Sterilizing-grade filtration of mRNA-encapsulated lipid nanoparticles (mRNA-LNP)

Membrane filters

Table 1. Cytiva membrane filters tested

Table 2. Competitor filters tested

Constant pressure (V_MAX) filterability testing

We performed constant pressure filtration studies to measure the flow decay of the process fluid through the filter membranes over time. Using the VMAX model, we determined the maximum throughput to allow scale-up calculations for larger volume filtrations.

We calculated V_MAX (the maximum volume that can be filtered at 100% plugging of the membrane) and Q₀ (initial flux) using equation 1 (Table 3), and determined the slope, y-intercept, and coefficient of determination (R²) via the least squares-fit analyses of the data. The model was only applied when the R² for the experimental data was larger than 0.95.

Table 3. Equations to determine flux and throughput

To introduce a safety margin for the scale-up, we calculated the filtrate volume at 90% (V₉₀), 75% (V₇₅), and 50% (V₅₀) flow decay. We used equation 2 to calculate V₇₅, and adapted the equation as needed to calculate V₅₀ and V₉₀ (Table 3). The simplified formula is shown in equation 3.

Material preparation

We thawed 0.12 g/L mRNA-LNPs, stored in 50 mL bottles, from -80°C to room temperature on the bench before use each day. The thawed bottles were then swirled to mix the contents before use.

Material sanitization

We sterilized all membrane filter and encapsulated discs by autoclaving at either 121°C for 60 min or 132°C for 6 min, prior to use. To make sure that the test preparation and testing conditions did not impact the membrane filter or encapsulated disc integrity, we subjected all filter and encapsulated discs to a pre-use installation test at > 50 psig and a 70% isopropyl alcohol (IPA) post-use bubble point (BP) integrity test. All other equipment was cleaned before use by soaking in 0.25 M NaOH, and neutralized by rinsing with nuclease-free RO grade water (Millipore Milli-Q system) until the rinse water reached pH 7. After sanitization, the equipment was allowed to dry in a biological safety cabinet level 2 (BSL-2), before assembly (Fig 2). All equipment was resanitized and neutralized after testing each membrane type before reuse.

Filter installation and testing

We placed the membrane filter discs into a 25 mm diameter polypropylene filter disc holder (Advantec, part no. 43303010) and the filters sealed in place. The holder was then connected to the reservoir (C3EP1 Novasip™ housing) (Fig 2) and we added approximately 50 mL of nuclease-free RO grade water to the reservoir. All luer valves were in the closed position at the start of this procedure. We pressurized the reservoir to 5 to 10 psig (measured using PendoTECH pressure transducers P/N MFLX19406-31 and displayed using a PMAT3P display) using compressed air. We opened the 3-way luer valve to allow water to flow to the 1-way luer valve to vent air from the system. Once vented, we opened the 3-way luer valve to allow water to flush through the membrane disc holder to wet the membrane. Once wetted, we pressurized each test membrane to > 50 psig and held for 1 min to make sure that there were no gross leaks in the filter assembly or the filter membrane. Any filter discs that showed a loss of pressure or bubbles emanating from the downstream side of the filter holder or device were replaced.

We depressurized the assembly after the leak test and removed any surplus water via the 1-way luer valve. Cryo-buffer was then added to the reservoir and the system pressurized to 2 to 3 psig. To displace water present in the membrane, we vented buffer through the 1-way luer valve. We depressurized the assembly and removed surplus buffer via the 1-way valve. Then, we added the mRNA-LNP fluid to the reservoir and pressurized the reservoir to 2 to 3 psig to allow air to vent through the 1-way valve.

Once vented, the pressure was set to the desired test pressure and the 3-way valve was opened to allow fluid to flow through the test membrane. We collected the filtered fluid in a beaker on a calibrated, tared Ohaus Scout balance (model no. SPX6201) and recorded the mass of filtrate per unit of time . After testing, we depressurized the reservoir and drained any surplus fluid from the holder. A 70% IPA solution was then added to the reservoir and the system was repressurized to flush the filters with the 70% IPA. A post-use bubble point integrity test was then conducted on each filter. See Appendix for test results.

