Difference between revisions of "Team:Marburg/Design"

One key feature of the Marburg Collection are the connectors that provide our toolbox with the required flexibility. Our design is inspired by the “Dueber Toolbox”, a golden-gate-based cloning method designed for applications in Saccharomyces cerevisiae (Lee et al. 2015) . To our knowledge, the “Dueber Toolbox” is the first cloning system that performs assembly through all levels using restriction sites located on independent basic parts, termed connectors, instead of destination plasmids for level 1 and level 2 plasmids (Weber et al. 2011) .

Design of insulators

Natural Competence

The Flp/frt system

There are different E. coli strains for a variety of applications. In every lab diverse strains are stored in freezers for methods like cloning, protein expression and protein interaction. And all of them are being used. For our project Vibrigens, we decided to establish the world’s fastest three lab strains suited for specific needs based on our V. natriegens wildtype strain. Here we describe how we designed our three strains VibriClone, VibriExpress and VibriInteract based on the genotype of commonly used E. coli strains.

VibriClone

To create a cloning strain that fulfills all the wishes of the scientific society, we have to think of a good strategy. A cloning strain is a kind of a tool in the laboratory day. We asked ourselves which features we needed urgently. The currently used chassis for cloning is E. coli. We started to look in the genotype, to see what modifications are important. Then we think about which one we would like to take over. Our choice was to knock out the exonucleases so that it would be possible to transform DNA without degrading. In addition, we would like to insert a point mutation in recA, that the homologous recombination between plasmids and into the genome. Moreover, we also want to make it possible to store in the fridge. Therefore we insert the katG gene encodes an enzyme with an catalase and peroxidase activity into the genome. Furthermore, we also wanted to make the blue-and-white selection possible for our VibriClone, we can achieve that by inserting the lactose Operon in the genome.

---

Exonucleases

In the cloning strain of E. coli, the exonuclease EndA was knocked out. This is very important for such a strain, otherwise the exonucleases degrade the DNA to be transformed. EndA makes double-strand breaks in duplex DNA und is the most abundant DNase activity in E. coli (Dürwaldet al.1968). In the present of the exonuclease no colonies were formed during transformation (Kanekoet al.2010) . It was recognized that by only the complete loss of host endA there were no degradation after 30h at 39°C. In addition, it was also found out that it creates a structural stability to give transformants and that higher molecular weight genomic DNA bands were recovered from the supernatant and it leads to certain discrete bands in digestion (Itayaet al.2012) . V. cholerae has two extracellular nucleases: Xds and Dns. The Dns is homologous to EndA from E. coli. Both proteins have been shown to be secreted into the medium. These two would degrade the free DNA, which is the source of a natural transformation. Because of this the method would not work or work very poorly. To enable the uptake of intact DNA during the competency process, the functions would have to be extinguished. It was shown that when you delete the nucleases, the transformation process works well. In a mutant in which the dns was knocked out, transformation frequency is more than hundred times higher than WT and when xds was knocked out the transformation frequency was 2,5-fold increased. This result showed that it is good if both nucleases are knocked out, but that the dns is the most important one (Itayaet al.2012) . In addition, you get through the deletion of the two nucleases a cleaner preparations of DNA. We wanted our cloning strain to have all the benefits, so it was very important to us that the exonucleases be knocked out. Otherwise, transformation efficiency would be very low and you don´t get clear preparations of DNA and that would not be great for a cloning strain. We started by searching for homologous genes of endA of E. coli and dns and xds of V. cholera. We were able to find out by blast analysis that in Vibrio natriegens there are two exonucleases homologous to those of E. coli and V. cholera. They are called Dns and ExeM.

In the alignment of the V. natriegens dns with the V. cholera dns and E. coli endA you can conclude that the three proteins are highly conserved among the three strains (see Alignment). The intermediate region and the C-terminus are highly conserved in all three proteins, while the N-terminus is variable. This suggests that the parts that are highly conserved are important for the function of nucleases. It can be concluded that the dns of V. natriegens has a very similar function to the two exonucleases of the other two organisms.

The most parts of the two proteins ExeM from V. natriegens and Xds from V. cholera are conserved, but both the N- and the C-Terminal vary a lot from each other. This suggests that the function of these two proteins has a similar function. Thus we can assume that it is very probable that the ExeM protein in V. natriegens behaves like the Xds from V. cholera. Because of this it is very important to delete this protein otherwise the DNA would be degraded. In 2016, Matthew T Weinstock showed that the quality of isolated plasmid DNA could be improved by deleting dns in V. natriegens (Weinstocket al.2016) . In addition, the growth and viability of cells with a deletion of dns in rich medium should not be impaired (Daliaet al.2017) .

After we made all these thoughts, we were able to start and plan the deletion of the exonucleases Dns and ExeM. We constructed cassettes to erase the genes. To first get to know the method and to check whether the natural transformation works as it has already been shown in our hands, we have started to knock out the dns. For this we constructed a cassette with a chloramphenicol resistance and the upstream and downstream region of the dns gene. To use the natural transformation it is enough to use 1000bp flanks but for the co transformation it is much better to use 3000bo flanks. By using Frt sequences between die single fragments, the antibiotics cassette can be excised from the flippase after integration. Because this knockout worked, we started planning for VibriClone. Our decision was made to perform the deletions with co-transformations. This was a big advantage because you only need one selected product. As selected product we chose the dns, due to the fact, it was shown that a deletion and insertion of an antibiotics cassette at this locus is not harmfulto the cells. This has not been shown for the other parts. The deletion of exeM is done by an unselected product. Therefore we create the upstream and downstream regions of this gene and take another gene that we would like to insert (katG) and build an unselected cassette by using fusion PCRs.
With this procedure, we are able to knock out the exonucleases and thus obtain a higher transformation efficiency.

