An engineered T7 RNA polymerase that produces mRNA free of immunostimulatory byproducts

Abstract

In vitro transcription (IVT) is a DNA-templated process for synthesizing long RNA transcripts, including messenger RNA (mRNA). For many research and commercial applications, IVT of mRNA is typically performed using bacteriophage T7 RNA polymerase (T7 RNAP) owing to its ability to produce full-length RNA transcripts with high fidelity; however, T7 RNAP can also produce immunostimulatory byproducts such as double-stranded RNA that can affect protein expression. Such byproducts require complex purification processes, using methods such as reversed-phase high-performance liquid chromatography, to yield safe and effective mRNA-based medicines. To minimize the need for downstream purification processes, we rationally and computationally engineered a double mutant of T7 RNAP that produces substantially less immunostimulatory RNA during IVT compared with wild-type T7 RNAP. The resulting mutant allows for a simplified production process with similar mRNA potency, lower immunostimulatory content and quicker manufacturing time compared with wild-type T7 RNAP. Herein, we describe the computational design and development of this improved T7 RNAP variant.

Fig. 1. T7 RNAP transcription cycle.

a, T7 RNAP takes on two distinct conformational states during catalysis, the IC and EC. b, In the IC, T7 RNAP binds the T7 promoter and generates short RNAs or abortive transcripts (2–10 nucleotides) until it transitions into the EC; this transition may require several initiations before it is successful. The short RNAs produced can interact with T7 RNAP via RNA-templated transcription, producing short dsRNAs. Only after transitioning into the EC is T7 RNAP highly processive and capable of generating full-length RNA. As an enzyme, T7 RNAP can catalyze the formation of hundreds of copies of full-length RNA. These full-length molecules can also be used as templates by T7 RNAP to generate long loopback dsRNA species. Both types of dsRNA impurities are immunostimulatory in vitro and in vivo.

Fig. 2. C-terminal ‘foot’ engineering to decrease dsRNA impurities.

a, Cutaway of T7 RNAP indicating the buried C-terminal ‘foot’ (FAFA⁸⁸³; purple) and associated void volume (red). The CTD (grey surface), bound DNA template (orange) and targeted helices (purple and green cylinders) in the NTD (ribbon) are shown for reference. b, Scatter plot of change in void volume (∆V_void; as a percent of total molecular volume) compared with the change in ∆∆E_mut due to C-terminal substitutions relative to 884A. The line indicates the best-fit linear regression (plotted with Seaborn.lmplot), with a Pearson correlation of −79% (calculated with the scipy.stats Python library); data are presented as mean ± s.d.; n = 25 for ∆∆E_mut; error bars for ∆V_void are excluded for clarity, n = 120. c, RNA yield as a function of C-terminal foot substitution. The x axis represents the identity of the added amino acid in single-letter code. Data indicate that yield generally decreases as a function of amino acid size; n = 2. d, Scatter plot of ∆∆E_mut versus RNA yield relative to WT. The line indicates the best-fit linear regression, with a Pearson correlation of −76%; data are presented as mean for ∆∆E_mut, n = 25; n = 1 for RNA yield. e, Scatter plot of 3′ homogeneity compared with total RNA yield of selected mutants; n = 1. WT T7 RNAP is represented by the dotted lines and blue circle. Here, 884G (shown as G) was selected owing to improved 3′ homogeneity while maintaining RNA yield. The 3′ homogeneity generally improved with increasing amino acid size. f, Scatter plot of ∆∆E_mut versus RNA 3′ homogeneity. The line indicates the best-fit linear regression, with a Pearson correlation of 29%; data are presented as mean for ∆∆E_mut, n = 25; n = 1 for RNA 3′ homogeneity. Source data

Fig. 3. NTD engineered mutants for decreased dsRNA impurities.

