Our Engineering Process

Gut Therapeutic

Our team began with a seemingly simple idea: create a gut therapeutic capable of secreting the compound pterostilbene for the treatment of Alzheimer’s disease (AD). We learnt about the Nissle strain of E. coli that could potentially be used to form a semi-stable colony within the gut from a paper by Hamimoto and colleagues (Hamimoto et al., 2022). However, a series of design choices needed to be considered. One of the major limitations of such a system was control of expression. The initial idea consisted of producing pterostilbene upon exposure to a biomarker of AD. One of the main candidates we identified from literature was short-chain fatty acids (SCFA) such as butyrate which are produced by bacterial fermentation of nondigestible starches; others include Vitamin D3 and advanced glycation end-products (Szczechowiak et al., 2019). Gut dysbiosis has been associated with increased permeability of the intestinal mucosa and a potential increase in blood-brain barrier permeability and so at the time, there was promise of potentially making a gut therapeutic treatment of AD. However, we soon learnt that many of the biomarkers present in the gut may not be the result of the disease pathology itself but instead of diet and so would not be able to provide the predictive effect desired for a gut therapeutic. Because AD is not a condition characterised by “flashes” or waves of pathological activity like epilepsy, any biomarker present of AD in the gut of a patient was likely to be continuous throughout the course of affliction and so not amenable to treatment by a conditionally expressed gut therapeutic. As such, our project changed focus. From the initial idea of pterostilbene production in the gut, we focused on the environmental benefit of producing pterostilbene biosynthetically as has been reported by researchers working in a similar field. Due to the low amounts pterostilbene that can be produced in plant material (99-520 ng/g of dry weight berry), we aimed to implement a biosynthetic pathway to produce pterostilbene in E. coli to create a less costly alternative to production (Heo et al., 2017).

Enzyme Choice

We learnt of four enzymes needed for production of pterostilbene from the precursor L-tyrosine. These enzymes are tyrosine ammonia-lyase, 4-coumaroyl:CoA ligase 1, stilbene synthase, and resveratrol o-methyltransferase. We considered alternative production pathways such as the natural production of stilbenes by Photorhabdus luminescens, which the literature identified as being able to produce the stilbene benvitimod. This stilbene is produced through a pathway driven by two P. luminescens enzymes: StlD and StlC, which catalyze condensation and aromatization reactions, respectively (Mori et al., 2016). While the benvitimod stilbene displays structural similarity to resveratrol (Smith et al., 2017), we could not identify any papers describing a possible conversion of benvitimod to resveratrol, for our potential use of this pathway to then produce pterostilbene. Therefore, we then looked at the StlD enzyme, which is the one using CoA-bound substrates, to determine whether it is able to, just like stilbene synthase (De Luca & Lauritano, 2020), use 4-coumaroyl-CoA and malonyl-CoA as substrates for resveratrol synthesis. While the StlD enzyme was shown to have a broad substrate tolerance, we did not identify any literature suggesting a resveratrol synthesis activity. Moreover, StlD even uses a different catalytic mechanism than the essential catalytic triad in stilbene synthase activity (Mori et al., 2016), and therefore it is unlikely to be able to produce resveratrol. For this reason, we selected the phenylpropanoid pathway commonly found in plants, comprised of the 4 enzymes mentioned.

Expression System Design

To learn more about our idea we conducted a literature search to identify competing expression systems that potentially yielded greater titres of pterostilbene. From this search, we investigated a promising paper by Yan et al. (2021). This paper suggested the resveratrol O-methyltransferase from Vitis vinifera to be a superior enzyme for pterostilbene synthesis; this was further supported by the achievement of higher pterostilbene titres than Heo et al. (2017). Furthermore, this paper proposed four mutant versions of a new set of enzyme homologues: tyrosine ammonia-lyase from Rhodotorula glutinis (RgTAL), 4-coumarate:CoA ligase from Arabidopsis thaliana (At4CL), and stilbene synthase (VvSTS) and resveratrol O-methyltransferase (VvROMT) both from Vitis vinifera. Many of the mutations found in these enzymes are close to the active site, suggesting a modulation in active site size, specificity or catalytic efficiency. Using these sequences, we designed our first set of parts on SnapGene with RFC10 BioBrick prefixes and suffixes to allow for 3A assembly. We then built our expression system – a co-transformation system with two enzymes on separate plasmids transformed into one E. coli (Figure 1).