Competitor filter membranes were all 25 mm diameter encapsulated filter discs (Table 2) with known effective filtration areas (EFA). These were substituted for the 25 mm diameter disc holder for competitor filter membrane tests (Fig 2), then tested as detailed above. Each filter membrane listed in Tables 1 and 2 was tested six times.

Fig 2. Setup for constant pressure filtration studies.

Analytical methods

We used the Zetasizer Ultra instrument to determine LNP particle size and distribution through dynamic light scattering (DLS). To make sure the measurements were accurate, we analyzed all samples using an attenuator between 7 and 9.

To measure RNA concentration and EE of LNPs, we used the RiboGreen assay. RiboGreen reagent was first added to intact LNPs with Tris EDTA (TE) assay buffer to measure the unencapsulated RNA concentration. LNPs were disrupted using detergent and RiboGreen reagent was used to quantitate total RNA (encapsulated and unencapsulated). We calculated the concentration of encapsulated mRNA, by subtracting the unencapsulated mRNA concentration from the total mRNA concentration.

We used high-performance liquid chromatography (HPLC) to determine the lipid composition within the LNPs. Lipid concentrations were measured by UHPLC-CAD on a water:methanol gradient and C18 column.

To measure the mRNA integrity, we extracted mRNA from LNPs by silica affinity purification and performed capillary electrophoresis employing the RNA9000 kit (SCIEX) according to the manufacturer's protocol.

To perform the potency assay, we plated BHK293 cells at 5000 cells/well in a 96-well plate, 24 h before transfecting with LNPs at a starting dose of 1 µg/mL. Twenty-four hours post transfection, GFP protein expression was evaluated via high-throughput fluorescence microscopy using a BioTek Cytation 7 imaging system.

To test the ability of the membranes to remove bacteria from mRNA-LNPs, we performed a bacterial retention test using Brevindimonas diminuta suspended in the mRNA-LNPs under simulated process-specific conditions. Filters were challenged with a minimum of 1 x 10⁷ colony forming units (CFU/cm²) of effective filtration area (EFA).

The analytical assays listed above were compared to the critical quality attributes listed in Table 4 below.

Table 4. Typical CQAs applied to the assays performed on the mRNA-LNPs produced for this study

Statistical analysis

To identify statistical significance of the data, we performed a two-tailed T-test or an analysis of variance test (ANOVA) and a Tukey-Kramer post hoc test. The statistical significance in bar graphs was represented by compact letter displays (CLD), where one or more letters were assigned to each bar in the graph. The bars that share at least one letter were not significantly different from each other. Statistical significance was defined as p < 0.05.

Uncapped mRNA-LNP throughput and flux

The uncapped mRNA-LNP average throughput for the six replicate membranes tested ranged from 10 to 68 L/m² at 10 psig test pressure and 62 to 76 L/m² at 40 psig test pressure (Fig 3). Interestingly, at 40 psig all filters tested showed an average throughput of approximately 70 L/m², irrespective of membrane configuration and manufacturer. The data showed no significant difference between the throughputs of all filters tested at 40 psig, with Supor™ Prime filters giving the highest overall average throughput (Fig 3). At 10 psig, the data also showed that there was no significant difference in throughput for all the Cytiva membranes tested (Fig 3). However, at this pressure Supor™ Prime filters were found to have a significantly higher throughput than the competitor filters, as well as the highest overall throughput.

Fig 3. Average throughput for the six replicates of each membrane type per filter surface area. Error bars are +/- 1 standard error of the mean (SEM). Letters represent Tukey-Kramer test compact letter displays (CLD). Bars sharing at least one letter are not significantly different from each other (p > 0.05). Millipore SHC filter discs tested at 10 psig, plugged immediately preventing any substantial filtrate collection.

Figure 4 shows the average throughputs for single 10 inch membrane filter cartridges calculated from the published EFA of each filter type. An ANOVA with Tukey-Kramer post hoc test showed that at 40 psig Supor™ Prime filters had the highest throughput of all 10 inch filters, and this difference was significant for all filters except for the Sartopore Platinum.

At 10 psig, the Supor™ Prime filters throughput was significantly higher than all other filters tested. All other Cytiva and competitor filters were not significantly different from each other.

