Process and analytical strategies for the safe production of mRNA vaccines and therapeutics

Studies proposed two main types of dsRNA molecules (Fig. 3) [12, 14]. The first type is 3'-loopback dsRNA, where a single RNA molecule folds back and hybridizes with itself, forming extended regions of intramolecular base pairing (Fig. 3A). The second type is sense–antisense dsRNA, in which two complementary RNA molecules bind to each other to form the double-stranded structure (Fig. 3B).

The exact mechanism of dsRNA formation during IVT is not yet fully understood. However, one study hypothesized three possible reaction pathways during IVT. The first, termed the "on-pathway", occurs early in the IVT process when T7 RNA polymerase binds to the promoter and initiates transcription, producing canonical run-off transcripts. The second, the "off-pathway", involves previously synthesized RNA hybridizing with itself or acting as a template for unintended RNA extension, potentially leading to the formation of 3′-loopback dsRNA. The third, named as "pseudo-on pathway", is sequence-dependent and may occur when specific motifs at the 3′ end of the template DNA facilitate the synthesis of antisense transcripts, which can then hybridize with the sense strand to form sense–antisense dsRNA [12].

The exact mechanism on why 3'-extended mRNA is being formed is unknown, but a likely contributing factor is a loose 3'-end that lacks stable secondary structure, allowing the polymerase to use it as a template in place of the DNA for further extension [52]. This self-extension can result in intramolecular (cis) duplexes when the RNA folds back on itself, or intermolecular (trans) duplexes when it anneals to another RNA molecule with complementary sequences [12, 14, 52 –56].

This 3'-loopback dsRNA has been identified as the predominant byproduct produced in IVT reactions [12, 53]. The formation of 3'-loopback dsRNA has been shown to be sequence-dependent with homopolymeric runs prone near the 3'terminus being particularly prone to self-priming and dsRNA formation [52, 53].

The second type of unwanted dsRNA formation arises from transcription initiation on the non-template strand of DNA, producing antisense RNA that pairs with the sense RNA transcript [12, 14]. This promoter-independent antisense transcription has been observed across a wide range of DNA sequences and appears to depend on specific motifs at the 3'-end of the DNA template [14]. These sequences can cause T7 RNA polymerase to initiate transcription on the non-template strand, generating antisense RNA and leading to sense–antisense dsRNA formation.

Importantly, the inclusion of a poly(T) sequence in the DNA template has been shown to reduce the formation of antisense dsRNA by-products, though it does not affect 3'-loopback dsRNA formation [12]. The precise mechanism remains under investigation. Wu et al., propose that certain 3'-terminal sequences promote strand-switching by T7 RNA polymerase, triggering antisense transcription. In the absence of these antisense-promoting motifs, 3′ self-extension and loopback-mediated dsRNA formation tend to predominate [12].

Due to their immunological impact the translation efficiency is low if mRNA therapeutics with dsRNA contamination are used. The innate immune system detects viral RNAs, such as dsRNAs, through pattern recognition receptors (PRRs). The signature of viral RNAs is the RNA duplex structure which initiate antiviral immune response through cytosolic sensors, retinoic acid-inducible gene I (RIG-I) and melanoma differentiation-associated 5 (MDA5), the two major sensors [14, 22, 57]. If viral dsRNA is recognized by RIG-I or MDA5 it activates the antiviral signalling pathways followed by the production of interferons type I and III (IFNs). Both receptors are similar in their domain structures and amino acid sequences but have different substrate preferences and mode of filament formation. MDA5 binds long dsRNA in a sequence dependant manner by interacting with the backbone of the RNA duplex. In contrast, RIG-1 recognizes the 5′-triphosphate (5′ppp) or 5′-diphosphate (5′pp) moieties at the ends of short dsRNA or blunt-ended duplexes [57].

Because T7 RNA polymerase transcripts naturally carry 5′ppp groups, they are inherently immunogenic and capable of activating RIG-I [14]. But removal of 5'ppp does not completely suppress the immunogenicity which can be explained by MDA5 activation which is independent of 5′ppp and responds primarily to the presence of dsRNA itself [14, 58].

