Phosphorodiamidate morpholino oligomers (PMOs) were purchased from Gene Tools LLC (Philomath, OR, United States), 2′ O-methyl phosphorothioate AOs were synthesised by TriLink BioTechnologies (San Diego CA, United States) and DNA phosphodiester AOs were synthesized by Integrated DNA Technologies (Singapore). The following sequences were evaluated; an AO encompassing the SMN intron 7 ISS-N1 target, but with a longer sequence SMN7D(-10-29) (5′AUU​CAC​UUU​CAU​AAU​GCU​GG 3′); the Gene Tools standard control oligomer, only known to be biologically active in reticulocytes carrying a splice mutation in the human beta-globin pre-mRNA, as an unrelated sham control AO sequence (5′CCU​CUU​ACC​UCA​GUU​ACA​AUU​UAU​A 3′), and identified as ‘control AO’ throughout this study; SMN7A(+13+32) (5′CAC​CUU​CCU​UCU​UUU​UGA​UU 3′), designed to induce SMN exon 7 skipping (Flynn et al., 2021) and a SMN sense AO (5′ CCA​GCA​UUA​UGA​AAG​UGA​AU 3′) complementary to SMN7D(-10-29).

The 2′ O-methyl phosphorothioate AOs were transfected into dermal fibroblasts as lipoplexes using 3 µl of Lipofectamine 3000 (Life Technologies, Melbourne, Australia) per 1 ml of OptiMEM, according to the manufacturer’s protocol. The transfection mix was applied to cells seeded at 10,000 per cover slip for immunofluorescence and 15,000 cells per well in 24 well plates for RNA extraction and incubated (37°C) for 24 h prior to RNA and protein analysis. SMA patient fibroblasts were used for the experiments shown in Figures 8D–F (Coriell Cell Repositories, GM03813) and fibroblasts from our in-house biobank, obtained from a healthy volunteer, with informed consent (Murdoch University Human Research Ethics Committee Approvals #2013/156 and 2017/101) were used for all other experiments, except live-cell imaging (see below).

The PMOs were delivered to cells either uncomplexed (10 µM) and incubated for 72 h, annealed to a complementary sense DNA oligonucleotide (phosphodiester leash) and transfected as a lipoplex (100 nM) using Lipofectamine 3000 and incubated for 24 h, or by electroporation using a Neon Electroporation System (ThermoFisher Scientific, Melbourne, Australia) according to the manufacturer’s instructions. PMOs were delivered at 50 or 10 μM, using the following parameters: 1,650 V, 10 ms, 3 pulses, and the cells incubated for 24 h.

RNA was extracted using the MagMAX-96 Total RNA Isolation Kit, including DNase treatment (Life Technologies), according to the manufacturer’s instructions. RT-PCRs were performed using the One-step Superscript III RT-PCR kit with Platinum Taq polymerase (Life Technologies) according to the manufacturer’s instructions. All primer sequences and PCR conditions used in this study are detailed in Supplementary Table S1.

cDNA was synthesised from (∼500 ng) total RNA extracted from treated and untreated cell cultures using the Superscript IV first-strand synthesis system (Life Technologies) as per the manufacturer’s instructions. Prior to qPCR amplification, cDNA was diluted in RNase/DNase free water (1:5). The qPCR reactions were performed using Fast SYBR™ Green Master Mix (ThermoFisher Scientific, Melbourne, Australia) and primers for ribosomal RNA subunits 5.8S (100% primer efficiency), 18S (95% primer efficiency), 45S (100% primer efficiency) and housekeeping Tata box protein (TBP;100% primer efficiency) and beta Tubulin (TUBB;94% primer efficiency) transcripts. All primer sequences and cycling conditions used in this study are detailed in Supplementary Table S1. Transcript abundance was measured using the CFX384 Touch™ Real Time PCR detection system (Bio-Rad Laboratories Pty., Ltd., Gladesville, Australia). The relative expression of each ribosomal RNA subunit to TBP and TUBB mRNA was calculated using the ΔΔCT method and presented as a fold change compared to untreated cells.

Approximately 10,000 fibroblasts were seeded onto 22 × 22 mm coverslips in 6-well plates and incubated for 24 h, prior to transfection. Following transfection, the cells were fixed using acetone:methanol (1:1) on ice for 4 min and air-dried. Fixed cells were washed in PBS containing 1% Triton X-100 to permeabilise the nuclear membrane, and then in PBS to remove excess Triton X-100. Primary antibodies were diluted in PBS containing 0.05% Tween20 and applied to cells for 1 h at room temperature. All antibody details and staining conditions are listed in Supplementary Table S2. Primary antibodies were detected using AlexaFluor488 anti-mouse (cat no. A11001), anti-rabbit (cat. No A11008) or AlexaFluor568 anti-mouse (cat. No A11004) or anti-rabbit (cat. No A11011) (1:400) after incubation for 1 h at room temperature, and counterstained with Hoechst 33342 (Sigma-Aldrich) for nuclei detection (1 mg/ml, diluted 1:125).

