Observations from a decade of oligonucleotide bioanalysis by LC-MS

ABSTRACT

There is a growing need for efficient bioanalysis of oligonucleotide therapeutics. This broad class of molecules presents numerous challenges relative to traditional small molecule therapeutics. Methodologies including ligand-binding assays or polymerase chain reaction may be fit-for-purpose in many instances, but liquid chromatography coupled to mass spectrometry (LC-MS) often delivers the best balance of sensitivity and selectivity. Over the last decade, we have engaged with many such molecules and derived insights into challenges and solutions. Herein, we provide four case studies illustrating challenges we have encountered. These issues include low or variable analyte recovery, poor resolution from related species, chromatographic abnormalities or challenging sensitivity. We present a summary of considerations, based on these experiences, to assist others working in the area.

Plain language summary

Executive summary.

• We have validated over 80 chromatographic methods for analysis of oligonucleotides (OGNs) in the past decade, with significant growth in demand noted over recent years.

Case studies

• We present four case studies highlighting the challenges associated to LLE and SPE-based extraction workflows, with recommendations on how best to optimize these. Common pitfalls such as low/variable recovery, dirty extracts and reagent integrity are discussed.

• The choice of triple-quadrupole MS/MS versus HRMS is important. We show case studies highlighting the benefits of each for optimum sensitivity and selectivity, respectively.

• Challenges unique to a fatty acid conjugated oligonucleotide are highlighted.

• The need to consider the potential impact of metabolites is shown.

Conclusion

• A list of considerations for LC-MS methods has been provided from our experience.

• Protease digestion and hybridization-based LC-FD or LC-MS approaches are gaining momentum. However, they will inevitably bring their own unique challenges.

• Hence, LLE/SPE with LC-MS workflows will likely remain popular.

Interest in oligonucleotide (OGN) therapeutics, and their ability to impact biological pathways which do not readily present active sites for traditional therapeutics, continues to grow. Several classes of such molecules present a variety of chemical and biological considerations for delivery and analysis [1,2], introducing unique complexities for developing methods for quantitative analysis.

Numerous publications exist that comprehensively review the technologies and techniques used for OGN analysis [^3-8]. It is not our intention to repeat what others have already done so well, so this article is intended as a sharing of specific experiences rather than as a review of the subject.

We encounter a wide range of potential drug molecules; indeed, we supported over 80% of all US FDA-approved drugs, through some aspect of their development, across our enterprise in 2022. However, as the molecules do not belong to us, we are often limited in what we can share. Therefore, rather than a scientific publication detailing the intricacies of a single program, our case-studies provide snapshots of selected examples, all using well-established techniques and instrumentation, where we feel the lessons learnt may be relevant to others.

In preparing this manuscript, we compiled statistics for all OGN bioanalytical methods validated using liquid chromatography-mass spectrometry (LC-MS) or liquid chromatography fluorescence detection (LC-FD) by Labcorp Bioanalytical Services during the last 10 years (2014–2023). These methods are primarily in support of pharmacokinetic analysis in plasma or urine samples. In total, 87 methods were identified, comprising a wide range of species, matrices, extraction and analysis approaches and lower limits of quantification (LLOQ). Inspection of this dataset yields several interesting insights.

First, sorting by year of validation (Figure 1A), a clear increase was observed in the number of OGN methods validated in each year, with 2019–2023 averaging over twice the number of methods validated per year as compared with 2014–2018, and 2023 presenting the highest activity to date. This increase is significant even when normalized against expanding capabilities for all molecule classes over the same period (data not shown). Furthermore, based on ongoing and awarded projects at the time of writing, this is a trend set to continue or even accelerate. As Labcorp is a large bioanalytical laboratory service provider with a broad client base, we believe this is likely representative of an overall trend within the field of bioanalysis/drug development.

Figure 1.

Statistical overview for the 87 OGN bioanalytical methods validated by Labcorp over the past 10 years, sorted by year validated (A), LLOQ associated with the assay (B), extraction type (C) and detection platform used (D).

Second, sorting methods by required LLOQ (Figure 1B), we have identified several tiers; 15 methods (17%) required an LLOQ <1 ng/ml, 40 methods (46%) required an LLOQ from 1 to 10 ng/ml, 26 methods (30%) required an LLOQ >10 to 100 ng/ml, and 6 methods (7%) required an LLOQ >100 ng/ml. These data indicate that most OGN programs (63%) require an LLOQ of 10 ng/ml or lower, with a significant proportion requiring <1 ng/ml, which can be particularly challenging.

These data are helpful in understanding what LLOQ is generally needed to support modern OGN programs, although there may be selection bias, as programs requiring lower LLOQs and/or having larger oligonucleotides (>30 nucleotides) may have been supported using ligand binding assays (LBA) due to actual or perceived limitations of the LC-MS platform. It is also worth noting that for some programs, due to the dose route or the study aims, samples spanning >1000-fold order of magnitude may need to be analyzed. Due to LC-MS assay range limitations associated to crosstalk, carryover, or linearity, two assay ranges (low and high) have been established for some programs.

Third, a range of extraction techniques are noted (Figure 1C). Several methods use liquidliquid extraction (LLE) with phenol: chloroform: isoamyl alcohol (25:24:1), which also precipitates proteins, or dichloromethane. Using either of these solvents, the analytes are retained in the upper aqueous layer. Many methods use mixed-mode solid phase extraction (MM-SPE), which retains the analyte by a combination of reversed phase and ion exchange, typically using a sorbent with C18 and weak base functionalities, for example, the Phenomenex Clarity^® OTX kit or Waters Oasis^® WAX. Other methods use reversed phase solid phase extraction (RP-SPE), for example, the Waters Oasis^® HLB sorbent. In many cases, a combination of these is applied to achieve a successful method. The final combinations used in our 87 methods are LLE + LLE (typically phenol-chloroform followed by dichloromethane extractions, 30 methods), MM-SPE (26 methods), LLE + RP-SPE (LLE usually being phenol-chloroform, 17 methods) and standalone RP-SPE (5 methods). All these combinations are discussed in our four case studies in this paper. Common to all these methods is the need to disrupt the strong binding of OGN drugs to matrix proteins at the start of the extraction process, which is achieved using phenol-chloroform extraction, or denaturation with guanidine hydrochloride.

