Mature microRNA-binding protein QKI suppresses extracellular microRNA let-7b release

MicroRNAs (miRNAs) assemble into the RNA-induced silencing complex (RISC) with argonaute (AGO) proteins and guide the RISC to target RNAs through base pairing with them (Iwakawa and Tomari, 2022). The miRISC associates with the 3′ untranslated region of a target mRNA, leading to translation repression and degradation (Bagga et al., 2005; Humphreys et al., 2005; Wilczynska and Bushell, 2015). The steady-state levels of miRNAs have been linked to multiple human diseases (Carè et al., 2007; Fiore et al., 2008; Lu et al., 2005), with dysregulation of miRNA particularly in neurological diseases including brain tumors and neurodegeneration (Hébert and De Strooper, 2009; Nicoloso and Calin, 2008). Considering these findings, understanding the molecular mechanisms of how the abundance of miRNAs is regulated could contribute to the prognosis, diagnosis and treatment of diseases.

The cellular concentration of total mature miRNAs was reported to be higher than that of AGO, with experimental limitation in the quantification of mature miRNAs (Janas et al., 2012). We also previously identified that there are plenty of mature miRNAs that are not engaged with AGOs, which was demonstrated by using serial depletion of AGOs, followed by miRNA sequencing and quantification of mature miRNAs from AGO-depleted cells (Min et al., 2024). In addition, numerous studies have shown that miRNAs exist in circulating blood and body fluids (Chen et al., 2008; Mitchell et al., 2008; Weber et al., 2010). These findings indicate that the abundance of cellular miRNAs could be governed by cellular machineries modulating miRNA decay and even excretion of mature miRNAs outside cells.

miRNAs released into extracellular fluids can be found either inside extracellular vesicles (EVs) or bound to AGO2 independently of vesicles (Arroyo et al., 2011; Gallo et al., 2012; O'Brien et al., 2018; Turchinovich et al., 2011; Weaver and Patton, 2020). Mature miRNAs in EVs contain the EXOmotif sequence (CGGAG), which is distinct from cellular retention motifs (CELLmotifs) (Garcia-Martin et al., 2022). Extracellular mature miRNAs are reported to be capable of mediating cell-to-cell communication (Makarova et al., 2021) and are associated with various pathological conditions (Mori et al., 2019). Accumulated studies have demonstrated that miRNAs transported to recipient cells can serve as ligands for Toll-like receptors (TLRs), which are part of the innate immune system that recognizes the invasion of pathogens (Chen et al., 2013; Fabbri et al., 2012; Lehmann et al., 2012). Notably, emerging findings emphasize the roles of TLR7 in sensing extracellular miRNAs in neurological diseases (Buonfiglioli et al., 2019; Lehmann et al., 2012; Yelamanchili et al., 2015). Additionally, extracellular let-7, a mature miRNA, was reported to induce neurodegeneration (Lehmann et al., 2012) and modulate microglial function and glioma growth (Buonfiglioli et al., 2019) by functioning as a ligand of TLR7. Despite the relevance of miRNAs to brain diseases, the regulatory mechanisms of the extracellular miRNA levels are not clearly understood. Notably, a previous study reported that AGO has not been detected in EVs, implicating the existence of novel miRNA-binding proteins (miRBPs) modulating extracellular miRNA release (Jeppesen et al., 2019). Based on these findings, we aimed to profile unknown miRBPs participating in intercellular communication via extracellular miRNAs.

In this study, we showed that several types of RNA-binding proteins (RBPs), including QKI, which is implicated in mRNA splicing, export and decay (De Bruin et al., 2016; Larocque et al., 2002; Thangaraj et al., 2017), and circular RNA biogenesis (Conn et al., 2015), directly bind to a subset of mature miRNAs (let-7, miR-21 and miR-130b). Remarkably, depletion of QKI enhanced extracellular release of let-7b managed by miRBP AU-rich element RBP 1 (AUF1, also known as hnRNPD) and hnRNPK, leading to activation of TLR7 in the recipient cells. Comprehensively, this study reveals a novel role for QKI in brain inflammation and provides novel insights into extracellular miRNA release.

The previous low-throughput approach for identifying RBPs interacting with mature miRNAs was limited to the one-by-one examination of RBPs and miRNAs (Zealy et al., 2017). In the present study, we advanced the previous approach for identifying miRBPs by using human protein arrays immobilized with 354 RNA-binding domains (Chen et al., 2020) or ∼17,000 recombinant proteins (Jeong et al., 2012). To identify human proteins interacting with mature miRNAs, we incubated the array slide with the tumor-suppressive miRNA let-7b (Johnson et al., 2005; Lee and Dutta, 2007; Takamizawa et al., 2004), the oncogenic miRNA miR-21 (Asangani et al., 2008; Chan et al., 2005; Meng et al., 2007; Seike et al., 2009) and miR-130b, an miRNA induced during differentiation of adipocytes and a negative regulator of adipogenesis (Lee et al., 2011; Luo et al., 2022; Wang et al., 2013). These miRNAs were labeled at the 5′ end with biotin (Fig. 1A) or internally labeled with Cy5 or DY647 (Jo et al., 2015; Yoon et al., 2015).

