mRNA manufacturing with fed-batch in vitro transcription

In vitro transcription

In this study, we chose a recipe based on historical batch IVT conditions (Table 1), unless otherwise mentioned, which produced yields ≥ 4 g/L. We optimized the starting NTP concentrations to maximize the initial rate of mRNA production. The DNA construct used in this study encoded enhanced green fluorescent protein (eGFP), with an encoded poly(A) tail.

Table 1. IVT reaction components

¹ Enzymes were sourced from Aldevron LLC

Batch experiments

We performed the IVT batch reactions at a small scale (0.8 mL) in the BioLector XT Microbioreactor using the MTP-48-FlowerPlate (M2P-MTP-48-BOH1 or M2P-MTP-48-BOH2). We added all reaction components, except for the T7 polymerase, to the wells of the MTP. We pre-heated the plate to 37 °C for 30 min inside the BioLector XT Microbioreactor. We added T7 polymerase to initiate the reaction and incubated the reaction at 37°C at 500 rpm for 2 to 6 h. Finally, we added ethylenediaminetetraacetic acid (EDTA) solution to a final concentration of 80 mM to quench the reaction.

Continuous feeding strategies

We used microfluidic FlowerPlates (M2P-MTP-MF32C-BOH1 or M2P-MTP-MF32C-BOH2) for the fed-batch experiments. These microplates have microfluidic channels in the bottom of the plate which enable the flow of fluids from reservoir wells (upper two rows of the MTP, row A and B) of a column into the cultivation wells of the same column (rows C to F) as controlled by pumps on the microfluidic chip that are actuated by compressed air (Fig 1).

Fig 1. (A) BioLector XT Microbioreactor with valve control unit. (B) Microfluidic microplates used in the BioLector XT Microbioreactor. The plate contains 32 wells for reactions and 16 reservoir wells for feed reagents. Adapted from Beckman Coulter Life Sciences. 2025.

For fed-batch reactions, we prepared up to 1.8 mL of feed stock consisting of magnesium acetate (50 mM) and NTPs, and added these to the reservoir wells of the MTP. The NTP concentrations were either equimolar (25 mM of each NTP), proportional to the nucleotide frequency of the RNA, or specifically adjusted to achieve a target consumption efficiency. We programmed the feeding profiles and rates using the BioLection 5 software. The reactions were incubated at 37°C at 500 rpm for 2 to 6 h before quenching by addition of EDTA solution to a final concentration of 80 mM. In experiments examining reaction kinetics, we prepared multiple wells using the same master mix and programmed identical feeding strategies. Each well was set to terminate at a different time point to capture the progression of the reaction. At the designated stopping time, we quenched the reactions by adding EDTA through a syringe, effectively halting transcription and preserving the intermediate state for purification and quantification.

In vitro transcription scale-up

We scaled up one of the fed-batch IVT conditions developed on the BioLector XT Microbioreactor on to the Cytiva ReadyToProcess WAVE™ 25 rocking bioreactor, which supports IVT volumes from 20 mL to 20 L, using either the enzyme micro reactor accessory (Fig 2B) or the standard WAVE™ trays (Fig 2A). In this study, we performed scaled-up reactions at 20 mL with the enzyme micro reactor accessory and 200 mL on the standard WAVE™ tray. In each case, we added the reaction mixture to the bag and preheated to a setpoint temperature of 37°C on the WAVE™ system, before adding T7 to initiate the reaction. The reaction utilized a constant feed of a stock solution prepared with NTP concentrations adjusted proportionally to the mRNA sequence nucleotide frequency. The feed was supplied for 6 h at a rate of 2.1 mL/h for the 20 mL reaction and 21 mL/h for the 200 mL reaction. We performed the feeding using either a syringe pump for the 20 mL reaction or the integrated WAVE™ 25 pumps for the 200 mL scale. During the IVT, we performed sampling where approximately 0.2 or 2 mL of sample was withdrawn and quenched to a final concentration of 80 mM EDTA. After 6 h, we treated the bulk reaction with 3.5 µg/mL DNAse I and 2.5 mM CaCl₂ for 40 min to digest the DNA template. Afterwards we quenched the reaction by addition of EDTA to a final concentration of 80 mM and harvested the cells from the Cellbag™ bioreactor container.

