Swapping uridine for N1-methylpseudouridine (m1Ψ) in synthetic mRNA — silencing RIG-I/TLR/PKR innate sensing while boosting ribosome throughput
Every therapeutic or vaccine mRNA begins as an in vitro transcription (IVT) reaction: a linearized DNA template (plasmid or PCR amplicon) bearing a T7 (or SP6) promoter is transcribed by recombinant bacteriophage RNA polymerase in the presence of the four ribonucleoside triphosphates. To produce a nucleoside-modified product, UTP is simply omitted from the reaction and replaced 1:1 by N1-methylpseudouridine-5′-triphosphate (m1Ψ-TP) — every position the genetic code specifies as U is instead transcribed as m1Ψ, with no change to the encoded amino acid sequence.
N1-methylpseudouridine differs from uridine at two connected positions. First, pseudouridine (Ψ) itself is a C5-glycoside isomer of uridine — the ribose is attached to C5 of the uracil ring via a carbon-carbon bond instead of the N1-nitrogen bond used by every other canonical nucleoside. This C-C linkage frees the N1 position, which in Ψ becomes an additional hydrogen-bond donor (N1-H), subtly stabilizing base stacking and Watson-Crick pairing with adenine. Second, N1-methylpseudouridine caps that newly exposed N1 with a methyl group, removing the extra H-bond donor again but leaving the structurally important C-glycosidic bond intact. The combination — C-C backbone linkage plus N1-methylation — is what distinguishes m1Ψ pharmacologically from both unmodified U and plain Ψ.
IVT reaction composition (typical 1 mL scale): • Linearized DNA template: 0.5–1 µg/µL, sequence-verified, EcoRI/BspQI or PCR-linearized downstream of poly(A) • T7 RNA polymerase: 4,000–8,000 U/mL, often with pyrophosphatase (0.1 U/mL) to prevent Mg-pyrophosphate precipitation • NTPs: ATP, GTP, CTP at 5–7.5 mM each; m1Ψ-TP substituted 1:1 for UTP at 5–7.5 mM (TriLink BioTechnologies and Hongene Biotech are the dominant commercial suppliers of m1Ψ-TP) • Reaction buffer: 40 mM Tris-HCl pH 7.9, 25 mM MgCl₂, 2 mM spermidine, 10 mM DTT • Co-transcriptional capping reagent (CleanCap AG, TriLink) at 4 mM, or post-transcriptional enzymatic capping (vaccinia capping enzyme D1/D2 + 2′-O-methyltransferase) to install Cap1 • Incubation: 37°C, 2–4 hours; DNase I treatment (30 min, 37°C) removes template; RNA recovered by LiCl precipitation, RP-HPLC, or magnetic-bead capture
T7 polymerase was evolved on natural nucleotides and shows measurably reduced catalytic efficiency (kcat/Km) when incorporating m1Ψ-TP opposite template adenines — typically a 15–20% drop in overall transcript yield and a mild increase in premature termination compared to an all-natural NTP mix. This is a well-characterized, manageable cost: yields of 3–5 mg RNA per mL of reaction are routinely achieved at GMP scale (Pfizer/BioNTech and Moderna both run m1Ψ IVT at multi-liter bioreactor scale for commercial vaccine manufacture), and downstream purification steps recover the vast majority of full-length, correctly capped product.
Unmodified exogenous RNA is a five-alarm signal to the innate immune system. Cytosolic RIG-I and MDA5 helicases, endosomal TLR3/7/8, and the interferon-inducible kinase PKR all evolved to detect foreign RNA and trigger a coordinated antiviral response — type I interferon secretion, translational shutdown, and RNA degradation. Karikó and Weissman’s foundational 2005 Immunity paper first showed that naturally occurring modified nucleosides, including pseudouridine, m5C, and m6A, are dramatically less immunostimulatory than unmodified bases when substituted into synthetic transcripts, and that m1Ψ in particular combines the lowest immunogenicity with the best translational performance of any modification tested.
