Combinatorial optimization of mRNA structure, stability, and translation for RNA-based therapeutics

Independent of the nucleotide identity 5' of the U, the bond leading to 3' U is a hotspot for in-line nucleophilic attack.

“Thus, independent of the nucleotide identity 5’ of the U, this bond is a hotspot for in-line nucleophilic attack63,65.”

• cites [63] Kaukinen, U., Lyytikäinen, S., Mikkola, S. & Lönnberg, H. The reactivity of phosphodiester bonds within linear single-stranded oligoribonucleotides is strongly dependent on the base sequence. Nucleic Acids Res. 30, 468–474 (2002). sourceDisputed

  The Results (page unavailable) report that cleavage of the studied 5′-UpA-3′ bond varies strongly with the surrounding sequence; in one sequence, no reaction was detected in three months. The Discussion (page unavailable) explicitly rejects the idea that this bond is inherently prone to cleavage. The paper studies a bond following U, rather than establishing that a bond leading to 3′ U is a hotspot independent of the preceding nucleotide. (LLM, conf=0.82)

  Checked by a reasoning model — may be wrong.

  Experimental — accuracy not yet measured
• cites [65] Seetin, M. G., Kladwang, W., Bida, J. P. & Das, R. Massively parallel RNA chemical mapping with a reduced bias MAP-seq protocol. Methods Mol. Biol. 1086, 95–117 (2014). sourceCould not verify

  The Introduction (page unavailable) discusses chemical probing and nucleotide reactivity, and the References list a work on in-line probing. Neither passage addresses whether the bond leading to a 3′ U is a hotspot for in-line nucleophilic attack regardless of the 5′ nucleotide; searches for that mechanism and sequence preference found no supporting passage in our copy. (LLM, conf=0.73)

  Experimental — accuracy not yet measured

Highly structured 5' UTRs, such as dengue virus (DEN2), can support efficient translation, contrary to previous reports.

“Most surprising, in contrast to previous reports47–49, we found 5’ UTRs that are highly structured, such as the dengue virus (DEN2), can support efficient translation.”

• cites [47] Zhang, H. et al. LinearDesign: efficient algorithms for optimized mRNA sequence design. Preprint at https://arxiv.org/abs/2004.10177 (2020). sourceNot checked yet

  Checked before — its earlier verdict is held back until a new check runs.

• cites [48] Ding, Y. et al. In vivo genome-wide profiling of RNA secondary structure reveals novel regulatory features. Nature 505, 696–700 (2014). sourceNot checked — no full text of the source
• cites [49] Shah, P., Ding, Y., Niemczyk, M., Kudla, G. & Plotkin, J. B. Rate-limiting steps in yeast protein translation. Cell 153, 1589–1601 (2013). sourceSupported

PERSIST-seq data across different CDS types confirmed the effect of CDS length on RNA stability.

“The strongest prediction of previous theoretical modeling7 was that length changes should drive the most variation in in-solution stability, and PERSIST-seq data across different different CDS types confirmed the effect of CDS length on RNA stability, (Fig. 2G).”

• cites [7] Wayment-Steele, H. K. et al. Theoretical basis for stabilizing messenger RNA through secondary structure design. Nucleic Acids Res. 49, 10604–10617 (2021). sourceSupported

• AssumptionThe authors state this as an assumption: The high pH buffer containing Mg2+ mimics the high effective pH and positively charged environment that can arise in lipid nanoparticles, protamine, and other formulations for mRNA therapeutics.

Nluc protein has negligible degradation as measured by luciferase activity in transiently Nluc-expressing HEK293 cells for at least 6 hours after cycloheximide treatment.

“kp is set to 0 since Nluc protein has negligible degradation as measured by luciferase activity in transiently Nluc-expressing HEK293 cells for at least 6 hours after cycloheximide treatment, which allows assessment of protein degradation in the absence of further translation91.”

• cites [91] Martin, M. Cutadapt removes adapter sequences from high-throughput sequencing reads. EMBnet J. 17, 10 (2011). sourceCould not verify

  The Introduction and Algorithm sections describe adapter trimming in sequencing reads, not protein degradation experiments. Searches for Nluc, HEK293, luciferase activity, and cycloheximide found no passage addressing the claimed six-hour stability. (LLM, conf=0.78)

  Experimental — accuracy not yet measured

The effect of high translation efficiency on stability may derive from overcrowding of ribosomes on mRNA coding regions due to rapid initiation, which can induce sterical ribosome collisions that lead to translation-dependent mRNA decay.

“This effect may derive from overcrowding of ribosomes on mRNA coding regions due to rapid initiation, which can induce sterical ribosome collisions that have recently been found to lead to translation-dependent mRNA decay78,79.”

• cites [78] Yoffe, A. M. et al. Predicting the sizes of large RNA molecules. Proc. Natl Acad. Sci. USA 105, 16153–16158 (2008). sourceNot checked — no full text of the source
• cites [79] Polack, F. P. et al. Safety and efficacy of the BNT162b2 mRNA Covid-19 vaccine. N. Engl. J. Med. 383, 2603–2615 (2020). sourceCould not verify

  The Methods and Discussion describe BNT162b2 vaccine efficacy and safety, not how translation efficiency affects mRNA stability. Searches for ribosome collisions, rapid initiation, and translation-dependent mRNA decay found no passage addressing the claimed mechanism. (LLM, conf=0.76)

  Experimental — accuracy not yet measured

112 of the mRNAs in the library contained varied 5' and/or 3' UTRs.