Finally, nine methods utilize Proteinase K digestion followed by hybridization to a fluorescent probe, which is accompanied by liquid chromatography-fluorescence detection (LC-FD). However, the focus of this paper and our case studies is ‘traditional’ LLE and SPE extractions followed by LC-MS analysis using triple-quadrupole MS/MS or high-resolution instruments. Therefore, we will not elaborate further on LC-FD, or the rapidly emerging technique of hybridization capture coupled to LC-MS detection.

Figure 1D groups our methods by detection platform. Most of our validated methods (54) use tandem mass spectrometry (MS/MS) via Sciex triple quadrupole instruments. A much smaller number (20) use high-resolution mass spectrometry (HRMS), via the Thermo Scientific Q-Exactive™ platform. The lower use of HRMS may be because, prior to the time of writing, HRMS instruments were not available at all our sites, but also because HRMS has historically struggled to achieve <10 ng/ml LLOQs, which Figure 1B shows to often be needed. Four of our methods (including one of our case studies) used both LC-MS/MS and HRMS in a single method. The nine methods which used proteinase K digestion and fluorescent hybridization were validated using LC-FD detection.

In summary, a number of extraction types and detection approaches have been utilized in the past 10 years. We see that OGN therapeutics can be very diverse in their behavior, dictated in-part by whether they are single or double-stranded, and by the modifications they present. The assay requirements will depend critically on potency, dose route, metabolism and study objectives. Having a diverse collection of platforms and expertise is important to support the equally diverse collection of OGN programs (and subsequent study needs) that currently exist in the drug development world.

Also beyond the scope of this publication is a wider review of chromatographic approaches. The examples herein use ion-pair reversed phase (IP-RP) chromatography, reviewed by several previous authors [3,^5-8], however, ion-exchange chromatography with fluorescence detection may be used [8], and hydrophilic interaction liquid chromatography (HILIC) with MS detection has been reported [9,10].

The following four case studies reflect many of the LLE and SPE-based extraction and MS detection techniques mentioned above, applied to assays for a variety of different OGN therapeutics. They provide specific examples of where we have encountered challenges or unexpected behavior, and where the lessons we have learnt may be beneficial for others pursuing similar workflows.

Case study 1: challenges of reagent integrity & extraction robustness using the Clarity^® OTX extraction kit & the use of HRMS & triple-quadrupole MS/MS in one method

Background

An LC-MS method was developed and validated to measure the sense and antisense strands of an investigational siRNA therapeutic in human plasma. Two metabolites (formed by truncation of the 3′ end of the antisense strand) were also measured. The method used MM-SPE with the Clarity^® OTX kit for the extraction of 150 μl of plasma, with a range of 10 to 1000 ng/ml for all four analytes.

The LC separations were performed using a Waters^® XSelect CSH column maintained at 80°C to ensure robust melting of the duplex, with water and methanol mobile phases containing diisopropylethylamine (DIPEA), 1,1,1,3,3,3 hexafluoroisopropanol (HFIP), and a small amount of ammonium hydroxide to promote robust deprotonation. Detection used the Q-Exactive™ (Thermo Fisher Scientific) Orbitrap™ HRMS system, with each of the analytes quantitated by summing the four most prominent isotopes of the three most abundant charge states (a sum of 12 m/z values).

The two internal standards (ISTDs), known as ISTD1 and ISTD2, were analog duplex oligonucleotides. The sense strand of ISTD1 was used to quantitate the parent siRNA sense strand, the antisense strand of ISTD1 was used to quantitate the parent siRNA antisense strand and one metabolite, and the antisense strand of ISTD2 used to quantitate the other metabolite. The ISTDs were measured using the four most abundant isotopes of the two most abundant charge states (a sum of 8 m/z values).

Methodological challenges

During sample analysis, full calibration curves were injected at the beginning and end of each analytical batch. Problems were encountered with these two curves showing divergent response (with one curve consistently showing higher response factor than the other) for each of the four analytes. Additionally, the overall signal associated with the second metabolite was found to be too low to enable robust performance at the LLOQ concentration.

Re-injection of first and second calibration curves showed an identical pattern, which implied the cause of the divergence was the extraction, rather than LC-MS system response drift during analysis of the run. During subsequent troubleshooting, recovery was measured at 20–30%, which was much lower than the 60–70% observed in the original validation. Several aspects of the extraction were interrogated to attempt to determine the source of the problem, including testing of load/wash solvents, and testing recovery prior to and following evaporative concentration and resuspension. During the investigation, it was noted that there were large and unexpected ion peaks present in the mass spectra obtained from extracted spiked plasma samples, which were absent from blank samples fortified with the analytes post-extraction (Figure 2). The additional ions were observed at -16 and -32 Da (relative to the targeted analyte ions), consistent with oxidation of the OGN phosphorothioate backbone.

Figure 2.

Representative spectra for the second metabolite of case study 1.

Mass spectra are shown for the z = -10 isotopes derived from a plasma sample spiked and the High-QC level and extracted (A), as well as blank plasma extracted, and spiked with the second metabolite post extraction (B). The low recovery was attributed to oxidation, as evidenced by the presence of large oxidation peaks.