Due to the triggering of the innate immune system dsRNA contaminations in mRNA therapeutics can lead to severe side-effects in the patients' body. Also, in vitro and in vivo studies can be affected, as triggering of the immune response shuts down the protein synthesis pathway and hence leads to lower translation efficiency [3, 14].

Various studies investigated how the formation of dsRNA can be reduced during IVT reaction (Table 1), which are described in the following.

The content of dsRNA impurities is commonly measured with immuno-Dot Blot assays or ELISA [66, 67]. Both techniques rely on monoclonal antibodies that specifically recognize dsRNA structures. The dot blot assay, while conceptually similar to a Western blot for proteins, is adapted for nucleic acids and is frequently used in academic research settings [21].

Native PAGE using either Sybr Gold or acridine orange can be used to visualize single- or double stranded RNA or DNA in native polyacrylamide gels.

It binds the charged backbone of nucleic acids and increases the fluorescence signal by more than 1000-fold [68]. Acridine orange (AO) is a dye that binds RNA regardless of its secondary structure but fluoresces differently depending on its secondary structure (14). With a gel scanner, the fluorescent gel images can be obtained, and AO emits green fluorescent when bound to ssRNA and red fluorescence when bound to dsRNA.

This method can be easily used to visualize and distinguish between ssRNA and dsRNA after IVT reaction, but it gives no information about the other by-products. Native PAGE is time-consuming, semi-quantitative, and low-throughput, making them less suitable for process development or GMP manufacturing workflows. More efficient, high-throughput, and quantitative analytical methods are needed to support scalable mRNA production.

A very sensitive method to detect not only dsRNA but also other impurities produced by T7 RNA polymerase, such as RNA fragments, is the radioactive gel electrophoresis in which ³²P-labeld NTPs. In this approach, a DNA template is designed with G-rich 5′ region followed by C-rich sequences later in the transcript. With α-³²P-GTP abortive species and full-length RNAs can be visualized, whereas IVT reactions that incorporates α-³²P-CTP label short and long dsRNAs (Fig. 4) [40]. The sample can then be analysed by gel electrophoresis.

In addition to molecular and immunobiological methods, chromatographic techniques have become well-established tools for the analysis of mRNA impurities, offering higher precision resolution, and quantifiability, —qualities particularly advantageous in a manufacturing and quality control context. One widely used technique is ion-pair reversed-phase high-performance liquid chromatography (IP-RP-HPLC), wherein the negatively charged sugar-phosphate backbone of mRNA interacts with quaternary ammonium ion-pairing agents present in the mobile phase [2, 69, 70]. This interaction increases the hydrophobicity of the RNA molecules, allowing them to bind to the nonpolar stationary phase of the reversed-phase column [71]. Analysis is typically performed under denaturing (75 °C) or partially denaturing (50 °C) conditions to resolve structural variants and by-products with on-line UV detection at 260 nm to monitor RNA species [72]. mRNA forms intramolecular loops and hairpins via base pairing which would lead to poor resolution and broad peaks. Hence, performing this analytical technique under 50–75 °C degrees is necessary to denature these structures into a more linear form which results in sharp peaks as shown in Fig. 5.

This method offers several advantages for mRNA impurity profiling. Process-related impurities, such as residual nucleoside triphosphates (NTPs), buffer components, and capping reagents, are not retained on the reversed-phase column and are thus readily separated from the mRNA product [73].Oligonucleotides are typically resolved based on size and secondary structure in IP-RP-HPLC, with longer sequences exhibiting increased retention times. As a result, product-related impurities such as short abortive transcripts, RNA fragments, short dsRNA, and full-length dsRNA can be effectively distinguished from the target single-stranded mRNA (ssRNA) [74, 75]. Notably, dsRNA displays greater hydrophobicity than ssRNA, leading to longer retention times and allowing its identification as a distinct peak in the chromatogram (Fig. 5) This has been shown by several studies [73, 75].

In a typical IP RP-HPLC chromatogram, the initial peak corresponds to process-related impurities, such as salts, unincorporated nucleotides, and capping agents, which are not retained on the column. Subsequent smaller peaks represent short RNA fragments, abortive transcripts and short double-stranded RNA (dsRNA), which elute earlier due to their lower molecular weight, structure and reduced hydrophobicity. The main peak corresponds to the desired single-stranded mRNA (ssRNA), followed by a later-eluting peak representing dsRNA, which exhibits increased hydrophobicity and retention time.