Cell lysates were prepared with 125 mM Tris/HCl pH 6.8, 15% SDS, 10% Glycerol, 1.25 µM PMSF (Sigma-Aldrich, NSW, Australia) and 1× protease inhibitor cocktail (Sigma-Aldrich) and sonicated 6 times (1 s pulses) before adding bromophenol blue (0.004%) and dithiothreitol (2.5 mM). Samples were heated at 94°C for 5 min, cooled on ice and centrifuged at 14,000 × g for 2 min before loading onto the gel. Total protein (10 µg), measured by BCA, was loaded per sample on a NuPAGE Novex 4–12% BIS/Tris gel (Life Technologies) and separated at 200 V for 55 min. Proteins were transferred onto a Pall Fluorotrans polyvinylidene fluoride (PVDF) membrane at 350 mA for 2 h. Following blocking for 1 h, the membrane was incubated in 5% skim milk powder in 1× TBST containing the primary antibody diluted as shown in Supplementary Table S2. Immunodetection was performed using the Western Breeze Chemiluminescent Immunodetection System (Life Technologies) according to the manufacturer’s instructions. Immunodetection of rabbit conjugated antibodies was performed using a HRP goat anti-rabbit secondary antibody (1:10,000, Dako, Denmark) and the Immobilon Western Chemiluminescent HRP substrate (Merck Millipore, VIC, Australia). Western blot images were captured on a Vilber Lourmat Fusion FX system using Fusion software and Bio- 1D software was used for image analysis.

U2OS cells, previously modified to tag the endogenous SFPQ gene with GFP were used for live-cell imaging (Li et al., 2017). Cells were seeded at 1.5 × 10⁵ cells per well in a 12-well plate, 24 h before transfection, with high glucose DMEM (Life Technologies) supplemented with 10% fetal calf serum and 1% penicillin-streptomycin. On the following day, cells in each well were transfected with 100 or 150 nM 2′ O-methyl phosphorothioate AOs complexed with Lipofectamine RNAiMAX (Life Technologies) following the manufacturer’s instructions. Immediately after transfection, the plate was transferred to an Incucyte S3 (Essen Bioscience) for live-cell imaging at 1 h intervals under 10 × optical zoom. SFPQ-GFP and activated caspase-3/7 were visualised under standard green and red channels. Incucyte Caspase-3/7 Red Apoptosis Assay Reagent (Essen Bioscience, Cat No. 4704) was used to reveal cell apoptosis events.

SIM imaging was performed using an N-SIM microscope (Nikon Corporation, Tokyo, Japan), with SR Apochromat TIRF 100 × 1.49 NA oil immersion objective. Spherical aberration was reduced using the Ti₂ automated correction collar at the beginning of the imaging session. Images were acquired using 405 nm, 488 and 561 nm lasers, with stacks of step size 0.12 µm (with top and bottom of samples determined visually), using 3D-SIM mode. Images were reconstructed with NIS Elements software (Nikon Corporation, Tokyo, Japan).

Following transfection, cells were washed in PBS and fixed in cold 2.5% phosphate buffered glutaraldehyde overnight. The fixed cells were scraped and centrifuged, then embedded in 4% agarose. The pellets were processed using a Leica tissue processor, moving through 1% aqueous osmium tetroxide, increasing graded alcohols, propylene oxide, propylene oxide/araldite mix, and finally pure araldite resin. The osmicated cells, surrounded by resin in a beem capsule were polymerised overnight at 80°C to form hard blocks. Semi-thin (0.5 micron) sections were taken from the blocks using glass knives, on an RMC ultratome, and stained with methylene blue (0.1% aqueous Methylene Blue with 0.1% Borax) to visualise available cells. Ultrathin sections were then cut from selected blocks, using a Diatome diamond knife, at approximately 95 nm thickness and mounted on copper mesh grids and stained with uranyl acetate (5% uranyl acetate solution in 5% aqueous acetic acid) and Reynold’s lead citrate. The grids were viewed using a JEOL 1400 TEM, at 80 kV and images captured by an 11-megapixel GATAN digital camera at varying magnifications.

RNA quality was confirmed using a Bioanalyser (Perkin Elmar, MA, United States) prior to RNAseq. Samples were sent to the Australian Genome Research Facility (AGRF, Perth and Melbourne, Australia) for whole transcriptome library preparation using the TruSeq Stranded Total RNA Library Prep Kit (Illumina, CA, United States) and ribosomal RNA depletion with the Ribo-Zero-Gold kit (Illumina, CA, United States). Sequencing was performed using an Illumina HiSeq 2500 (Illumina, CA, United States) to generate 100 base pair single end reads, resulting in an average 25 million reads per sample. Raw sequencing files were quality checked using FastQC (0.11.7), with all files passing. No adapter contamination was identified, and reads were not trimmed. Transcript quantification was performed with salmon (0.8.2), using an index constructed from the Ensembl GRCh38.93 annotation. Transcript abundance was summarised to gene-level counts and imported into R using tximport (1.4.0). Differential expression analysis was performed using DESeq2 (1.16.1) and the default parameters (alpha = 0.1). The heatmap of differentially expressed genes (p _adj < 0.1) was constructed using the heatmap.2 function from R package gplots (3.0.1). Gene ontology analysis was performed using GSEA (3.0) with 1000 gene set permutations. The gene ontology network was constructed using the Enrichment Map plugin for Cytoscape (3.5.1), with cut-offs: p < 0.05, FDR <0.1 and edge >0.3. The RNAseq data is available at https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE121713 (accession number: GSE121713).

