Stress Granules Contain Rbfox2 with Cell Cycle-related mRNAs

Rbfox RNA-binding proteins play important roles in the regulation of alternative pre-mRNA splicing, but their role in other gene regulatory mechanisms is not well understood. Here, we show that Rbfox2 is a novel constituent of cytoplasmic stress granules, the translational silencing machinery assembled in response to cellular stress. We also show that the RNA binding activity of the Rbfox family protein is crucial for its localization into stress granules. To investigate the role of Rbfox2 in stress granules we used RNA-immunoprecipitation sequencing to identify cytoplasmic transcriptome-wide targets of Rbfox2. We report that a subset of cell cycle-related genes including retinoblastoma 1 is the target of Rbfox2 in cytoplasmic stress granules, and Rbfox2 regulates the retinoblastoma 1 mRNA and protein expression levels during and following stress exposure. Our study proposes a novel function for Rbfox2 in cytoplasmic stress granules.


Results
Rbfox2 is a component of SGs. Rbfox2 is predominantly localized in the nucleus, as seen in human cervical carcinoma HeLa cells (Fig. 1). To investigate the role of Rbfox2 under stress conditions, we examined the subcellular localization of Rbfox2 following treatment with arsenite, a well-known oxidative stress inducer that has been shown to induce the formation of cytoplasmic SGs in a variety of different cell types 34 . Although nuclear localization of the splicing factor prolineand glutamine-rich (SFPQ) was not affected by arsenite-induced cellular stress, arsenite treatment induced the formation of cytoplasmic Rbfox2 granules in HeLa cells (arrowhead, Fig. 1). In order to determine the composition of these Rbfox2-containing cytoplasmic granules, Rbfox2 was co-immuno-stained with two SG markers, TIA-1 and G3BP1 and found to co-localize (Fig. 2a,b). However, Rbfox2-containing cytoplasmic granules were distinguished from GW182 (a 182 kDa protein with multiple glycine/tryptophan repeats) positive processing bodies (P-bodies) which are involved in mRNA decay (Fig. 2c). In addition, light chain 3 A/B (LC3A/B)-stained autophagy was also clearly separate from Rbfox2-containing cytoplasmic granules (Fig. 2d). These results support that Rbfox2 is a component of SGs. We further asked whether Rbfox2 is a bona-fide component of SGs using the strategy of trapping mRNP into polysomes with cycloheximide, an inhibitor of translation elongation, which stabilizes polysomes. HeLa cells were cultured in the presence of various doses of arsenite with cycloheximide. Cycloheximide led to a strong reduction of Rbfox2-containing SGs in various doses of arsenite (Fig. 3). Since an important property of SGs is that their formation is antagonistic to polysome formation 35 , these results suggest that Rbfox2 is a bona-fide component of SGs in arsenite-treated cells. In addition, we observed localization of Rbfox2 in G3BP1 positive SGs in human head and neck carcinoma HN13 cells, following arsenite treatment (Suppl. S1).
We then tested whether Rbfox2 localization into SGs could be induced by other cellular stresses. High salt and heat shock induced Rbfox2 to localize into G3BP1 positive SGs (Fig. 4a). We next asked whether Rbfox2 plays a role in SG formation. Depletion of Rbfox2 using a specific siRNA against endogenous Rbfox2 mRNA (siRbfox2) Figure 1. Sodium arsenite treatment induces cytoplasmic Rbfox2 granules. Immunofluorescence images of SFPQ (red) and Rbfox2 (green) proteins were visualized in untreated (Control) and sodium arsenitetreated (500 μM; 40 min) HeLa cells (Stress). DAPI stains nuclei. Arrowheads indicate cytoplasmic Rbfox2 granules. Scale bar, 20 μm. has no effect on SG formation in the presence of various doses of arsenite (Fig. 4b), indicating that Rbfox2 is not essential for SG formation. Next we examined the time course of SG assembly and disassembly in HeLa cells during exposure to and removal of cellular stress. Following 40 min of exposure to arsenite, cells were washed and allowed to recover in complete growth medium for various times. SGs were no longer detectable after a 3 h recovery period (Suppl. S2), showing that SG formation is highly dynamic and fully reversible upon removal of the stressor.