“112 of these mRNAs contained varied 5’ and/or 3’ UTRs (Fig. 1A).”

Chimeric fusion of the hHBB 5' UTR with TEV or 5' TOP sequence increased polysome loading.

“Moreover, the chimeric fusion of the hHBB 5’ UTR with elements such as the TEV or 5’ TOP sequence in the same 5’ UTR increased polysome loading.”

The mRNA designs with highest ribosome load were observed across 5' UTRs of cellular as well as viral origins.

“Overall, the mRNA designs with highest ribosome load were observed across 5’ UTRs of cellular as well as viral origins (Fig. 1E, Table S1).”

One-by-one tests of the 5' UTRs selected to have high ribosome load gave lower total protein output than the starting sequence.

“Interestingly, one-by-one tests of the 5’ UTRs selected to have high ribosome load gave lower total protein output than the starting sequence (see Supplemental Text); this observation foreshadowed a tradeoff between ribosome load and mRNA stability that we dissect in greater detail below.”

The integrated model accurately predicts protein output for a given mRNA based on its ribosome load and in-cell stability.

“We integrate our datasets to develop a model that accurately predicts protein output for a given mRNA based on its ribosome load and in-cell stability.”

• AssumptionThe authors state this as an assumption: Translation follows first-order kinetics and mRNA and protein decay are exponential.

Specific 5' UTRs (mouse COL1A2, Hoxa9 P4, Rpl18a TOP, plant RBCS1A, poxvirus poly(A) leader fused to scrambled 5' UTR, TEV, TMV) had higher ribosome load (1.7-2.3) than the hHBB 5' UTR (1.57).

“All of these sequences had a higher ribosome load (1.7–2.3) than the hHBB 5’ UTR (1.57), thereby identifying potential UTR design strategies to boost mRNA translational efficiency.”

Mutation of the uORF in the SARS-CoV-2 5' UTR (AUG to UUG; CoV2-UUG) resulted in higher ribosome load than the wildtype 5' UTR.

“We first observed that mutation of the uORF in the 5’ UTR (AUG mutated to UUG; CoV2-UUG) resulted in higher ribosome load than the wildtype 5’ UTR (Fig. 1F–G).”

Deletion of SL1 from the SARS-CoV-2 5' UTR increased ribosome load from 1.23 (CoV-2) to 1.5 (CoV-2-UUG-dSL-1), almost reaching the hHBB level (1.57).

“In particular, deletion of SL1 from the SARS-CoV-2 5’ UTR increased ribosome load from 1.23 (CoV-2) to 1.5 (CoV-2-UUG-dSL-1) which almost reaches the level of ribosome load of hHBB (1.57) (Fig. 1G).”

The most impactful effects of sequence k-mers were at the 5' and 3' end of the randomized 5' UTR stretch.

“We found the most impactful effects of sequence k-mers at the 5’ and 3’ end of the randomized 5’ UTR stretch (Fig. S2E, Table S3).”

The mRNA library contained 233 different mRNA sequences in total.

“Our mRNA library contained 233 different mRNA sequences in total (Table S1).”

The largest variation in ribosome load was observed for the 5' UTR variants group.

“The largest variation in ribosome load was observed for the 5’ UTR variants group (Fig. 1D).”

At 24 hours, 6 of 8 RiboTree-optimized mRNA designs demonstrated higher luciferase activities than the Nluc reference mRNA.

“6 of 8 RiboTree-optimized mRNA designs demonstrated higher luciferase activities than the Nluc reference mRNA at 24 hours”

RNA sequence in a given structure can dramatically affect degradation of the structure motif.

“When analyzed across known secondary structure motifs, the data revealed that the RNA sequence in a given structure can dramatically affect degradation of the structure motif.”

The majority of the mRNA stabilized with ψ and a structure-optimized CDS remains functional after 2 hours of accelerated solution degradation.

“the majority of the mRNA stabilized with ψ and a structure-optimized CDS remains functional after 2 hours of accelerated solution degradation”

Hydrolysis is suppressed in symmetric internal loops compared to asymmetric internal loops.

“A particularly salient characteristic was suppressed hydrolysis in symmetric internal loops compared to asymmetric internal loops (Fig. 3C).”

PERSIST-seq confirms predictions that extremely highly structured mRNAs can exhibit more than double the in-solution half-lives of conventionally designed mRNAs.

“PERSIST-seq confirms predictions that extremely highly structured mRNAs can exhibit more than double the in-solution half-lives of conventionally designed mRNAs”

Within mRNAs encoding for a single protein (Nluc, eGFP, or MEV), the variance in in-solution half-lives was greater than for the variance across UTR variants.

“Within mRNAs encoding for a single protein (Nluc, eGFP, or MEV), the variance in in-solution half-lives was greater than for the variance across UTR variants (Fig. 2G), and these values correlated well with different metrics for predicted structure (see below).”

• AssumptionThe authors state this as an assumption: The high pH buffer containing Mg2+ mimics the high effective pH and positively charged environment that can arise in lipid nanoparticles, protamine, and other formulations for mRNA therapeutics.

The effect of uridine susceptibility can be alleviated through inclusion of the nucleoside modification ψ or m1ψ.

“this effect can be alleviated through inclusion of the nucleoside modification ψ or m1ψ”

There is a 'sweet spot' for translation efficiency for maximal total protein output.

“there is a ‘sweet spot’ for translation efficiency for maximal total protein output.”

