Since the discovery of RNA interference (RNAi), substantial efforts have been made to develop RNAi therapeutics for human disease. However, despite ~50 clinical trials involving short interfering RNAs (siRNAs), only a single RNAi therapeutics has been approved for clinical use. This lack of success is largely due to a common problem: namely, ineffective endosomal escape. Achieving safe and effective cytosolic delivery of siRNAs remains the key technological obstacle in the design of RNAi therapeutics and solving this challenge has the potential to improve treatments for a diversity of diseases, including various cancers, infections, and genetic disorders. The ubiquitously expressed endo-lysosomal protein SIDT2 specifically recognizes and transports viral double-stranded RNA (dsRNA) from endosomes into the cytosol to mediate innate immune recognition. Here, we show that SIDT2 also interacts with siRNA, suggesting that SIDT2 similarly mediates endosomal escape of siRNA. Moreover, we find that this interaction with SIDT2 is disrupted by bulky chemical modifications of the siRNA (e.g. phosphothiorate, 2′-O-methyl) that are commonly used to improve RNA therapeutic stability, immunogenicity, and off-target effects. In this way, such modifications might be inadvertently contributing to ineffective endosomal escape, and we suggest that the success of future RNAi-based treatments would be improved if SIDT2-mediated endosomal escape is optimized.

RNAi is a fundamental biological process that causes sequence-specific, post-transcriptional gene silencing. Mechanistically, RNAi involves small molecules of double-stranded RNA (dsRNA), known as siRNAs, that bind complementary mRNAs and load them into the RNA-induced silencing complex (RISC) for destruction within the cytoplasm^[1]. There is much excitement surrounding the potential for RNAi to treat human disease since RNAi can in theory target any gene or mutant allele^[2]. The pharmaceutical industry has consequently invested billions of dollars to develop RNAi therapeutics over the past decade, and ~50 clinical trials have been completed for a wide variety of cancers, viral infections and genetic diseases^[3].

Despite their promise, only a single RNAi therapeutics have been approved for clinical use (to treat hereditary amyloidosis)^[4]. This is mainly due to inefficient delivery, which occurs for several reasons^[2]. First, siRNAs administered into the circulation are degraded by serum nucleases. Second, siRNAs are rapidly excreted via the kidney and must be specifically targeted to the desired tissue. Third, even with successful targeting, siRNAs must enter the cytosol for loading into the RISC. Considerable progress has been made in the first two areas, mostly by chemically modifying the siRNAs and/or conjugating them to specific delivery agents. The third problem remains unsolved since the vast majority of siRNAs become trapped within the endocytic compartment following uptake into target cells^[5][6]. In this way, transporting siRNAs from endosomes into the cytoplasm represents a major rate-limiting step, and achieving effective endosomal escape is critical to solving the RNAi delivery challenge^[5][7].

One organism in which RNAi is delivered extremely efficiently is the nematode worm, C. elegans, where organism-wide gene silencing can be triggered via injection of dsRNA into a single cell^[8]. The SID-1 protein is absolutely essential for this systemic delivery to occur^[9]. SID-1 is a large, multi-pass transmembrane protein^[10] that localizes to the endosomal compartment^[11] where it facilitates the release of dsRNA into the cytoplasm following endocytosis^[10][12]. Vertebrates all share a common SID-1 ortholog known as SIDT2. Like SID-1, SIDT2 is also capable of binding long dsRNA^[13] and has recently been shown to be important for trafficking endocytosed viral dsRNA into the cytoplasm to mediate antiviral immunity^[14]. However, whether SIDT2 can also interact with siRNAs in order to mediate endosomal escape remains unknown.

In this study, we investigated whether SIDT2 is capable of interacting with siRNA and whether various common chemical modifications currently being tested in siRNA clinical trials affect the ability of these siRNAs to interact with SIDT2.