The issue was found to be related to tetrahydrofuran (THF) used during the elution step (elution performed using 100 mM ammonium bicarbonate in acetonitrile:THF:water (4:1:5), pH 8.8). Use of freshly prepared elution solution, prepared from a freshly opened bottle of THF, was found to produce mass spectra containing minimal evidence of oxidation. However, use of previously opened THF or elution solution which was not freshly prepared would result in sample extracts showing significant oxidation. As a long-term solution, the use of butylated hydroxytoluene (BHT) stabilized THF was found to further inhibit oxidation. We have used BHT before in several methods as an antioxidant for lipophilic, oxidation-prone analytes. The divergent calibration curve issue was attributed to differences in oxidation experienced by the analytes, but less-so by the internal standards, during the 2-h sample evaporation, caused by slight differences across the wells of a 96-well plate.

Even with the use of BHT-stabilized THF and the resulting improvement to recovery, the signal was still too low for the second metabolite to be measured at the LLOQ on the HRMS platform. Efforts to improve the signal in the context of the original method were unsuccessful. However, adequate and robust signal for the second metabolite was observed upon injection on a Sciex API 6500+ triple quadrupole MS/MS instrument operating in negative electrospray mode and measuring an abundant product ion. The method was modified and partially revalidated to allow each extracted plate to be injected in two ways. The first injection utilized HRMS detection as described earlier for the sense strand, antisense strand, and first metabolite (and associated internal standards). The second injection was used to quantitate the second metabolite and utilized triple quadrupole MS/MS detection for greater sensitivity.

Further complications then arose after implementing the above changes. We found that the parent siRNA sense and antisense strands, and the first metabolite all showed unacceptable variability, causing batch failure. As part of the investigation, these analytes were quantitated using the alternative ISTD, which gave significantly improved results. This was interpreted to be due to variable recovery (from well to well within the 96-well plate) of the first internal standard. As the method used MM-SPE, this was believed to be due to variable pressure applied to each of the wells, as the load, wash, and elution steps were performed using a positive pressure manifold. Rather than re-validate with the alternative ISTD, the extraction was adjusted, with centrifugation for passing the load, wash, and elution solvents through the MM-SPE sorbent, with defined speed and duration, rather than using positive pressure. This adjustment was successful in addressing the variable recovery of the first internal standard, and assay performance was returned to acceptable levels, with minimal additional validation requirements.

Lessons learnt

There are several lessons from this case study which may be relevant to others using MM-SPE workflows. First, if THF is required for the extraction, as recommended by some manufacturer protocols, the use of freshly opened BHT-stabilized THF (commercially available, 100–300 ppm BHT), and fresh preparation of the reagents may be critical. Additionally, for SPE-based extractions of OGN therapeutics (RP-SPE and MM-SPE) where the capture or elution efficiency may be low, the use of low-speed centrifugation to pass liquid through the sorbent, rather than positive or negative pressure, may improve reproducibility day-to-day and between different operators.

Overall, the interaction of the OGNs with MM-SPE or RP-SPE sorbents (either initial retention of subsequent elution) is often very complex and prone to variability from several factors. For this reason, we are moving away from SPE-based approaches as a primary choice for many OGNs, and instead we favor LLE-LLE approaches, for example, phenol-chloroform extraction followed by dichloromethane extraction, as discussed in some of our examples below. Where such LLE-LLE workflows yield adequate cleanup and sensitivity, they seem to be more robust and less prone to uncontrolled variables.

Finally, an additional lesson learnt is that in some cases, triple quadrupole MS/MS analysis may offer superior sensitivity compared with HRMS. One benefit of HRMS is the ability to differentiate between structurally similar metabolites (e.g., truncated metabolites [11] or deamidated metabolites, see case study #4) where triple quadrupole instruments lack this mass resolution, especially when each OGN generates numerous minor precursor ions via multiple adducts of several charge states or losses and rearrangements of the molecule under the conditions of electrospray ionization. However, where the metabolism is well characterized or not expected to present an issue, and the risk associated with structurally similar metabolites are not considered significant, triple quadrupole MS/MS can deliver superior sensitivity. Both classes of instrument have their benefits, and where both are available, careful consideration of the study aims, assay requirements and molecule chemistry prior to deciding on an instrument platform is recommended.

Case study 2: challenging extraction recovery for a fatty acid modified oligonucleotide

Background

We were asked to establish rat and minipig plasma, and later, human plasma and urine methods for a C18-fatty acid conjugated single stranded OGN of ∼6500 Da, with common nucleotide stability-enhancing modifications. The target LLOQ was 5 ng/ml, using an analog internal standard OGN with an identical sequence but different lipid modification. The choice of detection technique was MS/MS with a Sciex API6500+ triple-quadrupole system, as HRMS was not available at the target site at the time.

Methodological challenges

At the start of this work, we were unclear how much impact the fatty acid would have on the behavior of the molecule. Initially, the molecule demonstrated LC-MS/MS behavior characteristic of, and defined by its OGN backbone, with several charge states observed, and requirement for the standard ion pair reagents Triethylamine (TEA) and HFIP to achieve good retention and peak shape by reversed phase chromatography, using approaches that have been extensively reviewed in the literature (3,5,6,8). The chromatographic retention was slightly stronger than most OGNs, owing to the fatty acid modification.

We tested extraction from plasma using one of our well-established approaches, phenol-chloroform (LLE) extraction followed by RP-SPE. We found that this analyte demonstrated very low recovery, despite the success of this workflow on other molecules [11,12].

