Skip to main content

Programming synthetic networks for bio-based production of value chemicals

Final Report Summary - PROMYS (Programming synthetic networks for bio-based production of value chemicals)

Executive Summary:
The 4 billion USD chemical industry is currently ongoing a long term transition towards greener and more sustainable production. A major factor limiting this transition is the development of stable high yielding cell factories. To address this challenge head on PROMYS was designed to develop the technology of ligand responsive regulation and selection systems to address the following three major bottlenecks:

- Synthetic pathway construction
- Cell factory optimization
- Cell factory stabilization

With regards to the development of the ligand responsive regulation and selection systems PROMYS has made substantial developments towards streamlining the development of both RNA and protein-based systems. These approaches have been applied to construct several new ligand responsive regulation and selection systems for a range of compounds relevant to the project.

Based on the developed ligand responsive regulation and selection systems PROMYS have discovered new enzymatic functionality for the targeted molecules, including a novel synthetic pathway for a confidential molecule to be commercialized by one of the SME partners and novel transporters identified from metagenomics libraries.

Based on the novel pathways and technology the partners of PROMYS have generated two cell factories. One for thiamine and one for the confidential molecule. Both cell factories are now world leading in terms of performance and shall be commercialized after project completion.

PROMYS then deployed the ligand responsive regulation and selection systems to develop synthetic circuits that stabilize cell factories in a high production state during timescales that are relevant for industrial fermentations. Importantly, PROMYS partners also identified genetic heterogeneity as an important limiting factor of the productivity of biobased processes further highlighting the relevance of the developed solutions.

In addition to delivering on the planned objectives PROMYS has also generated several new opportunities that have been assessed during the project. Indeed a dedicated business development officer hired in the last year of the project helped evaluate and further assist in the exploitation of these ideas. So far these efforts have led to the formation of one new spin out company with this sector.

Project Context and Objectives:
VISION AND IDEA:

The partners of PROMYS merged scientific and commercial aspirations with a clear vision to accelerate the bio-based revolution of the USD 4 trillion chemical industries (1). In fulfilling this ambition, PROMYS sought to develop, validate and implement a novel synthetic biology platform technology that drastically accelerates the construction, optimization and performance of cell factories by enabling industrial users to impose non-natural objectives on the engineered cell factory. Such non-natural objectives include overproduction of target chemicals at high yields and rates, a trait that normally conflicts with natural cellular objectives of survival and high growth rate. The PROMYS framework integrates the forward-engineering tools and concepts of synthetic biology for system design with identification of the most advantageous parameters through self-selective cycles of biological optimization. Our work program will enable the industrial user to re-engineer biological machinery for human defined objectives by overruling the natural inclinations of biological systems. Through its endeavours PROMYS has delivered innovative strategies that can be applied broadly within metabolic engineering with a strategic impact on industrial applications within the chemical industry.

To realize our goal we developed methods for the de novo design of what we term ligand responsive regulation and selection systems (Figure 1.1-1) which are robust biomolecular circuits for controlling cellular fate based on intracellular metabolic cues. Such synthetic cellular networks are used to impose non-natural selective pressure on the host cell. This selective pressure will be defined by the metabolic engineer to serve specific objectives of his/her application. Coupling selective pressures to the construction and optimization of metabolic processes is not novel; in fact this is the basis for the evolution of all biochemical processes within the natural world. The novelty of PROMYS results from integrating the principle of user defined self-selective cycles of biological optimization within the synthetic biology framework and applying it to solve major challenges within metabolic engineering.

CONCEPT AND BACKGROUND:

Our idea and platform technology relies on recent advances within the fields of RNA synthetic biology, transcription factor engineering, biosensors, de novo regulatory network construction, flux sensing and metabolic engineering. PROMYS brought together Partners that individually have made substantial contributions to each of these fields. Within PROMYS the Partners collaborate closely to integrate the advances of these fields and develop general strategies for the construction of ligand responsive regulation and selection systems that provide the mechanism by which non-natural and human-defined objectives are forced upon biological systems.

Ligand responsive regulation and selection systems (Figure 1) are biological devices that integrate biological sensing modules, within larger regulatory networks that control one or more cellular programs. Sensing modules include both synthetic riboswitches and transcription factors. Sensing modules integrated into more complex regulatory networks provide metabolic engineers with tight control of the execution of pre-defined cellular programs.

Over the past decade our capacity to assemble synthetic DNA and introduce combinatorial modifications to genomes has increased by several orders of magnitude with the advent of new synthetic biology methods. However, unlocking the full potential of these methods is limited by current screening strategies for discriminating between billions of modified strains based on analytical chemistry or fluorescent assays. Hence, the novel synthetic biology methods to construct diverse cell factory libraries need to be complemented by in vivo selective methods capable of discriminating between vast numbers of synthetic constructs. Specifically, PROMYS has used ligand responsive regulation and selection systems to solve the needle in a haystack problem, that arises when synthetic biology methods are used to construct cell factory libraries with billions of variants, of which, only a few fulfil the design objectives.

In addition to the above-mentioned technological opportunity, an important driver for the envisioned concept is the significant market demand: The bio-based segment of the chemical industry currently constitutes a USD 90 billion market, projected to reach USD 180 billion by 2020 (2). Thus, innovative technologies enabling this transition will be of substantial commercial value. A successful example of a bio-based process is the production of propanediol, by Dupont and Tate & Lyle, currently produced in quantities exceeding 120.000 tonnes per year. The development of this cell factory and process required R&D investments in excess of USD 100 million and took close to a decade. Such timelines and budgets are prohibitive for a broad transition of the chemical industry and thus the development cost and time of bio-based processes must be reduced.

