Amide Internucleoside Linkages Suppress MicroRNA-Like Off-Target Activity of Short Interfering RNA

Abstract

RNA interference (RNAi) rapidly matures as a novel therapeutic approach. In this field, chemical modifications have been critical for the clinical success of short interfering RNAs (siRNAs). Notwithstanding the significant advances, achieving robust durability and gene silencing in extrahepatic tissues, as well as reducing off-target effects of siRNA, are areas where chemical modifications can still improve siRNA performance. The present study developed the challenging synthesis of amide-linked guanosine dimers (G_AM1G and G_AM1A) and completed an “amide walk” one by one, systematically replacing every internucleoside phosphate with an amide linkage in a guide strand targeting the PIK3CB gene. Dual luciferase and RT-qPCR assays in HeLa cells showed that, in a model system of unmodified siRNAs, the amide linkage at position 3 (between nucleosides 3 and 4) suppressed cleavage of off-target YY1 and FADD mRNAs similarly to the industry gold standard modification, glycol nucleic acid (GNA). These results suggest that amide linkages in the seed region have a strong potential to improve the specificity of siRNAs by suppressing the microRNA-like off-target activity.

Graphical Abstract

Introduction

RNA interference (RNAi) is a well-established research tool in molecular biology and has also become a novel therapeutic approach.^1–3 Six short interfering RNA (siRNA) drugs are already on the market,^{2, 4} and new RNAi therapeutics are being rapidly developed.^1–3 Chemical modifications of siRNAs have been instrumental in the clinical success of currently approved drugs and continue offering opportunities to address the remaining challenges with delivery to extrahepatic tissues, efficacy, durability, and specificity.² The first clinical success, Patisiran (approved in 2018), used a partial chemical modification strategy that combined some 2′-OMe (Figure 1) modifications in the guide and passenger strands with two deoxythymidines in the 3′-overhangs.⁵ The relatively light modification pattern was sufficient for clinical use because Patisiran was delivered using a lipid nanoparticle formulation. The next generation of approved siRNA drugs used enhanced stabilization chemistry (ESC) that combined complete 2′-OMe and 2′-F modification throughout both guide and passenger strands with two phosphorothioate linkages (Figure 1) at both 3′- and 5′-ends the guide strand. The passenger strand had two 5′-phosphorothioate linkages and was conjugated with a trivalent N-acetylgalactosamine (GalNAc) motif at the 3′-end, which efficiently delivers siRNAs to hepatocytes.²

Figure 1.

Structures of RNA modifications to suppress off-target activity of siRNAs.

The latest addition to siRNA modifications, the ESC+ strategy includes glycol nucleic acid (GNA, Figure 1) at position 7 of the guide strand. In the context of fully 2′-OMe and 2′-F modified siRNAs, the GNA modification suppressed the undesired miRNA-like off-target activity with minimal loss of the desired on-target activity of siRNAs.^{6, 7} The ESC+ allowed Alnylam to proceed with a clinical trial of an siRNA that was previously found to have unacceptable off-target activity.⁷ These results demonstrate that chemical modifications still have significant potential to improve therapeutic siRNAs.

Chemical modifications have been previously explored to suppress the off-target activity of siRNAs. Early studies showed that a single 2′-OMe modification at position 2 of the guide strand mitigated microRNA-like activity.⁸ Later, several groups reported that a single unlocked nucleic acid (UNA, Figure 1) modification at position 7 of the guide strand reduced off-target activity, most likely by strongly destabilizing the partially complementary seed-mRNA duplexes formed by microRNAs.^{9, 10} Similar thermodynamic destabilization was initially proposed for GNA; however, later studies suggested that conformational fit and flexibility may be additional significant contributors to the enhanced specificity of siRNAs having GNA modifications.⁷

Except for the phosphorothioates, phosphate backbone modifications are underutilized in current therapeutic designs and remain an untapped resource for improving the properties of siRNAs. A research team at Wave Life Sciences reported that modifying siRNAs with Sp (but not Rp) phosphoryl guanidine linkages (PN, Figure 1) at positions 3 and 10 of the guide strand enhanced siRNA potency and durability in vivo in mouse liver.¹¹ The benefits were most likely due to improved loading of the PN-modified siRNAs in Ago2. Interestingly, the PN modification performed best in combination with a neighboring 2′-OMe and a 2′-F modification on the 3′-nucleoside, as shown in Figure 1.¹¹ In a most recent study, Yamada et al. reported that extended nucleic acid (exNA) modification having a methylene group inserted between phosphate and 5′-C (Figure 1) increased the enzymatic stability of RNA.¹² When combined with phosphorothioates (Figure 1), two exNA modifications at the 3′-end of a guide strand enhanced siRNA efficacy and durability in mice by improving tissue exposure and accumulation.¹² Taken together, these studies demonstrate that novel backbone modifications can still significantly improve the therapeutic potential of siRNAs.