Rules for hydrolytic degradation depended on the type of RNA structural loop in which a nucleotide appears.

“In addition, we noted rules for hydrolytic degradation that depended on the type of RNA structural loop in which a nucleotide appears.”

ψ and m1ψ substitutions did not change the global structure of the RNAs.

“Structure mapping data based on SHAPE and DMS profiling confirmed that ψ and m1ψ substitutions did not change the global structure of the RNAs; the suppression of in-line hydrolysis appears to be due to local chemical or structural effects.”

At 24 hours, 6 of 8 Eterna-derived designs tested had lower luciferase activity than the Nluc reference mRNA.

“At 24 hours, the trend of lower luciferase activity continued for 6 of 8 Eterna-derived designs tested compared to the Nluc reference mRNA”

For six constructs from the 233x-mRNA library, uniform substitution of U with ψ led to 1.2-2.7-fold stabilization of in-solution degradation lifetimes for longer Nluc-encoding mRNAs.

“Consistent with our in-line hydrolysis data on small RYOS RNAs, we observed a 1.2–2.7-fold stabilization for these longer Nluc-encoding mRNAs when U was uniformly substituted with ψ (Fig. S5C).”

PERSIST-seq gives a wide dynamic range of in-cell half-lives, ranging from less than 5 hours to over 15 hours.

“For our library of 233 mRNAs, PERSIST-seq gives a wide dynamic range of in-cell half-lives, ranging from less than 5 hours to over 15 hours (Fig. 2A).”

At 6 hours, most vendor-derived and GC-rich designs had slightly higher luciferase activities than the reference Nluc sequence.

“whereas most vendor-derived and GC-rich designs had slightly higher activities (Fig. 4C).”

RNA structure was a dominant predictor of in-line hydrolysis rate.

“At a broad level, the data confirmed that RNA structure was a dominant predictor of in-line hydrolysis rate (compare, e.g., SHAPE to in-line data; Fig. S4A), but a closer look revealed additional sequence and structure dependent rules for in-line hydrolysis.”

Highly structured CDSs can strongly affect both the in-solution stability of mRNAs and their protein synthesis in cells.

“Moreover, using our RNA degradation predictor DegScore and Eterna-derived designs, we saw that the highly structured CDSs can strongly affect both the in-solution stability of mRNAs and their protein synthesis in cells (Fig. 2, Fig. 3).”

In cells, modulation of UTR sequence produced large variation in in-cell stability.

“For example, in cells, modulation of UTR sequence produced large variation in in-cell stability (Fig. 2A), presumably through changes in recruitment of cellular machinery that affect mRNA decay.”

• AssumptionThe authors state this as an assumption: The high pH buffer containing Mg2+ mimics the high effective pH and positively charged environment that can arise in lipid nanoparticles, protamine, and other formulations for mRNA therapeutics.

In aqueous solution without cellular machinery, changing the UTRs produced comparably little change in RNA stability to hydrolytic degradation of the CDS.

“In contrast, in aqueous solution without such cellular machinery, changing the UTRs produced comparably little change in RNA stability to hydrolytic degradation of the CDS (Fig. 2G).”

• AssumptionThe authors state this as an assumption: The high pH buffer containing Mg2+ mimics the high effective pH and positively charged environment that can arise in lipid nanoparticles, protamine, and other formulations for mRNA therapeutics.

The widest in-cell stability variation was observed in the CDS and 5' UTR variant groups.

“However, surprisingly, we instead observed the widest in-cell stability variation in the CDS and 5’ UTR variant groups (Fig. 2A).”

DegScore gave the highest accuracy in predicting in-solution stability over all measured mRNAs, including low and high structure eGFP mRNAs from Moderna, Eterna, and Ribotree.

“Beyond the Nluc CDS variants, we confirmed that DegScore gave the highest accuracy in predicting in-solution stability when evaluated over all the measured mRNAs, including low and high structure eGFP mRNAs from Moderna researchers, Eterna, and Ribotree (Table S1 and Fig. S5A,B).”

For the most-sampled triloop secondary structure, in-line hydrolysis rates varied by up to 100-fold depending on sequence.

“For example, in the case of the most-sampled secondary structure of triloops, the in-line hydrolysis rates varied by up to 100-fold depending on sequence (Fig. S4B).”

DegScore correlated moderately with 24 hour protein expression (R = −0.47, p = 0.02).

“and moderately with 24 hour protein expression (R = −0.47, p = 0.02)”

The largest changes in in-solution stability occurred across CDS variants.

“The largest changes in in-solution stability occurred across CDS variants.”

• AssumptionThe authors state this as an assumption: The high pH buffer containing Mg2+ mimics the high effective pH and positively charged environment that can arise in lipid nanoparticles, protamine, and other formulations for mRNA therapeutics.

Both Nluc start and LinearDesign-1 CDSs with hHBB UTRs maintained high protein expression at 6 and 24 hours.

“Importantly, both Nluc start and LinearDesign-1 CDSs with hHBB UTRs maintained high protein expression at 6 and 24 hours (Fig. 4A) indicating that, despite ψ modification and highly structured CDSs, translation is sustained.”

The shortest mRNAs in the pool, encoding MEV, exhibited in-solution half-lives of 3.4±0.6 hours.

“The shortest mRNAs in the pool, encoding a multi-epitope SARS-CoV-2 vaccine CDS (MEV), exhibited in-solution half-lives of 3.4±0.6 hours.”

