ELX-02 generates protein via premature stop codon read-through without inducing native stop codon read-through proteins

ELX-02 is a clinical stage, small molecule Eukaryotic Ribosomal Selective Glycoside (ERSG) acting to induce read-through of premature stop codons (PSCs) that results in translation of full-length protein. However, improved read-through at PSCs has raised the question of whether native stop codon (NSC) fidelity would be impacted. Here, we compare read-through by ELX-02 in PSC and NSC contexts. DMS-114 cells containing a PSC in the TP53 gene were treated with ELX-02 and tested for increased nuclear p53 protein expression, while also monitoring two other proteins for NSC read-through. Additionally, blood samples were taken from healthy subjects pre- and post-treatment with ELX-02 (0.3-7.5mg/kg). These samples were processed to collect white blood cells, then analyzed by western blot to identify native and potentially elongated proteins from NSC read-through. In a separate experiment, lymphocytes cultivated with vehicle or ELX-02 (20 and 100 µ g/mL) were subjected to proteomic analysis. We found that ELX-02 produced significant read-through of the PSC found in TP53 mRNA in DMS-114 cells, resulting in increased p53 protein expression and consistent with decreased nonsense mediated mRNA degradation. NSC read-through protein products were not observed in either DMS-114 cells or in clinical samples from subjects dosed with ELX-02. The number of read-through proteins identified using proteomic analysis was lower than estimated and none of the NSC read-through products identified with >2 peptides showed dose-dependent responses to ELX-02. Our results demonstrate significant PSC read-through by ELX-02 with maintained NSC fidelity, thus supporting the therapeutic utility of ELX-02 in diseases resulting from nonsense alleles. unbiased proteomic analysis of isolated human peripheral blood mononuclear cells (PBMCs) read-through ERSGs. p53 quantified. Western blot assay: DMS-114 cells were cultured in 12 well plates (666,000 cells/well), then treated with compound (either vehicle, G418, or ELX-02 in an 8-point concentration response curve from 0.023-3mM) in biological duplicates starting 24 hours later. Cells were lysed in Radioimmunoprecipitation assay (RIPA) buffer (catalog # 89900, Thermo Fisher, Carlsbad, CA) containing protease and phosphatase inhibitors (catalog # 78443, Thermo Fisher, Carlsbad, CA) 48 hours post-treatment. Once protein content was quantified using the BCA protein determination kit (catalog # 23225, Thermo Fisher, Carlsbad, CA), 60 µ g of each sample was added into a lane of a 12% Bis-tris gel (catalog # NP0342, Thermo Fisher, Carlsbad, CA). For p53, cystatin C (CysC), and LaminB1, protein transfer was accomplished using a 0.2 µ m PVDF membrane in the iBlot2 dry blotting system (catalog # IB24001, Thermo Fisher, Carlsbad, CA). For Vascular Endothelial Growth Factor A (VEGF-A), protein transfer was accomplished by wet blotting onto a 0.45 µ m PVDF membrane using 1X NuPage transfer buffer (catalog # NP0006, Thermo Fisher, Carlsbad, CA) and 10% methanol in demineralized water for 1 hour at 30V. After blocking with 5% skimmed milk in TBS-T for one hour at room temperature, blots were probed with mouse anti-p53 primary antibody (DO-1) (1:200, catalog # sc-126, Santa Cruz Biotechnology, Dallas, TX) and goat anti-mouse secondary antibody (1:10,000, catalog # 925-32210, Li-Cor Biotechnology, Lincoln, NE). As a loading control, blots were also probed with rabbit anti-Lamin B1 primary antibody (1:1,000, catalog # ab133741, Abcam, RT-qPCR were cultured in 96 well plates (40,000 cells/well), then treated with compound (either vehicle, G418, or ELX-02 in an 8-point concentration response curve from 0.023-3mM) in biological duplicates starting 24 hours later. RNA was harvested 24 hours post-treatment using the MagMAX-96 Total RNA Isolation Kit (catalog # AM1830, Ambion, Austin, TX) and samples from two wells of each condition