We noted that other authors have tried to apply similar workflows to another lipid modified OGN (Imetelstat). They too reported issues of low recovery using conventional workflows that are otherwise usually successful for non-lipid modified OGNs [13]. We hypothesize that the lipid-modified OGNs may be more strongly bound to matrix proteins and are not fully released by phenol-chloroform LLE step (which concurrently precipitates the proteins). Alternatively, the fatty acid group may decrease overall molecular polarity to a degree preventing the OGN from partitioning effectively into the aqueous layer during LLE extractions.

The authors for the Imetelstat work [13] instead used the MM-SPE approach via the Clarity^® OTX kit, but experienced several challenges, resulting in a complex method. Based on our prior experience with this kit, which are detailed in our previous case-study, and the much lower LLOQ requirements of our assay (5 ng/ml), we did not see this as a viable solution.

Our previous work [11] had shown that one other potential technique was available to us; direct RP-SPE (with no prior phenol-chloroform extraction) using 6M guanidine to denature the interactions between the drug and plasma proteins. Because the retention by this technique relies entirely on reversed phase interactions, ion pair reagents must also be added to the sample in addition to the guanidine to reduce the apparent polarity of the OGN and enable retention on the RP-SPE sorbent.

The plasma sample (25 μl) was aliquotted into a 96-well plate, internal standard added (25 μl), and then 6M guanidine hydrochloride (250 μl). This was mixed, and Water: TEA: HFIP (100:0.5:2) was added (250 μl). The resulting mixture was loaded onto an equilibrated Waters^® Oasis^® HLB 10 mg plate, and then the plate was washed, eluted, and the eluate evaporated to dryness, and reconstituted using a workflow similar to one that we have previously reported [10].

The first internal standard tested (a cholesterol conjugated OGN) showed very poor results, with behavior and recovery not mirroring that of the analyte. However, the sponsor was able to supply an alternative, shorter chain fatty acid analog, which worked perfectly.

Chromatographic deterioration became problematic, particularly for the minipig assay. The use of guanidine for the sample pre-treatment causes the denaturation of many plasma proteins and releases a variety of other protein-bound molecules. Many of these molecules, and also the linearized proteins, are consequently able to retain on the RP-SPE sorbent whereas without guanidine they would simply flow through without interacting. After evaporation of the final extract, a large amount of dry residue was present which reconstitutes to give very yellow, dirty extracts.

The resolution to this was to reduce the amount of plasma loaded onto the SPE plate, through a lower aliquot volume, and to minimize the LC injection volume. This required the LLOQ for the minipig assay to be raised from 5 to 20 ng/ml. For rat plasma, the deterioration was much slower, and the original method was workable providing a fresh analytical column was used for each run.

The second issue, for minipig plasma only, was an accuracy mismatch between fresh calibration standards and frozen QC samples. The freezing of the samples seemed to affect the analyte protein binding, causing a change to the recovery, for which the internal standard did not compensate.

The resolution was twofold: First, to increase the concentration of the guanidine solution added from 6 to 8M, to ensure the guanidine concentration in the sample/guanidine mixture was higher than 6M during the denaturation step. Second, to then add the ion pair reagents (which are essential for capture of the drug on the RP-SPE sorbent) with a lower volume (100 μl) of a higher concentration reagent. We found that water:TEA:HFIP (100:2:4) was the absolute limit of HFIP solubility. The aim was to ensure the guanidine concentration remained at 5M or higher after addition of the ion pair reagent and hence during loading onto the SPE sorbent. Previously, the guanidine concentration was reduced to 3.6 M upon adding the ion pair reagents, which may have resulted in some partial refolding of proteins prior to SPE, and the re-binding of the analyte to these proteins, causing variable recovery.

The amended method worked well and validated successfully in both animal species, achieving approximately 55% recovery, and formed the basis for the subsequent clinical methods.

Lessons learnt

Lipid modification of an OGN may have minimal impact on the LC-MS behavior of the molecule relative to a similar unmodified OGN, however, it can have a substantial impact on the extraction, as shown here, and by other authors [13]. Releasing OGNs from plasma proteins, ensuring consistently high recovery from all batches of plasma and irrespective of sample storage and handling, is made more challenging by fatty acid modifications. Traditional techniques such as phenol-chloroform extraction may not yield good recovery. The use of RP-SPE (or MM-SPE) including strong chaotropes, such as guanidine, is required to robustly recover the analyte from matrix. Any technique that increases recovery of such an analyte will also increase recovery of the proteins themselves and other molecules, causing dirtier extracts, column deterioration and more interfering peaks. The challenges we observed were very different between rat and minipig, perhaps due to proteome differences between the two species. It is essential to carefully optimize the amount of chaotrope or denaturant used to ensure reproducible recovery, while ensuring the resulting extracts do not cause chromatographic deterioration.

In all cases, a good internal standard bearing very similar chemical modifications, is essential to achieving a robust method. Use of an off-the-shelf OGN without similar modifications is unlikely to be successful for such a unique class of molecule.

Case study 3: fine-tuning of mixed-mode SPE extraction of a GalNAc-conjugated siRNA oligonucleotide using the Clarity^® OTX extraction kit

Background

An LC-MS method was needed to measure the sense and antisense strands of an siRNA-GalNAc conjugate having an LLOQ of 10 ng/ml for each strand. A provisional MM-SPE method was developed using the Clarity^® OTX kit, as used in our first case study, largely following the manufacturer instructions. Briefly, 100 μl of plasma (fortified at known levels with duplex material) was combined with 50 μl of internal standard in water and 150 μl of the supplied lysis-loading buffer (LLB, which contains 56–58% guanidine hydrochloride, with 1–3% Triton X-100). Samples were then loaded to the SPE plate, which was pre-conditioned with 1000 μl of methanol, and 1000 μl equilibration buffer (EQB, 50 mM ammonium acetate (aq; pH 5.5)). After sample loading, each well was washed with EQB. Samples were then eluted twice with 500 μl of 100 mM Ammonium Bicarbonate pH 9.5 in acetonitrile:BHT-stabilized THF:water (40:10:50) into a 96 well plate containing 50 μl of [10 mM EDTA (in acetonitrile:water:ammonium hydroxide (15:85:0.4)]: ThermoFisher Scientific RNA Secure (100:4), the later component added as an RNAse inhibitor to preserve analyte stability. Samples were evaporated to dryness under nitrogen and reconstituted in 200 μl of water:DIPA:HFIP (100:0.14:0.264).