In the RNA-binding domain array, we identified 113 RBPs directly interacting with let-7b, 34 interacting with miR-21 and 69 with miR-130b (cutoff <500 original intensity). There were 29 RBPs commonly interacting with three miRNAs, whereas 44 RBPs uniquely bound exclusively with let-7b (Fig. 1B; Table S1). When we profiled RBPs differentially interacting with let-7 family miRNAs (let-7a, let-7b, let-7c and let-7d), we observed that 46 RBPs bound all four miRNAs and 42 bound only let-7b, let-7c and let-7d (Fig. S1A,B).

In the array with ∼17,000 proteins, we identified 60 human proteins that directly interacted with let-7b, 31 that directly interacted with miR-21 and 40 that directly interacted with miR-130b (as normalized to the intensity of GST binding, which was also used for quality control of individual protein spots). Among these 60 proteins, only six interacted with all three miRNAs: (1) ADP-ribosylation factor-like protein 10 (ARL10), (2) splicing factor homolog ISY1, (3) the RBP quaking (QKI), (4) Zic family member 3 (ZIC3), (5) zinc finger protein 490 (ZNF490) and (6) zinc finger protein 597 (ZNF597) (Fig. S1C, Table S2). We did not detect AGO family proteins and let-7b binders, such as HuR (also known as ELAVL1) and AUF1, possibly because of low levels of these proteins on the array (Yoon et al., 2013, 2015) and steric hindrance associated with immobilization of these proteins on the array.

Gene Ontology analysis of the 60 miRNA-interacting proteins revealed that they are involved in diverse cellular processes, such as cation transmembrane transport for let-7b, metal ion binding for miR-21 and DNA-binding for miR-130b (Table S3). Pulldown of mature miRNAs (let-7b, let-7i, miR-141c or miR-200) followed by mass spectrometry also profiled a series of canonical and non-canonical RBPs (Fig. S2A–D) with common protein–protein interaction networks (Fig. S2E). Overall, we successfully identified a number of potential novel miRBPs that interact with let-7b, miR-21 and miR-130b.

QKI is reported as an RBP implicated in mRNA splicing, export and decay (de Bruin et al., 2016; Larocque et al., 2002; Thangaraj et al., 2017), as well as in circular RNA biogenesis (Conn et al., 2015). Notably, photoactivatable ribonucleoside-enhanced crosslinking and immunoprecipitation (PAR-CLIP) analysis showed that mature miRNAs, including let-7b and miR-21, are associated with QKI (Hafner et al., 2010). In addition, we recently reported that QKI functions in miRNA-mediated gene silencing by promoting interaction of AGO2 with let-7b (Min et al., 2024). Considering these findings, we focused on QKI as the RBP that binds with mature miRNAs directly. To verify the protein array results, we performed binding assays of QKI with let-7b, our primary focus in this study, using mobility shift. We incubated recombinant histidine-tagged QKI (purified from Escherichia coli) with let-7b, followed by ultraviolet light (254 nm)-induced cross-linking, resolution using SDS-PAGE, and immunoblotting against the 6× histidine tag. Our western blot analysis confirmed that cross-linking of QKI with let-7b resulted in a significantly decreased electrophoretic mobility, suggesting an interaction between QKI and let-7b (Fig. 1C). Our fluorescence polarization assay revealed the binding affinities of recombinant QKI with let-7b (K_d=8.071±3 nM), miR-21 (K_d=48.45±17 nM) and miR-130b (K_d=7.032±3 nM) (Fig. 1D; Fig. S1D). YUAAY-mutant let-7b, lacking the sequence for QKI interaction, failed to bind recombinant QKI, as calculated for mutant let-7b compared to wild-type let-7b (K_d=8.071±3 nM) (Fig. 1D). This binding affinity is comparable to the high affinity of AGO2 for miRNAs (Tan et al., 2009), suggesting that QKI is a novel miRBP. Ribonucleoprotein (RNP) immunoprecipitation quantitative PCR (RIP-qPCR) of QKI enriched let-7b in HeLa cells and mouse embryonic fibroblasts (MEFs), regardless of the presence of AGO2 (Fig. 1E,F).

The QKI-binding sites on target mRNAs have previously been characterized with the conserved motif YUAAY (Teplova et al., 2013) and ACUAAY(N1–20)UAAY using systematic evolution of ligands by exponential enrichment (SELEX) (Galarneau and Richard, 2005), and AYUAAY using PAR-CLIP (Hafner et al., 2010). Although none of these motifs are present in let-7b (5ʹ-UGAGGUAGUAGGUUGUGUGGUU-3ʹ), miR-21 (5ʹ-UAGCUUAUCAGACUGAUGUUGA-3ʹ) or miR-130b (5ʹ-CAGUGCAAUGAUGAAAGGGCAU-3ʹ), the top 20 miRNAs identified from QKI PAR-CLIP share the motif WSUDSU (identified by the MEME suite) (Bailey et al., 2009), which is different from the aforementioned three motifs.

To support this finding, we performed simulated molecular docking of QKI with let-7b, miR-21 and miR-130b. Although the overall bound conformation of let-7b was not a perfect fit for that of the substrate (Fig. S3A,B), the UAA sequence in the YUAAY motif fit well with the UGU sequence in the let-7b RNA (Fig. S3C). However, our binding affinity calculations showed that there are no critical motifs in let-7b for binding QKI (Fig. S3D–F). These data indicate that QKI can bind let-7b, which does not have any consensus sequence (Galarneau and Richard, 2005; Hafner et al., 2010; Teplova et al., 2013). The fitting of both miR-21 and miR-130b was similar to that of let-7b, with no essential motifs for binding to QKI (Fig. S3G–L).