were pooled to obtain sufficient RNA. cDNA was synthesized in a separate reaction using reverse transcriptase (catalog # 4368814, Applied Biosystems, Foster City, CA). Amplification was performed using Taqman fast advanced master mix (catalog # 4444554, Applied Biosystems, Foster City, CA) in technical triplicates. The gene for β -actin ( ACTNB , assay ID Hs01060665_g1 with VIC-MGB dye, Thermo Fisher, Carlsbad, CA) was selected for multiplexing with the gene for p53 ( TP53 , assay ID Hs01034249_m1 with FAM-MGB dye, Thermo Fisher, Carlsbad, CA). Steady state gene expression was measured using the LightCycler 480 (Roche, Basel, Switzerland) using standard cycling conditions. Relative TP53 expression was then calculated based on the ΔΔC p method. that read-through peptides tend to have higher PEPs than standard peptides, consistent with the conclusion that a high proportion of identified read-through peptides are false approach to provide unbiased detection of read-through peptides for most human expressed proteins. Our analysis demonstrates that the number of read-through sequences in ELX-02 treated samples was lower than the predicted FDR levels of read-through peptides. Read-through peptides were also observed to have higher PEP distributions than standard peptides, consistent with the conclusion


INTRODUCTION
Approximately 10-12% of all mutations are single point mutations that lead to an in-frame UAA, UAG, or UGA codon in mRNA (Krawczak et al., 2000;Mort et al., 2008), thereby introducing a stop signal. These "nonsense" mutations that result in premature stop codons (PSCs) are the cause a significant number of genetic diseases (Linde and Kerem, 2008) including cystic fibrosis, cystinosis, Usher's Syndrome, primary ciliary dyskinesia, polycystic kidney disease, mucopolysaccharidosis I, and a variety of cancers (Bordeira-Carrico et al., 2012). Nonsense mutations are single nucleotide changes that may occur in germline or somatic deoxyribonucleic acid (DNA), by inaccurate or inefficient pre-messenger ribonucleic acid (mRNA) splicing, or in the context of improper RNA editing. The disease phenotypes caused by nonsense mutations are frequently more severe than those caused by missense mutations because nonsense mutations can completely eliminate protein function or can produce a small protein fragment which often yields a dominant negative effect with protein partners (Frischmeyer and Dietz, 1999;Miller and Pearce, 2014). Moreover, premature termination fails to remove all exon junction complexes (EJCs) from the mRNA, which signals nonsense-mediated decay (NMD) machinery to degrade mRNA and further limit protein production (Celik et al., 2015). Curative therapies do not exist for most nonsense mutation diseases and treatment options are needed beyond palliative care.
Premature translation termination can be overridden by small molecule intervention, allowing for readthrough and restoration of full-length protein. As an example, aminoglycoside antibiotics can induce translational read-through by binding to the decoding site in the small subunit of the ribosome (Burke and Mogg, 1985;Moazed et al., 1986;Pokrovskaya et al., 2010;Prokhorova et al., 2017;Dabrowski et al., 2018;Sabbavarapu et al., 2018). This site normally monitors proper codon-anticodon interactions and when these read-through molecules bind to the decoding site, they induce a conformational change in the ribosome which stabilizes the "exo" confirmation of two conserved adenosine residues Shulman et al., 2014). This change reduces the ability of the ribosome to discriminate between cognate and near-cognate aminoacyl-transfer RNAs (Fourmy et al., 1998) and increases the probability that translational read-through of PSCs will occur. This contrasts with native stop codons (NSCs), where the ability to induce read-through is markedly diminished as an empty decoding site will be This article has not been copyedited and formatted. The final version may differ from this version.
