Liquid chromatography-mass spectrometry has been widely implemented as a powerful tool for providing in-depth characterization of nucleic acid therapeutic modalities, such as anti-sense oligonucleotides and small interfering RNAs (siRNAs).
Trang 1Contents lists available at ScienceDirect
journal homepage: www.elsevier.com/locate/chroma
spectrometry
Ming Huang, Xiaobin Xu∗, Haibo Qiu∗, Ning Li
Regeneron Pharmaceuticals Inc., Tarrytown, NY 10591, USA
a r t i c l e i n f o
Article history:
Received 7 January 2021
Revised 28 March 2021
Accepted 18 April 2021
Available online 27 April 2021
Keywords:
Oligonucleotides
siRNAs
Hydrophilic interaction liquid
chromatography
Tandem mass spectrometry
Phosphorothioate
Synthetic metabolites
a b s t r a c t
Liquid chromatography-mass spectrometry has been widely implemented as a powerful tool for provid- ing in-depth characterization of nucleic acid therapeutic modalities, such as anti-sense oligonucleotides and small interfering RNAs (siRNAs) In this study, we developed a generic hydrophilic interaction liq- uid chromatography (HILIC) hyphenated with tandem mass spectrometry method in the absence of ion- pairing reagents and demonstrated its capability as an attractive and robust alternative for oligonu- cleotide and siRNA analysis HILIC separation of mixtures of unmodified and fully phosphorothioate- modified DNA oligonucleotides and their synthetic 3’ exonuclease-digested metabolites were also as- sessed High-resolution mass spectrometric (HRMS) analysis was used to determine the deconvoluted masses of oligonucleotide and siRNA standards and their impurities To enable unbiased sequence char- acterization with tandem mass spectrometry (MS/MS), we also optimized higher-energy C-trap dissocia- tion (HCD) on improving the sequence coverage of DNA and RNA oligonucleotides Lastly, we evaluated on-column sensitivity for a phosphorothioate oligonucleotide by performing targeted analysis with either targeted selected ion monitoring (tSIM) or parallel reaction monitoring (PRM) Higher on-column sensitiv- ity of 13 ng, equivalent to 2.0 pmol, of a phosphorothioate oligonucleotide was achieved by tSIM analysis
as compared to PRM analysis
© 2021 The Authors Published by Elsevier B.V This is an open access article under the CC BY-NC-ND license ( http://creativecommons.org/licenses/by-nc-nd/4.0/)
1 Introduction
The continued optimism for synthetic nucleic acid drugs has fu-
eled an increasing interest in antisense oligonucleotide therapy and
RNA interference (RNAi) therapy Due to their mechanisms of gene
silencing and modulation of gene expression, oligonucleotides and
small interfering RNAs (siRNAs) represent a distinct class of ther-
apeutic molecules Recently, an increasing number of these drug
candidates have been approved or have progressed to later-stage
clinical trials [ 1, 2] However, there are different regulatory strate-
gies for the approval pathway of nucleic acid therapeutics These
drug candidates are categorized by the Food and Drug Administra-
tion (FDA) as small molecule drug candidates while they are cate-
gorized as biologics by European Medicines Agency (EMA) [3] Both
oligonucleotides and siRNAs are large, highly negatively charged
∗ Corresponding authors
E-mail addresses: xiaobin.xu@regeneron.com (X Xu), haibo.qiu@regeneron.com
(H Qiu)
molecules and show fundamentally different chemical and physi- ological properties relative to small molecule and antibody drugs Driven by drug development needs and regulatory requirements, numerous analytical strategies have been explored to provide in- depth characterization of various drug facets, such as synthetic im- purities, chemical modifications, and metabolites
In 1990 Andrew Alpert coined the term “hydrophilic interac- tion chromatography” (HILIC) to describe a liquid chromatographic technique for separating polar or ionized analytes The HILIC re- tention mechanisms involved the partition of solutes between the bulk mobile phase and an enriched water layer partially immobi- lized on the stationary phase surface, as well as secondary electro- static and hydrogen bonding interactions [4] Since then, HILIC cou- pled to mass spectrometry (HILIC −MS) has been broadly adopted
in the characterization of different classes of biomolecules, includ- ing proteins, glycopeptides, and small molecules, such as glycans and saccharides [ 5, 6] Because polar ribosyl residues and ioniz- able phosphate groups in nucleic acid structures are hydrophilic, early applications of HILIC include bioanalysis of nucleosides and
https://doi.org/10.1016/j.chroma.2021.462184
0021-9673/© 2021 The Authors Published by Elsevier B.V This is an open access article under the CC BY-NC-ND license ( http://creativecommons.org/licenses/by-nc-nd/4.0/ )
Trang 2nucleotides in different types of sample matrices [7–9] Recently,
HILIC has gained more attention as a promising alternative to
the widely used ion-pairing reversed-phase (IP-RP) LC methods
for oligonucleotide analysis in an effort to mitigate ion-pairing
(IP) reagent disadvantages, including reduced MS signal intensity
caused by ionization suppression and potential contamination in-
troduced to multipurpose instruments [10] In addition, HILIC ex-
hibits high MS compatibility because it uses volatile mobile phase
additives (e.g ammonium acetate), whereas another commonly
used chromatographic method, anion-exchange chromatography
(AEX), requires high concentrations of non-volatile salts which im-
pair the coupling to MS detection [10]
Easter et al first demonstrated the applicability of HILIC to
oligonucleotide analysis [ 11, 12] HILIC in the presence of ion-
pairing reagents (IP-HILIC) was also pursued to render selective
and orthogonal analysis of oligonucleotides [ 11, 13] Although these
studies provided proof-of-concept evidence that HILIC can be uti-
lized for the separation of oligonucleotides, these methods still suf-
fer from poor column stability, long equilibration time between
runs, and limited resolution for oligonucleotides of similar chain
lengths Till recently, oligonucleotide LC −MS analyses facilitated by
HILIC in the absence of ion-pairing reagents has presented promise
for characterizing oligonucleotide therapeutics [14–17] From the
development of different HILIC stationary phases, increased appli-
cations of HILIC in oligonucleotide characterization emerged To
date, different types of column chemistries have been successfully
applied to oligonucleotide analysis to promote selectivity, including
weak ion-exchange and neutral (silica-, diol-, amide- and cyano-
