Advanced analytics and artificial intelligence (AI) will help make mRNA vaccines and therapeutics better by allowing biopharma engineers to fine-tune the enzymes used to manufacture such products.
Researchers at Australia’s Adelaide University put forward the case for process improvement in a new study, citing the in vitro transcription (IVT) step used to convert DNA templates into mRNA molecules as an example of an area in need of work.
Lukas Gerstweiler, PhD, told GEN, “It is well known that T7 RNA polymerase–based in vitro transcription is particularly prone to impurity formation.
“The enzyme can generate aberrant initiation and termination products, engage in template-independent transcription, or produce antisense RNA, all of which promote dsRNA formation and the generation of truncated fragments.”
And concerns about dsRNA—double-stranded RNA molecules—are already driving innovation in the analytics space, according to Gerstweiler.
“It has been demonstrated repeatedly in the literature that impurities can strongly affect the translation efficiency of mRNA, yet this is still not always monitored. Many impurities, such as dsRNA, are very similar to the product mRNA and are therefore hard to detect and quantify accurately.
“Legacy methods like dot blot are widely used but lack the precision of more advanced HPLC-based approaches. We believe that quality assessments of mRNA are often neglected. By integrating data-driven models and process control, these analytics enable rational optimization, predictive quality monitoring, and robust scale-up of mRNA manufacturing.”
Optimization
Other impurity reduction methods focus on optimizing transcription conditions by, for example, tweaking the ratio of Mg²⁺ ions to nucleoside triphosphate (NTP), reaction temperature, or altering template designs.
However, while such approaches can reduce the level of impurities, they are not fully effective. Moreover, they are process-specific, meaning manufacturers need to re-optimize for each new product, Gerstweiler says.
“We found that optimal IVT conditions can differ from template to template, while the percentage of dsRNA formed is also sequence-dependent. Manufacturers should consider reoptimizing conditions for each construct,” he says.
More recently, the focus has shifted, and researchers have started to use protein engineering to optimize the T7 RNA polymerase enzyme itself.
“There have been promising developments in engineered T7 polymerases that have been shown to lower dsRNA formation. They are also particularly promising, as modified T7 polymerases have been developed that can incorporate a broad range of modified nucleotides, which opens completely new avenues for mRNA modifications and stability enhancements,” Gerstweiler adds.
High-fidelity enzymes
Looking forward, AI is likely to play a significant role in these kinds of enzyme engineering and optimization processes, according to Gerstweiler.
“IVT using T7 RNA polymerase is a highly variable process with an immense number of degrees of freedom, far too many variables to explore exhaustively through traditional experimentation alone.
“Machine learning and AI approaches excel in this context by elucidating hidden patterns and correlations with substantially reduced experimental effort. This enables data-driven optimization of IVT processes, leading to higher yields, lower impurity formation, improved purity, and better overall process robustness,” Gerstweiler says.
He cited emerging techniques like ML–directed evolution, protein language models, and structure-based redesign as approaches that could produce T7 RNA pol variants with improved fidelity, higher thermostability, and better performance in mRNA synthesis.
Gerstweiler adds, “We believe the transformative impact of these techniques on developing more efficient, high-fidelity enzymes will be profound and difficult to overstate.”
