A Comprehensive Evaluation of Analytical Method Parameters Critical to the Reliable Assessment of Therapeutic mRNA Integrity by Capillary Gel Electrophoresis

ABSTRACT

In recent years, messenger ribonucleic acid (mRNA)‐lipid nanoparticle (LNP) biotherapeutics have demonstrated significant promise in disease treatment and prevention given their rapidly modifiable production processes and considerable capacity to adapt to complex or low‐yielding proteins of interest. As a result, many products are currently being developed in this space. Critically, well‐characterized and appropriately designed assays are required to monitor purity and integrity in order to maintain the efficacy and consistency of these novel products. Currently, capillary gel electrophoresis with laser‐induced fluorescence (CGE‐LIF) and ion‐pair reversed‐phase liquid chromatography (IP‐RPLC) are techniques of choice for mRNA integrity analysis. However, most methods proposed for biotherapeutic analysis have been developed using naked mRNA without LNP components or proprietary buffer formulations, which can obscure undiscovered impurities or complex interactions between mRNA and the sample matrix. In this study, we addressed these methodological challenges by using a biotherapeutically relevant commercial mRNA‐LNP sample (approx. 4200 b) to refine and optimize a customizable CGE‐LIF method currently under consideration for mRNA‐LNP biotherapeutic analysis. We systematically characterized how critical method parameters—such as denaturant type, concentration, and usage—and LNP disruption protocols can interfere with accurate mRNA integrity analysis in CGE‐LIF and IP‐RPLC. We found that optimal conditions for CGE‐LIF assay sensitivity, variability, and resolution included sample precipitation by isopropanol, high urea concentrations, no formamide as a sample diluent, and high concentrations of dye. Finally, the advantages and disadvantages of both CGE‐LIF and IP‐RPLC are highlighted, and a discussion of key considerations when using or designing methods for mRNA integrity assessment is presented.

Abbreviations

CGE‐LIF

capillary gel electrophoresis with laser‐induced fluorescence

CQAs

critical quality attributes

DBAA

dibutylammonium acetate

DMSO

dimethylsulfoxide

DSPC

1,2‐diasteroyl‐sn‐glycero‐3‐phosphocholine

IP‐RPLC

ion‐pair reversed‐phase liquid chromatography

IS

internal standard

LMS

late‐migrating smear

LNP

lipid nanoparticles

mRNA

messenger ribonucleic acid

PDA

photodiode array

PVP

polyvinylpyrrolidone

SD

standard deviation

TBE

Tris–Borate–EDTA

TEAA

triethylammonium acetate

UV

ultra‐violet

1. Introduction

Single‐stranded messenger ribonucleic acid (mRNA) is a highly promising biotherapeutic agent for disease treatment and prevention [1, 2]. Unlike protein‐based therapeutics, mRNA‐based therapies deliver transcripts used by host cell machinery to produce immunogenic or functional proteins of interest [3]. Typically, mRNA transcripts are larger transcripts (>2000 b), requiring encapsulation by lipid nanoparticles (LNPs) that protect against the environment and improve cellular uptake [4]. This mRNA‐based approach has significant advantages, including the delivery of complex or low‐yielding proteins [5] and facile adaptation to any target in emergency responses. The swift large‐scale development of COVID‐19 mRNA vaccines has also prompted significant discussion on assessing mRNA modalities as platform technologies instead of individual products characterized by manufacturing process and disease target [6]. Given the expected following of this novel modality, correctly identifying critical quality attributes (CQAs) and reaching a consensus on the means to assess them is an area of interest for both manufacturers and quality assessors.

The early identification of CQAs in biological products is critical for maintaining safety and efficacy in a product lifecycle. However, given the complex mode of action for mRNA‐LNP biotherapeutics, directly evaluating efficacy or potency can be difficult [7]. As a result, physicochemical properties related to potency‐determining attributes need to be extensively characterized [8]. One such CQA is transcript length or integrity, because critical elements at the 5′ and 3′ termini are required for efficient protein synthesis [9] and the manufacturing process necessarily results in a mixture of fragments [10]. RNA is also susceptible to hydrolytic cleavage from both spontaneous and RNAse‐induced degradation. Encapsulation offers some protection, but fragmentation remains a critical concern for long‐term stability [11]. Thus, as a stability‐indicating CQA, it is critical that the mRNA length profile is accurately assessed over time.