functionalized) stationary phases Easter et al first presented the
use of a diol-bonded HILIC column that couples crossed-linked
diol functional groups with the neutral, silica stationary phase [11]
Loube et al also explored HILIC separation of oligonucleotides us-
ing a diol-derivatized polymer-based column and showed moder-
ate separation efficiency for separating (n – x) shortmers of un-
modified or phosphorothioate (PS) oligonucleotides from the full-
length products (FLPs) [14] Demelenne et al’s more recent study
further demonstrated the excellent performance of a BEH amide
column functionalized with nonionic carbamoyl groups in render-
ing high peak capacities for unmodified and PS oligonucleotides
[16] Besides, with good orthogonality, HILIC has been coupled to
IP-RPLC or AEX in a two-dimensional (2D)-LC fashion to compre-
hensively characterize synthetic impurities of therapeutic oligonu-
cleotides [10] In this study, we developed a generic HILIC hy-
phenated with high-resolution mass spectrometric (HILIC −HRMS)
method to provide rapid, robust, and in-depth analytical character-
ization of DNA/RNA oligonucleotides This approach could be fur-
ther extended to the characterization of duplexed siRNAs Further-
more, HILIC separations of structurally similar synthetic metabo-
lites of unmodified and PS-modified DNA oligonucleotides were
assessed, and the coupling of HILIC to HRMS in profiling im-
purities and degradants was explored To facilitate unbiased se-
quence characterization, we also optimized higher-energy C-trap
dissociation (HCD) fragmentation conditions in sequencing DNA
and RNA oligonucleotides by MS/MS The study presented herein
demonstrated the analytical robustness of HILIC as an alterna-
tive chromatographic approach for in-depth characterization of
oligonucleotides and siRNAs, when completely liberated from any
IP reagents
2 Experimental
2.1 Materials and reagents
For mobile phase preparation, glacial acetic acid, ammonium
hydroxide (25% for LC −MS LiChropur TM), ammonium formate
(LiChropur TM LC −MS grade) and ammonium acetate (LiChropur TM
LC −MS grade) were purchased from Sigma-Aldrich (St Louis,
MO, USA) HPLC-grade acetonitrile was purchased from Honey- well (Charlotte, NC, USA) Deionized water was provided by a Milli-Q integral water purification system installed with a Milli- Pak Express 20 filter (Milli-Q system, Millipore, El Passo, TX, USA) All DNA and RNA oligonucleotides used in this study were pur- chased from Integrated DNA Technologies Inc (IDT; Coralville, IA, USA), and their sequences are listed in Supplementary Table 1 All synthetic oligonucleotides contained 3 and 5 -terminal hydroxyl groups The luciferase-targeted siRNA (siLuc) and its correspond- ing sense and anti-sense strands were purchased from IDT (Sup- plementary Table 1) MISSION® siRNA Universal Negative Control
#1 (SIC001, 13317 g/mol) and MISSION® siRNA Universal Negative Control #2 (SIC002, 13302 g/mol) were purchased from Millipore- Sigma (Burlington, MA, USA), and their sequences are proprietary All samples were dissolved in sterile nuclease-free OmniPur® Wa- ter (MilliporeSigma, Burlington, MA, USA) and stored as 200 μM stock solutions at –20 °C Aliquots were prepared to avoid repeated freeze-thaw cycles
HILIC mobile phases were prepared by pre-mixing 500 mM aqueous ammonium acetate (AA) or ammonium formate (AF), wa- ter and acetonitrile (ACN) to yield appropriate salt concentrations Mobile phase A (MPA) was composed of 70% ACN buffered with
AA or AF, and mobile phase B (MPB) was composed of 30% ACN buffered with AA or AF For HILIC analysis, sample diluent with identical buffer composition with MPA (weak solvent) was used for all analyzed oligonucleotides, and the on-column injection amount was 20 pmol (2 μL) if not otherwise specified The (n – x) oligonu- cleotide mixture was prepared by reconstitution of lyophilized equimolar mixtures of each oligonucleotide with sample diluent at
a nominal concentration of 10 μM
2.2 LC–MS/MS analysis
The LC–MS/MS platform was composed of a Waters Acquity I-class ultra-performance liquid chromatography (UPLC) system (Waters, Milford, MA, USA) interfaced to a Q Exactive Hybrid quadrupole Orbitrap mass spectrometer (Thermo Fisher Scientific, Waltham, MA, USA) by a heated electrospray ionization (H-ESI) source Samples were analyzed on a BEH Amide UHPLC column (Waters, Milford, MA, USA; 2.1 mm × 150 mm, 1.7 μm parti- cle size, 130 ˚A pore size) For regular analyses, DNA and RNA oligonucleotides were eluted with a linear gradient that was in- creased from 20% to 70% MPB over 10 min at a flow rate of 0.25 mL/min with column temperature set as 30 °C siRNAs were eluted with a 10-min linear gradient that was increased from 60% to 70% MPB For oligonucleotide impurity characterization, analytes were eluted with a 15-min linear gradient that was increased from 20%
to 65% MPB The gradient was then ramped to 80% MPB over 1 min and held for 2 min, before dropping to 20% MPA and then getting re-equilibrated at 20% MPA for 10 min before the next run The eluents were monitored at a wavelength of 260 nm us- ing either a photodiode array (PDA) or tunable ultraviolet (TUV) detector and then electro-sprayed into MS The following param- eters were used for MS analysis: negative polarity (static spray, 3.0 kV), ion funnel radiofrequency (RF) level at 60%, sheath gas (40 a.u.), auxiliary gas (15 a.u.), sweep gas (0 a.u.), ion transfer tube temperature of 325 °C, vaporizer temperature of 350 °C, mass range 40 0–20 0 0 m/z , mass resolution 70,0 0 0 full-width half max- imum (FWHM), automatic gain control (AGC) 5e5, injection time (IT) 100 ms, intensity threshold 1e3, charge state selection 1–4, and each full scan spectrum consisted of accumulation of 1 mi- croscan Data dependent MS/MS (top 7) were acquired by em- ploying higher-energy C-trap dissociation (HCD) with a 2.0 m/z
quadrupole isolation window, AGC of 1e5, IT 200 ms, and each MS/MS spectrum consisted of accumulation of 1 microscan For
Trang 3HCD fragmentation, normalized collision energy (NCE) was set at
15% and first mass m/z was set as 70 at a mass resolution of 30,0 0 0
FWHM For on-column sensitivity assessment, targeted selected
ion monitoring (tSIM) and parallel reaction monitoring (PRM) ac-
quisition were performed at a mass resolution of 70,0 0 0 FWHM
and a mass isolation window of 2.0 m/z , AGC of 1e6, and IT of
100 ms In PRM mode, the data were acquired according to a pre-
determined inclusion list containing the accurate mass and nor-
