Since the antisense oligonucleotide (ASO) concept (Zamecnik and Stephenson, 1978) and RNA interference mechanism (Fire et al., 1998) were disclosed, it took 20 years to introduce RNA therapeutics into the clinic. The first ASO drug, formivirsen, and the first small interfering (si)RNA drug, patisiran, were approved by the U.S. Food & Drug Administration in 1998 and 2018, respectively. The antisense concept could be translated into drugs owing to advances in many fields, including chemistry (medicinal, analytical, organic, and process), biology (RNA, genomics), and pharmacology. Specifically, innovations in medicinal chemistry, including endowing oligonucleotides with drug-likeness and sufficient binding affinity while maintaining target RNA degradation activity via natural enzymatic cleavage, were key. In addition to the several RNA-targeted drugs, including nine ASO drugs (Table 1) and five siRNA drugs, the two mRNA vaccines for COVID-19 approved in 2021 advanced the era of RNA therapy.

Table 1 Approved ASO drugs

Drug	Approval year	Target indication	Target gene	Mode of action	Chemistry
Formivirsen	1998	CMV retinitis	CMV	RNase H1	PS-ODN
Mipomersen	2013	HoFH	ApoB-100	RNase H1	PS-MOE gapmer
Nusinersen	2016	SMA	SMN2 intron 7	Exon inclusion	PS-MOE
Eteplirsen	2016	DMD	Dystrophin exon 51	Exon skipping	PMO
Inotersen	2018	hATTR	TTR	RNase H1	PS-MOE gapmer
Golodirsen	2019	DMD	Dystrophin exon 53	Exon skipping	PMO
Volanesorsen	2019	FCS	ApoC-III	RNase H1	PS-MOE gapmer
Viltolarsen	2020	DMD	Dystrophin exon 53	Exon skipping	PMO
Casimersen	2021	DMD	Dystrophin exon 45	Exon skipping	PMO

ApoB-100, apolipoprotein B-100; ApoC-III, apolipoprotein C-III; CMV, cytomegalovirus; DMD, Duchenne muscular dystrophy; hATTR, hereditary transthyretin-mediated amyloidosis; FCS, familial chylomicronemia syndrome; HoFH, homozygous familial hypercholesterolemia; MOE, 2′-methoxyethyl; ODN, oligodeoxynucleotide; PMO, phosphorodiamidate morpholino oligonucleotide; PS, phosphorothioate; SMA, spinal muscular atrophy; TTR, transthyretin. For details on chemistries, refer to the figures below.

RNA-targeted drugs bind to the complementary sequences of their target RNAs through Watson–Crick hybridization; therefore, they can modulate target RNA function very specifically, with few off-target effects, and they can be rationally designed based on sequence information. In addition, they are ideal for precision medicine or personalized medicine. For example, milasen was developed for a single patient named Mila (Kim et al., 2019). Most importantly, with RNA-targeted drugs, nearly all disease-related targets that are “undruggable” or “undrugged” by conventional medicinal approaches can theoretically be controlled. Only approximately 15% of all proteins are considered “druggable” (Makley and Gestwicki, 2013); therefore, 85% of proteins have not been targeted by small-molecule drugs or biologics and can be targeted by RNA therapeutics. Only 2% of human DNA is translated into proteins, while 90% of the genome is transcribed into RNAs, most of which are non-protein-coding (Djebali et al., 2012). Non-coding RNAs are also considered to be potential targets of RNA therapeutics.

RNA-targeted oligonucleotide drugs are either single-stranded or double-stranded. In contrast to double-stranded siRNAs, which can only degrade a target RNA to downregulate gene expression, ASOs possess more versatility as they can not only degrade RNA but also upregulate gene expression or manipulate mRNAs through diverse modes of action. Furthermore, ASOs can undergo more chemical modifications than siRNAs as the chemical modification of the latter is only limited to AGO2 enzyme substrates. Therefore, this review focuses on ASOs. It summarizes the past, present, and future of the medicinal chemistry of ASOs, including how to endow oligonucleotide derivatives with drug-likeness to allow them to overcome evolutionary defense systems that act against invading RNAs; such systems include lipid bilayers, RNases, the reticuloendothelial system, immunogenicity, and endocytosis (Dowdy, 2017). In addition, it discusses discoveries regarding the molecular mechanisms underlying the pharmacology, pharmacokinetics, and toxicology of ASOs, and perspectives for improving the therapeutic potential of ASOs by enhancing their therapeutic index and resolving the issue of tissue/cell-specific delivery.