Our research group has been studying the replacement of phosphates with amide linkages (AM1, Figure 1) to improve the properties of functional RNAs.^{13, 14} Early studies showed that amides were surprisingly good mimics of phosphate internucleoside linkages in A-type RNA helices.^{15, 16} Consequently, amides were well tolerated throughout both strands of siRNAs, except between the first and second nucleosides of the guide strand (position 1) where the RNA’s backbone takes a sharp turn away from the A-type conformation.^{17, 18} The intolerance of amide modification at position 1 could be used to improve the specificity of siRNAs. When placed at position 1 of the passenger strand, the amide modification suppressed its undesired off-target activity.¹⁹ More recently, we demonstrated that amide linkages at some positions of the seed region of the guide strand targeting PIK3CB gene suppressed the microRNA-like off-target activity in a dual luciferase (DLR) assay.²⁰ An amide at position 3 was especially beneficial by completely eliminating off-target cleavage with a relatively small decrease in on-target activity in the DLR assay.²⁰ In the present study, we followed up on these exciting results and completed the “amide walk” throughout the entire guide strand of an siRNA targeting the PIK3CB gene. Using DLR and qRT-PCR assays, we demonstrated that the siRNA modified with the amide linkage at position 3 had improved specificity for endogenous PIK3CB, similar to the industry gold standard GNA-modified siRNAs.

Results and Discussion

In the previous study,²⁰ we scanned most of the PIK3CB siRNA (Figure 2) except positions 4, 5, 16, 17, and 19, that required the amide-linked G_AM1G and G_AM1A dimers. We started the present study with the synthesis of 3′-C homologated carboxylic acid 4 that was further used to prepare the dimeric G_AM1G and G_AM1A phosphoramidites (Scheme 1) and the siRNAs G4, G5, G16, G17, and G19 missing in our original study.²⁰ Guanosine is usually the most challenging nucleoside to modify. In earlier studies, we selected siRNA sequences that did not require guanosine-containing amide-linked dimers.^17–19 In a more recent study on amide-modified CRISPR RNAs,²¹ we synthesized guanosine 5′-amine that allowed the preparation and testing of siRNAs containing amide-linked A_AM1G dimers (G3, G15, and G18 in Figure 2).²⁰ In the present study, for the first time, we report the synthesis of amide-linked G_AM1G and G_AM1A dimers made using guanosine 3′-C carboxylic acid 4 (Scheme 1).

Figure 2.

Sequences of PIK3CB mRNA (on-target), AM1-and GNA- modified siRNA guide strands, YY1 and FADD off-target mRNAs (off-target), and phosphoramidite building blocks to introduce the AM1 and GNA modifications. ‘G’ denotes guide strand targeting PIK3CB, and the number denotes the position of AM1 modification; ‘C’ denotes the GNA-modified control guide strands; ‘p’ denotes phosphodiester and ‘a’ denotes amide internucleoside linkages; the mismatched nucleotides in YY1 and FADD sequences are highlighted in pink. The modified guide strands that were not tested in our earlier study²⁰ are highlighted in bold.

Scheme 1.

Synthesis of amide-linked G_AM1G and G_AM1A phosphoramidite dimers for preparation of amide-modified siRNAs.