The existence of other miRBPs, in addition to AGO proteins, prompted us to speculate that QKI titrates the availability of AGO2-free miRNAs for packaging into EVs for paracrine signaling. Thus, we hypothesized that QKI regulates EV-mediated miRNA release. To test this, EVs were purified using sucrose density gradient centrifugation as performed previously (Kang et al., 2024; Kwon et al., 2016) (Fig. S4A,B). We first compared the levels of QKI target miRNAs (let-7b, miR-21 and miR-130b) in EVs from QKI shRNA-treated and control cells. Reverse transcription quantitative PCR (RT-qPCR) analysis showed that the level of let-7b in EVs was significantly higher in QKI shRNA-treated cells than in control cells (Fig. 2A); the levels of miR-21 and miR-130b in EVs were not increased significantly in QKI shRNA-treated cells compared to control cells. We also compared the expression of these three miRNAs inside cells and found that the levels of cellular miRNAs did not change significantly in QKI shRNA-treated cells (Fig. 2B). Small RNA-sequencing analysis demonstrated that, among let-7 family miRNAs, let-7b and let-7c were enriched in EVs after silencing QKI in HeLa cells (Fig. S4A). The abundance of let-7b from whole medium or mouse serum before EV purification exhibited similar changes to those after EV purification (Fig. S4C). In contrast, overexpression of QKI significantly decreased the abundance of let-7b in EVs compared to that in empty vector control-treated cells (Fig. 2C), with no significant difference in cytosolic miRNA expression (Fig. 2D) These results demonstrate that QKI prevents release of let-7b via EVs.

Next, we reasoned that if QKI depletion promotes let-7b accumulation in EVs, there should be cargo proteins to carry the let-7b out of cells. To address this question, we used mass spectrometry to profile proteins loaded into EVs. Among 321 proteins, we identified five RBPs (AUF1, DHX9, eIF4A1, hnRNPH and hnRNPK), but not QKI, AGO2 or HuR. The absence of AGO2 from our EV protein profiling is consistent with the recent finding in a previous publication (Jeppesen et al., 2019). RNA-binding domain array data also identified a KH domain critical for miRNA binding (Fig. 2E).

Among the five RBPs, AUF1 and hnRNPK are known to interact directly with let-7b and miR-21 (Ray et al., 2013; Yoon et al., 2015). Depletion of AUF1 or hnRNPK, in combination with QKI shRNA (Fig. 2F) suppressed let-7b enrichment in EVs (Fig. 2G), although the overall level of let-7b was unchanged (Fig. 2H). When we compared the levels of the exosomal protein CD63, its abundance did not change significantly (Fig. 2F). Our findings demonstrate that the let-7b-binding proteins AUF1 and hnRNPK are responsible for accelerated let-7b release when QKI is silenced. In this study, as we did not scrutinize the intracellular events underpinning the engagement of AUF1 and hnRNPK in the miRNA-packing process for exosomal release, it is plausible that these proteins could be involved in the LC3-conjugation machinery for specifying the cargo packages into EVs (Leidal et al., 2020). This suggests that increase of the AUF1/hnRNPK-let-7b complex when QKI levels are limited could contribute to let-7b release.

A subset of extracellular miRNAs has been reported as ligands for the TLR family sensing viral single-stranded RNAs, which in turn prompted us to hypothesize that EV-loaded let-7b activates the TLR7 signal in endosomes of recipient cells to promote inflammatory responses in the central nervous system (CNS)/brain and cancer cells (Buonfiglioli et al., 2019; Fabbri et al., 2012; Lehmann et al., 2012). This led us to examine whether accelerated exosomal release of let-7b, upon QKI depletion, activates TLR7 and subsequent inflammation.

To this end, we compared the abundance of four TLR7-target mRNAs [IFNA1, IFNB1, IL1B and TNFA (also known as TNF)] in the recipient cells (SH-SY5Y), after the cells were treated with EVs isolated from control or QKI shRNA-treated cells (HeLa) (Fig. 3A). Our RT-qPCR analysis showed that the recipient cells exposed to EVs from QKI shRNA-treated cells significantly increased the phosphorylation of STAT1 at Ser727 in a TLR7-dependent manner (indicating TLR7 activation) (Fig. 3B) (Ramirez-Ortiz et al., 2015).

We also observed that neuroblastoma cells exposed to EVs from QKI-depleted HeLa cells expressed significantly more IFNA1 and IL1B mRNA than did neuroblastoma cells exposed to EVs from control shRNA-treated cells. However, the abundance of IFNB1 and TNFA mRNA was not changed significantly (Fig. 3C). More strikingly, induction of IFNA1 and IL1B mRNAs was suppressed by depletion of TLR7 in the recipient cells (Fig. 3C). TLR7 activation was also confirmed by measuring IRF7 occupancy on the promoter regions of IFNA1 (Fig. 3D). Our results show that EVs originating from QKI-depleted donor cells significantly increase the degree of IRF7 binding to its target gene promoters in comparison with EVs from control shRNA-treated cells. Our findings above demonstrate that accelerated let-7b release via EVs after QKI depletion activates the TLR7 pathway in the recipient cells (Fig. 3E).