Currently, capillary gel electrophoresis (CGE) and ion‐pair reversed‐phase liquid chromatography (IP‐RPLC) are preferred methods for assessing transcript length and fragmentation [12, 13, 14, 15]. In CGE, RNA analytes are injected, typically electrokinetically, and electrophoresed through a sieving matrix containing denaturants, with larger fragments migrating slower and detection through an RNA‐binding dye [16]. Although CGE has been commonly used to analyze small DNA oligonucleotides, several groups have explored methods to improve analysis for large mRNA molecules, but mostly in research contexts [17, 18, 19, 20, 21]. IP‐RPLC is a chromatographic technique where ion‐pair agents complex the phosphate backbone of mRNA, thereby increasing affinity of longer fragments to the stationary phase [22]. Thus, longer RNA molecules require higher organic content for elution. Similar to CGE, IP‐RPLC is commonly used for small DNA oligonucleotide analysis [22, 23] but is also increasingly explored for longer nucleic acid analytes [24, 25, 26]. Although both provide a measure of mRNA size, their mechanisms of separation are vastly different, which can impact their utility to different steps of the biomanufacturing process. Defining these impacts would be valuable to the interpretability of mRNA quality evaluations.

However, in order to establish the necessary guidelines and reference standards for CQA analysis [27, 28], foundational studies on examining mRNA biotherapeutics are needed. As a contrast, the CQAs for novel protein‐based therapeutics can be based on the already extensive literature of physicochemical characterization and efficacy studies [29, 30]. Although this understanding is gradually forming in the biotherapeutic context for mRNA products [25, 31, 32, 33], many studies examine only naked mRNA [12, 26, 34] or use proprietary buffer formulations with closed instrumentation [35, 36]. Although informative, unknown impurities may go uninvestigated due to missing components or unpredictable interactions with undisclosed components. Indeed, lipids are important to the mRNA impurity profile as Packer et al. reported the discovery of novel mRNA‐lipid impurities, which were undetectable in a CGE method but present by IP‐RPLC [25]. Interestingly, late‐migrating impurities have been reported by CGE but only in the absence of formamide [27]. Both reports suggest that the differing method components of CGE with laser‐induced fluorescence (CGE‐LIF) and IP‐RPLC can result in different impurity profiles, which needs to be better understood. Given the emergence of this modality and the potential for novel impurities, it will be important to establish standard methods to accurately assess known CQAs while also maintaining the means to characterize unknown impurities through customizable methods.

To address the knowledge gaps of CGE‐LIF and IP‐RPLC method performance for mRNA‐LNPs, we used a large commercial mRNA‐LNP product (∼4200 b) to systematically evaluate critical method parameters for CGE‐LIF and compared them to IP‐RPLC. Both methods have been proposed for mRNA quality assessments [27]. In this study, we first examined how common gel additives—urea, formamide, and polyvinylpyrrolidone (PVP)—impacted CGE‐LIF resolution of an RNA ladder. Next, we evaluated how LNP disruption affects integrity measurements for the mRNA‐LNP. Leveraging these observations, an experimental design was implemented to investigate both main and interaction effects of urea, formamide, and dye percentage. CGE‐LIF performance was then compared to IP‐RPLC analysis. Finally, a discussion on designing methods for consistency and standardization from a biotherapeutic perspective will be presented.

2. Materials and Methods

2.1. Reagents and Samples

RNA markers (281 b–6583 b), control RNA encoding luciferase (1803 b), and kanamycin positive RNA used as a spike‐in internal standard (IS) (1.2 kb) were purchased from Promega Corporation (Wisconsin, USA). Water (both LC–MS grade and DEPC‐treated), acetonitrile, Tris–Borate–EDTA (TBE) buffer, dimethylsulfoxide (DMSO), SeaKem ME Agarose, deionized formamide, SYBR Green II, dialysis cassettes, and Dulbecco's phosphate‐buffered saline (DPBS) were purchased from Thermo Fisher Scientific (Massachusetts, USA). Isopropanol, Triton X‐100, urea, PVP, dibutylammonium acetate (DBAA), and Amicon Ultra centrifugal concentrators were purchased from MilliporeSigma (Missouri, USA). Triethylammonium acetate (TEAA) was purchased from TCI America (Oregon, USA). Ethanol was purchased from Greenfield Global (Toronto, Canada). Firefly luciferase (FLuc) mRNA capped with CleanCap AG analog modified with 5‐methoxyuridine (1922 b) was purchased from Trilink Biotechnologies (California, USA). The in‐house fabrication of LNPs used ionizable lipid ALC‐0315 and PEGylated lipid ALC‐0159 purchased from MedKoo Biosciences Inc. (North Carolina, USA) as well as 1,2‐diasteroyl‐sn‐glycero‐3‐phosphocholine (DSPC) and ovine cholesterol purchased from Avanti Polar Lipids Inc. (Alabama, USA).