malized collisional energies (NCE) of analytes Additionally, other
MS parameters were set as the AGC of 2e5, IT of 100 ms and iso-
lation window of 2.0 m/z
2.3 Data processing
High-resolution mass spectra were deconvoluted by the PMI-
Intact Mass TM software (Protein Metrics, San Carlos, CA, USA) to
obtain both deconvoluted monoisotopic mass and average mass
Automated MS/MS fragment annotation was achieved by using Bio-
Pharma Finder TM4.0 (Thermo Fisher Scientific, Waltham, MA, USA)
as described in the text Other DDA or tSIM raw files were an-
alyzed by Xcalibur Qual Browser software version 2.2 (Thermo
Fisher Scientific, Waltham, MA, USA) To generate response curves
for sensitivity evaluation, extracted ion chromatograms (EICs) were
generated for each standard injection from the most abundant
charge state with an m/z extraction window of 10 ppm The peak
areas of the EICs were plotted versus concentration to generate lin-
ear regression results (slope and R 2) The limit of detection (LOD)
of each method was determined based on a target signal-to-noise
ratio (S/N) above 3, as calculated in the Qual Browser software For
impurity profiling, the 5’ and 3’ shortmer sequences were calcu-
lated by Mongo Oligo Mass Calculator v2.08 (University at Albany,
SUNY) PRM data was analyzed by Skyline version 4.2.0 (University
of Washington)
3 Results and discussion
3.1 BEH amide column evaluation in oligonucleotide analysis
Many effort have been made to understand the HILIC reten-
tion mechanisms of oligonucleotides by exploring the use of dif-
ferent HILIC stationary phases Among them, the superior separa-
tion efficiency obtained using BEH amide column independently
reported by several research groups encouraged us to further eval-
uate the performance of this column for oligonucleotide analy-
sis [ 10, 15, 16] Mechanistically, the polar amide functional groups
may be more effective at interacting with the aqueous portion of
the mobile phase and forming the stagnant water layer required
for HILIC The carbamoyl groups within the amide phase can par-
ticipate in hydrogen-bonding as hydrogen-bond donors and fur-
ther interact with hydroxyl groups in the analytes [18] Addition-
ally, with smaller diameters of packing materials, improved chro-
matographic performance was achieved due to more effective mass
transfer (Van Deemter equation) As a result, we selected the BEH
amide column to develop a generic HILIC method suitable for rou-
tine oligonucleotide analysis To validate the method robustness,
we analyzed both single-stranded DNA and RNA oligonucleotides
as well as double-stranded siRNAs in five replicates with a 10-min
gradient and a column re-equilibration time of 10 min between
each injection High repeatability of retention times and peak areas
were obtained with < 10% coefficient of variance (CV) (Table S2) It
is also worth noting that our optimized HILIC method with 20-min
run time is 3 times faster than most HILIC applications for oligonu-
cleotides that employ a rather long gradient and re-equilibration
time (60-min run time) [ 14, 16]
Fig 1 HILIC retention of siRNAs, RNA oligonucleotides, DNA oligonucleotides and
phosphorothioate DNA oligonucleotides (grouped by color shaded areas) on a BEH amide column in mobile phases with (A) 25 mM ammonium formate or ammonium acetate; (B) ammonium acetate (pH 6.8) of various concentrations (2.5 mM, 7.5 mM,
15 mM, and 25 mM); (C) 15 mM of ammonium acetate of different pH (pH 5.5, pH 6.8, pH 9.0)
3.2 Impacts of mobile phase additives (salt types, concentrations and pH) on oligonucleotide/siRNA analysis
Next, we compared the influence of commonly used HILIC mo- bile phase additives on HILIC separations of oligonucleotides We first evaluated two commonly used additives ammonium formate (AF) and ammonium acetate (AA) at concentrations of 25 mM, and
we observed comparable peak capacities between the two groups
As illustrated by Fig 1A, regardless of salt types and concentra- tions, a general HILIC retention order was observed: duplexed RNA
> RNA oligonucleotide > DNA oligonucleotide > PS-modified DNA oligonucleotide Besides, the elution order of DNA oligonucleotides was dependent on their chain length, which could be attributed
to the net negative charges carried This is consistent with previ- ous studies that used poly-deoxy(thymidylic) acids (dT) ladders for benchmarking LC separations [14–16] Meanwhile, DNA oligonu- cleotides carrying backbone PS modifications showed diminished HILIC retention due to the higher hydrophobicity comparing to phosphate backbone, with the displacement of oxygen with the
Trang 4less electronegative sulfur [13] Moreover, AF-containing mobile
phases displayed slightly greater HILIC elution strength and dimin-
ished retention on the BEH amide column, as reflected by a de-
crease of the retention factor or k values ( Fig 1A) Such a phe-
nomenon was, however, opposed to the previous findings made on
a TSK-gel Amide-80 column that AA poses greater elution strength
compared to AF [ 11, 17] Weaker elution strength of AF in compar-
ison to AA was also indicated on a modified diol column, as the
analyte retention is more attributed to the pH rather than ionic
strength of the mobile phase [14] The differential elution strength
between AF- and AA-containing mobile phases on the BEH amide
column may be attributed to stronger ion exchange interactions of
the formate ions
Adjustments of ionic strengths in HILIC mobile phases can al-
ter method selectivity, column retention, and separation efficiency
[19] Lobue et al previously investigated buffer salt concentrations
ranging from 2.5 mM to 15 mM and determined 15 mM AA (pH
5.5) as the optimal concentration for good chromatographic peak
shapes and MS response on a polymer-based diol-bonded column
[14] The authors found that lower AA concentrations generally
gave rise to broadened LC peaks and worse peak shape Similar
mobile phase conditions, with 15 mM AA (pH 5.5) being chosen
as an optimal mobile phase additive, were reported in an indepen-
dent study from Demelenne et al on a BEH amide column [16]
Moreover, in another two studies that used BEH amide column,
MacNeill et al employed 10 mM AF (pH 9.0) for quantification of
a PS oligonucleotide, while Goyon et al chose 25 mM AA with no
pH adjustment [ 10, 15] Taken together, we reckoned that a system-
atic screening of salt concentrations and pH could benefit HILIC
method development for oligonucleotide using the BEH amide col-
umn In the present investigation, four different salt concentra-
tions of AA were tested: 2.5 mM, 7.5 mM, 15 mM and 25 mM