Despite its utility, IP-RP-HPLC has certain limitations, primarily associated with the resolution capacity of the column. Short abortive transcripts often produce overlapping or poorly resolved peaks. Moreover, longer aggregates or truncated dsRNA species with similar hydrodynamic properties, such as similar structure, to ssRNA may co-elute, making them difficult to distinguish. While the commonly used ion-pairing agent, triethylammonium acetate (TEAA), adds to the operational cost, IP-RP-HPLC remains one of the most rapid and reliable methods for detecting dsRNA in both manufacturing and quality control environments.

Additional critical quality attributes include mRNA sequence fidelity, structural identity, and poly(A) tail integrity. Emerging sequencing methods, such as the VAX-seq a nanopore-based, long-read cDNA sequencing protocol, enable comprehensive analysis of these features [76]. This approach detects mismatches introduced during IVT, characterizes poly(A) tail length and composition, and identifies degradation products and byproducts such as dsRNA. VAX-seq employs a 3' reverse adaptor for accurate poly(A) tail measurement and has shown greater reliability than direct RNA sequencing. Its low mRNA input requirement makes it suitable for routine quality control across manufacturing stages, including DNA template validation.

The production of mRNA fragments can be caused by the catalytic cycle of T7 RNA polymerase. The first step of IVT is initiation. The T7 RNA polymerase binds to the promotor forming the initiation complex (IC), which is relatively unstable. In this stage, as a part of early initiation, short abortive mRNAs are produced [40, 77].

Several studies in the past could reveal that the production process of abortive transcripts during the initiation stage, is a fundamental part of transcription. This process is named abortive cycle and has also been described for other RNA polymerases [43].

After the incorporation of about 10 nucleotides the ternary enzyme-DNA-RNA complex transits into the processive and more stable elongation complex (EC) [40]. This transition marks the end of the abortive initiation phase and the onset of productive RNA synthesis.

Martin et al. concluded that the abortive cycle is dependent on temperature and ionic strength [43]. The abortive cycling increased with increasing temperature and ionic strength but the amount of abortive transcripts varied depending on the initial sequence of the message (around 8–10 bases).

During the initial phase of transcription, two competing events can occur at each nucleotide position: the ternary enzyme–DNA–RNA complex may either dissociate, releasing a short abortive transcript, or successfully incorporate the next nucleotide. Martin et al. proposed a kinetic competition between nucleotide incorporation and dissociation of the ternary complex. Although dissociation is possible during initiation, elongation is generally kinetically favored, even in the earliest steps, and becomes strongly dominant once the complex transitions to a fully processive elongation state [43].

Approximately 90% of the transcripts that have escaped the abortive cycle successfully produce full-length transcripts of a 10,000-nucleotide template [43]. The results presented in this study suggest, that immediately after initiation the enzyme-DNA-RNA ternary complex possesses a higher probability of dissociation that after transition to a fully processive complex. Martin et al. provided three possible reasons for the transition from an unstable initiated complex to a highly processive ternary complex. The first hypothesis is that the formation of a DNA-RNA duplex stabilizes the ternary complex with increasing RNA length until the origin of the DNA-RNA duplex dissociates. Another reason could be a conformational change of the protein to an altered form with higher processivity. Other studies revealed this conformational change after synthesis of between 6 and 15 bases [43, 78]. Lastly, they proposed that nonspecific interactions between specific regions of the protein and the growing RNA transcript could stabilize the ternary complex. Assuming that the RNA-binding domain of the protein is spatially separate from its catalytic site, stabilization would not occur until the RNA chain reaches a certain length to allow stabilization interaction [43].

On a structural basis, the T7 RNA polymerase uses a specificity loop, an antiparallel β-hairpin motif with four side-chains on one side which is part of the N-terminal domain of the enzyme, to interact with the base pairs of the T7 promoter. This forms the basis for differentiation of T7 promoter and other DNA sequences [11]. Figure 6A shows the interactions between the T7 promoter and the specificity loop.

