Rouven
Stulz
a,
Michael
Lerche
b,
Olivia
Luige
cd,
Agnes
Taylor
b,
Stefan
Geschwindner
e and
Alice
Ghidini
*e
aOligonucleotide Chemistry, Discovery Sciences, BioPharmaceuticals R&D, AstraZeneca, Gothenburg, Sweden
bAdvanced Drug Delivery, Pharmaceutical Sciences, BioPharmaceuticals R&D, AstraZeneca, Gothenburg, Sweden
cDepartment of Biosciences and Nutrition, Karolinska Institutet, Neo, Huddinge, 14183, Sweden
dEarly Chemical Development, Pharmaceutical Sciences, BioPharmaceuticals R&D, AstraZeneca, Gothenburg, Sweden
eMechanistic and Structural Biology, Discovery Sciences, BioPharmaceuticals R&D, AstraZeneca, Gothenburg, Sweden
First published on 23rd October 2023
The recent and rapid increase in the discovery of new RNA therapeutics has created the perfect terrain to explore an increasing number of novel targets. In particular, antisense oligonucleotides (ASOs) have long held the promise of an accelerated and effective drug design compared to other RNA-based therapeutics. Although ASOs in silico design has advanced distinctively in the past years, especially thanks to the several predictive frameworks for RNA folding, it is somehow limited by the wide approximation of calculating sequence affinity based on RNA–RNA/DNA sequences. None of the ASO modifications are taken into consideration, losing hybridization information particularly fundamental to ASOs that elicit their function through RNase H1-mediated mechanisms. Here we present an inexpensive and enhanced biophysical screening strategy to investigate the affinity of ASOs for their target RNA using several biophysical techniques such as high throughput differential scanning fluorimetry (DSF), circular dichroism (CD), isothermal calorimetry (ITC), surface plasmon resonance (SPR) and small-angle X-ray scattering (SAXS).
In traditional small-molecule drug discovery, large libraries of chemical compounds are screened for activity in cellular, biochemical, and biophysical assays to identify genuine hits which ideally present specific activity towards a selected biological target. In contrast to the central small molecule drug discovery dogma, the identification of ASO sequences that are pharmaceutically active is based on a so-called HIT finding followed by RNA walk. First, the target mRNA regions are chosen based on the criteria of accessibility (e.g., loops and bulges), identified through a combination of several predictive bioinformatics programs like PFRED.1 Then the gapmer sequences are systematically walked along the mRNA target sequence (e.g., ±15nt from initial hit), generating a library of ASOs considered thermodynamically favourable. In this process the gapmer architecture is kept constant. One of the most used gapmer structures is the LNA–DNA–LNA, 3–10–3 configuration, where the flanking side consists of 3 LNA monomers and the central part is DNA, and the sequences is generally fully phosphorothioate.2,3 A major limitation of this strategy is that the computationally predicted hybridization properties are based only on the sequences without considering the nucleoside modifications influential for the affinity to the target. The second limitation is that most of the models consider only the most favourable secondary structure of the RNA, neglecting the fast dynamics of RNA structures and their tertiary structures. The third issue is that the relationship between the denaturation temperature of a duplex (Tm), which represents the temperature at which half of the ASO is in a double stranded conformation with its target RNA and half is free in solution, and the dissociation constant Kd is not linear. It has been suggested in fact that the correlation is parabolic and has a high variance, challenging the possibility of designing predicting tools.4
It has recently been shown experimentally that enhancing the stereochemical uniformity in the sequence, introducing modifications in one position in the DNA sequence,5,6 or changing the linkage type7 can dramatically change the toxicity and efficacy of gapmers. To date, there has not been any explanation for these phenomena, but there may be several components influencing it, including intracellular protein binding.6 Papargyri et al., at the Roche Innovation Centre Copenhagen, have recently evaluated 768 gapmers (both sequence and architecture diverse) in HeLa cells for target knockdown activity and cytotoxic potential.8 They found that ASO/RNA binding affinity and the presence of certain short sequence motifs in the gap region can explain differences between knockdown/cytotoxic ASO clusters. As a conclusion, they proposed machine-learning models to be used to predict region-specific, optimal LNA-modification architectures. A more extensive profiling of the ASO/RNA duplex formation, including binding kinetics, would provide a wealth of information that might be considerably more predictive of in vivo behaviour than Tm. High-throughput biophysical assays and target binding kinetics of oligonucleotide drugs are generally tedious to perform and are rarely reported. We therefore worked on designing a detailed biophysical screening process to generate affinity, kinetics, and structural data to be used for prospective machine learning.
