Separation of PS diastereomers in GalNAc-conjugated siRNA—a comparative study of chromatographic and ion mobility mass spectrometry approaches

Abstract

The separation and characterization of phosphorothioate (PS) diastereomers in GalNAc-conjugated siRNA presents a significant analytical challenge due to the high number of isomers present. To tackle this challenge, we explore and optimize various separation techniques, including chromatographic methods (ion-pair reversed-phase, anion exchange, and hydrophilic interaction liquid chromatography) and ion mobility mass spectrometry (IMS) to assess their efficacy in diastereomer separation. Our results indicate that AEX provides the highest diastereomer selectivity among the chromatographic techniques, although none achieved complete diastereomer separation for the chosen antisense and sense strand reference compounds. IMS, applied within a fragment-based tandem mass spectrometry approach, allows separation of all diastereomers of the antisense strand and partial resolution of the sense strand in the gas phase. The comparison of relative LC-UV quantification with IMS data reveals a strong correlation and suggests that IMS can effectively characterize diastereomer ratios not only qualitatively but also quantitatively, establishing IMS as a promising complementary analytical technique for diastereomer separation besides more established LC-UV methods.

Graphical abstract

Supplementary Information

The online version contains supplementary material available at 10.1007/s00216-025-06123-4.

Introduction

The rapid growth of oligonucleotide-based therapeutics marks a significant advancement in modern medicine, allowing the targeting of previously “undruggable” conditions. Among these, small interfering RNAs (siRNAs) have emerged as an important class, with six approved drugs currently on the market and numerous others in various stages of clinical development [1, 2]. siRNAs are duplex molecules consisting of two single strands, each single strand being 18 to 25 nucleotides long, which can be chemically modified to enhance in vivo stability and efficacy. A common modification is the sulfurization of the internucleotide phosphodiester (PO) to phosphorothioate (PS) linkages, which improves the stability of siRNAs in serum by increasing the resistance to nuclease digestion [3]. However, this modification also creates a new chiral center at the phosphorus atom, with each n PS linker having either Sp or Rp configuration leading to the formation of 2ⁿ diastereomers [4, 5]. The diastereomeric composition is significantly influenced by synthesis conditions, such as the activator used, and mock pooling during purification [6–8]. Minor changes in experimental procedures may lead to varying diastereomer distribution in the final product, which may influence the pharmacodynamic and pharmacokinetic properties of siRNAs [9, 10]. To ensure batch-to-batch consistency, regulatory agencies are therefore increasingly demanding control methods for the diastereomeric composition of oligonucleotide drugs, underscoring the growing importance of diastereomer characterization in pharmaceutical drug development.

However, distinguishing and quantifying stereoisomers of oligonucleotide single strands poses a significant challenge in quality assurance settings. Chromatographic separations, such as ion-pair reversed-phase (IP-RP) chromatography, anion exchange chromatography (AEX), and hydrophilic interaction chromatography (HILIC) represent the most popular options for oligonucleotide separation, but these methods are usually optimized to suppress diastereomer separation in order to maintain peak shape and separation for impurity characterization. While there are many approaches described in literature to enhance chromatographic diastereomer separation [11–13], it is unclear which of the separation modes provides the best separation of diastereomers for siRNA single strands.

Additionally, aside from the aforementioned chromatography techniques, alternative strategies like nuclear magnetic resonance (NMR), circular dichroism (CD), and enzymatic methods are available [14–16]. While these methods can potentially detect alterations in the overall diastereomeric distribution, they are often more complex to conduct and do not provide specific data for individual PS linkages.

Recently, ion mobility mass spectrometry (IMS) emerged as a potential alternative to chromatographic separation for impurity characterization as well as for diastereomer elucidation [17]. Demelenne et al. demonstrated the use of multiplexed linear drift tube IMS (DTIMS) to achieve full diastereomer separation of short DNA pentamers with 0–3 PS linkages [13] and Goyon et al. developed this approach further for short RNA model sequences [18]. Subsequently, Sharon et al. showed that this methodology can be applied to full-length siRNA single strands when using high-resolution ion mobility separation with a Cyclic™ IMS (cIMS) instrument [19]. Although diastereomer separation of the intact single strand was not directly feasible, the generation of small terminal fragments via collision-induced dissociation (CID) in the gas phase led to a reduced number of possible diastereomers per fragment and allowed the characterization of all diastereomers. Yu et al. performed a comparable workflow on single guide RNA compounds, but instead of gas-phase fragmentation, they relied on enzymatic digestion before measurement [20]. In addition to an impressive separation power, they demonstrated a quantitative correlation between LC-UV data and LC-cIMS data for diastereomer characterization. While digestion of siRNA single strands can be accomplished in a similar manner, the direct generation and measurement of fragments in the gas phase appears to be more straightforward. It remains to be determined, however, how well the observed diastereomeric separation of gas-phase fragments by IMS correlates with LC-UV data.

In this work, we investigate and compare the use of chromatographic separation and IMS for the diastereomer separation of individual strands of two GalNAc-conjugated siRNA model compounds. We optimize IP-RP, AEX, and HILIC methods for this intended purpose, and evaluate and compare their ability to generate a diastereomer fingerprint with good resolution and high selectivity. In addition, we assess the quantitative behavior of diastereomeric fragments measured by IMS and compare the gas-phase results with the LC-UV data as benchmark.

