Tumor treatment is another application of LNPs. The discontinuous and fenestrated endothelium in rapidly growing malignant tumors and the poor lymphatic system allow LNPs and internalized siRNAs greater accumulation and slower clearance in solid tumors, referred to as the enhanced permeation and retention (EPR) effect.161 The EPR effect has become a mainstay of antitumor nanodrug delivery. However, increasing evidence has shown that LNPs are still primarily enriched in hepatocytes; hence, it is difficult to achieve complete remission merely based on the EPR effect for cancer treatment. An advanced design adds active tumor-targeting agents on the surfaces of LNPs that can target (i) the tumor cell surface, (ii) the tumor extracellular matrix and (iii) the endothelial cell surface receptors of tumor vessels by biological or chemical means.162 Tumor-targeting moieties can be selected from antibodies or proteins (e.g., transferrin163,164), peptides (e.g., RGD165 or octreotide166), aptamers or small molecules (e.g., hyaluronan167,168,169 or folate170). These ligands, to some extent, facilitate siRNA-loaded LNP accumulation in tumor tissues.
After more than 10 years of technological evolution, some theories have been established to guide the rational design of novel artificial lipids. Manoharan et al.107 showed that the pKa of lipids or lipidoids is of great importance for delivering siRNA to hepatocytes in vivo. In addition, these researchers elucidated the influences of the PEG-lipid content of siRNA-LNPs on delivery, concluding that PEG-C14 showed the highest delivery efficiency compared with PEG-C16 and PEG-C18, and the incorporation of 1.5 mol% PEG-lipid exhibited ideal potency, as higher proportions of PEG-lipid may decrease the potency.171 Moreover, Anderson and colleagues172 identified four essential structural and pKa criteria for the design of degradable lipidoids: (i) tertiary amine, (ii) O13 tail, (iii) more than 2 hydrophobic tails and (iv) pKa ≥ 5.5. This finding is of importance for designing and developing liver-targeted delivery systems based on lipid-like materials.
In addition, Arcturus Therapeutics,173 Dicerna Pharmaceuticals,174 Silence Therapeutics,175,176 Sirna Therapeutics (acquired by Alnylam from Merck & Co., Inc. in 2014)177,178 and Suzhou Ribo Life Science142 also made great efforts to develop lipid-based siRNA delivery systems, some of which have been advanced to the clinical stage.
Lipid-employed siRNA therapeutics
ONPATTRO®78 is a chemically modified anti-transthyretin (TTR) siRNA formulated in liposomes (Figs. 5a, h and 7). The lipid components of ONPATTRO® include DLin-MC3-DMA (Fig. 6), DSPC, cholesterol and PEG-DMG ((R)-2,3-bis(octadecyloxy)propyl-1-(methoxy polyethylene glycol 2000) carbamate) at a molar ratio of 50/10/38.5/1.5.107 The active pharmaceutical ingredient (API, siRNA) of ONPATTRO® is modified with eleven 2′-OMe (nine in the sense strand and two in the antisense strand) and four 2′-deoxy thymidine (two in the sense strand and two in the antisense strand, all at the 3′-end) moieties. Before ONPATTRO®, Alnylam advanced ALN-TTR01 to a phase 1 clinical study (NCT01148953). The API of ALN-TTR01 is the same as that of ONPATTRO®, whereas the delivery system for ALN-TTR01 comprises Dlin-DMA,148,179 cholesterol, DSPC and PEG-DMG. A phase 1 study of ONPATTRO® (NCT01617967) showed a good dose-dependent pattern for TTR protein reduction.180 Overall, this drug was well tolerated by subjects, although pretreatment with antihistamines, nonsteroidal antihistamines or glucocorticoids was needed to reduce the incidence of infusion-related reactions. Phase 2 and 3 studies reached both the primary and second endpoints of the studies, according to the outcomes regarding the modified Neuropathy Impairment Score +7 (mNIS+7) and the Norfolk Quality of Life Diabetic Neuropathy (QoL-DN) score. Based on these achievements, the FDA and the EC approved the commercial use of ONPATTRO® in
Download PDF
Review Article
Open Access
Published: 19 June 2020
Therapeutic siRNA: state of the art
Bo Hu,
Liping Zhong,
[…]
Xing-Jie Liang
Signal Transduction and Targeted Therapy volume 5, Article number: 101 (2020) Cite this article
11k Accesses
14 Citations
22 Altmetric
Metricsdetails
ABSTRACT
RNA interference (RNAi) is an ancient biological mechanism used to defend against external invasion. It theoretically can silence any disease-related genes in a sequence-specific manner, making small interfering RNA (siRNA) a promising therapeutic modality. After a two-decade journey from its discovery, two approvals of siRNA therapeutics, ONPATTRO® (patisiran) and GIVLAARI™ (givosiran), have been achieved by Alnylam Pharmaceuticals. Reviewing the long-term pharmaceutical history of human beings, siRNA therapy currently has set up an extraordinary milestone, as it has already changed and will continue to change the treatment and management of human diseases. It can be administered quarterly, even twice-yearly, to achieve therapeutic effects, which is not the case for small molecules and antibodies. The drug development process was extremely hard, aiming to surmount complex obstacles, such as how to efficiently and safely deliver siRNAs to desired tissues and cells and how to enhance the performance of siRNAs with respect to their activity, stability, specificity and potential off-target effects. In this review, the evolution of siRNA chemical modifications and their biomedical performance are comprehensively reviewed. All clinically explored and commercialized siRNA delivery platforms, including the GalNAc (N-acetylgalactosamine)–siRNA conjugate, and their fundamental design principles are thoroughly discussed. The latest progress in siRNA therapeutic development is also summarized. This review provides a comprehensive view and roadmap for general readers working in the field.
Download PDF
Introduction
Gene therapy is a promising therapeutic platform because it targets disease-causing genes in a sequence-specific manner, which enables more precise and personalized treatment of diverse life-threatening diseases.1 By introducing a certain nucleic acid modality to the desired tissue of the patient, gene expression can be downregulated, augmented or corrected. Small interfering RNA (siRNA), microRNA (miRNA) and inhibitory antisense oligonucleotides (ASOs) are representative molecules used to trigger gene inhibition, whereas plasmid DNA, messenger RNA (mRNA), small activating RNA (saRNA), splicing-modulatory ASOs and CRISPR (clustered regularly interspaced short palindromic repeats)/Cas (CRISPR-associated protein) systems are usually employed to increase or correct target gene expression.2,3,4 Currently, many therapeutic programs have been explored to treat certain diseases.
RNA interference (RNAi) is a natural defense mechanism for the invasion of exogenous genes.5,6 RNAi modalities, e.g., siRNA and miRNA, can knockdown the expression of target genes in a sequence-specific way (Fig. 1) by mediating targeted mRNA degradation (for siRNA and miRNA) or mRNA translation repression (for miRNA). As a result of the slight differences between siRNA and miRNA, siRNA can typically trigger more efficient and specific gene silencing than miRNA, whereas one miRNA may compromise the expression of several different target genes simultaneously. Hence, siRNA and miRNA have different roles in pharmaceutical practice.
Fig. 1

Schematic illustrations of the working mechanisms of miRNA (a) and siRNA (b)
Full size image
Since the establishment of the RNAi concept in 1998, siRNA therapeutics have experienced many ups and downs. In 2001, Tuschl et al.7 successfully silenced the expression of a specific gene by introducing chemically synthesized siRNA into mammalian cells, leading to the emergence of a developmental upsurge. Although siRNA therapy once suffered due to the obstacles of its stability, specificity and delivery, advances in chemical modification and delivery brought the field to a robust and rapidly developing area of research again in recent years. After a 20-year journey, the United States Food and Drug Administration (FDA) and the European Commission (EC) approved ONPATTRO® (patisiran, ALN-TTR02) as the first commercial RNAi-based therapeutic for the treatment of hereditary amyloidogenic transthyretin (hATTR) amyloidosis with polyneuropathy in adults in 2018.2,8 Recently, the FDA-approved GIVLAARI™ (givosiran, ALN-AS1) for the treatment of adults with acute hepatic porphyria (AHP).9,10,11,12
siRNA has innate advantages over small molecular therapeutics and monoclonal antibody drugs because siRNA executes its function by complete Watson–Crick base pairing with mRNA, whereas small molecule and monoclonal antibody drugs need to recognize the complicated spatial conformation of certain proteins. As a result, there are many diseases that are not treatable by small molecule and monoclonal antibody drugs since a target molecule with high activity, affinity and specificity cannot be identified. In contrast, theoretically, any gene of interest can be targeted by siRNA since only the right nucleotide sequence along the targeting mRNA needs to be selected. This advantage confers the siRNA modality with a shorter research and development span and a wider therapeutic area than small molecule or antibody drugs, especially for those genes that are unfeasible for development with such strategies.
Although siRNA holds promising prospects in drug development, several intracellular and extracellular barriers limit its extensive clinical application. Naked and unmodified siRNA possesses some disadvantages, such as (1) unsatisfactory stability and poor pharmacokinetic behavior and (2) the possible induction of off-target effects. The phosphodiester bond of siRNA is vulnerable to RNases and phosphatases. Once it is systematically administered into circulation, endonucleases or exonucleases throughout the body will quickly degrade siRNA into fragments, thus preventing the accumulation of intact therapeutic siRNA in the intended tissue. In theory, siRNA only functions when its antisense strand is completely base-paired to the target mRNA. However, a few mismatches are tolerated by the RNA-induced silencing complex (RISC), which may lead to undesired silencing of those genes with a few nucleotide mismatches. In addition, the RISC-loaded sense strand of siRNA may also knockdown the expression of other irrelevant genes. Moreover, unformulated and unmodified siRNA may lead to the activation of Toll-like receptor 3 (TLR3) and adversely affect the blood and lymphatic systems.13 These discoveries have raised many concerns about the undesirable effects and pharmaceutical issues of siRNAs.
To maximize the treatment potency and reduce or avoid the side effects of siRNA, researchers have made great efforts to investigate various chemical modification geometries and to develop many different delivery systems. As a result, a series of modification patterns were proposed and evaluated preclinically and clinically with respect to their effects on activity, stability, specificity and biosafety. Delivery materials derived from lipids, lipid-like materials (lipidoids), polymers, peptides, exosomes, inorganic nanoparticles, etc., have been designed and investigated.2,8,14,15,16,17,18,19,20,21,22 As a result, several modification patterns and delivery platforms have been employed in clinical studies. Here, the detailed evolution and advances in the modification and delivery technologies of siRNA are comprehensively summarized and discussed. This review provides an overview and a handbook for reviewing siRNA therapeutic development.
siRNA modification
During the early stages of developing siRNA therapeutics, many agents were designed based on completely unmodified or slightly modified siRNA to arrive at appropriate tissue and then silence the target gene. These molecules can mediate gene silencing in vivo, especially in tissues that receive local drug administration, e.g., eyes. However, limited efficacy and potential off-target effects may be observed with these modalities. For example, Kleinman and colleagues23 observed that bevasiranib and AGN211745 triggered significant activation of toll-like receptor 3 (TLR3) and its adapter molecule TRIF, inducing the secretion of interleukin-12 and interferon-γ. Bevasiranib and AGN211745 are siRNA therapeutics developed for the treatment of age-related macular degeneration.24 The VEGFA-targeting bevasiranib25 and VEGFR1-targeting AGN21174526 are unmodified and slightly modified siRNAs, respectively. Moreover, Kleinman and colleagues further demonstrated that siRNA classes with sequences of 21 nucleotides or longer, regardless of which genes they target, can suppress CNV in mice compared to bevasiranib and AGN211745. These findings eventually led to the termination of the clinical investigation of bevasiranib in 2009.
