2.1 Chemicals and reagents

Methanol (LCMS grade), 2-propanol (LCMS grade), and Triethylamine (HPLC grade) were purchased from Fisher (Fair Lawn, NJ). Acetic Acid (99.7%+) was purchased from Acros (West Chester, PA). Formic Acid: Triethylamine Complex (5:2), pure ethanol (200 proof), 1-tetradecanoic acid–myristic acid (99% grade), formamide (99% grade), 30% hydrogen peroxide, sodium hydroxide (10.0N) were purchased from Sigma-Aldrich (St. Louis, MO). Cholesterol (99%) was purchased from Minakem (Dunkerque, FR). 1,2-Distearoyl-sn-glycero-3-phosphocholine (DSPC) (>99%), 1-stearoyl-2-hydroxy-sn-glycero-3-phosphocholine (1-LysoPC), and 2-stearoyl-sn-glycero-3-phosphocholine (2-LysoPC) were purchased from Avanti Polar Lipids (Alabaster, AL). 1,2-Dimyristoyl-rac-glycero-3-methylpolyoxyethylene (PEG-DMG), was purchased from NOF (White Plains, NY). Water was dispensed using a MilliQ filter system from Millipore (Burlington, MA). Octadecanoic acid (referred to as stearic acid) was purchased from European Pharmacopeia (Strausburg, FR). Cationic Lipid 1, Cationic Lipid 2, and PEG Lipid 2 are proprietary products that were manufactured at Merck & Co., Inc., Kenilworth, NJ, USA which operates as MSD outside of the USA and Canada. The LNP vaccine samples were prepared as previously described [18]. In the LNP vaccine manufacture process, a lipid mixture was prepared by dissolving cationic lipid, cholesterol, phospholipid, and PEG-conjugate lipid (molar ratio of 50–58:30–39:10:1–2) in ethanol. Then lipid nanoparticles were produced by simultaneous T-mixing of the lipid mixture with an aqueous solution mRNA, followed by stepwise dia-filtration [19]. Both starting solutions are submitted for analysis along with samples post diafiltration. Each submitted sample has a targeted mRNA dose concentration and is used for sample preparation.

2.2 Assay stock standard and LNP sample preparation

Each lipid was individually dissolved in pure ethanol for stock standard preparation and then diluted with assay diluent (described below). Each LNP sample was treated with a diluent consisting of ethanol:formamide (85:15 %V/V) mixture. A minimum of four-fold dilution relative to the concentration of mRNA is required to fully recover all the lipids based on developmental data (see Section 3.1 for more sample preparation information).

When analyzing samples containing Cationic Lipid 1, 0.5% Formic Acid:Triethylamine Complex (5:2) was added to the assay diluent. When analyzing samples containing Cationic Lipid 2, 19.2mM glacial acetic acid and 16.2mM triethylamine were added to the assay diluent. Samples were diluted to final concentrations that were within the prepared calibration curves.

2.3 Mobile phase composition

When measuring LNP samples containing ionizable Cationic Lipid 1, the following mobile phase compositions were used: mobile phase A consisted of water:methanol (1:1 %V/V) with 0.5% Formic Acid:Triethylamine Complex (5:2) – a premixed, purchasable reagent from Sigma Aldrich – at pH of 3.5, and mobile phase B was composed of methanol with 0.5% Formic Acid:Triethylamine Complex (5:2). When measuring LNP samples containing Cationic Lipid 2, the following mobile phases were used: mobile phase A was composed of water:methanol (1:1 %V/V) with 19.2 mM glacial acetic acid and 16.2 mM triethylamine, and mobile phase B was composed of methanol with 19.2 mM glacial acetic acid and 16.2 mM triethylamine. The pH of both mobile phases was measured after mixing by adding 1mL of mobile phase and 9 mL water, and adjusted with their respective acid or triethylamine when necessary.

2.4 UHPLC-CAD conditions

Samples were analyzed on a Waters Acquity Classic binary UPLC system (Waters Corp, Willmington, DE). Strong needle wash was composed of 2-propanol. Weak needle wash composed of water:methanol (80:20 %V/V). Seal wash was composed of water:2-propanol (1:1 %V/V).