Material and methods

Chemicals

Water was purified by a Milli-Q® system from Elix® Technology Inside (Merck Millipore, Darmstadt, Germany) or purchased (Corning Cell Culture Grade Water Tested to USP Sterile Water for Injection Specifications) from Fisher Scientific UK (Loughborough, UK). CHEMSOLUTE® LC-MS grade acetonitrile (ACN) and methanol (MeOH) as well as hydrochloric acid (HCl, 1 M), acetic acid (AcOH, ≥ 99.8%), and sodium hydroxide (NaOH, 1 M) were purchased from Th. Geyer (Renningen, Germany). MeOH was also purchased from Honeywell (Seelze, Germany). Triethylamine (TEA) was purchased from either Carlo Erba (Emmendingen, Germany) or Merck (Gillingham, UK). 1,1,1,3,3,3-Hexafluoro-2-propanol (HFIP, ≥ 99.5%) was either purchased from ChemPUR (Karlsruhe, Germany) or Merck (Gillingham, UK). Ethylenediaminetetraacetic acid (EDTA) and N,N-diisopropylethylamine (DIPEA) were purchased from Merck (Gillingham, UK). Trizma® (2-amino-2-(hydroxymethyl)−1,3-propanediol) hydrochloride solution (Tris, 1 M) was purchased from Sigma-Aldrich. Ammonium formate and sodium chloride (NaCl) were purchased from Honeywell. Anhydrous trisodium phosphate was purchased from AnalytiChem (Duisburg, Germany).

Synthesis and purification

Oligonucleotides were synthesized on an ÄKTA oligopilot™ 10 synthesizer using standard phosphoramidite chemistry. Base loaded solid supports, 2′OMe nucleotide phosphoramidites and 2′F nucleotide phosphoramidites (all standard protection), were commercially available (ChemGenes, Hongene Biotech Corporation, Union City, CA, USA) and used according to the manufacturers’ recommended procedures. Synthesis of GalNAc (GN) and C4XLT phosphoramidites was performed according to previously published procedures [21]. Ancillary reagents were purchased from EMP Biotech (Berlin, Germany). Synthesis was performed using a 0.1 M solution of the phosphoramidite in dry acetonitrile, and benzylthiotetrazole (BTT) was used as an activator (0.3 M in acetonitrile). Coupling time was 10 min. A Cap/OX/Cap or Cap/Thio/Cap cycle was applied (Cap: Ac₂O/NMI/lutidine/acetonitrile; oxidizer: 0.1 M I₂ in pyridine/H₂O). Phosphorothioates were introduced using 0.2 M XH (0.2 M xanthane hydride in pyridine). DMT cleavage was achieved by treatment with 3% dichloroacetic acid in toluene. Upon completion of the programmed synthesis cycles, a diethylamine (DEA) wash was performed. All oligonucleotides were synthesized in DMT-off mode. The single strands were cleaved off the support by 40% aq. methylamine treatment (90 min, room temperature). The crude oligonucleotide was purified by ion exchange chromatography (Source™ 15Q, 7.5 mL, GE HealthCare, Chicago, IL, USA) on an ÄKTA pure™ HPLC system using a sodium bromide gradient. Product-containing fractions were pooled, desalted on a size exclusion column (Zetadex-25 Medium, EMP Biotech, Berlin, Germany), and lyophilized.

Chromatographic equipment and conditions

Liquid chromatography analysis was conducted on 1290 Infinity II® UHPLC systems (Agilent Technologies, Santa Clara, CA, USA) using the OpenLab® ChemStation® software for data processing. The instruments were equipped with a vialsampler, a binary solvent delivery pump, a thermostatted column compartment, and a diode array detector for UV detection at 260 nm. The systems used for IP-RP and HILIC were furthermore coupled with an InfinityLab® LC/MSD XT single quadrupole mass detector equipped with an API-Electrospray ionization (ESI) source.

For IP-RP chromatography, an ACQUITY™ Premier Oligonucleotide BEH™ C18 column (1.7 µM, 2.1 × 100 mm, 130 Å; Waters Corporation, Milford, MA, USA) was used at 25 °C. The mobile phases consisted of 20 mM triethylammonium acetate (TEAA; based on an aqueous 1 M stock solution of equimolar amounts of TEA and AcOH adjusted to pH 7.0 using AcOH) in water (A) or ACN (B) and were used for gradient elution of 12–16% B in 20 min with a flow rate set to 0.3 mL/min.

For AEX analysis, a DNAPac™ PA200 RS (4 µM, 4.6 × 150 mm; Thermo Fisher Scientific Inc., Waltham, MA, USA) was used at 30 °C. Mobile phase A consisted of either 20 mM Tris in water adjusted to pH 7.0 or 9.0 using 1 M HCl or 1 M NaOH, or 20 mM Na₃PO₄ in water adjusted to pH 11.0 using 1 M HCl. Mobile phase B consisted of 1.25 M NaCl in A. Individual gradients with an increase of mobile phase B ranging from 2 to 11% in 12 or 22 min were used at a flow rate of 0.8 mL/min.

