Small molecule inhibitors of hnRNPA2B1-RNA interactions reveal a predictable sorting of RNA subsets into extracellular vesicles

Abstract

Extracellular vesicles (EVs) are cell-secreted membranous particles contributing to intercellular communication. Coding and noncoding RNAs can be detected as EV cargo, and RNA-binding proteins (RBPs), such as hnRNPA2B1, have been circumstantially implicated in EV-RNA sorting mechanisms. However, the contribution of competitive RBP-RNA interactions responsible for RNA-sorting outcomes is still unclear, especially for predicting the EV-RNA content. We designed a reverse proteomic analysis exploiting the EV-RNA to identify intracellular protein binders in vitro. Using cells expressing a recombinant hnRNPA2B1 to normalize competitive interactions, we prioritized a network of heterogeneous nuclear ribonucleoproteins and purine-rich RNA sequences subsequently validated in secreted EV-RNA through short fluorescent RNA oligos. Then, we designed a GGGAG-enriched RNA probe that efficiently interacted with a full-length human hnRNPA2B1 protein. We exploited the interaction to conduct a pharmacological screening and identify inhibitors of the protein-RNA binding. Small molecules were orthogonally validated through biochemical and cell-based approaches. Selected drugs remarkably impacted secreted EV-RNAs and reduced an RNA-dependent, EV-mediated paracrine activation of NF-kB in recipient cells. These results demonstrate the relevance of post-transcriptional mechanisms for EV-RNA sorting and the possibility of predicting the EV-RNA quality for developing innovative strategies targeting discrete paracrine functions.

Graphical Abstract

Figure 1.

Reverse proteomics using the EV-RNA to prioritize RBP binding partners. (A) Experimental workflow of reverse proteomic approach. The RNA was extracted from EVs and then enzymatically polyadenylated. A biotinylated oligo(dT) was hybridized and then incubated with magnetic Streptavidin beads to constitute the heterogeneous protein-capturing probe. Streptavidin beads with biotinylated oligo(dT) alone were included as technical negative controls. The probe was then incubated with cell lysates overexpressing or not hnRNPA2B1. Precipitated proteins were analysed by MS/MS and reported as a ratio between the two different cell lysates and from three independent experiments. The image was created on biorender.com. (B) Representative NTA profile of NSC34-recovered particles detected in scatter mode (left) and corresponding lipid particles positive to CMO detected in fluorescence mode (right). (C) Representative TEM negative stain acquisition of EVs recovered from NSC34 cells by NBI and then ultracentrifuged at 100 000 × g for 70 min at 4°C. A 20 500× magnification is shown. The scale bar corresponding to the original one embedded in gray, was added below for better recognition. (D) Immunoblotting using lysates of NSC34, HEK293T, and DU-145 cells and released EVs. Anti-GM130 and CALNEXIN primary antibodies were used as negative EV markers, while anti SYNTENIN antibody was used as positive EV marker. The gel is one representative of three independent experiments. (E) Representative bioanalyzer profile of the EV-RNA subjected to polyadenylation for reverse proteomics experiments. EV-RNAs were pooled as described in the experimental section. The black arrow indicates the enrichment of RNA fragments between 100 and 200 nt expected from vesicular RNAs. (F) Immunoblotting on NSC34 lysates expressing (OE) or not (Mock) the Myc-His-hnRNPA2B1 protein, recognized by anti-hnRNPA2B1 or anti-Myc primary antibodies. The shown gel is a representative of several independent experiments. Below, the volcano plot shows the differentially enriched or depleted proteins detected by MS and reported as a ratio between hnRNPA2B1 OE and Mock lysates. Enriched proteins (Class “+”), depleted proteins (Class “−”), and unchanged proteins (Class “=”) are indicated. (G) Relative abundance, in three independent experiments, of selected hnRNPs significantly enriched in the hnRNPA2B1 OE condition. (H) Network of experimentally validated interactions among the identified hnRNPs obtained from STRING database (https://string-db.org). Red circles indicate hnRNPs significantly enriched in the ratio OE/Mock; pink circles indicate hnRNPs showing a trend of enrichment; Green circles indicate hnRNPs showing a trend of displacement. Raw data are available in Supplementary Table S1. (I) Relative abundance of recombinant (Myc-His-hnRNPA2B1) and endogenous (hnrnpa2b1) proteins in the ratio OE/Mock. (J) Sequence motifs analysed by MEME toolbox (https://meme-suite.org/meme) using experimentally validated RNA sequences present in RPDB database (http://rbpdb.ccbr.utoronto.ca) and recognized by the identified hnRNPs..