Samples were injected using a Waters Acquity Classic autosampler using a fixed loop injection at 5 μL with a 50 μL sample loop. Lipids were separated using a Waters BEH C18 1.7 μm particle size, 2.1 mm ID x 150 mm column heated to 50°C (Waters Corp, Willmington, DE). Using a flow rate of 0.3mL/min, the mobile phase composition began at 0% B with one-minute hold time, then %B increased to 79% as the first step gradient change. At 7.5 min, %B increased to 98% as the second step gradient change; at 17.5 min, the %B decreased to the initial condition, 0%, and equilibrated for 3.5 min before the next sample injection. Analytes were detected using a Corona VEO RS CAD from Thermo Scientific (Waltham, MA). Filtered nitrogen was used at a preset manufacture pressure of 60.7 psi; data collection rate of 5 Hz; power function of 1; nebulizer temperature set to 35°C. Data were processed using Waters Empower 3 software.

2.5 Theoretical Log D calculations

Advanced Chemistry Development (ACD/Labs) Software V11.02 (© 1994–2020 ACD/Labs) was used to predict the logarithm of the distribution coefficient, Log D_7.4 from the compound structures. The distribution coefficient (D) considers the partitioning of all ionic species of a compound. In order to calculate the Log D_7.4 value for a test compound, both the logarithms of the partition coefficient (Log P) and the acid dissociation constants (pKa) are needed for weak acids or bases. To perform these calculations, the test compound structure is drawn graphically, after which the computational program PrologD automatically calculates Log D_7.4. The pKa of cationic lipids were also theoretically calculated using this software.

2.6 Forced lipid degradation preparation

Oxidation of Cationic Lipid 2 was performed by adding 60 μL 30% H₂O₂ to 5 mL of 85 mg/mL Cationic Lipid 2 in pure ethanol. This sample was stirred for 4 h before it was injected at 1 μL along with an untreated Cationic Lipid 2. PEG Lipid 2 hydrolysis was performed by treating 2 mL of 0.7 mg/mL PEG Lipid 2 solution in ethanol with 0.1 mL of 0.01 N NaOH.

3.1 Sample preparation diluent selection

In order to analyze lipid components in mRNA-LNP vaccines, the diluent selection is critical. The diluent ought to be able to dissolve each lipid component and dissociate any interactions between lipids and encapsulated mRNA. This is especially true for lipids with cationic functional groups that have strong ionic interactions with negatively charged mRNA molecules.

The first diluent evaluated was a 1:1 Ethanol:DMSO mixture. A sample of 0.05 mg/mL mRNA loaded LNP containing PEG-DMG, Cholesterol, Cationic Lipid 1 and DSPC was analyzed using four different dilutions - four, six, eight, and ten (Figure 1). All four lipid components were adequately recovered at a sixfold, eightfold, and tenfold dilutions with recovery greater than 95%. However, after a four-fold dilution, both Cationic Lipid 1 and DSPC were severely under recovered at ≤35%. In addition, the %molar fraction of each lipid was consistent with the targeted formulation (±1%), except for the fourfold dilution sample. The four-fold sample showed an increased molar fraction for PEG-DMG and cholesterol, and a decreased molar fraction for cation lipid 1 and DSPC. Both the decrease in lipid recovery and a decrease in molar ratio suggest that the LNP and mRNA were either not fully soluble or the not fully dissociated from each other. A new diluent was considered to achieve greater solubility of all lipid and mRNA components at a lower dilution.

When optimizing the new diluent, a similar dielectric solvent strength as ethanol:DMSO (1:1) was desired while also increasing the amount or strength of a nonpolar solvent for adequate solubility of all lipids. Linear combinations using the dielectric constants of the pure solvents and their volume were used to estimate solvent combinations with similar dielectric solvent strength as ethanol: DMSO (1:1). A diluent with a combination of ethanol and formamide at a 85:15 ratio was chosen.

Another dilution experiment measuring an LNP sample containing 0.05 mg/ml mRNA was analyzed using dilution factors of 4 and 10. The ethanol: formamide diluent showed all lipid components had greater than 93% recovery and the target molar fraction was observed in both dilutions tested (Figure 1b). These data suggest that the ethanol:formamide diluent was able to disrupt all non-covalent interactions between the mRNA and lipids and maintain solubility for both mRNA and lipids during analysis. Based on this data, ethanol: formamide (85:15) was chosen as the optimal diluent for mRNA-loaded LNP vaccine samples. Furthermore, when evaluating samples with various mRNA concentrations, samples were diluted based on the concentration of mRNA due to a fixed mRNA to lipid ratio. For example, a 0.05 mg/mL mRNA must be diluted 4-fold and a 0.1 mg/mL mRNA sample 8-fold for full recovery of lipids and be within the linear range.

3.2 Chromatographic separation

A UHPLC method that can quantitate neutral and cationic lipids was developed for the evaluation of LNPs containing mRNA. The final chromatographic conditions consisted of a step gradient with two isocratic holds after various chromatographic conditions were explored during method development.

