Skip to main content
NCBI home page
Protein Engineering, Design and Selection logoLink to Protein Engineering, Design and Selection
. 2026 Mar 10;39:gzag008. doi: 10.1093/protein/gzag008

Computational redesign of a thermostable T7 RNA polymerase

Zachary T Baumer 1, Timothy A Whitehead 2,✉
Editor: Robert Campbell
PMCID: PMC13633397  PMID: 41804789

Abstract

T7 RNA polymerase (T7 RNAP) is a foundational enzyme for biotechnology, but its utility for many potential applications is limited by low thermal stability of 43–44°C. While stabilized variants exist, the most stable commercial version has a proprietary sequence. In this work we developed a highly stable T7 RNAP using structure-based computational design. We combined mutations from previous stabilized variants (M5, M8, V7abcd) with new mutations identified by PROSS. These mutations were filtered using data-driven heuristics to preserve function. Our final design, T7T+, contains 30 point mutations from the original T7 RNAP and demonstrates a functional stability (T50) of 54.9°C in a thermal challenge assay, which is 2.4°C higher than the most stable, published open-source variant to date. Circular dichroism spectroscopy showed an apparent melting temperature of 53.8°C. T7T+ retains 59% of wild-type activity at 37°C. 16 of the 18 tested protein designs had higher stability against thermal challenge compared with the genetic background, attesting to the high success rates of existing non deep learning computational methods for the design of stable, functional proteins. A plasmid encoding T7T+ has been deposited in AddGene and is freely available for non-commercial use.

Keywords: RNA polymerase, thermal stability, computational design

Introduction

The single subunit T7 RNA polymerase (T7 RNAP) is a foundational enzyme for biotechnology. Native and engineered versions of T7 RNAP are widely used for instantiation of synthetic biology gene circuits (Baumer et al., 2024; McSweeney et al., 2025), for recombinant expression of proteins in bacteria (Studier, 2005), and for in vitro transcription of therapeutic mRNA (Dousis et al., 2023). Thermally stabilizing T7 RNAP can expand its utility in biotech. In vitro transcripts produced at higher temperatures have reduced dsRNA formation, which lessens immunogenicity (Wu et al., 2020). Further improvements to in vitro transcription processes may be realized using immobilized thermostable enzymes to improve downstream purification (Sheldon et al., 2021), enable continuous manufacturing (Thompson et al., 2019), and increase reusability of the immobilized enzymes (Maghraby et al., 2023). Additionally, a thermostable T7 RNAP could be applied in thermophilic organisms, where microbes grow at temperatures in excess of 55°C. Industrial use of thermophiles has comparative advantages in sterility, available metabolic pathways and feedstocks, and cooling costs (Vavitsas et al., 2022).

Native T7 RNAP has an apparent melting and inactivation temperature of 43–44°C at near-neutral pH. Earlier efforts to stabilize T7 RNAP used thermally challenged survival-based screens to identify mutations, which when combined resulted in the M5, M8, and V7abcd variants (Ikeda et al., 1992; Sugiyama et al., 2009; Boulain et al., 2013). The best variants with fully disclosed sequences have thermal stabilities of approximately 52°C (Boulain et al., 2013). A recent paper used PROSS 2.0 (Weinstein et al., 2021) and mutational combinations to stabilize a T7 RNAP variant with a melting temperature below 50°C (Liu et al., 2025). Recent deep learning approaches have identified stabilizing mutations (Jiang et al., 2024, 2025), but the published variants do not approach the thermal stabilities of the earlier described proteins. New England Biolabs scientists used a computational approach to model mutations which were predicted to be stabilizing using the BioLuminate® software package from Schrodinger (Ong et al., 2019). Combining stabilizing mutations led to the highly stabilized Hi-T7® with activity up to 56°C. The specific sequence has not been disclosed, which motivates development of an open source, thermally stable T7 RNAP. In this contribution we used structure-based computational design to further engineer stability into T7 RNAP. We verify stability using in vitro thermal challenge activity assays and biochemically characterize the best candidate. Our enzyme, T7T+, has the highest stability against thermal challenge (T50 54.9°C) of published sequences to date and has 59% of the relative activity of WT T7 RNAP at 37°C.