Chemically modified siRNAs, such as siRNAs with substitution of the 2′-OH with a 2′-O-methyl (2′-OMe)27 or 2′-methoxyethyl (2′-MOE)28 group or the substitution of certain nucleotides with locked nucleic acid (LNA),29 unlocked nucleic acid (UNA)30 or glycol nucleic acid (GNA)31 (Fig. 2), can efficiently suppress immunostimulatory siRNA-driven innate immune activation, enhance activity and specificity, and reduce off-target-induced toxicity. To enhance the potency and reduce the potential toxicity of siRNA, numerous chemical modification geometries have been established and tested.32
Fig. 2

Structures of chemical modifications and analogs used for siRNA and ASO decoration. According to the modification site in the nucleotide acid, these structures can be divided into three classes: phosphonate modification, ribose modification and base modification, which are marked in red, purple and blue, respectively. R = H or OH, for RNA or DNA, respectively. (S)-cEt-BNA (S)-constrained ethyl bicyclic nucleic acid, PMO phosphorodiamidate morpholino oligomer
Full size image
According to the natural structure of nucleotides, chemical modifications can be placed at the phosphate backbone, the ribose moiety or the base. Typically, these modifications are simultaneously introduced in siRNA. For example, the combination of 2′-OMe and phosphorothioate (PS) modifications facilitates the systemic administration of cholesterol-conjugated siRNA and achieves efficient gene silencing in vivo.33 In addition, the combination of 2′-OMe and 2′-deoxy-2′-fluoro (2′-F) has been used for ONPATTRO®34,35 (Figs. 2, 3). For QPI-1007, a naked and unformulated siRNA therapeutic,2 2′-OMe was alternatively incorporated into the antisense strand, and an inverted deoxyabasic moiety and an L-DNA cytidine nucleotide were used in the sense strand36 (Figs. 2,3). Moreover, modifications with phosphorothioate (PS), 2′-OMe, 2′-F and 2′-deoxy were employed for inclisiran (ALN-PCSsc), an siRNA therapeutic for the treatment of hypercholesterolemia37,38 (Figs. 2, 3). Overall, the precise modification of siRNA can increase its efficacy, specificity, and stability and reduce its toxicity and immunogenicity. Detailed information will be discussed in the following sections.
Fig. 3

Representative designs for the chemical modification of siRNA. The sequences and modification details for ONPATTRO®, QPI-1007, GIVLAARI™ and inclisiran are included. The representative siRNA modification patterns developed by Alnylam (STC, ESC, advanced ESC and ESC+) and arrowhead (AD1-3 and AD5) are shown. Dicerna developed four GalNAc moieties that can be positioned at the unpaired G–A–A–A nucleotides of the DsiRNA structure. 2′-OMe 2′-methoxy, 2′-F 2′-fluoro, GNA glycol nucleic acid, UNA unlocked nucleic acid, SS sense strand, AS antisense strand
Full size image
Phosphonate modification
Phosphorothioate (PS) linkage has previously been used in antisense oligonucleotide (ASO) modification.39 This modification was achieved by leveraging a sulfur atom to replace one nonbridging oxygen of a phosphodiester (Fig. 2). PS linkage endows modified oligonucleotides with resistance to nucleases,40 and these molecules more readily combine with plasma proteins,41 especially albumin42 than their unmodified counterparts, which may result in a longer circulation time. Although some data have shown that PS modification reduces the binding affinity between the oligonucleotide and its target sequence32 to some extent and aggravates chemistry-related toxicities with a high PS content,43,44 PS modification is still very important and necessary for both siRNA and ASO because, compared with unmodified oligonucleotides, PS-modified oligonucleotides are more hydrophobic and stable molecules with higher affinity to certain proteins.44 Appropriate protein binding may prolong the half-life of oligonucleotides in circulation, and the half-life of the elimination phases may be as long as a few days.41 Protein binding is typically beneficial for the cell entry of oligonucleotides; however, too much protein binding showed a positive correlation with in vivo toxicity.45,46 Hence, the position and number of PS linkages are pivotal for their application. For siRNA modification, Alnylam introduced two PS linkages at the first two nucleotides at the 5′-end of the sense strand and two PS linkages at the first two nucleotides at both the 5′-end and the 3′-end of the antisense strand of siRNA (Fig. 3). Moreover, compared to the parent siRNA without modifications, the PS linkage-containing siRNA has a barely changed in vivo biodistribution profile, as it predominantly accumulates in clearance organs, e.g., liver, kidneys and intestines, followed by glandular tissues, bone marrow, adipocytes and lymph nodes.47,48,49,50
Notably, there are two configurations for PS linkages, namely, Rp and Sp isomers (Fig. 2). Verdine and colleagues (team from Wave Life Sciences)51 demonstrated that the stereochemistry of PS significantly influences the pharmacologic performance of ASOs. PS linkages with the Sp configuration are more stable than those with the Rp configuration. More importantly, they unveiled that stereopure ASOs with a triplet stereochemical code, 3′-SpSpRp, make the target RNA more vulnerable to RNase H1 in vitro and achieve a more durable inhibition profile in vivo than stereorandom ASOs. However, Østergaard et al. (team from Ionis Pharmaceuticals)52 disclosed that controlling PS chirality markedly affects the interactions between the ASOs and RNase H1; however, this modulation failed to enhance the overall therapeutic effects. The sequence and design of ASOs are the primary drivers that determine the performances of gapmer ASOs. In addition, recently disclosed patents filed by Alnylam reported that the chirality of PS linkages may markedly influence siRNA performance.53 The terminal chiral PS linkages at the 3′- and 5′-ends of the antisense strand of siRNA prefer Sp and Rp configurations, respectively. The chiral PS number can be 1, 2 or 3 at the 3′-end and 1 or 2 at the 5′-end of the siRNA antisense strand. In addition, the chiral PS linkages at the 5′-end of the siRNA sense strand can adopt either an Sp or an Rp configuration and are positioned at the first internucleotide linkage at the 5′-end of the sense strand. Overall, the chirally modified siRNA can comprise four, five or more terminal, chirally modified PS linkages.
In addition, other residues have been identified and successfully used to replace the phosphodiester group in oligonucleotides and change the properties of intact strands, including phosphorodithioate (PS2),54 methylphosphonate (MP), methoxypropylphosphonate (MOP) and peptide nucleic acid (PNA) (Fig. 2). PS2 modification can increase the affinity between RISC and siRNA. The site-specific incorporation of MP and MOP modifications has been used to reduce ASO protein binding since the PS backbone is negatively charged, whereas alkylphosphonate linkages are charge neutral. As a result, the MOP linkage incorporated at position 2 or 3 from the 5′-end of the DNA gap significantly mitigated the hepatotoxicity of ASOs.46 In addition, both PNA55,56,57 (Fig. 2) and phosphotriesters58 are very meaningful modifications to make siRNA or ASO molecules more druggable, although these modifications are not as predominant as PS. PNA is also typically used to modify detection probes for capturing and detecting certain nucleic acid targets.59
In addition, phosphonate modification with various analogs at the 5′-end of siRNA or single-stranded siRNA (ss-siRNA) is a newly developed but important decoration strategy to enhance siRNA activity. The 5′-phosphate of exogenously introduced siRNA is required for RISC loading. A 5′-phosphate can be modified via either phosphorylation in cells mediated by Clp1 (cleavage and polyadenylation factor I subunit 1) or chemical synthesis in the laboratory.60 However, the natural 5′-phosphate can also be removed by dephosphorylation in cells. Together with the demand to enhance siRNA stability, researchers have identified a series of analogs that have similar conformation and steroidal electronic properties to natural phosphates but are resistant to dephosphorylases. As a result, 5′-(E)-vinyl phosphonate (5′-(E)-VP), 5′-methylphosphonate, (S)-5′-C-methyl with phosphate, 5′-phosphorothioate, etc., were designed and evaluated in vitro and in vivo61 (Fig. 2). Among these strategies, 5′-(E)-VP decoration, which substitutes bridge oxygen and carbon with E-vinyl phosphonate moieties at the 5′-end, is the most potent and metabolically stable mechanism. The replaced hydrophobic substitutions enhance the stability of intact oligonucleotides and are favorable for RISC loading. Ionis Pharmaceuticals first developed and own the intellectual property rights to the 5′-(E)-VP structure. This technology was first applied to ss-siRNA and achieved ideal pharmacodynamic and pharmacokinetic profiles. Modifying ss-siRNA with 5′-(E)-VP and other specific modification structures enables potent repression of targeted gene expression by ss-siRNA in vivo.61,62 Inspired by this finding, the 5′-(E)-VP modification was further applied to double-stranded siRNA. Data have revealed that this modification improves siRNA accumulation and residence time in tissue and enhances siRNA potency in vivo by elevating Argonaute-2 binding.63,64,65
Ribose modification
Ribose modifications at the 2′ position have been widely used to protect siRNA from attacking ribonucleases, which require the 2′-OH group for the hydrolysis of RNA. Notably, 2′-O-methyl (2′-OMe) (Fig. 2) is a naturally occurring ribosugar and the most frequently used modification in drug development so far66 because this molecule can enhance stability67 by blocking the nucleophilic 2′-OH group. In addition, 2′-OMe also shows increased affinity for its target mRNA68 and reduced immunogenicity in the body. Based on 2′-OMe, a series of analogs have been identified, among which 2′-O-methoxyethyl (2′-O-MOE) (Fig. 2) is one of the most useful and popular analogs of 2′-OMe. Moreover, 2′-O-MOE shows higher binding affinity for RNA than 2′-OMe with a 0.9–1.7 °C change in ΔTm per modified nucleotide, which further increases the capability of the modified siRNA to resist nuclease attack.32 Another widely used 2′-OH substitution is 2′-deoxy-2′-fluoro (2′-F)69 (Fig. 2). Highly electronegative fluorine makes the modified siRNA readily adopt a C3′-endo conformation, providing considerable benefits in binding affinity, with a ΔTm of ~2.5 °C per modified nucleotide.32 Interestingly, 2′-arabino-fluoro (2′-Ara-F), an alternative fluoro substitution form, can also be used to modify nucleic acid therapeutics, e.g., antisense oligonucleotides45,70 and siRNA.71 Both 2′-O-benzyl and 2′-O-methyl-4-pyridine (2′-O-CH2Py(4)) were well tolerated when they were incorporated on the guide strand of siRNA in vivo72 (Fig. 2). siRNA with modifications of four 2′-O-benzyl or six 2′-O-CH2Py(4) showed comparable activity with the unmodified siRNA in vivo, and increased activity was even achieved when these modifications were placed at the 8 and 15 positions on the siRNA guide strand.