Figure 2.

The interaction of recombinant hnRNPA2B1 with purine-rich RNA constitutes a platform for high-throughput drug screening. (A) Representative flow cytometry acquisition of NSC34 cells transfected for 24 h with 150 nM with transfection reagent alone (Mock, sense, and antisense fluorescent RNA oligos). Forward and side scatters were chosen to obtain the best cell distribution over the threshold line of the background (population P1). Subgatings in the SSC-A/SSC-W plot were considered to avoid including potential cell doublets, as shown by population P2, before recording the probe signals in the PE-A channel (population P3). Histogram reports mean and data distribution of three independent experiments. (B) Representative dot blot performed as described in the experimental section. Treatment with RNAse A was performed with 1 μg/ml of enzyme. Sense and antisense control probes were spotted at 1 nM concentration. Histograms reports data of several experiments; the indicated significance corresponds to a P-value < .01. (C) Representative REMSA performed at equilibrium with 110 nM of GST-hnRNPA2B1 or Biotin-GST and 25 nM of RNA probe. Arrows indicate protein–RNA complexes with increasing molecular weights and protein oligomerization phases on the same RNA molecules. Supershifts with primary antibody are poorly detectable for the different high molecular weight complexes formed in these conditions. (D) Representative plot of saturation binding AlphaScreen experiments to measure the GST-hnRNPA2B1 binding to different biotinylated RNA probes, as described in the text. Dissociation constants (K_d) were determined from nonlinear regression one-site binding model in GraphPad Prism^®, version 9.0. Mean and standard deviation values derive from two independent experiments with separate GST-hnRNPA2B1 protein purifications. (E) Kinetic experiment carried out with 30 nM of GST-hnRNPA2B1 and increasing concentrations of RNA EXOmotif. Association (K_on) and dissociation (K_off) rate constants and equilibrium dissociation constants (k_off/k_on) were determined using nonlinear regression association kinetic model of multiple ligand concentration in GraphPad Prism^®, version 9.0. Mean and standard deviation derive from two independent experiments with the two protein purifications used in panel (C). (F) Representative immunoblotting with anti-hnRNPA2B1 antibody following EXOmotif-, RNA 114-, and RNA 276-based RNA PD. These experiments were paralleled by an immunoblot showing the higher affinity of the EXOmotif RNA probe compared to the ARE probe. Densitometric quantification of protein fractions were normalized to the input. Mean and SD derive from three independent experiments. (G) Distribution of positive (EXOmotif) and negative (ARE) protein–RNA probe interactions calculated in the primary drug screening. Relative coefficient of variation (CV) and Z-factor value are indicated. (H) Plot of compounds ranked according to percent of control (POC) normalized (DMSO) values.

Figure 3.

Orthogonal validation of small molecules by biochemical and cell-based assays. (A) Counter screening by RNA electrophoresis mobility shift assay (REMSA) showing the activity of AlphaScreen-hit compounds on reducing the hnRNPA2B1–RNA complex formation. (B) AlphaScrenn dose-response curves with RNA EXOmotif. The IC₅₀ was determined from nonlinear regression using one site fitting model of GraphPad Prism 9. The “*” in the figure indicates “incomplete inhibition”. (C) AlphaScreen using 14 nM of RNA 114 probe and 250 nM of compounds. (D) Chemical structure of the six AlphaScreen/REMSA hit compounds. Right-hand arrow indicates a chemical space of Theaflavin digallate shared with a hematein. (E) Absorbance spectra relative to hematein and phenothrin mixed with RNA EXO motif or ARE probe. Arrows in the inserts magnify the main spectroscopic changes retrieved at 230 nm. (F) RNA PD and immunoblotting of Hnrnpa2b1 using the biotinylated RNA EXOmotif probe upon cell treatment with compounds. Densitometric quantification of the protein is reported as fraction normalized on input-FT and DMSO. Histogram reports data of two independent experiments.

