Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

# Quality by design modelling to support rapid RNA vaccine production against emerging infectious diseases

## Abstract

Rapid-response vaccine production platform technologies, including RNA vaccines, are being developed to combat viral epidemics and pandemics. A key enabler of rapid response is having quality-oriented disease-agnostic manufacturing protocols ready ahead of outbreaks. We are the first to apply the Quality by Design (QbD) framework to enhance rapid-response RNA vaccine manufacturing against known and future viral pathogens. This QbD framework aims to support the development and consistent production of safe and efficacious RNA vaccines, integrating a novel qualitative methodology and a quantitative bioprocess model. The qualitative methodology identifies and assesses the direction, magnitude and shape of the impact of critical process parameters (CPPs) on critical quality attributes (CQAs). The mechanistic bioprocess model quantifies and maps the effect of four CPPs on the CQA of effective yield of RNA drug substance. Consequently, the first design space of an RNA vaccine synthesis bioreactor is obtained. The cost-yield optimization together with the probabilistic design space contribute towards automation of rapid-response, high-quality RNA vaccine production.

## Introduction

The outbreak and spread of viral diseases, such as the COVID-19 pandemic caused by the SARS-CoV-2 virus, the 2015–2016 Zika virus epidemic in Brazil and American continents, the re-emerging Nipah outbreaks in South and Southeast Asia, and the 2013–2016 Ebola virus epidemic in West Africa, pose tremendous healthcare and economic challenges1,2,3. Vaccines are highly effective for stopping epidemics and pandemics. However, the development of vaccines using conventional production methods is becoming too slow to effectively respond to new viral outbreaks in the 21st century4, the frequency of which is predicted to increase3.

To address this pressing need, rapid-response vaccine production platform technologies are being deployed, such as the messenger RNA (mRNA) and self-amplifying RNA (saRNA) platforms, herein collectively referred to as RNA vaccine platforms. The mRNA and saRNA vaccine production process involves cell-free DNA-templated RNA synthesis based on the in vitro transcription (IVT) reaction catalysed by the T7 RNA polymerase enzyme (T7RNAP)5,6. The RNA (both mRNA and saRNA) drug substance is purified using tangential flow filtration (TFF) and chromatography techniques, such as ion-exchange or multimodal chromatography4,7. Then the RNA drug substance is formulated into lipid nanoparticles and filled into vials or other containers4,7. A process diagram showing RNA vaccine drug substance and drug product manufacturing are shown in Supplementary Fig. 1.

RNA vaccines involve rapid development and production timelines because the production platform is agnostic to the disease target as RNA sequences translating into any vaccine protein antigen can be produced using the same production process8. The only component in this production process that needs to be changed is the template DNA based on which the RNA is enzymatically synthesised. The rest of the materials, equipment, consumables, unit operations, formulation components, fill-to-finish processes as well as quality control and quality assurance methods remain unchanged when switching to the production of a new RNA sequence encoding for a new vaccine antigen. This is possible because the RNA vaccine manufacturing process produces only the genetic instructions for expressing an antigen in human cells, and not the actual antigen. Using this technology, candidate vaccines can be produced against any known or currently unknown future pathogens. For example, mRNA and saRNA vaccine candidates against COVID-19 have been recently produced with an unprecedented speed: in 2 weeks after obtaining the genetic sequence information of the antigen9,10. The mRNA vaccines developed by BioNTech and Moderna gained emergency use authorisation against Covid-19 at record speed, despite the RNA vaccine platform being a new technology that had not been approved by regulatory authorities in the past.

