity to either Penicillin G (red receptor) or Ampicillin (blue receptor).
Ultimately, we decided to use the two different β-lactam antibiotics: Penicillin G and Ampicillin. These are often used to cure diseases connected to bacterial infections (Yip et al. 2022, Peechakara et al. 2022). Both antibiotics are highly concentrated in wastewater, especially in developing countries as shown by a case study (Samandari et al. 2022), potentially connected to higher consumption rates. The intriguing aspect of these antibiotics lies in their remarkable similarity, which presents a significant challenge in the process of receptor design. To enhance binding affinity, it is essential to establish novel interactions between the protein and the antibiotic. However, when dealing with antibiotics that bear a high degree of similarity, enhancing binding affinity can only be attributed to the minimal chemical distinctions that exist among these closely related antibiotics.
Fig. 2 shows the structural formula of the two selected β-lactam antibiotics. Ampicillin contains an amino group which is absent in Penicillin G. The difference is rather marginal but detectable. Mutants could favor ampicillin by forming hydrogen bonds to the amino group located next to the aromatic ring.
Figure 2: Structural formula of Penicillin G (A) and Ampicillin (B) showing small structural distances.
To design mutants effectively, it is crucial to have a comprehensive understanding of how antibiotics interact with the receptor, ideally with a high degree of clarity. To that end, we used current literature and investigated crystal structures that already include a bound Penicillin G in the sensor domain (PDB: 1xa7). Fig 3. shows the crystal structure of BlaR1's sensor domain including the covalent bound Penicillin G (Wilke et al., 2004). By interacting with BlaR1, the highly-reactive β-lactam ring breaks, and binds covalently to the serine located in the binding pocket (Wilke et al 2004).
As indicated by Fig. 3B, the Penicillin G bound to the BlaR1 is fixed in the binding pocket by 7 hydrogen bonds, the carboxylate group is connected by an ionic bond, and the benzyl ring is held in place by multiple hydrophobic interactions. Overall, the antibiotic positioned well in the binding pocket with many present polar interactions.
Figure 3: Interactions between Penicillin G and BlaR1 as observed in the crystal structure (PDB: 1xa7). A: Formation of a covalent bond with the serine residue in an open conformation. B: Polar and non-polar interactions occurring between Penicillin G and the receptor estimated by PLIP (Adamse et al., 2021) Interactions: ion bond (yellow dashed line), hydrogen bond (blue solid lines), hydrophobic interactions (grey dashed lines)
The newest research by Alexander et al. (Alexander et al., 2023) revealed the entire structure of BlaR1, also covering the intercellular part. We used one monomer of this structure (PDB: 8exq - chain B) to determine the interaction between the receptor and the considered antibiotics, and prepared mutants with higher specificity.
Besides Penicillin G, no other β-lactam antibiotic was considered in a crystal structure experiment using BlaR1, as outlined before. Therefore, we first investigated the binding behavior of all selected antibiotics using covalent docking using the MOE software (MOE 2022).
Figure 4: PLIP results of the naive BlaR1 receptor docked covalent to Penicillin G (A) and Ampicillin (B) indicating similar polar and non-polar interactions. Interactions: ion bond (yellow dashed line), hydrogen bond (blue solid lines), hydrophobic interactions (grey dashed lines)
Fig. 4 illustrates the covalent docking of Penicillin G and ampicillin. Comparing the covalent dockings from the structure revealed by Alexander et al. to the crystal structure that includes the Penicillin G less hydrogen bond can be observed. Additionally, the experiment indicates that Penicillin G and Ampicillin exhibit similar interactions. As a result, there appears to be no residue engaging with the amino group adjacent to the benzyl ring in Ampicillin. This intriguing finding paves the way for the potential development of a mutant strain that could specifically enhance the affinity for Ampicillin due to its unique amino group.
BlaR1 exhibited no discernible preference in its binding affinity towards specific ß-lactam antibiotics. Consequently, we can proceed with designing mutants. To our advantage, the development of sophisticated algorithms offers an efficient solution to automate this protein design process, sparing us from a manual design approach. We considered the Rosetta interface design protocol (Moretti et al., 2016) to design our protein interface interacting with the antibiotics. In this process, certain amino acids in the binding pocket, which potentially get in contact with the antibiotics, are mutated. Those amino acids are automatically detected by Rosetta itself.