Potential degradants such as myristic acid, lyso-PC 1 and 2, and stearic acid are more hydrophilic than their parent lipids and eluted in the first isocratic hold. The four main lipid components are highly hydrophobic and eluted in the second isocratic hold. Two isocratic holds were chosen to minimize the effect of each analytes response factor based on mobile phase composition [20]. Impurities or degradants were monitored by loss of four main lipid peaks and percent impurity analysis for any addition of new, unspecified peaks. Percent impurity analysis was performed by integrating any peaks apart from the void, system, and parent lipid peaks. This analysis was monitored and compared in stability studies to the zero time point. As previously mentioned, based on the relative response of each analyte, the universal response of both parent lipids and their degradants are not achievable with this method. It would be worth exploring this method with the use of an inverse gradient or another means to a more universal response for impurity analysis utilizing a CAD.

Figure 2A shows the separation of an mRNA-LNP sample containing PEG-DMG, Cholesterol, Cationic Lipid 1, and DSPC. These lipid components are separated in order of increasing retention time 9.60, 10.15, 12.00, and 13.40 min. Figure 2B illustrates the separation of an mRNA-LNP sample containing the four lipids PEG Lipid 2, Cholesterol, Cationic Lipid 2, and DSPC in order of increasing retention time at 9.25, 10.15, 11.75, and 13.40 min. In both formulations, the four main lipids were baseline resolved for identification and content analysis.

Of the four lipid analytes, three of them (PEG-DMG, Cholesterol, and DSPC) have an overall neutral charge and are not affected by mobile phase pH. These neutral and zwitterionic lipids were well separated prior to introducing a mobile phase pH modifier. Each analyte's calculated LogD_7.4 values helped predict their retention under reversed phased conditions, that is, retention increases with an increase in LogD_7.4 value. DSPC has the highest calculated LogD_7.4 value of 13.60 of the neutral lipids and as predicted eluted after cholesterol and both PEG-conjugates. Cholesterol's calculated value was 9.85 and eluted well before DSPC. The predicted LogD_7.4 value for the neutral and zwitterionic lipids correlated well with their retention times. Neither PEG Lipid 2 or PEG-DMG LogD_7.4 values were predicted for this study due to the complexity of the polymer. It is conceivable that polyethyleneglycol's (PEG) large, hydrophilic head group plays an important role in reducing its hydrophobicity and retention in this method. Cationic Lipid 1 and Cationic Lipid 2's LogD_7.4 values were theoretically calculated to be 15.48 and 12.54, respectively. These values were somewhat helpful in predicting the strength of their hydrophobicity. Instead, their retention was greatly influenced from their ionizable amine funtional group.

The mobile phase pH was a crucial attribute that influenced the retention of the cationic lipids by controlling protonation of their amine functional groups. To start, triethylamine was chosen due to its volatility and its ability to prevent secondary interactions between the amine functional groups and silica base stationary phases thereby reducing tailing of the cationic lipid peaks (data not shown) [21, 22].

When studying LNP samples containing Cationic Lipid 1, the mobile phase pH was adjusted to 3.5 by mixing formic acid at a 5:2 ratio with triethylamine – purchasable from Sigma Aldrich. Optimizing the pH to 3.5 controlled the elution of Cationic Lipid 1, with a theoretical pKa of 9.37 - between cholesterol and DSPC (Figure 2A). Doing so provided sufficient resolution for impurity and degradant analysis.

When analyzing LNP samples containing Cationic Lipid 2 at pH 3.5, it eluted prior to cholesterol. The Cationic lipid 2, with a theoretical pKa of 9.70, eluted between PEG Lipid 2 and cholesterol, further demonstrating the effect of mobile phase pH on cationic lipid separation (Figure 3A). The separation between Cationic Lipid 2 and PEG Lipid 2 was not optimal for quantitation or impurity analysis. So, a mobile phase pH of 3.5 was not used for analyzing Cationic Lipid 2, and a higher mobile phase pH was investigated.

To increase the pH above 3.5, acetic acid was substituted for formic acid. This substitution, after pH adjusting to 5.4, eluted Cationic Lipid 2 between cholesterol and DSPC providing comparable retentions as with the more acidic mobile phase. Similarly, a retention time shift was observed for Cationic Lipid 1 when using a pH 5.4 mobile phase (Figure 3B). At higher pH, there was less ionization of the amine functional group resulting in a more hydrophobic molecule. Consequently, causing Cationic Lipid 1 to elute after DSPC.

