1. Introduction
The first use of mRNA to deliver a target protein in vivo was published in 1990 [
1] and the use as a therapeutic in 1992 [
2]. In the following decades, mRNA-based therapeutics were implemented for use in genetic engineering [
3], regenerative medicine [
4], protein replacement [
5], cancer immunotherapy [
6,
7], and vaccines [
8,
9]. These applications have been reviewed in refs. [
10,
11]. Perhaps most famously, lipid-encapsulated mRNA was utilized in the recent successful development of SARS-CoV-2 vaccines [
12,
13]. The use of mRNA vaccines is particularly attractive because of the ability to generate on-demand transcript sequences using platformed manufacturing methods to enable rapid response to emergent needs [
14]. The recent increase in mRNA vaccine research and development requires the adoption and implementation of new manufacturing techniques, as well as the proper control mechanisms to monitor them [
15].
mRNA vaccines encoding an antigen of interest are synthesized by in vitro transcription (IVT) using a linearized DNA template and an RNA polymerase (RNAP) such as T7, SP6, or T3 [
15]. T7 RNAP is most commonly used due to its high mRNA yield and fidelity, but T7 RNAP may also produce multiple IVT reaction byproducts, including immunostimulatory double-stranded RNA (dsRNA) [
16]. In contrast to the desired immunostimulatory effect of mRNA, which induces a strong adaptive response via antigen-presenting cell activation [
14], dsRNA can inhibit ribosomal protein translation, resulting in a 10- to 1000-fold decrease in target protein production [
17]. The dsRNA byproducts can generally be classified as short dsRNA, formed from the random annealing of abortive transcripts, and full-length dsRNA, formed from either loop-back elongation or spurious transcription from the non-coding strand of the DNA template [
17].
dsRNA is an innate immune system activator which activates cytosolic receptors such as RIG-I, MDA5, and PKR and endosomal receptor TOR3 [
14]. Collectively, the immune system activation creates efficacy concerns due to decreased ribosome expression and toxicity concerns due to cytokine secretion. Different aspects of dsRNA contribute to unique innate immune responses; RIG-I recognizes the 5′PPP end and is thus more sensitive to higher concentrations of short dsRNA, whereas MDA5 involves filament formation driven by longer dsRNA [
14]. To our knowledge, no regulatory pharmacopeia has provided target acceptable levels of dsRNA in mRNA vaccines. Existing guidance indicates that dsRNA impurities should be removed to the extent feasible, and any residual dsRNA should be well characterized and controlled [
18]. A reduction in dsRNA can be achieved by the optimization of the IVT reaction, such as the use of N1-Me-pseudo UTP, codon optimization, and lowering the Mg
2+ concentration [
19], and implementing purification strategies, such as ion-pairing reversed-phase HPLC (IP-RPHPLC) [
17] and cellulose purification [
20]. To implement these methods to minimize residual dsRNA in mRNA manufacturing, a sensitive and quantitative dsRNA analytical method is necessary.
The strategies to quantitate dsRNA involve cell-based assays, gel electrophoresis assays, chromatography-based methods, and immunochemistry assays [
20,
21]. Cell-based assays offer a direct measurement of innate immune activation but are complicated, laborious, and expensive. Gel electrophoresis-based assays can be used for characterizing dsRNA but are only semi-quantitative and lack sensitivity. Quantitation by IP-RPHPLC involves the digestion of mRNA followed by the successive resolution of intact or degraded dsRNA and offers the ability to both characterize and quantitate dsRNA [
21]. However, this method is limited by high cost and high level of expertise. The heterogeneity of residual dsRNA and presence of mRNA at orders of magnitude higher concentration than dsRNA precludes some methods of measuring nucleotides such as qPCR and next-generation sequencing.