Methods

Protein generation, expression, and purification

We constructed all T7 designs using the base plasmid pT7-911Q-His6-WT (Rio, 2013), which was a kind gift from the B. Seelig lab at the University of Minnesota. pT7-911Q-His6-WT encodes for an N-terminally His-tagged (MRGSHHHHHHGS-) T7 RNAP K387R expressed from a pT5-lac promoter. K378R existed in the original gene obtained from the Seelig lab, but is conserved in multiple sequence alignments and appears to not have any noticeable effects. M5 was generated by multisite nicking mutagenesis (Kirby et al., 2021) using pT7-911Q-His6-WT as a template. Other plasmids were constructed from gene fragments using Gibson Assembly (Gibson et al., 2009). All primers, gBlocks, and plasmid sequences are listed in the SOURCE_DATA_FILE.xlsx available from Zenodo (doi:http://doi.org/10.5281/zenodo.17593694). All gene sequences were confirmed using Sanger sequencing. Proteins were expressed in BL21(DE3) E. coli and purified (estimated >95%) by standard Ni-IMAC as described previously (Baumer et al., 2024). An example SDS-PAGE gel is shown in SOURCE_DATA_FILE.xlsx. Protein concentration was estimated using the Edelhoch method with calculated extinction coefficients (Edelhoch, 1967).

Computational and rational design

All fragments were constructed in the M5/K387R background (K378R, S430P, N433T, S633P, F849I, F880Y). To identify potential stabilizing mutations, five structures (PDB IDs: 1MSW, 1CEZ, 1QLN, 3E2E, 1H38)(Cheetham and Steitz, 1999; Cheetham et al., 1999; Tahirov et al., 2002; Yin and Steitz, 2002; Durniak et al., 2008) of T7 RNAP were input to PROSS (Goldenzweig et al., 2016) using the default settings. For each PDB file, separate runs were performed for each chain of multimeric structures (e.g. 1H38-A, 1H38-B, 1H38-C, and 1H38-D). Each run was also performed using both the Talaris and ref2015 energy functions, resulting in a total number of 144 designs (16 jobs; 9 designs per job). Mutations were selected by compiling and ranking the number of times each position and mutation were predicted to be stabilizing. Structural analysis of high ranking mutations were performed manually, selecting mutations which were surface exposed, Cα least 8 Å from the Cα of other mutated positions, and distant from the specificity loop and domains important to transcriptional activity. Functionally, these requirements were that mutations were solvent exposed and showed up in five or more jobs (out of 16). To determine distance from active site, positions were considered only if its Cα was more than 4 Å from DNA in any structure (1CEZ, 1MSW, 1QLN). We also did not include positions in previously described Motifs A, B, and C (Bonner et al., 1992), in the specificity loop (A: 532–554; B: 625–652; C: 805–818; Specificity Loop: 739–773), (Cheetham et al., 1999) or at the M5 mutations (430, 433, 633, 849, 880). The total mutations considered was reduced to 149 mutations through ranking (PROSS_outputs.xlsx available on Zenodo) before being manually curated to the final fragments and mutations listed in Table 1.

Table 1.

Screening of T7 RNAP designs. Screens represent individual replicates for indicated variants.

Variant Mutations Relative Activity (%) T50 (oC; screen 1) T50 (oC; screen 2) T50 (oC; screen 3) T50 (oC; screen 4)
WT K378R 100 ± 4 44.8 NT NT NT
M5 K378R, S430P, N433T, S633P, F849I, F880Y 106 ± 4 49.5 49.4 NT NT
A1 M5 + L16N, A49E, A70V, C125M, N165H 63 ± 4 48.9 NT NT NT
A2 M5 + R31D, A61R, F105Y, S128K, V177H, Y312D 65 ± 1 50.6 50.4 NT NT
B1 M5 + R379K, H411Y, E484D, E498D, T566K, I581Q, A615T 104 ± 3 49.8 49.7 NT NT
B2 M5 + D388N, F481W, S495N, N529V, S606T, G618Q 108 ± 3 50.1 49.9 NT NT
C M5 + S684A, W797H, A822E, S838E, N871D 97 ± 3 46.5 NT NT NT
A2-B1 A2 + B1 60 ± 1 NT 50.6 NT NT
A2-B2 A2 + B2 73 ± 2 NT 51.0 NT NT
V7abcd* M5 + C510R, Q744R, S767G, Q786L 60 ± 4 NT 53.3 53.0 NT
M8* M5 + K713E, S767G, H772R, Q786L 69 ± 4 NT 53.5 53.2 NT
A2-B1-V7abcd*-R510C B1 + Q744R, S767G, Q786L NT NT NT 52.6 NT
A2-B1-V7abcd* A2 + B1 + V7abcd* NT NT NT 53.7 NT
A2-B2-Vabcd* A2 + B2 + V7abcd* 45 ± 6 NT NT 54.2 NT
A2-B1/B2-V7abcd* A2 + B1 + B2 + V7abcd* NT NT NT 54.6 NT
A2-B1-M8*-C510R A2 + B1 + M8* + C510R NT NT NT 54.3 NT
A2-B2-M8* A2 + B2 + M8* NT NT NT 53.8 53.7
A2-B2-M8*-C510R A2 + B2 + M8* + C510R 54 ± 3 NT NT 54.8 54.7
A2-B1/B2-M8* A2 + B1 + B2 + M8* NT NT NT 53.9 54.0
T7T+ A2 + B1 + B2 + M8* + C510R 59 ± 2 NT NT 54.8 55.0
Open in a new tab

