Key Points

• Completed 4.5 iterations of the DBTL (Design-Build-Test-Learn) cycle.

• Established a new Golden Gate standard for NRPS engineering.

• Characterized 105 expression plasmids using our platform.

• Developed computational tools for guided NRPS engineering.

The Challenge

Natural products (NPs) are known for their enormous structural complexity and scaffold diversity features that often remain inaccessible by chemical synthesis^[1][2](Project description).

Unlocking this chemical space holds tremendous potential for drug discovery. Our project introduces a novel natural product derivatization platform designed to overcome these limitations, enabling the rapid generation of new-to-nature compound libraries.

At its core, the platform builds on NRPS engineering leveraged by synthetic biology tools. The concept is simple but powerful: every position within a natural product scaffold can, in principle, be accessed through the respective module in the NRPS assembly line. By exchanging a module in the NRPS enzyme, we can directly exchange the corresponding amino acid in its product, resulting in new-to-nature peptides (Fig.1).

🔍

However, despite recent progress in NRPS engineering technologies, reprogramming these enzymatic assembly lines remains a formidable challenge^[3]. Their intrinsic complexity makes it difficult to predict whether heterologously introduced modules will function together as intended. In fact, even replacing a single module within a native NRPS system, while maintaining activity, often presents a substantial hurdle. Additional obstacles include the difficulty of heterologous expression and the large size of NRPS gene clusters. Furthermore, modular exchange strategies based on standardized building blocks for derivatization are still lacking. Current Golden Gate–based approaches rely on non-standardized overhangs and require extensive cloning, while issues of module compatibility further limit their applicability^[4](Fig. 5).

To truly exploit the full potential of natural products, we set out to develop a platform that enables the standardized exchange of modular building blocks within NRPS enzymes. By making NRPS engineering broadly accessible and easy to use, we aimed to unlock the full potential of these enzymatic machineries, at the same time empowering other teams to design and build enzymes for producing novel peptides with ease.

Initial considerations

At the start of our project, we chose E. coli as our engineering platform. It is one of the best-established heterologous expression hosts, offers reliable systems for DNA manipulation, and is widely accessible to iGEM teams around the world, making it an ideal foundation for our NRPS engineering efforts (interview with Joachim Hug). To increase the likelihood of having functional biosynthetic gene clusters (BGCs) we aimed to express in our host, we focused on NRPS BGCs from the same class of gamma proteobacteria, more specifically on the closely related genera Photorhabdus and Xenorhabdus. This choice was guided by key considerations such as GC content, codon usage, and the availability of metabolite precursors required for efficient peptide biosynthesis. Additionally, it was suggested by Kim Lewis that NRPs produced by bacteria associated with parasitic nematodes likely possess favorable pharmacokinetic properties, enabling them to effectively disseminate throughout the insect host infected by the nematode^[5].

In general, each NRPS module is encoded by about 3 kb of DNA (NRPS and engineering). Since our clusters extend up to 11 modules, this corresponds to roughly 33 kb in total (Cluster selection)^[6].

Introducing such large inserts into plasmids is possible but they are very challenging to clone and transform, and therefore not an ideal basis for our platform.

To make the expression of larger native BGCs feasible, we split the NRPS BGCs across three separate, orthogonal plasmids, each capable of carrying up to 15 kb inserts. Once divided at the DNA level and co-expressed in E. coli, the challenge was to ensure that the separately expressed NRPS fragments could reassemble into a functional protein complex. To achieve this, we looked at different NPRS engineering tools and we decided to use split inteins(NRPS and engineering), a powerful tool in protein engineering that enables specific benefits compared to other engineering tools. Split inteins provide post-translational covalent linkage of two independently expressed proteins. We utilized two orthogonal pairs of split inteins, that were recently introduced as tools for NRPS engineering, to ensure efficient and specific reassembly of the NRPS fragments^[6].