Each sensor recognizes a distinct structural feature of RNA, and m1Ψ interferes with several simultaneously:
RIG-I (retinoic acid-inducible gene I): binds 5′-triphosphate or diphosphate blunt-ended dsRNA and short uridine/AU-rich single-stranded regions via its C-terminal domain (CTD) and helicase core, then oligomerizes on the RNA to trigger MAVS-dependent IRF3/NF-κB signaling. The CTD makes direct contacts with the uracil base-pairing edge; the altered stacking geometry and shifted hydrogen-bonding pattern of m1Ψ reduce CTD binding affinity and destabilize RIG-I filament formation along the RNA, cutting downstream IFN-β promoter activation by roughly 80–95% in reporter systems relative to unmodified transcripts of identical sequence.
TLR7/8: endosomal receptors on plasmacytoid dendritic cells and monocytes that recognize single-stranded, U-rich or GU-rich RNA delivered via endocytosis (the natural route for LNP-formulated mRNA). Guanosine- and uridine-rich motifs are the dominant agonist signature; substituting U for m1Ψ removes the canonical uridine recognition determinant almost entirely. HEK-Blue hTLR7 and hTLR8 reporter cell lines show roughly an order-of-magnitude reduction in NF-κB-driven SEAP output for m1Ψ transcripts compared to unmodified controls at matched RNA dose.
PKR (protein kinase R): a cytosolic dsRNA sensor that dimerizes on duplex RNA (>30 bp) and autophosphorylates, subsequently phosphorylating eIF2α to shut down cap-dependent translation globally — a self-defense mechanism that, left unchecked, would abolish the therapeutic protein output the mRNA drug is designed to produce. m1Ψ reduces PKR binding affinity for duplex regions and lowers autophosphorylation efficiency in vitro, partially uncoupling PKR activation from bulk mRNA dose.
2′-5′-OAS/RNase L: oligoadenylate synthetase (OAS1/2/3) is activated by dsRNA and synthesizes 2′-5′-linked oligoadenylates that allosterically activate RNase L, a broad-spectrum endoribonuclease that degrades both viral and host mRNA indiscriminately. Because OAS activation depends on duplex RNA content rather than the modified base directly, its suppression is achieved less by m1Ψ itself and more by the downstream dsRNA-depletion purification step (Stage 4) — the two effects are complementary rather than redundant.
In the pivotal Karikó, Muramatsu, Weissman et al. 2005 Immunity study, luciferase mRNA transcribed with Ψ or m5C induced roughly 10-fold less TNF-α and IL-12 from human dendritic cells than the unmodified control at equal RNA mass, while a parallel comparison later established that m1Ψ specifically also preserved — and in most cell types improved — protein output relative to Ψ, making it the modification ultimately selected for both authorized COVID-19 mRNA vaccines.
Reduced immune activation is only half the therapeutic rationale for m1Ψ. Because PKR autophosphorylation and eIF2α phosphorylation are blunted, cap-dependent translation initiation is not shut down mid-course the way it is for unmodified or under-purified mRNA. The net effect, measured directly by polysome profiling and reporter-protein assays, is substantially higher cumulative protein output per microgram of delivered mRNA — the property that made m1Ψ, rather than plain pseudouridine, the modification of choice for clinical mRNA products.
Three converging mechanisms explain the translational advantage of m1Ψ-modified mRNA:
1) Preserved eIF2α-dependent initiation: Unmodified or heavily dsRNA-contaminated mRNA activates PKR, which phosphorylates eIF2α at Ser51. Phospho-eIF2α sequesters the guanine nucleotide exchange factor eIF2B, starving the cell of ternary complex (eIF2-GTP-Met-tRNAi) and halting new rounds of translation initiation genome-wide, including on the therapeutic transcript itself. Because m1Ψ transcripts trigger far less PKR activation, initiation continues largely unimpeded, and polysome profiling on sucrose gradients shows visibly higher ribosome occupancy (more ribosomes bound per transcript, shifted toward heavier polysome fractions) for m1Ψ mRNA compared to unmodified controls at matched input dose.