Chromatography used a Waters Acquity BEH C18 column at 70°C and mobile phases of water and methanol containing 0.14% diisopropylamine (DIPA) and 0.264% HFIP. Detection was performed using a Q-Exactive HRMS platform operating in negative ESI mode and acquiring full MS scans from 1800 to 3000 m/z. Five isotopes from two charge states (10 m/z values in total) were summed for the sense and anti-sense strands and associated internal standards.

Methodological challenges

The first challenge encountered was blocking of the analytical column. As batches were injected, backpressure was observed to steadily increase, leading to the system going over-pressure and/or leaking. We observed that the sample extracts had a yellow color. Based on this observation, as well as experience with previous programs, we suspected that insufficient cleanup of samples was occurring, with an excessive amount of matrix components present in the final extracts. This situation closely mirrors the previous case study, which used guanidine denaturation for a RP-SPE workflow.

To mitigate this, the sample was subjected to additional dilution with LLB prior to loading onto the Clarity^® OTX SPE plate, with all other variables kept constant. This overcame the issues of column blocking but also dramatically reduced the analyte recovery and hence sensitivity. To screen analyte recovery as a function of guanidine concentration, we tested LLB and EQB mixed in varying proportion prior to loading to the SPE plate, with a total volume of 600 μl (Table 1). With LLB exceeding 1/6th of the total buffer, a persistent and negative impact was observed on the recovery of all compounds. The assay was amended to use the optimized loading conditions, with successful assay validation.

Table 1.

Mean recovery of sense/anti-sense strands and associated internal standards as a function of lysis-loading buffer/equilibration buffer proportion during optimization of Case Study 3.

lysis-loading buffer volume (ml)	Equilibration buffer volume (ml)	Mean recovery (%)†
0.100	0.500	100.0
0.200	0.400	64.2
0.300	0.300	53.6
0.400	0.200	38.7
0.500	0.100	34.4

^†Average recovery for the sense and antisense strands and associated internal standards.

The cause of the low recovery at higher proportions of LLB, which contains guanidine, was poor retention on the SPE plate (confirmed by analysis of the flow-through fraction). This is contrary to our previous case study, where recovery required much higher proportions of guanidine. We speculate that recovery in this case may vary with a multitude of parameters such as pH, ionic strength and detergent concentration, all of which would vary as a function of LLB:EQB proportion. This case study relies on an ion-exchange retention mechanism, whereas our previous case study relied only upon reversed-phase retention, which could explain the differences in requirements for high recovery. For example, higher guanidine concentrations could be more detrimental to ionic retention mechanisms. Alternatively, higher guanidine concentrations will result in greater protein denaturation, which could result in saturation of the Clarity^® OTX SPE plate with denatured protein or other molecules.

An additional observation during method development was that recovery was ∼20% higher at lower analyte concentrations, when measured against samples spiked after extraction to a concentration that mimics 100% recovery. One spiking solution was used to assess Low-QC level recovery, and one to assess High-QC level recovery, both prepared in water:DIPA:HFIP (100:0.14:0.264). The issue was found to be nonspecific binding losses of the OGNs from the solution used to spike after extraction for the Low-QC concentration assessment. For the high-QC solution, the adsorption loss was negligible. Upon a repeat of the experiment utilizing solutions containing 50 μg/ml bovine serum albumin (BSA) to block nonspecific binding, consistent recovery was achieved across the range.

Lessons learnt

This case study demonstrates the key differences between the requirements for sample pre-treatment when using different SPE modes (RP-SPE and MM-SPE), and between different analytes. In particular, our experiences with the Clarity^® OTX system closely mirror observations from other authors [14].

Our former case study showed that for RP-SPE of a lipid-modified oligonucleotide, extremely high guanidine concentrations were needed to achieve high recovery due to very strong protein binding, with the reversed phase capture mechanism not being affected by the high ionic strength. However, this case study for MM-SPE and a GalNAc modified siRNA shows that recovery drops with higher concentrations of the guanidine-based reagent, possibly due one or more of the following: the impact of the guanidine on ionic retention, the introduction of some pH change by the additional pH 5.5 EQB buffer, the impact of guanidine on a double-stranded oligonucleotide, possibly separating the two strands prior to SPE loading due to displacement of the H-bond interaction of the strands.

Careful consideration of how samples are treated prior to loading onto an SPE plate is critical. Some guanidine is likely to be needed; RP-SPE needs ion pair reagents too; MM-SPE may need careful control of ionic strength or pH to achieve optimum recovery, and this may vary depending on the nature of the OGN drug, for example, whether it is single or double stranded, and how it is modified. Furthermore, choices made at the start of the extraction can impact on extract cleanliness, and we have now seen in two examples that insufficient cleanup can cause issues of column blocking.

Finally, for recovery evaluations, which form a critical part of assessing extraction efficiency, great care is needed to ensure measured recovery is not being impacted by nonspecific binding in solutions used to prepared comparator samples. In such cases, introduction of a protein-rich diluent, or some organic solvent, for the spiking solution (e.g., BSA) may improve performance.