Rbfox family proteins are recruited into SGs. Mammalian genomes encode three Rbfox members: Rbfox1, Rbfox2, and Rbfox3. All three proteins contain an almost identical RRM in the middle of the molecule. Although there are a number of alternatively spliced isoforms for each gene, the C-terminal domains of the major isoforms from the three genes also show a high sequence similarity. The high degree of sequence identity raised the possibility that Rbfox family proteins other than Rbfox2 may also localize into SGs. To address this possibility, expression plasmids of myc-tagged Rbfox1, Rbfox2, and Rbfox3 were transfected into HeLa cells, and the exogenously expressed proteins were monitored by immunofluorescence microscopy using anti-myc and the SG marker TIA-1 antibodies. Anti-Rbfox2 and anti-myc immunoblot results indicate the expression levels of endogenous and exogenous Rbfox2 proteins, respectively (Supple. S3). As previously reported 22,36 , all three members of the Rbfox family showed strong nuclear localization in the absence of stress (Fig. 5a-c, upper rows, Control). Cellular stress induced SG localization of all exogenously expressed Rbfox family proteins (Fig. 5a-c, lower rows, Stress), suggesting that SG localization is a general property of Rbfox family proteins. We next determined the domain that is responsible for the localization of Rbfox into SGs. Since the domains of the Rbfox3 protein had previously been investigated by us 6 , we decided to determine whether the N-terminal and C-terminal domains flanking the RRM are required for SG localization using Rbfox3 as representative of the Rbfox family. The N-terminal half of Rbfox3 including the RRM (N-RRM) and the C-terminal half (RRM-C) tagged with myc were expressed at their expected size, as shown in Fig. 6a and b. Although N-RRM, where the C-terminal domain was deleted, distributed diffusely in both nucleus and cytoplasm, it still localized into SGs upon arsenite treatment (Fig. 6c). Similarly, RRM-C, where the N-terminal domain was deleted, showed a strong nuclear localization property and localized into SGs (Fig. 6d). These results suggest that neither the N-terminal nor C-terminal domain of Rbfox3 is critical for SG localization. Next, we investigated whether the RNA binding activity of Rbfox3 through its RRM is required for SG localization using two different Rbfox3 mutant proteins. Rbfox family proteins contain two critical phenylalanine (F) residues within the RRM domains (amino-acids 108 and 142 in Rbfox3) that are required for RNA-binding activities. The F142 mutation to alanine (F142A) in Rbfox3 abrogates its binding activity to both the canonical binding motif UGCAUG and noncanonical target RNAs. The F108A Rbfox3 mutant protein is unable to bind canonical RNAs but is able to bind to noncanonical RNA targets 22 . Surprisingly, the F142A mutant protein completely lost the ability to localize into SGs (Fig. 6f), while the F108A mutant protein still localized into SGs (Fig. 6e). These results suggest that the RNA binding activity of the Rbfox family protein is crucial for its localization into SG and that Rbfox target RNAs recruited to SGs do not necessarily contain the canonical sequence.
Rbfox2 binds to the mRNAs in SGs that are involved in the cell cycle. The fact that Rbfox is a component of SGs and its RNA binding activity is critical for its SG localization prompted us to investigate its role in SGs by identifying its target RNAs. To identify the RNA targets associated with the Rbfox2 protein, we performed  Fig. 6a. Immunofluorescence images of myc-Rbfox (anti-myc, red) and TIA-1 (green) proteins were visualized in untreated (Control) and sodium arsenite-treated HeLa cells (Stress). DAPI stains nuclei. Scale bar, 20 μm.