NT: not tested. Error bars represent 1 s.d. (n ranges from 3 to 9). Relative activity is defined as the activity of a given variant relative to WT.

Biophysical assays

Thermal challenge assays were performed to measure the functional stability of T7 RNAP. 10 μM of protein was heated in storage solution (50 mM Tris–HCl pH 7.9, 1 mM EDTA, 5 mM beta mercaptoethanol, 100 mM NaCl, 15 (v/v) % glycerol) for 25 minutes in a thermocycler set with a temperature gradient (35–60°C), then brought back to 4°C for at least 20 min. Thermally treated protein was then used in an in vitro transcriptional assay at 37°C as previously described (Baumer et al., 2024). Transcription of the Spinach aptamer (Paige et al., 2011) sequence yields fluorescence with addition of the ligand 3,5-difluor-4-hydroxybenzylidine imidizolanone. Activity is defined as the maximum rate of fluorescence increase as measured over 20 minute intervals [RFU/minute] which occurred in the first 30 minutes of the reaction start. For thermal challenge assays, the apparent T50 was calculated as the temperature at which activity drops to 50% of the maximum measured activity for each variant. The temperature for this analysis was the setpoint temperature for a row of wells in the thermocycler. The T50 was derived by fitting a modified power law to the normalized data in GraphPad PRISM.

Circular dichroism spectroscopy was performed on an Applied Photophysics Chirascan V100. Purified protein was buffer exchanged into 20 mM NaPO4, pH 7.8 using 7 kDa Zeba 0.5 mL spin columns per manufacturer’s protocol and diluted to a concentration in the range of 0.1–0.4 mg/mL. Spectral measurements were taken using 1 sec timepoints at 2 nm bandwidth and 0.5 nm step size. For CD melt curves, the wavelength for measuring ellipticity was maintained at 222 nm and the ellipticity was measured for 24 secs between 0.5°C temperature steps as the sample was heated at 1°C/min. The actual temperature of the sample was monitored and used for the analysis of the Tm,app. Instead of fitting data to a two state model, we evaluate the Tm,app as being the temperature where the maximum derivative of ellipticity as a function of temperature occurs. We wrote a custom Python script to evaluate the Tm,app. First, a Gaussian filter was applied to smooth the data (gaussian_filter1d, sigma = 1). Tm,app was taken as the maximum of the derivative evaluated by ‘numpy.gradient.’

Results

We used the computational design server PROSS 2.0 (Weinstein et al., 2021) to identify potentially stabilizing mutations for T7 RNAP. The stabilizing effect of individual mutations suggested by PROSS are, on average, small (Tsuboyama et al., 2023). Thus, many potential mutations need to be combined to see a potent stabilizing effect. However, there is a stability-activity tradeoff with proteins, where increases to stability often lead to decreases in activity, and vice-versa (Stimple et al., 2020). PROSS combines individually stabilizing mutations using Rosetta repacking, which mainly serves to prevent spatially proximate mutations. Given the large size of T7 RNAP (883 residues) and its complicated transitions between the initiation and elongation complex, we hypothesized that stacking many potential stabilizing mutations using default PROSS would be ultimately detrimental to function. Therefore, we combined PROSS suggested mutations using a more conservative approach. First, we performed PROSS runs on all available T7 RNAP experimentally determined structures. PROSS predicted 657 stabilizing mutations over 365 of the 883 positions. Second, we filtered these PROSS outputs by the number of times a mutation was observed, as well as selecting only positions that are (i.) distal from active sites, with (ii.) low contact numbers, that involve (iii.) mutations are evolutionary conserved, and (iv.) do not involve mutations to and from proline. We have found that filtering mutations by these relatively crude heuristics (Klesmith et al., 2017) improves the redesign success for functional, stabilized proteins (Wrenbeck et al., 2019; Beltrán et al., 2022; Daffern et al., 2024). Third, we tested 5–7 stabilizing mutations at a time in a gene fragment corresponding to a portion of the protein (Fig. 1A and B) and in the M5 background. Gene fragments containing stabilizing mutations were then recombined into the final design.

