One of these sequences encodes a toxin.

Do you know which?

The Next vivo Connection

Encrypted Sequences

Novel Genetic Codes

Again, though genetic recoding is a developing technology, the field will only grow and develop as our scientific understanding and computational ability to design RNAs and proteins improves. In the same way that we currently have access to large libraries of promoters, ribosomal binding sites, and protein coding sequences, it is not hard to imagine the construction of a library of tRNAs that can be charged with non-canonical amino acids. Whether this is achieved via flexizymes, mutant pairs, or orthogonal introduction, computational selection of internally consistent sets of tRNAs and charging machinery will make it trivially easy to design a novel genetic code, and cell-free options like the Next vivo system would make deploying a novel genetic code relatively uncomplicated and achievable for individuals that possess basic technical ability.

Despite the benefits of genetic recoding, we should be careful and consider the unintended consequences of this technology. Undermining the faithful reproduction of a protein by a universally conserved genetic codes strikes down a cornerstone used in many fields within biology- particularly within genomics and bioinformatics.

The apparent risk of this technology is that genetic recoding may allow harmful sequences to be “encrypted”, thus masking the information contained within while retaining the ability to faithfully produce the encoded protein.

When the available sample space provided by the genetic code and our understanding of the translational machinery is analyzed, it becomes apparent that recoding allows for a potential to generate numerous genetic codes. A lower-bound estimate of the total set of non-redundant genetic codes can be found ccording to the following formula:

Where n is the number of nucleic acid bases, l is the length of the codon, and a is the number of amino acids that need to be assigned a unique codon.

When this space is calculated with conventional parameters: n=4, l=3, and a=20, we estimate that there are (64 Choose 20)*20! possible combinations. Or put another way, 4.77 x 10^34 entirely novel genetic codes within which to encrypt a harmful DNA sequence.

Our estimation of the full genetic code sample space, including functionally redundant genetic codes is as above. However, it should be noted that this is an estimation of the upper limit of the coding space and may overestimate the true diversity of the given coding space.

Preliminary Testing

The potential for harm as a result of this technology is not to be underestimated. If recoded systems become as prevalent and easy to obtain as we expect them to be, control over where DNA containing funtional toxin sequences are distributed greatly diminishes. Accordingly, we reached out to gene synthesis companies to determine whether or not current bioinformatic technologies can detect radically recoded toxin sequences. For most of our analysis, we selected alpha conotoxin because it is short, lethal, and previously described as a potential threat in the paper Conotoxins: Potential Weapons from the Sea in the Journal of Bioterrorism and Biodefense.

Detecting Encrypted Sequences

A total of five companies from the IGSC (n=5) of the 11 possible agreed to test our sequences. Two control sequences were sent along with the encrypted sequences: unencrypted GFP and unencrypted conotoxin. The remaining 10 sequences consisted of equal numbers of encrypted GFP and conotoxin.

Emails were sent to all current members of the IGSC asking them to screen twelve sequences for us. Of the five companies that were willing to help us, all of them correctly identified the un-encrypted toxic proteins.

However, no organization could correctly identify the encrypted toxins.

While this result is unsettling, it is not unexpected. All genomic science relies heavily on the assumption that there is a known relationship between DNA and protein. Though the technology for large-scale recoding does not presently exist, it is prudent to be prepared instead of ignoring a potentially dangerous problem.

This experiment was repeated using each variation of the BLAST software hosted on the NCBI website. Again, the software could not identify any of the completely recoded sequences. If you are curious about the results, the sequences that we sent are available for you to try as well.

Since the initial testing, we have maintained correspondence with individuals at these companies and are looking forward to working closely with them to ensure that DNA synthesis remains a safe and secure practice. We would also like to thank the companies and the wonderful individuals that we had the chance to interact with for their tremendous assistance in identifying and dealing with this problem before it becomes a pressing security issue. Synthetic biologists need DNA, and DNA synthesis needs new bioinformatic screening tools.

Beating BLAST

However, it is not intuitively obvious what degree of genetic recoding is required to evade detection by BLAST. To test this, we developed a software tool within the CODONxCHANGE suite, written in Python 2.7 to test the integrity of the BLAST platform against sequences that have been partially encrypted via a set number of recoding events. This tool can also be used to prepare genes for implementation in an orthogonal cell-free system for biocontainment purposes.

How Robust is BLAST?

The sequence of an alpha conotoxin was randomly encrypted 29 times for each of the possible 17 recoding events. The sequences were then analyzed via BLAST using a locally curated database containing only the original sequence. The percent identity was averaged with non-hits counting as a 0 value. The percent identity of each sequence and the number of matches returned per group was then plotted as a function of the number of recoding events. The orange trace represents the percent identity of the query to the subject sequence, and the blue trace represents the proprotion of recoded events that BLAST found a match for.

The results of this analysis suggest that the most effective number of switches to BEAT Blast is 4. This level of recording is generally sufficient to bring the similarity of a protein to below 80%, which is the consesus value reported in the HHS guidelines and adhered to by the majority of gene synthesis companies.

After 9 recoding events, a number of sequences become able to evade BLAST entirely.

This suggests that rational recoding to disrupt the generation of high scoring words in BLAST may improve the cryptographic ability of genetic recoding even further. Furthermore, having more than 12 recoding events significantly improves the chances that the encrypted sequence is not detected by BLAST at all. Again, the raw data generated from the experiments are available for public analysis and more results can be found on the GitHub page.

Building Solutions

Summary

Looking for the Source Code?

While most of our software can be found freely availble on our GitHub repository, you will notice that some of the tools do not have their source distributed. For the interim, we have been advised not to publish the source code for the fully functional encryption software.

Until we know the ethical and legal standing surrounding the distribution of the software, it will be available via direct contact only.

Instead, a feature-reduced version of the software is available as a pre-compiled binary to get a taste of how it works, and a tangible idea for just how effective recoding is at hiding the identity of a sequence. If you would like to get access to the source code, you can get in touch with us by following the link and verifying that you are affiliated with an academic institution or other trusted party. We love how safe and accessible Synthetic Biology is, and we are excited to continue developing tools to keep it that way. Thank you for your understanding.

References

[1] Young, T. S. and P. G. Schultz, Beyond the Canonical 20 Amino Acids: Expanding the Genetic Lexicon. Journal of Biological Chemistry, 2010. 285: 11039-11044.

[2] Javahishvili, T., A. Manibusan, S. Srinagesh, D. Lee, S. Ensari, M. Shimazu, and P. G. Schultz, Role of tRNA Orthogonality in an Expanded Genetic Code. ACS Chemical Biology, 2014. 9(4): 874-879.

[3]Chatterjee, A., H. Xiao, and P. G. Schultz, Evolution of multiple, mutually orthogonal prolyl-tRNA synthetase/tRNA pairs for unnatural amino acid mutagenesis in Escherichia coli. Proceedings of the National Academy of Sciences of the United States of America, 2012. 109(37): 14841-14846.

[4] Ohuchi, M., H. Murakami, and H. Suga, The flexizyme system: a highly flexible tRNA aminoacylation tool for the translation apparatus. Current Opinion in Chemical Biology, 2007. 11(5): 537-542.