Samples of commercial mRNA products with modified nucleosides were used (0.1 mg/mL; approximately 4200 bases in length) encapsulated in a mixture of PEGylated lipids, ionizable lipids, phospholipid DSPC, and cholesterol were used for analysis.

An additional mRNA‐LNP control sample containing FLuc mRNA was fabricated in‐house by rapid microfluidic mixing using a NanoAssemblr Ignite system (Precision Nanosystems Inc.). ALC‐0315, DPSC, ovine cholesterol, and ALC‐0159 were combined at a molar ratio of 46.3:9.4:42.7:1.6 respectively. An aqueous solution of mRNA in 50 mM citrate buffer (pH 4.0) was mixed with lipids in ethanol at a 3:1 volumetric ratio with at a total flow rate of 5 mL/min. After mixing, particles were dialyzed with 10k MWCO cassettes at 4°C for 18 h against a 1000‐fold volume of DPBS. LNPs were 0.22 µm filtered and concentrated with an Amicon Ultra‐4 10k MWCO centrifugal concentrator. All particles were made with a nitrogen/phosphate (N/P) ratio of 6:1. The final concentration was adjusted to 0.1 mg/mL by DPBS.

2.2. LNP Disruption and Sample Preparation

Detergent‐based disruption was achieved by mixing 0.1 mg/mL of mRNA‐LNP with a solution containing either 2% or 20% Triton X‐100 (w/w) in a 1:1 ratio (v/v) and incubating the mixture at 30°C while shaking at 600 rpm for 20 min. Disruption by isopropanol precipitation was prepared as previously described [25]. For agarose gels, 0.3 µg of disrupted mRNA was mixed with loading dye, heat‐denatured at 70°C for 2 min, and cooled on ice for 5 min. For CGE‐LIF analysis, 12 µL of disrupted mRNA was added to 132 µL of either 0.05X TBE or 0.05X TBE with 25% (v/v) formamide, heat‐denatured at 70°C for 2 min and cooled on ice for 5 min. For IP‐RPLC analysis, disrupted mRNA was diluted to the same concentration as CGE‐LIF in nuclease‐free water before injection.

2.3. Capillary Gel Electrophoresis (CGE)

A 30 cm bare fused silica capillary (effective length: 20 cm and internal diameter: 50 µm) was used with a PA800 Plus system (Sciex) equipped with a solid‐state laser and fluorescence detector (λex = 488 nm/λem = 520 nm). Prior to first use, the capillary was conditioned: 0.1 M NaOH, DEPC‐treated water, 0.1 M HCl, and DEPC‐treated water, each injected at 20 psi for 5 min. For each separation, the capillary was filled with gel buffer (1% PVP, 4 M urea, 1X TBE, and 0.02% (v/v) SYBR Green II; prepared within 24 h of use) by applying 50 psi for 5 min. Samples were maintained at 4°C in the autosampler and electrokinetically injected at 5 KV reverse polarity for 5 s. Separation was 6 KV for 20 min, and cartridge temperature was maintained at 25°C. Analytes were detected by SYBR Green II fluorescence upon RNA binding. Injections were performed in triplicate and on multiple days. Data were acquired with 32Karat (version 10.1) and integration by Empower 3 (version 7.21). Velocity‐corrected areas (VCAs) were calculated by automated integration in Empower unless otherwise stated [37].

2.4. Statistical Analysis

With two randomized sequences of eight injections in duplicate (Sequences 1 and 2), the full factorial experiment was designed to remove potential time‐dependent confounding effects. Each sequence was analyzed on a separate day. Both sets of sequences were repeated three times (Sets 1, 2, and 3) to increase statistical power (the total number of observations = 96). Electropherograms were manually integrated with the Empower 3 (version 7.21) software (Waters) into % shoulder, % main peak, and % late‐migrating smear. Outliers were removed whenever the % main peak difference between duplicates in one assay was larger than 20% (n = 76). Descriptive statistics were generated using R and SAS Enterprise Guide 7.1. Generalized linear models were constructed using the GLMSELECT procedure of SAS Enterprise Guide 7.1 with Bayesian information criterion for % main peak, % shoulder, % late‐migrating smear, and main peak width at half‐height. Factors were urea concentration, the presence of formamide, and dye concentration.