Short DNA/RNA-like oligonucleotides with phosphodiester bonds are not drug-like because they are not able to cross the cell membrane, are rapidly degraded by nucleases and cleared by the kidneys through glomerular filtration, and activate innate immune responses. Even as they are taken up into the cells via endocytosis, they remain trapped inside the lipid bilayer of the endosome and cannot access their target RNAs in the cytosol or nucleus to exert pharmacological effects (Dowdy, 2017).

During the last decades, to endow ASOs with drug-likeness, various chemical modifications have been made and tested. Chemical modifications can be made in various sites of nucleotides, including the linkers, sugars, and nucleobases. Fig. 1 summarizes key chemical modifications that have already been applied in commercial products or are considered promising. Fig. 2 shows representative non-ribose sugar modifications, such as phosphorodiamidate morpholino oligomers (PMOs) and peptide nucleic acids (PNAs).

Figure 1. Chemical modifications in ASOs.

Figure 2. Non-ribose sugar modifications in ASOs.

Among them, however, only four chemical classes, including phosphorothioate (PS) oligodeoxynucleotides (ODNs), PS-2′-O-methoxyethyl (MOE) gapmers, and occupancy-only oligonucleotides such as PS-MOEs and PMOs, have been clinically accepted, as shown in Table 1 and Fig. 3. Therefore, in this section, the chemistries, designs, and pharmacological properties of these classes of ASOs will be described.

Figure 3. ASO designs. BNA, bridged nucleic acid; cEt, constrained ethyl; LNA, locked nucleic acid; MOE, methoxyethyl; PMO, phosphorodiamidate morpholino oligonucleotide; PNA, peptide nucleic acid; PS, phosphorothioate.

Nucleobase modifications

While numerous nucleobase modifications have been attempted to date, the only clinically accepted and commonly used modification to avoid CpG immune stimulation and enhance nuclease stability and binding affinity is 5-methyl cytosine (MeC, Fig. 1).

Linker modifications

The first breakthrough innovation for ASO drug-likeness was the introduction of a PS linkage (Fig. 1) (Stec et al., 1984). In this linker, one non-bridging oxygen of phosphodiester is replaced with a sulfur atom, which greatly alters the physicochemical and pharmacological properties of the resulting ASO (Sands et al., 1994; Eckstein, 2000). The PS modification greatly enhances nuclease resistance. Another key benefit of the PS linkage is that the negative charge in PS is more widely distributed than that in a phosphodiester (PO) bond, increasing lipophilicity. This prevents rapid renal clearance through stronger binding with plasma proteins, and enhances cellular uptake through stronger binding with cell surface proteins such as stabilin-1 and 2 (Miller et al., 2016). However, its stronger binding to immune receptors results in non-specific pro-inflammatory toxicity (Yanai et al., 2011), and the PS modification reduces ASO binding affinity to its complementary nucleic acids by –0.5°C per modification (Freier and Altmann, 1997). The lower potency and pro-inflammatory effect of PS-ODNs (Fig. 3) limit their clinical application and therefore, PS-ODNs require additional chemical modifications.

Sugar modifications

Sugar modification with 2′-alkyl substitution has proven to be a useful strategy to overcome the drawbacks of the PS modification. The most widely used modification, 2′-MOE (Fig. 1), has several advantages, including increased potency resulting from a higher binding affinity (T_m increase of 1.5°C per modification) (Freier and Altmann, 1997), enhanced nuclease resistance, decreased immune-stimulating effect (Henry et al., 2000), and prolonged tissue elimination half-life (Geary et al., 1997, 2001). The bulky alkoxy group at the 2′ position drives the conformational equilibrium of the furanose ring toward the RNA-like C3′-endo conformation (Plavec et al., 1993), which has a higher binding affinity for target RNA than the DNA-like C2′-endo conformation due to the gauche effect. Crystal structure analysis supported this conformation and revealed that hydration of MOE and phosphate oxygen plays a key role in overall stability (Teplova et al., 1999). As it does not support RNase H-mediated degradation, it has been used for occupancy-only oligonucleotides or as flanks or wings of RNase H-active MOE gapmers (Fig. 3).