Our initial attempts to prepare 4 following the 3′-C radical allylation strategy^16–21 we previously used to synthesize other amide-linked dimers were unsuccessful. The allylation step gave complex reaction mixtures from which the target compound could not be isolated in useful yields. However, we were able to synthesize 4 following the 3′-C homologation strategy, originally developed by Peterson et al.²² (Scheme 1) that we had previously used for the preparation of uridine and adenosine 5′-amino-3′-C carboxylic acids.²³ The synthesis started with Dess–Martin periodinane oxidation of protected guanosine 1 followed by Wittig reaction with a stabilized ylide, benzyl(triphenylphosphoranylidine)acetate (Scheme 1). Deprotection of 5′-O-TBS removed steric hindrance from the β-face and enabled stereoselective hydrogenation of alkene to make the key C-C bond with the correct stereochemistry. As in our previous study,²³ hydrogenation of the alkene led to the concomitant removal of the benzyl protecting group. The crude carboxylic acid was treated with 4-methoxytrityl chloride to give 4 in 32% yield over the three-step reaction sequence. Coupling of 4 with previously reported 5′-amino guanosine²¹ and adenosine¹⁶ derivatives gave the G_AM1G and G_AM1A dimers that were converted in the 3′-phosphoramidite building blocks for synthesizing modified siRNAs.

Amide-modified siRNAs were synthesized, deprotected, and purified as previously described.²⁰ To compare the performance of amide-modified siRNAs with the state of the art GNA-modified siRNAs, we also synthesized control sequences C6, C7, and C8 (Figure 2). The GNA monomers and GNA-modified siRNAs were synthesized following the literature procedures.²⁴ The on- and off-target activities of all newly synthesized siRNAs (bold in Figure 2) were measured using a dual luciferase (DLR) assay as in our previous study.²⁰ Briefly, we used three reporter plasmids (psiCHECK-2 vector) with the sequences of PIK3CB (a single copy), YY1 (four copies), and FADD (four copies) inserted into the 3′-UTR of the Renilla luciferase mRNA, following the methodology previously developed by Beal and co-workers.²⁵ The modified siRNAs were tested at eight or more concentrations to construct full IC₅₀ curves (Figures S17-S19). To illustrate the specificity gains for each modification, the results were normalized by dividing the IC₅₀ values of YY1 and FADD for each guide strand (G0-G20) by the IC₅₀ values of PIK3CB for each guide strand. Figure 3 shows the plot of relative selectivity further normalized to unmodified siRNA (G0). For full equations, see Page S18, in Supporting Information. In this plot, values greater than 1 indicate how many times a given modification increased the selectivity compared to the unmodified siRNA, while values less than 1 indicate a decrease in selectivity.

Figure 3.

Normalized activity ratios show fold change in siRNA specificity compared to unmodified siRNA. # at G2 and G3 indicates that the high selectivity cannot be precisely calculated because of the absence of IC₅₀ values for YY1 and FADD.

Taken together with our previous study, the newly tested G4, G5, G16, G17, and G19 complete a systematic “amide walk” through the guide strand of siRNA targeting the PIK3CB gene. Compared to the best previous results (e.g., G3 and G6), the selectivity improvements at the newly tested positions were modest (~2 fold). Amide-modified G3 siRNA remains the most effective in suppressing the off-target cleavage of YY1 and FADD mRNAs.

To further evaluate the performance of amide-modified G3, we used DLR and RT-qPCR assays to compare the on- and off-target activity of G3 with GNA-modified C6 and C7 that represent state of the art siRNA modifications for suppression of microRNA-like activity.^{6, 7} The GNA-modified C8 did not provide significant specificity enhancements in the initial DLR assays (Figures S17-S19) and was omitted from the RT-qPCR studies. As in our previous study,²⁰ the amide modifications caused relatively little effect on the thermal stability of siRNA duplexes. The T_ms of G4, G5, G16, G17, and G19 were in the range of 64.1 to 65.1 °C, staying close to 64.3 °C of the unmodified G0 (Table S3). Consistent with the previous study on GNA-modified siRNAs, GNA modification caused the T_ms of C6, C7, and C8 to decrease to the range of 56.0 to 60.2 °C (Table S3).