For all PCR and western blot data, averages were used, with error bars displaying the standard error of the mean. An unpaired t test was used to determine significance where the Fisher’s exact test showed equal variance between samples. Significance was determined at α = 0.05.

There is a significant body of work showing that phosphorothioate AOs can interact non-specifically with intracellular proteins (Crooke et al., 2020) and recruit paraspeckle proteins, forming nuclear inclusions when transfected into Hela cells (Shen W. et al., 2014; Shen et al., 2019). However, the dynamics of the formation of these structures, the long-term consequences on cell physiology, and whether the effects are recapitulated by transfection of 2′ O-fully modified phosphorothioate AOs into cells is unknown. In this study, we largely focused on two 2′ O-methyl fully modified phosphorothioate AOs, one that promotes SMN2 exon 7 retention during splicing by targeting the ISS-N1 intronic silencer motif (Singh et al., 2006) (termed ‘SMN7D(-10-29)’) and the other an oligomer sequence reported by Gene Tools LLC and widely used as a transfection control (termed ‘control AO’). Additional AO sequences were also tested in some experiments, as indicated below.

The bulk of studies on gapmer phosphorothioate AO-induced inclusions has been performed in cancer and other immortalized cell lines. We utilized primary human fibroblasts, widely used as patient-specific models, for the majority of our experiments. We transfected fibroblasts with 2′ O-methyl fully modified phosphorothioate AOs, fixed the cells 24 h following transfection, and immunostained the cells for the DBHS protein NONO. We observed that 50–70% of cells transfected with fully modified 2′ O-methyl phosphorothioate AOs, but not the un-transfected or transfection reagent only (Lipofectamine) treated control cells, developed large nuclear inclusions, identified by aberrant NONO staining (Figures 1A,B). Off-target binding of gapmer AOs by various paraspeckle proteins and the formation of nuclear inclusions, reported by Shen et al. (2015) was found to be phosphorothioate backbone dependent; the extent of the phosphorothioate backbone interaction with DBHS proteins can, in part, be dependent on further AO chemistry modifications. Consistent with the findings by Shen W. et al. (2014), we observed no nuclear inclusions containing NONO in cells transfected with a DNA ribose or 2′ O-methyl modified phosphodiester AO but did observe nuclear inclusions in cells transfected with a DNA AO on a phosphorothioate backbone, albeit with half as many cells displaying nuclear inclusions, when compared to the 2′ O-methyl modified equivalent, confirming that interaction with DBHS proteins is dependent on the phosphorothioate modification, and exacerbated by the 2′ O-methyl modification (Figures 1Ai–iv; Supplementary Figure S1). We next explored any correlation between cell death and the formation of 2′ O-methyl phosphorothioate AO-induced nuclear inclusions and observed that transfection with AOs induced nuclear inclusions that correlated with reduced fibroblast survival (Figures 1B,C).

To determine the distribution of the phosphorothioate AO within the cell and within the nuclear inclusions, we used a fluorescently labelled FITC-AO to visualize its distribution and interaction with NONO (Figure 2A). Consistent with reports by Shen W. et al. (2014), the phosphorothioate AO did indeed co-localize with NONO within the nucleus, confirming the phosphorothioate AO as the scaffold for NONO positive nuclear inclusions. We also observed that once formed, the inclusions appeared to be highly stable, and remained evident on the culture substrate following cell death and nuclear fragmentation and blebbing (Figures 2Bi,ii).

Others have shown that different modified gapmer phosphorothioate AO sequences induce variable toxicities in cell lines and that this effect is species independent (Shen et al., 2019). To further explore whether AO-induced nuclear inclusions are sequence independent, ninety different 2′ O-methyl phosphorothioate AOs (prepared for evaluation in other studies, targeting structural gene transcripts, transcription factors, splicing factors and enzymes), were transfected into fibroblasts at 100 nM and the cells were immunostained for SFPQ (Supplementary Table S3). We found that all AOs tested altered SFPQ distribution, and all but one AO formed nuclear inclusions by 24 h following AO transfection, irrespective of length (18–30 bases) or nucleotide sequence composition (Supplementary Table S3).

The nuclear inclusions appeared as either long ‘filaments’, punctate foci, or both (Figure 2Ci). Others reported cytoplasmic SFPQ aggregates induced by gapmer phosphorothioate AOs, (Liang et al., 2015), in addition to nuclear inclusions; also evident in our cells transfected with a variety of fully modified 2′ O-methyl phosphorothioate AO sequences (Figure 2Cii). While most AOs induced nuclear inclusions containing SFPQ, 20 of the AO sequences drew SFPQ from the nucleus to localise within the cytoplasm. Intriguingly, 18 of these AOs, including nucleotide equivalent scrambled sequences with no predicted target binding, had a guanine content of greater than 30%, suggesting sequence specific-effects on protein distribution, with guanine content potentially a driver of DBHS sequestration within the cytoplasm. We further report that one AO targeted SFPQ to the nuclear envelope in transfected cells (Figure 2Ciii). Thus, optimal transfection efficiency of any 2′ O-methyl phosphorothioate AO sequence over 18 bases in length is likely to induce either nuclear or cytoplasmic DBHS protein-enriched inclusions.