If QKI prevents extracellular let-7b release and subsequent inflammatory responses, Qki-expressing cells in mice could function in sequestering let-7b inside cells. To determine which cell types express Qki (Qki5, Qki6, Qki7 and Qki7b isoforms) in the cerebral cortex, we performed immunostaining with antibodies detecting proteins expressed exclusively in one cell type of mouse brain: NeuN (encoded by Rbfox3) or NF (encoded by Nefm) for neurons, GFAP for astrocytes, CNPase (encoded by Cnp) for oligodendrocytes, and Iba1 (encoded by Aif1) for microglia, respectively. We also used antibodies recognizing Qki5 only (anti-Qki5 antibody) or all Qki isoforms (anti-pan-Qki antibody) (see more details in the Materials and Methods).

First, we analyzed the expression patterns of Qki5 in Nestin-CreER^T2 Qki^+/+ and Nestin-CreER^T2 Qki^−/− mice, which have Qki deleted (exon 2 common in all isoforms; Shingu et al., 2017) (Fig. 4A,B) in neuroectoderm-originated cells, such as neurons, astrocytes and oligodendrocytes. Qki5 staining was mainly observed in nuclei, which overlapped with GFAP, CNPase and Iba1 but not with NeuN signals in Qki^+/+ mice (Fig. 4A). In the cerebral cortex from Plp-CreER^T2 Qki^+/+ mice, the Qki5 signal overlapped with signals from GFAP and CNPase as well (Fig. 4B). Strikingly, the Nestin-CreER^T2 Qki^−/− genotype resulted in the depletion of Qki5 in astrocytes and oligodendrocytes, whereas the Plp-CreER^T2 Qki^−/− genotype resulted in the depletion of Qki5 only in oligodendrocytes (Fig. 4A,B). Residual staining from anti-Qki5 and anti-pan-Qki antibodies in Qki^−/− mice might come from Qki in microglia or truncated Qki proteins containing at least exon 1, considering Qki^−/− as a functional knockout (KO).

In quantitative analyses of Qki5-positive cells, Nestin-CreER^T2 Qki KO showed the knockout from 64±8.3% to 3.7±3.7% in astrocytes and from 29.8±3.4% to 14.8±4.7% in oligodendrocytes (Fig. 4A, bottom). Plp-CreER^T2 Qki KO showed the knockout from 62.2±10.4% to 68.7±13% in astrocytes and from 28.7±4.1% to 8.3±3.1% in oligodendrocytes (Fig. 4B, bottom). Additionally, both Nestin-CreER^T2 and Plp-CreER^T2 Qki KO showed no alteration of Qki5 expression in microglia compared with that in each control, as microglia originate from the mesoderm but not the neuroectoderm, which are affected by the Nestin or Plp promoter, respectively. Intriguingly, both Nestin-CreER^T2 and Plp-CreER^T2 Qki KO showed increased Iba1 expression from 100 stained Qki5-positive cells, indicating increase of neuroinflammation. The structural abnormality of oligodendrocytes in Qki^−/− mice also supports our findings. Thus, these results demonstrate that Qki is expressed in astrocytes and oligodendrocytes of the normal cerebral cortex and its knockout was accomplished successfully for inducing microglial activation in both Nestin-CreER^T2 and Plp-CreER^T2 Qki KO mice.

To determine how Qki deficiency in the cerebral cortex promotes brain inflammation via let-7b, we traced cell types targeted by let-7b in Qki-deficient brains. As let-7b is recognized by its receptor, TLR7, we reasoned that TLR7-expressing cells are mainly affected by let-7b, the release of which is promoted by Qki deficiency. To address this, we performed immunostaining using sections from Nestin-CreER^T2 Qki^+/+ and Qki KO mice. Qki deletion increased TLR7 signals, which exclusively overlapped with NeuN signals (Fig. 4C) but not pan-Qki (Fig. S4E,F) expressed in microglia (Fig. S4G,H). Staining of double-stranded RNAs using the J2 antibody also mainly overlapped with NF only in Qki KO samples (Fig. S5A), without significant changes in staining of their receptors, TLR3 (Fig. S5B). Staining of J2 and TLR7 also overlapped only in Qki KO samples (Fig. S5D) implicating overlapping function of let-7b and double-stranded RNAs. Strikingly, microglia stained with Iba1 surrounded neurons stained with NeuN and TLR7 in Nestin-CreER^T2 Qki^−/− compared with Qki^+/+ samples (Fig. 4C). These results suggest that neuroinflammation induced by Qki deletion is initiated independent of microglia. Altogether, Qki deletion mainly happens in oligodendrocytes, resulting in destruction of their process structure and recruitment of microglia to neurons wrapped abnormally by processes of Qki-deficient oligodendrocytes.

To examine increase in inflammation in Qki-deficient brains, we studied sections of brains from Qki^+/+ and Qki^−/− mice (Shingu et al., 2017). Brain lysates from Qki^−/− mice exhibited significantly increased levels of phosphorylated Stat1 at Ser727 (Fig. 4D). Similar to our observations in EVs from QKI-depleted human cells, blood EVs purified from Qki^−/− mouse sera were enriched in the miRNA let-7b, but not miR-21 and miR-130b (Fig. 4E). The abundance of let-7b in the whole sera of Qki^−/− mice also showed similar changes compared to that in Qki^+/+ mice (Fig. S4C). The let-7b released from the host cells was indeed absorbed into the recipient cells, based on our results from miRNA labeling with ethynyl-labeled uridine (5-EU) and quantification in the host and recipient cells with RT-qPCR (Fig. 5). Our results support our hypothesis that let-7b in EVs acts as a paracrine signal. These results confirm that QKI is required for suppression of CNS/brain inflammation via regulation of the let-7b–TLR7 pathway.

