Design and use of synthetic regulatory small RNAs. to control gene expression in Escherichia coli

Transcription

1 Supplementary Discussion Design and use of synthetic regulatory small RNAs to control gene expression in Escherichia coli Seung Min Yoo, Dokyun Na, & Sang Yup Lee * Metabolic and Biomolecular Engineering National Research Laboratory, Department of Chemical and Biomolecular Engineering (BK21 program), BioProcess Engineering Research Center, Bioinformatics Research Center, Center for Systems and Synthetic Biotechnology, Institute for the BioCentury, Korea Advanced Institute of Science and Technology (KAIST), 291 Daehak-ro, Yuseong-gu, Daejeon , Republic of Korea. 1

2 Designing tailor-made synthetic srnas in assistance with computational tools. To select the best combination of a host strain and target genes simultaneously in performing metabolic engineering towards the enhanced production of a desired chemical compound, the use of synthetic srna library is time- and cost-efficient. When constructing a synthetic srna library, it is useful to check whether the target-binding sequences can bind to undesired genes with similar sequences, such as those encoding isozymes or evolutionarily closely related genes. Potential off-targets of synthetic srnas could be predicted by using several bioinformatics tools to identify potential target mrnas of a queried srna sequence. They employ the calculation of srna-mrna hybridization free energy and the accessibility of binding sites, and some of them use machine learning approaches to improve their accuracy further. The reported algorithms and their web sites are listed in table 1 in Supplementary Discussion. It should be noted that these algorithms can still produce many false predictions even though the accuracy of prediction is being improved. Among the available algorithms, we describe how to predict possible cross-reactions of a tailor-made synthetic srna by using the TargetRNA2 ( see table 1 in Supplementary Discussion). TargetRNA2 is based on the widely used features, such as accessibility of binding sites in srna and mrna, and their binding free energy, to predict the target mrnas. Its prediction accuracy is relatively higher than others. It allows searching for genomic mrnas of various bacteria including diverse strains of E. coli, so that one can easily check unexpected binding of synthetic srnas to the mrnas at genome-wide scale. In order to see whether a synthetic srna could bind and repress off-target mrnas, the whole synthetic srna sequence (that is, the target-binding sequence plus the scaffold sequence) should be designed first. On the TargetRNA2 website 2

3 ( one can find two input textboxes: one for setting bacterial species to search (named Replicon ) and the other for inputting srna sequences (named srna sequence ) (Supplementary Fig. 2). In order to set the species to E. coli, type Escherichia, then one will see a long list of E. coli strains. For example, we set to Escherichia coli str. K-12 substr. MG1655 chromosome and paste the sequence of anti-lacz synthetic srna in the below textbox. Next, change two parameters to fit synthetic srnas: change the value 20 in To 20 nts downstream of the mrna translation start site to 24, and change the value 20 in NTs in interaction region to 24, because the target-binding sequence is 24 nt-long. After clicking the SEARCH button below, the predicted target mrnas will be listed. The predicted target mrnas are ranked by their binding free energy. The top 10 target mrnas are shown in Supplementary Figure 2. The actual binding sequences are also shown at the bottom of the results page. As expected, in this example the top-ranking target is lacz with a binding free energy of kcal mol -1. The binding free energy is slightly different from what we got from DINAMelt, kcal mol -1, because the algorithm used in the TargetRNA2 to estimate binding free energy is different from that in DINAMelt and UNAfold. If one desires to convert the binding energy to the one that is calculated by other algorithms, copy and paste the binding sequences at the bottom of the page onto the DINAMelt web-tool as explained in the Experimental design section. Most of the predicted targets, except yfdc (6 th, kcal mol -1 ), yehb (7 th, kcal mol -1 ), and asma (10 th, kcal mol -1 ) were found to bind to the sequence of the scaffold of the synthetic srna. The scaffold was originated from E. coli MicC srna and it is wellknown that this srna specifically binds to the ompc mrna. Therefore, the predicted target mrnas that hybridize to the scaffold regions would be false predictions. Among the top 10 3