For HILIC analysis, an ACQUITY Premier BEH Amide column (1.7 µM, 2.1 × 50 mm, 130 Å; Waters Corporation, Milford, MA, USA) was used at 20 °C. The mobile phases consisted of 25 mM ammonium formate in water/ACN 30:70 (v/v, A) or 60:40 (v/v, B) which were delivered at a gradient of 15–50% B in 30 min and a flow rate of 0.5 mL/min.

Further details are outlined in the Electronic Supplementary Material (Tables S1 and S2).

Sample analysis with cyclic IMS

Samples were analyzed with a Waters SELECT SERIES™ cyclic IMS system equipped with an electrospray ionization source (Milford, MA, USA).

Data were acquired and processed using MassLynx™ v4.2 software with additional processing using DriftScope™ v3.0 software for IMS data and CONFIRM Sequence v1.4.0.13 software to aid MS/MS fragment ion data interpretation.

Sample introduction into the cyclic IMS system was initially via Waters ACQUITY Premier UPLC™ System (Milford, MA, USA) and subsequently via direct infusion for optimization of the cyclic sequence settings. For UHPLC-MS-IMS, mobile phase A was 8.6 mM TEA, 100 mM HFIP in water and mobile phase B was 4.3 mM TEA, 50 mM HFIP in water/MeOH 1:1 (v/v) at a flow rate of 0.25 mL/min. Mobile phase B composition was 26% for 0.2 min followed by linear gradients to 29.5% B at 1.0 min and 36% B at 6.0 min, and isocratic holds at 100% B from 6.1 to 7.0 min and 26% B from 7.1 to 9.0 min. The column was a Waters ACQUITY Premier Oligonucleotide BEH C18, 130 Å, 1.7 µm, 2.1 × 50 mm (Milford, MA, USA) with column oven temperature of 60 °C. For direct infusion, an external syringe pump set to 10 µL/min was used.

Electrospray ionization (ESI) was performed using a capillary voltage of between 2.2 and 2.5 kV for UHPLC-MS analysis or 1.5 kV for direct infusion experiments and a cone voltage of 40 V. Source temperature was set to 100 °C and desolvation temperature was set to 400 °C. Cone gas was set to 0 L/h and desolvation gas was set to 800 L/h.

Default settings for the cyclic IMS system were used throughout except for travelling wave height, pushes per bin, and separation time, which were optimized for separation of fragment ion diastereomers as shown in Tables S3 and S4.

Results and discussion

Diastereomer analysis of siRNA single strands via LC

The aim of this study was to compare and evaluate the ability of chromatographic and ion mobility methods to assess the diastereomeric composition of GalNAc-conjugated siRNAs. For this purpose, we designed and synthesized two generic model compounds, siRNA#1 and siRNA#2, adhering to a baseline oligonucleotide design utilized by Silence Therapeutics. As illustrated in Fig. 1a, this design represents a blunt-ended double strand consisting of an antisense (AS) and sense (S) strand, whereby each single strand contains 19 nucleobases and a fully modified backbone with alternating 2′OMe and 2′F riboses. In addition, the sense strand is conjugated to a linker and three GalNAc moieties (GalNAc cluster). Similar to already approved siRNA drugs on the market, the antisense strand incorporates four phosphorothioate (PS) linkages, with two terminal PS linkages at both the 3′ and 5′ end, resulting in a total of 16 potential diastereomers [2]. The sense strand contains two terminal PS linkages at the 3′ end and a fully thioated GalNAc cluster at the 5′ end which results in a total of six PS linkages and 64 possible diastereomers. The design of the sense strand is different from already approved siRNA drugs which typically introduce only two terminal PS linkages at the 3′ end. Despite individual design choices, the selected two model compounds reflect very well the typical PS pattern of conjugated siRNA drugs on the market and were therefore chosen as representative molecules to evaluate analytical methods for diastereomer characterization.

Fig. 1.

a Illustration of the siRNA model compound design. b Final results for the modified antisense strands of siRNA#1 with only 1× PS linkage in different positions after LC method optimization for AEX, IP-RP, and HILIC (from left to right); PS-1 to PS-4 describes the position of the PS linkage for the simplified model compounds derived from the siRNA#1 antisense. The number next to the peaks represents the relative peak area in percent for the respective diastereomer signal. For separated diastereomers, the resolution (R) calculated considering the peaks width measured at half peak height is stated. Samples for which no sufficient resolution was obtained are marked with an asterisk

To facilitate HPLC method optimization and subsequent comparison, we employed simplified versions of siRNA#1 and siRNA#2 containing just one PS linkage in the four different positions (PS‑1–PS‑4; see Fig. 1a). Thereby, we could first focus on the separation of only two diastereomers at a time and obtain preliminary insights into the selectivity of each chromatographic method (Fig. 1b). A list of all analyzed sequences can be found in the Electronic Supplementary Material (Table S5).