To address the clear market need, PROMYS exploited the technological opportunity to develop and apply ligand responsive selection systems to metabolic engineering. Specifically, PROMYS addressed three major challenges in metabolic engineering that limit the development of new cell factories (Figure 2):
1) Synthetic pathway construction
2) Cell factory optimization
3) Control of populations during fermentation

Ligand responsive regulation and selection systems directly couples the presence of a desired chemical product or flux state within a cell, to the survival of the cell. These systems have the potential to radically change the approaches to pathway construction and cell factory optimization, since they allow the in vivo identification of the needle (e.g. functional pathway or optimized cell factories) in a haystack (e.g. large libraries). In this way ligand responsive systems enables industrial users to harness the full potential of synthetic biology methods for the synthesis and engineering of pathways and genomes. In addition, the technology developed in PROMYS can be applied to improve fermentation yields by creating ligand responsive systems that counteract the natural tendency of cell factories to reduce overproduction of non-essential compounds and pathways in favour of survival and growth during fermentation. Such systems will lead to increased fermentation yields by continuously selecting for high yielding cell factories within the fermentation population.

The concept of ligand responsive selection systems that will be developed in PROMYS is general and the approaches will be applicable to cell factories of any biological chassis. To demonstrate the concept, PROMYS has focused on the application of these systems in Escherichia coli and Saccharomyces cerevisiae, since synthetic biology methods are well developed for these organisms and they are frequently used as cell factories by the chemical industry. Similarly, while the concept can be applied to almost any chemical, PROMYS focused mainly on four chemicals: Thiamine, mevalonate xanthine and a confidential compound from an SME partner of PROMYS.

Exploitation and commercial application of the technologies developed in PROMYS will be led by three European synthetic biology SMEs, that all fulfilled major roles in the research project together with the 4 academic partners. Furthermore, enabling technologies and competences developed within PROMYS would provide a strategic advantage for the European chemical industry and research communities within this important growth area.


(1) UNEP - United Nations Environment Programme (2012) Global Chemicals Outlook: Towards Sound Management of Chemicals.
(2) Frost & Sullivan (2011) A Bio-based Future for the Chemicals and Materials Market, Global Analysis of Opportunities for the
Chemicals and Materials Industry.


Project Results:
PROMYS has been industry driven and focused on the development of innovative approaches for the chemical industry, with a goal of subsequent commercialization. Accordingly, PROMYS was structured around three major scientific and technological challenges limiting the bio-based transition of the chemical industry:

1) Synthetic pathway construction
2) Cell factory optimization
3) Control of populations during fermentation

The overall progress within each of these 3 areas is summarized in the following in line with our original proposed aims. However, as a foundation a 4th objective was articulated in the original proposal and addressed in the project, related to the foundational technology required for addressing the first three problems.

1. LIGAND RESPONSIVE REGULATION AND SELECTION SYSTEMS

PROMYS objective: Development of general methodology for construction of ligand responsive regulation and selection systems.

Ligand responsive regulation and selection systems can be broadly divided into those systems that depend on RNA or protein based regulators. In PROMYS both strategies were addressed and we summarize below the main acheivements of the project.

1.1 CHARACTERIZATION OF SMALL MOLECULE BINDING APTAMERS AND ASSESSMENT OF THEIR SUITABILITY AS SENSING DOMAINS IN RIBOSWITCHES

Elucidation of aptamer properties required for riboswitch construction is a major challenge. A comprehensive list of known small molecule binding aptamers have been compiled. A literature screening based on various inherent properties of the aptamer sequence and the corresponding ligand as well as some general considerations yielded several subsets of aptamers that were subsequently tested for regulatory activity in vivo. The study revealed that only a small subset of aptamers harbor regulating properties. The characterization of the biophysical properties of such regulating aptamers clearly indicated that regulating aptamers have to possess two properties, a high affinity ligand binding and significant conformational changes and stabilization upon ligand binding. Based on these results a method for the identification of regulating aptamers has been established (3). This method was applied to identify aptamers with riboswitching properties and for the riboswitch development required for the new confidential molecule of the project. Once suitable aptamers where identified, they were subcloned into in vivo expression platform (E. coli or yeast), subjected to enrichment in vivo library using fluorescence activated cell sorting (FACS), in vivo characterization of functional riboswitches in microtiter plates, biochemical characterization and determination of binding affinity using MST, in silico RNA folding characterization. In this way successful riboswitches were constructed for several different molecules during the course of the project.

1.2 DEVELOPMENT OF A GLYCOTIC FLUX SENSORS

We exploited the fact that in cells the levels of the metabolite fructose-1,6-bisphosphate correlates with the flux through glycolysis. Here, we developed RNA- and protein-based systems that respond to FBP and thus the glycolytic flux.

First, harnessing the FBP-binding transcription factor Cra in E. coli, we developed a reporter system that can be used with flow cytometry to screen for cells with altered glycolytic flux (4). As yeast does not have an endogenous FBP-binding transcription factor, we transplanted the FBP-binding transcription factor CggR from B. subtilis into S. cerevisiae. We used CggR, because our biochemical studies showed that Cra in fact only binds F1B and not FBP. We engineered a synthetic CggR binding promoter, expressed CggR from different promoters (CMV, TEF2, TEF7), tested the CggR expression levels with proteomics, and tested the FBP- and flux-response of the engineered system on different nutrients, by means of CggR-regulated YFP expression versus constitutively expression of RFP. To modulate the FBP-affinity of CggR, we constructed a library of mutant CggR by site-directed mutagenesis, identified through computational protein design. Together, this provided us with a novel synthetic systems that responds to the glycolytic flux in yeast.

In parallel, we devised an RNA-based system responding to FBP level. Based on a SELEX-selected aptamer, we designed a plasmid-encoded synthetic system for ligand-dependent gene expression control in yeast. Specifically, we inserted the aptamer next to GFP with the purpose to couple FBP-binding to the aptamer to the expression of the GFP. To this aim, we used two different strategies (synthetic riboswitch and aptazyme) and sought for FBP-regulated GFP expression at flux-changing growth conditions, using a constitutively expressed RFP as control. For the synthetic riboswitch we inserted the aptamer in the 5’UTR of GFP and for the aptazyme we grafted the aptamer into one of the stem loops of a hammerhead ribozyme cloned in the 3' end of GFP. Then, we generated libraries by randomizing the aptamer-flanking sequences within GFP 5’UTR (riboswitch) and the sequence of the opposite loop of the ribozyme (aptazyme), and screened for variants showing a switching behavior (glycolytic flux-dependent GFP expression). Variants were sorted into 6 subpopulations based on GFP/RFP expression ratio using FACS. We identified all the variants in each subpopulation by NGS. Analysis of the sequencing data revealed promising sequences, which are being cloned and subjected to final tests.