It has been reported that the biosynthesis of miRNA, a major regulator of gene expression influencing various aspects within the cell, is tightly regulated by multiple RBPs (Rüegger and Großhans, 2012; Treiber et al., 2017). Recently, a number of studies have reported that RBPs directly interact with mature miRNA (Treiber et al., 2017; Yoon et al., 2014, 2015), indicating that RBPs are involved in the regulation of mature miRNA function as well as in the biosynthesis of miRNA. In this study, we profiled the RBPs associated with miRNAs with different functions to identify RBP candidates essential for the function of each miRNA (Fig. 1A,B). Interestingly, we discovered that the most RBPs binding with let-7b are not AGO family proteins (Fig. 1B), despite them being known to mediate miRNA-mediated gene silencing (Fabian and Sonenberg, 2012; Behm-Ansmant et al., 2006; Schmitter et al., 2006). These results imply that the activity or abundance of let-7b is tightly regulated by multiple RBPs.

Recently, extracellular let-7b was reported to act as a direct ligand of RNA-sensing TLR7 in macrophages and neurons (Lehmann et al., 2012). We hypothesized that the RBP identified by protein arrays might function in the extracellular release of let-7b and demonstrated that QKI-mediated suppression of let-7b release contributes to brain inflammation via regulation of intercellular communication (Fig. 4). In addition, we have shown that depletion of QKI in HeLa cells enhances let-7c abundance in EVs (Fig. S4A). Although let-7c is not a ligand of TLR7, these results indicate that QKI functions in the loading of let-7c as well as let-7b into EVs and suggest that let-7c is also associated with brain inflammation by silencing let-7 target genes in the recipient cells. Previously, we revealed that QKI promotes AGO2/let-7b-mediated gene silencing and functions as a tumor suppressor in HeLa cells (Min et al., 2024). We speculate that the presence of QKI enhances the role of let-7b in gene silencing, whereas the absence of QKI accelerates the loading of let-7b into EVs (Fig. 3E). In conclusion, QKI is a key mediator modulating the function of mature let-7b as a master regulator of brain inflammation (Fig. 3E).

Notably, our results showed that QKI depletion did not affect the abundance of miR-21 and miR-130b in the EVs (Fig. 2), which suggests that the regulatory mechanism of miR-21 and miR-130b loading into EVs could be different from that of let-7b. All in all, we profiled various RBP candidates for mature miRNAs, and our findings can be utilized to elucidate the regulatory mechanisms of miRNA loading into EVs in multiple human diseases. Collectively, our research contributes to our understanding of extracellular miRNA metabolism and provides novel insights into it.

Human HeLa, MEFs and SH-SY5Y cells were cultured in Dulbecco's modified Eagle medium (DMEM; Invitrogen) supplemented with 10% (v/v) fetal bovine serum (FBS; Corning, #35-015-CV) and antibiotics. Using polyethylenimine, the cells were transfected with shRNA plasmids designed to QKI (5′-GCTCAGAACAGAGCAGAAATC-3′) (Novikov et al., 2011), AUF1 (5′-CGAAGGAACAATATCAGCAA-3′), hnRNPK (5′-GATGTTTGATGACCGTCGCG-3′) or TLR7 (5′-CAGGAGTCTGACGAAGTATT-3′). pcDNA3 (Invitrogen) and pDESTmycQKI (Addgene, #19870) were used for overexpression experiments. Cells were collected for RT-qPCR, western blotting and EV purification 48 h after transfection.

For immunoprecipitation (IP) of RNP complexes from cell lysates (RIP analysis) (Yoon et al., 2012), cells were lysed in protein extraction buffer (PEB) with 20 mM Tris-HCl at pH 7.5, 100 mM KCl, 5 mM MgCl₂ and 0.5% NP-40 for 10 min on ice, and centrifuged at 10,000 g for 15 min at 4°C. The supernatant was then incubated for 1 h at 4°C with protein A-Sepharose beads coated with antibodies recognizing QKI (1 µg/1 ml lysate, ab126742, Abcam), AGO2 (1 µg/1ml lysate, ab57113, Abcam) or control IgG (1 µg/1 ml lysate, sc-2025, Santa Cruz Biotechnology). After the beads were washed with NT2 buffer (50 mM Tris-HCl at pH 7.5, 150 mM NaCl, 1 mM MgCl₂ and 0.05% NP-40), the complexes were incubated for 15 min at 37°C with 20 units of RNase-free DNase I (EN0521, Thermo Fisher Scientific). They were finally incubated for 15 min at 55°C with 0.1% SDS and 0.5 mg/ml Proteinase K (KB-0111, Bioneer) to remove the remaining DNA and proteins, respectively. The RNA isolated from the IP materials was further assessed by RT-qPCR analysis using the primers given in Table S4. Normalization of RIP results was carried out by quantifying, in parallel, the relative levels of housekeeping RNAs such as GAPDH mRNA and U6 snRNA in each IP sample. These abundant RNAs are non-specifically recovered during IP reactions.

