Maximizing Therapeutic Efficacy Through Codon Substitution Design

Synonymous gene recoding is a common technique used in gene therapeutic biomanufacturing to improve the production process or efficacy of a targeted therapeutic.  The most common example of this approach is synonymous codon substitution, where a three base-pair codon sequence is swapped for another one encoding the same amino acid, resulting in an identical protein sequence.  For many gene therapeutics, codon substitution is used to optimize the translation rate of the sequence thus increasing the number of successfully translated protein molecules.  Although synonymous codon substitutions do not change the amino acid sequence of the gene therapeutic it can still lead to undesirable and dangerous side-effects.  In this blog we discuss a recent publication by Lin et al, on the implications of the negative impacts and the role of artificial intelligence (AI) to overcome these challenges¹. ‍

Negative Effects of Synonymous Codon Substitution

Although synonymous codon substitutions do not change the amino acid sequence of the gene therapeutic it can still lead to undesirable and dangerous side-effects.  Some potential issues that may arise from synonymous codon substitutions are the following:

1. Changes in how messenger RNA transcripts are spliced, driving changes to the final amino acid sequence and subsequent protein expression. 
2. Changes in the structure and stability of the messenger RNA molecule, leading to faster degradation of the molecule or changes to the rate of translation that negatively impact the quantity of produced protein.
3. Impacts on microRNA binding capacity driving a change in the gene regulatory network of the cell.
4. Alternative folding of the final protein product, as well as changes in post-translational modifications of the protein, leading to deleterious changes in protein function.

These changes can reduce or eliminate the expression of the therapeutic protein, or alter its functionality and immunogenicity risk through changes to the protein structure.

Leveraging Computational Methods to Optimize Synonymous Codon Substitution and Therapy Production

In a recent opinion article by Lin et al., the authors advocate for use of emerging deep learning methods to help mitigate the negative side effects of synonymous gene recoding in therapeutic development¹.  There are a number of published neural network models (some of which are enumerated below) to predict the impact of sequence changes on messenger RNA splicing and structure, as well as micro-RNA binding affinity.  

MMSplice² (splicing)

SPOT-RNA³ (mRNA structure)

DMfold⁴ (mRNA structure)

CDPfold⁵ (mRNA structure)

miRDB⁶ (miRNA binding)

DeepMirTar⁷ (miRNA binding)

DeepTarget⁸ (miRNA binding)

PredDSM⁹  (pathogenic variants)

usDSM¹⁰ (pathogenic variants)

Using these tools will predict possible deleterious effects of synonymous codon substitutions allowing researchers to screen out unsafe or ineffective codon substitutions.  Although we are further away from having a reliable method to predict changes to protein structure due to synonymous changes to its encoding DNA sequence, big recent advancements in the field of protein folding (eg AlphaFold) indicate that there could be promising developments in the near future¹¹.  

In the gene therapy space, codon optimization has been tested as a strategy for improving the efficacy of lentiviral vector (LV) gene therapies, yet its application to RD114-TR glycoproteins revealed unexpected roadblocks. Despite increasing expression levels through codon optimization, subsequent modifications such as glycosylation and surface transport were reduced leading to inactive LVs with dramatically reduced potency¹².  Moreover, codon modification in AAV gene therapy vectors has shown to result in increased CpG content¹³. Unmethylated CpGs are known to elicit an immune response, which would be desirable in the context of vaccine design, therefore, for the purpose of codon optimization for gene therapy, the reduction of CpG sites is necessary to reduce the immunogenicity of AAV gene therapy¹³.

To build an optimal strategy in predicting functional synonymous variants, unique features that can identify one algorithm's superiority over another will be key to this process in addition to wet-lab experimental resources.  By combining both sets of approaches together, we can create a more comprehensive framework for gaining reliable results in gene therapy development now and into the future.

AI Disclosure: Feature image was generated by an AI image tool.