Preliminary Experiment

Testing heterologous expression of various BGCs in E. coli

To gain an understanding of which peptides, and therefore NRPS clusters, could serve as novel and promising lead molecules for antibiotic development, we contacted Prof. Felix Hausch (interview with Felix Hausch). He told us that macrocycles generally have more favorable properties than linear peptides, and can reach a wider range of drug targets. Thus, we decided to focus on five BGCs producing cyclic peptides: Gacamide (Pseudomonas), Photoditritide (Xenorhabdus), Chaiyaphumine (Xenorhabdus), Szentiamide (Xenorhabdus), Xentrivalpeptide (Xenorhabdus).

From these five clusters we were unsuccessful in cloning the Gacamide cluster, and we could not observe photoditritide production in E. coli. This left us with the three clusters Chaiyaphumine, Szentiamide, Xentrivalpeptide (all from Xenorhabdus), that successfully produced the expected cyclic peptides in E. coli (Fig. 2). In particular, chaiyaphumine and xentrivalpeptide were produced in high titers, while the Szentiamide cluster showed lower production. This confirms that splitting clusters across three plasmids and reconstitution with inteins is a reliable strategy to simplify expression. At the same time, these results validated our backbone design as a suitable system for expressing large BGCs, enabling high production yields(See Library characterization)

🔍

Both Chaiyaphumine and Xentrivalpeptide produce two distinct derivatives.

DBTL Cycle

Cycle 1 - Introducing a new Golden Gate Standard

Design & Build

To start building up our platform (see Our Platform), we decided on derivatizing the NRP chaiyaphumine by modifying its NRPS. Specifically, our goal was to systematically swap the modules that are responsible for the incorporation of the second, third and fourth amino acid in the cyclic part of chaiyaphumine. We decided not to exchange the first and the last amino acids as they are involved in depsipeptide bond formation (Fig. 3).

🔍

It is crucial to distinguish between the peptide, protein, and DNA levels. The peptide is synthesized by the NRPS, which is encoded on a plasmid. To modify the peptide, NRPS modules must be exchanged at the DNA level (Fig. 4). This cloning-based process involves replacing defined fragments inside of the Chaiyaphumine coding sequence with heterologous DNA fragments from other NRPSs. These inserted fragments encode for individual NRPS modules and will be referred to as donor cassettes.

🔍

At first glance, scarless assembly strategies seem ideal for inserting fragments into the ORF, since scars in the DNA would be translated into amino acids scar of the protein. Yet in practice, these methods pose a major bottleneck: they lack standardization and modularity. Each donor cassette would require unique overhangs depending on both the cluster and the insertion site, resulting in extensive cloning efforts and making the platform impractical.

A Golden-Gate based strategy to circumvent scars, but still use standardized overhangs, was recently introduced for NRPS engineering purposes. This strategy relies on using the FFxxGGxS motif as the insertion site, which corresponds to XUT^{IV [3]}(NRPS and engineering). This motif is highly conserved across different classes of NRPS. The codon ambiguity of the two glycines was used to generate a set of standardized overhangs (A, B, C, D, E, and F), which upon assembly restore the motif. In this system a pair of overhangs define the position in the cluster (Fig. 5).

However this strategy comes with two major limitations: 1. The insertion site is limited to XUT^IV. 2. Donor modules are not modular. The position of the donor module in the NRPS is fixed by the overhangs flanking the donor cassette. As a result, exchanging the same donor module at three different positions requires three separate donor plasmids. This issue scales rapidly, since each additional donor multiplies the cloning effort. For example, creating a donor library covering all 20 proteinogenic L-amino acids already requires 60 donor plasmids (3 × 20). Expanding this to four positions would require 80 plasmids, and to five positions already 100. These numbers grow out of control, ultimately making the method impractical for large-scale diversification.