We derived initial conditions for each chromatographic mode from literature and adjusted the LC conditions to optimize diastereomer separation for the model compounds. A short description together with critical parameters influencing diastereomer separation can be found in the Electronic Supplementary Material. The diastereomer separation for the simplified model compounds of siRNA#1 after optimization of chromatographic parameters is shown in Fig. 1b. Our findings indicate that all tested analytical methods provide some level of selectivity for the separation of most samples containing a single PS linkage (AEX: R = 0.9–6.3, IP-RP: R = 2.4–3.2, HILIC: R = 1.3–1.7). For the simplified model set of siRNA#1, HILIC is capable of separating two out of the four compounds (Fig. 1b, right column) while the optimized AEX (Fig. 1b, left column) and IP-RP (Fig. 1b, middle column) methods allow us to observe two distinct diastereomer peaks for three out of four samples. A similar trend can be observed when applying the three analytical methods for the simplified model set of siRNA#2 where HILIC is able to separate two compounds while AEX and IP-RP are able to separate all four compounds (Fig. S1). In general, AEX and IP‑RP demonstrate superior peak shape and resolution for single PS linkages compared to HILIC separation. However, neither method is capable of resolving all diastereomers and for some compounds, HILIC even seems to be complementary to AEX and IP-RP, for example for the PS-4 model of siRNA#1 (Fig. 1b, bottom row). Here, HILIC yielded two peaks while AEX and IP-RP could not provide sufficient separation. This indicates a sequence dependency of the chromatographic performance and that the position of the PS linkage also plays a crucial role. In addition, these results suggest that for simple compounds like single PS modified sample sets, not a single chromatographic method can resolve all diastereomers, but rather that HILIC can complement IP-RP and AEX. From a quantitative perspective, all three methods deliver comparable relative peak areas and diastereomer distributions for the individual samples. This is especially true comparing IP‑RP and AEX if baseline separation is given, e.g., for siRNA#1 PS-3 and PS-4 (Fig. 1b) and siRNA#2 PS-1 to PS-3 (Fig. S1). In terms of elution order of the two diastereomeric peaks, HILIC shows the opposite behavior compared to AEX and IP-RP, as can be seen in the reversed order of diastereomers for PS-3 of siRNA#1 and PS-2 and PS-3 of siRNA#2 (Fig. 1b and S1). This underlines the complementary character of HILIC for the analysis of simple 1 × PS oligonucleotides.

While the results for the single PS modified model compounds gave a preliminary insight into the resolution power of each chromatographic mode for diastereomer separation, they are not representative for actual siRNA drugs due to the reduced number of PS linkages. Therefore, the optimized LC methods were applied to the original antisense and sense strands of siRNA#1 (Fig. 2) and siRNA#2 (Fig. S2) which contain a more representative number of PS linkages. For the separation of the antisense of siRNA#1 in AEX (Fig. 2a), we can observe two peak clusters that are further divided into a total of 12 signals (out of 16 diastereomers). As diastereomer selectivity is likely not dependent on the adjacent PO/PS linkages [22], we can estimate the approximative equal ratio between R_P and S_P isomers observed for the single PS model compounds (Fig. 1), and assume coelution of isomers for the peaks 2, 3, 4, and 9. IP-RP is able to resolve six broad peaks of coelution-indicating shape (Fig. 2b), while HILIC only shows four incompletely resolved peaks (Fig. 2c). A similar trend is observed for the analysis of the sense strand, which is considered more challenging due to its higher number of PS linkages. Here, AEX can resolve 8 out of 64 possible diastereomers (Fig. 2d). Systematic removal of PS linkages in specific positions provided a strong indication that the diastereomers associated with the 3′ end result in four main peaks (data not shown). In some cases, those are further split into double peaks when the PS Linkage at the 5′ end is present that connects the oligonucleotide to the GalNAc cluster. Given the high flexibility of the alkyl linkers within the cluster, it is considered unlikely to resolve the corresponding diastereomers chromatographically. IP-RP (Fig. 2e) and HILIC (Fig. 2f) analyses show a single broadened peak making it difficult to assign the number of isomers. Overall, the same behavior can be observed for siRNA#2 as AEX is able to resolve the most peaks for both antisense and sense strand compared to IP-RP and HILIC (Fig. S2). Interestingly, the complementary behavior of HILIC compared to AEX and IP-RP that was observed before for the simple single PS compounds, could not be confirmed for more complex samples. Recent literature suggests that higher order structures resulting from intramolecular folding are critical to separate diastereomers via HILIC, while unordered structures are difficult to separate [12]. We therefore assume that the observed behavior is caused by the absence of a higher order structure in the single strands. UV melting curves are hinting in the same direction as they lack the typical sigmoidal shape that is characteristic for folded structures (Fig. S3). Based on our results, HILIC separation seems to be difficult to employ as standard quality control for diastereomers in batch release and for batch-to-batch comparison in a preclinical and clinical environment. Although no method was able to resolve all possible diastereomers, the data clearly indicates that AEX exhibits the highest diastereomer separation of the three probed LC modes for the employed model compounds. Therefore, we recognize the potential in implementing AEX for routine QC testing to assess the diastereomeric distribution and verify batch consistency. Although straightforward to implement as LC-UV method in regulated environments, AEX has the disadvantage that it cannot be coupled to mass spectrometry and therefore peak annotation becomes significantly more challenging. In addition, this increases the requirements in sample purity as high-level impurities might be mistaken as diastereomers. For other cases, IP-RP may be applicable as a compromise as it provides reasonable diastereomer selectivity and is compatible with MS. Overall, our results indicate that there is a need for alternative analytical approaches to overcome the above-mentioned problems for diastereomer separation.

Fig. 2.