1.3 DEVELOPMENT OF RIBOSWITCH FOR CONFIDENTIAL MOLECULE

To further apply the insights to a challenging case within the project we selected aptamers which specifically bind to the confidential ligands. In order to improve the specificity of the aptamers, a new selection method called Capture-SELEX was developed. After 15 rounds of selection, the aptamers specifically bind to the target ligand but not to closely related compounds. Using a ribozyme-based screening, a riboswitch for the confidential molecule has been identified which respond to the ligand with a 10-fold dynamic range.

1. 4 COMPUTATIONAL DESIGN OG REGAZYMES

We developed a computational design software that can be used to model intra- and inter-molecular secondary structure changes within and between RNA molecules, and we have designed a novel RNA system termed the Regazyme that takes an existing theophylline aptamer and combines it with a riboregulator acting as a signal mediator allowing us to produce cascades of multiple small-molecule/RNA inputs. We aimed to design switchable functional RNA domains by exploiting strand-displacement principles to generate cascades. We generated a model based around a standard physiochemical model predicting RNA secondary structure and free energy to be used in an optimization algorithm to select for the hierarchical activation of functional RNA modules in the cascade. The Regazyme consists of a theophylline aptazyme, that controls the cutting and release of a small RNA riboregulator, that then binds to a cis-repressed RBS controlling GFP expression (Figure 3A). We constructed a combinatorial optimization problem (Figure 3B) to explore the sequence space of the transducer module, where a nucleotide-level energy model considering the conformational states (uncleaved and cleaved) of the Regazyme was used to evaluate the performance of the generated sequences. For each state, the model accounts for its free energy and its secondary structure. As objectives to be optimized (computed as Hamming distances), the algorithm considers the energy of activation corresponding to the catalytic activity of the aptazyme (which we assume depends on the correct formation of the aptamer in the uncleaved state), and the degree of exposure to the solvent of the riboregulator seed before and after cleavage. In absence of signal molecule, the progression of the reaction is limited by the presence of a high-energy intermediate that prevents the interaction between the Regazyme and the 5’ UTR (Figure 3C). However, when the signal molecule is at sufficient concentration, a cleavage is produced and then the activation energy for the resulting riboregulatory element is lowered, which speeds up the reaction. As shown by a random sampling of 1,000 sequences (Figure 3D), an optimal score (zero, as our score is considered as a penalty) is very unlikely to be obtained arbitrarily. This means that this is a difficult design problem for a manual approach, requiring automated computation for efficient sequence design. Our algorithm designs by optimization the sequences implementing the intended signal transduction according to the objective function. The designed system was characterized experimentally and found to be functional in vivo.

2. SYNTHETIC PATHWAY CONSTRUCTION

Challenge: For the development of novel bioprocesses novel pathways need to be constructed. The identification of these parts (genes) is time consuming and relies typically on iteration between construction of a limited number of pathways followed by analytical screening. The limitations of this approach significantly contributes to the long development time of cell factories as well as limits the spectrum of chemical products for which biological production processes are pursued.

Approach: PROMYS will address this challenge by substituting the labour intensive and time-consuming analytical screening with ligand responsive selection systems that couple the intracellular presence of the desired chemical product to the survival of the cell. Using this technology millions of combinatorially designed synthetic pathway constructs can be tested in days compared to years using conventional analytical methods.

It is well know that the vast majority of biological function in microbial communities is not accessible using cultivation based techniques. Instead metagenomic approaches have been used to access this diversity. During the project we chose to apply the ligand response regulation and selection systems to the complicated task of metagenomic mining.

2.1 DEVELOPMENT OF XANTHINE ALKALOID SELECTION SYSTEM

We developed a xanthine alkaloid responsive strain of Escherichia coli by using the synthetic theophylline riboswitch (5) as an intracellular biosensor and regulator of antibiotic resistance. The theophylline riboswitch was developed based on the synthetic mTCT8-4 aptamer for specific binding of theophylline and related xanthine alkaloids (6) and has previously been implemented and used for theophylline-dependent chloramphenicol resistance in E. coli (5). We tested this system for theophylline conditional growth and found it to exhibit a high escape rate, i.e. false positive growth was observed under selective conditions in the absence of theophylline. For a more robust triggering of survival we constructed a new selection system using an enhanced variant of the theophylline riboswitch (variant 12.1) which has a reported ON/OFF ratio of ~90 (7). To counteract false positive growth, we coupled this riboswitch to two different antibiotic resistance markers such that binding of theophylline is required by two riboswitches, which individually induce resistance to chloramphenicol and spectinomycin respectively (fig. 4A). This dual selection architecture was adopted from a previous riboswitch-based functional selection system reported by our group, which proved highly successful in reducing false positive growth for riboswitch-based functional selections. When transformed in E. coli, the resulting strain displayed theophylline dependent growth in media containing chloramphenicol and spectinomycin (fig. 4B and C). The dual selection (using chloramphenicol and spectinomycin in combination) allowed plating of up to 108 cells on an agar plate in absence of inducer without appearance of any falsely growing colonies in 3 days (fig. 4D). In contrast, using single selection (chloramphenicol or spectinomycin alone) resulted in a bacterial lawn at high cell densities (fig. 4D), demonstrating the effect of the dual selection for reducing false positives. Characterization of colony formation and size across a gradient of theophylline concentrations indicated a ligand dynamic range of extracellular theophylline ranging from approximately 50 µM to 250 µM (fig. 4C).

2.2 INTEROGATION OF METAGONOMIC LIBRARIES FOR NEW ENZYMATIC FUNCTIONS

To assess the functionality and capability of the ligand responsive systems we used the synthetic selection system developed for xanthine alkaloids to select for metagenomics inserts that allowed for import of xanthine and other derivatives (Figure 5a-b). Using this approach we successfully identified several inserts that encoded xanthine transporters. Two of such inserts were further validated using growth assays and LC/MS (Figure 5c-d).