We set out to overcome this bottleneck by developing a new Golden Gate–based technology, which allows: 1. The flexible use of different engineering sites, such as XUT^I, XUT^VII, and XU. 2. The insertion of one donor cassette at any desired position within the NRPS, using one pair of overhangs for all positions. By combining Golden Gate cloning with intein technology, we avoided relying on overhangs to determine the position of donor cassettes. Instead, for each module targeted for exchange within the NRPS enzyme, we designed a separate acceptor plasmid in which the respective module was omitted. This allowed us to use the same overhangs, A and B. This modular and versatile strategy dramatically reduces the cloning effort and makes NRPS engineering more accessible.

Combining standardized overhangs with modular NRPS donors comes with the drawback of introducing assembly scars within the NRPS sequence. Our Golden Gate strategy deliberately avoids relying on conserved linker motifs. As a result, each assembly junction introduces an alanine and a serine in the resulting NRPS protein. We used the XUT^I site, which resides in the flexible, unstructured A-T linker region. It was shown that the insertion of additional three amino acids is tolerated by the NRPS^[6].

🔍

For our scar residues we chose alanine and serine instead of glycine, which is often used for linkers, since glycine at a position where it does not occur in the native protein can destabilize its backbone. Serine and alanine are good linker-scar residues because the –OH of serine supports solubility, while alanine has a neutral and small side chain that helps keeping the linker-scar compact^[7].

Nevertheless, introducing serine and alanine at this site could potentially interfere with function of the enzyme by changing the length and flexibility of A-T linker, increasing its rigidity, or even promoting the formation of secondary structures.

Introducing scar residues within the coding sequence becomes an even greater challenge when donor cassettes are inserted adjacent to a split intein, since this results in a combined scar of up to 10 amino acids: four from the Golden Gate assembly sites and six from the intein splicing reaction (Fig. 6).

🔍

Test

To experimentally assess the impact of the Golden Gate assembly scar on NRPS functionality, we introduced the two additional amino acids (Ala-Ser) into the respective A–T linker regions of the native, split chaiyaphumine synthetase. This design allowed us to simulate the outcome of a modular assembly. Each scar-containing plasmid was tested in combination with the corresponding two other native, scarless plasmids to isolate the effect of the insertion site. We evaluated the influence of the Ala-Ser scar at all three Golden Gate insertion sites (XUT2, XUT3, and XUT4), as well as in a construct containing scars at all three positions simultaneously. The production titers of these variants were compared to the split wild-type cluster, which served as the positive control (Fig.7) The expression cultures were done in triplicates. Gibson assembly was used to generate the constructs.

🔍

Learn

Comparative analysis of production levels revealed that only the XUT3 scar results in a production loss, with a statistical significance of p < 0.05. The scars in the XUT2 and XUT3 exchange site even showed higher production titers, compared to the wild type.

This trend may be explained by the scar length. The XUT3 module is flanked by two 10-amino-acid scars, originating from the overhangs A and B as well as the N-terminal gp481-8 and C-terminal NrdJ-1 inteins, potentially destabilizing the A-T domain-domain interaction. In contrast, the XUT2 and XUT4 modules are bordered by a shorter 2-amino-acid scar on one side and a single 10-amino-acid scar on the other.

Despite this reduction, the XUT3 variant still exhibited robust overall production levels. We therefore conclude that the introduction of the assembly scar does not critically impair NRPS functionality and remains well tolerated within engineered constructs, confirming the suitability of our Golden Gate-based approach for modular NRPS engineering.

Cycle 2 - Donor Module Selection

After validating that the scars in the A-T linker does not critically inhibit the functionality of the NRPS protein, we set out to establish our golden gate system, by generating a collection of donor plasmids and acceptor plasmids for Chaiyaphumine.

Design

In our Golden Gate design we distinguish between acceptor and donor plasmids. Each donor plasmids carries one donor cassette. Each cassette is flanked on both sides by inward-facing BsaI restriction sites, which generate the overhangs A and B, for the directional assembly into the acceptor plasmid. The acceptor plasmids carry a part of the splitted cluster, with a mCherry dropout cassette at the (BBa_25Q6N83H) position of the module to be exchanged (Fig. 8).