Diastereomer separation with optimized AEX, IP-RP, and HILIC methods (from left to right) for siRNA#1 antisense (top row) and sense strand (bottom row). The numbers in the chromatograms indicate visually identified peak tops

IMS as alternative for siRNA diastereomer analysis

It was demonstrated recently that ion mobility mass spectrometry (IMS) can be used to distinguish diastereomers of oligonucleotides and therefore represents a promising alternative to chromatographic separation [19]. IMS is an analytical technique that separates ions based on their size, shape, and charge in the gas phase. It operates by passing ions through a gas-filled chamber under the influence of an electric field, where they are separated according to their mobility. A recent and excellent summary on IMS for structural biology can be found elsewhere [23]. IMS separation allows for the resolution of isomeric and conformational variants, such as diastereomers, which may not be distinguishable by conventional mass spectrometry alone. It is particularly useful in providing insights into the structural characteristics of molecules, making it a powerful tool for characterizing complex biomolecules like siRNA diastereomers. Consequently, we tested the siRNA#1 antisense on a cyclic IMS (cIMS) instrument and the results are shown in Fig. 3. In the mass spectrum, we observed charge states ranging from − 3 to − 9 and examined the mobilograms for all of them (Fig. 3a). However, none of these charge states showed any separation of diastereomers beyond peak broadening. This behavior was also reported by Sharon et al. for different sequences and suggests that the conformational differences caused by diastereomers in such large molecules are minor, complicating separation via IMS on the level of intact single strands.

Fig. 3.

Mobilograms of the antisense strand of siRNA#1. a Mobilogram of the intact FLP for different charge states (z) ranging from –3 up to –9. b Mobilogram of the terminal 5-mer fragments from the 3′ end (b₅^2–) is shown on the left and from the 5′ end (w₅^3–) on the right side. The numbers in the chromatograms indicate visually identified peak tops

Sharon et al. demonstrated that the measurement can be simplified by applying a fragment-based approach. Activation of the full-length product (FLP) in the gas phase via collision-induced dissociation (CID) leads to the generation of shorter terminal pentamer fragments on the 3′ and 5′ end. This effectively isolates the two PS linkages from either side from each other, thereby reducing the total number of 16 diastereomers for the FLP to just four per fragment. This approach reduced the complexity, enabling the measurement of the mobility of each fragment and thereby allowing observation of all expected isomers via IMS with sufficient resolution (peak-to-peak resolution for the b5 fragment: R_1–2 = 0.8, R_2–3 = 2.2, R_3–4 = 1.0 and for the w5 fragment: R_1–2 = 1.5, R_2–3 = 1.9, R_3–4 = 0.8). Similar fragment-based approaches have been successfully applied to other biomolecules before, e.g., glycans, where IMS was used to separate isomeric fragments containing sialic acid motifs which were cleaved from intact N-glycans in the gas phase [24]. It is, however, still an open question if the IMS results can be interpreted solely qualitatively or if they represent the actual ratio of diastereomers present in solution. We therefore repeated the experimental approach from Sharon et al. with the antisense strand of siRNA#1 and fragmented it to generate the expected ladder-type fragments (Fig. S4). We analyzed the mobility data of all generated fragments from both sides of the single strands (data not shown) and found that the b₅²⁻ and w₅³⁻ fragments, following McLuckey fragmentation nomenclature [25], allow a good separation of four distinct peaks in their mobilogram (Fig. 3b). These peaks are hypothesized to correspond to the four possible diastereomers (RR, RS, SR, and SS). Currently, we cannot assign the specific diastereomer identity to each signal. However, it is anticipated that the identity of each peak can be confirmed by synthesizing stereopure compounds [3, 10, 26] or by modelling the structure of oligonucleotide fragments to derive theoretical collision cross-section (CCS) values in future studies.

Comparing the signal pattern from the resulting b₅²⁻ and w₅³⁻ fragments, it is clear that the diastereomeric distribution is different at both ends of the single strand (Fig. 3b). While the pentameric b₅²⁻ fragment exhibits a specific intensity pattern in the order of peak 1 > peak 2 ≈ peak 3 > peak 4, the w₅³⁻ fragment shows the pattern in the order of peak 3 > peak 2 ≈ peak 4 ≈ peak 1. The peak patterns seem to be different for both ends of the single strand and therefore support the assumption that the diastereomeric ratio is dependent on the individual sequence and position of the PS linkages in the single strand. A similar dependency could be observed for the simplified single PS samples in Fig. 1b, as the position of the single PS linkages had a significant influence on the observed diastereomeric ratio. In addition, it shows the benefits of the fragment-based approach that enables reduction of complexity and the monitoring of individually resolved PS linkages instead of overall changes between batches. Subsequently, we applied the same methodology to the antisense strand of siRNA#2 (Fig. S5), which confirmed that small terminal fragments (3- to 5-mers) are appropriate for IMS analysis and that we can qualitatively distinguish the four expected diastereomers for both ends of the antisense strand. In addition to the pentameric fragments, we checked for consistency in different w- and b-type fragments of different sizes, and it seems that smaller fragments (3-mers or 4-mers) exhibit no significant differences in peak distributions in the mobilograms compared to the larger 5-mer fragments (data not shown). A comprehensive study on the impact of fragment type, size, charge and applied collision energy was not conducted as it was out of the scope of the current study. Follow-up research will focus on elucidating oligonucleotide fragments and studying the corresponding IMS separation.