In relation to this, we find that preparation of spiking solutions of OGNs in water combined 1:1 with dimethylformamide (DMF), acetonitrile or methanol, to be best for avoidance of nonspecific binding. Recovery assessment solutions, prepared in such solvents for addition to the samples post-extraction, could have an adverse effect on the composition of the final extract, making it incompatible with the chromatographic start conditions. However, one approach to mitigate this is to spike the analytes post-extraction using a very low volume addition (e.g., 5 μl added to 100 μl final extract) such that the presence of adsorption blocking components such as BSA or organic solvent in the spiking solutions has no significant impact on peak shape or retention. This also requires the analyte to be present at much higher concentrations (e.g. 21-fold higher in the above example) to compensate for the dilution, thus also minimizing the impact of nonspecific binding. Second, in this case study, ISTD was added during the extraction in just water as a diluent, but for many OGNs, this will result in adsorption losses to the vessel and, worse, the pipette tip, which could cause downward drift in ISTD response across the run as it adsorbs more to the tip during the addition. We have observed this effect for both OGN and peptide assays, where an aqueous solvent is the instinctive choice. Inclusion of ∼20–50% organic solvent, BSA, or even a small percentage amount of control human plasma, in the internal standard working solution is recommended to help avoid risks of adsorption losses for all OGN assays. Entirely aqueous solutions without any kind of adsorption blocking component are generally not recommended at any stage.

Case study 4: unexpected metabolites & LLE-LLE as a more robust alternative to SPE workflows for a single stranded antisense oligonucleotide

Background

We were commissioned to establish an LC-MS/MS method in monkey plasma for a single stranded, intrathecal-dosed OGN (∼7000 Da), with commonly utilized second-generation nucleotide stability-enhancing modifications. The target LLOQ was 5 ng/ml, using an ISTD with similar modifications, two additional nucleotides and phosphodiester rather than phosphorothioate bonds. The approach used was LC-MS/MS with triple-quadrupole instruments.

Methodological challenges

During early optimization, the molecule demonstrated characteristic LC-MS/MS behavior for an antisense OGN, as would be expected from our own experience and previous reviews on the subject [6]. A reversed-phase LC method was established using a Waters Acquity BEH C18 50 × 2.1 mm 1.7 μm column maintained at 60°C, mobile phases of Water and Acetonitrile each containing 1% HFIP and 0.25% TEA, and a gradient ramping from 5 to 14% organic phase over 4 min at 0.45 ml/min. Such conditions are very typical for an antisense OGN assays [6,8,11,12].

Extraction of the OGN was performed using phenol-chloroform extraction as previously reported [10,11]. The aqueous layer following phenol-chloroform extraction was then diluted with ion pair reagents and further cleaned up using RP-SPE using a Waters Oasis HLB sorbent (10 mg/well). However, we observed issues with extremely variable recovery from the SPE. In some runs, as many as 10% of samples would show <10% recovery, even though the average recovery was >50%. Additionally, other samples showed randomly occurring very broad chromatographic peaks. We attributed this to phenol dissolving into the aqueous layer during the phenol-chloroform extraction (as phenol shows ∼8% aqueous solubility). The dissolved phenol could potentially cause saturation of the SPE sorbent leading to analyte breakthrough or it could elute into the final extract and cause chromatographic abnormalities.

Therefore, instead of RP-SPE, we used a secondary LLE extraction with dichloromethane (DCM). Non-polar interferences, including phenol, partition into the DCM, whereas the OGN remains in the upper aqueous later. This technique has proved to be robust for several molecules, including single and double stranded OGNs with a variety of modifications. However, as the aqueous layer cannot be fully recovered at each step due to the need to avoid aspirating too close to the phase partition, this approach results in high dilution of the analyte. For this reason, and also due to the complexities of SPE discussed in this and our other case studies, we now use LLE-LLE as a first choice only where the LLOQ is not challenging (≥1 ng/ml per 10 μl of plasma extracted, with triple-quadrupole MS/MS detection), and LLE + RP-SPE or MM-SPE are used only when lower LLOQs, HRMS detection, or lower sample volumes are needed.

Initial mock-validation testing of the method showed unusual adsorption issues which manifested as slight nonlinearity and poor accuracy at lower analyte concentrations. We identified that this was due to the low plasma volume being extracted (25 μl). We overcame this by adding commercially sourced control human plasma (75 μl) as a reagent at the start of the extraction to provide more matrix components through the extraction procedure to overcome nonspecific binding.

Briefly, the method involved aliquotting 25 μl of sample, adding 75 μl of Human Plasma (EDTA), 25 μl of internal standard and 350 μl of freshly prepared 10% ammonium hydroxide. The plate was mixed, 250 μl of Phenol: Chloroform: Isoamyl Alcohol (25:24:1) added, mixed again (30 min) and then centrifuged. The aqueous layer (300 μl) was withdrawn and mixed with 125 μl of water and 250 μl DCM, before mixing (10 min) and centrifuging. Finally, 150 μl of the aqueous layer was withdrawn, evaporated to dryness and reconstituted in 100 μl Acetonitrile: Water: HFIP: TEA (5:95:1:0.25), which matches perfectly the mobile phase starting conditions, prior to injection.

The second issue, encountered only during sample analysis, was the appearance of a wide, analyte-related peak eluting prior to the actual analyte peak. This was observed in all samples (both spiked and incurred) and the reproducibility was poor run-to-run. Remedial work identified that the column temperature (60°C) was just on the borderline of ‘melting’ secondary structures within the analyte molecule, and increasing to 80°C robustly overcame these issues without exceeding the maximum operating temperature for the column or adversely impacting on the chromatography.