A 1991 publication by Schönborn et al. described four antibodies (J2, J5, K1, and K2) that specifically bind dsRNA without binding single-stranded RNA and are highly sensitive for a wide range of sequences that are 40 bp in length [
22]. These antibodies have been used in several quantitative immunochemical assays such as the dot blot assay, [
20] sandwich ELISA [
16,
23], and lateral flow immunoassay [
24]. The sandwich ELISAs offer an appealing strategy for dsRNA quantitation because they offer an excellent balance of sensitivity, throughput, and simplicity and can be implemented using standard equipment found in most labs. Previous publications have presented data generated using a dsRNA sandwich ELISA but do not provide a detailed description of the development and final protocol for the assay [
16,
23]. To our knowledge, the full end-to-end development of an automated dsRNA sandwich ELISA has not been described in the literature. The objective of this manuscript is to provide a template for implementing a dsRNA sandwich ELISA to support mRNA vaccine manufacturing process development using commercially available J2 and K2 antibodies along with other readily available materials and reagents. To this end, we describe the development and application of an ELISA for quantifying residual dsRNA in non-adjuvanted mRNA vaccine samples ranging from crude IVT reaction products to fully purified bulk drug substance, as well as “lessons-learned” during assay development. We describe the identification and mitigation of assay interference by common IVT reaction components and the optimization of ELISA reagents, report the assay performance, provide examples of the application of the dsRNA, and finally demonstrate success in automating the ELISA in a high-throughput format (
Supplementary Figure S1).
2. Materials and Methods
2.1. Materials
Tris (hydroxymethyl)aminomethane, Sodium Chloride (NaCl), and Ethylenediaminetetraacetic acid (EDTA) were purchased from Invitrogen (Waltham, MA, USA). Sodium Hydroxide was purchased from Fisher Chemical™ (Pittsburgh, PA, USA). K2 and J2 dsRNA ELISA antibodies and polyinosinic–polycytidylic acid (poly I:C) positive control were purchased from Exalpha (Shirley, MA, USA), 142 bp dsRNA reference standard from Jena Bioscience (Jena, Thuringia, Germany), sample diluent from Cygnus (Southport, NC, USA), and secondary antibody from Jackson ImmunoResearch Laboratories (West Grove, PA, USA). Assay diluent was purchased from Teknova (Hollister, CA, USA), phosphate buffer saline from Cytiva (Marlborough, MA, USA), Tween 20 from MilliporeSigma (Burlington, MA, USA), and 4-methylumbelliferyl phosphate (4-MUP) from Virolabs (Chantilly, VA, USA). Black 96-well MaxiSorp® ELISA plates were purchased from Nunc (Roskilde, Denmark).
2.2. Quantitation of dsRNA
To quantitate residual dsRNA in mRNA samples, an ELISA was developed utilizing a Tecan EVO liquid handler configured with a Tecan Infinite F500 plate reader. Unless otherwise noted, assay diluent (Teknova P1391; 10 mM phosphate, 154 mM NaCl, 1% BSA) was used throughout. All incubation steps listed below were performed with plates left uncovered, without shaking at room temperature. Washing was performed with 3 wash cycles of 400 µL wash buffer (10 mM phosphate 154 mM NaCl, 0.05% Tween-20 pH 7.0) using either a Tecan HydroSpeed or BioTek EL406 plate washer. Unless otherwise stated, 96-well Black Maxisorp plates (Nunc 437111) were coated with 2 µg/mL mouse anti-dsRNA J2 IgG (Exalpha 10010500) in coating buffer (Hyclone SH29882925; 10 mM phosphate, 154 mM NaCl pH 7.0) per well for 1 h. Coating antibody was then aspirated prior to blocking with assay diluent for at least 15 min and no longer than 8 h. Blocked plates were washed as described above. Unless noted otherwise, mRNA samples were pre-diluted 2-fold in Sample Dilution Buffer (Cygnus I028) prior to serial diluting in Sample Dilution Buffer. After a 2 h sample incubation, plates were washed and incubated with mouse anti-dsRNA K2 IgM (Exalpha 10030010) for 1 h; unless otherwise stated, a 15-fold dilution of K2 IgM was used throughout. Plates were washed and then incubated with alkaline phosphatase (AP)-conjugated goat anti-mouse IgM (Jackson ImmunoResearch 115-056-020) for 1 h; unless otherwise stated, 0.3 µg/mL of conjugate antibody was used throughout. Samples were resolved using 4-MUP (Virolabs XPHOS-100) as the substrate, and the signal was detected by fluorescence at 360 nm excitation and 465 nm emission using a Tecan Infinite F500. The addition of capture antibody, blocking reagent, detection antibody, AP-conjugated secondary antibody, and 4-MUP substrate was performed by a single-step transfer utilizing the 96-well multichannel arm on the Tecan EVO. Eight 2-fold serial dilutions of mRNA samples were performed using a single column of tips on the Tecan EVO 96-well multichannel arm. Determination of dsRNA concentration in mRNA samples was calculated using parameters determined from a four-parameter logistic (4PL) fit of a standard curve generated with 142 bp dsRNA purchased from Jena Bioscience (RNT-sci-10080100). To correct for differences in total mRNA concentration of in-process samples, the concentration of dsRNA is normalized to the total mRNA concentration and reported as the percentage of dsRNA (w/w).