Figure 4.

Post-transcriptional activity of small molecules at intracellular and EV-RNA levels. (A) Immunoblotting using protein lysates from NSC34 cells transfected with the plasmid encoding the Myc-tagged hnRNPA2B1 (OE) or without DNA (Mock) and with siRNA pools targeting hnrnpa2b1 (siA2B1) or scramble (SCR). Densitometric quantification of all bands was plotted as a ratio with Calnexin. Mock and OE conditions in the histogram refer to the Myc-tag bands. (B) Particle concentration determined by NTA using Nanosight NS300 (left). Data were reported after normalization with relative cellular protein content in each condition. The EV-RNA concentration determined by Bioanalyzer was normalized on data reported on the left upon protein expression or silencing (right). Histograms report data of three independent experiments. (C) RIP analysis using cytoplasmic lysates of cell treated with compounds and anti-hnRNPA2B1–TDP43, and-GAPDH primary antibodies. IgG-conjugated magnetic beads were recovered and divided for protein detection and RNA isolation. Equivalent amount of IP RNA was used for cDNA synthesis and digital droplet PCR with TaqMan probes for miR-221, miR-1910, and miR-126 detection. The transcript copy number was normalized on relative protein levels and DMSO condition. Histogram reports data of three independent experiments. The indicated significance corresponds to a P-value < .05. (D) Relative abundance of vesicular RNA recovered after cell treatment and normalized on the particle number detected by NTA. (E) Relative miR-221–3p copy number detected by ddPCR from NSC34-derived EV-RNA and cDNA synthesis upon compound treatments. Transcript copy number was normalized on Y3 RNA levels and DMSO condition. (F) Relative miR-221–3p copy number detected by ddPCR from motor neurons-derived EVs upon compounds treatment. Transcript copy number was normalized on Y3 RNA levels and DMSO condition. *P-value < .05; *** P-value < .001; ****P-value < .0001.

Figure 5.

Paracrine effect on NF-kB activation of EVs deriving from RBP-conditioned and treated cells. (A) NF-kB activation in HEK293T upon acute treatment with EVs recovered from Mock or recombinant hnRNPA2B1-expressing cells (OE). Firefly luminescence was normalized on cellular protein content at the end point on the corresponding sample. Histogram reports mean and data of four independent experiments. The concentration of TNFα was 10 ng/ml. The concentration of RNAse A in the media was 2 μg/ml. Particles used to treat HEK293T cells were characterized by NTA (shown in Supplementary Fig. 5C) and the corresponding diameter is reported in the top right side. To have indication of NF-kB activation in EV-recipient cells, we performed an immunoblot using an anti-NF-kB antibody. *P-value < .05; ***P-value < .001; ****P-value < .0001. (B) Post-transcriptional control of EV RNA. Schematic representation of cytoplasmic events conveying RNA subsets into secreted EVs. RBPs, and in particular hnRNPs, could compete for and select RNAs. Interactions with other trans-acting factors, virtually RAB proteins participating in vesicular trafficking dynamics, could contribute to finalizing the RNA sorting process. Small molecule inhibitors of protein–RNA interactions, such as hematein or phenothrin, could alter intracellular RNA homeostasis and, ultimately, the distribution of the secreted counterpart, impacting paracrine functions. Direct consequences following the RNA recognition interference (?a) and vesicular packaging (?b) are still unclear. Created in BioRender. D’agostino, V. (2025) https://BioRender.com/h82u817.