For the 200 mL reaction, we used the 2 L Cellbag™ bioreactor container with an optical pH sensor. We calibrated the sensor with an external probe after preheating was completed and enabled real-time pH tracking as the reaction progressed.

Fig 2. Instruments used for IVT scale-up: (A) ReadyToProcess WAVE™ 25 rocking bioreactor and (B) enzyme micro reactor accessory.

Analytics

Unless otherwise specified, we performed all analytics from the IVT reactions on samples that were purified using lithium chloride (LiCl) precipitation. The mRNA concentration was measured spectrophotometrically and used to determine the yield, which is defined here as the total mass of RNA produced divided by the volume of the reaction (or initial volume in the case of fed-batch). We measured the integrity of RNA samples by capillary electrophoresis (Agilent 4150 TapeStation), defined as the ratio of the main peak area to the total area. Double-stranded RNA (dsRNA) content was measured by J2 immunoblot assay using a 142 basepair (bp) control dsRNA.

To determine protein expression (potency), purified mRNA samples were post-transcriptionally capped with vaccinia capping enzyme and 2’-O-methyltransferase. The samples were purified a second time with LiCl precipitation, and then transfected to HEK 293 cells using jetMESSENGER transfection reagent. The percentage of GFP positivie cells was determined 24 h after transfection using flow cytometry.

IVT initial conditions DoE

We conducted a small design of experiments (DoE) study to identify optimal magnesium (Mg²⁺) and NTP concentrations that would give a high initial rate of RNA production. In this case, we defined the initial rate as the RNA yield over the first 30 min of transcription. We used a central composite design, which consisted of nine different conditions, featuring one center point. The experiment was conducted in two blocks with the center point repeated in each block to serve as a control. The individual NTP and Mg²⁺ concentrations tested were 6, 8, and 10 mM and 35, 52.5, and 70 mM, respectively.

The DoE analysis revealed important trends in how Mg²⁺ and NTP concentrations influence the initial rate of RNA production (Fig 3). Interestingly, increasing the starting concentration of NTPs led to a noticeable decline in the initial transcription rate. This result suggests that excessively high NTP levels may inhibit the reaction, potentially due to excess salt addition (a component of NTP solutions) which disrupts polymerase-DNA binding (8). In contrast, a parabolic relationship was observed with Mg²⁺ concentration. This indicates that there is an optimal Mg²⁺ level that maximizes RNA synthesis, which depends on the NTP concentration. Furthermore, we found that a Mg²⁺ to total NTP ratio of less than one resulted in poor RNA production rates, underscoring the importance of a balanced Mg²⁺:NTP ratio for efficient transcription.

These initial findings suggest that implementing a fed-batch IVT strategy is a promising method to increase RNA production. The gradual addition of NTPs and Mg²⁺ mitigates against low transcription rates associated with high initial NTP concentrations while also preventing NTP depletion as the reaction proceeds.

Fig 3. DoE study investigation on initial rates of mRNA synthesis. Contour plot of model (R²: 0.98, p < 0.05) generated from DoE study showing impact of Mg²⁺ and NTP on initial reaction rates (mg of RNA per mL of reaction volume per h) in the first 30 min, determined by the absorbance at 260 nm (A₂₆₀) of LiCl purified samples. Significant model terms: Mg²⁺, Mg²⁺*NTP, and Mg²⁺*Mg²⁺. Black dots indicate experimental test conditions. NTP concentration is equimolar for each nucleotide, with total NTP concentration being 4× the amount shown. 