Melting the promoter duplex in order to initiate transcription is eased by the N-terminal of the T7 RNA polymerase. In this mechanism, which is called the transcription bubble [45], another complex called the intercalating hairpin is involved [11]. Due to interactions between residues of the specificity loop and the template strand in an open conformation the promoter is stabilized. It is important to note that the hydrogen-bonding interactions between the loop and the promoter bases, shown in Fig. 6A, are sequence-specific [11]. In the same study, Cheetham et al. could also create a model of the initiation complex based on the homology of T7 RNA polymerase and DNA polymerases which is shown in Fig. 6B.

This model suggested that the T7 RNA polymerase positions the DNA template strand exactly in its active site and therefore provides complementary base-pairing sites for the priming and incoming NTPs at the + 1 and + 2 positions.

Based on these findings, Cheetham et al. proposed the abortive cycling occurs due to accumulation of the template strand in the active site while the promoter duplex is still bound to the enzyme [11].

Several studies have demonstrated that a conformational change of the polymerase plays a crucial role in the transcriptional processivity. It is hypothesized that with increasing RNA length (after 3–7 nt) the RNA-DNA hybrid is 'pushing' on the N-terminal driving it to rotate around 40° which causes a structural change in the enzyme leading to weaker promoter contacts. This triggers the promoter release and forms the RNA exit channel. However, the enzyme must push back on the RNA-DNA hybrid which leads to destabilization of the latter and results in the abortive loss of the RNA [42, 50, 79 –81].

After 9 to 12 nt are synthesized a drastic 220° rotation of the enzyme occurs which causes a release of the upstream promoter interactions resulting in the transition of T7 RNA polymerase into the stable elongation complex. The abortive transcripts produced during the initiation phase can have a length of 2 to 13 nt [50, 79, 80, 82 –87].

The transcription initiation process was studied at near base-pair resolution [88]. They concluded two possible pathways to produce abortive transcripts (Fig. 7). First, the 'polymerase recycling' in which the T7 RNA polymerase re-starts RNA synthesis after releasing abortive transcripts while remaining bound to the template strand. Second, the 'polymerase exchange' where the T7 RNA polymerase with or after RNA release dissociates from the DNA and synthesis of a new RNA starts only after a new T7 RNA polymerase binds on the DNA [88].

Another important investigation of this study was that both pathways can occur on the same DNA molecule and the partitioning between both is sequence dependent. At high concentrations of GTP a preference of the polymerase recycling pathway was observed, probably due to kinetic interactions between synthesis and T7 RNA polymerase dissociation [88]. This leads to the suggestion that nucleotide availability, especially GTP, can modulate the choice of abortive transcript production pathway.

It is noteworthy that the presence of diphosphate (PPi) plays an important role during the initiation process. One study investigated the impact of diphosphate on the formation of abortive transcripts and found out that with increasing diphosphate the amount of abortive transcripts also increases [89].

Besides the formation during IVT, also hydrolysis of the mRNA or contamination with RNase can lead to the formation of fragmented mRNAs [24].

Hydrolysis likely occurs due to broken phosphodiester bonds that form the mRNA backbone. This is caused by a nucleophilic attack by a 2'OH group on the phosphate ester bond [90]. Previous studies observed that the secondary structure of the mRNA as well as the base sequence influence the rate of hydrolysis [91, 92]. Some cations like Mg²⁺ can also increase the activity of certain RNases which leads to hydrolysis of mRNA [93, 94].

These short mRNAs can interact with T7 RNA polymerase via promotor independent RNA-templated transcription, producing short complementary transcripts and forming dsRNAs.

Metal ions can have a crucial influence on the RNA secondary structure, its folding and stability [95, 96]. Contaminations with metal ions can occur due to storage bags, containers, etc [85]. Several studies revealed the possible impact of different metal ions on mRNA and some can cause hydrolysis of RNA due to the binding of sequence-specific metal ions such as Pb²⁺ (lead(II) ion) [85, 97].

Short mRNA fragments decrease mRNA purity and yield. Due to the incompletion of the IVT reaction the mRNA that is produced at this stage is truncated, so they are unable to translate the complete ORF resulting in reduced translation efficiency [24]. The more mRNA fragments are produced during manufacturing the lower is the potency of the therapeutics and therefore the therapeutic success for the patient. Besides translation efficiency mRNA fragments have also an impact on immunogenic responses since also short dsRNAs are formed [40] and therefore can trigger the cytosolic sensors of the immune system in the same way as long dsRNAs [14, 22, 57].