In addition, 2′-C, 4′-C, and even the whole sugar ring can also undergo modifications, resulting in molecules including UNAs, LNAs, GNAs, (S)-cEt-BNAs, tricyclo-DNA (tcDNA) and phosphorodiamidate morpholino oligomers (PMOs) (Fig. 2). UNA, with higher flexibility and thermal destabilization than the unmodified product due to unconnected 2′ and 3′ carbons, can block the entry of passenger strands and promote the RISC loading of the guide strand by introducing chemical asymmetry into duplex siRNAs. Similarly, glycol nucleic acid (GNA),31 another thermally destabilizing nucleotide, can be used to erase off-target effect-induced hepatotoxicity by including it in the seed region of the siRNA guide strand. LNA is a bicyclic structure that contains a bridge between the 2′ oxygen and the 4′ carbon. It "locks" the ribose into its preferred C3′-endo conformation and significantly increases the affinity of base pairing.40 Data have shown that the incorporation of LNAs increases the DNA melting temperature up to 8 °C per LNA.73 Given the successes of LNAs, many more bicyclic and even tricyclic analogs, including ethyl-bridged nucleic acids (ENAs),74 constrained ethyl (cEt) nucleic acids75 and tricyclo-DNA,76 have been engineered and incorporated into RNA or DNA strands in succession. Moreover, PMOs77 do not look like classic nucleotides, as the ribose subunit has been substituted with a morpholine subunit. PMOs are uncharged at physiological pH and are not substrates of RNase H. Therefore, they are mainly used to block RNA splicing or translation.78
Base modification
Base replacement is of great benefit to nucleic acid-based drug development. For instance, the substitution of pseudouridine,79 2-thiouridine,80 N6-methyladenosine,80 5-methylcytidine81 (Fig. 2) or other base analogs of uridine and cytidine residues can reduce innate immune recognition while making ASOs more resistant to nucleases. However, the artificial base substitution of ASOs, similar to siRNA, is basically at the stage of research and development. Pharmaceutical corporations still hold a prudent attitude toward these molecules. Instead, these companies prefer to use naturally occurring base structures, e.g., 5mC and 6 mA, to modify certain base(s), probably because of concerns about the safety of the metabolized unnatural residues that potentially might be incorporated into the genome.
N-ethylpiperidine triazole-modified adenosine analogs with Hoogsteen or Watson–Crick (WC) facial localization have also been used to modify siRNA,82,83,84,85 which can disrupt nucleotide/TLR8 interactions and therefore reduce the immunogenicity of siRNA. N-ethylpiperidine 7-EAA triazole (7-EAA, 7-ethynyl-8-aza-7-deazaadenosine) can pair well with uridine and form an A-form helix structure. This molecule will be recognized as adenosine by Avian myeloblastosis virus reverse transcriptase in RNA strands.83
In addition, 6′-phenylpyrrolocytosine (PhpC) is a cytosine mimic showing excellent base pairing fidelity, thermal stability and high fluorescence.86 PhpC-containing siRNA shows gene-silencing activity comparable to that of the parent molecule, and its fluorescent properties make it useful for fluorescence-based detection or monitoring or for exploring the cellular uptake and trafficking of siRNA in cells.
The internal uridine substitution of 2,4-difluorotolylribonucleoside (rF) is well tolerated by siRNA activity.87 The base pairing between rF and adenosine (A) is relatively unstable compared to the uridine–adenosine pair because the fluorine atom is more hydrophobic than uridine and cannot serve as a hydrogen bond acceptor.
Zhang and colleagues88 reported that a 5-nitroindole modification at position 15 of the siRNA passenger strand greatly reduced the activity of the passenger strand, whereas the effectiveness of the siRNA guide strand was barely affected. This modification provided a practical strategy to reduce the passenger strand-mediated off-target effects.
siRNAs containing 5-fluoro-2′-deoxyuridine (FdU) moieties can suppress targeted gene expression. The effects depend on the locations of the FdU substitution. In addition, these modified siRNAs quickly release FdU after entering the cell, thereby further inducing a variety of DNA-damage repair and apoptosis pathways and ultimately triggering cell death. These findings provide a strategy for siRNA-based cancer therapy.89
Moreover, many other less common base analogs have also been used in siRNA modifications. Their application helps researchers better understand the mechanism of gene silencing and helps develop new methods to mitigate off-target effects caused by harmful protein binding or undesired targeting of mRNA.90
Modification patterns used in clinical studies or for commercial siRNA
Early siRNA therapeutic programs typically employ partial or slight modifications by using 2′-OMe, 2′-F or PS,91,92 e.g., alternative 2′-OMe modifications were placed in the antisense strand of QPI-1007, whereas just one L-DNA was used in the sense strand. After making great efforts to understand the effects of various modifications on siRNA activity, specificity and toxicity, researchers have realized that heavy modification of siRNA, even full-length embellishment, probably do not affect siRNA activity. Furthermore, these judicious modification strategies can significantly enhance siRNA stability or biocompatibility. As a result, several different, interrelated modification patterns have been proposed and validated preclinically and clinically.
Standard template chemistry (STC) (Fig. 3) is a universal modification pattern established by Alnylam Pharmaceuticals. When the lengths of sense and antisense siRNA are 21 and 23 nt, respectively (referred to as "S21nt" and "AS23nt", respectively, in the following description), 2 PS linkages are placed at the 3′ terminus of the antisense strand, three consecutive 2′-F moieties are incorporated at positions 9, 10 and 11 of the sense strand from the 5′-end, and consecutive 2′-OMe modifications are placed at positions 11, 12 and 13 of the antisense strand from the 5′-end. Moreover, alternative 2′-OMe and 2′-F moieties are employed at other positions in both strands. It is worth mentioning that 2′-OMe and 2′-F are complementarily used for all positions of both strands of siRNA (Fig. 3). It is considered that placing three consecutive 2′-F moieties, instead of larger groups, at the middle of the sense strand aims to facilitate the cleavage and removal of the sense strand in RISC, as release of the sense strand is the prerequisite of RISC activation. PS linkages and OMe positioned at the terminals are used to enhance the capability of siRNA to resist degrading enzyme attack.
The STC design (e.g., as used in revusiran) has been proven to remarkably enhance siRNA stability and affinity without compromising intrinsic RNAi activity.93 ALN-AT3, another STC-modified RNAi therapeutic, also showed potent and durable gene silencing after single subcutaneous (s.c.) administration.94 However, the clinical investigation of revusiran was discontinued due to unbalanced death between revusiran- and placebo-treated groups. Although the detailed investigation report by Alnylam illustrated that the deaths of revusiran-treated patients were not correlated with the test drug, there are still strong concerns regarding STC-modified siRNA therapeutics. Hence, the next generation of modification patterns, called enhanced stabilization chemistry (ESC) (Fig. 3), was proposed. This design includes four more PS linkages at the 5′-end of the antisense strand and the 3′-end of the sense strand. Another important change was the reduction in 2′-F substitutions, probably because heavy use of 2′-F may magnify toxicity.95 These changes markedly enhanced siRNA potency and duration compared with the changes inherent to the STC design.9,96,97 This advancement allowed us to achieve the desired pharmacodynamic effect at a significantly lower dose, with a markedly reduced dosing frequency.98 Cemdisiran (ALN-CC5) achieved potent gene silencing as long as more than 1 year after receipt of a single dose of the testing therapeutic in a phase 1/2 clinical study (Fig. 4f), representing an unprecedented milestone in pharmaceutical history. Moreover, the newly approved GIVLAARI™ is an siRNA therapeutic based on the ESC design (Fig. 3).99
Fig. 4

Preclinical and clinical performance of siRNAs with comprehensive modification chemistries. a Sense and antisense strand optimization by using transthyretin (Ttr)-targeting siRNAs and the combinatorial designs explored across 10 siRNAs. The effect relative to the parent is described as the model-adjusted mean difference in the activity of the design variant (DV) compared to the parent. b–d The liver exposure, RISC loading, and silencing activities of the parent, DV18, and fully 2′-OMe conjugates. Data are shown as the mean ± SD. e Gene-silencing duration of antithrombin-targeted siRNAs modified with the designs of patent and DV22 evaluated in Cynomolgus monkeys (n = 3 per group). Asterisks indicate that a significant difference was observed between the parent and DV22. Error is shown as the SD. f C5 knockdown profile after applying a single dose of Cemdisiran that employs ESC modification in a phase 1/2 clinical study. Error is represented as the SEM. a–e Copyright of Elsevier Inc., 2018. f Copyright of Alnylam Pharmaceuticals, 2016
Full size image
In addition, Alnylam comprehensively optimized the modification design variants (DVs) of the ESC strategy across multiple siRNAs with an iterative screening approach to further enhance stability without compromising intrinsic RNAi activity. As a result, advanced ESC designs, e.g., DV18 and DV22, were proposed (Figs. 3 and 4a–e).96 Compared to the former scheme, the advanced strategy maintains six PS linkages at three strand terminals; however, this modification scheme significantly reduces the proportion of 2′-F moieties. Only 10 and 8 2′-F substitutions are used in two strands for the advanced ESC designs of DV18 and DV22, respectively. For DV18, the 2′-F modifications are positioned at sites 7, 9, 10 and 11 in the sense strand and sites 2, 6, 8, 9, 14 and 16 in the antisense strand (all from the 5′-ends of the strands). Compared with DV18, DV22 has the 2′-F modifications at 8 and 9 in the antisense strand further replaced with 2′-OMe (Figs. 3 and 4a). Both the DV18 and DV22 designs can achieve significantly higher liver exposure and RISC loading and more potent and durable gene silencing than the parent design in preclinical species, including nonhuman primates (Fig. 4b–e). The underlying mechanisms were not fully unveiled in Alnylam's literature. However, Zheng et al.100 demonstrated that modification of the 14th position of the siRNA guide strand could eliminate its gene-silencing activity by reducing RISC loading and target degradation, and the larger the modification group used was, the higher the reduction efficiency. Song et al.28 further proved that decorations at positions 9 and 10 (for siRNA with a 19 nt/19 nt structure) markedly reduced the activity of the unmodified strand of siRNA without disturbing the potency of the modified strand. In light of this, the specificity and activity of siRNAs could be improved by introducing 2′-MOE at the cleavage site of the siRNA. These observations are in line with the rationales of advanced ESC designs.
Furthermore, investigators from Alnylam have disclosed that the hepatotoxicity of GalNAc (N-acetylgalactosamine)-siRNA conjugates is attributed to off-target gene silencing mediated by miRNA-like recognition between siRNA and a mistargeted RNA.31 Researchers have demonstrated that disorganizing the nucleotides of the seed region without changing the 2′-OMe, 2′-F or PS content or placing a GNA at position 7 of the siRNA antisense strand can markedly alleviate off-target effects and mitigate hepatotoxicity because these modifications affect the binding of siRNA with undesired target mRNA in a seed region-specific manner. The employment of GNA in the siRNA seed region is the main technical characteristic for the ESC+ design compared to advanced ESC designs (Fig. 3). Several investigational siRNA therapeutics of Alnylam, e.g., ALN-AAT02, ALN-HBV02, and ALN-AGT, utilize the ESC+ design as the modification pattern.