2.5. Ion‐Pair Reversed‐Phase Liquid Chromatography (IP‐RPLC)

A DNAPac RP column (4 µm, 100 mm × 2.1 mm) was used on a Waters Alliance e2695 Separation Module (Waters) equipped with photodiode array (PDA) and fluorescence detectors. A previously reported method was adapted [25], with a flow rate at 0.35 mL/min and column temperature at 60°C. Mobile phase A was 50 mM DBAA, 100 mM TEAA, and mobile phase B was 50 mM DBAA, 100 mM TEAA, 50% (v/v) acetonitrile. Injections were 10 µL (40 ng mRNA) and detected by ultra‐violet (UV) at 260 nm. The gradient table is listed in Table S1. Empower 3 (version 7.21) was used for data acquisition and analysis.

3. Results

3.1. In‐Gel Additives Impact CGE‐LIF Signal‐To‐Noise and Resolution

Although convenient instruments and kits have been marketed for mRNA‐LNP analysis, the proprietary nature of these systems makes method customization and foundational research difficult. Thus, we chose an open‐source CGE‐LIF protocol [38] to clearly assess the impact of method parameters (Table 1). In addition to being highly accessible, a variant of this method has been proposed as an analytical procedure for integrity analysis of mRNA‐LNP biotherapeutics [27].

TABLE 1.

Method parameters for capillary gel electrophoresis with laser‐induced fluorescence (CGE‐LIF) analysis.

Parameter	Details
Gel buffer composition	1% PVP, 4 M urea, 1X TBE
Dye	0.02% (v/v) SYBR Green II
Injection voltage	5.0 KV
Injection time	5 s
Separation voltage	6.0 KV

Initial adaptation showed high reproducibility with prescribed standards: an RNA ladder (281 b–6583 b) and an RNA marker mix (1.8 and 1.2 kb) (Figure 1A). Some variation in intensity was observed, notably for the smallest 281 b fragment. On the basis of the ladder, the length of the RNA control was estimated at 2074 b, approximately 300 bases higher than the actual length; as calculated by linear regression of migration time and the natural log of RNA fragment length (Figure S1).

FIGURE 1.

CGE‐LIF analysis of RNA ladders under varying conditions: (A) ladder spiked with a 1.2 kb IS (black) compared to a control RNA spiked with the same 1.2 kb IS (red), (B) ladder analyzed under varying in‐gel urea concentrations, (C) ladder treated with in‐sample formamide under varying in‐gel urea concentrations, (D) ladder analyzed under varying in‐gel formamide concentrations, and (E) ladder analyzed under varying PVP concentrations. Traces have been stacked in panels with titrations. Light red (A and C) and light green traces (D and E) are the same traces as the corresponding red and green traces but with intensities normalized to allow visualization of the low‐intensity peaks. CGE‐LIF, capillary gel electrophoresis with laser‐induced fluorescence; IS, internal standard; PVR, polyvinylpyrrolidone; RNA, ribonucleic acid.

Because instrument parameters have been explored elsewhere [38], we focused on gel buffer additives. A characteristic property of RNA molecules is the propensity to form stable secondary structures, which interfere with size‐based separations [39]. Denaturants, such as urea and formamide, are commonly used to improve resolution of large mRNA [17]. Accordingly, decreased urea concentrations in the gel buffer broadened RNA peaks and increased electrophoretic mobility, resulting in lower migration times (Figure 1B) reflecting previous reports [40]. Interestingly, the absence of urea also resulted in a large relative peak area of smaller fragments (Figure S2), which could be due to an enhanced electrokinetic injection bias. Although past studies have reported a decrease of fluorescence signal in the presence of urea [41, 42], fragment‐dependent impacts on intensity have not been reported to our knowledge.