Fig 4. Average calculated throughput for the six replicates of each membrane type in L/10” filter. Error bars are +/- 1 standard error of the mean (SEM). Letters represent Tukey-Kramer test compact letter displays (CLD). Bars sharing at least one letter are not significantly different from each other (p > 0.05). Millipore SHC filter discs tested at 10 psig, plugged immediately preventing any substantial filtrate collection.

We measured the average flux of uncapped mRNA-LNP for the six membranes. As shown in Figure 5, the flux ranged from 231 to 425 LMH at 10 psig, and 333 to 1300 LMH at 40 psig.

The data analysis indicates that the flux data at 40 psig was not significantly different between the three Cytiva membranes, Supor™ Prime, Fluorodyne™ EX EDF, and Supor™ EX ECV, and the Sartopore II XLG membrane. However, Supor™ Prime and Fluorodyne™ EX EDF filters showed significantly higher flux than the Sartopore Platinum and Millipore Express membranes.

The flux obtained at 40 psig for Sartopore Platinum and Supor™ EX ECV membranes was not significantly different.

At 10 psig, all the filter membranes tested were not significantly different from each other.

Fig 5. Average flux for the six replicates of each membrane type in L/m²/h (LMH). Error bars are +/- 1 SEM. Letters represent Tukey-Kramer test compact letter displays (CLD). Bars sharing at least one letter are not significantly different from each other (p > 0.05). Millipore SHC filter discs tested at 10 psig plugged immediately, preventing any substantial filtrate collection.

Capped mRNA-LNP throughput

We measured the filtration throughput of mRNA-LNPs containing capped mRNA for Supor™ Prime and Fluorodyne™ EX EDF filters (Fig 6 and 7). Using Supor™ Prime membranes, the overall throughput of capped and uncapped mRNA-LNPs at 10 and 40 psig was not statistically different (81 vs 76 L/m² at 40 psig and 52 vs 50 L/m² at 10 psig, respectively).

Fig 6. Average throughput for the three replicates of each membrane type at 10 and 40 psig test pressures. Error bars are +/- 1 SEM.

Fig 7. Average throughput for the three replicates of each membrane type in liters/10 inch filter at 10 and 40 psig test pressures. Error bars are +/- 1 SEM.

The Fluorodyne™ EX EDF filters showed throughputs of 24 and 5.7 L/m² at 40 and 10 psig, respectively, after capping. These were significantly different at both pressures from the throughput before capping (69 and 30 L/m², respectively). However, this may be explained by the larger particle diameter of capped mRNA-LNPs compared to uncapped particles. The mean diameter of particles containing capped RNA was 127 nm (Fig 13), against 76 nm for the uncapped particles (Fig 8). Larger particles may have contributed to membrane plugging, leading to a reduced throughput for this membrane type.

Particle size and distribution of uncapped mRNA-LNPs

DLS sizing results show that immediately after filtration the average uncapped mRNA-LNP particle diameter was between 71 and 79 nm (Fig 8) with PDIs between 0.13 and 0.22 (Fig 9).

At 10 psig, we observed a significant difference between the average particle diameter before filtration (76 nm) and after filtration, for most membrane types. The exceptions were the Sartopore II XLG and Sartopore Platinum membranes, which showed no significant difference in average particle diameter before and after filtration. After filtration at 40 psig, all filters showed a significant difference in mean particle diameter compared to the diameter before filtration, except for the Sartopore Platinum filter. Even though the average particle diameter was lower after filtration, the values obtained were within the CQA specification for an mRNA-LNP product (Table 4).

After filtration at 10 psig, the data shows that the mean diameter of mRNA-LNP particles from Cytiva membrane filters was not significantly different from each other or from the Sartopore II XLG membrane. However, after filtration from Sartopore Platinum and Millipore Express SHC membranes the particle size was significantly larger compared to the other four filters tested. After filtration at 40 psig, the mean particle diameter showed no significant difference across all membrane filters tested.

Fig 8. Average particle diameter (nm) by DLS for the six replicates of each membrane type. Error bars are +/- 1 SEM. Letters represent Tukey-Kramer test compact letter displays (CLD). Bars sharing at least one letter are not significantly different from each other (p > 0.05). The unfiltered average mRNA-LNP diameter was 76 nm. Millipore SHC filter discs tested at 10 psig plugged immediately, preventing any substantial filtrate collection.