Figure 1.

Panel A: a cartoon of a gene segment and different gene blocks where mutations are encoded; Panel B: a cartoon of a initiation complex of T7 RNAP bound to promoter DNA (PDB ID: 1MSW). The protein structure is color coded with blocks identified in panel A. C. Relative activity plot for different variants. D. Relative activity vs. Temperature for T7 RNAP and different individual protein designs.

Open in a new tab

Fragment-based computational design of thermostable T7 RNAP. A. Division of the gene encoding T7 RNAP into fragments. Fragment boundaries listed are shown by amino acid positions. B. Fragmented T7 RNAP depicted by corresponding colors shown on the initiation complex of T7 RNAP (PDB ID: 1MSW). C. Relative activity of T7 RNAP designs compared with WT. Except for WT, all enzymes tested were in the M5 genetic background. p-values were determined using Brown-Forsyth and Welch’s ANOVA, corrected for multiple comparisons. *, <.05; **, <.01; ****, <1e-4; ns, not significant. D. Stability against thermal challenge for indicated variants. T7 RNAP designs are incubated at the indicated temperature for 25 minutes before a temperature ramp to 4°C for 20 min. Proteins are then diluted and assayed for activity at 37°C as in panel C. A T50 is calculated as the temperature with 50% relative activity of the starting protein. Screen 1 is shown; information on all screens is shown in Table 1.

In our initial screen, we tested the relative activity and stability after thermal challenge (T50) of five designs using a previously described fluorescence aptamer assay (Baumer et al., 2024; Delaney et al., 2024). We found that two of five designs (A1, A2) led to statistically significant decreases in relative activity (Fig. 1C,  Table 1) to 63-65% of the WT activity (p value <1e-4, Brown-Forsyth and Welch’s ANOVA, corrected for multiple comparisons). Interestingly, the B2 design led to a modest but statistically significant increase in relative activity (Fig. 1C) (p value <0.01). Three of the five designs (A2, B1, B2) showed increases in T50 compared with the M5 background (Fig. 1D,  Table 1). However, these stability increases were modest (ΔT50 ranging from 0.3–1.1°C) relative to the M5 background. Our one design incorporating mutations on the C-terminal domain of T7 RNAP (C; positions 651–883) was destabilizing. Thus, in a second round of screening we tested the addition of the M8 and the V7abcd mutations in the M5 background (Boulain et al., 2013) (M8*, V7abcd*, respectively). The stabilizing mutations target the C-terminal domain of T7 RNAP, which has lower stability than the rest of the protein (Protasevich et al., 1994; Meyer et al., 2015). The M8* and V7abcd* were considerably more stable than the PROSS designs but at the expense of lower relative activity compared to WT (Table 1). In the second round of screening we also tested A2 variants in combination with B1 and B2 mutations, resulting in A2-B1 and A2-B2 variants with nearly additive stabilization of the Tm,app and providing additional support for the feasibility of our approach. For the third screening round, we assessed the T50 of nine different combinations of A1, B1, B2, M8, and the V7abcd mutations. We also assessed the relative activity for designs with the highest T50.

Overall, testing these 18 proteins over the three screening rounds resulted in a lead candidate T7T+ (T7 RNAP R31D, A61R, F105Y, S128K, V177H, Y312D, K378R, R379K, D388N, H411Y, S430P, N433T, F481W, E484D, S495N, E498D, C510R, N529V, T566K, I581Q, S606T, A615T, G618Q, S633P, K713E, S767G, H772R, Q786L, F849I, F880Y) which had the highest T50 and comparatively highest relative activity. T7T+ contains 30 total mutations from WT T7 RNAP sourced from three PROSS-designed gene fragments (A2, B1, B2; this work) and three previous stabilized T7 RNAPs (Sugiyama et al., 2009; Boulain et al., 2013).