Because of the impacts of mRNA fragments described above it is necessary to remove those fragments or prevent their formation during IVT [2, 98].

Studies suggested that lowering the initiation complex to elongation complex transition barrier can reduce the production of short RNAs. As described in 3.3, one research group recently showed that T7 RNA polymerase can be engineered to produce mRNA free of immunostimulatory by-products and short mRNA fragments [40].

One such engineered variant, P226L, proved to drastically reduced mRNA fragments with a length of 4–7 nt [79, 99, 100]. However, longer fragments (9–11 nt) were still presented, suggesting that the transition from IC to EC was not fully stabilized with this mutant alone.

A hypothesis was that weakening the T7 RNA polymerases' interactions with the promotor would facilitate the promotor release, leading to a higher stability of the RNA-DNA hybrid and therefore reduced the formation of fragmented mRNA.

A combination of P226L modified T7 RNA polymerase and a modified promoter that weakens upstream promoter interactions reduced all species of abortive transcripts. This modified promoter had a cysteine instead of an adenosine and is called ϕ9(A-15 C). It was observed that this modified promoter was able to reduce longer abortive byproducts by facilitating the P266L mutant to undergo the transition from IC to EC at 9 nt [85, 100].

A study also investigated the use of truncated promoters to mitigate formation of abortive transcripts, but IVT reactions with truncated promoters led to similar levels of abortive byproducts [101].

Various promotor topologies were tested for reduced mRNA fragment production. These included, -insertion of six non-complementary nucleotides in the non-template strand (between pos. −5 and − 4) and non-complementary non-template strands on position − 4 to + 5. Additionally, nicks in the template strand between position − 5 and − 4 have been tested. Only the nicked DNA template strand on position − 5 resulted in significantly reduced mRNA fragments but also led to a reduced overall mRNA yield [102]. Thus, using truncated or nicked promotors to reduce fragment formation are not suitable strategies for mRNA manufacturing.

The presence of pyrophosphate (PPi) can increase the level of abortive transcripts. Therefore, the inclusion of pyrophosphatase into the IVT reaction can reduce the formation of these byproducts. Several studies showed that pyrophosphatase increased the total mRNA yield which made it an even more attractive mitigation strategy [103, 104].

Another approach that can help improving mRNA quality is the development of mRNA with enhanced structural properties that transit more rapidly into the elongation complex or exhibit reduced loopback folding [105].

This can be aided by novel artificial intelligence and machine learning (AI/ML) algorithms, like PETfold or FARFAR2 that can accurately predict mRNA secondary and tertiary structures [106]. This strategy allows for the improvement of mRNA quality without altering experimental parameters or requiring additional purification steps.

The formation of truncated byproducts caused by hydrolysis and RNase activity can be reduced by adding RNase Inhibitor into the IVT reaction and the use of RNase inactivating detergents for workspace cleaning. Metal ion contamination can be reduced by using high quality consumables, storage containers and bottles.

A common method to analyse mRNA integrity is Capillary Gel electrophoresis (CGE) [24]. This technique separates mRNA by their size under denaturing conditions, which eliminate secondary structures, using agents such as formamide or urea [24, 107].

It can efficiently separate oligonucleotides greater than 2000 nt in length and can also detect mRNA aggregates but its limitation lies in the long time of the separation which makes it less efficient for mRNA drug and process development. Recently, a study developed a microchip capillary electrophoresis (mCGE) to analyse purity and integrity of an RNA 2000 nt in length [108].

The separation time could be reduced to one minute and is suitable for analyse of various lengths of RNAs by separating an RNA mass ladder that ranged from 200 to 6000 nt.

This method is a useful approach to separate mRNAs and get information about the RNA integrity. Smaller RNA fragments like abortive transcripts (which are much shorter than 200 nt) cannot be detected with this method.

Fragmented mRNAs can also be analysed by IP-RP HPLC and radioactive labelled gel electrophoresis, as discussed in 3.4 [24, 40].

