Read time: 11 minutes

Interviewee: Adam Crowe (PhD), Sr Manager of Analytical Development, Cytiva

A few years ago, it was discovered that messenger RNA (mRNA) encapsulated in lipid nanoparticles (LNPs) could result in mRNA adducts due to the breakdown products of N-oxide impurities. The ionizable lipids used in LNPs are especially susceptible to forming N-oxide impurities.

In a recent webinar, available on demand, Dr. Adam Crowe—Sr. Manager of Analytical Development at Cytiva—explained the correlation between lipid purity and mRNA lipidation. He also demonstrated how liquid chromatography coupled to mass spectrometry (LC-MS) using electron activated dissociation (EAD) can be used to structurally elucidate lipid raw material, including localization of double bonds and saturated impurities and differentiation between oxidated species.

To help you improve your mRNA-LNP analysis, I have summarized Adam’s answers below to pressing questions from his webinar about the analytical landscape, effects of N-oxides on mRNA, lipid structure elucidation with EAD and more.

What are the best analytical techniques for assessing the purity of ionized lipids from your point of view?

There are two elements to that question, the routine work and the specialized work. Typically, routine work is done in the industry with charge aerosol detection (CAD) that allows you to have a good idea of the percentages of impurities. As the technology is agnostic towards the structure, the ionization, etc., we use it routinely for release testing and QC testing. However, mass spectrometry is essential when you want to dig deeper into impurity profiles to identify them. We typically use mass spectrometry (MS) as a characterization assay to do identity testing and CAD as our release assay.

Which analytical services does Cytiva offer to clients to test for the lipid impurities that you discussed today?

We offer different programs and related analytical services. Some clients need an LNP manufactured over a few days, some a multi-year clinical program. The type of analytical work we do depends on the clients’ program and needs. When a program is going towards GLP or clinical phases, we typically do a very thorough assessment—similar to what I described in my webinar. While we do not look for adduct formations routinely, we investigate with that focus, if we see a complication arise. Related to material characterization, we provide a full report on the lipid materials, including stability and impurities, reporting impurities down to between 1% and 0.1%.

Does adduct formation block the translation of the mRNA?

To my knowledge it has not been explicitly proven that adduct formation is blocking the translation, however, it is a very likely assumption. Apart from the adducts, we have not seen any differences in mRNA with and without adducts that could explain the loss of function. From a biology perspective, it is likely that the translation is hindered by the adducts, although there are other possible mechanisms that could be considered.

Can you comment on which levels of N-oxides are impacting mRNA efficacy?

This is a complicated question as the N-oxide formation is only the first of multiple steps, and may not be the reactive species causing adduct formation. If there are high N-oxide levels, the adduct formation is rapid, however the correlation between adduct formation and low N-oxide levels of 0.1 to 0.5% is not very strong.

Adduct formation is a multi-step process and a subsequent aldehyde is the species causing adducts. A Japanese research group published on the mechanisms recently (Hashiba, K., Taguchi, M., Sakamoto, S. et al. Commun Biol 7, 556 (2024)). To close this question with some guidance: If you are seeing N-oxide levels above 1%, I would certainly worry about adduct formation and if you are seeing 0.1 to 1%, I suggest evaluating your material and for less than 0.1%, it is likely not a huge risk.

Does the FDA require monitoring for RNA adducts in drug products? And if not, are they likely to request this in the near future?

At the end of 2021 when Meredith Packer et al. first reported on this, the clear answer was no. Regulatory bodies were not asking for this, but the industry has changed quite a bit. We are seeing a lot more specific questions about RNA adducts when working with clients on their Investigational New Drug (IND) applications, particularly those from the FDA, which are very targeted. Currently it is not requirement for which routine monitoring, but since it is a ubiquitous challenge faced with LNP-based genetic medicines, it is very possible it could become a routine characterization assay requested by the FDA as part of the drug master file or IND in the future.

Are there techniques available to directly assess the modified nucleobases to identify adducts?