The development of process monitoring and quality assurance approaches remains a key challenge for quickly and cost-effectively ensuring that the drug substance is produced with consistently high quality. This should be explored and developed prior to the production of a particular product, ideally in a disease- and product-agnostic manner to complement the flexible manufacturing platform of vaccine candidates against a wide range of pathogens. The quality by design (QbD) framework has been used to aid the regulatory approval and production of small molecule pharmaceuticals11,12 and monoclonal antibodies13,14 by establishing a design space (DS) in which the production process can be operated to consistently obtain the required quality target product profile. Regarding vaccines, some are currently in development based on QbD frameworks15,16. However, to the knowledge of the authors, there are currently no vaccines approved by regulatory authorities based on a full QbD filing. The QbD framework consists of two key steps: (1) a risk assessment based on the identification of product critical quality attributes (CQAs) and critical process parameters (CPPs) and (2) definition of the DS in the CPPs space which is obtained by defining mathematical relationships between CPPs and CQAs. For the first step, quality attributes are commonly ranked based on their impact and uncertainty scores for both product safety and efficacy, obtaining this way a severity score based on which the CQAs are identified15. Next, the CPPs are identified by assessing the impact of PP ranges on the identified CQAs, predominantly by using a binary, yes or no, approach based on expert knowledge and product-process understanding15. Alternatively, CQAs and CPPs can also be identified and ranked using fishbone diagrams, cause–effect matrices and failure mode effect analyses17,18,19. However, none of these existing methods is able to capture the direction, magnitude and shape of the impact of the CPPs on the CQAs. Therefore, more advanced approaches are needed to better describe the relationship between CPPs and CQAs in a data-poor environment, which is typical to the early phases of production process development.

To address this need, we developed and implemented a new qualitative QbD methodology that assesses the criticality of PPs considering the direction, magnitude and shape of the CPP–CQA relation. Furthermore, we developed a bioprocess model to map the multi-dimensional DS of RNA synthesis substantially faster and with fewer resources compared to the experimental design of experiments (DoE) protocols. This bioprocess model was built on previously published RNA synthesis kinetics20. Such mechanistic models tend to outperform statistical or data-driven (e.g. machine learning) models in data-poor environments, such as during the early stages of process development. This is the first bioprocess model of an RNA vaccine synthesis bioreactor in support of DS identification and optimisation. The proposed qualitative QbD methodology which maps the direction, magnitude and shape of the impact CPP–CQA relation together with the bioprocess model forms the QbD framework. Overall, the framework is to become universally applicable to mRNA and saRNA vaccine manufacturing using wild-type nucleotide triphosphate NTPs and is independent of the viral infectious disease indication, because both the RNA vaccine manufacturing process and the QbD framework can be applied to produce any antigen-encoding RNA sequence4,5. The QbD framework applied to the RNA platform further supports upstream process optimisation during both development and manufacturing and is anticipated to expedite the regulatory approval process by providing a form of “pre-qualification” by re-using and processing disease agnostic-prior knowledge4.

## Results

### QbD framework

The mRNA and saRNA and their intrinsic quality features are created during the in vitro transcription (IVT) reaction, therefore the QbD framework, consisting of a qualitative methodology and a quantitative bioprocess model, has been applied to this unit operation. As shown in Fig. 1, the QbD framework development cycle starts with patient need identification and quality target product profile definition. This is followed by CQA and CPP definition, CQA–CPP relation, DS and normal operating range (NOR) definition and, finally, production process automation and control using model predictive control (digital twins).

### CQAs and CPP identification

In the third step, the qualitative methodology is used to identify and rank the CQAs of the mRNA and saRNA vaccine, as shown in Fig. 1. The listing and ranking of CQAs are shown in Supplementary Table 1. The four CQAs identified were: RNA yield, sequence integrity, sequence identity and 5′ capping efficiency. CPP identification and CPP–CQA relation were then established using a novel qualitative ranking methodology, as shown in Table 1. This considers the direction, magnitude and shape of the impact of the CPPs on the CQAs, as described in the “Methods” section.