Although obvious nuclear inclusions were seen in numerous cells transfected at an AO concentration of 100 nM (Figures 1, 2A–C), we also detected evidence of some nuclear inclusions in fibroblasts transfected with AOs at concentrations as low as 12.5 nM, within 24 h following transfection (Figure 2D). Of note, the untreated cells show diffuse nucleoplasmic NONO signal, as well as the expected punctate NONO inside endogenous paraspeckle nuclear bodies (Fox et al., 2018) (e.g., Figure 2Dii). Nuclear inclusions were also observed following 2′ O-methyl phosphorothioate AO transfection into normal human myoblasts, transformed mouse H2K myoblasts, and the neuroblastoma SH-SY5Y cell line (Supplementary Figure S2).

To assess the dynamics of nuclear inclusion formation, we transfected 2′ O-methyl phosphorothioate AOs as lipoplexes (100 nM) into U2OS cells that had been genome engineered to express GFP-labelled SFPQ (Li et al., 2017) and performed live cell imaging (Figure 2E). We observed formation of nuclear inclusions within 5 h, and these intensified over a period of 12 h, by which time all of the microscopically identifiable GFP-SFPQ signal within the nucleus became sequestered into nuclear structures. At the point at which only nuclear inclusions were evident, these coalesced into a smaller number of discrete filaments (Figure 2Ei). Thus, nuclear inclusions form rapidly upon 2′ O-methyl phosphorothioate AO transfection. In a larger field view of U2OS cells transfected with the 2′ O-methyl phosphorothioate AO we observed that the normally even SFPQ nuclear distribution was clearly disturbed in many cells, forming both filament and foci-like inclusions (Figure 2Eii). Consistent with the reduced cell viability of fibroblasts transfected with the 2′ O-methyl AOs, transfection of U2OS cells was also associated with apparently lower cell confluence by 48 h (Supplementary Figure S3), implying either cytotoxicity or growth inhibition in these cells. Transfection of U2OS cells with a DNA phosphodiester AO and appropriate transfection controls did not result in perturbed SFPQ distribution of differences in cell confluence (Supplementary Figure S3), consistent with the findings in fibroblasts. The full time-course videos over the 48-h period are displayed in Supplementary Videos S1–S4. Taken together, we showed that transfection of fully modified 2′ O-methyl phosphorothioate AOs is a direct cause of the rapid formation of various DBHS protein inclusions and reduced cell survival. This effect is, to some extent, sequence specific.

The structure and composition of fully modified phosphorothioate AO nuclear inclusions was further investigated by immunostaining of additional selected nuclear proteins that are naturally found in a variety of nuclear structures, such as paraspeckles, the nuclear envelope, nucleoli, nuclear speckles, Cajal bodies and nuclear stress bodies. We observed that the paraspeckle-associated proteins PSPC1 and FUS were present in AO induced nuclear inclusions, as well as FUS accumulating within the cytoplasm (Supplementary Figures S4A,B). In contrast, TDP-43 and hnRNP-A1 were not evident in the nuclear inclusions but also appeared to have greater localization within the cytoplasm when compared to the untreated fibroblasts (Supplementary Figures S4C,D).

When examining coilin, a component of Cajal bodies, we observed a variety of abnormal staining patterns in 2′ O-methyl phosphorothioate AO transfected cells (Figure 3A). Specifically, we saw re-distribution of coilin to the cytoplasm and occasional co-localization within nuclear filaments with NONO, whereas coilin is present almost exclusively in the nucleus, segregated from NONO in untreated cells (Figure 3A). Similarly, both nucleolin and fibrillarin, markers for the nucleolus, showed altered distribution in nuclear inclusion-containing cells (Figures 3B,C) and the splicing factor and component of nuclear speckles, SC35, showed enhanced cytoplasmic localisation and was depleted from the nucleus in cells that also contained nuclear inclusions (Figure 3D). Line intensity profiling shows altered distribution of these nuclear proteins in individual cells that have nuclear inclusions (Supplementary Figure S5).

To further characterize the mechanism underpinning the displacement of nuclear paraspeckle proteins into the cytoplasm, we immunostained fibroblasts for hnRNP-A1, following transfection with the FITC-labeled 2′ O-methyl phosphorothioate sequence. Intriguingly, we observed some cytoplasmic co-localization of the AO with hnRNP-A1 in cells that did not appear to form nuclear inclusions (Figure 3E), suggesting that phosphorothioate AOs can indeed interact with nuclear proteins, drawing them to the cytoplasm. The cytoplasmic accumulation of hnRNP-A1 is reminiscent of its distribution within cytoplasmic stress granules, thus, to determine whether phosphorothioate AOs can interact with stress granules, we further immunostained for stress granule marker TIA-1 in FITC-AO transfected cells (Figure 3F). Cytoplasmic TIA-1 is clearly enhanced in 2′ O-methyl phosphorothioate transfected cells when compared to untransfected cells, with the appearance of defined stress granules. However, we observed no co-localisation of the FITC labeled AO with TIA-1, suggesting that the AO may not directly interact with the stress granules, and instead, it is possible that the stress of nuclear inclusions and associated toxicity may initiate the stress granule response. The difference in colocalization of the PS-AO between hnRNP-A1 and TIA-1 suggests that the phosphorothioate backbone does not draw hnRNP-A1 into stress granules in the cytoplasm within the 24 h timeframe evaluated.