Adam Crowe: When intact RNA shows hydrophobic peaks in the ion pairing reversed phase liquid chromatography (IP-RP-LC) data, hydrolyzing the mRNA and analyzing the nucleosides is a next logical step. We tried this and it turned out to be challenging. While an estimation of expected adducts can be done based on the IP-RP-LC data, we have been unable to directly observe the modified nucleosides using the RP-LC-MS protocol from Packer et al.- even in samples with >50% adducted mRNA. Currently, we are unsure if this is a methodological problem or a misunderstanding of the nature of RNA lipidation.  Interestingly, in a recent publication (Hashiba, K., Taguchi, M., Sakamoto, S. et al. Commun Biol 7, 556 (2024)), hydrophilic interaction chromatography (HILIC) coupled to MS was used, which seemed to work much better than RP-LC-MS. My assumption is that the industry is going the way of HILIC-MS to detect modified nucleosides, however, there is still more evaluation that needs to be done.

And can you expand on the differences between CID and EAD?

Adam Crowe: Collision-induced dissociation (CID) uses a collision gas, typically nitrogen with high purity. The ionized analyte is accelerated and collides with the gas molecules, which causes labile bonds to break. In contrast, electron-activated dissociation (EAD) uses electrons and is similar to the fragmentation used in gas chromatography MS. The electron causes bonds to be destabilized and to fragment in two ways: a two lone pair or through a radical chemistry reaction. Because EAD is a high-energy fragmentation technique it allows for the fragmentation of otherwise very stable carbon-carbon and carbon-nitrogen bonds. Further information can be obtained when contacting the SCIEX team.

Is MS/MS with EAD also applicable to monitor N-oxides in formulated LNPs?

Adam Crowe: Yes, absolutely. This is where high resolution instruments, such as the ZenoTOF 7600 system, really shine. We did a lot of the work intentionally with the mRNA and the ionizable lipid—what is referred to as the binary system. When you analyze a formulated LNP, there are complexities to consider, including other lipid species and excipients in your sample, and the de-formulation step.

Our approach is to do exploratory work in an untargeted way with data-dependent acquisition (DDA) first. Once we have a better understanding of the impurities for a given lipid, we do a more targeted approach. Either using high resolution instrumentation or often triple quadrupole instruments looking specifically for previously identified N-oxides.

Do you have a recommendation for the required purity of ionizable lipids for R&D usage?

Adam Crowe: The short answer is to use as high of a purity as you can. The long answer is that it is not the purity level itself that is the problem, but the types of impurities that are there. Of course, the less pure your material is, the more likely it is that you have reactive impurities present. However, some impurities are not likely to cause an impact on your final product. For example, differences in the lengths of the alkyl chains could lead to a low purity of the target material without huge impact. However, large amounts of aldehydes remaining from the synthesis or precursors or catalysts are problematic.

In addition, I recommend being cautious when screening lipids regarding missing hits based on impurity levels. For instance, when doing initial potency screening, a great ionizable lipid candidate could potentially show low activity because of the levels of reactive impurities.

Are there impurities from other LNP components than the ionizable lipid that can lead to RNA-lipid-adducts?

Adam Crowe: The answer is likely to be yes. While early generation LNPs used a simple system of ionizable lipids, PEG lipids, helper lipids and cholesterol, we are starting to see more and more complexities, including more type of components being added, which increases the likelihood of having reactive species.

Although I have not experienced any extreme case yet, I want to call attention to PEG‑ylated lipids as an area of potential concern. PEG is known to form peroxides under certain conditions and as the complexity of PEG-lipids is changing, there is a risk of increasing the peroxidation events inside LNPs, which can lead to adduct formations.

How much time do you spend on method optimization and data processing? Do you need to optimize for each new lipid?

Adam Crowe: Optimizing chromatography can be a challenge regardless of the detection system you are using. There are many different types of ionizable lipids, which are inherently charged, while optimization of the chromatography can be specific for each lipid, the MS method is quite interchangeable.