We prepared fresh protein preps (SOURCE_DATA_FILE.xlsx) and then assessed T7T+ compared with WT T7 RNAP, the M5 variant, and M8*. Our results are presented in Fig. 2. T7T+ has a relative activity of 59% evaluated at 37ºC, which is similar to that of the M8* variant. We used circular dichroism (CD) spectroscopy to examine the secondary structure and to test the apparent melting temperatures (Tm,app) of the four T7 RNAPs. All four enzymes showed essentially identical CD spectra at room temperature (Fig. 2). A previous study of the thermal melting and urea unfolding of T7 RNAP by CD (Griko et al., 2001) showed that wild-type T7 RNAP undergoes three distinct transitions during its temperature-induced unfolding, likely caused by separate domains unfolding. Thus, a two-state unfolding model is not appropriate to estimate the melting temperature of T7 RNAP; instead, we estimate the Tm,app by the midpoint of the molar ellipticity at 222 nm. Using this definition, WT has a Tm,app = 43.2°C, consistent with previous measurements. T7T+ has a Tm,app = 53.8°C, which is >2°C of any other variant. Interestingly, as we inspect the CD-melt curves for T7 RNAP, a steeper transition arises with increasing stability for the M8* and T7T+ designs, which may be interpreted as the individual unfolding transitions becoming closer in temperature. We also assessed T50 using the previously described thermal challenge assay. T7T+ has a T50 of 54.9°C, which is 2.4°C higher than any other variant.

Figure 2.

An image of four different assays for four different T7 RNAP sequences: WT, M5, M8*, and the T7T+ designs. The plots show from left to right: Activity assay at 37oC using a RNA aptamer assay; far-UV Circular dichroism spectra; Thermal melting evaluated by molar ellipticity at 222 nm; and a thermal challenge activity assay.

Open in a new tab

Comparison of activity and apparent Tm of differentially stabilized T7 RNAPs. A. Wild-type T7 RNAP; B. M5; C. M8*; D. T7T+; [Column I] Activity; [Column II] Circular Dichroism far-UV spectra; [Column III] Temperature melts using circular dichroism spectroscopy. Molar ellipticity was measured at 222 nm during the temperature ramp; [Column IV] Thermal challenge activity assay, normalized to activity at 37°C. Activity plots are the mean trace of two separate runs with the listed ‘relative activity.’ CD Spectra are given as a single spectra that was averaged from two traces during measurement. CD Melt curves are given as a single melt curve for each variant, except T7T+ which was performed in replicate. Thermal challenge activity assays are the average of two technical replicates.

Discussion and conclusion

Estimates from deep mutational scanning suggest that on the order of 1% of random mutations are stabilizing without impacting enzyme function (Klesmith et al., 2017). For T7 RNAP, this means that there are approximately 150 thermally stabilizing point mutations that could be recombined into stable variants. Our design, T7T+, contains up to 30 of these mutations, including 19 identified in the present work. Not all may be stabilizing, and further benefit may be found by testing these and other PROSS-suggested mutations individually. For example, the stabilizing effects of C510R first identified in the Vabcd variant were evident in all the constructs tested (Table 1). Further, there is minimal overlap between mutations in our design and recent thermostabilized T7 RNAPs. While the NEB patent (Ong et al., 2019) does not disclose the exact sequence, it does list potential mutations; 3 out of 19 mutations are shared between this list and T7T+. Taken together, the body of evidence suggests that there is further room for improvement in T7 RNAP with regards to thermal stability.

Our choice of 5–7 mutations per block was empirical. While the data from Rocklin et al. (Tsuboyama et al., 2023) suggest that combining more mutations maximizes stability (average ΔΔG −0.1 to −0.2 kcal/mol per PROSS mutation), their analysis did not account for functional trade-offs. If even 5% of filtered mutations abrogate function, accumulating 10 mutations results in a ~ 40% probability of inactivity. We therefore limited our block size to mitigate this cumulative risk while maintaining significant stability gains.

While deep learning is in vogue across science, our results show that existing computational methods grounded in biophysical principles still have considerable utility. We were able to identify T7T+ by screening only 18 proteins in total; of these, 16 were more thermally stable than the M5 genetic background used in the experiment. To expedite adoption of T7T+ in the broader community, we have deposited a plasmid expressing T7T+ in the AddGene plasmid repository (Kamens, 2014). This construct is freely available for non-commercial use.