The co-transformed plasmids would be transformed into E. coli which would, in turn, be exposed to the precursor tyrosine in the media, thus catalyzing the production of pterostilbene. We then designed our new method to involve production of pterostilbene in the lab, quantify the compound via high pressure liquid chromatography, then using this pterostilbene to attenuate the neuroinflammatory hallmark in AD. Firstly, it was necessary to decide the organisation of the four genes among the finalized expression plasmids. We designed two expression systems to produce pterostilbene. The first was an operon-like system where all four genes were under the control of a single promoter. This is a method similar to that which was utilized by the Uppsala 2013 iGEM team to produce resveratrol (Figure 2). However, instead of 4-coumaric acid as precursor, we selected the precursor L-tyrosine due to being cheaper to supply than the former.

The second expression system the expression of all four genes under their own promoter in a fashion similar to Heo et al. (2017). To learn whether this system was ideal, we contacted Dr Zhi-Bo Yan, the researcher who used the mutated version of the four enzymes as we wish to use for pterostilbene production. He recommended the use of a multi gene system design whereby the four genes are expressed under a single promoter in a “synthetic operon” system. Further research also showed that a synthetic operon was a better option for our system. Because the first option for the gene expression system involved using the same or similar promoter sequence repeatedly, there was a risk that during replication, certain sequences could be excised or recombined out of their plasmids or reorganised into an undesired construct. Therefore, since an operon system is a common natural expression system in E.coli (with several genes transcribed into one mRNA), our team decided to use this operon system over each gene being expressed under a its individual promoter.

Having decided on our expression system, our team wanted to learn about what plasmid copy number we wished to express our genes in. Yan et al. (2021) used a low plasmid copy number while Heo et al. (2017) used a medium plasmid copy number. While Yan et al. (2021) yielded higher titres of pterostilbene (80.04 mg/L) compared to Heo et al. (2017), the two research groups used a different o-methyltransferase genes, making direct comparisons difficult. Additionally, our team noted that whilst production of enzymes involved in stilbene synthesis were under the control of strong promoters in both papers, there is a risk that L-tyrosine could be drawn from other vital metabolic processes if there is either an excessive enzyme concentration or tyrosine depletion (Chavez et al., 2013). Because of the wide range of promoters and copy numbers available to our team, we decided to attempt to model the expression pathway in silico to allow for testing of different promoter strengths and plasmid copy numbers efficiently. Our team therefore decided to design a model of pterostilbene production under the conditions of varying plasmid copy numbers and promoter strength values within E. coli, using catalytic efficiency values available for the wild-type enzymes. We built this model on MATLAB based on a similar code used by the iGEM 2013 Uppsala team, and learnt from an initial version of this model that a medium copy number plasmid was ideal for pterostilbene synthesis. Therefore, we opted for strong promoters and a pBR322 medium copy number origin of replication as was done by Heo et al. (2017). The model made use of the Anderson promoter collection available on the iGEM registry as these have reported quantitative Polymerases Per Second (PoPS) values that could easily be translated into a model (Kelly et al., 2009).

Cloning Method

Another barrier to our project design was in identifying a method or standard to clone all four transcriptional units into a singular plasmid backbone and then transform them into E. coli. One method we learnt for implementing multiple transcriptional units into a single plasmid was type IIS assembly. This method utilizes type IIS enzymes to catalyze cuts that result in non-specific overhangs for the purposes of multigene ligation. More specifically, we used the JUMP type IIS assembly plasmids made available on the iGEM 2022 plate kits. This assembly method made use of the BsaI and BsmBI restriction enzymes. We designed our constructs using this system over 3A assembly due to its capacity to create multi-gene systems with greater flexibility in modulation of promoter strengths. We built our first set of sequences on SnapGene, adding BsaI sites to the level 0 parts in such a way that the ligated product would naturally order itself into an operon-style system. To allow for the construction of an operon system, we made use of linker sequences. There were two linker sequences we used: (1) a linker sequence replacing the promoter and (2) a linker sequence replacing the terminator. Because our final level 2 construct should only have one promoter and terminator in the entire 4-gene operon, we used these linker sequences (which did not contain promoter and terminator sequences) to allow for continuous transcription through the operon in a level 2 plasmid.