In addition, we applied the same fragment-based workflow to the sense strand of siRNA#1 (Fig. 4). While we see the expected separation of four peaks at the 3′ end, the characterization of the GalNAc cluster on the 5′ end seems to be more difficult. We observe a broadened signal with no apparent separation between peaks, although we would expect up to 16 diastereoisomers. A significant difference between the PS linkages in the GalNAc cluster and PS linkages inside the oligonucleotide backbone is the flexibility of the alkyl linkers. It seems that a certain level of structural rigidity is necessary to observe isomer separation, and too much flexibility may increase conformational diversity to a point that it diminishes separation. This behavior is observed in both gas phase (Fig. 4) as well as in solution (Fig. 2, bottom row) and therefore these specific PS diastereomers seem to be challenging to resolve. The well-resolved peaks for the 3′ end of the sense strand, however, demonstrate the applicability of the fragment-based methodology for the characterization of sense strand diastereomers.

Fig. 4.

Mobilograms of the fragments generated from the sense strand of siRNA#1. Mobilogram of the terminal fragment containing the GalNAc cluster from the 3′ end (c₁³⁻) is shown on the left, and a representative fragment from the 5′ end (w₃²⁻) on the right side

Up to this point, we successfully repeated the experimental procedure and workflows from Sharon et al. [19] and have demonstrated the feasibility of qualitatively characterizing all relevant PS linkages in an siRNA by generating characteristic fragments of each single strand. The subsequent step involves assessing whether these fragments can also be used quantitatively.

Quantitative assessment of diastereomers via IMS

It was demonstrated in the literature that IMS data can be used for relative quantification of smaller biomolecules [27] as well as for larger sgRNA [20], which would be useful to allow a more comprehensive characterization of the overall diastereomer ratio for a given siRNA. In theory, the relative quantification of the generated fragments should not be dependent on the FLP ionization, but rather on the fragmentation efficiency of the generated b and w fragments. In addition, as we activate PO bonds and generate the fragments outside the PS region (3-mer to 5-mer fragment size), there is reason to believe that fragmentation efficiency should be the same independent of the diastereomeric composition at the terminal end of the single strand. To prove the quantitative behavior of IMS fragments, we require an accepted benchmark to verify the IMS results. In our case, this is currently only possible by using LC-UV data, as most other types of analytical techniques, such as NMR spectroscopy, do not yield individual diastereomer separation and quantification [14]. Due to the nature of the IMS fragment-based approach, however, it is currently not straightforward to compare IMS results directly with LC-UV data, as IMS results are provided for both ends of the single strand with only respective two PS linkages, while LC-UV delivers the results for all four. A statistical evaluation of the gas phase results, however, might enable an indirect comparison (Table S6). Although the IMS results represent the diastereomer distribution for each individual side of the single strand, the Cartesian product of both data sets should reconstruct the relative intensities of the original 16 diastereomers for the intact single strand with four PS. This could be relevant for regulatory submissions or just in general to present overall diastereomeric ratios for complete siRNA single strands instead of fragment data only. We did this exercise and calculated the relative intensities for all possible diastereomers in the antisense strand of siRNA#1 (Table S6), but without knowing the identity of each specific diastereomer, it is difficult to compare to LC-UV datasets which exhibit a different specificity and therefore a different elution order. In addition, LC-UV data does only partially resolve all possible diastereomers, which increases the difficulty of finding similarities between the relative peak areas. Similar to the previous LC development, we therefore decided to reduce the complexity of the analysis by reducing the number of PS linkages in the antisense strand to enable a meaningful comparison of solution and gas-phase data. To evaluate the quantitative capabilities of the fragment-based IMS, we synthesized two variations of the antisense model compound siRNA#1 with only two PS linkages either on the 3′ end (termed siRNA#1 PS-1+2) or on the 5′ end (termed siRNA#1 PS-3+4). This design reduces the amount of diastereomers from 16 to 4 and should allow full resolution of the diastereomers via our optimized LC-UV methods. In addition, this design is tailored to allow a direct comparison between our results from the fragment-based IMS approach and LC-UV detection as quantitative benchmark.

The measurement of the two simplified model compounds via LC-UV confirms the complementary nature of IP-RP and AEX chromatography (Fig. 5, left side). The chromatogram of IP-RP for siRNA#1-PS1+2 shows insufficient separation of the expected four diastereomers as we only observe one large peak with a pronounced shoulder and a second sharp peak. For the same compound, AEX chromatography reveals four baseline-separated diastereomer signals. Even more interesting is the comparison of both chromatographic modes for siRNA#1 PS-3+4, which exhibits the same sequence but differs in the position of the two PS linkages, which are now on the other end of the single strand. Here, we can observe a nice baseline separation of the four diastereomers for IP-RP, while AEX shows four peak tops, of which two peaks are partially coeluting. The difference in resolution power for IP-RP and AEX between both model compounds underlines the sequence dependence of chromatography in regard to diastereomer separation. What is also apparent from the results of siRNA#1 PS-3+4 is that both chromatographic modes exhibit a different selectivity for diastereomers, which is observed in a distinct peak pattern. IP-RP shows a high-low-high-low intensity pattern, while AEX exhibits a high-high-low-low pattern.

Fig. 5.