Generally, high-energy fragmentation is required to achieve relevant bond breakage beyond the headgroup of lipids. We therefore used an electron kinetic energy of ~16 eV and this setting will be suitable for other lipids, too. However, EAD is highly tunable, so you can optimize the energy for the specific analyte you are aiming to fragment.

The advantage for processing of data from ionizable lipids is that SCIEX has a software that can process the EAD data. The Molecule Profiler software will predict the impurity and structure based on the detected fragments. Using experience and the wealth of fragments obtained by MS/MS with EAD, quite accurate predictions can be made that simplify the processing of data.

Which software do you recommend for interpreting data for lipid structure elucidation?

Adam Crowe: It’s a matter of the complexity of your data, which data acquisition strategy you were using and which questions you are trying to answer. For EAD data, you can use Molecule Profiler software (SCIEX), or manually review the data with SCIEX OS software in combination with ChemSpider. I have also seen groups taking a lipidomics-focused approach with the caveat that the diversity in biological systems is significantly higher than for LNPs.

 Are there libraries to assign fragments or do you rely on manual assignment?

Adam Crowe: The software I mentioned can assign fragments for impurities based on the known precursor structures or you can create libraries. In our case, we create libraries ourselves since most of the ionized lipids we are working with are proprietary and therefore there are no external libraries readily available.

How else do high resolution mass spectrometry instruments like the ZenoTOF 7600 system support the development of LNPs?

Adam Crowe: There are a lot of options how HRMS instrumentation can support LNP development. The mRNA drug substance needs to be characterized and optimized, which includes the capping structures: uncapped, G cap, cap 0 or cap1, the poly-adenylation on the 3’end of mRNA, the nucleoside content.

In addition, impurity analysis of lipids can be extended beyond ionizable lipids. Looking at the changes that occur inside the LNPs is another big question. Currently, some efforts are focused on fractionation of LNP subpopulations of large, small, empty, and filled particles. HRMS is the way to go to analyze these complex populations and understanding differences further. Lastly, HRMS is very well suited for characterizing and monitoring excipients or additives for stabilizing the LNPs, and their impurities.

Can you perform absolute quantitation of ionizable lipids with the EAD method you presented? Are isotope-labelled standards available?

Most of the ionized lipids are early stage, therefore, there are not a lot of internal standards or reference materials available. The groups we work with typically have a third-party manufacturer for their ionized lipids with their own synthesis processes, allowing for labelled standards. Standards are particularly important when attempting absolute quantitation of N-oxides because a tertiary amine changed to a quaternary amine is likely to make a difference in ionization compared to the main compound, hence comparing relative values is challenging.

You showed a correlation of percent impurities with amount of adduct formation (slide 16). Can you elaborate which impurities you were looking at?

Adam Crowe: The correlation shown was based on an initial screening we did using charge aerosol detection (CAD) and showed the overall percentage of impurities detected as sum, not any individual impurity species. CAD is great to understand relative levels as the response is dependent on the amount, not structures. Of course, you need to identify what these impurities are, which is why the second part of my presentation was focused on characterizing impurities using MS with EAD. Although the impurity levels correlate, they are not entirely predictive of adduct formation, which should be considered for any study.

In short, no, although we have evaluated a few design of experiment (DoE) approaches for LNP development and analysis. I am supportive of future in silico predictions of LNP behavior, however, I do not believe we currently have the necessary level of physiochemical understanding of LNPs to make this work a high priority. My rationale is threefold: firstly, I’m skeptical that in silico computations could accurately account for the complexity of the LNP system of 4+ lipid components, complex payload, huge influence of buffers/ excipients and the manufacturing process. Secondly, LNP drug products consist of a range of different LNP populations and therefore their therapeutic efficacy must be viewed holistically rather than for a single species. Finally, I do not believe we know which physiochemical characteristics should be selected to drive in silico predictions—let alone if simple technologies exist to assess these factors. For example, to my knowledge, the exact influence of particle size, surface charge, structure, etc. on therapeutic efficacy remain elusive.