Whole-cell lysates prepared in PEB were separated by SDS-polyacrylamide gel electrophoresis (SDS-PAGE) and transferred onto nitrocellulose membranes (Invitrogen iBlot Stack). Anti-β-actin (sc-47778), anti-IRF7 (sc-74472) and anti-α-tubulin (sc-5286) antibodies were purchased from Santa Cruz Biotechnology. Anti-Dicer (ab82539), anti-AGO2 (ab57113), anti-CD63 (ab59479) and anti-QKI (ab126742) antibodies were purchased from Abcam. Anti-pSTAT1 (8826S), anti-GAPDH (2118S), anti-hnRNPK (9081S) and anti-STAT1 (14994S) antibodies were purchased from Cell Signaling Technology. Anti-TLR7 (NBP2-24906) antibody was purchased from Novus Biologicals. Anti-AUF1 (07-260) antibody was purchased from Millipore. All primary antibodies were used at 1:1000. The HRP-conjugated secondary antibodies (1:5000, 7074S) were purchased from GE Healthcare.

For RIP analysis, Ribozole (Amresco) was used to extract total RNA, and acidic phenol (Ambion) was used to extract RNA. Reverse transcription was performed using random hexamers and reverse transcriptase (Maxima, Thermo Fisher Scientific), and real-time quantitative (q)PCR using gene-specific primers (Table S4) and SYBR Green Master Mix (Kapa Biosystems) using a Bio-Rad iCycler instrument.

miRNA quantitation was performed after RNA extraction from immunoprecipitated samples (with anti-AGO2 antibody or control IgG), polyadenylation (QuantiMiR kit, #RA420A, System Biosciences), and hybridization with oligo-dT adaptors. After reverse transcription, cDNAs were quantitated by qPCR with miRNA-specific primers, or with primers to detect the control transcript U6 snRNA, along with a universal primer. U6 snRNA was used for normalization of exosomal miRNA (Xu et al., 2017; Zhang et al., 2018).

This assay was performed according to modifications of a previously described method (Chen et al., 2020). GST recombinant proteins were designed to include reported and suspected RNA-binding domains and synthesized by Biomatik Corporation in the pGEX-4T1 vector (Amersham). Recombinant proteins were expressed and purified using the GST-fusion system. GST proteins were then arrayed, in duplicate, using an Aushon 2470 Microarrayer (Quanterix) onto nitrocellulose-coated glass slides (Grace Bio-Labs). The RNA-binding domains array contains approximately 354 GST recombinant proteins, which include SAP, RRM, KH, DSRM and YTH domains. Full-length GST- and His-fusion proteins were printed, in duplicate, onto nitrocellulose-coated glass slides.

To screen for potential RNA–protein interactions, the RNA-binding domains array was probed with biotinylated-miRNA oligonucleotides. miRNA oligonucleotides were designed with 5′ biotinylation and a 3′ Cy5 fluorophore and synthesized by Integrated DNA Technologies. Microarrays were blocked for 1 h with RNA-binding buffer (20 mM Tris-HCl at pH 7.4, 200 mM KCl, 0.2 mM EDTA, 0.05% NP40, 2 µg/µl heparin). miRNA oligonucleotides were prepared for microarray probing by incubating miRNA at 1 µM with streptavidin–Cy3 in RNA-binding buffer in presence of 0.4 U RNase inhibitor. Primed miRNA oligonucleotides were then incubated with microarrays overnight at 4°C in RNA-binding buffer with RNase inhibitor. Microarrays were dried and scanned using the InnoScan 1100 AL Fluorescence Scanner (Innopsys) and intensities were quantified using Mapix software (Innopsys).

The Invitrogen Ultimate 16,813 ORF collection, subcloned into a yeast expression vector that allows for galactose-dependent overexpression of N-terminal GST- and 6× His-tagged recombinant proteins, was spotted onto Fullmoon slides (Fullmoon Biosystem) using the NanoPrint LM210 system (ArrayIT) as previously described (Hu et al., 2009). The quality of these protein microarray chips, along with that of control proteins (such as IgG, GST and histones), was assayed using a randomly selected slide from the same batch, and at least 89% of the proteins were expressed correctly.

To identify miRBPs, the protein microarray chips were equilibrated and washed three times for 15 min with RNase-free PBS at pH 7.0 with 2 U/μl RNase inhibitor SUPERase-In (Thermo Fisher Scientific). The proteins on the chips were blocked with filtered 3% high quality IgG-/protease-free BSA (Jackson ImmunoResearch Laboratory) for 1 h at room temperature, then washed three times with RNase-free PBS buffer (filtered RNase-free PBS pH 7.0 with 1 U/μl RNase inhibitor). The protein microarray chips were incubated with internally Cy5- or DY647-labeled let-7b, miR-21, and miR-130b (100 ng) for 1 h, wrapped in aluminum foil on a shaking platform at 50 rpm. The chips were washed five more times with RNase-free PBS buffer for 10 min.

For detection of all possible proteins, the chips were then incubated with rabbit anti-GST antibody (71-7500, Invitrogen) at 1:5000 for 1 h at room temperature, washed three times for 15 min in RNase-free PBS buffer, and incubated with 1:1000 Cy3 goat anti-rabbit IgG antibody (A10520, Invitrogen) in RNase-free PBS buffer for 20 min at room temperature. The final washing was performed three times for 15 min in RNase-free PBS buffer and rinsed once in double-distilled H₂O. Protein microarray chips were dried via centrifugation at 200 g for 2 min using a 50 ml conical tube.