### Bioprocess model development

The four CQAs from Supplementary Table 1 were grouped into one output, termed effective RNA yield. The 5′ capping efficiency CQA was not modelled individually because the commercially available 5′ cap analogue, CleanCap (TriLink Biotechnologies, San Diego, CA, USA) yields 5′ capping efficiencies of ≈95% which is sufficient for the expression of the vaccine antigen from the RNA transcript in human cells21,22,23,24,25. The bioprocess model involves bi-substrate kinetic formulae for the transcription reaction, adapted from a previously published multiphysics kinetic model20, to compute the RNA transcription yield. Given the prior knowledge that RNA degrades in alkaline as well as acidic environments26, and that high Mg2+ concentration favours RNA degradation27, RNA degradation rate was modelled as a series of three power laws, each first order in RNA and first-order in either proton, hydroxy or Mg2+ concentration. Four CPPs identified in Table 1 were included in the model: initial total solution wild-type nucleotide triphosphate (NTP) and Mg concentrations, T7RNAP concentration, and reaction time. There is a clear distinction in notation between the use of Mg2+ and Mg. Mg2+ is used to refer to free solution magnesium, while Mg is used when referring to total magnesium concentration in free solution together with magnesium in complexes, often in the context of initial experiment conditions. The remaining nine CPPs were not considered in this RNA synthesis bioreactor model because these CPPs can be well controlled in commercially available bioreactor setups implemented in facilities following cGMP guidelines.

The model parameters were then fitted to a subset of 51 experimental samples from a statistical DoE dataset obtained from lab-scale saRNA synthesis experiments using wild-type NTPs28. This dataset includes NTP and T7RNAP screening experiments, and more thorough analysis of RNA yield surface response on Mg concentration. 33 samples correspond to the RNA yield at 0.04 M NTP and 1 × 10−8 M of T7RNAP vs. 11 concentrations of Mg ranging from 0.025 to 0.125 M after 2, 4 and 6 h (circles in Fig. 2). Twelve samples correspond to the RNA yield at 0.04 M NTP and 0.075 M Mg for 1.250 × 10−9, 2.5 × 10−9, 5 × 10−9 and 1 × 10−8 M of T7RNAP after 2, 4 and 6 h (crosses in Fig. 2) and 6 samples correspond to the RNA yield after 2 h at 0.02, 0.04 and 0.08 M NTP at 0.075 and 0.14 M Mg (squares in Fig. 2). For additional information about the experimental data see28. The kinetic equations describing the RNA yield response correspond to Eqs. (1)–(10). The parameter estimation found kapp to be 4.34 $$\frac{{\rm{L}}^2}{{\rm{mol}}\,{\rm{U}}\,{\rm{h}}}$$, K1 5.55 × 105 $$\frac{\rm{L}}{\rm{mol}}$$, K2 1.94 × 105 $$\frac{\rm{L}}{\rm{mol}}$$ and kac 1.20 × 106 $$\frac{\rm{L}}{{\rm{mol}}\,{\rm{h}}}$$, while the effect of kba and kMg was found to be negligible. It has to be noted that multiple parameters such as kapp, K1 and K2 were highly correlated, meaning multiple combinations thereof gave the same dynamic response.

To mitigate this co-correlation, insignificant parameters could be fixed. To this effect, a variance-based global sensitivity analysis was performed around the optimal parameter values as determined by the parameter estimation29,30,31,32,33. This analysis helps to evaluate how uncertainty propagates from the kinetic model parameters to the RNA yield and to quantify how much of the variation in the RNA yield can be attributed to the individual kinetic model parameters29,30,31,32,33. As expected, kba and kMg were found to be negligible, with Sobol indices below 0.001, thus contributing less than 0.1% to the variation in RNA yield computed by the model after 6 h of IVT reaction time, cf. Supplementary Table 3 in the SI document. On the other hand, kapp was found to be most significant as the only parameter driving the reaction forward, explaining over 60% of the model-predicted RNA yield variation after 6 h of IVT reaction time, cf. Supplementary Table 3 in the SI document. Higher values of these Sobol indices, which are ANOVA-decomposed variance contributions, indicate stronger dependence of the variation in the RNA yield on the respective kinetic model parameters29,30,31,32,33. Thus, the significance of parameters can be ranked in this decreasing order kapp, K1, K2, kac with Sobol indices of 0.61, 0.30, 0.06 and 0.03, respectively. kapp and K1 together explain over 91% variation in the RNA yield after 6 h of IVT reaction time. The Sobol index table as well as the scatter plots of RNA yield after 6 h plotted in function of kinetic model parameters can be found in Supplementary Table 3 and Supplementary Fig. 2.