In limited short-term studies, aminoglycoside antibiotics have shown potential to induce translational readthrough across PSCs, resulting in functional protein and clinical efficacy (Clancy et al., 2001;Wilschanski et al., 2003). Despite these promising results, these compounds have limited read-through efficacy and cannot be used as long-term therapies because of the potential to induce nephrotoxicity and ototoxicity following uptake and accumulation in renal proximal tubules and cochlear hair cells, respectively (Prayle et al., 2010;O'Sullivan et al., 2017). While nephrotoxicity is reversible through kidney tubule cell proliferation, aminoglycoside antibiotic-induced ototoxicity can lead to permanent hearing loss due an inability of cochlear hair cells to regenerate. Therefore, for effective long-term treatment of nonsense mutation diseases, compounds are needed which augment PSC read-through but lack the toxicities typically associated with this class of compounds.
The present studies compare PSC and NSC translational read-through following treatment with ELX-02 [6'-(R)-Methyl-5-O-(5-amino-5,6-dideoxy-α-L-talofuranosyl)-paromamine sulfate, also known as NB124 or (S)-9 from (Kandasamy et al., 2012)], a small molecule Eukaryotic Ribosomal Selective Glycoside (ERSG) engineered to read-through PSCs without the toxicities of aminoglycoside antibiotics (Leubitz et al., 2019). A cancer cell line carrying a nonsense mutation within the TP53 gene was evaluated for PSC read-through and full length p53 protein production was quantified following ELX-02 treatment, while also monitoring for NSC read-through protein products. Blood samples were also collected pre-and posttreatment with ELX-02 for protein analysis to further explore whether ELX-02 induces NSC read-through.
Finally, unbiased proteomic analysis of isolated human peripheral blood mononuclear cells (PBMCs) treated with ELX-02 was performed to evaluate differences in the read-through proteome. Such studies are important to understand the intended and potentially unintended read-through products following treatment with ERSGs. This article has not been copyedited and formatted. The final version may differ from this version.
JPET Fast Forward. Published on May 6, 2020 as DOI: 10.1124/jpet.120.265595 at ASPET Journals on July 9, 2020 jpet.aspetjournals.org contains a point mutation of C>T at nucleotide 637 of the TP53 gene, resulting in a UGA premature stop at codon 213 (R213X) which leads to a truncated p53 protein that lacks tumor suppressor activity. Cells were incubated at 37°C with 5% CO2 in Waymouth's MB 752/1 medium (catalog # 31220-023, Thermo Fisher, Carlsbad, CA), supplemented with 10% of non-heat-inactivated fetal bovine serum (FBS; catalog # F2442-500mL, Sigma-Aldrich, St. Louis, MO) and 1% PenStrep (catalog # 15140-122, Thermo Fisher, Carlsbad, CA). Compound treatments were performed in the same medium, but without PenStrep addition. Vehicle (1 or 2% water) was run in parallel on every plate as a negative control. G418 (0.75mM), considered to be one of the strongest read-through inducers (Bidou et al., 2017), was also run in parallel on every plate as a positive control.
Immunofluorescence assay: DMS-114 cells were cultured in 96 well plates (25,000 cells/well), then treated with compound (either vehicle, G418, or ELX-02 in an 8-point concentration response curve from 0.023-3mM) in biological triplicates starting 24 hours later. Edge wells were filled with water to avoid evaporation in the remaining wells of the plate. Cells were fixed 48 hours post-treatment with 4% paraformaldehyde for 15 minutes at room temperature. Blocking buffer containing 3% bovine serum albumin and 0.2% Triton X-100 in phosphate buffered saline (PBS) was applied, followed by incubation steps overnight at 4°C with mouse anti-p53 primary antibody (DO-1) (1:200, catalog # sc-126, Santa Cruz Biotechnology, Dallas, TX) and goat anti-mouse secondary antibody (1:1,000, catalog # A21236, Thermo Fisher, Carlsbad, CA). To stain cell nuclei, 4′,6-diamidino-2-phenylindole (DAPI, catalog # D3571, Invitrogen, Carlsbad, CA) was diluted in PBS and used at a final concentration of 1µg/mL in each well.