As shown in Fig.1B, the observed k values or HILIC retention in-
creased with elevated salt concentrations, consistent with previous
findings made on different HILIC stationary phases [ 14, 17, 20] Such
a phenomenon may suggest that solvation instead of ion exchange
remains the dominating retention mechanisms [19] For BEH amide
column, the underlying mechanisms accountable for this observa-
tion are likely to be the expansion of the aqueous layer adsorbed
on the stationary phase surface and as such, more solvated salt
ions accumulate in this layer and consequently contribute to a
thickening of the water layer and increased retention via hydrogen
bonding [ 17, 19, 20]
Packed with hybrid silica particles, the BEH amide column ex-
hibits greater tolerance to high pH ( >8) than pure silica parti-
cles, allowing more flexible method development By testing mo-
bile phases composed of 15 mM AA at different pH (pH 5.5, pH
6.8, and pH 9.0), we observed that elevated mobile phase pH with
addition of ammonium hydroxide (pH 9.0) led to lower HILIC re-
tention of most oligonucleotides, whereas addition of acetic acid
(pH 5.5) generally gave rise to higher HILIC retention ( Fig.1C) Our
findings were in line with a recent study by Kilanowska et al, in
which a minor increase of k was observed with decrease of pH val-
ues independent of salt types [17] However, opposing results indi-
cating increasing of HILIC retention of a PS oligonucleotide with in-
crease of pH were also seen, which was justified by higher charge
carried by oligonucleotides at higher pH [10] In theory, nucleotide
subunits contain both basic nitrogen atoms within the nucleobase
and acidic phosphate groups and, as such, mobile phase pH de-
termines the ionization and charge state This can in turn modu-
late the polarity or hydrophilicity of the analyte and likewise the
HILIC retention [19] At pH < 8, each nucleotide contributes to one
negative charge, so each additional nucleotide increases the over-
all charge on the molecule and consequently increases HILIC re-
tention As the eluent pH further increases to > 8, the tautomeric
oxygen on each G and T (U for RNA) becomes an oxyanion, thereby
further increasing the charge, in proportion to the T(U) + G con- tent [21] Interestingly, we also observed slightly more noticeable peak tailing at pH 5.5 while a more symmetrical peak profile at
pH 9.0, which was also reported by Goyon et al [10] Alterations of mobile phase pH, however, did not have obvious effects on modu- lating LC peak resolution
3.3 Assessing effects of column temperature on HILIC analysis of oligonucleotides and siRNAs
In HILIC mode, column temperatures can be optimized to en- hance selectivity, lower solvent viscosity, and increase mass trans- fer rates Most HILIC applications in oligonucleotides use rela- tively low temperature (40 °C) for separation [ 14, 16] In a study by Goyon et al, it was shown that higher column temperature of a BEH amide column led to increased peak capacity but lower sig- nal response when coupled to UV detection [10] In the present investigation, a panel of different column temperatures ranging from 30 °C to 80 °C (with 10 °C intervals) were tested As sug- gested by the HILIC–UV traces in Figure S1, increased column retention and peak capacity yet lowered UV response were ob- served with elevated temperature, which is in keeping with the trend observed by Goyon et al The mechanisms accounting for higher peak resolution at elevated column temperature could be attributed to minimized non-specific interactions that may arise from internal hydrogen bonding [22] However, analyte adsorption may occur at the surface of HILIC stationary phase through elec- trostatic interactions or hydrogen bonding, and increasing temper- ature may result in greater exposure of polar groups present in analytes [ 10, 18] Besides, higher column temperature could result
in reduction of the adsorbed water-rich layer on the stationary phase, thereby modulating the partition of the analytes between the bulk mobile phase and adsorbed water-rich layer and conse- quently retention behavior [23] Consequently, 30 °C was chosen as the column temperature for the HILIC separation of single-stranded oligonucleotides
In its native conformation, the sense (S) and anti-sense (AS) strands of duplex siRNA are primarily associated through non- covalent interactions such as hydrogen bond interaction and base stacking Increased column temperature is therefore expected to impact the peak shapes of duplexed siRNA compounds by disrupt- ing the double-stranded structures For further investigation, col- umn temperatures ranging from 30 °C to 80 °C were applied and the duplex, S and AS strands of an siRNA that targets luciferase (siLuc) were subject to HILIC–UV analysis As shown in Figure S2, we ob- served complete or partial melting of the duplex strands with el- evated column heating at 70 °C The results demonstrated that el- evating column temperature to as low as 70 °C led to peak broad- ening due to duplex melting, yet the conformation of S/AS strands remained largely unchanged (Figure S2) The denaturation or melt- ing of the duplexed structures was further confirmed by MS anal- ysis As such, it is generally advisable to employ a column tem- perature below the melting point (T m) of the double-stranded nu- cleic acids, under which the duplex structures remain largely in- tact That said, on-column T m depression can occur in the presence
of an organic solvent (i.e., ACN) in the mobile phase, and optimal column temperature should, therefore, be carefully evaluated [24] Furthermore, it is generally recognized that siRNAs appear to be more hydrophilic compared to their corresponding S/AS strands, as their hydrophobic bases are shielded from solvents in duplex con- formation [25] Indeed, we observed substantially elevated HILIC retention of the duplexed structures in contrast to the correspond- ing S and AS strands (Figure S2), inferring that our current HILIC conditions could preserve non-covalent duplexes of nucleic acids under a lower column temperature (30 °C)
Trang 5Fig 2 Full-scan and deconvoluted mass spectra for six DNA oligonucleotides (CMV-F, M13-R, SP6, TRC-F, T3, and T7) analyzed by HILIC −MS
3.4 Impacts of mobile phase additives on full-scan MS analysis of
oligonucleotides and siRNAs
The advantages of coupling HILIC to mass spectrometric (MS)
analysis include more efficient sample desolvation and ionization
facilitated by the high organic content in the eluents, as such con-