---

RecA

The classic allosterically regulated enzyme RecA is involved in the regulation of the SOS System and the homologous recombination (McGrewet al.2003) . If you think about what you need for a cloning strain, it quickly comes to your mind that homologous recombination in the genome and between plasmids is not only unnecessary but also undesirable.
In E. coli, a so-called recA1 mutation was performed. This mutation limits the recombination of an introduced plasmid with the genome and between plasmids thereby the stability of the plasmids are ensured. Another advantage is that the exogenous plasmid DNA is left unchanged by the modification of RecA. This allows the purification of this bacterial culture plasmid to allow high fidelity PCR amplification of the original plasmid sequence. The modification in this gene was a point mutation from G to A at position 720. Thereby the glycin at the place 160 chance to aspartic acid. We use this model to modify our strain VibriClone.
At position 158 in recA of V. natriegens is a glycine. Through an alignment, we have found out that this is homologous to that in E. coli. To get the same kind of function loss, this glyin needs to be changed to an aspartic acid. For the modification of recA we build an unselected cassette from the gene with a point mutation. To achieve that two primers bind at the future mutation place, one forward and one reverse were created. In addition two more primers with a gap of 3000 bp to the other both were ordered. The results are two fragments which end at the point mutation. By a fusion PCR with only the outside primers, you get the hole fragment containing the recA with the desired mutation.
All in all with these mutation we make VibriClone fine for all of us to use for cloning without recombineering of an introduced plasmid with the genome and between two plasmids.

---

KatG

Another feature, which would improve the applicability of V. natriegens as a cloning chassis, would be to engineer cold resistance, due to fact that we all are used to store our plates of E.coli in the fridge. It has been shown for other vibrio species that cold sensitivity is triggered by toxic peroxide present in media, which is normally removed for example in E.coli by reactive oxygen species detoxification systems. Therefore our approach is to introduce such a system into our cloning strain, in order to transfer that tolerance, which enables keeping the working procedures from E.coli. It was shown, that catalase supplemented on agar plates could improve the viability under cold conditions. Nevertheless a more usable approach would be to include this tolerance within the strain itself. Several approaches exist to realize this goal, for instance increasing the glutathione pool within the cell or by introducing a plasmid with an encoded catalase, such as KatG, KatE or the ahpCF operon.
We decided on introducing KatG into the genome, due to previous results, which were reported by Weinstock et al. They could show that KatG heterologous expressed from plasmid extends the time for fridge storage up to 3 weeks, which would suit for the needs for a cloning host. For integrating the katG into the genome, we combined introducing katG with the deletion of the second nuclease by utilizing 3000 bp homologous overhangs of the upstream and downstream flanking genomic region of exeM. These overhangs are fused to the katG transcription unit, that is amplified from the E.coli DH5𝛼 genome, by fusion PCR and subsequently transformed via natural transformation as a cotransformation.

---

The lac operon

The lac operon of E. coli consists of the three genes lacI, lacA and lacY. In the cloning strain, the fact that the β-galactokinase cleaves X-Gal into a blue dye became apparent. The lacZ gene encodes the β-galactokinase which consists of two parts: the lacZ-alpha and the lacZ-omega. It has been shown that if the alpha fragment was deleted, the omega fragment is not functional. By introduced the alpha fragment on a plasmid, alpha-complementation occurs. (Link alpha complementation) We also wanted to make the blue-and-white selection possible for our VibriClone, as it is in the DH5𝛼 strain of E. coli. Therefore we checked at first which lac genes are present in the wildtype of V. natriegens. We found the lacA gene on chromosom 1 and the lacZ and lacI gene on chromosom 2.
As you can see, several parts are missing or different in all three proteins (see Alignment). What indicates that the lac genes are not functional in V. natriegens, which makes sense as lactose can not be found in its natural habitat and therefore genes for lactose metabolism are not needed and no selection pressure is present. That's the reason why we decided to amplify the lac Operon from DH5𝛼 that has no lacZ alpha subunit.

Next we considered whether the wildtype lacZ from V. natriegens has a functional alpha subunit. If it exists, the 𝛼-complementation would not work anymore. To check this, we aligned the alpha subunit with the lacZ gene of V. natriegens (Figure 5). You can see that many parts of the subunit are missing, but a few sequences are conserved. To avoid that it can come to a complementation, we wanted to delete the part in our VibriClone. Since it appears that the entire lacZ may be nonfunctional, we decided to bring the amplified lactose operon on this place.
To achieve that we build a cassette consist of the upstream and downstream regions of lacZ and the lactose operon from DH5𝛼. The lactose operon of E. coli has a size of 6326bp. Since that is very large we decided to use smaller homologous flanks. It was shown that the natural co-transformation work well with 1000bp flanks, so we use this size. Otherwise the linear DNA Fragment has a size of ~12000bp. Thus, it is possible for us to introduce the lactose operon by cotransformation as an unselected product with the other described products.

With this kind of genome modification, it is possible for the entire scientific community to immediately recognize a successful transformation with VibriClone. This would save not only through the fast growth rate but also through the method of cloning an incredible amount of working time.

---

Outlook

After developing the first cloning strain of Vibrio natriegens to fully replace E. coli as the most used cloning host in synthetic biology, we asked ourselves what would be the next step. VibriClone provides a starting point for further improvements for instance in case of special cloning application. When we created this strain we completed a full round of the design build test cycle and as we want to bring engineering into biology we started with the next cycle by designing VibriClone 2.0.
Therefore several different strains should be established in parallel to full fill the demand for more specialized applications. For E. coli several different strains exist, which are utilized for example for different sized constructs (TOP10), for allowing genome engineering tool, such as MAGE (mutS deficient), for in vivo mutagenesis, for toxic gene cloning or for preventing unwanted methylation.
In the beginning of our project we did next generation sequencing for our Vibrio natriegens strain and performed bioinformatic analysis and identified several other candidates for deletion and other genome engineering methods. One design feature for our 2.0 strain would include deleting methylase and restriction modification systems, which are present in our genome. In common cloning strains for E.Coli (dH5𝛼, Top10) restriction modification systems are deleted (Δ(mrr-hsdRMS-mcrBC) to enable cloning with DNA from different sources and to protect the DNA from endogenous restriction enzymes. That's why a desirable goal for the next generation of our cloning strain, would be to design a strain, similar to E.coli, which is deficient for all for those modification systems.