We then tested our BsaI sites by running the assembly using the golden gate assembly feature on SnapGene. From this we discovered that the Joint Universal Modular Plasmid (JUMP) plasmids, present on the iGEM registry, crucially differed from those present on the supplementary material of Valenzuela-Ortega and French (2021); more specifically, the plasmid restriction site overhangs for multi-transcriptional unit assembly differed in a way that would alter the prefixes and suffixes required flanking each part. To learn more about these sequences and identify which plasmid sequences to use in designing, building, and testing our system, we contacted Dr Marcos Valenzuela-Ortega who initially designed these plasmids. From this, we learnt that the JUMP plasmids from the supplementary material (Valenzuela-Ortega & French 2021), and displayed by Addgene, were the appropriate ones to use. Dr Valenzuela-Ortega also very kindly offered protocols on how to execute JUMP assembly. Once we understood how to implement type IIS JUMP cloning into our design, we set about learning how to culture our E. coli for optimized pterostilbene production. We designed a 24-hour expression system in which we would add (at time = 0 h) 0.1 mM of IPTG for the expression of the four enzymes in E. coli, and in media containing the L-methionine (an intermediate needed for efficient dimethylation of resveratrol into pterostilbene). From this, our final expression system was ready, and thus we built an operon system consisting of four genes using a medium copy number plasmid of pBR322 origin. We used a LacR-repressible version of the strong BBa_J23100 promoter upstream of the RgTAL (first gene) coding sequence. Finally, E. coli lysate would be analyzed for pterostilbene production using HPLC-UV (Heo et al., 2017).

Future Implementations

For future implementations of our project, we also investigated other components aside from the broth and media we wanted to use to optimize pterostilbene production. We learnt that many papers focusing on the production of pterostilbene add in an antibiotic known as cerulenin to inhibit the fabB and fabF genes encoding enzymes involved in the fatty acid cycle, with the goal of increasing the crucial availability of malonyl-CoA in the pterostilbene production pathway (Lim et al., 2011). Since this cycle is a main consumer of malonyl-CoA, and the pterostilbene biosynthetic pathway requires 3 malonyl-CoA molecules compared to the one molecule of L-tyrosine needed for each molecule of pterostilbene produced, we deemed, as supported by literature, malonyl-CoA as the main limiting factor for pterostilbene synthesis. We learnt, however, that cerulenin is very expensive and could irreversibly inhibit fatty acid synthesis entirely which can be toxic to E. coli (Wu et al., 2017). In investigating alternatives, the KCL iGEM 2019 team conducted a project based on small RNA, coupled with different RNA scaffolds, which were capable of inhibiting GFP expression as proof of concept. We therefore investigated if we could use an RNAi-based method of fatty acid synthesis inhibition to increase the intracellular pool of malonyl-CoA. We aimed to determine whether similar technology could be used for inhibiting the fabD gene, encoding the enzyme responsible for converting malonyl-CoA into malonyl-ACP, an unusable compound in our synthetic system. We learnt from research about the MicC RNA scaffold which was used by the UT Tokyo iGEM team in 2013 and well supported in a paper by Yoo et al. (2013). Therefore, we designed a system in which two plasmids would be transformed into E. coli . The first plasmid would contain all four pterostilbene-producing genes under the control of a single strong BBa_J23100 promoter, and in a plasmid with a pBR322 origin of replication. We would also transform a second plasmid containing a fabD-targeting sRNA construct with the MicC scaffold to increase malonyl-CoA levels and therefore the final pterostilbene titre.

Kinetic Modelling for Copy Number Parameters

Part of our engineering process will aim to produce constructs yielding the highest amount of pterostilbene possible without overbearing the E. coli to ensure that our system will be fit for prolonged synthesis of pterostilbene. We are planning to use dry lab modelling to learn how to construct our laboratory design phase. Kinetic modelling will be used to work out parameters for plasmid copy number, promoter strength, and sufficient intracellular amount of L-tyrosine. The modelling will inform us about the best copy number range, promoter strength and optimal amount of tyrosine. This will help to design the final design of our constructs. Multiple iterations of the model would need to be constructed and tested until one proves satisfactory. Hence within the creation of the model there will be at least two or three engineering cycles, where the approach the model is constructed on would have to be changed.