Cells were imaged at 10X magnification. Immunofluorescence was detected using high content imaging (INCell6000 analyzer, GE Healthcare Life Sciences, Pittsburg, PA) and an algorithm was developed to quantify staining. The number of DAPI nuclei were counted and the mean density of p53 nuclear expression was quantified. This article has not been copyedited and formatted. The final version may differ from this version.

JPET # 265595
RT-qPCR assay: DMS-114 cells were cultured in 96 well plates (40,000 cells/well), then treated with compound (either vehicle, G418, or ELX-02 in an 8-point concentration response curve from 0.023-3mM) in biological duplicates starting 24 hours later. RNA was harvested 24 hours post-treatment using the MagMAX-96 Total RNA Isolation Kit (catalog # AM1830, Ambion, Austin, TX) and samples from two wells of each condition were pooled to obtain sufficient RNA. cDNA was synthesized in a separate reaction using reverse transcriptase (catalog # 4368814, Applied Biosystems, Foster City, CA). Amplification was performed using Taqman fast advanced master mix (catalog # 4444554, Applied Biosystems, Foster City, CA) in technical triplicates. The gene for β-actin (ACTNB, assay ID Hs01060665_g1 with VIC-MGB dye, Thermo Fisher, Carlsbad, CA) was selected for multiplexing with the gene for p53 (TP53, assay ID Hs01034249_m1 with FAM-MGB dye, Thermo Fisher, Carlsbad, CA). Steady state gene expression was measured using the LightCycler 480 (Roche, Basel, Switzerland) using standard cycling conditions.
Relative TP53 expression was then calculated based on the ΔΔCp method.

ΔCp = CpTP53 -CpACTNB
Data analysis: Data were graphed and analyzed using GraphPad Prism 8 software. Concentration response curves were subjected to non-linear regression analysis [log(agonist) vs. response -variable slope (four parameters)]. One-way ANOVA with Dunnett post-test was run to determine compound efficacy relative to vehicle, with significance threshold set as p < 0.05.

Clinical samples:
Collection and analysis: Blood samples were collected pre-dose, as well as at 24, 48 and 144 hours post-ELX-02 treatment (0.3-7.5mg/kg) from healthy male and female subjects as part of a randomized, doubleblinded, placebo-controlled single-ascending dose, Phase 1 clinical trial (Leubitz et al., 2019). Blood samples were similarly collected from subjects receiving placebo. In total, samples were collected from 51 subjects. EL-001 (NCT02807961) was conducted at Tel Aviv Sourasky Medical Center, Clinical Research This article has not been copyedited and formatted. The final version may differ from this version. Antwerp, Belgium) reviewed and approved the study protocols and the informed consent documents. All subjects provided written informed consent before any study-related procedures were performed.
Lysates were defrosted and protein extracts were diluted (ratios of 1:10 and 1:20) and the protein concentration measured using the Bradford assay reagent (catalog # 5000006, Bio-Rad Laboratories, Hercules, CA) with bovine serum albumin as a calibration standard. Crude extracts (5-10µg) were mixed with Laemmli buffer with β-mercaptoethanol as a reducing agent, to a final concentration of 1X. Extracts were analyzed by Western blot for the expression of two housekeeping proteins, CysC and β2 microglobulin (β2M), and two proteins with high NSC read-through, AGO1 and MDH1.
Gel Analysis of CysC and β2M: Samples were separated on 12% NUPAGE Bis-Tris gels (catalog # NP0342, Thermo Fisher, Carlsbad, CA), alongside a molecular weight standard ranging from 10 to 250kDa (catalog # 1610373, Bio-Rad Laboratories, Hercules, CA). Positive control proteins mimicking extended NSC read-through were also generated using plasmids harboring a point mutation in the NSC for each of the proteins. An in-frame tandem stop codon was present in each case downstream of the NSC. Expression of those plasmids in HEK293 cells gave rise to an elongated form of each of protein, This article has not been copyedited and formatted. The final version may differ from this version.