🔍

To select suitable overhangs for our Golden Gate system, we relied on the NEBridge Ligase Fidelity Viewer®. This tool is based on comprehensive experimental datasets in which all possible 4-bp DNA overhangs were systematically tested for ligation with T4 DNA ligase under controlled conditions. The tool predicts ligation fidelity and highlights risky combinations prone to mis-ligation. Although such predictions are approximate, since fidelity can depend on enzyme, buffer, and cycling conditions, they gave us the confidence to select two overhang sequences that are both robust and reliable^[8].

By introducing alanine and serine into the sequence, we were able to define the four base pair Golden Gate overhang within the corresponding six base pairs. The four-base overhang was designed to maximize the number of possible codon combinations. The four bases forming the overhang consisted of the last base of the first codon and the three bases of the second codon. This arrangement allowed us to keep the correct amino acids within the sequence while simultaneously using different codons for specific overhangs. In E. coli, alanine is encoded by four codons and serine by six. Therefore, alanine was positioned first and serine second, resulting in 24 potential four-base overhang combinations (Fig. 9). All 24 combinations were tested using the NEBridge Ligase Fidelity Viewer® to evaluate ligation fidelity. This analysis allowed us to define only those overhang pairs that were orthogonal and mutually compatible, thereby preventing cross-ligation during multi-fragment assembly.

🔍

In total, from the 24 possible combinations six distinct overhangs were defined, in addition to start and terminator positions. This configuration enables the directional assembly of up to seven modules on a single plasmid (SA, AB, BC, CD, DE, EF, FT). For the cloning of Chaiyaphumine, only two specific overhangs were required (A, B), but the complete set of combinations was defined by us to ensure future scalability and reuse in other engineering approaches that build on our technology.

We used the three plasmids making up the chayiaphumine cluster, to generate acceptor vectors for our golden gate system. The modules responsible for incorporating an amino acid at the three positions we wanted to exchange were removed. Instead of the NRPS module we introduced an mCherry dropout cassette. This allowed us easy by eye identification of a successful insertion of the donor cassette during cloning. Assuring an accessible cloning process. These plasmids serve as acceptor vectors (Fig. 10).

🔍

Build

To validate the functionality of our Golden Gate cloning system, we selected a set of donor modules. As described in the literature, a correlation can be observed between NRPS module compatibility and the phylogenetic distance between the species of origin. Modules derived from closely related organisms tend to exhibit higher functional compatibility, while increasing evolutionary divergence often leads to reduced interaction efficiency and product yield [5]. Therefore all donor modules we chose were derived from Xenorhabdus and Photorhabdus NRPS clusters. These modules corresponded to XUTI exchange units and were cloned into donor plasmids. As a proof of concept, these donor plasmids were subsequently used in Golden Gate assemblies with the middle Chaiyaphumine acceptor vector to verify that our modular cloning system enabled correct and efficient DNA fragment incorporation.

Test

We expressed the assembled hybrid NRPS in which the central module had been exchanged using our Golden Gate Intein system. The constructs were expressed and peptide production was analyzed by LC-MS (Measurement). As a positive control the original Chaiyaphumine sequence was reassembled using the same cloning workflow.

🔍

Learn

Through sequencing results and the functional positive control (Fig. 10, BBa_25UMSED) demonstrating that the system enables the functional exchange of NRPS modules at defined positions. When testing the donor modules, only three out of eleven constructs produced detectable peptide levels, and the titers were markedly low (Fig. 11, Fig 12). These results indicate that while the Golden Gate–Intein system is suitable for modular NRPS recombination, the success of such exchanges strongly depends on the compatibility between donor and acceptor modules. Consequently, a more systematic strategy for selecting compatible modules will be required to improve both the efficiency and the yield of engineered NRPS systems.