In comparison, formamide is often added to samples for denaturation prior to analysis. In our hands, this treatment only showed minor improvements in peak resolution (Figure 1C) compared to without treatment (Figure 1B), suggesting that denaturation in‐sample is insufficient for CGE‐LIF analysis. In‐sample formamide also increased assay variability and decreased signal intensity. Resolution was only improved when formamide was included in‐gel (Figure 1D), corroborated by work from Lu et al. [17]. However, in‐gel concentrations of >40% completely abrogated fluorescence, possibly due to interference with dye binding or sample injection [36]. Although both urea and in‐gel formamide improved resolution, there was substantial dissimilarity between profiles. With 40% in‐gel formamide, all nine peaks were present suggesting full denaturation similar to 8 M urea; however, the smallest 281 b fragment remained broad. With 8 M urea, the smallest fragment was consistently resolved into two peaks, suggesting that both the nature and concentration of the denaturant can resolve differential mRNA forms.

Finally, as the sieving polymer, PVP is often adjusted since higher concentrations improve the resolution of larger fragments. Overall, this was observed, albeit with a noticeable reduction in signal intensity, suggesting multiple mechanisms affecting assay sensitivity (Figure 1E). Interestingly, the smallest 281 b fragment appeared to split into two peaks but remained broad, more similar to 40% in‐gel formamide than 8 M urea, suggesting that the specific chemical properties of urea could be causing fragment separation.

3.2. LNP Disruption by Detergent Interferes With mRNA‐LNP Integrity Assessment

The analysis of mRNA‐LNPs requires disruption of the protective encapsulating LNP as only free mRNA molecules can be evaluated by CGE‐LIF. The profile of the released mRNA can then be visualized as a single sharp peak, containing intact mRNA, and a broad earlier eluting shoulder, containing mRNA fragments [12]. Later‐migrating peaks or tailing are often attributed to other mRNA‐related impurities [12, 25, 27].

Given the importance of disruption in sample preparation, two common disruption protocols were evaluated. These were mRNA‐LNP treatment with detergent (Triton X‐100) and precipitation by organic solvent (isopropanol) [25, 33, 36]. After mRNA‐LNP samples were disrupted by Triton X‐100, fluorescence intensity of the main peak was significantly impacted, even at low detergent concentrations (Figure 2A), which also resulted in lower integrity measurements and increased variability compared to isopropanol treatment (Table 2). These observations were recapitulated with an in‐house formulated FLuc‐LNP sample (Figure 2B), suggesting that isopropanol precipitation removal of non‐precipitating buffer components improves assay sensitivity. However, a comparison between the disrupted FLuc‐LNP and untreated FLuc mRNA at matching concentrations reveal that even isopropanol precipitation disruption can incur signal or sample loss on analysis (Figure 2C), suggesting that comparisons between drug substances and drug products must be carefully interpreted. This can be only partially attributed to the lipid content of mRNA‐LNP samples, as mock disruptions with RNA ladder and FLuc mRNA only were less impacted overall, though 10% Triton X‐100 addition still had the greatest effect (Figure 2D and Figure S3). Because these mock disruptions were performed on naked RNA, any detergent interactions with lipids that could further interfere with the fluorescence would not be observed. Interestingly, when control RNA and mRNA‐LNP samples were disrupted and analyzed by agarose gel electrophoresis, high detergent concentrations also appeared to disrupt electrophoretic mobility, causing faint bands and streakiness (Figure S4). Altogether, these results would suggest that detergent addition interferes with electrophoretic mobility, which is exacerbated by the presence of an LNP (Figure S5).

FIGURE 2.

CGE‐LIF analysis of (A) a commercial mRNA‐LNP sample disrupted with isopropanol precipitation and varying detergent concentrations, (B) FLuc‐LNP control samples disrupted with isopropanol and varying detergent concentrations, (C) non‐encapsulated FLuc mRNA compared to FLuc‐LNP disrupted with isopropanol precipitation, and (D) RNA ladder mock disrupted with the same disruption protocols. For ease of comparison, electropherograms have been normalized to 100 on the y‐axis. Electropherograms have been stacked in (D) with duplicate traces of “untreated” samples for clarity in visualization.

TABLE 2.

Corrected peak area composition (%) and standard deviation (SD) of capillary gel electrophoresis with laser‐induced fluorescence (CGE‐LIF) for messenger ribonucleic acid‐lipid nanoparticle (mRNA‐LNP) samples after varying disruption protocols.