tributing to superior electrospray ionization (ESI)-MS sensitivity
Because nucleic acid drugs are a class of molecules that exhibit
dominant negative charges, their charge state distribution could
be impacted by the generic p K aof the molecules, conformational
changes, and formation of cationic adducts We first compared the
MS response obtained with the use of mobile phases containing AF
(25 mM) and different concentrations of AA: 2.5 mM, 7.5 mM, 15
mM and 25 mM The highest MS response for most DNA oligonu-
cleotides was achieved with a salt concentration of 15 mM AA,
except for a longer 25-mer TRC-F (Figure S3) The lowest MS re-
sponse was observed for 25 mM AF (Figure S3) In addition, with
decrease of salt concentration, we observed a systematic shift of
charge state distributions of oligonucleotide polyanions to lower
m/z values, which align with the results shown by Guo et al via
direct infusion experiments [26] Together, to balance good chro-
matographic performance and reasonable MS signal intensities, 15
mM AA was chosen as the final mobile phase salt concentration Moreover, oligonucleotide precursor ions with longer chain lengths were populated in higher charge states (Figure S4A), and distinct differences in charge envelope profiles were noticed for three mo- bile phase pH tested: AA (pH 5.5), AA (pH 9.0) and AF (pH 5) (Figure S4B) The impacts of mobile phase composition on charge states were not solely dependent on buffer pH, but also can be re- lated to salt types Collectively, to ensure that less water content was needed for analyte elution and thereby higher ESI desolvation efficiency, we tend to use 15 mM AA (pH 9.0) as mobile phase ad- ditives optimized for the BEH amide column, which provided an equivalent chromatographic performance to the one obtained at
pH 5.5
Furthermore, we also showed that the acquired full-scan mass spectra of DNA/RNA oligonucleotides and siRNA duplexes acquired could be successfully deconvoluted using the PMI-Intact Mass TM
software (Protein Metrics, San Carlos, CA, USA) The accurate mass for each of the DNA oligonucleotides ( Fig.2) could be readily de- termined, with additional peaks of sodium (Na), potassium (K) and acetate (Ac) adducts observed Moreover, full-scan mass spectra
Trang 6Fig 3 Full-scan and deconvoluted mass spectra for intact siRNA duplexes (NT siRNA#1, NT siRNA#2 and siLuc) analyzed by HILIC −MS
Fig 4 (A) HILIC–UV chromatograms showing the separation of the mixture of synthetic 3’ (n – x) truncated sequences of a DNA oligonucleotide TRC-F (B) HILIC–UV
chromatograms showing the separation of the 4-oligo (upper panel) and 9-oligo (lower panel) mixture of synthetic 3’ (n – x) truncated sequences of a PS oligonucleotide TRC-FPS
of siRNA S/AS strands that were individually analyzed or partially
denatured during ESI process could also be deconvoluted (Figure
S5) While the analysis of duplexed nucleic acids by HILIC–MS ap-
pears to be absent from the current body of literature, it was
demonstrated by others that effective IP-RPLC–MS analysis of in-
tact duplex RNA could be achieved by using suitable mobile phases
under non-denaturing conditions [27–29] Nevertheless, some IP
reagents used in IP-RPLC applications can result in duplex dissoci-
ation in both chromatographic and MS detection, presumably due
to the disruption of hydrogen bonds with higher ionic strength
[ 29, 30] We showed in the previous section that the siRNA du-
plexes could be retained under HILIC conditions More importantly,
full-scan and deconvoluted mass spectra for native siRNA duplexes
corroborated the preservation of gas-phase duplex conformation by
MS analysis ( Fig 3) Taken together, the results have evidenced
that HILIC–MS could provide an attractive alternative to analyz-
ing oligonucleotides and siRNAs, and this is the first demonstra-
tion of HILIC–MS application in providing native analyses of siRNA duplexes
Besides, for both siRNA duplexes and single-stranded oligonu- cleotides, the metal adduction peaks observed in HILIC −MS were
of relatively low abundance ( Figs 2, 3 and S5) In line with our results, it was previously reported that the addition of ammonium acetate during ESI process can aid in the removal of alkali metal adducts with volatile ammonium ions (NH 4+) bound to the neg- atively charged backbone of oligonucleotides [31] Together, these observations support another advantage of HILIC in mitigating the adduct formation issues that are commonly encountered in IP- RPLC applications [32]
3.5 HILIC separation of synthetic 3’ (n − x) truncated sequences of oligonucleotides
In-depth characterization of nucleic acid drug products neces- sitates unambiguous profiling of structural variants of oligonu-
Trang 7Fig 5 (A) Full view and 60 × zoom-in view illustrating the HILIC–UV profiles of the DNA oligonucleotide CMV-F The dashed square highlights the peak region that was
further resolved by MS analysis Extracted ion chromatograms depicting a panel of (B) 3’ and (C) 5’ (n – x) shortmer impurities present in CMV-F using the most abundant [M – 4H] 4– precursor (D) Summary of the identified impurity sequences of CMV-F and their relative abundance present in the sample
cleotides or siRNA arising from synthetic impurities, degradants,
and metabolites generated in vitro or in vivo [33] Oligonu-
cleotide and siRNA therapeutics are degraded mainly by cleav-
age of the phosphodiester linkages by nucleases In vitro experi-
ments showed that cleavage at the 3 -terminus resulting from 3’-
exonucleolytic activities is the primary form of metabolites, fol-
lowed by 5 -exonuclease and endonuclease cleavage products [34]
Chemical modifications made to the ribose or phosphate backbone
are, hence, routinely incorporated into oligonucleotide and siRNA
molecule design in an effort to mitigate exo- and endo-nuclease
activity in order to ensure sustained drug efficacy [35] In addition
to metabolites, process-related impurities can arise from chemical
synthesis and often include families of shortmer (n – x) and long-
mer (n + x) sequences For example, an n – 1 family of impurities
would consist of multiple n – base (B), where B can be any base in
the sequence [36]
Structurally similar synthetic (n – x) sequences of oligonu-
cleotides have been largely employed to examine HILIC separations
of shortmer impurity or metabolite mimics [ 14, 16] Therefore, we
first evaluated separations of an equimolar mixture of up to (n
– 8) 3’ truncated synthetic sequences of a 25-mer DNA oligonu-
cleotide TRC-F By extending to a 20-minute linear HILIC gradient,