Supplementary Material

gzag008_SOURCE_DATA_FILE
gzag008_source_data_file.xlsx (882.1KB, xlsx)

Acknowledgements

This work was supported by the National Science Foundation (Award #: 2218330), and the National Science Foundation Graduate Research Fellowship Program (Z.T.B. DGE Award Number 2040434, fellow ID: 2021324468). We also thank the Shared Instruments Pool (RRID: SCR_018986) of the Department of Biochemistry at the University of Colorado Boulder for the use of the CD spectrometer. The CD is funded by NIH Shared Instrumentation Grant S10RR028036.

Contributor Information

Zachary T Baumer, Department of Chemical and Biological Engineering, University of Colorado Boulder, 596 UCB, Boulder 80309, CO, United States.

Timothy A Whitehead, Department of Chemical and Biological Engineering, University of Colorado Boulder, 596 UCB, Boulder 80309, CO, United States.

Author contributions

Timothy A Whitehead (Conceptualization [equal], Data curation [equal], Funding acquisition [equal], Investigation [equal], Methodology [equal], Project administration [equal], Resources [equal], Supervision [equal], Writing - original draft [equal], Writing - review & editing [equal]), Zachary Baumer (Conceptualization [equal], Data curation [equal], Formal analysis [equal], Funding acquisition [equal], Investigation [equal], Methodology [equal], Software [equal], Validation [equal], Visualization [equal], Writing - original draft [equal]).

Conflict of interest

None declared.

Funding

None declared.

Data and code availability

All data and novel code used in this paper is available in a Zenodo repository (doi:10.5281/zenodo.17593694). This repository contains:(1.) SOURCE_DATA_FILE.xlsx contains all experimental data presented in this paper. This file also contains all primers, plasmid sequences, and gBlock sequences used for plasmid construction. (2.) The Jupyter notebook ‘Evaluating Tmapp from CD.ipynb’ is a custom script for extracting Tm,app from circular dichroism temperature melts. (3.) PROSS_outputs.xlsx lists the full PROSS outputs and our relative ranking of the mutations for testing in this paper. Plasmid pZB246 encoding T7T+ is available on AddGene (ID: 250903).