Model fit to experimental data were generally good, capturing most of the non-linearities and overall trends with no consistent over- or underestimation bias, as seen in the prediction error plot in Fig. 2. The modelling mean absolute error (MAE) of 0.28 g/L is acceptable given that the standard deviation in experimental data samples can be as high as 0.95 g/L. The average prediction error is expected to decrease in future QbD framework iterations as more data and knowledge become available. However, there is one striking outlier predicting RNA yield to be high at both high Mg and NTP concentrations, while it should be close to zero. This outlier sample, as shown in Fig. 2, gives the RNA yield to be 0.01 g/L after 2 h at starting conditions of 0.14 M Mg, 0.08 M NTP and 1 × 10−8 M T7RNAP, corresponding to the encased dark blue square in Fig. 2. With the Mg concentration fixed at a high 0.14 M, in its current iteration, the model captures the increase in RNA yield from 0.02 to 0.04 M NTP but not the subsequent decrease thereof from 0.04 to 0.08 M NTP, after which the rate of RNA production should be almost zero. The failure of the model to predict this sample point can be explained by model overfitting to the many samples describing Mg dependence compared to the few samples at different NTP concentrations. The addition of other model terms could support a more accurate prediction at high Mg and NTP concentrations but would lead to even worse testing performance through over-parameterisation.

On top of the dataset being skewed towards the dependence on Mg and T7RNAP rather than NTP, many of the RNA yield values are clustered close to zero, c.f. Supplementary Table 4 for the descriptive statistics on the RNA yield dataset. Before performing a quantitative model-based DoE, qualitative suggestions for further experiments can be proposed to increase the statistical significance of the model and to more accurately account for the peak in RNA yield at increasing NTP concentrations. The following experiments would lead to a smoother regression around the experimentally optimal region: measure the RNA yield at each Mg concentration of 0.06, 0.075 and 0.090 M for NTP concentrations of 0.02, 0.03, 0.05 and 0.06 M after 2, 4 and 6 h of IVT reaction time. More useful still might be the inclusion of physical variable measurements other than RNA yield. These could include free solution NTP4− concentration (if analytically distinguishable from NTP in the transcribed RNA chain), solution turbidity due to Mg2PPi precipitating after the formation of PPi as a byproduct and pH. Through such measurements, one can more easily discriminate the relative importance of the physical phenomena contributing to the degradation of RNA. Ultimately, these measurements would also help in determining tighter bounds on kapp as the most significant parameter.

Despite these current limitations, the mechanistic model performs well in comparison to conventional statistical modelling techniques. Multiple linear regression (MLR) using four linear explanatory variables (four coefficients plus a constant) gave an R2 value of 0.398 and an MAE of 0.570 g/L due to its inability to capture non-linearities. Only after including squared terms in both Mg and NTP and their interaction term in the regression (seven coefficients plus a constant), did the fit of the statistical model increase to an R2 value of 0.766 and an MAE of 0.167 g/L, which are comparable to that of the mechanistic model (0.773 and 0.162 g/L, respectively). The summary of the models’ prediction plots and descriptive statistics can be found in Supplementary Fig. 3 and Supplementary Table 4 to Supplementary Table 6.

### Model implementation and DS definition

The current model performed well in the region of interest at medium-to-low Mg and NTP concentrations and hence this model was used to create the first DSs. In conjunction with cost and safety considerations, this leads to first recommendations about the desired operating region and subsequent experimental design. The limitations of the model at high Mg and NTP concentrations will be resolved in future iterations with the use of additional experimental data that will be obtained from optimal experiment designs.

The deterministic DS and concentration-cost-yield plots produced by the model are shown in Fig. 3. Figure 3A produces the deterministic DS after 6 h defined by the remaining three CPPs, with the optimum corresponding to high RNA effective yield shown by the green region. At fixed T7RNAP, the DS also shows that at fixed initial Mg or NTP, the concentration of the other component passes through an optimum. The optimum in Mg at fixed NTP can be seen more clearly at low NTP.

### Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.