In addition to Alnylam, Arrowhead Pharmaceuticals (Arrowhead) proposed a series of new modification designs (Fig. 3) by combining the modification patterns (e.g., inverted bases) of siRNAs with 2′-OMe/2′-F decorations similar to the aforementioned STC and ESC modification patterns. Typically, the modification strategies proposed by Arrowhead are characterized by placing inverted deoxythymine (idT) at the strand terminus and/or termini, including UNA and X (without nucleoside base), flanking the UAU or UAUAU motif(s), conjugating the siRNA with cholesterol or other hydrophobic substrates, etc. These designs are currently being investigated in clinical trials.101 Dicerna Pharmaceuticals (Dicerna)102 employed their proprietary dicer-substrate siRNA (DsiRNA) technology to develop RNAi therapeutics. In addition, these researchers also established a series of DsiRNA modification chemistries. For instance, three consecutive 2′-F moieties may be placed at sites 9, 10, and 11 in the sense strand of siRNA from the 5′-end, and the other sites may use alternative 2′-OMe and 2′-F moieties in the flank sequence. One example of a constant flank sequence is 'GCAGCCGAAAGGUGC', which contains inner complementary pairing motifs of 'GCAGCC' and 'GGCUGC'. Consecutive 2′-OMe moieties may be used to modify these motifs, and consecutive RNA or DNA nucleosides without additional 2′ modifications may be employed for the bubble motif of GAAA. Moreover, DNA nucleosides can also be applied at sites 2, 12, 16, 18, 20 and 21 in the antisense strand from the 5′-end. PS linkages may be incorporated into the antisense strand at specific sites, as well as within the constant flank sequence (Fig. 3). The ligands may be placed at the bubble sequence. Silence Therapeutics103 (Fig. 3) and other RNAi-based biotech companies, such as Arbutus Biopharma, OliX Pharmaceuticals and Suzhou Ribo Life Science, have all been devoted to developing state-of-the-art siRNA modification chemistries to establish fruitful drug development pipelines.2
siRNA delivery and drug development
Barriers to siRNA delivery
RNAi modalities (siRNAs and miRNAs) need to employ proteins (such as Argonaute and Tar RNA binding protein) in the cytoplasm. A series of barriers need to be circumvented for systemically administered exogenous siRNA before it can achieve gene silencing,104,105,106 such as nuclease degradation and short-lived circulation, immune recognition in blood circulation, accumulation in desired tissue, effective transmembrane trafficking, and escape from endosomes and lysosomes to the cytoplasm. By incorporating chemical modifications, the challenges of stability against serum nucleases and avoiding immune recognition have been substantially reduced. However, other problems remain to be solved.
In blood circulation, unspecific binding and glomerular filtration both hamper the accumulation of siRNA in desired tissues. The neutral surface charge of siRNA-loaded nanoformulations is beneficial for avoiding unfavorable protein binding in circulation.107 However, protein decoration or protein corona formation is required for nanoformulation and attachment to specific tissues or cells for some delivery systems.108,109 For instance, ionizable lipid nanoparticles (iLNPs) primarily accumulate in the liver by interacting with serum lipoproteins, e.g., apolipoprotein E3 (ApoE3), which leads to specific recognition by the receptor, e.g., low-density lipoprotein receptor (LDLR).108,109 In contrast, ligand–siRNA conjugates can transport siRNA to desired tissues and cells by specific recognition and interactions between the ligand, e.g., a carbohydrate, peptide, antibody, aptamer, small molecule, etc., and the surface receptor. This active targeting strategy not only reduces siRNA accumulation in unintended tissues, thus decreasing or avoiding undesired side effects and toxicity, but also achieves potent gene silencing at a low dosage.104
siRNA is ~7–8 nm in length and 2–3 nm in diameter. This molecule is too large to cross cell membranes but small enough to be freely cleared by glomeruli,110 as molecules with a size smaller than 8 nm111 are easily filtered into the urine. Hence, once siRNAs leave the bloodstream, they will accumulate in the bladder and be excreted out from the body quickly, within a few minutes to half an hour,48,49,50 which prevents them from accumulating in targeted tissues or cells. Encapsulating siRNA into vesicles or conjugating it to certain ligands can effectively avoid renal clearance and, more importantly, can deliver siRNA to the desired tissues or cells. Lipid nanoparticles (LNPs),106 dynamic polyconjugates (DPC™), and GalNAc–siRNA conjugates have all achieved efficient siRNA delivery to hepatocytes (Figs. 5, 6).112
Fig. 5

Schemes of representative clinically studied siRNA formulations and their pharmacodynamic performance. a–g Schemes of a lipid nanoparticles, b DPC™ or EX-1™, c TRiM™, d GalNAc–siRNA conjugates, e LODER™, f iExososme and g GalXC™. h Efficacy of ONPATTRO® (a liposome formulation) in healthy volunteers, in which siRNA targeting PCSK9 is formulated in lipid nanoparticles.34 Copyright Massachusetts Medical Society, 2013. i Levels of urinary aminolevulinic acid (ALA) and porphobilinogen (PBG) after once-monthly and once quarterly doses of GIVLAARI™ in acute hepatic porphyria (AHP) patients. Copyright Alnylam Pharmaceuticals, 2019. j Reduction of plasma PCSK9 after a single increasing dose of inclisiran.38 Copyright Massachusetts Medical Society, 2017. k Serum HBsAg reduction in hepatitis B patients who received the treatment of a single dose of ARC-520 (1–4 mg/kg) (formulated with DPC2.0) on a background of daily oral NUCs.264 PBO, patients on NUC therapy receiving placebo injection. NUCs nucleos(t)ide viral replication inhibitors. The error is shown as the SEM. Copyright American Association for the Advancement of Science, 2017. l Antitumor effect of siG12D-LODER™ combined with chemotherapy in locally advanced inoperable pancreatic cancer in a patient. A computed tomography (CT) scan was performed before and after (nine months later) the administration of siG12D-LODER™. The tumor measured 35.42 and 26.16 mm in the longest diameter, respectively. m PANC-1 tumor growth in animals receiving iExosomes or other control formulations.238N = 3 mice per group. Copyright Macmillan Publishers Limited, part of Springer Nature, 2017. n The 24-h Uox (urinary oxalate) values of healthy volunteers and patients treated with a single dose of Nedosiran (DCR-PHXC, a GalXC™ therapeutic) (3 and 6 mg/kg).265 Copyright Dicerna Pharmaceuticals, 2020. o Reduction of serum AAT in volunteers receiving a single dose of ARO-AAT (a TRiM™ therapeutic, 35–300 mg).266 The error bars show the SEM. Copyright Arrowhead Pharmaceuticals, 2019
Full size image
Fig. 6

siRNA delivery platforms that have been evaluated preclinically and clinically. Varieties of lipids or lipidoids, siRNA conjugates, peptides, polymers, exosomes, dendrimers, etc. have been explored and employed for siRNA therapeutic development by biotech companies or institutes. The chemical structures of the key component(s) of the discussed delivery platforms, including Dlin-DMA, Dlin-MC3-DMA, C12-200, cKK-E12, GalNAc–siRNA conjugates, MLP-based DPC2.0 (EX-1), PNP, PEI, PLGA-based LODER, PTMS, GDDC4, PAsp(DET), cyclodextrin-based RONDEL™ and dendrimer generation 3 are shown. DLin-DMA (1,2-dilinoleyloxy-3-dimethylaminopropane), DLin-MC3-DMA (6Z,9Z,28Z,31Z)-heptatriaconta-6,9,28,31-tetraen-19-yl-4-(dimethylamino) butanoate, DPC Dynamic PolyConjugates, MLP membrane-lytic peptide, CDM carboxylated dimethyl maleic acid, PEG polyethylene glycol, NAG N-acetylgalactosamine, PNP polypeptide nanoparticle, PEI poly(ethyleneimine), LODER LOcal Drug EluteR, PLGA poly(lactic-co-glycolic) acid, PTMS PEG-PTTMA-P(GMA-S-DMA) poly(ethylene glycol)-co-poly[(2,4,6-trimethoxybenzylidene-1,1,1-tris(hydroxymethyl))] ethane methacrylate-co-poly(dimethylamino glycidyl methacrylate), GDDC4 PG-P(DPAx-co-DMAEMAy)-PCB, where PG is guanidinated poly(aminoethyl methacrylate) PCB is poly(carboxybetaine) and P(DPAx-co-DMAEMAy) is poly(dimethylaminoethyl methacrylate-co-diisopropylethyl methacrylate), PEG-PAsp(DET) polyethylene glycol-b-poly(N′-(N-(2-aminoethyl)-2-aminoethyl) aspartamide), PBAVE polymer composed of butyl and amino vinyl ether, RONDEL™ RNAi/oligonucleotide nanoparticle delivery
Full size image
The relatively high molecular weight (~13–16 kD) and net negative charge prevent artificial siRNA from crossing the cell membrane. Hence, we attempted to determine whether any cell can internalize siRNA without a carrier, leading to the conclusion that naked siRNA can only be taken up by a few cell types, e.g., retinal ganglion cells (RGCs) and neurons. In addition, researchers have also been devoted to identifying various carriers to achieve efficient transmembrane delivery. As a result, cationic cell-penetrating peptides (CPPs) become a choice at the early stage. CPPs, typically tailed with arginine-rich sequences, can form bidentate bonds by the interaction between the guanidinium groups and the negative phosphates, sulfates and carboxylates on the cell surface.113 This interaction causes membrane pore formation, leading to the cellular uptake of siRNA. As another strategy, the negative charge of siRNA can be neutralized by positively charged lipids or polymers, conferring siRNA the ability to more readily bind to the membrane and become easily internalized via adsorptive pinocytosis.
Finally, siRNAs must effectively escape from endosomes and lysosomes to the cytoplasm, where antisense strands of siRNAs need to be loaded into RISCs. Many delivery systems employ a pH-sensitive unit to respond to pH changes in the endosome and lysosome,114,115 where they will absorb H+, presenting a positive charge on the surface. Then, the osmotic pressure will increase in the endosome or lysosome, resulting in the internal flow of Cl− and H2O. Finally, these changes may cause membrane disruption and siRNA release to the cytoplasm. This so-called 'proton sponge effect' or 'colloid osmotic pressure effect' results in membrane destabilization116,117,118,119 or membrane swelling,120,121 respectively. However, the underlying mechanism of endosomal release remains to be further illuminated. Only 1–2% of internalized LNP-loaded siRNAs were released into the cytoplasm, and this only occurred within a limited time frame after internalization.122,123 Hence, further understanding the escape mechanism and how to enhance the escape efficiency is of great importance for siRNA drug development. Recently, Wang and colleagues124 developed novel endoplasmic reticulum (ER) membrane-modified hybrid nanoplexes (EhCv/siRNA NPs). Compared with unmodified nanoplexes, they showed much higher RNAi activity in vitro and in vivo. The functional proteins on the ER membrane have an important role in intracellular trafficking of siRNA, helping siRNA reach the cytoplasm through the endosome–Golgi–ER pathway instead of the endosome–lysosome pathway, thereby avoiding the lysosomal degradation of siRNA. In addition, electroporation enables siRNA to directly cross the cell membrane, which also constitutes an effective approach to circumvent the endosomal escape issue.125,126,127,128,129,130,131
To date, two RNAi therapeutics, ONPATTRO® and GIVLAARI™, have been approved for commercial application, and two siRNAs, lumasiran (ALN-GO1) and inclisiran, have been submitted for new drug application (NDA) to the FDA. Seven siRNAs are undergoing phase 3 clinical studies, and more candidates are in the early developmental stage. Various delivery systems, e.g., LNPs, DPC™, TRiM™, GalNAc–siRNA conjugates, LODER™ polymers, exosomes and polypeptide nanoparticles (PNPs), have been explored (Figs. 5, 6). Based on these systems, plentiful drug pipelines have been established. We discuss this information in the following sections.