High-throughput differential scanning fluorimetry (DSF) is a screening method normally used to analyse the stability of a protein through the temperature transitions in response to environmental factors such as pH, salts, or the presence of ligands.11–13 Upon temperature increase, the protein gradually unfolds, and more hydrophobic side chains become accessible and the fluorescent reporter dye (e.g., SYPRO orange or 1-anilinonaphthalene-8-sulfonic acid) changes its fluorescence in response to a change in hydrophobic amino acid side chains presented to the environment. We found it timely to exploit the low-volume and high-throughput format of DSF for the characterization of the thermodynamic stability of libraries of ASOs and their corresponding RNA target.
DSF was previously and successfully applied for the measurement of RNA stabilisation in addition to additives such as mono- and divalent cations as well as polyamines.14 The assay makes use of an extrinsic fluorescent dye (Quant-iT RiboGreen RIBO from Life Technologies) that fluoresces with an intensity, several orders of magnitude greater when bound to nucleic acids compared to the unbound form. RiboGreen has been shown to have specificity for structured vs. unstructured RNA, specifically in being able to distinguish between double stranded RNA14,15 and triplex structures.16,17 We thus hypothesised that this specificity might be employed to assess the unfolding process of the ASO/RNA structure. As previously described by Silvers et al., the process goes through three main fluorescence phases.14 Initially, the dye binds in a competent way to the highly stable and structured ASO/RNA duplex. That process corresponds to the initial higher fluorescence value that decays with the increase of temperature. The curve of the fluorescence intensity reaches its inflection point at a temperature corresponding to the Tmax of the ASO/RNA duplex. The melting process of ASO/RNA continues, and the binding affinity of the fluorescent dye decreases until fluorescence completely disappears when the ASO and RNA are completely unfolded.
As a model sequence to test the Tmax measurement assay, we selected a PCSK9 specific LNA antisense oligonucleotide sequence which was previously evaluated in nonhuman primate pharmacology studies.‡18,19 The SPC5001 gapmer, complementary to the human sequence (Fig. 1, GenBank accession no. #NM174936), was shown to potentially reduce PCSK9 mRNA levels in treated cells. Early studies suggest that PCSK9 mRNA levels were reduced by 60% within 24 hours, and the effect was sustained for over 16 days.18 In non-human primates, the hepatic PCSK9 mRNA levels and plasma LDL-C levels decreased by 85 and 50%, respectively.19 Meanwhile, subcutaneous administration of SPC5001 lowered the PCSK9 mRNA and LDL-C levels in a dose-dependent manner by ∼50 and 25%, respectively, in healthy human volunteers.
The clinical development of SPC5001 was terminated due to transient renal tubular toxicity, and injection site reactions were also observed. The exact cause of acute kidney injury remains uncertain to date, reinforcing the need to collect more biophysical data on ASO sequences to generate correlation between their biophysical properties and their clinical output. The LNA–DNA–LNA gapmer sequence of SPC5001 (ASO 1) was used as a reference tool and compared to the corresponding MOE–DNA–MOE (ASO 2) and the full 2′OMe sequence (ASO 3) (Table 1, Fig. S1 and Table S1, ESI†). As expected, the substitution of LNA in the wings with MOE decreased the stability of the duplex ASO/RNA by about 14.5 °C (Fig. 2C). The full substitution of ASO 1 with 2′OMe retained its affinity. Presumably, the loss of affinity due to the substitution of LNA with 2′OMe was compensated through the gain of affinity due to the substitution of DNA. RNAs and ASOs alone did not show any fluorescent curve, confirming that the interaction of the dye is specific and not influenced by the chemistry of the ASOs. When looking at different RNA lengths, no difference in affinity was detected for the chosen ASOs. The 48-mer long RNA was designed to have a target site with a similar secondary structure compared to the 24-mer RNA, as a result the melting temperature curves and the Tm values resulted in being comparable, highlighting the flexibility of the DSF assay in terms of RNA sequence length. The melting experiment was repeated removing the salts from the buffer, but unexpectedly the Tm values were not significantly affected by the change (Fig. S2, ESI†).