If our dry lab modelling / wet lab tests show a flaw in our methodology, our design could be adjusted to meet our project needs by further confirming knowledge provided from dry lab like promoter strength. We could design 2 plasmids each containing different promoters, T7 and an anderson promoter from the iGEM registry. We can then build these plasmids in the lab using GFP as a reporter protein. We will then test the strength of each promoter by measuring the intensity of fluorescence produced by the constructs. This will enable us to determine which promoter will be more beneficial for our system and thus implement this new found knowledge in the design of our construct.

LEARN: Modelling Tools

Ordinary Differential Equations (ODEs)

Ordinary Differential Equations (ODEs) are one of the main tools used for modelling of all types, there is a form of the differential equation. Differential Equations are equations comprising of terms that are derivatives of one variable, (usually the dependent variable). The equation describes the derivative or derivatives of a function that are unknowns.

Differential equations are usually used to describe systems that have many rates of change that interact with each other. For example the rate of change of water volume of two interconnected water tanks (Figure 3).

LEARN: Modelling Tools

Law of Mass Action

The Law of mass action is often used as the foundation for the mathematical modelling of biochemical systems. It can be used to explain the behaviour exhibited by solutions that are in a state of balance between continuing processes, (dynamic equilibria)

If we apply the law of mass action to a reversible reaction with reactants A and B and products C and D, we can represent the equilibrium constant K_eq as follows:

For instances at which the system is not at equilibrium the ratio is known as the reaction quotient which is designated as Q, the expression for which is as follows:

LEARN: Types of Models

Stochastic vs. Deterministic

When we use ODEs alongside the law of mass action, we are creating a model that is deterministic and neglects noise. This means the model is therefore less accurate, as noise is intrinsic to biological systems

An alternative method is making use of the DCME (discrete chemical master equation) which is stochastic in nature. The chemical master equation is a system of ODEs that describes the evolution of a series of reactions as stochastic processes.

In general, a deterministic model operates under the assumption that when modelling large populations, known average rates have no random deviations. Whereas a stochastic model assumes that the systems vary in a random manner and models the potential outputs using time-ordered random variables

When used, deterministic models produce consistent outputs for a given set of input data, this is because none of the mathematical characteristics of the system is random and each problem has just one set of values and solution. For a deterministic model, the unknown components are external, it deals with explicit outcomes with no allowances for error. Henceforth this model is usually used when we can determine our result via known relationships, for example, accounting sheets

A stochastic model is unpredictable and unknown elements are integrated within the model. The model itself generates several outcomes with different variables and is repeated several times using different settings. They rely on estimating the probability of events. These systems are usually used in instances where relationships between variables are unknown to measure the growth of a bacterial population or the spatial variation of a gas molecule

When considering that we would be able to model how the concentration of substrates affects the concentration of products, we considered a deterministic approach would be the best suited for our model. However, we also saw value in exploring a stochastic approach as it would allow us to explore a variety of outcomes as a feature of the model. Hence the initial model we attempted to create was a stochastic model, shown below, with the intention of integrating it within a deterministic one.

DESIGN:

Model version 1

Stochastic Model

Goals

Types of Reactions

This is under the assumption that A≠B_i≠A_i

[A] is the concentration of A in moles per litre as a function of continuous time

[A]^• is the time derivative of [A]

Reactions we are considering

We assume that the reactions occur in a ‘well-stirred solution’, therefore the dynamics depend entirely on the concentrations of each chemical species.

In this instance, I have replaced the base rate (k) with the rate the enzyme is used up. Larger rates mean faster reactions.

We will now translate the reactions to ODEs, by creating a stoichiometric matrix we call N:

Stoichiometric Matrix

Let us consider our 4 chemical equations v₁, v₂, v₃, v₄ with corresponding reaction rates [TAL], [coA ligase], [stilbene synthase],[ROMT].

We create a stochiometric matrix, by allowing one row for each chemical species and one column for each reaction.

The number in each square of the matrix depends on whether or not molecules of the species are being produced or removed overall.

If the species are being produced the number is positive, if the species are being removed the number is negative, otherwise, the square contains zero.

Let us do that for our reactions:

For our reactions we get:

BUILD: MATLAB Code

Rate Equations

We then let X be the vector of chemical species. Hence the system of ODEs is then, in general, given by the following rate equation:

[X]•=N • l

We can then input the vectors into MATLAB to output the following equations.

The following MATLAB code was used in this instance:

The output equations were as follows:

We can then convert these equations to a more familiar format.