Naked siRNA-based therapeutics
Naked siRNA can be defined as a system that contains no delivery system that is associated with siRNA either covalently or noncovalently.40 Because no protective effect is offered by the delivery vehicle, siRNA should be modified carefully to be resistant to enzyme degradation and extend its circulation time in the bloodstream. As siRNA is naturally filtered to the kidney, it can be used to silence kidney-expressed genes and treat renal diseases.132 Alternatively, a feasible strategy for naked siRNA drug development is the local injection of this molecule into specific organs that are relatively closed off and contain few nucleases, e.g., the eye.
QPI-1002 (I5NP)132,133 and QPI-100736,134 were developed by Quark Pharmaceuticals (Table 1). QPI-1002 is a 19-base pair, 2′-O-methylated, blunt and naked siRNA135 targeting p53 for treating acute kidney injury (phase 2) and delayed graft function (phase 3). The sequence of the sense strand is 5′-GAAGAAAATTTCCGCAAAA-3′.135 Because of their small sizes, most siRNAs accumulate in the kidney after being intravenously administered, achieving concentrations 40 times greater than those in other organs, followed by rapid entry into proximal tubule cells. An in vitro study determined that QPI-1002 elicited near-complete p53 mRNA elimination at a concentration of ~1 nM, with an IC50 of ~0.23 nM. This molecule also effectively inhibited p53 protein expression, even at a transfection concentration of 0.5 nM. Furthermore, bilateral renal-clamp studies were conducted to identify the effect of siRNA on the preservation of kidney function. Compared with PBS-treated animals, siRNA-treated animals showed ischemia with decreased serum creatinine levels from 3.7 mg/dL to 1.9 mg/dL. Adverse effects (AEs) were found only when the dose was higher than 1000 mg/kg in nonhuman primates and higher than 800 mg/kg in rats.133 A phase 1 clinical study (NCT00554359) of QPI-1002 revealed an ideal safety profile.36,132
Download PDF
Review Article
Open Access
Published: 19 June 2020
Therapeutic siRNA: state of the art
Bo Hu,
Liping Zhong,
[…]
Xing-Jie Liang
Signal Transduction and Targeted Therapy volume 5, Article number: 101 (2020) Cite this article
11k Accesses
14 Citations
22 Altmetric
Metricsdetails
ABSTRACT
RNA interference (RNAi) is an ancient biological mechanism used to defend against external invasion. It theoretically can silence any disease-related genes in a sequence-specific manner, making small interfering RNA (siRNA) a promising therapeutic modality. After a two-decade journey from its discovery, two approvals of siRNA therapeutics, ONPATTRO® (patisiran) and GIVLAARI™ (givosiran), have been achieved by Alnylam Pharmaceuticals. Reviewing the long-term pharmaceutical history of human beings, siRNA therapy currently has set up an extraordinary milestone, as it has already changed and will continue to change the treatment and management of human diseases. It can be administered quarterly, even twice-yearly, to achieve therapeutic effects, which is not the case for small molecules and antibodies. The drug development process was extremely hard, aiming to surmount complex obstacles, such as how to efficiently and safely deliver siRNAs to desired tissues and cells and how to enhance the performance of siRNAs with respect to their activity, stability, specificity and potential off-target effects. In this review, the evolution of siRNA chemical modifications and their biomedical performance are comprehensively reviewed. All clinically explored and commercialized siRNA delivery platforms, including the GalNAc (N-acetylgalactosamine)–siRNA conjugate, and their fundamental design principles are thoroughly discussed. The latest progress in siRNA therapeutic development is also summarized. This review provides a comprehensive view and roadmap for general readers working in the field.
Download PDF
Introduction
Gene therapy is a promising therapeutic platform because it targets disease-causing genes in a sequence-specific manner, which enables more precise and personalized treatment of diverse life-threatening diseases.1 By introducing a certain nucleic acid modality to the desired tissue of the patient, gene expression can be downregulated, augmented or corrected. Small interfering RNA (siRNA), microRNA (miRNA) and inhibitory antisense oligonucleotides (ASOs) are representative molecules used to trigger gene inhibition, whereas plasmid DNA, messenger RNA (mRNA), small activating RNA (saRNA), splicing-modulatory ASOs and CRISPR (clustered regularly interspaced short palindromic repeats)/Cas (CRISPR-associated protein) systems are usually employed to increase or correct target gene expression.2,3,4 Currently, many therapeutic programs have been explored to treat certain diseases.
RNA interference (RNAi) is a natural defense mechanism for the invasion of exogenous genes.5,6 RNAi modalities, e.g., siRNA and miRNA, can knockdown the expression of target genes in a sequence-specific way (Fig. 1) by mediating targeted mRNA degradation (for siRNA and miRNA) or mRNA translation repression (for miRNA). As a result of the slight differences between siRNA and miRNA, siRNA can typically trigger more efficient and specific gene silencing than miRNA, whereas one miRNA may compromise the expression of several different target genes simultaneously. Hence, siRNA and miRNA have different roles in pharmaceutical practice.
Fig. 1

Schematic illustrations of the working mechanisms of miRNA (a) and siRNA (b)
Full size image
Since the establishment of the RNAi concept in 1998, siRNA therapeutics have experienced many ups and downs. In 2001, Tuschl et al.7 successfully silenced the expression of a specific gene by introducing chemically synthesized siRNA into mammalian cells, leading to the emergence of a developmental upsurge. Although siRNA therapy once suffered due to the obstacles of its stability, specificity and delivery, advances in chemical modification and delivery brought the field to a robust and rapidly developing area of research again in recent years. After a 20-year journey, the United States Food and Drug Administration (FDA) and the European Commission (EC) approved ONPATTRO® (patisiran, ALN-TTR02) as the first commercial RNAi-based therapeutic for the treatment of hereditary amyloidogenic transthyretin (hATTR) amyloidosis with polyneuropathy in adults in 2018.2,8 Recently, the FDA-approved GIVLAARI™ (givosiran, ALN-AS1) for the treatment of adults with acute hepatic porphyria (AHP).9,10,11,12
siRNA has innate advantages over small molecular therapeutics and monoclonal antibody drugs because siRNA executes its function by complete Watson–Crick base pairing with mRNA, whereas small molecule and monoclonal antibody drugs need to recognize the complicated spatial conformation of certain proteins. As a result, there are many diseases that are not treatable by small molecule and monoclonal antibody drugs since a target molecule with high activity, affinity and specificity cannot be identified. In contrast, theoretically, any gene of interest can be targeted by siRNA since only the right nucleotide sequence along the targeting mRNA needs to be selected. This advantage confers the siRNA modality with a shorter research and development span and a wider therapeutic area than small molecule or antibody drugs, especially for those genes that are unfeasible for development with such strategies.
Although siRNA holds promising prospects in drug development, several intracellular and extracellular barriers limit its extensive clinical application. Naked and unmodified siRNA possesses some disadvantages, such as (1) unsatisfactory stability and poor pharmacokinetic behavior and (2) the possible induction of off-target effects. The phosphodiester bond of siRNA is vulnerable to RNases and phosphatases. Once it is systematically administered into circulation, endonucleases or exonucleases throughout the body will quickly degrade siRNA into fragments, thus preventing the accumulation of intact therapeutic siRNA in the intended tissue. In theory, siRNA only functions when its antisense strand is completely base-paired to the target mRNA. However, a few mismatches are tolerated by the RNA-induced silencing complex (RISC), which may lead to undesired silencing of those genes with a few nucleotide mismatches. In addition, the RISC-loaded sense strand of siRNA may also knockdown the expression of other irrelevant genes. Moreover, unformulated and unmodified siRNA may lead to the activation of Toll-like receptor 3 (TLR3) and adversely affect the blood and lymphatic systems.13 These discoveries have raised many concerns about the undesirable effects and pharmaceutical issues of siRNAs.
To maximize the treatment potency and reduce or avoid the side effects of siRNA, researchers have made great efforts to investigate various chemical modification geometries and to develop many different delivery systems. As a result, a series of modification patterns were proposed and evaluated preclinically and clinically with respect to their effects on activity, stability, specificity and biosafety. Delivery materials derived from lipids, lipid-like materials (lipidoids), polymers, peptides, exosomes, inorganic nanoparticles, etc., have been designed and investigated.2,8,14,15,16,17,18,19,20,21,22 As a result, several modification patterns and delivery platforms have been employed in clinical studies. Here, the detailed evolution and advances in the modification and delivery technologies of siRNA are comprehensively summarized and discussed. This review provides an overview and a handbook for reviewing siRNA therapeutic development.
siRNA modification
During the early stages of developing siRNA therapeutics, many agents were designed based on completely unmodified or slightly modified siRNA to arrive at appropriate tissue and then silence the target gene. These molecules can mediate gene silencing in vivo, especially in tissues that receive local drug administration, e.g., eyes. However, limited efficacy and potential off-target effects may be observed with these modalities. For example, Kleinman and colleagues23 observed that bevasiranib and AGN211745 triggered significant activation of toll-like receptor 3 (TLR3) and its adapter molecule TRIF, inducing the secretion of interleukin-12 and interferon-γ. Bevasiranib and AGN211745 are siRNA therapeutics developed for the treatment of age-related macular degeneration.24 The VEGFA-targeting bevasiranib25 and VEGFR1-targeting AGN21174526 are unmodified and slightly modified siRNAs, respectively. Moreover, Kleinman and colleagues further demonstrated that siRNA classes with sequences of 21 nucleotides or longer, regardless of which genes they target, can suppress CNV in mice compared to bevasiranib and AGN211745. These findings eventually led to the termination of the clinical investigation of bevasiranib in 2009.
Chemically modified siRNAs, such as siRNAs with substitution of the 2′-OH with a 2′-O-methyl (2′-OMe)27 or 2′-methoxyethyl (2′-MOE)28 group or the substitution of certain nucleotides with locked nucleic acid (LNA),29 unlocked nucleic acid (UNA)30 or glycol nucleic acid (GNA)31 (Fig. 2), can efficiently suppress immunostimulatory siRNA-driven innate immune activation, enhance activity and specificity, and reduce off-target-induced toxicity. To enhance the potency and reduce the potential toxicity of siRNA, numerous chemical modification geometries have been established and tested.32
Fig. 2

Structures of chemical modifications and analogs used for siRNA and ASO decoration. According to the modification site in the nucleotide acid, these structures can be divided into three classes: phosphonate modification, ribose modification and base modification, which are marked in red, purple and blue, respectively. R = H or OH, for RNA or DNA, respectively. (S)-cEt-BNA (S)-constrained ethyl bicyclic nucleic acid, PMO phosphorodiamidate morpholino oligomer
Full size image
According to the natural structure of nucleotides, chemical modifications can be placed at the phosphate backbone, the ribose moiety or the base. Typically, these modifications are simultaneously introduced in siRNA. For example, the combination of 2′-OMe and phosphorothioate (PS) modifications facilitates the systemic administration of cholesterol-conjugated siRNA and achieves efficient gene silencing in vivo.33 In addition, the combination of 2′-OMe and 2′-deoxy-2′-fluoro (2′-F) has been used for ONPATTRO®34,35 (Figs. 2, 3). For QPI-1007, a naked and unformulated siRNA therapeutic,2 2′-OMe was alternatively incorporated into the antisense strand, and an inverted deoxyabasic moiety and an L-DNA cytidine nucleotide were used in the sense strand36 (Figs. 2,3). Moreover, modifications with phosphorothioate (PS), 2′-OMe, 2′-F and 2′-deoxy were employed for inclisiran (ALN-PCSsc), an siRNA therapeutic for the treatment of hypercholesterolemia37,38 (Figs. 2, 3). Overall, the precise modification of siRNA can increase its efficacy, specificity, and stability and reduce its toxicity and immunogenicity. Detailed information will be discussed in the following sections.