we successfully achieved good separation of all (n – x) shortmer
sequences and the FLP ( Fig.4A) However, when it comes to a fully
PS-modified version of TRC-F, i.e TRC-FPS, it was particularly dif-
ficult to resolve the 3’ (n – 1) sequence and the FLP sequences
Specifically, the 3’ (n – 3) and 3’ (n – 2) sequences could be well
resolved from the FLP, with the other 3’ (n – x) sequences eluted
closer together ( Fig.4B) The challenge of separating PS oligonu-
cleotides arises from the compromised LC peak resolution caused
by the inherent stereochemical configuration of the PS linkages
(“Sp ” or “Rp ”) One single PS linkage can introduce a chiral cen-
ter at phosphorus in addition to the D-ribose chiral centers, giv-
ing rise to 2 N diastereoisomers (number of PS bonds = N ) and
chromatographic peak splitting or broadening In IP-RPLC, nega- tive charges on the oligonucleotide phosphate backbone are neu- tralized by positively charged alkylammonium ions in the mobile phase Hydrophobic interactions between the oligonucleotide bases and the reversed-phase column play a major role in the sepa- ration, and the different hydrophobicity of the individual bases contributes to differences in retention In comparison, HILIC chro- matography depends on hydrophilicity, which is not highly variable between oligonucleotide units largely due to the presence of phos- phate groups Therefore, HILIC separation of oligonucleotides is not
as robust as it is with IP-RPLC approaches Further modifications with mobile phase modifiers or derivatization are needed to im- prove selectivity
3.6 Oligonucleotide impurity analysis by HILIC–MS
Although extensive purification process removes most of impu- rities from oligonucleotide therapeutics, low levels of remaining impurities or degraded products could significantly impact both drug safety and efficacy The commonly identified impurities of oligonucleotide therapeutics usually include chain shortened (n – x) products [37] Many hybridization-based oligonucleotide assays
do not accurately distinguish large impurities or degradants from the full-length oligonucleotide of interest, despite their ultra-high sensitivity in quantifying impurities or metabolites [38] In recent years, MS has gradually become the method of choice in that it can provide unambiguous and comprehensive impurity identifica- tion/quantitation and metabolite profiling for both oligonucleotide and siRNA modalities [39]
As illustrated by the HILIC–UV traces in Figure S6, various small impurity peaks could be readily separated with an extended linear gradient (20% to 65% MPB in 15 min) These impurities could re- sult from degradation through freeze-thaw cycles and the presence
of trace nuclease in the samples Interestingly, for the PS oligonu-
Trang 8Fig 6 (A) Annotated HCD-MS/MS fragmentation for the [M – 5H] 5– precursor ( m/z 1309.2167) of a DNA oligonucleotide CMV-F (NCE 15%) (B) MS/MS fragment coverage map for the results shown in (A), generated from Thermo Biopharma Finder 4.0
cleotide T7-FPS, there seemed to be a lesser amount of impuri-
ties compared to the amount of impurities for unmodified DNA
or RNA oligonucleotides, possibly due to the increased resistance
of PS bonds against nucleolytic activities Taking advantage of the
downstream HRMS analysis, we attempted to further characterize
the identities of some of these impurity peaks In the elution win-
dow highlighted by the dashed box in the HILIC–UV profile of the
20-mer DNA oligonucleotide CMV-F ( Fig.5A), we performed tar-
geted MS extraction using the m/z value of the most abundant pre-
cursor ions ([M – 4H] 4–) of each (n – x) sequence from either 3’ or
5’ terminus to calculate signal intensities Based on the peak in-
tensities extracted for the corresponding 3’ and 5’ (n – x) short-
mer sequences ( Fig 5B, C), our results evidenced that the gen-
eration of 3’ truncated degradants are more prevalent than those
generated from the 5’ end for CMV-F ( Fig.5D) For other impurity
peaks with earlier elution times, we speculated that there could be
both exo- and endo-nuclease activities present that led to oligonu-
cleotide impurities with shorter chain length Overall, we demon-
strated the feasibility of coupling the HILIC method to UV/MS de-
tection for impurity profiling of oligonucleotides
3.7 Oligonucleotide sequence characterization by HILIC–MS/MS
Sequencing elucidation of DNA or RNA oligonucleotides com- posed of up to 70 residues by MS/MS fragmentation of multiply charged oligonucleotide precursor ions in negative ESI mode has been intensively studied in the past few decades McLuckey et al indicated that the major fragmentation pathways by collisional- induced dissociation (CID) of oligonucleotides can proceed with multiple pathways [40] Among the backbone cleavage products, a–
B and w-type ions are the most dominant species for DNA oligonu- cleotides, while y- and c-type ions remain more favored for RNA oligonucleotides [41] The Q-Exactive mass spectrometer features beam-type CID, sometimes referred to as higher-energy C-trap dis- sociation (HCD), which relies on an octupole collision cell that op- erates at higher collisional energies than CID [42] Furthermore, it
is known that precursor charge states can affect the efficiency of fragmentation, as described in previous studies on peptides and other biomolecules [43] As such, we attempted to evaluate how the commonly observed precursor charge states in the HILIC–MS mode and the normalized collision energy (NCE) applied for HCD-
Trang 9Fig 7 Assessment of on-column sensitivity and linearity range of the HILIC–MS method by (A) tSIM scan and (B) PRM scan
MS/MS fragmentation could impact signature fragment ions and
sequence coverage for both DNA and RNA oligonucleotides
The complexity of a typical oligonucleotide mapping experi-
ment resulting from a great number of structurally diverse frag-
mentation ions presents a significant throughput hurdle Fortu-
nately, this challenge has been mitigated via recent developments
and advancements in software tools that facilitate automated and
unbiased data analysis In this study, a newly developed data anal-
ysis module incorporated into Biopharma Finder TM 4.0 (Thermo
Fisher Scientific) was utilized to achieve automated data process-
ing and sequence annotation for HCD-MS/MS data acquired from
oligonucleotides Benefited from this tool, the overall % sequence
coverage was systematically examined for oligonucleotides with
various base composition (deoxyribonucleic acid, ribonucleic acid),