The final, and most challenging issue encountered was the appearance of a second peak unique to incurred samples, eluting just on the front of the analyte peak, with less than 0.10 min separation. The integration software (Analyst 1.7.2, AB Sciex) was able to drop a perpendicular between the two peaks for most samples (Figure 3A & B), however, in some samples where the earlier eluting metabolite peak was larger than the analyte, achieving this integration consistently without sample-to-sample parameter adjustment (which is not permitted by our company policy), was not possible. These interferences were not present in spiked or blank samples (Figure 3C & D).

Figure 3.

Demonstrative chromatograms showing the analyte (left) and internal standard (right) from (A) An incurred sample showing a metabolite peak eluting just before the integrated analyte peak, (B) same chromatogram zoomed to analyte peak for clarity.

(C) For comparison, a spiked LLOQ (5 ng/ml) sample chromatogram and (D) blank plasma chromatogram on same scale as LLOQ chromatogram.

Subsequent identification work confirmed the metabolite to be a deamination of a deoxycytidine. The cytidines closest to the 3′ and 5′ ends of the molecule were protected from deamination by methylation, however, the interior cytidines were not. The deamination of one of the interior cytidines causes only 1 Da mass shift, which for the [M-9H]⁹-precursor ion being measured, gives only 0.11 m/z shift, meaning the metabolite and drug are indistinguishable by triple quadrupole MS/MS instruments.

Considerable efforts were made to chromatographically separate the two species, including a 20-min shallow gradient using a 100 mm Acquity BEH C18 column and mobile phases consisting of water and acetonitrile with 0.25% TEA and 1% HFIP. This achieved just enough separation to enable consistent integration, allowing samples to be analyzed for investigatory purposes.

After analysis of the data, the decision was made to continue with the original method. The metabolite was only observed for very late timepoints and did not impact on primary PK parameter determination for the molecule. The deamination may be a slow mechanism of metabolism or may only occur in specific compartments (e.g., tissue), requiring uptake of the drug and export of the deaminated metabolite back into the plasma before this starts to appear in systemic circulation.

Lessons learned

There are many excellent publications of OGN methods in the literature, providing a diversity of strategies for sample extraction (6). Phenol-chloroform extraction followed by RP-SPE is widely reported [11,12] but we see issues for some molecules. The use of a dichloromethane extraction of the aqueous layer after phenol-chloroform extraction provides a simple, quick and cheaper alternative, but may struggle to achieve LLOQs much below 1 ng/ml for many OGNs.

Use of too little plasma in the extraction can result in ‘over-cleaning’ of the extract which leads to nonspecific binding of some OGNs during downstream processing. As a general strategy, we now always aim to ensure at least 100 μl of total plasma is taken forward into the extraction using the LLE-LLE workflow. Control human plasma may be added as an extraction reagent to achieve this. The extracts generated by LLE-LLE are surprisingly cleaner than many SPE-only approaches due to the avoidance of guanidine, which our previous case studies have shown leads to very dirty extracts.

We also now prefer to utilize ≥80°C as a column temperature for all OGNs, to ensure more robust denaturation and avoid secondary peaks caused by the on-column separation of secondary structure forms of the drug. We see that the Acquity BEH C18 columns work robustly up to 85°C using TEA/HFIP mobile phases, and several runs can be analyzed without any deterioration of the column. Alternatives include the Thermo Scientific DNA Pac™ RP column, among others.

For any OGN drugs containing unmodified bases, particularly cytosine, deamination is a risk. A deamidated metabolite is unlikely to be mass resolved from the analyte on a unit-resolution triple quadrupole MS/MS system, and chromatographic separation is often insufficient. In many situations, particularly intravenous or intrathecal dosed OGNs, this may have minimal impact on pharmacokinetic outcomes as the metabolite may not appear at problematic concentrations until very late timepoints. However, for other dose routes or study aims, or for measurements in tissue, the appearance of a deaminated metabolite could be problematic. Consideration of the potential impact of metabolites (or reference material manufacturing impurities) is essential during the method development process, and if they need to be distinguished from the drug, HRMS detection may be required. In such situations, the trade-off between sensitivity and selectivity must be carefully considered.

Conclusion

The preceding case studies are not an exhaustive list of all the challenges that may be encountered in LC-MS bioanalysis of an OGN drug. Rather, they present many useful considerations for bioanalysts who are initiating new methods using LLE/SPE extraction and LC-MS detection.

Given the challenges and greater variability we have seen with RP-SPE and MM-SPE, we now favor the phenol-chloroform followed by dichloromethane (LLE-LLE) extraction approach. This is the simplest and, in our experience, often the most robust workflow. However, where this is unsuccessful, for example, for lipid-modified OGNs or where very low LLOQs are required (<1–2 ng/ml per 10 μl of matrix being extracted for triple-quadrupole instruments, higher for HRMS instruments), more complex SPE-based workflows may be required which will require bespoke optimization, for which our case studies emphasize some of the important considerations. We have discussed the choice of HRMS versus triple-quadrupole MS/MS, and strategies to avoid nonspecific binding. Finally, we have discussed how potential metabolites or degradants should be considered at the start of method development.

By understanding the properties of OGN therapeutics, including potential biotransformation, the tools used for analysis and the unique requirements of each program and assay, it is possible to start with an informed position that can streamline development and validation of bioanalytical methods.

A flowchart of considerations, based on the four case studies we have presented, can be found in Table 2. We hope this will help to summarize our experiences and provide a useful resource to others.

Table 2.

Considerations for oligonucleotide LC-MS based on case study outcomes.