Fig. 3

Representative designs for the chemical modification of siRNA. The sequences and modification details for ONPATTRO®, QPI-1007, GIVLAARI™ and inclisiran are included. The representative siRNA modification patterns developed by Alnylam (STC, ESC, advanced ESC and ESC+) and arrowhead (AD1-3 and AD5) are shown. Dicerna developed four GalNAc moieties that can be positioned at the unpaired G–A–A–A nucleotides of the DsiRNA structure. 2′-OMe 2′-methoxy, 2′-F 2′-fluoro, GNA glycol nucleic acid, UNA unlocked nucleic acid, SS sense strand, AS antisense strand
Full size image
Phosphonate modification
Phosphorothioate (PS) linkage has previously been used in antisense oligonucleotide (ASO) modification.39 This modification was achieved by leveraging a sulfur atom to replace one nonbridging oxygen of a phosphodiester (Fig. 2). PS linkage endows modified oligonucleotides with resistance to nucleases,40 and these molecules more readily combine with plasma proteins,41 especially albumin42 than their unmodified counterparts, which may result in a longer circulation time. Although some data have shown that PS modification reduces the binding affinity between the oligonucleotide and its target sequence32 to some extent and aggravates chemistry-related toxicities with a high PS content,43,44 PS modification is still very important and necessary for both siRNA and ASO because, compared with unmodified oligonucleotides, PS-modified oligonucleotides are more hydrophobic and stable molecules with higher affinity to certain proteins.44 Appropriate protein binding may prolong the half-life of oligonucleotides in circulation, and the half-life of the elimination phases may be as long as a few days.41 Protein binding is typically beneficial for the cell entry of oligonucleotides; however, too much protein binding showed a positive correlation with in vivo toxicity.45,46 Hence, the position and number of PS linkages are pivotal for their application. For siRNA modification, Alnylam introduced two PS linkages at the first two nucleotides at the 5′-end of the sense strand and two PS linkages at the first two nucleotides at both the 5′-end and the 3′-end of the antisense strand of siRNA (Fig. 3). Moreover, compared to the parent siRNA without modifications, the PS linkage-containing siRNA has a barely changed in vivo biodistribution profile, as it predominantly accumulates in clearance organs, e.g., liver, kidneys and intestines, followed by glandular tissues, bone marrow, adipocytes and lymph nodes.47,48,49,50
Notably, there are two configurations for PS linkages, namely, Rp and Sp isomers (Fig. 2). Verdine and colleagues (team from Wave Life Sciences)51 demonstrated that the stereochemistry of PS significantly influences the pharmacologic performance of ASOs. PS linkages with the Sp configuration are more stable than those with the Rp configuration. More importantly, they unveiled that stereopure ASOs with a triplet stereochemical code, 3′-SpSpRp, make the target RNA more vulnerable to RNase H1 in vitro and achieve a more durable inhibition profile in vivo than stereorandom ASOs. However, Østergaard et al. (team from Ionis Pharmaceuticals)52 disclosed that controlling PS chirality markedly affects the interactions between the ASOs and RNase H1; however, this modulation failed to enhance the overall therapeutic effects. The sequence and design of ASOs are the primary drivers that determine the performances of gapmer ASOs. In addition, recently disclosed patents filed by Alnylam reported that the chirality of PS linkages may markedly influence siRNA performance.53 The terminal chiral PS linkages at the 3′- and 5′-ends of the antisense strand of siRNA prefer Sp and Rp configurations, respectively. The chiral PS number can be 1, 2 or 3 at the 3′-end and 1 or 2 at the 5′-end of the siRNA antisense strand. In addition, the chiral PS linkages at the 5′-end of the siRNA sense strand can adopt either an Sp or an Rp configuration and are positioned at the first internucleotide linkage at the 5′-end of the sense strand. Overall, the chirally modified siRNA can comprise four, five or more terminal, chirally modified PS linkages.
In addition, other residues have been identified and successfully used to replace the phosphodiester group in oligonucleotides and change the properties of intact strands, including phosphorodithioate (PS2),54 methylphosphonate (MP), methoxypropylphosphonate (MOP) and peptide nucleic acid (PNA) (Fig. 2). PS2 modification can increase the affinity between RISC and siRNA. The site-specific incorporation of MP and MOP modifications has been used to reduce ASO protein binding since the PS backbone is negatively charged, whereas alkylphosphonate linkages are charge neutral. As a result, the MOP linkage incorporated at position 2 or 3 from the 5′-end of the DNA gap significantly mitigated the hepatotoxicity of ASOs.46 In addition, both PNA55,56,57 (Fig. 2) and phosphotriesters58 are very meaningful modifications to make siRNA or ASO molecules more druggable, although these modifications are not as predominant as PS. PNA is also typically used to modify detection probes for capturing and detecting certain nucleic acid targets.59
In addition, phosphonate modification with various analogs at the 5′-end of siRNA or single-stranded siRNA (ss-siRNA) is a newly developed but important decoration strategy to enhance siRNA activity. The 5′-phosphate of exogenously introduced siRNA is required for RISC loading. A 5′-phosphate can be modified via either phosphorylation in cells mediated by Clp1 (cleavage and polyadenylation factor I subunit 1) or chemical synthesis in the laboratory.60 However, the natural 5′-phosphate can also be removed by dephosphorylation in cells. Together with the demand to enhance siRNA stability, researchers have identified a series of analogs that have similar conformation and steroidal electronic properties to natural phosphates but are resistant to dephosphorylases. As a result, 5′-(E)-vinyl phosphonate (5′-(E)-VP), 5′-methylphosphonate, (S)-5′-C-methyl with phosphate, 5′-phosphorothioate, etc., were designed and evaluated in vitro and in vivo61 (Fig. 2). Among these strategies, 5′-(E)-VP decoration, which substitutes bridge oxygen and carbon with E-vinyl phosphonate moieties at the 5′-end, is the most potent and metabolically stable mechanism. The replaced hydrophobic substitutions enhance the stability of intact oligonucleotides and are favorable for RISC loading. Ionis Pharmaceuticals first developed and own the intellectual property rights to the 5′-(E)-VP structure. This technology was first applied to ss-siRNA and achieved ideal pharmacodynamic and pharmacokinetic profiles. Modifying ss-siRNA with 5′-(E)-VP and other specific modification structures enables potent repression of targeted gene expression by ss-siRNA in vivo.61,62 Inspired by this finding, the 5′-(E)-VP modification was further applied to double-stranded siRNA. Data have revealed that this modification improves siRNA accumulation and residence time in tissue and enhances siRNA potency in vivo by elevating Argonaute-2 binding.63,64,65
Ribose modification
Ribose modifications at the 2′ position have been widely used to protect siRNA from attacking ribonucleases, which require the 2′-OH group for the hydrolysis of RNA. Notably, 2′-O-methyl (2′-OMe) (Fig. 2) is a naturally occurring ribosugar and the most frequently used modification in drug development so far66 because this molecule can enhance stability67 by blocking the nucleophilic 2′-OH group. In addition, 2′-OMe also shows increased affinity for its target mRNA68 and reduced immunogenicity in the body. Based on 2′-OMe, a series of analogs have been identified, among which 2′-O-methoxyethyl (2′-O-MOE) (Fig. 2) is one of the most useful and popular analogs of 2′-OMe. Moreover, 2′-O-MOE shows higher binding affinity for RNA than 2′-OMe with a 0.9–1.7 °C change in ΔTm per modified nucleotide, which further increases the capability of the modified siRNA to resist nuclease attack.32 Another widely used 2′-OH substitution is 2′-deoxy-2′-fluoro (2′-F)69 (Fig. 2). Highly electronegative fluorine makes the modified siRNA readily adopt a C3′-endo conformation, providing considerable benefits in binding affinity, with a ΔTm of ~2.5 °C per modified nucleotide.32 Interestingly, 2′-arabino-fluoro (2′-Ara-F), an alternative fluoro substitution form, can also be used to modify nucleic acid therapeutics, e.g., antisense oligonucleotides45,70 and siRNA.71 Both 2′-O-benzyl and 2′-O-methyl-4-pyridine (2′-O-CH2Py(4)) were well tolerated when they were incorporated on the guide strand of siRNA in vivo72 (Fig. 2). siRNA with modifications of four 2′-O-benzyl or six 2′-O-CH2Py(4) showed comparable activity with the unmodified siRNA in vivo, and increased activity was even achieved when these modifications were placed at the 8 and 15 positions on the siRNA guide strand.
In addition, 2′-C, 4′-C, and even the whole sugar ring can also undergo modifications, resulting in molecules including UNAs, LNAs, GNAs, (S)-cEt-BNAs, tricyclo-DNA (tcDNA) and phosphorodiamidate morpholino oligomers (PMOs) (Fig. 2). UNA, with higher flexibility and thermal destabilization than the unmodified product due to unconnected 2′ and 3′ carbons, can block the entry of passenger strands and promote the RISC loading of the guide strand by introducing chemical asymmetry into duplex siRNAs. Similarly, glycol nucleic acid (GNA),31 another thermally destabilizing nucleotide, can be used to erase off-target effect-induced hepatotoxicity by including it in the seed region of the siRNA guide strand. LNA is a bicyclic structure that contains a bridge between the 2′ oxygen and the 4′ carbon. It "locks" the ribose into its preferred C3′-endo conformation and significantly increases the affinity of base pairing.40 Data have shown that the incorporation of LNAs increases the DNA melting temperature up to 8 °C per LNA.73 Given the successes of LNAs, many more bicyclic and even tricyclic analogs, including ethyl-bridged nucleic acids (ENAs),74 constrained ethyl (cEt) nucleic acids75 and tricyclo-DNA,76 have been engineered and incorporated into RNA or DNA strands in succession. Moreover, PMOs77 do not look like classic nucleotides, as the ribose subunit has been substituted with a morpholine subunit. PMOs are uncharged at physiological pH and are not substrates of RNase H. Therefore, they are mainly used to block RNA splicing or translation.78
Base modification
Base replacement is of great benefit to nucleic acid-based drug development. For instance, the substitution of pseudouridine,79 2-thiouridine,80 N6-methyladenosine,80 5-methylcytidine81 (Fig. 2) or other base analogs of uridine and cytidine residues can reduce innate immune recognition while making ASOs more resistant to nucleases. However, the artificial base substitution of ASOs, similar to siRNA, is basically at the stage of research and development. Pharmaceutical corporations still hold a prudent attitude toward these molecules. Instead, these companies prefer to use naturally occurring base structures, e.g., 5mC and 6 mA, to modify certain base(s), probably because of concerns about the safety of the metabolized unnatural residues that potentially might be incorporated into the genome.
N-ethylpiperidine triazole-modified adenosine analogs with Hoogsteen or Watson–Crick (WC) facial localization have also been used to modify siRNA,82,83,84,85 which can disrupt nucleotide/TLR8 interactions and therefore reduce the immunogenicity of siRNA. N-ethylpiperidine 7-EAA triazole (7-EAA, 7-ethynyl-8-aza-7-deazaadenosine) can pair well with uridine and form an A-form helix structure. This molecule will be recognized as adenosine by Avian myeloblastosis virus reverse transcriptase in RNA strands.83
In addition, 6′-phenylpyrrolocytosine (PhpC) is a cytosine mimic showing excellent base pairing fidelity, thermal stability and high fluorescence.86 PhpC-containing siRNA shows gene-silencing activity comparable to that of the parent molecule, and its fluorescent properties make it useful for fluorescence-based detection or monitoring or for exploring the cellular uptake and trafficking of siRNA in cells.