3.3 Forced lipid degradation study

This method was optimized to capture potential degradation products of the four lipids measured in each method. An impurity reference standard was prepared using myristic acid, Lyso-PC1, Lyso-PC2, stearic acid, and a known degradant of Cationic Lipid 1 dissolved in pure ethanol. Myristic acid was used because it is a hydrolysis degradant of both PEG lipids. Lyso-PC1, Lyso-PC2, and stearic acid were chosen because they are hydrolysis degradants of DSPC. Cholesterol is known to be stable and therefore no suspected degradants of cholesterol were not used in this experiment [23].

Another degradant is the oxidation product of Cationic Lipid 2 that eluted between cholesterol and Cationic Lipid 2 at 10.6 min (Figure 4B) [24]. Forced degradation products of PEG Lipid 2 were determined by hydrolysis under basic conditions. Three degradation peaks were observed, see Figure 4C. Though mRNA-LNP vaccines would not likely be exposed to NaOH or high concentrations of H₂0₂, these conditions were chosen in order to evaluate chromatographic selectivity of the analytes’ degradants, especially when these samples were monitored for stability over 12 months.

Figure 4A features a chromatogram of the lipid hydrolysis degradants, myristic acid, lyso-PC1, lyso-PC2, and stearic acid. All were resolved in order of increasing retention times 3.5, 3.8, 4.0, and 7.5 min in the first isocratic hold of the gradient at 79% mobile phase B. From 5.0 min to 7.50 min, mobile phase B is ramped to 98% to elute the four main lipid components, Cationic Lipid 2 oxidation, and the known degradant of Cationic Lipid 1.

3.4 Method validation

The method's accuracy, repeatability (intra-assay precision), inter-assay precision, specificity, linearity, and range were all validated for this UHPLC-CAD method. Validation results are summarized in Table 1.

The accuracy of this method was determined by spike recovery using a spike of 20%, 50%, and 70% of each lipid. The intra-assay precision of six injections was between 1 and 7% for all lipids (n = 6). Inter-assay precision was determined by evaluating the same sample between three separate runs on three different days. The inter-assay precision results for all lipids were between 1% and 6% (n = 3).

Four independent preparations of a negative control sample were prepared, and each preparation was injected once. The negative control sample consisted of the formulation buffer without lipids or mRNA. The negative control chromatogram and blank chromatogram were compared. Peaks with the same retention time in the blank were disregarded when compared to the negative control to obtain the peaks that were only related to the negative control. No interfering peaks from the negative control sample were observed.

The previously mentioned oxidation study of Cationic Lipid 2 was performed to demonstrate specificity for this method. Linearity was evaluated using a quadratic fitting between three independent runs using the concentration ranges listed in Table 1. The correlation coefficient (R²) of each quadratic fit was 0.993–1.000 for all lipids.

Assay performance was monitored using a control chart and a positive control sample. The performance was monitored over the course of 28 months with an mRNA-LNP sample containing Cationic lipid 1, and 15 months with an mRNA-LNP sample containing Cationic Lipid 2. Their results are listed in Table 2. All lipid concentration measurements had a % RSD of < 5% with the exception of PEG-DMG and PEG Lipid 2. Both PEG-lipid conjugates had a % RSD <10%. This is likely because each of the two PEG-conjugate lipids is the lowest concentrated relative to the other lipids. Their broad peak shape also likely contributes to this variability.

Finally, the LOQ of this method is listed as the lowest range of lipids in Table 1. LOD was not determined in this paper.

3.5 Lipid stability

This lipid assay was used to monitor lipid stability at four temperatures –70°C, –20°C, 4°C, and 25°C and wsd pulled for analysis at the following time points: initial (time zero), 1 month, 3 months, 6 months, and 10 months. The stability data are illustrated in Figure 5 for the following four main lipids: PEG-DMG, cholesterol, Cationic Lipid 1, and DSPC. It is a relative comparison of each time point with the initial –70°C time zero measurement. Each sample from each timepoint was prepared in duplicate, and each timepoint sample submission was tested once. PEG-DMG showed 0% change at –70°C, 4°C, and 25°C in 1 M, –20°C, and 4°C in 3 M and 4°C in 6 M. The stability study at elevated temperature, 25°C, was run for 6 months and discontinued due to significant mRNA degradation. The decrease of the lipid concentration was approximately 10–15% at 6-month time point under –70°C storage temperature, which was the most significant change among all the time points at all the temperatures. The trend of the stability data across the temperatures indicates it may be due to sample handling, variability from sample pull date to time of testing, or assay variability instead of chemical degradation. Samples stored at higher temperatures for longer times would be expected to show more lipid concentration decrease if it was caused by chemical degradations, which was not observed in the stability study. Therefore, the stability results demonstrated all four lipids were chemically stable in the mRNA vaccine product at all the evaluated temperatures.