Since components of subnuclear organelles were disorganized following transfection with phosphorothioate AOs, the levels of these proteins in the transfected cells were investigated. Western blot analysis showed that the overall abundance of each protein studied to be essentially unchanged following phosphorothioate AO transfection. This contrasts with the transfection of 2′fluoro-gapmer phosphorothioate AOs in Hela cells, that resulted in reduced levels of DBHS proteins (Shen et al., 2015). Representative western blots probed for SFPQ, NONO, TDP-43, hnRNP-A1, NCL, HSF1 and beta actin, are shown in Supplementary Figure S6. Thus, 2′ O-methyl phosphorothioate AO transfection into fibroblasts results in nuclear inclusions containing paraspeckle proteins, and mis-localisation of markers for other nuclear structures, including nucleoli and splicing speckles.

We next used transmission electron microscopy (TEM) on ultra-thin, osmium-stained sections to carry out detailed ultrastructural analysis of the 2′ O-methyl phosphorothioate SMN7D(-10-29) AO-induced nuclear inclusions in fibroblasts (Figures 4Ai–vi). We observed that the nuclear inclusions appeared to be microfibrillar or amyloid-like, approximately 200–250 nm in diameter, occurring mostly in groups or bundles. For reference, the width of a DNA double helix is ∼2 nm. If captured in a suitable plane of section, the structures appear to have very electron dense termini. Such nuclear inclusions have not previously been revealed by TEM in the investigating laboratory. The electron dense regions are reminiscent of perichromatin in size and electron density, the inclusions revealed by TEM extend to ∼2,000 nm in length. Numerous small, partially formed inclusions are also evident in nuclei of transfected cells when viewed at higher magnification (data not shown).

Since TEM can only capture segments of the presumed large fibril-like nuclear inclusions, super resolution microscopy was used to examine 2′ O-methyl phosphorothioate SMN7D(-10-29) AO transfected and untreated fibroblasts (Figures 4B,C). AO transfected fibroblasts co-stained for NONO and SFPQ show a network of many nuclear inclusions characterised by co-localisation of the two proteins, whereas the untreated cells show regular, punctate distribution of the proteins within naturally occurring paraspeckles (Figures 4Bi,ii, respectively). Strikingly, transfected fibroblasts co-stained for SFPQ and FUS reveal large, interconnected fibril-like structures decorated with FUS at mid-points and at the termini, suggesting the electron dense termini seen in the TEM could correspond to these FUS-specific regions (Figure 4Ci). SFPQ, NONO and FUS were all sequestered to the nuclear inclusions, unlike the even nuclear distribution of these proteins in untreated cells (Figure 4Cii).

We next examined the nuclear envelope protein, lamin B1 and observed altered distribution in a proportion of 2′ O-methyl phosphorothioate SMN7D(-10-29) transfected cells that exhibited nuclear inclusions (Figure 4Di,ii). In these cells, the nuclear envelope labelled by immunostaining of lamin B1 appears multi lobular and distorted, reminiscent of nuclei found in the premature ageing disease Hutchinson-Gilford progeria syndrome and in E145K cells (Taimen et al., 2009). The super resolution image in Figure 4Di clearly illustrates lamin B1 localized in 4 distinct nuclear lobules. In all images (Figures 4B–D), the nuclear inclusions appear to reside between areas that may reflect chromosomal territories. In summary, human fibroblasts with AO-induced nuclear inclusions exhibit striking ultrastructural features, including aberrant nuclear lamina and unusual arrangements of paraspeckle proteins within the inclusions.

Others have shown that some gapmer phosphorothioate AO sequences cause toxicity through sequestration of DBHS proteins to nucleoli and disrupted 5.8S RNA metabolism (Shen et al., 2019) [for review see (Crooke et al., 2020)]. Since Cajal bodies and fibrillarin, disrupted in our study, are also important in ribosomal RNA processing and assembly, we assessed rRNA levels in 2′ O-methyl phosphorothioate AO transfected cells and untreated cells. qPCR and bioanalyser data showed that unprocessed rRNA was markedly increased by 2′ O-methyl phosphorothioate AO transfection in fibroblasts (Figure 5). A Bioanalyser RNA trace from 2′ O-methyl phosphorothioate AO transfected fibroblasts (Figure 5A) showed low levels of 18S and 28S ribosomal subunits, as well as a several intermediate peaks, not present in RNA from untreated fibroblasts (Figure 5B), consistent with incomplete rRNA processing in the transfected cells. We ruled out that this effect was due to poor RNA quality, as a similar pattern was observed in all experimental repeats (n = 6) and that RNA integrity numbers (RIN) were measured at 10 for all samples from untreated, DNA transfected and lipofectamine treated cells, yet RINs of 6.4–8.8 were measured for the RNA collected from phosphorothioate AO treated cells, with all samples processed equally. This observation was supported by qPCR analysis of rRNA levels showing a 2–3.5-fold increase in the level of the unprocessed 45S rRNA in 2′ O-methyl phosphorothioate transfected compared to untreated fibroblasts (Figure 5C). Unprocessed rRNA accumulated over time, following the formation of nuclear inclusions (Figure 5D). Thus, fully modified phosphorothioate AOs that form nuclear inclusions sequestering DBHS proteins can also disrupt rRNA metabolism, without the involvement of RNase H1.