When we performed next generation sequencing in the beginning of our project for our wild type strain, we subsequently did a bioinformatic analysis and genome annotation. By utilizing these information we were able to identified several methylases and restriction modification systems.

Type I restriction - modification system subunit M	EC 2 . 1 . 1 . 72
Type I restriction . modification system , specificity subunit S	EC 3. 1.21.3
Type I restriction modification system restriction subunit R	EC 3 . 1 . 21 . 3
Type III restriction - modification system methylation subunit	EC 2.1 . 1.72

We are sure, that more methylases and restriction modification systems can be found in the genome of Vibrio natriegens by utilizing and combining different bioinformatic tools such as the rebase database.
Additionally we found several phage associated sequences and it has been shown that deleting phage related regions could make a bacteria more genetically accessible. In general a genome reduction could have advantages for a cloning host, as it has been shown that it could increase genome stability, reduce mutation rates and improved recombinational stability for cloning repetitive DNA sequences.
The most ambitious idea for our VibriClone 2.0 strain would be to free up chromosome II for utilization as a low copy plasmid in order to clone big DNA fragments or whole genomes and thereby establishing a synthetic chromosome for challenging cloning applications. In the future this could enable cloning of small of hole small genomes, such as the mycoplasma genome in order to make the genome engineering process easier and especially much faster, compared to common methods, where whole genome cloning heavily relies on yeast as a cloning chassis. Combining the fast growth of Vibrio natriegens with powerful genome engineering methods like Crispr Cas, MAGE, CAGE, the ability for natural transformation and MuGENT, would allow rapid genome prototyping and could open up complete new possibilities for building up synthetic genomes from scratch. This would enable engineering customizable microbes with complete new properties specifically tailored for the individual needs for a broad range of applications.

VibriExpress

Sometimes it is really hard to part with old things and adopt to new ones. But as scientists, it is our responsibility to try out new and unexplored things in order to advance progress and improve on already known. To convince you all, to start working with a new chassis we will show that all commonly used techniques are transferable from E. coli to V. natriegens, without much effort. Since it is important to overexpress and purify proteins in heterologous systems, a fast growing protein expression strain is required. Before starting to overexpress a protein, you need to make important decisions: “What kind of system would be the best?”, “Which chassis could be the most effective” or “Do I need a chassis at all, what about cell-free-expressions systems?”
There is not the one and only right way to overexpress a protein. For each protein, other aspects have to be considered in order to make a good decision. These decisions depend on different characteristics of the desired protein, like its solubility, if it is a bacterial protein or a eukaryotic one or if it is toxic or not.
In E. coli, several expression systems are established to cover a broad spectrum, but the most commonly used is the T7 expression system. To enable scientists, to overexpress their proteins much faster and purify a higher protein yield, we decided to establish the T7 system in the world’s fastest growing organism V. natriegens and did further analysis to learn which genetic modifications are needed to use its full potential. As a model, we were guided by the E. coli strain BL21 (DE3). This E. coli strain is often used for heterologous protein overexpression. In our opinion, the most important features of BL21 (DE3) are the genomic integration of the T7 RNA Polymerase gene and the deletion of proteases like Lon and OmpT to improve the protein yield by decreasing the protein degradation (Dunn & Studier 1983) , (Kim et al. 2015) . We will try two different strategies to create a genomic T7 cassette integration (Figure 2). Because it was shown, that the deletion of dns is not harmful ( Hesek et al. 2016) , we decided to integrate the sequence of the T7 cassette at the endogenous locus of dns. In parallel, we will integrate the T7 cassette in one of our successfully characterized integration sites.
Looking at the genotype of E. coli BL21 we noticed that lacI localizes at the upstream site of the T7 RNA Polymerase gene (Kim et al. 2015) . LacI is the negative regulator of the lac operon (Gilbert & Muller-Hill 1966) . In case of the T7 system, the lacUV5 promoter controls the T7 RNA Polymerase expression ( Studier & Moffatt 1986) . By integrating the repressor LacI at the same time into the V. natriegens genome we will reduce the leakiness of the lacUV5 promoter, as it was done in E. coli. Additionally we checked if there are genes encoding proteins similar to Lon or OmpT, because these two were deleted in E. coli BL21 (DE3) to minimize the degradation of proteins (Dunn & Studier 1983) ,(Kim et al. 2015) . By BLAST analysis, we could find a Lon protease with a high degree of similarity to the Lon protease of E. coli in V. natriegens (Figure 1). We could not find any OmpT homologs neither by BLAST analysis nor by an ompT annotation in the genome of V. natriegens. Because of the high similarity of the Lon proteases to each other, it is likely that the Lon of V. natriegens fulfils a similar function like the Lon protease in E. coli. This convinced us to delete the lon of V. natriegens.

As already mentioned we will follow two different strategies to create the T7 integration. For the first strategy, we will fuse the T7 cassette. Amplified from the BL21 genome to a chloramphenicol resistance cassette and to 1000-bp up- and downstream homologous regions to dns. Since it is possible to create two or more genomic modifications, using only one selected product in natural co-transformation of linear fragments (Hayes et al. 2017) , we will use another fragment with homologous up- and downstream regions of lon but without a selection marker in between, to delete lon.
For the second strategy, we will integrate the T7 cassette to one of our characterized integration sites. For this reason, we will fuse the T7 cassette to 1000 bp long up- and downstream homologous regions to integration site 19 (int19). For the deletion of lon, we will use the same linear fragment as in strategy one. As a selected product with an antibiotics marker in between we will utilize the deletion cassette as for Δdns in VibriClone.