To assess whether SIDT2 is able to interact with siRNA, we adapted a method combining Fluorescence Lifetime Imaging (FLIM) with Förster Resonance Energy Transfer (FRET) used in a previous study^[14] to assess the ability of SIDT2 to interact with long dsRNA in the form of poly(I:C). Unlike long dsRNA, we found that naked siRNAs showed a poor cellular uptake. To address this, we complexed fluorescein-conjugated siRNAs with a transfection reagent to facilitate endocytosis prior to the treatment of mouse embryonic fibroblasts (MEFs) that stably expressed SIDT2-mCherry. Using this approach, we observed co-localization of fluorescein-conjugated siRNAs with SIDT2-mCherry, consistent with efficient internalization via endocytosis (Fig. 1A). We subsequently performed FLIM-FRET analysis 16 h post-treatment. Notably, we observed a significant reduction in fluorescence lifetime for siRNA-fluorescein in the presence of SIDT2-mCherry (Fig. 1B) indicating a likely molecular interaction between SIDT2 and siRNA. To test whether this interaction was specific to siRNA, we performed the sam­­e experiment using fluorescein-conjugated dsDNA and consistent with the previous reports^[14], we did not observe a reduction in fluorescence lifetime dsDNA-fluorescein and SIDT2-mCherry (Fig. 1C), suggesting that SIDT2 can distinguish between siRNA and dsDNA as has previously been shown for C. elegans SID-1^[15].

Although unmodified siRNAs work well to silence gene activity in vitro, siRNAs are usually modified prior to delivery in vivo^[2][16] to counteract ribonuclease degradation, immunogenicity, off-target effects and unfavorable pharmacokinetics. The types of modifications that have been utilized are extensive, and include chemical changes to the ribose sugar, phosphodiester backbone, and nitrogenous bases (Fig. 1D). How such modifications affect transport by SIDT2 is currently unknown, but RNA transport by C. elegans SID-1 has been shown to be impaired when modifications are made to the ribose sugar^[12].

Structural analysis of SID-1 and it's mammalian orthologs^[1] suggest that these proteins likely function as multimeric transmembrane channels that transport dsRNA via a central pore (Fig. 1E). Given this, we hypothesized that while unmodified siRNAs can thread this pore, bulky chemical modifications to the ribose sugar or phosphodiester backbone will sterically hinder transit through the pore (Fig. 1F-G). To test this, we again used FLIM-FRET to determine whether common siRNA modifications previously used in clinical trials affect SIDT2 interaction. Notably, we find that bulky modifications such as phosphothiorate (PS) and 2′-O-methyl (2′-OMe) substitutions result in loss of interaction with SIDT2, while more compact modifications such as 2′-O-fluoro (2′-F) substitution maintain the interaction (Fig. 1H). This is consistent with our hypothesis regarding steric hindrance and provides evidence that bulky siRNA modifications are detrimental to SIDT2 interaction. Moreover, these results might help to also explain why gene knockdown is less efficient when PS and 2′-OMe groups are added to unmodified siRNAs^[17][18].

As highlighted in a recent Nature Biotechnology review, “getting RNA-based therapeutics out of the endosome and into the cytoplasm in a non-toxic manner is the key technological problem to solve before we can tap into the full potential of RNA-based therapeutics”^[19]. Here, we find that the endosomal RNA transporter SIDT2 interacts with internalized siRNA, but is sensitive to structural changes in the native siRNA molecule. This observation may in part explain the widespread failure to effectively deliver siRNA-based therapeutics thus far, since clinical trials have predominantly utilized bulky chemical modifications (e.g. PS and 2′-OMe)^[5] and may have thus inadvertently prevented SIDT2-mediated endosomal escape. Looking ahead, our data caution against the use of bulky modifications such as PS and 2′-OMe in the design of siRNA therapeutics and suggests that more compact modifications that do not impair interaction with SIDT2 such as 2′-F (or avoidance of chemical modifications altogether) might be more effective in future clinical trials.

The FLIM-FRET experiments conducted in this paper relied on the overexpression of SIDT2. Future studies involving siRNAs are required to further characterize SIDT2 transporter function at an endogenous level and could include the use of cells lacking SIDT2 function.

Moreover, interactions between SIDT2 and siRNAs were inferred here based solely on FLIM-FRET studies. It would, therefore, be helpful to confirm our observations using an independent experimental method, such as a pull-down assay.

Finally, the ultimate functional readout of successful siRNA delivery into the cytosol is gene knockdown. While we were able to identify siRNA modifications that ablate SIDT2 interaction via FLIM-FRET, further experiments are required to directly demonstrate that these modifications result in impaired gene silencing via SIDT2 and thus have a functional impact.