Another, more sensitive method to analyse short RNAs is Liquid-Chromatography-Mass spectrometry (LC-MS). The length of the transcripts can be measured and distinguished between single-stranded RNA, RNA aggregates, and double-stranded RNA (dsRNA) [24]. In one study it was shown that oligonucleotides as short as 4 nt can be analysed [109], making it an attractive and powerful tool for abortive transcript analysis.

Capping of mRNA is achieved by adding a 7-methylguanosine moiety to the 5'triphosphate end (m7GpppG) of the mRNA.

There are two pathways to cap mRNA in vitro: co-transcriptional using capping analogues that are directly incorporated during IVT and enzymatically after IVT (post-transcriptional) using the Vaccinia virus capping enzyme (Fig. 8). With the post-transcriptional pathway, the 5' end of the mRNA, pppN, is transformed by this enzyme using guanine from GTP and S-adenosylmethionine to form the terminal cap GpppN (G cap) which is then methylated by (guanine-7)-methyltransferase activity to the N7-methylated guanosine triphosphate cap (m7GpppG). This capping structure is referred to as Cap 0 [110 –114].

By 2'-O-methylation, Cap 0 is formed to m7GpppmG, called the Cap 1 type (Fig. 9A) [112, 113].

During co-transcriptional capping a capping analogue such as m7GpppG is added to the IVT reaction which is then incorporated by the T7 RNA polymerase at the 5' end instead of a GTP [112]. Beside Cap 0 and Cap 1, there are also other capping structures, Cap 2 and Cap 4 which are produced through additional 2′-O-methylation steps [115], although they are less commonly used.

During co-transcriptional capping some cap analogues can be incorporated in the reverse orientation (GpppGm7) resulting in the formation of wrong capped mRNA byproducts, lowering the translation efficiency of the mRNA. Up to 50% of co-transcriptional capping can be reverse orientated [112, 116, 117]. This is because the m7GpppG is linked by a 3'−5' phosphodiester bond to the first nucleotide residue of the mRNA chain [117].

A competitive interaction also occurs between cap analogues and GTP during transcription, often leading to partially uncapped mRNA species. To improve capping efficiency, the GTP concentration in IVT reactions is typically reduced and cap analogue concentration increased; however, this adjustment can compromise overall mRNA yield [112].

Currently, there are three common cap analogues for co-transcriptional capping that are broadly used for the studies of therapeutic mRNA: the m7GpppG dinucleotide, the 3'-O-Me-m⁷G(5')ppp(5')G RNA cap analogue (ARCA) (Fig. 9B), and CleanCap a trinucleotide (Fig. 9C). ARCA is a modified capping analogue blocked at the 3'-hydroxyl group of the m7G which ensures that the cap analogue is incorporated in the correct orientation and has reported capping efficiencies up to 80% [118].

The third capping analogue is CleanCap (cap 1 structure) a trinucleotide, which forms base pairing during IVT initiation and reported to achieve > 90% capped mRNA and requires a starting sequence of "AG" [119, 120].

The 5'-cap structure of mRNA plays a pivotal role in maintaining its stability and enabling efficient translation, making it essential for both natural mRNA function and mRNA-based therapeutic applications [112]. It participates in several key processes, including translation initiation, mRNA splicing, nuclear export, and turnover, all of which influence the steady state levels and functional lifespan of mRNA within the cell [115, 121, 122].

mRNAs that are uncapped or possess incorrectly oriented caps are not efficiently recognized by the ribosome, resulting in poor overall translation efficiency.

Moreover, the cap structure stabilizes the mRNA by shielding it from exonucleolytic degradation, thereby enhancing its half-life and allowing for sustained protein synthesis [123 –125].

Consequently, uncapped mRNA undergoes rapid degradation, compromising its stability and diminishing the therapeutic impact. Therefore, understanding and optimizing mRNA capping is fundamental to advancing effective mRNA-based therapies.

Cap analogues have been engineered to prevent incorporation in the reverse orientation during IVT, known as the anti-reverse cap analogues (ARCAs) [117]. A methyl group was included at the 3'-OH of the m7G ribose to create 3'-O-Me-m7GpppG capping analogues [126]. At this position the first dinucleotide caps are correctly incorporated during the IVT [127, 128]. This kind of modification is also used in COVID-19 mRNA vaccines [129]. Modified cap analogues with a higher affinity to the eukaryotic translation initiation factor 4E (eIF4E) have proven to increase the initiation of translation [121]. Additionally, the mRNA stability can also be improved by modifying the phosphate chain in the cap structure to make it more resistant to exonucleases that causes decapping. With this strategy, the lifetime of the mRNA in the cell can be increased [127, 130].