an RIP-seq analysis using cytosolic fractions from arsenite-treated HeLa cells. Consistent with our observations that arsenite treatment induces Rbfox2 localization into cytosolic SGs, Rbfox2 proteins in the cytosolic fraction were increased by arsenite treatment (Suppl. S4). Moreover, the RNA-binding protein SFPQ, in agreement with the microscopy results shown in Fig. 1, was undetectable in the cytosolic fraction in both unstressed and stressed conditions, confirming the absence of nuclear contamination. Immunoprecipitation was carried out using either anti-Rbfox2 ("Bethyl" Laboratories, hereafter referred to as RIP-seq-A) or anti-Rbfox2 (produced in this laboratory "LMC", hereafter referred to as RIP-seq-B) antibodies to exclude potential non-specific interactions and artifacts. Immunoprecipitated RNA (200-1,200 nt) was reverse transcribed to cDNA, ligated with adapters, PCR-amplified, and sequenced using an Illumina HiSeq.
We obtained 23 and 28 million 50-nucleotide paired-end reads uniquely mapped to the human genome from RIP-seq-A and RIP-seq-B in arsenite treated cells, respectively (Supplementary Table S1). Two replicates of untreated HeLa cell total RNA used for controls generated 61 and 75 million reads uniquely mapped to the human genome (Supplementary Table S1). More than 90% of the target genes were identified in both RIP-seq data sets (Fig. 7a, Venn diagram), demonstrating the high reproducibility of the RIP-seq experiments. Thus, further analyses were performed on the data set of RIP-seq-B. Twenty eight and 72% of the reads were mapped to intergenic and genic regions, respectively. The binding regions of Rbfox2 at the genic regions are distributed as follows: 64% intron only, 26% exon only, 7% exon-exon junction and 2% exon-intron junction regions (Fig. 7b). Unexpectedly, we observed a large amount of intronic read hits, suggesting that incompletely spliced pre-mRNAs were released from the nucleus to the cytoplasm upon stress exposure. Although another possible explanation is that this observation is partly due to the presence of nuclear fractions, the presence of only 2% reads from exon-intron junction regions compared with 7% reads from exon-exon junction regions suggests that pre-mRNA contamination from nuclear fractions is not the main reason. To explicitly identify the transcripts of Rbfox2-binding targets, we made a conservative estimate of the enrichment of RNAs in RIP versus controls. To visualize the enrichment data, scatter plots were constructed using the log-transformed and normalized read numbers (Fig. 7c). The dashed rectangle represents the enriched genes in Rbfox2-RIP (Supplementary Table S2). A gene ontology (GO) analysis of the enriched genes identified multiple GO terms, of which the highly significant terms were related to the cell cycle (Fig. 7d). Among the enriched genes, retinoblastoma 1 (RB1) mRNA was identified as the most significant target of Rbfox2 (Fig. 7c). Moreover, we observed the enrichment of Rbfox2 binding to the 3′-UTR of RB1 mRNA in both RIP-seq analyses (Fig. 7e), and the sequence of the 3′-UTR of RB1 mRNA is highly conserved in vertebrates (Fig. 7f). All other hits with smaller numbers corresponded to exon regions but not to introns, suggesting that Rbfox2 associated with mature RB1 mRNA. Although the results obtained with the RRM-mutant Rbfox3 (Fig. 6e,f) suggest that Rbfox target RNAs do not necessarily contain the Rbfox consensus sequence in order to be recruited to SGs, the human 3′-UTR of RB1 mRNA contains two (U)GCAUG elements, one of which is conserved in mouse. We then confirmed the increase in binding between Rbfox2 and RB1 mRNA during oxidative stress using RIP-PCR analysis. After co-immunoprecipitation of Rbfox2-RNA complexes from HeLa cell extracts with or without arsenite-treatment, the 3′-UTR region of RB1 mRNA was amplified by RT-PCR. Rbfox2 binding to RB1 mRNA was greatly increased by arsenite treatment (Fig. 7g). This increment was not detected in the control IgG or Gapdh mRNA, confirming that Rbfox2 binds specifically to RB1 mRNA and that their interaction is increased by cellular stress.