Following phosphorothioate AO transfection of GFP-SFPQ expressing U2OS cells (as in Figure 2) at 100 nM, live cell imaging to evaluate caspase activation was undertaken using a caspase-sensitive dye (Figure 6A). Focusing on individual cells with inclusions, we observed nuclear inclusions in two sister cells, 5 h after transfection. Nuclear inclusions became larger and more compact over time, followed by caspase activation, 22 h after transfection. No caspase activation was observed in the absence of intranuclear inclusions, suggesting that the presence of the inclusions was complicit in caspase activated apoptosis, leading to cell death. Others have shown when using a 2′ fluoro-modified phosphorothioate AO in mouse cells that cell death was tumour suppressor protein p53 dependent, although some p53-independent mechanisms also played a role (Shen et al., 2018). We therefore carried out western blotting of protein extracts from fibroblasts transfected with 2′ O-methyl phosphorothioate AOs, using antibodies against p53. We observed differently in this setting, in that p53 was decreased in transfected cells, relative to that in untreated control cells (Figures 6B,C). RT-PCR of BCL2-like isoforms also indicated an increase in the shorter, pro-apoptotic isoform in the 2′ O-methyl phosphorothioate AO transfected fibroblasts, compared to untransfected cells, again suggesting apoptosis as a result of the AO transfection (Figure 6E). As a comparison, we also measured p53 levels following transfection of PMOs of the same sequences and noted that p53 levels remained unchanged relative to those in untreated cells (Figures 6B,C). To further support the finding that fully modified phosphorothioate AOs suppressed p53 in primary cells, we investigated the expression of a number of p53 regulated transcripts following RNAseq analysis (Figure 6D). The p53 encoding transcript TP53 was also significantly downregulated in the RNAseq dataset, along with the p21 encoding transcript CDKN1A and BAX, two critical players in the p53 signaling pathway, supporting that in the primary cell model evaluated, caspase activation following fully modified phosphorothioate AO transfection was p53 independent.

We next evaluated the effect of 2′ O-methyl phosphorothioate AO transfection and nuclear inclusion formation on the transcriptome by RNAseq. We prepared RNA from fibroblasts transfected with the two 2′ O-methyl phosphorothioate AOs used throughout this study, as well as from untreated cells and cells transfected with a DNA phosphodiester AO lipoplex (100 nM for 24 h) for RNAseq analysis (n = 3). A heatmap comparing the expression of transcripts in 2′ O-methyl phosphorothioate AO transfected fibroblasts to that in cells transfected with the DNA phosphodiester oligonucleotide and untreated cells showed stark differences in expression between the groups (Figure 7A). A principal components analysis (PCA) clustered untreated and the DNA AO transfected cells perfectly, indicating that neither the lipofectamine transfection reagent or the DNA AO elicited alterations to the transcriptome (Figure 7B). A striking 91% variance was observed in the PCA between the two phosphorothioate AO transfected sample groups and the untreated and DNA-AO transfected groups, while only 3% variance in transcript profiles was observed between the two phosphorothioate AO transfected groups (Figure 7B).

Gene ontology (GO) analyses showed significant disruptions to major cellular pathways. The GO network (Figure 7C) revealed components of signaling pathways, chromatin silencing, and metabolic pathways to be over-represented, while pathways regulating both apoptosis and autophagy, membrane and organelle organisation, protein transmembrane transport and nucleotide excision repair components are under-expressed. Gene ontology pathways included in each cellular process are listed in Supplementary Table S4. Overall, these data show dramatic and widespread changes to gene expression in human fibroblasts, resulting from uptake of the 2′ O-methyl phosphorothioate AOs.

Given that the use of fully-modified AOs is common practice in developing splice-switching AOs, and that many paraspeckle proteins play a critical role in the splicing process, it was necessary to investigate changes in splicing events as a result of phosphorothioate AO transfection and protein interaction. Figure 7D illustrates differential splicing events as determined by RNAseq analysis, whereby the fold-change in the percentage spliced-in is plotted against transcript expression. We observed a notable difference in both increased and decreased exon retention events compared to baseline, as indicated by either positive or negative plots, respectively, suggesting global changes in pre-mRNA splicing.