Among available analogues, CleanCap™ trimer has demonstrated the highest capping efficiency and provides a natural Cap 1 structure, which supports superior translation compared to the Cap 0 structure achieved with ARCA [119].

From a manufacturing perspective, co-transcriptional is preferred over post-transcriptional capping. It is less time-consuming, and the manufacturing process can get more difficult to control if enzymatic reactions are involved due to their high fail rate based on enzyme quality and conditions [131].

The most effective co-transcriptional capping reagent is CleanCap reaching the highest reported capping efficiencies. However, it is costly, making the capping process itself to the most expensive part in mRNA manufacturing.

A common method to analyse uncapped mRNA is LC-MS [112]. However, LC-MS cannot analyse the full-length mRNA, therefore short sequences from the 5'-cap have to be cleaved using an enzyme or ribozyme [120, 132]. The cleaved fragment is then hybridized with an RNase H cleavage probe and can be analysed with LC-MS. Uncapped and capped mRNAs are investigated by their different masses and retention times [112, 133].

An alternative approach used to detect uncapped mRNA is to use enzymatic digestion into smaller fragments (< 50 base pairs) followed by polyacrylamide gel electrophoresis. The capped mRNA can be distinguished from un-capped mRNA by size. This is an in-expensive, semi-quantitative and simple method but includes many handling steps which lead to a higher risk for mistakes and the low throughput [133, 134].

Another effective method for assessing capping efficiency involves the selective enzymatic digestion of uncapped mRNA using a two-enzyme approach. This method, first described in 2017 by Chiron & Jais, involves the sequential use of RNA 5'polyphosphatase and terminator 5'phosphate dependent exonuclease. The 5′ polyphosphatase removes the γ and β phosphates from uncapped mRNA, converting it into a 5′ monophosphate form. This modified RNA is then specifically degraded by the exonuclease. The remaining capped transcripts can subsequently be quantified using HPLC, allowing for an accurate estimation of capping efficiency [135, 136].

Labelling the 5'-cap with a fluorescent marker by using capping analogues with a fluorescent tag or chemically modified guanosine nucleotides is another analysis technique [133, 137]. It is possible to label the phosphate with fluorescein, an organic compound [138]. It was shown that the translation is not affected by this method, however high signal background, fluorescence stability and labelling efficiency are reasons why this analytic approach is not widely used [133].

A recent approach is to detect uncapped and capped mRNA with Nanopore-sequencing, a method used for direct sequencing of RNA and its modifications [139]. In this technique, the mRNA molecules are translocated through a nanoscale protein pore, and the resulting changes in ionic current are measured to determine the sequence and structural features of the RNA [140]. This method has been improved by an engineered nanopore to directly identify various RNA modifications such as pseudouridine and N7-methylguanosin (m7G) with an accuracy of 99.6% [140]. Nanopore-sequencing has the potential to identify modifications such as cap structures and, therefore, to distinguish between capped and uncapped mRNA, but it requires further optimization. Apart from that, it is time consuming, as the samples must be clean and a library has to be prepared according to Oxford Nanopore Technologies' instructions [141].

Another method is quantitative splinted-ligation reverse-transcriptase PCR (qSL-RT-PCR) [142]. In this method, an overhang upstream at the 5'-end of the mRNA is created by using an oligonucleotide splint. This overhang is annealed with an oligonucleotide anchor on which the uncapped mRNA can be ligated with its free 5' monophosphate. Since capped mRNA is blocked and cannot be ligated onto the anchor only uncapped mRNA can be detected. With this method it is possible to quantify the amount of uncapped mRNA in the IVT product. This is a relatively complex method, making it unsuitable for high-throughput analysis. Only if multiplexed in a single PCR setup in which multiple targets are detected in the same reaction well or tube, this method becomes moderate high-throughput [143].

The process related impurities vary in their sizes, DNA is typically the largest undesired molecule in the crude IVT reaction mixture and is commonly removed by enzymatic digestion with DNaseI, which also serves to terminate transcription [2].