Additionally, we have applied a similar approach to thiamine and identified a wide range of novel enzymatic functionality from complex metagenomic libraries. In particular these genes include a novel family of thiamine transporters (Figure 6).

2.3 PATHWAY DISCOVERY FOR CONFIDENTIAL MOLECULE

During the project a significant effort was dedicated towards the discovery and establishment of a synthetic pathway for the confidential molecule of PROMYS. During the project several different approaches and biodiversity sources were tested ultimately leading to the successful identification of a synthetic pathway leading to the desired molecule produced in yeast.

3. CELL FACTORY OPTIMIZATION

Challenge: Once functional synthetic pathways have been constructed, they must be introduced and optimized in their host organism to create a consolidated cell factory. Substantial engineering of both the synthetic pathway and the host organism is required to identify cellular solutions that lead to competitive yields at high rates. This optimization effort is also limited by our current ability to separate highly yielding cell factories from the rest, since we again largely rely on analytical screening to make this distinction. As a result, our approaches to cell factory optimization are currently constrained, limiting the benefit we can derive from new synthetic biology methods.

A goal for PROMYS was to develop ligand responsive regulation and selection systems for the optimization of cell factories. Here we pursued a strategy using ligand responsive systems developed for thiamine. Our vision for synthetic pathway construction involves the rational and targeted combinatorial design of biosynthetic pathways using cutting edge DNA combinatorial technologies. However, unlocking the full potential of these new synthetic biology approaches is limited by current ex vivo strategies for discriminating between constructs based on analytical or fluorescent assays. Ligand responsive selection systems offer such a solution to the current limitations of synthetic biology. Within the field of pathway construction ligand responsive selection systems promise to drastically increase the throughput by which constructs can be evaluated, which was demonstrated in PROMYS for thiamine pyrophosphate (TPP).

3.1 CONSTRUCTION OF TTP COMBINATORIAL PATHWAY EXPRESSION LIBRARY

The 11 genes that are required for de novo TPP biosynthesis are distributed across 6 loci in the E. coli genome (Fig. 7a and 7b). Seven of the genes are organized in two operons, thiMD and thiCEFSGH, which are tightly feedback-regulated at the translational level by TPP-specific riboswitches (7). Of these thiM is required for THZ-P salvage and is therefore not considered necessary for de novo synthesis. To investigate the phenotypic consequences of overexpression of the TPP pathway, and to identify pathway variants leading to high TPP titers, we focused on the seven regulated genes exclusively involved in TPP biosynthesis (thiC, thiE, thiD, thiF, thiS, thiG and thiH). Genes that are not under feedback control or that also participate in other metabolic processes (thiL, thiI, iscS, and dxs) were excluded from this study. Overexpression of the seven genes was performed from a plasmid containing two operons (thiCED and thiFSGH) each transcribed by a strong constitutive promoter (Figure 7c). Organization of the genes in two operons was chosen to recapitulate the two biosynthetic branches. To vary gene expression levels, each gene was expressed by one of four synthetic ribosome binding site (RBS) sequences designed to have strengths ranging from weak to strong as predicted using the RBS Calculator (8, 9) (Figure7d). With seven genes, each expressed at four different levels, the total library size containing all possible RBS-gene variations is 47 = 16,384.
To construct the TPP pathway expression library we first developed a generic pathway-assembly and expression platform designed to facilitate large-scale combinatorial assembly of multi-gene pathways for overexpression in E. coli . Our system is based on the combinatorial and hierarchical assembly of metabolic pathways in vitro facilitated by Gibson (10) or USER cloning (11, 12). First the genetic parts to be assembled are prepared with the appropriate linker-overhangs to facilitate combinatorial operon assembly into plasmids containing a uniquely designed expression cassette . The cistron variants of the individual plasmids are subsequently combinatorially assembled by Gibson cloning supported by pre-designed linker overhangs. Briefly, the open reading frames (ORFs) of thiC, thiE, thiD, thiF, thiS, thiG, and thiH were amplified from genomic DNA by PCR (Figure 7c). Each gene was amplified four times using primers for the introduction of the four synthetic RBS sequences and appropriate linkers in the 5’ and 3’ ends to facilitate coordinated and directional assembly. First, the thiCED and thiFSGH cistrons were combinatorially constructed resulting in 64 and 256 cistron combinatorial variants respectively. Next, the two-cistron libraries were combined resulting in all 16,384 combinatorial pathway expression combinations (Figure 7c).

3.2 TTP RIBOSWTICH-BASED FUNCTIONAL SELECTIONS

The full combinatorial library was transformed into E. coli cells harboring the TPP riboswitch-based selection system. While the intracellular concentration of wild-type E. coli is ~1 μM TPP, induction of antibiotic resistance to chloramphenicol (20-40 μg/mL) and spectinomycin (50 μg/mL) using this system requires ~>5 μM intracellular TPP (13). Antibiotic-based selections should therefore enable the in vivo functional identification and isolation of constructed pathway variants that result in at least 5 fold higher intracellular TPP. Following transformation by electroporation and subsequent propagation of the combinatorial library, the cells were plated on selective agar plates to isolate overproducing clones (Figure 8). Colonies appeared for the combinatorial TPP overexpression pathway library, while no colonies were observed when the selection strain transformed with an empty vector was plated as a control (> 107 cells). Based on the number of colonies that emerged at selective and non-selective conditions, we estimate that the functional selection frequency was 1 in approximately 200 cells, indicating that TPP-overproducing phenotypes constitute approximately 0.5 % of the constructed variants in the library. To investigate genotype - phenotypic relationships of individual clones, we picked 42 colonies from selective conditions and 31 colonies from non-selective conditions.

For each clone we sequenced the TPP pathway (≥ 100 bp upstream and ≥ 400 bp downstream of each synthetic RBS), measured growth rate, determined protein expression levels of the seven refactored TPP pathway enzymes by mass spectrometry (MS), and quantified intra- and extracellular levels of thiamine, TMP and TPP using high-pressure liquid chromatography (HPLC) (Error! Reference source not found.c). An overview of the functional characterization of individual isolates is presented in Figure 9.