## Data availability

Experimental data are available from ref. 28. These data were obtained from saRNA synthesis experiments using wild-type, non-modified UTPs28. The model was calibrated using this data.

## Code availability

The documented code is available online at https://github.com/dv516/RNA-transcription-modelling-and-DS.

## References

1. 1.

Munster, V. J., Koopmans, M., van Doremalen, N., van Riel, D. & de Wit, E. A novel coronavirus emerging in China—key questions for impact assessment. N. Engl. J. Med. https://doi.org/10.1056/NEJMp2000929 (2020).

2. 2.

World Health Organization. “Novel Coronavirus (2019-nCoV): Situation Report—16—Erratum” (2020).

3. 3.

Sands, P. et al. Outbreak Readiness and Business Impact: Protecting Lives and Livelihoods Across the Global Economy (2019).

4. 4.

Kis, Z., Kontoravdi, C., Dey, A. K., Shattock, R. & Shah, N. Rapid development and deployment of high‐volume vaccines for pandemic response. J. Adv. Manuf. Process. 2, e10060 (2020).

5. 5.

Kis, Z., Shattock, R., Shah, N. & Kontoravdi, C. Emerging technologies for low-cost, rapid vaccine manufacture. Biotechnol. J. 14, 1800376 (2019).

6. 6.

Kis, Z., Papathanasiou, M., Calvo-Serrano, R., Kontoravdi, C. & Shah, N. A model-based quantification of the impact of new manufacturing technologies on developing country vaccine supply chain performance: a Kenyan case study. J. Adv. Manuf. Process. 1, e10025 (2019).

7. 7.

Kis, Z., Kontoravdi, C., Shattock, R., Shah, N. Resources, production scales and time required for producing RNA vaccines for the global pandemic demand. Vaccines 9, 1–14 (2021).

8. 8.

Pardi, N., Hogan, M. J., Porter, F. W. & Weissman, D. mRNA vaccines—a new era in vaccinology. Nat. Rev. Drug Discov. 17, 261–279 (2018).

9. 9.

O’Hare, R. & Lynch, P. First Novel COVID-19 Vaccine Candidate Commences Animal Testing. (Univadis, Medscape, 2020).

10. 10.

Le, T. T. et al. The COVID-19 vaccine development landscape. Nat. Rev. Drug Discov. 19, 305–306 (2020).

11. 11.

Yu, L. X. et al. Understanding pharmaceutical quality by design. AAPS J. 16, 771–783 (2014).

12. 12.

Sangshetti, J. N., Deshpande, M., Zaheer, Z., Shinde, D. B. & Arote, R. Quality by design approach: Regulatory need. Arab. J. Chem. 10, S3412–S3425 (2017).

13. 13.

Kelley, B. Quality by design risk assessments supporting approved antibody products. MAbs 8, 1435–1436 (2016).

14. 14.

Cooney, B., Jones, S. D. & Levine, L. Quality by design for monoclonal antibodies, part 1: establishing the foundations for process development. Bioprocess Int. 28–36 (2016).

15. 15.

CMC-Vaccines Working Group, A-Vax: applying quality by design to vaccines (2012).

16. 16.

Cox, M. M. J. & Onraedt, A. Innovations in vaccine development: can regulatory authorities keep up? Expert Rev. Vaccines 11, 1171–1173 (2012).

17. 17.

Schlindwein, W. S. & Gibson, M. eds. Pharmaceutical Quality by Design: A Practical Approach (Wiley-Blackwell, 2018).

18. 18.

Fahmy, R. et al. Quality by design I: application of failure mode effect analysis (FMEA) and Plackett–Burman design of experiments in the identification of “Main Factors” in the Formulation and Process Design Space for Roller-Compacted Ciprofloxacin Hydrochloride Immediat. AAPS PharmSciTech 13, 1243–1254 (2012).

19. 19.

Rathore, A. S., & Winkle, H. Quality by design for biopharmaceuticals. Nat. Biotechnol. 27, 26–34 (2009).

20. 20.

Akama, S., Yamamura, M. & Kigawa, T. A multiphysics model of in vitro transcription coupling enzymatic reaction and precipitation formation. Biophys. J. https://doi.org/10.1016/j.bpj.2011.12.014 (2012).