References

  1. Baumer  ZT, Newton  MS, Löfstrand  L  et al.  Engineered stop and go T7 RNA polymerases. ACS Synth Biol  2024;13:4165–74. 10.1021/acssynbio.4c00627. [DOI] [PubMed] [Google Scholar]
  2. Beltrán  J, Steiner  PJ, Bedewitz  M  et al.  Rapid biosensor development using plant hormone receptors as reprogrammable scaffolds. Nat Biotechnol  2022;40:1855–61. 10.1038/s41587-022-01364-5. [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Bonner  G, Patra  D, Lafer  EM  et al.  Mutations in T7 RNA polymerase that support the proposal for a common polymerase active site structure. EMBO J  1992;11:3767–75. 10.1002/j.1460-2075.1992.tb05462.x. [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Boulain  JC, Dassa  J, Mesta  L  et al.  Mutants with higher stability and specific activity from a single thermosensitive variant of T7 RNA polymerase. Protein Eng Des Sel  2013;26:725–34. 10.1093/protein/gzt040. [DOI] [PubMed] [Google Scholar]
  5. Cheetham  GM, Steitz  TA. Structure of a transcribing T7 RNA polymerase initiation complex. Science  1999;286:2305–9. 10.1126/science.286.5448.2305. [DOI] [PubMed] [Google Scholar]
  6. Cheetham  GMT, Jeruzalmi  D, Steitz  TA. Structural basis for initiation of transcription from an RNA polymerase–promoter complex. Nature  1999;399:80–3. 10.1038/19999. [DOI] [PubMed] [Google Scholar]
  7. Daffern  N, Johansson  KE, Baumer  ZT  et al.  GMMA can stabilize proteins across different functional constraints. J Mol Biol  2024;436:168586. 10.1016/j.jmb.2024.168586. [DOI] [PubMed] [Google Scholar]
  8. Delaney  R, Lamb  KA, Irvin  OM  et al.  Biochem Eng J  2024;215:109590. [Google Scholar]
  9. Dousis  A, Ravichandran  K, Hobert  EM  et al.  An engineered T7 RNA polymerase that produces mRNA free of immunostimulatory byproducts. Nat Biotechnol  2023;41:560–8. 10.1038/s41587-022-01525-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Durniak  KJ, Bailey  S, Steitz  TA. The structure of a transcribing T7 RNA polymerase in transition from initiation to elongation. Science  2008;322:553–7. 10.1126/science.1163433. [DOI] [PMC free article] [PubMed] [Google Scholar]
  11. Edelhoch  H. Spectroscopic determination of tryptophan and tyrosine in proteins*. Biochemistry  1967;6:1948–54. 10.1021/bi00859a010. [DOI] [PubMed] [Google Scholar]
  12. Gibson  DG, Young  L, Chuang  RY  et al.  Enzymatic assembly of DNA molecules up to several hundred kilobases. Nat Methods  2009;6:343–5. 10.1038/nmeth.1318. [DOI] [PubMed] [Google Scholar]
  13. Goldenzweig  A, Goldsmith  M, Hill  SE  et al.  Automated structure- and sequence-based Design of Proteins for high bacterial expression and stability. Mol Cell  2016;63:337–46. 10.1016/j.molcel.2016.06.012. [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Griko  Y, Sreerama  N, Osumi-Davis  P  et al.  Thermal and urea‐induced unfolding in T7 RNA polymerase: Calorimetry, circular dichroism and fluorescence study. Protein Sci  2001;10:845–53. 10.1110/ps.39701. [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Ikeda  RA, Ligman  CM, Warshamana  S. T7 promoter contacts essential for promoter activityin vivo. Nucleic Acids Res  1992;20:2517–24. 10.1093/nar/20.10.2517. [DOI] [PMC free article] [PubMed] [Google Scholar]
  16. Jiang  F, Li  M, Dong  J  et al.  A general temperature-guided language model to design proteins of enhanced stability and activity. Sci Adv  2024;10:eadr2641. 10.1126/sciadv.adr2641. [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Jiang  K, Yan  Z, Di Bernardo  M  et al.  Rapid in silico directed evolution by a protein language model with EVOLVEpro. Science  2025;387:eadr6006. 10.1126/science.adr6006. [DOI] [PubMed] [Google Scholar]
  18. Kamens  J. The Addgene repository: An international nonprofit plasmid and data resource. Nucleic Acids Res  2014;43:D1152–7. 10.1093/nar/gku893. [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Kirby  MB, Medina-Cucurella  AV, Baumer  ZT  et al.  Optimization of multi-site nicking mutagenesis for generation of large, user-defined combinatorial libraries. Protein Eng Des Sel  2021;34   10.1093/protein/gzab017. [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Klesmith  JR, Bacik  JP, Wrenbeck  EE  et al.  Trade-offs between enzyme fitness and solubility illuminated by deep mutational scanning. Proc Natl Acad Sci  2017;114:2265–70. 10.1073/pnas.1614437114. [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Liu  Q, Lu  C, Xu  R  et al.  Enhanced high-temperature performance of T7 RNA polymerase by greedy accumulation-directed evolution of thermostability. Syst Microbiol Biomanuf  2025;5:1154–67. 10.1007/s43393-025-00370-3. [DOI] [Google Scholar]