2.3. Preparation and Purification of mRNA Containing Samples
mRNA vaccine constructs were transcribed from linear DNA templates with encoded poly(A) tails following a standard IVT reaction protocol that uses a combination of nucleotide triphosphates (NTPs), T7 RNAP, pyrophosphatase, RNase inhibitor, and a Tris buffer system containing DTT, MgCl
2, and CleanCap AG, as previously described [
25]. The IVT reaction was incubated for 3–4 h at 37 °C followed by enzymatic digestion of the DNA template with DNase I at 37 °C for an additional 30–60 min. An optional enzyme removal step was included in selected samples by proteinase K/SDS addition after enzymatic digestion of the DNA template step. After the IVT reaction and digest steps, enzyme activity was quenched by the addition of 50 mM EDTA.
Samples that were not purified are referred to as IVT samples. Selected samples were purified using an mRNA purification process involving oligo dT chromatography, tangential flow filtration for concentration and buffer exchange into an mRNA storage buffer, and bioburden reduction filtration [
25] to generate bulk drug substance (BDS) samples. IVT samples, purification process intermediate samples, and BDS samples were analyzed by the dsRNA assay. Two samples were subjected to additional cellulose purification (
Section 3.4).
The total concentration of RNA was determined using a Nanodrop 8000 Spectrophotometer (ThermoScientific, Waltham, MA, USA) using ThermoScientific software V2.3.3. The instrument was blanked with subsequent measurements of 2 µL RNase-free water followed by 2 µL sample-specific matrix buffer. After blanking, the absorbance at 260 nm of the 2 µL sample was measured and converted into RNA concentration using an OD equivalence of 40.
The impact of selected IVT reaction and purification process parameters on mRNA yield and purity was investigated. For example, a concentrated stock of T7 RNAP was used in one IVT reaction condition resulting in additional matrix carry over relative to the standard usage practice. The additional matrix carried over into the IVT reaction is differentiated from the standard IVT reaction conditions below by referring to the former as “IVT Buffer 2” and the latter as “IVT Buffer 1”. As another iteration of the IVT reaction, several additives (numbered 1 to 3) were also included in the IVT reaction to evaluate the robustness of the IVT reaction and effect on dsRNA production by the T7 RNA polymerase when these additives were included. Several iterations were collectively incorporated into a procedure referred to below as “procedure 2” to distinguish against an earlier utilized “procedure 1”. These process changes may include a change in manufacturing facility, change in reagent sources, and an optimization of purification parameters within the general mRNA purification strategy outlined above.
4. Discussion
Commercially available dsRNA ELISA kits are available which could potentially support dsRNA quantitation at a small scale; however, a sandwich ELISA using bulk reagents is required to handle sample testing at a large scale. The automated ELISA described in this paper can support the testing of 72 samples per day (with a full titration curve/sample) with only a half-day analyst obligation. Within the past year, an in-house dsRNA sandwich ELISA has been utilized to analyze residual dsRNA in over 500 mRNA samples. Supporting 72 samples per day with a full titration curve using a typical five-plate commercial kit would require opening and combining three kits which would impart additional variability and analyst burden from preparing reagents from multiple sources. Furthermore, the cost per sample of testing with a kit is over twice as much as testing with the in-house dsRNA ELISA described in this paper. Two commercially available ELISA kits were evaluated in our lab and were seen to have unsatisfactorily high variability (
Supplementary Figure S6). This variability was suspected to be due to an inadequate amount of coating antibody included with the kit. More pertinently, the format of a kit precludes reagent optimization, which was found to be critical in the development of the dsRNA ELISA using bulk reagents. Furthermore, using a kit constrains the number of samples which can be tested within a study, as using multiple kits could contribute to variability. When compared to commercially available dsRNA ELISA kits, an in-house sandwich ELISA can assay a greater volume of samples, with reduced cost, reduced analyst time obligation, and reduced material waste.