ASO | Sequence | Sugar composition | Length |
---|---|---|---|
1 | TGCTACAAAACCCA | LNA gapmer 3-8-3 | 14 |
2 | TGCTACAAAACCCA | OMe gapmer 3-8-3 | 14 |
3 | UGCUACAAAACCCA | Full 2′OMe | 14 |
4 | TGCT*ACAAAACCCA | LNA gapmer 3-8-3 | 14 |
5 | TGCTA*CAAAACCCA | LNA gapmer 3-8-3 | 14 |
6 | AUUAAUAAAAAUGC | LNA gapmer 3-8-3 | 14 |
7 | AACCCAGAAUAAAU | LNA gapmer 3-8-3 | 14 |
8 | TGATACAAAACCCA | LNA gapmer 3-8-3 | 14 |
9 | TGATACAAAACCCA | MOE gapmer 3-8-3 | 14 |
10 | TGCTAAAAAACCCA | LNA gapmer 3-8-3 | 14 |
11 | TGCTAAAAAACCCA | MOE gapmer 3-8-3 | 14 |
12 | GTCTGTGGAAGCG | LNA gapmer 2-8-3 | 13 |
13 | CGTCAGTATGCGAATC | LNA gapmer 3-8-3-1 | 16 |
To evaluate the contribution of the dye on the stabilization of the duplex, the temperature values obtained by DSF were then compared to Tm values measured using UV absorbance (Fig. 2B and Fig. S4, ESI†). The difference between Tmax and Tm was about ±3 °C, confirming the potential of the DSF assay to deliver accurate data for several ASOs and RNAs while increasing the throughput significantly. The flexibility of the assay was also validated by measuring Tmax values for duplexes with the longer version of the RNA target. For our purpose, the 48-mer RNA was designed as an extension on both 3′ and 5′ of the 24-mer probe (Fig. 1B) and the Tmax values remained unchanged.
To validate the results obtained by DSF, we also performed circular dichroism (CD) spectroscopy. CD can be used to study the chirality of macromolecules in solution, influenced by secondary, tertiary, and quaternary structures.20–22 Increasing the temperature while observing changes in the CD spectra makes it possible to quantify temperature-dependent unfolding or dissociation of macromolecules without the need for dyes. CD spectra were recorded for ASO 1–3 and ASO 13 in the absence and presence of RNA. Temperature ramps were performed from 40 °C to 90 °C and melting temperatures were determined (Fig. S5, ESI†), showing consistent results with those obtained from UV-Vis spectroscopy and DSF. Additionally, the CD of the individual ASOs indicates some differences in the secondary structure. ASO 1 and ASO 2 share characteristic peaks at 220 and 280 nm (Fig. S5, ESI†), while the 280 nm peak has shifted to 270 nm for ASO 3. Introducing RNA to each of the ASOs resulted in a significant change in the CD spectrum for ASO 1–3 but not for ASO 13, confirming the latter's inability to bind RNA.
To date, in fact the main empirical characterization of ASOs revolves around their Tm. This evaluation allows the determination and comparison between the affinities of the ASOs and their mRNA targets at their respective melting temperatures but provides no information on the binding kinetics. For oligonucleotides, the off-rates are directly correlated to their length and chemistry, highlighting the relationship between off-rates and affinity. Therefore, improving the koff of an mRNA targeting oligonucleotide drug is relatively easy and could be modulated by adjusting these two components. In contrast, the optimization of on-rates, kon, is largely independent of oligonucleotide length.23 However, the value of these changes is questionable as mRNAs have short half-lives, and dissociation from cleaved targets needs to be fast for their catalytic mechanism.
A limited number of attempts have been made to study RNA and ASO binding by SPR.24 Using the same tool ASO sequences along with the 5′-biotinylated version of the 24-mer RNA, we developed a high through-put SPR assay which would allow for the analysis of several sequences. The SPR kinetic studies are conducted close to or at physiological temperatures (25 °C or 37 °C), which are below the melting temperature of the complex. One of the most arduous steps while developing the protocol has been finding regeneration conditions that are strong enough to release ASOs from the RNA, without degrading the RNA probe on the chip. This is particularly true while working with LNA gapmers due to their stronger affinity for the RNA target, which makes them less prone to separation without the help of high temperature, or strong ionic buffers. Although the DSF showed a substantial affinity difference between ASO 1 and ASO 2, the SPR assay at 25 °C showed Kd in the similar range (picomolar) (Table S2, ESI†), but the gap between the modifications becomes larger when the assay is run at 37 °C (Table S3, ESI†), where MOE and 2′OMe ASOs jump to nanomolar Kd. As observed previously through DSF (Fig. 2C and D), the introduction of a mismatch in the DNA stretch of the gapmer seems to be less tolerated than in the wings (ASO 8 and ASO 10, Fig. 3B and C). The koff was in this case considerably higher at 37 °C (Table S3, ESI†) than at 25 °C (Table S2, ESI†). At 37 °C, one mismatch was enough to dramatically reduce the affinity of the MOE gapmers (ASO 9 and ASO 11, Fig. 3B and C), confirming the decrease of the Tm values.