Recall that [A]^• is the change in concentration with respect to time of A, where [A] is the concentration.

[L-tyrosine]^• = [TAL]^•[L-tyrosine]

[P-coumaric acid]^• = ( [coA ligase]^•[(ATP+coA)][P-coumaric acid] ) -([TAL]^•[L-tyrosine])

[(ATP +coA]^• = [coA ligase]^•[P-coumaric acid][(ATP +coA)]

[4-cou coA]^• = ( [stil synthase]^•[4-cou coA][(H++3malonyl coA)] -[coA lig]^•[Pcou acid][(ATP coA)]

[(H++3malonyl coA)]^• = [stilbene synthase]^•[(H++3malonyl coA)][4 -coumaroyl coA]

[transresveratrol]^• = [ROMT]^•[S-adeno meth][transresveratrol] - [stil synth]^•[coum coA][(H+3m)]

[S-adenosyl methionine]^• = 2[ROMT]^•[S-adenosyl methionine][transresveratrol]

[pterostilbene]^• = -[ROMT]^•[S-adenosyl methionine][transresveratrol]

The following parameters were provided by our team members working within the wet lab:

Given this information, the equations are now as follows:

TEST: Model Outputs

Rate Equations

These equations, whilst useful to represent the vague relationships between reagents, without a system to properly construct an appropriate space. There is also a lack of information to further complete the model, specifically the probability of each outcome. Hence this model was deemed a waste of resources and left unfinished.

LEARN: What did we learn from this model?

Learned to identify different types of equations

Became familiar with the synthesis process

Collected parameters relating to enzymes, which proves useful later on.

Confirmed the stochastic approach was not ideal for the model requirements

DESIGN:

Model version 2a

Deterministic Model

Given this information, the next iteration of the model was purely deterministic, the construction of it described in further detail within the modelling section. This model began by building on two main sources, the first of which being the (Date) iGEM Engineering webinar on ‘Modelling using ODEs’ and the model of the 2013 Uppsala iGEM team used to predict p-coumaric acid yield.

BUILD

The initial version of the model was created with the assistance of the Michaelis Menten equations to create the majority of the rate equation ODEs. This model was quite simple with placeholder symbolic values for plasmid copy number, promoter strengths, k, kcat and kM values.

TEST

It produced graphs that made little physical sense and often had issues with resolution whilst using the MATLAB ODE solving function ode45, which we believed would give us the most accurate results.

LEARN: Choosing an ODE Solver

An ODE system consists of any number of coupled ODE equations. Our specific system at this point includes 15 ODEs. ode45 works well for most systems however we now know that ode45 may not work well for systems that exhibit stiffness or difficulty in evaluation.

Stiffness is generally considered to occur when there is a large difference in scaling within the problem. This would indeed apply to our system as the initial L-tyrosine concentration outside the cell is approximately 3-4 orders of magnitude larger than the concentrations of some of the enzymes within the cell.

LEARN: Creating arrays and using the switch case function

In order to test for different combinations of promoter strengths, we began to look into using a switch function to select what strength of promoter to use for each enzyme. A switch case is used in order to execute one of several statements depending on the value of a previously established parameter.

This would also require creating an array of promoter strengths to be used that could be switched between

DESIGN/BUILD: Creating plan for code

Using what we had learned from the online MATLAB resources we created a rough plan of the order of operations for the MATLAB script that went as follows.

1. Store promoter strength array

2. Request user to select promoter strength for each programme

3. Repeat until promoter strengths are selected for all four enzymes

4. Run ODE solver

5. Plot graphs

The ODE solver in use was changed and after consulting the MATLAB guide, we deduced that ode15s may be the best suited for our system, whilst retaining the highest possible accuracy for a stiff solver.

Model version 2b

Deterministic Model

The model itself appears to function correctly but it is restricted to only testing one set of promoter strength and plasmid copy number combinations at once.

LEARN: Converting script to a function

A function is more flexible than a script as it allows the script to reuse a series of commands. This is especially useful as you can pass input values into the function which can come from other functions and return output values.

The general format for creating a function looks as follows:

LEARN: Creating a 3D Model

Surface vs. Mesh Plots

The mesh and surf functions both create surface objects and display them in 3 dimensions. However, mesh produces a wireframe surface that only displays the lines connecting the data points in colour. Whereas surf displays both the connecting lines and the faces of the surface in colour.