Workflow step	Considerations
Extraction
Extraction type	Phenol-chloroform followed by a second LLE cleanup (e.g., with dichloromethane), is a simple approach that can robustly allow 1–5 ng/ml LLOQs for many single or double stranded OGNs. This is now our preferred workflow. Phenol-chloroform then SPE can work well but sometimes gives issues caused by residual phenol. Standalone RP-SPE or MM-SPE may be more robust for lower LLOQs or some modified OGNs
Pre-treatment	For SPE workflows, use of chaotropes such as guanidine to release OGNs from plasma proteins is essential, but concentrations must be optimized to balance recovery with extract cleanliness (to prevent column blocking) and this balance may differ between MM-SPE and RP-SPE workflows. For MM-SPE the ionic strength and pH of the sample can impact on recovery. Lipid-modified OGNs are particularly strongly protein bound
Reagent integrity	Reagents containing HFIP, ammonium hydroxide or ion-pairing bases may be volatile or unstable. If issues are encountered, consider freshly prepared reagents and/or from freshly opened bottles. Solubility of HFIP in water is poor and good mixing is needed. If using THF (as recommended in some kits), purchase as BHT-stabilized and prepare reagents fresh
Internal standard and crosstalk	Stable-labeled version preferred (e.g., 34S labeled on all phosphorothioate groups). For analogs, screen different charge states, which may differently compensate for drift in analyte MS response through a run. All co-eluting ISTDs carry risks of crosstalk between the analyte and ISTD when using triple-quadrupole MS/MS detection. Consider including salt-free EDTA in mobile phases (∼100 μM) and reduce the electrospray source temperature. These measures can help to avoid adduct formation from trace Na+/K+ in the system, and reduce in-source losses or structural rearrangements, both of which can introduce more ion species that can cause crosstalk
Nonspecific binding	Prepare all matrix-free solutions in a solvent that overcomes non-specific binding (e.g., 50% aqueous methanol, acetonitrile or DMF, or an aqueous solution containing BSA). Use DNA LoBind tubes or plates. Non-specific binding may occur in processed LLE-LLE samples. This is not observed for SPE workflows as the guanidine generates dirtier extracts which are less prone to non-specific binding. For LLE-LLE methods, or any method using a low-protein matrix (e.g., cerebrospinal fluid) consider adding control human plasma as a reagent at the start of the extraction to block non-specific binding throughout the workflow
Chromatography
Ion pairing reagents	TEA/HFIP preferred, but DIPEA/HFIP or DIPA/HFIP can also be used for more retention but can show solubility issues. optimization of the concentration of these modifiers is needed for each analyte. Mobile phases will benefit from being as fresh as possible (typically <7 days old) and must be stored well-sealed to preserve their integrity
Column	Waters BEH C18 or Thermo Scientific DNAPac RP are good examples. Must be stable at high temperatures at pH 7–10. Maintain at a minimum of 80°C to sharpen peaks and avoid secondary structures forming on-column (inter- or intra-molecular) which can cause additional analyte-related peaks
Mass spectrometry
Sensitivity	Triple quadrupole MS/MS systems are preferred for best sensitivity
Selectivity	If there is reason to suspect truncated, deamidated, oxidized or any other kind of metabolite, or crosstalk from an internal standard, HRMS may be a preferred technique. The accuracy of the mass measurement helps exclude the crossed-ion interference from such species that can be observed using lower resolution triple quadrupole MS/MS instruments
Data interrogation
Additional peaks	Additional peaks appearing close to, or partially merged with the analyte peak in incurred samples only, particularly using triple-quadrupole MS/MS, may suggest a metabolite. Possible metabolites include truncated, oxidized or deamidated. This may only impact later timepoints, so might not always impact on study outcomes, so careful assessment is needed

Not all our case studies align with our recommendations in Table 2, because our understanding has evolved over time through these studies. Even after a decade of OGN LC-MS assay development, we are still learning and identifying novel solutions to support increasingly more complex molecules. This article is not intended as a review, but more a sharing of specific LC-MS experiences. We have thus not offered discussion on alternative methods including ligand binding assays, LC-FD, polymerase chain reaction or hybrid-capture LC-MS techniques, which may be appropriate in many cases.

Future perspective

We expect to see continued growth in the number of OGN therapeutics in development over the next decade. Current LC-MS technologies seem sufficient to deliver on most sensitivity demands. In situations where truncation, deamidation or other metabolic modifications are an issue, HRMS is an excellent tool to differentiate between similar species. Future improvements in HRMS instruments will probably soon allow them to achieve similar sensitivity to current triple-quadrupole platforms, allowing for <10 ng/ml LLOQs to be routinely achieved by HRMS, which our data shows to be sufficient in most cases. We expect that extraction workflows will be enhanced, including new ‘kits’ coming to market as an alternative to the well-used Clarity^® OTX kit (for example, the Waters OligoWorks^® kit, very recently launched). Hybridization capture upstream of LC-MS is already being extensively reported, is an evolving area, and will likely become more prevalent [^14-19]. The availability of cleaner extracts will enable wider adoption of microflow LC techniques which have been shown to offer major sensitivity gains [20]. We also note that we have seen more demand for analysis of OGNs in tissues. We are preparing for more such work and look forward to the availability of more automated workflows for tissue processing.

Author contributions

M Ewles and A Ledvina contributed equally to the manuscript; B Powers also provided significant insights, text and data.

Financial disclosure

The authors have no financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript. This includes employment, consultancies, honoraria, stock ownership or options, expert testimony, grants or patents received or pending, or royalties.

Competing interests disclosure

The authors have no competing interests or relevant affiliations with any organization or entity with the subject matter or materials discussed in the manuscript. This includes employment, consultancies, stock ownership or options and expert testimony.

Writing disclosure

No writing assistance was utilized in the production of this manuscript.

Ethical conduct of research

The authors state that they have obtained appropriate institutional review board approval or have followed the principles outlined in the Declaration of Helsinki for all human or animal experimental investigations. In addition, for investigations involving human subjects, informed consent has been obtained from the participants involved.