Fig. 4 Isothermal calorimetry. Binding isotherms for a 1:1 system of ASO 1, 2, 3 and 24nt RNA at (A) 37 °C and (B) 45 °C. |
The entropic component was particularly favourable for the LNA gapmer at 37 °C compared to the other modifications, which is in line with the DSF data and can be attributed to an increased pre-organization of the single-stranded LNA oligonucleotide towards double strand RNA helices. Minor differences were observed for the MOE gapmer and the fully modified 2′OMe sequence compared to the LNA gapmers. All ASOs control injections showed a slight decrease in enthalpy while increasing the concentration of oligonucleotides in the cell, probably due to a change of the buffer conditions or due to some nonspecific aggregation of the PS oligonucleotides. Interestingly, the ΔG of different sequences was almost the same. Also, ΔG at different temperatures was almost identical; the enthalpic factor decreased substantially with temperature and compensated the increase of the entropic component.
The nature of the DSF assay allows the analysis of longer RNA target strands, adding flexibility to the protocol and a more accurate comparison between ASOs, resulting from the same RNA walk. The DSF throughput can be substantial in multi-well thermal cycling machines, thereby allowing the sensitive and simultaneous measurement of 384 samples in a single experiment. Compared to UV-Vis spectroscopy, the throughput is increased by a factor of 50, the experimental time is reduced by a factor of 5 along with the required amounts of RNA and ASO. The SPR assay here describes the distinction of sequences based on their kinetics properties in different temperature and salt conditions, allowing a better understanding of the influence of monomer and backbone modifications on the overall kinetic parameters.
A theoretical explanation of the observation that long, complementary oligonucleotides tend to be less potent than shorter versions, which consequently have a smaller affinity, has been proposed by Pedersen et al.4 However, to date, this theory is derived only from a small sample of data. They showed that the effect is due to affinity rather than length; we, therefore, designed biophysical techniques that will allow a fast generation of affinity, kinetics and structural data to accelerate an understanding on the development of ASOs.
After synthesis, the oligonucleotides were cleaved and deprotected for 1 hour at 65 °C using 1:1 methylamine (40%) aqueous ammonia (35%). After cleavage, the crude oligonucleotide was evaporated to dryness in a SpeedVac vacuum concentrator, purified via HPLC, and the pure fractions were freeze dried to give a white powder (see Fig. S1, ESI†).
In a typical experiment, RNA and ASOs were dissolved in a buffer and dispensed into the plate to reach a final concentration of 4 μM in 10 μL. Then 50 nL of Quant-iT™ RiboGreen® (RIBO, Life Technologies, Ex max = ∼500 nm, Em max = ∼525 nm) dye stock solution in DMSO was added to each well. The plates were then sealed with an optically clear adhesive seal (Thermo Scientific) and centrifuged at 1000 × g for 2 min.
The fluorescence intensity was monitored using the SYBR Green I/HRM dye filter combination (465–510 nm) from 20 to 95 °C at a ramp rate of 0.03 °C s−1 with 36 acquisitions per °C. Melting profiles were obtained through Tmax analysis using the LightCycler® 480 software. To ensure that the measurements were statistically valid, which means that they are reproducible, three to six runs were performed. The replicates were averaged, and the standard deviation was used for error determination. In some measurements, the concentration of RNA or ASO was also varied to see if this could influence the resulting melting curve/temperature.
This double-referenced data set has been fitted using a simple 1:1 interaction model with drift in order to extract kinetic and affinity data. All data were expressed as mean ± standard deviation and further analysed using GraphPad Prism 8.
Footnotes |
† Electronic supplementary information (ESI) available. See DOI: https://doi.org/10.1039/d3cb00072a |
‡ AstraZeneca is neither developing nor commercialising this PCSK9 specific LNA antisense oligonucleotide – SPC5001. |
This journal is © The Royal Society of Chemistry 2023 |