Rbfox2 regulates RB1 expression during and following stress exposure. SGs function to protect selected mRNAs and suppress their translation under stress conditions and SGs dissociate when the stress is removed or when cells adapt to stress, thereby allowing SGs to play a role in protein expression during and following cellular stress. Therefore we investigated the role of Rbfox2 in RB1 expression by examining its ability to regulate RB1 mRNA stability and translation, using an siRNA strategy. Immunoblot analysis showed that endogenous Rbfox2 protein is efficiently depleted by the specific siRNA against Rbfox2 (Fig. 8a). No differences were observed in the basal expression level of RB1 mRNA upon Rbfox2 depletion (data not shown). We then measured the level of RB1 mRNA by quantifying the rate of decay after transcription was blocked by actinomycin D treatment. The level of RB1 mRNA was not altered by arsenite treatment in control cells. In contrast arsenite treatment significantly reduced the level of RB1 mRNA in Rbfox2-depleted cells by the 3 h recovery time, when SGs dissociated (Fig. 8b, left panel). On the other hand, the stability of a control stable transcript, Gapdh mRNA, encoding a housekeeping gene product, was not affected by Rbfox2 depletion and arsenite treatment (Fig. 8b,  right panel), demonstrating that not all mRNA levels decreased rapidly following Rbfox2 depletion in stressed cells. These results suggest that Rbfox2 is required to protect RB1 mRNA from degradation presumably by incorporating RB1 mRNA into SGs under stress conditions. To further assess whether the binding of Rbfox2 to RB1 mRNA influences the protein level of RB1, immunoblot analysis, at two different recovery time points using anti-RB1 antibody, was carried out and the level of RB1 protein was determined. Although RB1 protein levels were the same in all conditions by 3 h of recovery, we observed reduced RB1 protein levels after Rbfox2 depletion and arsenite treatment at the 24 h recovery time point (Fig. 8c,d), presumably due to the reduced RB1 mRNA level. Our results demonstrate that Rbfox2 prevents RB1 mRNA degradation in SGs under cellular stress and that Rbfox2 is required to maintain RB1 protein expression in the post-stressed condition.

Discussion
The goal of our study was to investigate the cytoplasmic function of Rbfox2 under stress conditions. Here, we used RIP-seq analysis to identify target RNAs of Rbfox2 in cytoplasmic SGs and found that genes affecting the cell cycle are significantly enriched in Rbfox2 targets. Our findings underscore the importance of a novel function of Rbfox2 that allows significant gene regulatory flexibility which, in turn enables cells to adjust to external conditions. (f) Phylogenetic conservation of 3′ UTR of RB1. Conservation of the RB1 targeting 3′ UTR is shown by the phastCons track on the UCSC Genome Browser in the human hg19 genome. The phastCons score estimates the probability of each nucleotide, using a hidden markov model-based method, belonging to conserved elements. This track shows multiple alignments of 100 vertebrate species and measurements of evolution conservation. (g) After treatment of HeLa cells with sodium arsenite, RIP-qRT-PCR (IP followed by qRT-PCR) analysis was used to measure the levels of enrichment of RB1 mRNA and GAPDH mRNA associated with immune-complexes using anti-Rbfox2 (αRbfox2) or control IgG; the data represented as enrichment of each mRNA in immunocomplexes were compared with 5% input. Mean values ± SD (error bars) are shown for three independent experiments. *P < 0.01 (two-sided t test), comparing stressed with unstressed RB1 mRNAs.
Scientific REPORtS | 7: 11211 | DOI:10.1038/s41598-017-11651-w Each member of the Rbfox family of genes is known to generate a number of isoforms through splicing of multiple alternative exons and alternative promoters 36 . Both the N-terminal and C-terminal amino acid sequences affect their subcellular localization. Major isoforms from each gene predominantly localize in nuclei. Some isoforms with a reading frame shift at the C-terminus localize predominantly in the cytoplasm. However, to what extent each isoform of the Rbfox proteins shuttles between nucleus and cytoplasm has not been studied. In this report, we show that endogenous Rbfox2 isoforms and an exogenously expressed Rbfox2 isoform which are predominantly localized in nuclei are recruited to SGs in the cytoplasm under stress conditions. These observations suggest the idea that at least some isoforms of Rbfox2 translocate from the nucleus to the cytoplasm, although the amounts of translocated proteins by stress are small relative to the total amounts. Cytoplasmic form(s) of Rbfox2, which seems to be diffusely distributed in unstressed cells, might also be recruited to SGs, since the N-RRM deletion mutant of Rbfox3, which distributes to the cytoplasm to a significant extent, is also recruited to SGs.