Analysis of the PDI results before and after filtration at 10 psig, showed that there was a significant difference for all membranes tested, except for the Sartopore II XLG membrane. The average PDI of the unfiltered particles was 0.19. At 40 psig, the PDI before and after filtration with Cytiva Supor™ Prime and Fluorodyne™ EX EDF membranes filters, or Millipore Express SHC filters, showed no significant difference, with the remaining membranes showing a significantly different PDI value compared to that obtained before filtration.

We also compared the uncapped mRNA-LNP PDI between the different membranes. At 10 psig, the Supor™ Prime, Supor™ EX ECV, and Fluorodyne™ EX EDF filters were not significantly different from each other. Whereas the Sartopore Platinum and Sartopore II XLG showed a significantly higher PDI compared to all other filters tested.

Fig 9. Average PDI for the six replicates of each membrane type. Error bars are +/- 1 SEM. Letters represent Tukey-Kramer test compact letter displays (CLD). Bars sharing at least one letter are not significantly different from each other (p > 0.05). The unfiltered average PDI was 0.19. Millipore SHC filter discs tested at 10 psig plugged immediately, preventing any substantial filtrate collection.

At 40 psig there was no significant difference between all membranes, with the exception of the Fluorodyne™ EX EDF PDI being significantly lower than that obtained for Supor™ Prime. The PDI value obtained for the Sartopore Platinum membrane at 10 psig was higher than expected, as the PDI for a typical mRNA-LNP is normally ≤ 0.20. However, past experiments suggest that the presence of a small number of large aggregates can skew the PDI. Nevertheless, all PDI measured values were within the acceptable range (Table 4).

mRNA concentration and integrity of uncapped mRNA-LNPs

RiboGreen analysis of mRNA concentration (Fig 10) showed that the average concentration of uncapped mRNA samples before filtration (measured at 0.107 g/L) was significantly different from the concentration after filtration at 10 psig for all of the membranes tested. However, it is worth noting that the small sample volumes collected at 10 psig could lead to a reduction in the overall concentration due to dilution effects of any residual buffer remaining in the filter holder or encapsulated disc. After filtration at 40 psig, we found a significant difference in concentration for both the Sartopore filter membranes, and for the Supor™ Prime and Supor™ ECV filters, relative to the concentration before filtration. The Fluorodyne™ EDF and Millipore SHC membranes did not show a significant difference.

Note that there is no RNA concentration specification defined in the CQAs in Table 4 as it is normally adjusted to comply with the required dosing.

Fig 10. Average RNA concentration (µg/mL) for the six replicates of each membrane type. Error bars are +/- 1 SEM. Letters represent Tukey-Kramer test compact letter displays (CLD). Bars sharing at least one letter are not significantly different from each other (p > 0.05). The unfiltered average mRNA concentration was measured as 107 µg/mL.

The data further showed that there was no significant difference in mRNA concentration across all Cytiva membranes at 10 psig (Fig 10). However, both Sartopore membranes had significantly lower mean mRNA concentration compared to the Fluorodyne™ EX EDF membrane, but not to Supor™ Prime or Supor™ EX ECV filters. At 40 psig no significant differences in the mean mRNA concentration were observed.

We then analyzed the encapsulation efficiency (EE) of the uncapped samples at both test pressures. We observed no significant difference between the unfiltered (98%) and filtered samples at 10 psig across all membranes tested. At 40 psig, only the filtration using the Sartopore II XLG membrane showed a significant difference compared to the unfiltered samples. No significant difference was observed on other membranes.

Fig 11. Average mRNA EE for the six replicates of each membrane type. Error bars are +/- 1 SEM. Letters represent Tukey-Kramer test compact letter displays (CLD). Bars sharing at least one letter are not significantly different from each other (p > 0.05). The unfiltered EE was 98%.

In addition, comparison of the mean EE between all membranes after filtration at 10 psig also showed no significant difference. However, the mean EE for LNPs filtered at 40 psig using Supor™ Prime filters showed a significant difference compared to using Fluorodyne™ EX EDF and Sartopore II XLG membranes; however, this difference was small relative to the CQA specification.

According to the CQAs, an EE ≥ 85% is acceptable (Table 4). All tests conducted, which have a CQA specification, were within the acceptable ranges.