IVT feasibility study in BioLector XT Microbioreactor: Comparison to manual experiments in the incubator

Before developing a feeding strategy for fed-batch IVTs in the BioLector XT Microbioreactor, we conducted a series of tests to compare the performance of the system to a thermal shaking incubator, a traditional equipment for batch IVT reactions. We prepared a single IVT master mix, and placed three 0.8 mL aliquots in both the BioLector XT Microbioreactor and the shaking incubator. The results, as shown in Table 2, demonstrate no significant difference (p > 0.05) in mRNA yield and integrity with low dsRNA production (≤ 4%). In vitro potency testing of the mRNA from one of the replicates in each apparatus also showed comparable performance in terms of GFP positive cells and mean fluorescence intensity (Table 2).

Table 2. Comparison of mRNA yield and quality between a shaking incubator and the BioLector XT Microbioreactor

¹Percentage of GFP positive cells at 1 µg/mL mRNA

To test the variability within a BioLector microplate, we prepared a total of six different batch IVT conditions (Table 3). These conditions varied in Mg²⁺, NTP, and DNA concentrations and were selected based on historical data to encompass a range of yields (0 to 6 mg/mL). We prepared a large master mix for each IVT condition and split it into three wells distributed randomly across half of a BioLector microplate. The corresponding yields of all six conditions fell within the expected results and showed a coefficient of variation (CV) between 0.8% and 2.7%, excluding low yielding conditions (Fig 4A). For comparison, the IVT reactions performed in the shaking incubator had a variability of 1.7% CV.

Table 3. Batch IVT conditions tested in BioLector XT Microbioreactor

In addition, we monitored the pH for all conditions (Figure 4B). The largest change in pH was observed for the highest yielding condition (condition 5) with a drop of -0.40 over 2 h, while the two lowest yield conditions (condition 2 and 3) displayed the smallest pH change of +0.03 and -0.08, respectively.

These results show that the BioLector XT Microbioreactor is a reliable and robust system for performing IVT, offering the added benefits of real-time monitoring and high-throughput capability without compromising product quality or yield.

Fig 4. mRNA synthesis in the BioLector XT Microbioreactor. (A) mRNA yield for five different batch IVT conditions (Table 3) performed in the BioLector XT Microbioreactor (n = 3 wells). (B) pH traces (n = 3 wells). Error bars indicate standard deviation.

Fed-batch IVT development

Following the successful demonstration of the BioLector XT Microbioreactor for IVT, we leveraged the platform to evaluate different feeding strategies aimed at improving mRNA yield and reaction efficiency. Based on the results of the initial DoE study, we used starting concentrations of 52.5 mM Mg²⁺ and 6 mM NTP, which were the conditions resulting in the greatest reaction rate. The goal was to feed a mixture of Mg²⁺ and NTP feed stock that replenished free NTP and maintained the Mg²⁺ concentration while avoiding inhibition of transcription.

A feed solution comprising 50 mM Mg²⁺ and an equimolar NTP solution was delivered at various rates using customizable feeding profiles such as constant, linear, and exponential decay from the BioLector software (Fig 5).

mRNA yields from the fed-batch reactions ranged from 13.7 to 15.5 mg/mL of the initial IVT, which marked a considerable improvement over yields from previous batch reactions.

Fig 5. Comparison of IVT yields under different feeding strategies. (A) mRNA yields (mg per initial reaction volume) from fed-batch IVT reactions using various feeding profiles. The highest yield (15.5 mg/mL) was achieved with a constant feed of 85 µL/h, while other strategies including 0.5 h delayed start, 5 h feed duration, exponential decay, increased feed rate (100 µL/h), and linear increase produced similar or slightly lower yields. (B) Line graphs illustrate the feed rate profiles over time for each condition. Error bars represent standard deviation.

The highest yield (15.5 mg/mL initial IVT) was achieved with a constant feed of 85 μL/h. Increasing the feed rate to 100 µL/h did not result in a greater yield (15.2 mg/mL initial IVT). This suggests that NTPs are not the rate-limiting component and may be being supplied at a rate exceeding the mRNA production capacity which could inhibit reaction progress due to build-up of excess salts from NTP addition. Based on estimates calculated using the feed volume, mRNA sequence composition, and final yield, only approximately 50% to 55% of the NTPs supplied in the 85 µL/h condition were incorporated into the mRNA.