• AssumptionThe authors state this as an assumption: The high pH buffer containing Mg2+ mimics the high effective pH and positively charged environment that can arise in lipid nanoparticles, protamine, and other formulations for mRNA therapeutics.

The results for in-solution stability were strikingly different from the results for in-cell stability across the mRNA library.

“The results for in-solution stability were strikingly different from the results for in-cell stability across the mRNA library.”

• AssumptionThe authors state this as an assumption: The high pH buffer containing Mg2+ mimics the high effective pH and positively charged environment that can arise in lipid nanoparticles, protamine, and other formulations for mRNA therapeutics.

Ribotree_LinearDesign_degscoreall_1 mRNA had an in-solution half-life of 2.4 relative to Nluc start.

“yielded an mRNA that was both highly stable in solution (t1/2 = 2.4, relative to Nluc start)”

Preparation of mRNAs with ψ substituted for U led to increased in-solution stability across both Nluc start and LinearDesign-1 CDSs, independent of UTR.

“As expected, preparation of mRNAs with ψ substituted for U led to increased in-solution stability across both these CDSs, independent of UTR (Fig. S6).”

RNA linkages 5' of a uridine residue are particularly susceptible to degradation.

“RNA linkages 5’ of a uridine residue are particularly susceptible to degradation”

Unstable mRNAs include 5' UTR or 5'/3' UTR variants that were heavily polysome shifted.

“More specifically, these unstable mRNAs include 5’ UTR or 5’/3’ UTR variants that were heavily polysome shifted (Fig. 2B, polysome/monosome ratio).”

Increased association to monosomes and thus a more moderate increase in overall ribosome load is positively associated with mRNA stability.

“Increased association to monosomes and thus a more moderate increase in overall ribosome load is positively associated with mRNA stability (Fig. 2B, monosomes/pre-polysome(pre-80S) ratios).”

DegScore was uncorrelated with 6 hour protein expression.

“yet was uncorrelated with 6 hour protein expression (Fig. 4E).”

The largest spread of in-solution half-lives (2.8-fold) occurred in the CDS variants for Nluc mRNAs.

“The largest spread of in-solution half-lives (2.8-fold) occurred in the CDS variants for Nluc mRNAs.”

• AssumptionThe authors state this as an assumption: The high pH buffer containing Mg2+ mimics the high effective pH and positively charged environment that can arise in lipid nanoparticles, protamine, and other formulations for mRNA therapeutics.

By 24 hours, mRNA with LinearDesign-1 CDS displayed a two-fold increase in luciferase yield compared to Nluc start.

“Indeed, by 24 hours, the mRNA with the LinearDesign-1 CDS displayed an outstanding two-fold increase in luciferase yield compared to that of Nluc start (Fig. 4A).”

The optimized Ribotree_LinearDesign_degscoreall_1 mRNA exhibited higher resistance to degradation in-solution than Nluc Start mRNA.

“the optimized Ribotree_LinearDesign_degscoreall_1 mRNA exhibited higher resistance to degradation in-solution than the Nluc Start mRNA”

Substitution of U with either ψ or m1ψ led to suppression of in-line hydrolysis at nucleotides 1 to 2 positions 5' of the substitution.

“We also observed suppression of in-line hydrolysis at nucleotides 1 to 2 positions 5′ of the substitution (Fig. 3F), possibly due to locally enhanced base stacking72.”

Eterna-derived designs exhibited up to 2.6-fold higher in-solution half-life compared to the Nluc start reference sequence.

“Compared to the Nluc start reference sequence, designs originating from Eterna exhibited up to 2.6-fold higher in-solution half-life (see, e.g., RLT-10)”

Combining optimal UTRs, DegScore optimization with RiboTree, and ψ modification achieves improved protein expression.

“By combining optimal UTRs, DegScore optimization with RiboTree, and ψ modification, we are able to achieve high mRNA stability and improved protein expression.”

6 hour protein expression correlated strongly with Maximum Ladder Distance (R = −0.55, p<0.001).

“and the “Maximum Ladder Distance73,” or the maximum helix path length (R = −0.55, p<0.001)”

The UTR combinations that outperform hHBB include multiple cellular 5' UTRs and the dengue virus 5' and 3' UTRs.

“include multiple cellular 5’ UTRs as well as, unexpectedly, the dengue virus 5’ and 3’ UTRs”

DegScore accuracy outperformed dGMFE (R = -0.53) and SUP (R = -0.60) in predicting in-solution half-lives for Nluc CDS variants.

“Strikingly, the accuracy of DegScore outperformed the accuracy of two other metrics that have been used in prior studies to parametrize RNA structure but do not take into sequence or structure-motif dependences of RNA hydrolysis: the free energy of the predicted minimum free energy secondary structure, the metric used in several design algorithms including LinearDesign (dGMFE; R = −0.53), and the”

DegScore regression coefficients with the largest magnitude corresponded to the identity of the nucleotide 3' of a linkage.

“The DegScore regression coefficients with the largest magnitude corresponded to the identity of the nucleotide 3’ of a linkage (Fig. 3D).”

High translation efficiency mRNAs associated with the heaviest polysomes tend to be less stable.

“high translation efficiency mRNAs associated with the heaviest polysomes tend to be less stable”

LinearDesign-1 CDS mRNAs exhibited a particularly long in-solution half-life.

“LinearDesign-1 CDS mRNAs also exhibited a particularly long in-solution half-life (Fig. 3G–H).”

mRNAs designed to have highly structured CDSs by the LinearDesign algorithm exhibited polysome fraction profiles dominated by 80S monosomes.

“We did note that mRNAs designed to have highly structured CDSs by the LinearDesign algorithm exhibited polysome fraction profiles dominated by 80S monosomes (Fig. 1E), further discussed below.”

DegScore predictions strongly correlate with in-solution half-lives for Nluc CDS variants with the widest variance (Pearson R = -0.63, p<0.0001).

“For the Nluc CDS variants that showed the widest variance in in-solution half-lives, we observed a strong correlation of DegScore predictions to the in-solution half-lives (Pearson R = −0.63, p<0.0001).”

Designs from the LinearDesign server and modifications from Eterna participants and the RiboTree algorithm produced constructs with increased half-life up to approximately 2.5-fold over the Nluc reference sequence.

“Designs from the LinearDesign server and modifications to these designs from both Eterna participants (e.g., AB_rhiju_mod3) and the RiboTree algorithm produced constructs with increased half-life up to approximately 2.5-fold over the Nluc reference sequence”

Chemical structure mapping showed that the long-lived LinearDesign-1 was significantly more highly structured than Yellowstone, as assessed by DMS and SHAPE reactivities.

“Chemical structure mapping showed that the long-lived LinearDesign-1 was significantly more highly structured than Yellowstone, as assessed by dimethyl sulfate (DMS) and selective 2’-hydroxyl acylation with primer extension (SHAPE) reactivities59,60 (Fig. 2H, Fig. S3) and structure models guided by these data (Fig. 2I).”

• cites [59] Leonhardt, C. et al. Single-cell mRNA transfection studies: delivery, kinetics and statistics by numbers. Nanomedicine 10, 679–688 (2014). sourceNot checked yet

  Checked before — its earlier verdict is held back until a new check runs.

• cites [60] Wang, H. et al. Diversity of putative archaeal RNA viruses in metagenomic datasets of a yellowstone acidic hot spring. Springerplus 4, 189 (2015). sourceNot checked yet

  Checked before — its earlier verdict is held back until a new check runs.

• AssumptionThe authors state this as an assumption: The high pH buffer containing Mg2+ mimics the high effective pH and positively charged environment that can arise in lipid nanoparticles, protamine, and other formulations for mRNA therapeutics.

In many RNA loop types, linkages that lead to a 3' uridine were particularly prone to degradation.

“Furthermore, in many RNA loop types, it appeared that linkages that lead to a 3’ uridine were particularly prone to degradation (Fig. 3C, Fig. S4C), and this effect was reproduced in follow-up experiments using capillary electrophoresis readouts (Fig. S4D).”

PERSIST-seq identifies previously unknown tradeoffs and additive effects between different tunable aspects of mRNA performance as key determinants.

“This technology identifies previously unknown tradeoffs and additive effects between different tunable aspects of mRNA performance as key determinants.”

RiboTree's computational modifications from the starting LinearDesign solution are characterized by reducing Us in loops and shifting local base-pairing to minimize overall loop size, even if resulting in additional smaller loops.

“These computational modifications are characterized by reducing the presence of Us in loops, and shifting local base-pairing to minimize the overall size of loops, even if such shifts result in additional smaller loops.”

mRNAs with longer UTRs had consistently reduced in-solution half-life to hydrolytic degradation across all 6 CDSs.

“In terms of in-solution stability, we expected the mRNAs with the longer UTRs to have consistently reduced half-life to hydrolytic degradation across all 6 CDSs, and capillary electrophoresis experiments confirmed this expectation (Fig. S6).”

Substitution of U with either ψ or m1ψ led to remarkable suppression of in-line hydrolysis at the substituted residues.

“We observed that substitution of U with either ψ or m1ψ led to remarkable suppression of in-line hydrolysis at the substituted residues, presumably through changed nucleophilicity at the site of substitution (Fig. 3F).”

Selected CDS designs support a range of observed in-solution half-lives from 0.69- to 1.8-fold relative to Nluc start.

“The selected CDS designs represented CDSs from diverse origins that support a range of observed in-solution half-lives (from 0.69- to 1.8-fold relative to our reference ‘Nluc start’ sequence; Fig. 2G).”

PERSIST-seq enables massively parallel evaluation of the effects of UTR and CDS sequence and structure on in-cell mRNA translation efficiency, in-cell mRNA stability, and in-solution stability.

“Our first set of experimental findings derive from the integrated PERSIST-seq technology, which enables massively parallel evaluation of the effects of UTR and CDS sequence and structure on in-cell mRNA translation efficiency, in-cell mRNA stability, and in-solution stability.”

6 hour protein expression correlated strongly with the predicted number of hairpins (R = 0.70, p<0.001).

“6 hour protein expression correlated strongly with the predicted number of hairpins (R = 0.70, p<0.001)”

DegScore quantitatively captured features like increased hydrolysis rates at linkages leading to 3' U.

“To distill these observations into a predictive model, we trained a windowed ridge regression model called ‘DegScore’ based on these In-line-seq data (Fig. 3D; see Methods) which quantitatively captured features like the increased hydrolysis rates at linkages leading to 3’ U (Fig. 3D).”

The dengue virus 3' UTR individually increases ribosome loading.

“the dengue virus 5’ and 3’ UTRs, which both individually increase ribosome loading”

Less variance in ribosome loading was observed among CDS variants than for UTR variants.

“As noted above, we observed less variance in ribosome loading among CDS variants than for UTR variants (Fig. 1D–E).”

Designs from commercial algorithms or optimized GC-content tended to exhibit similar in-solution half-lives to the Nluc start reference sequence.

“while designs from commercial algorithms or optimized GC-content tended to exhibit similar in-solution half-lives (Fig. 4C).”

Incorporation of ψ into either mRNA further enhanced in-solution stability, with a readout based on mRNA function in cells.

“incorporation of ψ into either mRNA further enhanced in-solution stability, this time with a readout based on mRNA function in cells (Fig 4G).”

The longest mRNAs, encoding eGFP, exhibited much shorter in-solution half-lives of 1.1±0.08 hours.

“The longest mRNAs, encoding eGFP, exhibited much shorter in-solution half-lives of 1.1±0.08 hours (Fig. 2G), as expected given the larger number of sites of potential hydrolysis.”

• AssumptionThe authors state this as an assumption: The high pH buffer containing Mg2+ mimics the high effective pH and positively charged environment that can arise in lipid nanoparticles, protamine, and other formulations for mRNA therapeutics.

The starting mRNA sequence gives negligible in-cell activity after 2 hours of accelerated solution degradation.

“whereas our starting mRNA sequence gives negligible in-cell activity after the same degradation (Fig 4G).”

mRNAs harboring highly structured 5' and 3' UTRs can support high levels of protein synthesis in cells.

“Thus far we found that mRNAs harboring highly structured 5’ and 3’ UTRs can support high levels of protein synthesis in cells.”

Constructs with higher ribosome load values tended to be more unstable.

“We also noted that constructs with higher ribosome load values tended to be more unstable (Fig. 2B).”

The next largest source of variation in in-solution stability was driven by differences in mRNA structure.

“The next largest source of variation in in-solution stability was driven by differences in mRNA structure.”

• AssumptionThe authors state this as an assumption: The high pH buffer containing Mg2+ mimics the high effective pH and positively charged environment that can arise in lipid nanoparticles, protamine, and other formulations for mRNA therapeutics.

Combining the dengue virus 5' and 3' UTRs in one mRNA results in an additive effect on ribosome loading.

“combining them in one mRNA resulted in an additive effect”

Certain UTR combinations outperform hHBB in translation efficiency.

“certain UTR combinations outperform hHBB in translation efficiency”

The ribosome load of CDS variants had no obvious correlation with CAI, GC content, MFE, addition of signal peptides, or nonsynonymous mutations.

“The ribosome load of CDS variants had no obvious correlation with the codon adaptation index (CAI), GC content, minimum free energy (MFE), addition of signal peptides, or nonsynonymous mutations (Table S1).”

MEV-encoding CDS variants displayed less ribosome load than Nluc- or eGFP-encoding CDS variants, as expected from their shorter ORFs.

“The only clear effects on ribosome load were that MEV-encoding CDS variants displayed less ribosome load than Nluc- or eGFP-encoding CDS variants, as expected from their shorter ORFs.”

At 6 hours, Eterna- and Ribotree-derived designs tended to have similar or lower luciferase activities than the reference Nluc sequence.

“at 6 hours, both Eterna- and Ribotree-derived designs tended to have similar or lower luciferase activities than the reference Nluc sequence”

In-solution half-lives correlated well with different metrics for predicted structure.

“and these values correlated well with different metrics for predicted structure (see below).”

• AssumptionThe authors state this as an assumption: The high pH buffer containing Mg2+ mimics the high effective pH and positively charged environment that can arise in lipid nanoparticles, protamine, and other formulations for mRNA therapeutics.

DegScore correlated strongly with measured in-solution half-life (R = −0.55, p<0.001).

“DegScore correlated strongly with measured in-solution half-life (R = −0.55, p<0.001)”

Predicted and measured luciferase activities showed a correlation of r = 0.71.

“The predicted and measured luciferase activities showed a correlation of r = 0.71 (Fig. 2F), supporting the accuracy of our model.”

• AssumptionThe authors state this as an assumption: Translation follows first-order kinetics and mRNA and protein decay are exponential.

We obtained single-nucleotide-resolution measurements of 3030 RNA fragments using In-line-seq.

“In particular, we obtained single-nucleotide-resolution measurements of 3030 RNA fragments using In-line-seq, a version of a low-bias ligation and reverse transcription protocol (MAP-seq)64 adapted here for in-line hydrolysis profiles65 (Fig. 3B).”

• cites [64] Mikkola, S., Kaukinen, U. & Lönnberg, H. The effect of secondary structure on cleavage of the phosphodiester bonds of RNA. Cell Biochem. Biophys. 34, 95–119 (2001). sourceNot checked — no full text of the source
• cites [65] Seetin, M. G., Kladwang, W., Bida, J. P. & Das, R. Massively parallel RNA chemical mapping with a reduced bias MAP-seq protocol. Methods Mol. Biol. 1086, 95–117 (2014). sourceNot checked yet

  Checked before — its earlier verdict is held back until a new check runs.

After 6 hours, two CDSs (LinearDesign-1 and GCrich_2) achieved similarly high protein levels compared to Nluc start, when combined with hHBB 5' and 3' UTRs.

“After 6 hours, two CDSs (LinearDesign-1 (Fig. 3) and GCrich_2) achieved similarly high protein levels compared to Nluc start, when combined with hHBB 5’ and 3’ UTRs.”

Ribotree_LinearDesign_degscoreall_1 mRNA exhibited a 1.7-fold increase in protein expression relative to Nluc start.