Table 5. Average lipid concentration and yield for six replicates of each membrane type

¹ Average lipid yield = unfiltered lipid concentration / filtered lipid concentration.

We performed lipid analysis to compare the lipid concentration of uncapped mRNA-LNPs before and after filtration. The unfiltered average lipid concentration was 2.73 mg/mL. Postfiltration at 10 psig, we observed significant differences in lipid concentration for all membranes tested, except for Sartopore II XLG (Table 5). After filtration at 40 psig with Supor™ Prime, Supor™ EX ECV, or Sartopore Platinum filters, we observed no significant difference in lipid concentration compared to the unfiltered sample. Conversely, filtration with Fluorodyne™ EX EDF, Sartopore II XLG, or Millipore Express SHC membranes showed significantly different mean lipid concentration. We speculate that the results obtained at 10 psig are due in part to smaller sample volumes and dilution effects from residual buffer.

Generally, for mRNA-LNP dosing the production yield is defined in terms of mRNA concentration, rather than lipid. Therefore, more emphasis is placed on mRNA concentration and yield rather than on lipid concentration and yield.

RNA integrity was measured by capillary electrophoresis (CE). CE results are shown in Figure 12. Our results show significantly lower mRNA integrity after filtration with each filter compared to the prefiltered sample (94%). At 10 psig, filtration through the Sartopore II XLG membranes resulted in significantly lower mRNA integrity compared to the Supor™ EX ECV and Fluorodyne™ EX EDF membranes, but no significant difference was observed between the other membranes tested. At 40 psig, filtration through the Fluorodyne™ EX EDF membrane showed significantly lower mRNA integrity than all other membranes, which all yielded statistically similar mRNA integrity. It is likely that both membrane chemistry and pressure played a role in the forces experienced by the LNPs during filtration, resulting in the observed differences in final mRNA integrity.

Fig 12. Average mRNA integrity results for the six replicates of each membrane type. The unfiltered average mRNA integrity was measured as 94%. Error bars are +/- 1 SEM. Letters represent Tukey-Kramer test compact letter displays (CLD). Bars sharing at least one letter are not significantly different from each other (p > 0.05).

Particle size and distribution of capped mRNA-LNPs

We analyzed particle size and distribution of mRNA-LNPs containing capped mRNA. The average mRNA-LNP particle size immediately after filtration through the Supor™ Prime membrane was 125 nm with a PDI of 0.12 at 10 psig, and 126 nm with a mean PDI of 0.13 at 40 psig (Fig 13 and 14). The unfiltered mean particle diameter was 127 nm with a PDI of 0.13, which showed no significant difference to the post-filtration data at both test pressures. This indicates no change in the mRNA-LNPs mean diameter due to the filtration process.

After filtration through the Fluorodyne™ EX EDF membrane, the mean particle diameter was 125 nm with a PDI of 0.09 at 10 psig, and 128 nm with a PDI of 0.09 at 40 psig (Fig 13 and 14). Also in this case, we observed no significant difference in average particle size pre- and post-filtration at either test pressure. However, a significant difference was observed in PDI at 40 psig, indicating that at this pressure the particle size distribution was altered by the filtration through this membrane type. Nevertheless, the PDI for both membranes and at both test pressures was within the CQA specifications (Table 4).

Fig 13. Average particle size diameter (nm) by DLS for three replicates of each membrane type. Error bars are +/- 1 SEM.

Fig 14. Average PDI for three replicates of each membrane type. Error bars are +/- 1 SEM.

mRNA concentration and integrity of capped mRNA-LNPs

Next, we performed the RiboGreen analysis for the mRNA-LNPs containing capped mRNA. Filtration through the Supor™ Prime filters resulted in an mRNA concentration of 91 µg/mL at 10 psig and 96 µg/mL at 40 psig (Fig 15). Filtration through the Fluorodyne™ EX EDF filters, showed concentrations of 76 and 93 µg/mL at 10 and 40 psig, respectively. The average mRNA concentration before filtration was 114 µg/mL with an EE of 98%. All filtrate samples were found to be significantly different from the unfiltered mRNA concentration. The average RNA concentration does not have a defined CQA for mRNA-LNP products as the concentration must be adjusted to the particular application. The EE was not significantly different across all filtered samples (Fig 16).