JPET Fast Forward. Published on May 6, 2020 as DOI: 10.1124/jpet.120.265595 at ASPET Journals on July 9, 2020 jpet.aspetjournals.org The resulting raw files were then searched against the read-through proteome database (Goodenough et al., 2014), with the peptide and protein false discovery rates (FDR) set to 1%. An algorithm was used to match theoretical peptide spectra against observed spectra and assign them a score. To monitor the FDR, a decoy database was generated from the theoretically observable proteome (real search database), and the data searched against these fake proteins. A score threshold was identified such that the ratio of identifications from the decoy to the real search database was 1%. In addition, each peptide also had a Posterior Error Probability (PEP): the probability that the current identification was erroneous, equal to the modeled local ratio between decoy and real identifications for the peptide's score.
This article has not been copyedited and formatted. The final version may differ from this version.  Figure 1B). The decreased number of nuclei appears to be related to the increased TP53 read-through and is consistent with the hypothesis that restoring p53 protein and function in a cancer cell can trigger apoptosis (Mashima et al., 2005;Floquet et al., 2011;Shirole et al., 2016;Bidou et al., 2017). At the lowest concentration of ELX-02 (0.23 mM), p53 nuclear expression and DAPI-positive nuclei was not different from vehicle-treated cells. All other concentrations of ELX-02 had p53 nuclear expression that was significantly greater than vehicle-treated cells. High concentrations of ELX-02 (0.75-3 mM) induced p53 nuclear expression levels and loss of DAPI-positive nuclei to an equal or greater degree than G418-treated cells.
ELX-02 increased the truncated and full-length forms of the p53 protein in DMS-114 cells by western blot (Figure 2A-B), consistent with read-through of the TP53 gene. Similar results were observed by other groups after G418 treatment of either Calu-6 cells that harbor an R195X mutation in the TP53 gene (Wangen and Green, 2019) or HDQ-P1 cells that harbor an R213X mutation in the TP53 gene (Floquet et al., 2011;Bidou et al., 2017). At low concentrations of ELX-02 (0.047 mM), truncated p53 protein could be detected at levels greater than vehicle-treated controls. This increase in truncated p53 protein suggests small amounts of PSC read-through likely occurred, which then may have removed EJC proteins and led to less NMD, greater mRNA stability, and production of more protein overall (Keeling et al., 2014). The increase in truncated p53 protein by ELX-02 peaked at 0.38mM and then dropped off as ELX-02 effectively induced read-through of the PSC in TP53 mRNA and generated full length p53 protein. In samples taken from the same DMS-114 cells, the effect of ELX-02 on read-through of two other genes was tested: 1) CysC (a ~13 kDa housekeeping gene, which is a reversible competitive inhibitor of papainlike cysteine proteases) and 2) VEGF-A ( a dimeric glycoprotein that induces growth of blood vessels, with weights of ~38 kDA for normal protein and at ~43 kDa for NSC read-through protein). It was hypothesized that proteins showing an increase in NSC read-through by ELX-02 would be proteins with high expression across all cells or proteins with basal levels of native read-through higher than other proteins. A decrease in the amount of CysC protein at high concentrations of ELX-02 was observed ( Figure 2A, 2C). This effect may be due to cellular toxicity at high ELX-02 concentrations that impact CysC earlier than the other housekeeping gene Lamin B1, potentially because of differences in the halflife of each protein. No high molecular weight CysC protein product from NSC read-through could be detected. Additionally, there was no effect of ELX-02 on the expression of VEGF-A (Figure 2A, 2D) when measuring either the standard-length protein or the higher molecular weight protein that would result from NSC read-through.