To reduce total feed volume and thus the excess NTPs, we explored modifications to feed timing. Introducing a 0.5 h delay to allow consumption of initial NTPs or stopping the feed 1 h early to promote full utilization of residual NTPs, produced yields of 14.7 and 15.0 mg/mL initial IVT, respectively, while reducing feed volume by 8% and 17% compared to the 85 µL/h condition. We also explored an exponential decay of feed rate which resulted in a yield of 14.3 mg/mL initial IVT. In this case, the lower yield may reflect near-complete adenosine triphosphate (ATP) utilization (estimate based on mRNA yield, sequence identity, and feed rates). The lowest yield we observed (13.7 mg/mL initial IVT) resulted from a linear increase in the feed rate which is likely poorly suited to IVT due to the known decline in RNA production rate over time (data not shown). Overall, these findings reinforce the importance of an appropriate feeding strategy selection for IVT optimization.

It is important to note that while we tested some conditions in duplicate or triplicate, most were evaluated with only a single replicate, which limits the ability to assess variability or statistical significance. As such, the observed differences in RNA yield should be interpreted as preliminary trends.

In all fed-batch conditions tested, the estimated overall NTP incorporation was below 60%. This inefficiency is largely driven by a feedstock of equimolar nucleotides coupled with imbalanced nucleotide consumption rates, with ATP being utilized at the highest proportion due to the poly (A) tail in the mRNA construct, while the other nucleotides are incorporated at relatively lower rates. In the case of the 85 μL/h constant feed condition, we estimate that approximately 85% of the total ATP is consumed whereas the incorporation efficiencies for guanosine triphosphate (GTP), cytidine triphosphate (CTP), and (uridine triphosphate) UTP are all below 55%, with UTP being the lowest at just 35%.

To further improve nucleotide utilization, we evaluated six feeding strategies and two different feed stock compositions for their impact on NTP incorporation efficiency (Fig 6). For all conditions, an overall target of 85% incorporation efficiency was desired particularly by enhancing the usage of GTP, CTP, and UTP. NTP feed recipe 1 utilized a feed stock concentration that ensured that the total of each NTP supplied (i.e., initial + fed at 85 μL/h for 6 h) was directly proportional to the mRNA sequence frequency. As shown in Fig 6B , 83% efficiency was achieved with this strategy. Stopping the feed 1 h early or implementing an exponential decaying feed rate both resulted in lower yields (12.9 mg/mL initial IVT) without observed improvements in efficiency.

NTP feed recipe 2 utilized a simple proportional composition where the NTP concentrations were directly proportional to the mRNA sequence frequency. Attempting the same exponential decay, as used for feed recipe 1, resulted in a yield of 15.2 mg/mL initial IVT but an incorporation efficiency of 66% due to overfeeding (exponential decay). Two other exponentially decaying feed strategies were attempted: the first started at higher feeding and declined more sharply while the second combined that strategy with a 1 h initial delay. The goal was to match the feed rate as closely as possible to the consumption rate of NTP based on prior reaction kinetics data. Since the mRNA synthesis rate was greatest at the start of the reaction, a high initial feed rate was required that decayed through the run. Initial feed rates greater than the system limit were required to accomplish this. Thus, a constant feed of 13.82 μL/h was supplied from the second reservoir for the first 2 h before shutting off, in addition to an exponential decay feed for the 6 h of duration. These strategies showed final yields of 15.1 and 15.4 mg/mL initial IVT (two-stage exponential decay and two-stage exponential decay with 1 h delay) and improved efficiencies to 70% and 75%, respectively.

Fig 6. Impact of feeding strategies on NTP incorporation efficiency. (A) Comparison of mRNA yields (mg per initial reaction volume) from IVT reactions using six different NTP feeding strategies, grouped by two feed recipes (NTP feed recipes 1 and 2). The first three conditions (recipe 1) account for both starting and delivered NTP amounts, while the latter three (recipe 2) are based on sequence-proportional feed rates without adjusting for initial NTP concentrations. All IVTs started with 6 mM of each NTP. (B) Line graphs illustrate the feed rate profiles over time for each strategy. NTP incorporation efficiencies ranged from 66% to 84% and were estimated based on the mRNA yield, total NTP fed, and mRNA sequence.