As mentioned above, the C3′-endo conformation is preferred for strong binding with the target RNA. This conformation is fixed by creating a covalent tether between the 2′ and 4′ positions, leading to a 2′-O,4′-C-methylene-bridged nucleic acid (Fig. 1) or locked nucleic acid (LNA) (Obika et al., 1997; Koshkin et al., 1998) to afford unprecedentedly strong binding affinity for nucleic acids (T_m increase of ~5°C per modification). Owing to the high binding affinity, LNA gapmers (Fig. 3) show strongly improved potency, but are associated with an increased risk of hepatotoxicity, which may depend on ASO sequence and design (Swayze et al., 2007; Hagedorn et al., 2013). Interestingly, the addition of a methyl group to the LNA structure to form 2′,4′-constrained ethyl (cEt, Fig. 1) dramatically reduced hepatotoxicity without a loss in potency as evident in the cEt gapmer (Fig. 3) (Seth et al., 2009b). Although ASOs with such modifications (LNA/cEt gapmers) have not been approved for commercial use, numerous candidates are in clinical trials.

Non-ribose sugar modifications

PMOs (Fig. 2) are a neutral class of oligonucleotide analogs in which a morpholino ring replaces the ribofuranose and a phosphorodiamidate linker replaces the PO (Summerton and Weller, 1997). As they also do not support RNase H activity, they have been used as occupancy-only oligonucleotides for exon skipping and successfully marketed for the treatment of Duchenne muscular dystrophy (DMD). PMOs have the advantage of skeletal muscle-selective distribution, which makes them more efficacious for DMD treatment than other RNase H-inactive structures, such as 2′-MOE or cEt (Sheng et al., 2020).

PNAs (Fig. 2) are another neutral class of oligonucleotide analogs comprising a peptide scaffold and nucleobases for more efficient Watson–Crick hybridization with complementary nucleic acids. PNAs also do not support RNase H-mediated degradation and act as steric blockers or splicing modulators. However, they have low water solubility and poor cell permeability, limiting their application to molecular genetic diagnostics. There are no successful therapeutic PNAs as yet, despite several efforts to overcome the drawbacks, including structural modifications of the PNA backbone (Gupta et al., 2017), conjugation with cell-penetrating peptides (Taylor and Zahid, 2020) or N-acetylgalactosamine (GalNAc) (Bhingardeve et al., 2020), or lipid nanoparticle formulation (Gupta et al., 2016). The diverse efforts for PNA delivery have been reviewed elsewhere (Volpi et al., 2020).

ASO design and main mechanisms of action

All marketed ASO drugs can be classified in two groups based on their mechanisms of action: RNase H-mediated degradation for PS-ODN (formivirsen) and MOE gapmers (mipomersen, inotersen, and volanesorsen), or alternative splicing (exon inclusion or skipping) for occupancy-only oligonucleotides (PS-MOEs such as nusinersen and PMOs including eteplirsen, golodirsen, viltolarsen, and casimersen).

In RNase H-mediated degradation, the RNase H1 enzyme specifically recognizes and cleaves RNA-DNA-like heteroduplexes (Wu et al., 2004; Cerritelli and Crouch, 2009). For the ASO in an RNA-ASO duplex to be RNase H-active, at least five (Monia et al., 1993), but optimally, eight to ten contiguous deoxynucleotides are required. In addition, PS linkages can activate RNase H1 as PO linkages. Therefore, the central PS-ODN region of 8-10 nucleotides (gap) is flanked by various 3-5-nt 2′-modified nucleotides (wings) such as MOE, LNA, or cEt, which contribute to increased stability and affinity, to generate a gapmer structure (Fig. 3).

Since all 2′ modifications and non-ribose sugar modifications result in loss of RNase H activity, they result in occupancy-only oligonucleotides unless RNase H substrate ODNs are included (Fig. 3). The occupancy-only oligonucleotides bind to the target pre-mRNA and sterically block the spliceosome, leading to alternative splicing in the form of exon inclusion or skipping (nusinersen and four PMO drugs); as a result, they can be used in the treatment of spinal muscular atrophy (Garber, 2016) or DMD (Echevarría et al., 2018).