Global changes to the transcriptome following phosphorothioate AO transfection are evident, however the extent and consequences of deleterious effects are unknown. Table 1 shows key genes involved in critical cellular processes that were differentially expressed by a minimum Log₂FoldChange of ±0.7 (p _adj < 0.05) in fibroblasts following AO transfection. Transcripts encoding key paraspeckle protein NONO (Log2FoldChange −1.6, p = 1.33E-134) as well as for hnRNP-A1 (−0.89, p = 1.02E-89) were observed to be downregulated, while SFPQ was only modestly downregulated (−0.25, p = 2.5E-10) and PSPC1 modestly upregulated (0.55, p = 1.73E-5). Interesting, the long non-coding RNA NEAT1, essential for paraspeckle formation and displaced by the phosphorothioate AO interaction, was modestly increased (0.11, p = 0.026), while MALAT1 (NEAT2) was significantly upregulated (1.36, p = 1.6E-50). Given the observed changes in splicing events (Figure 7D), we highlight a number of transcripts critical in determining pre-mRNA splicing events, such as those involved in forming the spliceosome as well as major splicing factor families, including those encoding the serine/arginine splicing factor (SRSF) proteins, heterogeneous ribonuclear proteins (hnRNP) and the pre-mRNA processing factor (PRPF) proteins.

Of critical relevance to the use of AOs in treating neurodegenerative diseases and proteinopathies, we draw attention to transcripts encoding proteins involved in autophagy regulation as well as those linked to neurodegenerative diseases, such as ALS. Other protein families that were significantly dysregulated at the transcript level included those involved in transcription e.g. the SUPT proteins, mitochondrial function, chromatin structure and epigenetic regulation, e.g. the histone deacetylases (HDAC). Consistent with nucleolar disruption and mis-regulation of ribosomal RNA, key nucleolar stress markers were also dysregulated, including those encoding p53, c-Myc, E2F-1 and SP1 transcription factors (Yang et al., 2018). Heatmaps showing altered expression of transcripts regulated by these factors are shown in Supplementary Figure S7. Of particular concern for all studies evaluating phosphorothioate AOs, we observed that commonly utilized housekeeping genes, including those encoding β-actin, glyceraldehyde 3-phosphate dehydrogenase (GAPDH), lamin-A and β-tubulin were all downregulated, while cyclin-D was upregulated, suggesting caution be taken when normalizing expression to these transcripts. An expanded list of these transcripts, and the Log2FoldChange and p-value, is included in Supplementary Table S5.

While transfection of gapmer phosphorothioate AOs resulted in nuclear inclusions (Shen W. et al., 2014), PMOs are yet to be evaluated in this context. We therefore transfected fibroblasts with the AO sequences described above, synthesized as PMOs, and carried out immunofluorescence against NONO. Transfection of the PMO sequences into fibroblasts did not alter the apparent distribution of NONO, nor of any other nuclear proteins studied, when delivered using the Lipofectamine transfection reagent (Figures 8A,B). In order to deliver the PMO using the equivalent concentration and delivery technique to the 2′ O-methyl phosphorothioate AO, the PMO was annealed to a complementary DNA ‘leash’ and transfected as a lipoplex (100 nM, 24 h). While no altered NONO distribution was observed with the PMO, clear NONO positive nuclear inclusions were observed in the 2′ O-methyl transfected fibroblasts in the same experiment (Figure 8Aii). To further demonstrate robust cellular uptake of these PMOs, we used two additional established delivery techniques: Neon electroporation (50 and 10 µM), incubated for 24 h, and un-complexed ‘naked’ PMO transfection, relying on gymnotic uptake (10 µM), incubated for 72 h (Figures 8A,B). Immunofluorescent detection of NONO showed no PMO-induced nuclear inclusion formation following transfection, under any of the conditions used (Figures 8A,B), and was reproducible, irrespective of the PMO sequence (not shown). Interestingly, a proportion of the fibroblasts transfected with 50 µM of the PMO by Neon electroporation showed NONO distribution within the nucleolus at 24 h after transfection (Figure 8Aiv), yet this was not associated with nuclear inclusions, as observed in fibroblasts transfected with the 2′ O-methyl phosphorothioate AOs at the same concentration, with 90% of cells containing nuclear inclusions (Figure 8Av). Efficiency of transfection with SMN7A(+13+32) AO sequences and induced exon 7 skipping in the target SMN transcripts was assessed by RT-PCR across SMN exons 4-8, confirming that the absence of intranuclear inclusions in PMO transfected cells occurred despite robust cellular uptake of the PMO using each of the transfection techniques (Figure 8C; Supplementary Figure S8). The PMO chemistry is therefore a promising alternative to 2′ O-methyl phosphorothioate AOs for avoidance of nuclear inclusions in splice-switching applications.