DESIGN: Converting the script to a function

By converting the existing script to a function and removing the need for user input, we are able to loop the function over every single possible promoter strength and plasmid copy number combination.

Hence we are able to store the maximum pterostilbene yield of each iteration to create the 3-D model. This process must be done in a separate script.

BUILD

TEST:

Model Output

This iteration of the model is built using the Michaelis Menten equations as its base. It consists of 15 ODEs each of which tracks the concentration of each enzyme and reagent. The model outputs final concentrations of each reagent accounting for 256 different combinations of 4 possible promoter strengths for each gene, as well as a range of plasmid copy numbers ranging from 5 to 300. This data is represented by the 3D surface model on the wiki which represents the maximum concentration of pterostilbene for each iteration.

The model comprises 3 functions, one of which is an open source permutation MATLAB script sources externally, and one main script that outputs the final model. The other two functions include “Pterostilbene_rates_ver_2308”, which includes the 15 rate equations and “Pterostilbene_Production_Sim_function”, which carries out the ode solver.

Included on the wiki is also the script “Pterostilbene_Production_Sim_ver3008” which only carries out one iteration at a time. It was used in the engineering process to refine the model and outputs concentrations of each regent and enzyme specific to the promoter strength/copy number combination. It includes requests for input from the user such as initial concentrations, plasmid copy number and promoter strengths.

LEARN: Assessing the Model

Looking at the model, we can see that there is most likely a logical error in mathematics. This is because the model is implying the highest copy number and highest promoter strengths. Two issues that were brought to attention were the lack of restriction of L-tyrosine consumption as well as other trade-offs attributed to metabolic strain. This is made obvious by the model's implication that the highest possible copy number is ideal.

DESIGN: L-Tyrosine Consumption

We are able to exclude results that cause intracellular levels of L-tyrosine to dip below what we believe is required to retain normal cell function. We can do this using an if statement that zeroes pterostilbene yields that are the result of significantly-low l-tyrosine concentrations.

BUILD:

TEST: Surface Plot

Version 2

The code suggests a plasmid copy number of around 13 and promoter strength combination 254 which corresponds to the following strengths respective to enzyme:

RgTAL: 0.00043 PoPS

At4CL: 0.00043 PoPS

VvSTS: 0.00043 PoPS

VvROMT: 0.0200 PoPS

This is slightly different to the promoter strengths we ended up using hence, a different model needed to be constructed to account for the promoter strengths we ended up using, as follows:

LEARN: Metabolic Strain

Having discussed the model with the 2022 Warwick team we were able to pool resources regarding modelling metabolic strain.

Given more time, we would be able to implement the following changes to the model:

Pool m_RNA for all enzymes as one to allow for trade-off

Implement metabolic strain ODEs into existing simplistic scaffold

Adjust for overlap in parameter names

Remove redundant and inconsistent ODEs from previous iterations of model

Define rates and parameters

Modelling Engineering Cycle Summary

Please see our Contributions Page for the MATLAB scripts themselves: combinations code, user input, and promoter strengths.

References

Chávez-Béjar, M. I., Balderas-Hernandez, V. E., Gutiérrez-Alejandre, A., Martinez, A., Bolívar, F., & Gosset, G. (2013). Metabolic engineering of Escherichia coli to optimize melanin synthesis from glucose. Microbial cell factories, 12, 108. https://doi.org/10.1186/1475-2859-12-108

Harimoto, T., Hahn, J., Chen, Y.-Y., Im, J., Zhang, J., Hou, N. et Danino, T.. (2022). A programmable encapsulation system improves delivery of therapeutic bacteria in mice. Nature biotechnology, 40(8), 1259‑1269. doi:10.1038/s41587-022-01244-y

Heo, K. T., Kang, S.-Y. et Hong, Y.-S.. (2017). De novo biosynthesis of pterostilbene in an Escherichia coli strain using a new resveratrol O-methyltransferase from Arabidopsis. Microbial cell factories, 16(1). doi:10.1186/s12934-017-0644-6

Herrera, D. P., Chánique, A. M., Martínez-Márquez, A., Bru-Martínez, R., Kourist, R., Parra, L. P. et Schüller, A.. (2021). Rational Design of Resveratrol O-methyltransferase for the Production of Pinostilbene. International journal of molecular sciences, 22(9), 4345. doi:10.3390/ijms22094345