As we will attach FRT sites to the antibiotic cassette by PCR, we will be able to remove the selection marker after integration of the selected product, using a plasmid that encodes a Flip recombinase (link FLP recombinase). After the genomic modifications are proved by PCRs, the T7 system needs to be tested and VibriExpress to be characterized. First of all a growth curve of this new strain has to be measured in comparison to E. coli BL21 to show you, that V. antigens is still faster than E. coli. Next, we will prove, that the cells are healthy after several genomic modifications by microscopic analysis. To show you, that changing from E. coli to V. natriegens will be easy and to prove, that tools like the T7 system are transferable without much effort, we will test a construct containing a T7 promoter in both organisms and compare the results to each other. We are convinced that the T7 system in VibriXpress will deliver a much higher amount of protein in a shorter time. This way we would like to speed up your scientific progress. In 100 years of research with E. coli, BL21 is not the only used strain with a T7 expression system. Strains like Lemo21(DE3) (commercialized by NEB) are designed for fine-tuning of T7 expression especially for difficult targets. This fine-tuning is achieved due to the variation of the lysozyme (lysY) level, a gene encoding a variant of phage T7 lysozyme. The T7 Lysozyme is an inhibitor of T7 RNA Polymerase (Moffatt & Studier 1987) , (Zhang & Studier 1997) . By reducing the basal T7 RNA Polymerase activity toxic genes can be expressed in the same cell under control of a T7 promoter. Because of its ability to inhibit transcription the T7 lysozyme is often used for controlling expression in systems based on T7 RNA Polymerase ( Studier & Moffatt 1986).

In strains like Lemo21(DE3) the level of lysozyme can be optimized by inducing LysY with L-rhamnose to alleviate inclusion body formation or inhibitory effects from toxic proteins. In future, we will need to engineer this kind of specialized strains for troubleshooting as well in V. natriegens.
In addition, more research is necessary to study V. natriegens specific proteases, their function and their strength. Of course, we could not find an aspartyl protease similar to OmpT of E. coli, but this does not rule out that V. natriegens possesses other proteases that are crucial for protein degradation during overexpression. Finding highly active proteases and deleting them would further improve the amount of produced protein.
What about all the scientists that would like to use other expression systems than T7? As we mentioned before the T7 system is not always the first choice for protein overexpression. Other commonly used promoters like the synthetic trc promoter or T5 promoter are recognized from E. coli RNA Polymerases and do not require a genomic integration of a phage RNA polymerase (Erfle et al. 1985) , (Gentz & Bujard 1985) . We could show, that these promoters are recognized from the V. natriegens RNA polymerase as well (Link metabolic engineering), enabling you to use the potential of this impressing organism in a variety of tools and benefitting of its speed. This result, together with all promising facts about V. natriegens like the higher solubility of some proteins that are difficult to purify and the simple transferability of E. coli tools to this organism, have strengthen our vision that V. natriegens is the perfect chassis to accelerate scientific progress in synthetic biology.

Performing genome engineering in bacteria using CRISPR/Cas9 usually requires the presence of three components in a cell at the same time, namely: Cas9, gRNA and a donor template (Wasels et al. 2017). Therefore we aimed to design a convenient system that allows easy adaption of CRISPR/Cas9 to target various sequences.

Preparation of Cas9
A Cas9 and dCas9 LVL0 part was created as a CDS part in our Marburg Collection. The cas9 sequence was PCR amplified from pCas9CR4 (Reisch and Prather 2017). To make this sequence compatible with our toolbox and the iGEM registry, we removed a BsmBI and EcoRI recognition site in the parts sequence creating the Cas9 LVL0 part BBa_K2560047. For further experiments, a dCas9 part could prove useful and therefore we induced mutations the nuclease loop in the RuvC1 (D10A) & HNH (H840A) subunits creating BBa_K2560054.

To establish CRISPR/Cas9 in V. natriegens, we conceived a Cas9 plasmid where the expression is tightly controlled by an inducible promoter. We expect this to help during cloning and transformation steps, as the toxicity of Cas9 expression is reduced in the absence of inducer. We chose pTet for this purpose. This decision is based on the characterization of this part that showed tight control and a 40 fold induction of this promoter when induced with ATc (Link to Results) . Finally, we created LVL1-pTet-Cas9, a plasmid with the Cas9 CDS under control of pTet (figure 1). We chose to built this plasmid with p15A as the ori and a kanamycin resistance cassette. This enables cotransformation of this plasmid together with a second plasmid harboring the gRNA cassette.

Construction of a gRNA entry Vector
To enable simple cloning of gRNAs, we constructed BBa_K2560305 an entry vector for gRNA spacer (figure 2). The spacer is defined as the 20 bp at the 5’ end of the gRNA that confers sequence specificity to the target sequence in the genome. This parts design is based on BBa_K2457002 a part which harbors a gRNA targeting LacZ of E. coli. To achieve superior cloning flexibility we incorporated the approach of using a sfGFP dropout similar to the part entry vector of the Marburg Collection BBa_K2560002. Analogous to the cloning of small LVL0 parts, new spacer sequences can easily be cloned by annealing of oligos and subsequent ligation into the gRNA entry vector using a golden-gate reaction. This drop out helps to distinguish correctly assembled plasmids from the religated entry plasmid. The succesfully cloned plasmid then contains a gRNA expression cassette as well as a pMB1 ori and the chloramphenicol resistance cassette.

The two plasmids LVL1-pTet-Cas9 and BBa_K2560305 after integration of a spacer are compatible and can be cotransformed. As a third component a donor template has to be added to the transformation.

In most bacterial CRISPR/Cas9 approaches, a two plasmid system is used separating Cas9 and gRNA on two different plasmids (Wasels et al. 2017). The reason is that during the process of gRNA cloning, a functional Cas9 transcription unit is present in the same cell with the correctly cloned gRNA, resulting in double strand breaks (DSB) leading to the death of the cells.

Our approach to overcome this challenge is to use E. coli as cloning chassis for the construction of the Cas9-gRNA plasmid which is then transformed in V. natriegens together with a donor template to perform genome engineering. This is possible because the spacer sequence targeting the genome of V. natriegens should not match any sequence in the genome of E. coli and therefore we do not expect creation of DSB.

Designing of donor templates
Different approaches for the design of donor templates are described in literature. The mutation can be introduced either as dsDNA in a linear form or integrated into a plasmid or as ssDNA (Jiang et al. 2015).

We decided to use ssDNA because they can be easily ordered from companies offering synthesis. As a prove of concept we decided to tackle the dns gene on chromosome 1 of V. natriegens, as this mutation has already been shown not to be lethal and to significantly increase DNA yields of plasmid preparations (Lee et al. 2017). We used two ssDNA oligos each targeting the leading or lagging strand. The oligos consists of the desired mutation flanked by 50 bp long homolougous sequences flanking the position of the introduced DSB. The last three nucleotides at both ends are connected via phosphorothioate-bonds providing the stabilization of the donor oligonucleotides and protection against digestion (Reisch and Prather 2017).

The donor template was designed to exchange four basepairs at the beginning of the dns gene with a three basepair stop-codon (figure 4). In this way, the knock-out is ensured through the premature translation termination as well as a frameshift in the sequence. The spacer sequences targeting the genomic sequence was designed using the CRISPR design tool of Geneious which utilizes the algorithm developed by Doensch et al. (2014).

Our workflow

One of the central elements of synthetic biology is the Design-Build-Test-Learn cycle. It illustrates the workflow of how systems like metabolic pathways, genetic circuits or anything else are constructed and evaluated. The first step is to design the system basing on the existent knowledge about the system. Then the system is built and its functionality is tested. The results of these tests describe the ‘Learning’ part of the cycle. They are used to make an improved design. Finally, a new round in the DBTL cycle can be made and the system gets improved in each round of the cycle. For our approach we decided to establish a workflow for metabolic engineering, which bases on the DBTL cycle and combines the existing principles of metabolic engineering with the advantages of the fast-growing organism V. natriegens. First of all, we use the Marburg Toolbox to construct a library of pathways with different combinations of promoters, ribosome binding sites, coding regions, enzyme ratios and plasmid copy numbers. Next, we transform the plasmids for each pathway version together with the plasmid for the biosensor into our chassis. When the pathway is now performed, the cells produce a fluorescence or luminescence signal and its intensity correlates to the 3-hydroxypropionate ( 3HPA-) concentration. By using FACS we can select for the most efficient pathway versions. By using multiplex automated genome engineering (MAGE), refeeding 3HPA into the central metabolism and repetition of FACS selection we have a cycle of directed evolution-guided pathway optimization. The strains with the most promising pathway variants can then be isolated and their pathway composition elucidated. Thus, we get information about designing principles for our pathway and can propose an improved design for a new round in the cycle of metabolic engineering.

Our pathway

As mentioned in the description, there are many ways to produce 3HPA. We decided for a pathway, basing on the conversion of acetyl-CoA via malonyl-CoA to 3HPA. According to (Valdehuesa et al. 2013) who evaluated many pyruvate-derived production pathways from a thermodynamical point of view, this is one of the most efficient routes for 3HPA production. We decided for this route for several reasons. First it bases on glucose degradation and V. natriegens has an unbeaten glucose-uptake rate, enabling a high glucose consumption rate. Second there is no vitamin B12 necessary, the only cofactors which are involved are NADPH and biotin, both of them naturally occurring in the organism. Third there is theoretically just one further enzyme needed to complete the pathway. All enzymes necessary for the production of malonyl-CoA are still present in V. natriegens and we just have to integrate the last enzyme, malonyl-CoA reductase. Nevertheless, overexpression of acetyl-CoA carboxylase increases productivity, so we took both enzymes into consideration. (Kildegaard et al. 2016) proposed, that there are many ways to further optimize the pathway and direct the flow to a high product titer. Hence, implementing many variations of this pathway into V. natriegens and testing them in a fast manner promises a high gain of knowledge about pathway optimization and general principles of metabolic engineering.

Choice of enzymes

For the core pathway, we need to implement two enzymes, acetyl-CoA carboxylase (Acc) and malonyl-CoA reductase (Mcr). For Acc there are several candidates, which have still been tested for synthetic metabolisms ((Kildegaard et al. 2016); (Liu et al. 2013); (Vidra et al. 2017)). Our plan was to first test enzymes from different organisms and then optimize expression profiles. To benefit from the fast engineering workflow in V. natriegens, we decided to use enzymes from different organisms. We planned to use the native Acc from V. natriegens, because it is still present in the cells and we prevent unnecessary burden by overexpressing a heterologous Acc. The native Acc consists of four subunits together with an associated biotin ligase. The genetic organization is the same like in E. coli, which is shown in figure 3. Nevertheless, further looks into literature revealed, that heterologous expression of pathway enzymes could be more promising because housekeeping Accs often are strictly regulated while foreign genes have different regulation motifs. Hence, we also used the Acc from E. coli together with its associated biotin ligase (BirA). These enzymes have already been tested for 3HPA production (Liu et al. 2013). Additionaly, we decided to test the enzyme from Synechococcus elongatus. Since cyanobacteria require higher amounts of fatty acids to form thylakoid membranes, they might have more active enzymes. The S. elongatus Acc consists of four subunits, just like those from E. coli and V. natriegens. For this enzyme it was also recommended to use it with a biotin ligase, because it needs a covalently bound biotin to be activ. Unfortunately Accs consisting of four subunits form a very unstable complex and unbalanced expression levels of its subunits lead to gene repression (Abdel-Hamid et al. 2006). To circumvent this issue, we also chose the Acc from C. glutamicum. In this organism, AccB and AccC are fused to one subunit, as well as AccA and AccD. Balancing of two subunits instead of four is much easier, therefore this enzyme is beneficial for us. Gande et al., 2007 still characterized this enzyme and confirmed its applicability for synthetic biology. Since C. glutamicum is a gram-positive bacterium, we decided to codon optimize the genes for a smoother translation and coexpressed the associated biotin ligase (Accession numbers NCgl0670, NCgl0678, NCgl0679).

The second enzyme in our pathway is the malonyl-CoA reductase (Mcr). Since this enzyme only occurs in autotrophic organisms, performing the 3-hydroxypropionate bicycle, there aren’t many candidates for our enzyme choice. In literature Mcr from Chloroflexus aurantiacus has mostly been used, because it is easy to purify, has a high activity and is still active at moderate temperatures (Hügler et al. 2002) ; (Liu et al. 2017). We decided to use this enzyme and test it with different expression levels and Acc/Mcr ratios. However, recently it was shown, that a dissection of the enzyme strongly increases productivity (Liu et al. 2013). In general, Mcr has two functional domains, the C-terminal domain catalyzes the reduction of malonyl-CoA to 3-oxopropanoate and the N-terminal domain catalyzes the further reduction to 3HPA. In the mentioned publication there wasn’t an exact explanation for the increased productivity of the dissected enzymes. Our hypothesis is, that the complete enzyme, which has its temperature optimum at 60°C, has a suboptimal protein folding at 37°C while the dissected enzymes are smaller and less bulky and therefore can fold easier. Deeper investigations of enzyme behavior additionally revealed that the C-terminal domain is much slower than the N-terminal domain and thereby the C-domain slows down the overall reaction (Liu et al. 2013). Therefore, we decided to use the complete enzyme of Mcr as well as the dissected enzymes McrC and McrN. There is also another Mcr from Sulfolobus tokodaii, which only catalyzes the rate-limiting step from malonyl-CoA to 3-oxopropanoic acid. Since McrC has the same function, we planned to coexpress Mcr from S. tokodaii together with McrN from C. aurantiacus and test, if the activity increases. All in all, we planned to use the acetyl-CoA carboxylase from E. coli, V. natriegens, C. glutamicum and S. elongatus, the Mcr from S. tokodaii and the Mcr, McrC and McrN from C. aurantiacus. First, we wanted to test enzyme activity and afterwards use the Marburg toolbox to build combinations of enzymes and expression levels.

Rational design of plasmids

Synthetic biology proceeded significantly in recent years and we have much knowledge about rational design of circuits and pathways. Nevertheless, failures and unexpected outcomes slowdown scientific progress. To tackle this issue, we decided to follow two approaches. On the one hand we planned to model our pathway and predict a – theoretically – optimal pathway, which we can then construct and test. And on the other hand, we decided to build additionally to this particular pathway a whole library of many variations of this pathway with little differences. In this way we create many producer strains in parallel and speed up the metabolic engineering workflow. Our rational approach bases on a Michaelis-Menton-based metabolic analysis. For more details see our metabolic model. Since there are no stochiometric flux data about V. natriegens yet, we used the data from E. coli, which has a very comparable metabolome to V. natriegens (citation needed). For the model we only considered acc from C. glutamicum and mcr from C. aurantiacus, because they were from our point of view the most promising ones. The model predicted, that Acc is the rate limiting enzyme in the pathway and has to be expressed ten times stronger than Mcr. We used the promoter characterization data, obtained from our part collection team (For more details see the section about characterization of the promoters of our Marburg toolbox) and introduced them into the model. The resulting prediction was, that we have to use the promoter J23106 for acc and J23109 for mcr. Therefore, we designed the pathway plasmid according to this model. We built the plasmid based on the Marburg Toolbox with 4 transcriptional units. For the ori we chose ColEI, which has a broad host range and a high copy number in V. natriegens and E. coli (Jain et al. 2013), [Link zu qPCR von Carlos]). We decided for that one, because V. natriegens shall be a chassis for testing pathway variants, which then maybe will be transferred into other strains, therefore a broad host range vector is reasonable. For accBC and accD we chose the promoter J23106 and for mcr the promoter J23109 according to the model. For birA we also chose J23109 as promoter, because BirA is only relevant for activation of Acc and not for direct reaction catalysis. For ribosome binding sites and terminators, we used the same for each coding region, because otherwise we would have fluctuations which weren’t considered in the model. An illustration of the plasmid is shown in figure 5.

Pathway library

For the second approach of pathway design, where we build a whole library, we also used the Marburg Toolbox [Marburg Toolbox soll ein Link zur Marburg Toolbox sein] to combine different promoters, ribosome binding sites and coding regions. In theory, the toolbox enables the plasmid assembly of 21780 different part combinations per transcriptional unit (22 promoters x 5 rbs x 6 terminators x 3 oris x 11 resistance cassettes; connectors weren’t considered). For a plasmid containing accBC, accD, birA and mcr that would make 6 x 1012 possible constructs, but assembly of so many versions is irrational. The better way is to select a few parts which are combined and then for the most promising versions one can think of more combinations. Therefore, we focused on eight promoters and three ribosome binding sites with different strengths. Our idea was to build pathway plasmids containing the whole pathway, as well as plasmids containing only the acetyl-CoA carboxylase genes or only the mcr gene, which can then be cotransformed (Table 1). To enable cotransformation, we used the oris ColEI and p15a, which are compatible to each other.