chain length (17-mer to 25-mer), backbone modification (unmod-
ified, partially or fully PS-modified), and sequence composition
The most abundant precursors ([M – 4H] 4–, [M – 5H] 5– and [M
– 6H] 6–) observed in the HILIC–MS mode were chosen for anal-
ysis, and a panel of % normalized collisional energy (%NCE) val-
ues ranging from 13% to 23% were examined In general, precursor
ions of higher charge states require higher NCE to achieve good
sequence coverage, and an opposite trend was observed for lower-
charge-state precursors (Figure S7) For most DNA or RNA oligonu-
cleotides selected in this study, the optimal NCE was determined
to be 15% for their most abundant charge states For higher charge
states such as [M – 6H] 6–, only partial fragmentation of parent ions
was observed with low %NCE applied, resulting in low sequence
coverage As depicted in Fig 6A, Biopharma Finder TM 4.0 enabled
rapid and accurate fragment annotations (predominantly a–B and
w-type ions) for the DNA oligonucleotide CMV-F By employing 15%
NCE for the [M – 5H] 5– precursor ( m/z 1309.2167), 90% sequence
coverage was achieved, with the intensities of each fragment high-
lighted in the sequence coverage map ( Fig.6B)
3.8 Quantitative analysis of oligonucleotides
Quantitative analysis of therapeutic oligonucleotides often rep-
resents a fundamental part in determining their pharmacokinetic
and pharmacodynamic (PK/PD) properties during drug develop-
ment LC–MS based bioanalytical method can achieve the sensitiv-
ity of 1 ng/mL when coupled to upstream sample preparation tech-
niques such as solid-phase extraction (SPE) [44] Thus, we reckon
that it is essential to evaluate the sensitivity of our HILIC–MS
method to better suit bioanalysis requirements There have been
controversies on the sensitivity comparison between IP-RPLC–MS
and HILIC–MS in the field Loube et al showed enhanced MS signal response by HILIC–MS for oligonucleotides of shorter chain-lengths and lower hydrophobicity, as compared to IP-RPLC–MS [14] Nev- ertheless, Kilanowska et al also pointed out that ACN in the mo- bile phase may actually cause ionization suppression of oligonu- cleotides due to its aprotic properties [17] In the work by Easter
et al, the LOD values were observed at the picomolar level with the coupling of inductively coupled plasma (ICP)–MS to monitor the signature 31P 16O + ion [11] In another study, analyses of oligonu- cleotides were performed by HILIC–MS analysis with an LOD value
of 2.5 pmol loaded on-column for a 20-mer oligonucleotide [12] A similar LOD value of 2.0 pmol on-column was also reported for a 22-mer oligonucleotide using 2D-LC–MS analysis [45]
We first determined the optimal HCD energy for chemically modified oligonucleotides including the most prevalent therapeutic oligonucleotide design that involve full PS backbone modifications
on T7 (T7-FPS) and TRC-F (TRC-FPS) Notably, CID- or HCD-MS/MS fragmentation of oligonucleotides with PS modifications gave rise
to a diagnostic fragment ion PSO 2–( m/z 94.93), which was broadly
used for targeted MS analysis such as multiple-reaction monitoring (MRM) [ 17, 20] Our PRM results shown in Figure S8 indicated that higher %NCE may favor generation of the signature PSO 2–fragment ions for quantification purpose (sensitivity), whereas lower %NCE could benefit more comprehensive sequence annotation (speci- ficity) Next, we sought to examine the on-column sensitivity for T7-FPS by performing tSIM or PRM analysis As displayed in Fig.7, tSIM analysis displayed higher S/N ratios or sensitivity in com- parison to ratios or sensitivity of PRM analysis in all injections analyzed Even with the lowest injection of 13 ng (equivalent to 2.0 pmol), a clear S/N profile was achieved for the top two most abundant charge states Furthermore, we believe that PRM analysis could warrant higher selectivity for the analysis of structurally sim- ilar variants in complex mixtures by incorporating signature frag- ment ions
4 Conclusions
Until now, IP-RPLC remains the most frequently used technique
in providing analytical characterization of oligonucleotides There
is, however, a drastically increasing trend of using HILIC appli- cations to render complementary and more in-depth analysis of oligonucleotides and other nucleic acid therapeutics In this study,
we established a universal HILIC–MS/MS method that could pro- vide a comprehensive solution for the analysis of oligonucleotides and siRNAs, including separation, mass determination, sequence
Trang 10characterization, impurity profiling, as well as the potential in
quantitative analysis of oligonucleotides and siRNAs to support
drug development
We have evaluated the impacts of mobile phases additives
on both chromatographic separation and MS response of the
HILIC–MS method Accurate intact mass measurement for oligonu-
cleotides was successfully achieved by HRMS analysis More im-
portantly, for the first time, we presented the utility of this HILIC–
MS method that preserves the gas-phase duplexed conformation
in rendering native analysis of siRNA duplexes Furthermore, the
HILIC–MS described herein could facilitate straightforward impu-
rity analysis of oligonucleotides Another highlight of the estab-
lished method was the hyphenation of HILIC with MS/MS for un-
biased sequence annotation, which could be further applied to
oligonucleotides with backbone or ribose chemical modifications
for structural characterization We showed that employing low
NCE (15%) on an orbitrap instrument could facilitate comprehen-
sive HCD-MS/MS sequence coverages for 17- to 25-mer oligonu-
cleotides, and that automated MS/MS annotation could be achieved
with improved ease Additionally, we have demonstrated that the
use of HILIC–MS based on tSIM mode, as compared to PRM mode,
improved the overall on-column sensitivity However, for com-
plex mixtures of structurally similar oligonucleotide impurities and
metabolites, selectivity could be further enhanced with PRM anal-
ysis by selecting structure-related signature fragment ions
In-depth characterization and identification of synthetic impu-
rities present in drug products and metabolites generated in vitro
or in vivo remain critical tasks in risk mitigation and clinical stud-
ies Extensive work that relies on LC–MS approach has been done
previously to identify trace impurities [ 46, 47], in vivo metabo-
lites [48] and in vitro metabolites [49] for better risk mitigation
of manufacturing and clinical studies Importantly, the established
HILIC–MS/MS assay can be readily employed in combination with