Together, these experiments demonstrate how both feed composition and delivery strategy play critical roles in optimizing NTP utilization and mRNA yield. By tailoring feed recipes to match nucleotide demand and adjusting feed timing and profiles based on reaction kinetics, it is possible to improve incorporation efficiency while maintaining high yields.

Scale-up experiments

To evaluate the scalability of the fed-batch IVT process, the protocol that we developed at the 0.8 mL scale in the BioLector XT Microbioreactor with a constant feed (85 µL/h) of a sequence proportional solution was transferred to larger volumes using the Cytiva ReadytoProcess WAVE™ 25 rocking bioreactor. We scaled the feed rate proportionally based on initial reaction volumes of 20 and 200 mL. This direct scaling approach maintained the same feed-to-volume ratio across all scales, ensuring consistent reagent delivery kinetics.

Fig 7.Reaction kinetics across scales and instruments in fed-batch IVT. Time-course comparison of mRNA yield in fed-batch IVT reactions performed at three scales: 0.8 mL (BioLector XT Microbioreactor), and 20 mL and 200 mL (ReadyToProcess WAVE™ 25 rocking bioreactor). All conditions used a constant proportional feed strategy.

The reaction kinetics data shown in Figure 7 demonstrates the scalability of the fed-batch approach. mRNA yield increased steadily over time at all three scales, with very similar kinetic profiles. Furthermore, the quality of the mRNA produced (Table 4) using fed-batch methods was comparable for the various scales tested, as measured by mRNA integrity (Fig 8), dsRNA content, and in vitro potency (Fig 9). mRNA quality for the fed-batch process was also equivalent to a shorter 2 h batch IVT performed in the BioLector Microbioreactor.

Table 4. Comparison of mRNA quality from fed-batch IVTs across scales and instruments

¹Percentage of GFP positive cells at 1 µg/mL mRNA

This successful scale-up demonstrates the feasibility of translating small-scale fed-batch IVT processes to manufacturing-relevant volumes using proportional feed principles, providing a foundation for further process intensification and automation.

Fig 8. mRNA integrity measurements. The electropherograms from capillary electrophoresis (CE) show the analysis of produced mRNA. (A) 0.8 mL batch reaction in BioLector XT Microbioreactor, (B) 0.8 mL fed-batch reaction in BioLector XT Microbioreactor, (C) 20 mL fed-batch reaction in the WAVE™ 25 rocking bioreactor, and (D) 200 mL fed-batch reaction in the WAVE™ 25 rocking bioreactor. Percent integrity is the ratio of the peak within the blue boundaries to the total peak area.

Fig 9. In vitro potency data. Flow cytometry analysis of HEK 293 cells treated with mRNA produced using batch and fed-batch methods in the BioLector XT Microbioreactor and WAVE™ 25 bioreactor from 0.8 mL to 200 mL scales.

pH correlation

We measured the change in reaction pH for a fed-batch condition over time in both the BioLector XT and the WAVE™ 25 systems, using their in-line monitoring capabilities. For a constant feed condition (85 μL/h) with sequence proportional feeds, we observed that the pH declined rapidly over the first 2 h and then at a slower rate over the remainder of the reaction (Fig 10A). Comparing the pH traces of both systems, we observed a similar overall trend and change in pH (-0.76 to -0.86).

To further explore the relationship between pH and mRNA production kinetics, we performed time-course sampling of batch (data not shown) and fed-batch reactions. We observed a linear relationship (R² = 0.9765) between the mRNA crude concentration (concentration of the mRNA in the reaction) and the reaction pH (Fig 10B), in a 200 mL fed-batch run conducted on the WAVE™ 25 bioreactor.