  22. Maghraby  YR, El-Shabasy  RM, Ibrahim  AH  et al.  Enzyme immobilization technologies and industrial applications. ACS Omega  2023;8:5184–96. 10.1021/acsomega.2c07560. [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. McSweeney  MA, Patterson  AT, Loeffler  K  et al.  A modular cell-free protein biosensor platform using split T7 RNA polymerase. Sci Adv  2025;11:eado6280. 10.1126/sciadv.ado6280. [DOI] [PMC free article] [PubMed] [Google Scholar]
  24. Meyer  AJ, Garry  DJ, Hall  B  et al.  Transcription yield of fully 2′-modified RNA can be increased by the addition of thermostabilizing mutations to T7 RNA polymerase mutants. Nucleic Acids Res  2015;43:7480–8. 10.1093/nar/gkv734. [DOI] [PMC free article] [PubMed] [Google Scholar]
  25. Ong  J, Potapov  V, Hung  KC  et al. US11359184B2 Patent 2019. [Google Scholar]
  26. Paige  JS, Wu  KY, Jaffrey  SR. RNA mimics of green fluorescent protein. Science  2011;333:642–6. 10.1126/science.1207339. [DOI] [PMC free article] [PubMed] [Google Scholar]
  27. Protasevich  II, Memelova  LV, Kochetkov  SN  et al.  The studies of cooperative regions in T7 RNA polymerase. FEBS Lett  1994;349:429–32. 10.1016/0014-5793(94)00718-7. [DOI] [PubMed] [Google Scholar]
  28. Rio  DC. Reverse transcription–polymerase chain reaction. Cold Spring Harbor Protocols, 2014:pdb-prot080887. [DOI] [PubMed] [Google Scholar]
  29. Sheldon  RA, Basso  A, Brady  D. New frontiers in enzyme immobilisation: Robust biocatalysts for a circular bio-based economy. Chem Soc Rev  2021;50:5850–62. 10.1039/D1CS00015B. [DOI] [PubMed] [Google Scholar]
  30. Stimple  SD, Smith  MD, Tessier  PM. Directed evolution methods for overcoming trade‐offs between protein activity and stability. AICHE J  2020;66   10.1002/aic.16814. [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. Studier  FW. Protein production by auto-induction in high-density shaking cultures. Protein Expr Purif  2005;41:207–34. 10.1016/j.pep.2005.01.016. [DOI] [PubMed] [Google Scholar]
  32. Sugiyama  A, Nishiya  Y, Kawakami  B  RNA Polymerase Mutants with Increased Thermostability, US8551752B2 2009.
  33. Tahirov  TH, Temiakov  D, Anikin  M  et al.  Structure of a T7 RNA polymerase elongation complex at 2.9 Å resolution. Nature  2002;420:43–50. 10.1038/nature01129. [DOI] [PubMed] [Google Scholar]
  34. Thompson  MP, Peñafiel  I, Cosgrove  SC  et al.  Biocatalysis using immobilized enzymes in continuous flow for the synthesis of fine chemicals. Org Process Res Dev  2019;23:9–18. 10.1021/acs.oprd.8b00305. [DOI] [Google Scholar]
  35. Tsuboyama  K, Dauparas  J, Chen  J  et al.  Mega-scale experimental analysis of protein folding stability in biology and design. Nature  2023;620:434–44. 10.1038/s41586-023-06328-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
  36. Vavitsas  K, Glekas  PD, Hatzinikolaou  DG. Synthetic biology of thermophiles: Taking bioengineering to the extremes?  Appl Microbiol (Basel)  2022;2:165–74. 10.3390/applmicrobiol2010011. [DOI] [Google Scholar]
  37. Weinstein  JJ, Goldenzweig  A, Hoch  S  et al.  PROSS 2: A new server for the design of stable and highly expressed protein variants. Bioinformatics  2021;37:123–5. 10.1093/bioinformatics/btaa1071. [DOI] [PMC free article] [PubMed] [Google Scholar]
  38. Wrenbeck  EE, Bedewitz  MA, Klesmith  JR  et al.  An automated data-driven pipeline for improving heterologous enzyme expression. ACS Synth Biol  2019;8:474–81. 10.1021/acssynbio.8b00486. [DOI] [PMC free article] [PubMed] [Google Scholar]
  39. Wu  MZ, Asahara  H, Tzertzinis  G  et al.  Synthesis of low immunogenicity RNA with high-temperature in vitro transcription. RNA  2020;26:345–60. 10.1261/rna.073858.119. [DOI] [PMC free article] [PubMed] [Google Scholar]
  40. Yin  YW, Steitz  TA. Structural basis for the transition from initiation to elongation transcription in T7 RNA polymerase. Science  2002;298:1387–95. 10.1126/science.1077464. [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

gzag008_SOURCE_DATA_FILE
gzag008_source_data_file.xlsx (882.1KB, xlsx)

Data Availability Statement

All data and novel code used in this paper is available in a Zenodo repository (doi:10.5281/zenodo.17593694). This repository contains:(1.) SOURCE_DATA_FILE.xlsx contains all experimental data presented in this paper. This file also contains all primers, plasmid sequences, and gBlock sequences used for plasmid construction. (2.) The Jupyter notebook ‘Evaluating Tmapp from CD.ipynb’ is a custom script for extracting Tm,app from circular dichroism temperature melts. (3.) PROSS_outputs.xlsx lists the full PROSS outputs and our relative ranking of the mutations for testing in this paper. Plasmid pZB246 encoding T7T+ is available on AddGene (ID: 250903).

Articles from Protein Engineering, Design and Selection are provided here courtesy of Oxford University Press

RESOURCES