Although we propose that chemical modifications, such as PS and 2′-OMe, impair the ability of siRNAs to interact with SIDT2 due to increased steric hindrance, it is possible that these modifications might also introduce changes in molecular charge that influence SIDT2 interaction. Recently, SIDT2 was identified as having an arginine-rich motif that is important for RNA binding^[20], presumably as a result of the attraction between the positively charged arginine residues and the negatively charged phosphate groups of the siRNA. With that in mind, it is conceivable that the 2'-F modification may promote binding to the arginine-rich motif, while the 2′-OMe may impair binding relative to the native 2'-OH group.

In the present study, we provide evidence that the endo-lysosomal protein SIDT2 is able to interact with siRNA, suggesting that it may play a role in mediating endosomal escape of siRNA into the cytoplasm. Using FLIM-FRET, we were able to screen for siRNA modifications that appear to either impair or retain interaction with SIDT2. Building upon these results, future experiments could be conducted not only to test the ability of other commonly used modifications (e.g. locked nucleic acid) to interact with SIDT2 but also to identify novel modifications that enhance SIDT2 interaction.

FLIM-FRET analysis

SIDT2-mCherry transduced MEF cells were treated with 1 μg/mL doxycycline for 32 h to induce SIDT2-mCherry expression and transfected in Ibidi 8 well chamber µ-slides (Martinsried, Germany), with 100 nM siRNA-fluorescein (anti-luciferase (GL3), 100 nM chemically modified siRNA-AlexaFluor-488 or 100 nM dsDNA-fluorescein complexed with FuGENE HD for 1 h at 37°C. Cells were then washed and fixed in 4% paraformaldehyde for 10 min on ice. Fluorescein conjugated siRNA and AlexaFluor-488 conjugated modified siRNA was purchased from Bioneer Pacific and corresponded to the following sequence: 5’ CUU ACG CUG AGU ACU UCG A(TT), 5’ UCG AAG UAC UCA GCG UAA G(TT). dsDNA-fluorescein was purchased from Bioneer Pacific and corresponded to the following sequence: 5’ CTT ACG CTG AGT ACT TCG A(TT), 5’ GCT AAG TAC TCA GCG TAA G(TT). FLIM data were recorded using an Olympus FV1000 microscope equipped with a PicoHarp300 FLIM extension and a 485 nm pulsed laser diode from PicoQuant (Berlin, Germany). Cells were first imaged for green and red fluorescence by confocal microscopy using the Olympus FV1000 system with a 60X oil immersion objective to verify the presence of both donor (fluorescein or AlexaFluor-488) and acceptor (mCherry) dyes. Subsequently, corresponding FLIM images of green fluorescence were recorded using the PicoHarp extension. Pixel integration time for FLIM images was kept at 40 ms per pixel and fluorescence lifetime histograms were accumulated to at least 10,000 counts to ensure sufficient statistics for FRET FLIM analysis. Photon count rates were kept below 5% of the laser repetition rate to prevent pileup. FLIM-FRET analysis was performed using the SymPhoTime 64 software (PicoQuant). siRNA-fluorescein/AlexaFluor-488 or dsDNA-fluorescein positive compartments of individual cells were chosen as regions of interest (ROIs), then the fluorescence lifetime decay of each ROI was deconvolved with the measured instrument response function and fitted with a biexponential decay. The amplitude weighted average lifetime was extracted from each fit and averaged over all values of one sample condition. p-values were determined to assess the significance of donor lifetime changes in the absence and presence of the SIDT2-mCherry acceptor. For experiments with modified siRNAs, 2’-F siRNA, PS-siRNA, 2’OMe-siRNA, and 2’OMe-siRNA were purchased from Bioneer Pacific and used to treat cells, and FLIM-FRET analyses were conducted as above.

RNA molecular modeling

RNA and DNA models were generated with the 3DNA web interface^[21] using the GL3 Luciferase siRNA sequence. These models were then modified using the PyMOL Molecular Graphics System (Version 2.0 Schrödinger, LLC) to replace modified atoms for different RNA modifications, and the sculpt iterate command used to account for energy minimization around these modifications.

This work was supported by the CASS Foundation, Royal Australasian College of Physicians, and Murdoch Children's Research Institute. KCP was supported by the Royal Children’s Hospital Foundation.

The authors would like to thank Tiff Walsh, Craig Hunter, and Angus Johnstone for helpful advice.

The authors declare no conflicts of interest.

Not applicable.

No fraudulence is committed in performing these experiments or during processing of the data. We understand that in the case of fraudulence, the study can be retracted by ScienceMatters.