3.3 44-FOLD IMPROVEMENT IN A SINGLE ROUND OF COMBINATORIAL REFACTORING

HPLC measurements of thiamine, TMP and TPP concentrations in triplicate biological replicates, showed that the wild-type strain (E. coli DH10B harboring an empty expression vector) produces no detectable thiamine or TMP, but 1.1 ± 0.2 μM (mean ± standard deviation) intracellular TPP (cell specific) following 24 h growth. To confirm the functionality of the TPP selection system, we measured thiamine, TMP and TPP in all of the 73 clones picked from selective and non-selective conditions. We detected significantly higher intracellular TPP in all selected clones when compared to the wild-type strain with levels ranging from 4.5 μM (TS11H4) to 18.8 μM (TS11A2) with a mean of 10.9 ± 4.0 μM in all 42 clones. In contrast, the majority of the clones picked from the non-selective conditions produce TPP levels that are similar or only slightly higher than the wild-type strain, with a mean intracellular TPP concentration for all 31 non-selected clones of 2.4 ± 2.7 μM. By performing an ANOVA analysis of variance we found the difference in productivity of the non-selected clones to be to be highly significant (p value = 4E-15). As expected, a few non-selected clones did produce significantly more than the wild-type strain e.g. TS10G4, TS10B5, and TS10H4 with 11.5 10.6 and 7.3 μM intracellular TPP respectively. As a general rule, we observed large differences in intracellular TPP between clones, suggesting that our strategy to vary RBS sequences resulted in diverse balancing of the TPP pathway.

A comprehensive analysis of all thiamine moieties showed that the wild-type strain produces a total of 74 ± 18 nM TPP when considering intra and extra cellular fractions, which corresponds to ~32 μg/L. In contrast, the best performing strain (TS11A2) produces 3670 ± 150 nM total thiamines corresponding to 1.41 ± 0.15 mg/L. Overall, this corresponds to an impressive 44 fold improvement in thiamine titer as compared to the wild-type strain after a single round of strain engineering (Table 1). Previous investigations of TPP biosynthesis led to identification of a mutant strain of E. coli (E. coli PT-R1) (15) that was reported to accumulate 50 to 90 μg/L (120 to 210 nM TPP) (14) corresponding to a ~4-fold increase compared to E. coli DH10B as measured in this study. This strain was later found to contain a disruptive mutation in the riboswitch that feedback-regulates the thiCEFSGH operon (17). Efforts to construct cell factories overproducing thiamine have previously been based on the use of Bacillus subtilis as a host strain (16), and the state-of-the-art production data that is publically available is summarized in Error! Reference source not found.. By a combination of operon deregulation, enzyme overexpression, and engineering of transport, Schyns and co-workers achieved titers of 1.31 mg/L total thiamines (126 μg/L/OD600) (intra- + extracellular) with a strain of B. subtilis following 24 h cultures in shake flasks (16). The best strain isolated in this study (TS11A2) produces 170 μg/L/OD600 (Table 1) – a significant performance-leap in a single round of engineering, considering that only the core pathway has been modified in contrast to the wider engineering efforts in B. subtilis (16). Accordingly, the technology developed in PROMYS has shown the potential to accelerate and improve cell factory optimization.

3.4 CELL FACTORY OPTIMIZATION FOR CONFIDENTIAL MOLECULE

Following identification of the synthetic pathway for the confidential molecule PROMYS focused on the development of a cell factory yielding commercially relevant titers. This was achieved successfully towards the end of the project.

4.0 CONTROL OF POPULATIONS DURING FERMENTATION

Challenge: In fermentation processes, the engineered cell factories typically show reduced productivity due to the rapid appearance of non-optimal phenotypes in the fermentation culture. This results from the divergence between the objectives of the fermentation process and the selection pressure for increased growth rate of clones within the cell factory population. From an evolutionary perspective the overproduction of a chemical product is not desired and as a result even engineered cells will evolve away from the production objective and towards a state of lower production and higher growth rate.

Approach: PROMYS will address this challenge by integrating ligand responsive selection systems into the cell factory that sense the cellular state and couple the output of such circuits to cellular programs that either fix such stresses in real time or destroy individual cells that do not fulfil production objectives. In this way the fermentation population can be maintained in the desired state of high productivity resulting in higher fermentation yields.

Engineered cell factories are challenged with a wide variety of metabolic stresses and toxicities, which altogether contests synthetic cell performance. Especially when cultured for a prolonged cultivation span equivalent of large-scale biological production, we hypothesized that cells may deviate from the engineered high-level production. In PROMYS ligand-responsive networks have been developed, which allow highly selective identification of specific cells from large libraries. Here the knowledge from integrating ligand signals into cells was implemented to to dynamically control fermentations and maintain attractive cell factory behavior. Given the possible evolutionary selections that work against the stability of long-term, we first set out to test the possibility for genetic heterogeneity in such cultures and map the most important genetic modes to be controlled using ligand-responsive networks in long-term fermentations.

4.1 CHARACTERIZATION OF GENETIC HETOROGENEITY DURING FERMENTATIONS

Mevalonic acid is a precursor to the important secondary metabolite class of isoprenoids, acting as a chemical building block for colorants, medicines, flavors, fuels and fragrances (18) and therefore constitutes an interesting case fermentation product. We therefore wanted to study the phenotypic dynamics of high-level mevalonic acid-producing E. coli over industrially relevant time-scales. Inoculation of large fermenters typically involves gradual scale-up from an aliquot of a master cell bank by serial growth in vessels of increasing volume (19). During these cultivations, the original clone, giving rise to the master cell bank aliquots, proliferates through > 60 cell generations. To experimentally simulate this growth process, we serially transferred production strain lineages every eight hours for a total of nine times, corresponding to approximately 80 cell divisions (generations) (Fig. 10a). Specifically, we cultured five parallel lineages of an E. coli TOP10 clone harboring an induced mevalonic acid pathway plasmid (pMevT) maintained under constant antibiotic selection to prevent plasmid loss. To analyze phenotypic and genetic population dynamics, we sampled and freeze-stocked the growing populations every eight hours.