Spinal muscular atrophy is caused by homozygous loss of function of the survival of motor neuron 1 gene (SMN1). Although the redundant SMN2 gene produces limited amounts of full-length SMN protein, it frequently produces short and unstable protein because of exon 7 skipping during pre-mRNA splicing. Nusinersen is an 18-mer fully 2′-MOE-modified PS ASO that binds to and modifies the splicing of SMN2 pre-mRNA to include exon 7, thus restoring full-length SMN expression (Hua et al., 2007).

DMD is a rapidly progressing neuromuscular disorder. Mutations in the DMD gene that disrupt its reading frame introduce premature stop codons, halting gene expression. Eteplirsen repairs the reading frame by binding to exon 51 in the pre-mRNA, which is then skipped in translation, producing a shortened, but functional dystrophin (McDonald et al., 2021). Similarly, golodirsen (Servais et al., 2022) and viltolarsen (Clemens et al., 2020) adjust the splicing pattern of mutant DMD pre-mRNA by binding to exon 53. The absence of exon 53 in the mature mRNA restores dystrophin expression. Casimersen (Wagner et al., 2021) binds to and promotes the skipping of exon 45.

Additional molecular mechanisms for gene expression regulation by ASOs

In addition to the RNase H-mediated degradation or alternative splicing described above, ASOs have several other mechanisms to downregulate or upregulate gene expression, enhancing their versatility. Evolved RNA biology has identified diverse post-RNA-hybridization events of ASOs that are broadly classified into two mechanisms: enzymatic degradation and occupancy-only, as summarized in Table 2 (Crooke et al., 2021). Two enzymes degrade RNA-oligonucleotide duplexes. In addition to RNase H discussed above, ASO-RNA hybrids can recruit Argonaute 2 (AGO2) to degrade the target RNA. Occupancy-only can be further classified into alternative splicing and steric blocking. Alternative splicing of pre-mRNA leads to exon inclusion or skipping, as explained above, or it leads to nonsense-mediated decay (NMD) if the resulting mRNA contains a premature termination codon. The effects of steric blocking on the regulation of gene expression are known to vary based on the ASO binding site. The detailed mechanisms and structure-activity relationships in terms of the endpoint of gene regulation, i.e., downregulation or upregulation, are discussed below.

Table 2 Modes of action and relevant chemistry designs of ASOs

	Enzymatic degradation	Occupancy-only
		Alternative splicing	Steric blocking
		Downregulation	RNase H-mediated	NMD	NGD
AGO2-mediated	Translation arrest
	5′-cap inhibition
	Altering the polyadenylation site
Upregulation		Exon inclusion	miRNA function inhibition
		Exon skipping	Modulation of uORF/TIE
			Inhibition of NMD
Chemistry	RNase H: DNA (w/wo PS linkage) or gapmer	Fully 2′-modified RNA
	AGO2: dsRNA (siRNA w/wo 2′- F/OMe) or ssRNA (5′-phosphate)	PMO
		PNA

AGO2, Argonaute 2; NGD, no-go decay; NMD, nonsense-mediated decay; uORF, upstream open reading frame; TIE, translation inhibitory element.

Downregulation of gene expression: AGO2-mediated degradation is the main mechanism by which miRNA/siRNA cleaves its target RNA (Song et al., 2004). However, single-stranded siRNA stabilized with 2′-modification and a 5′-(E)-vinylphosphonate, which mimics the 5′-phosphate required for AGO2 activation, can be a substrate (guide strand) of AGO2 (Lima et al., 2012).

There are two RNA degradation mechanisms other than degradation by RNase H or AGO2: NMD and no-go decay (NGD). NMD of mRNA is initiated when ASO triggers splicing modulation of pre-mRNA to generate an mRNA that contains premature termination codons, which are NMD targets (Ward et al., 2014). Fully modified 2′-MOE ASOs that bind to the 3′ terminus of the open reading frame (ORF) degrade target RNA via NGD (Liang et al., 2019b). Fully modified cEt ASOs, which have a higher binding affinity to RNA are known to have a lower efficacy, which may be a result of the stronger binding to proteins. 2′-MOE ASOs with only one cEt have a similar efficacy, while less affined 2′-OMe ASOs have a lower efficacy.

A few reports have disclosed steric blocking as a minor mechanism of downregulation. ASOs can sterically block the translation of mature mRNAs (Melton, 1985; Iversen et al., 2003) and guide the cleavage of 5′-cap structures, inhibiting translation (Baker et al., 1999). Finally, ASOs can modulate the use of polyadenylation sites by binding to pre-mRNA sequences involved in cleavage and polyadenylation (Vickers et al., 2001).