Although we have not formally tested Rbfox2 deletion mutants, it is likely that the N-and C-terminal domains of Rbfox2 are not essential for targeting to SGs, based on the data from Rbfox3 deletion mutants. It is also unlikely that a unique amino acid sequence of Rbfox2 is required for SG localization, because all exogenously expressed Rbfox1, 2, and 3 show recruitment into SGs to a similar extent. However, we cannot completely rule out the possibility that the Rbfox3 deletion mutant or full-length Rbfox3 and 1 forms a dimer (or multimer) with endogenous intact Rbfox2 through the RRM domain and is recruited to SGs together with endogenous Rbfox2. Of note, RNA binding activity is essential for Rbfox recruitment to SGs, whereas neither the N-terminal nor C-terminal domain is essential. This suggests that translocation of nuclear Rbfox to cytoplasmic SGs might be navigated by its target RNAs. This observation is in contrast to the fact that the C-terminal and/or N-terminal domains, in addition to the RNA binding activity, are required for regulation of alternative splicing in many cases 5,6 . However, understanding the detailed mechanism by which Rbfox undergoes SG localization will require precise kinetic probing of Rbfox binding to target RNAs and determination of the molecular mechanism underlying posttranslational modification of Rbfox during cellular stress. To date cytoplasmic Rbfox1 has been reported to increase mRNA stability and translation by binding to the 3′-UTRs of its target RNAs in unstressed conditions, in some cases by competing with miRNAs 20 . Our direct evidence for SG localization of Rbfox2 during cellular stress provides a new gene regulatory mechanism for Rbfox. In this study we provide comprehensive analysis of cytoplasmic Rbfox2-associated RNAs that are generated under oxidative stress. The fact that more than half of genic reads are mapped to introns suggests a leak of incompletely processed pre-mRNAs from the nucleus and/or a splicing inhibition under the stress condition. It is also possible that the number of intron-mapped reads is apparently large because intron length is generally much longer than exon length. Shalgi et al. reported widespread inhibition of pre-mRNA splicing during heat stress and nuclear retention of intron-containing transcripts 37 . Interestingly, they observed splicing-inhibited transcripts were mostly spliced posttranscriptionally in the nucleoplasm and large intron-containing transcripts are more susceptible to splicing inhibition. The physiological significance of intron-containing RNAs associated with cytoplasmic Rbfox2 will be determined in a future work.
Importantly, we found that the target RNAs of cytoplasmic Rbfox2 are related to the cell cycle progression. Among Rbfox2-target RNAs in SGs, we demonstrated the significance of RB1 mRNA association with Rbfox2 to regulate RB1 mRNA stability and RB1 protein expression under oxidative stress. The generally accepted roles of SGs are to provide a pause in translation of incorporated mRNAs and to protect them from degradation 25,26 . SGs are transient and reversible structures. Upon removal of stress or adaptation of cells in a new environment, SGs are disassembled and then the protected mRNAs become readily available for translation. RB1 was initially identified as a tumor suppressor gene that is dysfunctional in several major cancers 38 . Since abnormal cells, including cancer cells, are under various cellular stresses such as oxidative stress, nutrient stress, and osmotic stress, we speculate that regulation of the RB1-mediated cell cycle by Rbfox2 could play a role in disease development and cancer progression.