Our first approach was to discover potential mutants by a non-covalent approach. Hereby, we placed the antibiotics in their activated state (closed β-lactam ring) in the binding pocket, created a rotamer library, and used the Rosetta interface design protocol to establish 1000 mutated BlaR1 receptors. Afterward, the top 3 mutants based on Rosetta's total score were considered and evaluated using MOE and PLIP (Adamse et al., 2021). Thereby, the interaction between the antibiotics and the receptor was assessed after docking them covalently again in the optimized receptors.
The approach is driven by the intricate process of the antibiotic's interaction with the protein. Specifically, the antibiotic must initially reach the serine residue while still in its activated state, before subsequently undergoing a reaction and forming a covalent connection with it.
In the case of Penicillin G, no receptor could be established that showed a significant increase in the binding affinity towards Penicillin G. However in the case of Ampicillin, a mutant was found that indicated a higher binding affinity compared to the state where the receptor binds covalent to Penicillin G indicated by more hydrogen bonds and hydrophobic interactions. In contrast to the naive receptor, the mutant receptor exhibited a noticeable reduction in critical interactions, notably the high-energy ionic bond that typically binds to the carboxylate group and specific hydrogen bonds. However, it did manage to establish a distinctive T-shaped π-stacking interaction between the benzyl ring and a phenylalanine residue.
Figure 5: Most optimized BlaR1 mutant for ampicillin after docking covalent: Quantification of polar and non-polar interactions between the mutated receptor and the targeted antibiotics (A). Visualization of interactions as 3D structures between the mutated receptor and Ampicillin (B) or Penicillin G (C). Results were discovered using PLIP. Interactions: π-stacking (green dashed line), hydrogen bond (blue solid lines), hydrophobic interactions (grey dashed lines).
Fine-tuning non-covalent bonds did indeed lead to a notable differentiation between the two antibiotics, particularly with Ampicillin. However, achieving a more pronounced distinction between them will require more substantial alterations in protein-antibiotic interactions.
Since our initial approach did not uncover tremendous differences in the interactions between the BlaR1 and the considered antibiotics, we explored an alternative method. In the previous non-covalent approach, we optimized the protein's interface considering potential conformations of the antibiotics. This approach could lead to changes in amino acids that do not interact with the antibiotic in its inactive state (open β-lactam ring). However, the protocol proposed by Rosetta was not suitable for dealing with covalent binding. Therefore, we pursued optimization of the protein interface using the inactive state of the antibiotics without actually forming a covalent bond with the protein. The rest of the pipeline remained. We call this the pseudo-covalent approach.
The pseudo-covalent approach, in a similar vein, did not produce a structure with markedly increased estimated binding affinities for Penicillin G. Surprisingly, the optimization consistently leaned towards Ampicillin, regardless of the intended antibiotic in the interface design. However, upon closer inspection of the top three results for Penicillin G, we did discover mutants that exhibited notably exceptional estimated binding affinities for Ampicillin.
Figure 6: Most optimized BlaR1 mutant optimized for Penicillin G show much higher estimated binding affinity towards Ampicillin in the pseudo-covalent approach: Quantification of polar and non-polar interactions between the mutated receptor and the targeted antibiotics (A). Visualization of interactions as 3D structures between the mutated receptor and Ampicillin (B) or Penicillin G (C). Results were discovered using PLIP. Interactions: ion bond (yellow dashed line), hydrogen bond (blue solid lines), hydrophobic interactions (grey dashed lines).
Fig. 6 shows the discovered mutant of BlaR1 which indicates many more hydrogen bonds for Ampicillin, resulting in a higher estimated binding affinity. This finding implies that utilizing similar antibiotics in the optimization process can help achieve results that are more optimal for the desired antibiotic.
Creating receptors with enhanced binding affinities for specific antibiotics is an essential requirement for our whole-cell biosensor to remain competitive in the field of antibiotic detection systems. In total, we successfully developed two mutants of BlaR1 that exhibited significantly increased estimated binding affinities for Ampicillin. In doing so, we leveraged Rosetta's interface design protocol, introducing a novel approach to identify receptor mutants tailored for specific ligands.