Disruption method	Shoulder	Main peak	Late‐migrating smear
Isopropanol precipitation	40.01% ± 2.90%	57.12% ± 2.67%	2.87% ± 0.32%
1% Triton X‐100	49.74% ± 4.77%	46.32% ± 4.06%	3.94% ± 1.58%
10% Triton X‐100	86.00% ± 1.91%	11.31% ± 2.36%	2.69% ± 1.26%

Because isopropanol precipitation least impacted the mRNA‐LNP fluorescence signal and did not appear to interfere significantly with the electrophoretic mobility, this method was chosen for further evaluation of conditions between mRNA‐LNP analysis conditions.

3.3. Different Denaturants Are Associated With Different mRNA‐LNP Impurities Independent of Dye Binding

Because denaturants were identified as critical method parameters affecting peak signal, resolution, and variability, we decided to systematically evaluate how in‐gel urea and in‐sample formamide affected integrity analyses given their prevalence in CGE‐LIF methods. Notably, both additives interfered with the signal‐to‐noise ratio, which could affect accurate integration, particularly in the shoulder [43]. Disproportionate binding to the SYBR Green II dye in the gel buffer was a particular concern because previous reports have suggested an impact on dye binding from denaturation or fragment nucleotide composition [42]. Given the interdependence of these variables and the challenge of defining separation between the main peak and shoulder, we decided to design a full factorial experiment to systematically examine relationships among dye, urea, and formamide (Table 3).

TABLE 3.

Full factorial design with three factors at two levels each.

No.	[Urea] (M)	Formamide presence	[Dye]
1	Low (2)	Yes	Low (0.005%)
2	Low (2)	Yes	High (0.04%)
3	Low (2)	No	Low (0.005%)
4	Low (2)	No	High (0.04%)
5	High (6)	Yes	Low (0.005%)
6	High (6)	Yes	High (0.04%)
7	High (6)	No	Low (0.005%)
8	High (6)	No	High (0.04%)

Two sequences were designed with randomized orders and duplicate injections (Tables S2 and S3) for an overall analysis time of <7 h to minimize 4°C autosampler storage. Storage times >7 h negatively impacted signal intensity (Figure S6).

Ideally, a comparable integration of peaks between electropherograms would include a correction for velocity broadening seen at longer migration times [44]. However, one immediately evident challenge was the markedly differing shoulder profiles, where apparent peaks were likely masking many analytes under the curve (Figure S7). Because automatic integration software typically calculates corrected peak areas using the migration time of peak apexes [37], the resulting corrections could introduce biases and amplify variance in the data (Figure S8). In order to represent the true profile shape, the three segments were thus integrated without correction (Figure 3A). In addition, the main peak width at half‐height as a measure of efficiency was also calculated. There were no time‐dependent effects over the sequence for uncorrected % main peak, % shoulder, or % LMS (Figure S9).

FIGURE 3.

(A) Integration regions of CGE‐LIF analyses showing the shoulder (yellow), the main peak (blue), and the LMS (pink). (B) Box plots of the % main peak within each condition. Descriptive statistics for each condition are summarized below the plot. (C) CGE‐LIF for results of one set and sequence in DOE design. Replicates within the sequence are overlaid in blue and black. CGE‐LIF, capillary gel electrophoresis with laser‐induced fluorescence; LMS, late‐migrating smear.

Overall, the least denaturing conditions, 3 and 4, resulted in the highest % main peak (Figure 3B). From a qualitative examination, low urea conditions resulted in shorter migration times, broadened main peaks, and larger shoulder areas (Figure 3C), suggesting that urea content had a larger effect on peak profile than velocity broadening. Compression of the shoulder into the main peak also appeared to contribute to increased material under the main peak, inflating integrity values.

Variability was highest in conditions 1 and 2, which both contained formamide. Although formamide conditions 5 and 6 had less variability, it was noted that most outliers were amongst these four conditions (Supporting Information section), which reflected prior observations that in‐sample formamide negatively impacted run‐to‐run reproducibility.

Several general linear models were constructed, and, surprisingly, few factors were associated with significant impact on responses (Figure S10). Interestingly, lower urea concentrations (Figure S11) and longer migration times were associated with lower relative shoulder content, whereas the absence of formamide was associated with lower relative LMS content (Figure S12). Noticeably, dye concentration was absent in all models despite a notable increase in signal intensity in high dye conditions, suggesting that any loss of signal or resolution was not compensated for by the addition of dye. For the main peak width at half‐height, high urea concentrations were associated with significantly lower peak widths or improved efficiency (Figure S13). The presence of formamide was also associated with lower peak widths but at a much lower effect size.