The chips were scanned with GenePix 4000B (Axon Instruments) for quantification and statistical analysis. Fluorescence from bound Cy5-/DY647-labeled let-7b, miR-21, and miR-130b was measured at 635 nm, and from Cy3-labeled anti-GST at 532 nm. Signal intensity values for each spot were obtained by scanning to find the ratio of foreground to background signals, and normalized with GST signal intensity. The mean and standard deviation of signal intensity of all proteins on the chip were calculated using R package, and P≤0.05 was considered statistically significant.

The list of 60 identified let-7b-binding proteins was submitted to the online DAVID functional annotation resource (v6.8) (Huang et al., 2009a,b), where the overrepresented proteins were identified and compared with the human proteome database.

EV isolation was performed as previously described (Kang et al., 2024; Kwon et al., 2016). Briefly, the cells were cultured with a freshly replaced serum-free medium. After 30 h, the conditioned medium containing EVs was collected and centrifuged at 500 g for 20 min. The supernatant was then incubated with 1 U/ml of benzonase (Millipore Sigma) for 15 min at room temperature and subsequently centrifuged at 2000 g for 20 min at 25°C. The obtained supernatant was then subjected to centrifugation at 10,000 g for 1 h at 25°C and subsequently filtered through a 0.2-μm asymmetric polyethersulfone membrane (Millipore Sigma). The filtered supernatant underwent tangential flow filtration using a 500-kDa cutoff Biomax cassette (Millipore Sigma). The feed rate was set to 4–5 ml/min with a poly pump (Buchler), ensuring that inlet and operating pressures remained below 68,947.6 Pa and 24,131.66 Pa, respectively. The retained particles were then washed with Dulbecco's phosphate-buffered saline (DPBS, Thermo Fisher Scientific). The volume of the final processed sample was typically concentrated 10- to 15-fold. The EV-containing retentate was subsequently centrifuged at 200,000 g using a SW60 Ti rotor (Beckman Coulter) for 1 h at 4°C. After resuspending the pellet in DPBS, the isolated EVs were stored at −80°C for future use. The size and number of EVs were measured using nanoparticle tracking analysis, and samples were stored at −20°C.

The crystal structure of QKI (PDB code 4JVH; Teplova et al., 2013) was used to calculate suitable binding conformations and associated binding affinities for all possible six-nucleotide subsequences of let-7b, miR-21 and miR-130b. Molecular dockings were carried out by the program PyRx Virtual Screening Tool Autodock Vina v0.9.7 (https://vina.scripps.edu/) for every RNA fragment, five times per model, with the determination of minimum docking energy among 15 calculated values. To run the software, the PDB file was converted to PDBQT format via OpenBabelGUI program (https://openbabel.org/docs/GUI/GUI.html). The docking models were presented by PyMOL software (The PyMOL Molecular Graphics System, v2.0 Schrödinger, LLC.).

To verify the RNA-binding affinity of QKI with let-7b, miR-21, miR-130b, let-7b mutant and N20 random RNA, recombinant QKI was incubated with 3 nM Cy3-labeled RNAs in assay buffer (20 mM HEPES pH 7.4, 150 mM NaCl, 2 mM MgCl₂, 1 U/µl RiboLock RNase inhibitor and 10 mM dithiothreitol) at room temperature. The assay was performed in 96-well non-binding black plate (Corning), with fluorescence polarization measured using a Spectramax iD5 microplate reader (Molecular Devices), with excitation and emission wavelengths at 530 nm and 570 nm, respectively. Millipolarization units (mP) were used to express fluorescence polarization values defined by the equation mP=1000×[(I_II−I_⊥)/(I_II+I_⊥)], where I_II and I_⊥ are parallel and perpendicular emission intensity measurements, respectively. The data were corrected for background (RNA sample alone) and G-factor, then fitted by a one-site binding model using the equation y=y₀+(y_maxx/K_d+x), where y is the corrected fluorescence polarization, y₀ is the polarization value without protein, y_max is the maximum binding fluorescence polarization signal, x is the protein concentration, and K_d is the dissociation constant. x and y_max were used as fitting parameters and nonlinear regression was performed using GraphPad Prism. Measurements were taken after 5 min incubation between the protein and RNA at room temperature.