SPE to fully support bioanalysis of clinical samples We envisioned
that the developed HILIC–MS/MS approach could efficiently sup-
port synthetic oligonucleotide and siRNA chemistry with a wide ar-
ray of modifications and provide rapid and flexible in-depth char-
acterization of oligonucleotide or siRNA drug products in the ab-
sence of IP reagents
Authors’ contribution
Ming Huang: Methodology, Data curation, Investigation, Writing
- original draft, Writing review & editing Xiaobin Xu: Concep-
tualization, Project administration, Methodology, Writing review
& editing Haibo Qiu: Conceptualization, Supervision, Writing re-
view & editing Ning Li: Supervision, Writing review & editing
Declaration of Competing Interest
All authors were employees of Regeneron Pharmaceuticals, Inc
while engaged in the study and may hold stock and/or stock op-
tions in the company All authors have patent applications that
were based on the research in the paper
Acknowledgement
This study was sponsored by Regeneron Pharmaceuticals, Inc
The authors would like to thank Ashley Roberts from Scien-
tific Writing Group for assistance in drafting and polishing this
manuscript
Supplementary materials
Supplementary material associated with this article can be
found, in the online version, at doi: 10.1016/j.chroma.2021.462184
References
[1] X Shen , D.R Corey , Chemistry, mechanism and clinical status of anti- sense oligonucleotides and duplex RNAs, Nucl Acids Res 46 (2018) 1584–
1600 [2] C.A Stein , D Castanotto , FDA-approved oligonucleotide therapies in 2017, Mol Ther 25 (2017) 1069–1075
[3] C Iglesias-Lopez , A Agusti , M Obach , A Vallano , Regulatory framework for advanced therapy medicinal products in Europe and United States, Front Phar- macol 10 (2019) 921
[4] A.J Alpert , Hydrophilic-interaction chromatography for the separation of pep- tides, nucleic acids and other polar compounds, J Chromatogr 499 (1990) 177–196
[5] Q Zhang , F.Q Yang , L Ge , Y.J Hu , Z.N Xia , Recent applications of hydrophilic interaction liquid chromatography in pharmaceutical analysis, J Sep Sci 40 (2017) 49–80
[6] T Ikegami , Hydrophilic interaction chromatography for the analysis of biophar- maceutical drugs and therapeutic peptides: a review based on the separation characteristics of the hydrophilic interaction chromatography phases, J Sep Sci 42 (2019) 130–213
[7] G Marrubini , B.E Mendoza , G Massolini , Separation of purine and pyrimidine bases and nucleosides by hydrophilic interaction chromatography, J Sep Sci
33 (2010) 803–816 [8] H.Q Zhao , X Wang , H.M Li , B Yang , H.J Yang , L Huang , Characterization of nucleosides and nucleobases in natural Cordyceps by HILIC-ESI/TOF/MS and HILIC-ESI/MS, Molecules 18 (2013) 9755–9769
[9] K Inoue , R Obara , T Hino , H Oka , Development and application of an HILIC-MS/MS method for the quantitation of nucleotides in infant formula, J Agric Food Chem 58 (2010) 9918–9924
[10] A Goyon , K Zhang , Characterization of antisense oligonucleotide impurities
by ion-pairing reversed-phase and anion exchange chromatography coupled to hydrophilic interaction liquid chromatography/mass spectrometry using a ver- satile two-dimensional liquid chromatography setup, Anal Chem 92 (2020) 5944–5951
[11] R.N Easter , K.K Kroning , J.A Caruso , P.A Limbach , Separation and identifi- cation of oligonucleotides by hydrophilic interaction liquid chromatography (HILIC)-inductively coupled plasma mass spectrometry (ICPMS), Analyst 135 (2010) 2560–2565
[12] L Gong , J.S McCullagh , Analysis of oligonucleotides by hydrophilic interaction liquid chromatography coupled to negative ion electrospray ionization mass spectrometry, J Chromatogr A 1218 (2011) 5480–5486
[13] L Gong , Analysis of oligonucleotides by ion-pairing hydrophilic interaction liq- uid chromatography/electrospray ionization mass spectrometry, Rapid Com- mun Mass Spectrom 31 (2017) 2125–2134
[14] P.A Lobue , M Jora , B Addepalli , P.A Limbach , Oligonucleotide analysis by hy- drophilic interaction liquid chromatography-mass spectrometry in the absence
of ion-pair reagents, J Chromatogr A 1595 (2019) 39–48 [15] R MacNeill , T Hutchinson , V Acharya , R Stromeyer , S Ohorodnik , An oligonu- cleotide bioanalytical LC-SRM methodology entirely liberated from ion-pairing, Bioanalysis 11 (2019) 1157–1169
[16] A Demelenne , M.-J Gou , G Nys , C Parulski , J Crommen , A.-C Servais , M Fil- let , Evaluation of hydrophilic interaction liquid chromatography, capillary zone electrophoresis and drift tube ion-mobility quadrupole time of flight mass spectrometry for the characterization of phosphodiester and phosphorothioate oligonucleotides, J Chromatogr A 1614 (2020) 460716
[17] A Kilanowska , B Buszewski , S Studzi ´nska , Application of hydrophilic interac- tion liquid chromatography coupled with tandem mass spectrometry for the retention and sensitivity studies of antisense oligonucleotides, J Chromatogr
A 1622 (2020) 461100 [18] R.-I Chirita , C West , A.-L Finaru , C Elfakir , Approach to hydrophilic interaction chromatography column selection: application to neurotransmitters analysis, J Chromatogr A 1217 (2010) 3091–3104
[19] B Buszewski , S Noga , Hydrophilic interaction liquid chromatography (HILIC)–a powerful separation technique, Anal Bioanal Chem 402 (2012) 231–
247 [20] S Studzinska , F Lobodzinski , B Buszewski , Application of hydrophilic interac- tion liquid chromatography coupled with mass spectrometry in the analysis
of phosphorothioate oligonucleotides in serum, J Chromatogr B Anal Technol Biomed Life Sci 1040 (2017) 282–288
[21] J.R Thayer , V Barreto , S Rao , C Pohl , Control of oligonucleotide retention
on a pH-stabilized strong anion exchange column, Anal Biochem 338 (2005) 39–47
[22] Z Hao , B Xiao , N Weng , Impact of column temperature and mobile phase components on selectivity of hydrophilic interaction chromatography (HILIC),
J Sep Sci 31 (2008) 1449–1464 [23] Y Guo , N Bhalodia , B Fattal , I Serris , Evaluating the adsorbed water layer
on polar stationary phases for hydrophilic interaction chromatography (HILIC), Separations 6 (2019)
[24] L Li , J.P Foley , R Helmy , Simultaneous separation of small interfering RNA and lipids using ion-pair reversed-phase liquid chromatography, J Chromatogr A
1601 (2019) 145–154 [25] S.T Crooke , J.L Witztum , C.F Bennett , B.F Baker , RNA-targeted therapeutics, Cell Metab 29 (2019) 501
[26] X Guo , M.F Bruist , D.L Davis , C.M Bentzley , Secondary structural characteri- zation of oligonucleotide strands using electrospray ionization mass spectrom- etry, Nucl Acids Res 33 (2005) 3659–3666