21. 21.

TriLink BioTechnologies, “CleanCap Reagent AG for Co-transcriptional Capping of mRNA” (2020) https:/doi.org/Catalog No. N-7113.

22. 22.

TriLink BioTechnologies, “CleanCap Reagent AU for Self-Amplifying mRNA” (2020) https:/doi.org/Catalog No. N-7114.

23. 23.

Wadhwa, A., Aljabbari, A., Lokras, A., Foged, C., & Thakur, A. Opportunities and challenges in the delivery of mRNA-based vaccines. Pharmaceutics 12, 1–27 (2020).

24. 24.

Vaidyanathan, S. et al. Uridine depletion and chemical modification increase Cas9 mRNA activity and reduce immunogenicity without HPLC purification. Mol. Ther. 12, 530–542 (2018).

25. 25.

26. 26.

Bernhardt, H. S. & Tate, W. P. Primordial soup or vinaigrette: did the RNA world evolve at acidic pH? Biol. Direct 7, 1–12 (2012).

27. 27.

Mikko Oivanen, Satu Kuusela & Lönnberg, H. Kinetics and mechanisms for the cleavage and isomerization of the phosphodiester bonds of RNA by brønsted acids and bases. https://doi.org/10.1021/CR960425X (1998).

28. 28.

Samnuan, K., Blakney, A. K., McKay, P. F. & Shattock, R. J. Design-of-experiments in vitro transcription yield optimization of self-amplifying RNA. bioRxiv 2021.01.08.425833 (2021).

29. 29.

Sobol’, I. M. On sensitivity estimation for nonlinear mathematical models. Mat. Model. 2, 112–118 (1990).

30. 30.

Sobol’, I. M. Sensitivity estimates for nonlinear mathematical models. Math. Model. Comput. Exp. https:/doi.org/1061-7590/93/04407-008 (1993).

31. 31.

Sobol’, I. M., Asotsky, D., Kreinin, A. & Kucherenko, S. Construction and comparison of high-dimensional sobol’ generators. Wilmott 2011, 64–79 (2011).

32. 32.

Bratley, P. & Fox, B. L. Algorithm 659: implementing sobol’s quasirandom sequence generator. ACM Trans. Math. Softw. 14, 88–100 (1988).

33. 33.

Kucherenko, S. SobolHDMR: a general-purpose modeling software. Methods Mol. Biol. 1073, 191–224 (2013).

34. 34.

McKay, P. F. et al. Self-amplifying RNA SARS-CoV-2 lipid nanoparticle vaccine candidate induces high neutralizing antibody titers in mice. Nat. Commun. 11, 3523 (2020).

35. 35.

Blakney, A. K., McKay, P. F., Yus, B. I., Aldon, Y. & Shattock, R. J. Inside out: optimization of lipid nanoparticle formulations for exterior complexation and in vivo delivery of saRNA. Gene Ther. 26, 363–372 (2019).

36. 36.

Ye, T., Zhong, Z., García-Sastre, A., Schotsaert, M. & De Geest, B. G. Current status of COVID-19 (pre)clinical vaccine. Dev. Angew. Chem. Int. Ed. 59, 18885–18897 (2020).

37. 37.

Kremsner, P. et al. Phase 1 assessment of the safety and immunogenicity of an mRNA-lipid nanoparticle vaccine candidate against SARS-CoV-2 in human volunteers. MEDRXIV 2020.11.09.20228551 (2020).

38. 38.

Walsh, E. E. et al. Safety and immunogenicity of two RNA-based covid-19 vaccine candidates. N. Engl. J. Med. https://doi.org/10.1056/NEJMoa2027906 (2020).

39. 39.

Sahin, U. et al. COVID-19 vaccine BNT162b1 elicits human antibody and TH1 T cell responses. Nature 586, 594–599 (2020).

40. 40.

Jackson, L. A. et al. An mRNA vaccine against SARS-CoV-2—preliminary report. N. Engl. J. Med. 383, 1920–1931 (2020).