4 target mrnas, the target-binding sequence of the anti-lacz synthetic srna is predicted to bind yfdc, yehb, and asma mrnas through partial complementarity, and thus these bindings are likely to occur. However, the binding energies are higher than -15 kcal mol -1, which is the minimum binding free energy that is able to repress translation, though the repression efficiency is quite low (~10%). In our previous study, the binding energy of -30 ~ -40 kcal mol -1 showed very high translation repression of up to >90% 1. Therefore, the binding energies of these three predicted targets are not in the range that is suitable for repression by the anti-lacz srna. As the developed algorithms so far suggest many false targets, it is recommended to run a couple of more algorithms (see table 1 in Supplementary Discussion), if possible, and compare the results. If a gene is consistently predicted as a target, then it is likely to be a real target. For the lacz gene we ran the RNAPredator and obtained a ranked list of predicted target mrnas. Except for the lacz gene itself, there was no predicted mrna that in the top 10 list of both algorithms. Furthermore, the predicted top 10 target mrnas, except for lacz, bind to the scaffold region and thus these bindings would be unlikely to occur. The estimated binding free energy for lacz is quite different from what we obtained from DINAMelt and TargetRNA2, because the RNAPredator utilizes a different algorithm to estimate the binding free energy. In this case, the binding energies should be converted by using the DINAMelt web-tool to get a better insight into the relationship between the binding free energy and repression efficiency. When the off-targets are predicted, it is recommended to increase or decrease the length of target-binding sequence within the range of binding energy of -30 ~ -40 kcal mol -1 to avoid the off-target binding. If this does not work, the target-binding sequence can be chosen at a different position (please refer to Supplementary Fig. 1b for more details) to identify one 4

5 that does not cause off-target binding. As long as the target-binding sequence masks the translation initiation region (TIR), which includes approximately 24 nt in the coding sequence, their repression efficiencies are similar 1. If off-target binding persists, try to target any region within the TIR spanning from the SD sequence. As the correlation between the binding energy and repression efficiency has not been investigated yet for this case, it should be carefully confirmed whether repression occurs as expected. High-throughput library construction and screening. The design of synthetic srnas is simple and clear (ATG and the following 21 bp with the resulting binding free energy range of 20 to 40 kcal mol -1 ). Thus, the primer sequences required for constructing a synthetic srna library covering the entire E. coli genome can be easily designed. Also, their cross-reactivities can be checked (see Supplementary Fig. 3). The PCR conditions are the same for all synthetic srnas because the same template is used and the primer-template annealing by using known srna prediction algorithms is the same (pwas-f and pwas-r in Table 1, Fig. 5b); thus, massive parallel PCR experiments can be easily performed. Once a synthetic srna library is constructed, the only step required for constructing a cell library for various high-throughput experiments is plasmid transformation. Each plasmid harboring a respective synthetic srna gene can be introduced into a strain of interest. For example, if the synthetic srna library covers all E. coli genes, it would result in about 4,000 different plasmids harboring respective synthetic srnas and accordingly 4,000 different transformed cells. If one wants simultaneously to screen for the best target genes and the best platform strain among different strains to achieve higher productivity of the desired chemical, plasmids can be transformed into a variety of strains simultaneously. This metabolic engineering application using synthetic srnas can be performed on a wild-type 5

6 strain from scratch to develop a producer of a desired chemical, but also can be done after minimal metabolic engineering is performed to make cells produce at least some amount of the desired chemical. The latter approach is recommended. Furthermore, this strategy can be applied to an already engineered strain decently producing the desired chemical in order to further enhance its production. The examples of applying synthetic srnas to engineer a wild-type E. coli strain for the production of tyrosine and to engineer an already metabolically engineered E. coli strain for the production of cadaverine are described in our previous report 1. The next step is to culture the engineered strains and compare the production titers to determine which strain and which synthetic srnas allow most efficient production of the desired product. As the number of transformants is large, high-throughput cultivation using multi-well plate systems (e.g., BioScreen C, Oy Growth Curves Ab Ltd) can be very helpful. However, it should be noted that those positive recombinant strains preliminary isolated through high-throughput screening should undergo normal cultivation in flasks and fermenters to make sure that they are indeed good producers under the actual production condition. When using cells harboring the dual regulatory system (arac-p BAD -ci ts2 ), cells in a pre-production phase should be incubated at 25 C and in the presence of 1% (wt vol -1 ) arabinose to prevent the production of introduced synthetic srnas during the pre-production experiment stages. Only when it is desired to produce synthetic srnas, cells are transferred into an appropriate medium that contains no arabinose and cultivated at 37 C. If the assay for the production of a chemical can be performed on plates (e.g., colorimetric assays), it is also possible to transform all the library plasmids into a single tube of competent cells and to spread transformants in plates on which cells can generate signals (e.g. color) in a product concentration-dependent manner. If certain colonies exhibit strong 6