Cells are constantly encountering numerous stresses such as an excess in hormone secretion, viral infections, oxidative stress, exposure to an abnormal pH, heat shock, hypoxia, and high salt concentrations. Recently, it was reported that cytoplasmic Rbfox3 levels were significantly higher in brain sections obtained from patients infected with human immunodeficiency virus type I (HIV) compared to those obtained from controls 39 . Although SG localization of HIV-induced-cytoplasmic Rbfox3 has not been confirmed, this is consistent with our observation that the subcellular localization of Rbfox2 is altered during cellular stress. Since the Rbfox family proteins are highly expressed in neuronal cells and all three members of this family are capable of responding to cellular stress, Rbfox proteins might play a role in development or prevention of neurodegenerative diseases which are often triggered by cellular stresses.

Materials and Methods
Cell culture, transfection, and chemicals. The human cervix adenocarcinoma cell line HeLa was maintained in Dulbecco's modified Eagle medium (DMEM) supplemented with 10% fetal bovine serum and antibiotics. Cells were transfected with plasmids or siRNA by electroporation (Amaxa Nucleofector, Lonza). Transfected cells were generally analyzed 36 hr later. Sodium arsenite and sodium chloride were purchased from Sigma-Aldrich. Stress conditions were induced by treating cells with different stress inducers such as 500 μM sodium arsenite, 150 mM sodium chloride, or 43 °C heat shock for 40 min.
Plasmids and siRNA. The preparation of the expression plasmids for Rbfox1, Rbfox2, and Rbfox3 in the pCS3 + MT vector, which contains an N-terminal myc-tag, has been described previously 7,36 . The N-RRM and RRM-C deletion constructs of Rbfox3 have been described previously 6 . The generation of F108A and F142A mutant clones of Rbfox3 has been described previously 22 . The generation of the siRNA designed for silencing the Rbfox2 (siRbfox2) has been described previously as a Fox2 siRNA 6 .

RNA-protein complex immunoprecipitation sequencing (RIP-seq) and PCR. A cytosolic fraction
of sodium arsenite-treated HeLa cells was isolated for SG enrichment using a Nuclear Extract Kit (Active Motif). Immunoprecipitation of RNA-Rbfox2 complexes was performed with an RNA ChIP-IT kit (Active Motif) using two different Rbfox2 antibodies; one obtained from Bethyl Laboratory, A-300-864A, for RIP-seq-A and a second generated in this laboratory (LMC) 6 for RIP-seq-B. The RNAs recovered from the complexes were subjected to cDNA-library preparation using a TruSeq Standard Total RNA library Prep Kit (Illumina). An Illumina Hiseq. 2000 platform was used to generate 100-bp and 50-bp paired-end reads for RIP-sequencing of Rbfox2 and control RNA-sequencing of total RNAs from untreated HeLa cells, respectively. For RIP-PCR, the RNAs recovered from the complexes of whole-cell extracts were subjected to RT-PCR using the primers 5′-CCA TGG AGA AGG CTG GGG-3′ and 5′-CAA AGT TGT CAT GGA TGA CC-3′ for Gapdh mRNA and 5′-GTT GGT CCT TCT CGG TCC TT-3′ and 5′-CAA AGC AGA AGG CAA CTT GA-3′ for RB1 mRNA.
Bioinformatics analysis of RIP sequencing. After the sequencing run, the resulting raw reads in FASTQ format were processed using in-house Perl scripts. This enabled clean reads to be obtained by removing those containing adapter sequences, poly-N sequences, or low quality bases (below a mean Phred score of 20). The trimmed paired-end reads were mapped to the UCSC human reference genome (hg19) using Tophat 2 40 under the default parameter setting. Only uniquely map-able reads with a maximum of two mismatches when compared with the reference were considered. The mapped results were fed to Cufflinks 41 for quantification of known human genes. Differential expression analysis was performed with Cuffdiff 42 . Genes with a false discovery rate (FDR) of less than 5% were marked as significant. Unless otherwise stated, all gene expression levels in our analyses are represented as FPKM (fragments per kilobase of exon per million sequences fragments) values on log 2 scales. The RNA-sequencing statistics on all four samples are listed in Supplementary Table S1.