The RT-qPCR assays measured cleavage of endogenous PIK3CB, YY1, and FADD mRNAs in contrast to the DLR assays that measured fluorescence signal from mRNA originating from transfected plasmids. In the RT-qPCR assays in HeLa cells, the IC₅₀ for the unmodified (G0) was 22 pM (Figure 4), which was in good agreement with the IC₅₀ of 25 pM observed in the DLR assays in our previous study.²⁰ In the RT-qPCR assays, all three modified siRNAs showed similar on-target activity with IC₅₀ values of 100 (G3), 200 (C6), and 275 (C7) pM, respectively (Figure 4), which were notably higher than the IC₅₀ values of 40 (G3), 30 (C6), and 150 (C7) pM obtained in the DLR assays (Figure S20, Table S2). It should be noted that the GNA-modified C6 and C7 did not reach complete on-target silencing (Figure 4); instead, their IC₅₀ curves plateaued at 0.6 and 0.7, respectively. In contrast, the amide-modified G3 and the unmodified G0 siRNAs silenced the PIK3CB gene down to 0.3 and 0.1, respectively (Figure 4). Thus, the RT-qPCR results showed that both amide and GNA modifications decreased the on-target activity, but the amide-modified G3 was somewhat less impaired than the GNA-modified siRNAs.

Figure 4.

Comparison of IC₅₀ curves of RT-qPCR assays of silencing of endogenous PIK3CB, YY1, and FADD by G0 (unmodified), G3, C6, and C7. Values in parentheses ()* do not represent true specificity because the silencing did not reach below 0.5; N/A – IC50 was not calculated because the silencing did not reach below 0.75.

In the RT-qPCR assays, both amide- and GNA-modified siRNAs suppressed the off-target cleavage of YY1 and FADD mRNAs (Figure 4). The effect was notable in both increased IC₅₀ and decreased maximum silencing by the modified siRNAs. GNA-modified C6 showed the highest specificity with no detectable cleavage of YY1 and very little cleavage of FADD at the highest siRNA concentrations. G3 and C7 showed similar suppression of off-target activity. The IC₅₀ curves plateaued out for G3 at 0.6 (YY1) and 0.7 (FADD) and for C7 at 0.7 (YY1) and 0.5 (FADD), which compared favorably with the maximum silencing of ~0.3 for G0. Because of the significant suppression of maximum silencing, the increase in specificity is higher than suggested by modest changes in IC₅₀.

Overall, the RT-qPCR results were consistent with the DLR results (Figure S20), except for C6. In the DLR assays, the GNA-modified C6 was more active than G3 or C7 but also had the highest off-target cleavage of YY1 and FADD with IC₅₀ of ~700 and ~800, respectively, and maximum silencing reaching 0.45 (Figure S20). It should be noted that the silencing results obtained by RT-qPCR and DLR cannot be directly compared because the reporter plasmids used in DLR had four copies of YY1 and FADD target sequences inserted in the psiCHECK-2 vector. Taken together, the RT-qPCR and DLR results demonstrated that, in the PIK3CB model system, the amide-modified G3 offered similar benefits of increased siRNA specificity as GNA-modified C6 and C7. In the present model system, the amide-modified siRNA had a slight advantage of higher on-target activity, as illustrated by lower IC₅₀ and higher maximum silencing in Figure 4.

Finally, we used RNA-Seq to compare the genome-wide specificity of G3, the best amide-modified siRNA, and the industry gold standard GNA-modified C7 (Figure 5). The results show that G3 (Figure 5B) and C7 (Figure 5C) provided similar improvements in siRNA specificity compared to unmodified G0 (Figure 5A). Overall, G3 induced somewhat smaller changes in the HeLa transcriptome than C7 (Figures 5D and S21), supporting the notion that the amide modification offers a similar increase in siRNA specificity than the well-established GNA modification.

Figure 5.

Volcano plots representing global gene expression changes in HeLa cells after 20 hours post transfection of the respective siRNA. (A) Unmodified G0/Nontargeting (NT) siRNA transfection; (B) Amide-modified G3/NT siRNA transfection, and (C) GNA-modified C7/NT transfection. The targeted gene, PIK3CB, is highlighted in red, with blue points indicating p≤ 0.05 and gray points indicating p> 0.05. The p-value for fold change was calculated in Deseq2 using the Wald test.²⁶ (D) Vann diagram representing the number of genes that are exclusively differentially expressed in amide-modified G3 (313, blue), GNA-modified C7 (581, pink), and unmodified G0 (1762, green). The overlapping areas represent genes that are differentially expressed in two or all three treatments (e.g., the central 784 genes are differentially expressed in all G3, C7 and G0 treatments).