2) Reduced RNase L-mediated decay: OAS/RNase L activation, driven primarily by residual dsRNA rather than the base modification per se, degrades mRNA nonspecifically once triggered. Because purified m1Ψ mRNA formulations carry markedly less dsRNA (see Stage 4) and generate a smaller interferon response overall (which itself upregulates OAS gene expression in a feed-forward loop), functional intracellular half-life is extended roughly 2–3-fold relative to unmodified or crudely purified transcripts, giving each mRNA copy more translation cycles before degradation.
3) Altered ribosome kinetics at the codon level: Biophysical and ribosome-profiling studies (Svitkin et al. 2017, Nucleic Acids Research; Eyler et al. 2019, PNAS) additionally report that m1Ψ can modestly affect ribosome dwell time and reading-frame fidelity at modified codons, generally without triggering the -1 ribosomal frameshifting or premature termination sometimes seen with plain Ψ at high substitution density — part of why m1Ψ outperforms Ψ as a translation-optimized modification despite both offering comparable immune evasion.
In combination, these effects compound: a reporter mRNA (e.g., firefly luciferase) encoding an identical protein sequence but synthesized with m1Ψ instead of UTP typically produces 8–16-fold more cumulative luminescence over a 48-hour cell-culture time course than the unmodified transcript, and the effect is dose-dependent on m1Ψ substitution fraction — partial substitution gives intermediate immune suppression and intermediate translational benefit, which is why clinical products use essentially complete (~100%) UTP replacement rather than partial modification.
Nucleoside modification alone does not fully solve the immunogenicity problem, because T7 IVT itself generates double-stranded RNA byproducts independent of base chemistry. RNA-dependent RNA polymerase activity intrinsic to T7 pol produces 3′-extended, self-complementary transcripts and RNA:DNA hybrid artifacts that fold into duplex structures potent enough to activate RIG-I, PKR, and OAS regardless of whether the single-stranded portion is m1Ψ-modified. Removing this contaminant is therefore a mandatory, independent purification step for any clinical-grade modified mRNA.
dsRNA contamination in IVT reactions arises from at least three characterized mechanisms: (i) T7 RNA polymerase RNA-dependent RNA polymerase (RdRp) activity, in which the enzyme uses a completed transcript as a template to synthesize a complementary strand, producing long hairpin or duplex species; (ii) 3′ self-priming, where the nascent transcript’s 3′ end folds back and primes template-independent extension; and (iii) abortive transcripts and RNA:DNA hybrids left over from incomplete template clearance. Depending on template sequence, reaction conditions, and polymerase lot, dsRNA can constitute anywhere from 3% to as much as 18% of total IVT RNA mass — more than enough to dominate the immunostimulatory profile of an otherwise well-modified product.
Two purification strategies dominate current practice:
Reverse-phase HPLC (Karikó/Weissman, Nucleic Acids Research 2011): Full-length ssRNA and dsRNA/aberrant species are separated on a C18 or similar hydrophobic stationary phase under denaturing conditions (elevated temperature, ion-pairing mobile phase with triethylammonium acetate), exploiting differential retention driven by RNA structure and length. This method reliably reduces dsRNA content to below 0.1% (often below the limit of detection by dsRNA-specific ELISA, e.g. the J2 monoclonal antibody dot blot) but has moderate throughput and RNA recovery losses that must be managed at manufacturing scale.
Cellulose batch chromatography (Baiersdorfer et al., Molecular Therapy Nucleic Acids 2019): dsRNA binds cellulose fibers under specific ethanol/salt conditions while ssRNA flows through, exploiting a simpler physicochemical partition rather than reversed-phase hydrophobic interaction. This method achieves comparable dsRNA clearance with somewhat better scalability and lower cost for GMP manufacturing, and has been widely adopted as a complementary or alternative step to HPLC in commercial mRNA vaccine production.