Upregulation of gene expression: While miRNAs generally downregulate gene expression, ASOs can upregulate gene expression through the inhibition of miRNA function by binding to miRNAs or miRNA-binding sites (Ørom et al., 2006; De Santi et al., 2020)

Upstream ORFs in the 5′-UTR are sequences that are defined by an initiation codon in frame with a termination codon located upstream or downstream of the main AUG. They are present in approximately half of the human transcripts and inhibit protein expression (Calvo et al., 2009). A fully 2′-modified PO or PS-ASO targeting an upstream ORF dose-dependently increased protein expression by 30-150% (Liang et al., 2016b). Eighteen-mer PS-MOE, 12-16-mer PS-2′-OMe, and 16-mer PO-2′-OMe ASOs have been suggested as optimal lengths and chemistries for gene upregulation.

Translation was enhanced approximately 3-fold when the recruitment of translation initiation factors to the target mRNA was improved through the binding of fully 2′-modified 16-mer PS-cEt to translation inhibitory elements in 5′-UTRs (Liang et al., 2017).

As NMD regulates the normal expression of many genes, ASOs can upregulate gene expression by inhibiting NMD either through the degradation of NMD factors (Huang et al., 2018) or through targeting the binding site of the exon junction complex in an mRNA (Nomakuchi et al., 2016; Kim et al., 2022). Splicing modulation of an ASO preventing non-productive (NMD-inducing) alternative splicing upregulated gene expression (Lim et al., 2020).

Many years of research on the mechanisms underlying the potency, pharmacokinetics, and toxicity of ASOs have revealed that ASO-protein interaction is a key factor (Crooke et al., 2020a, 2020b). In general, the binding affinity of ASOs for several proteins depends on the PS linkage, gapmer design, and sequence (Vickers and Crooke, 2016). In addition, more hydrophobic 2′ modifications, such as fluorine (F), LNA, or cEt, tend to enhance protein binding compared to more hydrophilic 2′ modifications, such as MOE or OMe, leading to an approximate 10-fold difference in binding affinity for diverse proteins. Many proteins prefer binding to the 5′-wing and the ~6 nucleotides of the 5′-gap position, which contain more hydrophobic 2′ modifications (Shen et al., 2019). This section discusses the structure-activity relationship of ASO-protein interaction and its effects on ASO properties, such as nuclease resistance, pharmacokinetic profile, and cellular uptake and endosomal escape capabilities.

Nucleases

Ubiquitous nucleases cleave the PO bonds of unmodified DNA-type oligonucleotides, which therefore have a half-life of only minutes. PS linkages stabilize ASOs against nuclease activity, increasing their plasma half-life to up to 1 h. Chemical modification of the 2′ position with MOE, cEt, or LNA eliminates exonuclease activity, extending tissue elimination half-life to four weeks. MOE gapmers are highly resistant to exonucleases, but are slowly metabolized by endonucleases at the DNA gap site to yield short metabolites of DNA ends, which are degraded further by exonucleases and/or excreted in the urine due to weak plasma protein binding.

Plasma proteins

Plasma protein binding is the main factor for improving absorption and distribution by suppressing renal clearance (Gaus et al., 2019). As mentioned above, PS linkages are the key structure for protein binding. As effective binding requires at least 10-12 PS moieties, metabolites from endonucleases and exonucleases that have less than 10 PS groups are cleared from tissues and excreted in the urine via glomerular filtration (Geary et al., 2015). As albumin is the most abundant protein in plasma (600 µM) and its binding affinity to a PS-MOE gapmer has a K_d of 12.7 µM (Gaus et al., 2019), it plays a major role in the plasma protein binding of PS-ASOs. The binding affinity is sufficiently high to effectively inhibit renal clearance, but sufficiently low to facilitate tissue uptake. In addition, the binding affinity depends on the chemical structure and sequence of the ASO. For example, occupancy-only full PS-MOE has a K_d of 26.9 µM, and T20 PS-ODN has a K_d of 0.94 µM, whereas A20 PS-ODN has a K_d of 204.5 µM.