The internal uridine substitution of 2,4-difluorotolylribonucleoside (rF) is well tolerated by siRNA activity.87 The base pairing between rF and adenosine (A) is relatively unstable compared to the uridine–adenosine pair because the fluorine atom is more hydrophobic than uridine and cannot serve as a hydrogen bond acceptor.
Zhang and colleagues88 reported that a 5-nitroindole modification at position 15 of the siRNA passenger strand greatly reduced the activity of the passenger strand, whereas the effectiveness of the siRNA guide strand was barely affected. This modification provided a practical strategy to reduce the passenger strand-mediated off-target effects.
siRNAs containing 5-fluoro-2′-deoxyuridine (FdU) moieties can suppress targeted gene expression. The effects depend on the locations of the FdU substitution. In addition, these modified siRNAs quickly release FdU after entering the cell, thereby further inducing a variety of DNA-damage repair and apoptosis pathways and ultimately triggering cell death. These findings provide a strategy for siRNA-based cancer therapy.89
Moreover, many other less common base analogs have also been used in siRNA modifications. Their application helps researchers better understand the mechanism of gene silencing and helps develop new methods to mitigate off-target effects caused by harmful protein binding or undesired targeting of mRNA.90
Modification patterns used in clinical studies or for commercial siRNA
Early siRNA therapeutic programs typically employ partial or slight modifications by using 2′-OMe, 2′-F or PS,91,92 e.g., alternative 2′-OMe modifications were placed in the antisense strand of QPI-1007, whereas just one L-DNA was used in the sense strand. After making great efforts to understand the effects of various modifications on siRNA activity, specificity and toxicity, researchers have realized that heavy modification of siRNA, even full-length embellishment, probably do not affect siRNA activity. Furthermore, these judicious modification strategies can significantly enhance siRNA stability or biocompatibility. As a result, several different, interrelated modification patterns have been proposed and validated preclinically and clinically.
Standard template chemistry (STC) (Fig. 3) is a universal modification pattern established by Alnylam Pharmaceuticals. When the lengths of sense and antisense siRNA are 21 and 23 nt, respectively (referred to as "S21nt" and "AS23nt", respectively, in the following description), 2 PS linkages are placed at the 3′ terminus of the antisense strand, three consecutive 2′-F moieties are incorporated at positions 9, 10 and 11 of the sense strand from the 5′-end, and consecutive 2′-OMe modifications are placed at positions 11, 12 and 13 of the antisense strand from the 5′-end. Moreover, alternative 2′-OMe and 2′-F moieties are employed at other positions in both strands. It is worth mentioning that 2′-OMe and 2′-F are complementarily used for all positions of both strands of siRNA (Fig. 3). It is considered that placing three consecutive 2′-F moieties, instead of larger groups, at the middle of the sense strand aims to facilitate the cleavage and removal of the sense strand in RISC, as release of the sense strand is the prerequisite of RISC activation. PS linkages and OMe positioned at the terminals are used to enhance the capability of siRNA to resist degrading enzyme attack.
The STC design (e.g., as used in revusiran) has been proven to remarkably enhance siRNA stability and affinity without compromising intrinsic RNAi activity.93 ALN-AT3, another STC-modified RNAi therapeutic, also showed potent and durable gene silencing after single subcutaneous (s.c.) administration.94 However, the clinical investigation of revusiran was discontinued due to unbalanced death between revusiran- and placebo-treated groups. Although the detailed investigation report by Alnylam illustrated that the deaths of revusiran-treated patients were not correlated with the test drug, there are still strong concerns regarding STC-modified siRNA therapeutics. Hence, the next generation of modification patterns, called enhanced stabilization chemistry (ESC) (Fig. 3), was proposed. This design includes four more PS linkages at the 5′-end of the antisense strand and the 3′-end of the sense strand. Another important change was the reduction in 2′-F substitutions, probably because heavy use of 2′-F may magnify toxicity.95 These changes markedly enhanced siRNA potency and duration compared with the changes inherent to the STC design.9,96,97 This advancement allowed us to achieve the desired pharmacodynamic effect at a significantly lower dose, with a markedly reduced dosing frequency.98 Cemdisiran (ALN-CC5) achieved potent gene silencing as long as more than 1 year after receipt of a single dose of the testing therapeutic in a phase 1/2 clinical study (Fig. 4f), representing an unprecedented milestone in pharmaceutical history. Moreover, the newly approved GIVLAARI™ is an siRNA therapeutic based on the ESC design (Fig. 3).99
Fig. 4

Preclinical and clinical performance of siRNAs with comprehensive modification chemistries. a Sense and antisense strand optimization by using transthyretin (Ttr)-targeting siRNAs and the combinatorial designs explored across 10 siRNAs. The effect relative to the parent is described as the model-adjusted mean difference in the activity of the design variant (DV) compared to the parent. b–d The liver exposure, RISC loading, and silencing activities of the parent, DV18, and fully 2′-OMe conjugates. Data are shown as the mean ± SD. e Gene-silencing duration of antithrombin-targeted siRNAs modified with the designs of patent and DV22 evaluated in Cynomolgus monkeys (n = 3 per group). Asterisks indicate that a significant difference was observed between the parent and DV22. Error is shown as the SD. f C5 knockdown profile after applying a single dose of Cemdisiran that employs ESC modification in a phase 1/2 clinical study. Error is represented as the SEM. a–e Copyright of Elsevier Inc., 2018. f Copyright of Alnylam Pharmaceuticals, 2016
Full size image
In addition, Alnylam comprehensively optimized the modification design variants (DVs) of the ESC strategy across multiple siRNAs with an iterative screening approach to further enhance stability without compromising intrinsic RNAi activity. As a result, advanced ESC designs, e.g., DV18 and DV22, were proposed (Figs. 3 and 4a–e).96 Compared to the former scheme, the advanced strategy maintains six PS linkages at three strand terminals; however, this modification scheme significantly reduces the proportion of 2′-F moieties. Only 10 and 8 2′-F substitutions are used in two strands for the advanced ESC designs of DV18 and DV22, respectively. For DV18, the 2′-F modifications are positioned at sites 7, 9, 10 and 11 in the sense strand and sites 2, 6, 8, 9, 14 and 16 in the antisense strand (all from the 5′-ends of the strands). Compared with DV18, DV22 has the 2′-F modifications at 8 and 9 in the antisense strand further replaced with 2′-OMe (Figs. 3 and 4a). Both the DV18 and DV22 designs can achieve significantly higher liver exposure and RISC loading and more potent and durable gene silencing than the parent design in preclinical species, including nonhuman primates (Fig. 4b–e). The underlying mechanisms were not fully unveiled in Alnylam's literature. However, Zheng et al.100 demonstrated that modification of the 14th position of the siRNA guide strand could eliminate its gene-silencing activity by reducing RISC loading and target degradation, and the larger the modification group used was, the higher the reduction efficiency. Song et al.28 further proved that decorations at positions 9 and 10 (for siRNA with a 19 nt/19 nt structure) markedly reduced the activity of the unmodified strand of siRNA without disturbing the potency of the modified strand. In light of this, the specificity and activity of siRNAs could be improved by introducing 2′-MOE at the cleavage site of the siRNA. These observations are in line with the rationales of advanced ESC designs.
Furthermore, investigators from Alnylam have disclosed that the hepatotoxicity of GalNAc (N-acetylgalactosamine)-siRNA conjugates is attributed to off-target gene silencing mediated by miRNA-like recognition between siRNA and a mistargeted RNA.31 Researchers have demonstrated that disorganizing the nucleotides of the seed region without changing the 2′-OMe, 2′-F or PS content or placing a GNA at position 7 of the siRNA antisense strand can markedly alleviate off-target effects and mitigate hepatotoxicity because these modifications affect the binding of siRNA with undesired target mRNA in a seed region-specific manner. The employment of GNA in the siRNA seed region is the main technical characteristic for the ESC+ design compared to advanced ESC designs (Fig. 3). Several investigational siRNA therapeutics of Alnylam, e.g., ALN-AAT02, ALN-HBV02, and ALN-AGT, utilize the ESC+ design as the modification pattern.
In addition to Alnylam, Arrowhead Pharmaceuticals (Arrowhead) proposed a series of new modification designs (Fig. 3) by combining the modification patterns (e.g., inverted bases) of siRNAs with 2′-OMe/2′-F decorations similar to the aforementioned STC and ESC modification patterns. Typically, the modification strategies proposed by Arrowhead are characterized by placing inverted deoxythymine (idT) at the strand terminus and/or termini, including UNA and X (without nucleoside base), flanking the UAU or UAUAU motif(s), conjugating the siRNA with cholesterol or other hydrophobic substrates, etc. These designs are currently being investigated in clinical trials.101 Dicerna Pharmaceuticals (Dicerna)102 employed their proprietary dicer-substrate siRNA (DsiRNA) technology to develop RNAi therapeutics. In addition, these researchers also established a series of DsiRNA modification chemistries. For instance, three consecutive 2′-F moieties may be placed at sites 9, 10, and 11 in the sense strand of siRNA from the 5′-end, and the other sites may use alternative 2′-OMe and 2′-F moieties in the flank sequence. One example of a constant flank sequence is 'GCAGCCGAAAGGUGC', which contains inner complementary pairing motifs of 'GCAGCC' and 'GGCUGC'. Consecutive 2′-OMe moieties may be used to modify these motifs, and consecutive RNA or DNA nucleosides without additional 2′ modifications may be employed for the bubble motif of GAAA. Moreover, DNA nucleosides can also be applied at sites 2, 12, 16, 18, 20 and 21 in the antisense strand from the 5′-end. PS linkages may be incorporated into the antisense strand at specific sites, as well as within the constant flank sequence (Fig. 3). The ligands may be placed at the bubble sequence. Silence Therapeutics103 (Fig. 3) and other RNAi-based biotech companies, such as Arbutus Biopharma, OliX Pharmaceuticals and Suzhou Ribo Life Science, have all been devoted to developing state-of-the-art siRNA modification chemistries to establish fruitful drug development pipelines.2
siRNA delivery and drug development
Barriers to siRNA delivery
RNAi modalities (siRNAs and miRNAs) need to employ proteins (such as Argonaute and Tar RNA binding protein) in the cytoplasm. A series of barriers need to be circumvented for systemically administered exogenous siRNA before it can achieve gene silencing,104,105,106 such as nuclease degradation and short-lived circulation, immune recognition in blood circulation, accumulation in desired tissue, effective transmembrane trafficking, and escape from endosomes and lysosomes to the cytoplasm. By incorporating chemical modifications, the challenges of stability against serum nucleases and avoiding immune recognition have been substantially reduced. However, other problems remain to be solved.