Both methods are typically paired with quality-control assays — dsRNA-specific ELISA (J2/K1 antibodies), dot-blot, or newer LC-based methods — with commercial release specifications commonly requiring dsRNA content below 1–3 ng per µg of total RNA (roughly <0.1–0.3% by mass) before a lot is deemed suitable for clinical use. Because dsRNA-driven signaling operates through OAS/RNase L and PKR largely independent of the U-versus-m1Ψ substitution at the single-stranded positions, purification and nucleoside modification act as additive, non-redundant layers of immune-stealth engineering — one addressing base-level recognition, the other addressing bulk structural contamination.
The clinical payoff of nucleoside modification and dsRNA depletion is measured in vivo: a lipid-nanoparticle-formulated, m1Ψ-substituted, dsRNA-depleted mRNA drug delivers robust, transient antigen or therapeutic protein expression with substantially lower local and systemic reactogenicity than earlier-generation unmodified mRNA formulations — the combination that made mRNA vaccines a viable platform for global deployment during the COVID-19 pandemic and that now underlies a broad pipeline of mRNA therapeutics beyond vaccines.
After intramuscular injection, LNP-encapsulated m1Ψ mRNA is taken up predominantly by myocytes and resident/infiltrating antigen-presenting cells at and near the injection site, with modest lymphatic drainage-mediated distribution to draining lymph nodes. Endosomal escape releases the mRNA into the cytosol, where it is translated by host ribosomes essentially as if it were an endogenous transcript. Protein expression typically peaks between 24 and 48 hours post-injection and remains detectable, at gradually diminishing levels, for approximately 10–14 days before the transcript is fully cleared by normal cellular RNA turnover — a pharmacokinetic profile intentionally distinct from DNA-based or viral-vector platforms, which can produce more sustained but less tunable expression.
The reactogenicity advantage of m1Ψ is not merely a cell-culture artifact: clinical and preclinical comparisons consistently show that unmodified mRNA-LNP formulations provoke substantially stronger local inflammation (injection-site swelling, myalgia) and systemic cytokine responses (transient fever, IFN-driven flu-like symptoms) than matched-dose m1Ψ formulations, both because the base modification suppresses the RIG-I/TLR7/PKR axis directly and because the ionizable-lipid LNP carrier itself contributes an independent, modification-independent innate stimulus that becomes the dominant remaining reactogenicity driver once the RNA cargo is optimized. This is why current-generation mRNA-LNP products are formulated with both an optimized ionizable lipid (e.g., ALC-0315 in BNT162b2, SM-102 in mRNA-1273) and fully m1Ψ-substituted, dsRNA-depleted RNA cargo — the two engineering axes are complementary, not substitutable.
The platform’s clinical validation is now extensive: BNT162b2 (Pfizer-BioNTech, m1Ψ-modified) and mRNA-1273 (Moderna, m1Ψ-modified) together account for well over a billion administered doses worldwide, generating the largest real-world safety and immunogenicity dataset for any nucleoside-modified mRNA product to date and establishing m1Ψ substitution as the default starting point for new mRNA vaccine and therapeutic programs — including in-development mRNA-encoded monoclonal antibodies, protein-replacement therapies (e.g., propionic acidemia, methylmalonic acidemia programs), and in vivo CAR-T/gene-editing payload delivery, all of which inherit the same stealth-and-yield rationale established first in infectious-disease vaccines.
A head-to-head comparison reported by Pardi and colleagues (multiple studies, 2015–2018) found that a nucleoside-modified, HPLC-purified mRNA-LNP influenza hemagglutinin vaccine elicited protective, broadly neutralizing antibody titers in mice and nonhuman primates at doses roughly 10-fold lower than unmodified mRNA-LNP required for comparable protection — directly translating the combined immune-evasion and translational-yield advantages of m1Ψ plus dsRNA depletion into a lower effective clinical dose, a result that anchored the dose-selection strategy later used in COVID-19 mRNA vaccine development.