Cell surface proteins

ASOs are taken up into cells via endocytosis (Doherty and McMahon, 2009). Non-productive uptake results in the accumulation of ASOs in late endosomes and lysosomes, whereas productive uptake involves clathrin- and caveolin-independent endocytosis (Koller et al., 2011). Among the numerous cell surface proteins, stabilin-1 and 2 receptors are responsible for ASO cellular uptake and internalization (Miller et al., 2016). A binding affinity study (Gaus et al., 2018) revealed that PS-ODN had a similar affinity to full PS-MOE, PS-MOE gapmer (5-10-5), and PS-cEt gapmer (5-10-5), which implied that 2′ modifications do not strongly contribute to the binding affinity. The binding affinity decreased as the number of PS linkages decreased (15 PS has a K_d of 83.9 µM, whereas 10 PS and PO-MOE gapmer (5-10-5) have a K_d of >1,000 µM), which indicated that PS linkage is the major contributor to stabilin-2 binding and that at least 15 PS linkages are required for appropriate cellular uptake.

Asialoglycoprotein receptor is highly expressed on hepatocytes (Stockert, 1995). GalNAc is a ligand of this receptor and has been conjugated to an ASO (Prakash et al., 2014) as well as an siRNA (Zimmermann et al., 2017) for liver/hepatocyte-specific delivery. ASOs can be selectively delivered to pancreatic β-cells via conjugation with glucagon-like peptide 1 receptor agonist (Ämmälä et al., 2018; Knerr et al., 2021).

Intracellular proteins

After cellular internalization, ASOs are trafficked through multiple membrane-bound intracellular compartments. They then slowly escape endosomes into the cytoplasm and nucleus, where they bind to their target RNAs. The interaction between ASOs and diverse intracellular proteins impacts their intracellular location, including intracellular trafficking, endosomal escape, and transport to the nucleus, affecting their pharmacological effects, such as potency and toxicity.

PS-ASOs interact with more than 50 intracellular proteins, such as P54nrb, which have RNA-recognition motifs, as well as chaperone proteins, such as HSP90, which lack RNA- or DNA-binding domains, and other proteins (Liang et al., 2015, 2016a). Some proteins, including La/SSB, NPM1, ANXA2, VARS, and PC4, appear to enhance ASO activity, likely through mechanisms related to subcellular distribution. VARS and ANXA2, which are co-localized with ASOs in endocytic organelles, may facilitate ASO release from the endosomes. La and NPM1 may play a role in ASO nuclear accumulation. In contrast, Ku70 and Ku80 proteins inhibit ASO activity, most likely by competing with RNase H1 for ASO/RNA duplex binding. Thus, PS-ASOs bind a set of cellular proteins that affect ASO activity via different mechanisms.

Endosomal escape is the rate-limiting step for ASO action. Several studies have suggested that intermediate endosomal compartments, such as multivesicular bodies and late endosomes, are key sites for productive ASO release into the cytosol (Crooke et al., 2017b; Juliano, 2018). The detailed mechanism has not yet been fully unraveled, but recent studies have revealed that proteins, such as M6PR (Liang et al., 2019a), and COPII vesicles (Liang et al., 2018) may play key roles in this process. No chemical modifications or conjugates for improving endosomal escape of ASOs have been discovered yet, although some small-molecular additives (Yang et al., 2015; Wang et al., 2017; Juliano et al., 2018), conjugation with synthetic endosomal escape domains (Lönn et al., 2016), and diverse cell-penetrating peptides (Kondow-McConaghy et al., 2020) have been evaluated. As it is estimated that the endosomal escape rate is quite low (<1%), ASO potency would be dramatically improved if the endosomal escape issue were to be resolved.

The toxicity of ASOs can be classified into sequence-dependent and sequence-independent toxicity. Sequence-dependent toxicity stems from the hybridization of ASOs with unintended targets, such as long pre-mRNA transcripts (Burel et al., 2016) or long introns (Kamola et al., 2015); however, it occurs less frequently than once thought (Lima et al., 2014). Sequence-independent toxicity stems from the interaction of ASOs with several proteins, such as P54nrb and RNase H1 (Vickers et al., 2019), which have been suggested to be key proteins. Toxic ASOs bind to paraspeckle proteins and RNase H1 to mislocalize them to the nucleolus and inhibit nucleolar RNA transcription and processing, which leads to nucleolar stress and apoptosis (Shen et al., 2019). The hepatotoxicity of ASOs may originate from sequence motifs, such as TCC/TGC, that typically bind to hepatocellular proteins, increasing the activity of the P53 and nuclear factor erythroid-2-related factor 2 stress pathways (Burdick et al., 2014).