To assess the dynamics of the population fitness during the experiment, we evaluated population growth rates. The average population growth rate gradually increased as a function of generation number, following a sigmoidal pattern that stabilized at a new level after 60-75 generations (Fig. 10b). The population growth rate at the beginning of the experiment was 28 % below the final population growth rate, highlighting a considerable change in fitness of the simulated fermentation populations (Fig. 10b). Next, we determined the mevalonic acid titer of each sampled population throughout the simulated fermentation (Fig. 10c). Starting from generation 34, product titers began a decline by several percent per generation before leveling off at undetectable concentrations around generation 70. The onset of the decline in mevalonic acid production coincided with the increase in population growth rate and followed an inversely proportional pattern to the increased growth rate. To test whether the production plasmid from the evolved populations still conferred mevalonic acid production to a non-evolved host plasmid populations were extracted from the five end-points and re-introduced into fresh E. coli TOP10 strains. The transformed cultures did not show any detectable mevalonic acid production, demonstrating that the mevalonic acid pathway had been disrupted to incapacitate its biosynthetic potential. Additionally, we found no SNPs in the genomes of nine randomly selected colonies from the end-point populations relative to the ancestral strain.

To investigate the genetic basis of production failure at the population level, we ultra-deep sequenced the heterologous mevalonic acid biosynthetic pathway from three lineages at five sampling points during the experimentally simulated fermentation, and all five at the generation 70 end-point (paired-end 2 x 150bp Illumina sequencing at average depth of 7,200 X). To quantify disruption dynamics, we tracked the fraction of position-specific coverage relative to corresponding coverage of non-disrupted reference sequences (Fig. 11a). We generally detected the position-specific presence of such disruptions in the pathway populations at frequencies down to 0.04 %. We found that during the experimentally simulated fermentation, six specific positions in atoB and ERG13 of the metabolic pathway were disrupted by IS10, IS186 and IS5 insertions, jointly constituting > 91 % at the end-points, generation 70 (Fig. 11a and 11b). While the atoB and ERG13 pathway genes became increasingly disrupted during the experiment, the final pathway gene tHMGR remained free of disruptions (Fig. 11a and 11b). This striking degree of preservation of the tHMGR gene is probably due to the cytotoxicity of HMG-CoA (22), the substrate of tHMGR. Spontaneous mobile element disruptions of tHMGR likely became toxic in cells with active atoB and ERG13, as these cells would accumulate cytotoxic HMG-CoA concentrations. Because several atoB disruptions were also enriched despite the presence of a chromosomal copy, it is very likely that enriched insertions within atoB also abolish ERG13 activity by means of IS-mediated transcriptional termination (20) owing to the operon structure of the mevalonic acid biosynthetic pathway.
Given that complete production loss was observed in the populations, we speculated that other mobile elements could explain the remaining 9 % fraction. As a strategy to fully resolve the population reads, we mapped all reads to the 24 unique mobile element subgroups in the E. coli DH10B genome (21) i.e. not detecting for loci-specific dynamics. We found that joint mobile element coverage relative to the original pMevT approached 99.9 % (s.e.m. = 1.1 %) at generation 70 (Fig. 11c). A spectrum of ten host mobile element subgroups each transposed to a frequency above 0.01 % in the end-point populations (Fig. 11d).

4.2 LIGAND RESPONSIVE REGULATION AND SELECTION SYSTEM FOR CONTROLLING FERMENTATIONS

Given the observed wide evolutionary pressure on faithful long-term fermentation observed (Fig. 10 and Fig 11.), we considered mevalonate fermentation as a relevant case to implement a ligand-responsive network. By employing prototrophic or antibiotic selection genes, a biosensor had previously been used in a production strain to enrich a phenotypically high-performing subpopulation (23). However, such medium-conditional selection genes are at risk for cross-cell leakage of the conditional nutrients or antibiotics-degrading enzymes, and constrain the medium composition in ways that can be prohibitive for large-scale fermentations. As an approach to synthetically limit the abundance of non-producing mutant cells in a population, we proposed a fermentation control system that punishes non-producing mutants without depending on conditioned medium. Conditioned growth medium makes engineering easier, but is poorly compatible with large-scale industrial fermentations (24). After evaluating toxin-antitoxin pairs, we pursued a system that could harness the regulation of essential endogenous cell processes to punish non-producing cells in the population.

To characterize the performance of our mevalonic acid fermentation controlled strain, we experimentally simulated durations of scale-up from master cell bank to large production bioreactors (>200 m3 volume). We used a concept of serial passaging (25) and compared the controlled e3.9 strain to a non-controlled strain pe1 with wildtype folP-glmM promoter. To avoid stationary-phase cultures not commonly desired in industrial processes, we passaged the four parallel lineages of each strain in exponential phase strictly every 16 hours for a total of 14 times, while freeze-storing samples for subsequent analysis. Towards the final passages, pe1 lineages accumulated slightly more biomass per passage, indicating a gain of fitness that was absent for the product-addicted strain e3.9. To investigate these dynamics, we quantified the population maximum growth rates over the long-term experiment. All pe1 lineages gradually increased maximum growth rates until they reached a new stable plateau of nearly double maximum growth rates around generations 80 (Fig. 13A). This new fitness state could result from a production load alleviated by subpopulations of detrimental genetic variants as previously described in long-term cultivations.

Next, we re-cultivated the stocked cell samples to investigate the mevalonic acid production dynamics over the long-term experiment. Initial production in product-addicted and non-addicted strains were equal at respectively 2.9 and 3.1 g/L mevalonic acid (Fig. 12B and D), corresponding to a yield of 0.38 g/g glucose (70 % of theoretical max). However, in accordance with their rising growth rates during the long-term experiment (Fig. 12A), the pe1 lineages gradually lost mevalonic acid production (Fig. 12B), likely due to selective enrichment of non-producing cells in the populations (25). Overall, these production declines corresponded to a half-life of the population productivity of 50 generations. In all four non-addicted lineages, mevalonic acid production ultimately fell to below 5% within the accumulated 95 generations studied. In contrast, the four product-addicted lineages of e3.9 retained their initial low growth rates and effectively endured 95 generations without statistically significant improvements of max. growth rates (Fig. 12C). Coherent with not gaining fitness, the four product-addicted lineages remained >95 % productive at the end of the large-scale simulated fermentation at generations 95 (Fig. 12D). Thus, these addicted populations maintained the functional metabolic pathway for a significantly prolonged cultivation period, postponing the beginning of a decline from 50 to at least 85 generations (Fig. 12B and D). This gain corresponds to a massive increase in terms of functional working volume due to exponential growth: 60 generations of cultivation would match the cell divisions required to saturate a 200 m3 bioreactor at high cell density. This evasion of otherwise evolving production declines strongly indicates that the fermentation control design functioned in the manner intended. It synthetically confined the cells to the otherwise unfavorable state of high-yield metabolite production (26) .