“and exhibited high levels of protein expression in cells (1.7-fold increase, relative to Nluc start) (Ribotree_LinearDesign_degscoreall_1; Fig. 4C).”

A key determinant of in-solution RNA degradation is the presence of uridine.

“a key determinant of in-solution RNA degradation is the presence of uridine”

Among the three UTR combinations and six CDS designs tested, most demonstrated lower overall luciferase activity than the reference start Nluc with hHBB UTRs.

“Among the three UTR combinations and six CDS designs tested, most demonstrated lower overall luciferase activity than the reference start Nluc with hHBB UTRs (Fig 4A).”

A closer look revealed additional sequence and structure dependent rules for in-line hydrolysis.

“but a closer look revealed additional sequence and structure dependent rules for in-line hydrolysis.”

The dengue virus 5' UTR individually increases ribosome loading.

“the dengue virus 5’ and 3’ UTRs, which both individually increase ribosome loading”

In-cell mRNA stability may be a greater driver of protein output than high ribosome load, particularly when proteins need to be expressed from a single dose of mRNAs for long periods of time.

“we find that in-cell mRNA stability may be a greater driver of protein output than high ribosome load particularly when proteins need to be expressed from a single dose of mRNAs for long periods of time.”

Combining optimal UTRs, DegScore optimization with RiboTree, and ψ modification achieves high mRNA stability.

“By combining optimal UTRs, DegScore optimization with RiboTree, and ψ modification, we are able to achieve high mRNA stability”

There is a tradeoff between ribosome load and mRNA stability.

“this observation foreshadowed a tradeoff between ribosome load and mRNA stability that we dissect in greater detail below.”

• Also stated in: § In-cell mRNA stability is a major predictor of total protein output · plain

  “Importantly, these findings identify an unexpected rule for the design of mRNA – excessively increasing translation efficiency may negatively impact mRNA stability.”

PERSIST-seq enables systematic determination of the effects of UTR, codon choice and RNA structure on mRNA translation rates in human cells and on mRNA stability in cells and in solution.

“We present a massively parallel reporter assay termed Pooled Evaluation of mRNA in-solution Stability, and In-cell Stability and Translation RNA-seq (PERSIST-seq), which enables systematic determination of the effects of UTR, codon choice and RNA structure on mRNA translation rates in human cells and on mRNA stability, both in cells and in solution.”

In-cell mRNA stability is a greater driver of total protein output than high ribosome load.

“We find that, surprisingly, in-cell stability is a greater driver of protein output than high ribosome load.”

• Also stated in: § In-cell mRNA stability is a major predictor of total protein output · boosted

  “Taken together, our large-scale and unbiased pooled measurement of translation and mRNA stability provides a simple but powerful platform to test large numbers of different mRNA sequence designs for desired protein expression and has allowed us to importantly infer that in-cell mRNA stability is the dominant predictor of total protein output (Fig. 2C,E).”

• AssumptionThe authors state this as an assumption: Translation follows first-order kinetics and mRNA and protein decay are exponential.

Different 5' UTRs have strong potential to tune translational efficiency of target mRNAs, more than 3' UTR or CDS regions.

“These data suggested strong potential for using different 5’ UTRs to tune translational efficiency of target mRNAs – more than for any other region (3’ UTR or CDS).”

This study represents the first large-scale screen of mRNA redesigned across its entire length.

“This represents the first large-scale screen of mRNA redesigned across its entire length.”

The combination of optimized UTRs for expression, DegScore-optimized CDS structure, and ψ modification provides a general technology that can be applied to stabilize and increase protein expression of candidate mRNA therapeutics.

“Together, the combination of optimized UTRs for expression, DegScore-optimized CDS structure, and ψ modification provides a general technology that can be applied to stabilize and increase protein expression of candidate mRNA therapeutics.”

The model enables identification of optimal mRNA designs without the need to individually test all mRNAs for protein output.

“This model enables identification of optimal mRNA designs without the need to individually test all mRNAs for protein output.”

• AssumptionThe authors state this as an assumption: Translation follows first-order kinetics and mRNA and protein decay are exponential.

In-cell mRNA stability is a primary driver of protein output.

“This finding is consistent with faster mRNA decay compared to hHBB UTRs and our results supporting that in-cell mRNA stability is a primary driver of protein output (Fig. 2C, E).”

• Also stated in: § One-by-one tests confirm additivity of UTR, CDS, and ψ improvements · boosted

  “Overall, these results demonstrated the importance of mRNA stability to protein output”

For short expression (early time points after mRNA transfection) or for "above-threshold" protein half-lives, translation can become the dominant predictor.

“However, for short expression (early time points after mRNA transfection) or for “above-threshold” protein half-lives, translation can become the dominant predictor.”

• AssumptionThe authors state this as an assumption: Translation follows first-order kinetics and mRNA and protein decay are exponential.

Downstream luciferase expression can be further amplified by ψ modification of mRNAs.

“Moreover, downstream luciferase expression can be further amplified by ψ modification of mRNAs.”

Combining UTRs that facilitate high levels of translation with structural optimization of CDSs via LinearDesign and DegScore-guided RiboTree design can enhance both in-solution stability and total in-cell protein output of mRNAs.

“Taken together, our detailed studies on individual mRNAs (Fig. 4) indicate that combining UTRs that facilitate high levels of translation alongside structural optimization of CDSs via LinearDesign and DegScore-guided RiboTree design, we can enhance both in-solution stability and total in-cell protein output of mRNAs.”