The idea behind the plasmids with the whole pathway is, that the cells only have to maintain one particular plasmid. The strategy for the single-enzyme plasmids is that by cotransformation different combinations can be made to increase the number of possible pathway variants. The different plasmid types can be seen in table 1, all in all we designed the construction of 1446 plasmids for our pathway library. For plasmids of category I (see table 1), we combined eight promoters with three ribosome binding sites, a kanamycin resistance gene and a p15a ori, giving us 24 constructs. For plasmids of category II and III we used the same promoters and ribosome binding sites for both transcriptional units, leading to 576 plasmids respectively. For plasmids of category IV we restricted our selection to three promoters and three ribosome binding sites. Since a misbalance of accBC- and accD expression levels leads to occlusion bodies and decreased activity we decided to always use the same promoters and rbs for both genes. In this category we get 81 plasmid variants. For categories V, VI and VII, containing the whole pathway, the combinatorial possibilities were very high, so we decided to further restrict the selection by always coupling strong promoters to strong rbs, medium promoter to medium rbs and weak promoter to weak rbs. Therefore, we got a total of 27 plasmids for category V and 81 plasmids for category VI and VII, respectively. We also considered to integrate one or several genes, therefore we planned to search for integration sites and to characterize them. [Link zu Bauersachs Teil]

Our biosensors

The design of our biosensors was largely guided by the goal of having as little impact on the host metabolism, while simultaneously being simple and easy to implement. We chose to design all parts to be compatible with the Marburg toolbox, RFC10, RFC25, andRFC1000, to make it as easy as possible for us and future users to implement it and adapt it to new requirements.

Annotation of the location for the regulatory sequences upstream of the HdpR coding sequence was insufficient for us to precisely remove the activator binding site or the constitutive promoter driving HdpR expression. So, we included an about 200bp long region upstream of the HdpR, which should be enough to include all regulatory motifs. Also included is the HdpR coding sequence with its own ribosomal binding site (RBS). and promoter. This results in a part that includes the whole sensing function, but does not quite fit our scheme of clearly defined functional units. Since expression of any part cloned behind the regulatory region is directly controlled by it, and the part serves as both, a promoter and RBS. It is basically an inducible promoter fused to an RBS. Thus, we decided to give the part the overhangs of an RBS. on the downstream and a promoter on the upstream end.

In order to obtain the desired region from the P. putida KT2440 genome, we planned to amplify it via PCR and then isolate the fragment by gel purification of the corresponding band. Into the primer-design we factored in the overhangs we needed to make it Marburg toolbox compatible. So, we gave them non-binding overhangs which automatically added the necessary BsmBI cut sites.

For the malonyl-CoA sensor, our proceeding was very similar. Just that we knew the exact location of all functional parts, except of the location of the promoter, form the 2015 paper by Liu et al and from annotations in the published genome (Liu et al. 2015). Therefore, we could implement it fully into the Marburg toolbox, by simply treating the FapR coding sequence as coding sequence. Not that straightforwardly was the design of the fapO regulator binding site. Since we did not know the promoter strength, and the mechanism of regulation was independent of the promoter sequence, we decided to implement a promoter from the Andersen library, J23100. By placing it right in front of the fapO site, we created a construct that, in theory, should give us a FapR controlled promoter. For the amplification of the individual parts from the B. subtilis 168 genome, again, we used primers sporting non-binding regions, adding the necessary overhangs for our toolbox.

For future projects, other sensors could also be employed. In a publication from 2016, Rogers et al. propose an alternative sensor for 3HPA with a lower perception threshold (Rogerset al.2016). We choose not to implement it in our design because of its more complicated mechanism that requires the implementation of more genes. Intriguingly, it works by first converting 3HPA to acrylate via a pathway of three heterologous enzymes. The acrylate is then sensed by a transcription factor, similarly to how our sensors described above work. They also report its successful implementation and application in increasing the 3HPA titer up to 23-fold. It is conceivable that with a little more effort our system could yield comparable results. Off cause, s we have the additional benefit of having a much shorter iteration interval.

Further product sceening

The usage of biosensors for product screening is a fast and cheap method enabling high throughput analysis of pathway libraries. However, to further characterize the pathway variants and implemented enzymes we planned to measure enzyme activities. In that way we can quantitatively benchmark each enzyme and establish in vitro assays as a prototyping engineering approach. For the malonyl-CoA reductase, it is easy to measure reaction velocity, because the enzyme converts NADPH into NAD+ and both compounds have different absorbances at 345nm (De Ruyck et al. 2007). Based on that we designed an enzyme assay, where we use cell extract from Mcr producing cells or directly purified Mcr and mixed it with a defined amount of NADPH and MgCl2. By measuring the absorbance, one can determine if NADPH concentration changes. By adding malonyl-CoA to the reaction mix we then start the reaction and are able to observe a decrease in absorbance because the reaction consumes NADPH. Here the starting concentrations of malonyl-CoA and NADPH have to be high enough to ensure, that nearly each Mcr molecule in the mixture is saturated and the reaction goes proportional to enzyme concentration. By determining the slope of the linear decrease of absorbance it is possible to calculate the specific activity. The specific activity is a value which characterizes the ability of enzymes, to convert a substrate into a product. It is defined as the moles of product formed by an enzyme in a given amount of time under given conditions per milligram of total proteins, the unit is µmol min-1mg-1. To calculate the specific activity, one needs the slope of the measured absorbance. This value defines the change of absorbance per time. By using the Lambert-Beer law it is possible to calculate the amount of NADPH converted into NADP+ per time (Lambert and Anding, 1760).

Directed evolution

Metabolic Engineering offers many possibilities to create suitable pathways but it consists of more than just building them. Pathway optimization is also crucial for efficient productivity. In our workflow for metabolic engineering we want to implement directed evolution to drive metabolic fluxes towards 3HPA production. We elaborated two strategies to do that. Our first strategy was to implement genes for the conversion of 3HPA into succinate. In that way, we lead our product back into the central metabolism. By deleting all other genes responsible for succinate production, the cells are forced to use the 3HPA bypass to get succinate, which is needed in the TCA cycle. The big advantage of using V. natriegens for that purpose is that its short doubling time increases the number of cell divisions per time and thereby increases the number of mutations in a given time frame. Our second strategy is to use MAGE [Link zu project→description→metabolic engineering→Directed evolution→MAGE erklärung] together with our biosensors and fluorescence-activated cell sorting ( FACS ), to cause various genome modifications and rapidly select for cells with increased 3HPA productivity (Figure 9). We looked into literature and found 16 possible genome modifications for 3HPA production made by other groups. Table 2 shows a list of all modifications. Our plan was to order oligos for each of these modifications, leading to upregulation, downregulation, deletion or site-directed mutagenesis of the respective gene. The aimed modification for each gene is shown in the table. For the next step we decided to use natural transformation for integration of these oligos into the strain. Because each cell will take up different numbers and different combinations of oligos, we will get a population of genomic variations [Link zu Project → Design → Strainies → Mugent]. The genetically diverse population can then be put into FACS and we can select for cells with the strongest signal and thereby the best productivity. Finally, the whole cycle of MAGE and FACS can be repeated, leading to continuously improved pathways. We believe that the combination of MAGE- FACS with the fast growth of V. natriegens will speed up the process of metabolic engineering and pathway optimization significantly!

Outlook

In our project we try to accelerate the process of metabolic engineering in all of its facets and thereby find a suitable pathway variant to achieve high 3HPA productivity. Nevertheless, there will still be many possibilities to further improve the production of 3HPA in V. natriegens. Table 2 shows a list of 16 genome modifications, that were made in other organisms with the same or a similar pathway to direct the flow towards 3HPA. Some of these modifications target for instance NADPH homeostasis. During glycolysis cells produce NADH while the production of 3HPA depends on NADPH. Therefore, a misbalance between NADH and NADPH could arise, but overexpression of pntAB (nicotinamide nucleotide transhydrogenase) or gapA (NADP+-dependent glycerol-3-phosphate dehydrogenase) has been shown to stabilize cofactor balance (See table for references). Another possibility to enrich the NADPH pool is to downregulate glycolysis while upregulating pentose phosphate pathway, which produces NADPH instead of NADH. This can be achieved by deletion of NAD+-dependent gapA or deletion of pfkA (phosphofructokinase). Another critical point in the 3HPA route are side reactions. Pyruvate can be reduced to lactate or decarboxylated to acetate, what could be suppressed by deletion of ldhA (lactate dehydrogenase) or pox (pyruvate decarboxylase). The flow from pyruvate to acetyl-CoA can be increased by overexpression of pdh (pyruvate dehydrogenase). For acetyl-CoA there are also side reactions, for instance hydrolysis to acetate. This reaction can be suppressed by deletion of pta (phosphotransacetylase) and ackA (acetate kinase) or by overexpression of ald6 (aldehyde dehydrogenase) and acs (acetyl-CoA synthase) for the reverse reaction to acetyl-CoA. Malonyl-CoA can also go into side reactions for fatty acid biosynthesis instead of the reduction to 3HPA. However, downregulation of fabAB (fatty acid biosynthesis) or inducible inactivation of fabAB has been showed to increase 3HPA titer. Additionally it has also been shown that site-directed mutagenesis of acc and mcr genes increases product titer (Kildegaard et al. 2016); (Liu et. al. 2016).

We chose the production of 3HPA as application to show the functionality of the Marburg toolbox and as proof-of-principle for our metabolic engineering workflow. However, our goal is not to produce high amounts of 3HPA, but rather to show that V. natriegens can be used as test chassis for rapid testing of many pathway variations to find the best one. The next step would therefore be to evaluate, if the pathway with its genetic composition can be transmitted to other organisms to see if the route still works. Besides that, there are many other compounds of medical and economical relevance, which still can’t be produced biotechnologically. The production of the anticancer drug taxol involves for instance 19 enzymes and is therefore a very challenging pathway with lot of bottlenecks and tunable reactions (Croteau et al., 2006). We think it would be beneficial to use V. natriegens as chassis for testing this challenging pathway or parts of it. Its short doubling time and genetic accessibility will accelerate the process of enzyme engineering and finetune expression levels or gene combinations. Another crucial aspect, that has to be tackled is the universality of our workflow. We involved biosensors for product screening but for many metabolites there are still no known sensor proteins. There are three possibilities to circumvent this issue. The first is to find new biosensors, what is complicated, because many compounds, especially drugs for medical purposes don’t occur in the nature, therefore it is unlikely that there are still proteins binding specifically to these compounds. The second possibility would be to engineer proteins to become drug-specific sensors. The third possibility is to upscale LC-MS techniques to enable larger sets of measurements and faster screening. LC-MS has an unbeaten accuracy but its bottleneck is the velocity. The average time needed for one sample lies between 1 and 10 minutes, making measurements of large sets of samples impossible. Furthermore LC-MS requires high amounts of buffer. However, as mentioned in the Labautomation-section [soll ein Link sein], the company Labcyte developed a system to use acoustic waves for pipetting nanoliter droplets into microtiter plates in a very fast manner. Now they invented an acoustic mass spectrometer, which combines acoustic nanodroplet formation with mass spectrometry. The device is able to measure 3 samples per second without the need of a mobile phase. Such a device could be used instead of biosensor-based flow cytometry to broader the applicability of our metabolic engineering workflow.

References