Comparison of diastereomer characterization via IP-RP (left column), AEX (middle column), and IMS (right column) for the model compounds siRNA#1 PS-1+2 (top row) and siRNA#1 PS-3+4 (bottom row). IP-RP and AEX were measured in solution for the intact single strand, while IMS data was generated in the gas-phase after activating the intact single strand to produce terminal fragments. The numbers represent the relative peak areas for the observed signals

We then also repeated the fragment-based IMS approach on the two model compounds and generated the same b₅²⁻ and w₅³⁻ fragments as used before (Fig. 3b), which revealed four peaks in a similar peak pattern and intensity as observed previously. The IMS results of the 2 × PS compounds are shown in Fig. 5 in the right column. After integration of the four signals observed in the mobilogram, we compared the quantitative results with the peak ratios from AEX and IP-RP UV measurements. We assume that the results between solution and gas phase can be directly compared although we are looking at an intact single strand versus a small fragment ion. The reason is that we are comparing the capability to distinguish diastereomers, and while additional PO linkages and building blocks might impact the IMS resolution, they should not impact the ratio of observed diastereomers.

As observed before for the 4× PS compounds, IMS exhibits a high selectivity for diastereomers that is expressed in an individual peak pattern and differs in elution order from the chromatographic modes. It therefore seems that all separation modes show complementary behavior to each other. AEX and IP-RP show a sequence-dependent resolution, as IP-RP is not capable of separating siRNA#1 PS-3+4, whereas AEX is unable to separate siRNA#1 PS-1+2, while only IMS is able to resolve all compounds in this study. It therefore seems that IMS could be more sequence-independent (at least for our model compounds). One reason for that could be in the easier tuning of the cyclic IMS instrument compared to chromatographic separation, as it is straightforward to tune the resolution to a certain degree and thereby the separation by simply optimizing the travelling wave height and increasing the number of rounds in the drift cell.

Although we observe the same number of signals in similar quantities in all three separation modes, a direct comparison is still challenging as observed in the datasets. To circumvent this problem, we assumed that the four signals observed in AEX and IP-RP can be identified as the same four signals in IMS. While the order of peaks is different and it is difficult to assign a specific diastereomer identity to each signal, the relative intensities are similar once ordered from low to high intensity (Table 1).

Table 1.

Comparison of UV and IMS relative peak areas for observed diastereomers. The left side of the table shows comparison for siRNA#1 PS-1+2 and the right side shows comparison for siRNA#1 PS-3+4. Difference describes absolute difference (in % peak area) between IMS compared to AEX and IP-RP

siRNA#1 PS-1+2				siRNA#1 PS-3+4
IP-RP	AEX	IMS	Difference	IP-RP	AEX	IMS	Difference
39%a	17%	16%	1%	18%	20%	17%	2–3%
	23%	21%	2%	22%	22%	21%	1%
27%	28%	29%	1–2%	29%	29%	30%	1%
34%	32%	34%	2%	30%	29%	32%	2–3%

^aIP-RP was not able to resolve these two diastereomeric peaks while AEX and IMS could, so IP-RP was not used for the comparison of these two signals

Once ordered in increasing intensities, we observe a remarkable agreement between the datasets from all three methods. The differences in relative peak areas are less than 2% for PS diastereomers from the 3′ end and less than 3% from the 5′ end, indicating a strong potential for IMS in quantitative diastereomer analysis. The distinct peak patterns for the 3′ and 5′ ends in IMS, as well as in AEX and IP-RP chromatography, suggest that the results are specific to the diastereomeric ratios rather than random. Our results support the hypothesis that IMS can provide meaningful quantitative insights while achieving full separation of all possible diastereomers.

From a quality control perspective, however, IMS represents a challenging task to be included in regulated environments. The largest hindrance right now is the general knowledge gap around the robust and predictable generation of suitable fragments for IMS separation for a variety of sequences and modifications. Once this challenge is overcome, e.g., by modelling the three-dimensional structure of fragment ions and theoretically calculating CCS values, it could serve as a benefit for IMS. An improved understanding of how to generate suitable fragments in combination with theoretical calculations would replace the need for stereopure reference standards and pave the way for automation. Right now, IMS can potentially be implemented in early drug discovery and drug development stages as a research tool. Here, it may provide quick preliminary insights into the diastereomeric composition of siRNA drug candidates without the need to optimize chromatographic methods for each new sequence. Furthermore, the IMS workflow demonstrated here has potential for further enhancement. Instead of stopping after generating 5-mer fragments in the gas phase to fingerprint the terminal two PS linkages, the IMS-IMS capabilities of the cyclic IMS instrument could be employed to generate sequential fragmentation of smaller terminal and internal fragments to measure their mobility in the gas phase. This approach allows reducing sample complexity even further and results in a single PS linkage per fragment. Such a workflow is described in Fig. S7, and preliminary data for the GalNAc-conjugated sense strand of siRNA#1 to isolate the two PS linkages from the 5′ end is shown in Fig. S8. While this workflow may not be absolutely necessary for siRNAs due to the relatively low number of PS linkages in the beginning, it could prove to be invaluable for the characterization of more complicated phosphorothioated compounds such as antisense oligonucleotides (ASOs).