All manipulations were performed with Institutional Animal Care and Use Committee (IACUC) approval from MD Anderson Cancer Center. Whole brains were collected from Qki^+/+ and Qki^−/− using Nestin- or Plp-Cre driver mice (Shingu et al., 2017). The frozen brain tissues were sliced and the resulting slides were post-fixed in 1× PBS containing 4% paraformaldehyde for 10 min, permeabilized with PBS containing 0.3% Triton X-100 (0.3% PBST) for 10 min, and then blocked with 0.3% PBST containing 10% bovine serum albumin (BSA) for 1 h at room temperature. The slides were then incubated with the corresponding primary antibodies overnight at 4°C. All antibodies were diluted 1:1000 in 0.3% PBST. The following primary antibodies were used: mouse anti-NeuN (Millipore, MAB377), rabbit anti-QKI5 (Bethyl Laboratories, IHC-00574), mouse anti-CNPase (Millipore, MAB326), rabbit anti-CNPase (Santa Cruz Biotechnology, sc-30158), mouse anti-GFAP (Millipore, MAB360), goat anti-GFAP (Santa Cruz Biotechnology, sc-6170), mouse anti-Iba1 (Millipore, MABN92), rabbit anti-Iba1 (Wako, 019-19741), rabbit anti-TLR7 (Novus Biologicals, NBP2-24906), mouse anti-J2 (SCICONS, 10010500, Hungary), rabbit anti-NF (BioLegend, 841001), rabbit anti-TLR3 (NB100-56571, Novus Biologicals) and mouse anti-pan-QKI (NeuroMab, 75-168, Davis, CA, USA). After primary antibody incubation, the slides were washed with PBS and incubated with Alexa Fluor 488- or 594-conjugated secondary antibodies (a21206 or a21203, Life Technologies) diluted in 0.3% PBST (1:1000) for 2 h at room temperature. After washing with PBS, to observe the nucleus, the slides were counterstained with 4′,6-diamidine-2′-phenylindole dihydrochloride (DAPI, Roche, 10236276001, Germany). The slides were then washed again with PBS, mounted and analyzed by confocal microscopy (LSM700; Carl Zeiss, Germany). Anti-pan-QKI antibody (NeuroMab, 75-168) was generated by using amino acids 1–341 (full-length) of human QKI5, which shares identity with mouse Qki (100%, 341/341 amino acids identical) and rat Qki (99%, 339/341 amino acids identical). The anti-Qki5 antibody was generated by using Qki5 C-terminal exon 8 (identical in human and mouse), which does not exist in Qki6, Qki7 and Qki7b) (Thermo Fisher Scientific, IHC-00574, previously from Bethyl Laboratories).

Chromatin immunoprecipitation assay was performed using SH-SY5Y cells treated with EVs isolated from conditioned media of either control or QKI shRNA-treated HeLa cells. For cross-linking, the cells exposed to EVs were treated with 1% formaldehyde for 10 min at room temperature. Crosslinking reactions were quenched with 125 mM glycine for 5 min at room temperature. The cells were washed with PBS twice and centrifuged at 1000 g for 5 min. After centrifugation, cell pellets were lysed in IP buffer composed of 0.1% SDS, 1% Triton X-100, 0.1% sodium deoxycholate and 140 mM NaCl containing protease inhibitors. The cell lysates were sonicated for 10 s with 10 s interval four times. Sheared chromatin was incubated with either 3 µg anti-IRF7 antibody (Santa Cruz Biotechnology, sc-74472) or normal IgG antibody (Santa Cruz Biotechnology, sc-2025) conjugated with protein A/G-Sepharose beads (#20349, Thermo Fisher Scientific) overnight at 4°C. After the beads were washed five times with wash buffer composed of 0.1% Triton-X, 20 mM Tris (pH 8.0), 150 mM NaCl and 1 mM EDTA, the beads were incubated with 10 unit/µl RNase A and 300 mM NaCl for 4 h at 65°C. Then, the beads were further incubated with proteinase K digestion buffer composed of 50 mM Tris-HCl (pH 8.0), 10 mM EDTA and 0.4 mg/ml Proteinase K for 2 h at 65°C. The immunopellets were subjected to phenol/chloroform/isoamyl alcohol (25:24:1) and the purified DNA was assayed with qPCR using the primers amplifying IFNA1 gene promoter sequences. Data were normalized to input and analyzed relative to the nonspecific IgG control.

De novo-synthesized let-7b miRNA via EVs was tracked using the Click-iT Nascent RNA Capture Kit (Invitrogen, C10365). Briefly, to label let-7b miRNA, QKI-depleted or control donor cells (HeLa) were treated with 0.28 mM 5-EU (E10345, Invitrogen) for 24 h and EVs containing 5-EU-labeled RNA were isolated from the conditioned media. The EVs were added to culture medium of SH-SY5Y cells and the cells were grown for 6 h; then, the total RNA of recipient cells was isolated by TRIzol (Invitrogen, 15596026) according to the manufacturer's protocol. For purification of 5-EU-labeled RNA, 5 µg of total RNA and 0.5 mM biotin-azide (B10184, Invitrogen) was used for RNA biotinylation, and precipitated with 7.5 M ammonium acetate in combination with ice-cold 100% ethanol. The biotinylated RNAs were subsequently incubated with streptavidin-conjugated beads for 1 h at room temperature with gentle agitation. Then, the beads were washed eight times with supplied wash buffer according to the manufacturer's instructions. The 5-EU-labeled RNAs on beads were extracted with elution buffer composed of 20 mM Tris (pH 7.4) and 1 mM EDTA for 4 min at 92°C, and the labeled let-7b was determined by qPCR in comparison with the standard condition from which EVs were isolated without 5-EU labeling and applied to the recipient cells.

Protein–protein interactions were sourced from BioGRID, build 4.3.194 (Stark et al., 2006). Networks were initially created using R software to select the appropriate protein–protein interactions from hit proteins and BioGRID data. Output tables were uploaded into Cytoscape 3.8.0 to create visual figures. Previous ‘known’ hits are from a previous publication (Treiber et al., 2017).

We thank D. Patel (Memorial Sloan Kettering Cancer Center, NY, USA) and M. J. Gamble (Albert Einstein College of Medicine, NY, USA) for providing plasmids. We appreciate G. M. Wilson (University of Maryland School of Medicine, MD, USA) for providing critical comments. We appreciate the core facilities at the University of Oklahoma, and the Medical University of South Carolina.

The peer review history is available online at https://journals.biologists.com/jcs/lookup/doi/10.1242/jcs.261575.reviewer-comments.pdf