Finally, ELX-02 dose-dependently decreased the Cp value for TP53 mRNA with an IC50 of 133µM, indicating an increase in TP53 steady state transcript levels in DMS-114 cells ( Figure 3A). This result is consistent with western blot data, suggesting decreased NMD of the TP53 mRNA product resulting in increased protein levels. Alternatively, the increased transcript and protein levels could indicate a direct effect of ELX-02 that produces increased expression from the TP53 gene. High concentrations of ELX-02 resulted in a Cp that was similar to G418-treated cells. ELX-02 had no effect on the Cp value for the housekeeping gene for β-actin, ACTNB, until the highest dose of 3 mM ( Figure 3B), suggesting toxicity at the highest dose. Using the relative Cp values for both TP53 and ACTNB, the fold change in expression for TP53 was calculated. Here, after disregarding the highest concentration which is confounded by cellular toxicity, ELX-02 treatment resulted in > 30-fold increase in TP53 transcript levels relative to vehicle controls ( Figure 3C), which is similar to the effect observed with G418 treatment. Furthermore, these findings are in consistent with results obtained from treating Calu-6 cells with G418, whereby a 16-This article has not been copyedited and formatted. The final version may differ from this version.

No Evidence of Native Stop Codon Read-through in Ex Vivo Clinical Samples
NSC read-through was examined for the housekeeping protein CysC in PBMCs from all clinical trial subjects treated with either placebo or ELX-02. Additionally, native read-through of β2M, a component of MHC class I molecules present on all nucleated cells, was examined following treatment of ELX-02. CysC and β2M each showed one band at the expected molecular weight for the native protein (~13kDa and ~12kDa for CysC and β2M, respectively; representative blot from subjects dosed with 5.0mg/kg ELX-02 are provided in Figure 4). The higher molecular weight species for these two proteins were not observed in any of the clinical samples, indicating no measurable NSC read-through translation of these housekeeping genes and no production of elongated protein variants.
PBMCs from two cohorts of clinical trial test subjects (placebo and 7.5 mg/kg ELX-02) were also analyzed for read-through of genes with high basal NSC read-through. AGO1 (a 90kDa protein that plays a central role in RNA silencing processes) and MDH1 (a 35kDa enzyme that catalyzes the reversible oxidation of malate to oxaloacetate) each use a native Opal, UGA, stop codon which is most susceptible to both NSC read-through and compound-induced PSC read-through (Cridge et al., 2018). We chose to focus on the 7.5 mg/kg ELX-02 group since this dose level provided the highest clinical exposure to ELX-02 (Leubitz et al., 2019) and therefore would be most likely to induce NSC read-through. Samples showed expression of the native AGO1 and MDH1 proteins at the expected molecular weights (representative blots from subjects receiving either placebo or 7.5 mg/kg ELX-02 are provided in Figure 5). The native AGO1 signal was lower than the limit of quantification (1.5ng recombinant AGO1) and at the limit of detection. Basal read-through appeared to be minimal for this protein and no noticeable increase was observed in samples collected after receiving ELX-02. The native MDH1 signal was within the range of quantitation (3ng recombinant MDH1). While a faint band consistent with a read-through product could be observed immediately above the primary band for MDH1, the basal read-through levels of this protein were below the limit of detection. ELX-02 treatment had no effect on the intensity of this potential MDH1 read-through This article has not been copyedited and formatted. The final version may differ from this version.

Proteomic analysis of ELX-02 effect on PBMCs
To explore the possibility of NSC read-through by ELX-02 using an unbiased proteomics approach, a read-through peptide database was generated from the NCBI RefSeq Database. In silico digestion of the real database generates about 1,895,000 unique peptide sequences. To monitor the false discovery rate (FDR), a decoy database was generated from the theoretically observable proteome, which generates an additional 1,527,000 unique peptides (not contained in the former). A total of 72,661 peptides were identified in ELX-02 and control treated PBMC extracts, corresponding to 49,133 modified peptide sequences. These were mapped to 6,086 protein group IDs. The mass spectrometry results were subsequently searched against our custom RefSeq database of human proteome sequences that include sequences at and after the NSC of each gene, to identify read-through-derived peptides while excluding masses that had matched within conventional coding regions.