Fig 15. Average RNA concentration (µg/mL) for three replicates of each membrane type. Error bars are +/- 1 SEM.

Fig 16. Average RNA EE for the three replicates of each membrane type tested. Error bars are +/- 1 SEM.

We then analyzed the average lipid concentration for the mRNA-LNPs containing capped mRNA samples (Table 6). The unfiltered average lipid concentration was 4.35 mg/mL. Both Supor™ Prime and Fluorodyne™ EX EDF filters showed significantly different lipid concentrations after filtration. However, there is no requirement for lipid concentration in the CQAs for mRNA-LNPs, as production yields for dosing purposes are defined in terms of mRNA rather than lipid, so more emphasis is typically placed on mRNA over lipid yield.

Table 6. Average lipid concentration and lipid yield for three replicates of each membrane type

¹ Average lipid yield = filtered lipid concentration/unfiltered concentration

We also analyzed the mRNA integrity of the mRNA-LNPs containing capped mRNA samples. The unfiltered average mRNA integrity was 87.5%. After filtration, the average mRNA integrity was significantly different for the Supor™ Prime membranes tested at 10 and 40 psig and for the Fluorodyne™ EX EDF membranes at 40 psig (Fig 17). After filtration at 10 psig through the Fluorodyne™ EX EDF membrane, the mRNA integrity was not significantly different from prefiltration levels. Since the post-filtration mRNA integrity was slightly higher than the unfiltered, it suggests that the difference is due to assay variability rather than mRNA degradation. Furthermore, mRNA integrity is not listed as a typical CQA value for production purposes.

Fig 17. Average mRNA integrity results for the three replicates of each membrane type tested. Error bars are +/- 1 SEM.

Potency assay for capped mRNA-LNPs

We assessed the average GFP expression of BHK cells transfected with mRNA-LNPs containing capped mRNA samples. The unfiltered average GFP expression was 62.3%. After filtration, the data showed that the GFP expression was not significantly different for both membrane types, at both test pressures (Fig 18). These data along with the inherent variability of in vitro work suggested that the different filtration conditions and membranes did not impact in vitro potency.

Fig 18. mRNA potency measured as the percentage of GFP expression for three replicates of each membrane type tested. Error bars are +/- 1 SEM. The unfiltered average GFP expression was 62.3%.

Finally, we tested the ability of Supor™ Prime (FTKASF) and Fluorodyne™ EX EDF (FTKEDF) filter membranes to remove a standard sterilizing-grade membrane challenge organism from an uncapped mRNA-LNP suspension at 10 and 30 psig, in accordance with the principles of ASTM (American Society for Testing and Materials International) F838-20 (1). Both removed 100% of incident B. diminuta bacteria from the challenge suspension at both test pressures (Table 7 and 8).

This application note was authored by Ian Johnson, Senior Scientist, Logan Ingalls, Associate Engineer II, and Kenneth J. Mabery, Principal Scientist. With thanks to the FastTrak™ team: Stephen W. Standring, Matt J. Leprohon, Brian Bergeron, Yaw Sarpong, Shawheen Fagan, Steve Turbayne, and Randolph M. Huelsman, who were involved in the development of methods and manufacture of the final mRNA sample. We also thank the analytical and development team composed of Erika Morata Perlera, Marilyn Patterson, Stephanie Purkis, Logan Molnar, Andrew Tapper, Aleksei Angell, Joanna Yovera, Mark Auger, Adam Hejmowski, Roman Kondra, Divakara Uppu, Aman Kumar, Ariel Zhang, and Ashley Braun. In addition, we acknowledge the validation team of Martha J. Folmsbee and Morven McAlister. Special thanks to David Sokolowski and Gary Pigeau for their patience and advice.

1. Standard Test Method for Determining Bacterial Retention of Membrane Filters Utilized for Liquid Filtration. ASTM International. ASTM F838-20 (2020). doi:10.1520/F0838-20.

Table 9. Post-use IPA/water bubble point integrity test values for the filter discs tested at 40 psig

¹ Wetted in 70% isopropyl alcohol.
² Wetted in 30% isopropyl alcohol

Table 10. Post-use IPA/water bubble point integrity test values for the filter discs tested at 10 psig

¹ Wetted in 70% isopropyl alcohol.
² Wetted in 30% isopropyl alcohol