(3) Schneider C, Suess B (2015)“Indentification of RNA aptamers with riboswitching properties”, Methods, DOI: 10.1016/j.ymeth.2015.12.001
(4) Lehning CE, Siedler S, Ellabaan MHH, Sommer MOA (2017 “Assessing glycotic flux alterations resulting from genetic perturbationsn in E. coli using a biosensor” Metabolic Engineering, DOI: 10.1016/*j.ymben.2017.07.002
(5) Desai SK, Gallivan JP (2004) “Genetic screens and selections for small molecules based on a synthetic riboswitch that activates protein translation, J.Am. Chem. Soc., DOI: 10.1021/ja048634j
(6) Jenison R, Gill S, Pardi A, Polisky B (1994) “High-resolution molecular discreimination by RNA”, Science DOI: 10.1126/science.7510417
(7) Winkler W, Nahvi A, Breaker RR (2002) “Thiamine derivatives bind messenger RNAs directly to regulate bacterial gene expression”, Nature DOI: 10.1038/nature01145
(8) Salis HM, Mersky EA, Voigh CA (2009) “Automated design of synthetic ribosome binding sites to control protein expression”, Nature Biotechnology, DOI: 10.1038/nbt.1568
(9) Salis HM (2011) “The ribosome binding site calculator”, Methods in Enzymology, DOI: 10.1016/B978-0-12-385120-8.00002-4
(10) Gibson DG (2011) “Enzymatic assembly of overlapping DNA fragments”, Methods in Enzymology, DOI: 10.1016/B978-0-12-385120-8.00015-2
(11) Bitinaite J, Nichols NM (2009) “DNA cloning and engineering by uracil excision”, Current Protocols in Molecular Biology, DOI: 10.1002/0471142727.mb0321s86
(12) Genee HJ, Bonde MT, Bagger FO, Jespersen JB, Sommer MOA, Wernersson R, Olsen LR (2014) “ Software-supported USER cloning strategies for site-directed mutagenesis and DNA assembly” ACS Synthetic Biology, DOI: 10.1021/sb500194z.
(13) GeneeHJ, Bali AP, Petersen SD, Bonde MT, Gronenberg LS, Sommer MOAS (2015) “ Synthetic selection enables functional discovery of a dominant microbial thiamine uptake system” submitted
(14) Schyns G, Haracz SM (2009) Thiamin production by fermentation – US 2009/0233296 A1. 1.
(15) Kawasaki T, Nose Y (1969) “Thiamine Regulatory Mutants in Escherichia coli”, Journal of Biochemistry, Vol 65, Issue 3, pp 417-&
(16) Schyns G, Geng Y, Barbosa TM, Henriques A, Perkins JB (2005) “Isolation and Characterization of New Thiamine-Deregulated Mutants of Bacillus subtilis”, Journal of Bacteriology, DOI: 10.1128/JB.187.23.8127-8136.2005
(17) Sudarsan N, Cohen-Chalamish S, Nakamura S, Emilsson GM, Breaker RR (2005) “Thiamine pyrophosphate riboswitches are targets for the antimicrobial compound pyrithiamine”, Chemistry and Biology, DOI: 10.1016/j.chembiol.2005.10.007
(18) Martin VJJ, Pitera DJ, Withers ST, Newman JD, Keasling JD (2003) “Engineering a mevalonate pathway in Escherichia coli for production of terpenoids” Nature Biotechnology, DOI: 10.1038/nbt833
(19) Ikeda M (2003) “Amino acid production processes”, Advanced in Biochemical Engineering/Biotechnology, 79, 1.35
(20) Mahillon J, Chandler M (1998) “Insertion Sequences”, Microbiology and Molecular Biology Reviews Vol 62, Issue 3, pp 725-774
(21) Durfee T, Nelson R, Baldwin S, Plunkett G, Burland V, Mau A, Csorgo B, Pasfai G, Weinstock GH, Blattner FR (2008) “The complete genome sequence of Escherichia coli DH10B: Insights into the biology of a laboratory workhorse”, Journal of Bacteriology, DOI: 10.1128/JB.01695-07
(22) Pitera DJ, Paddon CJ, Newman JD, Keasling JD (2007) “Balancing a heterologous mevalonate pathway for improved isoprenoid produktion in Escherichia coli”, Meetabolic Engineering, DOI: 10.1016/j.ymben.2006.11.002
(23) Xiao Y, Bown CH, Lui D, Zhang F (2016) “Exploiting nongentic cell-to-cell variation for enhanced Biosynthesis”, Nature Chemical Biology, DOI: 10.1038/nchembio.2046
(24) Lee SY, Kim HU (2015) “Systems strategies for developing industrial microbial strains”, Nature Biotechnology, DOI: 10.1038/nbt.3365
(25) Rugbjerg P, Myling-Petersen N, Porse A, Sarup-Lytzen K, Sommer MOA (2018) “ Diverse genetic error modes constrain large-scale bio-based production”, Nature Communications, DOI: 10.1038/s41467-018-03232-w
(26) Rugbjerg P, Sarup-Lytzen K, Nagy M, Sommer MOA (2018) “ Synthetic addiction extends the productive life time of engineered Escherichia coli populations”, Proceedings of the National Academy of Sciences of the United States of America, DOI: 10.1073/pnas.1718622115


Potential Impact:
PROMYS has had a very substantial impact and delivered a number of exploitable results. The confidential details of the commercially exploitable results were described in the deliverable 2.3 detailing the business plans for the individual exploitable results which here developed in the final year of the project. Here we provide a non-confidential description of the PROMYS outcomes.