a, Conformational change in the C-helix (green) and C-linker (purple) from the IC (PDB: 2PI4; left) to the EC (PBD: 1H38; right) for T7 RNAP. b, Computational prediction of EC-favoring substitutions using Rosetta. Any point below the dotted blue line represents a substitution for which the calculated ∆∆E_mut (reported as Rosetta Energy Units and distinguishable from ∆∆G_mut) was more favorable for the EC than the IC. ∆∆E_mut values for WT and all 19 possible substitutions at all 878 modeled positions within the T7 RNAP IC and EC crystal structures are plotted as grey dots (20 × 878 = 17560 total points). The C-helix and C-linker substitutions are shown as red dots (20 × 14 = 280 points), with blue dots representing 20 substitutions selected for experimental characterization. c, Scatter plot of 𝚪 (ΔΔE_mut,EC − ΔΔE_mut,IC) compared with RNA yield (percentage relative to WT) for NTD substitutions; the blue line represents the line of best fit; n = 1. d, Scatter plot of 𝚪 versus 3′ homogeneity for the NTD substitutions; Pearson correlation of 80%; n = 1. e, Scatter plot of RNA 3′ homogeneity compared with RNA yield for the NTD substitutions. f, Scatter plot of IFN-β response compared with RNA yield for the NTD mutants; n = 2. g, Scatter plot of IFN-β response compared with RNA 3′ homogeneity for the NTD mutants; n = 1. In e–f, black dotted lines and blue circle indicate WT T7 RNAP. Red dots represent all mutants; mutants with desired outcomes as compared to WT T7 RNAP are labeled with their identity. Source data

Fig. 4. Comparison of IVT impurities between WT and G47A + 884G T7 RNAPs.

a, Radioactive sequencing gels monitoring the formation of short RNA species and dsRNA by WT T7 RNAP and G47A + 884G T7 RNAP. Reactions using α-³²P-GTP selectively label abortive species, full-length RNA, run-on transcripts and large RNA species generated from RNA-templated and loopback transcription (blue panels). By contrast, reactions incorporating α-³²P-CTP label short and long dsRNA species in addition to full-length and run-on transcripts (red panels); n = 3. Full images of the radioactive gels and images from trials 2 and 3 are shown in Supplementary Fig. 3. b–d, 3′-homogeneity (n = 8 mRNA lots) (b), dsRNA levels (n = 6 mRNA lots) (c) and IFN-β responses (n = 4 mRNA lots) (d) between WT (black) and G47A + 884G (red) across eight different mRNAs of varying length and sequence composition. Data are presented as mean ± s.d. Statistical significance was determined using two-sided Brown–Forsythe analysis of variance followed by Dunnett’s multiple comparison test for hypothesis testing. Family-wise correction was used for multiple testing correction with alpha level of 0.05. nt, nucleotides. Source data

Fig. 5. In vitro and in vivo performance of mRNAs synthesized by WT and G47A + 884G T7 RNAPs.

a, Quantification of dsRNA using ELISA indicated that the RNAs produced by G47A + 884G had lower dsRNA content than WT T7 RNAP, whether purified by RP chromatography or not (all mRNAs were purified by oligo-dT affinity chromatography); n = 1 mRNA, n = 2 technical replicates. b, IFN-β response in BJ fibroblasts indicates that RNAs produced by G47A + 884G with (+RP) or without RP chromatography have baseline response, similar to the Lipofectamine control; n = 1 mRNA, n = 2 technical replicates. c, IP-10 response measured in MDMs confirmed that G47A + 884G generates immune-silent mRNA; n = 1 mRNA, n = 2 technical replicates. d, IP-10 response was measured in serum 6 h after mice were dosed intravenously with mRNA generated with WT T7 RNAP or G47A + 884G with or without RP chromatography. IP-10 response in vivo confirmed in vitro observations that G47A + 884G generates mRNA that does not stimulate a detectible innate immune response in vivo. Data are presented as mean ± s.d.; n = 4 mice for WT T7 RNAP and mutant, n = 3 mice for PBS and naive. Statistical significance was determined using Kruskal–Wallis testing followed by Dunn’s multiple comparison test for hypothesis testing. Family-wise correction was used for multiple testing correction with an alpha level of 0.05. e, hEPO expression was measured in serum 6 h after mice were dosed intravenously with mRNA synthesized with WT or G47A + 884G with or without RP chromatography. All mRNAs were expressed equally. Data are presented as mean ± s.d; n = 4 mice for WT T7 RNAP and mutant, n = 3 mice for PBS and naive. Source data