In the present study, we completed a systematic “amide walk” (Figure 2), replacing every phosphate of the guide stand with an AM1 amide linkage (Figure 1). We did not test G1 because previous studies showed that amide modification at the first position is not compatible with Ago2 loading.¹⁹ The results of DLR assays showed that the AM1 amide linkages at some positions in the seed region of siRNA targeting the PIK3CB mRNA strongly suppressed microRNA-like off-target cleavage of the YY1 and FADD mRNAs (Figure 3). In the DLR assays, amide modification at position 3 was especially effective, but positions 2, 6, 7, and 12 also offered notable specificity improvements.

The structure-function relationships of these improvements are not entirely clear, but some hypotheses may be suggested. Crystal structures of human Argonaute-2 (Ago2) loaded with a guide strand show that the seed nucleotides 2–5 are preorganized in an A-type helical conformation favorable for target recognition.^{27, 28} The amides fit very well in the A-type helical conformation, and^{15, 16} therefore, any improvements at positions 2 and 3 are most likely related to differences in hydrogen bonding interactions of Ago2 and phosphate or amide linkages. At the end of the seed region, Ago2 kinks the guide strand by inserting helix 7 between guide nucleotides 6 and 7;^{27, 28} insertion of two arginine kinks the guide again between nucleotides 9 and 10.²⁸ Pairing of seed nucleotides 2–5 act as the initial target recognition that triggers movement of helix 7 and PAZ domain as a single rigid body, revealing the kink and opening the Ago2 cleft for further target engagement.²⁹ Consequently, the specificity of target recognition is sensitive to chemical modifications of nucleotides 6 and 7 of the guide strand. It was proposed that GNA at position 7 mimics the kinked backbone conformation and destabilizes base pairing with the mRNA target, which collectively reduces the ability of Ago2 to accommodate imperfectly paired targets.⁷ In a similar manner, the conformationally rigid amide linkages may be better accommodated in the perfectly matched siRNA complexes compared to the more flexible and partially base-paired microRNA-like complexes. The fine details and reasons for such preferences will require additional structural studies.

It should be noted that in the present study, we compared amide and GNA as isolated modifications in an otherwise unmodified siRNA model system. In contrast, the clinically relevant GNA-modified siRNAs were developed as fully 2′-OMe and 2′-F modified compounds.⁷ It is reasonable to expect that GNA modification will perform better in the context of 2′-OMe and 2′-F modified siRNAs because the RNA duplex destabilization caused by GNA will be somewhat compensated by the stabilizing 2′-OMe and 2′-F modifications. To fully evaluate the potential of amide linkages, the specificity improvements by AM1 (including comparison with GNA) will need to be assessed in the context of fully 2′-OMe and 2′-F modified clinically relevant siRNA sequences.

As discussed in the Introduction, chemical modifications have tremendously impacted the clinical success of approved siRNA drugs. However, future progress is limited by challenges in achieving robust durability, delivery to extrahepatic tissues, and reducing off-target effects and toxicity of siRNA drugs. More chemistry is needed to address these challenges and accelerate the advancement of nucleic acid therapeutics. Backbone modifications, such as amides and others discussed in the Introduction, are well-positioned to solve the remaining problems in the RNA medicine field.

Conclusions

The best amide-modified siRNA, G3 was compared with the state of the art GNA-modified siRNAs, C6 and C7, using DLR assays with reporter plasmids and RT-qPCR assays for cleavage of endogenous PIK3CB, YY1, and FADD mRNAs. Overall, the results of DLR and RT-qPCR assays were similar, but notable differences were observed as the DLR assay (Figure S20) suggested that the amide-modified G3 was the most specific while the RT-qPCR assay showed the highest specificity for the GNA-modified C6 (Figure 4). Taken together, the results showed that in our model system, all modified siRNAs increased the PIK3CB siRNA specificity to a similar extent. In conclusion, the present study demonstrates that amide internucleoside linkages at specific positions in the guide strand have a strong potential to improve the specificity of siRNAs. While G3 appeared to be the most effective for PIK3CB, it is conceivable that the optimal modification pattern may be different for other siRNA sequences and would need to be individually optimized.

Supplementary Material

Supporting Information

General experimental procedures, synthesis, purification, and MS characterization of G_AM1G and G_AM1A phosphoramidites and amide-modified siRNAs; experimental details and results of dual luciferase, RT-qPCR, UV melting, and RNA-Seq assays; NMR spectra of new compounds.

The Supporting Information is available free of charge on the ACS Publications website.