Conclusion

In this study, we have explored and compared various analytical methods for the characterization and quantification of PS diastereomers in siRNA single strands. Our investigation focused on optimizing chromatographic methods, specifically IP-RP, AEX, and HILIC, and evaluating the potential of ion mobility mass spectrometry as an alternative approach to these chromatographic methods. Our findings indicate that among the chromatographic methods, AEX demonstrated the highest diastereomer selectivity, followed by IP-RP, while HILIC showed limited separation capabilities. However, neither AEX nor IP-RP achieved complete separation of all diastereomers, which underscores the complexity and challenges of chromatographic diastereomer analysis for siRNAs. Despite this, AEX showed promise as a routine quality control method in pharmaceutical drug development due to its superior resolution and ability to evaluate batch-to-batch consistency. The use of IMS provided a compelling alternative, particularly with the fragment-based approach, which allowed for full resolution of diastereomers in the gas phase. Our comparative analysis revealed that IMS not only qualitatively characterizes all relevant diastereomers but also exhibited a strong correlation with quantitative data generated by LC-UV methods. This suggested that IMS can be effectively utilized for both qualitative and quantitative diastereomer analysis. In conclusion, while chromatographic methods remain valuable for routine quality control, IMS presents a promising complementary approach, especially for comprehensive diastereomer characterization. The combination of these techniques offers a robust framework for ensuring the consistency and efficacy of siRNA therapeutics, thereby supporting the continued advancement of oligonucleotide-based medicines. Future studies should focus on further validating these methods with stereopure compounds and exploring their applications for a broader variety of oligonucleotide sequences and modifications.

Supplementary Information

Below is the link to the electronic supplementary material.

Biographies

Mona-Katharina Bill

is currently a senior scientist in analytical chemistry working in the CMC group at Silence Therapeutics. Her research focuses on the development and validation of LC-UV/MS methods for siRNA drug substances and structure elucidation of related impurities by high-resolution MS/MS sequencing. She holds a PhD in analytical chemistry and has several years of experience in natural product research focusing on the isolation and characterization of small molecules.

Micaela Graglia

is a senior scientist in analytical chemistry in the Oligonucleotide Chemistry group at Silence Therapeutics. She is focusing on the research and development of analytical methods for the characterization of oligonucleotide conjugates. She holds a PhD in colloidal chemistry from the Max Planck Institute and has several years of experience in the analytical characterization of biopolymers and small molecules.

Jo-Anne Riley

is an applications chemist in the EMEA Applications Laboratory at Waters in Wilmslow, UK. Since completing her chemistry PhD at the University of Southampton 12 years ago, her focus and passion has primarily been the characterization of oligonucleotide therapeutics using high-resolution mass spectrometry.

Mateusz Zugaj

is a senior research associate in analytical chemistry in the CMC group at Silence Therapeutics. His focus is on supporting the development and validation of chromatographic methods for the analysis of starting materials and siRNA-based drug substances.

Nadeschda Demtschenko

is a research associate in analytical chemistry in the Oligonucleotide Chemistry group at Silence Therapeutics. She focuses on the quality control of oligonucleotide conjugates and supporting the development of LC-MS-based analytical techniques. She has several years of experience in the characterization of small molecules in food and environmental matrices by LC-MS/MS.

Jonathan Fox

is currently a consulting scientist in the Biologics Team at Waters in Wilmslow, UK. He is passionate about working with customers and developing solutions to their analytical challenges utilizing Waters technologies and software. With over 24 years of mass spectrometry experience supporting customers in the field with training, developing and troubleshoot a wide range of applications across various Waters systems, he is now currently developing application solutions in the Biologics team. His areas of expertise include ion mobility, high-resolution MS for analysis of oligonucleotides, biologics, proteomics, DMPK, APGC, PFAS, food & environment and metabolomics.

Lucas Bethge

obtained his PhD in 2011 in bioorganic chemistry. Currently, he is Vice-President and leading the Oligonucleotide Chemistry group at Silence Therapeutics. He is an expert in the synthesis, modification, and conjugation of oligonucleotides with over 14 years of experience in industrial research and development at the interface between early research and clinical development.

Jeske J. Smink

obtained her PhD in 2003 and is currently Senior Director CMC at AiCuris Anti-infective Cures AG. As a CMC professional, she has over 13 years of experience in the biotech industry with a strong interest in advanced therapeutics, including cell therapies, biologics, and RNA therapies.

Christian Manz

is an analytical chemist specializing in the mass-spectrometric characterization of modified oligonucleotides, including ASOs and siRNAs. During his time at Silence Therapeutics, he developed and qualified advanced LC-MS methods for comprehensive oligonucleotide analysis and impurity profiling in drug manufacturing. Now at AstraZeneca in Sweden, he is responsible for the development of platform methods for high-throughput and semi-automated oligonucleotide characterization supporting preclinical studies. His research further integrates ion mobility spectrometry for isomer separation and MS/MS sequencing for detailed impurity analysis.

Author contribution

Conceptualization: MKB, MG, JAR, JF, CM. Formal analysis and investigation: MKB, MZ (AEX, IP-RP); MG, ND (HILIC); JAR (IMS). Visualization: MKB, CM. Writing—original draft: MKB, MG, JAR, CM. Writing—review and editing: MKB, MG, JAR, LB, JJS, CM. Supervision and resources: JF, LB, JJS.

Data availability

Due to confidentiality, the datasets supporting the conclusions of this article are not publicly available. Data may be made available from the corresponding author upon reasonable request and with permission from Silence Therapeutics.

Declarations

Conflict of interest

The authors declare no competing interests.