🔍

Cycle 3 - Phylogentic-based Donor Module Selection

Design

Building on insights from the previous cycle, we aimed to improve donor module selection. In contrast to our first approach, we used phylogenetic analysis to select donor modules from biosynthetic gene clusters, focusing on clusters phylogenetically related to the Chaiyaphumine NRPS.

For this purpose, phylogenetic trees were constructed based on thioesterase (TE) domain sequences. TE domains were chosen because they occur only once per NRPS and do not undergo duplication events, in contrast to adenylation or condensation domains^[9][10][11], which may lead to ambiguous mapping (Georg Hochberg).This approach was streamlined in one pipeline with our mATCHmaker software, making the approach ready-to-use for wet-lab application (link software). Using this TE-based approach allowed us to find NRPS clusters with close evolutionary proximity and to select donor modules from related clusters for the generation of NRPSieceS donor toolbox. The goal was to create a donor library covering both L- and D-amino acid specificities, thereby expanding the accessible chemical space of the resulting peptides.

Donor modules were preferentially chosen from clusters identified as closely related through a TE phylogenetic analysis. For amino acids not represented among these related clusters, additional donor modules were rationally selected according to the rules described in Phylogenetics .

Build

The selected donor modules were cloned into donor plasmids using Gibson assembly. Of the designed constructs, 35 were successfully assembled, forming the foundation of the NRPiecesS toolbox.

Subsequently, the donor plasmids were used in Golden Gate assemblies with all corresponding Chaiyaphumine acceptor vectors. In total, 35 starter, 35 elongation, and 35 termination modules were successfully cloned, resulting in a comprehensive collection of 105 plasmids, the NRPieceS expression plasmids.

Test

All donor modules were expressed in the Chaiyaphumine NRPS system to characterize functionality within the heterologous system. Each expression plasmid was individually tested with the two native chaiyaphumine units to evaluate peptide production and to verify the resulting products. Expression of the 105 newly assembled constructs led to the production of 63 distinct peptides.

Learn

The results revealed a clear correlation between module functionality and phylogenetic relation, confirming that evolutionary proximity elucidated by TE sequences shows strong correlation between donor and acceptor modules (See Library characterization)(Fig. 13).

🔍

Furthermore, the data indicated that successful incorporation of non-native modules was position-dependent, with higher success rates observed in the following order: XUT position 4 > 3 > 2, highlighting more flexibility towards the last position in the final peptide. This suggests that later positions within the assembly line are more tolerant to module exchange (Fig. 14).

🔍

Interestingly, when modules with an epimerization function were introduced after the threonine position, epimerization of threonine was observed, while the Chaiyaphumine TE domain remained capable of catalyzing cyclization. This indicates a catalytic flexibility regarding the configuration of this position.

Together, these findings demonstrate that phylogeny-guided module selection significantly improves the success rate of NRPS module exchanges and provides a rational framework for future NRPS engineering. This represents a substantial advancement in NRPS engineering, as it is, to the best of our knowledge, the first approach capable of predicting NRPS module compatibility. By providing a rational, predictive framework, it addresses one of the major bottlenecks in NRPS engineering and enables more efficient design of functional peptide synthetases.

Cycle 4 - Combination of Functionally Active Modules based on Bioactivity Testing

Following the workflow of the NRPieceS platform, we expressed the complete Golden Gate plasmid collection in combination with the native Chaiyaphumine cluster and identified which hybrid NRPS constructs produced detectable peptides. The produced NRPs were screened for antimicrobial activity against seven bacterial strains, including selected ESKAPE pathogens.

In the bioactivity assay the combination of the native Chaiyaphumine initiator and elongator with a terminator derivatized at XUT4 using a valine donor exhibited an inhibition zone against the high-sensitive strain Bacillus spizizenii (Fig. 15).

🔍

Design

Building on the observation that native Chaiyaphumine shows no effect against B. spizizenii while a derivatized variant produces an inhibitory zone, we concluded that the altered amino acid composition - specifically the exchange from proline to valine and the altered stereochemistry of the upstream amino acid - was responsible for this effect.