We hypothesized that nuclear inclusion formation may be circumvented by delivery of double stranded AOs formed by annealing two complimentary fully modified phosphorothioate AOs. We therefore evaluated several 2′ O-methyl phosphorothioate AOs, transfected either individually (Figure 8Di) or annealed using a decreasing thermal incubation protocol (Figure 8Diii). Different AO sequences were transfected in this manner and the cells were immuno-stained for SFPQ or NONO to identify any nuclear inclusion formation, with representative NONO staining images from the SMN7D(-10-29) sequence transfected cells shown in Figure 8D, with the percentage of cells containing nuclear inclusions shown in Figure 8E. Complementary 2′ O-methyl phosphorothioate antisense and sense AO sequences, delivered individually, both formed nuclear inclusions in up to 68% of cells (Figure 8Di). Of note, while the antisense sequence formed NONO-filaments (Figure 8Di), the sense sequence predominantly sequestered NONO into foci, as shown in Figure 2Ci, supporting our other observations of sequence-specific effects on the phosphorothioate AO-protein interactions. Annealing equal amounts of the two complementary AOs prior to transfection appears to abrogate NONO sequestration, with only 7% of cells showing inclusions after transfection with 100 nM of annealed AO duplex (Figure 8Diii), and no cells showing inclusions after transfection at 50 nM of the AO duplex (not shown). Cytoplasmic distribution of NONO was observed in a similar percentage of cells, however, diffuse NONO staining was also observed within the nucleus of these cells, suggesting that NONO was not entirely drawn from the nucleus, as with certain 2′ O-methyl phosphorothioate AOs. We further determined that the annealing of a DNA sense AO, much like a leash, to a fully modified antisense phosphorothioate 2′ O-methyl AO has a similar protective effect in preventing protein interaction and nuclear inclusions. Using the SMN7D(-10-29) sequence transfected into SMA patient fibroblasts, 8% of cells transfected with the 2′ O-methyl phosphorothioate:DNA hybrid contained nuclear inclusions (Figure 8Div). NONO was evident in the cytoplasm in some cells at a similar rate to that induced by the annealed 2′ O-methyl phosphorothioate AOs. Densitometry analysis of RT-PCR of SMN transcripts confirmed robust transfection efficiency, with the annealed all-modified phosphorothioate duplex and the 2′ O-methyl phosphorothioate:DNA hybrid duplex proving equally effective at inducing SMN2 exon 7 retention as the 2′ O-methyl phosphorothioate AO alone (Figure 8F; Supplementary Figure S8). The SMN7D(−10–29) AO is designed to reinforce exon 7 retention of SMN2 in SMA, increasing the production of full-length SMN (FL-SMN). Intriguingly, when examining total FL-SMN expression (Figure 8F), the annealed AOs induced up to 1.6- and 1.5-fold increase in FL-SMN with the 2′ O-methyl:2′ O-methyl and the 2′ O-methyl:DNA, respectively, compared to that of untransfected SMA fibroblasts, while the 2′ O-methyl phosphorothioate sequence alone only induced a 1.2-fold increase in FL-SMN, suggesting that antisense activity is in fact improved by delivering phosphorothioate AOs as double stranded molecules. Thus, pre-annealing complementary 2′ O-methyl phosphorothioate AOs appears to abolish the tendency to induce DBHS-protein enriched inclusions, presumably increasing the bioavailability and activity of the AO.

To determine whether preventing formation of DBHS protein positive nuclear inclusions by way of annealing complementary AOs or utilizing PMOs can prevent the downstream off-target effects and cellular toxicity, we measured cell viability and transcript expression of selected transcripts, shown to be dysregulated in the RNAseq analysis. Fibroblasts were transfected with three different 2′ O-methyl phosphorothioate AO sequences and corresponding 2′ O-methyl:2′ O-methyl fully phosphorothioate modified complex, 2′ O-methyl phosphorothioate:DNA hybrid and a PMO:DNA (leash) hybrid. Interesting, cell viability (Figure 9A) did not entirely correlate with the percentage of cells displaying nuclear inclusions (Figure 9B) in transfected cells. However, we did observe a significant increase in the viability of cells transfected with the annealed AOs, with both complementary, fully modified and DNA AOs, and with PMOs transfected using a leash. The PMO transfected cells showed the highest survival after transfection under all the conditions evaluated. AO-mediated effects on transcript expression were analyzed by assessing SMN splicing by densitometry analysis of RT-PCR across exons 4–8; TBP levels as a housekeeping/loading control, CCND1, HDAC2 and SUPT4H1 transcripts shown to be dysregulated by RNAseq, and BCL-xL shown to be decreased in our study (Figures 9C–H, respectively). Gel images of RT-PCRs are shown in Supplementary Figure S9. Interestingly, we observed each of these transcripts to be significantly downregulated following 2′ O-methyl phosphorothioate transfection, including TBP that was not dysregulated according to the RNAseq data. As such, we were unable to use TBP as a loading control for normalization and given that equal amounts of RNA were used for the RT-PCRs, with three different 2′ O-methyl phosphorothioate sequences evaluated (n = 4), we instead determined the fold-change in raw transcript expression between transfected and untreated samples. Expression of each of the transcripts evaluated was corrected to varying degrees by use of the double stranded annealed AOs and PMOs. The levels of transcript correction correlated with the percentage of cells containing nuclear inclusions, as well as improved cell viability, suggesting that the use of double stranded AOs or PMOs can overcome the propensity to form nuclear inclusions, as well as the toxic and off-target effects of phosphorothioate AOs.