3.4. IP‐RPLC Assays a Different Impurity Profile From CGE‐LIF

Finally, the performance of CGE‐LIF was compared to IP‐RPLC, an orthogonal technique for mRNA sizing. Analysis of the same RNA markers from Figure 1 showed generally lower resolution and sizing accuracy by IP‐RPLC (Figure 4A). When RNA ladder linearity was assessed for both techniques, the migration pattern was biphasic [40] with CGE‐LIF offering better separation across the entire fragment range. The resolution for IP‐RPLC was overall low, but particularly for higher molecular weight (>2.5 kb) fragments.

FIGURE 4.

IP‐RPLC analyses of (A) ladder spiked with a 1.2 kb IS (black) compared to a control RNA spiked with the same 1.2 kb IS (red) and (B) mRNA‐LNP samples treated with different disruption protocols. RNA, ribonucleic acid.

For mRNA‐LNPs, similar observations were made. Although a main peak was evident by IP‐RPLC, a clear shoulder could not be clearly delineated from the main peak. Integration values also differed (Table 4). Integrity by IP‐RPLC was consistently higher than CGE‐LIF, likely due to its lower resolving power and broader peaks obscuring similarly sized fragments within the main peak.

TABLE 4.

Peak area composition (%) by ion‐pair reversed‐phase liquid chromatography (IP‐RPLC) for messenger ribonucleic acid‐lipid nanoparticle (mRNA‐LNP) samples after different disruption protocols.

Disruption method	Shoulder	Main peak	SD
Isopropanol precipitation	19.86%	80.14%	2.05%
1% Triton X‐100	18.74%	81.26%	2.54%
10% Triton X‐100	N/A	N/A	N/A

Abbreviation: SD, standard deviation.

However, IP‐RPLC analysis of mRNA‐LNP samples disrupted with both 1% and 10% Triton X‐100 clearly revealed the interference of Triton X‐100 peaks with the main mRNA peak (Figure 4B), demonstrating that uncharged excipients or impurities can be efficiently detected. In comparison, in CGE‐LIF, these same samples only resulted in diminished signal intensity and altered mRNA peak shape, which could lead to different conclusions of sample purity or integrity. Method accuracy would need to be determined with a well‐characterized reference standard.

4. Discussion

The quality assessment of mRNA‐LNP biotherapeutics needs to be robust, reproducible, and, most importantly, interpretable. Although RNA analysis by CGE has long been a staple of research settings, we show that some conventional knowledge should be reconsidered in the context of mRNA biotherapeutic analysis. Indeed, many modern CGE method parameters have been derived from research work with slab gel electrophoresis and DNA [39], which can be problematic when considering problems such as preserving chemically unique impurities or adequately separating heterogeneous mixtures of fragments.

Denaturants, as an example, are often considered interchangeable in mechanism and head‐to‐head comparisons are often not performed. However, the mechanisms of denaturation for urea and formamide are not well understood, and they differ in chemical and physical properties, such as viscosity [45], which can impact the extent of denaturation and method performance. Although both urea and formamide have been reported to destabilize RNA secondary structure by hydrogen‐bonding [46, 47, 48], mechanistic studies of formamide denaturation are far less explored. In this study, we show that the usage and nature of the denaturant should be under consideration in biotherapeutic method development. First, formamide as a sample diluent does not effectively denature RNA fragments during CGE‐LIF analysis. Second, in‐gel urea and in‐gel formamide result in different peak profiles, suggesting that the nature of the diluent could reveal additional quality information. Indeed, one study by Stellwagen et al. showed that urea alone was unable to fully denature stable secondary structures in DNA [49]. Thus, assessing the extent of denaturation with a well‐characterized reference standard could be informative. Finally, both in‐gel formamide and urea decrease signal intensity, which has been reasonably attributed to interference of dye binding to RNA [36]. In effect, because SYBR Green II is an intercalating dye [50], its binding to RNA could be biased based on molecular structure [42]. However, it is worth noting that both the addition of sieving polymer and in‐sample formamide also reduced signal intensity without affecting denaturation, suggesting multiple mechanisms at play, including viscosity effects.