To assess whether additional modifications could further enhance activity, we aimed to generate Chaiyaphumine variants carrying derivatizations at two positions. Although three sites are theoretically accessible, each additional modification carries an increased risk of generating non-functional NRPS. Hence, we focused on double derivatization for a proof of concept.

Based on the results of our library expression, we selected initiator and elongator cassettes that had been characterized as functional to minimize the likelihood of unproductive constructs. In the experimental setup, the valine terminator was combined with seven additional derivatized modules at XUT2 and four XUT3, allowing us to systematically test the effect of multiple modifications while maintaining NRPS functionality(Fig. 16).

🔍

Build

To achieve our double derivatized NRPS hybrids we used intein shuffling. All required plasmids had previously been cloned as part of the library expression and are compatible with each other without further cloning effort due to the inteins.

Test

The selected combinations of the double derivatized NRPS were expressed heterologously and their peptide production was evaluated to verify abundance of the resulting products. Indeed, all eleven double derivatized peptides were detectable.

With all peptides available, we proceeded to evaluate their antimicrobial activity against a panel of bacterial strains. While the initial observation of an inhibitory zone against Bacillus spizizenii proved irreproducible, four compounds exhibited an inhibitory effects against Staphylococcus aureus and Enterococcus faecalis (Fig. 17).

🔍

Learn

The fact that the double derivatized peptides were detectable shows that the NRPieceS plasmid collection is a reliable tool to engineer NRPS not only at one but multiple positions.

In contrast to the hit against B. spizizenii, the double derivatives also showed to have an effect on two strains that were previously unaffected. Interestingly, all affected strains are gram-positive strains and nothing similar was observed with the gram-negative strains suggesting that the cell wall composition might pose a barrier for the peptides.

Cycle 5 Proposal - Training a Model to Predict Module Compatibility

The insights gained from previous iterations highlighted the need for additional tools to address the persistent challenge of unpredictable module compatibility in NRPS engineering.

In iteration 4, we observed that the TE sequence similarity score can provide a rationale for selecting donor modules for hybrid NRPS. However, high sequence similarity is just an indicator for the likelihood of module compatibility and does not guarantee that the resulting hybrid NRPS will function as intended. While there is a correlation between TE sequence similarity and compatibility, no fixed threshold exists below which module selection fails. In our data we observed that donors with less than 20% sequence similarity generally do not produce, but in our case have occasionally produced functional hybrids, too.

Another complication arises when no mapping is possible because the cluster lacks a TE domain. This underscores the complexity of predicting NRPS compatibility: At the core of NRPS peptide synthesis are the condensation complexes, which link amino acids into a growing peptide chain. In hybrid NRPS, these interfaces can be altered due to the introduction of foreign modules. Understanding their structural features is therefore key to explaining why some hybrid NRPS are functional while others fail to produce peptides.

At the core of NRPS peptide synthesis are the condensation complexes, which link amino acids into a growing peptide chain. In hybrid NRPS, these interfaces can be altered due to the introduction of foreign modules. Understanding their structural features is therefore key to explaining why some hybrid NRPS are functional while others fail to produce peptides.

Our software, mATChmaker, is designed to automate the prediction and annotation of condensation complexes and to store the resulting data in a database. Combined with phylogenetic information, this dataset could serve as the input for a predictive model capable of identifying previously unknown patterns in hybrid NRPS composition. Ultimately, this approach aims to address one of the greatest challenges in NRPS engineering.

AI or machine learning models in the future, can be helpful to detect underlying mechanisms, which determine module compatibility. Progress in AI development for NRPS design can benefit directly from our wet-lab results, as one of the main bottlenecks is the limited availability of high-quality, experimental data. We enable data generation in a high-throughput manner, our approach can help overcome this limitation and accelerate both discovery and optimization.