In blood circulation, unspecific binding and glomerular filtration both hamper the accumulation of siRNA in desired tissues. The neutral surface charge of siRNA-loaded nanoformulations is beneficial for avoiding unfavorable protein binding in circulation.107 However, protein decoration or protein corona formation is required for nanoformulation and attachment to specific tissues or cells for some delivery systems.108,109 For instance, ionizable lipid nanoparticles (iLNPs) primarily accumulate in the liver by interacting with serum lipoproteins, e.g., apolipoprotein E3 (ApoE3), which leads to specific recognition by the receptor, e.g., low-density lipoprotein receptor (LDLR).108,109 In contrast, ligand–siRNA conjugates can transport siRNA to desired tissues and cells by specific recognition and interactions between the ligand, e.g., a carbohydrate, peptide, antibody, aptamer, small molecule, etc., and the surface receptor. This active targeting strategy not only reduces siRNA accumulation in unintended tissues, thus decreasing or avoiding undesired side effects and toxicity, but also achieves potent gene silencing at a low dosage.104
siRNA is ~7–8 nm in length and 2–3 nm in diameter. This molecule is too large to cross cell membranes but small enough to be freely cleared by glomeruli,110 as molecules with a size smaller than 8 nm111 are easily filtered into the urine. Hence, once siRNAs leave the bloodstream, they will accumulate in the bladder and be excreted out from the body quickly, within a few minutes to half an hour,48,49,50 which prevents them from accumulating in targeted tissues or cells. Encapsulating siRNA into vesicles or conjugating it to certain ligands can effectively avoid renal clearance and, more importantly, can deliver siRNA to the desired tissues or cells. Lipid nanoparticles (LNPs),106 dynamic polyconjugates (DPC™), and GalNAc–siRNA conjugates have all achieved efficient siRNA delivery to hepatocytes (Figs. 5, 6).112
Fig. 5

Schemes of representative clinically studied siRNA formulations and their pharmacodynamic performance. a–g Schemes of a lipid nanoparticles, b DPC™ or EX-1™, c TRiM™, d GalNAc–siRNA conjugates, e LODER™, f iExososme and g GalXC™. h Efficacy of ONPATTRO® (a liposome formulation) in healthy volunteers, in which siRNA targeting PCSK9 is formulated in lipid nanoparticles.34 Copyright Massachusetts Medical Society, 2013. i Levels of urinary aminolevulinic acid (ALA) and porphobilinogen (PBG) after once-monthly and once quarterly doses of GIVLAARI™ in acute hepatic porphyria (AHP) patients. Copyright Alnylam Pharmaceuticals, 2019. j Reduction of plasma PCSK9 after a single increasing dose of inclisiran.38 Copyright Massachusetts Medical Society, 2017. k Serum HBsAg reduction in hepatitis B patients who received the treatment of a single dose of ARC-520 (1–4 mg/kg) (formulated with DPC2.0) on a background of daily oral NUCs.264 PBO, patients on NUC therapy receiving placebo injection. NUCs nucleos(t)ide viral replication inhibitors. The error is shown as the SEM. Copyright American Association for the Advancement of Science, 2017. l Antitumor effect of siG12D-LODER™ combined with chemotherapy in locally advanced inoperable pancreatic cancer in a patient. A computed tomography (CT) scan was performed before and after (nine months later) the administration of siG12D-LODER™. The tumor measured 35.42 and 26.16 mm in the longest diameter, respectively. m PANC-1 tumor growth in animals receiving iExosomes or other control formulations.238N = 3 mice per group. Copyright Macmillan Publishers Limited, part of Springer Nature, 2017. n The 24-h Uox (urinary oxalate) values of healthy volunteers and patients treated with a single dose of Nedosiran (DCR-PHXC, a GalXC™ therapeutic) (3 and 6 mg/kg).265 Copyright Dicerna Pharmaceuticals, 2020. o Reduction of serum AAT in volunteers receiving a single dose of ARO-AAT (a TRiM™ therapeutic, 35–300 mg).266 The error bars show the SEM. Copyright Arrowhead Pharmaceuticals, 2019
Full size image
Fig. 6

siRNA delivery platforms that have been evaluated preclinically and clinically. Varieties of lipids or lipidoids, siRNA conjugates, peptides, polymers, exosomes, dendrimers, etc. have been explored and employed for siRNA therapeutic development by biotech companies or institutes. The chemical structures of the key component(s) of the discussed delivery platforms, including Dlin-DMA, Dlin-MC3-DMA, C12-200, cKK-E12, GalNAc–siRNA conjugates, MLP-based DPC2.0 (EX-1), PNP, PEI, PLGA-based LODER, PTMS, GDDC4, PAsp(DET), cyclodextrin-based RONDEL™ and dendrimer generation 3 are shown. DLin-DMA (1,2-dilinoleyloxy-3-dimethylaminopropane), DLin-MC3-DMA (6Z,9Z,28Z,31Z)-heptatriaconta-6,9,28,31-tetraen-19-yl-4-(dimethylamino) butanoate, DPC Dynamic PolyConjugates, MLP membrane-lytic peptide, CDM carboxylated dimethyl maleic acid, PEG polyethylene glycol, NAG N-acetylgalactosamine, PNP polypeptide nanoparticle, PEI poly(ethyleneimine), LODER LOcal Drug EluteR, PLGA poly(lactic-co-glycolic) acid, PTMS PEG-PTTMA-P(GMA-S-DMA) poly(ethylene glycol)-co-poly[(2,4,6-trimethoxybenzylidene-1,1,1-tris(hydroxymethyl))] ethane methacrylate-co-poly(dimethylamino glycidyl methacrylate), GDDC4 PG-P(DPAx-co-DMAEMAy)-PCB, where PG is guanidinated poly(aminoethyl methacrylate) PCB is poly(carboxybetaine) and P(DPAx-co-DMAEMAy) is poly(dimethylaminoethyl methacrylate-co-diisopropylethyl methacrylate), PEG-PAsp(DET) polyethylene glycol-b-poly(N′-(N-(2-aminoethyl)-2-aminoethyl) aspartamide), PBAVE polymer composed of butyl and amino vinyl ether, RONDEL™ RNAi/oligonucleotide nanoparticle delivery
Full size image
The relatively high molecular weight (~13–16 kD) and net negative charge prevent artificial siRNA from crossing the cell membrane. Hence, we attempted to determine whether any cell can internalize siRNA without a carrier, leading to the conclusion that naked siRNA can only be taken up by a few cell types, e.g., retinal ganglion cells (RGCs) and neurons. In addition, researchers have also been devoted to identifying various carriers to achieve efficient transmembrane delivery. As a result, cationic cell-penetrating peptides (CPPs) become a choice at the early stage. CPPs, typically tailed with arginine-rich sequences, can form bidentate bonds by the interaction between the guanidinium groups and the negative phosphates, sulfates and carboxylates on the cell surface.113 This interaction causes membrane pore formation, leading to the cellular uptake of siRNA. As another strategy, the negative charge of siRNA can be neutralized by positively charged lipids or polymers, conferring siRNA the ability to more readily bind to the membrane and become easily internalized via adsorptive pinocytosis.
Finally, siRNAs must effectively escape from endosomes and lysosomes to the cytoplasm, where antisense strands of siRNAs need to be loaded into RISCs. Many delivery systems employ a pH-sensitive unit to respond to pH changes in the endosome and lysosome,114,115 where they will absorb H+, presenting a positive charge on the surface. Then, the osmotic pressure will increase in the endosome or lysosome, resulting in the internal flow of Cl− and H2O. Finally, these changes may cause membrane disruption and siRNA release to the cytoplasm. This so-called 'proton sponge effect' or 'colloid osmotic pressure effect' results in membrane destabilization116,117,118,119 or membrane swelling,120,121 respectively. However, the underlying mechanism of endosomal release remains to be further illuminated. Only 1–2% of internalized LNP-loaded siRNAs were released into the cytoplasm, and this only occurred within a limited time frame after internalization.122,123 Hence, further understanding the escape mechanism and how to enhance the escape efficiency is of great importance for siRNA drug development. Recently, Wang and colleagues124 developed novel endoplasmic reticulum (ER) membrane-modified hybrid nanoplexes (EhCv/siRNA NPs). Compared with unmodified nanoplexes, they showed much higher RNAi activity in vitro and in vivo. The functional proteins on the ER membrane have an important role in intracellular trafficking of siRNA, helping siRNA reach the cytoplasm through the endosome–Golgi–ER pathway instead of the endosome–lysosome pathway, thereby avoiding the lysosomal degradation of siRNA. In addition, electroporation enables siRNA to directly cross the cell membrane, which also constitutes an effective approach to circumvent the endosomal escape issue.125,126,127,128,129,130,131
To date, two RNAi therapeutics, ONPATTRO® and GIVLAARI™, have been approved for commercial application, and two siRNAs, lumasiran (ALN-GO1) and inclisiran, have been submitted for new drug application (NDA) to the FDA. Seven siRNAs are undergoing phase 3 clinical studies, and more candidates are in the early developmental stage. Various delivery systems, e.g., LNPs, DPC™, TRiM™, GalNAc–siRNA conjugates, LODER™ polymers, exosomes and polypeptide nanoparticles (PNPs), have been explored (Figs. 5, 6). Based on these systems, plentiful drug pipelines have been established. We discuss this information in the following sections.
Naked siRNA-based therapeutics
Naked siRNA can be defined as a system that contains no delivery system that is associated with siRNA either covalently or noncovalently.40 Because no protective effect is offered by the delivery vehicle, siRNA should be modified carefully to be resistant to enzyme degradation and extend its circulation time in the bloodstream. As siRNA is naturally filtered to the kidney, it can be used to silence kidney-expressed genes and treat renal diseases.132 Alternatively, a feasible strategy for naked siRNA drug development is the local injection of this molecule into specific organs that are relatively closed off and contain few nucleases, e.g., the eye.
QPI-1002 (I5NP)132,133 and QPI-100736,134 were developed by Quark Pharmaceuticals (Table 1). QPI-1002 is a 19-base pair, 2′-O-methylated, blunt and naked siRNA135 targeting p53 for treating acute kidney injury (phase 2) and delayed graft function (phase 3). The sequence of the sense strand is 5′-GAAGAAAATTTCCGCAAAA-3′.135 Because of their small sizes, most siRNAs accumulate in the kidney after being intravenously administered, achieving concentrations 40 times greater than those in other organs, followed by rapid entry into proximal tubule cells. An in vitro study determined that QPI-1002 elicited near-complete p53 mRNA elimination at a concentration of ~1 nM, with an IC50 of ~0.23 nM. This molecule also effectively inhibited p53 protein expression, even at a transfection concentration of 0.5 nM. Furthermore, bilateral renal-clamp studies were conducted to identify the effect of siRNA on the preservation of kidney function. Compared with PBS-treated animals, siRNA-treated animals showed ischemia with decreased serum creatinine levels from 3.7 mg/dL to 1.9 mg/dL. Adverse effects (AEs) were found only when the dose was higher than 1000 mg/kg in nonhuman primates and higher than 800 mg/kg in rats.133 A phase 1 clinical study (NCT00554359) of QPI-1002 revealed an ideal safety profile.36,132
QPI-1007 is a caspase-2-targeted siRNA without formulation that was developed to treat acute primary angle closure glaucoma (phase 2) and nonarteritic anterior ischemic optic neuropathy (NAION, phase 3). This molecule is administered via intravitreal injection. It is modified by 2′-OMe in the antisense strand, an L-DNA cytidine nucleotide is located before the last nucleotide at the 3′-end, and the sense strand has an inverted deoxyabasic residue at the 5′-end36 (Fig. 3).