Encapsulation by LNPs is a critical aspect of biotherapeutic mRNA analysis. As a result, LNP disruption is often required for product analysis [33]. Many bioanalytical studies, however, focus on naked mRNA analysis, which can overlook complex matrix interactions. To maintain relevance to biotherapeutics, we used commercial mRNA‐LNP products to show that LNP disruption does have an impact on the quality profile by CGE‐LIF. Common detergent disruption appeared to interfere with signal intensity and RNA mobility, negatively impacting integrity. Thus, care should be taken when interpreting analyses between detergent‐disrupted mRNA‐LNP samples and naked mRNA samples, such as between mRNA‐LNP drug products and mRNA drug substances. Analysis of in‐process intermediates should include relevant matrix controls and appropriate reference standards. Removing non‐precipitating buffer components would improve assay sensitivity.

Defining the standards for comparative analyses of CGE‐LIF profiles is challenging, especially in the shoulder region where a complexity of analytes can be masked in the area under the curve. Traditionally, peak areas are corrected based on the migration time of the peak; however, approaches based on identifying migration time by the peak apex may introduce variability or bias, especially when comparing “smoother” shoulders with fewer peaks to “noisier” shoulders. It may be beneficial to explore standard integration practices or the impact of alternative correction approaches such as weighted integrations. Resolution in CGE‐LIF analysis of mRNA is another criteria that can be difficult to define because the shoulder contains a mixture of fragments not fully separated from the main peak. Traditional resolution metrics, such as the separation of two well‐defined peaks [51], would not be applicable. Systematic approaches, such as full factorial design of experiments, allow for the comprehensive and quantitative evaluation of all factor combinations and their responses. In this study, we showed that reducing urea concentration (i.e., Conditions 1–4) would compress the peak profile, incorporating more shoulder into the main peak, thereby reducing resolution and artificially increasing integrity. Interestingly, different denaturants were correlated with different parts of the profile, with urea impacting the shoulder and formamide impacting the late‐migrating smear. Dye was not a significant factor, suggesting that any impact to dye binding by denaturants was proportional amongst all fragments and increasing dye concentrations would compensate for decreased sensitivity. Altogether, sample conditions with higher urea concentrations, no formamide, and higher dye concentrations (ex. Condition 8) would minimize assay variability, improve resolution or efficiency of the main peak, and increase detection sensitivity of the shoulder.

In comparison to IP‐RPLC, resolution in CGE‐LIF is higher, particularly for longer fragments (>2.5 kb), and mRNA sizing by IP‐RPLC should be considered broad estimates unless otherwise demonstrated. Our study did demonstrate one advantage of IP‐RPLC, which is the unbiased detection of additives in the sample matrix. Because IP‐RPLC functions by injecting a sample plug under pressure, all material, including excipients and impurities, will pass through the detector. In contrast, additives that cannot be electrophoresed will not be detected in CGE‐LIF. Modern LC systems are also highly reliable and accessible, making IP‐RPLC attractive for routine biopharmaceutical environments.

In summary, some method considerations can be made based on the results of this study. Given the sensitivity of CGE‐LIF analysis to gel buffer composition and sample matrix, incorporating an acceptance criterion that unambiguously defines resolution would be beneficial to consistent assay performance. For example, monitoring the separation of two RNA fragments close in size to the mRNA payload—either spiked or in a separate injection—could ensure that changes in the spread of the shoulder peak are attributed to the extent of product degradation and not reagent variability. Consequently, denaturants are critical method parameters, whose variation should be assessed in robustness studies. Changes in CGE‐LIF fluorescence intensity can be subject to assay variability and instrument calibration but may also be indicative of sample matrix changes and should be monitored. Finally, the incorporation of both CGE‐LIF and IP‐RPLC in an analytical testing plan would aid in controlling both known and unknown impurities.

5. Concluding Remarks

With commercial kits and user‐friendly instruments, method standardization across laboratories has never been easier. Indeed, common errors such as neglecting to adjust for the volume of urea salt in a high molarity solution can be easily avoided with pre‐purchased gel solutions. However, commercial products often contain proprietary formulations, which can overly simplify processes where complexities provide rich information. This is especially important for novel drug modalities where unexpected peaks and behaviors could be indicative of unknown quality characteristics. In this study, we reveal how a set of method parameters and their interactions impact mRNA quality profiles, systematically shedding light on knowledge gaps in the current understanding of mRNA‐LNP analysis, particularly for larger fragments relevant to biotherapeutics. We hope this work will provide a foundation for further studies in designing informative reference standards and mechanistic studies into large molecule separation with both CGE‐LIF and IP‐RPLC.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Supporting Information

Supporting Information

Data Availability Statement

Data are available in the article's Supporting Information section.