Valenzuela-Ortega, M. et French, C.. (2021). Joint universal modular plasmids (JUMP): a flexible vector platform for synthetic biology. Synthetic biology, 6(1). doi:10.1093/synbio/ysab003

Kallscheuer, N., Vogt, M., Bott, M., & Marienhagen, J. (2017). Functional expression of plant-derived O-methyltransferase, flavanone 3-hydroxylase, and flavonol synthase in Corynebacterium glutamicum for production of pterostilbene, kaempferol, and quercetin. Journal of biotechnology, 258, 190–196. https://doi.org/10.1016/j.jbiotec.2017.01.006

Lim, C. G., Fowler, Z. L., Hueller, T., Schaffer, S. et Koffas, M. A. G.. (2011). High-Yield Resveratrol Production in Engineered Escherichia coli. Applied and environmental microbiology, 77(10), 3451‑3460. doi:10.1128/aem.02186-10

Wu, J., Yu, O., Du, G., Zhou, J., & Chen, J. (2014). Fine-Tuning of the Fatty Acid Pathway by Synthetic Antisense RNA for Enhanced (2S)-Naringenin Production from l-Tyrosine in Escherichia coli. Applied and environmental microbiology, 80(23), 7283–7292. https://doi.org/10.1128/AEM.02411-14

Wu, J., Zhou, P., Zhang, X. et Dong, M.. (2017). Efficient de novo synthesis of resveratrol by metabolically engineered Escherichia coli. Journal of industrial microbiology and biotechnology, 44(7), 1083‑1095. doi:10.1007/s10295-017-1937-9

Yoo, S. M., Na, D. et Lee, S. Y.. (2013). Design and use of synthetic regulatory small RNAs to control gene expression in Escherichia coli. Nature protocols, 8(9), 1694‑1707. doi:10.1038/nprot.2013.105

Kim, K.-J., Kim, H.-E., Lee, K.-H., Han, W., Yi, M.-J., Jeong, J. et Oh, B.-H.. (2004). Two-promoter vector is highly efficient for overproduction of protein complexes. Protein science, 13(6), 1698‑1703. doi:10.1110/ps.04644504

Wang, Y., Bhuiya, M. W., Zhou, R., & Yu, O. (2015). Pterostilbene production by microorganisms expressing resveratrol O-methyltransferase. Annals of Microbiology, 65(2), 817-826. doi:10.1007/s13213-014-0922-z

De Luca, D., & Lauritano, C. (2020). In Silico Identification of Type III PKS Chalcone and Stilbene Synthase Homologs in Marine Photosynthetic Organisms. Biology, 9(5), 110. doi: 10.3390/biology9050110

Mori, T., Awakawa, T., Shimomura, K., Saito, Y., Yang, D., Morita, H., & Abe, I. (2016). Structural Insight into the Enzymatic Formation of Bacterial Stilbene. Cell Chemical Biology, 23(12), 1468-1479. doi: 10.1016/j.chembiol.2016.10.010

Smith, S., Jayawickreme, C., Rickard, D., Nicodeme, E., Bui, T., & Simmons, C. et al. (2017). Tapinarof Is a Natural AhR Agonist that Resolves Skin Inflammation in Mice and Humans. Journal Of Investigative Dermatology, 137(10), 2110-2119. doi: 10.1016/j.jid.2017.05.004

Szczechowiak, K., Diniz, B. S., & Leszek, J. (2019). Diet and Alzheimer's dementia – Nutritional approach to modulate inflammation. Pharmacology Biochemistry and Behavior, 184, 172743. doi:https://doi.org/10.1016/j.pbb.2019.172743

Gerngross, D., Beerenwinkel, N., & Panke, S. (2022). Systematic investigation of synthetic operon designs enables prediction and control of expression levels of multiple proteins. bioRxiv, 2022.2006.2010.495604. doi:10.1101/2022.06.10.495604

Kelly, J. R., Rubin, A. J., Davis, J. H., Ajo-Franklin, C. M., Cumbers, J., Czar, M. J., De Mora, K., Glieberman, A. L., Monie, D. D., & Endy, D.. (2009). Measuring the activity of BioBrick promoters using an in vivo reference standard. Journal of Biological Engineering, 3(1), 4. https://doi.org/10.1186/1754-1611-3-4