The number of false discoveries expected from the read-through database of human transcriptome was determined by calculating the number of peptides in the real and read-through databases. To identify true read-through peptides, the peptide and protein FDR was set to 1%, which translates to an expectation of about 491 of the peptides identified in the real database to be false hits. By scaling for the respective sizes of the peptide ensembles, we predicted an approximate number of false discoveries from the readthrough database of about 396. These calculations are based on the assumption that, on average, the chemistry of both peptide ensembles is similar. Our analysis demonstrates that the number of readthrough sequences in ELX-02 treated samples (124 peptides) was lower than the predicted FDR levels of read-through peptides (396 peptides). The distribution of posterior error probability (PEPs, Figure 6), the probability that there is an error in identification of standard (red) or read-through (blue) peptides, indicates that read-through peptides tend to have higher PEPs than standard peptides, consistent with the conclusion that a high proportion of identified read-through peptides are false hits. This article has not been copyedited and formatted. The final version may differ from this version.

JPET # 265595
To estimate if any of the identified 124 peptides correspond to an actual read-through peptide, we used a bootstrapping strategy to estimate the probability distribution of random samples of 124 peptides picked from the standard distribution. The mean of the median of 1,000,000 samplings of 124 PEP values from non-read-through standard peptides identified in our experiment was 0.0021. The median PEP for the 124 identified read-through peptides was approximately 0.020 (0.00079), 10 times higher than the standard peptides mean. Thus, we have a population of read-through peptides with a significantly higher PEP than expected for the population of similarly sized samples suggesting that at least some of the read-through peptides represent actual read-through. The 124 peptides map to 108 different proteins. Of those 108 proteins, only seven proteins were identified with more than a single peptide, indicating a higher confidence of read-through detection (Supplementary Table 1). To test if any of the detected readthrough peptides in ELX-02 treated samples are significant, we used the Benjamini Hochberg FDR procedure to analyze P values using a high accepted FDR of 30%. The seven proteins with more than one peptide identification were detected in both control and ELX-02 treated cells. Signal intensities were not statistically different between control and ELX-02 treated cells, indicating that these proteins undergo natural read-through. Thirty-nine read-through peptides were identified only in control cells Table 1, p value = NA). At the lower ELX-02 concentration (20µg/mL) none of the P values showed statistically significant change from control cells, while only one protein showed statistical significance at 100µg/mL. With FDR set at 30%, there is a 30% chance of this hit being a false discovery. Therefore, these proteomics results do not provide strong evidence for induction of NSC read-through by ELX-02 in human PBMCs.

(Supplementary
This article has not been copyedited and formatted. The final version may differ from this version.

DISCUSSION
The studies reported here demonstrate that ELX-02 produced significant PSC read-through of an exemplar nonsense mutation in the TP53 gene of a cancer cell line, but did not increase the presence of extended protein variants indicative of NSC read-through in either DMS-114 cells or PBMCs derived from normal healthy human subjects. These studies build on previous work testing the potential for translational read-through of nonsense mutations by ELX-02 using in vitro models (Kandasamy et al., 2012;Xue et al., 2014;Bidou et al., 2017;Friesen et al., 2018;Brasell et al., 2019) and in vivo mouse models of CF and cystinosis Brasell et al., 2019). Collectively, these studies show that ELX-02 restores functional activity to proteins encoded by genes containing a PSC yet has limited impact on NSCs and supports its therapeutic potential in diseases resulting from nonsense alleles.
Given that nonsense mutations comprise a significant fraction of all alleles causing inherited disorders (Krawczak et al., 2000;Mort et al., 2008), drugs that promote PSC read-through could have broad therapeutic potential. Translational read-through inducing drugs like ELX-02 work by suppressing the recognition of nonsense mutations by translation termination machinery, thereby allowing insertion of near-cognate amino acids and production of functional full-length proteins. However, one aspect of tolerability for therapeutic nonsense suppression is the assurance that the compound does not suppress NSCs and generate aberrant proteins. It was hypothesized that if any proteins would show an increase in NSC read-through as a result of ELX-02 exposure, it would be proteins showing high expression levels across all cells or proteins with basal levels of NSC read-through that are higher than average. We found that ELX-02 does not generate or increase extended protein variants that would be consistent with NSC read-through of housekeeping genes (CysC and β2M) or high native read-through genes (VEGF-A, AGO1 and MDH1).