This step is highly important as an exceeding amount of DNA in the vaccine would lead to severe inflammatory reactions [145]. Therefore, the amount of residual DNA recommended by the regulatory authorities is 10 ng DNA per dose [145, 146]. It has been proven, by performing a Qubit fluorometric quantitation assay and LC-MS, that in the COVID-19 vaccines from Moderna (Spikevax) and BioNTech (Comirnaty) the DNA content was below and in some samples 1.8-fold above this threshold [145]. It is worth noting that the Qubit system is not suitable for measuring mixtures of nucleic acids, as it exhibits non-specific interactions and requires enzymatic destruction of either RNA or DNA, depending on the target to be measured [147]. Another study used multiple orthogonal methods: qPCR, fluorometry, capillary electrophoresis, and sequencing with biological and technical replicates of the COVID-19 vaccines Spikevax and Comirnaty [147]. They could prove that in all samples and in all analytical methods the DNA content was consistently below the established threshold of 10 ng per dose. These findings confirmed the regulatory compliance of mRNA vaccines and the need of complementary approaches to achieve reliable vaccine analysis [147]. Another critical by-product that is being formed during the IVT process are RNA-DNA hybrids. Those are most likely also sensed by innate immune response sensors like dsRNA since several PRRs can detect them [18 –20]. By performing the Dot Blot assay with the antibody S9.6 mAb those impurities can be detected in the sample [21]. DNaseI removes RNA-DNA hybrids quite inefficiently, its activity is at least 100-fold lower than for dsDNA [148]. Additionally, RNA-DNA hybrids can also not be removed by cellulose-based chromatography [21]. Therefore, developing methods for efficient and reliable removal of RNA-DNA hybrids are needed, for example by optimizing chromatography approaches.

The residual T7 RNA polymerase can be effectively removed by using several purification techniques, such as chromatography.

Due to the fact that the DNA template and RNA polymerase are produced using bacterial expression systems such as E. coli control of microbial impurities, including viral, bacterial and endotoxin contaminants, is critical for ensuring product safety and compliance with international regulatory standards. The ICH Q5A(R2) guideline provides a framework of viral safety evaluation in biotechnological products, emphasizing testing strategies and risk assessment as well as validation for potential viral contaminants introduced through raw materials or process intermediates [149].

Characterization of expression constructs and production systems is supported by the ICH Q5B, which offers guidance on ensuring the stability and identity of recombinant materials used in upstream process [150]. E. coli is gram-negative and lipopolysaccharides from those, called Endotoxins, are of particular concern due to their potent immunogenicity and heat resistance. Through improperly purified reagents or contaminated labware these can enter the IVT workflow. Therefore, strict quality control, including the use of endotoxin-free materials and validated endotoxin testing is essential to meet ICH requirements and to ensure the safety of mRNA therapeutics.

Chromatography is the method of choice for efficient, large-scale purification of biopharmaceuticals and mRNA products due to selectivity and scalability.

A highly effective approach is oligo(dT) affinity chromatography. This affinity chromatography uses immobilized oligo-deoxythymidicil acid (oligo(dT)) ligand and its base-pairing with the poly(A) tail of the mRNA. Through complementary base-pairing only full length mRNA containing a poly(A) tail can bind to this ligand and impurities, such as residual NTPs and fragments remain in the flow through. Afterwards, it and can be eluted from the column leading to a high purity product with a single purification step [151].

Oligo(dT) columns cannot remove dsRNA impurities that contain a ploy(A) sequence and hence additional polishing steps are necessary [2].

One common polishing method is IP-RP-HPLC, which effectively removes dsRNA, thus, enhancing translation efficiency and reducing immunogenicity [69]. This method requires large amounts of toxic solvents such as acetonitrile and ion-pairing reagents, which must be carefully removed post-purification which poses environmental and scalability challenges.

Anion exchange chromatography (AEX) is another method for separation of mRNA from process-related impurities [2].

AEX is potentially limited in resolving mRNA product-related impurities such as fragments and dsRNA, and its full potential for mRNA manufacturing has not been fully explored yet.

To summarize, the purification of mRNA products needs further development as the currently applied methods all have limitations, and a platform process such as for monoclonal antibodies has not been developed yet.