MOE gapmer ASOs

Integrated safety assessment of MOE gapmer ASOs for kidney and liver function, hematology, and complement activation in non-human primates and humans has revealed that this chemical class is well tolerated in humans compared to the toxicities, such as complement activation and effects on platelets, found in non-human primates (Crooke et al., 2016). Although systemically administered MOE gapmer ASOs are distributed mainly to the kidneys and metabolites are cleared in the urine, no marked renal toxicity has been observed in numerous clinical trials using up to >175 mg/week dose and 52-week treatment (Crooke et al., 2018). After 4.6 years of treatment with up to 624 mg/week, MOE gapmer ASOs had no negative impact on platelet count (Crooke et al., 2017a).

Compared to PS-ODNs, MOE gapmers are substantially less immunostimulatory. In addition, immunotoxicity is dose-dependent; therefore, more potent MOE or cEt gapmer ASOs for which the therapeutic dose is lower have a lower risk of immunotoxicity. The CpG motif stimulates the innate immune response; however, 5-methyl cytosine replacement reduced this side effect.

A few MOE gapmers, including ISIS 147420, has been known to induce severe inflammatory response in mice. The adverse effects are characterized by the induction of interferon-β, followed by acute transaminitis and extensive hepatocyte apoptosis and necrosis (Burel et al., 2012). A specific sequence motif in viral DNA or RNA may trigger this acute innate immune response; thus, a single-nucleotide change can abolish this toxicity.

LNA/cEt gapmer ASOs

As mentioned above, LNA gapmers have higher risk of hepatotoxicity than MOE or cEt gapmers. The hepatotoxicity depends on a number of parameters, such as their length, gapmer design, and sequence. Shorter LNA gapmers have lower hepatotoxicity (Seth et al., 2009a). Regardless of 2′ modifications, toxic ASOs are able to bind to paraspeckle proteins, such as P54nrb and PSF, as well as RNase H1, delocalizing them to the nucleoli, which leads to nucleolar stress, P53 activation, and apoptosis (Shen et al., 2019). It has been reported that more toxic ASOs bind more proteins than non-toxic ones, and 5′-MOE-DNA-3′-cEt gapmer (mixmer) is less toxic than cEt gapmer or 5′-cEt-DNA-3′-MOE gapmer (mixmer), which implies that the 5′-side and more hydrophobic residues are preferred for protein binding. The introduction of a 2′-OMe at gap position 2 (Fig. 4) mitigated cytotoxicity by suppressing protein binding. Several chemical modifications with the aim of reducing toxicity and improving the therapeutic potential of ASOs are discussed in more detail below.

Figure 4. Site-specific modifications to mitigate toxicity. MOP, methoxy propyl phosphonate; MsPA, mesyl phosphoramidate.

As LNA and cEt gapmers with high affinity and potency have the issue of hepatotoxicity, diverse chemical modifications, including site-specific ones, have been developed, usually focused on the 5′ side of the gap, to mitigate toxicity (Fig. 4).

Nucleobase modifications

Recently, Yoshida et al. (2022) reported that some nucleobase modifications, such as 5-hydroxycytosine, 8-bromo, and amine guanine (Fig. 1), reduce the hepatotoxicity of LNA gapmers.

Linkage modifications

Alkylphosphonates: While the methyl phosphonate linkage (Fig. 1) was introduced long ago (Miller et al., 1979) and despite its advantages, such as having a neutral charge and nuclease resistance, it was not widely used due to RNase H inactivity, a relatively low affinity for RNA, and chemical instability under the basic condition required for the cleavage/deprotection step of oligonucleotide synthesis on a solid support. Recently, site-specific replacement of PS with alkyl phosphonates, such as methyl phosphonate or methoxy propyl phosphonate (Fig. 1), was reported to reduce toxicity by modulating the interactions with intracellular proteins (Migawa et al., 2019). For example, a 3-10-3 cEt gapmer (Fig. 3) in which one or two PS linkages at position 2 or 3 from the 5′-gap end were replaced with methyl phosphonate or methoxy propyl phosphonate (Fig. 4) showed comparable potency and substantially reduced hepatotoxicity, resulting in a higher therapeutic index than that of the parent ASO. The reduction in toxicity was accompanied by the absence of nucleolar mislocalization of paraspeckle protein P54nrb, ablation of P21 mRNA elevation, and caspase activation, which are thought to be key mechanisms of hepatotoxicity of PS-ASOs.