As targeted efforts to detect NSC read-through did not detect evidence of such activity, we undertook a proteomic approach to provide unbiased detection of read-through peptides for most human expressed proteins. Our analysis demonstrates that the number of read-through sequences in ELX-02 treated samples was lower than the predicted FDR levels of read-through peptides. Read-through peptides were also observed to have higher PEP distributions than standard peptides, consistent with the conclusion This article has not been copyedited and formatted. The final version may differ from this version. JPET Fast Forward. Published on May 6, 2020as DOI: 10.1124 at ASPET Journals on July 9, 2020 jpet.aspetjournals.org Downloaded from 21 JPET # 265595 that a high proportion of read-through peptides were false hits. Only seven proteins were identified with more than one peptide, which would be a signal for higher confidence in potential read-through. These peptides were observed in both control and ELX-02 treated samples and the signal intensity was not significantly different between control and treatment groups. Also, with only one protein with one peptide assignment showing a statistically higher intensity at 100 versus 20µg/mL ELX-02, in general no dosedependent response was observed. Since this top ELX-02 concentration corresponds to five-fold Cmax observed at the highest dose administered in a clinical study, these results suggest that ELX-02 does not induce general read-through at clinically relevant doses.
Recent work has similarly tested whether G418 induces stop codon read-through in either HEK293T cells carrying a reporter construct or Calu-6 cells that harbor an R195X mutation in the TP53 gene (Wangen and Green, 2019). Using ribosome profiling, they observed PSC read-through by G418 which was influenced by the stop codon identity and the surrounding mRNA sequence context, similar to what has been established in the read-through field (Floquet et al., 2011;Dabrowski et al., 2018). They also found that the majority of mRNAs were neither dramatically changed in abundance nor differentially translated in response to G418, but that significant NSC read-through by G418 led to substantial effects on translation of histone genes, selenoprotein genes, and S-adenosylmethionine decarboxylase (AMD1).
While this effect is worth attention, other work has found that translation past NSCs into the 3' untranslated region may encode peptide sequences that destabilize the attached protein, providing mitigation of unwelcome and varied translation errors (Arribere et al., 2016). This destabilization could explain the increased unfolded protein response observed in Calu-6 cells (Wangen and Green, 2019), as well as the lack of extended protein variants in our studies using Western blot detection of in vitro and ex vivo samples. Together, these data imply that ELX-02 could promote read-through at NSCs, but that evolutionary pressure has put cis-acting elements into place to limit NSC read-through and that cellular processes exist to provide efficient clearance of those potential NSC read-through protein products, thus limiting their abundance below detection limits.
In conclusion, ELX-02 produces significant read-through of PSCs leading to the production of full-length functional protein, while three complementary techniques suggest that ELX-02 does not promote read-This article has not been copyedited and formatted. The final version may differ from this version.      immediately above the primary band may represent this read-through product. Additional bands of higher molecular weight above these bands likely represent non-specific protein recognition by the antibody.
Recombinant proteins were run alongside each set of samples and provide information as to the lower limit of detection in each Western blot (1.5 and 3ng for AGO1 and MDH1, respectively). MW = molecular weight; M = MW marker; pre = pre-dose; 24h and 48h = samples taken 24 and 48 hours after ELX-02 treatment; D7 = samples taken 7 days after ELX-02 treatment. Samples were collected from clinical study This article has not been copyedited and formatted. The final version may differ from this version.
JPET Fast Forward. Published on May 6, 2020 as DOI: 10.1124/jpet.120.265595 at ASPET Journals on July 9, 2020 jpet.aspetjournals.org Downloaded from 32 JPET # 265595 EL-006. Codes above the blots refer to the patient identification number used to track the four samples collected from each trial participant. Figure 6. Distribution of PEPs for identified standard (red) and read-through (blue) peptides. Readthrough peptides tend to have higher PEPs than standard peptides, which is consistent with the idea that a high proportion of identified read-through peptides are false hits.
This article has not been copyedited and formatted. The final version may differ from this version.