Seven unique mRNA constructs have been assayed by the dsRNA sandwich ELISA with nucleotide lengths that differ by up to 2-fold (
Supplementary Table S1). Among these constructs, some are coding sequence variants of the same backbone, whereas others have different coding sequences and untranslated regions. The dsRNA sandwich ELISA has been successfully implemented toward the quantitation of residual dsRNA in partially purified mRNA samples across a range of manufacturing steps including oligo dT chromatography, tangential flow filtration, membrane filtration, and cellulose chromatography. mRNA samples across these manufacturing steps contain a range of background buffer components, including citrate buffer, Tris buffer, 0–600 mM NaCl, and pH values from 6.4 to 7.2. Furthermore, the dsRNA sandwich ELISA has been implemented towards testing unpurified IVT reaction products in a complicated matrix background with a wide range of components including various concentrations of NTPs, DNA template, RNA polymerase, polymerase inhibitor, MgCl
2, EDTA, DTT, and assorted IVT additives. In each of these conditions, the concentration of residual dsRNA was determined without significant interference from the IVT reaction components. The only IVT reaction components that were observed to completely inhibit meaningful quantitation were the enzyme removal components, proteinase K and SDS (
Figure 1B). The wide range of mRNA constructs and matrix backgrounds in the samples tested speaks to the broad applicability and robustness of the dsRNA sandwich ELISA.
Automating the dsRNA ELISA reduces the time burden of the assay, improves the assay reproducibility, and improves the throughput of continuous assay support. However, establishing an automated ELISA requires a high initial investment in time and expertise. A semi-automated or manual ELISA can provide a lower barrier of entry for use in the quantitation of residual dsRNA in mRNA samples. In this study, a semi-automated ELISA was implemented using resources and protocols readily available at the time, while a fully automated ELISA was being developed. Coating plates with 2 µg/mL of J2 antibody was sufficient for the semi-automated ELISA, whereas the fully automated ELISA required 4 µg/mL (
Supplementary Figure S5). This difference between the semi-automated and fully automated ELISA was partially attributed to the change from the BioTek plate washer used in the semi-automated ELISA to the Tecan HydroSpeed plate washer used for the fully automated ELISA. The BioTek plate washer was equipped with a manifold that contained side washing pins. This manifold was selected based on a historic use of the equipment for washing cell-coated plates and was not anticipated to be a critical feature of the assay. It is thought that the downward washing manifold equipped on the HydroSpeed is harsher and partially removes J2 antibody by sheer force, whereas the side washing allows for a gentler wash that does not disrupt the J2 coating antibody. By oversaturating the plates with J2 coating antibody, the partial removal of J2 antibody from washing does not reduce the remaining J2 below the assay saturation point. The plate washer, manifold, and wash settings are likely to be variable from lab to lab. The washer should be a major focus of attention for new labs adopting the ELISA, likely requiring optimization for each individual system.
During assay optimization, it was discovered that the mouse anti-dsRNA K2 monoclonal IgM detection antibody could be diluted 15-fold without a noticeable depreciation in assay quality. A large dilution of K2 was logistically advantageous as the K2 IgM monoclonal antibody is provided as a hybridoma supernatant of unspecified concentration, shipped in 10 mL quantities. A typical assay could utilize up to 140 mL in a single 12-plate run. However, it should be noted that the optimization and utilization of K2 detection antibody is based on a single lot of K2 antibody. Variations among lots and suppliers were not evaluated as a part of this study.