Mesyl phosphoramidate (MsPA): Oligonucleotide derivatives with an MsPA (Fig. 1) linkage, the synthetic method of which was recently developed (Chelobanov et al., 2017), have been demonstrated to have RNase H-activating ability, nuclease resistance, and anti-miRNA activity (Miroshnichenko et al., 2019). Their superior in vivo efficacy over PS (Patutina et al., 2020) and potential application in splice-switching (Hammond et al., 2021) have been demonstrated. A comprehensive study by (Anderson et al., 2021) on the impact of MsPA linkages on therapeutic properties showed that replacing up to 5 PS linkages in the gap with MsPA (Fig. 4) was well tolerated, and replacing specific PS linkages at appropriate locations greatly improved nuclease stability and potency and reduced both immune stimulation and cytotoxicity, leading to an improved therapeutic index, reduced pro-inflammatory effects, and extended effect duration.

Amides: Following the early attempts to replace PO with amide (1 in Fig. 5) (De Mesmaeker et al., 1994), the application of this neutral linkage to ASOs had not extensively been studied. However, recent systemic studies have investigated its impact on the pharmacological properties of an siRNA (Mutisya et al., 2017; Kotikam and Rozners, 2020) and an ASO (Epple et al., 2020). It was revealed that the amide linkage has no RNase H activity, but it could replace PO linkages at each wing of a gapmer to maintain RNA binding affinity and RNase H activity. The amide linkages combined with LNA (2 in Fig. 5) enhanced potency and cellular uptake, and reduced toxicity (Baker et al., 2022). The exon-skipping activity of an 18-mer ASO containing 4 LNA-amide linkages and 14 2′-OMe with 13 PS linkages (3 in Fig. 5) was higher than that of an 18-mer ASO with 17 PS linkages with all 2′-OMe or 8 LNA+10 2′-OMe. More importantly, the activity gap substantially increased when they were transfected without lipofectamine, which implied that the 18-mer ASO containing 4 LNA-amide linkages had substantially better cellular uptake ability.

Figure 5. Amide linkages and ASOs containing them. 1. Amide linkage 2. LNA-Amide linkage 3. Combination of LNA-amide linkage and 2-OMe-PS linkage.

Synergy by combination

The combination of 2′-OMe at the 5′-gap 2 position and two MsPA linkages at the 5′ end of the gap or in the 3′-wing improved the RNase H1 cleavage rate, considerably reduced the binding of proteins involved in cytotoxicity, and extended the elimination half-live (Zhang et al., 2022).

During the last decades, advances in medicinal chemistry have led to the development of drug-like ASO drugs that are sufficiently potent to treat various rare diseases. Among many chemical modifications, PS linkage, 2′-modification with MOE or cEt, and PMO are the key achievements that have been successfully clinically applied. As numerous molecular mechanisms underlying the pharmacology, pharmacokinetics, and toxicology of ASOs, such as ASO-protein interactions, have been revealed, it is now possible to manipulate the chemical modification and design of ASOs to broaden their therapeutic margin by fine-tuning or combining available chemistries based on structure-activity relationships. For example, site-specific modifications with 2′-OMe, MsPA, or an amide linkage are efficient in reducing toxicity while maintaining potency.

For ASOs to be more widely used for common diseases as well, two key challenges remain to be resolved. The first is tissue/cell-specific delivery to organs other than the liver. Besides GalNAc ligand conjugates targeting the asialoglycoprotein receptor on hepatocytes, which have been approved for clinical uses, another specific and efficacious ligand-receptor system should be discovered for targeting another tissues or cells. The second is chemical modification or conjugation to enhance cellular uptake and endosomal escape for improved potency. Some neutral linkers, such as amides that reduce the negative charge of PS-ASOs, are promising for this purpose. They are important because they will enhance potency and safety and lower the therapeutic dose, and as a result, reduce the cost and the risk of dose-dependent toxicity.

The author is the founder and CEO of Qmine Co., Ltd.