Longer or more branched double-stranded RNA stems may potentially hinder the ability of the ribosome to unwind RNA secondary structure.

“For instance, longer or more branched double-stranded RNA stems may potentially hinder the ability of the ribosome to unwind RNA secondary structure.”

In-solution RNA stability can be significantly further improved by incorporating modified U nucleosides, beyond redesigning RNA sequences to take on stable structure.

“This finding indicated that beyond redesigning RNA sequences to take on stable structure, in-solution RNA stability can be significantly further improved by incorporating modified U nucleosides.”

There does not have to be a tradeoff between mRNA structure, stability, and protein output.

“There does not have to be a tradeoff between mRNA structure, stability, and protein output.”

The model's accuracy is supported by the correlation between predicted and measured luciferase activities.

“The predicted and measured luciferase activities showed a correlation of r = 0.71 (Fig. 2F), supporting the accuracy of our model.”

• AssumptionThe authors state this as an assumption: Translation follows first-order kinetics and mRNA and protein decay are exponential.

Predicted protein expression per mole of transfected mRNA is modeled to be mostly driven by the in-cell mRNA half-life.

“Overall, predicted protein expression per mole of transfected mRNA is modeled to be mostly driven by the in-cell mRNA half-life (Fig. 2C, top panel), which more closely trends with predicted protein expression (Fig. 2E) than ribosome load does (Fig. 2C, lower panel).”

• AssumptionThe authors state this as an assumption: Translation follows first-order kinetics and mRNA and protein decay are exponential.

Increased polysome load appears counterproductive to maximizing the total amount of protein expressed over time.

“Stated differently, increased polysome load appears counterproductive to the goal of maximizing the total amount of protein expressed over time.”

DegScore enables in silico optimization of any RNA sequence.

“which enables in silico optimization of any RNA sequence”

DegScore, a model for hydrolytic degradation, was independently validated on PERSIST-seq data.

“DegScore, a model for hydrolytic degradation that was independently validated on PERSIST-seq data”

The study focused on characterizing degradation of naked, non-formulated mRNAs.

“our study has focused on characterizing degradation of naked, non-formulated mRNAs”

A new mRNA design methodology can dramatically enhance mRNA stability in aqueous solution.

“we report a new mRNA design methodology that can dramatically enhance mRNA stability in aqueous solution”

A new mRNA design methodology can sustain or even increase protein expression inside cells.

“we report a new mRNA design methodology that can dramatically enhance mRNA stability in aqueous solution while sustaining or even increasing protein expression inside cells”

Depending on the desired parameters (early burst of high protein expression or total protein output integrated over a longer time), the most optimal set of UTRs can vary.

“Therefore, depending on the desired parameters (early burst of high protein expression or total protein output integrated over a longer time), the most optimal set of UTRs can vary.”

• AssumptionThe authors state this as an assumption: Translation follows first-order kinetics and mRNA and protein decay are exponential.

Enhanced in-cell half-life of a structured mRNA can compensate for low translation efficiency.

“The result was nevertheless consistent with our model that enhanced in-cell half-life of a structured mRNA can compensate for low translation efficiency (Fig. 2C–E).”

The global assessment of in-solution RNA degradation using PERSIST-Seq reveals pronounced effects of RNA length and structure on RNA half-life in solution.

“Overall, the global assessment of in-solution RNA degradation using PERSIST-Seq reveals pronounced effects of RNA length and structure on RNA half-life in solution.”

• AssumptionThe authors state this as an assumption: The high pH buffer containing Mg2+ mimics the high effective pH and positively charged environment that can arise in lipid nanoparticles, protamine, and other formulations for mRNA therapeutics.

These modifications are consistent with mitigating hydrolysis as modeled in the DegScore predictor.

“These modifications are consistent with mitigating hydrolysis as modeled in the DegScore predictor (Fig. 3D).”

Measurements without Mg2+, at lower pH, and higher temperature gave strongly correlated results to the accelerated degradation conditions used for PERSIST-seq in-solution stability measurements.

“We matched the accelerated degradation conditions used for PERSIST-seq in-solution stability measurements, but also verified that measurements without Mg2+, at lower pH, and higher temperature gave strongly correlated results (Fig. S4A).”

Excessively increasing translation efficiency may negatively impact mRNA stability.

“Importantly, these findings identify an unexpected rule for the design of mRNA – excessively increasing translation efficiency may negatively impact mRNA stability.”

Despite ψ modification and highly structured CDSs, translation is sustained.

“indicating that, despite ψ modification and highly structured CDSs, translation is sustained.”

The simultaneous increase in in-solution stability and in-cell protein expression for Ribotree_LinearDesign_degscoreall_1 demonstrates the impact of the design rules for mRNA.

“This simultaneous increase in in-solution stability with improved and sustained in-cell protein expression provided a strong demonstration of the impact of our design rules for mRNA.”

Reducing the overall presence of Us in loops and reducing the number of hairpins to generate a 'linear' highly double-stranded mRNA can lead to enhanced mRNA stability and protein expression.

“Taken together, these data suggest that by reducing the overall presence of Us in loops and reducing the number of hairpins to generate a “linear” highly double-stranded mRNA can lead to enhanced mRNA stability and protein expression.”

The suppression of in-line hydrolysis by ψ/m1ψ appears to be due to local chemical or structural effects.

“the suppression of in-line hydrolysis appears to be due to local chemical or structural effects.”