The consistency of pH trajectories and reaction kinetics across scales and instruments further demonstrates the robustness of the fed-batch IVT method and the utility of the BioLector XT Microbioreactor and the WAVE™ 25 rocking bioreactor for scale-up. It also shows the potential for pH to serve as metric for real-time estimation of reaction progress which could leverage the in-line pH monitoring that both the BioLector XT Microbioreactor and the ReadyToProcess WAVE™ 25 rocking bioreactor offer. Furthermore, this correlation opens the possibility of developing custom feeding strategies that dynamically adjust reagent delivery based on pH trends, enabling more precise control over reaction kinetics and yield optimization.

Fig 10. pH profiles across scales and instruments in fed-batch IVT. (A) Time-course comparison of pH profiles from BioLector XT Microbioreactor (n = 2) and WAVE™ 25 rocking bioreactor (n =1) (B) pH and mRNA correlation for a reaction performed in the WAVE™ 25 rocking bioreactor. A linear regression is shown between pH and crude concentration. Feed condition used was a constant proportional feed strategy.

Cytiva: Darius Menezes, Patrick Francis, Gary Pigeau, Vinay Mayya, Emily Soon, Sree Gayathri Talluri, Miranda Fujisawa, Aman Kumar, Birgit Blank, and David Sokolowski.

Beckman Coulter Life Sciences: James Prescott, Rick Luedke, Anna Kress, and Maria Savino

1. Kis Z, Kontoravdi C, Shattock R, Shah N. Resources, production scales and time required for producing RNA vaccines for the global pandemic demand. Vaccines (Basel). 2020;9(1):3. doi:10.3390/vaccines9010003
2. Kis Z, Tak K, Ibrahim D et al. Pandemic-response adenoviral vector and RNA vaccine manufacturing. NPJ Vaccines 2022;7(1):29. doi:10.1038/s41541-022-00447-3
3. Kis Z. Process-cost modelling for producing 100 million COVID-19 mRNA vaccine doses per year at injectable medicines manufacturing sites. Médecins Sans Frontières. msfaccess.org/sites/default/files/2021-09/COVID19_TechBrief_Process_cost_modelling_ENG.pdf. Published 2021.
4. Daniel S, Kis Z, Kontoravdi C, Shah N. Quality by design for enabling RNA platform production processes. Trends Biotechnol. 2022;40(10):1213-1228. doi: 10.1016/j.tibtech.2022.03.012
5. Youssef M, Hitti C, Fulber JPC, Kamen AA. Enabling mRNA therapeutics: current landscape and challenges in manufacturing. Biomolecules. 2023;13(10):1497-1524. doi: 10.3390/biom13101497
6. Rosa SS, Nunes D, Antunes L, Prazeres DMF, Marques MPC, Azevedo AM. Maximizing mRNA vaccine production with Bayesian optimization. Biotechnol Bioeng. 2022;119(11):3127-3139. doi: 10.1002/bit.28216
7. Samnuan K, Blakney AK, McKay PF and Shattock RJ. Design-of-experiments in vitro transcription yield optimization of self-amplifying RNA. F1000Research. 2022;11:333. doi: 10.12688/f1000research.75677.1
8. Stover NM, Ahmadi S, Rosenfeld J, Destro F, Myerson AS, Braatz RD. Model-Based Optimization of Fed-Batch In Vitro Transcription. Chembiochem. 2025;26(21):e202500485. doi:10.1002/cbic.202500485

This application note is for demonstration only, and is not validated by Beckman Coulter. Beckman Coulter makes no warranties express or implied with respect to this protocol, including warranties of fitness for a particular purpose or merchantability or that the method is non-infringing. Your use of the method is solely at your own risk, without recourse to Beckman Coulter. BioLector XT not for use in diagnostic and therapeutic procedures. © 2025 Beckman Coulter, Inc. All rights reserved. Beckman Coulter, the Stylized Logo and Beckman Coulter product and service marks mentioned herein are trademarks or registered trademarks of Beckman Coulter, Inc. in the United States and other countries.