7 signals, then the synthetic srnas in those colonies can be identified by sequencing. The effects of combinatorial synthetic srnas on cellular phenotypic changes or metabolic performance can be examined by cloning them into the same plasmid or into separate, compatible plasmid. We confirmed that the introduction of up to three synthetic srnas did not exert significant burden on host cells 1. However, in our experience, srnas cloned into the same plasmid in close proximity can sometimes undergo homologous recombination. Thus, it is recommended to keep some distance (for example, separated by about 100 bp) between synthetic srnas cloned in the same plasmid to avoid unexpected recombination. After all screening experiments are done, one can actually knock out those target genes if desired. For large-scale industrial fermentation, this will allow strain stability. Also, one can transfer the desired srna knock-down system (e.g. the plasmid) into the chromosome if desired. TABLE. Tools to search for srna s potential binding to the target mrnas. Name URL and note Ref TargetRNA2 o o Based on srna conservation, accessibility of binding sites, and hybridization free energy. o Allows to search whole genomic mrnas 2 starpicker RNAPredator IntaRNA o o Based on two-step model for hybridization between srna and mrna. o o Based on RNAplex, but has additional post-processing steps. o o RNA-RNA interaction search algorithm based on hybridization free energy between two RNA molecules and accessibility of their interaction sites

8 RNAup srnatarget RNAplex o No default genomic search. Users have to paste whole mrna sequences in a FASTA format to search. o o RNA-RNA interaction search algorithm based on hybridization free energy and accessibility of binding sites. o No web-service. o Executable file is included in the Vienna RNA package. o o Based on Naïve Bayes method using RNA secondary structure profile as a feature o o RNA-RNA interaction search algorithm. o Executable file only References: 1. Na, D. et al. Metabolic engineering of Escherichia coli using synthetic small regulatory RNAs. Nat. Biotechnol. 31, (2013). 2. Tjaden, B. TargetRNA: a tool for predicting targets of small RNA action in bacteria. Nucleic Acids Res. 36, W109 W113 (2008). 3. Ying, X., Cao, Y., Wu, J., Liu, Q., Cha, L. & Li, W. starpicker: A Method for Efficient Prediction of Bacterial srna Targets Based on a Two-Step Model for Hybridization. PLoS ONE 6, e22705 (2011). 4. Eggenhofer, F., Tafer, H., Stadler, P. F. & Hofacker, I. L. RNApredator: fast accessibility-based prediction of srna targets. Nucleic Acids Res. 39, W149 W154 (2011). 5. Busch, A., Richter, A. S. & Backofen, R. IntaRNA: efficient prediction of bacterial srna targets incorporating target site accessibility and seed regions. Bioinformatics 24, (2008). 6. Mückstein, U. et al. Translational control by RNA-RNA interaction: improved computation of RNA-RNA binding thermodynamics. In Bioinformatics Research and Development, 13, (Springer Berlin Heidelberg, New York, USA, 2008) 7. Cao, Y., Zhao, Y., Cha, L., Ying, X., Wang, L., Shao, N. & Li, W. srnatarget: a web server for prediction of bacterial srna targets. Bioinformation 3, 364 (2009). 8. Tafer, H. & Hofacker, I. L. RNAplex: a fast tool for RNA-RNA interaction search. Bioinformatics 24, (2008). 8