Overall PROMYS has had impact in several different ways:
- Generation of foundational scientific knowledge
- Establishment of enabling technologies that can be commercialized by the SME partners
- Development of cell factories that will be commercialized by the SME partners
- New technologies that form the basis for new companies.

SCIENTIFIC KNOWLEDGE

During the PROMYS project a transition has occurred within the fields of metabolic engineering and synthetic biology. Indeed, the trends envisioned by the PROMYS proposal have begun to take root. Accordingly, the scientific knowledge developed in PROMYS is likely to be more widely applied and is expected to have an even wider impact.

In particular the work of PROMYS:
- Establishing the key parameters that are required in order to integrate aptamers into riboswitches can be applied to a vast range of molecules by several groups and companies enabling the broad application of ligand responsive regulation and selection systems.
- Generating new computational algorithms for predicting RNA circuits and designing regazymes, ligand-activated transcription factors and other RNA circuits shall find use broadly within the field of synthetic biology. Furthermore, such work on biological circuits might also find application within biomedical science where synthetic biology is increasingly delivering biological circuits and engineered cells as therapeutics.
- Developing approaches for effective selection of vast libraries using ligand responsive regulation and selection systems can be applied to mine vast libraries for enzymatic function. Coupled with the improvements in the design and development of circuits responsive to new ligands, these methodologies shall find application to the discovery of novel enzymes, pathways and cell factories.
- Assessment of the need for pathway balancing as evidenced by the work on the thiamine pathway. This work enabled by the ligand responsive regulation and selection system demonstrated how complex pathway must be carefully balanced in non-intuitive ways in order optimize production.
- Discovering the significant degree of genetic heterogeneity developing during mevalonate production highlights to the community that fermentation based processes may be severely limited by evolution. Furthermore, these findings highlight simple strategies that companies developing or commercializing fermentation based processes can take in order to assess the magnitude of this issue.

These findings have been disseminated through publications in scientific journals. Furthermore, the results have been disseminated to a wide audience through scientific conferences, trade shows and lectures to the general public.

ENABLING TECHNOLOGIES

In addition to the significant body of scientific knowledge generated by PROMYS a number of enabling technologies were also established with the participating SMEs. These enabling technologies are being commercialized or shall be commercialized in the near future to allow for further growth of their respective businesses.

Specifically, the following enabling technologies have been established:

- Biosynthetic selections have been established with Biosyntia as an enabling technology which allows Biosyntia to pursue new projects. Such project are established internally (at least one new project has been established) as well as through partnerships with larger chemical or ingredients companies (several projects are being evaluated). This technology has already resulted in additional revenue and investments to Biosyntia.
- Computational transcription factor design has been established as an enabling technology with Bacmine. Bacmine is using this technology to partner with other companies on the establishment of ligand response regulation and selection systems. This has already enabled new projects to be signed leading to additional revenue to Bacmine.
- Platform for biosynthesis of molecules related to the confidential molecule has been estasblished with Evolva. Based on the work of PROMYS several relevant biotransformations have been identified that could enable production of new molecules within a commercially interesting class of compounds.

CELL FACTORIES

PROMYS focused on the development of cell factories for thiamine and for the confidential molecule. During the cause of the project substantial progress has been made on both projects enabling Biosyntia and Evolva to take leadership positions within the development of cell factories for these molecules.

Specifically the following cell factories have been developed:
- Thiamine. Currently, Biosyntia owns the best cell factory for thiamine production. The cell factory is undergoing further development and is expected to be commercialized after the project finalization.
- Confidential molecule. Currently, Evolva owns the leading cell factory for production of this molecule that has substantial applications with the flavor and fragrance business. Commercialization of this bioprocess will commence after project finalization.

NEW BUSINESS IDEAS AND COMPANIES

During PROMYS several new ideas emerged that could potentially form the basis of new companies or projects. In the last year of the project a business development officer was hired from industry to evaluate and prioritize these ideas.

The following new ideas were also assessed for feasibility as new project within the participating SMEs:
- New opportunities within Flavor & Fragrances
- “plug-and-play” genetic circuitry that can be incorporated into industrial strains for continuous optimization of metabolic pathways
- New technologies benefiting from Biosynthetic selections.

The following ideas where developed and assessed in the project to determine appropriate business plans for new companies:
- RNA-aptamer assisted protein complementation for non-invasive and dynamic assessment of the concentration of intracellular biomolecules and metabolic fluxes
- In vivo phage-assisted evolution of riboswitches
- Enhancing production and product stability in industrial biomanufacturing

Already during the project a first spin out was launched based on proof-of-concept funding from DTU to enhance production and product stability in industrial biomanufacturing.

In conclusion, the PROMYS project is considered a significant success since it has delivered both on the stated objectives originally outlined in the proposal as well as delivered several new attractive opportunities for the participating partners.



List of Websites:
Project website: www.promys.eu

Coordinator:
Morten O A Sommer
Technical University of Denmark, Center for Biosustainability, Kemitorvet 220, 2800 Kongens Lyngby, Denmark
moas@biosustain.dtu.dk

Pablo Pomposiello
Bacmine, C/ de Santiago Grisolia, Laboratorio 151, Parque Cientifico de Madrid, Tres Cantos 28760, Spain
pablopompo@bacmine.com

Beatrix Suess
Technische Universitaet Darmstadt, Karolinenplatz 5, 64289 Darmstadt, Germany
bsuess@bio.tu-darmstadt.de

Matthias Heinemann
Rijksuniversiteit Groningen, Broerstraat 5, 9712 CP Groningen, Netherlands
m.heinemann@rug.nl

Alfonso Jaramillo
The University of Warwick, Kirby Corner Road, University House, Coventy, CV4 8UW, United Kingdom
alfonso.jaramillo@warwick.ac.uk

Michael Naesby
Evolva AG, Duggingerstrasse 23, Reinach 4153, Switzerland
michaeln@evolva.com

Hans Jasper Genee
Biosyntia ApS, Fruebjergvej 3, 2100 Copenhagen O, Denmark
hjg@biosyntia.com