One objective of this publication is to provide a template which could be easily adopted by other laboratories. If equipment availability and/or material cost preclude the adoption of the fluorescence-based reagents used in this study, the assay could be performed with horseradish peroxidase-conjugated goat anti-mouse IgM (Jackson Immunoresearch 115-036-075) as the conjugate antibody and resolved with 3,3′,5,5′-Tetramethylbenzidine (TMB) substrate (Sigma-Aldrich T0440). A commercially available polyinosinic–polycytidylic acid (poly I:C) reagent was evaluated as a potential alternative for either the reference standard or positive control for the dsRNA ELISA. Poly I:C is structurally similar to double-stranded RNA and is commonly used as a synthetic analog to dsRNA in immune research. It was observed that the dose–response curve generated by the poly I:C lacked parallelism with the intended in-house mRNA samples, resulting in a very high level of uncertainty in dsRNA concentration (
Supplementary Figure S7). For this reason, the poly I:C reagent is not recommended for use with the dsRNA ELISA when performed as described in this study.
The performance of the developed semi-automated ELISA was analyzed in terms of repeatability, precision, sensitivity, selectivity, and accuracy. Overall, the performance was satisfactory for its intended application. Increasing replicates improves precision at the cost of throughput. This decision should be made at a phase-appropriate level and with understanding of the precision required for informed decision making. The spike recovery of the 142 bp dsRNA reference standard indicated that the assay was linear from 400 ng/mL to 10,000 ng/mL in the presence of 1 mg/mL mRNA which equates to the detection of 40–1000 ng dsRNA per 100 µg mRNA. The lower limit of quantitation determined by this approach was constrained by the mRNA samples used to provide the 1 mg/mL background; the true limit of quantitation is likely far lower than presented. Based on a simple dilution of 142 bp dsRNA reference standard, we see that the ELISA can detect as low as 1 ng dsRNA suggesting a very high level of sensitivity. The dsRNA sandwich ELISA was successfully utilized to measure residual dsRNA ranging from 1% to 0.01% indicating a high level of both selectivity and sensitivity.
Two approaches were utilized to approximate the accuracy of the dsRNA ELISA described in this study: spike recovery of 142 bp dsRNA RS and differentiation of the apparent residual dsRNA concentration in a “high” and “low” dsRNA sample. The spike recovery was measured to be accurate within 30%; the deviation from the expected value is thought to be from variability in pipetting the low volumes implemented for this study. Without assurance that the affinities of the J2 and K2 antibody for dsRNA are fully sequence- and length-independent, the spike recovery of the 142 bp dsRNA RS can only approximate accuracy. Similar to the antibodies used in whole-cell protein assays [
26], the J2 and K2 antibodies do not target specific species but instead measure the aggregate of all dsRNA present. The 142 bp dsRNA is a uniform, well-defined sequence, whereas the dsRNA in mRNA samples is likely a heterogenous mixture of various lengths and uncertain sequence. An uncharacterized bias of J2 and K2 binding to dsRNA of a particular sequence or length could skew the apparent concentration as determined by the sandwich ELISA. Furthermore, it is known that the J2 antibody only recognizes sequences greater than 40 bp [
22]; the values presented in this paper could be an underestimate of the total amount of dsRNA if the predominant species are short fragments of <40 bp. The detection of a 20-fold increase in residual dsRNA in a mRNA sample produced using unmodified UTP relative to an mRNA sample produced using N1-Me-pseudo UTP gives credence to the accuracy of trending performed using the dsRNA ELISA. This is in agreement with the literature reports of a significant reduction in dsRNA using a N1-Me-pseudo UTP substitution for unmodified UTP [
19].
To better demonstrate the accuracy of the dsRNA ELISA when quantitating residual dsRNA in mRNA samples, it is imperative that the assay be investigated for potential affinity differences against mRNA of different sequences and lengths. The use of the uniform 142 bp dsRNA is only suitable for accurately quantitating dsRNA of unknown composition if the assay is shown to be reasonably independent of sequence and length. Otherwise, a bespoke reference standard, informed by the careful characterization of the residual dsRNA in an mRNA product of interest, is required for each individual project. Furthermore, the confirmation of residual dsRNA by an orthogonal method, such as ultraviolet (UV) absorbance spectroscopy following isolation by reversed-phased HPLC [
21], would improve confidence in the accuracy of the dsRNA ELISA. A representative dsRNA RS could be produced for each sequence by synthesizing and annealing the complementary RNA strand. However, this increases the complexity to qualify a new RS for each new sequence evaluated.