41. 41.

Fletcher, J. Clinical trial to assess the safety of a coronavirus vaccine in healthy men and women. ISRCTN Registry (2020).

42. 42.

Asprey, S. P., Macchietto, S. & Pantelides, C. C. Robust optimal designs for dynamic experiments. IFAC Proc.33, 845–850 (2000).

43. 43.

Asprey, S. P. & Macchietto, S. Statistical tools for optimal dynamic model building. Comput. Chem. Eng. 24, 1261–1267 (2000).

44. 44.

Thomen, P. et al. T7 RNA polymerase studied by force measurements varying cofactor concentration. Biophys. J. 95, 2423–2433 (2008).

45. 45.

Fjelstrup, S. et al. The effects of dithiothreitol on DNA. Sensors 17, 1201 (2017).

46. 46.

Frugier, M., Florentz, C., Hosseini, M. W., Lehn, J. Marie & Giegé, R. Synthetic polyamines stimulate in vitro transcription by T7 RNA polymerase. Nucleic Acids Res. 22, 2784–2790 (1994).

47. 47.

Garrett, R. H. & Grisham, C. M. Biochemistry (Cengage Learning, 2008).

48. 48.

Li, Y., & Breaker, R. R. Kinetics of RNA degradation by specific base catalysis of transesterification involving the 2′-hydroxyl group https://doi.org/10.1021/ja990592p (1999).

49. 49.

ICH Expert Working Group, “ICH harmonised tripartite guideline on pharmaceutical development Q8 (R2)” (2009).

50. 50.

ICH Expert Working Group, “ICH harmonised tripartite guideline on Good Manufacturing Practice Guide for Active Pharmaceutical Ingredients Q7” (2000).

51. 51.

Garrett, R. H. & Grisham, C. M. Biochemistry (Brooks/Cole, Cengage Learning, 2010).

52. 52.

Li, Y. & Breaker, R. R. Kinetics of RNA degradation by specific base catalysis of transesterification involving the 2‘-hydroxyl group. J. Am. Chem. Soc. 121, 5364–5372 (1999).

53. 53.

Oivanen, M., Kuusela, S. & Lönnberg, H. Kinetics and mechanisms for the cleavage and isomerization of the phosphodiester bonds of RNA by brønsted acids and bases. Chem. Rev. 98, 961–990 (1998).

## Acknowledgements

This research is funded by the Department of Health and Social Care using UK Aid funding and is managed by the Engineering and Physical Sciences Research Council (EPSRC, grant No. EP/R013764/1). The views expressed in this publication are those of the author(s) and not necessarily those of the Department of Health and Social Care. Funding from UK Research and Innovation (UKRI) via EPSRC grant number EP/V01479X/1 on COVID-19/SARS-CoV-2 vaccine manufacturing and supply chain optimisation is thankfully acknowledged.

## Author information

Authors

### Contributions

D.vdB., C.F.B., Z.K. and N.S. conceived and designed the study. D.vdB. created the model, wrote the python code and performed the simulations. Z.K. contributed to writing the python code. D.vdB. carried out the sensitivity analysis and the statistical modelling. D.vdB., C.F.B., Z.K., C.K. and N.S. evaluated the modelling results and provided feedback. D.vdB. and Z.K. prepared the figures. K.S., A.K.B. and R.S. provided the experimental data and participated in the discussions. D.vdB. created the GitHub repository for the project. D.vdB., C.F.B. and Z.K. wrote the paper. C.K., N.S., A.K.B. and R.S. reviewed the paper and provided feedback. Z.K. and N.S. supervised the project. C.K., N.S. and R.S. provided the grant funding.

### Corresponding author

Correspondence to Nilay Shah.

## Ethics declarations

### Competing interests

The authors declare no competing interests.

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

## Rights and permissions

Reprints and Permissions

van de Berg, D., Kis, Z., Behmer, C.F. et al. Quality by design modelling to support rapid RNA vaccine production against emerging infectious diseases. npj Vaccines 6, 65 (2021). https://doi.org/10.1038/s41541-021-00322-7

• Accepted:

• Published: